{"abstract": "  This paper proposes a new Jacobian-based inverse kinematics (IK) explicitly\nconsidering box-constrained joint space. To control humanoid robots, the\nreference pose of end effector(s) is planned in task space, then mapped into\nthe reference joints by IK. Due to the limited analytical solutions for IK,\niterative numerical IK solvers based on Jacobian between task and joint spaces\nhave become popular. However, the conventional Jacobian-based IK does not\nexplicitly consider the joint constraints, and therefore, they usually clamp\nthe obtained joints during iteration according to the constraints in practice.\nThe problem in clamping operation has been pointed out that it causes numerical\ninstability due to non-smoothed objective function. To alleviate the clamping\nproblem, this study explicitly considers the joint constraints, especially the\nbox constraints in this paper, inside the new IK solver. Specifically, instead\nof clamping, a mirror descent (MD) method with box-constrained real joint space\nand no-constrained mirror space is integrated with the conventional\nJacobian-based IK methods, so-called MD-IK. In addition, to escape local optima\nnearly on the boundaries of constraints, a heuristic technique, called\n$\\epsilon$-clamping, is implemented as margin in software level. As a result,\nMD-IK achieved more stable and enough fast i) regulation on the random\nreference poses and ii) tracking to the random trajectories compared to the\nconventional IK solvers.\n"}
{"abstract": "  The HGR is a quite challenging task as its performance is influenced by\nvarious aspects such as illumination variations, cluttered backgrounds,\nspontaneous capture, etc. The conventional CNN networks for HGR are following\ntwo stage pipeline to deal with the various challenges: complex signs,\nillumination variations, complex and cluttered backgrounds. The existing\napproaches needs expert expertise as well as auxiliary computation at stage 1\nto remove the complexities from the input images. Therefore, in this paper, we\nproposes an novel end-to-end compact CNN framework: fine grained feature\nattentive network for hand gesture recognition (Fit-Hand) to solve the\nchallenges as discussed above. The pipeline of the proposed architecture\nconsists of two main units: FineFeat module and dilated convolutional (Conv)\nlayer. The FineFeat module extracts fine grained feature maps by employing\nattention mechanism over multiscale receptive fields. The attention mechanism\nis introduced to capture effective features by enlarging the average behaviour\nof multi-scale responses. Moreover, dilated convolution provides global\nfeatures of hand gestures through a larger receptive field. In addition,\nintegrated layer is also utilized to combine the features of FineFeat module\nand dilated layer which enhances the discriminability of the network by\ncapturing complementary context information of hand postures. The effectiveness\nof Fit- Hand is evaluated by using subject dependent (SD) and subject\nindependent (SI) validation setup over seven benchmark datasets: MUGD-I,\nMUGD-II, MUGD-III, MUGD-IV, MUGD-V, Finger Spelling and OUHANDS, respectively.\nFurthermore, to investigate the deep insights of the proposed Fit-Hand\nframework, we performed ten ablation study.\n"}
{"abstract": "  The residue cocycle associated to a suitable spectral triple is the key\ncomponent of the Connes-Moscovici local index theorem in noncommutative\ngeometry. We review the relationship between the residue cocycle and heat\nkernel asymptotics. We use a modified version of the Getzler calculus to\ncompute the cocycle for a class of Dirac-type operators introduced by Bismut,\nobtained by deforming a Dirac operator by a closed 3-form B. We also compute\nthe cocycle in low-dimensions when the 3-form B is not closed.\n"}
{"abstract": "  We propose the adversarially robust kernel smoothing (ARKS) algorithm,\ncombining kernel smoothing, robust optimization, and adversarial training for\nrobust learning. Our methods are motivated by the convex analysis perspective\nof distributionally robust optimization based on probability metrics, such as\nthe Wasserstein distance and the maximum mean discrepancy. We adapt the\nintegral operator using supremal convolution in convex analysis to form a novel\nfunction majorant used for enforcing robustness. Our method is simple in form\nand applies to general loss functions and machine learning models. Furthermore,\nwe report experiments with general machine learning models, such as deep neural\nnetworks, to demonstrate that ARKS performs competitively with the\nstate-of-the-art methods based on the Wasserstein distance.\n"}
{"abstract": "  The dynamics and radiation of ultrarelativistic electrons in strong\ncounterpropagating laser beams are investigated. Assuming that the particle\nenergy is the dominant scale in the problem, an approximate solution of\nclassical equations of motion is derived and the characteristic features of the\nmotion are examined. A specific regime is found with comparable strong field\nquantum parameters of the beams, when the electron trajectory exhibits\nultrashort spike-like features, which bears great significance to the\ncorresponding radiation properties. An analytical expression for the spectral\ndistribution of spontaneous radiation is derived in the framework of the\nBaier-Katkov semiclassical approximation based on the classical trajectory. All\nthe analytical results are further validated by exact numerical calculations.\nWe consider a non-resonant regime of interaction, when the laser frequencies in\nthe electron rest frame are far from each other, avoiding stimulated emission.\nSpecial attention is devoted to settings when the description of radiation via\nthe local constant field approximation fails and to corresponding spectral\nfeatures. Periodic and non-periodic regimes are considered, when lab\nfrequencies of the laser waves are always commensurate. The sensitivity of\nspectra with respect to the electron beam spread, focusing and finite duration\nof the laser beams is explored.\n"}
{"abstract": "  Interactive driving scenarios, such as lane changes, merges and unprotected\nturns, are some of the most challenging situations for autonomous driving.\nPlanning in interactive scenarios requires accurately modeling the reactions of\nother agents to different future actions of the ego agent. We develop\nend-to-end models for conditional behavior prediction (CBP) that take as an\ninput a query future trajectory for an ego-agent, and predict distributions\nover future trajectories for other agents conditioned on the query. Leveraging\nsuch a model, we develop a general-purpose agent interactivity score derived\nfrom probabilistic first principles. The interactivity score allows us to find\ninteresting interactive scenarios for training and evaluating behavior\nprediction models. We further demonstrate that the proposed score is effective\nfor agent prioritization under computational budget constraints.\n"}
{"abstract": "  The performance of fine-tuning pre-trained language models largely depends on\nthe hyperparameter configuration. In this paper, we investigate the performance\nof modern hyperparameter optimization methods (HPO) on fine-tuning pre-trained\nlanguage models. First, we study and report three HPO algorithms' performances\non fine-tuning two state-of-the-art language models on the GLUE dataset. We\nfind that using the same time budget, HPO often fails to outperform grid search\ndue to two reasons: insufficient time budget and overfitting. We propose two\ngeneral strategies and an experimental procedure to systematically troubleshoot\nHPO's failure cases. By applying the procedure, we observe that HPO can succeed\nwith more appropriate settings in the search space and time budget; however, in\ncertain cases overfitting remains. Finally, we make suggestions for future\nwork. Our implementation can be found in\nhttps://github.com/microsoft/FLAML/tree/main/flaml/nlp/.\n"}
{"abstract": "  The relative abundance of alpha particles with respect to proton, usually\nexpressed as $A_{He}$ = ($n_\\alpha/n_p$)*100, is known to respond to solar\nactivity although changes in its behaviour in the last four solar cycles are\nnot known. In this letter, by systematically analysing inter-calibrated\n$A_{He}$ data obtained from the first Lagrangian point of the Sun-Earth system,\nwe show that $A_{He}$ variations are distinctively different in solar cycle 24\nas compared to the last three cycles. The frequency of $A_{He}$ = 2-3% events\nis significantly higher in slow/intermediate solar winds in cycle 24 as opposed\nto the dominance of the typical $A_{He}$ = 4-5% events in the previous three\ncycles. Further, the occurrence of $A_{He}$ $\\geq$ 10% events is significantly\nreduced in cycle 24. Not only that, the changes in delay of $A_{He}$ with\nrespect to peak sunspot numbers are less sensitive to changes in solar wind\nvelocity in cycle 24. The investigation suggests that the coronal magnetic\nfield configuration started undergoing systematic changes starting from cycle\n23 and this altered magnetic field configuration affected the way helium got\nprocessed and depleted in the solar atmosphere.\n"}
{"abstract": "  Large-area crop classification using multi-spectral imagery is a widely\nstudied problem for several decades and is generally addressed using classical\nRandom Forest classifier. Recently, deep convolutional neural networks (DCNN)\nhave been proposed. However, these methods only achieved results comparable\nwith Random Forest. In this work, we present a novel CNN based architecture for\nlarge-area crop classification. Our methodology combines both spatio-temporal\nanalysis via 3D CNN as well as temporal analysis via 1D CNN. We evaluated the\nefficacy of our approach on Yolo and Imperial county benchmark datasets. Our\ncombined strategy outperforms both classical as well as recent DCNN based\nmethods in terms of classification accuracy by 2% while maintaining a minimum\nnumber of parameters and the lowest inference time.\n"}
{"abstract": "  Man-in-The-Middle (MiTM) attacks present numerous threats to a smart grid. In\na MiTM attack, an intruder embeds itself within a conversation between two\ndevices to either eavesdrop or impersonate one of the devices, making it appear\nto be a normal exchange of information. Thus, the intruder can perform false\ndata injection (FDI) and false command injection (FCI) attacks that can\ncompromise power system operations, such as state estimation, economic\ndispatch, and automatic generation control (AGC). Very few researchers have\nfocused on MiTM methods that are difficult to detect within a smart grid. To\naddress this, we are designing and implementing multi-stage MiTM intrusions in\nan emulation-based cyber-physical power system testbed against a large-scale\nsynthetic grid model to demonstrate how such attacks can cause physical\ncontingencies such as misguided operation and false measurements. MiTM\nintrusions create FCI, FDI, and replay attacks in this synthetic power grid.\nThis work enables stakeholders to defend against these stealthy attacks, and we\npresent detection mechanisms that are developed using multiple alerts from\nintrusion detection systems and network monitoring tools. Our contribution will\nenable other smart grid security researchers and industry to develop further\ndetection mechanisms for inconspicuous MiTM attacks.\n"}
{"abstract": "  In this paper, we present a provably correct controller synthesis approach\nfor switched stochastic control systems with metric temporal logic (MTL)\nspecifications with provable probabilistic guarantees. We first present the\nstochastic control bisimulation function for switched stochastic control\nsystems, which bounds the trajectory divergence between the switched stochastic\ncontrol system and its nominal deterministic control system in a probabilistic\nfashion. We then develop a method to compute optimal control inputs by solving\nan optimization problem for the nominal trajectory of the deterministic control\nsystem with robustness against initial state variations and stochastic\nuncertainties. We implement our robust stochastic controller synthesis approach\non both a four-bus power system and a nine-bus power system under generation\nloss disturbances, with MTL specifications expressing requirements for the grid\nfrequency deviations, wind turbine generator rotor speed variations and the\npower flow constraints at different power lines.\n"}
{"abstract": "  It is often said that asymmetric dark matter is light compared to typical\nweakly interacting massive particles. Here we point out a simple scheme with a\nneutrino portal and $\\mathcal{O}(60 \\text{ GeV})$ asymmetric dark matter which\nmay be ''added'' to any standard baryogenesis scenario. The dark sector\ncontains a copy of the Standard Model gauge group, as well as (at least) one\nmatter family, Higgs, and right-handed neutrino. After baryogenesis, some\nlepton asymmetry is transferred to the dark sector through the neutrino portal\nwhere dark sphalerons convert it into a dark baryon asymmetry. Dark hadrons\nform asymmetric dark matter and may be directly detected due to the vector\nportal. Surprisingly, even dark anti-neutrons may be directly detected if they\nhave a sizeable electric dipole moment. The dark photons visibly decay at\ncurrent and future experiments which probe complementary parameter space to\ndark matter direct detection searches. Exotic Higgs decays are excellent\nsignals at future $e^+ e^-$ Higgs factories.\n"}
{"abstract": "  A high statistics $\\Sigma p$ scattering experiment has been performed at the\nJ-PARC Hadron Experimental Facility. Data for momentum-tagged $\\Sigma^{-}$\nrunning in a liquid hydrogen target were accumulated by detecting the $\\pi^{-}p\n\\to K^{+}\\Sigma^{-}$ reaction with a high intensity $\\pi^{-}$ beam of 20\nM/spill. Differential cross sections of the $\\Sigma^{-}p$ elastic scattering\nwere derived with a drastically improved accuracy by identifying the largest\nstatistics of about 4,500 events from 1.72 $\\times$ $10^{7}$ $\\Sigma^{-}$. The\nderived differential cross section shows a clear forward-peaking angular\ndistribution for a $\\Sigma^{-}$ momentum range from 470 to 850 MeV/$c$. The\naccurate data will impose a strong constraint on the theoretical models of the\nbaryon-baryon interactions.\n"}
{"abstract": "  Bound states in the continuum (BICs) are non-radiating solutions of the wave\nequation with a spectrum embedded in the continuum of propagating waves of the\nsurrounding space. The complete decoupling of BICs from the radiation continuum\nmakes their excitation impossible from the far-field. Here, we develop a\ngeneral theory of parametric excitation of BICs in nonlinear systems with\nKerr-type nonlinearity via spontaneous symmetry breaking, which results in a\ncoupling of a BIC and a bright mode of the system. Using the temporal\ncoupled-mode theory and perturbation analysis, we found the threshold intensity\nfor excitation of a BIC and study the possible stable and unstable solutions\ndepending on the pump intensity and frequency detuning between the pump and\nBIC. We revealed that at some parameters of the pump beam, there are no stable\nsolutions and the BIC can be used for frequency comb generation. Our findings\ncan be very promising for use in nonlinear photonic devices and all-optical\nnetworks.\n"}
{"abstract": "  Six significant new methodological developments of the previously-presented\n\"metastimuli architecture\" for human learning through machine learning of\nspatially correlated structural position within a user's personal information\nmanagement system (PIMS), providing the basis for haptic metastimuli, are\npresented. These include architectural innovation, recurrent (RNN) artificial\nneural network (ANN) application, a variety of atom embedding techniques\n(including a novel technique we call \"nabla\" embedding inspired by\nlinguistics), ANN hyper-parameter (one that affects the network but is not\ntrained, e.g. the learning rate) optimization, and meta-parameter (one that\ndetermines the system performance but is not trained and not a hyper-parameter,\ne.g. the atom embedding technique) optimization for exploring the large design\nspace. A technique for using the system for automatic atom categorization in a\nuser's PIMS is outlined. ANN training and hyper- and meta-parameter\noptimization results are presented and discussed in service of methodological\nrecommendations.\n"}
{"abstract": "  We present a full analysis of a broadband spectral line survey of Sagittarius\nB2 (Main), one of the most chemically rich regions in the Galaxy located within\nthe giant molecular cloud complex Sgr B2 in the Central Molecular Zone. Our\ngoal is to derive the molecular abundances and temperatures of the high-mass\nstar-forming region Sgr B2(M) and thus its physical and astrochemical\nconditions. Sgr B2(M) was observed using the Heterodyne Instrument for the\nFar-Infrared (HIFI) on board the Herschel Space Observatory in a spectral line\nsurvey from 480 to 1907 GHz at a spectral resolution of 1.1 MHz, which provides\none of the largest spectral coverages ever obtained toward this high-mass\nstar-forming region in the submillimeter with high spectral resolution and\nincludes frequencies > 1 THz unobservable from the ground. We model the\nmolecular emission from the submillimeter to the far-IR using the XCLASS\nprogram. For each molecule, a quantitative description was determined taking\nall emission and absorption features of that species across the entire spectral\nrange into account. Additionally, we derive velocity resolved ortho / para\nratios for those molecules for which ortho and para resolved molecular\nparameters are available. Finally, the temperature and velocity distributions\nare analyzed and the derived abundances are compared with those obtained for\nSgr B2(N) from a similar HIFI survey. A total of 92 isotopologues were\nidentified, arising from 49 different molecules, ranging from free ions to\ncomplex organic compounds and originating from a variety of environments from\nthe cold envelope to hot and dense gas within the cores. Sulfur dioxide,\nmethanol, and water are the dominant contributors. For the ortho / para ratios\nwe find deviations from the high temperature values between 13 and 27 %. In\ntotal 14 % of all lines remain unidentified.\n"}
{"abstract": "  This paper concerns the a priori generalization analysis of the Deep Ritz\nMethod (DRM) [W. E and B. Yu, 2017], a popular neural-network-based method for\nsolving high dimensional partial differential equations. We derive the\ngeneralization error bounds of two-layer neural networks in the framework of\nthe DRM for solving two prototype elliptic PDEs: Poisson equation and static\nSchr\\\"odinger equation on the $d$-dimensional unit hypercube. Specifically, we\nprove that the convergence rates of generalization errors are independent of\nthe dimension $d$, under the a priori assumption that the exact solutions of\nthe PDEs lie in a suitable low-complexity space called spectral Barron space.\nMoreover, we give sufficient conditions on the forcing term and the potential\nfunction which guarantee that the solutions are spectral Barron functions. We\nachieve this by developing a new solution theory for the PDEs on the spectral\nBarron space, which can be viewed as an analog of the classical Sobolev\nregularity theory for PDEs.\n"}
{"abstract": "  In this paper, we study the problem of fair sparse regression on a biased\ndataset where bias depends upon a hidden binary attribute. The presence of a\nhidden attribute adds an extra layer of complexity to the problem by combining\nsparse regression and clustering with unknown binary labels. The corresponding\noptimization problem is combinatorial, but we propose a novel relaxation of it\nas an \\emph{invex} optimization problem. To the best of our knowledge, this is\nthe first invex relaxation for a combinatorial problem. We show that the\ninclusion of the debiasing/fairness constraint in our model has no adverse\neffect on the performance. Rather, it enables the recovery of the hidden\nattribute. The support of our recovered regression parameter vector matches\nexactly with the true parameter vector. Moreover, we simultaneously solve the\nclustering problem by recovering the exact value of the hidden attribute for\neach sample. Our method uses carefully constructed primal dual witnesses to\nprovide theoretical guarantees for the combinatorial problem. To that end, we\nshow that the sample complexity of our method is logarithmic in terms of the\ndimension of the regression parameter vector.\n"}
{"abstract": "  Neutral atom arrays are promising for large-scale quantum computing\nespecially because it is possible to prepare large-scale qubit arrays. An\nunsolved issue is how to selectively excite one qubit deep in a 3D atomic array\nto Rydberg states. In this work, we show two methods for this purpose. The\nfirst method relies on a well-known result: in a dipole transition between two\nquantum states driven by two off-resonant fields of equal strength but opposite\ndetunings $\\pm\\Delta$, the transition is characterized by two counter-rotating\nRabi frequencies $\\Omega e^{\\pm i\\Delta t}$~[or $\\pm\\Omega e^{\\pm i\\Delta t}$\nif the two fields have a $\\pi$-phase difference]. This pair of detuned fields\nlead to a time-dependent Rabi frequency $2\\Omega \\cos(\\Delta t)$~[or $2i\\Omega\n\\sin(\\Delta t)$], so that a full transition between the two levels is\nrecovered. We show that when the two detuned fields are sent in different\ndirections, one atom in a 3D optical lattice can be selectively addressed for\nRydberg excitation, and when its state is restored, the state of any nontarget\natoms irradiated in the light path is also restored. Moreover, we find that the\nRydberg excitation by this method can significantly suppress the fundamental\nblockade error of a Rydberg gate, paving the way for a high-fidelity entangling\ngate with commonly used quasi-rectangular pulse that is easily obtained by\npulse pickers. Along the way, we find a second method for single-site Rydberg\naddressing in 3D, where a selected target atom can be excited to Rydberg state\nwhile preserving the state of any nontarget atom due to a spin echo sequence.\nThe capability to selectively address a target atom in 3D atomic arrays for\nRydberg excitation makes it possible to design large-scale neutral-atom\ninformation processor based on Rydberg blockade.\n"}
{"abstract": "  The $r$-index (Gagie et al., JACM 2020) represented a breakthrough in\ncompressed indexing of repetitive text collections, outperforming its\nalternatives by orders of magnitude. Its space usage, $\\mathcal{O}(r)$ where\n$r$ is the number of runs in the Burrows-Wheeler Transform of the text, is\nhowever larger than Lempel-Ziv and grammar-based indexes, and makes it\nuninteresting in various real-life scenarios of milder repetitiveness. In this\npaper we introduce the $sr$-index, a variant that limits the space to\n$\\mathcal{O}(\\min(r,n/s))$ for a text of length $n$ and a given parameter $s$,\nat the expense of multiplying by $s$ the time per occurrence reported. The\n$sr$-index is obtained by carefully subsampling the text positions indexed by\nthe $r$-index, in a way that we prove is still able to support pattern matching\nwith guaranteed performance. Our experiments demonstrate that the $sr$-index\nsharply outperforms virtually every other compressed index on repetitive texts,\nboth in time and space, even matching the performance of the $r$-index while\nusing 1.5--3.0 times less space. Only some Lempel-Ziv-based indexes achieve\nbetter compression than the $sr$-index, using about half the space, but they\nare an order of magnitude slower.\n"}
{"abstract": "  We present a subsampling strategy for the offline stage of the Reduced Basis\nMethod. The approach is aimed at bringing down the considerable offline costs\nassociated with using a finely-sampled training set. The proposed algorithm\nexploits the potential of the pivoted QR decomposition and the discrete\nempirical interpolation method to identify important parameter samples. It\nconsists of two stages. In the first stage, we construct a low-fidelity\napproximation to the solution manifold over a fine training set. Then, for the\navailable low-fidelity snapshots of the output variable, we apply the pivoted\nQR decomposition or the discrete empirical interpolation method to identify a\nset of sparse sampling locations in the parameter domain. These points reveal\nthe structure of the parametric dependence of the output variable. The second\nstage proceeds with a subsampled training set containing a by far smaller\nnumber of parameters than the initial training set. Different subsampling\nstrategies inspired from recent variants of the empirical interpolation method\nare also considered. Tests on benchmark examples justify the new approach and\nshow its potential to substantially speed up the offline stage of the Reduced\nBasis Method, while generating reliable reduced-order models.\n"}
{"abstract": "  We address the detection of material defects, which are inside a layered\nmaterial structure using compressive sensing based multiple-input and\nmultiple-output (MIMO) wireless radar. Here, the strong clutter due to the\nreflection of the layered structure's surface often makes the detection of the\ndefects challenging. Thus, sophisticated signal separation methods are required\nfor improved defect detection. In many scenarios, the number of defects that we\nare interested in is limited and the signaling response of the layered\nstructure can be modeled as a low-rank structure. Therefore, we propose joint\nrank and sparsity minimization for defect detection. In particular, we propose\na non-convex approach based on the iteratively reweighted nuclear and\n$\\ell_1-$norm (a double-reweighted approach) to obtain a higher accuracy\ncompared to the conventional nuclear norm and $\\ell_1-$norm minimization. To\nthis end, an iterative algorithm is designed to estimate the low-rank and\nsparse contributions. Further, we propose deep learning to learn the parameters\nof the algorithm (i.e., algorithm unfolding) to improve the accuracy and the\nspeed of convergence of the algorithm. Our numerical results show that the\nproposed approach outperforms the conventional approaches in terms of mean\nsquare errors of the recovered low-rank and sparse components and the speed of\nconvergence.\n"}
{"abstract": "  A two-dimensional string is simply a two-dimensional array. We continue the\nstudy of the combinatorial properties of repetitions in such strings over the\nbinary alphabet, namely the number of distinct tandems, distinct quartics, and\nruns. First, we construct an infinite family of $n\\times n$ 2D strings with\n$\\Omega(n^{3})$ distinct tandems. Second, we construct an infinite family of\n$n\\times n$ 2D strings with $\\Omega(n^{2}\\log n)$ distinct quartics. Third, we\nconstruct an infinite family of $n\\times n$ 2D strings with $\\Omega(n^{2}\\log\nn)$ runs. This resolves an open question of Charalampopoulos, Radoszewski,\nRytter, Wale\\'n, and Zuba [ESA 2020], who asked if the number of distinct\nquartics and runs in an $n\\times n$ 2D string is $\\mathcal{O}(n^{2})$.\n"}
{"abstract": "  Bit depth adaptation, where the bit depth of a video sequence is reduced\nbefore transmission and up-sampled during display, can potentially reduce data\nrates with limited impact on perceptual quality. In this context, we conducted\na subjective study on a UHD video database, BVI-BD, to explore the relationship\nbetween bit depth and visual quality. In this work, three bit depth adaptation\nmethods are investigated, including linear scaling, error diffusion, and a\nnovel adaptive Gaussian filtering approach. The results from a subjective\nexperiment indicate that above a critical bit depth, bit depth adaptation has\nno significant impact on perceptual quality, while reducing the amount\ninformation that is required to be transmitted. Below the critical bit depth,\nadvanced adaptation methods can be used to retain `good' visual quality (on\naverage) down to around 2 bits per color channel for the outlined experimental\nsetup - a large reduction compared to the typically used 8 bits per color\nchannel. A selection of image quality metrics were subsequently bench-marked on\nthe subjective data, and analysis indicates that a bespoke quality metric is\nrequired for bit depth adaptation.\n"}
{"abstract": "  Trialities of $\\mathcal{W}$-algebras are certain nontrivial isomorphisms\nbetween the affine cosets of three different $\\mathcal{W}$-(super)algebras, and\nwere first conjectured in the physics literature by Gaiotto and Rap\\v{c}\\'ak.\nIn this paper we prove trialities among eight families of\n$\\mathcal{W}$-(super)algebras of types $B$, $C$, and $D$. The key idea is to\nidentify the affine cosets of these algebras with one-parameter quotients of\nthe universal two-parameter even spin $\\mathcal{W}_{\\infty}$-algebra which was\nrecently constructed by Kanade and the second author. Our result is a vast\ngeneralization of both Feigin-Frenkel duality in types $B$, $C$, and $D$, and\nthe coset realization of principal $\\mathcal{W}$-algebras of type $D$ due to\nArakawa and us. It also provides a new coset realization of principal\n$\\mathcal{W}$-algebras of types $B$ and $C$. As an application, we prove the\nrationality of the affine vertex superalgebra $L_k(\\mathfrak{osp}_{1|2n})$, the\nminimal $\\mathcal{W}$-algebra $\\mathcal{W}_{k-1/2}(\\mathfrak{sp}_{2n+2},\nf_{\\text{min}})$, and the coset $\\text{Com}(L_k(\\mathfrak{sp}_{2m}),\nL_k(\\mathfrak{sp}_{2n}))$, for all integers $k,n,m \\geq 1$ with $m<n$. We also\nprove the rationality of some families of principal $\\mathcal{W}$-superalgebras\nof $\\mathfrak{osp}_{1|2n}$ and $\\mathfrak{osp}_{2|2n}$, and subregular\n$\\mathcal{W}$-algebras of $\\mathfrak{so}_{2n+1}$\n"}
{"abstract": "  Algorithmic stock trading has become a staple in today's financial market,\nthe majority of trades being now fully automated. Deep Reinforcement Learning\n(DRL) agents proved to be to a force to be reckon with in many complex games\nlike Chess and Go. We can look at the stock market historical price series and\nmovements as a complex imperfect information environment in which we try to\nmaximize return - profit and minimize risk. This paper reviews the progress\nmade so far with deep reinforcement learning in the subdomain of AI in finance,\nmore precisely, automated low-frequency quantitative stock trading. Many of the\nreviewed studies had only proof-of-concept ideals with experiments conducted in\nunrealistic settings and no real-time trading applications. For the majority of\nthe works, despite all showing statistically significant improvements in\nperformance compared to established baseline strategies, no decent\nprofitability level was obtained. Furthermore, there is a lack of experimental\ntesting in real-time, online trading platforms and a lack of meaningful\ncomparisons between agents built on different types of DRL or human traders. We\nconclude that DRL in stock trading has showed huge applicability potential\nrivalling professional traders under strong assumptions, but the research is\nstill in the very early stages of development.\n"}
{"abstract": "  The fractional quantum Hall effect (FQHE) is extensively studied, but the\nexplanation for Hall plateau widths and excitation energy gaps remains elusive.\nWe study the effective theory of FQHE built upon experimental inputs of Hall\ncurrent distribution, edge dynamics, and many-body correlations. We argue that\ncorrelated composites of integer spin, comprising electrons and their images,\nlocalized at the edge of the incompressible strip are the basic transport\nentity. We show in the lowest Landau level that Zeeman interactions of these\ncomposites produce all odd denominator plateaus and effective fractional\ncharges. Utilizing field-dependent chemical potential and effective g-factor,\nwe fully explain the observed Hall resistivity curve and excitation energy gaps\nof the half-filling family. The plateau heights are systematically generated by\nmulti-particle correlations, whereas the plateau widths and excitation energy\ngaps are determined by the correlation strengths. We explicitly show that the\nDrude-like behavior at half-filling follows from equal strength of\nmulti-particle correlations.\n"}
{"abstract": "  The cross-integration of spin-wave and superconducting technologies is a\npromising method for creating novel hybrid devices for future information\nprocessing technologies to store, manipulate, or convert data in both classical\nand quantum regimes. Hybrid magnon-polariton systems have been widely studied\nusing bulk Yttrium Iron Garnet (Y$_{3}$Fe$_{5}$O$_{12}$, YIG) and\nthree-dimensional microwave photon cavities. However, limitations in YIG growth\nhave thus far prevented its incorporation into CMOS compatible technology such\nas high quality factor superconducting quantum technology. To overcome this\nimpediment, we have used Plasma Focused Ion Beam (PFIB) technology -- taking\nadvantage of precision placement down to the micron-scale -- to integrate YIG\nwith superconducting microwave devices. Ferromagnetic resonance has been\nmeasured at millikelvin temperatures on PFIB-processed YIG samples using planar\nmicrowave circuits. Furthermore, we demonstrate strong coupling between\nsuperconducting resonator and YIG ferromagnetic resonance modes by maintaining\nreasonably low loss while reducing the system down to the micron scale. This\nachievement of strong coupling on-chip is a crucial step toward fabrication of\nfunctional hybrid quantum devices that advantage from spin-wave and\nsuperconducting components.\n"}
{"abstract": "  Vehicle Re-identification (re-id) over surveillance camera network with\nnon-overlapping field of view is an exciting and challenging task in\nintelligent transportation systems (ITS). Due to its versatile applicability in\nmetropolitan cities, it gained significant attention. Vehicle re-id matches\ntargeted vehicle over non-overlapping views in multiple camera network.\nHowever, it becomes more difficult due to inter-class similarity, intra-class\nvariability, viewpoint changes, and spatio-temporal uncertainty. In order to\ndraw a detailed picture of vehicle re-id research, this paper gives a\ncomprehensive description of the various vehicle re-id technologies,\napplicability, datasets, and a brief comparison of different methodologies. Our\npaper specifically focuses on vision-based vehicle re-id approaches, including\nvehicle appearance, license plate, and spatio-temporal characteristics. In\naddition, we explore the main challenges as well as a variety of applications\nin different domains. Lastly, a detailed comparison of current state-of-the-art\nmethods performances over VeRi-776 and VehicleID datasets is summarized with\nfuture directions. We aim to facilitate future research by reviewing the work\nbeing done on vehicle re-id till to date.\n"}
{"abstract": "  More than 2 billion mobile users worldwide type in multiple languages in the\nsoft keyboard. On a monolingual keyboard, 38% of falsely auto-corrected words\nare valid in another language. This can be easily avoided by detecting the\nlanguage of typed words and then validating it in its respective language.\nLanguage detection is a well-known problem in natural language processing. In\nthis paper, we present a fast, light-weight and accurate Language Detection\nEngine (LDE) for multilingual typing that dynamically adapts to user intended\nlanguage in real-time. We propose a novel approach where the fusion of\ncharacter N-gram model and logistic regression based selector model is used to\nidentify the language. Additionally, we present a unique method of reducing the\ninference time significantly by parameter reduction technique. We also discuss\nvarious optimizations fabricated across LDE to resolve ambiguity in input text\namong the languages with the same character pattern. Our method demonstrates an\naverage accuracy of 94.5% for Indian languages in Latin script and that of 98%\nfor European languages on the code-switched data. This model outperforms\nfastText by 60.39% and ML-Kit by 23.67% in F1 score for European languages. LDE\nis faster on mobile device with an average inference time of 25.91\nmicroseconds.\n"}
{"abstract": "  Collisions between interstellar gas clouds are potentially an important\nmechanism for triggering star formation. This is because they are able to\nrapidly generate large masses of dense gas. Observationally, cloud collisions\nare often identified in position-velocity (PV) space through bridging features\nbetween intensity peaks, usually of CO emission. Using a combination of\nhydrodynamical simulations, time-dependent chemistry, and radiative transfer,\nwe produce synthetic molecular line observations of overlapping clouds that are\ngenuinely colliding, and overlapping clouds that are just chance\nsuperpositions. Molecules tracing denser material than CO, such as NH$_3$ and\nHCN, reach peak intensity ratios of $0.5$ and $0.2$ with respect to CO in the\n`bridging feature' region of PV space for genuinely colliding clouds. For\noverlapping clouds that are just chance superpositions, the peak NH$_3$ and HCN\nintensities are co-located with the CO intensity peaks. This represents a way\nof confirming cloud collisions observationally, and distinguishing them from\nchance alignments of unrelated material.\n"}
{"abstract": "  Decay rates in quantum field theory (QFT) are typically calculated assuming\nthe particles are represented by momentum eigenstates (i.e. plane waves).\nHowever, strictly speaking, localized free particles should be represented by\nwave packets. This yields width corrections to the decay rate and to the clock\nbehaviour under Lorentz boosts. We calculate the decay rate of a particle of\nmass $M$ modeled as a Gaussian wavepacket of width $a$ and centered at zero\nmomentum. We find the decay rate to be $\\Gamma_0 \\big[1- \\frac{3 a^2}{4 M^2}\n+\\mathcal{O}\\big(\\tfrac{a^4}{M^4}\\big)\\big]$ where $\\Gamma_0$ is the decay rate\nof the particle at rest treated as a plane wave. The leading correction is then\nof order $\\tfrac{a^2}{M^2}$. We then perform a Lorentz boost of velocity $v$ on\nthe above Gaussian and find that its decay rate does not decrease\n\\textit{exactly} by the Lorentz factor $\\sqrt{1-v^2}$. There is a correction of\norder $\\tfrac{a^2v^2}{M^2}$. Therefore, the decaying wave packet does not act\nexactly like a typical clock under Lorentz boosts and we refer to it is a \"WP\nclock\" (wave packet clock). A WP clock does not move with a single velocity\nrelative to an observer but has a spread in velocities (more specifically, a\nspread in momenta). Nonetheless, it is best viewed as a single clock as the\nwave packet represents a one-particle state in QFT. WP clocks do not violate\nLorentz symmetry and are not based on new physics: they are a consequence of\nthe combined requirements of special relativity, quantum mechanics and\n\\textit{localized} free particles.\n"}
{"abstract": "  While the longitudinal field that dominates photospheric network regions has\nbeen studied extensively, small scale transverse fields have recently been\nfound to be ubiquitous in the quiet internetwork photosphere. Few observations\nhave captured how this field evolves. We aim to statistically characterise the\nmagnetic properties and observe the temporal evolution of small scale magnetic\nfeatures. We present two high spatial/temporal resolution observations that\nreveal the dynamics of two disk centre internetwork regions taken by the new\nGRIS/IFU (GREGOR Infrared Spectrograph Integral Field Unit) with the highly\nmagnetically sensitive Fe I line pair at 15648.52 {\\AA} and 15652.87 {\\AA}.\nWith the SIR code, we consider two inversion schemes: scheme 1 (S1), where a\nmagnetic atmosphere is embedded in a field free medium, and scheme 2 (S2), with\ntwo magnetic models and a fixed stray light component. S1 inversions returned a\nmedian magnetic field strength of 200 and 240 G for the two datasets,\nrespectively. We consider the median transverse (horizontal) component, among\npixels with Stokes Q or U, and the median unsigned longitudinal (vertical)\ncomponent, among pixels with Stokes V, above a noise threshold. We determined\nthe former to be 263 G and 267 G, and the latter to be 131 G and 145 G, for the\ntwo datasets, respectively. We present three regions of interest (ROIs),\ntracking the dynamics of small scale magnetic features. We apply S1 and S2\ninversions to specific profiles, and find S2 produces better approximations\nwhen there is evidence of mixed polarities. We find patches of linear\npolarization with magnetic flux density between 130 and 150 G, appearing\npreferentially at granule/intergranular lane (IGL) boundaries. The weak hG\nmagnetic field appears to be organised in terms of complex loop structures,\nwith transverse fields often flanked by opposite polarity longitudinal fields.\n"}
{"abstract": "  We consider the following variant of contextual linear bandits motivated by\nrouting applications in navigational engines and recommendation systems. We\nwish to learn a hidden $d$-dimensional value $w^*$. Every round, we are\npresented with a subset $\\mathcal{X}_t \\subseteq \\mathbb{R}^d$ of possible\nactions. If we choose (i.e. recommend to the user) action $x_t$, we obtain\nutility $\\langle x_t, w^* \\rangle$ but only learn the identity of the best\naction $\\arg\\max_{x \\in \\mathcal{X}_t} \\langle x, w^* \\rangle$. We design\nalgorithms for this problem which achieve regret $O(d\\log T)$ and $\\exp(O(d\n\\log d))$. To accomplish this, we design novel cutting-plane algorithms with\nlow \"regret\" -- the total distance between the true point $w^*$ and the\nhyperplanes the separation oracle returns. We also consider the variant where\nwe are allowed to provide a list of several recommendations. In this variant,\nwe give an algorithm with $O(d^2 \\log d)$ regret and list size\n$\\mathrm{poly}(d)$. Finally, we construct nearly tight algorithms for a weaker\nvariant of this problem where the learner only learns the identity of an action\nthat is better than the recommendation. Our results rely on new algorithmic\ntechniques in convex geometry (including a variant of Steiner's formula for the\ncentroid of a convex set) which may be of independent interest.\n"}
{"abstract": "  We present a catalogue of white dwarf candidates selected from Gaia early\ndata release three (EDR3). We applied several selection criteria in absolute\nmagnitude, colour, and Gaia quality flags to remove objects with unreliable\nmeasurements while preserving most stars compatible with the white dwarf locus\nin the Hertzsprung-Russell diagram. We then used a sample of over 30 000\nspectroscopically confirmed white dwarfs and contaminants from the Sloan\nDigital Sky Survey (SDSS) to map the distribution of these objects in the Gaia\nabsolute magnitude-colour space. Finally, we adopt the same method presented in\nour previous Gaia DR2 work to calculate a probability of being a white dwarf\n(Pwd) for $\\simeq$1.3 million sources which passed our quality selection. The\nPwd values can be used to select a sample of $\\simeq$359 000 high-confidence\nwhite dwarf candidates in the magnitude range 8< G <21. We calculated stellar\nparameters (effective temperature, surface gravity, and mass) for all these\nstars by fitting Gaia astrometry and photometry with synthetic models. We\nestimate an upper limit of 93 per cent for the overall completeness of our\ncatalogue for white dwarfs with G $\\leq$20 mag and effective temperature\n(Teff)>7000K, at high Galactic latitudes (|b|>20{\\deg}). Alongside the main\ncatalogue we include a reduced-proper-motion extension containing $\\simeq$10\n200 white dwarf candidates with unreliable parallax measurements which could,\nhowever be identified on the basis of their proper motion. We also performed a\ncross-match of our catalogues with SDSS DR16 spectroscopy and provide spectral\nclassification based on visual inspection for all resulting matches.\n"}
{"abstract": "  Zero-shot cross-domain dialogue state tracking (DST) enables us to handle\ntask-oriented dialogue in unseen domains without the expense of collecting\nin-domain data. In this paper, we propose a slot description enhanced\ngenerative approach for zero-shot cross-domain DST. Specifically, our model\nfirst encodes dialogue context and slots with a pre-trained self-attentive\nencoder, and generates slot values in an auto-regressive manner. In addition,\nwe incorporate Slot Type Informed Descriptions that capture the shared\ninformation across slots to facilitate cross-domain knowledge transfer.\nExperimental results on the MultiWOZ dataset show that our proposed method\nsignificantly improves existing state-of-the-art results in the zero-shot\ncross-domain setting.\n"}
{"abstract": "  Let $G$ be a connected semisimple real Lie group with finite center, and\n$\\mu$ a probability measure on $G$ whose support generates a Zariski-dense\nsubgroup of $G$. We consider the right $\\mu$-random walk on $G$ and show that\neach random trajectory spends most of its time at bounded distance of a\nwell-chosen Weyl chamber. We infer that if $G$ has rank one, and $\\mu$ has a\nfinite first moment, then for any discrete subgroup $\\Lambda \\subseteq G$, the\n$\\mu$-walk and the geodesic flow on $\\Lambda \\backslash G$ are either both\ntransient, or both recurrent and ergodic, thus extending a well known theorem\ndue to Hopf-Tsuji-Sullivan-Kaimanovich dealing with the Brownian motion.\n"}
{"abstract": "  Brillouin scattering has been widely exploited for advanced photonics\nfunctionalities such as microwave photonics, signal processing, sensing,\nlasing, and more recently in micro- and nano-photonic waveguides. So far, all\nthe works have focused on the opto-acoustic interaction driven from the core\nregion of micro- and nano-waveguides. Here we observe, for the first time, an\nefficient Brillouin scattering generated by an evanescent field nearby a\nsub-wavelength waveguide embedded in a pressurised gas cell, with a maximum\ngain coefficient of $18.90 \\pm 0.17$ m$^{-1}$W$^{-1}$. This gain is 11 times\nlarger than the highest Brillouin gain obtained in a hollow-core fibre and 79\ntimes larger than in a standard single-mode fibre. The realisation of strong\nfree-space Brillouin scattering from a waveguide benefits from the flexibility\nof confined light while providing a direct access to the opto-acoustic\ninteraction, as required in free-space optoacoustics such as Brillouin\nspectroscopy and microscopy. Therefore, our work creates an important bridge\nbetween Brillouin scattering in waveguides, Brillouin spectroscopy and\nmicroscopy, and opens new avenues in light-sound interactions, optomechanics,\nsensing, lasing and imaging.\n"}
{"abstract": "  The primate heteromodal cortex presents an evident functional modularity at a\nmesoscopic level, with physiological and anatomical evidence pointing to it as\nlikely substrate of long-term memory. In order to investigate some of its\nproperties, a model of multimodular autoassociator is studied. Each of the many\nmodules represents a neocortical functional ensemble of recurrently connected\nneurons and operates as a Hebbian autoassociator, storing a number of local\nfeatures which it can recall upon cue. The global memory patterns are made of\ncombinations of features sparsely distributed across the modules. Intermodular\nconnections are modelled as a finite-connectivity random graph. Any pair of\nfeatures in any respective pair of modules is allowed to be involved in several\nmemory patterns; the coarse-grained modular network dynamics is defined in such\na way as to overcome the consequent ambiguity of associations. Effects of\nlong-range homeostatic synaptic scaling on network performance are also\nassessed.\n  The dynamical process of cued retrieval almost saturates a natural upper\nbound while producing negligible spurious activation. The extent of finite-size\neffects on storage capacity is quantitatively evaluated. In the limit of\ninfinite size, the functional relationship between storage capacity and number\nof features per module reduces to that which other authors found by methods\nfrom equilibrium statistical mechanics, which suggests that the origin of the\nfunctional form is of a combinatorial nature. In contrast with its apparent\ninevitability at intramodular level, long-range synaptic scaling results to be\nof minor relevance to both retrieval and storage capacity, casting doubt on its\nexistence in the neocortex. A conjecture is also posited about how statistical\nfluctuation of connectivity across the network may underpin spontaneous\nemergence of semantic hierarchies through learning.\n"}
{"abstract": "  We consider the problem of bi-directional controlled quantum teleportation of\ninformation encoded in phase opposite coherent state among two distant partners\nAlice and Bob, with the consent of controller, Charlie. We use five-mode\ncluster-type entangled coherent state as the quantum resource to achieve this\ntask. The scheme uses linear optical devices such as beam splitter, phase\nshifters, and photon counters. We have shown that for moderately large coherent\namplitude, near-perfect bi-directional controlled teleportation can be obtained\nin terms of the average fidelity of teleportation.\n"}
{"abstract": "  Deep learning approaches may help radiologists in the early diagnosis and\ntimely treatment of cerebrovascular diseases. Accurate cerebral vessel\nsegmentation of Time-of-Flight Magnetic Resonance Angiographs (TOF-MRAs) is an\nessential step in this process. This study investigates deep learning\napproaches for automatic, fast and accurate cerebrovascular segmentation for\nTOF-MRAs. The performance of several data augmentation and selection methods\nfor training a 2D and 3D U-Net for vessel segmentation was investigated in five\nexperiments: a) without augmentation, b) Gaussian blur, c) rotation and\nflipping, d) Gaussian blur, rotation and flipping and e) different input patch\nsizes. All experiments were performed by patch-training both a 2D and 3D U-Net\nand predicted on a test set of MRAs. Ground truth was manually defined using an\ninteractive threshold and region growing method. The performance was evaluated\nusing the Dice Similarity Coefficient (DSC), Modified Hausdorff Distance and\nVolumetric Similarity, between the predicted images and the interactively\ndefined ground truth. The segmentation performance of all trained networks on\nthe test set was found to be good, with DSC scores ranging from 0.72 to 0.83.\nBoth the 2D and 3D U-Net had the best segmentation performance with Gaussian\nblur, rotation and flipping compared to other experiments without augmentation\nor only one of those augmentation techniques. Additionally, training on larger\npatches or slices gave optimal segmentation results. In conclusion, vessel\nsegmentation can be optimally performed on TOF-MRAs using a trained 3D U-Net on\nlarger patches, where data augmentation including Gaussian blur, rotation and\nflipping was performed on the training data.\n"}
{"abstract": "  Traffic monitoring cameras are powerful tools for traffic management and\nessential components of intelligent road infrastructure systems. In this paper,\nwe present a vehicle localization and traffic scene reconstruction framework\nusing these cameras, dubbed as CAROM, i.e., \"CARs On the Map\". CAROM processes\ntraffic monitoring videos and converts them to anonymous data structures of\nvehicle type, 3D shape, position, and velocity for traffic scene reconstruction\nand replay. Through collaborating with a local department of transportation in\nthe United States, we constructed a benchmarking dataset containing GPS data,\nroadside camera videos, and drone videos to validate the vehicle tracking\nresults. On average, the localization error is approximately 0.8 m and 1.7 m\nwithin the range of 50 m and 120 m from the cameras, respectively.\n"}
{"abstract": "  As technology becomes more widely available, millions of users worldwide have\ninstalled some form of smart device in their homes or workplaces. These devices\nare often off-the-shelf commodity systems, such as Google Home or Samsung\nSmartThings, that are installed by end-users looking to automate a small\ndeployment. In contrast to these \"plug-and-play\" systems, purpose-built\nEnterprise Internet-of-Things (E-IoT) systems such as Crestron, Control4, RTI,\nSavant offer a smart solution for more sophisticated applications (e.g.,\ncomplete lighting control, A/V management, security). In contrast to commodity\nsystems, E-IoT systems are usually closed source, costly, require certified\ninstallers, and are overall more robust for their use cases. Due to this, E-IoT\nsystems are often found in expensive smart homes, government and academic\nconference rooms, yachts, and smart private offices. However, while there has\nbeen plenty of research on the topic of commodity systems, no current study\nexists that provides a complete picture of E-IoT systems, their components, and\nrelevant threats. As such, lack of knowledge of E-IoT system threats, coupled\nwith the cost of E-IoT systems has led many to assume that E-IoT systems are\nsecure. To address this research gap, raise awareness on E-IoT security, and\nmotivate further research, this work emphasizes E-IoT system components, E-IoT\nvulnerabilities, solutions, and their security implications. In order to\nsystematically analyze the security of E-IoT systems, we divide E-IoT systems\ninto four layers: E-IoT Devices Layer, Communications Layer, Monitoring and\nApplications Layer, and Business Layer. We survey attacks and defense\nmechanisms, considering the E-IoT components at each layer and the associated\nthreats. In addition, we present key observations in state-of-the-art E-IoT\nsecurity and provide a list of open research problems that need further\nresearch.\n"}
{"abstract": "  We show that the bicategory of finite groupoids and right-free permutation\nbimodules is a quotient of the bicategory of Mackey 2-motives introduced in\narXiv:1808.04902, obtained by modding out the so-called cohomological\nrelations. This categorifies Yoshida's Theorem for ordinary cohomological\nMackey functors, and provides a direct connection between Mackey 2-motives and\nthe usual blocks of representation theory.\n"}
{"abstract": "  In this paper, we initiate the study of wave propagation in a recently\nproposed mathematical model for stretch-limited elastic strings. We consider\nthe longitudinal motion of a simple class of uniform, semi-infinite,\nstretch-limited strings under no external force with finite end held fixed and\nprescribed tension at the infinite end. We study a class of motions such that\nthe string has one inextensible segment, where the local stretch is maximized,\nand one extensible segment. The equations of motion reduce to a simple and\nnovel shock front problem in one spatial variable for which we prove existence\nand uniqueness of local-in-time solutions for appropriate initial data. We then\nprove the orbital asymptotic stability of an explicit two-parameter family of\npiece-wise constant stretched motions. If the prescribed tension at the\ninfinite end is increasing in time, our results provide an open set of initial\ndata launching motions resulting in the string becoming fully inextensible and\ntension blowing up in finite time.\n"}
{"abstract": "  In this paper, we study the spontaneous scalarization of an extended,\nself-gravitating system which is static, cylindrically symmetric and possesses\nelectromagnetic fields. We demonstrate that a real massive scalar field\ncondenses on this Melvin magnetic universe solution when introducing a\nnon-minimal coupling between the scalar field and (a) the magnetic field and\n(b) the curvature of the space-time, respectively. We find that in both cases,\nthe solutions exist on a finite interval of the coupling constant and that\nsolutions with a number of nodes $k$ in the scalar field exist. For case (a) we\nobserve that the intervals of existence are mutually exclusive for different\n$k$.\n"}
{"abstract": "  Self-supervised ASR-TTS models suffer in out-of-domain data conditions. Here\nwe propose an enhanced ASR-TTS (EAT) model that incorporates two main features:\n1) The ASR$\\rightarrow$TTS direction is equipped with a language model reward\nto penalize the ASR hypotheses before forwarding it to TTS. 2) In the\nTTS$\\rightarrow$ASR direction, a hyper-parameter is introduced to scale the\nattention context from synthesized speech before sending it to ASR to handle\nout-of-domain data. Training strategies and the effectiveness of the EAT model\nare explored under out-of-domain data conditions. The results show that EAT\nreduces the performance gap between supervised and self-supervised training\nsignificantly by absolute 2.6\\% and 2.7\\% on Librispeech and BABEL\nrespectively.\n"}
{"abstract": "  We present the Deception Detection and Physiological Monitoring (DDPM)\ndataset and initial baseline results on this dataset. Our application context\nis an interview scenario in which the interviewee attempts to deceive the\ninterviewer on selected responses. The interviewee is recorded in RGB,\nnear-infrared, and long-wave infrared, along with cardiac pulse, blood\noxygenation, and audio. After collection, data were annotated for\ninterviewer/interviewee, curated, ground-truthed, and organized into train /\ntest parts for a set of canonical deception detection experiments. Baseline\nexperiments found random accuracy for micro-expressions as an indicator of\ndeception, but that saccades can give a statistically significant response. We\nalso estimated subject heart rates from face videos (remotely) with a mean\nabsolute error as low as 3.16 bpm. The database contains almost 13 hours of\nrecordings of 70 subjects, and over 8 million visible-light, near-infrared, and\nthermal video frames, along with appropriate meta, audio and pulse oximeter\ndata. To our knowledge, this is the only collection offering recordings of five\nmodalities in an interview scenario that can be used in both deception\ndetection and remote photoplethysmography research.\n"}
{"abstract": "  Let $E$ be an elliptic curve defined over a number field $K$ and let $v$ be a\nfinite place of $K$. Write $K^{tv}$ the maximal extension of $K$ in which $v$\nis totally split and $L$ the field generated over $K^{tv}$ by all torsion\npoints of $E$. Under some conditions, we will show that the absolute\nlogarithmic Weil height (resp. N\\'eron-Tate height) of any element of $L^*$\n(resp. $E(L)$) is either $0$ or bounded from below by a positive constant\ndepending only on $E, K$ and $v$. This lower bound will be explicit in the\ntoric case when $K=\\mathbb{Q}$.\n"}
{"abstract": "  We consider integer programs (IP) defined by equations and box constraints,\nwhere it is known that determinants in the constraint matrix are one measure of\ncomplexity. For example, Artmann et al. showed that an IP can be solved in\nstrongly polynomial time if the constraint matrix is bimodular, that is, the\ndeterminants are bounded in absolute value by two. Determinants are also used\nto bound the $\\ell_1$-distance between IP solutions and solutions of its linear\nrelaxation. One of the first works to quantify the complexity of IPs with\nbounded determinants was that of Heller, who identified the maximum number of\ndiffering columns in a totally unimodular constraint matrix. So far, each\nextension of Heller's bound to general determinants has been exponential in the\ndeterminants or the number of equations. We provide the first column bound that\nis polynomial in both values. As a corollary, we give the first\n$\\ell_1$-distance bound that is polynomial in the determinants and the number\nof equations. We also show a tight bound on the number of differing columns in\na bimodular constraint matrix; this is the first tight bound since Heller's\nresult. Our analysis reveals combinatorial properties of bimodular IPs that may\nbe of independent interest, in particular in recognition algorithms for IPs\nwith bounded determinants.\n"}
{"abstract": "  In 1939 Weyl showed that the volume of spherical tubes around compact\nsubmanifolds M of Euclidean space depends solely on the induced Riemannian\nmetric on M. Can this intrinsic nature of the tube volume be preserved for\ntubes with more general cross sections D than the round ball? Under\nsufficiently strong symmetry conditions on D the answer turns out to be yes.\n"}
{"abstract": "  A variety of new physics scenarios allow for neutrinos to up-scatter into a\nheavy neutral lepton state. For a range of couplings and neutrino energies, the\nheavy neutrino may travel some distance before decaying to visible final\nstates. When both the up-scattering and decay occur within the detector volume,\nthese \"double bang\" events produce distinctive phenomenology with very low\nbackground. In this work, we first consider the current sensitivity at\nSuper-Kamiokande via the atmospheric neutrino flux, and find current data may\nalready provide new constraints. We then examine projected future sensitivity\nat DUNE and Hyper-Kamiokande, including both atmospheric and beam flux\ncontributions to double-bang signals.\n"}
{"abstract": "  We propose a simple yet reliable bottom-up approach with a good trade-off\nbetween accuracy and efficiency for the problem of multi-person pose\nestimation. Given an image, we employ an Hourglass Network to infer all the\nkeypoints from different persons indiscriminately as well as the guiding\noffsets connecting the adjacent keypoints belonging to the same persons. Then,\nwe greedily group the candidate keypoints into multiple human poses (if any),\nutilizing the predicted guiding offsets. And we refer to this process as greedy\noffset-guided keypoint grouping (GOG). Moreover, we revisit the\nencoding-decoding method for the multi-person keypoint coordinates and reveal\nsome important facts affecting accuracy. Experiments have demonstrated the\nobvious performance improvements brought by the introduced components. Our\napproach is comparable to the state of the art on the challenging COCO dataset\nunder fair conditions. The source code and our pre-trained model are publicly\navailable online.\n"}
{"abstract": "  We present a system for localizing sound sources in a room with several\nad-hoc microphone arrays. Each circular array performs direction of arrival\n(DOA) estimation independently using commercial software. The DOAs are fed to a\nfusion center, concatenated, and used to perform the localization based on two\nproposed methods, which require only few labeled source locations (anchor\npoints) for training. The first proposed method is based on principal component\nanalysis (PCA) of the observed DOA and does not require any knowledge of anchor\npoints. The array cluster can then perform localization on a manifold defined\nby the PCA of concatenated DOAs over time. The second proposed method performs\nlocalization using an affine transformation between the DOA vectors and the\nroom manifold. The PCA has fewer requirements on the training sequence, but is\nless robust to missing DOAs from one of the arrays. The methods are\ndemonstrated with five IoT 8-microphone circular arrays, placed at unspecified\nfixed locations in an office. Both the PCA and the affine method can easily map\nout a rectangle based on a few anchor points with similar accuracy. The\nproposed methods provide a step towards monitoring activities in a smart home\nand require little installation effort as the array locations are not needed.\n"}
{"abstract": "  We consider $\\rm R$-matrix realization of the quantum deformations of the\nloop algebras $\\tilde{\\mathfrak{g}}$ corresponding to non-exceptional affine\nLie algebras of type $\\widehat{\\mathfrak{g}}=A^{(1)}_{N-1}$, $B^{(1)}_n$,\n$C^{(1)}_n$, $D^{(1)}_n$, $A^{(2)}_{N-1}$. For each $U_q(\\tilde{\\mathfrak{g}})$\nwe investigate the commutation relations between Gauss coordinates of the\nfundamental $\\mathbb{L}$-operators using embedding of the smaller algebra into\nbigger one. The new realization of these algebras in terms of the currents is\ngiven. The relations between all off-diagonal Gauss coordinates and certain\nprojections from the ordered products of the currents are presented. These\nrelations are important in applications to the quantum integrable models.\n"}
{"abstract": "  Quantum logic gates for photonic qubits can be implemented using the quantum\nZeno effect based on strong two-photon absorption. The fidelity of quantum Zeno\ngates of this kind may be substantially reduced by photon loss. Heralding on\nthose outcomes in which both of the logical qubits emerge from the Zeno gate\ncan increase the fidelity at the expense of a limited success rate (P.M. Leung\net al., Phys. Rev. A 74, 062325 (2006)). We analyze the performance of heralded\nquantum Zeno gates by solving Schrodinger's equation for a system of photons\ncoupled to three-level atoms. This approach identifies several potential error\nsources that are not described by earlier models that assumed a fixed rate of\nsingle-photon loss and two-photon absorption.\n"}
{"abstract": "  In this work we address the task of observing the performance of a semantic\nsegmentation deep neural network (DNN) during online operation, i.e., during\ninference, which is of high importance in safety-critical applications such as\nautonomous driving. Here, many high-level decisions rely on such DNNs, which\nare usually evaluated offline, while their performance in online operation\nremains unknown. To solve this problem, we propose an improved online\nperformance prediction scheme, building on a recently proposed concept of\npredicting the primary semantic segmentation task's performance. This can be\nachieved by evaluating the auxiliary task of monocular depth estimation with a\nmeasurement supplied by a LiDAR sensor and a subsequent regression to the\nsemantic segmentation performance. In particular, we propose (i) sequential\ntraining methods for both tasks in a multi-task training setup, (ii) to share\nthe encoder as well as parts of the decoder between both task's networks for\nimproved efficiency, and (iii) a temporal statistics aggregation method, which\nsignificantly reduces the performance prediction error at the cost of a small\nalgorithmic latency. Evaluation on the KITTI dataset shows that all three\naspects improve the performance prediction compared to previous approaches.\n"}
{"abstract": "  We develop a new approach to $L^{\\infty}$-a priori estimates for degenerate\ncomplex Monge-Amp\\`ere equations on complex manifolds. It only relies on\ncompactness and envelopes properties of quasi-plurisubharmonic functions. In a\nprequel \\cite{GL21a} we have shown how this method allows one to obtain new and\nefficient proofs of several fundamental results in K\\\"ahler geometry. In\n\\cite{GL21b} we have studied the behavior of Monge-Amp\\`ere volumes on\nhermitian manifolds. We extend here the techniques of \\cite{GL21a} to the\nhermitian setting and use the bounds established in \\cite{GL21b}, producing new\nrelative a priori estimates, as well as several existence results for\ndegenerate complex Monge-Amp\\`ere equations on compact hermitian manifolds.\n"}
{"abstract": "  Though deep neural network models exhibit outstanding performance for various\napplications, their large model size and extensive floating-point operations\nrender deployment on mobile computing platforms a major challenge, and, in\nparticular, on Internet of Things devices. One appealing solution is model\nquantization that reduces the model size and uses integer operations commonly\nsupported by microcontrollers . To this end, a 1-bit quantized DNN model or\ndeep binary neural network maximizes the memory efficiency, where each\nparameter in a BNN model has only 1-bit. In this paper, we propose a\nreconfigurable BNN (RBNN) to further amplify the memory efficiency for\nresource-constrained IoT devices. Generally, the RBNN can be reconfigured on\ndemand to achieve any one of M (M>1) distinct tasks with the same parameter\nset, thus only a single task determines the memory requirements. In other\nwords, the memory utilization is improved by times M. Our extensive experiments\ncorroborate that up to seven commonly used tasks can co-exist (the value of M\ncan be larger). These tasks with a varying number of classes have no or\nnegligible accuracy drop-off on three binarized popular DNN architectures\nincluding VGG, ResNet, and ReActNet. The tasks span across different domains,\ne.g., computer vision and audio domains validated herein, with the prerequisite\nthat the model architecture can serve those cross-domain tasks. To protect the\nintellectual property of an RBNN model, the reconfiguration can be controlled\nby both a user key and a device-unique root key generated by the intrinsic\nhardware fingerprint. By doing so, an RBNN model can only be used per paid user\nper authorized device, thus benefiting both the user and the model provider.\n"}
{"abstract": "  In this paper, we propose a Collaboration of Experts (CoE) framework to pool\ntogether the expertise of multiple networks towards a common aim. Each expert\nis an individual network with expertise on a unique portion of the dataset,\nwhich enhances the collective capacity. Given a sample, an expert is selected\nby the delegator, which simultaneously outputs a rough prediction to support\nearly termination. To fulfill this framework, we propose three modules to impel\neach model to play its role, namely weight generation module (WGM), label\ngeneration module (LGM) and variance calculation module (VCM). Our method\nachieves the state-of-the-art performance on ImageNet, 80.7% top-1 accuracy\nwith 194M FLOPs. Combined with PWLU activation function and CondConv, CoE\nfurther achieves the accuracy of 80.0% with only 100M FLOPs for the first time.\nMore importantly, our method is hardware friendly and achieves a 3-6x speedup\ncompared with some existing conditional computation approaches.\n"}
{"abstract": "  We consider excited states in a CFT, obtained by applying a weak unitary\nperturbation to the vacuum. The perturbation is generated by the integral of a\nlocal operator $J^{(n)}$ of modular weight $n$ over a spacelike surface passing\nthrough $x = 0$. For $\\vert n \\vert \\geq 2$ the modular Hamiltonian associated\nwith a division of space at $x = 0$ picks up an endpoint contribution,\nsensitive to the details of the perturbation (including the shape of the\nspacelike surface) at $x = 0$. The endpoint contribution is a sum of light-ray\nmoments of the perturbing operator $J^{(n)}$ and its descendants. For\nperturbations on null planes only moments of $J^{(n)}$ itself contribute.\n"}
{"abstract": "  The analysis of longitudinal travel data enables investigating how mobility\npatterns vary across the population and identify the spatial properties\nthereof. The objective of this study is to identify the extent to which users\nexplore different parts of the network as well as identify distinctive user\ngroups in terms of the spatial extent of their mobility patterns. To this end,\nwe propose two means for representing spatial mobility profiles and clustering\ntravellers accordingly. We represent users patterns in terms of zonal visiting\nfrequency profiles and grid-cells spatial extent heatmaps. We apply the\nproposed analysis to a large-scale multi-modal mobility data set from the\npublic transport system in Stockholm, Sweden. We unravel three clusters -\nlocals, commuters and explorers - that best describe the zonal visiting\nfrequency and show that their composition varies considerably across users'\nplace of residence and related demographics. We also identify 18 clusters of\nvisiting spatial extent which form four groups that follow similar shapes of\ntravel extent yet oriented in different directions. The approach proposed and\napplied in this study could be applied for any longitudinal individual travel\ndemand data.\n"}
{"abstract": "  We study the radio and optical properties of the brightest group galaxies\n(BGGs) in a sample of galaxy groups from the SDSS DR7. The luminosity\ndifference between the BGG and the second ranked galaxy in the group (known as\nthe luminosity, or magnitude, gap) has been used as a probe for the level of\ngalaxy interaction for the BGG within the group. We study the properties of\nBGGs with magnitude gaps in the range 0-2.7 magnitudes, in order to investigate\nany relation between luminosity gap and the radio properties of the BGG. In\norder to eliminate selection biases, we ensure that all variations in stellar\nmass are accounted for. We then confirm that, at fixed stellar mass, there are\nno significant variations in the optical properties of the BGGs over the full\nrange of luminosity gaps studied. We compare these optical results with the\nEAGLE hydrodynamical simulations and find broad consistency with the\nobservational data. Using EAGLE we also confirm that no trends begin to arise\nin the simulated data at luminosity gaps beyond our observational limits.\nFinally, we find that, at fixed stellar mass, the fraction of BGGs that are\nradio-loud also shows no trends as a function of luminosity gap. We examine how\nthe BGG offset from the center of group may affect the radio results and find\nno significant trend for the fraction of radio-loud BGGs with magnitude gap in\neither the BGG samples with greater or less than 100kpc offset from the center\nof group.\n"}
{"abstract": "  This paper studies M-estimators with gradient-Lipschitz loss function\nregularized with convex penalty in linear models with Gaussian design matrix\nand arbitrary noise distribution. A practical example is the robust M-estimator\nconstructed with the Huber loss and the Elastic-Net penalty and the noise\ndistribution has heavy-tails. Our main contributions are three-fold. (i) We\nprovide general formulae for the derivatives of regularized M-estimators\n$\\hat\\beta(y,X)$ where differentiation is taken with respect to both $y$ and\n$X$; this reveals a simple differentiability structure shared by all convex\nregularized M-estimators. (ii) Using these derivatives, we characterize the\ndistribution of the residual $r_i = y_i-x_i^\\top\\hat\\beta$ in the intermediate\nhigh-dimensional regime where dimension and sample size are of the same order.\n(iii) Motivated by the distribution of the residuals, we propose a novel\nadaptive criterion to select tuning parameters of regularized M-estimators. The\ncriterion approximates the out-of-sample error up to an additive constant\nindependent of the estimator, so that minimizing the criterion provides a proxy\nfor minimizing the out-of-sample error. The proposed adaptive criterion does\nnot require the knowledge of the noise distribution or of the covariance of the\ndesign. Simulated data confirms the theoretical findings, regarding both the\ndistribution of the residuals and the success of the criterion as a proxy of\nthe out-of-sample error. Finally our results reveal new relationships between\nthe derivatives of $\\hat\\beta(y,X)$ and the effective degrees of freedom of the\nM-estimator, which are of independent interest.\n"}
{"abstract": "  In this article, we get classification theorems for a Ricci soliton on the\npseudo-Riemannian hypersurface of the Minkowski space $\\mathbb{E}^4_1$ taking\nthe potential vector field as the tangent component of the position vector of\nthe pseudo-Riemannian hypersurface, denoted by $(M,g,{\\bf x}^T,\\lambda)$ in\nboth Riemannian and Lorentzian settings. First, we obtain the necessary and\nsufficient condition that a pseudo-Riemannian hypersurface $(M,g)$ in\n$\\mathbb{E}^4_1$ admits a Ricci soliton $(M,g,{\\bf x}^T,\\lambda)$. In each of\nthe form of the shape operator of a pseudo-Riemannian hypersurface, we obtain\ncharacterization a Ricci soliton on a pseudo-Riemannian hypersurface. More\nprecisely, we show that totally umbilical hypersurfaces, hyperbolic and a\npseudo-spherical cylinder in $\\mathbb{E}^4_1$ is a shrinking Ricci soliton\nwhose the potential vector field is the tangent part of the position vector.\nFurthermore, we conclude that there exists only a shrinking Ricci soliton on a\nLorentzian isoparametric hypersurface in $\\mathbb{E}^4_1$ with\nnondiagonalizable shape operator whose the minimal polynomial has double real\nroots.\n"}
{"abstract": "  Characteristics of Oscillating Flow in a Straight Tube with varying thickness\nof porous layer at the wall is studied. Results for two different porous layer\nis discussed and compared in terms of Velocity Profile, phase differences and\nradial position of maximum velocity over the cross-section. Present problem is\nmodeled using Navier-Stokes Equation in clear medium at the core of tube while\nDarcy-BrinkmannForcheimmer Model is used for the outer porous layer. Solutions\nfor these equations have been obtained using Finite Volume Method based 3D\nunstructured solver. Velocity profile obtained for tube with high permeability\nporous layer is the main highlight of results.\n"}
{"abstract": "  We report polymer composite films containing fillers comprised of\nquasi-one-dimensional (1D) van der Waals materials, specifically transition\nmetal trichalcogenides containing 1D structural motifs that enable their\nexfoliation into bundles of atomic threads. These nanostructures are\ncharacterized by extremely large aspect ratios of up to 10^6. The polymer\ncomposites with low loadings of quasi-1D TaSe3 fillers (below 3 vol. %)\nrevealed excellent electromagnetic interference shielding in the X-band GHz and\nEHF sub-THz frequency ranges, while remaining DC electrically insulating. The\nunique electromagnetic shielding characteristics of these films are attributed\nto effective coupling of the electromagnetic waves to the high-aspect-ratio\nelectrically-conductive TaSe3 atomic-thread bundles even when the filler\nconcentration is below the electrical percolation threshold. These novel films\nare promising for high-frequency communication technologies, which require\nelectromagnetic shielding films that are flexible, lightweight, corrosion\nresistant, electrically insulating and inexpensive.\n"}
{"abstract": "  We present Affect2MM, a learning method for time-series emotion prediction\nfor multimedia content. Our goal is to automatically capture the varying\nemotions depicted by characters in real-life human-centric situations and\nbehaviors. We use the ideas from emotion causation theories to computationally\nmodel and determine the emotional state evoked in clips of movies. Affect2MM\nexplicitly models the temporal causality using attention-based methods and\nGranger causality. We use a variety of components like facial features of\nactors involved, scene understanding, visual aesthetics, action/situation\ndescription, and movie script to obtain an affective-rich representation to\nunderstand and perceive the scene. We use an LSTM-based learning model for\nemotion perception. To evaluate our method, we analyze and compare our\nperformance on three datasets, SENDv1, MovieGraphs, and the LIRIS-ACCEDE\ndataset, and observe an average of 10-15% increase in the performance over SOTA\nmethods for all three datasets.\n"}
{"abstract": "  With increasingly more data and computation involved in their training,\nmachine learning models constitute valuable intellectual property. This has\nspurred interest in model stealing, which is made more practical by advances in\nlearning with partial, little, or no supervision. Existing defenses focus on\ninserting unique watermarks in a model's decision surface, but this is\ninsufficient: the watermarks are not sampled from the training distribution and\nthus are not always preserved during model stealing. In this paper, we make the\nkey observation that knowledge contained in the stolen model's training set is\nwhat is common to all stolen copies. The adversary's goal, irrespective of the\nattack employed, is always to extract this knowledge or its by-products. This\ngives the original model's owner a strong advantage over the adversary: model\nowners have access to the original training data. We thus introduce $dataset$\n$inference$, the process of identifying whether a suspected model copy has\nprivate knowledge from the original model's dataset, as a defense against model\nstealing. We develop an approach for dataset inference that combines\nstatistical testing with the ability to estimate the distance of multiple data\npoints to the decision boundary. Our experiments on CIFAR10, SVHN, CIFAR100 and\nImageNet show that model owners can claim with confidence greater than 99% that\ntheir model (or dataset as a matter of fact) was stolen, despite only exposing\n50 of the stolen model's training points. Dataset inference defends against\nstate-of-the-art attacks even when the adversary is adaptive. Unlike prior\nwork, it does not require retraining or overfitting the defended model.\n"}
{"abstract": "  Operations in many essential industries including finance and banking are\noften characterized by the need to perform repetitive sequential tasks. Despite\ntheir criticality to the business, workflows are rarely fully automated or even\nformally specified, though there may exist a number of natural language\ndocuments describing these procedures for the employees of the company. Plan\nextraction methods provide us with the possibility of extracting structure\nplans from such natural language descriptions of the plans/workflows, which\ncould then be leveraged by an automated system. In this paper, we investigate\nthe utility of generalized language models in performing such extractions\ndirectly from such texts. Such models have already been shown to be quite\neffective in multiple translation tasks, and our initial results seem to point\nto their effectiveness also in the context of plan extractions. Particularly,\nwe show that GPT-3 is able to generate plan extraction results that are\ncomparable to many of the current state of the art plan extraction methods.\n"}
{"abstract": "  We consider a class of time-fractional phase field models including the\nAllen-Cahn and Cahn-Hilliard equations. We establish several weighted\npositivity results for functionals driven by the Caputo time-fractional\nderivative. Several novel criterions are examined for showing the\npositive-definiteness of the associated kernel functions. We deduce strict\nenergy-dissipation for a number of non-local energy functionals, thereby\nproving fractional energy dissipation laws.\n"}
{"abstract": "  The cooperation mechanism of indirect reciprocity has been studied by making\nmultiple variations of its parts. This research proposes a new variant of Nowak\nand Sigmund model, focused on agents' attitude; it is called Individualistic\nIndirect Reciprocity. In our model, an agent reinforces its strategy to the\nextent to which it makes a profit. We also include conditions related to the\nenvironment, visibility of agents, cooperation demand, and the attitude of an\nagent to maintain his cooperation strategy. Using Agent-Based Model and a Data\nScience method, we show on simulation results that the discriminatory stance of\nthe agents prevails in most cases. In general, cooperators only appear in\nconditions with low visibility of reputation and a high degree of cooperation\ndemand. The results also show that when the reputation of others is unknown,\nwith a high obstinacy and high cooperation demand, a heterogeneous society is\nobtained. The simulations show a wide diversity of scenarios, centralized,\npolarized, and mixed societies.\n"}
{"abstract": "  Ultra-precise astrometry from the Gaia mission is expected to lead to\nastrometric detections of more than 20,000 exoplanets in our Galaxy. One of the\nfactors that could hamper such detections is the astrometric jitter caused by\nthe magnetic activity of the planet host stars. In our previous study, we\nmodeled astrometric jitter for the Sun observed equator-on. In this work, we\ngeneralize our model and calculate the photocenter jitter as it would be\nmeasured by the Gaia and Small-JASMINE missions for stars with solar rotation\nrate and effective temperature, but with various values of the inclination\nangle of the stellar rotation axis. In addition, we consider the effect of\nmetallicity and of nesting of active regions (i.e. the tendency of active\nregions to emerge in the vicinity of each other). We find that, while the\njitter of stars observed equator-on does not have any long-term trends and can\nbe easily filtered out, the photocenters of stars observed out of their\nequatorial planes experience systematic shifts over the course of the activity\ncycle. Such trends allow the jitter to be detected with continuous\nmeasurements, in which case it can interfere with planet detectability. An\nincrease in the metallicity is found to increase the jitter caused by stellar\nactivity. Active-region nesting can further enhance the peak-to-peak amplitude\nof the photocenter jitter to a level that could be detected by Gaia.\n"}
{"abstract": "  As mathematical computing becomes more democratized in high-level languages,\nhigh-performance symbolic-numeric systems are necessary for domain scientists\nand engineers to get the best performance out of their machine without deep\nknowledge of code optimization. Naturally, users need different term types\neither to have different algebraic properties for them, or to use efficient\ndata structures. To this end, we developed Symbolics.jl, an extendable symbolic\nsystem which uses dynamic multiple dispatch to change behavior depending on the\ndomain needs. In this work we detail an underlying abstract term interface\nwhich allows for speed without sacrificing generality. We show that by\nformalizing a generic API on actions independent of implementation, we can\nretroactively add optimized data structures to our system without changing the\npre-existing term rewriters. We showcase how this can be used to optimize term\nconstruction and give a 113x acceleration on general symbolic transformations.\nFurther, we show that such a generic API allows for complementary\nterm-rewriting implementations. We demonstrate the ability to swap between\nclassical term-rewriting simplifiers and e-graph-based term-rewriting\nsimplifiers. We showcase an e-graph ruleset which minimizes the number of CPU\ncycles during expression evaluation, and demonstrate how it simplifies a\nreal-world reaction-network simulation to halve the runtime. Additionally, we\nshow a reaction-diffusion partial differential equation solver which is able to\nbe automatically converted into symbolic expressions via multiple dispatch\ntracing, which is subsequently accelerated and parallelized to give a 157x\nsimulation speedup. Together, this presents Symbolics.jl as a next-generation\nsymbolic-numeric computing environment geared towards modeling and simulation.\n"}
{"abstract": "  We demonstrate that a long-propagating plasma bubble executing undulatory\nmotion can be produced in the wake of two co-propagating laser pulses: a\nnear-single-cycle injector and a multi-cycle driver. When the undulation\namplitude exceeds the analytically derived threshold, highly-localized\ninjections of plasma electrons into the bubble are followed by their\nlong-distance acceleration. While the locations of the injection regions are\ncontrolled by the carrier-envelope phase (CEP) of the injector pulse, the\nmono-energetic spectrum of the accelerated sub-femtosecond high-charge electron\nbunches is shown to be nearly CEP-independent\n"}
{"abstract": "  What if deep neural networks can learn from sparsity-inducing priors? When\nthe networks are designed by combining layer modules (CNN, RNN, etc), engineers\nless exploit the inductive bias, i.e., existing well-known rules or prior\nknowledge, other than annotated training data sets. We focus on employing\nsparsity-inducing priors in deep learning to encourage the network to concisely\ncapture the nature of high-dimensional data in an unsupervised way. In order to\nuse non-differentiable sparsity-inducing norms as loss functions, we plug their\nproximal mappings into the automatic differentiation framework. We demonstrate\nunsupervised learning of U-Net for background subtraction using low-rank and\nsparse priors. The U-Net can learn moving objects in a training sequence\nwithout any annotation, and successfully detect the foreground objects in test\nsequences.\n"}
{"abstract": "  We report full vector mapping of local magnetization in CeAlSi, a Weyl\nsemimetal in which both inversion and time-reversal symmetries are broken. The\nvector maps reveal unanticipated features both within domains and at their\nboundaries. Boundaries between domains form two kinds of walls with distinct\ntopology and therefore different interactions with Weyl fermions. Domain walls\naligned along the tetragonal axes, e.g. (100), exhibit emergent chirality\nforbidden by the bulk space group, while diagonal walls are non-chiral. Within\nthe domains, we observe that the previously reported set of four easy axes\naligned along the in-plane diagonals of the tetragonal structure actually split\nto form an octet with decreasing temperature below the magnetic transition. All\nthe above phenomena are ultimately traced to the noncollinear magnetic\nstructure of CeAlSi.\n"}
{"abstract": "  This paper presents a multiple learner algorithm called the 'Three Ensemble\nClustering 3EC' algorithm that classifies unlabeled data into quality clusters\nas a part of unsupervised learning. It offers the flexibility to explore the\ncontext of new clusters formed by an ensemble of algorithms based on internal\nvalidation indices.\n  It is worth mentioning that the input data set is considered to be a cluster\nof clusters. An anomaly can possibly manifest as a cluster as well. Each\npartitioned cluster is considered to be a new data set and is a candidate to\nexplore the most optimal algorithm and its number of partition splits until a\npredefined stopping criteria is met. The algorithms independently partition the\ndata set into clusters and the quality of the partitioning is assessed by an\nensemble of internal cluster validation indices. The 3EC algorithm presents the\nvalidation index scores from a choice of algorithms and its configuration of\npartitions and it is called the Tau Grid. 3EC chooses the most optimal score.\nThe 3EC algorithm owes its name to the two input ensembles of algorithms and\ninternal validation indices and an output ensemble of final clusters.\n  Quality plays an important role in this clustering approach and it also acts\nas a stopping criteria from further partitioning. Quality is determined based\non the quality of the clusters provided by an algorithm and its optimal number\nof splits. The 3EC algorithm determines this from the score of the ensemble of\nvalidation indices. The user can configure the stopping criteria by providing\nquality thresholds for the score range of each of the validation indices and\nthe optimal size of the output cluster. The users can experiment with different\nsets of stopping criteria and choose the most 'sensible group' of quality\nclusters\n"}
{"abstract": "  We study scattering of itinerant electrons off a magnetic hopfion in a\nthree-dimensional metallic magnet described by a magnetization vector $\\mathbf\nS(\\mathbf r)$. A hopfion is a confined topological soliton of $\\mathbf\nS(\\mathbf r)$ characterized by an {\\it emergent} magnetic field\n$B_\\gamma(\\mathbf r) \\equiv \\epsilon_{\\alpha\\beta\\gamma} \\,\\mathbf\nS\\cdot(\\nabla_\\alpha \\mathbf S\\times \\nabla_\\beta \\mathbf S)/4 \\neq 0$ with\nvanishing average value $\\langle \\mathbf B(\\mathbf r)\\rangle = 0$. We evaluate\nthe scattering amplitude in the opposite limits of large and small hopfion\nradius $R$ using the eikonal and Born approximations, respectively. In both\nlimits, we find that the scattering cross-section contains a skew-scattering\ncomponent giving rise to the Hall effect within a hopfion plane. That\nconclusion contests the popular notion that the topological Hall effect in\nnon-collinear magnetic structures necessarily implies $\\langle \\mathbf\nB(\\mathbf r)\\rangle \\neq 0$. In the limit of small hopfion radius $pR \\ll 1$,\nwe expand the Born series in powers of momentum $p$ and identify different\nexpansion terms corresponding to the hopfion anisotropy, toroidal moment, and\nskew-scattering.\n"}
{"abstract": "  We develop an approximate quasi-static theory describing the low-frequency\nplasmonic resonances of slender nanometallic rings and configurations thereof.\nIn particular, we use asymptotic arguments to reduce the plasmonic eigenvalue\nproblem governing the geometric (material- and frequency-independent) modes of\na given ring structure to a 1D-periodic integro-differential problem in which\nthe eigenfunctions are represented by azimuthal voltage and polarization-charge\nprofiles associated with each ring. We obtain closed-form solutions to the\nreduced eigenvalue problem for azimuthally invariant rings (including\ntorus-shaped rings but also allowing for non-circular cross-sectional shapes),\nas well as coaxial dimers and chains of such rings. For more general\ngeometries, involving azimuthally non-uniform rings and non-coaxial structures,\nthe reduced eigenvalue problem is solved using a semi-analytical scheme based\non Fourier expansions of the reduced eigenfunctions. In conjunction with the\nquasi-static spectral theory of plasmonic resonance, the geometric modes of a\ngiven nanometallic structure explicitly determine its response to external\nradiation. Here, we employ the asymptotically approximated modes described\nabove to obtain comparable approximations for the frequency response of a wide\nrange of nanometallic slender-ring structures under plane-wave illumination.\nThese examples demonstrate how the theory can be employed to interpret and\ngeometrically tune the plasmonic properties of highly nontrivial 3D\nnanometallic structures.\n"}
{"abstract": "  We investigate the solution to the logistic equation involving non-local\noperators in time. In the linear case such operators lead to the well-known\ntheory of time changes. We provide the probabilistic representation for the\nnon-linear logistic equation with non-local operators in time. The so-called\nfractional logistic equation has been investigated by many researchers, the\nproblem to find the explicit representation of the solution on the whole real\nline is still open. In our recent work the solution on compact sets has been\nwritten in terms of Euler's numbers.\n"}
{"abstract": "  A graph $\\Ga=(V,E)$ is called a Cayley graph of some group $T$ if the\nautomorphism group $\\Aut(\\Ga)$ contains a subgroup $T$ which acts on regularly\non $V$. If the subgroup $T$ is normal in $\\Aut(\\Ga)$ then $\\Ga$ is called a\nnormal Cayley graph of $T$. Let $r$ be an odd prime. Fang et al. \\cite{FMW}\nproved that, with a finite number of exceptions for finite simple group $T$,\nevery connected symmetric Cayley graph of $T$ of valency $r$ is normal. In this\npaper, employing maximal factorizations of finite almost simple groups, we work\nout a possible list of those exceptions for $T$.\n"}
{"abstract": "  We examine some basic physics surrounding regenerative braking and air drag.\nFirst, we analyze under what conditions it becomes energetically favorable to\nuse aggressive regenerative braking to reach a lower speed over ``coasting''\nwhere one relies solely on air drag to slow down. We then proceed to\nreformulate the question as an optimization problem to find the\nvelocity-profile that maximizes battery charge. Making a simplifying assumption\non battery-charging efficiency, we formulate the recovered energy as an\nintegral quantity and solve the associated Euler-Lagrange equation to find the\noptimal braking curve. Using Lagrange-multipliers, we also explore the effect\nof adding a fixed-displacement constraint.\n"}
{"abstract": "  The physical mechanism for triggering the changing-look phenomenon in active\ngalactic nuclei (AGNs) is still unclear. We explore this issue based on the\nmulti-wavelength spectral and flux variations for a changing-look AGN Mrk~1018\nwith long-term observations in the X-ray, optical/ultraviolet(UV), and radio\nbands. Both the optical and the X-ray emission experience rapid decay in\nchanging-look phase during 2010--2015, where a re-flare appears in the\noptical/UV and X-ray bands. We find a time lag of $\\sim 20 $ days of optical/UV\nbehind X-ray variations in type 1.9 phase. The 5 GHz radio flux decreases by\n$\\sim 20$\\% in type 1.9 phase during 2016--2017. We find both X-ray photon\nindex ($\\Gamma$) and the optical-to-X-ray spectral index (\\alphaox\\,) are\nanti-correlated with the Eddington scaled 2--10~keV X-ray luminosity\n($L_\\mathrm{X}/L_\\mathrm{Edd}$) in the type 1.9 phase. However, the type 1\nphase deviates from these two anti-correlations, which suggests that the change\nof broad emission lines might be regulated by the evolution of accretion disk\n(e.g., disappearing of the inner cold disk in the type 1.9 phase).\n"}
{"abstract": "  This study on public-private research collaboration measures the variation\nover time of the propensity of academics to collaborate with colleagues from\nprivate companies. It also investigates the change in weights of the main\ndrivers underlying the academics' propensity to collaborate, and whether the\ntype profile of the collaborating academics changes. To do this, the study\napplies an inferential model on a dataset of professors working in Italian\nuniversities in consecutive periods, 2010-2013 and 2014-2017. The results,\nobtained at overall and field levels, support the formulation of policies aimed\nat fostering public-private research collaborations, and should be taken into\naccount in post-assessment of their effectiveness.\n"}
{"abstract": "  In recent years the use of convolutional layers to encode an inductive bias\n(translational equivariance) in neural networks has proven to be a very\nfruitful idea. The successes of this approach have motivated a line of research\ninto incorporating other symmetries into deep learning methods, in the form of\ngroup equivariant convolutional neural networks. Much of this work has been\nfocused on roto-translational symmetry of $\\mathbf R^d$, but other examples are\nthe scaling symmetry of $\\mathbf R^d$ and rotational symmetry of the sphere. In\nthis work, we demonstrate that group equivariant convolutional operations can\nnaturally be incorporated into learned reconstruction methods for inverse\nproblems that are motivated by the variational regularisation approach. Indeed,\nif the regularisation functional is invariant under a group symmetry, the\ncorresponding proximal operator will satisfy an equivariance property with\nrespect to the same group symmetry. As a result of this observation, we design\nlearned iterative methods in which the proximal operators are modelled as group\nequivariant convolutional neural networks. We use roto-translationally\nequivariant operations in the proposed methodology and apply it to the problems\nof low-dose computerised tomography reconstruction and subsampled magnetic\nresonance imaging reconstruction. The proposed methodology is demonstrated to\nimprove the reconstruction quality of a learned reconstruction method with a\nlittle extra computational cost at training time but without any extra cost at\ntest time.\n"}
{"abstract": "  In this paper, we introduce the NameRec* task, which aims to do highly\naccurate and fine-grained person name recognition. Traditional Named Entity\nRecognition models have good performance in recognising well-formed person\nnames from text with consistent and complete syntax, such as news articles.\nHowever, there are rapidly growing scenarios where sentences are of incomplete\nsyntax and names are in various forms such as user-generated contents and\nacademic homepages. To address person name recognition in this context, we\npropose a fine-grained annotation scheme based on anthroponymy. To take full\nadvantage of the fine-grained annotations, we propose a Co-guided Neural\nNetwork (CogNN) for person name recognition. CogNN fully explores the\nintra-sentence context and rich training signals of name forms. To better\nutilize the inter-sentence context and implicit relations, which are extremely\nessential for recognizing person names in long documents, we further propose an\nInter-sentence BERT Model (IsBERT). IsBERT has an overlapped input processor,\nand an inter-sentence encoder with bidirectional overlapped contextual\nembedding learning and multi-hop inference mechanisms. To derive benefit from\ndifferent documents with a diverse abundance of context, we propose an advanced\nAdaptive Inter-sentence BERT Model (Ada-IsBERT) to dynamically adjust the\ninter-sentence overlapping ratio to different documents. We conduct extensive\nexperiments to demonstrate the superiority of the proposed methods on both\nacademic homepages and news articles.\n"}
{"abstract": "  Fully Homomorphic Encryption (FHE) allows a third party to perform arbitrary\ncomputations on encrypted data, learning neither the inputs nor the computation\nresults. Hence, it provides resilience in situations where computations are\ncarried out by an untrusted or potentially compromised party. This powerful\nconcept was first conceived by Rivest et al. in the 1970s. However, it remained\nunrealized until Craig Gentry presented the first feasible FHE scheme in 2009.\n  The advent of the massive collection of sensitive data in cloud services,\ncoupled with a plague of data breaches, moved highly regulated businesses to\nincreasingly demand confidential and secure computing solutions. This demand,\nin turn, has led to a recent surge in the development of FHE tools. To\nunderstand the landscape of recent FHE tool developments, we conduct an\nextensive survey and experimental evaluation to explore the current state of\nthe art and identify areas for future development.\n  In this paper, we survey, evaluate, and systematize FHE tools and compilers.\nWe perform experiments to evaluate these tools' performance and usability\naspects on a variety of applications. We conclude with recommendations for\ndevelopers intending to develop FHE-based applications and a discussion on\nfuture directions for FHE tools development.\n"}
{"abstract": "  A search for R-parity violating supersymmetry in final states characterised\nby high jet multiplicity, at least one isolated light lepton and either zero or\nat least three $b$-tagged jets is presented. The search uses 139 fb$^{-1}$ of\n$\\sqrt{s} = 13$ TeV proton-proton collision data collected by the ATLAS\nexperiment during Run 2 of the Large Hadron Collider. The results are\ninterpreted in the context of R-parity-violating supersymmetry models that\nfeature gluino production, top-squark production, or electroweakino production.\nThe dominant sources of background are estimated using a data-driven model,\nbased on observables at medium jet multiplicity, to predict the $b$-tagged jet\nmultiplicity distribution at the higher jet multiplicities used in the search.\nMachine learning techniques are used to reach sensitivity to electroweakino\nproduction, extending the data-driven background estimation to the shape of the\nmachine learning discriminant. No significant excess over the Standard Model\nexpectation is observed and exclusion limits at the 95% confidence-level are\nextracted, reaching as high as 2.4 TeV in gluino mass, 1.35 TeV in top-squark\nmass, and 320 (365) GeV in higgsino (wino) mass.\n"}
{"abstract": "  Digital news outlets rely on a variety of outside contributors, from\nfreelance journalists, to political commentators, to executives and\npoliticians. These external dependencies create a network among news outlets,\ntraced along the contributors they share. Using connections between outlets, we\ndemonstrate how contributors' publishing trajectories tend to align with outlet\npolitical leanings. We also show how polarized clustering of outlets translates\nto differences in the topics of news covered and the style and tone of articles\npublished. In addition, we demonstrate how contributors who cross partisan\ndivides tend to focus on less explicitly political topics. This work addresses\nan important gap in the media polarization literature, by highlighting how\nstructural factors on the production side of news media create an ecosystem\nshaped by political leanings, independent of the priorities of any one person\nor organization.\n"}
{"abstract": "  Most data in cold-atom experiments comes from images, the analysis of which\nis limited by our preconceptions of the patterns that could be present in the\ndata. We focus on the well-defined case of detecting dark solitons -- appearing\nas local density depletions in a Bose-Einstein condensate (BEC) -- using a\nmethodology that is extensible to the general task of pattern recognition in\nimages of cold atoms. Studying soliton dynamics over a wide range of parameters\nrequires the analysis of large datasets, making the existing\nhuman-inspection-based methodology a significant bottleneck. Here we describe\nan automated classification and positioning system for identifying localized\nexcitations in atomic BECs utilizing deep convolutional neural networks to\neliminate the need for human image examination. Furthermore, we openly publish\nour labeled dataset of dark solitons, the first of its kind, for further\nmachine learning research.\n"}
{"abstract": "  Predefined evenly-distributed class centroids (PEDCC) can be widely used in\nmodels and algorithms of pattern classification, such as CNN classifiers,\nclassification autoencoders, clustering, and semi-supervised learning, etc. Its\nbasic idea is to predefine the class centers, which are evenly-distributed on\nthe unit hypersphere in feature space, to maximize the inter-class distance.\nThe previous method of generating PEDCC uses an iterative algorithm based on a\ncharge model, that is, the initial values of various centers (charge positions)\nare randomly set from the normal distribution, and the charge positions are\nupdated iteratively with the help of the repulsive force between charges of the\nsame polarity. The class centers generated by the algorithm will produce some\nerrors with the theoretically evenly-distributed points, and the generation\ntime will be longer. This paper takes advantage of regular polyhedron in\nhigh-dimensional space and the evenly distribution of points on the n\ndimensional hypersphere to generate PEDCC mathematically. Then, we discussed\nthe basic and extensive characteristics of the frames formed by PEDCC. Finally,\nexperiments show that new algorithm is not only faster than the iterative\nmethod, but also more accurate in position. The mathematical analysis and\nexperimental results of this paper can provide a theoretical tool for using\nPEDCC to solve the key problems in the field of pattern recognition, such as\ninterpretable supervised/unsupervised learning, incremental learning,\nuncertainty analysis and so on.\n"}
{"abstract": "  We leverage the interplay between microscopic variability and macroscopic\norder to connect physical descriptions across scales directly from data,\nwithout underlying equations. We reconstruct a state space by concatenating\nmeasurements in time, building a maximum entropy partition of the resulting\nsequences, and choosing the sequence length to maximize predictive information.\nTrading non-linear trajectories for linear, ensemble evolution, we analyze\nreconstructed dynamics through transfer operators. The evolution is\nparameterized by a transition time $\\tau$: capturing the source entropy rate at\nsmall $\\tau$ and revealing timescale separation with collective, coherent\nstates through the operator spectrum at larger $\\tau$. Applicable to both\ndeterministic and stochastic systems, we illustrate our approach through the\nLangevin dynamics of a particle in a double-well potential and the Lorenz\nsystem. Applied to the behavior of the nematode worm $C. elegans$, we derive a\n\"run-and-pirouette\" navigation strategy directly from posture dynamics. We\ndemonstrate how sequences simulated from the ensemble evolution capture both\nfine scale posture dynamics and large scale effective diffusion in the worm's\ncentroid trajectories and introduce a top-down, operator-based clustering which\nreveals subtle subdivisions of the \"run\" behavior.\n"}
{"abstract": "  The concept of bounded expansion provides a robust way to capture sparse\ngraph classes with interesting algorithmic properties. Most notably, every\nproblem definable in first-order logic can be solved in linear time on bounded\nexpansion graph classes. First-order interpretations and transductions of\nsparse graph classes lead to more general, dense graph classes that seem to\ninherit many of the nice algorithmic properties of their sparse counterparts.\n  In this work we introduce lacon- and shrub-decompositions and use them to\ncharacterize transductions of bounded expansion graph classes and other graph\nclasses. If one can efficiently compute sparse shrub- or lacon-decompositions\nof transductions of bounded expansion classes then one can solve every problem\ndefinable in first-order logic in linear time on these classes.\n"}
{"abstract": "  We study the initial value problem of fully nonlinear third-order dispersive\nequations on the torus. Under some conditions on the nonlinearity and the data,\nwe prove that the equation behaves like a parabolic one: there exists a unique\nlocal solution in one direction of time that is infinitely smooth and the\nproblem in not well-posed in the other direction. Under other conditions on the\nnonlinearity and the data, we prove that the equation behaves like a dispersive\none: there exists a unique local solution in both directions of time.\n"}
{"abstract": "  NEWAGE is a direction-sensitive dark matter search using a low-pressure\ngaseous time projection chamber. A low alpha-ray emission rate micro pixel\nchamber had been developed in order to reduce background for dark matter\nsearch. We conducted the dark matter search at the Kamioka Observatory in 2018.\nThe total live time was 107.6 days corresponding to an exposure of 1.1\nkg${\\cdot}$days. Two events remained in the energy region of 50-60 keV which\nwas consistent with 2.5 events of the expected background. A directional\nanalysis was carried out and no significant forward-backward asymmetry derived\nfrom the WIMP-nucleus elastic scatterings was found. Thus a 90% confidence\nlevel upper limit on Spin-Dependent WIMP-proton cross section of 50 pb for a\nWIMP mass of 100 GeV/c2 was derived. This limit is the most stringent yet\nobtained from direction-sensitive dark matter search experiments.\n"}
{"abstract": "  While current autonomous navigation systems allow robots to successfully\ndrive themselves from one point to another in specific environments, they\ntypically require extensive manual parameter re-tuning by human robotics\nexperts in order to function in new environments. Furthermore, even for just\none complex environment, a single set of fine-tuned parameters may not work\nwell in different regions of that environment. These problems prohibit reliable\nmobile robot deployment by non-expert users. As a remedy, we propose Adaptive\nPlanner Parameter Learning (APPL), a machine learning framework that can\nleverage non-expert human interaction via several modalities -- including\nteleoperated demonstrations, corrective interventions, and evaluative feedback\n-- and also unsupervised reinforcement learning to learn a parameter policy\nthat can dynamically adjust the parameters of classical navigation systems in\nresponse to changes in the environment. APPL inherits safety and explainability\nfrom classical navigation systems while also enjoying the benefits of machine\nlearning, i.e., the ability to adapt and improve from experience. We present a\nsuite of individual APPL methods and also a unifying cycle-of-learning scheme\nthat combines all the proposed methods in a framework that can improve\nnavigation performance through continual, iterative human interaction and\nsimulation training.\n"}
{"abstract": "  For nonlinear differential-algebraic equations (DAEs), we define two kinds of\nequivalences, namely, the external and internal equivalence. Roughly speaking,\nthe word \"external\" means that we consider a DAE (locally) everywhere and\n\"internal\" means that we consider the DAE on its (locally) maximal invariant\nsubmanifold (i.e., where its solutions exist) only. First, we revise the\ngeometric reduction method in DAEs solution theory and formulate an\nimplementable algorithm to realize that method. Then a procedure named\nexplicitation with driving variables is proposed to connect nonlinear DAEs with\nnonlinear control systems and we show that the driving variables of an\nexplicitation system can be reduced under some involutivity conditions.\nFinally, due to the explicitation, we will use some notions from nonlinear\ncontrol theory to derive two nonlinear generalizations of the Weierstrass form.\n"}
{"abstract": "  The multiple scattering of an ultrashort laser pulse by a turbid dispersive\nmedium (namely a cloud of bubbles in water) is investigated by means of Monte\nCarlo simulations. The theory of Gouesbet and Gr\\'ehan [Part. Part. Syst.\nCharact. 17 213-224 (2000)] is used to derive an energetic model of the\nscattering transient. The spreading and extinction of the pulse is decoupled\nfrom the transient of scattering to describe each phenomenon individually. The\ntransient of scattering is modeled with the Lorenz-Mie Theory, thus also valid\nfor a relative refractive index lower than one, contrary to the Debye series\nexpansion which does not converge close to the critical angle. To this aim, the\nScattering Impulse Response Function (SIRF) allows to detect the different\nmodes of scattering transient in time and direction. The present approach is\nmore generic and can simulate clouds of air bubbles in water. Two Monte-Carlo\napproaches are proposed. The first is a pure Monte Carlo approach where the\ndelay due to the scattering is randomly drawn at each event, while the second\nis based on the transport of the whole scattering signal. They are both\nembedded in the Monte Carlo code Scatter3D [JOSA A 24, 2206-2219 (2007)]. Both\nmodels produce equivalent trends and are validated against published numerical\nresults. They are applied to the multiple scattering of ultra short pulse by a\ncloud of bubble in water in the forward direction. The pulse spread due to the\npropagation in water is computed for a wide range of traveled distances and\npulse durations, and the optimal pulse duration is given to minimize the pulse\nspread at a given distance. The main result is that the scattered photons exit\nthe turbid medium earlier than the ballistic photons and produce a double peak\nrelated to the refraction in the bubble. This demonstrates the possibility to\ndevelop new diagnostics to characterize dynamic bubbly flows.\n"}
{"abstract": "  We derive two upper bounds for the probability of deviation of a\nvector-valued Lipschitz function of a collection of random variables from its\nexpected value. The resulting upper bounds can be tighter than bounds obtained\nby a direct application of a classical theorem due to Bobkov and G\\\"{o}tze.\n"}
{"abstract": "  We formulate a general statement of the problem of defining invariant\nmeasures with certain properties and suggest an ergodic method of perturbations\nfor describing such measures.\n"}
{"abstract": "  We study the extended Kitaev model called the K$\\Gamma$ model, using a\nperturbative expansion combined with a well-controlled mean-field approximation\nand a cutting-edge exact diagonalization. In the phase diagram, we discover a\nnematic Kitaev spin liquid and a Kekul\\'e Kitaev spin liquid. The former\npotentially explains the high-field nematic state with zero Chern number\nexperimentally observed in $\\alpha$-RuCl$_3$, even for sufficiently small\nvalues of $\\Gamma/|K|$. The latter has a Majorana zero mode in its $Z_3$ vortex\ncore, which can be potentially controlled by the domain wall motion. This opens\na possible application of the quantum liquid crystal phases in the K$\\Gamma$\nmodel to the topological quantum computation.\n"}
{"abstract": "  Intermittent motion, called stick--slip, is a friction instability that\ncommonly occurs during relative sliding of two elastic solids. In adhesive\npolymer contacts, where elasticity and interface adhesion are strongly coupled,\nstick--slip results from the propagation of slow detachment waves at the\ninterface. Using \\emph{in situ} imaging experiments at an adhesive contact, we\nshow the occurrence of two distinct detachment waves moving parallel\n(Schallamach wave) and anti-parallel (separation wave) to the applied remote\nsliding. Both waves cause slip in the same direction and travel at speeds much\nlesser than any elastic wave speed. We use an elastodynamic framework to\ndescribe the propagation of these slow detachment waves at an elastic-rigid\ninterface and obtain governing integral equations in the low wave speed limit.\nThese integral equations are solved in closed form when the elastic solid is\nincompressible. Two solution branches emerge, corresponding to opposite moving\ndetachment waves, just as seen in the experiments. A numerical scheme is used\nto obtain interface stresses and velocities for the incompressible case for\narbitrary Poisson ratio. Based on these results, we explicitly demonstrate a\ncorrespondence between propagating slow detachment waves and a static\nbi-material interface crack. Based on this, and coupled with a recently\nproposed fracture analogy for dynamic friction, we develop a phase diagram\nshowing domains of possible occurrence of stick--slip via detachment waves\nvis-\\'a-vis steady interface sliding.\n"}
{"abstract": "  The Hua-Radon and polarized Hua-Radon transform are two orthogonal\nprojections defined on holomorphic functions in the Lie sphere. Both\ntransformations can be written as integral transforms with respect to a\nsuitable reproducing kernel. Integrating both kernels over a Stiefel manifold\nyields a linear combination of zonal spherical monogenics. Using an Almansi\ntype decomposition of holomorphic functions and reproducing properties of the\nzonal monogenics, we obtain an inversion formula for both the Hua-Radon and the\npolarized Hua-Radon transform.\n"}
{"abstract": "  We estimate the effects on the decay constants of charmonium and on heavy\nmeson masses due to the charm quark in the sea. Our goal is to understand\nwhether for these quantities $N_f=2+1$ lattice QCD simulations provide results\nthat can be compared with experiments or whether $N_f=2+1+1$ QCD including the\ncharm quark in the sea needs to be simulated. We consider two theories, $N_f=0$\nQCD and QCD with $N_f=2$ charm quarks in the sea. The charm sea effects (due to\ntwo charm quarks) are estimated comparing the results obtained in these two\ntheories, after matching them and taking the continuum limit. The absence of\nlight quarks allows us to simulate the $N_f=2$ theory at lattice spacings down\nto $0.023$ fm that are crucial for reliable continuum extrapolations. We find\nthat sea charm quark effects are below $1\\%$ for the decay constants of\ncharmonium. Our results show that decoupling of charm works well up to energies\nof about $500$ MeV. We also compute the derivatives of the decay constants and\nmeson masses with respect to the charm mass. For these quantities we again do\nnot see a significant dynamical charm quark effect, albeit with a lower\nprecision. For mesons made of a charm quark and a heavy antiquark, whose mass\nis twice that of the charm quark, sea effects are only about $0.1\\%$ in the\nratio of vector to pseudoscalar masses.\n"}
{"abstract": "  In this paper, free Bosons are used to study some integrable properties of\nthe CKP hierarchy, from the aspects of tau functions. Firstly, the modified CKP\nhierarchy is constructed by using free Bosons and the corresponding Lax\nstructure is given. Then the constrained CKP hierarchy is found to be related\nwith the modified CKP hierarchy, and the corresponding solutions are derived by\nusing free Bosons. Next by using the relations between the Darboux\ntransformations and the squared eigenfunction symmetries, we express the\nDarboux transformations of the CKP hierarchy in terms of free Bosons, by which\none can better understand the essential of the CKP Darboux transformations. In\nparticular, the additional symmetries of the CKP hierarchy can be viewed as the\ninfinitesimal generator of the CKP Darboux transformations. Based upon these\nresults, we finally obtain the actions of the CKP additional symmetries on the\nCKP tau functions constructed by free Bosons.\n"}
{"abstract": "  Two-dimensional (2D) materials with robust magnetism have drawn immense\nattention for their promising applications in spintronics. Recently, intrinsic\nferromagnetic vanadium triiodide (VI3) has been synthesized experimentally. To\nenhance its spintronic property, we modified VI3 by interstitial doping with 3d\ntransition metals (TM) and used first-principles calculations to investigate\nthe geometric structure, formation energy, electronic property, and magnetism\nof pristine VI3 and 3d TM-doped VI3 monolayer. Among eight transition metal\n(Sc-, Ti-, V-, Cr-, Mn-, Fe-, Co-, and Ni-) doped VI3 materials, four of them\n(Ti-, V-, Mn-, and Ni-doped VI3) show robust magnetism with full spin\npolarization near the Fermi energy. Our research demonstrates that Ti-doped VI3\nresults in half-metallic semiconductor properties (HMS), while V-doped VI3 and\nNi-doped VI3 result in half-semiconductor properties (HSC). Surprisingly,\nMn-doped VI3 exhibits an unusual bipolar magnetic semiconductor property (BMS).\nThis unique combination of strong ferromagnetism and 100% spin polarization\nwith a half-metallic, half-semiconductor, or bipolar semiconductor property\nrenders 3d TM-doped VI3 as potential candidates for next generation\nsemiconductor spintronic applications. These spin-polarized materials will be\nextremely useful for spin-current generation and other spintronic applications.\n"}
{"abstract": "  This paper shows the usefulness of Perov's contraction principle, which\ngeneralizes Banach's contraction principle to a vector-valued metric, for\nstudying dynamic programming problems in which the discount factor can be\nstochastic. The discounting condition $\\beta<1$ is replaced by $\\rho(B)<1$,\nwhere $B$ is an appropriate nonnegative matrix and $\\rho$ denotes the spectral\nradius. Blackwell's sufficient condition is also generalized in this setting.\nApplications to asset pricing and optimal savings are discussed.\n"}
{"abstract": "  This paper is concerned with the linear theory of thermoelasticity with\nmicrotemperatures, based on the entropy balance proposed by Green and Naghdi,\nwhich permits the transmission of heat as thermal waves of finite speed. We\nanalyze the behavior of Rayleigh waves in an unbounded isotropic homogeneous\nstrongly elliptic thermoelastic material with microtemperatures. The related\nsolution of the Rayleigh surface wave problem is expressed as a linear\ncombination of the elements of the bases of the kernels of appropriate\nmatrices. The secular equation is established and afterwards an explicit form\nis written when some coupling constitutive coefficients vanish. Then, we solve\nnumerically the secular equation by mean of a grafical metod and by taking\narbitrary data for strongly elliptic thermoelastic material.\n"}
{"abstract": "  Computation of the instantaneous phase and amplitude via the Hilbert\nTransform is a powerful tool of data analysis. This approach finds many\napplications in various science and engineering branches but is not proper for\ncausal estimation because it requires knowledge of the signal's past and\nfuture. However, several problems require real-time estimation of phase and\namplitude; an illustrative example is phase-locked or amplitude-dependent\nstimulation in neuroscience. In this paper, we discuss and compare three causal\nalgorithms that do not rely on the Hilbert Transform but exploit well-known\nphysical phenomena, the synchronization and the resonance. After testing the\nalgorithms on a synthetic data set, we illustrate their performance computing\nphase and amplitude for the accelerometer tremor measurements and a\nParkinsonian patient's beta-band brain activity.\n"}
{"abstract": "  We investigate the creation and control of emergent collective behavior and\nquantum correlations using feedback in an emitter-waveguide system using a\nminimal model. Employing homodyne detection of photons emitted from a\nlaser-driven emitter ensemble into the modes of a waveguide allows to generate\nintricate dynamical phases. In particular, we show the emergence of a\ntime-crystal phase, the transition to which is controlled by the feedback\nstrength. Feedback enables furthermore the control of many-body quantum\ncorrelations, which become manifest in spin squeezing in the emitter ensemble.\nDeveloping a theory for the dynamics of fluctuation operators we discuss how\nthe feedback strength controls the squeezing and investigate its temporal\ndynamics and dependence on system size. The largely analytical results allow to\nquantify spin squeezing and fluctuations in the limit of large number of\nemitters, revealing critical scaling of the squeezing close to the transition\nto the time-crystal. Our study corroborates the potential of integrated\nemitter-waveguide systems -- which feature highly controllable photon emission\nchannels -- for the exploration of collective quantum phenomena and the\ngeneration of resources, such as squeezed states, for quantum enhanced\nmetrology.\n"}
{"abstract": "  Optical nanosensors are promising for hydrogen sensing because they are\nsmall, free from spark generation, and feasible for remote optical readout.\nConventional optical nanosensors require broadband excitation and\nspectrometers, rendering the devices bulky and complex. An alternative is\nspatial intensity-based optical sensing, which only requires an imaging system\nand a smartly designed platform to report the spatial distribution of\nanalytical optical signals. Here, we present a spatial intensity-based hydrogen\nsensing platform based on Fano-like spatial distribution of the transmission in\na Pd-Al2O3-Au metal-insulator-metal plasmonic Doppler grating (MIM-PDG). The\nMIM-PDG manifests the Fano resonance as an asymmetric spatial transmission\nintensity profile. The absorption of hydrogen changes the spatial Fano-like\ntransmission profiles, which can be analyzed with a \"spatial\" Fano-resonance\nmodel and the extracted Fano resonance parameters can be used to establish\nanalytical calibration lines. While gratings sensitive to hydrogen absorption\nare suitable for hydrogen sensing, we also found hydrogen insensitive gratings,\nwhich provide an unperturbed reference signal and may find applications in\nnanophotonic devices, that require a stable optical response under fluctuating\nhydrogen atmosphere. The MIM-PDG platform is a spectrometer-free and\nintensity-based optical sensor that requires only an imaging system, making it\npromising for cellphone-based optical sensing applications.\n"}
{"abstract": "  In this note we first consider a ternary matrix group related to the von\nNeumann regular semigroups and to the Artin braid group (in an algebraic way).\nThe product of a special kind of ternary matrices (idempotent and of finite\norder) reproduces the regular semigroups and braid groups with their binary\nmultiplication of components. We then generalize the construction to the higher\narity case, which allows us to obtain some higher degree versions (in our\nsense) of the regular semigroups and braid groups. The latter are connected\nwith the generalized polyadic braid equation and $R$-matrix introduced by the\nauthor, which differ from any version of the well-known tetrahedron equation\nand higher-dimensional analogs of the Yang-Baxter equation, $n$-simplex\nequations. The higher degree (in our sense) Coxeter group and symmetry groups\nare then defined, and it is shown that these are connected only in the\nnon-higher case.\n"}
{"abstract": "  Despite outstanding performance on public benchmarks, face recognition still\nsuffers due to domain mismatch between training (source) and testing (target)\ndata. Furthermore, these domains are not shared classes, which complicates\ndomain adaptation. Since this is also a fine-grained classification problem\nwhich does not strictly follow the low-density separation principle,\nconventional domain adaptation approaches do not resolve these problems. In\nthis paper, we formulate domain mismatch in face recognition as a style\nmismatch problem for which we propose two methods. First, we design a domain\ndiscriminator with human-level judgment to mine target-like images in the\ntraining data to mitigate the domain gap. Second, we extract style\nrepresentations in low-level feature maps of the backbone model, and match the\nstyle distributions of the two domains to find a common style representation.\nEvaluations on verification and open-set and closed-set identification\nprotocols show that both methods yield good improvements, and that performance\nis more robust if they are combined. Our approach is competitive with related\nwork, and its effectiveness is verified in a practical application.\n"}
{"abstract": "  We report on the generation of bulk photocurrents in materials driven by\nnon-resonant bi-chromatic fields that are circularly polarized and co-rotating.\nThe nonlinear photocurrents have a fully controllable directionality and\namplitude without requiring carrier-envelope-phase stabilization or few-cycle\npulses, and are generated with photon energies much smaller than the band gap\n(reducing heating in the photo-conversion process). We demonstrate with\nab-initio calculations that the photocurrent generation mechanism is universal\nand arises in gaped materials (Si, diamond, MgO, hBN), in semi-metals\n(graphene), and in two- and three-dimensional systems. Photocurrents are shown\nto rely on sub-laser-cycle asymmetries in the nonlinear response that build-up\ncoherently from cycle-to-cycle as the conduction band is populated.\nImportantly, the photocurrents are always transverse to the major axis of the\nco-circular lasers regardless of the material's structure and orientation\n(analogously to a Hall current), which we find originates from a generalized\ntime-reversal symmetry in the driven system. At high laser powers (~10^13\nW/cm^2) this symmetry can be spontaneously broken by vast electronic\nexcitations, which is accompanied by an onset of carrier-envelope-phase\nsensitivity and ultrafast many-body effects. Our results are directly\napplicable for efficient light-driven control of electronics, and for enhancing\nsub-band-gap bulk photovoltaic effects.\n"}
{"abstract": "  We investigate the effects of disorder in a simple model of a hybridized\ntwo-dimensional two-band s-wave superconductor. We take into account the\nsituations in which these bands are formed by electronic orbitals with angular\nmomentum such that the hybridization $V(\\mathbf{k})$ among them is\nantisymmetric, under inversion symmetry. The impurity potential is given by an\nindependent random variable $W$ which controls the strength of the on-site\ndisorder. We find that while the random disorder acts in detriment of\nsuperconductivity, hybridization proceeds favoring superconductivity. This\nshows that hybridization, which may be induced by pressure or doping, plays an\nimportant hole in two-band models of superconductivity, making them eligible\ncandidates to describe real materials. We also find that in moderate and strong\ndisorder, the system is broken into islands, with correlated local order\nparameters. These correlations persist to distances of several order lattice\nspacing which corresponds to the size of the SC-Islands.\n"}
{"abstract": "  A new mechanism for Topological Hall Effect (THE) was recently proposed for\nthe spiral magnet YMn$_{6}$Sn$_{6},$ which requires transverse conical spiral\n(TCS) magnetism, induced by external magnetic field, combined with thermally\nexcite helical spiral magnons. In principle, this mechanism should be\napplicable to other itinerant spiral magnets as well. In this paper, we show\nthat another metamagnetic compound, Fe$_{3}$Ga$_{4}$, where THE was observed\nexperimentally before, in one of its phases satisfies this condition, and the\nproposed theory of thermal-fluctuation driven THE is quantitatively consistent\nwith the experiment. This finding suggests that this mechanism is indeed rather\nuniversal, and the effect may have been observed in other compounds before, but\noverlooked.\n"}
{"abstract": "  We ascertain conditions and structures on categories and semigroups which\nadmit the construction of pseudo-products and trace products respectively,\nmaking their connection as precise as possible. This topic is modelled on the\nESN Theorem and its generalization to ample semigroups. Unlike some other\nvariants of ESN, it is self-dual (two-sided), and the condition of commuting\nprojections is relaxed. The condition that projections form a band (are closed\nunder multiplication) is shown to be a very natural one. One-sided reducts are\nconsidered, and compared to (generalized) D-semigroups. Finally the special\ncase when the category is a groupoid is examined.\n"}
{"abstract": "  A protocol for fast and robust magnon transport in a one-dimensional spin\nchain is devised. Employing an approximate mapping between the chain and a\nsingle harmonically trapped particle, we exploit the known analytic control\nprotocols for the latter and adopt them to achieve fast, high-fidelity\ntransport in the chain. We compare the performance with finite time adiabatic\nprotocols, establishing that the designed scheme allows for significantly\nfaster and more stable transport. Furthermore, we show that a sharp transition\nexists between regions in which the protocol is effective and when it breaks\ndown, giving rise to a heuristic speed limit for the process.\n"}
{"abstract": "  This paper is devoted to studying an augmented Lagrangian method for solving\na class of manifold optimization problems, which have nonsmooth objective\nfunctions and non-negative constraints. Under the constant positive linear\ndependence condition on manifolds, we show that the proposed method converges\nto a stationary point of the nonsmooth manifold optimization problem. Moreover,\nwe propose a globalized semismooth Newton method to solve the augmented\nLagrangian subproblem on manifolds efficiently. The local superlinear\nconvergence of the manifold semismooth Newton method is also established under\nsome suitable conditions. We also prove that the semismoothness on submanifolds\ncan be inherited from that in the ambient manifold. Finally, numerical\nexperiments on compressed modes and (constrained) sparse principal component\nanalysis illustrate the advantages of the proposed method.\n"}
{"abstract": "  In the setting of secure multiparty computation (MPC), a set of mutually\ndistrusting parties wish to jointly compute a function, while guaranteeing the\nprivacy of their inputs and the correctness of the output. An MPC protocol is\ncalled \\emph{fully secure} if no adversary can prevent the honest parties from\nobtaining their outputs. A protocol is called \\emph{fair} if an adversary can\nprematurely abort the computation, however, only before learning any new\ninformation.\n  We present highly efficient transformations from fair computations to fully\nsecure computations, assuming the fraction of honest parties is constant (e.g.,\n$1\\%$ of the parties are honest). Compared to previous transformations that\nrequire linear invocations (in the number of parties) of the fair computation,\nour transformations require super-logarithmic, and sometimes even\nsuper-constant, such invocations. The main idea is to delegate the computation\nto chosen random committees that invoke the fair computation. Apart from the\nbenefit of uplifting security, the reduction in the number of parties is also\nuseful, since only committee members are required to work, whereas the\nremaining parties simply \"listen\" to the computation over a broadcast channel.\n"}
{"abstract": "  Polarisation labelling spectroscopy technique was employed to study the\n3$^{1}\\Pi_{u}$ state of Cs$_2$ molecule. The main equlibrium constants are\n$T_e=20684.60$cm$^{-1}$, $\\omega_e=30.61$cm$^{-1}$ and $R_e=5.27$\\r{A}.\nVibrational levels $v=4-35$ of the 3$^{1}\\Pi_{u}$ state were found to be\nsubject to strong perturbations by the neighbouring electronic states. Energies\nof 3094 rovibronic levels of the perturbed complex were determined.\n"}
{"abstract": "  This paper presents a physics-based modeling framework for the analysis and\ntransient simulation of circuits containing Spin-Transfer Torque (STT) Magnetic\nTunnel Junction (MTJ) devices. The framework provides the tools to analyze the\nstochastic behavior of MTJs and to generate Verilog-A compact models for their\nsimulation in large VLSI designs, addressing the need for an industry-ready\nmodel accounting for real-world reliability and scalability requirements.\nDevice dynamics are described by the Landau-Lifshitz-Gilbert-Slonczewsky\n(s-LLGS ) stochastic magnetization considering Voltage-Controlled Magnetic\nAnisotropy (VCMA) and the non-negligible statistical effects caused by thermal\nnoise. Model behavior is validated against the OOMMF magnetic simulator and its\nperformance is characterized on a 1-Mb 28 nm Magnetoresistive-RAM (MRAM) memory\nproduct.\n"}
{"abstract": "  Microservice architecture advocates a number of technologies and practices\nsuch as lightweight container, container orchestration, and DevOps, with the\npromised benefits of faster delivery, improved scalability, and greater\nautonomy. However, microservice systems implemented in industry vary a lot in\nterms of adopted practices and achieved benefits, drastically different from\nwhat is advocated in the literature. In this article, we conduct an empirical\nstudy, including an online survey with 51 responses and 14 interviews for\nexperienced microservice experts to advance our understanding regarding to\nmicroservice practices in industry. As a part of our findings, the empirical\nstudy clearly revealed three levels of maturity of microservice systems (from\nbasic to advanced): independent development and deployment, high scalability\nand availability, and service ecosystem, categorized by the fulfilled benefits\nof microservices. We also identify 11 practical issues that constrain the\nmicroservice capabilities of organizations. For each issue, we summarize the\npractices that have been explored and adopted in industry, along with the\nremaining challenges. Our study can help practitioners better position their\nmicroservice systems and determine what infrastructures and capabilities are\nworth investing. Our study can also help researchers better understand\nindustrial microservice practices and identify useful research problems.\n"}
{"abstract": "  We study both the static properties of $^{11}$Be and its reaction dynamics\nduring electromagnetic breakup under a unified framework. A many-body approach\n- the antisymmetrized molecular dynamics (AMD) is used to describe the\nstructure of the neutron-halo nucleus, $^{11}$Be. The same AMD wave function is\nthen adapted as an input to the fully quantum theory of Coulomb breakup under\nthe aegis of the finite range distorted wave Born approximation theory. The\ncalculated observables are also compared with those obtained with a\nphenomenological Woods-Saxon potential model wave function. The experimental\ncore-valence neutron relative energy spectrum and dipole response along with\nother observables are well described by our calculations.\n"}
{"abstract": "  In this paper, we investigate problems which are dual to the unification\nproblem, namely the Fixed Point (FP) problem, Common Term (CT) problem and the\nCommon Equation (CE) problem for string rewriting systems. Our main motivation\nis computing fixed points in systems, such as loop invariants in programming\nlanguages. We show that the fixed point (FP) problem is reducible to the common\nterm problem. Our new results are: (i) the fixed point problem is undecidable\nfor finite convergent string rewriting systems whereas it is decidable in\npolynomial time for finite, convergent and dwindling string rewriting systems,\n(ii) the common term problem is undecidable for the class of dwindling string\nrewriting systems, and (iii) for the class of finite, monadic and convergent\nsystems, the common equation problem is decidable in polynomial time but for\nthe class of dwindling string rewriting systems, common equation problem is\nundecidable.\n"}
{"abstract": "  In this paper we analyze the singular set in the Stefan problem and prove the\nfollowing results:\n  - The singular set has parabolic Hausdorff dimension at most $n-1$.\n  - The solution admits a $C^\\infty$-expansion at all singular points, up to a\nset of parabolic Hausdorff dimension at most $n-2$.\n  - In $\\mathbb R^3$, the free boundary is smooth for almost every time $t$,\nand the set of singular times $\\mathcal S\\subset \\mathbb R$ has Hausdorff\ndimension at most $1/2$.\n  These results provide us with a refined understanding of the Stefan problem's\nsingularities and answer some long-standing open questions in the field.\n"}
{"abstract": "  Reconfigurable intelligent surface (RIS) is a promising technology for RF\nwireless power transfer (WPT) as it is capable of beamforming and beam focusing\nwithout using active and power-hungry components. In this paper, we propose a\nmulti-tile RIS beam scanning (MTBS) algorithm for powering up\ninternet-of-things (IoT) devices. Considering the hardware limitations of the\nIoT devices, the proposed algorithm requires only power information to enable\nthe beam focusing capability of the RIS. Specifically, we first divide the RIS\ninto smaller RIS tiles. Then, all RIS tiles and the phased array transmitter\nare iteratively scanned and optimized to maximize the receive power. We\nelaborately analyze the proposed algorithm and build a simulator to verify it.\nFurthermore, we have built a real-life testbed of RIS-aided WPT systems to\nvalidate the algorithm. The experimental results show that the proposed MTBS\nalgorithm can properly control the transmission phase of the transmitter and\nthe reflection phase of the RIS to focus the power at the receiver.\nConsequently, after executing the algorithm, about 20 dB improvement of the\nreceive power is achieved compared to the case that all unit cells of the RIS\nare in OFF state. By experiments, we confirm that the RIS with the MTBS\nalgorithm can greatly enhance the power transfer efficiency.\n"}
{"abstract": "  We show that in pool-based active classification without assumptions on the\nunderlying distribution, if the learner is given the power to abstain from some\npredictions by paying the price marginally smaller than the average loss $1/2$\nof a random guess, exponential savings in the number of label requests are\npossible whenever they are possible in the corresponding realizable problem. We\nextend this result to provide a necessary and sufficient condition for\nexponential savings in pool-based active classification under the model\nmisspecification.\n"}
{"abstract": "  We propose Skip-Convolutions to leverage the large amount of redundancies in\nvideo streams and save computations. Each video is represented as a series of\nchanges across frames and network activations, denoted as residuals. We\nreformulate standard convolution to be efficiently computed on residual frames:\neach layer is coupled with a binary gate deciding whether a residual is\nimportant to the model prediction,~\\eg foreground regions, or it can be safely\nskipped, e.g. background regions. These gates can either be implemented as an\nefficient network trained jointly with convolution kernels, or can simply skip\nthe residuals based on their magnitude. Gating functions can also incorporate\nblock-wise sparsity structures, as required for efficient implementation on\nhardware platforms. By replacing all convolutions with Skip-Convolutions in two\nstate-of-the-art architectures, namely EfficientDet and HRNet, we reduce their\ncomputational cost consistently by a factor of 3~4x for two different tasks,\nwithout any accuracy drop. Extensive comparisons with existing model\ncompression, as well as image and video efficiency methods demonstrate that\nSkip-Convolutions set a new state-of-the-art by effectively exploiting the\ntemporal redundancies in videos.\n"}
{"abstract": "  In this paper we investigate the \\emph{approximate string matching problem}\nwhen the allowed edit operations are \\emph{non-overlapping unbalanced\ntranslocations of adjacent factors}. Such kind of edit operations take place\nwhen two adjacent sub-strings of the text swap, resulting in a modified string.\nThe two involved substrings are allowed to be of different lengths.\n  Such large-scale modifications on strings have various applications. They are\namong the most frequent chromosomal alterations, accounted for 30\\% of all\nlosses of heterozygosity, a major genetic event causing inactivation of cancer\nsuppressor genes. In addition, among other applications, they are frequent\nmodifications accounted in musical or in natural language information\nretrieval. However, despite of their central role in so many fields of text\nprocessing, little attention has been devoted to the problem of matching\nstrings allowing for this kind of edit operation.\n  In this paper we present three algorithms for solving the problem, all of\nthem with a $\\bigO(nm^3)$ worst-case and a $\\bigO(m^2)$-space complexity, where\n$m$ and $n$ are the length of the pattern and of the text, respectively. % In\nparticular, our first algorithm is based on the dynamic-programming approach.\nOur second solution improves the previous one by making use of the Directed\nAcyclic Word Graph of the pattern. Finally our third algorithm is based on an\nalignment procedure. We also show that under the assumptions of equiprobability\nand independence of characters, our second algorithm has a\n$\\bigO(n\\log^2_{\\sigma} m)$ average time complexity, for an alphabet of size\n$\\sigma \\geq 4$.\n"}
{"abstract": "  Modern machine learning algorithms usually involve tuning multiple (from one\nto thousands) hyperparameters which play a pivotal role in terms of model\ngeneralizability. Black-box optimization and gradient-based algorithms are two\ndominant approaches to hyperparameter optimization while they have totally\ndistinct advantages. How to design a new hyperparameter optimization technique\ninheriting all benefits from both approaches is still an open problem. To\naddress this challenging problem, in this paper, we propose a new\nhyperparameter optimization method with zeroth-order hyper-gradients (HOZOG).\nSpecifically, we first exactly formulate hyperparameter optimization as an\nA-based constrained optimization problem, where A is a black-box optimization\nalgorithm (such as deep neural network). Then, we use the average zeroth-order\nhyper-gradients to update hyperparameters. We provide the feasibility analysis\nof using HOZOG to achieve hyperparameter optimization. Finally, the\nexperimental results on three representative hyperparameter (the size is from 1\nto 1250) optimization tasks demonstrate the benefits of HOZOG in terms of\nsimplicity, scalability, flexibility, effectiveness and efficiency compared\nwith the state-of-the-art hyperparameter optimization methods.\n"}
{"abstract": "  Our main goal is to compute the decomposition of arbitrary Kronecker powers\nof the Harmonics of $S_n$. To do this, we give a new way of decomposing the\ncharacter for the action of $S_n$ on polynomial rings with $k$ sets of $n$\nvariables. There are two aspects to this decomposition. The first is algebraic,\nin which formulas can be given for certain restrictions from $GL_n$ to $S_n$\noccurring in Schur-Weyl duality. The second is combinatorial. We give a\ngeneralization of the $comaj$ statistic on permutations which includes the\n$comaj$ statistic on standard tableaux. This statistic allows us to write a\ngeneralized principal evaluation for Schur functions and Gessel Fundamental\nquasisymmetric functions.\n"}
{"abstract": "  In this paper, a proactive dynamic spectrum sharing scheme between 4G and 5G\nsystems is proposed. In particular, a controller decides on the resource split\nbetween NR and LTE every subframe while accounting for future network states\nsuch as high interference subframes and multimedia broadcast single frequency\nnetwork (MBSFN) subframes. To solve this problem, a deep reinforcement learning\n(RL) algorithm based on Monte Carlo Tree Search (MCTS) is proposed. The\nintroduced deep RL architecture is trained offline whereby the controller\npredicts a sequence of future states of the wireless access network by\nsimulating hypothetical bandwidth splits over time starting from the current\nnetwork state. The action sequence resulting in the best reward is then\nassigned. This is realized by predicting the quantities most directly relevant\nto planning, i.e., the reward, the action probabilities, and the value for each\nnetwork state. Simulation results show that the proposed scheme is able to take\nactions while accounting for future states instead of being greedy in each\nsubframe. The results also show that the proposed framework improves\nsystem-level performance.\n"}
{"abstract": "  In this paper, we describe the birational structure of moduli spaces of rank\n2 logarithmic connections with the fixed trace connection. Our approach is to\nanalyze the underlying parabolic bundles and apparent singularities of the\nparabolic connections.\n"}
{"abstract": "  In this paper we consider an anisotropic Robin problem driven by the\n$p(x)$-Laplacian and a superlinear reaction. Applying variational tools along\nwith truncation and comparison techniques as well as critical groups, we prove\nthat the problem has at least five nontrivial smooth solutions to be ordered\nand with sign information: two positive, two negative and the fifth nodal.\n"}
{"abstract": "  Relations among fundamental invariants play an important role in algebraic\ngeometry. It is known that an $n$-dimensional variety of general type with nef\ncanonical divisor and canonical singularities, whose image $Y$ under the\ncanonical map is of maximal dimension, satisfies $K_X^n \\ge 2 (p_g-n)$. We\ninvestigate the very interesting extremal situation $K_X^n=2(p_g-n)$, which\nappears in a number of geometric situations. Since these extremal varieties are\nnatural higher dimensional analogues of Horikawa surfaces, we name them\nHorikawa varieties. These varieties have been previously dealt with inthe works\nof Fujita and Kobayashi. We carry out further studies of Horikawa varieties,\nproving new results on various geometric and topological issues concerning\nthem. In particular, we prove that the geometric genus of those Horikawa\nvarieties whose image under the canonical map is singular is bounded. We give\nan analogous result for polarized hyperelliptic subcanonical varieties, in\nparticular, for polarized Calabi-Yau and Fano varieties. The pleasing\nnumerology that emerges puts Horikawa's result on surfaces in a broader\nperspective. We obtain a structure theorem for Horikawa varieties and explore\ntheir pluriregularity. We use this to prove optimal results on projective\nnormality of pluricanonical linear systems. We study the fundamental groups of\nHorikawa varieties, showing that they are simply connected, even if $Y$ is\nsingular. We also prove results on deformations of Horikawa varieties, whose\nimplications on the moduli space make them the higher dimensional analogue of\ncurves of genus $2$.\n"}
{"abstract": "  We study the statistical mechanics of supercooled liquids when the system\nevolves at a temperature $T$ with a linear coupling of amplitude $\\epsilon$ to\nits overlap with a reference configuration of the same liquid sampled at a\ntemperature $T_0$. We use mean-field theory to fully characterize the influence\nof the reference temperature $T_0$, and we mainly study the case of a fixed,\nlow-$T_0$ value in computer simulations. We numerically investigate the\nextended phase diagram in the $(\\epsilon,T)$ plane of model glass-forming\nliquids in spatial dimensions $d=2$ and $d=3$, relying on umbrella sampling and\nreweighting techniques. For both $2d$ and $3d$ cases, a similar phenomenology\nwith nontrivial thermodynamic fluctuations of the overlap is observed at low\ntemperatures, but a detailed finite-size scaling analysis reveals qualitatively\ndistinct behaviors. We establish the existence of a first-order transition line\nfor nonzero $\\epsilon$ ending in a critical point in the universality class of\nthe random-field Ising model (RFIM) in $d=3$. In $d=2$ instead, no phase\ntransition exists in the thermodynamic limit at finite temperatures. Our\nresults show that glass-forming liquid samples of limited size display the\nthermodynamic fluctuations expected for finite systems undergoing a random\nfirst-order transition. They also establish the relevance of the physics of the\nRFIM for supercooled liquids, which in particular explains the qualitative\ndifference between $2d$ and $3d$ glass-formers.\n"}
{"abstract": "  We performed rheological measurements of the typical deep eutectic solvents\n(DESs) glyceline, ethaline, and reline in a very broad temperature and dynamic\nrange, extending from the low-viscosity to the high-viscosity\nsupercooled-liquid regime. We find that the mechanical compliance spectra can\nbe well described by the random free-energy barrier hopping model, while the\ndielectric spectra on the same materials involve significant contributions\narising from reorientational dynamics. The temperature-dependent viscosity and\nstructural relaxation time, revealing non-Arrhenius behavior typical for glassy\nfreezing, are compared to the ionic dc conductivity and relaxation times\ndetermined by broadband dielectric spectroscopy. For glyceline and ethaline we\nfind essentially identical temperature dependences for all dynamic quantities.\nThese findings point to a close coupling of the ionic and molecular\ntranslational and reorientational motions in these systems. However, for reline\nthe ionic charge transport appears decoupled from the structural and\nreorientational dynamics, following a fractional Walden rule. Especially, at\nlow temperatures the ionic conductivity in this DES is enhanced by about one\ndecade compared to expectations based on the temperature dependence of the\nviscosity. The results for all three DESs can be understood without invoking a\nrevolving-door mechanism previously considered as a possible charge-transport\nmechanism in DESs.\n"}
{"abstract": "  Gaussian processes (GPs) enable principled computation of model uncertainty,\nmaking them attractive for safety-critical applications. Such scenarios demand\nthat GP decisions are not only accurate, but also robust to perturbations. In\nthis paper we present a framework to analyse adversarial robustness of GPs,\ndefined as invariance of the model's decision to bounded perturbations. Given a\ncompact subset of the input space $T\\subseteq \\mathbb{R}^d$, a point $x^*$ and\na GP, we provide provable guarantees of adversarial robustness of the GP by\ncomputing lower and upper bounds on its prediction range in $T$. We develop a\nbranch-and-bound scheme to refine the bounds and show, for any $\\epsilon > 0$,\nthat our algorithm is guaranteed to converge to values $\\epsilon$-close to the\nactual values in finitely many iterations. The algorithm is anytime and can\nhandle both regression and classification tasks, with analytical formulation\nfor most kernels used in practice. We evaluate our methods on a collection of\nsynthetic and standard benchmark datasets, including SPAM, MNIST and\nFashionMNIST. We study the effect of approximate inference techniques on\nrobustness and demonstrate how our method can be used for interpretability. Our\nempirical results suggest that the adversarial robustness of GPs increases with\naccurate posterior estimation.\n"}
{"abstract": "  Modern vehicles, including connected vehicles and autonomous vehicles,\nnowadays involve many electronic control units connected through intra-vehicle\nnetworks to implement various functionalities and perform actions. Modern\nvehicles are also connected to external networks through vehicle-to-everything\ntechnologies, enabling their communications with other vehicles,\ninfrastructures, and smart devices. However, the improving functionality and\nconnectivity of modern vehicles also increase their vulnerabilities to\ncyber-attacks targeting both intra-vehicle and external networks due to the\nlarge attack surfaces. To secure vehicular networks, many researchers have\nfocused on developing intrusion detection systems (IDSs) that capitalize on\nmachine learning methods to detect malicious cyber-attacks. In this paper, the\nvulnerabilities of intra-vehicle and external networks are discussed, and a\nmulti-tiered hybrid IDS that incorporates a signature-based IDS and an\nanomaly-based IDS is proposed to detect both known and unknown attacks on\nvehicular networks. Experimental results illustrate that the proposed system\ncan detect various types of known attacks with 99.99% accuracy on the\nCAN-intrusion-dataset representing the intra-vehicle network data and 99.88%\naccuracy on the CICIDS2017 dataset illustrating the external vehicular network\ndata. For the zero-day attack detection, the proposed system achieves high\nF1-scores of 0.963 and 0.800 on the above two datasets, respectively. The\naverage processing time of each data packet on a vehicle-level machine is less\nthan 0.6 ms, which shows the feasibility of implementing the proposed system in\nreal-time vehicle systems. This emphasizes the effectiveness and efficiency of\nthe proposed IDS.\n"}
{"abstract": "  Subsampling is used in convolutional neural networks (CNNs) in the form of\npooling or strided convolutions, to reduce the spatial dimensions of feature\nmaps and to allow the receptive fields to grow exponentially with depth.\nHowever, it is known that such subsampling operations are not translation\nequivariant, unlike convolutions that are translation equivariant. Here, we\nfirst introduce translation equivariant subsampling/upsampling layers that can\nbe used to construct exact translation equivariant CNNs. We then generalise\nthese layers beyond translations to general groups, thus proposing group\nequivariant subsampling/upsampling. We use these layers to construct group\nequivariant autoencoders (GAEs) that allow us to learn low-dimensional\nequivariant representations. We empirically verify on images that the\nrepresentations are indeed equivariant to input translations and rotations, and\nthus generalise well to unseen positions and orientations. We further use GAEs\nin models that learn object-centric representations on multi-object datasets,\nand show improved data efficiency and decomposition compared to non-equivariant\nbaselines.\n"}
{"abstract": "  We investigate the electroweak phase transition in the real-singlet extension\nof the Standard Model at two-loop level, building upon existing one-loop\nstudies. We calculate the effective potential in the high-temperature\napproximation and detail the required resummations at two-loop order. In\ntypical strong-transition scenarios, we find deviations of order $20\\% - 50\\%$\nfrom one-loop results in transition strength and critical temperature for both\none- and two-step phase transitions. For extremely strong transitions, the\ndiscrepancy with one-loop predictions is even larger, presumably due to sizable\nscalar couplings in the tree-level potential. Along the way, we obtain a\ndimensionally-reduced effective theory applicable for non-perturbative lattice\nstudies of the model.\n"}
{"abstract": "  Micro-Doppler analysis has become increasingly popular in recent years owning\nto the ability of the technique to enhance classification strategies.\nApplications include recognising everyday human activities, distinguishing\ndrone from birds, and identifying different types of vehicles. However, noisy\ntime-frequency spectrograms can significantly affect the performance of the\nclassifier and must be tackled using appropriate denoising algorithms. In\nrecent years, deep learning algorithms have spawned many deep neural\nnetwork-based denoising algorithms. For these methods, noise modelling is the\nmost important part and is used to assist in training. In this paper, we\ndecompose the problem and propose a novel denoising scheme: first, a Generative\nAdversarial Network (GAN) is used to learn the noise distribution and\ncorrelation from the real-world environment; then, a simulator is used to\ngenerate clean Micro-Doppler spectrograms; finally, the generated noise and\nclean simulation data are combined as the training data to train a\nConvolutional Neural Network (CNN) denoiser. In experiments, we qualitatively\nand quantitatively analyzed this procedure on both simulation and measurement\ndata. Besides, the idea of learning from natural noise can be applied well to\nother existing frameworks and demonstrate greater performance than other noise\nmodels.\n"}
{"abstract": "  This paper is devoted to the study of the higher index theory of codimension\n$2$ submanifolds originated by Gromov-Lawson and Hanke-Pape-Schick. The first\nmain result is to construct the `codimension $2$ transfer' map from the\nHigson-Roe analytic surgery exact sequence of a manifold $M$ to that of its\ncodimension $2$ submanifold $N$ under some assumptions on homotopy groups. This\nmap sends the primary and secondary higher index invariants of $M$ to those of\n$N$. The second is to establish that the codimension 2 transfer map is adjoint\nto the co-transfer map in cyclic cohomology, defined by the cup product with a\ngroup cocycle. This relates the Connes-Moscovici higher index pairing and\nLott's higher $\\rho$-number of $M$ with those of $N$.\n"}
{"abstract": "  Circular Nim is a two-player impartial combinatorial game consisting of $n$\nstacks of tokens placed in a circle. A move consists of choosing $k$\nconsecutive stacks and taking at least one token from one or more of the\nstacks. The last player able to make a move wins. The question of interest is:\nWho can win from a given position if both players play optimally? In an\nimpartial combinatorial game, there are only two types of positions. An\n$\\mathcal{N}$-position is one from which the next player to move has a winning\nstrategy. A $\\mathcal{P}$-position is one from which the next player is bound\nto lose, no matter what moves s/he makes. Therefore, the question who wins is\nanswered by identifying the $\\mathcal{P}$-positions. We will prove results on\nthe structure of the $\\mathcal{P}$-positions for $n = 7$ and $k = 4$, extending\nknown results for other games in this family. The interesting feature of the\nset of $\\mathcal{P}$-positions of this game is that it splits into different\nsubsets, unlike the structure for the known games in this family.\n"}
{"abstract": "  Goal of this paper is to study positive semiclassical solutions of the\nnonlinear Schr\\\"odinger equation $$ \\varepsilon^{2s}(- \\Delta)^s u+ V(x) u=\nf(u), \\quad x \\in \\mathbb{R}^N,$$ where $s \\in (0,1)$, $N \\geq 2$, $V \\in\nC(\\mathbb{R}^N,\\mathbb{R})$ is a positive potential and $f$ is assumed critical\nand satisfying general Berestycki-Lions type conditions. We obtain existence\nand multiplicity for $\\varepsilon>0$ small, where the number of solutions is\nrelated to the cup-length of a set of local minima of $V$. Furthermore, these\nsolutions are proved to concentrate in the potential well, exhibiting a\npolynomial decay. We highlight that these results are new also in the limiting\nlocal setting $s=1$ and $N\\geq 3$, with an exponential decay of the solutions.\n"}
{"abstract": "  TV serials are a popular source of entertainment. The ongoing COVID19\nlockdown has a high probability of degrading the publics mental health. The\nGovernment of India started the retelecast of yesteryears popular TV serials on\npublic broadcaster Doordarshan from 28th March 2020 to 31st July 2020. Tweets\ncorresponding to the Doordarshan hashtag were mined to create a dataset. The\nexperiment aims to analyze the publics response to the retelecast of TV serials\nby calculating the sentiment score of the tweet dataset. Datasets mean\nsentiment score of 0.65 and high share 64.58% of positive tweets signifies the\nacceptance of Doordarshans retelecast decision. The sentiment analysis result\nalso reflects the positive state of mind of the public.\n"}
{"abstract": "  We consider optimal control of an elliptic two-point boundary value problem\ngoverned by functions of bounded variation (BV). The cost functional is\ncomposed of a tracking term for the state and the BV-seminorm of the control.\nWe use the mixed formulation for the state equation together with the\nvariational discretization approach, where we use the classical lowest order\nRaviart-Thomas finite elements for the state equation. Consequently the\nvariational discrete control is a piecewise constant function over the finite\nelement grid. We prove error estimates for the variational discretization\napproach in combination with the mixed formulation of the state equation and\nconfirm our analytical findings with numerical experiments.\n"}
{"abstract": "  When studied by small-angle neutron scattering the vortex lattice (VL) in\nUPt3 undergoes a gradual disordering as a function of time due to 235U fission.\nThis temporarily heats regions of the sample above the critical temperature,\nwhere, upon re-cooling, the vortices remain in a quenched vortex glass state.\nThe disordering rate is proportional to the magnetic field, suggesting that it\nis governed by collective VL properties such as the elastic moduli. An ordered\nVL can be re-formed by applying a small field oscillation, showing that the\nfission does not cause significant radiation damage to the UPt3 crystals, even\nafter long exposure.\n"}
{"abstract": "  In response to the COVID-19 pandemic, Bluetooth-based contact tracing has\nbeen deployed in many countries with the help of the developers of smartphone\noperating systems that provide APIs for privacy-preserving exposure\nnotification. However, it has been assumed by the design that the OS\ndevelopers, smartphone vendors, or governments will not violate people's\nprivacy. We propose a privacy-preserving exposure notification under situations\nwhere none of the middle entities can be trusted. We believe that it can be\nachieved with small changes to the existing mechanism: random numbers are\ngenerated on the application side instead of the OS, and the positive test\nresults are reported to a public ledger (e.g. blockchain) rather than to a\ngovernment server, with endorsements from the medical institutes with blind\nsignatures. We also discuss how to incentivize the peer-to-peer maintenance of\nthe public ledger if it should be newly built. We show that the level of\nverifiability is much higher with our proposed design if a consumer group were\nto verify the privacy protections of the deployed systems. We believe that this\nwill allow for safer contact tracing, and contribute to healthier lifestyles\nfor citizens who may want to or have to go out under pandemic situations.\n"}
{"abstract": "  We have studied systematically microscopic properties of a quantum vortex in\nneutron matter at finite temperatures and densities corresponding to different\nlayers of the inner crust of a neutron star. To this end and in preparation of\nfuture simulations of the vortex dynamics, we have carried out fully\nself-consistent 3D Hartree-Fock-Bogoliubov calculations, using one of the\nlatest nuclear energy-density functionals from the Brussels-Montreal family,\nwhich has been developed specifically for applications to neutron superfluidity\nin neutron-star crusts. By analyzing the flow around the vortex, we have\ndetermined the effective radius relevant for the vortex filament model. We have\nalso calculated the specific heat in the presence of the quantum vortex and\nhave shown that it is substantially larger than for a uniform system at low\ntemperatures. The low temperature limit of the specific heat has been\nidentified as being determined by Andreev states inside the vortex core. We\nhave shown that the specific heat in this limit does not scale linearly with\ntemperature. The typical energy scale associated with Andreev states is defined\nby the minigap, which we have extracted for various neutron-matter densities.\nOur results suggest that vortices may be spin-polarized in the crust of\nmagnetars. Finally, we have obtained a lower bound for the specific heat of a\ncollection of vortices with given surface density, taking into account both the\ncontributions from the vortex core states and from the hydrodynamic flow.\n"}
{"abstract": "  Multi-modal sentiment analysis plays an important role for providing better\ninteractive experiences to users. Each modality in multi-modal data can provide\ndifferent viewpoints or reveal unique aspects of a user's emotional state. In\nthis work, we use text, audio and visual modalities from MOSI dataset and we\npropose a novel fusion technique using a multi-head attention LSTM network.\nFinally, we perform a classification task and evaluate its performance.\n"}
{"abstract": "  Managing the data for Information Retrieval (IR) experiments can be\nchallenging. Dataset documentation is scattered across the Internet and once\none obtains a copy of the data, there are numerous different data formats to\nwork with. Even basic formats can have subtle dataset-specific nuances that\nneed to be considered for proper use. To help mitigate these challenges, we\nintroduce a new robust and lightweight tool (ir_datasets) for acquiring,\nmanaging, and performing typical operations over datasets used in IR. We\nprimarily focus on textual datasets used for ad-hoc search. This tool provides\nboth a Python and command line interface to numerous IR datasets and\nbenchmarks. To our knowledge, this is the most extensive tool of its kind.\nIntegrations with popular IR indexing and experimentation toolkits demonstrate\nthe tool's utility. We also provide documentation of these datasets through the\nir_datasets catalog: https://ir-datasets.com/. The catalog acts as a hub for\ninformation on datasets used in IR, providing core information about what data\neach benchmark provides as well as links to more detailed information. We\nwelcome community contributions and intend to continue to maintain and grow\nthis tool.\n"}
{"abstract": "  This paper uses data-driven operator theoretic approaches to explore the\nglobal phase space of a dynamical system. We defined conditions for discovering\nnew invariant subspaces in the state space of a dynamical system starting from\nan invariant subspace based on the spectral properties of the Koopman operator.\nWhen the system evolution is known locally in several invariant subspaces in\nthe state space of a dynamical system, a phase space stitching result is\nderived that yields the global Koopman operator. Additionally, in the case of\nequivariant systems, a phase space stitching result is developed to identify\nthe global Koopman operator using the symmetry properties between the invariant\nsubspaces of the dynamical system and time-series data from any one of the\ninvariant subspaces. Finally, these results are extended to topologically\nconjugate dynamical systems; in particular, the relation between the Koopman\ntuple of topologically conjugate systems is established. The proposed results\nare demonstrated on several second-order nonlinear dynamical systems including\na bistable toggle switch. Our method elucidates a strategy for designing\ndiscovery experiments: experiment execution can be done in many steps, and\nmodels from different invariant subspaces can be combined to approximate the\nglobal Koopman operator.\n"}
{"abstract": "  Optimization via simulation (OvS) procedures that assume the simulation\ninputs are generated from the real-world distributions are subject to the risk\nof selecting a suboptimal solution when the distributions are substituted with\ninput models estimated from finite real-world data -- known as input model\nrisk. Focusing on discrete OvS, this paper proposes a new Bayesian framework\nfor analyzing input model risk of implementing an arbitrary solution, $x$,\nwhere uncertainty about the input models is captured by a posterior\ndistribution. We define the $\\alpha$-level risk set of solution $x$ as the set\nof solutions whose expected performance is better than $x$ by a practically\nmeaningful margin $(>\\delta)$ given common input models with significant\nprobability ($>\\alpha$) under the posterior distribution. The user-specified\nparameters, $\\delta$ and $\\alpha$, control robustness of the procedure to the\ndesired level as well as guards against unnecessary conservatism. An empty risk\nset implies that there is no practically better solution than $x$ with\nsignificant probability even though the real-world input distributions are\nunknown. For efficient estimation of the risk set, the conditional mean\nperformance of a solution given a set of input distributions is modeled as a\nGaussian process (GP) that takes the solution-distributions pair as an input.\nIn particular, our GP model allows both parametric and nonparametric input\nmodels. We propose the sequential risk set inference procedure that estimates\nthe risk set and selects the next solution-distributions pair to simulate using\nthe posterior GP at each iteration. We show that simulating the pair expected\nto change the risk set estimate the most in the next iteration is the\nasymptotic one-step optimal sampling rule that minimizes the number of\nincorrectly classified solutions, if the procedure runs without stopping.\n"}
{"abstract": "  Many HPC applications suffer from a bottleneck in the shared caches,\ninstruction execution units, I/O or memory bandwidth, even though the remaining\nresources may be underutilized. It is hard for developers and runtime systems\nto ensure that all critical resources are fully exploited by a single\napplication, so an attractive technique for increasing HPC system utilization\nis to colocate multiple applications on the same server. When applications\nshare critical resources, however, contention on shared resources may lead to\nreduced application performance.\n  In this paper, we show that server efficiency can be improved by first\nmodeling the expected performance degradation of colocated applications based\non measured hardware performance counters, and then exploiting the model to\ndetermine an optimized mix of colocated applications. This paper presents a new\nintelligent resource manager and makes the following contributions: (1) a new\nmachine learning model to predict the performance degradation of colocated\napplications based on hardware counters and (2) an intelligent scheduling\nscheme deployed on an existing resource manager to enable application\nco-scheduling with minimum performance degradation. Our results show that our\napproach achieves performance improvements of 7% (avg) and 12% (max) compared\nto the standard policy commonly used by existing job managers.\n"}
{"abstract": "  We propose a new method for discriminating sub-micron nuclear recoil tracks\nfrom an instrumental background in fine-grain nuclear emulsions used in the\ndirectional dark matter search. The proposed method is based on the Deep\nLearning approach and uses a 3D Convolutional Neural Network architecture with\nparameters optimised by Bayesian search. Unlike previous studies focused on\nextracting the directional information, we focus on the signal/background\nseparation exploiting the polarisation dependence of the Localised Surface\nPlasmon Resonance phenomenon. Comparing the proposed method with the\nconventional cut-based approach shows a significant boost in the rejection\npower while keeping the signal efficiency at the same level.\n"}
{"abstract": "  We present the analysis of XMM-Newton European Photon Imaging Camera (EPIC)\nobservations of the nova shell IPHASX J210204.7$+$471015. We detect X-ray\nemission from the progenitor binary star with properties that resemble those of\nunderluminous intermediate polars such as DQ Her: an X-ray-emitting plasma with\ntemperature of $T_\\mathrm{X}=(6.4\\pm3.1)\\times10^{6}$ K, a non-thermal X-ray\ncomponent, and an estimated X-ray luminosity of $L_\\mathrm{X}=10^{30}$ erg\ns$^{-1}$. Time series analyses unveil the presence of two periods, the dominant\nwith a period of $2.9\\pm0.2$ hr, which might be attributed to the spin of the\nwhite dwarf, and a secondary of $4.5\\pm0.6$ hr that is in line with the orbital\nperiod of the binary system derived from optical observations. We do not detect\nextended X-ray emission as in other nova shells probably due to its relatively\nold age (130-170 yr) or to its asymmetric disrupted morphology which is\nsuggestive of explosion scenarios different to the symmetric ones assumed in\navailable numerical simulations of nova explosions.\n"}
{"abstract": "  Thermally conductive nanopapers fabricated from graphene and related\nmaterials are currently showing a great potential in thermal management\napplications. However, thermal contacts between conductive plates represent the\nbottleneck for thermal conductivity of nanopapers prepared in the absence of a\nhigh temperature step for graphitization. In this work, the problem of\nineffective thermal contacts is addressed by the use of bifunctional\npolyaromatic molecules designed to drive self-assembly of graphite nanoplates\n(GnP) and establish thermal bridges between them. To preserve the high\nconductivity associated to defect-free sp2 structure, non-covalent\nfunctionalization with bispyrene compounds, synthesised on purpose with\nvariable tethering chain length, was exploited. Pyrene terminal groups granted\nfor a strong {\\pi}-{\\pi} interaction with graphene surface, as demonstrated by\nUV-Vis, fluorescence and Raman spectroscopies. Bispyrene molecular junctions\nbetween GnP were found to control GnP organization and orientation within the\nnanopaper, delivering significant enhancement in both in-plane and cross-plane\nthermal diffusivity. Finally, nanopapers were validated as heat spreader\ndevices for electronic components, evidencing comparable or better thermal\ndissipation performance than conventional Cu foil, while delivering over 90%\nweight reduction.\n"}
{"abstract": "  Power grid parameter estimation involves the estimation of unknown\nparameters, such as inertia and damping coefficients, using observed dynamics.\nIn this work, we present a comparison of data-driven algorithms for the power\ngrid parameter estimation problem. First, we propose a new algorithm to solve\nthe parameter estimation problem based on the Sparse Identification of\nNonlinear Dynamics (SINDy) approach, which uses linear regression to infer the\nparameters that best describe the observed data. We then compare its\nperformance against two benchmark algorithms, namely, the unscented Kalman\nfilter (UKF) approach and the physics-informed neural networks (PINN) approach.\nWe perform extensive simulations on IEEE bus systems to examine the performance\nof the aforementioned algorithms. Our results show that the SINDy algorithm\noutperforms the PINN and UKF algorithms in being able to accurately estimate\nthe power grid parameters over a wide range of system parameters (including\nhigh and low inertia systems). Moreover, it is extremely efficient\ncomputationally and so takes significantly less time than the PINN algorithm,\nthus making it suitable for real-time parameter estimation.\n"}
{"abstract": "  As multi-robot systems (MRS) are widely used in various tasks such as natural\ndisaster response and social security, people enthusiastically expect an MRS to\nbe ubiquitous that a general user without heavy training can easily operate.\nHowever, humans have various preferences on balancing between task performance\nand safety, imposing different requirements onto MRS control. Failing to comply\nwith preferences makes people feel difficult in operation and decreases human\nwillingness of using an MRS. Therefore, to improve social acceptance as well as\nperformance, there is an urgent need to adjust MRS behaviors according to human\npreferences before triggering human corrections, which increases cognitive\nload. In this paper, a novel Meta Preference Learning (MPL) method was\ndeveloped to enable an MRS to fast adapt to user preferences. MPL based on meta\nlearning mechanism can quickly assess human preferences from limited\ninstructions; then, a neural network based preference model adjusts MRS\nbehaviors for preference adaption. To validate method effectiveness, a task\nscenario \"An MRS searches victims in an earthquake disaster site\" was designed;\n20 human users were involved to identify preferences as \"aggressive\", \"medium\",\n\"reserved\"; based on user guidance and domain knowledge, about 20,000\npreferences were simulated to cover different operations related to \"task\nquality\", \"task progress\", \"robot safety\". The effectiveness of MPL in\npreference adaption was validated by the reduced duration and frequency of\nhuman interventions.\n"}
{"abstract": "  In this work, we address the problem of cross-view geo-localization, which\nestimates the geospatial location of a street view image by matching it with a\ndatabase of geo-tagged aerial images. The cross-view matching task is extremely\nchallenging due to drastic appearance and geometry differences across views.\nUnlike existing methods that predominantly fall back on CNN, here we devise a\nnovel evolving geo-localization Transformer (EgoTR) that utilizes the\nproperties of self-attention in Transformer to model global dependencies, thus\nsignificantly decreasing visual ambiguities in cross-view geo-localization. We\nalso exploit the positional encoding of Transformer to help the EgoTR\nunderstand and correspond geometric configurations between ground and aerial\nimages. Compared to state-of-the-art methods that impose strong assumption on\ngeometry knowledge, the EgoTR flexibly learns the positional embeddings through\nthe training objective and hence becomes more practical in many real-world\nscenarios. Although Transformer is well suited to our task, its vanilla\nself-attention mechanism independently interacts within image patches in each\nlayer, which overlooks correlations between layers. Instead, this paper propose\na simple yet effective self-cross attention mechanism to improve the quality of\nlearned representations. The self-cross attention models global dependencies\nbetween adjacent layers, which relates between image patches while modeling how\nfeatures evolve in the previous layer. As a result, the proposed self-cross\nattention leads to more stable training, improves the generalization ability\nand encourages representations to keep evolving as the network goes deeper.\nExtensive experiments demonstrate that our EgoTR performs favorably against\nstate-of-the-art methods on standard, fine-grained and cross-dataset cross-view\ngeo-localization tasks.\n"}
{"abstract": "  We propose and analyze a randomized zeroth-order approach based on\napproximating the exact gradient byfinite differences computed in a set of\northogonal random directions that changes with each iteration. A number\nofpreviously proposed methods are recovered as special cases including\nspherical smoothing, coordinate descent, as wellas discretized gradient\ndescent. Our main contribution is proving convergence guarantees as well as\nconvergence ratesunder different parameter choices and assumptions. In\nparticular, we consider convex objectives, but also possiblynon-convex\nobjectives satisfying the Polyak-{\\L}ojasiewicz (PL) condition. Theoretical\nresults are complemented andillustrated by numerical experiments.\n"}
{"abstract": "  We demonstrate supermode-based second harmonic generation in an integrated\nnonlinear interferometer made of linear and nonlinear directional couplers. We\nuse a fully-fibered pump shaper to demonstrate second harmonic generation\npumped by the symmetric or anti-symmetric fundamental spatial modes. The\nselection of the pumping mode and thus of a specific SHG spectral profile is\nachieved through the selection of the fundamental wavelength and via a robust\nphase setting scheme. We use two methods: either post-selecting or actively\nsetting the pumping mode. Such a modal phase matching paves the way for\nclassical and quantum applications of coupled nonlinear photonic circuits,\nwhere multimode excitation, encoding and detection are a route for multiplexing\nand scaling up light-processing.\n"}
{"abstract": "  We propose an analytical framework based on stochastic geometry (SG)\nformulations to estimate a radar's detection performance under generalized\ndiscrete clutter conditions. We model the spatial distribution of discrete\nclutter scatterers as a homogeneous Poisson point process and the radar\ncross-section of each extended scatterer as a random variable of the Weibull\ndistribution. Using this framework, we derive a metric called the radar\ndetection coverage probability as a function of radar parameters such as\ntransmitted power, system noise temperature and radar bandwidth; and clutter\nparameters such as clutter density and mean clutter cross-section. We derive\nthe optimum radar bandwidth for maximizing this metric under noisy and\ncluttered conditions. We also derive the peak transmitted power beyond which\nthere will be no discernible improvement in radar detection performance due to\nclutter limited conditions. When both transmitted power and bandwidth are\nfixed, we show how the detection threshold can be optimized for best\nperformance. We experimentally validate the SG results with a hybrid of Monte\nCarlo and full wave electromagnetic solver based simulations using finite\ndifference time domain (FDTD) techniques.\n"}
{"abstract": "  Survival analysis is a challenging variation of regression modeling because\nof the presence of censoring, where the outcome measurement is only partially\nknown, due to, for example, loss to follow up. Such problems come up frequently\nin medical applications, making survival analysis a key endeavor in\nbiostatistics and machine learning for healthcare, with Cox regression models\nbeing amongst the most commonly employed models. We describe a new approach for\nsurvival analysis regression models, based on learning mixtures of Cox\nregressions to model individual survival distributions. We propose an\napproximation to the Expectation Maximization algorithm for this model that\ndoes hard assignments to mixture groups to make optimization efficient. In each\ngroup assignment, we fit the hazard ratios within each group using deep neural\nnetworks, and the baseline hazard for each mixture component\nnon-parametrically.\n  We perform experiments on multiple real world datasets, and look at the\nmortality rates of patients across ethnicity and gender. We emphasize the\nimportance of calibration in healthcare settings and demonstrate that our\napproach outperforms classical and modern survival analysis baselines, both in\nterms of discriminative performance and calibration, with large gains in\nperformance on the minority demographics.\n"}
{"abstract": "  In traditional models only an order one fraction of energy is transferred\nfrom the inflaton to radiation through nonperturbative resonance production in\npreheating immediately after inflation, due to backreaction effects. We propose\na particle production mechanism that could improve the depletion of the\ninflaton energy density by up to four orders of magnitude. The improvement\ncomes from the fast perturbative decays of resonantly produced daughter\nparticles. They act as a \"spillway\" to drain these daughter particles, reducing\ntheir backreaction on the inflaton and keeping the resonant production\neffective for a longer period. Thus we dub the scenario \"spillway preheating\".\nWe also show that the fraction of energy density remaining in the inflaton has\na simple inverse power-law scaling in the scenario. In general, spillway\npreheating is a much more efficient energy dissipation mechanism, which may\nhave other applications in model building for particle physics.\n"}
{"abstract": "  View synthesis aims to produce unseen views from a set of views captured by\ntwo or more cameras at different positions. This task is non-trivial since it\nis hard to conduct pixel-level matching among different views. To address this\nissue, most existing methods seek to exploit the geometric information to match\npixels. However, when the distinct cameras have a large baseline (i.e., far\naway from each other), severe geometry distortion issues would occur and the\ngeometric information may fail to provide useful guidance, resulting in very\nblurry synthesized images. To address the above issues, in this paper, we\npropose a novel deep generative model, called Self-Consistent Generative\nNetwork (SCGN), which synthesizes novel views from the given input views\nwithout explicitly exploiting the geometric information. The proposed SCGN\nmodel consists of two main components, i.e., a View Synthesis Network (VSN) and\na View Decomposition Network (VDN), both employing an Encoder-Decoder\nstructure. Here, the VDN seeks to reconstruct input views from the synthesized\nnovel view to preserve the consistency of view synthesis. Thanks to VDN, SCGN\nis able to synthesize novel views without using any geometric rectification\nbefore encoding, making it easier for both training and applications. Finally,\nadversarial loss is introduced to improve the photo-realism of novel views.\nBoth qualitative and quantitative comparisons against several state-of-the-art\nmethods on two benchmark tasks demonstrated the superiority of our approach.\n"}
{"abstract": "  We show that the set $\\ell^{\\infty}\\setminus c_0$ is dense-lineable and\ndense-algebrable answering a question posed by Nestoridis and complementing a\nresult by Garc\\'ia-Pacheco, Mart\\'in and Seoane-Sep\\'ulveda.\n"}
{"abstract": "  We report our progress on the project for solving larger scale quadratic\nassignment problems (QAPs). Our main approach to solve large scale NP-hard\ncombinatorial optimization problems such as QAPs is a parallel branch-and-bound\nmethod efficiently implemented on a powerful computer system using the Ubiquity\nGenerator (UG) framework that can utilize more than 100,000 cores. Lower\nbounding procedures incorporated in the branch-and-bound method play a crucial\nrole in solving the problems. For a strong lower bounding procedure, we employ\nthe Lagrangian doubly nonnegative (DNN) relaxation and the Newton-bracketing\nmethod developed by the authors' group. In this report, we describe some basic\ntools used in the project including the lower bounding procedure and branching\nrules, and present some preliminary numerical results.\n  Our next target problem is QAPs with dimension at least 50, as we have\nsucceeded to solve tai30a and sko42 from QAPLIB for the first time.\n"}
{"abstract": "  As a natural language generation task, it is challenging to generate\ninformative and coherent review text. In order to enhance the informativeness\nof the generated text, existing solutions typically learn to copy entities or\ntriples from knowledge graphs (KGs). However, they lack overall consideration\nto select and arrange the incorporated knowledge, which tends to cause text\nincoherence.\n  To address the above issue, we focus on improving entity-centric coherence of\nthe generated reviews by leveraging the semantic structure of KGs. In this\npaper, we propose a novel Coherence Enhanced Text Planning model (CETP) based\non knowledge graphs (KGs) to improve both global and local coherence for review\ngeneration. The proposed model learns a two-level text plan for generating a\ndocument: (1) the document plan is modeled as a sequence of sentence plans in\norder, and (2) the sentence plan is modeled as an entity-based subgraph from\nKG. Local coherence can be naturally enforced by KG subgraphs through\nintra-sentence correlations between entities. For global coherence, we design a\nhierarchical self-attentive architecture with both subgraph- and node-level\nattention to enhance the correlations between subgraphs. To our knowledge, we\nare the first to utilize a KG-based text planning model to enhance text\ncoherence for review generation. Extensive experiments on three datasets\nconfirm the effectiveness of our model on improving the content coherence of\ngenerated texts.\n"}
{"abstract": "  We present high-resolution optical images from the Hubble Space Telescope,\nX-ray images from the Chandra X-ray Observatory, and optical spectra from the\nNordic Optical Telescope for a newly-discovered galaxy cluster, CHIPS1911+4455,\nat z=0.485+/-0.005. CHIPS1911+4455 was discovered in the Clusters Hiding in\nPlain Sight (CHiPS) survey, which sought to discover galaxy clusters with\nextreme central galaxies that were misidentified as isolated X-ray point\nsources in the ROSAT All-Sky Survey. With new Chandra X-ray observations, we\nfind the core (r=10 kpc) entropy to be 17+2-9 keV cm^2, suggesting a strong\ncool core, which are typically found at the centers of relaxed clusters.\nHowever, the large-scale morphology of CHIPS1911+4455 is highly asymmetric,\npointing to a more dynamically active and turbulent cluster. Furthermore, the\nHubble images reveal a massive, filamentary starburst near the brightest\ncluster galaxy (BCG). We measure the star formation rate for the BCG to be\n140--190 Msun/yr, which is one of the highest rates measured in a central\ncluster galaxy to date. One possible scenario for CHIPS1911+4455 is that the\ncool core was displaced during a major merger and rapidly cooled, with cool,\nstar-forming gas raining back toward the core. This unique system is an\nexcellent case study for high-redshift clusters, where such phenomena are\nproving to be more common. Further studies of such systems will drastically\nimprove our understanding of the relation between cluster mergers and cooling,\nand how these fit in the bigger picture of active galactic nuclei (AGN)\nfeedback.\n"}
{"abstract": "  The membrane curvature of cells and intracellular compartments continuously\nadapts to enable cells to perform vital functions, from cell division to signal\ntrafficking. Understanding how membrane geometry affects these processes in\nvivo is challenging because of the membrane complexity as well as the short\ntime and small length scales involved. By contrast, in vitro model membranes\nwith engineered curvature provide a versatile platform for this investigation\nand applications to biosensing and biocomputing. However, a general route to\nthe fabrication of lipid membranes with prescribed curvature and high spatial\nresolution is still missing. Here, we present a strategy that overcomes these\nchallenges and achieve lipid membranes with designed shape by combining 3D\nmicro-printing and replica-molding lithography to create scaffolds with\nvirtually any geometry and high spatial resolution. The resulting supported\nlipid membranes are homogeneous, fluid, and can form chemically distinct lipid\ndomains. These features are essential for understanding curvature-dependent\ncellular processes and developing programmable bio-interfaces for living cells\nand nanostructures.\n"}
{"abstract": "  In the last twenty years, there have been significant advances in the study\nof the blow-up phenomenon for the critical generalized Korteweg-de Vries\nequation, including the determination of sufficient conditions for blowup, the\nstability of blowup in a refined topology and the classification of minimal\nmass blowup. Exotic blow-up solutions with a continuum of blow-up rates and\nmulti-point blow-up solutions were also constructed. However, all these\nresults, as well as numerical simulations, involve the bubbling of a solitary\nwave going at infinity at the blow-up time, which means that the blow-up\ndynamics and the residue are eventually uncoupled. Even at the formal level,\nthere was no indication whether blowup at a finite point could occur for this\nequation. In this article, we answer this question by constructing solutions\nthat blow up in finite time under the form of a single-bubble concentrating the\nground state at a finite point with an unforeseen blow-up rate. Finding a\nblow-up rate intermediate between the self-similar rate and other rates\npreviously known also reopens the question of which blow-up rates are actually\npossible for this equation.\n"}
{"abstract": "  In this paper we study the minimal faithful permutation representations of\n$SL_n(\\mathbb F_q)$ and $GL_n(\\mathbb F_q)$.\n"}
{"abstract": "  In this paper we study the phonon's effect on the position of the 1s\nexcitonic resonance of the fundamental absorption transition line in\ntwo-dimensional transition metal dichalcogenides. We apply our theory to\nWS$_{2}$a two-dimensional material where the shift in absorption peak position\nhas been measured as a function of temperature. The theory is composed of two\ningredients only: i) the effect of longitudinal optical phonons on the\nabsorption peak position, which we describe with second order perturbation\ntheory; ii) the effect of phonons on the value of the single particle energy\ngap, which we describe with the Huang Rhys model. Our results show an excellent\nagreement with the experimentally measured shift of the absorption peak with\nthe temperature.\n"}
{"abstract": "  The automatic recognition of food on images has numerous interesting\napplications, including nutritional tracking in medical cohorts. The problem\nhas received significant research attention, but an ongoing public benchmark to\ndevelop open and reproducible algorithms has been missing. Here, we report on\nthe setup of such a benchmark using publicly available food images sourced\nthrough the mobile MyFoodRepo app. Through four rounds, the benchmark released\nthe MyFoodRepo-273 dataset constituting 24,119 images and a total of 39,325\nsegmented polygons categorized in 273 different classes. Models were evaluated\non private tests sets from the same platform with 5,000 images and 7,865\nannotations in the final round. Top-performing models on the 273 food\ncategories reached a mean average precision of 0.568 (round 4) and a mean\naverage recall of 0.885 (round 3). We present experimental validation of round\n4 results, and discuss implications of the benchmark setup designed to increase\nthe size and diversity of the dataset for future rounds.\n"}
{"abstract": "  The traditional approach in FL tries to learn a single global model\ncollaboratively with the help of many clients under the orchestration of a\ncentral server. However, learning a single global model might not work well for\nall clients participating in the FL under data heterogeneity. Therefore, the\npersonalization of the global model becomes crucial in handling the challenges\nthat arise with statistical heterogeneity and the non-IID distribution of data.\nUnlike prior works, in this work we propose a new approach for obtaining a\npersonalized model from a client-level objective. This further motivates all\nclients to participate in federation even under statistical heterogeneity in\norder to improve their performance, instead of merely being a source of data\nand model training for the central server. To realize this personalization, we\nleverage finding a small subnetwork for each client by applying hybrid pruning\n(combination of structured and unstructured pruning), and unstructured pruning.\nThrough a range of experiments on different benchmarks, we observed that the\nclients with similar data (labels) share similar personal parameters. By\nfinding a subnetwork for each client ...\n"}
{"abstract": "  In the Madelung-Bohm approach to quantum mechanics, we consider a (time\ndependent) phase that depends quadratically on position and show that it leads\nto a Bohm potential that corresponds to a time dependent harmonic oscillator,\nprovided the time dependent term in the phase obeys an Ermakov equation.\n"}
{"abstract": "  Exponential random graphs are important to model the structure of real-world\ncomplex networks. Here we solve the two-star model with degree-degree\ncorrelations in the sparse regime. The model constraints the average\ncorrelation between the degrees of adjacent nodes (nearest neighbors) and\nbetween the degrees at the end-points of two-stars (next nearest neighbors). We\ncompute exactly the network free energy and show that this model undergoes a\nfirst-order transition to a condensed phase. For non-negative degree\ncorrelations between next nearest neighbors, the degree distribution inside the\ncondensed phase has a single peak at the largest degree, while for negative\ndegree correlations between next nearest neighbors the condensed phase is\ncharacterized by a bimodal degree distribution. We calculate the degree\nassortativities and show they are non-monotonic functions of the model\nparameters, with a discontinuous behavior at the first-order transition. The\nfirst-order critical line terminates at a second-order critical point, whose\nlocation in the phase diagram can be accurately determined. Our results can\nhelp to develop more detailed models of complex networks with correlated\ndegrees.\n"}
{"abstract": "  We discuss the non-Abelian artificial magnetic monopoles associated with\n$n$-level energy crossing in quantum systems. We found that hidden symmetries\nreveal themselves as observables such as spin, charge, and other physical\ndegrees of freedom. We illustrated our results on concrete examples of two and\nthree energy-level crossing. Our results can be useful for modeling of various\nphenomena in physical and biological systems.\n"}
{"abstract": "  Detector-based spectral computed tomography is a recent dual-energy CT (DECT)\ntechnology that offers the possibility of obtaining spectral information. From\nthis spectral data, different types of images can be derived, amongst others\nvirtual monoenergetic (monoE) images. MonoE images potentially exhibit\ndecreased artifacts, improve contrast, and overall contain lower noise values,\nmaking them ideal candidates for better delineation and thus improved\ndiagnostic accuracy of vascular abnormalities.\n  In this paper, we are training convolutional neural networks~(CNN) that can\nemulate the generation of monoE images from conventional single energy CT\nacquisitions. For this task, we investigate several commonly used\nimage-translation methods. We demonstrate that these methods while creating\nvisually similar outputs, lead to a poorer performance when used for automatic\nclassification of pulmonary embolism (PE). We expand on these methods through\nthe use of a multi-task optimization approach, under which the networks achieve\nimproved classification as well as generation results, as reflected by PSNR and\nSSIM scores. Further, evaluating our proposed framework on a subset of the\nRSNA-PE challenge data set shows that we are able to improve the Area under the\nReceiver Operating Characteristic curve (AuROC) in comparison to a na\\\"ive\nclassification approach from 0.8142 to 0.8420.\n"}
{"abstract": "  Let $X$ be an abstract orientable not necessarily compact CR manifold of\ndimension $2n+1$, $n\\geq1$, and let $L^k$ be the $k$-th tensor power of a CR\ncomplex line bundle $L$ over $X$. Suppose that condition $Y(q)$ holds at each\npoint of $X$, we establish asymptotics of the heat kernel of Kohn Laplacian\nwith values in $L^k$. As an application, we give a heat kernel proof of Morse\ninequalities on compact CR manifolds. When $X$ admits a transversal CR $\\mathbb\nR$-action, we also establish asymptotics of the $\\mathbb R$-equivariant heat\nkernel of Kohn Laplacian with values in $L^k$. As an application, we get\n$\\mathbb R$-equivariant Morse inequalities on compact CR manifolds with\ntransversal CR $\\mathbb R$-action.\n"}
{"abstract": "  An outlier is an observation or a data point that is far from rest of the\ndata points in a given dataset or we can be said that an outlier is away from\nthe center of mass of observations. Presence of outliers can skew statistical\nmeasures and data distributions which can lead to misleading representation of\nthe underlying data and relationships. It is seen that the removal of outliers\nfrom the training dataset before modeling can give better predictions. With the\nadvancement of machine learning, the outlier detection models are also\nadvancing at a good pace. The goal of this work is to highlight and compare\nsome of the existing outlier detection techniques for the data scientists to\nuse that information for outlier algorithm selection while building a machine\nlearning model.\n"}
{"abstract": "  Molecular dynamics is a powerful simulation tool to explore material\nproperties. Most of the realistic material systems are too large to be\nsimulated with first-principles molecular dynamics. Classical molecular\ndynamics has lower computational cost but requires accurate force fields to\nachieve chemical accuracy. In this work, we develop a symmetry-adapted graph\nneural networks framework, named molecular dynamics graph neural networks\n(MDGNN), to construct force fields automatically for molecular dynamics\nsimulations for both molecules and crystals. This architecture consistently\npreserves the translation, rotation and permutation invariance in the\nsimulations. We propose a new feature engineering method including higher order\ncontributions and show that MDGNN accurately reproduces the results of both\nclassical and first-principles molecular dynamics. We also demonstrate that\nforce fields constructed by the model has good transferability. Therefore,\nMDGNN provides an efficient and promising option for molecular dynamics\nsimulations of large scale systems with high accuracy.\n"}
{"abstract": "  We present an alternative approach to studying topology in open quantum\nsystems, relying directly on Green's functions and avoiding the need to\nconstruct an effective non-Hermitian Hamiltonian. We define an energy-dependent\nChern number based on the eigenstates of the inverse Green's function matrix of\nthe system which contains, within the self-energy, all the information about\nthe influence of the environment, interactions, gain or losses. We explicitly\ncalculate this topological invariant for a system consisting of a single 2D\nDirac cone and find that it is half-integer quantized when certain assumptions\nover the damping are made. Away from these conditions, which cannot or are not\nusually considered within the formalism of non-Hermitian Hamiltonians, we find\nthat such a quantization is usually lost and the Chern number vanishes, and\nthat in special cases, it can change to integer quantization.\n"}
{"abstract": "  We use a logical device called the Dutch Book to establish epistemic\nconfidence, defined as the sense of confidence \\emph{in an observed} confidence\ninterval. This epistemic property is unavailable -- or even denied -- in\northodox frequentist inference. In financial markets, including the betting\nmarket, the Dutch Book is also known as arbitrage or risk-free profitable\ntransaction. A numerical confidence is deemed epistemic if its use as a betting\nprice is protected from the Dutch Book by an external agent. Theoretically, to\nconstruct the Dutch Book, the agent must exploit unused information available\nin any relevant subset. Pawitan and Lee (2021) showed that confidence is an\nextended likelihood, and the likelihood principle states that the likelihood\ncontains all the information in the data, hence leaving no relevant subset.\nIntuitively, this implies that confidence associated with the full likelihood\nis protected from the Dutch Book, and hence is epistemic. Our aim is to provide\nthe theoretical support for this intuitive notion.\n"}
{"abstract": "  The use of continuum emission radio galaxies as cosmological tracers of the\nlarge-scale structure will soon move into a new phase. Upcoming surveys from\nthe Australian Square Kilometre Array Pathfinder (ASKAP), MeerKAT, and the\nSquare Kilometre Array project (SKA) will survey the entire available sky down\nto an ~100uJy flux limit, increasing the number of detected extra-galactic\nradio sources by several orders of magnitude. External data and machine\nlearning algorithms will also enable some low resolution radial selection\n(photometric redshift binning) of the sample, increasing the cosmological\nutility of the sample observed. In this paper, we discuss the flux limit\nrequired to detect enough galaxies to decrease the shot noise term in the error\nto be 10% of the total. We show how future surveys of this type will be limited\nby available technology. The confusion generated by the intrinsic sizes of\ngalaxies may have the consequence that surveys of this type eventually reach a\nhard flux limit of ~100nJy, as is predicted by the current modelling of AGN\nsizes by simulations such as the Tiered Radio Extragalactic Continuum\nSimulation (T-RECS). Finally, when considering the multi-tracer approach, where\ngalaxies are split by type to measure some bias ratio, we find that there are\nnot enough AGN present to achieve a reasonable level of shot noise for this\nkind of measurement.\n"}
{"abstract": "  Weakly-supervised anomaly detection aims at learning an anomaly detector from\na limited amount of labeled data and abundant unlabeled data. Recent works\nbuild deep neural networks for anomaly detection by discriminatively mapping\nthe normal samples and abnormal samples to different regions in the feature\nspace or fitting different distributions. However, due to the limited number of\nannotated anomaly samples, directly training networks with the discriminative\nloss may not be sufficient. To overcome this issue, this paper proposes a novel\nstrategy to transform the input data into a more meaningful representation that\ncould be used for anomaly detection. Specifically, we leverage an autoencoder\nto encode the input data and utilize three factors, hidden representation,\nreconstruction residual vector, and reconstruction error, as the new\nrepresentation for the input data. This representation amounts to encode a test\nsample with its projection on the training data manifold, its direction to its\nprojection and its distance to its projection. In addition to this encoding, we\nalso propose a novel network architecture to seamlessly incorporate those three\nfactors. From our extensive experiments, the benefits of the proposed strategy\nare clearly demonstrated by its superior performance over the competitive\nmethods.\n"}
{"abstract": "  We study the evolution of networks through `triplets' - three-node graphlets.\nWe develop a method to compute a transition matrix to describe the evolution of\ntriplets in temporal networks. To identify the importance of higher-order\ninteractions in the evolution of networks, we compare both artificial and\nreal-world data to a model based on pairwise interactions only. The significant\ndifferences between the computed matrix and the calculated matrix from the\nfitted parameters demonstrate that non-pairwise interactions exist for various\nreal-world systems in space and time, such as our data sets. Furthermore, this\nalso reveals that different patterns of higher-order interaction are involved\nin different real-world situations.\n  To test our approach, we then use these transition matrices as the basis of a\nlink prediction algorithm. We investigate our algorithm's performance on four\ntemporal networks, comparing our approach against ten other link prediction\nmethods. Our results show that higher-order interactions in both space and time\nplay a crucial role in the evolution of networks as we find our method, along\nwith two other methods based on non-local interactions, give the best overall\nperformance. The results also confirm the concept that the higher-order\ninteraction patterns, i.e., triplet dynamics, can help us understand and\npredict the evolution of different real-world systems.\n"}
{"abstract": "  Tunnel transport of interacting spin-polarized electrons through a\nsingle-level vibrating quantum dot in external magnetic field is studied. By\nusing density matrix method, the current-voltage characteristics and the\ndependence of conductance on temperature of a single-electron transistor were\ncalculated. We found that a lifting of Coulomb blockade in external magnetic\nfield happens in stages. The Franck-Condon steps associated with inelastic\nelectron tunneling in our case are doubled due to contribution of two\nZeeman-split levels in electron transport. The doubling of steps can be also\nobserved in the presence of Coulomb interaction. For strong electron-vibron\ninteraction the temperature dependence of conductance is shown to be\nnon-monotonic and anomalous growth of conductance maximum weakly depends both\non the Coulomb strength and the external magnetic field.\n"}
{"abstract": "  With the emergence of low-cost robotic systems, such as unmanned aerial\nvehicle, the importance of embedded high-performance image processing has\nincreased. For a long time, FPGAs were the only processing hardware that were\ncapable of high-performance computing, while at the same time preserving a low\npower consumption, essential for embedded systems. However, the recently\nincreasing availability of embedded GPU-based systems, such as the NVIDIA\nJetson series, comprised of an ARM CPU and a NVIDIA Tegra GPU, allows for\nmassively parallel embedded computing on graphics hardware. With this in mind,\nwe propose an approach for real-time embedded stereo processing on ARM and\nCUDA-enabled devices, which is based on the popular and widely used Semi-Global\nMatching algorithm. In this, we propose an optimization of the algorithm for\nembedded CUDA GPUs, by using massively parallel computing, as well as using the\nNEON intrinsics to optimize the algorithm for vectorized SIMD processing on\nembedded ARM CPUs. We have evaluated our approach with different configurations\non two public stereo benchmark datasets to demonstrate that they can reach an\nerror rate as low as 3.3%. Furthermore, our experiments show that the fastest\nconfiguration of our approach reaches up to 46 FPS on VGA image resolution.\nFinally, in a use-case specific qualitative evaluation, we have evaluated the\npower consumption of our approach and deployed it on the DJI Manifold 2-G\nattached to a DJI Matrix 210v2 RTK unmanned aerial vehicle (UAV), demonstrating\nits suitability for real-time stereo processing onboard a UAV.\n"}
{"abstract": "  The quantum approximate optimization algorithm (QAOA) has become a\ncornerstone of contemporary quantum applications development. In QAOA, a\nquantum circuit is trained -- by repeatedly adjusting circuit parameters -- to\nsolve a problem. Several recent findings have reported parameter concentration\neffects in QAOA and their presence has become one of folklore: while\nempirically observed, the concentrations have not been defined and analytical\napproaches remain scarce, focusing on limiting system and not considering\nparameter scaling as system size increases. We found that optimal QAOA circuit\nparameters concentrate as an inverse polynomial in the problem size, providing\nan optimistic result for improving circuit training. Our results are\nanalytically demonstrated for variational state preparations at $p=1,2$\n(corresponding to 2 and 4 tunable parameters respectively). The technique is\nalso applicable for higher depths and the concentration effect is cross\nverified numerically. Parameter concentrations allow for training on a fraction\n$w < n$ of qubits to assert that these parameters are nearly optimal on $n$\nqubits. Clearly this effect has significant practical importance.\n"}
{"abstract": "  Unprecedented winter storms that hit across Texas in February 2021 have\ncaused at least 69 deaths and 4.5 million customer interruptions due to the\nwide-ranging generation capacity outage and record-breaking electricity demand.\nWhile much remains to be investigated on what, how, and why such wide-spread\npower outages occurred across Texas, it is imperative for the broader macro\nenergy community to develop insights for policy making based on a coherent\nelectric grid model and data set. In this paper, we collaboratively release an\nopen-source extendable model that is synthetic but nevertheless provides a\nrealistic representation of the actual energy grid, accompanied by open-source\ncross-domain data sets. This simplified synthetic model is calibrated to the\nbest of our knowledge based on published data resources. Building upon this\nopen-source synthetic grid model, researchers could quantitatively assess the\nimpact of various policies on mitigating the impact of such extreme events. As\nan example, in this paper we critically assess several corrective measures that\ncould have mitigated the blackout under such extreme weather conditions. We\nuncover the regional disparity of load shedding. The analysis also quantifies\nthe sensitivity of several corrective measures with respect to mitigating the\nseverity of the power outage, as measured in Energy-not-Served (ENS). This\napproach and methodology are generalizable for other regions experiencing\nsignificant energy portfolio transitions.\n"}
{"abstract": "  We propose novel estimators for categorical and continuous treatments by\nusing an optimal covariate balancing strategy for inverse probability\nweighting. The resulting estimators are shown to be consistent and\nasymptotically normal for causal contrasts of interest, either when the model\nexplaining treatment assignment is correctly specified, or when the correct set\nof bases for the outcome models has been chosen and the assignment model is\nsufficiently rich. For the categorical treatment case, we show that the\nestimator attains the semiparametric efficiency bound when all models are\ncorrectly specified. For the continuous case, the causal parameter of interest\nis a function of the treatment dose. The latter is not parametrized and the\nestimators proposed are shown to have bias and variance of the classical\nnonparametric rate. Asymptotic results are complemented with simulations\nillustrating the finite sample properties. Our analysis of a data set suggests\na nonlinear effect of BMI on the decline in self reported health.\n"}
{"abstract": "  Inverse propensity weighting (IPW) is a popular method for estimating\ntreatment effects from observational data. However, its correctness relies on\nthe untestable (and frequently implausible) assumption that all confounders\nhave been measured. This paper introduces a robust sensitivity analysis for IPW\nthat estimates the range of treatment effects compatible with a given amount of\nunobserved confounding. The estimated range converges to the narrowest possible\ninterval (under the given assumptions) that must contain the true treatment\neffect. Our proposal is a refinement of the influential sensitivity analysis by\nZhao, Small, and Bhattacharya (2019), which we show gives bounds that are too\nwide even asymptotically. This analysis is based on new partial identification\nresults for Tan (2006)'s marginal sensitivity model.\n"}
{"abstract": "  Traversing a continuous phase transition at a finite rate leads to the\nbreakdown of adiabatic dynamics and the formation of topological defects, as\npredicted by the celebrated Kibble-Zurek mechanism (KZM). We investigate\nuniversal signatures beyond the KZM, by characterizing the distribution of\nvortices generated in a thermal quench leading to the formation of a\nholographic superconductor. The full counting statistics of vortices is\ndescribed by a binomial distribution, in which the mean value is dictated by\nthe KZM and higher-order cumulants share the universal power-law scaling with\nthe quench time. Extreme events associated with large fluctuations no longer\nexhibit a power-law behavior with the quench time and are characterized by a\nuniversal form of the Weibull distribution for different quench rates.\n"}
{"abstract": "  A common approach to tackle a combinatorial optimization problem is to first\nsolve a continuous relaxation and then round the obtained fractional solution.\nFor the latter, the framework of contention resolution schemes (or CR schemes),\nintroduced by Chekuri, Vondrak, and Zenklusen, is a general and successful\ntool. A CR scheme takes a fractional point $x$ in a relaxation polytope, rounds\neach coordinate $x_i$ independently to get a possibly non-feasible set, and\nthen drops some elements in order to satisfy the independence constraints.\nIntuitively, a CR scheme is $c$-balanced if every element $i$ is selected with\nprobability at least $c \\cdot x_i$.\n  It is known that general matroids admit a $(1-1/e)$-balanced CR scheme, and\nthat this is (asymptotically) optimal. This is in particular true for the\nspecial case of uniform matroids of rank one. In this work, we provide a simple\nand explicit monotone CR scheme with a balancedness of $1 -\n\\binom{n}{k}\\:\\left(1-\\frac{k}{n}\\right)^{n+1-k}\\:\\left(\\frac{k}{n}\\right)^k$,\nand show that this is optimal. As $n$ grows, this expression converges from\nabove to $1 - e^{-k}k^k/k!$. While this asymptotic bound can be obtained by\ncombining previously known results, these require defining an exponential-sized\nlinear program, as well as using random sampling and the ellipsoid algorithm.\nOur procedure, on the other hand, has the advantage of being simple and\nexplicit. Moreover, this scheme generalizes into an optimal CR scheme for\npartition matroids.\n"}
{"abstract": "  Multi-component integrable generalizations of the Fokas-Lenells equation,\nassociated with each irreducible Hermitian symmetric space are formulated.\nDescription of the underlying structures associated to the integrability, such\nas the Lax representation and the bi-Hamiltonian formulation of the equations\nis provided. Two reductions are considered as well, one of which leads to a\nnonlocal integrable model. Examples with Hermitian symmetric spaces of all\nclassical series of types A.III, BD.I, C.I and D.III are presented in details,\nas well as possibilities for further reductions in a general form.\n"}
{"abstract": "  The ultrafast hole dynamics triggered by the photoexcitation of molecular\ntargets is a highly correlated process even for those systems, like organic\nmolecules, having a weakly correlated ground state. We here provide a unifying\nframework and a numerically efficient matrix formulation of state-of-the-art\nnon-equilibrium Green's function (NEGF) methods like second-Born as well as\n$GW$ and $T$-matrix without and {\\em with} exchange diagrams. Numerical\nsimulations are presented for a paradigmatic, exactly solvable molecular system\nand the shortcomings of the established NEGF methods are highlighted. We then\ndevelop a NEGF scheme based on the Faddeev treatment of three-particle\ncorrelations; the exceptional improvement over established methods is explained\nand demonstrated. The Faddeev NEGF scheme scales linearly with the maximum\npropagation time, thereby opening prospects for femtosecond simulations of\nlarge molecules.\n"}
{"abstract": "  In the assembly process of printed circuit boards (PCB), most of the errors\nare caused by solder joints in Surface Mount Devices (SMD). In the literature,\ntraditional feature extraction based methods require designing hand-crafted\nfeatures and rely on the tiered RGB illumination to detect solder joint errors,\nwhereas the supervised Convolutional Neural Network (CNN) based approaches\nrequire a lot of labelled abnormal samples (defective solder joints) to achieve\nhigh accuracy. To solve the optical inspection problem in unrestricted\nenvironments with no special lighting and without the existence of error-free\nreference boards, we propose a new beta-Variational Autoencoders (beta-VAE)\narchitecture for anomaly detection that can work on both IC and non-IC\ncomponents. We show that the proposed model learns disentangled representation\nof data, leading to more independent features and improved latent space\nrepresentations. We compare the activation and gradient-based representations\nthat are used to characterize anomalies; and observe the effect of different\nbeta parameters on accuracy and on untwining the feature representations in\nbeta-VAE. Finally, we show that anomalies on solder joints can be detected with\nhigh accuracy via a model trained on directly normal samples without designated\nhardware or feature engineering.\n"}
{"abstract": "  The quasi-spherical Szekeres dust solutions are a generalization of the\nspherically symmetric Lemaitre-Tolman-Bondi dust models where the spherical\nshells of constant mass are non-concentric. The quasi-spherical Szekeres dust\nsolutions can be considered as cosmological models and are potentially models\nfor the formation of primordial black holes in the early universe. Any\ncollapsing quasi-spherical Szekeres dust solution where an apparent horizon\ncovers all shell-crossings that will occur can be considered as a model for the\nformation of a black hole. In this paper we will show that the apparent horizon\ncan be detected by a Cartan invariant. We will show that particular Cartan\ninvariants characterize properties of these solutions which have a physical\ninterpretation such as: the expansion or contraction of spacetime itself, the\nrelative movement of matter shells, shell-crossings and the appearance of necks\nand bellies.\n"}
{"abstract": "  We prove an estimate for multi-variable multiplicative character sums over\naffine subspaces of $\\mathbb A^n_k$, which generalize the well known estimates\nfor both classical Jacobi sums and one-variable polynomial multiplicative\ncharacter sums.\n"}
{"abstract": "  We calculate the quasiparticle dispersion and spectral weight of the\nquasiparticle that results when a hole is added to an antiferromagnetically\nordered CuO$_2$ plane of a cuprate superconductor. We also calculate the magnon\ncontribution to the quasiparticle spectral function. We start from a multiband\nmodel for the cuprates considered previously [Nat. Phys. \\textbf{10}, 951\n(2014)]. We map this model and the operator for creation of an O hole to an\neffective one-band generalized $t-J$ model, without free parameters. The\neffective model is solved using the state of the art self-consistent Born\napproximation. Our results reproduce all the main features of experiments. They\nalso reproduce qualitatively the dispersion of the multiband model, giving\nbetter results for the intensity near wave vector $(\\pi,\\pi)$, in comparison\nwith the experiments. In contrast to what was claimed in [Nat. Phys.\n\\textbf{10}, 951 (2014)], we find that spin fluctuations play an essential role\nin the dynamics of the quasiparticle, and hence in both its weight and\ndispersion.\n"}
{"abstract": "  The goal of this paper is to study the smallest brightening events observed\nin the EUV quiet Sun. We use commissioning data taken by the EUI instrument\nonboard the recently launched Solar Orbiter mission. On 2020 May 30, EUI was\nsituated at 0.556AU from the Sun. Its HRIEUV telescope 17.4nm passband reached\nan exceptionally high two-pixel spatial resolution of 400km. The size and\nduration of small-scale structures is determined in the HRIEUV data, while\ntheir height is estimated from triangulation with the simultaneous SDO/AIA\ndata. This is the first stereoscopy of small scale brightenings at high\nresolution. We observed small localised brightenings (\"campfires\") in a quiet\nSun region with lengthscales between 400km and 4000km and durations between 10\nand 200s. The smallest and weakest of these HRIEUV brightenings have not been\nobserved before. Simultaneous HRILYA observations do not show localised\nbrightening events, but the locations of the HRIEUV events correspond clearly\nto the chromospheric network. Comparison with simultaneous AIA images shows\nthat most events can also be identified in the 17.1nm, 19.3nm, 21.1nm, and\n30.4nm passbands of AIA, although they appear weaker and blurred. DEM analysis\nindicates coronal temperatures peaking at log(T)~6.1-6.15. We determined the\nheight of a few campfires, which is between 1000 and 5000km above the\nphotosphere. We conclude that \"campfires\" are mostly coronal in nature and are\nrooted in the magnetic flux concentrations of the chromospheric network. We\ninterpret these events as a new extension to the flare/microflare/nanoflare\nfamily. Given their low height, the EUI \"campfires\" could be a new element of\nthe fine structure of the transition region/low corona: apexes of small-scale\nloops that are internally heated to coronal temperatures.\n"}
{"abstract": "  Deciding what's next? is a fundamental problem in robotics and Artificial\nIntelligence. Under belief space planning (BSP), in a partially observable\nsetting, it involves calculating the expected accumulated belief-dependent\nreward, where the expectation is with respect to all future measurements. Since\nsolving this general un-approximated problem quickly becomes intractable, state\nof the art approaches turn to approximations while still calculating planning\nsessions from scratch. In this work we propose a novel paradigm, Incremental\nBSP (iX-BSP), based on the key insight that calculations across planning\nsessions are similar in nature and can be appropriately re-used. We calculate\nthe expectation incrementally by utilizing Multiple Importance Sampling\ntechniques for selective re-sampling and re-use of measurement from previous\nplanning sessions. The formulation of our approach considers general\ndistributions and accounts for data association aspects. We demonstrate how\niX-BSP could benefit existing approximations of the general problem,\nintroducing iML-BSP, which re-uses calculations across planning sessions under\nthe common Maximum Likelihood assumption. We evaluate both methods and\ndemonstrate a substantial reduction in computation time while statistically\npreserving accuracy. The evaluation includes both simulation and real-world\nexperiments considering autonomous vision-based navigation and SLAM. As a\nfurther contribution, we introduce to iX-BSP the non-integral wildfire\napproximation, allowing one to trade accuracy for computational performance by\naverting from updating re-used beliefs when they are \"close enough\". We\nevaluate iX-BSP under wildfire demonstrating a substantial reduction in\ncomputation time while controlling the accuracy sacrifice. We also provide\nanalytical and empirical bounds of the effect wildfire holds over the objective\nvalue.\n"}
{"abstract": "  Model-based Reinforcement Learning (MBRL) is a promising framework for\nlearning control in a data-efficient manner. MBRL algorithms can be fairly\ncomplex due to the separate dynamics modeling and the subsequent planning\nalgorithm, and as a result, they often possess tens of hyperparameters and\narchitectural choices. For this reason, MBRL typically requires significant\nhuman expertise before it can be applied to new problems and domains. To\nalleviate this problem, we propose to use automatic hyperparameter optimization\n(HPO). We demonstrate that this problem can be tackled effectively with\nautomated HPO, which we demonstrate to yield significantly improved performance\ncompared to human experts. In addition, we show that tuning of several MBRL\nhyperparameters dynamically, i.e. during the training itself, further improves\nthe performance compared to using static hyperparameters which are kept fixed\nfor the whole training. Finally, our experiments provide valuable insights into\nthe effects of several hyperparameters, such as plan horizon or learning rate\nand their influence on the stability of training and resulting rewards.\n"}
{"abstract": "  We introduce the Probabilistic Worldbuilding Model (PWM), a new\nfully-symbolic Bayesian model of semantic parsing and reasoning, as a first\nstep in a research program toward more domain- and task-general NLU and AI.\nHumans create internal mental models of their observations which greatly aid in\ntheir ability to understand and reason about a large variety of problems. In\nPWM, the meanings of sentences, acquired facts about the world, and\nintermediate steps in reasoning are all expressed in a human-readable formal\nlanguage, with the design goal of interpretability. PWM is Bayesian, designed\nspecifically to be able to generalize to new domains and new tasks. We derive\nand implement an inference algorithm that reads sentences by parsing and\nabducing updates to its latent world model that capture the semantics of those\nsentences, and evaluate it on two out-of-domain question-answering datasets:\n(1) ProofWriter and (2) a new dataset we call FictionalGeoQA, designed to be\nmore representative of real language but still simple enough to focus on\nevaluating reasoning ability, while being robust against heuristics. Our method\noutperforms baselines on both, thereby demonstrating its value as a\nproof-of-concept.\n"}
{"abstract": "  Video-to-speech is the process of reconstructing the audio speech from a\nvideo of a spoken utterance. Previous approaches to this task have relied on a\ntwo-step process where an intermediate representation is inferred from the\nvideo, and is then decoded into waveform audio using a vocoder or a waveform\nreconstruction algorithm. In this work, we propose a new end-to-end\nvideo-to-speech model based on Generative Adversarial Networks (GANs) which\ntranslates spoken video to waveform end-to-end without using any intermediate\nrepresentation or separate waveform synthesis algorithm. Our model consists of\nan encoder-decoder architecture that receives raw video as input and generates\nspeech, which is then fed to a waveform critic and a power critic. The use of\nan adversarial loss based on these two critics enables the direct synthesis of\nraw audio waveform and ensures its realism. In addition, the use of our three\ncomparative losses helps establish direct correspondence between the generated\naudio and the input video. We show that this model is able to reconstruct\nspeech with remarkable realism for constrained datasets such as GRID, and that\nit is the first end-to-end model to produce intelligible speech for LRW (Lip\nReading in the Wild), featuring hundreds of speakers recorded entirely `in the\nwild'. We evaluate the generated samples in two different scenarios -- seen and\nunseen speakers -- using four objective metrics which measure the quality and\nintelligibility of artificial speech. We demonstrate that the proposed approach\noutperforms all previous works in most metrics on GRID and LRW.\n"}
{"abstract": "  We use the NewHorizon simulation to study the redshift evolution of bar\nproperties and fractions within galaxies in the stellar masses range $M_{\\star}\n= 10^{7.25} - 10^{11.4} \\ \\rm{M}_{\\odot}$ over the redshift range $z = 0.25 -\n1.3$. We select disc galaxies using stellar kinematics as a proxy for galaxy\nmorphology. We employ two different automated bar detection methods, coupled\nwith visual inspection, resulting in observable bar fractions of $f_{\\rm bar} =\n0.070_{{-0.012}}^{{+0.018}}$ at $z\\sim$ 1.3, decreasing to $f_{\\rm bar} =\n0.011_{{-0.003}}^{{+0.014}}$ at $z\\sim$ 0.25. Only one galaxy is visually\nconfirmed as strongly barred in our sample. This bar is hosted by the most\nmassive disk and only survives from $z=1.3$ down to $z=0.7$. Such a low bar\nfraction, in particular amongst Milky Way-like progenitors, highlights a\nmissing bars problem, shared by literally all cosmological simulations with\nspatial resolution $<$100 pc to date. The analysis of linear growth rates,\nrotation curves and derived summary statistics of the stellar, gas and dark\nmatter components suggest that galaxies with stellar masses below\n$10^{9.5}-10^{10} \\ \\rm{M}_{\\odot}$ in NewHorizon appear to be too dominated by\ndark matter to bar, while more massive galaxies typically have formed large\nbulges that prevent bar persistence at low redshift. This investigation\nconfirms that the evolution of the bar fraction puts stringent constraints on\nthe assembly history of baryons and dark matter onto galaxies.\n"}
{"abstract": "  We study the relationship between the concept of a continuous ellipsoid\n$\\Theta$ cover of $\\mathbb{R}^n$, which was introduced by Dahmen, Dekel, and\nPetrushev, and the space of homogeneous type induced by $\\Theta$. We\ncharacterize the class of quasi-distances on $\\mathbb{R}^n$ (up to equivalence)\nwhich correspond to continuous ellipsoid covers. This places firmly continuous\nellipsoid covers as a subclass of spaces of homogeneous type on $\\mathbb{R}^n$\nsatisfying quasi-convexity and $1$-Ahlfors-regularity.\n"}
{"abstract": "  Multi-epoch, small-batch, Stochastic Gradient Descent (SGD) has been the\nmethod of choice for learning with large over-parameterized models. A popular\ntheory for explaining why SGD works well in practice is that the algorithm has\nan implicit regularization that biases its output towards a good solution.\nPerhaps the theoretically most well understood learning setting for SGD is that\nof Stochastic Convex Optimization (SCO), where it is well known that SGD learns\nat a rate of $O(1/\\sqrt{n})$, where $n$ is the number of samples. In this\npaper, we consider the problem of SCO and explore the role of implicit\nregularization, batch size and multiple epochs for SGD. Our main contributions\nare threefold:\n  (a) We show that for any regularizer, there is an SCO problem for which\nRegularized Empirical Risk Minimzation fails to learn. This automatically rules\nout any implicit regularization based explanation for the success of SGD.\n  (b) We provide a separation between SGD and learning via Gradient Descent on\nempirical loss (GD) in terms of sample complexity. We show that there is an SCO\nproblem such that GD with any step size and number of iterations can only learn\nat a suboptimal rate: at least $\\widetilde{\\Omega}(1/n^{5/12})$.\n  (c) We present a multi-epoch variant of SGD commonly used in practice. We\nprove that this algorithm is at least as good as single pass SGD in the worst\ncase. However, for certain SCO problems, taking multiple passes over the\ndataset can significantly outperform single pass SGD.\n  We extend our results to the general learning setting by showing a problem\nwhich is learnable for any data distribution, and for this problem, SGD is\nstrictly better than RERM for any regularization function. We conclude by\ndiscussing the implications of our results for deep learning, and show a\nseparation between SGD and ERM for two layer diagonal neural networks.\n"}
{"abstract": "  Social structures influence a variety of human behaviors including mobility\npatterns, but the extent to which one individual's movements can predict\nanother's remains an open question. Further, latent information about an\nindividual's mobility can be present in the mobility patterns of both social\nand non-social ties, a distinction that has not yet been addressed. Here we\ndevelop a \"colocation\" network to distinguish the mobility patterns of an ego's\nsocial ties from those of non-social colocators, individuals not socially\nconnected to the ego but who nevertheless arrive at a location at the same time\nas the ego. We apply entropy and predictability measures to analyse and bound\nthe predictive information of an individual's mobility pattern and the flow of\nthat information from their top social ties and from their non-social\ncolocators. While social ties generically provide more information than\nnon-social colocators, we find that significant information is present in the\naggregation of non-social colocators: 3-7 colocators can provide as much\npredictive information as the top social tie, and colocators can replace up to\n85% of the predictive information about an ego, compared with social ties that\ncan replace up to 94% of the ego's predictability. The presence of predictive\ninformation among non-social colocators raises privacy concerns: given the\nincreasing availability of real-time mobility traces from smartphones,\nindividuals sharing data may be providing actionable information not just about\ntheir own movements but the movements of others whose data are absent, both\nknown and unknown individuals.\n"}
{"abstract": "  The majority of existing methods for empathetic response generation rely on\nthe emotion of the context to generate empathetic responses. However, empathy\nis much more than generating responses with an appropriate emotion. It also\noften entails subtle expressions of understanding and personal resonance with\nthe situation of the other interlocutor. Unfortunately, such qualities are\ndifficult to quantify and the datasets lack the relevant annotations. To\naddress this issue, in this paper we propose an approach that relies on\nexemplars to cue the generative model on fine stylistic properties that signal\nempathy to the interlocutor. To this end, we employ dense passage retrieval to\nextract relevant exemplary responses from the training set. Three elements of\nhuman communication -- emotional presence, interpretation, and exploration, and\nsentiment are additionally introduced using synthetic labels to guide the\ngeneration towards empathy. The human evaluation is also extended by these\nelements of human communication. We empirically show that these approaches\nyield significant improvements in empathetic response quality in terms of both\nautomated and human-evaluated metrics. The implementation is available at\nhttps://github.com/declare-lab/exemplary-empathy.\n"}
{"abstract": "  The convergence of radar sensing and communication applications in the\nterahertz (THz) band has been envisioned as a promising technology, since it\nincorporates terabit-per-second (Tbps) data transmission and mm-level radar\nsensing in a spectrum- and cost-efficient manner, by sharing both the frequency\nand hardware resources. However, the joint THz radar and communication (JRC)\nsystem faces considerable challenges, due to the peculiarities of the THz\nchannel and front ends. To this end, the waveform design for THz-JRC systems\nwith ultra-broad bandwidth is investigated in this paper. Firstly, by\nconsidering THz-JRC systems based on the co-existence concept, where both\nfunctions operate in a time-domain duplex (TDD) manner, a novel multi-subband\nquasi-perfect (MS-QP) sequence, composed of multiple Zadoff-Chu (ZC) perfect\nsubsequences on different subbands, is proposed for target sensing, which\nachieves accurate target ranging and velocity estimation, whilst only requiring\ncost-efficient low-rate analog-to-digital converters (A/Ds) for sequence\ndetection. Furthermore, the root index of each ZC subsequence of the MS-QP\nsequence is designed to eliminate the influence of doppler shift on the THz\nradar sensing. Finally, a data-embedded MS-QP (DE-MS-QP) waveform is\nconstructed through time-domain extension of the MS-QP sequence, generating\nnull frequency points on each subband for data transmission. Unlike the THz-JRC\nsystem in TDD manner, the proposed DE-MS-QP waveform enables simultaneous\ninterference-free sensing and communication, whilst inheriting all the merits\nfrom MS-QP sequences. Numerical results validate the superiority of the\nproposed waveforms in terms of sensing performance, hardware cost and flexible\nresource allocation over their conventional counterparts.\n"}
{"abstract": "  This paper is concerned with matching feature vectors in a one-to-one fashion\nacross large collections of datasets. Formulating this task as a\nmultidimensional assignment problem with decomposable costs (MDADC), we develop\nextremely fast algorithms with time complexity linear in the number $n$ of\ndatasets and space complexity a small fraction of the data size. These\nremarkable properties hinge on using the squared Euclidean distance as\ndissimilarity function, which can reduce ${n \\choose 2}$ matching problems\nbetween pairs of datasets to $n$ problems and enable calculating assignment\ncosts on the fly. To our knowledge, no other method applicable to the MDADC\npossesses these linear scaling and low-storage properties necessary to\nlarge-scale applications. In numerical experiments, the novel algorithms\noutperform competing methods and show excellent computational and optimization\nperformances. An application of feature matching to a large neuroimaging\ndatabase is presented. The algorithms of this paper are implemented in the R\npackage matchFeat available at https://github.com/ddegras/matchFeat.\n"}
{"abstract": "  A new family of methods involving complex coefficients for the numerical\nintegration of differential equations is presented and analyzed. They are\nconstructed as linear combinations of symmetric-conjugate compositions obtained\nfrom a basic time-symmetric integrator of order 2n (n $\\ge$ 1). The new\nintegrators are of order 2(n + k), k = 1, 2, ..., and preserve time-symmetry up\nto order 4n + 3 when applied to differential equations with real vector fields.\nIf in addition the system is Hamiltonian and the basic scheme is symplectic,\nthen they also preserve symplecticity up to order 4n + 3. We show that these\nintegrators are well suited for a parallel implementation, thus improving their\nefficiency. Methods up to order 10 based on a 4th-order integrator are built\nand tested in comparison with other standard procedures to increase the order\nof a basic scheme.\n"}
{"abstract": "  Predicting user positive response (e.g., purchases and clicks) probability is\na critical task in Web applications. To identify predictive features from raw\ndata, the state-of-the-art extreme deep factorization machine model (xDeepFM)\nintroduces a new interaction network to leverage feature interactions at the\nvector-wise level explicitly. However, since each hidden layer in the\ninteraction network is a collection of feature maps, it can be viewed\nessentially as an ensemble of different feature maps. In this case, only using\na single objective to minimize the prediction loss may lead to overfitting and\ngenerate correlated errors. In this paper, an ensemble diversity enhanced\nextreme deep factorization machine model (DexDeepFM) is proposed, which designs\nthe ensemble diversity measure in each hidden layer and considers both ensemble\ndiversity and prediction accuracy in the objective function. In addition, the\nattention mechanism is introduced to discriminate the importance of ensemble\ndiversity measures with different feature interaction orders. Extensive\nexperiments on three public real-world datasets are conducted to show the\neffectiveness of the proposed model.\n"}
{"abstract": "  Several statistics-based detectors, based on unimodal matrix models, for\ndetermining the number of sources in a field are designed. A new variance ratio\nstatistic is proposed, and its asymptotic distribution is analyzed. The\nvariance ratio detector is shown to outperform the alternatives. It is shown\nthat further improvements are achievable via optimally selected rotations.\nNumerical experiments demonstrate the performance gains of our detection\nmethods over the baseline approach.\n"}
{"abstract": "  Topological spin textures have attracted much attention both for fundamental\nphysics and spintronics applications. Among them, antiskyrmions possess a\nunique spin configuration with Bloch-type and N\\'eel-type domain walls due to\nanisotropic Dzyaloshinskii-Moriya interaction (DMI) in the noncentrosymmetric\ncrystal structure. However, antiskyrmions have thus far only been observed in a\nfew Heusler compounds with $D_{2\\mathrm{d}}$ symmetry. Here, we report a new\nmaterial Fe$_{1.9}$Ni$_{0.9}$Pd$_{0.2}$P in a different symmetry class ($S_4$\nsymmetry), where antiskyrmions exist over a wide temperature region including\nroom temperature, and transform to skyrmions upon changing magnetic field and\nlamella thickness. The periodicity of magnetic textures greatly depends on\ncrystal thickness, and domains with anisotropic sawtooth fractals are observed\nat the surface of thick crystals, which are attributed to the interplay between\ndipolar interaction and DMI as governed by crystal symmetry. Our findings\nprovide a new arena to study antiskyrmions, and should stimulate further\nresearch on topological spin textures and their applications.\n"}
{"abstract": "  We present a hierarchical equations of motion approach, which allows a\nnumerically exact simulation of nonequilibrium transport in general open\nquantum systems involving multiple macroscopic bosonic and fermionic\nenvironments. The performance of the method is demonstrated for a model of a\nnanosystem, which involves interacting electronic and vibrational degrees of\nfreedom and is coupled to fermionic and bosonic baths. The results show the\nintricate interplay of electronic and vibrational degrees of freedom in this\nnonequilibrium transport scenario for both voltage and thermally driven\ntransport processes. Furthermore, the use of importance criteria to improve the\nefficiency of the method is discussed.\n"}
{"abstract": "  The standard collisional model paradigm consists of a system that interacts\nsequentially with identically prepared ancillas. After infinitely many\ncollisions, and under appropriate conditions, the system may converge to the\nsame state as the ancillas. This process, known as homogenization, is\nindependent of the ancilla initial state, being a property only of the\nunderlying dynamics. In this paper we extend this idea to locally identical,\nbut globally correlated, ancillas, and show that correlations break\nhomogenization. This is done numerically using a minimal qubit model, and\nanalytically using an exactly soluble Gaussian model. In both cases, we use\nHamiltonian graph states with cyclic graphs as the prototypical method for\nbuilding scalable many-body entangled ancillary states.\n"}
{"abstract": "  In this survey article we revisit Hilbert's $19^{\\text{th}}$ problem\nconcerning the regularity of minimizers of variational integrals. We first\ndiscuss the classical theory (that is, the statement and resolution of\nHilbert's problem in all dimensions). We then discuss recent results concerning\nthe regularity of minimizers of degenerate convex functionals. Finally, we\ndiscuss some open problems. Exercises are included for the benefit of\nresearchers who are entering the subject.\n"}
{"abstract": "  In spaces of metrics, we investigate topological distributions of the\ndoubling property, the uniform disconnectedness, and the uniform perfectness,\nwhich are the quasi-symmetrically invariant properties appearing in the\nDavid--Semmes theorem. We show that the set of all doubling metrics and the set\nof all uniformly disconnected metrics are dense in spaces of metrics on\nfinite-dimensional and zero-dimensional compact metrizable spaces,\nrespectively. Conversely, this denseness of the sets implies the\nfinite-dimensionality, zero-dimensionality, and the compactness of metrizable\nspaces. We also determine the topological distribution of the set of all\nuniformly perfect metrics in the space of metrics on the Cantor set.\n"}
{"abstract": "  Understanding students' misconceptions is important for effective teaching\nand assessment. However, discovering such misconceptions manually can be\ntime-consuming and laborious. Automated misconception discovery can address\nthese challenges by highlighting patterns in student data, which domain experts\ncan then inspect to identify misconceptions. In this work, we present a novel\nmethod for the semi-automated discovery of problem-specific misconceptions from\nstudents' program code in computing courses, using a state-of-the-art code\nclassification model. We trained the model on a block-based programming dataset\nand used the learned embedding to cluster incorrect student submissions. We\nfound these clusters correspond to specific misconceptions about the problem\nand would not have been easily discovered with existing approaches. We also\ndiscuss potential applications of our approach and how these misconceptions\ninform domain-specific insights into students' learning processes.\n"}
{"abstract": "  Let $M\\subset\\mathbb{C}^{n+1}$ be a smooth affine hypersurface defined by the\nequation $xy+p(z_1,\\cdots,z_{n-1})=1$, where $p$ is a Brieskorn-Pham polynomial\nand $n\\geq2$. We prove that if $L\\subset M$ is an orientable exact Lagrangian\nsubmanifold, then $L$ does not admit a Riemannian metric with non-positive\nsectional curvature. The key point of the proof is to establish a version of\nhomological mirror symmetry for the wrapped Fukaya category of $M$, from which\nthe finite-dimensionality of the symplectic cohomology group $\\mathit{SH}^0(M)$\nfollows by a Hochschild cohomology computation.\n"}
{"abstract": "  A direct consequence of quantization of gravity would be the existence of\ngravitons. Therefore, spontaneous transition of an atom from an excited state\nto a lower-lying energy state accompanied with the emission of a graviton is\nexpected. In this paper, we take the gravitons emitted by hydrogen and helium\nin the Universe after recombination as a possible source of high frequency\nbackground gravitational waves, and calculate the energy density spectrum.\nExplicit calculations show that the most prominent contribution comes from the\n$3d-1s$ transition of singly ionized helium $\\mathrm{He}^{+}$, which gives a\npeak in frequency at $\\sim10^{13}$ Hz. Although the corresponding energy\ndensity is too small to be detected even with state-of-the-art technology\ntoday, we believe that the spontaneous emission of $\\mathrm{He}^{+}$ is a\nnatural source of high frequency gravitational waves, since it is a direct\nconsequence if we accept that the basic quantum principles we are already\nfamiliar with apply as well to a quantum theory of gravity.\n"}
{"abstract": "  An important problem in quantum information is the practical demonstration of\nnon-classical long-range order on quantum computers. One of the best known\nexamples of a quantum system with non-classical long-range order is a\nsuperconductor. Here we achieve Cooper pairing of qubits on a quantum computer\nto represent superconducting or superfluid states. We rigorously confirm the\nquantum long-range order by measuring the large $O(N)$ eigenvalue of the\ntwo-electron reduced density matrix. The demonstration of maximal quantum\nlong-range order is an important step towards more complex modeling of\nsuperconductivity and superfluidity as well as other phenomena with significant\nquantum long-range order on quantum computers.\n"}
{"abstract": "  Solar and stellar flares are powerful events which produce intense radiation\nacross the electromagnetic spectrum. Multiwavelength observations are highly\nimportant for understanding the nature of flares, because different\nflare-related processes reveal themselves in different spectral ranges. To\nstudy the correlation between thermal and nonthermal processes in stellar\nflares, we have searched the databases of Kepler (optical observations) and\nXMM-Newton (soft X-rays) for the flares observed simultaneously with both\ninstruments; nine distinctive flares (with energies exceeding $10^{33}$ erg) on\nthree stars (of K-M spectral classes) have been found. We have analyzed and\ncompared the flare parameters in the optical and X-ray spectral ranges; we have\nalso compared the obtained results with similar observations of solar flares.\nMost of the studied stellar flares released more energy in the optical range\nthan in X-rays. In one flare, X-ray emission strongly dominated, which could be\ncaused either by soft spectrum of energetic electrons or by a near-limb\nposition of this flare. The X-ray flares were typically delayed with respect to\nand shorter than their optical counterparts, which is partially consistent with\nthe Neupert effect. Using the scaling laws based on the magnetic reconnection\ntheory, we have estimated the characteristic magnetic field strengths in the\nstellar active regions and the sizes of these active regions as about $25-70$ G\nand $250\\,000-500\\,000$ km, respectively. The observed stellar superflares\nappear to be scaled-up versions of solar flares, with a similar underlying\nmechanism and nearly the same characteristic magnetic field values, but with\nmuch larger active region sizes.\n"}
{"abstract": "  In this paper, a new method for the separation of noise categories based on\nFour-Wave Mixing is presented.\n  The theoretical analysis is grounded in the Gaussian Noise model and verified\nby split step simulations. The noise categories react differently to the\nintroduced perturbations, by performing a set of perturbations the behaviour of\nthe different categories can be separated by means of a least-square fitting.\nGiven ASE is independent of the induced perturbations, it is possible to\nseparate noise contributions. The analysis includes constant and variable power\nperturbations.\n  The estimation of the noise categories is discussed from two points of view:\nNSR evolution post-DSP processing, and over the power spectral density in a\nnotched region. The NSR estimation can only be performed at reception, whereas\nthe power spectral density approach can be performed along the optical link if\na high resolution Optical Spectrum Analyzer is available.\n  Additionally, we perform a simple experimental verification considering of\ntwo WaveLogic 3 transceivers for the NSR, successfully estimating the noise\ncontributions.\n"}
{"abstract": "  Quasirandomness is a general mathematical concept meant to encapsulate\nseveral characteristics usually satisfied by random combinatorial objects, and\nwhich we regard as describing when a given object 'looks random'. In this\nsurvey we explore this general concept as it applies to graphs, hypergraphs and\nadditive groups, making clear their many connections to each other and showing\nhow they can be used in order to better study these objects.\n"}
{"abstract": "  We propose a learning-based robust predictive control algorithm that\ncompensates for significant uncertainty in the dynamics for a class of\ndiscrete-time systems that are nominally linear with an additive nonlinear\ncomponent. Such systems commonly model the nonlinear effects of an unknown\nenvironment on a nominal system. We optimize over a class of nonlinear feedback\npolicies inspired by certainty equivalent \"estimate-and-cancel\" control laws\npioneered in classical adaptive control to achieve significant performance\nimprovements in the presence of uncertainties of large magnitude, a setting in\nwhich existing learning-based predictive control algorithms often struggle to\nguarantee safety. In contrast to previous work in robust adaptive MPC, our\napproach allows us to take advantage of structure (i.e., the numerical\npredictions) in the a priori unknown dynamics learned online through function\napproximation. Our approach also extends typical nonlinear adaptive control\nmethods to systems with state and input constraints even when we cannot\ndirectly cancel the additive uncertain function from the dynamics. Moreover, we\napply contemporary statistical estimation techniques to certify the system's\nsafety through persistent constraint satisfaction with high probability.\nFinally, we show in simulation that our method can accommodate more significant\nunknown dynamics terms than existing methods.\n"}
{"abstract": "  Suppose we choose a permutation $\\pi$ uniformly at random from $S_n$. Let\n$\\mathsf{runsort}(\\pi)$ be the permutation obtained by sorting the ascending\nruns of $\\pi$ into lexicographic order. Alexandersson and Nabawanda recently\nasked if the plot of $\\mathsf{runsort}(\\pi)$, when scaled to the unit square\n$[0,1]^2$, converges to a limit shape as $n\\to\\infty$. We answer their question\nby showing that the measures corresponding to the scaled plots of these\npermutations $\\mathsf{runsort}(\\pi)$ converge with probability $1$ to a\npermuton (limiting probability distribution) that we describe explicitly. In\nparticular, the support of this permuton is $\\{(x,y)\\in[0,1]^2:x\\leq\nye^{1-y}\\}$.\n"}
{"abstract": "  The gravitational wave event, GW190521 is the most massive binary black hole\nmerger observed by ground-based gravitational wave observatories LIGO/Virgo to\ndate. While the observed gravitational-wave signal is mainly in the merger and\nringdown phases, the inspiral gravitational-wave signal of GW190521-like binary\nwill be more visible by space-based detectors in the low-frequency band. In\naddition, the ringdown gravitational-wave signal will be more loud with the\nnext generation (3G) of ground-based detectors in the high-frequency band,\ndisplaying a great potential of the multiband gravitational wave observations.\nIn this paper, we explore the scientific potential of multiband observations of\nGW190521-like binaries with milli-Hz gravitational wave observatory: LISA,\ndeci-Hz observatory: B-DECIGO, and (next generation of) hecto-Hz observatories:\naLIGO and ET. In the case of quasicircular evolution, the triple-band\nobservation by LISA, B-DECIGO and ET will provide parameter estimation errors\nof the masses and spin amplitudes of component black holes at the level of\norder 1% -- 10%. This would allow consistency tests of general relativity in\nthe strong-field at an unparalleled precision, particularly with the \"B-DECIGO\n+ ET\" observation. In the case of eccentric evolution, the multiband\nsignal-to-noise ratio by \"B-DECIGO + ET\" observation would be larger than 100\nfor a five year observation prior to coalescence, even with high final\neccentricities.\n"}
{"abstract": "  We describe here a new algorithm to model the water contents of the\natmosphere from GNSS slant wet delays relative to a single receiver. We first\nmake the assumption that the water vapor contents are mainly governed by a\nscale height (exponential law), and secondly that the departures from this\ndecaying exponential can be mapped as a set of low degree 3D Zernike functions\n(w.r.t. space) and Tchebyshev polynomials (w.r.t. time.) We give an example of\ninversion with data acquired over a one day time span at the Geodesy\nObservatory of Tahiti.\n"}
{"abstract": "  This contribution studies the impact of the rotor-stator interaction in a\nsingle-stage progressing cavity pump on the flow rate and pressure.\nSpecifically, we investigate the effect of the rotor movement on the sealings\nformed with deformable stators for various speeds and pressures. Sealings are\nreconstructed with the help of a geometric 3D model. We analyze the tangential\nand radial deviation of the rotor from its reference path and show that the\nradial deviation affects the flow rate, whereas the tangential deviation\naffects the pressure dynamics. The conjectures are confirmed with a laboratory\ntest setup.\n"}
{"abstract": "  Explosive astrophysical transients at cosmological distances can be used to\nplace precision tests of the basic assumptions of relativity theory, such as\nLorentz invariance, the photon zero-mass hypothesis, and the weak equivalence\nprinciple (WEP). Signatures of Lorentz invariance violations (LIV) include\nvacuum dispersion and vacuum birefringence. Sensitive searches for LIV using\nastrophysical sources such as gamma-ray bursts, active galactic nuclei, and\npulsars are discussed. The most direct consequence of a nonzero photon rest\nmass is a frequency dependence in the velocity of light propagating in vacuum.\nA detailed representation of how to obtain a combined severe limit on the\nphoton mass using fast radio bursts at different redshifts through the\ndispersion method is presented. The accuracy of the WEP has been well tested\nbased on the Shapiro time delay of astrophysical messengers traveling through a\ngravitational field. Some caveats of Shapiro delay tests are discussed. In this\narticle, we review and update the status of astrophysical tests of fundamental\nphysics.\n"}
{"abstract": "  In this work, we firstly apply the Train-Tensor (TT) networks to construct a\ncompact representation of the classical Multilayer Perceptron, representing a\nreduction of up to 95% of the coefficients. A comparative analysis between\ntensor model and standard multilayer neural networks is also carried out in the\ncontext of prediction of the Mackey-Glass noisy chaotic time series and NASDAQ\nindex. We show that the weights of a multidimensional regression model can be\nlearned by means of TT network and the optimization of TT weights is a more\nrobust to the impact of coefficient initialization and hyper-parameter setting.\nFurthermore, an efficient algorithm based on alternating least squares has been\nproposed for approximating the weights in TT-format with a reduction of\ncomputational calculus, providing a much faster convergence than the well-known\nadaptive learning-method algorithms, widely applied for optimizing neural\nnetworks.\n"}
{"abstract": "  Reeb graphs are widely used in a range of fields for the purposes of\nanalyzing and comparing complex spaces via a simpler combinatorial object.\nFurther, they are closely related to extended persistence diagrams, which\nlargely but not completely encode the information of the Reeb graph. In this\npaper, we investigate the effect on the persistence diagram of a particular\ncontinuous operation on Reeb graphs; namely the (truncated) smoothing\noperation. This construction arises in the context of the Reeb graph\ninterleaving distance, but separately from that viewpoint provides a\nsimplification of the Reeb graph which continuously shrinks small loops. We\nthen use this characterization to initiate the study of inverse problems for\nReeb graphs using smoothing by showing which paths in persistence diagram space\n(commonly known as vineyards) can be realized by a path in the space of Reeb\ngraphs via these simple operations. This allows us to solve the inverse problem\non a certain family of piecewise linear vineyards when fixing an initial Reeb\ngraph.\n"}
{"abstract": "  Nonlinearity continuation method, applied to boundary value problems for\nsteady-state Richards equation, gradually approaches the solution through a\nseries of intermediate problems. Originally, the Newton method with simple line\nsearch algorithm was used to solve the intermediate problems. In this paper,\nother solvers such as Picard and mixed Picard-Newton methods are considered,\ncombined with slightly modified line search approach. Numerical experiments are\nperformed with advanced finite volume discretizations on model and real-life\nproblems.\n"}
{"abstract": "  Let $f$ be a Rademacher or a Steinhaus random multiplicative function. Let\n$\\varepsilon>0$ small. We prove that, as $x\\rightarrow +\\infty$, we almost\nsurely have $$\\bigg|\\sum_{\\substack{n\\leq x\\\\\nP(n)>\\sqrt{x}}}f(n)\\bigg|\\leq\\sqrt{x}(\\log\\log x)^{1/4+\\varepsilon},$$ where\n$P(n)$ stands for the largest prime factor of $n$. This gives an indication of\nthe almost sure size of the largest fluctuations of $f$.\n"}
{"abstract": "  This work proposes RaNNC (Rapid Neural Network Connector) as middleware for\nautomatic hybrid parallelism. In recent deep learning research, as exemplified\nby T5 and GPT-3, the size of neural network models continues to grow. Since\nsuch models do not fit into the memory of accelerator devices, they need to be\npartitioned by model parallelism techniques. Moreover, to accelerate training\nfor huge training data, we need a combination of model and data parallelisms,\ni.e., hybrid parallelism. Given a model description for PyTorch without any\nspecification for model parallelism, RaNNC automatically partitions the model\ninto a set of subcomponents so that (1) each subcomponent fits a device memory\nand (2) a high training throughput for pipeline parallelism is achieved by\nbalancing the computation times of the subcomponents. In our experiments, we\ncompared RaNNC with two popular frameworks, Megatron-LM (hybrid parallelism)\nand GPipe (originally proposed for model parallelism, but a version allowing\nhybrid parallelism also exists), for training models with increasingly greater\nnumbers of parameters. In the pre-training of enlarged BERT models, RaNNC\nsuccessfully trained models five times larger than those Megatron-LM could, and\nRaNNC's training throughputs were comparable to Megatron-LM's when pre-training\nthe same models. RaNNC also achieved better training throughputs than GPipe on\nboth the enlarged BERT model pre-training (GPipe with hybrid parallelism) and\nthe enlarged ResNet models (GPipe with model parallelism) in all of the\nsettings we tried. These results are remarkable, since RaNNC automatically\npartitions models without any modification to their descriptions; Megatron-LM\nand GPipe require users to manually rewrite the models' descriptions.\n"}
{"abstract": "  Active control of heat flow is of both fundamental and applied interest in\nthermal management and energy conversion. Here, we present a fluctuational\nelectrodynamic study of thermal radiation between twisted bilayer graphene\n(TBLG), motivated by its unusual and highly tunable plasmonic properties. We\nshow that near-field heat flow can vary by more than 10-fold over only a few\ndegrees of twist, and identify special angles leading to heat flow extrema.\nThese special angles are dictated by the Drude weight in the intraband optical\nconductivity of TBLG, and are roughly linear with the chemical potential.\nFurther, we observe multiband thermal transport due to the increasing role of\ninterband transitions as the twist angle decreases, in analogy to monolayer\ngraphene in a magnetic field. Our findings are understood via the surface\nplasmons in TBLG, and highlight its potential for manipulating radiative heat\nflow.\n"}
{"abstract": "  Marangoni contraction of sessile droplets occurs when a binary mixture of\nvolatile liquids is placed on a high-energy surface. Although the surface is\nwetted completely by the mixture and its components, a quasi-stationary\nnon-vanishing contact angle is observed. This seeming contradiction is caused\nby Marangoni flows that are driven by evaporative depletion of the volatile\ncomponent near the edge of the droplet. Here we show that the compositional\nevolution of such droplets is governed by Taylor-Aris dispersion. Breaking with\nprecedent, we demonstrate that Taylor-Aris dispersion can consistently be\nincluded in a long wave expansion for the evolution of thin films. Coupled to\ndiffusion limited evaporation, this model quantitatively reproduces not only\nthe apparent shape of Marangoni contracted droplets, but also their internal\nflows.\n"}
{"abstract": "  We demonstrate a 2-grating free electron Mach-Zehnder interferometer\nconstructed in a transmission electron microscope. A symmetric binary phase\ngrating and condenser lens system forms two spatially separated, focused probes\nat the sample which can be scanned while maintaining alignment. The two paths\ninterfere at a second grating, creating in constructive or destructive\ninterference in output beams. This interferometer has many notable features:\npositionable probe beams, large path separations relative to beam width,\ncontinuously tunable relative phase between paths, and real-time phase\ninformation. Here we use the electron interferometer to measure the relative\nphase shifts imparted to the electron probes by electrostatic potentials as\nwell as a demonstration of quantitative nanoscale phase imaging of a\npolystyrene latex nanoparticle.\n"}
{"abstract": "  This paper introduces a new controllability notion, termed partial strong\nstructural controllability (PSSC), on a structured system whose entries of\nsystem matrices are either fixed zero or indeterminate, which naturally extends\nthe conventional strong structural controllability (SSC) and bridges the gap\nbetween structural controllability and SSC. Dividing the indeterminate entries\ninto two categories, generic entries and unspecified entries, a system is PSSC,\nif for almost all values of the generic entries in the parameter space except\nfor a set of measure zero, and any nonzero (complex) values of the unspecified\nentries, the corresponding system is controllable. We highlight that this\nnotion generalizes the generic property embedded in the conventional structural\ncontrollability for single-input systems. We then give algebraic and\n(bipartite) graph-theoretic necessary and sufficient conditions for\nsingle-input systems to be PSSC. Conditions for multi-input systems are\nsubsequently given for a special case. It is shown the established results can\ninduce a new maximum matching based criterion for SSC over the system bipartite\ngraph representations.\n"}
{"abstract": "  We analyze 6.7 million case law documents to determine the presence of gender\nbias within our judicial system. We find that current bias detectino methods in\nNLP are insufficient to determine gender bias in our case law database and\npropose an alternative approach. We show that existing algorithms' inconsistent\nresults are consequences of prior research's definition of biases themselves.\nBias detection algorithms rely on groups of words to represent bias (e.g.,\n'salary,' 'job,' and 'boss' to represent employment as a potentially biased\ntheme against women in text). However, the methods to build these groups of\nwords have several weaknesses, primarily that the word lists are based on the\nresearchers' own intuitions. We suggest two new methods of automating the\ncreation of word lists to represent biases. We find that our methods outperform\ncurrent NLP bias detection methods. Our research improves the capabilities of\nNLP technology to detect bias and highlights gender biases present in\ninfluential case law. In order test our NLP bias detection method's\nperformance, we regress our results of bias in case law against U.S census data\nof women's participation in the workforce in the last 100 years.\n"}
{"abstract": "  This paper describes how the current lexical similarity and analogy gold\nstandards are built to conform to certain ideas about what the models they are\ndesigned to evaluate are used for. Topical relevance has always been the most\nimportant target notion for information access tools and related language\ntechnology technologies, and while this has proven a useful starting point for\nmuch of what information technology is used for, it does not always align well\nwith other uses to which technologies are being put, most notably use cases\nfrom digital scholarship in the humanities or social sciences. This paper\nargues for more systematic formulation of requirements from the digital\nhumanities and social sciences and more explicit description of the assumptions\nunderlying model design.\n"}
{"abstract": "  We study mean field portfolio games in incomplete markets with random market\nparameters, where each player is concerned with not only her own wealth but\nalso the relative performance to her competitors. We use the martingale\noptimality principle approach to characterize the unique Nash equilibrium in\nterms of a mean field FBSDE with quadratic growth, which is solvable under a\nweak interaction assumption. Motivated by the weak interaction assumption, we\nestablish an asymptotic expansion result in powers of the competition\nparameter. When the market parameters do not depend on the Brownian paths, we\nget the Nash equilibrium in closed form. Moreover, when all the market\nparameters become time-independent, we revisit the games in [21] and our\nanalysis shows that nonconstant equilibria do not exist in $L^\\infty$, and the\nconstant equilibrium obtained in [21] is unique in $L^\\infty$, not only in the\nspace of constant equilibria.\n"}
{"abstract": "  Despite extensive researches on manganites owing to widespread use in modern\nelectronics, this class of metal oxides does not cease to surprise with its\nunique properties and new phenomena. Here we have studied structural and\nmagnetic properties of non-Heisenberg manganite nanoparticles with a strong\nspin-electron coupling. During transition from low temperature ferromagnetic to\nhigh temperature paramagnetic state, the change in charge, valence and spin\n(magnetic moment) with the localization of eg-electrons on the manganese ions\nhave been detected. With decrease in a temperature, the overstated effective\nmagnetic moment of Mn ions in paramagnetic phase is reduced dramatically\ntestifying to spin-dependent magnetism. The critical behavior of magnetization\nwith determination of critical parameters near second-order phase transition\nhas been studied comprehensively. Based on unusual behavior of temperature and\nfield dependences of magnetic entropy change under different magnetic field, an\nadditional influence of superparamagnetism of nanoparticles on the\nmagnetocaloric effect has been found.\n"}
{"abstract": "  The Deep Underground Neutrino Experiment (DUNE) is a next-generation\nlong-baseline neutrino oscillation experiment consisting of a high-power,\nbroadband neutrino beam, a highly capable near detector located on site at\nFermilab, in Batavia, Illinois, and a massive liquid argon time projection\nchamber (LArTPC) far detector located at the 4850L of Sanford Underground\nResearch Facility in Lead, South Dakota. The long-baseline physics sensitivity\ncalculations presented in the DUNE Physics TDR, and in a related physics paper,\nrely upon simulation of the neutrino beam line, simulation of neutrino\ninteractions in the near and far detectors, fully automated event\nreconstruction and neutrino classification, and detailed implementation of\nsystematic uncertainties. The purpose of this posting is to provide a\nsimplified summary of the simulations that went into this analysis to the\ncommunity, in order to facilitate phenomenological studies of long-baseline\noscillation at DUNE. Simulated neutrino flux files and a GLoBES configuration\ndescribing the far detector reconstruction and selection performance are\nincluded as ancillary files to this posting. A simple analysis using these\nconfigurations in GLoBES produces sensitivity that is similar, but not\nidentical, to the official DUNE sensitivity. DUNE welcomes those interested in\nperforming phenomenological work as members of the collaboration, but also\nrecognizes the benefit of making these configurations readily available to the\nwider community.\n"}
{"abstract": "  A method for the detection of changes in the expectation in univariate\nsequences is provided. Moving sum processes are studied. These rely on the\nselection of a tuning bandwidth. Here, a framework to overcome bandwidth\nselection is presented - the bandwidth adjusts gradually. For that, moving sum\nprocesses are made dependent on both time and the bandwidth: the domain becomes\na triangle. On the triangle, paths are constructed which systematically lead to\nchange points. An algorithm is provided that estimates change points by\nsubsequent consideration of paths. Strong consistency for the number and\nlocation of change points is shown. Simulations support estimation precision. A\ncompanion R-package mscp is made available on CRAN.\n"}
{"abstract": "  We present a Satisfiability (SAT)-based approach for building Mixed Covering\nArrays with Constraints of minimum length, referred to as the Covering Array\nNumber problem. This problem is central in Combinatorial Testing for the\ndetection of system failures. In particular, we show how to apply Maximum\nSatisfiability (MaxSAT) technology by describing efficient encodings for\ndifferent classes of complete and incomplete MaxSAT solvers to compute optimal\nand suboptimal solutions, respectively. Similarly, we show how to solve through\nMaxSAT technology a closely related problem, the Tuple Number problem, which we\nextend to incorporate constraints. For this problem, we additionally provide a\nnew MaxSAT-based incomplete algorithm. The extensive experimental evaluation we\ncarry out on the available Mixed Covering Arrays with Constraints benchmarks\nand the comparison with state-of-the-art tools confirm the good performance of\nour approaches.\n"}
{"abstract": "  The magnetic van der Waals crystals MnBi2Te4/(Bi2Te3)n have drawn significant\nattention due to their rich topological properties and the tunability by\nexternal magnetic field. Although the MnBi2Te4/(Bi2Te3)n family have been\nintensively studied in the past few years, their close relatives, the\nMnSb2Te4/(Sb2Te3)n family, remain much less explored. In this work, combining\nmagnetotransport measurements, angle-resolved photoemission spectroscopy, and\nfirst principles calculations, we find that MnSb4Te7, the n = 1 member of the\nMnSb2Te4/(Sb2Te3)n family, is a magnetic topological system with versatile\ntopological phases which can be manipulated by both carrier doping and magnetic\nfield. Our calculations unveil that its A-type antiferromagnetic (AFM) ground\nstate stays in a Z_2 AFM topological insulator phase, which can be converted to\nan inversion-symmetry-protected axion insulator phase when in the ferromagnetic\n(FM) state. Moreover, when this system in the FM phase is slightly carrier\ndoped on either the electron or hole side, it becomes a Weyl semimetal with\nmultiple Weyl nodes in the highest valence bands and lowest conduction bands,\nwhich are manifested by the measured notable anomalous Hall effect. Our work\nthus introduces a new magnetic topological material with different topological\nphases which are highly tunable by carrier doping or magnetic field.\n"}
{"abstract": "  A number of recent studies indicate that the charge conduction of the\nLaAlO$_3$/SrTiO$_3$ interface at low temperature is confined to filaments which\nare linked to structural domain walls in the SrTiO$_3$ with drastic\nconsequences for example for the temperature dependence of local transport\nproperties. We demonstrate that as a consequences of this current carrying\nfilaments on the nano-scale the magnetotransport properties of the interface\nare highly anisotropic. Our magnetoresistance measurements reveal that the\nmagnetoresistance in different nanostructures ($<500nm$) is random in magnitude\nand sign, respectively. Warming up nanostructures above the structural phase\ntransition temperature (105K) results in the significant change in MR. Even a\nsign change of the magnetoresistance is possible. The results suggest that\ndomain walls that are differently oriented with respect to the surface exhibit\ndifferent respective magnetoresistance and the total magnetoresistance is a\nresult of a random domain wall pattern formed during the structural phase\ntransition in the SrTiO$_3$ at cool down.\n"}
{"abstract": "  We have successfully demonstrated electrical charging using the electromotive\nforce (EMF) generated in a ferromagnetic metal (FM) film under ferromagnetic\nresonance (FMR). In the case of Ni80Fe20 films, electrical charge due to the\nEMF generated under FMR can be accumulated in a capacitor; however, the amount\nof charge is saturated well below the charging limit of the capacitor.\nMeanwhile in the case of Co50Fe50, electrical charge generated under FMR can be\naccumulated in a capacitor and the amount of charge increases linearly with the\nFMR duration time. The difference between the Ni80Fe20 and Co50Fe50 films is\ndue to the respective magnetic field ranges for the FMR excitation. When the FM\nfilms were in equivalent thermal states during FMR experiments, Co50Fe50 films\ncould maintain FMR in a detuned condition, while Ni80Fe20 films were outside\nthe FMR excitation range. The EMF generation phenomenon in an FM film under FMR\ncan be used an energy harvesting technology by appropriately controlling the\nthermal conditions of the FM film.\n"}
{"abstract": "  A promising characteristic of Deep Reinforcement Learning (DRL) is its\ncapability to learn optimal policy in an end-to-end manner without relying on\nfeature engineering. However, most approaches assume a fully observable state\nspace, i.e. fully observable Markov Decision Processes (MDPs). In real-world\nrobotics, this assumption is unpractical, because of issues such as sensor\nsensitivity limitations and sensor noise, and the lack of knowledge about\nwhether the observation design is complete or not. These scenarios lead to\nPartially Observable MDPs (POMDPs). In this paper, we propose\nLong-Short-Term-Memory-based Twin Delayed Deep Deterministic Policy Gradient\n(LSTM-TD3) by introducing a memory component to TD3, and compare its\nperformance with other DRL algorithms in both MDPs and POMDPs. Our results\ndemonstrate the significant advantages of the memory component in addressing\nPOMDPs, including the ability to handle missing and noisy observation data.\n"}
{"abstract": "  Recently it was proposed that microscopic models of braneworld cosmology\ncould be realized in the context of AdS/CFT using black hole microstates\ncontaining an end-of-the-world brane. Motivated by a desire to establish the\nmicroscopic existence of such microstates, which so far have been discussed\nprimarily in bottom-up models, we have studied similar microstates in a simpler\nversion of AdS/CFT. On one side, we define and study boundary states in the\ncharged Sachdev-Ye-Kitaev model and show that these states typically look\nthermal with a certain pattern of symmetry breaking. On the other side, we\nstudy the dimensional reduction of microstates in Einstein-Maxwell theory\nfeaturing an end-of-the-world brane and show that they have an equivalent\ndescription in terms of 2D Jackiw-Teitelboim gravity coupled to an\nend-of-the-world particle. In particular, the same pattern of symmetry breaking\nis realized in both sides of the proposed duality. These results give\nsignificant evidence that such black hole microstates have a sensible\nmicroscopic realization.\n"}
{"abstract": "  In this paper, a hardware-optimized approach to emotion recognition based on\nthe efficient brain-inspired hyperdimensional computing (HDC) paradigm is\nproposed. Emotion recognition provides valuable information for human-computer\ninteractions, however the large number of input channels (>200) and modalities\n(>3) involved in emotion recognition are significantly expensive from a memory\nperspective. To address this, methods for memory reduction and optimization are\nproposed, including a novel approach that takes advantage of the combinatorial\nnature of the encoding process, and an elementary cellular automaton. HDC with\nearly sensor fusion is implemented alongside the proposed techniques achieving\ntwo-class multi-modal classification accuracies of >76% for valence and >73%\nfor arousal on the multi-modal AMIGOS and DEAP datasets, almost always better\nthan state of the art. The required vector storage is seamlessly reduced by 98%\nand the frequency of vector requests by at least 1/5. The results demonstrate\nthe potential of efficient hyperdimensional computing for low-power,\nmulti-channeled emotion recognition tasks.\n"}
{"abstract": "  Yet, their performance degrades in the presence of noisy labels at train\ntime. Inspired by the setting of learning with expert advice, where\nmultiplicative weights (MW) updates were recently shown to be robust to\nmoderate data corruptions in expert advice, we propose to use MW for\nreweighting examples during neural networks optimization. We theoretically\nestablish the convergence of our method when used with gradient descent and\nprove its advantage for label noise in 1d cases. We then validate empirically\nour findings for the general case by showing that MW improves neural networks\naccuracy in the presence of label noise on CIFAR-10, CIFAR-100 and Clothing1M.\nWe also show the impact of our approach on adversarial robustness.\n"}
{"abstract": "  The estimation of the polarization $P$ of extragalactic compact sources in\nCosmic Microwave Background images is a very important task in order to clean\nthese images for cosmological purposes -- as, for example, to constrain the\ntensor-to-scalar ratio of primordial fluctuations during inflation -- and also\nto obtain relevant astrophysical information about the compact sources\nthemselves in a frequency range, $\\nu \\sim 10$--$200$ GHz, where observations\nhave only very recently started to be available. In this paper we propose a\nBayesian maximum a posteriori (MAP) approach estimation scheme which\nincorporates prior information about the distribution of the polarization\nfraction of extragalactic compact sources between 1 and 100 GHz. We apply this\nBayesian scheme to white noise simulations and to more realistic simulations\nthat include CMB intensity, Galactic foregrounds and instrumental noise with\nthe characteristics of the QUIJOTE experiment Wide Survey at 11 GHz. Using\nthese simulations, we also compare our Bayesian method with the frequentist\nFiltered Fusion method that has been already used in WMAP data and in the\n\\emph{Planck} mission. We find that the Bayesian method allows us to decrease\nthe threshold for a feasible estimation of $P$ to levels below $\\sim 100$ mJy\n(as compared to $\\sim 500$ mJy that was the equivalent threshold for the\nfrequentist Filtered Fusion). We compare the bias introduced by the Bayesian\nmethod and find it to be small in absolute terms. Finally, we test the\nrobustness of the Bayesian estimator against uncertainties in the prior and in\nthe flux density of the sources. We find that the Bayesian estimator is robust\nagainst moderate changes in the parameters of the prior and almost insensitive\nto realistic errors in the estimated photometry of the sources.\n"}
{"abstract": "  We explore the implementation of hybridly protected quantum operations\ncombining the merits of holonomy, dynamical decoupling approach and\ndephasing-free feature based on a simple and experimentally achievable spin\nmodel. The implementation of the quantum operations can be achieved in\ndifferent physical systems with controllable parameters. The protected quantum\noperations are hence controllable, well-suited for resolving various quantum\ncomputation tasks, such as executing quantum error-correction codes or quantum\nerror mitigation. Our scheme is based on experimentally achievable Hamiltonian\nwith reduced requirement of computational resources and thus, it brings us\ncloser towards realizing protected quantum operations for resolving quantum\ncomputation tasks in near-term quantum devices.\n"}
{"abstract": "  We present methods for computing the distance from a Boolean polynomial on\n$m$ variables of degree $m-3$ (i.e., a member of the Reed-Muller code\n$RM(m-3,m)$) to the space of lower-degree polynomials ($RM(m-4,m)$). The\nmethods give verifiable certificates for both the lower and upper bounds on\nthis distance. By applying these methods to representative lists of\npolynomials, we show that the covering radius of $RM(4,8)$ in $RM(5,8)$ is 26\nand the covering radius of $RM(5,9)$ in $RM(6,9)$ is between 28 and 32\ninclusive, and we get improved lower bounds for higher~$m$. We also apply our\nmethods to various polynomials in the literature, thereby improving the known\nbounds on the distance from 2-resilient polynomials to $RM(m-4,m)$.\n"}
{"abstract": "  The interaction between the quantum emitter and topological photonic system\nmakes the photon behave in exotic ways. We here study the properties of a giant\natom coupled to two sites of a one-dimensional topological waveguide, which is\ndescribed by the Su-Schrieffer-Heeger (SSH) chain. We find that the giant atom\ncan act as an effective boundary and induce the chiral zero modes, which are\nsimilar to those in the SSH model with open boundary, for the waveguide under\nthe periodical boundary. Except for the boundary effect, we also find that the\ngiant atom can lift energy degeneracy inside the energy bands of the SSH chain\nand adjust spatial symmetry of the photon distributions for the states of the\ndressed giant atom and waveguide. That is, the giant atom can be used to change\nthe properties of the topological environment. Our work may stimulate more\nstudies on the interaction between matter and topological environment.\n"}
{"abstract": "  Atomically precise graphene nanoribbons are a promising emerging class of\ndesigner quantum materials with electronic properties that are tunable by\nchemical design. However, many challenges remain in the device integration of\nthese materials, especially regarding contacting strategies. We report on the\ndevice integration of uniaxially aligned and non-aligned 9-atom wide armchair\ngraphene nanoribbons (9-AGNRs) in a field-effect transistor geometry using\nelectron beam lithography-defined graphene electrodes. This approach yields\ncontrolled electrode geometries and enables higher fabrication throughput\ncompared to previous approaches using an electrical breakdown technique.\nThermal annealing is found to be a crucial step for successful device operation\nresulting in electronic transport characteristics showing a strong gate\ndependence. Raman spectroscopy confirms the integrity of the graphene\nelectrodes after patterning and of the GNRs after device integration. Our\nresults demonstrate the importance of the GNR-graphene electrode interface and\npave the way for GNR device integration with structurally well-defined\nelectrodes.\n"}
{"abstract": "  Using the gauge/string duality, we model a heavy quark-antiquark pair in a\ncolor singlet state moving through a cold medium and explore the consequences\nof temperature and velocity on string breaking. We show that the string\nbreaking distance slowly varies with temperature and velocity away from the\ncritical line but could fall near it.\n"}
{"abstract": "  We consider the optimal decision-making problem in a primary sample of\ninterest with multiple auxiliary sources available. The outcome of interest is\nlimited in the sense that it is only observed in the primary sample. In\nreality, such multiple data sources may belong to heterogeneous studies and\nthus cannot be combined directly. This paper proposes a new framework to handle\nheterogeneous studies and address the limited outcome simultaneously through a\nnovel calibrated optimal decision making (CODA) method, by leveraging the\ncommon intermediate outcomes in multiple data sources. Specifically, CODA\nallows the baseline covariates across different samples to have either\nhomogeneous or heterogeneous distributions. Under a mild and testable\nassumption that the conditional means of intermediate outcomes in different\nsamples are equal given baseline covariates and the treatment information, we\nshow that the proposed CODA estimator of the conditional mean outcome is\nasymptotically normal and more efficient than using the primary sample solely.\nIn addition, the variance of the CODA estimator can be easily obtained using\nthe simple plug-in method due to the rate double robustness. Extensive\nexperiments on simulated datasets demonstrate empirical validity and improved\nefficiency using CODA, followed by a real application to a MIMIC-III dataset as\nthe primary sample with the auxiliary data from eICU.\n"}
{"abstract": "  There were over 70,000 drug overdose deaths in the USA in 2017. Almost half\nof those involved the use of Opioids such as Heroin. This research supports\nefforts to combat the Opioid Epidemic by further understanding factors that\nlead to Heroin consumption. Previous research has debated the cause of Heroin\naddiction, with some explaining the phenomenon as a transition from\nprescription Opioids, and others pointing to various psycho-social factors.\nThis research used self-reported information about personality, demographics\nand drug consumption behavior to predict Heroin consumption. By applying a\nSupport Vector Machine algorithm optimized with a Genetic Algorithm (GA-SVM\nHybrid) to simultaneously identify predictive features and model parameters,\nthis research produced several models that were more accurate in predicting\nHeroin use than those produced in previous studies. Although all factors had\npredictive power, these results showed that consumption of other drugs (both\nprescription and illicit) were stronger predictors of Heroin use than\npsycho-social factors. The use of prescription drugs as a strong predictor of\nHeroin use is an important though disturbing discovery but that can help combat\nHeroin use.\n"}
{"abstract": "  Pretrained multilingual models are able to perform cross-lingual transfer in\na zero-shot setting, even for languages unseen during pretraining. However,\nprior work evaluating performance on unseen languages has largely been limited\nto low-level, syntactic tasks, and it remains unclear if zero-shot learning of\nhigh-level, semantic tasks is possible for unseen languages. To explore this\nquestion, we present AmericasNLI, an extension of XNLI (Conneau et al., 2018)\nto 10 indigenous languages of the Americas. We conduct experiments with XLM-R,\ntesting multiple zero-shot and translation-based approaches. Additionally, we\nexplore model adaptation via continued pretraining and provide an analysis of\nthe dataset by considering hypothesis-only models. We find that XLM-R's\nzero-shot performance is poor for all 10 languages, with an average performance\nof 38.62%. Continued pretraining offers improvements, with an average accuracy\nof 44.05%. Surprisingly, training on poorly translated data by far outperforms\nall other methods with an accuracy of 48.72%.\n"}
{"abstract": "  Insurance data can be asymmetric with heavy tails, causing inadequate\nadjustments of the usually applied models. To deal with this issue,\nhierarchical models for collective risk with heavy-tails of the claims\ndistributions that take also into account overdispersion of the number of\nclaims are proposed. In particular, the distribution of the logarithm of the\naggregate value of claims is assumed to follow a Student-t distribution.\nAdditionally, to incorporate possible overdispersion, the number of claims is\nmodeled as having a negative binomial distribution. Bayesian decision theory is\ninvoked to calculate the fair premium based on the modified absolute deviation\nutility. An application to a health insurance dataset is presented together\nwith some diagnostic measures to identify excess variability. The variability\nmeasures are analyzed using the marginal posterior predictive distribution of\nthe premiums according to some competitive models. Finally, a simulation study\nis carried out to assess the predictive capability of the model and the\nadequacy of the Bayesian estimation procedure.\n  Keywords: Continuous ranked probability score (CRPS); decision theory;\ninsurance data; marginal posterior predictive; tail value at risk; value at\nrisk.\n"}
{"abstract": "  Superadditivity relations characterize the distributions of coherence in\nmultipartite quantum systems. In this work, we investigate the superadditivity\nrelations related to the $l_1$-norm of coherence $C_{l_1}$ in multiqubit\nquantum systems. Tighter superadditivity inequalities based on the $\\alpha$-th\n($\\alpha\\geqslant 1$) power of $l_1$-norm of coherence are presented for\nmultiqubit states under certain conditions, which include the existing results\nas special cases. These superadditivity relations give rise to finer\ncharacterization of the coherence distributions among the subsystems of a\nmultipartite system. A detailed example is presented.\n"}
{"abstract": "  This paper investigates how to correct Chinese text errors with types of\nmistaken, missing and redundant characters, which is common for Chinese native\nspeakers. Most existing models based on detect-correct framework can correct\nmistaken characters errors, but they cannot deal with missing or redundant\ncharacters. The reason is that lengths of sentences before and after correction\nare not the same, leading to the inconsistence between model inputs and\noutputs. Although the Seq2Seq-based or sequence tagging methods provide\nsolutions to the problem and achieved relatively good results on English\ncontext, but they do not perform well in Chinese context according to our\nexperimental results. In our work, we propose a novel detect-correct framework\nwhich is alignment-agnostic, meaning that it can handle both text aligned and\nnon-aligned occasions, and it can also serve as a cold start model when there\nare no annotated data provided. Experimental results on three datasets\ndemonstrate that our method is effective and achieves the best performance\namong existing published models.\n"}
{"abstract": "  In this paper we present a hybrid feedback approach to solve the navigation\nproblem of a point mass in the n-dimensional space containing an arbitrary\nnumber of ellipsoidal shape obstacles. The proposed hybrid control algorithm\nguarantees both global asymptotic stabilization to a reference and avoidance of\nthe obstacles. The intuitive idea of the proposed hybrid feedback is to switch\nbetween two modes of control: stabilization and avoidance. The geometric\nconstruction of the flow and jump sets of the proposed hybrid controller,\nexploiting hysteresis regions, guarantees Zeno-free switching between the\nstabilization and the avoidance modes. Simulation results illustrate the\nperformance of the proposed hybrid control approach for 2-dimensional and\n3-dimensional scenarios.\n"}
{"abstract": "  The article is about algorithms for learning Bayesian hierarchical models,\nthe computational complexity of which scales linearly with the number of\nobservations and the number of parameters in the model. It focuses on crossed\nrandom effect and nested multilevel models, which are used ubiquitously in\napplied sciences, and illustrates the methodology on two challenging real data\nanalyses on predicting electoral results and real estate prices respectively.\nThe posterior dependence in both classes is sparse: in crossed random effects\nmodels it resembles a random graph, whereas in nested multilevel models it is\ntree-structured. For each class we develop a framework for scalable\ncomputation. We provide a number of negative (for crossed) and positive (for\nnested) results for the scalability (or lack thereof) of methods based on\nsparse linear algebra, which are relevant also to Laplace approximation methods\nfor such models. Our numerical experiments compare with off-the-shelf\nvariational approximations and Hamiltonian Monte Carlo. Our theoretical\nresults, although partial, are useful in suggesting interesting methodologies\nand lead to conclusions that our numerics suggest to hold well beyond the scope\nof the underlying assumptions.\n"}
{"abstract": "  In this paper, we consider the following three dimensional defocusing cubic\nnonlinear Schr\\\"odinger equation (NLS) with partial harmonic potential\n  \\begin{equation*}\\tag{NLS}\n  i\\partial_t u + \\left(\\Delta_{\\mathbb{R}^3 }-x^2 \\right) u = |u|^2 u, \\quad\nu|_{t=0} = u_0.\n  \\end{equation*}\n  Our main result shows that the solution $u$ scatters for any given initial\ndata $u_0$ with finite mass and energy.\n  The main new ingredient in our approach is to approximate (NLS) in the\nlarge-scale case by a relevant dispersive continuous resonant (DCR) system. The\nproof of global well-posedness and scattering of the new (DCR) system is\ngreatly inspired by the fundamental works of Dodson \\cite{D3,D1,D2} in his\nstudy of scattering for the mass-critical nonlinear Schr\\\"odinger equation. The\nanalysis of (DCR) system allows us to utilize the additional regularity of the\nsmooth nonlinear profile so that the celebrated\nconcentration-compactness/rigidity argument of Kenig and Merle applies.\n"}
{"abstract": "  Hyperbolic geometry has been successfully applied in modeling brain cortical\nand subcortical surfaces with general topological structures. However such\napproaches, similar to other surface based brain morphology analysis methods,\nusually generate high dimensional features. It limits their statistical power\nin cognitive decline prediction research, especially in datasets with limited\nsubject numbers. To address the above limitation, we propose a novel framework\ntermed as hyperbolic stochastic coding (HSC). Our preliminary experimental\nresults show that our algorithm achieves superior results on various\nclassification tasks. Our work may enrich surface based brain imaging research\ntools and potentially result in a diagnostic and prognostic indicator to be\nuseful in individualized treatment strategies.\n"}
{"abstract": "  This paper addresses the energy management of a grid-connected photovoltaic\nplant coupled with a battery energy storage device, within the capacity firming\nspecifications of the French Energy Regulatory Commission. The paper\ncontributions are positioned in the continuity of the studies adopting\nstochastic models for optimizing the bids of renewable energy sources in a\nday-ahead market by considering a storage device. The proposed deterministic\nand stochastic approaches are optimization problems formulated as quadratic\nproblems with linear constraints. The case study is a real microgrid with PV\nproduction monitored on-site. The results demonstrate the validity of the\nstochastic formulation by using an ideal predictor that produces unbiased PV\nscenarios.\n"}
{"abstract": "  Argumentation mining is a field of computational linguistics that is devoted\nto extracting from texts and classifying arguments and relations between them,\nas well as constructing an argumentative structure. A significant obstacle to\nresearch in this area for the Russian language is the lack of annotated\nRussian-language text corpora. This article explores the possibility of\nimproving the quality of argumentation mining using the extension of the\nRussian-language version of the Argumentative Microtext Corpus (ArgMicro) based\non the machine translation of the Persuasive Essays Corpus (PersEssays). To\nmake it possible to use these two corpora combined, we propose a Joint Argument\nAnnotation Scheme based on the schemes used in ArgMicro and PersEssays. We\nsolve the problem of classifying argumentative discourse units (ADUs) into two\nclasses - \"pro\" (\"for\") and \"opp\" (\"against\") using traditional machine\nlearning techniques (SVM, Bagging and XGBoost) and a deep neural network (BERT\nmodel). An ensemble of XGBoost and BERT models was proposed, which showed the\nhighest performance of ADUs classification for both corpora.\n"}
{"abstract": "  Empirical, theoretical and historical aspects of Post's \"problem of tag\" from\n1921 are explored. Evidence of strong computational irreducibility is found.\nDespite their deterministic origin, the lengths of successive sequences\ngenerated seem to closely approximate random walks. All 10^25 smallest initial\nconditions are found to eventually halt, although sometimes in > 6*10^11 steps.\nImplications of the Principle of Computational Equivalence are discussed, along\nwith examples of identifiable computational capabilities of tag systems.\nVarious minimal examples of complex behavior are found, including a less-biased\nanalog of the 3n+1 Collatz problem. There is also discussion of the history of\nEmil Post and of tag systems in the context of ideas about the foundations of\nmathematics and computation.\n"}
{"abstract": "  In recent years, there has been an ever increasing amount of multivariate\ntime series (MTS) data in various domains, typically generated by a large\nfamily of sensors such as wearable devices. This has led to the development of\nnovel learning methods on MTS data, with deep learning models dominating the\nmost recent advancements. Prior literature has primarily focused on designing\nnew network architectures for modeling temporal dependencies within MTS.\nHowever, a less studied challenge is associated with high dimensionality of MTS\ndata. In this paper, we propose a novel neural component, namely Neural Feature\nSelector (NFS), as an end-2-end solution for feature selection in MTS data.\nSpecifically, NFS is based on decomposed convolution design and includes two\nmodules: firstly each feature stream (a stream corresponds to an univariate\nseries of MTS) within MTS is processed by a temporal CNN independently; then an\naggregating CNN combines the processed streams to produce input for other\ndownstream networks. We evaluated the proposed NFS model on four real-world MTS\ndatasets and found that it achieves comparable results with state-of-the-art\nmethods while providing the benefit of feature selection. Our paper also\nhighlights the robustness and effectiveness of feature selection with NFS\ncompared to using recent autoencoder-based methods.\n"}
{"abstract": "  The twenty-first century is the century of encounter of the different races,\nnations, cultures, religions and customs. In the twenty-first century, man is\nmore and more exposed to various influences that leave a trace on the entire\nsphere of his social life, including education. Given that education systems\nplay one of the key roles in the formation of both physically and morally\nhealthy communities, it is of an enormous importance to analyze the phenomenon\nof a classroom composed of culturally diverse students.\n"}
{"abstract": "  These are lecture notes (by the first author) from a course (by the second\nauthor) given over two extended semesters at the University of Sydney. The\nfirst part provides an introduction to the Langlands correspondence from an\narithmetical point of view. The second part gives enough background in\ngeometric representation theory to understand Bezrukavnikov's equivalence,\nwhich is a categorification of Kazhdan and Lusztig's two realizations of the\naffine Hecke algebra.\n"}
{"abstract": "  The radiative transport equation is considered in the spatial frequency\ndomain. We extend the analytical discrete-ordinates method to three dimensions\nby rotating reference frames.\n"}
{"abstract": "  In a recent report, Carballo-Rubio [1] utilizes the semi-classical theory of\ngravity to obtain a generalized Tolman-Oppenheimer-Volkoff (TOV) equation. This\nmodel has a new coupling constant $l_p$ implying two different modified TOV\nequation forms characterized by the sign of $p'$. The negative branch reduces\nto the ordinary GR-TOV in the limit of $l_p\\to0$, while the positive one does\nnot. In the positive branch, Carballo-Rubio was able to find the exact\nsolutions using the constant-$\\lambda$ trick. In this work, we investigate\nwhether this theory's negative branch can also provide a different feature of\nthe ultra-compact object compared to those obtained from the GR-TOV equation.\nWe study ultra-compact objects with an isotropically ideal fluid matter where\nwe use a simple but physically motivated equation of state $\\rho=p/w+\\rho_0$\nwith $w$=1 and $w$=1/3. In general, we obtain that the range of $l_p$ is very\nrestricted and must not be equal to $r_c$. Here $r_c$ is the starting point of\nintegration located at the center of the star. While $l_p$ should be set to be\nmuch larger than Planck length $L_\\text{Pl}$. Consequently, the mass-radius\ncurves for the various value of $l_p$ for both $w$ cases are still\nindistinguishable from the standard GR-TOV results. Hence from the negative\nbranch of $p'(r)$, the additional free parameter $l_p$ does not provide a\nsignificant effect compared to the standard GR-TOV equation results, even\nthough $l_p$ is not in the limit of $l_p\\to0$ anymore. Therefore, similar to\nthe conclusion in Ref. [3] with GR theory that the ultra-compact objects from\nnegative branch of semi-classical gravity with a linear equation of state are\nunable to generate demanding gravitational echoes.\n"}
{"abstract": "  The high-Chern number phases with a Chern number C>1 have been observed in a\nrecent experiment that performed on the topological insulator (TI) multilayer\nstructures, consisting of the alternating magnetic-doped and undoped TI layers.\nIn this paper, we develop an effective method to determine the Chern numbers in\nthe TI multilayer structures and then make a systematic study on the Chern\nnumber phase diagrams that are modulated by the magnetic doping and the middle\nlayer thickness. We point out that in the multilayer structure, the high-C\nbehavior can be attributed to the band inversion mechanisms. Moreover, we find\nthat the lowest bands may be multifold degenerate around the Gamma point, and\nwhen they are inverted, the Chern number change will be larger than one.\nBesides the TI multilayer structures implemented in the experiment, we also\nexplore the high-C phase realizations in two other kinds of the TI multilayer\nstructures. The implications of our results for experiments are discussed.\n"}
{"abstract": "  The Sun, as an active star, is the driver of energetic phenomena that\nstructure interplanetary space and affect planetary atmospheres. The effects of\nSpace Weather on Earth and the solar system is of increasing importance as\nhuman spaceflight is preparing for lunar and Mars missions. This review is\nfocusing on the solar perspective of the Space Weather relevant phenomena,\ncoronal mass ejections (CMEs), flares, solar energetic particles (SEPs), and\nsolar wind stream interaction regions (SIR). With the advent of the STEREO\nmission (launched in 2006), literally, new perspectives were provided that\nenabled for the first time to study coronal structures and the evolution of\nactivity phenomena in three dimensions. New imaging capabilities, covering the\nentire Sun-Earth distance range, allowed to seamlessly connect CMEs and their\ninterplanetary counterparts measured in-situ (so called ICMEs). This vastly\nincreased our knowledge and understanding of the dynamics of interplanetary\nspace due to solar activity and fostered the development of Space Weather\nforecasting models. Moreover, we are facing challenging times gathering new\ndata from two extraordinary missions, NASA's Parker Solar Probe (launched in\n2018) and ESA's Solar Orbiter (launched in 2020), that will in the near future\nprovide more detailed insight into the solar wind evolution and image CMEs from\nview points never approached before. The current review builds upon the Living\nReviews paper by Schwenn from 2006, updating on the Space Weather relevant\nCME-flare-SEP phenomena from the solar perspective, as observed from multiple\nviewpoints and their concomitant solar surface signatures.\n"}
{"abstract": "  We propose a compact and efficient tetrahedral mesh representation to improve\nthe ray-tracing performance. We reorder tetrahedral mesh data using a\nspace-filling curve to improve cache locality. Most importantly, we propose an\nefficient ray traversal algorithm. We provide details of common ray tracing\noperations on tetrahedral meshes and give the GPU implementation of our\ntraversal method. We demonstrate our findings through a set of comprehensive\nexperiments. Our method outperforms existing tetrahedral mesh-based traversal\nmethods and yields comparable results to the traversal methods based on the\nstate of the art acceleration structures such as k-dimensional (k-d) trees and\nBounding Volume Hierarchies (BVHs).\n"}
{"abstract": "  In this paper we pose the question of how one can quantify the amount of\nentanglement necessary to create a given steering assemblage, when the original\nstate and measurements are unknown. To do this, we extend the concepts of\nentanglement cost and entanglement of formation to steering assemblages, and\nshow how to upper bound them with semidefinite programming. We prove that the\nentanglement of formation for assemblages is not generally continuous and not a\nflat roof extension; and use numerical analysis to illustrate these properties.\nFinally, we discuss the consequences of these results for\nassemblage-to-assemblage conversion and directions for further research.\n"}
{"abstract": "  To perform temporal and sequential machine learning tasks, the use of\nconventional Recurrent Neural Networks (RNNs) has been dwindling due to the\ntraining complexities of RNNs. To this end, accelerators for delayed feedback\nreservoir computing (DFRC) have attracted attention in lieu of RNNs, due to\ntheir simple hardware implementations. A typical implementation of a DFRC\naccelerator consists of a delay loop and a single nonlinear neuron, together\nacting as multiple virtual nodes for computing. In prior work, photonic DFRC\naccelerators have shown an undisputed advantage of fast computation over their\nelectronic counterparts. In this paper, we propose a more energy-efficient\nchip-scale DFRC accelerator that employs a silicon photonic microring (MR)\nbased nonlinear neuron along with on-chip photonic waveguides-based delayed\nfeedback loop. Our evaluations show that, compared to a well-known photonic\nDFRC accelerator from prior work, our proposed MR-based DFRC accelerator\nachieves 35% and 98.7% lower normalized root mean square error (NRMSE),\nrespectively, for the prediction tasks of NARMA10 and Santa Fe time series. In\naddition, our MR-based DFRC accelerator achieves 58.8% lower symbol error rate\n(SER) for the Non-Linear Channel Equalization task. Moreover, our MR-based DFRC\naccelerator has 98% and 93% faster training time, respectively, compared to an\nelectronic and a photonic DFRC accelerators from prior work.\n"}
{"abstract": "  This paper presents a novel neutral-pion reconstruction that takes advantage\nof the machine learning technique of semantic segmentation using MINERvA data\ncollected between 2013-2017, with an average neutrino energy of $6$ GeV.\nSemantic segmentation improves the purity of neutral pion reconstruction from\ntwo gammas from 71\\% to 89\\% and improves the efficiency of the reconstruction\nby approximately 40\\%. We demonstrate our method in a charged current neutral\npion production analysis where a single neutral pion is reconstructed. This\ntechnique is applicable to modern tracking calorimeters, such as the new\ngeneration of liquid-argon time projection chambers, exposed to neutrino beams\nwith $\\langle E_\\nu \\rangle$ between 1-10 GeV. In such experiments it can\nfacilitate the identification of ionization hits which are associated with\nelectromagnetic showers, thereby enabling improved reconstruction of\ncharged-current $\\nu_e$ events arising from $\\nu_{\\mu} \\rightarrow \\nu_{e}$\nappearance.\n"}
{"abstract": "  Pretrained language models demonstrate strong performance in most NLP tasks\nwhen fine-tuned on small task-specific datasets. Hence, these autoregressive\nmodels constitute ideal agents to operate in text-based environments where\nlanguage understanding and generative capabilities are essential. Nonetheless,\ncollecting expert demonstrations in such environments is a time-consuming\nendeavour. We introduce a two-stage procedure to learn from a small set of\ndemonstrations and further improve by interacting with an environment. We show\nthat language models fine-tuned with only 1.2% of the expert demonstrations and\na simple reinforcement learning algorithm achieve a 51% absolute improvement in\nsuccess rate over existing methods in the ALFWorld environment.\n"}
{"abstract": "  We address the problem of output prediction, ie. designing a model for\nautonomous nonlinear systems capable of forecasting their future observations.\nWe first define a general framework bringing together the necessary properties\nfor the development of such an output predictor. In particular, we look at this\nproblem from two different viewpoints, control theory and data-driven\ntechniques (machine learning), and try to formulate it in a consistent way,\nreducing the gap between the two fields. Building on this formulation and\nproblem definition, we propose a predictor structure based on the\nKazantzis-Kravaris/Luenberger (KKL) observer and we show that KKL fits well\ninto our general framework. Finally, we propose a constructive solution for\nthis predictor that solely relies on a small set of trajectories measured from\nthe system. Our experiments show that our solution allows to obtain an\nefficient predictor over a subset of the observation space.\n"}
{"abstract": "  A concise analytical formula is developed for the inverse of an invertible 3\nx 3 matrix using a telescoping method, and is generalized to larger square\nmatrices. The formula is confirmed using randomly generated matrices in Matlab\n"}
{"abstract": "  In this paper, kink scattering in the dimensional reduction of the bosonic\nsector of a one-parameter family of generalized Wess-Zumino models with three\nvacuum points is discussed. The value of the model parameter determines the\nspecific location of the vacua. The influence of the vacuum arrangements\n(evolving from three collinear vacua to three vacua placed at the vertices of\nan equilateral triangle) on the kink scattering is investigated. Two different\nregimes can be distinguished: in the first one, two symmetric BPS\nkinks/antikinks arise whereas in the second one a new different BPS\nkink/antikink emerges, with the exception of a three-fold rotational symmetry\ncase, where the three topological defects are identical. The scattering between\nthe two symmetric kinks is thoroughly analyzed. Two different scattering\nchannels have been found: kink-kink reflection and kink-kink hybridization. In\nthe last case, the collision between the two symmetric kinks gives rise to the\nthird different kink. Resonance phenomena also appear allowing a vibrating kink\nto split into two symmetric kinks moving away.\n"}
{"abstract": "  Machine learning potential enables molecular dynamics simulations of systems\nbeyond the capability of classical force fields. The traditional approach to\ndevelop structural sets for training machine learning potential typically\ngenerate a great number of redundant configurations, which will result in\nunnecessary computational costs. This work investigates the possibility of\nreducing redundancy in an initial data set containing 6183 configurations for a\nLi-Si machine learning potential. Starting from the initial data set, we\nconstructed a series of subsets ranging from 25 to 1500 configurations by\ncombining a structural similarity analysis algorithm and the farthest point\nsampling method. Results show that the machine learning potential trained from\na data set containing 400 configurations can achieve an accuracy comparable to\nthe one developed from the initial data set of 6183 configurations in\ndescribing potential energies, atomic forces, and structural properties of\nLi-Si systems. In addition, the redundancy reducing approach also demonstrates\nadvantages over the classic stochastic method for constructing a concise\ntraining data set for Li-Si systems.\n"}
{"abstract": "  Two-dimensional materials (2DMs) are a promising alternative to complement\nand upgrade high-frequency electronics. However, in order to boost their\nadoption, the availability of numerical tools and physically-based models able\nto support the experimental activities and to provide them with useful\nguidelines becomes essential. In this context, we propose a theoretical\napproach that combines numerical simulations and small-signal modeling to\nanalyze 2DM-based FETs for radio-frequency applications. This multi-scale\nscheme takes into account non-idealities, such as interface traps, carrier\nvelocity saturation, or short channel effects, by means of self-consistent\nphysics-based numerical calculations that later feed the circuit level via a\nsmall-signal model based on the dynamic intrinsic capacitances of the device.\nAt the circuit stage, the possibilities range from the evaluation of the\nperformance of a single device to the design of complex circuits combining\nmultiple transistors. In this work, we validate our scheme against experimental\nresults and exemplify its use and capability assessing the impact of the\nchannel scaling on the performance of MoS2-based FETs targeting RF\napplications.\n"}
{"abstract": "  Nystr\\\"om approximation is a fast randomized method that rapidly solves\nkernel ridge regression (KRR) problems through sub-sampling the n-by-n\nempirical kernel matrix appearing in the objective function. However, the\nperformance of such a sub-sampling method heavily relies on correctly\nestimating the statistical leverage scores for forming the sampling\ndistribution, which can be as costly as solving the original KRR. In this work,\nwe propose a linear time (modulo poly-log terms) algorithm to accurately\napproximate the statistical leverage scores in the stationary-kernel-based KRR\nwith theoretical guarantees. Particularly, by analyzing the first-order\ncondition of the KRR objective, we derive an analytic formula, which depends on\nboth the input distribution and the spectral density of stationary kernels, for\ncapturing the non-uniformity of the statistical leverage scores. Numerical\nexperiments demonstrate that with the same prediction accuracy our method is\norders of magnitude more efficient than existing methods in selecting the\nrepresentative sub-samples in the Nystr\\\"om approximation.\n"}
{"abstract": "  The goal of this work is the development of a motion model for sequentially\ntimed movement actions in robotic systems under specific consideration of\ntemporal stabilization, that is maintaining an approximately constant overall\nmovement time (isochronous behavior). This is demonstrated both in simulation\nand on a physical robotic system for the task of intercepting a moving target\nin three-dimensional space. Motivated from humanoid motion, timing plays a\nvital role to generate a naturalistic behavior in interaction with the dynamic\nenvironment as well as adaptively planning and executing action sequences\non-line. In biological systems, many of the physiological and anatomical\nfunctions follow a particular level of periodicity and stabilization, which\nexhibit a certain extent of resilience against external disturbances. A main\naspect thereof is stabilizing movement timing against limited perturbations.\nEspecially human arm movement, namely when it is tasked to reach a certain goal\npoint, pose or configuration, shows a stabilizing behavior. This work\nincorporates the utilization of an extended Kalman filter (EKF) which was\nimplemented to predict the target position while coping with non-linear system\ndynamics. The periodicity and temporal stabilization in biological systems was\nartificially generated by a Hopf oscillator, yielding a sinusoidal velocity\nprofile for smooth and repeatable motion.\n"}
{"abstract": "  Let $R=\\oplus_{n\\in \\N_0}R_n$ be a standard graded ring, $M$ be a finitely\ngenerated graded $R$-module and $R_+:=\\oplus_{n\\in \\N}R_n$ denotes the\nirrelevant ideal of $R$. In this paper, considering the new concept of linkage\nof ideals over a module, we study the graded components $H^i_{\\fa}(M)_n$ when\n$\\fa$ is an h-linked ideal over $M$. More precisely, we show that\n$H^i_{\\fa}(M)$ is tame in each of the following cases: \\begin{itemize}\n  \\item [(i)] $i=f_{\\fa}^{R_+}(M)$, the first integer $i$ for which\n$R_+\\nsubseteq \\sqrt{0:H^i_{\\fa}(M)}$;\n  \\item [(ii)] $i=\\cd(R_+,M)$, the last integer $i$ for which\n$H^{i}_{R_+}(M)\\neq 0$, and $\\fa=\\fb+R_+$ where $\\fb$ is an h-linked ideal with\n$R_+$ over $M$. \\end{itemize} Also, among other things, we describe the\ncomponents $H^i_{\\fa}(M)_n$ where $\\fa$ is radically h-$M$-licci with respect\nto $R_+$ of length 2.\n"}
{"abstract": "  We introduce a new quantized enveloping superalgebra\n$\\mathfrak{U}_q{\\mathfrak{p}}_n$ attached to the Lie superalgebra\n${\\mathfrak{p}}_n$ of type $P$. The superalgebra\n$\\mathfrak{U}_q{\\mathfrak{p}}_n$ is a quantization of a Lie bisuperalgebra\nstructure on ${\\mathfrak{p}}_n$ and we study some of its basic properties. We\nalso introduce the periplectic $q$-Brauer algebra and prove that it is the\ncentralizer of the $\\mathfrak{U}_q {\\mathfrak{p}}_n$-module structure on\n${\\mathbb C}(n|n)^{\\otimes l}$. We end by proposing a definition for a new\nperiplectic $q$-Schur superalgebra.\n"}
{"abstract": "  We propose a discontinuous least squares finite element method for solving\nthe Helmholtz equation. The method is based on the L2 norm least squares\nfunctional with the weak imposition of the continuity across the interior faces\nas well as the boundary conditions. We minimize the functional over the\ndiscontinuous polynomial spaces to seek numerical solutions. The wavenumber\nexplicit error estimates to our method are established. The optimal convergence\nrate in the energy norm with respect to a fixed wavenumber is attained. The\nleast squares functional can naturally serve as a posteriori estimator in the\nh-adaptive procedure. It is convenient to implement the code due to the usage\nof discontinuous elements. Numerical results in two and three dimensions are\npresented to verify the error estimates.\n"}
{"abstract": "  To generate safe and real-time trajectories for an autonomous vehicle in\ndynamic environments, path and speed decoupled planning methods are often\nconsidered. This paper studies speed planning, which mainly deals with dynamic\nobstacle avoidance given the planning path. The main challenges lie in the\ndecisions in non-convex space and the trade-off between safety, comfort and\nefficiency performances. This work uses dynamic programming to search heuristic\nwaypoints on the S-T graph and to construct convex feasible spaces. Further, a\npiecewise Bezier polynomials optimization approach with trapezoidal corridors\nis presented, which theoretically guarantees the safety and optimality of the\ntrajectory. The simulations verify the effectiveness of the proposed approach.\n"}
{"abstract": "  While multiple studies have explored the relation between inter-rater\nvariability and deep learning model uncertainty in medical segmentation tasks,\nlittle is known about the impact of individual rater style. This study\nquantifies rater style in the form of bias and consistency and explores their\nimpacts when used to train deep learning models. Two multi-rater public\ndatasets were used, consisting of brain multiple sclerosis lesion and spinal\ncord grey matter segmentation. On both datasets, results show a correlation\n($R^2 = 0.60$ and $0.93$) between rater bias and deep learning uncertainty. The\nimpact of label fusion between raters' annotations on this relationship is also\nexplored, and we show that multi-center consensuses are more effective than\nsingle-center consensuses to reduce uncertainty, since rater style is mostly\ncenter-specific.\n"}
{"abstract": "  We show how in the standard electroweak model three $SU(2)_L$ Nambu\nmonopoles, each carrying electromagnetic (EM) and Z- magnetic fluxes, can merge\n(through Z-strings) with a single $U(1)_Y$ Dirac monopole to yield a composite\nmonopole that only carries EM magnetic flux. Compatibility with the Dirac\nquantization condition requires this composite monopole to carry six quanta\n($12 \\pi /e$) of magnetic charge, independent of the electroweak mixing angle\n$\\theta_w$. The Dirac monopole is not regular at the origin and the energy of\nthe composite monopole is therefore divergent. We discuss how this problem is\ncured by embedding $U(1)_Y$ in a grand unified group such as $SU(5)$. A second\ncomposite configuration with only one Nambu monopole and a colored $U(1)_Y$\nDirac monopole that has minimal EM charge of $4\\pi/e$ is also described.\nFinally, there exists a configuration with an EM charge of $8\\pi/e$ as well as\nscreened color magnetic charge.\n"}
{"abstract": "  In this paper, we study the third-order nonlinear optical response due to\ntransitions between excitonic levels in two-dimensional transition metal\ndichalcogeniedes. To accomplish this, we use methods not applied to the\ndescription of excitons in two-dimensional materials so far and combined with a\nvariational approach to describe the $1s$ excitonic state. The aforementioned\ntransitions allow to probe dark states which are not revealed in absorption\nexperiments. We present general formulas capable of describing any third-order\nprocess. The specific case of two-photon absorption in WSe2 is studied. The\ncase of the circular well is also studied as a benchmark of the theory.\n"}
{"abstract": "  Recently, sparsification scale-spaces have been obtained as a sequence of\ninpainted images by gradually removing known image data. Thus, these\nscale-spaces rely on spatial sparsity. In the present paper, we show that\nsparsification of the co-domain, the set of admissible grey values, also\nconstitutes scale-spaces with induced hierarchical quantisation techniques.\nThese quantisation scale-spaces are closely tied to information theoretical\nmeasures for coding cost, and therefore particularly interesting for\ninpainting-based compression. Based on this observation, we propose a\nsparsification algorithm for the grey-value domain that outperforms uniform\nquantisation as well as classical clustering approaches.\n"}
{"abstract": "  Let $H$ be a cocommutative weak Hopf algebra and let $(B, \\varphi_{B})$ a\nweak left $H$-module algebra. In this paper, for a twisted convolution\ninvertible morphism $\\sigma:H\\otimes H\\rightarrow B$ we define its obstruction\n$\\theta_{\\sigma}$ as a degree three Sweedler 3-cocycle with values in the\ncenter of $B$. We obtain that the class of this obstruction vanish in third\nSweedler cohomology group $\\mathcal{H}^3_{\\varphi_{Z(B)}}(H, Z(B))$ if, and\nonly if, there exists a twisted convolution invertible 2-cocycle\n$\\alpha:H\\otimes H\\rightarrow B$ such that $H\\otimes B$ can be endowed with a\nweak crossed product structure with $\\alpha$ keeping a cohomological-like\nrelation with $\\sigma$. Then, as a consequence, the class of the obstruction of\n$\\sigma$ vanish if, and only if, there exists a cleft extension of $B$ by $H$.\n"}
{"abstract": "  Unsupervised Domain Adaptation (UDA) can tackle the challenge that\nconvolutional neural network(CNN)-based approaches for semantic segmentation\nheavily rely on the pixel-level annotated data, which is labor-intensive.\nHowever, existing UDA approaches in this regard inevitably require the full\naccess to source datasets to reduce the gap between the source and target\ndomains during model adaptation, which are impractical in the real scenarios\nwhere the source datasets are private, and thus cannot be released along with\nthe well-trained source models. To cope with this issue, we propose a\nsource-free domain adaptation framework for semantic segmentation, namely SFDA,\nin which only a well-trained source model and an unlabeled target domain\ndataset are available for adaptation. SFDA not only enables to recover and\npreserve the source domain knowledge from the source model via knowledge\ntransfer during model adaptation, but also distills valuable information from\nthe target domain for self-supervised learning. The pixel- and patch-level\noptimization objectives tailored for semantic segmentation are seamlessly\nintegrated in the framework. The extensive experimental results on numerous\nbenchmark datasets highlight the effectiveness of our framework against the\nexisting UDA approaches relying on source data.\n"}
{"abstract": "  The longstanding goals of federated learning (FL) require rigorous privacy\nguarantees and low communication overhead while holding a relatively high model\naccuracy. However, simultaneously achieving all the goals is extremely\nchallenging. In this paper, we propose a novel framework called hierarchical\nfederated learning (H-FL) to tackle this challenge. Considering the degradation\nof the model performance due to the statistic heterogeneity of the training\ndata, we devise a runtime distribution reconstruction strategy, which\nreallocates the clients appropriately and utilizes mediators to rearrange the\nlocal training of the clients. In addition, we design a compression-correction\nmechanism incorporated into H-FL to reduce the communication overhead while not\nsacrificing the model performance. To further provide privacy guarantees, we\nintroduce differential privacy while performing local training, which injects\nmoderate amount of noise into only part of the complete model. Experimental\nresults show that our H-FL framework achieves the state-of-art performance on\ndifferent datasets for the real-world image recognition tasks.\n"}
{"abstract": "  We present the BrainScaleS-2 mobile system as a compact analog inference\nengine based on the BrainScaleS-2 ASIC and demonstrate its capabilities at\nclassifying a medical electrocardiogram dataset. The analog network core of the\nASIC is utilized to perform the multiply-accumulate operations of a\nconvolutional deep neural network. We measure a total energy consumption of\n192uJ for the ASIC and achieve a classification time of 276us per\nelectrocardiographic patient sample. Patients with atrial fibrillation are\ncorrectly identified with a detection rate of 93.7(7)% at 14.0(10)% false\npositives. The system is directly applicable to edge inference applications due\nto its small size, power envelope and flexible I/O capabilities. Possible\nfuture applications can furthermore combine conventional machine learning\nlayers with online-learning in spiking neural networks on a single\nBrainScaleS-2 ASIC. The system has successfully participated and proven to\noperate reliably in the independently judged competition\n\"Pilotinnovationswettbewerb 'Energieeffizientes KI-System'\" of the German\nFederal Ministry of Education and Research (BMBF).\n"}
{"abstract": "  In runtime verification, a monitor watches a trace of a system and, if\npossible, decides after observing each finite prefix whether or not the unknown\ninfinite trace satisfies a given specification. We generalize the theory of\nruntime verification to monitors that attempt to estimate numerical values of\nquantitative trace properties (instead of attempting to conclude boolean values\nof trace specifications), such as maximal or average response time along a\ntrace. Quantitative monitors are approximate: with every finite prefix, they\ncan improve their estimate of the infinite trace's unknown property value.\nConsequently, quantitative monitors can be compared with regard to a\nprecision-cost trade-off: better approximations of the property value require\nmore monitor resources, such as states (in the case of finite-state monitors)\nor registers, and additional resources yield better approximations. We\nintroduce a formal framework for quantitative and approximate monitoring, show\nhow it conservatively generalizes the classical boolean setting for monitoring,\nand give several precision-cost trade-offs for monitors. For example, we prove\nthat there are quantitative properties for which every additional register\nimproves monitoring precision.\n"}
{"abstract": "  Federated learning is an emerging distributed machine learning framework for\nprivacy preservation. However, models trained in federated learning usually\nhave worse performance than those trained in the standard centralized learning\nmode, especially when the training data are not independent and identically\ndistributed (Non-IID) on the local devices. In this survey, we pro-vide a\ndetailed analysis of the influence of Non-IID data on both parametric and\nnon-parametric machine learning models in both horizontal and vertical\nfederated learning. In addition, cur-rent research work on handling challenges\nof Non-IID data in federated learning are reviewed, and both advantages and\ndisadvantages of these approaches are discussed. Finally, we suggest several\nfuture research directions before concluding the paper.\n"}
{"abstract": "  Deception detection is a task with many applications both in direct physical\nand in computer-mediated communication. Our focus is on automatic deception\ndetection in text across cultures. We view culture through the prism of the\nindividualism/collectivism dimension and we approximate culture by using\ncountry as a proxy. Having as a starting point recent conclusions drawn from\nthe social psychology discipline, we explore if differences in the usage of\nspecific linguistic features of deception across cultures can be confirmed and\nattributed to norms in respect to the individualism/collectivism divide. We\nalso investigate if a universal feature set for cross-cultural text deception\ndetection tasks exists. We evaluate the predictive power of different feature\nsets and approaches. We create culture/language-aware classifiers by\nexperimenting with a wide range of n-gram features based on phonology,\nmorphology and syntax, other linguistic cues like word and phoneme counts,\npronouns use, etc., and token embeddings. We conducted our experiments over 11\ndatasets from 5 languages i.e., English, Dutch, Russian, Spanish and Romanian,\nfrom six countries (US, Belgium, India, Russia, Mexico and Romania), and we\napplied two classification methods i.e, logistic regression and fine-tuned BERT\nmodels. The results showed that our task is fairly complex and demanding. There\nare indications that some linguistic cues of deception have cultural origins,\nand are consistent in the context of diverse domains and dataset settings for\nthe same language. This is more evident for the usage of pronouns and the\nexpression of sentiment in deceptive language. The results of this work show\nthat the automatic deception detection across cultures and languages cannot be\nhandled in a unified manner, and that such approaches should be augmented with\nknowledge about cultural differences and the domains of interest.\n"}
{"abstract": "  For a semisimple Lie algebra defined over a discrete valuation ring with\nfield of fractions $K$, we prove that any primitive ideal with rational central\ncharacter in the affinoid enveloping algebra, $\\widehat{U(\\mathfrak{g})_{K}},$\nis the annihilator of an affinoid highest weight module.\n"}
{"abstract": "  30 Doradus C is a superbubble which emits the brightest nonthermal X- and TeV\ngamma-rays in the Local Group. In order to explore detailed connection between\nthe high energy radiation and the interstellar medium, we have carried out new\nCO and HI observations using the Atacama Large Millimeter$/$Submillimeter Array\n(ALMA), Atacama Submillimeter Telescope Experiment, and the Australia Telescope\nCompact Array with resolutions of up to 3 pc. The ALMA data of $^{12}$CO($J$ =\n1-0) emission revealed 23 molecular clouds with the typical diameters of\n$\\sim$6-12 pc and masses of $\\sim$600-10000 $M_{\\odot}$. The comparison with\nthe X-rays of $XMM$-$Newton$ at $\\sim$3 pc resolution shows that X-rays are\nenhanced toward these clouds. The CO data were combined with the HI to estimate\nthe total interstellar protons. Comparison of the interstellar proton column\ndensity and the X-rays revealed that the X-rays are enhanced with the total\nproton. These are most likely due to the shock-cloud interaction modeled by the\nmagnetohydrodynamical simulations (Inoue et al. 2012, ApJ, 744, 71). Further,\nwe note a trend that the X-ray photon index varies with distance from the\ncenter of the high-mass star cluster, suggesting that the cosmic-ray electrons\nare accelerated by one or multiple supernovae in the cluster. Based on these\nresults we discuss the role of the interstellar medium in cosmic-ray particle\nacceleration.\n"}
{"abstract": "  Software-Defined Networking (SDN) is an evolutionary networking paradigm\nwhich has been adopted by large network and cloud providers, among which are\nTech Giants. However, embracing a new and futuristic paradigm as an alternative\nto well-established and mature legacy networking paradigm requires a lot of\ntime along with considerable financial resources and technical expertise.\nConsequently, many enterprises can not afford it. A compromise solution then is\na hybrid networking environment (a.k.a. Hybrid SDN (hSDN)) in which SDN\nfunctionalities are leveraged while existing traditional network\ninfrastructures are acknowledged. Recently, hSDN has been seen as a viable\nnetworking solution for a diverse range of businesses and organizations.\nAccordingly, the body of literature on hSDN research has improved remarkably.\nOn this account, we present this paper as a comprehensive state-of-the-art\nsurvey which expands upon hSDN from many different perspectives.\n"}
{"abstract": "  Liquid phase sensing applications at 1550~nm are highly desirable due to\nwidely available off-the-shelf components. Generally, liquids at 1550~nm induce\na high absorption loss that limits the overall sensor's sensitivity and\ndetection limit. One solution is to use an active fiber loop in conjunction\nwith cavity ring down spectroscopy to overcome these absorption losses.\nHowever, the amplifier inside the fiber loop suffers from inherent gain\nfluctuations that limit the sensing system's overall performance. Here, we\nprovide a novel sensor using the wavelength-scanned phase shift-cavity ring\ndown spectroscopy (PS-CRDS) in conjunction with a linear active fiber cavity\nthat potentially offers a more sensitive solution than traditional fiber loop\nsensors. We use a tapered fiber as a sensing head inside the active cavity\nbuilt from fiber Bragg gratings. We derive a theoretical phase shift expression\nfor our system and simulate it using the finite element method to determine\noptimum tapered fiber diameter for glucose sensing in DI water. Compared to a\nnon-amplified system, we find that our amplified system can increase the\nsensitivity by fourteen times via the amplifier gain tuning. We also conduct\nexperimental measurements using 0-15.5~mM glucose solutions and find them in\nexcellent agreement with our theoretical predictions. Experimentally we obtain\nthe sensor's sensitivity of 0.768~$^o$/mM (1164~$^o$/RIU) and detection limit\nof 0.75~mM ( 4.5~$\\times$~10$^{-4}$~RIU) without any temperature stabilization\nin the system. We anticipate that the present work will find a wide range of\nsensing applications in fiber cavities, ring resonators, and other microcavity\nstructures.\n"}
{"abstract": "  We introduce a mapping between Maximum Entropy Reinforcement Learning (MaxEnt\nRL) and Markovian processes conditioned on rare events. In the long time limit,\nthis mapping allows us to derive analytical expressions for the optimal policy,\ndynamics and initial state distributions for the general case of stochastic\ndynamics in MaxEnt RL. We find that soft-$\\mathcal{Q}$ functions in MaxEnt RL\ncan be obtained from the Perron-Frobenius eigenvalue and the corresponding left\neigenvector of a regular, non-negative matrix derived from the underlying\nMarkov Decision Process (MDP). The results derived lead to novel algorithms for\nmodel-based and model-free MaxEnt RL, which we validate by numerical\nsimulations. The mapping established in this work opens further avenues for the\napplication of novel analytical and computational approaches to problems in\nMaxEnt RL. We make our code available at:\nhttps://github.com/argearriojas/maxent-rl-mdp-scripts\n"}
{"abstract": "  The first-principles simulation of the electronic structure of organic\nsemiconductors in solution poses a number of challenges that are not trivial to\naddress simultaneously. In this work, we investigate the effects and the mutual\ninterplay of solvation, alkylization, and doping on the structural, electronic,\nand optical properties of sexithiophene, a representative organic semiconductor\nmolecule. To this end, we employ (time-dependent) density functional theory in\nconjunction with the polarizable-continuum model. We find that the torsion\nbetween adjacent monomer units plays a key role, as it strongly influences the\nelectronic structure of the molecule, including energy gap, ionization\npotential, and band widths. Alkylization promotes delocalization of the\nmolecular orbitals up to the first methyl unit, regardless of the chain length,\nleading to an overall shift of the energy levels. The alterations in the\nelectronic structure are reflected in the optical absorption, which is\nadditionally affected by dynamical solute-solvent interactions. Taking all\nthese effects into account, solvents decrease the optical gap by an amount that\ndepends on its polarity, and concomitantly increase the oscillator strength of\nthe first excitation. The interaction with a dopant molecule promotes\nplanarization. In such scenario, solvation and alkylization enhance charge\ntransfer both in the ground state and in the excited state.\n"}
{"abstract": "  Our objective is language-based search of large-scale image and video\ndatasets. For this task, the approach that consists of independently mapping\ntext and vision to a joint embedding space, a.k.a. dual encoders, is attractive\nas retrieval scales and is efficient for billions of images using approximate\nnearest neighbour search. An alternative approach of using vision-text\ntransformers with cross-attention gives considerable improvements in accuracy\nover the joint embeddings, but is often inapplicable in practice for\nlarge-scale retrieval given the cost of the cross-attention mechanisms required\nfor each sample at test time. This work combines the best of both worlds. We\nmake the following three contributions. First, we equip transformer-based\nmodels with a new fine-grained cross-attention architecture, providing\nsignificant improvements in retrieval accuracy whilst preserving scalability.\nSecond, we introduce a generic approach for combining a Fast dual encoder model\nwith our Slow but accurate transformer-based model via distillation and\nre-ranking. Finally, we validate our approach on the Flickr30K image dataset\nwhere we show an increase in inference speed by several orders of magnitude\nwhile having results competitive to the state of the art. We also extend our\nmethod to the video domain, improving the state of the art on the VATEX\ndataset.\n"}
{"abstract": "  We demonstrate single crystal growth of wafer-scale hexagonal boron nitride\n(hBN), an insulating atomic thin monolayer, on high-symmetry index surface\nplane Cu(111). The unidirectional epitaxial growth is guaranteed by large\nbinding energy difference, ~0.23 eV, between A- and B-steps edges on Cu(111)\ndocking with B6N7 clusters, confirmed by density functional theory\ncalculations.\n"}
{"abstract": "  Bimetallic nanoparticles are often superior candidates for a wide range of\ntechnological and biomedical applications, thanks to their enhanced catalytic,\noptical, and magnetic properties, which are often better than their\nmonometallic counterparts. Most of their properties strongly depend on their\nchemical composition, crystallographic structure, and phase distribution.\nHowever, little is known of how their crystal structure, on the nanoscale,\ntransforms over time at elevated temperatures, even though this knowledge is\nhighly relevant in case nanoparticles are used in, e.g., high-temperature\ncatalysis. Au-Fe is a promising bimetallic system where the low-cost and\nmagnetic Fe is combined with catalytically active and plasmonic Au. Here, we\nreport on the in situ temporal evolution of the crystalline ordering in Au-Fe\nnanoparticles, obtained from a modern laser ablation in liquids synthesis. Our\nin-depth analysis, complemented by dedicated atomistic simulations, includes a\ndetailed structural characterization by X-ray diffraction and transmission\nelectron microscopy as well as atom probe tomography to reveal elemental\ndistributions down to a single atom resolution. We show that the Au-Fe\nnanoparticles initially exhibit highly complex internal nested nanostructures\nwith a wide range of compositions, phase distributions, and size-depended\nmicrostrains. The elevated temperature induces a diffusion-controlled\nrecrystallization and phase merging, resulting in the formation of a single\nface-centered-cubic ultrastructure in contact with a body-centered cubic phase,\nwhich demonstrates the metastability of these structures. Uncovering these\nunique nanostructures with nested features could be highly attractive from a\nfundamental viewpoint as they could give further insights into the nanoparticle\nformation mechanism under non-equilibrium conditions.\n"}
{"abstract": "  Combinatorial Hantzsche-Wendt groups G(n) were defined by W. Craig and P. A.\nLinnell. For n = 2 it is a fundamental group of 3-dimensional oriented flat\nmanifold with no cyclic holonomy group. We calculate the Hilbert-Poincare\nseries of G(n), n > 0 with Q and F2 coefficients. Moreover, we prove that a\ncohomological dimension of G(n) is equal to n + 1. Some other properties of\nthis group are also considered.\n"}
{"abstract": "  Transfer learning is a standard technique to improve performance on tasks\nwith limited data. However, for medical imaging, the value of transfer learning\nis less clear. This is likely due to the large domain mismatch between the\nusual natural-image pre-training (e.g. ImageNet) and medical images. However,\nrecent advances in transfer learning have shown substantial improvements from\nscale. We investigate whether modern methods can change the fortune of transfer\nlearning for medical imaging. For this, we study the class of large-scale\npre-trained networks presented by Kolesnikov et al. on three diverse imaging\ntasks: chest radiography, mammography, and dermatology. We study both transfer\nperformance and critical properties for the deployment in the medical domain,\nincluding: out-of-distribution generalization, data-efficiency, sub-group\nfairness, and uncertainty estimation. Interestingly, we find that for some of\nthese properties transfer from natural to medical images is indeed extremely\neffective, but only when performed at sufficient scale.\n"}
{"abstract": "  A quantum Hall (QH) interface is different from an ordinary QH edge, as the\nlatter has its location determined by the confining potential, while the former\ncan be unpinned and behave like a free string. In this paper, we demonstrate\nthis difference by studying three different interfaces formed by (i) the\nLaughlin state and the vacuum, (ii) the Pfaffian state and the vacuum, and\n(iii) the Pfaffian and the anti-Pfaffian states. We find that string-like\ninterfaces propagating freely in the QH system lead to very different dynamical\nproperties from edges. This qualitative difference gives rise to fascinating\nnew physics and suggests a new direction in future research on QH physics. We\nalso discuss briefly possible analogies between QH interfaces and concepts in\nstring theory.\n"}
{"abstract": "  We introduce the concept of group state transfer on graphs, summarize its\nrelationship to other concepts in the theory of quantum walks, set up a basic\ntheory, and discuss examples.\n  Let $X$ be a graph with adjacency matrix $A$ and consider quantum walks on\nthe vertex set $V(X)$ governed by the continuous time-dependent unitary\ntransition operator $U(t)= \\exp(itA)$. For $S,T\\subseteq V(X)$, we says $X$\nadmits \"group state transfer\" from $S$ to $T$ at time $\\tau$ if the submatrix\nof $U(\\tau)$ obtained by restricting to columns in $S$ and rows not in $T$ is\nthe all-zero matrix. As a generalization of perfect state transfer, fractional\nrevival and periodicity, group state transfer satisfies natural monotonicity\nand transitivity properties. Yet non-trivial group state transfer is still\nrare; using a compactness argument, we prove that bijective group state\ntransfer (the optimal case where $|S|=|T|$) is absent for almost all $t$.\nFocusing on this bijective case, we obtain a structure theorem, prove that\nbijective group state transfer is \"monogamous\", and study the relationship\nbetween the projections of $S$ and $T$ into each eigenspace of the graph.\n  Group state transfer is obviously preserved by graph automorphisms and this\ngives us information about the relationship between the setwise stabilizer of\n$S\\subseteq V(X)$ and the stabilizers of naturally defined subsets obtained by\nspreading $S$ out over time and crudely reversing this process. These\noperations are sufficiently well-behaved to give us a topology on $V(X)$ which\nis likely to be simply the topology of subsets for which bijective group state\ntransfer occurs at that time. We illustrate non-trivial group state transfer in\nbipartite graphs with integer eigenvalues, in joins of graphs, and in symmetric\ndouble stars. The Cartesian product allows us to build new examples from old\nones.\n"}
{"abstract": "  In this paper, we explore multi-task learning (MTL) as a second pretraining\nstep to learn enhanced universal language representation for transformer\nlanguage models. We use the MTL enhanced representation across several natural\nlanguage understanding tasks to improve performance and generalization.\nMoreover, we incorporate knowledge distillation (KD) in MTL to further boost\nperformance and devise a KD variant that learns effectively from multiple\nteachers. By combining MTL and KD, we propose Robustly Optimized and Distilled\n(ROaD) modeling framework. We use ROaD together with the ELECTRA model to\nobtain state-of-the-art results for machine reading comprehension and natural\nlanguage inference.\n"}
{"abstract": "  A permutation $\\pi$ of a multiset is said to be a {\\em quasi-Stirling }\npermutation if there does not exist four indices $i<j<k<\\ell$ such that\n$\\pi_i=\\pi_k$ and $\\pi_j=\\pi_{\\ell}$. Define $$\n\\overline{Q}_{\\mathcal{M}}(t,u,v)=\\sum_{\\pi\\in\n\\overline{\\mathcal{Q}}_{\\mathcal{M}}}t^{des(\\pi)}u^{asc(\\pi)}v^{plat(\\pi)},$$\nwhere $\\overline{\\mathcal{Q}}_{\\mathcal{M}}$ denotes the set of quasi-Stirling\n  permutations on the multiset $\\mathcal{M}$, and $asc(\\pi)$ (resp. $des(\\pi)$,\n$plat(\\pi)$) denotes the number of ascents (resp. descents, plateaux) of $\\pi$.\nDenote by $\\mathcal{M}^{\\sigma}$ the multiset $\\{1^{\\sigma_1}, 2^{\\sigma_2},\n\\ldots, n^{\\sigma_n}\\}$, where $\\sigma=(\\sigma_1, \\sigma_2, \\ldots, \\sigma_n)$\nis an $n$-composition of $K$ for positive integers $K$ and $n$. In this paper,\nwe show that\n$\\overline{Q}_{\\mathcal{M}^{\\sigma}}(t,u,v)=\\overline{Q}_{\\mathcal{M}^{\\tau}}(t,u,v)$\nfor any two $n$-compositions $\\sigma$ and $\\tau$ of $K$. This is accomplished\nby establishing an $(asc, des, plat)$-preserving bijection\n  between $\\overline{\\mathcal{Q}}_{\\mathcal{M}^{\\sigma}}$ and\n$\\overline{\\mathcal{Q}}_{\\mathcal{M}^{\\tau}}$. As applications, we obtain\ngeneralizations of several results for quasi-Stirling permutations on\n$\\mathcal{M}=\\{1^k,2^k, \\ldots, n^k\\}$ obtained by Elizalde and solve an open\nproblem posed by Elizalde.\n"}
{"abstract": "  In this paper, we explore the collider phenomenology of the charged Higgs\nboson in the context of a generic Beyond Standard Model scenario with extended\ngauge and scalar sectors. In such scenarios, the charged Higgs boson can decay\nvia the $W^{'}Z/ WZ^{'}$ channels. We formulate a search strategy for the\n$H^{\\pm}$ in the channel $\\sigma(g b \\rightarrow H^{\\pm}t)\\mathcal{BR}(H^{\\pm}\n\\rightarrow W' Z)$ considering the interesting cascade decay chain $H^{\\pm}\n\\rightarrow W^{'} Z \\rightarrow W^{\\pm} Z Z$. We find that the charged Higgs\ncan be discovered in final states with multiple hard leptons and/ or b-quarks\nwhich future LHC experiments with sufficiently large luminosity ($\\mathcal{L} =\n1000 fb^{-1}$ and above) can probe.\n"}
{"abstract": "  There has been growing interest on forecasting mortality. In this article, we\npropose a novel dynamic Bayesian approach for modeling and forecasting the\nage-at-death distribution, focusing on a three-components mixture of a Dirac\nmass, a Gaussian distribution and a Skew-Normal distribution. According to the\nspecified model, the age-at-death distribution is characterized via seven\nparameters corresponding to the main aspects of infant, adult and old-age\nmortality. The proposed approach focuses on coherent modeling of multiple\ncountries, and following a Bayesian approach to inference we allow to borrow\ninformation across populations and to shrink parameters towards a common mean\nlevel, implicitly penalizing diverging scenarios. Dynamic modeling across years\nis induced trough an hierarchical dynamic prior distribution that allows to\ncharacterize the temporal evolution of each mortality component and to forecast\nthe age-at-death distribution. Empirical results on multiple countries indicate\nthat the proposed approach outperforms popular methods for forecasting\nmortality, providing interpretable insights on the evolution of mortality.\n"}
{"abstract": "  Face representation learning solutions have recently achieved great success\nfor various applications such as verification and identification. However, face\nrecognition approaches that are based purely on RGB images rely solely on\nintensity information, and therefore are more sensitive to facial variations,\nnotably pose, occlusions, and environmental changes such as illumination and\nbackground. A novel depth-guided attention mechanism is proposed for deep\nmulti-modal face recognition using low-cost RGB-D sensors. Our novel attention\nmechanism directs the deep network \"where to look\" for visual features in the\nRGB image by focusing the attention of the network using depth features\nextracted by a Convolution Neural Network (CNN). The depth features help the\nnetwork focus on regions of the face in the RGB image that contains more\nprominent person-specific information. Our attention mechanism then uses this\ncorrelation to generate an attention map for RGB images from the depth features\nextracted by CNN. We test our network on four public datasets, showing that the\nfeatures obtained by our proposed solution yield better results on the\nLock3DFace, CurtinFaces, IIIT-D RGB-D, and KaspAROV datasets which include\nchallenging variations in pose, occlusion, illumination, expression, and\ntime-lapse. Our solution achieves average (increased) accuracies of 87.3\\%\n(+5.0\\%), 99.1\\% (+0.9\\%), 99.7\\% (+0.6\\%) and 95.3\\%(+0.5\\%) for the four\ndatasets respectively, thereby improving the state-of-the-art. We also perform\nadditional experiments with thermal images, instead of depth images, showing\nthe high generalization ability of our solution when adopting other modalities\nfor guiding the attention mechanism instead of depth information\n"}
{"abstract": "  Product safety has been a concern in Europe ever since the early 1960s.\nDespite the long and relatively stable historical lineage of product safety\nregulations, new technologies, changes in the world economy, and other major\ntransformations have in recent years brought product safety again to the\nforefront of policy debates. As reforms are also underway, there is a\nmotivation to review the complex safety policy framework in the European Union\n(EU). Thus, building on deliberative policy analysis and interpretative\nliterature review, this paper reviews the safety policy for non-food consumer\nproducts in the EU. The review covers the historical background and the main\nlaws, administration and enforcement, standardization and harmonization, laws\nenacted for specific products, notifications delivered by national safety\nauthorities, recalls of dangerous products, and the liability of these. Based\non the review and analysis of these themes and the associated literature, some\ncurrent policy challenges are further discussed.\n"}
{"abstract": "  Anime is quite well-received today, especially among the younger generations.\nWith many genres of available shows, more and more people are increasingly\ngetting attracted to this niche section of the entertainment industry. As anime\nhas recently garnered mainstream attention, we have insufficient information\nregarding users' penchant and watching habits. Therefore, it is an uphill task\nto build a recommendation engine for this relatively obscure entertainment\nmedium. In this attempt, we have built a novel hybrid recommendation system\nthat could act both as a recommendation system and as a means of exploring new\nanime genres and titles. We have analyzed the general trends in this field and\nthe users' watching habits for coming up with our efficacious solution. Our\nsolution employs deep autoencoders for the tasks of predicting ratings and\ngenerating embeddings. Following this, we formed clusters using the embeddings\nof the anime titles. These clusters form the search space for anime with\nsimilarities and are used to find anime similar to the ones liked and disliked\nby the user. This method, combined with the predicted ratings, forms the novel\nhybrid filter. In this article, we have demonstrated this idea and compared the\nperformance of our implemented model with the existing state-of-the-art\ntechniques.\n"}
{"abstract": "  It is shown that every weak Jacobi form of weight zero and index one on a\ncongruence subgroup of the full Jacobi group can be decomposed into $N=4$\nsuperconformal characters. Additionally, a simple expression for the mock\nmodular form determining the superconformal character coefficients is obtained,\nas well as a universal completion structure. Along the way, a useful\nvector-valued mock modular form is also found and studied. These results are\napplied to analyze some Jacobi trace functions associated to super vertex\noperator algebras and a distinguished sector.\n"}
{"abstract": "  In this paper, we propose to unify the two aspects of voice synthesis, namely\ntext-to-speech (TTS) and vocoder, into one framework based on a pair of forward\nand reverse-time linear stochastic differential equations (SDE). The solutions\nof this SDE pair are two stochastic processes, one of which turns the\ndistribution of mel spectrogram (or wave), that we want to generate, into a\nsimple and tractable distribution. The other is the generation procedure that\nturns this tractable simple signal into the target mel spectrogram (or wave).\nThe model that generates mel spectrogram is called It$\\hat{\\text{o}}$TTS, and\nthe model that generates wave is called It$\\hat{\\text{o}}$Wave.\nIt$\\hat{\\text{o}}$TTS and It$\\hat{\\text{o}}$Wave use the Wiener process as a\ndriver to gradually subtract the excess signal from the noise signal to\ngenerate realistic corresponding meaningful mel spectrogram and audio\nrespectively, under the conditional inputs of original text or mel spectrogram.\nThe results of the experiment show that the mean opinion scores (MOS) of\nIt$\\hat{\\text{o}}$TTS and It$\\hat{\\text{o}}$Wave can exceed the current\nstate-of-the-art methods, and reached 3.925$\\pm$0.160 and 4.35$\\pm$0.115\nrespectively. The generated audio samples are available at\nhttps://shiziqiang.github.io/ito_audio/. All authors contribute equally to this\nwork.\n"}
{"abstract": "  In the infinite-dimensional separable complex Hilbert space we construct new\nabstract examples of unbounded maximal accretive and maximal sectorial\noperators $B$ for which ${\\rm dom\\,}B^{\\frac{1}{2}}\\ne{\\rm\ndom\\,}B^{*{\\frac{1}{2}}}$. New criterions for the equality are established.\n"}
{"abstract": "  Controlling the properties of organic/inorganic materials requires detailed\nknowledge of their molecular adsorption geometries. This is often unattainable,\neven with current state-of-the-art tools. Visualizing the structure of complex\nnon-planar adsorbates with atomic force microscopy (AFM) is challenging, and\nidentifying it computationally is intractable with conventional structure\nsearch. In this fresh approach, cross-disciplinary tools are integrated for a\nrobust and automated identification of 3D adsorbate configurations. Bayesian\noptimization is employed with first-principles simulations for accurate and\nunbiased structure inference of multiple adsorbates. The corresponding AFM\nsimulations then allow fingerprinting adsorbate structures that appear in AFM\nexperimental images. In the instance of bulky (1S)-camphor adsorbed on the\nCu(111) surface, three matching AFM image contrasts are found, which allow\ncorrelating experimental image features to distinct cases of molecular\nadsorption.\n"}
{"abstract": "  Millimeter wave (mmWave) communications are vulnerable to blockages and node\nmobility due to the highly directional signal beams. The emerging\nReconfigurable Intelligent Surfaces (RISs) technique can effectively mitigate\nthe blockage problem by exploring the non-line-of-sight (NLOS) path, where the\nbeam switching is realized by digitally configuring the phases of RIS elements.\nTo date, most efforts have been made in the stationary scenario. However, when\nconsidering node mobility, beam tracking algorithms designed specifically for\nRIS are needed in order to maintain the NLOS link. In this paper, a fast\nRIS-based beam tracking algorithm is developed by partly transforming the large\namount of signaling time into the calculation happens at base station in a\nmmWave system with mobile users. Specifically, the differential form of optimal\nRIS configuration is exploited as the updating beam tracking parameter to avoid\ncomplex channel estimation procedure. The RIS-based beam tracking problem is\nthen transformed into an optimization problem whose solution is found by a\ncalculation-based search. Finally, by training on a small set candidate,\nRIS-based beam tracking is realized. The effectiveness and efficiency of the\nproposed RIS-based beam tracking algorithm is evaluated by simulations. It\nshows that the proposed algorithm has near-optimal performance with dramatic\nsavings in terms of signaling time.\n"}
{"abstract": "  Under cold conditions in dense cores, gas-phase molecules and atoms are\ndepleted from the gas-phase to the surface of interstellar grains. Considering\nthe time scales and physical conditions within these cores, a portion of these\nmolecules has to be brought back into the gas-phase to explain their\nobservation by milimeter telescopes. We tested the respective efficiencies of\nthe different mechanisms commonly included in the models. We also tested the\naddition of sputtering of ice grain mantles via a collision with cosmic rays in\nthe electronic stopping power regime. The ice sputtering induced by cosmic rays\nhas been added to the Nautilus gas-grain model while the other processes were\nalready present. Each of these processes were tested on a 1D physical structure\ndetermined by observations in TMC1 cold cores. The resulting 1D chemical\nstructure was also compared to methanol gas-phase abundances observed in these\ncores. We found that all species are not sensitive in the same way to the\nnon-thermal desorption mechanisms, and the sensitivity also depends on the\nphysical conditions. Thus, it is mandatory to include all of them. Chemical\ndesorption seems to be essential in reproducing the observations for H\ndensities smaller than $4\\times 10^4$~cm$^{-3}$, whereas sputtering is\nessential above this density. The models are, however, systematically below the\nobserved methanol abundances. A more efficient chemical desorption and a more\nefficient sputtering could better reproduce the observations. In conclusion,\nthe sputtering of ices by cosmic-rays collisions may be the most efficient\ndesorption mechanism at high density (a few $10^4$~cm$^{-3}$ under the\nconditions studied here) in cold cores, whereas chemical desorption is still\nrequired at smaller densities. Additional works are needed on both mechanisms\nto assess their efficiency with respect to the main ice composition.\n"}
{"abstract": "  Most existing deep neural networks are static, which means they can only do\ninference at a fixed complexity. But the resource budget can vary substantially\nacross different devices. Even on a single device, the affordable budget can\nchange with different scenarios, and repeatedly training networks for each\nrequired budget would be incredibly expensive. Therefore, in this work, we\npropose a general method called MutualNet to train a single network that can\nrun at a diverse set of resource constraints. Our method trains a cohort of\nmodel configurations with various network widths and input resolutions. This\nmutual learning scheme not only allows the model to run at different\nwidth-resolution configurations but also transfers the unique knowledge among\nthese configurations, helping the model to learn stronger representations\noverall. MutualNet is a general training methodology that can be applied to\nvarious network structures (e.g., 2D networks: MobileNets, ResNet, 3D networks:\nSlowFast, X3D) and various tasks (e.g., image classification, object detection,\nsegmentation, and action recognition), and is demonstrated to achieve\nconsistent improvements on a variety of datasets. Since we only train the model\nonce, it also greatly reduces the training cost compared to independently\ntraining several models. Surprisingly, MutualNet can also be used to\nsignificantly boost the performance of a single network, if dynamic resource\nconstraint is not a concern. In summary, MutualNet is a unified method for both\nstatic and adaptive, 2D and 3D networks. Codes and pre-trained models are\navailable at \\url{https://github.com/taoyang1122/MutualNet}.\n"}
{"abstract": "  We generalize the local-global compatibility result in arXiv:1506.04022 to\nhigher dimensional cases, by examining the relation between Scholze's functor\nand cohomology of Kottwitz-Harris-Taylor type Shimura varieties. Along the way\nwe prove a cuspidality criterion from type theory. We also deal with\ncompatibility for torsion classes in the case of semisimple mod $p$ Galois\nrepresentations with distinct irreducible components under certain flatness\nhypotheses.\n"}
{"abstract": "  In this work, a first order relativistic dissipative hydrodynamic theory has\nbeen developed that redefines all the concerned hydrodynamic fields in order to\ninclude the out of equilibrium system dissipation from the medium. The\nresulting dispersion relation from the theory is observed to produce a\npropagating causal mode that was previously absent without the field\nredefinition. The condition for causality has been explicitly determined for\nlinear perturbations around a hydrostatic state. Finally, the causality has\nbeen tested with the field correction coefficients estimated from momentum\ndependent relaxation time approximation (MDRTA) in relativistic transport\nequation.\n"}
{"abstract": "  A practical algorithm has been developed for closeness analysis of sequential\ndata that combines closeness testing with algorithms based on the Markov chain\ntester. It was applied to reported sequential data for COVID-19 to analyze the\nevolution of COVID-19 during a certain time period (week, month, etc.).\n"}
{"abstract": "  In this paper, we present a novel unpaired point cloud completion network,\nnamed Cycle4Completion, to infer the complete geometries from a partial 3D\nobject. Previous unpaired completion methods merely focus on the learning of\ngeometric correspondence from incomplete shapes to complete shapes, and ignore\nthe learning in the reverse direction, which makes them suffer from low\ncompletion accuracy due to the limited 3D shape understanding ability. To\naddress this problem, we propose two simultaneous cycle transformations between\nthe latent spaces of complete shapes and incomplete ones. The insight of cycle\ntransformation is to promote networks to understand 3D shapes by learning to\ngenerate complete or incomplete shapes from their complementary ones.\nSpecifically, the first cycle transforms shapes from incomplete domain to\ncomplete domain, and then projects them back to the incomplete domain. This\nprocess learns the geometric characteristic of complete shapes, and maintains\nthe shape consistency between the complete prediction and the incomplete input.\nSimilarly, the inverse cycle transformation starts from complete domain to\nincomplete domain, and goes back to complete domain to learn the characteristic\nof incomplete shapes. We provide a comprehensive evaluation in experiments,\nwhich shows that our model with the learned bidirectional geometry\ncorrespondence outperforms state-of-the-art unpaired completion methods.\n"}
{"abstract": "  The infinite number of particles limit in the dual to elliptic Ruijsenaars\nmodel (coordinate trigonometric degeneration of quantum double elliptic model)\nis proposed using the Nazarov-Sklyanin approach. For this purpose we describe\ndouble-elliptization of the Cherednik construction. Namely, we derive explicit\nexpression in terms of the Cherednik operators, which reduces to the generating\nfunction of Dell commuting Hamiltonians on the space of symmetric functions.\nAlthough the double elliptic Cherednik operators do not commute, they can be\nused for construction of the $N\\rightarrow \\infty$ limit.\n"}
{"abstract": "  Attempt to fully discover the temporal diversity and chronological\ncharacteristics for self-supervised video representation learning, this work\ntakes advantage of the temporal dependencies within videos and further proposes\na novel self-supervised method named Temporal Contrastive Graph Learning\n(TCGL). In contrast to the existing methods that ignore modeling elaborate\ntemporal dependencies, our TCGL roots in a hybrid graph contrastive learning\nstrategy to jointly regard the inter-snippet and intra-snippet temporal\ndependencies as self-supervision signals for temporal representation learning.\nTo model multi-scale temporal dependencies, our TCGL integrates the prior\nknowledge about the frame and snippet orders into graph structures, i.e., the\nintra-/inter- snippet temporal contrastive graphs. By randomly removing edges\nand masking nodes of the intra-snippet graphs or inter-snippet graphs, our TCGL\ncan generate different correlated graph views. Then, specific contrastive\nlearning modules are designed to maximize the agreement between nodes in\ndifferent views. To adaptively learn the global context representation and\nrecalibrate the channel-wise features, we introduce an adaptive video snippet\norder prediction module, which leverages the relational knowledge among video\nsnippets to predict the actual snippet orders. Experimental results demonstrate\nthe superiority of our TCGL over the state-of-the-art methods on large-scale\naction recognition and video retrieval benchmarks.\n"}
{"abstract": "  Due to the need to store the intermediate activations for back-propagation,\nend-to-end (E2E) training of deep networks usually suffers from high GPUs\nmemory footprint. This paper aims to address this problem by revisiting the\nlocally supervised learning, where a network is split into gradient-isolated\nmodules and trained with local supervision. We experimentally show that simply\ntraining local modules with E2E loss tends to collapse task-relevant\ninformation at early layers, and hence hurts the performance of the full model.\nTo avoid this issue, we propose an information propagation (InfoPro) loss,\nwhich encourages local modules to preserve as much useful information as\npossible, while progressively discard task-irrelevant information. As InfoPro\nloss is difficult to compute in its original form, we derive a feasible upper\nbound as a surrogate optimization objective, yielding a simple but effective\nalgorithm. In fact, we show that the proposed method boils down to minimizing\nthe combination of a reconstruction loss and a normal cross-entropy/contrastive\nterm. Extensive empirical results on five datasets (i.e., CIFAR, SVHN, STL-10,\nImageNet and Cityscapes) validate that InfoPro is capable of achieving\ncompetitive performance with less than 40% memory footprint compared to E2E\ntraining, while allowing using training data with higher-resolution or larger\nbatch sizes under the same GPU memory constraint. Our method also enables\ntraining local modules asynchronously for potential training acceleration. Code\nis available at: https://github.com/blackfeather-wang/InfoPro-Pytorch.\n"}
{"abstract": "  Image enhancement aims at processing an input image so that the visual\ncontent of the output image is more pleasing or more useful for certain\napplications. Although histogram equalization is widely used in image\nenhancement due to its simplicity and effectiveness, it changes the mean\nbrightness of the enhanced image and introduces a high level of noise and\ndistortion. To address these problems, this paper proposes image enhancement\nusing fuzzy intensity measure and adaptive clipping histogram equalization\n(FIMHE). FIMHE uses fuzzy intensity measure to first segment the histogram of\nthe original image, and then clip the histogram adaptively in order to prevent\nexcessive image enhancement. Experiments on the Berkeley database and\nCVF-UGR-Image database show that FIMHE outperforms state-of-the-art histogram\nequalization based methods.\n"}
{"abstract": "  We study the problem of incorporating prior knowledge into a deep\nTransformer-based model,i.e.,Bidirectional Encoder Representations from\nTransformers (BERT), to enhance its performance on semantic textual matching\ntasks. By probing and analyzing what BERT has already known when solving this\ntask, we obtain better understanding of what task-specific knowledge BERT needs\nthe most and where it is most needed. The analysis further motivates us to take\na different approach than most existing works. Instead of using prior knowledge\nto create a new training task for fine-tuning BERT, we directly inject\nknowledge into BERT's multi-head attention mechanism. This leads us to a simple\nyet effective approach that enjoys fast training stage as it saves the model\nfrom training on additional data or tasks other than the main task. Extensive\nexperiments demonstrate that the proposed knowledge-enhanced BERT is able to\nconsistently improve semantic textual matching performance over the original\nBERT model, and the performance benefit is most salient when training data is\nscarce.\n"}
{"abstract": "  We study a model of interacting vector and Kalb-Ramond gauge fields in a\nnon-trivial Higgs vacuum generated by a charged and a neutral scalar field. The\nsystem admits different vacua for different v.e.v. of the two Higgs fields. Our\nprimary interest in this paper regards the \"mixed phase\" where both the photon\nand the Kalb-Ramond acquire a mass. In this phase we compute the interaction\npotential energy between static test charges. It turns out that the limit in\nwhich the photon becomes massless, while the Kalb-Ramond remains massive, leads\nto a Cornell confining potential between test charges.\n"}
{"abstract": "  Meta learning generalizes the empirical experience with different learning\ntasks and holds promise for providing important empirical insight into the\nbehaviour of machine learning algorithms. In this paper, we present a\ncomprehensive meta-learning study of data sets and methods for multi-label\nclassification (MLC). MLC is a practically relevant machine learning task where\neach example is labelled with multiple labels simultaneously. Here, we analyze\n40 MLC data sets by using 50 meta features describing different properties of\nthe data. The main findings of this study are as follows. First, the most\nprominent meta features that describe the space of MLC data sets are the ones\nassessing different aspects of the label space. Second, the meta models show\nthat the most important meta features describe the label space, and, the meta\nfeatures describing the relationships among the labels tend to occur a bit more\noften than the meta features describing the distributions between and within\nthe individual labels. Third, the optimization of the hyperparameters can\nimprove the predictive performance, however, quite often the extent of the\nimprovements does not always justify the resource utilization.\n"}
{"abstract": "  Diversity in personalized news recommender systems is often defined as\ndissimilarity, and based on topic diversity (e.g., corona versus farmers\nstrike). Diversity in news media, however, is understood as multiperspectivity\n(e.g., different opinions on corona measures), and arguably a key\nresponsibility of the press in a democratic society. While viewpoint diversity\nis often considered synonymous with source diversity in communication science\ndomain, in this paper, we take a computational view. We operationalize the\nnotion of framing, adopted from communication science. We apply this notion to\na re-ranking of topic-relevant recommended lists, to form the basis of a novel\nviewpoint diversification method. Our offline evaluation indicates that the\nproposed method is capable of enhancing the viewpoint diversity of\nrecommendation lists according to a diversity metric from literature. In an\nonline study, on the Blendle platform, a Dutch news aggregator platform, with\nmore than 2000 users, we found that users are willing to consume viewpoint\ndiverse news recommendations. We also found that presentation characteristics\nsignificantly influence the reading behaviour of diverse recommendations. These\nresults suggest that future research on presentation aspects of recommendations\ncan be just as important as novel viewpoint diversification methods to truly\nachieve multiperspectivity in online news environments.\n"}
{"abstract": "  The 21-cm signal from cosmic dawn and the epoch of reionization (EoR) probes\nthe characteristics of the high redshift galaxy population. Many of the\nastrophysical properties of galaxies at high redshifts are currently\nunconstrained due to the lack of observations. This creates a vast space of\npossible astrophysical scenarios where the 21-cm signal needs to be modeled in\norder to plan for, and eventually fit, future observations. This is done with\nfast numerical methods which make simplifying approximations for the underlying\nphysical processes. In this work we quantify the effect of Poisson fluctuations\nand scatter in the star formation efficiency; while Poisson fluctuations are\nincluded in some works and not in others, scatter in the star formation\nefficiency is usually neglected, and all galaxies of a given mass are assumed\nto have the same properties. We show that both features can have a significant\neffect on the 21-cm power spectrum, most importantly in scenarios where the\nsignal is dominated by massive galaxies. Scatter in the star formation\nefficiency does not simply enhance the effect of Poisson fluctuations; for\nexample we show that the power spectrum shape at cosmic dawn has a feature\ncorresponding to the width of the galaxy brightness distribution. We also\ndiscuss some of the consequences for 21-cm imaging, and the signature of\nreduced correlation between the density and radiation fields.\n"}
{"abstract": "  Finding and characterizing extrasolar Earth analogs will rely on\ninterpretation of the planetary system's environmental context. The total\nbudget and fractionation between C-H-O species sensitively affect the climatic\nand geodynamic state of terrestrial worlds, but their main delivery channels\nare poorly constrained. We connect numerical models of volatile chemistry and\npebble coagulation in the circumstellar disk with the internal compositional\nevolution of planetesimals during the primary accretion phase. Our simulations\ndemonstrate that disk chemistry and degassing from planetesimals operate on\ncomparable timescales and can fractionate the relative abundances of major\nwater and carbon carriers by orders of magnitude. As a result, individual\nplanetary systems with significant planetesimal processing display increased\ncorrelation in the volatile budget of planetary building blocks relative to no\ninternal heating. Planetesimal processing in a subset of systems increases the\nvariance of volatile contents across planetary systems. Our simulations thus\nsuggest that exoplanetary atmospheric compositions may provide constraints on\n$when$ a specific planet formed.\n"}
{"abstract": "  In the anomalous Hall effect (AHE), the magnetization, electric field and the\nHall current are presumed to be mutually vertical to each other. In this work,\nwe propose an unconventional AHE where the magnetization, the electric field\nand Hall current stay inside the same plane. Such an AHE is odd under\ntime-reversal and exists even when the magnetization is parallel to the\nelectric field or Hall current, different from the planar Hall effect which is\neven under time-reversal. Here, we term it parallel anomalous Hall effect\n(PAHE). We reveal that the PAHE exists when all the point group rotational and\nreflection symmetries are broken where the Berry curvature field is not\nnecessarily parallel to the magnetization axis. We further demonstrate the PAHE\nin a ferrimagnetic Weyl semimetal FeCr$_2$Te$_4$.\n"}
{"abstract": "  As neural networks become able to generate realistic artificial images, they\nhave the potential to improve movies, music, video games and make the internet\nan even more creative and inspiring place. Yet, at the same time, the latest\ntechnology potentially enables new digital ways to lie. In response, the need\nfor a diverse and reliable method toolbox arises to identify artificial images\nand other content. Previous work primarily relies on pixel-space CNN or the\nFourier transform. To the best of our knowledge, synthesized fake image\nanalysis and detection methods based on a multi-scale wavelet representation,\nwhich is localized in both space and frequency, have been absent thus far. This\npaper proposes to learn a model for the detection of synthetic images based on\nthe wavelet-packet representation of natural and GAN-generated images. We\nevaluate our method on FFHQ, CelebA, and LSUN source identification problems\nand find improved or competitive performance. Our forensic classifier has a\nsmall network size and can be learned efficiently. Furthermore, a comparison of\nthe wavelet coefficients from these two sources of images allows an\ninterpretation and identifies significant differences.\n"}
{"abstract": "  A great deal of recent research effort on speech spoofing countermeasures has\nbeen invested into back-end neural networks and training criteria. We\ncontribute to this effort with a comparative perspective in this study. Our\ncomparison of countermeasure models on the ASVspoof 2019 logical access task\ntakes into account recently proposed margin-based training criteria, widely\nused front ends, and common strategies to deal with varied-length input trials.\nWe also measured intra-model differences through multiple training-evaluation\nrounds with random initialization. Our statistical analysis demonstrates that\nthe performance of the same model may be significantly different when just\nchanging the random initial seed. Thus, we recommend similar analysis or\nmultiple training-evaluation rounds for further research on the database.\nDespite the intra-model differences, we observed a few promising techniques\nsuch as the average pooling to process varied-length inputs and a new\nhyper-parameter-free loss function. The two techniques led to the best single\nmodel in our experiment, which achieved an equal error rate of 1.92% and was\nsignificantly different in statistical sense from most of the other\nexperimental models.\n"}
{"abstract": "  Additional grid points are often introduced for the higher-order polynomial\nof a numerical solution with curvilinear elements. However, those points are\nlikely to be located slightly outside the domain, even when the vertices of the\ncurvilinear elements lie within the curved domain. This misallocation of grid\npoints generates a mesh error, called geometric approximation error. This error\nis smaller than the discretization error but large enough to significantly\ndegrade a long-time integration. Moreover, this mesh error is considered to be\nthe leading cause of conservation error. Two novel schemes are proposed to\nimprove conservation error and/or discretization error for long-time\nintegration caused by geometric approximation error: The first scheme retrieves\nthe original divergence of the original domain; the second scheme reconstructs\nthe original path of differentiation, called connection, thus retrieving the\noriginal connection. The increased accuracies of the proposed schemes are\ndemonstrated by the conservation error for various partial differential\nequations with moving frames on the sphere.\n"}
{"abstract": "  This article considers the $\\mathcal{H}_\\infty$ static output-feedback\ncontrol for linear time-invariant uncertain systems with polynomial dependence\non probabilistic time-invariant parametric uncertainties. By applying\npolynomial chaos theory, the control synthesis problem is solved using a\nhigh-dimensional expanded system which characterizes stochastic state\nuncertainty propagation. A closed-loop polynomial chaos transformation is\nproposed to derive the closed-loop expanded system. The approach explicitly\naccounts for the closed-loop dynamics and preserves the $\\mathcal{L}_2$-induced\ngain, which results in smaller transformation errors compared to existing\npolynomial chaos transformations. The effect of using finite-degree polynomial\nchaos expansions is first captured by a norm-bounded linear differential\ninclusion, and then addressed by formulating a robust polynomial chaos based\ncontrol synthesis problem. This proposed approach avoids the use of high-degree\npolynomial chaos expansions to alleviate the destabilizing effect of truncation\nerrors, which significantly reduces computational complexity. In addition, some\nanalysis is given for the condition under which the robustly stabilized\nexpanded system implies the robust stability of the original system. A\nnumerical example illustrates the effectiveness of the proposed approach.\n"}
{"abstract": "  We present high-resolution $^{12}$CO and $^{13}$CO 2-1 ALMA observations, as\nwell as optical and near-infrared spectroscopy, of the highly-inclined\nprotoplanetary disk around SSTC2D J163131.2-242627. The spectral type we derive\nfor the source is consistent with a $\\rm 1.2 \\, M_{\\odot}$ star inferred from\nthe ALMA observations. Despite its massive circumstellar disk, we find little\nto no evidence for ongoing accretion on the star. The CO maps reveal a disk\nthat is unusually compact along the vertical direction, consistent with its\nappearance in scattered light images. The gas disk extends about twice as far\naway as both the submillimeter continuum and the optical scattered light. CO is\ndetected from two surface layers separated by a midplane region in which CO\nemission is suppressed, as expected from freeze-out in the cold midplane. We\napply a modified version of the Topographically Reconstructed Distribution\nmethod presented by Dutrey et al. 2017 to derive the temperature structure of\nthe disk. We find a temperature in the CO-emitting layers and the midplane of\n$\\sim$33 K and $\\sim$20 K at $\\rm R<200$ au, respectively. Outside of $\\rm\nR>200$ au, the disk's midplane temperature increases to $\\sim$30 K, with a\nnearly vertically isothermal profile. The transition in CO temperature\ncoincides with a dramatic reduction in the sub-micron and sub-millimeter\nemission from the disk. We interpret this as interstellar UV radiation\nproviding an additional source of heating to the outer part of the disk.\n"}
{"abstract": "  The spin Hall conductivity is usually calculated for periodic solids, while\nthe knowledge of local contributions from inhomogeneous regions like interfaces\nor surfaces would be desirable. In this work we derive a local expression for\nthe spin Hall conductivity of two-dimensional systems in analogy to the local\nanomalous Hall conductivity discussed recently. Our approach is applicable to\nheterogeneous fully bounded systems and nanoribbons with both open and periodic\nboundary conditions.\n"}
{"abstract": "  Shared practices to assess the diversity of retrieval system results are\nstill debated in the Information Retrieval community, partly because of the\nchallenges of determining what diversity means in specific scenarios, and of\nunderstanding how diversity is perceived by end-users. The field of Music\nInformation Retrieval is not exempt from this issue. Even if fields such as\nMusicology or Sociology of Music have a long tradition in questioning the\nrepresentation and the impact of diversity in cultural environments, such\nknowledge has not been yet embedded into the design and development of music\ntechnologies. In this paper, focusing on electronic music, we investigate the\ncharacteristics of listeners, artists, and tracks that are influential in the\nperception of diversity. Specifically, we center our attention on 1)\nunderstanding the relationship between perceived diversity and computational\nmethods to measure diversity, and 2) analyzing how listeners' domain knowledge\nand familiarity influence such perceived diversity. To accomplish this, we\ndesign a user-study in which listeners are asked to compare pairs of lists of\ntracks and artists, and to select the most diverse list from each pair. We\ncompare participants' ratings with results obtained through computational\nmodels built using audio tracks' features and artist attributes. We find that\nsuch models are generally aligned with participants' choices when most of them\nagree that one list is more diverse than the other, while they present a mixed\nbehaviour in cases where participants have little agreement. Moreover, we\nobserve how differences in domain knowledge, familiarity, and demographics can\ninfluence the level of agreement among listeners, and between listeners and\ndiversity metrics computed automatically.\n"}
{"abstract": "  The structure tensor of $\\mathfrak{sl}_n$, denoted $T_{\\mathfrak{sl}_n}$, is\nthe tensor arising from the Lie bracket bilinear operation on the set of\ntraceless $n \\times n$ matrices over $\\mathbb{C}$. This tensor is intimately\nrelated to the well studied matrix multiplication tensor. Studying the\nstructure tensor of $\\mathfrak{sl}_n$ may provide further insight into the\ncomplexity of matrix multiplication and the \"hay in a haystack\" problem of\nfinding explicit sequences tensors with high rank or border rank. We aim to\nfind new bounds on the rank and border rank of this structure tensor in the\ncase of $\\mathfrak{sl}_3$ and $\\mathfrak{sl}_4$. We additionally provide bounds\nin the case of the lie algebras $\\mathfrak{so}_4$ and $\\mathfrak{so}_5$. The\nlower bounds on the border ranks were obtained via various recent techniques,\nnamely Koszul flattenings, border substitution, and border apolarity. Upper\nbounds on the rank of $T_{\\mathfrak{sl}_3}$ are obtained via numerical methods\nthat allowed us to find an explicit rank decomposition.\n"}
{"abstract": "  We study the behaviour of Yukawa and Newtonian gravitational forces in a\ncubic box with fully periodic boundaries commonly encountered in N-body\nsimulations of the structure formation. Placing a single gravitating body at\nthe origin of coordinates, we reveal the scales at which non-negligible\ndeviation from the Yukawa law occurs when the Newtonian approximation is\nemployed. We discuss the results in terms of the corresponding physical\ndistances today as well as earlier, back at the matter-dominated stage.\nRevisiting the problem for free boundaries, we also compare the periodic and\nplain gravitational forces for Yukawa-type interactions.\n"}
{"abstract": "  New methodologies are designed to reduce the time complexity of an implicit\ndiscrete-time differentiator and the simulation time to implement it. They rely\non Horner's method and the Shaw-Traub algorithm. The algorithms are compared\nfor differentiators of orders 3, 7, and 10. The Half-Horner and Full-Horner\nmethods showed the best performance and time complexity.\n"}
{"abstract": "  Measuring the obliquity distribution of stars hosting warm Jupiters may help\nus to understand the formation of close-orbiting gas giants. Few such\nmeasurements have been performed due to practical difficulties in scheduling\nobservations of the relatively infrequent and long-duration transits of warm\nJupiters. Here, we report a measurement of the Rossiter-McLaughlin effect for\nK2-232b, a warm Jupiter (M_P=0.39 M_Jup) on an 11.17-day orbit with an\neccentricity of 0.26. The data were obtained with the Automated Planet Finder\nduring two separate transits. The planet's orbit appears to be well-aligned\nwith the spin axis of the host star, with a projected spin-orbit angle of\nlambda = -11.1+/-6.6 deg. Combined with the other available data, we find that\nhigh obliquities are almost exclusively associated with planets that either\nhave an orbital separation greater than 10 stellar radii or orbit stars with\neffective temperatures hotter than 6,000K. This pattern suggests that the\nobliquities of the closest-orbiting giant planets around cooler stars have been\ndamped by tidal effects.\n"}
{"abstract": "  Thinning antenna arrays through quantum Fourier transform (QFT) is proposed.\nGiven the lattice of the candidate locations for the array elements, the\nproblem of selecting which antenna location has to be either occupied or not by\nan array element is formulated in the quantum computing (QC) framework and then\naddressed with an ad-hoc design method based on a suitable implementation of\nthe QFT algorithm. Representative numerical results are presented and discussed\nto point out the features and the advantages of the proposed QC-based thinning\ntechnique.\n"}
{"abstract": "  The general position number of a graph $G$ is the size of the largest set of\nvertices $S$ such that no geodesic of $G$ contains more than two elements of\n$S$. The monophonic position number of a graph is defined similarly, but with\n`induced path' in place of `geodesic'. In this paper we investigate some\nextremal problems for these parameters. Firstly we discuss the problem of the\nsmallest possible order of a graph with given general and monophonic position\nnumbers, with applications to a realisation result. We then solve a Tur\\'{a}n\nproblem for the size of graphs with given order and position numbers and\ncharacterise the possible diameters of graphs with given order and monophonic\nposition number. Finally we classify the graphs with given order and diameter\nand largest possible general position number.\n"}
{"abstract": "  We consider initial value problems of nonlinear dynamical systems, which\ninclude physical parameters. A quantity of interest depending on the solution\nis observed. A discretisation yields the trajectories of the quantity of\ninterest in many time points. We examine the mapping from the set of parameters\nto the discrete values of the trajectories. An evaluation of this mapping\nrequires to solve an initial value problem. Alternatively, we determine an\napproximation, where the evaluation requires low computation work, using a\nconcept of machine learning. We employ feedforward neural networks, which are\nfitted to data from samples of the trajectories. Results of numerical\ncomputations are presented for a test example modelling an electric circuit.\n"}
{"abstract": "  Entanglement is a unique property of quantum systems and an essential\nresource for many quantum technologies. The ability to transfer or swap\nentanglement between systems is an important protocol in quantum information\nscience. Entanglement between photons forms the basis of distributed quantum\nnetworks and the demonstration of photonic entanglement swapping is essential\nfor their realization. Here an experiment demonstrating entanglement swapping\nfrom two independent multimode time-frequency entangled sources is presented,\nresulting in multiple heralded temporal-mode Bell states. Entanglement in the\nheralded states is verified by measuring conditional anti-correlated joint\nspectra as well as quantum beating in two-photon interference. Our\nproof-of-concept experiment is able to distinguish up to five orthogonal Bell\npairs within the same setup, limited in principle only by the entanglement of\nthe sources.\n"}
{"abstract": "  Inhomogeneous big bang nucleosynthesis (BBN) produces a spatially\ninhomogeneous distribution of element abundances at $T \\sim 10^9$ K, but\nsubsequent element diffusion will tend to erase these inhomogeneities. We\ncalculate the cosmological comoving diffusion length for the BBN elements. This\ndiffusion length is limited by atomic scattering and is therefore dominated by\ndiffusion when the atoms are neutral, between the redshifts of recombination\nand reionization. We find that the comoving diffusion length today is $d_{com}\n\\approx 70$ pc for all of the elements of interest except $^7$Li, for which\n$d_{com}$ is an order of magnitude smaller because $^7$Li remains ionized\nthroughout the relevant epoch. This comoving diffusion length corresponds to a\nsubstellar baryonic mass scale and is roughly equal to the horizon scale at\nBBN. These results lend support to the possibility that inhomogeneities on\nscales larger than the horizon at BBN could lead to a spatially inhomogeneous\ndistribution of elements today, while purely subhorizon fluctuations at BBN can\nresult only in a homogeneous element distribution at present.\n"}
{"abstract": "  X-ray flux from the inner hot region around central compact object in a\nbinary system illuminates the upper surface of an accretion disc and it behaves\nlike a corona. This region can be photoionised by the illuminating radiation,\nthus can emit different emission lines. We study those line spectra in black\nhole X-ray binaries for different accretion flow parameters including its\ngeometry. The varying range of model parameters captures maximum possible\nobservational features. We also put light on the routinely observed Fe line\nemission properties based on different model parameters, ionization rate, and\nFe abundances. We find that the Fe line equivalent width $W_{\\rm E}$ decreases\nwith increasing disc accretion rate and increases with the column density of\nthe illuminated gas. Our estimated line properties are in agreement with\nobservational signatures.\n"}
{"abstract": "  We extend and generalize the construction of Sturm-Liouville problems for a\nfamily of Hamiltonians constrained to fulfill a third-order shape-invariance\ncondition and focusing on the \"$-2x/3$\" hierarchy of solutions to the fourth\nPainlev\\'e transcendent. Such a construction has been previously addressed in\nthe literature for some particular cases but we realize it here in the most\ngeneral case. The corresponding potential in the Hamiltonian operator is a\nrationally extended oscillator defined in terms of the conventional Okamoto\npolynomials, from which we identify three different zero-modes constructed in\nterms of the generalized Okamoto polynomials. The third-order ladder operators\nof the system reveal that the complete set of eigenfunctions is decomposed as a\nunion of three disjoint sequences of solutions, generated from a set of\nthree-term recurrence relations. We also identify a link between the\neigenfunctions of the Hamiltonian operator and a special family of exceptional\nHermite polynomial.\n"}
{"abstract": "  Artificial intelligence-powered pocket-sized air robots have the potential to\nrevolutionize the Internet-of-Things ecosystem, acting as autonomous,\nunobtrusive, and ubiquitous smart sensors. With a few cm$^{2}$ form-factor,\nnano-sized unmanned aerial vehicles (UAVs) are the natural befit for indoor\nhuman-drone interaction missions, as the pose estimation task we address in\nthis work. However, this scenario is challenged by the nano-UAVs' limited\npayload and computational power that severely relegates the onboard brain to\nthe sub-100 mW microcontroller unit-class. Our work stands at the intersection\nof the novel parallel ultra-low-power (PULP) architectural paradigm and our\ngeneral development methodology for deep neural network (DNN) visual pipelines,\ni.e., covering from perception to control. Addressing the DNN model design,\nfrom training and dataset augmentation to 8-bit quantization and deployment, we\ndemonstrate how a PULP-based processor, aboard a nano-UAV, is sufficient for\nthe real-time execution (up to 135 frame/s) of our novel DNN, called\nPULP-Frontnet. We showcase how, scaling our model's memory and computational\nrequirement, we can significantly improve the onboard inference (top energy\nefficiency of 0.43 mJ/frame) with no compromise in the quality-of-result vs. a\nresource-unconstrained baseline (i.e., full-precision DNN). Field experiments\ndemonstrate a closed-loop top-notch autonomous navigation capability, with a\nheavily resource-constrained 27-gram Crazyflie 2.1 nano-quadrotor. Compared\nagainst the control performance achieved using an ideal sensing setup, onboard\nrelative pose inference yields excellent drone behavior in terms of median\nabsolute errors, such as positional (onboard: 41 cm, ideal: 26 cm) and angular\n(onboard: 3.7$^{\\circ}$, ideal: 4.1$^{\\circ}$).\n"}
{"abstract": "  We report on the preparation of flexible polymer composite films with aligned\nmetallic fillers comprised of atomic chain bundles of the quasi-one-dimensional\n(1D) van der Waals material tantalum triselenide, TaSe3. The material\nfunctionality, embedded at the nanoscale level, is achieved by mimicking the\ndesign of an electromagnetic aperture grid antenna. The processed composites\nemploy chemically exfoliated TaSe3 nanowires as the grid building blocks\nincorporated within the thin film. Filler alignment is achieved using the\n\"blade coating\" method. Measurements conducted in the X-band frequency range\ndemonstrate that the electromagnetic transmission through such films can be\nvaried significantly by changing the relative orientations of the quasi-1D\nfillers and the polarization of the electromagnetic wave. We argue that such\npolarization-sensitive polymer films with quasi-1D fillers are applicable to\nadvanced electromagnetic interference shielding in future communication\nsystems.\n"}
{"abstract": "  The cell wall serves as a mechanical protection for the fungi and controls\nthe traffic of molecules entering the cell. However, the optical role of the\ncell wall has not been fully investigated. In this work, we use a computational\nsimulation to evaluate the optical response of the fungal cell wall. The main\nadvantage of this approach is that the scattering structure can be introduced\nwithin the simulation via transmission electronic microscope (TEM) images of\nits cross section which in this case correspond to real fungal cell walls. The\nobtained results indicate that the reflectivity of the fungal cell wall depends\non the incident wavelength, and then, it modulates the light that reaches the\ninterior of the cell. This suggests that the light stimulus that reaches the\ninner part of the cell could depend on the structural characteristics of the\ncell wall. Additionally, regarding the analysis of color, our finding opens up\na new and interesting approach to investigate the ecological role of UV light\nin relation to fungal organisms and their microhabitat.\n"}
{"abstract": "  Motivated by deviations with respect to Standard Model predictions in $b\\to\ns\\ell^+\\ell^-$ decays, we evaluate the global significance of the new physics\nhypothesis in this system by including the {\\it look-elsewhere effect} for the\nfirst time. We estimate the trial-factor with pseudo-experiments and find that\nit can be as large as seven. We calculate the global significance for the new\nphysics hypothesis by considering the most general description of a\nnon-standard $b\\to s\\ell^+\\ell^-$ amplitude of short-distance origin.\nTheoretical uncertainties are treated in a highly conservative way by absorbing\nthe corresponding effects into a redefinition of the Standard Model amplitude.\nUsing the most recent measurements of LHCb, ATLAS and CMS, we obtain the global\nsignificance to be $4.3$ standard deviations.\n"}
{"abstract": "  Continuous determinantal point processes (DPPs) are a class of repulsive\npoint processes on $\\mathbb{R}^d$ with many statistical applications. Although\nan explicit expression of their density is known, this expression is too\ncomplicated to be used directly for maximum likelihood estimation. In the\nstationary case, an approximation using Fourier series has been suggested, but\nit is limited to rectangular observation windows and no theoretical results\nsupport it. In this contribution, we investigate a different way to approximate\nthe likelihood by looking at its asymptotic behaviour when the observation\nwindow grows towards $\\mathbb{R}^d$. This new approximation is not limited to\nrectangular windows, is faster to compute than the previous one, does not\nrequire any tuning parameter, and some theoretical justifications are provided.\nThe performances of the associated estimator are assessed in a simulation study\non standard parametric models on $\\mathbb{R}^d$ and compare favourably to\ncommon alternative estimation methods for continuous DPPs.\n"}
{"abstract": "  Given a simple recursive function, we show how to extract from it a\nreversible and an classical iterative part. Those parts can synchronously\ncooperate under a Producer/Consumer pattern in order to implement the original\nrecursive function. The reversible producer is meant to run on reversible\nhardware. We also discuss how to extend the extraction to a more general\ncompilation scheme.\n"}
{"abstract": "  Since 2014 transfer learning has become the key driver for the improvement of\nspatial saliency prediction; however, with stagnant progress in the last 3-5\nyears. We conduct a large-scale transfer learning study which tests different\nImageNet backbones, always using the same read out architecture and learning\nprotocol adopted from DeepGaze II. By replacing the VGG19 backbone of DeepGaze\nII with ResNet50 features we improve the performance on saliency prediction\nfrom 78% to 85%. However, as we continue to test better ImageNet models as\nbackbones (such as EfficientNetB5) we observe no additional improvement on\nsaliency prediction. By analyzing the backbones further, we find that\ngeneralization to other datasets differs substantially, with models being\nconsistently overconfident in their fixation predictions. We show that by\ncombining multiple backbones in a principled manner a good confidence\ncalibration on unseen datasets can be achieved. This new model, \"DeepGaze IIE\",\nyields a significant leap in benchmark performance in and out-of-domain with a\n15 percent point improvement over DeepGaze II to 93% on MIT1003, marking a new\nstate of the art on the MIT/Tuebingen Saliency Benchmark in all available\nmetrics (AUC: 88.3%, sAUC: 79.4%, CC: 82.4%).\n"}
{"abstract": "  The largest anisotropy in the cosmic microwave background (CMB) is the 3 mK\nkinematic dipole reflecting our motion with respect to the CMB frame and\npointed in the direction $(l, b) = (264^\\circ, +48^\\circ)$ in Galactic\ncoordinates. We introduce the concept of the ring of attraction (RA), which is\northogonal to the axis of the kinematic dipole. These directions overlap with\nthe zone of percolation for the kinematic dipole, where its amplitude almost\nvanishes. We show that along this ring are oriented the directions of the\ndipole modulation of the CMB, and positions of the peaks responsible for\ngeneration of parity asymmetry. This coincidence is peculiar at around the 3\nsigma level. We analyzed the \"interaction\" of low multipoles of the CMB with RA\nand showed that for odd modes there is a sequence of peaks in the RA direction.\nThese peaks correlate with each other for different multipoles and result in\nmutual amplification of the odd $\\ell$ signal for the first 30 multipoles. Our\nmethod sheds new light on the nature of parity asymmetry. It consists of the\ndeficit of symmetrically located and equal in amplitude peaks in the CMB map in\ncomparison with asymmetric peaks.\n"}
{"abstract": "  In this work we investigate whether a multilayered planet can be approximated\nas a homogeneous planet, and in particular how well the dissipation rate of a\nmultilayered planet can be reproduced with a homogeneous rheology. We study the\ncase of a stratified body with an icy crust that, according to recent studies,\ndisplays a double peak feature in the tidal response that cannot be reproduced\nwith a homogeneous planet with an Andrade rheology. We revisit the problem with\na slightly more complex rheology for the homogeneous body, the Sundberg-Cooper\nrheology, which naturally has a double peak feature, and apply the model to the\nTRAPPIST-1e planet. Our results compare very well with the results obtained\nwhen employing a multilayered model, showing that it is possible to approximate\nthe behavior of a multilayer icy planet with a homogeneous planet using the\nSundberg-Cooper rheology. This highlights the fact that we do not need the\ncomplexity of the multilayer planet model in order to estimate the tidal\ndissipation of an icy planet.\n"}
{"abstract": "  The recently commissioned Dark Energy Spectroscopic Instrument (DESI) will\nmeasure the expansion historyof the universe using the Baryon Acoustic\nOscillation technique. The spectra of 35 million galaxies and quasarsover 14000\nsq deg will be measured during the life of the experiment. A new prime focus\ncorrector for theKPNO Mayall telescope delivers light to 5000 fiber optic\npositioners. The fibers in turn feed ten broad-bandspectrographs. We describe\nkey aspects and lessons learned from the development, delivery and installation\nofthe fiber system at the Mayall telescope.\n"}
{"abstract": "  Presence of noise in the labels of large scale facial expression datasets has\nbeen a key challenge towards Facial Expression Recognition (FER) in the wild.\nDuring early learning stage, deep networks fit on clean data. Then, eventually,\nthey start overfitting on noisy labels due to their memorization ability, which\nlimits FER performance. This work proposes an effective training strategy in\nthe presence of noisy labels, called as Consensual Collaborative Training (CCT)\nframework. CCT co-trains three networks jointly using a convex combination of\nsupervision loss and consistency loss, without making any assumption about the\nnoise distribution. A dynamic transition mechanism is used to move from\nsupervision loss in early learning to consistency loss for consensus of\npredictions among networks in the later stage. Inference is done using a single\nnetwork based on a simple knowledge distillation scheme. Effectiveness of the\nproposed framework is demonstrated on synthetic as well as real noisy FER\ndatasets. In addition, a large test subset of around 5K images is annotated\nfrom the FEC dataset using crowd wisdom of 16 different annotators and reliable\nlabels are inferred. CCT is also validated on it. State-of-the-art performance\nis reported on the benchmark FER datasets RAFDB (90.84%) FERPlus (89.99%) and\nAffectNet (66%). Our codes are available at https://github.com/1980x/CCT.\n"}
{"abstract": "  In this paper we present some very recent results regarding existence,\nuniqueness, and multiplicity of solutions for quasilinear elliptic equations\nand systems, exhibiting both singular and convective reaction terms. The\nimportance of boundary conditions (Dirichlet, Neumann, or Robin) is also\ndiscussed. Existence is achieved via sub-supersolution and truncation\ntechniques, fixed point theory, nonlinear regularity, and set-valued analysis,\nwhile uniqueness and multiplicity are obtained by monotonicity arguments.\n"}
{"abstract": "  Infinite Gray code has been introduced by Tsuiki~\\cite{ts} as a\nredundancy-free representation of the reals. In applications the signed digit\nrepresentation is mostly used which has maximal redundancy. Tsuiki presented a\nfunctional program converting signed digit code into infinite Gray code.\nMoreover, he showed that infinite Gray code can effectively be converted into\nsigned digit code, but the program needs to have some non-deterministic\nfeatures (see also \\cite{tsug}). Berger and Tsuiki~\\cite{btifp,bt} reproved the\nresult in a system of formal first-order intuitionistic logic extended by\ninductive and co-inductive definitions, as well as some new logical connectives\ncapturing concurrent behaviour. The programs extracted from the proofs are\nexactly the ones given by Tsuiki. In order to do so, co-inductive predicates\n$\\bS$ and $\\bG$ are defined and the inclusion $\\bS \\subseteq \\bG$ is derived.\nFor the converse inclusion the new logical connectives are used to introduce a\nconcurrent version $\\S_{2}$ of $S$ and $\\bG \\subseteq \\bS_{2}$ is shown. What\none is looking for, however, is an equivalence proof of the involved concepts.\nOne of the main aims of the present paper is to close the gap. A concurrent\nversion $\\bG^{*}$ of $\\bG$ and a modification $\\bS^{*}$ of $\\bS_{2}$ are\npresented such that $\\bS^{*} = \\bG^{*}$. A crucial tool in \\cite{btifp} is a\nformulation of the Archimedean property of the real numbers as an induction\nprinciple. We introduce a concurrent version of this principle which allows us\nto prove that $\\bS^{*}$ and $\\bG^{*}$ coincide. A further central contribution\nis the extension of the above results to the hyperspace of non-empty compact\nsubsets of the reals.\n"}
{"abstract": "  Ultra-hot Jupiters are the hottest exoplanets discovered so far. Observations\nbegin to provide insight into the composition of their extended atmospheres and\ntheir chemical day/night asymmetries. Both are strongly affected by cloud\nformation. We explore trends in cloud properties for a sample of five giant gas\nplanets: WASP-43b, WASP-18b, HAT-P-7b, WASP-103b, and WASP-121b. This provides\na reference frame for cloud properties for the JWST targets WASP-43b and\nWASP-121b. We further explore chemically inert tracers to observe geometrical\nasymmetries, and if the location of inner boundary of a 3D GCM matters for the\nclouds that form. The large day/night temperature differences of ultra-hot\nJupiters cause large chemical asymmetries: cloud-free days but cloudy nights,\natomic vs. molecular gases and respectively different mean molecular weights,\ndeep thermal ionospheres vs. low-ionised atmospheres, undepleted vs enhanced\nC/O. WASP-18b, as the heaviest planet in the sample, has the lowest global C/O.\nThe global climate may be considered as similar amongst ultra-hot Jupiters, but\ndifferent to that of hot gas giants. The local weather, however, is individual\nfor each planet since the local thermodynamic conditions, and hence the local\ncloud and gas properties, differ. The morning and the evening terminator of\nultra-hot Jupiters will carry signatures of their strong chemical asymmetry\nsuch that ingress/egress asymmetries can be expected. An increased C/O ratio is\na clear sign of cloud formation, making cloud modelling a necessity when\nutilizing C/O (or other mineral ratios) as tracer for planet formation. The\nchanging geometrical extension of the atmosphere from the day to the nightside\nmay be probed through chemically inert species like helium. Ultra-hot Jupiters\nare likely to develop deep atmospheric ionospheres which may impact the\natmosphere dynamics through MHD processes.\n"}
{"abstract": "  We consider the problem of determining the achievable region of parameters\nfor universal $1 \\to 2$ asymmetric quantum cloning. Measuring the cloning\nperformance with the figure of merit of singlet fraction, we show that the\nphysical region is a union of ellipses in the plane. Equivalently, we\ncharacterize the parameter region of quantum state compatibility of two\npossibly different isotropic states, considering, for the first time, negative\nsinglet fractions.\n"}
{"abstract": "  Unmanned Aerial Vehicles (UAVs) are an emerging computation platform known\nfor their safety-critical need. In this paper, we conduct an empirical study on\na widely used open-source UAV software framework, Paparazzi, with the goal of\nunderstanding the safety-critical concerns of UAV software from a bottom-up\ndeveloper-in-the-field perspective. We set our focus on the use of Bounding\nFunctions (BFs), the runtime checks injected by Paparazzi developers on the\nrange of variables. Through an in-depth analysis on BFs in the Paparazzi\nautopilot software, we found a large number of them (109 instances) are used to\nbound safety-critical variables essential to the cyber-physical nature of the\nUAV, such as its thrust, its speed, and its sensor values. The novel\ncontributions of this study are two fold. First, we take a static approach to\nclassify all BF instances, presenting a novel datatype-based 5-category\ntaxonomy with fine-grained insight on the role of BFs in ensuring the safety of\nUAV systems. Second, we dynamically evaluate the impact of the BF uses through\na differential approach, establishing the UAV behavioral difference with and\nwithout BFs. The two-pronged static and dynamic approach together illuminates a\nrarely studied design space of safety-critical UAV software systems.\n"}
{"abstract": "  In Schubert Puzzles and Integrability I we proved several \"puzzle rules\" for\ncomputing products of Schubert classes in K-theory (and sometimes equivariant\nK-theory) of d-step flag varieties. The principal tool was \"quantum\nintegrability\", in several variants of the Yang--Baxter equation; this let us\nrecognize the Schubert structure constants as q->0 limits of certain matrix\nentries in products of R- (and other) matrices of quantized affine algebra\nrepresentations. In the present work we give direct cohomological\ninterpretations of those same matrix entries but at finite q: they compute\nproducts of \"motivic Segre classes\", closely related to K-theoretic\nMaulik--Okounkov stable classes living on the cotangent bundles of the flag\nvarieties. Without q->0, we avoid some divergences that blocked fuller\nunderstanding of d=3,4. The puzzle computations are then explained (in\ncohomology onlyin this work, not K-theory) in terms of Lagrangian convolutions\nbetween Nakajima quiver varieties. More specifically, the conormal bundle to\nthe diagonal inclusion of a flag variety factors through a quiver variety that\nis not a cotangent bundle, and it is on that intermediate quiver variety that\nthe R-matrix calculation occurs.\n"}
{"abstract": "  Using finite difference operators, we define a notion of boundary and surface\nmeasure for configuration sets under Poisson measures. A Margulis-Russo type\nidentity and a co-area formula are stated with applications to deviation\ninequalities and functional inequalities, and bounds are obtained on the\nassociated isoperimetric constants.\n"}
{"abstract": "  Programming by Example (PBE) is a program synthesis paradigm in which the\nsynthesizer creates a program that matches a set of given examples. In many\napplications of such synthesis (e.g., program repair or reverse engineering),\nwe are to reconstruct a program that is close to a specific target program, not\nmerely to produce some program that satisfies the seen examples. In such\nsettings, we wish that the synthesized program generalizes well, i.e., has as\nfew errors as possible on the unobserved examples capturing the target function\nbehavior. In this paper, we propose the first framework (called SynGuar) for\nPBE synthesizers that guarantees to achieve low generalization error with high\nprobability. Our main contribution is a procedure to dynamically calculate how\nmany additional examples suffice to theoretically guarantee generalization. We\nshow how our techniques can be used in 2 well-known synthesis approaches: PROSE\nand STUN (synthesis through unification), for common string-manipulation\nprogram benchmarks. We find that often a few hundred examples suffice to\nprovably bound generalization error below $5\\%$ with high ($\\geq 98\\%$)\nprobability on these benchmarks. Further, we confirm this empirically: SynGuar\nsignificantly improves the accuracy of existing synthesizers in generating the\nright target programs. But with fewer examples chosen arbitrarily, the same\nbaseline synthesizers (without SynGuar) overfit and lose accuracy.\n"}
{"abstract": "  Mode competition is conventionally regarded as an unwanted phenomenon that\ninduces instability of laser emission. Here, we experimentally demonstrate that\nthe random dynamics behavior of mode competition provides hidden information of\nthe intracavity topology. We introduced the concept of self-assembled\nbiophotonics networks, where human insulin protein with network topology was\nsandwiched in a Fabry-Perot (FP) cavity. The optical coupling between insulin\nfibrils and enables strong mode competition. Thanks to the optical feedback\nprovided by the FP cavity, the interaction between the light and insulin\nnetwork was enhanced. Subtle changes in the network will cause significant\nchanges in laser mode dynamics. By recording the transverse mode dynamics, we\nreconstruct a graph that reveals the optical correlation in the insulin\nnetwork. Our results indicate that the graph represents an intrinsic feature\nand can sensitivity respond to the stimulus. The unclonable and sensitively\nresponsive property of the biophotonic lasing network can potentially be\nemployed in applications in sensing and information encryption.\n"}
{"abstract": "  Intensive research has been conducted on the verification and validation of\ndeep neural networks (DNNs), aiming to understand if, and how, DNNs can be\napplied to safety critical applications. However, existing verification and\nvalidation techniques are limited by their scalability, over both the size of\nthe DNN and the size of the dataset. In this paper, we propose a novel\nabstraction method which abstracts a DNN and a dataset into a Bayesian network\n(BN). We make use of dimensionality reduction techniques to identify hidden\nfeatures that have been learned by hidden layers of the DNN, and associate each\nhidden feature with a node of the BN. On this BN, we can conduct probabilistic\ninference to understand the behaviours of the DNN processing data. More\nimportantly, we can derive a runtime monitoring approach to detect in\noperational time rare inputs and covariate shift of the input data. We can also\nadapt existing structural coverage-guided testing techniques (i.e., based on\nlow-level elements of the DNN such as neurons), in order to generate test cases\nthat better exercise hidden features. We implement and evaluate the BN\nabstraction technique using our DeepConcolic tool available at\nhttps://github.com/TrustAI/DeepConcolic.\n"}
{"abstract": "  Aiming at the local overload of multi-controller deployment in\nsoftware-defined networks, a load balancing mechanism of SDN controller based\non reinforcement learning is designed. The initial paired migrate-out domain\nand migrate-in domain are obtained by calculating the load ratio deviation\nbetween the controllers, a preliminary migration triplet, contains migration\ndomain mentioned above and a group of switches which are subordinated to the\nmigrate-out domain, makes the migration efficiency reach the local optimum.\nUnder the constraint of the best efficiency of migration in the whole and\nwithout migration conflict, selecting multiple sets of triples based on\nreinforcement learning, as the final migration of this round to attain the\nglobal optimal controller load balancing with minimum cost. The experimental\nresults illustrate that the mechanism can make full use of the controllers'\nresources, quickly balance the load between controllers, reduce unnecessary\nmigration overhead and get a faster response rate of the packet-in request.\n"}
{"abstract": "  In this paper, we use a complete model of classical primordial standard\nclocks as an example to develop a methodology of directly comparing numerical\npredictions from complicated multi-field feature models with the Planck data,\nincluding the Planck 2018 Plik unbinned likelihood and the statistically most\npowerful CamSpec 2020 likelihood for temperature and polarization data. As this\ntwo-field inflationary model offers a plethora of primordial feature spectra\nthat represent combinations of sharp and resonance feature signals\nnon-trivially distributed over extended cosmological scales, its data\ncomparison has not been satisfactorily addressed by previous attempts using\nanalytical templates. The method of this paper, consisting of numerical\nprediction, effective parameter construction and nested sampling data\ncomparison, allows us to efficiently explore every possible spectra from the\nmodel. We classify the resulting feature candidates in three different\nfrequency ranges. We use the Bayesian evidences to assess the statistical\nsignificance of the candidates over the baseline model, taking into account the\neffect of additional parameters and the look-elsewhere effect. Although none of\nthe candidates is statistically significant, the methodology of this paper can\nbe used to facilitate the future model-building and data-screening process of\nprimordial features, and the candidates can be subjected to further tests with\ndata from the upcoming cosmic microwave background polarization observations\nand galaxy surveys.\n"}
{"abstract": "  Compared to the general news domain, information extraction (IE) from\nbiomedical text requires much broader domain knowledge. However, many previous\nIE methods do not utilize any external knowledge during inference. Due to the\nexponential growth of biomedical publications, models that do not go beyond\ntheir fixed set of parameters will likely fall behind. Inspired by how humans\nlook up relevant information to comprehend a scientific text, we present a\nnovel framework that utilizes external knowledge for joint entity and relation\nextraction named KECI (Knowledge-Enhanced Collective Inference). Given an input\ntext, KECI first constructs an initial span graph representing its initial\nunderstanding of the text. It then uses an entity linker to form a knowledge\ngraph containing relevant background knowledge for the the entity mentions in\nthe text. To make the final predictions, KECI fuses the initial span graph and\nthe knowledge graph into a more refined graph using an attention mechanism.\nKECI takes a collective approach to link mention spans to entities by\nintegrating global relational information into local representations using\ngraph convolutional networks. Our experimental results show that the framework\nis highly effective, achieving new state-of-the-art results in two different\nbenchmark datasets: BioRelEx (binding interaction detection) and ADE (adverse\ndrug event extraction). For example, KECI achieves absolute improvements of\n4.59% and 4.91% in F1 scores over the state-of-the-art on the BioRelEx entity\nand relation extraction tasks.\n"}
{"abstract": "  For a Banach space $X$ we demonstrate the equivalence of the following two\nproperties:\n  (1) $X$ is B-convex (that is, possesses a nontrivial infratype), and\n  (2) if ${F: [0,1] \\to 2^{X} \\setminus \\{\\varnothing\\}}$ is a {multifunction},\n$\\mathrm{conv} F$ denotes the mapping $t \\mapsto \\mathrm{conv} F(t)$, then the\nRiemann integrability of $\\mathrm{conv} F$ is equivalent to the Riemann\nintegrability of $F$.\n  For multifunctions with compact values the Riemann integrability of\n$\\mathrm{conv} F$ is equivalent to the Riemann integrability of $F$ without any\nrestrictions on the Banach space $X$.\n"}
{"abstract": "  This letter presents a novel framework termed DistSTN for the task of\nsynthetic aperture radar (SAR) automatic target recognition (ATR). In contrast\nto the conventional SAR ATR algorithms, DistSTN considers a more challenging\npractical scenario for non-cooperative targets whose aspect angles for training\nare incomplete and limited in a partial range while those of testing samples\nare unlimited. To address this issue, instead of learning the pose invariant\nfeatures, DistSTN newly involves an elaborated feature disentangling model to\nseparate the learned pose factors of a SAR target from the identity ones so\nthat they can independently control the representation process of the target\nimage. To disentangle the explainable pose factors, we develop a pose\ndiscrepancy spatial transformer module in DistSTN to characterize the intrinsic\ntransformation between the factors of two different targets with an explicit\ngeometric model. Furthermore, DistSTN develops an amortized inference scheme\nthat enables efficient feature extraction and recognition using an\nencoder-decoder mechanism. Experimental results with the moving and stationary\ntarget acquisition and recognition (MSTAR) benchmark demonstrate the\neffectiveness of our proposed approach. Compared with the other ATR algorithms,\nDistSTN can achieve higher recognition accuracy.\n"}
{"abstract": "  We give necessary and/or sufficient conditions for stochastic stability of\nsecond-order linear autonomous systems with parameters, which are perturbed by\na random process of the \"white noise\" type. The Ito's and Stratonovich's forms\nof stochastic differential equations representing the system with white noise\nare considered. The investigation of stochastic stability of the systems\nconsidered is based on the construction of special Lyapunov functions in the\nquadratic forms. The bifurcation value of the white noise intensity at which a\nsystem first becomes unstable is presented in the analytical expression. As an\nexample a damped harmonic oscillator with randomly perturbed parameters is\nconsidered.\n"}
{"abstract": "  Human ratings are one of the most prevalent methods to evaluate the\nperformance of natural language processing algorithms. Similarly, it is common\nto measure the quality of sentences generated by a natural language generation\nmodel using human raters. In this paper, we argue for exploring the use of\nsubjective evaluations within the process of training language generation\nmodels in a multi-task learning setting. As a case study, we use a\ncrowd-authored dialogue corpus to fine-tune six different language generation\nmodels. Two of these models incorporate multi-task learning and use subjective\nratings of lines as part of an explicit learning goal. A human evaluation of\nthe generated dialogue lines reveals that utterances generated by the\nmulti-tasking models were subjectively rated as the most typical, most moving\nthe conversation forward, and least offensive. Based on these promising first\nresults, we discuss future research directions for incorporating subjective\nhuman evaluations into language model training and to hence keep the human user\nin the loop during the development process.\n"}
{"abstract": "  We continue the study of the octagon form factor which helps to evaluate a\nclass of four-point correlation functions in $\\mathcal{N}=4$ SYM theory. The\noctagon is characterised, besides the kinematical parameters, by a \"bridge\" of\n$\\ell$ propagators connecting two non-adjacent operators. In this paper we\nconstruct an operator representation of the octagon with finite bridge as an\nexpectation value in the Fock space of free complex fermions. The bridge $\\ell$\nappears as the level of filling of the Dirac sea. We obtain determinant\nidentities relating octagons with different bridges, which we derive from the\nexpression of the octagon in terms of discrete fermionic oscillators. The\nderivation is based on the existence of a previously conjectured similarity\ntransformation, which we find here explicitly.\n"}
{"abstract": "  Diffuse gliomas are highly infiltrative tumors whose early diagnosis and\nfollow-up usually rely on magnetic resonance imaging (MRI). However, the\nlimited sensitivity of this technique makes it impossible to directly assess\nthe extent of the glioma cell invasion, leading to sub-optimal treatment\nplaning. Reaction-diffusion growth models have been proposed for decades to\nextrapolate glioma cell infiltration beyond margins visible on MRI and predict\nits spatial-temporal evolution. These models nevertheless require an initial\ncondition, that is the tumor cell density values at every location of the brain\nat diagnosis time. Several works have proposed to relate the tumor cell density\nfunction to abnormality outlines visible on MRI but the underlying assumptions\nhave never been verified so far. In this work we propose to verify these\nassumptions by stereotactic histological analysis of a non-operated brain with\nglioblastoma using a tailored 3D-printed slicer. Cell density maps are computed\nfrom histological slides using a deep learning approach. The density maps are\nthen registered to a postmortem MR image and related to an MR-derived geodesic\ndistance map to the tumor core. The relation between the edema outlines visible\non T2 FLAIR MRI and the distance to the core is also investigated. Our results\nsuggest that (i) the previously suggested exponential decrease of the tumor\ncell density with the distance to the tumor core is not unreasonable but (ii)\nthe edema outlines may in general not correspond to a cell density iso-contour\nand (iii) the commonly adopted tumor cell density value at these outlines is\nlikely overestimated. These findings highlight the limitations of using\nconventional MRI to derive glioma cell density maps and point out the need of\nvalidating other methods to initialize reaction-diffusion growth models and\nmake them usable in clinical practice.\n"}
{"abstract": "  Clustering is a fundamental problem in machine learning where distance-based\napproaches have dominated the field for many decades. This set of problems is\noften tackled by partitioning the data into K clusters where the number of\nclusters is chosen apriori. While significant progress has been made on these\nlines over the years, it is well established that as the number of clusters or\ndimensions increase, current approaches dwell in local minima resulting in\nsuboptimal solutions. In this work, we propose a new set of distance threshold\nmethods called Theta-based Algorithms (ThetA). Via experimental comparisons and\ncomplexity analyses we show that our proposed approach outperforms existing\napproaches in: a) clustering accuracy and b) time complexity. Additionally, we\nshow that for a large class of problems, learning the optimal threshold is\nstraightforward in comparison to learning K. Moreover, we show how ThetA can\ninfer the sparsity of datasets in higher dimensions.\n"}
{"abstract": "  This paper deals with random sensors initially randomly deployed on the line\naccording to general random process and on the plane according to two\nindependent general random processes. The mobile robot with carrying capacity\n$k$ placed at the origin point is to move the sensors to achieve the general\nscheduling requirement such as coverage, connectivity and thus to satisfy the\ndesired communication property in the network. We study tradeoffs between the\nenergy consumption in robot's movement, the numbers of sensors $n$, the sensor\nrange $r$, the interference distance $s$, and the robot capacity $k$ until\ncompletion of the coverage simultaneously with interference scheduling task. In\nthis work, we obtain upper bounds for the energy consumption in robot's\nmovement and obtain the sharp decrease in the total movement cost of the robot\nso as to provide the coverage simultaneously with interference requirement.\n"}
{"abstract": "  Parkinson's Disease (PD) is a chronic, degenerative disorder which leads to a\nrange of motor and cognitive symptoms. PD diagnosis is a challenging task since\nits symptoms are very similar to other diseases such as normal ageing and\nessential tremor. Much research has been applied to diagnosing this disease.\nThis project aims to automate the PD diagnosis process using deep learning,\nRecursive Neural Networks (RNN) and Convolutional Neural Networks (CNN), to\ndifferentiate between healthy and PD patients. Besides that, since different\ndatasets may capture different aspects of this disease, this project aims to\nexplore which PD test is more effective in the discrimination process by\nanalysing different imaging and movement datasets (notably cube and spiral\npentagon datasets). In addition, this project evaluates which dataset type,\nimaging or time series, is more effective in diagnosing PD.\n"}
{"abstract": "  In informal learning scenarios the popularity of multimedia content, such as\nvideo tutorials or lectures, has significantly increased. Yet, the users'\ninteractions, navigation behavior, and consequently learning outcome, have not\nbeen researched extensively. Related work in this field, also called search as\nlearning, has focused on behavioral or text resource features to predict\nlearning outcome and knowledge gain. In this paper, we investigate whether we\ncan exploit features representing multimedia resource consumption to predict of\nknowledge gain (KG) during Web search from in-session data, that is without\nprior knowledge about the learner. For this purpose, we suggest a set of\nmultimedia features related to image and video consumption. Our feature\nextraction is evaluated in a lab study with 113 participants where we collected\ndata for a given search as learning task on the formation of thunderstorms and\nlightning. We automatically analyze the monitored log data and utilize\nstate-of-the-art computer vision methods to extract features about the seen\nmultimedia resources. Experimental results demonstrate that multimedia features\ncan improve KG prediction. Finally, we provide an analysis on feature\nimportance (text and multimedia) for KG prediction.\n"}
{"abstract": "  A proper channel modeling methodology that characterizes the statistics of\nextreme events is key in the design of a system at an ultra-reliable regime of\noperation. The strict constraint of ultra-reliability corresponds to the packet\nerror rate (PER) in the range of $10^{-9}-10^{-5}$ within the acceptable\nlatency on the order of milliseconds. Extreme value theory (EVT) is a robust\nframework for modeling the statistical behavior of extreme events in the\nchannel data. In this paper, we propose a methodology based on EVT to model the\nextreme events of a non-stationary wireless channel for the ultra-reliable\nregime of operation. This methodology includes techniques for splitting the\nchannel data sequence into multiple groups concerning the environmental factors\ncausing non-stationarity, and fitting the lower tail distribution of the\nreceived power in each group to the generalized Pareto distribution (GPD). The\nproposed approach also consists of optimally determining the time-varying\nthreshold over which the tail statistics are derived as a function of time, and\nassessing the validity of the derived Pareto model. Finally, the proposed\napproach chooses the best model with minimum complexity that represents the\ntime variation behavior of the non-stationary channel data sequence. Based on\nthe data collected within the engine compartment of Fiat Linea under various\nengine vibrations and driving scenarios, we demonstrate the capability of the\nproposed methodology in providing the best fit to the extremes of the\nnon-stationary data, significantly outperforming the channel modeling approach\nin characterizing the extreme events with the stationary channel assumption.\n"}
{"abstract": "  What is the role of topology in the propagation of quantum light in photonic\nlattices? We address this question by studying the propagation of squeezed\nstates in a topological one-dimensional waveguide array, benchmarking our\nresults with those for a topologically trivial localized state, and studying\ntheir robustness against disorder. Specifically, we study photon statistics,\none-mode and two-mode squeezing, and entanglement generation when the localized\nstate is excited with squeezed light. These quantum properties inherit the\nshape of the localized state but, more interestingly, and unlike in the\ntopologically trivial case, we find that propagation of squeezed light in a\ntopologically protected state robustly preserves the phase of the squeezed\nquadrature as the system evolves. We show how this latter topological advantage\ncan be harnessed for quantum information protocols.\n"}
{"abstract": "  Batch codes are a useful notion of locality for error correcting codes,\noriginally introduced in the context of distributed storage and cryptography.\nMany constructions of batch codes have been given, but few lower bound\n(limitation) results are known, leaving gaps between the best known\nconstructions and best known lower bounds. Towards determining the optimal\nredundancy of batch codes, we prove a new lower bound on the redundancy of\nbatch codes. Specifically, we study (primitive, multiset) linear batch codes\nthat systematically encode $n$ information symbols into $N$ codeword symbols,\nwith the requirement that any multiset of $k$ symbol requests can be obtained\nin disjoint ways. We show that such batch codes need $\\Omega(\\sqrt{Nk})$\nsymbols of redundancy, improving on the previous best lower bounds of\n$\\Omega(\\sqrt{N}+k)$ at all $k=n^\\varepsilon$ with $\\varepsilon\\in(0,1)$. Our\nproof follows from analyzing the dimension of the order-$O(k)$ tensor of the\nbatch code's dual code.\n"}
{"abstract": "  Cosmic rays with energies up to a few PeV are known to be accelerated within\nthe Milky Way. Traditionally, it has been presumed that supernova remnants were\nthe main source of very-high-energy cosmic rays but theoretically it is\ndifficult to get protons to PeV energies and observationally there simply is no\nevidence to support the remnants as sources of hadrons with energies above a\nfew tens of TeV. One possible source of protons with those energies is the\nGalactic Center region. Here we report observations of 1-100 TeV gamma rays\ncoming from the 'Cygnus Cocoon', which is a superbubble surrounding a region of\nOB2 massive star formation. These gamma rays are likely produced by 10-1000 TeV\nfreshly accelerated CRs originating from the enclosed star forming region\nCygnus OB2. Hitherto it was not known that such regions could accelerate\nparticles to these energies. The measured flux is likely originated by hadronic\ninteractions. The spectral shape and the emission profile of the Cocoon changes\nfrom GeV to TeV energies, which reveals the transport of cosmic particles and\nhistorical activity in the superbubble.\n"}
{"abstract": "  It is well known that physical-layer key generation methods enable wireless\ndevices to harvest symmetric keys by accessing the randomness offered by the\nwireless channels. Although two-user key generation is well understood, group\nsecret-key (GSK) generation, wherein more than two nodes in a network generate\nsecret-keys, still poses open problems. Recently, Manish Rao et al., have\nproposed the Algebraic Symmetrically Quantized GSK (A-SQGSK) protocol for a\nnetwork of three nodes wherein the nodes share quantized versions of the\nchannel realizations over algebraic rings, and then harvest a GSK. Although\nA-SQGSK protocol guarantees confidentiality of common randomness to an\neavesdropper, we observe that the key-rate of the protocol is poor since only\none channel in the network is used to harvest GSK. Identifying this limitation,\nin this paper, we propose an opportunistic selection method wherein more than\none wireless channel is used to harvest GSKs without compromising the\nconfidentiality feature, thereby resulting in remarkable improvements in the\nkey-rate. Furthermore, we also propose a log-likelihood ratio (LLR) generation\nmethod for the common randomness observed at various nodes, so that the\nsoft-values are applied to execute LDPC codes based reconciliation to reduce\nthe bit mismatches among the nodes.\n"}
{"abstract": "  Computer security and user privacy are critical issues and concerns in the\ndigital era due to both increasing users and threats to their data. Separate\nissues arise between generic cybersecurity guidance (i.e., protect all user\ndata from malicious threats) and the individualistic approach of privacy (i.e.,\nspecific to users and dependent on user needs and risk perceptions). Research\nhas shown that several security- and privacy-focused vulnerabilities are\ntechnological (e.g., software bugs (Streiff, Kenny, Das, Leeth, & Camp, 2018),\ninsecure authentication (Das, Wang, Tingle, & Camp, 2019)), or behavioral\n(e.g., sharing passwords (Das, Dingman, & Camp, 2018); and compliance (Das,\nDev, & Srinivasan, 2018) (Dev, Das, Rashidi, & Camp, 2019)). This panel\nproposal addresses a third category of sociotechnical vulnerabilities that can\nand sometimes do arise from non-inclusive design of security and privacy. In\nthis panel, we will address users' needs and desires for privacy. The panel\nwill engage in in-depth discussions about value-sensitive design while focusing\non potentially vulnerable populations, such as older adults, teens, persons\nwith disabilities, and others who are not typically emphasized in general\nsecurity and privacy concerns. Human factors have a stake in and ability to\nfacilitate improvements in these areas.\n"}
{"abstract": "  We provide a new formulation of the Local Friendliness no-go theorem of Bong\net al [Nat. Phys. 16, 1199 (2020)] from fundamental causal principles,\nproviding another perspective on how it puts strictly stronger bounds on\nquantum reality than Bell's theorem. In particular, quantum causal models have\nbeen proposed as a way to maintain a peaceful coexistence between quantum\nmechanics and relativistic causality, while respecting Leibniz's methodological\nprinciple. This works for Bell's theorem but does not work for the Local\nFriendliness no-go theorem, which considers an extended Wigner's Friend\nscenario. More radical conceptual renewal is required; we suggest that cleaving\nto Leibniz's principle requires extending relativity to events themselves.\n"}
{"abstract": "  When solving two-player zero-sum games, multi-agent reinforcement learning\n(MARL) algorithms often create populations of agents where, at each iteration,\na new agent is discovered as the best response to a mixture over the opponent\npopulation. Within such a process, the update rules of \"who to compete with\"\n(i.e., the opponent mixture) and \"how to beat them\" (i.e., finding best\nresponses) are underpinned by manually developed game theoretical principles\nsuch as fictitious play and Double Oracle. In this paper, we introduce a novel\nframework -- Neural Auto-Curricula (NAC) -- that leverages meta-gradient\ndescent to automate the discovery of the learning update rule without explicit\nhuman design. Specifically, we parameterise the opponent selection module by\nneural networks and the best-response module by optimisation subroutines, and\nupdate their parameters solely via interaction with the game engine, where both\nplayers aim to minimise their exploitability. Surprisingly, even without human\ndesign, the discovered MARL algorithms achieve competitive or even better\nperformance with the state-of-the-art population-based game solvers (e.g.,\nPSRO) on Games of Skill, differentiable Lotto, non-transitive Mixture Games,\nIterated Matching Pennies, and Kuhn Poker. Additionally, we show that NAC is\nable to generalise from small games to large games, for example training on\nKuhn Poker and outperforming PSRO on Leduc Poker. Our work inspires a promising\nfuture direction to discover general MARL algorithms solely from data.\n"}
{"abstract": "  It is suggested that low-light image enhancement realizes one-to-many mapping\nsince we have different definitions of NORMAL-light given application scenarios\nor users' aesthetic. However, most existing methods ignore subjectivity of the\ntask, and simply produce one result with fixed brightness. This paper proposes\na neural network for multi-level low-light image enhancement, which is\nuser-friendly to meet various requirements by selecting different images as\nbrightness reference. Inspired by style transfer, our method decomposes an\nimage into two low-coupling feature components in the latent space, which\nallows the concatenation feasibility of the content components from low-light\nimages and the luminance components from reference images. In such a way, the\nnetwork learns to extract scene-invariant and brightness-specific information\nfrom a set of image pairs instead of learning brightness differences. Moreover,\ninformation except for the brightness is preserved to the greatest extent to\nalleviate color distortion. Extensive results show strong capacity and\nsuperiority of our network against existing methods.\n"}
{"abstract": "  We reproduce the electronic properties of FeSe in the high-temperature phase\nwithin an ab initio framework that includes screened Fock exchange and local\ndynamical correlations. We robustly capture the experimental band structure, as\nlong as the system is in the Hund's metal phase. In particular, we account for\nthe shrinking of the Fermi pockets and the sinking below the Fermi level of the\nhole pocket with $xy$ orbital character. This entails the elusive correct\nestimate of the Sommerfeld coefficient, and supports the interpretation of\nnon-compensated Fermi pockets seen in ARPES in terms of surface electron\ndoping. More stringently, our modeling matches well the experimental interband\noptical spectrum, and captures qualitatively the temperature dependence of the\nthermoelectric power, extremely sensitive to the details of the bands around\nthe Fermi level.\n"}
{"abstract": "  Adaptive optics (AO) systems using tomographic estimation of\nthree-dimensional structure of atmospheric turbulence requires vertical\natmospheric turbulence profile, which describes turbulence strength as a\nfunction of altitude as a prior information. We propose a novel method to\nreconstruct the profile by applying Multi Aperture Scintillation Sensor (MASS)\nmethod to scintillation data obtained by a Shack-Hartmann wavefront sensor\n(SH-WFS). Compared to the traditional MASS, which uses atmospheric\nscintillation within 4 concentric annular apertures, the new method utilizes\nscintillation in several hundreds of spatial patterns, which are created by\ncombinations of SH-WFS subapertures. Accuracy of the turbulence profile\nreconstruction is evaluated with Bayesian inference, and it is confirmed that\nturbulence profile with more than 10 layers can be reconstructed thanks to the\nlarge number of constraints. We demonstrate the new method with a SH-WFS\nattached to the 50 cm telescope at Tohoku university and confirm that general\ncharacteristics of atmospheric turbulence profile is reproduced.\n"}
{"abstract": "  Client contribution evaluation, also known as data valuation, is a crucial\napproach in federated learning(FL) for client selection and incentive\nallocation. However, due to restrictions of accessibility of raw data, only\nlimited information such as local weights and local data size of each client is\nopen for quantifying the client contribution. Using data size from available\ninformation, we introduce an empirical evaluation method called Federated\nClient Contribution Evaluation through Accuracy Approximation(FedCCEA). This\nmethod builds the Accuracy Approximation Model(AAM), which estimates a\nsimulated test accuracy using inputs of sampled data size and extracts the\nclients' data quality and data size to measure client contribution. FedCCEA\nstrengthens some advantages: (1) enablement of data size selection to the\nclients, (2) feasible evaluation time regardless of the number of clients, and\n(3) precise estimation in non-IID settings. We demonstrate the superiority of\nFedCCEA compared to previous methods through several experiments: client\ncontribution distribution, client removal, and robustness test to partial\nparticipation.\n"}
{"abstract": "  In this paper, we consider a single top quark production in association with\na neutral gauge boson (Z boson or Photon) at the future electron-positron\ncolliders. We utilize these channels to probe the top quark Flavour Changing\nNeutral Currents (FCNC) interactions in $tqZ$ and $tq\\gamma$ vertices. We\nperform two separate analyses for top-Z-jet and top-$\\gamma$-jet channels. The\nStandard Model Effective Field Theory (SMEFT) approach is employed to search\nfor these anomalous couplings. We consider parton showering, hadronization and\nfast detector simulation in our study and use a cut-and-count technique to\nseparate the signal from the Standard Model (SM) background processes. The\nupper limits on the FCNC couplings at $95\\%$ confidence level for the different\nintegrated luminosities are obtained. It is shown that the future lepton\ncollider would be able to probe the FCNC branching fractions to\n$\\rm{Br}(t\\rightarrow q \\gamma)< 10^{-4}$ and $\\rm{Br}(t\\rightarrow q Z) <\n10^{-4}$ with $3$ $\\rm{ab}^{-1}$of integrated luminosity of data at a center of\nmass energy of $350$ GeV.\n"}
{"abstract": "  For the biharmonic Steklov eigenvalue problem considered in this paper, we\nshow that among all bounded Euclidean domains of class $C^{1}$ with fixed\nmeasure, the ball maximizes the first positive eigenvalue.\n"}
{"abstract": "  We use the quantum Fisher information (QFI) to diagnose a dynamical phase\ntransition (DPT) in a closed quantum system, which is usually defined in terms\nof non-analytic behaviour of a time-averaged order parameter. Employing the\nLipkin-Meshkov-Glick model as an illustrative example, we find that the DPT\ncorrelates with a peak in the QFI that can be explained by a generic connection\nto an underlying excited-state quantum phase transition that also enables us to\nalso relate the scaling of the QFI with the behaviour of the order parameter.\nMotivated by the QFI as a quantifier of metrologically useful correlations and\nentanglement, we also present a robust interferometric protocol that can enable\nDPTs as a platform for quantum-enhanced sensing.\n"}
{"abstract": "  Vision transformer has achieved competitive performance on a variety of\ncomputer vision applications. However, their storage, run-time memory, and\ncomputational demands are hindering the deployment to mobile devices. Here we\npresent a vision transformer pruning approach, which identifies the impacts of\ndimensions in each layer of transformer and then executes pruning accordingly.\nBy encouraging dimension-wise sparsity in the transformer, important dimensions\nautomatically emerge. A great number of dimensions with small importance scores\ncan be discarded to achieve a high pruning ratio without significantly\ncompromising accuracy. The pipeline for vision transformer pruning is as\nfollows: 1) training with sparsity regularization; 2) pruning dimensions of\nlinear projections; 3) fine-tuning. The reduced parameters and FLOPs ratios of\nthe proposed algorithm are well evaluated and analyzed on ImageNet dataset to\ndemonstrate the effectiveness of our proposed method.\n"}
{"abstract": "  Social media are digitalising massive amounts of users' cognitions in terms\nof timelines and emotional content. Such Big Data opens unprecedented\nopportunities for investigating cognitive phenomena like perception,\npersonality and information diffusion but requires suitable interpretable\nframeworks. Since social media data come from users' minds, worthy candidates\nfor this challenge are cognitive networks, models of cognition giving structure\nto mental conceptual associations. This work outlines how cognitive network\nscience can open new, quantitative ways for understanding cognition through\nonline media, like: (i) reconstructing how users semantically and emotionally\nframe events with contextual knowledge unavailable to machine learning, (ii)\ninvestigating conceptual salience/prominence through knowledge structure in\nsocial discourse; (iii) studying users' personality traits like\nopenness-to-experience, curiosity, and creativity through language in posts;\n(iv) bridging cognitive/emotional content and social dynamics via multilayer\nnetworks comparing the mindsets of influencers and followers. These\nadvancements combine cognitive-, network- and computer science to understand\ncognitive mechanisms in both digital and real-world settings but come with\nlimitations concerning representativeness, individual variability and data\nintegration. Such aspects are discussed along the ethical implications of\nmanipulating socio-cognitive data. In the future, reading cognitions through\nnetworks and social media can expose cognitive biases amplified by online\nplatforms and relevantly inform policy making, education and markets about\nmassive, complex cognitive trends.\n"}
{"abstract": "  We study the effects of introducing a feedback channel in the two-receiver\nerasure source-broadcast problem in which a binary equiprobable source is to be\nsent over an erasure broadcast channel to two receivers subject to erasure\ndistortion constraints. The receivers each require a certain fraction of a\nsource sequence, and we are interested in the minimum latency, or transmission\ntime, required to serve them all. We first show that for a two-user broadcast\nchannel, a point-to-point outer bound can always be achieved. We further show\nthat the point-to-point outer bound can also be achieved if only one of the\nusers, the stronger user, has a feedback channel. Our coding scheme relies on a\nhybrid approach that combines transmitting both random linear combinations of\nsource symbols as well as a retransmission strategy.\n"}
{"abstract": "  We consider robust variants of the standard optimal transport, named robust\noptimal transport, where marginal constraints are relaxed via Kullback-Leibler\ndivergence. We show that Sinkhorn-based algorithms can approximate the optimal\ncost of robust optimal transport in\n$\\widetilde{\\mathcal{O}}(\\frac{n^2}{\\varepsilon})$ time, in which $n$ is the\nnumber of supports of the probability distributions and $\\varepsilon$ is the\ndesired error. Furthermore, we investigate a fixed-support robust barycenter\nproblem between $m$ discrete probability distributions with at most $n$ number\nof supports and develop an approximating algorithm based on iterative Bregman\nprojections (IBP). For the specific case $m = 2$, we show that this algorithm\ncan approximate the optimal barycenter value in\n$\\widetilde{\\mathcal{O}}(\\frac{mn^2}{\\varepsilon})$ time, thus being better\nthan the previous complexity\n$\\widetilde{\\mathcal{O}}(\\frac{mn^2}{\\varepsilon^2})$ of the IBP algorithm for\napproximating the Wasserstein barycenter.\n"}
{"abstract": "  CP4 3HDM is a unique three-Higgs-doublet model equipped with a higher-order\nCP symmetry in the scalar and Yukawa sector. Based on a single assumption (the\nminimal model with a CP-symmetry of order 4 and no accidental symmetry), it\nleads to a remarkable correlation between its scalar and Yukawa sectors, which\nechoes in its phenomenology. A recent scan of the parameter space of CP4 3HDM\nunder the assumption of scalar alignment identified a few dozens of points\nwhich passed many flavour constraints. In the present work we show, however,\nthat almost all of these points are now ruled out by the recent LHC searches of\n$t \\to H^+ b$ with subsequent hadronic decays of $H^+$. Apart from a few points\nwith charged Higgses heavier than the top quark, only one point survives all\nthe checks, the model with an exotic, non-2HDM-like generation pattern of $H^+$\ncouplings with quarks. One can expect many more points with exotic $H^+$\ncouplings to quarks if the scalar alignment assumption is relaxed.\n"}
{"abstract": "  In many real applications such as the data integration, social network\nanalysis, and the Semantic Web, the entity resolution (ER) is an important and\nfundamental problem, which identifies and links the same real-world entities\nfrom various data sources. While prior works usually consider ER over static\nand complete data, in practice, application data are usually collected in a\nstreaming fashion, and often incur missing attributes (due to the inaccuracy of\ndata extraction techniques). Therefore, in this paper, we will formulate and\ntackle a novel problem, topic-aware entity resolution over incomplete data\nstreams (TER-iDS), which online imputes incomplete tuples and detects pairs of\ntopic-related matching entities from incomplete data streams. In order to\neffectively and efficiently tackle the TER-iDS problem, we propose an effective\nimputation strategy, carefully design effective pruning strategies, as well as\nindexes/synopsis, and develop an efficient TER-iDS algorithm via index joins.\nExtensive experiments have been conducted to evaluate the effectiveness and\nefficiency of our proposed TER-iDS approach over real data sets.\n"}
{"abstract": "  We review the recent progress in Higgs inflation focusing on Higgs-$R^2$\ninflation, primordial black hole production and the $R^3$ term.\n"}
{"abstract": "  In many sequential decision-making problems, the individuals are split into\nseveral batches and the decision-maker is only allowed to change her policy at\nthe end of batches. These batch problems have a large number of applications,\nranging from clinical trials to crowdsourcing. Motivated by this, we study the\nstochastic contextual bandit problem for general reward distributions under the\nbatched setting. We propose the BatchNeuralUCB algorithm which combines neural\nnetworks with optimism to address the exploration-exploitation tradeoff while\nkeeping the total number of batches limited. We study BatchNeuralUCB under both\nfixed and adaptive batch size settings and prove that it achieves the same\nregret as the fully sequential version while reducing the number of policy\nupdates considerably. We confirm our theoretical results via simulations on\nboth synthetic and real-world datasets.\n"}
{"abstract": "  The magnetic properties of Fe$_3$O$_4$ nanoparticle assemblies have been\ninvestigated in detail through a combination of vibrating sample magnetometry\nand muon spin relaxation ($\\mu$SR) techniques. Two samples with average\nparticle sizes of 5 nm and 20 nm, respectively, were studied. For both samples,\nthe magnetometry and $\\mu$SR results exhibit clear signatures of the\nsuperparmagnetic state at high temperature and the magnetically blocked state\nat low temperature. The $\\mu$SR data demonstrate that the transition from the\nsuperparamagnetic to the blocked state occurs gradually throughout the sample\nvolume over a broad temperature range due to the finite particle size\ndistribution of each sample. The transition occurs between approximately 3 K\nand 45 K for the 5 nm sample and 150 K and 300 K for the 20 nm sample. The\nmagnetometry and $\\mu$SR data are further analyzed to yield estimates of\nmicroscopic magnetic parameters including the nanoparticle spin-flip activation\nenergy $E_A$, magnetic anisotropy $K$, and intrinsic nanoparticle spin reversal\nattempt time $\\tau_0$. These results highlight the complementary information\nabout magnetic nanoparticles that can be obtained by bulk magnetic probes such\nas magnetometry and local magnetic probes such as $\\mu$SR.\n"}
{"abstract": "  We propose a universal quantum circuit design that can estimate any arbitrary\none-dimensional periodic functions based on the corresponding Fourier\nexpansion. The quantum circuit contains N-qubits to store the information on\nthe different N-Fourier components and $M+2$ auxiliary qubits with $M =\n\\lceil{\\log_2{N}}\\rceil$ for control operations. The desired output will be\nmeasured in the last qubit $q_N$ with a time complexity of the computation of\n$O(N^2\\lceil \\log_2N\\rceil^2)$. We illustrate the approach by constructing the\nquantum circuit for the square wave function with accurate results obtained by\ndirect simulations using the IBM-QASM simulator. The approach is general and\ncan be applied to any arbitrary periodic function.\n"}
{"abstract": "  We study primary submodules and primary decompositions from a differential\nand computational point of view. Our main theoretical contribution is a general\nstructure theory and a representation theorem for primary submodules of an\narbitrary finitely generated module over a polynomial ring. We characterize\nprimary submodules in terms of differential operators and punctual Quot\nschemes. Moreover, we introduce and implement an algorithm that computes a\nminimal differential primary decomposition for a module.\n"}
{"abstract": "  This study presents an application of machine learning (ML) methods for\ndetecting the presence of stenoses and aneurysms in the human arterial system.\nFour major forms of arterial disease -- carotid artery stenosis (CAS),\nsubclavian artery stenosis (SAC), peripheral arterial disease (PAD), and\nabdominal aortic aneurysms (AAA) -- are considered. The ML methods are trained\nand tested on a physiologically realistic virtual patient database (VPD)\ncontaining 28,868 healthy subjects, which is adapted from the authors previous\nwork and augmented to include the four disease forms. Six ML methods -- Naive\nBayes, Logistic Regression, Support Vector Machine, Multi-layer Perceptron,\nRandom Forests, and Gradient Boosting -- are compared with respect to\nclassification accuracies and it is found that the tree-based methods of Random\nForest and Gradient Boosting outperform other approaches. The performance of ML\nmethods is quantified through the F1 score and computation of sensitivities and\nspecificities. When using all the six measurements, it is found that maximum F1\nscores larger than 0.9 are achieved for CAS and PAD, larger than 0.85 for SAS,\nand larger than 0.98 for both low- and high-severity AAAs. Corresponding\nsensitivities and specificities are larger than 90% for CAS and PAD, larger\nthan 85% for SAS, and larger than 98% for both low- and high-severity AAAs.\nWhen reducing the number of measurements, it is found that the performance is\ndegraded by less than 5% when three measurements are used, and less than 10%\nwhen only two measurements are used for classification. For AAA, it is shown\nthat F1 scores larger than 0.85 and corresponding sensitivities and\nspecificities larger than 85% are achievable when using only a single\nmeasurement. The results are encouraging to pursue AAA monitoring and screening\nthrough wearable devices which can reliably measure pressure or flow-rates\n"}
{"abstract": "  We show that the natural Glauber dynamics mixes rapidly and generates a\nrandom proper edge-coloring of a graph with maximum degree $\\Delta$ whenever\nthe number of colors is at least $q\\geq (\\frac{10}{3} + \\epsilon)\\Delta$, where\n$\\epsilon>0$ is arbitrary and the maximum degree satisfies $\\Delta \\geq C$ for\na constant $C = C(\\epsilon)$ depending only on $\\epsilon$. For edge-colorings,\nthis improves upon prior work \\cite{Vig99, CDMPP19} which show rapid mixing\nwhen $q\\geq (\\frac{11}{3}-\\epsilon_0 ) \\Delta$, where $\\epsilon_0 \\approx\n10^{-5}$ is a small fixed constant. At the heart of our proof, we establish a\nmatrix trickle-down theorem, generalizing Oppenheim's influential result, as a\nnew technique to prove that a high dimensional simplical complex is a local\nspectral expander.\n"}
{"abstract": "  Given a graph $G$, the burning number of $G$ is the smallest integer $k$ for\nwhich there are vertices $x_1, x_2,\\ldots,x_k$ such that $(x_1,x_2,\\ldots,x_k)$\nis a burning sequence of $G$. It has been shown that the graph burning problem\nis NP-complete, even for trees with maximum degree three, or linear forests. A\n$t$-unicyclic graph is a unicycle graph with exactly one vertex of degree\ngreater than $2$. In this paper, we first present the bounds for the burning\nnumber of $t$-unicyclic graphs, and then use the burning numbers of linear\nforests with at most three components to determine the burning number of all\n$t$-unicyclic graphs for $t\\le 2$.\n"}
{"abstract": "  Agricultural production using high technology is an inevitable trend in\nVietnam. Especially for material crops which typically need large growing\nareas, wireless sensor networks has been clearly playing a significant role in\nincreasing productivity, monitoring pests and diseases, mitigating the impact\nof climate change, and reducing the direct labor of cultivators. This paper\nconstructs an experimental model of agricultural crop field monitoring using a\ncombination of LoRa wireless sensor networks and unmanned aerial vehicles to\ncollect data on conditions of weather and soil, plant health, which helps\ngrowers easily making right decisions on solutions for irrigation, pest\ntreatment, and fertilization with the currently planted crops. The system has\nbeen developed and experimentized in the field to evaluate some basic features\nand justified the stability and reliability of the obtained data.\n"}
{"abstract": "  An array of low-frequency dipole antennas on the lunar farside surface will\nprobe a unique, unexplored epoch in the early Universe called the Dark Ages. It\nbegins at Recombination when neutral hydrogen atoms formed, first revealed by\nthe cosmic microwave background. This epoch is free of stars and astrophysics,\nso it is ideal to investigate high energy particle processes including dark\nmatter, early Dark Energy, neutrinos, and cosmic strings. A NASA-funded study\ninvestigated the design of the instrument and the deployment strategy from a\nlander of 128 pairs of antenna dipoles across a 10 kmx10 km area on the lunar\nsurface. The antenna nodes are tethered to the lander for central data\nprocessing, power, and data transmission to a relay satellite. The array, named\nFARSIDE, would provide the capability to image the entire sky in 1400 channels\nspanning frequencies from 100 kHz to 40 MHz, extending down two orders of\nmagnitude below bands accessible to ground-based radio astronomy. The lunar\nfarside can simultaneously provide isolation from terrestrial radio frequency\ninterference, the Earth's auroral kilometric radiation, and plasma noise from\nthe solar wind. It is thus the only location within the inner solar system from\nwhich sky noise limited observations can be carried out at sub-MHz frequencies.\nThrough precision calibration via an orbiting beacon and exquisite foreground\ncharacterization, the farside array would measure the Dark Ages global 21-cm\nsignal at redshifts z~35-200. It will also be a pathfinder for a larger 21-cm\npower spectrum instrument by carefully measuring the foreground with high\ndynamic range.\n"}
{"abstract": "  For a Galois extension $K/F$ with $\\text{char}(K)\\neq 2$ and $\\text{Gal}(K/F)\n\\simeq \\mathbb{Z}/2\\mathbb{Z}\\oplus\\mathbb{Z}/2\\mathbb{Z}$, we determine the\n$\\mathbb{F}_2[\\text{Gal}(K/F)]$-module structure of $K^\\times/K^{\\times 2}$.\nAlthough there are an infinite number of (pairwise non-isomorphic)\nindecomposable\n$\\mathbb{F}_2[\\mathbb{Z}/2\\mathbb{Z}\\oplus\\mathbb{Z}/2\\mathbb{Z}]$-modules, our\ndecomposition includes at most $9$ indecomposable types. This paper marks the\nfirst time that the Galois module structure of power classes of a field has\nbeen fully determined when the modular representation theory allows for an\ninfinite number of indecomposable types.\n"}
{"abstract": "  While coursework provides undergraduate data science students with some\nrelevant analytic skills, many are not given the rich experiences with data and\ncomputing they need to be successful in the workplace. Additionally, students\noften have limited exposure to team-based data science and the principles and\ntools of collaboration that are encountered outside of school. In this paper,\nwe describe the DSC-WAV program, an NSF-funded data science workforce\ndevelopment project in which teams of undergraduate sophomores and juniors work\nwith a local non-profit organization on a data-focused problem. To help\nstudents develop a sense of agency and improve confidence in their technical\nand non-technical data science skills, the project promoted a team-based\napproach to data science, adopting several processes and tools intended to\nfacilitate this collaboration. Evidence from the project evaluation, including\nparticipant survey and interview data, is presented to document the degree to\nwhich the project was successful in engaging students in team-based data\nscience, and how the project changed the students' perceptions of their\ntechnical and non-technical skills. We also examine opportunities for\nimprovement and offer insight to other data science educators who may want to\nimplement a similar team-based approach to data science projects at their own\ninstitutions.\n"}
{"abstract": "  Tensor computations overwhelm traditional general-purpose computing devices\ndue to the large amounts of data and operations of the computations. They call\nfor a holistic solution composed of both hardware acceleration and software\nmapping. Hardware/software (HW/SW) co-design optimizes the hardware and\nsoftware in concert and produces high-quality solutions. There are two main\nchallenges in the co-design flow. First, multiple methods exist to partition\ntensor computation and have different impacts on performance and energy\nefficiency. Besides, the hardware part must be implemented by the intrinsic\nfunctions of spatial accelerators. It is hard for programmers to identify and\nanalyze the partitioning methods manually. Second, the overall design space\ncomposed of HW/SW partitioning, hardware optimization, and software\noptimization is huge. The design space needs to be efficiently explored.\n  To this end, we propose an agile co-design approach HASCO that provides an\nefficient HW/SW solution to dense tensor computation. We use tensor syntax\ntrees as the unified IR, based on which we develop a two-step approach to\nidentify partitioning methods. For each method, HASCO explores the hardware and\nsoftware design spaces. We propose different algorithms for the explorations,\nas they have distinct objectives and evaluation costs. Concretely, we develop a\nmulti-objective Bayesian optimization algorithm to explore hardware\noptimization. For software optimization, we use heuristic and Q-learning\nalgorithms. Experiments demonstrate that HASCO achieves a 1.25X to 1.44X\nlatency reduction through HW/SW co-design compared with developing the hardware\nand software separately.\n"}
{"abstract": "  We have searched nearly 40,000 inorganic solids in the Inorganic Crystal\nStructural Database to identify compounds containing a transition metal or rare\nearth kagom\\'e sublattice, a geometrically magnetically frustrated lattice,\nultimately identifying $\\sim$500 materials. A broad analysis of the chemical\nand structural trends of these materials shows three types of kagom\\'e sheet\nstacking and several classes of magnetic complexity. Following the search and\nclassification, we rapidly screen the magnetic properties of a subset of the\nmaterials using density functional theory to eliminate those that are unlikely\nto exhibit magnetic frustration. From the results of our computational\nscreening, we rediscover six materials that have previously been explored for\ntheir low temperature magnetic behavior, albeit showing symmetry breaking\ndistortions, spin glass behavior, or magnetic ordering. However, all are\nmaterials with antiferromagnetic behavior, which we correctly predict. Finally,\nwe also report three materials that appear to be unexplored for their magnetic\nproperties.\n"}
{"abstract": "  Multiple-instance learning is a subset of weakly supervised learning where\nlabels are applied to sets of instances rather than the instances themselves.\nUnder the standard assumption, a set is positive only there is if at least one\ninstance in the set which is positive.\n  This paper introduces a series of multiple-instance learning benchmarks\ngenerated from MNIST, Fashion-MNIST, and CIFAR10. These benchmarks test the\nstandard, presence, absence, and complex assumptions and provide a framework\nfor future benchmarks to be distributed. I implement and evaluate several\nmultiple-instance learning techniques against the benchmarks. Further, I\nevaluate the Noisy-And method with label noise and find mixed results with\ndifferent datasets. The models are implemented in TensorFlow 2.4.1 and are\navailable on GitHub. The benchmarks are available from PyPi as mil-benchmarks\nand on GitHub.\n"}
{"abstract": "  To any combinatorial triangulation $T$ of a square, there is an associated\npolynomial relation $p_T$ among the areas of the triangles of $T$. With the\ngoal of understanding this polynomial, we consider polynomials obtained from\n$p_T$ by choosing $l$ of its variables and specializing $p_T$ to these\nvariables by zeroing out the remaining variables. We show that for fixed $l$,\nthe set ${\\mathcal E}_l$ of integer polynomials that appear as irreducible\nfactors of such specializations is finite. We compute this area encyclopedia\n${\\mathcal E}_l$ for $l\\leq 4$. We also show that in any dissection of a square\ninto $l$ triangles, the areas of the triangles must satisfy a polynomial in\n${\\mathcal E}_l$. Our results are obtained by studying the rational map that\nassociates to each drawing of $T$ the tuple of areas of the triangles in that\ndrawing. By analyzing the ways of approaching the base locus, we derive\nrestrictions on points of the closure of the image of this map.\n"}
{"abstract": "  BERT-based text ranking models have dramatically advanced the\nstate-of-the-art in ad-hoc retrieval, wherein most models tend to consider\nindividual query-document pairs independently. In the mean time, the importance\nand usefulness to consider the cross-documents interactions and the\nquery-specific characteristics in a ranking model have been repeatedly\nconfirmed, mostly in the context of learning to rank. The BERT-based ranking\nmodel, however, has not been able to fully incorporate these two types of\nranking context, thereby ignoring the inter-document relationships from the\nranking and the differences among queries. To mitigate this gap, in this work,\nan end-to-end transformer-based ranking model, named Co-BERT, has been proposed\nto exploit several BERT architectures to calibrate the query-document\nrepresentations using pseudo relevance feedback before modeling the relevance\nof a group of documents jointly. Extensive experiments on two standard test\ncollections confirm the effectiveness of the proposed model in improving the\nperformance of text re-ranking over strong fine-tuned BERT-Base baselines. We\nplan to make our implementation open source to enable further comparisons.\n"}
{"abstract": "  The multilinear Pagerank model [Gleich, Lim and Yu, 2015] is a tensor-based\ngeneralization of the Pagerank model. Its computation requires solving a system\nof polynomial equations that contains a parameter $\\alpha \\in [0,1)$. For\n$\\alpha \\approx 1$, this computation remains a challenging problem, especially\nsince the solution may be non-unique. Extrapolation strategies that start from\nsmaller values of $\\alpha$ and `follow' the solution by slowly increasing this\nparameter have been suggested; however, there are known cases where these\nstrategies fail, because a globally continuous solution curve cannot be defined\nas a function of $\\alpha$. In this paper, we improve on this idea, by employing\na predictor-corrector continuation algorithm based on a more general\nrepresentation of the solutions as a curve in $\\mathbb{R}^{n+1}$. We prove\nseveral global properties of this curve that ensure the good behavior of the\nalgorithm, and we show in our numerical experiments that this method is\nsignificantly more reliable than the existing alternatives.\n"}
{"abstract": "  As defined in [1], a Hausdorff space is strongly anti-Urysohn (in short: SAU)\nif it has at least two non-isolated points and any two infinite} closed subsets\nof it intersect. Our main result answers the two main questions of [1] by\nproviding a ZFC construction of a locally countable SAU space of cardinality\n$2^{\\mathfrak{c}}$. The construction hinges on the existence of\n$2^{\\mathfrak{c}}$ weak P-points in $\\omega^*$, a very deep result of Ken\nKunen.\n  It remains open if SAU spaces of cardinality $> 2^{\\mathfrak{c}}$ could\nexist, while it was shown in [1] that $2^{2^{\\mathfrak{c}}}$ is an upper bound.\nAlso, we do not know if crowded SAU spaces, i.e. ones without any isolated\npoints, exist in ZFC but we obtained the following consistency results\nconcerning such spaces.\n  (1) It is consistent that $\\mathfrak{c}$ is as large as you wish and there is\na locally countable and crowded SAU space of cardinality $\\mathfrak{c}^+$.\n  (2) It is consistent that both $\\mathfrak{c}$ and $2^\\mathfrak{c}$ are as\nlarge as you wish and there is a crowded SAU space of cardinality\n$2^\\mathfrak{c}$.\n  (3) For any uncountable cardinal ${\\kappa}$ the following statements are\nequivalent:\n  (i) ${\\kappa}=cof({[{\\kappa}]}^{{\\omega}},\\subseteq)$.\n  (ii) There is a locally countable and crowded SAU space of size ${\\kappa}$ in\nthe generic extension obtained by adding $\\kappa$ Cohen reals.\n  (iii) There is a locally countable and countably compact $T_1$-space of size\n${\\kappa}$ in some CCC generic extension.\n  [1] I. Juhasz, L. Soukup, and Z. Szentmiklossy, Anti-Urysohn spaces, Top.\nAppl., 213 (2016), pp. 8--23.\n"}
{"abstract": "  For unsupervised image-to-image translation, we propose a discriminator\narchitecture which focuses on the statistical features instead of individual\npatches. The network is stabilized by distribution matching of key statistical\nfeatures at multiple scales. Unlike the existing methods which impose more and\nmore constraints on the generator, our method facilitates the shape deformation\nand enhances the fine details with a greatly simplified framework. We show that\nthe proposed method outperforms the existing state-of-the-art models in various\nchallenging applications including selfie-to-anime, male-to-female and glasses\nremoval.\n"}
{"abstract": "  A neural ordinary differential equations network (ODE-Net)-enabled\nreachability method (Neuro-Reachability) is devised for the dynamic\nverification of networked microgrids (NMs) with unidentified subsystems and\nheterogeneous uncertainties. Three new contributions are presented: 1) An\nODENet-enabled dynamic model discovery approach is devised to construct the\ndata-driven state-space model which preserves the nonlinear and differential\nstructure of the NMs system; 2) A physics-data-integrated (PDI) NMs model is\nestablished, which empowers various NM analytics; and 3) A\nconformance-empowered reachability analysis is developed to enhance the\nreliability of the PDI-driven dynamic verification. Extensive case studies\ndemonstrate the efficacy of the ODE-Net-enabled method in microgrid dynamic\nmodel discovery, and the effectiveness of the Neuro-Reachability approach in\nverifying the NMs dynamics under multiple uncertainties and various operational\nscenarios.\n"}
{"abstract": "  We consider interacting Bose particles in an external local potential. It is\nshown that large class of external quasicrystal potentials cannot sustain any\ntype of Bose-Einstein condensates. Accordingly, at spatial dimensions $D\\leq 2$\nin such quasicrystal potentials a supersolid is not possible via Bose-Einstein\ncondensates at finite temperatures. The latter also hold true for the\ntwo-dimensional Fibonacci tiling. However, supersolids do arise at $D\\leq 2$\nvia Bose-Einstein condensates from infinitely long-range, nonlocal\ninterparticle potentials.\n"}
{"abstract": "  Weighted timed games are two-player zero-sum games played in a timed\nautomaton equipped with integer weights. We consider optimal reachability\nobjectives, in which one of the players, that we call Min, wants to reach a\ntarget location while minimising the cumulated weight. While knowing if Min has\na strategy to guarantee a value lower than a given threshold is known to be\nundecidable (with two or more clocks), several conditions, one of them being\nthe divergence, have been given to recover decidability. In such weighted timed\ngames (like in untimed weighted games in the presence of negative weights), Min\nmay need finite memory to play (close to) optimally. This is thus tempting to\ntry to emulate this finite memory with other strategic capabilities. In this\nwork, we allow the players to use stochastic decisions, both in the choice of\ntransitions and of timing delays. We give for the first time a definition of\nthe expected value in weighted timed games, overcoming several theoretical\nchallenges. We then show that, in divergent weighted timed games, the\nstochastic value is indeed equal to the classical (deterministic) value, thus\nproving that Min can guarantee the same value while only using stochastic\nchoices, and no memory.\n"}
{"abstract": "  Noisy intermediate-scale quantum (NISQ) computers could solve\nquantum-mechanical simulation problems that are beyond the capabilities of\nclassical computers. However, NISQ devices experience significant errors which,\nif not corrected, can render physical quantities measured in these simulations\ninaccurate or meaningless. Here we describe a method of reducing these errors\nwhich is tailored to quantum algorithms for simulating fermionic systems. The\nmethod is based on executing quantum circuits in the model of fermionic linear\noptics, which are known to be efficiently simulable classically, to infer the\nrelationship between exact and noisy measurement outcomes, and hence undo the\neffect of noise. We validated our method by applying it to the VQE algorithm\nfor estimating ground state energies of instances of the Fermi-Hubbard model.\nIn classical numerical simulations of 12-qubit examples with physically\nrealistic levels of depolarising noise, errors were reduced by a factor of\naround 34 compared with the uncorrected case. Smaller experiments on quantum\nhardware demonstrate an average reduction in errors by a factor of 10 or more.\n"}
{"abstract": "  Ecology and evolution under changing environments are important in many\nsubfields of biology with implications for medicine. Here, we explore an\nexample: the consequences of fluctuating environments on the emergence of\nantibiotic resistance, which is an immense and growing problem. Typically, high\ndoses of antibiotics are employed to eliminate the infection quickly and\nminimize the time under which resistance may emerge. However, this strategy may\nnot be optimal. Since competition can reduce fitness and resistance typically\nhas a reproductive cost, resistant mutants' fitness can depend on their\nenvironment. Here we show conditions under which environmental varying fitness\ncan be exploited to prevent the emergence of resistance. We develop a\nstochastic Lotka-Volterra model of a microbial system with competing\nphenotypes: a wild strain susceptible to the antibiotic, and a mutant strain\nthat is resistant. We investigate the impact of various pulsed applications of\nantibiotics on population suppression. Leveraging competition, we show how a\nstrategy of environmental switching can suppress the infection while avoiding\nresistant mutants. We discuss limitations of the procedure depending on the\nmicrobe and pharmacodynamics and methods to ameliorate them.\n"}
{"abstract": "  We demonstrate that a periodic waveguide comprising of uniform lossless\nsegments together with discrete gain and radiating elements supports\nexceptional points of degeneracy (EPDs). We provide analytical expressions for\nall possible conditions that guarantee the occurrence of an EPD, i.e., the\ncoalescence of eigenvalues and eigenvectors. We show that EPDs are not only\nachieved using symmetric gain and radiation periodic loading, but they are also\nobtained using asymmetric gain and radiation loss conditions. We illustrate the\ncharacteristics of the degenerate electromagnetic modes, showing the dispersion\ndiagram and discussing the tunability of the EPD frequency. We show a special\ncondition, we refer to it as parity-time (PT)-glide symmetry, which leads to a\ndegeneracy that is occurring at all frequencies of operation. The class of EPDs\nproposed in this work is very promising for many applications that incorporate\ndiscrete-distributed coherent sources and radiation-loss elements; operating in\nthe vicinity of such special degeneracy conditions leads to potential\nperformance enhancement in a variety of microwave and optical resonators,\nantennas, and devices and can be extended to a new class of active integrated\nantenna arrays and radiating laser arrays.\n"}
{"abstract": "  Due to the significant importance of Big Data analysis, especially in\nbusiness-related topics such as improving services, finding potential\ncustomers, and selecting practical approaches to manage income and expenses,\nmany companies attempt to collaborate with scientists to find how, why, and\nwhat they should analysis. In this work, we would like to compare and discuss\ntwo different approaches that employed in business analysis topic in Big Data\nwith more consideration on how they utilized Spark. Both studies have\ninvestigated Churn Prediction as their case study for their proposed approaches\nsince it is an essential topic in business analysis for companies to recognize\na customer intends to leave or stop using their services. Here, we focus on\nApache Spark since it has provided several solutions to handle a massive amount\nof data in recent years efficiently. This feature in Spark makes it one of the\nmost robust candidate tools to upfront with a Big Data problem, particularly\ntime and resource are concerns.\n"}
{"abstract": "  Rural electrification demands the use of inexpensive technologies such as\nsingle wire earth return (SWER) networks. There is a steadily growing energy\ndemand from remote consumers, and the capacity of existing lines may become\ninadequate soon. Furthermore, high impedance arcing faults (HIF) from SWER\nlines can cause catastrophic bushfires such as the 2009 Black Saturday event.\nAs a solution, reliable remote electricity networks can be established through\nbreaking the existing systems down into microgrids, and existing SWER lines can\nbe utilised to interconnect those microgrids. The development of such reliable\nnetworks with better energy demand management will rely on having an integrated\nnetwork-wide condition monitoring system. As the first contribution of this\nthesis, a distributed online monitoring platform is developed that incorporates\npower quality monitoring, real-time HIF identification and transient\nclassification in SWER network. Artificial Intelligence (AI) based techniques\nare developed to classify faults and transients. The proposed approach\ndemonstrates higher HIF detection accuracy (98.67%) and reduced detection\nlatency (115.2 ms). Secondly, a remote consumer load identification methodology\nis developed to detect the load type from its transients. An edge\ncomputing-based architecture is proposed to facilitate the high-frequency\nanalysis for load identification. The proposed approach is evaluated in\nreal-time, and it achieves an average accuracy of 98% in identifying different\nloads. Finally, a deep neural network-based energy disaggregation framework is\ndeveloped to separate the load specific energy usage from an aggregated signal.\nThe proposed framework is evaluated using a real-world data set. It improves\nthe signal aggregate error by 44% and mean aggregate error by 19% in comparison\nwith the state-of-the-art techniques.\n"}
{"abstract": "  With the invention of the COVID-19 vaccine, shipping and distributing are\ncrucial in controlling the pandemic. In this paper, we build a mean-field\nvariational problem in a spatial domain, which controls the propagation of\npandemic by the optimal transportation strategy of vaccine distribution. Here\nwe integrate the vaccine distribution into the mean-field SIR model designed in\nour previous paper arXiv:2006.01249. Numerical examples demonstrate that the\nproposed model provides practical strategies in vaccine distribution on a\nspatial domain.\n"}
{"abstract": "  We use double-resonant two-photon laser spectroscopy to measure the dynamic\nscalar polarizability of the rubidium 5$D_{3/2}$ level,\n$\\alpha^{S}_{5D_{3/2}}$, at a wavelength of $\\lambda=1064$~nm. Since $\\lambda$\nis shorter than the photoionization (PI) limit of the Rb 5$D_{3/2}$ level,\n$\\alpha^{S}_{5D_{3/2}}$ depends on both bound-bound and bound-free transition\nmatrix elements. The level also undergoes significant broadening due to PI. The\n$1064$-nm field is applied in the form of a deep optical lattice\n($\\sim10^{5}$~photon recoils) generated by an in-vacuum field-enhancement\ncavity. In our spectroscopic method, we use known dynamic polarizabilities to\neliminate the need to measure the light intensity. Our method yields, in atomic\nunits, $\\alpha^{S}_{5D_{3/2}}=-524(17)$, in agreement with estimates.\nAdditionally, we extract the $5D_{3/2}$ photoionization cross section $\\sigma$\nat $1064$~nm from spectral linewidths; we find $\\sigma=44(1)$~Mb.\n"}
{"abstract": "  Out-of-distribution (OOD) detection is essential to prevent anomalous inputs\nfrom causing a model to fail during deployment. While improved OOD detection\nmethods have emerged, they often rely on the final layer outputs and require a\nfull feedforward pass for any given input. In this paper, we propose a novel\nframework, multi-level out-of-distribution detection MOOD, which exploits\nintermediate classifier outputs for dynamic and efficient OOD inference. We\nexplore and establish a direct relationship between the OOD data complexity and\noptimal exit level, and show that easy OOD examples can be effectively detected\nearly without propagating to deeper layers. At each exit, the OOD examples can\nbe distinguished through our proposed adjusted energy score, which is both\nempirically and theoretically suitable for networks with multiple classifiers.\nWe extensively evaluate MOOD across 10 OOD datasets spanning a wide range of\ncomplexities. Experiments demonstrate that MOOD achieves up to 71.05%\ncomputational reduction in inference, while maintaining competitive OOD\ndetection performance.\n"}
{"abstract": "  Preparation of nanoemulsions of triglyceride oils in water usually requires\nhigh mechanical energy and sophisticated equipment. Recently, we showed that\nalpha-to-beta (viz. gel-to-crystal) phase transition, observed with most lipid\nsubstances (triglycerides, diglycerides, phospholipids, alkanes, etc.), may\ncause spontaneous disintegration of micro-particles of these lipids, dispersed\nin aqueous solutions of appropriate surfactants, into nanometer particles/drops\nusing a simple cooling/heating cycle of the lipid dispersion (Cholakova et al.\nACS Nano 14 (2020) 8594). In the current study we show that this \"cold-burst\nprocess\" is observed also with natural oils of high practical interest, incl.\ncoconut oil, palm kernel oil and cocoa butter. Mean drop diameters of ca. 50 to\n100 nm were achieved with some of the studied oils. From the results of\ndedicated model experiments we conclude that intensive nano-fragmentation is\nobserved when the following requirements are met: (1) The three phase contact\nangle at the air-water-solid lipid interface is below ca. 30 degrees; (2) The\nequilibrium surface tension of the surfactant solution is below ca. 30 mN/m and\nthe dynamic surface tension decreases rapidly. (3) The surfactant solution\ncontains non-spherical surfactant micelles. e.g. ellipsoidal micelles or bigger\nsupramolecular aggregates; (4) The three phase contact angle measured at the\ncontact line (frozen oil-melted oil-surfactant solution) is also relatively\nlow. The mechanism(s) of the particle bursting process is revealed and, on this\nbasis, the role of all these factors is clarified and discussed. We explain all\nmain effects observed experimentally and define guiding principles for\noptimization of the cold-burst process in various, practically relevant\nlipid-surfactant systems.\n"}
{"abstract": "  We establish a Kantorovich duality for the pseudometric $\\mathcal{E}_\\hbar$\nintroduced in [F. Golse, T. Paul, Arch. Rational Mech. Anal. 223 (2017),\n57--94], obtained from the usual Monge-Kantorovich distance $d_{MK,2}$ between\nclassical densities by quantization of one of the two densities involved. We\nshow several type of inequalities comparing $d_{MK,2}$, $\\mathcal{E}_\\hbar$ and\n$MK_\\hbar$, a full quantum analogue of $d_{MK,2}$ introduced in [F. Golse, C.\nMouhot, T. Paul, Commun. Math. Phys. 343 (2016), 165--205], including an up to\n$\\hbar$ triangle inequality for $MK_\\hbar$. Finally, we show that, when nice\noptimal Kantorovich potentials exist for $\\mathcal{E}_\\hbar$, optimal couplings\ninduce classical/quantum optimal transports and the potentials are linked by a\nsemiquantum Legendre type transform.\n"}
{"abstract": "  Federated learning (FL) is a promising privacy-preserving distributed machine\nlearning methodology that allows multiple clients (i.e., workers) to\ncollaboratively train statistical models without disclosing private training\ndata. Due to the characteristics of data remaining localized and the\nuninspected on-device training process, there may exist Byzantine workers\nlaunching data poisoning and model poisoning attacks, which would seriously\ndeteriorate model performance or prevent the model from convergence. Most of\nthe existing Byzantine-robust FL schemes are either ineffective against several\nadvanced poisoning attacks or need to centralize a public validation dataset,\nwhich is intractable in FL. Moreover, to the best of our knowledge, none of the\nexisting Byzantine-robust distributed learning methods could well exert its\npower in Non-Independent and Identically distributed (Non-IID) data among\nclients. To address these issues, we propose FedCom, a novel Byzantine-robust\nfederated learning framework by incorporating the idea of commitment from\ncryptography, which could achieve both data poisoning and model poisoning\ntolerant FL under practical Non-IID data partitions. Specifically, in FedCom,\neach client is first required to make a commitment to its local training data\ndistribution. Then, we identify poisoned datasets by comparing the Wasserstein\ndistance among commitments submitted by different clients. Furthermore, we\ndistinguish abnormal local model updates from benign ones by testing each local\nmodel's behavior on its corresponding data commitment. We conduct an extensive\nperformance evaluation of FedCom. The results demonstrate its effectiveness and\nsuperior performance compared to the state-of-the-art Byzantine-robust schemes\nin defending against typical data poisoning and model poisoning attacks under\npractical Non-IID data distributions.\n"}
{"abstract": "  We apply on-shell amplitude techniques to the study of neutrino oscillations\nin vacuum, focussing on processes involving W bosons. We start by determining\nthe 3-point amplitude involving one neutrino, one charged lepton and one W\nboson, highlighting all the allowed kinematic structures. The result we obtain\ncontains terms generated at all orders in an expansion in the cutoff scale of\nthe theory, and we explicitly determine the lower dimensional operators behind\nthe generation of the different structures. We then use this amplitude to\nconstruct the 4-point amplitude in which neutrinos are exchanged in the\ns-channel, giving rise to oscillations. We also study in detail how flavor\nenters in the amplitudes, and how the PMNS matrix emerges from the on-shell\nperspective.\n"}
{"abstract": "  By using the dry etching process of tantalum (Ta) film, we had obtained\ntransmon qubit with the best lifetime (T1) 503 us, suggesting that the dry\netching process can be adopted in the following multi-qubit fabrication with Ta\nfilm. We also compared the relaxation and coherence times of transmons made\nwith different materials (Ta, Nb and Al) with the same design and fabrication\nprocesses of Josephson junction, we found that samples prepared with Ta film\nhad the best performance, followed by those with Al film and Nb film. We\ninferred that the reason for this difference was due to the different loss of\noxide materials located at the metal-air interface.\n"}
{"abstract": "  Using logic gates is the traditional way of designing logic circuits.\nHowever, most of the minimization algorithms concern a limited set of gates\n(complete sets), like sum of products, exclusive-or sum of products, NAND\ngates, NOR gates e.t.c.. In this paper, a method is proposed for minimizing\nmulti-output Boolean functions using any kind of two-input gates although it\ncan easily be extended to multi-input gates. The method is based on non-linear\nmixed integer programming. The experimental results show that the method gives\nthe same or better results compared to other methods available in the\nliterature. However, other methods do not ensure that they produce the minimal\nsolution, while the main advantages of the proposed method are that it does\nguarantee minimality and it can also handle Boolean functions for incompletely\nspecified functions. The method is general enough and can easily be extended to\nmore complicated design modules than just basic gates.\n"}
{"abstract": "  Structured jets are recently invoked to explain the complex emission of gamma\nray bursts, such as GW 170817. Based on the accretion simulations, the jets are\nexpected to have a structure that is more complex than a simple top-hat. Also,\nthe structure of launching regions of blazar jets should influence their large\nscale evolution. This is recently revealed by the interactions of jet\ncomponents in TXS 0506+056, where the jet is observed at a viewing angle close\nto zero. Observational studies have also shown an anti-correlation between the\njet variability, measured e.g. by its minimum variability time scale, and the\nLorentz factor, that spans several orders of magnitude and covers both blazars\nand GRBs samples. Motivated by those observational properties of black hole\nsources, we investigate the accretion inflow and outflow properties, by means\nof numerical GR MHD simulations. We perform axisymmetric calculations of the\nstructure and evolution of central engine, composed of magnetized torus around\nKerr black hole that is launching a non-uniform jet. We probe the jet\nenergetics at different points along the line of sight, and we measure the jet\ntime variability as localized in these specific regions. We quantify our\nresults by computing the minimum variability timescales and power density\nspectra. We reproduce the MTS-$\\Gamma$ correlation and we attribute it to the\nblack hole spin as the main driving parameter of the engine. We also find that\nthe PDS slope is not strongly affected by the black hole spin, while it differs\nfor various viewing angles.\n"}
{"abstract": "  In dynamic and cramped industrial environments, achieving reliable Visual\nTeach and Repeat (VT&R) with a single-camera is challenging. In this work, we\ndevelop a robust method for non-synchronized multi-camera VT&R. Our\ncontribution are expected Camera Performance Models (CPM) which evaluate the\ncamera streams from the teach step to determine the most informative one for\nlocalization during the repeat step. By actively selecting the most suitable\ncamera for localization, we are able to successfully complete missions when one\nof the cameras is occluded, faces into feature poor locations or if the\nenvironment has changed. Furthermore, we explore the specific challenges of\nachieving VT&R on a dynamic quadruped robot, ANYmal. The camera does not follow\na linear path (due to the walking gait and holonomicity) such that precise\npath-following cannot be achieved. Our experiments feature forward and backward\nfacing stereo cameras showing VT&R performance in cluttered indoor and outdoor\nscenarios. We compared the trajectories the robot executed during the repeat\nsteps demonstrating typical tracking precision of less than 10cm on average.\nWith a view towards omni-directional localization, we show how the approach\ngeneralizes to four cameras in simulation. Video: https://youtu.be/iAY0lyjAnqY\n"}
{"abstract": "  Dynamic graphs are rife with higher-order interactions, such as co-authorship\nrelationships and protein-protein interactions in biological networks, that\nnaturally arise between more than two nodes at once. In spite of the ubiquitous\npresence of such higher-order interactions, limited attention has been paid to\nthe higher-order counterpart of the popular pairwise link prediction problem.\nExisting higher-order structure prediction methods are mostly based on\nheuristic feature extraction procedures, which work well in practice but lack\ntheoretical guarantees. Such heuristics are primarily focused on predicting\nlinks in a static snapshot of the graph. Moreover, these heuristic-based\nmethods fail to effectively utilize and benefit from the knowledge of latent\nsubstructures already present within the higher-order structures. In this\npaper, we overcome these obstacles by capturing higher-order interactions\nsuccinctly as \\textit{simplices}, model their neighborhood by face-vectors, and\ndevelop a nonparametric kernel estimator for simplices that views the evolving\ngraph from the perspective of a time process (i.e., a sequence of graph\nsnapshots). Our method substantially outperforms several baseline higher-order\nprediction methods. As a theoretical achievement, we prove the consistency and\nasymptotic normality in terms of the Wasserstein distance of our estimator\nusing Stein's method.\n"}
{"abstract": "  Recent works on sparse neural networks have demonstrated the possibility to\ntrain a sparse subnetwork independently from scratch, to match the performance\nof its corresponding dense network. However, identifying such sparse\nsubnetworks (winning tickets) either involves a costly iterative\ntrain-prune-retrain process (e.g., Lottery Ticket Hypothesis) or an\nover-extended training time (e.g., Dynamic Sparse Training). In this work, we\ndraw a unique connection between sparse neural network training and the deep\nensembling technique, yielding a novel ensemble learning framework called\nFreeTickets. Instead of starting from a dense network, FreeTickets randomly\ninitializes a sparse subnetwork and then trains the subnetwork while\ndynamically adjusting its sparse mask, resulting in many diverse sparse\nsubnetworks throughout the training process. FreeTickets is defined as the\nensemble of these sparse subnetworks freely obtained during this one-pass,\nsparse-to-sparse training, which uses only a fraction of the computational\nresources required by the vanilla dense training. Moreover, despite being an\nensemble of models, FreeTickets has even fewer parameters and training FLOPs\ncompared to a single dense model: this seemingly counter-intuitive outcome is\ndue to the high sparsity of each subnetwork. FreeTickets is observed to\ndemonstrate a significant all-round improvement compared to standard dense\nbaselines, in prediction accuracy, uncertainty estimation, robustness, and\nefficiency. FreeTickets easily outperforms the naive deep ensemble with\nResNet50 on ImageNet using only a quarter of the training FLOPs required by the\nlatter. Our results provide insights into the strength of sparse neural\nnetworks and suggest that the benefits of sparsity go way beyond the usually\nexpected inference efficiency.\n"}
{"abstract": "  In part 1 of this paper some linear weighted generalized Fibonacci number\nsummation identities were derived using the fact that the Fibonacci number is\nthe residue of a rational function. In this part, using the same method, some\nquadratic and cubic weighted generalized Fibonacci number summation identities\nare derived, including some infinite series and generating functions. In\naddition some quadratic and cubic weighted generalized Fibonacci number\nsummation identities with binomial coefficients, and some cubic and quartic\nFibonacci and Lucas number summation identities are derived.\n"}
{"abstract": "  Language implementation frameworks, e.g., RPython and Truffle/Graal, are\npractical tools for creating efficient virtual machines, including a\nwell-functioning just-in-time (JIT) compiler. It is demanding to support\nmulti-tier JIT compilation in such a framework for language developers. This\npaper presents an idea to generate threaded code by reusing an existing\nmeta-tracing JIT compiler, as well as an interpreter design for it. Our\napproach does not largely modify the RPython itself but constructs an effective\ninterpreter definition to enable threaded code generation in RPython. We expect\nour system to be extended to support multi-level JIT compilation in the RPython\nframework. We measured the potential performance of our threaded code\ngeneration by simulating its behavior in PyPy. We confirmed that our approach\nreduced code sizes by 80 \\% and compilation times by 60 \\% compared to PyPy's\nJIT compiler on average, and ran about 7 \\% faster than the interpreter-only\nexecution.\n"}
{"abstract": "  The Q-Star structures, also called as gray holes with a radius of two times\ngreater than the corresponding Schwarzschild radius of same mass and from the\ncompact neutron star family with a very high density, have been obtained by\nusing a generalization of $q$-deformed fermions with two parameters $q$ and $p$\nas the constituents of the star. Because the interaction between particles are\ncontrolled by deformation parameters in $q,p$-fermions instead of a complicated\ninteraction Lagrangian, we consider the deformation parameters giving the\nmaximum Q-Star pressure between the interacting particles of stars. The cold\nand hot Q-Stars in temperatures of $T=0$, $T=40\\ \\mathrm{MeV}$ and $T=60\\\n\\mathrm{MeV}$ have been investigated from the numerical solutions of TOV\nequations and it is obtained that the Q-Stars can reach up to $13$ solar mass\nand $75$ Km radius. Moreover, it is found that the total mass and radius of the\nstar increase when the star gets cool down, and a much denser state is reached.\n"}
{"abstract": "  Einstein's most famous equation -- $E=mc^2$ -- generated a short-circuit\nbetween the concepts of mass and energy, which also affects other concepts like\nmatter, radiation, and vacuum. Physics currently has a mixture of classical,\nrelativistic, and quantum concepts of mass, which generates a great deal of\nconfusion and many problems. Clear definitions need to be established if one\nwants to avoid ghost hunting. In particular, by abandoning the idea of mass and\nfocusing on time and energy, some interesting implications are emerging. It is\nnoted that the cutoff frequency of the quantum vacuum energy, consistent with\nthe observations of the cosmological constant, corresponds to that of the\nCosmic Microwave Background. This might be consistent with the hypothesis of a\nrotating and expanding universe described by a Kerr-de Sitter metric with the\nobserved cosmological constant. To verify this hypothesis is crucial to prove\nthe rotation of the universe.\n"}
{"abstract": "  In a quantum-noise limited system, weak-value amplification using\npost-selection normally does not produce more sensitive measurements than\nstandard methods for ideal detectors: the increased weak value is compensated\nby the reduced power due to the small post-selection probability. Here we\nexperimentally demonstrate recycled weak-value measurements using a pulsed\nlight source and optical switch to enable nearly deterministic weak-value\namplification of a mirror tilt. Using photon counting detectors, we demonstrate\na signal improvement by a factor of $4.4 \\pm 0.2$ and a signal-to-noise ratio\nimprovement of $2.10 \\pm 0.06$, compared to a single-pass weak-value\nexperiment, and also compared to a conventional direct measurement of the tilt.\nThe signal-to-noise ratio improvement could reach around 6 for the parameters\nof this experiment, assuming lower loss elements.\n"}
{"abstract": "  The massive spread of false information on social media has become a global\nrisk especially in a global pandemic situation like COVID-19. False information\ndetection has thus become a surging research topic in recent months.\nNLP4IF-2021 shared task on fighting the COVID-19 infodemic has been organised\nto strengthen the research in false information detection where the\nparticipants are asked to predict seven different binary labels regarding false\ninformation in a tweet. The shared task has been organised in three languages;\nArabic, Bulgarian and English. In this paper, we present our approach to tackle\nthe task objective using transformers. Overall, our approach achieves a 0.707\nmean F1 score in Arabic, 0.578 mean F1 score in Bulgarian and 0.864 mean F1\nscore in English ranking 4th place in all the languages.\n"}
{"abstract": "  We analyse three time integration schemes for unfitted methods in fluid\nstructure interaction. In Alghorithm 1 we propose a fully discrete monolithic\nalgorithm with P1 P1 stabilized finite elements for the fluid problem; for this\nalghorithm we prove well-posedness, unconditional stability and convergence in\nthe case of linearized problem (see Propositions 2.4.2, 2.4.3 and Theorem\n3.3.7, respectively). The analysis optimal convergence rates as expected from\nthe Euler scheme, and the supposed regularity of the solution to the continuous\nproblem. Moreover we introduce two algorithms that allow for a partitioning of\nthe coupled problem by exploiting an explicit-implicit treatment of the\ntransmission conditions. Algorithm 2 represents, essentially, a simplification\nof Algorithm 1 since it simply treat the solid elastic forces in explicit form\nusing the displacement and velocities of the structure evaluated in the\nprevious time steps. Instead, Algorithm 3, is really a splitting algorithm that\ninvolves the solution of two staggered problems. It splits the forces that\nsolid transfers to fluid in two contributions: the inertial contribution that\nis treated in implicit form and the elastic contribution that is treated in\nexplicit form. We perform the stability analysis for both the schemes in\nTheorems 4.3.1 and 4.3.3. Algorithm 2 results conditionally stable for all the\nextrapolations considered, instead Algorithm 3 is unconditionally stable, for\nextrapolations of order zero and one, and conditionally stable for the\nextrapolation of order two. Since Algorithm 3 is the most promising, we perform\nthe convergence analysis in the linearized case (see Theorem 4.4.2) obtaining\nresults in line with those of the monolithic case, in particular the splitting\nintroduced preserves optimal conevegence rates.\n"}
{"abstract": "  Traction adaptive motion planning and control has potential to improve an an\nautomated vehicle's ability to avoid accident in a critical situation. However,\nsuch functionality require an accurate friction estimate for the road ahead of\nthe vehicle that is updated in real time. Current state of the art friction\nestimation techniques include high accuracy local friction estimation in the\npresence of tire slip, as well as rough classification of the road surface\nahead of the vehicle, based on forward looking camera. In this paper we show\nthat neither of these techniques in isolation yield satisfactory behavior when\ndeployed with traction adaptive motion planning and control functionality.\nHowever, fusion of the two provides sufficient accuracy, availability and\nforesight to yield near optimal behavior. To this end, we propose a fusion\nmethod based on heteroscedastic gaussian process regression, and present\ninitial simulation based results.\n"}
{"abstract": "  New and unique opportunities now exist to look for technosignatures (TS)\nbeyond traditional SETI radio searches, motivated by tremendous advances in\nexoplanet science and observing capabilities in recent years. Space agencies,\nboth public and private, may be particularly interested in learning about the\ncommunity's views as to the optimal methods for future TS searches with current\nor forthcoming technology. This report is an effort in that direction. We put\nforward a set of possible mission concepts designed to search for TS, although\nthe data supplied by such missions would also benefit other areas of\nastrophysics. We introduce a novel framework to analyze a broad diversity of TS\nin a quantitative manner. This framework is based on the concept of ichnoscale,\nwhich is a new parameter related to the scale of a TS cosmic footprint,\ntogether with the number of potential targets where such TS can be searched\nfor, and whether or not it is continuous in time.\n"}
{"abstract": "  Materials with strong $\\chi^{(2)}$ optical nonlinearity, especially lithium\nniobate, play a critical role in building optical parametric oscillators\n(OPOs). However, chip-scale integration of low-loss $\\chi^{(2)}$ materials\nremains challenging and limits the threshold power of on-chip $\\chi^{(2)}$ OPO.\nHere we report the first on-chip lithium niobate optical parametric oscillator\nat the telecom wavelengths using a quasi-phase matched, high-quality microring\nresonator, whose threshold power ($\\sim$30 $\\mu$W) is 400 times lower than that\nin previous $\\chi^{(2)}$ integrated photonics platforms. An on-chip power\nconversion efficiency of 11% is obtained at a pump power of 93 $\\mu$W. The OPO\nwavelength tuning is achieved by varying the pump frequency and chip\ntemperature. With the lowest power threshold among all on-chip OPOs\ndemonstrated so far, as well as advantages including high conversion\nefficiency, flexibility in quasi-phase matching and device scalability, the\nthin-film lithium niobate OPO opens new opportunities for chip-based tunable\nclassical and quantum light sources and provides an potential platform for\nrealizing photonic neural networks.\n"}
{"abstract": "  In this paper, to improve wear resistance of components such as screws under\nsevere friction-wear, Fe-Cr-C-Mo-V-Ti-N hardfacing coatings were further\ndeveloped. The hardfacing coatings were acquired by shielded manual arc welding\n(SMAW) method. The ferroalloys added into the coating flux of the hardfaced\nelectrode were jointly nitrided. The microstructure of the coatings was carried\nout using X-ray diffraction(XRD), optical microscope(OM), field emission\nscanning electron microscope (FESEM) and energy dispersive Xray spectrometry\n(EDS). In addition, FactSage 7.0 software was employed to calculate the\nequilibrium phase diagram of the hardfacings. The wear resistance was performed\non a pin-on-disc machine. The Fe-Cr-C- Mo-V-Ti-N hardfacings exhibited higher\nwear resistance than cladding layer without nitrides.\n"}
{"abstract": "  This demo paper presents the first system for playing the popular Angry Birds\ngame using a domain-independent planner. Our system models Angry Birds levels\nusing PDDL+, a planning language for mixed discrete/continuous domains. It uses\na domain-independent PDDL+ planner to generate plans and executes them. In this\ndemo paper, we present the system's PDDL+ model for this domain, identify key\ndesign decisions that reduce the problem complexity, and compare the\nperformance of our system to model-specific methods for this domain. The\nresults show that our system's performance is on par with other domain-specific\nsystems for Angry Birds, suggesting the applicability of domain-independent\nplanning to this benchmark AI challenge.\n"}
{"abstract": "  Federated learning (FL) is a collaborative machine learning paradigm, which\nenables deep learning model training over a large volume of decentralized data\nresiding in mobile devices without accessing clients' private data. Driven by\nthe ever increasing demand for model training of mobile applications or\ndevices, a vast majority of FL tasks are implemented over wireless fading\nchannels. Due to the time-varying nature of wireless channels, however, random\ndelay occurs in both the uplink and downlink transmissions of FL. How to\nanalyze the overall time consumption of a wireless FL task, or more\nspecifically, a FL's delay distribution, becomes a challenging but important\nopen problem, especially for delay-sensitive model training. In this paper, we\npresent a unified framework to calculate the approximate delay distributions of\nFL over arbitrary fading channels. Specifically, saddle point approximation,\nextreme value theory (EVT), and large deviation theory (LDT) are jointly\nexploited to find the approximate delay distribution along with its tail\ndistribution, which characterizes the quality-of-service of a wireless FL\nsystem. Simulation results will demonstrate that our approximation method\nachieves a small approximation error, which vanishes with the increase of\ntraining accuracy.\n"}
{"abstract": "  People with Alzheimer's disease and related dementias (ADRD) often show the\nproblem of repetitive questioning, which brings a great burden on persons with\nADRD (PwDs) and their caregivers. Conversational robots hold promise of coping\nwith this problem and hence alleviating the burdens on caregivers. In this\npaper, we proposed a partially observable markov decision process (POMDP) model\nfor the PwD-robot interaction in the context of repetitive questioning, and\nused Q-learning to learn an adaptive conversation strategy (i.e., rate of\nfollow-up question and difficulty of follow-up question) towards PwDs with\ndifferent cognitive capabilities and different engagement levels. The results\nindicated that Q-learning was helpful for action selection for the robot. This\nmay be a useful step towards the application of conversational social robots to\ncope with repetitive questioning in PwDs.\n"}
{"abstract": "  In the paper we study what sets can be obtained as $\\alpha$-limit sets of\nbackward trajectories in graph maps. We show that in the case of mixing maps,\nall those $\\alpha$-limit sets are $\\omega$-limit sets and for all but finitely\nmany points $x$, we can obtain every $\\omega$-limits set as the $\\alpha$-limit\nset of a backward trajectory starting in $x$. For zero entropy maps, every\n$\\alpha$-limit set of a backward trajectory is a minimal set. In the case of\nmaps with positive entropy, we obtain a partial characterization which is very\nclose to complete picture of the possible situations.\n"}
{"abstract": "  Steering tokamak plasmas is commonly viewed as a way to avoid disruptions and\nrunaway electrons. Plasma steering sounds as safe as driving to work but will\nbe shown to more closely resemble driving at high speed through a dense fog on\nan icy road. The long time required to terminate an ITER discharge compared to\ntime over which dangers can be foreseen is analogous to driving in a dense fog.\nThe difficulty of regaining plasma control if it is lost resembles driving on\nan icy road. Disruptions and runaways are associated with three issues -- a\nsolution to one tends to complicate the solution to the other two: loss of\nplasma position control, excessive heat deposition, and wall melting due to\nrunaway electrons. All three risks must be addressed for ITER to achieve its\nmission and essentially eliminated before tokamak power plants can be deployed.\n"}
{"abstract": "  Modern day computing increasingly relies on specialization to satiate growing\nperformance and efficiency requirements. A core challenge in designing such\nspecialized hardware architectures is how to perform mapping space search,\ni.e., search for an optimal mapping from algorithm to hardware. Prior work\nshows that choosing an inefficient mapping can lead to multiplicative-factor\nefficiency overheads. Additionally, the search space is not only large but also\nnon-convex and non-smooth, precluding advanced search techniques. As a result,\nprevious works are forced to implement mapping space search using expert\nchoices or sub-optimal search heuristics.\n  This work proposes Mind Mappings, a novel gradient-based search method for\nalgorithm-accelerator mapping space search. The key idea is to derive a smooth,\ndifferentiable approximation to the otherwise non-smooth, non-convex search\nspace. With a smooth, differentiable approximation, we can leverage efficient\ngradient-based search algorithms to find high-quality mappings. We extensively\ncompare Mind Mappings to black-box optimization schemes used in prior work.\nWhen tasked to find mappings for two important workloads (CNN and MTTKRP), the\nproposed search finds mappings that achieve an average $1.40\\times$,\n$1.76\\times$, and $1.29\\times$ (when run for a fixed number of steps) and\n$3.16\\times$, $4.19\\times$, and $2.90\\times$ (when run for a fixed amount of\ntime) better energy-delay product (EDP) relative to Simulated Annealing,\nGenetic Algorithms and Reinforcement Learning, respectively. Meanwhile, Mind\nMappings returns mappings with only $5.32\\times$ higher EDP than a possibly\nunachievable theoretical lower-bound, indicating proximity to the global\noptima.\n"}
{"abstract": "  We demonstrate a passive down-conversion imaging system that converts\nbroadband ultraviolet light to narrow-band green light while preserving the\ndirectionality of rays, and thus enabling direct down-conversion imaging. At\nthe same time our system has high transparency in the visible, enabling\nsuperimposed visible and ultraviolet imaging. The frequency conversion is\nperformed by a subwavelength-thickness transparent downconverter based on\nhighly efficient CsPbBr3 nanocrystals incorporated into the focal plane of a\nsimple telescope or relay-lens geometry. The resulting imaging performance of\nthis down-conversion system approaches the diffraction limit. This\ndemonstration sets the stage for the incorporation of other high-efficiency\nperovskite nanocrystal materials to enable passive multi-frequency conversion\nimaging systems.\n"}
{"abstract": "  In this work we give a complete picture of how to in a direct simple way\ndefine the mass at null infinity in harmonic coordinates in three different\nways that we show satisfy the Bondi mass loss law. The first and second way\ninvolve only the limit of metric (Trautman mass) respectively the null second\nfundamental forms along asymptotically characteristic surfaces (asymptotic\nHawking mass) that only depend on the ADM mass. The last in an original way\ninvolves construction of special characteristic coordinates at null infinity\n(Bondi mass). The results here rely on asymptotics of the metric derived in\n[24].\n"}
{"abstract": "  The heaviest N=Z doubly-magic nucleus, $^{100}$Sn, and the neighboring nuclei\noffer unique opportunities to investigate the properties of nuclear interaction\nin extreme conditions. In particular, the Cd isotopes are expected to present\nfeatures similar to those found in the Sn isotopic chain, since they have only\ntwo proton holes in the Z=50 shell. In this manuscript, the lifetime\nmeasurements of low-lying states in the even-mass $^{102-108}$Cd is presented.\nThanks to the powerful detection capabilities of AGATA array and VAMOS++\nspectrometer, the unusual employment of multi-nucleon transfer reactions\npermitted to investigate the first 2$^+$ and 4$^+$ states in all these nuclei,\ntogether with various deformed bands in $^{106}$Cd. The results were\ninterpreted in the context of new state-of-the-art beyond-mean-field\ncalculations, using the symmetry-conserving configuration-mixing approach.\nDespite the similarities in the electromagnetic properties of the low-lying\nstates, there is a fundamental structural difference between the ground-state\nbands in the Z=48 and Z=50 isotopes. The comparison between experimental and\ntheoretical results revealed a rotational character of the Cd nuclei, which\nhave prolate-deformed ground states with $\\beta_2 \\approx 0.2$. At this\ndeformation Z=48 becomes a closed-shell configuration, which is favored with\nrespect to the spherical one.\n"}
{"abstract": "  In this paper, we consider a (nonlinear) transformation $\\Phi$ of invertible\npositive elements in $C^*$-algebras which preserves the norm of any of the\nthree fundamental means of positive elements; namely, $\\|\\Phi(A)\\mm \\Phi(B)\\| =\n\\|A\\mm B\\|$, where $\\mm$ stands for the arithmetic mean $A\\nabla B=(A+B)/2$,\nthe geometric mean $A\\#B=A^{1/2}(A^{-1/2}BA^{-1/2})^{1/2}A^{1/2}$, or the\nharmonic mean $A!B=2(A^{-1} + B^{-1})^{-1}$. Assuming that $\\Phi$ is surjective\nand preserves either the norm of the arithmetic mean or the norm of the\ngeometric mean, we show that $\\Phi$ extends to a Jordan $*$-isomorphism between\nthe underlying full algebras. If $\\Phi$ is surjective and preserves the norm of\nthe harmonic mean, then we obtain the same conclusion in the special cases\nwhere the underlying algebras are $AW^*$-algebras or commutative\n$C^*$-algebras.\n  In the commutative case, for a transformation $T: F(\\mathrm{X})\\subset\nC_0(\\mathrm{X})_+\\rightarrow C_0(\\mathrm{Y})_+$, we can relax the surjectivity\nassumption and show that $T$ is a generalized composition operator if $T$\npreserves the norm of the (arithmetic, geometric, harmonic, or in general any\npower) mean of any finite collection of positive functions, provided that the\ndomain $F(\\mathrm{X})$ contains sufficiently many elements to peak on compact\n$G_\\delta$ sets. When the image $T(F(\\mathrm{X}))$ also contains sufficiently\nmany elements to peak on compact $G_\\delta$ sets, $T$ extends to an algebra\n$*$-isomorphism between the underlying full function algebras.\n"}
{"abstract": "  The $k$-gonal models of random groups are defined as the quotients of free\ngroups on $n$ generators by cyclically reduced words of length $k$. As $k$\ntends to infinity, this model approaches the Gromov density model. In this\npaper we show that for any fixed $d_0 \\in (0, 1)$, if positive $k$-gonal random\ngroups satisfy Property (T) with overwhelming probability for densities $d\n>d_0$, then so do $nk$-gonal random groups, for any $n \\in \\mathbb{N}$. In\nparticular, this shows that for densities above 1/3, groups in $3k$-gonal\nmodels satisfy Property (T) with probability 1 as $n$ approaches infinity.\n"}
{"abstract": "  A data-centric approach with Natural Language Processing (NLP) to predict\npersonality types based on the MBTI (an introspective self-assessment\nquestionnaire that indicates different psychological preferences about how\npeople perceive the world and make decisions) through systematic enrichment of\ntext representation, based on the domain of the area, under the generation of\nfeatures based on three types of analysis: sentimental, grammatical and\naspects. The experimentation had a robust baseline of stacked models, with\npremature optimization of hyperparameters through grid search, with gradual\nfeedback, for each of the four classifiers (dichotomies) of MBTI. The results\nshowed that attention to the data iteration loop focused on quality,\nexplanatory power and representativeness for the abstraction of more\nrelevant/important resources for the studied phenomenon made it possible to\nimprove the evaluation metrics results more quickly and less costly than\ncomplex models such as the LSTM or state of the art ones as BERT, as well as\nthe importance of these results by comparisons made from various perspectives.\nIn addition, the study demonstrated a broad spectrum for the evolution and\ndeepening of the task and possible approaches for a greater extension of the\nabstraction of personality types.\n"}
{"abstract": "  We study the concomitant breaking of spatial translations and dilatations in\nGinzburg-Landau-like models, where the dynamics responsible for the symmetry\nbreaking is described by an effective Mexican hat potential for spatial\ngradients. We show that there are fractonic modes with either subdimensional\npropagation or no propagation altogether, namely, immobility. Such class of\neffective field theories encompasses instances of helical superfluids and\nmeta-fluids, where fractons can be connected to an emergent symmetry under\nhigher moment charges, leading in turns to the trivialization of some elastic\ncoefficients. The introduction of a finite charge density alters the mobility\nproperties of fractons and leads to a competition between the chemical\npotential and the superfluid velocity in determining the gap of the dilaton.\nThe mobility of fractons can also be altered at zero density upon considering\nadditional higher-derivative terms.\n"}
{"abstract": "  Employing high-resolution EUV imaging observations from SDO/AIA, we analyse a\ncompressive plasma oscillation in a hot coronal loop triggered by a C-class\nflare near one of its foot points as first studied by Kumar et al. We\ninvestigate the oscillation properties in both the 131{\\,}{\\AA} and 94{\\,}{\\AA}\nchannels and find that what appears as a pure sloshing oscillation in the\n131{\\,}{\\AA} channel actually transforms into a standing wave in the\n94{\\,}{\\AA} channel at a later time. This is the first clear evidence of such\ntransformation confirming the results of a recent numerical study which\nsuggests that these two oscillations are not independent phenomena. We\nintroduce a new analytical expression to properly fit the sloshing phase of an\noscillation and extract the oscillation properties. For the AIA 131{\\,}{\\AA}\nchannel, the obtained oscillation period and damping time are 608$\\pm$4{\\,}s\nand 431$\\pm$20{\\,}s, respectively during the sloshing phase. The corresponding\nvalues for the AIA 94{\\,}{\\AA} channel are 617$\\pm$3{\\,}s and 828$\\pm$50{\\,}s.\nDuring the standing phase that is observed only in the AIA 94{\\,}{\\AA} channel,\nthe oscillation period and damping time have increased to 791$\\pm$5{\\,}s and\n1598$\\pm$138{\\,}s, respectively. The plasma temperature obtained from the DEM\nanalysis indicates substantial cooling of the plasma during the oscillation.\nConsidering this, we show that the observed oscillation properties and the\nassociated changes are compatible with damping due to thermal conduction. We\nfurther demonstrate that the absence of a standing phase in the 131{\\,}{\\AA}\nchannel is a consequence of cooling plasma besides the faster decay of\noscillation in this channel.\n"}
{"abstract": "  In this paper we consider traces at initial times for functions with mixed\ntime-space smoothness. Such results are often needed in the theory of evolution\nequations. Our result extends and unifies many previous results. Our main\nimprovement is that we can allow general interpolation couples. The abstract\nresults are applied to regularity problems for fractional evolution equations\nand stochastic evolution equation, where uniform trace estimates on the\nhalf-line are shown.\n"}
{"abstract": "  Differentially private (DP) machine learning allows us to train models on\nprivate data while limiting data leakage. DP formalizes this data leakage\nthrough a cryptographic game, where an adversary must predict if a model was\ntrained on a dataset D, or a dataset D' that differs in just one example.If\nobserving the training algorithm does not meaningfully increase the adversary's\nodds of successfully guessing which dataset the model was trained on, then the\nalgorithm is said to be differentially private. Hence, the purpose of privacy\nanalysis is to upper bound the probability that any adversary could\nsuccessfully guess which dataset the model was trained on.In our paper, we\ninstantiate this hypothetical adversary in order to establish lower bounds on\nthe probability that this distinguishing game can be won. We use this adversary\nto evaluate the importance of the adversary capabilities allowed in the privacy\nanalysis of DP training algorithms.For DP-SGD, the most common method for\ntraining neural networks with differential privacy, our lower bounds are tight\nand match the theoretical upper bound. This implies that in order to prove\nbetter upper bounds, it will be necessary to make use of additional\nassumptions. Fortunately, we find that our attacks are significantly weaker\nwhen additional (realistic)restrictions are put in place on the adversary's\ncapabilities.Thus, in the practical setting common to many real-world\ndeployments, there is a gap between our lower bounds and the upper bounds\nprovided by the analysis: differential privacy is conservative and adversaries\nmay not be able to leak as much information as suggested by the theoretical\nbound.\n"}
{"abstract": "  A robot working in human-centric environments needs to know which kind of\nobjects exist in the scene, where they are, and how to grasp and manipulate\nvarious objects in different situations to help humans in everyday tasks.\nTherefore, object recognition and grasping are two key functionalities for such\nrobots. Most state-of-the-art tackles object recognition and grasping as two\nseparate problems while both use visual input. Furthermore, the knowledge of\nthe robot is fixed after the training phase. In such cases, if the robot faces\nnew object categories, it must retrain from scratch to incorporate new\ninformation without catastrophic interference. To address this problem, we\npropose a deep learning architecture with augmented memory capacities to handle\nopen-ended object recognition and grasping simultaneously. In particular, our\napproach takes multi-views of an object as input and jointly estimates\npixel-wise grasp configuration as well as a deep scale- and rotation-invariant\nrepresentation as outputs. The obtained representation is then used for\nopen-ended object recognition through a meta-active learning technique. We\ndemonstrate the ability of our approach to grasp never-seen-before objects and\nto rapidly learn new object categories using very few examples on-site in both\nsimulation and real-world settings.\n"}
{"abstract": "  In this paper we construct distance-regular graphs admitting a transitive\naction of the five sporadic simple groups discovered by E. Mathieu, the Mathieu\ngroups $M_{11}$, $M_{12}$, $M_{22}$, $M_{23}$ and $M_{24}$. From the code\nspanned by the adjacency matrix of the strongly regular graph with parameters\n(176,70,18,34) we obtain block designs having the full automorphism groups\nisomorphic to the Higman-Sims finite simple group. Further, we discuss a\npossibility of permutation decoding of the codes spanned by the adjacency\nmatrices of the graphs constructed and find small PD-sets for some of the\ncodes.\n"}
{"abstract": "  In hot dense plasmas of intermediate or high-Z elements in the state of local\nthermodynamic equilibrium, the number of electronic configurations contributing\nto key macroscopic quantities such as the spectral opacity and equation of\nstate, can be enormous. In this work we present systematic methods for the\nanalysis of the number of relativistic electronic configurations in a plasma.\nWhile the combinatoric number of configurations can be huge even for mid-Z\nelements, the number of configurations which have non negligible population is\nmuch lower and depends strongly and non-trivially on temperature and density.\nWe discuss two useful methods for the estimation of the number of populated\nconfigurations: (i) using an exact calculation of the total combinatoric number\nof configurations within superconfigurations in a converged\nsuper-transition-array (STA) calculation, and (ii) by using an estimate for the\nmultidimensional width of the probability distribution for electronic\npopulation over bound shells, which is binomial if electron exchange and\ncorrelation effects are neglected. These methods are analyzed, and the\nmechanism which leads to the huge number of populated configurations is\ndiscussed in detail. Comprehensive average atom finite temperature density\nfunctional theory (DFT) calculations are performed in a wide range of\ntemperature and density for several low, mid and high Z plasmas. The effects of\ntemperature and density on the number of populated configurations are discussed\nand explained.\n"}
{"abstract": "  We introduce a minimalist outbreak forecasting model that combines\ndata-driven parameter estimation with variational data assimilation. By\nfocusing on the fundamental components of nonlinear disease transmission and\nrepresenting data in a domain where model stochasticity simplifies into a\nprocess with independent increments, we design an approach that only requires\nfour core parameters to be estimated. We illustrate this novel methodology on\nCOVID-19 forecasts. Results include case count and deaths predictions for the\nUS and all of its 50 states, the District of Columbia, and Puerto Rico. The\nmethod is computationally efficient and is not disease- or location-specific.\nIt may therefore be applied to other outbreaks or other countries, provided\ncase counts and/or deaths data are available.\n"}
{"abstract": "  Moir\\'e lattices which consist of parallel but staggered periodic lattices\nhave been extensively explored due to their salient physical properties, such\nas van Hove singularities[1, 2], commensurable incommensurable transitions[3],\nnon-Abelian gauge potentials[4], fractional quantum Hall effects[5-7], van der\nWaals interfaces[8, 9] and unconventional superconductivity[10, 11]. However,\nthere are limited demonstrations of such concepts for classical wave systems.\nHere, we realized gauge fields in one-dimensional Moir\\'e phononic lattices\nconsisting of two superimposed periodic patterns which mismatched with each\nother along one direction. Benefiting from gauge fields, we generated Landau\nlevel flat bands near the Dirac cone and experimentally measured their spatial\nlocalization in pressure-field distributions. Then, by mismatching lattices\nalong both directions, we constructed two-dimensional Moir\\'e phononic lattices\nwith van der Waals interfaces. We found that acoustic waves efficiently\ntransported along van der Waals interfaces behaving as metallic networks. As\nmismatched lattices are well-controllable, our study offers a novel path to\nmanipulate sound waves which are inaccessible in traditional periodic acoustic\nsystems, and can be easily extended to mechanics, optics, electromagnetics and\nelectronics.\n"}
{"abstract": "  Stiffness variation and real-time position feedback are critical for any\nrobotic system but most importantly for active and wearable devices to interact\nwith the user and environment. Currently, for compact sizes, there is a lack of\nsolutions bringing high-fidelity feedback and maintaining design and functional\nintegrity. In this work, we propose a novel minimal clutch with integrated\nstiffness variation and real-time position feedback whose performance surpasses\nconventional jamming solutions. We introduce integrated design, modeling, and\nverification of the clutch in detail. Preliminary experimental results show the\nchange in impedance force of the clutch is close to 24-fold at the maximum\nforce density of 15.64 N/cm2. We validated the clutch experimentally in (1)\nenhancing the bending stiffness of a soft actuator to increase a soft\nmanipulator's gripping force by 73%; (2) enabling a soft cylindrical actuator\nto execute omnidirectional movement; (3) providing real-time position feedback\nfor hand posture detection and impedance force for kinesthetic haptic feedback.\nThis manuscript presents the functional components with a focus on the\nintegrated design methodology, which will have an impact on the development of\nsoft robots and wearable devices.\n"}
{"abstract": "  The light curve of KIC 8462852, a.k.a Boyajian's Star, undergoes deep dips\nthe origin of which remains unclear. A faint star $\\approx$2\\arcsec to the east\nwas discovered in Keck/NIRC2 imaging in Boyajian et al. (2016), but its status\nas a binary, and possible contribution to the observed variability, was\nunclear. Here, we use three epochs of Keck/NIRC2 imaging, spanning five years,\nin JHK near-infrared bands to obtain 1-mas precision astrometry. We show that\nthe two objects exhibit common proper motion, measure a relative velocity of\n$\\mu=0.14\\pm0.44$ mas yr$^{-1}$ ($\\mu=0.30\\pm0.93$ km s$^{-1}$) and conclude\nthat they are a binary pair at $880\\pm10$ AU projected separation. There is\nmarginal detection of possible orbital motion, but our astrometry is\ninsufficient to characterize the orbit. We show that two other point sources\nare not associated with KIC 8462852. We recommend that attempts to model KIC\n8462852 A's light curve should revisit the possibility that the bound stellar\ncompanion may play a role in causing the irregular brightness variations, for\nexample through disruption of the orbits of bodies around the primary due to\nlong-term orbital evolution of the binary orbit.\n"}
{"abstract": "  Two-dimensional array-based datasets are pervasive in a variety of domains.\nCurrent approaches for generative modeling have typically been limited to\nconventional image datasets and performed in the pixel domain which do not\nexplicitly capture the correlation between pixels. Additionally, these\napproaches do not extend to scientific and other applications where each\nelement value is continuous and is not limited to a fixed range. In this paper,\nwe propose a novel approach for generating two-dimensional datasets by moving\nthe computations to the space of representation bases and show its usefulness\nfor two different datasets, one from imaging and another from scientific\ncomputing. The proposed approach is general and can be applied to any dataset,\nrepresentation basis, or generative model. We provide a comprehensive\nperformance comparison of various combinations of generative models and\nrepresentation basis spaces. We also propose a new evaluation metric which\ncaptures the deficiency of generating images in pixel space.\n"}
{"abstract": "  In the present work, we study the large scale matter power spectrum as well\nas the observed galaxy power spectrum for non-canonical tachyon field dark\nenergy model considering the full general relativistic perturbation equations.\nWe form a set of coupled autonomous equations including both the background and\nlinearly perturbed quantities and obtain their solutions numerically with\nproper set of initial conditions. We consider different scalar field potentials\nfor our study. Deviations from concordance $\\Lambda$CDM model are studied for\ndifferent relevant quantities. Our study shows that non-canonical tachyon dark\nenergy model produces enhanced gravitational potentials, comoving density\ncontrast as well as linear growth factor for matter perturbations compared to\n$\\Lambda$CDM. It is also observed that for tachyon dark energy models, there is\nsuppression of power on large scales compared to both $\\Lambda$CDM model as\nwell as previously studied canonical scalar field models.\n"}
{"abstract": "  We present some problems related to the conjugacy classes of\n$\\mathsf{Aut}(\\mathbb{N}^*)$.\n"}
{"abstract": "  In the pseudospin-$\\frac{1}{2}$ honeycomb Mott insulators with strong\nspin-orbit coupling, there are two types of bond-dependent exchange\ninteractions, named Kitaev ($K$) and $\\Gamma$, leading to strong frustration.\nWhile the ground state of the Kitaev model is a quantum spin liquid with\nfractionalized excitations, the ground state of the $\\Gamma$ model remains\ncontroversial. In particular, the phase diagram of the $K\\Gamma$ model with\nferromagnetic $K$ and antiferromagnetic $\\Gamma$ interactions has been\nintensively studied because of its relevance to candidate materials such as\n$\\alpha$-RuCl$_{3}$. Numerical studies also included the effects of tuning the\nbond strengths, i.e., $z$-bond strength different from the other bonds.\nHowever, no clear consensus on the overall phase diagram has been reached yet.\nHere we study the classical $K\\Gamma$ model with the anisotropic bond strength\nusing Monte Carlo simulations to understand the emerging phases out of\ncompetition between the two frustrated limits. We also address how the\nanisotropic bond strength affects the phase diagram and strength of quantum\nfluctuations. We found various large unit cell phases due to the competing\nfrustrations, and analyzed their intrinsic degeneracy based on the symmetry of\nthe Hamiltonian. Using the linear spin wave theory we showed that the\nanisotropic bond strength enhances quantum fluctuations in the\n$\\Gamma$-dominant regime where a small reduced moment is observed. The\nimplications of our findings in relation to the quantum model are also\ndiscussed.\n"}
{"abstract": "  Many interesting applications of hyperbolic systems of equations are stiff,\nand require the time step to satisfy restrictive stability conditions. One way\nto avoid small time steps is to use implicit time integration. Implicit\nintegration is quite straightforward for first order schemes. High order\nschemes instead need also to control spurious oscillations, which requires\nlimiting in space and time also in the implicit case. We propose a framework to\nsimplify considerably the application of high order non oscillatory schemes\nthrough the introduction of a low order implicit predictor, which is used both\nto set up the nonlinear weights of a standard high order space reconstruction,\nand to achieve limiting in time. In this preliminary work, we concentrate on\nthe case of a third order scheme, based on DIRK integration in time and CWENO\nreconstruction in space. The numerical tests involve linear and nonlinear\nscalar conservation laws.\n"}
{"abstract": "  Friedlander, Mac\\^{e}do, and Pong recently introduced the projected polar\nproximal point algorithm (P4A) for solving optimization problems by using the\nclosed perspective transforms of convex objectives. We analyse a generalization\n(GP4A) which replaces the closed perspective transform with a more general\nclosed gauge. We decompose GP4A into the iterative application of two separate\noperators, and analyse it as a splitting method. By showing that GP4A and its\nunder-relaxations exhibit global convergence whenever a fixed point exists, we\nobtain convergence guarantees for P4A by letting the gauge specify to the\nclosed perspective transform for a convex function. We then provide\neasy-to-verify sufficient conditions for the existence of fixed points for the\nGP4A, using the Minkowski function representation of the gauge. Conveniently,\nthe approach reveals that global minimizers of the objective function for P4A\nform an exposed face of the dilated fundamental set of the closed perspective\ntransform.\n"}
{"abstract": "  Microservice and serverless computing systems open up massive versatility and\nopportunity to distributed and datacenter-scale computing. In the meantime, the\ndeployments of modern datacenter resources are moving to disaggregated\narchitectures. With the flourishing growths from both sides, we believe this is\nhigh time to write this vision paper to propose a potential research agenda to\nachieve efficient deployments, management, and executions of next-generation\nmicroservices on top of the emerging disaggregated datacenter architectures. In\nparticular, we envision a critical systems research direction of designing and\ndeveloping offloadable and migratable microservices on disaggregated\narchitectures. With this vision, we have surveyed the recent related work to\ndemonstrate the importance and necessity of researching it. We also outline the\nfundamental challenges that distributed systems and datacenter-scale computing\nresearch may encounter. We further propose a research roadmap to achieve our\nenvisioned objectives in a promising way. Within the roadmap, we identify\npotential techniques and methods that can be leveraged.\n"}
{"abstract": "  The cogirth, $g^\\ast(M)$, of a matroid $M$ is the size of a smallest\ncocircuit of $M$. Finding the cogirth of a graphic matroid can be done in\npolynomial time, but Vardy showed in 1997 that it is NP-hard to find the\ncogirth of a binary matroid. In this paper, we show that $g^\\ast(M)\\leq\n\\frac{1}{2}\\vert E(M)\\vert$ when $M$ is binary, unless $M$ simplifies to a\nprojective geometry. We also show that, when equality holds, $M$ simplifies to\na Bose-Burton geometry, that is, a matroid of the form $PG(r-1,2)-PG(k-1,2)$.\nThese results extend to matroids representable over arbitrary finite fields.\n"}
{"abstract": "  Understanding an opponent agent helps in negotiating with it. Existing works\non understanding opponents focus on preference modeling (or estimating the\nopponent's utility function). An important but largely unexplored direction is\nrecognizing an opponent's negotiation strategy, which captures the opponent's\ntactics, e.g., to be tough at the beginning but to concede toward the deadline.\nRecognizing complex, state-of-the-art, negotiation strategies is extremely\nchallenging, and simple heuristics may not be adequate for this purpose. We\npropose a novel data-driven approach for recognizing an opponent's s\nnegotiation strategy. Our approach includes a data generation method for an\nagent to generate domain-independent sequences by negotiating with a variety of\nopponents across domains, a feature engineering method for representing\nnegotiation data as time series with time-step features and overall features,\nand a hybrid (recurrent neural network-based) deep learning method for\nrecognizing an opponent's strategy from the time series of bids. We perform\nextensive experiments, spanning four problem scenarios, to demonstrate the\neffectiveness of our approach.\n"}
{"abstract": "  In the paper, we proposed the Dantzig selector based on the $\\ell_{1}-\\alpha\n\\ell_{2}$~$(0< \\alpha \\leq1)$ minimization for the signal recovery. In the\nDantzig selector, the constraint $\\|{\\bf A}^{\\top}({\\bf b}-{\\bf A}{\\bf\nx})\\|_\\infty \\leq \\eta$ for some small constant $\\eta>0$ means the columns of\n${\\bf A}$ has very weakly correlated with the error vector ${\\bf e}={\\bf A}{\\bf\nx}-{\\bf b}$. First, recovery guarantees based on the restricted isometry\nproperty (RIP) are established for signals. Next, we propose the effective\nalgorithm to solve the proposed Dantzig selector. Last, we illustrate the\nproposed model and algorithm by extensive numerical experiments for the\nrecovery of signals in the cases of Gaussian, impulsive and uniform noise. And\nthe performance of the proposed Dantzig selector is better than that of the\nexisting methods.\n"}
{"abstract": "  Malaria microscopy, microscopic examination of stained blood slides to detect\nparasite Plasmodium, is considered to be a gold-standard for detecting\nlife-threatening disease malaria. Detecting the plasmodium parasite requires a\nskilled examiner and may take up to 10 to 15 minutes to completely go through\nthe whole slide. Due to a lack of skilled medical professionals in the\nunderdeveloped or resource deficient regions, many cases go misdiagnosed;\nresulting in unavoidable complications and/or undue medication. We propose to\ncomplement the medical professionals by creating a deep learning-based method\nto automatically detect (localize) the plasmodium parasites in the photograph\nof stained film. To handle the unbalanced nature of the dataset, we adopt a\ntwo-stage approach. Where the first stage is trained to detect blood cells and\nclassify them into just healthy or infected. The second stage is trained to\nclassify each detected cell further into the life-cycle stage. To facilitate\nthe research in machine learning-based malaria microscopy, we introduce a new\nlarge scale microscopic image malaria dataset. Thirty-eight thousand cells are\ntagged from the 345 microscopic images of different Giemsa-stained slides of\nblood samples. Extensive experimentation is performed using different CNN\nbackbones including VGG, DenseNet, and ResNet on this dataset. Our experiments\nand analysis reveal that the two-stage approach works better than the one-stage\napproach for malaria detection. To ensure the usability of our approach, we\nhave also developed a mobile app that will be used by local hospitals for\ninvestigation and educational purposes. The dataset, its annotations, and\nimplementation codes will be released upon publication of the paper.\n"}
{"abstract": "  There are many sources of data giving information about the number of\nSARS-CoV-2 infections in the population, but all have major drawbacks,\nincluding biases and delayed reporting. For example, the number of confirmed\ncases largely underestimates the number of infections, deaths lag infections\nsubstantially, while test positivity rates tend to greatly overestimate\nprevalence. Representative random prevalence surveys, the only putatively\nunbiased source, are sparse in time and space, and the results come with a big\ndelay. Reliable estimates of population prevalence are necessary for\nunderstanding the spread of the virus and the effects of mitigation strategies.\nWe develop a simple Bayesian framework to estimate viral prevalence by\ncombining the main available data sources. It is based on a discrete-time SIR\nmodel with time-varying reproductive parameter. Our model includes likelihood\ncomponents that incorporate data of deaths due to the virus, confirmed cases,\nand the number of tests administered on each day. We anchor our inference with\ndata from random sample testing surveys in Indiana and Ohio. We use the results\nfrom these two states to calibrate the model on positive test counts and\nproceed to estimate the infection fatality rate and the number of new\ninfections on each day in each state in the USA. We estimate the extent to\nwhich reported COVID cases have underestimated true infection counts, which was\nlarge, especially in the first months of the pandemic. We explore the\nimplications of our results for progress towards herd immunity.\n"}
{"abstract": "  This article presents a new primal-dual weak Galerkin method for second order\nelliptic equations in non-divergence form. The new method is devised as a\nconstrained $L^p$-optimization problem with constraints that mimic the second\norder elliptic equation by using the discrete weak Hessian locally on each\nelement. An equivalent min-max characterization is derived to show the\nexistence and uniqueness of the numerical solution. Optimal order error\nestimates are established for the numerical solution under the discrete\n$W^{2,p}$ norm, as well as the standard $W^{1,p}$ and $L^p$ norms. An\nequivalent characterization of the optimization problem in term of a system of\nfixed-point equations via the proximity operator is presented. An iterative\nalgorithm is designed based on the fixed-point equations to solve the\noptimization problems. Implementation of the iterative algorithm is studied and\nconvergence of the iterative algorithm is established. Numerical experiments\nfor both smooth and non-smooth coefficients problems are presented to verify\nthe theoretical findings.\n"}
{"abstract": "  Data in the energy domain grows at unprecedented rates and is usually\ngenerated by heterogeneous energy systems. Despite the great potential that big\ndata-driven technologies can bring to the energy sector, general adoption is\nstill lagging. Several challenges related to controlled data exchange and data\nintegration are still not wholly achieved. As a result, fragmented applications\nare developed against energy data silos, and data exchange is limited to few\napplications. In this paper, we analyze the challenges and requirements related\nto energy-related data applications. We also evaluate the use of Energy Data\nEcosystems (EDEs) as data-driven infrastructures to overcome the current\nlimitations of fragmented energy applications. EDEs are inspired by the\nInternational Data Space (IDS) initiative launched in Germany at the end of\n2014 with an overall objective to take both the development and use of the IDS\nreference architecture model to a European/global level. The reference\narchitecture model consists of four architectures related to business,\nsecurity, data and service, and software aspects. This paper illustrates the\napplicability of EDEs and IDS reference architecture in real-world scenarios\nfrom the energy sector. The analyzed scenario is positioned in the context of\nthe EU-funded H2020 project PLATOON.\n"}
{"abstract": "  Non-generating resources such as thermostatically controlled loads (TCLs) can\narbitrage energy prices and provide balancing reserves when aggregated due to\ntheir thermal energy storage capacity. Based on a performed survey of Swedish\nsingle- and two-family dwellings with electric heating, this paper quantifies\nthe potential of TCLs to provide reserves to the power system in Sweden. To\nthis end, dwellings with heat pumps and direct electric heaters are modeled as\nthermal energy storage equivalents that can be included in a linear two-stage\nproblem formulation. We approach the operational flexibility of the TCLs by\nmodeling a risk-averse aggregator that controls decentralized TCLs and aims to\nmaximize its own profit. The results show a potential of 2 GW/0.1Hz averaged\nover a year, and up to 6.4 GW/0.1Hz peak capacity. Based on a sensitivity\nanalysis we derive policy implications regarding market timing and activation\nsignal.\n"}
{"abstract": "  In deformable registration, the geometric framework - large deformation\ndiffeomorphic metric mapping or LDDMM, in short - has inspired numerous\ntechniques for comparing, deforming, averaging and analyzing shapes or images.\nGrounded in flows, which are akin to the equations of motion used in fluid\ndynamics, LDDMM algorithms solve the flow equation in the space of plausible\ndeformations, i.e. diffeomorphisms. In this work, we make use of deep residual\nneural networks to solve the non-stationary ODE (flow equation) based on a\nEuler's discretization scheme. The central idea is to represent time-dependent\nvelocity fields as fully connected ReLU neural networks (building blocks) and\nderive optimal weights by minimizing a regularized loss function. Computing\nminimizing paths between deformations, thus between shapes, turns to find\noptimal network parameters by back-propagating over the intermediate building\nblocks. Geometrically, at each time step, ResNet-LDDMM searches for an optimal\npartition of the space into multiple polytopes, and then computes optimal\nvelocity vectors as affine transformations on each of these polytopes. As a\nresult, different parts of the shape, even if they are close (such as two\nfingers of a hand), can be made to belong to different polytopes, and therefore\nbe moved in different directions without costing too much energy. Importantly,\nwe show how diffeomorphic transformations, or more precisely bilipshitz\ntransformations, are predicted by our algorithm. We illustrate these ideas on\ndiverse registration problems of 3D shapes under complex topology-preserving\ntransformations. We thus provide essential foundations for more advanced shape\nvariability analysis under a novel joint geometric-neural networks\nRiemannian-like framework, i.e. ResNet-LDDMM.\n"}
{"abstract": "  Aims. The main purpose of this work is to provide a method to derive\ntabulated observational constraints on the halo mass function (HMF) by studying\nthe magnification bias effect on high-redshift submillimeter galaxies. Under\nthe assumption of universality, we parametrize the HMF according to two\ntraditional models, namely the Sheth and Tormen (ST) and Tinker fits and assess\ntheir performance in explaining the measured data within the {\\Lambda} cold\ndark matter ({\\Lambda}CDM) model. We also study the potential influence of the\nhalo occupation distribution (HOD) parameters in this analysis and discuss two\nimportant aspects regarding the HMF parametrization. Methods. We measure the\ncross-correlation function between a foreground sample of GAMA galaxies with\nredshifts in the range $0.2<z<0.8$ and a background sample of H-ATLAS galaxies\nwith redshifts in the range $1.2<z<4.0$ and carry out an MCMC algorithm to\ncheck this observable against its mathematical prediction within the halo model\nformalism. Results. If all HMF parameters are assumed to be positive, the ST\nfit only seems to fully explain the measurements by forcing the mean number of\nsatellite galaxies in a halo to increase substantially from its prior mean\nvalue. The Tinker fit, on the other hand, provides a robust description of the\ndata without relevant changes in the HOD parameters, but with some dependence\non the prior range of two of its parameters. When the normalization condition\nfor the HMF is dropped and we allow negative values of the $p_1$ parameter in\nthe ST fit, all the involved parameters are better determined, unlike the\nprevious models, thus deriving the most general HMF constraints. While all\ncases are in agreement with the traditional fits within the uncertainties, the\nlast one hints at a slightly higher number of halos at intermediate and high\nmasses, raising the important point of the allowed parameter range.\n"}
{"abstract": "  Small class-imbalanced datasets, common in many high-level semantic tasks\nlike discourse analysis, present a particular challenge to current\ndeep-learning architectures. In this work, we perform an extensive analysis on\nsentence-level classification approaches for the News Discourse dataset, one of\nthe largest high-level semantic discourse datasets recently published. We show\nthat a multitask approach can improve 7% Micro F1-score upon current\nstate-of-the-art benchmarks, due in part to label corrections across tasks,\nwhich improve performance for underrepresented classes. We also offer a\ncomparative review of additional techniques proposed to address resource-poor\nproblems in NLP, and show that none of these approaches can improve\nclassification accuracy in such a setting.\n"}
{"abstract": "  Inspired by the Hardy-Littlewood maximal function, we propose a novel pooling\nstrategy which is called maxfun pooling. It is presented both as a viable\nalternative to some of the most popular pooling functions, such as max pooling\nand average pooling, and as a way of interpolating between these two\nalgorithms. We demonstrate the features of maxfun pooling with two\napplications: first in the context of convolutional sparse coding, and then for\nimage classification.\n"}
{"abstract": "  This study proposes a generalized coordinates based smoothed particle\nhydrodynamics (GSPH) method with overset methods using a Total Lagrangian (TL)\nformulation for large deformation and crack propagation problems. In the\nproposed GSPH, the physical space is decomposed into multiple domains, each of\nwhich is mapped to a local coordinate space (generalized space) to avoid\ncoordinate singularities as well as to flexibly change the spatial resolution.\nThe smoothed particle hydrodynamics (SPH) particles are then non-uniformly,\ne.g., typically in the boundary-conforming way, distributed in the physical\nspace while they are defined uniformly in each generalized space similarly to\nthe normal SPH method, which are numerically related by a coordinate\ntransformation matrix. By solving a governing equation in each generalized\nspace, the shape and size of the SPH kernel can be spatially changed in the\nphysical space so that a spatial resolution is adaptively varied a priori\ndepending on the deformation characteristics, and thus, a low-cost calculation\nwith the less number of particles is achieved in complex shape structures.\n"}
{"abstract": "  In this paper, we analyze the spectra of the preconditioned matrices arising\nfrom discretized multi-dimensional Riesz spatial fractional diffusion\nequations. The finite difference method is employed to approximate the\nmulti-dimensional Riesz fractional derivatives, which will generate symmetric\npositive definite ill-conditioned multi-level Toeplitz matrices. The\npreconditioned conjugate gradient method with a preconditioner based on the\nsine transform is employed to solve the resulting linear system. Theoretically,\nwe prove that the spectra of the preconditioned matrices are uniformly bounded\nin the open interval (1/2,3/2) and thus the preconditioned conjugate gradient\nmethod converges linearly. The proposed method can be extended to multi-level\nToeplitz matrices generated by functions with zeros of fractional order. Our\ntheoretical results fill in a vacancy in the literature. Numerical examples are\npresented to demonstrate our new theoretical results in the literature and show\nthe convergence performance of the proposed preconditioner that is better than\nother existing preconditioners.\n"}
{"abstract": "  Let $\\alpha'$ and $\\mu_i$ denote the matching number of a non-empty simple\ngraph $G$ with $n$ vertices and the $i$-th smallest eigenvalue of its Laplacian\nmatrix, respectively. In this paper, we prove a tight lower bound $$\\alpha' \\ge\n\\min\\left\\{\\Big\\lceil\\frac{\\mu_2}{\\mu_n} (n -1)\\Big\\rceil,\\ \\\n\\Big\\lceil\\frac{1}{2}(n-1)\\Big\\rceil \\right\\}.$$ This bound strengthens the\nresult of Brouwer and Haemers who proved that if $n$ is even and $2\\mu_2 \\ge\n\\mu_n$, then $G$ has a perfect matching. A graph $G$ is factor-critical if for\nevery vertex $v\\in V(G)$, $G-v$ has a perfect matching. We also prove an\nanalogue to the result of Brouwer and Haemers mentioned above by showing that\nif $n$ is odd and $2\\mu_2 \\ge \\mu_n$, then $G$ is factor-critical. We use the\nseparation inequality of Haemers to get a useful lemma, which is the key idea\nin the proofs. This lemma is of its own interest and has other applications. In\nparticular, we prove similar results for the number of balloons, spanning even\nsubgraphs, as well as spanning trees with bounded degree.\n"}
{"abstract": "  Two-dimensional (2D) materials are suitable hosts for the intercalation of\nextrinsic guest ions such as Li+, Na+ and K+ as the interlayer coupling is\nweak. This allows ion intercalation engineering of 2D materials, which may be a\nkey to advancing technological applications in energy storage, neuromorphic\nelectronics, and bioelectronics. However, ions that are extrinsic to the host\nmaterials possess challenges in fabrication of devices as there are extra steps\nof ion intercalation. This results in degradation of the long-term stability of\nthe intercalated atomically thin structures. Here, we introduce large-area\nsingle-crystal ultra-thin layered MnO$_2$ via chemical vapor deposition,\nspontaneously intercalated by potassium ions during the synthesis. We studied\nthe ultra-thin 2D K-MnO$_2$ in detail and showed that charge transport in these\ncrystals is dominated by motion of hydrated potassium ions in the interlayer\nspace. Moreover, K-MnO$_2$ crystals exhibit reversible layered-to-spinel phase\ntransition accompanied by an optical contrast change based on the electrical\nand optical modulation of the potassium and the interlayer water concentration.\nWe used the electric field driven ionic motion in K-MnO$_2$ based devices to\ndemonstrate the memristive properties of two terminal devices. As a possible\napplication we showed that K-MnO$_2$ memristors display synapse-like behavior\nsuch as short and long-term potentiation and depression as well as ionic\ncoupling effects.\n"}
{"abstract": "  Extreme circumstances in which a local distribution system is electrically\nisolated from the main power supply may not always be avoidable. Efforts must\nbe made to keep the lights on for such an isolated distribution system (IDS)\nuntil reconnection to the main power source. In this paper, we propose a\nstrategy to enhance IDS survivability utilizing the coordination of two\nflexible approaches, namely, separable energy storage systems (SMESSs), which\nconstruct non-wires links for energy transmission between the IDS and the\nexternal live power sources, and demand response (DR), which adjusts the\ninternal electrical demand of the IDS to provide effective operating stress\nalleviation. Considering the uncertainty of renewable energy generation and\nloads, a two-stage robust optimization (RO) model involving the joint\nscheduling of these two approaches is constructed. The objective is to minimize\nthe fuel consumption and the decreased and nonserved demand under the\nworst-case scenario to endow the IDS with extended survivability. Finally, test\nis conducted and the results demonstrate the effectiveness of the proposed\nmethod in enhancing the survivability of IDS.\n"}
{"abstract": "  The Parker Solar Probe (PSP) mission presents a unique opportunity to study\nthe near-Sun solar wind closer than any previous spacecraft. During its fourth\nand fifth solar encounters, PSP had the same orbital trajectory, meaning that\nsolar wind was measured at the same latitudes and radial distances. We identify\ntwo streams measured at the same heliocentric distance ($\\sim$0.13au) and\nlatitude ($\\sim$-3.5$^{\\circ}$) across these encounters to reduce spatial\nevolution effects. By comparing the plasma of each stream, we confirm that they\nare not dominated by variable transient events, despite PSP's proximity to the\nheliospheric current sheet. Both streams are consistent with a previous slow\nAlfv\\'enic solar wind study once radial effects are considered, and appear to\noriginate at the Southern polar coronal hole boundary. We also show that the\nswitchback properties are not distinctly different between these two streams.\nLow $\\alpha$-particle abundance ($\\sim$ 0.6 %) is observed in the encounter 5\nstream, suggesting that some physical mechanism must act on coronal hole\nboundary wind to cause $\\alpha$-particle depletion. Possible explanations for\nour observations are discussed, but it remains unclear whether the depletion\noccurs during the release or the acceleration of the wind. Using a flux tube\nargument, we note that an $\\alpha$-particle abundance of $\\sim$ 0.6 % in this\nlow velocity wind could correspond to an abundance of $\\sim$ 0.9 % at 1 au.\nFinally, as the two streams roughly correspond to the spatial extent of a\nswitchback patch, we suggest that patches are distinct features of coronal hole\nwind.\n"}
{"abstract": "  Fluorescence microscopy has enabled a dramatic development in modern biology\nby visualizing biological organisms with micrometer scale resolution. However,\ndue to the diffraction limit, sub-micron/nanometer features are difficult to\nresolve. While various super-resolution techniques are developed to achieve\nnanometer-scale resolution, they often either require expensive optical setup\nor specialized fluorophores. In recent years, deep learning has shown the\npotentials to reduce the technical barrier and obtain super-resolution from\ndiffraction-limited images. For accurate results, conventional deep learning\ntechniques require thousands of images as a training dataset. Obtaining large\ndatasets from biological samples is not often feasible due to the\nphotobleaching of fluorophores, phototoxicity, and dynamic processes occurring\nwithin the organism. Therefore, achieving deep learning-based super-resolution\nusing small datasets is challenging. We address this limitation with a new\nconvolutional neural network-based approach that is successfully trained with\nsmall datasets and achieves super-resolution images. We captured 750 images in\ntotal from 15 different field-of-views as the training dataset to demonstrate\nthe technique. In each FOV, a single target image is generated using the\nsuper-resolution radial fluctuation method. As expected, this small dataset\nfailed to produce a usable model using traditional super-resolution\narchitecture. However, using the new approach, a network can be trained to\nachieve super-resolution images from this small dataset. This deep learning\nmodel can be applied to other biomedical imaging modalities such as MRI and\nX-ray imaging, where obtaining large training datasets is challenging.\n"}
{"abstract": "  We propose a new method for modeling the distribution function of high\ndimensional extreme value distributions. The Pickands dependence function\nmodels the relationship between the covariates in the tails, and we learn this\nfunction using a neural network that is designed to satisfy its required\nproperties. Moreover, we present new methods for recovering the spectral\nrepresentation of extreme distributions and propose a generative model for\nsampling from extreme copulas. Numerical examples are provided demonstrating\nthe efficacy and promise of our proposed methods.\n"}
{"abstract": "  Non-Hermitian systems with parity-time ($\\mathcal{PT}$) symmetry and\nanti-$\\mathcal{PT}$ symmetry give rise to exceptional points (EPs) with\nintriguing properties related to, e.g., chiral transport and enhanced\nsensitivity, due to the coalescence of eigenvectors. In this paper, we propose\na powerful and easily computable tool, based on the Hilbert-Schmidt speed\n(HSS), which does not require the diagonalization of the evolved density\nmatrix, to detect exactly the EPs and hence the critical behavior of the\n(anti-)$\\mathcal{PT}\\!-$symmetric systems, especially high-dimensional ones.\nOur theoretical predictions, made without the need for modification of the\nHilbert space, which is performed by diagonalizing one of the observables, are\ncompletely consistent with results extracted from recent experiments studying\nthe criticality in (anti-)$\\mathcal{PT}\\!-$symmetric systems. Nevertheless, not\nmodifying the Hilbert space of the non-Hermitian system, we find that the trace\ndistance, a measure of distinguishability of two arbitrary quantum states,\nwhose dynamics is known as a faithful witness of non-Markovianity in Hermitian\nsystems, may be non-contractive under the non-Hermitian evolution of the\nsystem. Therefore, it lacks one of the most important characteristics which\nmust be met by any standard witness of non-Markovianity.\n"}
{"abstract": "  Transformers recently are adapted from the community of natural language\nprocessing as a promising substitute of convolution-based neural networks for\nvisual learning tasks. However, its supremacy degenerates given an insufficient\namount of training data (e.g., ImageNet). To make it into practical utility, we\npropose a novel distillation-based method to train vision transformers. Unlike\nprevious works, where merely heavy convolution-based teachers are provided, we\nintroduce lightweight teachers with different architectural inductive biases\n(e.g., convolution and involution) to co-advise the student transformer. The\nkey is that teachers with different inductive biases attain different knowledge\ndespite that they are trained on the same dataset, and such different knowledge\ncompounds and boosts the student's performance during distillation. Equipped\nwith this cross inductive bias distillation method, our vision transformers\n(termed as CivT) outperform all previous transformers of the same architecture\non ImageNet.\n"}
{"abstract": "  In this note, I find a new property of the congruence lattice, Con$L$, of an\nSPS lattice $L$ (slim, planar, semimodular, where \"slim\" is the absence\nof~$\\mathsf M_3$ sublattices) with more than $2$ elements: \\emph{there are at\nleast two dual atoms in Con$L$}. So the three-element chain cannot be\nrepresented as the congruence lattice of an SPS lattice, supplementing a~result\nof G. Cz\\'edli.\n"}
{"abstract": "  We investigate in a systematic way hypercontractivity property in Orlicz\nspaces for Markov semi-groups related to homogeneous and non homogeneous\ndiffusions in $\\mathbb{R}^n$. We provide an explicit construction of a family\nof Orlicz functions for which we prove that the associated hypercontractivity\nproperty is equivalent to a suitable functional inequality.\n  Connections between our results and the Talagrand Conjecture on the\nregularization effect of the Ornstein-Uhlenbeck semi-group in $\\mathbb{L}_1$\nare also discussed.\n"}
{"abstract": "  Welcome to WeaSuL 2021, the First Workshop on Weakly Supervised Learning,\nco-located with ICLR 2021. In this workshop, we want to advance theory, methods\nand tools for allowing experts to express prior coded knowledge for automatic\ndata annotations that can be used to train arbitrary deep neural networks for\nprediction. The ICLR 2021 Workshop on Weak Supervision aims at advancing\nmethods that help modern machine-learning methods to generalize from knowledge\nprovided by experts, in interaction with observable (unlabeled) data. In total,\n15 papers were accepted. All the accepted contributions are listed in these\nProceedings.\n"}
{"abstract": "  Anomalous diffusion, which shows a deviation of transport dynamics from the\nframework of standard Brownian motion, is involved in the evolution of various\nphysical, chemical, biological, and economic systems. The study of such random\nprocesses is of fundamental importance in unveiling the physical properties of\nrandom walkers and complex systems. However, classical methods to characterize\nanomalous diffusion are often disqualified for individual short trajectories,\nleading to the launch of the Anomalous Diffusion (AnDi) Challenge. This\nchallenge aims at objectively assessing and comparing new approaches for single\ntrajectory characterization, with respect to three different aspects: the\ninference of the anomalous diffusion exponent; the classification of the\ndiffusion model; and the segmentation of trajectories. In this article, to\naddress the inference and classification tasks in the challenge, we develop a\nWaveNet-based deep neural network (WADNet) by combining a modified WaveNet\nencoder with long short-term memory networks, without any prior knowledge of\nanomalous diffusion. As the performance of our model has surpassed the current\n1st places in the challenge leaderboard on both two tasks for all dimensions (6\nsubtasks), WADNet could be the part of state-of-the-art techniques to decode\nthe AnDi database. Our method presents a benchmark for future research, and\ncould accelerate the development of a versatile tool for the characterization\nof anomalous diffusion.\n"}
{"abstract": "  A nonlinear small-strain elastic theory is constructed from a systematic\nexpansion in Biot strains, truncated at quadratic order. The primary motivation\nis the desire for a clean separation between stretching and bending energies\nfor shells, which appears to arise only from reduction of a bulk energy of this\ntype. An approximation of isotropic invariants, bypassing the solution of a\nquartic equation or computation of tensor square roots, allows stretches,\nrotations, stresses, and balance laws to be written in terms of derivatives of\nposition. Two-field formulations are also presented. Extensions to anisotropic\ntheories are briefly discussed.\n"}
{"abstract": "  This article describes a technique to augment a typical RGBD sensor by\nintegrating depth estimates obtained via Structure-from-Motion (SfM) with\nsensor depth measurements. Limitations in the RGBD depth sensing technology\nprevent capturing depth measurements in four important contexts: (1) distant\nsurfaces (>5m), (2) dark surfaces, (3) brightly lit indoor scenes and (4)\nsunlit outdoor scenes. SfM technology computes depth via multi-view\nreconstruction from the RGB image sequence alone. As such, SfM depth estimates\ndo not suffer the same limitations and may be computed in all four of the\npreviously listed circumstances. This work describes a novel fusion of RGBD\ndepth data and SfM-estimated depths to generate an improved depth stream that\nmay be processed by one of many important downstream applications such as\nrobotic localization and mapping, as well as object recognition and tracking.\n"}
{"abstract": "  A spatially homogeneous and locally rotationally symmetric Bianchi type-II\ncosmological model under the influence of both shear and bulk viscosity has\nbeen studied. Exact solutions are obtained with a barotropic equation of state\nbetween thermodynamics pressure and the energy density of the fluid, and\nconsidering the linear relationships amongst the energy density, the expansion\nscalar and the shear scalar. Special cases with vanishing bulk viscosity\ncoefficients and with the perfect fluid in the absence of viscosity have also\nbeen studied. The formal appearance of the solutions is the same for both the\nviscous as well as the perfect fluids. The difference is only in choosing a\nconstant parameter which appears in the solutions. In the cases of either a\nfluid with bulk viscosity alone or a perfect fluid, the barotropic equation of\nstate is no longer an additional assumption to be imposed; rather it follows\ndirectly from the field equations.\n"}
{"abstract": "  Inspired by the success of Deep Learning based approaches to English scene\ntext recognition, we pose and benchmark scene text recognition for three Indic\nscripts - Devanagari, Telugu and Malayalam. Synthetic word images rendered from\nUnicode fonts are used for training the recognition system. And the performance\nis bench-marked on a new IIIT-ILST dataset comprising of hundreds of real scene\nimages containing text in the above mentioned scripts. We use a segmentation\nfree, hybrid but end-to-end trainable CNN-RNN deep neural network for\ntranscribing the word images to the corresponding texts. The cropped word\nimages need not be segmented into the sub-word units and the error is\ncalculated and backpropagated for the the given word image at once. The network\nis trained using CTC loss, which is proven quite effective for\nsequence-to-sequence transcription tasks. The CNN layers in the network learn\nto extract robust feature representations from word images. The sequence of\nfeatures learnt by the convolutional block is transcribed to a sequence of\nlabels by the RNN+CTC block. The transcription is not bound by word length or a\nlexicon and is ideal for Indian languages which are highly inflectional.\nIIIT-ILST dataset, synthetic word images dataset and the script used to render\nsynthetic images are available at\nhttp://cvit.iiit.ac.in/research/projects/cvit-projects/iiit-ilst\n"}
{"abstract": "  Magnesium atoms fully embedded in helium nanodroplets are exposed to\ntwo-color laser pulses in order to exemplarily study the contribution of a\ndense and finite medium on multiphoton-triggered above-threshold ionization\n(ATI). The angular-resolved photoelectron spectra show striking differences\nwith respect to results obtained on free atoms. Scattering of the Mg\nphotoelectrons, when traversing the helium environment, causes the angular\ndistribution to become almost isotropic. Furthermore, we observe a marked\nincrease in the ATI order, pointing out the impact of the helium environment on\nthe concerted electron emission process. Phase-of-the-phase spectroscopy,\nhowever, reveals a marked loss in the $2\\omega-\\omega$ phase dependence of the\nelectron signal. Taking into account sideband formation on a quantitative\nlevel, a Monte-Carlo simulation which includes laser-assisted electron\nscattering can reproduce the experimental spectra and give insights into the\nstrong field-induced electron emission from disordered systems.\n"}
{"abstract": "  We develop the theory of fractional gradient flows: an evolution aimed at the\nminimization of a convex, l.s.c.~energy, with memory effects. This memory is\ncharacterized by the fact that the negative of the (sub)gradient of the energy\nequals the so-called Caputo derivative of the state. We introduce the notion of\nenergy solutions, for which we provide existence, uniqueness and certain\nregularizing effects. We also consider Lipschitz perturbations of this energy.\nFor these problems we provide an a posteriori error estimate and show its\nreliability. This estimate depends only on the problem data, and imposes no\nconstraints between consecutive time-steps. On the basis of this estimate we\nprovide an a priori error analysis that makes no assumptions on the smoothness\nof the solution.\n"}
{"abstract": "  Cosmogenic radio-nuclei are an important source of background for low-energy\nneutrino experiments. In Borexino, cosmogenic $^{11}$C decays outnumber solar\n$pep$ and CNO neutrino events by about ten to one. Highly efficient\nidentification of this background is mandatory for these neutrino analyses. We\npresent here the details of the most consolidated strategy, used throughout\nBorexino solar neutrino measurements. It hinges upon finding the space-time\ncorrelations between $^{11}$C decays, the preceding parent muons and the\naccompanying neutrons. This article describes the working principles and\nevaluates the performance of this Three-Fold Coincidence (TFC) technique in its\ntwo current implementations: a hard-cut and a likelihood-based approach. Both\nshow stable performances throughout Borexino Phases II (2012-2016) and III\n(2016-2020) data sets, with a $^{11}$C tagging efficiency of $\\sim$90 % and\n$\\sim$63-66 % of the exposure surviving the tagging. We present also a novel\ntechnique that targets specifically $^{11}$C produced in high-multiplicity\nduring major spallation events. Such $^{11}$C appear as a burst of events,\nwhose space-time correlation can be exploited. Burst identification can be\ncombined with the TFC to obtain about the same tagging efficiency of $\\sim$90 %\nbut with a higher fraction of the exposure surviving, in the range of\n$\\sim$66-68 %.\n"}
{"abstract": "  In Section 1, we revisit the partial differential equation (PDE) for the\nprobability generating function (PGF) of the time-nonhomogeneous BDI\n(birth-and-death-with-immigration) process and derive a closed form solution.\nTo the best of our knowledge, this is a new mathematical result. We state this\nresult as Proposition 1. We state as Corollary 1 that the negative binomial\ndistribution of the time-homogeneous BDI process discussed in Part I extends to\nthe general time-nonhomogeneous case, provided that the ratio of the\nimmigration rate to the birth rate is a constant. In section 1.2, we take up\nthe heuristic approach discussed by Bartlett and Bailey (1964), and carry it\nout to completion by arriving at the solution obtained above,.\n  In Section 2, we present the results of our extensive simulation experiments\nof the time-nonhomogeneous BD process that was analyzed in Part III-A and\nconfirm our analytic results. In Section 3, we undertake similar simulation\nexperiments for the BDI process that is analyzed in Section 1.\n  As we discuss in Section 4, our stochastic model now seems more promising and\npowerful than has been heretofore expected. In Appendix B, a closed form\nsolution for the M(t)/M(t)/infinity queue is obtained, as a special case of\nthis BDI process model.\n"}
{"abstract": "  We show that a strong `spin'-orbit coupled one-dimensional (1D) hole gas is\nachievable via applying a strong magnetic field to the original two-fold\ndegenerate (spin degeneracy) hole gas confined in a cylindrical Ge nanowire.\nBoth strong longitudinal and strong transverse magnetic fields are feasible to\nachieve this goal. Based on quasi-degenerate perturbation calculations, we show\nthe induced low-energy subband dispersion of the hole gas can be written as\n$E=\\hbar^{2}k^{2}_{z}/(2m^{*}_{h})+\\alpha\\sigma^{z}k_{z}+g^{*}_{h}\\mu_{B}B\\sigma^{x}/2$,\na form exactly the same as that of the electron gas in the conduction band.\nHere the Pauli matrices $\\sigma^{z,x}$ represent a pseudo spin (or `spin' ),\nbecause the real spin degree of freedom has been split off from the subband\ndispersions by the strong magnetic field. Also, for a moderate nanowire radius\n$R=10$ nm, the induced effective hole mass $m^{*}_{h}$ ($0.065\\sim0.08~m_{e}$)\nand the `spin'-orbit coupling $\\alpha$ ($0.35\\sim0.8$ eV~\\AA) have a small\nmagnetic field dependence in the studied magnetic field interval $1<B<15$ T,\nwhile the effective $g$-factor $g^{*}_{h}$ of the hole `spin' only has a small\nmagnetic field dependence in the large field region.\n"}
{"abstract": "  Guessing Random Additive Noise Decoding (GRAND) is a recently proposed\napproximate Maximum Likelihood (ML) decoding technique that can decode any\nlinear error-correcting block code. Ordered Reliability Bits GRAND (ORBGRAND)\nis a powerful variant of GRAND, which outperforms the original GRAND technique\nby generating error patterns in a specific order. Moreover, their simplicity at\nthe algorithm level renders GRAND family a desirable candidate for applications\nthat demand very high throughput. This work reports the first-ever hardware\narchitecture for ORBGRAND, which achieves an average throughput of up to $42.5$\nGbps for a code length of $128$ at an SNR of $10$ dB. Moreover, the proposed\nhardware can be used to decode any code provided the length and rate\nconstraints. Compared to the state-of-the-art fast dynamic successive\ncancellation flip decoder (Fast-DSCF) using a 5G polar $(128,105)$ code, the\nproposed VLSI implementation has $49\\times$ more average throughput while\nmaintaining similar decoding performance.\n"}
{"abstract": "  One of the main aspects of switched affine systems that makes their\nstabilizability study intricate is the existence of (generally) infinitely many\nequilibrium points in the state space. Thus, prior to designing the switched\ncontrol, the user must specify one of these equilibrium points to be the goal\nor reference. This can be a cumbersome task, especially if this goal is\npartially given or only defined as a set of constraints. To tackle this issue,\nin this paper we describe algorithms that can determine whether a given goal is\nan equilibrium point of the system and also jointly search for equilibrium\npoints and design globally stabilizing switching functions.\n"}
{"abstract": "  We consider the plurality consensus problem among $n$ agents. Initially, each\nagent has one of $k$ different opinions. Agents choose random interaction\npartners and revise their state according to a fixed transition function,\ndepending on their own state and the state of the interaction partners. The\ngoal is to reach a consensus configuration in which all agents agree on the\nsame opinion, and if there is initially a sufficiently large bias towards one\nopinion, that opinion should prevail.\n  We analyze a synchronized variant of the undecided state dynamics defined as\nfollows. The agents act in phases, consisting of a decision and a boosting\npart. In the decision part, any agent that encounters an agent with a different\nopinion becomes undecided. In the boosting part, undecided agents adopt the\nfirst opinion they encounter.\n  We consider this dynamics in the population model and the gossip model. For\nthe population model, our protocol reaches consensus (w.h.p.) in $O(\\log^2 n)$\nparallel time, providing the first polylogarithmic result for $k > 2$ (w.h.p.)\nin this model. Without any assumption on the bias, fast consensus has only been\nshown for $k = 2$ for the unsynchronized version of the undecided state\ndynamics [Clementi et al., MFCS'18]. We show that the synchronized variant of\nthe undecided state dynamics reaches consensus (w.h.p.) in time $O(\\log^2 n)$,\nindependently of the initial number, bias, or distribution of opinions. In both\nmodels, we guarantee that if there is an initial bias of $\\Omega(\\sqrt{n \\log\nn})$, then (w.h.p.) that opinion wins.\n  A simple extension of our protocol in the gossip model yields a dynamics that\ndoes not depend on $n$ or $k$, is anonymous, and has (w.h.p.) runtime $O(\\log^2\nn)$. This solves an open problem formulated by Becchetti et al.~[Distributed\nComputing,~2017].\n"}
{"abstract": "  Getting the distance to objects is crucial for autonomous vehicles. In\ninstances where depth sensors cannot be used, this distance has to be estimated\nfrom RGB cameras. As opposed to cars, the task of estimating depth from\non-board mounted cameras is made complex on drones because of the lack of\nconstrains on motion during flights. In this paper, we present a method to\nestimate the distance of objects seen by an on-board mounted camera by using\nits RGB video stream and drone motion information. Our method is built upon a\npyramidal convolutional neural network architecture and uses time recurrence in\npair with geometric constraints imposed by motion to produce pixel-wise depth\nmaps. In our architecture, each level of the pyramid is designed to produce its\nown depth estimate based on past observations and information provided by the\nprevious level in the pyramid. We introduce a spatial reprojection layer to\nmaintain the spatio-temporal consistency of the data between the levels. We\nanalyse the performance of our approach on Mid-Air, a public drone dataset\nfeaturing synthetic drone trajectories recorded in a wide variety of\nunstructured outdoor environments. Our experiments show that our network\noutperforms state-of-the-art depth estimation methods and that the use of\nmotion information is the main contributing factor for this improvement. The\ncode of our method is publicly available on GitHub; see\nhttps://github.com/michael-fonder/M4Depth\n"}
{"abstract": "  Fixed length summarization aims at generating summaries with a preset number\nof words or characters. Most recent researches incorporate length information\nwith word embeddings as the input to the recurrent decoding unit, causing a\ncompromise between length controllability and summary quality. In this work, we\npresent an effective length controlling unit Length Attention (LenAtten) to\nbreak this trade-off. Experimental results show that LenAtten not only brings\nimprovements in length controllability and ROGUE scores but also has great\ngeneralization ability. In the task of generating a summary with the target\nlength, our model is 732 times better than the best-performing length\ncontrollable summarizer in length controllability on the CNN/Daily Mail\ndataset.\n"}
{"abstract": "  The graphical structure of Probabilistic Graphical Models (PGMs) encodes the\nconditional independence (CI) relations that hold in the modeled distribution.\nGraph algorithms, such as d-separation, use this structure to infer additional\nconditional independencies, and to query whether a specific CI holds in the\ndistribution. The premise of all current systems-of-inference for deriving CIs\nin PGMs, is that the set of CIs used for the construction of the PGM hold\nexactly. In practice, algorithms for extracting the structure of PGMs from\ndata, discover approximate CIs that do not hold exactly in the distribution. In\nthis paper, we ask how the error in this set propagates to the inferred CIs\nread off the graphical structure. More precisely, what guarantee can we provide\non the inferred CI when the set of CIs that entailed it hold only\napproximately? It has recently been shown that in the general case, no such\nguarantee can be provided. We prove that such a guarantee exists for the set of\nCIs inferred in directed graphical models, making the d-separation algorithm a\nsound and complete system for inferring approximate CIs. We also prove an\napproximation guarantee for independence relations derived from marginal CIs.\n"}
{"abstract": "  Variable and function names are extremely important for program\ncomprehension. It is therefore also important to study how developers select\nnames. But controlled experiments on naming are hindered by the need to\ndescribe to experimental subjects what it is they need to name. Words appearing\nin these descriptions may then find their way into the names, leading to a bias\nin the results. We suggest that this problem can be alleviated by using emojis\nor other small graphics in lieu of key words in the descriptions. A replication\nof previous work on naming, this time including such emojis and graphics,\nindeed led to a more diverse and less biased choice of words in the names than\nwhen using English descriptions.\n"}
{"abstract": "  In this work, we consider the uplink of a scalable cell-free massive MIMO\nsystem where the users are served only by a subset of access points (APs) in\nthe network. The APs are physically grouped into predetermined \"cell-centric\nclusters\", which are connected to different cooperative central processing\nunits (CPUs). Given the cooperative nature of the considered communications\nnetwork, we assume that each user is associated with a \"virtual cluster\", that,\nin general, involves some APs belonging to different cell-centric clusters.\nAssuming the maximum-ratio-combining at the APs, we propose a user-association\nprocedure aimed at the maximization of the sum-rate of the users in the system.\nThe proposed procedure is based on the Hungarian Algorithm and exploits only\nthe knowledge of the position of the APs in the network. Numerical results\nreveal that the performance of the proposed approach is not always better than\nthe alternatives but it offers a considerably lower backhaul load with a\nnegligible performance loss compared to full-cell free approaches.\n"}
{"abstract": "  The neutron skin thickness $\\Delta r_{\\rm{np}}$ of heavy nuclei is\nessentially determined by the symmetry energy density slope $L({\\rho })$ at\n$\\rho_c = 0.11/0.16\\rho_0$ ($\\rho_0$ is nuclear saturation density), roughly\ncorresponding to the average density of finite nuclei. The PREX collaboration\nrecently reported a model-independent extraction of $\\Delta r^{208}_{\\rm{np}} =\n0.29 \\pm 0.07$ fm for the $\\Delta r_{\\rm{np}}$ of $^{208}$Pb, which suggests a\nrather stiff symmetry energy $E_{\\rm{sym}}({\\rho })$ with $L({\\rho_c }) \\ge 55$\nMeV. We demonstrate that the $E_{\\rm{sym}}({\\rho })$ cannot be too stiff and\n$L({\\rho_c }) \\le 73$ MeV is necessary to be compatible with (1) the\nground-state properties and giant monopole resonances of finite nuclei, (2) the\nconstraints on the equation of state of symmetric nuclear matter at\nsuprasaturation densities from flow data in heavy-ion collisions, (3) the\nlargest neutron star (NS) mass reported so far for PSR J0740+6620, (4) the NS\ntidal deformability extracted from gravitational wave signal GW170817 and (5)\nthe mass-radius of PSR J0030+045 measured simultaneously by NICER. This allow\nus to obtain $55 \\le L({\\rho_c }) \\le 73$ MeV and $0.22 \\le \\Delta\nr^{208}_{\\rm{np}} \\le 0.27$ fm, and further $E_{\\rm{sym}}({\\rho_0 }) = 34.5 \\pm\n1.5$ MeV, $L({\\rho_0 }) = 85.5 \\pm 22.2$ MeV, and $E_{\\rm{sym}}({2\\rho_0 }) =\n63.9 \\pm 14.8$ MeV. A number of critical implications on nuclear physics and\nastrophysics are discussed.\n"}
{"abstract": "  It is shown that the possibility of using Maxwell demon to cheating in\nquantum non-locality tests is prohibited by the Landauer's erasure principle.\n"}
{"abstract": "  Pseudo-labeling is the most adopted method for pre-training automatic speech\nrecognition (ASR) models. However, its performance suffers from the supervised\nteacher model's degrading quality in low-resource setups and under domain\ntransfer. Inspired by the successes of contrastive representation learning for\ncomputer vision and speech applications, and more recently for supervised\nlearning of visual objects, we propose Contrastive Semi-supervised Learning\n(CSL). CSL eschews directly predicting teacher-generated pseudo-labels in favor\nof utilizing them to select positive and negative examples. In the challenging\ntask of transcribing public social media videos, using CSL reduces the WER by\n8% compared to the standard Cross-Entropy pseudo-labeling (CE-PL) when 10hr of\nsupervised data is used to annotate 75,000hr of videos. The WER reduction jumps\nto 19% under the ultra low-resource condition of using 1hr labels for teacher\nsupervision. CSL generalizes much better in out-of-domain conditions, showing\nup to 17% WER reduction compared to the best CE-PL pre-trained model.\n"}
{"abstract": "  The planar topological Hall effect (PTHE), appeared when the magnetic field\ntended to be along the current, is believed to result from the real-space Berry\ncurvature of the spin spiral structure and has been experimentally observed in\nskyrmion-hosting materials. In this paper, we report an experimental\nobservation of the PTHE in a hexagonal ferromagnetic Fe5Sn3 single crystal.\nWith a current along the c axis of Fe5Sn3, the transverse resistivity curves\nexhibited obvious peaks near the saturation field as the magnetic field rotated\nto the current and appeared more obvious with increasing temperature, which was\nrelated to the noncoplanar spin structure in Fe5Sn3. This spin structure\ninduced nonzero scalar spin chirality, which acted as fictitious magnetic\nfields to conduction electrons and contributed the additional transverse\nsignal. These findings deepen the understanding of the interaction between\nconduction electrons and complex magnetic structures and are instructive for\nthe design of next-generation spintronic devices.\n"}
{"abstract": "  The Zr-based MAX phases have attracted considerable attention for their\noutstanding irradiation behavior and high neutron transparency relevant to\nnuclear power generation technologies. In spite of increased understanding of\nphysical behavior crystalline MAX phases, the high-temperature oxidation\nbehavior and reaction mechanism of disordered MAX phases both from theory and\nexperiments are not well understood due to increased system complexity. Here,\nwe present a detailed comparative assessment of high-temperature\nthermodynamic-stability and oxidation behavior (reaction-products and chemical\nactivity) of ordered Ti2AlC and disordered (Ti0.5Zr0.5)2AlC. We believe that\nthe new insights will enhance our understanding of oxidation process in\ndisordered MAX phases.\n"}
{"abstract": "  This is the second of two papers that study the asymptotic structure of\nspace-times with a non-negative cosmological constant $\\Lambda$. This paper\ndeals with the case $\\Lambda >0$. Our approach is founded on the `tidal\nenergies' built with the Weyl curvature and, specifically, we use the\nasymptotic super-Poynting vector computed from the rescaled Bel-Robinson tensor\nat infinity to provide a covariant, gauge-invariant, criterion for the\nexistence, or absence, of gravitational radiation at infinity. The fundamental\nidea we put forward is that the physical asymptotic properties are encoded in\n$(\\scri,h_{ab},D_{ab})$, where the first element of the triplet is a\n3-dimensional manifold, the second is a representative of a conformal class of\nRiemannian metrics on $\\scri$, and the third element is a traceless symmetric\ntensor field on $\\scri$. We similarly propose a no-incoming radiation criterion\nbased also on the triplet $(\\scri,h_{ab},D_{ab})$ and on radiant supermomenta\ndeduced from the rescaled Bel-Robinson tensor too. We search for news tensors\nand argue that any news-like object must be associated to, and depends on,\n2-dimensional cross-sections of $\\scri$. We identify one component of news for\nevery such cross-section and present a general strategy to find the second\ncomponent. We also introduce the concept of equipped $\\scri$, consider the\nlimit $\\Lambda\\rightarrow 0$ and apply all our results to selected exact\nsolutions of Einstein Field Equations. The full-length abstract is available in\nthe paper.\n"}
{"abstract": "  We propose an end-to-end network that takes a single perspective RGB image of\na complex road scene as input, to produce occlusion-reasoned layouts in\nperspective space as well as a top-view parametric space. In contrast to prior\nworks that require dense supervision such as semantic labels in perspective\nview, the only human annotations required by our method are for parametric\nattributes that are cheaper and less ambiguous to obtain. To solve this\nchallenging task, our design is comprised of modules that incorporate inductive\nbiases to learn occlusion-reasoning, geometric transformation and semantic\nabstraction, where each module may be supervised by appropriately transforming\nthe parametric annotations. We demonstrate how our design choices and proposed\ndeep supervision help achieve accurate predictions and meaningful\nrepresentations. We validate our approach on two public datasets, KITTI and\nNuScenes, to achieve state-of-the-art results with considerably lower human\nsupervision.\n"}
{"abstract": "  The Belle II experiment at the Super B factory SuperKEKB, an asymmetric\n$e^+e^-$ collider located in Tsukuba, Japan, is tailored to perform precision B\nphysics measurements. The centre of mass energy of the collisions is equal to\nthe rest mass of the $\\Upsilon(4S)$ resonance of $m_{\\Upsilon(4S)} = 10.58\\,\\rm\nGeV$. A high vertex resolution is essential for measuring the decay vertices of\nB mesons. Typical momenta of the decay products are ranging from a few tens of\nMeV to a few GeV and multiple scattering has a significant impact on the vertex\nresolution. The VerteX Detector (VXD) for Belle II is therefore designed to\nhave as little material as possible inside the acceptance region. Especially\nthe innermost two layers, populated by the PiXel Detector (PXD), have to be\nultra-thin. The PXD is based on DEpleted P-channel Field Effect Transistors\n(DEPFETs) with a thickness of only $75\\,\\rm\\mu m$. Spatial resolution and hit\nefficiency of production detector modules were studied in beam tests performed\nat the DESY test beam facility. The spatial resolution was investigated as a\nfunction of the incidence angle and improvements due to charge sharing are\ndemonstrated. The measured module performance is compatible with the\nrequirements for Belle II.\n"}
{"abstract": "  In this paper, we investigate the relationship between the weak min-max\nproperty and the diameter uniformity of domains in Banach spaces with\ndimensions at least $2$. As an application, we show that diameter uniform\ndomains are invariant under relatively quasim\\\"obius mappings.\n"}
{"abstract": "  The field of face recognition (FR) has witnessed great progress with the\nsurge of deep learning. Existing methods mainly focus on extracting\ndiscriminative features, and directly compute the cosine or L2 distance by the\npoint-to-point way without considering the context information. In this study,\nwe make a key observation that the local con-text represented by the\nsimilarities between the instance and its inter-class neighbors1plays an\nimportant role forFR. Specifically, we attempt to incorporate the local\nin-formation in the feature space into the metric, and pro-pose a unified\nframework calledInter-class DiscrepancyAlignment(IDA), with two dedicated\nmodules, Discrepancy Alignment Operator(IDA-DAO) andSupport Set\nEstimation(IDA-SSE). IDA-DAO is used to align the similarity scores considering\nthe discrepancy between the images and its neighbors, which is defined by\nadaptive support sets on the hypersphere. For practical inference, it is\ndifficult to acquire support set during online inference. IDA-SSE can provide\nconvincing inter-class neighbors by introducing virtual candidate images\ngenerated with GAN. Further-more, we propose the learnable IDA-SSE, which can\nimplicitly give estimation without the need of any other images in the\nevaluation process. The proposed IDA can be incorporated into existing FR\nsystems seamlessly and efficiently. Extensive experiments demonstrate that this\nframe-work can 1) significantly improve the accuracy, and 2) make the model\nrobust to the face images of various distributions.Without bells and whistles,\nour method achieves state-of-the-art performance on multiple standard FR\nbenchmarks.\n"}
{"abstract": "  The phenomena of intermittent and complete synchronization between two out of\nthree identical, magnetically coupled SQUIDs (Superconducting QUantum\nInterference Devices) are investigated numerically. SQUIDs are highly nonlinear\nsuperconducting oscillators/devices that exhibit strong resonant and tunable\nresponse to applied magnetic field(s). Single SQUIDs and SQUID arrays are\ntechnologically important solid state devices, and they also serve as a testbed\nfor exploring numerous complex dynamical phenomena. In SQUID oligomers, the\ndynamic complexity increases considerably with the number of SQUIDs. The SQUID\ntrimer, considered here in a linear geometrical configuration using a realistic\nmodel with accesible control parameters, exhibits chaotic and hyperchaotic\nbehavior in wide parameter regions. Complete chaos synchronization as well as\nintermittent chaos synchronization between two SQUIDs of the trimer is\nidentified and characterized using the complete Lyapunov spectrum of the system\nand appropriate measures. The passage from complete to intermittent\nsynchronization seems to be related to chaos-hyperchaos transitions as has been\nconjectured in the early days of chaos synchronization.\n"}
{"abstract": "  This paper contains two contributions in the study of optimal transport on\nmetric graphs. Firstly, we prove a Benamou-Brenier formula for the Wasserstein\ndistance, which establishes the equivalence of static and dynamical optimal\ntransport. Secondly, in the spirit of Jordan-Kinderlehrer-Otto, we show that\nMcKean-Vlasov equations can be formulated as gradient flow of the free energy\nin the Wasserstein space of probability measures. The proofs of these results\nare based on careful regularisation arguments to circumvent some of the\ndifficulties arising in metric graphs, namely, branching of geodesics and the\nfailure of semi-convexity of entropy functionals in the Wasserstein space.\n"}
{"abstract": "  In this paper, we will study the existence of finite time singularity to\nharmonic heat flow and their formation patterns. After works of\nCoron-Ghidaglia, Ding and Chen-Ding, one knows blow-up solutions under\nsmallness of initial energy for m>=3. soon later, 2 dimensional blowup\nsolutions were found by Chang-Ding-Ye. The first part of this paper is devoted\nto construction of new examples of finite time blow-up solutions without\nsmallness conditions for 3<=m<7. In fact, when considering rotational symmetric\nharmonic heat flow from B_1\\subset R^m to S^m\\subset R^{m+1}, we will prove\nthat the maximal solution blows up in finite time if b>\\vartheta_m, and exists\nfor all time if 0<b<\\pi/2. This result can be regarded as a generalization of\nresults of Chang-Ding-Ye nad Chang-Ding to higher dimensional case, which\nrelies on a completely different argument. The second part of the paper study\nthe rate of blow-up solutions. When M is a bounded domain in R^2 and consider\nDirichlet boundary condition on \\partial M, Hamilton has obtained that the\nblowup rate must be faster than (T-t)^{-1}. Under a similar setting, it was\nlater improved a litttle by Topping to (T-t)^{-1}|log(T-t)|. In this paper, we\nwill extend the results to all Riemmanian surfaces M and improve the rate of\nTopping to (T-t)^{-1}a(|log(T-t)|) for any positive nondecreasing function\na(\\tau) satisfying $\\int^\\infty_1\\frac{d\\tau}{a(\\tau)}=+\\infty$, which is\ncomparable to a recent result of Raphael-Schweyer for rotational symmetric\nsolutions. Turning to the higher dimensional case 3<=m<7, we will demonstrate a\ncompletely different phenomenon by showing that all rotational symmetric\nblow-up solutions can not be type II, which is different to the case m>=7 by\nBizon-Wasserman. Finally, we also present result of finite time type I blowup\nfor heat flow from S^m to S^m\\subset R^{m+1}, when 3<=m<7 and degree is no less\nthan 2.\n"}
{"abstract": "  In topological data analysis and visualization, topological descriptors such\nas persistence diagrams, merge trees, contour trees, Reeb graphs, and\nMorse-Smale complexes play an essential role in capturing the shape of scalar\nfield data. We present a state-of-the-art report on scalar field comparison\nusing topological descriptors. We provide a taxonomy of existing approaches\nbased on visualization tasks associated with three categories of data: single\nfields, time-varying fields, and ensembles. These tasks include symmetry\ndetection, periodicity detection, key event/feature detection, feature\ntracking, clustering, and structure statistics. Our main contributions include\nthe formulation of a set of desirable mathematical and computational properties\nof comparative measures, and the classification of visualization tasks and\napplications that are enabled by these measures.\n"}
{"abstract": "  In their papers, Lu et al. analysed experimental data from in-situ\ntransmission electron microscopy (TEM) and concluded that local stress\nconcentration at grain boundary-twin boundary (GB-TB) intersections in\nnanotwinned (nt-) metals increases as the TB thicknesses is reduced (below\n${\\lambda}$ $\\leq$ 18 nm). However, here, by analysing the raw experimental\ndata in Lu et al. together with finite element simulations and dimensional\nanalysis, we show that the measurement of local stress concentration in a\nnt-material is highly sensitive to the adopted averaging box size due to the\nhigh stress gradient near GB-TB intersections as well as to the TB location in\nthe nt-grain. By careful analysis of the raw data, a completely reverse trend,\nwhich is consistent with well-established laws of continuum mechanics, can be\ncaptured.\n"}
{"abstract": "  The mobile robotic testbed is an essential and critical support to verify the\neffectiveness of mobile robotics research. This paper introduces a novel\nmulti-robot testbed, named Robopheus, which exploits the ideas of\nvirtual-physical modeling in digital-twin. Unlike most existing testbeds, the\ndeveloped Robopheus constructs a bridge that connects the traditional physical\nhardware and virtual simulation testbeds, providing scalable, interactive, and\nhigh-fidelity simulations-tests on both sides. Another salient feature of the\nRobopheus is that it enables a new form to learn the actual models from the\nphysical environment dynamically and is compatible with heterogeneous robot\nchassis and controllers. In turn, the virtual world's learned models are\nfurther leveraged to approximate the robot dynamics online on the physical\nside. Extensive experiments demonstrate the extraordinary performance of the\nRobopheus. Significantly, the physical-virtual interaction design increases the\ntrajectory accuracy of a real robot by 300%, compared with that of not using\nthe interaction.\n"}
{"abstract": "  Distance-based regression model, as a nonparametric multivariate method, has\nbeen widely used to detect the association between variations in a distance or\ndissimilarity matrix for outcomes and predictor variables of interest. Based on\nit, a pseudo-$F$ statistic which partitions the variation in distance matrices\nis often constructed to achieve the aim. To the best of our knowledge, the\nstatistical properties of the pseudo-$F$ statistic has not yet been well\nestablished in the literature. To fill this gap, we study the asymptotic null\ndistribution of the pseudo-$F$ statistic and show that it is asymptotically\nequivalent to a mixture of chi-squared random variables. Given that the\npseudo-$F$ test statistic has unsatisfactory power when the correlations of the\nresponse variables are large, we propose a square-root $F$-type test statistic\nwhich replaces the similarity matric with its square root. The asymptotic null\ndistribution of the new test statistic and power of both tests are also\ninvestigated.\n  Simulation studies are conducted to validate the asymptotic distributions of\nthe tests and demonstrate that the proposed test has more robust power than the\npseudo-$F$ test. Both test statistics are exemplified with a gene expression\ndataset for a prostate cancer pathway.\n  Keywords: Asymptotic distribution, Chi-squared-type mixture, Nonparametric\ntest, Pseudo-$F$ test, Similarity matrix.\n"}
{"abstract": "  In general, the laws of physics are not invariant under a change of scale. To\nfind out whether the 'scale-invariance hypothesis' corresponds to nature or\nnot, a careful examination to its implications is required. As a consequence,\nthe scale-invariance cosmological models need to be carefully checked with many\ntests in order to confirm or disconfirm them. In this paper, three different\ntoy models have been introduced in the framework of a scale-invariance\ncosmology to examine dark energy and cosmic transit. Although cosmic transit\nexists in the three models, the pressure stays always negative during cosmic\nevolution. In addition, there is always a singularity in the evolution of the\nequation of state parameter which is not suitable for a complete investigation\nof dark energy evolution. The undesirable features of the parameters have been\ndiscussed, and a comparison with other cosmological contexts has been done.\n"}
{"abstract": "  Particle accelerators are essential tools in science, hospitals and industry.\nYet, their costs and large footprint, ranging in length from meters to several\nkilometres, limit their use. The recently demonstrated nanophotonics-based\nacceleration of charged particles can reduce the cost and size of these\naccelerators by orders of magnitude. In this approach, a carefully designed\nnano-structure transfers energy from laser light to the particles in a\nphase-synchronous manner, thereby accelerating them. However, so far,\nconfinement of the particle beam in the structure over extended distance has\nbeen elusive; it requires complex control of the electron beam phase space and\nis mandatory to accelerate particles to the MeV range and beyond with minimal\nparticle loss. Here, we demonstrate complex electron phase space control at\noptical frequencies in the 225 nanometre narrow channel of a record-long\nphotonic nanostructure. In particular, we experimentally show alternating phase\nfocusing, a particle propagation scheme for, in principle, arbitrarily long\nminimal-loss transport. We expect this work to enable MeV electron beam\ngeneration on a photonic chip, with direct ramification for new forms of\nradiotherapy and compact light sources, and other forms of electron phase space\ncontrol resulting in narrow energy or zeptosecond-bunched beams, for instance.\n"}
{"abstract": "  For a finite group $G$, we consider the problem of counting simultaneous\nconjugacy classes of $n$-tuples and simultaneous conjugacy classes of commuting\n$n$-tuples in $G$. Let $\\alpha_{G,n}$ denote the number of simultaneous\nconjugacy classes of $n$-tuples, and $\\beta_{G,n}$ the number of simultaneous\nconjugacy classes of commuting $n$-tuples in $G$. The generating functions\n$A_G(t) = \\sum_{n\\geq 0} \\alpha_{G,n}t^n,$ and $B_G(t) = \\sum_{n\\geq 0}\n\\beta_{G,n}t^n$ are rational functions of $t$. This paper concern studied of\nnormalized functions $A_G(t/|G|)$ and $B_G(t/|G|)$ for finite $p$-groups of\nrank at most $5$.\n"}
{"abstract": "  Maximum composite likelihood estimation is a useful alternative to maximum\nlikelihood estimation when data arise from data generating processes (DGPs)\nthat do not admit tractable joint specification. We demonstrate that generic\ncomposite likelihoods consisting of marginal and conditional specifications\npermit the simple construction of composite likelihood ratio-like statistics\nfrom which finite-sample valid confidence sets and hypothesis tests can be\nconstructed. These statistics are universal in the sense that they can be\nconstructed from any estimator for the parameter of the underlying DGP. We\ndemonstrate our methodology via a simulation study using a pair of\nconditionally specified bivariate models.\n"}
{"abstract": "  Johnson--Lindenstrauss Transforms are powerful tools for reducing the\ndimensionality of data while preserving key characteristics of that data, and\nthey have found use in many fields from machine learning to differential\nprivacy and more. This note explains what they are; it gives an overview of\ntheir use and their development since they were introduced in the 1980s; and it\nprovides many references should the reader wish to explore these topics more\ndeeply.\n"}
{"abstract": "  Let $K$ be a quartic CM field, that is, a totally imaginary quadratic\nextension of a real quadratic number field. In a 1962 article titled On the\nclassfields obtained by complex multiplication of abelian varieties, Shimura\nconsidered a particular family $\\{F_K(m) : m \\in \\mathbb{Z} >0 \\}$ of abelian\nextensions of $K$, and showed that the Hilbert class field $H_K$ of $K$ is\ncontained in $F_K(m)$ for some positive integer m. We make this m explicit. We\nthen give an algorithm that computes a set of defining polynomials for the\nHilbert class field using the field $F_K(m)$. Our proof-of-concept\nimplementation of this algorithm computes a set of defining polynomials much\nfaster than current implementations of the generic Kummer algorithm for certain\nexamples of quartic CM fields.\n"}
{"abstract": "  Minimization of a stochastic cost function is commonly used for approximate\nsampling in high-dimensional Bayesian inverse problems with Gaussian prior\ndistributions and multimodal posterior distributions. The density of the\nsamples generated by minimization is not the desired target density, unless the\nobservation operator is linear, but the distribution of samples is useful as a\nproposal density for importance sampling or for Markov chain Monte Carlo\nmethods. In this paper, we focus on applications to sampling from multimodal\nposterior distributions in high dimensions. We first show that sampling from\nmultimodal distributions is improved by computing all critical points instead\nof only minimizers of the objective function. For applications to\nhigh-dimensional geoscience problems, we demonstrate an efficient approximate\nweighting that uses a low-rank Gauss-Newton approximation of the determinant of\nthe Jacobian. The method is applied to two toy problems with known posterior\ndistributions and a Darcy flow problem with multiple modes in the posterior.\n"}
{"abstract": "  In the domain of Autonomic and Organic Computing, the entities of a\ndistributed system are variable as well as the efficiency and the intention of\ntheir work. Therefore, a scalable mechanism to incentivise/sanction entities\nwhich contribute towards/against the system goal is needed. Trust is a suited\nmetric to find benevolent entities. In this paper, we focus for one on the\nSimTrust model which introduces trust on entities when they share interest and\nopinions using tagging information. The second model is the Weighted Simple\nExponential Smoothing Trust metric (WSES) which functions on explicitly rated\nitems. WSES follows two basic rules which ensure a logic rating mechanism. When\nputting these two models in context, SimTrust has advantages on items that have\nnot been rated yet or can not easily be rated. WSES is a trust metric which\nreturns good results on explicit rank values. We propose concepts on combining\nboth approaches and state in which cases they are incompatible.\n"}
{"abstract": "  The simple structure of doubly torqued vectors allows for a natural\ncharacterization of doubly twisted down to warped spacetimes, as well as Kundt\nspacetimes down to PP waves. For the first ones the vectors are timelike, for\nthe others they are null. We also discuss some properties, and their connection\nto hypersurface orthogonal conformal Killing vectors, and null Killing vectors.\n"}
{"abstract": "  Current computer graphics research practices contain racial biases that have\nresulted in investigations into \"skin\" and \"hair\" that focus on the hegemonic\nvisual features of Europeans and East Asians. To broaden our research horizons\nto encompass all of humanity, we propose a variety of improvements to\nquantitative measures and qualitative practices, and pose novel, open research\nproblems.\n"}
{"abstract": "  Active cell-junction remodeling is important for tissue morphogenesis, yet\nits underlying physics is not understood. We study a mechanical model that\ndescribes junctions as dynamic active force dipoles. Their instability can\ntrigger cell intercalations by a critical collapse. Nonlinearities in tissue's\nelastic response can stabilize the collapse either by a limit cycle or\ncondensation of junction lengths at cusps of the energy landscape. Furthermore,\nactive junction networks undergo collective instability to drive active\nin-plane ordering or develop a limit cycle of collective oscillations, which\nextends over regions of the energy landscape corresponding to distinct network\ntopologies.\n"}
{"abstract": "  We determine the possible scaling limits in the quantum central limit theorem\nwith respect to the Gibbs state, for a growing distance-regular graph that has\nso-called classical parameters with base unequal to one. We also describe\nexplicitly the corresponding weak limits of the normalized spectral\ndistribution of the adjacency matrix. We demonstrate our results with the known\ninfinite families of distance-regular graphs having classical parameters and\nwith unbounded diameter.\n"}
{"abstract": "  Let $A$ be a rational function of one complex variable, and $z_0$ its\nrepelling fixed point with the multiplier $\\lambda.$ A Poincar\\'e function\nassociated with $z_0$ is a function $\\mathcal{P}_{A,z_0,\\lambda}$ meromorphic\non $\\mathbb C$ such that $\\mathcal{P}_{A,z_0,\\lambda}(0)=z_0$,\n$\\mathcal{P}_{A,z_0,\\lambda}'(0)\\neq 0,$ and\n$\\mathcal{P}_{A,z_0,\\lambda}(\\lambda z)=A\\circ \\mathcal{P}_{A,z_0,\\lambda}(z).$\nIn this paper, we investigate the following problem: given Poincar\\'e functions\n$\\mathcal{P}_{A_1,z_1,\\lambda_1}$ and $\\mathcal{P}_{A_2,z_2,\\lambda_2}$, find\nout if there is an algebraic relation\n$f(\\mathcal{P}_{A_1,z_1,\\lambda_1},\\mathcal{P}_{A_2,z_2,\\lambda_2})=0$ between\nthem and, if such a relation exists, describe the corresponding algebraic\ncurve. We provide a solution, which can be viewed as a refinement of the\nclassical Ritt theorem about commuting rational functions. We also complement\nprevious results concerning algebraic dependencies between B\\\"ottcher\nfunctions.\n"}
{"abstract": "  In this paper, we develop a PDE approach to consider the optimal strategy of\nmean field controlled stochastic system. Firstly, we discuss mean field SDEs\nand associated Fokker-Plank eqautions. Secondly, we consider a fully-coupled\nsystem of forward-backward PDEs. The backward one is the\nHamilton-Jacobi-Bellman equation while the forward one is the Fokker-Planck\nequation. Our main result is to show the existence of classical solutions of\nthe forward-backward PDEs in the class\n$H^{1+\\frac{1}{4},2+\\frac{1}{2}}([0,T]\\times\\mathbb{R}^n)$ by use of the\nSchauder fixed point theorem. Then, we use the solution to give the optimal\nstrategy of the mean field stochastic control problem. Finally, we give an\nexample to illustrate the role of our main result.\n"}
{"abstract": "  The equilibrium problem defined by the Nikaid\\^o-Isoda-Fan inequality\ncontains a number of problems such as optimization, variational inequality,\nKakutani fixed point, Nash equilibria, and others as special cases. This paper\npresents a picture for the relationship between the fixed points of the Moreau\nproximal mapping and the solutions of the equilibrium problem that satisfies\nsome kinds of monotonicity and Lipschitz-type condition.\n"}
{"abstract": "  We report the low-temperature properties of SrNd$_2$O$_4$, a geometrically\nfrustrated magnet. Magnetisation and heat capacity measurements performed on\npolycrystalline samples indicate the appearance of a magnetically ordered state\nat $T_{\\rm N}=2.28(4)$~K. Powder neutron diffraction measurements reveal that\nan \\afm\\ state with the propagation vector \\QV\\ is stabilised below this\ntemperature. The magnetic order is incomplete, as only one of the two Nd$^{3+}$\nsites carries a significant magnetic moment while the other site remains\nlargely disordered. The presence of a disordered magnetic component below\n$T_{\\rm N}$ is confirmed with polarised neutron diffraction measurements. In an\napplied magnetic field, the bulk properties measurements indicate a phase\ntransition at about 30~kOe. We construct a tentative $H$-$T$ phase diagram of\n\\sno\\ from these measurements.\n"}
{"abstract": "  Certifiable local robustness, which rigorously precludes small-norm\nadversarial examples, has received significant attention as a means of\naddressing security concerns in deep learning. However, for some classification\nproblems, local robustness is not a natural objective, even in the presence of\nadversaries; for example, if an image contains two classes of subjects, the\ncorrect label for the image may be considered arbitrary between the two, and\nthus enforcing strict separation between them is unnecessary. In this work, we\nintroduce two relaxed safety properties for classifiers that address this\nobservation: (1) relaxed top-k robustness, which serves as the analogue of\ntop-k accuracy; and (2) affinity robustness, which specifies which sets of\nlabels must be separated by a robustness margin, and which can be\n$\\epsilon$-close in $\\ell_p$ space. We show how to construct models that can be\nefficiently certified against each relaxed robustness property, and trained\nwith very little overhead relative to standard gradient descent. Finally, we\ndemonstrate experimentally that these relaxed variants of robustness are\nwell-suited to several significant classification problems, leading to lower\nrejection rates and higher certified accuracies than can be obtained when\ncertifying \"standard\" local robustness.\n"}
{"abstract": "  On the basis of a compartment model, the epidemic curve is investigated when\nthe net rate $\\lambda$ of change of the number of infected individuals $I$ is\ngiven by an ellipse in the $\\lambda$-$I$ plane which is supported in\n$[I_{\\ell}, I_h]$. With $a \\equiv (I_h - I_{\\ell})/(I_h + I_{\\ell})$, it is\nshown that (1) when $a < 1$ or $I_{\\ell} >0$, oscillation of the infection\ncurve is self-organized and the period of the oscillation is in proportion to\nthe ratio of the difference $ (I_h - I_{\\ell})$ and the geometric mean\n$\\sqrt{I_h I_{\\ell}}$ of $I_h$ and $I_{\\ell}$, (2) when $a = 1$, the infection\ncurve shows a critical behavior where it decays obeying a power law function\nwith exponent $-2$ in the long time limit after a peak, and (3) when $a > 1$,\nthe infection curve decays exponentially in the long time limit after a peak.\nThe present result indicates that the pandemic can be controlled by a measure\nwhich makes $I_{\\ell} < 0$.\n"}
{"abstract": "  Logistic regression (LR) is a widely used classification method for modeling\nbinary outcomes in many medical data classification tasks. Research that\ncollects and combines datasets from various data custodians and jurisdictions\ncan excessively benefit from the increased statistical power to support their\nanalyzing goals. However, combining data from these various sources creates\nsignificant privacy concerns that need to be addressed. In this paper, we\nproposed secret sharing-based privacy-preserving logistic regression protocols\nusing the Newton-Raphson method. Our proposed approaches are based on secure\nMulti-Party Computation (MPC) with different security settings to analyze data\nowned by several data holders. We conducted experiments on both synthetic data\nand real-world datasets and compared the efficiency and accuracy of them with\nthose of an ordinary logistic regression model. Experimental results\ndemonstrate that the proposed protocols are highly efficient and accurate. This\nstudy introduces iterative algorithms to simplify the federated training a\nlogistic regression model in a privacy-preserving manner. Our implementation\nresults show that our improved method can handle large datasets used in\nsecurely training a logistic regression from multiple sources.\n"}
{"abstract": "  We determine the relationship between the turnaround radius, $R_{\\rm t}$, and\nmass, $M_{\\rm t}$, in $\\Lambda$CDM, and in dark energy scenarios, using an\nextended spherical collapse model taking into account the effects of shear and\nvorticity. We find a more general formula than that usually described in\nliterature, showing a dependence of $R_{\\rm t}$ from shear, and vorticity. The\n$R_{\\rm t}-M_{\\rm t}$ relation differs from that obtained not taking into\naccount shear, and rotation, especially at galactic scales, differing $\\simeq\n30\\%$ from the result given in literature. This has effects on the constraint\nof the $w$ parameter of the equation of state. We compare the $R_{\\rm t}-M_{\\rm\nt}$ relationship obtained for the $\\Lambda$CDM, and different dark energy\nmodels to that obtained in the $f(R)$ modified gravity (MG) scenario. The\n$R_{\\rm t}-M_{\\rm t}$ relationship in $\\Lambda$CDM, and dark energy scenarios\nare tantamount to the prediction of the $f(R)$ theories. Then, the $R_{\\rm\nt}-M_{\\rm t}$ relationship is not a good probe to test gravity theories beyond\nEinstein's general relativity.\n"}
{"abstract": "  We introduce a new variational principle for the study of eigenvalues and\neigenfunctions of the Laplacians with Neumann and Dirichlet boundary conditions\non planar domains and apply it to the famous hot spots conjecture, which we\nprove for a large class of simply connected, not necessarily convex domains. We\nshow that each eigenfunction of the Neumann Laplacian corresponding to the\nlowest positive eigenvalue is strictly monotonous along certain directions if\nthe domain faces two opposite quadrants. In particular, the maximum and minimum\nof such an eigenfunction may only be located on the boundary.\n"}
{"abstract": "  An Objective Structured Practical Examination (OSPE) is an effective and\nrobust, but resource-intensive, means of evaluating anatomical knowledge. Since\nmost OSPEs employ short answer or fill-in-the-blank style questions, the format\nrequires many people familiar with the content to mark the exams. However, the\nincreasing prevalence of online delivery for anatomy and physiology courses\ncould result in students losing the OSPE practice that they would receive in\nface-to-face learning sessions. The purpose of this study was to test the\naccuracy of Decision Trees (DTs) in marking OSPE questions as a potential first\nstep to creating an intelligent, online OSPE tutoring system. The study used\nthe results of the winter 2020 semester final OSPE from McMaster University's\nanatomy and physiology course in the Faculty of Health Sciences (HTHSCI\n2FF3/2LL3/1D06) as the data set. Ninety percent of the data set was used in a\n10-fold validation algorithm to train a DT for each of the 54 questions. Each\nDT was comprised of unique words that appeared in correct, student-written\nanswers. The remaining 10% of the data set was marked by the generated DTs.\nWhen the answers marked by the DT were compared to the answers marked by staff\nand faculty, the DT achieved an average accuracy of 94.49% across all 54\nquestions. This suggests that machine learning algorithms such as DTs are a\nhighly effective option for OSPE grading and are suitable for the development\nof an intelligent, online OSPE tutoring system.\n"}
{"abstract": "  The choice of activation functions is crucial for modern deep neural\nnetworks. Popular hand-designed activation functions like Rectified Linear\nUnit(ReLU) and its variants show promising performance in various tasks and\nmodels. Swish, the automatically discovered activation function, has been\nproposed and outperforms ReLU on many challenging datasets. However, it has two\nmain drawbacks. First, the tree-based search space is highly discrete and\nrestricted, which is difficult for searching. Second, the sample-based\nsearching method is inefficient, making it infeasible to find specialized\nactivation functions for each dataset or neural architecture. To tackle these\ndrawbacks, we propose a new activation function called Piecewise Linear\nUnit(PWLU), which incorporates a carefully designed formulation and learning\nmethod. It can learn specialized activation functions and achieves SOTA\nperformance on large-scale datasets like ImageNet and COCO. For example, on\nImageNet classification dataset, PWLU improves 0.9%/0.53%/1.0%/1.7%/1.0% top-1\naccuracy over Swish for\nResNet-18/ResNet-50/MobileNet-V2/MobileNet-V3/EfficientNet-B0. PWLU is also\neasy to implement and efficient at inference, which can be widely applied in\nreal-world applications.\n"}
{"abstract": "  Neuroradiologists and neurosurgeons increasingly opt to use functional\nmagnetic resonance imaging (fMRI) to map functionally relevant brain regions\nfor noninvasive presurgical planning and intraoperative neuronavigation. This\napplication requires a high degree of spatial accuracy, but the fMRI\nsignal-to-noise ratio (SNR) decreases as spatial resolution increases. In\npractice, fMRI scans can be collected at multiple spatial resolutions, and it\nis of interest to make more accurate inference on brain activity by combining\ndata with different resolutions. To this end, we develop a new Bayesian model\nto leverage both better anatomical precision in high resolution fMRI and higher\nSNR in standard resolution fMRI. We assign a Gaussian process prior to the mean\nintensity function and develop an efficient, scalable posterior computation\nalgorithm to integrate both sources of data. We draw posterior samples using an\nalgorithm analogous to Riemann manifold Hamiltonian Monte Carlo in an expanded\nparameter space. We illustrate our method in analysis of presurgical fMRI data,\nand show in simulation that it infers the mean intensity more accurately than\nalternatives that use either the high or standard resolution fMRI data alone.\n"}
{"abstract": "  We consider the homogeneous and inhomogeneous Landau equation for very soft\nand Coulomb potentials and show that approximate Type I self-similar blow-up\nsolutions do not exist under mild decay assumptions on the profile. We extend\nour analysis to the Vlasov-Poisson-Landau system and to the Boltzmann equation\nwithout angular cut-off.\n"}
{"abstract": "  Recent experiments demonstrating storage of optical pulses in acoustic\nphonons based on stimulated Brillouin scattering raise a number of questions\nabout the spectral and temporal capacities of such protocols and the\nlimitations of the theoretical frameworks routinely used to describe them. In\nthis work, we consider the dynamics of photon-phonon scattering induced by\noptical pulses with temporal widths comparable to the period of acoustic\noscillations. We revisit the widely adopted classical formalism of coupled\nmodes and demonstrate its breakdown. We propose a simple extension to\ngeneralise the formulation and find potentially measurable consequences in the\ndynamics of Brillouin experiments involving ultra-short pulses.\n"}
{"abstract": "  In this paper we study the geometric structure of affine Deligne-Lusztig\nvarieties for $GL_3$ and $b$ basic. We completely determine the irreducible\ncomponents of the affine Deligne-Lusztig variety. In particular, we classify\nthe cases where all of the irreducible components are classical Deligne-Lusztig\nvarieties times finite-dimensional affine spaces. If this is the case, then the\nirreducible components are pairwise disjoint.\n"}
{"abstract": "  The amplification of evanescent waves by flat superlens requires a\nnear-resonance coupling which has been linked to resonant surface plasmons. A\nsubtle interplay has been proposed to exist between the excitation of\nwell-defined resonant surface plasmons and the focusing capability of a\nsuperlens. To gain insights into these resonant modes and their contributions\nto the amplification and recovery of evanescent waves, we performed simple but\nrobust full-wave FDTD simulations on causal negative index Lorentz models. We\nfound that well-defined pair of resonant surface plasmons is excited whenever a\ncoupling of a diverging transmitted beam and a converging refracted beam occurs\nat the second interface. The resonant coupling of these modes at the interface\nled to the excitation of a single interface resonance predicted by the theory.\nThe physical consequence of this resonance is that incident wave energy is\npumped into the negative-index material medium and beyond it. These phenomena\ncontribute substantially to the amplification and recovery of the near-fields\nin the image plane. It is noteworthy that, for evanescent wave refocusing to be\nachieved in a flat superlens, its thickness, and the source-to-superlens\ndistance should be optimized. This optimization is critical in that an optimal\nthickness alongside optimal source-to-superlens distance will allow evanescent\nwave refocusing to dominate material lost. The FDTD simulated result showed an\nimage of the point source inside the superlens and beyond it when its thickness\nwas optimized. The resolution of the image beyond the superlens was\n$\\sim.58\\lambda$ and this superlens behaves like a near-perfect lens. Overall,\nthese numerically simulated results could serve as useful approximations to the\ndegree of resonant amplification and refocusing of near-fields that can be\nachieved by flat superlens in near-fields experimental imaging setups.\n"}
{"abstract": "  Ontology Alignment is an important research problem applied to various fields\nsuch as data integration, data transfer, data preparation, etc.\nState-of-the-art (SOTA) Ontology Alignment systems typically use naive\ndomain-dependent approaches with handcrafted rules or domain-specific\narchitectures, making them unscalable and inefficient. In this work, we propose\nVeeAlign, a Deep Learning based model that uses a novel dual-attention\nmechanism to compute the contextualized representation of a concept which, in\nturn, is used to discover alignments. By doing this, not only is our approach\nable to exploit both syntactic and semantic information encoded in ontologies,\nit is also, by design, flexible and scalable to different domains with minimal\neffort. We evaluate our model on four different datasets from different domains\nand languages, and establish its superiority through these results as well as\ndetailed ablation studies. The code and datasets used are available at\nhttps://github.com/Remorax/VeeAlign.\n"}
{"abstract": "  In this paper, we prove two rigidity results for non-positively curved\nhomogeneous Finsler metrics. Our first main result yields an extension of\nHu-Deng's well-known result proven for the Randers metrics. Indeed, we prove\nthat every connected homogeneous Finsler space with non-positive flag curvature\nand isotropic S-curvature is Riemannian or locally Minkowskian. We extend the\nSzab\\'{o}'s rigidity theorem for Berwald surfaces and show that homogeneous\nisotropic Berwald metrics with non-positive flag curvature are Riemannian or\nlocally Minkowskian. We prove that a homogeneous $(\\alpha, \\beta)$-metrics has\nisotropic mean Berwald curvature if and only if it has vanishing mean Berwald\ncurvature generalizing result previously only known in the case of Randers\nmetrics. Our second main result is to show that every homogeneous\n$(\\alpha,\\beta)$-metric with non-positive flag curvature and almost isotropic\nS-curvature is Riemannian or locally Minkowskian.\n"}
{"abstract": "  This paper examines the short- and long-run effects of U.S. federal personal\nincome and corporate income tax cuts on a wide array of economic policy\nvariables in a data-rich environment. Using a panel of U.S. macroeconomic data\nset, made up of 132 quarterly macroeconomic series for 1959-2018, the study\nestimates factor-augmented vector autoregression (FAVARs) models where an\nextended narrative tax changes dataset combined with unobserved factors. The\nnarrative approach classifies if tax changes are exogenous or endogenous. This\npaper identifies narrative tax shocks in the vector autoregression model using\nthe sign restrictions with Uhlig's (2005) penalty function. Empirical findings\nshow a significant expansionary effect of tax cuts on the macroeconomic\nvariables. Cuts in personal and corporate income taxes cause a rise in output,\ninvestment, employment, and consumption; however, cuts in personal taxes appear\nto be a more effective fiscal policy tool than the cut in corporate income\ntaxes. Real GDP, employment, investment, and industrial production increase\nsignificantly and reach their maximum response values two years after personal\nincome tax cuts. The effects of corporate tax cuts have relatively smaller\neffects on output and consumption but show immediate and higher effects on\nfixed investment and price levels.\n"}
{"abstract": "  The classical Kubo-Martin-Schwinger (KMS) condition is a fundamental property\nof statistical mechanics characterizing the equilibrium of infinite classical\nmechanical systems. It was introduced in the seventies by G. Gallavotti and E.\nVerboven as an alternative to the Dobrushin-Lanford-Ruelle (DLR) equation. In\nthis article, we consider this concept in the framework of nonlinear\nHamiltonian PDEs and discuss its relevance. In particular, we prove that Gibbs\nmeasures are the unique KMS equilibrium states for such systems. Our proof is\nbased on Malliavin calculus and Gross-Sobolev spaces. The main feature of our\nwork is the applicability of our results to the general context of white noise,\nabstract Wiener spaces and Gaussian probability spaces, as well as to\nfundamental examples of PDEs like the nonlinear Schrodinger, Hartree, and wave\n(Klein-Gordon) equations.\n"}
{"abstract": "  Stochastic simulations such as large-scale, spatiotemporal, age-structured\nepidemic models are computationally expensive at fine-grained resolution. We\npropose Interactive Neural Process (INP), a Bayesian active learning framework\nto proactively learn a deep learning surrogate model and accelerate simulation.\nOur framework is based on the novel integration of neural process, deep\nsequence model and active learning. In particular, we develop a novel\nspatiotemporal neural process model to mimic the simulator dynamics. Our model\nautomatically infers the latent process which describes the intrinsic\nuncertainty of the simulator. This also gives rise to a new acquisition\nfunction based on the latent information gain. We design Bayesian active\nlearning algorithms to iteratively query the simulator, gather more data, and\ncontinuously improve the model. We perform theoretical analysis and demonstrate\nthat our approach reduces sample complexity compared with random sampling in\nhigh dimension. Empirically, we demonstrate our framework can faithfully\nimitate the behavior of a complex infectious disease simulator with a small\nnumber of examples, enabling rapid simulation and scenario exploration.\n"}
{"abstract": "  When compared to fault-tolerant quantum computational strategies, variational\nquantum algorithms stand as one of the candidates with the potential of\nachieving quantum advantage for real-world applications in the near term.\nHowever, the optimization of the circuit parameters remains arduous and is\nimpeded by many obstacles such as the presence of barren plateaus, many local\nminima in the optimization landscape, and limited quantum resources. A\nnon-random initialization of the parameters seems to be key to the success of\nthe parametrized quantum circuits (PQC) training. Drawing and extending ideas\nfrom the field of meta-learning, we address this parameter initialization task\nwith the help of machine learning and propose FLIP: a FLexible Initializer for\narbitrarily-sized Parametrized quantum circuits. FLIP can be applied to any\nfamily of PQCs, and instead of relying on a generic set of initial parameters,\nit is tailored to learn the structure of successful parameters from a family of\nrelated problems which are used as the training set. The flexibility advocated\nto FLIP hinges in the possibility of predicting the initialization of\nparameters in quantum circuits with a larger number of parameters from those\nused in the training phase. This is a critical feature lacking in other\nmeta-learning parameter initializing strategies proposed to date. We illustrate\nthe advantage of using FLIP in three scenarios: a family of problems with\nproven barren plateaus, PQC training to solve max-cut problem instances, and\nPQC training for finding the ground state energies of 1D Fermi-Hubbard models.\n"}
{"abstract": "  Micron-scale randomness during manufacturing can ensure anti-counterfeiting\nlabels are unclonable. However, this security typically comes at the expense of\ncomplex hardware being needed for authentication (e.g., microscopy systems). We\ndemonstrate unclonable labels that can be authenticated using a standard\nlight-emitting diode and smartphone camera. The labels consist of a microlens\narray laminated to a polymer film that is doped with luminescent\nmicroparticles. The micron-scale random overlap of focal volumes and\nmicroparticles leads to a pattern of bright points of visible light emission\nthat can be easily imaged by a smartphone camera. 10 000 comparisons of images\ndemonstrate that the labels can be robustly authenticated, and that the\nprobability of a false authentication is on the order of $10^{-15}$. The\nability for microlens arrays to simplify the hardware needed for authentication\nof unclonable labels is generalizable, and attractive for the implementation of\nunclonable labels in anti-counterfeiting systems.\n"}
{"abstract": "  A virtual flow meter (VFM) enables continuous prediction of flow rates in\npetroleum production systems. The predicted flow rates may aid the daily\ncontrol and optimization of a petroleum asset. Gray-box modeling is an approach\nthat combines mechanistic and data-driven modeling. The objective is to create\na computationally feasible VFM for use in real-time applications, with high\nprediction accuracy and scientifically consistent behavior. This article\ninvestigates five different gray-box model types in an industrial case study\nusing real, historical production data from 10 petroleum wells, spanning at\nmost four years of production. The results are diverse with an oil flow rate\nprediction error in the range of 1.8%-40.6%. Further, the study casts light\nupon the nontrivial task of balancing learning from both physics and data.\nConsequently, providing general recommendations towards the suitability of\ndifferent hybrid models is challenging. Nevertheless, the results are promising\nand indicate that gray-box VFMs may reduce the prediction error of a\nmechanistic VFM while remaining scientifically consistent. The findings\nmotivate further experimentation with gray-box VFM models and suggest several\nfuture research directions to improve upon the performance and scientific\nconsistency.\n"}
{"abstract": "  Person re-identification (re-ID) tackles the problem of matching person\nimages with the same identity from different cameras. In practical\napplications, due to the differences in camera performance and distance between\ncameras and persons of interest, captured person images usually have various\nresolutions. We name this problem as Cross-Resolution Person Re-identification\nwhich brings a great challenge for matching correctly. In this paper, we\npropose a Deep High-Resolution Pseudo-Siamese Framework (PS-HRNet) to solve the\nabove problem. Specifically, in order to restore the resolution of\nlow-resolution images and make reasonable use of different channel information\nof feature maps, we introduce and innovate VDSR module with channel attention\n(CA) mechanism, named as VDSR-CA. Then we reform the HRNet by designing a novel\nrepresentation head to extract discriminating features, named as HRNet-ReID. In\naddition, a pseudo-siamese framework is constructed to reduce the difference of\nfeature distributions between low-resolution images and high-resolution images.\nThe experimental results on five cross-resolution person datasets verify the\neffectiveness of our proposed approach. Compared with the state-of-the-art\nmethods, our proposed PS-HRNet improves 3.4\\%, 6.2\\%, 2.5\\%,1.1\\% and 4.2\\% at\nRank-1 on MLR-Market-1501, MLR-CUHK03, MLR-VIPeR, MLR-DukeMTMC-reID, and CAVIAR\ndatasets, respectively. Our code is available at\n\\url{https://github.com/zhguoqing}.\n"}
{"abstract": "  We investigate the spectral properties of one-dimensional lattices with\nposition-dependent hopping amplitudes and on-site potentials that are smooth\nbounded functions of position. We find an exact integral form for the density\nof states (DOS) in the limit of an infinite number of sites, which we derive\nusing a mixed Bloch-Wannier basis consisting of piecewise Wannier functions.\nNext, we provide an exact solution for the inverse problem of constructing the\nposition-dependence of hopping in a lattice model yielding a given DOS. We\nconfirm analytic results by comparing them to numerics obtained by exact\ndiagonalization for various incarnations of position-dependent hoppings and\non-site potentials. Finally, we generalize the DOS integral form to\nmulti-orbital tight-binding models with longer-range hoppings and in higher\ndimensions.\n"}
{"abstract": "  The semidirect product of the real Heisenberg group ${\\rm H}_1(\\mathbb{R})$\nwith ${\\rm SL}(2,\\mathbb{R})$, called the real Jacobi group\n$G^J_1(\\mathbb{R})$, admits a four-parameter invariant metric expressed in the\nS-coordinates. We determine the geodesic equations on the extended\nSiegel--Jacobi upper half-plane $\\tilde{\\mathcal{X}}^J_1\n=\\frac{G^J_1(\\R)}{\\rm{SO}(2)}\\approx\\mathcal{X}^J_1\\times\\mathbb{R}\\approx\n\\mathcal{X}_1 \\times\\mathbb{R}^3$, where $\\mathcal{X}^J_1$ ($\\mathcal{X}_1)$\ndenotes the Siegel-Jacobi upper half-plane (respectively Siegel upper\nhalf-plane). Equating successively with zero the values of the three parameters\nin the geodesic equations on $\\tilde{\\mathcal{X}}^J_1$, we get the geodesic\nequations on $\\mathcal{X}^J_1$, $\\mathcal{X}_1$ and ${\\rm H}_1(\\mathbb{R})$.\n"}
{"abstract": "  We propose a method for amplifying a train of sub-femtosecond pulses of\ncircularly or elliptically polarized extreme ultraviolet (XUV) radiation\nconstituted by high-order harmonics of an infrared (IR) laser field, in a\nneon-like active medium of a plasma-based X-ray laser, additionally irradiated\nwith a replica of a fundamental frequency IR field. It is shown that the\nellipticity of the pulses can be maintained or increased during the\namplification process. The experimental implementation is suggested in an\nactive medium of an X-ray laser based on neon-like Ti^{12+} ions irradiated by\nan IR laser field with a wavelength of 3.9 microns.\n"}
{"abstract": "  In this article, we investigate the problem of engineering synchronization in\nnon-Markovian quantum systems. First, a time-convoluted linear quantum\nstochastic differential equation is derived which describes the Heisenberg\nevolution of a localized quantum system driven by multiple colored noise\ninputs. Then, we define quantum expectation synchronization in an augmented\nsystem consisting of two subsystems. We prove that, for two homogenous\nsubsystems, synchronization can always be synthesized without designing direct\nHamiltonian coupling given that the degree of non-Markovianity is below a\ncertain threshold. System parameters are explicitly designed to achieve quantum\nsynchronization. Also, a numerical example is presented to illustrate our\nresults.\n"}
{"abstract": "  We present DeepIA, a deep neural network (DNN) framework for enabling fast\nand reliable initial access for AI-driven beyond 5G and 6G millimeter (mmWave)\nnetworks. DeepIA reduces the beam sweep time compared to a conventional\nexhaustive search-based IA process by utilizing only a subset of the available\nbeams. DeepIA maps received signal strengths (RSSs) obtained from a subset of\nbeams to the beam that is best oriented to the receiver. In both line of sight\n(LoS) and non-line of sight (NLoS) conditions, DeepIA reduces the IA time and\noutperforms the conventional IA's beam prediction accuracy. We show that the\nbeam prediction accuracy of DeepIA saturates with the number of beams used for\nIA and depends on the particular selection of the beams. In LoS conditions, the\nselection of the beams is consequential and improves the accuracy by up to 70%.\nIn NLoS situations, it improves accuracy by up to 35%. We find that, averaging\nmultiple RSS snapshots further reduces the number of beams needed and achieves\nmore than 95% accuracy in both LoS and NLoS conditions. Finally, we evaluate\nthe beam prediction time of DeepIA through embedded hardware implementation and\nshow the improvement over the conventional beam sweeping.\n"}
{"abstract": "  The study of hard-particle packings is of fundamental importance in physics,\nchemistry, cell biology, and discrete geometry. Much of the previous work on\nhard-particle packings concerns their densest possible arrangements. By\ncontrast, we examine kinetic effects inevitably present in both numerical and\nexperimental packing protocols. Specifically, we determine how changing the\ncompression/shear rate of a two-dimensional packing of noncircular particles\ncauses it to deviate from its densest possible configuration, which is always\nperiodic. The adaptive shrinking cell (ASC) optimization scheme maximizes the\npacking fraction of a hard-particle packing by first applying random\ntranslations and rotations to the particles and then isotropically compressing\nand shearing the simulation box repeatedly until a possibly jammed state is\nreached. We use a stochastic implementation of the ASC optimization scheme to\nmimic different effective time scales by varying the number of particle moves\nbetween compressions/shears. We generate dense, effectively jammed,\nmonodisperse, two-dimensional packings of obtuse scalene triangle, rhombus,\ncurved triangle, lens, and \"ice cream cone\" (a semicircle grafted onto an\nisosceles triangle) shaped particles, with a wide range of packing fractions\nand degrees of order. To quantify these kinetic effects, we introduce the\nkinetic frustration index $K$, which measures the deviation of a packing from\nits maximum possible packing fraction. To investigate how kinetics affect\nshort- and long-range ordering in these packings, we compute their spectral\ndensities and characterize their contact networks. We find that kinetic effects\nare most significant when the particles have greater asphericity, less\ncurvature, and less rotational symmetry. This work may be relevant to the\ndesign of laboratory packing protocols.\n"}
{"abstract": "  As a projection-free algorithm, Frank-Wolfe (FW) method, also known as\nconditional gradient, has recently received considerable attention in the\nmachine learning community. In this dissertation, we study several topics on\nthe FW variants for scalable projection-free optimization.\n  We first propose 1-SFW, the first projection-free method that requires only\none sample per iteration to update the optimization variable and yet achieves\nthe best known complexity bounds for convex, non-convex, and monotone\nDR-submodular settings. Then we move forward to the distributed setting, and\ndevelop Quantized Frank-Wolfe (QFW), a general communication-efficient\ndistributed FW framework for both convex and non-convex objective functions. We\nstudy the performance of QFW in two widely recognized settings: 1) stochastic\noptimization and 2) finite-sum optimization. Finally, we propose Black-Box\nContinuous Greedy, a derivative-free and projection-free algorithm, that\nmaximizes a monotone continuous DR-submodular function over a bounded convex\nbody in Euclidean space.\n"}
{"abstract": "  We propose Fit4CAD, a benchmark for the evaluation and comparison of methods\nfor fitting simple geometric primitives in point clouds representing CAD\nobjects. This benchmark is meant to help both method developers and those who\nwant to identify the best performing tools. The Fit4CAD dataset is composed by\n225 high quality point clouds, each of which has been obtained by sampling a\nCAD object. The way these elements were created by using existing platforms and\ndatasets makes the benchmark easily expandable. The dataset is already split\ninto a training set and a test set. To assess performance and accuracy of the\ndifferent primitive fitting methods, various measures are defined. To\ndemonstrate the effective use of Fit4CAD, we have tested it on two methods\nbelonging to two different categories of approaches to the primitive fitting\nproblem: a clustering method based on a primitive growing framework and a\nparametric method based on the Hough transform.\n"}
{"abstract": "  We investigate the effects of non-Hermiticity on interacting fermionic\nsystems. We do this by including non-Hermitian bilinear terms into the 3+1\ndimensional Nambu--Jona-Lasinio (NJL) model. Two possible bilinear\nmodifications give rise to $\\mathcal{PT}$ symmetric theories; this happens when\nthe standard NJL model is extended either by a pseudovector background field\n$ig \\bar\\psi\\gamma_5 B_\\mu \\gamma^\\mu \\psi$ or by an antisymmetric-tensor\nbackground field $g \\bar\\psi F_{\\mu\\nu}\\gamma^\\mu \\gamma^\\nu \\psi$. The three\nremaining bilinears are {\\it anti}-$\\mathcal{PT}$-symmetric in nature, $ig\n\\bar\\psi B_\\mu \\gamma^\\mu \\psi, ig\\bar\\psi \\gamma_5 \\psi$ and $ig\\bar\\psi\n{1}\\psi$, so that the Hamiltonian then has no overall symmetry. The\npseudovector $ig \\bar\\psi\\gamma_5 B_\\mu \\gamma^\\mu \\psi$ and the vector $ig\n\\bar\\psi B_\\mu \\gamma^\\mu \\psi$ combinations, are, in addition, chirally\nsymmetric. Thus, within this framework we are able to examine the effects that\nthe various combinations of non-Hermiticity, $\\mathcal{PT}$ symmetry, chiral\nsymmetry and the two-body interactions of the NJL model have on the existence\nand dynamical generation of a real effective fermion mass (a feature which is\nabsent in the corresponding modified massless free Dirac models) as well as on\nthe masses of the composite particles, the pseudoscalar and scalar mesonic\nmodes ($\\pi$ and $\\sigma$ mesons). Our findings demonstrate that $\\mathcal{PT}$\nsymmetry is not necessary for real fermion mass solutions to exist, rather the\ntwo-body interactions of the NJL model supersede the non-Hermitian bilinear\neffects. The effects of chiral symmetry are evident most clearly in the meson\nmodes, the pseudoscalar of which will always be Goldstone in nature if the\nsystem is chirally symmetric. Second solutions of the mesonic equations are\nalso discussed.\n"}
{"abstract": "  Bayesian optimization (BO) is a popular paradigm for global optimization of\nexpensive black-box functions, but there are many domains where the function is\nnot completely black-box. The data may have some known structure, e.g.\nsymmetries, and the data generation process can yield useful intermediate or\nauxiliary information in addition to the value of the optimization objective.\nHowever, surrogate models traditionally employed in BO, such as Gaussian\nProcesses (GPs), scale poorly with dataset size and struggle to incorporate\nknown structure or auxiliary information. Instead, we propose performing BO on\ncomplex, structured problems by using Bayesian Neural Networks (BNNs), a class\nof scalable surrogate models that have the representation power and flexibility\nto handle structured data and exploit auxiliary information. We demonstrate BO\non a number of realistic problems in physics and chemistry, including topology\noptimization of photonic crystal materials using convolutional neural networks,\nand chemical property optimization of molecules using graph neural networks. On\nthese complex tasks, we show that BNNs often outperform GPs as surrogate models\nfor BO in terms of both sampling efficiency and computational cost.\n"}
{"abstract": "  The use of quantitative broadband echosounders for biological studies and\nsurveys offers considerable advantages over narrowband echosounders. These\ninclude improved spectral-based target identification and significantly\nincreased ability to resolve individual targets. Biological studies and surveys\ntypically require accurate measures of backscatter strength and we present here\na systematic and comprehensive explanation of how to derive quantitative\nestimates of target strength and volume backscattering, as a function of\nfrequency from broadband echosounder signals.\n"}
{"abstract": "  It is a common practice to evaluate probability density function or matter\nspatial density function from statistical samples. Kernel density estimation is\na frequently used method, but to select an optimal bandwidth of kernel\nestimation, which is completely based on data samples, is a long-term issue\nthat has not been well settled so far. There exist analytic formulae of optimal\nkernel bandwidth, but they cannot be applied directly to data samples, since\nthey depend on the unknown underlying density functions from which the samples\nare drawn. In this work, we devise an approach to pick out the totally\ndata-based optimal bandwidth. First, we derive correction formulae for the\nanalytic formulae of optimal bandwidth to compute the roughness of the sample's\ndensity function. Then substitute the correction formulae into the analytic\nformulae for optimal bandwidth, and through iteration, we obtain the sample's\noptimal bandwidth. Compared with analytic formulae, our approach gives very\ngood results, with relative differences from the analytic formulae being only\n2%-3% for a sample size larger than 10^4. This approach can also be generalized\neasily to cases of variable kernel estimations.\n"}
{"abstract": "  We review spectral theory of soliton gases in integrable dispersive\nhydrodynamic systems. We first present a phenomenological approach based on the\nconsideration of phase shifts in pairwise soliton collisions and leading to the\nkinetic equation for a non-equilibrium soliton gas. Then a more detailed theory\nis presented in which soliton gas dynamics are modelled by a thermodynamic type\nlimit of modulated finite-gap spectral solutions of the Korteweg-de Vries and\nthe focusing nonlinear Schr\\\"odinger equations. For the focusing nonlinear\nSchr\\\"odinger equation the notions of soliton condensate and breather gas are\nintroduced that are related to the phenomena of spontaneous modulational\ninstability and the rogue wave formation. Integrability properties of the\nkinetic equation for soliton gas are discussed and some physically relevant\nsolutions are presented and compared with direct numerical simulations of\ndispersive hydrodynamic systems.\n"}
{"abstract": "  The Fibonacci cube of dimension n, denoted as $\\Gamma\\_n$, is the subgraph of\nthe hypercube induced by vertices with no consecutive 1's. The irregularity of\na graph G is the sum of |d(x)-d(y)| over all edges {x,y} of G. In two recent\npaper based on the recursive structure of $\\Gamma\\_n$ it is proved that the\nirregularity of $\\Gamma\\_n$ and $\\Lambda\\_n$ are two times the number of edges\nof $\\Gamma\\_{n-1}$ and $2n$ times the number of vertices of $\\Gamma\\_{n-4}$,\nrespectively. Using an interpretation of the irregularity in terms of couples\nof incident edges of a special kind (Figure 2) we give a bijective proof of\nboth results. For these two graphs we deduce also a constant time algorithm for\ncomputing the imbalance of an edge. In the last section using the same approach\nwe determine the number of edges and the sequence of degrees of the cube\ncomplement of $\\Gamma\\_n$.\n"}
{"abstract": "  Normalizing Flows (NFs) are universal density estimators based on Neuronal\nNetworks. However, this universality is limited: the density's support needs to\nbe diffeomorphic to a Euclidean space. In this paper, we propose a novel method\nto overcome this limitation without sacrificing universality. The proposed\nmethod inflates the data manifold by adding noise in the normal space, trains\nan NF on this inflated manifold, and, finally, deflates the learned density.\nOur main result provides sufficient conditions on the manifold and the specific\nchoice of noise under which the corresponding estimator is exact. Our method\nhas the same computational complexity as NFs and does not require computing an\ninverse flow. We also show that, if the embedding dimension is much larger than\nthe manifold dimension, noise in the normal space can be well approximated by\nGaussian noise. This allows to use our method for approximating arbitrary\ndensities on non-flat manifolds provided that the manifold dimension is known.\n"}
{"abstract": "  In Grammatical Error Correction (GEC), sequence labeling models enjoy fast\ninference compared to sequence-to-sequence models; however, inference in\nsequence labeling GEC models is an iterative process, as sentences are passed\nto the model for multiple rounds of correction, which exposes the model to\nsentences with progressively fewer errors at each round. Traditional GEC models\nlearn from sentences with fixed error rates. Coupling this with the iterative\ncorrection process causes a mismatch between training and inference that\naffects final performance. In order to address this mismatch, we propose a\nGAN-like sequence labeling model, which consists of a grammatical error\ndetector as a discriminator and a grammatical error labeler with Gumbel-Softmax\nsampling as a generator. By sampling from real error distributions, our errors\nare more genuine compared to traditional synthesized GEC errors, thus\nalleviating the aforementioned mismatch and allowing for better training. Our\nresults on several evaluation benchmarks demonstrate that our proposed approach\nis effective and improves the previous state-of-the-art baseline.\n"}
{"abstract": "  The Su-Schrieffer-Heeger (SSH) model on a two-dimensional square lattice has\nbeen considered as a significant platform for studying topological multipole\ninsulators. However, due to the highly-degenerate bulk energy bands protected\nby $ C_{4v} $ and chiral symmetry, the discussion of the zero-energy\ntopological corner states and the corresponding physical realization have been\nrarely presented. In this work, by tuning the hopping terms to break $ C_{4v} $\nsymmetry down to $ C_{2v} $ symmetry but with the topological phase invariant,\nwe show that the degeneracies can be removed and a complete band gap can be\nopened, which provides robust protection for the spectrally isolated\nzero-energy corner states. Meanwhile, we propose a rigorous acoustic\ncrystalline insulator and therefore these states can be observed directly. Our\nwork reveals the topological properties of the robust zero-energy states, and\nprovides a new way to explore novel topological phenomena.\n"}
{"abstract": "  We provide two closed-form geodesic formulas for a family of metrics on\nStiefel manifold, parameterized by two positive numbers, having both the\nembedded and canonical metrics as special cases. The closed-form formulas allow\nus to compute geodesics by matrix exponential in reduced dimension for low-rank\nmanifolds. Combining with the use of Fr{\\'e}chet derivatives to compute the\ngradient of the square Frobenius distance between a geodesic ending point to a\ngiven point on the manifold, we show the logarithm map and geodesic distance\nbetween two endpoints on the manifold could be computed by {\\it minimizing}\nthis square distance by a {\\it trust-region} solver. This leads to a new\nframework to compute the geodesic distance for manifolds with known geodesic\nformula but no closed-form logarithm map. We show the approach works well for\nStiefel as well as flag manifolds. The logarithm map could be used to compute\nthe Riemannian center of mass for these manifolds equipped with the above\nmetrics. We also deduce simple trigonometric formulas for the Riemannian\nexponential and logarithm maps on the Grassmann manifold.\n"}
{"abstract": "  In this article, we obtain an effectiveness result of strong openness\nproperty in $L^p$ with some applications.\n"}
{"abstract": "  In this paper, we propose a trust-centric privacy-preserving blockchain for\ndynamic spectrum access in IoT networks. To be specific, we propose a trust\nevaluation mechanism to evaluate the trustworthiness of sensing nodes and\ndesign a Proof-of-Trust (PoT) consensus mechanism to build a scalable\nblockchain with high transaction-per-second (TPS). Moreover, a privacy\nprotection scheme is proposed to protect sensors' real-time geolocatioin\ninformation when they upload sensing data to the blockchain. Two smart\ncontracts are designed to make the whole procedure (spectrum sensing, spectrum\nauction, and spectrum allocation) run automatically. Simulation results\ndemonstrate the expected computation cost of the PoT consensus algorithm for\nreliable sensing nodes is low, and the cooperative sensing performance is\nimproved with the help of trust value evaluation mechanism.In addition,\nincentivization and security are also analyzed, which show that our design not\nonly can encourage nodes' participation, but also resist to many kinds of\nattacks which are frequently encountered in trust-based blockchain systems.\n"}
{"abstract": "  LiDAR-based 3D detection in point cloud is essential in the perception system\nof autonomous driving. In this paper, we present LiDAR R-CNN, a second stage\ndetector that can generally improve any existing 3D detector. To fulfill the\nreal-time and high precision requirement in practice, we resort to point-based\napproach other than the popular voxel-based approach. However, we find an\noverlooked issue in previous work: Naively applying point-based methods like\nPointNet could make the learned features ignore the size of proposals. To this\nend, we analyze this problem in detail and propose several methods to remedy\nit, which bring significant performance improvement. Comprehensive experimental\nresults on real-world datasets like Waymo Open Dataset (WOD) and KITTI dataset\nwith various popular detectors demonstrate the universality and superiority of\nour LiDAR R-CNN. In particular, based on one variant of PointPillars, our\nmethod could achieve new state-of-the-art results with minor cost. Codes will\nbe released at https://github.com/tusimple/LiDAR_RCNN .\n"}
{"abstract": "  In an attempt to prove the Graceful Tree Conjecture, we present two\npropagation of graphs. The first is to propagate graceful graphs, and the\nsecond is to propagate trees from a gracefully labeled tree. The motivation in\npropagating such graphs is to see how graphs behave in the lens of their\nadjacency matrices. Thus, we provide the necessary algorithms for the said\nobjectives.\n"}
{"abstract": "  The next generations of ground-based cosmic microwave background experiments\nwill require polarisation sensitive, multichroic pixels of large focal planes\ncomprising several thousand detectors operating at the photon noise limit. One\napproach to achieve this goal is to couple light from the telescope to a\npolarisation sensitive antenna structure connected to a superconducting\ndiplexer network where the desired frequency bands are filtered before being\nfed to individual ultra-sensitive detectors such as Transition Edge Sensors.\nTraditionally, arrays constituted of horn antennas, planar phased antennas or\nanti-reflection coated micro-lenses have been placed in front of planar antenna\nstructures to achieve the gain required to couple efficiently to the telescope\noptics. In this paper are presented the design concept and a preliminary\nanalysis of the measured performances of a phase-engineered metamaterial\nflat-lenslet. The flat lens design is inherently matched to free space,\navoiding the necessity of an anti-reflection coating layer. It can be\nfabricated lithographically, making scaling to large format arrays relatively\nsimple. Furthermore, this technology is compatible with the fabrication process\nrequired for the production of large-format lumped element kinetic inductance\ndetector arrays which have already demonstrated the required sensitivity along\nwith multiplexing ratios of order 1000 detectors/channel.\n"}
{"abstract": "  Stellar and supernova nucleosynthesis in the first few billion years of the\ncosmic history have set the scene for early structure formation in the\nUniverse, while little is known about their nature. Making use of stellar\nphysical parameters measured by GALAH Data Release 3 with accurate astrometry\nfrom the Gaia EDR3, we have selected $\\sim 100$ old main-sequence turn-off\nstars (ages $\\gtrsim 12$ Gyrs) with kinematics compatible with the Milky Way\nstellar halo population in the Solar neighborhood. Detailed homogeneous\nelemental abundance estimates by GALAH DR3 are compared with supernova yield\nmodels of Pop~III (zero-metal) core-collapse supernovae (CCSNe), normal\n(non-zero-metal) CCSNe, and Type Ia supernovae (SN Ia) to examine which of the\nindividual yields or their combinations best reproduce the observed elemental\nabundance patterns for each of the old halo stars (\"OHS\"). We find that the\nobserved abundances in the OHS with [Fe/H]$>-1.5$ are best explained by\ncontributions from both CCSNe and SN~Ia, where the fraction of SN~Ia among all\nthe metal-enriching SNe is up to 10-20 % for stars with high [Mg/Fe] ratios and\nup to 20-27 % for stars with low [Mg/Fe] ratios, depending on the assumption\nabout the relative fraction of near-Chandrasekhar-mass SNe Ia progenitors. The\nresults suggest that, in the progenitor systems of the OHS with [Fe/H]$>-1.5$,\n$\\sim$ 50-60% of Fe mass originated from normal CCSNe at the earliest phases of\nthe Milky Way formation. These results provide an insight into the birth\nenvironments of the oldest stars in the Galactic halo.\n"}
{"abstract": "  The growing availability of open source projects has facilitated developers\nto reuse existing software artifacts and leverage them to develop new software.\nHowever, it is hard to understand the notion of similarity as it varies from\ndeveloper to developer. Some developers might search for repositories with\nsimilar source code, while some might be in search of repositories with similar\nrequirements or issues. Existing approaches tend to find similar projects by\ncomparing similar artifacts such as source-code to source-code, API usage to\nAPI usage, documentation to documentation, and so on. Even though there is a\ndissimilarity between two similar artifacts, there could be a similarity\nbetween two dissimilar artifacts. Hence, in this paper, we aim to answer the\nquestion - Can we find similarity of software repositories through dissimilar\nartifacts?. To this end, we conduct an experiment to find similarities between\nthree repositories, two similar and one different project comparing similar and\ndissimilar artifacts (documentation, commits, and source-code). We observed\nsimilarities between dissimilar artifacts such as Commits, Source Code, and\nReadme Files in the context of both similar and different repositories.\n"}
{"abstract": "  Phonon-mediated particle detectors based on single crystals and operated at\nmillikelvin temperatures are used in rare-event experiments for neutrino\nphysics and dark-matter searches. In general, these devices are not sensitive\nto the particle impact point, especially if the detection is mediated by\nthermal phonons. In this Letter, we demonstrate that excellent discrimination\nbetween interior and surface $\\beta$ and $\\alpha$ events can be achieved by\ncoating a crystal face with a thin metallic film, either continuous or in the\nform of a grid. The coating affects the phonon energy down-conversion cascade\nthat follows the particle interaction, leading to a modified signal shape for\nclose-to-film events. An efficient identification of surface events was\ndemonstrated with detectors based on a rectangular $20 \\times 20 \\times 10$\nmm$^3$ Li$_2$MoO$_4$ crystal coated with a Pd normal-metal film (10~nm thick)\nand with Al-Pd superconductive bi-layers (100~nm-10~nm thick) on a $20 \\times\n20$ mm$^2$ face. Discrimination capabilities were tested with $^{238}$U sources\nemitting both $\\alpha$ and $\\beta$ particles. Surface events are identified for\nenergy depositions down to millimeter-scale depths from the coated surface.\nWith this technology, a substantial improvement of the background figure can be\nachieved in experiments searching for neutrinoless double-beta decay.\n"}
{"abstract": "  Spiral acquisitions are preferred in real-time MRI because of their\nefficiency, which has made it possible to capture vocal tract dynamics during\nnatural speech. A fundamental limitation of spirals is blurring and signal loss\ndue to off-resonance, which degrades image quality at air-tissue boundaries.\nHere, we present a new CNN-based off-resonance correction method that\nincorporates an attention-gate mechanism. This leverages spatial and channel\nrelationships of filtered outputs and improves the expressiveness of the\nnetworks. We demonstrate improved performance with the attention-gate, on 1.5\nTesla spiral speech RT-MRI, compared to existing off-resonance correction\nmethods.\n"}
{"abstract": "  Low-precision training has become a popular approach to reduce compute\nrequirements, memory footprint, and energy consumption in supervised learning.\nIn contrast, this promising approach has not yet enjoyed similarly widespread\nadoption within the reinforcement learning (RL) community, partly because RL\nagents can be notoriously hard to train even in full precision. In this paper\nwe consider continuous control with the state-of-the-art SAC agent and\ndemonstrate that a na\\\"ive adaptation of low-precision methods from supervised\nlearning fails. We propose a set of six modifications, all straightforward to\nimplement, that leaves the underlying agent and its hyperparameters unchanged\nbut improves the numerical stability dramatically. The resulting modified SAC\nagent has lower memory and compute requirements while matching full-precision\nrewards, demonstrating that low-precision training can substantially accelerate\nstate-of-the-art RL without parameter tuning.\n"}
{"abstract": "  The operation of the future Cherenkov Telescope Array (CTA), the next\ngeneration of Imaging Atmospheric Cherenkov Telescopes (IACTs), as an Open\nObservatory requires data products and analysis tools to be accessible and\nusable by the scientific community. This requirement has impelled gamma-ray\nastronomers to develop open-source science tools and standardised data formats.\nThe objective of this presentation is to provide a perspective on the current\neffort to define a prototypical format for high-level gamma-ray astronomical\ndata. The initiative, started by the IACT community in 2016, has gained in the\nyears 2018 and 2019 full recognition thanks to the first public data releases\nin such format by current-generation IACTs, followed by a series of papers\nemploying it either to test the capabilities of open-source gamma-ray analysis\ntools or to showcase examples of reproducible, multi-instrument analysis.\n"}
{"abstract": "  In this paper, we focus on constructing numerical schemes preserving the\naveraged energy evolution law for nonlinear stochastic wave equations driven by\nmultiplicative noise. We first apply the compact finite difference method and\nthe interior penalty discontinuous Galerkin finite element method to discretize\nspace variable and present two semi-discrete schemes, respectively. Then we\nmake use of the discrete gradient method and the Pad\\'e approximation to\npropose efficient fully-discrete schemes. These semi-discrete and\nfully-discrete schemes are proved to preserve the discrete averaged energy\nevolution law. In particular, we also prove that the proposed fully-discrete\nschemes exactly inherit the averaged energy evolution law almost surely if the\nconsidered model is driven by additive noise. Numerical experiments are given\nto confirm theoretical findings.\n"}
{"abstract": "  We consider the problem of scheduling in constrained queueing networks with a\nview to minimizing packet delay. Modern communication systems are becoming\nincreasingly complex, and are required to handle multiple types of traffic with\nwidely varying characteristics such as arrival rates and service times. This,\ncoupled with the need for rapid network deployment, render a bottom up approach\nof first characterizing the traffic and then devising an appropriate scheduling\nprotocol infeasible.\n  In contrast, we formulate a top down approach to scheduling where, given an\nunknown network and a set of scheduling policies, we use a policy gradient\nbased reinforcement learning algorithm that produces a scheduler that performs\nbetter than the available atomic policies. We derive convergence results and\nanalyze finite time performance of the algorithm. Simulation results show that\nthe algorithm performs well even when the arrival rates are nonstationary and\ncan stabilize the system even when the constituent policies are unstable.\n"}
{"abstract": "  We present a necessary and sufficient condition for Alt's system to be\nrepresented by a continuous utility function. Moreover, we present a necessary\nand sufficient condition for this utility function to be concave. The latter\ncondition can be seen as an extension of Gossen's first law, and thus has an\neconomic interpretation. Together with the above results, we provide a\nnecessary and sufficient condition for Alt's utility to be continuously\ndifferentiable.\n"}
{"abstract": "  We propose UniT, a Unified Transformer model to simultaneously learn the most\nprominent tasks across different domains, ranging from object detection to\nnatural language understanding and multimodal reasoning. Based on the\ntransformer encoder-decoder architecture, our UniT model encodes each input\nmodality with an encoder and makes predictions on each task with a shared\ndecoder over the encoded input representations, followed by task-specific\noutput heads. The entire model is jointly trained end-to-end with losses from\neach task. Compared to previous efforts on multi-task learning with\ntransformers, we share the same model parameters across all tasks instead of\nseparately fine-tuning task-specific models and handle a much higher variety of\ntasks across different domains. In our experiments, we learn 7 tasks jointly\nover 8 datasets, achieving strong performance on each task with significantly\nfewer parameters. Our code is available in MMF at https://mmf.sh.\n"}
{"abstract": "  While studying systems driven out of equilibrium, one usually employs a drive\nthat is not directly coupled to the degrees of freedom of the system. In\ncontrast to such a case, we here unveil a hitherto unexplored situation of\nstate-dependent driving, whereby a direct coupling exists between the two. We\ndemonstrate the ubiquity of such a driving, and establish that it leads to a\nnontrivial steady-state that is qualitatively opposite to what is observed in\nother driven systems. Further, we show how state-dependent driving in a\nmany-body system can be effectively captured in terms of a single-particle\nmodel. The origin of this description may ultimately be traced to the fact that\nstate-dependent driving results in a force that undergoes repeated resetting in\ntime.\n"}
{"abstract": "  As we rapidly approach the frontiers of ultra large computing resources,\nsoftware optimization is becoming of paramount interest to scientific\napplication developers interested in efficiently leveraging all available\non-Node computing capabilities and thereby improving a requisite science per\nwatt metric. The scientific application of interest here is the Basic Math\nLibrary (BML) that provides a singular interface for linear algebra operation\nfrequently used in the Quantum Molecular Dynamics (QMD) community. The\nprovisioning of a singular interface indicates the presence of an abstraction\nlayer which in-turn suggests commonalities in the code-base and therefore any\noptimization or tuning introduced in the core of code-base has the ability to\npositively affect the performance of the aforementioned library as a whole.\nWith that in mind, we proceed with this investigation by performing a survey of\nthe entirety of the BML code-base, and extract, in form of micro-kernels,\ncommon snippets of code. We introduce several optimization strategies into\nthese micro-kernels including 1.) Strength Reduction 2.) Memory Alignment for\nlarge arrays 3.) Non Uniform Memory Access (NUMA) aware allocations to enforce\ndata locality and 4.) appropriate thread affinity and bindings to enhance the\noverall multi-threaded performance. After introducing these optimizations, we\nbenchmark the micro-kernels and compare the run-time before and after\noptimization for several target architectures. Finally we use the results as a\nguide to propagating the optimization strategies into the BML code-base. As a\ndemonstration, herein, we test the efficacy of these optimization strategies by\ncomparing the benchmark and optimized versions of the code.\n"}
{"abstract": "  We present the joint analysis of Neutral Hydrogen (HI) Intensity Mapping\nobservations with three galaxy samples: the Luminous Red Galaxy (LRG) and\nEmission Line Galaxy (ELG) samples from the eBOSS survey, and the WiggleZ Dark\nEnergy Survey sample. The HI intensity maps are Green Bank Telescope\nobservations of the redshifted 21cm emission on 100deg2 covering the redshift\nrange $0.6<z<1.0$. We process the data by separating and removing the\nforegrounds with FastICA, and construct a transfer function to correct for the\neffects of foreground removal on the HI signal. We cross-correlate the cleaned\nHI data with the galaxy samples and study the overall amplitude as well as the\nscale-dependence of the power spectrum. We also qualitatively compare our\nfindings with the predictions by a semi-analytic galaxy evolution simulation.\nThe cross-correlations constrain the quantity $\\Omega_{{HI}} b_{{HI}}\nr_{{HI},{opt}}$ at an effective scale $k_{eff}$, where $\\Omega_{HI}$ is the HI\ndensity fraction, $b_{HI}$ is the HI bias, and $r_{{HI},{opt}}$ the\ngalaxy-hydrogen correlation coefficient, which is dependent on the HI content\nof the optical galaxy sample. At $k_{eff}=0.31 \\, h/{Mpc}$ we find\n$\\Omega_{{HI}} b_{{HI}} r_{{HI},{Wig}} = [0.58 \\pm 0.09 \\, {(stat) \\pm 0.05 \\,\n{(sys)}}] \\times 10^{-3}$ for GBT-WiggleZ, $\\Omega_{{HI}} b_{{HI}}\nr_{{HI,{ELG}}} = [0.40 \\pm 0.09 \\, {(stat) \\pm 0.04 \\, {(sys)}}] \\times\n10^{-3}$ for GBT-ELG, and $\\Omega_{{HI}} b_{{HI}} r_{{HI},{LRG}} = [0.35 \\pm\n0.08 \\, {(stat) \\pm 0.03 \\, {(sys)}}] \\times 10^{-3}$ for GBT-LRG, at $z\\simeq\n0.8$. We also report results at $k_{eff}=0.24 \\, h/{Mpc}$ and $k_{eff}=0.48 \\,\nh/{Mpc}$. With little information on HI parameters beyond our local Universe,\nthese are amongst the most precise constraints on neutral hydrogen density\nfluctuations in an underexplored redshift range.\n"}
{"abstract": "  We gave an extensive study for the quasi-periodic perturbations on the time\nprofiles of the line of sight (LOS) magnetic field in 10x10 sub-areas in a\nsolar plage region (corresponds to a facula on the photosphere). The\nperturbations are found to be associated with enhancement of He I 10830 A\nabsorption in a moss region, which is connected to loops with million-degree\nplasma. FFT analysis to the perturbations gives a kind of spectrum similar to\nthat of Doppler velocity: a number of discrete periods around 5 minutes. The\namplitudes of the magnetic perturbations are found to be proportional to\nmagnetic field strength over these sub-areas. In addition, magnetic\nperturbations lag behind a quarter of cycle in phase with respect to the p-mode\nDoppler velocity. We show that the relationships can be well explained with an\nMHD solution for the magneto-acoustic oscillations in high-\\b{eta} plasma.\nObservational analysis also shows that, for the two regions with the stronger\nand weaker magnetic field, the perturbations are always anti-phased. All\nfindings show that the magnetic perturbations are actually magneto-acoustic\noscillations on the solar surface, the photosphere, powered by p-mode\noscillations. The findings may provide a new diagnostic tool for exploring the\nrelationship between magneto-acoustic oscillations and the heating of solar\nupper atmosphere, as well as their role in helioseismology.\n"}
{"abstract": "  The accurate determination of electronic temperatures in metallic\nnanostructures is essential for many technological applications, like\nplasmon-enhanced catalysis or lithographic nanofabrication procedures. In this\nLetter we demonstrate that the electronic temperature can be accurately\nmeasured by the shape of the tunnel electroluminescence emission edge in tunnel\nplasmonic nanocavities, which follows a universal thermal distribution with the\nbias voltage as the chemical potential of the photon population. A significant\ndeviation between electronic and lattice temperatures is found below 30 K for\ntunnel currents larger than 15 nA. This deviation is rationalized as the result\nof a two-electron process in which the second electron excites plasmon modes\nwith an energy distribution that reflects the higher temperature following the\nfirst tunneling event. These results dispel a long-standing controversy on the\nnature of overbias emission in tunnel junctions and adds a new method for the\ndetermination of electronic temperatures and quasiparticle dynamics.\n"}
{"abstract": "  Continued improvements on existing reconstruction methods are vital to the\nsuccess of high-energy physics experiments, such as the IceCube Neutrino\nObservatory. In IceCube, further challenges arise as the detector is situated\nat the geographic South Pole where computational resources are limited.\nHowever, to perform real-time analyses and to issue alerts to telescopes around\nthe world, powerful and fast reconstruction methods are desired. Deep neural\nnetworks can be extremely powerful, and their usage is computationally\ninexpensive once the networks are trained. These characteristics make a deep\nlearning-based approach an excellent candidate for the application in IceCube.\nA reconstruction method based on convolutional architectures and hexagonally\nshaped kernels is presented. The presented method is robust towards systematic\nuncertainties in the simulation and has been tested on experimental data. In\ncomparison to standard reconstruction methods in IceCube, it can improve upon\nthe reconstruction accuracy, while reducing the time necessary to run the\nreconstruction by two to three orders of magnitude.\n"}
{"abstract": "  In this study, we will study investor attention measurement using the Search\nVolume Index in the recent market. Since 2009, the popularity of mobile devices\nand the spread of the Internet have made the speed of information delivery\nfaster and the investment information retrieval data for obtaining investment\ninformation has increased dramatically. In these circumstances, investor\nattention measurement using search volume data can be measured more accurately\nand faster than before mobile device popularization. To confirm this, we will\ncompare the effect of measuring investor attention using search volume data\nbefore and after mobile device popularization. In addition, it is confirmed\nthat the measured investor attention is that of retail traders, not\ninstitutional traders or professional traders, and the relationship between\ninvestor attention and short-term price pressure theory. Using SVI data\nprovided by Google Trends, we will experiment with Russell 3000 stocks and IPO\nstocks and compare the results. In addition, the results of investigating the\ninvestor's interest using the search volume data from various angles through\nexperiments such as the comparison of the results based on the inclusion of the\nnoise ticker group, the comparison of the limitations of the existing investor\nattention measurement method, and the comparison of explanatory variables with\nexisting IPO related studies. We would like to verify its practicality and\nsignificance.\n"}
{"abstract": "  This paper proposes a new framework to evaluate unconditional quantile\neffects (UQE) in a data combination model. The UQE measures the effect of a\nmarginal counterfactual change in the unconditional distribution of a covariate\non quantiles of the unconditional distribution of a target outcome. Under rank\nsimilarity and conditional independence assumptions, we provide a set of\nidentification results for UQEs when the target covariate is continuously\ndistributed and when it is discrete, respectively. Based on these\nidentification results, we propose semiparametric estimators and establish\ntheir large sample properties under primitive conditions. Applying our method\nto a variant of Mincer's earnings function, we study the counterfactual\nquantile effect of actual work experience on income.\n"}
{"abstract": "  We prove that the genus-0 sector of the quartic analogue of the Kontsevich\nmodel is completely governed by an involution identity which expresses the\nmeromorphic differential $\\omega_{0,n}$ at a reflected point $\\iota z$ in terms\nof all $\\omega_{0,m}$ with $m\\leq n$ at the original point $z$. We prove that\nthe solution of the involution identity obeys blobbed topological recursion,\nwhich confirms a previous conjecture about the quartic Kontsevich model.\n"}
{"abstract": "  This compact review about gluonium focuses on a slate of theoretical efforts;\namong the many standing works, I have selected several that are meant to assist\nin the identification, among ordinary mesons, of the few Yang-Mills glueball\nconfigurations that populate the energy region below 3 GeV. This includes\n$J/\\psi$ radiative and vector-meson decays, studies of scalar meson mixing, of\nhigh-energy cross sections via the Pomeron and the odderon, glueball decays,\netc. The weight of accumulated evidence seems to support the $f_0(1710)$ as\nhaving a large (and the largest) glueball component among the scalars, although\nno single observable by itself is conclusive. Further tests would be welcome,\nsuch as exclusive $f_J$ production at asymptotically high $s$ and $t$. No clear\nexperimental candidates for the pseudoscalar or tensor glueball stand out yet,\nand continuing investigations trying to sort them out will certainly teach us\nmuch more about mesons.\n"}
{"abstract": "  Recently, some approaches are proposed to harness deep convolutional networks\nto facilitate superpixel segmentation. The common practice is to first evenly\ndivide the image into a pre-defined number of grids and then learn to associate\neach pixel with its surrounding grids. However, simply applying a series of\nconvolution operations with limited receptive fields can only implicitly\nperceive the relations between the pixel and its surrounding grids.\nConsequently, existing methods often fail to provide an effective context when\ninferring the association map. To remedy this issue, we propose a novel\n\\textbf{A}ssociation \\textbf{I}mplantation (AI) module to enable the network to\nexplicitly capture the relations between the pixel and its surrounding grids.\nThe proposed AI module directly implants the features of grid cells to the\nsurrounding of its corresponding central pixel, and conducts convolution on the\npadded window to adaptively transfer knowledge between them. With such an\nimplantation operation, the network could explicitly harvest the pixel-grid\nlevel context, which is more in line with the target of superpixel segmentation\ncomparing to the pixel-wise relation. Furthermore, to pursue better boundary\nprecision, we design a boundary-perceiving loss to help the network\ndiscriminate the pixels around boundaries in hidden feature level, which could\nbenefit the subsequent inferring modules to accurately identify more boundary\npixels. Extensive experiments on BSDS500 and NYUv2 datasets show that our\nmethod could not only achieve state-of-the-art performance but maintain\nsatisfactory inference efficiency.\n"}
{"abstract": "  We present a novel method to reconstruct a spectral central view and its\naligned disparity map from spatio-spectrally coded light fields. Since we do\nnot reconstruct an intermediate full light field from the coded measurement, we\nrefer to this as principal reconstruction. The coded light fields correspond to\nthose captured by a light field camera in the unfocused design with a\nspectrally coded microlens array. In this application, the spectrally coded\nlight field camera can be interpreted as a single-shot spectral depth camera.\n  We investigate several multi-task deep learning methods and propose a new\nauxiliary loss-based training strategy to enhance the reconstruction\nperformance. The results are evaluated using a synthetic as well as a new\nreal-world spectral light field dataset that we captured using a custom-built\ncamera. The results are compared to state-of-the art compressed sensing\nreconstruction and disparity estimation.\n  We achieve a high reconstruction quality for both synthetic and real-world\ncoded light fields. The disparity estimation quality is on par with or even\noutperforms state-of-the-art disparity estimation from uncoded RGB light\nfields.\n"}
{"abstract": "  Understanding unsupervised domain adaptation has been an important task that\nhas been well explored. However, the wide variety of methods have not analyzed\nthe role of a classifier's performance in detail. In this paper, we thoroughly\nexamine the role of a classifier in terms of matching source and target\ndistributions. We specifically investigate the classifier ability by matching\na) the distribution of features, b) probabilistic uncertainty for samples and\nc) certainty activation mappings. Our analysis suggests that using these three\ndistributions does result in a consistently improved performance on all the\ndatasets. Our work thus extends present knowledge on the role of the various\ndistributions obtained from the classifier towards solving unsupervised domain\nadaptation.\n"}
{"abstract": "  Cosmological phase transitions are thought to have taken place at the early\nUniverse imprinting their properties on the observable Universe. There is\nstrong evidence that, through the dynamics of a scalar field that lead a second\norder phase transition, inflation shaped the Universe accounting for the most\nconspicuous features of the observed Universe. It is argued that inflation has\nalso striking implications for the vacuum energy. Considerations for subsequent\nsecond order phase transitions are also discussed.\n"}
{"abstract": "  We investigate polariton quantum correlations in a coherently driven box\ncavity in the low driving regime, with a particular focus on accounting for the\npolarization degree of freedom. The possibility of having different interaction\nstrengths between co- and cross-circularly polarized polaritons as well as a\nrealistic linear-polarization splitting allows one to model the system as two\ncoupled nonlinear resonators with both self- and cross-Kerr-like\nnonlinearities, thus making our results potentially relevant for other\nexperimental platforms. Within an effective wave-function approach, we obtain\nanalytical expressions for the steady-state polarization-resolved polariton\npopulations and second-order correlation functions, which agree very well with\nour numerical results obtained from a Lindblad master equation. Notably, we\nhighlight that depending on the excitation polarization (circular or linear),\nboth the unconventional (interference-mediated) and conventional (mediated by\nnonlinearities) antibunchings can be investigated in a single cavity. Moreover,\nusing our results, we argue that recent experiments on confined fiber-cavity\npolaritons are likely to have probed a regime where the dominant interaction is\nbetween cross-polarized polaritons, which is characteristic of the polariton\nFeshbach resonance. We furthermore investigate the regime close to resonance\nusing a two-channel model, and we show that systems with large biexciton\nbinding energies, such as atomically thin semiconductors, are promising\nplatforms for realizing strong polariton antibunching.\n"}
{"abstract": "  In this paper, we study the single-source shortest-path (SSSP) problem with\npositive edge weights, which is a notoriously hard problem in the parallel\ncontext. In practice, the $\\Delta$-stepping algorithm proposed by Meyer and\nSanders has been widely adopted. However, $\\Delta$-stepping has no known\nworst-case bounds for general graphs. The performance of $\\Delta$-stepping also\nhighly relies on the parameter $\\Delta$. There have also been lots of\nalgorithms with theoretical bounds, such as Radius-stepping, but they either\nhave no implementations available or are much slower than $\\Delta$-stepping in\npractice.\n  We propose a stepping algorithm framework that generalizes existing\nalgorithms such as $\\Delta$-stepping and Radius-stepping. The framework allows\nfor similar analysis and implementations of all stepping algorithms. We also\npropose a new ADT, lazy-batched priority queue (LaB-PQ), that abstracts the\nsemantics of the priority queue needed by the stepping algorithms. We provide\ntwo data structures for LaB-PQ, focusing on theoretical and practical\nefficiency, respectively. Based on the new framework and LaB-PQ, we show two\nnew stepping algorithms, $\\rho$-stepping and $\\Delta^*$-stepping, that are\nsimple, with non-trivial worst-case bounds, and fast in practice.\n  The stepping algorithm framework also provides almost identical\nimplementations for three algorithms: Bellman-Ford, $\\Delta^*$-stepping, and\n$\\rho$-stepping. We compare our code with four state-of-the-art\nimplementations. On five social and web graphs, $\\rho$-stepping is 1.3--2.5x\nfaster than all the existing implementations. On two road graphs, our\n$\\Delta^*$-stepping is at least 14\\% faster than existing implementations,\nwhile $\\rho$-stepping is also competitive. The almost identical implementations\nfor stepping algorithms also allow for in-depth analyses and comparisons among\nthe stepping algorithms in practice.\n"}
{"abstract": "  Atomic-like defects in solids are not considered to be identical owing to the\nimperfections of host lattice. Here, we found that even under ambient\nconditions, negatively charged nitrogen-vacancy (NV$^-$) centers in diamond\ncould still manifest identical at Hz-precision level, corresponding to a\n10$^{-7}$-level relative precision, while the lattice strain can destroy the\nidentity by tens of Hz. All parameters involved in the NV$^-$-$^{14}$N\nHamiltonian are determined by formulating six nuclear frequencies at\n10-mHz-level precision and measuring them at Hz-level precision. The most\nprecisely measured parameter, the $^{14}$N quadrupole coupling $P$, is given by\n-4945754.9(8) Hz, whose precision is improved by nearly four orders of\nmagnitude compared with previous measurements. We offer an approach for\nperforming precision measurements in solids and deepening our understandings of\nNV centers as well as other solid-state defects. Besides, these high-precision\nresults imply a potential application of a robust and integrated atomic-like\nclock based on ensemble NV centers.\n"}
{"abstract": "  In this note, we completely classify left-invariant Riemann solitons on\nthree-dimensional Lorentzian Lie groups.\n"}
{"abstract": "  In our recent work, we build a propagation scenario to simultaneously explain\nthe spectra and anisotropy of cosmic rays (CRs) by considering spatially\ndependent propagation (SDP) model and nearby Geminga supernova remnant (SNR)\nsource. But the phase of anisotropy is still inconsistent with the experimental\ndata. Recent observations of CR anisotropy show that the phase is consistent\nwith local regular magnetic field (LRMF) observed by Interstellar Boundary\nExplorer (IBEX) below 100 TeV, which indicates that diffusion along LRMF is\nimportant. In this work, we further introduce the LRMF and take into account\nthe effect of corresponding anisotropic diffusion to explain the anisotropy of\nCRs. We find that when the diffusion coefficient perpendicular to the LRMF is\nmuch smaller than the parallel one, the phase of anisotropy points to $\\sim\nR.A.= 3^h$, which accords with experimental observation below 100 TeV. We also\nanalyze the influence of the ratio of perpendicular and parallel diffusion\ncoefficient on the anisotropy and the energy dependence of the ratio. The\nresults illustrate that with the decrease of perpendicular diffusion, the\nanisotropic phase changes from the direction of nearby source to the LRMF below\n100 TeV, meanwhile it changes from the galactic center (GC) to opposite\ndirection of LRMF above 100 TeV. When the perpendicular diffusion coefficient\ngrows faster than the parallel one with energy, the diffusion approaches to the\nisotropic at high energy, the phase of anisotropy shifts from the LRMF to the\nGC above 100 TeV. This could be helpful to ascertain the energy dependence of\ndiffusion coefficients.\n"}
{"abstract": "  The recent boost in data transfer rates puts a daring strain on information\ntechnology. Sustaining such a growth rate requires the development of sources,\ndetectors and systems working in the so-called TeraHertz (THz) gap covering the\nfrequency window from 0.1 to 10 THz (1 THz = 10$^{12}$~Hz). This gap represents\na challenge for conventional electronic devices due to carrier transit delays\n($\\sim$1-10ps), as well as for photonic devices due to thermal fluctuations\n(300K$\\sim$6THz). Nonetheless, designing efficient, room-temperature THz\nsources would constitute a key enabler to applications spanning from\nhigh-resolution imaging to extreme wide band wireless communication. Whereas\nhigh-harmonic generation in solid is usually limited to less than ten\nharmonics, broadband THz emission has been demonstrated using laser-induced\nsuperdiffusive spin currents in magnetic bilayers composed of a ferromagnet\ndeposited on top of a noble metal. While promising, this technique presents the\nmajor disadvantage of necessitating optical pumping and hence lacks\nscalability. Here, we demonstrate that extremely high harmonic emission can be\nachieved by exploiting conventional spin pumping, without the need of optical\nexcitation. We show that when the spin-orbit coupling strength is close to the\ns-d exchange energy, a strongly non linear regime resulting from resonant spin\nflip scattering occurs leading to the generation of a thousand of harmonics at\nrealistic antiferromagnetic precession frequencies, thereby enhancing both spin\nand charge dynamics by two orders of magnitude, and allowing for an emission at\nfrequencies above 300 THz.\n"}
{"abstract": "  Nowadays, neural vocoders can generate very high-fidelity speech when a bunch\nof training data is available. Although a speaker-dependent (SD) vocoder\nusually outperforms a speaker-independent (SI) vocoder, it is impractical to\ncollect a large amount of data of a specific target speaker for most real-world\napplications. To tackle the problem of limited target data, a data augmentation\nmethod based on speaker representation and similarity measurement of speaker\nverification is proposed in this paper. The proposed method selects utterances\nthat have similar speaker identity to the target speaker from an external\ncorpus, and then combines the selected utterances with the limited target data\nfor SD vocoder adaptation. The evaluation results show that, compared with the\nvocoder adapted using only limited target data, the vocoder adapted using\naugmented data improves both the quality and similarity of synthesized speech.\n"}
{"abstract": "  Disjoint paths are defined as paths between the source and destination nodes\nwhere the intermediate nodes in any two paths are disjoint. They are helpful in\nfault-tolerance routing and securing message distribution in the network.\nSeveral research papers were proposed to solve the problem of finding disjoint\npaths for variety of interconnection networks such as Hypercube, Generalized\nHypercube, Mesh, Torus, Gaussian, Eisenstein-Jacobi, and many other topologies.\nIn this research, we have developed a general real-time fault-tolerance\nalgorithm that constructs all node-to-node disjoint paths for symmetric\nnetworks where all paths are shortest or close to shortest. In addition, we\nhave simulated the proposed algorithm on different networks. The solution of\nunsolved problem in Cube-Connected-Cycles is given in the simulation results.\n"}
{"abstract": "  For classic Lotka-Volterra systems governing many interacting species, we\nestablish an exclusion principle that rules out the existence of linearly\nasymptotically stable steady states in subcommunitites of communities that\nadmit a stable state which is internally D-stable. This type of stability is\nknown to be ensured, e.g., by diagonal dominance or Volterra-Lyapunov stability\nconditions. By consequence, the number of stable steady states of this type is\nbounded by Sperner's lemma on anti-chains in a poset. The number of stable\nsteady states can nevertheless be very large if there are many groups of\nspecies that strongly inhibit outsiders but have weak interactions among\nthemselves. By examples we also show that in general it is possible for a\nstable community to contain a stable subcommunity consisting of a single\nspecies. Thus a recent empirical finding to the contrary, in a study of random\ncompetitive systems by Lischke and L\\\"offler (Theo. Pop. Biol. 115 (2017)\n24--34), does not hold without qualification.\n"}
{"abstract": "  Video relation detection problem refers to the detection of the relationship\nbetween different objects in videos, such as spatial relationship and action\nrelationship. In this paper, we present video relation detection with\ntrajectory-aware multi-modal features to solve this task.\n  Considering the complexity of doing visual relation detection in videos, we\ndecompose this task into three sub-tasks: object detection, trajectory proposal\nand relation prediction. We use the state-of-the-art object detection method to\nensure the accuracy of object trajectory detection and multi-modal feature\nrepresentation to help the prediction of relation between objects. Our method\nwon the first place on the video relation detection task of Video Relation\nUnderstanding Grand Challenge in ACM Multimedia 2020 with 11.74\\% mAP, which\nsurpasses other methods by a large margin.\n"}
{"abstract": "  The classical 1961 solution to the problem of determining the number of\nperfect matchings (or dimer coverings) of a rectangular grid graph -- due\nindependently to Kasteleyn and to Temperley and Fisher -- consists of changing\nthe sign of some of the entries in the adjacency matrix so that the Pfaffian of\nthe new matrix gives the number of perfect matchings, and then evaluating this\nPfaffian. Another classical method is to use the Lindstr\\\"om-Gessel-Viennot\ntheorem on non-intersecting lattice paths to express the number of perfect\nmatchings as a determinant, and then evaluate this determinant. In this paper\nwe present a new method for solving the two dimensional dimer problem, which\nrelies on the Cauchy-Binet theorem. It only involves facts that were known in\nthe mid 1930's when the dimer problem was phrased, so it could have been\ndiscovered while the dimer problem was still open.\n  We provide explicit product formulas for both the square and the hexagonal\nlattice. One advantage of our formula for the square lattice compared to the\noriginal formula of Kasteleyn, Temperley and Fisher is that ours has a linear\nnumber of factors, while the number of factors in the former is quadratic. Our\nresult for the hexagonal lattice yields a formula for the number of periodic\nstepped surfaces that fit in an infinite tube of given cross-section, which can\nbe regarded as a counterpart of MacMahon's boxed plane partition theorem.\n"}
{"abstract": "  This article proves the products, behaviors and simple zeros for the classes\nof the entire functions associated with the Weierstrass-Hadamard product and\nthe Taylor series.\n"}
{"abstract": "  Given a matrix $X\\in \\mathbb{R}_+^{m\\times n}$ with nonnegative entries, a\nPositive Semidefinite (PSD) factorization of $X$ is a collection of $r \\times\nr$-dimensional PSD matrices $\\{A_i\\}$ and $\\{B_j\\}$ satisfying $X_{ij}=\n\\mathrm{tr}(A_i B_j)$ for all $\\ i\\in [m],\\ j\\in [n]$. PSD factorizations are\nfundamentally linked to understanding the expressiveness of semidefinite\nprograms as well as the power and limitations of quantum resources in\ninformation theory. The PSD factorization task generalizes the Non-negative\nMatrix Factorization (NMF) problem where we seek a collection of\n$r$-dimensional nonnegative vectors $\\{a_i\\}$ and $\\{b_j\\}$ satisfying $X_{ij}=\na_i^\\top b_j$, for all $i\\in [m],\\ j\\in [n]$ -- one can recover the latter\nproblem by choosing matrices in the PSD factorization to be diagonal. The most\nwidely used algorithm for computing NMFs of a matrix is the Multiplicative\nUpdate algorithm developed by Lee and Seung, in which nonnegativity of the\nupdates is preserved by scaling with positive diagonal matrices. In this paper,\nwe describe a non-commutative extension of Lee-Seung's algorithm, which we call\nthe Matrix Multiplicative Update (MMU) algorithm, for computing PSD\nfactorizations. The MMU algorithm ensures that updates remain PSD by congruence\nscaling with the matrix geometric mean of appropriate PSD matrices, and it\nretains the simplicity of implementation that Lee-Seung's algorithm enjoys.\nBuilding on the Majorization-Minimization framework, we show that under our\nupdate scheme the squared loss objective is non-increasing and fixed points\ncorrespond to critical points. The analysis relies on Lieb's Concavity Theorem.\nBeyond PSD factorizations, we use the MMU algorithm as a primitive to calculate\nblock-diagonal PSD factorizations and tensor PSD factorizations. We demonstrate\nthe utility of our method with experiments on real and synthetic data.\n"}
{"abstract": "  The krypton electroluminescence yield was studied, at room temperature, as a\nfunction of electric field in the gas scintillation gap. A large area avalanche\nphotodiode has been used to allow the simultaneous detection of the\nelectroluminescence pulses as well as the direct interaction of x-rays, the\nlatter being used as a reference for the calculation of the number of charge\ncarriers produced by the electroluminescence pulses and, thus, the\ndetermination of the number of photons impinging the photodiode. An\namplification parameter of 113 photons per kV per drifting electron and a\nscintillation threshold of 2.7 Td ( 0.7 kV/cm/bar at 293 K ) was obtained, in\ngood agreement with the simulation data reported in the literature. On the\nother hand, the ionisation threshold in krypton was found to be around 13.5 Td\n(3.4 kV/cm/bar), less than what had been obtained by the most recent simulation\nwork-package. The krypton amplification parameter is about 80% and 140% of\nthose measured for xenon and argon, respectively. The electroluminescence yield\nin krypton is of great importance for modeling krypton-based double-phase or\nhigh-pressure gas detectors, which may be used in future rare event detection\nexperiments.\n"}
{"abstract": "  Many important challenges in science and technology can be cast as\noptimization problems. When viewed in a statistical physics framework, these\ncan be tackled by simulated annealing, where a gradual cooling procedure helps\nsearch for groundstate solutions of a target Hamiltonian. While powerful,\nsimulated annealing is known to have prohibitively slow sampling dynamics when\nthe optimization landscape is rough or glassy. Here we show that by\ngeneralizing the target distribution with a parameterized model, an analogous\nannealing framework based on the variational principle can be used to search\nfor groundstate solutions. Modern autoregressive models such as recurrent\nneural networks provide ideal parameterizations since they can be exactly\nsampled without slow dynamics even when the model encodes a rough landscape. We\nimplement this procedure in the classical and quantum settings on several\nprototypical spin glass Hamiltonians, and find that it significantly\noutperforms traditional simulated annealing in the asymptotic limit,\nillustrating the potential power of this yet unexplored route to optimization.\n"}
{"abstract": "  We study the excitation of asymmetric-top (including chiral) molecules by\ntwo-color femtosecond laser pulses. In the cases of non-chiral asymmetric-top\nmolecules excited by an orthogonally polarized two-color pulse, we demonstrate,\nclassically and quantum mechanically, three-dimensional orientation. For chiral\nmolecules, we show that the orientation induced by a cross-polarized two-color\npulse is enantioselective along the laser propagation direction, namely, the\ntwo enantiomers are oriented in opposite directions. On the short time scale,\nthe classical and quantum simulations give results that are in excellent\nagreement, whereas on the longer time scale, the enantioselective orientation\nexhibits quantum beats. These observations are qualitatively explained by\nanalyzing the interaction potential between the two-color pulse and molecular\n(hyper-)polarizability. The prospects for utilizing the long-lasting\norientation for measuring and using the enantioselective orientation for\nseparating the individual enantiomers are discussed.\n"}
{"abstract": "  We present new obstructions for a knot K in S^3 to admit purely cosmetic\nsurgeries, which arise from the study of Witten-Reshetikhin-Turaev invariants\nat fixed level. In particular, we strengthen a recent result of Hanselman,\nshowing that if K has purely cosmetic surgeries then the slopes of the\nsurgeries are of the form 1/5k except if the Jones polynomial of K evaluated at\na 5-th root of unity is 1. For any odd prime r, we also give an obstruction for\nK to have a 1/k surgery slope with k coprime to r that involves the values of\nthe first (r-3)/2 colored Jones polynomials of K at an r-th root of unity. We\nverify the purely cosmetic surgery conjecture for all knots with at most 17\ncrossings.\n"}
{"abstract": "  Despite the widespread availability of in-treatment room cone beam computed\ntomography (CBCT) imaging, due to the lack of reliable segmentation methods,\nCBCT is only used for gross set up corrections in lung radiotherapies. Accurate\nand reliable auto-segmentation tools could potentiate volumetric response\nassessment and geometry-guided adaptive radiation therapies. Therefore, we\ndeveloped a new deep learning CBCT lung tumor segmentation method. Methods: The\nkey idea of our approach called cross modality educed distillation (CMEDL) is\nto use magnetic resonance imaging (MRI) to guide a CBCT segmentation network\ntraining to extract more informative features during training. We accomplish\nthis by training an end-to-end network comprised of unpaired domain adaptation\n(UDA) and cross-domain segmentation distillation networks (SDN) using unpaired\nCBCT and MRI datasets. Feature distillation regularizes the student network to\nextract CBCT features that match the statistical distribution of MRI features\nextracted by the teacher network and obtain better differentiation of tumor\nfrom background.} We also compared against an alternative framework that used\nUDA with MR segmentation network, whereby segmentation was done on the\nsynthesized pseudo MRI representation. All networks were trained with 216\nweekly CBCTs and 82 T2-weighted turbo spin echo MRI acquired from different\npatient cohorts. Validation was done on 20 weekly CBCTs from patients not used\nin training. Independent testing was done on 38 weekly CBCTs from patients not\nused in training or validation. Segmentation accuracy was measured using\nsurface Dice similarity coefficient (SDSC) and Hausdroff distance at 95th\npercentile (HD95) metrics.\n"}
{"abstract": "  Input devices, such as buttons and sliders, are the foundation of any\ninterface. The typical user-centered design workflow requires the developers\nand users to go through many iterations of design, implementation, and\nanalysis. The procedure is inefficient, and human decisions highly bias the\nresults. While computational methods are used to assist various design tasks,\nthere has not been any holistic approach to automate the design of input\ncomponents. My thesis proposed a series of Computational Input Design\nworkflows: I envision a sample-efficient multi-objective optimization algorithm\nthat cleverly selects design instances, which are instantly deployed on\nphysical simulators. A meta-reinforcement learning user model then simulates\nthe user behaviors when using the design instance upon the simulators. The new\nworkflows derive Pareto-optimal designs with high efficiency and automation. I\ndemonstrate designing a push-button via the proposed methods. The resulting\ndesigns outperform the known baselines. The Computational Input Design process\ncan be generalized to other devices, such as joystick, touchscreen, mouse,\ncontroller, etc.\n"}
{"abstract": "  The Higgs sector of the standard model can be extended by introducing an\n$SU(2)_L$ Higgs triplet $\\Delta$ to generate the tiny neutrino masses in the\nframework of type-II seesaw mechanism. In this paper, we study the pair\nproduction of the introduced Higgs triplet at future $e^{-}p$ colliders. The\ncorresponding production cross sections via vector boson fusion process at\nFCC-ep and ILC$\\otimes$FCC are predicted, where the production of a pair of\ndoubly charged Higgs is found to be dominant and then used to investigate the\ncollider phenomenology of the Higgs triplet. Depending on the size of the Higgs\ntriplet vacuum expectation value, the doubly charged Higgs may decay into a\npair of same-sign charged leptons or a pair of same-sign $W$ bosons. In order\nto explore the discovery potential of the doubly charged Higgs at future\n$e^{-}p$ colliders, we discuss these two decay scenarios in detail and show\nrespectively the detection sensitivity on the mass of the doubly charged Higgs.\n"}
{"abstract": "  We introduce a very simple queue implementation with the singly linked list.\nWith the help of the rear blank node instead of the usual header node, we avoid\nadditional check steps for the dequeue operation in the traditional\nimplementations existing for many decades. The essence of our representation is\nthe half-opened pointer interval with the same direction of the queue\noperations, which can guarantee the uniform treatment even in the empty queue\ncase. The simplification of queue implementations cuts off unnecessary steps,\nand it minimizes the number of steps in the dequeue operation with the time\nlimitation of enqueue operation, which could contribute to the performance of\nthe real-time systems. We extend the linked queue to the circularly linked\nqueue, which can also be used to implement stack and take advantage of the\nmaximal information of the single direction in the circularly linked list, and\nit actually constructs the output-restricted deque. We also present a variant:\nlazy circularly linked queue, which is more efficient in some special cases,\nespecially for the dequeue operations.\n"}
{"abstract": "  Graphical calculi such as the ZH-calculus are powerful tools in the study and\nanalysis of quantum processes, with links to other models of quantum\ncomputation such as quantum circuits, measurement-based computing, etc.\n  A somewhat compact but systematic way to describe a quantum process is\nthrough the use of quantum multiple-valued decision diagrams (QMDDs), which\nhave already been used for the synthesis of quantum circuits as well as for\nverification.\n  We show in this paper how to turn a QMDD into an equivalent ZH-diagram, and\nvice-versa, and show how reducing a QMDD translates in the ZH-Calculus, hence\nallowing tools from one formalism to be used into the other.\n"}
{"abstract": "  We propose an effective consistency training framework that enforces a\ntraining model's predictions given original and perturbed inputs to be similar\nby adding a discrete noise that would incur the highest divergence between\npredictions. This virtual adversarial discrete noise obtained by replacing a\nsmall portion of tokens while keeping original semantics as much as possible\nefficiently pushes a training model's decision boundary. Moreover, we perform\nan iterative refinement process to alleviate the degraded fluency of the\nperturbed sentence due to the conditional independence assumption. Experimental\nresults show that our proposed method outperforms other consistency training\nbaselines with text editing, paraphrasing, or a continuous noise on\nsemi-supervised text classification tasks and a robustness benchmark.\n"}
{"abstract": "  We prove that a large class of Paschke dual algebras of simple unital\nC*-algebras are $K_1$-injective. As a consequence, we obtain interesting\n$KK$-uniqueness theorems which generalize the Brown--Douglas--Fillmore\nessential codimension property.\n"}
{"abstract": "  Digital services have been offered through remote systems for decades. The\nquestions of how these systems can be built in a trustworthy manner and how\ntheir security properties can be understood are given fresh impetus by recent\nhardware developments, allowing a fuller, more general, exploration of the\npossibilities than has previously been seen in the literature. Drawing on and\nconsolidating the disparate strains of research, technologies and methods\nemployed throughout the adaptation of confidential computing, we present a\nnovel, dedicated Confidential Remote Computing (CRC) model. CRC proposes a\ncompact solution for next-generation applications to be built on strong\nhardware-based security primitives, control of secure software products'\ntrusted computing base, and a way to make correct use of proofs and evidence\nreports generated by the attestation mechanisms. The CRC model illustrates the\ntrade-offs between decentralisation, task size and transparency overhead. We\nconclude the paper with six lessons learned from our approach, and suggest two\nfuture research directions.\n"}
{"abstract": "  We consider the problem of variance reduction in randomized controlled\ntrials, through the use of covariates correlated with the outcome but\nindependent of the treatment. We propose a machine learning regression-adjusted\ntreatment effect estimator, which we call MLRATE. MLRATE uses machine learning\npredictors of the outcome to reduce estimator variance. It employs\ncross-fitting to avoid overfitting biases, and we prove consistency and\nasymptotic normality under general conditions. MLRATE is robust to poor\npredictions from the machine learning step: if the predictions are uncorrelated\nwith the outcomes, the estimator performs asymptotically no worse than the\nstandard difference-in-means estimator, while if predictions are highly\ncorrelated with outcomes, the efficiency gains are large. In A/A tests, for a\nset of 48 outcome metrics commonly monitored in Facebook experiments the\nestimator has over 70% lower variance than the simple difference-in-means\nestimator, and about 19% lower variance than the common univariate procedure\nwhich adjusts only for pre-experiment values of the outcome.\n"}
{"abstract": "  Let $\\mathcal{A}_1,\\ldots,\\mathcal{A}_m$ be families of $k$-subsets of an\n$n$-set. Suppose that one cannot choose pairwise disjoint edges from $s+1$\ndistinct families. Subject to this condition we investigate the maximum of\n$|\\mathcal{A}_1|+\\ldots+|\\mathcal{A}_m|$. Note that the subcase $m=s+1$,\n$\\mathcal{A}_1=\\ldots=\\mathcal{A}_m$ is the Erd\\H{o}s Matching Conjecture, one\nof the most important open problems in extremal set theory. We provide some\nupper bounds, a general conjecture and its solution for the range $n\\geq\n4k^2s$.\n"}
{"abstract": "  In the cubic chiral magnet Fe$_{1-x}$Co$_{x}$Si a metastable state comprising\nof topologically nontrivial spin whirls, so-called skyrmions, may be preserved\ndown to low temperatures by means of field cooling the sample. This metastable\nskyrmion state is energetically separated from the topologically trivial ground\nstate by a considerable potential barrier, a phenomenon also referred to as\ntopological protection. Using magnetic force microscopy on the surface of a\nbulk crystal, we show that certain positions are preferentially and\nreproducibly decorated with metastable skyrmions, indicating that surface\npinning plays a crucial role. Increasing the magnetic field allows an\nincreasing number of skyrmions to overcome the potential barrier, and hence to\ntransform into the ground state. Most notably, we find that the unwinding of\nindividual skyrmions may be triggered by the magnetic tip itself, however, only\nwhen its magnetization is aligned parallel to the external field. This implies\nthat the stray field of the tip is key for locally overcoming the topological\nprotection. Both the control of the position of topologically nontrivial states\nas well as their creation and annihilation on demand pose important challenges\nin the context of potential skyrmionic applications.\n"}
{"abstract": "  One of the main issues in hadron spectroscopy is to identify the origin of\nthreshold or near-threshold enhancement. Prior to our study, there is no\nstraightforward way of distinguishing even the lowest channel\nthreshold-enhancement of the nucleon-nucleon system using only the\ncross-sections. The difficulty lies in the proximity of either a bound or\nvirtual state pole to the threshold which creates an almost identical structure\nin the scattering region. Identifying the nature of the pole causing the\nenhancement falls under the general classification problem and supervised\nmachine learning using a feed-forward neural network is known to excel in this\ntask. In this study, we discuss the basic idea behind deep neural network and\nhow it can be used to identify the nature of the pole causing the enhancement.\nThe applicability of the trained network can be explored by using an exact\nseparable potential model to generate a validation dataset. We find that within\nsome acceptable range of the cut-off parameter, the neural network gives high\naccuracy of inference. The result also reveals the important role played by the\nbackground singularities in the training dataset. Finally, we apply the method\nto nucleon-nucleon scattering data and show that the network was able to give\nthe correct nature of pole, i.e. virtual pole for ${}^1S_0$ partial\ncross-section and bound state pole for ${}^3S_0$.\n"}
{"abstract": "  Predicting the future interaction of objects when they come into contact with\ntheir environment is key for autonomous agents to take intelligent and\nanticipatory actions. This paper presents a perception framework that fuses\nvisual and tactile feedback to make predictions about the expected motion of\nobjects in dynamic scenes. Visual information captures object properties such\nas 3D shape and location, while tactile information provides critical cues\nabout interaction forces and resulting object motion when it makes contact with\nthe environment. Utilizing a novel See-Through-your-Skin (STS) sensor that\nprovides high resolution multimodal sensing of contact surfaces, our system\ncaptures both the visual appearance and the tactile properties of objects. We\ninterpret the dual stream signals from the sensor using a Multimodal\nVariational Autoencoder (MVAE), allowing us to capture both modalities of\ncontacting objects and to develop a mapping from visual to tactile interaction\nand vice-versa. Additionally, the perceptual system can be used to infer the\noutcome of future physical interactions, which we validate through simulated\nand real-world experiments in which the resting state of an object is predicted\nfrom given initial conditions.\n"}
{"abstract": "  In this paper, a novel convolutional neural network (CNN)-based framework is\ndeveloped for light field reconstruction from a sparse set of views. We\nindicate that the reconstruction can be efficiently modeled as angular\nrestoration on an epipolar plane image (EPI). The main problem in direct\nreconstruction on the EPI involves an information asymmetry between the spatial\nand angular dimensions, where the detailed portion in the angular dimensions is\ndamaged by undersampling. Directly upsampling or super-resolving the light\nfield in the angular dimensions causes ghosting effects. To suppress these\nghosting effects, we contribute a novel \"blur-restoration-deblur\" framework.\nFirst, the \"blur\" step is applied to extract the low-frequency components of\nthe light field in the spatial dimensions by convolving each EPI slice with a\nselected blur kernel. Then, the \"restoration\" step is implemented by a CNN,\nwhich is trained to restore the angular details of the EPI. Finally, we use a\nnon-blind \"deblur\" operation to recover the spatial high frequencies suppressed\nby the EPI blur. We evaluate our approach on several datasets, including\nsynthetic scenes, real-world scenes and challenging microscope light field\ndata. We demonstrate the high performance and robustness of the proposed\nframework compared with state-of-the-art algorithms. We further show extended\napplications, including depth enhancement and interpolation for unstructured\ninput. More importantly, a novel rendering approach is presented by combining\nthe proposed framework and depth information to handle large disparities.\n"}
{"abstract": "  Kernel methods are a highly effective and widely used collection of modern\nmachine learning algorithms. A fundamental limitation of virtually all such\nmethods are computations involving the kernel matrix that naively scale\nquadratically (e.g., constructing the kernel matrix and matrix-vector\nmultiplication) or cubically (solving linear systems) with the size of the data\nset $N.$ We propose the Fast Kernel Transform (FKT), a general algorithm to\ncompute matrix-vector multiplications (MVMs) for datasets in moderate\ndimensions with quasilinear complexity. Typically, analytically grounded fast\nmultiplication methods require specialized development for specific kernels. In\ncontrast, our scheme is based on auto-differentiation and automated symbolic\ncomputations that leverage the analytical structure of the underlying kernel.\nThis allows the FKT to be easily applied to a broad class of kernels, including\nGaussian, Matern, and Rational Quadratic covariance functions and physically\nmotivated Green's functions, including those of the Laplace and Helmholtz\nequations. Furthermore, the FKT maintains a high, quantifiable, and\ncontrollable level of accuracy -- properties that many acceleration methods\nlack. We illustrate the efficacy and versatility of the FKT by providing timing\nand accuracy benchmarks and by applying it to scale the stochastic neighborhood\nembedding (t-SNE) and Gaussian processes to large real-world data sets.\n"}
{"abstract": "  With the growing commercial interest in blockchain, permissioned\nimplementations have received increasing attention. Unfortunately, existing BFT\nconsensus protocols that are the backbone of permissioned blockchains, either\nscale poorly or offer limited throughput. Most of these algorithms require at\nleast one process to receive and validate the votes from all other processes\nand then broadcast the result, which is inherently non-scalable. Some\nalgorithms avoid this bottleneck by using aggregation trees to collect and\nvalidate votes. However, to the best of our knowledge, such algorithms offer\nlimited throughput and degrade quickly in the presence of faults. In this paper\nwe propose \\thesystem, the first BFT communication abstraction that organizes\nparticipants in a tree to perform scalable vote aggregation and that, in faulty\nruns, is able to terminate the protocol within an optimal number of\nreconfigurations ($f+1$). We define precisely which aggregation trees allow for\noptimal reconfiguration and show that, unlike previous protocols, when using\nthese configurations, \\thesystem scales to large number of processes and\noutperforms HotStuff's throughput by up to 38x.\n"}
{"abstract": "  In this article, we propose a method to design loss functions based on\ncomponent trees which can be optimized by gradient descent algorithms and which\nare therefore usable in conjunction with recent machine learning approaches\nsuch as neural networks. We show how the altitudes associated to the nodes of\nsuch hierarchical image representations can be differentiated with respect to\nthe image pixel values. This feature is used to design a generic loss function\nthat can select or discard image maxima based on various attributes such as\nextinction values. The possibilities of the proposed method are demonstrated on\nsimulated and real image filtering.\n"}
{"abstract": "  The analog of the Kepler system defined on the Heisenberg group introduced by\nMontgomery and Shanbrom in [Fields Inst. Commun., Vol. 73, Springer, New York,\n2015, 319-342, arXiv:1212.2713] is integrable on the zero level of the\nHamiltonian. We show that in all other cases the system is not Liouville\nintegrable due to the lack of additional meromorphic first integrals. We prove\nthat the analog of the two-body problem on the Heisenberg group is not\nintegrable in the Liouville sense.\n"}
{"abstract": "  It is shown that the general relativity has a one-parameter compact symmetry\nand this symmetry is analogous to the phase symmetry of quantum mechanics\n"}
{"abstract": "  We theoretically investigate how the spin susceptibility of a planar\nJosephson junction is affected when the system transits into the topological\nsuperconducting state. We show that the magnetic flux and magnetic field\ndependence of the spin susceptibility closely maps the phase diagram of the\nsystem. In the absence of an external magnetic flux the system can self-tune\ninto the topological superconducting state by minimizing its free energy.\nSelf-tuned topological transitions are accompanied by sharp peaks in the spin\nsusceptibility, which can therefore be use as measurable fingerprints for\ndetecting the topological superconducting state. Away from the peaks, the\namplitude of the spin susceptibility can provide qualitative information about\nthe relative size of the topological gap. The signatures in the spin\nsusceptibility are robust, even in junctions with narrow superconducting leads,\nwhere critical current minima may no longer serve as an indication of\ntopological phase transitions. Numerical simulations are complemented with\nsimplified analytical models capable of capturing the main features predicted\nfor the spin susceptibility behavior. The predicted results could be\nparticularly relevant for future experiments on realization and detection of\nthe topological superconducting state in planar Josephson junctions.\n"}
{"abstract": "  We present the first X-ray detections of ultracool dwarfs (UCDs) from the\nfirst all-sky survey of the extended ROentgen Survey with an Imaging Telescope\nArray (eROSITA) onboard the Russian Spektrum-Roentgen-Gamma (SRG) mission. We\nuse three publicly available input catalogs of spectroscopically confirmed UCDs\nand Gaia-selected UCD candidates that together comprise nearly 20000 objects.\nWe first extracted all X-ray sources from the catalog of the first eROSITA\nsurvey, eRASS1, that have a UCD or candidate within three times their\npositional uncertainty. Then we examined all Gaia objects in the vicinity of\nthese 96 X-ray sources and we associated them to the most plausible counterpart\non the basis of their spatial separation to the X-ray position and their\nmultiwavelength properties. This way we find 40 UCDs that have a secure\nidentification with an X-ray source and 18 plausible UCD X-ray emitters.\nTwenty-one of these X-ray emitting UCDs have a spectroscopic confirmation,\nwhile the others have been selected based on Gaia photometry and we computed\nspectral types from the G-J color. The spectral types of the X-ray emitting\nUCDs and candidates range between M5 and M9, and the distances range from 3.5\nto 190 pc. The majority of the UCDs from the eRASS1 sample show a ratio of\nX-ray to bolometric luminosity well above the canonical saturation limit of log\n(Lx/Lbol) ~ -3. For the two most extreme outliers, we showed through an\nanalysis of the eRASS1 light curve that these high values are due to flaring\nactivity. The X-ray spectra of the two brightest objects both reveal an\nemission-measure weighted plasma temperature of kT ~ 0.75 keV. These\nobservations demonstrate the potential of eROSITA for advancing our knowledge\non the faint coronal X-ray emission from UCDs by building statistical samples\nfor which the average X-ray brightness, flares, and coronal temperatures can be\nderived.\n"}
{"abstract": "  Let $G$ be a finite group, and let $\\Delta(G)$ be the prime graph built on\nits set of conjugacy class sizes: this is the (simple undirected) graph whose\nvertices are the prime numbers dividing some conjugacy class size of $G$, and\ntwo distinct vertices $p, q$ are adjacent if and only if $pq$ divides some\nclass size of $G$. In this paper, we characterise the structure of those groups\n$G$ whose prime graph $\\Delta(G)$ is a block square.\n"}
{"abstract": "  We investigate the magnetic nanoparticles (fluid) hyperthermia in\nnon-adiabatic conditions through the calorimetric method. Specifically, we\npropose a theoretical approach to magnetic hyperthermia from a thermodynamic\npoint of view. To test the robustness of the approach, we perform hyperthermia\nexperiments and analyze the thermal behavior of magnetite and magnesium ferrite\nmagnetic nanoparticles dispersed in water submitted to an alternating magnetic\nfield. From our findings, besides estimating the specific loss power value from\na non-adiabatic process, thus enhancing the accuracy in the determination of\nthis quantity, we provide physical meaning to parameters found in literature\nthat still remained not fully understood, and bring to light how they can be\nobtained experimentally.\n"}
{"abstract": "  We study possibilities of creation and detection of oscillating gravitational\nfields from lab-scale high energy, relativistic sources. The sources considered\nare high energy laser beams in an optical cavity and the ultra-relativistic\nproton bunches circulating in the beam of the Large Hadron Collider (LHC) at\nCERN. These sources allow for signal frequencies much higher and far narrower\nin bandwidth than what most celestial sources produce. In addition, by\nmodulating the beams, one can adjust the source frequency over a very broad\nrange, from Hz to GHz. The gravitational field of these sources and responses\nof a variety of detectors are analyzed.\n  We optimize a mechanical oscillator such as a pendulum or torsion balance as\ndetector and find parameter regimes such that -- combined with the planned\nhigh-luminosity upgrade of the LHC as a source -- a signal-to-noise ratio\nsubstantially larger than 1 should be achievable at least in principle,\nneglecting all sources of technical noise. This opens new perspectives of\nstudying general relativistic effects and possibly quantum-gravitational\neffects with ultra-relativistic, well-controlled terrestrial sources.\n"}
{"abstract": "  We theoretically and numerically study the terahertz auto oscillations in\nthin-film metallic non-collinear coplanar antiferromagnets (AFMs), such as\n$\\mathrm{Mn_{3}Sn}$ and $\\mathrm{Mn_{3}Ir}$, under the effect of anti-damping\nspin-torque with spin polarization perpendicular to the plane of the film. To\nobtain the order parameter dynamics in these AFMs, we solve three\nLandau-Lifshitz-Gilbert equations coupled by exchange interactions assuming\nboth single- and multi-domain (micromagnetics) dynamical processes. In the\nlimit of strong exchange interaction, the oscillatory dynamics of the order\nparameter in these AFMs, which have opposite chiralities, could be mapped to\nthat of a linear damped-driven pendulum in the case of $\\mathrm{Mn_{3}Sn}$, and\na non-linear damped-driven pendulum in case of $\\mathrm{Mn_{3}Ir}$. The\ntheoretical framework allows us to identify the input current requirements as a\nfunction of the material and geometry parameters for exciting an oscillatory\nresponse. We also obtain a closed-form approximate solution of the oscillation\nfrequency for large input currents in case of both $\\mathrm{Mn_{3}Ir}$ and\n$\\mathrm{Mn_{3}Sn}$. Our analytical predictions of threshold current and\noscillation frequency agree well with the numerical results and thus can be\nused as compact models to design and optimize the auto oscillator. Employing a\ncircuit model, based on the principle of tunnel anisotropy magnetoresistance,\nwe present detailed models of the output power and efficiency versus\noscillation frequency of the auto oscillator. Finally, we explore the spiking\ndynamics of two unidirectional as well as bidirectional coupled AFM oscillators\nusing non-linear damped-driven pendulum equations.\n"}
{"abstract": "  We investigate the training of sparse layers that use different parameters\nfor different inputs based on hashing in large Transformer models.\nSpecifically, we modify the feedforward layer to hash to different sets of\nweights depending on the current token, over all tokens in the sequence. We\nshow that this procedure either outperforms or is competitive with\nlearning-to-route mixture-of-expert methods such as Switch Transformers and\nBASE Layers, while requiring no routing parameters or extra terms in the\nobjective function such as a load balancing loss, and no sophisticated\nassignment algorithm. We study the performance of different hashing techniques,\nhash sizes and input features, and show that balanced and random hashes focused\non the most local features work best, compared to either learning clusters or\nusing longer-range context. We show our approach works well both on large\nlanguage modeling and dialogue tasks, and on downstream fine-tuning tasks.\n"}
{"abstract": "  We prove a factorization formula for the Taylor series coefficients of a zero\nof a polynomial as a function of the polynomial's coefficients. This result\nextends to more general functions which we call \"complex-exponent polynomials\".\nTo prove this formula, we prove theorems about derivations on commutative\nrings. We also show that, when applied to polynomials, our formula recovers the\nresults of Sturmfels obtained with GKZ systems (\"Solving algebraic equations in\nterms of $\\mathcal{A}$-hypergeometric series\". Discrete Math. 210 (2000) pp.\n171-181)\n"}
{"abstract": "  Motivated by work of Erd\\H{o}s, Ota determined the maximum size $g(n,k)$ of a\n$k$-connected nonhamiltonian graph of order $n$ in 1995. But for some pairs\n$n,k,$ the maximum size is not attained by a graph of connectivity $k.$ For\nexample, $g(15,3)=77$ is attained by a unique graph of connectivity $7,$ not\n$3.$ In this paper we obtain more precise information by determining the\nmaximum size of a nonhamiltonian graph of order $n$ and connectivity $k,$ and\ndetermining the extremal graphs. Consequently we solve the corresponding\nproblem for nontraceable graphs.\n"}
{"abstract": "  The ongoing COVID-19 pandemic has profoundly impacted people's life around\nthe world, including how they interact with mobile technologies. In this paper,\nwe seek to develop an understanding of how the dynamic trajectory of a pandemic\nshapes mobile phone users' experiences. Through the lens of app popularity, we\napproach this goal from a cross-country perspective. We compile a dataset\nconsisting of six-month daily snapshots of the most popular apps in the iOS App\nStore in China and the US, where the pandemic has exhibited distinct\ntrajectories. Using this longitudinal dataset, our analysis provides detailed\npatterns of app ranking during the pandemic at both category and individual app\nlevels. We reveal that app categories' rankings are correlated with the\npandemic, contingent upon country-specific development trajectories. Our work\noffers rich insights into how the COVID-19, a typical global public health\ncrisis, has influence people's day-to-day interaction with the Internet and\nmobile technologies.\n"}
{"abstract": "  We use spherical harmonic expansions to prove that any star body equal to its\nintersection body is a ball about the origin.\n"}
{"abstract": "  The performance of modern electronics is associated with the multi-layered\ninterconnects, encouraging the development of a low-k dielectrics. Herein, we\nstudied the effects of phase transition from crystalline to amorphous on\ndielectric, optical and electrical properties of MIL-100 (Fe) and Basolite F300\nmetal-organic framework (MOF) obtained using different synthesis techniques in\nboth the radio (4-1.5 MHz) and infrared (1.2-150 THz) frequency regimes, which\nare important for the microelectronics, infrared optical sensors, and\nhigh-frequency telecommunications. The impact of amorphization on broadband\ndielectric response was established based on: (1) The comparison in the\ndielectric characteristics of a commercially available amorphous Basolite F300\nversus a mechanochemically synthesized crystalline MIL-100 in the MHz region.\n(2) By tracking the frequency shifts in the vibrational modes of MIL-100\nstructure in the far-IR (phonons) and mid-IR regions. We showed that various\nparameters such as pelleting pressure, temperature, frequency, density, and\ndegree of amorphization greatly affect the dielectric properties of the\nframework. We also investigated the influence of temperature (20-100 {\\deg}C)\non the electric and dielectric response in the MHz region, crucial for all\nelectronic devices.\n"}
{"abstract": "  Results from the last FTU campaigns on the deuterium large (wrt FTU volume)\npellet REs suppression capability, mainly due to the induced burst MHD activity\nexpelling REs seed are presented for discharges with 0.5 MA and 5.3T. Clear\nindications of avalanche multiplication of REs following single pellet\ninjection on 0.36 MA flat-top discharges is shown together with quantitative\nindications of dissipative effects in terms of critical electrical field\nincrease due to fan-like instabilities. Analysis of large fan-like\ninstabilities on post-disruption RE beams, that seem to be correlated with low\nelectrical field and background density drops, reveal their strong RE energy\nsuppression capability suggesting a new strategy for RE energy suppression\ncontrolling large fan instabilities. We demonstrate how such density drops can\nbe induced using modulated ECRH power on post-disruption beams.\n"}
{"abstract": "  In this paper, we propose dualFace, a portrait drawing interface to assist\nusers with different levels of drawing skills to complete recognizable and\nauthentic face sketches. dualFace consists of two-stage drawing assistance to\nprovide global and local visual guidance: global guidance, which helps users\ndraw contour lines of portraits (i.e., geometric structure), and local\nguidance, which helps users draws details of facial parts (which conform to\nuser-drawn contour lines), inspired by traditional artist workflows in portrait\ndrawing. In the stage of global guidance, the user draws several contour lines,\nand dualFace then searches several relevant images from an internal database\nand displays the suggested face contour lines over the background of the\ncanvas. In the stage of local guidance, we synthesize detailed portrait images\nwith a deep generative model from user-drawn contour lines, but use the\nsynthesized results as detailed drawing guidance. We conducted a user study to\nverify the effectiveness of dualFace, and we confirmed that dualFace\nsignificantly helps achieve a detailed portrait sketch. see\nhttp://www.jaist.ac.jp/~xie/dualface.html\n"}
{"abstract": "  Many of the models used nowadays in mathematical epidemiology, in particular\nin COVID-19 research, belong to a certain sub-class of compartmental models\nwhose classes may be divided into three \"(x, y, z)\" groups, which we will call\nrespectively \"susceptible/entrance, diseased, and output\" (in the classic SIR\ncase, there is only one class of each type). Roughly, the ODE dynamics of these\nmodels contain only linear terms, with the exception of products between x and\ny terms. It has long been noticed that the basic reproduction number R has a\nvery simple formula (3.3) in terms of the matrices which define the model, and\nan explicit first integral formula (3.8) is also available. These results can\nbe traced back at least to [ABvdD+07] and [Fen07], respectively, and may be\nviewed as the \"basic laws of SIR-type epidemics\"; however many papers continue\nto reprove them in particular instances (by the next-generation matrix method\nor by direct computations, which are unnecessary). This motivated us to redraw\nthe attention to these basic laws and provide a self-contained reference of\nrelated formulas for (x, y, z) models. We propose to rebaptize the class to\nwhich they apply as matrix SIR epidemic models, abbreviated as SYR, to\nemphasize the similarity to the classic SIR case. For the case of one\nsusceptible class, we propose to use the name SIR-PH, due to a simple\nprobabilistic interpretation as SIR models where the exponential infection time\nhas been replaced by a PH-type distribution. We note that to each SIR-PH model,\none may associate a scalar quantity Y(t) which satisfies \"classic SIR\nrelations\", see (3.8). In the case of several susceptible classes, this\ngeneralizes to (5.10); in a future paper, we will show that (3.8), (5.10) may\nbe used to obtain approximate control policies which compare well with the\noptimal control of the original model.\n"}
{"abstract": "  A simple and natural explanation for the minimum period of millisecond\npulsars follows from a correlation between the accretion rate and the frozen\nsurface dipole magnetic field resulting from Ohmic diffusion through the\nneutron star crust in initial stages of accretion in low mass X-ray binaries.\n"}
{"abstract": "  About two dozens of exactly solvable Markov chains on one-dimensional finite\nand semi-infinite integer lattices are constructed in terms of convolutions of\northogonality measures of the Krawtchouk, Hahn, Meixner, Charlier, $q$-Hahn,\n$q$-Meixner and little $q$-Jacobi polynomials. By construction, the stationary\nprobability distributions, the complete sets of eigenvalues and eigenvectors\nare provided by the polynomials and the orthogonality measures. An interesting\nproperty possessed by these stationary probability distributions, called\n`convolutional self-similarity,' is demonstrated.\n"}
{"abstract": "  In threshold-based anomaly detection, we want to tune the threshold of a\ndetector to achieve an acceptable false alarm rate. However, tuning the\nthreshold is often a non-trivial task due to unknown detector output\ndistributions. A detector threshold that provides an acceptable false alarm\nrate is equivalent to a specific quantile of the detector output distribution.\nTherefore, we use quantile estimators based on order statistics to estimate the\ndetector threshold. The estimation of quantiles from sample data has a more\nthan a century long tradition and we provide three different distribution-free\nfinite sample guarantees for a class of quantile estimators. The first is based\non the Dworetzky-Kiefer-Wolfowitz inequality, the second utilizes Chebyshev's\ninequality, and the third is based on exact confidence intervals for a binomial\ndistribution. These guarantees are then compared and used in the detector\nthreshold tuning problem. We use both simulated data as well as data obtained\nfrom an experimental setup with the Temperature Control Lab to validate the\nguarantees provided.\n"}
{"abstract": "  We derive the optimal signed variable in general case kernels for the\nclassical statistic density estimation, which are some generalization of the\nfamous Epanechnikov's ones.\n"}
{"abstract": "  We give a brief review of the current understanding of renormalons of the\nstatic QCD potential in coordinate and momentum spaces. We also reconsider\nestimate of the normalization constant of the $u=3/2$ renormalon and propose a\nnew way to improve the estimate.\n"}
{"abstract": "  With the instantaneous growth of text information, retrieving domain-oriented\ninformation from the text data has a broad range of applications in Information\nRetrieval and Natural language Processing. Thematic keywords give a compressed\nrepresentation of the text. Usually, Domain Identification plays a significant\nrole in Machine Translation, Text Summarization, Question Answering,\nInformation Extraction, and Sentiment Analysis. In this paper, we proposed the\nMultichannel LSTM-CNN methodology for Technical Domain Identification for\nTelugu. This architecture was used and evaluated in the context of the ICON\nshared task TechDOfication 2020 (task h), and our system got 69.9% of the F1\nscore on the test dataset and 90.01% on the validation set.\n"}
{"abstract": "  Recent studies constructing direct interactions between the claim and each\nsingle user response (a comment or a relevant article) to capture evidence have\nshown remarkable success in interpretable claim verification. Owing to\ndifferent single responses convey different cognition of individual users\n(i.e., audiences), the captured evidence belongs to the perspective of\nindividual cognition. However, individuals' cognition of social things is not\nalways able to truly reflect the objective. There may be one-sided or biased\nsemantics in their opinions on a claim. The captured evidence correspondingly\ncontains some unobjective and biased evidence fragments, deteriorating task\nperformance. In this paper, we propose a Dual-view model based on the views of\nCollective and Individual Cognition (CICD) for interpretable claim\nverification. From the view of the collective cognition, we not only capture\nthe word-level semantics based on individual users, but also focus on\nsentence-level semantics (i.e., the overall responses) among all users and\nadjust the proportion between them to generate global evidence. From the view\nof individual cognition, we select the top-$k$ articles with high degree of\ndifference and interact with the claim to explore the local key evidence\nfragments. To weaken the bias of individual cognition-view evidence, we devise\ninconsistent loss to suppress the divergence between global and local evidence\nfor strengthening the consistent shared evidence between the both. Experiments\non three benchmark datasets confirm that CICD achieves state-of-the-art\nperformance.\n"}
{"abstract": "  Introduces HIVE-4-MAT - Helping Interdisciplinary Vocabulary Engineering for\nMaterials Science, an automatic linked data ontology application. Covers\ncontextual background for materials science, shared ontology infrastructures,\nand reviews the knowledge extraction and indexing process. HIVE-4-MAT's\nvocabulary browsing, term search and selection, and knowledge extraction and\nindexing are reviewed, and plans to integrate named entity recognition.\nConclusion highlights next steps with relation extraction to support better\nontologies.\n"}
{"abstract": "  The main motivation of this work is to assess the validity of a LWR traffic\nflow model to model measurements obtained from trajectory data, and propose\nextensions of this model to improve it. A formulation for a discrete dynamical\nsystem is proposed aiming at reproducing the evolution in time of the density\nof vehicles along a road, as observed in the measurements. This system is\nformulated as a chemical reaction network where road cells are interpreted as\ncompartments, the transfer of vehicles from one cell to the other is seen as a\nchemical reaction between adjacent compartment and the density of vehicles is\nseen as a concentration of reactant. Several degrees of flexibility on the\nparameters of this system, which basically consist of the reaction rates\nbetween the compartments, can be considered: a constant value or a function\ndepending on time and/or space. Density measurements coming from trajectory\ndata are then interpreted as observations of the states of this system at\nconsecutive times. Optimal reaction rates for the system are then obtained by\nminimizing the discrepancy between the output of the system and the state\nmeasurements. This approach was tested both on simulated and real data, proved\nsuccessful in recreating the complexity of traffic flows despite the\nassumptions on the flux-density relation.\n"}
{"abstract": "  In this paper, we derive the information theoretic performance bounds on\ncommunication data rates and errors in parameter estimation, for a joint radar\nand communication (JRC) system. We assume that targets are semi-passive, i.e.\nthey use active components for signal reception, and passive components to\ncommunicate their own information. Specifically, we let the targets to have\ncontrol over their passive reflectors in order to transmit their own\ninformation back to the radar via reflection-based beamforming or\nbackscattering. We derive the Cramer-Rao lower bounds (CRBs) for the mean\nsquared error in the estimation of target parameters. The concept of a target\nambiguity function (TAF) arises naturally in the derivation CRBs. Using these\nTAFs as cost function, we propose a waveform optimization technique based on\ncalculus of variations. Further, we derive lower bounds on the data rates for\ncommunication on forward and reverse channels, in radar-only and joint radar\nand communications scenarios. Through numerical examples, we demonstrate the\nutility of this framework for transmit waveform design, codebook construction,\nand establishing the corresponding data rate bounds.\n"}
{"abstract": "  Recently single-beam intracavity optical tweezers have been reported and\nachieved orders-of-magnitude higher confinement than standard optical tweezers.\nHowever, there is only one feedback loop between the trapped particle's\nthree-dimensional position and the scattering loss of the intracavity laser.\nThat leads to the coupling effect between the particle's radial and axial\nmotion, and aggravates the axial confinement efficiency. Here, we present and\ndemonstrate the dual-beam intracavity optical trap enabling independent radial\nand axial self-feedback control of the trapped particle, through offsetting the\nfoci of the clockwise and counter-clockwise beams. We have achieved the axial\nconfinement efficiency of 1.6*10^(-4) mW^(-1) experimentally at very low\nnumerical aperture (NA=0.25), which is the highest axial confinement efficiency\nof the optical trap to date, to the best of our knowledge. The dual-beam\nintracavity optical trap will significantly expand the range of applications in\nthe further studies of biology and physics, especially for a sample that is\nextremely sensitive to heat.\n"}
{"abstract": "  Here we present a liquid nitrogen-cooled Ca^+ optical clock with an overall\nsystematic uncertainty of 3*10^-18. In contrast with the room-temperature Ca^+\noptical clock that we have reported previously, the temperature of the\nblackbody radiation (BBR) shield in vacuum has been reduced to 82(5) K using\nliquid nitrogen. An ion trap with a lower heating rate and improved cooling\nlasers were also introduced. This allows cooling the ion temperature to the\nDoppler cooling limit during the clock operation, and the systematic\nuncertainty due to the ion's secular (thermal) motion is reduced to < 1*10^-18.\nThe uncertainty due to the probe laser light shift and the servo error are also\nreduced to < 1*10^-19 and 4*10^-19 with the hyper-Ramsey method and the\nhigher-order servo algorithm, respectively. By comparing the output frequency\nof the cryogenic clock to that of a room-temperature clock, the differential\nBBR shift between the two was measured with a fractional statistical\nuncertainty of 7*10^-18. The differential BBR shift was used to calculate the\nstatic differential polarizability, and it was found in excellent agreement\nwith our previous measurement with a different method. This work suggests that\nthe BBR shift of optical clocks can be well suppressed in a liquid nitrogen\nenvironment. This is advantageous because conventional liquid-helium cryogenic\nsystems for optical clocks are more expensive and complicated. Moreover, the\nproposed system can be used to suppress the BBR shift significantly in other\ntypes of optical clocks such as Yb^+, Sr^+, Yb, Sr, etc.\n"}
{"abstract": "  By linking to external websites, Wikipedia can act as a gateway to the Web.\nTo date, however, little is known about the amount of traffic generated by\nWikipedia's external links. We fill this gap in a detailed analysis of usage\nlogs gathered from Wikipedia users' client devices. Our analysis proceeds in\nthree steps: First, we quantify the level of engagement with external links,\nfinding that, in one month, English Wikipedia generated 43M clicks to external\nwebsites, in roughly even parts via links in infoboxes, cited references, and\narticle bodies. Official links listed in infoboxes have by far the highest\nclick-through rate (CTR), 2.47% on average. In particular, official links\nassociated with articles about businesses, educational institutions, and\nwebsites have the highest CTR, whereas official links associated with articles\nabout geographical content, television, and music have the lowest CTR. Second,\nwe investigate patterns of engagement with external links, finding that\nWikipedia frequently serves as a stepping stone between search engines and\nthird-party websites, effectively fulfilling information needs that search\nengines do not meet. Third, we quantify the hypothetical economic value of the\nclicks received by external websites from English Wikipedia, by estimating that\nthe respective website owners would need to pay a total of $7--13 million per\nmonth to obtain the same volume of traffic via sponsored search. Overall, these\nfindings shed light on Wikipedia's role not only as an important source of\ninformation, but also as a high-traffic gateway to the broader Web ecosystem.\n"}
{"abstract": "  Soft robots, in contrast to their rigid counter parts, have infinite degrees\nof freedom that are coupled with their interaction with the environment. We\nconsider the locomotion of an untethered robot, in the granular medium,\ncomprised of multiple flexible flagella that rotate about an axis by a motor.\nDrag from the grains causes the flagella to deform and the deformed shape\ngenerates a net forward propulsion. This external drag force depends on the\nshape of the flagella, while the change in flagellar shape is the result of the\ncompetition between the external loading and elastic forces. We introduce a\nnumerical tool that couples discrete differential geometry based simulation of\nelastic rods - our model for flagella - and a resistive force theory based\nmodel for the drag. In parallel with simulations, we conduct experiments to\nquantify the propulsive speed of this class of robots. We find reasonable\nquantitative agreement between experiments and simulations. Owing to a\nrod-based kinematic representation of the robot, the simulation runs faster\nthan real-time, and, therefore, we can use it as a design tool for this class\nof soft robots. We find that there is an optimal rotational speed at which\nmaximum efficiency is achieved. Moreover, both experiments and simulations show\nthat increasing the number of flagella decreases the speed of the robot. We\nalso gain insight into the mechanics of granular medium - while resistive force\ntheory can successfully describe the propulsion at low number of flagella, it\nfails when more flagella are added to the robot.\n"}
{"abstract": "  Automated proof search with connection tableaux, such as implemented by\nOtten's leanCoP prover, depends on backtracking for completeness. Otten's\nrestricted backtracking strategy loses completeness, yet for many problems, it\nsignificantly reduces the time required to find a proof. I introduce a new,\nless restricted backtracking strategy based on the notion of exclusive cuts. I\nimplement the strategy in a new prover called meanCoP and show that it greatly\nimproves upon the previous best strategy in leanCoP.\n"}
{"abstract": "  This paper proposes a flexible and analytically tractable class of\nfrequency-severity models based on neural networks to parsimoniously capture\nimportant empirical observations. In the proposed two-part model, mean\nfunctions of frequency and severity distributions are characterized by neural\nnetworks to incorporate the non-linearity of input variables. Furthermore, it\nis assumed that the mean function of the severity distribution is an affine\nfunction of the frequency variable to account for a potential linkage between\nfrequency and severity. We provide explicit closed-form formulas for the mean\nand variance of the aggregate loss within our modelling framework. Components\nof the proposed model including parameters of neural networks and distribution\nparameters can be estimated by minimizing the associated negative\nlog-likelihood functionals with neural network architectures. Furthermore, we\nleverage the Shapely value and recent developments in machine learning to\ninterpret the outputs of the model. Applications to a synthetic dataset and\ninsurance claims data illustrate that our method outperforms the existing\nmethods in terms of interpretability and predictive accuracy.\n"}
{"abstract": "  The current work is motivated by the need for robust statistical methods for\nprecision medicine; as such, we address the need for statistical methods that\nprovide actionable inference for a single unit at any point in time. We aim to\nlearn an optimal, unknown choice of the controlled components of the design in\norder to optimize the expected outcome; with that, we adapt the randomization\nmechanism for future time-point experiments based on the data collected on the\nindividual over time. Our results demonstrate that one can learn the optimal\nrule based on a single sample, and thereby adjust the design at any point t\nwith valid inference for the mean target parameter. This work provides several\ncontributions to the field of statistical precision medicine. First, we define\na general class of averages of conditional causal parameters defined by the\ncurrent context for the single unit time-series data. We define a nonparametric\nmodel for the probability distribution of the time-series under few\nassumptions, and aim to fully utilize the sequential randomization in the\nestimation procedure via the double robust structure of the efficient influence\ncurve of the proposed target parameter. We present multiple\nexploration-exploitation strategies for assigning treatment, and methods for\nestimating the optimal rule. Lastly, we present the study of the data-adaptive\ninference on the mean under the optimal treatment rule, where the target\nparameter adapts over time in response to the observed context of the\nindividual. Our target parameter is pathwise differentiable with an efficient\ninfluence function that is doubly robust - which makes it easier to estimate\nthan previously proposed variations. We characterize the limit distribution of\nour estimator under a Donsker condition expressed in terms of a notion of\nbracketing entropy adapted to martingale settings.\n"}
{"abstract": "  The interpretation of the nuclear cross sections measured using accelerator\nneutrino beams involve severe difficulties, arising primarily from the average\nover the incoming neutrino flux. The broad energy distribution of the beam\nparticles hampers the determination of the energy transfer to the nuclear\ntarget, the knowledge of which is needed to pin down the dominant reaction\nmechanism. Overcoming this problem requires the development of a theoretical\napproach suitable to describe neutrino interactions at energies ranging from\nhundreds of MeV to few GeV. In this paper, it is argued that the approach based\non the factorisation of the nuclear cross section provides a consistent\nframework for the calculation of neutrino-nucleus interactions in both the\nquasi elastic and inelastic channels. The near-degeneracy between theoretical\nmodels based on different assumptions, and the use of electron scattering data\nto advance the understanding of neutrino-nucleus cross sections are also\ndiscussed.\n"}
{"abstract": "  In Savage's classic decision-theoretic framework, actions are formally\ndefined as functions from states to outcomes. But where do the state space and\noutcome space come from? Expanding on recent work by Blume, Easley, and Halpern\n(BEH), we consider a language-based framework in which actions are identified\nwith (conditional) descriptions in a simple underlying language, while states\nand outcomes (along with probabilities and utilities) are constructed as part\nof a representation theorem. Our work expands the role of language from that of\nBEH by using it not only for the conditions that determine which actions are\ntaken, but also the effects. More precisely, we take the set of actions to be\nbuilt from those of the form \"do(phi)\", for formulas phi in the underlying\nlanguage. This presents a problem: how do we interpret the result of do(phi)\nwhen phi is underspecified (i.e., compatible with multiple states)? We answer\nthis using tools familiar from the semantics of counterfactuals: roughly\nspeaking, do(phi) maps each state to the \"closest\" phi-state. This notion of\n\"closest\" is also something we construct as part of the representation theorem;\nin effect, then, we prove that (under appropriate assumptions) the agent is\nacting as if each underspecified action is first made definite and then\nevaluated (i.e., by maximizing expected utility). Of course, actions in the\nreal world are often not presented in a fully precise manner, yet agents reason\nabout and form preferences among them all the same. Our work brings the\nabstract tools of decision theory into closer contact with such real-world\nscenarios.\n"}
{"abstract": "  We investigate a class of weighted planar stochastic lattice (WPSL1) created\nby random sequential nucleation of seed from which a crack is grown parallel to\none of the sides of the chosen block and ceases to grow upon hitting another\ncrack. It results in the partitioning of the square into contiguous and\nnon-overlapping blocks. Interestingly, we find that the dynamics of WPSL1 is\ngoverned by infinitely many conservation laws and each of the conserved\nquantities, except the trivial conservation of total mass or area, is a\nmultifractal measure. On the other hand, the dual of the lattice is a\nscale-free network as its degree distribution exhibits a power-law $P(k)\\sim\nk^{-\\gamma}$ with $\\gamma=4.13$. The network is also a small-world network as\nwe find that (i) the total clustering coefficient $C$ is high and independent\nof the network size and (ii) the mean geodesic path length grows\nlogarithmically with $N$. Besides, the clustering coefficient $C_k$ of the\nnodes which have degree $k$ decreases exactly as $2/(k-1)$ revealing that it is\nalso a nested hierarchical network.\n"}
{"abstract": "  Hypergraphs are a common model for multiway relationships in data, and\nhypergraph semi-supervised learning is the problem of assigning labels to all\nnodes in a hypergraph, given labels on just a few nodes. Diffusions and label\nspreading are classical techniques for semi-supervised learning in the graph\nsetting, and there are some standard ways to extend them to hypergraphs.\nHowever, these methods are linear models, and do not offer an obvious way of\nincorporating node features for making predictions. Here, we develop a\nnonlinear diffusion process on hypergraphs that spreads both features and\nlabels following the hypergraph structure, which can be interpreted as a\nhypergraph equilibrium network. Even though the process is nonlinear, we show\nglobal convergence to a unique limiting point for a broad class of\nnonlinearities, which is the global optimum of a interpretable, regularized\nsemi-supervised learning loss function. The limiting point serves as a node\nembedding from which we make predictions with a linear model. Our approach is\nmuch more accurate than several hypergraph neural networks, and also takes less\ntime to train.\n"}
{"abstract": "  The ability to persuade others is critical to professional and personal\nsuccess. However, crafting persuasive messages is demanding and poses various\nchallenges. We conducted nine exploratory case studies to identify adaptations\nthat professional and non-professional writers make in written scenarios to\nincrease their subjective persuasiveness. Furthermore, we identified challenges\nthat those writers faced and identified strategies to resolve them with\npersuasive natural language generation, i.e., artificial intelligence. Our\nfindings show that humans can achieve high degrees of persuasiveness (more so\nfor professional-level writers), and artificial intelligence can complement\nthem to achieve increased celerity and alignment in the process.\n"}
{"abstract": "  The widespread deployment of deep nets in practical applications has lead to\na growing desire to understand how and why such black-box methods perform\nprediction. Much work has focused on understanding what part of the input\npattern (an image, say) is responsible for a particular class being predicted,\nand how the input may be manipulated to predict a different class. We focus\ninstead on understanding which of the internal features computed by the neural\nnet are responsible for a particular class. We achieve this by mimicking part\nof the neural net with an oblique decision tree having sparse weight vectors at\nthe decision nodes. Using the recently proposed Tree Alternating Optimization\n(TAO) algorithm, we are able to learn trees that are both highly accurate and\ninterpretable. Such trees can faithfully mimic the part of the neural net they\nreplaced, and hence they can provide insights into the deep net black box.\nFurther, we show we can easily manipulate the neural net features in order to\nmake the net predict, or not predict, a given class, thus showing that it is\npossible to carry out adversarial attacks at the level of the features. These\ninsights and manipulations apply globally to the entire training and test set,\nnot just at a local (single-instance) level. We demonstrate this robustly in\nthe MNIST and ImageNet datasets with LeNet5 and VGG networks.\n"}
{"abstract": "  We review complicated problems in the Lifshitz theory describing the Casimir\nforce between real material plates made of metals and dielectrics including\ndifferent approaches to their resolution. It has been shown that both for\nmetallic plates with perfect crystal lattices and for any dielectric plates the\nCasimir entropy calculated in the framework of the Lifshitz theory violates the\nNernst heat theorem when the well approved dielectric functions are used in\ncomputations. The respective theoretical Casimir forces are excluded by the\nmeasurement data of numerous precision experiments. In the literature this\nsituation received the names of the Casimir puzzle and the Casimir conundrum\nfor the cases of metallic and dielectric plates, respectively. The review\npresents a summary of the main facts on this subject on both theoretical and\nexperimental sides. Next, we discuss the main approaches proposed in the\nliterature in order to bring the Lifshitz theory in agreement with the\nmeasurement data and with the laws of thermodynamics. Special attention is paid\nto the recently suggested spatially nonlocal Drude-like response functions\nwhich take into account the relaxation properties of conduction electrons, as\ndoes the standard Drude model, but lead to the theoretical results in agreement\nwith both thermodynamics and the measurement data through the alternative\nresponse to quantum fluctuations off the mass shell. Further advances and\ntrends in this field of research are discussed.\n"}
{"abstract": "  The dynamics of the 4-body problem allows for two binary collisions to occur\nsimultaneously. It is known that in the collinear 4-body problem this\nsimultaneous binary collision (SBC) can be block-regularised, but that the\nresulting block map is only $C^{8/3}$ differentiable. In this paper, it is\nproved that the $C^{8/3}$ differentiability persists for the SBC in the planar\n4-body problem. The proof uses several geometric tools, namely, blow-up, normal\nforms, dynamics near normally hyperbolic manifolds of equilibrium points, and\nDulac maps.\n"}
{"abstract": "  Eluder dimension and information gain are two widely used methods of\ncomplexity measures in bandit and reinforcement learning. Eluder dimension was\noriginally proposed as a general complexity measure of function classes, but\nthe common examples of where it is known to be small are function spaces\n(vector spaces). In these cases, the primary tool to upper bound the eluder\ndimension is the elliptic potential lemma. Interestingly, the elliptic\npotential lemma also features prominently in the analysis of linear\nbandits/reinforcement learning and their nonparametric generalization, the\ninformation gain. We show that this is not a coincidence -- eluder dimension\nand information gain are equivalent in a precise sense for reproducing kernel\nHilbert spaces.\n"}
{"abstract": "  Instance segmentation is an active topic in computer vision that is usually\nsolved by using supervised learning approaches over very large datasets\ncomposed of object level masks. Obtaining such a dataset for any new domain can\nbe very expensive and time-consuming. In addition, models trained on certain\nannotated categories do not generalize well to unseen objects. The goal of this\npaper is to propose a method that can perform unsupervised discovery of\nlong-tail categories in instance segmentation, through learning instance\nembeddings of masked regions. Leveraging rich relationship and hierarchical\nstructure between objects in the images, we propose self-supervised losses for\nlearning mask embeddings. Trained on COCO dataset without additional\nannotations of the long-tail objects, our model is able to discover novel and\nmore fine-grained objects than the common categories in COCO. We show that the\nmodel achieves competitive quantitative results on LVIS as compared to the\nsupervised and partially supervised methods.\n"}
{"abstract": "  Temporally shaping the density of electron beams using light forms the basis\nfor a wide range of established and emerging technologies, including\nfree-electron lasers and attosecond electron microscopy. The modulation depth\nof compressed electron pulses is a key figure of merit limiting applications.\nIn this work, we present an approach for generating background-free attosecond\nelectron pulse trains by sequential inelastic electron-light scattering.\nHarnessing quantum interference in the fractional Talbot effect, we suppress\nunwanted background density in electron compression by several orders of\nmagnitude. Our results will greatly enhance applications of coherent\nelectron-light scattering, such as stimulated cathodoluminescence and\nstreaking.\n"}
{"abstract": "  Suspensions of motile active particles with space dependent activity form\ncharacteristic polarization and density patterns. Recent single-particle\nstudies for planar activity landscapes identified several quantities associated\nwith emergent density-polarization patterns, including swim pressure, that are\nsolely determined by bulk variables and thus constitute state functions. We\nshow that for radially symmetric activity steps these variables depend on the\ncurvature of the active-passive interface. In the specific case of total\npolarization and swim pressure, we generalize this result for arbitrary\nradially symmetric activity landscapes. Our exact as well as approximate\nanalytical results agree with exact numerical calculations. We expect\nqualitatively the same results to hold for arbitrarily curved activity\nlandscapes.\n"}
{"abstract": "  Recent work on hyperparameters optimization (HPO) has shown the possibility\nof training certain hyperparameters together with regular parameters. However,\nthese online HPO algorithms still require running evaluation on a set of\nvalidation examples at each training step, steeply increasing the training\ncost. To decide when to query the validation loss, we model online HPO as a\ntime-varying Bayesian optimization problem, on top of which we propose a novel\n\\textit{costly feedback} setting to capture the concept of the query cost.\nUnder this setting, standard algorithms are cost-inefficient as they evaluate\non the validation set at every round. In contrast, the cost-efficient GP-UCB\nalgorithm proposed in this paper queries the unknown function only when the\nmodel is less confident about current decisions. We evaluate our proposed\nalgorithm by tuning hyperparameters online for VGG and ResNet on CIFAR-10 and\nImageNet100. Our proposed online HPO algorithm reaches human expert-level\nperformance within a single run of the experiment, while incurring only modest\ncomputational overhead compared to regular training.\n"}
{"abstract": "  We present simultaneous multiwavelength observations of the 4.66 ms redback\npulsar PSR J1048+2339. We performed phase-resolved spectroscopy with the Very\nLarge Telescope (VLT) searching for signatures of a residual accretion disk or\nintra-binary shock emission, constraining the companion radial velocity\nsemi-amplitude ($K_2$), and estimating the neutron star mass ($M_{\\rm NS}$).\nUsing the FORS2-VLT intermediate-resolution spectra, we measured a companion\nvelocity of $291 < K_2 < 348$ km s$^{-1}$ and a binary mass ratio of $0.209 < q\n< 0.250$. Combining our results for $K_2$ and $q$, we constrained the mass of\nthe neutron star and the companion to $(1.0 < M_{\\rm NS} < 1.6){\\rm\nsin}^{-3}i\\,M_{\\odot}$ and $(0.24 < M_2 < 0.33){\\rm sin}^{-3}i\\,M_{\\odot}$,\nrespectively, where $i$ is the system inclination. The Doppler map of the\nH$\\alpha$ emission line exhibits a spot feature at the expected position of the\ncompanion star and an extended bright spot close to the inner Lagrangian point.\nWe interpret this extended emission as the effect of an intra-binary shock\noriginating from the interaction between the pulsar relativistic wind and the\nmatter leaving the companion star. The mass loss from the secondary star could\nbe either due to Roche-lobe overflow or to the ablation of its outer layer by\nthe energetic pulsar wind. Contrastingly, we find no evidence for an accretion\ndisk. We report on the results of the SRT and the LOFAR simultaneous radio\nobservations at three different frequencies (150 MHz, 336 MHz, and 1400 MHz).\nNo pulsed radio signal is found in our search. This is probably due to both\nscintillation and the presence of material expelled from the system which can\ncause the absorption of the radio signal at low frequencies. Finally, we report\non an attempt to search for optical pulsations using IFI+Iqueye mounted at the\n1.2 m Galileo telescope at the Asiago Observatory.\n"}
{"abstract": "  As we enter the era of large-scale imaging surveys with the up-coming\ntelescopes such as LSST and SKA, it is envisaged that the number of known\nstrong gravitational lensing systems will increase dramatically. However, these\nevents are still very rare and require the efficient processing of millions of\nimages. In order to tackle this image processing problem, we present Machine\nLearning techniques and apply them to the Gravitational Lens Finding Challenge.\nThe Convolutional Neural Networks (CNNs) presented have been re-implemented\nwithin a new modular, and extendable framework, LEXACTUM. We report an Area\nUnder the Curve (AUC) of 0.9343 and 0.9870, and an execution time of 0.0061s\nand 0.0594s per image, for the Space and Ground datasets respectively, showing\nthat the results obtained by CNNs are very competitive with conventional\nmethods (such as visual inspection and arc finders) for detecting gravitational\nlenses.\n"}
{"abstract": "  A survey is given of approaches to the problem of distributed consensus,\nfocusing particularly on methods based on cellular automata and related\nsystems. A variety of new results are given, as well as a history of the field\nand an extensive bibliography. Distributed consensus is of current relevance in\na new generation of blockchain-related systems.\n"}
{"abstract": "  How do galaxies transform from blue, star-forming spirals to red, quiescent\nearly-type galaxies? To answer this question, we analyzed a set of 26 gas-rich,\nshocked post-starburst galaxies with Hubble Space Telescope (HST) imaging in B,\nI, and H bands, and Sloan Digital Sky Survey (SDSS) i-band imaging of similar\ndepth but lower resolution. We found that post-starbursts in our sample have\nintermediate morphologies between disk- and bulge-dominated (S\\'ersic\nn$=1.7^{+0.3}_{-0.0}$) and have red bulges, likely due to dust obscuration in\nthe cores.\n  Majority of galaxies in our sample are more morphologically disturbed than\nregular galaxies (88%, corresponding to >3$\\sigma$ significance) when observed\nwith HST, with asymmetry and S\\'ersic residual flux fraction being the most\nsuccessful measures of disturbance. Most disturbances are undetected at the\nlower resolution of SDSS imaging. Although ~27% galaxies are clear merger\nremnants, we found that disturbances in another ~30% of the sample are\ninternal, caused by small-scale perturbations or dust substructures rather than\ntidal features, and require high-resolution imaging to detect. We found a\n2.8$\\sigma$ evidence that asymmetry features fade on timescales ~200 Myr, and\nmay vanish entirely after ~750 Myr, so we do not rule out a possible merger\norigin of all post-starbursts given that asymmetric features may have already\nfaded. This work highlights the importance of small-scale disturbances,\ndetected only in high-resolution imaging, in understanding structural evolution\nof transitioning galaxies.\n"}
{"abstract": "  We propose a new coloring game on a graph, called the independence coloring\ngame, which is played by two players with opposite goals. The result of the\ngame is a proper coloring of vertices of a graph $G$, and Alice's goal is that\nas few colors as possible are used during the game, while Bob wants to maximize\nthe number of colors. The game consists of rounds, and in round $i$, where\n$i=1,2,,\\ldots$, the players are taking turns in selecting a previously\nunselected vertex of $G$ and giving it color $i$ (hence, in each round the\nselected vertices form an independent set). The game ends when all vertices of\n$G$ are selected (and thus colored), and the total number of rounds during the\ngame when both players are playing optimally with respect to their goals, is\ncalled the independence game chromatic number, $\\chi_{ig}(G)$, of $G$. In fact,\nfour different versions of the independence game chromatic number are\nconsidered, which depend on who starts a game and who starts next rounds. We\nprove that the new invariants lie between the chromatic number of a graph and\nthe maximum degree plus $1$, and characterize the graphs in which each of the\nfour versions of the game invariant equals $2$. We compare the versions of the\nindependence game chromatic number among themselves and with the classical game\nchromatic number. In addition, we prove that the independence game chromatic\nnumber of a tree can be arbitrarily large.\n"}
{"abstract": "  The description of chiral quantum incompressible fluids by the W-infinity\nsymmetry can be extended from the edge, where it encompasses the conformal\nfield theory approach, to the non-conformal bulk. The two regimes are\ncharacterized by excitations with different sizes, energies and momenta within\nthe disk geometry. In particular, the bulk quantities have a finite limit for\nlarge droplets. We obtain analytic results for the radial shape of excitations,\nthe edge reconstruction phenomenon and the energy spectrum of density\nfluctuations in Laughlin states.\n"}
{"abstract": "  Deep reinforcement learning has been applied for a variety of wireless tasks,\nwhich is however known with high training and inference complexity. In this\npaper, we resort to deep deterministic policy gradient (DDPG) algorithm to\noptimize predictive power allocation among K mobile users requesting video\nstreaming, which minimizes the energy consumption of the network under the\nno-stalling constraint of each user. To reduce the sampling complexity and\nmodel size of the DDPG, we exploit a kind of symmetric prior inherent in the\nactor and critic networks: permutation invariant and equivariant properties, to\ndesign the neural networks. Our analysis shows that the free model parameters\nof the DDPG can be compressed by 2/K^2. Simulation results demonstrate that the\nepisodes required by the learning model with the symmetric prior to achieve the\nsame performance as the vanilla policy reduces by about one third when K = 10.\n"}
{"abstract": "  Previous studies on nonspherical particle-fluid interaction were mostly\nconfined to elongated fiber-like particles, which were observed to induce\nturbulence drag reduction. However, with the presence of tiny disk-like\nparticles how wall turbulence is modulated and whether drag reduction occurs\nare still unknown. Motivated by those open questions, we performed two-way\ncoupled direct numerical simulations of inertialess spheroids in turbulent\nchannel flow by an Eulerian-Lagrangian approach. The additional stress accounts\nfor the feedback from inertialess spheroids on the fluid phase. The results\ndemonstrate that both rigid elongated fibers (prolate spheroids) and thin disks\n(oblate spheroids) can lead to significant turbulence modulations and drag\nreduction. However, the disk-induced drag reduction is less pronounced than\nthat of rigid fibers with the same volume fraction. Typical features of\ndrag-reduced flows by additives are observed in both flow statistics and\nturbulence coherent structures. Moreover, in contrast to one-way simulations,\nthe two-way coupled results of spheroidal particles exhibit stronger\npreferential alignments and lower rotation rates. At the end we propose a drag\nreduction mechanism by inertialess spheroids and explain the different\nperformance for drag reduction by fibers and disks. We find that the spheroidal\nparticles weaken the quasistreamwise vortices through negative work and,\ntherefore, the Reynolds shear stress is reduced. However, the mean shear stress\ngenerated by particles, which is shape-dependent, partly compensates for the\nreduction of Reynolds shear stress and thus affects the efficiency of drag\nreduction. The present study implies that tiny disk-like particles can be an\nalternative drag reduction agent in wall turbulence.\n"}
{"abstract": "  This work, presents a study of lifetime of carriers due to intrinsic\nscattering mechanisms $viz.$ electron-electron (EEI), electron-phonon (EPI) and\nphonon-phonon interactions (PPI) in a promising half-Heusler thermoelectric\nFeVSb. Using the full-$GW$ method, the effect of EEI and temperature on the\nvalence and conduction band extrema and band gap are studied. The lifetime of\ncarriers with temperature are estimated at these band extrema. At 300 K,\nestimated value of lifetime at VBM (CBM) is $\\sim$1.91 x10$^{-14}s$ ($\\sim$2.05\nx10$^{-14}s$). The estimated ground state band gap considering EEI is $\\sim$378\nmeV. Next, the effect of EPI on the lifetime of electrons and phonons with\ntemperature are discussed. The comparison of two electron lifetimes suggests\nthat EEI should be considered in transport calculations along with EPI. The\naverage acoustic, optical and overall phonon lifetimes due to EPI are studied\nwith temperature. Further, the effect of PPI is studied by computing average\nphonon lifetime for acoustic and optical phonon branches. The lifetime of the\nacoustic phonons are higher compared to optical phonons which indicates\nacoustic phonons contribute more to lattice thermal conductivity\n($\\kappa_{ph}$). The comparison of phonon lifetime due to EPI and PPI suggests\nthat, above 500 K EPI is the dominant phonon scattering mechanism and cannot be\nignored in $\\kappa_{ph}$ calculations. Lastly, a prediction of the power factor\nand figure of merit of n-type and p-type FeVSb is made by considering the\ntemperature dependent carrier lifetime for the electronic transport terms. This\nstudy shows the importance of considering EEI in electronic transport\ncalculations and EPI in phonon transport calculations. Our study is expected to\nprovide results to further explore the thermoelectric transport in this\nmaterial.\n"}
{"abstract": "  In this paper, we estimate epidemiological data of the COVID-19 pandemic in\nHungary using only the daily number of hospitalized patients, and applying\nwell-known techniques from systems and control theory. We use a previously\npublished and validated compartmental model for the description of epidemic\nspread. Exploiting the fact that an important subsystem of the model is linear,\nfirst we compute the number of latent infected persons in time. Then an\nestimate can be given for the number of people in other compartments. From\nthese data, it is possible to track the time dependent reproduction numbers via\na recursive least squares estimate. The credibility of the obtained results is\ndiscussed using available data from the literature.\n"}
{"abstract": "  We investigate the relationship between ideal membership of an operator and\nits pieces relative to several canonical types of partitions of the entries of\nits matrix representation with respect to a given orthonormal basis. Our main\ntheorems establish that if $T$ lies in an ideal $\\mathcal{I}$, then $\\sum P_n T\nP_n$ (or more generally $\\sum Q_n T P_n$) lies in the arithmetic mean closure\nof $\\mathcal{I}$ whenever $\\{P_n\\}$ (and also $\\{Q_n\\}$) is a sequence of\nmutually orthogonal projections; and in any basis for which $T$ is a block band\nmatrix, in particular, when in Patnaik--Petrovic--Weiss universal block\ntridiagonal form, then all the sub/super/main-block diagonals of $T$ are in\n$\\mathcal{I}$. And in particular, the principal ideal generated by this $T$ is\nthe finite sum of the principal ideals generated by each sub/super/main-block\ndiagonals.\n"}
{"abstract": "  Modeling the arrival process to an Emergency Department (ED) is the first\nstep of all studies dealing with the patient flow within the ED. Many of them\nfocus on the increasing phenomenon of ED overcrowding, which is afflicting\nhospitals all over the world. Since Discrete Event Simulation models are often\nadopted with the aim to assess solutions for reducing the impact of this\nproblem, proper nonstationary processes are taken into account to reproduce\ntime-dependent arrivals. Accordingly, an accurate estimation of the unknown\narrival rate is required to guarantee the reliability of the results. In this\nwork, an integer nonlinear black-box optimization problem is solved to\ndetermine the best piecewise constant approximation of the time-varying arrival\nrate function, by finding the optimal partition of the 24 hours into a suitable\nnumber of non equally spaced intervals. The black-box constraints of the\noptimization problem make the feasible solutions satisfy proper statistical\nhypotheses; these ensure the validity of the nonhomogeneous Poisson assumption\nabout the arrival process, commonly adopted in the literature, and prevent\nmixing overdispersed data for model estimation. The cost function includes a\nfit error term for the solution accuracy and a penalty term to select an\nadequate degree of regularity of the optimal solution. To show the\neffectiveness of this methodology, real data from one of the largest Italian\nhospital EDs are used.\n"}
{"abstract": "  We initiate the study of Dyson equations of perturbative QFT in AdS and their\nconsequences for large-N CFT. We show that the dressed one-particle AdS\npropagator features wavefunction renormalization and operator mixing, giving\nrise to finite corrections to OPE data. We show how the resummation of 1/N\neffects in the CFT emerges from the dressing in AdS. When a boundary-to-bulk\npropagator is dressed by propagators whose sum of conformal dimensions is lower\nthan the main dimension, it cannot map onto a CFT source following the standard\nprescription; we relate this to an AdS/CFT version of particle instability. We\ninvestigate the dressing of the two-particle propagator and obtain a conformal\nBethe-Salpeter equation for the conformal partial wave of a \"bound state''\noperator. In case of a ladder kernel, we briefly show that a bound state with\nconformal dimension equal to the sum of its constituents plus a 1/N-suppressed\n\"binding energy'' emerges. Resummation of the Dyson equations is essential to\nderiving these results.\n"}
{"abstract": "  This research devoted to the low-resource Veps and Karelian languages.\nAlgorithms for assigning part of speech tags to words and grammatical\nproperties to words are presented in the article. These algorithms use our\nmorphological dictionaries, where the lemma, part of speech and a set of\ngrammatical features (gramset) are known for each word form. The algorithms are\nbased on the analogy hypothesis that words with the same suffixes are likely to\nhave the same inflectional models, the same part of speech and gramset. The\naccuracy of these algorithms were evaluated and compared. 313 thousand Vepsian\nand 66 thousand Karelian words were used to verify the accuracy of these\nalgorithms. The special functions were designed to assess the quality of\nresults of the developed algorithms. 92.4% of Vepsian words and 86.8% of\nKarelian words were assigned a correct part of speech by the developed\nalgorithm. 95.3% of Vepsian words and 90.7% of Karelian words were assigned a\ncorrect gramset by our algorithm. Morphological and semantic tagging of texts,\nwhich are closely related and inseparable in our corpus processes, are\ndescribed in the paper.\n"}
{"abstract": "  Magnetic resonance (MR) image acquisition is an inherently prolonged process,\nwhose acceleration has long been the subject of research. This is commonly\nachieved by obtaining multiple undersampled images, simultaneously, through\nparallel imaging. In this paper, we propose the Dual-Octave Network (DONet),\nwhich is capable of learning multi-scale spatial-frequency features from both\nthe real and imaginary components of MR data, for fast parallel MR image\nreconstruction. More specifically, our DONet consists of a series of\nDual-Octave convolutions (Dual-OctConv), which are connected in a dense manner\nfor better reuse of features. In each Dual-OctConv, the input feature maps and\nconvolutional kernels are first split into two components (ie, real and\nimaginary), and then divided into four groups according to their spatial\nfrequencies. Then, our Dual-OctConv conducts intra-group information updating\nand inter-group information exchange to aggregate the contextual information\nacross different groups. Our framework provides three appealing benefits: (i)\nIt encourages information interaction and fusion between the real and imaginary\ncomponents at various spatial frequencies to achieve richer representational\ncapacity. (ii) The dense connections between the real and imaginary groups in\neach Dual-OctConv make the propagation of features more efficient by feature\nreuse. (iii) DONet enlarges the receptive field by learning multiple\nspatial-frequency features of both the real and imaginary components. Extensive\nexperiments on two popular datasets (ie, clinical knee and fastMRI), under\ndifferent undersampling patterns and acceleration factors, demonstrate the\nsuperiority of our model in accelerated parallel MR image reconstruction.\n"}
{"abstract": "  The Airborne Infrared Spectrometer (AIR-Spec) was commissioned during the\n2017 total solar eclipse, when it observed five infrared coronal emission lines\nfrom the Gulfstream V High-performance Instrumented Airborne Platform for\nEnvironmental Research (GV HIAPER), a research jet owned by the National\nScience Foundation (NSF) and operated by the National Center for Atmospheric\nResearch (NCAR). The second AIR-Spec research flight took place during the July\n2, 2019 total solar eclipse across the south Pacific. The 2019 eclipse flight\nresulted in seven minutes of observations, during which the instrument measured\nall four of its target emission lines: S XI 1.393 $\\mu$m, Si X 1.431 $\\mu$m, S\nXI 1.921 $\\mu$m, and Fe IX 2.853 $\\mu$m. The 1.393 $\\mu$m line, half of a\ndensity-sensitive S XI line pair, was detected for the first time. The 2017\nAIR-Spec detection of Fe IX was confirmed and the first observations were made\nof the Fe IX intensity as a function of solar radius. Observations of S XI and\nSi X were used to estimate the temperature and density above the east and west\nlimbs, the subject of a future paper. Atmospheric absorption was significant in\nthe 2019 data, and atmospheric modeling was required to extract accurate line\nintensities. Telluric absorption features were used to calibrate the wavelength\nmapping, instrumental broadening, and throughput of the instrument. AIR-Spec\nunderwent significant upgrades in preparation for the 2019 eclipse flight. The\nthermal background was reduced by a factor of 30, providing a 5.5x improvement\nin signal-to-noise ratio, and the pointing stability was improved by a factor\nof five to $<$10 arcsec RMS after image co-alignment. In addition, two imaging\nartifacts were identified and resolved, making the 2019 data easier to\ninterpret and improving the spectral resolution by up to 50%.\n"}
{"abstract": "  We prove that the Newton product of efficient polynomial projectors is still\nefficient. Various polynomial approximation theorems are established involving\nNewton product projectors on spaces of holomorphic functions on a neighborhood\nof a regular compact set, on spaces of entire functions of given growth and on\nspaces of differentiable functions. Efficient explicit new projectors are\npresented.\n"}
{"abstract": "  Neutrinos drive core-collapse supernovae, launch outflows from neutron star\nmerger accretion disks, and set the ratio of protons to neutrons in ejecta from\nboth systems that generate heavy elements in the universe. Neutrinos of\ndifferent flavors interact with matter differently, and much recent work has\nsuggested that fast flavor instabilities are likely ubiquitous in both systems,\nbut the final flavor content after the instability saturates has not been well\nunderstood. In this work we present particle-in-cell calculations which follow\nthe evolution of all flavors of neutrinos and antineutrinos through saturation\nand kinematic decoherence. We conduct one-dimensional three-flavor simulations\nof neutrino quantum kinetics to demonstrate the outcome of this instability in\na few example cases. We demonstrate the growth of both axially symmetric and\nasymmetric modes whose wavelength and growth rate match predictions from linear\nstability analysis. Finally, we vary the number density, flux magnitude, and\nflux direction of the neutrinos and antineutrinos and demonstrate that these\nfactors modify both the growth rate and post-saturation neutrino flavor\nabundances. Weak electron lepton number (ELN) crossings in these simulations\nproduce both slow growth of the instability and little difference between the\nflavor abundances in the initial and final states. In all of these calculations\nthe same number of neutrinos and antineutrinos change flavor, making the least\nabundant between them the limiting factor for post-saturation flavor change.\nMany more simulations and multi-dimensional simulations are needed to fully\nprobe the parameter space of the initial conditions.\n"}
{"abstract": "  We introduce a non-reversible generalization of the Vertex-Reinforced Jump\nProcess (VRJP), which we call *-Vertex-Reinforced Jump Process (*-VRJP). It can\nbe seen as the continuous-time counterpart of the *-Edge-Reinforced Random Walk\n(*-ERRW) (see [2,4]), which is itself a non-reversible generalization of the\noriginal ERRW introduced by Coppersmith and Diaconis in 1986. In contrast to\nthe classical VRJP, the *-VRJP is not exchangeable after time-change, but we\nshow that with some appropriate randomization of the initial local time, it\nbecomes partially exchangeable after time-change. We provide a representation\nof the \"randomized\" *-VRJP as mixture of Markov jump processes with an explicit\nmixing measure, and show that the \"non-randomized\" *-VRJP can be written as a\nmixture of conditioned Markov processes. We also give a counterpart of the\n$\\beta$-random potential introduced in [25] for the VRJP, which leads to new\nidentities between integrals.\n"}
{"abstract": "  Cross-lingual entity alignment, which aims to precisely connect the same\nentities in different monolingual knowledge bases (KBs) together, often suffers\nchallenges from feature inconsistency to sequence context unawareness. This\npaper presents a dual adversarial learning framework for cross-lingual entity\nalignment, DAEA, with two original contributions. First, in order to address\nthe structural and attribute feature inconsistency between entities in two\nknowledge graphs (KGs), an adversarial kernel embedding technique is proposed\nto extract graph-invariant information in an unsupervised manner, and project\ntwo KGs into the common embedding space. Second, in order to further improve\nsuccessful rate of entity alignment, we propose to produce multiple random\nwalks through each entity to be aligned and mask these entities in random\nwalks. With the guidance of known aligned entities in the context of multiple\nrandom walks, an adversarial knowledge translation model is developed to fill\nand translate masked entities in pairwise random walks from two KGs. Extensive\nexperiments performed on real-world datasets show that DAEA can well solve the\nfeature inconsistency and sequence context unawareness issues and significantly\noutperforms thirteen state-of-the-art entity alignment methods.\n"}
{"abstract": "  Graphene-based nanofolds (GNFs) are edge-connected 2D stacked monolayers\noriginated from single-layer graphene. Graphene-based nanoscrolls (GNSs) are\nnanomaterials with geometry resembling graphene layers rolled up into a spiral\n(papyrus-like) form. Both GNSs and GNFs structures induce significant changes\nin the mechanical and optoelectronic properties of single-layer graphene,\naggregating new functionalities in carbon-based applications. Here, we carried\nour fully atomistic reactive (ReaxFF) molecular dynamics simulations to study\nthe self-folding and self-scrolling of edge-deformed graphene sheets. We\nadopted initial armchair edge-scrolled graphene (AESG($\\phi$,$\\theta$))\nstructures with similar (or different) twist angles ($\\phi$,$\\theta$) in each\nedge, mimicking the initial configuration that was experimentally developed to\nform biscrolled sheets. Results showed that AESG(0,$2\\pi$) and\nAESG(2$\\pi$,$2\\pi$) evolved to single-folded and two-folded fully stacked\nmorphologies, respectively. As a general trend, for twist angles higher than\n$2\\pi$, the self-deformation process of AESG morphologies yields GNSs. Edge\ntwist angles lower than $\\pi$ are not enough for triggering the\nself-deformation processes. In the AESG(0,3$\\pi$) and AESG(3$\\pi$,3$\\pi$)\ncases, after a relaxation period, their morphology transition towards GNS\noccurred rapidly. In the AESG(3$\\pi$,3$\\pi$) dynamics, a metastable biscroll\nwas formed by the interplay between the left- and right-sided partial scrolling\nin forming a unique GNS. At high-temperature perturbations, the edge folding\nand scrolling transitions to GNFs and GNSs occurred within the ultrafast\nperiod. Remarkably, the AESG(2$\\pi$,3$\\pi$) evolved to a dual state that\ncombines folded and scrolled structures in a temperature-independent process.\n"}
{"abstract": "  Performing reliability analysis on complex systems is often computationally\nexpensive. In particular, when dealing with systems having high input\ndimensionality, reliability estimation becomes a daunting task. A popular\napproach to overcome the problem associated with time-consuming and expensive\nevaluations is building a surrogate model. However, these computationally\nefficient models often suffer from the curse of dimensionality. Hence, training\na surrogate model for high-dimensional problems is not straightforward.\nHenceforth, this paper presents a framework for solving high-dimensional\nreliability analysis problems. The basic premise is to train the surrogate\nmodel on a low-dimensional manifold, discovered using the active subspace\nalgorithm. However, learning the low-dimensional manifold using active subspace\nis non-trivial as it requires information on the gradient of the response\nvariable. To address this issue, we propose using sparse learning algorithms in\nconjunction with the active subspace algorithm; the resulting algorithm is\nreferred to as the sparse active subspace (SAS) algorithm. We project the\nhigh-dimensional inputs onto the identified low-dimensional manifold identified\nusing SAS. A high-fidelity surrogate model is used to map the inputs on the\nlow-dimensional manifolds to the output response. We illustrate the efficacy of\nthe proposed framework by using three benchmark reliability analysis problems\nfrom the literature. The results obtained indicate the accuracy and efficiency\nof the proposed approach compared to already established reliability analysis\nmethods in the literature.\n"}
{"abstract": "  Electronic Health Record Systems (EHR-S) are commonly developed in monolithic\narchitectures. This architectural style presents greater complexity and demands\nmore effort when we think of interoperability. A solution proposal is the\ncreation of Microservices that use HL7 FHIR as an interoperability strategy. In\nthis sense, it is presented the development of a prototype, based on a\nmicroservices architecture, to act in a real scenario of Patient Navigation\n(PN). The problem was subdivided into 3 steps: definition of architecture,\ndevelopment and construction of an interface to simulate the role of the\nnavigator. The Patient and Appointment microservices are capable of synchronous\ncommunication to query and record information. In general, the implemented\narchitectural style not only isolates information domains but can receive data\nfrom multiple sources while maintaining essential functionality. This type of\napproach plays a crucial role in a hospital environment, specifically in PN,\nhighlighting the importance of the standard and expanding the possibilities for\nfurther research to be conducted.\n"}
{"abstract": "  We measure diffusion of organic molecules located a few nanometers from the\ndiamond surface. To study molecular diffusion, we perform local detection of\nnuclear magnetic resonance, based on single shallow Nitrogen-Vacancy (NV)\ncenters in diamond. Specifically, we demonstrate measurements of translational\ndiffusion coefficient of phospholipids in an artificial cell membrane by\nemploying correlation spectroscopy. An analysis of correlation decay curves\nusing different diffusion models shows, that a simple 2D diffusion model gives\nsatisfactory diffusion coefficients, although the choice of the model affects\nthe extracted numbers. We find significant differences among the diffusion\ncoefficients measured by different single NV centers, which we attribute to\nlocal heterogeneities of the lipid layers, likely caused by the supporting\ndiamond substrate.\n"}
{"abstract": "  Steel pipes are widely used in high-risk and high-pressure scenarios such as\noil, chemical, natural gas, shale gas, etc. If there is some defect in steel\npipes, it will lead to serious adverse consequences. Applying object detection\nin the field of deep learning to pipe weld defect detection and identification\ncan effectively improve inspection efficiency and promote the development of\nindustrial automation. Most predecessors used traditional computer vision\nmethods applied to detect defects of steel pipe weld seams. However,\ntraditional computer vision methods rely on prior knowledge and can only detect\ndefects with a single feature, so it is difficult to complete the task of\nmulti-defect classification, while deep learning is end-to-end. In this paper,\nthe state-of-the-art single-stage object detection algorithm YOLOv5 is proposed\nto be applied to the field of steel pipe weld defect detection, and compared\nwith the two-stage representative object detection algorithm Faster R-CNN. The\nexperimental results show that applying YOLOv5 to steel pipe weld defect\ndetection can greatly improve the accuracy, complete the multi-classification\ntask, and meet the criteria of real-time detection.\n"}
{"abstract": "  The co-maximal subgroup graph $\\Gamma(G)$ of a group $G$ is a graph whose\nvertices are non-trivial proper subgroups of $G$ and two vertices $H$ and $K$\nare adjacent if $HK=G$. In this paper, we continue the study of $\\Gamma(G)$,\nespecially when $\\Gamma(G)$ has isolated vertices. We define a new graph\n$\\Gamma^*(G)$, which is obtained by removing isolated vertices from\n$\\Gamma(G)$. We characterize when $\\Gamma^*(G)$ is connected, a complete graph,\nstar graph, has an universal vertex etc. We also find various graph parameters\nlike diameter, girth, bipartiteness etc. in terms of properties of $G$.\n"}
{"abstract": "  GPUReplay is a novel way for deploying GPU-accelerated computation on mobile\nand embedded devices. It addresses high complexity of a modern GPU stack for\ndeployment ease and security. The idea is to record GPU executions on the full\nGPU stack ahead of time and replay the executions on new input at run time. We\naddress key challenges towards making GR feasible, sound, and practical to use.\nThe resultant replayer is a drop-in replacement of the original GPU stack. It\nis tiny (50 KB of executable), robust (replaying long executions without\ndivergence), portable (running in a commodity OS, in TEE, and baremetal), and\nquick to launch (speeding up startup by up to two orders of magnitude). We show\nthat GPUReplay works with a variety of integrated GPU hardware, GPU APIs, ML\nframeworks, and 33 neural network (NN) implementations for inference or\ntraining. The code is available at https://github.com/bakhi/GPUReplay.\n"}
{"abstract": "  In this paper, a mixed-integer linear programming formulation for the problem\nof obtaining task-relevant, multi-resolution, graph abstractions for\nresource-constrained agents is presented. The formulation leverages concepts\nfrom information-theoretic signal compression, specifically the information\nbottleneck (IB) method, to pose a graph abstraction problem as an optimal\nencoder search over the space of multi-resolution trees. The abstractions\nemerge in a task-relevant manner as a function of agent information-processing\nconstraints, and are not provided to the system a priori. We detail our\nformulation and show how the problem can be realized as an integer linear\nprogram. A non-trivial numerical example is presented to demonstrate the\nutility in employing our approach to obtain hierarchical tree abstractions for\nresource-limited agents.\n"}
{"abstract": "  In software reverse engineering, decompilation is the process of recovering\nsource code from binary files. Decompilers are used when it is necessary to\nunderstand or analyze software for which the source code is not available.\nAlthough existing decompilers commonly obtain source code with the same\nbehavior as the binaries, that source code is usually hard to interpret and\ncertainly differs from the original code written by the programmer. Massive\ncodebases could be used to build supervised machine learning models aimed at\nimproving existing decompilers. In this article, we build different\nclassification models capable of inferring the high-level type returned by\nfunctions, with significantly higher accuracy than existing decompilers. We\nautomatically instrument C source code to allow the association of binary\npatterns with their corresponding high-level constructs. A dataset is created\nwith a collection of real open-source applications plus a huge number of\nsynthetic programs. Our system is able to predict function return types with a\n79.1% F1-measure, whereas the best decompiler obtains a 30% F1-measure.\nMoreover, we document the binary patterns used by our classifier to allow their\naddition in the implementation of existing decompilers.\n"}
{"abstract": "  We study mappings that satisfy the inverse Poletskii-type inequality in a\ndomain of Euclidean space. It was proved that such mappings have a continuous\nextension to the boundary under some conditions on the geometry of the given\nand the mapped domains if the majorant responsible for the distortion of the\nfamilies of paths has finite integrals over the set of spheres of positive\nlinear measure.\n"}
{"abstract": "  Multi-modal delivery systems are a promising solution to the challenges posed\nby the increasing demand of e-commerce. Due to the potential benefit drones can\nhave on logistics networks such as delivery systems, some countries have taken\nsteps towards integrating drones into their airspace. In this paper we aim to\nquantify this potential by developing a mathematical model for a multi-modal\ndelivery system composed of trucks and drones. We propose an optimization\nformulation that can be efficiently solved in order to design socially-optimal\nrouting and allocation policies. We incorporate both societal cost in terms of\nroad congestion and parcel delivery latency in our formulation. Our model is\nable to quantify the effect drones have on mitigating road congestion, and can\nsolve for the path routing needed to minimize the chosen objective. To\naccurately capture the effect of stopping trucks on road latency, we use SUMO\nsimulations and derive a mathematical latency function for roads shared by\ntrucks and cars. Based on this, we show that the proposed framework is\ncomputationally feasible to scale due to its reliance on convex quadratic\noptimization techniques.\n"}
{"abstract": "  The size of convective cores remains uncertain, despite its substantial\ninfluence on stellar evolution, and thus on stellar ages. The seismic modeling\nof young subgiants can be used to obtain indirect constraints on the core\nstructure during main sequence, thanks to the high probing potential of mixed\nmodes. We selected the young subgiant KIC10273246, observed by Kepler, based on\nits mixed-mode properties. We thoroughly modeled this star, with the aim of\nplacing constraints on the size of its main sequence convective core. We first\nextracted the parameters of the oscillation modes of the star using the full\nKepler data set. To overcome the challenges posed by the seismic modeling of\nsubgiants, we proposed a method which is specifically tailored for subgiants\nwith mixed modes and consists in a nested optimization. We then applied this\nmethod to perform a detailed seismic modeling of KIC10273246. We obtained\nmodels that show good statistical agreements with the observations, both\nseismic and non-seismic. We showed that including core overshooting in the\nmodels significantly improves the quality of the seismic fit, optimal models\nbeing found for $\\alpha_{\\mathrm{ov}} = 0.15$. Higher amounts of core\novershooting strongly worsen the agreement with the observations and are thus\nfirmly ruled out. We also found that having access to two g-dominated mixed\nmodes in young subgiants allows us to place stronger constraints on the\ngradient of molecular weight in the core and on the central density. This study\nconfirms the high potential of young subgiants with mixed modes to investigate\nthe size of main-sequence convective cores. It paves the way for a more general\nstudy including the subgiants observed with Kepler, TESS, and eventually PLATO.\n"}
{"abstract": "  In this work, we explore the connection between the critical curves\n(\"shadows\") and the quasinormal mode frequencies (in the eikonal limit) of Kerr\nblack holes. This mapping has been previously established for non-rotating\nblack holes. We show that, the shadow seen by an distant observer at a given\ninclination angle, can be mapped to a family of quasinormal modes with\n$m/(\\ell+1/2)$ bounded within certain range, where $m$ is is azimuthal node\nnumber and $\\ell$ is the angular node number. We discuss the possibility of\ntesting such relation with space-borne gravitational wave detectors and the\nnext-generation Event Horizon Telescope.\n"}
{"abstract": "  The lack of equitable access to radiotherapy linear accelerators (LINACs) is\na substantial barrier to cancer care in Low and Middle-Income Countries\n(LMICs). Aside from the issue of cost, there are also issues of robustness of\nstate-of-the-art LINACs in LMICs, which are subject to mechanical and\nelectrical breakdowns, which can result in downtimes ranging from days to\nmonths. Recent research has identified disparities in failure frequency and\ndowntimes between LMICs (Nigeria, Botswana) and a High-Income Countries (HIC,\nthe UK), and highlighted the need for additional data and study particularly\nrelating to Multi-Leaf Collimators (MLCs). This study analyses data from 14\nIndonesian Hospitals with a total of 19 LINACs and shows the pathways to\nfailure of radiotherapy LINACs and frequency of breakdowns with an additional\nfocus on the Multi-Leaf Collimator (MLC) subsystem. We found that LINACs\nthroughout Indonesia are out of operation for 7 times longer than HICs and the\nMean Time Between Failures of a LINAC in Indonesia is 341.58 Hours, or about 14\ndays. Furthermore, of the LINACs with an MLC fitted, $59.02^{+1.98}_{-1.61}$%\nof mechanical faults are due to the MLC and $57.14^{+0.78}_{-1.27}$% of cases\nrequiring a replacement component are related to the MLC. These results outline\nthe need to reassess the current generation of RT LINACs and ultimately work\ntowards guiding future designs to be robust enough for all environments.\n"}
{"abstract": "  The performance demands of future particle-physics experiments investigating\nthe high-energy frontier pose a number of new challenges, forcing us to find\nimproved solutions for the detection, identification, and measurement of\nfinal-state particles in subnuclear collisions. One such challenge is the\nprecise measurement of muon momentum at very high energy, where an estimate of\nthe curvature provided by conceivable magnetic fields in realistic detectors\nproves insufficient for achieving good momentum resolution when detecting,\ne.g., a narrow, high mass resonance decaying to a muon pair.\n  In this work we show the feasibility of an entirely new avenue for the\nmeasurement of the energy of muons based on their radiative losses in a dense,\nfinely segmented calorimeter. This is made possible by exploiting spatial\ninformation of the clusters of energy from radiated photons in a regression\ntask. The use of a task-specific deep learning architecture based on\nconvolutional layers allows us to treat the problem as one akin to image\nreconstruction, where images are constituted by the pattern of energy released\nin successive layers of the calorimeter. A measurement of muon energy with\nbetter than 20% relative resolution is shown to be achievable for ultra-TeV\nmuons.\n"}
{"abstract": "  Under the global COVID-19 crisis, developing robust diagnosis algorithm for\nCOVID-19 using CXR is hampered by the lack of the well-curated COVID-19 data\nset, although CXR data with other disease are abundant. This situation is\nsuitable for vision transformer architecture that can exploit the abundant\nunlabeled data using pre-training. However, the direct use of existing vision\ntransformer that uses the corpus generated by the ResNet is not optimal for\ncorrect feature embedding. To mitigate this problem, we propose a novel vision\nTransformer by using the low-level CXR feature corpus that are obtained to\nextract the abnormal CXR features. Specifically, the backbone network is\ntrained using large public datasets to obtain the abnormal features in routine\ndiagnosis such as consolidation, glass-grass opacity (GGO), etc. Then, the\nembedded features from the backbone network are used as corpus for vision\ntransformer training. We examine our model on various external test datasets\nacquired from totally different institutions to assess the generalization\nability. Our experiments demonstrate that our method achieved the state-of-art\nperformance and has better generalization capability, which are crucial for a\nwidespread deployment.\n"}
{"abstract": "  We consider a model of a binary mixture of two immiscible compressible\nfluids. We propose a numerical scheme and discuss its basic properties:\nStability, consistency, convergence. The convergence is established via the\nmethod of generalized weak solutions combined with the weak-strong uniqueness\nprinciple.\n"}
{"abstract": "  In an effort to explore high-throughput processing of microscopic image data,\na method based on deep convolutional neural network is proposed. The\nstate-of-the-art computer vision algorithm, Faster R-CNN, was trained for the\ndetection of iron (II) phthalocyanines on Se-terminated Au(111) platform\nresolved by scanning probe microscopy. The construction of the feature pyramid\nenables the multi-scale molecule detection in images of different scales from\n10 nm to 50 nm. After the detection, the orientation of each molecule is\nmeasured by a following program. Based on the statistical distribution of the\norientation angles, the preferred adsorption configurations of iron (II)\nphthalocyanine on the platform are revealed. This method yields high accuracy\nand recall with F1 score close to 1 after optimization of hyperparameters and\ntraining. It is expected to be a feasible solution in the scenarios where\nautonomous and high-throughput processing of microscopic image data is needed.\n"}
{"abstract": "  The conformal bridge transformation (CBT) is reviewed in the light of the\n$\\mathcal{PT}$ symmetry. Originally, the CBT was presented as a non-unitary\ntransformation (a complex canonical transformation in the classical case) that\nrelates two different forms of dynamics in the sense of Dirac. Namely, it maps\nthe asymptotically free form into the harmonically confined form of dynamics\nassociated with the $\\mathfrak{so}(2,1)\\cong \\mathfrak{sl}(2,{\\mathbb R})$\nconformal symmetry. However, as the transformation relates the non-Hermitian\noperator $i\\hat{D}$, where $\\hat{D}$ is the generator of dilations, with the\ncompact Hermitian generator $\\hat{\\mathcal{J}}_0$ of the\n$\\mathfrak{sl}(2,{\\mathbb R})$ algebra, the CBT generator can be associated\nwith a $\\mathcal{PT}$-symmetric metric. In this work we review the applications\nof this transformation for one- and two-dimensional systems, as well as for\nsystems on a cosmic string background, and for a conformally extended charged\nparticle in the field of Dirac monopole. We also compare and unify the CBT with\nthe Darboux transformation. The latter is used to construct\n$\\mathcal{PT}$-symmetric solutions of the equations of the KdV hierarchy with\nthe properties of extreme waves. As a new result, by using a modified CBT we\nrelate the one-dimensional $\\mathcal{PT}$-regularized asymptotically free\nconformal mechanics model with the $\\mathcal{PT}$-regularized version of the de\nAlfaro, Fubini and Furlan system.\n"}
{"abstract": "  This paper presents a framework for solving the pure-state preparation\nproblem using numerical optimal control. As an example, we consider the case\nwhere a number of qubits are dispersively coupled to a readout cavity. We model\nopen system quantum dynamics using the Markovian Lindblad master equation,\ndriven by external control pulses. The main result of this paper develops a\nbasis of density matrices (a parameterization) where each basis element is a\ndensity matrix itself. Utilizing a specific objective function, we show how an\nensemble of the basis elements can be used as a single initial state throughout\nthe optimization process - independent of the system dimension. We apply the\ngeneral framework to the specific application of ground-state reset of one and\ntwo qubits coupled to a readout cavity.\n"}
{"abstract": "  The $J_1$-$J_2$ quantum spin sawtooth chain is a paradigmatic one-dimensional\nfrustrated quantum spin system exhibiting unconventional ground-state and\nfinite-temperature properties. In particular, it exhibits a flat energy band of\none-magnon excitations accompanied by an enhanced magnetocaloric effect for two\nsingular ratios of the basal interactions $J_1$ and the zigzag interactions\n$J_2$. In our paper, we demonstrate that one can drive the spin system into a\nflat-band scenario by applying an appropriate electric field, thus overcoming\nthe restriction of fine-tuned exchange couplings $J_1$ and $J_2$ and allowing\none to tune more materials towards flat-band physics, that is to show a\nmacroscopic magnetization jump when crossing the magnetic saturation field, a\nresidual entropy at zero temperature as well as an enhanced magnetocaloric\neffect. While the magnetic field acts on the spin system via the ordinary\nZeeman term, the coupling of an applied electric field with the spins is given\nby the sophisticated Katsura-Nagaosa-Balatsky (KNB) mechanism, where the\nelectric field effectively acts as a Dzyaloshinskii-Moriya spin-spin\ninteraction. The resulting novel features are corresponding reciprocal effects:\nWe find a magnetization jump driven by the electric field as well as a jump of\nthe electric polarization driven by the magnetic field, i.e.\\ the system\nexhibits an extraordinarily strong magnetoelectric effect. Moreover, in analogy\nto the enhanced magnetocaloric effect the system shows an enhanced\nelectrocaloric effect.\n"}
{"abstract": "  The thermal properties are one of the key parameters to control phase purity\nand microstructure of polycrystalline materials. The melting point of the\niron-based BaFe2As2 superconductor (Ba122), which foresees high-field\napplications, remains controversial. In this work,\nthermogravimetry-differential scanning calorimetry measurements (TG-DSC) of\nundoped and Co-doped Ba122 were carried out. Mixtures of elemental metals and\npre-reacted Ba122 powders were prepared to investigate the thermal responses\nduring in situ and ex situ synthesis routes, respectively. In addition, the\nphases and microstructures of the quenched samples were evaluated to elucidate\nthe observed exothermic/endothermic peaks. Our results suggest that the melting\npoint of Ba122 is ~1300{\\deg}C.\n"}
{"abstract": "  In the field of aesthetic design, log-aesthetic curves have a significant\nrole to meet the high industrial requirements. In this paper, we propose a new\ninteractive $G^1$ Hermite interpolation method based on the algorithm of\nYoshida et al. with a minor boundary condition. In this novel approach, we\ncompute an extended log-aesthetic curve segment that may include inflection\npoint (S-shaped curve) or cusp. The curve segment is defined by its endpoints,\na tangent vector at the first point, and a tangent direction at the second\npoint. The algorithm also determines the shape parameter of the log-aesthetic\ncurve based on the length of the first tangent that provides control over the\ncurvature of the first point and makes the method capable of joining\nlog-aesthetic curve segments with $G^2$ continuity.\n"}
{"abstract": "  Transformers have attracted increasing interests in computer vision, but they\nstill fall behind state-of-the-art convolutional networks. In this work, we\nshow that while Transformers tend to have larger model capacity, their\ngeneralization can be worse than convolutional networks due to the lack of the\nright inductive bias. To effectively combine the strengths from both\narchitectures, we present CoAtNets(pronounced \"coat\" nets), a family of hybrid\nmodels built from two key insights: (1) depthwise Convolution and\nself-Attention can be naturally unified via simple relative attention; (2)\nvertically stacking convolution layers and attention layers in a principled way\nis surprisingly effective in improving generalization, capacity and efficiency.\nExperiments show that our CoAtNets achieve state-of-the-art performance under\ndifferent resource constraints across various datasets: Without extra data,\nCoAtNet achieves 86.0% ImageNet top-1 accuracy; When pre-trained with 13M\nimages from ImageNet-21K, our CoAtNet achieves 88.56% top-1 accuracy, matching\nViT-huge pre-trained with 300M images from JFT-300M while using 23x less data;\nNotably, when we further scale up CoAtNet with JFT-3B, it achieves 90.88% top-1\naccuracy on ImageNet, establishing a new state-of-the-art result.\n"}
{"abstract": "  The decay $D^{+}_{s}\\rightarrow \\pi^{+}\\pi^{+}\\pi^{-}\\eta$ is observed for\nthe first time, using $e^+e^-$ collision data corresponding to an integrated\nluminosity of 6.32 fb$^{-1}$, collected by the BESIII detector at\ncenter-of-mass energies between 4.178 and 4.226 GeV. The absolute branching\nfraction for this decay is measured to be $\\mathcal{B}(D^+_s \\to \\pi^+ \\pi^+\n\\pi^- \\eta) = (3.12\\pm0.13_{\\rm stat.}\\pm0.09_{\\rm syst.})$%. The first\namplitude analysis of this decay reveals the sub-structures in\n$D^{+}_{s}\\rightarrow \\pi^{+}\\pi^{+}\\pi^{-}\\eta$ and determines the relative\nfractions and the phases among these sub-structures. The dominant intermediate\nprocess is $D^{+}_{s}\\rightarrow a_1(1260)^+ \\eta, a_1(1260)^+ \\rightarrow\n\\rho(770)^0\\pi^+$ with a branching fraction of $(1.73 \\pm 0.14_{\\rm stat.} \\pm\n0.08_{\\rm syst.})$%. We also observe the W-annihilation process\n$D^{+}_{s}\\rightarrow a_0(980)^+\\rho(770)^0$, $a_0(980)^+ \\to \\pi^+ \\eta$ with\na branching fraction of $(0.21\\pm0.08_{\\rm stat.}\\pm0.05_{\\rm syst.})$%, which\nis larger than the branching fractions of other measured pure W-annihilation\ndecays by one order of magnitude.\n"}
{"abstract": "  Conversational search systems, such as Google Assistant and Microsoft\nCortana, enable users to interact with search systems in multiple rounds\nthrough natural language dialogues. Evaluating such systems is very challenging\ngiven that any natural language responses could be generated, and users\ncommonly interact for multiple semantically coherent rounds to accomplish a\nsearch task. Although prior studies proposed many evaluation metrics, the\nextent of how those measures effectively capture user preference remains to be\ninvestigated. In this paper, we systematically meta-evaluate a variety of\nconversational search metrics. We specifically study three perspectives on\nthose metrics: (1) reliability: the ability to detect \"actual\" performance\ndifferences as opposed to those observed by chance; (2) fidelity: the ability\nto agree with ultimate user preference; and (3) intuitiveness: the ability to\ncapture any property deemed important: adequacy, informativeness, and fluency\nin the context of conversational search. By conducting experiments on two test\ncollections, we find that the performance of different metrics varies\nsignificantly across different scenarios whereas consistent with prior studies,\nexisting metrics only achieve a weak correlation with ultimate user preference\nand satisfaction. METEOR is, comparatively speaking, the best existing\nsingle-turn metric considering all three perspectives. We also demonstrate that\nadapted session-based evaluation metrics can be used to measure multi-turn\nconversational search, achieving moderate concordance with user satisfaction.\nTo our knowledge, our work establishes the most comprehensive meta-evaluation\nfor conversational search to date.\n"}
{"abstract": "  Gas morphology and kinematics in the Milky Way contain key information for\nunderstanding the formation and evolution of our Galaxy. We present a high\nresolution hydrodynamical simulation based on a realistic barred Milky Way\npotential constrained by recent observations. Our model can reproduce most\nfeatures in the observed longitude-velocity diagram, including the Central\nMolecular Zone, the Near and Far 3-kpc arms, the Molecular Ring, and the spiral\narm tangents. It can also explain the non-circular motions of masers obtained\nby the recent BeSSeL2 survey. The central gas kinematics are consistent with a\nmass of $6.9\\times10^8\\; {\\rm M}_{\\odot}$ in the Nuclear Stellar Disk. Our\nmodel predicts the formation of an elliptical gaseous ring surrounding the bar,\nwhich is composed of the 3-kpc arms, Norma arm, and the bar-spiral interfaces.\nThis ring is similar to those \"inner\" rings in some Milky Way analogs with a\nboxy/peanut-shaped bulge. The kinematics of gas near the solar neighbourhood\nare governed by the Local arm, which is induced by the four major stellar\nspiral arms. The bar pattern speed constrained by our gas model is $37.5-40\\;\n{\\rm km}\\;{\\rm s}^{-1}\\;{\\rm kpc}^{-1}$, corresponding to a corotation radius\nof $R_{\\rm CR}=6.0-6.4\\;{\\rm kpc}$. The rotation curve of our model rises\ngently within the central $\\sim5\\;{\\rm kpc}$, which is significantly less steep\nthan those predicted by modern zoom-in cosmological simulations such as Auriga.\n"}
{"abstract": "  We study the critical level statistics at the many-body localization (MBL)\ntransition region in random spin systems. By employing the inter-sample\nrandomness as indicator, we manage to locate the MBL transition point in both\northogonal and unitary models. We further count the $n$-th order gap ratio\ndistributions at the transition region up to $n=4$, and find they fit well with\nthe short-range plasma model (SRPM) with inverse temperature $\\beta =1$ for\northogonal model and $\\beta =2$ for unitary. These critical level statistics\nare argued to be universal by comparing results from systems both with and\nwithout total $S_{z}$ conservation. We also point out that these critical\ndistributions can emerge from the spectrum of a Poisson ensemble, which\nindicates the thermal-MBL transition point is more affected by the MBL phase\nrather than thermal phase.\n"}
{"abstract": "  Near-field radio holography is a common method for measuring and aligning\nmirror surfaces for millimeter and sub-millimeter telescopes. In instruments\nwith more than a single mirror, degeneracies arise in the holography\nmeasurement, requiring multiple measurements and new fitting methods. We\npresent HoloSim-ML, a Python code for beam simulation and analysis of radio\nholography data from complex optical systems. This code uses machine learning\nto efficiently determine the position of hundreds of mirror adjusters on\nmultiple mirrors with few micron accuracy. We apply this approach to the\nexample of the Simons Observatory 6m telescope.\n"}
{"abstract": "  Among fundamental problems in the context of distributed computing by\nautonomous mobile entities, one of the most representative and well studied is\n{\\sc Point Convergence}: given an arbitrary initial configuration of identical\nentities, disposed in the Euclidean plane, move in such a way that, for all\n$\\eps>0$, a configuration in which the separation between all entities is at\nmost $\\eps$ is eventually reached and maintained.\n  The problem has been previously studied in a variety of settings, including\nfull visibility, exact measurements (like distances and angles), and\nsynchronous activation of entities. Our study concerns the minimal assumptions\nunder which entities, moving asynchronously with limited and unknown visibility\nrange and subject to limited imprecision in measurements, can be guaranteed to\nconverge in this way.\n  We present an algorithm that solves {\\sc Point Convergence}, for entities in\nthe plane, in such a setting, provided the degree of asynchrony is bounded:\nwhile any one entity is active, any other entity can be activated at most $k$\ntimes, for some arbitrarily large but fixed $k$. This provides a strong\npositive answer to a decade old open question posed by Katreniak.\n  We also prove that in a comparable setting that permits unbounded asynchrony,\n{\\sc Point Convergence} in the plane is impossible, contingent on the natural\nassumption that algorithms maintain the (visible) connectivity among entities\npresent in the initial configuration. This variant, that we call {\\sc Cohesive\nConvergence}, serves to distinguish the power of bounded and unbounded\nasynchrony in the control of autonomous mobile entities, settling at the same\ntime a long-standing question whether in the Euclidean plane synchronously\nscheduled entities are more powerful than asynchronously scheduled entities.\n"}
{"abstract": "  We present a statistical analysis of more than two thousand bipolar\nelectrostatic solitary waves (ESW) collected from ten quasi-perpendicular\nEarth's bow shock crossings by Magnetospheric Multiscale spacecraft. We\ndeveloped and implemented a correction procedure for reconstruction of actual\nelectric fields, velocities, and other properties of ESW from measurements,\nwhose spatial scales are typically comparable with or smaller than spatial\ndistance between voltage-sensitive probes. We determined the optimal ratio\nbetween frequency response factors of axial and spin plane antennas to be\naround 1.65/1.8. We found that more than 95\\% of the ESW in the Earth's bow\nshock are of negative polarity and present an in depth analysis of properties\nof these ESW. They have spatial scales of about 10--100 m that is within a\nrange of $\\lambda_{D}$ to $10\\lambda_{D}$, amplitudes typically below a few\nVolts that is below 0.1 of local electron temperature, and velocities below a\nfew hundreds km/s in spacecraft and plasma rest frames that is on the order of\nlocal ion-acoustic speed. The spatial scales of ESW are distinctly correlated\nwith local Debye length $\\lambda_{D}$. ESW with amplitudes of 5--30 V or\n0.1--0.3 Te have the occurrence rate of a few percent. The ESW have electric\nfields generally oblique to local magnetic field and propagate highly oblique\nto shock normal ${\\bf N}$; more than 80\\% of ESW propagate within 30$^{\\circ}$\nof the shock plane. In the shock plane, ESW typically propagate within a few\ntens of degrees of local magnetic field projection ${\\bf B}_{\\rm LM}$ onto the\nshock plane and preferentially opposite to ${\\bf N}\\times {\\bf B}_{\\rm LM}$. We\nargue that the ESW of negative polarity are ion phase space holes produced in a\nnonlinear stage of ion-ion ion-streaming instabilities. We estimated lifetimes\nof the ion holes to be 10--100 ms, or 1--10 km in terms of spatial distance.\n"}
{"abstract": "  Representing deep neural networks (DNNs) in low-precision is a promising\napproach to enable efficient acceleration and memory reduction. Previous\nmethods that train DNNs in low-precision typically keep a copy of weights in\nhigh-precision during the weight updates. Directly training with low-precision\nweights leads to accuracy degradation due to complex interactions between the\nlow-precision number systems and the learning algorithms. To address this\nissue, we develop a co-designed low-precision training framework, termed\nLNS-Madam, in which we jointly design a logarithmic number system (LNS) and a\nmultiplicative weight update algorithm (Madam). We prove that LNS-Madam results\nin low quantization error during weight updates, leading to a stable\nconvergence even if the precision is limited. We further propose a hardware\ndesign of LNS-Madam that resolves practical challenges in implementing an\nefficient datapath for LNS computations. Our implementation effectively reduces\nenergy overhead incurred by LNS-to-integer conversion and partial sum\naccumulation. Experimental results show that LNS-Madam achieves comparable\naccuracy to full-precision counterparts with only 8 bits on popular computer\nvision and natural language tasks. Compared to a full-precision floating-point\nimplementation, LNS-Madam reduces the energy consumption by over 90.\n"}
{"abstract": "  This paper addressed the dark matter analysis on the spiral galaxy NGC 4321\n(M100) by considering the nine different dark matter profiles so far lacking in\nthe scientific literature, i.e. Pseudoisothermal, Burkert, NFW, Moore, Einasto,\ncore-modified, DC14, coreNFW and Lucky13 profiles. In this paper, we analyzed\nthe rotation curve analysis on the galaxy NGC 4321 by using nonlinear fitting\nof star, gaseous and dark matter halo equations with selected VLA HI\nobservation data. Among the nine dark matter profiles, four dark matter\nprofiles (DC14, Lucky13, Burkert and Moore profiles) showed declining features\nand hence found to be not suitable for this galaxy. This is concluded to be\nmainly due to the characteristics of those dark matter profiles and also due to\nthe varying levels of problems within the inner regions fittings. For the\nremaining five accepted dark matter profiles, we further conducted analysis by\nusing the reduced chi-square test. Four out of the five accepted dark matter\nprofiles lie within the range of 0.40 < reduced chi-square < 1.70, except for\nthe case of the core-modified profile. In addition, Pseudoisothermal profile\nachieved the best fitting i.e. reduced chi-square nearest to 1, mainly due to\nits linearity in the inner region and flatness at large radii.\n"}
{"abstract": "  Extracting relevant information from data is crucial for all forms of\nlearning. The information bottleneck (IB) method formalizes this, offering a\nmathematically precise and conceptually appealing framework for understanding\nlearning phenomena. However the nonlinearity of the IB problem makes it\ncomputationally expensive and analytically intractable in general. Here we\nderive a perturbation theory for the IB method and report the first complete\ncharacterization of the learning onset, the limit of maximum relevant\ninformation per bit extracted from data. We test our results on synthetic\nprobability distributions, finding good agreement with the exact numerical\nsolution near the onset of learning. We explore the difference and subtleties\nin our derivation and previous attempts at deriving a perturbation theory for\nthe learning onset and attribute the discrepancy to a flawed assumption. Our\nwork also provides a fresh perspective on the intimate relationship between the\nIB method and the strong data processing inequality.\n"}
{"abstract": "  Open-domain dialog systems have a user-centric goal: to provide humans with\nan engaging conversation experience. User engagement is one of the most\nimportant metrics for evaluating open-domain dialog systems, and could also be\nused as real-time feedback to benefit dialog policy learning. Existing work on\ndetecting user disengagement typically requires hand-labeling many dialog\nsamples. We propose HERALD, an efficient annotation framework that reframes the\ntraining data annotation process as a denoising problem. Specifically, instead\nof manually labeling training samples, we first use a set of labeling\nheuristics to label training samples automatically. We then denoise the weakly\nlabeled data using the Shapley algorithm. Finally, we use the denoised data to\ntrain a user engagement detector. Our experiments show that HERALD improves\nannotation efficiency significantly and achieves 86% user disengagement\ndetection accuracy in two dialog corpora.\n"}
{"abstract": "  Definite descriptions are phrases of the form 'the $x$ such that $\\varphi$',\nused to refer to single entities in a context. They are often more meaningful\nto users than individual names alone, in particular when modelling or querying\ndata over ontologies. We investigate free description logics with both\nindividual names and definite descriptions as terms of the language, while also\naccounting for their possible lack of denotation. We focus on the extensions of\n$\\mathcal{ALC}$ and, respectively, $\\mathcal{EL}$ with nominals, the universal\nrole, and definite descriptions. We show that standard reasoning in these\nextensions is not harder than in the original languages, and we characterise\nthe expressive power of concepts relative to first-order formulas using a\nsuitable notion of bisimulation. Moreover, we lay the foundations for automated\nsupport for definite descriptions generation by studying the complexity of\ndeciding the existence of definite descriptions for an individual under an\nontology. Finally, we provide a polynomial-time reduction of reasoning in other\nfree description logic languages based on dual-domain semantics to the case of\npartial interpretations.\n"}
{"abstract": "  Long-range interactions play a central role in electron transport. At the\nsame time, they present a challenge for direct computer simulations, since\nsufficiently large portions of the bath have to be included in the computation\nto accurately compute the Coulomb potential. This article presents a\nreduced-order approach, by deriving an open quantum model for the reduced\ndensity-matrix. To treat the transient dynamics, the problem is placed in a\nreduced-order framework. The dynamics, described by the Liouville von Neumann\nequation, is projected to subspaces using a Petrov-Galerkin projection. In\norder to recover the global electron density profile as a vehicle to compute\nthe Coulomb potential, we propose a domain decomposition approach, where the\ncomputational domain also includes segments of the bath that are selected using\nlogarithmic grids. This approach leads to a multi-component self-energy that\nenters the effective Hamiltonian. We demonstrate the accuracy of the reduced\nmodel using a molecular junction built from a Lithium chains.\n"}
{"abstract": "  Motivated by a spectral analysis of the generator of completely positive\ntrace-preserving semigroup, we analyze a real functional $$ A,B \\in\nM_n(\\mathbb{C}) \\to r(A,B) = \\frac{1}{2}\\Bigl(\\langle [B,A],BA\\rangle + \\langle\n[B,A^\\ast],BA^\\ast \\rangle \\Bigr) \\in \\mathbb{R} $$ where $\\langle A,B\\rangle\n:= {\\rm tr} (A^\\ast B)$ is the Hilbert-Schmidt inner product, and $[A,B]:= AB -\nBA$ is the commutator. In particular we discuss the upper and lower bounds of\nthe form $c_- \\|A\\|^2 \\|B\\|^2 \\le r(A,B) \\le c_+ \\|A\\|^2 \\|B\\|^2$ where $\\|A\\|$\nis the Frobenius norm. We prove that the optimal upper and lower bounds are\ngiven by $c_\\pm = \\frac{1 \\pm \\sqrt{2}}{2}$. If $A$ is restricted to be\ntraceless, the bounds are further improved to be $c_\\pm = \\frac{1 \\pm\n\\sqrt{2(1-\\frac{1}{n})}}{2}$. Interestingly, these upper bounds, especially the\nlatter one, provide new constraints on relaxation rates for the quantum\ndynamical semigroup tighter than previously known constraints in the\nliterature. A relation with B\\\"{o}ttcher-Wenzel inequality is also discussed.\n"}
{"abstract": "  Understanding why deep nets can classify data in large dimensions remains a\nchallenge. It has been proposed that they do so by becoming stable to\ndiffeomorphisms, yet existing empirical measurements support that it is often\nnot the case. We revisit this question by defining a maximum-entropy\ndistribution on diffeomorphisms, that allows to study typical diffeomorphisms\nof a given norm. We confirm that stability toward diffeomorphisms does not\nstrongly correlate to performance on benchmark data sets of images. By\ncontrast, we find that the stability toward diffeomorphisms relative to that of\ngeneric transformations $R_f$ correlates remarkably with the test error\n$\\epsilon_t$. It is of order unity at initialization but decreases by several\ndecades during training for state-of-the-art architectures. For CIFAR10 and 15\nknown architectures, we find $\\epsilon_t\\approx 0.2\\sqrt{R_f}$, suggesting that\nobtaining a small $R_f$ is important to achieve good performance. We study how\n$R_f$ depends on the size of the training set and compare it to a simple model\nof invariant learning.\n"}
{"abstract": "  Consider the transmission eigenvalue problem \\[ (\\Delta+k^2\\mathbf{n}^2)\nw=0,\\ \\ (\\Delta+k^2)v=0\\ \\ \\mbox{in}\\ \\ \\Omega;\\quad w=v,\\ \\ \\partial_\\nu\nw=\\partial_\\nu v=0\\ \\ \\mbox{on} \\ \\partial\\Omega. \\] It is shown in [12] that\nthere exists a sequence of eigenfunctions $(w_m, v_m)_{m\\in\\mathbb{N}}$\nassociated with $k_m\\rightarrow \\infty$ such that either\n$\\{w_m\\}_{m\\in\\mathbb{N}}$ or $\\{v_m\\}_{m\\in\\mathbb{N}}$ are surface-localized,\ndepending on $\\mathbf{n}>1$ or $0<\\mathbf{n}<1$. In this paper, we discover a\nnew type of surface-localized transmission eigenmodes by constructing a\nsequence of transmission eigenfunctions $(w_m, v_m)_{m\\in\\mathbb{N}}$\nassociated with $k_m\\rightarrow \\infty$ such that both\n$\\{w_m\\}_{m\\in\\mathbb{N}}$ and $\\{v_m\\}_{m\\in\\mathbb{N}}$ are\nsurface-localized, no matter $\\mathbf{n}>1$ or $0<\\mathbf{n}<1$. Though our\nstudy is confined within the radial geometry, the construction is subtle and\ntechnical.\n"}
{"abstract": "  Automatically and accurately identifying user intents and filling the\nassociated slots from their spoken language are critical to the success of\ndialogue systems. Traditional methods require manually defining the\nDOMAIN-INTENT-SLOT schema and asking many domain experts to annotate the\ncorresponding utterances, upon which neural models are trained. This procedure\nbrings the challenges of information sharing hindering, out-of-schema, or data\nsparsity in open-domain dialogue systems. To tackle these challenges, we\nexplore a new task of {\\em automatic intent-slot induction} and propose a novel\ndomain-independent tool. That is, we design a coarse-to-fine three-step\nprocedure including Role-labeling, Concept-mining, And Pattern-mining (RCAP):\n(1) role-labeling: extracting keyphrases from users' utterances and classifying\nthem into a quadruple of coarsely-defined intent-roles via sequence labeling;\n(2) concept-mining: clustering the extracted intent-role mentions and naming\nthem into abstract fine-grained concepts; (3) pattern-mining: applying the\nApriori algorithm to mine intent-role patterns and automatically inferring the\nintent-slot using these coarse-grained intent-role labels and fine-grained\nconcepts. Empirical evaluations on both real-world in-domain and out-of-domain\ndatasets show that: (1) our RCAP can generate satisfactory SLU schema and\noutperforms the state-of-the-art supervised learning method; (2) our RCAP can\nbe directly applied to out-of-domain datasets and gain at least 76\\%\nimprovement of F1-score on intent detection and 41\\% improvement of F1-score on\nslot filling; (3) our RCAP exhibits its power in generic intent-slot\nextractions with less manual effort, which opens pathways for schema induction\non new domains and unseen intent-slot discovery for generalizable dialogue\nsystems.\n"}
{"abstract": "  We investigate the evolution of spin polarization, spontaneous Hall angle\n(SHA), saturation magnetization and Curie temperature of $B2$-ordered\nFe$_{60}$Al$_{40}$ thin films under varying antisite disorder, induced by\nNe$^{+}$-ion irradiation. The spin polarization increases monotonically as a\nfunction of ion fluence. A relatively high polarization of 46 % and the SHA of\n3.1 % are achieved on 40 nm thick films irradiated with 2 $\\cdot$ 10$^{16}$\nions/cm$^2$ at 30 keV. An interesting divergence in the trends of the\nmagnetization and SHA is observed for low disorder concentrations. The high\nspin polarization and its broad tunability range make ion-irradiated\nFe$_{60}$Al$_{40}$ a promising material for application in spin electronic\ndevices.\n"}
{"abstract": "  We consider an ambient backscatter communication (AmBC) system aided by an\nintelligent reflecting surface (IRS). The optimization of the IRS to assist\nAmBC is extremely difficult when there is no prior channel knowledge, for which\nno design solutions are currently available. We utilize a deep reinforcement\nlearning-based framework to jointly optimize the IRS and reader beamforming,\nwith no knowledge of the channels or ambient signal. We show that the proposed\nframework can facilitate effective AmBC communication with a detection\nperformance comparable to several benchmarks under full channel knowledge.\n"}
{"abstract": "  Targeted evaluations have found that machine translation systems often output\nincorrect gender, even when the gender is clear from context. Furthermore,\nthese incorrectly gendered translations have the potential to reflect or\namplify social biases. We propose a gender-filtered self-training technique to\nimprove gender translation accuracy on unambiguously gendered inputs. This\napproach uses a source monolingual corpus and an initial model to generate\ngender-specific pseudo-parallel corpora which are then added to the training\ndata. We filter the gender-specific corpora on the source and target sides to\nensure that sentence pairs contain and correctly translate the specified\ngender. We evaluate our approach on translation from English into five\nlanguages, finding that our models improve gender translation accuracy without\nany cost to generic translation quality. In addition, we show the viability of\nour approach on several settings, including re-training from scratch,\nfine-tuning, controlling the balance of the training data, forward translation,\nand back-translation.\n"}
{"abstract": "  We propose an end-to-end pipeline to robustly generate high-quality\nquadrilateral meshes for complex CAD models. An initial quad-dominant mesh is\ngenerated with frontal point insertion guided by a locally integrable cross\nfield and a scalar size map adapted to the small CAD features. After triangle\ncombination and midpoint-subdivision into an all-quadrilateral mesh, the\ntopology of the mesh is modified to reduce the number of irregular vertices.\nThe idea is to preserve the irregular vertices matching cross-field\nsingularities and to eliminate the others. The topological modifications are\neither local and based on disk quadrangulations, or more global with the\nremeshing of patches of quads according to predefined patterns. Validity of the\nquad mesh is guaranteed by monitoring element quality during all operations and\nreverting the changes when necessary. Advantages of our approach include\nrobustness, strict respect of the CAD features and support for user-prescribed\nsize constraints. The quad mesher, which is available in Gmsh, is validated and\nillustrated on two datasets of CAD models.\n"}
{"abstract": "  Two-dimensional (2D) multiferroics have been casted great attention owing to\ntheir promising prospects for miniaturized electronic and memory devices.Here,\nwe proposed a highly stable 2D multiferroic, VOF monolayer, which is an\nintrinsic ferromagnetic half semiconductor with large spin polarization ~2\n$\\mu_{B}/V$ atom and a significant uniaxial magnetic anisotropy along a-axis\n(410 $\\mu eV/V$ atom). Meanwhile, it shows excellent ferroelectricity with a\nlarge spontaneous polarization 32.7 $\\mu C/cm^{2}$ and a moderate energy\nbarrier (~43 meV/atom) between two ferroelectric states, which can be ascribed\nto the Jahn-Teller distortion.Moreover, VOF monolayer harbors an ultra-large\nnegative Poisson's ratio in the in-plane direction (~-0.34). The Curie\ntemperature evaluated from the Monte Carlo simulations based on the Ising model\nis about 215 K, which can be enhanced room temperature under -4% compressive\nbiaxial strain.The combination of ferromagnetism and ferroelectricity in the\nVOF monolayer could provide a promising platform for future study of\nmultiferroic effects and next-generation multifunctional nanoelectronic device\napplications.\n"}
{"abstract": "  Homography estimation is often an indispensable step in many computer vision\ntasks. The existing approaches, however, are not robust to illumination and/or\nlarger viewpoint changes. In this paper, we propose bidirectional implicit\nHomography Estimation (biHomE) loss for unsupervised homography estimation.\nbiHomE minimizes the distance in the feature space between the warped image\nfrom the source viewpoint and the corresponding image from the target\nviewpoint. Since we use a fixed pre-trained feature extractor and the only\nlearnable component of our framework is the homography network, we effectively\ndecouple the homography estimation from representation learning. We use an\nadditional photometric distortion step in the synthetic COCO dataset generation\nto better represent the illumination variation of the real-world scenarios. We\nshow that biHomE achieves state-of-the-art performance on synthetic COCO\ndataset, which is also comparable or better compared to supervised approaches.\nFurthermore, the empirical results demonstrate the robustness of our approach\nto illumination variation compared to existing methods.\n"}
{"abstract": "  A new $\\beta$-metastable Ti-alloy is designed with the aim to obtain a TWIP\nalloy but positioned at the limit between the TRIP/TWIP and the TWIP dominated\nregime. The designed alloy exhibits a large ductility combined with an elevated\nand stable work-hardening rate. Deformation occurring by formation and\nmultiplication of {332}<113> twins is evidenced and followed by in-situ\nelectron microscopy, and no primary stress induced martensite is observed.\nSince microstructural investigations of the deformation mechanisms show a\nhighly heterogeneous deformation, the reason of the large ductility is then\ninvestigated. The spatial strain distribution is characterized by micro-scale\ndigital image correlation, and the regions highly deformed are found to stand\nat the crossover between twins, or at the intersection between deformation\ntwins and grain boundaries. Detailed electron back-scattered imaging in such\nregions of interest finally allowed to evidence the formation of thin needles\nof stress induced martensite. The latter is thus interpreted as an\naccommodation mechanism, relaxing the local high strain fields, which ensures a\nlarge and stable plastic deformation of this newly designed Ti-alloy.\n"}
{"abstract": "  The Cadmium Zinc Telluride Imager (CZTI) onboard AstroSat is designed for\nhard X-ray imaging and spectroscopy in the energy range of 20 - 100 keV. The\nCZT detectors are of 5 mm thickness and hence have good efficiency for Compton\ninteractions beyond 100 keV. The polarisation analysis using CZTI relies on\nsuch Compton events and have been verified experimentally. The same Compton\nevents can also be used to extend the spectroscopy up to 380 keV. Further, it\nhas been observed that about 20% pixels of the CZTI detector plane have low\ngain, and they are excluded from the primary spectroscopy. If these pixels are\nincluded, then the spectroscopic capability of CZTI can be extended up to 500\nkeV and further up to 700 keV with a better gain calibration in the future.\nHere we explore the possibility of using the Compton events as well as the low\ngain pixels to extend the spectroscopic energy range of CZTI for ON-axis bright\nX-ray sources. We demonstrate this technique using Crab observations and\nexplore its sensitivity.\n"}
{"abstract": "  Software obfuscation is a crucial technology to protect intellectual\nproperty. Despite its importance, commercial and academic state-of-the-art\nobfuscation approaches are vulnerable to a plethora of automated deobfuscation\nattacks, such as symbolic execution, taint analysis, or program synthesis.\nWhile several enhanced techniques were proposed to thwart taint analysis or\nsymbolic execution, they either impose a prohibitive runtime overhead or can be\nremoved by compiler optimizations. In general, they suffer from focusing on a\nsingle attack vector, allowing an attacker to switch to other more effective\ntechniques, such as program synthesis. In this work, we present Loki, an\napproach for code obfuscation that is resilient against all known automated\ndeobfuscation attacks. To this end, we deploy multiple techniques, including a\ngeneric approach to synthesize formally verified expressions of arbitrary\ncomplexity. Contrary to state-of-the-art approaches that rely on a few\nhardcoded generation rules, our expressions are more diverse and harder to\npattern match against. Moreover, Loki protects against previously unaccounted\nattack vectors such as program synthesis, for which it reduces the success rate\nto merely 19%. Overall, our design incurs significantly less overhead while\nproviding a much stronger protection level.\n"}
{"abstract": "  Recent progress in high-dispersion spectroscopy has revealed the presence of\nvaporized heavy metals and ions in the atmosphere of hot Jupiters whose dayside\ntemperature is larger than 2000 K, categorized as ultra hot Jupiters (UHJs).\nUsing the archival data of high resolution transmission spectroscopy obtained\nwith the Subaru telescope, we searched for neutral metals in HD149026b, a hot\nJupiter cooler than UHJs. By removing stellar and telluric absorption and using\na cross-correlation technique, we report tentative detection of neutral\ntitanium with 4.4 sigma and a marginal signal of neutral iron with 2.8 sigma in\nthe atmosphere. This is the first detection of neutral titanium in an\nexoplanetary atmosphere. In this temperature range, titanium tends to form\ntitanium oxide (TiO). The fact that we did not detect any signal from TiO\nsuggests that the C/O ratio in the atmosphere is higher than the solar value.\nThe detection of metals in the atmosphere of hot Jupiters cooler than UHJs will\nbe useful for understanding the atmospheric structure and formation history of\nhot Jupiters.\n"}
{"abstract": "  Nanofibers are an interesting phase into which amphiphilic molecules can\nself-assemble. Described for a large number of synthetic lipids, they were\nseldom reported for natural lipids like microbial amphiphiles, known as\nbiosurfactants. In this work, we show that the palmitic acid congener of\nsophorolipids (SLC16:0), one of the most studied families of biosurfactants,\nspontaneously forms a self-assembled fiber network (SAFiN) at pH below 6\nthrough a pH jump process. pH-resolved in-situ Small Angle X-ray Scattering\n(SAXS) shows a continuous micelle-to-fiber transition, characterized by an\nenhanced core-shell contrast between pH 9 and pH 7 and micellar fusion into\nflat membrane between pH 7 and pH 6, approximately. Below pH 6, homogeneous,\ninfinitely long nanofibers form by peeling off the membranes. Eventually, the\nnanofiber network spontaneously forms a thixotropic hydrogel with fast recovery\nrates after applying an oscillatory strain amplitude out of the linear\nviscoelastic regime (LVER): after being submitted to strain amplitudes during 5\nmin, the hydrogel recovers about 80% and 100% of its initial elastic modulus\nafter, respectively, 20 s and 10 min. Finally, the strength of the hydrogel\ndepends on the medium's final pH, with an elastic modulus fivefold higher at pH\n3 than at pH 6.\n"}
{"abstract": "  The field of 2D materials has gained a lot of attention for vast range of\napplications. Amongst others, the sensing ability towards harmful gases is the\napplication, which we explored in the present work using quantum-mechanical\nsimulations for the SnP3 material. Its electronic properties, namely 1 and 2\nlayers being semiconducting, while multilayers being metallic, offer a\npossibility to build a single-material logical junction. In addition, the\nharmful gases studied here show physical adsorption with charge transfer from\nthe substrate to the gas molecules. Calculated recovery times show promise of a\ngood sensing material. The I-V characteristics calculated for all cases\nindicates that SnP3 could be a viable sensing material towards NO gas via\nnegative differential resistance.\n"}
{"abstract": "  The Rouse model with internal friction (RIF), a widely used theoretical\nframework to interpret the effects of internal friction on conformational\ntransitions in biomolecules, is shown to be an approximate treatment that is\nbased on preaveraging internal friction. By comparison with Brownian dynamics\nsimulations of an exact coarse-grained model that incorporates fluctuations in\ninternal friction, the accuracy of the preaveraged model predictions is\nexamined both at and away from equilibrium. While the two models predict\nintrachain autocorrelations that approach each other for long enough chain\nsegments, they differ in their predictions for shorter segments. Furthermore,\nthe two models differ qualitatively in their predictions for the chain\nextension and viscosity in shear flow, which is taken to represent a\nprototypical out-of-equilibrium condition.\n"}
{"abstract": "  In order to develop reliable services using machine learning, it is important\nto understand the uncertainty of the model outputs. Often the probability\ndistribution that the prediction target follows has a complex shape, and a\nmixture distribution is assumed as a distribution that uncertainty follows.\nSince the output of mixture density estimation is complicated, its\ninterpretability becomes important when considering its use in real services.\nIn this paper, we propose a method for mixture density estimation that utilizes\nan interpretable tree structure. Further, a fast inference procedure based on\ntime-invariant information cache achieves both high speed and interpretability.\n"}
{"abstract": "  This is a perspective paper inspired from the study of Turing Test proposed\nby A.M. Turing (23 June 1912 - 7 June 1954) in 1950. Following one important\nimplication of Turing Test for enabling a machine with a human-like behavior or\nperformance, we define human-like robustness (HLR) for AI machines. The\nobjective of the new definition aims to enforce AI machines with HLR, including\nto evaluate them in terms of HLR. A specific task is discussed only on object\nidentification, because it is the most common task for every person in daily\nlife. Similar to the perspective, or design, position by Turing, we provide a\nsolution of how to achieve HLR AI machines without constructing them and\nconducting real experiments. The solution should consists of three important\nfeatures in the machines. The first feature of HLR machines is to utilize\ncommon sense from humans for realizing a causal inference. The second feature\nis to make a decision from a semantic space for having interpretations to the\ndecision. The third feature is to include a \"human-in-the-loop\" setting for\nadvancing HLR machines. We show an \"identification game\" using proposed design\nof HLR machines. The present paper shows an attempt to learn and explore\nfurther from Turing Test towards the design of human-like AI machines.\n"}
{"abstract": "  Let $M$ denote a finitely generated module over a Noetherian ring $R$. For an\nideal $I \\subset R$ there is a study of the endomorphisms of the local\ncohomology module $H^g_I(M), g = \\operatorname{grade} (I,M),$ and related\nresults. Another subject is the study of left derived functors of the $I$-adic\ncompletion $\\Lambda^I_i(H^g_I(M))$, motivated by a characterization of\nGorenstein rings given in the book by Simon and the author. This provides\nanother Cohen-Macaulay criterion. The results are illustrated by several\nexamples. There is also an extension to the case of homomorphisms of two\ndifferent local cohomology modules.\n"}
{"abstract": "  Alzheimer's disease (AD) is one of the most common neurodegenerative\ndiseases, with around 50 million patients worldwide. Accessible and\nnon-invasive methods of diagnosing and characterising AD are therefore urgently\nrequired. Electroencephalography (EEG) fulfils these criteria and is often used\nwhen studying AD. Several features derived from EEG were shown to predict AD\nwith high accuracy, e.g. signal complexity and synchronisation. However, the\ndynamics of how the brain transitions between stable states have not been\nproperly studied in the case of AD and EEG data. Energy landscape analysis is a\nmethod that can be used to quantify these dynamics. This work presents the\nfirst application of this method to both AD and EEG. Energy landscape assigns\nenergy value to each possible state, i.e. pattern of activations across brain\nregions. The energy is inversely proportional to the probability of occurrence.\nBy studying the features of energy landscapes of 20 AD patients and 20 healthy\nage-matched counterparts, significant differences were found. The dynamics of\nAD patients' brain networks were shown to be more constrained - with more local\nminima, less variation in basin size, and smaller basins. We show that energy\nlandscapes can predict AD with high accuracy, performing significantly better\nthan baseline models.\n"}
{"abstract": "  We consider here a family of singular Laplace-Beltrami operators, focussing\nour attention on the problem of so-called quantum confinement on the half-plane\nequipped with Riemannian metrics of Grushin type degenerate at the boundary. By\nintroducing a costant-fiber direct integral scheme we are able to rigorously\ncharacterize the presence or absence of self-adjointness of these operators. We\nalso compare our technique and results with the already studied problem of\nquantum confinement on the half-cylinder.\n"}
{"abstract": "  While the body of research focusing on Intelligent Environments (IEs)\nprogramming by adults is steadily growing, informed insights about children as\nprogrammers of such environments are limited. Previous work already established\nthat young children can learn programming basics. Yet, there is still a need to\ninvestigate whether this capability can be transferred in the context of IEs,\nsince encouraging children to participate in the management of their\nintelligent surroundings can enhance responsibility, independence, and the\nspirit of cooperation. We performed a user study (N=15) with children aged\n7-12, using a block-based, gamified AR spatial coding prototype allowing to\nmanipulate smart artifacts in an Intelligent Living room. Our results validated\nthat children understand and can indeed program IEs. Based on our findings, we\ncontribute preliminary implications regarding the use of specific technologies\nand paradigms (e.g. AR, trigger-action programming) to inspire future systems\nthat enable children to create enriching experiences in IEs.\n"}
{"abstract": "  New autonomous driving technologies are emerging every day and some of them\nhave been commercially applied in the real world. While benefiting from these\ntechnologies, autonomous trucks are facing new challenges in short-term\nmaintenance planning, which directly influences the truck operator's profit. In\nthis paper, we implement a vehicle health management system by addressing the\nmaintenance planning issues of autonomous trucks on a transport mission. We\nalso present a maintenance planning model using a risk-based decision-making\nmethod, which identifies the maintenance decision with minimal economic risk of\nthe truck company. Both availability losses and maintenance costs are\nconsidered when evaluating the economic risk. We demonstrate the proposed model\nby numerical experiments illustrating real-world scenarios. In the experiments,\ncompared to three baseline methods, the expected economic risk of the proposed\nmethod is reduced by up to $47\\%$. We also conduct sensitivity analyses of\ndifferent model parameters. The analyses show that the economic risk\nsignificantly decreases when the estimation accuracy of remaining useful life,\nthe maximal allowed time of delivery delay before order cancellation, or the\nnumber of workshops increases. The experiment results contribute to identifying\nfuture research and development attentions of autonomous trucks from an\neconomic perspective.\n"}
{"abstract": "  The paper reports on a theoretical description of work function of TiN, which\nis one of the most used materials for the realization of electrodes and gates\nin CMOS devices. Indeed, although the work function is a fundamental quantity\nin quantum mechanics and also in device physics, as it allows the understanding\nof band alignment at heterostructures and gap states formation at the\nmetal/semiconductor interface, the role of defects and contaminants is rarely\ntaken into account. Here, by using first principles simulations, we present an\nextensive study of the work function dependence on nitrogen vacancies and\nsurface oxidation for different TiN surface orientations. The results\ncomplement and explain a number of existent experimental data, and provide a\nuseful tool to tailoring transport properties of TiN electrodes in device\nsimulations.\n"}
{"abstract": "  Uncovering anomalies in attributed networks has recently gained popularity\ndue to its importance in unveiling outliers and flagging adversarial behavior\nin a gamut of data and network science applications including {the Internet of\nThings (IoT)}, finance, security, to list a few. The present work deals with\nuncovering anomalous edges in attributed graphs using two distinct formulations\nwith complementary strengths, which can be easily distributed, and hence\nefficient. The first relies on decomposing the graph data matrix into low rank\nplus sparse components to markedly improve performance. The second broadens the\nscope of the first by performing robust recovery of the unperturbed graph,\nwhich enhances the anomaly identification performance. The novel methods not\nonly capture anomalous edges linking nodes of different communities, but also\nspurious connections between any two nodes with different features. Experiments\nconducted on real and synthetic data corroborate the effectiveness of both\nmethods in the anomaly identification task.\n"}
{"abstract": "  The rapid development of social media changes the lifestyle of people and\nsimultaneously provides an ideal place for publishing and disseminating rumors,\nwhich severely exacerbates social panic and triggers a crisis of social trust.\nEarly content-based methods focused on finding clues from the text and user\nprofiles for rumor detection. Recent studies combine the stances of users'\ncomments with news content to capture the difference between true and false\nrumors. Although the user's stance is effective for rumor detection, the manual\nlabeling process is time-consuming and labor-intensive, which limits the\napplication of utilizing it to facilitate rumor detection.\n  In this paper, we first finetune a pre-trained BERT model on a small labeled\ndataset and leverage this model to annotate weak stance labels for users'\ncomment data to overcome the problem mentioned above. Then, we propose a novel\nStance-aware Reinforcement Learning Framework (SRLF) to select high-quality\nlabeled stance data for model training and rumor detection. Both the stance\nselection and rumor detection tasks are optimized simultaneously to promote\nboth tasks mutually. We conduct experiments on two commonly used real-world\ndatasets. The experimental results demonstrate that our framework outperforms\nthe state-of-the-art models significantly, which confirms the effectiveness of\nthe proposed framework.\n"}
{"abstract": "  We characterize the symbols $\\Phi$ for which there exists a weight w such\nthat the weighted composition operator M w C $\\Phi$ is compact on the weighted\nBergman space B 2 $\\alpha$. We also characterize the symbols for which there\nexists a weight w such that M w C $\\Phi$ is bounded but not compact. We also\ninvestigate when there exists w such that M w C $\\Phi$ is Hilbert-Schmidt on B\n2 $\\alpha$.\n"}
{"abstract": "  In this paper we prove that among pairs $K,\\,K' \\subset \\mathbb{C}$ of\nconformal dynamically defined Cantor sets with sum of Hausdorff dimensions\n$HD(K)+HD(K')>2$, there is an open and dense subset of such pairs verifying\n$\\text{int}(K-K')\\neq \\emptyset$. This is motivated by the work \\cite{MY},\nwhere Moreira and Yoccoz proved a similar statement for dynamically defined\nCantor sets in the real line. Here we adapt their argument to the context of\nconformal Cantor sets in the complex plane, this requires the introduction of\nseveral new concepts and a more detailed analysis in some parts of the\nargument.\n"}
{"abstract": "  This paper focuses on distributed learning-based control of decentralized\nmulti-agent systems where the agents' dynamics are modeled by Gaussian\nProcesses (GPs). Two fundamental problems are considered: the optimal design of\nexperiment for concurrent learning of the agents' GP models, and the\ndistributed coordination given the learned models. Using a Distributed Model\nPredictive Control (DMPC) approach, the two problems are formulated as\ndistributed optimization problems, where each agent's sub-problem includes both\nlocal and shared objectives and constraints. To solve the resulting complex and\nnon-convex DMPC problems efficiently, we develop an algorithm called\nAlternating Direction Method of Multipliers with Convexification (ADMM-C) that\ncombines a distributed ADMM algorithm and a Sequential Convexification method.\nThe computational efficiency of our proposed method comes from the facts that\nthe computation for solving the DMPC problem is distributed to all agents and\nthat efficient convex optimization solvers are used at the agents for solving\nthe convexified sub-problems. We also prove that, under some technical\nassumptions, the ADMM-C algorithm converges to a stationary point of the\npenalized optimization problem. The effectiveness of our approach is\ndemonstrated in numerical simulations of a multi-vehicle formation control\nexample.\n"}
{"abstract": "  In this work, we have applied the integrals of motion method in a nonunitary\napproach and so obtained the time-dependent displacement and squeezed\nparameters of the coherent squeezed states (CSS). On its turn, CSS for\none-dimensional systems with general time-dependent quadratic Hamiltonian are\nconstructed. We discuss the properties of these states, in particular,\nminimization of uncertainty relation and transition probabilities. As an\napplication, we calculate the CSS of an oscillator with a time-dependent\nfrequency and shown that the solution can be obtained from these well-known\nMathieu's equation.\n"}
{"abstract": "  This paper tackles the multi-objective optimization of the cost functional of\na path-following model predictive control for vehicle longitudinal and lateral\ncontrol. While the inherent optimal character of the model predictive control\nand the direct consideration of constraints gives a very powerful tool for many\napplications, is the determination of an appropriate cost functional a\nnon-trivial task. This results on the one hand from the number of degrees of\nfreedom or the multitude of adjustable parameters and on the other hand from\nthe coupling of these. To overcome this situation a Bayesian optimization\nprocedure is present, which gives the possibility to determine optimal cost\nfunctional parameters for a given desire. Moreover, a Pareto-front for a whole\nset of possible configurations can be computed.\n"}
{"abstract": "  Attosecond science promises to reveal the most fundamental electronic\ndynamics occurring in matter and it can develop further by meeting two linked\ntechnological goals related to high-order harmonic sources: higher photon flux\n(permitting to measure low cross-section processes) and improved spectral\ntunability (allowing selectivity in addressing specific electronic\ntransitions). New developments come through parametric waveform synthesis,\nwhich provides control over the shape of high-energy electric field transients,\nenabling the creation of highly-tunable isolated attosecond pulses via\nhigh-harmonic generation. Here we show that central energy, spectral\nbandwidth/shape and temporal duration of the attosecond pulses can be\ncontrolled by shaping the laser pulse waveform via two key parameters: the\nrelative-phase between two halves of the multi-octave spanning optical\nspectrum, and the overall carrier-envelope phase. These results not only\npromise to expand the experimental possibilities in attosecond science, but\nalso demonstrate coherent strong-field control of free-electron trajectories\nusing tailored optical waveforms.\n"}
{"abstract": "  Frequency modulation coherent anti-Stokes Raman scattering (FM CARS) is\npresented, using a compact as well as fast and widely tunable fiber-based light\nsource. With this light source Raman resonances between 700 cm$^{-1}$ and 3200\ncm$^{-1}$ can be addressed via wavelength tuning within only 5 ms, which allows\nfor FM CARS measurements with frame-to-frame wavelength switching. Moreover,\nthe functionality for high-sensitivity FM CARS measurements was integrated by\nmeans of fiber optics to keep a stable and reliable operation. The light source\naccomplished FM CARS measurements with a 40 times enhanced sensitivity at a\nlock-in amplifier (LIA) bandwidth of 1 Hz. For fast imaging with frame-to-frame\nwavelength switching at a LIA bandwidth of 1 MHz a 18-fold contrast enhancement\ncould be verified, making this light source ideal for routine and out-of-lab FM\nCARS measurements for medical diagnostics or environmental sensing.\n"}
{"abstract": "  In this study, we investigate the spatially nonuniform-temperature effects on\nthe QCD chiral phase transition in the heavy-ion collisions. Since the QCD\neffective theory and the Ising model belong to the same universality class, we\nstart our discussion by mimicking the QCD effective potential with an\nIsing-like effective potential. In contrast to the dynamical slowing down\neffects which delays the phase transition from quark-gluon-plasma to hadron\ngas, the spatially nonuniform-temperature effects show a possibility to lift\nthe phase transition temperature. Besides, both the fluctuations and the\ncorrelation length are enhanced in the phase transition region. Furthermore,\nthe critical phenomena is strongly suppressed like as the critical slowing down\neffects. The underlying mechanism is the nonzero-momentum mode fluctuations of\nthe order parameter induced by the nonuniform temperature. Our study provides a\nmethod to evaluate the nonuniform-temperature effects, and illustrate its\npotential influence on analyzing the QCD phase transition signals at RHIC.\n"}
{"abstract": "  Earth observation offers new insight into anthropogenic changes to nature,\nand how these changes are effecting (and are effected by) the built environment\nand the real economy. With the global availability of medium-resolution\n(10-30m) synthetic aperture radar (SAR) Sentinel-1 and multispectral Sentinel-2\nimagery, machine learning can be employed to offer these insights at scale,\nunbiased to the reporting of companies and countries. In this paper, I\nintroduce DeepSentinel, a data pipeline and experimentation framework for\nproducing general-purpose semantic embeddings of paired Sentinel-1 and\nSentinel-2 imagery. I document the development of an extensible corpus of\nlabelled and unlabelled imagery for the purposes of sensor fusion research.\nWith this new dataset I develop a set of experiments applying popular\nself-supervision methods and encoder architectures to a land cover\nclassification problem. Tile2vec spatial encoding with a self-attention enabled\nResNet model outperforms deeper ResNet variants as well as pretraining with\nvariational autoencoding and contrastive loss. All supporting and derived data\nand code are made publicly available.\n"}
{"abstract": "  Low gain avalanche detectors can measure charged particle fluences with high\nspeed and spatial precision, and are a promising technology for radiation\nmonitoring and dosimetry. A detector has been tested in a medical linac where\nsingle particles were observed with a time resolution of 50\\,ps. The integrated\nresponse is similar to a standard ionising chamber but with a spatial precision\ntwenty times finer, and a temporal precision over 100 million times better,\nwith the capability to measure the charge deposited by a single linac pulse.\nThe unprecedented resolving power allows the structure of the $\\sim 3\\,\\mu$s\nlinac pulses to be viewed and the 350\\,ps sub-pulses in the train to be\nobserved.\n"}
{"abstract": "  For the first time, simultaneous broadband spectral and timing study of the\natoll source 4U 1705-44 was performed using AstroSat Soft X-ray Telescope (SXT)\nand Large Area X-ray Proportional Counter (LAXPC) data (0.8-70 keV). Based on\nthe HID, the source was in the soft banana state during these observations.\nSpectral modeling was performed using the full reflection framework and an\ninner disk radii of 14 Rg was obtained. A hard powerlaw tail was noticed in the\nsoft state and hot component fluxes and varying powerlaw indices point towards\na varying corona/sub-keplerian flow. Based on the spectral fits the boundary\nlayer radius and magnetospheric radius were constrained to be $\\sim$ 14-18 km\nand $\\sim$ 9-19 km respectively. Cross- Correlation Function studies were\nperformed between the 0.8-3 keV soft SXT lightcurve and 10-20 keV hard LAXPC\nlightcurve and correlated and anticorrelated lags were found, which was used to\nconstrain the coronal height to 0.6-20 km (\\b{eta}=0.1). Since the inner disk\nradius is not varying during the observations, we conclude that the detected\nlags are possibly caused by a varying structure of corona/boundary layer in the\ninner region of the accretion disk. Based on the observations, a geometrical\nmodel is proposed for explaining the detected lags in the atoll source 4U\n1705-44.\n"}
{"abstract": "  UI design is an integral part of software development. For many developers\nwho do not have much UI design experience, exposing them to a large database of\nreal-application UI designs can help them quickly build up a realistic\nunderstanding of the design space for a software feature and get design\ninspirations from existing applications. However, existing keyword-based,\nimage-similarity-based, and component-matching-based methods cannot reliably\nfind relevant high-fidelity UI designs in a large database alike to the UI\nwireframe that the developers sketch, in face of the great variations in UI\ndesigns. In this article, we propose a deep-learning-based UI design search\nengine to fill in the gap. The key innovation of our search engine is to train\na wireframe image autoencoder using a large database of real-application UI\ndesigns, without the need for labeling relevant UI designs. We implement our\napproach for Android UI design search, and conduct extensive experiments with\nartificially created relevant UI designs and human evaluation of UI design\nsearch results. Our experiments confirm the superior performance of our search\nengine over existing image-similarity or component-matching-based methods and\ndemonstrate the usefulness of our search engine in real-world UI design tasks.\n"}
{"abstract": "  Medical diagnostic robot systems have been paid more and more attention due\nto its objectivity and accuracy. The diagnosis of mild cognitive impairment\n(MCI) is considered an effective means to prevent Alzheimer's disease (AD).\nDoctors diagnose MCI based on various clinical examinations, which are\nexpensive and the diagnosis results rely on the knowledge of doctors.\nTherefore, it is necessary to develop a robot diagnostic system to eliminate\nthe influence of human factors and obtain a higher accuracy rate. In this\npaper, we propose a novel Group Feature Domain Adversarial Neural Network\n(GF-DANN) for amnestic MCI (aMCI) diagnosis, which involves two important\nmodules. A Group Feature Extraction (GFE) module is proposed to reduce\nindividual differences by learning group-level features through adversarial\nlearning. A Dual Branch Domain Adaptation (DBDA) module is carefully designed\nto reduce the distribution difference between the source and target domain in a\ndomain adaption way. On three types of data set, GF-DANN achieves the best\naccuracy compared with classic machine learning and deep learning methods. On\nthe DMS data set, GF-DANN has obtained an accuracy rate of 89.47%, and the\nsensitivity and specificity are 90% and 89%. In addition, by comparing three\nEEG data collection paradigms, our results demonstrate that the DMS paradigm\nhas the potential to build an aMCI diagnose robot system.\n"}
{"abstract": "  Automatic differentiation (AD) is an important family of algorithms which\nenables derivative based optimization. We show that AD can be simply\nimplemented with effects and handlers by doing so in the Frank language. By\nconsidering how our implementation behaves in Frank's operational semantics, we\nshow how our code performs the dynamic creation of programs during evaluation.\n"}
{"abstract": "  The wire stretching measurement was completed on the prototype Double Quarter\nWave (DQW) crab cavity for operation practice and calibration of the\nmeasurement system. Four locations were defined to be on the electrical center\nplane of the crab cavity, and survey of the wire indicated all are on the same\nplane. The successful measurement validated the wire stretching system built at\nBrookhaven National Lab. The offset of the four wire locations to the fitted\nplane provided the error of the measurement.\n"}
{"abstract": "  Topological order in solid state systems is often calculated from the\nintegration of an appropriate curvature function over the entire Brillouin\nzone. At topological phase transitions where the single particle spectral gap\ncloses, the curvature function diverges and changes sign at certain high\nsymmetry points in the Brillouin zone. These generic properties suggest the\nintroduction of a supervised machine learning scheme that uses only the\ncurvature function at the high symmetry points as input data. We apply this\nscheme to a variety of interacting topological insulators in different\ndimensions and symmetry classes, and demonstrate that an artificial neural\nnetwork trained with the noninteracting data can accurately predict all\ntopological phases in the interacting cases with very little numerical effort.\nIntriguingly, the method uncovers a ubiquitous interaction-induced topological\nquantum multicriticality in the examples studied.\n"}
{"abstract": "  This paper aims at reviewing and analysing the method of reflections. The\nlatter is an iterative procedure designed to linear boundary value problems set\nin multiply connected domains. Being based on a decomposition of the domain\nboundary, this method is particularly well-suited to numerical solvers relying\non integral representation formulas. For the parallel and sequential forms of\nthe method appearing in the literature, we propose a general abstract\nformulation in a given Hilbert setting and interpret the procedure in terms of\nsubspace corrections. We then prove the unconditional convergence of the\nsequential form and propose a modification of the parallel one that makes it\nunconditionally converging. An alternative proof of convergence is provided in\na case which does not fit into the previous framework. We finally present some\nnumerical tests.\n"}
{"abstract": "  All the elements heavier than Fe are produced either by slow (-s) or rapid\n(-r) neutron-capture process. Neutron density prevailing in the stellar sites\nis one of the major factors that determines the type of neutron-capture\nprocesses. We present the results based on the estimates of corrected value of\nabsolute carbon abundance, [C/N] ratio, carbon isotopic ratio and [hs/ls] ratio\nobtained from the high resolution spectral analysis of six stars that include\nboth CH stars and CEMP stars. All the stars show enhancement of neutron-capture\nelements. Location of these objects in the A(C) vs. [Fe/H] diagram shows that\nthey are Group I objects, with external origin of carbon and neutron-capture\nelements. Low values of carbon isotopic ratios estimated for these objects may\nalso be attributed to some external sources. As the carbon isotopic ratio is a\ngood indicator of mixing, we have used the estimates of 12C/13C ratios to\nexamine the occurance of mixing in the stars. While the object HD 30443 might\nhave experienced an extra mixing process that usually occurs after red giant\nbranch (RGB) bump for stars with log(L/L0) > 2.0, the remaining objects do not\nshow any evidence of having undergone any such mixing process. The higher\nvalues of [C/N] ratios obtained for these objects also indicate that none of\nthese objects have experienced any strong internal mixing processes. Based on\nthe estimated abundances of carbon and the neutron-capture elements, and the\nabundance ratios, we have classified the objects into different groups. While\nthe objects HE 0110-0406, HD 30443 and CD-38 2151 are found to be CEMP-s stars,\nHE 0308-1612 and HD 176021 show characteristic properties of CH stars with\nmoderate enhancement of carbon. The object CD-28 1082 with enhancement of both\nr- and s-process elements is found to belong to the CEMP-r/s group.\n"}
{"abstract": "  This paper presents a brief overview of how astronomy can help society in\nCOVID-19 times, and the lessons that come from studying our place in the\nUniverse and the global coordination of scientific and outreach activities.\nSeveral examples coordinated by the International Astronomical Union are\npresented.\n"}
{"abstract": "  Point-spread function (PSF) estimation in spatially undersampled images is\nchallenging because large pixels average fine-scale spatial information. This\nis problematic when fine-resolution details are necessary, as in optimal\nphotometry where knowledge of the illumination pattern beyond the native\nspatial resolution of the image may be required. Here, we introduce a method of\nPSF reconstruction where point sources are artificially sampled beyond the\nnative resolution of an image and combined together via stacking to return a\nfinely sampled estimate of the PSF. This estimate is then deconvolved from the\npixel-gridding function to return a superresolution kernel that can be used for\noptimally weighted photometry. We benchmark against the < 1% photometric error\nrequirement of the upcoming SPHEREx mission to assess performance in a concrete\nexample. We find that standard methods like Richardson--Lucy deconvolution are\nnot sufficient to achieve this stringent requirement. We investigate a more\nadvanced method with significant heritage in image analysis called iterative\nback-projection (IBP) and demonstrate it using idealized Gaussian cases and\nsimulated SPHEREx images. In testing this method on real images recorded by the\nLORRI instrument on New Horizons, we are able to identify systematic pointing\ndrift. Our IBP-derived PSF kernels allow photometric accuracy significantly\nbetter than the requirement in individual SPHEREx exposures. This PSF\nreconstruction method is broadly applicable to a variety of problems and\ncombines computationally simple techniques in a way that is robust to\ncomplicating factors such as severe undersampling, spatially complex PSFs,\nnoise, crowded fields, or limited source numbers.\n"}
{"abstract": "  We present an accurate user-encounter trace generator based on analytical\nmodels. Our method generates traces of intercontact times faster than models\nthat explicitly generate mobility traces. We use this trace generator to study\nthe characteristics of pair-wise intercontact-time distributions and visualize,\nusing simulations, how they combine to form the aggregate intercontact-time\ndistribution. Finally, we apply our trace-generation model to the epidemic\nrouting protocol.\n"}
{"abstract": "  The need to accurately estimate the speed of road vehicles is becoming\nincreasingly important for at least two main reasons. First, the number of\nspeed cameras installed worldwide has been growing in recent years, as the\nintroduction and enforcement of appropriate speed limits is considered one of\nthe most effective means to increase the road safety. Second, traffic\nmonitoring and forecasting in road networks plays a fundamental role to enhance\ntraffic, emissions and energy consumption in smart cities, being the speed of\nthe vehicles one of the most relevant parameters of the traffic state. Among\nthe technologies available for the accurate detection of vehicle speed, the use\nof vision-based systems brings great challenges to be solved, but also great\npotential advantages, such as the drastic reduction of costs due to the absence\nof expensive range sensors, and the possibility of identifying vehicles\naccurately. This paper provides a review of vision-based vehicle speed\nestimation. We describe the terminology, the application domains, and propose a\ncomplete taxonomy of a large selection of works that categorizes all stages\ninvolved. An overview of performance evaluation metrics and available datasets\nis provided. Finally, we discuss current limitations and future directions.\n"}
{"abstract": "  We present a new compositional approach to compositional game theory (CGT)\nbased upon Arrows, a concept originally from functional programming, closely\nrelated to Tambara modules, and operators to build new Arrows from old. We\nmodel equilibria as a bimodule over an Arrow and define an operator to build a\nnew Arrow from such a bimodule over an existing Arrow. We also model strategies\nas graded Arrows and define an operator which builds a new Arrow by taking the\ncolimit of a graded Arrow. A final operator builds a graded Arrow from a graded\nbimodule. We use this compositional approach to CGT to show how known and\npreviously unknown variants of open games can be proven to form symmetric\nmonoidal categories.\n"}
{"abstract": "  We find an asymptotic enumeration formula for the number of simple\n$r$-uniform hypergraphs with a given degree sequence, when the number of edges\nis sufficiently large. The formula is given in terms of the solution of a\nsystem of equations. We give sufficient conditions on the degree sequence which\nguarantee existence of a solution to this system. Furthermore, we solve the\nsystem and give an explicit asymptotic formula when the degree sequence is\nclose to regular. This allows us to establish several properties of the degree\nsequence of a random $r$-uniform hypergraph with a given number of edges. More\nspecifically, we compare the degree sequence of a random $r$-uniform hypergraph\nwith a given number edges to certain models involving sequences of binomial or\nhypergeometric random variables conditioned on their sum.\n"}
{"abstract": "  Precise determination and assessment of bladder cancer (BC) extent of muscle\ninvasion involvement guides proper risk stratification and personalized therapy\nselection. In this context, segmentation of both bladder walls and cancer are\nof pivotal importance, as it provides invaluable information to stage the\nprimary tumour. Hence, multi region segmentation on patients presenting with\nsymptoms of bladder tumours using deep learning heralds a new level of staging\naccuracy and prediction of the biologic behaviour of the tumour. Nevertheless,\ndespite the success of these models in other medical problems, progress in\nmulti region bladder segmentation is still at a nascent stage, with just a\nhandful of works tackling a multi region scenario. Furthermore, most existing\napproaches systematically follow prior literature in other clinical problems,\nwithout casting a doubt on the validity of these methods on bladder\nsegmentation, which may present different challenges. Inspired by this, we\nprovide an in-depth look at bladder cancer segmentation using deep learning\nmodels. The critical determinants for accurate differentiation of muscle\ninvasive disease, current status of deep learning based bladder segmentation,\nlessons and limitations of prior work are highlighted.\n"}
{"abstract": "  In this paper we address the topic of a possible path to standardize the ICT\nservices expected to be delivered by the so-called \"Port of the Future\". How\nthe most relevant technologies and Information Systems are used by the Port\nCommunities for their businesses is discussed together with a detailed analysis\nof the on-going actions carried on by Standard Setting Organizations.\nConsidering the examples given by the C-ITS Platform and the C-Roads programme\nat EU level, a proposal of contents to be considered in a comprehensive\nstandardization action is given. The innovation services are therefore grouped\ninto four bundles: (i) Vessel \\& Marine Navigation, (ii) e-Freight \\&\n(Intermodal) Logistics, (iii) Passenger Transport, (iv) Environmental\nsustainability. The standardized version of these applications will be finally\nlabeled as C-Port services. Alongside the standardization plan, a proposal for\nranking the ports on the basis of a specially-defined C- Port vector is\ndiscussed with the purpose of addressing the well-known lack of consensus\naround the mathematical definition of the Smart Port Index. Considering the\ngood practice and the background offered by the Port of Livorno in terms of\ninnovation actions, the prospected final user applications are then labeled as\nDay 1, Day 1.5, and Day 2 services in consideration of the technical and\ncommercial gaps to be filled. As a case study about the evolution in the C-Port\nvector experienced by the Port of Livorno in the last years will also be\ndiscussed.\n"}
{"abstract": "  We study the homologous collapse of stellar nuclear core, the virial theorem\nfor hadron collisional relaxations, and photon productions from hadron\ncollisions. We thus show the gravo-thermal dynamical process that transforms\ngravitational energy to photon energy. The process is energetically and\nentropically favourable. The total baryon number conservation, Euler equation\nfor energy-momentum conservation and Poisson's equation for gravitational\npotential are adopted to describe homologous core collapses. The virial theorem\ndetermines the hadron collision energy gain from gravitational potential. The\nhadronic photon production rate determines the photon energy density. The time\nscales of macroscopic and microscopic processes are studied to verify\napproximations. As a result, we show the formation of opaque photon-pair\nspheres, whose total energy, size, temperature and number density, accounting\nfor the main energetic features of Gamma-Ray Burst progenitors. We obtain the\nintrinsic correlations of these quantities. They depend only on the averaged\nthermal index of the stellar core. We discuss the possibility to confront them\nwith observational data.\n"}
{"abstract": "  The etching process is one of the most important processes in semiconductor\nmanufacturing. We have introduced the state-of-the-art deep learning model to\npredict the etching profiles. However, the significant problems violating\nphysics have been found through various techniques such as explainable\nartificial intelligence and representation of prediction uncertainty. To\naddress this problem, this paper presents a novel approach to apply the\ninductive biases for etching process. We demonstrate that our approach fits the\nmeasurement faster than physical simulator while following the physical\nbehavior. Our approach would bring a new opportunity for better etching process\nwith higher accuracy and lower cost.\n"}
{"abstract": "  An isoparametric family in the unit sphere consists of parallel isoparametric\nhypersurfaces and their two focal submanifolds. The present paper has two\nparts. The first part investigates topology of the isoparametric families,\nnamely the homotopy, homeomorphism, or diffeomorphism types, parallelizability,\nas well as the Lusternik-Schnirelmann category. This part extends substantially\nthe results of Q.M.Wang in \\cite{Wa88}. The second part is concerned with their\ncurvatures, more precisely, we determine when they have non-negative sectional\ncurvatures or positive Ricci curvatures with the induced metric.\n"}
{"abstract": "  We rigorously compute the integrable system for the limiting\n$(N\\rightarrow\\infty)$ distribution function of the extreme momentum of $N$\nnoninteracting fermions when confined to an anharmonic trap $V(q)=q^{2n}$ for\n$n\\in\\mathbb{Z}_{\\geq 1}$ at positive temperature. More precisely, the edge\nmomentum statistics in the harmonic trap $n=1$ are known to obey the weak\nasymmetric KPZ crossover law which is realized via the finite temperature Airy\nkernel determinant or equivalently via a Painlev\\'e-II integro-differential\ntranscendent, cf. \\cite{LW,ACQ}. For general $n\\geq 2$, a novel higher order\nfinite temperature Airy kernel has recently emerged in physics literature\n\\cite{DMS} and we show that the corresponding edge law in momentum space is now\ngoverned by a distinguished Painlev\\'e-II integro-differential hierarchy. Our\nanalysis is based on operator-valued Riemann-Hilbert techniques which produce a\nLax pair for an operator-valued Painlev\\'e-II ODE system that naturally encodes\nthe aforementioned hierarchy. As byproduct, we establish a connection of the\nintegro-differential Painlev\\'e-II hierarchy to a novel integro-differential\nmKdV hierarchy.\n"}
{"abstract": "  We propose a novel type of a Bose-Hubbard ladder model based on an open\nquantum-gas--cavity-QED setup to study the physics of dynamical gauge\npotentials. Atomic tunneling along opposite directions in the two legs of the\nladder is mediated by photon scattering from transverse pump lasers to two\ndistinct cavity modes. The resulting interplay between cavity photon\ndissipation and the optomechanical atomic back-action then induces an\naverage-density-dependent dynamical gauge field. The dissipation-stabilized\nsteady-state atomic motion along the legs of the ladder leads either to a pure\nchiral current, screening the induced dynamical magnetic field as in the\nMeissner effect, or generates simultaneously chiral and particle currents. For\nsufficiently strong pump the system enters into a dynamically unstable regime\nexhibiting limit-cycle and period-doubled oscillations. Intriguingly, an\nelectromotive force is induced in this dynamical regime as expected from an\ninterpretation based on Faraday's law of induction for the time-dependent\nsynthetic magnetic flux.\n"}
{"abstract": "  In 2006, Biere, Jussila, and Sinz made the key observation that the\nunderlying logic behind algorithms for constructing Reduced, Ordered Binary\nDecision Diagrams (BDDs) can be encoded as steps in a proof in the extended\nresolution logical framework. Through this, a BDD-based Boolean satisfiability\n(SAT) solver can generate a checkable proof of unsatisfiability for a set of\nclauses. Such a proof indicates that the formula is truly unsatisfiable without\nrequiring the user to trust the BDD package or the SAT solver built on top of\nit.\n  We extend their work to enable arbitrary existential quantification of the\nformula variables, a critical capability for BDD-based SAT solvers. We\ndemonstrate the utility of this approach by applying a prototype solver to\nseveral problems that are very challenging for search-based SAT solvers,\nobtaining polynomially sized proofs on benchmarks for parity formulas, as well\nas the Urquhart, mutilated chessboard, and pigeonhole problems.\n"}
{"abstract": "  Ever since Machine Learning as a Service (MLaaS) emerges as a viable business\nthat utilizes deep learning models to generate lucrative revenue, Intellectual\nProperty Right (IPR) has become a major concern because these deep learning\nmodels can easily be replicated, shared, and re-distributed by any unauthorized\nthird parties. To the best of our knowledge, one of the prominent deep learning\nmodels - Generative Adversarial Networks (GANs) which has been widely used to\ncreate photorealistic image are totally unprotected despite the existence of\npioneering IPR protection methodology for Convolutional Neural Networks (CNNs).\nThis paper therefore presents a complete protection framework in both black-box\nand white-box settings to enforce IPR protection on GANs. Empirically, we show\nthat the proposed method does not compromise the original GANs performance\n(i.e. image generation, image super-resolution, style transfer), and at the\nsame time, it is able to withstand both removal and ambiguity attacks against\nembedded watermarks.\n"}
{"abstract": "  To obtain the accurate transient states of the big scale natural gas pipeline\nnetworks under the bad data and non-zero mean noises conditions, a robust\nKalman filter-based dynamic state estimation method is proposed using the\nlinearized gas pipeline transient flow equations in this paper. Firstly, the\ndynamic state estimation model is built. Since the gas pipeline transient flow\nequations are less than the states, the boundary conditions are used as\nsupplementary constraints to predict the transient states. To increase the\nmeasurement redundancy, the zero mass flow rate constraints at the sink nodes\nare taken as virtual measurements. Secondly, to ensure the stability under bad\ndata condition, the robust Kalman filter algorithm is proposed by introducing a\ntime-varying scalar matrix to regulate the measurement error variances\ncorrectly according to the innovation vector at every time step. At last, the\nproposed method is applied to a 30-node gas pipeline networks in several kinds\nof measurement conditions. The simulation shows that the proposed robust\ndynamic state estimation can decrease the effects of bad data and achieve\nbetter estimating results.\n"}
{"abstract": "  The Alcubierre warp drive metric is a spacetime construction where a massive\nparticle located inside a spacetime distortion, called warp bubble, travels at\nvelocities arbitrarily higher than the velocity of light. This theoretically\nconstructed spacetime geometry is a consequence of general relativity where\nglobal superluminal velocities, also known as warp speeds, are possible,\nwhereas local speeds are limited to subluminal ones as required by special\nrelativity. In this work we analyze the solutions of the Einstein equations\nhaving charged dust energy-momentum tensor as source for warp velocities. The\nEinstein equations with the cosmological constant are written and all solutions\nhaving energy-momentum tensor components for electromagnetic fields generated\nby charged dust are presented, as well as the respective energy conditions. The\nresults show an interplay between the energy conditions and the electromagnetic\nfield such that in some cases the former can be satisfied by both positive and\nnegative matter density. In other cases the dominant and null energy conditions\nare violated. A result connecting the electric energy density with the\ncosmological constant is also presented, as well as the effects of the\nelectromagnetic field on the bubble dynamics.\n"}
{"abstract": "  Annotated data has become the most important bottleneck in training accurate\nmachine learning models, especially for areas that require domain expertise. A\nrecent approach to deal with the above issue proposes using natural language\nexplanations instead of labeling individual data points, thereby increasing\nhuman annotators' efficiency as well as decreasing costs substantially. This\npaper focuses on the task of turning these natural language descriptions into\nPython labeling functions by following a novel approach to semantic parsing\nwith pre-trained text-to-text Transformers. In a series of experiments our\napproach achieves a new state of the art on the semantic parsing benchmark\nCoNaLa, surpassing the previous best approach by 3.7 BLEU points. Furthermore,\non a manually constructed dataset of natural language descriptions-labeling\nfunctions pairs we achieve a BLEU of 0.39. Our approach can be regarded as a\nstepping stone towards models that are taught how to label in natural language,\ninstead of being provided specific labeled samples. Our code, constructed\ndataset and models are available at\nhttps://github.com/ypapanik/t5-for-code-generation.\n"}
{"abstract": "  For sensitive text data to be shared among NLP researchers and practitioners,\nshared documents need to comply with data protection and privacy laws. There is\nhence a growing interest in automated approaches for text anonymization.\nHowever, measuring such methods' performance is challenging: missing a single\nidentifying attribute can reveal an individual's identity. In this paper, we\ndraw attention to this problem and argue that researchers and practitioners\ndeveloping automated text anonymization systems should carefully assess whether\ntheir evaluation methods truly reflect the system's ability to protect\nindividuals from being re-identified. We then propose TILD, a set of evaluation\ncriteria that comprises an anonymization method's technical performance, the\ninformation loss resulting from its anonymization, and the human ability to\nde-anonymize redacted documents. These criteria may facilitate progress towards\na standardized way for measuring anonymization performance.\n"}
{"abstract": "  Octupole deformations and related collective excitations are analyzed using\nthe framework of nuclear density functional theory. Axially-symmetric\nquadrupole-octupole constrained self-consistent mean-field (SCMF) calculations\nwith a choice of universal energy density functional and a pairing interaction\nare performed for Xe, Ba, and Ce isotopes from proton-rich to neutron-rich\nregions, and neutron-rich Se, Kr, and Sr isotopes, in which enhanced octupole\ncorrelations are expected to occur. Low-energy positive- and negative-parity\nspectra and transition strengths are computed by solving the\nquadrupole-octupole collective Hamiltonian, with the inertia parameters and\ncollective potential determined by the constrained SCMF calculations.\nOctupole-deformed equilibrium states are found in the potential energy surfaces\nof the Ba and Ce isotopes with $N\\approx 56$ and 88. The evolution of\nspectroscopic properties indicates enhanced octupole correlations in the\nregions corresponding to $N\\approx Z\\approx 56$, $Z\\approx 88$ and $Z\\approx\n56$, and $N\\approx 56$ and $Z\\approx 34$. The average $\\beta_{30}$ deformation\nparameter and its fluctuation exhibit signatures of octupole shape phase\ntransition around $N=56$ and 88.\n"}
{"abstract": "  We study the images of the Markov measures under transformations generated by\nthe Mealy automata. We find conditions under which the image measure is\nabsolutely continuous or singular relative to the Markov measure. Also, we\ndetermine statistical properties of the image of a generic sequence.\n"}
{"abstract": "  The bidomain model is the standard model for cardiac electrophysiology. In\nthis paper, we investigate the instability and asymptotic behavior of planar\nfronts and planar pulses of the bidomain Allen-Cahn equation and the bidomain\nFitzHugh-Nagumo equation in two spatial dimension. In previous work, it was\nshown that planar fronts of the bidomain Allen-Cahn equation can become\nunstable in contrast to the classical Allen-Cahn equation. We find that, after\nthe planar front is destabilized, a rotating zigzag front develops whose shape\ncan be explained by simple geometric arguments using a suitable Frank diagram.\nWe also show that the Hopf bifurcation through which the front becomes unstable\ncan be either supercritical or subcritical, by demonstrating a parameter regime\nin which a stable planar front and zigzag front can coexist. In our\ncomputational studies of the bidomain FitzHugh-Nagumo pulse solution, we show\nthat the pulses can also become unstable much like the bidomain Allen-Cahn\nfronts. However, unlike the bidomain Allen-Cahn case, the destabilized pulse\ndoes not necessarily develop into a zigzag pulse. For certain choice of\nparameters, the destabilized pulse can disintegrate entirely. These studies are\nmade possible by the development of a numerical scheme that allows for the\naccurate computation of the bidomain equation in a two dimensional strip domain\nof infinite extent.\n"}
{"abstract": "  Joint Energy-based Model (JEM) of Grathwohl et al. shows that a standard\nsoftmax classifier can be reinterpreted as an energy-based model (EBM) for the\njoint distribution p(x,y); the resulting model can be optimized to improve\ncalibration, robustness, and out-of-distribution detection, while generating\nsamples rivaling the quality of recent GAN-based approaches. However, the\nsoftmax classifier that JEM exploits is inherently discriminative and its\nlatent feature space is not well formulated as probabilistic distributions,\nwhich may hinder its potential for image generation and incur training\ninstability. We hypothesize that generative classifiers, such as Linear\nDiscriminant Analysis (LDA), might be more suitable for image generation since\ngenerative classifiers model the data generation process explicitly. This paper\ntherefore investigates an LDA classifier for image classification and\ngeneration. In particular, the Max-Mahalanobis Classifier (MMC), a special case\nof LDA, fits our goal very well. We show that our Generative MMC (GMMC) can be\ntrained discriminatively, generatively, or jointly for image classification and\ngeneration. Extensive experiments on multiple datasets show that GMMC achieves\nstate-of-the-art discriminative and generative performances, while\noutperforming JEM in calibration, adversarial robustness, and\nout-of-distribution detection by a significant margin. Our source code is\navailable at https://github.com/sndnyang/GMMC.\n"}
{"abstract": "  In this paper, we propose a novel joint intelligent trajectory design and\nresource allocation algorithm based on user's mobility and their requested\nservices for unmanned aerial vehicles (UAVs) assisted networks, where UAVs act\nas nodes of a network function virtualization (NFV) enabled network. Our\nobjective is to maximize energy efficiency and minimize the average delay on\nall services by allocating the limited radio and NFV resources. In addition,\ndue to the traffic conditions and mobility of users, we let some Virtual\nNetwork Functions (VNFs) to migrate from their current locations to other\nlocations to satisfy the Quality of Service requirements. We formulate our\nproblem to find near-optimal locations of UAVs, transmit power, subcarrier\nassignment, placement, and scheduling the requested service's functions over\nthe UAVs and perform suitable VNF migration. Then we propose a novel\nHierarchical Hybrid Continuous and Discrete Action (HHCDA) deep reinforcement\nlearning method to solve our problem. Finally, the convergence and\ncomputational complexity of the proposed algorithm and its performance analyzed\nfor different parameters. Simulation results show that our proposed HHCDA\nmethod decreases the request reject rate and average delay by 31.5% and 20% and\nincreases the energy efficiency by 40% compared to DDPG method.\n"}
{"abstract": "  Let $Y$ be a smooth hypersurface in a projective irreducible holomorphic\nsymplectic manifold X of dimension 2n. The characteristic foliation $F$ is the\nkernel of the symplectic form restricted to Y. In this article we prove that a\ngeneric leaf of the characteristic foliation is dense in Y if Y has positive\nBeauville-Bogomolov-Fujiki square.\n"}
{"abstract": "  This paper presents our findings for using activity modeling for simulation\n(validation), model checking (verification), and execution purposes. Each is\nneeded to tackle system complexity and further research into behavioral\nmodeling. We argue different models implicate different understandings and\nexpectations. We emphasize some distinctions with demonstrations using the\nDiscrete Event System Specification with an exemplary model. In particular, the\ncontinuous-time base in models helps observe some inherent limitations and\nstrengths in acquiring each capability. The temporal characterization of input,\noutput, and state, or the lack thereof, is at the core of developing behavioral\nspecifications. We use DEVS to arrive at the capability of validating\nsimulations for activity models. We use Constrained-DEVS for the verification\nof activity models. We show how some executions can be derived, whether from\nthe specification itself or with considerations for simulation and model\nchecking.\n"}
{"abstract": "  Urbanization of an area is known to increase the temperature of the\nsurrounding area. This phenomenon -- a so-called urban heat island (UHI) --\noccurs at a local level over a period of time and has lasting impacts for\nhistorical data analysis. We propose a methodology to examine if long-term\nchanges in temperature increases and decreases across time exist (and to what\nextent) at the local level for a given set of temperature readings at various\nlocations. Specifically, we propose a Bayesian change point model for\nspatio-temporally dependent data where we select the number of change points at\neach location using a \"forwards\" selection process using deviance information\ncriteria (DIC). We then fit the selected model and examine the linear slopes\nacross time to quantify changes in long-term temperature behavior. We show the\nutility of this model and method using a synthetic data set and temperature\nmeasurements from eight stations in Utah consisting of daily temperature data\nfor 60 years.\n"}
{"abstract": "  Variational quantum eigensolver (VQE) that solves the ground state of a\nHamiltonian is a promising application of noisy intermediate-scale quantum\ndevices. The quantum alternating operator Ansatz (QAOA) is one of the most\nwidely studied Ans\\\"atze for this purpose and solves many Hamiltonians\nreliably. However, because of inherited symmetries from the Hamiltonian, the\nQAOA is not necessarily good at solving problems with symmetry broken nature.\nIn this paper, we propose a variational Ansatz with symmetry breaking layers\nfor solving those systems. Notably, our Ansatz finds a constant-depth circuit\nfor solving a symmetry broken ground state which is impossible for the QAOA. We\nalso propose a simple learning technique that can choose a particular symmetry\nbroken state among degenerate ground states.\n"}
{"abstract": "  Here we consider the image of the principal minor map of symmetric matrices\nover an arbitrary unique factorization domain $R$. By exploiting a connection\nwith symmetric determinantal representations, we characterize the image of the\nprincipal minor map through the condition that certain polynomials coming from\nso-called Rayleigh differences are squares in the polynomial ring over $R$. In\nalmost all cases, one can characterize the image of the principal minor map\nusing the orbit of Cayley's hyperdeterminant under the action of $(SL_2(R))^{n}\n\\rtimes S_{n}$. Over the complex numbers, this recovers a characterization of\nOeding from 2011, and over the reals, the orbit of a single additional\nquadratic inequality suffices to cut out the image. Applications to other\nsymmetric determinantal representations are also discussed.\n"}
{"abstract": "  Due to the importance of the lower bounding distances and the attractiveness\nof symbolic representations, the family of symbolic aggregate approximations\n(SAX) has been used extensively for encoding time series data. However, typical\nSAX-based methods rely on two restrictive assumptions; the Gaussian\ndistribution and equiprobable symbols. This paper proposes two novel\ndata-driven SAX-based symbolic representations, distinguished by their\ndiscretization steps. The first representation, oriented for general data\ncompaction and indexing scenarios, is based on the combination of kernel\ndensity estimation and Lloyd-Max quantization to minimize the information loss\nand mean squared error in the discretization step. The second method, oriented\nfor high-level mining tasks, employs the Mean-Shift clustering method and is\nshown to enhance anomaly detection in the lower-dimensional space. Besides, we\nverify on a theoretical basis a previously observed phenomenon of the intrinsic\nprocess that results in a lower than the expected variance of the intermediate\npiecewise aggregate approximation. This phenomenon causes an additional\ninformation loss but can be avoided with a simple modification. The proposed\nrepresentations possess all the attractive properties of the conventional SAX\nmethod. Furthermore, experimental evaluation on real-world datasets\ndemonstrates their superiority compared to the traditional SAX and an\nalternative data-driven SAX variant.\n"}
{"abstract": "  In this paper, we focus on the rendezvous problem for the autonomous\ncooperative landing of an unmanned aerial vehicle (UAV) on an unmanned surface\nvehicle (USV). These heterogeneous agents with nonlinear dynamics are\ndynamically decoupled but share a common cooperative rendezvous task. The\nunderlying control scheme is based on the Distributed Model Predictive Control\n(MPC). One of our main contributions is a rendezvous algorithm with an online\nupdate rule of the rendezvous location. The algorithm requires that agents\nupdate the rendezvous location only when they are not guaranteed to reach it.\nTherefore, the exchange of information occurs aperiodically and proposed\nalgorithm improves the communication efficiency. Furthermore, we prove the\nrecursive feasibility of the algorithm. The simulation results show the\neffectiveness of our algorithm applied to the problem of autonomous cooperative\nlanding.\n"}
{"abstract": "  ASCENT is a fully automated methodology for extracting and consolidating\ncommonsense assertions from web contents (Nguyen et al., WWW 2021). It advances\ntraditional triple-based commonsense knowledge representation by capturing\nsemantic facets like locations and purposes, and composite concepts, i.e.,\nsubgroups and related aspects of subjects. In this demo, we present a web\nportal that allows users to understand its construction process, explore its\ncontent, and observe its impact in the use case of question answering. The demo\nwebsite and an introductory video are both available online.\n"}
{"abstract": "  In the context of statistical learning, the Information Bottleneck method\nseeks a right balance between accuracy and generalization capability through a\nsuitable tradeoff between compression complexity, measured by minimum\ndescription length, and distortion evaluated under logarithmic loss measure. In\nthis paper, we study a variation of the problem, called scalable information\nbottleneck, in which the encoder outputs multiple descriptions of the\nobservation with increasingly richer features. The model, which is of\nsuccessive-refinement type with degraded side information streams at the\ndecoders, is motivated by some application scenarios that require varying\nlevels of accuracy depending on the allowed (or targeted) level of complexity.\nWe establish an analytic characterization of the optimal relevance-complexity\nregion for vector Gaussian sources. Then, we derive a variational inference\ntype algorithm for general sources with unknown distribution; and show means of\nparametrizing it using neural networks. Finally, we provide experimental\nresults on the MNIST dataset which illustrate that the proposed method\ngeneralizes better to unseen data during the training phase.\n"}
{"abstract": "  This paper is intended to solve the nonconvex $\\ell_{p}$-ball constrained\nnonlinear optimization problems. An iteratively reweighted method is proposed,\nwhich solves a sequence of weighted $\\ell_{1}$-ball projection subproblems. At\neach iteration, the next iterate is obtained by moving along the negative\ngradient with a stepsize and then projecting the resulted point onto the\nweighted $\\ell_{1}$ ball to approximate the $\\ell_{p}$ ball. Specifically, if\nthe current iterate is in the interior of the feasible set, then the weighted\n$\\ell_{1}$ ball is formed by linearizing the $\\ell_{p}$ norm at the current\niterate. If the current iterate is on the boundary of the feasible set, then\nthe weighted $\\ell_{1}$ ball is formed differently by keeping those zero\ncomponents in the current iterate still zero. In our analysis, we prove that\nthe generated iterates converge to a first-order stationary point. Numerical\nexperiments demonstrate the effectiveness of the proposed method.\n"}
{"abstract": "  We study black and white hole analogues in Weyl semimetals with inhomogenous\nnodal tilts. We study how the presence of a microscopic lattice, giving rise to\nlow-energy fermion doubler states at large momenta that are not present for\nelementary particles, affects the analogy between Weyl Hamiltonians and general\nrelativity. Using a microscopic tight-binding lattice model, we find the\ndoubler states to give rise to Hawking fragmentation and Hawking attenuation of\nwavepackets by the analogue event horizon. These phenomena depend on an\nanalogue Hawking temperature, and can be measured in metamaterials and solids,\nas we confirm by numerical simulations.\n"}
{"abstract": "  In learning-phase clinical trials in drug development, adaptive designs can\nbe efficient and highly informative when used appropriately. In this article,\nwe extend the multiple comparison procedures with modeling techniques (MCP-Mod)\nprocedure with generalized multiple contrast tests (GMCTs) to two-stage\nadaptive designs for establishing proof-of-concept. The results of an interim\nanalysis of first-stage data are used to adapt the candidate dose-response\nmodels and the dosages studied in the second stage. GMCTs are used in both\nstages to obtain stage-wise p-values, which are then combined to determine an\noverall p-value. An alternative approach is also considered that combines the\nt-statistics across stages, employing the conditional rejection probability\n(CRP) principle to preserve the Type I error probability. Simulation studies\ndemonstrate that the adaptive designs are advantageous compared to the\ncorresponding tests in a non-adaptive design if the selection of the candidate\nset of dose-response models is not well informed by evidence from preclinical\nand early-phase studies.\n"}
{"abstract": "  Type- and token-based embedding architectures are still competing in lexical\nsemantic change detection. The recent success of type-based models in\nSemEval-2020 Task 1 has raised the question why the success of token-based\nmodels on a variety of other NLP tasks does not translate to our field. We\ninvestigate the influence of a range of variables on clusterings of BERT\nvectors and show that its low performance is largely due to orthographic\ninformation on the target word, which is encoded even in the higher layers of\nBERT representations. By reducing the influence of orthography we considerably\nimprove BERT's performance.\n"}
{"abstract": "  With 1.85 billion tons produced per year, steel is the most important\nmaterial class in terms of volume and environmental impact. While steel is a\nsustainability enabler, for instance through lightweight design, magnetic\ndevices, and efficient turbines, its primary production is not. For 3000 years,\niron has been reduced from ores using carbon. Today 2.1 tons CO2 are produced\nper ton of steel, causing 30% of the global CO2 emissions in the manufacturing\nsector, which translates to 6.5% of the global CO2 emissions. These numbers\nqualify iron- and steel-making as the largest single industrial greenhouse gas\nemission source. The envisaged future industrial route to mitigate these CO2\nemissions targets green hydrogen as a reductant. Although this reaction has\nbeen studied for decades, its kinetics is not well understood, particularly\nduring the wustite reduction step which is dramatically slower than the\nhematite reduction. Many rate-limiting factors of this reaction are set by the\nmicro- and nanostructure as well as the local chemistry of the ores. Their\nquantification allows knowledge-driven ore preparation and process optimization\nto make the hydrogen-based reduction of iron ores commercially viable, enabling\nthe required massive CO2 mitigation to ease global warming. Here, we report on\na multi-scale structure and composition analysis of iron reduced from hematite\nwith pure H2, reaching down to near-atomic scale.\n"}
{"abstract": "  In the past, the cumulative metallicity distribution function (CMDF) turned\nout as a useful tool to constrain the accretion history of various components\nof the Milky Way. In this Letter, by means of analytical, leaky-box chemical\nevolution models (i.e. including both infall and galactic outflows) we study\nthe CMDF of local star-forming galaxies that follow two fundamental empirical\nscaling relations, namely the mass-metallicity and main sequence relations. Our\nanalysis shows that galactic winds, which are dominant mostly in low-mass\nsystems, play a fundamental role in shaping this function and, in particular,\nin determining its steepness and curvature. We show that the CMDF of low-mass\n(M$_{\\star}$/M$_{\\odot} \\le 10^{9.5}$) and high-mass\n(M$_{\\star}$/M$_{\\odot}$>10$^{10.5}$) galaxies deviate substantially from the\nresults of a 'closed-box' model, as the evolution of the former (latter)\nsystems is mostly dominated by outflows (infall). In the context of galactic\ndownsizing, we show that downward-concave CMDFs (associated with systems with\nextremely small infall timescales and with very strong winds) are more frequent\nin low-mass galaxies, which include larger fractions of young systems and\npresent more substantial deviations from equilibrium between gas accretion and\nreprocessing (either via star formation or winds).\n"}
{"abstract": "  The problem of robustness under location deformations for deep convolutional\nneural networks (DCNNs) is of great theoretical and practical interest. This\nissue has been studied in pioneering works, especially for scattering-type\narchitectures, for deformation vector fields $\\tau(x)$ with some regularity -\nat least $C^1$. Here we address this issue for any field $\\tau\\in\nL^\\infty(\\mathbb{R}^d;\\mathbb{R}^d)$, without any additional regularity\nassumption, hence including the case of wild irregular deformations such as a\nnoise on the pixel location of an image. We prove that for signals in\nmultiresolution approximation spaces $U_s$ at scale $s$, whenever the network\nis Lipschitz continuous (regardless of its architecture), stability in $L^2$\nholds in the regime $\\|\\tau\\|_{L^\\infty}/s\\ll 1$, essentially as a consequence\nof the uncertainty principle. When $\\|\\tau\\|_{L^\\infty}/s\\gg 1$ instability can\noccur even for well-structured DCNNs such as the wavelet scattering networks,\nand we provide a sharp upper bound for the asymptotic growth rate. The\nstability results are then extended to signals in the Besov space\n$B^{d/2}_{2,1}$ tailored to the given multiresolution approximation. We also\nconsider the case of more general time-frequency deformations. Finally, we\nprovide stochastic versions of the aforementioned results, namely we study the\nissue of stability in mean when $\\tau(x)$ is modeled as a random field (not\nbounded, in general) with with identically distributed variables $|\\tau(x)|$,\n$x\\in\\mathbb{R}^d$.\n"}
{"abstract": "  We study endomorphisms and derivations of infinite dimensional cyclic Leibniz\nalgebra.\n"}
{"abstract": "  This paper presents an experimental and numerical investigation of\nsolid-liquid fluidized beds consisting of bonded spheres in very narrow tubes,\ni.e., when the ratio between the tube and grain diameters is small. In narrow\nbeds, high confinement effects have proved to induce crystallization, jamming\nand different patterns, which can be intensified or modified if some grains are\nbonded together. In order to investigate that, we produced duos and trios of\nbonded aluminum spheres with diameter of 4.8 mm, and formed beds consisting\neither of 150-300 duos or 100-200 trios in a 25.4 mm-ID pipe, which were\nsubmitted to water velocities above those necessary for fluidization. For the\nexperiments, we filmed the bed with high-speed and conventional cameras and\nprocessed the images, obtaining measurements at both the bed and grain scales.\nFor the numerical part, we computed the bed evolution for the same conditions\nwith a CFD-DEM (computational fluid dynamics - discrete element method) code.\nOur results show distinct motions for individual duos and trios, and different\nstructures within the bed. We also found that jamming may occur suddenly for\ntrios, where even the microscopic motion (fluctuation at the grain scale)\nstops, calling into question the fluidization conditions for those cases.\n"}
{"abstract": "  Star formation in galaxies is regulated by the heating and cooling in the\ninterstellar medium. In particular, the processing of molecular gas into stars\nwill depend strongly on the ratio of gas heating to gas cooling in the neutral\ngas around sites of recent star-formation. In this work, we combine\nmid-infrared (mid-IR) observations of Polycyclic Aromatic Hydrocarbons (PAHs),\nthe dominant heating mechanism of gas in the interstellar medium (ISM), with [C\nII], [O I], and [Si II] fine-structure emission, the strongest cooling channels\nin dense, neutral gas. The ratio of IR cooling line emission to PAH emission\nmeasures the photoelectric efficiency, a property of the ISM which dictates how\nmuch energy carried by ultraviolet photons gets transferred into the gas. We\nfind that star-forming, IR luminous galaxies in the Great Observatories All-Sky\nLIRG Survey (GOALS) with high IR surface densities have low photoelectric\nefficiencies. These systems also have, on average, higher ratios of radiation\nfield strength to gas densities, and larger average dust grain size\ndistributions. The data support a scenario in which the most compact galaxies\nhave more young star-forming regions per unit area, which exhibit less\nefficient gas heating. These conditions may be more common at high-z, and may\nhelp explain the higher star-formation rates at cosmic noon. We make\npredictions on how this can be investigated with JWST.\n"}
{"abstract": "  Nowadays, sensors play a major role in several contexts like science,\nindustry and daily life which benefit of their use. However, the retrieved\ninformation must be reliable. Anomalies in the behavior of sensors can give\nrise to critical consequences such as ruining a scientific project or\njeopardizing the quality of the production in industrial production lines. One\nof the more subtle kind of anomalies are uncalibrations. An uncalibration is\nsaid to take place when the sensor is not adjusted or standardized by\ncalibration according to a ground truth value. In this work, an online\nmachine-learning based uncalibration detector for temperature, humidity and\npressure sensors was developed. This solution integrates an Artificial Neural\nNetwork as main component which learns from the behavior of the sensors under\ncalibrated conditions. Then, after trained and deployed, it detects\nuncalibrations once they take place. The obtained results show that the\nproposed solution is able to detect uncalibrations for deviation values of 0.25\ndegrees, 1% RH and 1.5 Pa, respectively. This solution can be adapted to\ndifferent contexts by means of transfer learning, whose application allows for\nthe addition of new sensors, the deployment into new environments and the\nretraining of the model with minimum amounts of data.\n"}
{"abstract": "  360{\\deg} videos a.k.a. spherical videos are getting popular among users\nnevertheless, omnidirectional view of these videos demands high bandwidth and\nprocessing power at the end devices. Recently proposed viewport aware streaming\nmechanisms can reduce the amount of data transmitted by streaming a limited\nportion of the frame covering the current user viewport (VP). However, they\nstill suffer from sending a high amount of redundant data, as the fixed tile\nmechanisms can not provide finer granularity to the user VP. Though making the\ntiles smaller can provide a finer granularity for user viewport, high encoding\noverhead incurred. To overcome this trade-off, in this paper, we present a\ncomputational geometric approach based adaptive tiling mechanism named VAD360,\nwhich takes visual attention information on the 360{\\deg} video frame as the\ninput and provide a suitable non-overlapping variable size tile cover on the\nframe. Experimental results shows that VAD360 can save up to 31.1% of pixel\nredundancy before compression and 35.4% of bandwidth saving compared to\nrecently proposed fixed tile configurations, providing tile schemes within\n0.98($\\pm$0.11)s time frame.\n"}
{"abstract": "  While there has been significant progress towards modelling coherence in\nwritten discourse, the work in modelling spoken discourse coherence has been\nquite limited. Unlike the coherence in text, coherence in spoken discourse is\nalso dependent on the prosodic and acoustic patterns in speech. In this paper,\nwe model coherence in spoken discourse with audio-based coherence models. We\nperform experiments with four coherence-related tasks with spoken discourses.\nIn our experiments, we evaluate machine-generated speech against the speech\ndelivered by expert human speakers. We also compare the spoken discourses\ngenerated by human language learners of varying language proficiency levels.\nOur results show that incorporating the audio modality along with the text\nbenefits the coherence models in performing downstream coherence related tasks\nwith spoken discourses.\n"}
{"abstract": "  Short textual descriptions of entities provide summaries of their key\nattributes and have been shown to be useful sources of background knowledge for\ntasks such as entity linking and question answering. However, generating entity\ndescriptions, especially for new and long-tail entities, can be challenging\nsince relevant information is often scattered across multiple sources with\nvaried content and style. We introduce DESCGEN: given mentions spread over\nmultiple documents, the goal is to generate an entity summary description.\nDESCGEN consists of 37K entity descriptions from Wikipedia and Fandom, each\npaired with nine evidence documents on average. The documents were collected\nusing a combination of entity linking and hyperlinks to the Wikipedia and\nFandom entity pages, which together provide high-quality distant supervision.\nThe resulting summaries are more abstractive than those found in existing\ndatasets and provide a better proxy for the challenge of describing new and\nemerging entities. We also propose a two-stage extract-then-generate baseline\nand show that there exists a large gap (19.9% in ROUGE-L) between\nstate-of-the-art models and human performance, suggesting that the data will\nsupport significant future work.\n"}
{"abstract": "  In this paper, we determine the detectability of eclipsing binary star\ncompanions from eclipse timing variations using the Kepler mission dataset.\nExtensive and precise stellar time-series photometry from space-based missions\nenable searches for binary star companions. However, due to the large datasets\nand computational resources involved, these searches would benefit from\nguidance from detection simulations. Our simulations start with and benefit\nfrom the use of empirical Kepler mission data, into which we inject third\nbodies to predict the resulting timing of binary star eclipses. We find that\nthe orbital eccentricity of the third body and the orbital period of the host\nbinary star are the key factors in detecting companions. Target brightness is\nalso likely to be a factor in detecting companions. Detectable third body\nmasses and periods can be efficiently bound using just two equations. Our\nresults enable the setting of realistic expectations when planning searches for\neclipsing binary star planetary and brown dwarf companions. Our results also\nsuggest the brown dwarf desert is real rather than observational selection.\n"}
{"abstract": "  In a bid to resolve lingering problems in cosmology, more focus is being\ntilted towards cosmological models in which physical constants of nature are\nnot necessarily real constants but vary with cosmic time. In this paper, we\nstudy a cosmological model in nonlinear electrodynamics with Newton's\ngravitational constant $G$, which is not a constant but varies in terms of a\npower law of the scale factor of the universe. Hence, the evolution of the\nscale factor $a(t)$ is studied in this model, which depends on a fine-tuning\nterm of nonlinear electrodynamics, $\\alpha$. Finally, we verify the stability\nof the model using the speed of sound.\n"}
{"abstract": "  The jet composition and radiative efficiency of GRBs are poorly constrained\nfrom the data. If the jet composition is matter-dominated (i.e. a fireball),\nthe GRB prompt emission spectra would include a dominant thermal component\noriginating from the fireball photosphere, and a non-thermal component\npresumably originating from internal shocks whose radii are greater than the\nphotosphere radius. We propose a method to directly dissect the GRB fireball\nenergy budget into three components and measure their values by combining the\nprompt emission and early afterglow data. The measured parameters include the\ninitial dimensionless specific enthalpy density ($\\eta$), bulk Lorentz factors\nat the photosphere radius ($\\Gamma_{\\rm ph}$) and before fireball deceleration\n($\\Gamma_0$), the amount of mass loading ($M$), as well as the GRB radiative\nefficiency ($\\eta_\\gamma$). All the parameters can be derived from the data for\na GRB with a dominant thermal spectral component, a deceleration bump feature\nin the early afterglow lightcurve, and a measured redshift. The results only\nweakly depend on the density $n$ of the interstellar medium when the\ncomposition ${\\cal Y}$ parameter (typically unity) is specified.\n"}
{"abstract": "  The interaction of polycyclic aromatic hydrocarbons (PAHs) with vacuum\nultraviolet (VUV) photons triggers the emission of the well-known aromatic\ninfrared bands (AIBs) but other mechanisms such as fragmentation can be\ninvolved in this interaction. Fragmentation leads to selection effects that\nfavor specific sizes and structures. We investigate the impact of aliphatic\nbonds on the VUV photo-stability of PAH cations under conditions applicable for\nphotodissociation regions (PDRs). Cations of pyrene (C16H10) and coronene\n(C24H12) derivatives containing aliphatic bonds (methyl or ethyl sidegroups,\nsuperhydrogenation) are submitted to VUV (10.5 eV) photons over long timescales\n(~1000 s) in the cryogenic PIRENEA setup. The fragmentation cascades are\nanalyzed with a simple kinetics model; fragmentation pathways, rates and\nbranching ratios are derived. Aliphatic PAH derivatives are found to have a\nhigher fragmentation rate and carbon loss compared to regular PAHs. The\nfragmentation of PAHs with alkylated sidegroups forms species with peripheral\npentagonal cycles, which can be more stable than the bare PAH cations. This\nstability is quantified and the most stable species, for which there is an\neffective competition of the fragmentation with isomerization and radiative\ncooling are identified. This work supports a scenario in which the evaporation\nof nanograins with a mixed aliphatic and aromatic composition followed by VUV\nphotoprocessing results in both the production of the carriers of the 3.4um AIB\nby methyl sidegroups and in an abundant source of small hydrocarbons at the\nborder of PDRs. An additional side effect is the efficient formation of stable\nPAHs that contain some peripheral pentagonal rings. Our experiments also\nsupport the role of isomerization processes in PAH photofragmentation,\nincluding the H-migration process, which could lead to an additional\ncontribution to the 3.4um AIB.\n"}
{"abstract": "  The paper presents two variants of a Krylov-Simplex iterative method that\ncombines Krylov and simplex iterations to minimize the residual $r = b-Ax$. The\nfirst method minimizes $\\|r\\|_\\infty$, i.e. maximum of the absolute residuals.\nThe second minimizes $\\|r\\|_1$, and finds the solution with the least absolute\nresiduals. Both methods search for an optimal solution $x_k$ in a Krylov\nsubspace which results in a small linear programming problem. A specialized\nsimplex algorithm solves this projected problem and finds the optimal linear\ncombination of Krylov basis vectors to approximate the solution. The resulting\nsimplex algorithm requires the solution of a series of small dense linear\nsystems that only differ by rank-one updates. The $QR$ factorization of these\nmatrices is updated each iteration. We demonstrate the effectiveness of the\nmethods with numerical experiments.\n"}
{"abstract": "  A general framework for solving nonlinear least squares problems without the\nemployment of derivatives is proposed in the present paper together with a new\ngeneral global convergence theory. With the aim to cope with the case in which\nthe number of variables is big (for the standards of derivative-free\noptimization), two dimension-reduction procedures are introduced. One of them\nis based on iterative subspace minimization and the other one is based on\nspline interpolation with variable nodes. Each iteration based on those\nprocedures is followed by an acceleration step inspired in the Sequential\nSecant Method. The practical motivation for this work is the estimation of\nparameters in Hydraulic models applied to dam breaking problems. Numerical\nexamples of the application of the new method to those problems are given.\n"}
{"abstract": "  We establish the existence of chimera states, simultaneously supporting\nsynchronous and asynchronous dynamics, in a network consisting of two\nsymmetrically linked star subnetworks consisting of identical oscillators with\nshear and Kuramoto--Sakaguchi coupling. We show that the chimera states may be\nmetastable or asymptotically stable. If the intra-star coupling strength is of\norder $\\varepsilon$, the chimera states persist on time scales at least of\norder $1/\\varepsilon$ in general, and on time-scales at least of order\n$1/\\varepsilon^2$ if the intra-star coupling is of Kuramoto--Sakaguchi type. If\nthe intra-star coupling configuration is sparse, the chimeras are\nasymptotically stable. The analysis relies on a combination of dimensional\nreduction using a M\\\"obius symmetry group and techniques from averaging theory\nand normal hyperbolicity.\n"}
{"abstract": "  The paper explains the method to determine a short unit cell attached to any\nhyperplane given by its integer vector $\\mathbf{p}$. Equivalently, it gives all\nthe solutions of the $N$-dimensional Bezout identity associated with the\ncoordinates of $\\mathbf{p}$.\n"}
{"abstract": "  We investigate two-sided bounds for operator norms of random matrices with\nunhomogenous independent entries. We formulate a lower bound for Rademacher\nmatrices and conjecture that it may be reversed up to a universal constant. We\nshow that our conjecture holds up to $\\log\\log n$ factor for randomized\n$n\\times n$ circulant matrices and double logarithm may be eliminated under\nsome mild additional assumptions on the coefficients.\n"}
{"abstract": "  Generative adversarial networks (GANs) can be trained to generate 3D image\ndata, which is useful for design optimisation. However, this conventionally\nrequires 3D training data, which is challenging to obtain. 2D imaging\ntechniques tend to be faster, higher resolution, better at phase identification\nand more widely available. Here, we introduce a generative adversarial network\narchitecture, SliceGAN, which is able to synthesise high fidelity 3D datasets\nusing a single representative 2D image. This is especially relevant for the\ntask of material microstructure generation, as a cross-sectional micrograph can\ncontain sufficient information to statistically reconstruct 3D samples. Our\narchitecture implements the concept of uniform information density, which both\nensures that generated volumes are equally high quality at all points in space,\nand that arbitrarily large volumes can be generated. SliceGAN has been\nsuccessfully trained on a diverse set of materials, demonstrating the\nwidespread applicability of this tool. The quality of generated micrographs is\nshown through a statistical comparison of synthetic and real datasets of a\nbattery electrode in terms of key microstructural metrics. Finally, we find\nthat the generation time for a $10^8$ voxel volume is on the order of a few\nseconds, yielding a path for future studies into high-throughput\nmicrostructural optimisation.\n"}
{"abstract": "  Displays are an indispensable medium to visually convey information in our\ndaily life. Although conventional dye-based color displays have been rigorously\nadvanced by world leading companies, critical issues still remain. For\ninstance, color fading and wavelength-limited resolution restrict further\ndevelopments. Plasmonic colors emerging from resonant interactions between\nlight and metallic nanostructures can overcome these restrictions. With dynamic\ncharacteristics enabled by functional materials, dynamic plasmonic coloration\nmay find a variety of applications in display technologies. In this review, we\nelucidate basic concepts for dynamic plasmonic color generation and highlight\nrecent advances. In particular, we devote our review to a selection of dynamic\ncontrols endowed by functional materials, including magnesium, liquid crystals,\nelectrochromic polymers, and phase change materials. We also discuss their\nperformance in view of potential applications in current display technologies.\n"}
{"abstract": "  Industrial wireless sensor networks are becoming crucial for modern\nmanufacturing. If the sensors in those networks are mobile, the position\ninformation, besides the sensor data itself, can be of high relevance. E.g.\nthis position information can increase the trustability of a wireless sensor\nmeasurement by assuring that the sensor is not physically removed, off track,\nor otherwise compromised.\n  In certain applications, localization information at cell-level, whether the\nsensor is inside or outside a room or cell, is sufficient. For this,\nlocalization using Received Signal Strength Indicator (RSSI) measurements is\nvery popular since RSSI values are available in almost all existing\ntechnologies and no direct interaction with the mobile sensor node and its\ncommunication in the network is needed. For this scenario, we propose methods\nto improve the robustness and accuracy of common machine learning classifiers,\nby using features based on short-term moments and a second classification stage\nusing Hidden Markov Models. With the data from an extensive measurement\ncampaign, we show the applicability of our method and achieve a cell-level\nlocalization accuracy of 93.5\\%.\n"}
{"abstract": "  Multiple object tracking is a challenging problem in computer vision due to\ndifficulty in dealing with motion prediction, occlusion handling, and object\nre-identification. Many recent algorithms use motion and appearance cues to\novercome these challenges. But using appearance cues increases the computation\ncost notably and therefore the speed of the algorithm decreases significantly\nwhich makes them inappropriate for online applications. In contrast, there are\nalgorithms that only use motion cues to increase speed, especially for online\napplications. But these algorithms cannot handle occlusions and re-identify\nlost objects. In this paper, a novel online multiple object tracking algorithm\nis presented that only uses geometric cues of objects to tackle the occlusion\nand reidentification challenges simultaneously. As a result, it decreases the\nidentity switch and fragmentation metrics. Experimental results show that the\nproposed algorithm could decrease identity switch by 40% and fragmentation by\n28% compared to the state of the art online tracking algorithms. The code is\nalso publicly available.\n"}
{"abstract": "  This paper proposes a new method to provide the exponential convergence of\nboth the parameter and tracking errors of the composite adaptive control system\nwithout the requirement of the regressor persistent excitation (PE). Instead,\nthe composite adaptation law obtained in this paper requires the regressor to\nbe finitely exciting (FE) to guarantee the above-mentioned properties. Unlike\nknown solutions, not only does it relax the PE requirement, but also it\nfunctions effectively under the condition of a jump change of the plant\nuncertainty parameters. To derive such an adaptation law, an integral filter of\nregressor with damping and resetting is proposed. It provides the required\nproperties of the control system, and its output signal is bounded even when\nits input is subjected to noise and disturbances. A rigorous analytical proof\nof all mentioned properties of the developed adaptation law is presented. Such\nlaw is compared with the known composite ones relaxing the PE requirement. The\nwing-rock problem is used for the modeling of the developed composite MRAC\nsystem. The obtained results fully support the theoretical analysis and\ndemonstrate the advantages of the proposed method.\n"}
{"abstract": "  Distributed training across several quantum computers could significantly\nimprove the training time and if we could share the learned model, not the\ndata, it could potentially improve the data privacy as the training would\nhappen where the data is located. However, to the best of our knowledge, no\nwork has been done in quantum machine learning (QML) in federation setting yet.\nIn this work, we present the federated training on hybrid quantum-classical\nmachine learning models although our framework could be generalized to pure\nquantum machine learning model. Specifically, we consider the quantum neural\nnetwork (QNN) coupled with classical pre-trained convolutional model. Our\ndistributed federated learning scheme demonstrated almost the same level of\ntrained model accuracies and yet significantly faster distributed training. It\ndemonstrates a promising future research direction for scaling and privacy\naspects.\n"}
{"abstract": "  We propose a gauged $U(1)_{B-L}$ extended two Higgs doublet model to explain\nboth neutrino mass and lepton anomalous magnetic moments ($g-2$). Neutrino mass\nis generated via an inverse seesaw mechanism by introducing singlet fermions.\nEspecially, we update the result of muon $g-2$ in light of the very recent\nreport by E989 experiment at Fermilab, indicating $a_\\mu^{\\rm\nFNAL}=116592040(54)\\times 10^{-11}$. Combining BNL result, we have the\nfollowing deviation from the standard model prediction $\\Delta a_\\mu=(2.51\\pm\n5.9)\\times10^{-10}$ at 4.2 $\\sigma$. Thanks to an appropriate assignment for\n$U(1)_{B-L}\\times Z_2$ symmetry and larger $(20\\lesssim)\\tan\\beta$ that is\nfavoured by type-X model, we realize natural hierarchies among neutral\nfermions. The lepton anomalous magnetic moments can be induced at the one loop\nlevel by introducing an iso-spin singlet singly-charged boson. This charged\nscalar plays a significant role in evading chiral suppression of these\nphenomenologies. We show sizable lepton ($g-2$) can be obtained after\nsatisfying all the flavour constraints, such as $\\mu\\to e\\gamma$ and flavour\nconserving leptonic $Z$ boson decays.\n"}
{"abstract": "  The widespread diffusion of electric mobility requires a contextual expansion\nof the charging infrastructure. An extended collection and processing of\ninformation regarding charging of electric vehicles may turn each electric\nvehicle charging station into a valuable source of streaming data. Charging\npoint operators may profit from all these data for optimizing their operation\nand planning activities. In such a scenario, big data and machine learning\ntechniques would allow valorizing real-time data coming from electric vehicle\ncharging stations. This paper presents an architecture able to deal with data\nstreams from a charging infrastructure, with the final aim to forecast electric\ncharging station availability after a set amount of minutes from present time.\nBoth batch data regarding past charges and real-time data streams are used to\ntrain a streaming logistic regression model, to take into account recurrent\npast situations and unexpected actual events. The streaming model performs\nbetter than a model trained only using historical data. The results highlight\nthe importance of constantly updating the predictive model parameters in order\nto adapt to changing conditions and always provide accurate forecasts.\n"}
{"abstract": "  We review various characterizations of uniform convexity and smoothness on\nnorm balls in finite-dimensional spaces and connect results stemming from the\ngeometry of Banach spaces with \\textit{scaling inequalities} used in analysing\nthe convergence of optimization methods. In particular, we establish local\nversions of these conditions to provide sharper insights on a recent body of\ncomplexity results in learning theory, online learning, or offline\noptimization, which rely on the strong convexity of the feasible set. While\nthey have a significant impact on complexity, these strong convexity or uniform\nconvexity properties of feasible sets are not exploited as thoroughly as their\nfunctional counterparts, and this work is an effort to correct this imbalance.\nWe conclude with some practical examples in optimization and machine learning\nwhere leveraging these conditions and localized assumptions lead to new\ncomplexity results.\n"}
{"abstract": "  Open-world novelty occurs when the rules of an environment can change\nabruptly, such as when a game player encounters \"house rules\". To address\nopen-world novelty, game playing agents must be able to detect when novelty is\ninjected, and to quickly adapt to the new rules. We propose a model-based\nreinforcement learning approach where game state and rules are represented as\nknowledge graphs. The knowledge graph representation of the state and rules\nallows novelty to be detected as changes in the knowledge graph, assists with\nthe training of deep reinforcement learners, and enables imagination-based\nre-training where the agent uses the knowledge graph to perform look-ahead.\n"}
{"abstract": "  Gravitational-wave (GW) catalogs are rapidly increasing in number, allowing\nfor robust statistical analyses of the population of compact binaries.\nNonetheless, GW inference of cosmology has typically relied on additional\nelectromagnetic counterparts or galaxy catalogs. I present a new probe of\ncosmological modifications of general relativity with GW data only. I focus on\ndeviations of the GW luminosity distance constrained with the astrophysical\npopulation of binary black holes (BBHs). The three key observables are 1) the\nnumber of events as a function of luminosity distance, 2) the stochastic GW\nbackground of unresolved binaries and 3) the location of any feature in the\nsource mass distribution, such as the pair instability supernova (PISN) gap.\nDespite a priori degeneracies between modified gravity and the unknown\nevolution of the merger rate and source masses, a large damping of the GW\namplitude could be falsifiable since as redshift grows it reduces the events\nand lowers the edges of the PISN gap, which is against standard astrophysical\nexpectations. Applying a hierarchical Bayesian analysis to the current\nLIGO--Virgo catalog (GWTC-2), the strongest constraints to date are placed on\ndeviations from the GW luminosity distance, finding $c_{_M}=-3.2^{+3.4}_{-2.0}$\nat $68\\%$ C.L., which is $\\sim10$ times better than multi-messenger GW170817\nbounds. These modifications also affects the determination of the BBH masses,\nwhich is crucial to accommodate the high-mass binary GW190521 away from the\nPISN gap. In this analysis it is found that the maximum mass of $99\\%$ of the\npopulation shifts to lower masses with increased uncertainty,\n$m_{99\\%}=46.2^{+11.4}_{-9.1}M_\\odot$ at $68\\%$ C.L. Testing gravity at large\nscales with the population of BBHs will become increasingly relevant with\nfuture catalogs, providing an independent and self-contained test of the\nstandard cosmological model.\n"}
{"abstract": "  The recent availability of low-power neural accelerator hardware, combined\nwith improvements in end-to-end neural facial recognition algorithms provides,\nenabling technology for on-device facial authentication. The present research\nwork examines the effects of directional lighting on a State-of-Art(SoA) neural\nface recognizer. A synthetic re-lighting technique is used to augment data\nsamples due to the lack of public data-sets with sufficient directional\nlighting variations. Top lighting and its variants (top-left, top-right) are\nfound to have minimal effect on accuracy, while bottom-left or bottom-right\ndirectional lighting has the most pronounced effects. Following the fine-tuning\nof network weights, the face recognition model is shown to achieve close to the\noriginal Receiver Operating Characteristic curve (ROC)performance across all\nlighting conditions and demonstrates an ability to generalize beyond the\nlighting augmentations used in the fine-tuning data-set. This work shows that\nan SoA neural face recognition model can be tuned to compensate for directional\nlighting effects, removing the need for a pre-processing step before applying\nfacial recognition.\n"}
{"abstract": "  In this work, we have adopted gravitational decoupling by Minimal Geometric\nDeformation (MGD) approach and have developed an anisotropic version of\nwell-known Tolman VII isotropic solution in the framework of $f(R, T)$ gravity,\nwhere $R$ is Ricci scalar and $T$ is trace of energy momentum tensor. The set\nof field equations has been developed with respect to total energy momentum\ntensor, which combines effective energy momentum tensor in $f(R,T)$ gravity and\nadditional source $\\phi_{ij}$. Following MGD approach, the set of field\nequations has been separated into two sections. One section represents $f(R,\nT)$ field equations, while the other is related to the source $\\phi_{ij}$. The\nmatching conditions for inner and outer geometry have also been discussed and\nan anisotropic solution has been developed using mimic constraint for radial\npressure. In order to check viability of the solution, we have considered\nobservation data of three different compact star models, named PSR J1614-2230,\nPSR 1937+21 and SAX J1808.4-3658 and have discussed thermodynamical properties\nanalytically and graphically.\n  The energy conditions are found to be satisfied for the three compact stars.\nThe stability analysis has been presented through causality condition and\nHerrera's cracking concept, which ensures physical acceptability of the\nsolution.\n"}
{"abstract": "  Artificial Intelligence (AI) can augment and sometimes even replace human\ncognition. Inspired by efforts to value human agency alongside productivity, we\ndiscuss the benefits of solving Citizen Science (CS) tasks with Hybrid\nIntelligence (HI), a synergetic mixture of human and artificial intelligence.\nCurrently there is no clear framework or methodology on how to create such an\neffective mixture. Due to the unique participant-centered set of values and the\nabundance of tasks drawing upon both human common sense and complex 21st\ncentury skills, we believe that the field of CS offers an invaluable testbed\nfor the development of HI and human-centered AI of the 21st century, while\nbenefiting CS as well. In order to investigate this potential, we first relate\nCS to adjacent computational disciplines. Then, we demonstrate that CS projects\ncan be grouped according to their potential for HI-enhancement by examining two\nkey dimensions: the level of digitization and the amount of knowledge or\nexperience required for participation. Finally, we propose a framework for\ntypes of human-AI interaction in CS based on established criteria of HI. This\n\"HI lens\" provides the CS community with an overview of several ways to utilize\nthe combination of AI and human intelligence in their projects. It also allows\nthe AI community to gain ideas on how developing AI in CS projects can further\ntheir own field.\n"}
{"abstract": "  The goal of this paper is to present a methodology for the computation of\ninvariant tori in Hamiltonian systems combining flow map methods,\nparameterization methods, and symplectic geometry. While flow map methods\nreduce the dimension of the tori to be computed by one (avoiding Poincare\nmaps), parameterization methods reduce the cost of a single step of the derived\nNewton-like method to be proportional to the cost of a FFT. Symplectic\nproperties lead to some magic cancellations that make the methods work. The\nmultiple shooting version of the methods are applied to the computation of\ninvariant tori and their invariant bundles around librational equilibrium\npoints of the Restricted Three Body Problem. The invariant bundles are the\nfirst order approximations of the corresponding invariant manifolds, commonly\nknown as the whiskers, which are very important in the dynamical organization\nand have important applications in space mission design.\n"}
{"abstract": "  Trusted Execution Environments (TEEs) are used to protect sensitive data and\nrun secure execution for security-critical applications, by providing an\nenvironment isolated from the rest of the system. However, over the last few\nyears, TEEs have been proven weak, as either TEEs built upon security-oriented\nhardware extensions (e.g., Arm TrustZone) or resorting to dedicated secure\nelements were exploited multiple times. In this project, we introduce Trusted\nExecution Environments On-Demand (TEEOD), a novel TEE design that leverages the\nprogrammable logic (PL) in the heterogeneous system on chips (SoC) as the\nsecure execution environment. Unlike other TEE designs, TEEOD can provide\nhigh-bandwidth connections and physical on-chip isolation. We implemented a\nproof-of-concept (PoC) implementation targeting an Ultra96-V2 platform. The\nconducted evaluation demonstrated TEEOD can host up to 6 simultaneous enclaves\nwith a resource usage per enclave of 7.0%, 3.8%, and 15.3% of the total LUTs,\nFFs, and BRAMS, respectively. To demonstrate the practicability of TEEOD in\nreal-world applications, we successfully run a legacy open-source Bitcoin\nwallet.\n"}
{"abstract": "  In this paper, we investigate the worst-case robust beamforming design and\nresource block (RB) assignment problem for total transmit power minimization of\nthe central controller while guaranteeing each robot's transmission with target\nnumber of data bits and within required ultra-low latency and extremely high\nreliability. By using the property of the independence of each robot's\nbeamformer design, we can obtain the equivalent power control design form of\nthe original beamforming design. The binary RB mapping indicators are\ntransformed into continuous ones with additional $\\ell_0$-norm constraints to\npromote sparsity on each RB. A novel non-convex penalty (NCP) approach is\napplied to solve such $\\ell_0$-norm constraints. Numerical results demonstrate\nthe superiority of the NCP approach to the well-known reweighted $\\ell_1$\nmethod in terms of the optimized power consumption, convergence rate and\nrobustness to channel realizations. Also, the impacts of latency, reliability,\nnumber of transmit antennas and channel uncertainty on the system performance\nare revealed.\n"}
{"abstract": "  Unsupervised anomaly detection could be crucial in future analyses searching\nfor rare phenomena in large datasets, as for example collected at the LHC. To\nthis end, we introduce a physics inspired variational autoencoder (VAE)\narchitecture which performs competitively and robustly on the LHC Olympics\nMachine Learning Challenge datasets. We demonstrate how embedding some physical\nobservables directly into the VAE latent space, while at the same time keeping\nthe classifier manifestly agnostic to them, can help to identify and\ncharacterise features in measured spectra as caused by the presence of\nanomalies in a dataset.\n"}
{"abstract": "  We study the fundamental differences that separate: Litecoin; Bitcoin Gold;\nBitcoin Cash; Ethereum; and Zcash from Bitcoin, and draw analysis to how these\nfeatures are appreciated by the market, to ultimately make an inference as to\nhow future successful cryptocurrencies may behave. We use Google Trend data, as\nwell as price, volume and market capitalization data sourced from\ncoinmarketcap.com to support this analysis. We find that Litecoin's shorter\nblock times offer benefits in commerce, but drawbacks in the mining process\nthrough orphaned blocks. Zcash holds a niche use for anonymous transactions,\nbenefitting areas of the world lacking in economic freedom. Bitcoin Cash\nsuffers from centralization in the mining process, while the greater\ndecentralization of Bitcoin Gold has generally left it to stagnate. Ether's\ngreater functionality offers the greatest threat to Bitcoin's dominance in the\nmarket. A coin that incorporates several of these features can be technically\nbetter than Bitcoin, but the first-to-marketadvantage of Bitcoin should keep\nits dominant position in the market.\n"}
{"abstract": "  Theories with massive gravitons have peculiarity called the van\nDam-Veltman-Zakharov discontinuity in that the massive theory propagator does\nnot go to the massless graviton propagator in the zero graviton mass limit.\nThis results in large deviation in Newtons law for massive graviton theories\neven when the graviton mass vanishes. We test the vDVZ in massive graviton\ntheories for single graviton vertex process namely the gravitational radiation\nfrom a classical source. We calculate the gravitational radiation from compact\nbinaries using the perturbative Feynman diagram method. We perform this\ncalculation for Einstein's gravity with massless gravitons and verify that the\nFeynman diagram calculation reproduces the quadrupole formula. Using the same\nprocedure we calculate the gravitational radiation for three massive graviton\ntheories: (1) the Fierz-Pauli theory (2) the modified Fierz-Pauli theory\nwithout the vDVZ discontinuity and (3) the Dvali-Gabadadze-Porrati theory with\na momentum dependent graviton mass. We put limits on the graviton mass in each\nof these theories from observations of binary pulsar timings.\n"}
{"abstract": "  Nowadays, researchers have moved to platforms like Twitter to spread\ninformation about their ideas and empirical evidence. Recent studies have shown\nthat social media affects the scientific impact of a paper. However, these\nstudies only utilize the tweet counts to represent Twitter activity. In this\npaper, we propose TweetPap, a large-scale dataset that introduces temporal\ninformation of citation/tweets and the metadata of the tweets to quantify and\nunderstand the discourse of scientific papers on social media. The dataset is\npublicly available at https://github.com/lingo-iitgn/TweetPap\n"}
{"abstract": "  We study Landau gauge gluon propagators in two-color QCD at finite quark\nchemical potential ($\\mu_q$) and temperature ($T$). We include medium\npolarization effects at one-loop by quarks into massive gluon propagators, and\ncompared the analytic results with the available lattice data. We particularly\nfocus on the high density phase of color-singlet diquark condensates whose\ncritical temperature is $\\sim 100$ MeV with weak dependence on $\\mu_q$. At zero\ntemperature the color singlet condensates protect the IR limit of electric and\nmagnetic gluon propagators from the medium screening effects. At finite\ntemperature, this behavior remains true for the magnetic sector, but the\nelectric screening mass should be generated by thermal, and hence gapless,\nparticles which are unbound from the diquark condensates. Treating thermal\nexcitations as quasi-quarks, we found that the electric screening develops too\nfast compared to the lattice results. Beyond the critical temperature for\ndiquark condensates the analytic results are consistent with the lattice\nresults.\n"}
{"abstract": "  We have obtained the first constraint of the parameter space of\nScalar-Tensor-Vector-Gravity using the motion of the S2-star around the\nsupermassive black hole at the centre of the Milky Way, and we did not find any\nserious tension with General Relativity. We used the Schwarzschild-like metric\nof Scalar-Tensor-Vector-Gravity to predict the orbital motion of S2-star, and\nto compare it with the publicly available astrometric data, which include 145\nmeasurements of the positions, 44 measurements of the radial velocities of\nS2-star along its orbit, and the recent measurement of the orbital precession.\nWe employed a Monte Carlo Markov Chain algorithm to explore the parameter\nspace, and constrained the only one additional parameter of\nScalar-Tensor-Vector-Gravity to $\\alpha \\lesssim 0.410$ at 99,7$\\%$ confidence\nlevel, where $\\alpha=0$ reduces this modified theory of gravity to General\nRelativity.\n"}
{"abstract": "  Integrated photonics has enabled signal synthesis, modulation and conversion\nusing photonic integrated circuits (PIC). Many materials have been developed,\namong which silicon nitride (Si$_3$N$_4$) has emerged as a leading platform\nparticularly for nonlinear photonics. Low-loss Si$_3$N$_4$ PIC has been widely\nused for frequency comb generation, narrow-linewidth lasers, microwave\nphotonics, photonic computing networks, and even surface-electrode ion traps.\nYet, among all demonstrated functionalities for Si$_3$N$_4$ integrated\nphotonics, optical non-reciprocal devices, such as isolators and circulators,\nhave not been achieved. Conventionally, they are realized based on Faraday\neffect of magneto-optic materials under external magnetic field. However, it\nhas been challenging to integrate magneto-optic materials that are not\nCMOS-compatible and that require bulky external magnet. Here, we demonstrate a\nmagnetic-free optical isolator based on aluminum nitride (AlN) piezoelectric\nmodulators monolithically integrated on ultralow-loss Si$_3$N$_4$ PIC. The\ntransmission reciprocity is broken by spatio-temporal modulation of a\nSi$_3$N$_4$ microring resonator with three AlN bulk acoustic wave resonators\nthat are driven with a rotational phase. This design creates an effective\nrotating acoustic wave that allows indirect interband transition in only one\ndirection among a pair of strongly coupled optical modes. Maximum of 10 dB\nisolation is achieved under 100 mW RF power applied to each actuator, with\nminimum insertion loss of 0.1 dB. The isolation remains constant over nearly 30\ndB dynamic range of optical input power, showing excellent optical linearity.\nOur integrated, linear, magnetic-free, electrically driven optical isolator\ncould become key building blocks for integrated lasers, chip-scale LiDAR\nengines, as well as optical interfaces for superconducting circuits.\n"}
{"abstract": "  Automated construction of surface geometries of cardiac structures from\nvolumetric medical images is important for a number of clinical applications.\nWhile deep-learning-based approaches have demonstrated promising reconstruction\nprecision, these approaches have mostly focused on voxel-wise segmentation\nfollowed by surface reconstruction and post-processing techniques. However,\nsuch approaches suffer from a number of limitations including disconnected\nregions or incorrect surface topology due to erroneous segmentation and\nstair-case artifacts due to limited segmentation resolution. We propose a novel\ndeep-learning-based approach that directly predicts whole heart surface meshes\nfrom volumetric CT and MR image data. Our approach leverages a graph\nconvolutional neural network to predict deformation on mesh vertices from a\npre-defined mesh template to reconstruct multiple anatomical structures in a 3D\nimage volume. Our method demonstrated promising performance of generating whole\nheart reconstructions with as good or better accuracy than prior\ndeep-learning-based methods on both CT and MR data. Furthermore, by deforming a\ntemplate mesh, our method can generate whole heart geometries with better\nanatomical consistency and produce high-resolution geometries from lower\nresolution input image data. Our method was also able to produce temporally\nconsistent surface mesh predictions for heart motion from CT or MR cine\nsequences, and therefore can potentially be applied for efficiently\nconstructing 4D whole heart dynamics. Our code and pre-trained networks are\navailable at https://github.com/fkong7/MeshDeformNet\n"}
{"abstract": "  Extreme multi-label classification (XMC) aims to learn a model that can tag\ndata points with a subset of relevant labels from an extremely large label set.\nReal world e-commerce applications like personalized recommendations and\nproduct advertising can be formulated as XMC problems, where the objective is\nto predict for a user a small subset of items from a catalog of several million\nproducts. For such applications, a common approach is to organize these labels\ninto a tree, enabling training and inference times that are logarithmic in the\nnumber of labels. While training a model once a label tree is available is well\nstudied, designing the structure of the tree is a difficult task that is not\nyet well understood, and can dramatically impact both model latency and\nstatistical performance. Existing approaches to tree construction fall at an\nextreme point, either optimizing exclusively for statistical performance, or\nfor latency. We propose an efficient information theory inspired algorithm to\nconstruct intermediary operating points that trade off between the benefits of\nboth. Our algorithm enables interpolation between these objectives, which was\nnot previously possible. We corroborate our theoretical analysis with numerical\nresults, showing that on the Wiki-500K benchmark dataset our method can reduce\na proxy for expected latency by up to 28% while maintaining the same accuracy\nas Parabel. On several datasets derived from e-commerce customer logs, our\nmodified label tree is able to improve this expected latency metric by up to\n20% while maintaining the same accuracy. Finally, we discuss challenges in\nrealizing these latency improvements in deployed models.\n"}
{"abstract": "  The widespread deployment of surveillance cameras for facial recognition\ngives rise to many privacy concerns. This study proposes a privacy-friendly\nalternative to large scale facial recognition. While there are multiple\ntechniques to preserve privacy, our work is based on the minimization principle\nwhich implies minimizing the amount of collected personal data. Instead of\nrunning facial recognition software on all video data, we propose to\nautomatically extract a high quality snapshot of each detected person without\nrevealing his or her identity. This snapshot is then encrypted and access is\nonly granted after legal authorization. We introduce a novel unsupervised face\nimage quality assessment method which is used to select the high quality\nsnapshots. For this, we train a variational autoencoder on high quality face\nimages from a publicly available dataset and use the reconstruction probability\nas a metric to estimate the quality of each face crop. We experimentally\nconfirm that the reconstruction probability can be used as biometric quality\npredictor. Unlike most previous studies, we do not rely on a manually defined\nface quality metric as everything is learned from data. Our face quality\nassessment method outperforms supervised, unsupervised and general image\nquality assessment methods on the task of improving face verification\nperformance by rejecting low quality images. The effectiveness of the whole\nsystem is validated qualitatively on still images and videos.\n"}
{"abstract": "  Self-triggered control (STC) is a sample-and-hold control method aimed at\nreducing communications within networked-control systems; however, existing STC\nmechanisms often maximize how late the next sample is, and as such they do not\nprovide any sampling optimality in the long-term. In this work, we devise a\nmethod to construct self-triggered policies that provide near-maximal average\ninter-sample time (AIST) while respecting given control performance\nconstraints. To achieve this, we rely on finite-state abstractions of a\nreference event-triggered control, in which early triggers are also allowed.\nThese early triggers constitute controllable actions of the abstraction, for\nwhich an AIST-maximizing strategy can be computed by solving a mean-payoff\ngame. We provide optimality bounds, and how to further improve them through\nabstraction refinement techniques.\n"}
{"abstract": "  Adaptive gradient methods, such as Adam, have achieved tremendous success in\nmachine learning. Scaling gradients by square roots of the running averages of\nsquared past gradients, such methods are able to attain rapid training of\nmodern deep neural networks. Nevertheless, they are observed to generalize\nworse than stochastic gradient descent (SGD) and tend to be trapped in local\nminima at an early stage during training. Intriguingly, we discover that\nsubstituting the gradient in the second moment estimation term with the\nmomentumized version in Adam can well solve the issues. The intuition is that\ngradient with momentum contains more accurate directional information and\ntherefore its second moment estimation is a better choice for scaling than that\nof the raw gradient. Thereby we propose AdaMomentum as a new optimizer reaching\nthe goal of training fast while generalizing better. We further develop a\ntheory to back up the improvement in optimization and generalization and\nprovide convergence guarantees under both convex and nonconvex settings.\nExtensive experiments on a wide range of tasks and models demonstrate that\nAdaMomentum exhibits state-of-the-art performance consistently.\n"}
{"abstract": "  Modern models for event causality identification (ECI) are mainly based on\nsupervised learning, which are prone to the data lacking problem.\nUnfortunately, the existing NLP-related augmentation methods cannot directly\nproduce the available data required for this task. To solve the data lacking\nproblem, we introduce a new approach to augment training data for event\ncausality identification, by iteratively generating new examples and\nclassifying event causality in a dual learning framework. On the one hand, our\napproach is knowledge-guided, which can leverage existing knowledge bases to\ngenerate well-formed new sentences. On the other hand, our approach employs a\ndual mechanism, which is a learnable augmentation framework and can\ninteractively adjust the generation process to generate task-related sentences.\nExperimental results on two benchmarks EventStoryLine and Causal-TimeBank show\nthat 1) our method can augment suitable task-related training data for ECI; 2)\nour method outperforms previous methods on EventStoryLine and Causal-TimeBank\n(+2.5 and +2.1 points on F1 value respectively).\n"}
{"abstract": "  We study the black hole information problem within a semiclassically\ngravitating AdS$_d$ black hole coupled to and in equilibrium with a\n$d$-dimensional thermal conformal bath. We deform the bath state by a relevant\nscalar deformation, triggering a holographic RG flow whose \"trans-IR\" region\ndeforms from a Schwarzschild geometry to a Kasner universe. The setup manifests\ntwo independent scales which control both the extent of coarse-graining and the\nentanglement dynamics when counting Hawking degrees of freedom in the bath. In\ntuning either, we find nontrivial changes to the Page time and Page curve. We\nconsequently view the Page curve as a probe of the holographic RG flow, with a\nhigher Page time manifesting as a result of increased coarse-graining of the\nbath degrees of freedom.\n"}
{"abstract": "  A longstanding question in QCD is the origin of the mass gap in the\nYang-Mills sector of QCD, i.e., QCD without quarks. In Landau gauge QCD this\nmass gap, and hence confinement, is encoded in a mass gap of the gluon\npropagator, which is found both in lattice simulations and with functional\napproaches. While functional methods are well suited to unravel the mechanism\nbehind the generation of the mass gap, a fully satisfactory answer has not yet\nbeen found. In this work we solve the coupled Dyson-Schwinger equations for the\nghost propagator, gluon propagator and three-gluon vertex. We corroborate the\nfindings of earlier works, namely that the mass gap generation is tied to the\nlongitudinal projection of the gluon self-energy, which acts as an effective\nmass term in the equations. Because an explicit mass term is in conflict with\ngauge invariance, this leaves two possible scenarios: If it is viewed as an\nartifact, only the scaling solution survives; if it is dynamical, gauge\ninvariance can only be preserved if there are longitudinal massless poles in\neither of the vertices. We find that there is indeed a massless pole in the\nghost-gluon vertex, however in our approximation with the assumption of\ncomplete infrared dominance of the ghost this pole is only present for the\nscaling solution. We also put forward a possible mechanism that may reconcile\nthe scaling solution, with an infrared dominance of the ghost, with the\ndecoupling solutions based on longitudinal poles in the three-gluon vertex as\nseen in the PT-BFM scheme.\n"}
{"abstract": "  This article proves that Tate classes on Siegel modular 3-folds are spanned\nby the images of Hilbert modular surfaces at degree 2 and by the images of\nShimura curves at degree 4. The proof involves a careful study of the period\npairing between degree-4 rapidly decreasing cohomological forms and Hilbert\nmodular surfaces.\n"}
{"abstract": "  This letter investigates a Branching Dueling Q-Network (BDQ) based online\noperation strategy for a microgrid with distributed battery energy storage\nsystems (BESSs) operating under uncertainties. The developed deep reinforcement\nlearning (DRL) based microgrid online optimization strategy can achieve a\nlinear increase in the number of neural network outputs with the number of\ndistributed BESSs, which overcomes the curse of dimensionality caused by the\ncharge and discharge decisions of multiple BESSs. Numerical simulations\nvalidate the effectiveness of the proposed method.\n"}
{"abstract": "  Understanding cell fate selection remains a central challenge in\ndevelopmental biology. We present a class of simple yet biologically-motivated\nmathematical models for cell differentiation that generically generate\noscillations and hence suggest alternatives to the standard framework based on\nWaddington's epigenetic landscape. The models allow us to suggest two generic\ndynamical scenarios that describe the differentiation process. In the first\nscenario gradual variation of a single control parameter is responsible for\nboth entering and exiting the oscillatory regime. In the second scenario two\ncontrol parameters vary: one responsible for entering, and the other for\nexiting the oscillatory regime. We analyse the standard repressilator and four\nvariants of it and show the dynamical behaviours associated with each scenario.\nWe present a thorough analysis of the associated bifurcations and argue that\ngene regulatory networks with these repressilator-like characteristics are\npromising candidates to describe cell fate selection through an oscillatory\nprocess.\n"}
{"abstract": "  The Inverse Problem for the estimation of a point-wise approximation error\noccurring at the discretization and solving of the system of partial\ndifferential equations is addressed. The set of the differences between the\nnumerical solutions is used as the input data. The analyzed solutions are\nobtained by the numerical algorithms of the distinct inner structure on the\nsame grid. The approximation error is estimated by the Inverse Problem, which\nis posed in the variational statement with the zero order Tikhonov\nregularization. The numerical tests, performed for the two dimensional inviscid\ncompressible flows corresponding to Edney-I and Edney-VI shock wave\ninterference modes, are provided. The analyzed flowfields are computed using\nten different numerical algorithms. The comparison of the estimated\napproximation error and the true error, obtained by subtraction of numerical\nand analytic solutions, is presented.\n"}
{"abstract": "  We orchestrate a randomized algorithm for #$k$-SAT which counts the exact\nnumber of satisfying assignments in $2^{o(n)}$ time. The existence of such\nalgorithm signifies that the #ETH is hereby refuted, and so are $\\oplus$ETH,\nETH, #SETH, $\\oplus$SETH and SETH.\n"}
{"abstract": "  Manipulating the strain induced poly-domains and phase transition in\ncorrelated oxide material are important for high performance devices\nfabrication. Though the electronic transport in the strained oxide film at\nmacroscopic scales can be directly measured, the anisotropic electronic state\nand the controllable phase separation cross the insulator-to-metal transition\nwithin nanoscale size are still elusive. Here, we selected VO2 crystal film as\na prototypical oxide and achieved the manipulation of anisotropy electronic\nphase separation via injecting He+ nanobeam into VO2 film at room temperature.\nIn addition, this nanoscale phase separation was directly visualized by\ninfrared near-field imaging measurements, showing the pronounced and unique\ncR-axis dependent anisotropy on VO2 surface. Our results offered new insights\ntowards understanding the anisotropic nanoscale phase separation in strained\nmetal oxide films.\n"}
{"abstract": "  In this work, we propose \\texttt{TimeGrad}, an autoregressive model for\nmultivariate probabilistic time series forecasting which samples from the data\ndistribution at each time step by estimating its gradient. To this end, we use\ndiffusion probabilistic models, a class of latent variable models closely\nconnected to score matching and energy-based methods. Our model learns\ngradients by optimizing a variational bound on the data likelihood and at\ninference time converts white noise into a sample of the distribution of\ninterest through a Markov chain using Langevin sampling. We demonstrate\nexperimentally that the proposed autoregressive denoising diffusion model is\nthe new state-of-the-art multivariate probabilistic forecasting method on\nreal-world data sets with thousands of correlated dimensions. We hope that this\nmethod is a useful tool for practitioners and lays the foundation for future\nresearch in this area.\n"}
{"abstract": "  In this paper, we quantify the rate of convergence between the distribution\nof number of zeros of random trigonometric polynomials (RTP) with i.i.d.\ncentered random coefficients and the number of zeros of a stationary centered\nGaussian process G, whose covariance function is given by the sinc function.\nFirst, we find the convergence of the RTP towards G in the Wasserstein-1\ndistance, which in turn is a consequence of Donsker Theorem. Then, we use this\nresult to derive the rate of convergence between their respective number of\nzeros. Since the number of real zeros of the RTP is not a continuous function,\nwe use the Kac-Rice formula to express it as the limit of an integral and, in\nthis way, we approximate it by locally Lipschitz continuous functions.\n"}
{"abstract": "  Ceramic $BiFeO_{3}$ samples were prepared by rapid sintering at $880^OC$. Two\ncompositions were examined. A $56/44\\ Bi_{2}O_{3}/Fe_{2}O_{3}\\ mole\\%$\ncomposition and a $56Bi_{2}O_{3}\\cdot44Fe_{2}O_{3}+6.5wt\\%\\,NaCl$ composition.\nThe samples were heat treated at different times up to $8$ minutes and the\nphase content was examined as a function of the time using XRD measurements and\nanalysis. It was demonstrated that using both compositions, maximum $BiFeO_{3}$\nphase content is obtained after $3.5$ minutes. In the former approximately\n$50\\%$ of the material transformed to $BiFeO_{3}$ while in the latter $98.5\\%$.\n"}
{"abstract": "  The residual finite-dimensionality of a $\\mathrm{C}^*$-algebra is known to be\nencoded in a topological property of its space of representations, stating that\nfinite-dimensional representations should be dense therein. We extend this\nparadigm to general (possibly non-self-adjoint) operator algebras. While\nnumerous subtleties emerge in this greater generality, we exhibit novel tools\nfor constructing finite-dimensional approximations. One such tool is a notion\nof a residually finite-dimensional coaction of a semigroup on an operator\nalgebra, which allows us to construct finite-dimensional approximations for\noperator algebras of functions and operator algebras of semigroups. Our\ninvestigation is intimately related to the question of whether residual\nfinite-dimensionality of an operator algebra is inherited by its maximal\n$\\mathrm{C}^*$-cover, which we resolve in many cases of interest.\n"}
{"abstract": "  We propose a measure, which we call the dissipative spectral form factor\n(DSFF), to characterize the spectral statistics of non-Hermitian (and\nnon-Unitary) matrices. We show that DSFF successfully diagnoses dissipative\nquantum chaos, and reveals correlations between real and imaginary parts of the\ncomplex eigenvalues up to arbitrary energy (and time) scale. Specifically, we\nprovide the exact solution of DSFF for the GinUE and for a Poissonian random\nspectrum (Poisson) as minimal models of dissipative quantum chaotic and\nintegrable systems respectively. For dissipative quantum chaotic systems, we\nshow that DSFF exhibits an exact rotational symmetry in its complex time\nargument $\\tau$. Analogous to the spectral form factor (SFF) behaviour for GUE,\nDSFF for GinUE shows a ``dip-ramp-plateau'' behavior in $|\\tau|$: DSFF\ninitially decreases, increases at intermediate time scales, and saturates after\na generalized Heisenberg time which scales as the inverse mean level spacing.\nRemarkably, for large matrix size, the ``ramp'' of DSFF for GinUE increases\nquadratically in $|\\tau|$, in contrast to the linear ramp in SFF for Hermitian\nensembles. For dissipative quantum integrable systems, we show that DSFF takes\na constant value except for a region in complex time whose size and behavior\ndepends on the eigenvalue density. Numerically, we verify the above claims and\nshow that DSFF for real and quaternion real Ginibre ensembles coincides with\nthe GinUE behaviour except for a region in complex time plane of measure zero\nin the limit of large matrix size. As a physical example, we consider the\nquantum kicked top model with dissipation, and show that it falls under the\nGinibre universality class and Poisson as the `kick' is switched on or off.\nLastly, we study spectral statistics of ensembles of random classical\nstochastic matrices, and show that these models fall under the Ginibre\nuniversality class.\n"}
{"abstract": "  Analysis of online reviews has attracted great attention with broad\napplications. Often times, the textual reviews are coupled with the numerical\nratings in the data. In this work, we propose a probabilistic model to\naccommodate both textual reviews and overall ratings with consideration of\ntheir intrinsic connection for a joint sentiment-topic prediction. The key of\nthe proposed method is to develop a unified generative model where the topic\nmodeling is constructed based on review texts and the sentiment prediction is\nobtained by combining review texts and overall ratings. The inference of model\nparameters are obtained by an efficient Gibbs sampling procedure. The proposed\nmethod can enhance the prediction accuracy of review data and achieve an\neffective detection of interpretable topics and sentiments. The merits of the\nproposed method are elaborated by the case study from Amazon datasets and\nsimulation studies.\n"}
{"abstract": "  We use the energy method to study the well-posedness of initial-boundary\nvalue problems approximated by overset mesh methods in one and two space\ndimensions for linear constant-coefficient hyperbolic systems. We show that in\none space dimension, for both scalar equations and systems of equations, the\nproblem where one domain partially oversets another is well-posed when\ncharacteristic coupling conditions are used. If a system cannot be\ndiagonalized, as is ususally the case in multiple space dimensions, then the\nenergy method does not give proper bounds in terms of initial and boundary\ndata. For those problems, we propose a novel penalty approach. We show, by\nusing a global energy that accounts for the energy in the overlap region of the\ndomains, that under well-defined conditions on the coupling matrices the\npenalized overset domain problems are energy bounded, conservative, well-posed\nand have solutions equivalent to the original single domain problem.\n"}
{"abstract": "  Planning future operational scenarios of bulk power systems that meet\nsecurity and economic constraints typically requires intensive labor efforts in\nperforming massive simulations. To automate this process and relieve engineers'\nburden, a novel multi-stage control approach is presented in this paper to\ntrain centralized and decentralized reinforcement learning agents that can\nautomatically adjust grid controllers for regulating transmission line flows at\nnormal condition and under contingencies. The power grid flow control problem\nis formulated as Markov Decision Process (MDP). At stage one, centralized soft\nactor-critic (SAC) agent is trained to control generator active power outputs\nin a wide area to control transmission line flows against specified security\nlimits. If line overloading issues remain unresolved, stage two is used to\ntrain decentralized SAC agent via load throw-over at local substations. The\neffectiveness of the proposed approach is verified on a series of actual\nplanning cases used for operating the power grid of SGCC Zhejiang Electric\nPower Company.\n"}
{"abstract": "  In the article, hyperelastic material models which state consistent\npolynomial expansions of the stored energy function are discussed. The approach\nfollows from the muliplicative decomposition of the deformation gradient. Some\nadvantages of the third order expansion model over the five-parameter Rivlin\nmodel using Treloar's experimental data are shown. The models are qualitatively\nand quantitatively compared to highlight these advantages of the discussed MV\nmodel.\n"}
{"abstract": "  We investigate superconducting phase transition in\nsuperconductor(S)/ferromagnet(F) bilayer with Rasba spin-orbit interaction at\nS/F interface. This spin-orbit coupling produces spontaneous supercurrents\nflowing inside the atomic-thickness region near the interface, which are\ncompensated by the screening Meissner currents [S. Mironov and A. Buzdin, Phys.\nRev. Lett \\textbf{118}, 077001 (2017)]. In the case of thin superconducting\nfilm the emergence of the spontaneous surface currents causes the increase of\nthe superconducting critical temperature and we calculate the actual value of\nthe critical temperature shift. We also show that in the case of type-I\nsuperconducting film this phase transition can be of the first order. In the\nexternal magnetic field the critical temperature depends on the relative\norientation of the external magnetic field and the exchange field in the\nferromagnet. Also we predict the in-plane anisotropy of the critical current\nwhich may open an alternative way for the experimental observation of the\nspontaneous supercurrents generated by the SOC.\n"}
{"abstract": "  Warm magnets are magnets that function in normal ambient temperature\nconditions. These types are mostly using a soft steel yoke for field\namplification and either Copper or Aluminium coils or permanent magnets to\ngenerate the field. Magnets powered with such normal-conducting coils are often\nalso called classical, iron dominated or resistive magnets. Since decades these\ntypes of magnets are the workhorse for most linear and circular accelerators\nand beam transfer lines.\n"}
{"abstract": "  One of the main goals of online competitive games is increasing player\nengagement by ensuring fair matches. These games use rating systems for\ncreating balanced match-ups. Rating systems leverage statistical estimation to\nrate players' skills and use skill ratings to predict rank before matching\nplayers. Skill ratings of individual players can be aggregated to compute the\nskill level of a team. While research often aims to improve the accuracy of\nskill estimation and fairness of match-ups, less attention has been given to\nhow the skill level of a team is calculated from the skill level of its\nmembers. In this paper, we propose two new aggregation methods and compare them\nwith a standard approach extensively used in the research literature. We\npresent an exhaustive analysis of the impact of these methods on the predictive\nperformance of rating systems. We perform our experiments using three popular\nrating systems, Elo, Glicko, and TrueSkill, on three real-world datasets\nincluding over 100,000 battle royale and head-to-head matches. Our evaluations\nshow the superiority of the MAX method over the other two methods in the\nmajority of the tested cases, implying that the overall performance of a team\nis best determined by the performance of its most skilled member. The results\nof this study highlight the necessity of devising more elaborated methods for\ncalculating a team's performance -- methods covering different aspects of\nplayers' behavior such as skills, strategy, or goals.\n"}
{"abstract": "  A novel non-parametric quantile estimation method for continuous random\nvariables is introduced, based on a minimal neural network architecture\nconsisting of a single unit. Its advantage over estimations from ranking the\norder statistics is shown, specifically for small sample size. In a regression\ncontext, the method can be used to quantify predictive uncertainty under the\nsplit conformal prediction setting, where prediction intervals are estimated\nfrom the residuals of a pre-trained model on a held-out validation set to\nquantify the uncertainty in future predictions. Benchmarking experiments\ndemonstrate that the method is competitive in quality and coverage with\nstate-of-the-art solutions, with the added benefit of being more\ncomputationally efficient.\n"}
{"abstract": "  Cross-modal person re-identification (Re-ID) is critical for modern video\nsurveillance systems. The key challenge is to align inter-modality\nrepresentations according to semantic information present for a person and\nignore background information. In this work, we present AXM-Net, a novel CNN\nbased architecture designed for learning semantically aligned visual and\ntextual representations. The underlying building block consists of multiple\nstreams of feature maps coming from visual and textual modalities and a novel\nlearnable context sharing semantic alignment network. We also propose\ncomplementary intra modal attention learning mechanisms to focus on more\nfine-grained local details in the features along with a cross-modal affinity\nloss for robust feature matching. Our design is unique in its ability to\nimplicitly learn feature alignments from data. The entire AXM-Net can be\ntrained in an end-to-end manner. We report results on both person search and\ncross-modal Re-ID tasks. Extensive experimentation validates the proposed\nframework and demonstrates its superiority by outperforming the current\nstate-of-the-art methods by a significant margin.\n"}
{"abstract": "  Pre-compiled binary packages provide a convenient way of efficiently\ndistributing software that has been adopted by most Linux package management\nsystems. However, the heterogeneity of the Linux ecosystem, combined with the\ngrowing number of R extensions available, poses a scalability problem. As a\nresult, efforts to bring binary R packages to Linux have been scattered, and\nlack a proper mechanism to fully integrate them with R's package manager. This\nwork reviews past and present of binary distribution for Linux, and presents a\npath forward by showcasing the `cran2copr' project, an RPM-based\nproof-of-concept implementation of an automated scalable binary distribution\nsystem with the capability of building, maintaining and distributing thousands\nof packages, while providing a portable and extensible bridge to the system\npackage manager.\n"}
{"abstract": "  In performative prediction, predictions guide decision-making and hence can\ninfluence the distribution of future data. To date, work on performative\nprediction has focused on finding performatively stable models, which are the\nfixed points of repeated retraining. However, stable solutions can be far from\noptimal when evaluated in terms of the performative risk, the loss experienced\nby the decision maker when deploying a model. In this paper, we shift attention\nbeyond performative stability and focus on optimizing the performative risk\ndirectly. We identify a natural set of properties of the loss function and\nmodel-induced distribution shift under which the performative risk is convex, a\nproperty which does not follow from convexity of the loss alone. Furthermore,\nwe develop algorithms that leverage our structural assumptions to optimize the\nperformative risk with better sample efficiency than generic methods for\nderivative-free convex optimization.\n"}
{"abstract": "  Current deep quantization methods that produce binary code representations\nfor efficient image retrieval mostly learn codewords from data. They rarely\ninvestigate the effect of their inherent distribution on the quantization and\nthe learning metrics presently used are insufficient for face image retrieval\ntask. To address this, this paper integrates product quantization into an\nend-to-end deep learning framework to retrieve face images. We propose a novel\nscheme that uses predefined orthonormal vectors as codewords, to enhance the\nquantization informativeness and reduce redundancy in the codewords. A tailored\nloss function maximizes discriminability among identities in each quantization\nsubspace for both the quantized and original features. An entropy-based\nregularization term is imposed to reduce the quantization error. Experiments\nwere conducted on three commonly-used datasets for both single- and\ncross-domain retrieval. The proposed method outperformed all the deep\nhashing/quantization methods it was compared with under both settings. We\nobserve that the proposed orthonormal codewords consistently improved both\nmodels' standard retrieval performance and generalization ability. Therefore,\nthe proposed method is more suitable for scalable face image retrieval than\ndeep hashing methods.\n"}
{"abstract": "  The effect of ionisation quenching for ions is critical for experiments\nrelying on the measurement of low energy recoils, such as direct Dark Matter\nsearches. We present ionisation quenching factor estimates over a range of\nenergies for protons, $\\alpha$-particles, and heavier ions in H$_{2}$,\nCH$_{4}$, N$_{2}$, Ar, CO$_{2}$, and C$_{3}$H$_{8}$ gases, estimated from the\nrespective reference W-value measurements. The resulting ionisation quenching\nfactors are compared with predictions from SRIM.\n"}
{"abstract": "  Face Presentation Attack Detection (PAD) has drawn increasing attentions to\nsecure the face recognition systems that are widely used in many applications.\nConventional face anti-spoofing methods have been proposed, assuming that\ntesting is from the same domain used for training, and so cannot generalize\nwell on unseen attack scenarios. The trained models tend to overfit to the\nacquisition sensors and attack types available in the training data. In light\nof this, we propose an end-to-end learning framework based on Domain Adaptation\n(DA) to improve PAD generalization capability. Labeled source-domain samples\nare used to train the feature extractor and classifier via cross-entropy loss,\nwhile unsupervised data from the target domain are utilized in adversarial DA\napproach causing the model to learn domain-invariant features. Using DA alone\nin face PAD fails to adapt well to target domain that is acquired in different\nconditions with different devices and attack types than the source domain. And\nso, in order to keep the intrinsic properties of the target domain, deep\nclustering of target samples is performed. Training and deep clustering are\nperformed end-to-end, and experiments performed on several public benchmark\ndatasets validate that our proposed Deep Clustering guided Unsupervised Domain\nAdaptation (DCDA) can learn more generalized information compared with the\nstate-of-the-art classification error on the target domain.\n"}
{"abstract": "  We propose a new discretization method for PDEs on moving domains in the\nsetting of unfitted finite element methods, which is provably higher-order\naccurate in space and time. In the considered setting, the physical domain that\nevolves essentially arbitrarily through a time-independent computational\nbackground domain, is represented by a level set function. For the time\ndiscretization, the application of standard time stepping schemes that are\nbased on finite difference approximations of the time derivative is not\ndirectly possible, as the degrees of freedom may get active or inactive across\nsuch a finite difference stencil in time. In [Lehrenfeld, Olshanskii. An\nEulerian finite element method for PDEs in time-dependent domains. ESAIM: M2AN,\n53:585--614, 2019] this problem is overcome by extending the discrete solution\nat every timestep to a sufficiently large neighborhood so that all the degrees\nof freedom that are relevant at the next time step stay active. But that paper\nfocuses on low-order methods. We advance these results with introducing and\nanalyzing realizable techniques for the extension to higher order. To obtain\nhigher-order convergence in space and time, we combine the BDF time stepping\nwith the isoparametric unfitted FEM. The latter has been used and analyzed for\nseveral stationary problems before. However, for moving domains the key\ningredient in the method, the transformation of the underlying mesh, becomes\ntime-dependent which gives rise to some technical issues. We treat these with\nspecial care, carry out an a priori error analysis and two numerical\nexperiments.\n"}
{"abstract": "  Cross-ratio degrees count configurations of points $z_1,\\ldots, z_n \\in\n\\mathbb{P}^1$ satisfying $n - 3$ cross-ratio constraints, up to isomorphism.\nThese numbers arise in multiple contexts in algebraic and tropical geometry,\nand may be viewed as combinatorial invariants of certain hypergraphs. We prove\nan upper bound on cross-ratio degrees in terms of the theory of perfect\nmatchings on bipartite graphs. We also discuss several of the many perspectives\non cross-ratio degrees -- including a connection to Gromov-Witten theory -- and\ngive many example computations.\n"}
{"abstract": "  Let $R$ be a subring of $\\mathbb{C}[[z]]$, and let $X \\in \\mathbb{C}[[z]]$.\nThe Newton-Puiseux Theorem implies that if the coefficients of $X$ grow\nsufficiently rapidly relative to the coefficients of the series in $R$, then\n$X$ is transcendental over $R$. We prove an alternative proof of this result by\nestablishing a relationship between the coefficients of $A(X)$ and\n$A^\\prime(X)$, where $A(T)$ is a polynomial over $\\mathbb{C}[[z]]$.\n"}
{"abstract": "  Deterministic dynamics is an essential part of many MCMC algorithms, e.g.\nHybrid Monte Carlo or samplers utilizing normalizing flows. This paper presents\na general construction of deterministic measure-preserving dynamics using\nautonomous ODEs and tools from differential geometry. We show how Hybrid Monte\nCarlo and other deterministic samplers follow as special cases of our theory.\nWe then demonstrate the utility of our approach by constructing a continuous\nnon-sequential version of Gibbs sampling in terms of an ODE flow and extending\nit to discrete state spaces. We find that our deterministic samplers are more\nsample efficient than stochastic counterparts, even if the latter generate\nindependent samples.\n"}
{"abstract": "  Texture matrices are a way of mitigating the redundancy inherent in the\ndescription of flavor physics via Yukawa couplings by eliminating some entries\nin order to identify relevant parameters. A four-zero texture scheme has been\nused in the literature to successfully describe fermion masses and mixing.\nHowever, as we show in this work, improving experimental constraints require an\nupdate to this analysis. In this paper the implications of a 2-zero texture\nmass matrix is studied for quarks and leptons. We show that the introduction of\na new parameter in each mass matrix allow us to reach good results with\nrelative low cost in predictability. We report a numerical study using a\nhybridized nature-inspired/cellular automata search algorithm. We find that\nleptons and quarks can be described by the same 1-zero structure. We describe\nsome scenarios where a simplified description can be achieved, including a\nnarrow region in parameter space where the same values describe charged leptons\nand neutrinos, which is a stronger version of a previously proposed Universal\nTexture Constraint.\n"}
{"abstract": "  The paper addresses two unusual scaling regimes (types of critical behaviour)\npredicted by the field-theoretic renormalization group analysis for a\nself-organized critical system with turbulent motion of the environment. The\nsystem is modelled by the anisotropic stochastic equation for a \"running\nsandpile\" introduced by Hwa and Kardar in [{\\it Phys. Rev. Lett.} {\\bf 62}:\n1813 (1989)]. The turbulent motion is described by the isotropic\nKazantsev-Kraichnan \"rapid-change\" velocity ensemble for an incompressible\nfluid. The original Hwa-Kardar equation allows for independent scaling of the\nspatial coordinates $x_{\\parallel}$ (the coordinate along the preferred\ndimension) and ${\\bf x_{\\bot}}$ (the coordinates in the orthogonal subspace to\nthe preferred direction) that becomes impossible once the isotropic velocity\nensemble is coupled to the equation. However, it is found that one of the\nregimes of the system's critical behaviour (the one where the isotropic\nturbulent motion is irrelevant) recovers the anisotropic scaling through\n\"dimensional transmutation.\" The latter manifests as a dimensionless ratio\nacquiring nontrivial canonical dimension. The critical regime where both the\nvelocity ensemble and the nonlinearity of the Hwa-Kardar equation are relevant\nsimultaneously is also characterized by \"atypical\" scaling. While the ordinary\nscaling with fixed infra-red irrelevant parameters is impossible in this\nregime, the \"restricted\" scaling where the times, the coordinates, and the\ndimensionless ratio are scaled becomes possible. This result brings to mind\nscaling hypotheses modifications (Stell's weak scaling or Fisher's generalized\nscaling) for systems with significantly different characteristic scales.\n"}
{"abstract": "  This technical report presents our solution for temporal action detection\ntask in AcitivityNet Challenge 2021. The purpose of this task is to locate and\nidentify actions of interest in long untrimmed videos. The crucial challenge of\nthe task comes from that the temporal duration of action varies dramatically,\nand the target actions are typically embedded in a background of irrelevant\nactivities. Our solution builds on BMN, and mainly contains three steps: 1)\naction classification and feature encoding by Slowfast, CSN and ViViT; 2)\nproposal generation. We improve BMN by embedding the proposed Proposal Relation\nNetwork (PRN), by which we can generate proposals of high quality; 3) action\ndetection. We calculate the detection results by assigning the proposals with\ncorresponding classification results. Finally, we ensemble the results under\ndifferent settings and achieve 44.7% on the test set, which improves the\nchampion result in ActivityNet 2020 by 1.9% in terms of average mAP.\n"}
{"abstract": "  Recent work has focused on the potential and pitfalls of causal\nidentification in observational studies with multiple simultaneous treatments.\nOn the one hand, a latent variable model fit to the observed treatments can\nidentify essential aspects of the distribution of unobserved confounders. On\nthe other hand, it has been shown that even when the latent confounder\ndistribution is known exactly, causal effects are still not point identifiable.\nThus, the practical benefits of latent variable modeling in multi-treatment\nsettings remain unclear. We clarify these issues with a sensitivity analysis\nmethod that can be used to characterize the range of causal effects that are\ncompatible with the observed data. Our method is based on a copula\nfactorization of the joint distribution of outcomes, treatments, and\nconfounders, and can be layered on top of arbitrary observed data models. We\npropose a practical implementation of this approach making use of the Gaussian\ncopula, and establish conditions under which causal effects can be bounded. We\nalso describe approaches for reasoning about effects, including calibrating\nsensitivity parameters, quantifying robustness of effect estimates, and\nselecting models which are most consistent with prior hypotheses.\n"}
{"abstract": "  In this work, we introduce an unconditional video generative model,\nInMoDeGAN, targeted to (a) generate high quality videos, as well as to (b)\nallow for interpretation of the latent space. For the latter, we place emphasis\non interpreting and manipulating motion. Towards this, we decompose motion into\nsemantic sub-spaces, which allow for control of generated samples. We design\nthe architecture of InMoDeGAN-generator in accordance to proposed Linear Motion\nDecomposition, which carries the assumption that motion can be represented by a\ndictionary, with related vectors forming an orthogonal basis in the latent\nspace. Each vector in the basis represents a semantic sub-space. In addition, a\nTemporal Pyramid Discriminator analyzes videos at different temporal\nresolutions. Extensive quantitative and qualitative analysis shows that our\nmodel systematically and significantly outperforms state-of-the-art methods on\nthe VoxCeleb2-mini and BAIR-robot datasets w.r.t. video quality related to (a).\nTowards (b) we present experimental results, confirming that decomposed\nsub-spaces are interpretable and moreover, generated motion is controllable.\n"}
{"abstract": "  To broaden the scope of van der Waals 2D magnets, we report the synthesis and\nmagnetism of covalent 2D magnetic Cr$_2$Te$_3$ with a thickness down to\none-unit-cell. The 2D Cr$_2$Te$_3$ crystals exhibit robust ferromagnetism with\na Curie temperature of 180 K, a large perpendicular anisotropy of 7*105 J m-3,\nand a high coercivity of ~ 4.6 kG at 20 K. First-principles calculations\nfurther show a transition from canted to collinear ferromagnetism, a transition\nfrom perpendicular to in-plane anisotropy, and emergent half-metallic behavior\nin atomically-thin Cr$_2$Te$_3$, suggesting its potential application for\ninjecting carriers with high spin polarization into spintronic devices.\n"}
{"abstract": "  We give a lower bound of the $\\delta$-invariants of ample line bundles in\nterms of Seshadri constants. As applications, we prove the uniform K-stability\nof infinitely many families of Fano hypersurfaces of arbitrarily large index,\nas well as the uniform K-stability of most families of smooth Fano threefolds\nof Picard number one.\n"}
{"abstract": "  One of the main characteristics of the penumbra of sunspots is the radially\noutward-directed Evershed flow. Only recently have penumbral regions been\nreported with similar characteristics to normal penumbral filaments, but with\nan opposite direction of the flow. Such flows directed towards the umbra are\nknown as counter Evershed flows (CEFs). We aim to determine the frequency of\noccurrence of CEFs in active regions (ARs) and to characterize their lifetime\nand the prevailing conditions in the ARs. We analysed the continuum images,\nDopplergrams, and magnetograms recorded by SDO/HMI of 97 ARs that appeared from\n2011 to 2017. We followed the ARs for $9.6\\pm1.4$ days on average. We found 384\nCEFs in total, with a median value of 6 CEFs per AR. CEFs are a rather common\nfeature, they occur in 83.5% of all ARs regardless of the magnetic complexity\nof the AR. However, CEFs were observed on average only during 5.9% of the mean\ntotal duration of all the observations analyzed here. The lifetime of CEFs\nfollows a log-normal distribution with a median value of 10.6$_{-6.0}^{+12.4}$\nhr. In addition, we report two populations of CEFs depending on whether they\nare associated with light bridges, or not. We explain that the rarity of\nreports of CEFs in the literature is a result of highly incomplete coverage of\nARs with spectropolarimetric data. By using the continuous observations now\nroutinely available from space, we are able to overcome this limitation.\n"}
{"abstract": "  SHINE, as the first hard x-ray free-electron-laser (FEL) facility in China,\nis designed to provide high-brightness FEL lasing under high-repetition-rate\noperation mode. In order to drive x-ray FEL pulses with high qualities, the\nphotoinjector section is deployed to provide the specified electron beam with\nlow transverse emittance and high brightness. Normally the multi-objective\noptimization algorithm is employed in the injector beam dynamics design. In\nthis paper, the many-objective optimization algorithm NSGA-III is introduced to\nthe injector physical design for optimizing the 4 beam quality properties for\nthe first time. The results of the optimization are presented and analyzed. The\napplication of NSGA-III can provide guidance for further physical research as\nwell as improve the beam dynamics optimization efficiency.\n"}
{"abstract": "  Singular equations with rank-deficient Jacobians arise frequently in\nalgebraic computing applications. As shown in case studies in this paper,\ndirect and intuitive modeling of algebraic problems often results in\nnonisolated singular solutions. The challenges become formidable when the\nproblems need to be solved from empirical data of limited accuracy. A newly\ndiscovered low-rank Newton's iteration emerges as an effective regularization\nmechanism that enables solving singular equations accurately with an error\nbound in the same order as the data error. This paper elaborates applications\nof new methods on solving singular algebraic equations such as singular linear\nsystems, polynomial GCD and factorizations as well as matrix defective\neigenvalue problems.\n"}
{"abstract": "  We consider random filtered complexes built over marked point processes on\nEuclidean spaces. Examples of our filtered complexes include a filtration of\n$\\check{\\textrm{C}}$ech complexes of a family of sets with various sizes,\ngrowths, and shapes. We establish the law of large numbers for persistence\ndiagrams as the size of the convex window observing a marked point process\ntends to infinity.\n"}
{"abstract": "  This paper introduces and analyzes an improved Q-learning algorithm for\ndiscrete-time linear time-invariant systems. The proposed method does not\nrequire any knowledge of the system dynamics, and it enjoys significant\nefficiency advantages over other data-based optimal control methods in the\nliterature. This algorithm can be fully executed off-line, as it does not\nrequire to apply the current estimate of the optimal input to the system as in\non-policy algorithms. It is shown that a persistently exciting input, defined\nfrom an easily tested matrix rank condition, guarantees the convergence of the\nalgorithm. Moreover, the method avoids the use of linear matrix inequalities\n(LMIs) for control design, decreasing the corresponding computational\ncomplexity. A data-based method is proposed to design the initial stabilizing\nfeedback gain that the algorithm requires. Robustness of the algorithm in the\npresence of noisy measurements is analyzed. Both theoretical and simulation\ncomparisons are performed to show the advantages of this algorithm against\nother model-free control design methods.\n"}
{"abstract": "  Close to a magical angle, twisted bilayer graphene (TBLG) systems exhibit\nisolated flat electronic bands and, accordingly, strong electron localization.\nTBLGs have hence been ideal platforms to explore superconductivity, correlated\ninsulating states, magnetism, and quantized anomalous Hall states in reduced\ndimension. Below a threshold twist angle ($\\sim$ $1.1^\\circ$), the TBLG\nsuperlattice undergoes lattice reconstruction, leading to a periodic moir\\'e\nstructure which presents a marked atomic corrugation. Using a tight-binding\nframework, this research demonstrates that superlattice reconstruction affects\nsignificantly the electronic structure of small-angle TBLGs. The first magic\nangle at $\\sim$ $1.1^\\circ$ is found to be a critical case presenting globally\nmaximized electron localization, thus separating reconstructed TBLGs into two\nclasses with clearly distinct electronic properties. While low-energy Dirac\nfermions are still preserved at large twist angles $> 1.1 ^\\circ$, small-angle\n($\\lesssim 1.1^\\circ$) TBLG systems present common features such as large\nspatial variation and strong electron localization observed in unfavorable AA\nstacking regions. However, for small twist angles below $1.1 ^\\circ$, the\nrelative contribution of the local AA regions is progressively reduced, thus\nprecluding the emergence of further magic angles, in very good agreement with\nexisting experimental evidence.\n"}
{"abstract": "  The paper is concerned with an inverse point source problem for the Helmholtz\nequation. It consists of recovering the locations and amplitudes of a finite\nnumber of radiative point sources inside a given inhomogeneous medium from the\nknowledge of a single boundary measurement. The main result of the paper is a\nnew H\\\"{o}lder type stability estimate for the inversion under the assumption\nthat the point sources are well separated. The proof of the stability is based\non a combination of Carleman estimates and a technique for proving uniqueness\nof the Cauchy problem.\n"}
{"abstract": "  Using a combination of general-relativistic magnetohydrodynamics simulations\nand ray tracing of synchrotron emission, we study the effect of modest (24\ndegrees) misalignment between the black hole spin and plasma angular momentum,\nfocusing on the variability of total flux, image centroids, and image sizes. We\nconsider both millimeter and infrared (IR) observables motivated by Sagittarius\nA* (Sgr A*), though our results apply more generally to optically thin flows.\nFor most quantities, tilted accretion is more variable, primarily due to a\nsignificantly hotter and denser \"coronal\" region well off the disk midplane. We\nfind (1) a 150% increase in millimeter light curve variability when adding tilt\nto the flow; (2) the tilted image centroid in the millimeter shifts on a scale\nof 3.7 microarcseconds over 28 hours (5000 gravitational times) for some\nelectron temperature models; (3) tilted disk image diameters in the millimeter\ncan be 10% larger (52 versus 47 microarcseconds) than those of aligned disks at\ncertain viewing angles; (4) the tilted models produce significant IR flux,\nsimilar to that seen in Sgr A*, with comparable or even greater variability\nthan observed; (5) for some electron models, the tilted IR centroid moves by\nmore than 50 microarcseconds over several hours, in a similar fashion to the\ncentroid motion detected by the GRAVITY interferometer.\n"}
{"abstract": "  In many health domains such as substance-use, outcomes are often counts with\nan excessive number of zeros (EZ) - count data having zero counts at a rate\nsignificantly higher than that expected of a standard count distribution (e.g.,\nPoisson). However, an important gap exists in sample size estimation\nmethodology for planning sequential multiple assignment randomized trials\n(SMARTs) for comparing dynamic treatment regimens (DTRs) using longitudinal\ncount data. DTRs, also known as treatment algorithms or adaptive interventions,\nmimic the individualized and evolving nature of patient care through the\nspecification of decision rules guiding the type, timing and modality of\ndelivery, and dosage of treatments to address the unique and changing needs of\nindividuals. To close this gap, we develop a Monte Carlo-based approach to\nsample size estimation. A SMART for engaging alcohol and cocaine-dependent\npatients in treatment is used as motivation.\n"}
{"abstract": "  We study the phase synchronization problem with noisy measurements\n$Y=z^*z^{*H}+\\sigma W\\in\\mathbb{C}^{n\\times n}$, where $z^*$ is an\n$n$-dimensional complex unit-modulus vector and $W$ is a complex-valued\nGaussian random matrix. It is assumed that each entry $Y_{jk}$ is observed with\nprobability $p$. We prove that an SDP relaxation of the MLE achieves the error\nbound $(1+o(1))\\frac{\\sigma^2}{2np}$ under a normalized squared $\\ell_2$ loss.\nThis result matches the minimax lower bound of the problem, and even the\nleading constant is sharp. The analysis of the SDP is based on an equivalent\nnon-convex programming whose solution can be characterized as a fixed point of\nthe generalized power iteration lifted to a higher dimensional space. This\nviewpoint unifies the proofs of the statistical optimality of three different\nmethods: MLE, SDP, and generalized power method. The technique is also applied\nto the analysis of the SDP for $\\mathbb{Z}_2$ synchronization, and we achieve\nthe minimax optimal error $\\exp\\left(-(1-o(1))\\frac{np}{2\\sigma^2}\\right)$ with\na sharp constant in the exponent.\n"}
{"abstract": "  Noise injection-based regularization, such as Dropout, has been widely used\nin image domain to improve the performance of deep neural networks (DNNs).\nHowever, efficient regularization in the point cloud domain is rarely\nexploited, and most of the state-of-the-art works focus on data\naugmentation-based regularization. In this paper, we, for the first time,\nperform systematic investigation on noise injection-based regularization for\npoint cloud-domain DNNs. To be specific, we propose a series of regularization\ntechniques, namely DropFeat, DropPoint and DropCluster, to perform noise\ninjection on the point feature maps at the feature level, point level and\ncluster level, respectively. We also empirically analyze the impacts of\ndifferent factors, including dropping rate, cluster size and dropping position,\nto obtain useful insights and general deployment guidelines, which can\nfacilitate the adoption of our approaches across different datasets and DNN\narchitectures.\n  We evaluate our proposed approaches on various DNN models for different point\ncloud processing tasks. Experimental results show our approaches enable\nsignificant performance improvement. Notably, our DropCluster brings 1.5%, 1.3%\nand 0.8% higher overall accuracy for PointNet, PointNet++ and DGCNN,\nrespectively, on ModelNet40 shape classification dataset. On ShapeNet part\nsegmentation dataset, DropCluster brings 0.5%, 0.5% and 0.2% mean\nIntersection-over-union (IoU) increase for PointNet, PointNet++ and DGCNN,\nrespectively. On S3DIS semantic segmentation dataset, DropCluster improves the\nmean IoU of PointNet, PointNet++ and DGCNN by 3.2%, 2.9% and 3.7%,\nrespectively. Meanwhile, DropCluster also enables the overall accuracy increase\nfor these three popular backbone DNNs by 2.4%, 2.2% and 1.8%, respectively.\n"}
{"abstract": "  Self-supervised clustering methods have achieved increasing accuracy in\nrecent years but do not yet perform as well as supervised classification\nmethods. This contrasts with the situation for feature learning, where\nself-supervised features have recently surpassed the performance of supervised\nfeatures on several important tasks. We hypothesize that the performance gap is\ndue to the difficulty of specifying, without supervision, which features\ncorrespond to class differences that are semantic to humans. To reduce the\nperformance gap, we introduce the \"single-noun\" prior - which states that\nsemantic clusters tend to correspond to concepts that humans label by a\nsingle-noun. By utilizing a pre-trained network that maps images and sentences\ninto a common space, we impose this prior obtaining a constrained optimization\ntask. We show that our formulation is a special case of the facility location\nproblem, and introduce a simple-yet-effective approach for solving this\noptimization task at scale. We test our approach on several commonly reported\nimage clustering datasets and obtain significant accuracy gains over the best\nexisting approaches.\n"}
{"abstract": "  We consider high order approximations of the solution of the stochastic\nfiltering problem, derive their pathwise representation in the spirit of the\nearlier work of Clark and Davis and prove their robustness property. In\nparticular, we show that the high order discretised filtering functionals can\nbe represented by Lipschitz continuous functions defined on the observation\npath space. This property is important from the practical point of view as it\nis in fact the pathwise version of the filtering functional that is sought in\nnumerical applications. Moreover, the pathwise viewpoint will be a stepping\nstone into the rigorous development of machine learning methods for the\nfiltering problem. This work is a continuation of a recent work by two of the\nauthors where a discretisation of the solution of the filtering problem of\narbitrary order has been established. We expand the previous work by showing\nthat robust approximations can be derived from the discretisations therein.\n"}
{"abstract": "  We report multi-wavelength study of a complex M-class solar eruptive flare\nthat consists of three different sets of flare ribbons, viz. circular,\nparallel, and remote ribbons. Magnetic field modeling of source active region\nNOAA 12242 exhibits the presence of 3D null-point magnetic topology which\nencompasses an inner bipolar region. The event initiates with the faint\nsignatures of the circular ribbon along with remote brightening right from the\npre-flare phase which points toward the ongoing slow yet persistent null-point\nreconnection. We first detected flux cancellation and an associated\nbrightening, which are likely signatures of tether-cutting reconnection that\nbuilds the flux rope near the polarity inversion line (PIL) of the inner\nbipolar region. In the next stage, with the onset of M8.7 flare, there is a\nsubstantial enhancement in the brightening of circular ribbon which essentially\nsuggests an increase in the rate of ongoing null-point reconnection. Finally,\nthe eruption of underlying flux rope triggers ``standard flare reconnection\"\nbeneath it producing an abrupt rise in the intensity of the parallel ribbons as\nwell as enhancing the rate of null-point reconnection by external forcing. We\nshow that within the the fan dome, the region with magnetic decay index n>1.5\nborders the null-point QSL. Our analysis suggests that both the torus\ninstability and the breakout model have played role toward the triggering\nmechanism for the eruptive flare. This event is a nice example of the dynamical\nevolution of a flux rope initially confined in a null-point topology, that\nsubsequently activates and erupts with the progression of the circular -- cum\n-- parallel ribbon flare.\n"}
{"abstract": "  This paper considers a multiagent, connected, robotic fleet where the primary\nfunctionality of the agents is sensing. A distributed multi-sensor control\nstrategy maximizes the value of the collective sensing capability of the fleet,\nusing an information-driven approach. Each agent individually performs sensor\nprocessing (Kalman Filtering and Joint Probabilistic Data Association) to\nidentify trajectories (and associated distributions). Using communications with\nits neighbors the agents enhance the prediction of the trajectories using a\nConsensus of Information approach that iteratively calculates the\nKullback-Leibler average of trajectory distributions, enabling the calculation\nof the value of the collective information for the fleet. The dynamics of the\nagents, the evolution of the identified trajectories for each agent, and the\ndynamics of individual observed objects are captured as a Partially Observable\nMarkov Decision Process (POMDP). Using this POMDP and applying rollout with\nreceding horizon control, an optimized non-myopic control policy that maximizes\nthe collective fleet information value is synthesized. Simulations are\nperformed for a scenario with three heterogeneous UAVs performing coordinated\ntarget tracking that illustrate the proposed methodology and compare the\ncentralized approach with a contemporary sequential multiagent distributed\ndecision technique.\n"}
{"abstract": "  Most existing approaches to train a unified multi-organ segmentation model\nfrom several single-organ datasets require simultaneously access multiple\ndatasets during training. In the real scenarios, due to privacy and ethics\nconcerns, the training data of the organs of interest may not be publicly\navailable. To this end, we investigate a data-free incremental organ\nsegmentation scenario and propose a novel incremental training framework to\nsolve it. We use the pretrained model instead of its own training data for\nprivacy protection. Specifically, given a pretrained $K$ organ segmentation\nmodel and a new single-organ dataset, we train a unified $K+1$ organ\nsegmentation model without accessing any data belonging to the previous\ntraining stages. Our approach consists of two parts: the background label\nalignment strategy and the uncertainty-aware guidance strategy. The first part\nis used for knowledge transfer from the pretained model to the training model.\nThe second part is used to extract the uncertainty information from the\npretrained model to guide the whole knowledge transfer process. By combing\nthese two strategies, more reliable information is extracted from the\npretrained model without original training data. Experiments on multiple\npublicly available pretrained models and a multi-organ dataset MOBA have\ndemonstrated the effectiveness of our framework.\n"}
{"abstract": "  Automated and accurate segmentations of left atrium (LA) and atrial scars\nfrom late gadolinium-enhanced cardiac magnetic resonance (LGE CMR) images are\nin high demand for quantifying atrial scars. The previous quantification of\natrial scars relies on a two-phase segmentation for LA and atrial scars due to\ntheir large volume difference (unbalanced atrial targets). In this paper, we\npropose an inter-cascade generative adversarial network, namely JAS-GAN, to\nsegment the unbalanced atrial targets from LGE CMR images automatically and\naccurately in an end-to-end way. Firstly, JAS-GAN investigates an adaptive\nattention cascade to automatically correlate the segmentation tasks of the\nunbalanced atrial targets. The adaptive attention cascade mainly models the\ninclusion relationship of the two unbalanced atrial targets, where the\nestimated LA acts as the attention map to adaptively focus on the small atrial\nscars roughly. Then, an adversarial regularization is applied to the\nsegmentation tasks of the unbalanced atrial targets for making a consistent\noptimization. It mainly forces the estimated joint distribution of LA and\natrial scars to match the real ones. We evaluated the performance of our\nJAS-GAN on a 3D LGE CMR dataset with 192 scans. Compared with the\nstate-of-the-art methods, our proposed approach yielded better segmentation\nperformance (Average Dice Similarity Coefficient (DSC) values of 0.946 and\n0.821 for LA and atrial scars, respectively), which indicated the effectiveness\nof our proposed approach for segmenting unbalanced atrial targets.\n"}
{"abstract": "  This study proposes an efficient exact k-flexible aggregate nearest neighbor\n(k-FANN) search algorithm in road networks using the M-tree. The\nstate-of-the-art IER-kNN algorithm used the R-tree and pruned off unnecessary\nnodes based on the Euclidean coordinates of objects in road networks. However,\nIER-kNN made many unnecessary accesses to index nodes since the Euclidean\ndistances between objects are significantly different from the actual\nshortest-path distances between them. In contrast, our algorithm proposed in\nthis study can greatly reduce unnecessary accesses to index nodes compared with\nIER-kNN since the M-tree is constructed based on the actual shortest-path\ndistances between objects. To the best of our knowledge, our algorithm is the\nfirst exact FANN algorithm that uses the M-tree. We prove that our algorithm\ndoes not cause any false drop. In conducting a series of experiments using\nvarious real road network datasets, our algorithm consistently outperformed\nIER-kNN by up to 6.92 times.\n"}
{"abstract": "  We present a scenario for the formation and the morphology of the\narrow-shaped dwarf irregular galaxy NGC 1427A in the Fornax Cluster. This\ngalaxy shows intriguing stellar and gaseous tails pointing in different\ndirections for which alternative but not conclusive formation scenarios have\nbeen proposed in the literature. We performed N-body/SPH simulations of dwarf\ngalaxies falling into a model of the Fornax cluster, exhibiting a\njellyfish-like appearance while undergoing ram-pressure stripping. We noted\nthat some of our models show interesting tail morphologies similar to that of\nNGC 1427A. In this way, the peculiar NGC 1427A structure can be studied using\nmodels whose stellar and neutral gas photometry and kinematics are in good\nagreement with the observed ones, without the need of invoking an interaction\nwith a nearby galaxy. Thanks to the tails, we can identify the requirements for\na galaxy to expose such a structure and assess the possible position and\nvelocity of the galaxy in the cluster. This puts constraints on the orbit of\nthe galaxy, its position in the cluster and the time since its pericentre\npassage. From the statistics of identified snapshots following our modelling,\nwe found that the most likely position of the galaxy is around 200 kpc in front\nof the cluster centre, travelling towards the cluster with a velocity angle\nwith respect to the line-of-sight direction of around 50 deg. This analysis can\nbe useful in future observations of similar galaxies in clusters to\ncharacterise their position and velocity in the cluster and their formation.\n"}
{"abstract": "  Many mobile robotic platforms rely on an accurate knowledge of the extrinsic\ncalibration parameters, especially systems performing visual stereo matching.\nAlthough a number of accurate stereo camera calibration methods have been\ndeveloped, which provide good initial \"factory\" calibrations, the determined\nparameters can lose their validity over time as the sensors are exposed to\nenvironmental conditions and external effects. Thus, on autonomous platforms\non-board diagnostic methods for an early detection of the need to repeat\ncalibration procedures have the potential to prevent critical failures of\ncrucial systems, such as state estimation or obstacle detection. In this work,\nwe present a novel data-driven method to estimate the calibration quality and\ndetect discrepancies between the original calibration and the current system\nstate for stereo camera systems. The framework consists of a novel dataset\ngeneration pipeline to train CalQNet, a deep convolutional neural network.\nCalQNet can estimate the calibration quality using a new metric that\napproximates the degree of miscalibration in stereo setups. We show the\nframework's ability to predict from a single stereo frame if a state-of-the-art\nstereo-visual odometry system will diverge due to a degraded calibration in two\nreal-world experiments.\n"}
{"abstract": "  Weakly-supervised temporal action localization aims to localize action\ninstances temporal boundary and identify the corresponding action category with\nonly video-level labels. Traditional methods mainly focus on foreground and\nbackground frames separation with only a single attention branch and class\nactivation sequence. However, we argue that apart from the distinctive\nforeground and background frames there are plenty of semantically ambiguous\naction context frames. It does not make sense to group those context frames to\nthe same background class since they are semantically related to a specific\naction category. Consequently, it is challenging to suppress action context\nframes with only a single class activation sequence. To address this issue, in\nthis paper, we propose an action-context modeling network termed ACM-Net, which\nintegrates a three-branch attention module to measure the likelihood of each\ntemporal point being action instance, context, or non-action background,\nsimultaneously. Then based on the obtained three-branch attention values, we\nconstruct three-branch class activation sequences to represent the action\ninstances, contexts, and non-action backgrounds, individually. To evaluate the\neffectiveness of our ACM-Net, we conduct extensive experiments on two benchmark\ndatasets, THUMOS-14 and ActivityNet-1.3. The experiments show that our method\ncan outperform current state-of-the-art methods, and even achieve comparable\nperformance with fully-supervised methods. Code can be found at\nhttps://github.com/ispc-lab/ACM-Net\n"}
{"abstract": "  The log-aesthetic curve has a significant factor in the field of aesthetic\ndesign to meet the high industrial requirements. It has much exceptional\nproperty and a large number of research papers are published, since its\nintroduction. It can be parametrized by tangential angle, and it is\nautoevolute, meaning that its evolute is also a log-aesthetic curve. In this\npaper, we will examine the isoptics of the log-aesthetic curve and determine if\nit is autoisoptic as well. We will show that the log-aesthetic curve only\ncoincides its isoptic in some specific cases. We will also provide a new\nformula to calculate the isoptic curve of any tangential angle parametrized\ncurves.\n"}
{"abstract": "  With the advent of the Internet of Things (IoT), establishing a secure\nchannel between smart devices becomes crucial. Recent research proposes\nzero-interaction pairing (ZIP), which enables pairing without user assistance\nby utilizing devices' physical context (e.g., ambient audio) to obtain a shared\nsecret key. The state-of-the-art ZIP schemes suffer from three limitations: (1)\nprolonged pairing time (i.e., minutes or hours), (2) vulnerability to\nbrute-force offline attacks on a shared key, and (3) susceptibility to attacks\ncaused by predictable context (e.g., replay attack) because they rely on\nlimited entropy of physical context to protect a shared key. We address these\nlimitations, proposing FastZIP, a novel ZIP scheme that significantly reduces\npairing time while preventing offline and predictable context attacks. In\nparticular, we adapt a recently introduced Fuzzy Password-Authenticated Key\nExchange (fPAKE) protocol and utilize sensor fusion, maximizing their\nadvantages. We instantiate FastZIP for intra-car device pairing to demonstrate\nits feasibility and show how the design of FastZIP can be adapted to other ZIP\nuse cases. We implement FastZIP and evaluate it by driving four cars for a\ntotal of 800 km. We achieve up to three times shorter pairing time compared to\nthe state-of-the-art ZIP schemes while assuring robust security with\nadversarial error rates below 0.5%.\n"}
{"abstract": "  We review recent progress in understanding nearly integrable models within\nthe framework of generalized hydrodynamics (GHD). Integrable systems have\ninfinitely many conserved quantities and stable quasiparticle excitations: when\nintegrability is broken, only a few residual conserved quantities survive,\neventually leading to thermalization, chaotic dynamics and conventional\nhydrodynamics. In this review, we summarize recent efforts to take into account\nsmall integrability breaking terms, and describe the transition from GHD to\nstandard hydrodynamics. We discuss the current state of the art, with emphasis\non weakly inhomogeneous potentials, generalized Boltzmann equations and\ncollision integrals, as well as bound-state recombination effects. We also\nidentify important open questions for future works.\n"}
{"abstract": "  This paper describes the ExKaldi-RT online automatic speech recognition (ASR)\ntoolkit that is implemented based on the Kaldi ASR toolkit and Python language.\nExKaldi-RT provides tools for building online recognition pipelines. While\nsimilar tools are available built on Kaldi, a key feature of ExKaldi-RT that it\nworks on Python, which has an easy-to-use interface that allows online ASR\nsystem developers to develop original research, such as by applying neural\nnetwork-based signal processing and by decoding model trained with deep\nlearning frameworks. We performed benchmark experiments on the minimum\nLibriSpeech corpus, and it showed that ExKaldi-RT could achieve competitive ASR\nperformance in real-time recognition.\n"}
{"abstract": "  Answering questions about why characters perform certain actions is central\nto understanding and reasoning about narratives. Despite recent progress in QA,\nit is not clear if existing models have the ability to answer \"why\" questions\nthat may require commonsense knowledge external to the input narrative. In this\nwork, we introduce TellMeWhy, a new crowd-sourced dataset that consists of more\nthan 30k questions and free-form answers concerning why characters in short\nnarratives perform the actions described. For a third of this dataset, the\nanswers are not present within the narrative. Given the limitations of\nautomated evaluation for this task, we also present a systematized human\nevaluation interface for this dataset. Our evaluation of state-of-the-art\nmodels show that they are far below human performance on answering such\nquestions. They are especially worse on questions whose answers are external to\nthe narrative, thus providing a challenge for future QA and narrative\nunderstanding research.\n"}
{"abstract": "  We have revisited the chemistry of chlorine-bearing species in the diffuse\ninterstellar medium with new observations of the HCl$^+$ molecular ion and new\nastrochemical models. Using the GREAT instrument on board SOFIA, we observed\nthe $^2\\Pi_{3/2}\\, J = 5/2 - 3/2$ transition of HCl$^+$ near 1444 GHz toward\nthe bright THz continuum source W49N. We detected absorption by diffuse\nforeground gas unassociated with the background source, and were able to\nthereby measure the distribution of HCl$^+$ along the sight-line. We\ninterpreted the observational data using an updated version of an astrochemical\nmodel used previously in a theoretical study of Cl-bearing interstellar\nmolecules. The abundance of HCl$^+$ was found to be almost constant relative to\nthe related H$_2$Cl$^+$ ion, but the observed $n({\\rm H_2Cl^+})/n({\\rm HCl^+})$\nabundance ratio exceeds the predictions of our astrochemical model by an\norder-of-magnitude. This discrepancy suggests that the rate of the primary\ndestruction process for ${\\rm H_2Cl^+}$, dissociative recombination, has been\nsignificantly overestimated. For HCl$^+$, the model predictions can provide a\nsatisfactory fit to the observed column densities along the W49N sight-line\nwhile simultaneously accounting for the ${\\rm OH^+}$ and ${\\rm H_2O^+}$ column\ndensities.\n"}
{"abstract": "  We consider functional differential equations(FDEs) which are perturbations\nof smooth ordinary differential equations(ODEs). The FDE can involve multiple\nstate-dependent delays or distributed delays (forward or backward). We show\nthat, under some mild assumptions, if the ODE has a nondegenerate periodic\norbit, then the FDE has a smooth periodic orbit. Moreover, we get smooth\ndependence of the periodic orbit and its frequency on parameters with high\nregularity.\n  The result also applies to FDEs which are perturbations of some evolutionary\npartial differential equations(PDEs).\n  The proof consists in solving functional equations satisfied by the\nparameterization of the periodic orbit and the frequency using a fixed point\napproach. We do not need to consider the smoothness of the evolution or even\nthe phase space of the FDEs.\n"}
{"abstract": "  Microbially induced calcite precipitation (MICP) is a technology for sealing\nleakage paths to ensure the safe storage of CO$_2$ in geological formations. In\nthis work we introduce a numerical simulator of MICP for field-scale studies.\nThis simulator is implemented in the open porous media (OPM) framework. We\ncompare the numerical results to simulations using an upgraded implementation\nof the mathematical model in the MATLAB reservoir simulation toolbox (MRST).\nFinally, we consider a 3D system consisting of two aquifers separated by\ncaprock with a leakage path across the width of the reservoir. We study a\nstrategy where microbial solution is injected only at the beginning of the\ntreatment and subsequently either growth solution or cementation solution is\ninjected for biofilm development or calcite precipitation. By applying this\nstrategy, the numerical results show that the MICP technology could be used to\nseal these leakage paths.\n"}
{"abstract": "  pyGDM is a python toolkit for electro-dynamical simulations of individual\nnano-structures, based on the Green Dyadic Method (GDM). pyGDM uses the concept\nof a generalized propagator, which allows to solve cost-efficiently\nmonochromatic problems with a large number of varying illumination conditions\nsuch as incident angle scans or focused beam raster-scan simulations. We\nprovide an overview of new features added since the initial publication\n[Wiecha, Computer Physics Communications 233, pp.167-192 (2018)]. The updated\nversion of pyGDM is implemented in pure python, removing the former dependency\non fortran-based binaries. In the course of this re-write, the toolkit's\ninternal architecture has been completely redesigned to offer a much wider\nrange of possibilities to the user such as the choice of the dyadic Green's\nfunctions describing the environment. A new class of dyads allows to perform 2D\nsimulations of infinitely long nanostructures. While the Green's dyads in pyGDM\nare based on a quasistatic description for interfaces, we also provide as new\nexternal python package pyGDM2_retard a module with retarded Green's tensors\nfor an environment with two interfaces. We have furthermore added\nfunctionalities for simulations using fast-electron excitation, namely electron\nenergy loss spectroscopy and cathodoluminescence. Along with several further\nnew tools and improvements, the update includes also the possibility to\ncalculate the magnetic field and the magnetic LDOS inside nanostructures,\nfield-gradients in- and outside a nanoparticle, optical forces or the chirality\nof nearfields. All new functionalities remain compatible with the evolutionary\noptimization module of pyGDM for nano-photonics inverse design.\n"}
{"abstract": "  Bubble-sort star graphs are a combination of star graphs and bubble sort\ngraphs. They are bipartite graphs and also form a family of Cayley graphs. The\ndecycling number of a graph is the minimum number of vertices whose removal\nfrom the graph results in an acyclic subgraph. In this paper, we prove the\ndecycling number D(n) of an n-dimensional bubble-sort star graph for n <= 5. We\nalso show D(n) satisfies the inequalities for n >= 6.\n"}
{"abstract": "  In insulators, Born effective charges describe the electrical polarization\ninduced by the displacement of individual atomic sublattices. Such a physical\nproperty is at first sight irrelevant for metals and doped semiconductors,\nwhere the macroscopic polarization is ill-defined. Here we show that, in clean\nconductors, going beyond the adiabatic approximation results in nonadiabatic\nBorn effective charges that are well defined in the low-frequency limit. In\naddition, we find that the sublattice sum of the nonadiabatic Born effective\ncharges does not vanish as it does in the insulating case, but instead is\nproportional to the Drude weight. We demonstrate these formal results with\ndensity functional perturbation theory calculations of Al, and electron-doped\nSnS$_2$ and SrTiO$_3$.\n"}
{"abstract": "  The compact set of homogeneous quadratic polynomials in $n$ real variables\nwith modulus bounded by 1 on the unit sphere $S^{n-1}$ is trivially\nsemi-definite representable. The compact set of homogeneous ternary quartics\nwith modulus bounded by 1 on the unit sphere $S^2$ is also semi-definite\nrepresentable. This suggests that the compact set of homogeneous ternary cubics\nwith modulus bounded by 1 on $S^2$ is semi-definite representable. We deduce an\nexplicit semi-definite representation of this norm ball. More generally, we\nprovide a semi-definite description of the cone of inhomogeneous ternary cubics\nwhich are nonnegative on $S^2$.\n"}
{"abstract": "  We analyze dynamics of chiral gauge theories based on the $SU(N)$ gauge group\nwith one anti-symmetric tensor $A$ and $(N-4)$ anti-fundamentals $F_{i}$ when\n$N$ is odd. Based on the continuity to the supersymmetric gauge theories with\nanomaly-mediated supersymmetry breaking, we claim that the global $SU(N-4)$\nsymmetry is spontaneously broken to $Sp(N-5)$. There are $N-5$ massless\nfermions as a fundamental representation of $Sp(N-5)$, and another massless\nfermion, together saturating the anomaly matching conditions. When $N$ is even,\nthe unbroken flavor symmetry is $Sp(N-4)$ while there are no massless fermions.\nOur result is different from the dynamics suggested by tumbling where the full\n$SU(N-4)$ symmetry is unbroken, but the tumbling picture can be modified via\nthe addition of a second condensate to produce the symmetry breaking pattern\npredicted from our method.\n"}
{"abstract": "  In this paper, we expand on results from our previous paper \"The Case Against\nSmooth Null Infinity I: Heuristics and Counter-Examples\" [1] by showing that\nthe failure of \"peeling\" (and, thus, of smooth null infinity) in a\nneighbourhood of $i^0$ derived therein translates into logarithmic corrections\nat leading order to the well-known Price's law asymptotics near $i^+$. This\nsuggests that the non-smoothness of $\\mathcal{I}^+$ is physically measurable.\nMore precisely, we consider the linear wave equation $\\Box_g \\phi=0$ on a fixed\nSchwarzschild background ($M>0$), and we show the following: If one imposes\nconformally smooth initial data on an ingoing null hypersurface (extending to\n$\\mathcal{H}^+$ and terminating at $\\mathcal{I}^-$) and vanishing data on\n$\\mathcal{I}^-$ (this is the no incoming radiation condition), then the precise\nleading-order asymptotics of the solution $\\phi$ are given by\n$r\\phi|_{\\mathcal{I}^+}=C u^{-2}\\log u+\\mathcal{O}(u^{-2})$ along future null\ninfinity, $\\phi|_{r=R>2M}=2C\\tau^{-3}\\log\\tau+\\mathcal{O}(\\tau^{-3})$ along\nhypersurfaces of constant $r$, and $\\phi|_{\\mathcal{H}^+}=2Cv^{-3}\\log\nv+\\mathcal{O}(v^{-3})$ along the event horizon. Moreover, the constant $C$ is\ngiven by $C=4M I_0^{(\\mathrm{past})}[\\phi]$, where\n$I_0^{(\\mathrm{past})}[\\phi]:=\\lim_{u\\to -\\infty}\nr^2\\partial_u(r\\phi_{\\ell=0})$ is the past Newman--Penrose constant of $\\phi$\non $\\mathcal{I}^-$. Thus, the precise late-time asymptotics of $\\phi$ are\ncompletely determined by the early-time behaviour of the spherically symmetric\npart of $\\phi$ near $\\mathcal{I}^-$. Similar results are obtained for\npolynomially decaying timelike boundary data. The paper uses methods developed\nby Angelopoulos--Aretakis--Gajic and is essentially self-contained.\n"}
{"abstract": "  Compton scattering of gamma rays propagating in a pair plasma can drive the\nformation of a relativistic electron positron beam. This process is scrutinised\ntheoretically and numerically via particle-in-cell simulations. In addition, we\ndetermine in which conditions the beam can prompt a beam-plasma instability and\nconvert its kinetic energy into magnetic energy. We argue that such conditions\ncan be met at the photosphere radius of bright Gamma Ray Bursts.\n"}
{"abstract": "  With the emergence of Internet-of-Things (IoT) and ever-increasing demand for\nthe newly connected devices, there is a need for more effective storage and\nprocessing paradigms to cope with the data generated from these devices. In\nthis study, we have discussed different paradigms for data processing and\nstorage including Cloud, Fog, and Edge computing models and their suitability\nin integrating with the IoT. Moreover, a detailed discussion on low latency and\nmassive connectivity requirements of future cellular networks in accordance\nwith machine-type communication (MTC) is also presented. Furthermore, the need\nto bring IoT devices to Internet connectivity and a standardized protocol stack\nto regulate the data transmission between these devices is also addressed while\nkeeping in view the resource constraint nature of IoT devices.\n"}
{"abstract": "  Machine learning models that incorporate concept learning as an intermediate\nstep in their decision making process can match the performance of black-box\npredictive models while retaining the ability to explain outcomes in human\nunderstandable terms. However, we demonstrate that the concept representations\nlearned by these models encode information beyond the pre-defined concepts, and\nthat natural mitigation strategies do not fully work, rendering the\ninterpretation of the downstream prediction misleading. We describe the\nmechanism underlying the information leakage and suggest recourse for\nmitigating its effects.\n"}
{"abstract": "  We study the application of iterative first-order methods to the problem of\ncomputing equilibria of large-scale two-player extensive-form games.\nFirst-order methods must typically be instantiated with a regularizer that\nserves as a distance-generating function for the decision sets of the players.\nFor the case of two-player zero-sum games, the state-of-the-art theoretical\nconvergence rate for Nash equilibrium is achieved by using the dilated entropy\nfunction. In this paper, we introduce a new entropy-based distance-generating\nfunction for two-player zero-sum games, and show that this function achieves\nsignificantly better strong convexity properties than the dilated entropy,\nwhile maintaining the same easily-implemented closed-form proximal mapping.\nExtensive numerical simulations show that these superior theoretical properties\ntranslate into better numerical performance as well. We then generalize our new\nentropy distance function, as well as general dilated distance functions, to\nthe scaled extension operator. The scaled extension operator is a way to\nrecursively construct convex sets, which generalizes the decision polytope of\nextensive-form games, as well as the convex polytopes corresponding to\ncorrelated and team equilibria. By instantiating first-order methods with our\nregularizers, we develop the first accelerated first-order methods for\ncomputing correlated equilibra and ex-ante coordinated team equilibria. Our\nmethods have a guaranteed $1/T$ rate of convergence, along with linear-time\nproximal updates.\n"}
{"abstract": "  Retrosynthesis, of which the goal is to find a set of reactants for\nsynthesizing a target product, is an emerging research area of deep learning.\nWhile the existing approaches have shown promising results, they currently lack\nthe ability to consider availability (e.g., stability or purchasability) of the\nreactants or generalize to unseen reaction templates (i.e., chemical reaction\nrules). In this paper, we propose a new approach that mitigates the issues by\nreformulating retrosynthesis into a selection problem of reactants from a\ncandidate set of commercially available molecules. To this end, we design an\nefficient reactant selection framework, named RetCL (retrosynthesis via\ncontrastive learning), for enumerating all of the candidate molecules based on\nselection scores computed by graph neural networks. For learning the score\nfunctions, we also propose a novel contrastive training scheme with hard\nnegative mining. Extensive experiments demonstrate the benefits of the proposed\nselection-based approach. For example, when all 671k reactants in the USPTO\n{database} are given as candidates, our RetCL achieves top-1 exact match\naccuracy of $71.3\\%$ for the USPTO-50k benchmark, while a recent\ntransformer-based approach achieves $59.6\\%$. We also demonstrate that RetCL\ngeneralizes well to unseen templates in various settings in contrast to\ntemplate-based approaches.\n"}
{"abstract": "  In inverse optimal control, an optimal controller is synthesized with respect\nto a meaningful, a posteriori, defined cost functional. Our work illustrates\nthe usefulness of this approach in the control of converter-based power systems\nand networked systems in general, and thereby in designing controllers with\ntopological structure and known optimality properties. In particular, we design\nan inverse optimal feedback controller that stabilizes the phase angles of\nvoltage source-controlled DC/AC converters at an induced steady state with {\\em\nzero} frequency error. The distributed angular droop controller yields active\npower to angle droop behavior at steady state. Moreover, we suggest a practical\nimplementation of the controller and corroborate our results through\nsimulations on a three-converter system and a numerical comparison with\nstandard frequency droop control.\n"}
{"abstract": "  Automatic defect detection is a challenging task because of the variability\nin texture and type of fabric defects. An effective defect detection system\nenables manufacturers to improve the quality of processes and products.\nAutomation across the textile manufacturing systems would reduce fabric wastage\nand increase profitability by saving cost and resources. There are different\ncontemporary research on automatic defect detection systems using image\nprocessing and machine learning techniques. These techniques differ from each\nother based on the manufacturing processes and defect types. Researchers have\nalso been able to establish real-time defect detection system during weaving.\nAlthough, there has been research on patterned fabric defect detection, these\ndefects are related to weaving faults such as holes, and warp and weft defects.\nBut, there has not been any research that is designed to detect defects that\narise during such as spot and print mismatch. This research has fulfilled this\ngap by developing a print fabric database and implementing deep convolutional\nneural network (CNN).\n"}
{"abstract": "  Let $A$ be a square complex matrix and $z$ a complex number. The distance,\nwith respect to the spectral norm, from $A$ to the set of matrices which have\n$z$ as an eigenvalue is less than or equal to the distance from $z$ to the\nspectrum of $A$. If these two distances are equal for a sufficiently large\nfinite set of numbers $z$ which are not in the spectrum of $A$, then the matrix\n$A$ is normal.\n"}
{"abstract": "  Simulating inhomogeneous flows with different characteristic scales in\ndifferent coordinate directions using the collide-and-stream based lattice\nBoltzmann methods (LBM) can be accomplished efficiently using rectangular\nlattice grids. We develop and investigate a new rectangular central moment LBM\nbased on non-orthogonal moment basis (referred to as RC-LBM). The equilibria to\nwhich the central moments relax under collision in this approach are obtained\nfrom matching with those corresponding to the continuous Maxwell distribution.\nA Chapman-Enskog analysis is performed to derive the correction terms to the\nsecond order moment equilibria involving the grid aspect ratio and velocity\ngradients that restores the isotropy of the viscous stress tensor and\neliminates the non-Galilean invariant cubic velocity terms of the resulting\nhydrodynamical equations. A special case of this rectangular formulation\ninvolving the raw moments (referred to as the RNR-LBM) is also constructed. The\nresulting schemes represent a considerable simplification, especially for the\ntransformation matrices and isotropy corrections, and improvement over the\nexisting MRT-LB schemes on rectangular lattice grids that use orthogonal moment\nbasis. Numerical validation study of both the RC-LBM and RNR-LBM for a variety\nof benchmark flow problems are performed that show good accuracy at various\ngrid aspect ratios. The ability of our proposed schemes to simulate flows using\nrelatively lower grid aspect ratios than considered in prior rectangular LB\napproaches is demonstrated. Furthermore, simulations reveal the superior\nstability characteristics of the RC-LBM over RNR-LBM in handling shear flows at\nlower viscosities and/or higher characteristic velocities. In addition,\ncomputational advantages of using our rectangular LB formulation in lieu of\nthat based on the square lattice is shown.\n"}
{"abstract": "  We prove that among the finite dimensional algebras of finite representation\ntype those that are string algebras are precisely the ones that have the\nproperty that the middle term of an arbitrary extension of indecomposable\nmodules has at most two direct factors. On the other hand, we show that\nnon-domestic string algebras are very far from having that property.\n"}
{"abstract": "  We obtain Gabor frame characterisations of modulation spaces defined via a\nclass of translation-modulation invariant Banach spaces of distributions that\nwas recently introduced in $[10]$. We show that these spaces admit an atomic\ndecomposition through Gabor expansions and that they are characterised by\nsummability properties of their Gabor coefficients. Furthermore, we construct a\nlarge space of admissible windows. This generalises several fundamental results\nfor the classical modulation spaces $ M^{p,q}_{w}$. Due to the absence of\nsolidity assumptions on the Banach spaces defining these modulation spaces, the\nmethods used for the spaces $M^{p,q}_{w}$ (or, more generally, in coorbit space\ntheory) fail in our setting and we develop here a new approach based on the\ntwisted convolution.\n"}
{"abstract": "  We study the convergence of the discontinuous Galerkin (DG) method applied to\nthe advection-reaction equation on meshes with reentrant faces. On such meshes,\nthe upwind numerical flux is not smooth, and so the numerical integration of\nthe resulting face terms can only be expected to be first-order accurate.\nDespite this inexact integration, we prove that the DG method converges with\norder $\\mathcal{O}(h^{p+1/2})$, which is the same rate as in the case of exact\nintegration. Consequently, specialized quadrature rules that accurately\nintegrate the non-smooth numerical fluxes are not required for high-order\naccuracy. These results are numerically corroborated on examples of linear\nadvection and discrete ordinates transport equations.\n"}
{"abstract": "  Coulomb interaction, following an inverse-square force-law, quantifies the\namount of force between two stationary and electrically charged particles. The\nlong-range nature of Coulomb interactions poses a major challenge to molecular\ndynamics simulations which are major tools for problems at the nano-/micro-\nscale. Various algorithms are developed to calculate the pairwise Coulomb\ninteractions to a linear scaling but the poor scalability limits the size of\nsimulated systems. Here, we conduct an efficient molecular dynamics algorithm\nwith the random batch Ewald method on all-atom systems where the complete\nFourier components in the Coulomb interaction are replaced by randomly selected\nmini-batches. By simulating the $N$-body systems up to 100 million particles\nusing $10$ thousand CPU cores, we show that this algorithm furnishes $O(N)$\ncomplexity, almost perfect scalability and an order of magnitude faster\ncomputational speed when compared to the existing state-of-the-art algorithms.\nFurther examinations of our algorithm on distinct systems, including pure\nwater, micro-phase-separated electrolyte and protein solution demonstrate that\nthe spatiotemporal information on all time and length scales investigated and\nthermodynamic quantities derived from our algorithm are in perfect agreement\nwith those obtained from the existing algorithms. Therefore, our algorithm\nprovides a breakthrough solution on scalability of computing the Coulomb\ninteraction. It is particularly useful and cost-effective to simulate\nultra-large systems, which was either impossible or very costing to conduct\nusing existing algorithms, thus would benefit the broad community of sciences.\n"}
{"abstract": "  In this letter, we present the photonic nanojet as a phenomenon in a\nstructured light generator system that is implemented to modify the source\nfocal spot size and emission angle. The optical system comprises a microlens\narray that is illuminated by a focused Gaussian beam to generate a structured\npattern in the far-field. By introducing a spheroid with different aspect\nratios in the focus of the Gaussian beam, the source optical characteristics\nchange and a photonic nanojet is generated which will engineer the far-field\ndistribution. To probe the light fields we implement a high-resolution\ninterferometry setup to extract both the phase and intensity at different\nplanes. We both numerically and experimentally demonstrate that the pattern\ndistribution in the far-field can be engineered by photonic nanojet. As an\nexample, we examine prolate, sphere, and ablate geometries. An interesting\nfinding is that depending on the spheroid geometry, a smaller transverse FWHM\nof a photonic nanojet with a higher diverges angle produces an increased\npattern field of view at the same physical size of the optical system.\n"}
{"abstract": "  We investigate the groups of units of one-relator and special inverse\nmonoids. These are inverse monoids which are defined by presentations where all\nthe defining relations are of the form $r=1$. We develop new approaches for\nfinding presentations for the group of units of a special inverse monoid, and\napply these methods to give conditions under which the group admits a\npresentation with the same number of defining relations as the monoid. In\nparticular our results give sufficient conditions for the group of units of a\none-relator inverse monoid to be a one-relator group. When these conditions are\nsatisfied these results give inverse semigroup theoretic analogues of classical\nresults of Adjan for one-relator monoids, and Makanin for special monoids. In\ncontrast, we show that in general these classical results to not hold for\none-relator and special inverse monoids. In particular, we show that there\nexists a one-relator special inverse monoid whose group of units is not a\none-relator group (with respect to any generating set), and we show that there\nexists a finitely presented special inverse monoid whose group of units is not\nfinitely presented.\n"}
{"abstract": "  Nearly complete intersection ideals were introduced by A. Boocher and J.\nSeiner (2018) and defines a special class of monomial ideals in a polynomial\nring. These ideals were used to give a lower bound of the total sum of betti\nnumbers that appear a minimal free resolution of a monomial ideal. In this note\nwe give a graph theoretic classification of nearly complete intersection ideals\ngenerated in degree two. In doing so, we define a novel graph operation (the\ninversion) that is motivated by the definition of this new class of ideals.\n"}
{"abstract": "  Purpose: This study evaluates the effectiveness and impact of automated\norder-based protocol assignment for magnetic resonance imaging (MRI) exams\nusing natural language processing (NLP) and deep learning (DL).\n  Methods: NLP tools were applied to retrospectively process orders from over\n116,000 MRI exams with 200 unique sub-specialized protocols (\"Local\" protocol\nclass). Separate DL models were trained on 70\\% of the processed data for\n\"Local\" protocols as well as 93 American College of Radiology (\"ACR\") protocols\nand 48 \"General\" protocols. The DL Models were assessed in an \"auto-protocoling\n(AP)\" inference mode which returns the top recommendation and in a \"clinical\ndecision support (CDS)\" inference mode which returns up to 10 protocols for\nradiologist review. The accuracy of each protocol recommendation was computed\nand analyzed based on the difference between the normalized output score of the\ncorresponding neural net for the top two recommendations.\n  Results: The top predicted protocol in AP mode was correct for 82.8%, 73.8%,\nand 69.3% of the test cases for \"General\", \"ACR\", and \"Local\" protocol classes,\nrespectively. Higher levels of accuracy over 96% were obtained for all protocol\nclasses in CDS mode. However, at current validation performance levels, the\nproposed models offer modest, positive, financial impact on large-scale imaging\nnetworks.\n  Conclusions: DL-based protocol automation is feasible and can be tuned to\nroute substantial fractions of exams for auto-protocoling, with higher accuracy\nwith more general protocols. Economic analyses of the tested algorithms\nindicate that improved algorithm performance is required to yield a practical\nexam auto-protocoling tool for sub-specialized imaging exams.\n"}
{"abstract": "  INTRODUCTION: Lateral ventricles are reliable and sensitive indicators of\nbrain atrophy and disease progression in behavioral variant frontotemporal\ndementia (bvFTD). We aimed to investigate whether an automated tool using\nventricular features could improve diagnostic accuracy in bvFTD across\nneurodegenerative diseases. METHODS: Using 678 subjects (2750 timepoints),\ndifferences in ventricular features were assessed between bvFTD, normal\ncontrols and other dementia cohorts. RESULTS: Ventricular antero-posterior\nratio (APR) was the only feature that was significantly different and increased\nfaster in bvFTD compared to all other cohorts. We achieved a 10-fold\ncross-validation accuracy of 80% (77% sensitivity, 82% specificity) in\ndifferentiating bvFTD from all other cohorts with ventricular features, and 76%\naccuracy using only the single APR feature. DISCUSSION: APR could be a useful\nand easy-to-implement feature to aid bvFTD diagnosis. We have made our\nventricle feature estimation and bvFTD diagnostic tool publicly available,\nallowing application of our model in other studies.\n"}
{"abstract": "  We study the Duffing equation and its generalizations with polynomial\nnonlinearities. Recently, we have demonstrated that metamorphoses of the\namplitude response curves, computed by asymptotic methods in implicit form as\n$F\\left( \\Omega ,\\ A\\right) =0$, permit prediction of qualitative changes of\ndynamics occurring at singular points of the implicit curve $F\\left(\\Omega ,\\\nA\\right) =0$. In the present work we determine a global structure of singular\npoints of the amplitude profiles computing bifurcation sets, i.e. sets\ncontaining all points in the parameter space for which the amplitude profile\nhas a singular point. We connect our work with independent research on\ntangential points on amplitude profiles, associated with jump phenomena,\ncharacteristic for the Duffing equation. We also show that our techniques can\nbe applied to solutions of form $\\Omega _{\\pm }=f_{\\pm }\\left( A\\right) $,\nobtained within other asymptotic approaches.\n"}
{"abstract": "  We show a partial Boolean function $f$ together with an input $x\\in\nf^{-1}\\left(*\\right)$ such that both $C_{\\bar{0}}\\left(f,x\\right)$ and\n$C_{\\bar{1}}\\left(f,x\\right)$ are at least\n$C\\left(f\\right)^{2-o\\left(1\\right)}$. Due to recent results by Ben-David,\nG\\\"{o}\\\"{o}s, Jain, and Kothari, this result implies several other separations\nin query and communication complexity. For example, it gives a function $f$\nwith $C(f)=\\Omega(deg^{2-o\\left(1\\right)}(f))$ where $C$ and $deg$ denote\ncertificate complexity and polynomial degree of $f$. (This is the first\nimprovement over a separation between $C(f)$ and $deg(f)$ by Kushilevitz and\nNisan in 1995.) Other implications of this result are an improved separation\nbetween sensitivity and polynomial degree, a near-optimal lower bound on\nconondeterministic communication complexity for Clique vs. Independent Set\nproblem and a near-optimal lower bound on complexity of Alon--Saks--Seymour\nproblem in graph theory.\n"}
{"abstract": "  Tactoids are spindle shaped-droplets of a uniaxial nematic phase suspended in\nthe co-existing isotropic phase. They are found in dispersions of a wide\nvariety of elongated colloidal particles, including actin, fd virus, carbon\nnanotubes, vanadium peroxide and chitin nanocrystals. Recent experiments on\ntactoids of chitin nanocrystals in water show that electric fields can very\nstrongly elongate tactoids even though the dielectric properties of the\nco-existing isotropic and nematic phases differ only subtly. We develop a model\nfor partially bipolar tactoids, where the degree of bipolarness of the director\nfield is free to adjust to optimize the sum of the elastic, surface and Coulomb\nenergies of the system. By means of a combination of a scaling analysis and a\nnumerical study, we investigate the elongation and director field's behavior of\nthe tactoids as a function of their size, the strength of the electric field,\nthe surface tension, anchoring strength, the elastic constants and the electric\nsusceptibility anisotropy. We find that tactoids cannot elongate significantly\ndue to an external electric field, unless the director field is bipolar or\nquasi bipolar and is somehow frozen in the field-free configuration. Presuming\nthat this is the case, we find reasonable agreement with experimental data.\n"}
{"abstract": "  Sparse tensors appear frequently in distributed deep learning, either as a\ndirect artifact of the deep neural network's gradients, or as a result of an\nexplicit sparsification process. Existing communication primitives are agnostic\nto the peculiarities of deep learning; consequently, they impose unnecessary\ncommunication overhead. This paper introduces DeepReduce, a versatile framework\nfor the compressed communication of sparse tensors, tailored for distributed\ndeep learning. DeepReduce decomposes sparse tensors in two sets, values and\nindices, and allows both independent and combined compression of these sets. We\nsupport a variety of common compressors, such as Deflate for values, or\nrun-length encoding for indices. We also propose two novel compression schemes\nthat achieve superior results: curve fitting-based for values and bloom\nfilter-based for indices. DeepReduce is orthogonal to existing gradient\nsparsifiers and can be applied in conjunction with them, transparently to the\nend-user, to significantly lower the communication overhead. As proof of\nconcept, we implement our approach on Tensorflow and PyTorch. Our experiments\nwith large real models demonstrate that DeepReduce transmits fewer data and\nimposes lower computational overhead than existing methods, without affecting\nthe training accuracy.\n"}
{"abstract": "  Different properties of dark matter haloes, including growth rate,\nconcentration, interaction history, and spin, correlate with environment in\nunique, scale-dependent ways. While these halo properties are not directly\nobservable, galaxies will inherit their host haloes' correlations with\nenvironment. In this paper, we show how these characteristic environmental\nsignatures allow using measurements of galaxy environment to constrain which\ndark matter halo properties are most tightly connected to observable galaxy\nproperties. We show that different halo properties beyond mass imprint distinct\nscale-dependent signatures in both the galaxy two-point correlation function\nand the distribution of distances to galaxies' kth nearest neighbours, with\nfeatures strong enough to be accessible even with low-resolution (e.g., grism)\nspectroscopy at higher redshifts. As an application, we compute observed\ntwo-point correlation functions for galaxies binned by half-mass radius at z=0\nfrom the Sloan Digital Sky Survey, showing that classic galaxy size models\n(i.e., galaxy size being proportional to halo spin) as well as other recent\nproposals show significant tensions with observational data. We show that the\nagreement with observed clustering can be improved with a simple empirical\nmodel in which galaxy size correlates with halo growth.\n"}
{"abstract": "  SARS-CoV-2, the causative virus of COVID-19 continues to cause an ongoing\nglobal pandemic. Therapeutics are still needed to treat mild and severe\nCOVID-19. Drug repurposing provides an opportunity to deploy drugs for COVID-19\nmore rapidly than developing novel therapeutics. Some existing drugs have shown\npromise for treating COVID-19 in clinical trials. This in silico study uses\nstructural similarity to clinical trial drugs to identify two drugs with\npotential applications to treat early COVID-19. We apply in silico validation\nto suggest a possible mechanism of action for both. Triamcinolone is a\ncorticosteroid structurally similar to Dexamethasone. Gallopamil is a calcium\nchannel blocker structurally similar to Verapamil. We propose that both these\ndrugs could be useful to treat early COVID-19 infection due to the proximity of\ntheir targets within a SARS-CoV-2-induced protein-protein interaction network\nto kinases active in early infection, and the APOA1 protein which is linked to\nthe spread of COVID-19.\n"}
{"abstract": "  For NP-hard combinatorial optimization problems, it is usually difficult to\nfind high-quality solutions in polynomial time. The design of either an exact\nalgorithm or an approximate algorithm for these problems often requires\nsignificantly specialized knowledge. Recently, deep learning methods provide\nnew directions to solve such problems. In this paper, an end-to-end deep\nreinforcement learning framework is proposed to solve this type of\ncombinatorial optimization problems. This framework can be applied to different\nproblems with only slight changes of input (for example, for a traveling\nsalesman problem (TSP), the input is the two-dimensional coordinates of nodes;\nwhile for a capacity-constrained vehicle routing problem (CVRP), the input is\nsimply changed to three-dimensional vectors including the two-dimensional\ncoordinates and the customer demands of nodes), masks and decoder context\nvectors. The proposed framework is aiming to improve the models in literacy in\nterms of the neural network model and the training algorithm. The solution\nquality of TSP and the CVRP up to 100 nodes are significantly improved via our\nframework. Specifically, the average optimality gap is reduced from 4.53\\%\n(reported best \\cite{R22}) to 3.67\\% for TSP with 100 nodes and from 7.34\\%\n(reported best \\cite{R22}) to 6.68\\% for CVRP with 100 nodes when using the\ngreedy decoding strategy. Furthermore, our framework uses about 1/3$\\sim$3/4\ntraining samples compared with other existing learning methods while achieving\nbetter results. The results performed on randomly generated instances and the\nbenchmark instances from TSPLIB and CVRPLIB confirm that our framework has a\nlinear running time on the problem size (number of nodes) during the testing\nphase, and has a good generalization performance from random instance training\nto real-world instance testing.\n"}
{"abstract": "  A nanoparticle (NP) immersed in biological media rapidly forms a corona of\nadsorbed proteins, which later controls the eventual fate of the particle and\nthe route through which adverse outcomes may occur. The composition and\ntimescale for the formation of this corona are both highly dependent on both\nthe NP and its environment. The deposition of proteins on the surface of the NP\nis related to processes of random sequential adsorption and, based on this\nmodel, we develop a rate-equation treatment for the formation of a corona\nrepresented by hard spheres on spherical and cylindrical NPs. We find that the\ngeometry of the NP significantly alters the composition of the corona through a\nprocess independent of the rate constants assumed for adsorption and desorption\nof proteins, with the radius and shape of the NP both influencing the corona.\nMoreover, we demonstrate that in the condition of strong binding such that the\nadsorption is effectively irreversible the corona content reflects the protein\nmobility and concentration in solution rather than their binding affinity.\n"}
{"abstract": "  Travel demand forecasting is an essential part of transportation planning and\nmanagement. The four-step travel model is the traditional and most-common\nprocedure utilized for travel demand forecasting, and many models have been\nproposed in the literature to describe each step separately. However, there is\nstill a lack of a unified modeling framework that can successfully describe the\ncollective choice behavior of travelers interacting with each other at\ndifferent steps. This study uses the free utility model, whose objective\nfunction is mathematically consistent with the Helmholtz free energy in\nphysics, to separately and simultaneously describe travelers' mode,\ndestination, and route choice behaviors. The free utility model's basic\nassumption is that the travelers will trade off the expected utility and\ninformation-processing cost to maximize their own utility. This model provides\nnot only a unified modeling framework for traveler choice behavior, but also\nprovides a new perspective for understanding the user equilibrium model in\ntransportation science and the potential game model in game theory.\n"}
{"abstract": "  Explainable AI (XAI) is a research area whose objective is to increase\ntrustworthiness and to enlighten the hidden mechanism of opaque machine\nlearning techniques. This becomes increasingly important in case such models\nare applied to the chemistry domain, for its potential impact on humans'\nhealth, e.g, toxicity analysis in pharmacology. In this paper, we present a\nnovel approach to tackle explainability of deep graph networks in the context\nof molecule property prediction t asks, named MEG (Molecular Explanation\nGenerator). We generate informative counterfactual explanations for a specific\nprediction under the form of (valid) compounds with high structural similarity\nand different predicted properties. Given a trained DGN, we train a\nreinforcement learning based generator to output counterfactual explanations.\nAt each step, MEG feeds the current candidate counterfactual into the DGN,\ncollects the prediction and uses it to reward the RL agent to guide the\nexploration. Furthermore, we restrict the action space of the agent in order to\nonly keep actions that maintain the molecule in a valid state. We discuss the\nresults showing how the model can convey non-ML experts with key insights into\nthe learning model focus in the neighbourhood of a molecule.\n"}
{"abstract": "  Motivation: Early detection and isolation of COVID-19 patients are essential\nfor successful implementation of mitigation strategies and eventually curbing\nthe disease spread. With a limited number of daily COVID19 tests performed in\nevery country, simulating the COVID-19 spread along with the potential effect\nof each mitigation strategy currently remains one of the most effective ways in\nmanaging the healthcare system and guiding policy-makers. We introduce\nCOVIDHunter, a flexible and accurate COVID-19 outbreak simulation model that\nevaluates the current mitigation measures that are applied to a region and\nprovides suggestions on what strength the upcoming mitigation measure should\nbe. The key idea of COVIDHunter is to quantify the spread of COVID-19 in a\ngeographical region by simulating the average number of new infections caused\nby an infected person considering the effect of external factors, such as\nenvironmental conditions (e.g., climate, temperature, humidity) and mitigation\nmeasures.\n  Results: Using Switzerland as a case study, COVIDHunter estimates that the\npolicy-makers need to keep the current mitigation measures for at least 30 days\nto prevent demand from quickly exceeding existing hospital capacity. Relaxing\nthe mitigation measures by 50% for 30 days increases both the daily capacity\nneed for hospital beds and daily number of deaths exponentially by an average\nof 23.8x, who may occupy ICU beds and ventilators for a period of time. Unlike\nexisting models, the COVIDHunter model accurately monitors and predicts the\ndaily number of cases, hospitalizations, and deaths due to COVID-19. Our model\nis flexible to configure and simple to modify for modeling different scenarios\nunder different environmental conditions and mitigation measures.\n  Availability: https://github.com/CMU-SAFARI/COVIDHunter\n"}
{"abstract": "  Masses of classical Cepheids of 3 to 11 M$\\odot$ are predicted by theory but\nthose measured, clump between 3.6 and 5 M$\\odot$. As a result, their\nmass-luminosity relation is poorly constrained, impeding our understanding of\nbasic stellar physics and the Leavitt Law. All Cepheid masses come from the\nanalysis of 11 binary systems, including only 5 double-lined and well-suited\nfor accurate dynamical mass determination. We present a project to analyze a\nnew, numerous group of Cepheids in double-lined binary (SB2) systems to provide\nmass determinations in a wide mass interval and study their evolution. We\nanalyze a sample of 41 candidate binary LMC Cepheids spread along the P-L\nrelation, that are likely accompanied by luminous red giants, and present\nindirect and direct indicators of their binarity. In a spectroscopic study of a\nsubsample of 18 brightest candidates, for 16 we detected lines of two\ncomponents in the spectra, already quadrupling the number of Cepheids in SB2\nsystems. Observations of the whole sample may thus lead to quadrupling all the\nCepheid mass estimates available now. For the majority of our candidates,\nerratic intrinsic period changes dominate over the light travel-time effect due\nto binarity. However, the latter may explain the periodic phase modulation for\n4 Cepheids. Our project paves the way for future accurate dynamical mass\ndeterminations of Cepheids in the LMC, Milky Way, and other galaxies, which\nwill potentially increase the number of known Cepheid masses even 10-fold,\nhugely improving our knowledge about these important stars.\n"}
{"abstract": "  In the context of the ClearMind project, we measured the scintillating\nproperties, as induced from from gamma ray interactions, of today available\nPbWO4 crystal. We measured scintillation s yields and time constants by\nmeasuring the signal shape measured on a fast photo-multiplier and\ndeconvoluting it from the instrumental effects. For the doped crystals at room\ntemperature, we measured a fast scintillation component, with time constants of\n2 ns, 55 percent of the total light yield, and a slow component of 6 ns. We\nobserve a significant increase of the light yield for the slow component when\nthe temperature decreases and simultaneous increase of the time constants, but\nno increase in the fast component light yield. Our measurements reproduce the\nmain qualitative features of PbWO4 crystals quoted in the literature.\nQuantitatively though, we measured significantly shorter time constants and\nlarger light yields. This is explained by a rigorous treatment of the\ninstrumental contributions in the measurements. Results are discussed and\nprospect for future developments, tailored for the ClearMind project, are\npresented.\n"}
{"abstract": "  Many previous audio-visual voice-related works focus on speech, ignoring the\nsinging voice in the growing number of musical video streams on the Internet.\nFor processing diverse musical video data, voice activity detection is a\nnecessary step. This paper attempts to detect the speech and singing voices of\ntarget performers in musical video streams using audiovisual information. To\nintegrate information of audio and visual modalities, a multi-branch network is\nproposed to learn audio and image representations, and the representations are\nfused by attention based on semantic similarity to shape the acoustic\nrepresentations through the probability of anchor vocalization. Experiments\nshow the proposed audio-visual multi-branch network far outperforms the\naudio-only model in challenging acoustic environments, indicating the\ncross-modal information fusion based on semantic correlation is sensible and\nsuccessful.\n"}
{"abstract": "  We present Supervision by Registration and Triangulation (SRT), an\nunsupervised approach that utilizes unlabeled multi-view video to improve the\naccuracy and precision of landmark detectors. Being able to utilize unlabeled\ndata enables our detectors to learn from massive amounts of unlabeled data\nfreely available and not be limited by the quality and quantity of manual human\nannotations. To utilize unlabeled data, there are two key observations: (1) the\ndetections of the same landmark in adjacent frames should be coherent with\nregistration, i.e., optical flow. (2) the detections of the same landmark in\nmultiple synchronized and geometrically calibrated views should correspond to a\nsingle 3D point, i.e., multi-view consistency. Registration and multi-view\nconsistency are sources of supervision that do not require manual labeling,\nthus it can be leveraged to augment existing training data during detector\ntraining. End-to-end training is made possible by differentiable registration\nand 3D triangulation modules. Experiments with 11 datasets and a newly proposed\nmetric to measure precision demonstrate accuracy and precision improvements in\nlandmark detection on both images and video. Code is available at\nhttps://github.com/D-X-Y/landmark-detection.\n"}
{"abstract": "  The quantum singular value transformation is a powerful quantum algorithm\nthat allows one to apply a polynomial transformation to the singular values of\na matrix that is embedded as a block of a unitary transformation. This paper\nshows how to perform the quantum singular value transformation for a matrix\nthat can be embedded as a block of a Hamiltonian. The transformation can be\nimplemented in a purely Hamiltonian context by the alternating application of\nHamiltonians for chosen intervals: it is an example of the Quantum Alternating\nOperator Ansatz (generalized QAOA). We also show how to use the Hamiltonian\nquantum singular value transformation to perform inverse block encoding to\nimplement a unitary of which a given Hamiltonian is a block. Inverse block\nencoding leads to novel procedures for matrix multiplication and for solving\ndifferential equations on quantum information processors in a purely\nHamiltonian fashion.\n"}
{"abstract": "  A coming resurgence of super heavy-lift launch vehicles has precipitated an\nimmense interest in the future of crewed spaceflight and even future\ncolonisation efforts. While it is true that a bright future awaits this sector,\ndriven by commercial ventures and the reignited interest of old space-faring\nnations, and the joining of new ones, little of this attention has been\nreserved for the science-centric applications of these launchers. The Arcanum\nmission is a proposal to use these vehicles to deliver an L-class observatory\ninto a highly eccentric orbit around Neptune, with a wide-ranging suite of\nscience goals and instrumentation tackling Solar System science, planetary\nscience, Kuiper Belt Objects and exoplanet systems.\n"}
{"abstract": "  Reconstructing dynamic, time-varying scenes with computed tomography (4D-CT)\nis a challenging and ill-posed problem common to industrial and medical\nsettings. Existing 4D-CT reconstructions are designed for sparse sampling\nschemes that require fast CT scanners to capture multiple, rapid revolutions\naround the scene in order to generate high quality results. However, if the\nscene is moving too fast, then the sampling occurs along a limited view and is\ndifficult to reconstruct due to spatiotemporal ambiguities. In this work, we\ndesign a reconstruction pipeline using implicit neural representations coupled\nwith a novel parametric motion field warping to perform limited view 4D-CT\nreconstruction of rapidly deforming scenes. Importantly, we utilize a\ndifferentiable analysis-by-synthesis approach to compare with captured x-ray\nsinogram data in a self-supervised fashion. Thus, our resulting optimization\nmethod requires no training data to reconstruct the scene. We demonstrate that\nour proposed system robustly reconstructs scenes containing deformable and\nperiodic motion and validate against state-of-the-art baselines. Further, we\ndemonstrate an ability to reconstruct continuous spatiotemporal representations\nof our scenes and upsample them to arbitrary volumes and frame rates\npost-optimization. This research opens a new avenue for implicit neural\nrepresentations in computed tomography reconstruction in general.\n"}
{"abstract": "  A cornerstone of quantum mechanics is the characterisation of symmetries\nprovided by Wigner's theorem. Wigner's theorem establishes that every symmetry\nof the quantum state space must be either a unitary transformation, or an\nantiunitary transformation. Here we extend Wigner's theorem from quantum states\nto quantum evolutions, including both the deterministic evolution associated to\nthe dynamics of closed systems, and the stochastic evolutions associated to the\noutcomes of quantum measurements. We prove that every symmetry of the space of\nquantum evolutions can be decomposed into two state space symmetries that are\neither both unitary or both antiunitary. Building on this result, we show that\nit is impossible to extend the time reversal symmetry of unitary quantum\ndynamics to a symmetry of the full set of quantum evolutions. Our no-go theorem\nimplies that any time symmetric formulation of quantum theory must either\nrestrict the set of the allowed evolutions, or modify the operational\ninterpretation of quantum states and processes. Here we propose a time\nsymmetric formulation of quantum theory where the allowed quantum evolutions\nare restricted to a suitable set, which includes both unitary evolution and\nprojective measurements, but excludes the deterministic preparation of pure\nstates. The standard operational formulation of quantum theory can be retrieved\nfrom this time symmetric version by introducing an operation of conditioning on\nthe outcomes of past experiments.\n"}
{"abstract": "  Extending Aubin's construction of metrics with constant negative scalar\ncurvature, we prove that every $n$-dimensional closed manifold admits a\nRiemannian metric with constant negative scalar-Weyl curvature, that is\n$R+t|W|, t\\in\\mathbb{R}$. In particular, there are no topological obstructions\nfor metrics with $\\varepsilon$-pinched Weyl curvature and negative scalar\ncurvature.\n"}
{"abstract": "  We propose a new general training technique for attention mechanisms based on\nvirtual adversarial training (VAT). VAT can compute adversarial perturbations\nfrom unlabeled data in a semi-supervised setting for the attention mechanisms\nthat have been reported in previous studies to be vulnerable to perturbations.\nEmpirical experiments reveal that our technique (1) provides significantly\nbetter prediction performance compared to not only conventional adversarial\ntraining-based techniques but also VAT-based techniques in a semi-supervised\nsetting, (2) demonstrates a stronger correlation with the word importance and\nbetter agreement with evidence provided by humans, and (3) gains in performance\nwith increasing amounts of unlabeled data.\n"}
{"abstract": "  The $q$-Onsager algebra $O_q$ has a presentation involving two generators\n$W_0$, $W_1$ and two relations, called the $q$-Dolan/Grady relations. The\nalternating central extension $\\mathcal O_q$ has a presentation involving the\nalternating generators $\\lbrace \\mathcal W_{-k}\\rbrace_{k=0}^\\infty$, $\\lbrace\n\\mathcal W_{k+1}\\rbrace_{k=0}^\\infty$, $ \\lbrace \\mathcal\nG_{k+1}\\rbrace_{k=0}^\\infty$, $\\lbrace \\mathcal {\\tilde\nG}_{k+1}\\rbrace_{k=0}^\\infty$ and a large number of relations. Let $\\langle\n\\mathcal W_0, \\mathcal W_1 \\rangle$ denote the subalgebra of $\\mathcal O_q$\ngenerated by $\\mathcal W_0$, $\\mathcal W_1$. It is known that there exists an\nalgebra isomorphism $O_q \\to \\langle \\mathcal W_0, \\mathcal W_1 \\rangle$ that\nsends $W_0\\mapsto \\mathcal W_0$ and $W_1 \\mapsto \\mathcal W_1$. It is known\nthat the center $\\mathcal Z$ of $\\mathcal O_q$ is isomorphic to a polynomial\nalgebra in countably many variables.\n  It is known that the multiplication map $\\langle \\mathcal W_0, \\mathcal W_1\n\\rangle \\otimes \\mathcal Z \\to \\mathcal O_q$, $ w \\otimes z \\mapsto wz$ is an\nisomorphism of algebras. We call this isomorphism the standard tensor product\nfactorization of $\\mathcal O_q$. In the study of $\\mathcal O_q$ there are two\nnatural points of view: we can start with the alternating generators, or we can\nstart with the standard tensor product factorization. It is not obvious how\nthese two points of view are related. The goal of the paper is to describe this\nrelationship. We give seven main results; the principal one is an attractive\nfactorization of the generating function for some algebraically independent\nelements that generate $\\mathcal Z$.\n"}
{"abstract": "  Triangle counting is a fundamental technique in network analysis, that has\nreceived much attention in various input models. The vast majority of triangle\ncounting algorithms are targeted to static graphs. Yet, many real-world graphs\nare directed and temporal, where edges come with timestamps. Temporal triangles\nyield much more information, since they account for both the graph topology and\nthe timestamps.\n  Temporal triangle counting has seen a few recent results, but there are\nvarying definitions of temporal triangles. In all cases, temporal triangle\npatterns enforce constraints on the time interval between edges (in the\ntriangle). We define a general notion $(\\delta_{1,3}, \\delta_{1,2},\n\\delta_{2,3})$-temporal triangles that allows for separate time constraints for\nall pairs of edges.\n  Our main result is a new algorithm, DOTTT (Degeneracy Oriented Temporal\nTriangle Totaler), that exactly counts all directed variants of $(\\delta_{1,3},\n\\delta_{1,2}, \\delta_{2,3})$-temporal triangles. Using the classic idea of\ndegeneracy ordering with careful combinatorial arguments, we can prove that\nDOTTT runs in $O(m\\kappa\\log m)$ time, where $m$ is the number of (temporal)\nedges and $\\kappa$ is the graph degeneracy (max core number). Up to log\nfactors, this matches the running time of the best static triangle counters.\nMoreover, this running time is better than existing.\n  DOTTT has excellent practical behavior and runs twice as fast as existing\nstate-of-the-art temporal triangle counters (and is also more general). For\nexample, DOTTT computes all types of temporal queries in Bitcoin temporal\nnetwork with half a billion edges in less than an hour on a commodity machine.\n"}
{"abstract": "  Recent works have shown that deep neural networks benefit from multi-task\nlearning by learning a shared representation across several related tasks.\nHowever, performance of such systems depend on relative weighting between\nvarious losses involved during training. Prior works on loss weighting schemes\nassume that instances are equally easy or hard for all tasks. In order to break\nthis assumption, we let the training process dictate the optimal weighting of\ntasks for every instance in the dataset. More specifically, we equip every\ninstance in the dataset with a set of learnable parameters (instance-level task\nparameters) where the cardinality is equal to the number of tasks learned by\nthe model. These parameters model the weighting of each task for an instance.\nThey are updated by gradient descent and do not require hand-crafted rules. We\nconduct extensive experiments on SURREAL and CityScapes datasets, for human\nshape and pose estimation, depth estimation and semantic segmentation tasks. In\nthese tasks, our approach outperforms recent dynamic loss weighting approaches,\ne.g. reducing surface estimation errors by 8.97% on SURREAL. When applied to\ndatasets where one or more tasks can have noisy annotations, the proposed\nmethod learns to prioritize learning from clean labels for a given task, e.g.\nreducing surface estimation errors by up to 60%. We also show that we can\nreliably detect corrupt labels for a given task as a by-product from learned\ninstance-level task parameters.\n"}
{"abstract": "  We investigate a class of reducible yet indecomposable modules over the $N=2$\nsuperconformal algebras. These so-called staggered modules exhibit a\nnon-diagonalisable action of the Virasoro mode $L_{0}$. Using recent results on\nthe coset construction of $N=2$ minimal models, we explicitly construct such\nmodules for central charges $c = -1$ and $c = -6$. We also describe\nspectral-flow orbits and symmetries of the families of staggered modules which\narise via the coset.\n"}
{"abstract": "  Building on previous developments, we show that the Diagrammatic Monte Carlo\ntechnique allows to compute finite temperature response functions directly on\nthe real-frequency axis within any field-theoretical formulation of the\ninteracting fermion problem. There are no limitations on the type and nature of\nthe system's action or whether partial summation and self-consistent treatment\nof certain diagram classes are used. In particular, by eliminating the need for\nnumerical analytic continuation from a Matsubara representation, our scheme\nallows to study spectral densities of arbitrary complexity with controlled\naccuracy in models with frequency-dependent effective interactions. For\nillustrative purposes we consider the problem of the plasmon line-width in a\nhomogeneous electron gas (jellium).\n"}
{"abstract": "  I generalize a theorem of Predd, et al.~(2009) on domination and strictly\nproper scoring rules to the case of non-additive scoring rules.\n"}
{"abstract": "  We study the problem of online network change point detection. In this\nsetting, a collection of independent Bernoulli networks is collected\nsequentially, and the underlying distributions change when a change point\noccurs. The goal is to detect the change point as quickly as possible, if it\nexists, subject to a constraint on the number or probability of false alarms.\nIn this paper, on the detection delay, we establish a minimax lower bound and\ntwo upper bounds based on NP-hard algorithms and polynomial-time algorithms,\ni.e., \\[ \\mbox{detection delay} \\begin{cases} \\gtrsim \\log(1/\\alpha)\n\\frac{\\max\\{r^2/n, \\, 1\\}}{\\kappa_0^2 n \\rho},\\\\ \\lesssim \\log(\\Delta/\\alpha)\n\\frac{\\max\\{r^2/n, \\, \\log(r)\\}}{\\kappa_0^2 n \\rho}, & \\mbox{with NP-hard\nalgorithms},\\\\ \\lesssim \\log(\\Delta/\\alpha) \\frac{r}{\\kappa_0^2 n \\rho}, &\n\\mbox{with polynomial-time algorithms}, \\end{cases} \\] where $\\kappa_0, n,\n\\rho, r$ and $\\alpha$ are the normalised jump size, network size, entrywise\nsparsity, rank sparsity and the overall Type-I error upper bound. All the model\nparameters are allowed to vary as $\\Delta$, the location of the change point,\ndiverges. The polynomial-time algorithms are novel procedures that we propose\nin this paper, designed for quick detection under two different forms of Type-I\nerror control. The first is based on controlling the overall probability of a\nfalse alarm when there are no change points, and the second is based on\nspecifying a lower bound on the expected time of the first false alarm.\nExtensive experiments show that, under different scenarios and the\naforementioned forms of Type-I error control, our proposed approaches\noutperform state-of-the-art methods.\n"}
{"abstract": "  Over the years researchers have studied the evolution of Electrocardiogram\n(ECG) and the complex classification of cardiovascular diseases. This review\nfocuses on the evolution of the ECG, and covers the most recent signal\nprocessing schemes with milestones over last 150 years in a systematic manner.\nDevelopment phases of ECG, ECG leads, portable ECG monitors, Signal Processing\nSchemes and the Complex Transformations are discussed. It also provides\nrecommendations for the inclusion of certain important points based on the\nreview.\n"}
{"abstract": "  The density ratio model (DRM) provides a flexible and useful platform for\ncombining information from multiple sources. In this paper, we consider\nstatistical inference under two-sample DRMs with additional parameters defined\nthrough and/or additional auxiliary information expressed as estimating\nequations. We examine the asymptotic properties of the maximum empirical\nlikelihood estimators (MELEs) of the unknown parameters in the DRMs and/or\ndefined through estimating equations, and establish the chi-square limiting\ndistributions for the empirical likelihood ratio (ELR) statistics. We show that\nthe asymptotic variance of the MELEs of the unknown parameters does not\ndecrease if one estimating equation is dropped. Similar properties are obtained\nfor inferences on the cumulative distribution function and quantiles of each of\nthe populations involved. We also propose an ELR test for the validity and\nusefulness of the auxiliary information. Simulation studies show that correctly\nspecified estimating equations for the auxiliary information result in more\nefficient estimators and shorter confidence intervals. Two real-data examples\nare used for illustrations.\n"}
{"abstract": "  Denote by $Sof(G)$ the space of sofic representations of a countable group\n$G$. This space is known by a result of the second author, to have a\nconvex-like structure. We show that, in this space, minimal faces are extreme\npoints. We then construct uncountable many extreme points for\n$Sof(\\mathbb{F}_2)$ and show that there exists a decreasing chain of closed\nfaces with empty intersection. Finally we construct a strangely looking sofic\nrepresentation in $Sof(\\mathbb{F}_2)$ that we believe it is outside of the\nclosure of the convex hull of extreme points.\n"}
{"abstract": "  Decentralized training of deep learning models is a key element for enabling\ndata privacy and on-device learning over networks. In realistic learning\nscenarios, the presence of heterogeneity across different clients' local\ndatasets poses an optimization challenge and may severely deteriorate the\ngeneralization performance. In this paper, we investigate and identify the\nlimitation of several decentralized optimization algorithms for different\ndegrees of data heterogeneity. We propose a novel momentum-based method to\nmitigate this decentralized training difficulty. We show in extensive empirical\nexperiments on various CV/NLP datasets (CIFAR-10, ImageNet, and AG News) and\nseveral network topologies (Ring and Social Network) that our method is much\nmore robust to the heterogeneity of clients' data than other existing methods,\nby a significant improvement in test performance ($1\\% \\!-\\! 20\\%$). Our code\nis publicly available.\n"}
{"abstract": "  The phase-field method has been established as a de facto standard for\nsimulating the microstructural evolution of materials. In quantitative modeling\nthe assessment and compilation of thermodynamic/kinetic data is largely\ndominated by the CALPHAD approach, which has produced a large set of\nexperimentally and computationally generated Gibbs free energy and atomic\nmobility data in a standardized format: the thermodynamic database (TDB) file\nformat. Harnessing this data for the purpose of phase-field modeling is an\nongoing effort encompassing a wide variety of approaches. In this paper, we aim\nto directly link CALPHAD data to the phase-field method, without intermediate\nfitting or interpolation steps. We introduce a model based on the\nKim-Kim-Suzuki (KKS) approach. This model includes sublattice site fractions\nand can directly utilize data from TDB files. Using this approach, we\ndemonstrate the model on the U-Zr and Mo-Ni-Re systems.\n"}
{"abstract": "  General Relativity (GR) represents the most recent theory of gravity, on\nwhich all modern astrophysics is based, including some of the most astonishing\nresults of physics research. Nevertheless, its study is limited to university\ncourses, while being ignored at high school level. To introduceGR in high\nschool one of the approaches that can be used is the so-called rubber sheet\nanalogy, i.e. comparing the space-time to a rubber sheet which deforms under a\nweight. In this paper we analyze the efficacy of an activity for high school\nstudents held at the Department of Mathematics and Physics of Roma Tre\nUniversity that adopts the rubber sheet analogy to address several topics\nrelated to gravity. We present the results of the questionnaires we\nadministered to investigate the understanding of the topics treated to over 150\nItalian high school students who participated in this activity.\n"}
{"abstract": "  We study objective robustness failures, a type of out-of-distribution\nrobustness failure in reinforcement learning (RL). Objective robustness\nfailures occur when an RL agent retains its capabilities out-of-distribution\nyet pursues the wrong objective. This kind of failure presents different risks\nthan the robustness problems usually considered in the literature, since it\ninvolves agents that leverage their capabilities to pursue the wrong objective\nrather than simply failing to do anything useful. We provide the first explicit\nempirical demonstrations of objective robustness failures and present a partial\ncharacterization of its causes.\n"}
{"abstract": "  X-ray detectors made of crystalline silicon have several advantages including\nlow dark currents, fast charge collection and high energy resolution. For\nhigh-energy x-rays, however, silicon suffers from its low atomic number, which\nmight result in low detection efficiency, as well as low energy and spatial\nresolution due to Compton scattering. We have used a monte-carlo model to\ninvestigate the feasibility of a detector for pediatric CT with 30 to 40 mm of\nsilicon using x-ray spectra ranging from 80 to 140 kVp. A detection efficiency\nof 0.74 was found at 80 kVp, provided the noise threshold could be set low.\nScattered photons were efficiently blocked by a thin metal shielding between\nthe detector units, and Compton scattering in the detector could be well\nseparated from photo absorption at 80 kVp. Hence, the detector is feasible at\nlow acceleration voltages, which is also suitable for pediatric imaging. We\nconclude that silicon detectors may be an alternative to other designs for this\nspecial case.\n"}
{"abstract": "  Multi-objective controller synthesis concerns the problem of computing an\noptimal controller subject to multiple (possibly conflicting) objective\nproperties. The relative importance of objectives is often specified by human\ndecision-makers. However, there is inherent uncertainty in human preferences\n(e.g., due to different preference elicitation methods). In this paper, we\nformalize the notion of uncertain human preferences and present a novel\napproach that accounts for uncertain human preferences in the multi-objective\ncontroller synthesis for Markov decision processes (MDPs). Our approach is\nbased on mixed-integer linear programming (MILP) and synthesizes a sound,\noptimally permissive multi-strategy with respect to a multi-objective property\nand an uncertain set of human preferences. Experimental results on a range of\nlarge case studies show that our MILP-based approach is feasible and scalable\nto synthesize sound, optimally permissive multi-strategies with varying MDP\nmodel sizes and uncertainty levels of human preferences. Evaluation via an\nonline user study also demonstrates the quality and benefits of synthesized\n(multi-)strategies.\n"}
{"abstract": "  In this work, we study magnetic field effects on neutron star matter\ncontaining the baryon octet and additional heavier spin 3/2 baryons (the\n$\\Delta$'s). We make use of two different relativistic hadronic models that\ncontain an additional vector-isovector self interaction for the mesons: one\nversion of a relativistic mean field (RMF) model and the Chiral Mean Field\n(CMF) model. We find that both the additional interaction and a strong magnetic\nfield enhance the $\\Delta$ baryon population in dense matter, while decreasing\nthe relative density of hyperons. At the same time that the vector-isovector\nmeson interaction modifies neutron-star masses very little ($<0.1~M_\\odot$), it\ndecreases their radii considerably, allowing both models to be in better\nagreement with observations. Together, these features indicate that magnetic\nneutron stars are likely to contain $\\Delta$ baryons in their interior.\n"}
{"abstract": "  Proof-of-Work (PoW) is a fundamental underlying technology behind most major\nblockchain cryptocurrencies. It has been previously pointed out that quantum\ndevices provide a computational advantage in performing PoW in the context of\nBitcoin. Here we make the case that this quantum advantage extends not only to\nall existing PoW mechanisms, but to any possible PoW as well. This has strong\nconsequences regarding both quantum-based attacks on the integrity of the\nentirety of the blockchain, as well as more legitimate uses of quantum\ncomputation for the purpose of mining Bitcoin and other cryptocurrencies. For\nthe first case, we estimate when these quantum attacks will become feasible,\nfor various cryptocurrencies, and discuss the impact of such attacks. For the\nlatter, we derive a precise formula to calculate the economic incentive for\nswitching to quantum-based cryptocurrency miners. Using this formula, we\nanalyze several test scenarios, and conclude that investing in quantum hardware\nfor cryptocurrency mining has the potential to pay off immensely.\n"}
{"abstract": "  In this article we study a \\emph{non-directed} polymer model in dimension\n$d\\ge 2$: we consider a simple symmetric random walk on $\\mathbb{Z}^d$ which\ninteracts with a random environment, represented by i.i.d. random variables\n$(\\omega_x)_{x\\in \\mathbb{Z}^d}$. The model consists in modifying the law of\nthe random walk up to time (or length) $N$ by the exponential of $\\sum_{x\\in\n\\mathcal{R}_N}\\beta (\\omega_x-h)$ where $\\mathcal{R}_N$ is the range of the\nwalk, \\textit{i.e.} the set of visited sites up to time $N$, and $\\beta\\geq\n0,\\, h\\in \\mathbb{R}$ are two parameters. We study the behavior of the model in\na weak-coupling regime, that is taking $\\beta:=\\beta_N$ vanishing as the length\n$N$ goes to infinity, and in the case where the random variables $\\omega$ have\na heavy tail with exponent $\\alpha\\in (0,d)$. We are able to obtain precisely\nthe behavior of polymer trajectories under all possible weak-coupling regimes\n$\\beta_N = \\hat \\beta N^{-\\gamma}$ with $\\gamma \\geq 0$: we find the correct\ntransversal fluctuation exponent $\\xi$ for the polymer (it depends on $\\alpha$\nand $\\gamma$) and we give the limiting distribution of the rescaled\nlog-partition function. This extends existing works to the non-directed case\nand to higher dimensions.\n"}
{"abstract": "  In this paper, we aim to study the main properties and luminosity function\n(LF) of the [OII] emitters detected in the OTELO survey in order to\ncharacterise the star formation processes in low-mass galaxies at $z\\sim1.43$\nand to constrain the faint-end of the LF.\n  Here, we describe the selection method and analysis of the emitters obtained\nfrom narrow-band scanning techniques. In addition, we present several relevant\nproperties of the emitters and discuss the selection biases and uncertainties\nin the determination of the LF and the star formation rate density (SFRD).\n  We confirmed a total of 60 sources from a preliminary list of 332 candidates\nas [OII] emitters. Approximately 93% of the emitters have masses in the range\nof $10^{8}<M_{*}/{\\rm M_{\\odot}}<10^{9}$. All of our emitters are classified as\nlate-type galaxies, with a lower value of $(u-v)$\\, when compared with the rest\nof the emitters of the OTELO survey. We find that the cosmic variance strongly\naffects the normalisation ($\\phi^*$) of the LF and explains the discrepancy of\nour results when compared with those obtained from surveys of much larger\nvolumes. However, we are able to determine the faint-end slope of the LF,\nnamely, $\\alpha=-1.42\\pm0.06$, by sampling the LF down to $\\sim1$\\,dex lower\nthan in previous works. We present our calculation of the SFRD of our sample\nand compare it to the value obtained in previous studies from the literature.\n"}
{"abstract": "  We report that spin supercurrents can be induced in diffusive SFS Josephson\njunctions without any magnetic misalignment or intrinsic spin orbit coupling.\nInstead, the pathway to spin triplet generation is provided via geometric\ncurvature, and results in a long ranged Josephson effect. In addition, the\ncurvature is shown to induce a dynamically tunable $0-\\pi$ transition in the\njunction. We provide the analytic framework and discuss potential experimental\nand innovation implications.\n"}
{"abstract": "  In this paper, we open up new avenues for visual servoing systems built upon\nthe Path Integral (PI) optimal control theory, in which the non-linear partial\ndifferential equation (PDE) can be transformed into an expectation over all\npossible trajectories using the Feynman-Kac (FK) lemma. More precisely, we\npropose an MPPI-VS control strategy, a real-time and inversion-free control\nstrategy on the basis of sampling-based model predictive control (namely, Model\nPredictive Path Integral (MPPI) control) algorithm, for both image-based, 3D\npoint, and position-based visual servoing techniques, taking into account the\nsystem constraints (such as visibility, 3D, and control constraints) and\nparametric uncertainties associated with the robot and camera models as well as\nmeasurement noise. Contrary to classical visual servoing control schemes, our\ncontrol strategy directly utilizes the approximation of the interaction matrix,\nwithout the need for estimating the interaction matrix inversion or performing\nthe pseudo-inversion. We validate the MPPI-VS control strategy as well as the\nclassical control schemes on a 6-DoF Cartesian robot with an eye-in-hand camera\nbased on the utilization of four points in the image plane as visual features.\nTo better assess and demonstrate the robustness and potential advantages of our\nproposed control strategy compared to classical schemes, intensive simulations\nunder various operating conditions are carried out and then discussed. The\nobtained results demonstrate the effectiveness and capability of the proposed\nscheme in coping easily with the system constraints, as well as its robustness\nin the presence of large errors in camera parameters and measurements.\n"}
{"abstract": "  Recent improvements in deep learning models and their practical applications\nhave raised concerns about the robustness of these models against adversarial\nexamples. Adversarial training (AT) has been shown effective to reach a robust\nmodel against the attack that is used during training. However, it usually\nfails against other attacks, i.e. the model overfits to the training attack\nscheme. In this paper, we propose a simple modification to the AT that\nmitigates the mentioned issue. More specifically, we minimize the perturbation\n$\\ell_p$ norm while maximizing the classification loss in the Lagrangian form.\nWe argue that crafting adversarial examples based on this scheme results in\nenhanced attack generalization in the learned model. We compare our final model\nrobust accuracy against attacks that were not used during training to closely\nrelated state-of-the-art AT methods. This comparison demonstrates that our\naverage robust accuracy against unseen attacks is 5.9% higher in the CIFAR-10\ndataset and is 3.2% higher in the ImageNet-100 dataset than corresponding\nstate-of-the-art methods. We also demonstrate that our attack is faster than\nother attack schemes that are designed for unseen attack generalization, and\nconclude that it is feasible for large-scale datasets.\n"}
{"abstract": "  At high Reynolds number, the interaction between two vortex tubes leads to\nintense velocity gradients, which are at the heart of fluid turbulence. This\nvorticity amplification comes about through two different instability\nmechanisms of the initial vortex tubes, assumed anti-parallel and with a mirror\nplane of symmetry. At moderate Reynolds number, the tubes destabilize via a\nCrow instability, with the nonlinear development leading to strong flattening\nof the cores into thin sheets. These sheets then break down into filaments\nwhich can repeat the process. At higher Reynolds number, the instability\nproceeds via the elliptical instability, producing vortex tubes that are\nperpendicular to the original tube directions. In this work, we demonstrate\nthat these same transition between Crow and Elliptical instability occurs at\nmoderate Reynolds number when we vary the initial angle $\\beta$ between two\nstraight vortex tubes. We demonstrate that when the angle between the two tubes\nis close to $\\pi/2$, the interaction between tubes leads to the formation of\nthin vortex sheets. The subsequent breakdown of these sheets involves a\ntwisting of the paired sheets, followed by the appearance of a localized cloud\nof small scale vortex structures. At smaller values of the angle $\\beta$\nbetween the two tubes, the breakdown mechanism changes to an elliptic\ncascade-like mechanism. Whereas the interaction of two vortices depends on the\ninitial condition, the rapid formation of fine-scales vortex structures appears\nto be a robust feature, possibly universal at very high Reynolds numbers.\n"}
{"abstract": "  Emulator is widely used to build dynamic analysis frameworks due to its\nfine-grained tracing capability, full system monitoring functionality, and\nscalability of running on different operating systemsand architectures.\nHowever, whether the emulator is consistent with real devices is unknown. To\nunderstand this problem, we aim to automatically locate inconsistent\ninstructions, which behave differently between emulators and real devices.\n  We target ARM architecture, which provides machine readable specification.\nBased on the specification, we propose a test case generator by designing and\nimplementing the first symbolic execution engine for ARM architecture\nspecification language (ASL). We generate 2,774,649 representative instruction\nstreams and conduct differential testing with these instruction streams between\nfour ARM real devices in different architecture versions (i.e., ARMv5, ARMv6,\nARMv7-a, and ARMv8-a) and the state-of-the-art emulators (i.e., QEMU). We\nlocate 155,642 inconsistent instruction streams, which cover 30% of all\ninstruction encodings and 47.8% of the instructions. We find undefined\nimplementation in ARM manual and implementation bugs of QEMU are the major\ncauses of inconsistencies. Furthermore, we discover four QEMU bugs, which are\nconfirmed and patched by thedevelopers, covering 13 instruction encodings\nincluding the most commonly used ones (e.g.,STR,BLX). With the inconsistent\ninstructions, we build three security applications and demonstrate\nthecapability of these instructions on detecting emulators, anti-emulation, and\nanti-fuzzing.\n"}
{"abstract": "  We develop a special phase field/diffusive interface method to model the\nnuclear architecture reorganization process. In particular, we use a Lagrange\nmultiplier approach in the phase field model to preserve the specific physical\nand geometrical constraints for the biological events. We develop several\nefficient and robust linear and weakly nonlinear schemes for this new model. To\nvalidate the model and numerical methods, we present ample numerical\nsimulations which in particular reproduce several processes of nuclear\narchitecture reorganization from the experiment literature.\n"}
{"abstract": "  Episode discovery from an event is a popular framework for data mining tasks\nand has many real-world applications. An episode is a partially ordered set of\nobjects (e.g., item, node), and each object is associated with an event type.\nThis episode can also be considered as a complex event sub-sequence.\nHigh-utility episode mining is an interesting utility-driven mining task in the\nreal world. Traditional episode mining algorithms, by setting a threshold,\nusually return a huge episode that is neither intuitive nor saves time. In\ngeneral, finding a suitable threshold in a pattern-mining algorithm is a\ntrivial and time-consuming task. In this paper, we propose a novel algorithm,\ncalled Top-K High Utility Episode (THUE) mining within the complex event\nsequence, which redefines the previous mining task by obtaining the K highest\nepisodes. We introduce several threshold-raising strategies and optimize the\nepisode-weighted utilization upper bounds to speed up the mining process and\neffectively reduce the memory cost. Finally, the experimental results on both\nreal-life and synthetic datasets reveal that the THUE algorithm can offer six\nto eight orders of magnitude running time performance improvement over the\nstate-of-the-art algorithm and has low memory consumption.\n"}
{"abstract": "  As the potential applications of Artificial Intelligence (AI) in the\nfinancial sector increases, ethical issues become gradually latent. The\ndistrust of individuals, social groups, and governments about the risks arising\nfrom Fintech's activities is growing. Due to this scenario, the preparation of\nrecommendations and Ethics Guidelines is increasing and the risks of being\nchosen the principles and ethical values most appropriate to companies are\nhigh. Thus, this exploratory research aims to analyze the benefits of the\napplication of the stakeholder theory and the idea of Social License to build\nan environment of trust and for the realization of ethical principles by\nFintech. The formation of a Fintech association for the creation of a Social\nLicense will allow early-stage Fintech to participate from the beginning of its\nactivities in the elaboration of a dynamic ethical code and with the\nparticipation of stakeholders.\n"}
{"abstract": "  Automatic evaluation remains an open research question in Natural Language\nGeneration. In the context of Sentence Simplification, this is particularly\nchallenging: the task requires by nature to replace complex words with simpler\nones that shares the same meaning. This limits the effectiveness of n-gram\nbased metrics like BLEU. Going hand in hand with the recent advances in NLG,\nnew metrics have been proposed, such as BERTScore for Machine Translation. In\nsummarization, the QuestEval metric proposes to automatically compare two texts\nby questioning them.\n  In this paper, we first propose a simple modification of QuestEval allowing\nit to tackle Sentence Simplification. We then extensively evaluate the\ncorrelations w.r.t. human judgement for several metrics including the recent\nBERTScore and QuestEval, and show that the latter obtain state-of-the-art\ncorrelations, outperforming standard metrics like BLEU and SARI. More\nimportantly, we also show that a large part of the correlations are actually\nspurious for all the metrics. To investigate this phenomenon further, we\nrelease a new corpus of evaluated simplifications, this time not generated by\nsystems but instead, written by humans. This allows us to remove the spurious\ncorrelations and draw very different conclusions from the original ones,\nresulting in a better understanding of these metrics. In particular, we raise\nconcerns about very low correlations for most of traditional metrics. Our\nresults show that the only significant measure of the Meaning Preservation is\nour adaptation of QuestEval.\n"}
{"abstract": "  The Talbot effect, also referred to as self-imaging or lensless imaging, was\noriginally discovered in the 1830's by Henry Fox Talbot. Over the years,\nvarious investigators have found different aspects of this phenomenon, and a\ntheory of the Talbot effect capable of explaining the various observations\nbased on the classical theory of diffraction has emerged. Unfortunately, many\nof the standard Optics textbooks do not discuss the Talbot effect. The goal of\nthe present paper is to bring to the reader's attention the essential features\nas well as an elementary explanation of this wonderful phenomenon.\n"}
{"abstract": "  This paper explores multi-armed bandit (MAB) strategies in very short horizon\nscenarios, i.e., when the bandit strategy is only allowed very few interactions\nwith the environment. This is an understudied setting in the MAB literature\nwith many applications in the context of games, such as player modeling.\nSpecifically, we pursue three different ideas. First, we explore the use of\nregression oracles, which replace the simple average used in strategies such as\nepsilon-greedy with linear regression models. Second, we examine different\nexploration patterns such as forced exploration phases. Finally, we introduce a\nnew variant of the UCB1 strategy called UCBT that has interesting properties\nand no tunable parameters. We present experimental results in a domain\nmotivated by exergames, where the goal is to maximize a player's daily steps.\nOur results show that the combination of epsilon-greedy or epsilon-decreasing\nwith regression oracles outperforms all other tested strategies in the short\nhorizon setting.\n"}
{"abstract": "  We performed a large-scale mapping observation toward the W33 complex and its\nsurroundings, covering an area of $1.3^\\circ \\times 1.0^\\circ$ , in $^{12}$CO\n(1-0), $^{13}$CO (1-0), and C$^{18}$O (1-0) lines from the Purple Mountain\nObservatory (PMO). We found a new hub--filament system ranging from 30 to 38.5\n\\kms located at the W33 complex. Three supercritical filaments are directly\nconverging into the central hub W33. Velocity gradients are detected along the\nfilaments and the accretion rates are in order of $\\rm 10^{-3}\\,M_\\odot\\,\nyr^{-1}$. The central hub W33 has a total mass of $\\rm\\sim\n1.8\\times10^5\\,M_\\odot$, accounting for $\\sim 60\\%$ of the mass of the\nhub--filament system. This indicates that the central hub is the mass reservoir\nof the hub-filament system. Furthermore, 49 ATLASGAL clumps are associated with\nthe hub--filament system. We find $57\\%$ of the clumps to be situated in the\ncentral hub W33 and clustered at the intersections between the filaments and\nthe W33 complex. Moreover, the distribution of Class I young stellar objects\n(YSOs) forms a structure resembling the hub--filament system and peaks at where\nthe clumps group; it seems to suggest that the mechanisms of clump formation\nand star formation in this region are correlated. Gas flows along the filaments\nare likely to feed the materials into the intersections and lead to the\nclustering and formation of the clumps in the hub--filament system W33. The\nstar formation in the intersections between the filaments and the W33 complex\nmight be triggered by the motion of gas converging into the intersections.\n"}
{"abstract": "  The breakthrough of contrastive learning (CL) has fueled the recent success\nof self-supervised learning (SSL) in high-level vision tasks on RGB images.\nHowever, CL is still ill-defined for low-level vision tasks, such as joint\ndemosaicking and denoising (JDD), in the RAW domain. To bridge this\nmethodological gap, we present a novel CL approach on RAW images, residual\ncontrastive learning (RCL), which aims to learn meaningful representations for\nJDD. Our work is built on the assumption that noise contained in each RAW image\nis signal-dependent, thus two crops from the same RAW image should have more\nsimilar noise distribution than two crops from different RAW images. We use\nresiduals as a discriminative feature and the earth mover's distance to measure\nthe distribution divergence for the contrastive loss. To evaluate the proposed\nCL strategy, we simulate a series of unsupervised JDD experiments with\nlarge-scale data corrupted by synthetic signal-dependent noise, where we set a\nnew benchmark for unsupervised JDD tasks with unknown (random) noise variance.\nOur empirical study not only validates that CL can be applied on distributions\n(c.f. features), but also exposes the lack of robustness of previous non-ML and\nSSL JDD methods when the statistics of the noise are unknown, thus providing\nsome further insight into signal-dependent noise problems.\n"}
{"abstract": "  In this paper, we obtain two Lichnerowicz type formulas for sub-signature\noperators. And we give the proof of Kastler-Kalau-Walze type theorems for\nsub-signature operators on 4-dimensional and 6-dimensional compact manifolds\nwith (resp.without) boundary.\n"}
{"abstract": "  The Alcubierre warp drive metric is a spacetime geometry featuring a\nspacetime distortion, called warp bubble, where a massive particle inside it\nacquires global superluminal velocities, or warp speeds. This work presents\nsolutions of the Einstein equations for the Alcubierre metric having fluid\nmatter as gravity source. The energy-momentum tensor considered two fluid\ncontents, the perfect fluid and the parametrized perfect fluid (PPF), a\ntentative more flexible model whose aim is to explore the possibilities of warp\ndrive solutions with positive matter density content. Santos-Pereira et al.\n(2020; arXiv:2008.06560) have already showed that the Alcubierre metric having\ndust as source connects this geometry to the Burgers equation, which describes\nshock waves moving through an inviscid fluid, but led the solutions back to\nvacuum. The same happened for two out of four solutions subcases for the\nperfect fluid. Other solutions for the perfect fluid indicate the possibility\nof warp drive with positive matter density, but at the cost of a complex\nsolution for the warp drive regulating function. Regarding the PPF, solutions\nwere also obtained indicating that warp speeds could be created with positive\nmatter density. Weak, dominant, strong and null energy conditions were\ncalculated for all studied subcases, being satisfied for the perfect fluid and\ncreating constraints in the PPF quantities such that positive matter density is\nalso possible for creating a warp bubble. Summing up all\nresults,energy-momentum tensors describing more complex forms of matter, or\nfield, distributions generate solutions for the Einstein equations with the\nwarp drive metric where negative matter density might not be a strict\nprecondition for attaining warp speeds.\n"}
{"abstract": "  We present two fiberized vector magnetic-field sensors, based on\nnitrogen-vacancy (NV) centers in diamond. The sensors feature\nsub-nT/$\\sqrt{\\textrm{Hz}}$ magnetic sensitivity. We use commercially available\ncomponents to construct sensors with a small sensor size, high photon\ncollection, and minimal sensor-sample distance. Both sensors are located at the\nend of optical fibres with the sensor-head freely accessible and robust under\nmovement. These features make them ideal for mapping magnetic fields with high\nsensitivity and spatial resolution ($\\leq$\\,mm). As a demonstration we use one\nof the sensors to map the vector magnetic field inside the bore of a $\\geq$\n100\\,mT Halbach array. The vector field sensing protocol translates microwave\nspectroscopy data addressing all diamonds axes and including double quantum\ntransitions to a 3D magnetic field vector.\n"}
{"abstract": "  Robot apps are becoming more automated, complex and diverse. An app usually\nconsists of many functions, interacting with each other and the environment.\nThis allows robots to conduct various tasks. However, it also opens a new door\nfor cyber attacks: adversaries can leverage these interactions to threaten the\nsafety of robot operations. Unfortunately, this issue is rarely explored in\npast works.\n  We present the first systematic investigation about the function interactions\nin common robot apps. First, we disclose the potential risks and damages caused\nby malicious interactions. We introduce a comprehensive graph to model the\nfunction interactions in robot apps by analyzing 3,100 packages from the Robot\nOperating System (ROS) platform. From this graph, we identify and categorize\nthree types of interaction risks. Second, we propose RTron, a novel system to\ndetect and mitigate these risks and protect the operations of robot apps. We\nintroduce security policies for each type of risks, and design coordination\nnodes to enforce the policies and regulate the interactions. We conduct\nextensive experiments on 110 robot apps from the ROS platform and two complex\napps (Baidu Apollo and Autoware) widely adopted in industry. Evaluation results\nindicated RTron can correctly identify and mitigate all potential risks with\nnegligible performance cost. To validate the practicality of the risks and\nsolutions, we implement and evaluate RTron on a physical UGV (Turtlebot) with\nreal-word apps and environments.\n"}
{"abstract": "  We consider supersymmetric surface defects in compactifications of the $6d$\nminimal $(D_{N+3},D_{N+3})$ conformal matter theories on a punctured Riemann\nsurface. For the case of $N=1$ such defects are introduced into the\nsupersymmetric index computations by an action of the $BC_1\\,(\\sim A_1\\sim\nC_1)$ van Diejen model. We (re)derive this fact using three different field\ntheoretic descriptions of the four dimensional models. The three field\ntheoretic descriptions are naturally associated with algebras $A_{N=1}$,\n$C_{N=1}$, and $(A_1)^{N=1}$. The indices of these $4d$ theories give rise to\nthree different Kernel functions for the $BC_1$ van Diejen model. We then\nconsider the generalizations with $N>1$. The operators introducing defects into\nthe index computations are certain $A_{N}$, $C_N$, and $(A_1)^{N}$\ngeneralizations of the van Diejen model. The three different generalizations\nare directly related to three different effective gauge theory descriptions one\ncan obtain by compactifying the minimal $(D_{N+3},D_{N+3})$ conformal matter\ntheories on a circle to five dimensions. We explicitly compute the operators\nfor the $A_N$ case, and derive various properties these operators have to\nsatisfy as a consequence of $4d$ dualities following from the geometric setup.\nIn some cases we are able to verify these properties which in turn serve as\nchecks of said dualities. As a by-product of our constructions we also discuss\na simple Lagrangian description of a theory corresponding to compactification\non a sphere with three maximal punctures of the minimal $(D_5,D_5)$ conformal\nmatter and as consequence give explicit Lagrangian constructions of\ncompactifications of this 6d SCFT on arbitrary Riemann surfaces.\n"}
{"abstract": "  Adapter modules were recently introduced as an efficient alternative to\nfine-tuning in NLP. Adapter tuning consists in freezing pretrained parameters\nof a model and injecting lightweight modules between layers, resulting in the\naddition of only a small number of task-specific trainable parameters. While\nadapter tuning was investigated for multilingual neural machine translation,\nthis paper proposes a comprehensive analysis of adapters for multilingual\nspeech translation (ST). Starting from different pre-trained models (a\nmultilingual ST trained on parallel data or a multilingual BART (mBART) trained\non non-parallel multilingual data), we show that adapters can be used to: (a)\nefficiently specialize ST to specific language pairs with a low extra cost in\nterms of parameters, and (b) transfer from an automatic speech recognition\n(ASR) task and an mBART pre-trained model to a multilingual ST task.\nExperiments show that adapter tuning offer competitive results to full\nfine-tuning, while being much more parameter-efficient.\n"}
{"abstract": "  Entanglement measures quantify nonclassical correlations present in a quantum\nsystem, but can be extremely difficult to calculate, even more so, when\ninformation on its state is limited. Here, we consider broad families of\nentanglement criteria that are based on variances of arbitrary operators and\nanalytically derive the lower bounds these criteria provide for two relevant\nentanglement measures: the best separable approximation (BSA) and the\ngeneralized robustness (GR). This yields a practical method for quantifying\nentanglement in realistic experimental situations, in particular, when only few\nmeasurements of simple observables are available. As a concrete application of\nthis method, we quantify bipartite and multipartite entanglement in\nspin-squeezed Bose-Einstein condensates of $\\sim 500$ atoms, by lower bounding\nthe BSA and the GR only from measurements of first and second moments of the\ncollective spin operator.\n"}
{"abstract": "  Federated Learning allows training machine learning models by using the\ncomputation and private data resources of many distributed clients such as\nsmartphones and IoT devices. Most existing works on Federated Learning (FL)\nassume the clients have ground-truth labels. However, in many practical\nscenarios, clients may be unable to label task-specific data, e.g., due to a\nlack of expertise. This work considers a server that hosts a labeled dataset\nand wishes to leverage clients with unlabeled data for supervised learning. We\npropose a new Federated Learning framework referred to as SemiFL to address\nSemi-Supervised Federated Learning (SSFL). In SemiFL, clients have completely\nunlabeled data, while the server has a small amount of labeled data. SemiFL is\ncommunication efficient since it separates the training of server-side\nsupervised data and client-side unsupervised data. We demonstrate several\nstrategies of SemiFL that enhance efficiency and prediction and develop\nintuitions of why they work. In particular, we provide a theoretical\nunderstanding of the use of strong data augmentation for Semi-Supervised\nLearning (SSL), which can be interesting in its own right. Extensive empirical\nevaluations demonstrate that our communication efficient method can\nsignificantly improve the performance of a labeled server with unlabeled\nclients. Moreover, we demonstrate that SemiFL can outperform many existing FL\nresults trained with fully supervised data, and perform competitively with the\nstate-of-the-art centralized SSL methods. For instance, in standard\ncommunication efficient scenarios, our method can perform $93\\%$ accuracy on\nthe CIFAR10 dataset with only $4000$ labeled samples at the server. Such\naccuracy is only $2\\%$ away from the result trained from $50000$ fully labeled\ndata, and it improves about $30\\%$ upon existing SSFL methods in the\ncommunication efficient setting.\n"}
{"abstract": "  We consider the diffusive limit of a typical pure-jump Markovian control\nproblem as the intensity of the driving Poisson process tends to infinity. We\nshow that the convergence speed is provided by the H\\\"older constant of the\nHessian of the limit problem, and explain how correction terms can be\nconstructed. This provides an alternative efficient method for the numerical\napproximation of the optimal control of a pure-jump problem in situations with\nvery high intensity of jump. We illustrate this approach in the context of a\ndisplay advertising auction problem.\n"}
{"abstract": "  We give a formula for the bivariate generating function of a stationary\n1-dependent counting process in terms of its run probability generating\nfunction, with a probabilistic proof. The formula reduces to the well known\nbivariate generating function of the Eulerian distribution in the case of\ndescents of a sequence of indepependent and identically distributed random\nvariables. The formula is compared with alternative expressions from the theory\nof determinantal point processes and the combinatorics of sequences.\n"}
{"abstract": "  The purpose of this study is to examine the impact of board member\ncomposition and board members' social media presence on the performance of\nstartups. Using multiple sources, we compile a unique dataset of about 500\nUS-based technology startups. We find that startups with more venture\ncapitalists on the board and whose board members are active on Twitter attract\nadditional funding over the years, though they do not generate additional\nsales. By contrast, startups which have no venture capitalists on the board and\nwhose board members are not on Twitter show an increased ability to translate\nassets into sales. Consistent with other research, our results indicate that\nstartups potentially benefit from working with VCs because of the opportunity\nto access additional funding, although their presence does not necessarily\ntranslate into sales growth and operational efficiency. We use a number of\ncontrol variables, including board gender representation and board members'\nposition in the interlocking directorates' network.\n"}
{"abstract": "  Entity Linking (EL) seeks to align entity mentions in text to entries in a\nknowledge-base and is usually comprised of two phases: candidate generation and\ncandidate ranking. While most methods focus on the latter, it is the candidate\ngeneration phase that sets an upper bound to both time and accuracy performance\nof the overall EL system. This work's contribution is a significant improvement\nin candidate generation which thus raises the performance threshold for EL, by\ngenerating candidates that include the gold entity in the least candidate set\n(top-K). We propose a simple approach that efficiently embeds mention-entity\npairs in dense space through a BERT-based bi-encoder. Specifically, we extend\n(Wu et al., 2020) by introducing a new pooling function and incorporating\nentity type side-information. We achieve a new state-of-the-art 84.28% accuracy\non top-50 candidates on the Zeshel dataset, compared to the previous 82.06% on\nthe top-64 of (Wu et al., 2020). We report the results from extensive\nexperimentation using our proposed model on both seen and unseen entity\ndatasets. Our results suggest that our method could be a useful complement to\nexisting EL approaches.\n"}
{"abstract": "  We have conducted an Arecibo 327 MHz search of two dwarf irregular galaxies\nin the Local Group, Leo A and T, for radio pulsars and single pulses from fast\nradio bursts and other giant pulse emitters. We detected no astrophysical\nsignals in this search, and we estimate flux density limits on both periodic\nand burst emission. Our derived luminosity limits indicate that only the most\nluminous radio pulsars known in our Galaxy and in the Magellanic Clouds (MCs)\nwould have been detectable in our search if they were at the distances of Leo A\nand T. Given the much smaller stellar mass content and star formation rates of\nLeo A and T compared to the Milky Way and the MCs, there are likely to be few\n(if any) extremely luminous pulsars in these galaxies. It is therefore not\nsurprising that we detected no pulsars in our search.\n"}
{"abstract": "  We derive low-order, inf-sup stable and divergence-free finite element\napproximations for the Stokes problem using Worsey-Farin splits in three\ndimensions and Powell-Sabin splits in two dimensions. The velocity space simply\nconsists of continuous, piecewise linear polynomials, where as the pressure\nspace is a subspace of piecewise constants with weak continuity properties at\nsingular edges (3D) and singular vertices (2D). We discuss implementation\naspects that arise when coding the pressure space, and in particular, show that\nthe pressure constraints can be enforced at an algebraic level.\n"}
{"abstract": "  The system size dependence of baryon-strangeness (BS) correlations ($C_{BS}$)\nare investigated with a multiphase transport (AMPT) model for various collision\nsystems from $\\mathrm{^{10}B+^{10}B}$, $\\mathrm{^{12}C+^{12}C}$,\n$\\mathrm{^{16}O+^{16}O}$, $\\mathrm{^{20}Ne+^{20}Ne}$,\n$\\mathrm{^{40}Ca+^{40}Ca}$, $\\mathrm{^{96}Zr+^{96}Zr}$, and\n$\\mathrm{^{197}Au+^{197}Au}$ at RHIC energies $\\sqrt{s_{NN}}$ of 200, 39, 27,\n20, and 7.7 GeV. Both effects of hadron rescattering and a combination of\ndifferent hadrons play a leading role for baryon-strangeness correlations. When\nthe kinetic window is limited to absolute rapidity $|y|>3$, these correlations\ntend to be constant after the final-state interaction whatever kind of hadrons\nsubset we chose based on the AMPT framework. The correlation is found to\nsmoothly increase with baryon chemical potential $\\mu_B$, corresponding to the\ncollision system or energy from the quark-gluon-plasma-like phase to the\nhadron-gas-like phase. Besides, the influence of initial nuclear geometrical\nstructures of $\\alpha$-clustered nuclear collision systems of\n$\\mathrm{^{12}C+^{12}C}$ as well as $\\mathrm{^{16}O+^{16}O}$ collisions is\ndiscussed but the effect is found negligible. The current model studies provide\nbaselines for searching for the signals of Quantum Chromodynamics (QCD) phase\ntransition and critical point in heavy-ion collisions through the BS\ncorrelation.\n"}
{"abstract": "  Scalar Field Dark Matter (SFDM) comprised of ultralight ($\\gtrsim 10^{-22}$\neV) bosons is an alternative to standard, collisionless Cold Dark Matter (CDM)\nthat is CDM-like on large scales but inhibits small-scale structure formation.\nAs a Bose-Einstein condensate, its free-field (\"fuzzy\") limit (FDM) suppresses\nstructure below the de Broglie wavelength, $\\lambda_\\text{deB}$, creating\nvirialized haloes with central cores of radius $\\sim\\lambda_\\text{deB}$,\nsurrounded by CDM-like envelopes, and a halo mass function (HMF) with a sharp\ncut-off on small scales. With a strong enough repulsive self-interaction (SI),\nstructure is inhibited, instead, below the Thomas-Fermi (TF) radius,\n$R_\\text{TF}$ (the size of an SI-pressure-supported ($n=1$)-polytrope), when\n$R_\\text{TF} > \\lambda_\\text{deB}$. Previously, we developed tools to describe\nSFDM dynamics on scales above $\\lambda_\\text{deB}$ and showed that SFDM-TF\nhaloes formed by Jeans-unstable collapse from non-cosmological initial\nconditions have $R_\\text{TF}$-sized cores, surrounded by CDM-like envelopes.\nRevisiting SFDM-TF in the cosmological context, we simulate halo formation by\ncosmological infall and collapse, and derive its transfer function from linear\nperturbation theory to produce cosmological initial conditions and predict\nstatistical measures of structure formation, such as the HMF. Since FDM and\nSFDM-TF transfer functions both have small-scale cut-offs, we can align them to\nlet observational constraints on FDM proxy for SFDM-TF, finding FDM with\nparticle masses $1 \\lesssim m/(10^{-22} \\text{ eV}/c^2) \\lesssim 30$\ncorresponds to SFDM-TF with $10 \\gtrsim R_\\text{TF}/(1 \\text{ pc}) \\gtrsim 1$,\nfavoring sub-galactic (sub-kpc) core-size. The SFDM-TF HMF cuts off gradually,\nhowever, leaving more small-mass haloes: its Jeans mass shrinks so fast, scales\nfiltered early can still recover and grow!\n"}
{"abstract": "  This study applies convolutional neural network (CNN)-based automatic\nsegmentation and distensibility measurement of the ascending and descending\naorta from 2D phase-contrast cine magnetic resonance imaging (PC-cine MRI)\nwithin the large MESA cohort with subsequent assessment on an external cohort\nof thoracic aortic aneurysm (TAA) patients. 2D PC-cine MRI images of the\nascending and descending aorta at the pulmonary artery bifurcation from the\nMESA study were included. Train, validation, and internal test sets consisted\nof 1123 studies (24282 images), 374 studies (8067 images), and 375 studies\n(8069 images), respectively. An external test set of TAAs consisted of 37\nstudies (3224 images). A U-Net based CNN was constructed, and performance was\nevaluated utilizing dice coefficient (for segmentation) and concordance\ncorrelation coefficients (CCC) of aortic geometric parameters by comparing to\nmanual segmentation and parameter estimation. Dice coefficients for aorta\nsegmentation were 97.6% (CI: 97.5%-97.6%) and 93.6% (84.6%-96.7%) on the\ninternal and external test of TAAs, respectively. CCC for comparison of manual\nand CNN maximum and minimum ascending aortic areas were 0.97 and 0.95,\nrespectively, on the internal test set and 0.997 and 0.995, respectively, for\nthe external test. CCCs for maximum and minimum descending aortic areas were\n0.96 and 0. 98, respectively, on the internal test set and 0.93 and 0.93,\nrespectively, on the external test set. We successfully developed and validated\na U-Net based ascending and descending aortic segmentation and distensibility\nquantification model in a large multi-ethnic database and in an external cohort\nof TAA patients.\n"}
{"abstract": "  The paper describes an algorithm to compute a consensus sequence from a set\nof DNA sequences of approximatively identical length generated by 3rd\nsequencing generation technologies. Its purpose targets DNA storage and is\nguided by specific features that cannot be exhibited from biological data such\nas the exact length of the consensus sequences, the precise location of known\npatterns, the kmer composition, etc.\n"}
{"abstract": "  This paper studies the complexity for finding approximate stationary points\nof nonconvex-strongly-concave (NC-SC) smooth minimax problems, in both general\nand averaged smooth finite-sum settings. We establish nontrivial lower\ncomplexity bounds of $\\Omega(\\sqrt{\\kappa}\\Delta L\\epsilon^{-2})$ and\n$\\Omega(n+\\sqrt{n\\kappa}\\Delta L\\epsilon^{-2})$ for the two settings,\nrespectively, where $\\kappa$ is the condition number, $L$ is the smoothness\nconstant, and $\\Delta$ is the initial gap. Our result reveals substantial gaps\nbetween these limits and best-known upper bounds in the literature. To close\nthese gaps, we introduce a generic acceleration scheme that deploys existing\ngradient-based methods to solve a sequence of crafted\nstrongly-convex-strongly-concave subproblems. In the general setting, the\ncomplexity of our proposed algorithm nearly matches the lower bound; in\nparticular, it removes an additional poly-logarithmic dependence on accuracy\npresent in previous works. In the averaged smooth finite-sum setting, our\nproposed algorithm improves over previous algorithms by providing a\nnearly-tight dependence on the condition number.\n"}
{"abstract": "  We perform a finite-size Kirkwood--Buff analysis on model supercooled\nliquids, with and without attractive interactions, to compare thermodynamic\nquantities affected by long-range fluctuations in two-body correlation\nfunctions. The attractive potential induces anomalous structural behaviour,\napparent in the $k\\to 0$ limit of density and composition structure factors,\nanalogous to the nucleation of Fischer clusters. By contrast, the purely\nrepulsive system remains perfectly miscible. This difference contradicts the\nwidespread belief that the two systems exhibit similar two-body structures.\nMoreover, the anomaly disappears at high density, where both systems display\nnearly identical dynamical properties. This result suggests that long-range\nstructures might contribute to the dynamic slowing down of glass-forming\nliquids.\n"}
{"abstract": "  Is a geometric model required to synthesize novel views from a single image?\nBeing bound to local convolutions, CNNs need explicit 3D biases to model\ngeometric transformations. In contrast, we demonstrate that a transformer-based\nmodel can synthesize entirely novel views without any hand-engineered 3D\nbiases. This is achieved by (i) a global attention mechanism for implicitly\nlearning long-range 3D correspondences between source and target views, and\n(ii) a probabilistic formulation necessary to capture the ambiguity inherent in\npredicting novel views from a single image, thereby overcoming the limitations\nof previous approaches that are restricted to relatively small viewpoint\nchanges. We evaluate various ways to integrate 3D priors into a transformer\narchitecture. However, our experiments show that no such geometric priors are\nrequired and that the transformer is capable of implicitly learning 3D\nrelationships between images. Furthermore, this approach outperforms the state\nof the art in terms of visual quality while covering the full distribution of\npossible realizations. Code is available at https://git.io/JOnwn\n"}
{"abstract": "  Reconfigurable intelligent surfaces (RISs) have recently received widespread\nattention in the field of wireless communication. An RIS can be controlled to\nreflect incident waves from the transmitter towards the receiver; a feature\nthat is believed to fundamentally contribute to beyond 5G wireless technology.\nThe typical RIS consists of entirely passive elements, which requires the\nhigh-dimensional channel estimation to be done elsewhere. Therefore, in this\npaper, we present a semi-passive large-scale RIS architecture equipped with\nonly a small fraction of simplified receiver units with only 1-bit\nquantization. Based on this architecture, we first propose an alternating\ndirection method of multipliers (ADMM)-based approach to recover the training\nsignals at the passive RIS elements, We then obtain the global channel by\ncombining a channel sparsification step with the generalized approximate\nmessage passing (GAMP) algorithm. Our proposed scheme exploits both the\nsparsity and low-rankness properties of the channel in the joint\nspatial-frequency domain of a wideband mmWave multiple-input-multiple-output\n(MIMO) communication system. Simulation results show that the proposed\nalgorithm can significantly reduce the pilot signaling needed for accurate\nchannel estimation and outperform previous methods, even with fewer receiver\nunits.\n"}
{"abstract": "  In this paper, we show that replacing the annealed disorder by the quenched\none can result in the switch from the continuous to discontinuous phase\ntransition. This is an opposite phenomenon to the one, which is usually seen.\nSo far, it was indicated by numerous studies that the quenched disorder can\nround or completely destroy existence of discontinuous phase transition. To\nshow the possibility of such a behavior, we study a generalized multi-state\n$q$-voter model with two types of disorder, annealed and quenched, that\nrepresents so-called anticonformity. We analyze the model on the complete graph\nanalytically and via Monte Carlo simulations and show that in such a model the\nquenched disorder promotes discontinuous phase transitions.\n"}
{"abstract": "  Background subtraction (BGS) is a fundamental video processing task which is\na key component of many applications. Deep learning-based supervised algorithms\nachieve very good perforamnce in BGS, however, most of these algorithms are\noptimized for either a specific video or a group of videos, and their\nperformance decreases dramatically when applied to unseen videos. Recently,\nseveral papers addressed this problem and proposed video-agnostic supervised\nBGS algorithms. However, nearly all of the data augmentations used in these\nalgorithms are limited to the spatial domain and do not account for temporal\nvariations that naturally occur in video data. In this work, we introduce\nspatio-temporal data augmentations and apply them to one of the leading\nvideo-agnostic BGS algorithms, BSUV-Net. We also introduce a new\ncross-validation training and evaluation strategy for the CDNet-2014 dataset\nthat makes it possible to fairly and easily compare the performance of various\nvideo-agnostic supervised BGS algorithms. Our new model trained using the\nproposed data augmentations, named BSUV-Net 2.0, significantly outperforms\nstate-of-the-art algorithms evaluated on unseen videos of CDNet-2014. We also\nevaluate the cross-dataset generalization capacity of BSUV-Net 2.0 by training\nit solely on CDNet-2014 videos and evaluating its performance on LASIESTA\ndataset. Overall, BSUV-Net 2.0 provides a ~5% improvement in the F-score over\nstate-of-the-art methods on unseen videos of CDNet-2014 and LASIESTA datasets.\nFurthermore, we develop a real-time variant of our model, that we call Fast\nBSUV-Net 2.0, whose performance is close to the state of the art.\n"}
{"abstract": "  We study charged massless fermionic perturbations in the background of\n$4$-dimensional linear dilaton black holes in Einstein-Maxwell-dilaton theory\nwith double Liouville-type potentials. We present the analytical fermionic\nquasinormal modes, whose Dirac equations are solved in terms of hypergeometric\nfunctions. We also discuss the stability of these black holes under the charged\nfermionic perturbations.\n"}
{"abstract": "  Both linear complementary dual (LCD) codes and maximum distance separable\n(MDS) codes have good algebraic structures, and they have interesting practical\napplications such as communication systems, data storage, quantum codes, and so\non. So far, most of LCD MDS codes have been constructed by employing\ngeneralized Reed-Solomon codes. In this paper we construct some classes of new\nEuclidean LCD MDS codes and Hermitian LCD MDS codes which are not monomially\nequivalent to Reed-Solomon codes, called LCD MDS codes of non-Reed- Solomon\ntype. Our method is based on the constructions of Beelen et al. (2017) and Roth\nand Lempel (1989). To the best of our knowledge, this is the first paper on the\nconstruction of LCD MDS codes of non-Reed-Solomon type; any LCD MDS code of\nnon- Reed-Solomon type constructed by our method is not monomially equivalent\nto any LCD code constructed by the method of Carlet et al. (2018).\n"}
{"abstract": "  We show that a positive proportion of quartic fields are not monogenic,\ndespite having no local obstruction to being monogenic. Our proof builds on the\ncorresponding result for cubic fields that we obtained in a previous work.\nAlong the way, we also prove that a positive proportion of quartic rings of\nintegers do not arise as the invariant order of an integral binary quartic form\ndespite having no local obstruction.\n"}
{"abstract": "  In real acoustic environment, speech enhancement is an arduous task to\nimprove the quality and intelligibility of speech interfered by background\nnoise and reverberation. Over the past years, deep learning has shown great\npotential on speech enhancement. In this paper, we propose a novel real-time\nframework called DBNet which is a dual-branch structure with alternate\ninterconnection. Each branch incorporates an encoder-decoder architecture with\nskip connections. The two branches are responsible for spectrum and waveform\nmodeling, respectively. A bridge layer is adopted to exchange information\nbetween the two branches. Systematic evaluation and comparison show that the\nproposed system substantially outperforms related algorithms under very\nchallenging environments. And in INTERSPEECH 2021 Deep Noise Suppression (DNS)\nchallenge, the proposed system ranks the top 8 in real-time track 1 in terms of\nthe Mean Opinion Score (MOS) of the ITU-T P.835 framework.\n"}
{"abstract": "  This paper introduces the F3ORNITS non-iterative co-simulation algorithm in\nwhich F3 stands for the 3 flexible aspects of the method: flexible polynomial\norder representation of coupling variables, flexible time-stepper applying\nvariable co-simulation step size rules on subsystems allowing it and flexible\nscheduler orchestrating the meeting times among the subsystems and capable of\nasynchronousness when subsystems constraints requires it. The motivation of the\nF3ORNITS method is to accept any kind of co-simulation model, including any\nkind of subsystem, regardless on their available capabilities. Indeed, one the\nmajor problems in industry is that the subsystems usually have constraints or\nlack of advanced capabilities making it impossible to implement most of the\nadvanced co-simulation algorithms on them. The method makes it possible to\npreserve the dynamics of the coupling constraints when necessary as well as to\navoid breaking C1 smoothness at communication times, and also to adapt the\nco-simulation step size in a way that is robust both to zero-crossing variables\n(contrary to classical relative error-based criteria) and to jumps. Two test\ncases are presented to illustrate the robustness of the F3ORNITS method as well\nas its higher accuracy than the non-iterative Jacobi coupling algorithm (the\nmost commonly used method in industry) for a smaller number of co-simulation\nsteps.\n"}
{"abstract": "  The Hall effects comprise one of the oldest but most vital fields in\ncondensed matter physics, and they persistently inspire new findings, such as\nquantum Hall effects and topological phases of matter. The recently discovered\nnonlinear Hall effect is a new member of the family of Hall effects. It is\ncharacterized as a transverse Hall voltage in response to two longitudinal\ncurrents in the Hall measurement, but it does not require time-reversal\nsymmetry to be broken. It has deep connections to symmetry and topology and,\nthus, opens new avenues by which to probe the spectral, symmetry and\ntopological properties of emergent quantum materials and phases of matter. In\nthis Perspective, we present an overview of the recent progress regarding the\nnonlinear Hall effect. We discuss the open problems, the prospects of the use\nof the nonlinear Hall effect in spectroscopic and device applications, and\ngeneralizations to other nonlinear transport effects.\n"}
{"abstract": "  Accessible machine learning algorithms, software, and diagnostic tools for\nenergy-efficient devices and systems are extremely valuable across a broad\nrange of application domains. In scientific domains, real-time near-sensor\nprocessing can drastically improve experimental design and accelerate\nscientific discoveries. To support domain scientists, we have developed hls4ml,\nan open-source software-hardware codesign workflow to interpret and translate\nmachine learning algorithms for implementation with both FPGA and ASIC\ntechnologies. We expand on previous hls4ml work by extending capabilities and\ntechniques towards low-power implementations and increased usability: new\nPython APIs, quantization-aware pruning, end-to-end FPGA workflows, long\npipeline kernels for low power, and new device backends include an ASIC\nworkflow. Taken together, these and continued efforts in hls4ml will arm a new\ngeneration of domain scientists with accessible, efficient, and powerful tools\nfor machine-learning-accelerated discovery.\n"}
{"abstract": "  In the last few years, image denoising has benefited a lot from the fast\ndevelopment of neural networks. However, the requirement of large amounts of\nnoisy-clean image pairs for supervision limits the wide use of these models.\nAlthough there have been a few attempts in training an image denoising model\nwith only single noisy images, existing self-supervised denoising approaches\nsuffer from inefficient network training, loss of useful information, or\ndependence on noise modeling. In this paper, we present a very simple yet\neffective method named Neighbor2Neighbor to train an effective image denoising\nmodel with only noisy images. Firstly, a random neighbor sub-sampler is\nproposed for the generation of training image pairs. In detail, input and\ntarget used to train a network are images sub-sampled from the same noisy\nimage, satisfying the requirement that paired pixels of paired images are\nneighbors and have very similar appearance with each other. Secondly, a\ndenoising network is trained on sub-sampled training pairs generated in the\nfirst stage, with a proposed regularizer as additional loss for better\nperformance. The proposed Neighbor2Neighbor framework is able to enjoy the\nprogress of state-of-the-art supervised denoising networks in network\narchitecture design. Moreover, it avoids heavy dependence on the assumption of\nthe noise distribution. We explain our approach from a theoretical perspective\nand further validate it through extensive experiments, including synthetic\nexperiments with different noise distributions in sRGB space and real-world\nexperiments on a denoising benchmark dataset in raw-RGB space.\n"}
{"abstract": "  In this work we aim to characterise the dust motion in the inner coma of\ncomet 67P/Churyumov- Gerasimenko to provide constraints for theoretical 3D coma\nmodels. The OSIRIS camera onboard the Rosetta mission was able for the first\ntime to acquire images of single dust particles from inside the cometary coma,\nvery close to the nucleus. We analyse a large number of particles, performing a\nsignificant statistic of their behaviour during the post perihelion period,\nwhen the spacecraft covered distances from the nucleus ranging between 80 and\n400 km. We describe the particle trajectories, investigating their orientation\nand finding highly radial motion with respect to the nucleus. Then, from the\nparticle brightness profiles, we derive a particle rotational frequency of v <\n3.6 Hz, revealing that they are slow rotators and do not undergo fragmentation.\nWe use scattering models to compare the observed spectral radiance of the\nparticles with the simulated ones in order to estimate their size, finding\nvalues that range from millimetres up to centimetres. The statistics performed\nin this paper provide useful parameters to constrain the cometary coma\ndynamical models.\n"}
{"abstract": "  We present RAWLSNET, a system for altering Bayesian Network (BN) models to\nsatisfy the Rawlsian principle of fair equality of opportunity (FEO).\nRAWLSNET's BN models generate aspirational data distributions: data generated\nto reflect an ideally fair, FEO-satisfying society. FEO states that everyone\nwith the same talent and willingness to use it should have the same chance of\nachieving advantageous social positions (e.g., employment), regardless of their\nbackground circumstances (e.g., socioeconomic status). Satisfying FEO requires\nalterations to social structures such as school assignments. Our paper\ndescribes RAWLSNET, a method which takes as input a BN representation of an FEO\napplication and alters the BN's parameters so as to satisfy FEO when possible,\nand minimize deviation from FEO otherwise. We also offer guidance for applying\nRAWLSNET, including on recognizing proper applications of FEO. We demonstrate\nthe use of our system with publicly available data sets. RAWLSNET's altered BNs\noffer the novel capability of generating aspirational data for FEO-relevant\ntasks. Aspirational data are free from the biases of real-world data, and thus\nare useful for recognizing and detecting sources of unfairness in machine\nlearning algorithms besides biased data.\n"}
{"abstract": "  This work presents an in-depth analysis of four gravitationally lensed red\ngalaxies at z = 1.6-3.2. The sources are magnified by factors of 2.7-30 by\nforeground clusters, enabling spectral and morphological measurements that are\notherwise challenging. Our sample extends below the characteristic mass of the\nstellar mass function and is thus more representative of the quiescent galaxy\npopulation at z > 1 than previous spectroscopic studies. We analyze deep\nVLT/X-SHOOTER spectra and multi-band Hubble Space Telescope photometry that\ncover the rest-frame UV-to-optical regime. The entire sample resembles stellar\ndisks as inferred from lensing-reconstructed images. Through stellar population\nsynthesis analysis we infer that the targets are young (median age = 0.1-1.2\nGyr) and formed 80% of their stellar masses within 0.07-0.47 Gyr. Mg II\n$\\lambda\\lambda 2796,2803$ absorption is detected across the sample.\nBlue-shifted absorption and/or redshifted emission of Mg II is found in the two\nyoungest sources, indicative of a galactic-scale outflow of warm ($T\\sim10^{4}$\nK) gas. The [O III] $\\lambda5007$ luminosity is higher for the two young\nsources (median age less than 0.4 Gyr) than the two older ones, perhaps\nsuggesting a decline in nuclear activity as quenching proceeds. Despite\nhigh-velocity ($v\\approx1500$ km s$^{-1}$) galactic-scale outflows seen in the\nmost recently quenched galaxies, warm gas is still present to some extent long\nafter quenching. Altogether our results indicate that star formation quenching\nat high redshift must have been a rapid process (< 1 Gyr) that does not\nsynchronize with bulge formation or complete gas removal. Substantial bulge\ngrowth is required if they are to evolve into the metal-rich cores of\npresent-day slow-rotators.\n"}
{"abstract": "  ATLAS is the first Portuguese radar system that aims to detect space debris.\nThe article introduces the system and provides a brief description of its\ncapabilities. The system is capable of synthesizing arbitrary amplitude\nmodulated pulse shapes with a resolution of 10 ns. Given that degree of freedom\nwe decided to test an amplitude modulated chirp signal developed by us and a\nnested barker code. These waveforms are explained as well as their advantages\nand drawbacks for space debris detection. An experimental setup was developed\nto test the system receiver and waveforms are processed by digital matched\nfiltering. The experiments test the system using different waveform shapes and\nnoise levels. Experimental results are in agreement with simulation and show\nthat the chirp signal is more resilient to Doppler shifts, has higher range\nresolution and lower peak-to-sidelobe ratio in comparison with the nested\nbarker code. Future work in order to increase detection capabilities is\ndiscussed at the end.\n"}
{"abstract": "  Metaphor is widely used in human communication. The cohort of scholars\nstudying metaphor in various fields is continuously growing, but very few work\nhas been done in bibliographical analysis of metaphor research. This paper\nexamines the advancements in metaphor research from 2000 to 2017. Using data\nretrieved from Microsoft Academic Graph and Web of Science, this paper makes a\nmacro analysis of metaphor research, and expounds the underlying patterns of\nits development. Taking into consideration sub-fields of metaphor research, the\ninternal analysis of metaphor research is carried out from a micro perspective\nto reveal the evolution of research topics and the inherent relationships among\nthem. This paper provides novel insights into the current state of the art of\nmetaphor research as well as future trends in this field, which may spark new\nresearch interests in metaphor from both linguistic and interdisciplinary\nperspectives.\n"}
{"abstract": "  We prove Sklar's theorem in infinite dimensions via a topological argument\nand the notion of inverse systems.\n"}
{"abstract": "  Cross-resolution face recognition (CRFR), which is important in intelligent\nsurveillance and biometric forensics, refers to the problem of matching a\nlow-resolution (LR) probe face image against high-resolution (HR) gallery face\nimages. Existing shallow learning-based and deep learning-based methods focus\non mapping the HR-LR face pairs into a joint feature space where the resolution\ndiscrepancy is mitigated. However, little works consider how to extract and\nutilize the intermediate discriminative features from the noisy LR query faces\nto further mitigate the resolution discrepancy due to the resolution\nlimitations. In this study, we desire to fully exploit the multi-level deep\nconvolutional neural network (CNN) feature set for robust CRFR. In particular,\nour contributions are threefold. (i) To learn more robust and discriminative\nfeatures, we desire to adaptively fuse the contextual features from different\nlayers. (ii) To fully exploit these contextual features, we design a feature\nset-based representation learning (FSRL) scheme to collaboratively represent\nthe hierarchical features for more accurate recognition. Moreover, FSRL\nutilizes the primitive form of feature maps to keep the latent structural\ninformation, especially in noisy cases. (iii) To further promote the\nrecognition performance, we desire to fuse the hierarchical recognition outputs\nfrom different stages. Meanwhile, the discriminability from different scales\ncan also be fully integrated. By exploiting these advantages, the efficiency of\nthe proposed method can be delivered. Experimental results on several face\ndatasets have verified the superiority of the presented algorithm to the other\ncompetitive CRFR approaches.\n"}
{"abstract": "  The memorization effect of deep neural network (DNN) plays a pivotal role in\nmany state-of-the-art label-noise learning methods. To exploit this property,\nthe early stopping trick, which stops the optimization at the early stage of\ntraining, is usually adopted. Current methods generally decide the early\nstopping point by considering a DNN as a whole. However, a DNN can be\nconsidered as a composition of a series of layers, and we find that the latter\nlayers in a DNN are much more sensitive to label noise, while their former\ncounterparts are quite robust. Therefore, selecting a stopping point for the\nwhole network may make different DNN layers antagonistically affected each\nother, thus degrading the final performance. In this paper, we propose to\nseparate a DNN into different parts and progressively train them to address\nthis problem. Instead of the early stopping, which trains a whole DNN all at\nonce, we initially train former DNN layers by optimizing the DNN with a\nrelatively large number of epochs. During training, we progressively train the\nlatter DNN layers by using a smaller number of epochs with the preceding layers\nfixed to counteract the impact of noisy labels. We term the proposed method as\nprogressive early stopping (PES). Despite its simplicity, compared with the\nearly stopping, PES can help to obtain more promising and stable results.\nFurthermore, by combining PES with existing approaches on noisy label training,\nwe achieve state-of-the-art performance on image classification benchmarks.\n"}
{"abstract": "  The problem of causal inference with panel data is a central econometric\nquestion. The following is a fundamental version of this problem: Let $M^*$ be\na low rank matrix and $E$ be a zero-mean noise matrix. For a `treatment' matrix\n$Z$ with entries in $\\{0,1\\}$ we observe the matrix $O$ with entries $O_{ij} :=\nM^*_{ij} + E_{ij} + \\mathcal{T}_{ij} Z_{ij}$ where $\\mathcal{T}_{ij} $ are\nunknown, heterogenous treatment effects. The problem requires we estimate the\naverage treatment effect $\\tau^* := \\sum_{ij} \\mathcal{T}_{ij} Z_{ij} /\n\\sum_{ij} Z_{ij}$. The synthetic control paradigm provides an approach to\nestimating $\\tau^*$ when $Z$ places support on a single row. This paper extends\nthat framework to allow rate-optimal recovery of $\\tau^*$ for general $Z$, thus\nbroadly expanding its applicability. Our guarantees are the first of their type\nin this general setting. Computational experiments on synthetic and real-world\ndata show a substantial advantage over competing estimators.\n"}
{"abstract": "  The recognition of personalized content, such as contact names, remains a\nchallenging problem for end-to-end speech recognition systems. In this work, we\ndemonstrate how first and second-pass rescoring strategies can be leveraged\ntogether to improve the recognition of such words. Following previous work, we\nuse a shallow fusion approach to bias towards recognition of personalized\ncontent in the first-pass decoding. We show that such an approach can improve\npersonalized content recognition by up to 16% with minimum degradation on the\ngeneral use case. We describe a fast and scalable algorithm that enables our\nbiasing models to remain at the word-level, while applying the biasing at the\nsubword level. This has the advantage of not requiring the biasing models to be\ndependent on any subword symbol table. We also describe a novel second-pass\nde-biasing approach: used in conjunction with a first-pass shallow fusion that\noptimizes on oracle WER, we can achieve an additional 14% improvement on\npersonalized content recognition, and even improve accuracy for the general use\ncase by up to 2.5%.\n"}
{"abstract": "  Designing feasible and effective architectures under diverse computation\nbudgets incurred by different applications/devices is essential for deploying\ndeep models in practice. Existing methods often perform an independent\narchitecture search for each target budget, which is very inefficient yet\nunnecessary. Moreover, the repeated independent search manner would inevitably\nignore the common knowledge among different search processes and hamper the\nsearch performance. To address these issues, we seek to train a general\narchitecture generator that automatically produces effective architectures for\nan arbitrary budget merely via model inference. To this end, we propose a\nPareto-Frontier-aware Neural Architecture Generator (NAG) which takes an\narbitrary budget as input and produces the Pareto optimal architecture for the\ntarget budget. We train NAG by learning the Pareto frontier (i.e., the set of\nPareto optimal architectures) over model performance and computational cost\n(e.g., latency). Extensive experiments on three platforms (i.e., mobile, CPU,\nand GPU) show the superiority of the proposed method over existing NAS methods.\n"}
{"abstract": "  In oxide heterostructures, different materials are integrated into a single\nartificial crystal, resulting in a breaking of inversion-symmetry across the\nheterointerfaces. A notable example is the interface between polar and\nnon-polar materials, where valence discontinuities lead to otherwise\ninaccessible charge and spin states. This approach paved the way to the\ndiscovery of numerous unconventional properties absent in the bulk\nconstituents. However, control of the geometric structure of the electronic\nwavefunctions in correlated oxides remains an open challenge. Here, we create\nheterostructures consisting of ultrathin SrRuO$_3$, an itinerant ferromagnet\nhosting momentum-space sources of Berry curvature, and LaAlO$_3$, a polar\nwide-bandgap insulator. Transmission electron microscopy reveals an atomically\nsharp LaO/RuO$_2$/SrO interface configuration, leading to excess charge being\npinned near the LaAlO$_3$/SrRuO$_3$ interface. We demonstrate through\nmagneto-optical characterization, theoretical calculations and transport\nmeasurements that the real-space charge reconstruction modifies the\nmomentum-space Berry curvature in SrRuO$_3$, driving a reorganization of the\ntopological charges in the band structure. Our results illustrate how the\ntopological and magnetic features of oxides can be manipulated by engineering\ncharge discontinuities at oxide interfaces.\n"}
{"abstract": "  Score-based generative models and diffusion probabilistic models have been\nsuccessful at generating high-quality samples in continuous domains such as\nimages and audio. However, due to their Langevin-inspired sampling mechanisms,\ntheir application to discrete and sequential data has been limited. In this\nwork, we present a technique for training diffusion models on sequential data\nby parameterizing the discrete domain in the continuous latent space of a\npre-trained variational autoencoder. Our method is non-autoregressive and\nlearns to generate sequences of latent embeddings through the reverse process\nand offers parallel generation with a constant number of iterative refinement\nsteps. We apply this technique to modeling symbolic music and show strong\nunconditional generation and post-hoc conditional infilling results compared to\nautoregressive language models operating over the same continuous embeddings.\n"}
{"abstract": "  Over the last several decades, entangled photon pairs generated by\nspontaneous parametric down conversion processes in both second-order and\nthird-order nonlinear optical materials have been intensively studied for\nvarious quantum features such as Bell inequality violation and anticorrelation.\nIn an interferometric scheme, anticorrelation results from photon bunching\nbased on randomness when entangled photon pairs coincidently impinge on a beam\nsplitter. Compared with post-measurement-based probabilistic confirmation, a\ncoherence version has been recently proposed using the wave nature of photons.\nHere, the origin of quantum features in a coupled interferometric scheme is\ninvestigated using pure coherence optics. In addition, a deterministic method\nof entangled photon-pair generation is proposed for on-demand coherence control\nof quantum processing.\n"}
{"abstract": "  Developing novel techniques for depositing transition metal dichalcogenides\nis crucial for the industrial adoption of 2D materials in optoelectronics. In\nthis work, the lateral growth of molybdenum disulfide (MoS2) over an insulating\nsurface is demonstrated using electrochemical deposition. By fabricating a new\ntype of microelectrodes, MoS2 2D films grown from TiN electrodes across\nopposite sides have been connected over an insulating substrate, hence, forming\na lateral device structure through only one lithography and deposition step.\nUsing a variety of characterization techniques, the growth rate of MoS2 has\nbeen shown to be highly anisotropic with lateral to vertical growth ratios\nexceeding 20-fold. Electronic and photo-response measurements on the device\nstructures demonstrate that the electrodeposited MoS2 layers behave like\nsemiconductors, confirming their potential for photodetection applications.\nThis lateral growth technique paves the way towards room temperature, scalable\nand site-selective production of various transition metal dichalcogenides and\ntheir lateral heterostructures for 2D materials-based fabricated devices.\n"}
{"abstract": "  The sea-ice cover of the Arctic Ocean is an important element of the climate\nand ocean system in the Northern Hemisphere as it impacts albedo, atmospheric\npressure regimes, CO2-exchange at the ocean/atmosphere interface as well as the\nNorth Atlantic freshwater budget and thermohaline circulation [1]. Due to\nglobal warming, the Arctic sea-ice cover is presently evolving at an\nunprecedent rate towards full melt during the summer season, driving the\nso-called \"Arctic amplification\" [2]. However, the Arctic sea-ice has also\nexperienced large amplitude variations, from seasonal to orbital (Milankovitch)\ntime scales, in the past. Recent studies led to suggest that whereas insolation\nhas been a major driver of Arctic sea-ice variability through time, sea-level\nchanges governed the development of \"sea-ice factories\" over shelves (Figure\n1), thus fine-tuning the response of the Arctic Ocean to glacial/interglacial\noscillations that is slightly out of phase compared to lower latitudes [3,4].\nWe discuss below how insolation and sea-level changes may have interacted and\ncontrolled the sea-ice cover of the Arctic Ocean during warm past intervals and\nhow they could still interfere in the future.\n"}
{"abstract": "  The increase in the resistivity with decreasing temperature followed by a\ndrop by more than one order of magnitude is observed on the metallic side near\nthe zero-magnetic-field metal-insulator transition in a strongly interacting\ntwo-dimensional electron system in ultra-clean SiGe/Si/SiGe quantum wells. We\nfind that the temperature $T_{\\text{max}}$, at which the resistivity exhibits a\nmaximum, is close to the renormalized Fermi temperature, in agreement with the\ndynamical mean-field theory. However, rather than increasing along with the\nFermi temperature, the value $T_{\\text{max}}$ decreases appreciably for\nspinless electrons in spin-polarizing magnetic fields, which is in\ncontradiction with the theory in its current form. Remarkably, the\ncharacteristic scaling of the resistivity, predicted by the theory, holds in\nboth spin-unpolarized and completely spin-polarized systems.\n"}
{"abstract": "  We initiate the study of the $C^0$ symplectic mapping class group, i.e. the\ngroup of isotopy classes of symplectic homeomorphisms. We prove that none of\nthe different powers of the square of the Dehn-Seidel twist belong to the same\nconnected component of the group of symplectic homeomorphisms of certain\nLiouville domains. This generalizes to the $C^0$ setting a celebrated result of\nSeidel. In other words, we obtain the non-triviality of the $C^0$ symplectic\nmapping class group in these domains and in fact an element of infinite order.\nFor that purpose, we develop a method coming from Floer theory and the theory\nof barcodes. This builds on recent developments of $C^0$-symplectic topology.\nIn particular, we adapt and generalize to our context results by\nBuhovsky-Humili\\`ere-Seyfaddini and Kislev-Shelukhin.\n"}
{"abstract": "  In this paper, we present a qualitative study on speleology that aims to\nwiden the current understanding of people's practices in Nature and identify a\ndesign space for technology that supports such practices. Speleology is a\npractice based on the discovery, study, and dissemination of natural cavities.\nSpeleologists are amateur experts who often collaborate with scientists and\nlocal institutions to understand the geology, hydrology, and biology of a\nterritory. Their skills are at the same time physical, technical, and\ntheoretical; this is why speleology is defined as a 'sporting science'. Being\nat the boundary between outdoor adventure sports and citizen science,\nspeleology is an interesting case study for investigating the variety and\ncomplexity of activities carried out in the natural context. We interviewed 15\nexperienced speleologists to explore their goals, routines, vision of the\noutdoors, and attitude towards technology. From our study, it emerged that i)\nthe excitement of discovery and the unpredictability of an explorative trip are\nthe strongest motivations for people to engage in speleology; ii) physical\nskilfulness is a means for knowledge generation; iii) the practice is\nnecessarily collective and requires group coordination. From these findings, an\nambivalent attitude towards technology emerged: on the one hand, the scientific\nvocation of speleology welcomes technology supporting the development of\nknowledge; on the other hand, aspects typical of adventure sports lead to\nresistance to technology facilitating the physical performance. We conclude the\narticle by presenting design considerations for devices supporting speleology,\nas well as a few reflections on how communities of speleologists can inspire\ncitizen science projects.\n"}
{"abstract": "  We perform a feasibility study to search for axion-like particles (ALPs)\nusing vector boson fusion (VBF) processes at the LHC. We work in an effective\nfield theory framework with cutoff scale $\\Lambda$ and ALP mass $m_{a}$, and\nassume that ALPs couple to photons with strength $\\propto 1/\\Lambda$. Assuming\nproton-proton collisions at $\\sqrt{s} = 13$ TeV, we present the total VBF ALP\nproduction cross sections, ALP decay widths and lifetimes, and relevant\nkinematic distributions as a function of $m_{a}$ and $\\Lambda$. We consider the\n$a\\to\\gamma\\gamma$ decay mode to show that the requirement of an energetic\ndiphoton pair combined with two forward jets with large dijet mass and\npseudorapidity separation can significantly reduce the Standard Model\nbackgrounds, leading to a $5\\sigma$ discovery reach for $10 \\text{ MeV}\n\\lesssim m_{a} \\lesssim 1$ TeV with $\\Lambda \\lesssim 2$ TeV, assuming an\nintegrated luminosity of 3000 fb$^{-1}$. In particular, this extends the LHC\nsensitivity to a previously unstudied region of the ALP parameter space.\n"}
{"abstract": "  Many chemical concepts can be well defined in the context of quantum chemical\ntheories. Examples are the electronegativity scale of Mulliken and Jaffe and\nthe hard and soft acids and bases concept of Pearson. The sound theoretical\nbasis allows for a systematic definition of such concepts. However, while they\nare often used to describe and compare chemical processes in terms of\nreactivity, their predictive power remains unclear. In this work, we elaborate\non the predictive potential of chemical reactivity concepts, which can be\ncrucial for autonomous reaction exploration protocols to guide them by\nfirst-principles heuristics that expoit these concepts.\n"}
{"abstract": "  Embedding entities and relations of a knowledge graph in a low-dimensional\nspace has shown impressive performance in predicting missing links between\nentities. Although progresses have been achieved, existing methods are\nheuristically motivated and theoretical understanding of such embeddings is\ncomparatively underdeveloped. This paper extends the random walk model (Arora\net al., 2016a) of word embeddings to Knowledge Graph Embeddings (KGEs) to\nderive a scoring function that evaluates the strength of a relation R between\ntwo entities h (head) and t (tail). Moreover, we show that marginal loss\nminimisation, a popular objective used in much prior work in KGE, follows\nnaturally from the log-likelihood ratio maximisation under the probabilities\nestimated from the KGEs according to our theoretical relationship. We propose a\nlearning objective motivated by the theoretical analysis to learn KGEs from a\ngiven knowledge graph. Using the derived objective, accurate KGEs are learnt\nfrom FB15K237 and WN18RR benchmark datasets, providing empirical evidence in\nsupport of the theory.\n"}
{"abstract": "  This paper introduces an approach for formally verifying the safety of the\nflight controller of an octorotor platform. Our method involves finding regions\nof the octorotor's state space that are considered safe, and which can be\nproven to be invariant with respect to the dynamics. Specifically, exponential\nbarrier functions are used to construct candidate invariant regions near\ndesired commanded states. The proof that these regions are invariant is\ndiscovered automatically using the dReal SMT solver, which ensures the accurate\ncommand tracking of the octorotor to within a certain margin of error. Rotor\nfailures in which rotor thrusts become stuck at fixed values are considered and\naccounted for via a pseudo-inverse control allocator. The safety of the control\nallocator is verified in dReal by checking that the thrusts demanded by the\nallocator never exceed the capability of the rotors. We apply our approach on a\nspecific octorotor example and verify the desired command tracking properties\nof the controller under normal conditions and various combinations of rotor\nfailures.\n"}
{"abstract": "  Lower extremity amputees face challenges in natural locomotion, which is\npartially compensated using powered assistive systems, e.g., micro-processor\ncontrolled prosthetic leg. In this paper, a radar-based perception system is\nproposed to assist prosthetic legs for autonomous obstacle traversing, focusing\non multiple-step staircases. The presented perception system is composed of a\nradar module operating with a multiple-input-multiple-output (MIMO)\nconfiguration to localize consecutive stair corners. An inertial measurement\nunit (IMU) is integrated for coordinates correction due to the angular\ndis-positioning that occurs because of the knee angular motion. The captured\ninformation from both sensors is used for staircase dimensioning (depth and\nheight). A shallow neural network (NN) is proposed to model the error due to\nthe hardware limitations and enhance the dimension estimation accuracy (1 cm).\nThe algorithm is implemented on a microcontroller subsystem of the radar kit to\nqualify the perception system for embedded integration in powered prosthetic\nlegs.\n"}
{"abstract": "  The increasing frequency, impact, consequence and sophistication of\ncybersecurity attacks is becoming a strategic concern for boards and executive\nmanagement of organisations. Consequently, in addition to focusing on\nproductivity and performance, organisations are prioritizing Information\nSecurity Management (ISM). However, research has revealed little or no\nconceptualisation of a dynamic ISM capability and its link to organisational\nperformance. In this research, we set out to 1) define and describe an\norganisational level dynamic ISM capability, 2) to develop a strategic model\nthat links resources with this dynamic capability, and then 3) empirically\ndemonstrate how dynamic ISM capability contributes to firm performance. By\ndrawing on Resource-Based Theory (RBT) and Dynamic Capabilities View (DCV), we\nhave developed the Dynamic ISM Capability model to address the identified gap.\nAs we develop this research, we will empirically test this model to demonstrate\ncausality between ISM capability and organisational performance.\n"}
{"abstract": "  Topological lattices have recently generated a great deal of interest based\non the unique mechanical properties rooted in their topological polarization,\nincluding the ability to support localized modes at certain floppy edges. The\nstudy of these systems has been predominantly restricted to the realm of\nin-plane mechanics, to which many topological effects are germane. In this\nstudy, we stretch this paradigm by exploring the possibility to export certain\ntopological attributes to the flexural wave behavior of thin lattice sheets. To\ncouple the topological modes to the out-of-plane response, we assemble a\nbilayer lattice by stacking a thick topological kagome layer onto a thin\ntwisted kagome lattice. The band diagram reveals the existence of modes whose\nout-of-plane character is controlled by the edge modes of the topological\nlayer, a behavior elucidated via simulations and confirmed via laser vibrometer\nexperiments on a bilayer prototype specimen. These results open an alternative\ndirection for topological mechanics whereby flexural waves are controlled by\nthe in-plane topology, leading to potential applications for flexural wave\ndevices with engineered polarized response.\n"}
{"abstract": "  The idea of using a Reconfigurable Intelligent Surface (RIS) consisting of a\nlarge array of passive scattering elements to assist wireless communication\nsystems has recently attracted much attention from academia and industry. A\ncentral issue with RIS is how much power they can effectively convey to the\ntarget radio nodes. Regarding this question, several power level models exist\nin the literature but few have been validated through experiments. In this\npaper, we propose a radar cross section-based received power model for an\nRIS-aided wireless communication system that is rooted in the physical\nproperties of RIS. Our proposed model follows the intuition that the received\npower is related to the distances from the transmitter/receiver to the RIS, the\nangles in the TX-RIS-RX triangle, the effective area of each element, and the\nreflection coefficient of each element. To the best of our knowledge, this\npaper is the first to model the angle-dependent phase shift of the reflection\ncoefficient, which is typically ignored in existing literature. We further\nmeasure the received power with our experimental platform in different\nscenarios to validate our model. The measurement results show that our model is\nappropriate both in near field and far field and can characterize the impact of\nangles well.\n"}
{"abstract": "  Novae and supernovae play a key role in many fields of Astrophysics and\nCosmology. Despite their importance, an accurate description of which objects\nexplode and why and how they explode is still lacking. One of the main\ncharacteristics of such explosions is that they are the main suppliers of newly\nsynthesized chemical elements in the Galaxy. Since some of these isotopes are\nradioactive, it is possible to use the corresponding gamma-rays as a diagnostic\ntool of the explosion thanks to their independence on the thermal state of the\ndebris. The drawback is the poor sensitivity of detectors in the MeV energy\ndomain. As a consequence, the radioactive lines have only been detected in one\ncore collapse supernova (SN 1987A), one Type Ia supernova (SN 2014J), and one\nsupernova remnant (Cas A). Nevertheless these observations have provided and\nare providing important information about the explosion mechanisms.\nUnfortunately, novae are still eluding detection. These results emphasize the\nnecessity to place as soon as possible a new instrument in orbit with enough\nsensitivity to noticeably enlarge the sample of detected events\n"}
{"abstract": "  Simplicial complexes form an important class of topological spaces that are\nfrequently used to in many applications areas such as computer-aided design,\ncomputer graphics, and simulation. The representation learning on graphs, which\nare just 1-d simplicial complexes, has witnessed a great attention and success\nin the past few years. Due to the additional complexity higher dimensional\nsimplicial hold, there has not been enough effort to extend representation\nlearning to these objects especially when it comes to learn entire-simplicial\ncomplex representation. In this work, we propose a method for simplicial\ncomplex-level representation learning that embeds a simplicial complex to a\nuniversal embedding space in a way that complex-to-complex proximity is\npreserved. Our method utilizes a simplex-level embedding induced by a\npre-trained simplicial autoencoder to learn an entire simplicial complex\nrepresentation. To the best of our knowledge, this work presents the first\nmethod for learning simplicial complex-level representation.\n"}
{"abstract": "  Interaction-aware planning for autonomous driving requires an exploration of\na combinatorial solution space when using conventional search- or\noptimization-based motion planners. With Deep Reinforcement Learning, optimal\ndriving strategies for such problems can be derived also for higher-dimensional\nproblems. However, these methods guarantee optimality of the resulting policy\nonly in a statistical sense, which impedes their usage in safety critical\nsystems, such as autonomous vehicles. Thus, we propose the\nExperience-Based-Heuristic-Search algorithm, which overcomes the statistical\nfailure rate of a Deep-reinforcement-learning-based planner and still benefits\ncomputationally from the pre-learned optimal policy. Specifically, we show how\nexperiences in the form of a Deep Q-Network can be integrated as heuristic into\na heuristic search algorithm. We benchmark our algorithm in the field of path\nplanning in semi-structured valet parking scenarios. There, we analyze the\naccuracy of such estimates and demonstrate the computational advantages and\nrobustness of our method. Our method may encourage further investigation of the\napplicability of reinforcement-learning-based planning in the field of\nself-driving vehicles.\n"}
{"abstract": "  Recent studies show that the inner Galactic regions host genuine bulge\nglobular clusters, but also halo intruders, complex remnants of primordial\nbuilding blocks, and objects likely accreted during major merging events. In\nthis study we focus on the properties of M 28, a very old and massive cluster\ncurrently located in the Galactic bulge. We analysed wide-field infrared\nphotometry collected by the VVV survey, VVV proper motions, and\nintermediate-resolution spectra in the calcium triplet range for 113 targets in\nthe cluster area. Our results in general confirm previous estimates of the\ncluster properties available in the literature. We find no evidence of\ndifferences in metallicity between cluster stars, setting an upper limit of\nDelta[Fe/H]<0.08 dex to any internal inhomogeneity. We confirm that M 28 is one\nof the oldest objects in the Galactic bulge (13-14 Gyr). From this result and\nthe literature data, we find evidence of a weak age-metallicity relation among\nbulge globular clusters that suggests formation and chemical enrichment. In\naddition, wide-field density maps show that M 28 is tidally stressed and that\nit is losing mass into the general bulge field. Our study indicates that M 28\nis a genuine bulge globular cluster, but its very old age and its mass loss\nsuggest that this cluster could be the remnant of a larger structure, possibly\na primeval bulge building block.\n"}
{"abstract": "  The Laser Interferometer Space Antenna (LISA) mission, scheduled for launch\nin the early 2030s, is a gravitational wave observatory in space designed to\ndetect sources emitting in the milli-Hertz band. In contrast to the present\nground based detectors, the LISA data are expected to be a signaldominated,\nwith strong and weak gravitational wave signals overlapping in time and in\nfrequency. Astrophysical population models predict a sufficient number of\nsignals in the LISA band to blend together and form an irresolvable foreground\nnoise. In this work, we present a generic method for characterizing the\nforeground signals originating from a given astrophysical population of\ncoalescing compact binaries. Assuming idealized detector conditions and perfect\ndata analysis technique capable of identifying and removing the bright sources,\nwe apply an iterative procedure which allows us to predict the different levels\nof foreground noise.\n"}
{"abstract": "  Graphene is a material with enormous potential for numerous applications.\nTherefore, significant efforts are dedicated to large-scale graphene production\nusing a chemical vapor deposition (CVD) technique. In addition, research is\ndirected at developing methods to incorporate graphene in established\nproduction technologies and process flows. In this paper, we present a brief\nreview of available CVD methods for graphene synthesis. We also discuss\nscalable methods to transfer graphene onto desired substrates. Finally, we\ndiscuss potential applications that would benefit from a fully scaled,\nsemiconductor technology compatible production process.\n"}
{"abstract": "  We consider the self-adjoint Schr\\\"odinger operator in $L^2(\\mathbb{R}^d)$,\n$d\\geq 2$, with a $\\delta$-potential supported on a hyperplane\n$\\Sigma\\subseteq\\mathbb{R}^d$ of strength $\\alpha=\\alpha_0+\\alpha_1$, where\n$\\alpha_0\\in\\mathbb{R}$ is a constant and $\\alpha_1\\in L^p(\\Sigma)$ is a\nnonnegative function. As the main result, we prove that the lowest spectral\npoint of this operator is not smaller than that of the same operator with\npotential strength $\\alpha_0+\\alpha_1^*$, where $\\alpha_1^*$ is the symmetric\ndecreasing rearrangement of $\\alpha_1$. The proof relies on the\nBirman-Schwinger principle and the reduction to an analogue of the\nP\\'{o}lya-Szeg\\H{o} inequality for the relativistic kinetic energy in\n$\\mathbb{R}^{d-1}$.\n"}
{"abstract": "  Manual evaluation is essential to judge progress on automatic text\nsummarization. However, we conduct a survey on recent summarization system\npapers that reveals little agreement on how to perform such evaluation studies.\nWe conduct two evaluation experiments on two aspects of summaries' linguistic\nquality (coherence and repetitiveness) to compare Likert-type and ranking\nannotations and show that best choice of evaluation method can vary from one\naspect to another. In our survey, we also find that study parameters such as\nthe overall number of annotators and distribution of annotators to annotation\nitems are often not fully reported and that subsequent statistical analysis\nignores grouping factors arising from one annotator judging multiple summaries.\nUsing our evaluation experiments, we show that the total number of annotators\ncan have a strong impact on study power and that current statistical analysis\nmethods can inflate type I error rates up to eight-fold. In addition, we\nhighlight that for the purpose of system comparison the current practice of\neliciting multiple judgements per summary leads to less powerful and reliable\nannotations given a fixed study budget.\n"}
{"abstract": "  We prove existence results of two solutions of the problem \\[ \\begin{cases}\nL(u)+u^{m-1}=\\lambda u^{p-1} & \\text{ in $\\Omega$}, \\\\ \\quad u>0 &\\text{ in\n$\\Omega$}, \\\\ \\quad u=0 & \\text{ on $\\partial \\Omega$}, \\end{cases} \\] where\n$L(v)=-{\\rm div}(M(x)\\nabla v)$ is a linear operator, $p\\in (2,2^{*}]$ and\n$\\lambda$ and $ m$ sufficiently large. Then their asymptotical limit as $m\\to\n+\\infty$ is investigated showing different behaviors.\n"}
{"abstract": "  Human language has been described as a system that makes \\textit{use of\nfinite means to express an unlimited array of thoughts}. Of particular interest\nis the aspect of compositionality, whereby, the meaning of a compound language\nexpression can be deduced from the meaning of its constituent parts. If\nartificial agents can develop compositional communication protocols akin to\nhuman language, they can be made to seamlessly generalize to unseen\ncombinations. Studies have recognized the role of curiosity in enabling\nlinguistic development in children. In this paper, we seek to use this\nintrinsic feedback in inducing a systematic and unambiguous protolanguage. We\ndemonstrate how compositionality can enable agents to not only interact with\nunseen objects but also transfer skills from one task to another in a zero-shot\nsetting: \\textit{Can an agent, trained to `pull' and `push twice', `pull\ntwice'?}.\n"}
{"abstract": "  We show that the property of being rationally $K$-stable passes from the\nfibers of a continuous $C(X)$-algebra to the ambient algebra, under the\nassumption that the underlying space $X$ is compact, metrizable, and of finite\ncovering dimension. As an application, we show that a crossed product\nC*-algebra is (rationally) $K$-stable provided the underlying C*-algebra is\n(rationally) $K$-stable, and the action has finite Rokhlin dimension with\ncommuting towers.\n"}
{"abstract": "  A physically-based method to derive well-posed instances of the two-fluid\ntransport equations for two-phase flow, from the Hamilton principle, is\npresented. The state of the two-fluid flow is represented by the superficial\nvelocity and the drift-flux, instead of the average velocities of each fluid.\nThis generates the conservation equations of the two principal motion modes\nnaturally: the global center-of-mass flow and the relative velocity between\nfluids. Well-posed equations can be obtained by modelling the storage of\nkinetic energy in fluctuations structures induced by the interaction between\nfluids, like wakes and vortexes. In this way, the equations can be regularized\nwithout losing in the process the instabilities responsible for flow-patterns\nformation and transition. A specific case of vertical air-water flow is\nanalyzed showing the capability of the present model to predict the formation\nof the slug flow regime as trains of non-linear waves.\n"}
{"abstract": "  Modern risk modelling approaches deal with vectors of multiple components.\nThe components could be, for example, returns of financial instruments or\nlosses within an insurance portfolio concerning different lines of business.\nOne of the main problems is to decide if there is any type of dependence\nbetween the components of the vector and, if so, what type of dependence\nstructure should be used for accurate modelling.\n  We study a class of heavy-tailed multivariate random vectors under a\nnon-parametric shape constraint on the tail decay rate. This class contains,\nfor instance, elliptical distributions whose tail is in the intermediate\nheavy-tailed regime, which includes Weibull and lognormal type tails. The study\nderives asymptotic approximations for tail events of random walks.\nConsequently, a full large deviations principle is obtained under, essentially,\nminimal assumptions. As an application, an optimisation method for a large\nclass of Quota Share (QS) risk sharing schemes used in insurance and finance is\nobtained.\n"}
{"abstract": "  Convolutional neural networks (CNNs) have been used quite successfully for\nsemantic segmentation of brain tumors. However, current CNNs and attention\nmechanisms are stochastic in nature and neglect the morphological indicators\nused by radiologists to manually annotate regions of interest. In this paper,\nwe introduce a channel and spatial wise asymmetric attention (CASPIAN) by\nleveraging the inherent structure of tumors to detect regions of saliency. To\ndemonstrate the efficacy of our proposed layer, we integrate this into a\nwell-established convolutional neural network (CNN) architecture to achieve\nhigher Dice scores, with less GPU resources. Also, we investigate the inclusion\nof auxiliary multiscale and multiplanar attention branches to increase the\nspatial context crucial in semantic segmentation tasks. The resulting\narchitecture is the new CASPIANET++, which achieves Dice Scores of 91.19% whole\ntumor, 87.6% for tumor core and 81.03% for enhancing tumor. Furthermore, driven\nby the scarcity of brain tumor data, we investigate the Noisy Student method\nfor segmentation tasks. Our new Noisy Student Curriculum Learning paradigm,\nwhich infuses noise incrementally to increase the complexity of the training\nimages exposed to the network, further boosts the enhancing tumor region to\n81.53%. Additional validation performed on the BraTS2020 data shows that the\nNoisy Student Curriculum Learning method works well without any additional\ntraining or finetuning.\n"}
{"abstract": "  We present the first provable Least-Squares Value Iteration (LSVI) algorithms\nthat have runtime complexity sublinear in the number of actions. We formulate\nthe value function estimation procedure in value iteration as an approximate\nmaximum inner product search problem and propose a locality sensitive hashing\n(LSH) [Indyk and Motwani STOC'98, Andoni and Razenshteyn STOC'15, Andoni,\nLaarhoven, Razenshteyn and Waingarten SODA'17] type data structure to solve\nthis problem with sublinear time complexity. Moreover, we build the connections\nbetween the theory of approximate maximum inner product search and the regret\nanalysis of reinforcement learning. We prove that, with our choice of\napproximation factor, our Sublinear LSVI algorithms maintain the same regret as\nthe original LSVI algorithms while reducing the runtime complexity to sublinear\nin the number of actions. To the best of our knowledge, this is the first work\nthat combines LSH with reinforcement learning resulting in provable\nimprovements. We hope that our novel way of combining data-structures and\niterative algorithm will open the door for further study into cost reduction in\noptimization.\n"}
{"abstract": "  The aim of this paper is to derive a new uncertainty principle for the\ngeneralized $q$-Bessel wavelet transform studied earlier in \\cite{Rezguietal}.\nIn this paper, an uncertainty principle associated with wavelet transforms in\nthe $q$-calculus framework has been established. A two-parameters extension of\nthe classical Bessel operator is applied to generate a wavelet function which\nis exploited next to explore a wavelet uncertainty principle already in the\n$q$-calculus framework.\n"}
{"abstract": "  We perform high-fidelity, two-dimensional (2D), fluid-structure interaction\n(FSI) simulations at a Reynolds number of $Re=200$ of uniform flow past an\ninverted flag (i.e., clamped at its trailing edge). The inverted flag system\ncan exhibit large-amplitude flapping motions (on the order of the flag length)\nthat can be converted to electricity via, e.g., piezoelectric materials. We\ninvestigate the effect of structural nonuniformity in altering the FSI dynamics\ncompared with the uniform-stiffness scenario that has been thoroughly\ncharacterized. We consider linear, quadratic, and cubic stiffness\ndistributions, and demonstrate that the FSI dynamics mirror those of a\nuniform-stiffness flag with an appropriately defined effective stiffness. We\nshow that this effective stiffness can be computed simply via analysis of an\n\\emph{in-vacuo} Euler-Bernoulli beam. When expressed in terms of the effective\nstiffness, the FSI dynamics of the nonuniform-stiffness flag exhibit the same\nregimes -- with many similarities in the detailed dynamics -- as a\nuniform-stiffness flag. This study opens questions about (i) what the optimal\nstiffness distribution is for, e.g., energy harvesting capacity, and (ii) how\nto use nonuniform (and possibly time-varying) stiffness distributions to\ncontrol the flag dynamics towards a desired state.\n"}
{"abstract": "  We report microscale friction experiments for diamond/metal and\ndiamond/silica contacts under gigapascal contact pressures. Using a new\nnanoprobe technique which has sufficient dynamic range of force and stiffness,\nwe demonstrate the processes involved in the transition from purely interface\nsliding at the nanoscale to the situation where at least one of the sliding\nbodies undergoes some plastic deformation. For sliding of micrometer-scale\ndiamond spherical tips on metallic substrates, additional local plastic\nyielding of the substrate resulting from tangential tractions causes the tip to\nsink into the surface, increasing the contact area in the direction of loading\nand resulting in a static friction coefficient higher than the kinetic during\nploughing. This sink-in is largely absent in fused silica, and no friction drop\nis observed, along with lower friction in general. The transition from sinking\nin within the static friction regime to ploughing in the sliding friction\nregime is mediated by failure of the contact interface, indicated by a sharp\nincrease in energy dissipation. At lower contact pressures, the elastic\ninterfacial sliding behaviour characteristic of scanning probe or surface force\napparatus experiments is recovered, bridging the gap between the exotic realm\nof nanotribology and plasticity dominated macroscale friction. We delineate the\nmaterial and geometric factors which determine the transition. We also find\nthat some unexpected light is cast on the origins of the difference between\nstatic and dynamic friction.\n"}
{"abstract": "  The Standard Model Extension (SME) is a generic parametrization for Lorentz\nviolation and the phenomenological consequences of the minimal gravity sector\nof the SME are usually studied using a post-Newtonian expansion that requires\nspacetime to be asymptotically flat. However, there is a term in this sector\nfor which these approximations are unable to make predictions; this is known as\nthe $t$ puzzle. The present paper studies a model of spontaneous Lorentz\nviolation in the minimal gravity sector of the SME in a static and spherically\nsymmetric situation, when no additional matter fields are present. It is shown\nthat, under the above mentioned assumptions, $t$ is the only term in the\nminimal gravity sector for which no asymptotically flat solutions exist. This\nstems from the fact that the $t$ term fixes the asymptotic behavior of all the\npieces of the curvature tensor.\n"}
{"abstract": "  We present the first characterization of the circumgalactic medium of\nLy$\\alpha$ emitters (LAEs), using a sample of 96 $z\\approx3.3$ LAEs detected\nwith the VLT/MUSE in fields centered on 8 bright background quasars. The LAEs\nhave low Ly$\\alpha$ luminosities ($\\sim 10^{42}\\,\\text{erg}\\,\\text{s}^{-1}$)\nand star formation rates (SFRs) $\\sim 1~\\text{M}_\\odot\\,\\text{yr}^{-1}$, which\nfor main sequence galaxies corresponds to stellar masses of only $\\sim\n10^{8.6}\\,\\text{M}_\\odot$. The median transverse distance between the LAEs and\nthe quasar sightlines is 165 proper kpc (pkpc). We stacked the high-resolution\nquasar spectra and measured significant excess HI and CIV absorption near the\nLAEs out to 500 $\\text{km}\\,\\text{s}^{-1}$ and at least $\\approx 250$ pkpc\n(corresponding to $\\approx 7$ virial radii). At $\\lesssim\n30~\\text{km}\\,\\text{s}^{-1}$ from the galaxies the median HI and CIV optical\ndepths are enhanced by an order of magnitude. The absorption is significantly\nstronger around the $\\approx 1/3$ of our LAEs that are part of `groups', which\nwe attribute to the large-scale structures in which they are embedded. We do\nnot detect any strong dependence of either the HI or CIV absorption on\ntransverse distance (over the range $\\approx 50-250$ pkpc), redshift, or the\nproperties of the Ly$\\alpha$ emission line (luminosity, full width at half\nmaximum, or equivalent width). However, for HI, but not CIV, the absorption at\n$\\lesssim 100\\,\\text{km}\\,\\text{s}^{-1}$ from the LAE does increase with the\nSFR. This suggests that LAEs surrounded by more HI tend to have higher SFRs.\n"}
{"abstract": "  We consider the $k$-clustering problem with $\\ell_p$-norm cost, which\nincludes $k$-median, $k$-means and $k$-center cost functions, under an\nindividual notion of fairness proposed by Jung et al. [2020]: given a set of\npoints $P$ of size $n$, a set of $k$ centers induces a fair clustering if for\nevery point $v\\in P$, $v$ can find a center among its $n/k$ closest neighbors.\nRecently, Mahabadi and Vakilian [2020] showed how to get a\n$(p^{O(p)},7)$-bicriteria approximation for the problem of fair $k$-clustering\nwith $\\ell_p$-norm cost: every point finds a center within distance at most $7$\ntimes its distance to its $(n/k)$-th closest neighbor and the $\\ell_p$-norm\ncost of the solution is at most $p^{O(p)}$ times the cost of an optimal fair\nsolution. In this work, for any $\\varepsilon>0$, we present an improved $(16^p\n+\\varepsilon,3)$-bicriteria approximation for the fair $k$-clustering with\n$\\ell_p$-norm cost. To achieve our guarantees, we extend the framework of\n[Charikar et al., 2002, Swamy, 2016] and devise a $16^p$-approximation\nalgorithm for the facility location with $\\ell_p$-norm cost under matroid\nconstraint which might be of an independent interest. Besides, our approach\nsuggests a reduction from our individually fair clustering to a clustering with\na group fairness requirement proposed by Kleindessner et al. [2019], which is\nessentially the median matroid problem [Krishnaswamy et al., 2011].\n"}
{"abstract": "  Autonomous driving vehicles and robotic systems rely on accurate perception\nof their surroundings. Scene understanding is one of the crucial components of\nperception modules. Among all available sensors, LiDARs are one of the\nessential sensing modalities of autonomous driving systems due to their active\nsensing nature with high resolution of sensor readings. Accurate and fast\nsemantic segmentation methods are needed to fully utilize LiDAR sensors for\nscene understanding. In this paper, we present Lite-HDSeg, a novel real-time\nconvolutional neural network for semantic segmentation of full $3$D LiDAR point\nclouds. Lite-HDSeg can achieve the best accuracy vs. computational complexity\ntrade-off in SemanticKitti benchmark and is designed on the basis of a new\nencoder-decoder architecture with light-weight harmonic dense convolutions as\nits core. Moreover, we introduce ICM, an improved global contextual module to\ncapture multi-scale contextual features, and MCSPN, a multi-class Spatial\nPropagation Network to further refine the semantic boundaries. Our experimental\nresults show that the proposed method outperforms state-of-the-art semantic\nsegmentation approaches which can run real-time, thus is suitable for robotic\nand autonomous driving applications.\n"}
{"abstract": "  Automotive engineering is recognized as a combination of software and\nmechanical engineering due to the ever-increasing number of software-based\ncomponents in vehicles. Since vehicles have become more sophisticated than\nbefore to ensure robustness, testing of automotive electronics is performed in\nhigh volume, producing immense test-related data.\n  This study investigates how unstructured and decentralized test-related data\nfrom testing of automotive electronics creates issues in decision making during\nthe testing and analysis process of test artifacts by performing an exploratory\ncase-study at one of the leading automotive companies, Volvo Cars. From the\nfindings of the exploratory study, a prototype was designed to improve the data\nand information structure and presentation for test analysis and diagnostics\nfor automotive electronics. The prototype's results showed that providing\nbetter data and information structure significantly increases the efficiency\nand reduces the workload for testers when conducting test analysis and\ndiagnostics. Testers showed a decrease in task load for tasks related to\ntesting due to better information structure, presentation, correctness and\naccessibility. Hence, the improvements aided the testers to arrive at decisions\nregarding root cause analysis of failed tests efficiently. The findings of this\nstudy can assist automotive companies in systematically investigating and\nimproving the testing process of automotive electronics in regards to managing\nand structuring test-related data.\n  Keywords: Testing, Automotive Electronics, Electronic Control Unit, ECU,\nUnstructured Data.\n"}
{"abstract": "  We study Birkhoff sums as distributions. We obtain regularity results on such\ndistributions for various dynamical systems with hyperbolicity, as hyperbolic\nlinear maps on the torus and piecewise expanding maps on the interval. We also\ngive some applications, as the study of advection in discrete dynamical\nsystems.\n"}
{"abstract": "  Although several impurity solvers in the dynamical mean field theory (DMFT)\nhave been proposed, especially in multi-band systems, there are practical\ndifficulties arising from a trade-off between numerical costs and reliability.\nIn this study, we re-interpret the iterative perturbation theory (IPT) as an\napproximation which captures the strong correlation effects by mimicking the\nparticular frequency structures of the exact full vertex, and extend it such\nthat it can have efficiency and reliability simultaneously by modifying IPT\nvertex using the parquet equations. We apply this method to several models to\nevaluate their validity. We confirm that our method shows good agreements with\nthe numerically exact continuous-time quantum Monte Carlo method in the\nsingle-site DMFT calculation.\n"}
{"abstract": "  With the large-scale deployment of industrial internet of things (IIoT)\ndevices, the number of vulnerabilities that threaten IIoT security is also\ngrowing dramatically, including a mass of undisclosed IIoT vulnerabilities that\nlack mitigation measures. Coordination Vulnerabilities Disclosure (CVD) is one\nof the most popular vulnerabilities sharing solutions, in which some security\nworkers (SWs) can develop undisclosed vulnerabilities patches together.\nHowever, CVD assumes that sharing participants (SWs) are all honest, and thus\noffering chances for dishonest SWs to leak undisclosed IIoT vulnerabilities. To\ncombat such threats, we propose an Undisclosed IIoT Vulnerabilities Trusted\nSharing Protection (UIV-TSP) scheme with dynamic token. In this article, a\ndynamic token is an implicit access credential for an SW to acquire an\nundisclosed vulnerability information, which is only held by the system and\nconstantly updated as the SW access. Meanwhile, the latest updated token can be\nstealthily sneaked into the acquired information as the traceability token.\nOnce the undisclosed vulnerability information leaves the SW host, the embedded\nself-destruct program will be automatically triggered to prevent leaks since\nthe destination MAC address in the traceability token has changed. To quickly\ndistinguish dishonest SWs, trust mechanism is adopted to evaluate the trust\nvalue of SWs. Moreover, we design a blockchain-assisted continuous logs storage\nmethod to achieve the tamper-proofing of dynamic token and the transparency of\nundisclosed IIoT vulnerabilities sharing. The simulation results indicate that\nour proposed scheme is resilient to suppress dishonest SWs and protect the IoT\nundisclosed vulnerabilities effectively.\n"}
{"abstract": "  In this work, we introduce X-FACT: the largest publicly available\nmultilingual dataset for factual verification of naturally existing real-world\nclaims. The dataset contains short statements in 25 languages and is labeled\nfor veracity by expert fact-checkers. The dataset includes a multilingual\nevaluation benchmark that measures both out-of-domain generalization, and\nzero-shot capabilities of the multilingual models. Using state-of-the-art\nmultilingual transformer-based models, we develop several automated\nfact-checking models that, along with textual claims, make use of additional\nmetadata and evidence from news stories retrieved using a search engine.\nEmpirically, our best model attains an F-score of around 40%, suggesting that\nour dataset is a challenging benchmark for evaluation of multilingual\nfact-checking models.\n"}
{"abstract": "  We describe computer searches that prove the graph reconstruction conjecture\nfor graphs with up to 13 vertices and some limited classes on larger sizes. We\nalso investigate the reconstructibility of tournaments up to 13 vertices,\ndigraphs up to 9 vertices, and posets up to 13 points. In all cases, our\nresults also apply to the set reconstruction problem that uses the\nisomorph-reduced deck.\n"}
{"abstract": "  In atomic physics, the Hund rule says that the largest spin and orbital state\nis realized due to the interplay of the spin-orbit coupling (SOC) and the\nCoulomb interactions. Here, we show that in ferromagnetic solids the effective\nSOC and the orbital magnetic moment can be dramatically enhanced by a factor of\n$1/[1-(2U^\\prime-U-J_H)\\rho_0]$, where $U$ and $U^\\prime$ are the on-site\nCoulomb interaction within the same oribtals and between different orbitals,\nrespectively, $J_H$ is the Hund coupling, and $\\rho_0$ is the average density\nof states. This factor is obtained by using the two-orbital as well as\nfive-orbital Hubbard models with SOC. We also find that the spin polarization\nis more favorable than the orbital polarization, being consistent with\nexperimental observations. This present work provides a fundamental basis for\nunderstanding the enhancements of SOC and orbital moment by Coulomb\ninteractions in ferromagnets, which would have wide applications in\nspintronics.\n"}
{"abstract": "  The possibility of formation of Bose-Einstein Condensation (BEC) is studied\nin $pp$ collisions at $\\sqrt s$ = 7 TeV at the Large Hadron Collider. A\nthermodynamically consistent non-extensive formulation of the identified hadron\ntransverse momentum distributions is used to estimate the critical temperature\nrequired to form BEC of charged pions, which are the most abundant species in a\nmulti-particle production process in hadronic and nuclear collisions. The\nobtained results have been contrasted with the systems produced in Pb-Pb\ncollisions to have a better understanding. We observe an explicit dependency of\nBEC critical temperature and number of particles in the pion condensates on the\nnon-extensive parameter $q$, which is a measure of degree of non-equilibrium --\nas $q$ decreases, the critical temperature increases and approaches to the\ncritical temperature obtained from Bose-Einstein statistics without\nnon-extensivity. Studies are performed on the final state multiplicity\ndependence of number of particles in the pion condensates in a wide range of\nmultiplicity covering hadronic and heavy-ion collisions, using the inputs from\nexperimental transverse momentum spectra.\n"}
{"abstract": "  One of the last unexplored windows to the cosmos, the Dark Ages and Cosmic\nDawn, can be opened using a simple low frequency radio telescope from the\nstable, quiet lunar farside to measure the Global 21-cm spectrum. This frontier\nremains an enormous gap in our knowledge of the Universe. Standard models of\nphysics and cosmology are untested during this critical epoch. The messenger of\ninformation about this period is the 1420 MHz (21-cm) radiation from the\nhyperfine transition of neutral hydrogen, Doppler-shifted to low radio\nastronomy frequencies by the expansion of the Universe. The Global 21-cm\nspectrum uniquely probes the cosmological model during the Dark Ages plus the\nevolving astrophysics during Cosmic Dawn, yielding constraints on the first\nstars, on accreting black holes, and on exotic physics such as dark\nmatter-baryon interactions. A single low frequency radio telescope can measure\nthe Global spectrum between ~10-110 MHz because of the ubiquity of neutral\nhydrogen. Precise characterizations of the telescope and its surroundings are\nrequired to detect this weak, isotropic emission of hydrogen amidst the bright\nforeground Galactic radiation. We describe how two antennas will permit\nobservations over the full frequency band: a pair of orthogonal wire antennas\nand a 0.3-m$^3$ patch antenna. A four-channel correlation spectropolarimeter\nforms the core of the detector electronics. Technology challenges include\nadvanced calibration techniques to disentangle covariances between a bright\nforeground and a weak 21-cm signal, using techniques similar to those for the\nCMB, thermal management for temperature swings of >250C, and efficient power to\nallow operations through a two-week lunar night. This simple telescope sets the\nstage for a lunar farside interferometric array to measure the Dark Ages power\nspectrum.\n"}
{"abstract": "  Pre-trained language models (PrLM) has been shown powerful in enhancing a\nbroad range of downstream tasks including various dialogue related ones.\nHowever, PrLMs are usually trained on general plain text with common language\nmodel (LM) training objectives, which cannot sufficiently capture dialogue\nexclusive features due to the limitation of such training setting, so that\nthere is an immediate need to fill the gap between a specific dialogue task and\nthe LM task. As it is unlikely to collect huge dialogue data for\ndialogue-oriented pre-training, in this paper, we propose three strategies to\nsimulate the conversation features on general plain text. Our proposed method\ndiffers from existing post-training methods that it may yield a general-purpose\nPrLM and does not individualize to any detailed task while keeping the\ncapability of learning dialogue related features including speaker awareness,\ncontinuity and consistency. The resulted Dialog-PrLM is fine-tuned on three\npublic multi-turn dialogue datasets and helps achieve significant and\nconsistent improvement over the plain PrLMs.\n"}
{"abstract": "  A prevalent theory circulating among the non-scientific community is that the\nintensive deployment of base stations over the territory significantly\nincreases the level of electromagnetic field (EMF) exposure and affects\npopulation health. To alleviate this concern, in this work, we propose a\nnetwork architecture that introduces tethered unmanned aerial vehicles (TUAVs)\ncarrying green antennas to minimize the EMF exposure while guaranteeing a high\ndata rate for users. In particular, each TUAV can attach itself to one of the\npossible ground stations at the top of some buildings. The location of the\nTUAVs, transmit power of user equipment and association policy are optimized to\nminimize the EMF exposure. Unfortunately, the problem turns out to be\nmixed-integer non-linear programming (MINLP), which is non-deterministic\npolynomial-time (NP) hard. We propose an efficient low-complexity algorithm\ncomposed of three submodules. Firstly, we propose an algorithm based on the\ngreedy principle to determine the optimal association matrix between the users\nand base stations. Then, we offer two approaches, a modified K-mean and shrink\nand realign (SR) process, to associate each TUAV with a ground station.\nFinally, we put forward two algorithms based on the golden search and SR\nprocess to adjust the TUAV's position within the hovering area over the\nbuilding. After that, we consider the dual problem that maximizes the sum rate\nwhile keeping the exposure below a predefined value, such as the level enforced\nby the regulation. Next, we perform extensive simulations to show the\neffectiveness of the proposed TUAVs to reduce the exposure compared to various\narchitectures. Eventually, we show that TUAVs with green antennas can\neffectively mitigate the EMF exposure by more than 20% compared to fixed green\nsmall cells while achieving a higher data rate.\n"}
{"abstract": "  Astronomy has always been propelled by the discovery of new phenomena lacking\nprecedent, often followed by new theories to explain their existence and\nproperties. In the modern era of large surveys tiling the sky at ever high\nprecision and sampling rates, these serendipitous discoveries look set to\ncontinue, with recent examples including Boyajian's Star, Fast Radio Bursts and\n`Oumuamua. Accordingly, we here look ahead and aim to provide a statistical\nframework for interpreting such events and providing guidance to future\nobservations, under the basic premise that the phenomenon in question\nstochastically repeat at some unknown, constant rate, $\\lambda$. Specifically,\nexpressions are derived for 1) the a-posteriori distribution for $\\lambda$, 2)\nthe a-posteriori distribution for the recurrence time, and, 3) the\nbenefit-to-cost ratio of further observations relative to that of the inaugural\nevent. Some rule-of-thumb results for each of these are found to be 1) $\\lambda\n< \\{0.7, 2.3, 4.6\\}\\,t_1^{-1}$ to $\\{50, 90, 95\\}\\%$ confidence (where $t_1=$\ntime to obtain the first detection), 2) the recurrence time is $t_2 < \\{1, 9,\n99\\}\\,t_1$ to $\\{50, 90, 95\\}\\%$ confidence, with a lack of repetition by time\n$t_2$ yielding a $p$-value of $1/[1+(t_2/t_1)]$, and, 3) follow-up for\n$\\lesssim 10\\,t_1$ is expected to be scientifically worthwhile under an array\nof differing assumptions about the object's intrinsic scientific value. We\napply these methods to the Breakthrough Listen Candidate 1 signal and tidal\ndisruption events observed by TESS.\n"}
{"abstract": "  More than 50 years after the discovery of pulsars and confirmation of their\nassociation with supernova explosions, the origin of the initial spin and\nvelocity of pulsars remains largely a mystery. The typical space velocities of\nseveral hundred km/s have been attributed to \"kicks\" resulting from asymmetries\neither in the supernova ejecta or in the neutrino emission. Observations have\nshown a strong tendency for alignment of the pulsar space velocity and spin\naxis in young pulsars but, up to now, these comparisons have been restricted to\ntwo dimensions. We report here the first evidence for three-dimensional\nalignment between the spin and velocity vectors, largely based on observations\nmade with the Five-hundred-meter Aperture Spherical radio Telescope of the\npulsar PSR~J0538+2817 and its associated supernova remnant S147. Analysis of\nthese and related observations has enabled us to determine the location of the\npulsar within the supernova remnant and hence its radial velocity. Current\nsimulations of supernova explosions have difficulty producing such\nthree-dimensional alignment. Our results, which depend on the unprecedented\nsensitivity of the new observations, add another dimension to the intriguing\ncorrelation between pulsar spin-axis and birth-kick directions, thus deepening\nthe mysteries surrounding the birth of neutron stars.\n"}
{"abstract": "  By specializing the parameters in the partition function of the 8VSOS model\nwith domain wall boundary conditions and diagonal reflecting end, we find\nconnections between the three-color model and certain polynomials $p_n(z)$ of\nBazhanov and Mangazeev appearing in the eigenvectors of the Hamiltonian of the\nXYZ spin chain. This work is a continuation of a previous paper where we\ninvestigated the related polynomials $q_n(z)$ of Bazhanov and Mangazeev, also\nappearing in the eigenvectors of the XYZ spin chain.\n"}
{"abstract": "  #StopAsianHate and #StopAAPIHate are two of the most commonly used hashtags\nthat represent the current movement to end hate crimes against the Asian\nAmerican and Pacific Islander community. We conduct a social media study of\npublic opinion on the #StopAsianHate and #StopAAPIHate movement based on 46,058\nTwitter users across 30 states in the United States ranging from March 18 to\nApril 11, 2021. The movement attracts more participation from women, younger\nadults, Asian and Black communities. 51.56% of the Twitter users show direct\nsupport, 18.38% are news about anti-Asian hate crimes, while 5.43% show a\nnegative attitude towards the movement. Public opinion varies across user\ncharacteristics. Furthermore, among the states with most racial bias motivated\nhate crimes, the negative attitude towards the #StopAsianHate and #StopAAPIHate\nmovement is the weakest. To our best knowledge, this is the first large-scale\nsocial media-based study to understand public opinion on the #StopAsianHate and\n#StopAAPIHate movement. We hope our study can provide insights and promote\nresearch on anti-Asian hate crimes, and ultimately help address such a serious\nsocietal issue for the common benefits of all communities.\n"}
{"abstract": "  We gave a classification of P and Q with a finite number of K-orbits of a\ndouble flag variety G/P*K/Q for a symmetric pair (G, K) when G=GL_{m+n} and\nK=GL_{m}*GL_{n}, and a description of K-orbits when the number of K-orbits of\nG/P*K/Q is finite. We solved the problem by providing a correspondence between\nthe K-orbits and the quiver representations.\n"}
{"abstract": "  Abnormal event detection is a challenging task that requires effectively\nhandling intricate features of appearance and motion. In this paper, we present\nan approach of detecting anomalies in videos by learning a novel LSTM based\nself-contained network on normal dense optical flow. Due to their sigmoid\nimplementations, standard LSTM's forget gate is susceptible to overlooking and\ndismissing relevant content in long sequence tasks like abnormality detection.\nThe forget gate mitigates participation of previous hidden state for\ncomputation of cell state prioritizing current input. In addition, the\nhyperbolic tangent activation of standard LSTMs sacrifices performance when a\nnetwork gets deeper. To tackle these two limitations, we introduce a bi-gated,\nlight LSTM cell by discarding the forget gate and introducing sigmoid\nactivation. Specifically, the LSTM architecture we come up with fully sustains\ncontent from previous hidden state thereby enabling the trained model to be\nrobust and make context-independent decision during evaluation. Removing the\nforget gate results in a simplified and undemanding LSTM cell with improved\nperformance effectiveness and computational efficiency. Empirical evaluations\nshow that the proposed bi-gated LSTM based network outperforms various LSTM\nbased models verifying its effectiveness for abnormality detection and\ngeneralization tasks on CUHK Avenue and UCSD datasets.\n"}
{"abstract": "  This papers presents a novel quantised transform (the Sinclair-Town or ST\ntransform for short) that subsumes the rolls of both edge-detector, MSER style\nregion detector and corner detector. The transform is similar to the $unsharp$\ntransform but the difference from the local mean is quantised to 3 values\n(dark-neutral-light). The transform naturally leads to the definition of an\nappropriate local scale. A range of methods for extracting shape features form\nthe transformed image are presented. The generalized feature provides a robust\nbasis for establishing correspondence between images. The transform readily\nadmits more complicated kernel behaviour including multi-scale and asymmetric\nelements to prefer shorter scale or oriented local features.\n"}
{"abstract": "  Technologies play an important role in the hiring process for software\nprofessionals. Within this process, several studies revealed misconceptions and\nbad practices which lead to suboptimal recruitment experiences. In the same\ncontext, grey literature anecdotally coined the term R\\'esum\\'e-Driven\nDevelopment (RDD), a phenomenon describing the overemphasis of trending\ntechnologies in both job offerings and resumes as an interaction between\nemployers and applicants. While RDD has been sporadically mentioned in books\nand online discussions, there are so far no scientific studies on the topic,\ndespite its potential negative consequences. We therefore empirically\ninvestigated this phenomenon by surveying 591 software professionals in both\nhiring (130) and technical (558) roles and identified RDD facets in substantial\nparts of our sample: 60% of our hiring professionals agreed that trends\ninfluence their job offerings, while 82% of our software professionals believed\nthat using trending technologies in their daily work makes them more attractive\nfor prospective employers. Grounded in the survey results, we conceptualize a\ntheory to frame and explain R\\'esum\\'e-Driven Development. Finally, we discuss\ninfluencing factors and consequences and propose a definition of the term. Our\ncontribution provides a foundation for future research and raises awareness for\na potentially systemic trend that may broadly affect the software industry.\n"}
{"abstract": "  Deploying distributed renewable energy at the demand side is an important\nmeasure to implement a sustainable society. However, the massive small solar\nand wind generation units are beyond the control of a central operator. To\nencourage users to participate in energy management and reduce the dependence\non dispatchable resources, a peer-to-peer energy sharing scheme is proposed\nwhich releases the flexibility at the demand side. Every user makes decision\nindividually considering only local constraints; the microgrid operator\nannounces the sharing prices subjective to the coupling constraints without\nknowing users' local constraints. This can help protect privacy. We prove that\nthe proposed mechanism can achieve the same disutility and flexibility as\ncentralized dispatch, and develop an effective modified best response based\nalgorithm for reaching the market equilibrium. The concept of absorbable region\nis presented to measure the operating flexibility under the proposed energy\nsharing mechanism. A linear programming based polyhedral projection algorithm\nis developed to compute that region. Case studies validate the theoretical\nresults and show that the proposed method is scalable.\n"}
{"abstract": "  Interstellar dust grains are non-spherical and, in some environments,\npartially aligned along the direction of the interstellar magnetic field.\nNumerous alignment theories have been proposed, all of which examine the grain\nrotational dynamics. In 1999, Lazarian & Draine introduced the important\nconcept of thermal flipping, in which internal relaxation processes induce the\ngrain body to flip while its angular momentum remains fixed. Through detailed\nnumerical simulations, we study the role of thermal flipping on the grain\ndynamics during periods of relatively slow rotation, known as `crossovers', for\nthe special case of a spheroidal grain with a non-uniform mass distribution.\nLazarian & Draine proposed that rapid flipping during a crossover would lead to\n`thermal trapping', in which a systematic torque, fixed relative to the grain\nbody, would time average to zero, delaying spin-up to larger rotational speeds.\nWe find that the time-averaged systematic torque is not zero during the\ncrossover and that thermal trapping is not prevalent. As an application, we\nexamine whether the classic Davis-Greenstein alignment mechanism is viable, for\ngrains residing in the cold neutral medium and lacking superparamagnetic\ninclusions. We find that Davis-Greenstein alignment is not hindered by thermal\ntrapping, but argue that it is, nevertheless, too inefficient to yield the\nalignment of large grains responsible for optical and infrared starlight\npolarization. Davis-Greenstein alignment of small grains could potentially\ncontribute to the observed ultraviolet polarization. The theoretical and\ncomputational tools developed here can also be applied to analyses of alignment\nvia radiative torques and rotational disruption of grains.\n"}
{"abstract": "  Biopharmaceutical manufacturing is a rapidly growing industry with impact in\nvirtually all branches of medicine. Biomanufacturing processes require close\nmonitoring and control, in the presence of complex bioprocess dynamics with\nmany interdependent factors, as well as extremely limited data due to the high\ncost and long duration of experiments. We develop a novel model-based\nreinforcement learning framework that can achieve human-level control in\nlow-data environments. The model uses a probabilistic knowledge graph to\ncapture causal interdependencies between factors in the underlying stochastic\ndecision process, leveraging information from existing kinetic models from\ndifferent unit operations while incorporating real-world experimental data. We\nthen present a computationally efficient, provably convergent stochastic\ngradient method for policy optimization. Validation is conducted on a realistic\napplication with a multi-dimensional, continuous state variable.\n"}
{"abstract": "  A key challenge in scaling Gaussian Process (GP) regression to massive\ndatasets is that exact inference requires computation with a dense n x n kernel\nmatrix, where n is the number of data points. Significant work focuses on\napproximating the kernel matrix via interpolation using a smaller set of m\ninducing points. Structured kernel interpolation (SKI) is among the most\nscalable methods: by placing inducing points on a dense grid and using\nstructured matrix algebra, SKI achieves per-iteration time of O(n + m log m)\nfor approximate inference. This linear scaling in n enables inference for very\nlarge data sets; however the cost is per-iteration, which remains a limitation\nfor extremely large n. We show that the SKI per-iteration time can be reduced\nto O(m log m) after a single O(n) time precomputation step by reframing SKI as\nsolving a natural Bayesian linear regression problem with a fixed set of m\ncompact basis functions. With per-iteration complexity independent of the\ndataset size n for a fixed grid, our method scales to truly massive data sets.\nWe demonstrate speedups in practice for a wide range of m and n and apply the\nmethod to GP inference on a three-dimensional weather radar dataset with over\n100 million points.\n"}
{"abstract": "  Alkaline-earth-like transition-metal atoms such as Zn and Cd are promising\ncandidates for precision measurements and quantum many-body physics\nexperiments. Here, we theoretically investigate the properties of diatomic\nmolecules containing these closed-shell atoms. We calculate potential energy\ncurves, permanent electric dipole moments, and spectroscopic constants for\nmolecules consisting of either a Zn or Cd atom interacting with an alkali-metal\n(Li, Na, K, Rb, Cs, Fr) or alkaline-earth-metal (Be, Mg, Ca, Sr, Ba, Ra) atom.\nWe use the ab initio electronic structure coupled cluster method with single,\ndouble, and triple excitations combined with large Gaussian basis sets and\nsmall-core relativistic energy-consistent pseudopotentials for heavier atoms.\nWe predict that the studied molecules in the ground electronic state are\nchemically reactive weakly bound van der Waals complexes with small permanent\nelectric dipole moments. The present results may be useful for spectroscopy and\napplication of the studied molecules in modern ultracold physics and chemistry\nexperiments.\n"}
{"abstract": "  The long-tailed distribution datasets poses great challenges for deep\nlearning based classification models on how to handle the class imbalance\nproblem. Existing solutions usually involve class-balacing strategies or\ntransfer learing from head- to tail-classes or use two-stages learning strategy\nto re-train the classifier. However, the existing methods are difficult to\nsolve the low quality problem when images are obtained by SAR. To address this\nproblem, we establish a novel three-stages training strategy, which has\nexcellent results for processing SAR image datasets with long-tailed\ndistribution. Specifically, we divide training procedure into three stages. The\nfirst stage is to use all kinds of images for rough-training, so as to get the\nrough-training model with rich content. The second stage is to make the rough\nmodel learn the feature expression by using the residual dataset with the class\n0 removed. The third stage is to fine tune the model using class-balanced\ndatasets with all 10 classes (including the overall model fine tuning and\nclassifier re-optimization). Through this new training strategy, we only use\nthe information of SAR image dataset and the network model with very small\nparameters to achieve the top 1 accuracy of 22.34 in development phase.\n"}
{"abstract": "  Video-based human pose estimation (HPE) is a vital yet challenging task.\nWhile deep learning methods have made significant progress for the HPE, most\napproaches to this task detect each joint independently, damaging the pose\nstructural information. In this paper, unlike the prior methods, we propose a\nRelation-based Pose Semantics Transfer Network (RPSTN) to locate joints\nassociatively. Specifically, we design a lightweight joint relation extractor\n(JRE) to model the pose structural features and associatively generate heatmaps\nfor joints by modeling the relation between any two joints heuristically\ninstead of building each joint heatmap independently. Actually, the proposed\nJRE module models the spatial configuration of human poses through the\nrelationship between any two joints. Moreover, considering the temporal\nsemantic continuity of videos, the pose semantic information in the current\nframe is beneficial for guiding the location of joints in the next frame.\nTherefore, we use the idea of knowledge reuse to propagate the pose semantic\ninformation between consecutive frames. In this way, the proposed RPSTN\ncaptures temporal dynamics of poses. On the one hand, the JRE module can infer\ninvisible joints according to the relationship between the invisible joints and\nother visible joints in space. On the other hand, in the time, the propose\nmodel can transfer the pose semantic features from the non-occluded frame to\nthe occluded frame to locate occluded joints. Therefore, our method is robust\nto the occlusion and achieves state-of-the-art results on the two challenging\ndatasets, which demonstrates its effectiveness for video-based human pose\nestimation. We will release the code and models publicly.\n"}
{"abstract": "  A subset $A$ of the integers is a $B_k[g]$ set if the number of multisets\nfrom $A$ that sum to any fixed integer is at most $g$. Let $F_{k,g}(n)$ denote\nthe maximum size of a $B_k[g]$ set in $\\{1,\\dots, n\\}$. In this paper we\nimprove the best-known upper bounds on $F_{k,g}(n)$ for $g>1$ and $k$ large.\nWhen $g=1$ we match the best upper bound of Green with an improved error term.\nAdditionally, we give a lower bound on $F_{k,g}(n)$ that matches a construction\nof Lindstr\\\"om while removing one of the hypotheses.\n"}
{"abstract": "  A theory of local convexity for a second order differential equation (SODE)\non a Lie algebroid is developed. The particular case when the SODE is\nhomogeneous quadratic is extensively discussed.\n"}
{"abstract": "  Tidal perturbations play an important role in the study of the dynamics in\nthe classical two-body system. Understanding tidal effects in strong-field\nregions may allow one to use gravitational-wave or electromagnetic observations\nto locate or constraint the location of possible companions. Here, we\ninvestigate how timelike and null geodesics of a Schwarzschild black hole are\naffected in the presence of a companion. There is a panoply of new effects. In\nsome limiting cases, we find analytical solutions for closed null or timelike\ngeodesics. Our results show that light ring period as measured by a far-away\nobserver can be eiter shorter or longer, depending on the location of the\ncompanion. We also show that there are closed lightlike trajectories which are\nelliptic (for equatorial companions), and that timelike particles are affected\nin a similar manner. Finally, we attempt at estimating the ringdown from\ntidally perturbed geometries. Our results indicate that there are two stages in\nthe relaxation of such geometries, one associated with a prompt decay of waves\naround the deformed photonsphere, and a later relaxation of the global\ngeometry. These results are consistent with previous, full numerical studies.\n"}
{"abstract": "  Solar cycle 24 is one of the weakest solar cycles recorded, but surprisingly\nthe declining phase of it had a slow CME which evolved without any low coronal\nsignature and is classified as a stealth CME which was responsible for an\nintense geomagnetic storm at Earth (Dst = -176 nT). The impact of this space\nweather event on the terrestrial ionosphere has been reported. However,the\npropagation of this CME beyond 1 au and the impact of this CME on other\nplanetary environments have not been studied so far. In this paper, we analyse\nthe data from Sun-Earth L1 point as well as from the Martian orbit (near 1.5\nau) to understand the characteristics of the stealth CME as observed beyond 1\nau. The observations near Earth are using data from the Solar Dynamics\nObservatory (SDO) and the Advanced Composition Explorer (ACE) satellite located\nat L1 point whereas those near Mars are from the instruments for plasma and\nmagnetic field measurements on board Mars Atmosphere and Volatile EvolutioN\n(MAVEN) mission. The observations show that the stealth CME has reached 1.5 au\nafter 7 days of its initial observations at the Sun and caused depletion in the\nnightside topside ionosphere of Mars, as observed during the inbound phase\nmeasurements of the Langmuir Probe and Waves (LPW) instrument on board MAVEN.\nThese observations have implications on the ion escape rates from the Martian\nupper atmosphere.\n"}
{"abstract": "  This work presents a system identification procedure based on Convolutional\nNeural Networks (CNN) for human posture control using the DEC (Disturbance\nEstimation and Compensation) parametric model. The modular structure of the\nproposed control model inspired the design of a modular identification\nprocedure, in the sense that the same neural network is used to identify the\nparameters of the modules controlling different degrees of freedom. In this way\nthe presented examples of body sway induced by external stimuli provide several\ntraining samples at once\n"}
{"abstract": "  Aerosolized droplets play a central role in the transmission of various\ninfectious diseases, including Legionnaire's disease, gastroenteritis-causing\nnorovirus, and most recently COVID-19. Respiratory droplets are known to be the\nmost prominent source of transmission for COVID-19, however, alternative routes\nmay exist given the discovery of small numbers of viable viruses in urine and\nstool samples. Flushing biomatter can lead to the aerosolization of\nmicroorganisms, thus, there is a likelihood that bioaerosols generated in\npublic restrooms may pose a concern for the transmission of COVID-19,\nespecially since these areas are relatively confined, experience heavy foot\ntraffic, and may suffer from inadequate ventilation. To quantify the extent of\naerosolization, we measure the size and number of droplets generated by\nflushing toilets and urinals in a public restroom. The results indicate that\nthe particular designs tested in the study generate a large number of droplets\nin the size range 0.3$\\mu m$ to 3$\\mu m$, which can reach heights of at least\n1.52m. Covering the toilet reduced aerosol levels but did not eliminate them\ncompletely, suggesting that aerosolized droplets escaped through small gaps\nbetween the cover and the seat. In addition to consistent increases in aerosol\nlevels immediately after flushing, there was a notable rise in ambient aerosol\nlevels due to the accumulation of droplets from multiple flushes conducted\nduring the tests. This highlights the need for incorporating adequate\nventilation in the design and operation of public spaces, which can help\nprevent aerosol accumulation in high occupancy areas and mitigate the risk of\nairborne disease transmission.\n"}
{"abstract": "  Einstein claimed that the fundamental dynamical insight of special relativity\nwas the equivalence of mass and energy. I disagree. Not only are mass and\nenergy not equivalent (whatever exactly that means) but talk of such\nequivalence obscures the real dynamical insight of special relativity, which\nconcerns the nature of 4-forces and interactions more generally. In this paper\nI present and defend a new ontology of special relativistic particle dynamics\nthat makes this insight perspicuous and I explain how alleged cases of\nmass-energy conversion can be accommodated within that ontology.\n"}
{"abstract": "  Triangles with integer length sides and integer area are known as Heron\ntriangles. Taking rescaling freedom into account, one can apply the same name\nwhen all sides and the area are rational numbers. A perfect triangle is a Heron\ntriangle with all three medians being rational, and it is a longstanding\nconjecture that no such triangle exists. However, Buchholz and Rathbun showed\nthat there are infinitely many Heron triangles with two rational medians, an\ninfinite subset of which are associated with rational points on an elliptic\ncurve $E(\\mathbb{Q})$ with Mordell-Weil group $\\mathbb{Z}\\times\n\\mathbb{Z}/2\\mathbb{Z}$, and they observed a connection with a pair of Somos-5\nsequences. Here we make the latter connection more precise by providing\nexplicit formulae for the integer side lengths, the two rational medians, and\nthe area in this infinite family of Heron triangles. The proof uses a combined\napproach to Somos-5 sequences and associated Quispel-Roberts-Thompson (QRT)\nmaps in the plane, from several different viewpoints: complex analysis, real\ndynamics, and reduction modulo a prime.\n"}
{"abstract": "  This paper studies the convergence of the tamed Euler-Maruyama (EM) scheme\nfor a class of McKean-Vlasov neutral stochastic differential delay equations\n(MV-NSDDEs) that the drift coefficients satisfy the super-linear growth\ncondition. We provide the existence and uniqueness of strong solutions to\nMV-NSDDEs. Then, we use a stochastic particle method, which is based upon the\ntheory of the propagation of chaos between particle system and the original\nMV-NSDDE, to deal with the approximation of the law. Moreover, we obtain the\nconvergence rate of tamed EM scheme with respect to the corresponding particle\nsystem. Combining the result of propagation of chaos and the convergence rate\nof the numerical solution to the particle system, we get a convergence error\nbetween the numerical solution and exact solution of the original MV-NSDDE in\nthe stepsize and number of particles.\n"}
{"abstract": "  We investigate the prospects of next-generation neutrino oscillation\nexperiments DUNE, T2HK and JUNO including TAO within Standard Model Effective\nField Theory (SMEFT). We also re-interpret COHERENT data in this framework.\nConsidering both charged and neutral current neutrino Non-Standard Interactions\n(NSIs), we analyse dimension-6 SMEFT operators and derive lower bounds to UV\nscale $\\Lambda$. The most powerful probe is obtained on ${\\cal\nO}_{{ledq}_{1211}}$ with $\\Lambda \\gtrsim$ 450 TeV due to the electron neutrino\nsample in T2HK near detector. We find DUNE and JUNO to be complementary to T2HK\nin exploring different subsets of SMEFT operators at about 25 TeV. We conclude\nthat near detectors play a significant role in each experiment. We also find\nCOHERENT with CsI and LAr targets to be sensitive to new physics up to\n$\\sim$900 GeV.\n"}
{"abstract": "  Mutation testing is used to evaluate the effectiveness of test suites. In\nrecent years, a promising variation called extreme mutation testing emerged\nthat is computationally less expensive. It identifies methods where their\nfunctionality can be entirely removed, and the test suite would not notice it,\ndespite having coverage. These methods are called pseudo-tested. In this paper,\nwe compare the execution and analysis times for traditional and extreme\nmutation testing and discuss what they mean in practice. We look at how extreme\nmutation testing impacts current software development practices and discuss\nopen challenges that need to be addressed to foster industry adoption. For\nthat, we conducted an industrial case study consisting of running traditional\nand extreme mutation testing in a large software project from the semiconductor\nindustry that is covered by a test suite of more than 11,000 unit tests. In\naddition to that, we did a qualitative analysis of 25 pseudo-tested methods and\ninterviewed two experienced developers to see how they write unit tests and\ngathered opinions on how useful the findings of extreme mutation testing are.\nOur results include execution times, scores, numbers of executed tests and\nmutators, reasons why methods are pseudo-tested, and an interview summary. We\nconclude that the shorter execution and analysis times are well noticeable in\npractice and show that extreme mutation testing supplements writing unit tests\nin conjunction with code coverage tools. We propose that pseudo-tested code\nshould be highlighted in code coverage reports and that extreme mutation\ntesting should be performed when writing unit tests rather than in a decoupled\nsession. Future research should investigate how to perform extreme mutation\ntesting while writing unit tests such that the results are available fast\nenough but still meaningful.\n"}
{"abstract": "  In this paper, we interpret the theta rank of an irreducible character of a\nfinite classical group in terms of the data from the Lusztig classification.\nThen we prove the following two results: (1) the agreement of the $U$-rank and\nthe theta rank for any irreducible character of low Theta rank; (2) the\nagreement of the parabolic asymptotic rank and the theta rank for any\nirreducible character.\n"}
{"abstract": "  The IDIA Visualisation Laboratory based at the University of Cape Town is\nexploring the use of virtual reality technology to visualise and analyse\nastronomical data. The iDaVIE software suite currently under development reads\nfrom both volumetric data cubes and sparse multi-dimensional catalogs,\nrendering them in a room-scale immersive environment that allows the user to\nintuitively view, navigate around and interact with features in three\ndimensions. This paper will highlight how the software imports from common\nastronomy data formats and processes the information for loading into the Unity\ngame engine. It will also describe what tools are currently available to the\nuser and the various performance optimisations made for seamless use.\nApplications by astronomers will be reviewed in addition to the features we\nplan to include in future releases.\n"}
{"abstract": "  We study robustly optimal mechanisms for selling multiple items. The seller\nmaximizes revenue against a worst-case distribution of a buyer's valuations\nwithin a set of distributions, called an \"ambiguity\" set. We identify the exact\nforms of robustly optimal selling mechanisms and the worst-case distributions\nwhen the ambiguity set satisfies a variety of moment conditions on the values\nof subsets of goods. We also identify general properties of the ambiguity set\nthat lead to the robust optimality of partial bundling which includes separate\nsales and pure bundling as special cases.\n"}
{"abstract": "  Multi-scale structures are prevalent in both natural and artificial systems,\nas they can handle increasing complexity. Several terms are employed almost\ninterchangeably across various application domains to refer to the multi-scale\nconcept - e.g., hierarchy, holarchy, multi-level, multi-layer, nested,\nembedded, micro-macro or coarse graining. While the concrete meanings behind\nthese terms may differ slightly, several core commonalities persist across all\ncases. In this position paper we aim to highlight these common features of the\nmulti-scale concept, as a preliminary basis for a generic theory of multi-scale\nsystems. We discuss the concepts of scale and multi-scale systems in general,\nand then of multi-scale feedback systems in particular, focusing on the role\nplayed by information in such systems. Our long-term objective is to develop a\ngeneral theory of multi-scale feedback systems, applicable across all domains\ndealing with complex systems.\n"}
{"abstract": "  The tightest and most robust cosmological results of the next decade will be\nachieved by bringing together multiple surveys of the Universe. This endeavor\nhas to happen across multiple layers of the data processing and analysis, e.g.,\nenhancements are expected from combining Euclid, Rubin, and Roman (as well as\nother surveys) not only at the level of joint processing and catalog\ncombination, but also during the post-catalog parts of the analysis such as the\ncosmological inference process. While every experiment builds their own\nanalysis and inference framework and creates their own set of simulations,\ncross-survey work that homogenizes these efforts, exchanges information from\nnumerical simulations, and coordinates details in the modeling of astrophysical\nand observational systematics of the corresponding datasets is crucial.\n"}
{"abstract": "  A rigorous approach for solving canonical circular open-ended\ndielectric-lined waveguide diffraction problems is presented. This is\ncontinuation of our recent paper [1] where a simpler case of uniform dielectric\nfilling has been considered. Here we deal with the case of an open-ended\ncircular waveguide with layered dielectric filling which is closer to potential\napplications. The presented method uses the solution of corresponding\nWiener-Hopf-Fock equation and leads to an infinite linear system for reflection\ncoefficients (S-parameters) of the waveguide, the latter can be efficiently\nsolved numerically using the reducing technique. As a specific example directly\napplicable to beam-driven radiation sources based on dielectric-lined\ncapillaries, diffraction of a slow TM symmetrical mode at the open end of the\ndescribed waveguide is considered. A series of such modes forms the wakefield\n(Cherenkov radiation field) generated by a charged particle bunch during its\npassage along the vacuum channel axis. Calculated S-parameters were compared\nwith those obtained from COMSOL simulation and an excellent agreement was\nshown. This method is expected to be very convenient for analytical\ninvestigation of various electromagnetic interactions of Terahertz (THz) waves\n(both free and guided) and charged particle bunches with slow-wave structures\nprospective in context of modern beam-driven THz emitters, THz accererators and\nTHz-based bunch manipulation and bunch diagnostic systems.\n"}
{"abstract": "  The concept of smart cities is driven by the need to enhance citizens'\nquality of life. It is estimated that 70% of the world population will live in\nurban areas by 2050. The electric grid is the energy backbone of smart city\ndeployments. An electric energy system immune to adverse events, both cyber and\nphysical risks, and able to support the integration of renewable sources will\ndrive a transformational development approach for future smart cities. This\narticle describes how the future electric energy system with 100% electricity\nsupply from renewable energy sources requires the \"birth of security and\nresiliency\" incorporated with its ecosystem.\n"}
{"abstract": "  Multi-robot teams can achieve more dexterous, complex and heavier payload\ntasks than a single robot, yet effective collaboration is required. Multi-robot\ncollaboration is extremely challenging due to the different kinematic and\ndynamics capabilities of the robots, the limited communication between them,\nand the uncertainty of the system parameters. In this paper, a Decentralized\nAbility-Aware Adaptive Control is proposed to address these challenges based on\ntwo key features. Firstly, the common manipulation task is represented by the\nproposed nominal task ellipsoid, which is used to maximize each robot force\ncapability online via optimizing its configuration. Secondly, a decentralized\nadaptive controller is designed to be Lyapunov stable in spite of heterogeneous\nactuation constraints of the robots and uncertain physical parameters of the\nobject and environment. In the proposed framework, decentralized coordination\nand load distribution between the robots is achieved without communication,\nwhile only the control deficiency is broadcast if any of the robots reaches its\nforce limits. In this case, the object reference trajectory is modified in a\ndecentralized manner to guarantee stable interaction. Finally, we perform\nseveral numerical and physical simulations to analyse and verify the proposed\nmethod with heterogeneous multi-robot teams in collaborative manipulation\ntasks.\n"}
{"abstract": "  We discuss the generation of field-induced entanglement between two objects\neach in a superposition of two trajectories.The objects have currents coupled\nto local quantum fields, and the currents are evaluated around each trajectory\nof the objects.The fields have only dynamical degrees of freedom and satisfy\nthe microcausality condition. We find that the superposed state of trajectories\ncannot be entangled when the objects are spacelike separated.This means that\nthe quantum fields do not generate spacelike entanglement in the superposition\nof two trajectories of each object.\n"}
{"abstract": "  We investigate the asymptotic variance of Gaussian nodal excursions in the\nEuclidean space, focusing on the case where the spectral measure has\nincommensurable atoms. This study requires to establish fine recurrence\nproperties in 0 for the associated irrational random walk on the torus. We show\nin particular that the recurrence magnitude depends strongly on the diophantine\nproperties of the atoms, and the same goes for the variance asymptotics of\nnodal excursions. More specifically, if the spectral measures contains atoms\nwhich ratios are well approximable by rationals, the variance is likely to have\nlarge fluctuations as the observation window grows, whereas the variance is\nbounded by the (d -- 1)-dimensional measure of the window boundary if these\nratio are badly approximable. We also show that, given any reasonable function,\nthere are uncountably many sets of parameters for which the variance is\nasymptotically equivalent to this function.\n"}
{"abstract": "  In this work we present a generalized Kirchhoff-Love shell theory that can\ncapture anisotropy not only in stretching and out-of-plane bending, but also in\nin-plane bending. This setup is particularly suitable for heterogeneous and\nfibrous materials such as textiles, biomaterials, composites and pantographic\nstructures. The presented theory is a direct extension of existing\nKirchhoff-Love shell theory to incorporate the in-plane bending resistance of\nfibers. It also extends existing high gradient Kirchhoff-Love shell theory for\ninitially straight fibers to initially curved fibers. To describe the\nadditional kinematics of multiple fiber families, a so-called in-plane\ncurvature tensor -- which is symmetric and of second order -- is proposed. The\neffective stress tensor and the in-plane and out-of-plane moment tensors are\nthen identified from the mechanical power balance. These tensors are all second\norder and symmetric for general materials. The constitutive equations for\nhyperelastic materials are derived from different expressions of the mechanical\npower balance. The weak form is also presented as it is required for\ncomputational shell formulations based on rotation-free finite element\ndiscretizations.\n"}
{"abstract": "  A detector emulator method has been developed. It allows for testing any\nreadout electronics with realistic detector-like signals under controlled\nconditions, beyond the limits of any experiment based on radiation sources. It\ncan substitute for expensive test-beam campaigns, or predict their results. The\ndetector emulator is a powerful tool for the engineering of detector concepts,\nwith its capability to reproduce any detector signal feature to test the\nelectronics response. Here, the method is applied to the case of the VMM3a ASIC\ntested with sTGC detector emulated signals. Measurements of charge spectra and\nmuon detection efficiency under intense gamma background are reproduced. The\neffect of different attenuation circuits is studied.\n"}
{"abstract": "  This work addresses the problem of optimizing communications between server\nand clients in federated learning (FL). Current sampling approaches in FL are\neither biased, or non optimal in terms of server-clients communications and\ntraining stability. To overcome this issue, we introduce \\textit{clustered\nsampling} for clients selection. We prove that clustered sampling leads to\nbetter clients representatitivity and to reduced variance of the clients\nstochastic aggregation weights in FL. Compatibly with our theory, we provide\ntwo different clustering approaches enabling clients aggregation based on 1)\nsample size, and 2) models similarity. Through a series of experiments in\nnon-iid and unbalanced scenarios, we demonstrate that model aggregation through\nclustered sampling consistently leads to better training convergence and\nvariability when compared to standard sampling approaches. Our approach does\nnot require any additional operation on the clients side, and can be seamlessly\nintegrated in standard FL implementations. Finally, clustered sampling is\ncompatible with existing methods and technologies for privacy enhancement, and\nfor communication reduction through model compression.\n"}
{"abstract": "  Here we report a developed high performance and simplified version of the\ncode denominated RIO, which can be easily extended, for the generalized BSSN\nformulation. We implement a code which is regular at the center of symmetry,\nwithout use a special procedure for regularization, as usual. We get\nexponential convergence for constraints. The numerical algorithm is based on\nthe Galerkin-Collocation method developed successfully for diverse physical\nscenarios by the Numerical Relativity Group at UERJ. For the sake of clarity in\npresentation, we consider here the most simple case to display the most salient\nfeatures of the procedure. Thus, we focus on the definite tests of the new\nnumerical framework. The timing and performance of the code show that we can\nreach a better accuracy close to the machine precision, for the Hamiltonian and\nmomentum constraints. RIO will be an open source code; currently is under\ncontinuous development to consider more general and realistic problems.\n"}
{"abstract": "  We unveil the nature of the structural disorder in bottom-up zigzag graphene\nnanoribbons along with its effect on the magnetism and electronic transport on\nthe basis of scanning probe microscopies and first-principles calculations. We\nfind that edge-missing m-xylene units emerging during the cyclodehydrogenation\nstep of the on-surface synthesis are the most common point defects. These\n\"bite'' defects act as spin-1 paramagnetic centers, severely disrupt the\nconductance spectrum around the band extrema, and give rise to spin-polarized\ncharge transport. We further show that the electronic conductance across\ngraphene nanoribbons is more sensitive to \"bite\" defects forming at the zigzag\nedges than at the armchair ones. Our work establishes a comprehensive\nunderstanding of the low-energy electronic properties of disordered bottom-up\ngraphene nanoribbons.\n"}
{"abstract": "  A power system electromechanical wave propagates from the disturbance\nlocation to the rest of system, influencing various types of protections. In\naddition, since more power-electronics-interfaced generation and energy storage\ndevices are being integrated into power systems, electromechanical wave\npropagation speeds in the future power systems are likely to change\naccordingly. In this paper, GPS-synchronized measurement data from a wide-area\nsynchrophasor measurement system FNET/GridEye are used to analyze the\ncharacteristics of electromechanical wave propagation in the U.S. Eastern\nInterconnection (EI) system. Afterwards, high levels of photovoltaic (PV)\npenetration are modeled in the EI to investigate the influences of a typical\npower-electronics--interfaced resource on the electromechanical wave\npropagation speed. The result shows a direct correlation between the local\npenetration level of inverter-based generation and the electromechanical wave\npropagation speed.\n"}
{"abstract": "  Modern deep unsupervised learning methods have shown great promise for\ndetecting diseases across a variety of medical imaging modalities. While\nprevious generative modeling approaches successfully perform anomaly detection\nby learning the distribution of healthy 2D image slices, they process such\nslices independently and ignore the fact that they are correlated, all being\nsampled from a 3D volume. We show that incorporating the 3D context and\nprocessing whole-body MRI volumes is beneficial to distinguishing anomalies\nfrom their benign counterparts. In our work, we introduce a multi-channel\nsliding window generative model to perform lesion detection in whole-body MRI\n(wbMRI). Our experiments demonstrate that our proposed method significantly\noutperforms processing individual images in isolation and our ablations clearly\nshow the importance of 3D reasoning. Moreover, our work also shows that it is\nbeneficial to include additional patient-specific features to further improve\nanomaly detection in pediatric scans.\n"}
{"abstract": "  Channel state information (CSI) rapidly becomes outdated in high mobility\nscenarios, degrading the performance of wireless communication systems. In\nthese cases, time series prediction techniques can be applied to combat the\neffects of outdated CSI. Recently, it has been shown that recurrent neural\nnetworks (RNNs) exhibit outstanding performance in time series prediction\ntasks. In this paper, we investigate the performance of RNN and long short term\nmemory (LSTM) predictors in a simple Rayleigh flat-fading channel. We conduct\nnumerical experiments to evaluate whether these machine-learning (ML)-based\npredictors can outperform the optimal linear minimum mean square error Wiener\npredictor. Our simulation results indicate that the considered neural network\npredictors outperform the Wiener predictor for small observation window lengths\nand are more robust under weak channel correlation as well as in the presence\nof noise. Furthermore, we show that simple shallow RNNs are sufficient to model\nRayleigh channels over a wide range of Doppler shifts.\n"}
{"abstract": "  Multimedia technologies are now more practical and deployable in real life,\nand the algorithms are widely used in various researching areas such as deep\nlearning, signal processing, haptics, computer vision, robotics, and medical\nmultimedia processing. This survey provides an overview of multimedia\ntechnologies and robust algorithms in multimedia data processing, medical\nmultimedia processing, human facial expression tracking and pose recognition,\nand multimedia in education and training. This survey will also analyze and\npropose a future research direction based on the overview of current robust\nalgorithms and multimedia technologies. We want to thank the research and\nprevious work done by the Multimedia Research Centre (MRC), the University of\nAlberta, which is the inspiration and starting point for future research.\n"}
{"abstract": "  The sub-seasonal and synoptic-scale variability of the Indian summer monsoon\nrainfall are controlled primarily by monsoon intra-seasonal oscillations (MISO)\nand low pressure systems (LPS), respectively. The positive and negative phases\nof MISO lead to alternate epochs of above-normal (active) and below-normal\n(break) spells of rainfall. LPSs are embedded within the different phases of\nMISO and are known to produce heavy precipitation events over central India.\nWhether the interaction with the MISO phases modulates the precipitation\nresponse of LPSs, and thereby the characteristics of extreme rainfall events\n(EREs) remains unaddressed in the available literature. In this study, we\nanalyze the LPSs that produce EREs of various spatial extents viz., Small,\nMedium, and Large over central India from 1979 to 2012. We also compare them\nwith the LPSs that pass through central India and do not give any ERE\n(LPS-noex). We find that thermodynamic characteristics of LPSs that trigger\ndifferent spatial extents of EREs are similar. However, they show differences\nin their dynamic characteristics. The ERE producing LPSs are slower, moister\nand more intense than LPS-noex. The LPSs that lead to Medium and Large EREs\ntend to occur during the positive phase of MISO when an active monsoon trough\nis present over central India. On the other hand, LPS-noex and the LPSs that\ntrigger Small EREs occur mainly during the neutral or negative phases of the\nMISO. The large-scale dynamic forcing, intensification of LPSs, and diabatic\ngeneration of low-level potential vorticity due to the presence of active\nmonsoon trough help in the organization of convection and lead to Medium and\nLarge EREs. On the other hand, the LPSs that form during the negative or\nneutral phases of MISO do not intensify much during their lifetime and trigger\nscattered convection, leading to EREs of small size.\n"}
{"abstract": "  At peak, long-duration gamma-ray bursts are the most luminous sources of\nelectromagnetic radiation known. Since their progenitors are massive stars,\nthey provide a tracer of star formation and star-forming galaxies over the\nwhole of cosmic history. Their bright power-law afterglows provide ideal\nbacklights for absorption studies of the interstellar and intergalactic medium\nback to the reionization era. The proposed THESEUS mission is designed to\ndetect large samples of GRBs at $z>6$ in the 2030s, at a time when supporting\nobservations with major next generation facilities will be possible, thus\nenabling a range of transformative science. THESEUS will allow us to explore\nthe faint end of the luminosity function of galaxies and the star formation\nrate density to high redshifts; constrain the progress of re-ionisation beyond\n$z\\gtrsim6$; study in detail early chemical enrichment from stellar explosions,\nincluding signatures of Population III stars; and potentially characterize the\ndark energy equation of state at the highest redshifts.\n"}
{"abstract": "  This white paper presents a summary of the discussions regarding critical\nconsiderations to develop an extensive repository of online videos annotated\nwith labels indicating questionable content. The main discussion points\ninclude: 1) the type of appropriate labels that will result in a valuable\nrepository for the larger AI community; 2) how to design the collection and\nannotation process, as well as the distribution of the corpus to maximize its\npotential impact; and, 3) what actions we can take to reduce risk of trauma to\nannotators.\n"}
{"abstract": "  We consider the Cauchy problem for one-dimensional dispersive equations with\na general nonlinearity in the periodic setting. Our main hypotheses are both\nthat the dispersive operator behaves for high frequencies as a Fourier\nmultiplier by $ i |\\xi|^\\alpha \\xi $, with $ 1\\le \\alpha \\le 2 $, and that the\nnonlinear term is of the form $ \\partial_x f(u) $ where $ f $ is the sum of an\nentire series with infinite radius of convergence. Under these conditions, we\nprove the unconditional local well-posedness of the Cauchy problem in\n$H^{s}(\\mathbb{T})$ for $ s\\ge 1-\\frac{\\alpha}{2(\\alpha+1)}$. This leads to\nsome global existence results above the energy space $ H^{\\alpha/2}(\\mathbb{T})\n$, for $ \\alpha \\in [\\sqrt{2},2]$.\n"}
{"abstract": "  The Hubble tension is shown to be solvable, without any free parameter,\nconceptually and quantitatively, within the approach of modified weak-field\nGeneral Relativity involving the cosmological constant $\\Lambda$. That approach\nenables one to describe in a unified picture both the dynamics of dark matter\ncontaining galaxies and the accelerated expansion of the Universe, thus\ndefining a {\\it local} Hubble constant of a local flow and the {\\it global}\none. The data on the dark matter content of peculiar galaxy samples are shown\nto be compatible to that unified picture. Future more refined surveys of galaxy\ndistribution, hierarchical dynamics and flows within the vicinity of the Local\ngroup and the Virgo supercluster can be decisive in revealing the possible\ncommon nature of the dark sector.\n"}
{"abstract": "  In this paper, we investigate the optimal output tracking problem for linear\ndiscrete-time systems with unknown dynamics using reinforcement learning and\nrobust output regulation theory. This output tracking problem only allows to\nutilize the outputs of the reference system and the controlled system, rather\nthan their states, and differs from most existing tracking results that depend\non the state of the system. The optimal tracking problem is formulated into a\nlinear quadratic regulation problem by proposing a family of dynamic\ndiscrete-time controllers. Then, it is shown that solving the output tracking\nproblem is equivalent to solving output regulation equations, whose solution,\nhowever, requires the knowledge of the complete and accurate system dynamics.\nTo remove such a requirement, an off-policy reinforcement learning algorithm is\nproposed using only the measured output data along the trajectories of the\nsystem and the reference output. By introducing re-expression error and\nanalyzing the rank condition of the parameterization matrix, we ensure the\nuniqueness of the proposed RL based optimal control via output feedback.\n"}
{"abstract": "  As an essential component of human cognition, cause-effect relations appear\nfrequently in text, and curating cause-effect relations from text helps in\nbuilding causal networks for predictive tasks. Existing causality extraction\ntechniques include knowledge-based, statistical machine learning(ML)-based, and\ndeep learning-based approaches. Each method has its advantages and weaknesses.\nFor example, knowledge-based methods are understandable but require extensive\nmanual domain knowledge and have poor cross-domain applicability. Statistical\nmachine learning methods are more automated because of natural language\nprocessing (NLP) toolkits. However, feature engineering is labor-intensive, and\ntoolkits may lead to error propagation. In the past few years, deep learning\ntechniques attract substantial attention from NLP researchers because of its'\npowerful representation learning ability and the rapid increase in\ncomputational resources. Their limitations include high computational costs and\na lack of adequate annotated training data. In this paper, we conduct a\ncomprehensive survey of causality extraction. We initially introduce primary\nforms existing in the causality extraction: explicit intra-sentential\ncausality, implicit causality, and inter-sentential causality. Next, we list\nbenchmark datasets and modeling assessment methods for causal relation\nextraction. Then, we present a structured overview of the three techniques with\ntheir representative systems. Lastly, we highlight existing open challenges\nwith their potential directions.\n"}
{"abstract": "  We present a consistent \\emph{ab initio} computation of the longitudinal\nresponse function $R_L$ in $^{40}$Ca using the coupled-cluster and Lorentz\nintegral transform methods starting from chiral nucleon-nucleon and\nthree-nucleon interactions. We validate our approach by comparing our results\nfor $R_L$ in $^4$He and the Coulomb sum rule in $^{40}$Ca against experimental\ndata and other calculations. For $R_L$ in $^{40}$Ca we obtain a very good\nagreement with experiment in the quasi-elastic peak up to intermediate momentum\ntransfers, and we find that final state interactions are essential for an\naccurate description of the data. This work presents a milestone towards\n\\emph{ab initio} computations of neutrino-nucleus cross sections relevant for\nexperimental long-baseline neutrino programs.\n"}
{"abstract": "  Scholarly journals are increasingly using social media to share their latest\nresearch publications and communicate with their readers. Having a presence on\nsocial media gives journals a platform to raise their profile and promote their\ncontent. This study compares the number of clicks received when journals\nprovide two types of links to subscription articles: open access (OA) and paid\ncontent links. We examine the OA effect using unique matched-pair data for the\njournal Nature Materials. Our study finds that OA links perform better than\npaid content links. In particular, when the journal does not indicate that a\nlink to an article is an OA link, there is an obvious drop in performance\nagainst clicks on links indicating OA status. OA has a positive effect on the\nnumber of clicks in all countries, but its positive impact is slightly greater\nin developed countries. The results suggest that free content is more\nattractive to users than paid content. Social media exposure of scholarly\narticles promotes the use of research outputs. Combining social media\ndissemination with OA appears to enhance the reach of scientific information.\nHowever, extensive further efforts are needed to remove barriers to OA.\n"}
{"abstract": "  It has long been thought that high-dimensional data encountered in many\npractical machine learning tasks have low-dimensional structure, i.e., the\nmanifold hypothesis holds. A natural question, thus, is to estimate the\nintrinsic dimension of a given population distribution from a finite sample. We\nintroduce a new estimator of the intrinsic dimension and provide finite sample,\nnon-asymptotic guarantees. We then apply our techniques to get new sample\ncomplexity bounds for Generative Adversarial Networks (GANs) depending only on\nthe intrinsic dimension of the data.\n"}
{"abstract": "  We extend the planar Markus-Yamabe Jacobian Conjecture to differential\nsystems having jacobian matrix with eigenvalues with negative or zero real\nparts.\n"}
{"abstract": "  This editorial presents the quantum field theoretic ambiance in which Elko\nand mass dimension one fermions come to exist. This may serve not only to\nintroduce Elko, and the associated quantum field, but it may also open a door\nto a new class of spin one half fermions and bosons. These new particles, as it\nwill be argued in this volume, are first-principle candidates for dark matter.\n"}
{"abstract": "  We combine numerical and analytical methods to study two dimensional active\ncrystals formed by permanently linked swimmers and with two distinct alignment\ninteractions. The system admits a stationary phase with quasi long range\ntranslational order, as well as a moving phase with quasi-long range velocity\norder. The translational order in the moving phase is significantly influenced\nby alignment interaction. For Vicsek-like alignment, the translational order is\nshort-ranged, whereas the bond-orientational order is quasi-long ranged,\nimplying a moving hexatic phase. For elasticity-based alignment, the\ntranslational order is quasi-long ranged parallel to the motion and\nshort-ranged in perpendicular direction, whereas the bond orientational order\nis long-ranged. We also generalize these results to higher dimensions.\n"}
{"abstract": "  Quantum memory effects can be related to a bidirectional exchange of\ninformation between an open system and its environment, which in turn modifies\nthe state and dynamical behavior of the last one. Nevertheless,\nnon-Markovianity can also be induced by environments whose dynamics is not\naffected during the system evolution, implying the absence of any physical\ninformation exchange. An unsolved open problem in the formulation of quantum\nmemory measures is the apparent impossibility of discerning between both\nparadigmatic cases. Here, we present an operational scheme that, based on the\noutcomes of successive measurements processes performed over the system of\ninterest, allows to distinguishing between both kinds of memory effects. The\nmethod accurately detects bidirectional information flows in diverse\ndissipative and dephasing non-Markovian open system dynamics.\n"}
{"abstract": "  Deep convolutional neural networks have shown outstanding performance in\nmedical image segmentation tasks. The usual problem when training supervised\ndeep learning methods is the lack of labeled data which is time-consuming and\ncostly to obtain. In this paper, we propose a novel uncertainty-guided\nsemi-supervised learning based on a student-teacher approach for training the\nsegmentation network using limited labeled samples and a large number of\nunlabeled images. First, a teacher segmentation model is trained from the\nlabeled samples using Bayesian deep learning. The trained model is used to\ngenerate soft segmentation labels and uncertainty maps for the unlabeled set.\nThe student model is then updated using the softly segmented samples and the\ncorresponding pixel-wise confidence of the segmentation quality estimated from\nthe uncertainty of the teacher model using a newly designed loss function.\nExperimental results on a retinal layer segmentation task show that the\nproposed method improves the segmentation performance in comparison to the\nfully supervised approach and is on par with the expert annotator. The proposed\nsemi-supervised segmentation framework is a key contribution and applicable for\nbiomedical image segmentation across various imaging modalities where access to\nannotated medical images is challenging\n"}
{"abstract": "  The hat guessing number is a graph invariant based on a hat guessing game\nintroduced by Winkler. Using a new vertex decomposition argument involving an\nedge density theorem of Erd\\H{o}s for hypergraphs, we show that the hat\nguessing number of all outerplanar graphs is less than $2^{125000}$. We also\ndefine the class of layered planar graphs, which contains outerplanar graphs,\nand we show that every layered planar graph has bounded hat guessing number.\n"}
{"abstract": "  Negation Scope Resolution is an extensively researched problem, which is used\nto locate the words affected by a negation cue in a sentence. Recent works have\nshown that simply finetuning transformer-based architectures yield\nstate-of-the-art results on this task. In this work, we look at Negation Scope\nResolution as a Cloze-Style task, with the sentence as the Context and the cue\nwords as the Query. We also introduce a novel Cloze-Style Attention mechanism\ncalled Orthogonal Attention, which is inspired by Self Attention. First, we\npropose a framework for developing Orthogonal Attention variants, and then\npropose 4 Orthogonal Attention variants: OA-C, OA-CA, OA-EM, and OA-EMB. Using\nthese Orthogonal Attention layers on top of an XLNet backbone, we outperform\nthe finetuned XLNet state-of-the-art for Negation Scope Resolution, achieving\nthe best results to date on all 4 datasets we experiment with: BioScope\nAbstracts, BioScope Full Papers, SFU Review Corpus and the *sem 2012 Dataset\n(Sherlock).\n"}
{"abstract": "  In the rare physics events search, liquid and solid detector plays an\nimportant role. Its analysis is based on the detection of excess events over\nthe expected background or on the detection of an annual event rate modulation.\nGermanium detectors sensitivities have been demonstrated as efficient means to\nprobe Weakly Interacting Massive Particles (WIMPs). Germanium detectors having\nsensitivity less than 100 eV opens new platform for the investigation of\nneutrinos and dark matter physics. There are various working configuration\navailable for germanium detectors such as semi-coaxial, true coaxial and point\ncontact. The point contact germanium detectors have been widely used for dark\nmatter and low energy neutrino searches because of their sub KeV energy\nthreshold and low electronic noise. Pulse shape of P-type point contact (P-PC)\ngermanium detector has been simulated in the present work and compared to the\npulse shape of Majorana Demonstrator(MJD).\n"}
{"abstract": "  Multisource image analysis that leverages complementary spectral, spatial,\nand structural information benefits fine-grained object recognition that aims\nto classify an object into one of many similar subcategories. However, for\nmultisource tasks that involve relatively small objects, even the smallest\nregistration errors can introduce high uncertainty in the classification\nprocess. We approach this problem from a weakly supervised learning perspective\nin which the input images correspond to larger neighborhoods around the\nexpected object locations where an object with a given class label is present\nin the neighborhood without any knowledge of its exact location. The proposed\nmethod uses a single-source deep instance attention model with parallel\nbranches for joint localization and classification of objects, and extends this\nmodel into a multisource setting where a reference source that is assumed to\nhave no location uncertainty is used to aid the fusion of multiple sources in\nfour different levels: probability level, logit level, feature level, and pixel\nlevel. We show that all levels of fusion provide higher accuracies compared to\nthe state-of-the-art, with the best performing method of feature-level fusion\nresulting in 53% accuracy for the recognition of 40 different types of trees,\ncorresponding to an improvement of 5.7% over the best performing baseline when\nRGB, multispectral, and LiDAR data are used. We also provide an in-depth\ncomparison by evaluating each model at various parameter complexity settings,\nwhere the increased model capacity results in a further improvement of 6.3%\nover the default capacity setting.\n"}
{"abstract": "  The spectrum of a real and symmetric $N\\times N$ matrix determines the matrix\nup to unitary equivalence. More spectral data is needed together with some sign\nindicators to remove the unitary ambiguities. In the first part of this work we\nspecify the spectral and sign information required for a unique reconstruction\nof general matrices. More specifically, the spectral information consists of\nthe spectra of the $N$ nested main minors of the original matrix of the sizes\n$1,2,\\dots,N$. However, due to the complicated nature of the required sign\ndata, improvements are needed in order to make the reconstruction procedure\nfeasible. With this in mind, the second part is restricted to banded matrices\nwhere the amount of spectral data exceeds the number of the unknown matrix\nentries. It is shown that one can take advantage of this redundancy to\nguarantee unique reconstruction of {\\it generic} matrices, in other words, this\nsubset of matrices is open, dense and of full measure in the set of real,\nsymmetric and banded matrices. It is shown that one can optimize the ratio\nbetween redundancy and genericity by using the freedom of choice of the\nspectral information input. We demonstrate our constructions in detail for\npentadiagonal matrices.\n"}
{"abstract": "  Trust is of paramount concern for tenants to deploy their security-sensitive\nservices in the cloud. The integrity of VMs in which these services are\ndeployed needs to be ensured even in the presence of powerful adversaries with\nadministrative access to the cloud. Traditional approaches for solving this\nchallenge leverage trusted computing techniques, e.g., vTPM, or hardware CPU\nextensions, e.g., AMD SEV. But, they are vulnerable to powerful adversaries, or\nthey provide only load time (not runtime) integrity measurements of VMs.\n  We propose WELES, a protocol allowing tenants to establish and maintain trust\nin VM runtime integrity of software and its configuration. WELES is transparent\nto the VM configuration and setup. It performs an implicit attestation of VMs\nduring a secure login and binds the VM integrity state with the secure\nconnection. Our prototype's evaluation shows that WELES is practical and incurs\nlow performance overhead.\n"}
{"abstract": "  We establish the pointwise continuity of bounded weak solutions to of a class\nof scalar parabolic equations and strongly coupled parabolic systems. Our\napproach to the regularity theory of parabolic scalar equations is quite\nelementary and its applications to strongly coupled systems does not require\nhigher $L^p$ integrability of derivatives.\n"}
{"abstract": "  Transfer learning has become the dominant paradigm for many natural language\nprocessing tasks. In addition to models being pretrained on large datasets,\nthey can be further trained on intermediate (supervised) tasks that are similar\nto the target task. For small Natural Language Inference (NLI) datasets,\nlanguage modelling is typically followed by pretraining on a large (labelled)\nNLI dataset before fine-tuning with each NLI subtask. In this work, we explore\nGradient Boosted Decision Trees (GBDTs) as an alternative to the commonly used\nMulti-Layer Perceptron (MLP) classification head. GBDTs have desirable\nproperties such as good performance on dense, numerical features and are\neffective where the ratio of the number of samples w.r.t the number of features\nis low. We then introduce FreeGBDT, a method of fitting a GBDT head on the\nfeatures computed during fine-tuning to increase performance without additional\ncomputation by the neural network. We demonstrate the effectiveness of our\nmethod on several NLI datasets using a strong baseline model (RoBERTa-large\nwith MNLI pretraining). The FreeGBDT shows a consistent improvement over the\nMLP classification head.\n"}
{"abstract": "  Non-negative Matrix Factorization (NMF) methods offer an appealing\nunsupervised learning method for real-time analysis of streaming spectral data\nin time-sensitive data collection, such as $\\textit{in situ}$ characterization\nof materials. However, canonical NMF methods are optimized to reconstruct a\nfull dataset as closely as possible, with no underlying requirement that the\nreconstruction produces components or weights representative of the true\nphysical processes. In this work, we demonstrate how constraining NMF weights\nor components, provided as known or assumed priors, can provide significant\nimprovement in revealing true underlying phenomena. We present a PyTorch based\nmethod for efficiently applying constrained NMF and demonstrate this on several\nsynthetic examples. When applied to streaming experimentally measured spectral\ndata, an expert researcher-in-the-loop can provide and dynamically adjust the\nconstraints. This set of interactive priors to the NMF model can, for example,\ncontain known or identified independent components, as well as functional\nexpectations about the mixing of components. We demonstrate this application on\nmeasured X-ray diffraction and pair distribution function data from $\\textit{in\nsitu}$ beamline experiments. Details of the method are described, and general\nguidance provided to employ constrained NMF in extraction of critical\ninformation and insights during $\\textit{in situ}$ and high-throughput\nexperiments.\n"}
{"abstract": "  Learning visual knowledge from massive weakly-labeled web videos has\nattracted growing research interests thanks to the large corpus of easily\naccessible video data on the Internet. However, for video action recognition,\nthe action of interest might only exist in arbitrary clips of untrimmed web\nvideos, resulting in high label noises in the temporal space. To address this\nissue, we introduce a new method for pre-training video action recognition\nmodels using queried web videos. Instead of trying to filter out, we propose to\nconvert the potential noises in these queried videos to useful supervision\nsignals by defining the concept of Sub-Pseudo Label (SPL). Specifically, SPL\nspans out a new set of meaningful \"middle ground\" label space constructed by\nextrapolating the original weak labels during video querying and the prior\nknowledge distilled from a teacher model. Consequently, SPL provides enriched\nsupervision for video models to learn better representations. SPL is fairly\nsimple and orthogonal to popular teacher-student self-training frameworks\nwithout extra training cost. We validate the effectiveness of our method on\nfour video action recognition datasets and a weakly-labeled image dataset to\nstudy the generalization ability. Experiments show that SPL outperforms several\nexisting pre-training strategies using pseudo-labels and the learned\nrepresentations lead to competitive results when fine-tuning on HMDB-51 and\nUCF-101 compared with recent pre-training methods.\n"}
{"abstract": "  It is known that the contribution of torsion to the equation for the chiral\nWeyl fermions can be equivalently considered in terms of the axial $U(1)$ gauge\nfield. In this scenario the gravitational field transforms to the $U(1)$ gauge\nfield. Here we show that in chiral superconductors the opposite scenario takes\nplace: the electromagnetic $U(1)$ field serves as the spin connection for the\nBogoliubov fermionic quasiparticles. As a result the electromagnetic field\ngives rise to the gravitational anomaly, which contains the extra factor $1/3$\nin the corresponding Adler-Bell-Jackiw equation as compared with the\nconventional chiral anomaly. We also consider the gravitational anomaly\nproduced in neutral Weyl superfluids by the analog of the gravitational\ninstanton, the process of creation and annihilation of the 3D topological\nobjects -- hopfions. The gravitational instanton leads to creation of the\nchiral charge.\n"}
{"abstract": "  Recently there has been increased interest in semi-supervised classification\nin the presence of graphical information. A new class of learning models has\nemerged that relies, at its most basic level, on classifying the data after\nfirst applying a graph convolution. To understand the merits of this approach,\nwe study the classification of a mixture of Gaussians, where the data\ncorresponds to the node attributes of a stochastic block model. We show that\ngraph convolution extends the regime in which the data is linearly separable by\na factor of roughly $1/\\sqrt{D}$, where $D$ is the expected degree of a node,\nas compared to the mixture model data on its own. Furthermore, we find that the\nlinear classifier obtained by minimizing the cross-entropy loss after the graph\nconvolution generalizes to out-of-distribution data where the unseen data can\nhave different intra- and inter-class edge probabilities from the training\ndata.\n"}
{"abstract": "  Regardless of their different types of progenitors and central engines,\ngamma-ray bursts (GRBs) were always assumed to be standalone systems after they\nformed. Little attention has been paid to the possibility that a stellar\ncompanion can still accompany a GRB itself. This paper investigates such a\nGRB-involved binary system and studies the effects of the stellar companion on\nthe observed GRB emission when it is located inside the jet opening angle.\nAssuming a typical emission radius of $\\sim10^{15}\\,$cm, we show that the\nblockage by a companion star with a radius of\n$R_\\mathrm{c}\\sim67\\,\\mathrm{R_\\odot}$ becomes non-negligible when it is\nlocated within a typical GRB jet opening angle (e.g., $\\sim10$ degrees) and\nbeyond the GRB emission site. In such a case, an on-axis observer will see a\nGRB with a similar temporal behavior but 25% dimmer. On the other hand, an\noff-axis observer outside the jet opening angle (hence missed the original GRB)\ncan see a delayed \"reflected\" GRB, which is much fainter in brightness, much\nwider in the temporal profile and slightly softer in energy. Our study can\nnaturally explain the origin of some low-luminosity GRBs. Moreover, we also\npoint out that the companion star may be shocked if it is located inside the\nGRB emission site, which can give rise to an X-ray transient or a GRB followed\nby a delayed X-ray bump on top of X-ray afterglows.\n"}
{"abstract": "  We present a far-UV (FUV) study of the star-forming complexes (SFCs) in three\nnearby galaxies using the Ultraviolet Imaging Telescope (UVIT). The galaxies\nare close to face-on and show significant outer disk star formation. Two of\nthem are isolated (NGC 628, NGC 6946), and one is interacting with distant\ncompanions (NGC 5457). We compared the properties of the SFCs inside and\noutside the optical radius (R$_{25}$). We estimated the sizes, star formation\nrates (SFRs), metallicities, and the Toomre Q parameter of the SFCs. We find\nthat the outer disk SFCs are at least ten times smaller in area than those in\nthe inner disk. The SFR per unit area ($\\Sigma_{SFR}$) in both regions have\nsimilar mean values, but the outer SFCs have a much smaller range of\n$\\Sigma_{SFR}$. They are also metal-poor compared to the inner disk SFCs. The\nFUV emission is well correlated with the neutral hydrogen gas (\\HI)\ndistribution and is detected within and near several \\HI~holes. Our estimation\nof the Q parameter in the outer disks of the two isolated galaxies suggests\nthat their outer disks are stable (Q$>$1). However, their FUV images indicate\nthat there is ongoing star formation in these regions. This suggests that there\nmay be some non-luminous mass or dark matter in their outer disks, which\nincreases the disk surface density and supports the formation of local\ngravitational instabilities. In the interacting galaxy, NGC 5457, the baryonic\nsurface density is sufficient (Q$<$1) to trigger local disk instabilities in\nthe outer disk.\n"}
{"abstract": "  We study a new application for text generation -- idiomatic sentence\ngeneration -- which aims to transfer literal phrases in sentences into their\nidiomatic counterparts. Inspired by psycholinguistic theories of idiom use in\none's native language, we propose a novel approach for this task, which\nretrieves the appropriate idiom for a given literal sentence, extracts the span\nof the sentence to be replaced by the idiom, and generates the idiomatic\nsentence by using a neural model to combine the retrieved idiom and the\nremainder of the sentence. Experiments on a novel dataset created for this task\nshow that our model is able to effectively transfer literal sentences into\nidiomatic ones. Furthermore, automatic and human evaluations show that for this\ntask, the proposed model outperforms a series of competitive baseline models\nfor text generation.\n"}
{"abstract": "  Extracting temporal relationships over a range of scales is a hallmark of\nhuman perception and cognition -- and thus it is a critical feature of machine\nlearning applied to real-world problems. Neural networks are either plagued by\nthe exploding/vanishing gradient problem in recurrent neural networks (RNNs) or\nmust adjust their parameters to learn the relevant time scales (e.g., in\nLSTMs). This paper introduces DeepSITH, a network comprising\nbiologically-inspired Scale-Invariant Temporal History (SITH) modules in series\nwith dense connections between layers. SITH modules respond to their inputs\nwith a geometrically-spaced set of time constants, enabling the DeepSITH\nnetwork to learn problems along a continuum of time-scales. We compare DeepSITH\nto LSTMs and other recent RNNs on several time series prediction and decoding\ntasks. DeepSITH achieves state-of-the-art performance on these problems.\n"}
{"abstract": "  The core problem of Magnetic Resonance Imaging (MRI) is the trade off between\nacceleration and image quality. Image reconstruction and super-resolution are\ntwo crucial techniques in Magnetic Resonance Imaging (MRI). Current methods are\ndesigned to perform these tasks separately, ignoring the correlations between\nthem. In this work, we propose an end-to-end task transformer network\n(T$^2$Net) for joint MRI reconstruction and super-resolution, which allows\nrepresentations and feature transmission to be shared between multiple task to\nachieve higher-quality, super-resolved and motion-artifacts-free images from\nhighly undersampled and degenerated MRI data. Our framework combines both\nreconstruction and super-resolution, divided into two sub-branches, whose\nfeatures are expressed as queries and keys. Specifically, we encourage joint\nfeature learning between the two tasks, thereby transferring accurate task\ninformation. We first use two separate CNN branches to extract task-specific\nfeatures. Then, a task transformer module is designed to embed and synthesize\nthe relevance between the two tasks. Experimental results show that our\nmulti-task model significantly outperforms advanced sequential methods, both\nquantitatively and qualitatively.\n"}
{"abstract": "  In recent years, weak lensing of the cosmic microwave background (CMB) has\nemerged as a powerful tool to probe fundamental physics, such as neutrino\nmasses, primordial non-Gaussianity, dark energy, and modified gravity. The\nprime target of CMB lensing surveys is the lensing potential, which is\nreconstructed from observed CMB temperature $T$ and polarization $E$ and $B$\nfields. Until very recently, this reconstruction has been performed with\nquadratic estimators (QEs), which, although known to be suboptimal for\nhigh-sensitivity experiments, are numerically efficient, and useful to make\nforecasts and cross-check the results of more sophisticated likelihood-based\nmethods. It is expected that ongoing and near-future CMB experiments such as\nAdvACT, SPT-3G and the Simons Observatory (SO), will also rely on QEs. Here, we\nreview different QEs, and clarify their differences. In particular, we show\nthat the Hu-Okamoto (HO02) estimator is not the absolute optimal lensing\nestimator that can be constructed out of quadratic combinations of $T, E$ and\n$B$ fields. Instead, we derive the global-minimum-variance (GMV) lensing\nquadratic estimator. Although this estimator can be found elsewhere in the\nliterature, it was erroneously described as equivalent to the HO02 estimator,\nand has never been used in real data analyses. Here, we show explicitly that\nthe HO02 estimator is suboptimal to the GMV estimator, with a reconstruction\nnoise larger by up to $\\sim 9\\%$ for a SO-like experiment. We further show that\nthe QE used in the Planck, and recent SPT lensing analysis are suboptimal to\nboth the HO02 and GMV estimator, and would have a reconstruction noise up to\n$\\sim 11\\%$ larger than that of the GMV estimator for a SO-like experiment. In\naddition to clarifying differences between different QEs, this work should thus\nprovide motivation to implement the GMV estimator in future lensing analyses\nrelying on QEs.\n"}
{"abstract": "  We propose a neural network approach for solving high-dimensional optimal\ncontrol problems arising in real-time applications. Our approach yields\ncontrols in a feedback form, where the policy function is given by a neural\nnetwork (NN). Specifically, we fuse the Hamilton-Jacobi-Bellman (HJB) and\nPontryagin Maximum Principle (PMP) approaches by parameterizing the value\nfunction with an NN. We can therefore synthesize controls in real-time without\nhaving to solve an optimization problem. Once the policy function is trained,\ngenerating a control at a given space-time location takes milliseconds; in\ncontrast, efficient nonlinear programming methods typically perform the same\ntask in seconds. We train the NN offline using the objective function of the\ncontrol problem and penalty terms that enforce the HJB equations. Therefore,\nour training algorithm does not involve data generated by another algorithm. By\ntraining on a distribution of initial states, we ensure the controls'\noptimality on a large portion of the state-space. Our grid-free approach scales\nefficiently to dimensions where grids become impractical or infeasible. We\ndemonstrate the effectiveness of our approach on several multi-agent\ncollision-avoidance problems in up to 150 dimensions. Furthermore, we\nempirically observe that the number of parameters in our approach scales\nlinearly with the dimension of the control problem, thereby mitigating the\ncurse of dimensionality.\n"}
{"abstract": "  We report measurements of branching fractions ($\\mathcal B$) and direct ${\\it\nCP}$-violating asymmetries ($\\mathcal A_{\\it CP}$) for the decays $B^+\\to\nK^+\\pi^0$ and $B^+ \\to \\pi^+\\pi^0$ reconstructed with the Belle II detector in\na sample of asymmetric-energy electron-positron collisions at the\n$\\Upsilon(4S)$ resonance corresponding to 62.8 $\\text{fb}^{-1}$ of integrated\nluminosity. The results are $\\mathcal{B}(B^+ \\to K^+\\pi^0) = [11.9\n^{+1.1}_{-1.0} (\\rm stat)\\pm 1.6(\\rm syst)]\\times 10^{-6}$, $\\mathcal{B}(B^+\n\\to \\pi^+\\pi^0) = [5.5 ^{+1.0}_{-0.9} (\\rm stat)\\pm 0.7(\\rm syst)]\\times\n10^{-6}$, $\\mathcal A_{\\it CP}(B^+ \\to K^+\\pi^0) = -0.09 \\pm 0.09 (\\rm stat)\\pm\n0.03(\\rm syst)$, and $\\mathcal A_{\\it CP}(B^+ \\to \\pi^+\\pi^0) = -0.04 \\pm 0.17\n(\\rm stat)\\pm 0.06(\\rm syst)$. The results are consistent with previous\nmeasurements and show a detector performance comparable with early Belle\nperformance.\n"}
{"abstract": "  Myers shows that every compact, connected, orientable $3$--manifold with no\n$2$--sphere boundary components contains a hyperbolic knot. We use work of\nIkeda with an observation of Adams-Reid to show that every $3$--manifold\nsubject to the above conditions contains a hyperbolic knot which admits a\nnon-trivial non-hyperbolic surgery, a toroidal surgery in particular. We\nconclude with a question and a conjecture about reducible surgeries.\n"}
{"abstract": "  Inspired by the way human brain works, the emerging hyperdimensional\ncomputing (HDC) is getting more and more attention. HDC is an emerging\ncomputing scheme based on the working mechanism of brain that computes with\ndeep and abstract patterns of neural activity instead of actual numbers.\nCompared with traditional ML algorithms such as DNN, HDC is more\nmemory-centric, granting it advantages such as relatively smaller model size,\nless computation cost, and one-shot learning, making it a promising candidate\nin low-cost computing platforms. However, the robustness of HDC models have not\nbeen systematically studied. In this paper, we systematically expose the\nunexpected or incorrect behaviors of HDC models by developing HDXplore, a\nblackbox differential testing-based framework. We leverage multiple HDC models\nwith similar functionality as cross-referencing oracles to avoid manual\nchecking or labeling the original input. We also propose different perturbation\nmechanisms in HDXplore. HDXplore automatically finds thousands of incorrect\ncorner case behaviors of the HDC model. We propose two retraining mechanisms\nand using the corner cases generated by HDXplore to retrain the HDC model, we\ncan improve the model accuracy by up to 9%.\n"}
{"abstract": "  The Ladder of Causation describes three qualitatively different types of\nactivities an agent may be interested in engaging in, namely, seeing\n(observational), doing (interventional), and imagining (counterfactual) (Pearl\nand Mackenzie, 2018). The inferential challenge imposed by the causal hierarchy\nis that data is collected by an agent observing or intervening in a system\n(layers 1 and 2), while its goal may be to understand what would have happened\nhad it taken a different course of action, contrary to what factually ended up\nhappening (layer 3). While there exists a solid understanding of the conditions\nunder which cross-layer inferences are allowed from observations to\ninterventions, the results are somewhat scarcer when targeting counterfactual\nquantities. In this paper, we study the identification of nested\ncounterfactuals from an arbitrary combination of observations and experiments.\nSpecifically, building on a more explicit definition of nested counterfactuals,\nwe prove the counterfactual unnesting theorem (CUT), which allows one to map\narbitrary nested counterfactuals to unnested ones. For instance, applications\nin mediation and fairness analysis usually evoke notions of direct, indirect,\nand spurious effects, which naturally require nesting. Second, we introduce a\nsufficient and necessary graphical condition for counterfactual identification\nfrom an arbitrary combination of observational and experimental distributions.\nLastly, we develop an efficient and complete algorithm for identifying nested\ncounterfactuals; failure of the algorithm returning an expression for a query\nimplies it is not identifiable.\n"}
{"abstract": "  The thermal properties of light mesons, including the temperature dependence\nof their masses (both screening and pole masses) and thermal widths, are\nstudied in a two-flavor ($N_f=2$) soft-wall AdS/QCD model. By solving the\nspatial correlation functions, we extract the screening masses ($m_{\\rm{scr}}$)\nfrom their poles. The screening masses of pseudo-scalar ($\\pi$) and\naxial-vector ($a_1$) mesons increase almost monotonously with the increase of\ntemperature. The screening masses of scalar ($\\sigma$) and vector ($\\rho$)\nmesons decrease at low temperature and increase at high temperature. The pole\nmasses ($m_{\\rm{pole}}$) and the thermal widths ($\\Gamma$) are extracted from\nthe temporal correlation functions and the corresponding spectral functions.\nThe results indicate that the pole masses have local minima at low temperature\nand increase at high temperature. The thermal widths increase rapidly above the\nchiral crossover temperature $T_{cp}$, indicating the dissociations of mesons\nat high temperature. Furthermore, the degeneration of the chiral partners\n($\\pi$ and $\\sigma$, $\\rho$ and $a_1$) above $T_{cp}$ is observed from the\nscreening and pole masses, revealing the chiral symmetry restoration at the\nhadronic spectrum level. Finally, we numerically verify that the spectral\nfunctions in the temporal regime are strongly related to the quasi-normal modes\nwith complex frequencies $\\omega_0=m_{\\rm{pole}}-i \\Gamma/2$.\n"}
{"abstract": "  In this work, we explore the phenomenon of cosmic evolution using curved FLRW\nspace-time bounded by apparent horizon with a specific holographic cut-off. To\nthis end, we use the framework of Rastall gravity (RG) and universe is assumed\nto be consists of interacting/ non-interacting dark energy (DE) and dark matter\n(DM). In both scenarios, we evaluate exact solutions of the dynamical equations\nand constraint the holographic parameter $c^{2}(z)$ assuming a slowly varying\nfunction of red-shift during non-interacting model. For interacting model, we\nconsider $c^{2}$ as a constant function of red-shift. Moreover, we analyze\nnature of the obtained results via deceleration parameter $(q)$, statefinder\npair $(j,s)$ and $Om(z)$-diagnostic by constraining the involved model\nparameters using latest observational data. The graphical analysis showed that\ninteracting model is very close to $\\Lambda$ cold DM ($\\Lambda$CDM) model as\ncompared to non-interacting case. We conclude that this holographic proposal is\nenough to describe the cosmic evolution at an accelerating rate.\n"}
{"abstract": "  End-to-end production of object tracklets from high resolution video in\nreal-time and with high accuracy remains a challenging problem due to the cost\nof object detection on each frame. In this work we present Localization-based\nTracking (LBT), an extension to any tracker that follows the tracking by\ndetection or joint detection and tracking paradigms. Localization-based\nTracking focuses only on regions likely to contain objects to boost detection\nspeed and avoid matching errors. We evaluate LBT as an extension to two example\ntrackers (KIOU and SORT) on the UA-DETRAC and MOT20 datasets. LBT-extended\ntrackers outperform all other reported algorithms in terms of PR-MOTA, PR-MOTP,\nand mostly tracked objects on the UA-DETRAC benchmark, establishing a new\nstate-of-the art. relative to tracking by detection with KIOU, LBT-extended\nKIOU achieves a 25% higher frame-rate and is 1.1% more accurate in terms of\nPR-MOTA on the UA-DETRAC dataset. LBT-extended SORT achieves a 62% speedup and\na 3.2% increase in PR-MOTA on the UA-DETRAC dataset. On MOT20, LBT-extended\nKIOU has a 50% higher frame-rate than tracking by detection and is 0.4% more\naccurate in terms of MOTA. As of submission time, our LBT-extended KIOU tracker\nplaces 10th overall on the MOT20 benchmark.\n"}
{"abstract": "  In this report, we present a theoretical support of the continual learning\nmethod \\textbf{Elastic Weight Consolidation}, introduced in paper titled\n`Overcoming catastrophic forgetting in neural networks'. Being one of the most\ncited paper in regularized methods for continual learning, this report\ndisentangles the underlying concept of the proposed objective function. We\nassume that the reader is aware of the basic terminologies of continual\nlearning.\n"}
{"abstract": "  Domain specific neural network accelerators have garnered attention because\nof their improved energy efficiency and inference performance compared to CPUs\nand GPUs. Such accelerators are thus well suited for resource-constrained\nembedded systems. However, mapping sophisticated neural network models on these\naccelerators still entails significant energy and memory consumption, along\nwith high inference time overhead. Binarized neural networks (BNNs), which\nutilize single-bit weights, represent an efficient way to implement and deploy\nneural network models on accelerators. In this paper, we present a novel\noptical-domain BNN accelerator, named ROBIN, which intelligently integrates\nheterogeneous microring resonator optical devices with complementary\ncapabilities to efficiently implement the key functionalities in BNNs. We\nperform detailed fabrication-process variation analyses at the optical device\nlevel, explore efficient corrective tuning for these devices, and integrate\ncircuit-level optimization to counter thermal variations. As a result, our\nproposed ROBIN architecture possesses the desirable traits of being robust,\nenergy-efficient, low latency, and high throughput, when executing BNN models.\nOur analysis shows that ROBIN can outperform the best-known optical BNN\naccelerators and also many electronic accelerators. Specifically, our\nenergy-efficient ROBIN design exhibits energy-per-bit values that are ~4x lower\nthan electronic BNN accelerators and ~933x lower than a recently proposed\nphotonic BNN accelerator, while a performance-efficient ROBIN design shows ~3x\nand ~25x better performance than electronic and photonic BNN accelerators,\nrespectively.\n"}
{"abstract": "  For the 2D compressible isentropic Euler equations of polytropic gases with\nan initial perturbation of size $\\ve$ of a rest state, it has been known that\nif the initial data are rotationnally invariant or irrotational, then the\nlifespan $T_{\\ve}$ of the classical solutions is of order $O(\\f{1}{\\ve^2})$; if\nthe initial vorticity is of size $\\ve^{1+\\al}$ ($0\\le\\al\\le 1$), then $T_{\\ve}$\nis of $O(\\f{1}{\\ve^{1+\\al}})$. In the present paper, for the 2D compressible\nisentropic Euler equations of Chaplygin gases, if the initial data are a\nperturbation of size $\\ve$, and the initial vorticity is of any size $\\dl$ with\n$0<\\dl\\le \\ve$, we will establish the lifespan $T_{\\dl}=O(\\f{1}{\\dl})$. For\nexamples, if $\\dl=e^{-\\f{1}{\\ve^2}}$ or $\\dl=e^{-e^{\\f{1}{\\ve^2}}}$ are chosen,\nthen $T_{\\dl}=O(e^{\\f{1}{\\ve^2}})$ or $T_{\\dl}=O(e^{e^{\\f{1}{\\ve^2}}})$\nalthough the perturbations of the initial density and the divergence of the\ninitial velocity are only of order $O(\\ve)$. Our main ingredients are: finding\nthe null condition structures in 2D compressible Euler equations of Chaplygin\ngases and looking for the good unknown; establishing a new class of weighted\nspace-time $L^\\infty$-$L^\\infty$ estimates for the solution itself and its\ngradients of 2D linear wave equations; introducing some suitably weighted\nenergies and taking the $L^p$ $(1<p<\\infty)$ estimates on the vorticity.\n"}
{"abstract": "  We survey determinantal singularities, their deformations, and their\ntopology. This class of singularities generalizes the well studied case of\ncomplete intersections in several different aspects, but exhibits a plethora of\nnew phenomena such as for instance non-isolated singularities which are\nfinitely determined, or smoothings with low connectivity; already the union of\nthe coordinate axes in $(\\mathbb{C}^3,0)$ is determinantal, but not a complete\nintersection. We start with the algebraic background and then continue by\ndiscussing the subtle interplay of unfoldings and deformations in this setting,\nincluding a survey of the case of determinantal hypersurfaces, Cohen-Macaulay\ncodimension $2$ and Gorenstein codimension $3$ singularities, and determinantal\nrational surface singularities. We conclude with a discussion of essential\nsmoothings and provide an appendix listing known classifications of simple\ndeterminantal singularities.\n"}
{"abstract": "  The solid-on-solid model is a model of height functions, introduced to study\nthe interface separating the $+$ and $-$ phase in the Ising model. The planar\nsolid-on-solid model thus corresponds to the three-dimensional Ising model.\nDelocalisation of this model at high temperature and at zero slope was first\nderived by Fr\\\"ohlich and Spencer, in parallel to proving the\nKosterlitz-Thouless phase transition. The first main result of this article\nconsists of a simple, alternative proof for delocalisation of the\nsolid-on-solid model. In fact, the argument is more general: it works on any\nplanar graph -- not just the square lattice -- and implies that the interface\ndelocalises at any slope rather than exclusively at the zero slope. The second\nmain result is that the absolute value of the height function in this model\nsatisfies the FKG lattice condition. This property is believed to be intimately\nlinked to the (quantitative) understanding of delocalisation, given the recent\nsuccesses in the context of the square ice and (more generally) the six-vertex\nmodel. The new FKG inequality is shown to hold true for both the solid-on-solid\nmodel as well as for the discrete Gaussian model, which implies immediately\nthat the two notions of delocalisation, namely delocalisation in finite volume\nand delocalisation of shift-invariant Gibbs measures, coincide.\n"}
{"abstract": "  The dark (or hidden) photon is a massive U(1) gauge boson theorized as a dark\nforce mediator and as a dark matter candidate. Dark photons can be detected\nwith axion cavity haloscopes by probing for a power excess caused by the dark\nphoton's kinetic mixing with Standard Model photons. Haloscope axion exclusion\nlimits may therefore be converted into competitive dark photon parameter limits\nvia the calculation of a corresponding dark photon to photon coupling factor.\nThis calculation allows for an improvement in sensitivity of around four orders\nof magnitude relative to other dark photon exclusions and may be attained using\nexisting data. We present how one converts haloscope axion search limits and a\nsummary of relevant experimental parameters from published searches. In\naddition, we have included the code that can be used to generate our dark\nphoton exclusion limits for the cases described in this paper. Finally, we\npresent limits on the kinetic mixing coefficient between dark photons and the\nStandard Model photons based on existing haloscope axion searches.\n"}
{"abstract": "  It is widely assumed that a star and its protoplanetary disk are initially\naligned, with the stellar equator parallel to the disk plane. When observations\nreveal a misalignment between stellar rotation and the orbital motion of a\nplanet, the usual interpretation is that the initial alignment was upset by\ngravitational perturbations that took place after planet formation. Most of the\npreviously known misalignments involve isolated hot Jupiters, for which\nplanet-planet scattering or secular effects from a wider-orbiting planet are\nthe leading explanations. In theory, star/disk misalignments can result from\nturbulence during star formation or the gravitational torque of a wide-orbiting\ncompanion star, but no definite examples of this scenario are known. An ideal\nexample would combine a coplanar system of multiple planets -- ruling out\nplanet-planet scattering or other disruptive post-formation events -- with a\nbackward-rotating star, a condition that is easier to obtain from a primordial\nmisalignment than from post-formation perturbations. There are two previously\nknown examples of a misaligned star in a coplanar multi-planet system, but in\nneither case has a suitable companion star been identified, nor is the stellar\nrotation known to be retrograde. Here, we show that the star K2-290 A is tilted\nby $124\\pm 6$ degrees compared to the orbits of both of its known planets, and\nhas a wide-orbiting stellar companion that is capable of having tilted the\nprotoplanetary disk. The system provides the clearest demonstration that stars\nand protoplanetary disks can become grossly misaligned due to the gravitational\ntorque from a neighbouring star.\n"}
{"abstract": "  Humans can synchronize with musical events whilst coordinating their\nmovements with others. Interpersonal entrainment phenomena, such as dance,\ninvolve multiple body parts and movement directions. Along with being\nmultidimensional, dance movement interaction is plurifrequential, since it can\noccur at different frequencies simultaneously. Moreover, it is prone to\nnonstationarity, due to, for instance, displacements around the dance floor.\nVarious methodological approaches have been adopted to study entrainment, but\nonly spectrogram-based techniques allow for an integral analysis thereof. This\narticle proposes an alternative approach based upon the cross-wavelet\ntransform, a technique for nonstationary and plurifrequential analysis of\nunivariate interaction. The presented approach generalizes the cross-wavelet\ntransform to multidimensional signals. It allows to identify, for different\nfrequencies of movement, interaction estimates of interaction and\nleader-follower dynamics across body parts and movement directions. Further,\nthe generalized cross-wavelet transform can be used to quantify the\nfrequency-wise contribution of individual body parts and movement directions to\noverall synchrony. The article provides a thorough mathematical description of\nthe method and includes proofs of its invariance under translation, rotation,\nand reflection. Finally, its properties and performance are illustrated via\nexamples using simulated data and behavioral data collected through a mirror\ngame task and a free dance movement task.\n"}
{"abstract": "  This report presents the application of object detection on a database of\nunderwater images of different species of crabs, as well as aerial images of\nsea lions and finally the Pascal VOC dataset. The model is an end-to-end object\ndetection neural network based on a convolutional network base and a Long\nShort-Term Memory detector.\n"}
{"abstract": "  Several classes of near-MDS codes of ${\\rm PG}(3,q)$ are described. They are\nobtained either by considering the intersection of an elliptic quadric ovoid\nand a Suzuki-Tits ovoid of a symplectic polar space ${\\cal W}(3, q)$ or\nstarting from the $q+1$ points of a twisted cubic of ${\\rm PG}(3, q)$. As a\nby-product two classes of complete caps of ${\\rm PG}(4,q)$ of size $2q^2-q \\pm\n\\sqrt{2q} + 2$ are exhibited.\n"}
{"abstract": "  We showcase the advantages of orbital-free density-potential functional\ntheory (DPFT), a more flexible variant of Hohenberg-Kohn density functional\ntheory. DPFT resolves the usual trouble with the gradient-expanded kinetic\nenergy functional by facilitating systematic semiclassical approximations in\nterms of an effective potential energy that incorporates all interactions. With\nthe aid of two systematic approximation schemes we demonstrate that DPFT is not\nonly scalable, universally applicable in both position and momentum space, and\nallows kinetic and interaction energy to be approximated consistently, but can\nalso compete with highly accurate, yet restricted, methods. As two- and\nthree-dimensional geometries are extensively covered elsewhere, our focus here\nis on one-dimensional settings, with semiclassical observables systematically\nderived from both the Wigner function formalism and a split-operator approach.\nThe high quality of our results for Fermi gases in Morse potentials invites the\nuse of DPFT for describing more exotic systems, such as trapped large-spin\nfermion mixtures with contact or dipole-dipole interactions.\n"}
{"abstract": "  We aim to characterize a diverse selection of dense, potentially star-forming\ncores, clumps, and clouds within the Milky Way in terms of their dust emission\nand SF activity. We studied 53 fields that have been observed in the JCMT\nSCUBA-2 continuum survey SCOPE and have been mapped with Herschel. We estimated\ndust properties by fitting Herschel observations with modified blackbody\nfunctions, studied the relationship between dust temperature and dust opacity\nspectral index $\\beta$, and estimated column densities. We extracted clumps\nfrom the SCUBA-2 850 $\\mu$m maps with the FellWalker algorithm and examined\ntheir masses and sizes. Clumps are associated with young stellar objects found\nin several catalogs. We estimated the gravitational stability of the clumps\nwith virial analysis. The clumps are categorized as unbound starless,\nprestellar, or protostellar. We find 529 dense clumps, typically with high\ncolumn densities from (0.3-4.8)$\\times 10^{22}$ cm$^{-2}$, with a mean of\n(1.5$\\pm$0.04)$\\times10^{22}$ cm$^{-2}$, low temperatures ($T\\sim $10-20 K),\nand estimated submillimeter $\\beta$ =1.7$\\pm$0.1. We detect a slight increase\nin opacity spectral index toward millimeter wavelengths. Masses of the sources\nrange from 0.04 $M_\\odot$ to 4259 $M_\\odot$. Mass, linear size, and temperature\nare correlated with distance. Furthermore, the estimated gravitational\nstability is dependent on distance, and more distant clumps appear more\nvirially bound. Finally, we present a catalog of properties of the clumps.Our\nsources present a large array of SF regions, from high-latitude, nearby diffuse\nclouds to large SF complexes near the Galactic center. Analysis of these\nregions will continue with the addition of molecular line data, which will\nallow us to study the densest regions of the clumps in more detail.\n"}
{"abstract": "  Progress in deep reinforcement learning (RL) research is largely enabled by\nbenchmark task environments. However, analyzing the nature of those\nenvironments is often overlooked. In particular, we still do not have agreeable\nways to measure the difficulty or solvability of a task, given that each has\nfundamentally different actions, observations, dynamics, rewards, and can be\ntackled with diverse RL algorithms. In this work, we propose policy information\ncapacity (PIC) -- the mutual information between policy parameters and episodic\nreturn -- and policy-optimal information capacity (POIC) -- between policy\nparameters and episodic optimality -- as two environment-agnostic,\nalgorithm-agnostic quantitative metrics for task difficulty. Evaluating our\nmetrics across toy environments as well as continuous control benchmark tasks\nfrom OpenAI Gym and DeepMind Control Suite, we empirically demonstrate that\nthese information-theoretic metrics have higher correlations with normalized\ntask solvability scores than a variety of alternatives. Lastly, we show that\nthese metrics can also be used for fast and compute-efficient optimizations of\nkey design parameters such as reward shaping, policy architectures, and MDP\nproperties for better solvability by RL algorithms without ever running full RL\nexperiments.\n"}
{"abstract": "  Recently, topological phases in non-Hermitian systems have attracted much\nattention because non-Hermiticity sometimes gives rise to unique phases with no\nHermitian counterparts. Non-Hermitian Bloch Hamiltonians can always be mapped\nto doubled Hermitianized Hamiltonians with chiral symmetry, which enables us to\nutilize the existing framework for Hermitian systems into the classification of\nnon-Hermitian topological phases. While this strategy succeeded in the\ntopological classification of non-Hermitian Bloch Hamiltonians in the presence\nof internal symmetries, the generalization of symmetry indicators -- a way to\nefficiently diagnose topological phases -- to non-Hermitian systems is still\nelusive. In this work, we study a theory of symmetry indicators for\nnon-Hermitian systems. We define space group symmetries of non-Hermitian Bloch\nHamiltonians as ones of the doubled Hermitianized Hamiltonians. Consequently,\nsymmetry indicator groups for chiral symmetric Hermitian systems are equivalent\nto those for non-Hermitian systems. Based on this equivalence, we list symmetry\nindicator groups for non-Hermitian systems in the presence of space group\nsymmetries. We also discuss the physical implications of symmetry indicators\nfor some symmetry classes. Furthermore, explicit formulas of symmetry\nindicators for spinful electronic systems are included in appendices.\n"}
{"abstract": "  A channel from a process p to a process q satisfies the ADD property if there\nare constants K and D, unknown to the processes, such that in any sequence of K\nconsecutive messages sent by p to q, at least one of them is delivered to q at\nmost D time units after it has been sent. This paper studies implementations of\nan eventual leader, namely, an {\\Omega} failure detector, in an arbitrarily\nconnected network of eventual ADD channels, where processes may fail by\ncrashing. It first presents an algorithm that assumes that processes initially\nknow n, the total number of processes, sending messages of size O( log n).\nThen, it presents a second algorithm that does not assume the processes know n.\nEventually the size of the messages sent by this algorithm is also O( log n).\nThese are the first implementations of leader election in the ADD model. In\nthis model, only eventually perfect failure detectors were considered, sending\nmessages of size O(n log n).\n"}
{"abstract": "  In this paper we present the dataset of Himachali low resource endangered\nlanguage, Kangri (ISO 639-3xnr) listed in the United Nations Educational,\nScientific and Cultural Organization (UNESCO). The compilation of kangri corpus\nhas been a challenging task due to the non-availability of the digitalized\nresources. The corpus contains 1,81,552 Monolingual and 27,362 Hindi-Kangri\nParallel corpora. We shared pre-trained kangri word embeddings. We also\nreported the Bilingual Evaluation Understudy (BLEU) score and Metric for\nEvaluation of Translation with Explicit ORdering (METEOR) score of Statistical\nMachine Translation (SMT) and Neural Machine Translation (NMT) results for the\ncorpus. The corpus is freely available for non-commercial usages and research.\nTo the best of our knowledge, this is the first Himachali low resource\nendangered language corpus. The resources are available at\n(https://github.com/chauhanshweta/Kangri_corpus)\n"}
{"abstract": "  We consider the yielding behavior of amorphous solids under cyclic shear\ndeformation and show that it can be mapped into a random walk in a confining\npotential with an absorbing boundary. The resulting dynamics is governed by the\nfirst passage time into the absorbing state and suffices to capture the\nessential qualitative features recently observed in atomistic simulations of\namorphous solids. Our results provide insight into the mechanism underlying\nyielding and its robustness. When the possibility of activated escape from\nabsorbing states is added, it leads to a unique determination of a threshold\nenergy and yield strain, suggesting thereby an appealing approach to\nunderstanding fatigue failure.\n"}
{"abstract": "  Following the work of Gao and Xie in [2], we state some properties of the\ninverse Kazhdan-Lusztig polynomial of a matroid. We also give partial answers\nto a conjecture that states that regular connected matroids are non-degenerate.\nWe link the degeneracy of a matroid to the inverse Kazhdan-Lusztig polynomial\nand we show that the Conjecture holds for modular matroids, by proving that\ndegenerate modular matroids are not regular.\n"}
{"abstract": "  In this article it is reported a formulation of the solenoidal nature of\nquantum electronic currents at the nanoscale whose divergence is expressed as\nthe coupling of a magnetic field, interacting with a quantum body, and a\nweighted Cern invariant vector making then a direct topological interpretation\nof this quantum magnetic phenomenon. Also a Fourier analysis of the signaling\nof electronic waves is reported in an ab initio formalism from first\npinciples\\cite{Martin}.\n"}
{"abstract": "  Problem definition: Mining for heterogeneous responses to an intervention is\na crucial step for data-driven operations, for instance to personalize\ntreatment or pricing. We investigate how to estimate price sensitivity from\ntransaction-level data. In causal inference terms, we estimate heterogeneous\ntreatment effects when (a) the response to treatment (here, whether a customer\nbuys a product) is binary, and (b) treatment assignments are partially observed\n(here, full information is only available for purchased items).\nMethodology/Results: We propose a recursive partitioning procedure to estimate\nheterogeneous odds ratio, a widely used measure of treatment effect in medicine\nand social sciences. We integrate an adversarial imputation step to allow for\nrobust inference even in presence of partially observed treatment assignments.\nWe validate our methodology on synthetic data and apply it to three case\nstudies from political science, medicine, and revenue management. Managerial\nImplications: Our robust heterogeneous odds ratio estimation method is a simple\nand intuitive tool to quantify heterogeneity in patients or customers and\npersonalize interventions, while lifting a central limitation in many revenue\nmanagement data.\n"}
{"abstract": "  In this paper, we investigate the recent studies on multimedia edge\ncomputing, from sensing not only traditional visual/audio data but also\nindividuals' geographical preference and mobility behaviors, to performing\ndistributed machine learning over such data using the joint edge and cloud\ninfrastructure and using evolutional strategies like reinforcement learning and\nonline learning at edge devices to optimize the quality of experience for\nmultimedia services at the last mile proactively. We provide both a\nretrospective view of recent rapid migration (resp. merge) of cloud multimedia\nto (resp. and) edge-aware multimedia and insights on the fundamental guidelines\nfor designing multimedia edge computing strategies that target satisfying the\nchanging demand of quality of experience. By showing the recent research\nstudies and industrial solutions, we also provide future directions towards\nhigh-quality multimedia services over edge computing.\n"}
{"abstract": "  Hydrodynamics is shown to induce non-Hermitian topological phenomena in\nordinary, passive soft matter. This is demonstrated for the first time by\nsubjecting a 2D elastic lattice to a low-Reynolds viscous flow. The interplay\nof hydrodynamics and elasticity splits Dirac cones into bulk Fermi arcs,\npairing exceptional points with opposite half-integer topological charges. The\nbulk Fermi arc is a generic hallmark of the system exhibited in all lattice and\nflow symmetries. Analytic model and simulations explain how the emergent\nsingularities shape the spectral bands and give rise to a web of van Hove\nsingularity lines in the density of states. The present findings suggest that\nnon-Hermitian physics can be explored in a broad class of ordinary soft matter,\nliving and artificial alike, opening avenues for topology-based technology in\nthis regime.\n"}
{"abstract": "  This paper provides a robust epistemic foundation for predicting and\nimplementing collective actions when only the proportions that take specific\nactions in the population matter. We apply $\\Delta$-rationalizability to\nanalyze strategic sophistication entailed in (structural) quantal response\nequilibrium (QRE); the former is called $\\Delta(p)$-rationalization to\nemphasize the only requirement on first-order beliefs is that they should be\nconsistent with the transparent knowledge of the distributions of errors in the\npopulation. We show that each QRE is a $\\Delta(p)$-rationalizable outcome. We\nalso give conditions under which the converse also holds, and prove that the\ncondition is almost never satisfied in generic games. It implies that QRE may\nbe too demanding as a predictor in general, and $\\Delta(p)$-rationalizable\noutcomes can be a robust benchmark to start from.\n"}
{"abstract": "  Reinforcement learning is about learning agent models that make the best\nsequential decisions in unknown environments. In an unknown environment, the\nagent needs to explore the environment while exploiting the collected\ninformation, which usually forms a sophisticated problem to solve.\nDerivative-free optimization, meanwhile, is capable of solving sophisticated\nproblems. It commonly uses a sampling-and-updating framework to iteratively\nimprove the solution, where exploration and exploitation are also needed to be\nwell balanced. Therefore, derivative-free optimization deals with a similar\ncore issue as reinforcement learning, and has been introduced in reinforcement\nlearning approaches, under the names of learning classifier systems and\nneuroevolution/evolutionary reinforcement learning. Although such methods have\nbeen developed for decades, recently, derivative-free reinforcement learning\nexhibits attracting increasing attention. However, recent survey on this topic\nis still lacking. In this article, we summarize methods of derivative-free\nreinforcement learning to date, and organize the methods in aspects including\nparameter updating, model selection, exploration, and parallel/distributed\nmethods. Moreover, we discuss some current limitations and possible future\ndirections, hoping that this article could bring more attentions to this topic\nand serve as a catalyst for developing novel and efficient approaches.\n"}
{"abstract": "  Distributional collision resistance is a relaxation of collision resistance\nthat only requires that it is hard to sample a collision $(x,y)$ where $x$ is\nuniformly random and $y$ is uniformly random conditioned on colliding with $x$.\nThe notion lies between one-wayness and collision resistance, but its exact\npower is still not well-understood. On one hand, distributional collision\nresistant hash functions cannot be built from one-way functions in a black-box\nway, which may suggest that they are stronger. On the other hand, so far, they\nhave not yielded any applications beyond one-way functions.\n  Assuming distributional collision resistant hash functions, we construct\n\\emph{constant-round} statistically hiding commitment scheme. Such commitments\nare not known based on one-way functions and are impossible to obtain from\none-way functions in a black-box way. Our construction relies on the reduction\nfrom inaccessible entropy generators to statistically hiding commitments by\nHaitner et al.\\ (STOC '09). In the converse direction, we show that two-message\nstatistically hiding commitments imply distributional collision resistance,\nthereby establishing a loose equivalence between the two notions.\n  A corollary of the first result is that constant-round statistically hiding\ncommitments are implied by average-case hardness in the class $SZK$ (which is\nknown to imply distributional collision resistance). This implication seems to\nbe folklore, but to the best of our knowledge has not been proven explicitly.\nWe provide yet another proof of this implication, which is arguably more direct\nthan the one going through distributional collision resistance.\n"}
{"abstract": "  Deep neural network-based classifiers trained with the categorical\ncross-entropy (CCE) loss are sensitive to label noise in the training data. One\ncommon type of method that can mitigate the impact of label noise can be viewed\nas supervised robust methods; one can simply replace the CCE loss with a loss\nthat is robust to label noise, or re-weight training samples and down-weight\nthose with higher loss values. Recently, another type of method using\nsemi-supervised learning (SSL) has been proposed, which augments these\nsupervised robust methods to exploit (possibly) noisy samples more effectively.\nAlthough supervised robust methods perform well across different data types,\nthey have been shown to be inferior to the SSL methods on image classification\ntasks under label noise. Therefore, it remains to be seen that whether these\nsupervised robust methods can also perform well if they can utilize the\nunlabeled samples more effectively. In this paper, we show that by initializing\nsupervised robust methods using representations learned through contrastive\nlearning leads to significantly improved performance under label noise.\nSurprisingly, even the simplest method (training a classifier with the CCE\nloss) can outperform the state-of-the-art SSL method by more than 50\\% under\nhigh label noise when initialized with contrastive learning. Our implementation\nwill be publicly available at\n{\\url{https://github.com/arghosh/noisy_label_pretrain}}.\n"}
{"abstract": "  Recently, several authors have adopted new alternative approaches in the\nstudy of some classical notions of modules. Among them, we find the notion of\nsubprojectivity which was introduced to measure in a way the degree of\nprojectivity of modules. The study of subprojectivity has recently been\nextended to the context of abelian categories, which has brought to light some\ninteresting new aspects. For instance, in the category of complexes, it gives a\nnew way to measure, among other things, the exactness of complexes. In this\npaper, we prove that the subprojectivity notion provides a new sight of\nnull-homotopic morphisms in the category of complexes. This will be proven\nthrough two main results. Moreover, various results which emphasize the\nimportance of subprojectivity in the category of complexes are also given.\nNamely, we give some applications by characterizing some classical rings and\nestablish various examples that allow us to reflect the scope and limits of our\nresults.\n"}
{"abstract": "  In this paper we investigate the possibility of modeling a single antenna\nalone and in close proximity to a physical object by means of discrete point\nsource scatterers. The scatter point model allows joint modeling of a physical\nantenna and the human body as a single extended object with direction dependent\nscattering coefficients for the scatter points. We introduce the term extended\nantenna describing antenna and human body together. To investigate the\nidentifiability of the model parameters we make use of ultrawideband channel\nmeasurements and accurate ground truth position and orientation measurements\nobtained with an optical tracking system. By comparing measurements of the\nantenna attached directly to the user with measurements for the antenna without\nthe user nearby, we show the shadowing and scattering effects of the human body\nand the antenna.\n"}
{"abstract": "  Flexibility design problems are a class of problems that appear in strategic\ndecision-making across industries, where the objective is to design a ($e.g.$,\nmanufacturing) network that affords flexibility and adaptivity. The underlying\ncombinatorial nature and stochastic objectives make flexibility design problems\nchallenging for standard optimization methods. In this paper, we develop a\nreinforcement learning (RL) framework for flexibility design problems.\nSpecifically, we carefully design mechanisms with noisy exploration and\nvariance reduction to ensure empirical success and show the unique advantage of\nRL in terms of fast-adaptation. Empirical results show that the RL-based method\nconsistently finds better solutions compared to classical heuristics.\n"}
{"abstract": "  We consider the order parameter $u=\\left<{\\rm Tr}\\phi^2\\right>$ as function\nof the running coupling constant $\\tau \\in \\mathbb{H}$ of asymptotically free\n$\\mathcal{N}=2$ QCD with gauge group $SU(2)$ and $N_f\\leq 3$ massive\nhypermultiplets. If the domain for $\\tau$ is restricted to an appropriate\nfundamental domain $\\mathcal{F}_{N_f}$, the function $u$ is one-to-one. We\ndemonstrate that these domains consist of six or less images of an ${\\rm\nSL}(2,\\mathbb{Z})$ keyhole fundamental domain, with appropriate identifications\nof the boundaries. For special choices of the masses, $u$ does not give rise to\nbranch points and cuts, such that $u$ is a modular function for a congruence\nsubgroup $\\Gamma$ of ${\\rm SL}(2,\\mathbb{Z})$ and the fundamental domain is\n$\\Gamma\\backslash\\mathbb{H}$. For generic masses, however, branch points and\ncuts are present, and subsets of $\\mathcal{F}_{N_f}$ are being cut and glued\nupon varying the mass. We study this mechanism for various phenomena, such as\ndecoupling of hypermultiplets, merging of local singularities, as well as\nmerging of non-local singularities which give rise to superconformal\nArgyres-Douglas theories.\n"}
{"abstract": "  Stratifying commercial product portfolios into multiple classes of decreasing\npriority, ABCD analysis, is a common supply chain tool. Key planning parameters\nthat drive strategic and execution priorities are tied to the resulting\nsegmentation. These priorities in turn drive supply chain performance. For\nlarge product assortments, manual segmentation is infeasible so an automated\nalgorithm is needed. We therefore advocate that careful attention be paid to\nthe design of such an ABCD algorithm and present three key features that can be\nincorporated into such a calculation to improve its quality and commercial\nutility.\n"}
{"abstract": "  There are relatively few conventions followed in reinforcement learning (RL)\nenvironments to structure the action spaces. As a consequence the application\nof RL algorithms to tasks with large action spaces with multiple components\nrequire additional effort to adjust to different formats. In this paper we\nintroduce {\\em Conditional Action Trees} with two main objectives: (1) as a\nmethod of structuring action spaces in RL to generalise across several action\nspace specifications, and (2) to formalise a process to significantly reduce\nthe action space by decomposing it into multiple sub-spaces, favoring a\nmulti-staged decision making approach. We show several proof-of-concept\nexperiments validating our scheme, ranging from environments with basic\ndiscrete action spaces to those with large combinatorial action spaces commonly\nfound in RTS-style games.\n"}
{"abstract": "  Let $\\mathcal{B}$ be a collection of rectangular parallelepipeds in\n$\\mathbb{R}^3$ whose sides are parallel to the coordinate axes and such that\n$\\mathcal{B}$ contains parallelepipeds with side lengths of the form $s,\n\\frac{2^N}{s} , t $, where $s, t > 0$ and $N$ lies in a nonempty subset $S$ of\nthe natural numbers. We show that if $S$ is an infinite set, then the\nassociated geometric maximal operator $M_\\mathcal{B}$ satisfies the weak type\nestimate\n  $$\\left|\\left\\{x \\in \\mathbb{R}^3 : M_{\\mathcal{B}}f(x) >\n\\alpha\\right\\}\\right| \\leq C \\int_{\\mathbb{R}^3} \\frac{|f|}{\\alpha} \\left(1 +\n\\log^+ \\frac{|f|}{\\alpha}\\right)^{2}$$ but does not satisfy an estimate of the\nform\n  $$\\left|\\left\\{x \\in \\mathbb{R}^3 : M_{\\mathcal{B}}f(x) >\n\\alpha\\right\\}\\right| \\leq C \\int_{\\mathbb{R}^3}\n\\phi\\left(\\frac{|f|}{\\alpha}\\right)$$ for any convex increasing function $\\phi:\n\\mathbb[0, \\infty) \\rightarrow [0, \\infty)$ satisfying the condition $$\\lim_{x\n\\rightarrow \\infty}\\frac{\\phi(x)}{x (\\log(1 + x))^2} = 0\\;.$$\n"}
{"abstract": "  We study the evolution of the low-temperature field-induced magnetic defects\nobserved under an applied magnetic field in a series of frustrated amorphous\nferromagnets (Fe$_{1-x}$Mn$_{x}$)$_{75}$P$_{16}$B$_{3}$Al$_{3}$ (a-FeMn).\nCombining small-angle neutron scattering and Monte Carlo simulations, we show\nthat the morphology of these defects resemble that of quasi-bidimensional spin\nvortices. They are observed in the reentrant spin-glass (RSG) phase, up to the\ncritical concentration $x_{\\rm C} \\approx 0.36$ which separates the RSG and\n\"true\" spin glass (SG) within the low temperature part of the magnetic phase\ndiagram of a-FeMn. These vortices systematically decrease in size with\nincreasing magnetic field or decreasing the average exchange interaction, and\nthey finally disappear in the SG sample ($x = 0.41$), being replaced by\nfield-induced correlations over finite length scales. We argue that the study\nof these nanoscopic defects could be used to probe the nature of the critical\nline between the RSG and SG phases.\n"}
{"abstract": "  In this letter, we discuss a new method to probe the redshift evolution of\nthe gas depletion factor, i.e. the ratio by which the gas mass fraction of\ngalaxy clusters is depleted with respect to the universal mean of baryon\nfraction. The dataset we use for this purpose consists of 40 gas mass fraction\nmeasurements measured at $r_{2500}$ using Chandra X-ray observations, strong\ngravitational lensing sub-samples obtained from SLOAN Lens ACS + BOSS\nEmission-line Lens Survey (BELLS) + Strong Legacy Survey SL2S + SLACS. For our\nanalysis, the validity of cosmic distance duality relation is assumed. We find\na mildly decreasing trend for the gas depletion factor as a function of\nredshift at about 2.7$\\sigma$. This is the first result in literature which\ndoes not find a constant gas depletion factor as a function of redshift using\ngas mass fraction measurements at $r_{2500}$.\n"}
{"abstract": "  In a 1976 landmark paper, Gordon James defined the regularisation maps on\ninteger partition, yielding certain decomposition numbers for modular\nrepresentations of $\\mathfrak{S}_n$. We describe a generalisation of James's\nregularisation map and give with proof an algorithm for such maps in the\nabacus.\n"}
{"abstract": "  Interactions between the winds of stars and the magnetospheres and\natmospheres of planets involve many processes, including the acceleration of\nparticles, heating of upper atmospheres, and a diverse range of atmospheric\nloss processes. Winds remove angular momentum from their host stars causing\nrotational spin-down and a decay in magnetic activity, which protects\natmospheres from erosion. While wind interactions are strongly influenced by\nthe X-ray and ultraviolet activity of the star and the chemical composition of\nthe atmosphere, the role of planetary magnetic fields is unclear. In this\nchapter, I review our knowledge of the properties and evolution of stellar\nactivity and winds and discuss the influences of these processes on the long\nterm evolution of planetary atmospheres. I do not consider the large number of\nimportant processes taking place at the surfaces of planets that cause\nexchanges between the atmosphere and the planet's interior.\n"}
{"abstract": "  Dimension reduction of data sets is a standard problem in the realm of\nmachine learning and knowledge reasoning. They affect patterns in and\ndependencies on data dimensions and ultimately influence any decision-making\nprocesses. Therefore, a wide variety of reduction procedures are in use, each\npursuing different objectives. A so far not considered criterion is the\nconceptual continuity of the reduction mapping, i.e., the preservation of the\nconceptual structure with respect to the original data set. Based on the notion\nscale-measure from formal concept analysis we present in this work a) the\ntheoretical foundations to detect and quantify conceptual errors in data\nscalings; b) an experimental investigation of our approach on eleven data sets\nthat were respectively treated with a variant of non-negative matrix\nfactorization.\n"}
{"abstract": "  Traditional opinion dynamics models are simple and yet, enough to explore the\nconsequences in basic scenarios. But, to better describe problems such as\npolarization and extremism, we might need to include details about human biases\nand other cognitive characteristics. In this paper, I explain how we can\ndescribe and use mental models and assumptions of the agents using\nBayesian-inspired model building. The relationship between human rationality\nand Bayesian methods will be explored, and we will see that Bayesian ideas can\nindeed be used to explain how humans reason. We will see how to use\nBayesian-inspired rules using the simplest version of the Continuous Opinions\nand Discrete Actions (CODA) model. From that, we will explore how we can obtain\nupdate rules that include human behavioral characteristics such as confirmation\nbias, motivated reasoning, or our tendency to change opinions much less than we\nshould.\n  Keywords: Opinion dynamics, Bayesian methods, Cognition, CODA, Agent-based\nmodels\n"}
{"abstract": "  Joint network topology inference represents a canonical problem of jointly\nlearning multiple graph Laplacian matrices from heterogeneous graph signals. In\nsuch a problem, a widely employed assumption is that of a simple common\ncomponent shared among multiple networks. However, in practice, a more\nintricate topological pattern, comprising simultaneously of sparse, homogeneity\nand heterogeneity components, would exhibit in multiple networks. In this\npaper, we propose a general graph estimator based on a novel structured fusion\nregularization that enables us to jointly learn multiple graph Laplacian\nmatrices with such complex topological patterns, and enjoys both high\ncomputational efficiency and rigorous theoretical guarantee. Moreover, in the\nproposed regularization term, the topological pattern among networks is\ncharacterized by a Gram matrix, endowing our graph estimator with the ability\nof flexible modelling different types of topological patterns by different\nchoices of the Gram matrix. Computationally, the regularization term, coupling\nthe parameters together, makes the formulated optimization problem intractable\nand thus, we develop a computationally-scalable algorithm based on the\nalternating direction method of multipliers (ADMM) to solve it efficiently.\nTheoretically, we provide a theoretical analysis of the proposed graph\nestimator, which establishes a non-asymptotic bound of the estimation error\nunder the high-dimensional setting and reflects the effect of several key\nfactors on the convergence rate of our algorithm. Finally, the superior\nperformance of the proposed method is illustrated through simulated and real\ndata examples.\n"}
{"abstract": "  This paper presents a cross channel context model for latents in deep image\ncompression. Generally, deep image compression is based on an autoencoder\nframework, which transforms the original image to latents at the encoder and\nrecovers the reconstructed image from the quantized latents at the decoder. The\ntransform is usually combined with an entropy model, which estimates the\nprobability distribution of the quantized latents for arithmetic coding.\nCurrently, joint autoregressive and hierarchical prior entropy models are\nwidely adopted to capture both the global contexts from the hyper latents and\nthe local contexts from the quantized latent elements. For the local contexts,\nthe widely adopted 2D mask convolution can only capture the spatial context.\nHowever, we observe that there are strong correlations between different\nchannels in the latents. To utilize the cross channel correlations, we propose\nto divide the latents into several groups according to channel index and code\nthe groups one by one, where previously coded groups are utilized to provide\ncross channel context for the current group. The proposed cross channel context\nmodel is combined with the joint autoregressive and hierarchical prior entropy\nmodel. Experimental results show that, using PSNR as the distortion metric, the\ncombined model achieves BD-rate reductions of 6.30% and 6.31% over the baseline\nentropy model, and 2.50% and 2.20% over the latest video coding standard\nVersatile Video Coding (VVC) for the Kodak and CVPR CLIC2020 professional\ndataset, respectively. In addition, when optimized for the MS-SSIM metric, our\napproach generates visually more pleasant reconstructed images.\n"}
{"abstract": "  We study the generalized multidimensional bin packing problem (GVBP) that\ngeneralizes both geometric packing and vector packing. Here, we are given $n$\nrectangular items where the $i^{\\textrm{th}}$ item has width $w(i)$, height\n$h(i)$, and $d$ nonnegative weights $v_1(i), v_2(i), \\ldots, v_{d}(i)$. Our\ngoal is to get an axis-parallel non-overlapping packing of the items into\nsquare bins so that for all $j \\in [d]$, the sum of the $j^{\\textrm{th}}$\nweight of items in each bin is at most 1. This is a natural problem arising in\nlogistics, resource allocation, and scheduling. Despite being well studied in\npractice, surprisingly, approximation algorithms for this problem have rarely\nbeen explored.\n  We first obtain two simple algorithms for GVBP having asymptotic\napproximation ratios $6(d+1)$ and $3(1 + \\ln(d+1) + \\varepsilon)$. We then\nextend the Round-and-Approx (R&A) framework [Bansal-Khan, SODA'14] to wider\nclasses of algorithms, and show how it can be adapted to GVBP. Using more\nsophisticated techniques, we obtain better approximation algorithms for GVBP,\nand we get further improvement by combining them with the R&A framework. This\ngives us an asymptotic approximation ratio of $2(1+\\ln((d+4)/2))+\\varepsilon$\nfor GVBP, which improves to $2.919+\\varepsilon$ for the special case of $d=1$.\nWe obtain further improvement when the items are allowed to be rotated. We also\npresent algorithms for a generalization of GVBP where the items are high\ndimensional cuboids.\n"}
{"abstract": "  We introduce the COMBO-based approach for EUD parsing and its implementation,\nwhich took part in the IWPT 2021 EUD shared task. The goal of this task is to\nparse raw texts in 17 languages into Enhanced Universal Dependencies (EUD). The\nproposed approach uses COMBO to predict UD trees and EUD graphs. These\nstructures are then merged into the final EUD graphs. Some EUD edge labels are\nextended with case information using a single language-independent expansion\nrule. In the official evaluation, the solution ranked fourth, achieving an\naverage ELAS of 83.79%. The source code is available at\nhttps://gitlab.clarin-pl.eu/syntactic-tools/combo.\n"}
{"abstract": "  We study the mobility edges in a variety of one-dimensional tight binding\nmodels with slowly varying quasi-periodic disorders. It is found that the\nquasi-periodic disordered models can be approximated by an ensemble of periodic\nmodels. The mobility edges can be determined by the overlaps of the energy\nbands of these periodic models. We demonstrate that this method provides an\nefficient way to find out the precise location of mobility edge in\nqusi-periodic disordered models. Based on this approximate method, we also\npropose an index to indicate the degree of localization of each eigenstate.\n"}
{"abstract": "  The concept of linear set in projective spaces over finite fields was\nintroduced by Lunardon in 1999 and it plays central roles in the study of\nblocking sets, semifields, rank-distance codes and etc. A linear set with the\nlargest possible cardinality and rank is called maximum scattered. Despite two\ndecades of study, there are only a limited number of maximum scattered linear\nsets of a line $\\mathrm{PG}(1,q^n)$. In this paper, we provide a large family\nof new maximum scattered linear sets over $\\mathrm{PG}(1,q^n)$ for any even\n$n\\geq 6$ and odd $q$. In particular, the relevant family contains at least\n  \\[ \\begin{cases}\n  \\left\\lfloor\\frac{q^t+1}{8rt}\\right\\rfloor,& \\text{ if }t\\not\\equiv\n2\\pmod{4};\\\\[8pt]\n  \\left\\lfloor\\frac{q^t+1}{4rt(q^2+1)}\\right\\rfloor,& \\text{ if }t\\equiv\n2\\pmod{4},\n  \\end{cases} \\]\n  inequivalent members for given $q=p^r$ and $n=2t>8$, where\n$p=\\mathrm{char}(\\mathbb{F}_q)$. This is a great improvement of previous\nresults: for given $q$ and $n>8$, the number of inequivalent maximum scattered\nlinear sets of $\\mathrm{PG}(1,q^n)$ in all classes known so far, is smaller\nthan $q^2$. Moreover, we show that there are a large number of new maximum\nrank-distance codes arising from the constructed linear sets.\n"}
{"abstract": "  The advent of comprehensive synaptic wiring diagrams of large neural circuits\nhas created the field of connectomics and given rise to a number of open\nresearch questions. One such question is whether it is possible to reconstruct\nthe information stored in a recurrent network of neurons, given its synaptic\nconnectivity matrix. Here, we address this question by determining when solving\nsuch an inference problem is theoretically possible in specific attractor\nnetwork models and by providing a practical algorithm to do so. The algorithm\nbuilds on ideas from statistical physics to perform approximate Bayesian\ninference and is amenable to exact analysis. We study its performance on three\ndifferent models and explore the limitations of reconstructing stored patterns\nfrom synaptic connectivity.\n"}
{"abstract": "  Intestinal parasitic infection leads to several morbidities to humans\nworldwide, especially in tropical countries. The traditional diagnosis usually\nrelies on manual analysis from microscopic images which is prone to human error\ndue to morphological similarity of different parasitic eggs and abundance of\nimpurities in a sample. Many studies have developed automatic systems for\nparasite egg detection to reduce human workload. However, they work with high\nquality microscopes, which unfortunately remain unaffordable in some rural\nareas. Our work thus exploits a benefit of a low-cost USB microscope. This\ninstrument however provides poor quality of images due to limitation of\nmagnification (10x), causing difficulty in parasite detection and species\nclassification. In this paper, we propose a CNN-based technique using transfer\nlearning strategy to enhance the efficiency of automatic parasite\nclassification in poor-quality microscopic images. The patch-based technique\nwith sliding window is employed to search for location of the eggs. Two\nnetworks, AlexNet and ResNet50, are examined with a trade-off between\narchitecture size and classification performance. The results show that our\nproposed framework outperforms the state-of-the-art object recognition methods.\nOur system combined with final decision from an expert may improve the real\nfaecal examination with low-cost microscopes.\n"}
{"abstract": "  Minimizing cross-entropy over the softmax scores of a linear map composed\nwith a high-capacity encoder is arguably the most popular choice for training\nneural networks on supervised learning tasks. However, recent works show that\none can directly optimize the encoder instead, to obtain equally (or even more)\ndiscriminative representations via a supervised variant of a contrastive\nobjective. In this work, we address the question whether there are fundamental\ndifferences in the sought-for representation geometry in the output space of\nthe encoder at minimal loss. Specifically, we prove, under mild assumptions,\nthat both losses attain their minimum once the representations of each class\ncollapse to the vertices of a regular simplex, inscribed in a hypersphere. We\nprovide empirical evidence that this configuration is attained in practice and\nthat reaching a close-to-optimal state typically indicates good generalization\nperformance. Yet, the two losses show remarkably different optimization\nbehavior. The number of iterations required to perfectly fit to data scales\nsuperlinearly with the amount of randomly flipped labels for the supervised\ncontrastive loss. This is in contrast to the approximately linear scaling\npreviously reported for networks trained with cross-entropy.\n"}
{"abstract": "  Simultaneous trajectory prediction for multiple heterogeneous traffic\nparticipants is essential for the safe and efficient operation of connected\nautomated vehicles under complex driving situations in the real world. The\nmulti-agent prediction task is challenging, as the motions of traffic\nparticipants are affected by many factors, including their individual dynamics,\ntheir interactions with surrounding agents, the traffic infrastructures, and\nthe number and modalities of the target agents. To further advance the\ntrajectory prediction techniques, in this work we propose a three-channel\nframework together with a novel Heterogeneous Edge-enhanced graph ATtention\nnetwork (HEAT), which is able to deal with the heterogeneity of the target\nagents and traffic participants involved. Specifically, the agent's dynamics\nare extracted from their historical states using type-specific encoders. The\ninter-agent interactions are represented with a directed edge-featured\nheterogeneous graph, and then interaction features are extracted using the\nproposed HEAT network. Besides, the map features are shared across all agents\nby introducing a selective gate mechanism. And finally, the trajectories of\nmulti-agent are executed simultaneously. Validations using both urban and\nhighway driving datasets show that the proposed model can realize simultaneous\ntrajectory predictions for multiple agents under complex traffic situations,\nand achieve state-of-the-art performance with respect to prediction accuracy,\ndemonstrating its feasibility and effectiveness.\n"}
{"abstract": "  We show the existence of a natural Dirichlet-to-Neumann map on Riemannian\nmanifolds with boundary and bounded geometry, such that the bottom of the\nDirichlet spectrum is positive. This map regarded as a densely defined operator\nin the $L^2$-space of the boundary admits Friedrichs extension. We focus on the\nspectrum of this operator on covering spaces and total spaces of Riemannian\nprincipal bundles over compact manifolds.\n"}
{"abstract": "  This paper proceeds an approximate calculation of ultimate time survival\nprobability for bi-seasonal discrete time risk model when premium rate equals\ntwo. The same model with income rate equal to one was investigated in 2014 by\nDamarackas and \\v{S}iaulys. In general, discrete time and related risk models\ndeal with possibility for a certain version of random walk to hit a certain\nthreshold at least once in time. In this research, the mentioned threshold is\nthe line $u+2t$ and random walk consists from two interchangeably occurring\nindependent but not necessarily identically distributed random variables. Most\nof proved theoretical statements are illustrated via numerical calculations.\nAlso, there are raised a couple of conjectures on a certain recurrent\ndeterminants non-vanishing.\n"}
{"abstract": "  We study the orientifold of the ${\\mathcal{N}} = 1$ superconformal field\ntheories describing D3-branes probing the Suspended Pinch Point singularity, as\nwell as the orientifolds of non-chiral theories obtained by a specific orbifold\n$\\mathbb{Z}_n$ of SPP. We find that these models realize a mechanism analogous\nto the one recently found for the orientifold of the complex Calabi-Yau cone\nover the Pseudo del Pezzo surface PdP$_{3c}$: they all flow to a new IR fixed\npoint such that the value of the $a$-charge is less than half the one of the\noriented theory. We also find that the value of $a$ coincides with the charge\nof specific orientifolds of the toric singularities\n$L^{(\\bar{n},\\bar{n},\\bar{n})}$ with $\\bar{n}=3n/2$ for $n$ even or\n$L^{(\\bar{n},\\bar{n}+1,\\bar{n})}$ with $\\bar{n}=(3n{-}1)/2$ for $n$ odd,\nsuggesting the existence of an IR duality.\n"}
{"abstract": "  With the increase of variable renewable energy sources (VRES) share in\nelectricity systems, manystudies were developed in order to determine their\noptimal technological and spatial mix. Modern PortfolioTheory (MPT) has been\nfrequently applied in this context. However, some crucial aspects, important\ninenergy planning, are not addressed by these analyses. We, therefore, propose\nseveral improvements andevaluate how each change in formulation impacts\nresults. More specifically, we address generation costs, system demand, and\nfirm energy output, present a formal model and apply it to the case of Brazil.\nWefound that, after including our proposed modifications, the resulting\nefficient frontier differs strongly fromthe one obtained in the original\nformulation. Portfolios with high output standard deviation are not ableto\nprovide a firm output level at competitive costs. Furthermore, we show that\ndiversification plays animportant role in smoothing output from VRES portfolios\n"}
{"abstract": "  Standard methods for the meta-analysis of medical tests without a gold\nstandard are limited to dichotomous data. Multivariate probit models are used\nto analyze correlated binary data, and can be extended to multivariate ordered\nprobit models to model ordinal data. Within the context of an imperfect gold\nstandard, they have previously been used for the analysis of dichotomous and\nordinal tests in a single study, and for the meta-analysis of dichotomous\ntests. In this paper, we developed a hierarchical, latent class multivariate\nprobit model for the simultaneous meta-analysis of ordinal and dichotomous\ntests without assuming a gold standard. The model can accommodate a\nhierarchical partial pooling model on the conditional within-study\ncorrelations, enabling one to obtain summary estimates of joint test accuracy.\nDichotomous tests use probit regression likelihoods and ordinal tests use\nordered probit regression likelihoods. We fitted the models using Stan, which\nuses a state-of-the-art Hamiltonian Monte Carlo algorithm. We applied the\nmodels to a dataset in which studies evaluated the accuracy of tests, and test\ncombinations, for deep vein thrombosis. We first demonstrated the issues with\ndichotomising test accuracy data a priori without a gold standard by fitting\nmodels which dichotomised the ordinal test data, and then we applied models\nwhich do not dichotomise the data. Furthermore, we fitted and compared a\nvariety of other models, including those which assumed conditional independence\nand dependence between tests, and those assuming perfect and an imperfect gold\nstandard.\n"}
{"abstract": "  We present the first set of cosmological baryonic zoom-in simulations of\ngalaxies including dissipative self-interacting dark matter (dSIDM). These\nsimulations utilize the Feedback In Realistic Environments (FIRE-2) galaxy\nformation physics, but allow the dark matter to have dissipative\nself-interactions analogous to Standard Model forces, parameterized by the\nself-interaction cross-section per unit mass, $(\\sigma/m)$, and the\ndimensionless degree of dissipation, $0<f_{\\rm diss}<1$. We survey this\nparameter space, including constant and velocity-dependent cross-sections, and\nfocus on structural and kinematic properties of dwarf galaxies with $M_{\\rm\nhalo} \\simeq 10^{10-11} {\\rm M}_{\\odot}$. Central density profiles of simulated\ndwarfs become cuspy when $(\\sigma/m)_{\\rm eff} \\gtrsim 0.1\\,{\\rm\ncm^{2}\\,g^{-1}}$ (and $f_{\\rm diss}=0.5$ as fiducial). The power-law slopes\nasymptote to $\\alpha \\approx -1.5$ in low-mass dwarfs independent of\ncross-section, which arises from a dark matter \"cooling flow\". Through\ncomparisons with dark matter only simulations, we find the profile in this\nregime is insensitive to the inclusion of baryons. However, when\n$(\\sigma/m)_{\\rm eff} \\ll 0.1\\,{\\rm cm^{2}\\,g^{-1}}$, baryonic effects can\nproduce cored density profiles comparable to non-dissipative cold dark matter\n(CDM) runs but at smaller radii. Simulated galaxies with $(\\sigma/m) \\gtrsim\n10\\,{\\rm cm^{2}\\,g^{-1}}$ develop significant coherent rotation of dark matter,\naccompanied by halo deformation, but this is unlike the well-defined thin \"dark\ndisks\" often attributed to baryon-like dSIDM. The density profiles in this high\ncross-section model exhibit lower normalizations given the onset of halo\ndeformation. For our surveyed dSIDM parameters, halo masses and galaxy stellar\nmasses do not show appreciable difference from CDM, but dark matter kinematics\nand halo concentrations/shapes can differ.\n"}
{"abstract": "  The celebrated CLEAN algorithm has been the cornerstone of deconvolution\nalgorithms in radio interferometry almost since its conception in the 1970s.\nFor all its faults, CLEAN is remarkably fast, robust to calibration artefacts\nand in its ability to model point sources. We demonstrate how the same\nassumptions that afford CLEAN its speed can be used to accelerate more\nsophisticated deconvolution algorithms.\n"}
{"abstract": "  Generative adversarial networks (GANs) have shown impressive results in both\nunconditional and conditional image generation. In recent literature, it is\nshown that pre-trained GANs, on a different dataset, can be transferred to\nimprove the image generation from a small target data. The same, however, has\nnot been well-studied in the case of conditional GANs (cGANs), which provides\nnew opportunities for knowledge transfer compared to unconditional setup. In\nparticular, the new classes may borrow knowledge from the related old classes,\nor share knowledge among themselves to improve the training. This motivates us\nto study the problem of efficient conditional GAN transfer with knowledge\npropagation across classes. To address this problem, we introduce a new GAN\ntransfer method to explicitly propagate the knowledge from the old classes to\nthe new classes. The key idea is to enforce the popularly used conditional\nbatch normalization (BN) to learn the class-specific information of the new\nclasses from that of the old classes, with implicit knowledge sharing among the\nnew ones. This allows for an efficient knowledge propagation from the old\nclasses to the new ones, with the BN parameters increasing linearly with the\nnumber of new classes. The extensive evaluation demonstrates the clear\nsuperiority of the proposed method over state-of-the-art competitors for\nefficient conditional GAN transfer tasks. The code is available at:\nhttps://github.com/mshahbazi72/cGANTransfer\n"}
{"abstract": "  We analyze two different fermionic systems that defy Mott localization\nshowing a metallic ground state at integer filling and very large Coulomb\nrepulsion. The first is a multiorbital Hubbard model with a Hund's coupling,\nwhere this physics has been widely studied and the new metallic state is called\na Hund's metal, and the second is a SU(3) Hubbard model with a patterned\nsingle-particle potential designed to display a similar interaction-resistant\nmetal in a set-up which can be implemented with SU($N$) ultracold atoms. With\nsimple analytical arguments and exact numerical diagonalization of the\nHamiltonians for a minimal three-site system, we demonstrate that the\ninteraction-resistant metal emerges in both cases as a compromise between two\ndifferent insulating solutions which are stabilized by different terms of the\nmodels. This provides a strong evidence that the Hund's metal is a specific\nrealization of a more general phenomenon which can be realized in various\nstrongly correlated systems.\n"}
{"abstract": "  Motivated by recent experiments, we calculate particle emission from a\nBose-Einstein condensate trapped in a single deep well of a one-dimensional\nlattice when the interaction strength is modulated. In addition to pair\nemission, which has been widely studied, we observe single-particle emission.\nWithin linear response, we are able to write closed-form expressions for the\nsingle-particle emission rates and reduce the pair emission rates to\none-dimensional integrals. The full nonlinear theory of single-particle\nemission is reduced to a single variable integrodifferential equation, which we\nnumerically solve.\n"}
{"abstract": "  We show that, under finitely many ergodicity assumptions, any\nmulticorrelation sequence defined by invertible measure preserving\n$\\mathbb{Z}^d$-actions with multivariable integer polynomial iterates is the\nsum of a nilsequence and a null sequence, extending a recent result of the\nsecond author. To this end, we develop a new seminorm bound estimate for\nmultiple averages by improving the results in a previous work of the first,\nthird and fourth authors. We also use this approach to obtain new criteria for\njoint ergodicity of multiple averages with multivariable polynomial iterates on\n$\\mathbb{Z}^{d}$-systems.\n"}
{"abstract": "  We report promising results for high-throughput on-field soybean pod count\nwith small mobile robots and machine-vision algorithms. Our results show that\nthe machine-vision based soybean pod counts are strongly correlated with\nsoybean yield. While pod counts has a strong correlation with soybean yield,\npod counting is extremely labor intensive, and has been difficult to automate.\nOur results establish that an autonomous robot equipped with vision sensors can\nautonomously collect soybean data at maturity. Machine-vision algorithms can be\nused to estimate pod-counts across a large diversity panel planted across\nexperimental units (EUs, or plots) in a high-throughput, automated manner. We\nreport a correlation of 0.67 between our automated pod counts and soybean\nyield. The data was collected in an experiment consisting of 1463 single-row\nplots maintained by the University of Illinois soybean breeding program during\nthe 2020 growing season. We also report a correlation of 0.88 between automated\npod counts and manual pod counts over a smaller data set of 16 plots.\n"}
{"abstract": "  The paper presents a novel asymptotic distribution for a mle when the\nlog--likelihood is strictly concave in the parameter for all data points; for\nexample, the exponential family. The new asymptotic distribution can be seen as\na refinement of the usual normal asymptotic distribution and is comparable to\nan Edgeworth expansion. However, it is obtained with weaker conditions than\neven those for asymptotic normality. The same technique is then used to find\nthe exact distribution of the weighted likelihood bootstrap sampler.\n"}
{"abstract": "  Green communications have always been a target for the information industry\nto alleviate energy overhead and reduce fossil fuel usage. In current 5G and\nfuture 6G era, there is no doubt that the volume of network infrastructure and\nthe number of connected terminals will keep exponentially increasing, which\nresults in the surging energy cost. It becomes growing important and urgent to\ndrive the development of green communications. However, 6G will inevitably have\nincreasingly stringent and diversified requirements for Quality of Service\n(QoS), security, flexibility, and even intelligence, all of which challenge the\nimprovement of energy efficiency. Moreover, the dynamic energy harvesting\nprocess, which will be adopted widely in 6G, further complicates the power\ncontrol and network management. To address these challenges and reduce human\nintervene, Artificial Intelligence (AI) has been widely recognized and\nacknowledged as the only solution. Academia and industry have conducted\nextensive research to alleviate energy demand, improve energy efficiency, and\nmanage energy harvesting in various communication scenarios. In this paper, we\npresent the main considerations for green communications and survey the related\nresearch on AI-based green communications. We focus on how AI techniques are\nadopted to manage the network and improve energy harvesting toward the green\nera. We analyze how state-of-the-art Machine Learning (ML) and Deep Learning\n(DL) techniques can cooperate with conventional AI methods and mathematical\nmodels to reduce the algorithm complexity and optimize the accuracy rate to\naccelerate the applications in 6G. Finally, we discuss the existing problems\nand envision the challenges for these emerging techniques in 6G.\n"}
{"abstract": "  The paper studies a probabilistic notion of causes in Markov chains that\nrelies on the counterfactuality principle and the probability-raising property.\nThis notion is motivated by the use of causes for monitoring purposes where the\naim is to detect faulty or undesired behaviours before they actually occur. A\ncause is a set of finite executions of the system after which the probability\nof the effect exceeds a given threshold. We introduce multiple types of costs\nthat capture the consumption of resources from different perspectives, and\nstudy the complexity of computing cost-minimal causes.\n"}
{"abstract": "  The field of neuromorphic computing is in a period of active exploration.\nWhile many tools have been developed to simulate neuronal dynamics or convert\ndeep networks to spiking models, general software libraries for learning rules\nremain underexplored. This is partly due to the diverse, challenging nature of\nefforts to design new learning rules, which range from encoding methods to\ngradient approximations, from population approaches that mimic the Bayesian\nbrain to constrained learning algorithms deployed on memristor crossbars. To\naddress this gap, we present Neko, a modular, extensible library with a focus\non aiding the design of new learning algorithms. We demonstrate the utility of\nNeko in three exemplar cases: online local learning, probabilistic learning,\nand analog on-device learning. Our results show that Neko can replicate the\nstate-of-the-art algorithms and, in one case, lead to significant\noutperformance in accuracy and speed. Further, it offers tools including\ngradient comparison that can help develop new algorithmic variants. Neko is an\nopen source Python library that supports PyTorch and TensorFlow backends.\n"}
{"abstract": "  How much credit (or blame) should an action taken in a state get for a future\nreward? This is the fundamental temporal credit assignment problem in\nReinforcement Learning (RL). One of the earliest and still most widely used\nheuristics is to assign this credit based on a scalar coefficient $\\lambda$\n(treated as a hyperparameter) raised to the power of the time interval between\nthe state-action and the reward. In this empirical paper, we explore heuristics\nbased on more general pairwise weightings that are functions of the state in\nwhich the action was taken, the state at the time of the reward, as well as the\ntime interval between the two. Of course it isn't clear what these pairwise\nweight functions should be, and because they are too complex to be treated as\nhyperparameters we develop a metagradient procedure for learning these weight\nfunctions during the usual RL training of a policy. Our empirical work shows\nthat it is often possible to learn these pairwise weight functions during\nlearning of the policy to achieve better performance than competing approaches.\n"}
{"abstract": "  1.) Spatio-temporal datasets that are difficult to analyze are common in\necological surveys. There are software packages available to analyze these\ndatasets, but many of them require advanced coding skills. There is a growing\nneed for easy to use packages that researchers can use to analyze common\necological datasets 2.) We develop a particular generalized linear mixed model\nfor spatio-temporal point-referenced data that is flexible enough to\naccommodate data from most ecological surveys while being structured enough to\nfacilitate analyses without advanced coding. Our implementation in the staRVe\npackage uses a computationally efficient version of a nearest neighbour\nGaussian process enabling analysis of relatively large datasets. 3.) A brief\nsimulation study shows our model produces accurate predictions and forecasts,\nwhile a tutorial analysis of a Carolina wren survey suggests a recommended\nworkflow for analyses. We also analyze a more complicated scientific survey of\nhaddock to showcase the capabilities of our model. 4.) Our model and package\nare tools that can easily be added to researchers' workflow to help make sense\nof data from ecological surveys. We emphasize the ability of our model to\ncreate useful visualisations of data which can then lead to identification of\nimportant trends in species distributions.\n"}
{"abstract": "  A collisional-radiative non-equilibrium impurity treatment for JOREK 3D\nnonlinear magneto-hydrodynamic (MHD) simulations has been developed. The\nimpurities are represented by super-particles flowing along the fluid velocity\nfield lines, while ionizing and recombining independently according to ADAS\ndata and local fluid density and temperature. The non-equilibrium impurity\ncontributions are then projected back to the fluid field for self-consistent\ntime evolution. A 2D test case is used to compare the new non-equilibrium\nimpurity model against previous Coronal Equilibrium (CE) impurity treatment, as\nwell as to compare the non-equilibrium impurity behavior between the single and\nthe two temperature model. Further, we conduct benchmark with previously\npublished coronal non-equilibrium results by other 3D nonlinear MHD codes such\nas M3D-C1 and NIMROD. The new non-equilibrium treatment is shown to\nsuccessfully capture the early phase cooling by weakly ionized impurities which\nthe CE model missed. The benchmarks with M3D-C1 and NIMROD show general\nagreement in both the integrated quantities and the 2D profile evolution,\ndespite the difference in the atomic model used. The above comparison and\nbenchmark cases demonstrate the capability of the non-equilibrium impurity\nmodel for JOREK, paving the way for more sophisticated 3D non-linear Massive\nMaterial Injection (MMI) simulations which have important applications in\ndisruption mitigation studies.\n"}
{"abstract": "  Sharing data from clinical studies can facilitate innovative data-driven\nresearch and ultimately lead to better public health. However, sharing\nbiomedical data can put sensitive personal information at risk. This is usually\nsolved by anonymization, which is a slow and expensive process. An alternative\nto anonymization is sharing a synthetic dataset that bears a behaviour similar\nto the real data but preserves privacy. As part of the collaboration between\nNovartis and the Oxford Big Data Institute, we generate a synthetic dataset\nbased on COSENTYX (secukinumab) Ankylosing Spondylitis clinical study. We apply\nan Auxiliary Classifier GAN to generate synthetic MRIs of vertebral units. The\nimages are conditioned on the VU location (cervical, thoracic and lumbar). In\nthis paper, we present a method for generating a synthetic dataset and conduct\nan in-depth analysis on its properties along three key metrics: image fidelity,\nsample diversity and dataset privacy.\n"}
{"abstract": "  In this paper we propose two models describing the dynamics of heavy and\nlight vehicles on a road network, taking into account the interactions between\nthe two classes. The models are tailored for two-lane highways where heavy\nvehicles cannot overtake. This means that heavy vehicles cannot saturate the\nwhole road space, while light vehicles can. In these conditions the creeping\nphenomenon can appear, i.e. one class of vehicles can proceed even if the other\nclass has reached the maximal density. The first model we propose couples two\nfirst-order macroscopic LWR models, while the second model couples a\nsecond-order microscopic Follow-the-Leader model with a first-order macroscopic\nLWR model. Numerical results show that both models are able to catch some\nsecond-order (inertial) phenomena like stop & go waves. Models are calibrated\nby means of real data measured by fixed sensors placed along the A4 Italian\nhighway Trieste-Venice and its branches, provided by Autovie Venete S.p.A.\n"}
{"abstract": "  In this paper we present DMC, a model and associated tool for polarimetric\nimaging of very long baseline interferometry datasets that simultaneously\nreconstructs the full-Stokes emission structure along with the station-based\ngain and leakage calibration terms. DMC formulates the imaging problem in terms\nof posterior exploration, which is achieved using Hamiltonian Monte Carlo\nsampling. The resulting posterior distribution provides a natural\nquantification of uncertainty in both the image structure and in the data\ncalibration. We run DMC on both synthetic and real datasets, the results of\nwhich demonstrate its ability to accurately recover both the image structure\nand calibration quantities as well as to assess their corresponding\nuncertainties. The framework underpinning DMC is flexible, and its specific\nimplementation is under continued development.\n"}
{"abstract": "  We study the general embedding of a $ P(X, \\varphi) $ inflationary theory\ninto a two-field theory with curved field space metric, which was proposed as a\npossible way to examine the relation between de Sitter Swampland conjecture and\n\\textit{k}-inflation. We show that this embedding method fits into the special\ntype of two-field model in which the heavy field can be integrated out at the\nfull action level. However, this embedding is not exact due to the upper bound\nof the effective mass of the heavy field. We quantify the deviation between the\nspeed of sound calculated via the $ P(X, \\varphi) $ theory and the embedding\ntwo-field picture to next leading order terms. We especially focus on the first\npotential slow roll parameter defined in the two-field picture and obtain an\nupper bound on it.\n"}
{"abstract": "  We construct solutions of the fast diffusion equation, which exist for all\n$t\\in\\mathbb{R}$ and are singular on the set $\\Gamma(t):= \\{ \\xi(s) ; -\\infty\n<s \\leq ct \\}$, $c>0$, where $\\xi\\in C^3(\\mathbb{R};\\mathbb{R}^n)$, $n\\geq 2$.\nWe also give a precise description of the behavior of the solutions near\n$\\Gamma(t)$.\n"}
{"abstract": "  We present a compact formula in Mellin space for the four-point tree-level\nholographic correlators of chiral primary operators of arbitrary conformal\nweights in $(2,0)$ supergravity on $AdS_3 \\times S^3$, with two operators in\ntensor multiplet and the other two in gravity multiplet. This is achieved by\nsolving the recursion relation arising from a hidden six-dimensional conformal\nsymmetry. We note the compact expression is obtained after carefully analysing\nthe analytic structures of the correlators. Various limits of the correlators\nare studied, including the maximally R-symmetry violating limit and flat-space\nlimit.\n"}
{"abstract": "  Destructive quantum interference (DQI) and its effects on electron transport\nis studied in chemical molecules and finite physical lattices that can be\ndescribed by a discrete Hamiltonian. Starting from a bipartite system whose\nconductance zeros are known to exist between any two points of a specially\ndesignated set, the interference set, we use the Dyson equation to develop a\ngeneral algorithm of determining the zero conductance points in complex\nsystems, which are not necessarily bipartite. We illustrate this procedure as\nit applies to the fulvene molecule. The stability of the conductance zeros is\nanalyzed in respect with external perturbations.\n"}
{"abstract": "  We conduct an audit of pricing algorithms employed by companies in the\nItalian car insurance industry, primarily by gathering quotes through a popular\ncomparison website. While acknowledging the complexity of the industry, we find\nevidence of several problematic practices. We show that birthplace and gender\nhave a direct and sizeable impact on the prices quoted to drivers, despite\nnational and international regulations against their use. Birthplace, in\nparticular, is used quite frequently to the disadvantage of foreign-born\ndrivers and drivers born in certain Italian cities. In extreme cases, a driver\nborn in Laos may be charged 1,000 euros more than a driver born in Milan, all\nelse being equal. For a subset of our sample, we collect quotes directly on a\ncompany website, where the direct influence of gender and birthplace is\nconfirmed. Finally, we find that drivers with riskier profiles tend to see\nfewer quotes in the aggregator result pages, substantiating concerns of\ndifferential treatment raised in the past by Italian insurance regulators.\n"}
{"abstract": "  The growth of crystals confined in porous or cellular materials is ubiquitous\nin Nature and industry. Confinement affects the formation of biominerals in\nliving organisms, of minerals in the Earth's crust and of salt crystals\ndamaging porous limestone monuments, and is also used to control the growth of\nartificial crystals. However, the mechanisms by which confinement alters\ncrystal shapes and growth rates are still not elucidated. Based on novel\n\\textit{in situ} optical observations of (001) surfaces of NaClO$_3$ and\nCaCO$_3$ crystals at nanometric distances from a glass substrate, we\ndemonstrate that new molecular layers can nucleate homogeneously and propagate\nwithout interruption even when in contact with other solids, raising the\nmacroscopic crystal above them. Confined growth is governed by the peculiar\ndynamics of these molecular layers controlled by the two-dimensional transport\nof mass through the liquid film from the edges to the center of the contact,\nwith distinctive features such as skewed dislocation spirals, kinetic\nlocalization of nucleation in the vicinity of the contact edge, and directed\ninstabilities. Confined growth morphologies can be predicted from the values of\nthree main dimensionless parameters.\n"}
{"abstract": "  With the increasing amount of reliance on digital data and computer networks\nby corporations and the public in general, the occurrence of cyber attacks has\nbecome a great threat to the normal functioning of our society. Intrusion\ndetection systems seek to address this threat by preemptively detecting attacks\nin real time while attempting to block them or minimizing their damage. These\nsystems can function in many ways being some of them based on artificial\nintelligence methods. Datasets containing both normal network traffic and cyber\nattacks are used for training these algorithms so that they can learn the\nunderlying patterns of network-based data. The CIDDS-001 is one of the most\nused datasets for network-based intrusion detection research. Regarding this\ndataset, in the majority of works published so far, the Class label was used\nfor training machine learning algorithms. However, there is another label in\nthe CIDDS-001, AttackType, that seems very promising for this purpose and\nremains considerably unexplored. This work seeks to make a comparison between\ntwo machine learning models, K-Nearest Neighbours and Random Forest, which were\ntrained with both these labels in order to ascertain whether AttackType can\nproduce reliable results in comparison with the Class label.\n"}
{"abstract": "  For the wave equation $\\partial_t^2-\\Delta+V$ on $\\mathbb{R}^d$ with\ncompactly supported, real valued potential $V$, we establish a sharp relation\nbetween Sobolev regularity of $V$ and the existence of finite order expansions\nas $t\\rightarrow 0$ for the relative trace of the wave group.\n"}
{"abstract": "  The experimental verification of quantum fluctuation relations for driven\nopen quantum system is currently a challenge, due to the conceptual and\noperative difficulty of distinguishing work and heat. The Nitrogen-Vacancy\ncenter in diamond has been recently proposed as a controlled test bed to study\nfluctuation relations in the presence of an engineered dissipative channel, in\nabsence of work [Hern\\'andez-G\\'omez et al., Phys. Rev. Research 2, 023327\n(2020)]. Here, we extend those studies to exploring the validity of quantum\nfluctuation relations in a driven-dissipative scenario, where the spin\nexchanges energy both with its surroundings because of a thermal gradient, and\nwith an external work source. We experimentally prove the validity of the\nquantum fluctuation relations in the presence of cyclic driving in two cases,\nwhen the spin exchanges energy with an effective infinite-temperature\nreservoir, and when the total work vanishes at stroboscopic times -- although\nthe power delivered to the NV center is non-null. Our results represent the\nfirst experimental study of quantum fluctuation relation in driven open quantum\nsystems.\n"}
{"abstract": "  This work concerns with the description of the topological phases of band\ninsulators of class DIII by using the equivariant cohomology. The main result\nis the definition of a cohomology class for general systems of class DIII which\ngeneralizes the well known $\\mathbb{Z}_2$-invariant given by the Teo-Kane\nformula in the one-dimension case. In the two-dimensional case this cohomology\ninvariant allows a complete description of the strong and weak phases. The\nrelation with the KR-theory, the Noether-Fredholm index and the classification\nof \"Real\" gerbes are also discussed.\n"}
{"abstract": "  Safety in human-robot interaction can be divided into physical safety and\nperceived safety, where the latter is still under-addressed in the literature.\nInvestigating perceived safety in human-robot interaction requires a\nmultidisciplinary perspective. Indeed, perceived safety is often considered as\nbeing associated with several common factors studied in other disciplines,\ni.e., comfort, predictability, sense of control, and trust. In this paper, we\ninvestigated the relationship between these factors and perceived safety in\nhuman-robot interaction using subjective and objective measures. We conducted a\ntwo-by-five mixed-subjects design experiment. The five within-subjects\nconditions correspond to (1) baseline, and the manipulations of robot behaviors\nto stimulate: (2) discomfort, (3) decreased perceived safety, (4) decreased\nsense of control and (5) distrust. Twenty-seven young adult participants took\npart in the experiments. Participants were asked to answer questionnaires that\nmeasure the manipulated factors after within-subjects conditions. Besides\nquestionnaire data, we collected objective measures such as videos and\nphysiological data. The questionnaire results show a correlation between\ncomfort, sense of control, trust, and perceived safety. We also discuss the\neffect of individual human characteristics (such as personality and gender)\nthat they could be predictors of perceived safety. We used the physiological\nsignal data and facial affect from videos for estimating perceived safety where\nparticipants' subjective ratings were utilized as labels. The data from\nobjective measures revealed that the prediction rate was higher from\nphysiological signal data. This paper can play an important role in the goal of\nbetter understanding perceived safety in human-robot interaction.\n"}
{"abstract": "  Single-phase multiferroic materials that allow the coexistence of\nferroelectric and magnetic ordering above room temperature are highly\ndesirable, motivating an ongoing search for mechanisms for unconventional\nferroelectricity in magnetic oxides. Here, we report an antisite defect\nmechanism for room temperature ferroelectricity in epitaxial thin films of\nyttrium orthoferrite, YFeO3, a perovskite-structured canted antiferromagnet. A\ncombination of piezoresponse force microscopy, atomically resolved elemental\nmapping with aberration corrected scanning transmission electron microscopy and\ndensity functional theory calculations reveals that the presence of YFe\nantisite defects facilitates a non-centrosymmetric distortion promoting\nferroelectricity. This mechanism is predicted to work analogously for other\nrare earth orthoferrites, with a dependence of the polarization on the radius\nof the rare earth cation. Furthermore, a vertically aligned nanocomposite\nconsisting of pillars of a magnetoelastic oxide CoFe2O4 embedded epitaxially in\nthe YFeO3 matrix exhibits both robust ferroelectricity and ferrimagnetism at\nroom temperature, as well as a noticeable strain-mediated magnetoelectric\ncoupling effect. Our work uncovers the distinctive role of antisite defects in\nproviding a novel mechanism for ferroelectricity in a range of magnetic\northoferrites and further augments the functionality of this family of complex\noxides for multiferroic applications.\n"}
{"abstract": "  Patients with Frontotemporal Dementia (FTD) have impaired cognitive\nabilities, executive and behavioral traits, loss of language ability, and\ndecreased memory capabilities. Based on the distinct patterns of cortical\natrophy and symptoms, the FTD spectrum primarily includes three variants:\nbehavioral variant FTD (bvFTD), non-fluent variant primary progressive aphasia\n(nfvPPA), and semantic variant primary progressive aphasia (svPPA). The purpose\nof this study is to classify MRI images of every single subject into one of the\nspectrums of the FTD in a hierarchical order by applying data-driven techniques\nof Artificial Intelligence (AI) on cortical thickness data. This data is\ncomputed by FreeSurfer software. We used the Smallest Univalue Segment\nAssimilating Nucleus (SUSAN) technique to minimize the noise in cortical\nthickness data. Specifically, we took 204 subjects from the frontotemporal\nlobar degeneration neuroimaging initiative (NIFTD) database to validate this\napproach, and each subject was diagnosed in one of the diagnostic categories\n(bvFTD, svPPA, nfvPPA and cognitively normal). Our proposed automated\nclassification model yielded classification accuracy of 86.5, 76, and 72.7 with\nsupport vector machine (SVM), linear discriminant analysis (LDA), and Naive\nBayes methods, respectively, in 10-fold cross-validation analysis, which is a\nsignificant improvement on a traditional single multi-class model with an\naccuracy of 82.7, 73.4, and 69.2.\n"}
{"abstract": "  Long-term complex activity recognition and localisation can be crucial for\nthe decision-making process of several autonomous systems, such as smart cars\nand surgical robots. Nonetheless, most current methods are designed to merely\nlocalise short-term action/activities or combinations of atomic actions that\nonly last for a few frames or seconds. In this paper, we address the problem of\nlong-term complex activity detection via a novel deformable, spatiotemporal\nparts-based model. Our framework consists of three main building blocks: (i)\naction tube detection, (ii) the modelling of the deformable geometry of parts,\nand (iii) a sparsity mechanism. Firstly, action tubes are detected in a series\nof snippets using an action tube detector. Next, a new 3D deformable RoI\npooling layer is designed for learning the flexible, deformable geometry of the\nconstellation of parts. Finally, a sparsity strategy differentiates between\nactivated and deactivate features. We also provide temporal complex activity\nannotation for the recently released ROAD autonomous driving dataset and the\nSARAS-ESAD surgical action dataset, to validate our method and show the\nadaptability of our framework to different domains. As they both contain long\nvideos portraying long-term activities they can be used as benchmarks for\nfuture work in this area.\n"}
{"abstract": "  Deep-learning-based image processing has emerged as a valuable tool in recent\nyears owing to its high performance. However, the quality of\ndeep-learning-based methods relies heavily on the amount of training data, and\nthe cost of acquiring a large amount of data is often prohibitive in medical\nfields. Therefore, we performed CT modality conversion based on deep learning\nrequiring only a small number of unsupervised images. The proposed method is\nbased on generative adversarial networks (GANs) with several extensions\ntailored for CT images. This method emphasizes the preservation of the\nstructure in the processed images and reduction in the amount of training data.\nThis method was applied to realize the conversion of mega-voltage computed\ntomography (MVCT) to kilo-voltage computed tomography (kVCT) images. Training\nwas performed using several datasets acquired from patients with head and neck\ncancer. The size of the datasets ranged from 16 slices (for two patients) to\n2745 slices (for 137 patients) of MVCT and 2824 slices of kVCT for 98 patients.\nThe quality of the processed MVCT images was considerably enhanced, and the\nstructural changes in the images were minimized. With an increase in the size\nof training data, the image quality exhibited a satisfactory convergence from a\nfew hundred slices. In addition to statistical and visual evaluations, these\nresults were clinically evaluated by medical doctors in terms of the accuracy\nof contouring. We developed an MVCT to kVCT conversion model based on deep\nlearning, which can be trained using a few hundred unpaired images. The\nstability of the model against the change in the data size was demonstrated.\nThis research promotes the reliable use of deep learning in clinical medicine\nby partially answering the commonly asked questions: \"Is our data enough? How\nmuch data must we prepare?\"\n"}
{"abstract": "  Optimal transport (OT) has recently found widespread interest in machine\nlearning. It allows to define novel distances between probability measures,\nwhich have shown promise in several applications. In this work, we discuss how\nto computationally approach general non-linear OT problems within the framework\nof Riemannian manifold optimization. The basis of this is the manifold of\ndoubly stochastic matrices (and their generalization). Even though the manifold\ngeometry is not new, surprisingly, its usefulness for solving general\nnon-linear OT problems has not been popular. To this end, we specifically\ndiscuss optimization-related ingredients that allow modeling the OT problem on\nsmooth Riemannian manifolds by exploiting the geometry of the search space. We\nalso discuss extensions where we reuse the developed optimization ingredients.\nWe make available the Manifold optimization-based Optimal Transport, or MOT,\nrepository with codes useful in solving OT problems in Python and Matlab. The\ncodes are available at \\url{https://github.com/SatyadevNtv/MOT}.\n"}
{"abstract": "  Light diffraction at an aperture is a basic problem that has generated a\ntremendous amount of interest in optics. Some of the most significant\ndiffraction results are the Fresnel-Kirchhoff and Rayleigh-Sommerfeld formulas.\nThese theories are based on solving the wave equation using Green's theorem and\nresult in slightly different expressions depending on the particular boundary\nconditions employed. In this paper, we propose another approach for solving the\ndiffraction by a thin screen, which includes apertures, gratings,\ntransparencies etc. We show that, applying the first order Born approximation\nto 2D objects, we obtain a general diffraction formula, without angular\napproximations. We discuss several common approximations and place our results\nin the context of existing theories.\n"}
{"abstract": "  Decoder diversity is a powerful error correction framework in which a\ncollection of decoders collaboratively correct a set of error patterns\notherwise uncorrectable by any individual decoder. In this paper, we propose a\nnew approach to design the decoder diversity of finite alphabet iterative\ndecoders (FAIDs) for Low-Density Parity Check (LDPC) codes over the binary\nsymmetric channel (BSC), for the purpose of lowering the error floor while\nguaranteeing the waterfall performance. The proposed decoder diversity is\nachieved by training a recurrent quantized neural network (RQNN) to\nlearn/design FAIDs. We demonstrated for the first time that a machine-learned\ndecoder can surpass in performance a man-made decoder of the same complexity.\nAs RQNNs can model a broad class of FAIDs, they are capable of learning an\narbitrary FAID. To provide sufficient knowledge of the error floor to the RQNN,\nthe training sets are constructed by sampling from the set of most problematic\nerror patterns - trapping sets. In contrast to the existing methods that use\nthe cross-entropy function as the loss function, we introduce a\nframe-error-rate (FER) based loss function to train the RQNN with the objective\nof correcting specific error patterns rather than reducing the bit error rate\n(BER). The examples and simulation results show that the RQNN-aided decoder\ndiversity increases the error correction capability of LDPC codes and lowers\nthe error floor.\n"}
{"abstract": "  The freeze-in production of Feebly Interacting Massive Particle (FIMP) dark\nmatter in the early universe is an appealing alternative to the well-known -\nand constrained - Weakly Interacting Massive Particle (WIMP) paradigm. Although\nchallenging, the phenomenology of FIMP dark matter has been receiving growing\nattention and is possible in a few scenarios. In this work, we contribute to\nthis endeavor by considering a $Z^\\prime$ portal to fermionic dark matter, with\nthe $Z^\\prime$ having both vector and axial couplings and a mass ranging from\nMeV up to PeV. We evaluate the bounds on both freeze-in and freeze-out from\ndirect detection, atomic parity violation, leptonic anomalous magnetic moments,\nneutrino-electron scattering, collider, and beam dump experiments. We show that\nFIMPs can already be tested by most of these experiments in a complementary\nway, whereas WIMPs are especially viable in the $Z^\\prime$ low mass regime, in\naddition to the $Z^\\prime$ resonance region. We also discuss the role of the\naxial couplings of $Z^\\prime$ in our results. We therefore hope to motivate\nspecific realizations of this model in the context of FIMPs, as well as\nsearches for these elusive dark matter candidates.\n"}
{"abstract": "  Smooth periodic travelling waves in the Camassa--Holm (CH) equation are\nrevisited. We show that these periodic waves can be characterized in two\ndifferent ways by using two different Hamiltonian structures. The standard\nformulation, common to the Korteweg--de Vries (KdV) equation, has several\ndisadvantages, e.g., the period function is not monotone and the quadratic\nenergy form may have two rather than one negative eigenvalues. We explore the\nnonstandard formulation and prove that the period function is monotone and the\nquadratic energy form has only one simple negative eigenvalue. We deduce a\nprecise condition for the spectral and orbital stability of the smooth periodic\ntravelling waves and show numerically that this condition is satisfied in the\nopen region where the smooth periodic waves exist.\n"}
{"abstract": "  Given the great success of Deep Neural Networks(DNNs) and the black-box\nnature of it,the interpretability of these models becomes an important\nissue.The majority of previous research works on the post-hoc interpretation of\na trained model.But recently, adversarial training shows that it is possible\nfor a model to have an interpretable input-gradient through\ntraining.However,adversarial training lacks efficiency for interpretability.To\nresolve this problem, we construct an approximation of the adversarial\nperturbations and discover a connection between adversarial training and\namplitude modulation. Based on a digital analogy,we propose noise modulation as\nan efficient and model-agnostic alternative to train a model that interprets\nitself with input-gradients.Experiment results show that noise modulation can\neffectively increase the interpretability of input-gradients model-agnosticly.\n"}
{"abstract": "  The Gamma Factory is a proposal to back-scatter laser photons off a beam of\npartially-stripped ions at the LHC, producing a beam of $\\sim 10$ MeV to $1$\nGeV photons with intensities of $10^{16}$ to $10^{18}~\\text{s}^{-1}$. This\nimplies $\\sim 10^{23}$ to $10^{25}$ photons on target per year, many orders of\nmagnitude greater than existing accelerator light sources and also far greater\nthan all current and planned electron and proton fixed target experiments. We\ndetermine the Gamma Factory's discovery potential through \"dark Compton\nscattering,\" $\\gamma e \\to e X$, where $X$ is a new, weakly-interacting\nparticle. For dark photons and other new gauge bosons with masses in the\n1~to~100 MeV range, the Gamma Factory has the potential to discover extremely\nweakly-interacting particles with just a few hours of data and will probe\ncouplings as low as $\\sim 10^{-9}$ with a year of running. The Gamma Factory\ntherefore may probe couplings lower than all other terrestrial experiments and\nis highly complementary to astrophysical probes. We outline the requirements of\nan experiment to realize this potential and determine the sensitivity reach for\nvarious experimental configurations.\n"}
{"abstract": "  Real-world classification domains, such as medicine, health and safety, and\nfinance, often exhibit imbalanced class priors and have asynchronous\nmisclassification costs. In such cases, the classification model must achieve a\nhigh recall without significantly impacting precision. Resampling the training\ndata is the standard approach to improving classification performance on\nimbalanced binary data. However, the state-of-the-art methods ignore the local\njoint distribution of the data or correct it as a post-processing step. This\ncan causes sub-optimal shifts in the training distribution, particularly when\nthe target data distribution is complex. In this paper, we propose Radial-Based\nCombined Cleaning and Resampling (RB-CCR). RB-CCR utilizes the concept of class\npotential to refine the energy-based resampling approach of CCR. In particular,\nRB-CCR exploits the class potential to accurately locate sub-regions of the\ndata-space for synthetic oversampling. The category sub-region for oversampling\ncan be specified as an input parameter to meet domain-specific needs or be\nautomatically selected via cross-validation. Our $5\\times2$ cross-validated\nresults on 57 benchmark binary datasets with 9 classifiers show that RB-CCR\nachieves a better precision-recall trade-off than CCR and generally\nout-performs the state-of-the-art resampling methods in terms of AUC and\nG-mean.\n"}
{"abstract": "  We describe in this short note a technique to convert an implicit surface\ninto a Signed Distance Function (SDF) while exactly preserving the zero\nlevel-set of the implicit. The proposed approach relies on embedding the input\nimplicit in the final layer of a neural network, which is trained to minimize a\nloss function characterizing the SDF.\n"}
{"abstract": "  We develop a second order correction to commonly used density functional\napproximations (DFA) to eliminate the systematic delocalization error. The\nmethod, based on the previously developed global scaling correction (GSC), is\nan exact quadratic correction to the DFA for the fractional charge behavior and\nuses the analytical second derivatives of the total energy with respect to\nfractional occupation numbers of the canonical molecular orbitals. For small\nand medium-size molecules, this correction leads to ground-state orbital\nenergies that are highly accurate approximation to the corresponding\nquasiparticle energies. It provides excellent predictions of ionization\npotentials, electron affinities, photoemission spectrum and photoexcitation\nenergies beyond previous approximate second order approaches, thus showing\npotential for broad applications in computational spectroscopy.\n"}
{"abstract": "  This paper does not describe a working system. Instead, it presents a single\nidea about representation which allows advances made by several different\ngroups to be combined into an imaginary system called GLOM. The advances\ninclude transformers, neural fields, contrastive representation learning,\ndistillation and capsules. GLOM answers the question: How can a neural network\nwith a fixed architecture parse an image into a part-whole hierarchy which has\na different structure for each image? The idea is simply to use islands of\nidentical vectors to represent the nodes in the parse tree. If GLOM can be made\nto work, it should significantly improve the interpretability of the\nrepresentations produced by transformer-like systems when applied to vision or\nlanguage\n"}
{"abstract": "  The structural flexibility at three substitution sites in LaFeAsO enabled\ninvestigation of the relation between superconductivity and structural\nparameters over a wide range of crystal compositions. Substitutions of Nd for\nLa, Sb or P for As, and F or H for O were performed. All these substitutions\nmodify the local structural parameters, while the F/H-substitution also changes\nband filling. It was found that the superconducting transition temperature\n$T_{c}$ is strongly affected by the pnictogen height $h_{Pn}$ from the Fe-plane\nthat controls the electron correlation strength and the size of the $d_{xy}$\nhole Fermi surface (FS). With increasing $h_{Pn}$, weak coupling\nsuperconductivity switches to the strong coupling one where the $d_{xy}$ hole\nFS is crucially important.\n"}
{"abstract": "  The aim of the paper is to recover some results of\nCardaliaguet-Nolen-Souganidis in \\cite{CNS} and Xin-Yu in \\cite{XY} about the\nhomogenization of the G--equation, using different and simpler techniques. The\nmain mathematical issue is the lack of coercivity of the Hamiltonians. In our\napproach we consider a multivalued dynamics without periodic invariants sets, a\nfamily of intrinsic distances and perform an approximation by a sequence of\ncoercive Hamiltonians.\n"}
{"abstract": "  Primordial black hole (PBH) is a fascinating candidate for the origin of\nbinary merger events observed by LIGO-Virgo collaboration. The spatial\ndistribution of PBHs at formation is an important feature to estimate the\nmerger rate. We investigate the clustering of PBHs formed by Affleck-Dine (AD)\nbaryogenesis, where dense baryon bubbles collapse to form PBHs. We found that\nformed PBHs show a strong clustering due to the stochastic dynamics of the AD\nfield. Including the clustering, we evaluate the merger rate and isocurvature\nperturbations of PBHs, which show significant deviations from those without\nclustering.\n"}
{"abstract": "  The ratios $R_{\\tau/P}\\equiv \\Gamma(\\tau \\to P \\nu_\\tau [\\gamma]) / \\Gamma(P\n\\to\\mu \\nu_\\mu[\\gamma])$ ($P=\\pi, K$) provide sensitive tests of lepton\nuniversality $\\left|g_\\tau/g_\\mu\\right|=1$ and are a useful tool for new\nphysics searches. The radiative corrections to $R_{\\tau/P}$ are computed\nfollowing a large-$N_C$ expansion to deal with hadronic effects: Chiral\nPerturbation Theory is enlarged by including the lightest multiplets of\nspin-one heavy states such that the relevant Green functions are well-behaved\nat high energies. We find $\\delta R_{\\tau/\\pi}=(0.18\\pm 0.57 )\\%$ and $\\delta\nR_{\\tau/K}=(0.97\\pm 0.58 )\\%$, which imply\n$\\left|g_\\tau/g_\\mu\\right|_\\pi=0.9964\\pm 0.0038$ and\n$\\left|g_\\tau/g_\\mu\\right|_K=0.9857\\pm 0.0078$, compatible with and at\n$1.8\\sigma$ of lepton universality, respectively. We test unitarity and bind\nnon-standard effective interactions with the $\\tau \\to P \\nu_\\tau [\\gamma]$\ndecays.\n"}
{"abstract": "  The development of four-dimensional (4D) scanning transmission electron\nmicroscopy (STEM) using fast detectors has opened-up new avenues for addressing\nsome of long-standing challenges in electron imaging. One of these challenges\nis how to image severely distorted crystal lattices, such as at a dislocation\ncore. Here we introduce a new 4D-STEM technique, called Cepstral STEM, for\nimaging disordered crystals using electron diffuse scattering. Local\nfluctuations of diffuse scattering are captured by scanning electron\nnanodiffraction (SEND) using a coherent probe. The harmonic signals in electron\ndiffuse scattering are detected through Cepstral analysis and used for imaging.\nBy integrating Cepstral analysis with 4D-STEM, we demonstrate that information\nabout the distortive part of electron scattering potential can be separated and\nimaged at nm spatial resolution. We apply our technique to the analysis of a\ndislocation core in SiGe and lattice distortions in high entropy alloy.\n"}
{"abstract": "  Recent work in human-computer interaction has explored the use of\nconversational agents as facilitators for group goal-oriented discussions.\nInspired by this work and by the apparent lack of tooling available to support\nit, we created Diplomat, a Python-based framework for building conversational\nagent facilitators. Diplomat is designed to support simple specification of\nagent functionality as well as customizable integration with online chat\nservices. We document a preliminary user study we conducted to help inform the\ndesign of Diplomat. We also describe the architecture, capabilities, and\nlimitations of our tool, which we have shared on GitHub.\n"}
{"abstract": "  Max-plus equations are derived from tropically discretized Sel'kov model via\nultradiscretization. These max-plus equations possess common dynamical\nstructures with the discretized model: Neimark-Sacker bifurcation and limit\ncycles. The limit cycles of the ultradiscrete max-plus equations have seven\ndiscrete states. Furthermore, these max-plus equations exhibit excitability.\nRelationship between the tropically discretized model and the derived max-plus\nequations is also discussed based on numerical results. It is found that the\nderived max-plus equations correspond to a limiting case of the tropically\ndiscretized model.\n"}
{"abstract": "  We present a number of new piecewise-polynomial kernels for image\ninterpolation. The kernels are constructed by optimizing a measure of\ninterpolation quality based on the magnitude of anisotropic artifacts. The\nkernel design process is performed symbolically using Mathematica computer\nalgebra system. Experimental evaluation involving 14 image quality assessment\nmethods demonstrates that our results compare favorably with the existing\nlinear interpolators.\n"}
{"abstract": "  AQ Men is a nova-like variable which is presumed to have a tilted, precessing\naccretion disc. Grazing eclipses in this system have been speculated to be\nuseful in exploring the geometry of its accretion disc. In this work we\nanalysed TESS observations of AQ Men, which provide the best light curve of\nthis object thus far. We show that the depths of the eclipses are changing with\nthe orientation of the accretion disc, which means that they can serve as a\ndirect test of the tilted accretion disc models. The precession period of the\naccretion disc is increasing during the TESS observations. However, it is still\nshorter than the period determined in the previous studies. The amplitude of\nthe variability related to the precession of the accretion disc varies, and so\ndoes the shape of this variability. Moreover, we have detected a positive\nsuperhump that was previously unseen in AQ Men. Interestingly, the positive\nsuperhump has a strongly non-sinusoidal shape, which is not expected for a\nnova-like variable.\n"}
{"abstract": "  Fooling people with highly realistic fake images generated with Deepfake or\nGANs brings a great social disturbance to our society. Many methods have been\nproposed to detect fake images, but they are vulnerable to adversarial\nperturbations -- intentionally designed noises that can lead to the wrong\nprediction. Existing methods of attacking fake image detectors usually generate\nadversarial perturbations to perturb almost the entire image. This is redundant\nand increases the perceptibility of perturbations. In this paper, we propose a\nnovel method to disrupt the fake image detection by determining key pixels to a\nfake image detector and attacking only the key pixels, which results in the\n$L_0$ and the $L_2$ norms of adversarial perturbations much less than those of\nexisting works. Experiments on two public datasets with three fake image\ndetectors indicate that our proposed method achieves state-of-the-art\nperformance in both white-box and black-box attacks.\n"}
{"abstract": "  In an online survey in December 2020, 385 journalists in Germany were\nsurveyed about platforms in journalism and about their frequency of use and\nwillingness to adopt emerging technologies. Journalists have a commitment to\npublish on a journalism platform on a full-time basis. Freelancers have a\nhigher commitment than employed journalists. A platform subscription model is\nrated more attractive than advertising for a platform. Employed journalists on\nthe other hand consider advertising more attractive than freelance journalists.\nFor German journalists it is important that the platform is developed in Europe\nor Germany and that it sets high standards on data protection. Multimedia forms\nand interactive elements are used occasionally, often or always. Stories or\nReels are predominantly not used. AI software as well as editorial analytics\nare rarely or never used. Apart from stories or reels, journalists intend to\nuse multimedia forms and interactive elements more often in the future. They\nare receptive to software for research process documentation as well as to the\nanalysis of indicators of their own publications. Software as a support for\ntext production, image selection or headline suggestions is mostly rejected.\n"}
{"abstract": "  Privacy scoring aims at measuring the privacy violation risk of a user over\nan online social network (OSN). Existing work in the field rely on possibly\nbiased or emotional survey data and focus only on personel purpose OSNs like\nFacebook. In contrast to existing work, in this thesis, we work with real-world\nOSN data collected from LinkedIn, the most popular professional-purpose OSN\n(ProOSN). Towards this end, we developed an extensive crawler to collect all\nrelevant profile data of 5,389 LinkedIn users, modelled these data using both\nrelational and graph databases and quantitatively analyzed all privacy risk\nscoring methods in the literature. Additionally, we propose a novel scoring\nmethod that consider the granularity of data an OSN user shares on her profile\npage. Extensive experimental evaluation of existing and proposed scoring\nmethods indicates the effectiveness of the proposed solution.\n"}
{"abstract": "  Given a polynomial with integral coefficients, one can inquire about the\npossible residues it can take in its image modulo a prime $p$. The sum over\nthese residues can sometimes be computed independently of the choice of prime\n$p$; for example, Gauss showed that the sum over quadratic residues vanishes\nmodulo a prime. In this paper we provide a closed form for the sum over\ndistinct residues in the image of Lucas polynomials of arbitrary degree over\nall primes, and prove a complete characterization of the size of the image set.\nOur result provides the first non-trivial classification of such a sum for a\nfamily of polynomials of unbounded degree.\n"}
{"abstract": "  Blockchains have a two-sided reputation: they are praised for disrupting some\nof our institutions through innovative technology for good, yet notorious for\nbeing slow and expensive to use. In this work, we tackle this issue with\nconcurrency, yet we aim to take a radically different approach by valuing\nsimplicity. We embrace the simplicity through two steps: first, we formulate a\nsimple runtime mechanism to deal with conflicts called concurrency delegation.\nThis method is much simpler and has less overhead, particularly in scenarios\nwhere conflicting transactions are relatively rare. Moreover, to further reduce\nthe number of conflicting transactions, we propose using static annotations\nattached to each transaction, provided by the programmer. These annotations are\npseudo-static: they are static with respect to the lifetime of the transaction,\nand therefore are free to use information such as the origin and parameters of\nthe transaction. We propose a distributor component that can use the output of\nthis pseudo-static annotations and use them to effectively distribute\ntransactions between threads in the least-conflicting way. We name the outcome\nof a system combining concurrency delegation and pseudo-static annotations as\nSonicChain. We evaluate SonicChain for both validation and authoring tasks\nagainst a common workload in blockchains, namely, balance transfers, and\nobserve that it performs expectedly well while introducing very little overhead\nand additional complexity to the system.\n"}
{"abstract": "  We revisit the critical two-dimensional Ashkin-Teller model, i.e. the\n$\\mathbb{Z}_2$ orbifold of the compactified free boson CFT at $c=1$. We solve\nthe model on the plane by computing its three-point structure constants and\nproving crossing symmetry of four-point correlation functions. We do this not\nonly for affine primary fields, but also for Virasoro primary fields, i.e.\nhigher twist fields and degenerate fields.\n  This leads us to clarify the analytic properties of Virasoro conformal blocks\nand fusion kernels at $c=1$. We show that blocks with a degenerate channel\nfield should be computed by taking limits in the central charge, rather than in\nthe conformal dimension. In particular, Al. Zamolodchikov's simple explicit\nexpression for the blocks that appear in four-twist correlation functions is\nonly valid in the non-degenerate case: degenerate blocks, starting with the\nidentity block, are more complicated generalized theta functions.\n"}
{"abstract": "  Radar has long been a common sensor on autonomous vehicles for obstacle\nranging and speed estimation. However, as a robust sensor to all-weather\nconditions, radar's capability has not been well-exploited, compared with\ncamera or LiDAR. Instead of just serving as a supplementary sensor, radar's\nrich information hidden in the radio frequencies can potentially provide useful\nclues to achieve more complicated tasks, like object classification and\ndetection. In this paper, we propose a new dataset, named CRUW, with a\nsystematic annotator and performance evaluation system to address the radar\nobject detection (ROD) task, which aims to classify and localize the objects in\n3D purely from radar's radio frequency (RF) images. To the best of our\nknowledge, CRUW is the first public large-scale dataset with a systematic\nannotation and evaluation system, which involves camera RGB images and radar RF\nimages, collected in various driving scenarios.\n"}
{"abstract": "  Small-sized unmanned surface vehicles (USV) are coastal water devices with a\nbroad range of applications such as environmental control and surveillance. A\ncrucial capability for autonomous operation is obstacle detection for timely\nreaction and collision avoidance, which has been recently explored in the\ncontext of camera-based visual scene interpretation. Owing to curated datasets,\nsubstantial advances in scene interpretation have been made in a related field\nof unmanned ground vehicles. However, the current maritime datasets do not\nadequately capture the complexity of real-world USV scenes and the evaluation\nprotocols are not standardised, which makes cross-paper comparison of different\nmethods difficult and hiders the progress. To address these issues, we\nintroduce a new obstacle detection benchmark MODS, which considers two major\nperception tasks: maritime object detection and the more general maritime\nobstacle segmentation. We present a new diverse maritime evaluation dataset\ncontaining approximately 81k stereo images synchronized with an on-board IMU,\nwith over 60k objects annotated. We propose a new obstacle segmentation\nperformance evaluation protocol that reflects the detection accuracy in a way\nmeaningful for practical USV navigation. Seventeen recent state-of-the-art\nobject detection and obstacle segmentation methods are evaluated using the\nproposed protocol, creating a benchmark to facilitate development of the field.\n"}
{"abstract": "  With vast mmWave spectrum and narrow beam antenna technology, precise\nposition location is now possible in 5G and future mobile communication\nsystems. In this article, we describe how centimeterlevel localization accuracy\ncan be achieved, particularly through the use of map-based techniques. We show\nhow data fusion of parallel information streams, machine learning, and\ncooperative localization techniques further improve positioning accuracy.\n"}
{"abstract": "  Batch Normalization (BN) is a commonly used technique to accelerate and\nstabilize training of deep neural networks. Despite its empirical success, a\nfull theoretical understanding of BN is yet to be developed. In this work, we\nanalyze BN through the lens of convex optimization. We introduce an analytic\nframework based on convex duality to obtain exact convex representations of\nweight-decay regularized ReLU networks with BN, which can be trained in\npolynomial-time. Our analyses also show that optimal layer weights can be\nobtained as simple closed-form formulas in the high-dimensional and/or\noverparameterized regimes. Furthermore, we find that Gradient Descent provides\nan algorithmic bias effect on the standard non-convex BN network, and we design\nan approach to explicitly encode this implicit regularization into the convex\nobjective. Experiments with CIFAR image classification highlight the\neffectiveness of this explicit regularization for mimicking and substantially\nimproving the performance of standard BN networks.\n"}
{"abstract": "  The variational discrete element method developed in [28] for dynamic\nelasto-plastic computations is adapted to compute the deformation of elastic\nCosserat materials. In addition to cellwise displacement degrees of freedom\n(dofs), cellwise rotational dofs are added. A reconstruction is devised to\nobtain P 1 non-conforming polynomials in each cell and thus constant strains\nand stresses in each cell. The method requires only the usual macroscopic\nparameters of a Cosserat material and no microscopic parameter. Numerical\nexamples show the robustness of the method for both static and dynamic\ncomputations in two and three dimensions.\n"}
{"abstract": "  Surgical state estimators in robot-assisted surgery (RAS) - especially those\ntrained via learning techniques - rely heavily on datasets that capture surgeon\nactions in laboratory or real-world surgical tasks. Real-world RAS datasets are\ncostly to acquire, are obtained from multiple surgeons who may use different\nsurgical strategies, and are recorded under uncontrolled conditions in highly\ncomplex environments. The combination of high diversity and limited data calls\nfor new learning methods that are robust and invariant to operating conditions\nand surgical techniques. We propose StiseNet, a Surgical Task Invariance State\nEstimation Network with an invariance induction framework that minimizes the\neffects of variations in surgical technique and operating environments inherent\nto RAS datasets. StiseNet's adversarial architecture learns to separate\nnuisance factors from information needed for surgical state estimation.\nStiseNet is shown to outperform state-of-the-art state estimation methods on\nthree datasets (including a new real-world RAS dataset: HERNIA-20).\n"}
{"abstract": "  We examine the dielectron invariant mass spectrum and the variation of the\npolarization asymmetry as a function of that mass in the event generator of the\nfive-dimensional Bethe-Heitler differential cross section of the conversion of\nlinearly polarized photons that we have implemented recently as a Geant4\nPhysics Model. We compare the results obtained with simulated samples to\nanalytical expressions.\n  Studies of $b \\to s \\gamma$ decays at LHC and Belle II experiments could\nprovide hints of physics beyond the standard model in flavor-changing neutral\ncurrent processes; issues have been reported in the simulation of the\nproperties of the background induced by genuine $\\gamma$-ray conversions in\ntheir detectors. Our results demonstrate that using the five-dimensional model\nwould help solve these issues.\n"}
{"abstract": "  The underground muon detector of the Pierre Auger Observatory is aimed at\nattaining direct measurements of the muonic component of extensive air showers\nproduced by cosmic rays with energy from $10^{16.5}$ eV up to the region of the\nankle (around $10^{18.7}$ eV). It consists of two nested triangular grids of\nunderground scintillators with 433 m, and 750 m spacings and a total of 71\npositions, each with 192 scintillator strips (30 m$^2$) deployed 2.3 m\nunderground. The light produced by impinging muons in the scintillators is\npropagated with optical fibers towards an array of silicon photomultipliers. In\nthis work, we present the development, validation, and performance of an\nend-to-end tool for simulating the response of the underground muon detector to\nsingle-muon signals, which constitutes the basis for further simulations of the\nwhole array. Laboratory data and simulation outcomes are found consistent,\nshowing that with the underground muon detector we can measure single muons,\nwith an efficiency of 99 %, up to about 1050 particles arriving at exactly the\nsame time in 30 m$^2$ of scintillator.\n"}
{"abstract": "  We study $N$ vicious Brownian bridges propagating from an initial\nconfiguration $\\{a_1 < a_2 < \\ldots< a_N \\}$ at time $t=0$ to a final\nconfiguration $\\{b_1 < b_2 < \\ldots< b_N \\}$ at time $t=t_f$, while staying\nnon-intersecting for all $0\\leq t \\leq t_f$. We first show that this problem\ncan be mapped to a non-intersecting Dyson's Brownian bridges with Dyson index\n$\\beta=2$. For the latter we derive an exact effective Langevin equation that\nallows to generate very efficiently the vicious bridge configurations. In\nparticular, for the flat-to-flat configuration in the large $N$ limit, where\n$a_i = b_i = (i-1)/N$, for $i = 1, \\cdots, N$, we use this effective Langevin\nequation to derive an exact Burgers' equation (in the inviscid limit) for the\nGreen's function and solve this Burgers' equation for arbitrary time $0 \\leq\nt\\leq t_f$. At certain specific values of intermediate times $t$, such as\n$t=t_f/2$, $t=t_f/3$ and $t=t_f/4$ we obtain the average density of the\nflat-to-flat bridge explicitly. We also derive explicitly how the two edges of\nthe average density evolve from time $t=0$ to time $t=t_f$. Finally, we discuss\nconnections to some well known problems, such as the Chern-Simons model, the\nrelated Stieltjes-Wigert orthogonal polynomials and the Borodin-Muttalib\nensemble of determinantal point processes.\n"}
{"abstract": "  We propose and analyze a specific asymptotic stochastic order for random\nprocesses based on the measure of departure discussed in the literature. As\napplications, we stochastically compare mixtures of order statistics and record\nvalues coming from two different homogeneous samples, as the sample size\nbecomes large.\n"}
{"abstract": "  We report on observations of the Be/X-ray binary system SwiftJ1626.6-5156\nperformed with NuSTAR during a short outburst in March 2021, following its\ndetection of by the MAXI monitor and Spektrum-Roentgen-Gamma (SRG) observatory.\nOur analysis of the broadband X-ray spectrum of the source confirms the\npresence of two absorption-like features at energies ~9 and ~17 keV previously\nreported in literature and interpreted as the fundamental cyclotron resonance\nscattering feature (CRSF) and its first harmonic (based on RXTE data). The\nbetter sensitivity and energy resolution of NuSTAR, combined with the low\nenergy coverage of NICER, allowed us to detect two additional absorption-like\nfeatures at ~4.9 keV and ~13 keV. We conclude, therefore, that in total four\ncyclotron lines are observed in the spectrum of SwiftJ1626.6-5156: the\nfundamental CRSF at ~4.9 keV and three higher spaced harmonics. This discovery\nmakes SwiftJ1626.6-5156 the second accreting pulsar, after 4U0115+63, whose\nspectrum is characterized by more than three lines of a cyclotronic origin, and\nimplies the source has the weakest confirmed magnetic field among all X-ray\npulsars B~4E11 G. This discovery makes SwiftJ1626.6-5156 one of prime targets\nfor the upcoming X-ray polarimetry missions covering soft X-ray band such as\nIXPE and eXTP.\n"}
{"abstract": "  Sentiment analysis is one of the most fundamental tasks in Natural Language\nProcessing. Popular languages like English, Arabic, Russian, Mandarin, and also\nIndian languages such as Hindi, Bengali, Tamil have seen a significant amount\nof work in this area. However, the Marathi language which is the third most\npopular language in India still lags behind due to the absence of proper\ndatasets. In this paper, we present the first major publicly available Marathi\nSentiment Analysis Dataset - L3CubeMahaSent. It is curated using tweets\nextracted from various Maharashtrian personalities' Twitter accounts. Our\ndataset consists of ~16,000 distinct tweets classified in three broad classes\nviz. positive, negative, and neutral. We also present the guidelines using\nwhich we annotated the tweets. Finally, we present the statistics of our\ndataset and baseline classification results using CNN, LSTM, ULMFiT, and\nBERT-based deep learning models.\n"}
{"abstract": "  Most of the turbulence models in practice are based on the assumption of a\nlinear relation between Reynolds stresses and mean flow strain rates which\ngenerally provides a good approximation in case of attached and fully turbulent\nflows. A two dimensional numerical study has been carried out over NACA 0021\nwith k-\\omega SST model with non-linear correction at Re = 120,000 for various\nangles of attack which experiences the formation of a laminar separation bubble\n(LSB). A correct prediction of LSB requires an accurate resolution of\nanisotropy in Reynolds stresses. For comparison with other linear models, the\nsimulations are also performed with k-kl-\\omega, k-\\omega SST and Spalart\nAllmaras. The performance of these models is assessed through aerodynamic lift,\ndrag, pressure and friction coefficients. It is found that the non-linear\nk-\\omega SST and k-kl-\\omega transition model provide comparable quality of\nprediction in lift and drag coefficients (in spite of the fact that non-linear\nk-\\omega SST involves solving less number of transport equation than the\ntransition model) as observed in the experiments whereas k-\\omega SST and SA\nmodels under predict the drag coefficient value at low angle of attack due to\ninability to capture the separation induced transition. It is also observed\nthat the location of laminar separation bubble is captured accurately when\nnonlinear or transition model is used as opposed to the SA or linear SST\nmodels, which lack in the ability to predict the same.\n"}
{"abstract": "  Adaptive control is subject to stability and performance issues when a\nlearned model is used to enhance its performance. This paper thus presents a\ndeep learning-based adaptive control framework for nonlinear systems with\nmultiplicatively-separable parametrization, called adaptive Neural Contraction\nMetric (aNCM). The aNCM approximates real-time optimization for computing a\ndifferential Lyapunov function and a corresponding stabilizing adaptive control\nlaw by using a Deep Neural Network (DNN). The use of DNNs permits real-time\nimplementation of the control law and broad applicability to a variety of\nnonlinear systems with parametric and nonparametric uncertainties. We show\nusing contraction theory that the aNCM ensures exponential boundedness of the\ndistance between the target and controlled trajectories in the presence of\nparametric uncertainties of the model, learning errors caused by aNCM\napproximation, and external disturbances. Its superiority to the existing\nrobust and adaptive control methods is demonstrated using a cart-pole balancing\nmodel.\n"}
{"abstract": "  We present a general approach to constructing permutation tests that are both\nexact for the null hypothesis of equality of distributions and asymptotically\ncorrect for testing equality of parameters of distributions while allowing the\ndistributions themselves to differ. These robust permutation tests transform a\ngiven test statistic by a consistent estimator of its limiting distribution\nfunction before enumerating its permutation distribution. This transformation,\nknown as prepivoting, aligns the unconditional limiting distribution for the\ntest statistic with the probability limit of its permutation distribution.\nThrough prepivoting, the tests permute one minus an asymptotically valid\n$p$-value for testing the null of equality of parameters. We describe two\napproaches for prepivoting within permutation tests, one directly using\nasymptotic normality and the other using the bootstrap. We further illustrate\nthat permutation tests using bootstrap prepivoting can provide improvements to\nthe order of the error in rejection probability relative to competing\ntransformations when testing equality of parameters, while maintaining\nexactness under equality of distributions. Simulation studies highlight the\nversatility of the proposal, illustrating the restoration of asymptotic\nvalidity to a wide range of permutation tests conducted when only the\nparameters of distributions are equal.\n"}
{"abstract": "  PredDiff is a model-agnostic, local attribution method that is firmly rooted\nin probability theory. Its simple intuition is to measure prediction changes\nwhile marginalizing features. In this work, we clarify properties of PredDiff\nand its connection to Shapley values. We stress important differences between\nclassification and regression, which require a specific treatment within both\nformalisms. We extend PredDiff by introducing a new, well-founded measure for\ninteraction effects between arbitrary feature subsets. The study of interaction\neffects represents an inevitable step towards a comprehensive understanding of\nblack-box models and is particularly important for science applications. As\nopposed to Shapley values, our novel measure maintains the original linear\nscaling and is thus generally applicable to real-world problems.\n"}
{"abstract": "  Automatically tuning software configuration for optimizing a single\nperformance attribute (e.g., minimizing latency) is not trivial, due to the\nnature of the configuration systems (e.g., complex landscape and expensive\nmeasurement). To deal with the problem, existing work has been focusing on\ndeveloping various effective optimizers. However, a prominent issue that all\nthese optimizers need to take care of is how to avoid the search being trapped\nin local optima -- a hard nut to crack for software configuration tuning due to\nits rugged and sparse landscape, and neighboring configurations tending to\nbehave very differently. Overcoming such in an expensive measurement setting is\neven more challenging. In this paper, we take a different perspective to tackle\nthis issue. Instead of focusing on improving the optimizer, we work on the\nlevel of optimization model. We do this by proposing a meta\nmulti-objectivization model (MMO) that considers an auxiliary performance\nobjective (e.g., throughput in addition to latency). What makes this model\nunique is that we do not optimize the auxiliary performance objective, but\nrather use it to make similarly-performing while different configurations less\ncomparable (i.e. Pareto nondominated to each other), thus preventing the search\nfrom being trapped in local optima.\n  Experiments on eight real-world software systems/environments with diverse\nperformance attributes reveal that our MMO model is statistically more\neffective than state-of-the-art single-objective counterparts in overcoming\nlocal optima (up to 42% gain), while using as low as 24% of their measurements\nto achieve the same (or better) performance result.\n"}
{"abstract": "  We present a method for generating comparative summaries that highlights\nsimilarities and contradictions in input documents. The key challenge in\ncreating such summaries is the lack of large parallel training data required\nfor training typical summarization systems. To this end, we introduce a hybrid\ngeneration approach inspired by traditional concept-to-text systems. To enable\naccurate comparison between different sources, the model first learns to\nextract pertinent relations from input documents. The content planning\ncomponent uses deterministic operators to aggregate these relations after\nidentifying a subset for inclusion into a summary. The surface realization\ncomponent lexicalizes this information using a text-infilling language model.\nBy separately modeling content selection and realization, we can effectively\ntrain them with limited annotations. We implemented and tested the model in the\ndomain of nutrition and health -- rife with inconsistencies. Compared to\nconventional methods, our framework leads to more faithful, relevant and\naggregation-sensitive summarization -- while being equally fluent.\n"}
{"abstract": "  In this article we investigate the unitary dynamics of squashed entanglement\nand concurrence measures in Werner state and maximally entangled mixed states\n(MEMS) under two different Hamiltonians. The aim of the present study is two\nfold. The first part of the study deals with the dynamics under Heisenberg\nHamiltonian and the second part deals under bi-linear bi-quadratic Hamiltonian\nwhich is the extension of first Hamiltonian. In both the parts we investigate\nthe dynamical trade off and balancing points for squashed entanglement and\nconcurrence. During the study, we also found the results of entanglement sudden\ndeath (ESD) with Heisenberg Hamiltonian in Werner state under concurrence\nmeasure. In second part, we investigate the special result for bi-linear\nbi-quadratic Hamiltonian; it does not disturb squashed entanglement and\nconcurrence in both the states and exhibit the robust character for both of the\nstates.\n"}
{"abstract": "  Using nonlinear mathematical models and experimental data from laboratory and\nclinical studies, we have designed new combination therapies against COVID-19.\n"}
{"abstract": "  In this work, we introduce the new scene understanding task of Part-aware\nPanoptic Segmentation (PPS), which aims to understand a scene at multiple\nlevels of abstraction, and unifies the tasks of scene parsing and part parsing.\nFor this novel task, we provide consistent annotations on two commonly used\ndatasets: Cityscapes and Pascal VOC. Moreover, we present a single metric to\nevaluate PPS, called Part-aware Panoptic Quality (PartPQ). For this new task,\nusing the metric and annotations, we set multiple baselines by merging results\nof existing state-of-the-art methods for panoptic segmentation and part\nsegmentation. Finally, we conduct several experiments that evaluate the\nimportance of the different levels of abstraction in this single task.\n"}
{"abstract": "  This report aims to survey multi-agent Q-Learning algorithms, analyze\ndifferent game theory frameworks used, address each framework's applications,\nand report challenges and future directions. The target application for this\nstudy is resource management in the wireless sensor network.\n  In the first section, the author provided an introduction regarding the\napplications of wireless sensor networks. After that, the author presented a\nsummary of the Q-Learning algorithm, a well-known classic solution for\nmodel-free reinforcement learning problems.\n  In the third section, the author extended the Q-Learning algorithm for\nmulti-agent scenarios and discussed its challenges.\n  In the fourth section, the author surveyed sets of game-theoretic frameworks\nthat researchers used to address this problem for resource allocation and task\nscheduling in the wireless sensor networks. Lastly, the author mentioned some\ninteresting open challenges in this domain.\n"}
{"abstract": "  The competition between kinetic energy and Coulomb interactions in electronic\nsystems can lead to complex many-body ground states with competing\nsuperconducting, charge density wave, and magnetic orders. Here we study the\nlow temperature phases of a strongly interacting zinc-oxide-based high mobility\ntwo dimensional electron system that displays a tunable metal-insulator\ntransition. Through a comprehensive analysis of the dependence of electronic\ntransport on temperature, carrier density, in-plane and perpendicular magnetic\nfields, and voltage bias, we provide evidence for the existence of competing\ncorrelated metallic and insulating states with varying degrees of spin\npolarization. Our system features an unprecedented level of agreement with the\nstate-of-the-art Quantum Monte Carlo phase diagram of the ideal jellium model,\nincluding a Wigner crystallization transition at a value of the interaction\nparameter $r_s\\sim 30$ and the absence of a pure Stoner transition. In-plane\nfield dependence of transport reveals a new low temperature state with partial\nspin polarization separating the spin unpolarized metal and the Wigner crystal,\nwhich we examine against possible theoretical scenarios such as an\nanti-ferromagnetic crystal, Coulomb induced micro-emulsions, and disorder\ndriven puddle formation.\n"}
{"abstract": "  Multiple imputation is increasingly used in dealing with missing data. While\nsome conventional multiple imputation approaches are well studied and have\nshown empirical validity, they entail limitations in processing large datasets\nwith complex data structures. Their imputation performances usually rely on\nexpert knowledge of the inherent relations among variables. In addition, these\nstandard approaches tend to be computationally inefficient for medium and large\ndatasets. In this paper, we propose a scalable multiple imputation framework\nmixgb, which is based on XGBoost, bootstrapping and predictive mean matching.\nXGBoost, one of the fastest implementations of gradient boosted trees, is able\nto automatically retain interactions and non-linear relations in a dataset\nwhile achieving high computational efficiency. With the aid of bootstrapping\nand predictive mean matching, we show that our approach obtains less biased\nestimates and reflects appropriate imputation variability. The proposed\nframework is implemented in an R package misle. Supplementary materials for\nthis article are available online.\n"}
{"abstract": "  We derive the chiral effective Lagrangian for excited heavy-light mesons from\nQCD under proper approximations. We focus on the chiral partners with\n$j_l^P=\\frac{3}{2}^+$ and $j_l^P=\\frac{3}{2}^-$ which amounts to ($1^+,2^+$)\nand ($1^-,2^-$) states respectively. The low energy constants including the\nmasses of the chiral partners are calculated. The calculated spectrum for the\nexcited mesons are found roughly consistent with experimental data. In\naddition, our results indicate that quantum numbers of $B_J(5970)$ can be\nidentified with $1^-$ or $2^-$.\n"}
{"abstract": "  One dimensional analysis is presented of solitary positive potential plasma\nstructures whose velocity lies within the range of ion distribution velocities\nthat are strongly populated: so called \"slow\" electron holes. It is shown that\nto avoid the self-acceleration of the hole velocity away from ion velocities it\nmust lie within a local minimum in the ion velocity distribution. Quantitative\ncriteria for the existence of stable equilibria are obtained. The background\nion distributions required are generally stable to ion-ion modes unless the\nelectron temperature is much higher than the ion temperature. Since slow\npositive potential solitons are shown not to be possible without a significant\ncontribution from trapped electrons, it seems highly likely that such observed\nslow potential structures are indeed electron holes.\n"}
{"abstract": "  A well-known formula of Whipple relates certain hypergeometric values\n$_7F_6(1)$ and $_4F_3(1)$. In this paper we revisit this relation from the\nviewpoint of the underlying hypergeometric data $HD$, to which there are also\nassociated hypergeometric character sums and Galois representations. We explain\na special structure behind Whipple's formula when the hypergeometric data $HD$\nare primitive and self-dual. If the data are also defined over $\\mathbb Q$, by\nthe work of Katz, Beukers, Cohen, and Mellit, there are compatible families of\n$\\ell$-adic representations of the absolute Galois group of $\\mathbb Q$\nattached to $HD$. For specialized choices of $HD$, these Galois representations\nare shown to be decomposable and automorphic. As a consequence, the values of\nthe corresponding hypergeometric character sums can be explicitly expressed in\nterms of Fourier coefficients of certain modular forms. We further relate the\nhypergeometric values $_7F_6(1)$ in Whipple's formula to the periods of these\nmodular forms.\n"}
{"abstract": "  A theory for analyzing the radiative and reactive energies for pulse\nradiators in free space is presented. With the proposed definition of reactive\nenergies and radiative energies, power balance at arbitrarily chosen\nobservation surfaces are established, which intuitively shows that the Poynting\nvector contains not only the power flux density associated with the radiative\nenergies, but also the influence of the fluctuation of the reactive energies\ndragging by the sources. A new vector is defined for the radiative power flux\ndensity. The radiative energies passing through observation surfaces enclosing\nthe radiator are accurately calculated. Numerical results verifies that the\nproposed radiative flux density is more proper for expressing the radiative\npower flux density than the Poynting vector.\n"}
{"abstract": "  In this thesis we study the momentum space approach to the solution of the\nCWI's of CFT's in higher dimensions. Our work's goal is to illustrate the\nessential steps needed to build tensor correlators starting from the scalar\nsolutions, for 3-point functions. In the case of 4-point functions, our\nattention is centred around scalar correlators for which the CWI's are\nsufficient to isolate the unique solution if we enhance the symmetry with the\naddition of a dual conformal symmetry. Dual conformal symmetry in momentum\nspace is obtained once the momentum variables are rewritten in a dual form, as\nthe difference of coordinate-like variables and treated as ordinary correlators\nin such variables, mirroring the action of coordinate space. This enhancement\nof the symmetry is sufficient to fix the solutions also for such correlators.\nThe solution of the conformal constraints are given in terms of triple-$K$\nintegrals and are expressed in terms of a set of constants, specific for each\ncorrelator and spacetime dimension. We present a discussion of the intermediate\nsteps in the description of two nontrivial correlators, the $TTO$ and the\n$TTT$, in a more pedagogical way, offering details that could help extend such\nmethods to higher point function.\n"}
{"abstract": "  Machine learning models are often implemented in cohort with humans in the\npipeline, with the model having an option to defer to a domain expert in cases\nwhere it has low confidence in its inference. Our goal is to design mechanisms\nfor ensuring accuracy and fairness in such prediction systems that combine\nmachine learning model inferences and domain expert predictions. Prior work on\n\"deferral systems\" in classification settings has focused on the setting of a\npipeline with a single expert and aimed to accommodate the inaccuracies and\nbiases of this expert to simultaneously learn an inference model and a deferral\nsystem. Our work extends this framework to settings where multiple experts are\navailable, with each expert having their own domain of expertise and biases. We\npropose a framework that simultaneously learns a classifier and a deferral\nsystem, with the deferral system choosing to defer to one or more human experts\nin cases of input where the classifier has low confidence. We test our\nframework on a synthetic dataset and a content moderation dataset with biased\nsynthetic experts, and show that it significantly improves the accuracy and\nfairness of the final predictions, compared to the baselines. We also collect\ncrowdsourced labels for the content moderation task to construct a real-world\ndataset for the evaluation of hybrid machine-human frameworks and show that our\nproposed learning framework outperforms baselines on this real-world dataset as\nwell.\n"}
{"abstract": "  The global gravitational-wave detector network achieves higher detection\nrates, better parameter estimates, and more accurate sky localisation, as the\nnumber of detectors, $\\mathcal{I}$ increases. This paper quantifies network\nperformance as a function of $\\mathcal{I}$ for BayesWave, a source-agnostic,\nwavelet-based, Bayesian algorithm which distinguishes between true\nastrophysical signals and instrumental glitches. Detection confidence is\nquantified using the signal-to-glitch Bayes factor,\n$\\mathcal{B}_{\\mathcal{S},\\mathcal{G}}$. An analytic scaling is derived for\n$\\mathcal{B}_{\\mathcal{S},\\mathcal{G}}$ versus $\\mathcal{I}$, the number of\nwavelets, and the network signal-to-noise ratio, SNR$_\\text{net}$, which is\nconfirmed empirically via injections into detector noise of the\nHanford-Livingston (HL), Hanford-Livingston-Virgo (HLV), and\nHanford-Livingston-KAGRA-Virgo (HLKV) networks at projected sensitivities for\nthe fourth observing run (O4). The empirical and analytic scalings are\nconsistent; $\\mathcal{B}_{\\mathcal{S},\\mathcal{G}}$ increases with\n$\\mathcal{I}$. The accuracy of waveform reconstruction is quantified using the\noverlap between injected and recovered waveform, $\\mathcal{O}_\\text{net}$. The\nHLV and HLKV network recovers $87\\%$ and $86\\%$ of the injected waveforms with\n$\\mathcal{O}_\\text{net}>0.8$ respectively, compared to $81\\%$ with the HL\nnetwork. The accuracy of BayesWave sky localisation is $\\approx 10$ times\nbetter for the HLV network than the HL network, as measured by the search area,\n$\\mathcal{A}$, and the sky areas contained within $50\\%$ and $90\\%$ confidence\nintervals. Marginal improvement in sky localisation is also observed with the\naddition of KAGRA.\n"}
{"abstract": "  VI$_{3}$ is a ferromagnet with planar honeycomb sheets of bonded V$^{3+}$\nions held together by van der Waals forces. We apply neutron spectroscopy to\nmeasure the two dimensional ($J/J_{c} \\approx 17$) magnetic excitations in the\nferromagnetic phase, finding two energetically gapped ($\\Delta \\approx k_{B}\nT_{c} \\approx$ 55 K) and dispersive excitations. We apply a multi-level spin\nwave formalism to describe the spectra in terms of two coexisting domains\nhosting differing V$^{3+}$ orbital ground states built from contrasting\ndistorted octahedral environments. This analysis fits a common nearest neighbor\nin-plane exchange coupling ($J$=-8.6 $\\pm$ 0.3 meV) between V$^{3+}$ sites. The\ndistorted local crystalline electric field combined with spin-orbit coupling\nprovides the needed magnetic anisotropy for spatially long-ranged\ntwo-dimensional ferromagnetism in VI$_{3}$.\n"}
{"abstract": "  We introduce weak barycenters of a family of probability distributions, based\non the recently developed notion of optimal weak transport of mass by Gozlanet\nal. (2017) and Backhoff-Veraguas et al. (2020). We provide a theoretical\nanalysis of this object and discuss its interpretation in the light of convex\nordering between probability measures. In particular, we show that, rather than\naveraging the input distributions in a geometric way (as the Wasserstein\nbarycenter based on classic optimal transport does) weak barycenters extract\ncommon geometric information shared by all the input distributions, encoded as\na latent random variable that underlies all of them. We also provide an\niterative algorithm to compute a weak barycenter for a finite family of input\ndistributions, and a stochastic algorithm that computes them for arbitrary\npopulations of laws. The latter approach is particularly well suited for the\nstreaming setting, i.e., when distributions are observed sequentially. The\nnotion of weak barycenter and our approaches to compute it are illustrated on\nsynthetic examples, validated on 2D real-world data and compared to standard\nWasserstein barycenters.\n"}
{"abstract": "  In this paper, we compute the Bott connections and their curvature on\nthree-dimensional Lorentzian Lie groups with three different distributions, and\nwe classify affine Ricci solitons associated to the Bott connection on\nthree-dimensional Lorentzian Lie groups with three different distributions.\n"}
{"abstract": "  The relative motion of three impenetrable particles on a ring, in our case\ntwo identical fermions and one impurity, is isomorphic to a triangular quantum\nbilliard. Depending on the ratio $\\kappa$ of the impurity and fermion masses,\nthe billiards can be integrable or non-integrable (also referred to in the main\ntext as chaotic). To set the stage, we first investigate the energy level\ndistributions of the billiards as a function of $1/\\kappa\\in [0,1]$ and find no\nevidence of integrable cases beyond the limiting values $1/\\kappa=1$ and\n$1/\\kappa=0$. Then, we use machine learning tools to analyze properties of\nprobability distributions of individual quantum states. We find that\nconvolutional neural networks can correctly classify integrable and\nnon-integrable states.The decisive features of the wave functions are the\nnormalization and a large number of zero elements, corresponding to the\nexistence of a nodal line. The network achieves typical accuracies of 97%,\nsuggesting that machine learning tools can be used to analyze and classify the\nmorphology of probability densities obtained in theory or experiment.\n"}
{"abstract": "  In this paper we study the effects of dynamic wetting on the immiscible\ndisplacement of a high viscosity fluid subject to the radial injection of a\nless viscous fluid in a Hele-Shaw cell. The displaced fluid can leave behind a\ntrailing film that coats the cell walls, dynamically affecting the pressure\ndrop at the fluid interface. By considering the non-linear pressure drop in a\nboundary element formulation, we construct a Picard scheme to iteratively\npredict the interfacial velocity and subsequent displacement in finite mobility\nratio flow regimes. Dynamic wetting delays the onset of finger bifurcation in\nthe late stages of interfacial growth, and at high local capillary numbers can\nalter the fundamental mode of bifurcation, producing vastly different finger\nmorphologies. In low mobility ratio regimes, we see that finger interaction is\nreduced and characteristic finger breaking mechanisms are delayed but never\nfully inhibited. In high mobility ratio regimes, finger shielding is reduced\nwhen dynamic wetting is present. Finger bifurcation is delayed which allows the\nprimary fingers to advance further into the domain before secondary fingers are\ngenerated, reducing the level of competition.\n"}
{"abstract": "  The symmetry axes of active galactic nuclei (AGN) are randomly distributed in\nspace but highly inclined sources are heavily obscured and are not seen as\nquasars with broad emission lines. The obscuring torus geometry determines the\naverage viewing angle, and if the torus geometry changes with the redshift,\nthis average viewing angle will also change. Thus the ratio between the\nisotropic luminosity and observed luminosity may change systematically with\nredshift. Therefore, if we use quasars to measure the luminosity distance by\nevaluating the isotropic absolute luminosity and measuring the observed flux,\nwe can have a redshift-dependent bias which can propagate to cosmological\nparameters. We propose a toy model for testing the effect of viewing angle\nuncertainty on measurement of the luminosity distance. The model is based on\nanalytical description of the obscuring torus applied to one-parameter\nobservational data. It illustrates the possible change of the torus covering\nfactor between the two chosen redshift ranges. We have estimated the possible\nerror on specific cosmological parameters (H0,Omega_m) for the flat Lambda-CDM\ncosmology if a method is calibrated at low redshift and applied to the higher\nredshift. The errors on cosmological parameters due to potential dependence of\nviewing angle on redshift are found to be potentially significant, and the\neffect will have to be accommodated in the future in all quasar-based\ncosmological methods. A careful systematic study of AGN mean viewing angle\nacross redshift is necessary, with the use of appropriate samples and models\nwhich uniquely determine the inclination of each source.\n"}
{"abstract": "  The electric dipole moment (EDM) of electron is studied in the supersymmetric\n$\\rm A_4$ modular invariant theory of flavors with CP invariance. The CP\nsymmetry of the lepton sector is broken by fixing the modulus $\\tau$. Lepton\nmass matrices are completely consistent with observed lepton masses and mixing\nangles in our model. In this framework, a fixed $\\tau$ also causes the CP\nviolation in the soft SUSY breaking terms. The elecrton EDM arises from the CP\nnon-conserved soft SUSY breaking terms. The experimental upper bound of the\nelectron EDM excludes the SUSY mass scale below $2-6$ TeV for five cases of the\nlepton mass matrices. In order to see the effect of CP phase of the modulus\n$\\tau$, we examine the correlation between the electron EDM and the decay rate\nof the $\\mu \\rightarrow e \\gamma$ decay, which is also predicted by the soft\nSUSY breaking terms. The correlations are clearly predicted in contrast to\nmodels of the conventional flavor symmetry. The SUSY mass scale will be\nconstrained by the future sensitivity of the electron EDM, $|d_e/e| \\simeq\n10^{-30}$. Indeed, it could probe the SUSY mass range of $10-20$ TeV in our\nmodel. Thus, the electron EDM provides a severe test of the CP violation via\nthe modulus $\\tau$ in the supersymmetric modular invariant theory of flavors.\n"}
{"abstract": "  This paper is concerned with a class of nonlinear boundary value problem\ninvolving fractional derivative in the $\\varphi$-Riemann-Liouville sense. Some\nProperties of the Green's function for this problem are mentioned. By means of\nthe Banach contraction principle in $b$-metric space and the technique of the\n$\\gamma$-$\\psi$- Geraghty contractive maps, existence and uniqueness results\nare obtained. Two examples are given to support the theoretical results.\n"}
{"abstract": "  The Frank-Wolfe algorithm is a method for constrained optimization that\nrelies on linear minimizations, as opposed to projections. Therefore, a\nmotivation put forward in a large body of work on the Frank-Wolfe algorithm is\nthe computational advantage of solving linear minimizations instead of\nprojections. However, the discussions supporting this advantage are often too\nsuccinct or incomplete. In this paper, we review the complexity bounds for both\ntasks on several sets commonly used in optimization. Projection methods onto\nthe $\\ell_p$-ball, $p\\in\\left]1,2\\right[\\cup\\left]2,+\\infty\\right[$, and the\nBirkhoff polytope are also proposed.\n"}
{"abstract": "  In this paper, we get the sharp bound for $|G/O_p(G)|_p$ under the assumption\nthat either $p^2 \\nmid \\chi(1)$ for all $\\chi \\in {\\rm Irr}(G)$ or $p^2 \\nmid\n\\phi(1)$ for all $\\phi \\in {\\rm IBr}_p(G)$. This would settle two conjectures\nraised by Lewis, Navarro, Tiep, and Tong-Viet.\n"}
{"abstract": "  Let $G$ be a group with identity $e$. Let $R$ be a $G$-graded commutative\nring and $M$ a graded $R$-module. A proper graded submodule $N$ of $M$ is\ncalled \\textit{a graded semiprime submodule} if whenever $r\\in h(R)$, $m\\in\nh(M)$ and $n\\in Z^{+}$ with $r^{n}m\\in N$, then $rm\\in N$. In this paper, we\nintroduce the concept of graded quasi-semiprime submodule as a generalization\nof graded semiprime submodule and show a number of results in this class. We\nsay that a proper graded submodule $N$ of $M$ is \\emph{a graded quasi-semiprime\nsubmodule} if $(N:_{R}M)$ is a graded semiprime ideal of $R$.\n"}
{"abstract": "  In 2018, Solar Cycle 24 entered into a solar minimum phase. During this\nperiod, 11 million zenithal night sky brightness (NSB) data were collected at\ndifferent dark sites around the planet, including astronomical observatories\nand natural protected areas, with identical broadband Telescope Encoder and Sky\nSensor photometers (based on the Unihedron Sky Quality Meter TSL237 sensor). A\ndetailed observational review of the multiple effects that contribute to the\nNSB measurement has been conducted with optimal filters designed to avoid\nbrightening effects by the Sun, the Moon, clouds, and other astronomical\nsources (the Galaxy and zodiacal light). The natural NSB has been calculated\nfrom the percentiles for 44 different photometers by applying these new\nfilters. The pristine night sky was measured to change with an amplitude of 0.1\nmag/arcsec$^2$ in all the photometers, which is suggested to be due to NSB\nvariations on scales of up to months and to be compatible with semiannual\noscillations. We report the systematic observation of short-time variations in\nNSB on the vast majority of the nights and find these to be related to airglow\nevents forming above the mesosphere.\n"}
{"abstract": "  In this paper, the soft gluon resummation effect in double heavy quarkonium\nproduction at the LHC is studied. By applying the transverse momentum dependent\nfactorization formalism, the large logarithms introduced by the small total\ntransverse momentum of heavy quarkonium pair final state system, are resummed\nto all orders in the expansion of the strong interaction coupling at the\nNext-to-Leading Logarithm accuracy. We also compare our result with the LHC\ndata. We find that the distribution shape predicted by resummation calculation\nis consistent with experimental data very well. Since this process is mainly\ninitiated by gluon fusion, it supplies us an important channel to study the\ngluon parton property in hadrons.\n"}
{"abstract": "  This paper presents a machine learning approach for the automatic\nidentification of rip currents with breaking waves. Rip currents are dangerous\nfast moving currents of water that result in many deaths by sweeping people out\nto sea. Most people do not know how to recognize rip currents in order to avoid\nthem. Furthermore, efforts to forecast rip currents are hindered by lack of\nobservations to help train and validate hazard models. The presence of web cams\nand smart phones have made video and still imagery of the coast ubiquitous and\nprovide a potential source of rip current observations. These same devices\ncould aid public awareness of the presence of rip currents. What is lacking is\na method to detect the presence or absence of rip currents from coastal\nimagery. This paper provides expert labeled training and test data sets for rip\ncurrents. We use Faster-RCNN and a custom temporal aggregation stage to make\ndetections from still images or videos with higher measured accuracy than both\nhumans and other methods of rip current detection previously reported in the\nliterature.\n"}
{"abstract": "  The increased penetration of Electric Vehicles (EVs) in the transportation\nsector has increased the requirement of Fast Charging Direct Current (FCDC)\nstations to meet customer's speedy charging requirements. However, both\ncharging stations and EVs connection to the communication infrastructure as\nwell as the power grid makes it vulnerable to cyber attacks. In this paper, the\nvulnerability of the EV charging process is initially studied. We then show how\na botnet of compromised EVs and FCDC stations can be utilized to launch cyber\nattacks on the power grid resulting in an increase in the load at a specific\ntime. The effect of such attacks on the distribution network in terms of line\ncongestion and voltage limit violations is investigated. Moreover, the effect\nof the botnet of the transmission network is also studied.\n"}
{"abstract": "  We propose an alternate approach to quantifying how well language models\nlearn natural language: we ask how well they match the statistical tendencies\nof natural language. To answer this question, we analyze whether text generated\nfrom language models exhibits the statistical tendencies present in the\nhuman-generated text on which they were trained. We provide a framework--paired\nwith significance tests--for evaluating the fit of language models to these\ntrends. We find that neural language models appear to learn only a subset of\nthe tendencies considered, but align much more closely with empirical trends\nthan proposed theoretical distributions (when present). Further, the fit to\ndifferent distributions is highly-dependent on both model architecture and\ngeneration strategy. As concrete examples, text generated under the nucleus\nsampling scheme adheres more closely to the type--token relationship of natural\nlanguage than text produced using standard ancestral sampling; text from LSTMs\nreflects the natural language distributions over length, stopwords, and symbols\nsurprisingly well.\n"}
{"abstract": "  The Galactic halo is supposed to form from merging with nearby dwarf\ngalaxies. In order to probe different components of the Galactic halo, we have\napplied the Gaussian Mixture Models method to a selected sample of metal poor\nstars with [Fe/H] $< -0.7$ dex in the APOGEE DR16 catalogue based on\nfour-parameters, metallicity, [Mg/Fe] ratio and spatial velocity\n(\\textit{$V_R$, $V_\\phi$}). Nine groups are identified with four from the halo\n(group 1, 3, 4 and 5), one from the thick disk (group 6), one from the thin\ndisk (group 8) and one from dwarf galaxies (group 7) by analyzing their\ndistributions in the ([M/H], [Mg/Fe]), ($V_R$, $V_\\phi$), (\\textit{Zmax},\n\\textit{eccentricity}), (\\textit{Energy}, \\textit{Lz}) and ([Mg/Mn], [Al/Fe])\ncoordinates. The rest two groups are respectively caused by observational\neffect (group 9) and the cross section component (group 2) between the thin\ndisk and the thick disk. It is found that in the extremely outer accreted halo\n(group 1), stars born in the Milky Way can not be distinguished from those\naccreted from other galaxies either chemically or kinematically. In the\nintermediate metallicity of $-$1.6 $<$ [Fe/H] $< -0.7$ dex, the accreted halo\nmainly composed of the Gaia-Enceladus-Sausage substructure (group 5), which can\nbe easily distinguished from group 4 (the in-situ halo group) in both chemical\nand kinematic space. Some stars of group 4 may come from the disk and some disk\nstars can be scattered to high orbits by resonant effects as shown in the\n\\textit{Zmax} versus Energy coordinate. We also displayed the spatial\ndistribution of main components of the halo and the ratio of accreted\ncomponents do not show clear relation to Galactic radius.\n"}
{"abstract": "  We present deep, narrowband imaging of the nearby spiral galaxy M101 and its\ngroup environment to search for star-forming dwarf galaxies and outlying HII\nregions. Using the Burrell Schmidt telescope, we target the brightest emission\nlines of star-forming regions, H$\\alpha$, H$\\beta$, and [OIII], to detect\npotential outlying star-forming regions. Our survey covers $\\sim$6 square\ndegrees around M101, and we detect objects in emission down to an H$\\alpha$\nflux level of $5.7 \\times 10^{-17}$ erg s$^{-1}$ cm$^{-2}$ (equivalent to a\nlimiting SFR of $1.7 \\times 10^{-6}$ $M_\\odot$ yr$^{-1}$ at the distance of\nM101). After careful removal of background contaminants and foreground M stars,\nwe detect 19 objects in emission in all three bands, and 8 objects in emission\nin H$\\alpha$ and [OIII]. We compare the structural and photometric properties\nof the detected sources to Local Group dwarf galaxies and star-forming galaxies\nin the 11HUGS and SINGG surveys. We find no large population of outlying HII\nregions or undiscovered star-forming dwarfs in the M101 Group, as most sources\n(93%) are consistent with being M101 outer disk HII regions. Only two sources\nwere associated with other galaxies: a faint star-forming satellite of the\nbackground galaxy NGC 5486, and a faint outlying HII region near the M101\ncompanion NGC 5474. We also find no narrowband emission associated with\nrecently discovered ultradiffuse galaxies and starless HI clouds near M101. The\nlack of any hidden population of low luminosity star-forming dwarfs around M101\nsuggests a rather shallow faint end slope (as flat as $\\alpha \\sim -1.0$) for\nthe star-forming luminosity function in the M101 Group. We discuss our results\nin the context of tidally-triggered star formation models and the interaction\nhistory of the M101 Group.\n"}
{"abstract": "  Human learners can readily understand speech, or a melody, when it is\npresented slower or faster than usual. Although deep convolutional neural\nnetworks (CNNs) are extremely powerful in extracting information from time\nseries, they require explicit training to generalize to different time scales.\nThis paper presents a deep CNN that incorporates a temporal representation\ninspired by recent findings from neuroscience. In the mammalian brain, time is\nrepresented by populations of neurons with temporal receptive fields.\nCritically, the peaks of the receptive fields form a geometric series, such\nthat the population codes a set of temporal basis functions over log time.\nBecause memory for the recent past is a function of log time, rescaling the\ninput results in translation of the memory. The Scale-Invariant Temporal\nHistory Convolution network (SITHCon) builds a convolutional layer over this\nlogarithmically-distributed temporal memory. A max-pool operation results in a\nnetwork that is invariant to rescalings of time modulo edge effects. We compare\nperformance of SITHCon to a Temporal Convolution Network (TCN). Although both\nnetworks can learn classification and regression problems on both univariate\nand multivariate time series f(t), only SITHCon generalizes to rescalings\nf(at). This property, inspired by findings from contemporary neuroscience and\nconsistent with findings from cognitive psychology, may enable networks that\nlearn with fewer training examples, fewer weights and that generalize more\nrobustly to out of sample data.\n"}
{"abstract": "  Emergency Departments (EDs) overcrowding is a well recognized worldwide\nphenomenon. The consequences range from long waiting times for visits and\ntreatment of patients up to life-threatening health conditions. The\ninternational community is devoting greater and greater efforts to analyze this\nphenomenon aiming at reducing waiting times, improving the quality of the\nservice. Within this framework, we propose a Discrete Event Simulation (DES)\nmodel to study the patient flows through a medium-size ED located in a region\nof Central Italy recently hit by a severe earthquake. In particular, our aim is\nto simulate unusual ED conditions, corresponding to critical events (like a\nnatural disaster) that cause a sudden spike in the number of patient arrivals.\nThe availability of detailed data concerning the ED processes enabled to build\nan accurate DES model and to perform extensive scenario analyses. The model\nprovides a valid decision support system for the ED managers also in defining\nspecific emergency plans to be activated in case of mass casualty disasters.\n"}
{"abstract": "  We consider an arbitrary metric graph, to which we glue another graph with\nedges of lengths proportional to $\\varepsilon$, where $\\varepsilon$ is a small\npositive parameter. On such graph, we consider a general self-adjoint second\norder differential operator $\\mathcal{H}_\\varepsilon$ with varying coefficients\nsubject to general vertex conditions; all coefficients in differential\nexpression and vertex conditions are supposed to be analytic in $\\varepsilon$.\nWe introduce a special operator on a certain graph obtained by rescaling the\naforementioned small edges and assume that it has no embedded eigenvalues at\nthe threshold of its essential spectrum. Under such assumption, we show that\ncertain parts of the resolvent of $\\mathcal{H}_\\varepsilon$ are analytic in\n$\\varepsilon$. This allows us to represent the resolvent of\n$\\mathcal{H}_\\varepsilon$ by an uniformly converging Taylor-like series and its\npartial sums can be used for approximating the resolvent up to an arbitrary\npower of $\\varepsilon$. In particular, the zero-order approximation reproduces\nrecent convergence results by G. Berkolaiko, Yu. Latushkin, S. Sukhtaiev and by\nC. Cacciapuoti, but we additionally show that next-to-leading terms in\n$\\varepsilon$-expansions of the coefficients in the differential expression and\nvertex conditions can contribute to the limiting operator producing the Robin\npart at the vertices, to which small edges are incident. We also discuss\npossible generalizations of our model including both the cases of a more\ngeneral geometry of the small parts of the graph and a non-analytic\n$\\varepsilon$-dependence of the coefficients in the differential expression and\nvertex conditions.\n"}
{"abstract": "  In this work, we consider the following elliptic partial differential\nequations: \\begin{equation*} \\left\\{ \\begin{aligned}{}\n  - b_{ij} \\; \\frac{\\partial^{2} w}{\\partial x_{i} \\partial x_{j}} &= g\n  \\;\\;\\; \\text{in} \\;\\; \\Omega, w &= 0 \\;\\;\\;\\text{on} \\;\\partial \\Omega,\n\\end{aligned} \\right. \\end{equation*} \\noindent where the domain $\\Omega\n\\subset \\mathbb{R}^{n}$ is convex, the matrix $\\big(b_{ij}\\big)_{n \\times n}$\nsatisfies the uniform ellipticity conditions. For $g$ in the scaling critical\nLorentz space $ L(n,\\; 1)(\\Omega)$, we establish boundary differentiability of\nsolutions to the above problem. We also prove $C^{\\mathrm{Log-Lip}}$ regularity\nestimate at a boundary point in the case when $g \\in L^{n}(\\Omega)$.\n"}
{"abstract": "  Recent researches in data assimilation lead to the introduction of the\nparametric Kalman filter (PKF): an implementation of the Kalman filter, where\nthe covariance matrices are approximated by a parameterized covariance model.\nIn the PKF, the dynamics of the covariance during the forecast step relies on\nthe prediction of the covariance parameters. Hence, the design of the parameter\ndynamics is crucial while it can be tedious to do this by hand. This\ncontribution introduces a python package, SymPKF, able to compute PKF dynamics\nfor univariate statistics and when the covariance model is parameterized from\nthe variance and the local anisotropy of the correlations. The ability of\nSymPKF to produce the PKF dynamics is shown on a non-linear diffusive advection\n(Burgers equation) over a 1D domain and the linear advection over a 2D domain.\nThe computation of the PKF dynamics is performed at a symbolic level, but an\nautomatic code generator is also introduced to perform numerical simulations. A\nfinal multivariate example illustrates the potential of SymPKF to go beyond the\nunivariate case.\n"}
{"abstract": "  In this paper we study the multiplicity and concentration of positive\nsolutions for the following $(p, q)$-Laplacian problem: \\begin{equation*}\n\\left\\{ \\begin{array}{ll} -\\Delta_{p} u -\\Delta_{q} u +V(\\varepsilon x)\n\\left(|u|^{p-2}u + |u|^{q-2}u\\right) = f(u) &\\mbox{ in } \\mathbb{R}^{N}, \\\\\nu\\in W^{1, p}(\\mathbb{R}^{N})\\cap W^{1, q}(\\mathbb{R}^{N}), \\quad u>0 \\mbox{ in\n} \\mathbb{R}^{N}, \\end{array} \\right. \\end{equation*} where $\\varepsilon>0$ is\na small parameter, $1< p<q<N$, $\\Delta_{r}u=\\mbox{div}(|\\nabla u|^{r-2}\\nabla\nu)$, with $r\\in \\{p, q\\}$, is the $r$-Laplacian operator,\n$V:\\mathbb{R}^{N}\\rightarrow \\mathbb{R}$ is a continuous function satisfying\nthe global Rabinowitz condition, and $f:\\mathbb{R}\\rightarrow \\mathbb{R}$ is a\ncontinuous function with subcritical growth. Using suitable variational\narguments and Ljusternik-Schnirelmann category theory, we investigate the\nrelation between the number of positive solutions and the topology of the set\nwhere $V$ attains its minimum for small $\\varepsilon$.\n"}
{"abstract": "  Building cross-lingual voice conversion (VC) systems for multiple speakers\nand multiple languages has been a challenging task for a long time. This paper\ndescribes a parallel non-autoregressive network to achieve bilingual and\ncode-switched voice conversion for multiple speakers when there are only\nmono-lingual corpora for each language. We achieve cross-lingual VC between\nMandarin speech with multiple speakers and English speech with multiple\nspeakers by applying bilingual bottleneck features. To boost voice cloning\nperformance, we use an adversarial speaker classifier with a gradient reversal\nlayer to reduce the source speaker's information from the output of encoder.\nFurthermore, in order to improve speaker similarity between reference speech\nand converted speech, we adopt an embedding consistency loss between the\nsynthesized speech and its natural reference speech in our network.\nExperimental results show that our proposed method can achieve high quality\nconverted speech with mean opinion score (MOS) around 4. The conversion system\nperforms well in terms of speaker similarity for both in-set speaker conversion\nand out-set-of one-shot conversion.\n"}
{"abstract": "  By using electrochemical etching, we fabricated conductive ultrathin SrVO3\n(SVO) films that exhibited metallic behavior down to 3 monolayers (ML). From an\nobserved systematic change in transport properties with decreasing film\nthickness, it was found that the disorder in the films remained nearly\nunchanged during etching, and only the thickness was reduced. This is in\ncontrast to the insulating behavior found for as-deposited SVO ultrathin films.\nFor the etched films, the electron mobility at 200 K decreased with decreasing\nfilm thickness below 10 ML, originating from an increased scattering rate and\nelectron effective mass near the metal-insulator transition. A slight upturn in\nthe resistivity and a positive magnetoresistance at low temperatures were\ntypically observed for the etched films down to 3 ML, which was explained by\nweak anti-localization of electrons in a weakly disordered metal.\n"}
{"abstract": "  We study the non-parametric isotonic regression problem for bivariate\nelicitable functionals that are given as an elicitable univariate functional\nand its Bayes risk. Prominent examples for functionals of this type are (mean,\nvariance) and (Value-at-Risk, Expected Shortfall), where the latter pair\nconsists of important risk measures in finance. We present our results for\ntotally ordered covariates but extenstions to partial orders are given in the\nappendix.\n"}
{"abstract": "  Characterizing the predicted environments of dwarf galaxies like the Large\nMagellanic Cloud (LMC) is becoming increasingly important as next generation\nsurveys push sensitivity limits into this low-mass regime at cosmological\ndistances. We study the environmental effects of LMC-mass halos ($M_{200m} \\sim\n10^{11}$ M$_\\odot$) on their populations of satellites ($M_\\star \\geq 10^4$\nM$_\\odot$) using a suite of zoom-in simulations from the Feedback In Realistic\nEnvironments (FIRE) project. Our simulations predict significant hot coronas\nwith $T\\sim10^5$ K and $M_\\text{gas}\\sim10^{9.5}$ M$_\\odot$. We identify\nsignatures of environmental quenching in dwarf satellite galaxies, particularly\nfor satellites with intermediate mass ($M_\\star = 10^{6-7}$ M$_\\odot$). The gas\ncontent of such objects indicates ram-pressure as the likely quenching\nmechanism, sometimes aided by star formation feedback. Satellites of LMC-mass\nhosts replicate the stellar mass dependence of the quiescent fraction found in\nsatellites of MW mass hosts (i.e. that the quiescent fraction increases as\nstellar mass decreases). Satellites of LMC-mass hosts have a wider variety of\nquenching times when compared to the strongly bi-modal distribution of\nquenching times of nearby centrals. Finally, we identify significant tidal\nstellar structures around four of our six LMC-analogs, suggesting that stellar\nstreams may be common. These tidal features originated from satellites on close\norbits, extend to $\\sim$80 kpc from the central galaxy, and contain\n$\\sim10^{6-7}$ M$_\\odot$~of stars.\n"}
{"abstract": "  Optical metasurfaces have been heralded as the platform to integrate multiple\nfunctionalities in a compact form-factor, potentially replacing bulky\ncomponents. A central stepping stone towards realizing this promise is the\ndemonstration of multifunctionality under several constraints (e.g. at multiple\nincident wavelengths and/or angles) in a single device -- an achievement being\nhampered by design limitations inherent to single-layer planar geometries.\nHere, we propose a general framework for the inverse design of volumetric 3D\nmetaoptics via topology optimization, showing that even few-wavelength thick\ndevices can achieve high-efficiency multifunctionality. We embody our framework\nin multiple closely-spaced patterned layers of a low-index polymer. We\nexperimentally demonstrate our approach with an inverse-designed 3d-printed\nlight concentrator working at five different non-paraxial angles of incidence.\nOur framework paves the way towards realizing multifunctional ultra-compact 3D\nnanophotonic devices.\n"}
{"abstract": "  Recently, significant progress has been made on semantic segmentation.\nHowever, the success of supervised semantic segmentation typically relies on a\nlarge amount of labelled data, which is time-consuming and costly to obtain.\nInspired by the success of semi-supervised learning methods in image\nclassification, here we propose a simple yet effective semi-supervised learning\nframework for semantic segmentation. We demonstrate that the devil is in the\ndetails: a set of simple design and training techniques can collectively\nimprove the performance of semi-supervised semantic segmentation significantly.\nPrevious works [3, 27] fail to employ strong augmentation in pseudo label\nlearning efficiently, as the large distribution change caused by strong\naugmentation harms the batch normalisation statistics. We design a new batch\nnormalisation, namely distribution-specific batch normalisation (DSBN) to\naddress this problem and demonstrate the importance of strong augmentation for\nsemantic segmentation. Moreover, we design a self correction loss which is\neffective in noise resistance. We conduct a series of ablation studies to show\nthe effectiveness of each component. Our method achieves state-of-the-art\nresults in the semi-supervised settings on the Cityscapes and Pascal VOC\ndatasets.\n"}
{"abstract": "  Given a group $G$ and an endomorphism $\\varphi$ of $G$, two elements $x, y\n\\in G$ are said to be $\\varphi$-conjugate if $x = gy \\varphi(g)^{-1}$ for some\n$g \\in G$. The number of equivalence classes for this relation is the\nReidemeister number $R(\\varphi)$ of $\\varphi$. The set $\\{R(\\psi) \\mid \\psi \\in\n\\mathrm{Aut}(G)\\}$ is called the Reidemeister spectrum of $G$. We investigate\nReidemeister numbers and spectra on direct products of finitely many groups and\ndetermine what information can be derived from the individual factors.\n"}
{"abstract": "  Mobile communication via high-altitude platforms operating in the\nstratosphere is an idea that has been on the table for decades. In the past few\nyears, however, with recent advances in technology and parallel progress in\nstandardization and regulatory bodies like 3GPP and ITU, these ideas have\ngained considerable momentum. In this article, we present a comprehensive\noverview of HIBS - High Altitude Platform Stations as IMT Base Stations. We lay\nout possible use cases and summarize the current status of the development,\nfrom a technological point of view as well as from standardization in 3GPP, and\nregarding spectrum aspects. We then present preliminary system level simulation\nresults to shed light on the performance of HIBS. We conclude with pointing out\nseveral directions for future research.\n"}
{"abstract": "  If gravitational perturbations are quantized into gravitons in analogy with\nthe electromagnetic field and photons, the resulting graviton interactions\nshould lead to an entangling interaction between massive objects. We suggest a\ntest of this prediction. To do this, we introduce the concept of interactive\nquantum information sensing. This novel sensing protocol is tailored to\nprovable verification of weak dynamical entanglement generation between a pair\nof systems. We show that this protocol is highly robust to typical thermal\nnoise sources. The sensitivity can moreover be increased both using an initial\nthermal state and/or an initial phase of entangling via a non-gravitational\ninteraction. We outline a concrete implementation testing the ability of the\ngravitational field to generate entanglement between an atomic interferometer\nand mechanical oscillator. Preliminary numerical estimates suggest that\nnear-term devices could feasibly be used to perform the experiment.\n"}
{"abstract": "  Accelerating Entrepreneurial Decision-Making Through Hybrid Intelligence\nDESIGN PARADIGMS AND PRINCIPLES FOR DECISIONAL GUIDANCE IN ENTREPRENEURSHIP\n"}
{"abstract": "  Traditional numerical methods for calculating matrix eigenvalues are\nprohibitively expensive for high-dimensional problems. Iterative random\nsparsification methods allow for the estimation of a single dominant eigenvalue\nat reduced cost by leveraging repeated random sampling and averaging. We\npresent a general approach to extending such methods for the estimation of\nmultiple eigenvalues and demonstrate its performance for several benchmark\nproblems in quantum chemistry.\n"}
{"abstract": "  Impacts of dust grains on spacecraft are known to produce typical impulsive\nsignals in the voltage waveform recorded at the terminals of electric antennas.\nSuch signals are routinely detected by the Time Domain Sampler (TDS) system of\nthe Radio and Plasma Waves (RPW) instrument aboard Solar Orbiter. We\ninvestigate the capabilities of RPW in terms of interplanetary dust studies and\npresent the first analysis of dust impacts recorded by this instrument. We\ndiscuss previously developed models of voltage pulses generation after a dust\nimpact onto a spacecraft and present the relevant technical parameters for\nSolar Orbiter RPW as a dust detector. Then we present the statistical analysis\nof the dust impacts recorded by RPW/TDS from April 20th, 2020 to February 27th,\n2021 between 0.5 AU and 1 AU. The study shows that the dust population studied\npresents a radial velocity component directed outward from the Sun, the order\nof magnitude of which can be roughly estimated as $v_{r, dust} \\simeq 50$\nkm.$s^{-1}$. This is consistent with the flux of impactors being dominated by\n$\\beta$-meteoroids. We estimate the cumulative flux of these grains at 1 AU to\nbe roughly $F_\\beta \\simeq 8\\times 10^{-5} $ m$^{-2}$s$^{-1}$, for particles of\nradius $r \\gtrsim 100$ nm. The power law index $\\delta$ of the cumulative mass\nflux of the impactors is evaluated by two differents methods (direct\nobservations of voltage pulses and indirect effect on the impact rate\ndependency on the impact speed). Both methods give a result $\\delta \\simeq\n0.3-0.4$. Solar Orbiter RPW proves to be a suitable instrument for\ninterplanetary dust studies. These first results are promising for the\ncontinuation of the mission, in particular for the in-situ study of the dust\ncloud outside the ecliptic plane, which Solar Orbiter will be the first\nspacecraft to explore.\n"}
{"abstract": "  Combining two classical notions in extremal combinatorics, the study of\nRamsey-Tur\\'an theory seeks to determine, for integers $m\\le n$ and $p \\leq q$,\nthe number $\\mathsf{RT}_p(n,K_q,m)$, which is the maximum size of an $n$-vertex\n$K_q$-free graph in which every set of at least $m$ vertices contains a $K_p$.\n  Two major open problems in this area from the 80s ask: (1) whether the\nasymptotic extremal structure for the general case exhibits certain periodic\nbehaviour, resembling that of the special case when $p=2$; (2) constructing\nanalogues of Bollob\\'as-Erd\\H{o}s graphs with densities other than $1/2$.\n  We refute the first conjecture by witnessing asymptotic extremal structures\nthat are drastically different from the $p=2$ case, and address the second\nproblem by constructing Bollob\\'as-Erd\\H{o}s-type graphs using high dimensional\ncomplex spheres with all rational densities. Some matching upper bounds are\nalso provided.\n"}
{"abstract": "  In order to further develop and implement novel drift scan imaging\nexperiments to undertake wide field, high time resolution surveys for\nmillisecond optical transients, an appropriate telescope drive system is\nrequired. This paper describes the development of a simple and inexpensive\nhardware and software system to monitor, characterise, and correct the primary\ncategory of telescope drive errors, periodic errors due to imperfections in the\ndrive and gear chain. A model for the periodic errors is generated from direct\nmeasurements of the telescope drive shaft rotation, verified by comparison to\nastronomical measurements of the periodic errors. The predictive model is\ngenerated and applied in real-time in the form of corrections to the drive\nrate. A demonstration of the system shows that that inherent periodic errors of\npeak-to-peak amplitude ~100'' are reduced to below the seeing limit of ~3''.\nThis demonstration allowed an estimate of the uncertainties on the transient\nsensitivity timescales of the prototype survey of Tingay & Joubert (2021), with\nthe nominal timescale sensitivity of 21 ms revised to be in the range of 20 -\n22 ms, which does not significantly affect the results of the experiment. The\ncorrection system will be adopted into the final version of high cadence\nimaging experiment, which is currently under construction. The correction\nsystem is inexpensive (<$A100) and composed of readily available hardware, and\nis readily adaptable to other applications. Design details and codes are\ntherefore made publicly available.\n"}
{"abstract": "  We show that an invertible bilateral weighted shift is strongly structurally\nstable if and only if it has the shadowing property. We also exhibit a\nK{\\\"o}the sequence space supporting a frequently hypercyclic weighted shift,\nbut no chaotic weighted shifts.\n"}
{"abstract": "  Simple, controllable models play an important role to learn how to manipulate\nand control quantum resources. We focus here on quantum non-Markovianity and\nmodel the evolution of open quantum systems by quantum renewal processes. This\nclass of quantum dynamics provides us with a phenomenological approach to\ncharacterise dynamics with a variety of non-Markovian behaviours, here\ndescribed in terms of the trace distance between two reduced states. By\nadopting a trajectory picture for the open quantum system evolution, we analyse\nhow non-Markovianity is influenced by the constituents defining the quantum\nrenewal process, namely the time-continuous part of the dynamics, the type of\njumps and the waiting time distributions. We focus not only on the mere value\nof the non-Markovianity measure, but also on how different features of the\ntrace distance evolution are altered, including times and number of revivals.\n"}
{"abstract": "  Security is of critical importance for the Internet of Things (IoT). Many IoT\ndevices are resource-constrained, calling for lightweight security protocols.\nPhysical unclonable functions (PUFs) leverage integrated circuits' variations\nto produce responses unique for individual devices, and hence are not\nreproducible even by the manufacturers. Implementable with simplistic circuits\nof thousands of transistors and operable with low energy, Physical unclonable\nfunctions are promising candidates as security primitives for\nresource-constrained IoT devices. Arbiter PUFs (APUFs) are a group of\ndelay-based PUFs which are highly lightweight in resource requirements but\nsuffer from high susceptibility to machine learning attacks. To defend APUF\nvariants against machine learning attacks, we introduce challenge input\ninterface, which incurs low resource overhead. With the interface, experimental\nattack study shows that all tested PUFs have substantially improved their\nresistance against machine learning attacks, rendering interfaced APUF variants\npromising candidates for security critical applications.\n"}
{"abstract": "  We present explicit estimates of right and left tails and exact (up to\nuniversal, multiplicative constants) estimates of tails and moments of hitting\ntimes of Bessel processes. The latter estimates are obtained from more general\nestimates of moments and tails established for convolutions of elementary\nmixtures of exponential distributions.\n"}
{"abstract": "  Modern information retrieval systems, including web search, ads placement,\nand recommender systems, typically rely on learning from user feedback. Click\nmodels, which study how users interact with a ranked list of items, provide a\nuseful understanding of user feedback for learning ranking models. Constructing\n\"right\" dependencies is the key of any successful click model. However,\nprobabilistic graphical models (PGMs) have to rely on manually assigned\ndependencies, and oversimplify user behaviors. Existing neural network based\nmethods promote PGMs by enhancing the expressive ability and allowing flexible\ndependencies, but still suffer from exposure bias and inferior estimation. In\nthis paper, we propose a novel framework, Adversarial Imitation Click Model\n(AICM), based on imitation learning. Firstly, we explicitly learn the reward\nfunction that recovers users' intrinsic utility and underlying intentions.\nSecondly, we model user interactions with a ranked list as a dynamic system\ninstead of one-step click prediction, alleviating the exposure bias problem.\nFinally, we minimize the JS divergence through adversarial training and learn a\nstable distribution of click sequences, which makes AICM generalize well across\ndifferent distributions of ranked lists. A theoretical analysis has indicated\nthat AICM reduces the exposure bias from $O(T^2)$ to $O(T)$. Our studies on a\npublic web search dataset show that AICM not only outperforms state-of-the-art\nmodels in traditional click metrics but also achieves superior performance in\naddressing the exposure bias and recovering the underlying patterns of click\nsequences.\n"}
{"abstract": "  We associate to a graphon $\\gamma$ the sequence of $W$-random graphs\n$(G_n(\\gamma))_{n \\geq 1}$. We say that the graphon is singular if, for any\nfinite graph $F$, the homomorphism density $t(F,G_n(\\gamma))$ has a variance of\norder $O(n^{-2})$. This behavior is singular because generically, the density\nof a fixed finite graph $F$ in a $W$-random graph has a variance of order\n$O(n^{-1})$. We conjecture that the only singular graphons are the constant\ngraphons $\\gamma_p$ with $p \\in [0,1]$, corresponding to the Erd\\H{o}s-R\\'enyi\nrandom graphs $G(n,p)$. In this paper, we investigate the general properties of\nthe singular graphons, and we show that they share many properties with the\nErd\\H{o}s-R\\'enyi random graphs. In particular, if $\\gamma$ is a singular\ngraphon, then the scaled densities\n$n(t(F,G_n(\\gamma))-\\mathbb{E}[t(F,G_n(\\gamma))])$ converge in joint\ndistribution. This generalises the central limit theorem satisfied by the\nErd\\H{o}s-R\\'enyi random graphs $G(n,p)$; however, the limiting distribution\nmight be non-Gaussian if the conjecture does not hold. We also establish an\nequation satisfied by the characteristic polynomial of the Laplacian of the\ngraph $G_n(\\gamma)$ associated to a singular graphon; this opens the way to a\nspectral approach of the conjecture.\n"}
{"abstract": "  Many businesses and industries nowadays rely on large quantities of time\nseries data making time series forecasting an important research area. Global\nforecasting models that are trained across sets of time series have shown a\nhuge potential in providing accurate forecasts compared with the traditional\nunivariate forecasting models that work on isolated series. However, there are\ncurrently no comprehensive time series archives for forecasting that contain\ndatasets of time series from similar sources available for the research\ncommunity to evaluate the performance of new global forecasting algorithms over\na wide variety of datasets. In this paper, we present such a comprehensive time\nseries forecasting archive containing 20 publicly available time series\ndatasets from varied domains, with different characteristics in terms of\nfrequency, series lengths, and inclusion of missing values. We also\ncharacterise the datasets, and identify similarities and differences among\nthem, by conducting a feature analysis. Furthermore, we present the performance\nof a set of standard baseline forecasting methods over all datasets across\neight error metrics, for the benefit of researchers using the archive to\nbenchmark their forecasting algorithms.\n"}
{"abstract": "  Operads are algebraic devices offering a formalization of the concept of\noperations with several inputs and one output. Such operations can be naturally\ncomposed to form bigger and more complex ones. Coming historically from\nalgebraic topology, operads intervene now as important objects in computer\nscience and in combinatorics. The theory of operads, together with the\nalgebraic setting and the tools accompanying it, promises advances in these two\nareas. On the one hand, operads provide a useful abstraction of formal\nexpressions, and also, provide connections with the theory of rewrite systems.\nOn the other hand, a lot of operads involving combinatorial objects highlight\nsome of their properties and allow to discover new ones.\n  This book presents the theory of nonsymmetric operads under a combinatorial\npoint of view. It portrays the main elements of this theory and the links it\nmaintains with several areas of computer science and combinatorics. A lot of\nexamples of operads appearing in combinatorics are studied and some\nconstructions relating operads with known algebraic structures are presented.\nThe modern treatment of operads consisting in considering the space of formal\npower series associated with an operad is developed. Enrichments of\nnonsymmetric operads as colored, cyclic, and symmetric operads are reviewed.\n  This text is addressed to any computer scientist or combinatorist who looks a\ncomplete and a modern description of the theory of nonsymmetric operads.\nEvenly, this book is intended to an audience of algebraists who are looking for\nan original point of view fitting in the context of combinatorics.\n"}
{"abstract": "  The (open-high-low-close) OHLC data is the most common data form in the field\nof finance and the investigate object of various technical analysis. With\nincreasing features of OHLC data being collected, the issue of extracting their\nuseful information in a comprehensible way for visualization and easy\ninterpretation must be resolved. The inherent constraints of OHLC data also\npose a challenge for this issue. This paper proposes a novel approach to\ncharacterize the features of OHLC data in a dataset and then performs dimension\nreduction, which integrates the feature information extraction method and\nprincipal component analysis. We refer to it as the pseudo-PCA method.\nSpecifically, we first propose a new way to represent the OHLC data, which will\nfree the inherent constraints and provide convenience for further analysis.\nMoreover, there is a one-to-one match between the original OHLC data and its\nfeature-based representations, which means that the analysis of the\nfeature-based data can be reversed to the original OHLC data. Next, we develop\nthe pseudo-PCA procedure for OHLC data, which can effectively identify\nimportant information and perform dimension reduction. Finally, the\neffectiveness and interpretability of the proposed method are investigated\nthrough finite simulations and the spot data of China's agricultural product\nmarket.\n"}
{"abstract": "  We introduce an algebraic system which can be used as a model for spaces with\ngeodesic paths between any two of their points. This new algebraic structure is\nbased on the notion of mobility algebra which has recently been introduced as a\nmodel for the unit interval of real numbers. We show that there is a strong\nconnection between modules over a ring and affine mobility spaces over a\nmobility algebra. However, geodesics in general fail to be affine thus giving\nrise to the new algebraic structure of mobility space. We show that the so\ncalled formula for spherical linear interpolation, which gives geodesics on the\nn-sphere, is an example of a mobility space over the unit interval mobility\nalgebra.\n"}
{"abstract": "  GANs largely increases the potential impact of generative models. Therefore,\nwe propose a novel knowledge transfer method for generative models based on\nmining the knowledge that is most beneficial to a specific target domain,\neither from a single or multiple pretrained GANs. This is done using a miner\nnetwork that identifies which part of the generative distribution of each\npretrained GAN outputs samples closest to the target domain. Mining effectively\nsteers GAN sampling towards suitable regions of the latent space, which\nfacilitates the posterior finetuning and avoids pathologies of other methods,\nsuch as mode collapse and lack of flexibility. Furthermore, to prevent\noverfitting on small target domains, we introduce sparse subnetwork selection,\nthat restricts the set of trainable neurons to those that are relevant for the\ntarget dataset. We perform comprehensive experiments on several challenging\ndatasets using various GAN architectures (BigGAN, Progressive GAN, and\nStyleGAN) and show that the proposed method, called MineGAN, effectively\ntransfers knowledge to domains with few target images, outperforming existing\nmethods. In addition, MineGAN can successfully transfer knowledge from multiple\npretrained GANs.\n"}
{"abstract": "  The Peace-Athabasca Delta (PAD) of northwestern Alberta is one of the largest\ninland freshwater deltas in the world, laying at the confluence of the Peace\nand Athabasca Rivers. The PAD is recognized as a having unique ecological\nsignificance and periodic ice jam flooding from both rivers is an important\nfeature of its current ecology. Past studies have debated whether a change in\nice jam flood (IJF) frequency on the Peace River has recently occurred, and\nwhat factors might be driving any perceived changes. This study contributes to\nthis debate by addressing two questions: (1) what factors are most predictive\nof Peace River IJFs, and (2) how might climate change impact IJF frequency?\nThis work starts with a physically-based conceptual model of the necessary\nconditions for a large Peace River IJF, and the factors that indicate whether\nthose conditions are met. Logistic regression is applied to the historical\nflood record to determine which combination of hydroclimatic and riverine\nfactors best predict IJFs and the uncertainty in those relationships given the\navailable data. Winter precipitation and temperature are most predictive of\nPeace River IJFs, while freeze-up elevation contains little predictive power\nand is not closely related to IJF occurrence. The best logistic regression\nmodel is forced with downscaled climate change scenarios from multiple climate\nmodels to project IJF frequency for a variety of plausible futures. Parametric\nuncertainty in the best logistic regression model is propagated into the\nprojections using a parametric bootstrap to sample many plausible statistical\nmodels. Although there is variability across emissions scenarios and climate\nmodels, all projections indicate that the frequency of Peace River IJFs is\nlikely to decrease substantially in the coming decades, and that average\nwaiting times between future IJFs will likely surpass recent experience.\n"}
{"abstract": "  We investigate the impact on cosmological observables of $f(Q)$-gravity, a\nspecific class of modified gravity models in which gravity is described by the\nnon-metricity scalar, $Q$. In particular we focus on a specific model which is\nindistinguishable from the $\\Lambda$-cold-dark-matter ($\\Lambda$CDM) model at\nthe background level, while showing peculiar and measurable signatures at\nlinear perturbation level. These are attributed to a time-dependent Planck mass\nand are regulated by a single dimensionless parameter, $\\alpha$. In comparison\nto the $\\Lambda$CDM model, we find for positive values of $\\alpha$ a suppressed\nmatter power spectrum and lensing effect on the Cosmic Microwave Background\nradiation (CMB) angular power spectrum and an enhanced integrated-Sachs-Wolfe\ntail of CMB temperature anisotropies. The opposite behaviors are present when\nthe $\\alpha$ parameter is negative. We also investigate the modified\nGravitational Waves (GWs) propagation and show the prediction of the GWs\nluminosity distance compared to the standard electromagnetic one. Finally, we\ninfer the accuracy on the free parameter of the model with standard sirens at\nfuture GWs detectors.\n"}
{"abstract": "  Sakurai et al. (J Comput Phys, 2019) presented a flux-based volume\npenalization (VP) approach for imposing inhomogeneous Neumann boundary\nconditions on embedded interfaces. The flux-based VP method modifies the\ndiffusion coefficient of the original elliptic (Poisson) equation and uses a\nflux-forcing function as a source term in the equation to impose the Neumann\nboundary conditions. As such, the flux-based VP method can be easily\nincorporated into existing fictitious domain codes. Sakurai et al. relied on an\nanalytical construction of flux-forcing functions, which limits the\npracticality of the approach. Because of the analytical approach taken in the\nprior work, only (spatially) constant flux values on simple interfaces were\nconsidered. In this paper, we present a numerical technique for constructing\nflux-forcing functions for arbitrarily complex boundaries. The imposed flux\nvalues are also allowed to vary spatially in our approach. Furthermore, the\nflux-based VP method is extended to include (spatially varying) Robin boundary\nconditions, which makes the flux-based VP method even more general. We consider\nseveral two- and three-dimensional test examples to access the spatial accuracy\nof the numerical solutions. The method is also used to simulate flux-driven\nthermal convection in a concentric annular domain. We formally derive the\nflux-based volume penalized Poisson equation satisfying Neumann/Robin boundary\ncondition in strong form; such a derivation was not presented in Sakurai et\nal., where the equation first appeared for the Neumann problem. The derivation\nreveals that the flux-based VP approach relies on a surface delta function to\nimpose inhomogeneous Neumann/Robin boundary conditions. However, explicit\nconstruction of the delta function is not necessary for the flux-based VP\nmethod, which makes it different from other diffuse domain equations presented\nin the literature.\n"}
{"abstract": "  Two extremely red main-belt asteroids: 203 Pompeja and 269 Justitia, were\nidentified from combined visible and near-infrared spectroscopic observations\ncollected at the IRTF and SAO observatories. These two asteroids have a redder\nspectral slope than any other D-type body, which are the reddest objects in the\nasteroid belt, and similar to RR and IR-class objects found in the outer Solar\nSystem among trans-Neptunian objects and Centaurs. Spectroscopic results\nsuggest the presence of complex organic materials on the surface layer of these\nasteroids, implying that they could have formed in the vicinity of Neptune and\nbeen transplanted to the main belt region during a phase of planetary\nmigration. 203 Pompeia is the only very red asteroid known so far among the\n~250 bodies with diameter larger than 110 km (i.e. presumably structurally\nintact) found in the asteroid belt. These discoveries add another piece of\nevidence that the main asteroid belt hosts a population of bodies that were\nformed in the outskirt of the Solar System.\n"}
{"abstract": "  Freestanding semiconductor nanowires have opened up new possibilities for\nsemiconductor devices, enabling geometries, material combinations and strain\nstates which were not previously possible. Along these lines, spontaneous\nbending in asymmetric core-shell nanowire heterostructures has recently been\nproposed as a means to realize previously unimagined device geometries, novel\nstrain-gradient engineering and bottom-up device fabrication. The synthesis of\nthese nanostructures exploits the nanowire geometry and the directionality of\nthe shell deposition process. Here, we explore the underlying mechanisms of\nthis bending process by modeling the evolution of nanowires during asymmetric\nshell deposition. We show how bending can lead to dramatic local shell\nthickness, curvature and strain variations along the length of the nanowire,\nand we elucidate the dependence of shell growth and bending on parameters such\nas the core and shell dimensions and materials, and angle of incidence of the\ndeposition source. In addition, deposition shadowing by neighboring nanowires\nis explored. We show that shadowing can easily be employed to connect nanowire\npairs, which could be used to fabricate novel nanowire sensors. Model results\nare compared with GaAs-InP and GaAs-(Al,In)As core-shell nanowire growth\nexperiments. These results can be used to guide future experiments and to help\npave the way to bent nanowire devices.\n"}
{"abstract": "  We report on the Shubnikov de Haas (SdH) oscillations in the quasi\ntwo-dimensional molecular conductor $\\alpha-$(BETS)$_{2}$I$_{3}$ [BETS:\nbis(ethylenedithio)tetraselenafulvalene] laminated on polyimide films at 1.7 K.\nFrom the SdH phase factor, we verified experimentally that the material is in\nthe Dirac fermion phase under pressure. $\\alpha-$(BETS)$_{2}$I$_{3}$ is in the\nvicinity of the phase transition between strongly correlated insulating and\nDirac fermion phases, and is a possible candidate for an ambient-pressure\nmolecular Dirac fermion system. However, the SdH oscillations indicate that the\nBerry phase is zero at ambient pressure. Under pressure, a $\\pi$ Berry phase\nemerges when the metal-insulator crossover is almost suppressed at $\\sim$0.5\nGPa. The results contrast those for the pioneering molecular Dirac fermion\nsystem $\\alpha-$(BEDT-TTF)$_{2}$I$_{3}$ [BEDT-TTF:\nbis(ethylenedithio)tetrathiafulvalene] in which Dirac fermions and\nsemiconducting behavior are simultaneously observed.\n"}
{"abstract": "  Variability is a general property of accretion discs and their associated\njets. We introduce a semi-analytic model for particle acceleration and radio\njet/lobe evolution and explore the effect of Myr timescale jet variability on\nthe particles accelerated by an AGN jet. Our work is motivated by the need for\nlocal powerful ultrahigh energy cosmic ray (UHECR) sources and evidence for\nvariability in AGN and radio galaxies. Our main results are: i) UHECR and\nnonthermal radiative luminosities track the jet power but with a response set\nby the escape and cooling times, respectively; ii) jet variability produces\nstructure in the electron, synchrotron and UHECR spectra that deviates from\nthat produced for a constant jet power - in particular, spectral hardening\nfeatures may be signatures of variability; iii) the cutoff in the integrated CR\nspectrum is stretched out due to the variation in jet power (and, consequently,\nmaximum CR energy). The resulting spectrum is the convolution of the jet power\ndistribution and the source term. We derive an approximate form for a\nlog-normal distribution of powers; iv) we introduce the idea of $\\sim 10$ GeV\n'proxy electrons' that are cooling at the same rate that UHECRs of rigidity 10\nEV are escaping from the source, and determine the corresponding photon\nfrequencies that probe escaping UHECRs. Our results demonstrate the link\nbetween the history of an astrophysical particle accelerator and its particle\ncontents, nonthermal emission and UHECR spectrum, with consequences for\nobservations of radio galaxies and UHECR source models.\n"}
{"abstract": "  Optimal transport (OT), which provides a distance between two probability\ndistributions by considering their spatial locations, has been applied to\nwidely diverse applications. Computing an OT problem requires solution of\nlinear programming with tight mass-conservation constraints. This requirement\nhinders its application to large-scale problems. To alleviate this issue, the\nrecently proposed relaxed-OT approach uses a faster algorithm by relaxing such\nconstraints. Its effectiveness for practical applications has been\ndemonstrated. Nevertheless, it still exhibits slow convergence. To this end,\naddressing a convex semi-relaxed OT, we propose a fast block-coordinate\nFrank-Wolfe (BCFW) algorithm, which gives sparse solutions. Specifically, we\nprovide their upper bounds of the worst convergence iterations, and equivalence\nbetween the linearization duality gap and the Lagrangian duality gap. Three\nfast variants of the proposed BCFW are also proposed. Numerical evaluations in\ncolor transfer problem demonstrate that the proposed algorithms outperform\nstate-of-the-art algorithms across different settings.\n"}
{"abstract": "  We are concerned below with the characterization in a unital commutative real\nBanach algebra $\\mathbb{A}$ of continuous solutions of the\nGo{\\l}\\k{a}b-Schinzel functional equation (below), the general Popa groups they\ngenerate and the associated Goldie functional equation. This yields general\nstructure theorems involving both linear and exponential homogeneity in\n$\\mathbb{A}$ for both these functional equations and also explict forms, in\nterms of the recently developed theory of multi-Popa groups [BinO3,4], both for\nthe ring $C[0,1]$ and for the case of $\\mathbb{R}^{d}$ with componentwise\nproduct, clarifying the context of recent developments in [RooSW]. The case\n$\\mathbb{A}=\\mathbb{C}$ provides a new viewpoint on continuous complex-valued\nsolutions of the primary equation by distinguishing analytic from real-analytic\nones.\n"}
{"abstract": "  The LPMWPC can be used as the {\\Delta}E detector for the low-energy charged\nparticle detection. In order to increase the transmittance, the wires were\nadopted as the cathode. This work investigated the LPMWPC signal\ncharacteristics of this configuration and measured the gas gain with a mixture\nof 90%Ar and 10% CO2 from 1E3 to 1E5 Pa. From the test, the second pulse after\nthe avalanche signal was observed, which proved to be caused by the ions'\ndrifting near the cathode wire.\n"}
{"abstract": "  We show that the position-momentum duality offers a transparent\ninterpretation of the band geometry at the topological band crossings. Under\nthis duality, the band geometry with Berry connection is dual to the\nfree-electron motion under gauge field. This identifies the trace of quantum\nmetric as the dual energy in momentum space. The band crossings with Berry\ndefects thus induce the dual energy quantization in the trace of quantum\nmetric. For a $\\mathbb Z$ nodal-point or nodal-surface semimetal, a dual Landau\nlevel quantization occurs owing to the Berry charge. Meanwhile, a nodal-loop\nsemimetal exhibits a Berry vortex line, leading to the quantized dual\nrotational energy about the nodal loop. A $\\mathbb Z_2$ monopole brings about\nanother dual rotational energy, which originates from the link with an\nadditional nodal line. Nontrivial band geometry generically induces finite\nspread in the Wannier functions. While the spread manifests a quantized lower\nbound in a $\\mathbb Z$ nodal-point or nodal-surface semimetal, logarithmic\ndivergence occurs in a nodal-loop semimetal. The band geometry at the band\ncrossings may be probed experimentally by a periodic-drive measurement.\n"}
{"abstract": "  In this paper, we introduce new generalizations of higher-order Changhee of\nthe first and second kind. Moreover, we derive some new results for these\nnumbers and polynomials. Furthermore, some interesting special cases of the\ngeneralized higher-order Changhee numbers and polynomials are deduced.\n"}
{"abstract": "  It is known that the right key of a Kashiwara-Nakashima tableau can be\ncomputed using the Lecouvey-Sheats symplectic jeu de taquin. Motivated by\nWillis' direct way of computing type A right keys, we also give a way of\ncomputing symplectic right keys without the use of jeu de taquin.\n"}
{"abstract": "  Atom-interferometer gravitational-wave (GW) observatory, as a new design of\nground-based GW detector for the near future, is sensitive at a relatively low\nfrequency for GW observations. Taking the proposed atom interferometer Zhaoshan\nLong-baseline Atom Interferometer Gravitation Antenna (ZAIGA), and its\nillustrative upgrade (Z+) as examples, we investigate how the atom\ninterferometer will complement ground-based laser interferometers in testing\nthe gravitational dipole radiation from binary neutron star (BNS) mergers. A\ntest of such kind is important for a better understanding of the strong\nequivalence principle laying at the heart of Einstein's general relativity. To\nobtain a statistically sound result, we sample BNS systems according to their\nmerger rate and population, from which we study the expected bounds on the\nparameterized dipole radiation parameter $B$. Extracting BNS parameters and the\ndipole radiation from the combination of ground-based laser interferometers and\nthe atom-interferometer ZAIGA/Z+, we are entitled to obtain tighter bounds on\n$B$ by a few times to a few orders of magnitude, compared to ground-based laser\ninterferometers alone, ultimately reaching the levels of $|B| \\lesssim 10^{-9}$\n(with ZAIGA) and $|B| \\lesssim 10^{-10}$ (with Z+).\n"}
{"abstract": "  Grid turbulence is investigated using cross-correlation digital Particle\nImage Velocimetry (PIV) over a range of Taylor Reynolds Number (Re{\\lambda})\nfrom 5 to 44. Instantaneous velocity is measured directly and vorticity and\nvelocity gradients are obtained indirectly. Measurements are taken at various\ndownstream locations from the generating grid. Probability distribution\nfunctions (PDFs) are calculated for the fluctuating component of the velocity,\nthe spatial velocity gradients and vorticity. The PDF of the velocity\nfluctuations has a Gaussian distribution while velocity gradients and vorticity\nare found to have non-Gaussian PDF distributions. The structure of the flow is\ninvestigated by calculating spatial autocorrelations for all measured and\nderived data. The spatial velocity autocorrelations differ from previous\nexperimental measurements of grid turbulence, most of which have been\ndetermined from single-point measurements. This difference is believed to be\ndue to differences in the way in which the measurements are made, and\nparticularly to differences in the spatial size of the experimental domain.\n"}
{"abstract": "  Photosynthesis is the primary energy production process of ecosystem.\nThroughout the green plants, two different types of chlorophylls, Chl-a and -b,\nare bound to the light-harvesting complex for photosystem II. While the\nstructure and functional characteristics of the two types of chlorophylls have\nbeen well documented as individual, the features of the coexistence has yet to\nbe understood clearly. Here, we present a principled framework for discerning\nand quantifying the chlorophylls' role during the excited energy transfer. Our\napproach combines network science and Monte Carlo Markov analysis based on the\nFoerster resonance energy transfer rates between all the chlorophyll pairs in\nthe light-harvesting antennae for photosystem II. We find that the excitation\nenergy can be rapidly confined to the specific chlorophyll clusters in the\nlight-harvesting complex, where excess energy dissipation possibly takes place,\nonly when both chlorophyll species are involved. Comparing to other models with\nhypothetical chlorophyll compositions, we show that the light harvesting system\nwith natural chlorophyll a and b ratio is the most advantageous for the\nphotosystem II supercomplex to balance the light-harvesting and photoprotection\ncapabilities.\n"}
{"abstract": "  A class of models of long waves in dispersive media with coupled quadratic\nnonlinearities on a periodic domain $\\mathbb{T}$ are studied. We used two\ndistributed controls, supported in $\\omega\\subset\\mathbb{T}$ and assumed to be\ngenerated by a linear feedback law conserving the \"mass\" (or \"volume\"), to\nprove global control results. The first result, using spectral analysis,\nguarantees that the system in consideration is locally controllable in\n$H^s(\\mathbb{T})$, for $s\\geq0$. After that, by certain properties of Bourgain\nspaces we show a property of global exponential stability. This property\ntogether with the local exact controllability ensures for the first time in the\nliterature that long waves in nonlinear dispersive media are globally exactly\ncontrollable in large time. Precisely, our analysis relies strongly on the\nbilinear estimates using the Fourier restriction spaces in two different\ndispersions that will guarantee a global control result for coupled systems of\nthe Korteweg-de Vries type. This result, of independent interest in the area of\ncontrol of coupled dispersive systems, provides a necessary first step for the\nstudy of global control properties to the coupled dispersive systems in\nperiodic domains.\n"}
{"abstract": "  Using the unification of the chiral SU(3) model and QCD sum rules, we deduce\nthe in-medium properties of $K^\\pm_1$ meson. Within the chiral SU(3) model,\nmedium modified gluon and quark condensates are evaluated through their\ninteractions with the scalar fields ($\\sigma$, $\\zeta$, $\\delta$, and $\\chi$).\nThese condensates are further used as input in the Borel transformed equations\nof QCD sum rules to evaluate the in-medium mass of strange $K^\\pm_1$ meson. The\nin-medium property of the above meson can be used to study the restoration of\nchiral symmetry in nuclear matter.\n"}
{"abstract": "  As two valuable quantum resources, Einstein-Podolsky-Rosen entanglement and\nsteering play important roles in quantum-enhanced communication protocols.\nDistributing such quantum resources among multiple remote users in a network is\na crucial precondition underlying various quantum tasks. We experimentally\ndemonstrate the deterministic distribution of two- and three-mode Gaussian\nentanglement and steering by transmitting separable states in a network\nconsisting of a quantum server and multiple users. In our experiment, entangled\nstates are not prepared solely by the quantum server, but are created among\nindependent users during the distribution process. More specifically, the\nquantum server prepares separable squeezed states and applies classical\ndisplacements on them before spreading out, and users simply perform local\nbeam-splitter operations and homodyne measurements after they receive separable\nstates. We show that the distributed Gaussian entanglement and steerability are\nrobust against channel loss. Furthermore, one-way Gaussian steering is achieved\namong users that is useful for further directional or highly asymmetric quantum\ninformation processing.\n"}
{"abstract": "  In the present paper, we prove that any finite non-trivial irreducible module\nover a rank two Lie conformal algebra $\\mathcal{H}$ is of rank one. We also\ndescribe the actions of $\\mathcal{H}$ on its finite irreducible modules\nexplicitly. Moreover, we show that all finite non-trivial irreducible modules\nof finite Lie conformal algebras whose semisimple quotient is the Virasoro Lie\nconformal algebra are of rank one.\n"}
{"abstract": "  Predicting all applicable labels for a given image is known as multi-label\nclassification. Compared to the standard multi-class case (where each image has\nonly one label), it is considerably more challenging to annotate training data\nfor multi-label classification. When the number of potential labels is large,\nhuman annotators find it difficult to mention all applicable labels for each\ntraining image. Furthermore, in some settings detection is intrinsically\ndifficult e.g. finding small object instances in high resolution images. As a\nresult, multi-label training data is often plagued by false negatives. We\nconsider the hardest version of this problem, where annotators provide only one\nrelevant label for each image. As a result, training sets will have only one\npositive label per image and no confirmed negatives. We explore this special\ncase of learning from missing labels across four different multi-label image\nclassification datasets for both linear classifiers and end-to-end fine-tuned\ndeep networks. We extend existing multi-label losses to this setting and\npropose novel variants that constrain the number of expected positive labels\nduring training. Surprisingly, we show that in some cases it is possible to\napproach the performance of fully labeled classifiers despite training with\nsignificantly fewer confirmed labels.\n"}
{"abstract": "  The Cayley-Klein (CK) formalism is applied to the real algebra ${so}(5)$ by\nmaking use of four graded contraction parameters describing in a unified\nsetting 81 Lie algebras, which cover the (anti-)de Sitter, Poincar\\'e,\nNewtonian and Carrollian algebras. Starting with the Drinfel'd-Jimbo real Lie\nbialgebra for ${so}(5)$ together with its Drinfel'd double structure, we obtain\nthe corresponding CK bialgebra and the CK $r$-matrix coming from a Drinfel'd\ndouble. As a novelty, we construct the (first-order) noncommutative CK spaces\nof points, lines, 2-planes and 3-hyperplanes, studying their structural\nproperties. By requiring to deal with real structures, it comes out that there\nexist 63 specific real Lie bialgebras together with their sets of four\nnoncommutative spaces. Furthermore, we find 14 classical $r$-matrices coming\nfrom Drinfel'd doubles, obtaining new results for the de Sitter ${so}(4,1)$ and\nanti-de Sitter ${so}(3,2)$ and for some of their contractions. These geometric\nresults are exhaustively applied onto the (3+1)D kinematical algebras, not only\nconsidering the usual (3+1)D spacetime but also the 6D space of lines. We\nestablish different assignations between the geometrical CK generators and the\nkinematical ones which convey physical identifications for the CK contraction\nparameters in terms of the cosmological constant/curvature $\\Lambda$ and speed\nof light $c$. We finally obtain four classes of kinematical $r$-matrices\ntogether with their noncommutative spacetimes and spaces of lines, comprising\nall $\\kappa$-deformations as particular cases.\n"}
{"abstract": "  In this paper, we revisit math word problems~(MWPs) from the cross-lingual\nand multilingual perspective. We construct our MWP solvers over pretrained\nmultilingual language models using sequence-to-sequence model with copy\nmechanism. We compare how the MWP solvers perform in cross-lingual and\nmultilingual scenarios. To facilitate the comparison of cross-lingual\nperformance, we first adapt the large-scale English dataset MathQA as a\ncounterpart of the Chinese dataset Math23K. Then we extend several English\ndatasets to bilingual datasets through machine translation plus human\nannotation. Our experiments show that the MWP solvers may not be transferred to\na different language even if the target expressions have the same operator set\nand constants. But for both cross-lingual and multilingual cases, it can be\nbetter generalized if problem types exist on both source language and target\nlanguage.\n"}
{"abstract": "  The black hole information paradox has been hotly debated for the last few\ndecades, without full resolution. This makes it desirable to find analogs of\nthis paradox in simple and experimentally accessible systems, whose resolutions\nmay shed light on this long-standing and fundamental problem. Here we identify\nand resolve an apparent \"information paradox\" in a quantum Hall interface\nbetween the Halperin-331 and Pfaffian states. Information carried by pseudospin\ndegree of freedom of the Abelian 331 quasiparticles gets scrambled when they\ncross the interface to enter non-Abelian Pfaffian state, and becomes\ninaccessible to local measurements; in this sense the Pfaffian region is an\nanalog of black hole interior while the interface plays a role similar to its\nhorizon. We demonstrate that the \"lost\" information gets recovered once the\n\"black hole\" evaporates and the quasiparticles return to the 331 region, albeit\nin a highly entangled form. Such recovery is quantified by the Page curve of\nthe entropy carried by these quasiparticles, which are analogs of Hawking\nradiation.\n"}
{"abstract": "  Modern deep Super-Resolution (SR) networks have established themselves as\nvaluable techniques in image reconstruction and enhancement. However, these\nnetworks are normally trained and tested on benchmark image data that lacks the\ntypical image degrading noise present in real images. In this paper, we test\nthe feasibility of using deep SR in real remote sensing payloads by assessing\nSR performance in reconstructing realistically degraded satellite images. We\ndemonstrate that a state-of-the-art SR technique called Enhanced Deep\nSuper-Resolution Network (EDSR), without domain specific pre-training, can\nrecover encoded pixel data on images with poor ground sampling distance,\nprovided the ground resolved distance is sufficient. However, this recovery\nvaries amongst selected geographical types. Our results indicate that custom\ntraining has potential to further improve reconstruction of overhead imagery,\nand that new satellite hardware should prioritise optical performance over\nminimising pixel size as deep SR can overcome a lack of the latter but not the\nformer.\n"}
{"abstract": "  Clock configuration within constrained general-purpose microcontrollers takes\na key role in tuning performance, power consumption, and timing accuracy of\napplications in the Internet of Things (IoT). Subsystems governing the\nunderlying clock tree must nonetheless cope with a huge parameter space,\ncomplex dependencies, and dynamic constraints. Manufacturers expose the\nunderlying functions in very diverse ways, which leads to specialized\nimplementations of low portability. In this paper, we propose FlexClock, an\napproach for generic online clock reconfiguration on constrained IoT devices.\nWe argue that (costly) generic clock configuration of general purpose computers\nand powerful mobile devices need to slim down to the lower end of the device\nspectrum. In search of a generalized solution, we identify recurring patterns\nand building blocks, which we use to decompose clock trees into independent,\nreusable components. With this segmentation we derive an abstract\nrepresentation of vendor-specific clock trees, which then can be dynamically\nreconfigured at runtime. We evaluate our implementation on common hardware. Our\nmeasurements demonstrate how FlexClock significantly improves peak power\nconsumption and energy efficiency by enabling dynamic voltage and frequency\nscaling (DVFS) in a platform-agnostic way.\n"}
{"abstract": "  Top-N recommendation is a challenging problem because complex and sparse\nuser-item interactions should be adequately addressed to achieve high-quality\nrecommendation results. The local latent factor approach has been successfully\nused with multiple local models to capture diverse user preferences with\ndifferent sub-communities. However, previous studies have not fully explored\nthe potential of local models, and failed to identify many small and coherent\nsub-communities. In this paper, we present Local Collaborative Autoencoders\n(LOCA), a generalized local latent factor framework. Specifically, LOCA adopts\ndifferent neighborhood ranges at the training and inference stages. Besides,\nLOCA uses a novel sub-community discovery method, maximizing the coverage of a\nunion of local models and employing a large number of diverse local models. By\nadopting autoencoders as the base model, LOCA captures latent non-linear\npatterns representing meaningful user-item interactions within sub-communities.\nOur experimental results demonstrate that LOCA is scalable and outperforms\nstate-of-the-art models on several public benchmarks, by 2.99~4.70% in Recall\nand 1.02~7.95% in NDCG, respectively.\n"}
{"abstract": "  We study the mod $p^r$ Milnor $K$-groups of $p$-adically complete and\n$p$-henselian rings, establishing in particular a Nesterenko-Suslin style\ndescription in terms of the Milnor range of syntomic cohomology. In the case of\nsmooth schemes over complete discrete valuation rings we prove the mod $p^r$\nGersten conjecture for Milnor $K$-theory locally in the Nisnevich topology. In\ncharacteristic $p$ we show that the Bloch-Kato-Gabber theorem remains true for\nvaluation rings, and for regular formal schemes in a pro sense.\n"}
{"abstract": "  Deterministic and randomized, row-action and column-action linear solvers\nhave become increasingly popular owing to their simplicity, low computational\nand memory complexities, and ease of composition with other techniques.\nMoreover, in order to achieve high-performance, such solvers must often be\nadapted to the given problem structure and to the hardware platform on which\nthe problem will be solved. Unfortunately, determining whether such adapted\nsolvers will converge to a solution has required equally unique analyses. As a\nresult, adapted, reliable solvers are slow to be developed and deployed. In\nthis work, we provide a general set of assumptions under which such adapted\nsolvers are guaranteed to converge with probability one, and provide worst case\nrates of convergence. As a result, we can provide practitioners with guidance\non how to design highly adapted, randomized or deterministic, row-action or\ncolumn-action linear solvers that are also guaranteed to converge.\n"}
{"abstract": "  Deep neural networks (DNNs) have been found to be vulnerable to adversarial\nexamples. Adversarial examples are malicious images with visually imperceptible\nperturbations. While these carefully crafted perturbations restricted with\ntight $\\Lp$ norm bounds are small, they are still easily perceivable by humans.\nThese perturbations also have limited success rates when attacking black-box\nmodels or models with defenses like noise reduction filters. To solve these\nproblems, we propose Demiguise Attack, crafting ``unrestricted'' perturbations\nwith Perceptual Similarity. Specifically, we can create powerful and\nphotorealistic adversarial examples by manipulating semantic information based\non Perceptual Similarity. Adversarial examples we generate are friendly to the\nhuman visual system (HVS), although the perturbations are of large magnitudes.\nWe extend widely-used attacks with our approach, enhancing adversarial\neffectiveness impressively while contributing to imperceptibility. Extensive\nexperiments show that the proposed method not only outperforms various\nstate-of-the-art attacks in terms of fooling rate, transferability, and\nrobustness against defenses but can also improve attacks effectively. In\naddition, we also notice that our implementation can simulate illumination and\ncontrast changes that occur in real-world scenarios, which will contribute to\nexposing the blind spots of DNNs.\n"}
{"abstract": "  Let $n,k,t$ be positive integers. What is the maximum number of arcs in a\ndigraph on $n$ vertices in which there are at most $t$ distinct walks of length\n$k$ with the same endpoints? In this paper, we prove that the maximum number is\nequal to $n(n-1)/2$ and the extremal digraph are the transitive tournaments\nwhen $k\\ge n-1\\ge \\max\\{2t+1,2\\left\\lceil \\sqrt{2t+9/4}+1/2\\right\\rceil+3\\}$.\nBased on this result, we may determine the maximum numbers and the extremal\ndigraphs for $k\\ge \\max\\{2t+1,2\\left\\lceil \\sqrt{2t+9/4}+1/2\\right\\rceil+3\\}$\nand $n$ is sufficiently large, which generalises the existing results. A\nconjecture is also presented.\n"}
{"abstract": "  We explore entanglement generation between multiple optically levitated\nnanospheres interacting with a common optical cavity via the Coherent\nScattering optomechanical interaction. We derive the many-particle Hamiltonian\ngoverning the unitary evolution of the system and show that it gives rise to\nquantum correlations among the various partitions of the setup, following a\nnon-Markovian dynamics of entanglement birth, death and revivals. We also\nconsider the effects of coupling the system to external environments and show\nthat under reasonable experimental conditions entanglement between the\nmechanical modes can survive even at room temperature. Its dependence upon the\nnumber of nanoparticles, their initial temperature and coupling strength is\nstudied. A numerical toolbox to simulate the closed and open dynamics of\nGaussian optomechanical states and their informational measures is developed.\n"}
{"abstract": "  The geometry of black hole spacetimes can be probed with exquisite precision\nin the gravitational-wave window, and possibly also in the optical regime. We\nstudy the accretion of bright spots -- objects which emit strongly in the\noptical or in gravitational waves -- by non-spinning black holes. The emission\nas the object falls down the hole is dominated by photons or gravitons orbiting\nthe light ring, causing the total luminosity to decrease exponentially as\n${\\cal L}_o\\sim e^{-t/(3\\sqrt{3}\\,M)}$. Late-time radiation is blueshifted, due\nto its having been emitted during the infall, trapped at the light ring and\nsubsequently re-emitted. These universal properties are a clear signature of\nthe existence of light rings in the spacetime, and not particularly sensitive\nto near horizon details.\n"}
{"abstract": "  One-shot neural architecture search (NAS) methods significantly reduce the\nsearch cost by considering the whole search space as one network, which only\nneeds to be trained once. However, current methods select each operation\nindependently without considering previous layers. Besides, the historical\ninformation obtained with huge computation cost is usually used only once and\nthen discarded. In this paper, we introduce a sampling strategy based on Monte\nCarlo tree search (MCTS) with the search space modeled as a Monte Carlo tree\n(MCT), which captures the dependency among layers. Furthermore, intermediate\nresults are stored in the MCT for the future decision and a better\nexploration-exploitation balance. Concretely, MCT is updated using the training\nloss as a reward to the architecture performance; for accurately evaluating the\nnumerous nodes, we propose node communication and hierarchical node selection\nmethods in the training and search stages, respectively, which make better uses\nof the operation rewards and hierarchical information. Moreover, for a fair\ncomparison of different NAS methods, we construct an open-source NAS benchmark\nof a macro search space evaluated on CIFAR-10, namely NAS-Bench-Macro.\nExtensive experiments on NAS-Bench-Macro and ImageNet demonstrate that our\nmethod significantly improves search efficiency and performance. For example,\nby only searching $20$ architectures, our obtained architecture achieves\n$78.0\\%$ top-1 accuracy with 442M FLOPs on ImageNet. Code (Benchmark) is\navailable at: \\url{https://github.com/xiusu/NAS-Bench-Macro}.\n"}
{"abstract": "  The sources of IceCube neutrinos are as yet unknown. The multi-messenger\nobservation of their emission in $\\gamma$-rays can be a guide to their\nidentification, as exemplified by the case of TXS 0506+056. We suggest a new\nmethod of searching for $\\gamma$-rays with Imaging Air Cherenkov Telescopes\nfrom sources in coincidence with possible astrophysical neutrinos. We propose\nthat searches of $\\gamma$-rays are extended, from the current practice of only\na few days, to up to one month from a neutrino alert. We test this strategy on\nsimulated sources modeled after the blazar \\emph{TXS 0506+056-like}, emitting\nneutrinos and $\\gamma$-rays via photohadronic interactions: the $\\gamma$-rays\nare subsequently reprocessed in the VHE range. Using MAGIC as a benchmark\nexample, we show that current Cherenkov Telescopes should be able to\ndetect$\\gamma$-ray counterparts to neutrino alerts with a rate of approximately\none per year. It has been proposed that the high-energy diffuse neutrino flux\ncan be explained by $\\sim$ 5\\% of all blazars flaring in neutrinos once every\n10 years, with a neutrino luminosity similar to that of TXS 0506+056 during the\n2014-2015 neutrino flare. The implementation of our strategy could lead, over a\ntimescale of one or few years, either to the detection of this subclass of\nblazars contributing to the diffuse neutrino flux, or to a constraint on this\nmodel.\n"}
{"abstract": "  Deep Q Network (DQN) firstly kicked the door of deep reinforcement learning\n(DRL) via combining deep learning (DL) with reinforcement learning (RL), which\nhas noticed that the distribution of the acquired data would change during the\ntraining process. DQN found this property might cause instability for training,\nso it proposed effective methods to handle the downside of the property.\nInstead of focusing on the unfavourable aspects, we find it critical for RL to\nease the gap between the estimated data distribution and the ground truth data\ndistribution while supervised learning (SL) fails to do so. From this new\nperspective, we extend the basic paradigm of RL called the Generalized Policy\nIteration (GPI) into a more generalized version, which is called the\nGeneralized Data Distribution Iteration (GDI). We see massive RL algorithms and\ntechniques can be unified into the GDI paradigm, which can be considered as one\nof the special cases of GDI. We provide theoretical proof of why GDI is better\nthan GPI and how it works. Several practical algorithms based on GDI have been\nproposed to verify the effectiveness and extensiveness of it. Empirical\nexperiments prove our state-of-the-art (SOTA) performance on Arcade Learning\nEnvironment (ALE), wherein our algorithm has achieved 9620.98% mean human\nnormalized score (HNS), 1146.39% median HNS and 22 human world record\nbreakthroughs (HWRB) using only 200M training frames. Our work aims to lead the\nRL research to step into the journey of conquering the human world records and\nseek real superhuman agents on both performance and efficiency.\n"}
{"abstract": "  Motivated by various possible generalizations of Taubes's \\(SW=Gr\\) theorem\n[T] to Floer-theoretic setting, we prove certain variants of Taubes's\nconvergence theorem in \\cite{T} (the first part of his proof of \\(SW=Gr\\)). In\nplace of the closed symplectic 4-manifold considered in [T], this article\nconsiders non-compact manifolds with cylindrical ends, equipped with a\nself-dual harmonic 2-form with non-degenerate zeroes. This extends and\nsimplifies some central technical ingredients of the author's prior work in\n[LT] and [KLT5]. Other expected applications include: extending the \\(HM=PFH\\)\ntheorem in [T] and the \\(HM=HF\\) theorem in [KLT1]-[KLT5] to TQFTs on both\nsides [L1]; definitions of large-perturbation Seiberg-Witten analogs of\nHeegaard Floer theory's link Floer homologies and link cobordism invariants.\n"}
{"abstract": "  Multimodal pre-training with text, layout, and image has achieved SOTA\nperformance for visually-rich document understanding tasks recently, which\ndemonstrates the great potential for joint learning across different\nmodalities. In this paper, we present LayoutXLM, a multimodal pre-trained model\nfor multilingual document understanding, which aims to bridge the language\nbarriers for visually-rich document understanding. To accurately evaluate\nLayoutXLM, we also introduce a multilingual form understanding benchmark\ndataset named XFUND, which includes form understanding samples in 7 languages\n(Chinese, Japanese, Spanish, French, Italian, German, Portuguese), and\nkey-value pairs are manually labeled for each language. Experiment results show\nthat the LayoutXLM model has significantly outperformed the existing SOTA\ncross-lingual pre-trained models on the XFUND dataset. The pre-trained\nLayoutXLM model and the XFUND dataset are publicly available at\nhttps://aka.ms/layoutxlm.\n"}
{"abstract": "  Symmetry-protected topological $\\left(SPT\\right)$ phases are gapped\nshort-range entangled states with symmetry $G$, which can be systematically\ndescribed by group cohomology theory. $SU(3)$ and $SU(2)\\times{U(1)}$ are\nconsidered as the basic groups of Quantum Chromodynamics and\nWeak-Electromagnetic unification, respectively. In two dimension $(2D)$,\nnonlinear-sigma models with a quantized topological Theta term can be used to\ndescribe nontrivial SPT phases. By coupling the system to a probe field and\nintegrating out the group variables, the Theta term becomes the effective\naction of Chern-Simons theory which can derive the response current density. As\na result, the current shows a spin Hall effect, and the quantized number of the\nspin Hall conductance of SPT phases $SU(3)$ and $SU(2)\\times{U(1)}$ are same.\nIn addition, relationships between $SU(3)$ and $SU(2)\\times{U(1)}$ which maps\n$SU(3)$ to $SU(2)$ with a rotation $U(1)$ will be given.\n"}
{"abstract": "  We present a novel method for real-time pose and shape reconstruction of two\nstrongly interacting hands. Our approach is the first two-hand tracking\nsolution that combines an extensive list of favorable properties, namely it is\nmarker-less, uses a single consumer-level depth camera, runs in real time,\nhandles inter- and intra-hand collisions, and automatically adjusts to the\nuser's hand shape. In order to achieve this, we embed a recent parametric hand\npose and shape model and a dense correspondence predictor based on a deep\nneural network into a suitable energy minimization framework. For training the\ncorrespondence prediction network, we synthesize a two-hand dataset based on\nphysical simulations that includes both hand pose and shape annotations while\nat the same time avoiding inter-hand penetrations. To achieve real-time rates,\nwe phrase the model fitting in terms of a nonlinear least-squares problem so\nthat the energy can be optimized based on a highly efficient GPU-based\nGauss-Newton optimizer. We show state-of-the-art results in scenes that exceed\nthe complexity level demonstrated by previous work, including tight two-hand\ngrasps, significant inter-hand occlusions, and gesture interaction.\n"}
{"abstract": "  Non-target screening consists in searching a sample for all present\nsubstances, suspected or unknown, with very little prior knowledge about the\nsample. This approach has been introduced more than a decade ago in the field\nof water analysis, together with dedicated compound identification tools, but\nis still very scarce for indoor and atmospheric trace gas measurements, despite\nthe clear need for a better understanding of the atmospheric trace gas\ncomposition.For a systematic detection of emerging trace gases in the\natmosphere, a new and powerful analytical method is gas chromatography (GC) of\npreconcentrated samples, followed by electron ionisation, high resolution mass\nspectrometry (EI-HRMS). In this work, we present data analysis tools to enable\nautomated fragment formula annotation for unknown compounds measured by\nGC-EI-HRMS. Based on co-eluting mass/charge fragments, we developed an\ninnovative data analysis method to reliably reconstruct the chemical formulae\nof the fragments, using efficient combinatorics and graph theory. The method\ndoes not require the presence of the molecular ion, which is absent in ~40% of\nEI spectra. Our method has been trained and validated on \\textgreater50\nhalocarbons and hydrocarbons, with 3 to 20 atoms and molar masses of 30 to 330\ng mol-1, measured with a mass resolution of approx.~3500. For 90% of the\ncompounds, more than 90% of the annotated fragment formulae are correct. Cases\nof wrong identification can be attributed to the scarcity of detected fragments\nper compound or the lack of isotopic constraint (no minor isotopocule\ndetected).Our method enables to reconstruct most probable chemical formulae\nindependently from spectral databases. Therefore, it demonstrates the\nsuitability of EI-HRMS data for non-target analysis and paves the way for the\nidentification of substances for which no EI mass spectrum is registered in\ndatabases. We illustrate the performances of our method for atmospheric trace\ngases and suggest that it may be well suited for many other types of samples.\nThe L-GPL licenced Python code is released under the name ALPINAC for\nALgorithmic Process for Identification of Non-targeted Atmospheric Compounds.\n"}
{"abstract": "  Collaborative filtering has been largely used to advance modern recommender\nsystems to predict user preference. A key component in collaborative filtering\nis representation learning, which aims to project users and items into a low\ndimensional space to capture collaborative signals. However, the scene\ninformation, which has effectively guided many recommendation tasks, is rarely\nconsidered in existing collaborative filtering methods. To bridge this gap, we\nfocus on scene-based collaborative recommendation and propose a novel\nrepresentation model SceneRec. SceneRec formally defines a scene as a set of\npre-defined item categories that occur simultaneously in real-life situations\nand creatively designs an item-category-scene hierarchical structure to build a\nscene-based graph. In the scene-based graph, we adopt graph neural networks to\nlearn scene-specific representation on each item node, which is further\naggregated with latent representation learned from collaborative interactions\nto make recommendations. We perform extensive experiments on real-world\nE-commerce datasets and the results demonstrate the effectiveness of the\nproposed method.\n"}
{"abstract": "  This paper formalizes constraint-based structure learning of the \"true\"\ncausal graph from observed data when unobserved variables are also existent. We\nprovide conditions for a \"natural\" family of constraint-based\nstructure-learning algorithms that output graphs that are Markov equivalent to\nthe causal graph. Under the faithfulness assumption, this natural family\ncontains all exact structure-learning algorithms. More importantly, we provide\nclear and testable assumptions, as an alternative to faithfulness, under which\nany natural structure-learning algorithm outputs Markov equivalent graphs to\nthe causal graph. We provide these definitions and results for the general\nclass of models under the assumption that the distribution is Markovian to the\ntrue causal graph, and we specialize the definitions and results for structural\ncausal models.\n"}
{"abstract": "  The dynamical critical exponent $z$ of natural swarms is calculated using the\nrenormalization group. To order $\\epsilon = 4-d$, a novel fixed point emerges,\nwhere both off-equilibrium activity and mode-coupling inertial interactions are\nrelevant. In three dimensions the critical exponent at the new fixed point is\n$z=1.3$, in fair agreement with experiments.\n"}
{"abstract": "  We explore the benefits of adapted gauges to small mass ratio binary black\nhole evolutions in the moving puncture formulation. We find expressions that\napproximate the late time behavior of the lapse and shift,\n$(\\alpha_0,\\beta_0)$, and use them as initial values for their evolutions. We\nalso use a position and black hole mass dependent damping term,\n$\\eta[\\vec{x}_1(t),\\vec{x}_2(t),m_1,m_2]$, in the shift evolution, rather than\na constant or conformal-factor dependent choice. We have found that this\nsubstantially reduces noise generation at the start of the numerical\nintegration and keeps the numerical grid stable around both black holes,\nallowing for more accuracy with lower resolutions. We test our choices for this\ngauge in detail in a case study of a binary with a 7:1 mass ratio, and then use\n15:1 and 32:1 binaries for a convergence study. Finally, we apply our new gauge\nto a 64:1 binary and a 128:1 binary to well cover the comparable and small mass\nratio regimes.\n"}
{"abstract": "  A rainbow matching in an edge-colored graph is a matching in which no two\nedges have the same color. The color degree of a vertex v is the number of\ndifferent colors on edges incident to v. Kritschgau [Electron. J. Combin.\n27(2020)] studied the existence of rainbow matchings in edge-colored graph G\nwith average color degree at least 2k, and proved some sufficient conditions\nfor a rainbow marching of size k in G. The sufficient conditions include that\n|V(G)|>=12k^2+4k, or G is a properly edge-colored graph with |V(G)|>=8k. In\nthis paper, we show that every edge-colored graph G with |V(G)|>=4k-4 and\naverage color degree at least 2k-1 contains a rainbow matching of size k. In\naddition, we also prove that every strongly edge-colored graph G with average\ndegree at least 2k-1 contains a rainbow matching of size at least k. The bound\nis sharp for complete graphs.\n"}
{"abstract": "  Modern machine learning models often employ a huge number of parameters and\nare typically optimized to have zero training loss; yet surprisingly, they\npossess near-optimal prediction performance, contradicting classical learning\ntheory. We examine how these benign overfitting phenomena occur in a two-layer\nneural network setting where sample covariates are corrupted with noise. We\naddress the high dimensional regime, where the data dimension $d$ grows with\nthe number $n$ of data points. Our analysis combines an upper bound on the bias\nwith matching upper and lower bounds on the variance of the interpolator (an\nestimator that interpolates the data). These results indicate that the excess\nlearning risk of the interpolator decays under mild conditions. We further show\nthat it is possible for the two-layer ReLU network interpolator to achieve a\nnear minimax-optimal learning rate, which to our knowledge is the first\ngeneralization result for such networks. Finally, our theory predicts that the\nexcess learning risk starts to increase once the number of parameters $s$ grows\nbeyond $O(n^2)$, matching recent empirical findings.\n"}
{"abstract": "  Starting from a general many-body fermionic Hamiltonian, we derive the\nequations of motion (EOM) for nucleonic propagators in a superfluid system. The\nresulting EOM is of the Dyson type formulated in the basis of Bogoliubov's\nquasiparticles. As the leading contributions to the dynamical kernel of this\nEOM in strongly-coupled regimes contain phonon degrees of freedom in various\nchannels, an efficient method of calculating phonon's characteristics is\nrequired to successfully model these kernels. The traditional quasiparticle\nrandom phase approximation (QRPA) solvers are typically used for this purpose\nin nuclear structure calculations, however, they become very prohibitive in\nnon-spherical geometries. In this work, by linking the notion of the\nquasiparticle-phonon vertex to the variation of the Bogoliubov's Hamiltonian,\nwe show that the recently developed finite-amplitude method (FAM) can be\nefficiently employed to compute the vertices within the FAM-QRPA. To illustrate\nthe validity of the method, calculations based on the relativistic\ndensity-dependent point-coupling Lagrangian are performed for the\nsingle-nucleon states in heavy and medium-mass nuclei with axial deformations.\nThe cases of $^{38}$Si and $^{250}$Cf are presented and discussed.\n"}
{"abstract": "  This study describes a method to detect hijacked journals based on the\nanalysis of the archives of clone journals. This approach is most effective in\ndiscovering a network of hijacked journals that have the same organizer(s).\nAnalysis of the archives of clone journals allowed to detect 62 URLs of\nhijacked journals. It also provided the possibility to predict two clone\nwebsites before they became operational. This study shows that most detected\nhijacked journals represent a network of clone journals organized by one or\nseveral fraudulent individuals. The information and content of nine legitimate\njournals were compromised in international and national scientometric\ndatabases.\n"}
{"abstract": "  The stellar mass-stellar metallicity relation (MZR) is an essential approach\nto probe the chemical evolution of galaxies. It reflects the balance between\ngalactic feedback and gravitational potential as a function of stellar mass.\nHowever, the current MZR of local dwarf satellite galaxies (M* <~ 10^8 Msun,\nmeasured from resolved stellar spectroscopy) may not be reconcilable with that\nof more massive galaxies (M* >~ 10^9.5 Msun, measured from integrated-light\nspectroscopy). Such a discrepancy may result from a systematic difference\nbetween the two methods, or it may indicate a break in the MZR around 10^9\nMsun. To address this question, we measured the stellar metallicity of NGC 147\nfrom integrated light using the Palomar Cosmic Web Imager (PCWI). We compared\nthe stellar metallicity estimates from integrated light with the measurements\nfrom resolved stellar spectroscopy and found them to be consistent within 0.1\ndex. On the other hand, the high-mass MZR overpredicts the metallicity by 0.6\ndex at the mass of NGC 147. Therefore, our results tentatively suggest that the\ndiscrepancy between the low-mass MZR and high-mass MZR should not be attributed\nto a systematic difference in techniques. Instead, real physical processes\ncause the transition in the MZR. In addition, we discovered a positive age\ngradient in the innermost region and a negative metallicity gradient from the\nresolved stars at larger radii, suggesting a possible outside-in formation of\nNGC 147.\n"}
{"abstract": "  The explosion of IoT and wearable devices determined a rising attention\ntowards energy harvesting as source for powering these systems. In this\ncontext, many applications cannot afford the presence of a battery because of\nsize, weight and cost issues. Therefore, due to the intermittent nature of\nambient energy sources, these systems must be able to save and restore their\nstate, in order to guarantee progress across power interruptions. In this work,\nwe propose a specialized backup/restore controller that dynamically tracks the\nmemory accesses during the execution of the program. The controller then\ncommits the changes to a snapshot in a Non-Volatile Memory (NVM) when a power\nfailure is detected. Our approach does not require complex hybrid memories and\ncan be implemented with standard components. % and integrated in any MCU with\nResults on a set of benchmarks show an average $8\\times$ reduction in backup\nsize. Thanks to our dedicated controller, the backup time is further reduced by\nmore than $100\\times$, with an area and power overhead of only 0.4\\% and 0.8\\%,\nrespectively, w.r.t. a low-end IoT node.\n"}
{"abstract": "  Based on crystal chemistry considerations and quantum density functional\ntheory ground state calculations, rhombohedral rh-C3 and hexagonal h-C6 carbon\nallotropes are proposed and energetically calculated as new stable ultra-hard\nphases likewise lonsdaleite. Along the two kinds of carbon in linear C2-C1-C2\nlattice, distorted tetrahedra C2(sp3) with an angle of 106.17{\\deg} (smaller\nthan ideal 109.4{\\deg}) and C1(sp)-like hybridizations are inferred from charge\ndensity projections. The calculated elastic constants point to a strong\nanisotropy of mechanical properties of rh-C3 and h-C6 with an exceptionally\nlarge C33 values (1636 GPa and 1610 GPa, respectively), exceeding that of\nlonsdaleite (1380 GPa), due to the presence of aligned tricarbon units along\nthe hexagonal c-axis. Both phases are characterized by large bulk moduli and\nhigh hardness values that are slightly less than those of lonsdaleite and\ndiamond. Weak metallic behavior of both new phases is identified from\nelectronic band structure calculations.\n"}
{"abstract": "  Voltage collapse is a type of blackout-inducing dynamic instability that\noccurs when power demand exceeds the maximum power that can be transferred\nthrough a network. The traditional (preventive) approach to avoid voltage\ncollapse is based on ensuring that the network never reaches its maximum\ncapacity. However, such an approach leads to inefficient use of network\nresources and does not account for unforeseen events. To overcome this\nlimitation, this paper seeks to initiate the study of voltage collapse\nstabilization, i.e., the design of load controllers aimed at stabilizing the\npoint of voltage collapse. We formulate the problem of voltage stability for a\nstar direct current network as a dynamic problem where each load seeks to\nachieve a constant power consumption by updating its conductance as the voltage\nchanges. We introduce a voltage collapse stabilization controller and show that\nthe high-voltage equilibrium is stabilized. More importantly, we are able to\nachieve proportional load shedding under extreme loading conditions. We further\nhighlight the key features of our controller using numerical illustrations.\n"}
{"abstract": "  An algorithm is proposed, analyzed, and tested experimentally for solving\nstochastic optimization problems in which the decision variables are\nconstrained to satisfy equations defined by deterministic, smooth, and\nnonlinear functions. It is assumed that constraint function and derivative\nvalues can be computed, but that only stochastic approximations are available\nfor the objective function and its derivatives. The algorithm is of the\nsequential quadratic optimization variety. A distinguishing feature of the\nalgorithm is that it allows inexact subproblem solutions to be employed, which\nis particularly useful in large-scale settings when the matrices defining the\nsubproblems are too large to form and/or factorize. Conditions are imposed on\nthe inexact subproblem solutions that account for the fact that only stochastic\nobjective gradient estimates are available. Convergence results in expectation\nare established for the method. Numerical experiments show that it outperforms\nan alternative algorithm that employs highly accurate subproblem solutions in\nevery iteration.\n"}
{"abstract": "  The motivation of the present study is to discuss the global (in time)\nexistence of small data solutions to the following semi-linear structurally\ndamped $\\sigma$-evolution models: \\begin{equation*}\n\\partial_{tt}u+(-\\Delta)^{\\sigma}u+(-\\Delta)^{\\sigma/2}\\partial_{t}u=\\left|u\\right|\n^{p}, \\ \\sigma\\geq 1, \\ \\ p>1, \\end{equation*} where the Cauchy data $(u(0,x),\n\\partial_{t}u(0,x))$ will be chosen from energy space on the base of $L^{q}$\nwith different additional $L^{m}$ regularity, namely \\begin{equation*}\nu(0,x)\\in H^{\\sigma,q}(\\mathbb{R}^{n})\\cap L^{m_{1}}(\\mathbb{R}^{n}) , \\ \\\n\\partial_{t}u(0,x)\\in L^{q}(\\mathbb{R}^{n})\\cap L^{m_{2}}(\\mathbb{R}^{n}), \\ \\\nq\\in(1,\\infty),\\ \\ m_{1}, m_{2}\\in [1,q). \\end{equation*} Our new results will\nshow that the critical exponent which guarantees the global (in time) existence\nis really affected by these different additional regularities and will take\n\\textit{two different values} under some restrictions on $m_{1}, m_{2}$, $q$,\n$\\sigma$ and the space dimension $n\\geq1$. Moreover, in each case, we have no\nloss of decay estimates of the unique solution with respect to the\ncorresponding linear models.\n"}
{"abstract": "  It was conjectured by Bott-Grove-Halperin that a compact simply connected\nRiemannian manifold $M$ with nonnegative sectional curvature is rationally\nelliptic. We confirm this conjecture under the stronger assumption that $M$ has\nentire Grauert tube,i.e.,$M$ is a real analytic Riemannian manifold that has a\nunique adapted complex structure defined on the whole tangent bundle $TM$.\n"}
{"abstract": "  An analytical solution of the nonlinear Richards equation is presented, for\none-dimensional infiltration into a soil of uniform initial moisture content\nsubject to a constant depth of surface ponded water. Adopted mathematical forms\nof the soil water diffusivity and conductivity are flexible enough to model a\nrange of real soils. The solution takes the form of a power series in $\\sqrt\nt$, but is observed to converge not only for small times but also for\nrelatively large times at which travelling-wave-like behavior is evident. The\nsolution is used to tabulate exact infiltration coefficients with higher-order\ncorrections as the natural nonlinear limit of soil properties is approached.\nPreviously published approximate solutions that apply for a wide range of soil\nproperties are tested against the exact solution and found to be sufficiently\naccurate.\n"}
{"abstract": "  Normalizing flows have recently been applied to the problem of accelerating\nMarkov chains in lattice field theory. We propose a generalization of\nnormalizing flows that allows them to applied to theories with a sign problem.\nThese complex normalizing flows are closely related to contour deformations\n(i.e. the generalized Lefschetz thimble method), which been applied to sign\nproblems in the past. We discuss the question of the existence of normalizing\nflows: they do not exist in the most general case, but we argue that exact\nnormalizing flows are likely to exist for many physically interesting problems,\nincluding cases where the Lefschetz thimble decomposition has an intractable\nsign problem. Finally, normalizing flows can be constructed in perturbation\ntheory. We give numerical results on their effectiveness across a range of\ncouplings for the Schwinger-Keldysh sign problem associated to a real scalar\nfield in $0+1$ dimensions.\n"}
{"abstract": "  Observations of binary pulsars and pulsars in globular clusters suggest that\nat least some pulsars must receive weak natal kicks at birth. If all pulsars\nreceived strong natal kicks above \\unit[50]{\\kms}, those born in globular\nclusters would predominantly escape, while wide binaries would be disrupted. On\nthe other hand, observations of transverse velocities of isolated radio pulsars\nindicate that only $5\\pm2\\%$ have velocities below \\unit[50]{\\kms}. We explore\nthis apparent tension with rapid binary population synthesis modelling. We\npropose a model in which supernovae with characteristically low natal kicks\n(e.g., electron-capture supernovae) only occur if the progenitor star has been\nstripped via binary interaction with a companion. We show that this model\nnaturally reproduces the observed pulsar speed distribution and without\nreducing the predicted merging double neutron star yield. We estimate that the\nzero-age main sequence mass range for non-interacting progenitors of\nelectron-capture supernovae should be no wider than ${\\approx}0.2 M_\\odot$.\n"}
{"abstract": "  We present a non existence result of complete, Einstein hypersurfaces tangent\nto the Reeb vector field of a regular Sasakian manifold which fibers onto a\ncomplex Stein manifold.\n"}
{"abstract": "  We show that the inclusion of a recently found additional term of the spin\npolarization vector at local equilibrium which is linear in the symmetrized\ngradients of the velocity field, and the assumption of hadron production at\nconstant temperature restore the quantitative agreement between hydrodynamic\nmodel predictions and local polarization measurements in relativistic heavy ion\ncollisions at $\\sqrt{s_{\\rm NN}} = 200$ GeV. The longitudinal component of the\nspin polarization vector turns out to be very sensitive to the temperature\nvalue, with a good fit around 155 MeV. The implications of this finding are\ndiscussed.\n"}
{"abstract": "  We present an approach for visualizing mobile robots through an Augmented\nReality headset when there is no line-of-sight visibility between the robot and\nthe human. Three elements are visualized in Augmented Reality: 1) Robot's 3D\nmodel to indicate its position, 2) An arrow emanating from the robot to\nindicate its planned movement direction, and 3) A 2D grid to represent the\nground plane. We conduct a user study with 18 participants, in which each\nparticipant are asked to retrieve objects, one at a time, from stations at the\ntwo sides of a T-junction at the end of a hallway where a mobile robot is\nroaming. The results show that visualizations improved the perceived safety and\nefficiency of the task and led to participants being more comfortable with the\nrobot within their personal spaces. Furthermore, visualizing the motion intent\nin addition to the robot model was found to be more effective than visualizing\nthe robot model alone. The proposed system can improve the safety of automated\nwarehouses by increasing the visibility and predictability of robots.\n"}
{"abstract": "  In this work, we represent Lex-BERT, which incorporates the lexicon\ninformation into Chinese BERT for named entity recognition (NER) tasks in a\nnatural manner. Instead of using word embeddings and a newly designed\ntransformer layer as in FLAT, we identify the boundary of words in the\nsentences using special tokens, and the modified sentence will be encoded\ndirectly by BERT. Our model does not introduce any new parameters and are more\nefficient than FLAT. In addition, we do not require any word embeddings\naccompanying the lexicon collection. Experiments on Ontonotes and ZhCrossNER\nshow that our model outperforms FLAT and other baselines.\n"}
{"abstract": "  Very recently, the Muon $g-2$ experiment at Fermilab has confirmed the E821\nBrookhaven result, which hinted at a deviation of the muon anomalous magnetic\nmoment from the Standard Model (SM) expectation. The combined results from\nBrookhaven and Fermilab show a difference with the SM prediction $\\delta a_\\mu\n= (251 \\pm 59) \\times 10^{-11}$ at a significance of $4.2\\sigma$, strongly\nindicating the presence of new physics. Motivated by this new result we\nreexamine the contributions to the muon anomalous magnetic moment from both:\n(i)~the ubiquitous $U(1)$ gauge bosons of D-brane string theory constructions\nand (ii)~the Regge excitations of the string. We show that, for a string scale\n${\\cal O} ({\\rm PeV})$, the contribution from anomalous $U(1)$ gauge bosons\nwhich couple to hadrons could help to reduce (though not fully eliminate) the\ndiscrepancy reported by the Muon $g-2$ Collaboration. Consistency with null\nresults from LHC searches of new heavy vector bosons imparts the dominant\nconstraint. We demonstrate that the contribution from Regge excitations is\nstrongly suppressed as it was previously conjectured. We also comment on\ncontributions from Kaluza-Klein (KK) modes, which could help resolve the\n$\\delta a_\\mu$ discrepancy. In particular, we argue that for 4-stack\nintersecting D-brane models, the KK excitations of the $U(1)$ boson living on\nthe lepton brane would not couple to hadrons and therefore can evade the LHC\nbounds while fully bridging the $\\delta a_\\mu$ gap observed at Brookhaven and\nFermilab.\n"}
{"abstract": "  Agile and DevOps are widely adopted by the industry. Hence, integrating\nsecurity activities with industrial practices, such as continuous integration\n(CI) pipelines, is necessary to detect security flaws and adhere to regulators'\ndemands early. In this paper, we analyze automated security activities in CI\npipelines of enterprise-driven open source software (OSS). This shall allow us,\nin the long-run, to better understand the extent to which security activities\nare (or should be) part of automated pipelines. In particular, we mine publicly\navailable OSS repositories and survey a sample of project maintainers to better\nunderstand the role that security activities and their related tools play in\ntheir CI pipelines. To increase transparency and allow other researchers to\nreplicate our study (and to take different perspectives), we further disclose\nour research artefacts. Our results indicate that security activities in\nenterprise-driven OSS projects are scarce and protection coverage is rather\nlow. Only 6.83% of the analyzed 8,243 projects apply security automation in\ntheir CI pipelines, even though maintainers consider security to be rather\nimportant. This alerts industry to keep the focus on vulnerabilities of 3rd\nParty software and it opens space for other improvements of practice which we\noutline in this manuscript.\n"}
{"abstract": "  In this paper, we learn dynamics models for parametrized families of\ndynamical systems with varying properties. The dynamics models are formulated\nas stochastic processes conditioned on a latent context variable which is\ninferred from observed transitions of the respective system. The probabilistic\nformulation allows us to compute an action sequence which, for a limited number\nof environment interactions, optimally explores the given system within the\nparametrized family. This is achieved by steering the system through\ntransitions being most informative for the context variable. We demonstrate the\neffectiveness of our method for exploration on a non-linear toy-problem and two\nwell-known reinforcement learning environments.\n"}
{"abstract": "  In this paper, we propose a wrapper for feature subset selection (FSS) based\non parallel and distributed hybrid evolutionary algorithms viz., parallel\nbinary differential evolution and threshold accepting (PB-DETA), parallel\nbinary threshold accepting and differential evolution (PB-TADE) under the\nApache Spark environment. Here, the FSS is formulated as a combinatorial\noptimization problem. PB-TADE comprises invoking two optimization algorithms\ni.e., TA and BDE in tandem in every iteration, while in PB-DETA, BDE is invoked\nfirst before TA takes over in tandem in every iteration. In addition to these\nhybrids, parallel binary differential evolution (P-BDE), is also developed to\ninvestigate the role played by TA and for baseline comparison. For all the\nthree proposed approaches, logistic regression (LR) is used to compute the\nfitness function namely, the area under ROC curve (AUC) score. The\neffectiveness of the parallel and distributed wrappers is assessed over five\nlarge datasets of varying feature space dimension pertaining to the cyber\nsecurity and biology domains. It is noteworthy that the PB-TADE turned out to\nbe statistically significant compared to P-BDE and PB-DETA. The speed up is\nreported with respect to the sequential version of the three wrappers. Average\nAUC score obtained, most repeated feature subsets, feature subsets with least\ncardinality having best AUC score are also reported. Further, our proposed\nmethods outperformed the state-of-the-art results, wherever the results were\nreported.\n"}
{"abstract": "  Two-particle long-range azimuthal correlations are measured in photonuclear\ncollisions using 1.7 nb$^{-1}$ of 5.02 TeV Pb+Pb collision data collected by\nthe ATLAS experiment at the LHC. Candidate events are selected using a\ndedicated high-multiplicity photonuclear event trigger, a combination of\ninformation from the zero-degree calorimeters and forward calorimeters, and\nfrom pseudorapidity gaps constructed using calorimeter energy clusters and\ncharged-particle tracks. Distributions of event properties are compared between\ndata and Monte Carlo simulations of photonuclear processes. Two-particle\ncorrelation functions are formed using charged-particle tracks in the selected\nevents, and a template-fitting method is employed to subtract the non-flow\ncontribution to the correlation. Significant nonzero values of the second- and\nthird-order flow coefficients are observed and presented as a function of\ncharged-particle multiplicity and transverse momentum. The results are compared\nwith flow coefficients obtained in proton-proton and proton-lead collisions in\nsimilar multiplicity ranges, and with theoretical expectations. The unique\ninitial conditions present in this measurement provide a new way to probe the\norigin of the collective signatures previously observed only in hadronic\ncollisions.\n"}
{"abstract": "  We report direct measurements via angle-resolved photoemission spectroscopy\n(ARPES) of the electronic dispersion of single-layer CoO$_2$. The Fermi contour\nconsists of a large hole pocket centered at the $\\overline{\\Gamma}$ point. To\ninterpret the ARPES results, we use density functional theory (DFT) in\ncombination with the multi-orbital Gutzwiller Approximation (DFT+GA), basing\nour calculations on crystalline structure parameters derived from x-ray\nphotoelectron diffraction and low-energy electron diffraction. Our calculations\nare in good agreement with the measured dispersion. We conclude that the\nmaterial is a moderately correlated metal. We also discuss substrate effects,\nand the influence of hydroxylation on the CoO$_2$ single-layer electronic\nstructure.\n"}
{"abstract": "  Accreting supermassive binary black holes (SMBBHs) are potential\nmulti-messenger sources because they emit both gravitational wave and\nelectromagnetic (EM) radiation. Past work has shown that their EM output may be\nperiodically modulated by an asymmetric density distribution in the\ncircumbinary disk, often called an \"overdensity\" or \"lump;\" this modulation\ncould possibly be used to identify a source as a binary. We explore the\nsensitivity of the overdensity to SMBBH mass ratio and magnetic flux through\nthe accretion disk. We find that the relative amplitude of the overdensity and\nits associated EM periodic signal both degrade with diminishing mass ratio,\nvanishing altogether somewhere between 1:2 and 1:5. Greater magnetization also\nweakens the lump and any modulation of the light output. We develop a model to\ndescribe how lump formation results from internal stress degrading faster in\nthe lump region than it can be rejuvenated through accretion inflow, and\npredicts a threshold value in specific internal stress below which lump\nformation should occur and which all our lump-forming simulations satisfy.\nThus, detection of such a modulation would provide a constraint on both\nmass-ratio and magnetic flux piercing the accretion flow.\n"}
{"abstract": "  Ultra-reliable low-latency communication (URLLC) in 5G New Radio has been\noriginally defined only for licensed spectrum. However, due to new use cases in\nthe Industry 4.0 scenarios, URLLC operation is currently being extended to\nunlicensed spectrum in the ongoing Release 17 of the 3rd Generation Partnership\nProject. Although in such controlled environments we can guarantee the absence\nof any other technology sharing the channel on a long-term basis, the\nuncertainty of obtaining channel access through load based equipment (LBE) or\nframe based equipment (FBE) can impede with the latency requirements of URLLC.\nIn FBE, the transmitters can be prioritized to support data with different\nrequirements and have lower energy consumption and latency compared to LBE with\na big contention window size. In this paper we analyze the performance of FBE\nin an unlicensed controlled environment through a Markov chain. Based on this\nanalysis, we propose two schemes to improve the URLLC performance in FBE: The\nfirst scheme allows the transmitters to use multiple fixed frame period (FFP)\nconfigurations while the second scheme configures the FFP's starting point of\neach transmitter based on its priority. The simulations show the benefits of\nthese schemes compared to the URLLC transmission of existing schemes.\n"}
{"abstract": "  Micro-solid oxide fuel cells based on thin films have strong potential for\nuse in portable power devices. However, devices based on silicon substrates\ntypically involve thin-film metallic electrodes which are unstable at high\ntemperatures. Devices based on bulk metal substrates overcome these\nlimitations, though performance is hindered by the challenge of growing\nstate-of-the-art epitaxial materials on metals. Here, we demonstrate for the\nfirst time the growth of epitaxial cathode materials on metal substrates\n(stainless steel) commercially supplied with epitaxial electrolyte layers (1.5\n{um (Y2O3)0.15(ZrO2)0.85 (YSZ) + 50 nm CeO2). We create epitaxial mesoporous\ncathodes of (La0.60Sr0.40)0.95Co0.20Fe0.80O3 (LSCF) on the substrate by growing\nLSCF/MgO vertically aligned nanocomposite films by pulsed laser deposition,\nfollowed by selectively etching out the MgO. To enable valid comparison with\nthe literature, the cathodes are also grown on single-crystal substrates,\nconfirming state-of-the-art performance with an area specific resistance of\n100ohmegacm2 at 500dC and activation energy down to 0.97 eV. The work marks an\nimportant step toward the commercialization of high-performance micro-solid\noxide fuel cells for portable power applications.\n"}
{"abstract": "  This manuscript introduces a space of functions, termed occupation kernel\nHilbert space (OKHS), that operate on collections of signals rather than real\nor complex functions. To support this new definition, an explicit class of\nOKHSs is given through the consideration of a reproducing kernel Hilbert space\n(RKHS). This space enables the definition of nonlocal operators, such as\nfractional order Liouville operators, as well as spectral decomposition methods\nfor corresponding fractional order dynamical systems. In this manuscript, a\nfractional order DMD routine is presented, and the details of the finite rank\nrepresentations are given. Significantly, despite the added theoretical content\nthrough the OKHS formulation, the resultant computations only differ slightly\nfrom that of occupation kernel DMD methods for integer order systems posed over\nRKHSs.\n"}
{"abstract": "  We present supremum Lagrange Multiplier tests to compare a linear ARMA\nspecification against its threshold ARMA extension. We derive the asymptotic\ndistribution of the test statistics both under the null hypothesis and\ncontiguous local alternatives. Moreover, we prove the consistency of the tests.\nThe Monte Carlo study shows that the tests enjoy good finite-sample properties,\nare robust against model mis-specification and their performance is not\naffected if the order of the model is unknown. The tests present a low\ncomputational burden and do not suffer from some of the drawbacks that affect\nthe quasi-likelihood ratio setting. Lastly, we apply our tests to a time series\nof standardized tree-ring growth indexes and this can lead to new research in\nclimate studies.\n"}
{"abstract": "  For a projective hyperplane arrangement, we study sufficient conditions in\nterms of combinatorial data for ESV-calculability of the monodromy eigenspaces\nof the first Milnor fiber cohomology for eigenvalues of order $m>1$. This can\nbe reduced to the line arrangement case by Artin's theorem. These sufficient\nconditions are often unsatisfied if efficiency or complexity of the\ncombinatorics of arrangement is high. In order to measure these, we introduce\nthe notions of $m$-efficiency and $m$-complexity for $m\\ge 3$. The former is\ndefined to be the number of points with multiplicity divisible by $m$ lying on\none line in average. In many cases, one of the above sufficient conditions is\nsatisfied if it is at most 2, although there are certain exceptional cases,\nespecially when $m=3$. The $m$-complexity is defined to be the maximal number\nof edges containing one vertex of the associated $m$-graph. We can show that\none of the sufficient condition holds if it is at most $(m+1)/2$.\n"}
{"abstract": "  Bayesian variable selection is a powerful tool for data analysis, as it\noffers a principled method for variable selection that accounts for prior\ninformation and uncertainty. However, wider adoption of Bayesian variable\nselection has been hampered by computational challenges, especially in\ndifficult regimes with a large number of covariates or non-conjugate\nlikelihoods. Generalized linear models for count data, which are prevalent in\nbiology, ecology, economics, and beyond, represent an important special case.\nHere we introduce an efficient MCMC scheme for variable selection in binomial\nand negative binomial regression that exploits Tempered Gibbs Sampling (Zanella\nand Roberts, 2019) and that includes logistic regression as a special case. In\nexperiments we demonstrate the effectiveness of our approach, including on\ncancer data with seventeen thousand covariates.\n"}
{"abstract": "  The remaining Useful Life (RUL) of equipment is defined as the duration\nbetween the current time and its failure. An accurate and reliable prognostic\nof the remaining useful life provides decision-makers with valuable information\nto adopt an appropriate maintenance strategy to maximize equipment utilization\nand avoid costly breakdowns. In this work, we propose an end-to-end deep\nlearning model based on multi-layer perceptron and long short-term memory\nlayers (LSTM) to predict the RUL. After normalization of all data, inputs are\nfed directly to an MLP layers for feature learning, then to an LSTM layer to\ncapture temporal dependencies, and finally to other MLP layers for RUL\nprognostic. The proposed architecture is tested on the NASA commercial modular\naero-propulsion system simulation (C-MAPSS) dataset. Despite its simplicity\nwith respect to other recently proposed models, the model developed outperforms\nthem with a significant decrease in the competition score and in the root mean\nsquare error score between the predicted and the gold value of the RUL. In this\npaper, we will discuss how the proposed end-to-end model is able to achieve\nsuch good results and compare it to other deep learning and state-of-the-art\nmethods.\n"}
{"abstract": "  Using the off-shell formulation for ${\\cal N}=2$ conformal supergravity in\nfour dimensions, we describe superconformal higher-spin multiplets of conserved\ncurrents in a curved background and present their associated unconstrained\ngauge prepotentials. The latter are used to construct locally superconformal\nchiral actions, which are demonstrated to be gauge invariant in arbitrary\nconformally flat backgrounds. The main ${\\cal N}=2$ results are then\ngeneralised to the ${\\cal N}$-extended case. We also present the\ngauge-invariant field strengths for on-shell massless higher-spin ${\\cal N}=2$\nsupermultiplets in anti-de Sitter space. These field strengths prove to furnish\nrepresentations of the ${\\cal N}=2$ superconformal group.\n"}
{"abstract": "  We investigate the null controllability property of the parabolic equation\nassociated with the Grushin operator defined by the canonical almost-Riemannian\nstructure on the 2-dimensional sphere $\\mathbb S^2$. This is the natural\ngeneralization of the Grushin operator $\\mathcal G = \\partial_x^2 +\nx^2\\partial_y^2$ on $\\mathbb R^2$ to this curved setting, and presents a\ndegeneracy at the equator of $\\mathbb S^2$. Namely, its expression in\nlatitude-longitude spherical coordinates $(x,y)\\in(-\\pi/2,\\pi/2)\\times\n[0,2\\pi)$ is\n  $$\\mathcal{L} = \\partial_{x}^{2}-\\tan\nx\\partial_{x}+\\tan^2x\\partial_{y}^{2}.$$\n  We prove that the null controllability is verified in large time when the\ncontrol acts as a source term distributed on a subset $\\omega = \\{\n(x_1,x_2,x_3)\\in \\mathbb S^2\\mid \\alpha<|x_3|<\\beta \\}$ for some\n$0\\le\\alpha<\\beta\\le 1$. More precisely, we show the existence of a positive\ntime $T^{*}>0$ such that the system is null controllable from $\\omega$ in any\ntime $T>T^{*}$, and that the minimal time of control from $\\omega$ satisfies\n$T_{\\min}\\ge\\log(1/\\sqrt{1-\\alpha^2})$. Here, the lower bound corresponds to\nthe Agmon distance of $\\omega$ from the equator.\n  These results are obtained by applying the same ideas as in (Beauchard et\nal., 2014), consisting in proving a suitable Carleman estimate, and then taking\nadvantage of a Poincar\\'e-Hardy type inequality due to (Chisholm and Everitt,\n1970) to show the positive null-controllability result. On the other hand, the\nnegative statement is proved by exploiting an appropriate family of spherical\nharmonics, which concentrate at the equator, to falsify the uniform\nobservability inequality.\n"}
{"abstract": "  Significant understanding has been achieved over the last few decades\nregarding chirality-dependent properties of single-wall carbon nanotubes\n(SWCNTs), primarily through single-tube studies. However, macroscopic\nmanifestations of chirality dependence have been limited, especially in\nelectronic transport, despite the fact that such distinct behaviors are needed\nfor many applications of SWCNT-based devices. In addition, developing reliable\ntransport theory is challenging since a description of localization phenomena\nin an assembly of nanoobjects requires precise knowledge of disorder on\nmultiple spatial scales, particularly if the ensemble is heterogeneous. Here,\nwe report an observation of pronounced chirality-dependent electronic\nlocalization in temperature and magnetic field dependent conductivity\nmeasurements on macroscopic films of single-chirality SWCNTs. The samples\nincluded large-gap semiconducting (6,5) and (10,3) films, narrow-gap\nsemiconducting (7,4) and (8,5) films, and armchair metallic (6,6) films.\nExperimental data and theoretical calculations revealed Mott\nvariable-range-hopping dominated transport in all samples, while localization\nlengths fall into three distinct categories depending on their band gaps.\nArmchair films have the largest localization length. Our detailed analyses on\nelectronic transport properties of single-chirality SWCNT films provide\nsignificant new insight into electronic transport in ensembles of nanoobjects,\noffering foundations for designing and deploying macroscopic SWCNT solid-state\ndevices.\n"}
{"abstract": "  The pythagorean fuzzy set (PFS) which is developed based on intuitionistic\nfuzzy set, is more efficient in elaborating and disposing uncertainties in\nindeterminate situations, which is a very reason of that PFS is applied in\nvarious kinds of fields. How to measure the distance between two pythagorean\nfuzzy sets is still an open issue. Mnay kinds of methods have been proposed to\npresent the of the question in former reaserches. However, not all of existing\nmethods can accurately manifest differences among pythagorean fuzzy sets and\nsatisfy the property of similarity. And some other kinds of methods neglect the\nrelationship among three variables of pythagorean fuzzy set. To addrees the\nproplem, a new method of measuring distance is proposed which meets the\nrequirements of axiom of distance measurement and is able to indicate the\ndegree of distinction of PFSs well. Then some numerical examples are offered to\nto verify that the method of measuring distances can avoid the situation that\nsome counter? intuitive and irrational results are produced and is more\neffective, reasonable and advanced than other similar methods. Besides, the\nproposed method of measuring distances between PFSs is applied in a real\nenvironment of application which is the medical diagnosis and is compared with\nother previous methods to demonstrate its superiority and efficiency. And the\nfeasibility of the proposed method in handling uncertainties in practice is\nalso proved at the same time.\n"}
{"abstract": "  We consider a decision maker who must choose an action in order to maximize a\nreward function that depends also on an unknown parameter {\\Theta}. The\ndecision maker can delay taking the action in order to experiment and gather\nadditional information on {\\Theta}. We model the decision maker's problem using\na Bayesian sequential experimentation framework and use dynamic programming and\ndiffusion-asymptotic analysis to solve it. For that, we scale our problem in a\nway that both the average number of experiments that is conducted per unit of\ntime is large and the informativeness of each individual experiment is low.\nUnder such regime, we derive a diffusion approximation for the sequential\nexperimentation problem, which provides a number of important insights about\nthe nature of the problem and its solution. Our solution method also shows that\nthe complexity of the problem grows only quadratically with the cardinality of\nthe set of actions from which the decision maker can choose. We illustrate our\nmethodology and results using a concrete application in the context of\nassortment selection and new product introduction. Specifically, we study the\nproblem of a seller who wants to select an optimal assortment of products to\nlaunch into the marketplace and is uncertain about consumers' preferences.\nMotivated by emerging practices in e-commerce, we assume that the seller is\nable to use a crowdvoting system to learn these preferences before a final\nassortment decision is made. In this context, we undertake an extensive\nnumerical analysis to assess the value of learning and demonstrate the\neffectiveness and robustness of the heuristics derived from the diffusion\napproximation.\n"}
{"abstract": "  Prostate cancer biopsy benefits from accurate fusion of transrectal\nultrasound (TRUS) and magnetic resonance (MR) images. In the past few years,\nconvolutional neural networks (CNNs) have been proved powerful in extracting\nimage features crucial for image registration. However, challenging\napplications and recent advances in computer vision suggest that CNNs are quite\nlimited in its ability to understand spatial correspondence between features, a\ntask in which the self-attention mechanism excels. This paper aims to develop a\nself-attention mechanism specifically for cross-modal image registration. Our\nproposed cross-modal attention block effectively maps each of the features in\none volume to all features in the corresponding volume. Our experimental\nresults demonstrate that a CNN network designed with the cross-modal attention\nblock embedded outperforms an advanced CNN network 10 times of its size. We\nalso incorporated visualization techniques to improve the interpretability of\nour network. The source code of our work is available at\nhttps://github.com/DIAL-RPI/Attention-Reg .\n"}
{"abstract": "  This paper focuses on an important problem of detecting offensive analogy\nmeme on online social media where the visual content and the texts/captions of\nthe meme together make an analogy to convey the offensive information. Existing\noffensive meme detection solutions often ignore the implicit relation between\nthe visual and textual contents of the meme and are insufficient to identify\nthe offensive analogy memes. Two important challenges exist in accurately\ndetecting the offensive analogy memes: i) it is not trivial to capture the\nanalogy that is often implicitly conveyed by a meme; ii) it is also challenging\nto effectively align the complex analogy across different data modalities in a\nmeme. To address the above challenges, we develop a deep learning based\nAnalogy-aware Offensive Meme Detection (AOMD) framework to learn the implicit\nanalogy from the multi-modal contents of the meme and effectively detect\noffensive analogy memes. We evaluate AOMD on two real-world datasets from\nonline social media. Evaluation results show that AOMD achieves significant\nperformance gains compared to state-of-the-art baselines by detecting offensive\nanalogy memes more accurately.\n"}
{"abstract": "  Semiparametric accelerated failure time (AFT) models are a useful alternative\nto Cox proportional hazards models, especially when the assumption of constant\nhazard ratios is untenable. However, rank-based criteria for fitting AFT models\nare often non-differentiable, which poses a computational challenge in\nhigh-dimensional settings. In this article, we propose a new alternating\ndirection method of multipliers algorithm for fitting semiparametric AFT models\nby minimizing a penalized rank-based loss function. Our algorithm scales well\nin both the number of subjects and number of predictors; and can easily\naccommodate a wide range of popular penalties. To improve the selection of\ntuning parameters, we propose a new criterion which avoids some common problems\nin cross-validation with censored responses. Through extensive simulation\nstudies, we show that our algorithm and software is much faster than existing\nmethods (which can only be applied to special cases), and we show that\nestimators which minimize a penalized rank-based criterion often outperform\nalternative estimators which minimize penalized weighted least squares\ncriteria. Application to nine cancer datasets further demonstrates that\nrank-based estimators of semiparametric AFT models are competitive with\nestimators assuming proportional hazards model in high-dimensional settings,\nwhereas weighted least squares estimators are often not. A software package\nimplementing the algorithm, along with a set of auxiliary functions, is\navailable for download at github.com/ajmolstad/penAFT.\n"}
{"abstract": "  The Dzyaloshinskii-Moriya interaction (DMI) is at the heart of many modern\ndevelopments in the research field of spintronics. DMI is known to generate\nnoncollinear magnetic textures, and can take two forms in antiferromagnets:\nhomogeneous or inter-sublattice, leading to small, canted moments and\ninhomogeneous or intra-sublattice, leading to formation of chiral structures.\nIn this work, we first determine the strength of the effective field created by\nthe DMI, using SQUID based magnetometry and transport measurements, in thin\nfilms of the antiferromagnetic iron oxide hematite, $\\alpha$-Fe$_2$O$_3$. We\ndemonstrate that DMI additionally introduces reconfigurability in the long\ndistance magnon transport in these films under different orientations of a\nmagnetic field. This arises as a hysteresis centred around the easy-axis\ndirection for an external field rotated in opposing directions whose width\ndecreases with increasing magnetic field as the Zeeman energy competes with the\neffective field created by the DMI.\n"}
{"abstract": "  The lack of specifications is a key difference between traditional software\nengineering and machine learning. We discuss how it drastically impacts how we\nthink about divide-and-conquer approaches to system design, and how it impacts\nreuse, testing and debugging activities. Traditionally, specifications provide\na cornerstone for compositional reasoning and for the divide-and-conquer\nstrategy of how we build large and complex systems from components, but those\nare hard to come by for machine-learned components. While the lack of\nspecification seems like a fundamental new problem at first sight, in fact\nsoftware engineers routinely deal with iffy specifications in practice: we face\nweak specifications, wrong specifications, and unanticipated interactions among\ncomponents and their specifications. Machine learning may push us further, but\nthe problems are not fundamentally new. Rethinking machine-learning model\ncomposition from the perspective of the feature interaction problem, we may\neven teach us a thing or two on how to move forward, including the importance\nof integration testing, of requirements engineering, and of design.\n"}
{"abstract": "  Gyroscopes are widely used in various field. The instability of the low-cost\ngyroscopes makes them need to be calibrated on every boot. To meet the\nrequirement of frequency calibration, finding an efficient in-field calibration\nmethod is essential. This paper proposes a fast calibration method that does\nnot require any external equipment. We use the manual rotation angle as a\ncalibration reference and linearize the calibration model. On the basis of this\nmodel, a G-optimal experimental design scheme is proposed, which can get enough\ncalibration information with the least number of experiments. The simulations\nindicate that the calibration error is relatively low, and the results are\nunbiased. We empirically validate the effectiveness of the proposed method on\ntwo commonly used low-cost gyroscope and achieve real-time calibration on a\nlow-energy microcontroller. We validate the proposed method by comparing the\nabove method with the conventional turntable method. The experiment result\nshows that the error between these two methods is less than $\\pm3 \\times\n10^{-2}$ and the calibration process takes less than 30 seconds. This method\nmight have a practical implication for low-cost gyroscope calibration.\n"}
{"abstract": "  According to the FNAL+BNL measurements for the muon $g-2$ and the Berkeley\n$^{133}$Cs measurement for the electron $g-2$, the SM prediction for the muon\n(electron) $g-2$ is $4.2\\sigma$ ($2.4\\sigma$) below (above) the experimental\nvalue. A joint explanation requires a positive contribution to the muon $g-2$\nand a negative contribution to the electron $g-2$. In this work we explore the\npossibility of such a joint explanation in the minimal supersymmetric standard\nmodel (MSSM). Assuming no universality between smuon and selectron soft masses,\nwe find out a part of parameter space for a joint explanation of muon and\nelectron $g-2$ anomalies at $2\\sigma$ level. This part of parameter space can\nsurvive the LHC and LEP constraints, but gives an over-abundance for the dark\nmatter if the bino-like lightest neutralino is assumed to be the dark matter\ncandidate. With the assumption that the dark matter candidate is a superWIMP\n(say a pseudo-goldstino in multi-sector SUSY breaking scenarios, whose mass can\nbe as light as GeV and produced from the late-dacay of the thermally freeze-out\nlightest neutralino), the dark matter problem can be avoided. So, the MSSM may\ngive a joint explanation for the muon and electron $g-2$ anomalies at $2\\sigma$\nlevel (the muon $g-2$ anomaly can be ameliorated to $1\\sigma$).\n"}
{"abstract": "  It remains an open question how to determine the winner of an election given\nincomplete or uncertain voter preferences. One solution is to assume some\nprobability space for the voting profile and declare the candidates having the\nbest chance of winning to be the (co-)winners. We refer to this as the Most\nProbable Winner (MPW). In this paper, we propose an alternative winner\ninterpretation for positional scoring rules - the Most Expected Winner (MEW),\nbased on the expected performance of the candidates. This winner interpretation\nenjoys some desirable properties that the MPW does not. We establish the\ntheoretical hardness of MEW over incomplete voter preferences, then identify a\ncollection of tractable cases for a variety of voting profiles. An important\ncontribution of this work is to separate the voter preferences into the\ngeneration step and the observation step, which gives rise to a unified voting\nprofile combining both incomplete and probabilistic voting profiles.\n"}
{"abstract": "  The aims of this work are to study Rees algebras of filtrations of monomial\nideals associated to covering polyhedra of rational matrices with non-negative\nentries and non-zero columns using combinatorial optimization and integer\nprogramming, and to study powers of monomial ideals and their integral closures\nusing irreducible decompositions and polyhedral geometry. We study the\nWaldschmidt constant and the ic-resurgence of the filtration associated to a\ncovering polyhedron and show how to compute these constants using linear\nprogramming. Then we show a lower bound for the ic-resurgence of the ideal of\ncovers of a graph and prove that the lower bound is attained when the graph is\nperfect. We also show lower bounds for the ic-resurgence of the edge ideal of a\ngraph and give an algorithm to compute the asymptotic resurgence of squarefree\nmonomial ideals. A classification of when Newton's polyhedron is the\nirreducible polyhedron is presented using integral closure.\n"}
{"abstract": "  Polariton-based devices require materials where light-matter coupling under\nambient conditions exceeds losses, but our current selection of such materials\nis limited. Here we measured the dispersion of polaritons formed by the $A$ and\n$B$ excitons in thin MoS$_2$ slabs by imaging their optical near fields. We\ncombined fully tunable laser excitation in the visible with a scattering\nnear-field optical microscope to excite polaritons and image their optical near\nfields. We obtained the properties of bulk MoS$_2$ from fits to the slab\ndispersion. The in-plane excitons are in the strong regime of light-matter\ncoupling with a coupling strength ($40-100\\,$meV) that exceeds their losses by\nat least a factor of two. The coupling becomes comparable to the exciton\nbinding energy, which is known as very strong coupling. MoS$_2$ and other\ntransition metal dichalcogenides are excellent materials for future polariton\ndevices.\n"}
{"abstract": "  Domain adaptation (DA) enables knowledge transfer from a labeled source\ndomain to an unlabeled target domain by reducing the cross-domain distribution\ndiscrepancy. Most prior DA approaches leverage complicated and powerful deep\nneural networks to improve the adaptation capacity and have shown remarkable\nsuccess. However, they may have a lack of applicability to real-world\nsituations such as real-time interaction, where low target inference latency is\nan essential requirement under limited computational budget. In this paper, we\ntackle the problem by proposing a dynamic domain adaptation (DDA) framework,\nwhich can simultaneously achieve efficient target inference in low-resource\nscenarios and inherit the favorable cross-domain generalization brought by DA.\nIn contrast to static models, as a simple yet generic method, DDA can integrate\nvarious domain confusion constraints into any typical adaptive network, where\nmultiple intermediate classifiers can be equipped to infer \"easier\" and\n\"harder\" target data dynamically. Moreover, we present two novel strategies to\nfurther boost the adaptation performance of multiple prediction exits: 1) a\nconfidence score learning strategy to derive accurate target pseudo labels by\nfully exploring the prediction consistency of different classifiers; 2) a\nclass-balanced self-training strategy to explicitly adapt multi-stage\nclassifiers from source to target without losing prediction diversity.\nExtensive experiments on multiple benchmarks are conducted to verify that DDA\ncan consistently improve the adaptation performance and accelerate target\ninference under domain shift and limited resources scenarios\n"}
{"abstract": "  We report an ultra long range energy transfer in a layered\ndonor:spacer:acceptor structures, which consist of the typical organic\nphotovoltaic material system of P3HT as the donor and PCBM as the acceptor. By\nvarying the thicknesses of spacer and acceptor layers we show that the energy\ntransfer is to the full volume of the acceptor and not just to its nearest\ninterface, and we find an effective energy transfer range on the order of about\n100nm. Our efforts to elucidate the origin of this process, both theoretically\nand experimentally, are discussed as well. Although it was recently implied\nthat an exceptionally large energy transfer may take place in this material\nsystem, this is the first time, to the best of our knowledge, that the energy\ntransfer mechanism is characterized in a quantitative way and compared to the\ncommon models. Our analysis offers new prospects for the familiar photovoltaic\nbi-layer configurations, which may be very efficient if the ultra long range\nenergy transfer is utilized through a suitable architecture.\n"}
{"abstract": "  We study large-scale applications using a GPU-accelerated version of the\nmassively parallel J\\\"ulich universal quantum computer simulator (JUQCS--G).\nFirst, we benchmark JUWELS Booster, a GPU cluster with 3744 NVIDIA A100 Tensor\nCore GPUs. Then, we use JUQCS--G to study the relation between quantum\nannealing (QA) and the quantum approximate optimization algorithm (QAOA). We\nfind that a very coarsely discretized version of QA, termed approximate quantum\nannealing (AQA), performs surprisingly well in comparison to the QAOA. It can\neither be used to initialize the QAOA, or to avoid the costly optimization\nprocedure altogether. Furthermore, we study the scaling of the success\nprobability when using AQA for problems with 30 to 40 qubits. We find that the\ncase with largest discretization error performs most favorably, surpassing the\nbest result obtained from the QAOA.\n"}
{"abstract": "  We prove that there exists an open and dense subset $\\mathcal{U}$ in the\nspace of $C^{2}$ expanding self-maps of the circle $\\mathbb{T}$ such that the\nLyapunov minimizing measures of any $T\\in{\\mathcal U}$ are uniquely supported\non a periodic orbit.This answers a conjecture of Jenkinson-Morris in the $C^2$\ntopology.\n"}
{"abstract": "  Cryptocurrencies redefined how money can be stored and transferred among\nusers. However, independent of the amount being sent, public blockchain-based\ncryptocurrencies suffer from high transaction waiting times and fees. These\ndrawbacks hinder the wide use of cryptocurrencies by masses. To address these\nchallenges, payment channel network concept is touted as the most viable\nsolution to be used for micro-payments. The idea is exchanging the ownership of\nmoney by keeping the state of the accounts locally. The users inform the\nblockchain rarely, which decreases the load on the blockchain. Specifically,\npayment channel networks can provide transaction approvals in seconds by\ncharging a nominal fee proportional to the payment amount. Such attraction on\npayment channel networks inspired many recent studies which focus on how to\ndesign them and allocate channels such that the transactions will be secure and\nefficient. However, as payment channel networks are emerging and reaching large\nnumber of users, privacy issues are becoming more relevant that raise concerns\nabout exposing not only individual habits but also businesses' revenues. In\nthis paper, we first propose a categorization of the existing payment networks\nformed on top of blockchain-backed cryptocurrencies. After discussing several\nemerging attacks on user/business privacy in these payment channel networks, we\nqualitatively evaluate them based on a number of privacy metrics that relate to\nour case. Based on the discussions on the strengths and weaknesses of the\napproaches, we offer possible directions for research for the future of privacy\nbased payment channel networks.\n"}
{"abstract": "  The accurate measurement of stellar masses over a wide range of galaxy\nproperties is essential for better constraining models of galaxy evolution.\nEmission line galaxies (ELGs) tend to have better redshift estimates than\ncontinuum-selected objects and have been shown to span a large range of\nphysical properties, including stellar mass. Using data from the 3D-HST\nTreasury program, we construct a carefully vetted sample of 4350 ELGs at\nredshifts 1.16<z<1.90. We combine the 3D-HST emission line fluxes with far-UV\nthrough near-IR photometry and use the MCSED spectral energy distribution\nfitting code to constrain the galaxies' physical parameters, such as their star\nformation rate (SFRs) and stellar masses. Our sample is consistent with the z~2\nmass-metallicity relation. More importantly, we show there is a simple but\ntight correlation between stellar mass and absolute magnitude in a near-IR\nfilter that will be particularly useful in quickly calculating accurate stellar\nmasses for millions of galaxies in upcoming missions such as Euclid and the\nNancy Grace Roman Space Telescope.\n"}
{"abstract": "  The impact of momentum anisotropy on the heavy quark transport coefficients\ndue to collisional and radiative processes in the QCD medium has been studied\nwithin the ambit of kinetic theory. Anisotropic aspects (momentum) are\nincorporated into the heavy quark dynamics through the non-equilibrium momentum\ndistribution function of quarks, antiquarks, and gluons. These non-equilibrium\ndistribution functions that encode the physics of momentum anisotropy and\nturbulent chromo-fields have been obtained by solving the ensemble-averaged\ndiffusive Vlasov-Boltzmann equation. The momentum dependence of heavy quark\ntransport coefficients in the medium is seen to be sensitive to the strength of\nthe anisotropy for both collisional and radiative processes. In addition, the\ncollisional and radiative energy loss of the heavy quark in the anisotropic hot\nQCD medium have been analyzed. The effects of anisotropy on the drag and\ndiffusion coefficients are observed to have a visible impact on the nuclear\nsuppression factor both at the RHIC and LHC.\n"}
{"abstract": "  For years, independence has been considered as an important concept in many\ndisciplines. Nevertheless, we present the first research that investigates the\ndiscovery problem of independence in data. In its arguably simplest form,\nindependence is a statement between two sets of columns expressing that for\nevery two rows in a table there is also a row in the table that coincides with\nthe first row on the first set of columns and with the second row on the second\nset of columns. We show that the problem of deciding whether there is an\nindependence statement that holds on a given table is not only NP-complete but\n$W[3]$-complete in its arguably most natural parameter, namely its arity. We\nestablish the first algorithm to discover all independence statement that hold\non a given table. We illustrate in experiments with benchmark data that our\nalgorithm performs well within the limits established by our hardness results.\nIn practice, it is often useful to determine the ratio with which independence\nstatements hold on a given table. For that purpose, we show that our treatment\nof independence and the design of our algorithm enables us to extend our\nfindings to approximate independence. In our final experiments, we provide some\ninsight into the trade-off between run time and the approximation ratio.\nNaturally, the smaller the ratio, the more approximate independence statements\nhold, and the more time it takes to discover all of them. While this research\nestablishes first insight into the computational properties of discovering\nindependence from data, we hope to initiate research into more sophisticated\nnotions of independence, including embedded multivalued dependencies, as well\nas their context-specific and probabilistic variants.\n"}
{"abstract": "  Momentum-space topology of electrons under strong spin-orbit coupling\ncontributes to the electrically induced torques exerting on magnetic textures\ninsensitively to disorder or thermal fluctuation. We present a direct\nconnection between band topology and the torques by classifying the whole\ntorques phenomenologically. As well as the intrinsic anomalous Hall effect, the\ntorques also emerge intrinsically from the anomalous velocity of electrons\nregardless of a nonequilibrium transport current. We especially point out the\nintrinsic contribution arising exclusively in magnetic textures, which we call\nthe \"topological Hall torque (THT).\" The THT emerges in bulk crystals without\nany interface or surface structures. We numerically demonstrate the enhancement\nof the THT in comparison with the conventional spin-transfer torque in the bulk\nmetallic ferromagnet, which accounts for the giant current-induced torque\nmeasured in ferromagnetic SrRuO3.\n"}
{"abstract": "  Recently, the end-to-end approach has been successfully applied to\nmulti-speaker speech separation and recognition in both single-channel and\nmultichannel conditions. However, severe performance degradation is still\nobserved in the reverberant and noisy scenarios, and there is still a large\nperformance gap between anechoic and reverberant conditions. In this work, we\nfocus on the multichannel multi-speaker reverberant condition, and propose to\nextend our previous framework for end-to-end dereverberation, beamforming, and\nspeech recognition with improved numerical stability and advanced frontend\nsubnetworks including voice activity detection like masks. The techniques\nsignificantly stabilize the end-to-end training process. The experiments on the\nspatialized wsj1-2mix corpus show that the proposed system achieves about 35%\nWER relative reduction compared to our conventional multi-channel E2E ASR\nsystem, and also obtains decent speech dereverberation and separation\nperformance (SDR=12.5 dB) in the reverberant multi-speaker condition while\ntrained only with the ASR criterion.\n"}
{"abstract": "  Structure and thermodynamics in restricted primitive model electrolytes are\nexamined using three recently developed versions of a linear form of the\nmodified Poisson-Boltzmann equation. Analytical expressions for the osmotic\ncoefficient and the electrical part of the mean activity coefficient are\nobtained and the results for the osmotic and the mean activity coefficients are\ncompared with that from the more established mean spherical approximation,\nsymmetric Poisson-Boltzmann, modified Poisson-Boltzmann theories, and available\nMonte Carlo simulation results. The linear theories predict the thermodynamics\nto a remarkable degree of accuracy relative to the simulations and are\nconsistent with the mean spherical approximation and modified Poisson-Boltzmann\nresults. The predicted structure in the form of the radial distribution\nfunctions and the mean electrostatic potential also compare well with the\ncorresponding results from the formal theories. The excess internal energy and\nthe electrical part of the mean activity coefficient are shown to be identical\nanalytically for the mean spherical approximation and the linear modified\nPoisson-Boltzmann theories.\n"}
{"abstract": "  This paper describes the training process of the first Czech monolingual\nlanguage representation models based on BERT and ALBERT architectures. We\npre-train our models on more than 340K of sentences, which is 50 times more\nthan multilingual models that include Czech data. We outperform the\nmultilingual models on 9 out of 11 datasets. In addition, we establish the new\nstate-of-the-art results on nine datasets. At the end, we discuss properties of\nmonolingual and multilingual models based upon our results. We publish all the\npre-trained and fine-tuned models freely for the research community.\n"}
{"abstract": "  This paper presents a novel computer vision the algorithm proposed for the\non-line range and bearing detection of the robot swarms. Results demonstrated\nthe reliability of the proposed vision system such that it can be used for the\nrobot swarms applications.\n"}
{"abstract": "  The concept of Hybrid Twin (HT) has recently received a growing interest\nthanks to the availability of powerful machine learning techniques. This twin\nconcept combines physics-based models within a model-order reduction\nframework-to obtain real-time feedback rates-and data science. Thus, the main\nidea of the HT is to develop on-the-fly data-driven models to correct possible\ndeviations between measurements and physics-based model predictions. This paper\nis focused on the computation of stable, fast and accurate corrections in the\nHybrid Twin framework. Furthermore, regarding the delicate and important\nproblem of stability, a new approach is proposed, introducing several\nsub-variants and guaranteeing a low computational cost as well as the\nachievement of a stable time-integration.\n"}
{"abstract": "  DeepLLL algorithm (Schnorr, 1994) is a famous variant of LLL lattice basis\nreduction algorithm, and PotLLL algorithm (Fontein et al., 2014) and $S^2$LLL\nalgorithm (Yasuda and Yamaguchi, 2019) are recent polynomial-time variants of\nDeepLLL algorithm developed from cryptographic applications. However, the known\npolynomial bounds for computational complexity are shown only for parameter\n$\\delta < 1$; for \"optimal\" parameter $\\delta = 1$ which ensures the best\noutput quality, no polynomial bounds are known, and except for LLL algorithm,\nit is even not formally proved that the algorithm always halts within finitely\nmany steps. In this paper, we prove that these four algorithms always halt also\nwith optimal parameter $\\delta = 1$, and furthermore give explicit upper bounds\nfor the numbers of loops executed during the algorithms. Unlike the known bound\n(Akhavi, 2003) applicable to LLL algorithm only, our upper bounds are deduced\nin a unified way for all of the four algorithms.\n"}
{"abstract": "  We search for microlensing planets with signals exhibiting no\ncaustic-crossing features, considering the possibility that such signals may be\nmissed due to their weak and featureless nature. For this purpose, we reexamine\nthe lensing events found by the KMTNet survey before the 2019 season. From this\ninvestigation, we find two new planetary lensing events, KMT-2018-BLG-1976 and\nKMT-2018-BLG-1996. We also present the analysis of the planetary event\nOGLE-2019-BLG-0954, for which the planetary signal was known, but no detailed\nanalysis has been presented before. We identify the genuineness of the\nplanetary signals by checking various interpretations that can generate\nshort-term anomalies in lensing light curves. From Bayesian analyses conducted\nwith the constraint from available observables, we find that the host and\nplanet masses are $(M_1, M_2)\\sim (0.65~M_\\odot, 2~M_{\\rm J})$ for\nKMT-2018-BLG-1976L, $\\sim (0.69~M_\\odot, 1~M_{\\rm J})$ for KMT-2018-BLG-1996L,\nand $\\sim (0.80~M_\\odot, 14~M_{\\rm J})$ for OGLE-2019-BLG-0954L. The estimated\ndistance to OGLE-2019-BLG-0954L, $3.63^{+1.22}_{-1.64}$~kpc, indicates that it\nis located in the disk, and the brightness expected from the mass and distance\nmatches well the brightness of the blend, indicating that the lens accounts for\nmost of the blended flux. The lens of OGLE-2019-BLG-0954 could be resolved from\nthe source by conducting high-resolution follow-up observations in and after\n2024.\n"}
{"abstract": "  The random q-state quantum Potts model is studied on hypercubic lattices in\ndimensions 2 and 3 using the numerical implementation of the Strong Disorder\nRenormalization Group introduced by Kovacs and Igl{\\'o}i [Phys. Rev. B 82,\n054437 (2010)]. Critical exponents $\\nu$, d f and $\\psi$ at the Infinite\nDisorder Fixed Point are estimated by Finite-Size Scaling for several numbers\nof states q between 2 and 50. When scaling corrections are not taken into\naccount, the estimates of both d f and $\\psi$ systematically increase with q.\nIt is shown however that q-dependent scaling corrections are present and that\nthe exponents are compatible within error bars, or close to each other, when\nthese corrections are taking into account. This provides evidence of the\nexistence of a super-universality of all 2D and 3D random Potts models.\n"}
{"abstract": "  Blockchain is an emerging technology with potential to address issues related\nto sustainability. Literature reviews on blockchain and sustainability exist,\nbut there is a need to consolidate existing results, in particular, in terms of\nSustainable Development Goals (SDG). This extended abstract presents an ongoing\ntertiary study based on existing literature reviews to investigate the\nrelationship between blockchain and sustainability in terms of SDGs. Results\nfrom a pilot analysis of 18 reviews using thematic analysis are presented.\n"}
{"abstract": "  We prove pointwise convergence for the scattering data of a Dirac system of\ndifferential equations. Equivalently, we prove an analog of Carleson's theorem\non almost everywhere convergence of Fourier series for a version of the\nnon-linear Fourier transform. Our proofs are based on the study of resonances\nof Dirac systems using families of meromorphic inner functions, generated by a\nRicatti equation corresponding to the system.\n"}
{"abstract": "  We show that the Liouville function changes sign infinitely often for\n$x^2+d$.\n"}
{"abstract": "  We show that any compact surface of genus zero in Euclidean 3-space that\nsatisfies a quasiconformal inequality between its principal curvatures is a\nround sphere. This solves an old open problem by H. Hopf, and gives a spherical\nversion of Simon's quasiconformal Bernstein theorem. The result generalizes,\namong others, Hopf's theorem for constant mean curvature spheres, the\nclassification of round spheres as the only compact elliptic Weingarten\nsurfaces of genus zero, and the uniqueness theorem for ovaloids by Han,\nNadirashvili and Yuan. The proof relies on the Bers-Nirenberg representation of\nsolutions to linear elliptic equations with discontinuous coefficients.\n"}
{"abstract": "  To convert a fractional solution to an instance of a constraint satisfaction\nproblem into a solution, a rounding scheme is needed, which can be described by\na collection of symmetric operations with one of each arity. An intriguing\npossibility, raised in a recent paper by Carvalho and Krokhin, would imply that\nany clone of operations on a set $D$ which contains symmetric operations of\narities $1, 2, \\ldots, \\lvert D \\rvert$ contains symmetric operations of all\narities in the clone. If true, then it is possible to check whether any given\nfamily of constraint satisfaction problems is solved by its linear programming\nrelaxation.\n  We characterize all idempotent clones containing symmetric operations of\narities $1, 2, \\ldots, \\lvert D \\rvert$ for all sets $D$ with size at most four\nand prove that each one contains symmetric operations of every arity, proving\nthe conjecture above for $\\lvert D \\rvert \\leq 4$.\n"}
{"abstract": "  The emergence of nano-scale memristive devices encouraged many different\nresearch areas to exploit their use in multiple applications. One of the\nproposed applications was to implement synaptic connections in bio-inspired\nneuromorphic systems. Large-scale neuromorphic hardware platforms are being\ndeveloped with increasing number of neurons and synapses, having a critical\nbottleneck in the online learning capabilities. Spike-timing-dependent\nplasticity (STDP) is a widely used learning mechanism inspired by biology which\nupdates the synaptic weight as a function of the temporal correlation between\npre- and post-synaptic spikes. In this work, we demonstrate experimentally that\nbinary stochastic STDP learning can be obtained from a memristor when the\nappropriate pulses are applied at both sides of the device.\n"}
{"abstract": "  In the framework of the study of helium-like atomic systems possessing the\ncollinear configuration, we propose a simple method for computing compact but\nvery accurate wave functions describing the relevant $S$ state. It is worth\nnoting that the considered states include the well-known states of the\nelectron-nucleus and electron-electron coalescences as a particular case. The\nsimplicity and compactness imply that the considered wave functions represent a\nlinear combinations of few single exponentials. We have calculated such model\nwave functions for the ground state of helium and the two-electron ions with\nnucleus charge $1 \\leq Z \\leq 5$. The parameters and the accompanying\ncharacteristics of these functions are presented in tables for number of\nexponential from 3 to 6. The accuracy of the resulting wave functions are\nconfirmed graphically. The specific properties of the relevant codes by Wolfram\nMathematica are discussed. An example of application of the compact wave\nfunctions under consideration is reported.\n"}
{"abstract": "  Recent work on promoting cooperation in multi-agent learning has resulted in\nmany methods which successfully promote cooperation at the cost of becoming\nmore vulnerable to exploitation by malicious actors. We show that this is an\nunavoidable trade-off and propose an objective which balances these concerns,\npromoting both safety and long-term cooperation. Moreover, the trade-off\nbetween safety and cooperation is not severe, and you can receive exponentially\nlarge returns through cooperation from a small amount of risk. We study both an\nexact solution method and propose a method for training policies that targets\nthis objective, Accumulating Risk Capital Through Investing in Cooperation\n(ARCTIC), and evaluate them in iterated Prisoner's Dilemma and Stag Hunt.\n"}
{"abstract": "  Utilizing the unbiased time-dependent density-matrix renormalization group\ntechnique, we examine the photoemission spectra in the extended Falicov-Kimball\nmodel at zero and finite temperatures, particularly with regard to the\nexcitonic insulator state most likely observed in the quasi-one-dimensional\nmaterial Ta$_2$NiSe$_5$. Working with infinite boundary conditions, we are able\nto simulate all dynamical correlation functions directly in the thermodynamic\nlimit. For model parameters best suited for Ta$_2$NiSe$_5$ the photoemission\nspectra show a weak but clearly visible two-peak structure, around the Fermi\nmomenta $k\\simeq\\pm k_{\\rm F}$, which suggests that Ta$_2$NiSe$_5$ develops an\nexcitonic insulator of BCS-like type. At higher temperatures, the leakage of\nthe conduction-electron band beyond the Fermi energy becomes distinct, which\nprovides a possible explanation for the bare non-interacting band structure\nseen in time- and angle-resolved photoemission spectroscopy experiments.\n"}
{"abstract": "  Quantitative measures of uptake in caudate, putamen, and globus pallidus in\ndopamine transporter (DaT) brain SPECT have potential as biomarkers for the\nseverity of Parkinson disease. Reliable quantification of uptake requires\naccurate segmentation of these regions. However, segmentation is challenging in\nDaT SPECT due to partial-volume effects, system noise, physiological\nvariability, and the small size of these regions. To address these challenges,\nwe propose an estimation-based approach to segmentation. This approach\nestimates the posterior mean of the fractional volume occupied by caudate,\nputamen, and globus pallidus within each voxel of a 3D SPECT image. The\nestimate is obtained by minimizing a cost function based on the binary\ncross-entropy loss between the true and estimated fractional volumes over a\npopulation of SPECT images, where the distribution of the true fractional\nvolumes is obtained from magnetic resonance images from clinical populations.\nThe proposed method accounts for both the sources of partial-volume effects in\nSPECT, namely the limited system resolution and tissue-fraction effects. The\nmethod was implemented using an encoder-decoder network and evaluated using\nrealistic clinically guided SPECT simulation studies, where the ground-truth\nfractional volumes were known. The method significantly outperformed all other\nconsidered segmentation methods and yielded accurate segmentation with dice\nsimilarity coefficients of ~ 0.80 for all regions. The method was relatively\ninsensitive to changes in voxel size. Further, the method was relatively robust\nup to +/- 10 degrees of patient head tilt along transaxial, sagittal, and\ncoronal planes. Overall, the results demonstrate the efficacy of the proposed\nmethod to yield accurate fully automated segmentation of caudate, putamen, and\nglobus pallidus in 3D DaT-SPECT images.\n"}
{"abstract": "  An old theorem of Ad\\'amek constructs initial algebras for sufficiently\ncocontinuous endofunctors via transfinite iteration over ordinals in classical\nset theory. We prove a new version that works in constructive logic, using\n\"inflationary\" iteration over a notion of size that abstracts from limit\nordinals just their transitive, directed and well-founded properties. Borrowing\nfrom Taylor's constructive treatment of ordinals, we show that sizes exist with\nupper bounds for any given signature of indexes. From this it follows that\nthere is a rich class of endofunctors to which the new theorem applies,\nprovided one admits a weak form of choice (WISC) due to Streicher, Moerdijk,\nvan den Berg and Palmgren, and which is known to hold in the internal\nconstructive logic of many kinds of elementary topos.\n"}
{"abstract": "  Self-Sovereign Identity (SSI), a Blockchain-based technology for digital\nidentity management, is a promising concept for handling health data. It could\nrepresent a step forward in empowering users, granting them control over their\ndata. This work conducts a systematic literature review to investigate\nstate-of-the-art measures based on SSI and Blockchain technologies for dealing\nwith electronic health records (EHRs), identifying gaps, and determining the\nkey questions for future research. As a result, this review shows a growing\ninterest in Blockchain methods to handle EHRs, but few works consider using the\nself-sovereign identity approaches. The results obtained in this work also\nsuggest that: Blockchain technologies provide a viable alternative to deliver\nEHR solutions such as patient monitoring, healthcare data trading, and\nprescription control; consolidated Blockchain technologies are the preferred\ncore components of most effective strategies; keeping raw health data off-chain\nhelps to create scalable solutions; health data standards make searching\nmedical records in Blockchain structures feasible; Smart Contracts are\nessential components of Blockchain-based EHR solutions; the concepts of data\nownership and Self-Sovereign Identity have been neither adequately defined nor\nemployed in the health context.\n"}
{"abstract": "  This paper proposes a new defense approach for counteracting state-of-the-art\nwhite and black-box adversarial attack algorithms. Our approach fits into the\nimplicit reactive defense algorithm category since it does not directly\nmanipulate the potentially malicious input signals. Instead, it reconstructs a\nsimilar signal with a synthesized spectrogram using a cyclic generative\nadversarial network. This cyclic framework helps to yield a stable generative\nmodel. Finally, we feed the reconstructed signal into the speech-to-text model\nfor transcription. The conducted experiments on targeted and non-targeted\nadversarial attacks developed for attacking DeepSpeech, Kaldi, and Lingvo\nmodels demonstrate the proposed defense's effectiveness in adverse scenarios.\n"}
{"abstract": "  In transportation, communication, social and other real complex networks,\nsome critical edges act a pivotal part in controlling the flow of information\nand maintaining the integrity of the structure. Due to the importance of\ncritical edges in theoretical studies and practical applications, the\nidentification of critical edges gradually become a hot topic in current\nresearches. Considering the overlap of communities in the neighborhood of\nedges, a novel and effective metric named subgraph overlap (SO) is proposed to\nquantifying the significance of edges. The experimental results show that SO\noutperforms all benchmarks in identifying critical edges which are crucial in\nmaintaining the integrity of the structure and functions of networks.\n"}
{"abstract": "  We propose a new scenario of using the dark axion portal at one-loop level to\nexplain the recently observed muon anomalous magnetic moment by the Fermilab\nMuon g-2 experiment. Both axion/axion-like particle (ALP) and dark photon are\ninvolved in the same vertex with photon. Although ALP or dark photon alone\ncannot explain muon $g-2$, since the former provides only negative contribution\nwhile the latter has very much constrained parameter space, dark axion portal\ncan save the situation and significantly extend the allowed parameter space.\nThe observed muon anomalous magnetic moment provides a robust probe of the dark\naxion portal scenario.\n"}
{"abstract": "  Finding the appropriate words to convey concepts (i.e., lexical access) is\nessential for effective communication. Reverse dictionaries fulfill this need\nby helping individuals to find the word(s) which could relate to a specific\nconcept or idea. To the best of our knowledge, this resource has not been\navailable for the Persian language. In this paper, we compare four different\narchitectures for implementing a Persian reverse dictionary (PREDICT).\n  We evaluate our models using (phrase,word) tuples extracted from the only\nPersian dictionaries available online, namely Amid, Moein, and Dehkhoda where\nthe phrase describes the word. Given the phrase, a model suggests the most\nrelevant word(s) in terms of the ability to convey the concept. The model is\nconsidered to perform well if the correct word is one of its top suggestions.\n  Our experiments show that a model consisting of Long Short-Term Memory (LSTM)\nunits enhanced by an additive attention mechanism is enough to produce\nsuggestions comparable to (or in some cases better than) the word in the\noriginal dictionary. The study also reveals that the model sometimes produces\nthe synonyms of the word as its output which led us to introduce a new metric\nfor the evaluation of reverse dictionaries called Synonym Accuracy accounting\nfor the percentage of times the event of producing the word or a synonym of it\noccurs. The assessment of the best model using this new metric also indicates\nthat at least 62% of the times, it produces an accurate result within the top\n100 suggestions.\n"}
{"abstract": "  It is known that not only classical semantics but also intuitionistic Kripke\nsemantics can be generalized so that it can treat arbitrary propositional\nconnectives characterized by truth tables, or truth functions. In our previous\nwork, it has been shown that the set of Kripke-valid propositional sequents and\nthat of classically valid propositional sequents coincide if and only if all\navailable propositional connectives are monotonic. The present paper extend\nthis result to frst-order logic showing that, in the case of predicate logic,\nthe condition that all available propositional connectives are monotonic is a\nnecessary and sufficient condition for the set of sequents valid in all\nconstant domain Kripke models, not the set of Kripke-valid sequents, and the\nset of classically valid sequents to coincide.\n"}
{"abstract": "  Hadamard matrices are square $n\\times n$ matrices whose entries are ones and\nminus ones and whose rows are orthogonal to each other with respect to the\nstandard scalar product in $\\Bbb R^n$. Each Hadamard matrix can be transformed\nto a matrix whose entries are zeros and ones. This presentation of Hadamard\nmatrices is investigated in the paper and based on it an algorithm for\ngenerating them is designed.\n"}
{"abstract": "  Artificial intelligence (AI) is becoming increasingly more popular and can be\nfound in workplaces and homes around the world. However, how do we ensure trust\nin these systems? Regulation changes such as the GDPR mean that users have a\nright to understand how their data has been processed as well as saved.\nTherefore if, for example, you are denied a loan you have the right to ask why.\nThis can be hard if the method for working this out uses \"black box\" machine\nlearning techniques such as neural networks. TrustyAI is a new initiative which\nlooks into explainable artificial intelligence (XAI) solutions to address\ntrustworthiness in ML as well as decision services landscapes.\n  In this paper we will look at how TrustyAI can support trust in decision\nservices and predictive models. We investigate techniques such as LIME, SHAP\nand counterfactuals, benchmarking both LIME and counterfactual techniques\nagainst existing implementations. We also look into an extended version of\nSHAP, which supports background data selection to be evaluated based on\nquantitative data and allows for error bounds.\n"}
{"abstract": "  We investigate multi-speaker speech recognition from ultrasound images of the\ntongue and video images of the lips. We train our systems on imaging data from\nmodal speech, and evaluate on matched test sets of two speaking modes: silent\nand modal speech. We observe that silent speech recognition from imaging data\nunderperforms compared to modal speech recognition, likely due to a\nspeaking-mode mismatch between training and testing. We improve silent speech\nrecognition performance using techniques that address the domain mismatch, such\nas fMLLR and unsupervised model adaptation. We also analyse the properties of\nsilent and modal speech in terms of utterance duration and the size of the\narticulatory space. To estimate the articulatory space, we compute the convex\nhull of tongue splines, extracted from ultrasound tongue images. Overall, we\nobserve that the duration of silent speech is longer than that of modal speech,\nand that silent speech covers a smaller articulatory space than modal speech.\nAlthough these two properties are statistically significant across speaking\nmodes, they do not directly correlate with word error rates from speech\nrecognition.\n"}
{"abstract": "  Clear cell renal cell carcinoma (ccRCC) is one of the most common forms of\nintratumoral heterogeneity in the study of renal cancer. ccRCC originates from\nthe epithelial lining of proximal convoluted renal tubules. These cells undergo\nabnormal mutations in the presence of Ki67 protein and create a lump-like\nstructure through cell proliferation. Manual counting of tumor cells in the\ntissue-affected sections is one of the strongest prognostic markers for renal\ncancer. However, this procedure is time-consuming and also prone to\nsubjectivity. These assessments are based on the physical cell appearance and\nsuffer wide intra-observer variations. Therefore, better cell nucleus detection\nand counting techniques can be an important biomarker for the assessment of\ntumor cell proliferation in routine pathological investigations. In this paper,\nwe introduce a deep learning-based detection model for cell classification on\nIHC stained histology images. These images are classified into binary classes\nto find the presence of Ki67 protein in cancer-affected nucleus regions. Our\nmodel maps the multi-scale pyramid features and saliency information from local\nbounded regions and predicts the bounding box coordinates through regression.\nOur method validates the impact of Ki67 expression across a cohort of four\nhundred histology images treated with localized ccRCC and compares our results\nwith the existing state-of-the-art nucleus detection methods. The precision and\nrecall scores of the proposed method are computed and compared on the clinical\ndata sets. The experimental results demonstrate that our model improves the F1\nscore up to 86.3% and an average area under the Precision-Recall curve as\n85.73%.\n"}
{"abstract": "  Understanding surface reactivity is crucial in many fields, going from\nheterogeneous catalysis to materials oxidation and corrosion. In order to\ndecipher the surface reactions of ZrB2 exposed to the harsh environment of\naerospace components, the chemical activity of both Zr- and B-surfaces is\npredicted and compared by using density functional theory and nudged elastic\nband methods. In particular the adsorption, dissociation and diffusion of O2,\nCO and H2O are extensively examined through the calculation of surface\nadsorption energies and reaction pathways. We find the dissociative adsorption\nof O2 dominating the reactivity of ZrB2 surfaces, while the dissociation of H2O\nand CO is weakly active on Zr-surfaces, and even less activated on B-terminated\nones. Importantly, we discover that the reaction of O2 and H2O can trigger\nstrong surface reconstruction at B-surfaces. Our work thus provides significant\ninsights into the diverse adsorption and reaction mechanisms of ZrB2 surfaces.\n"}
{"abstract": "  kNN-MT, recently proposed by Khandelwal et al. (2020a), successfully combines\npre-trained neural machine translation (NMT) model with token-level\nk-nearest-neighbor (kNN) retrieval to improve the translation accuracy.\nHowever, the traditional kNN algorithm used in kNN-MT simply retrieves a same\nnumber of nearest neighbors for each target token, which may cause prediction\nerrors when the retrieved neighbors include noises. In this paper, we propose\nAdaptive kNN-MT to dynamically determine the number of k for each target token.\nWe achieve this by introducing a light-weight Meta-k Network, which can be\nefficiently trained with only a few training samples. On four benchmark machine\ntranslation datasets, we demonstrate that the proposed method is able to\neffectively filter out the noises in retrieval results and significantly\noutperforms the vanilla kNN-MT model. Even more noteworthy is that the Meta-k\nNetwork learned on one domain could be directly applied to other domains and\nobtain consistent improvements, illustrating the generality of our method. Our\nimplementation is open-sourced at https://github.com/zhengxxn/adaptive-knn-mt.\n"}
{"abstract": "  This paper presents VTN, a transformer-based framework for video recognition.\nInspired by recent developments in vision transformers, we ditch the standard\napproach in video action recognition that relies on 3D ConvNets and introduce a\nmethod that classifies actions by attending to the entire video sequence\ninformation. Our approach is generic and builds on top of any given 2D spatial\nnetwork. In terms of wall runtime, it trains $16.1\\times$ faster and runs\n$5.1\\times$ faster during inference while maintaining competitive accuracy\ncompared to other state-of-the-art methods. It enables whole video analysis,\nvia a single end-to-end pass, while requiring $1.5\\times$ fewer GFLOPs. We\nreport competitive results on Kinetics-400 and present an ablation study of VTN\nproperties and the trade-off between accuracy and inference speed. We hope our\napproach will serve as a new baseline and start a fresh line of research in the\nvideo recognition domain. Code and models are available at:\nhttps://github.com/bomri/SlowFast/blob/master/projects/vtn/README.md\n"}
{"abstract": "  This monograph, along with a self-consistent presentation of the theory of\nq-W-algebras including the construction of algebraic group analogues of Slodowy\nslices, contains a description of q-W-algebras in terms of Zhelobenko type\noperators introduced in the book. This description is applied to prove the De\nConcini-Kac-Procesi conjecture on the dimensions of irreducible modules over\nquantum groups at roots of unity.\n"}
{"abstract": "  Facial expressions vary from the visible to the subtle. In recent years, the\nanalysis of micro-expressions $-$ a natural occurrence resulting from the\nsuppression of one's true emotions, has drawn the attention of researchers with\na broad range of potential applications. However, spotting microexpressions in\nlong videos becomes increasingly challenging when intertwined with normal or\nmacro-expressions. In this paper, we propose a shallow optical flow\nthree-stream CNN (SOFTNet) model to predict a score that captures the\nlikelihood of a frame being in an expression interval. By fashioning the\nspotting task as a regression problem, we introduce pseudo-labeling to\nfacilitate the learning process. We demonstrate the efficacy and efficiency of\nthe proposed approach on the recent MEGC 2020 benchmark, where state-of-the-art\nperformance is achieved on CAS(ME)$^{2}$ with equally promising results on SAMM\nLong Videos.\n"}
{"abstract": "  Face recognition is a popular and well-studied area with wide applications in\nour society. However, racial bias had been proven to be inherent in most State\nOf The Art (SOTA) face recognition systems. Many investigative studies on face\nrecognition algorithms have reported higher false positive rates of African\nsubjects cohorts than the other cohorts. Lack of large-scale African face image\ndatabases in public domain is one of the main restrictions in studying the\nracial bias problem of face recognition. To this end, we collect a face image\ndatabase namely CASIA-Face-Africa which contains 38,546 images of 1,183 African\nsubjects. Multi-spectral cameras are utilized to capture the face images under\nvarious illumination settings. Demographic attributes and facial expressions of\nthe subjects are also carefully recorded. For landmark detection, each face\nimage in the database is manually labeled with 68 facial keypoints. A group of\nevaluation protocols are constructed according to different applications,\ntasks, partitions and scenarios. The performances of SOTA face recognition\nalgorithms without re-training are reported as baselines. The proposed database\nalong with its face landmark annotations, evaluation protocols and preliminary\nresults form a good benchmark to study the essential aspects of face biometrics\nfor African subjects, especially face image preprocessing, face feature\nanalysis and matching, facial expression recognition, sex/age estimation,\nethnic classification, face image generation, etc. The database can be\ndownloaded from our http://www.cripacsir.cn/dataset/\n"}
{"abstract": "  We study supervised learning algorithms in which a quantum device is used to\nperform a computational subroutine - either for prediction via probability\nestimation, or to compute a kernel via estimation of quantum states overlap. We\ndesign implementations of these quantum subroutines using Boson Sampling\narchitectures in linear optics, supplemented by adaptive measurements. We then\nchallenge these quantum algorithms by deriving classical simulation algorithms\nfor the tasks of output probability estimation and overlap estimation. We\nobtain different classical simulability regimes for these two computational\ntasks in terms of the number of adaptive measurements and input photons. In\nboth cases, our results set explicit limits to the range of parameters for\nwhich a quantum advantage can be envisaged with adaptive linear optics compared\nto classical machine learning algorithms: we show that the number of input\nphotons and the number of adaptive measurements cannot be simultaneously small\ncompared to the number of modes. Interestingly, our analysis leaves open the\npossibility of a near-term quantum advantage with a single adaptive\nmeasurement.\n"}
{"abstract": "  We present calculations of the photoionization (PI) cross sections of\nrubidium and cesium Rydberg atoms for light with wavelengths ranging from the\ninfrared to the ultraviolet, using model potentials from [M. Marinescu, H. R.\nSadeghpour, and A. Dalgarno, Phys. Rev. A 49, 982 (1994)]. The origins of\npronounced PI minima are identified by investigating the free-electron\nwavefunctions. These include broad PI minima in the $nS$ to $\\epsilon P$ PI\nchannels of both Rb and Cs, with free-electron energy $\\epsilon$, which are\nidentified as Cooper minima. Much narrower PI minima in the $nD$ to $\\epsilon\nF$ channels are due to shape resonances of the free-electron states. We\ndescribe possible experimental procedures for measuring the PI minima, and we\ndiscuss their implications in fundamental atomic physics as well as in\npractical applications.\n"}
{"abstract": "  We study the problem of how to construct a set of policies that can be\ncomposed together to solve a collection of reinforcement learning tasks. Each\ntask is a different reward function defined as a linear combination of known\nfeatures. We consider a specific class of policy compositions which we call set\nimproving policies (SIPs): given a set of policies and a set of tasks, a SIP is\nany composition of the former whose performance is at least as good as that of\nits constituents across all the tasks. We focus on the most conservative\ninstantiation of SIPs, set-max policies (SMPs), so our analysis extends to any\nSIP. This includes known policy-composition operators like generalized policy\nimprovement. Our main contribution is a policy iteration algorithm that builds\na set of policies in order to maximize the worst-case performance of the\nresulting SMP on the set of tasks. The algorithm works by successively adding\nnew policies to the set. We show that the worst-case performance of the\nresulting SMP strictly improves at each iteration, and the algorithm only stops\nwhen there does not exist a policy that leads to improved performance. We\nempirically evaluate our algorithm on a grid world and also on a set of domains\nfrom the DeepMind control suite. We confirm our theoretical results regarding\nthe monotonically improving performance of our algorithm. Interestingly, we\nalso show empirically that the sets of policies computed by the algorithm are\ndiverse, leading to different trajectories in the grid world and very distinct\nlocomotion skills in the control suite.\n"}
{"abstract": "  In this paper, we present an update to the NISQA speech quality prediction\nmodel that is focused on distortions that occur in communication networks. In\ncontrast to the previous version, the model is trained end-to-end and the\ntime-dependency modelling and time-pooling is achieved through a Self-Attention\nmechanism. Besides overall speech quality, the model also predicts the four\nspeech quality dimensions Noisiness, Coloration, Discontinuity, and Loudness,\nand in this way gives more insight into the cause of a quality degradation.\nFurthermore, new datasets with over 13,000 speech files were created for\ntraining and validation of the model. The model was finally tested on a new,\nlive-talking test dataset that contains recordings of real telephone calls.\nOverall, NISQA was trained and evaluated on 81 datasets from different sources\nand showed to provide reliable predictions also for unknown speech samples. The\ncode, model weights, and datasets are open-sourced.\n"}
{"abstract": "  We extend the GENEVA Monte Carlo framework using the transverse momentum of a\ncolour-singlet system as the resolution variable. This allows us to use\nnext-to-next-to-next-to leading logarithm (N$^3$LL) resummation via the RadISH\nformalism to obtain precise predictions for any colour-singlet production\nprocess at the fully exclusive level. Thanks to the implementation of two\ndifferent resolution variables within the GENEVA framework, we are able to\nassess the impact of such a choice on differential observables for the first\ntime. As a first application we present predictions for Drell-Yan lepton pair\nproduction at next-to-next-to-leading order (NNLO) in QCD interfaced to a\nparton shower simulation that includes additional all-order radiative\ncorrections. We provide fully showered and hadronised events using PYTHIA 8,\nwhile retaining the NNLO QCD accuracy for observables which are inclusive over\nthe additional radiation. We compare our final predictions to LHC data at 13\nTeV, finding good agreement across several distributions.\n"}
{"abstract": "  We show that ultradilute quantum liquids can be formed with ultracold bosonic\ndipolar atoms in a bilayer geometry. Contrary to previous realizations of\nultradilute liquids, there is no need of stabilizing the system with an\nadditional repulsive short-range potential. The advantage of the proposed\nsystem is that dipolar interactions on their own are sufficient for creation of\na self-bound state and no additional short-range potential is needed for the\nstabilization. We perform quantum Monte Carlo simulations and find a rich\nground state phase diagram that contains quantum phase transitions between\nliquid, solid, atomic gas, and molecular gas phases. The stabilization\nmechanism of the liquid phase is consistent with the microscopic scenario in\nwhich the effective dimer-dimer attraction is balanced by an effective\nthree-dimer repulsion. The equilibrium density of the liquid, which is\nextremely small, can be controlled by the interlayer distance. From the\nequation of state, we extract the spinodal density, below which the homogeneous\nsystem breaks into droplets. Our results offer a new example of a\ntwo-dimensional interacting dipolar liquid in a clean and highly controllable\nsetup.\n"}
{"abstract": "  We study the jumps in the archimedean height of the Ceresa cycle, as\nintroduced by R. Hain in his work on normal functions on moduli spaces of\ncurves, and as further analyzed by P. Brosnan and G. Pearlstein in terms of\nasymptotic Hodge theory. Our work is based on a study of the asymptotic\nbehavior of the Hain-Reed beta-invariant in degenerating families of curves. We\nshow that the height jump of the Ceresa cycle at a given stable curve is equal\nto the so-called \"slope\" of the dual graph of the curve, and we characterize\nthose stable curves for which the height jump vanishes. We also obtain an\nanalytic formula for the height of the Ceresa cycle for a curve over a function\nfield over the complex numbers, and characterize in analytic terms when the\nheight of the Ceresa cycle vanishes.\n"}
{"abstract": "  The N-body gauge allows the introduction of relativistic effects in Newtonian\ncosmological simulations. Here we extend this framework to general Horndeski\ngravity theories, and investigate the relativistic effects that the scalar\nfield introduces in the matter power spectrum at intermediate and large scales.\nIn particular, we show that the kineticity function at these scales enhances\nthe amplitude of the signal of contributions coming from the extra degree of\nfreedom. Using the Quasi-Static Approximation (QSA), we separate modified\ngravity effects into two parts: one that only affects small-scale physics, and\none that is due to relativistic effects. This allows our formalism to be\nreadily implemented in modified gravity N-body codes in a straightforward\nmanner, e.g., relativistic effects can be included as an additional linear\ndensity field in simulations. We identify the emergence of gravity acoustic\noscillations (GAOs) in the matter power spectrum at large scales, $k \\sim\n10^{-3}-10^{-2}$ Mpc$^{-1}$. GAO features have a purely relativistic origin,\ncoming from the dynamical nature of the scalar field. GAOs may be enhanced to\ndetectable levels by the rapid evolution of the dark energy sound horizon in\ncertain modified gravity models and can be seen as a new test of gravity at\nscales probed by future galaxy and intensity-mapping surveys.\n"}
{"abstract": "  Simultaneous Localization And Mapping (SLAM) is a task to estimate the robot\nlocation and to reconstruct the environment based on observation from sensors\nsuch as LIght Detection And Ranging (LiDAR) and camera. It is widely used in\nrobotic applications such as autonomous driving and drone delivery. Traditional\nLiDAR-based SLAM algorithms mainly leverage the geometric features from the\nscene context, while the intensity information from LiDAR is ignored. Some\nrecent deep-learning-based SLAM algorithms consider intensity features and\ntrain the pose estimation network in an end-to-end manner. However, they\nrequire significant data collection effort and their generalizability to\nenvironments other than the trained one remains unclear. In this paper we\nintroduce intensity features to a SLAM system. And we propose a novel full SLAM\nframework that leverages both geometry and intensity features. The proposed\nSLAM involves both intensity-based front-end odometry estimation and\nintensity-based back-end optimization. Thorough experiments are performed\nincluding both outdoor autonomous driving and indoor warehouse robot\nmanipulation. The results show that the proposed method outperforms existing\ngeometric-only LiDAR SLAM methods.\n"}
{"abstract": "  Locomotion over soft terrain remains a challenging problem for legged robots.\nMost of the work done on state estimation for legged robots is designed for\nrigid contacts, and does not take into account the physical parameters of the\nterrain. That said, this letter answers the following questions: how and why\ndoes soft terrain affect state estimation for legged robots? To do so, we\nutilized a state estimator that fuses IMU measurements with leg odometry that\nis designed with rigid contact assumptions. We experimentally validated the\nstate estimator with the HyQ robot trotting over both soft and rigid terrain.\nWe demonstrate that soft terrain negatively affects state estimation for legged\nrobots, and that the state estimates have a noticeable drift over soft terrain\ncompared to rigid terrain.\n"}
{"abstract": "  When a single-layer graphene sheet is contacted with metallic electrodes,\ntunnel barriers are formed as a result of the doping of graphene by the metal\nin the contact region. If the Fermi energy level is modulated by a gate\nvoltage, the phenomenon of Klein tunneling results in specific features in the\nconductance and noise. Here we obtain analytically exact solutions for the\ntransmission and reflection probability amplitudes using a trapezoidal\npotential barrier, allowing us to calculate the differential conductance and\nthe Fano factor for a graphene sheet in the ballistic regime. We put in\nevidence an unexpected global symmetry - the transmission probability is the\nsame for energies symmetric with respect to half of the barrier height. We\noutline a proposal for the experimental verification of these ideas using\nrealistic sample parameters.\n"}
{"abstract": "  A maximum sequence $S$ of vertices in a graph $G$, so that every vertex in\n$S$ has a neighbor which is independent, or is itself independent, from all\nprevious vertices in $S$, is called a Grundy dominating sequence. The Grundy\ndomination number, $\\gamma_{gr}(G)$, is the length of $S$. We show that for any\nforest $F$, $\\gamma_{gr}(F)=|V(T)|-|\\mathcal{P}|$ where $\\mathcal{P}$ is a\nminimum partition of the non-isolate vertices of $F$ into caterpillars in which\nif two caterpillars of $\\mathcal{P}$ have an edge between them in $F$, then\nsuch an edge must be incident to a non-leaf vertex in at least one of the\ncaterpillars. We use this result to show the strong product conjecture of B.\nBre\\v{s}ar, Cs. Bujt\\'{a}s, T. Gologranc, S. Klav\\v{z}ar, G. Ko\\v{s}mrlj, B.\nPatk\\'{o}s, Zs. Tuza, and M. Vizer, Dominating sequences in grid-like and\ntoroidal graphs, Electron. J. Combin. 23(4): P4.34 (2016), for all forests.\nNamely, we show that for any forest $G$ and graph $H$, $\\gamma_{gr}(G \\boxtimes\nH) = \\gamma_{gr}(G) \\gamma_{gr}(H)$. We also show that every connected graph\n$G$ has a spanning tree $T$ so that $\\gamma_{gr}(G)\\le \\gamma_{gr}(T)$ and that\nevery non-complete connected graph contains a Grundy dominating set $S$ so that\nthe induced subgraph of $S$ contains no isolated vertices.\n"}
{"abstract": "  A positive integer is called an $E_j$-number if it is the product of $j$\ndistinct primes. We prove that there are infinitely many triples of\n$E_2$-numbers within a gap size of $32$ and infinitely many triples of\n$E_3$-numbers within a gap size of $15$. Assuming the Elliot-Halberstam\nconjecture for primes and $E_2$-numbers, we can improve these gaps to $12$ and\n$5$, respectively. We can obtain even smaller gaps for almost primes, almost\nprime powers, or integers having the same exponent pattern in the their prime\nfactorizations. In particular, if $d(x)$ denotes the number of divisors of $x$,\nwe prove that there are integers $a,b$ with $1\\leq a < b \\leq 9$ such that\n$d(x)=d(x+a)=d(x+b) = 192$ for infinitely many $x$. Assuming Elliot-Halberstam,\nwe prove that there are integers $a,b$ with $1\\leq a < b \\leq 4$ such that\n$d(x)=d(x+a)=d(x+b) = 24$ for infinitely many $x$.\n"}
{"abstract": "  We study the Thermodynamic Bethe Ansatz (TBA) equations for pure $T\\bar{T}$\nperturbations of some simple integrable quantum field theories with a single\nbosonic or fermionic particle, in particular the massive sinh-Gordon model and\nits ultraviolet (UV) limit which is a deformation of the conformally invariant\nfree massless boson. Whereas the TBA equations for $T\\bar{T}$ deformations of\nmassive theories are in principle known, the TBA equations we propose for the\ndeformations of conformal field theories (CFT's) are relatively new and require\na special factorization in rapidity variables of the CDD factor for the\nscattering of the massless particles. The latter TBA equations can be solved\nexactly and reproduce the known results for the ground state energy on a\ncylinder of circumference $R$ which were previously obtained using different\nmethods based for instance on the Burgers differential equation. Special\nattention is paid to the c-theorem in this context which is discussed in some\ndetail. For positive infra-red (IR) central charge $c_{IR}$, for flows\nconsistent with the c-theorem the ground state energy develops a (previously\nknown) square-root singularity towards the UV, which strongly suggests the\ntheories are UV incomplete in these physically important cases. We suggest that\nthe singularity indicates a tachyonic vacuum instability. Other cases with\n$c_{IR} < 0$ do not have this singularity and are interpreted as being UV\ncomplete with $c_{UV} = 0$. We extend our results to a continuously variable\n$c_{IR}$ by introducing a chemical potential and suggest this as a possible toy\nmodel for the $T\\bar{T}$ perturbed Liouville theory.\n"}
{"abstract": "  This paper focuses on the numerical simulation of geothermal systems in\ncomplex geological settings. The physical model is based on two-phase Darcy\nflows coupling the mass conservation of the water component with the energy\nconservation and the liquid vapor thermodynamical equilibrium. The\ndiscretization exploits the flexibility of unstructured meshes to model complex\ngeology including conductive faults as well as complex wells. The polytopal and\nessentially nodal Vertex Approximate Gradient scheme is used for the\napproximation of the Darcy and Fourier fluxes combined with a Control Volume\napproach for the transport of mass and energy. Particular attention is paid to\nthe faults which are modelled as two-dimensional interfaces defined as\ncollection of faces of the mesh and to the flow inside deviated or multi-branch\nwells defined as collection of edges of the mesh with rooted tree data\nstructure. By using an explicit pressure drop calculation, the well model\nreduces to a single equation based on complementarity constraints with only one\nwell implicit unknown. The coupled systems are solved fully implicitely at each\ntime step using efficient nonlinear and linear solvers on parallel distributed\narchitectures. The convergence of the discrete model is investigated\nnumerically on a simple test case with a Cartesian geometry and a single\nvertical producer well. Then, the ability of our approach to deal efficiently\nwith realistic test cases is assessed on a high energy faulted geothermal\nreservoir operated using a doublet of two deviated wells.\n"}
{"abstract": "  Cosmological domain walls can be formed as a result of symmetry breaking at\nany epoch during the evolution of our universe. We study their interaction with\na classical macroscopic object, like Earth or a satellite in Earth's orbit. We\nset up an action that includes the interaction term between the massive\nclassical object and the scalar field that the domain wall is made of. We use\nnumerical calculations to solve the coupled equations of motion which describe\nthe crossing between the domain wall and the classical object. Depending on the\nstrength of the interaction, relative velocity and size, the object can be\neither stopped by the wall, or it can pass through it inducing deformations in\nthe wall that cost energy. At the same time, the coupling to the scalar filed\nmight change the object's mass during the crossover. The fact that satellites\nin Earth's orbit (or planets in Sun's orbit) can change their mass and/or lose\nenergy interacting with walls can be used as a new domain wall detection probe.\nFor example, a typical velocity precision of a satellite is about $0.5$ mm/s,\nwhich directly puts an upper limit on its mass change to $\\Delta M/M\n\\lessapprox 5\\times 10^{-17} $. Alternatively, a known satellite flyby anomaly\ncan easily be explained as an interaction with a closed domain wall. We also\nshow that the presence of matter modifies the scalar filed potential and can\nlocally create a bubble of the true vacuum, and thus trigger the decay of the\nfalse vacuum. For a critical bubble which is able to expand, such an\ninteraction with the domain wall must be strong enough.\n"}
{"abstract": "  In active galactic nuclei (AGN), fluorescent Fe K$\\alpha$ (iron) line\nemission is generally interpreted as originating from obscuring material around\na supermassive black hole (SMBH) on the scale of a few parsecs (pc). However,\nrecent Chandra studies indicate the existence of iron line emission extending\nto kpc scales in the host galaxy. The connection between iron line emission and\nlarge-scale material can be spatially resolved directly only in nearby\ngalaxies, but could be inferred in more distant AGNs by a connection between\nline emission and star-forming gas and dust that is more extended than the\npc-scale torus. Here we present the results from a stacking analysis and X-ray\nspectral fitting performed on sources in the Chandra Deep Field South (CDFS) 7\nMs observations. From the deep stacked spectra, we select sources with stellar\nmass $\\log(M_*/M_\\odot)>10$ at $0.5<z<2$, obtaining 25 sources with high\ninfrared luminosity ($ {\\rm SFR}_{\\rm FIR} \\geq 17\\;M_{\\odot}\\;{\\rm yr}^{-1}$)\nand 32 sources below this threshold. We find that the equivalent width of the\niron line EW(Fe) is a factor of three higher with 3$\\sigma$ significance for\nhigh infrared luminosity measured from Herschel observations, indicating a\nconnection between iron line emission and star-forming material on galaxy\nscales. We show that there is no significant dependence in EW(Fe) on $M_*$ or\nX-ray luminosity, suggesting the reflection of AGN X-ray emission over large\nscales in their host galaxies may be widespread.\n"}
{"abstract": "  For fixed functions $G,H:[0,\\infty)\\to[0,\\infty)$, consider the rotationally\ninvariant probability density on $\\mathbb{R}^n$ of the form \\[ \\mu^n(ds) =\n\\frac{1}{Z_n} G(\\|s\\|_2)\\, e^{ - n H( \\|s\\|_2)} ds. \\] We show that when $n$ is\nlarge, the Euclidean norm $\\|Y^n\\|_2$ of a random vector $Y^n$ distributed\naccording to $\\mu^n$ satisfies a Gaussian thin-shell property: the distribution\nof $\\|Y^n\\|_2$ concentrates around a certain value $s_0$, and the fluctuations\nof $\\|Y^n\\|_2$ are approximately Gaussian with the order $1/\\sqrt{n}$. We apply\nthis thin shell property to the study of rotationally invariant random\nsimplices, simplices whose vertices consist of the origin as well as\nindependent random vectors $Y_1^n,\\ldots,Y_p^n$ distributed according to\n$\\mu^n$. We show that the logarithmic volume of the resulting simplex exhibits\nhighly Gaussian behavior, providing a generalizing and unifying setting for the\nobjects considered in Grote-Kabluchko-Th\\\"ale [Limit theorems for random\nsimplices in high dimensions, ALEA, Lat. Am. J. Probab. Math. Stat. 16,\n141--177 (2019)]. Finally, by relating the volumes of random simplices to\nrandom determinants, we show that if $A^n$ is an $n \\times n$ random matrix\nwhose entries are independent standard Gaussian random variables, then there\nare explicit constants $c_0,c_1\\in(0,\\infty)$ and an absolute constant\n$C\\in(0,\\infty)$ such that \\[\\sup_{ s \\in \\mathbb{R}} \\left| \\mathbb{P} \\left[\n\\frac{ \\log \\mathrm{det}(A^n) - \\log(n-1)! - c_0 }{ \\sqrt{ \\frac{1}{2} \\log n +\nc_1 }} < s \\right] - \\int_{-\\infty}^s \\frac{e^{ - u^2/2} du}{ \\sqrt{ 2 \\pi }}\n\\right| < \\frac{C}{\\log^{3/2}n}, \\] sharpening the $1/\\log^{1/3 + o(1)}n$ bound\nin Nguyen and Vu [Random matrices: Law of the determinant, Ann. Probab. 42 (1)\n(2014), 146--167].\n"}
{"abstract": "  The compression quality losses of depth sequences determine quality of view\nsynthesis in free-viewpoint video. The depth map intra prediction in 3D\nextensions of the HEVC applies intra modes with auxiliary depth modeling modes\n(DMMs) to better preserve depth edges and handle motion discontinuities.\nAlthough such modes enable high efficiency compression, but at the cost of very\nhigh encoding complexity. Skipping conventional intra coding modes and DMMs in\ndepth coding limits practical applicability of the HEVC for 3D display\napplications. In this paper, we introduce a novel low-complexity scheme for\ndepth video compression based on low-rank tensor decomposition and HEVC intra\ncoding. The proposed scheme leverages spatial and temporal redundancy by\ncompactly representing the depth sequence as a high-order tensor. Tensor\nfactorization into a set of factor matrices following CANDECOMP PARAFAC (CP)\ndecomposition via alternating least squares give a low-rank approximation of\nthe scene geometry. Further, compression of factor matrices with HEVC intra\nprediction support arbitrary target accuracy by flexible adjustment of bitrate,\nvarying tensor decomposition ranks and quantization parameters. The results\ndemonstrate proposed approach achieves significant rate gains by efficiently\ncompressing depth planes in low-rank approximated representation. The proposed\nalgorithm is applied to encode depth maps of benchmark Ballet and Breakdancing\nsequences. The decoded depth sequences are used for view synthesis in a\nmulti-view video system, maintaining appropriate rendering quality.\n"}
{"abstract": "  Adversarial regularization has been shown to improve the generalization\nperformance of deep learning models in various natural language processing\ntasks. Existing works usually formulate the method as a zero-sum game, which is\nsolved by alternating gradient descent/ascent algorithms. Such a formulation\ntreats the adversarial and the defending players equally, which is undesirable\nbecause only the defending player contributes to the generalization\nperformance. To address this issue, we propose Stackelberg Adversarial\nRegularization (SALT), which formulates adversarial regularization as a\nStackelberg game. This formulation induces a competition between a leader and a\nfollower, where the follower generates perturbations, and the leader trains the\nmodel subject to the perturbations. Different from conventional approaches, in\nSALT, the leader is in an advantageous position. When the leader moves, it\nrecognizes the strategy of the follower and takes the anticipated follower's\noutcomes into consideration. Such a leader's advantage enables us to improve\nthe model fitting to the unperturbed data. The leader's strategic information\nis captured by the Stackelberg gradient, which is obtained using an unrolling\nalgorithm. Our experimental results on a set of machine translation and natural\nlanguage understanding tasks show that SALT outperforms existing adversarial\nregularization baselines across all tasks. Our code is publicly available.\n"}
{"abstract": "  Substructured domain decomposition (DD) methods have been extensively\nstudied, and they are usually associated with nonoverlapping decompositions. We\nintroduce here a substructured version of Restricted Additive Schwarz (RAS)\nwhich we call SRAS, and we discuss its advantages compared to the standard\nvolume formulation of the Schwarz method when they are used both as iterative\nsolvers and preconditioners for a Krylov method. To extend SRAS to nonlinear\nproblems, we introduce SRASPEN (Substructured Restricted Additive Schwarz\nPreconditioned Exact Newton), where SRAS is used as a preconditioner for\nNewton's method. We study carefully the impact of substructuring on the\nconvergence and performance of these methods as well as their implementations.\nWe finally introduce two-level versions of nonlinear SRAS and SRASPEN.\nNumerical experiments confirm the advantages of formulating a Schwarz method at\nthe substructured level.\n"}
{"abstract": "  More than 10% of the barred galaxies with a direct measurement of the bar\npattern speed host an ultrafast bar. These bars extend beyond the corotation\nradius and challenge our understanding of the orbital structure of barred\ngalaxies. Most of them are found in spiral galaxies, rather than in lenticular\nones. We analysed the properties of the ultrafast bars detected in the CALIFA\nSurvey to investigate whether they are an artefact resulting from an\noverestimation of the bar radius and/or an underestimation of the corotation\nradius or a new class of bars, whose orbital structure has not yet been\nunderstood. We revised the available measurements of the bar radius based on\nellipse fitting and Fourier analysis and of the bar pattern speed from the\nTremaine-Weinberg method. In addition, we measured the bar radius from the\nanalysis of the maps tracing the transverse-to-radial force ratio, which we\nobtained from the deprojected i-band images of the galaxies retrieved from the\nSDSS Survey. We found that nearly all the sample galaxies are spirals with an\ninner ring or pseudo-ring circling the bar and/or strong spiral arms, which\nhamper the measurement of the bar radius from the ellipse fitting and Fourier\nanalysis. According to these methods, the bar ends overlap the ring or the\nspiral arms making the adopted bar radius unreliable. On the contrary, the bar\nradius from the ratio maps are shorter than the corotation radius. This is in\nagreement with the theoretical predictions and findings of numerical\nsimulations about the extension and stability of the stellar orbits supporting\nthe bars. We conclude that ultrafast bars are no longer observed when the\ncorrect measurement of the bar radius is adopted. Deriving the bar radius in\ngalaxies with rings and strong spiral arms is not straightforward and a solid\nmeasurement method based on both photometric and kinematic data is still\nmissing.\n"}
{"abstract": "  We present an elegant derivation of the Doppler factor in the\nLienard-Wiechert potentials of an extended particle, based on a theorem\nauthored by J. L. Synge and dealing with cross-sections through worldtubes.\n"}
{"abstract": "  We introduce a notion of emergence for coarse-grained macroscopic variables\nassociated with highly-multivariate microscopic dynamical processes, in the\ncontext of a coupled dynamical environment. Dynamical independence instantiates\nthe intuition of an emergent macroscopic process as one possessing the\ncharacteristics of a dynamical system \"in its own right\", with its own\ndynamical laws distinct from those of the underlying microscopic dynamics. We\nquantify (departure from) dynamical independence by a transformation-invariant\nShannon information-based measure of dynamical dependence. We emphasise the\ndata-driven discovery of dynamically-independent macroscopic variables, and\nintroduce the idea of a multiscale \"emergence portrait\" for complex systems. We\nshow how dynamical dependence may be computed explicitly for linear systems via\nstate-space modelling, in both time and frequency domains, facilitating\ndiscovery of emergent phenomena at all spatiotemporal scales. We discuss\napplication of the state-space operationalisation to inference of the emergence\nportrait for neural systems from neurophysiological time-series data. We also\nexamine dynamical independence for discrete- and continuous-time deterministic\ndynamics, with potential application to Hamiltonian mechanics and classical\ncomplex systems such as flocking and cellular automata.\n"}
{"abstract": "  Crossbar resistive memory with the 1 Selector 1 Resistor (1S1R) structure is\nattractive for nonvolatile, high-density, and low-latency storage-class memory\napplications. As technology scales down to the single-nm regime, the increasing\nresistivity of wordline/bitline becomes a limiting factor to device\nreliability. This paper presents write/read communication channels while\nconsidering the line resistance and device variabilities by statistically\nrelating the degraded write/read margins and the channel parameters. Binary\nasymmetric channel (BAC) models are proposed for the write/read operations.\nSimulations based on these models suggest that the bit-error rate of devices\nare highly non-uniform across the memory array. These models provide\nquantitative tools for evaluating the trade-offs between memory reliability and\ndesign parameters, such as array size, technology nodes, and aspect ratio, and\nalso for designing coding-theoretic solutions that would be most effective for\ncrossbar memory. Method for optimizing the read threshold is proposed to reduce\nthe raw bit-error rate (RBER). We propose two schemes for efficient channel\ncoding based on Bose-Chaudhuri-Hocquenghem (BCH) codes. An interleaved coding\nscheme is proposed to mitigate the non-uniformity of reliability and a location\ndependent coding framework is proposed to leverage this non-uniformity. Both of\nour proposed coding schemes effectively reduce the undetected bit-error rate\n(UBER).\n"}
{"abstract": "  In this paper we investigate the amount of spatial context required for\nchannel attention. To this end we study the popular squeeze-and-excite (SE)\nblock which is a simple and lightweight channel attention mechanism. SE blocks\nand its numerous variants commonly use global average pooling (GAP) to create a\nsingle descriptor for each channel. Here, we empirically analyze the amount of\nspatial context needed for effective channel attention and find that limited\nlocalcontext on the order of seven rows or columns of the original image is\nsufficient to match the performance of global context. We propose tiled\nsqueeze-and-excite (TSE), which is a framework for building SE-like blocks that\nemploy several descriptors per channel, with each descriptor based on local\ncontext only. We further show that TSE is a drop-in replacement for the SE\nblock and can be used in existing SE networks without re-training. This implies\nthat local context descriptors are similar both to each other and to the global\ncontext descriptor. Finally, we show that TSE has important practical\nimplications for deployment of SE-networks to dataflow AI accelerators due to\ntheir reduced pipeline buffering requirements. For example, using TSE reduces\nthe amount of activation pipeline buffering in EfficientDetD2 by 90% compared\nto SE (from 50M to 4.77M) without loss of accuracy. Our code and pre-trained\nmodels will be publicly available.\n"}
{"abstract": "  When data is of an extraordinarily large size or physically stored in\ndifferent locations, the distributed nearest neighbor (NN) classifier is an\nattractive tool for classification. We propose a novel distributed adaptive NN\nclassifier for which the number of nearest neighbors is a tuning parameter\nstochastically chosen by a data-driven criterion. An early stopping rule is\nproposed when searching for the optimal tuning parameter, which not only speeds\nup the computation but also improves the finite sample performance of the\nproposed Algorithm. Convergence rate of excess risk of the distributed adaptive\nNN classifier is investigated under various sub-sample size compositions. In\nparticular, we show that when the sub-sample sizes are sufficiently large, the\nproposed classifier achieves the nearly optimal convergence rate. Effectiveness\nof the proposed approach is demonstrated through simulation studies as well as\nan empirical application to a real-world dataset.\n"}
{"abstract": "  We investigate the exact overlaps between eigenstates of integrable spin\nchains and a special class of states called \"integrable initial/final states\".\nThese states satisfy a special integrability constraint, and they are closely\nrelated to integrable boundary conditions. We derive new algebraic relations\nfor the integrable states, which lead to a set of recursion relations for the\nexact overlaps. We solve these recursion relations and thus we derive new\noverlap formulas, valid in the XXX Heisenberg chain and its integrable higher\nspin generalizations. Afterwards we generalize the integrability condition to\ntwisted boundary conditions, and derive the corresponding exact overlaps.\nFinally, we embed the integrable states into the \"Separation of Variables\"\nframework, and derive an alternative representation for the exact overlaps of\nthe XXX chain. Our derivations and proofs are rigorous, and they can form the\nbasis of future investigations involving more complicated models such as nested\nor long-range deformed systems.\n"}
{"abstract": "  An important class of cyber-physical systems relies on multiple agents that\njointly perform a task by coordinating their actions over a wireless network.\nExamples include self-driving cars in intelligent transportation and production\nrobots in smart manufacturing. However, the scalability of existing\ncontrol-over-wireless solutions is limited as they cannot resolve overload\nsituations in which the communication demand exceeds the available bandwidth.\nThis paper presents a novel co-design of distributed control and wireless\ncommunication that overcomes this limitation by dynamically allocating the\navailable bandwidth to agents with the greatest need to communicate.\nExperiments on a real cyber-physical testbed with 20 agents, each consisting of\na wireless node and a cart-pole system, demonstrate that our solution achieves\nsignificantly better control performance under overload than the state of the\nart. We further prove that our co-design guarantees closed-loop stability for\nphysical systems with stochastic linear time-invariant dynamics.\n"}
{"abstract": "  We analyze behavior of weak solutions to compressible fluid flows in a\nbounded domain in $\\mathbb{R}^3$, randomly perforated by tiny balls with random\nsize. Assuming the radii of the balls scale like $\\varepsilon^\\alpha$, $\\alpha\n> 3$, with $\\varepsilon$ denoting the average distance between the balls, the\nproblem homogenize with the same limiting equation. Our main contribution is a\nconstruction of the Bogovski\\u{\\i} operator, uniformly in $\\varepsilon$,\nwithout any assumptions on the minimal distance between the balls.\n"}
{"abstract": "  Cuspy shadow was first reported for hairy rotating black holes, whose metrics\ndeviate significantly from the Kerr one. The non-smooth edge of the shadow is\nattributed to a transition between different branches of unstable but bounded\norbits, known as the fundamental photon orbits, which end up at the light\nrings. In searching for a minimal theoretical setup to reproduce such a salient\nfeature, in this work, we devise a toy model with axisymmetry, a slowly\nrotating Kerr black hole enveloped by a thin slowly rotating dark matter shell.\nDespite its simplicity, we show rich structures regarding fundamental photon\norbits explicitly in such a system. We observe two disconnected branches of\nunstable spherical photon orbits, and the jump between them gives rise to a\npair of cusps in the resultant black hole shadow. Besides the cuspy shadow, we\nexplore other intriguing phenomena when the Maxwell construction cannot be\nestablished. We find that it is possible to have an incomplete arc of Einstein\nrings and a \"fractured\" shadow. The potential astrophysical significance of the\ncorresponding findings is addressed.\n"}
{"abstract": "  The ever-continuing explosive growth of on-demand content requests has\nimposed great pressure on mobile/wireless network infrastructures. To ease\ncongestion in the network and increase perceived user experience, caching\npopular content closer to the end-users can play a significant role and as such\nthis issue received significant attention over the last few years.\nAdditionally, energy efficiency is treated as a fundamental requirement in the\ndesign of next-generation mobile networks. However, there has been little\nattention to the overlapping area between energy efficiency and network caching\nespecially when considering multipath routing. To this end, this paper proposes\nan energy-efficient caching with multipath routing support. The proposed scheme\nprovides a joint anchoring of popular content into a set of potential caching\nnodes with optimized multi-path support whilst ensuring a balance between\ntransmission and caching energy cost. The proposed model also considers\ndifferent content delivery modes, such as multicast and unicast. Two separated\nInteger-Linear Programming (ILP) models are formulated for each delivery mode.\nTo tackle the curse of dimensionality we then provide a greedy simulated\nannealing algorithm, which not only reduces the time complexity but also\nprovides a competitive performance as ILP models. A wide set of numerical\ninvestigations has shown that the proposed scheme is more energy-efficient\ncompared with other widely used approaches in caching under the premise of\nnetwork resource limitation.\n"}
{"abstract": "  Consider an elliptic self-adjoint pseudodifferential operator $A$ acting on\n$m$-columns of half-densities on a closed manifold $M$, whose principal symbol\nis assumed to have simple eigenvalues. Relying on a basis of pseudodifferential\nprojections commuting with $A$, we construct an almost-unitary\npseudodifferential operator that diagonalizes $A$ modulo an infinitely\nsmoothing operator. We provide an invariant algorithm for the computation of\nits full symbol, as well as an explicit closed formula for its subprincipal\nsymbol. Finally, we give a quantitative description of the relation between the\nspectrum of $A$ and the spectrum of its approximate diagonalization, and\ndiscuss the implications at the level of spectral asymptotics.\n"}
{"abstract": "  The Ces\\`aro limit - the asymptotic average of a sequence of real numbers -\nis an operator of fundamental importance in probability, statistics and\nmathematical analysis. To better understand sequences with Ces\\`aro limits,\nthis paper considers the space $\\mathcal{F}$ comprised of all binary sequences\nwith a Ces\\`aro limit, and the associated functional $\\nu: \\mathcal{F}\n\\rightarrow [0,1]$ mapping each such sequence to its Ces\\`aro limit. The basic\nproperties of $\\mathcal{F}$ and $\\nu$ are enumerated, and chains (totally\nordered sets) in $\\mathcal{F}$ on which $\\nu$ is countably additive are studied\nin detail. The main result of the paper concerns a structural property of the\npair $(\\mathcal{F},\\nu)$, specifically that $\\mathcal{F}$ can be factored (in a\ncertain sense) to produce a monotone class on which $\\nu$ is countably\nadditive. In the process, a slight generalisation and clarification of the\nmonotone class theorem for Boolean algebras is proved.\n"}
{"abstract": "  We establish the local existence and uniqueness of solutions to the\nfree-boundary ideal compressible magnetohydrodynamic equations with surface\ntension in three spatial dimensions by a suitable modification of the\nNash--Moser iteration scheme. The main ingredients in proving the convergence\nof the scheme are the tame estimates and unique solvability of the linearized\nproblem in the anisotropic Sobolev spaces $H_*^m$ for $m$ large enough. In\norder to derive the tame estimates, we make full use of the boundary regularity\nenhanced from the surface tension. The unique solution of the linearized\nproblem is constructed by designing some suitable $\\varepsilon$--regularization\nand passing to the limit $\\varepsilon\\to 0$.\n"}
{"abstract": "  In this work, we argue that the observed differences in the value of the\nvector coupling constant extracted from the decays $\\rho\\to \\pi\\pi$ ($g_{\\rho}\n= 6.0$), $\\rho\\to l^+l^-$ ($g_{\\rho} = 5.0 $) and $\\omega\\to l^+ l^-$\n($g_{\\rho} = 5.7$), where $l=e,\\mu $, are an indication of the important role\nplayed by the $1/N_c$ corrections in the description of these processes. We\nshow that an emission of a photon by charged meson loops in the $\\rho^0,\n\\omega,\\phi\\to\\gamma$ transitions is a key process that allows to describe\nabove vector meson decays into two leptons with a single value $g_\\rho = 6.0$.\nOur result supports the idea of universality of neutral vector mesons and\nclarifies the role of accounting of $1/N_c$ corrections to its fulfilment.\n"}
{"abstract": "  Starting from the pseudo-differential decomposition\n$\\mathbf{D}=(-\\Delta)^{\\frac{1}{2}}\\mathcal{H}$ of the Dirac operator\n$\\displaystyle \\mathbf{D}=\\sum_{j=1}^n\\mathbf{e}_j\\partial_{x_j}$ in terms of\nthe fractional operator $(-\\Delta)^{\\frac{1}{2}}$ of order $1$ and of the\nRiesz-Hilbert type operator $\\mathcal{H}$ we will investigate the fundamental\nsolutions of the space-fractional Dirac equation of L\\'evy-Feller type\n$$\\partial_t\\Phi_\\alpha(\\mathbf{x},t;\\theta)=-(-\\Delta)^{\\frac{\\alpha}{2}}\\exp\\left(\n\\frac{i\\pi\\theta}{2} \\mathcal{H}\\right)\\Phi_\\alpha(\\mathbb{x},t;\\theta) $$\ninvolving the fractional Laplacian $-(-\\Delta)^{\\frac{\\alpha}{2}}$ of order\n$\\alpha$, with $2m\\leq \\alpha <2m+2$ ($m\\in \\mathbb{N}$), and the\nexponentiation operator $\\exp\\left( \\frac{i\\pi\\theta}{2} \\mathcal{H}\\right)$ as\nthe hypercomplex counterpart of the fractional Riesz-Hilbert transform carrying\nthe \\textit{skewness parameter} $\\theta$, with values in the range\n$|\\theta|\\leq \\min\\{\\alpha-2m,2m+2-\\alpha\\}$. Such model problem permits us to\nobtain hypercomplex counterparts for the fundamental solutions of higher-order\nheat-type equations $\\partial_t F_M(x,t)=\\kappa_M(\\partial_x)^M F_M(x,t)$\n$(M=2,3,\\ldots)$ in case where the even powers resp. odd powers\n$\\mathbf{D}^{2m}=(-\\Delta)^{m}$ ($M=\\alpha=2m$) resp.\n$\\mathbf{D}^{2m+1}=(-\\Delta)^{m+\\frac{1}{2}}\\mathcal{H}$ ($M=\\alpha=2m+1$) of\n$\\mathbf{D}$ are being considered.\n"}
{"abstract": "  We compute the second (and the first) cohomology groups of $^*$-algebras\nassociated to the universal quantum unitary groups of not neccesarily Kac type,\nextending our earlier results for the free unitary group $U_d^+$. The extended\nsetup forces us to use infinite-dimensional representations to construct the\ncocycles.\n"}
{"abstract": "  In the recent COVID-19 pandemic, a wide range of epidemiological modelling\napproaches have been used to predict the effective reproduction number, R(t),\nand other COVID-19 related measures such as the daily rate of exponential\ngrowth, r(t). These candidate models use different modelling approaches or\ndiffering assumptions about spatial or age mixing, and some capture genuine\nuncertainty in scientific understanding of disease dynamics. Combining\nestimates using appropriate statistical methodology from multiple candidate\nmodels is important to better understand the variation of these outcome\nmeasures to help inform decision making. In this paper, we combine these\nestimates for specific UK nations and regions using random effects meta\nanalyses techniques, utilising the restricted maximum likelihood (REML) method\nto estimate the heterogeneity variance parameter, and two approaches to\ncalculate the confidence interval for the combined estimate: the standard\nWald-type intervals; and the Knapp and Hartung (KNHA) method. As estimates in\nthis setting are derived using model predictions, each with varying degrees of\nuncertainty, equal weighting is favoured over the more standard\ninverse-variance weighting in order avoid potential up-weighting of models\nproviding estimates with lower levels of uncertainty that are not fully\naccounting for inherent uncertainties. Utilising these meta-analysis techniques\nhas allowed for statistically robust combined estimates to be calculated for\nkey COVID-19 outcome measures. This in turn allows timely and informed decision\nmaking based on all of the available information.\n"}
{"abstract": "  A cosmological zoom-in simulation which develops into a Milky Way-like halo\nis started at redshift 7. The initial dark matter distribution is seeded with\ndense star clusters, median mass $5\\times 10^5 M_\\sun$, placed in the largest\nsub-halos present, which have a median peak circular velocity of 25 \\kms. Three\nsimulations are initialized using the same dark matter distribution, with the\nstar clusters started on approximately circular orbits having initial median\nradii 6.8 kpc, 0.14 kpc, and, at the exact center of the sub-halos. The\nsimulations are evolved to the current epoch at which time the median galactic\norbital radii of the three sets of clusters are 30, 5 and 16 kpc, with the\nclusters losing about 2, 50 and 15\\% of their mass, respectively. Clusters\nstarted at small orbital radii have so much tidal forcing that they are often\nnot in equilibrium. Clusters started at larger sub-halo radii have a velocity\ndispersion that declines smoothly to $\\simeq$20\\% of the central value at\n$\\simeq$20 half mass radii. The clusters started at the sub-halo centers can\nshow a rise in velocity dispersion beyond 3-5 half mass radii. That is, the\nclusters formed without local dark matter always have stellar mass dominated\nkinematics at all radii, whereas about 25\\% of the clusters started at sub-halo\ncenters have remnant local dark matter.\n"}
{"abstract": "  Off-policy evaluation (OPE) holds the promise of being able to leverage\nlarge, offline datasets for both evaluating and selecting complex policies for\ndecision making. The ability to learn offline is particularly important in many\nreal-world domains, such as in healthcare, recommender systems, or robotics,\nwhere online data collection is an expensive and potentially dangerous process.\nBeing able to accurately evaluate and select high-performing policies without\nrequiring online interaction could yield significant benefits in safety, time,\nand cost for these applications. While many OPE methods have been proposed in\nrecent years, comparing results between papers is difficult because currently\nthere is a lack of a comprehensive and unified benchmark, and measuring\nalgorithmic progress has been challenging due to the lack of difficult\nevaluation tasks. In order to address this gap, we present a collection of\npolicies that in conjunction with existing offline datasets can be used for\nbenchmarking off-policy evaluation. Our tasks include a range of challenging\nhigh-dimensional continuous control problems, with wide selections of datasets\nand policies for performing policy selection. The goal of our benchmark is to\nprovide a standardized measure of progress that is motivated from a set of\nprinciples designed to challenge and test the limits of existing OPE methods.\nWe perform an evaluation of state-of-the-art algorithms and provide open-source\naccess to our data and code to foster future research in this area.\n"}
{"abstract": "  Recent research has recognized interpretability and robustness as essential\nproperties of trustworthy classification. Curiously, a connection between\nrobustness and interpretability was empirically observed, but the theoretical\nreasoning behind it remained elusive. In this paper, we rigorously investigate\nthis connection. Specifically, we focus on interpretation using decision trees\nand robustness to $l_{\\infty}$-perturbation. Previous works defined the notion\nof $r$-separation as a sufficient condition for robustness. We prove upper and\nlower bounds on the tree size in case the data is $r$-separated. We then show\nthat a tighter bound on the size is possible when the data is linearly\nseparated. We provide the first algorithm with provable guarantees both on\nrobustness, interpretability, and accuracy in the context of decision trees.\nExperiments confirm that our algorithm yields classifiers that are both\ninterpretable and robust and have high accuracy. The code for the experiments\nis available at https://github.com/yangarbiter/interpretable-robust-trees .\n"}
{"abstract": "  In this paper, we propose a novel edge and corner detection algorithm for an\nunorganized point cloud. Our edge detection method classifies a query point as\nan edge point by evaluating the distribution of local neighboring points around\nthe query point. The proposed technique has been tested on generic items such\nas dragons, bunnies, and coffee cups from the Stanford 3D scanning repository.\nThe proposed technique can be directly applied to real and unprocessed point\ncloud data of random clutter of objects. To demonstrate the proposed\ntechnique's efficacy, we compare it to the other solutions for 3D edge\nextractions in an unorganized point cloud data. We observed that the proposed\nmethod could handle the raw and noisy data with little variations in parameters\ncompared to other methods. We also extend the algorithm to estimate the 6D pose\nof known objects in the presence of dense clutter while handling multiple\ninstances of the object. The overall approach is tested for a warehouse\napplication, where an actual UR5 robot manipulator is used for robotic pick and\nplace operations in an autonomous mode.\n"}
{"abstract": "  Robot arm placements are oftentimes a limitation in surgical preoperative\nprocedures, relying on trained staff to evaluate and decide on the optimal\npositions for the arms. Given new and different patient anatomies, it can be\nchallenging to make an informed choice, leading to more frequently colliding\narms or limited manipulator workspaces. In this paper, we develop a method to\ngenerate the optimal manipulator base positions for the multi-port da Vinci\nsurgical system that minimizes self-collision and environment-collision, and\nmaximizes the surgeon's reachability inside the patient. Scoring functions are\ndefined for each criterion so that they may be optimized over. Since for\nmulti-manipulator setups, a large number of free parameters are available to\nadjust the base positioning of each arm, a challenge becomes how one can\nexpediently assess possible setups. We thus also propose methods that perform\nfast queries of each measure with the use of a proxy collision-checker. We then\ndevelop an optimization method to determine the optimal position using the\nscoring functions. We evaluate the optimality of the base positions for the\nrobot arms on canonical trajectories, and show that the solution yielded by the\noptimization program can satisfy each criterion. The metrics and optimization\nstrategy are generalizable to other surgical robotic platforms so that\npatient-side manipulator positioning may be optimized and solved.\n"}
{"abstract": "  Quantification represents the problem of predicting class distributions in a\ngiven target set. It also represents a growing research field in supervised\nmachine learning, for which a large variety of different algorithms has been\nproposed in recent years. However, a comprehensive empirical comparison of\nquantification methods that supports algorithm selection is not available yet.\nIn this work, we close this research gap by conducting a thorough empirical\nperformance comparison of 24 different quantification methods. To consider a\nbroad range of different scenarios for binary as well as multiclass\nquantification settings, we carried out almost 3 million experimental runs on\n40 data sets. We observe that no single algorithm generally outperforms all\ncompetitors, but identify a group of methods including the Median Sweep and the\nDyS framework that perform significantly better in binary settings. For the\nmulticlass setting, we observe that a different, broad group of algorithms\nyields good performance, including the Generalized Probabilistic Adjusted\nCount, the readme method, the energy distance minimization method, the EM\nalgorithm for quantification, and Friedman's method. More generally, we find\nthat the performance on multiclass quantification is inferior to the results\nobtained in the binary setting. Our results can guide practitioners who intend\nto apply quantification algorithms and help researchers to identify\nopportunities for future research.\n"}
{"abstract": "  Industrial processes rely on sensory data for decision-making processes, risk\nassessment, and performance evaluation. Extracting actionable insights from the\ncollected data calls for an infrastructure that can ensure the dissemination of\ntrustworthy data. For the physical data to be trustworthy, it needs to be\ncross-validated through multiple sensor sources with overlapping fields of\nview. Cross-validated data can then be stored on the blockchain, to maintain\nits integrity and trustworthiness. Once trustworthy data is recorded on the\nblockchain, product lifecycle events can be fed into data-driven systems for\nprocess monitoring, diagnostics, and optimized control. In this regard, Digital\nTwins (DTs) can be leveraged to draw intelligent conclusions from data by\nidentifying the faults and recommending precautionary measures ahead of\ncritical events. Empowering DTs with blockchain in industrial use-cases targets\nkey challenges of disparate data repositories, untrustworthy data\ndissemination, and the need for predictive maintenance. In this survey, while\nhighlighting the key benefits of using blockchain-based DTs, we present a\ncomprehensive review of the state-of-the-art research results for\nblockchain-based DTs. Based on the current research trends, we discuss a\ntrustworthy blockchain-based DTs framework. We highlight the role of Artificial\nIntelligence (AI) in blockchain-based DTs. Furthermore, we discuss current and\nfuture research and deployment challenges of blockchain-supported DTs that\nrequire further investigation.\n"}
{"abstract": "  We present an overview of the EmotionGIF2020 Challenge, held at the 8th\nInternational Workshop on Natural Language Processing for Social Media\n(SocialNLP), in conjunction with ACL 2020. The challenge required predicting\naffective reactions to online texts, and included the EmotionGIF dataset, with\ntweets labeled for the reaction categories. The novel dataset included 40K\ntweets with their reaction GIFs. Due to the special circumstances of year 2020,\ntwo rounds of the competition were conducted. A total of 84 teams registered\nfor the task. Of these, 25 teams success-fully submitted entries to the\nevaluation phase in the first round, while 13 teams participated successfully\nin the second round. Of the top participants, five teams presented a technical\nreport and shared their code. The top score of the winning team using the\nRecall@K metric was 62.47%.\n"}
{"abstract": "  In this work we provide a detailed specification of how we extended and\nimplemented Classic Paxos (CP) to execute Read-Modify-Writes. In addition, we\nalso specify how we implemented All-aboard Paxos over CP and how we use\ncarstamps, to also add ABD reads and writes, to accelerate the common case,\nwhere RMWs are not needed. Our specification targets a Key-Value-Store that is\ndeployed within the datacenter, is replicated across 3 to 7 machines and\nsupports reads, writes and RMWs.\n"}
{"abstract": "  We report the first detection of a hydroxyl radical (OH) emission signature\nin the planetary atmosphere outside the solar system, in this case, in the\nday-side of WASP-33b. We analyze high-resolution near-infrared emission spectra\nof WASP-33b taken using the InfraRed Doppler spectrograph on the 8.2-m Subaru\ntelescope. The telluric and stellar lines are removed using a de-trending\nalgorithm, SysRem. The residuals are then cross-correlated with OH and H$_{2}$O\nplanetary spectrum templates produced using several different line-lists. We\ncheck and confirm the accuracy of OH line-lists by cross-correlating with the\nspectrum of GJ 436. As a result, we detect the emission signature of OH at\n$K_\\mathrm{p}$ of 230.9$^{+6.9}_{-7.4}$ km s$^{-1}$ and $v_{\\mathrm{sys}}$ of\n$-$0.3$^{+5.3}_{-5.6}$ km s$^{-1}$ with S/N of 5.4 and significance of\n5.5$\\sigma$. Additionally, we marginally detect H$_{2}$O emission in the H-band\nwith S/N of 4.0 and significance of 5.2$\\sigma$ using the POKAZATEL line-list.\nHowever, no significant signal is detected using the HITEMP 2010, which might\nbe due to differences in line positions and strengths, as well as the\nincompleteness of the line-lists. Nonetheless, this marginal detection is\nconsistent with the prediction that H$_{2}$O is mostly thermally dissociated in\nthe upper atmosphere of the ultra-hot Jupiters. Therefore, along with CO, OH is\nexpected to be one of the most abundant O-bearing molecules in the day-side\natmosphere of ultra-hot Jupiters and should be considered when studying their\natmosphere.\n"}
{"abstract": "  Mathematical models describing the behavior of viscoelastic materials are\noften based on evolution equations that measure the change in stress depending\non its material parameters such as stiffness, viscosity or relaxation time. In\nthis article, we introduce a Maxwell-based rheological model, define the\nassociated forward operator and the inverse problem in order to determine the\nnumber of Maxwell elements and the material parameters of the underlying\nviscoelastic material. We perform a relaxation experiment by applying a strain\nto the material and measure the generated stress. Since the measured data\nvaries with the number of Maxwell elements, the forward operator of the\nunderlying inverse problem depends on parts of the solution. By introducing\nassumptions on the relaxation times, we propose a clustering algorithm to\nresolve this problem. We provide the calculations that are necessary for the\nminimization process and conclude with numerical results by investigating\nunperturbed as well as noisy data. We present different reconstruction\napproaches based on minimizing a least squares functional. Furthermore, we look\nat individual stress components to analyze different displacement rates.\nFinally, we study reconstructions with shortened data sets to obtain assertions\non how long experiments have to be performed to identify conclusive material\nparameters.\n"}
{"abstract": "  Algebraic structures involving both multiplications and comultiplications\n(such as, e.g., bialgebras or Hopf algebras) can be encoded using PROPs\n(categories with PROducts and Permutations) of Adams and MacLane. To encode\nsuch structures on objects of a braided monoidal category, we need PROBs\n(braided analogs of PROPs). Colored PROBs correspond to multi-sorted\nstructures.\n  In particular, we have a colored PROB B governing non-negatively graded\nbialgebras in braided categories. As a category, B splits into blocks B_n\naccording to the grading. We relate B_n with the category P_n of perverse\nsheaves on the n-th symmetric product of the complex line, smooth with respect\nto the natural stratification by multiplicities. More precisely, we show that\nP_n is equivalent to the category of functors from B_n to vector spaces. This\ngives a natural quiver description of P_n.\n"}
{"abstract": "  The flow-based generative model is a deep learning generative model, which\nobtains the ability to generate data by explicitly learning the data\ndistribution. Theoretically its ability to restore data is stronger than other\ngenerative models. However, its implementation has many limitations, including\nlimited model design, too many model parameters and tedious calculation. In\nthis paper, a bi-parallel linear flow model for facial emotion generation from\nemotion set images is constructed, and a series of improvements have been made\nin terms of the expression ability of the model and the convergence speed in\ntraining. The model is mainly composed of several coupling layers superimposed\nto form a multi-scale structure, in which each coupling layer contains 1*1\nreversible convolution and linear operation modules. Furthermore, this paper\nsorted out the current public data set of facial emotion images, made a new\nemotion data, and verified the model through this data set. The experimental\nresults show that, under the traditional convolutional neural network, the\n3-layer 3*3 convolution kernel is more conducive to extracte the features of\nthe face images. The introduction of principal component decomposition can\nimprove the convergence speed of the model.\n"}
{"abstract": "  We present the first observational constraint on the Br-gamma recombination\nline emission associated with the supermassive black hole at the center of our\nGalaxy, known as Sgr A*. By combining 13 years of data with the Adaptive Optics\nfed integral field spectrograph OSIRIS at the W. M. Keck Observatory obtained\nas part of the Galactic Center Orbits Initiative, we extract the near-infrared\nspectrum within ~0.2'' of the black hole and we derive an upper limit on the\nBr-gamma flux. The aperture was set to match the size of the disk-like\nstructure that was recently reported based on millimeter-wave ALMA observations\nof the hydrogen recombination line, H30-alpha. Our stringent upper limit is at\nleast a factor of 80 (and up to a factor of 245) below what would be expected\nfrom the ALMA measurements and strongly constrains possible interpretation of\nemission from this highly under-luminous supermassive black hole.\n"}
{"abstract": "  I searched for the ground state 6.8 and 9.2 GHz hyperfine transitions of\nrubidium and cesium toward M- and L-dwarfs that show Rb and Cs optical\nresonance lines. The optical lines can pump the hyperfine transitions,\npotentially forming masers. These spin-flip transitions of Rb and Cs are the\nprincipal transitions used in atomic clocks (the $^{133}$Cs hyperfine\ntransition defines the second). If they are detected in stellar atmospheres,\nthese transitions would provide exceptionally precise clocks that can be used\nas accelerometers, as exoplanet detectors, as probes of the predictions of\ngeneral relativity, as probes of light propagation effects, and as a means to\ndo fundamental physics with telescopes. Observations of 21 M- and L-dwarfs,\nhowever, show no evidence for Rb or Cs maser action, and a previous survey of\ngiant stars made no Rb maser detections.\n"}
{"abstract": "  A central question in a large class of strongly correlated electron systems,\nincluding heavy fermion compounds and iron pnictides, is the identification of\ndifferent phases and their origins. It has been shown that the\nantiferromagnetic (AFM) phase in some heavy fermion compounds is induced by\nRuderman-Kittel-Kasuya-Yosida (RKKY) interaction between localized moments, and\nthat the competition between this interaction and Kondo effect is responsible\nfor quantum criticality. However, conclusive experimental evidence of the RKKY\ninteraction in pnictides is lacking. Here, using high resolution $^{23}$Na NMR\nmeasurements on lightly Cu-doped metallic single crystals of\nNaFe$_{1-x}$Cu$_{x}$As ~($x \\approx 0.01$) and numerical simulation, we show\ndirect evidence of the RKKY interaction in this pnictide system. Aided by\ncomputer simulation, we identify the $^{23}$Na NMR satellite resonances with\nthe RKKY oscillations of spin polarization at Fe sites. Our $^{23}$Na\nspin-lattice and spin-spin relaxation data exhibits signature of an itinerant\nand inhomogeneous AFM phase in this system, accompanied by a simulation of\nCu-induced perturbation to the ordered moments on the Fe sites. Our NMR results\nindicate coexistence of local and itinerant magnetism in lightly Cu-doped\nNaFe$_{1-x}$Cu$_{x}$As.\n"}
{"abstract": "  Neural-network quantum states (NQS) have been shown to be a suitable\nvariational ansatz to simulate out-of-equilibrium dynamics in two-dimensional\nsystems using time-dependent variational Monte Carlo (t-VMC). In particular,\nstable and accurate time propagation over long time scales has been observed in\nthe square-lattice Heisenberg model using the Restricted Boltzmann machine\narchitecture. However, achieving similar performance in other systems has\nproven to be more challenging. In this article, we focus on the two-leg\nHeisenberg ladder driven out of equilibrium by a pulsed excitation as a\nbenchmark system. We demonstrate that unmitigated noise is strongly amplified\nby the nonlinear equations of motion for the network parameters, which causes\nnumerical instabilities in the time evolution. As a consequence, the achievable\naccuracy of the simulated dynamics is a result of the interplay between network\nexpressiveness and measures required to remedy these instabilities. We show\nthat stability can be greatly improved by appropriate choice of regularization.\nThis is particularly useful as tuning of the regularization typically imposes\nno additional computational cost. Inspired by machine learning practice, we\npropose a validation-set based diagnostic tool to help determining optimal\nregularization hyperparameters for t-VMC based propagation schemes. For our\nbenchmark, we show that stable and accurate time propagation can be achieved in\nregimes of sufficiently regularized variational dynamics.\n"}
{"abstract": "  This paper addresses the problem of sheet-image-based on-line audio-to-score\nalignment also known as score following. Drawing inspiration from object\ndetection, a conditional neural network architecture is proposed that directly\npredicts x,y coordinates of the matching positions in a complete score sheet\nimage at each point in time for a given musical performance. Experiments are\nconducted on a synthetic polyphonic piano benchmark dataset and the new method\nis compared to several existing approaches from the literature for\nsheet-image-based score following as well as an Optical Music Recognition\nbaseline. The proposed approach achieves new state-of-the-art results and\nfurthermore significantly improves the alignment performance on a set of\nreal-world piano recordings by applying Impulse Responses as a data\naugmentation technique.\n"}
{"abstract": "  Point-of-Interest recommendation is an increasing research and developing\narea within the widely adopted technologies known as Recommender Systems. Among\nthem, those that exploit information coming from Location-Based Social Networks\n(LBSNs) are very popular nowadays and could work with different information\nsources, which pose several challenges and research questions to the community\nas a whole. We present a systematic review focused on the research done in the\nlast 10 years about this topic. We discuss and categorize the algorithms and\nevaluation methodologies used in these works and point out the opportunities\nand challenges that remain open in the field. More specifically, we report the\nleading recommendation techniques and information sources that have been\nexploited more often (such as the geographical signal and deep learning\napproaches) while we also alert about the lack of reproducibility in the field\nthat may hinder real performance improvements.\n"}
{"abstract": "  It is shown that a singular equivalence induced by tensoring with a suitable\ncomplex of bimodules defines a singular equivalence of Morita type with level,\nin the sense of Wang. This result is applied to homological ideals and\nidempotents to produce new reduction techniques for testing the properties of\nsyzygy-finite and injectives generation of finite dimensional algebras over a\nfield.\n"}
{"abstract": "  In the past decades, massive efforts involving companies, non-profit\norganizations, governments, and others have been put into supporting the\nconcept of data democratization, promoting initiatives to educate people to\nconfront information with data. Although this represents one of the most\ncritical advances in our free world, access to data without concrete facts to\ncheck or the lack of an expert to help on understanding the existing patterns\nhampers its intrinsic value and lessens its democratization. So the benefits of\ngiving full access to data will only be impactful if we go a step further and\nsupport the Data Analytics Democratization, assisting users in transforming\nfindings into insights without the need of domain experts to promote\nunconstrained access to data interpretation and verification. In this paper, we\npresent Explainable Patterns (ExPatt), a new framework to support lay users in\nexploring and creating data storytellings, automatically generating plausible\nexplanations for observed or selected findings using an external (textual)\nsource of information, avoiding or reducing the need for domain experts. ExPatt\napplicability is confirmed via different use-cases involving world demographics\nindicators and Wikipedia as an external source of explanations, showing how it\ncan be used in practice towards the data analytics democratization.\n"}
{"abstract": "  This article presents a novel computational model to study the selective\nfiltering of biological hydrogels due to the surface charge and size of\ndiffusing particles. It is the first model that includes the random 3D fiber\norientation and connectivity of the biopolymer network and that accounts for\nelastic deformations of the fibers by means of beam theory. As a key component\nof the model, novel formulations are proposed both for the electrostatic and\nrepulsive steric interactions between a spherical particle and a beam. In\naddition to providing a thorough validation of the model, the presented\ncomputational studies yield new insights into the underlying mechanisms of\nhindered particle mobility, especially regarding the influence of the\naforementioned aspects that are unique to this model. It is found that the\nprecise distribution of fiber and thus charge agglomerations in the network\nhave a crucial influence on the mobility of oppositely charged particles and\ngives rise to distinct motion patterns. Considering the high practical\nsignificance for instance with respect to targeted drug release or infection\ndefense, the provided proof of concept motivates further advances of the model\ntoward a truly predictive computational tool that allows a case- and\npatient-specific assessment for real (biological) systems.\n"}
{"abstract": "  To perform manipulation tasks in the real world, robots need to operate on\nobjects with various shapes, sizes and without access to geometric models. It\nis often unfeasible to train monolithic neural network policies across such\nlarge variance in object properties. Towards this generalization challenge, we\npropose to learn modular task policies which compose object-centric task-axes\ncontrollers. These task-axes controllers are parameterized by properties\nassociated with underlying objects in the scene. We infer these controller\nparameters directly from visual input using multi-view dense correspondence\nlearning. Our overall approach provides a simple, modular and yet powerful\nframework for learning manipulation tasks. We empirically evaluate our approach\non multiple different manipulation tasks and show its ability to generalize to\nlarge variance in object size, shape and geometry.\n"}
{"abstract": "  We present new results of the backscattering dominated prompt emission model\nin which the photons generated through pair annihilation at the centre of a\ngamma ray burst (GRB) are backscattered through Compton scattering by an\noutflowing stellar cork. %We show that the produced spectral features are\ncapable of producing spectral features that are widely observed in GRBs. Using\nComptonized pair annihilation spectrum accompanied by bremsstrahlung radiation\nfor seed photons, we show that the obtained spectra produce low energy photon\nindex in the range $\\alpha \\sim -1.95$ to $-1.1$, steeper high energy slopes\n$\\beta \\sim -3.5$ to $-2.4$ and spectral peak energies $\\sim$ few KeV to few\n$10\\times$ MeV. These findings are consistent with the values covered in GRB\nprompt phase observations.\n"}
{"abstract": "  We study the confinement/deconfinement phase transition of the radion field\nin a warped model with a polynomial bulk potential. The backreaction of the\nradion on the metric is taken into account by using the superpotential\nformalism, while the radion effective potential is obtained from a novel\nformulation which can incorporate the backreaction. The phase transition leads\nto a stochastic gravitational wave background that depends on the energy scale\nof the first Kaluza-Klein resonance, $m_{\\textrm{KK}}$. This work completes\nprevious studies in the following aspects: i) we detail the evaluation of the\nradion spectrum; ii) we report on the mismatches between the thick wall\napproximation and the numerical bounce solution; iii) we include a suppression\nfactor in the spectrum of sound waves accounting for their finite lifetime;\nand, iv) we update the bound on $m_{\\textrm{KK}}$ in view of the O3 LIGO and\nVirgo data. We find that the forthcoming gravitational wave interferometers can\nprobe scenarios where $m_{\\textrm{KK}} \\lesssim 10^9$ TeV, while the O3-run\nbounds rule out warped models with $10^4 \\textrm{TeV} \\lesssim m_{\\textrm{KK}}\n\\lesssim 10^7$ TeV exhibiting an extremely strong confinement/deconfinement\nphase transition.\n"}
{"abstract": "  When obtaining interior 3D voxel data from triangular meshes, most existing\nmethods fail to handle low quality meshes which happens to take up a big\nportion on the internet. In this work we present a robust voxelization method\nthat is based on tetrahedral mesh generation within a user defined error bound.\nComparing to other tetrahedral mesh generation methods, our method produces\nmuch higher quality tetrahedral meshes as the intermediate outcome, which\nallows us to utilize a faster voxelization algorithm that is based on a\nstronger assumption. We show the results comparing to various methods including\nthe state-of-the-art. Our contribution includes a framework which takes\ntriangular mesh as an input and produces voxelized data, a proof to an unproved\nalgorithm that performs better than the state-of-the-art, and various\nexperiments including parallelization built on the GPU and CPU. We further\ntested our method on various dataset including Princeton ModelNet and Thingi10k\nto show the robustness of the framework, where near 100% availability is\nachieved, while others can only achieve around 50%.\n"}
{"abstract": "  The recently proposed RCube network is a cube-based server-centric data\ncenter network (DCN), including two types of heterogeneous servers, called core\nand edge servers. Remarkably, it takes the latter as backup servers to deal\nwith server failures and thus achieve high availability. This paper first\npoints out that RCube is suitable as a candidate topology of DCNs for edge\ncomputing. Three transmission types are among core and edge servers based on\nthe demand for applications' computation and instant response. We then employ\nprotection routing to analyze the transmission failure of RCube DCNs. Unlike\ntraditional protection routing, which only tolerates a single link or node\nfailure, we use the multi-protection routing scheme to improve fault-tolerance\ncapability. To configure a protection routing in a network, according to\nTapolcai's suggestion, we need to construct two completely independent spanning\ntrees (CISTs). A logic graph of RCube, denoted by $L$-$RCube(n,m,k)$, is a\nnetwork with a recursive structure. Each basic building element consists of $n$\ncore servers and $m$ edge servers, where the order $k$ is the number of\nrecursions applied in the structure. In this paper, we provide algorithms to\nconstruct $\\min\\{n,\\lfloor(n+m)/2\\rfloor\\}$ CISTs in $L$-$RCube(n,m,k)$ for\n$n+m\\geqslant 4$ and $n>1$. From a combination of the multiple CISTs, we can\nconfigure the desired multi-protection routing. In our simulation, we configure\nup to 10 protection routings for RCube DCNs. As far as we know, in past\nresearch, there were at most three protection routings developed in other\nnetwork structures. Finally, we summarize some crucial analysis viewpoints\nabout the transmission efficiency of DCNs with heterogeneous edge-core servers\nfrom the simulation results.\n"}
{"abstract": "  We provide a microscopic-level derivation of earlier results showing that, in\nthe critical vicinity of the superconductor-to-insulator transition (SIT),\ndisorder and localization become negligible and the structure of the emergent\nphases is determined by topological effects arising from the competition\nbetween two quantum orders, superconductivity and superinsulation. We find\nthat, around the critical point, the ground state is a composite incompressible\nquantum fluid of Cooper pairs and vortices coexisting with an intertwined\nWigner crystal for the excess (with respect to integer filling) excitations of\nthe two types.\n"}
{"abstract": "  We study the zero-field optical spin noise spectra (OSN) for a thermal state\nhot 87Rb vapor with strong spin exchange coupling. Our main findings are: (1)\nThe OSN spectrum consists of two components representing a positive and a\nnegative hyperfine spin correlation (HSC), the relative power of which varies\ndramatically with the detuning frequency of the probe. (2) There exist two\npolar frequencies at which the OSN spectrum is completely polarized with one\ntype of HSC. (3) At the limit of far detuning, the power ratio of the positive\nand negative HSC component of the OSN is 4:5. (4) The total power of the OSN is\nindependent of the strength of SE coupling. (5) We give a simple way of\nderiving the OSN using the eigensolution of the density matrix equation.\n"}
{"abstract": "  Planning for legged-wheeled machines is typically done using trajectory\noptimization because of many degrees of freedom, thus rendering legged-wheeled\nplanners prone to falling prey to bad local minima. We present a combined\nsampling and optimization-based planning approach that can cope with\nchallenging terrain. The sampling-based stage computes whole-body\nconfigurations and contact schedule, which speeds up the optimization\nconvergence. The optimization-based stage ensures that all the system\nconstraints, such as non-holonomic rolling constraints, are satisfied. The\nevaluations show the importance of good initial guesses for optimization.\nFurthermore, they suggest that terrain/collision (avoidance) constraints are\nmore challenging than the robot model's constraints. Lastly, we extend the\noptimization to handle general terrain representations in the form of elevation\nmaps.\n"}
{"abstract": "  Nonlinear interactions are recognized as potential resources for quantum\nmetrology, facilitating parameter estimation precisions that scale as the\nexponential Heisenberg limit of $2^{-N}$. We explore such nonlinearity and\npropose an associated quantum measurement scenario based on the nonlinear\ninteraction of $N$-probe entanglement generating form. This scenario provides\nan enhanced precision scaling of $D^{-N}/(N-1)!$ with $D > 2$ a tunable\nparameter. In addition, it can be readily implemented in a variety of\nexperimental platforms and applied to measurements of a wide range of\nquantities, including local gravitational acceleration $g$, magnetic field, and\nits higher-order gradients.\n"}
{"abstract": "  Ionic gating is a very popular tool to investigate and control the electric\ncharge transport and electronic ground state in a wide variety of different\nmaterials. This is due to its capability to induce large modulations of the\nsurface charge density by means of the electric-double-layer field-effect\ntransistor (EDL-FET) architecture, and has been proven to be capable of tuning\neven the properties of metallic systems. In this short review, I summarize the\nmain results which have been achieved so far in controlling the superconducting\n(SC) properties of thin films of conventional metallic superconductors by means\nof the ionic gating technique. I discuss how the gate-induced charge doping,\ndespite being confined to a thin surface layer by electrostatic screening,\nresults in a long-range \"bulk\" modulation of the SC properties by the coherent\nnature of the SC condensate, as evidenced by the observation of suppressions in\nthe critical temperature of films much thicker than the electrostatic screening\nlength, and by the pronounced thickness-dependence of their magnitude. I review\nhow this behavior can be modelled in terms of proximity effect between the\ncharge-doped surface layer and the unperturbed bulk with different degrees of\napproximation, and how first-principles calculations have been employed to\ndetermine the origin of an anomalous increase in the electrostatic screening\nlength at ultrahigh electric fields, thus fully confirming the validity of the\nproximity effect model. Finally, I discuss a general framework - based on the\ncombination of ab-initio Density Functional Theory and the Migdal-Eliashberg\ntheory of superconductivity - by which the properties of any gated thin film of\na conventional metallic superconductor can be determined purely from first\nprinciples.\n"}
{"abstract": "  Heterogeneity in medical data, e.g., from data collected at different sites\nand with different protocols in a clinical study, is a fundamental hurdle for\naccurate prediction using machine learning models, as such models often fail to\ngeneralize well. This paper leverages a recently proposed\nnormalizing-flow-based method to perform counterfactual inference upon a\nstructural causal model (SCM), in order to achieve harmonization of such data.\nA causal model is used to model observed effects (brain magnetic resonance\nimaging data) that result from known confounders (site, gender and age) and\nexogenous noise variables. Our formulation exploits the bijection induced by\nflow for the purpose of harmonization. We infer the posterior of exogenous\nvariables, intervene on observations, and draw samples from the resultant SCM\nto obtain counterfactuals. This approach is evaluated extensively on multiple,\nlarge, real-world medical datasets and displayed better cross-domain\ngeneralization compared to state-of-the-art algorithms. Further experiments\nthat evaluate the quality of confounder-independent data generated by our model\nusing regression and classification tasks are provided.\n"}
{"abstract": "  The low-energy excitations of graphene are relativistic massless Dirac\nfermions with opposite chiralities at valleys K and K'. Breaking the chiral\nsymmetry could lead to gap opening in analogy to dynamical mass generation in\nparticle physics. Here we report direct experimental evidences of chiral\nsymmetry breaking (CSB) from both microscopic and spectroscopic measurements in\na Li-intercalated graphene. The CSB is evidenced by gap opening at the Dirac\npoint, Kekul\\'e-O type modulation, and chirality mixing near the gap edge. Our\nwork opens up opportunities for investigating CSB related physics in a\nKekul\\'e-ordered graphene.\n"}
{"abstract": "  Artificial Neural networks are mathematical models at their core. This\ntruismpresents some fundamental difficulty when networks are tasked with\nNatural Language Processing. A key problem lies in measuring the similarity or\ndistance among vectors in NLP embedding space, since the mathematical concept\nof distance does not always agree with the linguistic concept. We suggest that\nthe best way to measure linguistic distance among vectors is by employing the\nLanguage Model (LM) that created them. We introduce Language Model Distance\n(LMD) for measuring accuracy of vector transformations based on the\nDistributional Hypothesis ( LMD Accuracy ). We show the efficacy of this metric\nby applying it to a simple neural network learning the Procrustes algorithm for\nbilingual word mapping.\n"}
{"abstract": "  We present a straightforward and efficient way to control unstable robotic\nsystems using an estimated dynamics model. Specifically, we show how to exploit\nthe differentiability of Gaussian Processes to create a state-dependent\nlinearized approximation of the true continuous dynamics that can be integrated\nwith model predictive control. Our approach is compatible with most Gaussian\nprocess approaches for system identification, and can learn an accurate model\nusing modest amounts of training data. We validate our approach by learning the\ndynamics of an unstable system such as a segway with a 7-D state space and 2-D\ninput space (using only one minute of data), and we show that the resulting\ncontroller is robust to unmodelled dynamics and disturbances, while\nstate-of-the-art control methods based on nominal models can fail under small\nperturbations. Code is open sourced at\nhttps://github.com/learning-and-control/core .\n"}
{"abstract": "  We consider weighted tree automata over strong bimonoids (for short: wta). A\nwta $\\mathcal{A}$ has the finite-image property if its recognized weighted tree\nlanguage $[\\![\\mathcal{A}]\\!]$ has finite image; moreover, $\\mathcal{A}$ has\nthe preimage property if the preimage under $[\\![\\mathcal{A}]\\!]$ of each\nelement of the underlying strong bimonoid is a recognizable tree language. For\neach wta $\\mathcal{A}$ over a past-finite monotonic strong bimonoid we prove\nthe following results. In terms of $\\mathcal{A}$'s structural properties, we\ncharacterize whether it has the finite-image property. We characterize those\npast-finite monotonic strong bimonoids such that for each wta $\\mathcal{A}$ it\nis decidable whether $\\mathcal{A}$ has the finite-image property. In\nparticular, the finite-image property is decidable for wta over past-finite\nmonotonic semirings. Moreover, we prove that $\\mathcal{A}$ has the preimage\nproperty. All our results also hold for weighted string automata.\n"}
{"abstract": "  We give a proof the monodromy conjecture relating the poles of motivic zeta\nfunctions with roots of b-functions for isolated quasihomogeneous\nhypersurfaces, and more generally for semi-quasihomogeneous hypersurfaces. We\nalso give a strange generalization allowing a twist by certain differential\nforms.\n"}
{"abstract": "  Recently, generalizations of the classical Three Gap Theorem to higher\ndimensions attracted a lot of attention. In particular, upper bounds for the\nnumber of nearest neighbor distances have been established for the Euclidean\nand the maximum metric. It was proved that a generic multi-dimensional\nKronecker attains the maximal possible number of different gap lengths for\nevery sub-exponential subsequence. We mirror this result in dimension $d \\in\n\\left\\{ 2, 3 \\right\\}$ by constructing Kronecker sequences which have a\nsurprisingly low number of different nearest neighbor distances for infinitely\n$N \\in \\mathbb{N}$. Our proof relies on simple arguments from the theory of\ncontinued fractions.\n"}
{"abstract": "  We study radiation pressure due to Lyman alpha line photons, obtaining and\nexploring analytical expressions for the force-multiplier, $M_F(N_H, Z) =\nF_\\alpha/(L_\\alpha/c)$, as a function of gas column density, $N_H$, and\nmetallicity, $Z$, for both dust-free and dusty media, employing a WKB approach\nfor the latter case. Solutions for frequency offset emission to emulate\nnon-static media moving with a bulk velocity $v$, have also been obtained. We\nfind that, in static media, Ly$\\alpha$ pressure dominates over both\nphotoionization and dust-mediated UV radiation pressure in a very wide\nparameter range ($16 < \\log N_H < 23$; $-4 < \\log[Z/Z_\\odot] < 0$). For\nexample, it overwhelms the other two forces by 10 (300) times in standard\n(low-$Z$) star-forming clouds. Thus, in agreement with previous studies, we\nconclude that Ly$\\alpha$ pressure plays a dominant role in the initial\nacceleration of the gas around luminous sources, and must be implemented in\ngalaxy formation, evolution and outflow models and simulations.\n"}
{"abstract": "  Super-Earths and sub-Neptunes are commonly thought to have accreted\nhydrogen/helium envelopes, consisting of a few to ten percent of their total\nmass, from the primordial gas disk. Subsequently, hydrodynamic escape driven by\ncore-powered mass-loss and/or photo-evaporation likely stripped much of these\nprimordial envelopes from the lower-mass and closer-in planets to form the\nsuper-Earth population. In this work we show that after undergoing core-powered\nmass-loss, some super-Earths can retain small residual H/He envelopes. This\nretention is possible because, for significantly depleted atmospheres, the\ndensity at the radiative-convective boundary drops sufficiently such that thhe\ncooling time-scale becomes less than the mass-loss time-scale. The residual\nenvelope is therefore able to contract, terminating further mass loss. Using\nanalytic calculations and numerical simulations, we show that the mass of\nprimordial H/He envelope retained as a fraction of the planet's total mass,\n$f_{ret}$, increases with increasing planet mass, $M_{c}$, and decreases with\nincreasing equilibrium temperature, $T_{eq}$, scaling as $f_{ret} \\propto\nM_{c}^{3/2} T_{eq}^{-1/2} \\exp{[M_{c}^{3/4} T_{eq}^{-1}]}$. $f_{ret}$ varies\nfrom $< 10^{-8}$ to about $10^{-3}$ for typical super-Earth parameters. To\nfirst order, the exact amount of left-over H/He depends on the initial envelope\nmass, the planet mass, its equilibrium temperature, and the envelope's opacity.\nThese residual hydrogen envelopes reduce the atmosphere's mean molecular weight\ncompared to a purely secondary atmosphere, a signature observable by current\nand future facilities. These remnant atmospheres may, however, in many cases be\nvulnerable to long-term erosion by photo-evaporation. Any residual hydrogen\nenvelope likely plays an important role in the long-term physical evolution of\nsuper-Earths, including their geology and geochemistry.\n"}
{"abstract": "  Reaction-diffusion equations are one of the most common mathematical models\nin the natural sciences and are used to model systems that combine reactions\nwith diffusive motion. However, rather than normal diffusion, anomalous\nsubdiffusion is observed in many systems and is especially prevalent in cell\nbiology. What are the reaction-subdiffusion equations describing a system that\ninvolves first-order reactions and subdiffusive motion? In this paper, we\nanswer this question. We derive fractional reaction-subdiffusion equations\ndescribing an arbitrary number of molecular species which react at first-order\nrates and move subdiffusively with general space-dependent diffusivities and\ndrifts. Importantly, different species may have different diffusivities and\ndrifts, which contrasts previous approaches to this question which assume that\neach species has the same movement dynamics. We derive the equations by\ncombining results on time-dependent fractional Fokker-Planck equations with\nmethods of analyzing stochastically switching evolution equations. Furthermore,\nwe construct the stochastic description of individual molecules whose\ndeterministic concentrations follow these reaction-subdiffusion equations. This\nstochastic description involves subordinating a diffusion process whose\ndynamics are controlled by a subordinated Markov jump process. We illustrate\nour results in several examples and show that solutions of the\nreaction-subdiffusion equations agree with stochastic simulations of individual\nmolecules.\n"}
{"abstract": "  Zariski decomposition is a fundamental tool for studying linear systems of\ndivisors on algebraic surfaces. Bauer, K\\\"uronya and Szemberg obtained a\ndecomposition of the big cone of a smooth projective surface into chambers,\ncalled Zariski chambers, in each of which the support of the negative part of\nthe Zariski decomposition is constant. On the other hand, thanks to the work of\nBoucksom, on a compact complex manifold we have the divisorial Zariski\ndecomposition which, on an irreducible holomorphic symplectic manifold (IHS),\nis a higher dimensional Zariski decomposition with respect to the\nBeauville-Bogomolov-Fujiki quadratic form. Similarly to the case of surfaces,\nwe provide for the big cone of a projective irreducible holomorphic symplectic\nmanifold a decomposition into chambers (which we describe in detail), in each\nof which the support of the negative part of the divisorial Zariski\ndecomposition is constant. As for surfaces, the obtained decomposition of the\nbig cone allows to describe the volume function, which turns out to behave very\nwell. Moreover, always inspired by the surface case, we also see that the big\ncone of a projective IHS manifold admits a decomposition into simple Weyl\nchambers, which we compare to that induced by the divisorial Zariski\ndecomposition. To conclude, we investigate the boundary of the big cone and\ndetermine the structure of the algebraic pseudo-effective cone.\n"}
{"abstract": "  Quantum shell effects are known to affect the formation of fragments in\nnuclear fission. Shell effects also affect quasi-fission reactions occurring in\nheavy-ion collisions. Systematic time-dependent Hartree-Fock simulations of\n50Ca+176Yb collisions show that the mass equilibration between the fragments in\nquasi-fission is stopped when they reach similar properties to those in the\nasymmetric fission mode of the 226Th compound nucleus. Similar shell effects\nare then expected to determine the final repartition of nucleons between the\nnascent fragments in both mechanisms. Future experimental studies that could\ntest these observations are discussed.\n"}
{"abstract": "  Solar grade silicon (SoG-Si) is a key material for the development of\ncrystalline silicon photovoltaics (PV), which is expected to reach the\ntera-watt level in the next years and around 50TW in 2050. Upgraded\nmetallurgical grade silicon (UMG-Si) has already demonstrated to be a viable\nalternative to standard polysilicon in terms of cost and quality. This study\npresents the life cycle assessment (LCA) of UMG obtained by the FerroSolar\nprocess. Moreover, it shows the environmental impacts of PV modules and\nelectricity generation based on this material. For this, an exhaustive review\nof the life cycle inventory (LCI) of PV value chain, from metallurgical grade\nsilicon (MG-Si) down to electricity generation, has been carried out updating\ninputs for all processes. The Balance of System (BoS) has also been updated\nwith real state of the art data for a fixed open ground large PV site (100 MW).\nTwo different electricity mixes, with low and high carbon intensities, have\nbeen considered. The results reveal that for PV electricity generation using\nUMG instead of polysilicon leads to an overall reduction of Climate change (CC)\nemissions of over 20%, along with an improvement of the Energy Payback Time\n(EPBT) of 25%, achieving significantly low values, 12 gCO2eq / kWhe and 0.52\nyears, respectively. Moreover, it is shown that UMG silicon feedstock is not\nthe main contributor to the carbon and energy footprint of the produced\nelectricity, leaving the first place to PV module manufacturing.\n"}
{"abstract": "  Common domain adaptation techniques assume that the source domain and the\ntarget domain share an identical label space, which is problematic since when\ntarget samples are unlabeled we have no knowledge on whether the two domains\nshare the same label space. When this is not the case, the existing methods\nfail to perform well because the additional unknown classes are also matched\nwith the source domain during adaptation. In this paper, we tackle the open set\ndomain adaptation problem under the assumption that the source and the target\nlabel spaces only partially overlap, and the task becomes when the unknown\nclasses exist, how to detect the target unknown classes and avoid aligning them\nwith the source domain. We propose to utilize an instance-level reweighting\nstrategy for domain adaptation where the weights indicate the likelihood of a\nsample belonging to known classes and to model the tail of the entropy\ndistribution with Extreme Value Theory for unknown class detection. Experiments\non conventional domain adaptation datasets show that the proposed method\noutperforms the state-of-the-art models.\n"}
{"abstract": "  Animal interval timing is often studied through the peak interval (PI)\nprocedure. In this procedure, the animal is rewarded for the first response\nafter a fixed delay from the stimulus onset, but on some trials, the stimulus\nremains and no reward is given. The common methods and models to analyse the\nresponse pattern describe it as break-run-break, a period of low rate response\nfollowed by rapid responding, followed by a low rate of response. The study of\nthe pattern has found correlations between start, stop, and duration of the run\nperiod that hold across species and experiment.\n  It is commonly assumed that in order to achieve the statistics with a\npacemaker accumulator model it is necessary to have start and stop thresholds.\nIn this paper we will develop a new model that varies response rate in relation\nto the likelihood of event occurrence, as opposed to a threshold, for changing\nthe response rate. The new model reproduced the start and stop statistics that\nhave been observed in 14 different PI experiments from 3 different papers. The\ndeveloped model is also compared to the Time-adaptive Drift-diffusion Model\n(TDDM), the latest accumulator model subsuming the scalar expectancy theory\n(SET), on all 14 data-sets. The results show that it is unnecessary to have\nexplicit start and stop thresholds or an internal equivalent to break-run-break\nstates to reproduce the individual trials statistics and population behaviour\nand get the same break-run-break analysis results. The new model also produces\nmore realistic individual trials compared to TDDM.\n"}
{"abstract": "  Background: Spin-triplet ($S=1$) proton-neutron (pn) pairing in nuclei has\nbeen under debate. It is well known that the dynamical pairing affects the\nnuclear matrix element of the Gamow-Teller (GT) transition and the double beta\ndecay. Purpose: We investigate the effect of the pn-pair interaction in the\n$T=0, S=1$ channel on the low-lying spin-dipole (SD) transition. We then aim at\nclarifying the distinction of the role in between the SD and GT transitions.\nMethod: We perform a three-body model calculation for the transition\n${}^{80}\\mathrm{Ni}\\to{}^{80}\\mathrm{Cu}$, where ${}^{78}\\mathrm{Ni}$ is taken\nas a core. The strength of the pair interaction is varied to see the effect on\nthe SD transition-strength distribution. To fortify the finding obtained by the\nthree-body model, we employ the nuclear energy-density functional method for\nthe SD transitions in several nuclei, where one can expect a strong effect.\nResults: The effect of the $S=1$ pn-pair interaction depends on the spatial\noverlap of the pn pair and the angular momentum of the valence nucleons; the\nhigher the angular momentum of the orbitals, the more significant the effect.\nConclusions: The dynamical $S=1$ pairing is effective even for SD states\nalthough the spatial overlap of the pn pair can be smaller than GT states. The\nSD transition involving high-$\\ell$ orbitals with the same principal quantum\nnumber is strongly affected by the dynamical $S=1$ pairing.\n"}
{"abstract": "  Graph Neural Networks (GNNs) are a new and increasingly popular family of\ndeep neural network architectures to perform learning on graphs. Training them\nefficiently is challenging due to the irregular nature of graph data. The\nproblem becomes even more challenging when scaling to large graphs that exceed\nthe capacity of single devices. Standard approaches to distributed DNN\ntraining, such as data and model parallelism, do not directly apply to GNNs.\nInstead, two different approaches have emerged in the literature: whole-graph\nand sample-based training.\n  In this paper, we review and compare the two approaches. Scalability is\nchallenging with both approaches, but we make a case that research should focus\non sample-based training since it is a more promising approach. Finally, we\nreview recent systems supporting sample-based training.\n"}
{"abstract": "  Many potential applications of reinforcement learning (RL) require guarantees\nthat the agent will perform well in the face of disturbances to the dynamics or\nreward function. In this paper, we prove theoretically that standard maximum\nentropy RL is robust to some disturbances in the dynamics and the reward\nfunction. While this capability of MaxEnt RL has been observed empirically in\nprior work, to the best of our knowledge our work provides the first rigorous\nproof and theoretical characterization of the MaxEnt RL robust set. While a\nnumber of prior robust RL algorithms have been designed to handle similar\ndisturbances to the reward function or dynamics, these methods typically\nrequire adding additional moving parts and hyperparameters on top of a base RL\nalgorithm. In contrast, our theoretical results suggest that MaxEnt RL by\nitself is robust to certain disturbances, without requiring any additional\nmodifications. While this does not imply that MaxEnt RL is the best available\nrobust RL method, MaxEnt RL does possess a striking simplicity and appealing\nformal guarantees.\n"}
{"abstract": "  The Internet of Value (IOV) with its distributed ledger technology (DLT)\nunderpinning has created new forms of lending markets. As an integral part of\nthe decentralised finance (DeFi) ecosystem, lending protocols are gaining\ntremendous traction, holding an aggregate liquidity supply of over $40 billion\nat the time of writing. In this paper, we enumerate the challenges of\ntraditional money markets led by banks and lending platforms, and present\nadvantageous characteristics of DeFi lending protocols that might help resolve\ndeep-rooted issues in the conventional lending environment. With the examples\nof Maker, Compound and Aave, we describe in detail the mechanism of DeFi\nlending protocols. We discuss the persisting reliance of DeFi lending on the\ntraditional financial system, and conclude with the outlook of the lending\nmarket in the IOV era.\n"}
{"abstract": "  Highly nonclassical character of optical quantum detectors, such as\nsingle-photon detectors, is essential for preparation of quantum states of\nlight and a vast majority of applications in quantum metrology and quantum\ninformation processing. Therefore, it is both fundamentally interesting and\npractically relevant to investigate the nonclassical features of optical\nquantum measurements. Here we propose and experimentally demonstrate a\nprocedure for direct certification of quantum non-Gaussianity and Wigner\nfunction negativity, two crucial nonclassicality levels, of photonic quantum\ndetectors. Remarkably, we characterize the highly nonclassical properties of\nthe detector by probing it with only two classical thermal states and a vacuum\nstate. We experimentally demonstrate the quantum non-Gaussianity of\nsingle-photon avalanche diode even under the presence of background noise, and\nwe also certify the negativity of the Wigner function of this detector. Our\nresults open the way for direct benchmarking of photonic quantum detectors with\na few measurements on classical states.\n"}
{"abstract": "  Massive single-cell profiling efforts have accelerated our discovery of the\ncellular composition of the human body, while at the same time raising the need\nto formalise this new knowledge. Here, we review current cell ontology efforts\nto harmonise and integrate different sources of annotations of cell types and\nstates. We illustrate with examples how a unified ontology can consolidate and\nadvance our understanding of cell types across scientific communities and\nbiological domains.\n"}
{"abstract": "  A model of unified dark matter and dark energy based on a Dynamical Spacetime\nTheory (DST) is studied. By introducing a Dynamical Spacetime vector field\n$\\chi_\\mu$, a conservation of an energy momentum tensor $T^{\\mu\\nu}_{(\\chi)}$\nemerges. The action allows for two different potentials, while one represents a\ndark energy. For constant potentials, the cosmological solution yields a non\nsingular bouncing solutions that rapidly approaches the $\\Lambda$CDM model. The\nDynamical Time corresponds to the cosmic time as well. The theory fits with the\nlate time expansion data of the Universe. With higher dimensions a mechanism\nfor inflation and compactification appears, with exponential growth for some\ndimensions and exponential contraction of the others. By demanding that the\nDynamical Spacetime vector field be a gradient of a scalar the DST becomes a\ntheory with diffusive interacting, which asymptotically returns to the\n$\\Lambda$CDM model as a stable point. These formulations lead to scenarios\nwhich address our understanding about the origin of the Universe.\n"}
{"abstract": "  With the COVID-19 pandemic entering a second phase and vaccination strategies\nbeing applied by countries and governments worldwide, there is an increasing\nexpectation by people to return to normal life. There is currently a debate\nabout immunity passports, privacy, and the enablement of individuals to safely\nenter everyday social life, workplace, and travel. Such digital immunity\npassports and vaccination certificates should meet people's expectations for\nprivacy while enabling them to present to 3rd party verifiers tamper-evident\ncredentials. This paper provides a comprehensive answer to the technological,\nethical and security challenges, by proposing an architecture that provides to\nindividuals, employers, and government agencies, a digital, decentralized,\nportable, immutable, and non-refutable health status cryptographic proof. It\ncan be used to evaluate the risk of allowing individuals to return to work,\ntravel, and public life activities.\n"}
{"abstract": "  In this paper, we show that for sufficiently strong atomic interactions,\nthere exist analytical solutions of current-carrying nonlinear Bloch states at\nthe Brillouin zone edge to the model of spin-orbit-coupled Bose-Einstein\ncondensates (BECs) with symmetric spin interaction loaded into optical\nlattices. These simple but generic exact solutions provide an analytical\ndemonstration of some intriguing properties which have neither an analog in the\nregular BEC lattice systems nor in the uniform spin-orbit-coupled BEC systems.\nIt is an analytical example for understanding the superfluid and other related\nproperties of the spin-orbit-coupled BEC lattice systems.\n"}
{"abstract": "  Insertion-deletion codes (insdel codes for short) play an important role in\nsynchronization error correction. The higher the minimum insdel distance, the\nmore insdel errors the code can correct. Haeupler and Shahrasbi established the\nSingleton bound for insdel codes: the minimum insdel distance of any $[n,k]$\nlinear code over $\\mathbb{F}_q$ satisfies $d\\leq2n-2k+2.$ There have been some\nconstructions of insdel codes through Reed-Solomon codes with high\ncapabilities, but none has come close to this bound. Recently, Do Duc {\\it et\nal.} showed that the minimum insdel distance of any $[n,k]$ Reed-Solomon code\nis no more than $2n-2k$ if $q$ is large enough compared to the code length $n$;\noptimal codes that meet the new bound were also constructed explicitly. The\ncontribution of this paper is twofold. We first show that the minimum insdel\ndistance of any $[n,k]$ linear code over $\\mathbb{F}_q$ satisfies $d\\leq2n-2k$\nif $n>k>1$. This result improves and generalizes the previously known results\nin the literature. We then give a sufficient condition under which the minimum\ninsdel distance of a two-dimensional Reed-Solomon code of length $n$ over\n$\\mathbb{F}_q$ is exactly equal to $2n-4$. As a consequence, we show that the\nsufficient condition is not hard to achieve; we explicitly construct an\ninfinite family of optimal two-dimensional Reed-Somolom codes meeting the\nbound.\n"}
{"abstract": "  This paper examines the dynamics of saturated vapour in a corner formed by\ntwo walls. It is shown that, if the angle $\\phi$ between the walls is\nsufficiently small, the vapor becomes unstable and spontaneous condensation\noccurs near the corner, giving rise to an ever-growing liquid meniscus. The\ndiffuse-interface model and the lubrication approximation are used to\ndemonstrate that the condensation occurs if and only if $\\phi+2\\theta<\\pi$,\nwhere $\\theta$ is the contact angle corresponding to the fluid-solid\ncombination under consideration. This criterion has a simple physical\nexplanation: if it holds, the meniscus surface is concave -- hence, the\nso-called Kelvin effect causes condensation. If the near-vertex region of the\ncorner is smoothed, the instability can be triggered off only by a\n\\emph{finite}-size perturbation, with enough liquid to cover the smoothed aria\nby a microscopically-thin liquid film.\n"}
{"abstract": "  Motivated by a wide range of assemble-to-order systems and systems of the\ncollaborative economy applications, we introduce a stochastic matching model on\nhypergraphs and multigraphs, extending the model introduced by Mairesse and\nMoyal 2016. In this thesis, the stochastic matching model\n$(\\mathbb{S},\\Phi,\\mu)$ on general graph structures are defined as follows:\ngiven a compatibility general graph structure\n$\\mathbb{S}=(\\mathcal{V},\\mathcal{S})$ which of a set of nodes denoted by\n$\\mathcal{V}$ that represent the classes of items and by a set of edges denoted\nby $\\mathcal{S}$ that allows matching between different classes of items. Items\narrive at the system at a random time, by a sequence (assumed to be $i.i.d.$)\nthat consists of different classes of $\\mathcal{V}$, and request to be matched\ndue to their compatibility according to $\\mathbb{S}$. The compatibility by\ngroups of two or more (hypergraphical cases) and by groups of two with\npossibilities of matching between the items of the same classes (multigraphical\ncases). The unmatched items are stored in the system and wait for a future\ncompatible item and as soon as they are matched they leave it together. Upon\narrival, an item may find several possible matches, the items that leave the\nsystem depend on a matching policy $\\Phi$ to be specified. We study the\nstability of the stochastic matching model on hypergraphs, for different\nhypergraphical topologies. Then, the stability of the stochastic matching model\non multigraphs using the maximal subgraph and minimal blow-up to distinguish\nthe zone of stability.\n"}
{"abstract": "  A line-of-sight towards the Galactic Center (GC) offers the largest number of\npotentially habitable systems of any direction in the sky. The Breakthrough\nListen program is undertaking the most sensitive and deepest targeted SETI\nsurveys towards the GC. Here, we outline our observing strategies with Robert\nC. Byrd Green Bank Telescope (GBT) and Parkes telescope to conduct 600 hours of\ndeep observations across 0.7--93 GHz. We report preliminary results from our\nsurvey for ETI beacons across 1--8 GHz with 7.0 and 11.2 hours of observations\nwith Parkes and GBT, respectively. With our narrowband drifting signal search,\nwe were able to place meaningful constraints on ETI transmitters across 1--4\nGHz and 3.9--8 GHz with EIRP limits of $\\geq$4$\\times$10$^{18}$ W among 60\nmillion stars and $\\geq$5$\\times$10$^{17}$ W among half a million stars,\nrespectively. For the first time, we were able to constrain the existence of\nartificially dispersed transient signals across 3.9--8 GHz with EIRP\n$\\geq$1$\\times$10$^{14}$ W/Hz with a repetition period $\\leq$4.3 hours. We also\nsearched our 11.2 hours of deep observations of the GC and its surrounding\nregion for Fast Radio Burst-like magnetars with the DM up to 5000 pc cm$^{-3}$\nwith maximum pulse widths up to 90 ms at 6 GHz. We detected several hundred\ntransient bursts from SGR J1745$-$2900, but did not detect any new transient\nburst with the peak luminosity limit across our observed band of\n$\\geq$10$^{31}$ erg s$^{-1}$ and burst-rate of $\\geq$0.23 burst-hr$^{-1}$.\nThese limits are comparable to bright transient emission seen from other\nGalactic radio-loud magnetars, constraining their presence at the GC.\n"}
{"abstract": "  Systems for transport and processing of granular media are challenging to\nanalyse, operate and optimise. In the mining and mineral processing industries\nthese systems are chains of processes with complex interplay between the\nequipment, control, and the processed material. The material properties have\nnatural variations that are usually only known at certain locations. Therefore,\nwe explore a material-oriented approach to digital twins with a particle\nrepresentation of the granular media. In digital form, the material is treated\nas pseudo-particles, each representing a large collection of real particles of\nvarious sizes, shapes and, mineral properties. Movements and changes in the\nstate of the material are determined by the combined data from control systems,\nsensors, vehicle telematics, and simulation models at locations where no real\nsensors can see. The particle-based representation enables material tracking\nalong the chain of processes. Each digital particle can act as a carrier of\nobservational data generated by the equipment as it interacts with the real\nmaterial. This makes it possible to better learn material properties from\nprocess observations, and to predict the effect on downstream processes. We\ntest the technique on a mining simulator and demonstrate analysis that can be\nperformed using data from cross-system material tracking.\n"}
{"abstract": "  We studied the creep motion of granular materials in a gradient potential\nfield that is created using a slow spin-up experiment device. Natural sand\nconfined in the acrylic box is spun up by a controlled turntable and the\nsurface flows are captured using video-based measurements. Various spin-up\naccelerations were considered to understand the responses of creep motion on\ndifferent accelerating paths. Convergent behaviors in the morphological change\nof sand surface were observed in the final steady state. To quantify the\nquasi-static spin-up process, we examined the net flux and the surface slope as\na function of the spin rate and offset from the rotation axis. Evolution of\nsand creep motion demonstrated behaviors similar to regolith migration in\nnumeric simulations, showing intermittency like general sheared granular\nsystems. We noticed the sand surface approaches criticality as the spin-up\nproceeding, consistent with the observation that top-shaped asteroids near\nlimiting spin rate take on critical shape. Comparisons to large-scale numeric\nsimulations and analytical solutions reveal underlying similarities between our\nexperiments and the million-year evolution of asteroid regolith under YORP\nacceleration, which raises the possibility of studying asteroid surface\nprocesses in laboratory analogue experiments.\n"}
{"abstract": "  Weakly-supervised text classification aims to induce text classifiers from\nonly a few user-provided seed words. The vast majority of previous work assumes\nhigh-quality seed words are given. However, the expert-annotated seed words are\nsometimes non-trivial to come up with. Furthermore, in the weakly-supervised\nlearning setting, we do not have any labeled document to measure the seed\nwords' efficacy, making the seed word selection process \"a walk in the dark\".\nIn this work, we remove the need for expert-curated seed words by first mining\n(noisy) candidate seed words associated with the category names. We then train\ninterim models with individual candidate seed words. Lastly, we estimate the\ninterim models' error rate in an unsupervised manner. The seed words that yield\nthe lowest estimated error rates are added to the final seed word set. A\ncomprehensive evaluation of six binary classification tasks on four popular\ndatasets demonstrates that the proposed method outperforms a baseline using\nonly category name seed words and obtained comparable performance as a\ncounterpart using expert-annotated seed words.\n"}
{"abstract": "  Quantum theory of a test field on a quantum cosmological spacetime may be\nviewed as a theory of the test field on an emergent classical background. In\nsuch a case, the resulting dressed metric for the field propagation is a\nfunction of the quantum fluctuations of the original geometry. When the\nbackreaction is negligible, massive modes can experience an anisotropic Bianchi\ntype I background. The field modes propagating on such a\nquantum-gravity-induced spacetime can then unveil interesting phenomenological\nconsequences of the super-Planckian scales, such as gravitational particle\nproduction. The aim of this paper is to address the issue of gravitational\nparticle production associated to the massive modes in such an anisotropic\ndressed spacetime. By imposing a suitable adiabatic condition on the vacuum\nstate and computing the energy density of the created particles, the\nsignificance of the particle production on the dynamics of the universe in\nPlanck era is discussed.\n"}
{"abstract": "  The compositions of planet-forming disks are set by a combination of material\ninherited from the interstellar medium and material reprocessed during disk\nformation and evolution. Indeed, comets and primitive meteorites exhibit\ninterstellar-like isotopic ratios and/or volatile compositions, supporting that\nsome pristine material was incorporated intact into icy planetesimals in the\nSolar Nebula. To date, the survival of volatile interstellar material in the\ndisk stage has not been modeled using realistic disk physics. Here, we present\na modeling framework to track the destruction of interstellar ices on dust\ngrains undergoing transport processes within a disk, with a particular focus on\nexplaining the incorporation of pristine material into icy planetesimals. We\nfind it is difficult to explain inheritance through the local assembly of\ncomets, as ice destruction is rapid for small (<10um) grains in the inner few\ntens of au. Instead, a plausible pathway to inheritance is to form pebbles at\nlarger disk radii, which then drift inwards to the comet-forming zone with\ntheir ices mostly preserved. Small grains beyond ~100 au can experience ice\nphotodissociation at the tens of percent level, however little of the ice is\nactually lost from the grain, likely making this a robust site for in situ ice\nchemistry. Our models also indicate that many complex organic species should\nsurvive passage through the disk intact. This raises the possibility that\norganics synthesized in the interstellar medium can be delivered to terrestrial\nplanets by icy body impact and thus potentially participate in origins of life\nchemistry.\n"}
{"abstract": "  Pauli blocking of spontaneous emission is responsible for the stability of\natoms. Higher electronic orbitals cannot decay to lower-lying states if they\nare already occupied -- this is Pauli blocking due to occupation of internal\nstates. Pauli blocking also occurs when free atoms scatter light elastically\n(Rayleigh scattering) and the final external momentum states are already\noccupied. A suppression of the total rate of light scattering requires a\nquantum-degenerate Fermi gas with a Fermi energy larger than the photon recoil\nenergy. This has been predicted more than 30 years ago, but never realized.\nHere we report the creation of a dense Fermi gas of ultracold lithium atoms and\nshow that at low temperatures light scattering is suppressed. We also explore\nthe suppression of inelastic light scattering when two colliding atoms emit\nlight shifted in frequency.\n"}
{"abstract": "  It is shown how the extended conformal Einstein field equations and a gauge\nbased on the properties of conformal geodesics can be used to analyse the\nnon-linear stability of Einstein spaces with negative scalar curvature. This\nclass of spacetimes admits a smooth conformal extension with a spacelike\nconformal boundary. Central to the analysis is the use of conformal Gaussian\nsystems to obtain a hyperbolic reduction of the conformal Einstein field\nequations for which standard Cauchy stability results for symmetric hyperbolic\nsystems can be employed. The use of conformal methods allows to rephrase the\nquestion of global existence of solutions to the Einstein field equations into\nconsiderations of finite existence time for the conformal evolution system.\n"}
{"abstract": "  The specific energy loss (dE/dx) resolution is critical for particle\nidentification and separation in TPC. In connection with the R&D for the TPC at\nthe Circular Electron Positron Collider (CEPC), we find the laser ionisation\nobeys the Gaussian distribution theoretically and experimentally. Based on\nthis, we develop a novel method using a 266 nm UV laser to measure the dE/dx\nresolution. Comparing to the traditional methods with particle beams or cosmic\nrays, the novel method with 266 nm UV laser is more convenient and flexible. We\nverify the method's feasibility with a TPC prototype and extrapolate the dE/dx\nresolution to the CEPC TPC. The estimation of the dE/dx resolution is without\nthe magnetic field.\n"}
{"abstract": "  Linear Kinematic Features (LKFs) are found everywhere in the Arctic sea ice\ncover. They are strongly localized deformations often associated with the\nformation of leads and pressure ridges. Viscous-plastic sea ice models start to\ngenerate LKFs at high spatial grid resolution, typically with a grid spacing\nbelow 5 km. Besides grid spacing, other aspects of a numerical implementation,\nsuch as discretization details, may affect the number and definition of\nsimulated LKFs. To explore these effects, the solutions of sea ice models with\ndifferent grid spacings, mesh types, and numerical discretization techniques\nare compared in an idealized configuration, which could also serve as a\nbenchmark problem in the future. The A, B, and C-grid discretizations of sea\nice dynamics on quadrilateral meshes leads to a similar number of LKFs as the\nA-grid approximation on triangular meshes (with the same number of vertices).\nThe discretization on an Arakawa CD-grid on both structured quadrilateral and\ntriangular meshes resolves the same number of LKFs as conventional Arakawa\nA-grid, B-grid, and C-grid approaches, but on two times coarser meshes. This is\ndue to the fact that the CD-grid approach has a higher number of degrees of\nfreedom to discretize the velocity field. Due to its enhanced resolving\nproperties, the CD-grid discretization is an attractive alternative to\nconventional discretizations.\n"}
{"abstract": "  To mitigate the spread of COVID-19 pandemic, decision-makers and public\nauthorities have announced various non-pharmaceutical policies. Analyzing the\ncausal impact of these policies in reducing the spread of COVID-19 is important\nfor future policy-making. The main challenge here is the existence of\nunobserved confounders (e.g., vigilance of residents). Besides, as the\nconfounders may be time-varying during COVID-19 (e.g., vigilance of residents\nchanges in the course of the pandemic), it is even more difficult to capture\nthem. In this paper, we study the problem of assessing the causal effects of\ndifferent COVID-19 related policies on the outbreak dynamics in different\ncounties at any given time period. To this end, we integrate data about\ndifferent COVID-19 related policies (treatment) and outbreak dynamics (outcome)\nfor different United States counties over time and analyze them with respect to\nvariables that can infer the confounders, including the covariates of different\ncounties, their relational information and historical information. Based on\nthese data, we develop a neural network based causal effect estimation\nframework which leverages above information in observational data and learns\nthe representations of time-varying (unobserved) confounders. In this way, it\nenables us to quantify the causal impact of policies at different\ngranularities, ranging from a category of policies with a certain goal to a\nspecific policy type in this category. Besides, experimental results also\nindicate the effectiveness of our proposed framework in capturing the\nconfounders for quantifying the causal impact of different policies. More\nspecifically, compared with several baseline methods, our framework captures\nthe outbreak dynamics more accurately, and our assessment of policies is more\nconsistent with existing epidemiological studies of COVID-19.\n"}
{"abstract": "  Keystroke dynamics can be used to analyze the way that users type by\nmeasuring various aspects of keyboard input. Previous work has demonstrated the\nfeasibility of user authentication and identification utilizing keystroke\ndynamics. In this research, we consider a wide variety of machine learning and\ndeep learning techniques based on fixed-text keystroke-derived features, we\noptimize the resulting models, and we compare our results to those obtained in\nrelated research. We find that models based on extreme gradient boosting\n(XGBoost) and multi-layer perceptrons (MLP)perform well in our experiments. Our\nbest models outperform previous comparable research.\n"}
{"abstract": "  Transformer based knowledge tracing model is an extensively studied problem\nin the field of computer-aided education. By integrating temporal features into\nthe encoder-decoder structure, transformers can processes the exercise\ninformation and student response information in a natural way. However, current\nstate-of-the-art transformer-based variants still share two limitations. First,\nextremely long temporal features cannot well handled as the complexity of\nself-attention mechanism is O(n2). Second, existing approaches track the\nknowledge drifts under fixed a window size, without considering different\ntemporal-ranges. To conquer these problems, we propose MUSE, which is equipped\nwith multi-scale temporal sensor unit, that takes either local or global\ntemporal features into consideration. The proposed model is capable to capture\nthe dynamic changes in users knowledge states at different temporal-ranges, and\nprovides an efficient and powerful way to combine local and global features to\nmake predictions. Our method won the 5-th place over 3,395 teams in the Riiid\nAIEd Challenge 2020.\n"}
{"abstract": "  We introduce a new concept of dissipative varifold solution to models of two\nphase compressible viscous fluids. In contrast with the existing approach based\non the Young measure description, the new formulation is variational combining\nthe energy and momentum balance in a single inequality. We show the existence\nof dissipative varifold solutions for a large class of general viscous fluids\nwith non--linear dependence of the viscous stress on the symmetric velocity\ngradient.\n"}
{"abstract": "  We formulate dynamical phase transitions in subsystems embedded in larger\nquantum systems. Introducing the entanglement echo as an overlap of the initial\nand instantaneous entanglement ground states, we show its analytic structure\nafter a quench provides natural definition of dynamical phase transitions in\nthe subsystem. These transitions come in two varieties, the entanglement-type\ntransitions and the bulk-type Loschmidt transitions. The entanglement-type\ntransitions arise from periodic reorganization of quantum correlations between\nthe subsystem and its environment, manifesting in instantaneous entanglement\nground state degeneracies. Furthermore, the entanglement echo distinguishes the\ndirection of the quench, resolves spatially distinct dynamical phase\ntransitions for non-uniform quenches and give rise to sharply-defined\ntransitions for mixed initial states. We propose an experimental probe to\nidentify entanglement-type transitions through temporal changes in subsystem\nfluctuations.\n"}
{"abstract": "  This work presents an ab-initio study of a few-layers hexagonal boron nitride\n(hBN) and hBN-graphene heterostructure sandwiched between Ni(111) layers. The\naim of this study is to understand the electron transmission process through\nthe interface. Spin-polarized density functional theory calculations and\ntransmission probability calculations were conducted on Ni(111)/$n$hBN/Ni(111)\nwith $n$ = 2, 3, 4, and 5 as well as on Ni(111)/hBN-Gr-hBN/Ni(111). Slabs with\nmagnetic alignment in an anti-parallel configuration (APC) and parallel\nconfiguration (PC) were considered. The pd-hybridizations at both the upper and\nlower interfaces between the Ni slabs and hBN were found to stabilize the\nsystem. The Ni/nhBN/Ni magnetic tunnel junction (MTJ) was found to exhibit a\nhigh tunneling magnetoresistance (TMR) ratio at ~0.28 eV for $n$ = 2 and 0.34\neV for $n$ > 2, which are slightly higher than the Fermi energy. The observed\nshifting of this high TMR ratio originates from the transmission of electrons\nthrough the surface states of the $d_{z^2}$-orbital of Ni atoms at interfaces\nwhich are hybridized with the $p_z$-orbital of N atoms. In the case of $n$ > 2,\nthe proximity effect causes an evanescent wave, contributing to decreasing\ntransmission probability but increasing the TMR ratio. However, TMR ratio, as\nwell as transmission probability, was found to be increased, by replacing the\nunhybridized hBN layer of the Ni/3hBN/Ni MTJ with graphene, thus becoming\nNi/hBN-Gr-hBN/Ni. A TMR ratio as high as ~1200% was observed at an energy of\n0.34 eV, which is higher than the Fermi energy. Furthermore, a design is\nproposed for a device based on a new reading mechanism using the high TMR\nobserved just above the Fermi energy level.\n"}
{"abstract": "  We give a detailed presentation of the theory and numerical implementation of\nan expression for the adiabatic energy flux in extended systems, derived from\ndensity-functional theory. This expression can be used to estimate the heat\nconductivity from equilibrium ab initio molecular dynamics, using the\nGreen-Kubo linear response theory of transport coefficients. Our expression is\nimplemented in an open-source component of the QE suite of computer codes for\nquantum mechanical materials modelling, which is being made publicly available.\n"}
{"abstract": "  We propose a method to reveal axions and axion-like particles based on\ninterferometric measurement of neutron beams. We consider an interferometer in\nwhich the neutron beam is split in two sub-beams propagating in regions with\ndifferently oriented magnetic fields. The beam paths and the strength of the\nmagnetic fields are set in such a way that all the contributions to the phase\ndifference but the one due to axion-induced interactions are removed. The\nresulting phase difference is directly related to the presence of axions. Our\nresults show that such a phase is in principle detectable with neutron\ninterferometry, possibly proving the existence of axions and axion-like\nparticles.\n"}
{"abstract": "  We provide a combined theoretical and experimental study of the electronic\nstructure and the optical absorption edge of the orthorhombic perovskite\nLaInO$_{3}$. Employing density-functional theory and many-body perturbation\ntheory, we predict a direct electronic quasiparticle band gap of about 5 eV and\nan effective electron (hole) mass of 0.31 (0.48) m$_{0}$. We find the\nlowest-energy excitation at 0.2 eV below the fundamental gap, reflecting a\nsizeable electron-hole attraction. Since the transition from the valence band\nmaximum (VBM, $\\Gamma$ point) is, however, dipole forbidden the onset is\ncharacterized by weak excitations from transitions around it. The first intense\nexcitation appears about 0.32 eV above. Interestingly, this value coincides\nwith an experimental value obtained by ellipsometry (4.80 eV) which is higher\nthan the onset from optical absorption spectroscopy (4.35 eV). The latter\ndiscrepancy is attributed to the fact that the weak transitions that define the\noptical gap are not resolved by the ellipsometry measurement. The absorption\nedge shows a strong dependency on the light polarization, reflecting the\ncharacter of the involved valence states. Temperature-dependent measurements\nshow a redshift of the optical gap by about 120 meV by increasing the\ntemperature from 5 to 300 K. Renormalization due to zero-point vibrations is\nextrapolated from the latter measurement to amount to 150 meV. By adding the\nexcitonic binding energy of 0.2 eV obtained theoretically to the experimental\noptical absorption onset, we determine the fundamental band gap at room\ntemperature to be 4.55 eV.\n"}
{"abstract": "  We investigate the coexistence of underlay spectrum sharing in cell-free\nmassive multiple-input multiple-output (MIMO) systems. A primary system with\ngeographically distributed primary access points (P-APs) serves a multitude of\nprimary users (PUs), while a secondary system serves a large number of\nsecondary users (SUs) in the same primary/licensed spectrum by exploiting the\nunderlay spectrum sharing. To mitigate the secondary co-channel interference\ninflected at PUs, stringent secondary transmit power constraints are defined\nfor the secondary access points (S-APs). A generalized pilots sharing scheme is\nused to locally estimate the uplink channels at P-APs/S-APs, and thereby,\nconjugate precoders are adopted to serve PUs/SUs in the same time-frequency\nresource element. Moreover, the effect of a user-centric AP clustering scheme\nis investigated by assigning a suitable set of APs to a particular user. The\nimpact of estimated downlink (DL) channels at PUs/SUs via DL pilots beamformed\nby P-APs/S-APs is investigated. The achievable primary/secondary rates at\nPUs/SUs are derived for the statistical DL and estimated DL CSI cases.\nUser-fairness for PUs/SUs is achieved by designing efficient transmit power\ncontrol policies based on a multi-objective optimization problem formulation of\njoint underlay spectrum sharing and max-min criteria. The proposed orthogonal\nmultiple-access based analytical framework is also extended to facilitate\nnon-orthogonal multiple-access. Our analysis and numerical results manifest\nthat the primary/secondary performance of underlay spectrum sharing can be\nboosted by virtue of the average reduction of transmit powers/path-losses,\nuniform coverage/service, and macro-diversity gains, which are inherent to\ndistributed transmissions/receptions of cell-free massive MIMO.\n"}
{"abstract": "  The resistance to failure through tearing is a crucial mechanical property\nfor the application of different elastomers. In this work, graphene\nnanoplatelets (GNPs) were introduced into a fluoroelastomer (FKM) matrix with\nthe aim of improving its tear resistance. The fracture energy through tearing\nwas evaluated using the pure shear test. It was found that the tearing energy\nincreased linearly with the volume fraction of the GNPs. At the maximum GNP\ncontent, the tearing resistance was 3 times higher, suggesting efficient\ntoughening from the GNPs. Theoretical analysis of the micromechanics was\nconducted by considering debonding and pull-out of the nanoplatelets as\npossible toughening mechanisms. It was determined quantitatively that the main\ntoughening mechanism was debonding of the interface rather than pull-out. The\nformation of cavities at flake ends during the deformation, as confirmed by\nscanning electron microscopy, was found to contribute to the remarkably high\ninterfacial debonding energy (~1 kJ/m2).\n"}
{"abstract": "  Recently, van der Waals heterostructure has attracted interest both\ntheoretically and experimentally for their potential applications in\nphotoelectronic devices, photovoltaic devices, plasmonic devices and\nphotocatalysis. Inspired by this, we design a lepidocrocite-type TiO2/GaSe\nheterostructure. Via first-principles simulations, we show that such a\nheterostructure is a direct bandgap semiconductor with a strong and broad\noptical absorption, ranging from visible light to UV region, exhibiting its\npotential application in photoelectronic and photovoltaic devices. With the\nplanar-averaged electron density difference and Bader charge analysis, the\nheterostructure shows a strong capacity of enhancing the charge redistribution\nespecially at the interface, prolonging the lifetime of excitons, and hence\nimproving photocatalytic performance. By applying biaxial strain and interlayer\ncoupling, the heterostructure exhibits a direct-indirect bandgap transition and\nshows a potential for mechanical sensors due to the smooth and linear variation\nof bandgaps. Furthermore, our result indicates that a lower interlayer distance\nleads to a stronger charge redistribution. The calculation of irradiating\nultrafast on the heterostructure further reveals a semiconductor-metal\ntransition for the heterostructure. Moreover, we find an enhanced induced\nplasmonic current in the heterostructure under both x-polarized and z-polarized\nlaser, which is beneficial to plasmonic devices designs. Our research provides\nvaluable insight in applying the lepidocrocite-type TiO2/GaSe heterostructure\nin photoelectronic, photovoltaic, photocatalytic, mechanical sensing and\nplasmonic realms.\n"}
{"abstract": "  In the first part of this paper we analyzed finite non-deterministic matrix\nsemantics for propositional non-normal modal logics as an alternative to the\nstandard Kripke's possible world semantics. This kind of modal systems\ncharacterized by finite non-deterministic matrices was originally proposed by\nJu. Ivlev in the 70's. The aim of this second paper is to introduce a formal\nnon-deterministic semantical framework for the quantified versions of some\nIvlev-like non-normal modal logics. It will be shown that several well-known\ncontroversial issues of quantified modal logics, relative to the identity\npredicate, Barcan's formulas, and de dicto and de re modalities, can be tackled\nfrom a new angle within the present framework.\n"}
{"abstract": "  In this paper, we explore the effects of language variants, data sizes, and\nfine-tuning task types in Arabic pre-trained language models. To do so, we\nbuild three pre-trained language models across three variants of Arabic: Modern\nStandard Arabic (MSA), dialectal Arabic, and classical Arabic, in addition to a\nfourth language model which is pre-trained on a mix of the three. We also\nexamine the importance of pre-training data size by building additional models\nthat are pre-trained on a scaled-down set of the MSA variant. We compare our\ndifferent models to each other, as well as to eight publicly available models\nby fine-tuning them on five NLP tasks spanning 12 datasets. Our results suggest\nthat the variant proximity of pre-training data to fine-tuning data is more\nimportant than the pre-training data size. We exploit this insight in defining\nan optimized system selection model for the studied tasks.\n"}
{"abstract": "  Predicting the final ischaemic stroke lesion provides crucial information\nregarding the volume of salvageable hypoperfused tissue, which helps physicians\nin the difficult decision-making process of treatment planning and\nintervention. Treatment selection is influenced by clinical diagnosis, which\nrequires delineating the stroke lesion, as well as characterising cerebral\nblood flow dynamics using neuroimaging acquisitions. Nonetheless, predicting\nthe final stroke lesion is an intricate task, due to the variability in lesion\nsize, shape, location and the underlying cerebral haemodynamic processes that\noccur after the ischaemic stroke takes place. Moreover, since elapsed time\nbetween stroke and treatment is related to the loss of brain tissue, assessing\nand predicting the final stroke lesion needs to be performed in a short period\nof time, which makes the task even more complex. Therefore, there is a need for\nautomatic methods that predict the final stroke lesion and support physicians\nin the treatment decision process. We propose a fully automatic deep learning\nmethod based on unsupervised and supervised learning to predict the final\nstroke lesion after 90 days. Our aim is to predict the final stroke lesion\nlocation and extent, taking into account the underlying cerebral blood flow\ndynamics that can influence the prediction. To achieve this, we propose a\ntwo-branch Restricted Boltzmann Machine, which provides specialized data-driven\nfeatures from different sets of standard parametric Magnetic Resonance Imaging\nmaps. These data-driven feature maps are then combined with the parametric\nMagnetic Resonance Imaging maps, and fed to a Convolutional and Recurrent\nNeural Network architecture. We evaluated our proposal on the publicly\navailable ISLES 2017 testing dataset, reaching a Dice score of 0.38, Hausdorff\nDistance of 29.21 mm, and Average Symmetric Surface Distance of 5.52 mm.\n"}
{"abstract": "  The motion of Brownian particles in nonlinear baths, such as, e.g.,\nviscoelastic fluids, is of great interest. We theoretically study a simple\nmodel for such bath, where two particles are coupled via a sinusoidal\npotential. This model, which is an extension of the famous Prandtl Tomlinson\nmodel, has been found to reproduce some aspects of recent experiments, such as\nshear-thinning and position oscillations [J. Chem. Phys. {\\bf 154}, 184904\n(2021)]. Analyzing this model in detail, we show that the predicted behavior of\nposition oscillations agrees qualitatively with experimentally observed trends;\n(i) oscillations appear only in a certain regime of velocity and trap stiffness\nof the confining potential, and (ii), the amplitude and frequency of\noscillations increase with driving velocity, the latter in a linear fashion.\nIncreasing the potential barrier height of the model yields a rupture\ntransition as a function of driving velocity, where the system abruptly changes\nfrom a mildly driven state to a strongly driven state. The frequency of\noscillations scales as $(v_0-v_0^*)^{1/2}$ near the rupture velocity $v_0^*$,\nfound for infinite trap stiffness. Investigating the (micro-)viscosity for\ndifferent parameter ranges, we note that position oscillations leave their\nsignature by an additional (mild) plateau in the flow curves, suggesting that\noscillations influence the micro-viscosity. For a time-modulated driving, the\nmean friction force of the driven particle shows a pronounced resonance\nbehavior, i.e, it changes strongly as a function of driving frequency. The\nmodel has two known limits: For infinite trap stiffness, it can be mapped to\ndiffusion in a tilted periodic potential. For infinite bath friction, the\noriginal Prandtl Tomlinson model is recovered. We find that the flow curve of\nthe model (roughly) crosses over between these two limiting cases.\n"}
{"abstract": "  As a part of Advanced Driver Assistance Systems (ADASs), Consensus-based\nSpeed Advisory Systems (CSAS) have been proposed to recommend a common speed to\na group of vehicles for specific application purposes, such as emission control\nand energy management. With Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure\n(V2I) technologies and advanced control theories in place, state-of-the-art\nCSAS can be designed to get an optimal speed in a privacy-preserving and\ndecentralized manner. However, the current method only works for specific cost\nfunctions of vehicles, and its execution usually involves many algorithm\niterations leading long convergence time. Therefore, the state-of-the-art\ndesign method is not applicable to a CSAS design which requires real-time\ndecision making. In this paper, we address the problem by introducing MPC-CSAS,\na Multi-Party Computation (MPC) based design approach for privacy-preserving\nCSAS. Our proposed method is simple to implement and applicable to all types of\ncost functions of vehicles. Moreover, our simulation results show that the\nproposed MPC-CSAS can achieve very promising system performance in just one\nalgorithm iteration without using extra infrastructure for a typical CSAS.\n"}
{"abstract": "  We study the gradient flow for a relaxed approximation to the\nKullback-Leibler (KL) divergence between a moving source and a fixed target\ndistribution. This approximation, termed the KALE (KL approximate lower-bound\nestimator), solves a regularized version of the Fenchel dual problem defining\nthe KL over a restricted class of functions. When using a Reproducing Kernel\nHilbert Space (RKHS) to define the function class, we show that the KALE\ncontinuously interpolates between the KL and the Maximum Mean Discrepancy\n(MMD). Like the MMD and other Integral Probability Metrics, the KALE remains\nwell defined for mutually singular distributions. Nonetheless, the KALE\ninherits from the limiting KL a greater sensitivity to mismatch in the support\nof the distributions, compared with the MMD. These two properties make the KALE\ngradient flow particularly well suited when the target distribution is\nsupported on a low-dimensional manifold. Under an assumption of sufficient\nsmoothness of the trajectories, we show the global convergence of the KALE\nflow. We propose a particle implementation of the flow given initial samples\nfrom the source and the target distribution, which we use to empirically\nconfirm the KALE's properties.\n"}
{"abstract": "  Photo- and thermally stimulated luminescence characteristics of\nGd3(Ga,Al)5O12:Ce single crystals co-doped with W and Mo are investigated in\nthe 85 - 510 K temperature range and compared with the corresponding\ncharacteristics of the undoped and Ce3+ - doped Gd3(Ga,Al)5O12 single crystals\nof similar composition. A strong effect of the W and Mo impurity ions appears\nin the photoluminescence spectra and temperature dependences of the\nphotoluminescence intensity, afterglow intensity and decay kinetics, thermally\nstimulated luminescence (TSL) intensity and TSL glow curves, excitation spectra\nof the TSL glow curve peaks and activation energy of their creation. The\nobtained results are explained by the enhancement of the intrinsic emission\ncontribution into the luminescence spectrum of Gd3(Ga,Al)5O12:Ce,W and\nGd3(Ga,Al)5O12:Ce,Mo due to a large concentration of various W - and Mo -\nrelated electron traps in these crystals and, consequently, the O- - type hole\ncenters, as well as intrinsic crystal lattice defects (e.g., cation vacancies\nneeded for the excess positive charge compensation of the W6+ and Mo6+ ions).\nThe influence of the co-doping with W and Mo ions on the scintillation\ncharacteristics of Gd3(Ga,Al)5O12:Ce is discussed.\n"}
{"abstract": "  Domestic and service robots have the potential to transform industries such\nas health care and small-scale manufacturing, as well as the homes in which we\nlive. However, due to the overwhelming variety of tasks these robots will be\nexpected to complete, providing generic out-of-the-box solutions that meet the\nneeds of every possible user is clearly intractable. To address this problem,\nrobots must therefore not only be capable of learning how to complete novel\ntasks at run-time, but the solutions to these tasks must also be informed by\nthe needs of the user. In this paper we demonstrate how behaviour trees, a well\nestablished control architecture in the fields of gaming and robotics, can be\nused in conjunction with natural language instruction to provide a robust and\nmodular control architecture for instructing autonomous agents to learn and\nperform novel complex tasks. We also show how behaviour trees generated using\nour approach can be generalised to novel scenarios, and can be re-used in\nfuture learning episodes to create increasingly complex behaviours. We validate\nthis work against an existing corpus of natural language instructions,\ndemonstrate the application of our approach on both a simulated robot solving a\ntoy problem, as well as two distinct real-world robot platforms which,\nrespectively, complete a block sorting scenario, and a patrol scenario.\n"}
{"abstract": "  The many-body theory of interacting electrons poses an intrinsically\ndifficult problem that requires simplifying assumptions. For the determination\nof electronic screening properties of the Coulomb interaction, the Random Phase\nApproximation (RPA) provides such a simplification. Here, we explicitly show\nthat this approximation is justified for band structures with sizeable band\ngaps. This is when the electronic states responsible for the screening are\nenergetically far away from the Fermi level, which is equivalent to a short\nelectronic propagation length of these states. The RPA contains exactly those\ndiagrams in which the classical Coulomb interaction covers all distances,\nwhereas neglected vertex corrections involve quantum tunneling through the\nbarrier formed by the band gap. Our analysis of electron-electron interactions\nprovides a real-space analogy to Migdal's theorem on the smallness of vertex\ncorrections in electron-phonon problems. An important application is the\nincreasing use of constrained Random Phase Approximation (cRPA) calculations of\neffective interactions. We find that their usage of Kohn-Sham energies already\naccounts for the leading local (excitonic) vertex correction in insulators.\n"}
{"abstract": "  We consider the mathematical analysis and numerical approximation of a system\nof nonlinear partial differential equations that arises in models that have\nrelevance to steady isochoric flows of colloidal suspensions. The symmetric\nvelocity gradient is assumed to be a monotone nonlinear function of the\ndeviatoric part of the Cauchy stress tensor. We prove the existence of a unique\nweak solution to the problem, and under the additional assumption that the\nnonlinearity involved in the constitutive relation is Lipschitz continuous we\nalso prove uniqueness of the weak solution. We then construct mixed finite\nelement approximations of the system using both conforming and nonconforming\nfinite element spaces. For both of these we prove the convergence of the method\nto the unique weak solution of the problem, and in the case of the conforming\nmethod we provide a bound on the error between the analytical solution and its\nfinite element approximation in terms of the best approximation error from the\nfinite element spaces. We propose first a Lions-Mercier type iterative method\nand next a classical fixed-point algorithm to solve the finite-dimensional\nproblems resulting from the finite element discretisation of the system of\nnonlinear partial differential equations under consideration and present\nnumerical experiments that illustrate the practical performance of the proposed\nnumerical method.\n"}
{"abstract": "  This paper analyzes the global dynamics of 1-dimensional agent arrays with\nnearest neighbor linear couplings. The equations of motion are second order\nlinear ODEs with constant coeffcients. The novel part of this research is that\nthe couplings are different for each distinct agent. We allow the forces to\ndepend on the positions and velocity (damping terms) but the magnitudes of both\nthe position and velocity couplings are different for each agent. We, also, do\nnot assume that the forces are \"Newtonian\" (i.e. the force due to A on B equals\nthe minus the force of B on A) as this assumption does not apply to certain\nsituations, such as traffic modeling. For example, driver A reacting to driver\nB does not imply the opposite reaction in driver B. There are no known\nanalytical means to solve these systems, even though they are linear, and so\nrelatively little is known about them. This paper is a generalization of\nprevious work that computed the global dynamics of 1-dimensional sequences of\nidentical agents [3] assuming periodic boundary conditions. In this paper, we\npush that method further, similar to [2], and use an extended periodic boundary\ncondition to to gain quantitative insights to the systems under consideration.\nWe find that we can approximate the global dynamics of such a system by\ncarefully analyzing the low-frequency behavior of the system with (generalized)\nperiodic boundary conditions.\n"}
{"abstract": "  Deploying deep learning models in time-critical applications with limited\ncomputational resources, for instance in edge computing systems and IoT\nnetworks, is a challenging task that often relies on dynamic inference methods\nsuch as early exiting. In this paper, we introduce a novel architecture for\nearly exiting based on the vision transformer architecture, as well as a\nfine-tuning strategy that significantly increase the accuracy of early exit\nbranches compared to conventional approaches while introducing less overhead.\nThrough extensive experiments on image and audio classification as well as\naudiovisual crowd counting, we show that our method works for both\nclassification and regression problems, and in both single- and multi-modal\nsettings. Additionally, we introduce a novel method for integrating audio and\nvisual modalities within early exits in audiovisual data analysis, that can\nlead to a more fine-grained dynamic inference.\n"}
{"abstract": "  The discovery of new multicomponent inorganic compounds can provide direct\nsolutions to many scientific and engineering challenges, yet the vast size of\nthe uncharted material space dwarfs current synthesis throughput. While the\ncomputational crystal structure prediction is expected to mitigate this\nfrustration, the NP-hardness and steep costs of density functional theory (DFT)\ncalculations prohibit material exploration at scale. Herein, we introduce\nSPINNER, a highly efficient and reliable structure-prediction framework based\non exhaustive random searches and evolutionary algorithms, which is completely\nfree from empiricism. Empowered by accurate neural network potentials, the\nprogram can navigate the configuration space faster than DFT by more than\n10$^{2}$-fold. In blind tests on 60 ternary compositions diversely selected\nfrom the experimental database, SPINNER successfully identifies experimental\n(or theoretically more stable) phases for ~80% of materials within 5000\ngenerations, entailing up to half a million structure evaluations for each\ncomposition. When benchmarked against previous data mining or DFT-based\nevolutionary predictions, SPINNER identifies more stable phases in the majority\nof cases. By developing a reliable and fast structure-prediction framework,\nthis work opens the door to large-scale, unbounded computational exploration of\nundiscovered inorganic crystals.\n"}
{"abstract": "  Training deep neural networks for automatic speech recognition (ASR) requires\nlarge amounts of transcribed speech. This becomes a bottleneck for training\nrobust models for accented speech which typically contains high variability in\npronunciation and other semantics, since obtaining large amounts of annotated\naccented data is both tedious and costly. Often, we only have access to large\namounts of unannotated speech from different accents. In this work, we leverage\nthis unannotated data to provide semantic regularization to an ASR model that\nhas been trained only on one accent, to improve its performance for multiple\naccents. We propose Accent Pre-Training (Acc-PT), a semi-supervised training\nstrategy that combines transfer learning and adversarial training. Our approach\nimproves the performance of a state-of-the-art ASR model by 33% on average over\nthe baseline across multiple accents, training only on annotated samples from\none standard accent, and as little as 105 minutes of unannotated speech from a\ntarget accent.\n"}
{"abstract": "  Estimating health benefits of reducing fossil fuel use from improved air\nquality provides important rationales for carbon emissions abatement.\nSimulating pollution concentration is a crucial step of the estimation, but\ntraditional approaches often rely on complicated chemical transport models that\nrequire extensive expertise and computational resources. In this study, we\ndevelop a novel and succinct machine learning framework that is able to provide\nprecise and robust annual average fine particle (PM2.5) concentration\nestimations directly from a high-resolution fossil energy use data set. The\naccessibility and applicability of this framework show great potentials of\nmachine learning approaches for integrated assessment studies. Applications of\nthe framework with Chinese data reveal highly heterogeneous health benefits of\nreducing fossil fuel use in different sectors and regions in China with a mean\nof \\$34/tCO2 and a standard deviation of \\$84/tCO2. Reducing rural and\nresidential coal use offers the highest co-benefits with a mean of \\$360/tCO2.\nOur findings prompt careful policy designs to maximize cost-effectiveness in\nthe transition towards a carbon-neutral energy system.\n"}
{"abstract": "  Recent advances in bioimaging have provided scientists a superior high\nspatial-temporal resolution to observe dynamics of living cells as 3D\nvolumetric videos. Unfortunately, the 3D biomedical video analysis is lagging,\nimpeded by resource insensitive human curation using off-the-shelf 3D analytic\ntools. Herein, biologists often need to discard a considerable amount of rich\n3D spatial information by compromising on 2D analysis via maximum intensity\nprojection. Recently, pixel embedding-based cell instance segmentation and\ntracking provided a neat and generalizable computing paradigm for understanding\ncellular dynamics. In this work, we propose a novel spatial-temporal\nvoxel-embedding (VoxelEmbed) based learning method to perform simultaneous cell\ninstance segmenting and tracking on 3D volumetric video sequences. Our\ncontribution is in four-fold: (1) The proposed voxel embedding generalizes the\npixel embedding with 3D context information; (2) Present a simple multi-stream\nlearning approach that allows effective spatial-temporal embedding; (3)\nAccomplished an end-to-end framework for one-stage 3D cell instance\nsegmentation and tracking without heavy parameter tuning; (4) The proposed 3D\nquantification is memory efficient via a single GPU with 12 GB memory. We\nevaluate our VoxelEmbed method on four 3D datasets (with different cell types)\nfrom the ISBI Cell Tracking Challenge. The proposed VoxelEmbed method achieved\nconsistent superior overall performance (OP) on two densely annotated datasets.\nThe performance is also competitive on two sparsely annotated cohorts with\n20.6% and 2% of data-set having segmentation annotations. The results\ndemonstrate that the VoxelEmbed method is a generalizable and memory-efficient\nsolution.\n"}
{"abstract": "  Topic model evaluation, like evaluation of other unsupervised methods, can be\ncontentious. However, the field has coalesced around automated estimates of\ntopic coherence, which rely on the frequency of word co-occurrences in a\nreference corpus. Contemporary neural topic models surpass classical ones\naccording to these metrics. At the same time, topic model evaluation suffers\nfrom a validation gap: automated coherence, developed for classical models, has\nnot been validated using human experimentation for neural models. In addition,\na meta-analysis of topic modeling literature reveals a substantial\nstandardization gap in automated topic modeling benchmarks. To address the\nvalidation gap, we compare automated coherence with the two most widely\naccepted human judgment tasks: topic rating and word intrusion. To address the\nstandardization gap, we systematically evaluate a dominant classical model and\ntwo state-of-the-art neural models on two commonly used datasets. Automated\nevaluations declare a winning model when corresponding human evaluations do\nnot, calling into question the validity of fully automatic evaluations\nindependent of human judgments.\n"}
{"abstract": "  Low rank matrix recovery problems, including matrix completion and matrix\nsensing, appear in a broad range of applications. In this work we present GNMR\n-- an extremely simple iterative algorithm for low rank matrix recovery, based\non a Gauss-Newton linearization. On the theoretical front, we derive recovery\nguarantees for GNMR in both the matrix sensing and matrix completion settings.\nA key property of GNMR is that it implicitly keeps the factor matrices\napproximately balanced throughout its iterations. On the empirical front, we\nshow that for matrix completion with uniform sampling, GNMR performs better\nthan several popular methods, especially when given very few observations close\nto the information limit.\n"}
{"abstract": "  Novel Coronavirus disease (COVID-19) is a highly contagious respiratory\ninfection that has had devastating effects on the world. Recently, new COVID-19\nvariants are emerging making the situation more challenging and threatening.\nEvaluation and quantification of COVID-19 lung abnormalities based on chest\nComputed Tomography (CT) scans can help determining the disease stage,\nefficiently allocating limited healthcare resources, and making informed\ntreatment decisions. During pandemic era, however, visual assessment and\nquantification of COVID-19 lung lesions by expert radiologists become expensive\nand prone to error, which raises an urgent quest to develop practical\nautonomous solutions. In this context, first, the paper introduces an open\naccess COVID-19 CT segmentation dataset containing 433 CT images from 82\npatients that have been annotated by an expert radiologist. Second, a Deep\nNeural Network (DNN)-based framework is proposed, referred to as the\nCOVID-Rate, that autonomously segments lung abnormalities associated with\nCOVID-19 from chest CT scans. Performance of the proposed COVID-Rate framework\nis evaluated through several experiments based on the introduced and external\ndatasets. The results show a dice score of 0:802 and specificity and\nsensitivity of 0:997 and 0:832, respectively. Furthermore, the results indicate\nthat the COVID-Rate model can efficiently segment COVID-19 lesions in both 2D\nCT images and whole lung volumes. Results on the external dataset illustrate\ngeneralization capabilities of the COVID-Rate model to CT images obtained from\na different scanner.\n"}
{"abstract": "  Scepticism towards childhood vaccines and genetically modified food has grown\ndespite scientific evidence of their safety. Beliefs about scientific issues\nare difficult to change because they are entrenched within many related moral\nconcerns and beliefs about what others think. We propose a cognitive network\nmodel which estimates the relationships, dissonance, and randomness between all\nrelated beliefs to derive predictions of the circumstances under which beliefs\nchange. Using a probabilistic nationally representative longitudinal study, we\nfound support for our model's predictions: Randomness of the belief networks\ndecreased over time, for many participants their estimated dissonance related\npositively to their self-reported dissonance, and individuals who had high\nestimated dissonance of their belief network were more likely to change their\nbeliefs to reduce this dissonance. This study is the first to combine a\nunifying predictive model with an experimental intervention and sheds light on\ndynamics of dissonance reduction leading to belief change.\n"}
{"abstract": "  The excitation of the filamentary gas structures surrounding giant elliptical\ngalaxies at the center of cool-core clusters, a.k.a BCGs (brightest cluster\ngalaxies), is key to our understanding of active galactic nucleus feedback, and\nof the impact of environmental and local effects on star formation. We\ninvestigate the contribution of the thermal radiation from the cooling flow\nsurrounding BCGs to the excitation of the filaments. We explore the effects of\nsmall levels of extra-heating (turbulence), and of metallicity, on the optical\nand infrared lines. Using the Cloudy code, we model the photoionization and\nphotodissociation of a slab of gas of optical depth AV{\\leq}30mag at constant\npressure, in order to calculate self-consistently all of the gas phases, from\nionized gas to molecular gas. The ionizing source is the EUV and soft X-ray\nradiation emitted by the cooling gas. We test these models comparing their\npredictions to the rich multi-wavelength observations, from optical to\nsubmillimeter. These models reproduce most of the multi-wavelength spectra\nobserved in the nebulae surrounding the BCGs, not only the LINER-like optical\ndiagnostics: [O iii]{\\lambda} 5007 {\\AA}/H\\b{eta}, [N ii]{\\lambda} 6583\n{\\AA}/H{\\alpha} and ([S ii]{\\lambda} 6716 {\\AA}+[S ii]{\\lambda} 6731\n{\\AA})/H{\\alpha} but also the infrared emission lines from the atomic gas. The\nmodeled ro-vib H2 lines also match observations, which indicates that near and\nmid-IR H2 lines are mostly excited by collisions between H2 molecules and\nsecondary electrons produced naturally inside the cloud by the interaction\nbetween the X-rays and the cold gas in the filament. However, there is still\nsome tension between ionized and molecular line tracers (i.e. CO), which\nrequires to optimize the cloud structure and the density of the molecular zone.\n"}
{"abstract": "  Considering the energy-efficient emergency response, subject to a given set\nof constraints on emergency communication networks (ECN), this article proposes\na hybrid device-to-device (D2D) and device-to-vehicle (D2V) network for\ncollecting and transmitting emergency information. First, we establish the D2D\nnetwork from the perspective of complex networks by jointly determining the\noptimal network partition (ONP) and the temporary data caching centers (TDCC),\nand thus emergency data can be forwarded and cached in TDCCs. Second, based on\nthe distribution of TDCCs, the D2V network is established by unmanned aerial\nvehicles (UAV)-based waypoint and motion planning, which saves the time for\nwireless transmission and aerial moving. Finally, the amount of time for\nemergency response and the total energy consumption are simultaneously\nminimized by a multiobjective evolutionary algorithm based on decomposition\n(MOEA/D), subject to a given set of minimum signal-to-interference-plus-noise\nratio (SINR), number of UAVs, transmit power, and energy constraints.\nSimulation results show that the proposed method significantly improves\nresponse efficiency and reasonably controls the energy, thus overcoming\nlimitations of existing ECNs. Therefore, this network effectively solves the\nkey problem in the rescue system and makes great contributions to post-disaster\ndecision-making.\n"}
{"abstract": "  In this paper, we deal with (angular cut-off) Boltzmann equation with soft\npotential ($-3<\\gamma<0$). In particular, we construct a unique global solution\nin $L^\\infty_{x,v}$ which converges to global equilibrium asymptotically\nprovided that initial data has a large amplitude but with sufficiently small\nrelative entropy. Because frequency multiplier is not uniformly positive\nanymore, unlike hard potential case, time-involved velocity weight will be used\nto derive sub-exponential decay of the solution. Motivated by recent\ndevelopment of $L^2\\mbox{-}L^\\infty$ approach also, we introduce some modified\nestimates of quadratic nonlinear terms. Linearized collision kernel will be\ntreated in a subtle manner to control singularity of soft potential kernel.\n"}
{"abstract": "  The lack of explainability of a decision from an Artificial Intelligence (AI)\nbased \"black box\" system/model, despite its superiority in many real-world\napplications, is a key stumbling block for adopting AI in many high stakes\napplications of different domain or industry. While many popular Explainable\nArtificial Intelligence (XAI) methods or approaches are available to facilitate\na human-friendly explanation of the decision, each has its own merits and\ndemerits, with a plethora of open challenges. We demonstrate popular XAI\nmethods with a mutual case study/task (i.e., credit default prediction),\nanalyze for competitive advantages from multiple perspectives (e.g., local,\nglobal), provide meaningful insight on quantifying explainability, and\nrecommend paths towards responsible or human-centered AI using XAI as a medium.\nPractitioners can use this work as a catalog to understand, compare, and\ncorrelate competitive advantages of popular XAI methods. In addition, this\nsurvey elicits future research directions towards responsible or human-centric\nAI systems, which is crucial to adopt AI in high stakes applications.\n"}
{"abstract": "  Coronal holes are well accepted to be source regions of the fast solar wind.\nAs one of the common structures in coronal holes, coronal plumes might\ncontribute to the origin of the nascent solar wind. To estimate the\ncontribution of coronal plumes to the nascent solar wind, we make the first\nattempt to estimate their populations in solar polar coronal holes. By\ncomparing the observations viewed from two different angles taken by the twin\nsatellites of STEREO and the results of Monte Carlo simulations, we estimate\nabout 16--27 plumes rooted in an area of $4\\times10^4$ arcsec$^2$ of the polar\ncoronal holes near the solar minimum, which occupy about 2--3.4% of the area.\nBased on these values, the contribution of coronal plumes to the nascent solar\nwind has also been discussed. A further investigation indicates that more\nprecise number of coronal plumes can be worked out with observations from three\nor more viewing angles.\n"}
{"abstract": "  Borrowing from the transformer models that revolutionized the field of\nnatural language processing, self-supervised feature learning for visual tasks\nhas also seen state-of-the-art success using these extremely deep, isotropic\nnetworks. However, the typical AI researcher does not have the resources to\nevaluate, let alone train, a model with several billion parameters and\nquadratic self-attention activations. To facilitate further research, it is\nnecessary to understand the features of these huge transformer models that can\nbe adequately studied by the typical researcher. One interesting characteristic\nof these transformer models is that they remove most of the inductive biases\npresent in classical convolutional networks. In this work, we analyze the\neffect of these and more inductive biases on small to moderately-sized\nisotropic networks used for unsupervised visual feature learning and show that\ntheir removal is not always ideal.\n"}
{"abstract": "  App builders commonly use security challenges, a form of step-up\nauthentication, to add security to their apps. However, the ethical\nimplications of this type of architecture has not been studied previously. In\nthis paper, we present a large-scale measurement study of running an existing\nanti-fraud security challenge, Boxer, in real apps running on mobile devices.\nWe find that although Boxer does work well overall, it is unable to scan\neffectively on devices that run its machine learning models at less than one\nframe per second (FPS), blocking users who use inexpensive devices. With the\ninsights from our study, we design Daredevil, anew anti-fraud system for\nscanning payment cards that work swell across the broad range of performance\ncharacteristics and hardware configurations found on modern mobile devices.\nDaredevil reduces the number of devices that run at less than one FPS by an\norder of magnitude compared to Boxer, providing a more equitable system for\nfighting fraud. In total, we collect data from 5,085,444 real devices spread\nacross 496 real apps running production software and interacting with real\nusers.\n"}
{"abstract": "  A spherical conical metric $g$ on a surface $\\Sigma$ is a metric of constant\ncurvature $1$ with finitely many isolated conical singularities. The\nuniformization problem for such metrics remains largely open when at least one\nof the cone angles exceeds $2\\pi$. The eigenfunctions of the Friedrichs\nLaplacian $\\Delta_g$ with eigenvalue $\\lambda=2$ play a special role in this\nproblem, as they represent local obstructions to deformations of the metric $g$\nin the class of spherical conical metrics. In the present paper we apply the\ntheory of multivalued harmonic maps to spheres to the question of existence of\nsuch eigenfunctions. In the first part we establish a new criterion for the\nexistence of $2$-eigenfunctions, given in terms of a certain meromorphic data\non $\\Sigma$. As an application we give a description of all $2$-eigenfunctions\nfor metrics on the sphere with at most three conical singularities. The second\npart is an algebraic construction of metrics with large number of\n$2$-eigenfunctions via the deformation of multivalued harmonic maps. We provide\nnew explicit examples of metrics with many $2$-eigenfunctions via both\napproaches, and describe the general algorithm to find metrics with arbitrarily\nlarge number of $2$-eigenfunctions.\n"}
{"abstract": "  State estimation problems without absolute position measurements routinely\narise in navigation of unmanned aerial vehicles, autonomous ground vehicles,\netc., whose proper operation relies on accurate state estimates and reliable\ncovariances. Unaware of absolute positions, these problems have immanent\nunobservable directions. Traditional causal estimators, however, usually gain\nspurious information on the unobservable directions, leading to over-confident\ncovariance inconsistent with actual estimator errors. The consistency problem\nof fixed-lag smoothers (FLSs) has only been attacked by the first estimate\nJacobian (FEJ) technique because of the complexity to analyze their\nobservability property. But the FEJ has several drawbacks hampering its wide\nadoption. To ensure the consistency of a FLS, this paper introduces the right\ninvariant error formulation into the FLS framework. To our knowledge, we are\nthe first to analyze the observability of a FLS with the right invariant error.\nOur main contributions are twofold. As the first novelty, to bypass the\ncomplexity of analysis with the classic observability matrix, we show that\nobservability analysis of FLSs can be done equivalently on the linearized\nsystem. Second, we prove that the inconsistency issue in the traditional FLS\ncan be elegantly solved by the right invariant error formulation without\nartificially correcting Jacobians. By applying the proposed FLS to the\nmonocular visual inertial simultaneous localization and mapping (SLAM) problem,\nwe confirm that the method consistently estimates covariance similarly to a\nbatch smoother in simulation and that our method achieved comparable accuracy\nas traditional FLSs on real data.\n"}
{"abstract": "  This work introduces a concept of explanations with respect to the violation\nof safe behaviours within software defined networks (SDNs) expressible in\nNetKAT. The latter is a network programming language based on a well-studied\nmathematical structure, namely, Kleene Algebra with Tests (KAT). Amongst\nothers, the mathematical foundation of NetKAT gave rise to a sound and complete\nequational theory. In our setting, a safe behaviour is characterised by a\nNetKAT policy, or program, which does not enable forwarding packets from an\ningress i to an undesirable egress e. We show how explanations for safety\nviolations can be derived in an equational fashion, according to a modification\nof the existing NetKAT axiomatisation. We propose an approach based on the\nMaude system for actually computing the undesired behaviours witnessing the\nforwarding of packets from i to e as above. SDN-SafeCheck is a tool based on\nMaude equational theories satisfying important properties such as Church-Rosser\nand termination. SDN-SafeCheck automatically identifies all the undesired\nbehaviours leading to e, covering forwarding paths up to a user specified size.\n"}
{"abstract": "  The Standard Model Neutrino Effective Field Theory (SMNEFT) is the Standard\nModel Effective Field Theory (SMEFT) augmented with right-handed neutrinos.\nBuilding on our previous work, arXiv:2010.12109, we calculate the Yukawa\ncoupling contributions to the one-loop anomalous dimension matrix for the 11\ndimension-six four-fermion SMNEFT operators. We also present the new\ncontributions to the anomalous dimension matrix for the 14 four-fermion SMEFT\noperators that mix with the SMNEFT operators through the Yukawa couplings of\nthe right-handed neutrinos.\n"}
{"abstract": "  We present a spinning black hole solution in $d$ dimensions with a maximal\nnumber of rotation parameters in the context of the Einstein-Maxwell-Dilaton\ntheory. An interesting feature of such a solution is that it accommodates\nLifshitz black holes when the rotation parameters are set to zero. We verify\nthe rotating nature of the black hole solution by performing the quasi-local\nanalysis of conserved charges and defining the corresponding angular momenta.\nIn addition, we perform the thermodynamical analysis of the black hole\nconfiguration, show that the first law of thermodynamics is completely\nconsistent, and obtain a Smarr-like formula. We further study the thermodynamic\nstability of the constructed solution from a local viewpoint, by computing the\nassociated specific heats, and from a global perspective, by using the\nso-called new thermodynamic geometry. We finally make some comments related to\na pathology found in the causal structure of the obtained rotating black hole\nspacetime and compute some of its curvature invariants.\n"}
{"abstract": "  In this paper we address the large-scale regularity theory for the stationary\nNavier-Stokes equations in highly oscillating bumpy John domains. These domains\nare very rough, possibly with fractals or cusps, at the microscopic scale, but\nare amenable to the mathematical analysis of the Navier-Stokes equations. We\nprove: (i) a large-scale Calder\\'on-Zygmund estimate, (ii) a large-scale\nLipschitz estimate, (iii) large-scale higher-order regularity estimates,\nnamely, $C^{1,\\gamma}$ and $C^{2,\\gamma}$ estimates. These nice regularity\nresults are inherited only at mesoscopic scales, and clearly fail in general at\nthe microscopic scales. We emphasize that the large-scale $C^{1,\\gamma}$\nregularity is obtained by using first-order boundary layers constructed via a\nnew argument. The large-scale $C^{2,\\gamma}$ regularity relies on the\nconstruction of second-order boundary layers, which allows for certain boundary\ndata with linear growth at spatial infinity. To the best of our knowledge, our\nwork is the first to carry out such an analysis. In the wake of many works in\nquantitative homogenization, our results strongly advocate in favor of\nconsidering the boundary regularity of the solutions to fluid equations as a\nmultiscale problem, with improved regularity at or above a certain scale.\n"}
{"abstract": "  Sparse Bayesian learning (SBL) can be implemented with low complexity based\non the approximate message passing (AMP) algorithm. However, it does not work\nwell for a generic measurement matrix, which may cause AMP to diverge. Damped\nAMP has been used for SBL to alleviate the problem at the cost of reducing\nconvergence speed. In this work, we propose a new SBL algorithm based on\nstructured variational inference, leveraging AMP with a unitary transformation\n(UAMP). Both single measurement vector and multiple measurement vector problems\nare investigated. It is shown that, compared to state-of-the-art AMP-based SBL\nalgorithms, the proposed UAMP-SBL is more robust and efficient, leading to\nremarkably better performance.\n"}
{"abstract": "  Charge order has recently been identified as a leading competitor of\nhigh-temperature superconductivity in moderately doped cuprates. We provide a\nsurvey of universal and materials-specific aspects of this phenomenon, with\nemphasis on results obtained by scattering methods. In particular, we discuss\nthe structure, periodicity, and stability range of the charge-ordered state,\nits response to various external perturbations, the influence of disorder, the\ncoexistence and competition with superconductivity, as well as collective\ncharge dynamics. In the context of this journal issue which honors Roger\nCowley's legacy, we also discuss the connection of charge ordering with lattice\nvibrations and the central-peak phenomenon. We end the review with an outlook\non research opportunities offered by new synthesis methods and experimental\nplatforms, including cuprate thin films and superlattices.\n"}
{"abstract": "  Neural networks are a prominent tool for identifying and modeling complex\npatterns, which are otherwise hard to detect and analyze. While machine\nlearning and neural networks have been finding applications across many areas\nof science and technology, their use in decoding ultrafast dynamics of quantum\nsystems driven by strong laser fields has been limited so far. Here we use deep\nneural networks to analyze simulated noisy spectra of highly nonlinear optical\nresponse of a 2-dimensional gapped graphene crystal to intense few-cycle laser\npulses. We show that a computationally simple 1-dimensional system provides a\nuseful \"nursery school\" for our neural network, allowing it to be easily\nretrained to treat more complex systems, recovering the band structure and\nspectral phases of the incident few-cycle pulse with high accuracy, in spite of\nsignificant amplitude noise and phase jitter. Our results both offer a new tool\nfor attosecond spectroscopy of quantum dynamics in solids and also open a route\nto developing all-solid-state devices for complete characterization of\nfew-cycle pulses, including their nonlinear chirp and the carrier envelope\nphase.\n"}
{"abstract": "  We probe the effect of diffeomorphism symmetry on the critical exponents\nvalues for massive O($N$) $\\lambda\\phi^{4}$ scalar field theories in curved\nspacetime. We apply field-theoretic renormalization group tools, where we use\nonly momentum coordinates as opposed to the earlier standard procedure with\nboth space and momentum coordinates, for computing analytically the critical\nexponents up to three-loop order for one of them and up to two-loop level for\nthe remaining ones. We found that the curved spacetime critical exponents are\nthe same as their flat spacetime counterparts, at least up to the level\nconsidered. At the conclusions we present the physical interpretation for the\nresults.\n"}
{"abstract": "  A set $D$ of vertices in $G$ is a disjunctive dominating set in $G$ if every\nvertex not in $D$ is adjacent to a vertex of $D$ or has at least two vertices\nin $D$ at distance $2$ from it in $G$. The disjunctive domination number,\n$\\gamma^{d}_2(G)$, of $G$ is the minimum cardinality of a disjunctive\ndominating set in $G$. In this paper, we show that if $G$ be a graph of order\nat least $3$, $\\delta(G)\\geq 2$ and with no component isomorphic to any of\neight forbidden graphs, then $\\gamma^{d}_2(G)\\leq \\frac{|G|}{3}$. Moreover, we\nprovide an infinite family of graphs attaining this bound. In addition, we also\nstudy the case that $G$ is a claw-free graph with minimum degree at least two.\n"}
{"abstract": "  A growing interest in designing high-alumina MgO-bonded refractory castables\nhas been identified in recent years due to the magnesia ability to react: (i)\nwith water at the initial processing stages of these materials (inducing the\nprecipitation of brucite phase) or (ii) with alumina, giving rise to in situ\nMgAl2O4 generation at high temperatures. Nevertheless, despite the great\npotential of caustic magnesia to be used as a binder in such systems due to its\nhigh reactivity, it is still a challenge to control the hydration reaction rate\nof this oxide and the negative effects derived from the expansive feature of\nMg(OH)2 formation. Thus, this work evaluated the incorporation of different\ncontents of aluminum hydroxyl lactate (AHL) into caustic magnesia-bonded\ncastables, aiming to control the brucite precipitation during the curing and\ndrying steps of the prepared samples, resulting in crack-free refractories. The\ndesigned compositions were characterized via flowability, setting behavior,\nX-ray diffraction, cold flexural strength, porosity, permeability and\nthermogravimetric measurements. According to the results, instead of Mg(OH)2,\nhydrotalcite-like phases [Mg6Al2(OH)16(OH)2.4.5H2O and Mg6Al2(OH)16(CO3).4H2O]\nwere the main hydrated phases identified in the AHL-containing compositions.\nThe addition of 1.0 wt.% of aluminum hydroxyl lactate to the designed castable\nproved to be, so far, the best option for this magnesia source, resulting in\nthe development of a crack-free refractory with enhanced properties and greater\nspalling resistance under heating.\n"}
{"abstract": "  Rapid miniaturization of electronic devices and circuits demands profound\nunderstanding of fluctuation phenomena at the nanoscale. Superconducting\nnanowires -- serving as important building blocks for such devices -- may\nseriously suffer from fluctuations which tend to destroy long-range order and\nsuppress superconductivity. In particular, quantum phase slips (QPS)\nproliferating at low temperatures may turn a quasi-one-dimensional\nsuperconductor into a resistor or an insulator. Here, we introduce a physical\nconcept of QPS-controlled localization of Cooper pairs that may occur even in\nuniform nanowires without any dielectric barriers being a fundamental\nmanifestation of the flux-charge duality in superconductors. We demonstrate --\nboth experimentally and theoretically -- that deep in the \"insulating\" state\nsuch nanowires actually exhibit non-trivial superposition of superconductivity\nand weak Coulomb blockade of Cooper pairs generated by quantum tunneling of\nmagnetic fluxons across the wire.\n"}
{"abstract": "  Up to date, only lower and upper bounds for the optimal configuration of a\nSquare Array (A2) Group Testing (GT) algorithm are known. We establish exact\nanalytical formulae and provide a couple of applications of our result. First,\nwe compare the A2 GT scheme to several other classical GT schemes in terms of\nthe gain per specimen attained at optimal configuration. Second, operating\nunder objective Bayesian framework with the loss designed to attain minimum at\noptimal GT configuration, we suggest the preferred choice of the group size\nunder natural minimal assumptions: the prior information regarding the\nprevalence suggests that grouping and application of A2 is better than\nindividual testing. The same suggestion is provided for the Minimax strategy.\n"}
{"abstract": "  Let $ \\mathcal{H}(\\mathbb{D}) $ be the linear space of analytic functions on\nthe unit disk $ \\mathbb{D}=\\{z\\in\\mathbb{C}: |z|<1\\} $ and let $\n\\mathcal{B}=\\{w\\in \\mathcal{H}(\\mathbb{D}: |w(z)|<1)\\} $. The classical Bohr's\ninequality states that if a power series $ f(z)=\\sum_{n=0}^{\\infty}a_nz^n $\nconverges in $ \\mathbb{D} $ and $ |f(z)|<1 $ for $ z\\in\\mathbb{D} $, then\n\\begin{equation*}\n  \\sum_{n=0}^{\\infty}|a_n|r^n\\leq 1\\;\\;\\mbox{for}\\;\\; r\\leq \\frac{1}{3}\n\\end{equation*} and the constant $ 1/3 $ is the best possible. The constant $\n1/3 $ is known as Bohr radius. A function $ f : \\mathbb{D}\\rightarrow\\mathbb{C}\n$ is said to be log-harmonic if there is a $ w\\in\\mathcal{B} $ such that $ f $\nis a non-constant solution of the non-linear elliptic partial differential\nequation \\begin{equation*}\n  \\bar{f}_{\\bar{z}}(z)/\\bar{f}(z)=w(z)f_{z}(z)/f(z). \\end{equation*} The class\nof log-harmonic mappings is denoted by $ \\mathcal{S}_{LH} $. The set of all\nstarlike log-harmonic mapping is defined by \\begin{equation*}\n\\mathcal{ST}_{LH}=\\bigg\\{f\\in\\mathcal{S}_{LH}:\\frac{\\partial}{\\partial\\theta}{\\rm\nArg}(f(e^{i\\theta}))={\\rm\nRe}\\left(\\frac{zf_{z}-\\bar{z}f_{\\bar{z}}}{f}\\right)>0\\;\\; \\mbox{in}\\;\\;\n\\mathbb{D}\\bigg\\}. \\end{equation*} In this paper, we study several improved\nBohr radius for the class $ \\mathcal{ST}^{0}_{LH} $, a subclass of $\n\\mathcal{ST}_{LH} $, consisting of functions $ f\\in\\mathcal{ST}_{LH} $ which\nmap the unit disk $ \\mathbb{D} $ onto a starlike domain (with respect to the\norigin).\n"}
{"abstract": "  Jellyfish (Medusozoa) have been deemed the most energy-efficient animals in\nthe world. Their bell morphology and relatively simple nervous systems make\nthem attractive to robotocists. Although, the science community has devoted\nmuch attention to understanding their swimming performance, there is still much\nto be learned about the jet propulsive locomotive gait displayed by prolate\njellyfish. Traditionally, computational scientists have assumed uniform duty\ncycle kinematics when computationally modeling jellyfish locomotion. In this\nstudy we used fluid-structure interaction modeling to determine possible\nenhancements in performance from shuffling different duty cycles together\nacross multiple Reynolds numbers and contraction frequencies. Increases in\nspeed and reductions in cost of transport were observed as high as 80% and 50%,\nrespectively. Generally, the net effects were greater for cases involving lower\ncontraction frequencies. Overall, robust duty cycle combinations were\ndetermined that led to enhanced or impeded performance.\n"}
{"abstract": "  The Turing Award is recognized as the most influential and prestigious award\nin the field of computer science(CS). With the rise of the science of science\n(SciSci), a large amount of bibliographic data has been analyzed in an attempt\nto understand the hidden mechanism of scientific evolution. These include the\nanalysis of the Nobel Prize, including physics, chemistry, medicine, etc. In\nthis article, we extract and analyze the data of 72 Turing Award laureates from\nthe complete bibliographic data, fill the gap in the lack of Turing Award\nanalysis, and discover the development characteristics of computer science as\nan independent discipline. First, we show most Turing Award laureates have\nlong-term and high-quality educational backgrounds, and more than 61% of them\nhave a degree in mathematics, which indicates that mathematics has played a\nsignificant role in the development of computer science. Secondly, the data\nshows that not all scholars have high productivity and high h-index; that is,\nthe number of publications and h-index is not the leading indicator for\nevaluating the Turing Award. Third, the average age of awardees has increased\nfrom 40 to around 70 in recent years. This may be because new breakthroughs\ntake longer, and some new technologies need time to prove their influence.\nBesides, we have also found that in the past ten years, international\ncollaboration has experienced explosive growth, showing a new paradigm in the\nform of collaboration. It is also worth noting that in recent years, the\nemergence of female winners has also been eye-catching. Finally, by analyzing\nthe personal publication records, we find that many people are more likely to\npublish high-impact articles during their high-yield periods.\n"}
{"abstract": "  The Jacobian group (also known as the critical group or sandpile group) is an\nimportant invariant of a finite, connected graph $X$; it is a finite abelian\ngroup whose cardinality is equal to the number of spanning trees of $X$\n(Kirchhoff's Matrix Tree Theorem). A specific type of covering graph, called a\nderived graph, that is constructed from a voltage graph with voltage group $G$\nis the object of interest in this paper. Towers of derived graphs are studied\nby using aspects of classical Iwasawa Theory (from number theory). Formulas for\nthe orders of the Sylow $p$-subgroups of Jacobians in an infinite voltage\n$p$-tower, for any prime $p$, are obtained in terms of classical $\\mu$ and\n$\\lambda$ invariants by using the decomposition of a finitely generated module\nover the Iwasawa Algebra.\n"}
{"abstract": "  We propose an experiment that the entanglement between two macroscopic\nmirrors suspended at the end of an equal-arm interferometer is destroyed by the\nnoise of gravitons through bremsstrahlung. By calculating the correlation\nfunction of the noise, we obtain the decoherence time from the decoherence\nfunctional. We estimate that the decoherence time induced by the noise of\ngravitons in squeezed states stemming from inflation is approximately 20\nseconds for 40 km long arms and 40 kg mirrors. Our analysis shows that\nobservation of the decoherence time of quantum entanglement has the potential\nto detect gravitons indirectly. This indirect detection of gravitons would give\nstrong evidence of quantum gravity.\n"}
{"abstract": "  In general, graph representation learning methods assume that the train and\ntest data come from the same distribution. In this work we consider an\nunderexplored area of an otherwise rapidly developing field of graph\nrepresentation learning: The task of out-of-distribution (OOD) graph\nclassification, where train and test data have different distributions, with\ntest data unavailable during training. Our work shows it is possible to use a\ncausal model to learn approximately invariant representations that better\nextrapolate between train and test data. Finally, we conclude with synthetic\nand real-world dataset experiments showcasing the benefits of representations\nthat are invariant to train/test distribution shifts.\n"}
{"abstract": "  A low noise linearly frequency modulated continuous wave laser source with a\nwide frequency bandwidth is demonstrated. By two-dimensional thermal tuning,\nthe laser source shows 42 GHz continuous frequency tuning with 49.86 Hz\nintrinsic linewidth under static operation. For dynamical FMCW, the laser\nsource has 10.25 GHz frequency bandwidth at 100 Hz chirped frequency and 5.56\nGHz at 1 kHz chirped frequency. With pre-distortion linearization, it can\ndistinguish 3 m length difference at 45 km distance in the fibre length\nmeasured experiment, which demonstrate a potential for application in\nultra-long fibre sensing and FMCW LiDAR.\n"}
{"abstract": "  The temporal coincidences of events detected in four neutrino detectors and\ntwo gravitational antennas still remains among the most puzzling phenomena\nassociated with SN1987A. The coincidences form a six-hour signal approximately\ncoincident in time with the well-known LSD signal at 2h52m UT on 23/02/1987.\nAfter 30 years of research, the characteristics and the shape of the six-hour\nsignal have been studied quite well, but the mechanisms of its formation have\nnot been fully understood as of yet. Here we suggest that data obtained from\nanother technology, radioactive decays, might provide new insights into the\norigin of signals previously seen in neutrino detectors and gravity wave\ndetectors. On August 17, 2017, at 12h41m UT, the GW170817 signal was detected\nby LIGO and Virgo. At the same time, an approximately 7-hour long signal\ncoincident with GW170817 was detected in the Si/Cl experiment on precision\nmeasurement of the $^{32}$Si half-life. We show that the Si/Cl signal is\nunexpectedly similar to the six-hour signal from SN1987A. In addition, we\nestablish that the sources of the coinciding events are similar to those of the\nSi/Cl signal. To explain the surprising similarities in both signals, we\npresent a mechanism which could in principle account for this phenomenon in\nterms of a local increase in the density of axionic dark matter induced by a\ngravity wave.\n"}
{"abstract": "  A major difficulty in debugging distributed systems lies in manually\ndetermining which of the many available debugging tools to use and how to query\nits logs. Our own study of a production debugging workflow confirms the\nmagnitude of this burden. This paper explores whether a machine-learning model\ncan assist developers in distributed systems debugging. We present Revelio, a\ndebugging assistant which takes user reports and system logs as input, and\noutputs debugging queries that developers can use to find a bug's root cause.\nThe key challenges lie in (1) combining inputs of different types (e.g.,\nnatural language reports and quantitative logs) and (2) generalizing to unseen\nfaults. Revelio addresses these by employing deep neural networks to uniformly\nembed diverse input sources and potential queries into a high-dimensional\nvector space. In addition, it exploits observations from production systems to\nfactorize query generation into two computationally and statistically simpler\nlearning tasks. To evaluate Revelio, we built a testbed with multiple\ndistributed applications and debugging tools. By injecting faults and training\non logs and reports from 800 Mechanical Turkers, we show that Revelio includes\nthe most helpful query in its predicted list of top-3 relevant queries 96% of\nthe time. Our developer study confirms the utility of Revelio.\n"}
{"abstract": "  References are an essential part of Wikipedia. Each statement in Wikipedia\nshould be referenced. In this paper, we explore the creation and collection of\nreferences for new Wikipedia articles from an editors' perspective. We map out\nthe workflow of editors when creating a new article, emphasising how they\nselect references.\n"}
{"abstract": "  We continue the study of partition functions of 5d supersymmetric theories on\nmanifolds taking the form of a twisted product $\\mathcal{M}_3\\times\n\\Sigma_{\\mathfrak{g}}$ with $\\Sigma_{\\mathfrak{g}}$ denoting a Riemann surface\nof genus $\\mathfrak{g}$. The 5d theory compactified on $\\Sigma_{\\mathfrak{g}}$\nleads to a novel class of 3d theories in IR, whose existence at large $N$ is\nexpected from holography. Focussing on $\\mathcal{M}_3$ being $S^2\\times S^1$\nwithout or with a topological twist on the 2-sphere leads to the superconformal\nindex or topologically twisted index, respectively, for such a class of 3d\ntheories. We discuss the large $N$ limit of these partition functions and find\nnew relations between them and other well-known 5d partition functions, with\ninteresting consequences for the 3d indices.\n"}
{"abstract": "  We derive novel results on the ergodic theory of irreducible, aperiodic\nMarkov chains. We show how to optimally steer the network flow to a stationary\ndistribution over a finite or infinite time horizon. Optimality is with respect\nto an entropic distance between distributions on feasible paths. When the prior\nis reversible, it shown that solutions to this discrete time and space steering\nproblem are reversible as well. A notion of temperature is defined for\nBoltzmann distributions on networks, and problems analogous to cooling (in this\ncase, for evolutions in discrete space and time) are discussed.\n"}
{"abstract": "  Temperature-dependent reflectivity measurements on the kagome metal\nCsV$_3$Sb$_5$ in a broad frequency range of $50-20000$ cm$^{-1}$ down to $T$=10\nK are reported. The charge-density wave (CDW) formed below $T_{\\rm CDW}$ = 94 K\nmanifests itself in a prominent spectral-weight transfer from low to higher\nenergy regions. The CDW gap of 60-75 meV is observed at the lowest temperature\nand shows significant deviations from an isotropic BCS-type mean-field\nbehavior. Absorption peaks appear at frequencies as low as 200 cm$^{-1}$ and\ncan be identified with interband transitions according to density-functional\ncalculations. The change in the interband absorption compared to KV$_3$Sb$_5$\nreflects the inversion of band saddle points between the K and Cs compounds.\nAdditionally, a broader and strongly temperature-dependent absorption feature\nis observed below 1000 cm$^{-1}$ and assigned to a displaced Drude peak. It\nreflects localization effects on charge carriers.\n"}
{"abstract": "  False Data Injection (FDI) attacks against powersystem state estimation are a\ngrowing concern for operators.Previously, most works on FDI attacks have been\nperformedunder the assumption of the attacker having full knowledge ofthe\nunderlying system without clear justification. In this paper, wedevelop a\ntopology-learning-aided FDI attack that allows stealthycyber-attacks against AC\npower system state estimation withoutprior knowledge of system information. The\nattack combinestopology learning technique, based only on branch and bus\npowerflows, and attacker-side pseudo-residual assessment to performstealthy FDI\nattacks with high confidence. This paper, for thefirst time, demonstrates how\nquickly the attacker can developfull-knowledge of the grid topology and\nparameters and validatesthe full knowledge assumptions in the previous work.\n"}
{"abstract": "  A rapidly growing bubble close to a free surface induces jetting: a central\njet protruding outwards and a crown surrounding it at later stages. While the\nformation mechanism of the central jet is known and documented, that of the\ncrown remains unsettled. We perform axisymmetric simulations of the problem\nusing the free software program basilisk, where a finite-volume compressible\nsolver has been implemented, that uses a geometric Volume-of-Fluid method (VoF)\nfor the tracking of the interface. We show that the mechanism of crown\nformation is a combination of a pressure distortion over the curved interface,\ninducing flow focusing, and of a flow reversal, caused by the second expansion\nof the toroidal bubble that drives the crown. The work culminates in a\nparametric study with the Weber number, the Reynolds number, the pressure ratio\nand the dimensionless bubble distance to the free surface as control\nparameters. Their effects on both the central jet and the crown are explored.\nFor high Weber numbers, we observe the formation of weaker \"secondary crowns\",\nhighly correlated with the third oscillation cycle of the bubble.\n"}
{"abstract": "  We prove that finitely generated virtually free groups are stable in\npermutations. As an application, we show that almost-periodic\nalmost-automorphisms of labelled graphs are close to periodic automorphisms.\n"}
{"abstract": "  Labeling data for modern machine learning is expensive and time-consuming.\nLatent variable models can be used to infer labels from weaker,\neasier-to-acquire sources operating on unlabeled data. Such models can also be\ntrained using labeled data, presenting a key question: should a user invest in\nfew labeled or many unlabeled points? We answer this via a framework centered\non model misspecification in method-of-moments latent variable estimation. Our\ncore result is a bias-variance decomposition of the generalization error, which\nshows that the unlabeled-only approach incurs additional bias under\nmisspecification. We then introduce a correction that provably removes this\nbias in certain cases. We apply our decomposition framework to three scenarios\n-- well-specified, misspecified, and corrected models -- to 1) choose between\nlabeled and unlabeled data and 2) learn from their combination. We observe\ntheoretically and with synthetic experiments that for well-specified models,\nlabeled points are worth a constant factor more than unlabeled points. With\nmisspecification, however, their relative value is higher due to the additional\nbias but can be reduced with correction. We also apply our approach to study\nreal-world weak supervision techniques for dataset construction.\n"}
{"abstract": "  Having timely and fresh knowledge about the current state of information\nsources is critical in a variety of applications. In particular, a status\nupdate may arrive at the destination later than its generation time due to\nprocessing and communication delays. The freshness of the status update at the\ndestination is captured by the notion of age of information. In this study, we\nanalyze a multiple sensing network with multiple sources, multiple servers, and\na monitor (destination). Each source corresponds to an independent piece of\ninformation and its age is measured individually. Given a particular source,\nthe servers independently sense the source of information and send the status\nupdate to the monitor. We assume that updates arrive at the servers according\nto Poisson random processes. Each server sends its updates to the monitor\nthrough a direct link, which is modeled as a queue. The service time to\ntransmit an update is considered to be an exponential random variable. We\nexamine both homogeneous and heterogeneous service and arrival rates for the\nsingle-source case, and only homogeneous arrival and service rates for the\nmultiple-source case. We derive a closed-form expression for the average age of\ninformation under a last-come-first-serve (LCFS) queue for a single source and\narbitrary number of homogeneous servers. Using a recursive method, we derive\nthe explicit average age of information for any number of sources and\nhomogeneous servers. We also investigate heterogeneous servers and a single\nsource, and present algorithms for finding the average age of information.\n"}
{"abstract": "  We propose a new model to determine the ages and masses of red giant branch\n(RGB) stars from the low-resolution large sky area multi-object fiber\nspectroscopic telescope (LAMOST) spectra. The ages of RGB stars are difficult\nto determine using classical isochrone fitting techniques in the\nHertzsprung-Russell diagram, because isochrones of RGB stars are tightly\ncrowned. With the help of the asteroseismic method, we can determine the masses\nand ages of RGB stars accurately. Using the ages derived from the asteroseismic\nmethod, we train a deep learning model based on DenseNet to calculate the ages\nof RGB stars directly from their spectra. We then apply this model to determine\nthe ages of 512 272 RGB stars from LAMOST DR7 spectra (see\nhttp://dr7.lamost.org/). The results show that our model can estimate the ages\nof RGB stars from low-resolution spectra with an accuracy of 24.3%. The results\non the open clusters M 67, Berkeley 32, and NGC 2420 show that our model\nperforms well in estimating the ages of RGB stars. Through comparison, we find\nthat our method performs better than other methods in determining the ages of\nRGB stars. The proposed method can be used in the stellar parameter pipeline of\nupcoming large surveys such as 4MOST, WEAVES, and MOONS.\n"}
{"abstract": "  We present SICK-NL (read: signal), a dataset targeting Natural Language\nInference in Dutch. SICK-NL is obtained by translating the SICK dataset of\nMarelli et al. (2014)from English into Dutch. Having a parallel inference\ndataset allows us to compare both monolingual and multilingual NLP models for\nEnglish and Dutch on the two tasks. In the paper, we motivate and detail the\ntranslation process, perform a baseline evaluation on both the original SICK\ndataset and its Dutch incarnation SICK-NL, taking inspiration from Dutch\nskipgram embeddings and contextualised embedding models. In addition, we\nencapsulate two phenomena encountered in the translation to formulate stress\ntests and verify how well the Dutch models capture syntactic restructurings\nthat do not affect semantics. Our main finding is all models perform worse on\nSICK-NL than on SICK, indicating that the Dutch dataset is more challenging\nthan the English original. Results on the stress tests show that models don't\nfully capture word order freedom in Dutch, warranting future systematic\nstudies.\n"}
{"abstract": "  In preparation for the Dark Energy Spectroscopic Instrument (DESI), a new top\nend was installed on the Mayall 4-meter telescope at Kitt Peak National\nObservatory. The refurbished telescope and the DESI instrument were\nsuccessfully commissioned on sky between 2019 October and 2020 March. Here we\ndescribe the pointing, tracking and imaging performance of the Mayall telescope\nequipped with its new DESI prime focus corrector, as measured by six guider\ncameras sampling the outer edge of DESI's focal plane. Analyzing ~500,000\nguider images acquired during commissioning, we find a median delivered image\nFWHM of 1.1 arcseconds (in the r-band at 650 nm), with the distribution\nextending to a best-case value of ~0.6 arcseconds. The point spread function is\nwell characterized by a Moffat profile with a power-law index of $\\beta$ ~ 3.5\nand little dependence of $\\beta$ on FWHM. The shape and size of the PSF\ndelivered by the new corrector at a field angle of 1.57 degrees are very\nsimilar to those measured with the old Mayall corrector on axis. We also find\nthat the Mayall achieves excellent pointing accuracy (several arcseconds RMS)\nand minimal open-loop tracking drift (< 1 milliarcsecond per second),\nimprovements on the telecope's pre-DESI performance. In the future, employing\nDESI's active focus adjustment capabilities will likely further improve the\nMayall/DESI delivered image quality.\n"}
{"abstract": "  In the field of topology optimization, the homogenization approach has been\nrevived as an important alternative to the established, density-based methods\nbecause it can represent the microstructural design at a much finer\nlength-scale than the computational grid. The optimal microstructure for a\nsingle load case is an orthogonal rank-3 laminate. A rank-3 laminate can be\ndescribed in terms of frame fields, which are also an important tool for mesh\ngeneration in both 2D and 3D.\n  We propose a method for generating multi-laminar structures from frame\nfields. Rather than relying on integrative approaches that find a\nparametrization based on the frame field, we find stream surfaces, represented\nas point clouds aligned with frame vectors, and we solve an optimization\nproblem to find well-spaced collections of such stream surfaces. The stream\nsurface tracing is unaffected by the presence of singularities outside the\nregion of interest. Neither stream surface tracing nor selecting well-spaced\nsurface rely on combed frame fields.\n  In addition to stream surface tracing and selection, we provide two methods\nfor generating structures from stream surface collections. One of these methods\nproduces volumetric solids by summing basis functions associated with each\npoint of the stream surface collection. The other method reinterprets point\nsampled stream surfaces as a spatial twist continuum and produces a\nhexahedralization by dualizing a graph representing the structure.\n  We demonstrate our methods on several frame fields produced using the\nhomogenization approach for topology optimization, boundary-aligned, algebraic\nframe fields, and frame fields computed from closed-form expressions.\n"}
{"abstract": "  We present a logic named L_{LF} whose intended use is to formalize properties\nof specifications developed in the dependently typed lambda calculus LF. The\nlogic is parameterized by the LF signature that constitutes the specification.\nAtomic formulas correspond to typing derivations relative to this signature.\nThe logic includes a collection of propositional connectives and quantifiers.\nQuantification ranges over expressions that denote LF terms and LF contexts.\nQuantifiers of the first variety are qualified by simple types that describe\nthe functional structure associated with the variables they bind; deeper,\ndependency related properties are expressed by the body of the formula.\nContext-level quantifiers are qualified by context schemas that identify\npatterns of declarations out of which actual contexts may be constructed. The\nsemantics of variable-free atomic formulas is articulated via the derivability\nin LF of the judgements they encode. Propositional constants and connectives\nare understood in the usual manner and the meaning of quantifiers is explicated\nthrough substitutions of expressions that adhere to the type qualifications.\nThe logic is complemented by a proof system that enables reasoning that is\nsound with respect to the described semantics. The main novelties of the proof\nsystem are the provision for case-analysis style reasoning about LF judgements,\nsupport for inductive reasoning over the heights of LF derivations and the\nencoding of LF meta-theorems. The logic is motivated by the paradigmatic\nexample of type assignment in the simply-typed lambda calculus and the proof\nsystem is illustrated through the formalization of a proof of type uniqueness\nfor this calculus.\n"}
{"abstract": "  In the pilot-wave theory of quantum mechanics particles have definite\npositions and velocities and the system evolves deterministically. The velocity\nof a particle is determined by the wave function of the system (the guidance\nequation) and the wave function evolves according to Schrodinger's equation. In\nthis paper I first construct a Hamiltonian that gives Schrodinger's equation\nand the guidance equation for the particle. I then find the Hamiltonian for a\nrelativistic particle in Dirac's theory and for a quantum scalar field.\n"}
{"abstract": "  Music is an essential component in our everyday lives and experiences, as it\nis a way that we use to express our feelings, emotions and cultures. In this\nstudy, we explore the association between music genre preferences, demographics\nand moral values by exploring self-reported data from an online survey\nadministered in Canada. Participants filled in the moral foundations\nquestionnaire, while they also provided their basic demographic information,\nand music preferences. Here, we predict the moral values of the participants\ninferring on their musical preferences employing classification and regression\ntechniques. We also explored the predictive power of features estimated from\nfactor analysis on the music genres, as well as the generalist/specialist (GS)\nscore for revealing the diversity of musical choices for each user. Our results\nshow the importance of music in predicting a person's moral values (.55-.69\nAUROC); while knowledge of basic demographic features such as age and gender is\nenough to increase the performance (.58-.71 AUROC).\n"}
{"abstract": "  Where information grows abundant, attention becomes a scarce resource. As a\nresult, agents must plan wisely how to allocate their attention in order to\nachieve epistemic efficiency. Here, we present a framework for multi-agent\nepistemic planning with attention, based on Dynamic Epistemic Logic (DEL, a\npowerful formalism for epistemic planning). We identify the framework as a\nfragment of standard DEL, and consider its plan existence problem. While in the\ngeneral case undecidable, we show that when attention is required for learning,\nall instances of the problem are decidable.\n"}
{"abstract": "  A conceptual model can be used to manage complexity in both the design and\nimplementation phases of the system development life cycle. Such a model\nrequires a firm grasp of the abstract principles on which a system is based, as\nwell as an understanding of the high-level nature of the representation of\nentities and processes. In this context, models can have distinct architectural\ncharacteristics. This paper discusses model multiplicity (e.g., unified\nmodeling language [UML]), model singularity (e.g., object-process methodology\n[OPM], thinging machine [TM]), and a heterogeneous model that involves\nmultiplicity and singularity. The basic idea of model multiplicity is that it\nis not possible to present all views in a single representation, so a number of\nmodels are used, with each model representing a different view. The model\nsingularity approach uses only a single unified model that assimilates its\nsubsystems into one system. This paper is concerned with current approaches,\nespecially in software engineering texts, where multimodal UML is introduced as\nthe general-purpose modeling language (i.e., UML is modeling). In such a\nsituation, we suggest raising the issue of multiplicity versus singularity in\nmodeling. This would foster a basic appreciation of the UML advantages and\ndifficulties that may be faced during modeling, especially in the educational\nsetting. Furthermore, we advocate the claim that a multiplicity of views does\nnot necessitate a multiplicity of models. The model singularity approach can\nrepresent multiple views (static, behavior) without resorting to a collection\nof multiple models with various notations. We present an example of such a\nmodel where the static representation is developed first. Then, the dynamic\nview and behavioral representations are built by incorporating a decomposition\nstrategy interleaved with the notion of time.\n"}
{"abstract": "  The demand for on-device document recognition systems increases in\nconjunction with the emergence of more strict privacy and security\nrequirements. In such systems, there is no data transfer from the end device to\na third-party information processing servers. The response time is vital to the\nuser experience of on-device document recognition. Combined with the\nunavailability of discrete GPUs, powerful CPUs, or a large RAM capacity on\nconsumer-grade end devices such as smartphones, the time limitations put\nsignificant constraints on the computational complexity of the applied\nalgorithms for on-device execution.\n  In this work, we consider document location in an image without prior\nknowledge of the document content or its internal structure. In accordance with\nthe published works, at least 5 systems offer solutions for on-device document\nlocation. All these systems use a location method which can be considered\nHough-based. The precision of such systems seems to be lower than that of the\nstate-of-the-art solutions which were not designed to account for the limited\ncomputational resources.\n  We propose an advanced Hough-based method. In contrast with other approaches,\nit accounts for the geometric invariants of the central projection model and\ncombines both edge and color features for document boundary detection. The\nproposed method allowed for the second best result for SmartDoc dataset in\nterms of precision, surpassed by U-net like neural network. When evaluated on a\nmore challenging MIDV-500 dataset, the proposed algorithm guaranteed the best\nprecision compared to published methods. Our method retained the applicability\nto on-device computations.\n"}
{"abstract": "  Financial regulatory agencies are struggling to manage the systemic risks\nattributed to negative economic shocks. Preventive interventions are prominent\nto eliminate the risks and help to build a more resilient financial system.\nAlthough tremendous efforts have been made to measure multi-risk severity\nlevels, understand the contagion behaviors and other risk management problems,\nthere still lacks a theoretical framework revealing what and how regulatory\nintervention measurements can mitigate systemic risk. Here we demonstrate\nregshock, a practical visual analytical approach to support the exploration and\nevaluation of financial regulation measurements. We propose risk-island, an\nunprecedented risk-centered visualization algorithm to help uncover the risk\npatterns while preserving the topology of financial networks. We further\npropose regshock, a novel visual exploration and assessment approach based on\nthe simulation-intervention-evaluation analysis loop, to provide a heuristic\nsurgical intervention capability for systemic risk mitigation. We evaluate our\napproach through extensive case studies and expert reviews. To our knowledge,\nthis is the first practical systemic method for the financial network\nintervention and risk mitigation problem; our validated approach potentially\nimproves the risk management and control capabilities of financial experts.\n"}
{"abstract": "  Observations of the Galactic Center supermassive black hole Sagittarius A*\n(Sgr A*) with very long baseline interferometry (VLBI) are affected by\ninterstellar scattering along our line of sight. At long radio observing\nwavelengths ($\\gtrsim1\\,$cm), the scattering heavily dominates image\nmorphology. At 3.5 mm (86 GHz), the intrinsic source structure is no longer\nsub-dominant to scattering, and thus the intrinsic emission from Sgr A* is\nresolvable with the Global Millimeter VLBI Array (GMVA). Long-baseline\ndetections to the phased Atacama Large Millimeter/submillimeter Array (ALMA) in\n2017 provided new constraints on the intrinsic and scattering properties of Sgr\nA*, but the stochastic nature of the scattering requires multiple observing\nepochs to reliably estimate its statistical properties. We present new\nobservations with the GMVA+ALMA, taken in 2018, which confirm non-Gaussian\nstructure in the scattered image seen in 2017. In particular, the ALMA-GBT\nbaseline shows more flux density than expected for an anistropic Gaussian\nmodel, providing a tight constraint on the source size and an upper limit on\nthe dissipation scale of interstellar turbulence. We find an intrinsic source\nextent along the minor axis of $\\sim100\\,\\mu$as both via extrapolation of\nlonger wavelength scattering constraints and direct modeling of the 3.5 mm\nobservations. Simultaneously fitting for the scattering parameters, we find an\nat-most modestly asymmetrical (major-to-minor axis ratio of $1.5\\pm 0.2$)\nintrinsic source morphology for Sgr A*.\n"}
{"abstract": "  The field of neuromorphic computing promises extremely low-power and\nlow-latency sensing and processing. Challenges in transferring learning\nalgorithms from traditional artificial neural networks (ANNs) to spiking neural\nnetworks (SNNs) have so far prevented their application to large-scale, complex\nregression tasks. Furthermore, realizing a truly asynchronous and fully\nneuromorphic pipeline that maximally attains the abovementioned benefits\ninvolves rethinking the way in which this pipeline takes in and accumulates\ninformation. In the case of perception, spikes would be passed as-is and\none-by-one between an event camera and an SNN, meaning all temporal integration\nof information must happen inside the network. In this article, we tackle these\ntwo problems. We focus on the complex task of learning to estimate optical flow\nfrom event-based camera inputs in a self-supervised manner, and modify the\nstate-of-the-art ANN training pipeline to encode minimal temporal information\nin its inputs. Moreover, we reformulate the self-supervised loss function for\nevent-based optical flow to improve its convexity. We perform experiments with\nvarious types of recurrent ANNs and SNNs using the proposed pipeline.\nConcerning SNNs, we investigate the effects of elements such as parameter\ninitialization and optimization, surrogate gradient shape, and adaptive\nneuronal mechanisms. We find that initialization and surrogate gradient width\nplay a crucial part in enabling learning with sparse inputs, while the\ninclusion of adaptivity and learnable neuronal parameters can improve\nperformance. We show that the performance of the proposed ANNs and SNNs are on\npar with that of the current state-of-the-art ANNs trained in a self-supervised\nmanner.\n"}
{"abstract": "  We analyze experimentally and theoretically the transport spectra of a gated\nlateral GaAs double quantum dot containing two holes. The strong spin-orbit\ninteraction present in the hole subband lifts the Pauli spin blockade and\nallows to map out the complete spectra of the two-hole system. By performing\nmeasurements in both source-drain voltage directions, at different detunings\nand magnetic fields, we carry out quantitative fitting to a Hubbard two-site\nmodel accounting for the tunnel coupling to the leads and the spin-flip\nrelaxation process. We extract the singlet-triplet gap and the magnetic field\ncorresponding to the singlet-triplet transition in the double-hole ground\nstate. Additionally, at the singlet-triplet transition we find a resonant\nenhancement (in the blockaded direction) and suppression of current (in the\nconduction direction). The current enhancement stems from the multiple\nresonance of two-hole levels, opening several conduction channels at once. The\ncurrent suppression arises from the quantum interference of spin-conserving and\nspin flipping tunneling processes.\n"}
{"abstract": "  In this contribution, the estimates for the response of time delay systems\nwith nonlinear homogeneous right-hand side of degree strictly greater than one\nare constructed. The existing results obtained via the Lyapunov--Razumikhin\napproach are reminded. Their proofs, revisited in the appendix, lead to\nexplicit expressions of the involved constants. Based on a recently introduced\nLyapunov--Krasovskii functional and known estimates of the domain of\nattraction, we present new estimates of the system response. We compare both\napproaches and discuss the illustrative examples.\n"}
{"abstract": "  Let G be a combinatorial graph with vertices V and edges E. A proper coloring\nof G is an assignment of colors to the vertices such that no edge connects two\nvertices of the same color. These are the colorings considered in the famous\nFour Color Theorem. It turns out that the number of proper colorings of G using\nt colors is a polynomial in t, called the chromatic polynomial of G. This\npolynomial has many wonderful properties. It also has the surprising habit of\nappearing in contexts which, a priori, have nothing to do with graph coloring.\nWe will survey three such instances involving acyclic orientations, hyperplane\narrangements, and increasing forests. In addition, connections to symmetric\nfunctions and algebraic geometry will be mentioned.\n"}
{"abstract": "  Annotated medical images are typically rarer than labeled natural images\nsince they are limited by domain knowledge and privacy constraints. Recent\nadvances in transfer and contrastive learning have provided effective solutions\nto tackle such issues from different perspectives. The state-of-the-art\ntransfer learning (e.g., Big Transfer (BiT)) and contrastive learning (e.g.,\nSimple Siamese Contrastive Learning (SimSiam)) approaches have been\ninvestigated independently, without considering the complementary nature of\nsuch techniques. It would be appealing to accelerate contrastive learning with\ntransfer learning, given that slow convergence speed is a critical limitation\nof modern contrastive learning approaches. In this paper, we investigate the\nfeasibility of aligning BiT with SimSiam. From empirical analyses, different\nnormalization techniques (Group Norm in BiT vs. Batch Norm in SimSiam) are the\nkey hurdle of adapting BiT to SimSiam. When combining BiT with SimSiam, we\nevaluated the performance of using BiT, SimSiam, and BiT+SimSiam on CIFAR-10\nand HAM10000 datasets. The results suggest that the BiT models accelerate the\nconvergence speed of SimSiam. When used together, the model gives superior\nperformance over both of its counterparts. We hope this study will motivate\nresearchers to revisit the task of aggregating big pre-trained models with\ncontrastive learning models for image analysis.\n"}
{"abstract": "  Inertial cavitation in soft matter is an important phenomenon featured in a\nwide array of biological and engineering processes. Recent advances in\nexperimental, theoretical, and numerical techniques have provided access into a\nworld full of nonlinear physics, yet most of our quantitative understanding to\ndate has been centered on a spherically symmetric description of the cavitation\nprocess. However, cavitation bubble growth and collapse rarely occur in a\nperfectly symmetrical fashion, particularly in soft materials. Predicting the\nonset of dynamically arising, non-spherical instabilities has remained a\nsignificant, unresolved challenge in part due to the additional constitutive\ncomplexities introduced by the surrounding nonlinear viscoelastic solid. Here,\nwe provide a new theoretical model capable of accurately predicting the onset\nof non-spherical instability shapes of a bubble in a soft material by\nexplicitly accounting for all pertinent nonlinear interactions between the\nfluid-like cavitation bubble and the solid-like surroundings. Comparison\nagainst high-resolution experimental images from laser-induced cavitation\nevents in a polyacrylamide (PA) hydrogel show excellent agreement.\nInterestingly, and consistent with experimental findings, our model predicts\nthe emergence of various dynamic instability shapes for hoop stretch ratios\ngreater than one in contrast to most quasi-static investigations. Our new\ntheoretical framework not only provides unprecedented insight into the\ncavitation dynamics in a soft solid, but it also provides a quantitative means\nof interpreting bubble dynamics relevant to a wide array of engineering and\nmedical applications as well as natural phenomena.\n"}
{"abstract": "  Growing individualization of products up to lot-size-1 and high volatility of\nproduct mixes lead to new challenges in the manufacturing domain, including the\nneed for frequent reconfiguration of the system and reacting to changing\norders. Thus, apart from functional aspects, safety aspects of the production\nsystem as well as product quality assurance aspects must be addressed for\nflexible and reconfigurable manufacturing systems at runtime. To cope with the\nmentioned challenges, we present an integrated model-based approach SQUADfps\n(machine Safety and product QUAlity for flexible proDuction systems) to support\nthe automatic conduct of the risk assessment of flexible production scenarios\nin terms of safety as well as the process-FMEA to ensure that the requirements\nw.r.t. the quality of the production process and the resulting product are met.\nOur approach is based on a meta-model which captures all information needed to\nconduct both risk assessment and process-FMEA dynamically during the runtime,\nand thus enables flexible manufacturing scenarios with frequent changes of the\nproduction system and orders up to a lot-size of one while guaranteeing safety\nand product quality requirements. The automatically generated results will\nassist human in making further decisions. To demonstrate the feasibility of our\napproach, we apply it to a case study.\n"}
{"abstract": "  We present an experimental and theoretical study of core-level ionization of\nsmall hetero- and homo-nuclear molecules employing circularly polarized light\nand address molecular-frame photoelectron angular distributions in the light's\npolarization plane (CP-MFPADs). We find that the main forward-scattering peaks\nof CP-MFPADs are slightly tilted with respect to the molecular axis. We show\nthat this tilt angle can be directly connected to the molecular bond length by\na simple, universal formula. The extraction of the bond length becomes more\naccurate as the photoelectron energy is increased. We apply the derived formula\nto several examples of CP-MFPADs of C 1s and O 1s photoelectrons of CO, which\nhave been measured experimentally or obtained by means of ab initio modeling.\nThe photoelectron kinetic energies range from 70 to 1000~eV and the extracted\nbond lengths agree well with the known bond length of the CO molecule in its\nground state. In addition, we discuss the influence of the back-scattering\ncontribution that is superimposed over the analyzed forward-scattering peak in\ncase of homo-nuclear diatomic molecules as N$_2$.\n"}
{"abstract": "  Secure wireless access in ultra-reliable low-latency communications (URLLC),\nwhich is a critical aspect of 5G security, has become increasingly important\ndue to its potential support of grant-free configuration. In grant-free URLLC,\nprecise allocation of different pilot resources to different users that share\nthe same time-frequency resource is essential for the next generation NodeB\n(gNB) to exactly identify those users under access collision and to maintain\nprecise channel estimation required for reliable data transmission. However,\nthis process easily suffers from attacks on pilots. We in this paper propose a\nquantum learning based nonrandom superimposed coding method to encode and\ndecode pilots on multidimensional resources, such that the uncertainty of\nattacks can be learned quickly and eliminated precisely. Particularly,\nmultiuser pilots for uplink access are encoded as distinguishable subcarrier\nactivation patterns (SAPs) and gNB decodes pilots of interest from observed\nSAPs, a superposition of SAPs from access users, by joint design of attack mode\ndetection and user activity detection though a quantum learning network (QLN).\nWe found that the uncertainty lies in the identification process of codeword\ndigits from the attacker, which can be always modelled as a black-box model,\nresolved by a quantum learning algorithm and quantum circuit. Novel analytical\nclosed-form expressions of failure probability are derived to characterize the\nreliability of this URLLC system with short packet transmission. Simulations\nhow that our method can bring ultra-high reliability and low latency despite\nattacks on pilots.\n"}
{"abstract": "  The ability to reconstruct the kinematic parameters of hand movement using\nnon-invasive electroencephalography (EEG) is essential for strength and\nendurance augmentation using exosuit/exoskeleton. For system development, the\nconventional classification based brain computer interface (BCI) controls\nexternal devices by providing discrete control signals to the actuator. A\ncontinuous kinematic reconstruction from EEG signal is better suited for\npractical BCI applications. The state-of-the-art multi-variable linear\nregression (mLR) method provides a continuous estimate of hand kinematics,\nachieving maximum correlation of upto 0.67 between the measured and the\nestimated hand trajectory. In this work, three novel source aware deep learning\nmodels are proposed for motion trajectory prediction (MTP). In particular,\nmulti layer perceptron (MLP), convolutional neural network - long short term\nmemory (CNN-LSTM), and wavelet packet decomposition (WPD) CNN-LSTM are\npresented. Additional novelty of the work includes utilization of brain source\nlocalization (using sLORETA) for the reliable decoding of motor intention\nmapping (channel selection) and accurate EEG time segment selection.\nPerformance of the proposed models are compared with the traditionally utilised\nmLR technique on the real grasp and lift (GAL) dataset. Effectiveness of the\nproposed framework is established using the Pearson correlation coefficient and\ntrajectory analysis. A significant improvement in the correlation coefficient\nis observed when compared with state-of-the-art mLR model. Our work bridges the\ngap between the control and the actuator block, enabling real time BCI\nimplementation.\n"}
{"abstract": "  Rapid growth in scientific data and a widening gap between computational\nspeed and I/O bandwidth makes it increasingly infeasible to store and share all\ndata produced by scientific simulations. Instead, we need methods for reducing\ndata volumes: ideally, methods that can scale data volumes adaptively so as to\nenable negotiation of performance and fidelity tradeoffs in different\nsituations. Multigrid-based hierarchical data representations hold promise as a\nsolution to this problem, allowing for flexible conversion between different\nfidelities so that, for example, data can be created at high fidelity and then\ntransferred or stored at lower fidelity via logically simple and mathematically\nsound operations. However, the effective use of such representations has been\nhindered until now by the relatively high costs of creating, accessing,\nreducing, and otherwise operating on such representations. We describe here\nhighly optimized data refactoring kernels for GPU accelerators that enable\nefficient creation and manipulation of data in multigrid-based hierarchical\nforms. We demonstrate that our optimized design can achieve up to 264 TB/s\naggregated data refactoring throughput -- 92% of theoretical peak -- on 1024\nnodes of the Summit supercomputer. We showcase our optimized design by applying\nit to a large-scale scientific visualization workflow and the MGARD lossy\ncompression software.\n"}
{"abstract": "  Heat kernels arise in a variety of contexts including probability, geometry,\nand functional analysis; the automorphic heat kernel is particularly important\nin number theory and string theory. The typical construction of an automorphic\nheat kernel as a Poincar\\'{e} series presents analytic difficulties, which can\nbe dealt with in special cases (e.g. hyperbolic spaces) but are often\nsidestepped in higher rank by restricting to the compact quotient case. In this\npaper, we present a new approach, using global automorphic Sobolev theory, a\nrobust framework for solving automorphic PDEs that does not require any\nsimplifying assumptions about the rank of the symmetric space or the\ncompactness of the arithmetic quotient. We construct an automorphic heat kernel\nvia its automorphic spectral expansion in terms of cusp forms, Eisenstein\nseries, and residues of Eisenstein series. We then prove uniqueness of the\nautomorphic heat kernel as an application of operator semigroup theory.\nFinally, we prove the smoothness of the automorphic heat kernel by proving that\nits automorphic spectral expansion converges in the $C^\\infty$-topology.\n"}
{"abstract": "  We complete the description of automorphism groups of all nonsplit extensions\nof elementary abelian $2$-groups by $PSL_2(q)$, with $q$ odd, for an\nirreducible induced action. An application of this result to the theory of\n$\\pi$-submaximal subgroups is given.\n"}
{"abstract": "  We propose a simple recursive algorithm that allows the computation of the\nfirst- and second-order derivatives with respect to the inputs of an arbitrary\ndeep feed forward neural network (DFNN). The algorithm naturally incorporates\nthe derivatives with respect to the network parameters. To test the algorithm,\nwe apply it to the study of the quantum mechanical variational problem for few\ncases of simple potentials, modeling the ground-state wave function in terms of\na DFNN.\n"}
{"abstract": "  We present a statistical testing framework to detect if a given machine\nlearning classifier fails to satisfy a wide range of group fairness notions.\nThe proposed test is a flexible, interpretable, and statistically rigorous tool\nfor auditing whether exhibited biases are intrinsic to the algorithm or due to\nthe randomness in the data. The statistical challenges, which may arise from\nmultiple impact criteria that define group fairness and which are discontinuous\non model parameters, are conveniently tackled by projecting the empirical\nmeasure onto the set of group-fair probability models using optimal transport.\nThis statistic is efficiently computed using linear programming and its\nasymptotic distribution is explicitly obtained. The proposed framework can also\nbe used to test for testing composite fairness hypotheses and fairness with\nmultiple sensitive attributes. The optimal transport testing formulation\nimproves interpretability by characterizing the minimal covariate perturbations\nthat eliminate the bias observed in the audit.\n"}
{"abstract": "  It is well-known that the densest lattice sphere packings also typically have\nthe largest kissing numbers. The sphere packing density maximization problem is\nknown to have a solution among well-rounded lattices, of which the integer\nlattice $\\mathbb{Z}^n$ is the simplest example. The integer lattice is also an\nexample of a generic well-rounded lattice, i.e., a well-rounded lattice with a\nminimal kissing number. However, the integer lattice has the worst density\namong well-rounded lattices. In this paper, the problem of constructing\nexplicit generic well-rounded lattices with dense sphere packings is\nconsidered. To this end, so-called tame lattices recently introduced by Damir\net al. are utilized. Tame lattices came to be as a generalization of the ring\nof integers of certain abelian number fields. The sublattices of tame lattices\nconstructed in this paper are shown to always result in either a generic\nwell-rounded lattice or the lattice $A_n$, with density ranging between that of\n$\\mathbb{Z}^n$ and $A_n$. Further, explicit deformations of some known densest\nlattice packings are constructed, yielding a family of generic well-rounded\nlattices with densities arbitrarily close to the optimum. In addition to being\nan interesting mathematical problem on its own, the constructions are also\nmotivated from a more practical point of view. Namely, the constructed lattices\nprovide good candidates for lattice codes used in secure wireless\ncommunications.\n"}
{"abstract": "  The quantum anomalous Hall (QAH) effect in magnetic topological insulator\n(TI) represents a new state of matter originated from the interplay between\ntopology and magnetism. The defining characteristics of the QAH ground state\nare the quantized Hall resistivity ($\\rho_{yx}$) and vanishing longitudinal\nresistivity ($\\rho_{xx}$) in the absence of external magnetic field. A\nfundamental question concerning the QAH effect is whether it is merely a\nzero-magnetic-field quantum Hall (QH) effect, or if it can host unique quantum\nphases and phase transitions that are unavailable elsewhere. The most dramatic\ndeparture of the QAH systems from other QH systems lies in the strong magnetic\ndisorders that induce spatially random magnetization. Because disorder and\nmagnetism play pivotal roles in the phase diagram of two-dimensional electron\nsystems, the high degree of magnetic disorders in QAH systems may create novel\nphases and quantum critical phenomena. In this work, we perform systematic\ntransport studies of a series of magnetic TIs with varied strength of magnetic\ndisorders. We find that the ground state of QAH effect can be categorized into\ntwo distinct classes: the QAH insulator and anomalous Hall (AH) insulator\nphases, as the zero-magnetic-field counterparts of the QH liquid and Hall\ninsulator in the QH systems. In the low disorder limit of the QAH insulator\nregime, we observe a universal quantized longitudinal resistance $\\rho_{xx} =\nh/e^{2}$ at the coercive field. In the AH insulator regime, we find that a\nmagnetic field can drive it to the QAH insulator phase through a quantum\ncritical point with distinct scaling behaviors from that in the QH phase\ntransition. We propose that the transmission between chiral edge states at\ndomain boundaries, tunable by disorder and magnetic fields, is the key for\ndetermining the QAH ground state.\n"}
{"abstract": "  In this article we deal with the free Banach lattice $FBL\\langle \\mathbb{L}\n\\rangle$ generated by a lattice $\\mathbb{L}$. We prove that if $FBL\\langle\n\\mathbb{L} \\rangle$ is projective then $\\mathbb{L}$ has a maximum and a\nminimum. On the other hand, we show that if $\\mathbb{L}$ has maximum and\nminimum then $FBL\\langle \\mathbb{L} \\rangle$ is $2$-lattice isomorphic to a\n$C(K)$-space. As a consequence, $FBL\\langle \\mathbb{L} \\rangle$ is projective\nif and only if it is lattice isomorphic to a $C(K)$-space with $K$ being an\nabsolute neighborhood retract. As an application, we characterize those\nlinearly ordered sets and Boolean algebras for which the corresponding free\nBanach lattice is projective.\n"}
{"abstract": "  In this work, we focus on two crucial components in the cross-domain\ntext-to-SQL semantic parsing task: schema linking and value filling. To\nencourage the model to learn better encoding ability, we propose a column\nselection auxiliary task to empower the encoder with the relevance matching\ncapability by using explicit learning targets. Furthermore, we propose two\nvalue filling methods to build the bridge from the existing zero-shot semantic\nparsers to real-world applications, considering most of the existing parsers\nignore the values filling in the synthesized SQL. With experiments on Spider,\nour proposed framework improves over the baselines on the execution accuracy\nand exact set match accuracy when database contents are unavailable, and\ndetailed analysis sheds light on future work.\n"}
{"abstract": "  The routing number is a graph invariant introduced by Alon, Chung, and Graham\nin 1994, and it has been studied for trees and other classes of graphs such as\nhypercubes. It gives the minimum number of routing steps needed to sort a set\nof distinct tokens, placed one on each vertex, where each routing step swaps a\nset of disjoint pairs of adjacent tokens. Our main theorem generalizes the\nknown estimate that a rectangular grid graph R with width w(R) and height h(R)\nhas routing number rt(R) in O(w(R)+h(R)). We show that for the subgraph P of\nthe infinite square lattice enclosed by any convex polygon, its routing number\nrt(P) is in O(w(P)+h(P)).\n"}
{"abstract": "  Diodes made of heterostructures of the 2D material graphene and conventional\n3D materials are reviewed in this manuscript. Several applications in high\nfrequency electronics and optoelectronics are highlighted. In particular,\nadvantages of metal-insulator-graphene (MIG) diodes over conventional\nmetal-insulator-metal diodes are discussed with respect to relevant\nfigures-of-merit. The MIG concept is extended to 1D diodes. Several\nexperimentally implemented radio frequency circuit applications with MIG diodes\nas active elements are presented. Furthermore, graphene-silicon Schottky diodes\nas well as MIG diodes are reviewed in terms of their potential for\nphotodetection. Here, graphene-based diodes have the potential to outperform\nconventional photodetectors in several key figures-of-merit, such as overall\nresponsivity or dark current levels. Obviously, advantages in some areas may\ncome at the cost of disadvantages in others, so that 2D/3D diodes need to be\ntailored in application-specific ways.\n"}
{"abstract": "  It is well-known that GPS is vulnerable to signal spoofing attacks. Although\nseveral spoofing detection techniques exist, they are incapable of mitigation\nand recovery from stealthy attackers. In this work, we present SemperFi, a\nsingle antenna GPS receiver capable of tracking legitimate GPS satellite\nsignals and estimating the true location even during a spoofing attack. The\nmain challenge in building SemperFi is, unlike most wireless systems where\n\\emph{the data} contained in the wireless signals is important, GPS relies on\nthe time of arrival (ToA) of satellite signals. SemperFi is capable of\ndistinguishing spoofing signals and recovering legitimate GPS signals that are\neven completely overshadowed by a strong adversary. We exploit the short-term\nstability of inertial sensors to identify the spoofing signal and extend the\nsuccessive interference cancellation algorithm to preserve the legitimate\nsignal's ToA. We implement SemperFi in GNSS-SDR, an open-source\nsoftware-defined GNSS receiver, and evaluate its performance using UAV\nsimulators, real drones, a variety of real-world GPS datasets, and various\nembedded platforms. Our evaluation results indicate that in many scenarios,\nSemperFi can identify adversarial peaks by executing flight patterns that are\nless than 50 m long and recover the true location within 10 seconds (Jetson\nXavier). We show that our receiver is secure against stealthy attackers who\nexploit inertial sensor errors and execute seamless takeover attacks. We design\nSemperFi as a pluggable module capable of generating a spoofer-free GPS signal\nfor processing on any commercial-off-the-shelf GPS receiver available today.\nFinally, we release our implementation to the community for usage and further\nresearch.\n"}
{"abstract": "  We present an implementation of the trimmed serendipity finite element\nfamily, using the open source finite element package Firedrake. The new\nelements can be used seamlessly within the software suite for problems\nrequiring $H^1$, \\hcurl, or \\hdiv-conforming elements on meshes of squares or\ncubes. To test how well trimmed serendipity elements perform in comparison to\ntraditional tensor product elements, we perform a sequence of numerical\nexperiments including the primal Poisson, mixed Poisson, and Maxwell cavity\neigenvalue problems. Overall, we find that the trimmed serendipity elements\nconverge, as expected, at the same rate as the respective tensor product\nelements while being able to offer significant savings in the time or memory\nrequired to solve certain problems.\n"}
{"abstract": "  The multi-color CCD photometric study of 27 Delta Scuti stars is presented.\nBy using approximately three years of photometric observations, we obtained the\ntimes of maxima and magnitude changes during the observation time interval for\neach star. The ephemerides of our Delta Scuti stars were calculated based on\nthe Markov Chain Monte Carlo (MCMC) method using the observed times of maxima\nand the period of the stars' oscillations. We used the Gaia EDR3 parallaxes to\ncalculate the luminosities and also the absolute magnitudes of these Delta\nScuti stars. The fundamental physical parameters of all the stars in our sample\nsuch as masses and radii were estimated. We determined the pulsation modes of\nthe stars based on the pulsation constants. Moreover, the period-luminosity\n(P-L) relation of Delta Scuti stars was investigated and discussed. Then, by\nusing a machine learning classification, new P-L relations for fundamental and\novertone modes are presented.\n"}
{"abstract": "  GPS technology has revolutionized the way we localize and navigate outdoors.\nHowever, the poor reception of GPS signals in buildings makes it unsuitable for\nindoor localization. WiFi fingerprinting-based indoor localization is one of\nthe most promising ways to meet this demand. Unfortunately, most work in the\ndomain fails to resolve challenges associated with deployability on\nresource-limited embedded devices. In this work, we propose a compression-aware\nand high-accuracy deep learning framework called CHISEL that outperforms the\nbest-known works in the area while maintaining localization robustness on\nembedded devices.\n"}
{"abstract": "  Inferring social relations from dialogues is vital for building emotionally\nintelligent robots to interpret human language better and act accordingly. We\nmodel the social network as an And-or Graph, named SocAoG, for the consistency\nof relations among a group and leveraging attributes as inference cues.\nMoreover, we formulate a sequential structure prediction task, and propose an\n$\\alpha$-$\\beta$-$\\gamma$ strategy to incrementally parse SocAoG for the\ndynamic inference upon any incoming utterance: (i) an $\\alpha$ process\npredicting attributes and relations conditioned on the semantics of dialogues,\n(ii) a $\\beta$ process updating the social relations based on related\nattributes, and (iii) a $\\gamma$ process updating individual's attributes based\non interpersonal social relations. Empirical results on DialogRE and MovieGraph\nshow that our model infers social relations more accurately than the\nstate-of-the-art methods. Moreover, the ablation study shows the three\nprocesses complement each other, and the case study demonstrates the dynamic\nrelational inference.\n"}
{"abstract": "  In this paper, we consider the renormalization operator $\\mathcal R$ for\nmultimodal maps. We prove the renormalization operator $\\mathcal R$ is a\nself-homeomorphism on any totally $\\mathcal R$-invariant set. As a corollary,\nwe prove the existence of the full renormalization horseshoe for multimodal\nmaps.\n"}
{"abstract": "  The interplay between quadrupole and octupole degrees of freedom is discussed\nin a series of neutron-rich actinides and superheavy nuclei with $92 \\le$ Z\n$\\le 110$ and $186 \\le$ N $\\le 202$. In addition to the static\nHartree-Fock-Bogoliubov approach, dynamical beyond-mean-field correlations are\ntaken into account via both parity restoration and symmetry-conserving\nGenerator Coordinate Method calculations based on the Gogny-D1M energy density\nfunctional. Physical properties such as correlation energies, negative-parity\nexcitation energies as well as reduced transition probabilities $B(E1)$ and\n$B(E3)$ are discussed in detail. It is shown that, for the studied nuclei, the\nquadrupole-octupole coupling is weak and to a large extent the properties of\nnegative parity states can be reasonably well described in terms of the\noctupole degree of freedom alone.\n"}
{"abstract": "  BINGO (Baryon Acoustic Oscillations from Integrated Neutral Gas\nObservations.) is a radio telescope designed to survey from 980 MHz to 1260\nMHz, observe the neutral Hydrogen (HI) 21-cm line and detect BAO (Baryon\nAcoustic Oscillation) signal with Intensity Mapping technique. Here we present\nour method to generate mock maps of the 21-cm Intensity Mapping signal covering\nthe BINGO frequency range and related test results. (Abridged)\n"}
{"abstract": "  Domain discretization is an essential part of the solution procedure in\nnumerical simulations. Meshless methods simplify the domain discretization to\npositioning of nodes in the interior and on the boundary of the domain.\nHowever, generally speaking, the shape of the boundary is often undefined and\nthus needs to be constructed before it can be discretized with a desired\ninternodal spacing. Domain shape construction is far from trivial and is the\nmain challenge of this paper. We tackle the simulation of moving boundary\nproblems where the lack of domain shape information can introduce difficulties.\nWe present a solution for 2D surface reconstruction from discretization points\nusing cubic splines and thus providing a surface description anywhere in the\ndomain. We also demonstrate the presented algorithm in a simulation of\nphase-change-like problem.\n"}
{"abstract": "  The key idea of power-domain non-orthogonal multiple access (NOMA) is to\nexploit the superposition coding (SC) combined with successive interference\ncancellation (SIC) technique (called SC-SIC) while reducing the receivers'\ncomplexity as well as error propagation. Actually, NOMA suggests a\nlow-complexity scheme, where users are grouped into multiple clusters operating\nin isolated resource blocks, and SC-SIC is performed among users within each\ncluster. In this paper, we propose a globally optimal joint intra- and\ninter-cluster power allocation for any arbitrary user grouping to maximize\nusers' sum-rate. In this algorithm, we exploit the closed-form of optimal\nintra-cluster power allocation obtained in our previous work. Then, by\ntransforming network-NOMA to an equivalent virtual network-OMA, we show that\nthe optimal power allocation can be obtained based on the very fast\nwater-filling algorithm. Interestingly, we observe that each NOMA cluster acts\nas a virtual OMA user whose effective channel gain is obtained in closed form.\nAlso, each virtual OMA user requires a minimum power to satisfy the minimum\nrate demand of its real multiplexed user. In simulation results, we evaluate\nthe performance gap between fully SC-SIC, NOMA, and OMA, in terms of users\nsum-rate, and outage probability.\n"}
{"abstract": "  Understanding the formation of star clusters with low star-formation\nefficiency (SFE) is very important to know about the star-formation history. In\nN-body models of star cluster evolution after gas expulsion, the Plummer model\nwith outer power law density profile has been used massively. We study the\nimpact of the density profile slopes on the survivability of the low-SFE star\nclusters after instantaneous gas expulsion. We compare cases when stellar\ncluster has Plummer profile and Dehnen profiles with cusp of different slopes\nat the time of formation. We determine the corresponding density profile of the\nresidual gas for a given global SFE, assuming that our model clusters formed\nwith a constant efficiency per free-fall time and hence have shallower density\nprofile of gas than that of stars. We perform direct $N$-body simulations of\nevolution of clusters initially in virial equilibrium within gas potential\nafter gas removal. We find that the violent relaxation lasts no longer than\n20~Myr independently of the density profile power law slopes. Dehnen model\nclusters survive after violent relaxation with significantly lower SFEs when\nthe global SFE measured within the Jacobi radius or within a half-mass radius.\nDehnen $\\gamma=0$ model clusters show similar final bound fraction with the\nPlummer model clusters if global SFE is measured within 10 scale radii. The\nfinal bound fraction increases with $\\gamma$ values for a given global SFE. We\nconclude that Dehnen clusters better resist the consequences of the violent\nrelaxation followed the instantaneous gas expulsion than the Plummer clusters.\nThus the shallower the outer density slope of the low-SFE clusters, the better\nfor their survivability after gas expulsion. Among Dehnen clusters we find that\nthe steeper the inner slope (cusp) the higher the bound mass fraction is\nretained after violent relaxation for a given global SFE.\n"}
{"abstract": "  Deep neural networks are vulnerable to input deformations in the form of\nvector fields of pixel displacements and to other parameterized geometric\ndeformations e.g. translations, rotations, etc. Current input deformation\ncertification methods either 1. do not scale to deep networks on large input\ndatasets, or 2. can only certify a specific class of deformations, e.g. only\nrotations. We reformulate certification in randomized smoothing setting for\nboth general vector field and parameterized deformations and propose\nDeformRS-VF and DeformRS-Par, respectively. Our new formulation scales to large\nnetworks on large input datasets. For instance, DeformRS-Par certifies rich\ndeformations, covering translations, rotations, scaling, affine deformations,\nand other visually aligned deformations such as ones parameterized by\nDiscrete-Cosine-Transform basis. Extensive experiments on MNIST, CIFAR10, and\nImageNet show competitive performance of DeformRS-Par achieving a certified\naccuracy of $39\\%$ against perturbed rotations in the set\n$[-10\\degree,10\\degree]$ on ImageNet.\n"}
{"abstract": "  We study the motion of charged particles in a family of five-dimensional\nsolutions describing either a black hole or topological star with a fifth\ncompact dimension stabilised by a magnetic flux. The particle's trajectory is\nshown to move along the surface of a Poincar\\'{e} cone. The radial motion shows\na rich structure where the existence of various bound, plunging, or escaping\ntrajectories depend on the constants of motion. Curves of energy and angular\nmomentum corresponding to spherical orbits show a swallow-tail structure highly\nreminiscent to phase transitions of thermodynamics. When the momentum along the\ncompact direction is varied, the is a critical point beyond which the\nswallow-tail kink disappears and becomes a smooth curve.\n"}
{"abstract": "  We consider dual-primal isogeometric tearing and interconnection (IETI-DP)\nsolvers for multi-patch geometries in Isogeometric Analysis. Recently, the\nauthors have published a convergence analysis for those solvers that is\nexplicit in both the grid size and the spline degree for conforming\ndiscretizations of two dimensional computational domains. In the present paper,\nwe shortly revisit these results and provide numerical experiments that\nindicate that similar results may hold for three dimensional domains.\n"}
{"abstract": "  Localization is a characteristic phenomenon of space-inhomogeneous quantum\nwalks in one dimension, where particles remain localized around their initial\nposition. The existence of eigenvalues of time evolution operators is a\nnecessary and sufficient condition for the occurrence of localization, and\ntheir associated eigenvectors are deeply related to the amount of localization,\ni.e., the probability that the walker stays around the starting position in the\nlong-time limit. In a previous study by authors, the eigenvalues of two-phase\nquantum walks with one defect were studied using a transfer matrix, which\nfocused on the occurrence of localization (Quantum Inf. Process 20(5), 2021).\nIn this paper, we introduce the analytical method to calculate eigenvectors\nusing the transfer matrix and also extend our results to characterize\neigenvalues not only for two-phase quantum walks with one defect but also for a\nmore general space-inhomogeneous model.\n  With these results, we quantitatively evaluate localization and study the\nstrong trapping property by deriving the time-averaged limit distributions of\nfive models studied previously.\n"}
{"abstract": "  Darboux theorem in symplectic geometry is the crux of emergent gravity in\nwhich the gravitational metric emerges from a noncommutative U(1)-theory.\nTopological T-duality, on the other hand, is a relation between two a priori\ndifferent backgrounds (with different geometries, different fluxes and even\ntopologically distinct manifolds) which nevertheless behave identically from a\nphysical point of view. For us these backgrounds are principal torus bundles on\nthe same base manifold. In this article we review how these theories can be\nnaturally understood in the light of generalized geometry. Generalized geometry\nprovides an unifying framework for such a systematic approach and gives rise to\nthe group of Courant automorphism $Diff(M) \\ltimes \\Omega_{closed}^{2}(M)$ for\nthe $TM \\oplus T^*M$ bundle. Here we propose a novel geometric construction for\nthe T-dual of an emergent gravity theory implemented between the\n$\\mathbb{T}$-bundles and this duality is realized using the\nGualtieri-Cavalcanti map that establishes an isomorphism between Courant\nalgebroids. In the case of flat spacetime we obtain that, under mild\nassumptions, the T-dual of emergent gravity is again an emergent theory of\ngravity. In the general case we obtained formulas for the T-dual of an emergent\nmetric in a $\\mathbb{T}^2$-fibration, however due to the appearance of H-flux\nafter T-dualizing, the theory thus obtained can no longer be considered in the\nusual framework of emergent gravity. This motivates the study of emergent\ngravity with non trivial H-flux.\n"}
{"abstract": "  Urban Air Mobility, the scenario where hundreds of manned and Unmanned\nAircraft System (UAS) carry out a wide variety of missions (e.g. moving humans\nand goods within the city), is gaining acceptance as a transportation solution\nof the future. One of the key requirements for this to happen is safely\nmanaging the air traffic in these urban airspaces. Due to the expected density\nof the airspace, this requires fast autonomous solutions that can be deployed\nonline. We propose Learning-'N-Flying (LNF) a multi-UAS Collision Avoidance\n(CA) framework. It is decentralized, works on-the-fly and allows autonomous UAS\nmanaged by different operators to safely carry out complex missions,\nrepresented using Signal Temporal Logic, in a shared airspace. We initially\nformulate the problem of predictive collision avoidance for two UAS as a\nmixed-integer linear program, and show that it is intractable to solve online.\nInstead, we first develop Learning-to-Fly (L2F) by combining: a) learning-based\ndecision-making, and b) decentralized convex optimization-based control. LNF\nextends L2F to cases where there are more than two UAS on a collision path.\nThrough extensive simulations, we show that our method can run online\n(computation time in the order of milliseconds), and under certain assumptions\nhas failure rates of less than 1% in the worst-case, improving to near 0% in\nmore relaxed operations. We show the applicability of our scheme to a wide\nvariety of settings through multiple case studies.\n"}
{"abstract": "  By revising the application of the open quantum system approach to the early\nuniverse and extending it to the conditions beyond the Markovian approximation,\nwe obtain a new non-Markovian quantum Boltzmann equation. Throughout the paper,\nwe also develop an extension of the quantum Boltzmann equation to describe the\nprocesses that are irreversible at the macroscopic level. This new kinetic\nequation is, in principle, applicable to a wide variety of processes in the\nearly universe. For instance, using this equation one can accurately study the\nmicroscopic influence of a cosmic environment on a system of cosmic background\nphotons or stochastic gravitational waves. In this paper, we apply the\nnon-Markovian quantum Boltzmann equation to study the damping of gravitational\nwaves propagating in a medium consisting of decoupled ultra-relativistic\nneutrinos. For such a system, we study the time evolution of the intensity and\nthe polarization of the gravitational waves. It is shown that, in contrast to\nintensity and linear polarization which are damped, the circular polarization\n(V-mode) of the gravitational wave (if present) is amplified by propagating\nthrough such a medium.\n"}
{"abstract": "  Genetic programming is the practice of evolving formulas using crossover and\nmutation of genes representing functional operations. Motivated by genetic\nevolution we develop and solve two combinatorial games, and we demonstrate some\nadvantages and pitfalls of using genetic programming to investigate Grundy\nvalues. We conclude by investigating a combinatorial game whose ruleset and\nstarting positions are inspired by genetic structures.\n"}
{"abstract": "  We prove a large-deviation principle (LDP) for the sample paths of jump\nMarkov processes in the small noise limit when, possibly, all the jump rates\nvanish uniformly, but slowly enough, in a region of the state space. We further\ndiscuss the optimality of our assumptions on the decay of the jump rates. As a\ndirect application of this work we relax the assumptions needed for the\napplication of LDPs to, e.g., Chemical Reaction Network dynamics, where\nvanishing reaction rates arise naturally particularly the context of mass\naction kinetics.\n"}
{"abstract": "  The multidisciplinarity of robotics creates a need for robust integration\nmethodologies that can facilitate the adoption of state-of-the-art research\ncomponents in an industrial application. Unfortunately, there are no clear,\ncommunity accepted guidelines or standards that define the integration of such\ncomponents in a single robotic system. In this paper, we propose a methodology\nthat assesses the software components of a candidate system on the basis of the\neffort required to integrate them and the impact their integration will have on\na target system. We demonstrate how this methodology can be applied using an\nindustrial tool packing system as an example. The system integrates a wide\nrange of both in-house and third-party research outputs and software\ncomponents. We prove the effectiveness of our approach by evaluating system\nperformance with an experimental benchmark that assesses the robustness,\nreliability and operational speed of the system for the given packing task. We\nalso demonstrate how our methodology can be used to predict the amount of\nintegration time required for a component. The proposed integration methodology\ncan be applied to any robotic system to facilitate its transition from the\nresearch to an industrial environment.\n"}
{"abstract": "  The interaction of Galactic-Centre (GC) super bubbles (GSB) with the gaseous\ndisc and halo of the Milky Way is investigated using radio continuum, X-ray, HI\nand CO line surveys. The radio North Polar Spur (NPS) constitutes the brightest\neastern ridge of GSB, brightening towards the galactic plane and reaching $ l =\n22\\deg, \\ b = + 2\\deg$ at the sharpest end, where it intersects the tangential\ndirection of the 3-kpc expanding ring and crater. Examination of the spur\nridges reveals that the entire GSB, including the NPS and its counter spurs,\nconstitutes a GC-symmetrical $\\Omega /$\\rotatebox[origin=c]{180}{$\\Omega$}\nshape. The thickness and gas density of the HI and CO discs are shown to\nincrease sharply from the inside (lower longitude) to the outside of the 3-kpc\ncrater. Formation of crater is explained by the sweeping of the upper layer of\ndisc gas by the shock wave from the GC by the explosion $ \\sim 10 $ My ago with\nthe emitted energy of several $10 ^ {55} $ ergs. Based on the discussion, a\nunified view on the structure and formation mechanism of GSB is presented.\n"}
{"abstract": "  Causal Optimal Transport (COT) results from imposing a temporal causality\nconstraint on classic optimal transport problems, which naturally generates a\nnew concept of distances between distributions on path spaces. The first\napplication of the COT theory for sequential learning was given in Xu et al.\n(2020), where COT-GAN was introduced as an adversarial algorithm to train\nimplicit generative models optimized for producing sequential data. Relying on\nXu et al. (2020), the contribution of the present paper is twofold. First, we\ndevelop a conditional version of COT-GAN suitable for sequence prediction. This\nmeans that the dataset is now used in order to learn how a sequence will evolve\ngiven the observation of its past evolution. Second, we improve on the\nconvergence results by working with modifications of the empirical measures via\na specific type of quantization due to Backhoff et al. (2020). The resulting\nquantized conditional COT-GAN algorithm is illustrated with an application for\nvideo prediction.\n"}
{"abstract": "  Following an eruptive accretion event in NGC6334I-MM1, flares in the various\nmaser species, including water masers, were triggered. We report the observed\nrelative proper motion of the highly variable water masers associated with the\nmassive star-forming region, NGC6334I. High velocity H$_2$O maser proper\nmotions were detected in 5 maser clusters, CM2-W2 (bow-shock structure),\nMM1-W1, MM1-W3, UCHII-W1 and UCHII-W3. The overall average of the derived\nrelative proper motion is 85 km s$^{-1}$. This mean proper motion is in\nagreement with the previous results from VLA multi-epoch observations. Our\nposition and velocity variance and co-variance matrix analyses of the maser\nproper motions show its major axis to have a position angle of $-$79.4$^\\circ$,\ncutting through the dust cavity around MM1B and aligned in the\nnorthwest-southeast direction. We interpret this as the axis of the jet driving\nthe CM2 shock and the maser motion. The complicated proper motions in MM1-W1\ncan be explained by the combined influence of the MM1 northeast-southwest\nbipolar outflow, CS(6-5) north-south collimated bipolar outflow, and the radio\njet. The relative proper motions of the H$_2$O masers in UCHII-W1 are likely\nnot driven by the jets of MM1B protostar but by MM3-UCHII. Overall, the\npost-accretion burst relative proper motions of the H$_2$O masers trace shocks\nof jet motion.\n"}
{"abstract": "  The effects of second-neighbor interactions in Kekule patterned graphene\nelectronic properties are studied starting from a tight-binding Hamiltonian.\nThereafter, a low-energy effective Hamiltonian is obtained by projecting the\nhigh energy bands at the Gamma point into the subspace defined by the Kekule\nwave vector. The spectrum of the low energy Hamiltonian is in excellent\nagreement with the one obtained from a numerical diagonalization of the full\ntight-binding Hamiltonian. The main effect of the second-neighbour interaction\nis that a set of bands gains an effective mass and a shift in energy, thus\nlifting the degeneracy of the conduction bands at the Dirac point. This band\nstructure is akin to a spin-one Dirac cone, a result expected for honeycomb\nlattices with a distinction between one third of the atoms in one sublattice.\nFinally, we present a study of Kekule patterned graphene nanoribbons. This\nshows that the previous effects are enhanced as the width decreases. Moreover,\nedge states become dispersive, as expected due to second neighbors interaction,\nbut here the Kek-Y bond texture results in an hybridization of both edge\nstates. The present study shows the importance of second neighbors in realistic\nmodels of Kekule patterned graphene, specially at surfaces.\n"}
{"abstract": "  When can a unimodular random planar graph be drawn in the Euclidean or the\nhyperbolic plane in a way that the distribution of the random drawing is\nisometry-invariant? This question was answered for one-ended unimodular graphs\nin earlier work of Benjamini and Tim\\'ar, using the fact that such graphs\nautomatically have locally finite (simply connected) drawings into the plane.\nFor the case of graphs with multiple ends the question was left open. We\nprovide a graph theoretic characterization of graphs that have a locally finite\nembedding into the plane. Then we prove that such graphs do have a locally\nfinite invariant embedding into the Euclidean or the hyperbolic plane,\ndepending on whether the graph is amenable or not.\n"}
{"abstract": "  Consider a multitype coalescent process in which each block has a colour in\n$\\{1,\\ldots,d\\}$. Individual blocks may change colour, and some number of\nblocks of various colours may merge to form a new block of some colour. We show\nthat if the law of a multitype coalescent process is invariant under\npermutations of blocks of the same colour, has consistent Markovian\nprojections, and has asychronous mergers, then it is a multitype\n$\\Lambda$-coalescent: a process in which single blocks may change colour, two\nblocks of like colour may merge to form a single block of that colour, or large\nmergers across various colours happen at rates governed by a $d$-tuple of\nmeasures on the unit cube $[0,1]^d$. We go on to identify when such processes\ncome down from infinity. Our framework generalises Pitman's celebrated\nclassification theorem for singletype coalescent processes, and provides a\nunifying setting for numerous examples that have appeared in the literature\nincluding the seed-bank model, the island model and the coalescent structure of\ncontinuous-state branching processes.\n"}
{"abstract": "  In strong-field physics experiments with intense lasers, it is of paramount\nimportance to single-shot diagnose the temporal contrast between laser pulse\npeak and its noise pedestal. This allows fast optimization of pulse contrast\nand meaningful comparison with theory for each pulse shot, and it can help new\noutcomes from clean laser-plasma interactions. Thus far, high contrast ratios\nup to ~10^10, required by present petawatt (PW) class lasers, have been\naccessible in both generation and single-shot characterization. However,\nultrahigh contrast ~10^13, required by the planned 200-PW lasers, challenges\nintense laser technology and remains an open question. This paper reports on\nthe first demonstration of such an ultrahigh-contrast measurement by adapting\nsingle-shot cross-correlator (SSCC). We introduce an ultrafast method that\nenables to determine the SSCC detection limit. Our strategy mimics the test\nlaser having known ultrahigh contrast in the measurement frame of time-to-space\nmapping. The ultimate contrast-measurement limit of 10^13 is achieved, which\ncorresponds to the highest pulse intensity set by SSCC damage threshold and the\nlowest noise pedestal set by single-photon detection. As a consequence, photon\nnoise in the detection is observed and increases as the noise pedestal reduces.\nThe demonstrated measurement ability at the photon noise limit is applied to a\nhigh-contrast laser system based on second-harmonic generation and optical\nparametric chirped-pulse amplification, suggesting accessible of ultrahigh\ncontrast pulses.\n"}
{"abstract": "  In the late 1990's, observations of 93 Type Ia supernovae were analysed in\nthe framework of the FLRW cosmology assuming these to be `standard(isable)\ncandles'. It was thus inferred that the Hubble expansion rate is accelerating\nas if driven by a positive Cosmological Constant $\\Lambda$. This is still the\nonly direct evidence for the `dark energy' that is the dominant component of\nthe standard $\\Lambda$CDM cosmological model. Other data such as BAO, CMB\nanisotropies, stellar ages, the rate of structure growth, etc are all\n`concordant' with this model but do not provide independent evidence for\naccelerated expansion. Analysis of a larger sample of 740 SNe Ia shows that\nthese are not quite standard candles, and highlights the \"corrections\" applied\nto analyse the data in the FLRW framework. The latter holds in the reference\nframe in which the CMB is isotropic, whereas observations are made in our\nheliocentric frame in which the CMB has a large dipole anisotropy. This is\nassumed to be of kinematic origin i.e. due to our non-Hubble motion driven by\nlocal inhomogeneity in the matter distribution. The $\\Lambda$CDM model predicts\nhow this peculiar velocity should fall off as the averaging scale is raised and\nthe universe becomes sensibly homogeneous. However observations of the local\n`bulk flow' are inconsistent with this expectation and convergence to the CMB\nframe is not seen. Moreover the kinematic interpretation implies a\ncorresponding dipole in the sky distribution of high redshift quasars, which is\nrejected by observations at 4.9$\\sigma$. The acceleration of the Hubble\nexpansion rate is also anisotropic at 3.9$\\sigma$ and aligned with the bulk\nflow. Thus dark energy may be an artefact of analysing data assuming that we\nare idealised observers in an FLRW universe, when in fact the real universe is\ninhomogeneous and anisotropic out to distances large enough to impact on\ncosmological analyses.\n"}
{"abstract": "  Li metal is an ideal anode material for use in state-of-the-art secondary\nbatteries. However, Li-dendrite growth is a safety concern and results in low\ncoulombic efficiency, which significantly restricts the commercial application\nof Li secondary batteries. Unfortunately, the Li deposition (growth) mechanism\nis poorly understood on the atomic scale. Here, we used machine learning to\nconstruct a Li potential model with quantum-mechanical computational accuracy.\nMolecular dynamics simulations in this study with this model revealed two\nself-healing mechanisms in a large Li-metal system, viz. surface self-healing\nand bulk self-healing, and identified three Li-dendrite morphologies under\ndifferent conditions, viz. \"needle\", \"mushroom\", and \"hemisphere\". Finally, we\nintroduce the concepts of local current density and variance in local current\ndensity to supplement the critical current density when evaluating the\nprobability of self-healing.\n"}
{"abstract": "  Lockdowns imposed in most of the countries were lifted following a decline in\nthe COVID-19 cases towards May-June 2020. A recent surge (second wave) in the\nCOVID-19 cases in Europe and other temperate countries as compared to the\ntropical regions suggests the likely role of solar radiation. We hypothesized\nthat ultraviolet radiation's effect might be a significant factor moderating\nthe spread of the COVID-19 across countries. Regression analysis was done for\nthe UV radiation data for seven hotspot cities (New Delhi, Mumbai, Milan,\nMadrid, New York, Melbourne and Sydney) with the daily COVID-19 cases. Global\nerythemal UV radiation values were lower during winter and higher during\nsummer. In general, the daily new COVID-19 cases registered were higher during\nthe winter months having low UV radiation dose (0.5-3.7 kJ m-2). Cases began to\ndecline with summer onset that corresponded to increased UV radiation (2.5-6.7\nkJ m-2). Our studies suggest that the natural UV radiation could be a strong\ndetermining factor moderating the spread of COVID-19 cases. The importance of\nUV radiation in natural sunlight as a disinfectant for SARS-CoV-2 cannot be\nignored since the lockdowns were lifted; further, it can be considered as a\nfactor for COVID management.\n"}
{"abstract": "  We present a system to measure the ripeness of fruit with a hyperspectral\ncamera and a suitable deep neural network architecture. This architecture did\noutperform competitive baseline models on the prediction of the ripeness state\nof fruit. For this, we recorded a data set of ripening avocados and kiwis,\nwhich we make public. We also describe the process of data collection in a\nmanner that the adaption for other fruit is easy. The trained network is\nvalidated empirically, and we investigate the trained features. Furthermore, a\ntechnique is introduced to visualize the ripening process.\n"}
{"abstract": "  We introduce and describe the $2$-category $\\mathsf{Grt}_{\\flat}$ of\nGrothendieck categories and flat morphisms between them. First, we show that\nthe tensor product of locally presentable linear categories $\\boxtimes$\nrestricts nicely to $\\mathsf{Grt}_{\\flat}$. Then, we characterize exponentiable\nobjects with respect to $\\boxtimes$: these are continuous Grothendieck\ncategories. In particular, locally finitely presentable Grothendieck categories\nare exponentiable. Consequently, we have that, for a quasi-compact\nquasi-separated scheme $X$, the category of quasi-coherent sheaves\n$\\mathsf{Qcoh}(X)$ is exponentiable. Finally, we provide a family of examples\nand concrete computations of exponentials.\n"}
{"abstract": "  Recent observations indicated that the photoelectric conversion properties of\nperovskite materials are intimately related to the presence of superlattice\nstructures and other unusual nanoscale features in them. The low dimensional or\nmixed dimensional halide perovskite family are found to be more efficient\nmaterials for device application compared to 3-dimensional halide perovskites.\nThe emergence of perovskite solar cell has revolutionized the solar cell\nindustry because of their flexible architecture and rapidly increased\nefficiency. Tuning the dielectric constant, charge separation are the main\nobjective in designing a photovoltaic device that can be explored using\n2-dimensional perovskite family. Thus, revisiting the fundamental properties of\nperovskite crystals could reveal further possibilities for recognizing these\nimprovements towards device functionality. In this context, this review\ndiscusses the material properties of 2-dimensional halide perovskite and\nrelated optoelectronic devices aiming particularly for solar cell application.\n"}
{"abstract": "  Let $K$ be a compact convex set and $m$ be a positive integer. The covering\nfunctional of $K$ with respect to $m$ is the smallest $\\lambda\\in[0,1]$ such\nthat $K$ can be covered by $m$ translates of $\\lambda K$. Estimations of the\ncovering functionals of convex hulls of two or more compact convex sets are\npresented. It is proved that, if a three-dimensional convex body $K$ is the\nconvex hull of two compact convex sets having no interior points, then the\nleast number $c(K)$ of smaller homothetic copies of $K$ needed to cover $K$ is\nnot greater than $8$ and $c(K)=8$ if and only if $K$ is a parallelepiped.\n"}
{"abstract": "  Experimental and theoretical studies of colloidal nanoparticles have\nprimarily focused on accurate characterization and simulation of observable\ncharacteristics, such as resonant wavelength. In this Letter, we tackle the\noptimal design of colloidal-nanoparticle ensembles: what is the largest\npossible optical response, which designs might achieve them, and can such\nresponse be experimentally demonstrated? We combine theory and experiment to\nanswer each of these questions. We derive general bounds on the maximum\ncross-sections per volume, and we apply an analytical antenna model to show\nthat resonant nanorods should nearly achieve such bounds. We use a modified\nseed-mediated synthesis approach to synthesize ensembles of gold nanorods with\nsmall polydispersity, i.e., small variations in size and aspect ratio.\nPolydispersity is the key determinant of how closely such ensembles can\napproach their respective bounds yet is difficult to characterize\nexperimentally without near-field measurements. We show that a certain\n\"extinction metric,\" connecting extinction cross-section per volume with the\nradiative efficiencies of the nanoparticles, offers a quantitative prediction\nof polydispersity via quantities that can be rapidly measured with far-field\ncharacterization tools. Our predictions apply generally across all plasmonic\nmaterials and offers a roadmap to the largest possible optical response of\nnanoparticle ensembles.\n"}
{"abstract": "  The problem of adaptive Kalman filtering for a discrete observable linear\ntime-varying system with unknown noise covariance matrices is addressed in this\npaper. The measurement difference autocovariance method is used to formulate a\nlinear least squares cost function containing the measurements and the process\nand measurement noise covariance matrices. Subsequently, a Riemannian\ntrust-region optimization approach is designed to minimize the least squares\ncost function and ensure symmetry and positive definiteness for the estimates\nof the noise covariance matrices. The noise covariance matrix estimates, under\nsufficient excitation of the system, are shown to converge to their unknown\ntrue values. Saliently, the exponential stability and convergence guarantees\nfor the proposed adaptive Kalman filter to the optimal Kalman filter with known\nnoise covariance matrices is shown to be achieved under the relatively mild\nassumptions of uniform observability and uniform controllability. Numerical\nsimulations on a linear time-varying system demonstrate the effectiveness of\nthe proposed adaptive filtering algorithm.\n"}
{"abstract": "  We establish fundamental limits of tracking a moving target over the unit\ncube under the framework of 20 questions with measurement-dependent noise. In\nthis problem, there is an oracle who knows the instantaneous location of a\ntarget. Our task is to query the oracle as few times as possible to accurately\nestimate the trajectory of the moving target, whose initial location and\nvelocity is \\emph{unknown}. We study the case where the oracle's answer to each\nquery is corrupted by random noise with query-dependent discrete distribution.\nIn our formulation, the performance criterion is the resolution, which is\ndefined as the maximal absolute value between the true location and estimated\nlocation at each discrete time during the searching process. We are interested\nin the minimal resolution of any non-adaptive searching procedure with a finite\nnumber of queries and derive approximations to this optimal resolution via the\nsecond-order asymptotic analysis.\n"}
{"abstract": "  Given a $t$-$(v, k, \\lambda)$ design, $\\mathcal{D}=(X,\\mathcal{B})$, a\nzero-sum $n$-flow of $\\mathcal{D}$ is a map $f : \\mathcal{B}\\longrightarrow\n\\{\\pm1,\\ldots, \\pm(n-1)\\}$ such that for any point $x\\in X$, the sum of $f$\nover all blocks incident with $x$ is zero. For a positive integer $k$, we find\na zero-sum $k$-flow for an STS$(u w)$ and for an STS$(2v+7)$ for $v\\equiv\n1~(\\mathrm{mod}~4)$, if there are STS$(u)$, STS$(w)$ and STS$(v)$ such that the\nSTS$(u)$ and STS$(v)$ both have a zero-sum $k$-flow. In 2015, it was\nconjectured that for $v>7$ every STS$(v)$ admits a zero-sum $3$-flow. Here, it\nis shown that many cyclic STS$(v)$ have a zero-sum $3$-flow. Also, we\ninvestigate the existence of zero-sum flows for some Steiner quadruple systems.\n"}
{"abstract": "  Trusted computing defines how to securely measure, store, and verify the\nintegrity of software controlling a computer. One of the major challenges that\nmake them hard to be applied in practice is the issue with software updates.\nSpecifically, an operating system update causes the integrity violation because\nit changes the well-known initial state trusted by remote verifiers, such as\nintegrity monitoring systems. Consequently, the integrity monitoring of remote\ncomputers becomes unreliable due to the high amount of false positives. We\naddress this problem by adding an extra level of indirection between the\noperating system and software repositories. We propose a trusted software\nrepository (TSR), a secure proxy that overcomes the shortcomings of previous\napproaches by sanitizing software packages. Sanitization consists of modifying\nunsafe installation scripts and adding digital signatures in a way software\npackages can be installed in the operating system without violating its\nintegrity. TSR leverages shielded execution, i.e., Intel SGX, to achieve\nconfidentiality and integrity guarantees of the sanitization process. TSR is\ntransparent to package managers, and requires no changes in the software\npackages building and distributing processes. Our evaluation shows that running\nTSR inside SGX is practical; since it induces only ~1.18X performance overhead\nduring package sanitization compared to the native execution without SGX. TSR\nsupports 99.76% of packages available in the main and community repositories of\nAlpine Linux while increasing the total repository size by 3.6%.\n"}
{"abstract": "  During magnetic reconnection in collisionless space plasma, the electron\nfluid decouples from the magnetic field within narrow current layers, and\ntheoretical models for this process can be distinguished in terms of their\npredicted current layer widths. From theory, the off-diagonal stress in the\nelectron pressure tensor is related to thermal non-circular orbit motion of\nelectrons around the magnetic field lines. This stress becomes significant when\nthe width of the reconnecting current layer approaches the small characteristic\nlength scale of the electron motion. To aid in situ spacecraft and numerical\ninvestigations of reconnection, the structure of the electron diffusion region\nis here investigated using the Terrestrial Reconnection EXperiment (TREX). In\nagreement with the closely matched kinetic simulations, laboratory observations\nreveal the presence of electron-scale current layer widths. Although the layers\nare modulated by a current-driven instability, 3D simulations demonstrate that\nit is the off-diagonal stress that is responsible for breaking the frozen-in\ncondition of the electron fluid.\n"}
{"abstract": "  Here, we have analysed a GARCH(1,1) model with the aim to fit higher order\nmoments for different companies' stock prices. When we assume a gaussian\nconditional distribution, we fail to capture any empirical data when fitting\nthe first three even moments of financial time series. We show instead that a\ndouble gaussian conditional probability distribution better captures the higher\norder moments of the data. To demonstrate this point, we construct regions\n(phase diagrams), in the fourth and sixth order standardised moment space,\nwhere a GARCH(1,1) model can be used to fit these moments and compare them with\nthe corresponding moments from empirical data for different sectors of the\neconomy. We found that the ability of the GARCH model with a double gaussian\nconditional distribution to fit higher order moments is dictated by the time\nwindow our data spans. We can only fit data collected within specific time\nwindow lengths and only with certain parameters of the conditional double\ngaussian distribution. In order to incorporate the non-stationarity of\nfinancial series, we assume that the parameters of the GARCH model have time\ndependence.\n"}
{"abstract": "  Over the last few years, we have witnessed tremendous progress on many\nsubtasks of autonomous driving, including perception, motion forecasting, and\nmotion planning. However, these systems often assume that the car is accurately\nlocalized against a high-definition map. In this paper we question this\nassumption, and investigate the issues that arise in state-of-the-art autonomy\nstacks under localization error. Based on our observations, we design a system\nthat jointly performs perception, prediction, and localization. Our\narchitecture is able to reuse computation between both tasks, and is thus able\nto correct localization errors efficiently. We show experiments on a\nlarge-scale autonomy dataset, demonstrating the efficiency and accuracy of our\nproposed approach.\n"}
{"abstract": "  Berezinskii-Kosterlitz-Thouless transition of the classical XY model is\nre-investigated, combining the Tensor Network Renormalization (TNR) and the\nLevel Spectroscopy method based on the finite-size scaling of the Conformal\nField Theory. By systematically analyzing the spectrum of the transfer matrix\nof the systems of various moderate sizes which can be accurately handled with a\nfinite bond dimension, we determine the critical point removing the logarithmic\ncorrections. This improves the accuracy by an order of magnitude over previous\nstudies including those utilizing TNR. Our analysis also gives a visualization\nof the celebrated Kosterlitz Renormalization Group flow based on the numerical\ndata.\n"}
{"abstract": "  Many popular learning-rate schedules for deep neural networks combine a\ndecaying trend with local perturbations that attempt to escape saddle points\nand bad local minima. We derive convergence guarantees for bandwidth-based\nstep-sizes, a general class of learning rates that are allowed to vary in a\nbanded region. This framework includes many popular cyclic and non-monotonic\nstep-sizes for which no theoretical guarantees were previously known. We\nprovide worst-case guarantees for SGD on smooth non-convex problems under\nseveral bandwidth-based step sizes, including stagewise $1/\\sqrt{t}$ and the\npopular step-decay (constant and then drop by a constant), which is also shown\nto be optimal. Moreover, we show that its momentum variant converges as fast as\nSGD with the bandwidth-based step-decay step-size. Finally, we propose novel\nstep-size schemes in the bandwidth-based family and verify their efficiency on\nseveral deep neural network training tasks.\n"}
{"abstract": "  In this letter, we propose a whole-body control strategy for humanoid robots\nin multi-contact settings that enables switching between fixed and sliding\ncontacts under active balance. We compute, in real-time, a safe center-of-mass\nposition and wrench distribution of the contact points based on the Chebyshev\ncenter. Our solution is formulated as a quadratic programming problem without a\npriori computation of balance regions. We assess our approach with experiments\nhighlighting switches between fixed and sliding contact modes in multi-contact\nconfigurations. A humanoid robot demonstrates such contact interchanges from\nfully-fixed to multi-sliding and also shuffling of the foot. The scenarios\nillustrate the performance of our control scheme in achieving the desired\nforces, CoM position attractor, and planned trajectories while actively\nmaintaining balance.\n"}
{"abstract": "  It is known that the membership in a given reproducing kernel Hilbert space\n(RKHS) of the samples of a Gaussian process $X$ is controlled by a certain\nnuclear dominance condition. However, it is less clear how to identify a\n\"small\" set of functions (not necessarily a vector space) that contains the\nsamples. This article presents a general approach for identifying such sets. We\nuse scaled RKHSs, which can be viewed as a generalisation of Hilbert scales, to\ndefine the sample support set as the largest set which is contained in every\nelement of full measure under the law of $X$ in the $\\sigma$-algebra induced by\nthe collection of scaled RKHS. This potentially non-measurable set is then\nshown to consist of those functions that can be expanded in terms of an\northonormal basis of the RKHS of the covariance kernel of $X$ and have their\nsquared basis coefficients bounded away from zero and infinity, a result\nsuggested by the Karhunen-Lo\\`{e}ve theorem.\n"}
{"abstract": "  The use of Internet of Things (IoT) devices in homes and the immediate\nproximity of an individual communicates to create Personal IoT (PIoT) networks.\nThe exploratory study of PIoT is in its infancy, which will explore the\nexpansion of new use cases, service requirements, and the proliferation of PIoT\ndevices. This article provides a big picture of PIoT architecture, vision, and\nfuture research scope.\n"}
{"abstract": "  The nonlinear optical response of materials is the foundation upon which\napplications such as frequency conversion, all-optical signal processing,\nmolecular spectroscopy, and nonlinear microscopy are built. However, the\nutility of all such parametric nonlinear optical processes is hampered by\nphase-matching requirements. Quasi-phase-matching, birefringent phase matching,\nand higher-order-mode phase matching have all been developed to address this\nconstraint, but the methods demonstrated to date suffer from the inconvenience\nof only being phase-matched for a single, specific arrangement of beams,\ntypically co-propagating, resulting in cumbersome experimental configurations\nand large footprints for integrated devices. Here, we experimentally\ndemonstrate that these phase-matching requirements may be satisfied in a\nparametric nonlinear optical process for multiple, if not all, configurations\nof input and output beams when using low-index media. Our measurement\nconstitutes the first experimental observation of direction-independent phase\nmatching for a medium sufficiently long for phase matching concerns to be\nrelevant. We demonstrate four-wave mixing from spectrally distinct co- and\ncounter-propagating pump and probe beams, the backward-generation of a\nnonlinear signal, and excitation by an out-of-plane probe beam. These results\nexplicitly show that the unique properties of low-index media relax traditional\nphase-matching constraints, which can be exploited to facilitate nonlinear\ninteractions and miniaturize nonlinear devices, thus adding to the established\nexceptional properties of low-index materials.\n"}
{"abstract": "  Three-dimensional geophysical fluids support both internal and\nboundary-trapped waves. To obtain the normal modes in such fluids we must solve\na differential eigenvalue problem for the vertical structure (for simplicity,\nwe only consider horizontally periodic domains). If the boundaries are\ndynamically inert (e.g., rigid boundaries in the Boussinesq internal wave\nproblem, flat boundaries in the quasigeostrophic Rossby wave problem) the\nresulting eigenvalue problem typically has a Sturm-Liouville form and the\nproperties of such problems are well-known. However, when restoring forces are\nalso present at the boundaries, then the equations of motion contain a\ntime-derivative in the boundary conditions and this leads to an eigenvalue\nproblem where the eigenvalue correspondingly appears in the boundary\nconditions. In certain cases, the eigenvalue problem can be formulated as an\neigenvalue problem in the Hilbert space $L^2\\oplus \\mathbb{C}$ and this theory\nis well-developed. Less explored is the case when the eigenvalue problem takes\nplace in a Pontryagin space, as in the Rossby wave problem over sloping\ntopography. This article develops the theory of such problems and explores the\nproperties of wave problems with dynamically-active boundaries. The theory\nallows us to solve the initial value problem for quasigeostrophic Rossby waves\nin a region with sloping bottom (we also apply the theory to two Boussinesq\nproblems with a free-surface). For a step-function perturbation at a\ndynamically-active boundary, we find that the resulting time-evolution consists\nof waves present in proportion to their projection onto the dynamically-active\nboundary.\n"}
{"abstract": "  Light front wave functions motivated by holographic constructions are used to\nstudy Bloom-Gilman duality of deep inelastic scattering. Separate expressions\nfor structure functions in terms of quark and hadronic degrees of freedom are\npresented, with a goal of relating the two expressions. A two-parton model is\ndefined and resonance transition form factors are computed using previously\nderived light front wave functions. A new form of global duality is derived\nfrom the valence quark-number sum rule. Using a complete set of hadronic states\nis necessary for this new global duality to be achieved. Previous original work\ndoes not provide such a set. This is remedied by amending the model to include\na longitudinal confining potential, and the resulting complete set is\nsufficient to carry out the study of Bloom-Gilman duality. Expressions for\ntransition form factors are obtained and all are shown to fall asymptotically\nas 1/Q2. The Feynman mechanism dominates the asymptotic behavior of the model.\nThese transition form factors are used to assess the validity of the global and\nlocal duality sum rules, with the result that both neither are satisfied.\nEvaluations of the hadronic expression for q(x,Q2) provide more details about\nthis lack. This result shows that the observed validity of both global and\nlocal forms of duality for deep inelastic scattering must be related to a\nfeature of QCD that is deeper than completeness. Our simple present model\nsuggests a prediction that Bloom-Gilman duality would not be observed if deep\ninelastic scattering experiments were to be made on the pion. The underlying\norigin of the duality phenomenon in deep inelastic scattering is deeply buried\nwithin the confinement aspects of QCD, and remains a mystery.\n"}
{"abstract": "  This paper deals with Poisson approximation to weighted sums of\ndouble-indexed independent random variables using Stein's method. The obtained\nresults improved the existing bounds. As applications, limit theorems for sums\nof weighted Bernoulli and geometric random variables are discussed.\n"}
{"abstract": "  In this work, we introduce DGO, a new MATLAB toolbox for derivative-free\nglobal optimization. DGO collects various deterministic derivative-free\nDIRECT-type algorithms for box-constrained, generally-constrained, and problems\nwith hidden constraints. Each sequential algorithm is implemented in two\ndifferent ways: using static and dynamic data structures for more efficient\ninformation storage and organization. Furthermore, parallel schemes are applied\nto some promising algorithms within DGO. The toolbox is equipped with a\ngraphical user interface (GUI), which ensures the user-friendly use of all\nfunctionalities available in DGO. Available features are demonstrated in\ndetailed computational studies using a created comprehensive library of global\noptimization problems. Additionally, eleven classical engineering design\nproblems are used to illustrate the potential of DGO to solve challenging\nreal-world problems.\n"}
{"abstract": "  We construct a proof of the second law of thermodynamics in an arbitrary\ndiffeomorphism invariant theory of gravity working within the approximation of\nlinearized dynamical fluctuations around stationary black holes. We achieve\nthis by establishing the existence of an entropy current defined on the horizon\nof the dynamically perturbed black hole in such theories. By construction, this\nentropy current has non-negative divergence, suggestive of a mechanism for the\ndynamical black hole to approach a final equilibrium configuration via entropy\nproduction as well as the spatial flow of it on the null horizon. This enables\nus to argue for the second law in its strongest possible form, which has a\nmanifest locality at each space-time point. We explicitly check that the form\nof the entropy current that we construct in this paper exactly matches with\npreviously reported expressions computed considering specific four derivative\ntheories of higher curvature gravity. Using the same set up we also provide an\nalternative proof of the physical process version of the first law applicable\nto arbitrary higher derivative theories of gravity.\n"}
{"abstract": "  Presheaves on a small category are well-known to correspond via a category of\nelements construction to ordinary discrete fibrations over that same small\ncategory. Work of R. Par\\'e proposes that presheaves on a small double category\nare certain lax functors valued in the double category of sets with spans. This\npaper isolates the discrete fibration concept corresponding to this presheaf\nnotion and shows that the category of elements construction introduced by\nPar\\'e leads to an equivalence of virtual double categories.\n"}
{"abstract": "  We present the results of a study of the prospect of detecting habitable\nTrojan planets in the Kepler Habitable Zone circumbinary planetary systems\n(Kepler-16, -47, -453, -1647, -1661). We integrated the orbits of 10,000\nseparate N-body systems (N=4,6), each with a one Earth-mass body in a randomly\nselected orbit near the L4 and L5 Lagrangian points of the host HZ circumbinary\nplanet. We find that stable Trojan planets are restricted to a narrow range of\nsemimajor axes in all five systems and limited to small eccentricities in\nKepler-16, -47, and -1661. To assess the prospect of the detection of these\nhabitable Trojan planets, we calculated the amplitudes of the variations they\ncause in the transit timing of their host bodies. Results show that the mean\namplitudes of the transit timing variations (TTVs) correlate with the mass of\nthe transiting planet and range from 70 minutes for Kepler-16b to 390 minutes\nfor Kepler-47c. Our analysis indicates that the TTVs of the circumbinary\nplanets caused by these Trojan bodies fall within the detectable range of\ntiming precision obtained from the Kepler telescope's long-cadence data. The\nlatter points to Kepler data as a viable source to search for habitable Trojan\nplanets.\n"}
{"abstract": "  Randomized methods are becoming increasingly popular in numerical linear\nalgebra. However, few attempts have been made to use them in developing\npreconditioners. Our interest lies in solving large-scale sparse symmetric\npositive definite linear systems of equations where the system matrix is\npreordered to doubly bordered block diagonal form (for example, using a nested\ndissection ordering). We investigate the use of randomized methods to construct\nhigh quality preconditioners. In particular, we propose a new and efficient\napproach that employs Nystr\\\"om's method for computing low rank approximations\nto develop robust algebraic two-level preconditioners. Construction of the new\npreconditioners involves iteratively solving a smaller but denser symmetric\npositive definite Schur complement system with multiple right-hand sides.\nNumerical experiments on problems coming from a range of application areas\ndemonstrate that this inner system can be solved cheaply using block conjugate\ngradients and that using a large convergence tolerance to limit the cost does\nnot adversely affect the quality of the resulting Nystr\\\"om--Schur two-level\npreconditioner.\n"}
{"abstract": "  In this work, we consider the smoothing effect of Vlasov-Poisson-Landau\nsystem for both hard and soft potential. In particular, we prove that any\nclassical solutions becomes immediately smooth with respect to all variables.\nWe also give a proof on the global existence to Vlasov-Poisson-Landau system\nwith optimal large time decay. These results give the regularity to\nVlasov-Poisson-Landau system. The proof is based on the time-weighted energy\nmethod building upon the pseudo-differential calculus.\n"}
{"abstract": "  In this paper, we perform an analysis of 13 outflows in the Cygnus X\nstar-forming region. We use the James Clerk Maxwell Telescope observations of\n$^{13}$CO(3-2) and C$^{18}$O(3-2) molecular emission lines combined with\narchival $^{12}$CO(3-2) data. Using these new observations, we measure the\nmechanical properties of the outflows, and identify the associated protostars,\nfinding their properties consistent with previous surveys of outflows\nthroughout the Milky Way. Finally, we develop and test a method to measure the\nsame properties using the existing $^{12}$CO(3-2) line data alone, finding the\nproperties agree to within a factor of 2.\n"}
{"abstract": "  Fekete, Jord\\'an and Kaszanitzky [4] characterised the graphs which can be\nrealised as 2-dimensional, infinitesimally rigid, bar-joint frameworks in which\ntwo given vertices are coincident. We formulate a conjecture which would extend\ntheir characterisation to an arbitrary set T of vertices and verify our\nconjecture when |T| = 3.\n"}
{"abstract": "  In this article, we investigate meandric systems having one shallow side: the\narch configuration on that side has depth at most two. This class of meandric\nsystems was introduced and extensively examined by I. P. Goulden, A. Nica, and\nD. Puder in 2020. Shallow arch configurations are in bijection with the set of\ninterval partitions. We study meandric systems by using moment-cumulant\ntransforms for non-crossing and interval partitions, corresponding to the\nnotions of free and boolean independence, respectively, in non-commutative\nprobability. We obtain formulas for the generating series of different classes\nof meandric systems with one shallow side, by explicitly enumerating the\nsimpler, irreducible objects. In addition, we propose random matrix models for\nthe corresponding meandric polynomials, which can be described in the language\nof quantum information theory, in particular that of quantum channels.\n"}
{"abstract": "  Just-in-time adaptive interventions (JITAIs) are time-varying adaptive\ninterventions that use frequent opportunities for the intervention to be\nadapted--weekly, daily, or even many times a day. The micro-randomized trial\n(MRT) has emerged for use in informing the construction of JITAIs. MRTs can be\nused to address research questions about whether and under what circumstances\nJITAI components are effective, with the ultimate objective of developing\neffective and efficient JITAI. The purpose of this article is to clarify why,\nwhen, and how to use MRTs; to highlight elements that must be considered when\ndesigning and implementing an MRT; and to review primary and secondary analyses\nmethods for MRTs. We briefly review key elements of JITAIs and discuss a\nvariety of considerations that go into planning and designing an MRT. We\nprovide a definition of causal excursion effects suitable for use in primary\nand secondary analyses of MRT data to inform JITAI development. We review the\nweighted and centered least-squares (WCLS) estimator which provides consistent\ncausal excursion effect estimators from MRT data. We describe how the WCLS\nestimator along with associated test statistics can be obtained using standard\nstatistical software such as R (R Core Team, 2019). Throughout we illustrate\nthe MRT design and analyses using the HeartSteps MRT, for developing a JITAI to\nincrease physical activity among sedentary individuals. We supplement the\nHeartSteps MRT with two other MRTs, SARA and BariFit, each of which highlights\ndifferent research questions that can be addressed using the MRT and\nexperimental design considerations that might arise.\n"}
{"abstract": "  For a prime number $\\ell$ we say that an oriented pro-$\\ell$ group\n$(G,\\theta)$ has the Bogomolov-Positselski property if the kernel of the\ncanonical projection on its maximal $\\theta$-abelian quotient\n$\\pi^{ab}_{G,\\theta}\\colon G\\to G(\\theta)$ is a free pro-$\\ell$ group contained\nin the Frattini subgroup of $G$. We show that oriented pro-$\\ell$ groups of\nelementary type have the Bogomolov-Positselski property. This shows that\nEfrat's Elementary Type Conjecture implies a positive answer to Positselski's\nversion of Bogomolov's Conjecture on maximal pro-$\\ell$ Galois groups of a\nfield $K$ in case that $K^\\times/(K^\\times)^\\ell$ is finite. Secondly, it is\nshown that for an $H^\\bullet$-quadratic oriented pro-$\\ell$ group $(G,\\theta)$\nthe Bogomolov-Positselski property can be expressed by the injectivity of the\ntransgression map $d_2^{2,1}$ in the Hochschild-Serre spectral sequence.\n"}
{"abstract": "  Aditya-L1 is India's first solar mission with Visible Emission Line\nCoronagraph (VELC) consisting of three spectral channels taking high-resolution\nspectroscopic observations of the inner corona up to 1.5 Rsun at 5303 \\AA, 7892\n\\AA, and 10747 \\AA. In this work, we present the strategy for the slit width\noptimization for the VELC using synthetic line profiles by taking into account\nthe instrument characteristics and coronal conditions for log(T) varying from 6\nto 6.5. The synthetic profiles are convolved with simulated instrumental\nscattered light and noise to estimate the signal-to-noise ratio (SNR), which\nwill be crucial to design the future observation plans. We find that the\noptimum slit width for VELC turns out to be 50 microns providing sufficient SNR\nfor observations in different solar conditions. We also analyzed the effect of\nplasma temperature on the SNR at different heights in the VELC field of view\nfor the optimized slit width. We also studied the expected effect of the\npresence of a CME on the spectral channel observations. This analysis will help\nto plan the science observations of VELC in different solar conditions.\n"}
{"abstract": "  We study the lattice QCD $\\Lambda_c N$ phase shifts for the $^3S_1-{}^3D_1$\ncoupled channel using both the leading order covariant chiral effective theory\nand the next-to-leading order non-relativistic chiral effective field theory.\nWe show that although it is possible to describe simultaneously the $^3S_1$ and\n$^3D_1$ phase shifts and the inelasticity $\\eta_1$, the fitted energy range is\npretty small, only up to $E_\\mathrm{c.m.}=5$ MeV. This raises concerns\nregarding the consistency between leading/next-to-leading order chiral\neffective field theory and the lattice QCD simulations.\n"}
{"abstract": "  Mesons comprising a beauty quark and a strange quark can oscillate between\nparticle (B0s) and antiparticle (B0sbar) flavour eigenstates, with a frequency\ngiven by the mass difference between heavy and light mass eigenstates, deltams.\nHere we present a measurement of deltams using B0s2DsPi decays produced in\nproton-proton collisions collected with the LHCb detector at the Large Hadron\nCollider. This measurement improves upon the current deltams precision by a\nfactor of two. The oscillation frequency is found to be deltams = 17.7683 +-\n0.0051 +- 0.0032 ps-1 where the first uncertainty is statistical and the second\nsystematic. We combine this result with previous LHCb measurements to determine\ndeltams = 17.7656 +- 0.0057 ps-1.\n"}
{"abstract": "  We present an empirical investigation of a new mapping system based on a\ngraph of panoramic depth images. Panoramic images efficiently capture range\nmeasurements taken by a spinning lidar sensor, recording fine detail on the\norder of a few centimeters within maps of expansive scope on the order of tens\nof millions of cubic meters. The flexibility of the system is demonstrated by\nrunning the same mapping software against data collected by hand-carrying a\nsensor around a laboratory space at walking pace, moving it outdoors through a\ncampus environment at running pace, driving the sensor on a small wheeled\nvehicle on- and off-road, flying the sensor through a forest, carrying it on\nthe back of a legged robot navigating an underground coal mine, and mounting it\non the roof of a car driven on public roads. The full 3D maps are built online\nwith a median update time of less than ten milliseconds on an embedded NVIDIA\nJetson AGX Xavier system.\n"}
{"abstract": "  We aim to reveal the nature of the reddest known substellar companion HD\n206893 B by studying its near-infrared colors and spectral morphology and by\ninvestigating its orbital motion. We fit atmospheric models for giant planets\nand brown dwarfs and perform spectral retrievals with petitRADTRANS and ATMO on\nthe observed GRAVITY, SPHERE, and GPI spectra of HD 206893 B. To recover its\nunusual spectral features, we include additional extinction by high-altitude\ndust clouds made of enstatite grains in the atmospheric model fits. We also\ninfer the orbital parameters of HD 206893 B by combining the $\\sim\n100~\\mu\\text{as}$ precision astrometry from GRAVITY with data from the\nliterature and constrain the mass and position of HD 206893 C based on the Gaia\nproper motion anomaly of the system. The extremely red color and the very\nshallow $1.4~\\mu\\text{m}$ water absorption feature of HD 206893 B can be fit\nwell with the adapted atmospheric models and spectral retrievals. Altogether,\nour analysis suggests an age of $\\sim 3$-$300~\\text{Myr}$ and a mass of $\\sim\n5$-$30~\\text{M}_\\text{Jup}$ for HD 206893 B, which is consistent with previous\nestimates but extends the parameter space to younger and lower-mass objects.\nThe GRAVITY astrometry points to an eccentric orbit ($e =\n0.29^{+0.06}_{-0.11}$) with a mutual inclination of $< 34.4~\\text{deg}$ with\nrespect to the debris disk of the system. While HD 206893 B could in principle\nbe a planetary-mass companion, this possibility hinges on the unknown influence\nof the inner companion on the mass estimate of\n$10^{+5}_{-4}~\\text{M}_\\text{Jup}$ from radial velocity and Gaia as well as a\nrelatively small but significant Argus moving group membership probability of\n$\\sim 61\\%$. However, we find that if the mass of HD 206893 B is $<\n30~\\text{M}_\\text{Jup}$, then the inner companion HD 206893 C should have a\nmass between $\\sim 8$-$15~\\text{M}_\\text{Jup}$.\n"}
{"abstract": "  We mimic random nanowire networks by the homogeneous, isotropic, and random\ndeposition of conductive zero-width sticks onto an insulating substrate. The\nnumber density (the number of objects per unit area of the surface) of these\nsticks is supposed to exceed the percolation threshold, i.e., the system under\nconsideration is a conductor. To identify any current-carrying part (the\nbackbone) of the percolation cluster, we have proposed and implemented a\nmodification of the well-known wall follower algorithm -- one type of maze\nsolving algorithm. The advantage of the modified algorithm is its\nidentification of the whole backbone without visiting all the edges. The\ncomplexity of the algorithm depends significantly on the structure of the graph\nand varies from $O\\left(\\sqrt{N_\\text{V}}\\right)$ to $\\Theta(N_\\text{V})$. The\nalgorithm has been applied to backbone identification in networks with\ndifferent number densities of conducting sticks. We have found that (i) for\nnumber densities of sticks above the percolation threshold, the strength of the\npercolation cluster quickly approaches unity as the number density of the\nsticks increases; (ii) simultaneously, the percolation cluster becomes\nidentical to its backbone plus simplest dead ends, i.e., edges that are\nincident to vertices of degree 1. This behavior is consistent with the\npresented analytical evaluations.\n"}
{"abstract": "  Equational reasoning is among the most important tools that functional\nprogramming provides us. Curiously, relatively less attention has been paid to\nreasoning about monadic programs.\n  In this report we derive a backtracking algorithm for problem specifications\nthat use a monadic unfold to generate possible solutions, which are filtered\nusing a $\\mathit{scanl}$-like predicate. We develop theorems that convert a\nvariation of $\\mathit{scanl}$ to a $\\mathit{foldr}$ that uses the state monad,\nas well as theorems constructing hylomorphism. The algorithm is used to solve\nthe $n$-queens puzzle, our running example. The aim is to develop theorems and\npatterns useful for the derivation of monadic programs, focusing on the\nintricate interaction between state and non-determinism.\n"}
{"abstract": "  We consider compactifications of rank $Q$ E-string theory on a genus zero\nsurface with no punctures but with flux for various subgroups of the\n$\\text{E}_8\\times \\text{SU}(2)$ global symmetry group of the six dimensional\ntheory. We first construct a simple Wess-Zumino model in four dimensions\ncorresponding to the compactification on a sphere with one puncture and a\nparticular value of flux, the cap model. Using this theory and theories\ncorresponding to two punctured spheres with flux, one can obtain a large number\nof models corresponding to spheres with a variety of fluxes. These models\nexhibit interesting IR enhancements of global symmetry as well as duality\nproperties. As an example we will show that constructing sphere models\nassociated to specific fluxes related by an action of the Weyl group of\n$\\text{E}_8$ leads to the S-confinement duality of the $\\text{USp}(2Q)$ gauge\ntheory with six fundamentals and a traceless antisymmetric field. Finally, we\nshow that the theories we discuss possess an $\\text{SU}(2)_{\\text{ISO}}$\nsymmetry in four dimensions that can be naturally identified with the isometry\nof the two-sphere. We give evidence in favor of this identification by\ncomputing the `t Hooft anomalies of the $\\text{SU}(2)_{\\text{ISO}}$ in 4d and\ncomparing them with the predicted anomalies from 6d.\n"}
{"abstract": "  Non-Hermitian systems with aperiodic order display phase transitions that are\nbeyond the paradigm of Hermitian physics. Unfortunately, owing to the\nincommensurability of the potential most of known non-Hermitian models are not\nintegrable. This motivates the search for exactly solvable models, where\nlocalization/delocalization phase transitions, mobility edges in complex plane\nand their topological nature can be unraveled. Here we present an exactly\nsolvable model of quasi crystal, which is a non-pertrurbative non-Hermitian\nextension of a famous integrable model of quantum chaos proposed by Grempel\n{\\it at al.} [Phys. Rev. Lett. {\\bf 49}, 833 (1982)] and dubbed the Maryland\nmodel. Contrary to the Hermitian Maryland model, its non-Hermitian extension\nshows a richer scenario, with a localization-delocalization phase transition\nvia topological mobility edges in complex energy plane.\n"}
{"abstract": "  We use example of the Davey-Stewartson hierarchy to show that in addition to\nthe standard equations given by Lax operator and evolutions of times with\npositive numbers, one can consider time evolutions with negative numbers and\nthe same Lax operator. We derive corresponding Lax pairs and integrable\nequations.\n"}
{"abstract": "  Besides the well-known classification task, these days neural networks are\nfrequently being applied to generate or transform data, such as images and\naudio signals. In such tasks, the conventional loss functions like the mean\nsquared error (MSE) may not give satisfactory results. To improve the\nperceptual quality of the generated signals, one possibility is to increase\ntheir similarity to real signals, where the similarity is evaluated via a\ndiscriminator network. The combination of the generator and discriminator nets\nis called a Generative Adversarial Network (GAN). Here, we evaluate this\nadversarial training framework in the articulatory-to-acoustic mapping task,\nwhere the goal is to reconstruct the speech signal from a recording of the\nmovement of articulatory organs. As the generator, we apply a 3D convolutional\nnetwork that gave us good results in an earlier study. To turn it into a GAN,\nwe extend the conventional MSE training loss with an adversarial loss component\nprovided by a discriminator network. As for the evaluation, we report various\nobjective speech quality metrics such as the Perceptual Evaluation of Speech\nQuality (PESQ), and the Mel-Cepstral Distortion (MCD). Our results indicate\nthat the application of the adversarial training loss brings about a slight,\nbut consistent improvement in all these metrics.\n"}
{"abstract": "  Our Milky Way (MW) has witnessed a series of major accretion events. One of\nthe later additions, Gaia-Enceladus, has contributed a considerable mass to the\ninner Galaxy, but also generously donated to the outer halo. So far,\nassociations with present-day MW globular clusters (GCs) have been chiefly\nbased on their kinematics and ages. Here, we present a chemical abundance study\nof the outer halo (R$_{\\rm GC}$=18 kpc) GC NGC 1261, which has been suggested\nto be an accreted object. We measured 31 species of 29 elements in two stars\nfrom high-resolution Magellan/MIKE spectra and find that the cluster is\nmoderately metal poor, at [Fe/H]=-1.26. NGC 1261 is moderately\n$\\alpha$-enhanced to the 0.3-dex level. While from the small sample alone it is\ndifficult to assert any abundance correlations, the light elements Na,O,Mg, and\nAl differ significantly between the two stars in contrast to the majority of\nother elements with smaller scatter; this argues in favour of multiple\ngenerations of stars coexisting in this GC. Intriguingly for its metallicity,\nNGC 1261 shows heavy element abundances that are consistent with $r$-process\nnucleosynthesis and we discuss their origin in various sites. In particular the\nEu overabundance quantitatively suggests that one single $r$-process event,\nsuch as a neutron-star neutron-star merger or a rare kind of supernova, can be\nresponsible for the stellar enhancement or even the enrichment of the cluster\nwith the excess $r$-material. Its heavy element pattern makes NGC 1261 resemble\nthe moderately enhanced r-I stars that are commonly found in the halo and have\nbeen detected in Gaia-Enceladus as well. Therefore, combining all kinematical,\nage, and chemical evidence we conclude that NGC 1261 is a chemically intriguing\nGC that was born in Gaia-Enceladus and has been subsequently accreted into the\nMW halo. [abridged]\n"}
{"abstract": "  Given a topological ring $R$, we study semitopological $R$-modules, construct\ntheir completions, Bohr and borno modifications. For every topological space\n$X$, we construct the free (semi)topological $R$-module over $X$ and prove that\nfor a $k$-space $X$ its free semitopological $R$-module is a topological\n$R$-module. Also we construct a Tychonoff space $X$ whose free semitopological\n$R$-module is not a topological $R$-module.\n"}
{"abstract": "  Let $G$ be a classical linear algebraic group over an algebraically closed\nfield, and let $\\mathfrak{n}$ denote the subset of nilpotent elements in its\nLie algebra. In this paper we study a partial order on the $G$-orbits in\n$\\mathfrak{n}$ given by taking limits along cocharacters of $G$. This gives\nrise to the so-called accessibility order on the nilpotent orbits. Our main\nresults show that for general and special linear algebras, this new order\ncoincides with the usual dominance order on nilpotent orbits, but for\nsymplectic and orthogonal algebras this is not the case.\n"}
{"abstract": "  We present measurements of the electron helicity asymmetry in quasi-elastic\nproton knockout from $^{2}$H and $^{12}$C nuclei by polarized electrons. This\nasymmetry depends on the fifth structure function, is antisymmetric with\nrespect to the scattering plane, and vanishes in the absence of final-state\ninteractions, and thus it provides a sensitive tool for their study. Our\nkinematics cover the full range in off-coplanarity angle $\\phi_{pq}$, with a\npolar angle $\\theta_{pq}$ coverage up to about 8 degrees. The missing energy\nresolution enabled us to determine the asymmetries for knock-out resulting in\ndifferent states of the residual $^{11}$B system. We find that the helicity\nasymmetry for $p$-shell knockout from $^{12}$C depends on the final state of\nthe residual system and is relatively large (up to $\\approx 0.16$), especially\nat low missing momentum. It is considerably smaller (up to $\\approx 0.01$) for\n$s$-shell knockout from both $^{12}$C and $^2$H. The data for $^2$H are in very\ngood agreement with theoretical calculations, while the predictions for\n$^{12}$C exhibit differences with respect to the data.\n"}
{"abstract": "  We analyze quantum fluctuation effects at the onset of charge or spin density\nwave order in two-dimensional metals with an incommensurate nesting ($2k_F$)\nwave vector connecting two pairs of hot spots on the Fermi surface. We first\ncompute the momentum and frequency dependence of the fermion self-energy near\nthe hot spots to leading order in a perturbation expansion (one loop).\nNon-Fermi liquid behavior with a linear (in energy) quasi-particle decay rate\nand a logarithmically vanishing quasi-particle weight is obtained. The momentum\ndependence of the self-energy entails only finite renormalizations of the Fermi\nvelocity and the Fermi surface curvature at the hot spots. The perturbative\none-loop result is not self-consistent and casts doubt on the stability of the\n$2k_F$ quantum critical point. We construct a self-consistent solution of the\none-loop equations with self-energy feedback, where the quantum critical point\nis stabilized rather than being destroyed by fluctuations, while the non-Fermi\nliquid behavior as found in the perturbative one-loop calculation is confirmed.\n"}
{"abstract": "  We consider the fundamental problems of determining the rooted and global\nedge and vertex connectivities (and computing the corresponding cuts) in\ndirected graphs. For rooted (and hence also global) edge connectivity with\nsmall integer capacities we give a new randomized Monte Carlo algorithm that\nruns in time $\\tilde{O}(n^2)$. For rooted edge connectivity this is the first\nalgorithm to improve on the $\\Omega(n^3)$ time bound in the dense-graph\nhigh-connectivity regime. Our result relies on a simple combination of sampling\ncoupled with sparsification that appears new, and could lead to further\ntradeoffs for directed graph connectivity problems.\n  We extend the edge connectivity ideas to rooted and global vertex\nconnectivity in directed graphs. We obtain a $(1 + \\epsilon)$-approximation for\nrooted vertex connectivity in $\\tilde{O}(nW/\\epsilon)$ time where $W$ is the\ntotal vertex weight (assuming integral vertex weights); in particular this\nyields an $\\tilde{O}(n^2/\\epsilon)$ time randomized algorithm for unweighted\ngraphs. This translates to a $\\tilde{O}(\\kappa nW)$ time exact algorithm where\n$\\kappa$ is the rooted connectivity. We build on this to obtain similar bounds\nfor global vertex connectivity.\n  Our results complement the known results for these problems in the low\nconnectivity regime due to work of Gabow [9] for edge connectivity from 1991,\nand the very recent work of Nanongkai et al. [24] and Forster et al. [7] for\nvertex connectivity.\n"}
{"abstract": "  We prove that every positively-weighted tree T can be realized as the cut\nlocus C(x) of a point x on a convex polyhedron P, with T weights matching C(x)\nlengths. If T has n leaves, P has (in general) n+1 vertices. We show there are\nin fact a continuum of polyhedra P each realizing T for some x on P. Three main\ntools in the proof are properties of the star unfolding of P, Alexandrov's\ngluing theorem, and a cut-locus partition lemma. The construction of P from T\nis surprisingly simple.\n"}
{"abstract": "  The evolution of the atomic structures of the combinatorial library of\nSm-substituted thin film BiFeO3 along the phase transition boundary from the\nferroelectric rhombohedral phase to the non-ferroelectric orthorhombic phase is\nexplored using scanning transmission electron microscopy (STEM). Localized\nproperties including polarization, lattice parameter, and chemical composition\nare parameterized from atomic-scale imaging and their causal relationships are\nreconstructed using a linear non-Gaussian acyclic model (LiNGAM). This approach\nis further extended toward exploring the spatial variability of the causal\ncoupling using the sliding window transform method, which revealed that new\ncausal relationships emerged both at the expected locations, such as domain\nwalls and interfaces, but also at additional regions forming clusters in the\nvicinity of the walls or spatially distributed features. While the exact\nphysical origins of these relationships are unclear, they likely represent\nnanophase separated regions in the morphotropic phase boundaries. Overall, we\npose that an in-depth understanding of complex disordered materials away from\nthermodynamic equilibrium necessitates understanding not only of the generative\nprocesses that can lead to observed microscopic states, but also the causal\nlinks between multiple interacting subsystems.\n"}
{"abstract": "  We prove distality of quantifier-free relations on valued fields with finite\nresidue field. By a result of Chernikov-Galvin-Starchenko, this yields\nSzemer\\'edi-Trotter-like incidence bounds for function fields over finite\nfields. We deduce a version of the Elekes-Szab\\'o theorem for such fields.\n"}
{"abstract": "  Time-series is ubiquitous across applications, such as transportation,\nfinance and healthcare. Time-series is often influenced by external factors,\nespecially in the form of asynchronous events, making forecasting difficult.\nHowever, existing models are mainly designated for either synchronous\ntime-series or asynchronous event sequence, and can hardly provide a synthetic\nway to capture the relation between them. We propose Variational Synergetic\nMulti-Horizon Network (VSMHN), a novel deep conditional generative model. To\nlearn complex correlations across heterogeneous sequences, a tailored encoder\nis devised to combine the advances in deep point processes models and\nvariational recurrent neural networks. In addition, an aligned time coding and\nan auxiliary transition scheme are carefully devised for batched training on\nunaligned sequences. Our model can be trained effectively using stochastic\nvariational inference and generates probabilistic predictions with Monte-Carlo\nsimulation. Furthermore, our model produces accurate, sharp and more realistic\nprobabilistic forecasts. We also show that modeling asynchronous event\nsequences is crucial for multi-horizon time-series forecasting.\n"}
{"abstract": "  In this paper it presents, develops and discusses the existence of a process\nwith long scope memory structure, representing of the independence between the\ndegree of randomness of the traffic generated by the sources and flow pattern\nexhibited by the network. The process existence is presented in term of a new\nalgorithmic that is a variant of the maximum likelihood estimator (MLE) of\nWhittle, for the calculation of the Hurst exponent (H) of self-similar\nstationary second order time series of the flows of the individual sources and\ntheir aggregation. Also, it is discussed the additional problems introduced by\nthe phenomenon of the locality of the Hurst exponent, that appears when the\ntraffic flows consist of diverse elements with different Hurst exponents. The\ninstance is exposed with the intention of being considered as a new and\nalternative approach for modeling and simulating traffic in existing computer\nnetworks.\n"}
{"abstract": "  In the present work we have studied the dynamics of dipolar atom-molecular\nBose Einstein Condensates coupled via Feshbach Resonance in a double well\npotential. We have numerically solved four coupled GP like equations, two for\nleft well and two for right well for this atom molecular coupled system. Our\nnumerical results show that both the long-range dipole-dipole interaction\n(chosen to be positive) and the coherent coupling interaction (which is\npositive for bosons) facilitate the transmission of atoms and molecules from\nleft well to the right well when the population in the right well dominates\nover that in the left well and is trapped for a period of time. Whereas in\nabsence of any one of these interactions probability of transient transmission\ndecreases. However in absence of both the interactions (dipole-dipole and\ncoherent coupling) i.e. when only the repulsive contact interaction is present,\nit leads to self trapping in the left well for a period of time. It is also\nshown that the signature of coherent coupling between atoms and molecules on\nthe density distribution of atoms in the double well potential is present both\nin absence and presence of dipole-dipole interaction.\n"}
{"abstract": "  Like ESG investing, climate change is an important concern for asset managers\nand owners, and a new challenge for portfolio construction. Until now,\ninvestors have mainly measured carbon risk using fundamental approaches, such\nas with carbon intensity metrics. Nevertheless, it has not been proven that\nasset prices are directly impacted by these fundamental-based measures. In this\npaper, we focus on another approach, which consists in measuring the\nsensitivity of stock prices with respect to a carbon risk factor. In our\nopinion, carbon betas are market-based measures that are complementary to\ncarbon intensities or fundamental-based measures when managing investment\nportfolios, because carbon betas may be viewed as an extension or\nforward-looking measure of the current carbon footprint. In particular, we show\nhow this new metric can be used to build minimum variance strategies and how\nthey impact their portfolio construction.\n"}
{"abstract": "  Developers need to perform adequate testing to ensure the quality of\nAutomatic Speech Recognition (ASR) systems. However, manually collecting\nrequired test cases is tedious and time-consuming. Our recent work proposes\nCrossASR, a differential testing method for ASR systems. This method first\nutilizes Text-to-Speech (TTS) to generate audios from texts automatically and\nthen feed these audios into different ASR systems for cross-referencing to\nuncover failed test cases. It also leverages a failure estimator to find\nfailing test cases more efficiently. Such a method is inherently\nself-improvable: the performance can increase by leveraging more advanced TTS\nand ASR systems. So in this accompanying tool demo paper, we devote more\nengineering and propose CrossASR++, an easy-to-use ASR testing tool that can be\nconveniently extended to incorporate different TTS and ASR systems, and failure\nestimators. We also make CrossASR++ chunk texts from a given corpus dynamically\nand enable the estimator to work in a more effective and flexible way. We\ndemonstrate that the new features can help CrossASR++ discover more failed test\ncases. Using the same TTS and ASR systems, CrossASR++ can uncover 26.2% more\nfailed test cases for 4 ASRs than the original tool. Moreover, by simply adding\none more ASR for cross-referencing, we can increase the number of failed test\ncases uncovered for each of the 4 ASR systems by 25.07%, 39.63%, 20.9\\% and\n8.17% respectively. We also extend CrossASR++ with 5 additional failure\nestimators. Compared to worst estimator, the best one can discover 10.41% more\nfailed test cases within the same amount of time.\n"}
{"abstract": "  Existing compartmental models in epidemiology are limited in terms of\noptimizing the resource allocation to control an epidemic outbreak under\ndisease growth uncertainty. In this study, we address this core limitation by\npresenting a multi-stage stochastic programming compartmental model, which\nintegrates the uncertain disease progression and resource allocation to control\nan infectious disease outbreak. The proposed multi-stage stochastic program\ninvolves various disease growth scenarios and optimizes the distribution of\ntreatment centers and resources while minimizing the total expected number of\nnew infections and funerals. We define two new equity metrics, namely infection\nand capacity equity, and explicitly consider equity for allocating treatment\nfunds and facilities over multiple time stages. We also study the multi-stage\nvalue of the stochastic solution (VSS), which demonstrates the superiority of\nthe proposed stochastic programming model over its deterministic counterpart.\nWe apply the proposed formulation to control the Ebola Virus Disease (EVD) in\nGuinea, Sierra Leone, and Liberia of West Africa to determine the optimal and\nfair resource-allocation strategies. Our model balances the proportion of\ninfections over all regions, even without including the infection equity or\nprevalence equity constraints. Model results also show that allocating\ntreatment resources proportional to population is sub-optimal, and enforcing\nsuch a resource allocation policy might adversely impact the total number of\ninfections and deaths, and thus resulting in a high cost that we have to pay\nfor the fairness. Our multi-stage stochastic epidemic-logistics model is\npractical and can be adapted to control other infectious diseases in\nmeta-populations and dynamically evolving situations.\n"}
{"abstract": "  The deformation behaviour of the intermetallic Al$_{2}$Cu-phase was\ninvestigated using atomistic simulations and micropillar compression, where\nslip on the unexpected {211} and {022} slip planes was revealed. Additionally,\nall possible slip systems for the intermetallic phases were further evaluated\nand a preference for the activation of slip systems based on their effective\ninterplanar distances as well as the effective Burgers vector is proposed. The\neffective interplanar distance corresponds to the manually determined\ninterplanar distance, whereas the effective Burgers vector takes a potential\ndislocation dissociation into account. This new order is: {211}1/2<111>,\n{022}1/2<111> and {022}<100>, {110}<001>, {310}<001>, {022}<011>,\n{110}1/2<111>, {112}<110> and {112}1/2<111> from high to low ratio of\ndeff/beff. Also, data on the critical resolved shear stresses of several of\nthese slip systems were measured.\n"}
{"abstract": "  Many processes in psychology are complex, such as dyadic interactions between\ntwo interacting partners (e.g. patient-therapist, intimate relationship\npartners). Nevertheless, many basic questions about interactions are difficult\nto investigate because dyadic processes can be within a person and between\npartners, they are based on multimodal aspects of behavior and unfold rapidly.\nCurrent analyses are mainly based on the behavioral coding method, whereby\nhuman coders annotate behavior based on a coding schema. But coding is\nlabor-intensive, expensive, slow, focuses on few modalities. Current approaches\nin psychology use LIWC for analyzing couples' interactions. However, advances\nin natural language processing such as BERT could enable the development of\nsystems to potentially automate behavioral coding, which in turn could\nsubstantially improve psychological research. In this work, we train machine\nlearning models to automatically predict positive and negative communication\nbehavioral codes of 368 German-speaking Swiss couples during an 8-minute\nconflict interaction on a fine-grained scale (10-seconds sequences) using\nlinguistic features and paralinguistic features derived with openSMILE. Our\nresults show that both simpler TF-IDF features as well as more complex BERT\nfeatures performed better than LIWC, and that adding paralinguistic features\ndid not improve the performance. These results suggest it might be time to\nconsider modern alternatives to LIWC, the de facto linguistic features in\npsychology, for prediction tasks in couples research. This work is a further\nstep towards the automated coding of couples' behavior which could enhance\ncouple research and therapy, and be utilized for other dyadic interactions as\nwell.\n"}
{"abstract": "  Although rapid progress in-vehicle automated technology has sped up the\npossibility of using fully automated technology for public use, little research\nhas been done on the possible influences of autonomous vehicles (AVs)\ntechnology on long-distance travel. This technology has the potential to have a\nsignificant effect on intercity trips. This study analyzed a travel survey to\nanticipate the impact of this technology on long-distance trips. We have\ndivided trips into two different categories including trips for pleasure and\ntrips for business. Different hypotheses based on the authors' knowledge and\nassisted by existing literature have been defined for each type of trip. By\nusing the Pearson method these hypotheses have been tested and the positive or\nnegative responses from respondents have been evaluated. The findings show that\nusing AVs for pleasure trips can increase the number of travelers and stimulate\npeople to choose longer distances for their trips. In addition, people enjoy\nmore and will be interested to travel more frequently. For business trips, AV\ntechnology can reduce travel costs and job-related stress. Unlike pleasure\ntrips for which people are not interested in traveling at night, business\ntravelers prefer to travel at night.\n"}
{"abstract": "  The production of axionlike particles (ALPs) with small masses in\nultraperipheral $Pb - p$ and $Pb - Pb$ collisions at the LHC is investigated.\nThe cross section and kinematical distributions associated to the diphoton\nfinal state produced in the $\\gamma \\gamma \\rightarrow a \\rightarrow \\gamma\n\\gamma$ subprocesses are estimated considering a realistic set of kinematical\ncuts. A detailed analysis of the backgrounds is performed and the expected\nsensitivity to the ALP production is derived. Our results demonstrate that a\nfuture experimental analysis of the exclusive diphoton production for the\nforward rapidities probed by the LHCb detector can improve the existing\nexclusion limits on the ALP - photon coupling in the mass range 2 GeV $\\le m_a\n\\le$ 5 GeV.\n"}
{"abstract": "  Let $G$ be a graph with $m$ edges and spectral radius $\\lambda_{1}$. Let\n$bk\\left( G\\right) $ stand for the maximal number of triangles with a common\nedge in $G$.\n  In 1970 Nosal proved that if $\\lambda_{1}^{2}>m,$ then $G$ contains a\ntriangle. In this paper we show that the same premise implies that \\[ bk\\left(\nG\\right) >\\frac{1}{12}\\sqrt[4]{m}. \\] This result settles a conjecture of Zhai,\nLin, and Shu.\n  Write $\\lambda_{2}$ for the second largest eigenvalue of $G$. Recently, Lin,\nNing, and Wu showed that if $G$ is a triangle-free graph of order at least\nthree, then \\[ \\lambda_{1}^{2}+\\lambda_{2}^{2}\\leq m, \\] thereby settling the\nsimplest case of a conjecture of Bollob\\'{a}s and the author. We give a simpler\nproof of their result.\n"}
{"abstract": "  The origin of weakly bound objects like clusters and hypernuclei, observed in\nheavy-ion collisions, is of theoretical and experimental interest. It is in the\nfocus of the experiments at RHIC and LHC since it is not evident how such\nweakly bound objects can survive in an environment whose hadronic decay\nproducts point to a temperature of the order of 150 MeV. It is as well one of\nthe key research topics in the future facilities of FAIR and NICA which are\nunder construction in Darmstadt (Germany) and Dubna (Russia), respectively. We\npresent here first results on the cluster dynamics within the model-independent\ncluster recognition library \"phase-space Minimum Spanning Tree\" (psMST) applied\nto different transport approaches: PHQMD, PHSD, SMASH and UrQMD. The psMST is\nbased on the \"Minimum Spanning Tree\" (MST) algorithm for the cluster\nrecognition which exploits correlations in coordinate space, and it is extended\nto correlations of baryons in the clusters in momentum space. We show the\nsensitivity of the cluster formation on the microscopic realization of the\nn-body dynamics and on the potential interactions in heavy-ion collisions.\n"}
{"abstract": "  Due to their large mass and small aspect ratio, icebergs pose a threat to\nboats and offshore structures. Small icebergs and bergy bits can cause harm to\nplatform hulls and are more difficult to discover remotely. As icebergs are\ndynamic mediums, the study of icebergs in relation to safe human operations\nrequires the rigorous analysis of the ice-ocean interaction, in particular with\nwaves and currents. In this paper, we present iceberg towing experiments and\nanalyze iceberg stability from GPS tracks and inertial motion unit data. The\ntowline tension as well as the boat motion relative to the iceberg was\nmeasured. Different scenarios were investigated by changing the towing strategy\nwith regards to towing speed, direction (straight or curved trajectory) and\nacceleration. Large amplitude roll oscillations with period of approximately 30\ns were observed immediately after the load dropped and the iceberg returned to\na stable static position. In two of the cases, the iceberg flipped over partly\nor entirely after some towing time. From the load cell, we observed\noscillations in the system with periods of approximately 6 s, which were\nattributed to the rope elastic properties and the iceberg response. The load\noscillations increased when the towing direction was against the waves as\nopposed to perpendicular to the waves.\n"}
{"abstract": "  Teaching an anthropomorphic robot from human example offers the opportunity\nto impart humanlike qualities on its movement. In this work we present a\nreinforcement learning based method for teaching a real world bipedal robot to\nperform movements directly from human motion capture data. Our method\nseamlessly transitions from training in a simulation environment to executing\non a physical robot without requiring any real world training iterations or\noffline steps. To overcome the disparity in joint configurations between the\nrobot and the motion capture actor, our method incorporates motion re-targeting\ninto the training process. Domain randomization techniques are used to\ncompensate for the differences between the simulated and physical systems. We\ndemonstrate our method on an internally developed humanoid robot with movements\nranging from a dynamic walk cycle to complex balancing and waving. Our\ncontroller preserves the style imparted by the motion capture data and exhibits\ngraceful failure modes resulting in safe operation for the robot. This work was\nperformed for research purposes only.\n"}
{"abstract": "  A short review of scalar curvature invariants in gravity theories is\npresented. We introduce how these invariants are constructed and discuss the\nminimal number of invariants required for a given spacetime. We then discuss\napplications of these invariants and focus on three topics that are of\nparticular interest in modern gravity theories.\n"}
{"abstract": "  Navigation guided by natural language instructions is particularly suitable\nfor Domestic Service Robots that interacts naturally with users. This task\ninvolves the prediction of a sequence of actions that leads to a specified\ndestination given a natural language navigation instruction. The task thus\nrequires the understanding of instructions, such as ``Walk out of the bathroom\nand wait on the stairs that are on the right''. The Visual and Language\nNavigation remains challenging, notably because it requires the exploration of\nthe environment and at the accurate following of a path specified by the\ninstructions to model the relationship between language and vision. To address\nthis, we propose the CrossMap Transformer network, which encodes the linguistic\nand visual features to sequentially generate a path. The CrossMap transformer\nis tied to a Transformer-based speaker that generates navigation instructions.\nThe two networks share common latent features, for mutual enhancement through a\ndouble back translation model: Generated paths are translated into instructions\nwhile generated instructions are translated into path The experimental results\nshow the benefits of our approach in terms of instruction understanding and\ninstruction generation.\n"}
{"abstract": "  Approximate computing (AC) leverages the inherent error resilience and is\nused in many big-data applications from various domains such as multimedia,\ncomputer vision, signal processing, and machine learning to improve systems\nperformance and power consumption. Like many other approximate circuits and\nalgorithms, the memory subsystem can also be used to enhance performance and\nsave power significantly. This paper proposes an efficient and effective\nsystematic methodology to construct an approximate non-volatile\nmagneto-resistive RAM (MRAM) framework using consumer-off-the-shelf (COTS) MRAM\nchips. In the proposed scheme, an extensive experimental characterization of\nmemory errors is performed by manipulating the write latency of MRAM chips\nwhich exploits the inherent (intrinsic/extrinsic process variation) stochastic\nswitching behavior of magnetic tunnel junctions (MTJs). The experimental\nresults and error-resilient image application reveal that the proposed AC\nframework provides a significant performance improvement and demonstrates a\nmaximum reduction in MRAM write current of ~66% on average with negligible or\nno loss in output quality.\n"}
{"abstract": "  In practical optimisation the dominant characteristics of the problem are\noften not known prior. Therefore, there is a need to develop general solvers as\nit is not always possible to tailor a specialised approach to each application.\nThe hybrid form of Multi-Level Selection Genetic Algorithm (MLSGA) already\nshows good performance on range of problems due to its diversity-first\napproach, which is rare among Evolutionary Algorithms. To increase the\ngenerality of its performance this paper proposes a distinct set of\nco-evolutionary mechanisms, which defines co-evolution as competition between\ncollectives rather than individuals. This distinctive approach to\nco-evolutionary provides less regular communication between sub-populations and\ndifferent fitness definitions between individuals and collectives. This\nencourages the collectives to act more independently creating a unique\nsub-regional search, leading to the development of co-evolutionary MLSGA\n(cMLSGA). To test this methodology nine genetic algorithms are selected to\ngenerate several variants of cMLSGA, which incorporates these approaches at the\nindividual level. The new mechanisms are tested on over 100 different functions\nand benchmarked against the 9 state-of-the-art competitors in order to find the\nbest general solver. The results show that the diversity of co-evolutionary\napproaches is more important than their individual performances. This allows\nthe selection of two competing algorithms that improve the generality of\ncMLSGA, without large loss of performance on any specific problem type. When\ncompared to the state-of-the-art, the proposed methodology is the most\nuniversal and robust, leading to an algorithm more likely to solve complex\nproblems with limited knowledge about the search space.\n"}
{"abstract": "  Context. The physical processes driving the formation of Galactic spiral arms\nare still under debate. Studies using open clusters favour the description of\nthe Milky Way spiral arms as long-lived structures following the classical\ndensity wave theory. Current studies comparing the Gaia DR2 field stars\nkinematic information of the Solar neighbourhood to simulations, find a better\nagreement with short-lived arms with a transient behaviour. Aims. Our aim is to\nprovide an observational, data-driven view of the Milky Way spiral structure\nand its dynamics using open clusters as the main tracers, and to contrast it\nwith simulation-based approaches. We use the most complete catalogue of Milky\nWay open clusters, with astrometric Gaia EDR3 updated parameters, estimated\nastrophysical information and radial velocities, to re-visit the nature of the\nspiral pattern of the Galaxy. Methods. We use a Gaussian mixture model to\ndetect overdensities of open clusters younger than 30 Myr that correspond to\nthe Perseus, Local, Sagittarius and Scutum spiral arms, respectively. We use\nthe birthplaces of the open cluster population younger than 80 Myr to trace the\nevolution of the different spiral arms and compute their pattern speed. We\nanalyse the age distribution of the open clusters across the spiral arms to\nexplore the differences in the rotational velocity of stars and spiral arms.\nResults. We are able to increase the range in Galactic azimuth where\npresent-day spiral arms are described, better estimating its parameters by\nadding 264 young open clusters to the 84 high-mass star-forming regions used so\nfar, thus increasing by a 314% the number of tracers. We use the evolution of\nthe open clusters from their birth positions to find that spiral arms nearly\nco-rotate with field stars at any given radius, discarding a common spiral\npattern speed for the spiral arms explored. [abridged]\n"}
{"abstract": "  This study analyses the dose difference for a variety of phantom materials\nthat can be used for Leksell Gamma Knife. These materials have properties that\nare very similar to the human tissue not including the skull bone. Geant4 was\nemployed in the analysis the dose distributions for collimator helmet sizes of\n4mm and 8mm. The phantom had a radius of 80mm. Water, brain, PMMA (Poly-methyl\nmethacrylate) and polystyrene were used as the material types. Results showed\nno considerable differences for radiation dosimetries depending on the material\ntypes. In addition, the polystyrene and PMMA (Poly-methyl methacrylate) phantom\nare also suitable for measuring the dose profiles of the Gamma Knife unit.\n"}
{"abstract": "  In satellite-based free-space continuous-variable QKD (CV-QKD), the parameter\nestimation for the atmospheric channel fluctuations due to the turbulence\neffects and attenuation is crucial for analyzing and improving the protocol\nperformance. In this paper, compressive sensing (CS) theory is applied to\nfree-space CV-QKD to achieve the channel parameter estimation with low\ncomputational complexity and small amount of data. According to CS theory, the\npossibility of the sparse representation for free-space channel is analyzed and\nthe two types of sparse reconstruction models for the channel parameters are\nconstructed combining with the stability of the sub-channels. The most part of\nvariable for parameter estimation is saved by using the model relying on the\nvariables in the quantum signals, while all the variables can be used to\ngenerate the secret key by using the model relying on the second-order\nstatistics of the variables. The methods are well adapted for the cases with\nthe limited communication time since a little or no variable is sacrificed for\nparameter estimation. Finally, simulation results are given to verify the\neffectiveness of the proposed methods.\n"}
{"abstract": "  We estimate chemical abundances and ionization parameters in the nuclear\nregion of a sample of 143 galaxies from the Palomar Spectroscopic Survey,\ncomposed by Star-Forming Galaxies (87), Seyferts 2 (16) and LINERs (40) using\nthe \\textsc{Hii-Chi-mistry} code. We also study for each spectral type the\ncorrelation of the derived quantities with other different properties of the\nhost galaxies, such as morphology, stellar mass, luminosity and mass of their\nSupermassive Black Holes. The results obtained for Star-Forming Galaxies are\nused to check the soundness of our methodology. Then, we replicate a similar\nstudy for our sample of AGN, distinguishing between Seyferts 2 and LINERs. We\nreport a saturation of Oxygen abundances for the nuclear regions of SFG. The\ncorrelations between chemical abundances and their host galaxy properties for\nSFG are in good agreement with previous studies. We find that Seyferts 2\npresent slightly higher chemical abundances but this result must be reexamined\nin larger samples of Seyfert galaxies. In contrast, we obtain lower chemical\nabundances for LINERs than for SFG. We confirm these relatively lower\nabundances for another sample of infrared luminous LINERs in the same stellar\nmass range. Our analysis of AGNs (both LINERs and Seyferts) shows that their\nhost galaxy properties are not correlated with our estimated chemical\nabundances.\n"}
{"abstract": "  We investigate critical properties of the stacked-$J_1$-$J_2$ Ising model on\na cubic lattice. Using Monte Carlo simulations and renormalization group, we\nfind a single phase transition of the first order for $J_2/J_1>1/2$. The\nrenormgroup approach predicts that a transition can be of the second order from\nthe universality class of the $O(2)$ model, but the Monte Carlo results show\nanother set of critical exponents: exponents continuously vary form the values\ntypical for a first-order transition in the finite-size scaling theory at\n$1/2<J_2/J_1<1$ to the Ising values in the limit $J_2/J_1\\to\\infty$. We also\nexclude the pseudo-first-order behavior observed in the $J_1$-$J_2$ Ising model\non a square lattice for $0.67\\lesssim J_2/J_1\\lesssim0.9$.\n"}
{"abstract": "  Given a pair of translation surfaces it is very difficult to determine\nwhether they are supported on the same algebraic curve. In fact, there are very\nfew examples of such pairs. In this note we present infinitely many examples of\nfinite collections of translation surfaces supported on the same algebraic\ncurve.\n  The underlying curves are hyperelliptic curves with many automorphisms. For\neach curve, the automorphism of maximal order acts on the space of holomorphic\n1-forms. We present a translation surface corresponding to each of the\neigenforms of this action.\n"}
{"abstract": "  In many urban areas, traffic load and noise pollution are constantly\nincreasing. Automated systems for traffic monitoring are promising\ncountermeasures, which allow to systematically quantify and predict local\ntraffic flow in order to to support municipal traffic planning decisions. In\nthis paper, we present a novel open benchmark dataset, containing 2.5 hours of\nstereo audio recordings of 4718 vehicle passing events captured with both\nhigh-quality sE8 and medium-quality MEMS microphones. This dataset is well\nsuited to evaluate the use-case of deploying audio classification algorithms to\nembedded sensor devices with restricted microphone quality and hardware\nprocessing power. In addition, this paper provides a detailed review of recent\nacoustic traffic monitoring (ATM) algorithms as well as the results of two\nbenchmark experiments on vehicle type classification and direction of movement\nestimation using four state-of-the-art convolutional neural network\narchitectures.\n"}
{"abstract": "  In this paper, we propose a whole-body planning framework that unifies\ndynamic locomotion and manipulation tasks by formulating a single multi-contact\noptimal control problem. We model the hybrid nature of a generic multi-limbed\nmobile manipulator as a switched system, and introduce a set of constraints\nthat can encode any pre-defined gait sequence or manipulation schedule in the\nformulation. Since the system is designed to actively manipulate its\nenvironment, the equations of motion are composed by augmenting the robot's\ncentroidal dynamics with the manipulated-object dynamics. This allows us to\ndescribe any high-level task in the same cost/constraint function. The\nresulting planning framework could be solved on the robot's onboard computer in\nreal-time within a model predictive control scheme. This is demonstrated in a\nset of real hardware experiments done in free-motion, such as base or\nend-effector pose tracking, and while pushing/pulling a heavy resistive door.\nRobustness against model mismatches and external disturbances is also verified\nduring these test cases.\n"}
{"abstract": "  IoT devices have become popular targets for various network attacks due to\ntheir lack of industry-wide security standards. In this work, we focus on smart\nhome IoT device identification and defending them against Distributed Denial of\nService (DDoS) attacks. The proposed framework protects smart homes by using\nVLAN-based network isolation. This architecture has two VLANs: one with\nnon-verified devices and the other with verified devices, both of which are\nmanaged by the SDN controller. Lightweight stateless flow-based features,\nincluding ICMP, TCP, and UDP protocol percentage, packet count and size, and IP\ndiversity ratio, are proposed for efficient feature collections. Further\nanalysis is performed to minimize training data to run on resource-constrained\nedge devices in smart home networks. Three popular machine learning algorithms,\nincluding K-Nearest-Neighbors, Random Forest, and Support Vector Machines, are\nused to classify IoT devices and detect different types of DDoS attacks,\nincluding TCP-SYN, UDP, and ICMP. The system's effectiveness and efficiency are\nevaluated by emulating a network consisting of an Open vSwitch, Faucet SDN\ncontroller, and several IoT device traces from two different testbeds.\n"}
{"abstract": "  This paper extends the project initiated in arXiv:2002.00201 and studies a\nlifecycle portfolio choice problem with borrowing constraints and finite\nretirement time in which an agent receives labor income that adjusts to\nfinancial market shocks in a path dependent way. The novelty here, with respect\nto arXiv:2002.00201, is the fact that we have a finite retirement time, which\nmakes the model more realistic, but harder to solve. The presence of both\npath-dependency, as in arXiv:2002.00201, and finite retirement, leads to a\ntwo-stages infinite dimensional stochastic optimal control problem, a family of\nproblems which, to our knowledge, has not yet been treated in the literature.\nWe solve the problem completely, and find explicitly the optimal controls in\nfeedback form. This is possible because we are able to find an explicit\nsolution to the associated infinite dimensional Hamilton-Jacobi-Bellman (HJB)\nequation, even if state constraints are present. Note that, differently from\narXiv:2002.00201 , here the HJB equation is of parabolic type, hence the work\nto identify the solutions and optimal feedbacks is more delicate, as it\ninvolves, in particular, time-dependent state constraints, which, as far as we\nknow, have not yet been treated in the infinite dimensional literature. The\nexplicit solution allows us to study the properties of optimal strategies and\ndiscuss their financial implications.\n"}
{"abstract": "  The normalization of brain recordings from multiple subjects responding to\nthe natural stimuli is one of the key challenges in auditory neuroscience. The\nobjective of this normalization is to transform the brain data in such a way as\nto remove the inter-subject redundancies and to boost the component related to\nthe stimuli. In this paper, we propose a deep learning framework to improve the\ncorrelation of electroencephalography (EEG) data recorded from multiple\nsubjects engaged in an audio listening task. The proposed model extends the\nlinear multi-way canonical correlation analysis (CCA) for audio-EEG analysis\nusing an auto-encoder network with a shared encoder layer. The model is trained\nto optimize a combined loss involving correlation and reconstruction. The\nexperiments are performed on EEG data collected from subjects listening to\nnatural speech and music. In these experiments, we show that the proposed deep\nmulti-way CCA (DMCCA) based model significantly improves the correlations over\nthe linear multi-way CCA approach with absolute improvements of 0.08 and 0.29\nin terms of the Pearson correlation values for speech and music tasks\nrespectively.\n"}
{"abstract": "  To what extent can the Planck satellite observations be interpreted as\nconfirmation of the quantum part of the inflationary paradigm? Has it \"seen\"\nthe Bunch-Davies state? We compare and contrast the Bunch-Davies interpretation\nwith one using a so-called entangled state in which the fluctuations of a\nspectator scalar field are entangled with those of the metric perturbations\n$\\zeta$. We first show how a spectator scalar field $\\Sigma$, with an\nexpectation value $\\sigma(t)$ that evolves in time, will generically generate\nsuch a state. We then use this state to compute the power spectrum\n$P_{\\zeta}(k)$ and thence the temperature anisotropies $C_l$ in the Cosmic\nMicrowave Background (CMB). We find interesting differences from the standard\ncalculations using the Bunch-Davies (BD) state. We argue that existing data may\nalready be used to place interesting bounds on this class of deviations from\nthe BD state and that, for some values of the parameters of the state, the\npower spectra may be consistent with the Planck satellite data.\n"}
{"abstract": "  We study notions of complexity for link complement states in Chern Simons\ntheory with compact gauge group $G$. Such states are obtained by the Euclidean\npath integral on the complement of $n$-component links inside a 3-manifold\n$M_3$. For the Abelian theory at level $k$ we find that a natural set of\nfundamental gates exists and one can identify the complexity as differences of\nlinking numbers modulo $k$. Such linking numbers can be viewed as coordinates\nwhich embeds all link complement states into $\\mathbb{Z}_k ^{\\otimes n(n-1)/2}$\nand the complexity is identified as the distance with respect to a particular\nnorm. For non-Abelian Chern Simons theories, the situation is much more\ncomplicated. We focus here on torus link states and show that the problem can\nbe reduced to defining complexity for a single knot complement state. We\nsuggest a systematic way to choose a set of minimal universal generators for\nsingle knot complement states and then evaluate the complexity using such\ngenerators. A detailed illustration is shown for $SU(2)_k$ Chern Simons theory\nand the results can be extended to general compact gauge group.\n"}
{"abstract": "  Dialogue systems in the form of chatbots and personal assistants are being\nincreasingly integrated into people's lives. Modern dialogue systems may\nconsider adopting anthropomorphic personas, mimicking societal demographic\ngroups to appear more approachable and trustworthy to users. However, the\nadoption of a persona can result in the adoption of biases. In this paper, we\npresent the first large-scale study on persona biases in dialogue systems and\nconduct analyses on personas of different social classes, sexual orientations,\nraces, and genders. We define persona biases as harmful differences in\nresponses (e.g., varying levels of offensiveness, agreement with harmful\nstatements) generated from adopting different demographic personas.\nFurthermore, we introduce an open-source framework, UnitPersonaBias, to explore\nand aggregate persona biases in dialogue systems. By analyzing the Blender and\nDialoGPT dialogue systems, we observe that adopting personas can actually\ndecrease harmful responses, compared to not using any personas. Additionally,\nwe find that persona choices can affect the degree of harms in generated\nresponses and thus should be systematically evaluated before deployment. We\nalso analyze how personas can result in different amounts of harm towards\nspecific demographics.\n"}
{"abstract": "  Recent researches have suggested that the predictive accuracy of neural\nnetwork may contend with its adversarial robustness. This presents challenges\nin designing effective regularization schemes that also provide strong\nadversarial robustness. Revisiting Vicinal Risk Minimization (VRM) as a\nunifying regularization principle, we propose Adversarial Labelling of\nPerturbed Samples (ALPS) as a regularization scheme that aims at improving the\ngeneralization ability and adversarial robustness of the trained model. ALPS\ntrains neural networks with synthetic samples formed by perturbing each\nauthentic input sample towards another one along with an adversarially assigned\nlabel. The ALPS regularization objective is formulated as a min-max problem, in\nwhich the outer problem is minimizing an upper-bound of the VRM loss, and the\ninner problem is L$_1$-ball constrained adversarial labelling on perturbed\nsample. The analytic solution to the induced inner maximization problem is\nelegantly derived, which enables computational efficiency. Experiments on the\nSVHN, CIFAR-10, CIFAR-100 and Tiny-ImageNet datasets show that the ALPS has a\nstate-of-the-art regularization performance while also serving as an effective\nadversarial training scheme.\n"}
{"abstract": "  Remote sensing for archaeological investigations using surface response is\nreasonably well established, however, remote subsurface exploration is limited\nby depth and penetration and ground resolution. Furthermore, the conservation\nof archaeological sites requires constant monitoring capability, which is often\nnot feasible between annual field seasons, but may be provided by modern\nsatellite coverage. Here we develop an approach using Sentinel-1 C-band radar\nbackscatter, and Sentinel-2 multispectral data, to map and characterise the\nsite of Qubbet el-Hawa, Egypt. The multispectral bands analysed show similar\nsensitivity to satellite imagery. However, the radar backscatter is sensitive\nto exposed known structures, as well as disturbances to soil\ntextural/composition profile due to excavation/erosion. Sub-resolution features\nsuch as causeways manifest as a 'radar-break' in the backscatter - a\ndiscontinuity in otherwise continuous radar units. Furthermore, the finite\nsubsurface response in the backscatter under the arid conditions of the site\nmeans we are able to delineate some shallow subsurface structures and map their\norientation beneath the surface in areas not yet excavated. The sensitivity of\nSentinel-1 backscatter to soil disturbance and human activity at Qubbet\nel-Hawa, and the short (~12 day) recurrence time of the satellites, makes it an\nimportant tool in heritage conservation.\n"}
{"abstract": "  The $24^{\\circ}$ N jet borders the North Tropical Belt and North Tropical\nZone, and is the fastest prograde jet on Jupiter, reaching speeds above $170$\nm/s. In this region, observations have shown several periodic convective\nplumes, likely from latent heat release from water condensation, which affect\nthe cloud and zonal wind structure of the jet. We model this region with the\nExplicit Planetary hybrid-Isentropic Coordinate model using its active\nmicrophysics scheme to study the phenomenology of water and ammonia clouds\nwithin the jet region. On perturbing the atmosphere, we find that an upper\ntropospheric wave develops that directly influences the cloud structure within\nthe jet. This wave travels at $\\sim75$ m/s in our model, and leads to periodic\nchevron-shaped features in the ammonia cloud deck. These features travel with\nthe wave speed, and are subsequently much slower than the zonal wind at the\ncloud deck. The cloud structure, and the slower drift rate, were both observed\nfollowing the convective outbreak in this region in 2016 and 2020. We find that\nan upper level circulation is responsible for these cloud features in the\naftermath of the convective outbursts. The comparatively slower observed drift\nrates of these features, relative to the wind speed of the jet, provides\nconstraints on the vertical wind shear above the cloud tops, and we suggest\nthat wind velocities determined from cloud tracking should correspond to a\ndifferent altitude compared to the $680$ hPa pressure level. We also diagnose\nthe convective potential of the atmosphere due to water condensation, and find\nthat it is strongly coupled to the wave.\n"}
{"abstract": "  Over an arbitrary compact complex space or an arbitrary germ of complex space\n$X$, we provide fine resolutions of pure Hodge modules with strict supports\n$IC_X(\\mathbb{V})$ via differential forms with locally $L^2$ boundary\nconditions. When $\\mathbb{V}=\\mathbb{C}_{X_{\\rm reg}}$ is the trivial variation\nof Hodge structure, we give a solution to a Cheeger-Goresky-MacPherson type\nconjecture: For any compact complex space $X$, there is a complete hermitian\nmetric $ds^2$ on $X_{\\rm reg}$ such that there is a canonical isomorphism\n  $$H^i_{(2)}(X_{\\rm reg},ds^2)\\simeq IH^i(X),\\quad \\forall i.$$\n  Such metric $ds^2$ could be K\\\"ahler if $X$ is a K\\\"ahler space.\n  As an application, we give a differential geometrical proof of the K\\\"ahler\npackage of the hypercohomology of pure Hodge modules. We also prove the\nK\\\"ahler version of Kashiwara's conjecture in the absolute case.\n"}
{"abstract": "  We report a theoretical description of the synthetic momentum-state lattices\nwith a 3D Gross-Pitaevskii equation (GPE), where both the external trap\npotential and the mean-field spatial-density-dependent many-body interactions\nare naturally included and exactly treated. The GPE models exhibit better\nperformance than the tight-binding model to depict the experimental\nobservations. Since the trap modifies the dispersion relation for free\nparticles and shapes the spatial density distribution that leads to\ninhomogeneous interactions, decoherences (damping oscillation) appear even for\na short-time evolution. Our parametric calculations for the two-state\noscillation suggest that we should work with a relatively shallow trap in the\nweakly interacting regime, especially when the long-term dynamics are\nconcerned. The impact of the mean-field interaction, i.e., the self-trapping\nbehavior, on the transport dynamics and the topological phase transition in a\nfinite multiple-state lattice chain is also specifically investigated. Such an\naccurate treatment of the inhomogeneous interactions allows for further\ninvestigations on the interplay with disorder, the pair correlation dynamics,\nand the thermalization process in momentum space.\n"}
{"abstract": "  We present LAMOST J0140355+392651 (hereafter J0140), a close ($P_{\\rm orb} =\n3.81$ hours) binary containing a bloated, low-mass ($M \\approx 0.15 M_{\\odot}$)\nproto-white dwarf (WD) and a massive ($M\\approx 0.95\\,M_{\\odot}$) WD companion.\nThe system's optical light curve is dominated by large-amplitude ellipsoidal\nvariability but also exhibits additional scatter, likely driven by pulsations.\nThe proto-WD is cooler ($T_{\\rm eff} = 6800\\pm 100$ K) and more puffy\n($\\log\\left[g/\\left({\\rm cm\\,s^{-2}}\\right)\\right]=4.74\\pm0.07$) than any known\nextremely low mass (ELM) WD, but hotter than any known cataclysmic variable\n(CV) donor. It either completely or very nearly fills its Roche lobe\n($R/R_{{\\rm Roche\\,lobe}}=0.99\\pm0.01$), suggesting ongoing or recently\nterminated mass transfer. No dwarf nova-like outbursts have been observed. The\nspectrum is dominated by the proto-WD but shows tentative hints of H$\\alpha$\nemission, perhaps due to accretion onto the massive WD. The properties of the\nsystem are well-matched by MESA binary evolution models of CVs with donors that\nunderwent significant nuclear evolution before the onset of mass transfer. In\nthese models, the bloated proto-WD is either still losing mass via stable Roche\nlobe overflow or was doing so until very recently. In either case, it is\nevolving toward higher temperatures at near-constant luminosity to become an\nELM WD. If the system is detached, mass transfer likely ended when the donor\nbecame too hot for magnetic braking to remain efficient. Evolutionary models\npredict that the binary will shrink to $P_{\\rm orb}\\lesssim 10$ minutes within\na few Gyr, when it will either merge or become an AM CVn binary. J0140 provides\nan observational link between the formation channels of CVs, ELM WDs, detached\nultracompact WD binaries, and AM CVn systems.\n"}
{"abstract": "  Using determinant Quantum Monte Carlo, we compare three methods of evaluating\nthe dc Hall coefficient $R_H$ of the Hubbard model: the direct measurement of\nthe off-diagonal current-current correlator $\\chi_{xy}$ in a system coupled to\na finite magnetic field (FF), $\\chi_{xy}^{\\text{FF}}$; the three-current linear\nresponse to an infinitesimal field as measured in the zero-field (ZF) Hubbard\nHamiltonian, $\\chi_{xy}^{\\text{ZF}}$; and the leading order of the recurrent\nexpansion $R_H^{(0)}$ in terms of thermodynamic susceptibilities. The two\nquantities $\\chi_{xy}^{\\text{FF}}$ and $\\chi_{xy}^{\\text{ZF}}$ can be compared\ndirectly in imaginary time. Proxies for $R_H$ constructed from the\nthree-current correlator $\\chi_{xy}^{\\text{ZF}}$ can be determined under\ndifferent simplifying assumptions and compared with $R_H^{(0)}$. We find these\ndifferent quantities to be consistent with one another, validating previous\nconclusions about the close correspondence between Fermi surface topology and\nthe sign of $R_H$, even for strongly correlated systems. These various\nquantities also provide a useful set of numerical tools for testing theoretical\npredictions about the full behavior of the Hall conductivity for strong\ncorrelations.\n"}
{"abstract": "  Pooling is a critical operation in convolutional neural networks for\nincreasing receptive fields and improving robustness to input variations. Most\nexisting pooling operations downsample the feature maps, which is a lossy\nprocess. Moreover, they are not invertible: upsampling a downscaled feature map\ncan not recover the lost information in the downsampling. By adopting the\nphilosophy of the classical Lifting Scheme from signal processing, we propose\nLiftPool for bidirectional pooling layers, including LiftDownPool and\nLiftUpPool. LiftDownPool decomposes a feature map into various downsized\nsub-bands, each of which contains information with different frequencies. As\nthe pooling function in LiftDownPool is perfectly invertible, by performing\nLiftDownPool backward, a corresponding up-pooling layer LiftUpPool is able to\ngenerate a refined upsampled feature map using the detail sub-bands, which is\nuseful for image-to-image translation challenges. Experiments show the proposed\nmethods achieve better results on image classification and semantic\nsegmentation, using various backbones. Moreover, LiftDownPool offers better\nrobustness to input corruptions and perturbations.\n"}
{"abstract": "  Modern radio telescopes produce unprecedented amounts of data, which are\npassed through many processing pipelines before the delivery of scientific\nresults. Hyperparameters of these pipelines need to be tuned by hand to produce\noptimal results. Because many thousands of observations are taken during a\nlifetime of a telescope and because each observation will have its unique\nsettings, the fine tuning of pipelines is a tedious task. In order to automate\nthis process of hyperparameter selection in data calibration pipelines, we\nintroduce the use of reinforcement learning. We test two reinforcement learning\ntechniques, twin delayed deep deterministic policy gradient (TD3) and soft\nactor-critic (SAC), to train an autonomous agent to perform this fine tuning.\nFor the sake of generalization, we consider the pipeline to be a black-box\nsystem where the summarized state of the performance of the pipeline is used by\nthe autonomous agent. The autonomous agent trained in this manner is able to\ndetermine optimal settings for diverse observations and is therefore able to\nperform 'smart' calibration, minimizing the need for human intervention.\n"}
{"abstract": "  One of the most important aims of grain boundary modeling is to predict the\nevolution of a large collection of grains in phenomena such as abnormal grain\ngrowth, coupled grain boundary motion, and recrystallization that occur under\nextreme thermomechanical loads. A unified framework to study the coevolution of\ngrain boundaries with bulk plasticity has recently been developed by Admal et\nal. (2018), which is based on modeling grain boundaries as continuum\ndislocations governed by an energy based on the Kobayashi--Warren--Carter (KWC)\nmodel (Kobayashi et al., 1998, 2000). While the resulting unified model\ndemonstrates coupled grain boundary motion and polygonization (seen in\nrecrystallization), it is restricted to grain boundary energies of the\nRead--Shockley type, which applies only to small misorientation angles. In\naddition, the implementation of the unified model using finite elements\ninherits the computational challenges of the KWC model that originate from the\nsingular diffusive nature of its governing equations. The main goal of this\nstudy is to generalize the KWC functional to grain boundary energies beyond the\nRead--Shockley-type that respect the bicrystallography of grain boundaries. The\ncomputational challenges of the KWC model are addressed by developing a\nthresholding method that relies on a primal dual algorithm and the fast\nmarching method, resulting in an O(NlogN) algorithm, where N is the number of\ngrid points. We validate the model by demonstrating the Herring angle relation,\nfollowed by a study of the grain microstructure evolution in a two-dimensional\nface-centered cubic copper polycrystal with crystal symmetry-invariant grain\nboundary energy data obtained from the lattice matching method of Runnels et\nal. (2016a,b).\n"}
{"abstract": "  3D face reconstruction plays a very important role in many real-world\nmultimedia applications, including digital entertainment, social media,\naffection analysis, and person identification. The de-facto pipeline for\nestimating the parametric face model from an image requires to firstly detect\nthe facial regions with landmarks, and then crop each face to feed the deep\nlearning-based regressor. Comparing to the conventional methods performing\nforward inference for each detected instance independently, we suggest an\neffective end-to-end framework for multi-face 3D reconstruction, which is able\nto predict the model parameters of multiple instances simultaneously using\nsingle network inference. Our proposed approach not only greatly reduces the\ncomputational redundancy in feature extraction but also makes the deployment\nprocedure much easier using the single network model. More importantly, we\nemploy the same global camera model for the reconstructed faces in each image,\nwhich makes it possible to recover the relative head positions and orientations\nin the 3D scene. We have conducted extensive experiments to evaluate our\nproposed approach on the sparse and dense face alignment tasks. The\nexperimental results indicate that our proposed approach is very promising on\nface alignment tasks without fully-supervision and pre-processing like\ndetection and crop. Our implementation is publicly available at\n\\url{https://github.com/kalyo-zjl/WM3DR}.\n"}
{"abstract": "  We propose a successive convex approximation based off-policy optimization\n(SCAOPO) algorithm to solve the general constrained reinforcement learning\nproblem, which is formulated as a constrained Markov decision process (CMDP) in\nthe context of average cost. The SCAOPO is based on solving a sequence of\nconvex objective/feasibility optimization problems obtained by replacing the\nobjective and constraint functions in the original problems with convex\nsurrogate functions. At each iteration, the convex surrogate problem can be\nefficiently solved by Lagrange dual method even the policy is parameterized by\na high-dimensional function. Moreover, the SCAOPO enables to reuse old\nexperiences from previous updates, thereby significantly reducing the\nimplementation cost when deployed in the real-world engineering systems that\nneed to online learn the environment. In spite of the time-varying state\ndistribution and the stochastic bias incurred by the off-policy learning, the\nSCAOPO with a feasible initial point can still provably converge to a\nKarush-Kuhn-Tucker (KKT) point of the original problem almost surely.\n"}
{"abstract": "  This paper presents our solution for the ICDAR 2021 Competition on Scientific\nTable Image Recognition to LaTeX. This competition has two sub-tasks: Table\nStructure Reconstruction (TSR) and Table Content Reconstruction (TCR). We treat\nboth sub-tasks as two individual image-to-sequence recognition problems. We\nleverage our previously proposed algorithm MASTER \\cite{lu2019master}, which is\noriginally proposed for scene text recognition. We optimize the MASTER model\nfrom several perspectives: network structure, optimizer, normalization method,\npre-trained model, resolution of input image, data augmentation, and model\nensemble. Our method achieves 0.7444 Exact Match and 0.8765 Exact Match @95\\%\non the TSR task, and obtains 0.5586 Exact Match and 0.7386 Exact Match 95\\% on\nthe TCR task.\n"}
{"abstract": "  The QCD axion or axion-like particles are candidates of dark matter of the\nuniverse. On the other hand, axion-like excitations exist in certain condensed\nmatter systems, which implies that there can be interactions of dark matter\nparticles with condensed matter axions. We discuss the relationship between the\ncondensed matter axion and a collective spin-wave excitation in an\nanti-ferromagnetic insulator at the quantum level. The conversion rate of the\nlight dark matter, such as the elementary particle axion or hidden photon, into\nthe condensed matter axion is estimated for the discovery of the dark matter\nsignals.\n"}
{"abstract": "  We present practical methods to show a Brunnian link is hyperbolic. The\nmethods, belonging to geometric topology, are based on observation, worked by\nhand and generally applicable. Typical examples illustrate their efficiency. We\nthus determine hyperbolicity for all of the known Brunnian links in literatures\nexcept two series. Especially, we discover 6 infinite series of hyperbolic\nBrunnian links.\n"}
{"abstract": "  We introduce DELFT, a factoid question answering system which combines the\nnuance and depth of knowledge graph question answering approaches with the\nbroader coverage of free-text. DELFT builds a free-text knowledge graph from\nWikipedia, with entities as nodes and sentences in which entities co-occur as\nedges. For each question, DELFT finds the subgraph linking question entity\nnodes to candidates using text sentences as edges, creating a dense and high\ncoverage semantic graph. A novel graph neural network reasons over the\nfree-text graph-combining evidence on the nodes via information along edge\nsentences-to select a final answer. Experiments on three question answering\ndatasets show DELFT can answer entity-rich questions better than machine\nreading based models, bert-based answer ranking and memory networks. DELFT's\nadvantage comes from both the high coverage of its free-text knowledge\ngraph-more than double that of dbpedia relations-and the novel graph neural\nnetwork which reasons on the rich but noisy free-text evidence.\n"}
{"abstract": "  High-entropy alloys (HEAs) composed of multiple principal elements have been\nshown to offer improved radiation resistance over their elemental or\ndilute-solution counterparts. Using NiCoFeCrMn HEA as a model, here we\nintroduce carbon and nitrogen interstitial alloying elements to impart chemical\nheterogeneities in the form of the local chemical order (LCO) and associated\ncompositional variations. Density functional theory simulations predict\nchemical short-range order (CSRO) (nearest neighbors and the next couple of\natomic shells) surrounding C and N, due to the chemical affinity of C with (Co,\nFe) and N with (Cr, Mn). Atomic-resolution chemical mapping of the elemental\ndistribution confirms marked compositional variations well beyond statistical\nfluctuations. Ni+ irradiation experiments at elevated temperatures demonstrate\na remarkable reduction in void swelling by at least one order of magnitude\ncompared to the base HEA without C and N alloying. The underlying mechanism is\nthat the interstitial-solute-induced chemical heterogeneities roughen the\nlattice as well as the energy landscape, impeding the movements of, and\nconstraining the path lanes for, the normally fast-moving self-interstitials\nand their clusters. The irradiation-produced interstitials and vacancies\ntherefore recombine more readily, delaying void formation. Our findings thus\nopen a promising avenue towards highly radiation-tolerant alloys.\n"}
{"abstract": "  Deoxyribonucleic acid (DNA) has shown great promise in enabling computational\napplications, most notably in the fields of DNA digital data storage and DNA\ncomputing. Information is encoded as DNA strands, which will naturally bind in\nsolution, thus enabling search and pattern-matching capabilities. Being able to\ncontrol and predict the process of DNA hybridisation is crucial for the\nambitious future of Hybrid Molecular-Electronic Computing. Current tools are,\nhowever, limited in terms of throughput and applicability to large-scale\nproblems. We present the first comprehensive study of machine learning methods\napplied to the task of predicting DNA hybridisation. For this purpose, we\nintroduce an in silico-generated hybridisation dataset of over 2.5 million data\npoints, enabling the use of deep learning. Depending on hardware, we achieve a\nreduction in inference time ranging from one to over two orders of magnitude\ncompared to the state-of-the-art, while retaining high fidelity. We then\ndiscuss the integration of our methods in modern, scalable workflows.\n"}
{"abstract": "  The Mg-Zn and Al-Zn binary alloys have been investigated theoretically under\nstatic isotropic pressure. The stable phases of these binaries on both\ninitially hexagonal-close-packed (HCP) and face-centered-cubic (FCC) lattices\nhave been determined by utilizing an iterative approach that uses a\nconfigurational cluster expansion method, Monte Carlo search algorithm, and\ndensity functional theory (DFT) calculations. Based on 64-atom models, it is\nshown that the most stable phases of the Mg-Zn binary alloy under ambient\ncondition are $\\rm MgZn_3$, $\\rm Mg_{19}Zn_{45}$, $\\rm MgZn$, and $\\rm\nMg_{34}Zn_{30}$ for the HCP, and $\\rm MgZn_3$ and $\\rm MgZn$ for the FCC\nlattice, whereas the Al-Zn binary is energetically unfavorable throughout the\nentire composition range for both the HCP and FCC lattices under all\nconditions. By applying an isotropic pressure in the HCP lattice, $\\rm\nMg_{19}Zn_{45}$ turns into an unstable phase at P$\\approx$$10$~GPa, a new\nstable phase $\\rm Mg_{3}Zn$ appears at P$\\gtrsim$$20$~GPa, and $\\rm\nMg_{34}Zn_{30}$ becomes unstable for P$\\gtrsim$$30$~GPa. For FCC lattice, the\n$\\rm Mg_{3}Zn$ phase weakly touches the convex hull at P$\\gtrsim$$20$~GPa while\nthe other stable phases remain intact up to $\\approx$$120$~GPa. Furthermore,\nmaking use of the obtained DFT results, bulk modulus has been computed for\nseveral compositions up to pressure values of the order of $\\approx$$120$~GPa.\nThe findings suggest that one can switch between $\\rm Mg$-rich and $\\rm\nZn$-rich early-stage clusters simply by applying external pressure. $\\rm\nZn$-rich alloys and precipitates are more favorable in terms of stiffness and\nstability against external deformation.\n"}
{"abstract": "  Convolutional layers in CNNs implement linear filters which decompose the\ninput into different frequency bands. However, most modern architectures\nneglect standard principles of filter design when optimizing their model\nchoices regarding the size and shape of the convolutional kernel. In this work,\nwe consider the well-known problem of spectral leakage caused by windowing\nartifacts in filtering operations in the context of CNNs. We show that the\nsmall size of CNN kernels make them susceptible to spectral leakage, which may\ninduce performance-degrading artifacts. To address this issue, we propose the\nuse of larger kernel sizes along with the Hamming window function to alleviate\nleakage in CNN architectures. We demonstrate improved classification accuracy\non multiple benchmark datasets including Fashion-MNIST, CIFAR-10, CIFAR-100 and\nImageNet with the simple use of a standard window function in convolutional\nlayers. Finally, we show that CNNs employing the Hamming window display\nincreased robustness against various adversarial attacks.\n"}
{"abstract": "  The anomalies recently reported in lepton pair transitions of $^8$Be$^*$ and\n$^4$He nuclei may be attributed to the existence of a feebly interacting light\nvector boson $X17$. We study the effects of this hypothetical particle in the\nsemileptonic $H^* \\to H e^+e^-$ decays ($H$ a $Q\\bar{q}$ meson) in the\nframework of the HQET+VMD model. Using current bounds and the universality\nassumption of the $X17$ boson to quarks, we find that decays of $D^{*+}$ and\n$D_s^{*+}$ mesons can be importantly enhanced relative to the dominant\nphoton-mediated contributions. Dedicated experimental searches at current heavy\nmeson factories may confirm the existence of this light boson or set stronger\nbounds of their couplings to ordinary matter.\n"}
{"abstract": "  As the \"Mobile AI\" revolution continues to grow, so does the need to\nunderstand the behaviour of edge-deployed deep neural networks. In particular,\nMobileNets are the go-to family of deep convolutional neural networks (CNN) for\nmobile. However, they often have significant accuracy degradation under\npost-training quantization. While studies have introduced quantization-aware\ntraining and other methods to tackle this challenge, there is limited\nunderstanding into why MobileNets (and potentially depthwise-separable CNNs\n(DWSCNN) in general) quantize so poorly compared to other CNN architectures.\nMotivated to gain deeper insights into this phenomenon, we take a different\nstrategy and study the multi-scale distributional dynamics of MobileNet-V1, a\nset of smaller DWSCNNs, and regular CNNs. Specifically, we investigate the\nimpact of quantization on the weight and activation distributional dynamics as\ninformation propagates from layer to layer, as well as overall changes in\ndistributional dynamics at the network level. This fine-grained analysis\nrevealed significant dynamic range fluctuations and a \"distributional mismatch\"\nbetween channelwise and layerwise distributions in DWSCNNs that lead to\nincreasing quantized degradation and distributional shift during information\npropagation. Furthermore, analysis of the activation quantization errors show\nthat there is greater quantization error accumulation in DWSCNN compared to\nregular CNNs. The hope is that such insights can lead to innovative strategies\nfor reducing such distributional dynamics changes and improve post-training\nquantization for mobile.\n"}
{"abstract": "  Modeling user preference from his historical sequences is one of the core\nproblems of sequential recommendation. Existing methods in this field are\nwidely distributed from conventional methods to deep learning methods. However,\nmost of them only model users' interests within their own sequences and ignore\nthe dynamic collaborative signals among different user sequences, making it\ninsufficient to explore users' preferences. We take inspiration from dynamic\ngraph neural networks to cope with this challenge, modeling the user sequence\nand dynamic collaborative signals into one framework. We propose a new method\nnamed Dynamic Graph Neural Network for Sequential Recommendation (DGSR), which\nconnects different user sequences through a dynamic graph structure, exploring\nthe interactive behavior of users and items with time and order information.\nFurthermore, we design a Dynamic Graph Recommendation Network to extract user's\npreferences from the dynamic graph. Consequently, the next-item prediction task\nin sequential recommendation is converted into a link prediction between the\nuser node and the item node in a dynamic graph. Extensive experiments on three\npublic benchmarks show that DGSR outperforms several state-of-the-art methods.\nFurther studies demonstrate the rationality and effectiveness of modeling user\nsequences through a dynamic graph.\n"}
{"abstract": "  Face obfuscation (blurring, mosaicing, etc.) has been shown to be effective\nfor privacy protection; nevertheless, object recognition research typically\nassumes access to complete, unobfuscated images. In this paper, we explore the\neffects of face obfuscation on the popular ImageNet challenge visual\nrecognition benchmark. Most categories in the ImageNet challenge are not people\ncategories; however, many incidental people appear in the images, and their\nprivacy is a concern. We first annotate faces in the dataset. Then we\ndemonstrate that face blurring -- a typical obfuscation technique -- has\nminimal impact on the accuracy of recognition models. Concretely, we benchmark\nmultiple deep neural networks on face-blurred images and observe that the\noverall recognition accuracy drops only slightly (no more than 0.68%). Further,\nwe experiment with transfer learning to 4 downstream tasks (object recognition,\nscene recognition, face attribute classification, and object detection) and\nshow that features learned on face-blurred images are equally transferable. Our\nwork demonstrates the feasibility of privacy-aware visual recognition, improves\nthe highly-used ImageNet challenge benchmark, and suggests an important path\nfor future visual datasets. Data and code are available at\nhttps://github.com/princetonvisualai/imagenet-face-obfuscation.\n"}
{"abstract": "  In this paper, we survey our recent results on the Benjamin-Ono equation on\nthe torus. As an application of the methods developed we construct large\nfamilies of periodic or quasiperiodic solutions, which are not\n$C^\\infty$-smooth.\n"}
{"abstract": "  In this paper, we first provide a brief review of the effective dynamics of\ntwo recently well-studied models of modified loop quantum cosmologies (mLQCs),\nwhich arise from different regularizations of the Hamiltonian constraint and\nshow the robustness of a generic resolution of the big bang singularity,\nreplaced by a quantum bounce due to non-perturbative Planck scale effects. As\nin loop quantum cosmology (LQC), in these modified models the slow-roll\ninflation happens generically. We consider the cosmological perturbations\nfollowing the dressed and hybrid approaches and clarify some subtle issues\nregarding the ambiguity of the extension of the effective potential of the\nscalar perturbations across the quantum bounce, and the choice of initial\nconditions. Both of the modified regularizations yield primordial power spectra\nthat are consistent with current observations for the Starobinsky potential\nwithin the framework of either the dressed or the hybrid approach. But\ndifferences in primordial power spectra are identified among the mLQCs and LQC.\nIn addition, for mLQC-I, striking differences arise between the dressed and\nhybrid approaches in the infrared and oscillatory regimes. While the\ndifferences between the two modified models can be attributed to differences in\nthe Planck scale physics, the permissible choices of the initial conditions and\nthe differences between the two perturbation approaches have been reported for\nthe first time. All these differences, due to either the different\nregularizations or the different perturbation approaches in principle can be\nobserved in terms of non-Gaussianities.\n"}
{"abstract": "  Among various recommender techniques, collaborative filtering (CF) is the\nmost successful one. And a key problem in CF is how to represent users and\nitems. Previous works usually represent a user (an item) as a vector of latent\nfactors (aka. \\textit{embedding}) and then model the interactions between users\nand items based on the representations. Despite its effectiveness, we argue\nthat it's insufficient to yield satisfactory embeddings for collaborative\nfiltering. Inspired by the idea of SVD++ that represents users based on\nthemselves and their interacted items, we propose a general collaborative\nfiltering framework named DNCF, short for Dual-embedding based Neural\nCollaborative Filtering, to utilize historical interactions to enhance the\nrepresentation. In addition to learning the primitive embedding for a user (an\nitem), we introduce an additional embedding from the perspective of the\ninteracted items (users) to augment the user (item) representation. Extensive\nexperiments on four publicly datasets demonstrated the effectiveness of our\nproposed DNCF framework by comparing its performance with several traditional\nmatrix factorization models and other state-of-the-art deep learning based\nrecommender models.\n"}
{"abstract": "  In celestial conformal field theory, gluons are represented by primary fields\nwith dimensions $\\Delta=1+i\\lambda$, $\\lambda\\in\\mathbb{R}$ and spin $J=\\pm 1$,\nin the adjoint representation of the gauge group. All two- and three-point\ncorrelation functions of these fields are zero as a consequence of\nfour-dimensional kinematic constraints. Four-point correlation functions\ncontain delta-function singularities enforcing planarity of four-particle\nscattering events. We relax these constraints by taking a shadow transform of\none field and perform conformal block decomposition of the corresponding\ncorrelators. We compute the conformal block coefficients. When decomposed in\nchannels that are \"compatible\" in two and four dimensions, such four-point\ncorrelators contain conformal blocks of primary fields with dimensions\n$\\Delta=2+M+i\\lambda$, where $M\\ge 0$ is an integer, with integer spin\n$J=-M,-M+2,\\dots,M-2,M$. They appear in all gauge group representations\nobtained from a tensor product of two adjoint representations. When decomposed\nin incompatible channels, they also contain primary fields with continuous\ncomplex spin, but with positive integer dimensions.\n"}
{"abstract": "  Prior work has shown Convolutional Neural Networks (CNNs) trained on\nsurrogate Computer Aided Design (CAD) models are able to detect and classify\nreal-world artefacts from photographs. The applications of which support\ntwinning of digital and physical assets in design, including rapid extraction\nof part geometry from model repositories, information search \\& retrieval and\nidentifying components in the field for maintenance, repair, and recording. The\nperformance of CNNs in classification tasks have been shown dependent on\ntraining data set size and number of classes. Where prior works have used\nrelatively small surrogate model data sets ($<100$ models), the question\nremains as to the ability of a CNN to differentiate between models in\nincreasingly large model repositories. This paper presents a method for\ngenerating synthetic image data sets from online CAD model repositories, and\nfurther investigates the capacity of an off-the-shelf CNN architecture trained\non synthetic data to classify models as class size increases. 1,000 CAD models\nwere curated and processed to generate large scale surrogate data sets,\nfeaturing model coverage at steps of 10$^{\\circ}$, 30$^{\\circ}$, 60$^{\\circ}$,\nand 120$^{\\circ}$ degrees. The findings demonstrate the capability of computer\nvision algorithms to classify artefacts in model repositories of up to 200,\nbeyond this point the CNN's performance is observed to deteriorate\nsignificantly, limiting its present ability for automated twinning of physical\nto digital artefacts. Although, a match is more often found in the top-5\nresults showing potential for information search and retrieval on large\nrepositories of surrogate models.\n"}
{"abstract": "  We present the results of ALMA band 3 observations of a nearby type Ic\nsupernova (SN) 2020oi. Under the standard assumptions on the SN-circumstellar\nmedium (CSM) interaction and the synchrotron emission, the data indicate that\nthe CSM structure deviates from a smooth distribution expected from the\nsteady-state mass loss in the very vicinity of the SN (~10^{15} cm), which is\nthen connected to the outer smooth distribution (~10^{16} cm). This structure\nis further confirmed through the light curve modeling of the whole radio data\nset as combined with data at lower frequency previously reported. Being an\nexplosion of a bare carbon-oxygen (C+O) star having a fast wind, we can trace\nthe mass-loss history of the progenitor of SN 2020oi in the final year. The\ninferred non-smooth CSM distribution corresponds to fluctuations on the\nsub-year time scale in the mass-loss history toward the SN explosion. Our\nfinding suggests that the pre-SN activity is likely driven by the accelerated\nchange in the nuclear burning stage in the last moments just before the massive\nstar's demise. The structure of the CSM derived in this study is beyond the\napplicability of the other methods at optical wavelengths, highlighting an\nimportance and uniqueness of quick follow-up observations of SNe by ALMA and\nother radio facilities.\n"}
{"abstract": "  We investigate the asymmetry properties of quantum states in relation to the\nHamiltonian responsible for the magnetic dipolar interaction (MDI) dynamics,\nand we evaluate its relationship to entanglement production. We consider some\nclasses of pure and mixed quantum states of two qubits evolved under MDI and,\nusing the asymmetry measure defined via the Wigner-Yanase skew information, we\ndescribe the asymmetry dependence on the Hamiltonian parameters and initial\nconditions of the system. In addition, we define and calculate the dynamics of\nthe asymmetry of local states, characterizing their temporal and interaction\nparameters dependence. Finally, because the MDI Hamiltonian has a null\neigenvalue, the group generator-based asymmetry measure does not adequately\nquantify the state susceptibility with respect to the action of the subspace\ngenerated by the eigenvectors associated with this eigenvalue. For this reason,\nwe also define and study the group element-based asymmetry measure with\nrelation to the unitary operator associated with the MDI Hamiltonian.\n"}
{"abstract": "  We report exact black hole solutions in asymptotically flat or (A)dS\nfour-dimensional spacetime with a conformally coupled self-interacting scalar\nfield in $f(R)$ gravity. We first consider the asymptotically flat model $f(R)\n= R -2\\alpha \\sqrt{R}$ and derive an exact black hole solution. Then, we\nconsider the asymptotically (A)dS model $f(R) =R -2 \\Lambda -2 \\alpha \\sqrt{R-4\n\\Lambda }$ and derive an exact black hole solution. In both cases the modified\ngravity parameter $\\alpha$, which has the dimension of the inverse mass, cannot\nbe set to zero and the self-interacting potential is determined from the\nKlein-Gordon equation, preserving the conformal invariance. The thermodynamics\nof the solutions is also studied.\n"}
{"abstract": "  We experimentally demonstrate the generation of spin-wave frequency combs\nbased on the nonlinear interaction of propagating spin waves in a\nmicrostructured waveguide. By means of time and space-resolved Brillouin light\nscattering spectroscopy, we show that the simultaneous excitation of spin waves\nwith different frequencies leads to a cascade of four-magnon scattering events\nwhich ultimately results in well-defined frequency combs. Their spectral weight\ncan be tuned by the choice of amplitude and frequency of the input signals.\nFurthermore, we introduce a model for stimulated four-magnon scattering which\ndescribes the formation of spin-wave frequency combs in the frequency and time\ndomain.\n"}
{"abstract": "  Light-manipulation of correlated electronic phases in solids offers the\ntantalizing prospect of realizing electronic devices operating at the ultrafast\ntime-scale. In this context, the experimental realization of non-equilibrium\ntransitions from a metal to a band or Mott insulator has shown to be\nparticularly elusive. Using dynamical mean-field theory, we study a simple\nmodel representing the main physical properties of the oxygen-enriched compound\nLaTiO$_{3+x}$. By properly optimizing the photo-doping of electrons from a\nlow-energy band into the valence states of the system, we show it is possible\nto induce a valence transition from a correlated metallic state to a Mott\ninsulator at ultrashort time scales and to contain the heating during this\nprocess, with the final non-thermal valence insulator having almost the same\neffective temperature of the starting metal.\n"}
{"abstract": "  We show that the dynamics of an epidemic are governed by exponential time\nrelaxation since the start of the outbreak, and explain this counterintuitive\nresult using an analogy with chemical reactions where an exponential relaxation\nof the thermodynamic driving force, or affinity, occurs. The theoretical\nreaction rate takes a form similar to the infection rate equation of\nmathematical epidemiology, while the affinity measured in epidemic data relaxes\nexponentially. The rise and fall of an epidemic wave directly follows from this\nexponential relaxation, but change in public health policies, major travels or\nnew variants induce visible perturbations on the relaxation dynamics.\n"}
{"abstract": "  Capturing interactions among event arguments is an essential step towards\nrobust event argument extraction (EAE). However, existing efforts in this\ndirection suffer from two limitations: 1) The argument role type information of\ncontextual entities is mainly utilized as training signals, ignoring the\npotential merits of directly adopting it as semantically rich input features;\n2) The argument-level sequential semantics, which implies the overall\ndistribution pattern of argument roles over an event mention, is not well\ncharacterized. To tackle the above two bottlenecks, we formalize EAE as a\nSeq2Seq-like learning problem for the first time, where a sentence with a\nspecific event trigger is mapped to a sequence of event argument roles. A\nneural architecture with a novel Bi-directional Entity-level Recurrent Decoder\n(BERD) is proposed to generate argument roles by incorporating contextual\nentities' argument role predictions, like a word-by-word text generation\nprocess, thereby distinguishing implicit argument distribution patterns within\nan event more accurately.\n"}
{"abstract": "  Motivated by the need for compact descriptions of the evolution of\nnon-classical wakes behind yawed wind turbines, we develop an analytical model\nto predict the shape of curled wakes. Interest in such modelling arises due to\nthe potential of wake steering as a strategy for mitigating power reduction and\nunsteady loading of downstream turbines in wind farms. We first estimate the\ndistribution of the shed vorticity at the wake edge due to both yaw offset and\nrotating blades. By considering the wake edge as an ideally thin vortex sheet,\nwe describe its evolution in time moving with the flow. Vortex sheet equations\nare solved using a power series expansion method, and an approximate solution\nfor the wake shape is obtained. The vortex sheet time evolution is then mapped\ninto a spatial evolution by using a convection velocity. Apart from the wake\nshape, the lateral deflection of the wake including ground effects is modelled.\nOur results show that there exists a universal solution for the shape of curled\nwakes if suitable dimensionless variables are employed. For the case of\nturbulent boundary layer inflow, the decay of vortex sheet circulation due to\nturbulent diffusion is included. Finally, we modify the Gaussian wake model by\nincorporating the predicted shape and deflection of the curled wake, so that we\ncan calculate the wake profiles behind yawed turbines. Model predictions are\nvalidated against large-eddy simulations and laboratory experiments for\nturbines with various operating conditions.\n"}
{"abstract": "  The uniform field assumption used to derive semi-classical cutoff energies of\n$10U_p$ for electron emission and $3.17U_p$ for high harmonic generation is\napplicable for ponderomotive amplitudes ($\\propto E\\lambda^2$) much smaller\nthan the field drop-off scale. For large wavelength and high field experiments\nat nanoscale structures this assumption may break down by predicting energies\nbeyond the true classical energy limits. Here we provide generalized\ncalculations for these cutoff energies by taking into account the spatial field\ndrop-off. The modified cutoff energies vary significantly from the uniform\nfield results even with ponderomotive amplitudes still an order of magnitude\nbelow the field drop-off scale. Electron emission and scattering energy as a\nfunction of the time-of-ionization is considered for the nanotip ($\\sim1/r^2$)\nfield profile. The cutoff energies as a function of the adiabaticity parameter\n$\\delta$, which may be easily calculated for given wavelength, apex field\nstrength, and nanostructure scale, are then determined through maximization for\nnanotip, nanoblade ($\\sim1/r$), and exponential field profiles. These profiles\ndeviate from each other in electron emission energy by up to nearly a factor of\nthe ponderomotive energy, indicating the importance of mid-field profile\nbehavior. The electron emission energy cutoff also attains an additional factor\nof $U_p$ due to the smooth integrated ponderomotive force in the adiabatic\ndrop-off and very long pulse regime. These results also provided as\ndouble-exponential fits for ease of use. We then compare the nanoblade electron\nemission cutoffs with a quantum simulation of the electron rescattering\nprocess. We also consider a short (few-cycle) pulsed field, focusing on a\ncosine-like pulse and overviewing the general carrier-envelope phase\ndependencies.\n"}
{"abstract": "  Analogous to conventional charge-based electronics, valleytronics aims at\nencoding data via the valley degree of freedom, enabling new routes for\ninformation processing. Long-lived interlayer excitons (IXs) in van der Waals\nheterostructures (HSs) stacked by transition metal dichalcogenides (TMDs) carry\nvalley-polarized information and thus could find promising applications in\nvalleytronic devices. Although great progress of studies on valleytronic\ndevices has been achieved, nonvolatile valleytronic memory, an indispensable\ndevice in valleytronics, is still lacking up to date. Here, we demonstrate an\nIX-based nonvolatile valleytronic memory in a WS2/WSe2 HS. In this device, the\nemission characteristics of IXs exhibit a large excitonic/valleytronic\nhysteresis upon cyclic-voltage sweeping, which is ascribed to the\nchemical-doping of O2/H2O redox couple trapped between the TMDs and substrate.\nTaking advantage of the large hysteresis, the first nonvolatile valleytronic\nmemory has been successfully made, which shows a good performance with\nretention time exceeding 60 minutes. These findings open up an avenue for\nnonvolatile valleytronic memory and could stimulate more investigations on\nvalleytronic devices.\n"}
{"abstract": "  Communication-avoiding algorithms for Linear Algebra have become increasingly\npopular, in particular for distributed memory architectures. In practice, these\nalgorithms assume that the data is already distributed in a specific way, thus\nmaking data reshuffling a key to use them. For performance reasons, a\nstraightforward all-to-all exchange must be avoided.\n  Here, we show that process relabeling (i.e. permuting processes in the final\nlayout) can be used to obtain communication optimality for data reshuffling,\nand that it can be efficiently found by solving a Linear Assignment Problem\n(Maximum Weight Bipartite Perfect Matching). Based on this, we have developed a\nCommunication-Optimal Shuffle and Transpose Algorithm (COSTA): this\nhighly-optimised algorithm implements $A=\\alpha\\cdot \\operatorname{op}(B) +\n\\beta \\cdot A,\\ \\operatorname{op} \\in \\{\\operatorname{transpose},\n\\operatorname{conjugate-transpose}, \\operatorname{identity}\\}$ on distributed\nsystems, where $A, B$ are matrices with potentially different (distributed)\nlayouts and $\\alpha, \\beta$ are scalars. COSTA can take advantage of the\ncommunication-optimal process relabeling even for heterogeneous network\ntopologies, where latency and bandwidth differ among nodes. The implementation\nnot only outperforms the best available ScaLAPACK redistribute and transpose\nroutines multiple times, but is also able to deal with more general matrix\nlayouts, in particular it is not limited to block-cyclic layouts. Finally, we\nuse COSTA to integrate a communication-optimal matrix multiplication algorithm\ninto the CP2K quantum chemistry simulation package. This way, we show that\nCOSTA can be used to unlock the full potential of recent Linear Algebra\nalgorithms in applications by facilitating interoperability between algorithms\nwith a wide range of data layouts, in addition to bringing significant\nredistribution speedups.\n"}
{"abstract": "  Food waste is a major challenge for the present world. It is the precursor to\nseveral socioeconomic problems that are plaguing the modern society. To counter\nthe same and to, simultaneously, stand by the undernourished, surplus food\nredistribution has surfaced as a viable solution. Information and\nCommunications Technology (ICT)-mediated food redistribution is a highly\nscalable approach and it percolates into the masses far better. Even if ICT is\nnot brought into the picture, the presence of food surplus redistribution in\ndeveloping countries like India is scarce and is limited to only a few of the\nmajor cities. The discussion of a surplus food redistribution framework under\nstrategic settings is a less discussed topic around the globe. This paper aims\nat addressing a surplus food redistribution framework under strategic settings,\nthereby facilitating a smoother exchange of surplus food in the smart cities of\ndeveloping countries, and beyond. As ICT is seamlessly available in smart\ncities, the paper aims to focus the framework in these cities. However, this\ncan be extended beyond the smart cities to places with greater human\ninvolvement.\n"}
{"abstract": "  Power system transient stability has been translated into a Lyapunov\nstability problem of the post-disturbance equilibrium for decades. Despite\nsubstantial results, conventional theories suffer from the stringent\nrequirement of knowing the post-disturbance equilibrium a priori. In contrast,\nthe wisdom from practice, which certificates stability by only the observation\nof converging frequencies and voltages, seems to provide an\nequilibrium-independent approach. Here, we formulate the empirical wisdom by\nthe concept of augmented synchronization and aim to bridge such a\ntheory-practice gap. First, we derive conditions under which the convergence to\naugmented synchronization implies the convergence to the equilibrium set,\nlaying the first theoretical foundation for the empirical wisdom. Then, we\nreveal from what initial values the power system can achieve augmented\nsynchronization. Our results open the possibility of an equilibrium-independent\npower system stability analytic that re-defines the nominal motion as augmented\nsynchronization rather than certain equilibrium. Single-machine examples and\nthe IEEE 9-bus system well verify our results and illustrate promising\nimplications.\n"}
{"abstract": "  We compare two properties for a CW-space $X$: (1) being homotopy equivalent\nto a CW-complex without $j$-cells for $k\\leq j\\leq \\ell$ (($k,\\ell$)-cellfree)\nand (2) $H^j(X;R)=0$ for any $\\mathbb Z\\pi_1(X)$-module $R$ when $k\\leq j\\leq\n\\ell$ (cohomogy ($k,\\ell$)-silent). Using the technique of Wall's finiteness\nobstruction, we show that a connected CW-space $X$ of finite type which is\ncohomogy ($k,\\ell$)-silent determines a \"cell-dispensability obstruction''\n$w_k(X)\\in\\tilde K_0(\\mathbb Z\\pi_1(X))$ which vanishes if and only if $X$ is\n($k,\\ell$)-cellfree ($k\\geq 4$). Any class in $\\tilde K_0(\\mathbb Z\\pi)$ may\noccur as the cell-dispensability obstruction $w_k(X)$ for a CW-space $X$ with\n$\\pi_1(X)$ identified with $\\pi$. Using projective surgery, a similar theory is\nobtained for manifolds, replacing \"cells\" by \"handles\" (antisimple manifolds).\n"}
{"abstract": "  This paper presents a real-time Acoustic Echo Cancellation (AEC) algorithm\nsubmitted to the AEC-Challenge. The algorithm consists of three modules:\nGeneralized Cross-Correlation with PHAse Transform (GCC-PHAT) based time delay\ncompensation, weighted Recursive Least Square (wRLS) based linear adaptive\nfiltering and neural network based residual echo suppression. The wRLS filter\nis derived from a novel semi-blind source separation perspective. The neural\nnetwork model predicts a Phase-Sensitive Mask (PSM) based on the aligned\nreference and the linear filter output. The algorithm achieved a mean\nsubjective score of 4.00 and ranked 2nd in the AEC-Challenge.\n"}
{"abstract": "  This work is to tackle the problem of point cloud semantic segmentation for\n3D hybrid scenes under the framework of zero-shot learning. Here by hybrid, we\nmean the scene consists of both seen-class and unseen-class 3D objects, a more\ngeneral and realistic setting in application. To our knowledge, this problem\nhas not been explored in the literature. To this end, we propose a network to\nsynthesize point features for various classes of objects by leveraging the\nsemantic features of both seen and unseen object classes, called PFNet. The\nproposed PFNet employs a GAN architecture to synthesize point features, where\nthe semantic relationship between seen-class and unseen-class features is\nconsolidated by adapting a new semantic regularizer, and the synthesized\nfeatures are used to train a classifier for predicting the labels of the\ntesting 3D scene points. Besides we also introduce two benchmarks for\nalgorithmic evaluation by re-organizing the public S3DIS and ScanNet datasets\nunder six different data splits. Experimental results on the two benchmarks\nvalidate our proposed method, and we hope our introduced two benchmarks and\nmethodology could be of help for more research on this new direction.\n"}
{"abstract": "  Analyzing multi-source data, which are multiple views of data on the same\nsubjects, has become increasingly common in molecular biomedical research.\nRecent methods have sought to uncover underlying structure and relationships\nwithin and/or between the data sources, and other methods have sought to build\na predictive model for an outcome using all sources. However, existing methods\nthat do both are presently limited because they either (1) only consider data\nstructure shared by all datasets while ignoring structures unique to each\nsource, or (2) they extract underlying structures first without consideration\nto the outcome. We propose a method called supervised joint and individual\nvariation explained (sJIVE) that can simultaneously (1) identify shared (joint)\nand source-specific (individual) underlying structure and (2) build a linear\nprediction model for an outcome using these structures. These two components\nare weighted to compromise between explaining variation in the multi-source\ndata and in the outcome. Simulations show sJIVE to outperform existing methods\nwhen large amounts of noise are present in the multi-source data. An\napplication to data from the COPDGene study reveals gene expression and\nproteomic patterns that are predictive of lung function. Functions to perform\nsJIVE are included in the R.JIVE package, available online at\nhttp://github.com/lockEF/r.jive .\n"}
{"abstract": "  The A-polynomial is a knot invariant related to the space of\n$SL_2(\\mathbb{C})$ representations of the knot group. In this paper our\ninterests lies in the logarithmic Gauss map of the A-polynomial. We develop a\nhomological point of view on this slope by extending the constructions of\nDegtyarev, the second author and Lecuona to the setting of non-abelian\nrepresentations. It defines a rational function on the character variety, which\nunifies various known invariants such as the change of curves in the\nReidemeister function, the modulus of boundary-parabolic representations, the\nboundary slope of some incompressible surfaces embedded in the exterior of the\nknot $K$ or equivalently the slopes of the sides of the Newton polygon of the\nA-polynomial $A_K$. We also present a method to compute $s_K$ in terms of\nAlexander matrices and Fox calculus.\n"}
{"abstract": "  Time evolution of quantum many-body systems typically leads to a state with\nmaximal entanglement allowed by symmetries. Two distinct routes to impede\nentanglement growth are inducing localization via spatial disorder, or\nsubjecting the system to non-unitary evolution, e.g., via projective\nmeasurements. Here we employ the idea of space-time rotation of a circuit to\nexplore the relation between systems that fall into these two classes. In\nparticular, by space-time rotating unitary Floquet circuits that display a\nlocalization transition, we construct non-unitary circuits that display a rich\nvariety of entanglement scaling and phase transitions. One outcome of our\napproach is a non-unitary circuit for free fermions in 1d that exhibits an\nentanglement transition from logarithmic scaling to volume-law scaling. This\ntransition is accompanied by a 'purification transition' analogous to that seen\nin hybrid projective-unitary circuits. We follow a similar strategy to\nconstruct a non-unitary 2d Clifford circuit that shows a transition from area\nto volume-law entanglement scaling. Similarly, we space-time rotate a 1d spin\nchain that hosts many-body localization to obtain a non-unitary circuit that\nexhibits an entanglement transition. Finally, we introduce an unconventional\ncorrelator and argue that if a unitary circuit hosts a many-body localization\ntransition, then the correlator is expected to be singular in its non-unitary\ncounterpart as well.\n"}
{"abstract": "  We show that a curve is a soliton solution to the curve shortening flow if\nand only if its geodesic curvature can be written as the inner product between\nits tangent vector field and a fixed vector of the 3-dimensional Minkowski\nspace. We use this characterization to provide a qualitative study of the\nsolitons. We show that for each fixed vector there is a 2-parameter family of\nsoliton solutions to the curve shortening flow on the 2-dimensional hyperbolic\nspace. Moreover, we prove that each soliton is defined on the entire real line,\nit is embedded and its geodesic curvature converges to a constant at each end.\n"}
{"abstract": "  The outstanding performance of transformer-based language models on a great\nvariety of NLP and NLU tasks has stimulated interest in exploring their inner\nworkings. Recent research has focused primarily on higher-level and complex\nlinguistic phenomena such as syntax, semantics, world knowledge, and common\nsense. The majority of the studies are anglocentric, and little remains known\nregarding other languages, precisely their morphosyntactic properties. To this\nend, our work presents Morph Call, a suite of 46 probing tasks for four\nIndo-European languages of different morphology: English, French, German and\nRussian. We propose a new type of probing task based on the detection of guided\nsentence perturbations. We use a combination of neuron-, layer- and\nrepresentation-level introspection techniques to analyze the morphosyntactic\ncontent of four multilingual transformers, including their less explored\ndistilled versions. Besides, we examine how fine-tuning for POS-tagging affects\nthe model knowledge. The results show that fine-tuning can improve and decrease\nthe probing performance and change how morphosyntactic knowledge is distributed\nacross the model. The code and data are publicly available, and we hope to fill\nthe gaps in the less studied aspect of transformers.\n"}
{"abstract": "  In this study we present a mixture of deep experts (MoDE) neural-network\narchitecture for single microphone speech enhancement. Our architecture\ncomprises a set of deep neural networks (DNNs), each of which is an 'expert' in\na different speech spectral pattern such as phoneme. A gating DNN is\nresponsible for the latent variables which are the weights assigned to each\nexpert's output given a speech segment. The experts estimate a mask from the\nnoisy input and the final mask is then obtained as a weighted average of the\nexperts' estimates, with the weights determined by the gating DNN. A soft\nspectral attenuation, based on the estimated mask, is then applied to enhance\nthe noisy speech signal. As a byproduct, we gain reduction at the complexity in\ntest time. We show that the experts specialization allows better robustness to\nunfamiliar noise types.\n"}
{"abstract": "  Data augmentation has recently seen increased interest in NLP due to more\nwork in low-resource domains, new tasks, and the popularity of large-scale\nneural networks that require large amounts of training data. Despite this\nrecent upsurge, this area is still relatively underexplored, perhaps due to the\nchallenges posed by the discrete nature of language data. In this paper, we\npresent a comprehensive and unifying survey of data augmentation for NLP by\nsummarizing the literature in a structured manner. We first introduce and\nmotivate data augmentation for NLP, and then discuss major methodologically\nrepresentative approaches. Next, we highlight techniques that are used for\npopular NLP applications and tasks. We conclude by outlining current challenges\nand directions for future research. Overall, our paper aims to clarify the\nlandscape of existing literature in data augmentation for NLP and motivate\nadditional work in this area. We also present a GitHub repository with a paper\nlist that will be continuously updated at\nhttps://github.com/styfeng/DataAug4NLP\n"}
{"abstract": "  Different types of graphene-related materials (GRM) are industrially\navailable and have been exploited for thermal conductivity enhancement in\npolymers. These include materials with very different features, in terms of\nthickness, lateral size and composition, especially concerning the oxygen to\ncarbon ratio and the possible presence of surface functionalization. Due to the\nvariability of GRM properties, the differences in polymer nanocomposites\npreparation methods and the microstructures obtained, a large scatter of\nthermal conductivity performance is found in literature. However, detailed\ncorrelations between GRM-based nanocomposites features, including nanoplatelets\nthickness and size, defectiveness, composition and dispersion, with their\nthermal conductivity remain mostly undefined. In the present paper, the thermal\nconductivity of GRM-based polymer nanocomposites, prepared by melt\npolymerization of cyclic polybutylene terephtalate oligomers, exploiting 13\ndifferent GRM grades, was investigated. The selected GRM, covering a wide range\nof specific surface area, size and defectiveness, secure a sound basis for the\nunderstanding of the effect of GRM properties on the thermal conductivity of\ntheir relevant polymer nanocomposites. Indeed, the thermal conductivity\nobtained appeared to depend on the interplay between the above GRM feature. In\nparticular, the combination of low GRM defectiveness and high filler\npercolation density was found to maximize the nanocomposites thermal\nconductivity.\n"}
{"abstract": "  We propose a novel method for protecting trained models with a secret key so\nthat unauthorized users without the correct key cannot get the correct\ninference. By taking advantage of transfer learning, the proposed method\nenables us to train a large protected model like a model trained with ImageNet\nby using a small subset of a training dataset. It utilizes a learnable\nencryption step with a secret key to generate learnable transformed images.\nModels with pre-trained weights are fine-tuned by using such transformed\nimages. In experiments with the ImageNet dataset, it is shown that the\nperformance of a protected model was close to that of a non-protected model\nwhen the correct key was given, while the accuracy tremendously dropped when an\nincorrect key was used. The protected model was also demonstrated to be robust\nagainst key estimation attacks.\n"}
{"abstract": "  Recent research on the time-domain audio separation networks (TasNets) has\nbrought great success to speech separation. Nevertheless, conventional TasNets\nstruggle to satisfy the memory and latency constraints in industrial\napplications. In this regard, we design a low-cost high-performance\narchitecture, namely, globally attentive locally recurrent (GALR) network.\nAlike the dual-path RNN (DPRNN), we first split a feature sequence into 2D\nsegments and then process the sequence along both the intra- and inter-segment\ndimensions. Our main innovation lies in that, on top of features recurrently\nprocessed along the inter-segment dimensions, GALR applies a self-attention\nmechanism to the sequence along the inter-segment dimension, which aggregates\ncontext-aware information and also enables parallelization. Our experiments\nsuggest that GALR is a notably more effective network than the prior work. On\none hand, with only 1.5M parameters, it has achieved comparable separation\nperformance at a much lower cost with 36.1% less runtime memory and 49.4% fewer\ncomputational operations, relative to the DPRNN. On the other hand, in a\ncomparable model size with DPRNN, GALR has consistently outperformed DPRNN in\nthree datasets, in particular, with a substantial margin of 2.4dB absolute\nimprovement of SI-SNRi in the benchmark WSJ0-2mix task.\n"}
{"abstract": "  We investigated Wosret, a region located on the small lobe of the\n67P/Churyumov-Gerasimenko comet subject to strong heating during the perihelion\npassage. This region includes Abydos, the final landing site of the Philae\nlander. We analyzed high-resolution images of the Wosret region acquired\nbetween 2015 and 2016 by the OSIRIS instrument on board the Rosetta spacecraft.\nWe observed a few morphological changes in Wosret, related to local dust\ncoating removal with an estimated depth of $\\sim$ 1 m, along with the formation\nof a cavity measuring 30 m in length and 6.5 m in depth, for a total estimated\nmass loss of 1.2 $\\times$ 10$^6$ kg. The spectrophotometry of the region is\ntypical of medium-red regions of comet 67P, with spectral slope values of 15-16\n\\%/(100 nm) in pre-perihelion data acquired at phase angle 60$^o$. Wosret has a\nspectral phase reddening of 0.0546 $\\times 10^{-4}$ nm$^{-1} deg^{-1}$, which\nis about a factor of 2 lower than what was determined for the nucleus northern\nhemisphere regions, possibly indicating a reduced surface micro-roughness due\nto the lack of widespread dust coating. A few tiny bright spots are observed.\nMorphological features such as \"goosebumps\" or clods are widely present and\nlarger in size than similar features located in the big lobe. Compared to Anhur\nand Khonsu, two southern hemisphere regions in the big lobe which are also\nexposed to high insolation during perihelion, Wosret exhibits fewer exposed\nvolatiles and less morphological variations due to activity events. Our\nanalysis indicates that the small lobe has different physical and mechanical\nproperties than the big one and a lower volatile content, at least in its\nuppermost layers. These results support the hypothesis that comet 67P\noriginated from the merging of two distinct bodies in the early Solar System.\n"}
{"abstract": "  Muscle fatigue is usually defined as a decrease in the ability to produce\nforce. The surface electromyography (sEMG) signals have been widely used to\nprovide information about muscle activities including detecting muscle fatigue\nby various data-driven techniques such as machine learning and statistical\napproaches. However, it is well-known that sEMG signals are weak signals (low\namplitude of the signals) with a low signal-to-noise ratio, data-driven\ntechniques cannot work well when the quality of the data is poor. In\nparticular, the existing methods are unable to detect muscle fatigue coming\nfrom static poses. This work exploits the concept of weak monotonicity, which\nhas been observed in the process of fatigue, to robustly detect muscle fatigue\nin the presence of measurement noises and human variations. Such a population\ntrend methodology has shown its potential in muscle fatigue detection as\ndemonstrated by the experiment of a static pose.\n"}
{"abstract": "  We have previously shown for powders that Mo substitution into the CuO chains\nof YBa2Cu3O7 can create effective pinning centres which significantly increase\nthe critical current density (j_c) in 7 T field by a factor of 4 and 10 at 50\nand 60 K, respectively. The present work reports on the influence of the Mo\nsubstitution and high-pressure oxygen annealing on the pinning properties and\ncritical currents of YBa_2Cu_{3-x}Mo_xO_{7-d} by comparing pure (x = 0, d > 0)\nand substituted (x = 0.03, d < 0) single crystals. Pinning properties have been\ninvestigated by measurements of magnetization loops and calculations of j_c in\nthe ab-plane, in the temperature range from 2 to 90 K and in fields up to 14 T.\nDepending on the Mo substitution and the oxygen treatment, several types of\npinning centres increasing j_c have been revealed and analysed in the frame of\nDew-Hughes' and Kramer's models.\n"}
{"abstract": "  Significant efforts are being invested to bring state-of-the-art\nclassification and recognition to edge devices with extreme resource\nconstraints (memory, speed and lack of GPU support). Here, we demonstrate the\nfirst deep network for acoustic recognition that is small enough for an\noff-the-shelf microcrocontroller, yet achieves state-of-the-art performance on\nstandard benchmarks. Rather than handcrafting a once-off solution, we present a\nuniversal pipeline that converts a large deep convolutional network\nautomatically via compression and quantization into a network for\nresource-impoverished edge devices. After introducing ACDNet, which produces\nabove state-of-the-art accuracy on ESC-10 (96.65%) and ESC-50 (87.1%), we\ndescribe the compression pipeline and show that it allows us to achieve 97.22%\nsize reduction and 97.28% FLOP reduction while maintaining close to\nstate-of-the-art accuracy (83.65% on ESC-50). We describe a successful\nimplementation on a standard off-the-shelf microcontroller and, beyond\nlaboratory benchmarks, report successful tests on real-world data sets.\n"}
{"abstract": "  Despite coverage enhancement in rural areas is one of the main requirements\nin next generations of wireless networks (i.e., 5G and 6G), the low expected\nprofit prevents telecommunication providers from investing in such sparsely\npopulated areas. Hence, it is required to design and deploy cost efficient\nalternatives for extending the cellular infrastructure to these regions. A\nconcrete mathematical model that characterizes and clearly captures the\naforementioned problem might be a key-enabler for studying the efficiency of\nany potential solution. Unfortunately, the commonly used mathematical tools\nthat model large scale wireless networks are not designed to capture the\nunfairness, in terms of cellular coverage, suffered by exurban and rural areas.\nIn big cities, in fact, cellular deployment is essentially capacity driven and\nthus cellular base station densities are maximum in the town centers and\ndecline when getting far from them. In this paper, a new stochastic\ngeometry-based model is implemented in order to show the coverage spatial\nvariation among urban, suburban, and exurban settlements. Indeed, by\nimplementing inhomogeneous Poisson point processes (PPPs) it is possible to\nstudy the performance metrics in a realistic scenario where terrestrial base\nstations (TBSs) are clustered around the urban center while outer aerial base\nstations (ABSs) are uniformly distributed outside an urban exclusion zone.\nBased on this, our simulation results can quantify the improvement, in terms of\ncoverage probability, that even a surprisingly low density of ABSs can bring to\nperipheral regions depending on the extension of the exclusion zone, enabling\nus to draw insightful considerations.\n"}
{"abstract": "  The El Ni\\~no Southern Oscillation (ENSO) is the most important driver of\ninterannual climate variability and can trigger extreme weather events and\ndisasters in various parts of the globe. Recently, we have developed two\napproaches for the early forecasting of El Ni\\~no. The climate network-based\napproach allows forecasting the onset of an El Ni\\~no event about 1 year ahead.\nThe complexity-based approach allows additionally to forecast the magnitude of\nan upcoming El Ni\\~no event in the calendar year before. Here we communicate\nthe forecasts of both methods for 2021. The methods indicate a high probability\n(about 89% and 90%, respectively) for the absence of an El Ni\\~no in 2021.\n"}
{"abstract": "  Land cover mapping is essential for monitoring global environmental change\nand managing natural resources. Unfortunately, traditional classification\nmodels are plagued by limited training data available in existing land cover\nproducts and data heterogeneity over space and time. In this survey, we provide\na structured and comprehensive overview of challenges in land cover mapping and\nmachine learning methods used to address these problems. We also discuss the\ngaps and opportunities that exist for advancing research in this promising\ndirection.\n"}
{"abstract": "  Video affective understanding, which aims to predict the evoked expressions\nby the video content, is desired for video creation and recommendation. In the\nrecent EEV challenge, a dense affective understanding task is proposed and\nrequires frame-level affective prediction. In this paper, we propose a\nmulti-granularity network with modal attention (MGN-MA), which employs\nmulti-granularity features for better description of the target frame.\nSpecifically, the multi-granularity features could be divided into frame-level,\nclips-level and video-level features, which corresponds to visual-salient\ncontent, semantic-context and video theme information. Then the modal attention\nfusion module is designed to fuse the multi-granularity features and emphasize\nmore affection-relevant modals. Finally, the fused feature is fed into a\nMixtures Of Experts (MOE) classifier to predict the expressions. Further\nemploying model-ensemble post-processing, the proposed method achieves the\ncorrelation score of 0.02292 in the EEV challenge.\n"}
{"abstract": "  5G networks are going to support a variety of vertical services, with a\ndiverse set of key performance indicators (KPIs), by using enabling\ntechnologies such as software-defined networking and network function\nvirtualization. It is the responsibility of the network operator to efficiently\nallocate the available resources to the service requests in such a way to honor\nKPI requirements, while accounting for the limited quantity of available\nresources and their cost. A critical challenge is that requests may be highly\nvarying over time, requiring a solution that accounts for their dynamic\ngeneration and termination. With this motivation, we seek to make joint\ndecisions for request admission, resource activation, VNF placement, resource\nallocation, and traffic routing. We do so by considering real-world aspects\nsuch as the setup times of virtual machines, with the goal of maximizing the\nmobile network operator profit. To this end, first, we formulate a one-shot\noptimization problem which can attain the optimum solution for small size\nproblems given the complete knowledge of arrival and departure times of\nrequests over the entire system lifespan. We then propose an efficient and\npractical heuristic solution that only requires this knowledge for the next\ntime period and works for realistically-sized scenarios. Finally, we evaluate\nthe performance of these solutions using real-world services and large-scale\nnetwork topologies. {Results demonstrate that our heuristic solution performs\nbetter than a state-of-the-art online approach and close to the optimum.\n"}
{"abstract": "  This paper studies bilinear saddle point problems $\\min_{\\bf{x}}\n\\max_{\\bf{y}} g(\\bf{x}) + \\bf{x}^{\\top} \\bf{A} \\bf{y} - h(\\bf{y})$, where the\nfunctions $g, h$ are smooth and strongly-convex. When the gradient and proximal\noracle related to $g$ and $h$ are accessible, optimal algorithms have already\nbeen developed in the literature \\cite{chambolle2011first,\npalaniappan2016stochastic}. However, the proximal operator is not always easy\nto compute, especially in constraint zero-sum matrix games\n\\cite{zhang2020sparsified}. This work proposes a new algorithm which only\nrequires the access to the gradients of $g, h$. Our algorithm achieves a\ncomplexity upper bound $\\tilde{\\mathcal{O}}\\left(\n\\frac{\\|\\bf{A}\\|_2}{\\sqrt{\\mu_x \\mu_y}} + \\sqrt[4]{\\kappa_x \\kappa_y (\\kappa_x\n+ \\kappa_y)} \\right)$ which has optimal dependency on the coupling condition\nnumber $\\frac{\\|\\bf{A}\\|_2}{\\sqrt{\\mu_x \\mu_y}}$ up to logarithmic factors.\n"}
{"abstract": "  While convolutional neural networks have shown a tremendous impact on various\ncomputer vision tasks, they generally demonstrate limitations in explicitly\nmodeling long-range dependencies due to the intrinsic locality of the\nconvolution operation. Initially designed for natural language processing\ntasks, Transformers have emerged as alternative architectures with innate\nglobal self-attention mechanisms to capture long-range dependencies. In this\npaper, we propose TransDepth, an architecture that benefits from both\nconvolutional neural networks and transformers. To avoid the network losing its\nability to capture local-level details due to the adoption of transformers, we\npropose a novel decoder that employs attention mechanisms based on gates.\nNotably, this is the first paper that applies transformers to pixel-wise\nprediction problems involving continuous labels (i.e., monocular depth\nprediction and surface normal estimation). Extensive experiments demonstrate\nthat the proposed TransDepth achieves state-of-the-art performance on three\nchallenging datasets. Our code is available at:\nhttps://github.com/ygjwd12345/TransDepth.\n"}
{"abstract": "  This report presents design considerations for automatically generating\nsatellite imagery datasets for training machine learning models with emphasis\nplaced on dense classification tasks, e.g. semantic segmentation. The\nimplementation presented makes use of freely available Sentinel-2 data which\nallows generation of large scale datasets required for training deep neural\nnetworks. We discuss issues faced from the point of view of deep neural network\ntraining and evaluation such as checking the quality of ground truth data and\ncomment on the scalability of the approach. Accompanying code is provided in\nhttps://github.com/michaeltrs/DeepSatData.\n"}
{"abstract": "  The sparse LiDAR point clouds become more and more popular in various\napplications, e.g., the autonomous driving. However, for this type of data,\nthere exists much under-explored space in the corresponding compression\nframework proposed by MPEG, i.e., geometry-based point cloud compression\n(G-PCC). In G-PCC, only the distance-based similarity is considered in the\nintra prediction for the attribute compression. In this paper, we propose a\nnormal-based intra prediction scheme, which provides a more efficient lossless\nattribute compression by introducing the normals of point clouds. The angle\nbetween normals is used to further explore accurate local similarity, which\noptimizes the selection of predictors. We implement our method into the G-PCC\nreference software. Experimental results over LiDAR acquired datasets\ndemonstrate that our proposed method is able to deliver better compression\nperformance than the G-PCC anchor, with $2.1\\%$ gains on average for lossless\nattribute coding.\n"}
{"abstract": "  The numerical analysis of time fractional evolution equations with the\nsecond-order elliptic operator including general time-space dependent variable\ncoefficients is challenging, especially when the classical weak initial\nsingularities are taken into account. In this paper, we introduce a concise\ntechnique to construct efficient time-stepping schemes with variable time step\nsizes for two-dimensional time fractional sub-diffusion and diffusion-wave\nequations with general time-space dependent variable coefficients. By means of\nthe novel technique, the nonuniform Alikhanov type schemes are constructed and\nanalyzed for the sub-diffusion and diffusion-wave problems. For the\ndiffusion-wave problem, our scheme is constructed by employing the recently\nestablished symmetric fractional-order reduction (SFOR) method. The\nunconditional stability of proposed schemes is rigorously discussed under mild\nassumptions on variable coefficients and, based on reasonable regularity\nassumptions and weak time mesh restrictions, the second-order convergence is\nobtained with respect to discrete $H^1$-norm. Numerical experiments are given\nto demonstrate the theoretical statements.\n"}
{"abstract": "  This work investigates the problem of learning temporal interaction networks.\nA temporal interaction network consists of a series of chronological\ninteractions between users and items. Previous methods tackle this problem by\nusing different variants of recurrent neural networks to model sequential\ninteractions, which fail to consider the structural information of temporal\ninteraction networks and inevitably lead to sub-optimal results. To this end,\nwe propose a novel Deep Structural Point Process termed as DSPP for learning\ntemporal interaction networks. DSPP simultaneously incorporates the topological\nstructure and long-range dependency structure into our intensity function to\nenhance model expressiveness. To be specific, by using the topological\nstructure as a strong prior, we first design a topological fusion encoder to\nobtain node embeddings. An attentive shift encoder is then developed to learn\nthe long-range dependency structure between users and items in continuous time.\nThe proposed two modules enable our model to capture the user-item correlation\nand dynamic influence in temporal interaction networks. DSPP is evaluated on\nthree real-world datasets for both tasks of item prediction and time\nprediction. Extensive experiments demonstrate that our model achieves\nconsistent and significant improvements over state-of-the-art baselines.\n"}
{"abstract": "  We study decomposition of $SU(2)$ gauge field into monopole and monopoleless\ncomponents. After fixing the Maximal Abelian gauge in $SU(2)$ lattice gauge\ntheory we decompose the nonabelian gauge field into the Abelian field created\nby monopoles and the modified nonabelian field with monopoles removed. We then\ncalculate respective static potentialis and show that the potential due to the\nmodified nonabelian field is nonconfining while, as is well known, the Abelian\nfield produces linear potential. We further find that the sum of these\npotentials approximates the nonabelian static potential with good precision at\nall distances considered. We conclude that at large distances the monopole\nfield potential describes the classical energy of the hadronic string while the\nstatic potential due to the modified nonabelian field describes the string\nfluctuations energy.\n"}
{"abstract": "  We show that the Mabuchi energy of any polarized manifold (X,L) is (bounded\nbelow) proper on the full space of Kahler metrics in the first Chern class of L\nif and only if (X,L) is asymptotically (semi)stable. In particular it now\nfollows from work of Xiuxiong Chen and Jinguri Cheng that there exists a cscK\nmetric in the first Chern class of L if and only if (X,L) is asymptotically\nstable, provided the reduced automorphism group of (X,L) is finite.\n"}
{"abstract": "  In the infinite dimensional Heisenberg group, we construct a left invariant\nweak Riemannian metric that gives a degenerate geodesic distance. The same\nconstruction yields a degenerate sub-Riemannian distance. We show how the\nstandard notion of sectional curvature adapts to our framework, but it cannot\nbe defined everywhere and it is unbounded on suitable sequences of planes. The\nvanishing of the distance precisely occurs along this sequence of planes, so\nthat the degenerate Riemannian distance appears in connection with an unbounded\nsectional curvature. In the 2005 paper by Michor and Mumford, this phenomenon\nwas first observed in some specific Fr\\'echet manifolds.\n"}
{"abstract": "  We report the ground-level detection of a Galactic Cosmic-Ray (GCR) flux\nenhancement lasting $\\sim$ 17 hr and associated with the passage of a magnetic\nflux rope (MFR) over the Earth. The MFR was associated with a slow Coronal Mass\nEjection (CME) caused by the eruption of a filament on 2016 October 9. Due to\nthe quiet conditions during the eruption and the lack of interactions during\nthe interplanetary CME transport to the Earth, the associated MFR preserved its\nconfiguration and reached the Earth with a strong magnetic field, low density,\nand a very low turbulence level compared to the local background, thus\ngenerating the ideal conditions to redirect and guide GCRs (in the $\\sim$ 8 to\n60 GV rigidity range) along the magnetic field of the MFR. An important\nnegative $B_Z$ component inside the MFR caused large disturbances in the\ngeomagnetic field and a relatively strong geomagnetic storm. However, these\ndisturbances are not the main factors behind the GCR enhancement. Instead, we\nfound that the major factor was the alignment between the MFR axis and the\nasymptotic direction of the observer.\n"}
{"abstract": "  By using the relativistic precession model, we have studied frequencies of\nquasi-periodic oscillations in the spacetime of a disformal Kerr black hole.\nThis black hole owns an extra disformal parameter and belongs to a class of\nnon-stealth solutions in quadratic degenerate higher-order scalar-tensor\n(DHOST) theories. Our result shows that only the periastron precession\nfrequency is related to the disformal parameter, while the azimuthal frequency\nand the nodal precession frequency are identical with those in the usual Kerr\nblack hole in general relativity. Combing with the observation data of GRO\nJ1655-40, we fit parameters of the disformal Kerr black hole, and find that the\ndisformal parameter $\\alpha$ is almost negative in the range of $1 \\sigma$,\nwhich implies the negative disformal parameter $\\alpha$ is favored by the\nobservation data of GRO J1655-40.\n"}
{"abstract": "  Migration of halogen vacancies is one of the primary sources of phase\nsegregation and material degradation in lead-halide perovskites. Here we use\nfirst principles density functional theory to compare migration energy barriers\nand paths of bromine vacancies in the bulk and at a (001) surface of cubic\nCsPbBr$_3$. Our calculations indicate that surfaces might facilitate bromine\nvacancy migration in these perovskites, due to their soft structure that allows\nfor bond lengths variations larger than in the bulk. We calculate the migration\nenergy for axial-to-axial bromine vacancy migration at the surface to be only\nhalf of the value in the bulk. Furthermore, we study the effect of modifying\nthe surface with four different alkali halide monolayers, finding an increase\nof the migration barrier to almost the bulk value for the NaCl-passivated\nsystem. Migration barriers are found to be correlated to the lattice mismatch\nbetween the CsPbBr$_3$ surface and the alkali halide monolayer. Our\ncalculations suggest that surfaces might play a significant role in mediating\nvacancy migration in halide perovskites, a result with relevance for perovskite\nnanocrystals with large surface-to-volume ratios. Moreover, we propose viable\nways for suppressing this undesirable process through passivation with alkali\nhalide salts.\n"}
{"abstract": "  Best match graphs (BMGs) are a class of colored digraphs that naturally\nappear in mathematical phylogenetics and can be approximated with the help of\nsimilarity measures between gene sequences, albeit not without errors. The\ncorresponding graph editing problem can be used as a means of error correction.\nSince the arc set modification problems for BMGs are NP-complete, efficient\nheuristics are needed if BMGs are to be used for the practical analysis of\nbiological sequence data. Since BMGs have a characterization in terms of\nconsistency of a certain set of rooted triples, we consider heuristics that\noperate on triple sets. As an alternative, we show that there is a close\nconnection to a set partitioning problem that leads to a class of top-down\nrecursive algorithms that are similar to Aho's supertree algorithm and give\nrise to BMG editing algorithms that are consistent in the sense that they leave\nBMGs invariant. Extensive benchmarking shows that community detection\nalgorithms for the partitioning steps perform best for BMG editing.\n"}
{"abstract": "  The present status of the field theoretical model studies of the deep\ninelastic scattering induced by (anti)neutrino on the nuclear targets in a wide\nrange of Bjorken variable $x$ and four momentum transfer square $Q^2$, has been\nreviewed~\\cite{Haider:2011qs,Haider:2012nf,Haider:2016zrk,Zaidi:2019mfd,Zaidi:2019asc,Ansari:2020xne}.\n  The effect of the nonperturbative corrections such as target mass correction\nand higher twist effects, perturbative evolution of the parton densities,\nnuclear medium modifications in the nucleon structure functions, nuclear\nisoscalarity corrections on the weak nuclear structure functions have been\ndiscussed. These structure functions have been used to obtain the differential\nscattering cross sections. The various nuclear medium effects like the Fermi\nmotion, binding energy, nucleon correlations, mesonic contributions, shadowing\nand antishadowing corrections relevant in the different regions of $x$ and\n$Q^2$ have been discussed. The numerical results for the structure functions\nand the cross sections are compared with some of the available experimental\ndata including the recent results from MINERvA. The predictions are made in\nargon nuclear target which is planned to be used as a target material in DUNE\nat the Fermilab.\n"}
{"abstract": "  We present a test platform for the Athena X-IFU detection chain, which will\nserve as the first demonstration of the representative end-to-end detection and\nreadout chain for the X-IFU, using prototypes of the future flight electronics\nand currently available subsystems. This test bench, housed in a commercial\ntwo-stage ADR cryostat, includes a focal plane array placed at the 50 mK cold\nstage of the ADR with a kilopixel array of transition-edge sensor\nmicrocalorimeter spectrometers and associated cold readout electronics.\nPrototype room temperature electronics for the X-IFU provide the readout, and\nwill evolve over time to become more representative of the X-IFU mission\nbaseline. The test bench yields critical feedback on subsystem designs and\ninterfaces, in particular the warm readout electronics, and will provide an\nin-house detection system for continued testing and development of the warm\nreadout electronics and for the validation of X-ray calibration sources. In\nthis paper, we describe the test bench subsystems and design, characterization\nof the cryostat, and current status of the project.\n"}
{"abstract": "  Controlling network systems has become a problem of paramount importance.\nOptimally controlling a network system with linear dynamics and minimizing a\nquadratic cost is a particular case of the well-studied linear-quadratic\nproblem. When the specific topology of the network system is ignored, the\noptimal controller is readily available. However, this results in a\n\\emph{centralized} controller, facing limitations in terms of implementation\nand scalability. Finding the optimal \\emph{distributed} controller, on the\nother hand, is intractable in the general case. In this paper, we propose the\nuse of graph neural networks (GNNs) to parametrize and design a distributed\ncontroller. GNNs exhibit many desirable properties, such as being naturally\ndistributed and scalable. We cast the distributed linear-quadratic problem as a\nself-supervised learning problem, which is then used to train the GNN-based\ncontrollers. We also obtain sufficient conditions for the resulting closed-loop\nsystem to be input-state stable, and derive an upper bound on the trajectory\ndeviation when the system is not accurately known. We run extensive simulations\nto study the performance of GNN-based distributed controllers and show that\nthey are computationally efficient and scalable.\n"}
{"abstract": "  The major part of the vanilla vision transformer (ViT) is the attention block\nthat brings the power of mimicking the global context of the input image. For\nbetter performance, ViT needs large-scale training data. To overcome this data\nhunger limitation, many ViT-based networks, or hybrid-ViT, have been proposed\nto include local context during the training. The robustness of ViTs and its\nvariants against adversarial attacks has not been widely investigated in the\nliterature like CNNs. This work studies the robustness of ViT variants 1)\nagainst different Lp-based adversarial attacks in comparison with CNNs, 2)\nunder adversarial examples (AEs) after applying preprocessing defense methods\nand 3) under the adaptive attacks using expectation over transformation (EOT)\nframework. To that end, we run a set of experiments on 1000 images from\nImageNet-1k and then provide an analysis that reveals that vanilla ViT or\nhybrid-ViT are more robust than CNNs. For instance, we found that 1) Vanilla\nViTs or hybrid-ViTs are more robust than CNNs under Lp-based attacks and under\nadaptive attacks. 2) Unlike hybrid-ViTs, Vanilla ViTs are not responding to\npreprocessing defenses that mainly reduce the high frequency components.\nFurthermore, feature maps, attention maps, and Grad-CAM visualization jointly\nwith image quality measures, and perturbations' energy spectrum are provided\nfor an insight understanding of attention-based models.\n"}
{"abstract": "  The theory for multiplier empirical processes has been one of the central\ntopics in the development of the classical theory of empirical processes, due\nto its wide applicability to various statistical problems. In this paper, we\ndevelop theory and tools for studying multiplier $U$-processes, a natural\nhigher-order generalization of the multiplier empirical processes. To this end,\nwe develop a multiplier inequality that quantifies the moduli of continuity of\nthe multiplier $U$-process in terms of that of the (decoupled) symmetrized\n$U$-process. The new inequality finds a variety of applications including (i)\nmultiplier and bootstrap central limit theorems for $U$-processes, (ii) general\ntheory for bootstrap $M$-estimators based on $U$-statistics, and (iii) theory\nfor $M$-estimation under general complex sampling designs, again based on\n$U$-statistics.\n"}
{"abstract": "  Motivated by recent data analyses in biomedical imaging studies, we consider\na class of image-on-scalar regression models for imaging responses and scalar\npredictors. We propose using flexible multivariate splines over triangulations\nto handle the irregular domain of the objects of interest on the images, as\nwell as other characteristics of images. The proposed estimators of the\ncoefficient functions are proved to be root-n consistent and asymptotically\nnormal under some regularity conditions. We also provide a consistent and\ncomputationally efficient estimator of the covariance function. Asymptotic\npointwise confidence intervals and data-driven simultaneous confidence\ncorridors for the coefficient functions are constructed. Our method can\nsimultaneously estimate and make inferences on the coefficient functions while\nincorporating spatial heterogeneity and spatial correlation. A highly efficient\nand scalable estimation algorithm is developed. Monte Carlo simulation studies\nare conducted to examine the finite-sample performance of the proposed method,\nwhich is then applied to the spatially normalized positron emission tomography\ndata of the Alzheimer's Disease Neuroimaging Initiative.\n"}
{"abstract": "  Given a spectral triple on a $C^*$-algebra $\\mathcal A$ together with a\nunital injective endomorphism $\\alpha$, the problem of defining a suitable\ncrossed product $C^*$-algebra endowed with a spectral triple is addressed. The\nproposed construction is mainly based on the works of Cuntz and of Hawkins,\nSkalski, White and Zacharias, and on our previous papers. The embedding of\n$\\alpha(\\mathcal A)$ in $\\mathcal A$ can be considered as the dual form of a\ncovering projection between noncommutative spaces. A main assumption is the\nexpansiveness of the endomorphism, which takes the form of the local\nisometricity of the covering projection and is expressed via the compatibility\nof the Lip-norms on $\\mathcal A$ and $\\alpha(\\mathcal A)$.\n"}
{"abstract": "  Neural ODE Processes approach the problem of meta-learning for dynamics using\na latent variable model, which permits a flexible aggregation of contextual\ninformation. This flexibility is inherited from the Neural Process framework\nand allows the model to aggregate sets of context observations of arbitrary\nsize into a fixed-length representation. In the physical sciences, we often\nhave access to structured knowledge in addition to raw observations of a\nsystem, such as the value of a conserved quantity or a description of an\nunderstood component. Taking advantage of the aggregation flexibility, we\nextend the Neural ODE Process model to use additional information within the\nLearning Using Privileged Information setting, and we validate our extension\nwith experiments showing improved accuracy and calibration on simulated\ndynamics tasks.\n"}
{"abstract": "  There are two elementary superconducting qubit types that derive directly\nfrom the quantum harmonic oscillator. In one the inductor is replaced by a\nnonlinear Josephson junction to realize the widely used charge qubits with a\ncompact phase variable and a discrete charge wavefunction. In the other the\njunction is added in parallel, which gives rise to an extended phase variable,\ncontinuous wavefunctions and a rich energy level structure due to the loop\ntopology. While the corresponding rf-SQUID Hamiltonian was introduced as a\nquadratic, quasi-1D potential approximation to describe the fluxonium qubit\nimplemented with long Josephson junction arrays, in this work we implement it\ndirectly using a linear superinductor formed by a single uninterrupted aluminum\nwire. We present a large variety of qubits all stemming from the same circuit\nbut with drastically different characteristic energy scales. This includes flux\nand fluxonium qubits but also the recently introduced quasi-charge qubit with\nstrongly enhanced zero point phase fluctuations and a heavily suppressed flux\ndispersion. The use of a geometric inductor results in high precision of the\ninductive and capacitive energy as guaranteed by top-down lithography - a key\ningredient for intrinsically protected superconducting qubits. The geometric\nfluxonium also exhibits a large magnetic dipole, which renders it an\ninteresting new candidate for quantum sensing applications.\n"}
{"abstract": "  The development of pathological speech systems is currently hindered by the\nlack of a standardised objective evaluation framework. In this work, (1) we\nutilise existing detection and analysis techniques to propose a general\nframework for the consistent evaluation of synthetic pathological speech. This\nframework evaluates the voice quality and the intelligibility aspects of speech\nand is shown to be complementary using our experiments. (2) Using our proposed\nevaluation framework, we develop and test a dysarthric voice conversion system\n(VC) using CycleGAN-VC and a PSOLA-based speech rate modification technique. We\nshow that the developed system is able to synthesise dysarthric speech with\ndifferent levels of speech intelligibility.\n"}
{"abstract": "  Using the BRST approach to higher spin field theories we develop a generic\ntechnique for constructing the cubic interaction vertices for N=1\nsupersymmetric massless higher spin fields on four, six and ten dimensional\nflat backgrounds. Such an approach allows formulation of the equations for\ncubic vertices including bosonic and fermionic higher spin fields, and the\nproblem of finding the vertices is reduced to finding the consistent solutions\nto these equations. As a realization of this procedure, we present the\nparticular solutions for the vertices where the fields obey some off-shell\nconstraints. It is shown that the supersymmetry imposes additional constraints\non the vertices and singles out a particular subclass of the solutions. As a\nconcrete application of the generic scheme, we consider supersymmetric\nYang-Mills-like systems in four, six and ten dimensions where the higher spin\nfields transform under some internal symmetry group, as well as\nsupergravity-like systems in the same dimensions.\n"}
{"abstract": "  Reliable platforms for data collation during airline schedule operations have\nsignificantly increased the quality and quantity of available information for\neffectively managing airline schedule disruptions. To that effect, this paper\napplies macroscopic and microscopic techniques by way of basic statistics and\nmachine learning, respectively, to analyze historical scheduling and operations\ndata from a major airline in the United States. Macroscopic results reveal that\nmajority of irregular operations in airline schedule that occurred over a\none-year period stemmed from disruptions due to flight delays, while\nmicroscopic results validate different modeling assumptions about key drivers\nfor airline disruption management like turnaround as a Gaussian process.\n"}
{"abstract": "  Logistic regression model is widely used in many studies to investigate the\nrelationship between a binary response variable $Y$ and a set of potential\npredictors $\\mathbf X$. The binary response may represent, for example, the\noccurrence of some outcome of interest ($Y=1$ if the outcome occurred and $Y=0$\notherwise). When the dependent variable $Y$ represents a rare event, the\nlogistic regression model shows relevant drawbacks. In order to overcome these\ndrawbacks we propose the Generalized Extreme Value (GEV) regression model. In\nparticular, we suggest the quantile function of the GEV distribution as link\nfunction, so our attention is focused on the tail of the response curve for\nvalues close to one. A sample of observations is said to contain a cure\nfraction when a proportion of the study subjects (the so-called cured\nindividuals, as opposed to the susceptibles) cannot experience the outcome of\ninterest. One problem arising then is that it is usually unknown who are the\ncured and the susceptible subjects, unless the outcome of interest has been\nobserved. In these settings, a logistic regression analysis of the relationship\nbetween $\\mathbf X$ and $Y$ among the susceptibles is no more straightforward.\nWe develop a maximum likelihood estimation procedure for this problem, based on\nthe joint modeling of the binary response of interest and the cure status. We\ninvestigate the identifiability of the resulting model. Then, we conduct a\nsimulation study to investigate its finite-sample behavior, and application to\nreal data.\n"}
{"abstract": "  This paper describes a bounded generation result concerning the minimal\nnatural number $K$ such that for\n$Q(C_2,2R):=\\{A\\varepsilon_{\\phi}(2x)A^{-1}|x\\in R,A\\in{\\rm Sp}_4(R),\\phi\\in\nC_2\\}$, one has $N_{C_2,2R}=\\{X_1\\cdots X_K|\\forall 1\\leq i\\leq K:X_i\\in\nQ(C_2,2R)\\}$ for rings of algebraic integers $R$ and the principal congruence\nsubgroup $N_{C_2,2R}$ in ${\\rm Sp}_4(R).$ This gives an explicit version of an\nabstract bounded generation result of a similar type as presented by Morris.\nFurthermore, the result presented does not depend on several number-theoretic\nquantities unlike Morris' result. Using this bounded generation result, we\nfurther give explicit bounds for the strong boundedness of ${\\rm Sp}_4(R)$ for\ncertain examples of rings $R,$ thereby giving explicit versions of results in\nan earlier paper. We further give a classification of normally generating\nsubsets of ${\\rm Sp}_4(R)$ for $R$ a ring of algebraic integers.\n"}
{"abstract": "  A connected, locally finite graph $\\Gamma$ is a Cayley--Abels graph for a\ntotally disconnected, locally compact group $G$ if $G$ acts vertex-transitively\nwith compact, open vertex stabilizers on $\\Gamma$. Define the minimal degree of\n$G$ as the minimal degree of a Cayley--Abels graph of $G$. We relate the\nminimal degree in various ways to the modular function, the scale function and\nthe structure of compact open subgroups. As an application, we prove that if\n$T_d$ denotes the $d$-regular tree, then the minimal degree of ${\\rm Aut}(T_d)$\nis $d$ for all $d\\geq 2$.\n"}
{"abstract": "  Via a unified geometric approach, a class of generalized trigonometric\nfunctions with two parameters are analytically extended to maximal domains on\nwhich they are univalent. Some consequences are deduced concerning radius of\nconvergence for the Maclaurin series, commutation with rotation, continuation\nbeyond the domain of univalence, and periodicity.\n"}
{"abstract": "  Modelling joint dynamics of liquid vanilla options is crucial for\narbitrage-free pricing of illiquid derivatives and managing risks of option\ntrade books. This paper develops a nonparametric model for the European options\nbook respecting underlying financial constraints and while being practically\nimplementable. We derive a state space for prices which are free from static\n(or model-independent) arbitrage and study the inference problem where a model\nis learnt from discrete time series data of stock and option prices. We use\nneural networks as function approximators for the drift and diffusion of the\nmodelled SDE system, and impose constraints on the neural nets such that\nno-arbitrage conditions are preserved. In particular, we give methods to\ncalibrate \\textit{neural SDE} models which are guaranteed to satisfy a set of\nlinear inequalities. We validate our approach with numerical experiments using\ndata generated from a Heston stochastic local volatility model.\n"}
{"abstract": "  Multi-task learning is a framework that enforces different learning tasks to\nshare their knowledge to improve their generalization performance. It is a hot\nand active domain that strives to handle several core issues; particularly,\nwhich tasks are correlated and similar, and how to share the knowledge among\ncorrelated tasks. Existing works usually do not distinguish the polarity and\nmagnitude of feature weights and commonly rely on linear correlation, due to\nthree major technical challenges in: 1) optimizing the models that regularize\nfeature weight polarity, 2) deciding whether to regularize sign or magnitude,\n3) identifying which tasks should share their sign and/or magnitude patterns.\nTo address them, this paper proposes a new multi-task learning framework that\ncan regularize feature weight signs across tasks. We innovatively formulate it\nas a biconvex inequality constrained optimization with slacks and propose a new\nefficient algorithm for the optimization with theoretical guarantees on\ngeneralization performance and convergence. Extensive experiments on multiple\ndatasets demonstrate the proposed methods' effectiveness, efficiency, and\nreasonableness of the regularized feature weighted patterns.\n"}
{"abstract": "  We investigate the performance of Gaussianmodulated coherent-state QKD\nprotocols in the presence of canonical attacks, which are collective Gaussian\nattacks resulting in Gaussian channels described by one of the possible\ncanonical forms. We present asymptotic key rates and then we extend the results\nto the finite-size regime using a recently-developed toolbox for composable\nsecurity.\n"}
{"abstract": "  The defect corrections to polarization and dielectric functions of Bloch\nelectrons in quantum wells are first calculated. Following this, we derive the\nfirst two moment equations from Boltzmann transport theory and apply them to\nexplore defect effects on magneto-transport of Bloch electrons. Meanwhile, we\nobtain analytically the momentum-relaxation time and mobility tensor for Bloch\nelectrons making use of the screened defect-corrected polarization function.\nBased on quantum-statistical theory, we further investigate the defect capture\nand charging dynamics by employing a parameterized physics model for defects to\nobtain defect wave functions. After this, both capture and relaxation rates, as\nwell as density for captured Bloch electrons, are calculated self-consistently\nas functions of temperature, doping density and different defect types. By\napplying the energy-balance equation, the number of occupied energy levels and\nchemical potential of defects are determined, with which the transition rate\nfor defect capturing is obtained. By using these results, defect\nenergy-relaxation, capture and escape rates, and Bloch-electron chemical\npotential are obtained self-consistently. At the same time, the Bloch-electron\nenergy- and momentum-relaxation rates, as well as the current suppression\nfactor, are also investigated quantitatively. Finally, by combining all these\nstudies together, the temperature dependence of the Hall and longitudinal\nmobilities is demonstrated for Bloch electrons in either single- or\nmulti-quantum wells, which can be utilized for quantifying burst noise in\ntransistors and blinking noise in photo-detectors.\n"}
{"abstract": "  Earlier there was no mode of online communication between users. In big or\nsmall organizations communication between users posed a challenge. There was a\nrequirement to record these communications and store the data for further\nevaluation. The idea is to automate the existing Simple Chat Room system and\nmake the users utilize the software so that their valuable information is\nstored digitally and can be retrieved for further management purposes. There\nwas no online method of communicating with different users. There were many\ndifferent interfaces available in the market but this method of using windows\nsockets to communicate between nodes would be fast and reliable. The main\nobjective of our Simple Chat Room project is to create a chat application that\nhelps different users to communicate with each other through a server\nconnected. This is a simple chat program with a server and can have many\nclients. The server needs to be started first and clients can be connected\nlater. Simple Chat Room provides bidirectional communication between client and\nserver. It enables users to seamlessly communicate with each other. The user\ncan chat using this chat application. If the user at the other end is active\nthen they can start a chat session. The chat is recorded in this application.\n"}
{"abstract": "  We have designed, optimized, fabricated and characterized highly reflective\nquasi-omnidirectional (angular range of $0-60^\\circ$) multilayered structures\nwith a wide spectral range. Two techniques, chirping (a continuous change in\nthicknesses) and stacking of Bragg-type sub-structures, have been used to\nenhance the reflectance with minimum thickness for a given pair of refractive\nindices. Numerical calculations were carried out employing the transfer matrix\nmethod and we optimized the design parameters to obtain maximal reflectance\naveraged over different spectral ranges for all angles. We fabricated some of\nthe optimized structures with porous silicon dielectric multilayers with low\nrefractive contrast and compared their measured optical properties with the\ncalculations. Two chirped structures with thicknesses 21.6 $\\mu$m and 60.4\n$\\mu$m, resulting in quasi omnidirectional mirrors with bandwidths of 360 nm\nand 1800 nm, centered at 1160 nm and 1925 nm respectively have been shown. In\naddition, we fabricated a stacked sub-Bragg mirror structure with a quasi\nomnidirectional bandwidth of 1800 nm (centered at 1850 nm) and a thickness of\n41.5 $\\mu$m, which is almost two third (in thickness) of the chirped structure.\nThus, our techniques allowed us to obtain relatively thin quasi-omnidirectional\nmirrors with wide bands over different wavelength ranges. Our analysis\ntechniques can be used for the optimization of the reflectance not only in\nmultilayered PS systems with different refractive index contrasts but also for\nsystems with other types of materials with low refractive index contrast. The\npresent study could also be useful for obtaining omnidirectional dielectric\nmirrors in large spectral regions using different materials, flat focusing\nreflectors, thermal regulators, or, if defects are included, as filters or\nremote chemical/biosensors with a wide angular independent response.\n"}
{"abstract": "  In this paper, we investigate the question: Given a small number of\ndatapoints, for example N = 30, how tight can PAC-Bayes and test set bounds be\nmade? For such small datasets, test set bounds adversely affect generalisation\nperformance by withholding data from the training procedure. In this setting,\nPAC-Bayes bounds are especially attractive, due to their ability to use all the\ndata to simultaneously learn a posterior and bound its generalisation risk. We\nfocus on the case of i.i.d. data with a bounded loss and consider the generic\nPAC-Bayes theorem of Germain et al. While their theorem is known to recover\nmany existing PAC-Bayes bounds, it is unclear what the tightest bound derivable\nfrom their framework is. For a fixed learning algorithm and dataset, we show\nthat the tightest possible bound coincides with a bound considered by Catoni;\nand, in the more natural case of distributions over datasets, we establish a\nlower bound on the best bound achievable in expectation. Interestingly, this\nlower bound recovers the Chernoff test set bound if the posterior is equal to\nthe prior. Moreover, to illustrate how tight these bounds can be, we study\nsynthetic one-dimensional classification tasks in which it is feasible to\nmeta-learn both the prior and the form of the bound to numerically optimise for\nthe tightest bounds possible. We find that in this simple, controlled scenario,\nPAC-Bayes bounds are competitive with comparable, commonly used Chernoff test\nset bounds. However, the sharpest test set bounds still lead to better\nguarantees on the generalisation error than the PAC-Bayes bounds we consider.\n"}
{"abstract": "  Co-part segmentation is an important problem in computer vision for its rich\napplications. We propose an unsupervised learning approach for co-part\nsegmentation from images. For the training stage, we leverage motion\ninformation embedded in videos and explicitly extract latent representations to\nsegment meaningful object parts. More importantly, we introduce a dual\nprocedure of part-assembly to form a closed loop with part-segmentation,\nenabling an effective self-supervision. We demonstrate the effectiveness of our\napproach with a host of extensive experiments, ranging from human bodies,\nhands, quadruped, and robot arms. We show that our approach can achieve\nmeaningful and compact part segmentation, outperforming state-of-the-art\napproaches on diverse benchmarks.\n"}
{"abstract": "  String theory on ${\\rm AdS}_3\\times {\\rm S}^3 \\times\\mathbb{T}^4$ with one\nunit ($k=1$) of NS-NS flux is considered in the hybrid formalism of Berkovits,\nVafa & Witten (BVW). Using the free field realisation of the world-sheet theory\nat $k=1$, we identify explicitly the BRST cohomology classes corresponding to\nsome of the low-lying states of the dual CFT. In particular, we do this for the\n${\\cal N}=4$ superconformal generators of the symmetric orbifold theory, and we\nconfirm these identifications by showing that the worldsheet correlators\nreproduce the expected dual CFT answer. Along the way we note that the physical\nvertex operators on the worldsheet have a simpler form if one works with a\ndifferent, but equivalent, choice for the BRST operators relative to BVW.\n"}
{"abstract": "  The reliability of Deep Learning systems depends on their accuracy but also\non their robustness against adversarial perturbations to the input data.\nSeveral attacks and defenses have been proposed to improve the performance of\nDeep Neural Networks under the presence of adversarial noise in the natural\nimage domain. However, robustness in computer-aided diagnosis for volumetric\ndata has only been explored for specific tasks and with limited attacks. We\npropose a new framework to assess the robustness of general medical image\nsegmentation systems. Our contributions are two-fold: (i) we propose a new\nbenchmark to evaluate robustness in the context of the Medical Segmentation\nDecathlon (MSD) by extending the recent AutoAttack natural image classification\nframework to the domain of volumetric data segmentation, and (ii) we present a\nnovel lattice architecture for RObust Generic medical image segmentation (ROG).\nOur results show that ROG is capable of generalizing across different tasks of\nthe MSD and largely surpasses the state-of-the-art under sophisticated\nadversarial attacks.\n"}
{"abstract": "  We prove that the topological recursion formalism can be used to quantize any\ngeneric classical spectral curve with smooth ramification points and simply\nramified away from poles. For this purpose, we build both the associated\nquantum curve, i.e. the differential operator quantizing the algebraic equation\ndefining the classical spectral curve considered, and a basis of wave\nfunctions, that is to say a basis of solutions of the corresponding\ndifferential equation. We further build a Lax pair representing the resulting\nquantum curve and thus present it as a point in an associated space of\nmeromorphic connections on the Riemann sphere, a first step towards\nisomonodromic deformations. We finally propose two examples: the derivation of\na 2-parameter family of formal trans-series solutions to Painlev\\'e 2 equation\nand the quantization of a degree three spectral curve with pole only at\ninfinity.\n"}
{"abstract": "  In this paper, approximate outage probability (OP) expressions are derived\nfor uplink cell-free massive multiple-input-multiple-output (CF-mMIMO) system.\nThe access points (APs) of the system considered have imperfect channel state\ninformation (CSI). The approximate expressions are derived first using\nconditional expectations and then using a novel dimension reduction method that\napproximates higher-order integration by several single order integrations.\nUsing the same approach, closed-form approximations are also derived for\nconventional massive MIMO (mMIMO) systems. The OP approximations are then used\nto characterize the performance of cell-edge users of CF-mMIMO systems and\ncompare the designs of CF-mMIMO and mMIMO systems. The derived expressions have\na close match with the simulated expression for OP.\n"}
{"abstract": "  In this paper, we propose an interesting semi-sparsity smoothing algorithm\nbased on a novel sparsity-inducing optimization framework. This method is\nderived from the multiple observations, that is, semi-sparsity prior knowledge\nis more universally applicable, especially in areas where sparsity is not fully\nadmitted, such as polynomial-smoothing surfaces. We illustrate that this\nsemi-sparsity can be identified into a generalized $L_0$-norm minimization in\nhigher-order gradient domains, thereby giving rise to a new \"feature-aware\"\nfiltering method with a powerful simultaneous-fitting ability in both sparse\nfeatures (singularities and sharpening edges) and non-sparse regions\n(polynomial-smoothing surfaces). Notice that a direct solver is always\nunavailable due to the non-convexity and combinatorial nature of $L_0$-norm\nminimization. Instead, we solve the model based on an efficient half-quadratic\nsplitting minimization with fast Fourier transforms (FFTs) for acceleration. We\nfinally demonstrate its versatility and many benefits to a series of\nsignal/image processing and computer vision applications.\n"}
{"abstract": "  Multicore fibers are expected to be a game-changer in the coming decades\nthanks to their intrinsic properties, allowing a larger transmission bandwidth\nand a lower footprint in optical communications. In addition, multicore fibers\nhave recently been explored for quantum communication, attesting their\nuniqueness in transporting high-dimensional quantum states. However,\ninvestigations and experiments reported in literature have been carried out in\nresearch laboratories, typically making use of short fiber links in controlled\nenvironments. Thus, the possibility of using long distance multicore fibers for\nquantum applications is still to be proven. We here characterize for the first\ntime, in terms of phase stability, multiple strands of a 4-core multicore fiber\ninstalled underground in the city of L'Aquila, with an overall fiber length up\nto about 25 km. In this preliminary study, we investigate the possibility of\nusing such an infrastructure to implement quantum-enhanced schemes, such as\nhigh-dimensional quantum key distribution, quantum-based environmental sensors,\nand more in general quantum communication protocols.\n"}
{"abstract": "  The standard classic special relativistic transformation of the\nelectromagnetic (EM) field under proper Lorentz transformations is revisited.\nAs to the pure Lorentz-boosts, popular treatments on EM transformation\ncontemplate ideal geometries generating special static charge and steady\ncurrent distributions and in conjunction, invoke parallel and perpendicular (to\nthe boost-velocity) components of the fields so engendered; the outcomes\nsubsequently being suitably generalized. We demonstrate ab initio, the EM field\ntransformation under arbitrary oblique uniform relative boost velocities from a\nbroader, rigorous yet remarkably lucid perspective, exploiting the antisymmetry\nof the EM-field tensor and symmetry of the general Lorentz-boost transformation\npreemptively eliminating the labor of performing complicated matrix\nmultiplications. Gratifyingly, this instructive exercise manifestly encompasses\ntime varying fields and yields the transformation relations in a\ncoordinate-free form. Further, the EM field transformation under proper,\ninstantaneous 3D rotations is demonstrated to transform as a passive rotation\nof the coordinate axes, where the electric and magnetic fields transform\nindependently as purely electric and magnetic fields, respectively. The present\ndidactical exercise, amenable to be incorporated in graduate courses on\nrelativistic electrodynamics, should serve as an excellent illustration on how\npowerful symmetry arguments simplify the otherwise tedious calculations while\nendowing complete generality to the connections obtained.\n"}
{"abstract": "  We consider the Navier--Stokes--Fourier system describing the motion of a\ncompressible, viscous, and heat conducting fluid in a bounded domain with\ngeneral non-homogeneous Dirichlet boundary conditions for the velocity and the\nabsolute temperature, with the associated boundary conditions for the density\non the inflow part. We introduce a new concept of weak solution based on the\nsatisfaction of the entropy inequality together with a balance law for the\nballistic energy. We show the weak-strong uniqueness principle as well as the\nexistence of global-in-time solutions.\n"}
{"abstract": "  In this paper, we propose a known-plaintext attack (KPA) method based on deep\nlearning for traditional chaotic encryption scheme. We employ the convolutional\nneural network to learn the operation mechanism of chaotic cryptosystem, and\naccept the trained network as the final decryption system. To evaluate the\nattack performance of different networks on different chaotic cryptosystem, we\nadopt two neural networks to perform known-plaintext attacks on two distinct\nchaotic encryption schemes. The experimental results demonstrate the potential\nof deep learning-based method for known-plaintext attack against chaotic\ncryptosystem. Different from the previous known-plaintext attack methods, which\nwere usually limited to a specific chaotic cryptosystem, a neural network can\nbe applied to the cryptanalysis of various chaotic cryptosystems with deep\nlearning-based approach, while several different networks can be designed for\nthe cryptanalysis of chaotic cryptosystems. This paper provides a new idea for\nthe cryptanalysis of chaotic image encryption algorithm.\n"}
{"abstract": "  Let $R$ be the power series ring or the polynomial ring over a field $k$ and\nlet $I $ be an ideal of $R.$ Macaulay proved that the Artinian Gorenstein\n$k$-algebras $R/I$ are in one-to-one correspondence with the cyclic\n$R$-submodules of the divided power series ring $\\Gamma. $ The result is\neffective in the sense that any polynomial of degree $s$ produces an Artinian\nGorenstein $k$-algebra of socle degree $s.$ In a recent paper, the authors\nextended Macaulay's correspondence characterizing the $R$-submodules of $\\Gamma\n$ in one-to-one correspondence with Gorenstein d-dimensional $k$-algebras.\nHowever, these submodules in positive dimension are not finitely generated. Our\ngoal is to give constructive and finite procedures for the construction of\nGorenstein $k$-algebras of dimension one and any codimension. This has been\nachieved through a deep analysis of the $G$-admissible submodules of $\\Gamma. $\nApplications to the Gorenstein linkage of zero-dimensional schemes and to\nGorenstein affine semigroup rings are discussed.\n"}
{"abstract": "  The glissadic overshoot is characterized by an unwanted type of movement\nknown as glissades. The glissades are a short ocular movement that describe the\nfailure of the neural programming of saccades to move the eyes in order to\nreach a specific target. In this paper we develop a procedure to determine if a\nspecific saccade have a glissade appended to the end of it. The use of the\nthird partial sum of the Gauss series as mathematical model, a comparison\nbetween some specific parameters and the RMSE error are the steps made to reach\nthis goal. Finally a machine learning algorithm is trained, returning expected\nresponses of the presence or not of this kind of ocular movement.\n"}
{"abstract": "  In this article, we present a $T$-matrix method for numerical computation of\nsecond-harmonic generation from clusters of arbitrarily distributed spherical\nparticles made of centrosymmetric optical materials. The electromagnetic fields\nat the fundamental and second-harmonic (SH) frequencies are expanded in series\nof vector spherical wave functions, and the single sphere $T$-matrix entries\nare computed by imposing field boundary conditions at the surface of the\nparticles. Different from previous approaches, we compute the SH fields by\ntaking into account both local surface and nonlocal bulk polarization sources,\nwhich allows one to accurately describe the generation of SH in arbitrary\nclusters of spherical particles. Our numerical method can be used to\nefficiently analyze clusters of spherical particles made of various optical\nmaterials, including metallic, dielectric, semiconductor, and polaritonic\nmaterials.\n"}
{"abstract": "  Gross and Siebert identified a class of singular Lagrangian torus fibrations\nwhich arise when smoothing toroidal degenerations, and which come in pairs that\nare related by mirror symmetry. We identify an immersed Lagrangian in each of\nthese local models which supports a moduli space of objects that is isomorphic\nto the mirror space, and prove a homological mirror statement along the way.\n"}
{"abstract": "  In their seminal paper Artemov and Protopopescu provide Hilbert formal\nsystems, Brower-Heyting-Kolmogorov and Kripke semantics for the logics of\nintuitionistic belief and knowledge. Subsequently Krupski has proved that the\nlogic of intuitionistic knowledge is PSPACE-complete and Su and Sano have\nprovided calculi enjoying the subformula property. This paper continues the\ninvestigations around to sequent calculi for Intuitionistic Epistemic Logics by\nproviding sequent calculi that have the subformula property and that are\nterminating in linear depth. Our calculi allow us to design a procedure that\nfor invalid formulas returns a Kripke model of minimal depth. Finally we also\ndiscuss refutational sequent calculi, that is sequent calculi to prove the\ninvalidity.\n"}
{"abstract": "  We present the results of an all-sky search for continuous gravitational wave\nsignals with frequencies in the 20-500 Hz range from neutron stars with\nellipticity of 1e-8. This frequency region is particularly hard to probe\nbecause of the quadratic dependence of signal strength on frequency. The search\nemploys the Falcon analysis pipeline on LIGO O2 public data. Compared to\nprevious Falcon analyses the coherence length has been quadrupled, with a\ncorresponding increase in sensitivity. This enables us to search for small\nellipticity neutron stars in this low frequency region up to 44 pc away. The\nfrequency derivative range is up to 3e-13 Hz/s easily accommodating sources\nwith ellipticities of 1e-7 and a corresponding factor of 10 increase in reach.\nNew outliers are found, many of which we are unable to associate with any\ninstrumental cause.\n"}
{"abstract": "  This paper investigates the design and performance of delayed bit-interleaved\ncoded modulation (DBICM) with low-density parity-check (LDPC) codes. For Gray\nlabeled square $M$-ary quadrature amplitude modulation (QAM) constellations, we\ninvestigate the optimal delay scheme with the largest spectrum efficiency of\nDBICM for a fixed maximum number of delayed time slots and a given\nsignal-to-noise ratio. When analyzing the capacity of DBICM, we find two\nimportant properties: the capacity improvement due to delayed coded bits being\nmapped to the real and imaginary parts of the transmitted symbols are\nindependent of each other; a pair of delay schemes with delayed coded bits\nhaving identical bit-channel capacity lead to equivalent DBICM capacity. Using\nthese two properties, we efficiently optimize the delay scheme for any uniform\nGray-QAM systems. Furthermore, these two properties enable efficient LDPC code\ndesigns regarding unequal error protection via bit-channel type\nclassifications. Moreover, we use protograph-based extrinsic information\ntransfer charts to jointly optimize degree distributions and channel\nassignments of LDPC codes and propose a constrained progressive edge growth\nlike algorithm to jointly construct LDPC codes and bit-interleavers for DBICM,\ntaking distinctive bit-channel's capacity into account. Simulation results\ndemonstrate that the designed LDPC coded DBICM systems significantly outperform\nLDPC coded BICM systems.\n"}
{"abstract": "  In recent years the number of CubeSats (U-class spacecrafts) launched into\nspace has increased exponentially marking the dawn of the nanosatellite\ntechnology. In general these satellites have a much smaller mass budget\ncompared to conventional scientific satellites which limits shielding of\nscientific instruments against direct and indirect radiation in space. In this\npaper we present a simulation framework to quantify the signal in large\nfield-of-view gamma-ray scintillation detectors of satellites induced by\nX-ray/gamma-ray transients, by taking into account the response of the\ndetector. Furthermore, we quantify the signal induced by X-ray and particle\nbackground sources at a Low-Earth Orbit outside South Atlantic Anomaly and\npolar regions. Finally, we calculate the signal-to-noise ratio taking into\naccount different energy threshold levels. Our simulation can be used to\noptimize material composition and predict detectability of various\nastrophysical sources by CubeSats. We apply the developed simulation to a\nsatellite belonging to a planned CAMELOT CubeSat constellation. This project\nmainly aims to detect short and long gamma-ray bursts (GRBs) and as a secondary\nscience objective, to detect soft gamma-ray repeaters (SGRs) and terrestrial\ngamma-ray flashes (TGFs). The simulation includes a detailed computer-aided\ndesign (CAD) model of the satellite to take into account the interaction of\nparticles with the material of the satellite as accurately as possible. Results\nof our simulations predict that CubeSats can complement the large space\nobservatories in high-energy astrophysics for observations of GRBs, SGRs and\nTGFs. For the detectors planned to be on board of the CAMELOT CubeSats the\nsimulations show that detections with signal-to-noise ratio of at least 9 for\nmedian GRB and SGR fluxes are achievable.\n"}
{"abstract": "  We study real-time routing policies in smart transit systems, where the\nplatform has a combination of cars and high-capacity vehicles (e.g., buses or\nshuttles) and seeks to serve a set of incoming trip requests. The platform can\nuse its fleet of cars as a feeder to connect passengers to its high-capacity\nfleet, which operates on fixed routes. Our goal is to find the optimal set of\n(bus) routes and corresponding frequencies to maximize the social welfare of\nthe system in a given time window. This generalizes the Line Planning Problem,\na widely studied topic in the transportation literature, for which existing\nsolutions are either heuristic (with no performance guarantees), or require\nextensive computation time (and hence are impractical for real-time use). To\nthis end, we develop a $1-\\frac{1}{e}-\\varepsilon$ approximation algorithm for\nthe Real-Time Line Planning Problem, using ideas from randomized rounding and\nthe Generalized Assignment Problem. Our guarantee holds under two assumptions:\n$(i)$ no inter-bus transfers and $(ii)$ access to a pre-specified set of\nfeasible bus lines. We moreover show that these two assumptions are crucial by\nproving that, if either assumption is relaxed, the Real-Time Line Planning\nProblem does not admit any constant-factor approximation. Finally, we\ndemonstrate the practicality of our algorithm via numerical experiments on\nreal-world and synthetic datasets, in which we show that, given a fixed time\nbudget, our algorithm outperforms Integer Linear Programming-based exact\nmethods.\n"}
{"abstract": "  Neural networks have proved to be very robust at processing unstructured data\nlike images, text, videos, and audio. However, it has been observed that their\nperformance is not up to the mark in tabular data; hence tree-based models are\npreferred in such scenarios. A popular model for tabular data is boosted trees,\na highly efficacious and extensively used machine learning method, and it also\nprovides good interpretability compared to neural networks. In this paper, we\ndescribe a novel architecture XBNet, which tries to combine tree-based models\nwith that of neural networks to create a robust architecture trained by using a\nnovel optimization technique, Boosted Gradient Descent for Tabular Data which\nincreases its interpretability and performance.\n"}
{"abstract": "  This article summarizes some of the open questions in the field of active\nmatter that have emerged during Active20, a nine-week program held at the Kavli\nInstitute for Theoretical Physics (KITP) in Spring 2020. The article does not\nprovide a review of the field, but rather a personal view of the authors,\ninformed by contributions of all participants, on new directions in active\nmatter research. The topics highlighted include: the ubiquitous occurrence of\nspontaneous flows and active turbulence and the theoretical and experimental\nchallenges associated with controlling and harnessing such flows; the role of\nmotile topological defects in ordered states of active matter and their\npossible biological relevance; the emergence of non-reciprocal effective\ninteractions and the role of chirality in active systems and their intriguing\nconnections to non-Hermitian quantum mechanics; the progress towards a\nformulation of the thermodynamics of active systems thanks to the feedback\nbetween theory and experiments; the impact of the active matter framework on\nour understanding of the emergent mechanics of biological tissue. These\nseemingly diverse phenomena all stem from the defining property of active\nmatter - assemblies of self-driven entities that individually break\ntime-reversal symmetry and collectively organize in a rich variety of\nnonequilibrium states.\n"}
{"abstract": "  This paper is devoted to investigating the Freidlin-Wentzell's large\ndeviation principle for a class of McKean-Vlasov quasilinear SPDEs perturbed by\nsmall multiplicative noise. We adopt the variational framework and the modified\nweak convergence criteria to prove the Laplace principle for McKean-Vlasov type\nSPDEs, which is equivalent to the large deviation principle. Moreover, we do\nnot assume any compactness condition of embedding in the Gelfand triple to\nhandle both the cases of bounded and unbounded domains in applications. The\nmain results can be applied to various McKean-Vlasov type SPDEs such as\ndistribution dependent stochastic porous media type equations and stochastic\np-Laplace type equations.\n"}
{"abstract": "  Students in quantum mechanics class are taught that the wave function\ncontains all knowable information about an isolated system. Later in the\ncourse, this view seems to be contradicted by the mysterious density matrix,\nwhich introduces a new set of probabilities in addition to those that are built\ninto the wave function. This paper brings attention to the fact that the\ndensity matrix can be reconciled with the underlying quantum-mechanical\ndescription using a two-particle entangled state with a one-particle subsystem\nas the simplest illustration of the basic principle. The extra-quantum\nprobabilities are traced to the coefficients of superposition of the quantum\nstate vector and the seemingly irreversible exponential population decay is\nshown to be compatible with the unitary time evolution of a pure state when the\ntwo particles interact. The two-particle universe thus provides the student\nwith a tool for understanding how the density operator, with all its richness,\nemerges from quantum mechanics.\n"}
{"abstract": "  We study the electronic properties of a position-dependent effective mass\nelectron on a bilayer graphene catenoid bridge. We propose a position-dependent\nmass (PDM) as a function of both gaussian and mean curvature. The hamiltonian\nexhibits parity and time-reversal steaming from the bridge symmetry. The\neffective potential contains the da Costa, centrifugal and PDM terms which are\nconcentrated around the catenoid bridge. For zero angular momentum states, the\nPDM term provides a transition between a reflectionless to a double-well\npotential. As a result, the bound states undergo a transition from a single\nstate around the bridge throat into two states each one located at rings around\nthe bridge. Above some critical value of the PDM coupling constant, the\ndegeneracy is restored due to double-well tunneling resonance.\n"}
{"abstract": "  Realizing optical manipulation of microscopic objects is crucial in the\nresearch fields of life science, condensed matter physics and physical\nchemistry. In non-liquid environments, this task is commonly regarded as\ndifficult due to strong adhesive surface force ($\\sim\\mu\\rm N$) between solid\ninterfaces that makes tiny optical driven force ($\\sim\\rm pN$) insignificant.\nHere, by recognizing the microscopic interaction mechanism between friction\nforce -- the parallel component of surface force on the contact surface -- and\nthermoelastic waves induced by pulsed optical absorption, we establish a\ngeneral principle enabling the actuation of micro-objects on dry frictional\nsurfaces based on the opto-thermo-mechanical effects. Theoretically, we predict\nthat nanosecond pulsed optical absorption with mW-scale peak power is\nsufficient to tame $\\mu\\rm N$-scale friction force. Experimentally, we\ndemonstrate that two-dimensional spiral motion of gold plates on micro-fibers\ndriven by a nanosecond pulsed laser, and reveal the specific rules of motion\ncontrol. Our results pave the way for future development of micro-scale\nactuators in nonliquid environments.\n"}
{"abstract": "  We consider the Cauchy problem for the heat diffusion equation in the whole\nEuclidean space consisting of two media with different constant conductivities.\nThe large time behavior of temperature, the solution of the problem, is studied\nwhen initially temperature is assigned to be 0 on one medium and $1$ on the\nother. We show that under a certain geometric condition of the configuration of\nthe media, temperature is stabilized to a constant as time tends to infinity.\nWe also show by examples that temperature in general oscillates and is not\nstabilized.\n"}
{"abstract": "  We report the discovery of one super-Earth- (TOI-1749b) and two\nsub-Neptune-sized planets (TOI-1749c and TOI-1749d) transiting an early M dwarf\nat a distance of 100~pc, which were first identified as planetary candidates\nusing data from the TESS photometric survey. We have followed up this system\nfrom the ground by means of multiband transit photometry, adaptive-optics\nimaging, and low-resolution spectroscopy, from which we have validated the\nplanetary nature of the candidates. We find that TOI-1749b, c, and d have\norbital periods of 2.39, 4.49, and 9.05 days, and radii of 1.4, 2.1, and 2.5\n$R_\\oplus$, respectively. We also place 95\\% confidence upper limits on the\nmasses of 57, 14, and 15 $M_\\oplus$ for TOI-1749b, c, and d, respectively, from\ntransit timing variations. The periods, sizes, and tentative masses of these\nplanets are in line with a scenario in which all three planets initially had a\nhydrogen envelope on top of a rocky core, and only the envelope of the\ninnermost planet has been stripped away by photoevaporation and/or core-powered\nmass loss mechanisms. These planets are similar to other planetary trios found\naround M dwarfs, such as TOI-175b,c,d and TOI-270b,c,d, in the sense that the\nouter pair has a period ratio within 1\\% of 2. Such a characteristic orbital\nconfiguration, in which an additional planet is located interior to a near 2:1\nperiod-ratio pair, is relatively rare around FGK dwarfs.\n"}
{"abstract": "  In order to investigate the tomographical structure of light pseudoscalar\nmesons, particularly, pion and kaon, we study the valence quark distribution\nfunction (PDF) and the generalized parton distributions (GPDs) using light-cone\nquark model (LCQM).\n"}
{"abstract": "  Estimation of population size using incomplete lists (also called the\ncapture-recapture problem) has a long history across many biological and social\nsciences. For example, human rights and other groups often construct partial\nand overlapping lists of victims of armed conflicts, with the hope of using\nthis information to estimate the total number of victims. Earlier statistical\nmethods for this setup either use potentially restrictive parametric\nassumptions, or else rely on typically suboptimal plug-in-type nonparametric\nestimators; however, both approaches can lead to substantial bias, the former\nvia model misspecification and the latter via smoothing. Under an identifying\nassumption that two lists are conditionally independent given measured\ncovariate information, we make several contributions. First, we derive the\nnonparametric efficiency bound for estimating the capture probability, which\nindicates the best possible performance of any estimator, and sheds light on\nthe statistical limits of capture-recapture methods. Then we present a new\nestimator, and study its finite-sample properties, showing that it has a double\nrobustness property new to capture-recapture, and that it is near-optimal in a\nnon-asymptotic sense, under relatively mild nonparametric conditions. Next, we\ngive a method for constructing confidence intervals for total population size\nfrom generic capture probability estimators, and prove non-asymptotic\nnear-validity. Finally, we study our methods in simulations, and apply them to\nestimate the number of killings and disappearances attributable to different\ngroups in Peru during its internal armed conflict between 1980 and 2000.\n"}
{"abstract": "  With the increasing popularity of cloud based machine learning (ML)\ntechniques there comes a need for privacy and integrity guarantees for ML data.\nIn addition, the significant scalability challenges faced by DRAM coupled with\nthe high access-times of secondary storage represent a huge performance\nbottleneck for ML systems. While solutions exist to tackle the security aspect,\nperformance remains an issue. Persistent memory (PM) is resilient to power loss\n(unlike DRAM), provides fast and fine-granular access to memory (unlike disk\nstorage) and has latency and bandwidth close to DRAM (in the order of ns and\nGB/s, respectively). We present PLINIUS, a ML framework using Intel SGX\nenclaves for secure training of ML models and PM for fault tolerance\nguarantees. P LINIUS uses a novel mirroring mechanism to create and maintain\n(i) encrypted mirror copies of ML models on PM, and (ii) encrypted training\ndata in byte-addressable PM, for near-instantaneous data recovery after a\nsystem failure. Compared to disk-based checkpointing systems,PLINIUS is 3.2x\nand 3.7x faster respectively for saving and restoring models on real PM\nhardware, achieving robust and secure ML model training in SGX enclaves.\n"}
{"abstract": "  Non-interacting spinless electrons in one-dimensional quasicrystals,\ndescribed by the Aubry-Andr\\'{e}-Harper (AAH) Hamiltonian with nearest\nneighbour hopping, undergoes metal to insulator transition (MIT) at a critical\nstrength of the quasi-periodic potential. This transition is related to the\nself-duality of the AAH Hamiltonian. Interestingly, at the critical point,\nwhich is also known as the self-dual point, all the single particle wave\nfunctions are multifractal or non-ergodic in nature, while they are ergodic and\ndelocalized (localized) below (above) the critical point. In this work, we have\nstudied the one dimensional quasi-periodic AAH Hamiltonian in the presence of\nspin-orbit (SO) coupling of Rashba type, which introduces an additional spin\nconserving complex hopping and a spin-flip hopping. We have found that,\nalthough the self-dual nature of AAH Hamiltonian remains unaltered, the\nself-dual point gets affected significantly. Moreover, the effect of the\ncomplex and spin-flip hoppings are identical in nature. We have extended the\nidea of Kohn's localization tensor calculations for quasi-particles and\ndetected the critical point very accurately. These calculations are followed by\ndetailed multifractal analysis along with the computation of inverse\nparticipation ratio and von Neumann entropy, which clearly demonstrate that the\nquasi-particle eigenstates are indeed multifractal and non-ergodic at the\ncritical point. Finally, we mapped out the phase diagram in the parameter space\nof quasi-periodic potential and SO coupling strength.\n"}
{"abstract": "  This paper presents a novel deep learning-based approach for automatically\nvectorizing and synthesizing the clipart of man-made objects. Given a raster\nclipart image and its corresponding object category (e.g., airplanes), the\nproposed method sequentially generates new layers, each of which is composed of\na new closed path filled with a single color. The final result is obtained by\ncompositing all layers together into a vector clipart image that falls into the\ntarget category. The proposed approach is based on an iterative generative\nmodel that (i) decides whether to continue synthesizing a new layer and (ii)\ndetermines the geometry and appearance of the new layer. We formulated a joint\nloss function for training our generative model, including the shape\nsimilarity, symmetry, and local curve smoothness losses, as well as vector\ngraphics rendering accuracy loss for synthesizing clipart recognizable by\nhumans. We also introduced a collection of man-made object clipart, ClipNet,\nwhich is composed of closed-path layers, and two designed preprocessing tasks\nto clean up and enrich the original raw clipart. To validate the proposed\napproach, we conducted several experiments and demonstrated its ability to\nvectorize and synthesize various clipart categories. We envision that our\ngenerative model can facilitate efficient and intuitive clipart designs for\nnovice users and graphic designers.\n"}
{"abstract": "  Recently, a new measure of information called extropy has been introduced by\nLad, Sanfilippo and Agr\\`o as the dual version of Shannon entropy. In the\nliterature, Tsallis introduced a measure for a discrete random variable, named\nTsallis entropy, as a generalization of Boltzmann-Gibbs statistics. In this\nwork, a new measure of discrimination, called Tsallis extropy, is introduced\nand some of its properties are then discussed. The relation between Tsallis\nextropy and entropy is given and some bounds are also presented. Finally, an\napplication of this extropy to pattern recognition is demonstrated.\n"}
{"abstract": "  The near threshold photo or electroproduction of heavy vector quarkonium off\nthe proton is studied in quantum chromodynamics. Similar to the high-energy\nlimit, the production amplitude can be factorized in terms of gluonic\nGeneralized Parton Distributions and the quarkonium distribution amplitude. At\nthe threshold, the threshold kinematics has a large skewness parameter $\\xi$,\nleading to the dominance of the spin-2 contribution over higher-spin twist-2\noperators. Thus threshold production data are useful to extract the gluonic\ngravitational form factors, allowing studying the gluonic contributions to the\nquantum anomalous energy, mass radius, spin and mechanical pressure in the\nproton. We use the recent GlueX data on the $J/\\psi$ photoproduction to\nillustrate the potential physics impact from the high-precision data from\nfuture JLab 12 GeV and EIC physics program.\n"}
{"abstract": "  Electron microscopy (EM) enables the reconstruction of neural circuits at the\nlevel of individual synapses, which has been transformative for scientific\ndiscoveries. However, due to the complex morphology, an accurate reconstruction\nof cortical axons has become a major challenge. Worse still, there is no\npublicly available large-scale EM dataset from the cortex that provides dense\nground truth segmentation for axons, making it difficult to develop and\nevaluate large-scale axon reconstruction methods. To address this, we introduce\nthe AxonEM dataset, which consists of two 30x30x30 um^3 EM image volumes from\nthe human and mouse cortex, respectively. We thoroughly proofread over 18,000\naxon instances to provide dense 3D axon instance segmentation, enabling\nlarge-scale evaluation of axon reconstruction methods. In addition, we densely\nannotate nine ground truth subvolumes for training, per each data volume. With\nthis, we reproduce two published state-of-the-art methods and provide their\nevaluation results as a baseline. We publicly release our code and data at\nhttps://connectomics-bazaar.github.io/proj/AxonEM/index.html to foster the\ndevelopment of advanced methods.\n"}
{"abstract": "  The negative sign of the anomalous Nernst thermopower ($S_\\text{ANE}$)\nobserved in Mn-Ga ordered alloys is an attractive property for thermoelectric\napplications exploiting the anomalous Nernst effect (ANE); however, its origin\nhas not been clarified. In this study, to gain insight into the negative\n$S_\\text{ANE}$, we prepared epitaxial thin films of Mn$_{x}$Ga$_{100-x}$ with\n$x$ ranging from 56.2 to 71.7, and systematically investigated the structural,\nmagnetic, and transport properties including the anomalous Hall effect (AHE)\nand the ANE. The measured $S_\\text{ANE}$ is negative for all samples and shows\nclose to one order of magnitude difference among different compositions.\nTogether with the measured transport properties, we were able to separate the\ntwo different contributions of the ANE, i.e., one originating from the\ntransverse thermoelectric coefficient ($\\alpha_{xy}$), and the other one\noriginating from the AHE acting on the longitudinal carrier flow induced by the\nSeebeck effect. Both contributions are found to be negative for all samples,\nwhile the experimentally obtained negative $\\alpha_{xy}$ exhibits a monotonic\nincrease towards zero with increasing $x$, which is consistent with the\ntendency indicated by first-principles calculations. Our results show that the\nlarge difference in the negative $S_\\text{ANE}$ is mostly attributed to\n$\\alpha_{xy}$, and thus shed light on further enhancement of the ANE in\nMn-based ordered alloys.\n"}
{"abstract": "  The goal of this paper is to present a theoretical and practical introduction\nto generalized eigendecomposition (GED), which is a robust and flexible\nframework used for dimension reduction and source separation in multichannel\nsignal processing. In cognitive electrophysiology, GED is used to create\nspatial filters that maximize a researcher-specified contrast, such as relative\nspectral power or experiment condition differences. GED is fast and easy to\ncompute, performs well in simulated and real data, and is easily adaptable to a\nvariety of specific research goals. This paper introduces GED in a way that\nties together myriad individual publications and applications of GED in\nelectrophysiology, and provides sample MATLAB and Python code that can be\ntested and adapted. Practical considerations and issues that often arise in\napplications are discussed.\n"}
{"abstract": "  Thin-film optical filters can nowadays be integrated onto pixels of\ncommercial image sensors used for spectral imaging. A drawback of having more\nfilters on an image sensor is a loss in spatial resolution which could be\nregained by using smaller pixels. However, this work shows that miniaturization\ncauses a reduction in transmittance and requires a trade-off between pixel size\nand filter bandwidth. This is caused by diffraction and the reduced number of\nreflections in the filter. Analytical wave and ray optics models are developed\nand used for fast transmittance prediction and this is validated using a\ncommercial spectral sensor and numerical simulations. The identified trade-off\nis an intrinsic limitation of pixel-integrated thin-film filter technology and\nthe developed models will be useful tools for exploring new hardware and\ncomputational solutions.\n"}
{"abstract": "  Topological insulator lasers (TILs) are a recently introduced family of\nlasing arrays in which phase locking is achieved through synthetic gauge\nfields. These single frequency light source arrays operate in the spatially\nextended edge modes of topologically non-trivial optical lattices. Because of\nthe inherent robustness of topological modes against perturbations and defects,\nsuch topological insulator lasers tend to demonstrate higher slope efficiencies\nas compared to their topologically trivial counterparts. So far, magnetic and\nnon-magnetic optically pumped topological laser arrays as well as electrically\npumped TILs that are operating at cryogenic temperatures have been\ndemonstrated. Here we present the first room temperature and electrically\npumped topological insulator laser. This laser array, using a structure that\nmimics the quantum spin Hall effect for photons, generates light at telecom\nwavelengths and exhibits single frequency emission. Our work is expected to\nlead to further developments in laser science and technology, while opening up\nnew possibilities in topological photonics.\n"}
{"abstract": "  The unprecedented behavioural responses of societies have been evidently\nshaping the COVID-19 pandemic, yet it is a significant challenge to accurately\nmonitor the continuously changing social mixing patterns in real-time. Contact\nmatrices, usually stratified by age, summarise interaction motifs efficiently,\nbut their collection relies on conventional representative survey techniques,\nwhich are expensive and slow to obtain. Here we report a data collection effort\ninvolving over $2.3\\%$ of the Hungarian population to simultaneously record\ncontact matrices through a longitudinal online and sequence of representative\nphone surveys. To correct non-representative biases characterising the online\ndata, by using census data and the representative samples we develop a\nreconstruction method to provide a scalable, cheap, and flexible way to\ndynamically obtain closer-to-representative contact matrices. Our results\ndemonstrate the potential of combined online-offline data collections to\nunderstand the changing behavioural responses determining the future evolution\nof the outbreak, and inform epidemic models with crucial data.\n"}
{"abstract": "  Star-forming galaxies (SFGs) emit non-thermal radiation from radio to\ngamma-rays. We aim to investigate the main mechanisms of global CR transport\nand cooling in SFGs. The way they contribute in shaping the relations between\nnon-thermal luminosities and SFR could shed light onto their nature. We develop\na model to compute the CR populations of SFGs, taking into account their\nproduction, transport, and cooling. The model is parameterised only through\nglobal galaxy properties, and describes the non-thermal emission in both radio\nand gamma-rays. We focus on the role of diffusive and advective transport by\ngalactic winds, either driven by turbulent or thermal instabilities. We compare\nmodel predictions to observations, for which we compile a homogeneous set of\nluminosities in these radio bands, and update those available in gamma-rays.\nOur model reproduces reasonably well the observed relations between the\ngamma-ray or 1.4 GHz radio luminosities and the SFR, assuming a single\npower-law scaling of the magnetic field with the latter with index beta=0.3,\nand winds blowing either at Alfvenic speeds or typical starburst wind\nvelocities. Escape of CR is negligible for > 30 Mo/yr. A constant ionisation\nfraction of the interstellar medium fails to reproduce the 150 MHz radio\nluminosity throughout the whole SFR range. Our results reinforce the idea that\ngalaxies with high SFR are CR calorimeters, and that the main mechanism driving\nproton escape is diffusion, whereas electron escape also proceeds via wind\nadvection. They also suggest that these winds should be CR or thermally-driven\nat low and intermediate SFR, respectively. Our results globally support that\nmagnetohydrodynamic turbulence is responsible for the dependence of the\nmagnetic field strength on the SFR and that the ionisation fraction is strongly\ndisfavoured to be constant throughout the whole SFR range.\n"}
{"abstract": "  Link Prediction, addressing the issue of completing KGs with missing facts,\nhas been broadly studied. However, less light is shed on the ubiquitous\nhyper-relational KGs. Most existing hyper-relational KG embedding models still\ntear an n-ary fact into smaller tuples, neglecting the indecomposability of\nsome n-ary facts. While other frameworks work for certain arity facts only or\nignore the significance of primary triple. In this paper, we represent an n-ary\nfact as a whole, simultaneously keeping the integrity of n-ary fact and\nmaintaining the vital role that the primary triple plays. In addition, we\ngeneralize hyperbolic Poincar\\'e embedding from binary to arbitrary arity data,\nwhich has not been studied yet. To tackle the weak expressiveness and high\ncomplexity issue, we propose HYPER^2 which is qualified for capturing the\ninteraction between entities within and beyond triple through information\naggregation on the tangent space. Extensive experiments demonstrate HYPER^2\nachieves superior performance to its translational and deep analogues,\nimproving SOTA by up to 34.5\\% with relatively few dimensions. Moreover, we\nstudy the side effect of literals and we theoretically and experimentally\ncompare the computational complexity of HYPER^2 against several best performing\nbaselines, HYPER^2 is 49-61 times quicker than its counterparts.\n"}
{"abstract": "  All global symmetries are expected to be explicitly broken by quantum\ngravitational effects, and yet may play an important role in Particle Physics\nand Cosmology. As such, any evidence for a well-preserved global symmetry would\ngive insight into an important feature of gravity. We argue that a recently\nreported $2.4\\sigma$ detection of cosmic birefringence in the Cosmic Microwave\nBackground could be the first observational indication of a well-preserved\n(although spontaneously broken) global symmetry in nature. A compelling\nsolution to explain this measurement is a very light pseudoscalar field that\ninteracts with electromagnetism. In order for gravitational effects not to lead\nto large corrections to the mass of this scalar field, we show that the\nbreaking of global symmetries by gravity should be bounded above. Finally, we\nhighlight that any bound of this type would have clear implications for the\nconstruction of theories of quantum gravity, as well as for many particle\nphysics scenarios.\n"}
{"abstract": "  Intrusion Detection Systems (IDS) are now an essential element when it comes\nto securing computers and networks. Despite the huge research efforts done in\nthe field, handling sources' reliability remains an open issue. To address this\nproblem, this paper proposes a novel contextual discounting method based on\nsources' reliability and their distinguishing ability between normal and\nabnormal behavior. Dempster-Shafer theory, a general framework for reasoning\nunder uncertainty, is used to construct an evidential classifier. The NSL-KDD\ndataset, a significantly revised and improved version of the existing KDDCUP'99\ndataset, provides the basis for assessing the performance of our new detection\napproach. While giving comparable results on the KDDTest+ dataset, our approach\noutperformed some other state-of-the-art methods on the KDDTest-21 dataset\nwhich is more challenging.\n"}
{"abstract": "  We study the fragmentation of He$_2$ dimers into He$^+$ ions by relativistic\nhighly-charged projectiles. It is demonstrated that the interaction between an\nultrafast projectile with an extremely extended object -- the helium dimer --\npossesses interesting features which are absent in collisions with \"normal\"\nmolecules. It is also shown that such projectiles, due to their enormous\ninteraction range, can accurately probe the ground state of the dimer and even\nbe used for a precise determination of its binding energy.\n"}
{"abstract": "  Let $E$ be a holomorphic vector bundle over a compact K\\\"{a}hler manifold\n$(X,\\omega)$ with negative sectional curvature $sec\\leq -K<0$, $\\Delta_{E}$ be\nthe Chern connection on $E$. In this article we show that if\n$C:=|[\\Lambda,i\\Theta(E)]|\\leq c_{n}K$, then $(X,E)$ satisfy a family of Chern\nnumber inequalities. The main idea in our proof is study the $L^{2}$\n$\\bar{\\partial}_{\\tilde{E}}$-harmonic forms on lifting bundle $\\tilde{E}$ over\nthe universal covering space $\\tilde{X}$. We also observe that there is a\nclosely relationship between the eigenvalue of the Laplace-Beltrami operator\n$\\Delta_{\\bar{\\partial}_{\\tilde{E}}}$ and the Euler characteristic of $X$.\nPrecisely, if there is a line bundle $L$ on $X$ such that\n$\\chi^{p}(X,L^{\\otimes m})$ is not constant for some integers $p\\in[0,n]$, then\nthe Euler characteristic of $X$ satisfies $(-1)^{n}\\chi(X)\\geq\n(n+1)+\\lfloor\\frac{c_{n}K}{2nC} \\rfloor$.\n"}
{"abstract": "  Music information is often conveyed or recorded across multiple data\nmodalities including but not limited to audio, images, text and scores.\nHowever, music information retrieval research has almost exclusively focused on\nsingle modality recognition, requiring development of separate models for each\nmodality. Some multi-modal works require multiple coexisting modalities given\nto the model as inputs, constraining the use of these models to the few cases\nwhere data from all modalities are available. To the best of our knowledge, no\nexisting model has the ability to take inputs from varying modalities, e.g.\nimages or sounds, and classify them into unified music categories. We explore\nthe use of cross-modal retrieval as a pretext task to learn modality-agnostic\nrepresentations, which can then be used as inputs to classifiers that are\nindependent of modality. We select instrument classification as an example task\nfor our study as both visual and audio components provide relevant semantic\ninformation. We train music instrument classifiers that can take both images or\nsounds as input, and perform comparably to sound-only or image-only\nclassifiers. Furthermore, we explore the case when there is limited labeled\ndata for a given modality, and the impact in performance by using labeled data\nfrom other modalities. We are able to achieve almost 70% of best performing\nsystem in a zero-shot setting. We provide a detailed analysis of experimental\nresults to understand the potential and limitations of the approach, and\ndiscuss future steps towards modality-agnostic classifiers.\n"}
{"abstract": "  One of the most celebrated contributions to the study of the mechanical\nbehavior of materials is due to J.D. Eshelby, who in the late 50s\nrevolutionized our understanding of the elastic stress and strain fields due to\nan ellipsoidal inclusion/inhomogeneity that undergoes a transformation of shape\nand size. While Eshelby's work laid the foundation for significant advancements\nin various fields, including fracture mechanics, theory of phase transitions,\nand homogenization methods, its extension into the range of large deformations,\nand to situations in which the material can actively reorganize in response to\nthe finite transformation strain, is in a nascent state. Beyond the theoretical\ndifficulties imposed by highly nonlinear material response, a major hindrance\nhas been the absence of experimental observations that can elucidate the\nintricacies that arise in this regime. To address this limitation, our\nexperimental observations reveal the key morphogenesis steps of Vibrio cholerae\nbiofilms embedded in hydrogels, as they grow by four orders of magnitude from\ntheir initial size. Using the biofilm growth as a case study, our theoretical\nmodel considers various growth scenarios and employs two different and\ncomplimentary methods -- a minimal analytical model and finite element\ncomputations -- to obtain approximate equilibrium solutions. A particular\nemphasis is put on determining the natural growth path of an inclusion that\noptimizes its shape in response to the confinement, and the onset of damage in\nthe matrix, which together explain the observed behavior of biofilms. Beyond\nbacterial biofilms, this work sheds light on the role of mechanics in\ndetermining the morphogenesis pathways of confined growing bodies and thus\napplies to a broad range of phenomena that are ubiquitous in both natural and\nengineered material systems.\n"}
{"abstract": "  In this paper we define the algebraic entropy test for face-centered quad\nequations, which are equations defined on vertices of a quadrilateral plus an\nadditional interior vertex. This notion of algebraic entropy is applied to a\nrecently introduced class of these equations that satisfy a new form of\nmultidimensional consistency called consistency-around-a-face-centered-cube\n(CAFCC), whereby the system of equations is consistent on a face-centered cubic\nunit cell. It is found that for certain arrangements of equations (or pairs of\nequations) in the square lattice, all known CAFCC equations pass the algebraic\nentropy test possessing either quadratic or linear growth.\n"}
{"abstract": "  Stress and driving are a dangerous combination which can lead to crashes, as\nevidenced by the large number of road traffic crashes that involve stress.\nMotivated by the need to address the significant costs of driver stress, it is\nessential to build a practical system that can classify driver stress level\nwith high accuracy. However, the performance of an accurate driving stress\nlevels classification system depends on hyperparameter optimization choices\nsuch as data segmentation (windowing hyperparameters). The configuration\nsetting of hyperparameters, which has an enormous impact on the system\nperformance, are typically hand-tuned while evaluating the algorithm. This\ntuning process is time consuming and often depends on personal experience.\nThere are also no generic optimal values for hyperparameters values. In this\nwork, we propose a meta-heuristic approach to support automated hyperparameter\noptimization and provide a real-time driver stress detection system. This is\nthe first systematic study of optimizing windowing hyperparameters based on\nElectrocardiogram (ECG) signal in the domain of driving safety. Our approach is\nto propose a framework based on Particle Swarm Optimization algorithm (PSO) to\nselect an optimal/near optimal windowing hyperparameters values. The\nperformance of the proposed framework is evaluated on two datasets: a public\ndataset (DRIVEDB dataset) and our collected dataset using an advanced\nsimulator. DRIVEDB dataset was collected in a real time driving scenario, and\nour dataset was collected using an advanced driving simulator in the control\nenvironment. We demonstrate that optimising the windowing hyperparameters\nyields significant improvement in terms of accuracy. The most accurate built\nmodel applied to the public dataset and our dataset, based on the selected\nwindowing hyperparameters, achieved 92.12% and 77.78% accuracy, respectively.\n"}
{"abstract": "  This paper investigates the quadratic irrationals that arise as periodic\npoints of the Gauss type shift associated to the odd continued fraction\nexpansion (Schweiger). It is shown that these numbers, which we call O-reduced,\nare equidistributed with respect to the Lebesgue absolutely continuous\ninvariant probability measure of the Odd Gauss shift.\n"}
{"abstract": "  We calculate the energy levels and corresponding eigenstates of an\ninteracting scalar quantum field theory on a lattice using a\ncontinuous-variable version of the quantum imaginary time evolution algorithm.\nOnly a single qumode is needed for the simulation of the field at each point on\nthe lattice. Our quantum algorithm avoids the use of non-Gaussian quantum gates\nand relies, instead, on detectors projecting onto eigenstates of the\nphoton-number operator. Using Xanadu's Strawberry Fields simulator, we obtain\nresults on energy levels that are in very good agreement with results from\nexact calculations. We propose an experimental setup that can be realized with\nexisting technology.\n"}
{"abstract": "  Current computational-emotion research has focused on applying acoustic\nproperties to analyze how emotions are perceived mathematically or used in\nnatural language processing machine learning models. While recent interest has\nfocused on analyzing emotions from the spoken voice, little experimentation has\nbeen performed to discover how emotions are recognized in the singing voice --\nboth in noiseless and noisy data (i.e., data that is either inaccurate,\ndifficult to interpret, has corrupted/distorted/nonsense information like\nactual noise sounds in this case, or has a low ratio of usable/unusable\ninformation). Not only does this ignore the challenges of training machine\nlearning models on more subjective data and testing them with much noisier\ndata, but there is also a clear disconnect in progress between advancing the\ndevelopment of convolutional neural networks and the goal of emotionally\ncognizant artificial intelligence. By training a new model to include this type\nof information with a rich comprehension of psycho-acoustic properties, not\nonly can models be trained to recognize information within extremely noisy\ndata, but advancement can be made toward more complex biofeedback applications\n-- including creating a model which could recognize emotions given any human\ninformation (language, breath, voice, body, posture) and be used in any\nperformance medium (music, speech, acting) or psychological assistance for\npatients with disorders such as BPD, alexithymia, autism, among others. This\npaper seeks to reflect and expand upon the findings of related research and\npresent a stepping-stone toward this end goal.\n"}
{"abstract": "  The present paper is a follow-up of our previous paper that derives a\nslightly simplified model equation for the Klein-Gordon equation, describing\nthe propagation of a scalar field of mass $\\mu$ in the background of a rotating\nblack hole and, among others, supports the instability of the field down to\n$a/M \\approx 0.97$. The latter result was derived numerically. This paper gives\ncorresponding rigorous results, supporting instability of the field down to\n$a/M \\approx 0.979796$.\n"}
{"abstract": "  Building neural network classifiers with an ability to distinguish between in\nand out-of distribution inputs is an important step towards faithful deep\nlearning systems. Some of the successful approaches for this, resort to\narchitectural novelties, such as ensembles, with increased complexities in\nterms of the number of parameters and training procedures. Whereas some other\napproaches make use of surrogate samples, which are easy to create and work as\nproxies for actual out-of-distribution (OOD) samples, to train the networks for\nOOD detection. In this paper, we propose a very simple approach for enhancing\nthe ability of a pretrained network to detect OOD inputs without even altering\nthe original parameter values. We define a probabilistic trust interval for\neach weight parameter of the network and optimize its size according to the\nin-distribution (ID) inputs. It allows the network to sample additional weight\nvalues along with the original values at the time of inference and use the\nobserved disagreement among the corresponding outputs for OOD detection. In\norder to capture the disagreement effectively, we also propose a measure and\nestablish its suitability using empirical evidence. Our approach outperforms\nthe existing state-of-the-art methods on various OOD datasets by considerable\nmargins without using any real or surrogate OOD samples. We also analyze the\nperformance of our approach on adversarial and corrupted inputs such as\nCIFAR-10-C and demonstrate its ability to clearly distinguish such inputs as\nwell. By using fundamental theorem of calculus on neural networks, we explain\nwhy our technique doesn't need to observe OOD samples during training to\nachieve results better than the previous works.\n"}
{"abstract": "  A path system $\\mathcal{P}$ in a graph $G=(V,E)$ is said to be irreducible if\nthere does not exist a partition $V= A\\sqcup B$ such that $\\mathcal{P}$\nrestricts to a path system on both $G[A]$ and $G[B]$. In this paper, we\nconstruct an infinite family of non-metrizable irreducible path systems defined\non certain Paley graphs.\n"}
{"abstract": "  Moir\\'e superlattices in van der Waals heterostructures are gaining\nincreasing attention because they offer new opportunities to tailor and explore\nunique electronic phenomena when stacking 2D materials with small twist angles.\nHere, we reveal local surface potentials associated with stacking domains in\ntwisted double bilayer graphene (TDBG) moir\\'e superlattices. Using a\ncombination of both lateral Piezoresponse Force Microscopy (LPFM) and Scanning\nKelvin Probe Microscopy (SKPM), we distinguish between Bernal (ABAB) and\nrhombohedral (ABCA) stacked graphene and directly correlate these stacking\nconfigurations with local surface potential. We find that the surface potential\nof the ABCA domains is ~15 mV higher (smaller work function) than that of the\nABAB domains. First-principles calculations based on density functional theory\nfurther show that the different work functions between ABCA and ABAB domains\narise from the stacking dependent electronic structure. We show that, while the\nmoir\\'e superlattice visualized by LPFM can change with time, imaging the\nsurface potential distribution via SKPM appears more stable, enabling the\nmapping of ABAB and ABCA domains without tip-sample contact-induced effects.\nOur results provide a new means to visualize and probe local domain stacking in\nmoir\\'e superlattices along with its impact on electronic properties.\n"}
{"abstract": "  Many techniques have been proposed for quickly detecting and containing\nmalware-generated network attacks such as large-scale denial of service\nattacks; unfortunately, much damage is already done within the first few\nminutes of an attack, before it is identified and contained. There is a need\nfor an early warning system that can predict attacks before they actually\nmanifest, so that upcoming attacks can be prevented altogether by blocking the\nhosts that are likely to engage in attacks. However, blocking responses may\ndisrupt legitimate processes on blocked hosts; in order to minimise user\ninconvenience, it is important to also foretell the time when the predicted\nattacks will occur, so that only the most urgent threats result in\nauto-blocking responses, while less urgent ones are first manually\ninvestigated. To this end, we identify a typical infection sequence followed by\nmodern malware; modelling this sequence as a Markov chain and training it on\nreal malicious traffic, we are able to identify behaviour most likely to lead\nto attacks and predict 98\\% of real-world spamming and port-scanning attacks\nbefore they occur. Moreover, using a Semi-Markov chain model, we are able to\nforetell the time of upcoming attacks, a novel capability that allows\naccurately predicting the times of 97% of real-world malware attacks. Our work\nrepresents an important and timely step towards enabling flexible threat\nresponse models that minimise disruption to legitimate users.\n"}
{"abstract": "  A classic application of description length is for model selection with the\nminimum description length (MDL) principle. The focus of this paper is to\nextend description length for data analysis beyond simple model selection and\nsequences of scalars. More specifically, we extend the description length for\ndata analysis in Gaussian graphical models. These are powerful tools to model\ninteractions among variables in a sequence of i.i.d Gaussian data in the form\nof a graph. Our method uses universal graph coding methods to accurately\naccount for model complexity, and therefore provide a more rigorous approach\nfor graph model selection. The developed method is tested with synthetic and\nelectrocardiogram (ECG) data to find the graph model and anomaly in Gaussian\ngraphical models. The experiments show that our method gives better performance\ncompared to commonly used methods.\n"}
{"abstract": "  In this article we have conducted a reanalysis of the phase III aducanumab\n(ADU) summary statistics announced by Biogen, in particular the result of the\nClinical Dementia Rating-Sum of Boxes (CDR-SB). The results showed that the\nevidence on the efficacy of the drug is very low and a more clearer view of the\nresults of clinical trials are presented in the Bayesian framework that can be\nuseful for future development and research in the field.\n"}
{"abstract": "  Fuzzy dark matter (FDM) is an attractive dark matter candidate motivated by\nsmall scale problems in astrophysics and with a rich phenomenology on those\nscales. We scrutinize the FDM model, more specifically the mass of the FDM\nparticle, through a dynamical analysis for the Galactic ultra-faint dwarf (UFD)\ngalaxies. We use a sample of 18 UFDs to place the strongest constraints to date\non the mass of the FDM particle, updating on previous bounds using a subset of\nthe sample used here. We find that most of the sample UFDs prefer a FDM\nparticle mass heavier than $10^{-21}\\mathrm{eV}$. In particular, Segue 1\nprovides the strongest constraint, with\n$m_\\psi=1.1^{+8.3}_{-0.7}\\times10^{-19}\\mathrm{eV}$. The constraints found here\nare the first that are compatible with various other independent cosmological\nand astrophysical bounds found in the literature, in particular with the latest\nbounds using the Lyman-$\\alpha$ forest. We also find that the constraints\nobtained in this work are not compatible with the bounds from luminous dwarf\ngalaxies, as already pointed out in the previous work using UFDs. This could\nindicate that although a viable dark matter model, it might be challenging for\nthe FDM model to solve the small scale problems.\n"}
{"abstract": "  We demonstrate that a quantum annealer can be used to solve the NP-complete\nproblem of graph partitioning into subgraphs containing Hamiltonian cycles of\nconstrained length. We present a method to find a partition of a given directed\ngraph into Hamiltonian subgraphs with three or more vertices, called vertex\n3-cycle cover. We formulate the problem as a quadratic unconstrained binary\noptimisation and run it on a D-Wave Advantage quantum annealer. We test our\nmethod on synthetic graphs constructed by adding a number of random edges to a\nset of disjoint cycles. We show that the probability of solution is independent\nof the cycle length, and a solution is found for graphs up to 4000 vertices and\n5200 edges, close to the number of physical working qubits available on the\nquantum annealer.\n"}
{"abstract": "  We study the link between lobbying and industrial concentration. Using data\nfor the past 20 years in the US, we show how lobbying increases when an\nindustry becomes more concentrated, using mergers as shocks to concentration.\nThis holds true both for expenditures on federal lobbying as well as\nexpenditures on campaign contributions. Results are in line with the\npredictions of a model where lobbying is akin to a public good for incumbents,\nand thus typically underprovided, while a merger solves the coordination\nproblem.\n"}
{"abstract": "  Most data is automatically collected and only ever \"seen\" by algorithms. Yet,\ndata compressors preserve perceptual fidelity rather than just the information\nneeded by algorithms performing downstream tasks. In this paper, we\ncharacterize the bit-rate required to ensure high performance on all predictive\ntasks that are invariant under a set of transformations, such as data\naugmentations. Based on our theory, we design unsupervised objectives for\ntraining neural compressors. Using these objectives, we train a generic image\ncompressor that achieves substantial rate savings (more than $1000\\times$ on\nImageNet) compared to JPEG on 8 datasets, without decreasing downstream\nclassification performance.\n"}
{"abstract": "  Approximate Dynamic Programming (ADP) is a methodology to solve multi-stage\nstochastic optimization problems in multi-dimensional discrete or continuous\nspaces. ADP approximates the optimal value function by adaptively sampling both\naction and state space. It provides a tractable approach to very large\nproblems, but can suffer from the exploration-exploitation dilemma. We propose\na novel approach for selecting actions using importance sampling weighted by\nthe value function approximation in continuous decision spaces to address this\ndilemma. An advantage of this approach is it balances exploration and\nexploitation without any tuning parameters when sampling actions compared to\nother exploration approaches such as Epsilon Greedy, instead relying only on\nthe approximate value function. We compare the proposed algorithm with other\nexploration strategies in continuous action space in the context of a\nmulti-stage generation expansion planning problem under uncertainty.\n"}
{"abstract": "  Network embedding has been widely used in social recommendation and network\nanalysis, such as recommendation systems and anomaly detection with graphs.\nHowever, most of previous approaches cannot handle large graphs efficiently,\ndue to that (i) computation on graphs is often costly and (ii) the size of\ngraph or the intermediate results of vectors could be prohibitively large,\nrendering it difficult to be processed on a single machine. In this paper, we\npropose an efficient and effective distributed algorithm for network embedding\non large graphs using Apache Spark, which recursively partitions a graph into\nseveral small-sized subgraphs to capture the internal and external structural\ninformation of nodes, and then computes the network embedding for each subgraph\nin parallel. Finally, by aggregating the outputs on all subgraphs, we obtain\nthe embeddings of nodes in a linear cost. After that, we demonstrate in various\nexperiments that our proposed approach is able to handle graphs with billions\nof edges within a few hours and is at least 4 times faster than the\nstate-of-the-art approaches. Besides, it achieves up to $4.25\\%$ and $4.27\\%$\nimprovements on link prediction and node classification tasks respectively. In\nthe end, we deploy the proposed algorithms in two online games of Tencent with\nthe applications of friend recommendation and item recommendation, which\nimprove the competitors by up to $91.11\\%$ in running time and up to $12.80\\%$\nin the corresponding evaluation metrics.\n"}
{"abstract": "  Practical learning-based autonomous driving models must be capable of\ngeneralizing learned behaviors from simulated to real domains, and from\ntraining data to unseen domains with unusual image properties. In this paper,\nwe investigate transfer learning methods that achieve robustness to domain\nshifts by taking advantage of the invariance of spatio-temporal features across\ndomains. In this paper, we propose a transfer learning method to improve\ngeneralization across domains via transfer of spatio-temporal features and\nsalient data augmentation. Our model uses a CNN-LSTM network with Inception\nmodules for image feature extraction. Our method runs in two phases: Phase 1\ninvolves training on source domain data, while Phase 2 performs training on\ntarget domain data that has been supplemented by feature maps generated using\nthe Phase 1 model. Our model significantly improves performance in unseen test\ncases for both simulation-to-simulation transfer as well as simulation-to-real\ntransfer by up to +37.3\\% in test accuracy and up to +40.8\\% in steering angle\nprediction, compared to other SOTA methods across multiple datasets.\n"}
{"abstract": "  In this article we study a coarse version of the $K$-theoretic Farrell--Jones\nconjecture we call coarse or bounded isomorphism conjecture. Using controlled\ncategory theory we are able to translate this conjecture for asymptotically\nfaithful covers into a more familiar form. This allows us to prove the\nconjecture for box spaces of residually finite groups whose Farrell--Jones\nassembly map with coefficients is an isomorphism.\n"}
{"abstract": "  This article presents stability and convergence analyses of subgrid\nmultiscale stabilized finite element formulation of non-Newtonian power-law\nfluid flow model strongly coupled with variable coefficients\nAdvection-Diffusion-Reaction ($VADR$) equation. Considering the highly\nnon-linear viscosity coefficient as solute concentration dependent makes the\ncoupling two way. The stabilized formulation of the transient coupled system is\ndeveloped based upon time dependent subscales, which ensures inherent\nconsistency of the method. The proposed algebraic expressions of the\nstabilization parameters appropriately shape up the apriori and aposteriori\nerror estimates. Both the shear thinning and shear thickening properties,\nindicated by different power-law indices are properly highlighted in\ntheoretical derivations as well as in numerical validations. The numerical\nexperiments carried out for different combinations of small and large Reynolds\nnumbers and power-law indices establish far better performance of time\ndependent $ASGS$ method in approximating the solution of this coupled system\nfor all the cases over the other well known stabilized finite element methods.\n"}
{"abstract": "  We present a novel approach for time series classification where we represent\ntime series data as plot images and feed them to a simple CNN, outperforming\nseveral state-of-the-art methods. We propose a simple and highly replicable way\nof plotting the time series, and feed these images as input to a non-optimized\nshallow CNN, without any normalization or residual connections. These\nrepresentations are no more than default line plots using the time series data,\nwhere the only pre-processing applied is to reduce the number of white pixels\nin the image. We compare our method with different state-of-the-art methods\nspecialized in time series classification on two real-world non public\ndatasets, as well as 98 datasets of the UCR dataset collection. The results\nshow that our approach is very promising, achieving the best results on both\nreal-world datasets and matching / beating the best state-of-the-art methods in\nsix UCR datasets. We argue that, if a simple naive design like ours can obtain\nsuch good results, it is worth further exploring the capabilities of using\nimage representation of time series data, along with more powerful CNNs, for\nclassification and other related tasks.\n"}
{"abstract": "  Theoretical predictions of top-quark cross sections depend on the values of\nthe parameters of the quantum chromodynamics Lagrangian, such as the strong\ncoupling constant and the mass of the top quark, but also on parameters of the\nelectroweak sector of the standard model and parton distribution functions.\nComparisons between state-of-the-art calculations and recent measurements at\nthe ATLAS and CMS experiments at the CERN LHC allow for the precise\ndetermination of these parameters. In this proceeding, the most recent results\nby the two collaborations are summarized and discussed.\n"}
{"abstract": "  We adapt the argument of Dodson-Murphy to give a simple proof of scattering\nbelow the ground state for the intercritical inhomogeneous nonlinear\nSchr\\\"odinger equation. The decaying factor in the nonlinearity obviates the\nneed for a radial assumption.\n"}
{"abstract": "  We consider a class of inhomogeneous self-similar cosmological models in\nwhich the perfect fluid flow is tangential to the orbits of a three-parameter\nsimilarity group. We restrict the similarity group to possess both an Abelian\n$G_{2}$, and a single hypersurface orthogonal Killing vector field, and we\nrestrict the fluid flow to be orthogonal to the orbits of the Abelian $G_{2}$.\nThe temporal evolution of the models is forced to be power law, due to the\nsimilarity group, and the Einstein field equations reduce to a\nthree-dimensional autonomous system of ordinary differential equations which is\nqualitatively analysed in order to determine the spatial structure of the\nmodels. The existence of two classes of well-behaved models is demonstrated.\nThe first of these is asymptotically spatially homogeneous and matter\ndominated, and the second is vacuum dominated and either asymptotically\nspatially homogeneous or acceleration dominated, at large spatial distances.\n"}
{"abstract": "  In dimension two, we investigate a free energy and the ground state energy of\nthe Schr\\\"odinger-Poisson system coupled with a logarithmic nonlinearity in\nterms of underlying functional inequalities which take into account the scaling\ninvariances of the problem. Such a system can be considered as a nonlinear\nSchr\\\"odinger equation with a cubic but nonlocal Poisson nonlinearity, and a\nlocal logarithmic nonlinearity. Both cases of repulsive and attractive forces\nare considered. We also assume that there is an external potential with minimal\ngrowth at infinity, which turns out to have a logarithmic growth. Our estimates\nrely on new logarithmic interpolation inequalities which combine logarithmic\nHardy-Littlewood-Sobolev and logarithmic Sobolev inequalities. The\ntwo-dimensional model appears as a limit case of more classical problems in\nhigher dimensions.\n"}
{"abstract": "  Training deep neural models in the presence of corrupted supervision is\nchallenging as the corrupted data points may significantly impact the\ngeneralization performance. To alleviate this problem, we present an efficient\nrobust algorithm that achieves strong guarantees without any assumption on the\ntype of corruption and provides a unified framework for both classification and\nregression problems. Unlike many existing approaches that quantify the quality\nof the data points (e.g., based on their individual loss values), and filter\nthem accordingly, the proposed algorithm focuses on controlling the collective\nimpact of data points on the average gradient. Even when a corrupted data point\nfailed to be excluded by our algorithm, the data point will have a very limited\nimpact on the overall loss, as compared with state-of-the-art filtering methods\nbased on loss values. Extensive experiments on multiple benchmark datasets have\ndemonstrated the robustness of our algorithm under different types of\ncorruption.\n"}
{"abstract": "  We present a method for assessing the sensitivity of the true causal effect\nto unmeasured confounding. The method requires the analyst to set two intuitive\nparameters. Otherwise, the method is assumption-free. The method returns an\ninterval that contains the true causal effect, and whose bounds are sharp, i.e.\nattainable. We show experimentally that our bounds can be sharper than those\nobtained by the method of Ding and VanderWeele (2016a) which, moreover,\nrequires to set one more parameter than our method. Finally, we extend our\nmethod to bound the natural direct and indirect effects when there are measured\nmediators and unmeasured exposure-outcome confounding.\n"}
{"abstract": "  Suppose $G$ is a finite group acting on an Abelian variety $A$ such that the\ncoarse moduli $A/G$ is smooth. Using the recent classification result due to\nAuffarth, Lucchini, and Quezada, we construct a motivic semiorthogonal\ndecomposition for $\\mathcal{D}[A/G]$ provided $G = T\\rtimes H$ with $T$ a\nsubgroup of translations and $H$ is a subgroup of group automorphisms.\n"}
{"abstract": "  It is shown that the Cotton tensor can describe the effects of gravity beyond\ngeneral relativity. Any solution of the Einstein equations with or without the\ncosmological constant satisfies the field equations described by the Cotton\ntensor. It implies that the cosmological constant is an integration constant. A\nvacuum of a theory is represented by the vanishing of the Cotton tensor, rather\nthan the vanishing of the Ricci tensor. An exact Schwarzschild-like solution\nfor a static and spherically symmetric source is discovered. Although the field\nequations involve the third order of derivative, it is found that they reduce\nto the second-order differential equations due to a variational principle.\n"}
{"abstract": "  Nanostructures can be used for boosting the light outcoupling of color\ncenters in diamond; however, the fiber coupling performance of these\nnanostructures is rarely investigated. Here, we use a finite element method for\ncomputing the emission from color centers in inverted nanocones and the overlap\nof this emission with the propagation mode in a single-mode fiber. Using\ndifferent figures of merit, the inverted nanocone parameters are optimized to\nobtain maximal fiber coupling efficiency, free-space collection efficiency, or\nrate enhancement. The optimized inverted nanocone designs show promising\nresults with 66% fiber coupling or 83% free-space coupling efficiency at the\ntin-vacancy center zero-phonon line wavelength of 619 nm. Moreover, when\nevaluated for broadband performance, the optimized designs show 55% and 76% for\nfiber coupling and free-space efficiencies respectively, for collecting the\nfull tin-vacancy emission spectrum at room temperature. An analysis of\nfabrication insensitivity indicates that these nanostructures are robust\nagainst imperfections. For maximum emission rate into a fiber mode, a design\nwith a Purcell factor of 2.34 is identified. Finally, possible improvements\noffered by a hybrid inverted nanocone, formed by patterning into two different\nmaterials, are investigated, and increases the achievable fiber coupling\nefficiency to 71%.\n"}
{"abstract": "  Many popular models from the networks literature can be viewed through a\ncommon lens. We describe it here and call the class of models log-linear ERGMs.\nIt includes degree-based models, stochastic blockmodels, and combinations of\nthese. Given the interest in combined node and block effects in network\nformation mechanisms, we introduce a general directed relative of the\ndegree-corrected stochastic blockmodel: an exponential family model we call\n$p_1$-SBM. It is a generalization of several well-known variants of the\nblockmodel.\n  We study the problem of testing model fit for the log-linear ERGM class.\n  The model fitting approach we take, through the use of quick estimation\nalgorithms borrowed from the contingency table literature and effective\nsampling methods rooted in graph theory and algebraic statistics, results in an\nexact test whose $p$-value can be approximated efficiently in networks of\nmoderate sizes.\n  We showcase the performance of the method on two data sets from biology: the\nconnectome of \\emph{C. elegans} and the interactome of \\emph{Arabidopsis\nthaliana}. These two networks, a neuronal network and a protein-protein\ninteraction network, have been popular examples in the network science\nliterature, but a model-based approach to studying them has been missing thus\nfar.\n"}
{"abstract": "  Artificial neural networks (ANNs) have gained significant popularity in the\nlast decade for solving narrow AI problems in domains such as healthcare,\ntransportation, and defense. As ANNs become more ubiquitous, it is imperative\nto understand their associated safety, security, and privacy vulnerabilities.\nRecently, it has been shown that ANNs are susceptible to a number of\nadversarial evasion attacks--inputs that cause the ANN to make high-confidence\nmisclassifications despite being almost indistinguishable from the data used to\ntrain and test the network. This work explores to what degree finding these\nexamples maybe aided by using side-channel information, specifically switching\npower consumption, of hardware implementations of ANNs. A black-box threat\nscenario is assumed, where an attacker has access to the ANN hardware's input,\noutputs, and topology, but the trained model parameters are unknown. Then, a\nsurrogate model is trained to have similar functional (i.e. input-output\nmapping) and switching power characteristics as the oracle (black-box) model.\nOur results indicate that the inclusion of power consumption data increases the\nfidelity of the model extraction by up to 30 percent based on a mean square\nerror comparison of the oracle and surrogate weights. However, transferability\nof adversarial examples from the surrogate to the oracle model was not\nsignificantly affected.\n"}
{"abstract": "  This manuscript explores a new class of non-autonomous second-order\nstochastic inclusions of Clarke's subdifferential form with non-instantaneous\nimpulses (NIIs), unbounded delay and the Rosenblatt process in Hilbert spaces.\nThe existence of a solution is deduced by employing a fixed point strategy for\na set-valued map together with the evolution operator and stochastic analysis\napproach. An example is analyzed for theoretical developments.\n"}
{"abstract": "  Combined bounds on the Majorana neutrino mass for light and heavy neutrino\nexchange mechanisms are derived from current neutrinoless double beta decay\n(0{\\nu}\\b{eta}\\b{eta}) search results for a variety of nuclear matrix element\n(NME) models. The approach requires self-consistency of a given model to\npredict NMEs across different isotopes. The derived bounds are notably stronger\nthan those from any single experiment and show less model-to-model variation,\nhighlighting the advantages of using multiple isotopes in such searches.\nProjections indicate that the combination of near-term experiments should be\nable to probe well into the inverted mass hierarchy region. A method to\nvisually represent 0{\\nu}\\b{eta}\\b{eta} experimental results is also suggested\nto more transparently compare across different isotopes and explicitly track\nmodel dependencies.\n"}
{"abstract": "  We show that all longest cycles intersect in 2-connected partial 3-trees.\n"}
{"abstract": "  Structural materials are broadly used in applications such as nuclear\nvessels, high-temperature processes, and civil construction. Usually, during\ntheir placing and lifespan, they may present free or chemically bonded liquid\nphases in their structure, demanding careful attention when exposed to high\nheating rates. Their behavior in such conditions is a challenging problem as it\ncomprises numerous highly nonlinear properties (not easily measured via\nexperimental tests), strongly coupled equations and unreliable experimental\nbenchmarks. Nonetheless, such simulations are of great interest. This work aims\nto provide a numerical study, checking whether its solution indeed converges\nand yields reliable results. Additionally, as the model needs several input\nparameters, this work conducts a sensitivity analysis and also assesses its\napplicability to more complex scenarios, as such issues remain open in the\nliterature. In order to do that, a simple model that can be easily adapted for\nmixed formulations and complex geometries was proposed. It was found out that\nwhen considering unidimensional models the choices regarding the interpolation\nof the sorption isotherms are not essential to the numerical stability of the\nsystem. Besides that, the permeability and thermal conductivity of the material\nare the most important parameters that affect the simulation results of\npressure, temperature and evaporable water content profiles. Finally, the 2D\nmesoscale simulation of concrete with polymeric fibers (based on the mixed\nformulation of the problem) yielded results that agreed with experimental\nobservations. Thus, the model proposed herein can provide a solid base for\nfuture works and also important insights towards simpler methodologies.\n"}
{"abstract": "  Data-driven, model-free analytics are natural choices for discovery and\nforecasting of complex, nonlinear systems. Methods that operate in the system\nstate-space require either an explicit multidimensional state-space, or, one\napproximated from available observations. Since observational data are\nfrequently sampled with noise, it is possible that noise can corrupt the\nstate-space representation degrading analytical performance. Here, we evaluate\nthe synthesis of empirical mode decomposition with empirical dynamic modeling,\nwhich we term empirical mode modeling, to increase the information content of\nstate-space representations in the presence of noise. Evaluation of a\nmathematical, and, an ecologically important geophysical application across\nthree different state-space representations suggests that empirical mode\nmodeling may be a useful technique for data-driven, model-free, state-space\nanalysis in the presence of noise.\n"}
{"abstract": "  During the COVID-19 epidemic, many health professionals started using mass\ncommunication on social media to relay critical information and persuade\nindividuals to adopt preventative health behaviors. Our group of clinicians and\nnurses developed and recorded short video messages to encourage viewers to stay\nhome for the Thanksgiving and Christmas Holidays. We then conducted a two-stage\nclustered randomized controlled trial in 820 counties (covering 13 States) in\nthe United States of a large-scale Facebook ad campaign disseminating these\nmessages. In the first level of randomization, we randomly divided the counties\ninto two groups: high intensity and low intensity. In the second level, we\nrandomly assigned zip codes to either treatment or control such that 75% of zip\ncodes in high intensity counties received the treatment, while 25% of zip codes\nin low intensity counties received the treatment. In each treated zip code, we\nsent the ad to as many Facebook subscribers as possible (11,954,109 users\nreceived at least one ad at Thanksgiving and 23,302,290 users received at least\none ad at Christmas). The first primary outcome was aggregate holiday travel,\nmeasured using mobile phone location data, available at the county level: we\nfind that average distance travelled in high-intensity counties decreased by\n-0.993 percentage points (95% CI -1.616, -0.371, p-value 0.002) the three days\nbefore each holiday. The second primary outcome was COVID-19 infection at the\nzip-code level: COVID-19 infections recorded in the two-week period starting\nfive days post-holiday declined by 3.5 percent (adjusted 95% CI [-6.2 percent,\n-0.7 percent], p-value 0.013) in intervention zip codes compared to control zip\ncodes.\n"}
{"abstract": "  We prove some $C^\\infty$ and Gevrey well-posedness results for hyperbolic\nequations whose coefficients lose regularity at one point.\n"}
{"abstract": "  Software traceability establishes and leverages associations between diverse\ndevelopment artifacts. Researchers have proposed the use of deep learning trace\nmodels to link natural language artifacts, such as requirements and issue\ndescriptions, to source code; however, their effectiveness has been restricted\nby availability of labeled data and efficiency at runtime. In this study, we\npropose a novel framework called Trace BERT (T-BERT) to generate trace links\nbetween source code and natural language artifacts. To address data sparsity,\nwe leverage a three-step training strategy to enable trace models to transfer\nknowledge from a closely related Software Engineering challenge, which has a\nrich dataset, to produce trace links with much higher accuracy than has\npreviously been achieved. We then apply the T-BERT framework to recover links\nbetween issues and commits in Open Source Projects. We comparatively evaluated\naccuracy and efficiency of three BERT architectures. Results show that a\nSingle-BERT architecture generated the most accurate links, while a\nSiamese-BERT architecture produced comparable results with significantly less\nexecution time. Furthermore, by learning and transferring knowledge, all three\nmodels in the framework outperform classical IR trace models. On the three\nevaluated real-word OSS projects, the best T-BERT stably outperformed the VSM\nmodel with average improvements of 60.31% measured using Mean Average Precision\n(MAP). RNN severely underperformed on these projects due to insufficient\ntraining data, while T-BERT overcame this problem by using pretrained language\nmodels and transfer learning.\n"}
{"abstract": "  Understanding the high-tech industrial agglomeration from a spatial-spillover\nperspective is essential for cities to gain economic and technological\ncompetitive advantages. Along with rapid urbanization and the development of\nfast transportation networks, socioeconomic interactions between cities have\nbeen ever-increasing, traditional spatial metrics are not enough to describe\nactual inter-city connections. High-skilled labor flow between cities strongly\ninfluences the high-tech industrial agglomeration, yet receives less attention.\nBy exploiting unique large-scale datasets and tools from complex network and\ndata mining, we construct an inter-city high-skilled labor flow network, which\nwas integrated into spatial econometric models. Our regression results indicate\nthat spatial-spillover effects exist in the development of high-tech industries\nin the Yangtze River Delta Urban Agglomeration region. Moreover, the\nspatial-spillover effects are stronger among cities with a higher volume of\nhigh-skilled labor flows than among cities with just stronger geographic\nconnections. Additionally, we investigate the channels for the spillover\neffects and discover that inadequate local government expenses on science and\ntechnology likely hamper the high-tech industrial agglomeration, so does the\ninadequate local educational provision. The increasing foreign direct\ninvestments in one city likely encourages the high-tech industrial\nagglomeration in other cities because of the policy inertia toward traditional\nindustries.\n"}
{"abstract": "  We prove how the universal enveloping algebra constructions for Lie-Rinehart\nalgebras and anchored Lie algebras are naturally left adjoint functors. This\nprovides a conceptual motivation for the universal properties these\nconstructions satisfy. As a supplement, the categorical approach offers new\ninsights into the definitions of Lie-Rinehart algebra morphisms, of modules\nover Lie-Rinehart algebras and of the infinitesimal gauge algebra of a module.\n"}
{"abstract": "  Analysis-suitable T-splines (AST-splines) are a promising candidate to\nachieve a seamless integration between the design and the analysis of\nthin-walled structures in industrial settings. In this work, we generalize\nAST-splines to allow multiple extraordinary points within the same face. This\ngeneralization drastically increases the flexibility to build geometries using\nAST-splines; e.g., much coarser meshes can be generated to represent a certain\ngeometry. The AST-spline spaces detailed in this work have $C^1$ inter-element\ncontinuity near extraordinary points and $C^2$ inter-element continuity\nelsewhere. We mathematically show that AST-splines with multiple extraordinary\npoints per face are linearly independent and their polynomial basis functions\nform a non-negative partition of unity. We numerically show that AST-splines\nwith multiple extraordinary points per face lead to optimal convergence rates\nfor second- and fourth-order linear elliptic problems. To illustrate a possible\nisogeometric framework that is already available, we design the B-pillar and\nthe side outer panel of a car using T-splines with the commercial software\nAutodesk Fusion360, import the control nets into our in-house code to build\nAST-splines, and import the B\\'ezier extraction information into the commercial\nsoftware LS-DYNA to solve eigenvalue problems. The results are compared with\nconventional finite elements. Good agreement is found, but conventional finite\nelements require significantly more degrees of freedom to reach a converged\nsolution than AST-splines.\n"}
{"abstract": "  This paper develops asymptotic normality results for individual coordinates\nof robust M-estimators with convex penalty in high-dimensions, where the\ndimension $p$ is at most of the same order as the sample size $n$, i.e,\n$p/n\\le\\gamma$ for some fixed constant $\\gamma>0$. The asymptotic normality\nrequires a bias correction and holds for most coordinates of the M-estimator\nfor a large class of loss functions including the Huber loss and its smoothed\nversions regularized with a strongly convex penalty.\n  The asymptotic variance that characterizes the width of the resulting\nconfidence intervals is estimated with data-driven quantities. This estimate of\nthe variance adapts automatically to low ($p/n\\to0)$ or high ($p/n \\le \\gamma$)\ndimensions and does not involve the proximal operators seen in previous works\non asymptotic normality of M-estimators. For the Huber loss, the estimated\nvariance has a simple expression involving an effective degrees-of-freedom as\nwell as an effective sample size. The case of the Huber loss with Elastic-Net\npenalty is studied in details and a simulation study confirms the theoretical\nfindings. The asymptotic normality results follow from Stein formulae for\nhigh-dimensional random vectors on the sphere developed in the paper which are\nof independent interest.\n"}
{"abstract": "  The presence of disorder and inhomogeneities in quantum networks has often\nbeen unexpectedly beneficial for both quantum and classical resources. Here, we\nexperimentally realize a controllable inhomogenous Quantum Walk dynamics, which\ncan be exploited to investigate the effect of coherent disorder on the quantum\ncorrelations between two indistinguishable photons. Through the imposition of\nsuitable disorder configurations, we observe two photon states which exhibit an\nenhancement in the quantum correlations between two modes of the network,\ncompared to the case of an ordered Quantum Walk. Different configurations of\ndisorder can steer the system towards different realizations of such an\nenhancement, thus allowing spatial and temporal manipulation of quantum\ncorrelations.\n"}
{"abstract": "  Aerial pixel-wise scene perception of the surrounding environment is an\nimportant task for UAVs (Unmanned Aerial Vehicles). Previous research works\nmainly adopt conventional pinhole cameras or fisheye cameras as the imaging\ndevice. However, these imaging systems cannot achieve large Field of View\n(FoV), small size, and lightweight at the same time. To this end, we design a\nUAV system with a Panoramic Annular Lens (PAL), which has the characteristics\nof small size, low weight, and a 360-degree annular FoV. A lightweight\npanoramic annular semantic segmentation neural network model is designed to\nachieve high-accuracy and real-time scene parsing. In addition, we present the\nfirst drone-perspective panoramic scene segmentation dataset Aerial-PASS, with\nannotated labels of track, field, and others. A comprehensive variety of\nexperiments shows that the designed system performs satisfactorily in aerial\npanoramic scene parsing. In particular, our proposed model strikes an excellent\ntrade-off between segmentation performance and inference speed suitable,\nvalidated on both public street-scene and our established aerial-scene\ndatasets.\n"}
{"abstract": "  We introduce and study the dynamics of an \\emph{immortal} critical branching\nprocess. In the classic, critical branching process, particles give birth to a\nsingle offspring or die at the same rates. Even though the average population\nis constant in time, the ultimate fate of the population is extinction. We\naugment this branching process with immortality by positing that either: (a) a\nsingle particle cannot die, or (b) there exists an immortal stem cell that\ngives birth to ordinary cells that can subsequently undergo critical branching.\nWe discuss the new dynamical aspects of this immortal branching process.\n"}
{"abstract": "  Gaussian processes (GPs) are non-parametric Bayesian models that are widely\nused for diverse prediction tasks. Previous work in adding strong privacy\nprotection to GPs via differential privacy (DP) has been limited to protecting\nonly the privacy of the prediction targets (model outputs) but not inputs. We\nbreak this limitation by introducing GPs with DP protection for both model\ninputs and outputs. We achieve this by using sparse GP methodology and\npublishing a private variational approximation on known inducing points. The\napproximation covariance is adjusted to approximately account for the added\nuncertainty from DP noise. The approximation can be used to compute arbitrary\npredictions using standard sparse GP techniques. We propose a method for\nhyperparameter learning using a private selection protocol applied to\nvalidation set log-likelihood. Our experiments demonstrate that given\nsufficient amount of data, the method can produce accurate models under strong\nprivacy protection.\n"}
{"abstract": "  This paper presents a methodology for using LLVM-based tools to tune the\nDCA++ (dynamical clusterapproximation) application that targets the new ARM\nA64FX processor. The goal is to describethe changes required for the new\narchitecture and generate efficient single instruction/multiple data(SIMD)\ninstructions that target the new Scalable Vector Extension instruction set.\nDuring manualtuning, the authors used the LLVM tools to improve code\nparallelization by using OpenMP SIMD,refactored the code and applied\ntransformation that enabled SIMD optimizations, and ensured thatthe correct\nlibraries were used to achieve optimal performance. By applying these code\nchanges, codespeed was increased by 1.98X and 78 GFlops were achieved on the\nA64FX processor. The authorsaim to automatize parts of the efforts in the\nOpenMP Advisor tool, which is built on top of existingand newly introduced LLVM\ntooling.\n"}
{"abstract": "  Experiments with diblock co-polymer melts display undulated bilayers that\nemanate from defects such as triple junctions and endcaps,\n\\cite{batesjain_2004}. Undulated bilayers are characterized by oscillatory\nperturbations of the bilayer width, which decay on a spatial length scale that\nis long compared to the bilayer width. We mimic defects within the\nfunctionalized Cahn-Hillard free energy by introducing spatially localized\ninhomogeneities within its parameters. For length parameter $\\varepsilon\\ll1$,\nwe show that this induces undulated bilayer solutions whose width perturbations\ndecay on an $O\\!\\left(\\varepsilon^{-1/2}\\right)$ inner length scale that is\nlong in comparison to the $O(1)$ scale that characterizes the bilayer width.\n"}
{"abstract": "  In this paper we investigate whether the features of the non-equilibrium\ncascade, which have been identified in recent studies using high-fidelity\ntools, can be captured in the case of the classical dissipation scaling by\nturbulence closures based on the statistical description of freely decaying\nisotropic turbulence. Numerical results obtained using the EDQNM model over a\nvery large range of Reynolds numbers (from $Re_{\\lambda}=50$ up to\n$Re_{\\lambda}=10^6$) are analyzed to perform an extensive investigation of the\nscaling region identified as inertial range in Kolmogorov's theory. It is\nobserved that EDQNM results are in agreement with the results of Lundgren's\nmatched asymptotic expansion approach to the Karman-Howarth equation. Both\npredict that the Kolmogorov inertial range equilibrium is never obtained\nirrespective of Reynolds number. Equilibrium is reached in the vicinity of the\nTaylor length $\\lambda$ (which depends on viscosity) as Reynolds number tends\nto infinity and there is a gradual departure from equilibrium as the length\nscale moves away from $\\lambda$, in particular towards scales larger than\n$\\lambda$ all the way to the integral length-scale.\n"}
{"abstract": "  Cache coherence protocols such as MESI that use writer-initiated invalidation\nhave high complexity and sometimes have poor performance and energy usage,\nespecially under false sharing. Such protocols require numerous transient\nstates, a shared directory, and support for core-to-core communication, while\nalso suffering under false sharing. An alternative to MESI's writer-initiated\ninvalidation is self-invalidation, which achieves lower complexity than MESI\nbut adds high performance costs or relies on programmer annotations or specific\ndata access patterns.\n  This paper presents Neat, a low-complexity, efficient cache coherence\nprotocol. Neat uses self-invalidation, thus avoiding MESI's transient states,\ndirectory, and core-to-core communication requirements. Neat uses novel\nmechanisms that effectively avoid many unnecessary self-invalidations. An\nevaluation shows that Neat is simple and has lower verification complexity than\nthe MESI protocol. Neat not only outperforms state-of-the-art self-invalidation\nprotocols, but its performance and energy consumption are comparable to MESI's,\nand it outperforms MESI under false sharing.\n"}
{"abstract": "  This White Paper summarizes the authors' discussion regarding objectionable\ncontent for the University of Houston (UH) Research Team to outline a strategy\nfor building an extensive repository of online videos to support research into\nautomated multimodal approaches to detect objectionable content. The workshop\nfocused on defining what harmful content is, to whom it is harmful, and why it\nis harmful.\n"}
{"abstract": "  I present a numerical study of the crossover between the low temperature\nchirally broken phase and the high temperature chirally restored phase in\n$SU(N_c)$ gauge theory with $N_c=3-5$ colors and $N_f=2$ degenerate fermion\nflavors. Fermion masses span a range of intermediate values (represented by the\nsquared ratio of pseudoscalar to vector meson masses $(m_{PS}/m_V)^2\\sim 0.25$\nto 0.63). Observables include the temperature dependent chiral condensate and\nscreening masses. At each fermion mass these quantities show nearly identical\ntemperature dependence across $N_c$.\n"}
{"abstract": "  Recently, artificial neural networks have been gaining momentum in the field\nof gravitational wave astronomy, for example in surrogate modelling of\ncomputationally expensive waveform models for binary black hole inspiral and\nmerger. Surrogate modelling yields fast and accurate approximations of\ngravitational waves and neural networks have been used in the final step of\ninterpolating the coefficients of the surrogate model for arbitrary waveforms\noutside the training sample. We investigate the existence of underlying\nstructures in the empirical interpolation coefficients using autoencoders. We\ndemonstrate that when the coefficient space is compressed to only two\ndimensions, a spiral structure appears, wherein the spiral angle is linearly\nrelated to the mass ratio. Based on this finding, we design a spiral module\nwith learnable parameters, that is used as the first layer in a neural network,\nwhich learns to map the input space to the coefficients. The spiral module is\nevaluated on multiple neural network architectures and consistently achieves\nbetter speed-accuracy trade-off than baseline models. A thorough experimental\nstudy is conducted and the final result is a surrogate model which can evaluate\nmillions of input parameters in a single forward pass in under 1ms on a desktop\nGPU, while the mismatch between the corresponding generated waveforms and the\nground-truth waveforms is better than the compared baseline methods. We\nanticipate the existence of analogous underlying structures and corresponding\ncomputational gains also in the case of spinning black hole binaries.\n"}
{"abstract": "  The question of finding sets of monomials which are removable from a generic\nhomogeneous polynomial through a linear change of coordinates was raised by E.\nK. Wakeford in [32]. This linear algebra question motivated C. K. Fan and J.\nLosonczy to define the concept of acyclic matchings in $\\mathbb{Z}^n$ in [16]\nwhich was later generalized to abelian groups by the latter author [26].\nConcepts of matchings and acyclic matchings have linear analogues developed in\nthe context of vector subspaces in a field extension [13, 1]. We discuss the\nacyclic matching and weak acyclic matching properties and we provide results on\nthe existence of acyclic matchings infinite cyclic groups. As for field\nextensions, we classify field extensions with the linear acyclic matching\nproperty generalizing a theorem from [1]. The analogy between matchings in\nabelian groups and in field extensions is highlighted throughout the paper and\nnumerous open questions are presented for further inquiry.\n"}
{"abstract": "  The Weyl double copy is a procedure for relating exact solutions in biadjoint\nscalar, gauge and gravity theories, and relates fields in spacetime directly.\nWhere this procedure comes from, and how general it is, have until recently\nremained mysterious. In this paper, we show how the current form and scope of\nthe Weyl double copy can be derived from a certain procedure in twistor space.\nThe new formalism shows that the Weyl double copy is more general than\npreviously thought, applying in particular to gravity solutions with arbitrary\nPetrov types. We comment on how to obtain anti-self-dual as well as self-dual\nfields, and clarify some conceptual issues in the twistor approach.\n"}
{"abstract": "  We evaluate what will be the effectiveness of the ExoMars Raman Laser\nSpectrometer (RLS) to determine the degree of serpentinization of olivine-rich\nunits on Mars. We selected terrestrial analogues of martian ultramafic rocks\nfrom the Leka Ophiolite Complex (LOC) and analyzed them with both laboratory\nand flight-like analytical instruments. We first studied the mineralogical\ncomposition of the samples (mostly olivine and serpentine) with\nstate-of-the-art diffractometric and spectroscopic laboratory systems. We\ncompared these results with those obtained using our RLS ExoMars Simulator. Our\nwork shows that the RLS ExoMars Simulator successfully identified all major\nphases. Moreover, when emulating the automatic operating mode of the flight\ninstrument, the RLS ExoMars simulator also detected several minor compounds,\nsome of which were not observed by NIR and XRD. Thereafter, we produced RLS\ndedicated calibration curves (R2 between 0.9993 and 0.9995 with an uncertainty\nbetween 3.0% and 5.2% with a confidence interval of 95%) to estimate the\nrelative content of olivine and serpentine in the samples. Our results show\nthat RLS can be very effective to identify serpentine, a scientific target of\nprimary importance for the potential detection of biosignatures on Mars the\nmain objective of the ExoMars rover mission.\n"}
{"abstract": "  Post-click conversion, as a strong signal indicating the user preference, is\nsalutary for building recommender systems. However, accurately estimating the\npost-click conversion rate (CVR) is challenging due to the selection bias,\ni.e., the observed clicked events usually happen on users' preferred items.\nCurrently, most existing methods utilize counterfactual learning to debias\nrecommender systems. Among them, the doubly robust (DR) estimator has achieved\ncompetitive performance by combining the error imputation based (EIB) estimator\nand the inverse propensity score (IPS) estimator in a doubly robust way.\nHowever, inaccurate error imputation may result in its higher variance than the\nIPS estimator. Worse still, existing methods typically use simple\nmodel-agnostic methods to estimate the imputation error, which are not\nsufficient to approximate the dynamically changing model-correlated target\n(i.e., the gradient direction of the prediction model). To solve these\nproblems, we first derive the bias and variance of the DR estimator. Based on\nit, a more robust doubly robust (MRDR) estimator has been proposed to further\nreduce its variance while retaining its double robustness. Moreover, we propose\na novel double learning approach for the MRDR estimator, which can convert the\nerror imputation into the general CVR estimation. Besides, we empirically\nverify that the proposed learning scheme can further eliminate the high\nvariance problem of the imputation learning. To evaluate its effectiveness,\nextensive experiments are conducted on a semi-synthetic dataset and two\nreal-world datasets. The results demonstrate the superiority of the proposed\napproach over the state-of-the-art methods. The code is available at\nhttps://github.com/guosyjlu/MRDR-DL.\n"}
{"abstract": "  Reconfigurable intelligent surfaces (RISs) are set to be a revolutionary\ntechnology in the 6th generation of wireless systems. In this work, we study\nthe application of RIS in a multi-user passive localization scenario, where we\nhave one transmitter (Tx) and multiple asynchronous receivers (Rxs) with known\nlocations. We aim to estimate the locations of multiple users equipped with\nRISs. The RISs only reflect the signal from the Tx to the Rxs and are not used\nas active transceivers themselves. Each Rx receives the signal from the Tx (LOS\npath) and the reflected signal from the RISs (NLOS path). We show that users'\n3D position can be estimated with submeter accuracy in a large area around the\ntransmitter, using the LOS and NLOS time-of-arrival measurements at the Rxs. We\ndo so, by developing the signal model, deriving the Cramer-Rao bounds, and\ndevising an estimator that attains these bounds. Furthermore, by\northogonalizing the RIS phase profiles across different users, we circumvent\ninter-path interference.\n"}
{"abstract": "  In this paper we prove the existence of isoperimetric regions of any volume\nin Riemannian manifolds with Ricci bounded below and with a mild assumption at\ninfinity, that is Gromov-Hausdorff asymptoticity to simply connected models of\nconstant sectional curvature. The previous result is a consequence of a general\nstructure theorem for perimeter-minimizing sequences of sets of fixed volume on\nnoncollapsed Riemannian manifolds with a lower bound on the Ricci curvature. We\nshow that, without assuming any further hypotheses on the asymptotic geometry,\nall the mass and the perimeter lost at infinity, if any, are recovered by at\nmost countably many isoperimetric regions sitting in some Gromov-Hausdorff\nlimits at infinity. The Gromov-Hausdorff asymptotic analysis conducted allows\nus to provide, in low dimensions, a result of nonexistence of isoperimetric\nregions in Cartan-Hadamard manifolds that are Gromov-Hausdorff asymptotic to\nthe Euclidean space. While studying the isoperimetric problem in the smooth\nsetting, the nonsmooth geometry naturally emerges, and thus our treatment\ncombines techniques from both the theories.\n"}
{"abstract": "  Practical quantum networking architectures are crucial for scaling the\nconnection of quantum resources. Yet quantum network testbeds have thus far\nunderutilized the full capabilities of modern lightwave communications, such as\nflexible-grid bandwidth allocation. In this work, we implement flex-grid\nentanglement distribution in a deployed network for the first time, connecting\nnodes in three distinct campus buildings time-synchronized via the Global\nPositioning System (GPS). We quantify the quality of the distributed\npolarization entanglement via log-negativity, which offers a generic metric of\nlink performance in entangled bits per second. After demonstrating successful\nentanglement distribution for two allocations of our eight dynamically\nreconfigurable channels, we demonstrate remote state preparation -- the first\nrealization on deployed fiber -- showcasing one possible quantum protocol\nenabled by the distributed entanglement network. Our results realize an\nadvanced paradigm for managing entanglement resources in quantum networks of\never-increasing complexity and service demands.\n"}
{"abstract": "  This paper addresses black-box optimization over multiple information sources\nwhose both fidelity and query cost change over the search space, that is they\nare location dependent. The approach uses: (i) an Augmented Gaussian Process,\nrecently proposed in multi-information source optimization as a single model of\nthe objective function over search space and sources, and (ii) a Gaussian\nProcess to model the location-dependent cost of each source. The former is used\ninto a Confidence Bound based acquisition function to select the next source\nand location to query, while the latter is used to penalize the value of the\nacquisition depending on the expected query cost for any source-location pair.\nThe proposed approach is evaluated on a set of Hyperparameters Optimization\ntasks, consisting of two Machine Learning classifiers and three datasets of\ndifferent sizes.\n"}
{"abstract": "  Adaptive networks have the capability to pursue solutions of global\nstochastic optimization problems by relying only on local interactions within\nneighborhoods. The diffusion of information through repeated interactions\nallows for globally optimal behavior, without the need for central\ncoordination. Most existing strategies are developed for cooperative learning\nsettings, where the objective of the network is common to all agents. We\nconsider in this work a team setting, where a subset of the agents form a team\nwith a common goal while competing with the remainder of the network. We\ndevelop an algorithm for decentralized competition among teams of adaptive\nagents, analyze its dynamics and present an application in the decentralized\ntraining of generative adversarial neural networks.\n"}
{"abstract": "  We study the condition for the dilaton stabilization in Type IIB flux\ncompactifications consistent with the KKLT scenario. Since the\nGukov-Vafa-Witten superpotential depends linearly on the dilaton, the dilaton\nmass squared is given by a sum of the gravitino mass squared and additional\nterms determined by the complex structure moduli stabilization. If the dilaton\nmass is not much enhanced from the gravitino mass, the mass mixing with the\nK\\\"ahler modulus in the presence of the non-perturbative effect generates the\nsaddle point at the supersymmetric field values, hence the potential becomes\nunstable. When the complex structure moduli other than the conifold modulus are\nneglected, the saddle point problem arises over the controllable parameter\nspace. We also point out that the dilaton stabilization condition is equivalent\nto the condition on the NS 3-form fluxes, $|H_{(1,2)}| > | H_{(0,3)}|$.\n"}
{"abstract": "  This paper studies the differential lattice, defined to be a lattice $L$\nequipped with a map $d:L\\to L$ that satisfies a lattice analog of the Leibniz\nrule for a derivation. Isomorphic differential lattices are studied and\nclassifications of differential lattices are obtained for some basic lattices.\nSeveral families of derivations on a lattice are explicitly constructed, giving\nrealizations of the lattice as lattices of derivations. Derivations on a finite\ndistributive lattice are shown to have a natural structure of lattice.\nMoreover, derivations on a complete infinitely distributive lattice form a\ncomplete lattice. For a general lattice, it is conjectured that its poset of\nderivations is a lattice that uniquely determines the given lattice.\n"}
{"abstract": "  Recent advancements in computer vision promise to automate medical image\nanalysis. Rheumatoid arthritis is an autoimmune disease that would profit from\ncomputer-based diagnosis, as there are no direct markers known, and doctors\nhave to rely on manual inspection of X-ray images. In this work, we present a\nmulti-task deep learning model that simultaneously learns to localize joints on\nX-ray images and diagnose two kinds of joint damage: narrowing and erosion.\nAdditionally, we propose a modification of label smoothing, which combines\nclassification and regression cues into a single loss and achieves 5% relative\nerror reduction compared to standard loss functions. Our final model obtained\n4th place in joint space narrowing and 5th place in joint erosion in the global\nRA2 DREAM challenge.\n"}
{"abstract": "  We propose a new Arithmetic Distribution Neural Network (ADNN) for learning\nthe distributions of temporal pixels during background subtraction. In our\nADNN, the arithmetic distribution operations are utilized to propose the\narithmetic distribution layers, including the product distribution layer and\nthe sum distribution layer. Furthermore, in order to improve the accuracy of\nthe proposed approach, an improved Bayesian refinement model based on\nneighboring information, with a GPU implementation, is introduced. In the\nforward pass and backpropagation of the proposed arithmetic distribution\nlayers, histograms are considered as probability density functions rather than\nmatrices. Thus, the proposed approach is able to utilize the probability\ninformation of the histogram and achieve promising results with a very simple\narchitecture compared to traditional convolutional neural networks. Evaluations\nusing standard benchmarks demonstrate the superiority of the proposed approach\ncompared to state-of-the-art traditional and deep learning methods. To the best\nof our knowledge, this is the first method to propose network layers based on\narithmetic distribution operations for learning distributions during background\nsubtraction.\n"}
{"abstract": "  Owing to the development of research on local aggregation operators, dramatic\nbreakthrough has been made in point cloud analysis models. However, existing\nlocal aggregation operators in the current literature fail to attach decent\nimportance to the local information of the point cloud, which limits the power\nof the models. To fit this gap, we propose an efficient Vector Attention\nConvolution module (VAConv), which utilizes K-Nearest Neighbor (KNN) to extract\nthe neighbor points of each input point, and then uses the elevation and\nazimuth relationship of the vectors between the center point and its neighbors\nto construct an attention weight matrix for edge features. Afterwards, the\nVAConv adopts a dual-channel structure to fuse weighted edge features and\nglobal features. To verify the efficiency of the VAConv, we connect the VAConvs\nwith different receptive fields in parallel to obtain a Multi-scale graph\nconvolutional network, VA-GCN. The proposed VA-GCN achieves state-of-the-art\nperformance on standard benchmarks including ModelNet40, S3DIS and ShapeNet.\nRemarkably, on the ModelNet40 dataset for 3D classification, VA-GCN increased\nby 2.4% compared to the baseline.\n"}
{"abstract": "  The Neolithic Revolution began c. 10000 years ago and is characterised by the\nultimate, near complete transition from hunting and gathering to agricultural\nfood production on land. The Neolithic Revolution is thought to have been\ncatalysed by a combination of local population pressure, cultural diffusion,\nproperty rights and climate change. We undertake a thought experiment that\nexamines trends in these key hypothesised catalysts and patters of today to\nexplore whether society could be on a path towards another paradigm shift in\nfood production: away from hunting of wild fish towards a transition to mostly\nfish farming. We find similar environmental and cultural pressures have driven\nthe rapid rise of aquaculture, during a period that has now been coined the\nBlue Revolution, providing impetus for such a transition in coming decades to\ncenturies. We also highlight the interacting and often mutually reinforcing\nimpacts of 1)technological and scientific advancement, 2)environmental\nawareness and collective action and 3)globalisation and trade influencing the\ntrajectory and momentum of the Blue Revolution. We present two qualitative\nnarratives that broadly fall within two future trajectories: 1)a ubiquitous\naquaculture transition and 20commercial aquaculture and fisheries coexistence.\nThis scenarios approach aims to encourage logical, forward thinking, and\ninnovative solutions to complex systems dynamics. Scenario-based thought\nexperiments are useful to explore large scale questions, increase the\naccessibility to a wider readership and ideally catalyse discussion around\nproactive governance mechanisms. We argue the future is not fixed and society\nnow has greater foresight and capacity to choose the workable balance between\nfisheries sand aquaculture that supports economic, environmental, cultural and\nsocial objectives through combined planning, policies and management.\n"}
{"abstract": "  Exclusive rare Higgs decays into lepton pair plus one light hadron, such as\n$h\\to \\rho^0(\\omega)\\ell\\bar{\\ell}$, $h\\to \\pi^0\\ell\\bar{\\ell}$, $h\\to \\pi^+\n(K^+)\\ell^-\\bar{\\nu}_\\ell$, and $h\\to \\rho^+(K^{*+})\\ell^-\\bar{\\nu}_\\ell$, have\nbeen explored in the standard model. Decay amplitudes are dominantly from the\nHiggs couplings to gauge bosons and to charged leptons, and their branching\nratios are predicted in the range of $10^{-8}\\sim 10^{-5}$. We have also\nanalyzed the differential dilepton invariant mass and angular distributions of\n$h\\to \\rho^0 \\ell^+\\ell^-$ decays. It will be challenging to search for these\nrare processes. Nevertheless, experimental studies of them, in particular,\n$h\\to \\rho^0\\ell^+\\ell^-$ with $\\ell=e,\\mu$, might be interesting both to help\ndeepen our understanding of the standard model and to probe new physics beyond\nthe standard model in the future high-precision experiments.\n"}
{"abstract": "  A desirable goal for autonomous agents is to be able to coordinate on the fly\nwith previously unknown teammates. Known as \"ad hoc teamwork\", enabling such a\ncapability has been receiving increasing attention in the research community.\nOne of the central challenges in ad hoc teamwork is quickly recognizing the\ncurrent plans of other agents and planning accordingly. In this paper, we focus\non the scenario in which teammates can communicate with one another, but only\nat a cost. Thus, they must carefully balance plan recognition based on\nobservations vs. that based on communication. This paper proposes a new metric\nfor evaluating how similar are two policies that a teammate may be following -\nthe Expected Divergence Point (EDP). We then present a novel planning algorithm\nfor ad hoc teamwork, determining which query to ask and planning accordingly.\nWe demonstrate the effectiveness of this algorithm in a range of increasingly\ngeneral communication in ad hoc teamwork problems.\n"}
{"abstract": "  One commonly used clinical approach towards detecting melanomas recognises\nthe existence of Ugly Duckling nevi, or skin lesions which look different from\nthe other lesions on the same patient. An automatic method of detecting and\nanalysing these lesions would help to standardize studies, compared with manual\nscreening methods. However, it is difficult to obtain expertly-labelled images\nfor ugly duckling lesions. We therefore propose to use self-supervised machine\nlearning to automatically detect outlier lesions. We first automatically detect\nand extract all the lesions from a wide-field skin image, and calculate an\nembedding for each detected lesion in a patient image, based on automatically\nidentified features. These embeddings are then used to calculate the L2\ndistances as a way to measure dissimilarity. Using this deep learning method,\nUgly Ducklings are identified as outliers which should deserve more attention\nfrom the examining physician. We evaluate through comparison with\ndermatologists, and achieve a sensitivity rate of 72.1% and diagnostic accuracy\nof 94.2% on the held-out test set.\n"}
{"abstract": "  Let $A$ be an abelian variety over a number field $\\mathrm E\\subset \\mathbb\nC$ and let $\\mathbf G$ denote the Mumford--Tate group of $A$. After replacing\n$\\mathrm E$ by a finite extension, the action of the absolute Galois group\n$\\mathrm{Gal}(\\overline{\\mathrm E}/\\mathrm E)$ on the $\\ell$-adic cohomology\n$\\mathrm{H}^1_{\\mathrm{\\acute{e}t}}(A_{\\overline{\\mathrm E}},\\mathbb Q_\\ell)$\nfactors through $\\mathbf G(\\mathbb Q_\\ell).$ We show that for $v$ an odd prime\nof $\\mathrm E$ where $A$ has good reduction, the conjugacy class of Frobenius\n$\\mathrm{Frob}_v$ in $\\mathbf G(\\mathbb Q_\\ell)$ is independent of $\\ell$.\nAlong the way we prove that every point in the $\\mu$-ordinary locus of the\nspecial fiber of Shimura varieties has a special point lifting it.\n"}
{"abstract": "  A stable map of a closed orientable $3$-manifold into the real plane is\ncalled a stable map of a link in the manifold if the link is contained in the\nset of definite fold points. We give a complete characterization of the\nhyperbolic links in the $3$-sphere that admit stable maps into the real plane\nwith exactly one (connected component of a) fiber having two singular points.\n"}
{"abstract": "  In this paper, we present a neural model for joint dropped pronoun recovery\n(DPR) and conversational discourse parsing (CDP) in Chinese conversational\nspeech. We show that DPR and CDP are closely related, and a joint model\nbenefits both tasks. We refer to our model as DiscProReco, and it first encodes\nthe tokens in each utterance in a conversation with a directed Graph\nConvolutional Network (GCN). The token states for an utterance are then\naggregated to produce a single state for each utterance. The utterance states\nare then fed into a biaffine classifier to construct a conversational discourse\ngraph. A second (multi-relational) GCN is then applied to the utterance states\nto produce a discourse relation-augmented representation for the utterances,\nwhich are then fused together with token states in each utterance as input to a\ndropped pronoun recovery layer. The joint model is trained and evaluated on a\nnew Structure Parsing-enhanced Dropped Pronoun Recovery (SPDPR) dataset that we\nannotated with both two types of information. Experimental results on the SPDPR\ndataset and other benchmarks show that DiscProReco significantly outperforms\nthe state-of-the-art baselines of both tasks.\n"}
{"abstract": "  Given high-dimensional time series data (e.g., sensor data), how can we\ndetect anomalous events, such as system faults and attacks? More challengingly,\nhow can we do this in a way that captures complex inter-sensor relationships,\nand detects and explains anomalies which deviate from these relationships?\nRecently, deep learning approaches have enabled improvements in anomaly\ndetection in high-dimensional datasets; however, existing methods do not\nexplicitly learn the structure of existing relationships between variables, or\nuse them to predict the expected behavior of time series. Our approach combines\na structure learning approach with graph neural networks, additionally using\nattention weights to provide explainability for the detected anomalies.\nExperiments on two real-world sensor datasets with ground truth anomalies show\nthat our method detects anomalies more accurately than baseline approaches,\naccurately captures correlations between sensors, and allows users to deduce\nthe root cause of a detected anomaly.\n"}
{"abstract": "  We carry out a study in the approach of a dimension-8 effective field theory\non the anomalous quartic $W^+W^-\\gamma\\gamma$ couplings that arise from the\nprocess $pp \\to W\\gamma\\gamma$ at the Future Circular Collider-hadron hadron\n(FCC-hh). The process is sensitive to the $W^+W^-\\gamma\\gamma$ vertex and\nlimits may be placed on the anomalous $f_ {M,i}/\\Lambda^4$ and $ f_\n{T,j}/\\Lambda^4$ couplings by measurements of $W\\gamma\\gamma$ production. We\nillustrate the physics opportunities and reach of the FCC-hh operating for a\nhigh integrated luminosity scenario of ${\\cal L}=1, 5, 10, 30$ ${\\rm ab}^{-1}$\nof proton-proton collisions at a center-of-mass energy of $\\sqrt{s}=100$ TeV.\nProduction cross-sections of the signal and the sensitivity constraints on the\nanomalous $f_ {M,i}/\\Lambda^4$ and $ f_ {T,j}/\\Lambda^4$ couplings at $95\\%$\nC.L. are identified through the leptonic and hadronic decay modes of the\n$W$-boson. The best FCC-hh sensitivity limits on these anomalous couplings\nmight reach up to the order of magnitude ${\\cal O}(10^{-4}-10^1)$. These\nresults show that the $pp \\to W\\gamma\\gamma$ channel at the FCC-hh has a\ngreater potential to research for the anomalous quartic $W^+W^-\\gamma\\gamma$\ncouplings and to research for new physics.\n"}
{"abstract": "  In this work, a robust and efficient text-to-speech (TTS) synthesis system\nnamed Triple M is proposed for large-scale online application. The key\ncomponents of Triple M are: 1) A sequence-to-sequence model adopts a novel\nmulti-guidance attention to transfer complementary advantages from guiding\nattention mechanisms to the basic attention mechanism without in-domain\nperformance loss and online service modification. Compared with single\nattention mechanism, multi-guidance attention not only brings better\nnaturalness to long sentence synthesis, but also reduces the word error rate by\n26.8%. 2) A new efficient multi-band multi-time vocoder framework, which\nreduces the computational complexity from 2.8 to 1.0 GFLOP and speeds up LPCNet\nby 2.75x on a single CPU.\n"}
{"abstract": "  Statistical models of real world data typically involve continuous\nprobability distributions such as normal, Laplace, or exponential\ndistributions. Such distributions are supported by many probabilistic modelling\nformalisms, including probabilistic database systems. Yet, the traditional\ntheoretical framework of probabilistic databases focusses entirely on finite\nprobabilistic databases.\n  Only recently, we set out to develop the mathematical theory of infinite\nprobabilistic databases. The present paper is an exposition of two recent\npapers which are cornerstones of this theory. In (Grohe, Lindner; ICDT 2020) we\npropose a very general framework for probabilistic databases, possibly\ninvolving continuous probability distributions, and show that queries have a\nwell-defined semantics in this framework. In (Grohe, Kaminski, Katoen, Lindner;\nPODS 2020) we extend the declarative probabilistic programming language\nGenerative Datalog, proposed by (B\\'ar\\'any et al.~2017) for discrete\nprobability distributions, to continuous probability distributions and show\nthat such programs yield generative models of continuous probabilistic\ndatabases.\n"}
{"abstract": "  We are concerned with the dependence of the lowest positive eigenvalue of the\nDirac operator on the geometry of rectangles, subject to infinite-mass boundary\nconditions. We conjecture that the square is a global minimiser both under the\narea or perimeter constraints. Contrary to well-known non-relativistic\nanalogues, we show that the present spectral problem does not admit explicit\nsolutions. We prove partial optimisation results based on a variational\nreformulation and newly established lower and upper bounds to the Dirac\neigenvalue. We also propose an alternative approach based on symmetries of\nrectangles and a non-convex minimisation problem; this implies a sufficient\ncondition formulated in terms of a symmetry of the minimiser which guarantees\nthe conjectured results.\n"}
{"abstract": "  We give an $O(k^3 n \\log n \\min(k,\\log^2 n) \\log^2(nC))$-time algorithm for\ncomputing maximum integer flows in planar graphs with integer arc {\\em and\nvertex} capacities bounded by $C$, and $k$ sources and sinks. This improves by\na factor of $\\max(k^2,k\\log^2 n)$ over the fastest algorithm previously known\nfor this problem [Wang, SODA 2019].\n  The speedup is obtained by two independent ideas. First we replace an\niterative procedure of Wang that uses $O(k)$ invocations of an $O(k^3 n \\log^3\nn)$-time algorithm for maximum flow algorithm in a planar graph with $k$ apices\n[Borradaile et al., FOCS 2012, SICOMP 2017], by an alternative procedure that\nonly makes one invocation of the algorithm of Borradaile et al. Second, we show\ntwo alternatives for computing flows in the $k$-apex graphs that arise in our\nmodification of Wang's procedure faster than the algorithm of Borradaile et al.\nIn doing so, we introduce and analyze a sequential implementation of the\nparallel highest-distance push-relabel algorithm of Goldberg and Tarjan~[JACM\n1988].\n"}
{"abstract": "  TxGraffiti is an automated conjecturing program that produces graph theoretic\nconjectures in the form of conjectured inequalities. This program written and\nmaintained by the second author since 2017 was inspired by the successes of\nprevious automated conjecturing programs including Fajtlowicz's GRAFFITI and\nDeLaVi\\~{n}a's GRAFFITI.pc. In this paper we prove and generalize several\nconjectures generated by TxGraffiti when it was prompted to conjecture on the\n\\emph{independence number}, the \\emph{domination number}, and the\n\\emph{matching number} (and generalizations of each of these graph invariants).\nMoreover, in several instances we also show the proposed inequalities relating\nthese graph invariants are sharp.\n"}
{"abstract": "  The PC and FCI algorithms are popular constraint-based methods for learning\nthe structure of directed acyclic graphs (DAGs) in the absence and presence of\nlatent and selection variables, respectively. These algorithms (and their\norder-independent variants, PC-stable and FCI-stable) have been shown to be\nconsistent for learning sparse high-dimensional DAGs based on partial\ncorrelations. However, inferring conditional independences from partial\ncorrelations is valid if the data are jointly Gaussian or generated from a\nlinear structural equation model -- an assumption that may be violated in many\napplications. To broaden the scope of high-dimensional causal structure\nlearning, we propose nonparametric variants of the PC-stable and FCI-stable\nalgorithms that employ the conditional distance covariance (CdCov) to test for\nconditional independence relationships. As the key theoretical contribution, we\nprove that the high-dimensional consistency of the PC-stable and FCI-stable\nalgorithms carry over to general distributions over DAGs when we implement\nCdCov-based nonparametric tests for conditional independence. Numerical studies\ndemonstrate that our proposed algorithms perform nearly as good as the\nPC-stable and FCI-stable for Gaussian distributions, and offer advantages in\nnon-Gaussian graphical models.\n"}
{"abstract": "  We present elemental abundance results for HE 2148-2039 and HE 2155-2043\nbased on a detailed high-resolution spectroscopic analysis. The high-resolution\nSUBARU/HDS spectra used for our analysis have a resolution of R~60 000.\nAlthough limited information based on photometry and low-resolution\nspectroscopy are available, we present for the first time an abundance analysis\nbased on high-resolution spectra for both the objects. Our analysis shows that\nthe two objects are extremely metal-poor with [Fe/H]< -3. Among the\nneutron-capture elements, abundances of only Sr and Ba could be determined in\nour programme stars. For both the objects [Ba/Fe] is found to be < 0. While\nstrontium is under abundant in HE 2148-2039 with [Sr/Fe]~ -2.02, Sr is near\nsolar in HE 2155-2043. The locations of the programme stars in the absolute\ncarbon abundance, A(C) vs. [Fe/H] diagram show that HE 2148-2039 is a CEMP-no\nGroup II object and HE 2155-2043 is a CEMP-no Group III object. Observed\n[Sr/Ba] ratios are characteristics of a fast rotating massive star progenitor\nfor HE 2155-2043 and a metal-poor Asymptotic giant branch (AGB) star for HE\n2148-2039. The estimated [Sc/Mn] as well as [C/Cr] ratios in HE 2155-2043 show\nthat the surface chemical composition of this object is mono-enriched. The\nsurface chemical composition of HE 2148-2039 is also found to be mono-enriched\nbased on [Mg/C] ratio. With respect to their locations in the [C/N] vs. T eff\ndiagram, HE 2148-2039 shows signatures of mixing, and HE 2155-2043 falls in the\nunmixed region of [C/N] vs. Teff plot. Kinematic analysis shows that both the\nobjects belong to Galactic halo population.\n"}
{"abstract": "  Estimators for mutual information are typically biased. However, in the case\nof the Kozachenko-Leonenko estimator for metric spaces, a type of nearest\nneighbour estimator, it is possible to calculate the bias explicitly.\n"}
{"abstract": "  This work reports the general design and characterization of two exotic,\nanomalous nonequilibrium topological phases. In equilibrium systems, the Weyl\nnodes or the crossing points of nodal lines may become the transition points\nbetween higher-order and first-order topological phases defined on\ntwo-dimensional slices, thus featuring both hinge Fermi arc and surface Fermi\narc. We advance this concept by presenting a strategy to obtain, using\ntime-sequenced normal insulator phases only, Floquet higher-order Weyl\nsemimetals and Floquet higher-order nexus semimetals, where the concerned\ntopological singularities in the three-dimensional Brillouin zone border\nanomalous two-dimensional higher-order Floquet phases. The fascinating\ntopological phases we obtain are previously unknown and can be experimentally\nstudied using, for example, a three-dimensional lattice of coupled ring\nresonators.\n"}
{"abstract": "  It is showed that complex scalar fields with a self-interaction potential may\npropagate along null geodesics on flat Friedmann--Lema\\^itre--Robertson--Walker\nuniverses with different time-dependent scale factors. This occurs provided\nthey self interact adequately, for different forms of potentials, and\n  even for the massive case.\n"}
{"abstract": "  While many real-world data streams imply that they change frequently in a\nnonstationary way, most of deep learning methods optimize neural networks on\ntraining data, and this leads to severe performance degradation when dataset\nshift happens. However, it is less possible to annotate or inspect newly\nstreamed data by humans, and thus it is desired to measure model drift at\ninference time in an unsupervised manner. In this paper, we propose a novel\nmethod of model drift estimation by exploiting statistics of batch\nnormalization layer on unlabeled test data. To remedy possible sampling error\nof streamed input data, we adopt low-rank approximation to each\nrepresentational layer. We show the effectiveness of our method not only on\ndataset shift detection but also on model selection when there are multiple\ncandidate models among model zoo or training trajectories in an unsupervised\nway. We further demonstrate the consistency of our method by comparing model\ndrift scores between different network architectures.\n"}
{"abstract": "  Measurement plays a quintessential role in the control of quantum systems.\nBeyond initialization and readout which pertain to projective measurements,\nweak measurements in particular, through their back-action on the system, may\nenable various levels of coherent control. The latter ranges from observing\nquantum trajectories to state dragging and steering. Furthermore, just like the\nadiabatic evolution of quantum states that is known to induce the Berry phase,\nsequential weak measurements may lead to path-dependent geometric phases. Here\nwe measure the geometric phases induced by sequences of weak measurements and\ndemonstrate a topological transition in the geometric phase controlled by\nmeasurement strength. This connection between weak measurement induced quantum\ndynamics and topological transitions reveals subtle topological features in\nmeasurement-based manipulation of quantum systems. Our protocol could be\nimplemented for classes of operations (e.g. braiding) which are topological in\nnature. Furthermore, our results open new horizons for measurement-enabled\nquantum control of many-body topological states.\n"}
{"abstract": "  Evolution algebras are non-associative algebras that describe non-Mendelian\nhereditary processes and have connections with many other areas. In this paper\nwe obtain necessary and sufficient conditions for a given algebra $A$ to be an\nevolution algebra. We prove that the problem is equivalent to the so-called\n$SDC$ $problem$, that is, the $simultaneous$ $diagonalisation$ $via$\n$congruence$ of a given set of matrices. More precisely we show that an\n$n$-dimensional algebra $A$ is an evolution algebra if, and only if, a certain\nset of $n$ symmetric $n\\times n$ matrices $\\{M_{1}, \\ldots, M_{n}\\}$ describing\nthe product of $A$ are $SDC$. We apply this characterisation to show that while\ncertain classical genetic algebras (representing Mendelian and auto-tetraploid\ninheritance) are not themselves evolution algebras, arbitrarily small\nperturbations of these are evolution algebras. This is intriguing as evolution\nalgebras model asexual reproduction unlike the classical ones.\n"}
{"abstract": "  In this work we investigate symmetry breaking in the presence of a turbulent\nenvironment. The transition from a symmetric state to a symmetry-breaking state\nis demonstrated using two examples: (i) the transition of a two-dimensional\nflow to a three dimensional flow as the fluid layer thickness is varied and\n(ii) the dynamo instability in a thin layer flow as the magnetic Reynolds\nnumber is varied. We show that these examples have similar critical exponents\nthat differ from the mean-field predictions. The critical behavior can be\nrelated to the multiplicative nature of the fluctuations and can be predicted\nin certain limits using results from the statistical properties of random\ninterfaces. Our results indicate the possibility of existence of a new class of\nout-of-equilibrium phase transition controlled by the multiplicative noise.\n"}
{"abstract": "  To characterize the \"average\" of a sample of graphs, one can compute the\nsample Frechet mean (or median) graph, which provides an interpretable summary\nof the graph sample. In this paper, we address the following foundational\nquestion: does the mean or median graph inherit the structural properties of\nthe graphs in the sample? An important graph property is the edge density.\nBecause sparse graphs provide prototypical models for real networks, one would\nlike to guarantee that the edge density be preserved when computing the sample\nmean (or median). In this paper, we prove that the edge density is an\nhereditary property, which can be transmitted from a graph sample to its sample\nFrechet mean (or median), irrespective of the method used to estimate the mean\nor the median. Specifically, we prove the following result: the number of edges\nof the Frechet mean (or median) graph of a set of graphs is bounded by the\nmaximal number of edges amongst all the graphs in the sample. We prove the\nresult for the graph Hamming distance, and the spectral adjacency pseudometric,\nusing very different arguments.\n"}
{"abstract": "  The combination of the Internet of Things and the Edge Computing gives many\nopportunities to support innovative applications close to end users. Numerous\ndevices present in both infrastructures can collect data upon which various\nprocessing activities can be performed. However, the quality of the outcomes\nmay be jeopardized by the presence of outliers. In this paper, we argue on a\nnovel model for outliers detection by elaborating on a `soft' approach. Our\nmechanism is built upon the concepts of candidate and confirmed outliers. Any\ndata object that deviates from the population is confirmed as an outlier only\nafter the study of its sequence of magnitude values as new data are\nincorporated into our decision making model. We adopt the combination of a\nsliding with a landmark window model when a candidate outlier is detected to\nexpand the sequence of data objects taken into consideration. The proposed\nmodel is fast and efficient as exposed by our experimental evaluation while a\ncomparative assessment reveals its pros and cons.\n"}
{"abstract": "  The newly established Heisenberg limit in arbitrary waveform estimation is\nquite different with parameter estimation and shows a unique characteristic of\na future quantum version of oscilloscope. However, it is still a non-trivial\nchallenge to generate a large number of exotic quantum entangled states to\nachieve this quantum limit. Here, by employing the time-domain quantum\ndifference detection method, we demonstrate Heisenberg-limited waveform quantum\nestimation with diamond spins under ambient condition in the experiment.\nPeriodic dynamical decoupling is applied to enhance both the dynamic range and\nsensitivity by one order of magnitude. Using this quantum-enhanced estimation\nscheme, the estimation error of an unknown waveform is reduced by more than $5$\ndB below the standard quantum limit with $N\\sim{\\text{2}} \\times\n{\\text{1}}{{\\text{0}}^3}$ resources, where more than ${1 \\times\n{\\text{1}}{{\\text{0}}^5}}$ resources would be required to achieve a similar\nerror level using classical detection. This work provides an essential step\ntowards realizing quantum-enhanced structure recognition in a continuous space\nand time.\n"}
{"abstract": "  Content moderation is often performed by a collaboration between humans and\nmachine learning models. However, it is not well understood how to design the\ncollaborative process so as to maximize the combined moderator-model system\nperformance. This work presents a rigorous study of this problem, focusing on\nan approach that incorporates model uncertainty into the collaborative process.\nFirst, we introduce principled metrics to describe the performance of the\ncollaborative system under capacity constraints on the human moderator,\nquantifying how efficiently the combined system utilizes human decisions. Using\nthese metrics, we conduct a large benchmark study evaluating the performance of\nstate-of-the-art uncertainty models under different collaborative review\nstrategies. We find that an uncertainty-based strategy consistently outperforms\nthe widely used strategy based on toxicity scores, and moreover that the choice\nof review strategy drastically changes the overall system performance. Our\nresults demonstrate the importance of rigorous metrics for understanding and\ndeveloping effective moderator-model systems for content moderation, as well as\nthe utility of uncertainty estimation in this domain.\n"}
{"abstract": "  There is a recent interest in investigating few-shot NER, where the\nlow-resource target domain has different label sets compared with a\nresource-rich source domain. Existing methods use a similarity-based metric.\nHowever, they cannot make full use of knowledge transfer in NER model\nparameters. To address the issue, we propose a template-based method for NER,\ntreating NER as a language model ranking problem in a sequence-to-sequence\nframework, where original sentences and statement templates filled by candidate\nnamed entity span are regarded as the source sequence and the target sequence,\nrespectively. For inference, the model is required to classify each candidate\nspan based on the corresponding template scores. Our experiments demonstrate\nthat the proposed method achieves 92.55% F1 score on the CoNLL03 (rich-resource\ntask), and significantly better than fine-tuning BERT 10.88%, 15.34%, and\n11.73% F1 score on the MIT Movie, the MIT Restaurant, and the ATIS\n(low-resource task), respectively.\n"}
{"abstract": "  A new method for calculating the symmetry-projected energy of coupled-cluster\nsingles and doubles (CCSD) wave function through the Monte Carlo method is\nproposed. We present benchmark calculations in considering the three-level\nLipkin model which is a simple and minimal model with two phases: spherical and\ndeformed ones. It is demonstrated that this new method gives good ground state\nenergy and low-lying spectra.\n"}
{"abstract": "  Internet of Things (IoT)-based smart home applications are rising in\npopularity. However, this trend attracts malicious activity, which causes\ncost-efficient security to be in high demand. This paper proposes a low-end\ndesign that reinforces the security of a home network. It uses private\nblockchain technology and localization via RSSI-based trilateration. We\ninvestigated the benefits of private blockchains over their public counterpart,\nand we improve the precision of the localization algorithm by testing it\nagainst different wireless technologies. The results conclude that using a\nprivate blockchain with a WiFi-based communication system produces the most\nefficient iteration of the proposed design.\n"}
{"abstract": "  The beam aperture of a particle accelerator defines the clearance available\nfor the circulating beams and is a parameter of paramount importance for the\naccelerator performance. At the CERN Large Hadron Collider (LHC), the knowledge\nand control of the available aperture is crucial because the nominal proton\nbeams carry an energy of 362 MJ stored in a superconducting environment. Even a\ntiny fraction of beam losses could quench the superconducting magnets or cause\nsevere material damage. Furthermore, in a circular collider, the performance in\nterms of peak luminosity depends to a large extent on the aperture of the inner\ntriplet quadrupoles, which are used to focus the beams at the interaction\npoints. In the LHC, this aperture represents the smallest aperture at\ntop-energy with squeezed beams and determines the maximum potential reach of\nthe peak luminosity. Beam-based aperture measurements in these conditions are\ndifficult and challenging. In this paper, we present different methods that\nhave been developed over the years for precise beam-based aperture measurements\nin the LHC, highlighting applications and results that contributed to boost the\noperational LHC performance in Run 1 (2010-2013) and Run 2 (2015-2018).\n"}
{"abstract": "  Propylene oxide is one of the simplest organic chiral molecules and has\nattracted considerable interest from the scientific community a few years ago,\nwhen it was discovered in the interstellar medium. Here, we report a\npreliminary study on the interaction between propylene oxide and rare-gas\natoms, specifically He, Ne, and Ar. The interaction potentials as a function of\nthe distance between the center-of-mass of propylene oxide and the\nrare-gas-atom are calculated for fourteen leading configurations at\nCCSD(T)/aug-cc-pVDZ level of theory. Symmetry Adapted Perturbation Theory has\nbeen employed for the analysis of the intermolecular potential, revealing that\nmost of the contribution is given by dispersion and exchange forces.\n"}
{"abstract": "  This work studies a two-component Fornberg-Whitham (FW) system, which can be\nconsidered as a model for the propagation of shallow water waves. It's known\nthat its solutions depend continuously on their initial data from the local\nwell-posedness result. In this paper, we further show that such dependence is\nnot uniformly continuous in $H^{s}(R)\\times H^{s-1}(R)$ for $s>\\frac{3}{2}$,\nbut H\\\"{o}ler continuous in a weaker topology. Besides, we also establish that\nthe FW system is ill-posed in the critical Sobolev space\n$H^{\\frac{3}{2}}(R)\\times H^{\\frac{1}{2}}(R)$ by proving the norm inflation.\n"}
{"abstract": "  Versatile nanoscale sensors that are susceptible to changes in a variety of\nphysical quantities often exhibit limited selectivity. This paper reports a\nnovel scheme based on microwave-dressed spin states for optically probed\nnanoscale temperature detection using diamond quantum sensors, which provides\nselective sensitivity to temperature changes. By combining this scheme with a\ncontinuous pump-probe scheme using ensemble nitrogen-vacancy centers in\nnanodiamonds, a sub-microsecond temporal resolution with thermal sensitivity of\n3.7 K$\\cdot$Hz$^{-1/2}$ that is insensitive to variations in external magnetic\nfields on the order of 2 G is demonstrated. The presented results are favorable\nfor the practical application of time-resolved nanoscale quantum sensing, where\ntemperature imaging is required under fluctuating magnetic fields.\n"}
{"abstract": "  First, we construct some families of nonsolvable anticommutative algebras,\nsolvable Lie algebras and even nilpotent Lie algebras, that can be endowed with\nthe structure of simple Hom-Lie algebras. This situation shows that a\nclassification of simple Hom-Lie algebras would be unrealistic without any\nfurther restrictions. Therefore, we introduce the class of \\emph{strongly\nsimple Hom-Lie algebras}, which is the class of anticommutative algebras that\nare simple Hom-Lie with respect to all their twisting maps. We show some of its\nproperties, provide a characterisation and explore some of its subclasses.\nThen, we classify completely regular simple Hom-Lie algebras over any arbitrary\nfield. Furthermore, we establish that every simple anticommutative algebra of\ndimension $3$ turns out to be a simple Lie algebra where its Lie bracket is\ndeformed by a bijective linear map, and also we determine all the simple\nHom-Lie algebras in dimension $2$, that were wrongly claimed to be nonexistent\nin \\cite{CH}.\n"}
{"abstract": "  We propose OmniLytics, a blockchain-based secure data trading marketplace for\nmachine learning applications. Utilizing OmniLytics, many distributed data\nowners can contribute their private data to collectively train an ML model\nrequested by some model owners, and receive compensation for data contribution.\nOmniLytics enables such model training while simultaneously providing 1) model\nsecurity against curious data owners; 2) data security against the curious\nmodel and data owners; 3) resilience to malicious data owners who provide\nfaulty results to poison model training; and 4) resilience to malicious model\nowners who intend to evade payment. OmniLytics is implemented as a blockchain\nsmart contract to guarantee the atomicity of payment. In OmniLytics, a model\nowner splits its model into the private and public parts and publishes the\npublic part on the contract. Through the execution of the contract, the\nparticipating data owners securely aggregate their locally trained models to\nupdate the model owner's public model and receive reimbursement through the\ncontract. We implement a working prototype of OmniLytics on Ethereum blockchain\nand perform extensive experiments to measure its gas cost, execution time, and\nmodel quality under various parameter combinations. For training a CNN on the\nMNIST dataset, the MO is able to boost its model accuracy from 62% to 83%\nwithin 500ms in blockchain processing time.This demonstrates the effectiveness\nof OmniLytics for practical deployment.\n"}
{"abstract": "  We introduce a general approach for modeling the dynamic of multivariate time\nseries when the data are of mixed type (binary/count/continuous). Our method is\nquite flexible and conditionally on past values, each coordinate at time $t$\ncan have a distribution compatible with a standard univariate time series model\nsuch as GARCH, ARMA, INGARCH or logistic models whereas past values of the\nother coordinates play the role of exogenous covariates in the dynamic. The\nsimultaneous dependence in the multivariate time series can be modeled with a\ncopula. Additional exogenous covariates are also allowed in the dynamic. We\nfirst study usual stability properties of these models and then show that\nautoregressive parameters can be consistently estimated equation-by-equation\nusing a pseudo-maximum likelihood method, leading to a fast implementation even\nwhen the number of time series is large. Moreover, we prove consistency results\nwhen a parametric copula model is fitted to the time series and in the case of\nGaussian copulas, we show that the likelihood estimator of the correlation\nmatrix is strongly consistent. We carefully check all our assumptions for two\nprototypical examples: a GARCH/INGARCH model and logistic/log-linear INGARCH\nmodel. Our results are illustrated with numerical experiments as well as two\nreal data sets.\n"}
{"abstract": "  Inferring the probability distribution of sentences or word sequences is a\nkey process in natural language processing. While word-level language models\n(LMs) have been widely adopted for computing the joint probabilities of word\nsequences, they have difficulty in capturing a context long enough for sentence\nprobability estimation (SPE). To overcome this, recent studies introduced\ntraining methods using sentence-level noise-contrastive estimation (NCE) with\nrecurrent neural networks (RNNs). In this work, we attempt to extend it for\ncontextual SPE, which aims to estimate a conditional sentence probability given\na previous text. The proposed NCE samples negative sentences independently of a\nprevious text so that the trained model gives higher probabilities to the\nsentences that are more consistent with \\textcolor{blue}{the} context. We apply\nour method to a simple word-level RNN LM to focus on the effect of the\nsentence-level NCE training rather than on the network architecture. The\nquality of estimation was evaluated against multiple-choice cloze-style\nquestions including both human and automatically generated questions. The\nexperimental results show that the proposed method improved the SPE quality for\nthe word-level RNN LM.\n"}
{"abstract": "  The way molecules absorb, transfer, and emit light can be modified by\ncoupling them to optical cavities. The extent of the modification is often\ndefined by the cavity-molecule coupling strength, which depends on the number\nof coupled molecules. We experimentally and numerically study the evolution of\nphotoemission from a thin layered J-aggregated molecular material strongly\ncoupled to a Fabry-Perot microcavity as a function of the number of coupled\nlayers. We unveil an important difference between the strong coupling\nsignatures obtained from reflection spectroscopy and from polariton assisted\nphotoluminescence. We also study the effect of the vibrational modes supported\nby the molecular material on the polariton assisted emission both for a focused\nlaser beam and for normally incident excitation, for two different excitation\nwavelengths: a laser in resonance with the lower polariton branch, and a laser\nnot in resonance. We found that the Raman scattered photons play an important\nrole in populating the lower polariton branch, especially when the system was\nexcited with a laser in resonance with the lower polariton branch. We also\nfound that the polariton assisted photoemission depends on the extent of\nmodification of the molecular absorption induced by the molecule-cavity\ncoupling.\n"}
{"abstract": "  Motivated by recent experiments on the phonon contribution to the thermal\nHall effect in the cuprates, we present an analysis of chiral phonon transport.\nWe assume the chiral behavior arises from a non-zero phonon Hall vicosity,\nwhich is likely induced by the coupling to electrons. Phonons with a non-zero\nphonon Hall viscosity have an intrinsic thermal Hall conductivity, but Chen et\nal. (Phys. Rev. Lett. 124, 167601 (2020)) have argued that a significantly\nlarger thermal Hall conductivity can arise from an extrinsic contribution which\nis inversely proportional to the density of impurities. We solve the Boltzmann\nequation for phonon transport and compute the temperature ($T$) dependence of\nthe thermal Hall conductivity originating from skew scattering off point-like\nimpurities. We find that the dominant source for thermal Hall transport is an\ninterference between impurity skew scattering channels with opposite parity.\nThe thermal Hall conductivity $\\sim T^{d+2}$ at low $T$ in $d$ dimensions, and\nhas a window of $T$-independent behavior for $T > T_{\\rm imp}$, where $T_{\\rm\nimp}$ is determined by the ratio of scattering potentials with opposite parity.\nWe also consider the role of non-specular scattering off the sample boundary,\nand find that it leads to negligible corrections to thermal Hall transport at\nlow $T$.\n"}
{"abstract": "  The nonanalyticity and the sign problem in the Z3-symmetric heavy quark model\nat low temperature are studied phenomenologically. For the free heavy quarks,\nthe nonanalyticity is analyzed in the relation to the zeros of the grand\ncanonical partition function. The Z3-symmetric effective Polyakov-line model\n(EPLM) in strong coupling limit is also considered as an phenomenological model\nof Z3-symmetric QCD with large quark mass at low temperature. We examine how\nthe Z3-symmetric EPLM approaches to the original one in the zero-temperature\nlimit. The effects of the Z3-symmetry affect the structure of zeros of the\nmicroscopic probability density function at the nonanalytic point. The average\nvalue of the Polyakov line can detect the structure, while the other\nthermodynamic quantities are not sensible to the structure in the\nzero-temperature limit. The effect of the imaginary quark chemical potential is\nalso discussed. The imaginary part of the quark number density is very\nsensitive to the symmetry structure at the nonanalytical point. For a\nparticular value of the imaginary quark number chemical potential, large quark\nnumber may be induced in the vicinity of the nonanalytical point.\n"}
{"abstract": "  In the residential sector, electric water heaters are appliances with a\nrelatively high power consumption and a significant thermal inertia, which is\nparticularly suitable for Demand Response schemes. The success of efficient DR\nschemes via the control of water heaters presupposes an accurate estimate of\ntheir power demand at each instant. Although the load of water heaters is\nrarely directly measured, a large penetration of Smart Meters (SMs) in\ndistribution grids enables to indirectly infer this information on a large\nscale via load disaggregation. For that purpose, a considerable number of\nNon-Intrusive Load Monitoring (NILM) approaches are suggested in the\nliterature. However, they require data streams at a time resolution in the\nrange of one second or higher, which is not realistic for standard SMs. Hence,\nthis paper proposes an unsupervised approach to detect and disaggregate the\nload profile of water heaters from standard SM data with a time resolution in\nthe minute range. Evaluated on multiple real loads with sub-metering, the\nproposed approach achieves a Normalized Mean Absolute Error (NMAE) lower than\n2% and a precision generally higher than 92% with time resolutions between 5\nand 15 minutes.\n"}
{"abstract": "  Field measurement campaigns have grown exponentially in recent years,\nstemming from the need for reliable data to validate urban climate models and\nobtain a better understanding of urban climate features. Also contributing to\nthis growth is the Local Climate Zone (LCZ) scheme, firstly developed to\nenhance the accuracy in the contextualisation of urban measurements, and lately\nused for characterising urban areas. Due to its relative novelty, researchers\nare still investigating the potential of LCZs and its indicators for urban\ntemperature variability detection. In this respect, the present study\nintroduces the results of an extensive monitoring campaign carried out in the\ncity of Madrid over a two-year period (2016-2018). The aim of this work is to\nfurther examine the relationships between LCZs and air temperature differences,\nwith emphasis on their hourly and seasonal evolution. A graphical and\nstatistical analysis to identify temperature variability trends for each LCZ is\nperformed. Results support the existing evidence suggesting a high level of\neffectiveness in capturing the heat island (UHI) profile of different urban\nareas, while underperforming when it comes to capturing diurnal temperature\nvariability. The incorporation of indicators that explain the daytime\ntemperature variation phenomenon into the LCZ scheme is therefore recommended,\nwarranting further research.\n"}
{"abstract": "  The polyhedral model allows a structured way of defining semantics-preserving\ntransformations to improve the performance of a large class of loops. Finding\nprofitable points in this space is a hard problem which is usually approached\nby heuristics that generalize from domain-expert knowledge. Existing problem\nformulations in state-of-the-art heuristics depend on the shape of particular\nloops, making it hard to leverage generic and more powerful optimization\ntechniques from the machine learning domain. In this paper, we propose PolyGym,\na shape-agnostic formulation for the space of legal transformations in the\npolyhedral model as a Markov Decision Process (MDP). Instead of using\ntransformations, the formulation is based on an abstract space of possible\nschedules. In this formulation, states model partial schedules, which are\nconstructed by actions that are reusable across different loops. With a simple\nheuristic to traverse the space, we demonstrate that our formulation is\npowerful enough to match and outperform state-of-the-art heuristics. On the\nPolybench benchmark suite, we found transformations that led to a speedup of\n3.39x over LLVM O3, which is 1.83x better than the speedup achieved by ISL. Our\ngeneric MDP formulation enables using reinforcement learning to learn\noptimization policies over a wide range of loops. This also contributes to the\nemerging field of machine learning in compilers, as it exposes a novel problem\nformulation that can push the limits of existing methods.\n"}
{"abstract": "  Prediction models often fail if train and test data do not stem from the same\ndistribution. Out-of-distribution (OOD) generalization to unseen, perturbed\ntest data is a desirable but difficult-to-achieve property for prediction\nmodels and in general requires strong assumptions on the data generating\nprocess (DGP). In a causally inspired perspective on OOD generalization, the\ntest data arise from a specific class of interventions on exogenous random\nvariables of the DGP, called anchors. Anchor regression models, introduced by\nRothenhaeusler et al. (2021), protect against distributional shifts in the test\ndata by employing causal regularization. However, so far anchor regression has\nonly been used with a squared-error loss which is inapplicable to common\nresponses such as censored continuous or ordinal data. Here, we propose a\ndistributional version of anchor regression which generalizes the method to\npotentially censored responses with at least an ordered sample space. To this\nend, we combine a flexible class of parametric transformation models for\ndistributional regression with an appropriate causal regularizer under a more\ngeneral notion of residuals. In an exemplary application and several simulation\nscenarios we demonstrate the extent to which OOD generalization is possible.\n"}
{"abstract": "  Shannon entropy in position ($S_{\\rvec}$) and momentum ($S_{\\pvec}$) spaces,\nalong with their sum ($S_t$) are presented for unit-normalized densities of He,\nLi$^+$ and Be$^{2+}$ ions, spatially confined at the center of an impenetrable\nspherical enclosure defined by a radius $r_c$. Both ground as well as some\nselected low-lying singly excited states, \\emph{viz.,} 1sns (n $=$ 2-4) $^3$S,\n1snp (n $=$ 2-3) $^3$P, 1s3d $^3$D are considered within a density functional\nmethodology that makes use of a work-function-based exchange potential along\nwith two correlation potentials (local Wigner-type parametrized functional as\nwell as the more involved non-linear gradient- and Laplacian-dependent\nLee-Yang-Parr functional). The radial Kohn-Sham (KS) equation is solved using\nan optimal spatial discretization scheme via the generalized pseudospectral\n(GPS) method. A detailed systematic analysis of the confined system (relative\nto corresponding free system) has been performed for these quantities with\nrespect to $r_c$ in tabular and graphical forms, \\emph{with and without}\nelectron correlation. Due to compression, the pattern of entropy in\naforementioned states gets characterized by various crossovers at intermediate\nand lower $r_c$ regions. The impact of electron correlation is more pronounced\nin weaker confinement limit, and appears to decay with rise in confinement\nstrength. The exchange-only results are quite good to provide a decent\nqualitative discussion. The lower-bounds provided by entropic uncertainty\nrelation holds good in all cases. Several other new interesting features are\nobserved.\n"}
{"abstract": "  Many non-linear scalar field theories possess a screening mechanism that can\nsuppress any associated fifth force in dense environments. As a result, these\ntheories can evade local experimental tests of new forces. Chameleon-like\nscreening, which occurs because of non-linearities in the scalar potential or\nthe coupling to matter, is well understood around extended objects. However,\nmany experimental tests of these theories involve objects with spatial extent\nmuch smaller than the scalar field's Compton wavelength, and which could\ntherefore be considered point-like. In this work, we determine how the fifth\nforces are screened in the limit that the source objects become extremely\ncompact.\n"}
{"abstract": "  Controllable Image Captioning (CIC) -- generating image descriptions\nfollowing designated control signals -- has received unprecedented attention\nover the last few years. To emulate the human ability in controlling caption\ngeneration, current CIC studies focus exclusively on control signals concerning\nobjective properties, such as contents of interest or descriptive patterns.\nHowever, we argue that almost all existing objective control signals have\noverlooked two indispensable characteristics of an ideal control signal: 1)\nEvent-compatible: all visual contents referred to in a single sentence should\nbe compatible with the described activity. 2) Sample-suitable: the control\nsignals should be suitable for a specific image sample. To this end, we propose\na new control signal for CIC: Verb-specific Semantic Roles (VSR). VSR consists\nof a verb and some semantic roles, which represents a targeted activity and the\nroles of entities involved in this activity. Given a designated VSR, we first\ntrain a grounded semantic role labeling (GSRL) model to identify and ground all\nentities for each role. Then, we propose a semantic structure planner (SSP) to\nlearn human-like descriptive semantic structures. Lastly, we use a role-shift\ncaptioning model to generate the captions. Extensive experiments and ablations\ndemonstrate that our framework can achieve better controllability than several\nstrong baselines on two challenging CIC benchmarks. Besides, we can generate\nmulti-level diverse captions easily. The code is available at:\nhttps://github.com/mad-red/VSR-guided-CIC.\n"}
{"abstract": "  More than 20 years after Fickett attempted to prove the Hyers-Ulam stability\nof isometries defined on bounded subsets of $\\mathbb{R}^n$ in 1981,\nV\\\"{a}is\\\"{a}l\\\"{a} improved Fickett's result significantly. In this paper, we\nwill improve Fickett's theorem by proving the Hyers-Ulam stability of\nisometries defined on bounded subsets of $\\mathbb{R}^n$ using a more intuitive\nmethod different from that used by V\\\"{a}is\\\"{a}l\\\"{a}.\n"}
{"abstract": "  We extend Bourke and Garner's idempotent adjunction between monads and\npretheories to the framework of $\\infty$-categories and we use this to prove\nmany classical results about monads in the $\\infty$-categorical framework.\nAmongst other things, we show that the category of algebras for an accessible\nmonads on a locally presentable $\\infty$-category $\\mathcal{E}$ is again\nlocally presentable, and that a diagram of accessible monads on a locally\npresentable $\\infty$-category admits a colimit. Our results also provide a new\nand simpler way to construct and describe monads in terms of theories.\n"}
{"abstract": "  Generalization of machine learning models trained on a set of source domains\non unseen target domains with different statistics, is a challenging problem.\nWhile many approaches have been proposed to solve this problem, they only\nutilize source data during training but do not take advantage of the fact that\na single target example is available at the time of inference. Motivated by\nthis, we propose a method that effectively uses the target sample during\ninference beyond mere classification. Our method has three components - (i) A\nlabel-preserving feature or metric transformation on source data such that the\nsource samples are clustered in accordance with their class irrespective of\ntheir domain (ii) A generative model trained on the these features (iii) A\nlabel-preserving projection of the target point on the source-feature manifold\nduring inference via solving an optimization problem on the input space of the\ngenerative model using the learned metric. Finally, the projected target is\nused in the classifier. Since the projected target feature comes from the\nsource manifold and has the same label as the real target by design, the\nclassifier is expected to perform better on it than the true target. We\ndemonstrate that our method outperforms the state-of-the-art Domain\nGeneralization methods on multiple datasets and tasks.\n"}
{"abstract": "  Optical spectroscopy and density-functional calculations reveal electronic\nproperties of the nonmagnetickagome metal KV$_3$Sb$_5$. Temperature and\nfrequency-dependent optical measurements down to 10K and up to 2 eV energy\nrange confirm bulk nature of the charge-density-wave (CDW) state below 78 K and\ngauge the charge gap of $\\Delta_{CDW} \\approx$ 60 meV at 10 K. We further\ndetect strong phonon anomalies and the prominent low-energy localization peak\nindicative of the unconventional charge transport caused by electron-phonon or\nelectron-electron interactions. Possible CDW structures of KV$_3$Sb$_5$, the\nstar and hexagon (inverse star), are strongly reminiscent of $p$-wave states\nexpected in the Hubbard model on the kagome lattice at the filling level of the\nvan Hove singularity. The proximity to this regime may have intriguing and\nfar-reaching implications for the physics of KV$_3$Sb$_5$ and related\nmaterials.\n"}
{"abstract": "  In this paper, we investigate the evolution of spacelike curves in\nLorentz-Minkowski plane $\\mathbb{R}^{2}_{1}$ along prescribed geometric flows\n(including the classical curve shortening flow or mean curvature flow as a\nspecial case), which correspond to a class of quasilinear parabolic initial\nboundary value problems, and can prove that this flow exists for all time.\nMoreover, we can also show that the evolving spacelike curves converge to a\nspacelike straight line or a spacelike Grim Reaper curve as time tends to\ninfinity.\n"}
{"abstract": "  Higher-order networks have so far been considered primarily in the context of\nstudying the structure of complex systems, i.e., the higher-order or multi-way\nrelations connecting the constituent entities. More recently, a number of\nstudies have considered dynamical processes that explicitly account for such\nhigher-order dependencies, e.g., in the context of epidemic spreading processes\nor opinion formation. In this chapter, we focus on a closely related, but\ndistinct third perspective: how can we use higher-order relationships to\nprocess signals and data supported on higher-order network structures. In\nparticular, we survey how ideas from signal processing of data supported on\nregular domains, such as time series or images, can be extended to graphs and\nsimplicial complexes. We discuss Fourier analysis, signal denoising, signal\ninterpolation, and nonlinear processing through neural networks based on\nsimplicial complexes. Key to our developments is the Hodge Laplacian matrix, a\nmulti-relational operator that leverages the special structure of simplicial\ncomplexes and generalizes desirable properties of the Laplacian matrix in graph\nsignal processing.\n"}
{"abstract": "  It has been known that even though two elemental metals, $X$ and $Y$, are\nimmiscible, they can form alloys on surfaces of other metal $Z$. In order to\nunderstand such surface alloying of immiscible metals, we study the energetic\nstability of binary alloys, $XZ$ and $YZ$, in several structures with various\ncoordination numbers (CNs). By analyzing the formation energy modified to\nenhance the subtle energy difference between metastable structures, we find\nthat $XZ$ and $YZ$ with B2-type structure (CN$=$8) become energetically stable\nwhen the $X$ and $Y$ metals form an alloy on the $Z$ metal surface. This is\nconsistent with the experimental results for Pb-Sn alloys on metal surfaces\nsuch as Rh(111) and Ru(0001). Some suitable metal substrates are also predicted\nto form Pb-Sn alloys.\n"}
{"abstract": "  The moduli space of $n$ chains in the plane with generic side lengths that\nterminate on a fixed line is a smooth, closed manifold of dimension $n-1$. This\nmanifold is also equipped with a locally standard action of\n$\\mathbb{Z}_2^{n-1}$. The orbit space of this action is a simple polytope\ncalled the moment polytope. Interestingly, this manifold is also the fixed\npoint set of an involution on a toric manifold known as the abelian polygon\nspace. In this article we show that the moment polytope of the moduli space of\nchains is completely characterized by the combinatorial data, called the\n\\emph{short code} of the length vector. We also classify aspherical chain\nspaces using a result of Davis, Januszkiewicz and Scott.\n"}
{"abstract": "  Open Cloud Robot Table Organization Challenge (OCRTOC) is one of the most\ncomprehensive cloud-based robotic manipulation competitions. It focuses on\nrearranging tabletop objects using vision as its primary sensing modality. In\nthis extended abstract, we present our entry to the OCRTOC2020 and the key\nchallenges the team has experienced.\n"}
{"abstract": "  Let $G=(V,E)$ be a finite simple graph. The Sombor index $SO(G)$ of $G$ is\ndefined as $\\sum_{uv\\in E(G)}\\sqrt{d_u^2+d_v^2}$, where $d_u$ is the degree of\nvertex $u$ in $G$. Let $G$ be a connected graph constructed from pairwise\ndisjoint connected graphs $G_1,\\ldots ,G_k$ by selecting a vertex of $G_1$, a\nvertex of $G_2$, and identifying these two vertices. Then continue in this\nmanner inductively. We say that $G$ is a polymer graph, obtained by\npoint-attaching from monomer units $G_1,...,G_k$. In this paper, we consider\nsome particular cases of these graphs that are of importance in chemistry and\nstudy their Sombor index.\n"}
{"abstract": "  We revisit Wagoner et al. (1967), a classic contribution in the development\nof Big Bang Nucleosynthesis. We demonstrate that it presents an incorrect\nexpression for the temperature of the early universe as a function of time in\nthe high temperature limit, $T \\gtrsim 10^{10}$K. As this incorrect expression\nhas been reproduced elsewhere, we present a corrected form for the initial\nconditions required for calculating the formation of the primordial elements in\nthe Big Bang.\n"}
{"abstract": "  Stable organic radicals serve as model systems for investigating metal-free\nmagnetic phenomena at (sub)nanometer length scales. We have investigated at the\nsingle-molecule level the stable $\\pi$-radical 2,2-diphenyl-1-picrylhydrazyl\n(DPPH) adsorbed at the elbow sites of the $22\\times\\sqrt{3}$ reconstructed\nAu(111) surface. Low-temperature scanning tunneling microscopy (STM) and\n-spectroscopy under ultrahigh vacuum conditions at 8~K reveal structural as\nwell as frontier-orbital related electronic details of DPPH/Au(111). A\nKondo-like spectroscopic signature indicates preservation of the unpaired\nelectron spin state.\n"}
{"abstract": "  Language coverage bias, which indicates the content-dependent differences\nbetween sentence pairs originating from the source and target languages, is\nimportant for neural machine translation (NMT) because the target-original\ntraining data is not well exploited in current practice. By carefully designing\nexperiments, we provide comprehensive analyses of the language coverage bias in\nthe training data, and find that using only the source-original data achieves\ncomparable performance with using full training data. Based on these\nobservations, we further propose two simple and effective approaches to\nalleviate the language coverage bias problem through explicitly distinguishing\nbetween the source- and target-original training data, which consistently\nimprove the performance over strong baselines on six WMT20 translation tasks.\nComplementary to the translationese effect, language coverage bias provides\nanother explanation for the performance drop caused by back-translation. We\nalso apply our approach to both back- and forward-translation and find that\nmitigating the language coverage bias can improve the performance of both the\ntwo representative data augmentation methods and their tagged variants.\n"}
{"abstract": "  In this paper, a third-order compact gas-kinetic scheme (GKS) on unstructured\ntetrahedral mesh is constructed for the compressible Euler and Navier-Stokes\nsolutions. The time-dependent gas distribution function at a cell interface is\nused to calculate the fluxes for the updating the cell-averaged flow variables\nand to evaluate the time accurate cell-averaged flow variables as well for\nevolving the cell-averaged gradients of flow variables. With the accurate\nevolution model for both flow variables and their slopes, the quality of the\nscheme depends closely on the accuracy and reliability of the initial\nreconstruction of flow variables. The reconstruction scheme becomes more\nchallenge on tetrahedral mesh, where the conventional second-order unlimited\nleast-square reconstruction can make the scheme be linearly unstable when using\ncell-averaged conservative variables alone with von Neumann neighbors.\nBenefiting from the evolved cell-averaged slopes, on tetrahedral mesh the GKS\nis linearly stable from a compact third-order smooth reconstruction with a\nlarge CFL number. In order to further increase the robustness of the high-order\ncompact GKS for capturing discontinuous solution, a new two-step\nmulti-resolution weighted essentially non-oscillatory (WENO) reconstruction\nwill be proposed. The novelty of the reconstruction includes the following.\nFirstly, it releases the stability issue from a second-order compact\nreconstruction through the introduction of a pre-reconstruction step. Secondly,\nin the third-order non-linear reconstruction, only one more large stencil is\nadded beside those in the second-order one, which significantly simplifies the\nhigh-order reconstruction. The proposed third-order scheme shows good\nrobustness in high speed flow computation and favorable mesh adaptability in\ncases with complex geometry.\n"}
{"abstract": "  We propose a new prediction method for multivariate linear regression\nproblems where the number of features is less than the sample size but the\nnumber of outcomes is extremely large. Many popular procedures, such as\npenalized regression procedures, require parameter tuning that is\ncomputationally untenable in such large-scale problems. We take a different\napproach, motivated by ideas from simultaneous estimation problems, that\nperforms linear shrinkage on ordinary least squares parameter estimates. Our\napproach is extremely computationally efficient and tuning-free. We show that\nit can asymptotically outperform ordinary least squares without any structural\nassumptions on the true regression coefficients and illustrate its good\nperformance in simulations and an analysis of single-cell RNA-seq data.\n"}
{"abstract": "  We study the performance of low-density parity-check (LDPC) codes over finite\ninteger rings, over two channels that arise from the Lee metric. The first\nchannel is a discrete memory-less channel (DMC) matched to the Lee metric. The\nsecond channel adds to each codeword an error vector of constant Lee weight,\nwhere the error vector is picked uniformly at random from the set of vectors of\nconstant Lee weight. It is shown that the marginal conditional distribution of\nthe two channels coincides, in the limit of large blocklengths. The performance\nof selected LDPC code ensembles is analyzed by means of density evolution and\nfinite-length simulations, with belief propagation decoding and with a\nlow-complexity symbol message passing algorithm.\n"}
{"abstract": "  We present a new linearly implicit exponential integrator that preserves the\npolynomial first integrals or Lyapunov functions for the conservative and\ndissipative stiff equations, respectively. The method is tested by both\noscillated ordinary differential equations and partial differential equations,\ne.g., an averaged system in wind-induced oscillation, the Fermi-Pasta-Ulam\nsystems, and the polynomial pendulum oscillators. The numerical simulations\nconfirm the conservative properties of the proposed method and demonstrate its\ngood behavior in superior running speed when compared with fully implicit\nschemes for long-time simulations.\n"}
{"abstract": "  Annotation is a major hurdle in the semantic segmentation of microscopy\nimages and volumes due to its prerequisite expertise and effort. This work\nenables the training of semantic segmentation networks on images with only a\nsingle point for training per instance, an extreme case of weak supervision\nwhich drastically reduces the burden of annotation. Our approach has two key\naspects: (1) we construct a graph-theoretic soft-segmentation using individual\nseeds to be used within a regularizer during training and (2) we use an\nobjective function that enables learning from the constructed soft-labels. We\nachieve competitive results against the state-of-the-art in point-supervised\nsemantic segmentation on challenging datasets in digital pathology. Finally, we\nscale our methodology to point-supervised segmentation in 3D fluorescence\nmicroscopy volumes, obviating the need for arduous manual volumetric\ndelineation. Our code is freely available.\n"}
{"abstract": "  The Landauer principle states that any logically irreversible information\nprocessing must be accompanied by dissipation into the environment. In this\nstudy, we investigate the heat dissipation associated with finite-time\ninformation erasure and the effect of quantum coherence in such processes. By\nconsidering a scenario wherein information is encoded in an open quantum system\nwhose dynamics are described by the Markovian Lindblad equation, we show that\nthe dissipated heat is lower-bounded by the conventional Landauer cost, as well\nas a correction term inversely proportional to the operational time. To clarify\nthe relation between quantum coherence and dissipation, we derive a lower bound\nfor heat dissipation in terms of quantum coherence. This bound quantitatively\nimplies that the creation of quantum coherence in the energy eigenbasis during\nthe erasure process inevitably leads to additional heat costs. The obtained\nbounds hold for arbitrary operational time and control protocol. By following\nan optimal control theory, we numerically present an optimal protocol and\nillustrate our findings by using a single-qubit system.\n"}
{"abstract": "  The charged and neutral pion mass difference can be attributed to both the\nQED and QCD contributions. The current quark mass difference ($\\Delta m$) is\nthe source of the QCD contribution. Here, in a two flavour non-local NJL model,\nwe try to estimate the QCD contribution. Interestingly, we find that the\nstrength of the $U(1)_A$ symmetry breaking parameter $c$ plays a crucial role\nfor obtaining the pion mass difference while intertwined with the current quark\nmass difference. To obtain the QCD contribution for the pion mass difference we\ncan scan the parameter space in $\\{\\Delta m,\\; c\\}$ and by comparing this with\nthe existing results this parameter space is constrained. Further, using a\nfitted value of $c$ we can determine the allowed range for the current quark\nmass difference in the model.\n"}
{"abstract": "  We obtain lower norm bounds for multivariate functions in weighted Lebesgue\nspaces on unbounded domains in finite Euclidean space. These are identified by\nthe solution of a system of nonlinear integral equations, with linear\nconstrains on one-dimensional marginals. We recognize the importance of the\nSmirnov condition for the corresponding weights, which guarantees uniqueness of\nsolutions of the system. Sufficient conditions on weights are formulated so to\nsatisfy the Smirnov property, and counterexamples are constructed, when the\nproperty is violated.\n"}
{"abstract": "  We prove the two-scale transformation method which allows rigorous\nhomogenisation of problems defined on locally periodic domains by\ntransformation on periodic domains. The idea to consider periodic substitute\nproblems was originally proposed by M. A. Peter for the homogenisation on\nevolving microstructure and is applied in several works. However, only the\nhomogenisation of the periodic substitute problems was proven, whereas the\nmethod itself was just postulated (i.e. the equivalence to the homogenisation\nof the actual problem had to be assumed). In this work, we develop this idea\nfurther and formulate a rigorous two-scale convergence concept for microscopic\ntransformation to prove this method. Moreover, we show a new two-scale\ntransformation rule for gradients which allows to derive new limit problems\nthat are now transformationally independent.\n"}
{"abstract": "  Long gamma ray bursts (LGRBs) are associated to the collapse of a massive\nstar and the formation of a relativistic jet. As the jet propagates through the\nstar, it forms an extended, hot cocoon. The dynamical evolution of the\njet/cocoon system and its interaction with the environment has been studied\nextensively both analytically and numerically. On the other hand, the role\nplayed by the supernova (SN) explosion associated with LGRBs in determining the\noutcome of the system has been barely considered. In this paper, we discuss the\nlarge landscape of outcomes resulting from the interaction of the SN, jet and\ncocoon. We show that the outcome depends mainly on three timescales: the times\nfor the cocoon and supernova shock wave to break through the surface of the\nprogenitor star, and the time needed for the cocoon to engulf completely the\nprogenitor star. The delay between the launch of the SN shock moving through\nthe progenitor star and the jet can be related to these three timescales.\nDepending on the ordering of these time scales, the jet-cocoon might propagate\ninside the SN ejecta or the other way around, and the outcome for the\nproperties of the explosion would be different. We discuss the imprint of the\ncomplex interaction between the jet-cocoon and the supernova shock on the\nemergent thermal and non-thermal radiation.\n"}
{"abstract": "  We study some aspects of dynamical compactification scenario where\nstabilisation of extra dimensions occurs due to presence the Gauss-Bonnet term\nand non-zero spatial curvature. In the framework of the model under\nconsideration there exists two-stages scenario of evolution of a Universe: on\nthe first stage, the space evolves from a totally anisotropic state to the\nstate with 3-dimensional (corresponding to our \"real\", world) expanding and\n$D$-dimensional contracting isotropic subspaces; on the second stage, constant\ncurvature of extra dimensions begins to play role and provide compactification\nof extra dimensions. It is already known that such a scenario is realizable\nwhen constant curvature of extra dimensions is negative. Here we show that a\nrange of coupling constants for which exponential solutions with 3-dimensional\nexpanding and $D$-dimensional contracting isotropic subspaces are stable is\nlocated in a zone where compactification solutions with positively curved extra\nspace are unstable, so that two-stage scenario analogous to the one described\nabove is \\emph{not} realizable. Also we study \"nearly-Friedmann\", regime for\nthe case of arbitrary constant curvature of extra dimensions and describe new\nparametrization of the general solution for the model under consideration which\nprovide elegant way of describing areas of existence over parameters space.\n"}
{"abstract": "  Reinforcement learning synthesizes controllers without prior knowledge of the\nsystem. At each timestep, a reward is given. The controllers optimize the\ndiscounted sum of these rewards. Applying this class of algorithms requires\ndesigning a reward scheme, which is typically done manually. The designer must\nensure that their intent is accurately captured. This may not be trivial, and\nis prone to error. An alternative to this manual programming, akin to\nprogramming directly in assembly, is to specify the objective in a formal\nlanguage and have it \"compiled\" to a reward scheme. Mungojerrie\n(https://plv.colorado.edu/mungojerrie/) is a tool for testing reward schemes\nfor $\\omega$-regular objectives on finite models. The tool contains\nreinforcement learning algorithms and a probabilistic model checker.\nMungojerrie supports models specified in PRISM and $\\omega$-automata specified\nin HOA.\n"}
{"abstract": "  The ever-growing popularity and rapid improving of artificial intelligence\n(AI) have raised rethinking on the evolution of wireless networks. Mobile edge\ncomputing (MEC) provides a natural platform for AI applications since it\nprovides rich computation resources to train AI models, as well as low-latency\naccess to the data generated by mobile and Internet of Things devices. In this\npaper, we present an infrastructure to perform machine learning tasks at an MEC\nserver with the assistance of a reconfigurable intelligent surface (RIS). In\ncontrast to conventional communication systems where the principal criteria are\nto maximize the throughput, we aim at optimizing the learning performance.\nSpecifically, we minimize the maximum learning error of all users by jointly\noptimizing the beamforming vectors of the base station and the phase-shift\nmatrix of the RIS. An alternating optimization-based framework is proposed to\noptimize the two terms iteratively, where closed-form expressions of the\nbeamforming vectors are derived, and an alternating direction method of\nmultipliers (ADMM)-based algorithm is designed together with an error level\nsearching framework to effectively solve the nonconvex optimization problem of\nthe phase-shift matrix. Simulation results demonstrate significant gains of\ndeploying an RIS and validate the advantages of our proposed algorithms over\nvarious benchmarks.\n"}
{"abstract": "  Queueing-network models are developed for enhanced and next-generation 911\n(E911 and NG911) systems, which capture both their cyber-components\n(communications, data-processing) and physical-world-elements (call takers,\nvehicle dispatch). The models encompass both the call-processing and dispatch\nfunctions of 911, and can represent the interdependencies between multiple\nPSAPs enabled by NG911. An instantiation of the model for a future NG911 system\nfor Charlotte, North Carolina, is developed and used to assess performance\nmetrics. Representation of cyber-threats (e.g. Distributed Denial-of-Service\nattacks) within the queueing-network model is undertaken. Based on these\nrepresentations, the model is used for analysis of holistic threat impacts, as\na step toward risk and vulnerability assessment for future emergency response\nsystems.\n"}
{"abstract": "  We give an upper bound for the dimension of the bounded derived categories of\n$(m,n)$-Igusa-Todorov algebras which is a generalization of $n$-Igusa-Todorov\nalgebras, where $m,n$ are two nonnegative integers. As an applications, we get\na new upper bound for the dimension of bounded derived categories in terms of\nthe projective dimensions of certain of simple modules as well as radical layer\nlength of artin algebra $\\Lambda$.\n"}
{"abstract": "  In this paper we study approximations of functions of Sobolev spaces\n$W^2_{p,\\loc}(\\Omega)$, $\\Omega\\subset\\mathbb R^n$, by Lipschitz continuous\nfunctions. We prove that if $f\\in W^2_{p,\\loc}(\\Omega)$, $1\\leq p<\\infty$, then\nthere exists a sequence of closed sets $\\{A_k\\}_{k=1}^{\\infty},A_k\\subset\nA_{k+1}\\subset \\Omega$, such that the restrictions $f \\vert_{A_k}$ are\nLipschitz continuous functions and $\\cp_p\\left(S\\right)=0$,\n$S=\\Omega\\setminus\\bigcup_{k=1}^{\\infty}A_k$. Using these approximations we\nprove the change of variables formula in the Lebesgue integral for mappings of\nSobolev spaces $W^2_{p,\\loc}(\\Omega;\\mathbb R^n)$ with the Luzin\ncapacity-measure $N$-property.\n"}
{"abstract": "  We present a new conceptual model of the Earth's glacial-interglacial cycles,\none leading to governing equations for which the vector field has a hyperplane\nof discontinuities. This work extends the classic Budyko- and Sellers-type\nconceptual energy balance models of temperature-albedo feedback by removing the\nstandard assumption of planetary symmetry about the equator. The dynamics of\nseparate Northern and Southern Hemisphere ice caps are coupled to an equation\nrepresenting the annual global mean surface temperature. The system has a\ndiscontinuous switching mechanism based on mass balance principles for the\nNorthern Hemisphere ice sheet. We show the associated Filippov system admits a\nunique nonsmooth and attracting limit cycle that represents the cycling between\nglacial and interglacial states. Due to the vastly different time scales\ninvolved, the model presents a nonsmooth geometric perturbation problem, for\nwhich we use ad hoc mathematical techniques to produce the periodic orbit. We\nfind climatic changes in the Northern Hemisphere drive synchronous changes in\nthe Southern Hemisphere, as is observed for the Earth on orbital time scales.\n"}
{"abstract": "  Marginal-likelihood based model-selection, even though promising, is rarely\nused in deep learning due to estimation difficulties. Instead, most approaches\nrely on validation data, which may not be readily available. In this work, we\npresent a scalable marginal-likelihood estimation method to select both\nhyperparameters and network architectures, based on the training data alone.\nSome hyperparameters can be estimated online during training, simplifying the\nprocedure. Our marginal-likelihood estimate is based on Laplace's method and\nGauss-Newton approximations to the Hessian, and it outperforms cross-validation\nand manual-tuning on standard regression and image classification datasets,\nespecially in terms of calibration and out-of-distribution detection. Our work\nshows that marginal likelihoods can improve generalization and be useful when\nvalidation data is unavailable (e.g., in nonstationary settings).\n"}
{"abstract": "  We prove that there exist Delone sets in $\\mathbb{R}^d$, $d \\geq 2$, which\ncannot be mapped onto the standard lattice $\\mathbb{Z}^d$ by Lipschitz\nbijections. The impossibility of the Lipschitz-rectifiability crucially uses\nideas of Lipschitz regular maps recently introduced by M. Dymond, V. Kalu\\v{z}a\nand E. Kopeck\\'a.\n"}
{"abstract": "  In this work, we explore the dynamics of entanglement of an isolated quantum\nsystem consisting of two time-dependent, coupled harmonic oscillators. Through\nthe use of a numerical method that relies on the estimation of the system's\nWigner representation by a specific Gaussian function, we investigate the time\nevolution of the entanglement entropy after an instant quench in the inherent\nparameters of the system. Besides, from the comparison of the results obtained\nfrom the analytical expression for the time-dependent von Neumann entropy with\nthe numerically computed entropy data, the effectiveness of the numerical\nmethod is tested for a variety of angular frequency combinations. Also, we\nanalyze how the entropy of entanglement change as a function of time.\n"}
{"abstract": "  Energy providers are moving to the smart meter era, encouraging consumers to\ninstall, free of charge, these devices in their homes, automating consumption\nreadings submission and making consumers life easier. However, the increased\ndeployment of such smart devices brings a lot of security and privacy risks. In\norder to overcome such risks, Intrusion Detection Systems are presented as\npertinent tools that can provide network-level protection for smart devices\ndeployed in home environments. In this context, this paper is exploring the\nproblems of Advanced Metering Infrastructures (AMI) and proposing a novel\nMachine Learning (ML) Intrusion Prevention System (IPS) to get optimal\ndecisions based on a variety of factors and graphical security models able to\ntackle zero-day attacks.\n"}
{"abstract": "  The development of neural networks has greatly improved the performance in\nvarious computer vision tasks. In the filed of image denoising, convolutional\nneural network based methods such as DnCNN break through the limits of\nclassical methods, achieving better quantitative results. However, the\nepistemic uncertainty existing in neural networks limits further improvements\nin their performance over denoising tasks. Therefore, we develop and study\ndifferent solutions to minimize uncertainty and further improve the removal of\nnoise. From the perspective of ensemble learning, we implement manipulations to\nnoisy images from the point of view of spatial and frequency domains and then\ndenoise them using pre-trained denoising networks. We propose a fusion model\nconsisting of two attention modules, which focus on assigning the proper\nweights to pixels and channels. The experimental results show that our model\nachieves better performance on top of the baseline of regular pre-trained\ndenoising networks.\n"}
{"abstract": "  Auger-like exciton-exciton annihilation (EEA) is considered the key\nfundamental limitation to quantum yield in devices based on excitons in\ntwo-dimensional (2d) materials. Since it is challenging to experimentally\ndisentangle EEA from competing processes, guidance of a quantitative theory is\nhighly desirable. The very nature of EEA requires a material-realistic\ndescription that is not available to date. We present a many-body theory of EEA\nbased on first-principle band structures and Coulomb interaction matrix\nelements that goes beyond an effective bosonic picture. Applying our theory to\nmonolayer MoS$_2$ encapsulated in hexagonal BN, we obtain an EEA coefficient in\nthe order of $10^{-3}$ cm$^{2}$s$^{-1}$ at room temperature, suggesting that\ncarrier losses are often dominated by other processes, such as defect-assisted\nscattering. Our studies open a perspective to quantify the efficiency of\nintrinsic EEA processes in various 2d materials in the focus of modern\nmaterials research.\n"}
{"abstract": "  We present the results of computer simulations on a class of percolative\nsystems, called protected percolation, that violates the Harris criterion. The\nHarris criterion states whether the critical behavior at a phase transition\nfrom a disordered state to an ordered state will be altered by impurities. We\nhave incorporated impurities into our simulations to test whether the critical\nexponents for protected percolation are altered by impurities. We find that the\ncritical exponents for three-dimensional protected percolation simulations\nindeed change with impurities in the form of missing sites and immortal sites.\nOn the other hand, the critical exponents for both standard percolation and\nprotected percolation in two dimensions are stable against impurities.\n"}
{"abstract": "  Creating presentation materials requires complex multimodal reasoning skills\nto summarize key concepts and arrange them in a logical and visually pleasing\nmanner. Can machines learn to emulate this laborious process? We present a\nnovel task and approach for document-to-slide generation. Solving this involves\ndocument summarization, image and text retrieval, slide structure and layout\nprediction to arrange key elements in a form suitable for presentation. We\npropose a hierarchical sequence-to-sequence approach to tackle our task in an\nend-to-end manner. Our approach exploits the inherent structures within\ndocuments and slides and incorporates paraphrasing and layout prediction\nmodules to generate slides. To help accelerate research in this domain, we\nrelease a dataset about 6K paired documents and slide decks used in our\nexperiments. We show that our approach outperforms strong baselines and\nproduces slides with rich content and aligned imagery.\n"}
{"abstract": "  The interplay between activity and elasticity often found in active and\nliving systems triggers a plethora of autonomous behaviors ranging from\nself-assembly and collective motion to actuation. Amongst these, spontaneous\nself-oscillations of mechanical structures is perhaps the simplest and most\nwide-spread type of non-equilibrium phenomenon. Yet, we lack experimental model\nsystems to investigate the various dynamical phenomena that may appear. Here,\nwe report self-oscillation and synchronization transitions in a\ncentimeter-sized model system for one-dimensional elasto-active structures. By\ncombining precision-desktop experiments of elastically coupled self-propelled\nparticles with numerical simulations and analytical perturbative theory, we\ndemonstrate that the dynamics of single chain follows a Hopf bifurcation. We\nshow that this instability is controlled by a single non-dimensional\nelasto-active number that quantifies the interplay between activity and\nelasticity. Finally, we demonstrate that pairs of coupled elasto-active chains\ncan undergo a synchronization transition: the oscillations phases of both\nchains lock when the coupling link is sufficiently stiff. Beyond the canonical\ncase considered here, we anticipate our work to open avenues for the\nunderstanding and design of the self-organisation and response of active\nartificial and biological solids, e.g. in higher dimensions and for more\nintricate geometries.\n"}
{"abstract": "  Recent studies on Question Answering (QA) and Conversational QA (ConvQA)\nemphasize the role of retrieval: a system first retrieves evidence from a large\ncollection and then extracts answers. This open-retrieval ConvQA setting\ntypically assumes that each question is answerable by a single span of text\nwithin a particular passage (a span answer). The supervision signal is thus\nderived from whether or not the system can recover an exact match of this\nground-truth answer span from the retrieved passages. This method is referred\nto as span-match weak supervision. However, information-seeking conversations\nare challenging for this span-match method since long answers, especially\nfreeform answers, are not necessarily strict spans of any passage. Therefore,\nwe introduce a learned weak supervision approach that can identify a\nparaphrased span of the known answer in a passage. Our experiments on QuAC and\nCoQA datasets show that the span-match weak supervisor can only handle\nconversations with span answers, and has less satisfactory results for freeform\nanswers generated by people. Our method is more flexible as it can handle both\nspan answers and freeform answers. Moreover, our method can be more powerful\nwhen combined with the span-match method which shows it is complementary to the\nspan-match method. We also conduct in-depth analyses to show more insights on\nopen-retrieval ConvQA under a weak supervision setting.\n"}
{"abstract": "  Browder (1960) proved that for every continuous function $F : X \\times Y \\to\nY$, where $X$ is the unit interval and $Y$ is a nonempty, convex, and compact\nsubset of $\\dR^n$, the set of fixed points of $F$, defined by $C_F := \\{ (x,y)\n\\in X \\times Y \\colon F(x,y)=y\\}$ has a connected component whose projection to\nthe first coordinate is $X$. We extend this result to the case where $X$ is a\nconnected and compact Hausdorff space.\n"}
{"abstract": "  Data-efficiency and generalization are key challenges in deep learning and\ndeep reinforcement learning as many models are trained on large-scale,\ndomain-specific, and expensive-to-label datasets. Self-supervised models\ntrained on large-scale uncurated datasets have shown successful transfer to\ndiverse settings. We investigate using pretrained image representations and\nspatio-temporal attention for state representation learning in Atari. We also\nexplore fine-tuning pretrained representations with self-supervised techniques,\ni.e., contrastive predictive coding, spatio-temporal contrastive learning, and\naugmentations. Our results show that pretrained representations are at par with\nstate-of-the-art self-supervised methods trained on domain-specific data.\nPretrained representations, thus, yield data and compute-efficient state\nrepresentations. https://github.com/PAL-ML/PEARL_v1\n"}
{"abstract": "  Recently, the importance of the electronic many-body effect in the dark\nmatter (DM) detection has been recognized and a coherent formulation of the\nDM-electron scattering in terms of the dielectric response of the target\nmaterial has been well established in literatures. In this paper, we put\nrelevant formulas into practical density functional theory (DFT) estimation of\nthe excitation event rates for the diamond and silicon semiconductor targets.\nMoreover, we compare the event rates calculated from the energy loss functions\nwith and without the local field effects. For a consistency check of this\nnumerical method, we also compare the differential spectrum and detection reach\nof the silicon with those computed with the $\\mathtt{GPAW}$ code. It turns out\nthat this DFT approach is quite consistent and robust. As an interesting\nextension, we also investigate the in-medium effect on the detection of the\nsolar-reflected DM flux in silicon-based detectors, where the screening effect\nis found to be also remarkable in the optically thick regime, and to turn\ninsignificant in the optically thin regime, depending on the energies of the\nreflected DM particles.\n"}
{"abstract": "  As an application of the simultaneous and continuous measurement of\nnoncommutative observables formulated in our previous paper [C. Jiang and G.\nWatanabe, Phys. Rev. A 102, 062216 (2020)], we propose a scheme to generate the\npure ideal quadrature squeezed state in an one-dimensional harmonic oscillator\nsystem by the feedback control based on such type of measurement of\nnoncommutative quadrature observables. We find that, by appropriately setting\nthe strengths of the measurement and the feedback control, the pure ideal\nquadrature squeezed state with arbitrary squeezedness can be produced. This is\nin contrast to the scheme based on the single-observable measurement and the\nfeedback control, where only nonideal squeezed state with squeezing of the\nmeasured quadrature are produced.\n"}
{"abstract": "  Recent work has shown that Neural Ordinary Differential Equations (ODEs) can\nserve as generative models of images using the perspective of Continuous\nNormalizing Flows (CNFs). Such models offer exact likelihood calculation, and\ninvertible generation/density estimation. In this work we introduce a\nMulti-Resolution variant of such models (MRCNF), by characterizing the\nconditional distribution over the additional information required to generate a\nfine image that is consistent with the coarse image. We introduce a\ntransformation between resolutions that allows for no change in the log\nlikelihood. We show that this approach yields comparable likelihood values for\nvarious image datasets, with improved performance at higher resolutions, with\nfewer parameters, using only 1 GPU. Further, we examine the out-of-distribution\nproperties of (Multi-Resolution) Continuous Normalizing Flows, and find that\nthey are similar to those of other likelihood-based generative models.\n"}
{"abstract": "  Young exoplanets can offer insight into the evolution of planetary\natmospheres, compositions, and architectures. We present the discovery of the\nyoung planetary system TOI 451 (TIC 257605131, Gaia DR2 4844691297067063424).\nTOI 451 is a member of the 120-Myr-old Pisces--Eridanus stream (Psc--Eri). We\nconfirm membership in the stream with its kinematics, its lithium abundance,\nand the rotation and UV excesses of both TOI 451 and its wide binary companion,\nTOI 451 B (itself likely an M dwarf binary). We identified three candidate\nplanets transiting in the TESS data and followed up the signals with photometry\nfrom Spitzer and ground-based telescopes. The system comprises three validated\nplanets at periods of 1.9, 9.2 and 16 days, with radii of 1.9, 3.1, and 4.1\nEarth radii, respectively. The host star is near-solar mass with V=11.0 and\nH=9.3 and displays an infrared excess indicative of a debris disk. The planets\noffer excellent prospects for transmission spectroscopy with HST and JWST,\nproviding the opportunity to study planetary atmospheres that may still be in\nthe process of evolving.\n"}
{"abstract": "  We explore changes in the adiabatic low-order g-mode pulsation periods of\n0.526, 0.560, and 0.729 M$_\\odot$ carbon-oxygen white dwarf models with\nhelium-dominated envelopes due to the presence, absence, and enhancement of\n$^{22}$Ne in the interior. The observed g-mode pulsation periods of such white\ndwarfs are typically given to 6$-$7 significant figures of precision. Usually\nwhite dwarf models without $^{22}$Ne are fit to the observed periods and other\nproperties. The root-mean-square residuals to the $\\simeq$ 150$-$400 s\nlow-order g-mode periods are typically in the range of $\\sigma_{rms}$\n$\\lesssim$ 0.3 s, for a fit precision of $\\sigma_{rms}/ P$ $\\lesssim$ 0.3 %. We\nfind average relative period shifts of $\\Delta P/P$ $\\simeq$ $\\pm$ 0.5 % for\nthe low-order dipole and quadrupole g-mode pulsations within the observed\neffective temperature window, with the range of $\\Delta P/P$ depending on the\nspecific g-mode, abundance of $^{22}$Ne, effective temperature, and mass of the\nwhite dwarf model. This finding suggests a systematic offset may be present in\nthe fitting process of specific white dwarfs when $^{22}$Ne is absent. As part\nof the fitting processes involves adjusting the composition profiles of a white\ndwarf model, our study on the impact of $^{22}$Ne can provide new inferences on\nthe derived interior mass fraction profiles. We encourage routinely including\n$^{22}$Ne mass fraction profiles, informed by stellar evolution models, to\nfuture generations of white dwarf model fitting processes.\n"}
{"abstract": "  A simple geometrical model with event-by-event fluctuations is suggested to\nstudy elliptical and triangular eccentricities in the initial state of\nrelativistic heavy-ion collisions. This model describes rather well the ALICE\nand ATLAS data for Pb+Pb collisions at center-of-mass energy $\\sqrt{s_{NN}} =\n5.02$~TeV per nucleon pair, assuming that the second, $v_2$, and third, $v_3$,\nharmonics of the anisotropic flow are simply linearly proportional to the\neccentricities $\\varepsilon_2$ and $\\varepsilon_3$, respectively. We show that\nthe eccentricity $\\varepsilon_3$ has a pure fluctuation origin and is\nsubstantially dependent on the size of the overlap area only, while the\neccentricity $\\varepsilon_2$ is mainly related to the average collision\ngeometry. Elliptic flow, therefore, is weakly dependent on the event-by-event\nfluctuations everywhere except of the very central collisions 0--2%, whereas\ntriangular flow is mostly determined by the fluctuations. The scaling\ndependence of the magnitude of the flow harmonics on atomic number, $v_n\n\\propto A^{-1/3}$, is predicted for this centrality interval.\n"}
{"abstract": "  Let $n,k$ be positive integers such that $n\\geq 3$, $k < \\frac {n}{2} $. Let\n$q$ be a power of a prime $p$ and $\\mathbb{F}_q$ be a finite field of order\n$q$. Let $V(q,n)$ be a vector space of dimension $n$ over $\\mathbb{F}_q$. We\ndefine the graph $S(q,n,k)$ as a graph with the vertex set $V=V_k \\cup\nV_{k+1}$, where $V_k$ and $V_{k+1}$ are the family of subspaces in $V(q,n)$ of\ndimension $k$ and $k+1$ respectively, in which two vertices $v$ and $w$ are\nadjacent whenever $v$ is a subspace of $w$ or $w$ is a subspace of $v$.\n  It is clear that the graph $S(q,n,k)$ is a bipartite graph. In this paper, we\nstudy some properties of this graph. In particular, we determine the spectrum\nof the graph $S(q,n,k)$.\n"}
{"abstract": "  3D point cloud semantic segmentation is a challenging topic in the computer\nvision field. Most of the existing methods in literature require a large amount\nof fully labeled training data, but it is extremely time-consuming to obtain\nthese training data by manually labeling massive point clouds. Addressing this\nproblem, we propose a superpoint-guided semi-supervised segmentation network\nfor 3D point clouds, which jointly utilizes a small portion of labeled scene\npoint clouds and a large number of unlabeled point clouds for network training.\nThe proposed network is iteratively updated with its predicted pseudo labels,\nwhere a superpoint generation module is introduced for extracting superpoints\nfrom 3D point clouds, and a pseudo-label optimization module is explored for\nautomatically assigning pseudo labels to the unlabeled points under the\nconstraint of the extracted superpoints. Additionally, there are some 3D points\nwithout pseudo-label supervision. We propose an edge prediction module to\nconstrain features of edge points. A superpoint feature aggregation module and\na superpoint feature consistency loss function are introduced to smooth\nsuperpoint features. Extensive experimental results on two 3D public datasets\ndemonstrate that our method can achieve better performance than several\nstate-of-the-art point cloud segmentation networks and several popular\nsemi-supervised segmentation methods with few labeled scenes.\n"}
{"abstract": "  For a subset $A$ of $\\{1,2,\\ldots,N\\}^2$ of size $\\alpha N^2$ we show\nexistence of $(m,n)\\neq(0,0)$ such that the set $A$ contains at least\n$(\\alpha^3 - o(1))N^2$ triples of points of the form $(a,b)$, $(a+m,b+n)$,\n$(a-n,b+m)$. This answers a question by Ackelsberg, Bergelson, and Best from\narXiv:2101.02811. The same approach also establishes the corresponding result\nfor compact abelian groups. Furthermore, for a finite field $\\mathbb{F}_q$ we\ncomment on exponential smallness of subsets of $(\\mathbb{F}_q^n)^2$ that avoid\nthe aforementioned configuration. The proofs are minor modifications of the\nexisting proofs regarding three-term arithmetic progressions.\n"}
{"abstract": "  We present analytic cooling and diffusion rates for a simplified model of\ncoherent electron cooling (CEC), based on a proton energy kick at each turn.\nThis model also allows to estimate analytically the rms value of electron beam\ndensity fluctuations in the \"kicker\" section. Having such analytic expressions\nshould allow for better understanding of the CEC mechanism, and for a quicker\nanalysis and optimization of main system parameters. Our analysis is applicable\nto any CEC amplification mechanism, as long as the wake (kick) function is\navailable.\n"}
{"abstract": "  In this short paper, we argue that the chiral central charge $c_-$ of a\n(2+1)d topological ordered state is sometimes strongly constrained by 't Hooft\nanomaly of anti-unitary global symmetry. For example, if a (2+1)d fermionic\nTQFT has a time reversal anomaly with $T^2=(-1)^F$ labeled as\n$\\nu\\in\\mathbb{Z}_{16}$, the TQFT must have $c_-=1/4$ mod $1/2$ for odd $\\nu$,\nwhile $c_-=0$ mod $1/2$ for even $\\nu$. This generalizes the fact that the\nbosonic TQFT with $T$ anomaly in a particular class must carry $c_-=4$ mod $8$\nto fermionic cases. We also study such a constraint for fermionic TQFT with\n$U(1)\\times CT$ symmetry, which is regarded as a gapped surface of the\ntopological superconductor in class AIII.\n"}
{"abstract": "  The application of the Cauchy distribution has often been discussed as a\npotential model of the financial markets. In particular the way in which single\nextreme, or \"Black Swan\", events can impact long term historical moments, is\noften cited. In this article we show how one can construct Martingale\nprocesses, which have marginal distributions that tend to the Cauchy\ndistribution in the large volatility limit. This provides financial\njustification to approaches discussed by other authors, and highlights an\nexample of how quantum probability can be used to construct non-Gaussian\nMartingales. We go on to illustrate links with hyperbolic diffusion, and\ndiscuss the insight this provides.\n"}
{"abstract": "  In the standard model of charge density wave (CDW) transitions, the\ndisplacement along a single phonon mode lowers the total electronic energy by\ncreating a gap at the Fermi level, making the CDW a metal--insulator\ntransition. Here, using scanning tunneling microscopy and spectroscopy and ab\ninitio calculations, we show that VS$_2$ realizes a CDW which stands out of\nthis standard model. There is a full CDW gap residing in the unoccupied states\nof monolayer VS$_2$. At the Fermi level, the CDW induces a topological\nmetal-metal (Lifshitz) transition. Non-linear coupling of transverse and\nlongitudinal phonons is essential for the formation of the CDW and the full gap\nabove the Fermi level. Additionally, x-ray magnetic circular dichroism reveals\nthe absence of net magnetization in this phase, pointing to coexisting charge\nand spin density waves in the ground state.\n"}
{"abstract": "  Recent gravitational-wave observations have begun to constrain the internal\nphysics of neutron stars. However, current detection searches for neutron star\nsystems assume that potential neutron stars are low-mass black holes, ignoring\nany affect on the gravitational-wave signal due to the internal neutron-star\nphysics. We wish to create a template bank of binary neutron star waveforms\nincluding the effect of tidal deformability. However, we find that the Fisher\nmatrix, which is commonly used to approximate match calculations when placing\ntemplate banks, is unsuitable to predict the match between two binary neutron\nstar waveforms. We find that the Fisher matrix can predict errors on the\nmismatch that are larger than $100\\%$ when attempting to identify waveforms\nwith a match of $0.97$. We explore the regime in which the Fisher matrix cannot\nbe trusted and examine why it breaks down. We demonstrate that including\nhigher-order terms in the Taylor series expansion of the match can reliably\ncompute matches for these examples, but that it is prohibitively\ncomputationally expensive to do so. Finally, we demonstrate that stochastic\nplacement can still be used to construct a template bank of tidally deformed\nneutron-star waveforms.\n"}
{"abstract": "  This paper targets the problem of procedural multimodal machine comprehension\n(M3C). This task requires an AI to comprehend given steps of multimodal\ninstructions and then answer questions. Compared to vanilla machine\ncomprehension tasks where an AI is required only to understand a textual input,\nprocedural M3C is more challenging as the AI needs to comprehend both the\ntemporal and causal factors along with multimodal inputs. Recently Yagcioglu et\nal. [35] introduced RecipeQA dataset to evaluate M3C. Our first contribution is\nthe introduction of two new M3C datasets- WoodworkQA and DecorationQA with 16K\nand 10K instructional procedures, respectively. We then evaluate M3C using a\ntextual cloze style question-answering task and highlight an inherent bias in\nthe question answer generation method from [35] that enables a naive baseline\nto cheat by learning from only answer choices. This naive baseline performs\nsimilar to a popular method used in question answering- Impatient Reader [6]\nthat uses attention over both the context and the query. We hypothesized that\nthis naturally occurring bias present in the dataset affects even the best\nperforming model. We verify our proposed hypothesis and propose an algorithm\ncapable of modifying the given dataset to remove the bias elements. Finally, we\nreport our performance on the debiased dataset with several strong baselines.\nWe observe that the performance of all methods falls by a margin of 8% - 16%\nafter correcting for the bias. We hope these datasets and the analysis will\nprovide valuable benchmarks and encourage further research in this area.\n"}
{"abstract": "  Let $M$ be a cancellative monoid. It is known~\\cite{Ta54} that if $M$ is left\namenable then the monoid ring $K[M]$ satisfies Ore condition, that is, there\nexist nontrivial common right multiples for the elements of this ring.\nIn~\\cite{Don10} Donnelly shows that a partial converse to this statement is\ntrue. Namely, if the monoid $\\mathbb Z^{+}[M]$ of all elements of $\\mathbb\nZ[M]$ with positive coefficients has nonzero common right multiples, then $M$\nis left amenable. He asks whether the converse is true for this particular\nstatement.\n  We show that the converse is false even for the case of groups. If $M$ is a\nfree metabelian group, then $M$ is amenable but the Ore condition fails for\n$\\mathbb Z^{+}[M]$. Besides, we study the case of the monoid $M$ of positive\nelements of R.\\,Thompson's group $F$. The amenability problem for it is a\nfamous open question. It is equivalent to left amenability of the monoid $M$.\nWe show that for this case the monoid $\\mathbb Z^{+}[M]$ does not satisfy Ore\ncondition. That is, even if $F$ is amenable, this cannot be shown using the\nabove sufficient condition.\n"}
{"abstract": "  In this paper we present an extensive analysis of the GW190521 gravitational\nwave event with the current (fourth) generation of phenomenological waveform\nmodels for binary black hole coalescences. GW190521 stands out from other\nevents since only a few wave cycles are observable. This leads to a number of\nchallenges, one being that such short signals are prone to not resolve\napproximate waveform degeneracies, which may result in multi-modal posterior\ndistributions. The family of waveform models we use includes a new fast\ntime-domain model IMRPhenomTPHM, which allows us extensive tests of different\npriors and robustness with respect to variations in the waveform model,\nincluding the content of spherical harmonic modes. We clarify some issues\nraised in a recent paper [Nitz&Capano], associated with possible support for a\nhigh-mass ratio source, but confirm their finding of a multi-modal posterior\ndistribution, albeit with important differences in the statistical significance\nof the peaks. In particular, we find that the support for both masses being\noutside the PISN mass-gap, and the support for an intermediate mass ratio\nbinary are drastically reduced with respect to what Nitz&Capano found. We also\nprovide updated probabilities for associating GW190521 to the potential\nelectromagnetic counterpart from ZTF.\n"}
{"abstract": "  An increasing number of models require the control of the spectral norm of\nconvolutional layers of a neural network. While there is an abundance of\nmethods for estimating and enforcing upper bounds on those during training,\nthey are typically costly in either memory or time. In this work, we introduce\na very simple method for spectral normalization of depthwise separable\nconvolutions, which introduces negligible computational and memory overhead. We\ndemonstrate the effectiveness of our method on image classification tasks using\nstandard architectures like MobileNetV2.\n"}
{"abstract": "  We consider performing phase estimation under the following conditions: we\nare given only one copy of the input state, the input state does not have to be\nan eigenstate of the unitary, and the state must not be measured. Most quantum\nestimation algorithms make assumptions that make them unsuitable for this\n'coherent' setting, leaving only the textbook approach. We present novel\nalgorithms for phase, energy, and amplitude estimation that are both\nconceptually and computationally simpler than the textbook method, featuring\nboth a smaller query complexity and ancilla footprint. They do not require a\nquantum Fourier transform, and they do not require a quantum sorting network to\ncompute the median of several estimates. Instead, they use block-encoding\ntechniques to compute the estimate one bit at a time, performing all\namplification via singular value transformation. These improved subroutines\naccelerate the performance of quantum Metropolis sampling and quantum Bayesian\ninference.\n"}
{"abstract": "  We propose NeuMIP, a neural method for representing and rendering a variety\nof material appearances at different scales. Classical prefiltering\n(mipmapping) methods work well on simple material properties such as diffuse\ncolor, but fail to generalize to normals, self-shadowing, fibers or more\ncomplex microstructures and reflectances. In this work, we generalize\ntraditional mipmap pyramids to pyramids of neural textures, combined with a\nfully connected network. We also introduce neural offsets, a novel method which\nallows rendering materials with intricate parallax effects without any\ntessellation. This generalizes classical parallax mapping, but is trained\nwithout supervision by any explicit heightfield. Neural materials within our\nsystem support a 7-dimensional query, including position, incoming and outgoing\ndirection, and the desired filter kernel size. The materials have small storage\n(on the order of standard mipmapping except with more texture channels), and\ncan be integrated within common Monte-Carlo path tracing systems. We\ndemonstrate our method on a variety of materials, resulting in complex\nappearance across levels of detail, with accurate parallax, self-shadowing, and\nother effects.\n"}
{"abstract": "  Plant leaf diseases pose a significant danger to food security and they cause\ndepletion in quality and volume of production. Therefore accurate and timely\ndetection of leaf disease is very important to check the loss of the crops and\nmeet the growing food demand of the people. Conventional techniques depend on\nlab investigation and human skills which are generally costly and inaccessible.\nRecently, Deep Neural Networks have been exceptionally fruitful in image\nclassification. In this research paper, plant leaf disease detection employing\ntransfer learning is explored in the JPEG compressed domain. Here, the JPEG\ncompressed stream consisting of DCT coefficients is, directly fed into the\nNeural Network to improve the efficiency of classification. The experimental\nresults on JPEG compressed leaf dataset demonstrate the efficacy of the\nproposed model.\n"}
{"abstract": "  We provide an extension to previous analysis of the localised beading\ninstability of soft slender tubes under surface tension and axial stretching.\nThe primary questions pondered here are: under what loading conditions, if any,\ncan bifurcation into circumferential buckling modes occur, and do such\nsolutions dominate localisation and periodic axial modes? Three distinct\nboundary conditions are considered; in case 1 the tube's curved surfaces are\ntraction free and under surface tension, whilst in cases 2 and 3 the inner and\nouter surfaces (respectively) are fixed to prevent radial displacement and\nsurface tension. A linear bifurcation analysis is conducted to determine\nnumerically the existence of circumferential mode solutions. In case 1 we focus\non the tensile stress regime given the preference of slender compressed tubes\ntowards Euler buckling over axial wrinkling. We show that tubes under several\nloading paths are highly sensitive to circumferential modes; in contrast,\nlocalised and periodic axial modes are absent, suggesting that the\ncircumferential buckling is dominant by default. In case 2, circumferential\nmode solutions are associated with negative surface tension values and thus are\nphysically implausible. Circumferential buckling solutions are shown to exist\nin case 3 for tensile and compressive axial loads, and we demonstrate for\nmultiple loading scenarios their dominance over localisation and periodic axial\nmodes within specific parameter regimes.\n"}
{"abstract": "  We review the hypothetical interactions predicted beyond the Standard Model\nwhich could be constrained by using the results of tabletop laboratory\nexperiments. These interactions are described by the power-type potentials with\ndifferent powers, Yukawa potential, other spin-independent potentials, and by\nthe spin-dependent potentials of different kinds. In all these cases the\ncurrent constraints on respective hypothetical interactions are considered\nwhich follow from the Casimir effect and some other tabletop physics. The\nexotic particles and constraints on them are discussed in the context of\nproblems of the quantum vacuum, dark energy, and the cosmological constant.\n"}
{"abstract": "  Despite the success of reinforcement learning methods, they have yet to have\ntheir breakthrough moment when applied to a broad range of robotic manipulation\ntasks. This is partly due to the fact that reinforcement learning algorithms\nare notoriously difficult and time consuming to train, which is exacerbated\nwhen training from images rather than full-state inputs. As humans perform\nmanipulation tasks, our eyes closely monitor every step of the process with our\ngaze focusing sequentially on the objects being manipulated. With this in mind,\nwe present our Attention-driven Robotic Manipulation (ARM) algorithm, which is\na general manipulation algorithm that can be applied to a range of\nsparse-rewarded tasks, given only a small number of demonstrations. ARM splits\nthe complex task of manipulation into a 3 stage pipeline: (1) a Q-attention\nagent extracts interesting pixel locations from RGB and point cloud inputs, (2)\na next-best pose agent that accepts crops from the Q-attention agent and\noutputs poses, and (3) a control agent that takes the goal pose and outputs\njoint actions. We show that current learning algorithms fail on a range of\nRLBench tasks, whilst ARM is successful.\n"}
{"abstract": "  Motivated by the recently observed failure of the kinetic theory for the bulk\nviscosity, we in turn revisit the shear viscosity and the thermal conductivity\nof two-component fermions with a zero-range interaction both in two and three\ndimensions. In particular, we show that their Kubo formula evaluated exactly in\nthe high-temperature limit to the lowest order in fugacity is reduced to the\nlinearized Boltzmann equation. Previously, such a microscopic derivation of the\nlatter was achieved only incompletely corresponding to the relaxation-time\napproximation. Here, we complete it by resuming all contributions that are\nnaively higher orders in fugacity but become comparable in the zero-frequency\nlimit due to the pinch singularity, leading to a self-consistent equation for a\nvertex function identical to the linearized Boltzmann equation. We then compute\nthe shear viscosity and the thermal conductivity in the high-temperature limit\nfor an arbitrary scattering length and find that the Prandtl number exhibits a\nnonmonotonic behavior slightly below the constant value in the relaxation-time\napproximation.\n"}
{"abstract": "  The generalized gamma convolutions class of distributions appeared in\nThorin's work while looking for the infinite divisibility of the log-Normal and\nPareto distributions. Although these distributions have been extensively\nstudied in the univariate case, the multivariate case and the dependence\nstructures that can arise from it have received little interest in the\nliterature. Furthermore, only one projection procedure for the univariate case\nwas recently constructed, and no estimation procedures are available. By\nexpanding the densities of multivariate generalized gamma convolutions into a\ntensorized Laguerre basis, we bridge the gap and provide performant estimation\nprocedures for both the univariate and multivariate cases. We provide some\ninsights about performance of these procedures, and a convergent series for the\ndensity of multivariate gamma convolutions, which is shown to be more stable\nthan Moschopoulos's and Mathai's univariate series. We furthermore discuss some\nexamples.\n"}
{"abstract": "  Non-invasive cortical neural interfaces have only achieved modest performance\nin cortical decoding of limb movements and their forces, compared to invasive\nbrain-computer interfaces (BCIs). While non-invasive methodologies are safer,\ncheaper and vastly more accessible technologies, signals suffer from either\npoor resolution in the space domain (EEG) or the temporal domain (BOLD signal\nof functional Near Infrared Spectroscopy, fNIRS). The non-invasive BCI decoding\nof bimanual force generation and the continuous force signal has not been\nrealised before and so we introduce an isometric grip force tracking task to\nevaluate the decoding. We find that combining EEG and fNIRS using deep neural\nnetworks works better than linear models to decode continuous grip force\nmodulations produced by the left and the right hand. Our multi-modal deep\nlearning decoder achieves 55.2 FVAF[%] in force reconstruction and improves the\ndecoding performance by at least 15% over each individual modality. Our results\nshow a way to achieve continuous hand force decoding using cortical signals\nobtained with non-invasive mobile brain imaging has immediate impact for\nrehabilitation, restoration and consumer applications.\n"}
{"abstract": "  Quantum mass function has been applied in lots of fields because of its\nefficiency and validity of managing uncertainties in the form of quantum which\ncan be regarded as an extension of classical Dempster-Shafer (D-S) evidence\ntheory. However, how to handle uncertainties in the form of quantum is still an\nopen issue. In this paper, a new method is proposed to dispose uncertain\nquantum information, which is called two-dimensional quantum mass function\n(TDQMF). A TDQMF is consist of two elements, TQ = (Qoriginal, Qindicative),\nboth of the Qs are quantum mass functions, in which the Qindicative is an\nindicator of the reliability on Qoriginal. More flexibility and effectiveness\nare offered in handling uncertainty in the field of quantum by the proposed\nmethod compared with primary quantum mass function. Besides, some numerical\nexamples are provided and some practical applications are given to verify its\ncorrectness and validity\n"}
{"abstract": "  We investigate a simple quantitative genetics model subjet to a gradual\nenvironmental change from the viewpoint of the phylogenies of the living\nindividuals. We aim to understand better how the past traits of their ancestors\nare shaped by the adaptation to the varying environment. The individuals are\ncharacterized by a one-dimensional trait. The dynamics -- births and deaths --\ndepend on a time-changing mortality rate that shifts the optimal trait to the\nright at constant speed. The population size is regulated by a nonlinear\nnon-local logistic competition term. The macroscopic behaviour can be described\nby a PDE that admits a unique positive stationary solution. In the stationary\nregime, the population can persist, but with a lag in the trait distribution\ndue to the environmental change. For the microscopic (individual-based)\nstochastic process, the evolution of the lineages can be traced back using the\nhistorical process, that is, a measure-valued process on the set of continuous\nreal functions of time. Assuming stationarity of the trait distribution, we\ndescribe the limiting distribution, in large populations, of the path of an\nindividual drawn at random at a given time $T$. Freezing the non-linearity due\nto competition allows the use of a many-to-one identity together with\nFeynman-Kac's formula. This path, in reversed time, remains close to a simple\nOrnstein-Uhlenbeck process. It shows how the lagged bulk of the present\npopulation stems from ancestors once optimal in trait but still in the tail of\nthe trait distribution in which they lived.\n"}
{"abstract": "  This study focuses on comparing the individual polymer chain dynamics in an\nentangled polymeric liquid under different shear and extension rates. Polymer\nchains under various shear rates and extension rates were simulated using a\nstochastic-tube model [J. Rheol. 56: 1057 (2012)]. We developed a Matlab code\nto visualize and analyze the simulated configurations from the stochastic-tube\nmodel. We introduced new variables to determine how the extent of linearity\nchanges with time for different shear rates, which is more useful than a\ntypical end-to-end distance analysis. We identified whether the polymer chains\nundergo a tumbling rotation (slight elongation not accompanying contraction) or\nflipping rotation (elongation accompanying contraction). The simulation results\nindicate that the polymer chains exhibit a significant tendency to elongate at\nhigher shear rates and occasionally experience flipping, while lower shear\nrates tend to exhibit very frequent tumbling. Furthermore, no rotations were\nobserved under extensional flows. These results help clarifying uncertainty of\npreviously proposed polymer deformation mechanisms of the convective constraint\nrelease and the configuration-dependent friction coefficient.\n"}
{"abstract": "  Electric vector potential $\\Theta(\\boldsymbol{r})$ is a legitimate but rarely\nused tool to calculate the steady electric field in free-charge regions.\nCommonly, it is preferred to employ the scalar electric potential\n$\\Phi(\\boldsymbol{r})$ rather than $\\Theta(\\boldsymbol{r})$ in most of the\nelectrostatic problems. However, the electric vector potential formulation can\nbe a viable representation to study certain systems. One of them is the surface\nelectrode SE, a planar finite region $\\mathcal{A}_{-}$ kept at a fixed electric\npotential with the rest grounded including a gap of thickness $\\nu$ between\nelectrodes. In this document we use the \\textit{Helmholtz Decomposition\nTheorem} and the electric vector potential formulation to provide integral\nexpressions for the surface charge density and the electric field of the SE of\narbitrary contour $\\partial\\mathcal{A}$. We also present an alternative\nderivation of the result found in [M. Oliveira and J. A. Miranda 2001 Eur. J.\nPhys. 22 31] for the gapless ($\\nu=0$) surface electrode GSE without invoking\nany analogy between the GSE and magnetostatics. It is shown that electric\nvector potential and the electric field of the gapped circular SE at any point\ncan be obtained from an average of the gapless solution on the gap. Keywords:\nElectric vector potential, surface-electrode, Helmholtz Decomposition, Green's\ntheorem.\n"}
{"abstract": "  Thermodynamic properties of a system of interacting boson particles and\nantiparticles at finite temperatures are studied within the framework of the\nthermodynamically consistent Skyrme-like mean-field model. The mean field\ncontains both attractive and repulsive terms. Self-consistency relations\nbetween the mean field and thermodynamic functions are derived. We assume\nconservation of the isospin density for all temperatures. It is shown that,\nindependently of the strength of the attractive mean field, at the critical\ntemperature $T_{\\rm c}$ the system undergoes the phase transition of\nsecond-order to the Bose-Einstein condensate, which exists in the temperature\ninterval $0 \\le T \\le T_{\\rm c}$. We obtained that the condensation represents\na discontinuity of the derivative of the heat capacity at $T = T_{\\rm c}$, and\ncondensate occurs only for the component with a higher particle-number density\nin the particle-antiparticle system.\n"}
{"abstract": "  We provide strong evidence for single-electron spatial dynamics with a\nmacroscopic discontinuity. The dynamics is observed in an electron quantum\nphase consisted of macroscopic quantum orbits similar to those responsible for\nthe integer quantum Hall effect. The dynamics is consistent with the standard\nQM that predicts a \"cloud-like\" behavior of electron in quantum orbits\nregardless of their lengthscale. Sinse the dynamics is beyond the relativistic\nparadigm of movement, it may well be nonlocal. It thereby can be the basis for\na quantum concept of absolute simultaneity, which now rests not on the\nmeaningless notion of infinite speed but on empirically-tested quantum effect.\nThis concept revives the Bell-Popper idea to replace the current \"kinematic\"\nversion of relativity by the \"dynamic\" version by Lorentz and Poincare with a\nrevival of the classical view of space and time where the problem of preferred\nreference frame now can be solved without to involve the vague notion \"aether\".\nWe argue that the concept of quantum absolute simultaneity opens the door to a\nrealistic interpretation of standard QM through a deeper insight of what is\ncalled \"reality\" and ultimately to a harmony between quantum and relativity\ntheories.\n"}
{"abstract": "  We present a model for differentiating between images that are authentic\ncopies of ones published by photographers, and images that have been\nmanipulated by cropping, splicing or downsampling after publication. The model\ncomprises an encoder that resides with the photographer and a matching decoder\nthat is available to observers. The encoder learns to embed imperceptible\npositional signatures into image values prior to publication. The decoder\nlearns to use these steganographic positional (\"stegapos\") signatures to\ndetermine, for each small image patch, the 2D positional coordinates that were\nheld by the patch in its originally-published image. Crop, splice and\ndownsample edits become detectable by the inconsistencies they cause in the\nhidden positional signatures. We find that training the encoder and decoder\ntogether produces a model that imperceptibly encodes position, and that enables\nsuperior performance on established benchmarks for splice detection and high\naccuracy on a new benchmark for crop detection.\n"}
{"abstract": "  Extending DiNezza-Lu's approach to the setting of big cohomology classes, we\nprove that solutions of degenerate complex Monge-Amp{\\`e}re equations on\ncompact K{\\\"a}hler manifolds are continuous on a Zariski open set. This allows\nus to show that singular K{\\\"a}hler-Einstein metrics on log canonical varieties\nof general type have continuous potentials on the ample locus outside of the\nnon-klt part.\n"}
{"abstract": "  The task of sequential sentence classification enables the semantic\nstructuring of research papers. This can enhance academic search engines to\nsupport researchers in finding and exploring research literature more\neffectively. However, previous work has not investigated the potential of\ntransfer learning with datasets from different scientific domains for this task\nyet. We propose a uniform deep learning architecture and multi-task learning to\nimprove sequential sentence classification in scientific texts across domains\nby exploiting training data from multiple domains. Our contributions can be\nsummarised as follows: (1) We tailor two common transfer learning methods,\nsequential transfer learning and multi-task learning, and evaluate their\nperformance for sequential sentence classification; (2) The presented\nmulti-task model is able to recognise semantically related classes from\ndifferent datasets and thus supports manual comparison and assessment of\ndifferent annotation schemes; (3) The unified approach is capable of handling\ndatasets that contain either only abstracts or full papers without further\nfeature engineering. We demonstrate that models, which are trained on datasets\nfrom different scientific domains, benefit from one another when using the\nproposed multi-task learning architecture. Our approach outperforms the state\nof the art on three benchmark datasets.\n"}
{"abstract": "  For $n\\leq d$, a family ${\\cal F}=\\{C_0,C_1,\\ldots, C_n\\}$ of compact convex\nsets in $R^d$ is called an $n$-critical family provided any $n$ members of\n${\\cal F}$ have a non-empty intersection, but $\\bigcap_{i=0}^n\nC_i=\\varnothing$. If $n=d$ then a lemma on the intersection of convex sets due\nto Klee implies that the $d+1$ members of the $d$-critical family enclose a\n`hollow' in $R^d$, a bounded connected component of\n$R^d\\setminus\\bigcup_{i=0}^n C_i.$ Here we prove that the closure of the convex\nhull of a hollow in $R^d$ is a $d$-simplex.\n"}
{"abstract": "  In this article, the effect of geometrical parameters and flow conditions on\nthe performance of a swirl atomizer is studied. Dimensional analysis and\nexperimental investigations are utilized to define significant terms. The PDA\nsystem used for the measurements was able to supply information about the size,\nconcentration, and particle velocity at each measurement location. The orifice\ndiameter, the spiral cone angle, and also the flow Reynolds number, which is\ndefined based on the injector orifice diameter, play an important role in spray\nquality, and their significance is summarized in a correlation. In order to\nachieve the appropriate combination of design variables that satisfy the design\nconstraints, a GA-based program was used in a reverse analysis process.\nFinally, the advantages of human inspection were employed to provide true best\nperformers from a small group of final answers.\n"}
{"abstract": "  There is no exact closed form formula for pricing of European options with\ndiscrete cash dividends under the model where the underlying asset price\nfollows a piecewise lognormal process with jumps at dividend ex-dates. This\npaper presents alternative expansions based on the technique of Etore and\nGobet, leading to more robust first, second and third-order expansions across\nthe range of strikes and the range of dividend dates.\n"}
{"abstract": "  Existing malware detection approaches suffer from a simplistic trade-off\nbetween false positive rate (FPR) and true positive rate (TPR) due to a single\ntier classification approach, where the two measures adversely affect one\nanother. The practical implication for malware detection is that FPR must be\nkept at an acceptably low level while TPR remains high. To this end, we propose\na two-tiered learning, called ``Echelon\", from raw byte data with no need for\nhand-crafted features. The first tier locks FPR at a specified target level,\nwhereas the second tier improves TPR while maintaining the locked FPR. The core\nof Echelon lies at extracting activation information of the hidden layers of\nfirst tier model for constructing a stronger second tier model. Echelon is a\nframework in that it allows any existing CNN based model to be adapted in both\ntiers. We present experimental results of evaluating Echelon by adapting the\nstate-of-the-art malware detection model ``Malconv\" in the first and second\ntiers.\n"}
{"abstract": "  We develop several formal analogies between the logistic equation and the\nspatially homogeneous and isotropic relativistic cosmology described by the\nEinstein-Friedmann equations. These analogies produce an effective Lagrangian\nand Hamiltonian and new symmetries for the logistic equation.\n"}
{"abstract": "  We consider the manner in which the spacetime manifold emerges from a quantum\nsubstratum through the transactional process, in which spacetime events and\ntheir connections are established. In this account, there is no background\nspacetime as is generally assumed in physical theorizing. Instead, the usual\nnotion of a background spacetime is replaced by the quantum substratum,\ncomprising quantum systems with nonvanishing rest mass. Rest mass corresponds\nto internal periodicities that function as internal clocks defining proper\ntimes, and in turn, inertial frames that are not themselves aspects of the\nspacetime manifold, but are pre-spacetime reference structures. Specific\nprocesses in the quantum substratum serve to distinguish absolute from relative\nmotion.\n"}
{"abstract": "  A plethora of observational data obtained over the last couple of decades has\nallowed cosmology to enter into a precision era and has led to the foundation\nof the standard cosmological constant and cold dark matter paradigm, known as\nthe $\\Lambda$CDM model. Given the many possible extensions of this concordance\nmodel, we present here several novel consistency tests which could be used to\nprobe for deviations from $\\Lambda$CDM. First, we derive a joint consistency\ntest for the spatial curvature $\\Omega_{k,0}$ and the matter density\n$\\Omega_\\textrm{m,0}$ parameters, constructed using only the Hubble rate\n$H(z)$, which can be determined directly from observations. Second, we present\na new test of possible deviations from homogeneity using the combination of two\ndatasets, either the baryon acoustic oscillation (BAO) and $H(z)$ data or the\ntransversal and radial BAO data, while we also introduce two consistency tests\nfor $\\Lambda$CDM which could be reconstructed via the transversal and radial\nBAO data. We then reconstruct the aforementioned tests using the currently\navailable data in a model independent manner using a particular machine\nlearning approach, namely the Genetic Algorithms. Finally, we also report on a\n$\\sim 4\\sigma$ tension on the transition redshift as determined by the $H(z)$\nand radial BAO data.\n"}
{"abstract": "  Within the $SP(2,R)$ symmetry, the Two-time model (2T model) has six\ndimensions with two dimensions of time and the dilaton field that can be\nidentified as inflaton in a warm inflation scenario with potential of the form\n$\\sim\\phi^4$. From that consideration, we derive the range of parameters for\nthe Higgs-Dilaton potential, the coupling constant between Higgs and Dialton\n($\\alpha$) is smaller than $3.85\\times 10^{-7}$ when the mass of Dilaton is\nlager than $200$ GeV. Therefore, the 2T-model indirectly suggests that\nextra-dimension can also be a source of inflation.\n"}
{"abstract": "  We study a general convergence theory for the analysis of numerical solutions\nto the magnetohydrodynamic system describing the time evolution of\ncompressible, viscous, electrically conducting fluids in space dimension d (=\n2; 3). First, we introduce the concept of dissipative weak solutions and prove\nthe weak-strong uniqueness property for dissipative weak solutions, meaning a\ndissipative weak solution coincides with a classical solution emanating from\nthe same initial data on the lifespan of the latter. Next, we introduce the\nconcept of consistent approximations and prove the convergence of consistent\napproximations towards the dissipative weak solution as well as the classical\nsolution. Interpreting the consistent approximation as the energy stability and\nconsistency of numerical solutions, we have built a nonlinear variant of the\ncelebrated Lax-equivalence theorem. Finally, as an application of this theory,\nwe show the convergence analysis for two numerical methods.\n"}
{"abstract": "  Interacting spin networks are fundamental to quantum computing. Data-based\ntomography of time-independent spin networks has been achieved, but an open\nchallenge is to ascertain the structures of time-dependent spin networks using\ntime series measurements taken locally from a small subset of the spins.\nPhysically, the dynamical evolution of a spin network under time-dependent\ndriving or perturbation is described by the Heisenberg equation of motion.\nMotivated by this basic fact, we articulate a physics-enhanced machine learning\nframework whose core is Heisenberg neural networks. In particular, we develop a\ndeep learning algorithm according to some physics motivated loss function based\non the Heisenberg equation, which \"forces\" the neural network to follow the\nquantum evolution of the spin variables. We demonstrate that, from local\nmeasurements, not only the local Hamiltonian can be recovered but the\nHamiltonian reflecting the interacting structure of the whole system can also\nbe faithfully reconstructed. We test our Heisenberg neural machine on spin\nnetworks of a variety of structures. In the extreme case where measurements are\ntaken from only one spin, the achieved tomography fidelity values can reach\nabout 90%. The developed machine learning framework is applicable to any\ntime-dependent systems whose quantum dynamical evolution is governed by the\nHeisenberg equation of motion.\n"}
{"abstract": "  Radio interferometer arrays such as HERA consist of many close-packed dishes\narranged in a regular pattern, giving rise to a large number of `redundant'\nbaselines with the same length and orientation. Since identical baselines\nshould see an identical sky signal, this provides a way of finding a relative\ngain/bandpass calibration without needing an explicit sky model. In reality,\nthere are many reasons why baselines will not be exactly identical, giving rise\nto a host of effects that spoil the redundancy of the array and induce spurious\nstructure in the calibration solutions if not accounted for. In this paper, we\nseek to build an understanding of how differences in the primary beam response\nbetween antennas affect redundantly-calibrated interferometric visibilities and\ntheir resulting frequency (delay-space) power spectra. We use simulations to\nstudy several generic types of primary beam variation, including differences in\nthe width of the main lobe, the angular and frequency structure of the\nsidelobes, and the beam ellipticity and orientation. For all of these types, we\nfind that additional temporal structure is induced in the gain solutions,\nparticularly when bright point sources pass through the beam. In comparison,\nonly a low level of additional spectral structure is induced. The temporal\nstructure modulates the cosmological 21cm power spectrum, but only at the level\nof a few percent in our simulations. We also investigate the possibility of\nsignal loss due to decoherence effects when non-redundant visibilities are\naveraged together, finding that the decoherence is worst when bright point\nsources pass through the beam, and that its magnitude varies significantly\nbetween baseline groups and types of primary beam variation. Redundant\ncalibration absorbs some of the decoherence effect however, reducing its impact\ncompared to if the visibilities were perfectly calibrated.\n"}
{"abstract": "  Recent work has shown constrained Bayesian optimization to be a powerful\ntechnique for the optimization of industrial processes. In complex\nmanufacturing processes, the possibility to run extensive sequences of\nexperiments with the goal of finding good process parameters is severely\nlimited by the time required for quality evaluation of the produced parts. To\naccelerate the process parameter optimization, we introduce a parallel\nacquisition procedure tailored on the process characteristics. We further\npropose an algorithm that adapts to equipment status to improve run-to-run\nreproducibility. We validate our optimization method numerically and\nexperimentally, and demonstrate that it can efficiently find input parameters\nthat produce the desired outcome and minimize the process cost.\n"}
{"abstract": "  Learning from label proportions (LLP) aims at learning an instance-level\nclassifier with label proportions in grouped training data. Existing deep\nlearning based LLP methods utilize end-to-end pipelines to obtain the\nproportional loss with Kullback-Leibler divergence between the bag-level prior\nand posterior class distributions. However, the unconstrained optimization on\nthis objective can hardly reach a solution in accordance with the given\nproportions. Besides, concerning the probabilistic classifier, this strategy\nunavoidably results in high-entropy conditional class distributions at the\ninstance level. These issues further degrade the performance of the\ninstance-level classification. In this paper, we regard these problems as noisy\npseudo labeling, and instead impose the strict proportion consistency on the\nclassifier with a constrained optimization as a continuous training stage for\nexisting LLP classifiers. In addition, we introduce the mixup strategy and\nsymmetric crossentropy to further reduce the label noise. Our framework is\nmodel-agnostic, and demonstrates compelling performance improvement in\nextensive experiments, when incorporated into other deep LLP models as a\npost-hoc phase.\n"}
{"abstract": "  We propose a novel framework for off-chain execution and verification of\ncomputationally-intensive smart contracts. Our framework is the first solution\nthat avoids duplication of computing effort across multiple contractors, does\nnot require trusted execution environments, supports computations that do not\nhave deterministic results, and supports general-purpose computations written\nin a high-level language. Our experiments reveal that some intensive\napplications may require as much as 141 million gas, approximately 71x more\nthan the current block gas limit for computation in Ethereum today, and can be\navoided by utilizing the proposed framework.\n"}
{"abstract": "  In this paper, we propose reconfigurable intelligent surface (RIS)-assisted\nunmanned aerial vehicles (UAVs) networks that can utilise both advantages of\nUAV's agility and RIS's reflection for enhancing the network's performance. To\naim at maximising the energy efficiency (EE) of the considered networks, we\njointly optimise the power allocation of the UAVs and the phase-shift matrix of\nthe RIS. A deep reinforcement learning (DRL) approach is proposed for solving\nthe continuous optimisation problem with time-varying channels in a centralised\nfashion. Moreover, a parallel learning approach is also proposed for reducing\nthe information transmission requirement of the centralised approach. Numerical\nresults show a significant improvement of our proposed schemes compared with\nthe conventional approaches in terms of EE, flexibility, and processing time.\nOur proposed DRL methods for RIS-assisted UAV networks can be used for\nreal-time applications due to their capability of instant decision-making and\nhandling the time-varying channel with the dynamic environmental setting.\n"}
{"abstract": "  A magnetic field parallel to an electrical current does not produce a Lorentz\nforce on the charge carriers. Therefore, orbital longitudinal magnetoresistance\nis unexpected. Here we report on the observation of a large and non saturating\nmagnetoresistance in lightly doped SrTiO$_{3-x}$ independent of the relative\norientation of current and magnetic field. We show that this quasi-isotropic\nmagnetoresistance can be explained if the carrier mobility along all\norientations smoothly decreases with magnetic field. This anomalous regime is\nrestricted to low concentrations when the dipolar correlation length is longer\nthan the distance between carriers. We identify cyclotron motion of electrons\nin a potential landscape tailored by polar domains as the cradle of\nquasi-isotropic orbital magnetoresistance. The result emerges as a challenge to\ntheory and may be a generic feature of lightly-doped quantum paralectric\nmaterials.\n"}
{"abstract": "  We explore four approaches to the question of defectivity for a complex\nprojective toric variety $X_A$ associated with an integral configuration $A$.\nThe explicit tropicalization of the dual variety $X_A^\\vee$ due to Dickenstein,\nFeichtner, and Sturmfels allows for the computation of the defect in terms of\nan affine combinatorial invariant $\\rho(A)$. We express $\\rho(A)$ in terms of\naffine invariants $\\iota(A)$ associated to Esterov's iterated circuits and\n$\\lambda(A)$, an invariant defined by Curran and Cattani in terms of a Gale\ndual of $A$. Thus we obtain formulae for the dual defect in terms of iterated\ncircuits and Gale duals. An alternative expression for the dual defect of $X_A$\nis given by Furukawa-Ito in terms of Cayley decompositions of $A$. We give a\nGale dual interpretation of these decompositions and apply it to the study of\ndefective configurations.\n"}
{"abstract": "  We have measured the angular dependence of ferromagnetic resonance (FMR)\nspectra for Fibonacci-distorted, Kagome artificial spin ice (ASI). The number\nof strong modes in the FMR spectra depend on the orientation of the applied DC\nmagnetic field. In addition, discontinuities observed in the FMR\nfield-frequency dispersion curves also depend on DC field orientation, and\nsignal a multi-step DC magnetization reversal, which is caused by the reduced\nenergy degeneracy of Fibonacci-distorted vertices. The results suggest the\norientation of applied magnetic field and severity of Fibonacci distortion\nconstitute control variables for FMR modes and multi-step reversal in future\nmagnonic devices and magnetic switching systems.\n"}
{"abstract": "  The existence of a gapped chiral spin liquid has been recently suggested in\nthe vicinity of the metal-insulator transition of the Hubbard model on the\ntriangular lattice, by intensive density-matrix renormalization group (DMRG)\nsimulations [A. Szasz, J. Motruk, M.P. Zaletel, and J.E. Moore, Phys. Rev. X\n{\\bf 10}, 021042 (2020)]. Here, we report the results obtained within the\nvariational Monte Carlo technique based upon Jastrow-Slater wave functions,\nimplemented with backflow correlations. As in DMRG calculations, we consider\n$N$-leg cylinders. For $N=4$ and in the presence of a next-nearest neighbor\nhopping, a chiral spin liquid emerges between the metal and the insulator with\nmagnetic quasi-long-range order. Within our approach, the chiral state is\ngapped and breaks the reflection symmetry. By contrast, for both $N=5$ and\n$N=6$, the chiral spin liquid is not the state with the lowest variational\nenergy: in the former case a nematic spin liquid is found in the entire\ninsulating regime, while for the less frustrated case with $N=6$ the results\nare very similar to the one obtained on two-dimensional clusters [L.F. Tocchio,\nA. Montorsi, and F. Becca, Phys. Rev. B {\\bf 102}, 115150 (2020)], with an\nantiferromagnetic phase close to the metal-insulator transition and a nematic\nspin liquid in the strong-coupling regime.\n"}
{"abstract": "  The earth system is exceedingly complex and often chaotic in nature, making\nprediction incredibly challenging: we cannot expect to make perfect predictions\nall of the time. Instead, we look for specific states of the system that lead\nto more predictable behavior than others, often termed \"forecasts of\nopportunity.\" When these opportunities are not present, scientists need\nprediction systems that are capable of saying \"I don't know.\" We introduce a\nnovel loss function, termed the \"NotWrong loss\", that allows neural networks to\nidentify forecasts of opportunity for classification problems. The NotWrong\nloss introduces an abstention class that allows the network to identify the\nmore confident samples and abstain (say \"I don't know\") on the less confident\nsamples. The abstention loss is designed to abstain on a user-defined fraction\nof the samples via a PID controller. Unlike many machine learning methods used\nto reject samples post-training, the NotWrong loss is applied during training\nto preferentially learn from the more confident samples. We show that the\nNotWrong loss outperforms other existing loss functions for multiple climate\nuse cases. The implementation of the proposed loss function is straightforward\nin most network architectures designed for classification as it only requires\nthe addition of an abstention class to the output layer and modification of the\nloss function.\n"}
{"abstract": "  Direct reconstruction methods have been developed to estimate parametric\nimages directly from the measured PET sinograms by combining the PET imaging\nmodel and tracer kinetics in an integrated framework. Due to limited counts\nreceived, signal-to-noise-ratio (SNR) and resolution of parametric images\nproduced by direct reconstruction frameworks are still limited. Recently\nsupervised deep learning methods have been successfully applied to medical\nimaging denoising/reconstruction when large number of high-quality training\nlabels are available. For static PET imaging, high-quality training labels can\nbe acquired by extending the scanning time. However, this is not feasible for\ndynamic PET imaging, where the scanning time is already long enough. In this\nwork, we proposed an unsupervised deep learning framework for direct parametric\nreconstruction from dynamic PET, which was tested on the Patlak model and the\nrelative equilibrium Logan model. The patient's anatomical prior image, which\nis readily available from PET/CT or PET/MR scans, was supplied as the network\ninput to provide a manifold constraint, and also utilized to construct a kernel\nlayer to perform non-local feature denoising. The linear kinetic model was\nembedded in the network structure as a 1x1 convolution layer. The training\nobjective function was based on the PET statistical model. Evaluations based on\ndynamic datasets of 18F-FDG and 11C-PiB tracers show that the proposed\nframework can outperform the traditional and the kernel method-based direct\nreconstruction methods.\n"}
{"abstract": "  Weak-scale secluded sector dark matter can reproduce the observed dark matter\nrelic density with thermal freeze-out within that sector. If nature is\nsupersymmetric, three portals to the visible sector - a gauge portal, a Higgs\nportal, and a gaugino portal - are present. We present gamma ray spectra\nrelevant for indirect detection of dark matter annihilation in such setups.\nSince symmetries in the secluded sector can stabilize dark matter, $R$-parity\nis unnecessary, and we investigate the impact of $R$-parity violation on\nannihilation spectra. We present limits from the Fermi Large Area Telescope\nobservations of dwarf galaxies and projections for Cherenkov Telescope Array\nobservations of the galactic center. Many of our results are also applicable to\ngeneric, non-supersymmetric setups.\n"}
{"abstract": "  The movement of large-scale (tens of Terabytes and larger) data sets between\nhigh performance computing (HPC) facilities is an important and increasingly\ncritical capability. A growing number of scientific collaborations rely on HPC\nfacilities for tasks which either require large-scale data sets as input or\nproduce large-scale data sets as output. In order to enable the transfer of\nthese data sets as needed by the scientific community, HPC facilities must\ndesign and deploy the appropriate data transfer capabilities to allow users to\ndo data placement at scale.\n  This paper describes the Petascale DTN Project, an effort undertaken by four\nHPC facilities, which succeeded in achieving routine data transfer rates of\nover 1PB/week between the facilities. We describe the design and configuration\nof the Data Transfer Node (DTN) clusters used for large-scale data transfers at\nthese facilities, the software tools used, and the performance tuning that\nenabled this capability.\n"}
{"abstract": "  Influencing spectral and directional features of exciton emission\ncharacteristics from 2D transition metal dichalcogenides by coupling it to\nplasmonic nano-cavities has emerged as an important prospect in nanophotonics\nof 2D materials. In this paper we experimentally study the directional\nphotoluminescence emission from Tungsten disulfide (WS2) monolayer sandwiched\nbetween a single-crystalline plasmonic silver nanowire (AgNW) waveguide and a\ngold (Au) mirror, thus forming an AgNW-WS2-Au cavity. By employing\npolarization-resolved Fourier plane optical microscopy, we quantify the\ndirectional emission characteristics from the distal end of the AgNW-WS2-Au\ncavity. Given that our geometry simultaneously facilitates local field\nenhancement and waveguiding capability, we envisage its utility in 2D\nmaterial-based, on-chip nanophotonic signal processing, including nonlinear and\nquantum optical regimes.\n"}
{"abstract": "  Metamaterials are artificially created media, which allow introducing\nadditional degrees of freedom into electromagnetic design by controlling\nconstitutive material parameters. Reconfigurable time-dependent metamaterials\ncan further enlarge those capabilities by introducing a temporal variable as an\nadditional controllable parameter. Here we demonstrate a first-of-its-kind\nreconfigurable volumetric metamaterial-based scatterer, wherein the\nelectromagnetic properties are controlled dynamically with light. In\nparticular, hybridized resonances in arrays of split ring resonators give rise\nto a collective mode that presents properties of artificial high-frequency\nmagnetism for centimeter waves. Resonant behavior of each individual ring is\ncontrolled with a photocurrent, which allows the fast tuning of macroscopic\neffective permeability. Thus, the artificial gigahertz magnon resonant\nexcitation within a subwavelength spherical scatterer is governed by light\nintensity. Four-dimensional control over electromagnetic scattering in both\nspace and time opens new venues for modern applications, including wireless\ncommunications and automotive radars to name just a few.\n"}
{"abstract": "  Progress towards the energy breakthroughs needed to combat climate change can\nbe significantly accelerated through the efficient simulation of atomic\nsystems. Simulation techniques based on first principles, such as Density\nFunctional Theory (DFT), are limited in their practical use due to their high\ncomputational expense. Machine learning approaches have the potential to\napproximate DFT in a computationally efficient manner, which could dramatically\nincrease the impact of computational simulations on real-world problems.\nApproximating DFT poses several challenges. These include accurately modeling\nthe subtle changes in the relative positions and angles between atoms, and\nenforcing constraints such as rotation invariance or energy conservation. We\nintroduce a novel approach to modeling angular information between sets of\nneighboring atoms in a graph neural network. Rotation invariance is achieved\nfor the network's edge messages through the use of a per-edge local coordinate\nframe and a novel spin convolution over the remaining degree of freedom. Two\nmodel variants are proposed for the applications of structure relaxation and\nmolecular dynamics. State-of-the-art results are demonstrated on the\nlarge-scale Open Catalyst 2020 dataset. Comparisons are also performed on the\nMD17 and QM9 datasets.\n"}
{"abstract": "  Connected and Automated Vehicles (CAVs) rely on the correctness of position\nand other vehicle kinematics information to fulfill various driving tasks such\nas vehicle following, lane change, and collision avoidance. However, a\nmalicious vehicle may send false sensor information to the other vehicles\nintentionally or unintentionally, which may cause traffic inconvenience or loss\nof human lives. Here, we take the advantage of cloud-computing and increase the\nresilience of CAVs to malicious vehicles by assuming each vehicle shares its\nlocal sensor information with other vehicles to create information redundancy\non the cloud side. We exploit this redundancy and propose a sensor fusion\nalgorithm for the cloud, capable of providing a robust state estimation of all\nvehicles in the cloud under the condition that the number of malicious\ninformation is sufficiently small. Using the proposed estimator, we provide an\nalgorithm for isolating malicious vehicles. We use numerical examples to\nillustrate the effectiveness of our methods.\n"}
{"abstract": "  The kinetic term of the $N$-body Hamiltonian system defined on the surface of\nthe sphere is non-separable. As a result, standard explicit symplectic\nintegrators are inapplicable. We exploit an underlying hierarchy in the\nstructure of the kinetic term to construct an explicit time-reversible\nsymplectic scheme of second order. We use iterative applications of the method\nto construct a fourth order scheme and demonstrate its efficiency.\n"}
{"abstract": "  In this paper, we study properties of quadratic variations of c\\`{a}dl\\`{a}g\npaths within the framework of the It\\^{o}--F\\\"{o}llmer calculus in Banach\nspaces. We prove a generalized $C^1$-transformation formula for quadratic\nvariations. We also investigate relations between tensor and scalar quadratic\nvariations.\n"}
{"abstract": "  We study interactions between strategic players and markets whose behavior is\nguided by an algorithm. Algorithms use data from prior interactions and a\nlimited set of decision rules to prescribe actions. While as-if rational play\nneed not emerge if the algorithm is constrained, it is possible to guide\nbehavior across a rich set of possible environments using limited details.\nProvided a condition known as weak learnability holds, Adaptive Boosting\nalgorithms can be specified to induce behavior that is (approximately) as-if\nrational. Our analysis provides a statistical perspective on the study of\nendogenous model misspecification.\n"}
{"abstract": "  This tutorial paper surveys training alternatives to end-to-end\nbackpropagation (E2EBP) -- the de facto standard for training deep\narchitectures. Modular training refers to strictly local training without both\nthe forward and the backward pass, i.e., dividing a deep architecture into\nseveral nonoverlapping modules and training them separately without any\nend-to-end operation. Between the fully global E2EBP and the strictly local\nmodular training, there are \"weakly modular\" hybrids performing training\nwithout the backward pass only. These alternatives can match or surpass the\nperformance of E2EBP on challenging datasets such as ImageNet, and are gaining\nincreased attention primarily because they offer practical advantages over\nE2EBP, which will be enumerated herein. In particular, they allow for greater\nmodularity and transparency in deep learning workflows, aligning deep learning\nwith the mainstream computer science engineering that heavily exploits\nmodularization for scalability. Modular training has also revealed novel\ninsights about learning and has further implications on other important\nresearch domains. Specifically, it induces natural and effective solutions to\nsome important practical problems such as data efficiency and transferability\nestimation.\n"}
{"abstract": "  Beyond generating long and topic-coherent paragraphs in traditional\ncaptioning tasks, the medical image report composition task poses more\ntask-oriented challenges by requiring both the highly-accurate medical term\ndiagnosis and multiple heterogeneous forms of information including impression\nand findings. Current methods often generate the most common sentences due to\ndataset bias for individual case, regardless of whether the sentences properly\ncapture key entities and relationships. Such limitations severely hinder their\napplicability and generalization capability in medical report composition where\nthe most critical sentences lie in the descriptions of abnormal diseases that\nare relatively rare. Moreover, some medical terms appearing in one report are\noften entangled with each other and co-occurred, e.g. symptoms associated with\na specific disease. To enforce the semantic consistency of medical terms to be\nincorporated into the final reports and encourage the sentence generation for\nrare abnormal descriptions, we propose a novel framework that unifies template\nretrieval and sentence generation to handle both common and rare abnormality\nwhile ensuring the semantic-coherency among the detected medical terms.\nSpecifically, our approach exploits hybrid-knowledge co-reasoning: i) explicit\nrelationships among all abnormal medical terms to induce the visual attention\nlearning and topic representation encoding for better topic-oriented symptoms\ndescriptions; ii) adaptive generation mode that changes between the template\nretrieval and sentence generation according to a contextual topic encoder.\nExperimental results on two medical report benchmarks demonstrate the\nsuperiority of the proposed framework in terms of both human and metrics\nevaluation.\n"}
{"abstract": "  We investigate the repeated prisoner's dilemma game where both players\nalternately use reinforcement learning to obtain their optimal memory-one\nstrategies. We theoretically solve the simultaneous Bellman optimality\nequations of reinforcement learning. We find that the Win-stay Lose-shift\nstrategy, the Grim strategy, and the strategy which always defects can form\nsymmetric equilibrium of the mutual reinforcement learning process amongst all\ndeterministic memory-one strategies.\n"}
{"abstract": "  We apply a modified vector meson dominance (VMD) model to analyze the\nelectromagnetic form factors of the $\\Lambda_c$ hyperon in the time-like\nreaction $e^+e^-\\rightarrow \\Lambda_c^+ \\bar{\\Lambda}_c^-$. In the model, we\ninclude the contributions from the vector charmed mesons and their excitations\n$\\psi(1S)$, $\\psi(2S)$, $\\psi(3770)$, $\\psi(4040)$, $\\psi(4160)$ and\n$\\psi(4415)$. We perform a combined fit to the available data on the Born cross\nsection in reaction $e^+e^-\\rightarrow \\Lambda_c^+ \\bar{\\Lambda}_c^-$ and the\nratio of the electromagnetic form factors $|G_E/G_M|$ to obtain the values of\nthe model parameters. Our results show that the VMD model can simultaneously\ndescribe the data of electromagnetic form factors from the Belle and BESIII\nCollaborations, and the behavior of $|G_E/G_M|$ for the BESIII data can be\nqualitatively reproduced by the VMD model prediction at the threshold region.\nMoreover, we predict the single and double polarization observables in\n$e^+e^-\\rightarrow \\Lambda_c^+ \\bar{\\Lambda}_c^-$ reactions, which are\nexperimentally accessible in the polarized process. We also obtain the form\nfactors of the $\\Lambda_c$ hyperon in the space-like region via analytic\ncontinuing the time-like form factors.\n"}
{"abstract": "  We report on an unusually bright observation of PSR J2051$-$0827 recorded\nduring a regular monitoring campaign of black-widow pulsar systems with the\nEffelsberg 100-m telescope. Through fortunate coincidence, a particularly\nbright scintillation maximum is simultaneous with the eclipse by the companion,\nenabling precise measurements of variations in the flux density, dispersion\nmeasure (DM), and scattering strength throughout the eclipse. The flux density\nis highly variable throughout the eclipse, with a peak 1.7 times the average\naway from the eclipse, and yet does not significantly decrease on average. We\nrecover the flux density variations from the measured DM variations using\ngeometric optics, with a relative velocity as the only free parameter. We\nmeasure an effective velocity of (470 $\\pm$ 10) km/s, consistent with the\nrelative orbital motion of the companion, suggesting that the outflow velocity\nof the lensing material is low, or is directly along the line of sight. The 2\nper cent uncertainty on the effective velocity is a formal error; systematics\nrelated to our current model are likely to dominate, and we detail several\nextensions to the model to be considered in a full treatment of lensing. This\nis a demonstration of the causal link between DM and lensing; the flux density\nvariations can be predicted directly through the derivatives of DM. Going\nforward, this approach can be applied to investigate the dynamics of other\neclipsing systems, and to investigate the physical nature of scintillation and\nlensing in the ionized interstellar medium.\n"}
{"abstract": "  The proton elastic form factor ratio shows a discrepancy between measurements\nusing the Rosenbluth technique in unpolarized beam and target experiments and\nmeasurements using polarization degrees of freedom. The proposed explanation of\nthis discrepancy is uncorrected hard two-photon exchange (TPE), a type of\nradiative correction that is conventionally neglected. The effect size and\nagreement with theoretical predictions has been tested recently by three\nexperiments. While the results support the existence of a small two-photon\nexchange effect, they cannot establish that theoretical treatments are valid.\nAt larger momentum transfers, theory remains untested. This proposal aims to\nmeasure two-photon exchange over an extended and so far largely untested $Q^2$\nand $\\varepsilon$ range with high precision using the {\\tt CLAS12} experiment.\nSuch data are crucial to clearly confirm or rule out TPE as the driver for the\ndiscrepancy as well as test several theoretical approaches, believed valid in\ndifferent parts of the tested $Q^2$ range.\n"}
{"abstract": "  We study a smart grid with wind power and battery storage. Traditionally,\nday-ahead planning aims to balance demand and wind power, yet actual wind\nconditions often deviate from forecasts. Short-term flexibility in storage and\ngeneration fills potential gaps, planned on a minutes time scale for 30-60\nminute horizons. Finding the optimal flexibility deployment requires solving a\nsemi-infinite non-convex stochastic program, which is generally intractable to\ndo exactly. Previous approaches rely on sampling, yet such critical problems\ncall for rigorous approaches with stronger guarantees. Our method employs\nprobabilistic model checking techniques. First, we cast the problem as a\ncontinuous-space Markov decision process with discretized control, for which an\noptimal deployment strategy minimizes the expected grid frequency deviation. To\nmitigate state space explosion, we exploit specific structural properties of\nthe model to implement an iterative exploration method that reuses pre-computed\nvalues as wind data is updated. Our experiments show the method's feasibility\nand versatility across grid configurations and time scales.\n"}
{"abstract": "  The Leray-Serre and the Eilenberg-Moore spectral sequence are fundamental\ntools for computing the cohomology of a group or, more generally, of a space.\nWe describe the relationship between these two spectral sequences when both of\nthem share the same abutment. There exists a joint tri-graded refinement of the\nLeray-Serre and the Eilenberg-Moore spectral sequence. This refinement involves\ntwo more spectral sequences, the preludes from the title, which abut to the\ninitial terms of the Leray-Serre and the Eilenberg-Moore spectral sequence,\nrespectively. We show that one of these always degenerates from its second page\non and that the other one satisfies a local-to-global property: It degenerates\nfor all possible base spaces if and only if it does so when the base space is\ncontractible. When the preludes degenerate early enough, they appear to echo\nDeligne's d\\'ecalage, but in general, this is an illusion. We discuss several\nprincipal fibrations to illustrate the possible cases and give applications, in\nparticular, to Lie groups, torus bundles, and generalizations.\n"}
{"abstract": "  Motivated by recent developments of hydrodynamical quantum mechanical analogs\n[J. W. M. Bush, Annu. Rev. Fluid Mech. 47, 269-292 (2015)] we provide a\nrelativistic model for a classical particle coupled to a scalar wave-field\nthrough a holonomic constraint. In presence of an external Coulomb field we\ndefine a regime where the particle is guided by the wave in a way similar to\nthe old de Broglie phase-wave proposal. Moreover, this dualistic mechanical\nanalog of the quantum theory is reminiscent of the double-solution approach\nsuggested by de Broglie in 1927 and is able to reproduce the Bohr-Sommerfeld\nsemiclassical quantization formula for an electron moving in a atom.\n"}
{"abstract": "  Tellurium trioxide, TeO3, is the only example of a trioxide adopting at\nambient conditions the VF3-type structure (a distorted variant of the cubic\nReO3 structure). Here we present a combined experimental (Raman scattering) and\ntheoretical (DFT modelling) study on the influence of high pressure (exceeding\n100 GPa) on the phase stability of this compound. In experiment the\nambient-pressure VF3-type structure (R-3c symmetry) is preserved up to 110 GPa.\nIn contrast, calculations indicate that above 66 GPa the R-3c structure should\ntransform to a YF3-type polymorph (Pnma symmetry) with the coordination number\nof Te6+ increasing from 6 to 8 upon the transition. The lack of this transition\nin the room-temperature experiment is most probably connected with energetic\nbarriers, in analogy to what is found for compressed WO3. The YF3-type phase is\npredicted to be stable up to 220 GPa when it should transform to a novel\nstructure of R-3 symmetry and Z = 18. We analyse the influence of pressure on\nthe band gap of TeO3, and discuss the present findings in the context of\nstructural transformations of trioxides and trifluorides adopting an extended\nstructure in the solid state\n"}
{"abstract": "  We present a source-to-source transformation framework, GOCC, that consumes\nlock-based pessimistic concurrency programs in the Go language and transforms\nthem into optimistic concurrency programs that use Hardware Transactional\nMemory (HTM). The choice of the Go language is motivated by the fact that\nconcurrency is a first-class citizen in Go, and it is widely used in Go\nprograms. GOCC performs rich inter-procedural program analysis to detect and\nfilter lock-protected regions and performs AST-level code transformation of the\nsurrounding locks when profitable. Profitability is driven by both static\nanalyses of critical sections and dynamic analysis via execution profiles. A\ncustom HTM library, using perceptron, learns concurrency behavior and\ndynamically decides whether to use HTM in the rewritten lock/unlock points.\nGiven the rich history of transactional memory research but its lack of\nadoption in any industrial setting, we believe this workflow, which ultimately\nproduces source-code patches, is more apt for industry-scale adoption. Results\non widely adopted Go libraries and applications demonstrate significant (up to\n10x) and scalable performance gains resulting from our automated transformation\nwhile avoiding major performance regressions.\n"}
{"abstract": "  Wearable biosensor technology enables real-time, convenient, and continuous\nmonitoring of users behavioral signals. Such include signals relative to body\nmotion, body temperature, biological or biochemical markers, and individual\ngrip forces, which are studied in this paper. A four step pick and drop image\nguided and robot assisted precision task has been designed for exploiting a\nwearable wireless sensor glove system. Individual spatio temporal grip forces\nare analyzed on the basis of thousands of individual sensor data, collected\nfrom different locations on the dominant and non-dominant hands of each of\nthree users in ten successive task sessions. Statistical comparisons reveal\nspecific differences between grip force profiles of the individual users as a\nfunction of task skill level (expertise) and time.\n"}
{"abstract": "  MDS codes are elegant constructions in coding theory and have mode important\napplications in cryptography, network coding, distributed data storage,\ncommunication systems et. In this study, a method is given which MDS codes are\nlifted to a higher finite field. The presented method satisfies the protection\nof the distance and creating the MDS code over the $F_q$ by using MDS code over\n$F_p.$\n"}
{"abstract": "  The mechanism of the generation of dark matter and dark radiation from the\nevaporation of primordial black holes is very interesting. We consider the case\nof Kerr black holes to generalize previous results obtained in the\nSchwarzschild case. For dark matter, the results do not change dramatically and\nthe bounds on warm dark matter apply similarly: in particular, the Kerr case\ncannot save the scenario of black hole domination for light dark matter. For\ndark radiation, the expectations for $\\Delta N_{eff}$ do not change\nsignificantly with respect to the Schwarzschild case, but for an enhancement in\nthe case of spin 2 particles: in the massless case, however, the projected\nexperimental sensitivity would be reached only for extremal black holes.\n"}
{"abstract": "  In this work, we introduce, justify and demonstrate the Corrective Source\nTerm Approach (CoSTA) -- a novel approach to Hybrid Analysis and Modeling\n(HAM). The objective of HAM is to combine physics-based modeling (PBM) and\ndata-driven modeling (DDM) to create generalizable, trustworthy, accurate,\ncomputationally efficient and self-evolving models. CoSTA achieves this\nobjective by augmenting the governing equation of a PBM model with a corrective\nsource term generated using a deep neural network. In a series of numerical\nexperiments on one-dimensional heat diffusion, CoSTA is found to outperform\ncomparable DDM and PBM models in terms of accuracy -- often reducing predictive\nerrors by several orders of magnitude -- while also generalizing better than\npure DDM. Due to its flexible but solid theoretical foundation, CoSTA provides\na modular framework for leveraging novel developments within both PBM and DDM.\nIts theoretical foundation also ensures that CoSTA can be used to model any\nsystem governed by (deterministic) partial differential equations. Moreover,\nCoSTA facilitates interpretation of the DNN-generated source term within the\ncontext of PBM, which results in improved explainability of the DNN. These\nfactors make CoSTA a potential door-opener for data-driven techniques to enter\nhigh-stakes applications previously reserved for pure PBM.\n"}
{"abstract": "  Background: The quantitative analysis of microscope videos often requires\ninstance segmentation and tracking of cellular and subcellular objects. The\ntraditional method consists of two stages: (1) performing instance object\nsegmentation of each frame, and (2) associating objects frame-by-frame.\nRecently, pixel-embedding-based deep learning approaches these two steps\nsimultaneously as a single stage holistic solution. In computer vision,\nannotated training data with consistent segmentation and tracking is resource\nintensive, the severity of which is multiplied in microscopy imaging due to (1)\ndense objects (e.g., overlapping or touching), and (2) high dynamics (e.g.,\nirregular motion and mitosis). Adversarial simulations have provided successful\nsolutions to alleviate the lack of such annotations in dynamics scenes in\ncomputer vision, such as using simulated environments (e.g., computer games) to\ntrain real-world self-driving systems. Methods: In this paper, we propose an\nannotation-free synthetic instance segmentation and tracking (ASIST) method\nwith adversarial simulation and single-stage pixel-embedding based learning.\nContribution: The contribution of this paper is three-fold: (1) the proposed\nmethod aggregates adversarial simulations and single-stage pixel-embedding\nbased deep learning; (2) the method is assessed with both the cellular (i.e.,\nHeLa cells) and subcellular (i.e., microvilli) objects; and (3) to the best of\nour knowledge, this is the first study to explore annotation-free instance\nsegmentation and tracking study for microscope videos. Results: The ASIST\nmethod achieved an important step forward, when compared with fully supervised\napproaches: ASIST shows 7% to 11% higher segmentation, detection and tracking\nperformance on microvilli relative to fully supervised methods, and comparable\nperformance on Hela cell videos.\n"}
{"abstract": "  An effective approach to improve the detection efficiency of nanoscale light\nsources relies on a planar antenna configuration, which beams the emitted light\ninto a narrow cone. Planar antennas operate like optical Yagi-Uda antennas,\nwhere reflector and director elements are made of metal films. Here we\nintroduce and investigate, both theoretically and experimentally, a scanning\nimplementation of a planar antenna. Using a small ensemble of molecules\ncontained in fluorescent nanobeads placed in the antenna, we independently\naddress the intensity, the radiation pattern and the decay rate as a function\nof distance between a flat-tip scanning gold wire (reflector) and a thin gold\nfilm coated on a glass coverslip (director). The scanning planar antenna\nchanges the radiation pattern of a single fluorescent bead and it beams light\ninto a narrow cone down to angles of 45{\\deg} (full width at half maximum).\nMoreover, the collected signal compared to the case of a glass coverslip is\nlarger than a factor of 3, which is mainly due to the excitation enhancement.\nThese results offer a better understanding of the modification of light-matter\ninteraction by planar antennas and they hold promise for applications, such as\nsensing, imaging and diagnostics.\n"}
{"abstract": "  Large MIMO transceivers are integral components of next-generation wireless\nnetworks. However, for such systems to be practical, their channel estimation\nprocess needs to be fast and reliable. Although several solutions for fast\nestimation of sparse channels do exist, there is still a gap in understanding\nthe fundamental limits governing this problem. Specifically, we need to better\nunderstand the lower bound on the number of measurements under which accurate\nchannel estimates can be obtained. This work bridges that knowledge gap by\nderiving a tight asymptotic lower bound on the number of measurements. This not\nonly helps develop a better understanding for the sparse MIMO channel\nestimation problem, but it also provides a benchmark for evaluating current and\nfuture solutions.\n"}
{"abstract": "  Here we study the transport properties of a nickelate compound, namely LaNiO2\nusing density functional theory (DFT) in conjunction with dynamical mean field\ntheory (DMFT). An interacting multi-orbital spin-resolved scenario for LaNiO2\nyields a metallic ground state with ferromagnetic correlations. The latter\ncontrasts an antiferromagnetic order reported earlier. The metallic behaviour\npersists even at large values of the interaction energies. Further support of\nthe metallic state is provided by the angular resolved photoemission\nspectroscopy (ARPES) data.\n"}
{"abstract": "  Large-scale distributed training of Deep Neural Networks (DNNs) on\nstate-of-the-art platforms is expected to be severely communication\nconstrained. To overcome this limitation, numerous gradient compression\ntechniques have been proposed and have demonstrated high compression ratios.\nHowever, most existing methods do not scale well to large scale distributed\nsystems (due to gradient build-up) and/or fail to evaluate model fidelity (test\naccuracy) on large datasets. To mitigate these issues, we propose a new\ncompression technique, Scalable Sparsified Gradient Compression (ScaleCom),\nthat leverages similarity in the gradient distribution amongst learners to\nprovide significantly improved scalability. Using theoretical analysis, we show\nthat ScaleCom provides favorable convergence guarantees and is compatible with\ngradient all-reduce techniques. Furthermore, we experimentally demonstrate that\nScaleCom has small overheads, directly reduces gradient traffic and provides\nhigh compression rates (65-400X) and excellent scalability (up to 64 learners\nand 8-12X larger batch sizes over standard training) across a wide range of\napplications (image, language, and speech) without significant accuracy loss.\n"}
{"abstract": "  We present a method for comparing the classical and quantum calculations of\nthe electric quadrupolar Love number $k_2$ and show that our previous\nderivation of the quantum Love number of a quantum blackhole matches exactly\nthe classical calculation of $k_2$ when quantum expectation values are replaced\nby the corresponding classical quantities, as dictated by the Bohr\ncorrespondence principle. The standard derivation of $k_2$ for classical\nrelativistic stars relies on fixing boundary conditions on the surface of the\nstar for the Einstein equations in the presence of an external perturbing\nfield. An alternative method for calculating $k_2$ uses properties of the\nspectrum of the non-relativistic fluid modes of the star. We adopt this\nalternative method and use it to derive an effective description of the\ninterior modes in terms of a collection of driven harmonic oscillators\ncharacterized by different frequencies and amplitudes. We compare these two\nclassical methods and find that most of the interior information can be\nintegrated out, reducing the problem of calculating $k_2$ to fixing a single\nboundary condition for the perturbed Einstein equations on the surface of the\ndeformed star. We then determine this single boundary condition in terms of the\nspectrum of the object and proceed to identify the relationship between\nclassical quantities and quantum expectation values and to verify the agreement\nbetween the results of the effective classical calculation and the quantum\ncalculation.\n"}
{"abstract": "  In this paper we consider the moduli space of complete, conformally flat\nmetrics on a sphere with k punctures having constant positive Q-curvature and\npositive scalar curvature. Previous work has shown that such metrics admit an\nasymptotic expansion near each puncture, allowing one to define an asymptotic\nnecksize of each singular point. We prove that any set in the moduli space such\nthat the distances between distinct punctures and the asymptotic necksizes all\nremain bounded away from zero is sequentially compact, mirroring a theorem of\nD. Pollack about singular Yamabe metrics. Along the way we define a radial\nPohozaev invariant at each puncture and refine some a priori bounds of the\nconformal factor, which may be of independent interest.\n"}
{"abstract": "  Unmanned Aerial Vehicle (UAV) offers lots of applications in both commerce\nand recreation. With this, monitoring the operation status of UAVs is crucially\nimportant. In this work, we consider the task of tracking UAVs, providing rich\ninformation such as location and trajectory. To facilitate research on this\ntopic, we propose a dataset, Anti-UAV, with more than 300 video pairs\ncontaining over 580k manually annotated bounding boxes. The releasing of such a\nlarge-scale dataset could be a useful initial step in research of tracking\nUAVs. Furthermore, the advancement of addressing research challenges in\nAnti-UAV can help the design of anti-UAV systems, leading to better\nsurveillance of UAVs. Besides, a novel approach named dual-flow semantic\nconsistency (DFSC) is proposed for UAV tracking. Modulated by the semantic flow\nacross video sequences, the tracker learns more robust class-level semantic\ninformation and obtains more discriminative instance-level features.\nExperimental results demonstrate that Anti-UAV is very challenging, and the\nproposed method can effectively improve the tracker's performance. The Anti-UAV\nbenchmark and the code of the proposed approach will be publicly available at\nhttps://github.com/ucas-vg/Anti-UAV.\n"}
{"abstract": "  Dealing with nonlinear effects of the radio-frequency(RF) chain is a key\nissue in the realization of very large-scale multi-antenna (MIMO) systems.\nAchieving the remarkable gains possible with massive MIMO requires that the\nsignal processing algorithms systematically take into account these effects.\nHere, we present a computationally efficient linear precoding method satisfying\nthe requirements for low peak-to-average power ratio (PAPR) and low-resolution\nD/A-converters (DACs). The method is based on a sparse regularization of the\nprecoding matrix and offers advantages in terms of precoded signal PAPR as well\nas processing complexity. Through simulation, we find that the method\nsubstantially improves conventional linear precoders.\n"}
{"abstract": "  The stochastic SIRS model is a continuous-time Markov chain modelling the\nspread of infectious diseases with temporary immunity, in a\nhomogeneously-mixing population of fixed size $N$. We study the scaling\nbehaviour of the extinction time of stochastic SIRS models as $N$ tends to\ninfinity. When the initial size of infected population is small, we obtain the\nclosed-form expression of the asymptotic distribution of this extinction time,\nand compare it with the data from Monte Carlo simulation.\n"}
{"abstract": "  We prove a ``positive'' Jantzen sum formula for the Specht modules of the\ncyclotomic Hecke algebras of type~$A$. That is, in the Grothendieck group, we\nshow that the sum of the pieces of the Jantzen filtration is equal to an\nexplicit non-negative linear combination of modules $E^\\nu_{f,e}$, which are\nmodular reductions of simple modules for closely connected Hecke algebras in\ncharacteristic zero. The coefficient of $E^\\nu_{f,e}$ in the sum formula is\ndetermined by the graded decomposition numbers in characteristic zero, which\nare known, and the characteristic of the field. As a consequence we see that\nthe decomposition numbers of a cyclotomic Hecke algebra at an $e$th root of\nunity in characteristic $p$ depend on the decomposition numbers of related\ncyclotomic Hecke algebras at $ep^r$th roots of unity in characteristic zero,\nfor $r\\ge0$.\n"}
{"abstract": "  By using small computing devices deployed at user premises, Autonomous Demand\nResponse (ADR) adapts users electricity consumption to given time-dependent\nelectricity tariffs. This allows end-users to save on their electricity bill\nand Distribution System Operators to optimise (through suitable time-dependent\ntariffs) management of the electric grid by avoiding demand peaks.\nUnfortunately, even with ADR, users power consumption may deviate from the\nexpected (minimum cost) one, e.g., because ADR devices fail to correctly\nforecast energy needs at user premises. As a result, the aggregated power\ndemand may present undesirable peaks. In this paper we address such a problem\nby presenting methods and a software tool (APD-Analyser) implementing them,\nenabling Distribution System Operators to effectively verify that a given\ntime-dependent electricity tariff achieves the desired goals even when\nend-users deviate from their expected behaviour. We show feasibility of the\nproposed approach through a realistic scenario from a medium voltage Danish\ndistribution network.\n"}
{"abstract": "  We present a brief review of power laws and correlation functions as measures\nof criticality and the relation between them. By comparing phenomenology from\nrain, brain and the forest fire model we discuss the relevant features of\nself-organisation to the vicinity about a critical state. We conclude that\norganisation to a region of extended correlations and approximate power laws\nmay be behaviour of interest shared between the three considered systems.\n"}
{"abstract": "  We evaluated the DiscovErr system, in which we had previously implemented a\nnew methodology for assessment of compliance to continuous application of\nevidence-based clinical guidelines, based on a bidirectional search from the\nguideline objectives to the patient's longitudinal data, and vice versa. We\ncompared the system comments on 1584 transactions regarding the management,\nover a mean of 5.23 years, of 10 randomly selected Type 2 diabetes patients, to\nthose of two diabetes experts and a senior family practitioner. After providing\ntheir own comments, the experts assessed both the correctness (precision) and\nthe importance of each of the DiscovErr system comments. The completeness\n(recall or coverage) of the system was computed by comparing its comments to\nthose made by the experts. The system made 279 comments. The experts made 181\nunique comments. The completeness of the system was 91% compared to comments\nmade by at least two experts, and 98% when compared to comments made by all\nthree. 172 comments were evaluated by the experts for correctness and\nimportance: All 114 medication-related comments, and a random 35% of the 165\nmonitoring-related comments. The system's correctness was 81% compared to\ncomments judged as correct by both diabetes experts, and 91% compared to\ncomments judged as correct by a diabetes expert and at least as partially\ncorrect by the other. 89% of the comments were judged as important by both\ndiabetes experts, 8% were judged as important by one expert, 3% were judged as\nless important by both experts. The completeness scores of the three experts\n(compared to the comments of all experts plus the validated system comments)\nwere 75%, 60%, and 55%; the experts' correctness scores (compared to their\nmajority) were respectively 99%, 91%, and 88%. Conclusion: Systems such as\nDiscovErr can assess the quality of continuous guideline-based care.\n"}
{"abstract": "  We study the mod $p$ motivic cohomology of homogeneous varieties such as\n$GL_{n}/GL_{r}$ or $Sp_{2n}/Sp_{2n-2}$ along with the action of the Steenrod\noperations, without restrictions on the characteristic of the base field. In\nparticular, we prove that certain quotient maps do not admit sections.\n"}
{"abstract": "  Machine learning-based systems are rapidly gaining popularity and in-line\nwith that there has been a huge research surge in the field of explainability\nto ensure that machine learning models are reliable, fair, and can be held\nliable for their decision-making process. Explainable Artificial Intelligence\n(XAI) methods are typically deployed to debug black-box machine learning models\nbut in comparison to tabular, text, and image data, explainability in time\nseries is still relatively unexplored. The aim of this study was to achieve and\nevaluate model agnostic explainability in a time series forecasting problem.\nThis work focused on proving a solution for a digital consultancy company\naiming to find a data-driven approach in order to understand the effect of\ntheir sales related activities on the sales deals closed. The solution involved\nframing the problem as a time series forecasting problem to predict the sales\ndeals and the explainability was achieved using two novel model agnostic\nexplainability techniques, Local explainable model-agnostic explanations (LIME)\nand Shapley additive explanations (SHAP) which were evaluated using human\nevaluation of explainability. The results clearly indicate that the\nexplanations produced by LIME and SHAP greatly helped lay humans in\nunderstanding the predictions made by the machine learning model. The presented\nwork can easily be extended to any time\n"}
{"abstract": "  Single-image super-resolution (SR) and multi-frame SR are two ways to super\nresolve low-resolution images. Single-Image SR generally handles each image\nindependently, but ignores the temporal information implied in continuing\nframes. Multi-frame SR is able to model the temporal dependency via capturing\nmotion information. However, it relies on neighbouring frames which are not\nalways available in the real world. Meanwhile, slight camera shake easily\ncauses heavy motion blur on long-distance-shot low-resolution images. To\naddress these problems, a Blind Motion Deblurring Super-Reslution Networks,\nBMDSRNet, is proposed to learn dynamic spatio-temporal information from single\nstatic motion-blurred images. Motion-blurred images are the accumulation over\ntime during the exposure of cameras, while the proposed BMDSRNet learns the\nreverse process and uses three-streams to learn Bidirectional spatio-temporal\ninformation based on well designed reconstruction loss functions to recover\nclean high-resolution images. Extensive experiments demonstrate that the\nproposed BMDSRNet outperforms recent state-of-the-art methods, and has the\nability to simultaneously deal with image deblurring and SR.\n"}
{"abstract": "  Analyzing political ideology and polarization is of critical importance in\nadvancing our understanding of the political context in society. Recent\nresearch has made great strides towards understanding the ideological bias\n(i.e., stance) of news media along a left-right spectrum. In this work, we take\na novel approach and study the ideology of the policy under discussion teasing\napart the nuanced co-existence of stance and ideology. Aligned with the\ntheoretical accounts in political science, we treat ideology as a\nmulti-dimensional construct, and introduce the first diachronic dataset of news\narticles whose political ideology under discussion is annotated by trained\npolitical scientists and linguists at the paragraph-level. We showcase that\nthis framework enables quantitative analysis of polarization, a temporal,\nmultifaceted measure of ideological distance. We further present baseline\nmodels for ideology prediction.\n"}
{"abstract": "  A single Majorana Kramers pair hosts only one component of the magnetic\nmultipole. This can be used to determine the bulk Cooper-pair symmetry through\nsurface-sensitive spectroscopic measurements, either by applying a magnetic\nfield or by using a ferromagnet/superconductor junction. This paper proposes\nthat the electric response, which is free from the Meissner effect, can be used\nan alternative method to measure the bulk Cooper-pair symmetry in\ntime-reversal-invariant superconductors with double Majorana Kramers pairs. The\nrelationships among electric multipoles, strain tensors and superconducting\nsymmetries under a given wallpaper group on the surfaces of topological\ncrystalline superconductors are shown. This study also reveals that only a\nspecific irreducible representation of a uniform strain yields a gap in the\ndouble Majorana Kramers pairs for the topological-crystalline-superconductor\ncandidate Sr$_3$SnO. This highlights the viability of electric detection\nregarding Cooper-pair symmetry.\n"}
{"abstract": "  The Higgs mode associated with amplitude fluctuations of the superconducting\ngap in uniform superconductors usually is heavy, which makes its excitation and\ndetection difficult. We report on the existence of a gapless Higgs mode in the\nFulde-Ferrell-Larkin-Ovchinnikov states. This feature is originated from the\nGoldstone mode associated with the translation symmetry breaking. The existence\nof the gapless Higgs mode is demonstrated by using both a phenomenological\nmodel and microscopic Bardeen-Cooper-Schrieffer (BCS) theory. The gapless Higgs\nmode can avoid the decay into other low energy excitations, which renders it\nstable and detectable.\n"}
{"abstract": "  In this paper, we use a model of large random network game where the agents\nplays selfishly and are affected by their neighbors, to explore the conditions\nunder which the Nash equilibrium (NE) of the game is affected by a perturbation\nin the network. We use a phase transition phenomenon observed in finite rank\ndeformations of large random matrices, to study how the NE changes on crossing\ncritical threshold points. Our main contribution is as follows: when the\nperturbation strength is greater than a critical point, it impacts the NE of\nthe game, whereas when this perturbation is below this critical point, the NE\nremains independent of the perturbation parameter. This demonstrates a phase\ntransition in NE which alludes that perturbations can affect the behavior of\nthe society only if their strength is above a critical threshold. We provide\nnumerical examples for this result and present scenarios under which this\nphenomenon could potentially occur in real world applications.\n"}
{"abstract": "  Sequences and time-series often arise in robot tasks, e.g., in activity\nrecognition and imitation learning. In recent years, deep neural networks\n(DNNs) have emerged as an effective data-driven methodology for processing\nsequences given sufficient training data and compute resources. However, when\ndata is limited, simpler models such as logic/rule-based methods work\nsurprisingly well, especially when relevant prior knowledge is applied in their\nconstruction. However, unlike DNNs, these \"structured\" models can be difficult\nto extend, and do not work well with raw unstructured data. In this work, we\nseek to learn flexible DNNs, yet leverage prior temporal knowledge when\navailable. Our approach is to embed symbolic knowledge expressed as linear\ntemporal logic (LTL) and use these embeddings to guide the training of deep\nmodels. Specifically, we construct semantic-based embeddings of automata\ngenerated from LTL formula via a Graph Neural Network. Experiments show that\nthese learnt embeddings can lead to improvements in downstream robot tasks such\nas sequential action recognition and imitation learning.\n"}
{"abstract": "  Stochastic gravitational wave backgrounds (SGWBs) receive increasing\nattention and provide a new possibility to directly probe the early Universe.\nIn the preheating process at the end of inflation, parametric resonance can\ngenerate large energy density perturbations and efficiently produce\ngravitational waves (GWs) which carry unique information about inflation. Since\nthe peak frequency of such GWs is approximately proportional to the\ninflationary energy scale, $\\Lambda_{\\mathrm{inf}}$, GWs from preheating are\nexpected to be observed by interferometer GW detectors in low-scale\ninflationary models. We investigate the dependence of the amplitude of such GWs\non $\\Lambda_{\\mathrm{inf}}$, and find that the present energy spectrum of these\nGWs does not depend on $\\Lambda_{\\mathrm{inf}}$ only in the case of\n$\\Lambda_{\\mathrm{inf}}$ is above a critical value $\\Lambda_{c}$, a parameter\ndepending on the resonance strength. We numerically obtain $\\Lambda_{c}$ in\nterms of the model parameters in linear approximation and then conduct lattice\nsimulations to verify this result. For\n$\\Lambda_{\\mathrm{inf}}\\lesssim\\Lambda_{c}$, the amplitude of GWs quickly\ndecreases with $\\Lambda_{\\mathrm{inf}}$ and becomes challenging to observe. In\nturn, observing such GWs in interferometer detectors also helps to determine\n$\\Lambda_{\\mathrm{inf}}$ and the resonance strength during the preheating.\n"}
{"abstract": "  We exhibit a non-hyperelliptic curve C of genus 3 such that the class of the\nCeresa cycle [C]-[-C] in the intermediate Jacobian of JC is torsion.\n"}
{"abstract": "  Cascade prediction estimates the size or the state of a cascade from either\nmicroscope or macroscope. It is of paramount importance for understanding the\ninformation diffusion process such as the spread of rumors and the propagation\nof new technologies in social networks. Recently, instead of extracting\nhand-crafted features or embedding cascade sequences into feature vectors for\ncascade prediction, graph neural networks (GNNs) are introduced to utilize the\nnetwork structure which governs the cascade effect. However, these models do\nnot take into account social factors such as personality traits which drive\nhuman's participation in the information diffusion process. In this work, we\npropose a novel multitask framework for enhancing cascade prediction with a\npersonality recognition task. Specially, we design a general plug-and-play GNN\ngate, named PersonalityGate, to couple into existing GNN-based cascade\nprediction models to enhance their effectiveness and extract individuals'\npersonality traits jointly. Experimental results on two real-world datasets\ndemonstrate the effectiveness of our proposed framework in enhancing GNN-based\ncascade prediction models and in predicting individuals' personality traits as\nwell.\n"}
{"abstract": "  We prove a new rigorous bound for the mean convective heat transport $\\langle\nw T \\rangle$, where $w$ and $T$ are the nondimensional vertical velocity and\ntemperature, in internally heated convection between an insulating lower\nboundary and an upper boundary with a fixed heat flux. The quantity $\\langle wT\n\\rangle$ is equal to half the ratio of convective to conductive vertical heat\ntransport, and also to $\\frac12$ plus the mean temperature difference between\nthe top and bottom boundaries. An analytical application of the background\nmethod based on the construction of a quadratic auxiliary function yields\n$\\langle w T \\rangle \\leq \\tfrac{1}{2}\\big(\\tfrac{1}{2}+ \\tfrac{1}{\\sqrt{3}}\n\\big) - 1.6552\\, R^{-\\frac13}$ uniformly in the Prandtl number, where $R$ is\nthe nondimensional control parameter measuring the strength of the internal\nheating. Numerical optimisation of the auxiliary function suggests that the\nasymptotic value of this bound and the $-1/3$ exponent are optimal within our\nbounding framework. This new result halves the best existing (uniform in $R$)\nbound (Goluskin 2016, Springer, Table 1.2) and its dependence on $R$ is\nconsistent with previous conjectures and heuristic scaling arguments. Contrary\nto physical intuition, however, it does not rule out a mean heat transport\nlarger than $\\frac12$ at high $R$, which corresponds to the top boundary being\nhotter than the bottom one on average.\n"}
{"abstract": "  Convolutional neural networks (CNN) have been frequently used to extract\nsubject-invariant features from electroencephalogram (EEG) for classification\ntasks. This approach holds the underlying assumption that electrodes are\nequidistant analogous to pixels of an image and hence fails to explore/exploit\nthe complex functional neural connectivity between different electrode sites.\nWe overcome this limitation by tailoring the concepts of convolution and\npooling applied to 2D grid-like inputs for the functional network of electrode\nsites. Furthermore, we develop various graph neural network (GNN) models that\nproject electrodes onto the nodes of a graph, where the node features are\nrepresented as EEG channel samples collected over a trial, and nodes can be\nconnected by weighted/unweighted edges according to a flexible policy\nformulated by a neuroscientist. The empirical evaluations show that our\nproposed GNN-based framework outperforms standard CNN classifiers across ErrP,\nand RSVP datasets, as well as allowing neuroscientific interpretability and\nexplainability to deep learning methods tailored to EEG related classification\nproblems. Another practical advantage of our GNN-based framework is that it can\nbe used in EEG channel selection, which is critical for reducing computational\ncost, and designing portable EEG headsets.\n"}
{"abstract": "  We present a novel yet simple approach to produced multiple entangled photon\npairs through spontaneous parametric downconversion. We have developed Gaussian\nmasks to subdivide the pump beam before passing it through a nonlinear medium.\nIn this way, we are able to observe simultaneous separate down-converted\nemission cones with spatial overlap. The technique we employ can be used to\ngreatly increase the dimensionality of entangled photonic systems generated\nfrom spontaneous parametric down conversion, affording greater scalability to\noptical quantum computing than previously explored.\n"}
{"abstract": "  In this work, we investigate the asymptotic spectral density of the random\nfeature matrix $M = Y Y^\\ast$ with $Y = f(WX)$ generated by a\nsingle-hidden-layer neural network, where $W$ and $X$ are random rectangular\nmatrices with i.i.d. centred entries and $f$ is a non-linear smooth function\nwhich is applied entry-wise. We prove that the Stieltjes transform of the\nlimiting spectral distribution approximately satisfies a quartic\nself-consistent equation, which is exactly the equation obtained by\n[Pennington, Worah] and [Benigni, P\\'ech\\'e] with the moment method. We extend\nthe previous results to the case of additive bias $Y=f(WX+B)$ with $B$ being an\nindependent rank-one Gaussian random matrix, closer modelling the neural\nnetwork infrastructures encountered in practice. Our key finding is that in the\ncase of additive bias it is impossible to choose an activation function\npreserving the layer-to-layer singular value distribution, in sharp contrast to\nthe bias-free case where a simple integral constraint is sufficient to achieve\nisospectrality. To obtain the asymptotics for the empirical spectral density we\nfollow the resolvent method from random matrix theory via the cumulant\nexpansion. We find that this approach is more robust and less combinatorial\nthan the moment method and expect that it will apply also for models where the\ncombinatorics of the former become intractable. The resolvent method has been\nwidely employed, but compared to previous works, it is applied here to\nnon-linear random matrices.\n"}
{"abstract": "  We address inference for a partially observed nonlinear non-Gaussian latent\nstochastic system comprised of interacting units. Each unit has a state, which\nmay be discrete or continuous, scalar or vector valued. In biological\napplications, the state may represent a structured population or the abundances\nof a collection of species at a single location. Units can have spatial\nlocations, allowing the description of spatially distributed interacting\npopulations arising in ecology, epidemiology and elsewhere. We consider models\nwhere the collection of states is a latent Markov process, and a time series of\nnoisy or incomplete measurements is made on each unit. A model of this form is\ncalled a spatiotemporal partially observed Markov process (SpatPOMP). The R\npackage spatPomp provides an environment for implementing SpatPOMP models,\nanalyzing data, and developing new inference approaches. We describe the\nspatPomp implementations of some methods with scaling properties suited to\nSpatPOMP models. We demonstrate the package on a simple Gaussian system and on\na nontrivial epidemiological model for measles transmission within and between\ncities. We show how to construct user-specified SpatPOMP models within\nspatPomp.\n"}
{"abstract": "  This paper considers Bayesian revenue maximization in the $k$-unit setting,\nwhere a monopolist seller has $k$ copies of an indivisible item and faces $n$\nunit-demand buyers (whose value distributions can be non-identical). Four basic\nmechanisms among others have been widely employed in practice and widely\nstudied in the literature: {\\sf Myerson Auction}, {\\sf Sequential\nPosted-Pricing}, {\\sf $(k + 1)$-th Price Auction with Anonymous Reserve}, and\n{\\sf Anonymous Pricing}. Regarding a pair of mechanisms, we investigate the\nlargest possible ratio between the two revenues (a.k.a.\\ the revenue gap), over\nall possible value distributions of the buyers.\n  Divide these four mechanisms into two groups: (i)~the discriminating\nmechanism group, {\\sf Myerson Auction} and {\\sf Sequential Posted-Pricing}, and\n(ii)~the anonymous mechanism group, {\\sf Anonymous Reserve} and {\\sf Anonymous\nPricing}. Within one group, the involved two mechanisms have an asymptotically\ntight revenue gap of $1 + \\Theta(1 / \\sqrt{k})$. In contrast, any two\nmechanisms from the different groups have an asymptotically tight revenue gap\nof $\\Theta(\\log k)$.\n"}
{"abstract": "  We present the extension of a modeling technique for Lagrangian tracer\nparticles [B. Viggiano et al., J. Fluid Mech.(2020), vol. 900, A27] which\naccounts for the effects of particle inertia. Thereby, the particle velocity\nfor several Stokes numbers is modeled directly by a multi-layered\nOrnstein-Uhlenbeck process and a comparison of key statistical quantities\n(second-order velocity structure function, acceleration correlation function,\nand root mean square acceleration) to expressions derived from Batchelor's\nmodel as well as to direct numerical simulations (DNS) is performed. In both\napproaches, Stokes' drag is treated by an approximate ``linear filter'' which\nreplaces the particle position entering the fluid velocity field by the\ncorresponding ideal tracer position. Effects of preferential concentration of\ninertial particles are taken into account in terms of an effective Stokes\nnumber that is determined from the zero crossing of the acceleration\ncorrelation function from DNS. This approximation thus allows the modeling of\ninertial particle statistics through stochastic methods and models for the\nLagrangian velocity; the particle velocity is effectively decoupled from the\nparticle position. In contrast to the ordinary filtering technique [Cencini et\nal., J. Turbul. (2006), 7, N36], our method captures the effects of\npreferential concentration of particles at low Stokes numbers which manifest\nthemselves for instance by a sharp decrease of the acceleration variance for\nincreasing Stokes numbers.\n"}
{"abstract": "  This paper addresses the challenge of understanding the waiting dependencies\nbetween the threads and hardware resources required to complete a task. The\nobjective is to improve software performance by detecting the underlying\nbottlenecks caused by system-level blocking dependencies. In this paper, we use\na system level tracing approach to extract a Waiting Dependency Graph that\nshows the breakdown of a task execution among all the interleaving threads and\nresources. The method allows developers and system administrators to quickly\ndiscover how the total execution time is divided among its interacting threads\nand resources. Ultimately, the method helps detecting bottlenecks and\nhighlighting their possible causes. Our experiments show the effectiveness of\nthe proposed approach in several industry-level use cases. Three performance\nanomalies are analysed and explained using the proposed approach. Evaluating\nthe method efficiency reveals that the imposed overhead never exceeds 10.1%,\ntherefore making it suitable for in-production environments.\n"}
{"abstract": "  In [6], D'Angelo introduced the notion of finite type for points $p$ of a\nreal hypersurface $M$ of $\\mathbb C^n$ by defining the order of contact\n$\\Delta_q(M,p)$ of complex analytic $q$-dimensional varieties with $M$ at $p$.\nLater, Catlin [4] defined $q$-type, $D_q(M,p)$ for points of hypersurfaces by\nconsidering generic $(n-q+1)$-dimensional complex affine subspaces of $\\mathbb\nC^n$. We define a generalization of the Catlin's $q$-type for an arbitrary\nsubset $M$ of $\\mathbb C^n$ in a similar way that D'Angelo's 1-type,\n$\\Delta_1(M,p)$, is generalized in [13]. Using recent results connecting the\nD'Angelo and Catlin $q$-types in [1] and building on D'Angelo's work on the\nopenness of the set of points of finite $\\Delta_q$-type, we prove the openness\nof the set of points of finite Catlin $q$-type for an arbitrary subset\n$M\\subset \\mathbb C^n$.\n"}
{"abstract": "  Manipulation of light-induced magnetization has become a fundamentally hot\ntopic with a potentially high impact for atom trapping, confocal and magnetic\nresonance microscopy, and data storage. The control of the magnetization\norientation mainly relies on the direct methods composed of amplitude, phase\nand polarization modulations of the incident light under the tight focusing\ncondition, leaving the achievement of arbitrary desirable three-dimensional\n(3D) magnetization orientation complicated, inflexible and inefficient. Here,\nwe propose a facile approach called machine learning inverse design to achieve\nexpected vectorial opto-magnetization orientation. This pathway is\ntime-efficient and accurate to produce the demanded incident beam for arbitrary\nprescribed 3D magnetization orientation. It is highlighted that the machine\nlearning method is not only applied for magnetization orientations, but also\nwidely used in the control of magnetization structures.\n"}
{"abstract": "  Measurement-based quantum correlation mimics several characteristics of\nmultipartite quantum correlations and at the same time, it reduces the parent\nsystem to a smaller subsystem. On the other hand, genuine multipartite\nentanglement measures can capture certain features of a multisite composite\nsystem that are inaccessible via bipartite quantum correlation quantifiers. We\nmerge these two concepts by introducing localizable genuine multimode\nentanglement for continuous variable systems, both for Gaussian and\nnon-Gaussian multimode parent states. We report a compact form of localizable\ngeneralized geometric measure for multimode Gaussian states when Gaussian\nmeasurements are performed in some of the modes. We show that non-Gaussian\nmeasurements can concentrate more genuine multimode entanglement compared to\nthe Gaussian ones. For non-Gaussian states with non-Gaussian measurements, we\nfind that although four-mode squeezed vacuum state has permutation symmetry\nwith respect to the exchange of first and third modes as well as the second and\nthe fourth modes, the symmetry can be broken by performing measurements in one\nof the modes in case of addition while for subtraction, such symmetry is\npreserved, thereby providing a method for distinguishing multimode photon-added\nand -subtracted states via localizations.\n"}
{"abstract": "  Quantum sensors offer the capability to reach unprecedented precision by\noperating at the standard quantum limit (SQL) or beyond by using quantum\nentanglement. But an emerging class of quantum sensors that use Rydberg\nelectromagnetically-induced transparency (EIT) to detect rf electric fields\nhave yet to reach the SQL. In this work we prove that this discrepancy is due\nto fundamental limitations in the EIT probing mechanism. We derive the optimum\nsensitivity of a three-level quantum sensor based on EIT, or more generally\ncoherent spectroscopy, and compare this to the SQL. We apply a minimal set of\nassumptions, while allowing strong probing fields, thermal broadening, and\nlarge optical depth. We derive the optimal laser intensities and optical depth,\nproviding specific guidelines for sensitive operation under common experimental\nconditions. Clear boundaries of performance are established, revealing that\nladder-EIT can not achieve the SQL due to unavoidable absorption loss. The\nresults may be applied to any EIT-based quantum sensor, but we particularly\nemphasize our results' importance to the growing field of Rydberg quantum\nsensing.\n"}
{"abstract": "  Point cloud semantic segmentation is a crucial task in 3D scene\nunderstanding. Existing methods mainly focus on employing a large number of\nannotated labels for supervised semantic segmentation. Nonetheless, manually\nlabeling such large point clouds for the supervised segmentation task is\ntime-consuming. In order to reduce the number of annotated labels, we propose a\nsemi-supervised semantic point cloud segmentation network, named SSPC-Net,\nwhere we train the semantic segmentation network by inferring the labels of\nunlabeled points from the few annotated 3D points. In our method, we first\npartition the whole point cloud into superpoints and build superpoint graphs to\nmine the long-range dependencies in point clouds. Based on the constructed\nsuperpoint graph, we then develop a dynamic label propagation method to\ngenerate the pseudo labels for the unsupervised superpoints. Particularly, we\nadopt a superpoint dropout strategy to dynamically select the generated pseudo\nlabels. In order to fully exploit the generated pseudo labels of the\nunsupervised superpoints, we furthermore propose a coupled attention mechanism\nfor superpoint feature embedding. Finally, we employ the cross-entropy loss to\ntrain the semantic segmentation network with the labels of the supervised\nsuperpoints and the pseudo labels of the unsupervised superpoints. Experiments\non various datasets demonstrate that our semi-supervised segmentation method\ncan achieve better performance than the current semi-supervised segmentation\nmethod with fewer annotated 3D points. Our code is available at\nhttps://github.com/MMCheng/SSPC-Net.\n"}
{"abstract": "  We study the Auslander ring of a basic left K\\\"othe ring $\\Lambda$ and give a\ncharacterization of basic left K\\\"othe rings in terms of their Auslander rings.\nWe also study the functor category Mod$((\\Lambda$-mod$)^{\\rm op})$ and\ncharacterize basic left K\\\"othe rings $\\Lambda$ by using functor categories\nMod$((\\Lambda$-mod$)^{\\rm op})$. As a consequence we show that there exists a\nbijection between the Morita equivalence classes of left Kawada rings and the\nMorita equivalence classes of Auslander generalized right QF-2 rings.\n"}
{"abstract": "  Many tasks in robot-assisted surgery require planning and controlling\nmanipulators' motions that interact with highly deformable objects. This study\nproposes a realistic, time-bounded simulator based on Position-based Dynamics\n(PBD) simulation that mocks brain deformations due to catheter insertion for\npre-operative path planning and intra-operative guidance in keyhole surgical\nprocedures. It maximizes the probability of success by accounting for\nuncertainty in deformation models, noisy sensing, and unpredictable actuation.\nThe PBD deformation parameters were initialized on a parallelepiped-shaped\nsimulated phantom to obtain a reasonable starting guess for the brain white\nmatter. They were calibrated by comparing the obtained displacements with\ndeformation data for catheter insertion in a composite hydrogel phantom.\nKnowing the gray matter brain structures' different behaviors, the parameters\nwere fine-tuned to obtain a generalized human brain model. The brain\nstructures' average displacement was compared with values in the literature.\nThe simulator's numerical model uses a novel approach with respect to the\nliterature, and it has proved to be a close match with real brain deformations\nthrough validation using recorded deformation data of in-vivo animal trials\nwith a mean mismatch of 4.73$\\pm$2.15%. The stability, accuracy, and real-time\nperformance make this model suitable for creating a dynamic environment for KN\npath planning, pre-operative path planning, and intra-operative guidance.\n"}
{"abstract": "  Neural dialogue generation models trained with the one-hot target\ndistribution suffer from the over-confidence issue, which leads to poor\ngeneration diversity as widely reported in the literature. Although existing\napproaches such as label smoothing can alleviate this issue, they fail to adapt\nto diverse dialog contexts. In this paper, we propose an Adaptive Label\nSmoothing (AdaLabel) approach that can adaptively estimate a target label\ndistribution at each time step for different contexts. The maximum probability\nin the predicted distribution is used to modify the soft target distribution\nproduced by a novel light-weight bi-directional decoder module. The resulting\ntarget distribution is aware of both previous and future contexts and is\nadjusted to avoid over-training the dialogue model. Our model can be trained in\nan end-to-end manner. Extensive experiments on two benchmark datasets show that\nour approach outperforms various competitive baselines in producing diverse\nresponses.\n"}
{"abstract": "  The contact geometric structure of the thermodynamic phase space is used to\nintroduce a novel symplectic structure on the tangent bundle of the equilibrium\nspace. Moreover, it turns out that the equilibrium space can be interpreted as\na Lagrange submanifold of the corresponding tangent bundle, if the fundamental\nequation is known explicitly. As a consequence, the Hamiltonian description of\nthermodynamics is obtained in a natural way and a Hamilton-Jacobi-like equation\nis introduced whose solutions determine entire families of thermodynamic\nsystems.\n"}
{"abstract": "  We study the prospects of future gravitational wave (GW) detectors in probing\nprimordial black hole (PBH) binaries. We show that across a broad mass range\nfrom $10^{-5}M_\\odot$ to $10^7M_\\odot$, future GW interferometers provide a\npotential probe of the PBH abundance that is more sensitive than any currently\nexisting experiment. In particular, we find that galactic PBH binaries with\nmasses as low as $10^{-5}M_\\odot$ may be probed with ET, AEDGE and LISA by\nsearching for nearly monochromatic continuous GW signals. Such searches could\nindependently test the PBH interpretation of the ultrashort microlensing events\nobserved by OGLE. We also consider the possibility of observing GWs from\nasteroid mass PBH binaries through graviton-photon conversion.\n"}
{"abstract": "  Let $K$ be a field and let $S=K[x_1,\\dots,x_n]$ be a standard polynomial ring\nover a field $K$. We characterize the extremal Betti numbers, values as well\npositions, of a $t$-spread strongly stable ideal of $S$. Our approach is\nconstructive. Indeed, given some positive integers $a_1,\\dots,a_r$ and some\npairs of positive integers $(k_1,\\ell_1),\\dots,(k_r,\\ell_r)$, we are able to\ndetermine under which conditions there exist a $t$-spread strongly stable ideal\n$I$ of $S$ with $\\beta_{k_i, k_i\\ell_i}(I)=a_i$, $i=1, \\ldots, r$, as extremal\nBetti numbers, and then to construct it.\n"}
{"abstract": "  Inferential (or soft) sensors are used in industry to infer the values of\nimprecisely and rarely measured (or completely unmeasured) variables from\nvariables measured online (e.g., pressures, temperatures). The main challenge,\nakin to classical model overfitting, in designing an effective inferential\nsensor is the selection of a correct structure of the sensor. The sensor\nstructure is represented by the number of inputs to the sensor, which\ncorrespond to the variables measured online and their (simple) combinations.\nThis work is focused on the design of inferential sensors for product\ncomposition of an industrial distillation column in two oil refinery units, a\nFluid Catalytic Cracking unit and a Vacuum Gasoil Hydrogenation unit. As the\nfirst design step, we use several well-known data pre-treatment (gross error\ndetection) methods and compare the ability of these approaches to indicate\nsystematic errors and outliers in the available industrial data. We then study\neffectiveness of various methods for design of the inferential sensors taking\ninto account the complexity and accuracy of the resulting model. The\neffectiveness analysis indicates that the improvements achieved over the\ncurrent inferential sensors are up to 19 %.\n"}
{"abstract": "  How can we identify similar repositories and clusters among a large online\narchive, such as GitHub? Determiningrepository similarity is an essential\nbuilding block in studying the dynamics and the evolution of such software\necosystems. The key challenge is to determine the right representation for the\ndiverse repository features in a way that: (a) it captures all aspects of the\navailable information, and (b) it is readily usable by MLalgorithms. We propose\nRepo2Vec, a comprehensive embedding approach to represent a repository as a\ndistributed vector by combining features from three types of information\nsources. As our key novelty, we consider three types of information:\n(a)metadata, (b) the structure of the repository, and (c) the source code. We\nalso introduce a series of embedding approaches to represent and combine these\ninformation types into a single embedding. We evaluate our method with two real\ndatasets from GitHub for a combined 1013 repositories. First, we show that our\nmethod outperforms previous methods in terms of precision (93%vs 78%), with\nnearly twice as many Strongly Similar repositories and 30% fewer False\nPositives. Second, we show how Repo2Vecprovides a solid basis for: (a)\ndistinguishing between malware and benign repositories, and (b) identifying a\nmeaningful hierarchical clustering. For example, we achieve 98% precision and\n96%recall in distinguishing malware and benign repositories. Overall, our work\nis a fundamental building block for enabling many repository analysis functions\nsuch as repository categorization by target platform or intention, detecting\ncode-reuse and clones, and identifying lineage and evolution.\n"}
{"abstract": "  Sponsored search ads appear next to search results when people look for\nproducts and services on search engines. In recent years, they have become one\nof the most lucrative channels for marketing. As the fundamental basis of\nsearch ads, relevance modeling has attracted increasing attention due to the\nsignificant research challenges and tremendous practical value. Most existing\napproaches solely rely on the semantic information in the input query-ad pair,\nwhile the pure semantic information in the short ads data is not sufficient to\nfully identify user's search intents. Our motivation lies in incorporating the\ntremendous amount of unsupervised user behavior data from the historical search\nlogs as the complementary graph to facilitate relevance modeling. In this\npaper, we extensively investigate how to naturally fuse the semantic textual\ninformation with the user behavior graph, and further propose three novel\nAdsGNN models to aggregate topological neighborhood from the perspectives of\nnodes, edges and tokens. Furthermore, two critical but rarely investigated\nproblems, domain-specific pre-training and long-tail ads matching, are studied\nthoroughly. Empirically, we evaluate the AdsGNN models over the large industry\ndataset, and the experimental results of online/offline tests consistently\ndemonstrate the superiority of our proposal.\n"}
{"abstract": "  This paper highlights the trends in the field of predictive maintenance with\nthe use of machine learning. With the continuous development of the Fourth\nIndustrial Revolution, through IoT, the technologies that use artificial\nintelligence are evolving. As a result, industries have been using these\ntechnologies to optimize their production. Through scientific research\nconducted for this paper, conclusions were drawn about the trends in Predictive\nMaintenance applications with the use of machine learning bridging Artificial\nIntelligence and IoT. These trends are related to the types of industries in\nwhich Predictive Maintenance was applied, the models of artificial intelligence\nwere implemented, mainly of machine learning and the types of sensors that are\napplied through the IoT to the applications. Six sectors were presented and the\nproduction sector was dominant as it accounted for 54.54% of total\npublications. In terms of artificial intelligence models, the most prevalent\namong ten were the Artificial Neural Networks, Support Vector Machine and\nRandom Forest with 27.84%, 17.72% and 13.92% respectively. Finally, twelve\ncategories of sensors emerged, of which the most widely used were the sensors\nof temperature and vibration with percentages of 60.71% and 46.42%\ncorrespondingly.\n"}
{"abstract": "  Reward decomposition is a critical problem in centralized training with\ndecentralized execution~(CTDE) paradigm for multi-agent reinforcement learning.\nTo take full advantage of global information, which exploits the states from\nall agents and the related environment for decomposing Q values into individual\ncredits, we propose a general meta-learning-based Mixing Network with Meta\nPolicy Gradient~(MNMPG) framework to distill the global hierarchy for delicate\nreward decomposition. The excitation signal for learning global hierarchy is\ndeduced from the episode reward difference between before and after \"exercise\nupdates\" through the utility network. Our method is generally applicable to the\nCTDE method using a monotonic mixing network. Experiments on the StarCraft II\nmicromanagement benchmark demonstrate that our method just with a simple\nutility network is able to outperform the current state-of-the-art MARL\nalgorithms on 4 of 5 super hard scenarios. Better performance can be further\nachieved when combined with a role-based utility network.\n"}
{"abstract": "  The high-frequency part of spectrum of electromagnetic waves propagating\nparallel to the external magnetic field is considered for the macroscopically\nmotionless plasmas with the relativistic temperatures $T\\sim m_{e}c^{2}$, where\n$m_{e}$ is the mass of electron, $c$ is the speed of light. The analysis is\nbased on the novel hydrodynamic model based on four equations for the material\nfields which can be combined in two four vectors. These material fields are the\nconcentration and the velocity field \\emph{and} the average reverse\nrelativistic $\\gamma$ functor and the flux of the reverse relativistic $\\gamma$\nfunctor. In the nonrelativistic regime we have three waves (the ions are\nassumed to be motionless). Strong thermal effects lead to a coefficient in\nfront of cyclotron frequency which decreases the effective contribution of the\ncyclotron frequency. At $T=0.1m_{e}c^{2}$ we have a decrease of area of\nexistence of fast magneto-sound wave from the area of the large frequencies.\nWhile the area of existence of extraordinary waves becomes larger towards\nsmaller frequencies. The strong magnetic field limit $\\mid\\Omega_{e}\\mid >\n\\omega_{Le}$ additional wave appears with frequency below thermally decreased\ncyclotron frequency, where $\\mid\\Omega_{e}\\mid$ is the electron cyclotron\nfrequency, and $\\omega_{Le}$ is the Langmuir frequency. Further increase of\ntemperature leads to the disappearance of fast magneto-sound wave and to the\nconsiderable increase of area of existence of extraordinary towards smaller\nfrequencies.\n"}
{"abstract": "  Recently, spatiotemporal optical vortex pulses carrying a purely transverse\nintrinsic orbital angular momentum were generated experimentally [{\\it Optica}\n{\\bf 6}, 1547 (2019); {\\it Nat. Photon.} {\\bf 14}, 350 (2020)]. However, an\naccurate theoretical analysis of such states and their angular-momentum\nproperties remains elusive. Here we provide such analysis, including scalar and\nvector spatiotemporal Bessel-type solutions as well as descrption of their\npropagational, polarization, and angular-momentum properties. Most importantly,\nwe calculate both local densities and integral values of the spin and orbital\nangular momenta, and predict observable spin-orbit interaction phenomena\nrelated to the coupling between the trasnverse spin and orbital angular\nmomentum. Our analysis is readily extended to spatiotemporal vortex pulses of\nother natures (e.g., acoustic).\n"}
{"abstract": "  Story visualization is an under-explored task that falls at the intersection\nof many important research directions in both computer vision and natural\nlanguage processing. In this task, given a series of natural language captions\nwhich compose a story, an agent must generate a sequence of images that\ncorrespond to the captions. Prior work has introduced recurrent generative\nmodels which outperform text-to-image synthesis models on this task. However,\nthere is room for improvement of generated images in terms of visual quality,\ncoherence and relevance. We present a number of improvements to prior modeling\napproaches, including (1) the addition of a dual learning framework that\nutilizes video captioning to reinforce the semantic alignment between the story\nand generated images, (2) a copy-transform mechanism for\nsequentially-consistent story visualization, and (3) MART-based transformers to\nmodel complex interactions between frames. We present ablation studies to\ndemonstrate the effect of each of these techniques on the generative power of\nthe model for both individual images as well as the entire narrative.\nFurthermore, due to the complexity and generative nature of the task, standard\nevaluation metrics do not accurately reflect performance. Therefore, we also\nprovide an exploration of evaluation metrics for the model, focused on aspects\nof the generated frames such as the presence/quality of generated characters,\nthe relevance to captions, and the diversity of the generated images. We also\npresent correlation experiments of our proposed automated metrics with human\nevaluations. Code and data available at:\nhttps://github.com/adymaharana/StoryViz\n"}
{"abstract": "  Let $M$ be a topological space that admits a free involution $\\tau$, and let\n$N$ be a topological space. A homotopy class $\\beta \\in [ M,N ]$ is said to\nhave the Borsuk-Ulam property with respect to $\\tau$ if for every\nrepresentative map $f: M \\to N$ of $\\beta$, there exists a point $x \\in M$ such\nthat $f(\\tau(x))= f(x)$. In this paper, we determine the homotopy class of maps\nfrom the $2$-torus $T^2$ to the Klein bottle $K^2$ that possess the Borsuk-Ulam\nproperty with respect to any free involution of $T^2$ for which the orbit space\nis $K^2$. Our results are given in terms of a certain family of homomorphisms\ninvolving the fundamental groups of $T^2$ and $K^2$. This completes the\nanalysis of the Borsuk-Ulam problem for the case $M=T^2$ and $N=K^2$, and for\nany free involution $\\tau$ of $T^2$.\n"}
{"abstract": "  We explore the cosmological viability of a theory of gravity defined by the\nLagrangian $f(\\mathcal{R})=\\mathcal{R}^{n\\left(\\mathcal{R}\\right)}$ in the\nPalatini formalism, where $n\\left(\\mathcal{R}\\right)$ is a dimensionless\nfunction of the Palatini scalar curvature $\\mathcal{R}$ that interpolates\nbetween general relativity when $n\\left(\\mathcal{R}\\right)=1$ and a locally\nscale-invariant and superficially renormalizable theory when\n$n\\left(\\mathcal{R}\\right)=2$. We refer to this model as asymptotically\nWeyl-invariant gravity (AWIG).\n  We analyse perhaps the simplest possible implementation of AWIG. A phase\nspace analysis yields three fixed points with effective equation of states\ncorresponding to de Sitter, radiation and matter-dominated phases. An analysis\nof the deceleration parameter suggests our model is consistent with an early\nand late period of accelerated cosmic expansion, with an intermediate period of\ndecelerated expansion. We show that the model contains no obvious curvature\nsingularities. Therefore, AWIG appears to be cosmologically viable, at least\nfor the simple implementation explored.\n"}
{"abstract": "  Generating personalized responses is one of the major challenges in natural\nhuman-robot interaction. Current researches in this field mainly focus on\ngenerating responses consistent with the robot's pre-assigned persona, while\nignoring the user's persona. Such responses may be inappropriate or even\noffensive, which may lead to the bad user experience. Therefore, we propose a\nBilateral Personalized Dialogue Generation (BPDG) method for dyadic\nconversation, which integrates user and robot personas into dialogue generation\nvia designing a dynamic persona-aware fusion method. To bridge the gap between\nthe learning objective function and evaluation metrics, the Conditional Mutual\nInformation Maximum (CMIM) criterion is adopted with contrastive learning to\nselect the proper response from the generated candidates. Moreover, a bilateral\npersona accuracy metric is designed to measure the degree of bilateral\npersonalization. Experimental results demonstrate that, compared with several\nstate-of-the-art methods, the final results of the proposed method are more\npersonalized and consistent with bilateral personas in terms of both automatic\nand manual evaluations.\n"}
{"abstract": "  Multi-Pixel Photon Counter (MPPC) has been recently emerged and realized as a\ngreat type of Silicon Photomultiplier to replace or compensate for the\nconventional vacuum-based Photomultiplier tubes. MPPC provides many striking\nfeatures such as high electrical gain, outstanding photon detection efficiency,\nfast timing response, immunity to the magnetic field, low-voltage operation,\ncompactness, portability, and cost-effectiveness. The report introduces and\nexamines the electrical and optical characteristics of the MPPC under loosely\ncontrolled environmental conditions. Also, we report a measurement of the light\nyield captured by the MPPC when the cosmic ray passes through the plastic\nscintillator, demonstrating that such setup is suitable to build a simple,\ncost-effective tabletop cosmic-ray detector for educational and research\npurposes.\n"}
{"abstract": "  Along with the development of modern computing technology and social\nsciences, both theoretical research and practical applications of social\ncomputing have been continuously extended. In particular with the boom of\nartificial intelligence (AI), social computing is significantly influenced by\nAI. However, the conventional technologies of AI have drawbacks in dealing with\nmore complicated and dynamic problems. Such deficiency can be rectified by\nhybrid human-artificial intelligence (H-AI) which integrates both human\nintelligence and AI into one unity, forming a new enhanced intelligence. H-AI\nin dealing with social problems shows the advantages that AI can not surpass.\nThis paper firstly introduces the concept of H-AI. AI is the intelligence in\nthe transition stage of H-AI, so the latest research progresses of AI in social\ncomputing are reviewed. Secondly, it summarizes typical challenges faced by AI\nin social computing, and makes it possible to introduce H-AI to solve these\nchallenges. Finally, the paper proposes a holistic framework of social\ncomputing combining with H-AI, which consists of four layers: object layer,\nbase layer, analysis layer, and application layer. It represents H-AI has\nsignificant advantages over AI in solving social problems.\n"}
{"abstract": "  The compressive learning framework reduces the computational cost of training\non large-scale datasets. In a sketching phase, the data is first compressed to\na lightweight sketch vector, obtained by mapping the data samples through a\nwell-chosen feature map, and averaging those contributions. In a learning\nphase, the desired model parameters are then extracted from this sketch by\nsolving an optimization problem, which also involves a feature map. When the\nfeature map is identical during the sketching and learning phases, formal\nstatistical guarantees (excess risk bounds) have been proven.\n  However, the desirable properties of the feature map are different during\nsketching and learning (e.g. quantized outputs, and differentiability,\nrespectively). We thus study the relaxation where this map is allowed to be\ndifferent for each phase. First, we prove that the existing guarantees carry\nover to this asymmetric scheme, up to a controlled error term, provided some\nLimited Projected Distortion (LPD) property holds. We then instantiate this\nframework to the setting of quantized sketches, by proving that the LPD indeed\nholds for binary sketch contributions. Finally, we further validate the\napproach with numerical simulations, including a large-scale application in\naudio event classification.\n"}
{"abstract": "  Recently, we proposed a novel speaker diarization method called\nEnd-to-End-Neural-Diarization-vector clustering (EEND-vector clustering) that\nintegrates clustering-based and end-to-end neural network-based diarization\napproaches into one framework. The proposed method combines advantages of both\nframeworks, i.e. high diarization performance and handling of overlapped speech\nbased on EEND, and robust handling of long recordings with an arbitrary number\nof speakers based on clustering-based approaches. However, the method was only\nevaluated so far on simulated 2-speaker meeting-like data. This paper is to (1)\nreport recent advances we made to this framework, including newly introduced\nrobust constrained clustering algorithms, and (2) experimentally show that the\nmethod can now significantly outperform competitive diarization methods such as\nEncoder-Decoder Attractor (EDA)-EEND, on CALLHOME data which comprises real\nconversational speech data including overlapped speech and an arbitrary number\nof speakers. By further analyzing the experimental results, this paper also\ndiscusses pros and cons of the proposed method and reveals potential for\nfurther improvement. A set of the code to reproduce the results is available at\nhttps://github.com/nttcslab-sp/EEND-vector-clustering.\n"}
{"abstract": "  Regular sequences generalize the extensively studied automatic sequences. Let\n$S$ be an abstract numeration system. When the numeration language $L$ is\nprefix-closed and regular, a sequence is said to be $S$-regular if the module\ngenerated by its $S$-kernel is finitely generated.\n  In this paper, we give a new characterization of such sequences in terms of\nthe underlying numeration tree $T(L)$ whose nodes are words of $L$. We may\ndecorate these nodes by the sequence of interest following a breadth-first\nenumeration. For a prefix-closed regular language $L$, we prove that a sequence\nis $S$-regular if and only if the tree $T(L)$ decorated by the sequence is\nlinear, i.e., the decoration of a node depends linearly on the decorations of a\nfixed number of ancestors.\n  Next, we introduce and study regular sequences in a rational base numeration\nsystem, whose numeration language is known to be highly non-regular. We\nmotivate and comment our definition that a sequence is $\\frac{p}{q}$-regular if\nthe underlying numeration tree decorated by the sequence is linear. We give the\nfirst few properties of such sequences, we provide a few examples of them, and\nwe propose a method for guessing $\\frac{p}{q}$-regularity. Then we discuss the\nrelationship between $\\frac{p}{q}$-automatic sequences and\n$\\frac{p}{q}$-regular sequences. We finally present a graph directed linear\nrepresentation of a $\\frac{p}{q}$-regular sequence. Our study permits us to\nhighlight the places where the regularity of the numeration language plays a\npredominant role.\n"}
{"abstract": "  Mason and Remmel introduced a basis for quasisymmetric functions known as the\nrow-strict quasisymmetric Schur functions. This basis is generated\ncombinatorially by fillings of composition diagrams that are analogous to the\nrow-strict tableaux that generate Schur functions. We introduce a modification\nknown as Young row-strict quasisymmetric Schur functions, which are generated\nby row-strict Young composition fillings. After discussing basic combinatorial\nproperties of these functions, we define a skew Young row-strict quasisymmetric\nSchur function using the Hopf algebra of quasisymmetric functions and then\nprove this is equivalent to a combinatorial description. We also provide a\ndecomposition of the skew Young row-strict quasisymmetric Schur functions into\na sum of Gessel's fundamental quasisymmetric functions and prove a\nmultiplication rule for the product of a Young row-strict quasisymmetric Schur\nfunction and a Schur function.\n"}
{"abstract": "  The paper is devoted to the analysis of relationships between principal\nobjects of the spectral theory of dynamical systems (transfer and weighted\nshift operators) and basic characteristics of information theory and\nthermodynamic formalism (entropy and topological pressure). We present explicit\nformulae linking these objects with $t$-entropy and spectral potential.\nHerewith we uncover the role of inverse rami-rate, forward entropy along with\nessential set and the property of non-contractibility of a dynamical system.\n"}
{"abstract": "  We reconstruct the history of reionization using Gaussian process regression.\nUsing the UV luminosity data compilation from Hubble Frontiers Fields we\nreconstruct the redshift evolution of UV luminosity density and thereby the\nevolution of the source term in the ionization equation. This model-independent\nreconstruction rules out single power-law evolution of the luminosity density\nbut supports the logarithmic double power-law parametrization. We obtain\nreionization history by integrating ionization equations with the reconstructed\nsource term. Using optical depth constraint from Planck Cosmic Microwave\nBackground observation, measurement of UV luminosity function integrated till\ntruncation magnitude of -17 and -15, and derived ionization fraction from high\nredshift quasar, galaxies and gamma-ray burst observations, we constrain the\nhistory of reionization. In the conservative case we find the constraint on the\noptical depth as $\\tau =0.052\\pm0.001\\pm0.002$ at 68% and 95% confidence\nintervals. We find the redshift duration between 10% and 90% ionization to be\n$2.05_{-0.21-0.30}^{+0.11+0.37}$. Longer duration of reionization is supported\nif UV luminosity density data with truncation magnitude of -15 is used in the\njoint analysis. Our results point out that even in a conservative\nreconstruction, a combination of cosmological and astrophysical observations\ncan provide stringent constraints on the epoch of reionization.\n"}
{"abstract": "  The meme stock phenomenon is yet to be explored. In this note, we provide\nevidence that these stocks display common stylized facts on the dynamics of\nprice, trading volume, and social media activity. Using a regime-switching\ncointegration model, we identify the meme stock \"mementum\" which exhibits a\ndifferent characterization with respect to other stocks with high volumes of\nactivity (persistent and not) on social media. Understanding these properties\nhelps the investors and market authorities in their decision.\n"}
{"abstract": "  Existing speech enhancement methods mainly separate speech from noises at the\nsignal level or in the time-frequency domain. They seldom pay attention to the\nsemantic information of a corrupted signal. In this paper, we aim to bridge\nthis gap by extracting phoneme identities to help speech enhancement.\nSpecifically, we propose a phoneme-based distribution regularization (PbDr) for\nspeech enhancement, which incorporates frame-wise phoneme information into\nspeech enhancement network in a conditional manner. As different phonemes\nalways lead to different feature distributions in frequency, we propose to\nlearn a parameter pair, i.e. scale and bias, through a phoneme classification\nvector to modulate the speech enhancement network. The modulation parameter\npair includes not only frame-wise but also frequency-wise conditions, which\neffectively map features to phoneme-related distributions. In this way, we\nexplicitly regularize speech enhancement features by recognition vectors.\nExperiments on public datasets demonstrate that the proposed PbDr module can\nnot only boost the perceptual quality for speech enhancement but also the\nrecognition accuracy of an ASR system on the enhanced speech. This PbDr module\ncould be readily incorporated into other speech enhancement networks as well.\n"}
{"abstract": "  In this paper, we introduce a new distribution called Burr III-Weibull(BW)\ndistribution using the concept of competing risk. We derive moments,\nconditional moments, mean deviation and quantiles of the proposed distribution.\nAlso the Renyi's entropy and order statistics of the distribution are obtained.\nEstimation of parameters of the distribution is performed via maximum\nlikelihood method. A simulation study is performed to validate the maximum\nlikelihood estimator (MLE). A real practical data set is analyzed for\nillustration.\n"}
{"abstract": "  In this paper, we consider the problem of optimally allocating tasks,\nexpressed as global Linear Temporal Logic (LTL) specifications, to teams of\nheterogeneous mobile robots. The robots are classified in different types that\ncapture their different capabilities, and each task may require robots of\nmultiple types. The specific robots assigned to each task are immaterial, as\nlong as they are of the desired type. Given a discrete workspace, our goal is\nto design paths, i.e., sequences of discrete states, for the robots so that the\nLTL specification is satisfied. To obtain a scalable solution to this complex\ntemporal logic task allocation problem, we propose a hierarchical approach that\nfirst allocates specific robots to tasks using the information about the tasks\ncontained in the Nondeterministic Buchi Automaton (NBA) that captures the LTL\nspecification, and then designs low-level executable plans for the robots that\nrespect the high-level assignment. Specifically, we first prune and relax the\nNBA by removing all negative atomic propositions. This step is motivated by\n\"lazy collision checking\" methods in robotics and allows to simplify the\nplanning problem by checking constraint satisfaction only when needed. Then, we\nextract sequences of subtasks from the relaxed NBA along with their temporal\norders, and formulate a Mixed Integer Linear Program (MILP) to allocate these\nsubtasks to the robots. Finally, we define generalized multi-robot path\nplanning problems to obtain low-level executable robot plans that satisfy both\nthe high-level task allocation and the temporal constraints captured by the\nnegative atomic propositions in the original NBA. We provide theoretical\nresults showing completeness and soundness of our proposed method and present\nnumerical simulations demonstrating that our method can generate robot paths\nwith lower cost, considerably faster than existing methods.\n"}
{"abstract": "  Wikipedia is one of the main repositories of free knowledge available today,\nwith a central role in the Web ecosystem. For this reason, it can also be a\nbattleground for actors trying to impose specific points of view or even\nspreading disinformation online. There is a growing need to monitor its\n\"health\" but this is not an easy task. Wikipedia exists in over 300 language\neditions and each project is maintained by a different community, with their\nown strengths, weaknesses and limitations. In this paper, we introduce a\ntaxonomy of knowledge integrity risks across Wikipedia projects and a first set\nof indicators to assess internal risks related to community and content issues,\nas well as external threats such as the geopolitical and media landscape. On\ntop of this taxonomy, we offer a preliminary analysis illustrating how the lack\nof editors' geographical diversity might represent a knowledge integrity risk.\nThese are the first steps of a research project to build a Wikipedia Knowledge\nIntegrity Risk Observatory.\n"}
{"abstract": "  Keyphrase generation aims to summarize long documents with a collection of\nsalient phrases. Deep neural models have demonstrated a remarkable success in\nthis task, capable of predicting keyphrases that are even absent from a\ndocument. However, such abstractiveness is acquired at the expense of a\nsubstantial amount of annotated data. In this paper, we present a novel method\nfor keyphrase generation, AutoKeyGen, without the supervision of any human\nannotation. Motivated by the observation that an absent keyphrase in one\ndocument can appear in other places, in whole or in part, we first construct a\nphrase bank by pooling all phrases in a corpus. With this phrase bank, we then\ndraw candidate absent keyphrases for each document through a partial matching\nprocess. To rank both types of candidates, we combine their lexical- and\nsemantic-level similarities to the input document. Moreover, we utilize these\ntop-ranked candidates as to train a deep generative model for more absent\nkeyphrases. Extensive experiments demonstrate that AutoKeyGen outperforms all\nunsupervised baselines and can even beat strong supervised methods in certain\ncases.\n"}
{"abstract": "  Although Convolutional Neural Networks (CNNs) are widely used, their\ntranslation invariance (ability to deal with translated inputs) is still\nsubject to some controversy. We explore this question using\ntranslation-sensitivity maps to quantify how sensitive a standard CNN is to a\ntranslated input. We propose the use of Cosine Similarity as sensitivity metric\nover Euclidean Distance, and discuss the importance of restricting the\ndimensionality of either of these metrics when comparing architectures. Our\nmain focus is to investigate the effect of different architectural components\nof a standard CNN on that network's sensitivity to translation. By varying\nconvolutional kernel sizes and amounts of zero padding, we control the size of\nthe feature maps produced, allowing us to quantify the extent to which these\nelements influence translation invariance. We also measure translation\ninvariance at different locations within the CNN to determine the extent to\nwhich convolutional and fully connected layers, respectively, contribute to the\ntranslation invariance of a CNN as a whole. Our analysis indicates that both\nconvolutional kernel size and feature map size have a systematic influence on\ntranslation invariance. We also see that convolutional layers contribute less\nthan expected to translation invariance, when not specifically forced to do so.\n"}
{"abstract": "  We propose and investigate thermometry of a setup employing dual quantum dots\nwith staircase ground state configuration. The stair-case ground state\nconfiguration actuates thermally controlled inelastic tunnelling, which\ntranslates into a temperature sensitive conductance, thereby inducing\nthermometry. The performance of the set-up is then analyzed employing quantum\nmaster equation (QME) for such systems in the sequential tunnelling regime. In\nparticular, it is demonstrated that the system performance, in terms of\ntemperature sensitivity and efficiency, is maximum in the regime of low\ntemperature, making such system suitable for cryogenic thermometry. The\nproposed set-up can pave the path towards realization of high performance\ncryogenic nano temperature sensors.\n"}
{"abstract": "  Global coordination is required to solve a wide variety of challenging\ncollective action problems from network colorings to the tragedy of the\ncommons. Recent empirical study shows that the presence of a few noisy\nautonomous agents can greatly improve collective performance of humans in\nsolving networked color coordination games. To provide further analytical\ninsights into the role of behavioral randomness, here we study myopic\nartificial agents attempt to solve similar network coloring problems using\ndecision update rules that are only based on local information but allow random\nchoices at various stages of their heuristic reasonings. We consider that\nagents are distributed over a random bipartite network which is guaranteed to\nbe solvable with two colors. Using agent-based simulations and theoretical\nanalysis, we show that the resulting efficacy of resolving color conflicts is\ndependent on the specific implementation of random behavior of agents,\nincluding the fraction of noisy agents and at which decision stage noise is\nintroduced. Moreover, behavioral randomness can be finely tuned to the specific\nunderlying population structure such as network size and average network degree\nin order to produce advantageous results in finding collective coloring\nsolutions. Our work demonstrates that distributed greedy optimization\nalgorithms exploiting local information should be deployed in combination with\noccasional exploration via random choices in order to overcome local minima and\nachieve global coordination.\n"}
{"abstract": "  We present the ongoing NorLM initiative to support the creation and use of\nvery large contextualised language models for Norwegian (and in principle other\nNordic languages), including a ready-to-use software environment, as well as an\nexperience report for data preparation and training. This paper introduces the\nfirst large-scale monolingual language models for Norwegian, based on both the\nELMo and BERT frameworks. In addition to detailing the training process, we\npresent contrastive benchmark results on a suite of NLP tasks for Norwegian.\nFor additional background and access to the data, models, and software, please\nsee http://norlm.nlpl.eu\n"}
{"abstract": "  This paper presents a novel method, termed Bridge to Answer, to infer correct\nanswers for questions about a given video by leveraging adequate graph\ninteractions of heterogeneous crossmodal graphs. To realize this, we learn\nquestion conditioned visual graphs by exploiting the relation between video and\nquestion to enable each visual node using question-to-visual interactions to\nencompass both visual and linguistic cues. In addition, we propose bridged\nvisual-to-visual interactions to incorporate two complementary visual\ninformation on appearance and motion by placing the question graph as an\nintermediate bridge. This bridged architecture allows reliable message passing\nthrough compositional semantics of the question to generate an appropriate\nanswer. As a result, our method can learn the question conditioned visual\nrepresentations attributed to appearance and motion that show powerful\ncapability for video question answering. Extensive experiments prove that the\nproposed method provides effective and superior performance than\nstate-of-the-art methods on several benchmarks.\n"}
{"abstract": "  We show that every isometric action on a Cantor set is conjugate to an\ninverse limit of actions on finite sets; and that every isometric action by a\nfinitely generated amenable group is residually finite.\n"}
{"abstract": "  The Multiple Travelling Salesman Problem (MTSP) is among the most interesting\ncombinatorial optimization problems because it is widely adopted in real-life\napplications, including robotics, transportation, networking, etc. Although the\nimportance of this optimization problem, there is no survey dedicated to\nreviewing recent MTSP contributions. In this paper, we aim to fill this gap by\nproviding a comprehensive review of existing studies on MTSP. In this survey,\nwe focus on MTSP's recent contributions to both classical vehicles/robots and\nunmanned aerial vehicles. We highlight the approaches applied to solve the MTSP\nas well as its application domains. We analyze the MTSP variants and propose a\ntaxonomy and a classification of recent studies.\n"}
{"abstract": "  Milner (1984) introduced a process semantics for regular expressions as\nprocess graphs. Unlike for the language semantics, where every regular (that\nis, DFA-accepted) language is the interpretation of some regular expression,\nthere are finite process graphs that are not bisimilar to the process\ninterpretation of any regular expression. For reasoning about graphs that are\nexpressible by regular expressions modulo bisimilarity it is desirable to have\nstructural representations of process graphs in the image of the\ninterpretation.\n  For '1-free' regular expressions, their process interpretations satisfy the\nstructural property LEE (loop existence and elimination). But this is not in\ngeneral the case for all regular expressions, as we show by examples. Yet as a\nremedy, we describe the possibility to recover the property LEE for a close\nvariant of the process interpretation. For this purpose we refine the process\nsemantics of regular expressions to yield process graphs with 1-transitions,\nsimilar to silent moves for finite-state automata.\n"}
{"abstract": "  The temperature distribution in the battery significantly impacts the\nshort-term and long-term performance of battery systems. Therefore, efficient,\nsafe, and reliable battery system operation requires an accurate estimation of\nthe temperature field. The current industry standard for sensors to battery\ncell ratio is quite frugal. Thus, the problem of sensor placement for accurate\ntemperature estimation becomes non-trivial, especially for large-scale systems.\nIn this paper, we explore a greedy approach for sensor placement suitable for\nlarge-scale battery systems. An observer to estimate the thermal field is\ndesigned in an $\\mathcal{H}_{\\infty}$ framework while simultaneously minimizing\nthe sensor precisions, thus lowering the overall thermal management system's\neconomic cost.\n"}
{"abstract": "  In this paper we study a 2-type linear-fractional branching process in\nvarying environment with asymptotically constant mean matrices. Let $\\nu$ be\nthe extinction time and for $k\\ge1$ let $M_k$ be the mean matrix of offspring\ndistribution of individuals of the $(k-1)$-th generation. Under certain\nconditions, we show that $P(\\nu=n)$ and $P(\\nu>n)$ are asymptotically\nequivalent to some functions of products of spectral radii of the mean\nmatrices. This paper complements a former result [arXiv: 2007.07840] which\nrequires in addition a condition $\\forall k\\ge1,\\rm{det}(M_k)<-\\varepsilon$ for\nsome $\\varepsilon>0.$ Such a condition excludes a large class of mean matrices.\nAs byproducts, we also get some results on asymptotics of products of\nnonhomogeneous matrices which have their own interests.\n"}
{"abstract": "  When hexagonal boron nitride (hBN) and graphene are aligned at zero or small\ntwist angle, a moir\\'e structure is formed due to the small lattice constant\nmismatch between the two structures. In this work, we analyze magnetic ordering\ntendencies, driven by onsite Coulomb interactions, of encapsulated bilayer\ngraphene (BG) forming a moir\\'e structure with one (hBN-BG) or both hBN layers\n(hBN-BG-hBN), using the random phase approximation. The calculations are\nperformed in a fully atomistic Hubbard model that takes into account all\n$\\pi$-electrons of the carbon atoms in one moir\\'e unit cell. We analyze the\ncharge neutral case and find that the dominant magnetic ordering instability is\nuniformly antiferromagnetic. Furthermore, at low temperatures, the critical\nHubbard interaction $U_c$ required to induce magnetic order is slightly larger\nin those systems where the moir\\'e structure has caused a band gap opening in\nthe non-interacting picture, although the difference is less than 6%.\nMean-field calculations are employed to estimate how such an\ninteraction-induced magnetic order may change the observable single-particle\ngap sizes.\n"}
{"abstract": "  High-resolution Fourier-transform spectroscopy using table-top sources in the\nextreme ultraviolet (XUV) spectral range is still in its infancy. In this\ncontribution a significant advance is presented based on a Michelson-type\nall-reflective split-and-delay autocorrelator operating in a quasi amplitude\nsplitting mode. The autocorrelator works under a grazing incidence angle in a\nbroad spectral range $\\mathrm{(10\\,nm - 1\\,\\mu m)}$ providing collinear\npropagation of both pulse replicas and thus a constant phase difference across\nthe beam profile. The compact instrument allows for XUV pulse autocorrelation\nmeasurements in the time domain with a single-digit attosecond precision\nresulting in a resolution of $\\mathrm{E/\\Delta E=2000}$. Its performance for\nspectroscopic applications is demonstrated by characterizing a very sharp\nelectronic transition at $\\mathrm{26.6\\,eV}$ in Ar gas induced by the\n$\\mathrm{11^{th}}$ harmonic of a frequency-doubled Yb-fiber laser leading to\nthe characteristic $\\mathrm{3s3p^{6}4p^{1}P^{1}}$ Fano-resonance of Ar atoms.\nWe benchmark our time-domain interferometry results with a high-resolution XUV\ngrating spectrometer and find an excellent agreement. The common-path\ninterferometer opens up new opportunities for short-wavelength femtosecond and\nattosecond pulse metrology and dynamic studies on extreme time scales in\nvarious research fields.\n"}
{"abstract": "  Existing neural ranking models follow the text matching paradigm, where\ndocument-to-query relevance is estimated through predicting the matching score.\nDrawing from the rich literature of classical generative retrieval models, we\nintroduce and formalize the paradigm of deep generative retrieval models\ndefined via the cumulative probabilities of generating query terms. This\nparadigm offers a grounded probabilistic view on relevance estimation while\nstill enabling the use of modern neural architectures. In contrast to the\nmatching paradigm, the probabilistic nature of generative rankers readily\noffers a fine-grained measure of uncertainty. We adopt several current neural\ngenerative models in our framework and introduce a novel generative ranker\n(T-PGN), which combines the encoding capacity of Transformers with the Pointer\nGenerator Network model. We conduct an extensive set of evaluation experiments\non passage retrieval, leveraging the MS MARCO Passage Re-ranking and TREC Deep\nLearning 2019 Passage Re-ranking collections. Our results show the\nsignificantly higher performance of the T-PGN model when compared with other\ngenerative models. Lastly, we demonstrate that exploiting the uncertainty\ninformation of deep generative rankers opens new perspectives to\nquery/collection understanding, and significantly improves the cut-off\nprediction task.\n"}
{"abstract": "  Interpreting mathematical expressions at runtime is a standard task in\nscientific software engineering. There are different approaches to this problem\nfrom creating an embedded domain-specific language (eDSL) with its own parser\nand interpreter specifically for that task, to using a full-fledged embedded\ncompiler. This article is dedicated to a middle-ground solution implemented in\nthe KMath library, which uses the Kotlin object builder DSL and its own\nalgebraic abstractions to generate an AST for mathematical operations. This AST\nis then compiled just-in-time to generate JVM bytecode. A similar approach is\ntested on other Kotlin platforms, where its performance is compared across a\nvariety of supported platforms.\n"}
{"abstract": "  We demonstrate that multiple-Higgs production at the LHC is the most\nsensitive probe of first and second-generation quark flavor in the Higgs\nsector. In models where new scalars couple to light quarks, gigantic di-Higgs\nand even sizable tri-Higgs production rates can be obtained, which can be used\nto either discover or severely constrain such theories. As an example, we show\nthat the most stringent bounds on enhanced interactions of the\n$125\\,\\textrm{GeV}$ Higgs to the down quark in extended Higgs sectors are\nobtained by looking for the extra Higgs bosons that provide for such\nenhancements using the di-Higgs and $Zh$ topologies. In this context, we set\nnew limits on the 125 GeV Higgs coupling to the down quark as strong as\n$\\lambda_{hd\\bar{d}} \\lesssim 30 \\lambda_{hd\\bar{d}}^{\\textrm{SM}}$ -- a\ndramatic improvement over previously available bounds. Regarding\nsecond-generation quark flavor, we obtain new limits in the coupling to strange\nas strong as $\\lambda_{hs\\bar{s}} \\lesssim 10\n\\lambda_{hs\\bar{s}}^{\\textrm{SM}}$. In addition, we show that the currently\nunexplored triple-Higgs production topology could be a potential discovery\nchannel of a wide variety of extended Higgs sectors at the LHC, including not\nonly models where extra Higgses couple to light quarks, but also popular\ntheories where they have preferential couplings to the the top.\n"}
{"abstract": "  Building on the deep learning based acoustic echo cancellation (AEC) in the\nsingle-loudspeaker (single-channel) and single-microphone setup, this paper\ninvestigates multi-channel AEC (MCAEC) and multi-microphone AEC (MMAEC). We\ntrain a deep neural network (DNN) to predict the near-end speech from\nmicrophone signals with far-end signals used as additional information. We find\nthat the deep learning approach avoids the non-uniqueness problem in\ntraditional MCAEC algorithms. For the AEC setup with multiple microphones,\nrather than employing AEC for each microphone, a single DNN is trained to\nachieve echo removal for all microphones. Also, combining deep learning based\nAEC with deep learning based beamforming further improves the system\nperformance. Experimental results show the effectiveness of both bidirectional\nlong short-term memory (BLSTM) and convolutional recurrent network (CRN) based\nmethods for MCAEC and MMAEC. Furthermore, deep learning based methods are\ncapable of removing echo and noise simultaneously and work well in the presence\nof nonlinear distortions.\n"}
{"abstract": "  Semi-quantum inspired lightweight protocol is an important research issue in\nrealization of quantum protocols. However, the previous semi-quantum inspired\nlightweight mediated quantum key distribution (SQIL-MQKD) protocols need to use\nthe Bell states or measure the Bell states. The generation and measurement of\nBell states are more difficult and expensive than those of single photons. To\nsolve this problem, a semi-quantum inspired lightweight mediated quantum key\ndistribution with limited resource protocol is proposed. In this protocol, an\nuntrusted third party (TP) only needs to perform the quantum operations related\nto single photons and the participants only have to perform two quantum\noperations: (1) reflecting qubits without disturbance (2) performing unitary\noperations on single photons. In addition, this protocol is showed to be robust\nunder the collective attack.\n"}
{"abstract": "  In this work we propose a topological valley phononic crystal plate and we\nextensively investigate the refraction of valley modes into the surrounding\nhomogeneous medium. This phononic crystal includes two sublattices of\nresonators (A and B) modeled by mass-spring systems. We show that two edge\nstates confined at the AB/BA and BA/AB type domain walls exhibit different\nsymmetries in physical space and energy peaks in the Fourier space. As a\nresult, distinct refraction behaviors, especially through an armchair cut edge,\nare observed. On the other hand, the decay depth of these localized topological\nmodes, which is found to be solely determined by the relative resonant strength\nbetween the scatterers, significantly affects the refraction patterns. More\ninterestingly, the outgoing traveling wave through a zigzag interface becomes\nevanescent when operating at deep sub-wavelength scale. This is realized by\ntuning the average resonant strength. We show that the evanescent modes only\nexist along a particular type of outlet edge, and that they can couple with\nboth topological interface states. We also present two designs of topological\nfunctional devices, including an elastic one-way transmission waveguide and a\nnear-ideal monopole/dipole emitter, both based on our phononic structure.\n"}
{"abstract": "  As the role of algorithmic systems and processes increases in society, so\ndoes the risk of bias, which can result in discrimination against individuals\nand social groups. Research on algorithmic bias has exploded in recent years,\nhighlighting both the problems of bias, and the potential solutions, in terms\nof algorithmic transparency (AT). Transparency is important for facilitating\nfairness management as well as explainability in algorithms; however, the\nconcept of diversity, and its relationship to bias and transparency, has been\nlargely left out of the discussion. We reflect on the relationship between\ndiversity and bias, arguing that diversity drives the need for transparency.\nUsing a perspective-taking lens, which takes diversity as a given, we propose a\nconceptual framework to characterize the problem and solution spaces of AT, to\naid its application in algorithmic systems. Example cases from three research\ndomains are described using our framework.\n"}
{"abstract": "  Relocalization is a fundamental task in the field of robotics and computer\nvision. There is considerable work in the field of deep camera relocalization,\nwhich directly estimates poses from raw images. However, learning-based methods\nhave not yet been applied to the radar sensory data. In this work, we\ninvestigate how to exploit deep learning to predict global poses from Emerging\nFrequency-Modulated Continuous Wave (FMCW) radar scans. Specifically, we\npropose a novel end-to-end neural network with self-attention, termed RadarLoc,\nwhich is able to estimate 6-DoF global poses directly. We also propose to\nimprove the localization performance by utilizing geometric constraints between\nradar scans. We validate our approach on the recently released challenging\noutdoor dataset Oxford Radar RobotCar. Comprehensive experiments demonstrate\nthat the proposed method outperforms radar-based localization and deep camera\nrelocalization methods by a significant margin.\n"}
{"abstract": "  The development of next generation perovskite-based optoelectronic devices\nrelies critically on the understanding of the interaction between charge\ncarriers and the polar lattice in out-of-equilibrium conditions. While it has\nbecome increasingly evident for CsPbBr3 perovskites that the Pb-Br framework\nflexibility plays a key role in their light-activated functionality, the\ncorresponding local structural rearrangement has not yet been unambiguously\nidentified. In this work, we demonstrate that the photoinduced lattice changes\nin the system are due to a specific polaronic distortion, associated with the\nactivation of a longitudinal optical phonon mode at 18 meV by electron-phonon\ncoupling, and we quantify the associated structural changes with atomic-level\nprecision. Key to this achievement is the combination of time-resolved and\ntemperature-dependent studies at Br K-edge and Pb L3-edge X-ray absorption with\nrefined ab-initio simulations, which fully account for the screened core-hole\nfinal state effects on the X-ray absorption spectra. From the temporal\nkinetics, we show that carrier recombination reversibly unlocks the structural\ndeformation at both Br and Pb sites. The comparison with the\ntemperature-dependent XAS results rules out thermal effects as the primary\nsource of distortion of the Pb-Br bonding motif during photoexcitation. Our\nwork provides a comprehensive description of the CsPbBr3 perovskites\nphotophysics, offering novel insights on the light-induced response of the\nsystem and its exceptional optoelectronic properties.\n"}
{"abstract": "  NeRF synthesizes novel views of a scene with unprecedented quality by fitting\na neural radiance field to RGB images. However, NeRF requires querying a deep\nMulti-Layer Perceptron (MLP) millions of times, leading to slow rendering\ntimes, even on modern GPUs. In this paper, we demonstrate that real-time\nrendering is possible by utilizing thousands of tiny MLPs instead of one single\nlarge MLP. In our setting, each individual MLP only needs to represent parts of\nthe scene, thus smaller and faster-to-evaluate MLPs can be used. By combining\nthis divide-and-conquer strategy with further optimizations, rendering is\naccelerated by three orders of magnitude compared to the original NeRF model\nwithout incurring high storage costs. Further, using teacher-student\ndistillation for training, we show that this speed-up can be achieved without\nsacrificing visual quality.\n"}
{"abstract": "  We consider the category of Harish-Chandra modules for ${\\rm SL}_2(\\mathbb\nR)$ as a module over the category of finite-dimensional representations of\n${\\rm SL}(2)$ with respect to the tensor product. In this note we use classical\nresults about principal series to obtain a classification of\n$\\otimes$-submodules of the category of Harish-Chandra modules of ${\\rm\nSL}_2(\\mathbb R)$. The resulting classification recovers the classical\nclassification of irreducible Harish-Chandra modules. Our methods are expected\nto generalize to arbitrary reductive pairs and more general base fields.\n"}
{"abstract": "  The effectiveness of collective navigation of biological or artificial agents\nrequires to accommodate for contrasting requirements, such as staying in a\ngroup while avoiding close encounters and at the same time limiting the energy\nexpenditure for manoeuvring. Here, we address this problem by considering a\nsystem of active Brownian particles in a finite two-dimensional domain and ask\nwhat is the control that realizes the optimal tradeoff between collision\navoidance and control expenditure. We couch this problem in the language of\noptimal stochastic control theory and by means of a mean-field game approach we\nderive an analytic mean-field solution, characterized by a second-order phase\ntransition in the alignment order parameter. We find that a mean-field version\nof a classical model for collective motion based on alignment interactions\n(Vicsek model) performs remarkably close to the optimal control. Our results\nsubstantiate the view that observed group behaviors may be explained as the\nresult of optimizing multiple objectives and offer a theoretical ground for\nbiomimetic algorithms used for artificial agents.\n"}
{"abstract": "  The observation of electromagnetic radiation emitted or absorbed by matter\nwas instrumental in revealing the quantum properties of atoms and molecules in\nthe early XX century, and constituted a turning-point in the development of the\nquantum theory. Quantum mechanics changes dramatically the way radiation and\nmatter interact, making the probability of emission and absorption of light\nstrongly frequency dependent, as clearly manifested in atomic spectra. In this\nessay, we advocate that gravitational radiation can play, for the quantum\naspects of black holes, a similar role as electromagnetic radiation did for\natoms, and that the advent of gravitational-wave astronomy can bring this\nfascinating possibility to the realm of observations.\n"}
{"abstract": "  This paper addresses the abstract dynamic programming (DP) in the online\nscenario, where the abstract DP mapping is time-varying, instead of static. In\nthis case, optimal costs and policies at different time instants are not the\nsame in general, and the problem amounts to tracking time-varying optimal costs\nand policies, which is of interest to many practical problems. It is thus\nnecessary to analyze the performance of classical value iteration (VI) and\npolicy iteration (PI) algorithms in the online case. In doing so, this paper\ndevelops and provides the theoretical analysis for several online algorithms,\nincluding approximate online VI, online PI, approximate online PI, online\noptimistic PI, approximate online optimistic PI, and asynchronous online PI and\nVI algorithms. It is proved that the tracking error bounds for all algorithms\ncritically depend upon the largest difference between any two consecutive\nabstract mappings. Meanwhile, examples are presented to illustrate the\ntheoretical results.\n"}
{"abstract": "  News recommendation is critical for personalized news access. Existing news\nrecommendation methods usually infer users' personal interest based on their\nhistorical clicked news, and train the news recommendation models by predicting\nfuture news clicks. A core assumption behind these methods is that news click\nbehaviors can indicate user interest. However, in practical scenarios, beyond\nthe relevance between user interest and news content, the news click behaviors\nmay also be affected by other factors, such as the bias of news presentation in\nthe online platform. For example, news with higher positions and larger sizes\nare usually more likely to be clicked. The bias of clicked news may bring\nnoises to user interest modeling and model training, which may hurt the\nperformance of the news recommendation model.\n  In this paper, we propose a bias-aware personalized news recommendation\nmethod named DebiasRec, which can handle the bias information for more accurate\nuser interest inference and model training. The core of our method includes a\nbias representation module, a bias-aware user modeling module, and a bias-aware\nclick prediction module. The bias representation module is used to model\ndifferent kinds of news bias and their interactions to capture their joint\neffect on click behaviors. The bias-aware user modeling module aims to infer\nusers' debiased interest from the clicked news articles by using their bias\ninformation to calibrate the interest model. The bias-aware click prediction\nmodule is used to train a debiased news recommendation model from the biased\nclick behaviors, where the click score is decomposed into a preference score\nindicating user's interest in the news content and a news bias score inferred\nfrom its different bias features. Experiments on two real-world datasets show\nthat our method can effectively improve the performance of news recommendation.\n"}
{"abstract": "  The management of the anthropogenic water cycle must ensure the preservation\nof the quality and quantity of water resources and their careful allocation to\nthe different uses. Protection of water resources requires the control of\npollution sources that may deteriorate them. This is a challenging task in\nmulti-stressed catchments. This work presents an approach that combines\npesticide occurrence patterns and stable isotope analyses of nitrogen\n(delta15N-NO3-, delta15N-NH4+), oxygen (delta18O-NO3-), and boron (delta11B) to\ndiscriminate the origin of pesticides and nitrogenpollution to tackle this\nchallenge. The approach has been applied to a Mediterranean subcatchment\nsubject to a variety of natural and anthropogenic pressures. Combining the\nresults from both analytical approaches in selected locations of the basin, the\nurban/industrial activity was identified as the main pressure on the quality of\nthe surface water resources, and to a large extent also on the groundwater\nresources, although agriculture may play also an important role, mainly in\nterms of nitrate and ammonium pollution. Total pesticide concentrations in\nsurface waters were one order of magnitude higher than in groundwaters and\nbelieved to originate mainly from soil and/or sediments desorption processes\nand urban and industrial use, as they were mainly associated with treated\nwastewaters. These findings are supported by the stable isotope results, that\npointed to an organic origin of nitrate in surface waters and most groundwater\nsamples. Ammonium pollution observed in some aquifer locations is probably\ngenerated by nitrate reduction. Overall, no significant attenuation processes\ncould be inferred for nitrate pollution. The approach presented here\nexemplifies the investigative monitoring envisioned in the Water Framework\nDirective.\n"}
{"abstract": "  Nuclear magnetic resonance (NMR) schemes can be applied to micron-, and\nnanometer-sized samples by the aid of quantum sensors such as nitrogen-vacancy\n(NV) color centers in diamond. These minute devices allow for magnetometry of\nnuclear spin ensembles with high spatial and frequency resolution at ambient\nconditions, thus having a clear impact in different areas such as chemistry,\nbiology, medicine, and material sciences. In practice, NV quantum sensors are\ndriven by microwave (MW) control fields with a twofold objective: On the one\nhand, MW fields bridge the energy gap between NV and nearby nuclei which\nenables a coherent and selective coupling among them while, on the other hand,\nMW fields remove environmental noise on the NV leading to enhanced\ninterrogation time. In this work we review distinct MW radiation patterns, or\ndynamical decoupling techniques, for nanoscale NMR applications.\n"}
{"abstract": "  Explaining the outcome of deep learning decisions based on affect is\nchallenging but necessary if we expect social companion robots to interact with\nusers on an emotional level. In this paper, we present a commonsense approach\nthat utilizes an interpretable hybrid neural-symbolic system to associate\nextracted targets, noun chunks determined to be associated with the expressed\nemotion, with affective labels from a natural language expression. We leverage\na pre-trained neural network that is well adapted to tree and sub-tree\nprocessing, the Dependency Tree-LSTM, to learn the affect labels of dynamic\ntargets, determined through symbolic rules, in natural language. We find that\nmaking use of the unique properties of the recursive network provides higher\naccuracy and interpretability when compared to other unstructured and\nsequential methods for determining target-affect associations in an\naspect-based sentiment analysis task.\n"}
{"abstract": "  Modeling the Pauli energy, the contribution to the kinetic energy caused by\nPauli statistics, without using orbitals is the open problem of orbital-free\ndensity functional theory. An important aspect of this problem is correctly\nreproducing the Pauli potential, the response of the Pauli kinetic energy to a\nchange in density. We analyze the behavior of the Pauli potential of\nnon-relativistic neutral atoms under Lieb-Simon scaling -- the process of\ntaking nuclear charge and particle number to infinity, in which the kinetic\nenergy tends to the Thomas-Fermi limit. We do this by mathematical analysis of\nthe near-nuclear region and by calculating the exact orbital-dependent Pauli\npotential using the approach of Ouyang and Levy for closed-shell atoms out to\nelement Z=976. In rough analogy to Lieb and Simon's own findings for the charge\ndensity, we find that the potential does not converge smoothly to the\nThomas-Fermi limit on a point-by-point basis but separates into several\ndistinct regions of behavior. Near the nucleus, the potential approaches a\nconstant given by the difference in energy between the lowest and highest\noccupied eigenvalues. We discover a transition region in the outer core where\nthe potential deviates unexpectedly and predictably from both the Thomas-Fermi\npotential and the gradient expansion correction to it. These results may\nprovide insight into semi-classical description of Pauli statistics, and new\nconstraints to aid the improvement of orbital-free DFT functionals.\n"}
{"abstract": "  The EDGES collaboration has reported the detection of a global 21-cm signal\nwith a plateau centered at 76 MHz (i.e., redshift 17.2), with an amplitude of\n500^(+200)_(-500) mK. This anomalous measurement does not comport with standard\ncosmology, which can only accommodate an amplitude < 230 mK. Nevertheless, the\nline profile's redshift range (15 < z < 20) suggests a possible link to Pop III\nstar formation and an implied evolution out of the `dark ages.' Given this\ntension with the standard model, we here examine whether the observed 21-cm\nsignal is instead consistent with the results of recent modeling based on the\nalternative Friedmann-Lemaitre-Robertson-Walker cosmology known as the R_h=ct\nuniverse, showing that--in this model--the CMB radiation might have been\nrethermalized by dust ejected into the IGM by the first-generation stars at\nredshift z < 16. We find that the requirements for this process to have\noccurred would have self-consistently established an equilibrium spin\ntemperature T_s~3.4 K in the neutral hydrogen, via the irradiation of the IGM\nby deep penetrating X-rays emitted at the termination shocks of Pop III\nsupernova remnants. Such a dust scenario has been strongly ruled out for the\nstandard model, so the spin temperature (~3.3 K) inferred from the 21-cm\nabsorption feature appears to be much more consistent with the R_h=ct profile\nthan that implied by LCDM, for which adiabatic cooling would have established a\nspin temperature T_s(z=17.2)~6 K.\n"}
{"abstract": "  Quantum key distribution (QKD) has been developed for decades and several\ndifferent QKD protocols have been proposed. But two difficulties limit the\nimplementation of most QKD protocols. First, the involved participants are\nrequired to have heavy quantum capabilities, such as quantum joint operation,\nquantum register, and so on. Second, a hypothetical authenticated classical\nchannel is used in most of the existing QKD protocols and this assumed channel\ndoes not exist in reality. To solve both the above limitations at the same\ntime, this study proposes three lightweight authenticated QKD protocols with\nkey recycling and shows these proposed protocols are robust under the\ncollective attack.\n"}
{"abstract": "  Let $\\Lambda$ be a finite-dimensional algebra. A wide subcategory of\n$\\mathsf{mod}\\Lambda$ is called left finite if the smallest torsion class\ncontaining it is functorially finite. In this paper, we prove that the wide\nsubcategories of $\\mathsf{mod}\\Lambda$ arising from $\\tau$-tilting reduction\nare precisely the Serre subcategories of left finite wide subcategories. As a\nconsequence, we show that the class of such subcategories is closed under\nfurther $\\tau$-tilting reduction. This leads to a natural way to extend the\ndefinition of the \"$\\tau$-cluster morphism category\" of $\\Lambda$ to arbitrary\nfinite-dimensional algebras. This category was recently constructed by\nBuan-Marsh in the $\\tau$-tilting finite case and by Igusa-Todorov in the\nhereditary case.\n"}
{"abstract": "  Despite considerable advancements with deep neural language models (LMs),\nneural text generation still suffers from degeneration: the generated text is\nrepetitive, generic, self-contradictory, and often lacks commonsense. Our\nanalyses on sentence-level attention patterns in LMs reveal that neural\ndegeneration may be associated with insufficient learning of task-specific\ncharacteristics by the attention mechanism. This finding motivates on-the-fly\nattention modulation -- a simple but effective method that enables the\ninjection of priors into attention computation during inference. Automatic and\nhuman evaluation results on three text generation benchmarks demonstrate that\nattention modulation helps LMs generate text with enhanced fluency, creativity,\nand commonsense reasoning, in addition to significantly reduce sentence-level\nrepetition.\n"}
{"abstract": "  Community-level bans are a common tool against groups that enable online\nharassment and harmful speech. Unfortunately, the efficacy of community bans\nhas only been partially studied and with mixed results. Here, we provide a\nflexible unsupervised methodology to identify in-group language and track user\nactivity on Reddit both before and after the ban of a community (subreddit). We\nuse a simple word frequency divergence to identify uncommon words\noverrepresented in a given community, not as a proxy for harmful speech but as\na linguistic signature of the community. We apply our method to 15 banned\nsubreddits, and find that community response is heterogeneous between\nsubreddits and between users of a subreddit. Top users were more likely to\nbecome less active overall, while random users often reduced use of in-group\nlanguage without decreasing activity. Finally, we find some evidence that the\neffectiveness of bans aligns with the content of a community. Users of dark\nhumor communities were largely unaffected by bans while users of communities\norganized around white supremacy and fascism were the most affected.\nAltogether, our results show that bans do not affect all groups or users\nequally, and pave the way to understanding the effect of bans across\ncommunities.\n"}
{"abstract": "  Goal-conditioned reinforcement learning endows an agent with a large variety\nof skills, but it often struggles to solve tasks that require more temporally\nextended reasoning. In this work, we propose to incorporate imagined subgoals\ninto policy learning to facilitate learning of complex tasks. Imagined subgoals\nare predicted by a separate high-level policy, which is trained simultaneously\nwith the policy and its critic. This high-level policy predicts intermediate\nstates halfway to the goal using the value function as a reachability metric.\nWe don't require the policy to reach these subgoals explicitly. Instead, we use\nthem to define a prior policy, and incorporate this prior into a KL-constrained\npolicy iteration scheme to speed up and regularize learning. Imagined subgoals\nare used during policy learning, but not during test time, where we only apply\nthe learned policy. We evaluate our approach on complex robotic navigation and\nmanipulation tasks and show that it outperforms existing methods by a large\nmargin.\n"}
{"abstract": "  State-of-the-art training algorithms for deep learning models are based on\nstochastic gradient descent (SGD). Recently, many variations have been\nexplored: perturbing parameters for better accuracy (such as in Extragradient),\nlimiting SGD updates to a subset of parameters for increased efficiency (such\nas meProp) or a combination of both (such as Dropout). However, the convergence\nof these methods is often not studied in theory. We propose a unified\ntheoretical framework to study such SGD variants -- encompassing the\naforementioned algorithms and additionally a broad variety of methods used for\ncommunication efficient training or model compression. Our insights can be used\nas a guide to improve the efficiency of such methods and facilitate\ngeneralization to new applications. As an example, we tackle the task of\njointly training networks, a version of which (limited to sub-networks) is used\nto create Slimmable Networks. By training a low-rank Transformer jointly with a\nstandard one we obtain superior performance than when it is trained separately.\n"}
{"abstract": "  During the evolution of coupled nonlinear oscillators on a lattice, with\ndynamics dictated by the discrete nonlinear Schr\\\"odinger equation (DNLSE\nsystems), two quantities are conserved: system energy (Hamiltonian) and system\ndensity (number of particles). If the number of system oscillators is large\nenough, a significant portion of the array can be considered to be an \"open\nsystem\", in intimate energy and density contact with a \"bath\" - the rest of the\narray. Thus, as indicated in previous works, the grand canonical formulation\ncan be exploited in order to determine equilibrium statistical properties of\nthermalized DNLSE systems. In this work, given the values of the two conserved\nquantities, we have calculated the necessary values of the two Lagrange\nparameters (typically designated $\\beta,\\mu$) associated with the grand\ncanonical partition function in two different ways. One is numerical and the\nother is analytic, based on a published approximate entropy expression. In\naddition we have accessed a purposely-derived approximate PDF expression of\nsite-densities. Applying these mathematical tools we have generated maps of\ntemperatures, chemical potentials, and field correlations for DNLSE systems\nover the entire thermalization zone of the DNLSE phase diagram, subjected to\nall system-nonlinearity levels. The end result is a rather complete picture,\ncharacterizing equilibrated large DNLSE systems.\n"}
{"abstract": "  We present an overview of a proposal in relativistic proton-proton ($pp$)\ncollisions emphasizing the thermal or kinetic freeze-out stage in the framework\nof the Tsallis distribution. In this paper we take into account the chemical\npotential present in the Tsallis distribution by following a two step\nprocedure. In the first step we used the redundancy present in the variables\nsuch as the system temperature, $T$, volume, $V$, Tsallis exponent, $q$,\nchemical potential, $\\mu$, and performed all fits by effectively setting to\nzero the chemical potential. In the second step the value $q$ is kept fixed at\nthe value determined in the first step. This way the complete set of variables\n$T, q, V$ and $\\mu$ can be determined. The final results show a weak energy\ndependence in $pp$ collisions at the centre-of-mass energy $\\sqrt{s}= 6$ GeV to\n13 TeV. The chemical potential $\\mu$ at kinetic freeze-out shows an increase\nwith beam energy. This simplifies the description of the thermal freeze-out\nstage in $pp$ collisions as the values of $T$ and of the freeze-out radius $R$\nvary only mildly over a wide range of beam energies.\n"}
{"abstract": "  We prove a general Mosco convergence theorem for bounded Euclidean domains\nsatisfying a set of mild geometric hypotheses. For bounded domains, this notion\nimplies norm-resolvent convergence for the Dirichlet Laplacian which in turn\nensures spectral convergence. A key element of the proof is the development of\na novel, explicit Poincar\\'e-type inequality. These results allow us to\nconstruct a universal algorithm capable of computing the eigenvalues of the\nDirichlet Laplacian on a wide class of rough domains. Many domains with fractal\nboundaries, such as the Koch snowflake and certain filled Julia sets, are\nincluded among this class. Conversely, we construct a counter example showing\nthat there does not exist a universal algorithm of the same type capable of\ncomputing the eigenvalues of the Dirichlet Laplacian on an arbitrary bounded\ndomain.\n"}
{"abstract": "  Two dimensional layered van der Waals (vdW) magnets have demonstrated their\npotential to study both fundamental and applied physics due to their remarkable\nelectronic properties. However, the connection of vdW magnets to spintronics as\nwell as quantum information science is not clear. In particular, it remains\nelusive whether there are novel magnetic phenomena only belonging to vdW\nmagnets, but absent in the widely studied crystalline magnets. Here we consider\nthe quantum correlations of magnons in a layered vdW magnet and identify an\nentanglement channel of magnons across the magnetic layers, which can be\neffectively tuned and even deterministically switched on and off by both\nmagnetic and electric means. This is a unique feature of vdW magnets in which\nthe underlying physics is well understood in terms of the competing roles of\nexchange and anisotropy fields that contribute to the magnon excitation.\nFurthermore, we show that such a tunable entanglement channel can mediate the\nelectrically controllable entanglement of two distant qubits, which also\nprovides a protocol to indirectly measure the entanglement of magnons. Our\nfindings provide a novel avenue to electrically manipulate the qubits and\nfurther open up new opportunities to utilize vdW magnets for quantum\ninformation science.\n"}
{"abstract": "  A detailed study of the BL Lacertae PKS 0903-57 has been done for the first\ntime with 12 years of Fermi Large Area Telescope data. We have identified two\nbright gamma-ray flares in 2018 and 2020. Many sub-structures were observed\nduring multiple time binning of these flares. We have performed detailed\ntemporal and spectral study on all the sub-structures separately. A single-zone\nemission model is used for time-dependent leptonic modeling of the\nmulti-wavelength spectral energy distributions. Our estimated values of\nvariability time scale, magnetic field in the emission region, jet power\nobtained from leptonic modeling of PKS 0903-57 are presented in this work.\nCurrently, we have a minimal number of observations in X-rays and other bands.\nHence, more simultaneous multi-wavelength monitoring of this source is required\nto have a better understanding of the physical processes happening in the jet\nof the blazar PKS 0903-57.\n"}
{"abstract": "  We theoretically studied the quantum Cram\\'{e}r-Rao bound of an actively\ncorrelated Mach-Zehnder interferometer (ACMZI), where the quantum Fisher\ninformation obtained by the phase-averaging method can give the proper\nphase-sensing limit without any external phase reference. We numerically\ncalculate the phase sensitivities with the method of homodyne detection and\nintensity detection in the presence of losses. Under lossless and very low loss\nconditions, the ACMZI is operated in a balanced case to beat the standard\nquantum limit (SQL). As the loss increases, the reduction in sensitivity\nincreases. However within a certain range, we can adjust the gain parameters of\nthe beam recombination process to reduce the reduction in sensitivity and\nrealize the sensitivity can continue to beat the SQL in an unbalanced\nsituation. Our scheme provides an optimization method of phase estimation in\nthe presence of losses.\n"}
{"abstract": "  The infinite Atlas model describes a countable system of competing Brownian\nparticles where the lowest particle gets a unit upward drift and the rest\nevolve as standard Brownian motions. The stochastic process of gaps between the\nparticles in the infinite Atlas model does not have a unique stationary\ndistribution and in fact for every $a \\ge 0$, $\\pi_a :=\n\\bigotimes_{i=1}^{\\infty} \\operatorname{Exp}(2 + ia)$ is a stationary measure\nfor the gap process. We say that an initial distribution of gaps is in the weak\ndomain of attraction of the stationary measure $\\pi_a$ if the time averaged\nlaws of the stochastic process of the gaps, when initialized using that\ndistribution, converge to $\\pi_a$ weakly in the large time limit. We provide\ngeneral sufficient conditions on the initial gap distribution of the Atlas\nparticles for it to lie in the weak domain of attraction of $\\pi_a$ for each\n$a\\ge 0$. The cases $a=0$ and $a>0$ are qualitatively different as is seen from\nthe analysis and the sufficient conditions that we provide. Proofs are based on\nthe analysis of synchronous couplings, namely, couplings of the ranked particle\nsystems started from different initial configurations, but driven using the\nsame set of Brownian motions.\n"}
{"abstract": "  While the large paramagnetic response measured in certain ensembles of\nmetallic nanoparticles has been assigned to orbital effects of conduction\nelectrons, the spin-orbit coupling has been pointed out as a possible origin of\nthe anomalously large diamagnetic response observed in other cases. Such a\nrelativistic effect, arising from the inhomogeneous electrostatic potential\nseen by the conduction electrons, might originate from the host ionic lattice,\nimpurities, or the self-consistent confining potential. Here we theoretically\ninvestigate the effect of the spin-orbit coupling arising from the confining\npotential, quantifying its contribution to the zero-field magnetic\nsusceptibility and gauging it against the ones generated by other\nweakly-relativistic corrections. Two ideal geometries are considered in detail,\nthe sphere and the half-sphere, focusing on the expected increased role of the\nspin-orbit coupling upon a symmetry reduction, and the application of these\nresults to actual metallic nanoparticles is discussed. The matrix elements of\nthe different weakly-relativistic corrections are obtained and incorporated in\na perturbative treatment of the magnetic field, leading to tractable\nsemi-analytical and semiclassical expressions for the case of the sphere, while\na numerical treatment becomes necessary for the half-sphere. The correction to\nthe zero-field susceptibility arising from the spin-orbit coupling in a single\nsphere is quite small, and it is dominated by the weakly-relativistic kinetic\nenergy correction, which in turn remains considerably smaller than the typical\nvalues of the nonrelativistic zero-field susceptibility. Moreover, the\nspin-orbit contribution to the average response for ensembles of nanoparticles\nwith a large size dispersion is shown to vanish. The symmetry reduction in\ngoing from the single sphere to the half-sphere does not translate into a\nsignificant (...)\n"}
{"abstract": "  In this paper we consider new geometric flow equations, called D-flow, which\ndescribe the variation of space-time geometries under the change of the number\nof dimensions. The D-flow is originating from the non-trivial dependence of the\nvolume of space-time manifolds on the number of space-time dimensions and it is\ndriven by certain curvature invariants. We will work out specific examples of\nD-flow equations and their solutions for the case of D-dimensional spheres and\nFreund-Rubin Compactification. The discussion of the paper is motivated from\nrecent swampland considerations, where the number $D$ of space-time dimensions\nis treated as a new swampland parameter.\n"}
{"abstract": "  Simulation-based optimization using agent-based models is typically carried\nout under the assumption that the gradient describing the sensitivity of the\nsimulation output to the input cannot be evaluated directly. To still apply\ngradient-based optimization methods, which efficiently steer the optimization\ntowards a local optimum, gradient estimation methods can be employed. However,\nmany simulation runs are needed to obtain accurate estimates if the input\ndimension is large. Automatic differentiation (AD) is a family of techniques to\ncompute gradients of general programs directly. Here, we explore the use of AD\nin the context of time-driven agent-based simulations. By substituting common\ndiscrete model elements such as conditional branching with smooth\napproximations, we obtain gradient information across discontinuities in the\nmodel logic. On the example of microscopic traffic models and an epidemics\nmodel, we study the fidelity and overhead of the differentiable models, as well\nas the convergence speed and solution quality achieved by gradient-based\noptimization compared to gradient-free methods. In traffic signal timing\noptimization problems with high input dimension, the gradient-based methods\nexhibit substantially superior performance. Finally, we demonstrate that the\napproach enables gradient-based training of neural network-controlled\nsimulation entities embedded in the model logic.\n"}
{"abstract": "  Since Transformer has found widespread use in NLP, the potential of\nTransformer in CV has been realized and has inspired many new approaches.\nHowever, the computation required for replacing word tokens with image patches\nfor Transformer after the tokenization of the image is vast(e.g., ViT), which\nbottlenecks model training and inference. In this paper, we propose a new\nattention mechanism in Transformer termed Cross Attention, which alternates\nattention inner the image patch instead of the whole image to capture local\ninformation and apply attention between image patches which are divided from\nsingle-channel feature maps capture global information. Both operations have\nless computation than standard self-attention in Transformer. By alternately\napplying attention inner patch and between patches, we implement cross\nattention to maintain the performance with lower computational cost and build a\nhierarchical network called Cross Attention Transformer(CAT) for other vision\ntasks. Our base model achieves state-of-the-arts on ImageNet-1K, and improves\nthe performance of other methods on COCO and ADE20K, illustrating that our\nnetwork has the potential to serve as general backbones. The code and models\nare available at \\url{https://github.com/linhezheng19/CAT}.\n"}
{"abstract": "  Sequential whole-body 18F-Fluorodeoxyglucose (FDG) positron emission\ntomography (PET) scans are regarded as the imaging modality of choice for the\nassessment of treatment response in the lymphomas because they detect treatment\nresponse when there may not be changes on anatomical imaging. Any computerized\nanalysis of lymphomas in whole-body PET requires automatic segmentation of the\nstudies so that sites of disease can be quantitatively monitored over time.\nState-of-the-art PET image segmentation methods are based on convolutional\nneural networks (CNNs) given their ability to leverage annotated datasets to\nderive high-level features about the disease process. Such methods, however,\nfocus on PET images from a single time-point and discard information from other\nscans or are targeted towards specific organs and cannot cater for the multiple\nstructures in whole-body PET images. In this study, we propose a\nspatio-temporal 'dual-stream' neural network (ST-DSNN) to segment sequential\nwhole-body PET scans. Our ST-DSNN learns and accumulates image features from\nthe PET images done over time. The accumulated image features are used to\nenhance the organs / structures that are consistent over time to allow easier\nidentification of sites of active lymphoma. Our results show that our method\noutperforms the state-of-the-art PET image segmentation methods.\n"}
{"abstract": "  Building on work by Chabauty from 1941, Coleman proved in 1985 an explicit\nbound for the number of rational points of a curve $C$ of genus $g\\ge 2$\ndefined over a number field $F$, with Jacobian of rank at most $g-1$. Namely,\nin the case $F=\\mathbb{Q}$, if $p>2g$ is a prime of good reduction, then the\nnumber of rational points of $C$ is at most the number of $\\mathbb{F}_p$-points\nplus a contribution coming from the canonical class of $C$. We prove a result\nanalogous to Coleman's bound in the case of a hyperbolic surface $X$ over a\nnumber field, embedded in an abelian variety $A$ of rank at most one, under\nsuitable conditions on the reduction type at the auxiliary prime. This provides\nthe first extension of Coleman's explicit bound beyond the case of curves. The\nmain innovation in our approach is a new method to study the intersection of a\n$p$-adic analytic subgroup with a subvariety of $A$ by means of overdetermined\nsystems of differential equations in positive characteristic.\n"}
{"abstract": "  Automatic Identification System (AIS) data represents a rich source of\ninformation about maritime traffic and offers a great potential for data\nanalytics and predictive modeling solutions, which can help optimizing logistic\nchains and to reduce environmental impacts. In this work, we address the main\nlimitations of the validity of AIS navigational data fields, by proposing a\nmachine learning-based data-driven methodology to detect and (to the possible\nextent) also correct erroneous data. Additionally, we propose a metric that can\nbe used by vessel operators and ports to express numerically their business and\nenvironmental efficiency through time and spatial dimensions, enabled with the\nobtained validated AIS data. We also demonstrate Port Area Vessel Movements\n(PARES) tool, which demonstrates the proposed solutions.\n"}
{"abstract": "  Transformers have offered a new methodology of designing neural networks for\nvisual recognition. Compared to convolutional networks, Transformers enjoy the\nability of referring to global features at each stage, yet the attention module\nbrings higher computational overhead that obstructs the application of\nTransformers to process high-resolution visual data. This paper aims to\nalleviate the conflict between efficiency and flexibility, for which we propose\na specialized token for each region that serves as a messenger (MSG). Hence, by\nmanipulating these MSG tokens, one can flexibly exchange visual information\nacross regions and the computational complexity is reduced. We then integrate\nthe MSG token into a multi-scale architecture named MSG-Transformer. In\nstandard image classification and object detection, MSG-Transformer achieves\ncompetitive performance and the inference on both GPU and CPU is accelerated.\nCode is available at https://github.com/hustvl/MSG-Transformer.\n"}
{"abstract": "  Terahertz (THz) technology is promising in several applications such as\nimaging, spectroscopy and communications. Among several methods in the\ngeneration and detection of THz waves, a THz time domain system (TDS) that is\ndeveloped using photoconductive antennas (PCA) as emitter and detector presents\nseveral advantages such as simple alignment, low cost, high performance etc. In\nthis work, we report the design, fabrication and characterization of a 2-D PCA\narray that is capable of detecting both the amplitude and phase of the THz\npulse. The PCA array is fabricated using LT-GaAs and has 8 channels with 64\npixels (8x8). The infrared probe beam is steered and focused towards each pixel\nof the PCA array using a spatial light modulator (SLM). The measured\nphotocurrent (amplitude and phase) from each channel is recorded separately and\nthe frequencies up to 1.4 THz can be detected. Furthermore, the parameters such\nas directional time delay of the THz pulse, crosstalk between the channels\netc., were characterized. Finally, we show that the proposed 2D PCA array\ndesign is flexible and can be used for accelerated THz spectral image\nacquisition.\n"}
{"abstract": "  Through lockdowns and other severe changes to daily life, almost everyone is\naffected by the COVID-19 pandemic. Scientists and medical doctors are - among\nothers - mainly interested in researching, monitoring, and improving physical\nand mental health of the general population. Mobile health apps (mHealth), and\napps conducting ecological momentary assessments (EMA) respectively, can help\nin this context. However, developing such mobile applications poses many\nchallenges like costly software development efforts, strict privacy rules,\ncompliance with ethical guidelines, local laws, and regulations. In this paper,\nwe present TrackYourHealth (TYH), a highly configurable, generic, and modular\nmobile data collection and EMA platform, which enabled us to develop and\nrelease two mobile multi-platform applications related to COVID-19 in just a\nfew weeks. We present TYH and highlight specific challenges researchers and\ndevelopers of similar apps may also face, especially when developing apps\nrelated to the medical field.\n"}
{"abstract": "  In this paper we show that a third order PDE system that is a general form of\na CR-geometry PDE system has at most a ten-dimensional Lie symmetry algebra. We\nalso show that this estimate is precise.\n"}
{"abstract": "  Radio Active Galactic Nuclei (RAGNs) are mainly found in dense structures\n(i.e., clusters/groups) at redshifts of z$<$2 and are commonly used to detect\nprotoclusters at higher redshift. Here, we attempt to study the host and\nenvironmental properties of two relatively faint ($\\mathrm L_\\mathrm{1.4GHz}\n\\sim10^{25}$ W Hz$^{-1}$) RAGNs in a known protocluster at z=3.3 in the PCl\nJ0227-0421 field, detected using the latest radio observation obtained as part\nof the Observations of Redshift Evolution in Large-Scale Environments (ORELSE)\nSurvey. Using new spectroscopic observations obtained from Keck/MOSFIRE as part\nof the Charting Cluster Construction with the VIMOS Ultra-Deep Survey (VUDS)\nand ORELSE (C3VO) survey and previous spectroscopic data obtained as part of\nthe VIMOS-VLT Deep Survey (VVDS) and VUDS, we revise the three-dimensional\noverdensity field around this protocluster. The protocluster is embedded in a\nlarge scale overdensity protostructure. This protostructure has an estimated\ntotal mass of $\\sim$2.6$\\times10^{15} M_\\odot$ and contains several overdensity\npeaks. Both RAGNs are hosted by very bright and massive galaxies, while their\nhosts show extreme differences color, indicating that they have different ages\nand are in different evolutionary stages. Furthermore, we find that they are\nnot in the most locally dense parts of the protostructure, but are fairly close\nto the centers of their parent overdensity peaks. We propose a scenario where\nmerging might already have happened in both cases, which lowered the local\ndensity of their surrounding area and boosted their stellar mass. This work is\nthe first time that two RAGNs at low luminosity have been found and studied\nwithin a high redshift protostructure.\n"}
{"abstract": "  Many social events and policies generate spillover effects in both time and\nspace. Their occurrence influences not only the outcomes of interest in the\nfuture, but also these outcomes in nearby areas. In this paper, we propose a\ndesign-based approach to estimate the direct and indirect/spillover treatment\neffects of any event or policy under the assumption of sequential ignorability,\nwhen both temporal and spatial interference are allowed to present. The\nproposed estimators are shown to be consistent and asymptotically Normal if the\ndegree of interference dependence does not grow too fast relative to the sample\nsize. The conventional difference-in-differences (DID) or two-way fixed effects\nmodel, nevertheless, leads to biased estimates in this scenario. We apply the\nmethod to examine the impact of Hong Kong's Umbrella Movement on the result of\nthe ensuing election and how an institutional reform affects real estate\nassessment in New York State.\n"}
{"abstract": "  Biological systems need to react to stimuli over a broad spectrum of\ntimescales. If and how this ability can emerge without external fine-tuning is\na puzzle. We consider here this problem in discrete Markovian systems, where we\ncan leverage results from random matrix theory. Indeed, generic large\ntransition matrices are governed by universal results, which predict the\nabsence of long timescales unless fine-tuned. We consider an ensemble of\ntransition matrices and motivate a temperature-like variable that controls the\ndynamic range of matrix elements, which we show plays a crucial role in the\napplicability of the large matrix limit: as the dynamic range increases, a\nphase transition occurs whereby the random matrix theory result is avoided, and\nlong relaxation times ensue, in the entire `ordered' phase. We furthermore show\nthat this phase transition is accompanied by a drop in the entropy rate and a\npeak in complexity, as measured by predictive information (Bialek, Nemenman,\nTishby Neural Computation 13(21) 2001). Extending the Markov model to a Hidden\nMarkov model (HMM), we show that observable sequences inherit properties of the\nhidden sequences, allowing HMMs to be understood in terms of more accessible\nMarkov models. We then apply our findings to fMRI data from 820 human subjects\nscanned at wakeful rest. We show that the data can be quantitatively understood\nin terms of the random model, and that brain activity lies close to the phase\ntransition when engaged in unconstrained, task-free cognition -- supporting the\nbrain criticality hypothesis in this context.\n"}
{"abstract": "  Under any Multiclass Classification (MCC) setting defined by a collection of\nlabeled point-cloud specified by a feature-set, we extract only stochastic\npartial orderings from all possible triplets of point-cloud without explicitly\nmeasuring the three cloud-to-cloud distances. We demonstrate that such a\ncollective of partial ordering can efficiently compute a label embedding tree\ngeometry on the Label-space. This tree in turn gives rise to a predictive\ngraph, or a network with precisely weighted linkages. Such two multiscale\ngeometries are taken as the coarse scale information content of MCC. They\nindeed jointly shed lights on explainable knowledge on why and how labeling\ncomes about and facilitates error-free prediction with potential multiple\ncandidate labels supported by data. For revealing within-label heterogeneity,\nwe further undergo labeling naturally found clusters within each point-cloud,\nand likewise derive multiscale geometry as its fine-scale information content\ncontained in data. This fine-scale endeavor shows that our computational\nproposal is indeed scalable to a MCC setting having a large label-space.\nOverall the computed multiscale collective of data-driven patterns and\nknowledge will serve as a basis for constructing visible and explainable\nsubject matter intelligence regarding the system of interest.\n"}
{"abstract": "  In the article we discuss the notion of the generalized invariant manifold\nintroduced in our previous study. In the literature the method of the\ndifferential constraints is well known as a tool for constructing particular\nsolutions for the nonlinear partial differential equations. Its essence is in\nadding to the nonlinear PDE, a much simpler, as a rule ordinary, differential\nequation, compatible with the given one. Then any solution of the ODE is a\nparticular solution of the PDE as well. However the main problem is to find\nthis compatible ODE. Our generalization is that we look for an ordinary\ndifferential equation that is compatible not with the nonlinear partial\ndifferential equation itself, but with its linearization. Such a generalized\ninvariant manifold is effectively sought. Moreover, it allows one to construct\nsuch important attributes of integrability theory as Lax pairs and recursion\noperators for integrable nonlinear equations. In this paper, we show that they\nprovide a way to construct particular solutions to the equation as well.\n"}
{"abstract": "  The ability to control the spin textures in semiconductors is a fundamental\nstep toward novel spintronic devices, while seeking desirable materials\nexhibiting persistent spin texture (PST) remains a key challenge. The PST is\nthe property of materials preserving a unidirectional spin orientation in the\nmomentum space, which has been predicted to support an extraordinarily long\nspin lifetime of carriers. Herein, by using first-principles density functional\ntheory calculations, we report the emergence of the PST in the two-dimensional\n(2D) germanium monochalcogenides (GeMC). By considering two stable formations\nof the 2D GeMC, namely the pure GeX and Janus Ge2XY monolayers (X, Y = S, Se,\nand Te), we observed the PST around the valence band maximum where the spin\norientation is enforced by the lower point group symmetry of the crystal. In\nthe case of the pure GeX monolayers, we found that the PST is characterized by\nfully out-of-plane spin orientation protected by C2v point group, while the\ncanted PST in the y-z plane is observed in the case of the Janus Ge2XY\nmonolayers due to the lowering symmetry into Cs point group. More importantly,\nwe find large spin-orbit coupling (SOC) parameter in which the PST sustains,\nwhich could be effectively tuned by in-plane strain. The large SOC parameter\nobserved in the present systems leads to the small wavelength of the spatially\nperiodic mode of the spin polarization, which is promising for short spin\nchannel in the spin Hall transistor devices.\n"}
{"abstract": "  There is a need to address the urban-rural disparities in healthcare\nregarding equal access and quality of care. Due to higher rates of chronic\ndisease, reduced access to providers, and a continuous decline in rural\nhospitals, it is imperative that Appalachian cancer patients adopt the use of\nhealth information technology (HIT). The NCCN Distress Thermometer and Problem\nList (DT) is under-utilized, not patient-centered, does not consider provider\nneeds, and is outdated in the current digital landscape. Digitizing patient\ndistress screening poses advantages, such as allowing for more frequent\nscreenings, removing geographical barriers, and rural patient autonomy. In this\npaper, we discuss how knowledge gained from patient-centered design led to the\nunderpinnings of developing a rural remote patient monitoring app that provides\ndelightful and insightful experiences to users.\n"}
{"abstract": "  Functional Electrical Stimulation (FES) can restore motion to a paralysed\nperson's muscles. Yet, control stimulating many muscles to restore the\npractical function of entire limbs is an unsolved problem. Current\nneurostimulation engineering still relies on 20th Century control approaches\nand correspondingly shows only modest results that require daily tinkering to\noperate at all. Here, we present our state of the art Deep Reinforcement\nLearning (RL) developed for real time adaptive neurostimulation of paralysed\nlegs for FES cycling. Core to our approach is the integration of a personalised\nneuromechanical component into our reinforcement learning framework that allows\nus to train the model efficiently without demanding extended training sessions\nwith the patient and working out of the box. Our neuromechanical component\nincludes merges musculoskeletal models of muscle and or tendon function and a\nmultistate model of muscle fatigue, to render the neurostimulation responsive\nto a paraplegic's cyclist instantaneous muscle capacity. Our RL approach\noutperforms PID and Fuzzy Logic controllers in accuracy and performance.\nCrucially, our system learned to stimulate a cyclist's legs from ramping up\nspeed at the start to maintaining a high cadence in steady state racing as the\nmuscles fatigue. A part of our RL neurostimulation system has been successfully\ndeployed at the Cybathlon 2020 bionic Olympics in the FES discipline with our\nparaplegic cyclist winning the Silver medal among 9 competing teams.\n"}
{"abstract": "  When a stable ordered phase and a metastable disordered phase are separated\nby a flat interface, the metastable state changes to the stable state through\nthe propagation of the interface. For cases in which latent heat is generated,\nthe interface displacement during some time interval is proportional to the\nsquare root of the time interval when the extent of supercooling is less than a\ncertain value. We demonstrate this behavior by deriving a perturbative solution\nfor a propagating interface in the phase field model. We calculate the\nleading-order contribution explicitly, and find that the interface temperature\ndeviates from the equilibrium transition temperature in proportion to the\ninterface velocity.\n"}
{"abstract": "  We study the algorithmic stability of Nesterov's accelerated gradient method.\nFor convex quadratic objectives, Chen et al. (2018) proved that the uniform\nstability of the method grows quadratically with the number of optimization\nsteps, and conjectured that the same is true for the general convex and smooth\ncase. We disprove this conjecture and show, for two notions of algorithmic\nstability (including uniform stability), that the stability of Nesterov's\naccelerated method in fact deteriorates exponentially fast with the number of\ngradient steps. This stands in sharp contrast to the bounds in the quadratic\ncase, but also to known results for non-accelerated gradient methods where\nstability typically grows linearly with the number of steps.\n"}
{"abstract": "  A prominent class of model FQH ground states is those realized as correlation\nfunction of $\\mathbb{Z}_k^{(r)}$-algebras. In this paper, we study the\ninterplay between these algebras and their corresponding wavefunctions. In the\nhopes of realizing these wavefunctions as a unique densest zero energy state,\nwe propose a generalization for the projection Hamiltonians\n$\\mathscr{H}_{k+1}^{2r-1}$. Finally, using techniques from invariants of binary\nforms, an ansatz for computation of correlations $\\langle\n\\psi(z_1)\\cdots\\psi(z_{2k})\\rangle \\prod_{i<j}(z_i-z_j)^{2r/k}$ is devised. We\nprovide some evidence that, at least when $r=2$, our proposed Hamiltonian\nrealizes $\\mathbb{Z}_k^{(2)}$-wavefunctions as a \\emph{unique} ground state.\n"}
{"abstract": "  We introduce new message passing algorithms for inference with graphical\nmodels. The standard min-sum and sum-product belief propagation algorithms are\nguaranteed to converge when the graph is tree-structured, but may not converge\nand can be sensitive to the initialization when the graph contains cycles. This\npaper describes modifications to the standard belief propagation algorithms\nthat are guaranteed to converge to a unique solution regardless of the topology\nof the graph.\n"}
{"abstract": "  Stochastic spreading models defined on complex network topologies are used to\nmimic the diffusion of diseases, information, and opinions in real-world\nsystems. Existing theoretical approaches to the characterization of the models\nin terms of microscopic configurations rely on some approximation of\nindependence among dynamical variables, thus introducing a systematic bias in\nthe prediction of the ground-truth dynamics. Here, we develop a combinatorial\nframework based on the approximation that spreading may occur only along the\nshortest paths connecting pairs of nodes. The approximation overestimates\ndynamical correlations among node states and leads to biased predictions.\nSystematic bias is, however, pointing in the opposite direction of existing\napproximations. We show that the combination of the two biased approaches\ngenerates predictions of the ground-truth dynamics that are more accurate than\nthe ones given by the two approximations if used in isolation. We further take\nadvantage of the combinatorial approximation to characterize theoretical\nproperties of some inference problems, and show that the reconstruction of\nmicroscopic configurations is very sensitive to both the place where and the\ntime when partial knowledge of the system is acquired.\n"}
{"abstract": "  The Randomized Incremental Construction (RIC) of search DAGs for point\nlocation in planar subdivisions, nearest-neighbor search in 2D points, and\nextreme point search in 3D convex hulls, are well known to take ${\\cal O}(n\n\\log n)$ expected time for structures of ${\\cal O}(n)$ expected size. Moreover,\nsearching takes w.h.p. ${\\cal O}(\\log n)$ comparisons in the first and w.h.p.\n${\\cal O}(\\log^2 n)$ comparisons in the latter two DAGs. However, the expected\ndepth of the DAGs and high probability bounds for their size are unknown.\n  Using a novel analysis technique, we show that the three DAGs have w.h.p. i)\na size of ${\\cal O}(n)$, ii) a depth of ${\\cal O}(\\log n)$, and iii) a\nconstruction time of ${\\cal O}(n \\log n)$. One application of these new and\nimproved results are \\emph{remarkably simple} Las Vegas verifiers to obtain\nsearch DAGs with optimal worst-case bounds. This positively answers the\nconjectured logarithmic search cost in the DAG of Delaunay triangulations\n[Guibas et al.; ICALP 1990] and a conjecture on the depth of the DAG of\nTrapezoidal subdivisions [Hemmer et al.; ESA 2012]. It also shows that\nhistory-based RIC circumvents a lower bound on runtime tail estimates of\nconflict-graph RICs [Sen; STACS 2019].\n"}
{"abstract": "  Recently, there emerges a series of vision Transformers, which show superior\nperformance with a more compact model size than conventional convolutional\nneural networks, thanks to the strong ability of Transformers to model\nlong-range dependencies. However, the advantages of vision Transformers also\ncome with a price: Self-attention, the core part of Transformer, has a\nquadratic complexity to the input sequence length. This leads to a dramatic\nincrease of computation and memory cost with the increase of sequence length,\nthus introducing difficulties when applying Transformers to the vision tasks\nthat require dense predictions based on high-resolution feature maps. In this\npaper, we propose a new vision Transformer, named Glance-and-Gaze Transformer\n(GG-Transformer), to address the aforementioned issues. It is motivated by the\nGlance and Gaze behavior of human beings when recognizing objects in natural\nscenes, with the ability to efficiently model both long-range dependencies and\nlocal context. In GG-Transformer, the Glance and Gaze behavior is realized by\ntwo parallel branches: The Glance branch is achieved by performing\nself-attention on the adaptively-dilated partitions of the input, which leads\nto a linear complexity while still enjoying a global receptive field; The Gaze\nbranch is implemented by a simple depth-wise convolutional layer, which\ncompensates local image context to the features obtained by the Glance\nmechanism. We empirically demonstrate our method achieves consistently superior\nperformance over previous state-of-the-art Transformers on various vision tasks\nand benchmarks. The codes and models will be made available at\nhttps://github.com/yucornetto/GG-Transformer.\n"}
{"abstract": "  Electronic Health Records (EHRs) provide a wealth of information for machine\nlearning algorithms to predict the patient outcome from the data including\ndiagnostic information, vital signals, lab tests, drug administration, and\ndemographic information. Machine learning models can be built, for example, to\nevaluate patients based on their predicted mortality or morbidity and to\npredict required resources for efficient resource management in hospitals. In\nthis paper, we demonstrate that an attacker can manipulate the machine learning\npredictions with EHRs easily and selectively at test time by backdoor attacks\nwith the poisoned training data. Furthermore, the poison we create has\nstatistically similar features to the original data making it hard to detect,\nand can also attack multiple machine learning models without any knowledge of\nthe models. With less than 5% of the raw EHR data poisoned, we achieve average\nattack success rates of 97% on mortality prediction tasks with MIMIC-III\ndatabase against Logistic Regression, Multilayer Perceptron, and Long\nShort-term Memory models simultaneously.\n"}
{"abstract": "  A detailed analysis of the stimulated parametric down-conversion (PDC)\nprocess is performed to investigate the effects of the spatiotemporal degrees\nof freedom. The analysis provides information that would be useful for\nPDC-based metrology applications. Using a Wigner functional approach, we obtain\nthe parametric down-converted state as the Bogoliubov transformed input state,\nin terms of Bogoliubov kernel functions. The result is used to consider the\ncase for a coherent state seeding stimulated PDC. We also compute the\nbackground which is obtained from spontaneous PDC.\n"}
{"abstract": "  We present a study of metallicities in a sample of main sequence stars with\nspectral types M, K, G and F ($T_{\\rm eff}$ $\\sim$ 3200 -- 6500K and log $g$\n$\\sim$ 4.3 -- 5.0 dex) belonging to the solar neighborhood young open cluster\nComa Berenices. Metallicities were determined using the high-resolution\n(R=$\\lambda$/$\\Delta$ $\\lambda$ $\\sim$ 22,500) NIR spectra ($\\lambda$1.51 --\n$\\lambda$1.69 $\\mu$m) of the SDSS-IV APOGEE survey. Membership to the cluster\nwas confirmed using previous studies in the literature along with APOGEE radial\nvelocities and Gaia DR2. An LTE analysis using plane-parallel MARCS model\natmospheres and the APOGEE DR16 line list was adopted to compute synthetic\nspectra and derive atmospheric parameters ($T_{\\rm eff}$ and log $g$) for the M\ndwarfs and metallicities for the sample. The derived metallicities are near\nsolar and are homogeneous at the level of the expected uncertainties, in\nparticular when considering stars from a given stellar class. The mean\nmetallicity computed for the sample of G, K, and M dwarfs is\n$\\langle$[Fe/H]$\\rangle$ = +0.04 $\\pm$ 0.02 dex; however, the metallicities of\nthe F-type stars are slightly lower, by about 0.04 dex, when compared to cooler\nand less massive members. Models of atomic diffusion can explain this modest\nabundance dip for the F dwarfs, indicating that atomic diffusion operates in\nComa Berenices stars. The [Fe/H] dip occurs in nearly the same effective\ntemperature range as that found in previous analyses of the lithium and\nberyllium abundances in Coma Berenices.\n"}
{"abstract": "  Email is a channel of communication which is considered to be a confidential\nmedium of communication for exchange of information among individuals and\norganisations. The confidentiality consideration about e-mail is no longer the\ncase as attackers send malicious emails to users to deceive them into\ndisclosing their private personal information such as username, password, and\nbank card details, etc. In search of a solution to combat phishing cybercrime\nattacks, different approaches have been developed. However, the traditional\nexiting solutions have been limited in assisting email users to identify\nphishing emails from legitimate ones. This paper reveals the different email\nand website phishing solutions in phishing attack detection. It first provides\na literature analysis of different existing phishing mitigation approaches. It\nthen provides a discussion on the limitations of the techniques, before\nconcluding with an exploration into how phishing detection can be improved.\n"}
{"abstract": "  Mapping new and old buildings are of great significance for understanding\nsocio-economic development in rural areas. In recent years, deep neural\nnetworks have achieved remarkable building segmentation results in\nhigh-resolution remote sensing images. However, the scarce training data and\nthe varying geographical environments have posed challenges for scalable\nbuilding segmentation. This study proposes a novel framework based on Mask\nR-CNN, named HTMask R-CNN, to extract new and old rural buildings even when the\nlabel is scarce. The framework adopts the result of single-object instance\nsegmentation from the orthodox Mask R-CNN. Further, it classifies the rural\nbuildings into new and old ones based on a dynamic grayscale threshold inferred\nfrom the result of a two-object instance segmentation task where training data\nis scarce. We found that the framework can extract more buildings and achieve a\nmuch higher mean Average Precision (mAP) than the orthodox Mask R-CNN model. We\ntested the novel framework's performance with increasing training data and\nfound that it converged even when the training samples were limited. This\nframework's main contribution is to allow scalable segmentation by using\nsignificantly fewer training samples than traditional machine learning\npractices. That makes mapping China's new and old rural buildings viable.\n"}
{"abstract": "  This work presents a quantitative approach to estimate the total time spent\nin the vicinity of a bus stop including the deceleration time, the boarding and\nalighting time (developed in an earlier study), the acceleration time, and\nre-entry time (time required to merge into the adjacent lane). Different\nstatistical models were used to compute the deceleration, acceleration and\nmerge times. Typical deceleration and acceleration levels were computed using\nkinematic equations that were then used to compute both the deceleration and\nacceleration times. The adopted method to estimate both the deceleration time\nand the acceleration time was validated utilizing transit data from Blacksburg\nTransit (BT) using the mean absolute percentage error (MAPE) and root mean\nsquare error (RMSE). The MAPE and RMSE values were calculated to be 0.3 and\n13.3 percent for the deceleration time and 0.42 and 2.72 percent for the\nacceleration time, respectively. The re-entry time was estimated to be a\nfunction of the adjacent roadway traffic density using both a multiple linear\nregression and Bayesian regression approach. Both methods showed consistency in\nestimating the merge-time model coefficients. The proposed models can be\nintegrated with transit applications to estimate transit vehicle travel times.\n"}
{"abstract": "  Deep convolutional models often produce inadequate predictions for inputs\nforeign to the training distribution. Consequently, the problem of detecting\noutlier images has recently been receiving a lot of attention. Unlike most\nprevious work, we address this problem in the dense prediction context in order\nto be able to locate outlier objects in front of in-distribution background.\nOur approach is based on two reasonable assumptions. First, we assume that the\ninlier dataset is related to some narrow application field (e.g.~road driving).\nSecond, we assume that there exists a general-purpose dataset which is much\nmore diverse than the inlier dataset (e.g.~ImageNet-1k). We consider pixels\nfrom the general-purpose dataset as noisy negative training samples since most\n(but not all) of them are outliers. We encourage the model to recognize borders\nbetween known and unknown by pasting jittered negative patches over inlier\ntraining images. Our experiments target two dense open-set recognition\nbenchmarks (WildDash 1 and Fishyscapes) and one dense open-set recognition\ndataset (StreetHazard). Extensive performance evaluation indicates competitive\npotential of the proposed approach.\n"}
{"abstract": "  Countermeasures to effectively fight the ever increasing hate speech online\nwithout blocking freedom of speech is of great social interest. Natural\nLanguage Generation (NLG), is uniquely capable of developing scalable\nsolutions. However, off-the-shelf NLG methods are primarily\nsequence-to-sequence neural models and they are limited in that they generate\ncommonplace, repetitive and safe responses regardless of the hate speech (e.g.,\n\"Please refrain from using such language.\") or irrelevant responses, making\nthem ineffective for de-escalating hateful conversations. In this paper, we\ndesign a three-module pipeline approach to effectively improve the diversity\nand relevance. Our proposed pipeline first generates various counterspeech\ncandidates by a generative model to promote diversity, then filters the\nungrammatical ones using a BERT model, and finally selects the most relevant\ncounterspeech response using a novel retrieval-based method. Extensive\nExperiments on three representative datasets demonstrate the efficacy of our\napproach in generating diverse and relevant counterspeech.\n"}
{"abstract": "  Graph-based Neural Networks (GNNs) are recent models created for learning\nrepresentations of nodes (and graphs), which have achieved promising results\nwhen detecting patterns that occur in large-scale data relating different\nentities. Among these patterns, financial fraud stands out for its\nsocioeconomic relevance and for presenting particular challenges, such as the\nextreme imbalance between the positive (fraud) and negative (legitimate\ntransactions) classes, and the concept drift (i.e., statistical properties of\nthe data change over time). Since GNNs are based on message propagation, the\nrepresentation of a node is strongly impacted by its neighbors and by the\nnetwork's hubs, amplifying the imbalance effects. Recent works attempt to adapt\nundersampling and oversampling strategies for GNNs in order to mitigate this\neffect without, however, accounting for concept drift. In this work, we conduct\nexperiments to evaluate existing techniques for detecting network fraud,\nconsidering the two previous challenges. For this, we use real data sets,\ncomplemented by synthetic data created from a new methodology introduced here.\nBased on this analysis, we propose a series of improvement points that should\nbe investigated in future research.\n"}
{"abstract": "  In this paper, we propose a new macro-micro approach to modeling parking. We\nfirst develop a microscopic parking simulation model considering both on- and\noff-street parking with limited capacity. In the microscopic model, a parking\nsearch algorithm is proposed to mimic cruising-for-parking based on the\nprinciple of proximity, and a parking-related state tracking algorithm is\nproposed to acquire an event-based simulated data set. Some key aspects of\nparking modeling are discussed based on the sim-ulated evidence and theoretical\nanalysis. Results suggest (i) although the low cruising speed reduces the\nnetwork performance, it does not significantly alter the macroscopic or network\nfundamental diagram (MFD or NFD) unless the cruising vehicles dominate the\ntraffic stream; (ii) distance to park is not uniquely determined by parking\noccupancy because factors such as cruising speed and parking dura-tion also\ncontribute; and (iii) multiscale parking occupancy-driven intelligent parking\nguidance can re-duce distance to park yielding considerable network efficiency\ngains. Using the microscopic model, we then extend, calibrate, and validate a\nmacroscopic parking dynamics model with an NFD representation. The demonstrated\nconsistency between the macro- and micro-models permits integration of the two\nfor online parking pricing optimization via model predictive control. Numerical\nexperiments highlight the effectiveness of the proposed approach as well as one\ncaveat. That is, when pricing on-street parking, the road network connected to\nthe alternate off-street parking lots must have sufficient capacity to\nac-commodate the increased parking demand; otherwise, local congestion may\narise that violates the ho-mogeneity assumption underlying the macroscopic\nmodel.\n"}
{"abstract": "  HD 106906 is a young, binary stellar system, located in the Lower Centaurus\nCrux (LCC) group. This system is unique among discovered systems in that it\ncontains an asymmetrical debris disk, as well as an 11 M$_{Jup}$ planet\ncompanion, at a separation of $\\sim$735 AU. Only a handful of other systems are\nknown to contain both a disk and directly imaged planet, where HD 106906 is the\nonly one in which the planet has apparently been scattered. The debris disk is\nnearly edge on, and extends roughly to $>$500 AU, where previous studies with\nHST have shown the outer regions to have high asymmetry. To better understand\nthe structure and composition of the disk, we have performed a deep\npolarimetric study of HD 106906's asymmetrical debris disk using newly obtained\n$H$-, $J$-, and $K1$-band polarimetric data from the Gemini Planet Imager\n(GPI). An empirical analysis of our data supports a disk that is asymmetrical\nin surface brightness and structure, where fitting an inclined ring model to\nthe disk spine suggests that the disk may be highly eccentric ($e\\gtrsim0.16$).\nA comparison of the disk flux with the stellar flux in each band suggests a\nblue color that also does not significantly vary across the disk. We discuss\nthese results in terms of possible sources of asymmetry, where we find that\ndynamical interaction with the planet companion, HD 106906b, is a likely\ncandidate.\n"}
{"abstract": "  Engineering phase change materials (PCM) to realize superior data storage\ndevices requires a detailed understanding of crystallization kinetics and its\ntemperature dependence. The temperature dependence of crystallization differs\ndistinctly between crystallizing from the glassy phase and the undercooled\nliquid (UCL). Hence, knowing the phase from which crystallization occurs is\nnecessary for predicting the switching ability. Here, we measure the glassy\ndynamics and crystallization kinetics using calorimetry for heating rates\nspanning over six orders of magnitude. Our results show that the prominent PCM\n(Ag,In)-doped Sb2Te (AIST) exhibits a change from crystallizing from the glassy\nphase to crystallizing from the UCL at a critical heating rate of 5,000 K/s.\nAbove the glass transition, the activation energy of crystallization changes\ndrastically enabling rapid crystallization at elevated temperatures.\n"}
{"abstract": "  Team Semantics is a generalization of Tarskian Semantics that can be used to\nadd to First Order Logic atoms and connectives expressing dependencies between\nthe possible values of variables. Some of these extensions are more expressive\nthan First Order Logic, while others are reducible to it. In this work, I fully\ncharacterize the (relativizable) atoms and families of atoms that do not\nincrease the expressive power of First Order Logic with Team Semantics when\nthey and their negations are added to it, separately or jointly (or,\nequivalently, when they are added to First Order Logic augmented with a\ncontradictory negation connective applicable only to literals and dependency\natoms).\n"}
{"abstract": "  A method to minimize the cost of the structural design of reinforced concrete\nstructures using Biogeography-Based Optimization, an evolutionary algorithm, is\npresented. SAP2000 is used as computational engine, taking into account\nmodelling aspects such as static soil-structure interaction (SSSI). The\noptimization problem is formulated to properly reflect an actual design\nproblem, limiting e.g. the size of reinforcement bars to commercially available\nsections. Strategies to reduce the computational cost of the optimization\nprocedure are proposed and an extensive parameter tuning was performed. The\nresulting tuned optimization algorithm allows to reduce the direct cost of the\nconstruction of a particular structure project with 21% compared to a design\nbased on traditional criteria. We also evaluate the effect on the cost of the\nsuperstructure when SSSI is takeninto account.\n"}
{"abstract": "  Machine learning (ML) has been used to develop increasingly accurate link\nquality estimators for wireless networks. However, more in-depth questions\nregarding the most suitable class of models, most suitable metrics and model\nperformance on imbalanced datasets remain open. In this paper, we propose a new\ntree-based link quality classifier that meets high performance and fairly\nclassifies the minority class and, at the same time, incurs low training cost.\nWe compare the tree-based model, to a multilayer perceptron (MLP) non-linear\nmodel and two linear models, namely logistic regression (LR) and SVM, on a\nselected imbalanced dataset and evaluate their results using five different\nperformance metrics. Our study shows that 1) non-linear models perform slightly\nbetter than linear models in general, 2) the proposed non-linear tree-based\nmodel yields the best performance trade-off considering F1, training time and\nfairness, 3) single metric aggregated evaluations based only on accuracy can\nhide poor, unfair performance especially on minority classes, and 4) it is\npossible to improve the performance on minority classes, by over 40% through\nfeature selection and by over 20% through resampling, therefore leading to\nfairer classification results.\n"}
{"abstract": "  A neural multimodal machine translation (MMT) system is one that aims to\nperform better translation by extending conventional text-only translation\nmodels with multimodal information. Many recent studies report improvements\nwhen equipping their models with the multimodal module, despite the controversy\nof whether such improvements indeed come from the multimodal part. We revisit\nthe contribution of multimodal information in MMT by devising two interpretable\nMMT models. To our surprise, although our models replicate similar gains as\nrecently developed multimodal-integrated systems achieved, our models learn to\nignore the multimodal information. Upon further investigation, we discover that\nthe improvements achieved by the multimodal models over text-only counterparts\nare in fact results of the regularization effect. We report empirical findings\nthat highlight the importance of MMT models' interpretability, and discuss how\nour findings will benefit future research.\n"}
{"abstract": "  We prove that any surjective self-morphism with $\\delta_f > 1$ on a\npotentially dense smooth projective surface defined over a number field $K$ has\ndensely many $L$-rational points for a finite extension $L/K$.\n"}
{"abstract": "  Following Abbatiello et al. [ DCCDS-A (41), 2020], we introduce dissipative\nturbulent solutions to a simple model of a mixture of two non interacting\ncompressible fluids {\\tc filling a bounded domain with general non zero\ninflow/outflow boundary conditions.} We prove existence of such solutions for\nall adiabatic coefficients $\\gamma>1$, their compatibility with classical\nsolutions, the relative energy inequality, and the weak strong uniqueness\nprinciple in this class. The class of dissipative turbulent solutions is so far\nthe largest class of generalized solutions which still enjoys the weak strong\nuniqueness property.\n"}
{"abstract": "  We study the radiation from charged particles crossing a cholesteric plate in\nthe shortwave approximation when the wavelength of photons is much smaller than\nthe pitch of the cholesteric helix whereas the escaping angle of the photon and\nthe anisotropy of the permittivity tensor can be arbitrary. The radiation of\nphotons is treated in the framework of quantum electrodynamics with classical\ncurrents. The radiation of the plane-wave photons and the photons with definite\nprojection of the angular momentum (the twisted photons) produced by charged\nparticles crossing the cholesteric plate and moving rectilinearly and uniformly\nis considered. The explicit expressions for the average number of radiated\nphotons and their spectra with respect to the energy and the projection of the\nangular momentum are obtained in this case. It turns out that in the paraxial\napproximation the projection of the orbital angular momentum, $l$, of radiated\ntwisted photons is related to the harmonic number $n\\in \\mathbb{Z}$ as\n$l=2n+1$, i.e., the given system is a pure source of twisted photons as\nexpected. It is shown that in the paraxial shortwave regime the main part of\nradiated photons is linearly polarized with $l=\\pm1$ at the harmonics\n$n=\\{-1,0\\}$. The applicability conditions of the approach developed are\ndiscussed. As the examples, we consider the production of $6.3$ eV twisted\nphotons from uranium nuclei and the production of X-ray twisted photons from\n$120$ MeV electrons.\n"}
{"abstract": "  Understanding the formation of molecules under conditions relevant to\ninterstellar chemistry is fundamental to characterize the chemical evolution of\nthe universe. Using reactive molecular dynamics simulations with model-based or\nhigh-quality potential energy surfaces provides a means to specifically and\nquantitatively probe individual reaction channels at a molecular level. The\nformation of CO$_2$ from collision of CO($^1 \\Sigma$) and O($^1$D) is\ncharacterized on amorphous solid water (ASW) under conditions typical in cold\nmolecular clouds. Recombination takes place on the sub-nanosecond time scale\nand internal energy redistribution leads to stabilization of the product with\nCO$_2$ remaining adsorbed on the ASW on extended time scales. Using a\nhigh-level, reproducing kernel-based potential energy surface for CO$_2$,\nformation into and stabilization of CO$_2$ and COO is observed.\n"}
{"abstract": "  Many dynamical systems exhibit abrupt transitions or tipping as the control\nparameter is varied. In scenarios where the parameter is varied continuously,\nthe rate of change of control parameter greatly affects the performance of\nearly warning signals (EWS) for such critical transitions.We study the impact\nof variation of the control parameter with a finite rate on the performance of\n\\textcolor{black}{EWS for critical transitions} in a thermoacoustic system (a\nhorizontal Rijke tube) exhibiting subcritical Hopf bifurcation. There is a\ngrowing interest in developing early warning signals for tipping in real\nsystems. Firstly, we explore the efficacy of early warning signals based on\ncritical slowing down and fractal characteristics. From this study, lag-1\nautocorrelation (AC) and Hurst exponent H are found to be good measures to\npredict the transition well-before the tipping point. The warning time,\nobtained using AC and $H$, reduces with an increase in the rate of change of\nthe control parameter following an inverse power law relation. Hence, for very\nfast rates, the warning time may be too short to perform any control action.\nFurthermore, we report the observation of a hyperexponential scaling relation\nbetween the AC and the variance of fluctuations during such dynamic Hopf\nbifurcation. We construct a theoretical model for noisy Hopf bifurcation\nwherein the control parameter is continuously varied at different rates to\nstudy the effect of rate of change of parameter on EWS. Similar results,\nincluding the hyperexponential scaling, are observed in the model as well.\n"}
{"abstract": "  Individuals of modern societies share ideas and participate in collective\nprocesses within a pervasive, variable, and mostly hidden ecosystem of content\nfiltering technologies that determine what information we see online. Despite\nthe impact of these algorithms on daily life and society, little is known about\ntheir effect on information transfer and opinion formation. It is thus unclear\nto what extent algorithmic bias has a harmful influence on collective\ndecision-making, such as a tendency to polarize debate. Here we introduce a\ngeneral theoretical framework to systematically link models of opinion\ndynamics, social network structure, and content filtering. We showcase the\nflexibility of our framework by exploring a family of binary-state opinion\ndynamics models where information exchange lies in a spectrum from pairwise to\ngroup interactions. All models show an opinion polarization regime driven by\nalgorithmic bias and modular network structure. The role of content filtering\nis, however, surprisingly nuanced; for pairwise interactions it leads to\npolarization, while for group interactions it promotes coexistence of opinions.\nThis allows us to pinpoint which social interactions are robust against\nalgorithmic bias, and which ones are susceptible to bias-enhanced opinion\npolarization. Our framework gives theoretical ground for the development of\nheuristics to tackle harmful effects of online bias, such as information\nbottlenecks, echo chambers, and opinion radicalization.\n"}
{"abstract": "  The question of how and why the phenomenon of mode connectivity occurs in\ntraining deep neural networks has gained remarkable attention in the research\ncommunity. From a theoretical perspective, two possible explanations have been\nproposed: (i) the loss function has connected sublevel sets, and (ii) the\nsolutions found by stochastic gradient descent are dropout stable. While these\nexplanations provide insights into the phenomenon, their assumptions are not\nalways satisfied in practice. In particular, the first approach requires the\nnetwork to have one layer with order of $N$ neurons ($N$ being the number of\ntraining samples), while the second one requires the loss to be almost\ninvariant after removing half of the neurons at each layer (up to some\nrescaling of the remaining ones). In this work, we improve both conditions by\nexploiting the quality of the features at every intermediate layer together\nwith a milder over-parameterization condition. More specifically, we show that:\n(i) under generic assumptions on the features of intermediate layers, it\nsuffices that the last two hidden layers have order of $\\sqrt{N}$ neurons, and\n(ii) if subsets of features at each layer are linearly separable, then no\nover-parameterization is needed to show the connectivity. Our experiments\nconfirm that the proposed condition ensures the connectivity of solutions found\nby stochastic gradient descent, even in settings where the previous\nrequirements do not hold.\n"}
{"abstract": "  We report a dramatic current reduction, or a resistance increase, by a few\norders of magnitude of two common-anode Au/Ti/pGaN Schottky junctions annealed\nwithin a certain annealing condition window (600 - 700 oC, 1 - 4 min). Results\nfrom similar common-anode Schottky junctions made of Au/p-GaN, Al/Ti/p-GaN and\nAu/Ti/graphene/p-GaN junctions demonstrated that all the three layers (Au, Ti\nand p-GaN) are essential for the increased resistance. Raman characterization\nof the p-GaN showed a decrease of the Mg-N bonding, i.e., the deactivation of\nMg, which is consistent with the Hall measurement results. Moreover, this\nhigh-resistance junction structure was employed in p-GaN gate AlGaN/GaN HEMTs.\nIt was shown to be an effective gate technology that was capable to boost the\ngate breakdown voltage from 9.9 V to 13.8 V with a negligible effect on the\nthreshold voltage or the sub-threshold slope.\n"}
{"abstract": "  A description of tilting complexes is given for a class of PI algebras whose\nprime spectrum is canonically homeomorphic to the prime spectrum of its center.\nSome Sklyanin algebras are the kind of algebras considered. As an application,\nit is proved that any algebra derived equivalent to such kind of algebra, is\nMorita equivalent to it.\n"}
{"abstract": "  In this paper we propose a system capable of tracking multiple soccer players\nin different types of video quality. The main goal, in contrast to most\nstate-of-art soccer player tracking systems, is the ability of execute\neffectively tracking in videos of low-quality. We adapted a state-of-art\nMultiple Object Tracking to the task. In order to do that adaptation, we\ncreated a Detection and a Tracking Dataset for 3 different qualities of video.\nThe results of our system are conclusive of its high performance.\n"}
{"abstract": "  In this article, we investigate inverse source problems for a wide range of\nPDEs of parabolic and hyperbolic types as well as time-fractional evolution\nequations by partial interior observation. Restricting the source terms to the\nform of separated variables, we establish uniqueness results for simultaneously\ndetermining both temporal and spatial components without non-vanishing\nassumptions at $t=0$, which seems novel to the best of our knowledge.\nRemarkably, mostly we allow a rather flexible choice of the observation time\nnot necessarily starting from $t=0$, which fits into various situations in\npractice. Our main approach is based on the combination of the Titchmarsh\nconvolution theorem with unique continuation properties and time-analyticity of\nthe PDEs under consideration.\n"}
{"abstract": "  The Crouzeix-Raviart triangular finite elements are $\\inf$-$\\sup$ stable for\nthe Stokes equations for any mesh with at least one interior vertex. This\nresult affirms a {\\em conjecture of Crouzeix-Falk} from 1989 for $p=3$. Our\nproof applies to {\\em any odd degree} $p\\ge 3$ and hence Crouzeix-Raviart\ntriangular finite elements of degree $p$ in two dimensions and the piecewise\npolynomials of degree $p-1$ with vanishing integral form a stable Stokes pair\n{\\em for all positive integers} $p$.\n"}
{"abstract": "  What topological spaces can be partitioned into copies of the Cantor space\n$2^\\omega$? An obvious necessary condition is that a space can be partitioned\ninto copies of $2^\\omega$ only if it can be covered with copies of $2^\\omega$.\nWe prove three theorems concerning when this necessary condition is also\nsufficient.\n  If $X$ is a metrizable space and $|X| \\leq \\mathfrak{c}^{+\\omega}$ (the least\nlimit cardinal $>\\!\\mathfrak{c}$), then $X$ can be partitioned into copies of\n$2^\\omega$ if and only if $X$ can be covered with copies of $2^\\omega$. To show\nthis cardinality bound is sharp, we construct a metrizable space of size\n$\\mathfrak{c}^{+(\\omega+1)}$ that can be covered with copies of $2^\\omega$, but\nnot partitioned into copies of $2^\\omega$.\n  Similarly, if $X$ is first countable and $|X| \\leq \\mathfrak{c}$, then $X$\ncan be partitioned into copies of $2^\\omega$ if and only if $X$ can be covered\nwith copies of $2^\\omega$. On the other hand, there is a first countable space\nof size $\\mathfrak{c}^+$ that can be covered with copies of $2^\\omega$, but not\npartitioned into copies of $2^\\omega$.\n  Finally, we show that a completely metrizable space can be partitioned into\ncopies of $2^\\omega$ if and only if it can be covered with copies of $2^\\omega$\nif and only if it has no isolated points.\n"}
{"abstract": "  The product of two zero mean correlated normal random variables has received\nmuch attention in the statistics literature and appears in many application\nareas. However, many important distributional properties are yet to tabulated.\nThis paper fills this gap by providing the basic distributional theory for the\nproduct of two zero mean correlated normal random variables.\n"}
{"abstract": "  Simultaneous object recognition and pose estimation are two key\nfunctionalities for robots to safely interact with humans as well as\nenvironments. Although both object recognition and pose estimation use visual\ninput, most state-of-the-art tackles them as two separate problems since the\nformer needs a view-invariant representation while object pose estimation\nnecessitates a view-dependent description. Nowadays, multi-view Convolutional\nNeural Network (MVCNN) approaches show state-of-the-art classification\nperformance. Although MVCNN object recognition has been widely explored, there\nhas been very little research on multi-view object pose estimation methods, and\neven less on addressing these two problems simultaneously. The pose of virtual\ncameras in MVCNN methods is often pre-defined in advance, leading to bound the\napplication of such approaches. In this paper, we propose an approach capable\nof handling object recognition and pose estimation simultaneously. In\nparticular, we develop a deep object-agnostic entropy estimation model, capable\nof predicting the best viewpoints of a given 3D object. The obtained views of\nthe object are then fed to the network to simultaneously predict the pose and\ncategory label of the target object. Experimental results showed that the views\nobtained from such positions are descriptive enough to achieve a good accuracy\nscore. Code is available online at: https://github.com/tparisotto/more_mvcnn\n"}
{"abstract": "  Few-shot learning (FSL) aims to classify images under low-data regimes, where\nthe conventional pooled global feature is likely to lose useful local\ncharacteristics. Recent work has achieved promising performances by using deep\ndescriptors. They generally take all deep descriptors from neural networks into\nconsideration while ignoring that some of them are useless in classification\ndue to their limited receptive field, e.g., task-irrelevant descriptors could\nbe misleading and multiple aggregative descriptors from background clutter\ncould even overwhelm the object's presence. In this paper, we argue that a\nMutual Nearest Neighbor (MNN) relation should be established to explicitly\nselect the query descriptors that are most relevant to each task and discard\nless relevant ones from aggregative clutters in FSL. Specifically, we propose\nDiscriminative Mutual Nearest Neighbor Neural Network (DMN4) for FSL. Extensive\nexperiments demonstrate that our method outperforms the existing\nstate-of-the-arts on both fine-grained and generalized datasets.\n"}
{"abstract": "  In the last few years, the release of BERT, a multilingual transformer based\nmodel, has taken the NLP community by storm. BERT-based models have achieved\nstate-of-the-art results on various NLP tasks, including dialog tasks. One of\nthe limitation of BERT is the lack of ability to handle long text sequence. By\ndefault, BERT has a maximum wordpiece token sequence length of 512. Recently,\nthere has been renewed interest to tackle the BERT limitation to handle long\ntext sequences with the addition of new self-attention based architectures.\nHowever, there has been little to no research on the impact of this limitation\nwith respect to dialog tasks. Dialog tasks are inherently different from other\nNLP tasks due to: a) the presence of multiple utterances from multiple\nspeakers, which may be interlinked to each other across different turns and b)\nlonger length of dialogs. In this work, we empirically evaluate the impact of\ndialog length on the performance of BERT model for the Next Response Selection\ndialog task on four publicly available and one internal multi-turn dialog\ndatasets. We observe that there is little impact on performance with long\ndialogs and even the simplest approach of truncating input works really well.\n"}
{"abstract": "  Different categories of visual stimuli activate different responses in the\nhuman brain. These signals can be captured with EEG for utilization in\napplications such as Brain-Computer Interface (BCI). However, accurate\nclassification of single-trial data is challenging due to low signal-to-noise\nratio of EEG. This work introduces an EEG-ConvTranformer network that is based\non multi-headed self-attention. Unlike other transformers, the model\nincorporates self-attention to capture inter-region interactions. It further\nextends to adjunct convolutional filters with multi-head attention as a single\nmodule to learn temporal patterns. Experimental results demonstrate that\nEEG-ConvTransformer achieves improved classification accuracy over the\nstate-of-the-art techniques across five different visual stimuli classification\ntasks. Finally, quantitative analysis of inter-head diversity also shows low\nsimilarity in representational subspaces, emphasizing the implicit diversity of\nmulti-head attention.\n"}
{"abstract": "  Let $K$ be a complex bi-quadratic field with ring of integers\n$\\mathcal{O}_{K}$. For $K = \\mathbb{Q}(\\sqrt{-m}$, $\\sqrt{n}$), where $ m\n\\equiv 3 \\pmod 4 $ and $ n \\equiv 1 \\pmod 4$, we prove that every algebraic\ninteger can be written as sum of integral squares. Using this, we prove that\nfor any complex bi-quadratic field $K$, every element of $4\\mathcal{O}_K$ can\nbe written as sum of five integral squares. In addition, we show that the\nPythagoras number of ring of integers of any CM field is at most five.\nMoreover, we give two classes of complex bi-quadratic fields for which\n$p(\\mathcal{O}_{K})= 3$ and $p(4\\mathcal{O}_{K})=3 $ respectively. Here,\n$p(\\mathcal{O}_{K})$ is the Pythagoras number of ring of integers of $K$ and\n$p(4\\mathcal{O}_{K})$ is the smallest positive integer $t$ such that every\nelement of $4\\mathcal{O}_{K}$ can be written as sum of $t$ integral squares.\n"}
{"abstract": "  In the past two decades, machine learning potentials (MLP) have reached a\nlevel of maturity that now enables applications to large-scale atomistic\nsimulations of a wide range of systems in chemistry, physics and materials\nscience. Different machine learning algorithms have been used with great\nsuccess in the construction of these MLPs. In this review, we discuss an\nimportant group of MLPs relying on artificial neural networks to establish a\nmapping from the atomic structure to the potential energy. In spite of this\ncommon feature, there are important conceptual differences, which concern the\ndimensionality of the systems, the inclusion of long-range electrostatic\ninteractions, global phenomena like non-local charge transfer, and the type of\ndescriptor used to represent the atomic structure, which can either be\npredefined or learnable. A concise overview is given along with a discussion of\nthe open challenges in the field.\n"}
{"abstract": "  We apply the Tangent Method of Colomo and Sportiello to predict the arctic\ncurves of the Six Vertex model with reflecting (U-turn) boundary and of the\nrelated Twenty Vertex model with suitable domain wall boundary conditions on a\nquadrangle, both in their Disordered phase.\n"}
{"abstract": "  Android malware detection is a significat problem that affects billions of\nusers using millions of Android applications (apps) in existing markets. This\npaper proposes PetaDroid, a framework for accurate Android malware detection\nand family clustering on top of static analyses. PetaDroid automatically adapts\nto Android malware and benign changes over time with resilience to common\nbinary obfuscation techniques. The framework employs novel techniques\nelaborated on top of natural language processing (NLP) and machine learning\ntechniques to achieve accurate, adaptive, and resilient Android malware\ndetection and family clustering. PetaDroid identifies malware using an ensemble\nof convolutional neural network (CNN) on proposed Inst2Vec features. The\nframework clusters the detected malware samples into malware family groups\nutilizing sample feature digests generated using deep neural auto-encoder. For\nchange adaptation, PetaDroid leverages the detection confidence probability\nduring deployment to automatically collect extension datasets and periodically\nuse them to build new malware detection models. Besides, PetaDroid uses\ncode-fragment randomization during the training to enhance the resiliency to\ncommon obfuscation techniques. We extensively evaluated PetaDroid on multiple\nreference datasets. PetaDroid achieved a high detection rate (98-99% f1-score)\nunder different evaluation settings with high homogeneity in the produced\nclusters (96%). We conducted a thorough quantitative comparison with\nstate-of-the-art solutions MaMaDroid, DroidAPIMiner, MalDozer, in which\nPetaDroid outperforms them under all the evaluation settings.\n"}
{"abstract": "  One of the most significant differences of M5 over previous forecasting\ncompetitions is that it was held on Kaggle, an online platform of data\nscientists and machine learning practitioners. Kaggle provides a gathering\nplace, or virtual community, for web users who are interested in the M5\ncompetition. Users can share code, models, features, loss functions, etc.\nthrough online notebooks and discussion forums. This paper aims to study the\nsocial influence of virtual community on user behaviors in the M5 competition.\nWe first research the content of the M5 virtual community by topic modeling and\ntrend analysis. Further, we perform social media analysis to identify the\npotential relationship network of the virtual community. We study the roles and\ncharacteristics of some key participants that promote the diffusion of\ninformation within the M5 virtual community. Overall, this study provides\nin-depth insights into the mechanism of the virtual community's influence on\nthe participants and has potential implications for future online competitions.\n"}
{"abstract": "  Recent interest in exploiting Deep Learning techniques for Noise Suppression,\nhas led to the creation of Hybrid Denoising Systems that combine classic Signal\nProcessing with Deep Learning. In this paper, we concentrated our efforts on\nextending the RNNoise denoising system (arXiv:1709.08243) with the inclusion of\ncomplementary features during the training phase. We present a comprehensive\nexplanation of the set-up process of a modified system and present the\ncomparative results derived from a performance evaluation analysis, using a\nreference version of RNNoise as control.\n"}
{"abstract": "  Contrary to a paradigm of metal-enhanced fluorescence, unusually thick\ndielectric coatings can be very favorable to achieve extreme values of averaged\nfluorescence enhancement factor $\\bar F\\gtrsim 3000$ for emitters located on\nthe surface, or in the interior, of the shell of Au@dielectric spherical\ncore-shell particles under realistic conditions, even for the emitters with\n100% intrinsic quantum yield. Thick dielectric coatings facilitate high-quality\ntransverse electric (TE) multipole ($\\ell=7$) resonances which are shown as the\nmajor cause for the reported extraordinary values of $\\bar F$.\n"}
{"abstract": "  Embedded systems represent a billionaire market and are present in most of\nthe processors produced in the world.Embedded software interacts with external\nperipherals such as sensors and actuators through drivers. The strong\ninteraction between drivers and external peripherals often hamper embedded\nsoftware development, in special the testing task, giving that the physical\nenvironment may not be deterministic and difficult to be recreated during the\ntests, the hardware may not be available or its presence may not be desirable\nor costly. Aiming at addressing these problems, this paper introduces a\nsolution to test drivers on microcontrollers, based on a method that uses three\ncomponents: a Device Under Test (DUT), a Double device, and a computer. The\ncomputer runs a test orchestration code, while the DUT runs the test target\ncode, and the Double plays the role of the real external peripherals that\ninteract with the DUT. The proposed solution was successfully implemented and\nvalidated using different protocols.\n"}
{"abstract": "  We study the total $\\alpha$-powered length of the rooted edges in a random\nminimal directed spanning tree - first introduced in \\cite{BR04} - on a Poisson\nprocess with intensity $s \\ge 1$ on the unit cube $[0,1]^d$ for $d \\ge 3$.\nWhile a Dickman limit was proved in \\cite{PW04} in the case of $d=2$, in\ndimensions three and higher, \\cite{BLP06} showed a Gaussian central limit\ntheorem when $\\alpha=1$, with a rate of convergence of the order $(\\log\ns)^{-(d-2)/4} (\\log \\log s)^{(d+1)/2}$. In this paper, we extend these results\nand prove a central limit theorem in any dimension $d \\ge 3$ for any\n$\\alpha>0$. Moreover, making use of recent results in Stein's method for\nregion-stabilizing functionals, we provide presumably optimal non-asymptotic\nbounds of the order $(\\log s)^{-(d-2)/2}$ on the Wasserstein and the Kolmogorov\ndistances between the distribution of the total $\\alpha$-powered length of\nrooted edges, suitably normalized, and that of a standard Gaussian random\nvariable.\n"}
{"abstract": "  Nodal-line semimetals (NLSMs) contains Dirac/Weyl type band-crossing nodes\nextending into shapes of line, loop and chain in the reciprocal space, leading\nto novel band topology and transport responses. Robust NLSMs against spin-orbit\ncoupling typically occur in three-dimensional materials with more symmetry\noperations to protect the line nodes of band crossing, while the possibilities\nin lower-dimensional materials are rarely discussed. Here we demonstrate robust\nNLSM phase in a quasi-one-dimensional nonmagnetic semimetal TaNiTe5. Combining\nangle-resolved photoemission spectroscopy measurements and first-principles\ncalculations, we reveal how reduced dimension can interact with nonsymmorphic\nsymmetry and result into multiple Dirac-type nodal lines with four-fold\ndegeneracy. Our findings suggest rich physics and application in\n(quasi-)one-dimensional topological materials and call for further\ninvestigation on the interplay between the quantum confinement and nontrivial\nband topology.\n"}
{"abstract": "  The Multi-Phase Transport model (AMPT) is used to study the effects of the\nparton-scattering cross-sections ($\\sigma_{pp}$) and hadronic re-scattering on\nthe linear contributions to the flow harmonic $\\textit{v}_{4}$, the non-linear\nresponse coefficients, and the correlations between different order flow\nsymmetry planes in Au+Au collisions at 200~GeV. The model results, which agree\nwith current experimental measurements, indicate that the higher-order flow\nharmonics are sensitive to the $\\sigma_{pp}$ variations. However, the\nnon-linear response coefficients and the correlations between different order\nflow symmetry planes are $\\sigma_{pp}$ independent. These results suggest that\nfurther detailed experimental measurements which span a broad range of\ncollision systems and beam energies could serve as an additional constraint for\nthe theoretical models' calculations.\n"}
{"abstract": "  This is the problem of the 8$^\\mathrm{th}$ International Experimental Physics\nOlympiad (EPO). The task of the EPO8 is to determine Plank constant\n$\\hbar=h/2\\pi$ using the given set-up with LED. If you have an idea how to do\nit, do it and send us the result; skip the reading of the detailed step by step\ninstructions with increasing difficulties. We expect the participants to follow\nthe suggested items -- they are instructive for physics education in\ngeneral.Only the reading of historical remarks given in the first section can\nbe omitted during the Olympiad without loss of generality. Participants should\ntry solving as much tasks as they can without paying attention to the age\ncategories: give your best.\n"}
{"abstract": "  In the paper, we consider a stochastic hybrid Korteweg - de Vries - Burgers\ntype equation with multiplicative noise in the form of cylindrical Wiener\nprocess. We prove the existence of a martingale solution to the equation\nstudied. The proof of the existence of the solution is based on two\napproximations of the considered problem and the compactness method. First, we\nintroduce an auxiliary problem corresponding to the equation studied. Then, we\nprove the existence of a martingale solution to this problem. Finally, we show\nthat the solution of the auxiliary problem converges, in some sense, to the\nsolution of the equation under consideration.\n"}
{"abstract": "  Distributed quantum information processing is essential for building quantum\nnetworks and enabling more extensive quantum computations. In this regime,\nseveral spatially separated parties share a multipartite quantum system, and\nthe most natural set of operations is Local Operations and Classical\nCommunication (LOCC). As a pivotal part in quantum information theory and\npractice, LOCC has led to many vital protocols such as quantum teleportation.\nHowever, designing practical LOCC protocols is challenging due to LOCC's\nintractable structure and limitations set by near-term quantum devices. Here we\nintroduce LOCCNet, a machine learning framework facilitating protocol design\nand optimization for distributed quantum information processing tasks. As\napplications, we explore various quantum information tasks such as entanglement\ndistillation, quantum state discrimination, and quantum channel simulation. We\ndiscover protocols with evident improvements, in particular, for entanglement\ndistillation with quantum states of interest in quantum information. Our\napproach opens up new opportunities for exploring entanglement and its\napplications with machine learning, which will potentially sharpen our\nunderstanding of the power and limitations of LOCC. An implementation of\nLOCCNet is available in Paddle Quantum, a quantum machine learning Python\npackage based on PaddlePaddle deep learning platform.\n"}
{"abstract": "  The $F_\\beta$ score is a commonly used measure of classification performance,\nwhich plays crucial roles in classification tasks with imbalanced data sets.\nHowever, the $F_\\beta$ score cannot be used as a loss function by\ngradient-based learning algorithms for optimizing neural network parameters due\nto its non-differentiability. On the other hand, commonly used loss functions\nsuch as the binary cross-entropy (BCE) loss are not directly related to\nperformance measures such as the $F_\\beta$ score, so that neural networks\noptimized by using the loss functions may not yield optimal performance\nmeasures. In this study, we investigate a relationship between classification\nperformance measures and loss functions in terms of the gradients with respect\nto the model parameters. Then, we propose a differentiable surrogate loss\nfunction for the optimization of the $F_\\beta$ score. We show that the gradient\npaths of the proposed surrogate $F_\\beta$ loss function approximate the\ngradient paths of the large sample limit of the $F_\\beta$ score. Through\nnumerical experiments using ResNets and benchmark image data sets, it is\ndemonstrated that the proposed surrogate $F_\\beta$ loss function is effective\nfor optimizing $F_\\beta$ scores under class imbalances in binary classification\ntasks compared with other loss functions.\n"}
{"abstract": "  Social events provide valuable insights into group social behaviors and\npublic concerns and therefore have many applications in fields such as product\nrecommendation and crisis management. The complexity and streaming nature of\nsocial messages make it appealing to address social event detection in an\nincremental learning setting, where acquiring, preserving, and extending\nknowledge are major concerns. Most existing methods, including those based on\nincremental clustering and community detection, learn limited amounts of\nknowledge as they ignore the rich semantics and structural information\ncontained in social data. Moreover, they cannot memorize previously acquired\nknowledge. In this paper, we propose a novel Knowledge-Preserving Incremental\nHeterogeneous Graph Neural Network (KPGNN) for incremental social event\ndetection. To acquire more knowledge, KPGNN models complex social messages into\nunified social graphs to facilitate data utilization and explores the\nexpressive power of GNNs for knowledge extraction. To continuously adapt to the\nincoming data, KPGNN adopts contrastive loss terms that cope with a changing\nnumber of event classes. It also leverages the inductive learning ability of\nGNNs to efficiently detect events and extends its knowledge from previously\nunseen data. To deal with large social streams, KPGNN adopts a mini-batch\nsubgraph sampling strategy for scalable training, and periodically removes\nobsolete data to maintain a dynamic embedding space. KPGNN requires no feature\nengineering and has few hyperparameters to tune. Extensive experiment results\ndemonstrate the superiority of KPGNN over various baselines.\n"}
{"abstract": "  In this article we disprove the conjectures asserting the positivity of the\ncoefficients of the Ehrhart polynomial of matroid polytopes by De Loera, Haws\nand K\\\"oppe (2007) and of generalized permutohedra by Castillo and Liu (2018).\nWe prove constructively that for every $n\\geq 19$ there exist connected\nmatroids on $n$ elements that are not Ehrhart positive. Also, we prove that for\nevery $k\\geq 3$ there exist connected matroids of rank $k$ that are not Ehrhart\npositive.\n"}
{"abstract": "  High-resolution cosmological hydrodynamic simulations are currently limited\nto relatively small volumes due to their computational expense. However, much\nlarger volumes are required to probe rare, overdense environments, and measure\nclustering statistics of the large scale structure. Typically, zoom simulations\nof individual regions are used to study rare environments, and semi-analytic\nmodels and halo occupation models applied to dark matter only (DMO) simulations\nare used to study the Universe in the large-volume regime. We propose a new\napproach, using a machine learning framework to explore the halo-galaxy\nrelationship in the periodic EAGLE simulations, and zoom C-EAGLE simulations of\ngalaxy clusters. We train a tree based machine learning method to predict the\nbaryonic properties of galaxies based on their host dark matter halo\nproperties. The trained model successfully reproduces a number of key\ndistribution functions for an infinitesimal fraction of the computational cost\nof a full hydrodynamic simulation. By training on both periodic simulations as\nwell as zooms of overdense environments, we learn the bias of galaxy evolution\nin differing environments. This allows us to apply the trained model to a\nlarger DMO volume than would be possible if we only trained on a periodic\nsimulation. We demonstrate this application using the $(800 \\; \\mathrm{Mpc})^3$\nP-Millennium simulation, and present predictions for key baryonic distribution\nfunctions and clustering statistics from the EAGLE model in this large volume.\n"}
{"abstract": "  We work with continuous-time, continuous-space stochastic dynamical systems\ndescribed by stochastic differential equations (SDE). We present a new approach\nto compute probabilistic safety regions, namely sets of initial conditions of\nthe SDE associated to trajectories that are safe with a probability larger than\na given threshold. We introduce a functional that is minimised at the border of\nthe probabilistic safety region, then solve an optimisation problem using\ntechniques from Malliavin Calculus that computes such region. Unlike existing\nresults in the literature, this new approach allows one to compute\nprobabilistic safety regions without gridding the state space of the SDE.\n"}
{"abstract": "  Principal component analysis (PCA) is a versatile tool to reduce the\ndimensionality which has wide applications in statistics and machine learning\ncommunity. It is particularly useful to model data in high-dimensional\nscenarios where the number of variables $p$ is comparable to, or much larger\nthan the sample size $n$. Despite extensive literature on this topic,\nresearchers have focused on modeling static principal eigenvectors or\nsubspaces, which are not suitable for stochastic processes that are dynamic in\nnature. To characterize the change in the whole course of high-dimensional data\ncollection, we propose a unified framework to directly estimate dynamic\nprincipal subspaces spanned by leading eigenvectors of covariance matrices. We\nformulate an optimization problem by combining the local linear smoothing and\nregularization penalty together with the orthogonality constraint, which can be\neffectively solved by the proximal gradient method for manifold optimization.\nWe show that our method is suitable for high-dimensional data observed under\nboth common and irregular designs. In addition, theoretical properties of the\nestimators are investigated under $l_q (0 \\leq q \\leq 1)$ sparsity. Extensive\nexperiments demonstrate the effectiveness of the proposed method in both\nsimulated and real data examples.\n"}
{"abstract": "  Self-supervised learning of depth map prediction and motion estimation from\nmonocular video sequences is of vital importance -- since it realizes a broad\nrange of tasks in robotics and autonomous vehicles. A large number of research\nefforts have enhanced the performance by tackling illumination variation,\nocclusions, and dynamic objects, to name a few. However, each of those efforts\ntargets individual goals and endures as separate works. Moreover, most of\nprevious works have adopted the same CNN architecture, not reaping\narchitectural benefits. Therefore, the need to investigate the inter-dependency\nof the previous methods and the effect of architectural factors remains. To\nachieve these objectives, we revisit numerous previously proposed\nself-supervised methods for joint learning of depth and motion, perform a\ncomprehensive empirical study, and unveil multiple crucial insights.\nFurthermore, we remarkably enhance the performance as a result of our study --\noutperforming previous state-of-the-art performance.\n"}
{"abstract": "  We study the performance of the hybrid template-machine-learning photometric\nredshift (photo-$z$) algorithm Delight, which uses Gaussian processes, on a\nsubset of the early data release of the Physics of the Accelerating Universe\nSurvey (PAUS). We calibrate the fluxes of the $40$ PAUS narrow bands with $6$\nbroadband fluxes ($uBVriz$) in the COSMOS field using three different methods,\nincluding a new method which utilises the correlation between the apparent size\nand overall flux of the galaxy. We use a rich set of empirically derived galaxy\nspectral templates as guides to train the Gaussian process, and we show that\nour results are competitive with other standard photometric redshift\nalgorithms. Delight achieves a photo-$z$ $68$th percentile error of\n$\\sigma_{68}=0.0081(1+z)$ without any quality cut for galaxies with\n$i_\\mathrm{auto}<22.5$ as compared to $0.0089(1+z)$ and $0.0202(1+z)$ for the\nBPz and ANNz2 codes, respectively. Delight is also shown to produce more\naccurate probability distribution functions for individual redshift estimates\nthan BPz and ANNz2. Common photo-$z$ outliers of Delight and BCNz2 (previously\napplied to PAUS) are found to be primarily caused by outliers in the narrowband\nfluxes, with a small number of cases potentially indicating spectroscopic\nredshift failures in the reference sample. In the process, we introduce\nperformance metrics derived from the results of BCNz2 and Delight, allowing us\nto achieve a photo-$z$ quality of $\\sigma_{68}<0.0035(1+z)$ at a magnitude of\n$i_\\mathrm{auto}<22.5$ while keeping $50$ per cent objects of the galaxy\nsample.\n"}
{"abstract": "  Convolutional neural networks (CNNs) are highly successful for\nsuper-resolution (SR) but often require sophisticated architectures with heavy\nmemory cost and computational overhead, significantly restricts their practical\ndeployments on resource-limited devices. In this paper, we proposed a novel\ncontrastive self-distillation (CSD) framework to simultaneously compress and\naccelerate various off-the-shelf SR models. In particular, a channel-splitting\nsuper-resolution network can first be constructed from a target teacher network\nas a compact student network. Then, we propose a novel contrastive loss to\nimprove the quality of SR images and PSNR/SSIM via explicit knowledge transfer.\nExtensive experiments demonstrate that the proposed CSD scheme effectively\ncompresses and accelerates several standard SR models such as EDSR, RCAN and\nCARN. Code is available at https://github.com/Booooooooooo/CSD.\n"}
{"abstract": "  We report on the onset of anti-resonant behaviour of mass transport systems\ndriven by time-dependent forces. Anti-resonances arise from the coupling of a\nsufficiently high number of space-time modes of the force. The presence of\nforces having a wide space-time spectrum, a necessary condition for the\nformation of an anti-resonance, is typical of confined systems with uneven and\ndeformable walls that induce entropic forces dependent on space and time. We\nhave analyzed, in particular, the case of polymer chains confined in a flexible\nchannel and shown how they can be sorted and trapped. The presence of\nresonance-antiresonance pairs found can be exploited to design protocols able\nto engineer optimal transport processes and to manipulate the dynamics of\nnano-objects.\n"}
{"abstract": "  The very first detection of gravitational waves from a neutron star binary\nmerger, GW170817, exceeded all expectations. The event was relatively nearby,\nwhich may point to a relatively high merger rate. It was possible to extract\nfinite-size effects from the gravitational-wave signal, which constrains the\nnuclear equation of state. Also, an electromagnetic counterpart was detected at\nmany wavebands from radio to gamma rays marking the begin of a new\nmulti-messenger era involving gravitational waves. We describe how\nmulti-messenger observations of GW170817 are employed to constrain the nuclear\nequation of state. Combining the information from the optical emission and the\nmass measurement through gravitational waves leads to a lower limit on neutron\nstar radii. According to this conservative analysis, which employs a minimum\nset of assumptions, the radii of neutron stars with typical masses should be\nlarger than about 10.7~km. This implies a lower limit on the tidal\ndeformability of about 210, while much stronger lower bounds are not supported\nby the data of GW170817. The multi-messenger interpretation of GW170817 rules\nout very soft nuclear matter and complements the upper bounds on NS radii which\nare derived from the measurement of finite-size effects during the pre-merger\nphase. We highlight the future potential of multi-messenger observations and of\nGW measurements of the postmerger phase for constraining the nuclear equation\nof state. Finally, we propose an observing strategy to maximize the scientific\nyield of future multi-messenger observations.\n"}
{"abstract": "  We describe the installation, commissioning, and characterization of the new\ninjection kicker system in the Muon $g-2$ Experiment (E989) at Fermilab, which\nmakes a precision measurement of the muon magnetic anomaly. Three Blumlein\npulsers drive each of the 1.27-m-long non-ferric kicker magnets, which reside\nin a storage ring vacuum (SRV) that is subjected to a 1.45 T magnetic field.\nThe new system has been redesigned relative to Muon $g-2$'s predecessor\nexperiment, and we present those details in this manuscript.\n"}
{"abstract": "  Aim: to evaluate the ability of the nonlinear 24-HRV as a predictor of MI\nusing Machine Learning Methods: The sample was composed of 218 patients divided\ninto two groups (Healthy, n=128; MI n=90). The sample dataset is part of the\nTelemetric and Holter Electrocardiogram Warehouse (THEW) database, from the\nUniversity of Rochester Medical Center. We used the most common ML algorithms\nfor accuracy comparison with a setting of 10-fold cross-validation (briefly,\nLinear Regression, Linear Discriminant Analysis, k-Nearest Neighbour, Random\nForest, Supporting Vector Machine, Na\\\"ive Bayes, C 5.0 and Stochastic Gradient\nBoosting). Results: The main findings of this study show that the combination\nof SD1nu + SD2nu has greater predictive power for MI in comparison to other HRV\nindexes. Conclusion: The ML model using nonlinear HRV indexes showed to be more\neffective than the linear domain, evidenced through the application of ML,\nrepresented by a good precision of the Stochastic Gradient Boosting model.\n  Keywords: heart rate variability, machine learning, nonlinear domain,\ncardiovascular disease\n"}
{"abstract": "  Dialect and standard language identification are crucial tasks for many\nArabic natural language processing applications. In this paper, we present our\ndeep learning-based system, submitted to the second NADI shared task for\ncountry-level and province-level identification of Modern Standard Arabic (MSA)\nand Dialectal Arabic (DA). The system is based on an end-to-end deep Multi-Task\nLearning (MTL) model to tackle both country-level and province-level MSA/DA\nidentification. The latter MTL model consists of a shared Bidirectional Encoder\nRepresentation Transformers (BERT) encoder, two task-specific attention layers,\nand two classifiers. Our key idea is to leverage both the task-discriminative\nand the inter-task shared features for country and province MSA/DA\nidentification. The obtained results show that our MTL model outperforms\nsingle-task models on most subtasks.\n"}
{"abstract": "  Given a family of null hypotheses $H_{1},\\ldots,H_{s}$, we are interested in\nthe hypothesis $H_{s}^{\\gamma}$ that at most $\\gamma-1$ of these null\nhypotheses are false. Assuming that the corresponding $p$-values are\nindependent, we are investigating combined $p$-values that are valid for\ntesting $H_{s}^{\\gamma}$. In various settings in which $H_{s}^{\\gamma}$ is\nfalse, we determine which combined $p$-value works well in which setting. Via\nsimulations, we find that the Stouffer method works well if the null $p$-values\nare uniformly distributed and the signal strength is low, and the Fisher method\nworks better if the null $p$-values are conservative, i.e. stochastically\nlarger than the uniform distribution. The minimum method works well if the\nevidence for the rejection of $H_{s}^{\\gamma}$ is focused on only a few\nnon-null $p$-values, especially if the null $p$-values are conservative.\nMethods that incorporate the combination of $e$-values work well if the null\nhypotheses $H_{1},\\ldots,H_{s}$ are simple.\n"}
{"abstract": "  This work is concerned with practical stabilization of nonlinear systems by\nmeans of inf-convolution-based sample-and-hold control. It is a fairly general\nstabilization technique based on a generic non-smooth control Lyapunov function\n(CLF) and robust to actuator uncertainty, measurement noise, etc. The\nstabilization technique itself involves computation of descent directions of\nthe CLF. It turns out that non-exact realization of this computation leads not\njust to a quantitative, but also qualitative obstruction in the sense that the\nresult of the computation might fail to be a descent direction altogether and\nthere is also no straightforward way to relate it to a descent direction.\nDisturbance, primarily measurement noise, complicate the described issue even\nmore. This work suggests a modified inf-convolution-based control that is\nrobust w. r. t. system and measurement noise, as well as computational\nuncertainty. The assumptions on the CLF are mild, as, e. g., any piece-wise\nsmooth function, which often results from a numerical LF/CLF construction,\nsatisfies them. A computational study with a three-wheel robot with dynamical\nsteering and throttle under various tolerances w. r. t. computational\nuncertainty demonstrates the relevance of the addressed issue and the necessity\nof modifying the used stabilization technique. Similar analyses may be extended\nto other methods which involve optimization, such as Dini aiming or steepest\ndescent.\n"}
{"abstract": "  Heisenberg's uncertainty principle is often cited as an example of a \"purely\nquantum\" relation with no analogue in the classical limit where $\\hbar \\to 0$.\nHowever, this formulation of the classical limit is problematic for many\nreasons, one of which is dimensional analysis. Since $\\hbar$ is a dimensionful\nconstant, we may always work in natural units in which $\\hbar = 1$. Dimensional\nanalysis teaches us that all physical laws can be expressed purely in terms of\ndimensionless quantities. This indicates that the existence of a dimensionally\nconsistent constraint on $\\Delta x \\Delta p$ requires the existence of a\ndimensionful parameter with units of action, and that any definition of the\nclassical limit must be formulated in terms of dimensionless quantities (such\nas quantum numbers). Therefore, bounds on classical uncertainty (formulated in\nterms of statistical ensembles) can only be written in terms of dimensionful\nscales of the system under consideration, and can be readily compared to their\nquantum counterparts after being non-dimensionalized. We compare the\nuncertainty of certain coupled classical systems and their quantum counterparts\n(such as harmonic oscillators and particles in a box), and show that they\nconverge in the classical limit. We find that since these systems feature\nadditional dimensionful scales, the uncertainty bounds are dependent on\nmultiple dimensionless parameters, in accordance with dimensional\nconsiderations.\n"}
{"abstract": "  Here we propose an NPT metadynamics simulation scheme for pressure-induced\nstructural phase transitions, using coordination number and volume as\ncollective variables, and apply it to the reconstructive structural\ntransformation $B1$-$B2$ in NaCl. By studying systems with size up to 64 000\natoms we reach a regime beyond collective mechanism and observe transformations\nproceeding via nucleation and growth. We also reveal the crossover of the\ntransition mechanism from Buerger-like for smaller systems to\nWatanabe-Tol\\'{e}dano for larger ones. The scheme is likely to be applicable to\na broader class of pressure-induced structural transitions, allowing study of\ncomplex nucleation effects and bringing simulations closer to realistic\nconditions.\n"}
{"abstract": "  Using a geometric formalism of elasticity theory we develop a systematic\ntheoretical method for controlling and manipulating the mechanical response of\nslender solids to external loads. We formally express global mechanical\nproperties associated with non-euclidean thin sheets, and interpret the\nexpressions as inverse problem for designing desired mechanical properties. We\nshow that by wisely designing geometric frustration, extreme mechanical\nproperties can be encoded into a material using accessible experimental\ntechniques. To test the methodology we derive a family of geometries that\nresult with anomalous mechanical behavior such as tunable, an-harmonic, and\neven vanishing rigidities. The presented formalism can be discretized and thus\nopens a new pathway for the design of both continuum and discrete solids and\nstructures.\n"}
{"abstract": "  Crossover and mutation are the two main operators that lead to new solutions\nin evolutionary approaches. In this article, a new method of performing the\ncrossover phase is presented. The problem of choice is evolutionary decision\ntree construction. The method aims at finding such individuals that together\ncomplement each other. Hence we say that they are diversely specialized. We\npropose the way of calculating the so-called complementary fitness. In several\nempirical experiments, we evaluate the efficacy of the method proposed in four\nvariants and compare it to a fitness-rank-based approach. One variant emerges\nclearly as the best approach, whereas the remaining ones are below the\nbaseline.\n"}
{"abstract": "  Magnesium and its alloys have aroused tremendous interests because of their\npromising mechanical properties and biocompatibility. However, their\nexcessively fast corrosion rate hinders the development of Mg alloys in the\nbiomedical fields. Inspired by conventional ion implantation, a less-toxic\nfunctional group (hydroxyl) is used as the ion source to bombard the ZK60 Mg\nalloy surface to form a functionalized oxide layer. This functionalized oxide\nlayer significantly facilitates the corrosion resistance of the ZK60 Mg alloy\nsubstrate and the proliferation of MC3T3-E1 cells, which is confirmed by\nelectrochemical, immersion, and in vitro cytocompatibility tests. In comparison\nwith results of ZK60 alloy implanted with carboxyl ions in our previous work,\nit is concluded that hydroxyl-treated alloys exhibit slightly higher corrosion\nrate while better biocompatibility. In summary, less-toxic functional ion\nimplantation can be an effective strategy for inhibiting corrosion of Mg alloy\nimplants and promoting their biocompatibility.\n"}
{"abstract": "  Let $G = (V, E)$ be a simple graph of order $n$. The total dominating set of\n$G$ is a subset $D$ of $V$ that every vertex of $V$ is adjacent to some\nvertices of $D$. The total domination number of $G$ is equal to minimum\ncardinality of total dominating set in $G$ and is denoted by $\\gamma_t(G)$. The\ntotal domination polynomial of $G$ is the polynomial\n$D_t(G,x)=\\sum_{i=\\gamma_t(G)}^n d_t(G,i)x^i$, where $d_t(G,i)$ is the number\nof total dominating sets of $G$ of size $i$. Two graphs $G$ and $H$ are said to\nbe total dominating equivalent or simply $\\mathcal{D}_t$-equivalent, if\n$D_t(G,x)=D_t(H,x)$. The equivalence class of $G$, denoted $[G]$, is the set of\nall graphs $\\mathcal{D}_t$-equivalent to $G$. In this paper, we investigate\n$\\mathcal{D}_t$-equivalence classes of some graphs. Also we introduce some\nfamilies of graphs whose total domination polynomials are unimodal.\n"}
{"abstract": "  We evaluate critically the use of pulsed gradient spin-echo nuclear magnetic\nresonance (PGSE NMR) to measure molecular mobility during chemical reactions.\nWith raw NMR spectra available in a public depository, we confirm boosted\nmobility during the click chemical reaction (Science 2020, 369, 537) regardless\nof the order of magnetic field gradient (linearly-increasing,\nlinearly-decreasing, random sequence). We also confirm boosted mobility for the\nDiels-Alder chemical reaction. The conceptual advantage of the former chemical\nsystem is that constant reaction rate implies constant catalyst concentration,\nwhereas that of the latter is the absence of a paramagnetic catalyst,\nprecluding paramagnetism as objection to the measurements. Data and discussion\nin this paper show the reliability of experiments when one avoids convection,\nallows decay of nuclear spin magnetization between successive pulses and\nrecovery of its intensity between gradients, and satisfies quasi-steady state\nduring the time window to acquire each datum. Especially important is to make\ncomparisons on the time scale of actual chemical reaction kinetics. We discuss\npossible sources of mistaken conclusions that are desirable to avoid.\n"}
{"abstract": "  The adoption of neural networks and deep learning in non-Euclidean domains\nhas been hindered until recently by the lack of scalable and efficient learning\nframeworks. Existing toolboxes in this space were mainly motivated by research\nand education use cases, whereas practical aspects, such as deploying and\nmaintaining machine learning models, were often overlooked.\n  We attempt to bridge this gap by proposing TensorFlow RiemOpt, a Python\nlibrary for optimization on Riemannian manifolds in TensorFlow. The library is\ndesigned with the aim for a seamless integration with the TensorFlow ecosystem,\ntargeting not only research, but also streamlining production machine learning\npipelines.\n"}
{"abstract": "  A new $\\gamma$-ray source with a significance level of approximately\n5$\\sigma$ was reported in the region of SNR G272.2-3.2, analysing the\napproximately 12.4 years of observation data from the Fermi Large Area\nTelescope (Fermi-LAT). Its $\\gamma$-ray spatial distribution did not show\nextended feature, and it had a soft spectrum with the spectral index of\n2.56$\\pm$0.01 of a power-law model. No significant variability of its light\ncurve (LC) with 10 time bins was identified, and its spatial positions in the\nX-ray and GeV bands overlapped. We suggest that the new $\\gamma$-ray source is\na likely counterpart of SNR G272.2-3.2. Analysing its spectrum, we discussed\nthe likely origins of the $\\gamma$-ray emission.\n"}
{"abstract": "  We describe an infinite family of graphs $G_n$, where $G_n$ has $n$ vertices,\nindependence number at least $n/4$, and no set of less than $\\sqrt{n}/2$\nvertices intersects all its maximum independent sets. This is motivated by a\nquestion of Bollob\\'as, Erd\\H{o}s and Tuza, and disproves a recent conjecture\nof Friedgut, Kalai and Kindler. Motivated by a related question of the last\nauthors, we show that for every graph $G$ on $n$ vertices with independence\nnumber $(1/4+\\eps)n$, the average independence number of an induced subgraph of\n$G$ on a uniform random subset of the vertices is at most\n$(1/4+\\eps-\\Omega(\\eps^2)) n$.\n"}
{"abstract": "  In the current paradigm of planet formation research, it is believed that the\nfirst step to forming massive bodies (such as asteroids and planets) requires\nthat small interstellar dust grains floating through space collide with each\nother and grow to larger sizes. The initial formation of these pebbles is\ngoverned by an integro-differential equation known as the Smoluchowski\ncoagulation equation, to which analytical solutions are intractable for all but\nthe simplest possible scenarios. While brute-force methods of approximation\nhave been developed, they are computationally costly, currently making it\ninfeasible to simulate this process including other physical processes relevant\nto planet formation, and across the very large range of scales on which it\noccurs. In this paper, we take a machine learning approach to designing a\nsystem for a much faster approximation. We develop a multi-output random forest\nregression model trained on brute-force simulation data to approximate\ndistributions of dust particle sizes in protoplanetary disks at different\npoints in time. The performance of our random forest model is measured against\nthe existing brute-force models, which are the standard for realistic\nsimulations. Results indicate that the random forest model can generate highly\naccurate predictions relative to the brute-force simulation results, with an\n$R^{2}$ of 0.97, and do so significantly faster than brute-force methods.\n"}
{"abstract": "  In this paper, we get an elementary and important lemma(See Lemma 3.2) which\nis about pushout and pullback of modules. And we prove a weak form of a long\nopen conjecture on vanishing of group cohomology for blocks.\n"}
{"abstract": "  A mismatch of atomic registries between single-layer transition metal\ndichalcogenides (TMDs) in a two dimensional van der Waals heterostructure\nproduces a moir\\'e superlattice with a periodic potential, which can be\nfine-tuned by introducing a twist angle between the materials. This approach is\npromising both for controlling the interactions between the TMDs and for\nengineering their electronic band structures, yet direct observation of the\nchanges to the electronic structure introduced with varying twist angle has so\nfar been missing. Here, we probe heterobilayers comprised of single-layer\nMoS$_2$ and WS$_2$ with twist angles of $(2.0 \\pm 0.5)^{\\circ}$, $(13.0 \\pm\n0.5)^{\\circ}$, and $(20.0 \\pm 0.5)^{\\circ}$ and investigate the differences in\ntheir electronic band structure using micro-focused angle-resolved\nphotoemission spectroscopy. We find strong interlayer hybridization between\nMoS$_2$ and WS$_2$ electronic states at the $\\bar{\\mathrm{\\Gamma}}$-point of\nthe Brillouin zone, leading to a transition from a direct bandgap in the\nsingle-layer to an indirect gap in the heterostructure. Replicas of the\nhybridized states are observed at the centre of twist angle-dependent moir\\'e\nmini Brillouin zones. We confirm that these replica features arise from the\ninherent moir\\'e potential by comparing our experimental observations with\ndensity functional theory calculations of the superlattice dispersion. Our\ndirect visualization of these features underscores the potential of using\ntwisted heterobilayer semiconductors to engineer hybrid electronic states and\nsuperlattices that alter the electronic and optical properties of 2D\nheterostructures.\n"}
{"abstract": "  We propose an efficient encoding algorithm for the binary and non-binary\nlow-density parity-check codes. This algorithm transforms the parity part of\nthe parity-check matrix into a block triangular matrix with low weight diagonal\nsubmatrices by row and column permutations in the preprocessing stage. This\nalgorithm determines the parity part of a codeword by block back-substitution\nto reduce the encoding complexity in the encoding stage. Numerical experiments\nshow that this algorithm has a lower encoding complexity than existing encoding\nalgorithms. Moreover, we show that this algorithm encodes the non-binary cycle\ncodes in linear time.\n"}
{"abstract": "  We describe our single submission to task 1 of COLIEE 2021. Our vanilla BM25\ngot second place, well above the median of submissions. Code is available at\nhttps://github.com/neuralmind-ai/coliee.\n"}
{"abstract": "  We have applied nuclear magnetic resonance spectroscopy to study the\ndistinctive network of nodal lines in the Dirac semimetal ZrSiTe. The low-$T$\nbehavior is dominated by a symmetry-protected nodal line, with NMR providing a\nsensitive probe of the diamagnetic response of the associated carriers. A sharp\nlow-$T$ minimum in NMR shift and $(T_1T)^{-1}$ provides a quantitative measure\nof the dispersionless, quasi-2D behavior of this nodal line. We also identify a\nvan Hove singularity closely connected to this nodal line, and an associated\n$T$-induced Lifshitz transition. A disconnect in the NMR shift and line width\nat this temperature indicates the change in electronic behavior associated with\nthis topological change. These features have an orientation-dependent behavior\nindicating a field-dependent scaling of the associated band energies.\n"}
{"abstract": "  The growing success of graph signal processing (GSP) approaches relies\nheavily on prior identification of a graph over which network data admit\ncertain regularity. However, adaptation to increasingly dynamic environments as\nwell as demands for real-time processing of streaming data pose major\nchallenges to this end. In this context, we develop novel algorithms for online\nnetwork topology inference given streaming observations assumed to be smooth on\nthe sought graph. Unlike existing batch algorithms, our goal is to track the\n(possibly) time-varying network topology while maintaining the memory and\ncomputational costs in check by processing graph signals sequentially-in-time.\nTo recover the graph in an online fashion, we leverage proximal gradient (PG)\nmethods to solve a judicious smoothness-regularized, time-varying optimization\nproblem. Under mild technical conditions, we establish that the online graph\nlearning algorithm converges to within a neighborhood of (i.e., it tracks) the\noptimal time-varying batch solution. Computer simulations using both synthetic\nand real financial market data illustrate the effectiveness of the proposed\nalgorithm in adapting to streaming signals to track slowly-varying network\nconnectivity.\n"}
{"abstract": "  In autonomous driving, using a variety of sensors to recognize preceding\nvehicles in middle and long distance is helpful for improving driving\nperformance and developing various functions. However, if only LiDAR or camera\nis used in the recognition stage, it is difficult to obtain necessary data due\nto the limitations of each sensor. In this paper, we proposed a method of\nconverting the tracking data of vision into bird's eye view (BEV) coordinates\nusing an equation that projects LiDAR points onto an image, and a method of\nfusion between LiDAR and vision tracked data. Thus, the newly proposed method\nwas effective through the results of detecting closest in-path vehicle (CIPV)\nin various situations. In addition, even when experimenting with the EuroNCAP\nautonomous emergency braking (AEB) test protocol using the result of fusion,\nAEB performance is improved through improved cognitive performance than when\nusing only LiDAR. In experimental results, the performance of the proposed\nmethod was proved through actual vehicle tests in various scenarios.\nConsequently, it is convincing that the newly proposed sensor fusion method\nsignificantly improves the ACC function in autonomous maneuvering. We expect\nthat this improvement in perception performance will contribute to improving\nthe overall stability of ACC.\n"}
{"abstract": "  In this paper we give a self-contained treatment of finite group quotients of\nadmissible (formal) schemes and adic spaces that are locally topologically\nfinite type over a locally strongly noetherian adic space.\n"}
{"abstract": "  We present a scalable approach for simplifying the stability analysis of\ncluster synchronization patterns on directed networks. When a network has\ndirectional couplings, decomposition of the coupling matrix into independent\nblocks (which in turn decouples the variational equation) is no longer adequate\nto reveal the full relations among perturbation modes. Instead, it is often\nnecessary to introduce directional dependencies among the blocks and establish\nhierarchies among perturbation modes. For this purpose, we develop an algorithm\nthat finds the simultaneous block upper triangularization of sets of asymmetric\nmatrices, which generalizes the Jordan canonical decomposition from a single\nmatrix to an arbitrary number of matrices. The block upper triangularization\norders subspaces of the variational equation in a directional manner, allowing\nthe stability of perturbation modes to be analyzed in sequence. We show that\nour algorithm gives the greatest possible simplification under mild\nassumptions, both in terms of the sizes of the blocks and in terms of the\nnumber of nonzero upper triangular entries linking the blocks.\n"}
{"abstract": "  Natural language generation (NLG) benchmarks provide an important avenue to\nmeasure progress and develop better NLG systems. Unfortunately, the lack of\npublicly available NLG benchmarks for low-resource languages poses a\nchallenging barrier for building NLG systems that work well for languages with\nlimited amounts of data. Here we introduce IndoNLG, the first benchmark to\nmeasure natural language generation (NLG) progress in three low-resource -- yet\nwidely spoken -- languages of Indonesia: Indonesian, Javanese, and Sundanese.\nAltogether, these languages are spoken by more than 100 million native\nspeakers, and hence constitute an important use case of NLG systems today.\nConcretely, IndoNLG covers six tasks: summarization, question answering,\nchit-chat, and three different pairs of machine translation (MT) tasks. We\ncollate a clean pretraining corpus of Indonesian, Sundanese, and Javanese\ndatasets, Indo4B-Plus, which is used to pretrain our models: IndoBART and\nIndoGPT. We show that IndoBART and IndoGPT achieve competitive performance on\nall tasks -- despite using only one-fifth the parameters of a larger\nmultilingual model, mBART-LARGE (Liu et al., 2020). This finding emphasizes the\nimportance of pretraining on closely related, local languages to achieve more\nefficient learning and faster inference for very low-resource languages like\nJavanese and Sundanese.\n"}
{"abstract": "  We present implicit displacement fields, a novel representation for detailed\n3D geometry. Inspired by a classic surface deformation technique, displacement\nmapping, our method represents a complex surface as a smooth base surface plus\na displacement along the base's normal directions, resulting in a\nfrequency-based shape decomposition, where the high frequency signal is\nconstrained geometrically by the low frequency signal. Importantly, this\ndisentanglement is unsupervised thanks to a tailored architectural design that\nhas an innate frequency hierarchy by construction. We explore implicit\ndisplacement field surface reconstruction and detail transfer and demonstrate\nsuperior representational power, training stability and generalizability.\n"}
{"abstract": "  We study the quantum-droplet state in a 3-dimensional (3D) Bose gas in the\npresence of 1D spin-orbit-coupling and Raman coupling, especially the stripe\nphase with density modulation, by numerically computing the ground state energy\nincluding the mean-field energy and Lee-HuangYang correction. In this droplet\nstate, the stripe can exist in a wider range of Raman coupling, compared with\nthe BEC-gas state. More intriguingly, both spin-orbit-coupling and Raman\ncoupling strengths can be used to tune the droplet density.\n"}
{"abstract": "  We have studied the masse spectra for the\n$cc\\bar{b}\\bar{b}$/$bb\\bar{c}\\bar{c}$ tetraquark states with quantum numbers\n$J^{P}=0^{\\pm},1^{\\pm}$, and $2^{+}$. We systematically construct the\ninterpolating currents with various spin-parity quantum numbers and calculate\ntheir two-point correlation functions in the framework of QCD moment sum rule\nmethod. Our calculations show that the masses are about $12.3-12.4$ GeV for the\npositive parity $cc\\bar{b}\\bar{b}$ tetraquark ground states with $J^{P}=0^+,\n1^+, 2^+$, while $12.8-13.1$ GeV for the negative parity channels with\n$J^{P}=0^-, 1^-$. The mass predictions for the positive parity\n$cc\\bar{b}\\bar{b}$ ground states are lower than the $B_{c}B_{c}$ threshold,\nimplying that these tetraquarks can only decay via weak interaction and thus\nare expected to be stable and narrow.\n"}
{"abstract": "  Measles is considered as a highly contagious disease that leads to serious\ncomplications around the world. Thus, the paper determined the trend and the\nfive-year forecasted data of the Measles in the Philippines. This study\nutilized the time series data for trend analysis and data forecasting using the\nARIMA model to visualize the measles cases. Figures for the time-series and\nforecasted results are individually presented with the use of GRETL software.\nResults showed that there was an increasing pattern of the disease from 2016 to\n2019. However, there was a decreasing pattern of its occurrence in the next\nfive years based on the five-year forecast. Nevertheless, with the results of\nthe study, there is still a need to improve the different intervention plans of\nthe authority in alleviating the occurrence of the disease though it yielded a\ndecreasing pattern in the future since it is evident that the figure of the\nforecasted data is still approximately 15,000 and above.\n"}
{"abstract": "  This paper considers an angle-domain intelligent reflecting surface (IRS)\nsystem. We derive maximum likelihood (ML) estimators for the effective angles\nfrom the base station (BS) to the user and the effective angles of propagation\nfrom the IRS to the user. It is demonstrated that the accuracy of the estimated\nangles improves with the number of BS antennas. Also, deploying the IRS closer\nto the BS increases the accuracy of the estimated angle from the IRS to the\nuser. Then, based on the estimated angles, we propose a joint optimization of\nBS beamforming and IRS beamforming, which achieves similar performance to two\nbenchmark algorithms based on full CSI and the multiple signal classification\n(MUSIC) method respectively. Simulation results show that the optimized BS beam\nbecomes more focused towards the IRS direction as the number of reflecting\nelements increases. Furthermore, we derive a closed-form approximation, upper\nbound and lower bound for the achievable rate. The analytical findings indicate\nthat the achievable rate can be improved by increasing the number of BS\nantennas or reflecting elements. Specifically, the BS-user link and the\nBS-IRS-user link can obtain power gains of order $N$ and $NM^2$, respectively,\nwhere $N$ is the antenna number and $M$ is the number of reflecting elements.\n"}
{"abstract": "  Because of the unique physical properties and potential applications, the\nexploration of quantum materials with diverse symmetry-protected topological\nstates has attracted considerable interest in the condensed-matter community in\nrecent years. Most of the topologically nontirvial materials identified thus\nfar have two-dimensional or three-dimensional structural characteristics, while\nthe quasi-one-dimensional (quasi-1D) analogs are rare. Here we report on\nanisotropic magnetoresistance, Hall effect, and quantum de Haas$-$van Alphen\n(dHvA) oscillations in TaPtTe$_5$ single crystals, which possess a layered\ncrystal structure with quasi-1D PtTe$_2$ chains. TaPtTe$_5$ manifests an\nanisotropic magnetoresistance and a nonlinear Hall effect at low temperatures.\nThe analysis of the dHvA oscillations reveals two major oscillation frequencies\n(63.5 T and 95.2 T). The corresponding light effective masses and the nonzero\nBerry phases suggest the nontrivial band topology in TaPtTe$_5$, which is\nfurther corroborated by the first-principles calculations. Our results suggest\nthat TaPtTe$_5$, in analogy with its sister compounds TaPdTe$_5$ and\nTaNiTe$_5$, is another quasi-1D material hosting topological Dirac fermions.\n"}
{"abstract": "  Classical results from Sturm-Liouville theory state that the number of\nunstable eigenvalues of a scalar, second-order linear operator is equal to the\nnumber of associated conjugate points. Recent work has extended these results\nto a much more general setting, thus allowing for spectral stability of\nnonlinear waves in a variety of contexts to be determined by counting conjugate\npoints. However, in practice, it is not yet clear whether it is easier to\ncompute conjugate points than to just directly count unstable eigenvalues. We\naddress this issue by developing a framework for the computation of conjugate\npoints using validated numerics. Moreover, we apply our method to a\nparameter-dependent system of bistable equations and show that there exist both\nstable and unstable standing fronts. This application can be seen as\ncomplimentary to the classical result via Sturm-Louiville theory that in scalar\nreaction-diffusion equations pulses are unstable whereas fronts are stable, and\nto the more recent result of \"Instability of pulses in gradient\nreaction-diffusion systems: a symplectic approach,\" by Beck et. al., that\nsymmetric pulses in reaction-diffusion systems with gradient nonlinearity are\nalso necessarily unstable.\n"}
{"abstract": "  For the classes of finite dimensional linear time-invariant semi-dissipative\nHamiltonian ordinary differential equations and differential-algebraic\nequations, stability and hypocoercivity are discussed and related to concepts\nfrom control theory. On the basis of staircase forms the solution behavior is\ncharacterized and connected to the hypocoercivity index of these evolution\nequations. The results are applied to two infinite dimensional flow problems.\n"}
{"abstract": "  The Fradkin-Schwinger functional methods to represent a Green function in an\nexternal gravitational field are used to study the eikonal and the\nnext-to-eikonal limit, including the nonlinear gravitational interactions, of\nthe scattering amplitudes of an ultra-relativistic scalar particle on a static\nsuper-massive scalar target in the nearly forward limit. The functional\napproach confirms the exponentiation of the leading eikonal which also applies\nto the first non-leading power in the energy of the light particle, moreover\nincludes the interaction at impact parameter much larger than the Schwarzschild\nradius associated with the center of mass energy in the ultra-relativistic\nlimit.\n"}
{"abstract": "  In this paper, the problem of vertical handover in software-defined network\n(SDN) based heterogeneous networks (HetNets) is studied. In the studied model,\nHetNets are required to offer diverse services for mobile users. Using an SDN\ncontroller, HetNets have the capability of managing users' access and mobility\nissues but still have the problems of ping-pong effect and service interruption\nduring vertical handover. To solve these problems, a mobility-aware seamless\nhandover method based on multipath transmission control protocol (MPTCP) is\nproposed. The proposed handover method is executed in the controller of the\nsoftware-defined HetNets (SDHetNets) and consists of three steps: location\nprediction, network selection, and handover execution. In particular, the\nmethod first predicts the user's location in the next moment with an echo state\nnetwork (ESN). Given the predicted location, the SDHetNet controller can\ndetermine the candidate network set for the handover to pre-allocate network\nwireless resources. Second, the target network is selected through fuzzy\nanalytic hierarchical process (FAHP) algorithm, jointly considering user\npreferences, service requirements, network attributes, and user mobility\npatterns. Then, seamless handover is realized through the proposed MPTCP-based\nhandover mechanism. Simulations using real-world user trajectory data from\nKorea Advanced Institute of Science & Technology show that the proposed method\ncan reduce the handover times by 10.85% to 29.12% compared with traditional\nmethods. The proposed method also maintains at least one MPTCP subflow\nconnected during the handover process and achieves a seamless handover.\n"}
{"abstract": "  A wide range of intuitionistic type theories may be presented as equational\ntheories within a logical framework. This method was formulated by Per\nMartin-L\\\"{o}f in the mid-1980's and further developed by Uemura, who used it\nto prove an initiality result for a class of models. Herein is presented a\nlogical framework for type theories that includes an extensional equality type\nso that a type theory may be given by a signature of constants. The framework\nis illustrated by a number of examples of type-theoretic concepts, including\nidentity and equality types, and a hierarchy of universes.\n"}
{"abstract": "  By exploiting the superiority of non-orthogonal multiple access (NOMA),\nNOMA-aided mobile edge computing (MEC) can provide scalable and low-latency\ncomputing services for the Internet of Things. However, given the prevalent\nstochasticity of wireless networks and sophisticated signal processing of NOMA,\nit is critical but challenging to design an efficient task offloading algorithm\nfor NOMA-aided MEC, especially under a large number of devices. This paper\npresents an online algorithm that jointly optimizes offloading decisions and\nresource allocation to maximize the long-term system utility (i.e., a measure\nof throughput and fairness). Since the optimization variables are temporary\ncoupled, we first apply Lyapunov technique to decouple the long-term stochastic\noptimization into a series of per-slot deterministic subproblems, which does\nnot require any prior knowledge of network dynamics. Second, we propose to\ntransform the non-convex per-slot subproblem of optimizing NOMA power\nallocation equivalently to a convex form by introducing a set of auxiliary\nvariables, whereby the time-complexity is reduced from the exponential\ncomplexity to $\\mathcal{O} (M^{3/2})$. The proposed algorithm is proved to be\nasymptotically optimal, even under partial knowledge of the device states at\nthe base station. Simulation results validate the superiority of the proposed\nalgorithm in terms of system utility, stability improvement, and the overhead\nreduction.\n"}
{"abstract": "  The rules of a game of dice are extended to a \"hyper-die\" with\n$n\\in\\mathbb{N}$ equally probable faces, numbered from 1 to $n$. We derive\nrecursive and explicit expressions for the probability mass function and the\ncumulative distribution function of the gain $G_n$ for arbitrary values of $n$.\nA numerical study suggests the conjecture that for $n \\to \\infty$ the\nexpectation of the scaled gain $\\mathbb{E}[H_n]=\\mathbb{E}[G_n/\\sqrt{n}\\,]$\nconverges to $\\sqrt{\\pi/\\,2}$. The conjecture is proved by deriving an analytic\nexpression of the expected gain $\\mathbb{E}[G_n]$. An analytic expression of\nthe variance of the gain $G_n$ is derived by a similar technique. Finally, it\nis proved that $H_n$ converges weakly to the Rayleigh distribution with scale\nparameter~1.\n"}
{"abstract": "  We revisit the computation of the trace anomaly for Weyl fermions using\ndimensional regularization. For a consistent treatment of the chiral gamma\nmatrix $\\gamma_*$ in dimensional regularization, we work in $n$ dimensions from\nthe very beginning and use the Breitenlohner-Maison scheme to define\n$\\gamma_*$. We show that the parity-odd contribution to the trace anomaly\nvanishes (for which the use of dimension-dependent identities is crucial), and\nthat the parity-even contribution is half the one of a Dirac fermion. To arrive\nat this result, we compute the full renormalized expectation value of the\nfermion stress tensor to second order in perturbations around Minkowski\nspacetime, and also show that it is conserved.\n"}
{"abstract": "  The chirality-induced spin selectivity (CISS) effect enables the detection of\nchirality as electrical charge signals. It is often studied using a\ntwo-terminal circuit geometry where a ferromagnet is connected to a chiral\ncomponent, and a change of electrical resistance is reported upon magnetization\nreversal. This is however not expected in the linear response regime because of\ncompensating reciprocal processes, limiting the interpretation of experimental\nresults. Here we show that magnetoresistance effects can indeed appear even in\nthe linear response regime, either by changing the magnitude or the direction\nof the magnetization or an applied magnetic field. We illustrate this in a\nspin-valve device and in a chiral thin film as the CISS-induced Hanle\nmagnetoresistance (CHMR) effect. This effect helps to distinguish\nspin-transport-related effects from other effects, and can thereby provide\nfurther insight into the origin of CISS.\n"}
{"abstract": "  Semi-supervised domain adaptation (SSDA), which aims to learn models in a\npartially labeled target domain with the assistance of the fully labeled source\ndomain, attracts increasing attention in recent years. To explicitly leverage\nthe labeled data in both domains, we naturally introduce a conditional GAN\nframework to transfer images without changing the semantics in SSDA. However,\nwe identify a label-domination problem in such an approach. In fact, the\ngenerator tends to overlook the input source image and only memorizes\nprototypes of each class, which results in unsatisfactory adaptation\nperformance. To this end, we propose a simple yet effective Relaxed conditional\nGAN (Relaxed cGAN) framework. Specifically, we feed the image without its label\nto our generator. In this way, the generator has to infer the semantic\ninformation of input data. We formally prove that its equilibrium is desirable\nand empirically validate its practical convergence and effectiveness in image\ntransfer. Additionally, we propose several techniques to make use of unlabeled\ndata in the target domain, enhancing the model in SSDA settings. We validate\nour method on the well-adopted datasets: Digits, DomainNet, and Office-Home. We\nachieve state-of-the-art performance on DomainNet, Office-Home and most digit\nbenchmarks in low-resource and high-resource settings.\n"}
{"abstract": "  In many different fields interactions between objects play a critical role in\ndetermining their behavior. Graph neural networks (GNNs) have emerged as a\npowerful tool for modeling interactions, although often at the cost of adding\nconsiderable complexity and latency. In this paper, we consider the problem of\nspatial interaction modeling in the context of predicting the motion of actors\naround autonomous vehicles, and investigate alternative approaches to GNNs. We\nrevisit convolutions and show that they can demonstrate comparable performance\nto graph networks in modeling spatial interactions with lower latency, thus\nproviding an effective and efficient alternative in time-critical systems.\nMoreover, we propose a novel interaction loss to further improve the\ninteraction modeling of the considered methods.\n"}
{"abstract": "  This article introduces a novel method for the implementation of shape\noptimisation with Lipschitz domains. We propose to use the shape derivative to\ndetermine deformation fields which represent steepest descent directions of the\nshape functional in the $W^{1,\\infty}-$ topology. The idea of our approach is\ndemonstrated for shape optimisation of $n$-dimensional star-shaped domains,\nwhich we represent as functions defined on the unit $(n-1)$-sphere. In this\nsetting we provide the specific form of the shape derivative and prove the\nexistence of solutions to the underlying shape optimisation problem. Moreover,\nwe show the existence of a direction of steepest descent in the $W^{1,\\infty}-$\ntopology. We also note that shape optimisation in this context is closely\nrelated to the $\\infty-$Laplacian, and to optimal transport, where we highlight\nthe latter in the numerics section. We present several numerical experiments in\ntwo dimensions illustrating that our approach seems to be superior over a\nwidely used Hilbert space method in the considered examples, in particular in\ndeveloping optimised shapes with corners.\n"}
{"abstract": "  It is important to know the accurate trajectory of a free fall object in\nfluid (such as a spacecraft), whose motion might be chaotic in many cases.\nHowever, it is impossible to accurately predict its chaotic trajectory in a\nlong enough duration by traditional numerical algorithms in double precision.\nIn this paper, we give the accurate predictions of the same problem by a new\nstrategy, namely the Clean Numerical Simulation (CNS). Without loss of\ngenerality, a free fall disk in water is considered, whose motion is governed\nby the Andersen-Pesavento-Wang model. We illustrate that convergent and\nreliable trajectories of a chaotic free fall disk in a long enough interval of\ntime can be obtained by means of the CNS, but different traditional algorithms\nin double precision give disparate trajectories. Besides, unlike the\ntraditional algorithms in double precision, the CNS can predict the accurate\nposture of the free fall disk near the vicinity of the bifurcation point of\nsome physical parameters in a long duration. Therefore, the CNS can provide\nreliable prediction of chaotic systems in a long enough interval of time.\n"}
{"abstract": "  We elucidate how the interplay between diffusive mobility and number\nconservation can make nearly phase-ordered active XY spins on a substrate\nstable. For wide-ranging model parameters, it has stable uniform phases with\norientational order logarithmically stronger or weaker than in equilibrium,\ntogether with miniscule (i.e., hyperuniform) or giant number fluctuations,\nrespectively, forming a new universality class. For other parameters, it has no\nstable uniformly ordered phase. Our theory sheds light on wide-ranging systems,\ne.g., active superfluids on substrates, synchronization of oscillators, active\ncarpets of cilia and bacterial flagella, and sandblasting.\n"}
{"abstract": "  Intersections where vehicles are permitted to turn and interact with\nvulnerable road users (VRUs) like pedestrians and cyclists are among some of\nthe most challenging locations for automated and accurate recognition of road\nusers' behavior. In this paper, we propose a deep conditional generative model\nfor interaction detection at such locations. It aims to automatically analyze\nmassive video data about the continuity of road users' behavior. This task is\nessential for many intelligent transportation systems such as traffic safety\ncontrol and self-driving cars that depend on the understanding of road users'\nlocomotion. A Conditional Variational Auto-Encoder based model with Gaussian\nlatent variables is trained to encode road users' behavior and perform\nprobabilistic and diverse predictions of interactions. The model takes as input\nthe information of road users' type, position and motion automatically\nextracted by a deep learning object detector and optical flow from videos, and\ngenerates frame-wise probabilities that represent the dynamics of interactions\nbetween a turning vehicle and any VRUs involved. The model's efficacy was\nvalidated by testing on real--world datasets acquired from two different\nintersections. It achieved an F1-score above 0.96 at a right--turn intersection\nin Germany and 0.89 at a left--turn intersection in Japan, both with very busy\ntraffic flows.\n"}
{"abstract": "  For a root system R, a field K and an invertible element q in K let U be the\nassociated quantum group, defined via Lusztig's divided powers construction. We\nstudy the irreducible characters of this algebra with integral (but not\nnecessarily dominant) highest weight. If the l-th cyclotomic polynomial\nvanishes when evaluated at q, then these characters exhibit a certain\nl-periodicity.\n"}
{"abstract": "  Abraham, Dolev, Geffner, and Halpern proved that, in asynchronous systems, a\n$(k,t)$-robust equilibrium for $n$ players and a trusted mediator can be\nimplemented without the mediator as long as $n > 4(k+t)$, where an equilibrium\nis $(k,t)$-robust if, roughly speaking, no coalition of $t$ players can\ndecrease the payoff of any of the other players, and no coalition of $k$\nplayers can increase their payoff by deviating. We prove that this bound is\ntight, in the sense that if $n \\le 4(k+t)$ there exist $(k,t)$-robust\nequilibria with a mediator that cannot be implemented by the players alone.\nEven though implementing $(k,t)$-robust mediators seems closely related to\nimplementing asynchronous multiparty $(k+t)$-secure computation \\cite{BCG93},\nto the best of our knowledge there is no known straightforward reduction from\none problem to another. Nevertheless, we show that there is a non-trivial\nreduction from a slightly weaker notion of $(k+t)$-secure computation, which we\ncall $(k+t)$-strict secure computation, to implementing $(k,t)$-robust\nmediators. We prove the desired lower bound by showing that there are functions\non $n$ variables that cannot be $(k+t)$-strictly securely computed if $n \\le\n4(k+t)$. This also provides a simple alternative proof for the well-known lower\nbound of $4t+1$ on asynchronous secure computation in the presence of up to $t$\nmalicious agents.\n"}
{"abstract": "  Artificial audition aims at providing hearing capabilities to machines,\ncomputers and robots. Existing frameworks in robot audition offer interesting\nsound source localization, tracking and separation performance, but involve a\nsignificant amount of computations that limit their use on robots with embedded\ncomputing capabilities. This paper presents ODAS, the Open embeddeD Audition\nSystem framework, which includes strategies to reduce the computational load\nand perform robot audition tasks on low-cost embedded computing systems. It\npresents key features of ODAS, along with cases illustrating its uses in\ndifferent robots and artificial audition applications.\n"}
{"abstract": "  In 1957 Feynman suggested that the quantum/classical character of gravity can\nbe assessed by the presence/absence of entanglement between gravitationally\ninteracting test masses. However, in all proposed experimental realisations\nusing matter-wave interferometry the extreme weakness of this interaction\nrequires pure initial states with extreme squeezing to achieve measurable\nentanglement for reasonable interaction times. In practice, the systems that\ncan be prepared in such nonclassical states are limited to small masses, which\nin turn limits the rate at which they get entangled. Here we address this key\nchallenge - the weakness of gravitational interaction - by using a massive body\nas an amplifying mediator of gravitational interaction between two\ntest-systems. Our analysis shows, that this results in an effective interaction\nbetween the two test-systems that grows with the mass of the mediator and is\nindependent of its initial state and, therefore, its temperature. This greatly\nreduces the requirement on the mass and degree of delocalization of the test\nsystems and, while still highly challenging, brings experiments on\ngravitational source masses a step closer to reality.\n"}
{"abstract": "  In this paper we present an adaptive deep density approximation strategy\nbased on KRnet (ADDA-KR) for solving the steady-state Fokker-Planck (F-P)\nequations. F-P equations are usually high-dimensional and defined on an\nunbounded domain, which limits the application of traditional grid based\nnumerical methods. With the Knothe-Rosenblatt rearrangement, our newly proposed\nflow-based generative model, called KRnet, provides a family of probability\ndensity functions to serve as effective solution candidates for the\nFokker-Planck equations, which has a weaker dependence on dimensionality than\ntraditional computational approaches and can efficiently estimate general\nhigh-dimensional density functions. To obtain effective stochastic collocation\npoints for the approximation of the F-P equation, we develop an adaptive\nsampling procedure, where samples are generated iteratively using the\napproximate density function at each iteration. We present a general framework\nof ADDA-KR, validate its accuracy and demonstrate its efficiency with numerical\nexperiments.\n"}
{"abstract": "  We report on the demonstration of metalorganic vapor phase epitaxy-regrown\n(MOVPE) ohmic contacts in an all MOVPE-grown $\\beta$-Ga$_2$O$_3$\nmetal-semiconductor field effect transistor (MESFET). The low-temperature\n(600$^{\\circ}$C) heavy (n$^{+}$) Si-doped regrown layers exhibit extremely high\nconductivity with sheet resistance of 73 $\\Omega$/$\\square$ and record low\nmetal/n$^{+}$-Ga$_2$O$_3$ contact resistance of 80 m$\\Omega$.mm and specific\ncontact resistivity of 8.3$\\times$10$^{-7}$ $\\Omega$.cm$^{2}$ were achieved.\nThe fabricated MESFETs exhibit a maximum drain-to-source current of 130 mA/mm,\na high I$_{ON}$/I$_{OFF}$ of $>$10$^{10}$ with a high power FOM of 25\nMW/cm$^{2}$ were achieved without any field plates. Nanoparticle-assisted Raman\nthermometry, thermal modeling, and infrared thermography were performed to\nassess the device self-heating under the high current and power conditions.\nThis demonstration shows the promise of MOVPE technique for the realization of\nhigh-performance lateral $\\beta$-Ga$_2$O$_3$ devices and also highlights the\nneed for device-level thermal management.\n"}
{"abstract": "  We find a late times approximation for the SYK spectral form factor from a\nlarge $N$ steepest descent version of the path integral over two replica\ncollective fields. Main ingredients are a suitable uv regularization of the two\nreplica kinetic operator, the property of its Fourier transform and some\nspectral analysis of the four point function two replica ladder kernel.\n"}
{"abstract": "  $\\mathrm{Cu_2IrO_3}$ is among the newest layered honeycomb iridates and a\npromising candidate to harbor a Kitaev quantum spin liquid state. Here, we\ninvestigate the pressure and temperature dependence of its structure through a\ncombination of powder x-ray diffraction and x-ray absorption fine structure\nmeasurements, as well as $ab$-$initio$ evolutionary structure search. At\nambient pressure, we revise the previously proposed $C2/c$ solution with a\nrelated but notably more stable $P2_1/c$ structure. Pressures below 8 GPa drive\nthe formation of Ir-Ir dimers at both ambient and low temperatures, similar to\nthe case of $\\mathrm{Li_2IrO_3}$. At higher pressures, the structural evolution\ndramatically depends on temperature. A large discontinuous reduction of the Ir\nhoneycomb interplanar distance is observed around 15 GPa at room temperature,\nlikely driven by a collapse of the O-Cu-O dumbbells. At 15 K, pressures beyond\n20 GPa first lead to an intermediate phase featuring a continuous reduction of\nthe interplanar distance, which then collapses at 30 GPa across yet another\nphase transition. However, the resulting structure around 40 GPa is not the\nsame at room and low temperatures. Remarkably, the reduction in interplanar\ndistance leads to an apparent healing of the stacking faults at room\ntemperature, but not at 15 K. Possible implications on the evolution of\nelectronic structure of $\\mathrm{Cu_2IrO_3}$ with pressure are discussed.\n"}
{"abstract": "  The present paper shows a proposal of the characteristics Cloud Risk\nAssessment Models should have and presents the review of the literature\nconsidering those characteristics in order to identify current gaps. This work\nshows a ranking of Cloud RA models and their degree of compliance with the\ntheoretical reference Cloud Risk Assessment model. The review of literature\nshows that RA approaches leveraging CSA (Cloud Security Alliance) STAR Registry\nthat have into account organizations security requirements present higher\ndegree of compliance, but they still lack risk economic quantification. The\nmyriad of conceptual models, methodologies and frameworks although based on\ncurrent NIST SP 800:30, ISO 27001, ISO 27005, ISO 30001, ENISA standards could\nbe enhanced by the use of techno-economic models like UTEM, created by the\nauthor, in order to conceive more simplified models for effective Risk\nAssessment and Mitigation closer to the theoretical reference model for Cloud\nRisk Assessment, available for all cloud models (IaaS, PaaS, SaaS) and easy to\nuse for all stakeholders.\n"}
{"abstract": "  The tension between current expansion rate H$_0$ using Planck data and direct\nmodel-independent measurements in the local universe has reached a tension\nabove 5$\\sigma$ in the context of the $\\Lambda$CDM model. The growing tension\namong early time and local measurements of H$_0$ has not ameliorated and\nremains a crucial and open question in cosmology. Solutions to understand this\ntension are possible hidden sources of systematic error in the observable\nmeasurements or modifications to the concordance $\\Lambda$CDM model. In this\nwork, we investigate a solution to the H$_0$ tension by modifying $\\Lambda$CDM\nand we add at early times extra relativistic energy density $\\rho_{ex}$. In\nsome context this $\\rho_{ex}$ corresponds to EDE or Bound Dark Energy. We\nimplement our model in CAMB and perform a full COSMO-MC (MCMC) allowing to\nsimultaneously fit the latest data from CMB anisotropies and the value of H$_0\n= 74.03\\pm 1.42,\\mathrm{km\\, s^{-1}Mpc^{-1}}$ from A. Riess [R-19]\n\\ci{Riess:2019cxk}. The inclusion of $\\rho_{ex}$ ameliorates the tension\nbetween early and late time measurements only slightly and we obtain a value\nH$_0 = (68.70\\pm 0.45$\\ksm) still in conflict with local measurements [R-19].\nWe follow up our analysis by proposing two forecasting standard deviation\n$\\sigma_H=1$ and $\\sigma_H=0.5$ (in units of \\ksm) for local distance\nmeasurements H$_0 = (74.03\\pm \\sigma_H)$ \\ksm. From MCMC we obtain a value of\nH$_0 =(72.83\\pm 0.47)$ \\ksm\\, at 68\\% confidence level for $\\sigma_H =0.5$,\nfully consistent with [R-19], while the price to pay is a percentage increase\nof $0.12\\%$ in CMB $\\chi_{cmb}^2$. Finally, the extra energy density\n$\\rho_{ex}$ leaves distinctive imprints in the matter power spectrum at scales\n$k \\sim k_c$ with $k_c=a_cH(a_c)$ and in the CMB power spectrum, allowing for\nindependent verification of our analysis.\n"}
{"abstract": "  We present optical follow-up observations with the DDOTI telescope of\ngravitational-wave events detected during the Advanced LIGO and Advanced Virgo\nO3 observing run. DDOTI is capable of responding to an alert in a few minutes,\nhas an instantaneous field of about 69 deg$^{2}$, and obtains $10\\sigma$ upper\nlimits of $w_{\\rm lim}=18.5$ to 20.5 AB mag in 1000~s of exposure, depending on\nthe conditions. We observed 54\\% (26 out of 48) of the unretracted\ngravitational-wave alerts and did not find any electromagnetic counterparts. We\ncompare our upper limits to various possible counterparts: the kilonova\nAT~2017gfo, models of radioactive- and magnetar-powered kilonovae, short\ngamma-ray burst afterglows, and AGN flares. Although the large positional\nuncertainties of GW sources do not allow us to place strong constraints during\nO3, DDOTI observations of well-localized GW events in O4 and beyond could\nmeaningfully constrain models of compact binary mergers. We show that DDOTI is\nable to detect kilonovae similar to AT~2017gfo up to about 200~Mpc and\nmagnetar-powered kilonovae up to 1~Gpc. We calculate that nearby ($\\lesssim$200\nMpc) afterglows have a high chance ($\\approx$70\\%) to be detected by rapid\n($\\lesssim$3 hours) DDOTI observations if observed on-axis, whereas off-axis\nafterglows are unlikely to be seen. Finally, we suggest that long-term\nmonitoring of massive BBH events with DDOTI could confirm or rule out late AGN\nflares associated with these events.\n"}
{"abstract": "  This paper proposes a equivariant filtering (EqF) framework for the\ninertial-integrated state estimation problem. As the kinematic system of the\ninertial-integrated navigation can be naturally modeling on the matrix Lie\ngroup $SE_2(3)$, the symmetry of the Lie group can be exploited to design a\nequivariant filter which extends the invariant extended Kalman filtering on the\ngroup affine system. Furthermore, details of the analytic state transition\nmatrices for left invariant error and right invariant error are given.\n"}
{"abstract": "  Let $(\\pi, \\mathcal{H})$ be a strongly continuous unitary representation of a\n1-connected Lie group $G$ such that the Lie algebra $\\mathfrak{g}$ of $G$ is\ngenerated by the positive cone $C_\\pi := \\{x \\in \\mathfrak{g} : -i\\partial\n\\pi(x) \\geq 0\\}$ and an element $h$ for which the adjoint representation of $h$\ninduces a 3-grading of $\\mathfrak{g}$. Moreover, suppose that $(\\pi,\n\\mathcal{H})$ extends to an antiunitary representation of the extended Lie\ngroup $G_\\tau := G \\rtimes \\{\\mathbf{1}, \\tau_G\\}$, where $\\tau_G$ is an\ninvolutive automorphism of $G$ with $\\mathbf{L}(\\tau_G) = e^{i\\pi\\mathrm{ad}\nh}$.\n  In a recent work by Neeb and \\'Olafsson, a method for constructing nets of\nstandard subspaces of $\\mathcal{H}$ indexed by open regions of $G$ has been\nintroduced and applied in the case where $G$ is semisimple. In this paper, we\nextend this construction to general Lie groups $G$, provided the above\nassumptions are satisfied and the center of the ideal $\\mathfrak{g}_C = C_\\pi -\nC_\\pi$ of $\\mathfrak{g}$ is one-dimensional. The case where the center of\n$\\mathfrak{g}_C$ has more than one dimension will be discussed in a separate\npaper.\n"}
{"abstract": "  In this book there will be found an introduction to transcendental number\ntheory, starting at the beginning and ending at the frontiers. The emphasis is\non the conceptual aspects of the subject, thus the effective theory has been\nmore or less completely ignored, as has been the theory of $E$-functions and\n$G$-functions. Still, a fair amount of ground is covered and while I take\ncertain results without proof, this is done primarily so as not to get bogged\ndown in technicalities, otherwise the exposition is detailed and little is left\nto the reader.\n"}
{"abstract": "  Babies born with low and very low birthweights -- i.e., birthweights below\n2,500 and 1,500 grams, respectively -- have an increased risk of complications\ncompared to other babies, and the proportion of babies with a low birthweight\nis a common metric used when evaluating public health in a population. While\nmany factors increase the risk of a baby having a low birthweight, many can be\nlinked to the mother's socioeconomic status, which in turn contributes to large\nracial disparities in the incidence of low weight births. Here, we employ\nBayesian statistical models to analyze the proportion of babies with low\nbirthweight in Pennsylvania counties by race/ethnicity. Due to the small number\nof births -- and low weight births -- in many Pennsylvania counties when\nstratified by race/ethnicity, our methods must walk a fine line. On one hand,\nleveraging spatial structure can help improve the precision of our estimates.\nOn the other hand, we must be cautious to avoid letting the model overwhelm the\ninformation in the data and produce spurious conclusions. As such, we first\ndevelop a framework by which we can measure (and control) the informativeness\nof our spatial model. After demonstrating the properties of our framework via\nsimulation, we analyze the low birthweight data from Pennsylvania and examine\nthe extent to which the commonly used conditional autoregressive model can lead\nto oversmoothing. We then reanalyze the data using our proposed framework and\nhighlight its ability to detect (or not detect) evidence of racial disparities\nin the incidence of low birthweight.\n"}
{"abstract": "  We initiate the study of Regge theory in a bottom-up holographic model for\nQCD in the Veneziano limit, where the backreaction of the quarks to the gluon\ndynamics is included. We determine the parameters of the model by carrying out\na precise fit to the meson spectrum in QCD. The spectrum for spin-one and\npseudoscalar mesons is well reproduced. We then generalise the model to incluce\nhigher spin fields in the bulk trajectories dual to the Pomeron and meson Regge\ntrajectories at the boundary. With this setting, we fit the masses of the\nmesons with spins $J=2$, $3$, and $4$,as well as the experimental data of the\ntotal cross-sections $\\sigma(\\gamma \\gamma \\to X)$, $\\sigma(\\gamma p \\to X)$\nand $\\sigma(p p \\to X)$. For the cross sections we obtain a\n$\\chi^2_{\\mathrm{d.o.f.}}$ of $0.74$ for a total of 199 experimental points.\n"}
{"abstract": "  Absorption-selected galaxies offer an effective way to study low-mass\ngalaxies at high redshift. However, the physical properties of the underlying\ngalaxy population remains uncertain. In particular, the multiphase\ncircum-galactic medium is thought to hold key information on gas flows into and\nout of galaxies that are vital for galaxy evolution models. Here we present\nALMA observations of CO molecular gas in host galaxies of H_2-bearing\nabsorbers. In our sample of six absorbers we detect molecular gas-rich galaxies\nin five absorber fields although we did not target high-metallicity (>50 per\ncent solar) systems for which previous studies reported the highest detection\nrate. Surprisingly, we find that the majority of the absorbers are associated\nwith multiple galaxies rather than single haloes. Together with the large\nimpact parameters these results suggest that the H_2-bearing gas seen in\nabsorption is not part of an extended disk, but resides in dense gas pockets in\nthe circum-galactic and intra-group medium.\n"}
{"abstract": "  We report on the development of multi-beam RF linear ion accelerators that\nare formed from stacks of low cost wafers and describe the status of beam power\nscale-up using an array of 120 beams. The total argon ion current extracted\nfrom the 120-beamlet extraction column was 0.5 mA. The measured energy gain in\neach RF gap reached as high as 7.25 keV. We present a path of using this\ntechnology to achieve ion currents >1 mA and ion energies >100 keV for\napplications in materials processing.\n"}
{"abstract": "  The notion of a tensor captures three great ideas: equivariance,\nmultilinearity, separability. But trying to be three things at once makes the\nnotion difficult to understand. We will explain tensors in an accessible and\nelementary way through the lens of linear algebra and numerical linear algebra,\nelucidated with examples from computational and applied mathematics.\n"}
{"abstract": "  Millions of cameras at edge are being deployed to power a variety of\ndifferent deep learning applications. However, the frames captured by these\ncameras are not always pristine - they can be distorted due to lighting issues,\nsensor noise, compression etc. Such distortions not only deteriorate visual\nquality, they impact the accuracy of deep learning applications that process\nsuch video streams. In this work, we introduce AQuA, to protect application\naccuracy against such distorted frames by scoring the level of distortion in\nthe frames. It takes into account the analytical quality of frames, not the\nvisual quality, by learning a novel metric, classifier opinion score, and uses\na lightweight, CNN-based, object-independent feature extractor. AQuA accurately\nscores distortion levels of frames and generalizes to multiple different deep\nlearning applications. When used for filtering poor quality frames at edge, it\nreduces high-confidence errors for analytics applications by 17%. Through\nfiltering, and due to its low overhead (14ms), AQuA can also reduce computation\ntime and average bandwidth usage by 25%.\n"}
{"abstract": "  A theory is NIP (resp. stable) if and only if every formula with parameters\nin two single variables is NIP (resp. does not have the order property).\n"}
{"abstract": "  Increasing effort is being directed to understanding the personality profiles\nof highly engaged information systems (IS) developers and the impact of such\nprofiles on development outcomes. However, there has been a lesser degree of\nattention paid to studying attitudes at a fine-grained level, and relating such\nattitudes to developers' in-process activities, in spite of the fact that\nsocial motivation theory notes the importance of such a relationship in general\ngroup work. We have therefore applied linguistic analysis, text mining and\nvisualization, and statistical analysis techniques to artefacts developed by\n474 developers to study these issues. Our results indicate that our sample of\nIS developers conveyed a range of attitudes while working to deliver systems\nfeatures, and those practitioners who communicated the most were also the most\nengaged. Additionally, of eight linguistic dimensions considered, expressions\nregarding work and achievement, as well as insightful attitudes, were most\nclosely related to developers' engagement. Accordingly, team diversity and the\nprovision of active support for outcome-driven developers may contribute\npositively to maintaining team balance and performance.\n"}
{"abstract": "  Text summarization is the process of extracting the most important\ninformation from the text and presenting it concisely in fewer sentences. Call\ntranscript is a text that involves textual description of a phone conversation\nbetween a customer (caller) and agent(s) (customer representatives). This paper\npresents an indigenously developed method that combines topic modeling and\nsentence selection with punctuation restoration in condensing ill-punctuated or\nun-punctuated call transcripts to produce summaries that are more readable.\nExtensive testing, evaluation and comparisons have demonstrated the efficacy of\nthis summarizer for call transcript summarization.\n"}
{"abstract": "  Axions were first introduced in connection with chiral symmetry but are now\nbeing looked for mainly as dark matter. In this paper we introduce a nonabelian\nanalogue of axions which can also be potential candidates for dark matter.\nTheir nonabelian symmetries, which are generalizations of the Peccei-Quinn\nsymmetry, are interesting in their own right. Detailed analysis, using fermion\nmeasure and zeta function approaches shows that these symmetries are not\nanomalous.\n"}
{"abstract": "  Concept-based explanations have emerged as a popular way of extracting\nhuman-interpretable representations from deep discriminative models. At the\nsame time, the disentanglement learning literature has focused on extracting\nsimilar representations in an unsupervised or weakly-supervised way, using deep\ngenerative models. Despite the overlapping goals and potential synergies, to\nour knowledge, there has not yet been a systematic comparison of the\nlimitations and trade-offs between concept-based explanations and\ndisentanglement approaches. In this paper, we give an overview of these fields,\ncomparing and contrasting their properties and behaviours on a diverse set of\ntasks, and highlighting their potential strengths and limitations. In\nparticular, we demonstrate that state-of-the-art approaches from both classes\ncan be data inefficient, sensitive to the specific nature of the\nclassification/regression task, or sensitive to the employed concept\nrepresentation.\n"}
{"abstract": "  Although Generative Adversarial Networks (GANs) are successfully applied to\ndiverse fields, training GANs on synthetic aperture radar (SAR) data is a\nchallenging task mostly due to speckle noise. On the one hands, in a learning\nperspective of human's perception, it is natural to learn a task by using\nvarious information from multiple sources. However, in the previous GAN works\non SAR target image generation, the information on target classes has only been\nused. Due to the backscattering characteristics of SAR image signals, the\nshapes and structures of SAR target images are strongly dependent on their pose\nangles. Nevertheless, the pose angle information has not been incorporated into\nsuch generative models for SAR target images. In this paper, we firstly propose\na novel GAN-based multi-task learning (MTL) method for SAR target image\ngeneration, called PeaceGAN that uses both pose angle and target class\ninformation, which makes it possible to produce SAR target images of desired\ntarget classes at intended pose angles. For this, the PeaceGAN has two\nadditional structures, a pose estimator and an auxiliary classifier, at the\nside of its discriminator to combine the pose and class information more\nefficiently. In addition, the PeaceGAN is jointly learned in an end-to-end\nmanner as MTL with both pose angle and target class information, thus enhancing\nthe diversity and quality of generated SAR target images The extensive\nexperiments show that taking an advantage of both pose angle and target class\nlearning by the proposed pose estimator and auxiliary classifier can help the\nPeaceGAN's generator effectively learn the distributions of SAR target images\nin the MTL framework, so that it can better generate the SAR target images more\nflexibly and faithfully at intended pose angles for desired target classes\ncompared to the recent state-of-the-art methods.\n"}
{"abstract": "  Broad and unexplored kinematic regions can be accessed at the LHC with\nfixed-target $pp$, $pA$ and $PbA$ collisions at\n$\\sqrt{s_{\\rm{NN}}}=72-115~\\rm{GeV}$. The LHCb detector is a fully-instrumented\nforward spectrometer able to run in fixed-target mode, and currently hosts a\ntarget gas cell to take data in the upcoming Run 3. The LHCspin project aims at\nextending this physics program to Run 4 and to bring polarised physics at the\nLHC. An overview of the physics potential and a description of the LHCspin\nexperimental setup are presented.\n"}
{"abstract": "  Motivation: A Chemical Reaction Network (CRN) is a set of chemical reactions,\nwhich can be very complex and difficult to analyze. Indeed, dynamical\nproperties of CRNs can be described by a set of non-linear differential\nequations that rarely can be solved in closed-form, but that can instead be\nused to reason on the system dynamics. In this context, one of the possible\napproaches is to perform numerical simulations, which may require a high\ncomputational effort. In particular, in order to investigate some dynamical\nproperties, such as robustness or global sensitivity, many simulations have to\nbe performed by varying the initial concentration of chemical species. Results:\nIn order to reduce the computational effort required when many simulations are\nneeded to assess a property, we exploit a new notion of monotonicity of the\noutput of the system (the concentration of a target chemical species at the\nsteady-state) with respect to the input (the initial concentration of another\nchemical species). To assess such monotonicity behavior, we propose a new\ngraphical approach that allows us to state sufficient conditions for ensuring\nthat the monotonicity property holds. Our sufficient conditions allow us to\nefficiently verify the monotonicity property by exploring a graph constructed\non the basis of the reactions involved in the network. Once established, our\nmonotonicity property allows us to drastically reduce the number of simulations\nrequired to assess some dynamical properties of the CRN.\n"}
{"abstract": "  We use machine learning to classify rational two-dimensional conformal field\ntheories. We first use the energy spectra of these minimal models to train a\nsupervised learning algorithm. We find that the machine is able to correctly\npredict the nature and the value of critical points of several strongly\ncorrelated spin models using only their energy spectra. This is in contrast to\nprevious works that use machine learning to classify different phases of\nmatter, but do not reveal the nature of the critical point between phases.\nGiven that the ground-state entanglement Hamiltonian of certain topological\nphases of matter is also described by conformal field theories, we use\nsupervised learning on R\\'{e}yni entropies and find that the machine is able to\nidentify which conformal field theory describes the entanglement Hamiltonian\nwith only the lowest few R\\'{e}yni entropies to a high degree of accuracy.\nFinally, using autoencoders, an unsupervised learning algorithm, we find a\nhidden variable that has a direct correlation with the central charge and\ndiscuss prospects for using machine learning to investigate other conformal\nfield theories, including higher-dimensional ones. Our results highlight that\nmachine learning can be used to find and characterize critical points and also\nhint at the intriguing possibility to use machine learning to learn about more\ncomplex conformal field theories.\n"}
{"abstract": "  Operational urban transport models require to gather heterogeneous sets of\ndata and often integrate different sub-models. Their systematic validation and\nreproducible application therefore remains problematic. We propose in this\ncontribution to build transport models from the bottom-up using scientific\nworkflow systems with open-source components and data. These open models are\naimed in particular at estimating congestion of public transport in all UK\nurban areas. This allows us building health indicators related to public\ntransport density in the context of the COVID-19 crisis, and testing related\npolicies.\n"}
{"abstract": "  Networks analysis has been commonly used to study the interactions between\nunits of complex systems. One problem of particular interest is learning the\nnetwork's underlying connection pattern given a single and noisy instantiation.\nWhile many methods have been proposed to address this problem in recent years,\nthey usually assume that the true model belongs to a known class, which is not\nverifiable in most real-world applications. Consequently, network modeling\nbased on these methods either suffers from model misspecification or relies on\nadditional model selection procedures that are not well understood in theory\nand can potentially be unstable in practice. To address this difficulty, we\npropose a mixing strategy that leverages available arbitrary models to improve\ntheir individual performances. The proposed method is computationally efficient\nand almost tuning-free; thus, it can be used as an off-the-shelf method for\nnetwork modeling. We show that the proposed method performs equally well as the\noracle estimate when the true model is included as individual candidates. More\nimportantly, the method remains robust and outperforms all current estimates\neven when the models are misspecified. Extensive simulation examples are used\nto verify the advantage of the proposed mixing method. Evaluation of link\nprediction performance on 385 real-world networks from six domains also\ndemonstrates the universal competitiveness of the mixing method across multiple\ndomains.\n"}
{"abstract": "  Vertical federated learning (VFL) attracts increasing attention due to the\nemerging demands of multi-party collaborative modeling and concerns of privacy\nleakage. In the real VFL applications, usually only one or partial parties hold\nlabels, which makes it challenging for all parties to collaboratively learn the\nmodel without privacy leakage. Meanwhile, most existing VFL algorithms are\ntrapped in the synchronous computations, which leads to inefficiency in their\nreal-world applications. To address these challenging problems, we propose a\nnovel {\\bf VF}L framework integrated with new {\\bf b}ackward updating mechanism\nand {\\bf b}ilevel asynchronous parallel architecture (VF{${\\textbf{B}}^2$}),\nunder which three new algorithms, including VF{${\\textbf{B}}^2$}-SGD, -SVRG,\nand -SAGA, are proposed. We derive the theoretical results of the convergence\nrates of these three algorithms under both strongly convex and nonconvex\nconditions. We also prove the security of VF{${\\textbf{B}}^2$} under\nsemi-honest threat models. Extensive experiments on benchmark datasets\ndemonstrate that our algorithms are efficient, scalable and lossless.\n"}
{"abstract": "  In this paper we present a large sample of jellyfish galaxies in low redshift\nclusters (z<0.05), identified through 120-168 MHz radio continuum from the\nLOFAR Two-metre Sky Survey (LoTSS). From a parent sample of 29 X-ray-detected\nSDSS galaxy clusters and their spectroscopic members, we visually identify 95\nstar-forming, LoTSS jellyfish galaxies with 144 MHz radio tails. Star formation\nrates (SFRs) and stellar masses are obtained for all galaxies from SED fits.\nFor each jellyfish galaxy we determine the tail orientation with respect to the\ncluster centre and quantify the prominence of the radio tails with the 144 MHz\nshape asymmetry. After carefully accounting for redshift-dependent selection\neffects, we find that the frequency of jellyfish galaxies is relatively\nconstant from cluster to cluster. LoTSS jellyfish galaxies are preferentially\nfound at small clustercentric radius and large velocity offsets within their\nhost clusters and have radio tails that are oriented away from the cluster\ncentre. These galaxies also show enhanced star formation, relative to both\n'normal' cluster galaxies and isolated field galaxies, but generally fall\nwithin the scatter of the L144MHz - SFR relation. The properties of the LoTSS\njellyfish galaxies identified in this work are fully consistent with\nexpectations from ram pressure stripping. This large sample of jellyfish\ngalaxies will be valuable for further constraining ram pressure stripping and\nstar formation quenching in nearby galaxy clusters. We show that LOFAR is a\npowerful instrument for identifying ram pressure stripped galaxies across\nextremely wide fields. Moving forward we will push the search for jellyfish\ngalaxies beyond this initial cluster sample, including a comprehensive survey\nof the galaxy group regime.\n"}
{"abstract": "  Many techniques were proposed for detecting software misconfigurations in\ncloud systems and for diagnosing unintended behavior caused by such\nmisconfigurations. Detection and diagnosis are steps in the right direction:\nmisconfigurations cause many costly failures and severe performance issues.\nBut, we argue that continued focus on detection and diagnosis is symptomatic of\na more serious problem: configuration design and implementation are not yet\nfirst-class software engineering endeavors in cloud systems. Little is known\nabout how and why developers evolve configuration design and implementation,\nand the challenges that they face in doing so.\n  This paper presents a source-code level study of the evolution of\nconfiguration design and implementation in cloud systems. Our goal is to\nunderstand the rationale and developer practices for revising initial\nconfiguration design/implementation decisions, especially in response to\nconsequences of misconfigurations. To this end, we studied 1178\nconfiguration-related commits from a 2.5 year version-control history of four\nlarge-scale, actively-maintained open-source cloud systems (HDFS, HBase, Spark,\nand Cassandra). We derive new insights into the software configuration\nengineering process. Our results motivate new techniques for proactively\nreducing misconfigurations by improving the configuration design and\nimplementation process in cloud systems. We highlight a number of future\nresearch directions.\n"}
{"abstract": "  Region modification-based data augmentation techniques have shown to improve\nperformance for high level vision tasks (object detection, semantic\nsegmentation, image classification, etc.) by encouraging underlying algorithms\nto focus on multiple discriminative features. However, as these techniques\ndestroy spatial relationship with neighboring regions, performance can be\ndeteriorated when using them to train algorithms designed for low level vision\ntasks (low light image enhancement, image dehazing, deblurring, etc.) where\ntextural consistency between recovered and its neighboring regions is important\nto ensure effective performance. In this paper, we examine the efficacy of a\nsimple copy-blend data augmentation technique that copies patches from noisy\nimages and blends onto a clean image and vice versa to ensure that an\nunderlying algorithm localizes and recovers affected regions resulting in\nincreased perceptual quality of a recovered image. To assess performance\nimprovement, we perform extensive experiments alongside different region\nmodification-based augmentation techniques and report observations such as\nimproved performance, reduced requirement for training dataset, and early\nconvergence across tasks such as low light image enhancement, image dehazing\nand image deblurring without any modification to baseline algorithm.\n"}
{"abstract": "  Power grids, as well as neuronal networks with synaptic plasticity, describe\nreal-world systems of tremendous importance for our daily life. The\ninvestigation of these seemingly unrelated types of dynamical networks has\nattracted increasing attention over the last decade. In this work, we exploit\nthe recently established relation between these two types of networks to gain\ninsights into the dynamical properties of multifrequency clusters in power grid\nnetworks. For this, we consider the model of Kuramoto-Sakaguchi phase\noscillators with inertia and describe the emergence of multicluster states.\nBuilding on this, we provide a new perspective on solitary states in power grid\nnetworks by introducing the concept of pseudo coupling weights.\n"}
{"abstract": "  Heisenberg's uncertainty principle tells us that it is impossible to\ndetermine simultaneously the position of a photon crossing a telescope's\naperture and its momentum. Super-resolution imaging techniques rely on\nmodification of the observed sample, or on entangling photons. In astronomy we\nhave no access to the object, but resolution may be improved by optical\namplification. Unfortunately, spontaneous emission contributes noise and\nnegates the possible gain from stimulated emissions. We show that it is\npossible to increase the weight of the stimulated photons by considering photon\nstatistics, and observe an improvement in resolution. Most importantly, we\ndemonstrate a method which can apply for all imaging purposes.\n"}
{"abstract": "  Object insertion is a classic contact-rich manipulation task. The task\nremains challenging, especially when considering general objects of unknown\ngeometry, which significantly limits the ability to understand the contact\nconfiguration between the object and the environment. We study the problem of\naligning the object and environment with a tactile-based feedback insertion\npolicy. The insertion process is modeled as an episodic policy that iterates\nbetween insertion attempts followed by pose corrections. We explore different\nmechanisms to learn such a policy based on Reinforcement Learning. The key\ncontribution of this paper is to demonstrate that it is possible to learn a\ntactile insertion policy that generalizes across different object geometries,\nand an ablation study of the key design choices for the learning agent: 1) the\ntype of learning scheme: supervised vs. reinforcement learning; 2) the type of\nlearning schedule: unguided vs. curriculum learning; 3) the type of sensing\nmodality: force/torque (F/T) vs. tactile; and 4) the type of tactile\nrepresentation: tactile RGB vs. tactile flow. We show that the optimal\nconfiguration of the learning agent (RL + curriculum + tactile flow) exposed to\n4 training objects yields an insertion policy that inserts 4 novel objects with\nover 85.0% success rate and within 3~4 attempts. Comparisons between F/T and\ntactile sensing, shows that while an F/T-based policy learns more efficiently,\na tactile-based policy provides better generalization.\n"}
{"abstract": "  The diffraction of light imposes a fundamental limit on the resolution of\nlight microscopes. This limit can be circumvented by creating and exploiting\nindependent behaviors of the sample at length scales below the diffraction\nlimit. In super-resolution single molecule localization microscopy (SMLM), the\nindependence arises from individual fluorescent labels stochastically switching\nbetween dark and fluorescent states, which in turn allows the pinpointing of\nfluorophores post experimentally using a sequence of acquired sparse image\nframes. Finally, the resulting list of fluorophore coordinates is utilized to\nproduce high resolution images or to gain quantitative insight into the\nunderlying biological structures. Therefore, image processing and\npost-processing are essential stages of SMLM. Here, we review the latest\nprogress on SMLM data processing and post-processing.\n"}
{"abstract": "  The quantum spin liquid (QSL) is an exotic phase of magnetic materials where\nthe spins continue to fluctuate without any symmetry breaking down to zero\ntemperature. Among the handful reports of QSL with spin $S\\ge$1, examples with\nmagnetic ions on a three-dimensional magnetic lattice are extremely rare since\nboth larger spin and higher dimension tend to suppress quantum fluctuations. In\nthis work, we offer a new strategy to achieve 3-D QSL with high spin by\nutilizing two types of transition metal ions, both are magnetically active but\nlocated at crystallographically inequivalent positions. We design a 3-D\nmagnetic system Ba$_3$NiIr$_2$O$_9$ consisting of interconnected corner shared\nNiO$_6$ octahedra and face shared Ir$_2$O$_9$ dimer, both having triangular\narrangements in \\textit{a-b} plane. X-ray absorption spectroscopy measurements\nconfirm the presence of Ni$^{2+}$ ($S$=1). Our detailed thermodynamic and\nmagnetic measurements reveal that this compound is a realization of gapless QSL\nstate down to at least 100 mK. Ab-initio calculations find a strong magnetic\nexchange between Ir and Ni sublattices and in-plane antiferromagnetic coupling\nbetween the dimers, resulting in dynamically fluctuating magnetic moments on\nboth the Ir and Ni sublattice.\n"}
{"abstract": "  Energetic electrons inside Earth's outer Van Allen belt pose a major\nradiation threat to space-borne electronics that often play vital roles in our\nmodern society. Ultra-relativistic electrons with energies greater than or\nequal to two Megaelectron-volt (MeV) are of particular interest due to their\nhigh penetrating ability, and thus forecasting these >=2 MeV electron levels\nhas significant meaning to all space sectors. Here we update the latest\ndevelopment of the predictive model for MeV electrons inside the Earth's outer\nradiation belt. The new version, called PreMevE-2E, focuses on forecasting\nultra-relativistic electron flux distributions across the outer radiation belt,\nwith no need of in-situ measurements except for at the geosynchronous (GEO)\norbit. Model inputs include precipitating electrons observed in\nlow-Earth-orbits by NOAA satellites, upstream solar wind conditions (speeds and\ndensities) from solar wind monitors, as well as ultra-relativistic electrons\nmeasured by one Los Alamos GEO satellite. We evaluated a total of 32 supervised\nmachine learning models that fall into four different classes of linear and\nneural network architectures, and also successfully tested ensemble forecasting\nby using groups of top-performing models. All models are individually trained,\nvalidated, and tested by in-situ electron data from NASA's Van Allen Probes\nmission. It is shown that the final ensemble model generally outperforms\nindividual models overs L-shells, and this PreMevE-2E model provides reliable\nand high-fidelity 25-hr (~1-day) and 50-hr (~2-day) forecasts with high mean\nperformance efficiency values. Our results also suggest this new model is\ndominated by non-linear components at low L-shells (< ~4) for\nultra-relativistic electrons, which is different from the dominance of linear\ncomponents at all L-shells for 1 MeV electrons as previously discovered.\n"}
{"abstract": "  This article proposes Rydberg multi-dimensional discrete-time quantum walk\n(DTQW) implementation that could ideally simulate different classes of\ntopological insulators. Using an exchange interaction between Rydberg excited\natoms in an atomic-array with dual lattice-constants, the new setup operates\nboth coined and coin-less models of DTQW. Here, complicated coupling\ntesselations are performed by global laser excitation. The long-range\ninteraction provides a new feature of designing different topologically ordered\nperiodic boundary conditions. Limiting the Rydberg population to 1.5\nexcitations, coherent QW over thousand lattice sites and steps are achievable\nwith the current technology. These features would improve the performance of\nthis quantum machine in running the quantum search algorithm over topologically\nordered databases as well as diversifying the range of topological insulators\nthat could be simulated.\n"}
{"abstract": "  We study the ring of polyfunctions over a commutative ring $R$ with unit\nelement, i.e., the ring of functions $f:R\\to R$ which admit a polynomial\nrepresentative $p\\in R[x]$ in the sense that $f(x)= p(x)$ for all $x\\in R$.\nThis allows to define a ring invariant $s$ which associates to a commutative\nring $R$ with unit element a value in $\\mathbb N\\cup\\{\\infty\\}$. The function\n$s$ generalizes the number theoretic Smarandache function. For the ring\n$R=\\mathbb Z/n\\mathbb Z$ we provide a unique representation of polynomials\nwhich vanish as a function. This yields a new formula for the number $\\Psi(n)$\nof polyfunctions over $\\mathbb Z/n\\mathbb Z$. We also investigate algebraic\nproperties of the ring of polyfunctions over $\\mathbb Z/n\\mathbb Z$. In\nparticular, we identify the additive subgroup of the ring and the ring\nstructure itself. Moreover we derive a new formula for the size of the ring of\npolyfunctions in several variables over $\\mathbb Z/n\\mathbb Z$.\n"}
{"abstract": "  Schubert polynomials form a basis of all polynomials and appear in the study\nof cohomology rings of flag manifolds. The vanishing problem for Schubert\npolynomials asks if a coefficient of a Schubert polynomial is zero. We give a\ntableau criterion to solve this problem, from which we deduce the first\npolynomial time algorithm. These results are obtained from new\ncharacterizations of the Schubitope, a generalization of the permutahedron\ndefined for any subset of the n x n grid. In contrast, we show that computing\nthese coefficients explicitly is #P-complete.\n"}
{"abstract": "  Topologically stable field configurations appear in many fields of physics,\nfrom elementary particles to condensed matter and cosmology. During the last\ndecade, chiral liquid crystals and chiral magnets took on the role of model\nobjects for experimental investigation of topological solitons and\nunderstanding of their nonsingular field configurations. Here we introduce a\nparadigm of facile skyrmionic networks with mutually-orthogonal orientations of\nconstituent isolated skyrmions. Such networks are envisioned as a novel concept\nof spintronic devices and are presumably responsible for precursor phenomena\nnear the ordering temperatures of bulk cubic helimagnets. In particular, we\ndemonstrate an interconversion between mutually orthogonal skyrmions:\nhorizontal skyrmions swirl into an intermediate spring-like states and\nsubsequently squeeze into vertical skyrmions with both polarities. Skyrmion\ntubes are also considered as building blocks for particle-like states with more\ninvolved internal structure. A family of target-skyrmions, which includes an\noverlooked so far type with a multiple topological charge, is formed owing to\nthe tendency to minimize the interaction energy between vertical and horizontal\nskyrmions. The conical phase serves as a suitable background for considered\nskyrmion evolution. It substantializes the attracting skyrmion-skyrmion\ninteraction in the skyrmionic networks, shapes their internal structure and\nguides the nucleation processes. Alternatively, intricate textural changes of\nisolated skyrmions result not only in the structural deformations of a host\nconical state, but may lead to the formation of an exotic skyrmion order with\npairs of merons being the core of the game. Generically, the fundamental\ninsights provided by this work emphasize a three-dimensional character of\nskyrmion metamorphoses and can also be extended to three-dimensional solitons,\nsuch as hopfions.\n"}
{"abstract": "  With the aim to parallelize and monitor biological or biochemical phenomena,\ntrapping and immobilization of objects such as particles, droplets or cells in\nmicrofluidic devices has been an intense area of research and engineering so\nfar. Either being passive or active, these microfluidic devices are usually\ncomposed of arrays of elementary traps with various levels of sophistication.\nFor a given array, it is important to have an efficient and fast immobilization\nof the highest number of objects, while optimizing the spatial homogeneity of\nthe trapping over the whole chip. For passive devices, this has been achieved\nwith two-layers structures, making the fabrication process more complex. In\nthis work, we designed small microfluidic traps by single-layer direct laser\nwriting into a photoresist, and we show that even in this simplest case, the\norientation of the main flow of particles with respect to the traps have a\ndrastic effect on the trapping efficiency and homogeneity. To better understand\nthis phenomenon, we have considered two different flow geometries: parallel and\noblique with respect to the traps array, and compared quantitatively the\nimmobilization of particles with various sizes and densities. Using image\nanalysis, we show that diagonal flows gives a spatial distribution of the trap\nloading that is more homogeneous over the whole chip as compared to the\nstraight ones, and by performing FEM and trapping simulation, we propose a\nqualitative explanation of this phenomenon.\n"}
{"abstract": "  Our gargantuan ab initio all-electron fully relativistic Dirac-Fock (DF),\nnonrelativistic (NR) Hartree-Fock(HF) and Dirac-Fock-Breit-Gaunt(DFBG)\nmolecular SCF calculations for the superheavy octahedral Oganesson\nhexatenniside OgTs$_6$ predict atomization energy (Ae) of 9.47, -5.54and 9.37\neV, at the optimized Os-Ts bond distances of 3.35, 3.34 and 3.36 angstroms,\nrespectively. There are dramatic effects of relativity for the atomization\nenergy of OgTs$_6$ (with seven superheavy elements and 820 electrons) of ~ 15.0\neV each at both the DF and DFBG levels of theory, respectively. Our calculated\nenergy of reaction for the titled superheavy reaction Og + 3Ts$_2$ -> OgTs$_6$\nat the DF, NR and DFBG levels of theory is 6.33, 8.81, and 6.26 eV,\nrespectively. Mulliken analysis as implemented in the DIRAC code for our DF and\nNR calculations (using the dyall.ev4z basis) yields the charges Og(+0.60) and\nOg(+0.96), respectively on the central Og atom indicating that our relativistic\nDF calculations predict octahedral OgTs$_6$ to be less ionic. However, due\ncaution must be used to interpret the results of Mulliken's population\nanalysis, which is highly basis set dependent.\n"}
{"abstract": "  Accurate photo-z measurements are important to construct a large-scale\nstructure map of X-ray Universe in the ongoing SRG/eROSITA All-Sky Survey. We\npresent machine learning Random Forest-based models for probabilistic photo-z\npredictions based on information from 4 large photometric surveys (SDSS,\nPan-STARRS, DESI Legacy Imaging Survey, and WISE). Our models are trained on\nthe large sample of $\\approx$580000 quasars and galaxies selected from the SDSS\nDR14 spectral catalog and take into account Galactic extinction and\nuncertainties in photometric measurements for target objects. On the Stripe82X\ntest sample we obtained photo-z accuracy for X-ray sources: $NMAD=0.034$\n(normalized median absolute deviation) and $n_{>0.15}=0.088$ (catastrophic\noutliers fraction), which is almost $\\sim2$ times better than best photo-z\nresults available in the literature.\n"}
{"abstract": "  Temporal semantic scene understanding is critical for self-driving cars or\nrobots operating in dynamic environments. In this paper, we propose 4D panoptic\nLiDAR segmentation to assign a semantic class and a temporally-consistent\ninstance ID to a sequence of 3D points. To this end, we present an approach and\na point-centric evaluation metric. Our approach determines a semantic class for\nevery point while modeling object instances as probability distributions in the\n4D spatio-temporal domain. We process multiple point clouds in parallel and\nresolve point-to-instance associations, effectively alleviating the need for\nexplicit temporal data association. Inspired by recent advances in benchmarking\nof multi-object tracking, we propose to adopt a new evaluation metric that\nseparates the semantic and point-to-instance association aspects of the task.\nWith this work, we aim at paving the road for future developments of temporal\nLiDAR panoptic perception.\n"}
{"abstract": "  Fingerspelling, in which words are signed letter by letter, is an important\ncomponent of American Sign Language. Most previous work on automatic\nfingerspelling recognition has assumed that the boundaries of fingerspelling\nregions in signing videos are known beforehand. In this paper, we consider the\ntask of fingerspelling detection in raw, untrimmed sign language videos. This\nis an important step towards building real-world fingerspelling recognition\nsystems. We propose a benchmark and a suite of evaluation metrics, some of\nwhich reflect the effect of detection on the downstream fingerspelling\nrecognition task. In addition, we propose a new model that learns to detect\nfingerspelling via multi-task training, incorporating pose estimation and\nfingerspelling recognition (transcription) along with detection, and compare\nthis model to several alternatives. The model outperforms all alternative\napproaches across all metrics, establishing a state of the art on the\nbenchmark.\n"}
{"abstract": "  We propose a dynamic encoder transducer (DET) for on-device speech\nrecognition. One DET model scales to multiple devices with different\ncomputation capacities without retraining or finetuning. To trading off\naccuracy and latency, DET assigns different encoders to decode different parts\nof an utterance. We apply and compare the layer dropout and the collaborative\nlearning for DET training. The layer dropout method that randomly drops out\nencoder layers in the training phase, can do on-demand layer dropout in\ndecoding. Collaborative learning jointly trains multiple encoders with\ndifferent depths in one single model. Experiment results on Librispeech and\nin-house data show that DET provides a flexible accuracy and latency trade-off.\nResults on Librispeech show that the full-size encoder in DET relatively\nreduces the word error rate of the same size baseline by over 8%. The\nlightweight encoder in DET trained with collaborative learning reduces the\nmodel size by 25% but still gets similar WER as the full-size baseline. DET\ngets similar accuracy as a baseline model with better latency on a large\nin-house data set by assigning a lightweight encoder for the beginning part of\none utterance and a full-size encoder for the rest.\n"}
{"abstract": "  Charge-$4e$ superconductivity as a novel phase of matter remains elusive so\nfar. Here we show that charge-$4e$ phase can arise as a vestigial order above\nthe nematic superconducting transition temperature in time-reversal-invariant\nnematic superconductors. On the one hand, the nontrivial topological defect --\nnematic vortex -- is energetically favored over the superconducting phase\nvortex when the nematic stiffness is less than the superfluid stiffness;\nconsequently the charge-$4e$ phase emerges by proliferation of nematic vortices\nupon increasing temperatures. On the other hand, the Ginzburg-Landau theory of\nthe nematic superconductors has two distinct decoupling channels to either\ncharge-$4e$ orders or nematic orders; by analyzing the competition between the\neffective mass of the charge-$4e$ order and the cubic potential of the nematic\norder, we find a sizable regime where the charge-$4e$ order is favored. These\ntwo analysis consistently show that nematic superconductors can provide a\npromising route to realize charge-$4e$ phases, which may apply to candidate\nnematic superconductors such as PbTaSe$_2$ and twisted bilayer graphene.\n"}
{"abstract": "  In search for the possible astrophysical sources behind origination of the\ndiverse gamma-ray bursts, cluster analyses are performed to find homogeneous\ngroups, which discover an intermediate group other than the conventional short\nand long bursts. However, very recently, few studies indicate a possibility of\nthe existence of more than three (namely five) groups. Therefore, in this\npaper, fuzzy clustering is conducted on the gamma-ray bursts from the final\n'Burst and Transient Source Experiment' catalog to cross-check the significance\nof these new groups. Meticulous study on individual bursts based on their\nmemberships in the fuzzy clusters confirms the previously well-known three\ngroups against the newly found five.\n"}
{"abstract": "  As reading on mobile devices is becoming more ubiquitous, content is consumed\nin shorter intervals and is punctuated by frequent interruptions. In this work,\nwe explore the best way to mitigate the effects of reading interruptions on\nlonger text passages. Our hypothesis is that short summaries of either\npreviously read content (reviews) or upcoming content (previews) will help the\nreader re-engage with the reading task. Our target use case is for students who\nstudy using electronic textbooks and who are frequently mobile. We present a\nseries of pilot studies that examine the benefits of different types of\nsummaries and their locations, with respect to variations in text content and\nparticipant cohorts. We find that users prefer reviews after an interruption,\nbut that previews shown after interruptions have a larger positive influence on\ncomprehension. Our work is a first step towards smart reading applications that\nproactively provide text summaries to mitigate interruptions on the go.\n"}
{"abstract": "  Crack-like objects that propagate along frictional interfaces,\ni.e.~frictional shear cracks, play a major role in a broad range of frictional\nphenomena. Such frictional cracks are commonly assumed to feature the universal\nsquare root near-edge singularity of ideal shear cracks, as predicted by Linear\nElastic Fracture Mechanics. Here we show that this is not the generic case due\nto the intrinsic dependence of the frictional strength on the slip rate, even\nif the bodies forming the frictional interface are identical and predominantly\nlinear elastic. Instead, frictional shear cracks feature unconventional\nsingularities characterized by a singularity order $\\xi$ that differs from the\nconventional $-\\tfrac{1}{2}$ one. It is shown that $\\xi$ depends on the\nfriction law, on the propagation speed and on the symmetry mode of loading. We\ndiscuss the general structure of a theory of unconventional singularities,\nalong with their implications for the energy balance and dynamics of frictional\ncracks. Finally, we present explicit calculations of $\\xi$ and the associated\nnear-edge fields for linear viscous-friction -- which may emerge as a\nperturbative approximation of nonlinear friction laws or on its own -- for both\nin-plane (mode-II) and anti-plane (mode-III) shear loadings.\n"}
{"abstract": "  Protoplanetary disks drive some of the formation process (e.g., accretion,\ngas dissipation, formation of structures, etc.) of stars and planets.\nUnderstanding such physical processes is one of the main astrophysical\nquestions. HD 163296 is an interesting young stellar object for which infrared\nand sub-millimeter observations have shown a prominent circumstellar disk with\ngaps plausibly created by forming planets. This study aims at characterizing\nthe morphology of the inner disk in HD 163296 with multi-epoch near-infrared\ninterferometric observations performed with GRAVITY at the Very Large Telescope\nInterferometer (VLTI). Our goal is to depict the K-band (lambda_0 ~ 2.2 um)\nstructure of the inner rim with milliarcsecond (sub-au) angular resolution. Our\ndata is complemented with archival PIONIER (H-band; lambda_0 ~ 1.65 um) data of\nthe source. We performed a Gradient Descent parametric model fitting to recover\nthe sub-au morphology of our source. Our analysis shows the existence of an\nasymmetry in the disk surrounding the central star of HD 163296. We confirm\nvariability of the disk structure in the inner ~2 mas (0.2 au). While\nvariability of the inner disk structure in this source has been suggested by\nprevious interferometric studies, this is the first time that it is confirmed\nin the H- and K-bands by using a complete analysis of the closure phases and\nsquared visibilities over several epochs. Because of the separation from the\nstar, position changes, and persistence of this asymmetric structure on\ntimescales of several years, we argue that it is a dusty feature (e.g., a\nvortex or dust clouds), probably, made by a mixing of sillicate and carbon dust\nand/or refractory grains, inhomogeneously distributed above the mid-plane of\nthe disk.\n"}
{"abstract": "  The explosive growth of bandwidth hungry Internet applications has led to the\nrapid development of new generation mobile network technologies that are\nexpected to provide broadband access to the Internet in a pervasive manner. For\nexample, 6G networks are capable of providing high-speed network access by\nexploiting higher frequency spectrum; high-throughout satellite communication\nservices are also adopted to achieve pervasive coverage in remote and isolated\nareas. In order to enable seamless access, Integrated Satellite-Terrestrial\nCommunication Networks (ISTCN) has emerged as an important research area. ISTCN\naims to provide high speed and pervasive network services by integrating\nbroadband terrestrial mobile networks with satellite communication networks. As\nterrestrial mobile networks began to use higher frequency spectrum (between\n3GHz to 40GHz) which overlaps with that of satellite communication (4GHz to\n8GHz for C band and 26GHz to 40GHz for Ka band), there are opportunities and\nchallenges. On one hand, satellite terminals can potentially access terrestrial\nnetworks in an integrated manner; on the other hand, there will be more\ncongestion and interference in this spectrum, hence more efficient spectrum\nmanagement techniques are required. In this paper, we propose a new technique\nto improve spectrum sharing performance by introducing Non-orthogonal Frequency\nDivision Multiplexing (NOMA) and Cognitive Radio (CR) in the spectrum sharing\nof ISTCN. In essence, NOMA technology improves spectrum efficiency by allowing\ndifferent users to transmit on the same carrier and distinguishing users by\nuser power levels while CR technology improves spectrum efficiency through\ndynamic spectrum sharing. Furthermore, some open researches and challenges in\nISTCN will be discussed.\n"}
{"abstract": "  Camera-based passive dietary intake monitoring is able to continuously\ncapture the eating episodes of a subject, recording rich visual information,\nsuch as the type and volume of food being consumed, as well as the eating\nbehaviours of the subject. However, there currently is no method that is able\nto incorporate these visual clues and provide a comprehensive context of\ndietary intake from passive recording (e.g., is the subject sharing food with\nothers, what food the subject is eating, and how much food is left in the\nbowl). On the other hand, privacy is a major concern while egocentric wearable\ncameras are used for capturing. In this paper, we propose a privacy-preserved\nsecure solution (i.e., egocentric image captioning) for dietary assessment with\npassive monitoring, which unifies food recognition, volume estimation, and\nscene understanding. By converting images into rich text descriptions,\nnutritionists can assess individual dietary intake based on the captions\ninstead of the original images, reducing the risk of privacy leakage from\nimages. To this end, an egocentric dietary image captioning dataset has been\nbuilt, which consists of in-the-wild images captured by head-worn and\nchest-worn cameras in field studies in Ghana. A novel transformer-based\narchitecture is designed to caption egocentric dietary images. Comprehensive\nexperiments have been conducted to evaluate the effectiveness and to justify\nthe design of the proposed architecture for egocentric dietary image\ncaptioning. To the best of our knowledge, this is the first work that applies\nimage captioning to dietary intake assessment in real life settings.\n"}
{"abstract": "  With the advancement of the accelerator systems and the requirements of high\nluminosity particle beams to reach different physics goals, detectors with good\nposition resolution and high rate handling capability have become essential for\ndesigning any High Energy Physics (HEP) experiments. The Gas Electron\nMultiplier (GEM) detectors are widely used in many HEP experiments as a\ntracking device because of their good spatial resolution and rate handling\ncapability. The presence of the dielectric medium inside the active volume of\nthe GEM detector changes its behaviour when exposed to external radiation. This\nmechanism is commonly referred as the charging-up effect. In this article, the\neffect of the charging-up phenomenon and the initial polarisation effect of the\ndielectric on the gain of the chamber are reported for a single mask triple GEM\nchamber with Ar/CO2 gas mixture.\n"}
{"abstract": "  We study the regret of reinforcement learning from offline data generated by\na fixed behavior policy in an infinite-horizon discounted Markov decision\nprocess (MDP). While existing analyses of common approaches, such as fitted\n$Q$-iteration (FQI), suggest a $O(1/\\sqrt{n})$ convergence for regret,\nempirical behavior exhibits much faster convergence. In this paper, we present\na finer regret analysis that exactly characterizes this phenomenon by providing\nfast rates for the regret convergence. First, we show that given any estimate\nfor the optimal quality function $Q^*$, the regret of the policy it defines\nconverges at a rate given by the exponentiation of the $Q^*$-estimate's\npointwise convergence rate, thus speeding it up. The level of exponentiation\ndepends on the level of noise in the decision-making problem, rather than the\nestimation problem. We establish such noise levels for linear and tabular MDPs\nas examples. Second, we provide new analyses of FQI and Bellman residual\nminimization to establish the correct pointwise convergence guarantees. As\nspecific cases, our results imply $O(1/n)$ regret rates in linear cases and\n$\\exp(-\\Omega(n))$ regret rates in tabular cases.\n"}
{"abstract": "  The comb-like spectrum of a white light-illuminated Fabry-P\\'{e}rot etalon\ncan serve as a cost-effective and stable reference for precise Doppler\nmeasurements. Understanding the stability of these devices across their broad\n(100's of nm) spectral bandwidths is essential to realize their full potential\nas Doppler calibrators. However, published descriptions remain limited to small\nbandwidths or short timespans. We present a $\\sim6$ month broadband stability\nmonitoring campaign of the Fabry-P\\'{e}rot etalon system deployed with the\nnear-infrared Habitable Zone Planet Finder spectrograph (HPF). We monitor the\nwavelengths of each of $\\sim3500$ resonant modes measured in HPF spectra of\nthis Fabry-P\\'{e}rot etalon (free spectral range = 30 GHz, bandwidth = 820 -\n1280 nanometers), leveraging the accuracy and precision of an electro-optic\nfrequency comb reference. These results reveal chromatic structure in the\nFabry-P\\'{e}rot mode locations and in their evolution with time. We measure an\naverage drift on the order of 2 cm s $^{-1}$ d$^{-1}$, with local departures up\nto $\\pm5$ cm s $^{-1}$ d$^{-1}$. We discuss these behaviors in the context of\nthe Fabry-P\\'{e}rot etalon mirror dispersion and other optical properties of\nthe system, and the implications for the use of similar systems for precise\nDoppler measurements. Our results show that this system supports the wavelength\ncalibration of HPF at the $\\lesssim10$ cm s $^{-1}$ level over a night and at\nthe $\\lesssim30$ cm s $^{-1}$ level over $\\sim10$ d. Our results also highlight\nthe need for long-term and spectrally-resolved study of similar systems that\nwill be deployed to support Doppler measurement precision approaching $\\sim10$\ncm s $^{-1}$.\n"}
{"abstract": "  This paper explores the calibration of a classifier output score in binary\nclassification problems. A calibrator is a function that maps the arbitrary\nclassifier score, of a testing observation, onto $[0,1]$ to provide an estimate\nfor the posterior probability of belonging to one of the two classes.\nCalibration is important for two reasons; first, it provides a meaningful\nscore, that is the posterior probability; second, it puts the scores of\ndifferent classifiers on the same scale for comparable interpretation. The\npaper presents three main contributions: (1) Introducing multi-score\ncalibration, when more than one classifier provides a score for a single\nobservation. (2) Introducing the idea that the classifier scores to a\ncalibration process are nothing but features to a classifier, hence proposing\nexpanding the classifier scores to higher dimensions to boost the calibrator's\nperformance. (3) Conducting a massive simulation study, in the order of 24,000\nexperiments, that incorporates different configurations, in addition to\nexperimenting on two real datasets from the cybersecurity domain. The results\nshow that there is no overall winner among the different calibrators and\ndifferent configurations. However, general advices for practitioners include\nthe following: the Platt's\ncalibrator~\\citep{Platt1999ProbabilisticOutputsForSupport}, a version of the\nlogistic regression that decreases bias for a small sample size, has a very\nstable and acceptable performance among all experiments; our suggested\nmulti-score calibration provides better performance than single score\ncalibration in the majority of experiments, including the two real datasets. In\naddition, expanding the scores can help in some experiments.\n"}
{"abstract": "  We trained deep neural networks (DNNs) as a function of the neutrino energy\ndensity, flux, and the fluid velocity to reproduce the Eddington tensor for\nneutrinos obtained in our first-principles core-collapse supernova (CCSN)\nsimulations. Although the moment method, which is one of the most popular\napproximations for neutrino transport, requires a closure relation, none of the\nanalytical closure relations commonly employed in the literature captures all\naspects of the neutrino angular distribution in momentum space. In this paper,\nwe developed a closure relation by using the DNN that takes the neutrino energy\ndensity, flux, and the fluid velocity as the input and the Eddington tensor as\nthe output. We consider two kinds of DNNs: a conventional DNN named a\ncomponent-wise neural network (CWNN) and a tensor-basis neural network (TBNN).\nWe found that the diagonal component of the Eddington tensor is reproduced\nbetter by the DNNs than the M1-closure relation especially for low to\nintermediate energies. For the off-diagonal component, the DNNs agree better\nwith the Boltzmann solver than the M1 closure at large radii. In the comparison\nbetween the two DNNs, the TBNN has slightly better performance than the CWNN.\nWith the new closure relations at hand based on the DNNs that well reproduce\nthe Eddington tensor with much smaller costs, we opened up a new possibility\nfor the moment method.\n"}
{"abstract": "  We show that a suitable weak solution to the incompressible Navier-Stokes\nequations on ${\\mathbb{R}^3\\times(-1,1)}$ is regular on\n$\\mathbb{R}^3\\times(0,1]$ if $\\partial_3 u $ belongs to $M^{2p/(2p-3),\\alpha }\n((-1,0);L^p (\\mathbb{R}^3 ))$ for any $\\alpha >1$ and $p\\in (3/2,\\infty)$,\nwhich is a logarithmic-type variation of a Morrey space in time. For each\n$\\alpha >1$ this space is, up to a logarithm, critical with respect to the\nscaling of the equations, and contains all spaces $L^q ((-1,0);L^p\n(\\mathbb{R}^3 ))$ that are subcritical, that is for which $2/q+3/p<2$.\n"}
{"abstract": "  Resource-disaggregated data centres (RDDC) propose a resource-centric, and\nhigh-utilisation architecture for data centres (DC), avoiding resource\nfragmentation and enabling arbitrarily sized resource pools to be allocated to\ntasks, rather than server-sized ones. RDDCs typically impose greater demand on\nthe network, requiring more infrastructure and increasing cost and power, so\nnew resource allocation algorithms that co-manage both server and networks\nresources are essential to ensure that allocation is not bottlenecked by the\nnetwork, and that requests can be served successfully with minimal networking\nresources. We apply reinforcement learning (RL) to this problem for the first\ntime and show that an RL policy based on graph neural networks can learn\nresource allocation policies end-to-end that outperform previous\nhand-engineered heuristics by up to 22.0\\%, 42.6\\% and 22.6\\% for acceptance\nratio, CPU and memory utilisation respectively, maintain performance when\nscaled up to RDDC topologies with $10^2\\times$ more nodes than those seen\nduring training and can achieve comparable performance to the best baselines\nwhile using $5.3\\times$ less network resources.\n"}
{"abstract": "  Vortex induced vibration (VIV) occurs when vortex shedding frequency falls\nclose to the natural frequency of a structure. Investigation on VIV is of great\nvalue in disaster mitigation, energy extraction and other applications.\nFollowing recent development in machine learning on VIV in laminar flow, this\nstudy extends it to the turbulent region by employing the state-of-the-art\nparameterised Navier-Stokes equations based physics informed neural network\n(PNS-PINN). Turbulent flow with Reynolds number $Re$ = $10^4$, passing through\na cylinder undergoing VIV motion, was considered as an example. Within the\nPNS-PINN, an artificial viscosity $\\nu_t$ is introduced in the Navier-Stokes\nequations and treated as a hidden output variable. A recently developed\nNavier-Stokes equations based PINN, termed NSFnets (Jin \\textit{et al.}, J\nCOMPUT PHYS 426: 109951, 2021), was also considered for comparison. Training\ndatasets of scattered velocity and dye trace concentration snapshots, from\ncomputational fluid dynamics (CFD) simulations, were prepared for both the\nPNS-PINN and NSFnets. Results show that, compared with the NSFnets, the\nPNS-PINN is more effective in inferring and reconstructing turbulent flow under\nVIV circumstances. The PNS-PINN also shows the capability to deal with unsteady\nand multi-scale flows in VIV. Inspired by the Reynolds-Averaged Navier-Stokes\n(RANS) formulation, by implanting an additional artificial viscosity, the\nPNS-PINN is free of complex turbulence model closure, thus may be applied for\nmore general and complex turbulent flows.\n"}
{"abstract": "  We develop a framework for model checking infinite-state systems by\nautomatically augmenting them with auxiliary variables, enabling\nquantifier-free induction proofs for systems that would otherwise require\nquantified invariants. We combine this mechanism with a counterexample-guided\nabstraction refinement scheme for the theory of arrays. Our framework can thus,\nin many cases, reduce inductive reasoning with quantifiers and arrays to\nquantifier-free and array-free reasoning. We evaluate the approach on a wide\nset of benchmarks from the literature. The results show that our implementation\noften outperforms state-of-the-art tools, demonstrating its practical\npotential.\n"}
{"abstract": "  This work discusses an extension to conventional low-coherence interferometry\nby the introduction of dispersion-encoding. The extension facilitates the\nmeasurement of surface height profiles with sub-nm resolution. The selection of\na dispersive element for encoding allows for tuning of the axial measurement\nrange and resolution of the setup. The approach is theoretically designed and\nimplemented for applications such as surface profilometry, the characterization\nof polymeric cross-linking and as a tool for the determination of layer\nthicknesses in thin-film processing. During the characterization of the\nimplemented setup, it was shown that an axial measurement range of 79.91 $\\mu\nm$ with a resolution of 0.1 nm was achievable in the evaluation of surface\nprofiles. Simultaneously, profiles of up to 1.5 mm length could be obtained\nwithout the need for mechanical scanning. This marked a significant improvement\nin relation to state-of-the-art technologies in terms of dynamic range. It was\nalso shown that axial and lateral measurement range can be decoupled partially.\nAdditionally, functional parameters such as surface roughness were\ncharacterized with the same tool. The characterization of the degree of\npolymeric cross-linking was performed as a function of the refractive index.\nHere, the refractive index could be acquired in a spatially-resolved manner\nwith an index resolution of down to $3.36 \\pm 10^{-5}$. This was achieved by\nthe development of a novel mathematical analysis approach. For the acquisition\nof layer thicknesses of thin-films, an advanced setup was developed which could\nbe used to characterize the thickness thin-films and its (flexible) substrate\nsimultaneously.\n"}
{"abstract": "  Where do cosmic X-rays come from? Every new unidentified X-ray source could\npotentially revolutionize our understanding of the universe. The international\ncollaborative astronomy project EXTraS aimed at automatically classifying new\nsources of X-ray emission (e.g., stars or galaxies) in the large observation\ndatabase of the X-ray satellite XMM-Newton. Because data archives have reached\ndimensions of big data astronomers used different machine-learning (ML) random\nforest decision tree algorithms that performed the classification process. In\nthis bachelor thesis in information design, I was interested in the challenge\nto visualize these big data sets and the results of the ML algorithms in an\ninteractive and intuitive way to facilitate the visual exploration of its\ninternal structures and relationships. The VIRTUAL DATA COSMOS is an\ninteractive data visualization tool in virtual reality (VR) for scientists to\nexplore multidimensional data sets.\n"}
{"abstract": "  It was established by Boalch that Euler continuants arise as Lie group valued\nmoment maps for a class of wild character varieties described as moduli spaces\nof points on $\\mathbb{P}^1$ by Sibuya. Furthermore, Boalch noticed that these\nvarieties are multiplicative analogues of certain Nakajima quiver varieties\noriginally introduced by Calabi, which are attached to the quiver $\\Gamma_n$ on\ntwo vertices and $n$ equioriented arrows. In this article, we go a step further\nby unveiling that the Sibuya varieties can be understood using noncommutative\nquasi-Poisson geometry modeled on the quiver $\\Gamma_n$. We prove that the\nPoisson structure carried by these varieties is induced, via the\nKontsevich-Rosenberg principle, by an explicit Hamiltonian double quasi-Poisson\nalgebra defined at the level of the quiver $\\Gamma_n$ such that its\nnoncommutative multiplicative moment map is given in terms of Euler\ncontinuants. This result generalises the Hamiltonian double quasi-Poisson\nalgebra associated with the quiver $\\Gamma_1$ by Van den Bergh. Moreover, using\nthe method of fusion, we prove that the Hamiltonian double quasi-Poisson\nalgebra attached to $\\Gamma_n$ admits a factorisation in terms of $n$ copies of\nthe algebra attached to $\\Gamma_1$.\n"}
{"abstract": "  The Gibbs-Thomson (GT) equation describes the shift of the crystallization\ntemperature for a confined fluid with respect to the bulk as a function of pore\nsize. While this century old relation is successfully used to analyze\nexperiments, its derivations found in the literature often rely on nucleation\ntheory arguments (i.e. kinetics instead of thermodynamics) or fail to state\ntheir assumptions, therefore leading to similar but different expressions.\nHere, we revisit the derivation of the GT equation to clarify the system\ndefinition, corresponding thermodynamic ensemble, and assumptions made along\nthe way. We also discuss the role of the thermodynamic conditions in the\nexternal reservoir on the final result. We then turn to numerical simulations\nof a model system to compute independently the various terms entering in the GT\nequation, and compare the predictions of the latter with the melting\ntemperatures determined under confinement by means of hyper-parallel tempering\ngrand canonical Monte Carlo simulations. We highlight some difficulties related\nto the sampling of crystallization under confinement in simulations. Overall,\ndespite its limitations, the GT equation may provide an interesting alternative\nroute to predict the melting temperature in large pores, using molecular\nsimulations to evaluate the relevant quantities entering in this equation. This\napproach could for example be used to investigate the nanoscale capillary\nfreezing of ionic liquids recently observed experimentally between the tip of\nan Atomic Force Microscope and a substrate.\n"}
{"abstract": "  Recently, we have introduced and modified two graph-decomposition theorems\nbased on a new graph product, motivated by applications in the context of\nsynchronising periodic real-time processes. This vertex-removing synchronised\nproduct (VRSP), is based on modifications of the well-known Cartesian product,\nand is closely related to the synchronised product due to Wohrle and Thomas.\nHere, we recall the definition of the VRSP and the two modified\ngraph-decompositions and introduce three new graph-decomposition theorems. The\nfirst new theorem decomposes a graph with respect to the semicomplete bipartite\nsubgraphs of the graph. For the second new theorem, we introduce a matrix\ngraph, which is used to decompose a graph in a manner similar to the\ndecomposition of graphs using the Cartesian product. In the third new theorem,\nwe combine these two types of decomposition. Ultimately, the goal of these\ngraph-decomposition theorems is to come to a prime-graph decomposition.\n"}
{"abstract": "  Frictional sliding is an intrinsically complex phenomenon, emerging from the\ninterplay between driving forces, elasto-frictional instabilities, interfacial\nnonlinearity and dissipation, material inertia and bulk geometry. We show that\nhomogeneous rate-and-state dependent frictional systems, driven at a prescribed\nboundary velocity -- as opposed to a prescribed stress -- in a range where the\nfrictional interface is rate-weakening, generically host self-healing slip\npulses, a sliding mode not yet fully understood. Such velocity-driven\nfrictional systems are then shown to exhibit coarsening dynamics saturated at\nthe system length in the sliding direction, independently of the system's\nheight, leading to steadily propagating pulse trains. The latter may be viewed\nas a propagating phase-separated state, where slip and stick characterize the\ntwo phases. While pulse trains' periodicity is coarsening-limited by the\nsystem's length, the single pulse width, characteristic slip velocity and\npropagation speed exhibit rich properties, which are comprehensively understood\nusing theory and extensive numerics. Finally, we show that for sufficiently\nsmall system heights, pulse trains are accompanied by periodic\nelasto-frictional instabilities.\n"}
{"abstract": "  To ease the burden of labeling, unsupervised domain adaptation (UDA) aims to\ntransfer knowledge in previous and related labeled datasets (sources) to a new\nunlabeled dataset (target). Despite impressive progress, prior methods always\nneed to access the raw source data and develop data-dependent alignment\napproaches to recognize the target samples in a transductive learning manner,\nwhich may raise privacy concerns from source individuals. Several recent\nstudies resort to an alternative solution by exploiting the well-trained\nwhite-box model from the source domain, yet, it may still leak the raw data\nthrough generative adversarial learning. This paper studies a practical and\ninteresting setting for UDA, where only black-box source models (i.e., only\nnetwork predictions are available) are provided during adaptation in the target\ndomain. To solve this problem, we propose a new two-step knowledge adaptation\nframework called DIstill and fine-tuNE (DINE). Taking into consideration the\ntarget data structure, DINE first distills the knowledge from the source\npredictor to a customized target model, then fine-tunes the distilled model to\nfurther fit the target domain. Besides, neural networks are not required to be\nidentical across domains in DINE, even allowing effective adaptation on a\nlow-resource device. Empirical results on three UDA scenarios (i.e.,\nsingle-source, multi-source, and partial-set) confirm that DINE achieves highly\ncompetitive performance compared to state-of-the-art data-dependent approaches.\nCode is available at \\url{https://github.com/tim-learn/DINE/}.\n"}
{"abstract": "  We study the dynamics of a small number of impurity particles coupled to the\nideal Fermi gas in a $d$-dimensional box. The impurities interact with the\nfermions via a two-body potential $\\lambda v(x)$ where $\\lambda$ is a coupling\nconstant and $v(x)$ for instance a screened Coulomb potential. After taking the\nlarge-volume limit at positive Fermi momentum $k_F$ we consider the regime of\nhigh density of the fermions, that is, $k_F$ large compared to one. For\ncoupling constants that scale like $\\lambda^2 \\sim k_F^{(2-d)}$ we show that\nthe impurity particles effectively decouple from the fermions but evolve with\nan attractive pair interaction among each other which is induced by\nfluctuations in the Fermi gas.\n"}
{"abstract": "  We perform a covariant (Lagrangian) quantization of perturbative gravity in\nthe background of a Schwarzschild black hole. The key tool is a decomposition\nof the field into spherical harmonics. We fix Regge-Wheeler gauge for modes\nwith angular momentum quantum number $l \\geq 2$, while for low multipole modes\nwith $l$ $=$ $0$ or $1$ -- for which Regge-Wheeler gauge is inapplicable -- we\npropose a set of gauge fixing conditions which are 2D background covariant and\nperturbatively well-defined. We find that the corresponding Faddeev-Popov\nghosts are non-propagating for the $l\\geq2$ modes, but are in general\nnontrivial for the low multipole modes with $l = 0,1$. However, in\nSchwarzschild coordinates, all time derivatives acting on the ghosts drop from\nthe action and the low multipole ghosts have instantaneous propagators. Up to\npossible subtleties related to quantizing gravity in a space with a horizon,\nFaddeev's theorem suggests the possibility of an underlying canonical\n(Hamiltonian) quantization with a manifestly ghost-free Hilbert space.\n"}
{"abstract": "  Although deep neural networks hold the state-of-the-art in several remote\nsensing tasks, their black-box operation hinders the understanding of their\ndecisions, concealing any bias and other shortcomings in datasets and model\nperformance. To this end, we have applied explainable artificial intelligence\n(XAI) methods in remote sensing multi-label classification tasks towards\nproducing human-interpretable explanations and improve transparency. In\nparticular, we utilized and trained deep learning models with state-of-the-art\nperformance in the benchmark BigEarthNet and SEN12MS datasets. Ten XAI methods\nwere employed towards understanding and interpreting models' predictions, along\nwith quantitative metrics to assess and compare their performance. Numerous\nexperiments were performed to assess the overall performance of XAI methods for\nstraightforward prediction cases, competing multiple labels, as well as\nmisclassification cases. According to our findings, Occlusion, Grad-CAM and\nLime were the most interpretable and reliable XAI methods. However, none\ndelivers high-resolution outputs, while apart from Grad-CAM, both Lime and\nOcclusion are computationally expensive. We also highlight different aspects of\nXAI performance and elaborate with insights on black-box decisions in order to\nimprove transparency, understand their behavior and reveal, as well, datasets'\nparticularities.\n"}
{"abstract": "  In a time division broadcast positioning system (TDBPS), a user device (UD)\ndetermines its position by obtaining sequential time-of-arrival (TOA) or\npseudorange measurements from signals broadcast by multiple synchronized base\nstations (BSs). The existing localization method using sequential pseudorange\nmeasurements and a linear clock drift model for the TDPBS, namely LSPM-D, does\nnot compensate the position displacement caused by the UD movement and will\nresult in position error. In this paper, depending on the knowledge of the UD\nvelocity, we develop a set of optimal localization methods for different cases.\nFirst, for known UD velocity, we develop the optimal localization method,\nnamely LSPM-KVD, to compensate the movement-caused position error. We show that\nthe LSPM-D is a special case of the LSPM-KVD when the UD is stationary with\nzero velocity. Second, for the case with unknown UD velocity, we develop a\nmaximum likelihood (ML) method to jointly estimate the UD position and\nvelocity, namely LSPM-UVD. Third, in the case that we have prior distribution\ninformation of the UD velocity, we present a maximum a posteriori (MAP)\nestimator for localization, namely LSPM-PVD. We derive the Cramer-Rao lower\nbound (CRLB) for all three estimators and analyze their localization error\nperformance. We show that the position error of the LSPM-KVD increases as the\nassumed known velocity deviates from the true value. As expected, the LSPM-KVD\nhas the smallest position error while the LSPM-PVD and the LSPM-UVD are more\nrobust when the prior knowledge of the UD velocity is limited. Numerical\nresults verify the theoretical analysis on the optimality and the positioning\naccuracy of the proposed methods.\n"}
{"abstract": "  Modeling various aspects that make a music piece unique is a challenging\ntask, requiring the combination of multiple sources of information. Deep\nlearning is commonly used to obtain representations using various sources of\ninformation, such as the audio, interactions between users and songs, or\nassociated genre metadata. Recently, contrastive learning has led to\nrepresentations that generalize better compared to traditional supervised\nmethods. In this paper, we present a novel approach that combines multiple\ntypes of information related to music using cross-modal contrastive learning,\nallowing us to learn an audio feature from heterogeneous data simultaneously.\nWe align the latent representations obtained from playlists-track interactions,\ngenre metadata, and the tracks' audio, by maximizing the agreement between\nthese modality representations using a contrastive loss. We evaluate our\napproach in three tasks, namely, genre classification, playlist continuation\nand automatic tagging. We compare the performances with a baseline audio-based\nCNN trained to predict these modalities. We also study the importance of\nincluding multiple sources of information when training our embedding model.\nThe results suggest that the proposed method outperforms the baseline in all\nthe three downstream tasks and achieves comparable performance to the\nstate-of-the-art.\n"}
{"abstract": "  PageRank (PR) is a fundamental tool for assessing the relative importance of\nthe nodes in a network. In this paper, we propose a measure, weighted PageRank\n(WPR), extended from the classical PR for weighted, directed networks with\npossible non-uniform node-specific information that is dependent or independent\nof network structure. A tuning parameter leveraging node degree and strength is\nintroduced. An efficient algorithm based on R program has been developed for\ncomputing WPR in large-scale networks. We have tested the proposed WPR on\nwidely used simulated network models, and found it outperformed other competing\nmeasures in the literature. By applying the proposed WPR to the real network\ndata generated from World Input-Output Tables, we have seen the results that\nare consistent with the global economic trends, which renders it a preferred\nmeasure in the analysis.\n"}
{"abstract": "  Background: Distinction of IDH mutant and wildtype GBMs is challenging on\nMRI, since conventional imaging shows considerable overlap. While few studies\nemployed deep-learning in a mixed low/high grade glioma population, a\nGBM-specific model is still lacking in the literature. Our objective was to\ndevelop a deep-learning model for IDH prediction in GBM by using Convoluted\nNeural Networks (CNN) on multiparametric MRI. Methods: We included 100 adult\npatients with pathologically proven GBM and IDH testing. MRI data included:\nmorphologic sequences, rCBV and ADC maps. Tumor area was obtained by a bounding\nbox function on the axial slice with widest tumor extension on T2 images and\nwas projected on every sequence. Data was split into training and test (80:20)\nsets. A 4 block 2D - CNN architecture was implemented for IDH prediction on\nevery MRI sequence. IDH mutation probability was calculated with softmax\nactivation function from the last dense layer. Highest performance was\ncalculated accounting for model accuracy and categorical cross-entropy loss\n(CCEL) in the test cohort. Results: Our model achieved the following\nperformance: T1 (accuracy 77%, CCEL 1.4), T2 (accuracy 67%, CCEL 2.41), FLAIR\n(accuracy 77%, CCEL 1.98), MPRAGE (accuracy 66%, CCEL 2.55), rCBV (accuracy\n83%, CCEL 0.64). ADC achieved lower performance. Conclusion: We built a\nGBM-tailored deep-learning model for IDH mutation prediction, achieving\naccuracy of 83% with rCBV maps. High predictivity of perfusion images may\nreflect the known correlation between IDH, hypoxia inducible factor (HIF) and\nneoangiogenesis. This model may set a path for non-invasive evaluation of IDH\nmutation in GBM.\n"}
{"abstract": "  In this paper we study the distributed average consensus problem in\nmulti-agent systems with directed communication links that are subject to\nquantized information flow. Specifically, we present and analyze a distributed\naveraging algorithm which operates exclusively with quantized values (i.e., the\ninformation stored, processed and exchanged between neighboring agents is\nsubject to deterministic uniform quantization) and relies on event-driven\nupdates (e.g., to reduce energy consumption, communication bandwidth, network\ncongestion, and/or processor usage). The main idea of the proposed algorithm is\nthat each node (i) models its initial state as two quantized fractions which\nhave numerators equal to the node's initial state and denominators equal to\none, and (ii) transmits one fraction randomly while it keeps the other stored.\nThen, every time it receives one or more fractions, it averages their\nnumerators with the numerator of the fraction it stored, and then transmits\nthem to randomly selected out-neighbors. We characterize the properties of the\nproposed distributed algorithm and show that its execution, on any static and\nstrongly connected digraph, allows each agent to reach in finite time a fixed\nstate that is equal (within one quantisation level) to the average of the\ninitial states. We extend the operation of the algorithm to achieve finite-time\nconvergence in the presence of a dynamic directed communication topology\nsubject to some connectivity conditions. Finally, we provide examples to\nillustrate the operation, performance, and potential advantages of the proposed\nalgorithm. We compare against state-of-the-art quantized average consensus\nalgorithms and show that our algorithm's convergence speed significantly\noutperforms most existing protocols.\n"}
{"abstract": "  The behaviour of information cascades (such as retweets) has been modelled\nextensively. While point process-based generative models have long been in use\nfor estimating cascade growths, deep learning has greatly enhanced diverse\nfeature integration. We observe two significant temporal signals in cascade\ndata that have not been emphasized or reported to our knowledge. First, the\npopularity of the cascade root is known to influence cascade size strongly; but\nthe effect falls off rapidly with time. Second, there is a measurable positive\ncorrelation between the novelty of the root content (with respect to a\nstreaming external corpus) and the relative size of the resulting cascade.\nResponding to these observations, we propose GammaCas, a new cascade growth\nmodel as a parametric function of time, which combines deep influence signals\nfrom content (e.g., tweet text), network features (e.g., followers of the root\nuser), and exogenous event sources (e.g., online news). Specifically, our model\nprocesses these signals through a customized recurrent network, whose states\nthen provide the parameters of the cascade rate function, which is integrated\nover time to predict the cascade size. The network parameters are trained\nend-to-end using observed cascades. GammaCas outperforms seven recent and\ndiverse baselines significantly on a large-scale dataset of retweet cascades\ncoupled with time-aligned online news -- it beats the best baseline with an\n18.98% increase in terms of Kendall's $\\tau$ correlation and $35.63$ reduction\nin Mean Absolute Percentage Error. Extensive ablation and case studies unearth\ninteresting insights regarding retweet cascade dynamics.\n"}
{"abstract": "  Acoustic emission analysis is a promising technique to investigate the\nphysiological events leading to drought-induced injuries and mortality.\nHowever, the nature and the source of the acoustic emissions are not fully\nunderstood and make the use of this technique difficult as a direct measure of\nthe loss of xylem hydraulic conductance. In this study, acoustic emissions were\nrecorded during severe dehydration in lavender plants and compared to the\ndynamics of embolism development and cell lysis. The timing and characteristics\nof acoustic signals from two independent recording systems were compared by\nprincipal component analysis. In parallel, changes in water potential, branch\ndiameter, loss of hydraulic conductance and electrolyte leakage were measured\nto quantify drought-induced damages. Two distinct phases of acoustic emissions\nwere observed during dehydration: the first one associated with a rapid loss of\ndiameter and a significant increase in loss of xylem conductance (90%) and the\nsecond one with a significant increase in electrolyte leakage and slower\ndiameter changes. This second phase corresponds to a complete loss of recovery\ncapacity. The acoustic signals of both phases were discriminated by the third\nand fourth principal components. The loss of hydraulic conductance during the\nfirst acoustic phase suggests the hydraulic origin of these signals (i.e.\ncavitation events). For the second phase, the signals showed much higher\nvariability between plants and acoustic systems suggesting that the sources of\nthese signals may be plural, although likely including cellular damage. A\nsimple algorithm was developed to discriminate hydraulic-related acoustic\nsignals from other sources, allowing the reconstruction of dynamic hydraulic\nvulnerability curves. However, hydraulic failure precedes cellular damage and\nlack of whole plant recovery is associated to these latter.\n"}
{"abstract": "  We investigate the problem of active learning in the streaming setting in\nnon-parametric regimes, where the labels are stochastically generated from a\nclass of functions on which we make no assumptions whatsoever. We rely on\nrecently proposed Neural Tangent Kernel (NTK) approximation tools to construct\na suitable neural embedding that determines the feature space the algorithm\noperates on and the learned model computed atop. Since the shape of the label\nrequesting threshold is tightly related to the complexity of the function to be\nlearned, which is a-priori unknown, we also derive a version of the algorithm\nwhich is agnostic to any prior knowledge. This algorithm relies on a regret\nbalancing scheme to solve the resulting online model selection problem, and is\ncomputationally efficient. We prove joint guarantees on the cumulative regret\nand number of requested labels which depend on the complexity of the labeling\nfunction at hand. In the linear case, these guarantees recover known minimax\nresults of the generalization error as a function of the label complexity in a\nstandard statistical learning setting.\n"}
{"abstract": "  We show that the physical conditions which induce the Thakurta metric,\nrecently studied by B{\\oe}hm {\\it et al.} in the context of time-dependent\nblack hole masses, correspond to a single accreting black hole in the entire\nUniverse filled with isotropic non-interacting dust. In such a case, the\nphysics of black hole accretion is not local but tied to the properties of the\nwhole Universe. We show that radiation, primordial black holes or particle dark\nmatter cannot produce the specific energy flux required for supporting the mass\ngrowth of Thakurta black holes. In particular, this solution does not apply to\nblack hole binaries. We conclude that cosmological black holes and their mass\ngrowth cannot be described by the Thakurta metric, and thus existing\nconstraints on the primordial black hole abundance from the LIGO-Virgo and the\nCMB measurements remain valid.\n"}
{"abstract": "  We propose a simple yet effective framework for instance and panoptic\nsegmentation, termed CondInst (conditional convolutions for instance and\npanoptic segmentation). In the literature, top-performing instance segmentation\nmethods typically follow the paradigm of Mask R-CNN and rely on ROI operations\n(typically ROIAlign) to attend to each instance. In contrast, we propose to\nattend to the instances with dynamic conditional convolutions. Instead of using\ninstance-wise ROIs as inputs to the instance mask head of fixed weights, we\ndesign dynamic instance-aware mask heads, conditioned on the instances to be\npredicted. CondInst enjoys three advantages: 1.) Instance and panoptic\nsegmentation are unified into a fully convolutional network, eliminating the\nneed for ROI cropping and feature alignment. 2.) The elimination of the ROI\ncropping also significantly improves the output instance mask resolution. 3.)\nDue to the much improved capacity of dynamically-generated conditional\nconvolutions, the mask head can be very compact (e.g., 3 conv. layers, each\nhaving only 8 channels), leading to significantly faster inference time per\ninstance and making the overall inference time almost constant, irrelevant to\nthe number of instances. We demonstrate a simpler method that can achieve\nimproved accuracy and inference speed on both instance and panoptic\nsegmentation tasks. On the COCO dataset, we outperform a few state-of-the-art\nmethods. We hope that CondInst can be a strong baseline for instance and\npanoptic segmentation. Code is available at: https://git.io/AdelaiDet\n"}
{"abstract": "  Domain adaptation is to transfer the shared knowledge learned from the source\ndomain to a new environment, i.e., target domain. One common practice is to\ntrain the model on both labeled source-domain data and unlabeled target-domain\ndata. Yet the learned models are usually biased due to the strong supervision\nof the source domain. Most researchers adopt the early-stopping strategy to\nprevent over-fitting, but when to stop training remains a challenging problem\nsince the lack of the target-domain validation set. In this paper, we propose\none efficient bootstrapping method, called Adaboost Student, explicitly\nlearning complementary models during training and liberating users from\nempirical early stopping. Adaboost Student combines the deep model learning\nwith the conventional training strategy, i.e., adaptive boosting, and enables\ninteractions between learned models and the data sampler. We adopt one adaptive\ndata sampler to progressively facilitate learning on hard samples and aggregate\n\"weak\" models to prevent over-fitting. Extensive experiments show that (1)\nWithout the need to worry about the stopping time, AdaBoost Student provides\none robust solution by efficient complementary model learning during training.\n(2) AdaBoost Student is orthogonal to most domain adaptation methods, which can\nbe combined with existing approaches to further improve the state-of-the-art\nperformance. We have achieved competitive results on three widely-used scene\nsegmentation domain adaptation benchmarks.\n"}
{"abstract": "  Analyzing the effects of ocean eddies is important in oceanology for gaining\ninsights into transport of energy and biogeochemical particles. We present an\napplication of statistical visualization algorithms for the analysis of the Red\nSea eddy simulation ensemble. Specifically, we demonstrate the applications of\nstatistical volume rendering and statistical Morse complex summary maps to a\nvelocity magnitude field for studying the eddy positions in the flow dataset.\nIn statistical volume rendering, we model per-voxel data uncertainty using\nnoise models, such as parametric and nonparametric, and study the propagation\nof uncertainty into the volume rendering pipeline. In the statistical Morse\ncomplex summary maps, we derive histograms charactering uncertainty of gradient\nflow destinations to understand Morse complex topological variations across the\nensemble. We demonstrate the utility of our statistical visualizations for an\neffective analysis of the potential eddy positions and their spatial\nuncertainty.\n"}
{"abstract": "  We prove that, for almost all $r \\leq N^{1/2}/\\log^{O(1)}N$, for any given\n$b_1 \\mod r$ with $(b_1, r) = 1$, and for almost all $b_2 \\mod r$ with $(b_2,\nr) = 1$, we have that almost all natural numbers $2n \\leq N$ with $2n \\equiv\nb_1 + b_2 \\mod r$ can be written as the sum of two prime numbers $2n = p_1 +\np_2$, where $p_1 \\equiv b_1 \\mod r$ and $p_2 \\equiv b_2 \\mod r$. This improves\nthe previous result which required $r \\leq N^{1/3}/\\log^{O(1)}N$ instead of $r\n\\leq N^{1/2}/\\log^{O(1)}N$. We also improve some other results concerning\nvariations of the problem.\n"}
{"abstract": "  A common observation regarding adversarial attacks is that they mostly give\nrise to false activation at the penultimate layer to fool the classifier.\nAssuming that these activation values correspond to certain features of the\ninput, the objective becomes choosing the features that are most useful for\nclassification. Hence, we propose a novel approach to identify the important\nfeatures by employing counter-adversarial attacks, which highlights the\nconsistency at the penultimate layer with respect to perturbations on input\nsamples. First, we empirically show that there exist a subset of features,\nclassification based in which bridge the gap between the clean and robust\naccuracy. Second, we propose a simple yet efficient mechanism to identify those\nfeatures by searching the neighborhood of input sample. We then select features\nby observing the consistency of the activation values at the penultimate layer.\n"}
{"abstract": "  In this paper, we study the exact dynamics of open quantum systems to the\ncase with periodic driving field. It is shown that different from the static\nadjustment of the system on-site energy that can either generate or destroy the\ndissipationless localized bound states, the periodic driving can either\npreserve the existed localized bound states or destroy some of them but cannot\ngenerate new localized bound states. With the picture of energy transfer\ninvolved with the driving field, we find the condition for the survival of the\nlocalized bound states when the driving amplitude is weak. For the strong\ndriving case, the condition breaks down because of the strong energy\nrenormalization to the originally existed localized bound states. These\nproperties of decoherence dynamics may help in controlling the quantum state\nagainst decoherence for the sake of its sensitivity to the fundamental\nfrequency of the driving field.\n"}
{"abstract": "  This Overview, Design Concepts, and Details Protocol (ODD) provides a\ndetailed description of an agent-based model (ABM) that was developed to\nsimulate hospitalizations during the COVID-19 pandemic. Using the descriptions\nof submodels, provided parameters, and the links to data sources, modelers will\nbe able to replicate the creation and results of this model.\n"}
{"abstract": "  Non-intrusive load monitoring (NILM) is essential for understanding\ncustomer's power consumption patterns and may find wide applications like\ncarbon emission reduction and energy conservation. The training of NILM models\nrequires massive load data containing different types of appliances. However,\ninadequate load data and the risk of power consumer privacy breaches may be\nencountered by local data owners during the NILM model training. To prevent\nsuch potential risks, a novel NILM method named Fed-NILM which is based on\nFederated Learning (FL) is proposed in this paper. In Fed-NILM, local model\nparameters instead of local load data are shared among multiple data owners.\nThe global model is obtained by weighted averaging the parameters. Experiments\nbased on two measured load datasets are conducted to explore the generalization\nability of Fed-NILM. Besides, a comparison of Fed-NILM with locally-trained\nNILMs and the centrally-trained NILM is conducted. The experimental results\nshow that Fed-NILM has superior performance in scalability and convergence.\nFed-NILM outperforms locally-trained NILMs operated by local data owners and\napproximates the centrally-trained NILM which is trained on the entire load\ndataset without privacy protection. The proposed Fed-NILM significantly\nimproves the co-modeling capabilities of local data owners while protecting\npower consumers' privacy.\n"}
{"abstract": "  Virtualization technology is nowadays adopted in security-critical embedded\nsystems to achieve higher performance and more design flexibility. However, it\nalso comes with new security threats, where attackers leverage timing covert\nchannels to exfiltrate sensitive information from a partition using a trojan.\n  This paper presents a novel approach for the experimental assessment of\ntiming covert channels in embedded hypervisors, with a case study on security\nassessment of a commercial hypervisor product (Wind River VxWorks MILS), in\ncooperation with a licensed laboratory for the Common Criteria security\ncertification. Our experimental analysis shows that it is indeed possible to\nestablish a timing covert channel, and that the approach is useful for system\ndesigners for assessing that their configuration is robust against this kind of\ninformation leakage.\n"}
{"abstract": "  Recommender systems (RS) have started to employ knowledge distillation, which\nis a model compression technique training a compact model (student) with the\nknowledge transferred from a cumbersome model (teacher). The state-of-the-art\nmethods rely on unidirectional distillation transferring the knowledge only\nfrom the teacher to the student, with an underlying assumption that the teacher\nis always superior to the student. However, we demonstrate that the student\nperforms better than the teacher on a significant proportion of the test set,\nespecially for RS. Based on this observation, we propose Bidirectional\nDistillation (BD) framework whereby both the teacher and the student\ncollaboratively improve with each other. Specifically, each model is trained\nwith the distillation loss that makes to follow the other's prediction along\nwith its original loss function. For effective bidirectional distillation, we\npropose rank discrepancy-aware sampling scheme to distill only the informative\nknowledge that can fully enhance each other. The proposed scheme is designed to\neffectively cope with a large performance gap between the teacher and the\nstudent. Trained in the bidirectional way, it turns out that both the teacher\nand the student are significantly improved compared to when being trained\nseparately. Our extensive experiments on real-world datasets show that our\nproposed framework consistently outperforms the state-of-the-art competitors.\nWe also provide analyses for an in-depth understanding of BD and ablation\nstudies to verify the effectiveness of each proposed component.\n"}
{"abstract": "  There is a growing interest in Behavior Trees (BTs) as a tool to describe and\nimplement robot behaviors. BTs were devised in the video game industry and\ntheir adoption in robotics resulted in the development of ad-hoc libraries to\ndesign and execute BTs that fit complex robotics software architectures.\n  While there is broad consensus on how BTs work, some characteristics rely on\nthe implementation choices done in the specific software library used.\n  In this letter, we outline practical aspects in the adoption of BTs and the\nsolutions devised by the robotics community to fully exploit the advantages of\nBTs in real robots. We also overview the solutions proposed in open-source\nlibraries used in robotics, we show how BTs fit in robotic software\narchitecture, and we present a use case example.\n"}
{"abstract": "  The use of millimeter waves for wireless communications is one of the main\ntechnological innovations of 5G systems with respect to previous generations of\ncellular systems. Their consideration, however, has been up to now mainly\nrestricted to the case in which analog or, at most, hybrid analog-digital\nbeamforming structures were used, thus posing a limitation on the multiplexing\ncapabilities and peak data rates that could be theoretically achieved at these\nfrequencies. Recent progress in the field of electronics, however, has made the\nenergy consumption of digital beamforming structures at least on par with that\nof analog beamforming, thus redeeming them from the ghetto they had been placed\nin over the last year. Digital beamforming, coupled with the use of large\nantenna arrays at both sides of the communication link, promises thus to be one\nof the secret weapons of future 6G networks, capable of unleashing\nunprecedented values of spectral and energy efficiency for ultra-broadband\nconnectivity.\n"}
{"abstract": "  The Ensemble Kalman Filter (EnKF) has achieved great successes in data\nassimilation in atmospheric and oceanic sciences, but its failure in\nconvergence to the right filtering distribution precludes its use for\nuncertainty quantification. We reformulate the EnKF under the framework of\nLangevin dynamics, which leads to a new particle filtering algorithm, the\nso-called Langevinized EnKF. The Langevinized EnKF inherits the\nforecast-analysis procedure from the EnKF and the use of mini-batch data from\nthe stochastic gradient Langevin-type algorithms, which make it scalable with\nrespect to both the dimension and sample size. We prove that the Langevinized\nEnKF converges to the right filtering distribution in Wasserstein distance\nunder the big data scenario that the dynamic system consists of a large number\nof stages and has a large number of samples observed at each stage. We\nreformulate the Bayesian inverse problem as a dynamic state estimation problem\nbased on the techniques of subsampling and Langevin diffusion process. We\nillustrate the performance of the Langevinized EnKF using a variety of\nexamples, including the Lorenz-96 model, high-dimensional variable selection,\nBayesian deep learning, and Long Short Term Memory (LSTM) network learning with\ndynamic data.\n"}
{"abstract": "  We introduce dissipative solutions to the compressible Navier-Stokes system\nwith potential temperature transport motivated by the concept of Young\nmeasures. We prove their global-in-time existence by means of convergence\nanalysis of a mixed finite element-finite volume method. If a classical\nsolution to the compressible Navier-Stokes system with potential temperature\ntransport exists, we prove the strong convergence of numerical solutions. Our\nresults hold for the full range of adiabatic indices including the physically\nrelevant cases in which the existence of global-in-time weak solutions is open.\n"}
{"abstract": "  A graph neural network (GNN) enables deep learning on structured graph data.\nThere are two major GNN training obstacles: 1) it relies on high-end servers\nwith many GPUs which are expensive to purchase and maintain, and 2) limited\nmemory on GPUs cannot scale to today's billion-edge graphs. This paper presents\nDorylus: a distributed system for training GNNs. Uniquely, Dorylus can take\nadvantage of serverless computing to increase scalability at a low cost.\n  The key insight guiding our design is computation separation. Computation\nseparation makes it possible to construct a deep, bounded-asynchronous pipeline\nwhere graph and tensor parallel tasks can fully overlap, effectively hiding the\nnetwork latency incurred by Lambdas. With the help of thousands of Lambda\nthreads, Dorylus scales GNN training to billion-edge graphs. Currently, for\nlarge graphs, CPU servers offer the best performance-per-dollar over GPU\nservers. Just using Lambdas on top of CPU servers offers up to 2.75x more\nperformance-per-dollar than training only with CPU servers. Concretely, Dorylus\nis 1.22x faster and 4.83x cheaper than GPU servers for massive sparse graphs.\nDorylus is up to 3.8x faster and 10.7x cheaper compared to existing\nsampling-based systems.\n"}
{"abstract": "  The problem of dynamic prediction with time-dependent covariates, given by\nbiomarkers, repeatedly measured over time, has received much attention over the\nlast decades. Two contrasting approaches have become in widespread use. The\nfirst is joint modelling, which attempts to jointly model the longitudinal\nmarkers and the event time. The second is landmarking, a more pragmatic\napproach that avoids modelling the marker process. Landmarking has been shown\nto be less efficient than correctly specified joint models in simulation\nstudies, when data are generated from the joint model. When the mean model is\nmisspecified, however, simulation has shown that joint models may be inferior\nto landmarking.\n  The objective of this paper is to develop methods that improve the predictive\naccuracy of landmarking, while retaining its relative simplicity and\nrobustness. We start by fitting a working longitudinal model for the biomarker,\nincluding a temporal correlation structure. Based on that model, we derive a\npredictable time-dependent process representing the expected value of the\nbiomarker after the landmark time, and we fit a time-dependent Cox model based\non the predictable time-dependent covariate. Dynamic predictions based on this\napproach for new patients can be obtained by first deriving the expected values\nof the biomarker, given the measured values before the landmark time point, and\nthen calculating the predicted probabilities based on the time-dependent Cox\nmodel.\n  We illustrate the approach in predicting overall survival in liver cirrhosis\npatients based on prothrombin index.\n"}
{"abstract": "  Imitation learning (IL) algorithms use expert demonstrations to learn a\nspecific task. Most of the existing approaches assume that all expert\ndemonstrations are reliable and trustworthy, but what if there exist some\nadversarial demonstrations among the given data-set? This may result in poor\ndecision-making performance. We propose a novel general frame-work to directly\ngenerate a policy from demonstrations that autonomously detect the adversarial\ndemonstrations and exclude them from the data set. At the same time, it's\nsample, time-efficient, and does not require a simulator. To model such\nadversarial demonstration we propose a min-max problem that leverages the\nentropy of the model to assign weights for each demonstration. This allows us\nto learn the behavior using only the correct demonstrations or a mixture of\ncorrect demonstrations.\n"}
{"abstract": "  We consider the problem of designing video games (modeled here by choosing\nthe structure of a linear program solved by players) so that players with\ndifferent resources play diverse strategies. In particular, game designers hope\nto avoid scenarios where players use the same ``weapons'' or ``tactics'' even\nas they progress through the game. We model this design question as a choice\nover the constraint matrix $A$ and cost vector $c$ that seeks to maximize the\nnumber of possible supports of unique optimal solutions (what we call loadouts)\nof Linear Programs $\\max\\{c^\\top x \\mid Ax \\le b, x \\ge 0\\}$ with nonnegative\ndata considered over all resource vectors $b$. We provide an upper bound on the\noptimal number of loadouts and provide a family of constructions that have an\nasymptotically optimal number of loadouts. The upper bound is based on a\nconnection between our problem and the study of triangulations of point sets\narising from polyhedral combinatorics, and specifically the combinatorics of\nthe cyclic polytope. Our asymptotically optimal construction also draws\ninspiration from the properties of the cyclic polytope. Our construction\nprovides practical guidance to game designers seeking to offer a diversity of\nplay for their plays.\n"}
{"abstract": "  We investigate further the properties of axially symmetrical\nmonopole-antimonopole pair in the standard Weinberg-Salam theory. These\nnumerical solutions behave quite differently from their counterparts in the\nSU(2) Yang-Mills-Higgs theory as the poles are now bounded by a flux string.\nOur results confirm that the energy of these solutions resides in a range of\n13.17 - 21.02 TeV. In addition, we found strong evidence suggesting this\nconfiguration is time-dependent and discovered an energy oscillation phenomenon\nbetween the monopole and anti-monopole, which is associated with the symmetry\nof the solution. Finally, we calculate numerically the magnetic charge of the\nsolution and confirm that its value is indeed\n$\\frac{4\\pi}{e}\\sin^2\\theta_{\\scalebox{.5} {W}}$, as predicted by Y. Nambu in\nthe 1970s. Counterintuitively, the magnetic charge is spread out and\ndistributed along the string instead of concentrated near the poles.\n"}
{"abstract": "  Ultra-short period (USP) planets are an enigmatic subset of exoplanets\ndefined by having orbital periods $<$ 1 day. It is still not understood how USP\nplanets form, or to what degree they differ from planets with longer orbital\nperiods. Most USP planets have radii $<$ 2 $R_{\\oplus}$, while planets that\norbit further from their star extend to Jupiter size ($>$ 10 $R_{\\oplus}$).\nSeveral theories attempt to explain the formation and composition of USP\nplanets: they could be remnant cores of larger gas giants that lost their\natmospheres due to photo-evaporation or Roche lobe overflow, or they could have\ndeveloped through mass accretion in the innermost part of the protoplanetary\ndisk. The radius and mass distribution of USP planets could provide important\nclues to distinguish between potential formation mechanisms. In this study, we\nfirst verify and update the Kepler catalog of USP planet host star properties,\nincorporating new data collected by the Gaia mission where applicable. We then\nuse the transit depths measured by Kepler to derive a radius distribution and\npresent occurrence rates for USP planets. Using spherical and tidally distorted\nplanet models, we then derive a mass distribution for USP planets. Comparisons\nbetween the updated USP planet mass distribution and simulated planetary\nsystems offer further insights into the formation and evolutionary processes\nshaping USP planet populations.\n"}
{"abstract": "  We illustrate with several new applications the power and elegance of the\nBendixson Dulac theorem to obtain upper bounds of the number of limit cycles\nfor several families of planar vector fields. In some cases we propose to use a\nfunction related with the curvature of the orbits of the vector field as a\nDulac function. We get some general results for Lienard type equations and for\nrigid planar systems. We also present a remarkable phenomenon: for each integer\nm greater than one, we provide a simple one parametric differential system for\nwhich we prove that it has limit cycles only for the values of the parameter in\na subset of an interval that decreases exponentially when m grows. One of the\nstrengths of the results presented in this work is that although they are\nobtained with simple calculations, that can be easily checked by hand, they\nimprove and extend previous studies. Another one is that, for certain systems,\nit is possible to reduce the question of the number of limit cycles to the\nstudy of the shape of a planar curve and the sign of an associated function in\none or two variables.\n"}
{"abstract": "  In the interaction between control and mathematics, mathematical tools are\nfundamental for all the control methods, but it is unclear how control impacts\nmathematics. This is the first part of our paper that attempts to give an\nanswer with focus on solving linear algebraic equations (LAEs) from the\nperspective of systems and control, where it mainly introduces the\ncontrollability-based design results. By proposing an iterative method that\nintegrates a learning control mechanism, a class of tracking problems for\niterative learning control (ILC) is explored for the problem solving of LAEs. A\ntrackability property of ILC is newly developed, by which analysis and\nsynthesis results are established to disclose the equivalence between the\nsolvability of LAEs and the controllability of discrete control systems. Hence,\nLAEs can be solved by equivalently achieving the perfect tracking tasks of\nresulting ILC systems via the classic state feedback-based design and analysis\nmethods. It is shown that the solutions for any solvable LAE can all be\ncalculated with different selections of the initial input. Moreover, the\npresented ILC method is applicable to determining all the least squares\nsolutions of any unsolvable LAE. In particular, a deadbeat design is\nincorporated to ILC such that the solving of LAEs can be completed within\nfinite iteration steps. The trackability property is also generalized to\nconventional two-dimensional ILC systems, which creates feedback-based methods,\ninstead of the common used contraction mapping-based methods, for the design\nand convergence analysis of ILC.\n"}
{"abstract": "  The presented report evaluates Contextualizing Hate Speech Classifiers with\nPost-hoc Explanation paper within the scope of ML Reproducibility Challenge\n2020. Our work focuses on both aspects constituting the paper: the method\nitself and the validity of the stated results. In the following sections, we\nhave described the paper, related works, algorithmic frameworks, our\nexperiments and evaluations.\n"}
{"abstract": "  Despite their outstanding accuracy, semi-supervised segmentation methods\nbased on deep neural networks can still yield predictions that are considered\nanatomically impossible by clinicians, for instance, containing holes or\ndisconnected regions. To solve this problem, we present a Context-aware Virtual\nAdversarial Training (CaVAT) method for generating anatomically plausible\nsegmentation. Unlike approaches focusing solely on accuracy, our method also\nconsiders complex topological constraints like connectivity which cannot be\neasily modeled in a differentiable loss function. We use adversarial training\nto generate examples violating the constraints, so the network can learn to\navoid making such incorrect predictions on new examples, and employ the\nReinforce algorithm to handle non-differentiable segmentation constraints. The\nproposed method offers a generic and efficient way to add any constraint on top\nof any segmentation network. Experiments on two clinically-relevant datasets\nshow our method to produce segmentations that are both accurate and\nanatomically-plausible in terms of region connectivity.\n"}
{"abstract": "  In this work, we propose an ensemble forecasting approach based on randomized\nneural networks. Improved randomized learning streamlines the fitting abilities\nof individual learners by generating network parameters in accordance with the\ndata and target function features. A pattern-based representation of time\nseries makes the proposed approach suitable for forecasting time series with\nmultiple seasonality. We propose six strategies for controlling the diversity\nof ensemble members. Case studies conducted on four real-world forecasting\nproblems verified the effectiveness and superior performance of the proposed\nensemble forecasting approach. It outperformed statistical models as well as\nstate-of-the-art machine learning models in terms of forecasting accuracy. The\nproposed approach has several advantages: fast and easy training, simple\narchitecture, ease of implementation, high accuracy and the ability to deal\nwith nonstationarity and multiple seasonality in time series.\n"}
{"abstract": "  When the transition temperature of a continuous phase transition is tuned to\nabsolute zero, new ordered phases and physical behaviour emerge in the vicinity\nof the resulting quantum critical point. Sr3Ru2O7 can be tuned through quantum\ncriticality with magnetic field at low temperature. Near its critical field Bc\nit displays the hallmark T-linear resistivity and a T log(1/T) electronic heat\ncapacity behaviour of strange metals. However, these behaviours have not been\nrelated to any critical fluctuations. Here we use inelastic neutron scattering\nto reveal the presence of collective spin fluctuations whose relaxation time\nand strength show a nearly singular variation with magnetic field as Bc is\napproached. The large increase in the electronic heat capacity and entropy near\nBc can be understood quantitatively in terms of the scattering of conduction\nelectrons by these spin-fluctuations. On entering the spin density wave (SDW)\nphase present near Bc, the fluctuations become stronger suggesting that the SDW\norder is stabilised through an \"order-by-disorder\" mechanism.\n"}
{"abstract": "  We study the large time behavior of small data solutions to the\nVlasov-Navier-Stokes system set on $\\Omega \\times \\mathbb{R}^3$, for a smooth\nbounded domain $\\Omega$ of $\\mathbb{R}^3$, with homogeneous Dirichlet boundary\ncondition for the fluid and absorption boundary condition for the kinetic\nphase. We prove that the fluid velocity homogenizes to $0$ while the\ndistribution function concentrates towards a Dirac mass in velocity centered at\n$0$, with an exponential rate. The proof, which follows the methods introduced\nin [Han-Kwan - Moussa - Moyano, arXiv:1902.03864v2], requires a careful\nanalysis of the boundary effects. We also exhibit examples of classes of\ninitial data leading to a variety of asymptotic behaviors for the kinetic\ndensity, from total absorption to no absorption at all.\n"}
{"abstract": "  The nature of the three narrow $P_c$ states, i.e., $P_c(4312)$, $P_c(4440)$\nand $P_c(4457)$, is under intense discussion since their discovery from the\nupdated analysis by LHCb. In this work we extend our previous coupled-channel\napproach [Phys. Rev. Lett. \\bf{124}, 072001 (2020)] by including the\n$\\Lambda_c\\bar{D}^{(*)}$ and $\\eta_cp$ as explicitly in addition to the $J/\\psi\np$, as required by unitarity and heavy quark spin symmetry (HQSS). Since\ninelastic parameters are very badly constrained by the current data, three\nschemes are considered: (a) scheme I with pure contact interactions and without\nthe $\\Lambda_c\\bar D^{(*)}$ interactions, (b) scheme II, where the one-pion\nexchange is added to scheme I, and (c) scheme III, where the $\\Lambda_c \\bar\nD^{(*)}$ are included in addition. It is shown that to obtain cutoff\nindependent results, OPE in the multichannel system is to be supplemented with\n$S$-$D$ mixing contact terms. We demonstrate that the experimental data for the\n$J/\\psi p$ invariant mass distribution are consistent with the interpretation\nof the $P_c(4312)$ and $P_c(4440)/P_c(4457)$ as $\\Sigma_c\\bar{D}$ and $\\Sigma_c\n\\bar{D}^{*}$ hadronic molecules, respectively, and that the data show clear\nevidence for a new narrow $P_c(4380)$, as a $\\Sigma_c^*\\bar D$ molecule, which\nshould exist as a consequence of HQSS. While two equally good solutions are\nfound in scheme I, only one of these solutions with the quantum numbers of the\n$P_c(4440)$ and $P_c(4457)$ being $J^P=3/2^-$ and $1/2^-$, respectively,\nsurvives the requirement of regulator independence once the OPE is included.\nMoreover, we predict the line shapes in the elastic and inelastic channels and\ndemonstrate that those related to the $P_c(4440)$ and the $P_c(4457)$ in the\ncorresponding $\\Sigma_c^{(*)}\\bar{D}$ and $\\eta_cp$ mass distributions allow\none to confirm the quantum numbers given above, once the data are available.\n"}
{"abstract": "  In this work we introduce a concept of expansiveness for actions of connected\nLie groups. We study some of its properties and investigate some implications\nof expansiveness. We study the centralizer of expansive actions and introduce\nCW-expansiveness for pseudo-group actions. As an application, we prove\npositiveness of geometric entropy for expansive foliations and expansive group\nactions.\n"}
{"abstract": "  The growth of meromorphic solutions of linear difference equations containing\nAskey-Wilson divided difference operators is estimated. The $\\varphi$-order is\nused as a general growth indicator, which covers the growth spectrum between\nthe logarithmic order $\\rho_{\\log}(f)$ and the classical order $\\rho(f)$ of a\nmeromorphic function $f$.\n"}
{"abstract": "  The two-parameter Poisson--Dirichlet diffusion takes values in the infinite\nordered simplex and has a two-parameter Poisson-Dirichlet stationary\ndistribution ${\\cal PD}(\\alpha, \\theta)$ . The dual process is in terms of a\nline-of-descent process which tracks the evolution of non-mutant frequencies\nfrom time zero, and Remarkably does not depend on the parameter $\\alpha$.\nMethods of proof use the sampling distributions associated to the two-parameter\nPoisson-Dirichlet process. We also exploit a connection between the sampling\ndistribution and a generalized Blackwell and MacQueen P{\\'o}lya urn model to\nobtain a direct derivation of the transition density.\n"}
{"abstract": "  We report on the controlled switching of domain wall (DW) magnetization in\naligned stripe domain structures, stabilized in [Co (0.44 nm)/Pt (0.7 nm)]$_X$\n($X = 48$, 100, 150) multilayers with perpendicular magnetic anisotropy. The\nswitching process, induced by an external magnetic field, is monitored by\nmeasuring the evolution of the in-plane magnetization. % We show that the\nremanent in-plane magnetization originates from the polarization of the\nBloch-type DWs. With micromagnetic simulations, we reveal that the reversal of\nthe DW polarization is the result of the emergence and collapse of horizontal\nBloch lines within the DWs at particular strengths of the external magnetic\nfield, applied opposite to the DW polarization. Our findings are relevant for\nDW-based magnonics and bubble skyrmion applications in magnetic multilayers.\n"}
{"abstract": "  Macular holes are a common eye condition which result in visual impairment.\nWe look at the application of deep convolutional neural networks to the problem\nof macular hole segmentation. We use the 3D U-Net architecture as a basis and\nexperiment with a number of design variants. Manually annotating and measuring\nmacular holes is time consuming and error prone. Previous automated approaches\nto macular hole segmentation take minutes to segment a single 3D scan. Our\nproposed model generates significantly more accurate segmentations in less than\na second. We found that an approach of architectural simplification, by greatly\nsimplifying the network capacity and depth, exceeds both expert performance and\nstate-of-the-art models such as residual 3D U-Nets.\n"}
{"abstract": "  Many effective field theories describing gravity cannot arise from an\nunderlying theory based on Riemann geometry or its extensions to include\ntorsion and nonmetricity but may instead emerge from another geometry or may\nhave a nongeometric basis. The Lagrange density for a broad class of such\ntheories is investigated. The action for fermions coupled to gravity is\nlinearized about a Minkowski background and is found to include terms\ndescribing small deviations from Lorentz invariance and gravitational gauge\ninvariance. The corresponding nonrelativistic hamiltonian is derived at second\norder in the fermion momentum. The implications for laboratory experiments and\nastrophysical observations with fermions are studied, with primary focus on\nanomalous spin-gravity couplings. First constraints on some coefficients are\nextracted from existing data obtained via measurements at different potentials,\ncomparisons of gravitational accelerations, interferometric methods, and\ninvestigations of gravitational bound states. Some prospects for future\nexperimental studies are discussed.\n"}
{"abstract": "  We discuss the model robustness of the infinite two-dimensional square grid\nwith respect to symmetry breakings due to the presence of defects, that is,\nlacks of finitely or infinitely many edges. Precisely, we study how these\ntopological perturbations of the square grid affect the so-called dimensional\ncrossover identified in [Adami et al. 2019]. Such a phenomenon has two\nevidences: the coexistence of the one and the two-dimensional Sobolev\ninequalities and the appearence of a continuum of $L^2$-critical exponents for\nthe ground states at fixed mass of the nonlinear Schr\\\"odinger equation. From\nthis twofold perspective, we investigate which classes of defects do preserve\nthe dimensional crossover and which classes do not.\n"}
{"abstract": "  Enumerating integral orbits in prehomogeneous vector spaces plays an\nimportant role in arithmetic statistics. We describe a method of proving\nsubconvexity of the zeta function enumerating the integral orbits, illustrated\nby proving a subconvex estimate for the Shintani $\\zeta$ function enumerating\nclass numbers of binary cubic forms. The method also obtains a subconvex\nestimate for the zeta function twisted by a Maass cusp form.\n"}
{"abstract": "  A widely used tool for analysing the Covid-19 pandemic is the standard SIR\nmodel. It seems often to be used as a black box, not taking into account that\nthis model was derived as a special case of the seminal Kermack-McKendrick\ntheory from 1927. This is our starting point. We explain the setup of the\nKermack-McKendrick theory (passing to a discrete approach) and use medical\ninformation for specializing to a model which we call {\\em adapted\nK-McK-model}. This includes effects of vaccination, mass testing and mutants.\nWe demonstrate the use of the model by applying it to the development in\nGermany. As a striking application we demonstrate that a comparatively mild\nintervention reducing the time until quarantine by one day leads to a drastic\nimprovement. A similar effect can be obtained by certain mass testings as we\nwill demonstrate. We discuss possibilities to apply the model both for\npredictions and as an analysis tool. We compare the adapted K-McK-model to the\nstandard SIR model and observe condiderable differences if the contact rates\nare not constant. Finally we compare the model reproduction rate with the\nempirical reproduction rate determined by the Robert Koch-Institut.\n"}
{"abstract": "  By a combined study with first-principles calculations and symmetry analysis,\nwe theoretically investigate the electronic properties of monolayer\nMoSi$_2$N$_4$. While the spin-orbital coupling results in bands splitting, the\nhorizontal mirror symmetry locks the spin polarization along z-direction. In\naddition, a three-band tight-binding model is constructed to describe the\nlow-energy quasi-particle states of monolayer MoSi$_2$N$_4$, which can be\ngeneralized to strained MoSi$_2$N$_4$ and its derivatives. The calculations\nusing the tight-binding model show an undamped $\\sqrt{q}$-dependent plasmon\nmode that agrees well with the results of first-principles calculations. Our\nmodel can be extended to be suitable for future theoretical and numerical\nstudies of low-energy properties in MoSi$_2$N$_4$ family materials.\nFurthermore, the study of electronic properties of monolayer MoSi$_2$N$_4$\npaves a way for its applications in spintronics and plasmonics.\n"}
{"abstract": "  Stars in the mass range from 8 to 10 solar masses are expected to produce one\nof two types of supernovae (SNe), either electron-capture supernovae (ECSNe) or\ncore-collapse supernovae (CCSNe), depending on their previous evolution. Either\nof the associated progenitors retain extended and massive hydrogen-rich\nenvelopes, the observables of these SNe are, therefore, expected to be similar.\nIn this study we explore the differences in these two types of SNe.\nSpecifically, we investigate three different progenitor models: a\nsolar-metallicity ECSN progenitor with an initial mass of 8.8 solar masses, a\nzero-metallicity progenitor with 9.6 solar masses, and a solar-metallicity\nprogenitor with 9 solar masses, carrying out radiative transfer simulations for\nthese progenitors. We present the resulting light curves for these models. The\nmodels exhibit very low photospheric velocity variations of about 2000 km/s,\ntherefore, this may serve as a convenient indicator of low-mass SNe. The ECSN\nhas very unique light curves in broad bands, especially the U band, and does\nnot resemble any currently observed SN. This ECSN progenitor being part of a\nbinary will lose its envelope for which reason the light curve becomes short\nand undetectable. The SN from the 9.6 solar masses progenitor exhibits also\nquite an unusual light curve, explained by the absence of metals in the initial\ncomposition. The artificially iron polluted 9.6 solar masses model demonstrates\nlight curves closer to normal SNe IIP. The SN from the 9 solar masses\nprogenitor remains the best candidate for so-called low-luminosity SNe IIP like\nSN 1999br and SN 2005cs.\n"}
{"abstract": "  In this paper, we study communication efficient distributed algorithms for\ndistributionally robust federated learning via periodic averaging with adaptive\nsampling. In contrast to standard empirical risk minimization, due to the\nminimax structure of the underlying optimization problem, a key difficulty\narises from the fact that the global parameter that controls the mixture of\nlocal losses can only be updated infrequently on the global stage. To\ncompensate for this, we propose a Distributionally Robust Federated Averaging\n(DRFA) algorithm that employs a novel snapshotting scheme to approximate the\naccumulation of history gradients of the mixing parameter. We analyze the\nconvergence rate of DRFA in both convex-linear and nonconvex-linear settings.\nWe also generalize the proposed idea to objectives with regularization on the\nmixture parameter and propose a proximal variant, dubbed as DRFA-Prox, with\nprovable convergence rates. We also analyze an alternative optimization method\nfor regularized cases in strongly-convex-strongly-concave and non-convex (under\nPL condition)-strongly-concave settings. To the best of our knowledge, this\npaper is the first to solve distributionally robust federated learning with\nreduced communication, and to analyze the efficiency of local descent methods\non distributed minimax problems. We give corroborating experimental evidence\nfor our theoretical results in federated learning settings.\n"}
{"abstract": "  We model the dynamics of the cryptocurrency (CC) asset class via a stochastic\nvolatility with correlated jumps (SVCJ) model with rolling-window parameter\nestimates. By analyzing the time-series of parameters, stylized patterns are\nobservable which are robust to changes of the window size and supported by\ncluster analysis. During bullish periods, volatility stabilizes at low levels\nand the size and volatility of jumps in mean decreases. In bearish periods\nthough, volatility increases and takes longer to return to its long-run trend.\nFurthermore, jumps in mean and jumps in volatility are independent. With the\nrise of the CC market in 2017, a level shift of the volatility of volatility\noccurred. All codes are available on Quantlet.com.\n"}
{"abstract": "  Quantum emitters (QEs) in two-dimensional transition metal dichalcogenides\n(2D TMDCs) have advanced to the forefront of quantum communication and\ntransduction research due to their unique potentials in accessing valley\npseudo-spin degree of freedom (DOF) and facile integration into\nquantum-photonic, electronic and sensing platforms via the\nlayer-by-layer-assembly approach. To date, QEs capable of operating in O-C\ntelecommunication bands have not been demonstrated in TMDCs. Here we report a\ndeterministic creation of such telecom QEs emitting over the 1080 to 1550 nm\nwavelength range via coupling of 2D molybdenum ditelluride (MoTe2) to strain\ninducing nano-pillar arrays. Our Hanbury Brown and Twiss experiment conducted\nat 10 K reveals clear photon antibunching with 90% single photon purity.\nUltra-long lifetimes, 4-6 orders of magnitude longer than that of the 2D\nexciton, are also observed. Polarization analysis further reveals that while\nsome QEs display cross-linearly polarized doublets with ~1 meV splitting\nresulting from the strain induced anisotropic exchange interaction, valley\ndegeneracy is preserved in other QEs. Valley Zeeman splitting as well as\nrestoring of valley symmetry in cross-polarized doublets are observed under 8T\nmagnetic field. In contrast to other telecom QEs, our QEs which offer the\npotential to access valley DOF through single photons, could lead to\nunprecedented advantages in optical fiber-based quantum networks.\n"}
{"abstract": "  In the presence of both space and time reversal symmetries, an s-wave A1g\nsuperconducting state is usually topologically trivial. Here we demonstrate\nthat an exception can take place in a type of nonsymmorphic lattice structures.\nWe specify the demonstration in a system with a centrosymmetric space group\nP4/nmm, the symmetry group that governs iron-based superconductors, by showing\nthe existence of a second-order topological state protected by a mirror\nsymmetry. The topological superconductivity is featured by 2Z degenerate Dirac\ncones on the (1,0) edge, and Z pairs of Majorana modes at the intersection\nbetween the (1,1) and (1,-1) edges. The topological invariance and Fermi\nsurface criterion for the topological state are provided. Moreover, we point\nout that the previously proposed s-wave state in iron-based superconductors,\nwhich features a sign-changed superconducting order parameter between two\nelectron pockets, is such a topological state. Thus, these results not only\nopen a new route to pursue topological superconductivity, but also establish a\nmeasurable quantity to settle one long-lasting debate on the pairing nature of\niron-based superconductors.\n"}
{"abstract": "  Automated audio captioning (AAC) aims at generating summarizing descriptions\nfor audio clips. Multitudinous concepts are described in an audio caption,\nranging from local information such as sound events to global information like\nacoustic scenery. Currently, the mainstream paradigm for AAC is the end-to-end\nencoder-decoder architecture, expecting the encoder to learn all levels of\nconcepts embedded in the audio automatically. This paper first proposes a topic\nmodel for audio descriptions, comprehensively analyzing the hierarchical audio\ntopics that are commonly covered. We then explore a transfer learning scheme to\naccess local and global information. Two source tasks are identified to\nrespectively represent local and global information, being Audio Tagging (AT)\nand Acoustic Scene Classification (ASC). Experiments are conducted on the AAC\nbenchmark dataset Clotho and Audiocaps, amounting to a vast increase in all\neight metrics with topic transfer learning. Further, it is discovered that\nlocal information and abstract representation learning are more crucial to AAC\nthan global information and temporal relationship learning.\n"}
{"abstract": "  A long-standing question in the theory of measures of noncompactness is that\nfor the Kuratowski measure of noncompactness $\\alpha$ defined on a metric space\n$M$, and for every bounded subset $B\\subset M$, is there a countable subset\n$B_0\\subset B$ such that $\\alpha(B_0)=\\alpha(B)$? In this paper, we give an\naffirmative answer to the question above. It is done by showing that for each\nnonempty set $B$ of a Banach space, there is a countable subset $B_0\\subset B$\nso that $B$ is strongly finitely representable in $B_0$, and that there is a\nfree ultrafilter $\\mathcal U$ so that $B$ is affinely isometric to a subset of\nthe ultrapower $[{\\rm co}(B_0)]_\\mathcal U$ of ${\\rm co}(B_0)$.\n"}
{"abstract": "  The main contributions of this paper are the proposition and the convergence\nanalysis of a class of inertial projection-type algorithm for solving\nvariational inequality problems in real Hilbert spaces where the underline\noperator is monotone and uniformly continuous. We carry out a unified analysis\nof the proposed method under very mild assumptions. In particular, weak\nconvergence of the generated sequence is established and nonasymptotic $O(1/n)$\nrate of convergence is established, where $n$ denotes the iteration counter. We\nalso present some experimental results to illustrate the profits gained by\nintroducing the inertial extrapolation steps.\n"}
{"abstract": "  Recent works in medical image segmentation have actively explored various\ndeep learning architectures or objective functions to encode high-level\nfeatures from volumetric data owing to limited image annotations. However, most\nexisting approaches tend to ignore cross-volume global context and define\ncontext relations in the decision space. In this work, we propose a novel\nvoxel-level Siamese representation learning method for abdominal multi-organ\nsegmentation to improve representation space. The proposed method enforces\nvoxel-wise feature relations in the representation space for leveraging limited\ndatasets more comprehensively to achieve better performance. Inspired by recent\nprogress in contrastive learning, we suppressed voxel-wise relations from the\nsame class to be projected to the same point without using negative samples.\nMoreover, we introduce a multi-resolution context aggregation method that\naggregates features from multiple hidden layers, which encodes both the global\nand local contexts for segmentation. Our experiments on the multi-organ dataset\noutperformed the existing approaches by 2% in Dice score coefficient. The\nqualitative visualizations of the representation spaces demonstrate that the\nimprovements were gained primarily by a disentangled feature space.\n"}
{"abstract": "  What makes an artificial neural network easier to train and more likely to\nproduce desirable solutions than other comparable networks? In this paper, we\nprovide a new angle to study such issues under the setting of a fixed number of\nmodel parameters which in general is the most dominant cost factor. We\nintroduce a notion of variability and show that it correlates positively to the\nactivation ratio and negatively to a phenomenon called {Collapse to Constants}\n(or C2C), which is closely related but not identical to the phenomenon commonly\nknown as vanishing gradient. Experiments on a styled model problem empirically\nverify that variability is indeed a key performance indicator for fully\nconnected neural networks. The insights gained from this variability study will\nhelp the design of new and effective neural network architectures.\n"}
{"abstract": "  This paper describes the use of Distributed Ledger Technologies as a mean to\nenforce social contracts and to orchestrate the behaviour of agents in a smart\ncity environment. Specifically, we present a scheme to price personalised risk\nin sharing economy applications. We provide proofs for the convergence of the\nproposed stochastic system and we validate our approach through the use of\nextensive Monte Carlo simulations.\n"}
{"abstract": "  Ammonia (NH3) is commonly used as group V precursor in gallium nitride (GaN)\nmetalorganic chemical vapor deposition (MOCVD). The high background carbon (C)\nimpurity in MOCVD GaN is related to the low pyrolysis efficiency of NH3, which\nrepresents one of the fundamental challenges hindering the development of high\npurity thick GaN for vertical high power device applications. This work uses a\nlaser-assisted MOCVD (LA-MOCVD) growth technique to address the high-C issue in\nMOCVD GaN. Carbon dioxide (CO2) laser with wavelength of 9.219 um was utilized\nto facilitate NH3 decomposition via resonant vibrational excitation. The\nLA-MOCVD GaN growth rate (as high as 10 um/hr) shows a strong linear\nrelationship with the trimethylgallium (TMGa) flow rate, indicating high\neffective V/III ratios and hence efficient NH3 decomposition. Pits-free surface\nmorphology of LA-MOCVD GaN was demonstrated for films with growth rate as high\nas 8.5 um/hr. The background [C] in LA-MOCVD GaN films decreases monotonically\nas the laser power increases. A low [C] at 5.5E15 cm-3 was achieved in LA-MOCVD\nGaN film grown with the growth rate of 4 um/hr. Charge transport\ncharacterization of LA-MOCVD GaN films reveals high crystalline quality with\nroom temperature mobility >1000 cm2/Vs. LA-MOCVD growth technique provides an\nenabling route to achieve high quality GaN epitaxy with low-C impurity and fast\ngrowth rate simultaneously. This technique can also be extended for epitaxy of\nother nitride-based semiconductors.\n"}
{"abstract": "  In recent years, compressed sensing (CS) based image coding has become a hot\ntopic in image processing field. However, since the bit depth required for\nencoding each CS sample is too large, the compression performance of this\nparadigm is unattractive. To address this issue, a novel CS-based image coding\nsystem by using gray transformation is proposed. In the proposed system, we use\na gray transformation to preprocess the original image firstly and then use CS\nto sample the transformed image. Since gray transformation makes the\nprobability distribution of CS samples centralized, the bit depth required for\nencoding each CS sample is reduced significantly. Consequently, the proposed\nsystem can considerably improve the compression performance of CS-based image\ncoding. Simulation results show that the proposed system outperforms the\ntraditional one without using gray transformation in terms of compression\nperformance.\n"}
{"abstract": "  Human teams exhibit both implicit and explicit intention sharing. To further\ndevelopment of human-robot collaboration, intention recognition is crucial on\nboth sides. Present approaches rely on a vast sensor suite on and around the\nrobot to achieve intention recognition. This relegates intuitive human-robot\ncollaboration purely to such bulky systems, which are inadequate for\nlarge-scale, real-world scenarios due to their complexity and cost. In this\npaper we propose an intention recognition system that is based purely on a\nportable head-mounted display. In addition robot intention visualisation is\nalso supported. We present experiments to show the quality of our human goal\nestimation component and some basic interactions with an industrial robot. HAIR\nshould raise the quality of interaction between robots and humans, instead of\nsuch interactions raising the hair on the necks of the human coworkers.\n"}
{"abstract": "  There has been a recent surge in adversarial attacks on deep learning based\nautomatic speech recognition (ASR) systems. These attacks pose new challenges\nto deep learning security and have raised significant concerns in deploying ASR\nsystems in safety-critical applications. In this work, we introduce WaveGuard:\na framework for detecting adversarial inputs that are crafted to attack ASR\nsystems. Our framework incorporates audio transformation functions and analyses\nthe ASR transcriptions of the original and transformed audio to detect\nadversarial inputs. We demonstrate that our defense framework is able to\nreliably detect adversarial examples constructed by four recent audio\nadversarial attacks, with a variety of audio transformation functions. With\ncareful regard for best practices in defense evaluations, we analyze our\nproposed defense and its strength to withstand adaptive and robust attacks in\nthe audio domain. We empirically demonstrate that audio transformations that\nrecover audio from perceptually informed representations can lead to a strong\ndefense that is robust against an adaptive adversary even in a complete\nwhite-box setting. Furthermore, WaveGuard can be used out-of-the box and\nintegrated directly with any ASR model to efficiently detect audio adversarial\nexamples, without the need for model retraining.\n"}
{"abstract": "  In this study we investigate the nuclear quantum effects (NQEs) on the\nacidity constant (pKA) of liquid water isotopologues at the ambient condition\nby path integral molecular dynamics (PIMD) simulations. We compared simulations\nusing a fully explicit solvent model with a classical polarizable force field,\ndensity functional tight binding, and ab initio density functional theory,\nwhich correspond to empirical, semiempirical, and ab initio PIMD simulations,\nrespectively. The centroid variable with respect to the proton coordination\nnumber of a water molecule was restrained to compute the gradient of the free\nenergy, which measures the reversible work of the proton abstraction for the\nquantum mechanical system. The free energy curve obtained by thermodynamic\nintegration was used to compute the pKA value based on probabilistic\ndetermination. This technique not only reproduces the pKA value of liquid D2O\nexperimentally measured (14.86) but also allows for a theoretical prediction of\nthe pKA values of liquid T2O, aqueous HDO and HTO which are unknown due to its\nscarcity. It is also shown that the NQEs on the free energy curve can result in\na downshift of 4.5 +/- 0.9 pKA units in the case of liquid water, which\nindicates that the NQEs plays an indispensable role in the absolute\ndetermination of pKA. The results of this study can help to inform further\nextensions into the calculation of the acidity constants of isotope substituted\nspecies with high accuracy.\n"}
{"abstract": "  The FitzHugh-Rinzel system is able to describe some biophysical phenomena,\nsuch as bursting oscillations, and the study of its solutions can help to\nbetter understand several behaviours of the complex dynamics of biological\nsystems. We express the solutions by means of an integral equation involving\nthe fundamental solution $ H(x,t) $ related to a non linear\nintegro-differential equation. Properties of $ H(x,t) $ allow us to obtain a\npriori estimates for solutions determined in the whole space, showing both the\ninfluence of the initial data and the source term.\n"}
{"abstract": "  We study the position distribution $P(\\vec{R},N)$ of a run-and-tumble\nparticle (RTP) in arbitrary dimension $d$, after $N$ runs. We assume that the\nconstant speed $v>0$ of the particle during each running phase is independently\ndrawn from a probability distribution $W(v)$ and that the direction of the\nparticle is chosen isotropically after each tumbling. The position distribution\nis clearly isotropic, $P(\\vec{R},N)\\to P(R,N)$ where $R=|\\vec{R}|$. We show\nthat, under certain conditions on $d$ and $W(v)$ and for large $N$, a\ncondensation transition occurs at some critical value of $R=R_c\\sim O(N)$\nlocated in the large deviation regime of $P(R,N)$. For $R<R_c$ (subcritical\nfluid phase), all runs are roughly of the same size in a typical trajectory. In\ncontrast, an RTP trajectory with $R>R_c$ is typically dominated by a\n`condensate', i.e., a large single run that subsumes a finite fraction of the\ntotal displacement (supercritical condensed phase). Focusing on the family of\nspeed distributions $W(v)=\\alpha(1-v/v_0)^{\\alpha-1}/v_0$, parametrized by\n$\\alpha>0$, we show that, for large $N$, $P(R,N)\\sim\n\\exp\\left[-N\\psi_{d,\\alpha}(R/N)\\right]$ and we compute exactly the rate\nfunction $\\psi_{d,\\alpha}(z)$ for any $d$ and $\\alpha$. We show that the\ntransition manifests itself as a singularity of this rate function at $R=R_c$\nand that its order depends continuously on $d$ and $\\alpha$. We also compute\nthe distribution of the condensate size for $R>R_c$. Finally, we study the\nmodel when the total duration $T$ of the RTP, instead of the total number of\nruns, is fixed. Our analytical predictions are confirmed by numerical\nsimulations, performed using a constrained Markov chain Monte Carlo technique,\nwith precision $\\sim 10^{-100}$.\n"}
{"abstract": "  The fractional Brownian motion (fBm) extends the standard Brownian motion by\nintroducing some dependence between non-overlapping increments. Consequently,\nif one considers for example that log-prices follow an fBm, one can exploit the\nnon-Markovian nature of the fBm to forecast future states of the process and\nmake statistical arbitrages. We provide new insights into forecasting an fBm,\nby proposing theoretical formulas for accuracy metrics relevant to a systematic\ntrader, from the hit ratio to the expected gain and risk of a simple strategy.\nIn addition, we answer some key questions about optimizing trading strategies\nin the fBm framework: Which lagged increments of the fBm, observed in discrete\ntime, are to be considered? If the predicted increment is close to zero, up to\nwhich threshold is it more profitable not to invest? We also propose empirical\napplications on high-frequency FX rates, as well as on realized volatility\nseries, exploring the rough volatility concept in a forecasting perspective.\n"}
{"abstract": "  This paper provides a unified stochastic operator framework to analyze the\nconvergence of iterative optimization algorithms for both static problems and\nonline optimization and learning. In particular, the framework is well suited\nfor algorithms that are implemented in an inexact or stochastic fashion because\nof (i) stochastic errors emerging in algorithmic steps, and because (ii) the\nalgorithm may feature random coordinate updates. To this end, the paper focuses\non separable operators of the form $T x = (T_1 x, \\ldots, T_n x)$, defined over\nthe direct sum of possibly infinite-dimensional Hilbert spaces, and\ninvestigates the convergence of the associated stochastic Banach-Picard\niteration. Results in terms of convergence in mean and in high-probability are\npresented when the errors affecting the operator follow a sub-Weibull\ndistribution and when updates $T_i x$ are performed based on a Bernoulli random\nvariable. In particular, the results are derived for the cases where $T$ is\ncontractive and averaged in terms of convergence to the unique fixed point and\ncumulative fixed-point residual, respectively. The results do not assume\nvanishing errors or vanishing parameters of the operator, as typical in the\nliterature (this case is subsumed by the proposed framework), and links with\nexiting results in terms of almost sure convergence are provided. In the online\noptimization context, the operator changes at each iteration to reflect changes\nin the underlying optimization problem. This leads to an online Banach-Picard\niteration, and similar results are derived where the bounds for the convergence\nin mean and high-probability further depend on the evolution of fixed points\n(i.e., optimal solutions of the time-varying optimization problem).\n"}
{"abstract": "  We construct a Leray model for a discrete polymatroid with arbitrary building\nset and we prove a generalized Goresky-MacPherson formula. The first row of the\nmodel is the Chow ring of the polymatroid; we prove Poincar\\'e duality, Hard\nLefschetz, and Hodge-Riemann theorems for the Chow ring. Furthermore we provide\na relative Lefschetz decomposition with respect to the deletion of an element.\n"}
{"abstract": "  Confinement of atoms inside various cavities has been studied for nearly\neight decades. However, the Compton profile for such systems has not yet been\ninvestigated. Here we construct the Compton profile (CP) for a H atom radially\nconfined inside a \\emph{hard} spherical enclosure, as well as in \\emph{free\ncondition}. Some exact analytical relations for the CP's of circular or\nnodeless states of free atom is presented. By means of a scaling idea, this has\nbeen further extended to the study of an H-like atom trapped inside an\nimpenetrable cavity. The accuracy of these constructed CP has been confirmed by\ncomputing various momentum moments. Apart from that, several information\ntheoretical measures, like Shannon entropy ($S$) and Onicescu energy ($E$) have\nbeen exploited to characterize these profiles. Exact closed form expressions\nare derived for $S$ and $E$ using the ground state CP in free H-like atoms. A\ndetailed study reveals that, increase in confinement inhibits the rate of\ndissipation of kinetic energy. At a fixed $\\ell$, this rate diminishes with\nrise in $n$. However, at a certain $n$, this rate accelerates with progress in\n$\\ell$. A similar analysis on the respective free counterpart displays an\nexactly opposite trend as that in confined system. However, in both free and\nconfined environments, CP generally gets broadened with rise in $Z$.\nRepresentative calculations are done numerically for low-lying states of the\nconfined systems, taking two forms of position-space wave functions: (a) exact\n(b) highly accurate eigenfunctions through a generalized pseudospectral method.\nIn essence, CPs are reported for confined H atom (and isoelectronic series) and\ninvestigated adopting an information-theoretic framework.\n"}
{"abstract": "  Capturing high-dimensional (HD) data is a long-term challenge in signal\nprocessing and related fields. Snapshot compressive imaging (SCI) uses a\ntwo-dimensional (2D) detector to capture HD ($\\ge3$D) data in a {\\em snapshot}\nmeasurement. Via novel optical designs, the 2D detector samples the HD data in\na {\\em compressive} manner; following this, algorithms are employed to\nreconstruct the desired HD data-cube. SCI has been used in hyperspectral\nimaging, video, holography, tomography, focal depth imaging, polarization\nimaging, microscopy, \\etc.~Though the hardware has been investigated for more\nthan a decade, the theoretical guarantees have only recently been derived.\nInspired by deep learning, various deep neural networks have also been\ndeveloped to reconstruct the HD data-cube in spectral SCI and video SCI. This\narticle reviews recent advances in SCI hardware, theory and algorithms,\nincluding both optimization-based and deep-learning-based algorithms. Diverse\napplications and the outlook of SCI are also discussed.\n"}
{"abstract": "  Traditional breast cancer image classification methods require manual\nextraction of features from medical images, which not only require professional\nmedical knowledge, but also have problems such as time-consuming and\nlabor-intensive and difficulty in extracting high-quality features. Therefore,\nthe paper proposes a computer-based feature fusion Convolutional neural network\nbreast cancer image classification and detection method. The paper pre-trains\ntwo convolutional neural networks with different structures, and then uses the\nconvolutional neural network to automatically extract the characteristics of\nfeatures, fuse the features extracted from the two structures, and finally use\nthe classifier classifies the fused features. The experimental results show\nthat the accuracy of this method in the classification of breast cancer image\ndata sets is 89%, and the classification accuracy of breast cancer images is\nsignificantly improved compared with traditional methods.\n"}
{"abstract": "  Thin films are gaining ground in photonics and optoelectronics, promising\nimprovements in their efficiency and functionality as well as decreased\nmaterial usage as compared to bulk technologies. However, proliferation of thin\nfilms would benefit not only from continuous improvements in their fabrication,\nbut also from a unified and accurate theoretical framework of the interplay of\nphotons and charge carriers. In particular, such a framework would need to\naccount quantitatively and self-consistently for photon recycling and\ninterference effects. To this end, here we combine the drift-diffusion\nformalism of charge carrier dynamics and the fluctuational electrodynamics of\nphoton transport self-consistently using the recently introduced\ninterference-extended radiative transfer equations. The resulting equation\nsystem can be solved numerically using standard simulation tools, and as an\nexample, here we apply it to study well-known GaAs thin-film solar cells. In\naddition to obtaining the expected device characteristics, we analyze the\nunderlying complex photon transport and recombination-generation processes,\ndemonstrating the physical insight provided for unevenly excited structures\nthrough the direct and self-consistent description of photons and charge\ncarriers. The methodology proposed in this work is general and can be used to\nobtain accurate physical insight into a wide range of planar optoelectronic\ndevices, of which the thin-film single-junction solar cells studied here are\nonly one example.\n"}
{"abstract": "  The notions of synchronizing and partitioning sets are recently introduced\nvariants of locally consistent parsings with great potential in\nproblem-solving. In this paper we propose a deterministic algorithm that\nconstructs for a given readonly string of length $n$ over the alphabet\n$\\{0,1,\\ldots,n^{\\mathcal{O}(1)}\\}$ a variant of $\\tau$-partitioning set with\nsize $\\mathcal{O}(b)$ and $\\tau = \\frac{n}{b}$ using $\\mathcal{O}(b)$ space and\n$\\mathcal{O}(\\frac{1}{\\epsilon}n)$ time provided $b \\ge n^\\epsilon$, for\n$\\epsilon > 0$. As a corollary, for $b \\ge n^\\epsilon$ and constant $\\epsilon >\n0$, we obtain linear construction algorithms with $\\mathcal{O}(b)$ space on top\nof the string for two major small-space indexes: a sparse suffix tree, which is\na compacted trie built on $b$ chosen suffixes of the string, and a longest\ncommon extension (LCE) index, which occupies $\\mathcal{O}(b)$ space and allows\nus to compute the longest common prefix for any pair of substrings in\n$\\mathcal{O}(n/b)$ time. For both, the $\\mathcal{O}(b)$ construction storage is\nasymptotically optimal since the tree itself takes $\\mathcal{O}(b)$ space and\nany LCE index with $\\mathcal{O}(n/b)$ query time must occupy at least\n$\\mathcal{O}(b)$ space by a known trade-off (at least for $b \\ge \\Omega(n /\n\\log n)$). In case of arbitrary $b \\ge \\Omega(\\log^2 n)$, we present\nconstruction algorithms for the partitioning set, sparse suffix tree, and LCE\nindex with $\\mathcal{O}(n\\log_b n)$ running time and $\\mathcal{O}(b)$ space,\nthus also improving the state of the art.\n"}
{"abstract": "  We obtain entropy formulas for SRB measures with finite entropy given by\ninducing schemes. In the first part of the work, we obtain Pesin entropy\nformula for the class of noninvertible systems whose SRB measures are given by\nGibbs-Markov induced maps. In the second part, we obtain Pesin entropy formula\nfor invertible maps whose SRB measures given by Young sets, taking into account\na classical compression technique along the stable direction that allows a\nreduction of the return map associated with a Young set to a Gibbs-Markov map.\nIn both cases, we give applications of our main results to several classes of\ndynamical systems with singular sets, where the classical results by Ruelle and\nPesin cannot be applied. We also present examples of systems with SRB measures\ngiven by inducing schemes for which Ruelle inequality does not hold.\n"}
{"abstract": "  Technical Debt is a metaphor used to describe the situation in which\nlong-term software artifact quality is traded for short-term goals in software\nprojects. In recent years, the concept of self-admitted technical debt (SATD)\nwas proposed, which focuses on debt that is intentionally introduced and\ndescribed by developers. Although prior work has made important observations\nabout admitted technical debt in source code, little is known about SATD in\nbuild systems. In this paper, we set out to better understand the\ncharacteristics of SATD in build systems. To do so, through a qualitative\nanalysis of 500 SATD comments in the Maven build system of 291 projects, we\ncharacterize SATD by location and rationale (reason and purpose). Our results\nshow that limitations in tools and libraries, and complexities of dependency\nmanagement are the most frequent causes, accounting for 50% and 24% of the\ncomments. We also find that developers often document SATD as issues to be\nfixed later. As a first step towards the automatic detection of SATD rationale,\nwe train classifiers to detect the two most frequently occurring reasons and\nthe four most frequently occurring purposes of SATD in the content of comments\nin Maven build systems. The classifier performance is promising, achieving an\nF1-score of 0.71-0.79. Finally, within 16 identified 'ready-to-be-addressed'\nSATD instances, the three SATD submitted by pull requests and the five SATD\nsubmitted by issue reports were resolved after developers were made aware. Our\nwork presents the first step towards understanding technical debt in build\nsystems and opens up avenues for future work, such as tool support to track and\nmanage SATD backlogs.\n"}
{"abstract": "  Currently, practitioners follow a top-down approach in automotive development\nprojects. However, recent studies have shown that this top-down approach is not\nsuitable for the implementation and testing of modern automotive systems.\nSpecifically, practitioners increasingly fail to specify requirements and tests\nfor systems with complex component interactions (e.g., e-mobility systems). In\nthis paper, we address this research gap and propose an integrated and\niterative scenario-based technique for the specification of requirements and\ntest scenarios. Our idea is to combine both a top-down and a bottom-up\nintegration strategy. For the top-down approach, we use a behavior-driven\ndevelopment (BDD) technique to drive the modeling of high-level system\ninteractions from the user's perspective. For the bottom-up approach, we\ndiscovered that natural language processing (NLP) techniques are suited to make\ntextual specifications of existing components accessible to our technique. To\nintegrate both directions, we support the joint execution and automated\nanalysis of system-level interactions and component-level behavior. We\ndemonstrate the feasibility of our approach by conducting a case study at\nKostal (Tier1 supplier). The case study corroborates, among other things, that\nour approach supports practitioners in improving requirements and test\nspecifications for integrated system behavior.\n"}
{"abstract": "  The generalized Langevin equation is a model for the motion of coarse-grained\nparticles where dissipative forces are represented by a memory term. The\nnumerical realization of such a model requires the implementation of a\nstochastic delay-differential equation and the estimation of a corresponding\nmemory kernel. Here we develop a new approach for computing a data-driven\nMarkov model for the motion of the particles, given equidistant samples of\ntheir velocity autocorrelation function. Our method bypasses the determination\nof the underlying memory kernel by representing it via up to about twenty\nauxiliary variables. The algorithm is based on a sophisticated variant of the\nProny method for exponential interpolation and employs the Positive Real Lemma\nfrom model reduction theory to extract the associated Markov model. We\ndemonstrate the potential of this approach for the test case of anomalous\ndiffusion, where data are given analytically, and then apply our method to\nvelocity autocorrelation data of molecular dynamics simulations of a colloid in\na Lennard-Jones fluid. In both cases, the VACF and the memory kernel can be\nreproduced very accurately. Moreover, we show that the algorithm can also\nhandle input data with large statistical noise. We anticipate that it will be a\nvery useful tool in future studies that involve dynamic coarse-graining of\ncomplex soft matter systems.\n"}
{"abstract": "  In this paper, we investigate the contributions of anomalous quartic gauge\ncouplings (aQGCs) to $Z\\gamma jj$ production at the Large Hadron Collider (LHC)\nin the context of Standard Model effective theory (SMEFT). When energy scale is\nlarge, the validity of SMEFT becomes an important issue. To ensure the\nvalidity, the unitarity bound is applied in a model independent approach, which\nis found to have significant suppressive effects on the signals of $O_{M_i}$\noperators. The kinematic and polarization features of the aQGC signals are also\nstudied. The polarization effect is useful to highlight the signals of\n$O_{T_i}$ operators. The sensitivity estimates on dimension-8 operators with\nunitarity bounds at $\\sqrt{s}=14$ TeV are obtained.\n"}
{"abstract": "  Intelligent reflecting surface (IRS) is a promising technology for achieving\nspectrum and energy efficient wireless networks cost-effectively. Most existing\nworks on IRS have focused on exploiting IRS to enhance the performance of\nwireless communication or wireless information transmission (WIT), while its\npotential for boosting the efficiency of radio-frequency (RF) wireless energy\ntransmission (WET) still remains largely open. Although IRS-aided WET shares\nsimilar characteristics with IRS-aided WIT, they differ fundamentally in terms\nof design objective, receiver architecture, and practical constraints. In this\npaper, we provide a tutorial overview on how to efficiently design IRS-aided\nWET systems as well as IRS-aided systems with both WIT and WET, namely\nIRS-aided simultaneous wireless information and power transfer (SWIPT) and\nIRS-aided wireless powered communication network (WPCN), mainly from a\ncommunication and signal processing perspective. In particular, we present\nstate-of-the-art solutions to tackle the unique challenges in operating these\nsystems, such as IRS passive reflection optimization, channel estimation and\ndeployment. In addition, we also propose new solution approaches and point out\nimportant directions for future research and investigation.\n"}
{"abstract": "  By coloring a signed graph by signed colors, one obtains the signed chromatic\npolynomial of the signed graph. For each signed graph we construct graded\ncohomology groups whose graded Euler characteristic yields the signed chromatic\npolynomial of the signed graph. We show that the cohomology groups satisfy a\nlong exact sequence which corresponds to signed deletion-contraction rule. This\nwork is motivated by Helme-Guizon and Rong's construction of the\ncategorification for the chromatic polynomial of unsigned graphs.\n"}
{"abstract": "  Quantum circuit theory has become a powerful and indispensable tool to\npredict the dynamics of superconducting circuits. Surprisingly however, the\nquestion of how to properly account for a time-dependent driving via external\nmagnetic fields has hardly been addressed so far. Here, we derive a general\nrecipe to construct a low-energy Hamiltonian, taking as input only the circuit\ngeometry and the solution of the external magnetic fields. A gauge fixing\nprocedure for the scalar and vector potentials is given which assures that\ntime-varying magnetic fluxes make contributions only to the potential function\nin the Schr\\\"odinger equation. Our proposed procedure is valid for continuum\ngeometries and thus significantly generalizes previous efforts, which were\nbased on discrete circuits. We study some implications of our results for the\nconcrete example of a parallel-plate SQUID circuit. We show that if we insist\non representing the response of this SQUID with individual, discrete\ncapacitances associated with each individual Josephson junction, this is only\npossible if we permit the individual capacitance values to be negative,\ntime-dependent or even momentarily singular. Finally, we provide some\nexperimentally testable predictions, such as a strong enhancement of the qubit\nrelaxation rates arising from the effective negative capacitances, and the\nemergence of a Berry phase due to time dependence of these capacitances.\n"}
{"abstract": "  Action quality assessment (AQA) aims at automatically judging human action\nbased on a video of the said action and assigning a performance score to it.\nThe majority of works in the existing literature on AQA transform RGB videos to\nhigher-level representations using C3D networks. These higher-level\nrepresentations are used to perform action quality assessment. Due to the\nrelatively shallow nature of C3D, the quality of extracted features is lower\nthan what could be extracted using a deeper convolutional neural network. In\nthis paper, we experiment with deeper convolutional neural networks with\nresidual connections for learning representations for action quality\nassessment. We assess the effects of the depth and the input clip size of the\nconvolutional neural network on the quality of action score predictions. We\nalso look at the effect of using (2+1)D convolutions instead of 3D convolutions\nfor feature extraction. We find that the current clip level feature\nrepresentation aggregation technique of averaging is insufficient to capture\nthe relative importance of features. To overcome this, we propose a\nlearning-based weighted-averaging technique that can perform better. We achieve\na new state-of-the-art Spearman's rank correlation of 0.9315 (an increase of\n0.45%) on the MTL-AQA dataset using a 34 layer (2+1)D convolutional neural\nnetwork with the capability of processing 32 frame clips, using our proposed\naggregation technique.\n"}
{"abstract": "  Most mobile health apps employ data visualization to help people view their\nhealth and activity data, but these apps provide limited support for visual\ndata exploration. Furthermore, despite its huge potential benefits, mobile\nvisualization research in the personal data context is sparse. This work aims\nto empower people to easily navigate and compare their personal health data on\nsmartphones by enabling flexible time manipulation with speech. We designed and\ndeveloped Data@Hand, a mobile app that leverages the synergy of two\ncomplementary modalities: speech and touch. Through an exploratory study with\n13 long-term Fitbit users, we examined how multimodal interaction helps\nparticipants explore their own health data. Participants successfully adopted\nmultimodal interaction (i.e., speech and touch) for convenient and fluid data\nexploration. Based on the quantitative and qualitative findings, we discuss\ndesign implications and opportunities with multimodal interaction for better\nsupporting visual data exploration on mobile devices.\n"}
{"abstract": "  Reinforcement learning is finding its way to real-world problem application,\ntransferring from simulated environments to physical setups. In this work, we\nimplement vision-based alignment of an optical Mach-Zehnder interferometer with\na confocal telescope in one arm, which controls the diameter and divergence of\nthe corresponding beam. We use a continuous action space; exponential scaling\nenables us to handle actions within a range of over two orders of magnitude.\nOur agent trains only in a simulated environment with domain randomizations. In\nan experimental evaluation, the agent significantly outperforms an existing\nsolution and a human expert.\n"}
{"abstract": "  This paper introduces the first open-source FPGA-based infrastructure,\nMetaSys, with a prototype in a RISC-V core, to enable the rapid implementation\nand evaluation of a wide range of cross-layer techniques in real hardware.\nHardware-software cooperative techniques are powerful approaches to improve the\nperformance, quality of service, and security of general-purpose processors.\nThey are however typically challenging to rapidly implement and evaluate in\nreal hardware as they require full-stack changes to the hardware, OS, system\nsoftware, and instruction-set architecture (ISA).\n  MetaSys implements a rich hardware-software interface and lightweight\nmetadata support that can be used as a common basis to rapidly implement and\nevaluate new cross-layer techniques. We demonstrate MetaSys's versatility and\nease-of-use by implementing and evaluating three cross-layer techniques for:\n(i) prefetching for graph analytics; (ii) bounds checking in memory unsafe\nlanguages, and (iii) return address protection in stack frames; each technique\nonly requiring ~100 lines of Chisel code over MetaSys.\n  Using MetaSys, we perform the first detailed experimental study to quantify\nthe performance overheads of using a single metadata management system to\nenable multiple cross-layer optimizations in CPUs. We identify the key sources\nof bottlenecks and system inefficiency of a general metadata management system.\nWe design MetaSys to minimize these inefficiencies and provide increased\nversatility compared to previously-proposed metadata systems. Using three use\ncases and a detailed characterization, we demonstrate that a common metadata\nmanagement system can be used to efficiently support diverse cross-layer\ntechniques in CPUs.\n"}
{"abstract": "  We propose a black-box reduction that turns a certain reinforcement learning\nalgorithm with optimal regret in a (near-)stationary environment into another\nalgorithm with optimal dynamic regret in a non-stationary environment,\nimportantly without any prior knowledge on the degree of non-stationarity. By\nplugging different algorithms into our black-box, we provide a list of examples\nshowing that our approach not only recovers recent results for (contextual)\nmulti-armed bandits achieved by very specialized algorithms, but also\nsignificantly improves the state of the art for (generalized) linear bandits,\nepisodic MDPs, and infinite-horizon MDPs in various ways. Specifically, in most\ncases our algorithm achieves the optimal dynamic regret\n$\\widetilde{\\mathcal{O}}(\\min\\{\\sqrt{LT}, \\Delta^{1/3}T^{2/3}\\})$ where $T$ is\nthe number of rounds and $L$ and $\\Delta$ are the number and amount of changes\nof the world respectively, while previous works only obtain suboptimal bounds\nand/or require the knowledge of $L$ and $\\Delta$.\n"}
{"abstract": "  Rather than using wooden sticks to simulate the breakage of trees in high\nwinds as in most research, we employed fresh samples with branches and leaves\nto certify the crucial role played by the tree crown. By using the blowdown\nwind tunnel with a maximum wind speed of 60 m/s, we purposely reduce the number\nof leaves and show that the drag force will drop by as much as two thirds when\nhalf pruned. Based on real observations, we model the leaf by an open and full\ncone in the presence of light and strong wind, and calculate how their\ncorresponding cross-sectional area and drag force vary with wind speed.\nDifferent power-law relations are predicted and confirmed by experiments for\nthese properties before and after the formation of a full cone. Compared to the\nempirical value of 1/3 and 3/4, our simple model gave 2/5 and 2/3 for the\npower-law exponent of cross-sectional area at low and high winds. Discrepancy\ncan be accounted for by including further details, such as the reorientation of\nopen cones and the movement of branches.\n"}
{"abstract": "  Motivated by biological considerations, we study sparse neural maps from an\ninput layer to a target layer with sparse activity, and specifically the\nproblem of storing $K$ input-target associations $(x,y)$, or memories, when the\ntarget vectors $y$ are sparse. We mathematically prove that $K$ undergoes a\nphase transition and that in general, and somewhat paradoxically, sparsity in\nthe target layers increases the storage capacity of the map. The target vectors\ncan be chosen arbitrarily, including in random fashion, and the memories can be\nboth encoded and decoded by networks trained using local learning rules,\nincluding the simple Hebb rule. These results are robust under a variety of\nstatistical assumptions on the data. The proofs rely on elegant properties of\nrandom polytopes and sub-gaussian random vector variables. Open problems and\nconnections to capacity theories and polynomial threshold maps are discussed.\n"}
{"abstract": "  We characterize the magnetic properties and domain structure of Pt/Ni/Co\nasymmetric superlattices in comparison to the more established Pt/Co/Ni\nstructure. This reversal in stacking sequence leads to a marked drop in\ninterfacial magnetic anisotropy and the magnitude of the interfacial\nDzyaloshinskii-Moriya interaction (DMI) as inferred from the DW structure,\nwhich we speculate could be related to a degradation of the Pt/Co interface\nwhen Pt is deposited on top of the Co layer. Lorentz transmission electron\nmicroscopy reveals exclusively N\\'eel type domain walls and, with a\nperpendicular field, N\\'eel skyrmions in the Pt/Co/Ni films. Conversely, the\nPt/Ni/Co samples show only achiral Bloch domain walls, which leads to the\nformation of achiral Bloch ($Q=1$) and type II bubbles ($Q=0$) at increased\nperpendicular field. Combined with the reduced anisotropy leading to greater\nbubble densities, the latter case makes for an excellent test bed to examine\nthe benefits of topological charge on stability. Simultaneous observation of\nBloch and type II bubbles shows a roughly 50 mT larger annihilation field for\nthe former. An in-plane component to the magnetic field is shown to both impact\nthe structure of the formed bubbles and separately suppress the topological\nbenefit.\n"}
{"abstract": "  We show that a unified framework based on an $SU(2)_H$ horizontal symmetry\nwhich generates a naturally large neutrino transition magnetic moment and\nexplains the XENON1T electron recoil excess also predicts a positive shift in\nthe muon anomalous magnetic moment. This shift is of the right magnitude to be\nconsistent with the Brookhaven measurement as well as the recent Fermilab\nmeasurement of the muon $g-2$. A relatively light neutral scalar from a Higgs\ndoublet with mass near 100 GeV contributes to muon $g-2$, while its charged\npartner induces the neutrino magnetic moment. We analyze the collider tests of\nthis framework and find that the HL-LHC can probe the entire parameter space of\nthese models.\n"}
{"abstract": "  We propose a novel pairwise distance measure between image keypoint sets, for\nthe purpose of large-scale medical image indexing. Our measure generalizes the\nJaccard index to account for soft set equivalence (SSE) between keypoint\nelements, via an adaptive kernel framework modeling uncertainty in keypoint\nappearance and geometry. A new kernel is proposed to quantify the variability\nof keypoint geometry in location and scale. Our distance measure may be\nestimated between $O(N^2)$ image pairs in $O(N~\\log~N)$ operations via keypoint\nindexing. Experiments report the first results for the task of predicting\nfamily relationships from medical images, using 1010 T1-weighted MRI brain\nvolumes of 434 families including monozygotic and dizygotic twins, siblings and\nhalf-siblings sharing 100%-25% of their polymorphic genes. Soft set equivalence\nand the keypoint geometry kernel improve upon standard hard set equivalence\n(HSE) and appearance kernels alone in predicting family relationships.\nMonozygotic twin identification is near 100%, and three subjects with uncertain\ngenotyping are automatically paired with their self-reported families, the\nfirst reported practical application of image-based family identification. Our\ndistance measure can also be used to predict group categories, sex is predicted\nwith an AUC=0.97. Software is provided for efficient fine-grained curation of\nlarge, generic image datasets.\n"}
{"abstract": "  Deep Neural Network (DNN)-based physical layer techniques are attracting\nconsiderable interest due to their potential to enhance communication systems.\nHowever, most studies in the physical layer have tended to focus on the\napplication of DNN models to wireless communication problems but not to\ntheoretically understand how does a DNN work in a communication system. In this\nletter, we aim to quantitatively analyse why DNNs can achieve comparable\nperformance in the physical layer comparing with traditional techniques and\ntheir cost in terms of computational complexity. We further investigate and\nalso experimentally validate how information is flown in a DNN-based\ncommunication system under the information theoretic concepts.\n"}
{"abstract": "  Current face forgery detection methods achieve high accuracy under the\nwithin-database scenario where training and testing forgeries are synthesized\nby the same algorithm. However, few of them gain satisfying performance under\nthe cross-database scenario where training and testing forgeries are\nsynthesized by different algorithms. In this paper, we find that current\nCNN-based detectors tend to overfit to method-specific color textures and thus\nfail to generalize. Observing that image noises remove color textures and\nexpose discrepancies between authentic and tampered regions, we propose to\nutilize the high-frequency noises for face forgery detection. We carefully\ndevise three functional modules to take full advantage of the high-frequency\nfeatures. The first is the multi-scale high-frequency feature extraction module\nthat extracts high-frequency noises at multiple scales and composes a novel\nmodality. The second is the residual-guided spatial attention module that\nguides the low-level RGB feature extractor to concentrate more on forgery\ntraces from a new perspective. The last is the cross-modality attention module\nthat leverages the correlation between the two complementary modalities to\npromote feature learning for each other. Comprehensive evaluations on several\nbenchmark databases corroborate the superior generalization performance of our\nproposed method.\n"}
{"abstract": "  We theoretically investigate advanced multi-functional integrated photonic\nfilters formed by three waveguide coupled Sagnac loop reflectors (3WC-SLRs). By\ntailoring the coherent mode interference, the spectral response of the 3WC-SLR\nresonators is engineered to achieve diverse filtering functions with high\nperformance. These include optical analogues of Fano resonances that yield\nultrahigh spectral extinction ratios (ERs) and slope rates, resonance mode\nsplitting with high ERs and low free spectral ranges, and classical\nButterworth, Bessel, Chebyshev, and elliptic filters. A detailed analysis of\nthe impact of the structural parameters and fabrication tolerances is provided\nto facilitate device design and optimization. The requirements for practical\napplications are also considered. These results theoretically verify the\neffectiveness of using 3WC-SLR resonators as multi-functional integrated\nphotonic filters for flexible spectral engineering in diverse applications.\n"}
{"abstract": "  Motivated by experiments and formal asymptotic expansions in the physics\nliterature, Maor and Shachar (J. Elasticity 134 (2019), 149-173) studied the\nbehaviour of a model elastic energy of maps between manifolds with incompatible\nmetrics. For thin objects they analysed the scaling of the minimal elastic\nenergy as a function of the thickness. In particular they showed that for maps\nfrom a ball of radius h in an oriented Riemannian manifold to Euclidean space,\nthe infimum of a model elastic energy per unit volume scales like the fourth\npower of h and after rescaling one gets convergence to a quadratic expression\nin the curvature tensor R(p), where p denotes the centre of the ball. In this\npaper we show the same result for general compact oriented Riemannian targets\nwith R(p) replaced by a suitable difference of the curvature tensors in the\ntarget and the domain, thus answering Open Question 1 in the paper by Maor and\nShachar. The result extends noncompact targets provided they satisfy a uniform\nregularity condition. A key idea in the proof is to use Lipschitz\napproximations to define a suitable notion of convergence.\n"}
{"abstract": "  In this paper, we consider the global well-posedness of the incompressible\nHall-MHD equations in $\\mathbb{R}^3$. We prove that the solution of this system\nis globally regular if the initial data is axisymmetric and the swirl\ncomponents of the velocity and magnetic vorticity are trivial. It should be\npointed out that the initial data without any smallness and in low regularity\nspaces. This improves a previous result established in \\cite{Fan2013}.\n"}
{"abstract": "  The capability of the traditional semi-supervised learning (SSL) methods is\nfar from real-world application since they do not consider (1) class imbalance\nand (2) class distribution mismatch between labeled and unlabeled data. This\npaper addresses such a relatively under-explored problem, imbalanced\nsemi-supervised learning, where heavily biased pseudo-labels can harm the model\nperformance. Interestingly, we find that the semantic pseudo-labels from a\nsimilarity-based classifier in feature space and the traditional pseudo-labels\nfrom the linear classifier show the complementary property. To this end, we\npropose a general pseudo-labeling framework to address the bias motivated by\nthis observation. The key idea is to class-adaptively blend the semantic\npseudo-label to the linear one, depending on the current pseudo-label\ndistribution. Thereby, the increased semantic pseudo-label component suppresses\nthe false positives in the majority classes and vice versa. We term the novel\npseudo-labeling framework for imbalanced SSL as Distribution-Aware\nSemantics-Oriented (DASO) Pseudo-label. Extensive evaluation on CIFAR10/100-LT\nand STL10-LT shows that DASO consistently outperforms both recently proposed\nre-balancing methods for label and pseudo-label. Moreover, we demonstrate that\ntypical SSL algorithms can effectively benefit from unlabeled data with DASO,\nespecially when (1) class imbalance and (2) class distribution mismatch exist\nand even on recent real-world Semi-Aves benchmark.\n"}
{"abstract": "  The vast majority of binaries containing a compact object and a regular star\nspend most of their time in a quiescent state where no strong interactions\noccur between components. Detection of these binaries is extremely challenging\nand only few candidates have been detected through optical spectroscopy.\nSelf-lensing represents a new means of detecting compact objects in binaries,\nwhere gravitational lensing of the light from the visible component by the\ncompact object produces periodic optical flares. Here we show that current and\nplanned large-area optical surveys can detect a significant number ($\\sim\n100$-$10,000$s) of these self-lensing binaries and provide insights into the\nproperties of the compact lenses. We show that many of the predicted population\nof observable self-lensing binaries will be observed with multiple self-lensing\nflares; this both improves the chances of detection and also immediately\ndistinguishes them from chance-alignment micro-lensing events. Through\nself-lensing we can investigate long - but previously hidden - stages of binary\nevolution and consequently provide new constraints on evolutionary models which\nimpact on the number and nature of double compact object mergers.\n"}
{"abstract": "  A $[k,n,1]$-graph is a $k$-partite graph with parts of order $n$ such that\nthe bipartite graph induced by any pair of parts is a matching. An independent\ntransversal in such a graph is an independent set that intersects each part in\na single vertex. A factor of independent transversals is a set of $n$\npairwise-disjoint independent transversals. Let $f(k)$ be the smallest integer\n$n_0$ such that every $[k,n,1]$-graph has a factor of independent transversals\nassuming $n \\ge n_0$. Several known conjectures imply that for $k \\ge 2$,\n$f(k)=k$ if $k$ is even and $f(k)=k+1$ if $k$ is odd. While a simple greedy\nalgorithm based on iterating Hall's Theorem shows that $f(k) \\le 2k-2$, no\nbetter bound is known and in fact, there are instances showing that the bound\n$2k-2$ is tight for the greedy algorithm. Here we significantly improve upon\nthe greedy algorithm bound and prove that $f(k) \\le 1.78k$ for all $k$\nsufficiently large, answering a question of MacKeigan.\n"}
{"abstract": "  The coronavirus disease 2019 (COVID-19) is a severe global pandemic that has\nclaimed millions of lives and continues to overwhelm public health systems in\nmany countries. The spread of COVID-19 pandemic has negatively impacted the\nhuman mobility patterns such as daily transportation-related behavior of the\npublic. There is a requirement to understand the disease spread patterns and\nits routes among neighboring individuals for the timely implementation of\ncorrective measures at the required placement. To increase the effectiveness of\ncontact tracing, countries across the globe are leveraging advancements in\nmobile technology and Internet of Things (IoT) to aid traditional manual\ncontact tracing to track individuals who have come in close contact with\nidentified COVID-19 patients. Even as the first administration of vaccines\nbegins in 2021, the COVID-19 management strategy will continue to be\nmulti-pronged for the foreseeable future with digital contact tracing being a\nvital component of the response along with the use of preventive measures such\nas social distancing and the use of face masks. After some months of deployment\nof digital contact tracing technology, deeper insights into the merits of\nvarious approaches and the usability, privacy, and ethical trade-offs involved\nare emerging. In this paper, we provide a comprehensive analysis of digital\ncontact tracing solutions in terms of their methodologies and technologies in\nthe light of the new data emerging about international experiences of\ndeployments of digital contact tracing technology. We also provide a discussion\non open challenges such as scalability, privacy, adaptability and highlight\npromising directions for future work.\n"}
{"abstract": "  In this note we observe that for constrained convex minimization problems\n$\\min_{x \\in P}f(x)$ over a polytope $P$, dual prices for the linear program\n$\\min_{z \\in P} \\nabla f(x) z$ obtained from linearization at approximately\noptimal solutions $x$ have a similar interpretation of rate of change in\noptimal value as for linear programming, providing a convex form of sensitivity\nanalysis. This is of particular interest for Frank--Wolfe algorithms (also\ncalled conditional gradients), forming an important class of first-order\nmethods, where a basic building block is linear minimization of gradients of\n$f$ over $P$, which in most implementations already compute the dual prices as\na by-product.\n"}
{"abstract": "  For the model of a multi-level system in the rotating wave approximation we\nobtain the corrections for a usual weak coupling limit dynamics by means of\nperturbation theory with Bogolubov-van Hove scaling. It generalizes our\nprevious results on a spin-boson model in the rotating wave approximation.\nAdditionally, in this work we take into account some dependence of the system\nHamiltonian on the small parameter. We show that the dynamics is long-time\nMarkovian, i.e. after the bath correlation time all the non-Markovianity could\nbe captured by the renormalization of initial condition and correlation\nfunctions.\n"}
{"abstract": "  In the problem \"Localization and trilateration with the minimum number of\nlandmarks\", we faced the 3-Guard and classic Art Gallery Problems. The goal of\nthe art gallery problem is to find the minimum number of guards within a simple\npolygon to observe and protect its entirety. It has many applications in\nrobotics, telecommunications, etc. There are some approaches to handle the art\ngallery problem that is theoretically NP-hard. This paper offers an efficient\nmethod based on the Particle Filter algorithm which solves the most fundamental\nstate of the problem in a nearly optimal manner. The experimental results on\nthe random polygons generated by Bottino et al. \\cite{bottino2011nearly} show\nthat the new method is more accurate with fewer or equal guards. Furthermore,\nwe discuss resampling and particle numbers to minimize the run time.\n"}
{"abstract": "  Three-dimensional (3D) magnetohydrodynamic simulations are carried out to\nexplore magnetic reconnections in the presence of 3D magnetic nulls and\nquasi-separatrix layers (QSLs). The initial magnetic fields are created by\nsuperposing uniform vertical magnetic fields of two different magnitudes on a\nlinear force-free field. The interior of the numerical box contains two 3D\nnulls with separatrix domes separated by a quasi-separator (or hyperbolic flux\ntube) with QSLs. In the first simulation, the uniform vertical field is so\nlarge that the nulls are located at low heights and the domes are separate.\nInitially unbalanced Lorentz forces drive rotational flows that form strong\nelectric currents and strong torsional fan reconnection at the 3D nulls and\nweak QSL reconnection at the hyperbolic flux tube. Flipping or slipping of\nfield lines is observed in both cases. In the second simulation, with a weaker\nvertical field and larger domes, the separatrix surfaces meet at the central\nquasi-separator and their rotation drives stronger QSL reconnection than\nbefore.\n"}
{"abstract": "  Encryption provides a method to protect data outsourced to a DBMS provider,\ne.g., in the cloud. However, performing database operations over encrypted data\nrequires specialized encryption schemes that carefully balance security and\nperformance. In this paper, we present a new encryption scheme that can\nefficiently perform equi-joins over encrypted data with better security than\nthe state-of-the-art. In particular, our encryption scheme reduces the leakage\nto equality of rows that match a selection criterion and only reveals the\ntransitive closure of the sum of the leakages of each query in a series of\nqueries. Our encryption scheme is provable secure. We implemented our\nencryption scheme and evaluated it over a dataset from the TPC-H benchmark.\n"}
{"abstract": "  To interact seamlessly with robots, users must infer the causes of a robot's\nbehavior and be confident about that inference. Hence, trust is a necessary\ncondition for human-robot collaboration (HRC). Despite its crucial role, it is\nlargely unknown how trust emerges, develops, and supports human interactions\nwith nonhuman artefacts. Here, we review the literature on trust, human-robot\ninteraction, human-robot collaboration, and human interaction at large. Early\nmodels of trust suggest that trust entails a trade-off between benevolence and\ncompetence, while studies of human-to-human interaction emphasize the role of\nshared behavior and mutual knowledge in the gradual building of trust. We then\nintroduce a model of trust as an agent's best explanation for reliable sensory\nexchange with an extended motor plant or partner. This model is based on the\ncognitive neuroscience of active inference and suggests that, in the context of\nHRC, trust can be cast in terms of virtual control over an artificial agent. In\nthis setting, interactive feedback becomes a necessary component of the\ntrustor's perception-action cycle. The resulting model has important\nimplications for understanding human-robot interaction and collaboration, as it\nallows the traditional determinants of human trust to be defined in terms of\nactive inference, information exchange and empowerment. Furthermore, this model\nsuggests that boredom and surprise may be used as markers for under and\nover-reliance on the system. Finally, we examine the role of shared behavior in\nthe genesis of trust, especially in the context of dyadic collaboration,\nsuggesting important consequences for the acceptability and design of\nhuman-robot collaborative systems.\n"}
{"abstract": "  The unsupervised search for overdense regions in high-dimensional feature\nspaces, where locally high population densities may be associated with\nanomalous contaminations to an otherwise more uniform population, is of\nrelevance to applications ranging from fundamental research to industrial use\ncases. Motivated by the specific needs of searches for new phenomena in\nparticle collisions, we propose a novel approach that targets signals of\ninterest populating compact regions of the feature space. The method consists\nin a systematic scan of subspaces of a standardized copula of the feature\nspace, where the minimum p-value of a hypothesis test of local uniformity is\nsought by gradient descent. We characterize the performance of the proposed\nalgorithm and show its effectiveness in several experimental situations.\n"}
{"abstract": "  As for the unique continuation property (UCP) of solutions in\n$(0,T)\\times\\Omega$ with a domain $\\Omega\\subset{\\mathbb R}^n,\\,n\\in{\\mathbb\nN}$ for a multi-terms time fractional diffusion equation, we have already shown\nit by assuming that the solutions are zero for $t\\le0$ (see \\cite{LN2019}).\nHere the strongly elliptic operator for this diffusion equation can depend on\ntime and the orders of its time fractional derivatives are in $(0,2)$. This\npaper is a continuation of the previous study. The aim of this paper is to drop\nthe assumption that the solutions are zero for $t\\le0$. We have achieved this\naim by first using the usual Holmgren transformation together with the argument\nin \\cite{LN2019} to derive the UCP in $(T_1,T_2)\\times B_1$ for some\n$0<T_1<T_2<T$ and a ball $B_1\\subset\\Omega$. Then if $u$ is the solution of the\nequation with $u=0$ in $(T_1,T_2)\\times B_1$, we show $u=0$ also in\n$((0,T_1]\\cup[T_2,T))\\times B_r$ for some $r<1$ by using the argument in\n\\cite{LN2019} which uses two Holmgren type transformations different from the\nusual one. This together with spatial coordinates transformation, we can obtain\nthe usual UCP which we call it the classical UCP given in the title of this\npaper for our time fractional diffusion equation.\n"}
{"abstract": "  Magnetic topological materials have attracted much attention due to the\ncorrelation between topology and magnetism. Recent studies suggest that\nEuCd$_2$As$_2$ is an antiferromagnetic topological material. Here by carrying\nout thorough magnetic, electrical and thermodynamic property measurements, we\ndiscover a long time relaxation of the magnetic susceptibility in\nEuCd$_2$As$_2$. The (001) in-plane magnetic susceptibility at 5 K is found to\ncontinuously increase up to $\\sim$10% over the time of $\\sim$14 hours. The\nmagnetic relaxation is anisotropic and strongly depends on the temperature and\nthe applied magnetic field. These results will stimulate further theoretical\nand experimental studies to understand the origin of the relaxation process and\nits effect on the electronic structure and physical properties of the magnetic\ntopological materials.\n"}
{"abstract": "  We present highly optimized data structures for the dynamic predecessor\nproblem, where the task is to maintain a set $S$ of $w$-bit numbers under\ninsertions, deletions, and predecessor queries (return the largest element in\n$S$ no larger than a given key). The problem of finding predecessors can be\nviewed as a generalized form of the membership problem, or as a simple version\nof the nearest neighbour problem. It lies at the core of various real-world\nproblems such as internet routing.\n  In this work, we engineer (1) a simple implementation of the idea of universe\nreduction, similar to van-Emde-Boas trees (2) variants of y-fast tries\n[Willard, IPL'83], and (3) B-trees with different strategies for organizing the\nkeys contained in the nodes, including an implementation of dynamic fusion\nnodes [P\\v{a}tra\\c{s}cu and Thorup, FOCS'14]. We implement our data structures\nfor $w=32,40,64$, which covers most typical scenarios.\n  Our data structures finish workloads faster than previous approaches while\nbeing significantly more space-efficient, e.g., they clearly outperform\nstandard implementations of the STL by finishing up to four times as fast using\nless than a third of the memory. Our tests also provide more general insights\non data structure design, such as how small sets should be stored and handled\nand if and when new CPU instructions such as advanced vector extensions pay\noff.\n"}
{"abstract": "  This article proves a Pythagoras-type formula for the sides and diagonals of\na polygon inscribed in a semicircle having one of the sides of the polygon as\ndiameter.\n"}
{"abstract": "  In the Einstein-bumblebee gravity, the Lorentz symmetry is spontaneously\nbroken by a vector field. In this paper, we attempt to test the Lorentz\nsymmetry via the observation of the shadow cast by the Kerr-like black hole\nwith or without plasma. A novel phenomenon of the Lorentz-violating parameter\non the shadow is observed. The result shows that when the observer gradually\nmoves from the poles to the equatorial plane, the shadow radius $R_{\\rm s}$\nfirstly decreases and then increases with the Lorentz-violating parameter. Such\nnonmonotonic behavior provides us an important understanding on the black hole\nshadow in the Einstein-bumblebee gravity. Besides, three more distortion\nobservables are calculated, and found to increase with the Lorentz-violating\nparameter. Moreover, when a homogeneous plasma is present, the motion of the\nphoton is analyzed. We further observe that the refractive index shrinks the\nsize, while enhances the deformation of the shadow. Finally, adopting the\nobserved data of the diameter of M87$^*$, we find the refractive index is more\nfavored in (0.914, 1).\n"}
{"abstract": "  We present recent updates in our development of a library for Universal\nAlgebra in the UniMath proof assistant. The code here discussed concerns\nmulti-sorted signatures and their algebras, along with the basics for equation\nsystems. Moreover, we give neat constructions of the corresponding univalent\ncategories by using the formalism of displayed categories, and show that the\nterm algebra over a signature is the initial object of the category.\n  Besides the formalization, we reflect on the methodological principles --\nbased on the idea of evaluability of our elementary construction by the\nbuilt-in normalization procedure of the system -- leading our coding style, and\nshow that this path is practicable indeed by sketching simple examples.\n"}
{"abstract": "  Recent controversies related to topics such as fake news, privacy, and\nalgorithmic bias have prompted increased public scrutiny of digital\ntechnologies and soul-searching among many of the people associated with their\ndevelopment. In response, the tech industry, academia, civil society, and\ngovernments have rapidly increased their attention to \"ethics\" in the design\nand use of digital technologies (\"tech ethics\"). Yet almost as quickly as\nethics discourse has proliferated across the world of digital technologies, the\nlimitations of these approaches have also become apparent: tech ethics is vague\nand toothless, is subsumed into corporate logics and incentives, and has a\nmyopic focus on individual engineers and technology design rather than on the\nstructures and cultures of technology production. As a result of these\nlimitations, many have grown skeptical of tech ethics and its proponents,\ncharging them with \"ethics-washing\": promoting ethics research and discourse to\ndefuse criticism and government regulation without committing to ethical\nbehavior. By looking at how ethics has been taken up in both science and\nbusiness in superficial and depoliticizing ways, I recast tech ethics as a\nterrain of contestation where the central fault line is not whether it is\ndesirable to be ethical, but what \"ethics\" entails and who gets to define it.\nThis framing highlights the significant limits of current approaches to tech\nethics and the importance of studying the formulation and real-world effects of\ntech ethics. In order to identify and develop more rigorous strategies for\nreforming digital technologies and the social relations that they mediate, I\ndescribe a sociotechnical approach to tech ethics, one that reflexively applies\nmany of tech ethics' own lessons regarding digital technologies to tech ethics\nitself.\n"}
{"abstract": "  Detecting small targets at range is difficult because there is not enough\nspatial information present in an image sub-region containing the target to use\ncorrelation-based methods to differentiate it from dynamic confusers present in\nthe scene. Moreover, this lack of spatial information also disqualifies the use\nof most state-of-the-art deep learning image-based classifiers. Here, we use\ncharacteristics of target tracks extracted from video sequences as data from\nwhich to derive distinguishing topological features that help robustly\ndifferentiate targets of interest from confusers. In particular, we calculate\npersistent homology from time-delayed embeddings of dynamic statistics\ncalculated from motion tracks extracted from a wide field-of-view video stream.\nIn short, we use topological methods to extract features related to target\nmotion dynamics that are useful for classification and disambiguation and show\nthat small targets can be detected at range with high probability.\n"}
{"abstract": "  Images are an indispensable part of the news content we consume. Highly\nemotional images from sources of misinformation can greatly influence our\njudgements. We present two studies on the effects of emotional facial images on\nusers' perception of bias in news content and the credibility of sources. In\nstudy 1, we investigate the impact of happy and angry facial images on users'\ndecisions. In study 2, we focus on sources' systematic emotional treatment of\nspecific politicians. Our results show that depending on the political\norientation of the source, the cumulative effect of angry facial emotions\nimpacts users' perceived content bias and source credibility. When sources\nsystematically portray specific politicians as angry, users are more likely to\nfind those sources as less credible and their content as more biased. These\nresults highlight how implicit visual propositions manifested by emotions in\nfacial expressions might have a substantial effect on our trust of news content\nand sources.\n"}
{"abstract": "  In network design problems, such as compact routing, the goal is to route\npackets between nodes using the (approximated) shortest paths. A desirable\nproperty of these routes is a small number of hops, which makes them more\nreliable, and reduces the transmission costs. Following the overwhelming\nsuccess of stochastic tree embeddings for algorithmic design, Haeupler,\nHershkowitz, and Zuzic (STOC'21) studied hop-constrained Ramsey-type metric\nembeddings into trees. Specifically, embedding $f:G(V,E)\\rightarrow T$ has\nRamsey hop-distortion $(t,M,\\beta,h)$ (here $t,\\beta,h\\ge1$ and $M\\subseteq V$)\nif $\\forall u,v\\in M$, $d_G^{(\\beta\\cdot h)}(u,v)\\le d_T(u,v)\\le t\\cdot\nd_G^{(h)}(u,v)$. $t$ is called the distortion, $\\beta$ is called the\nhop-stretch, and $d_G^{(h)}(u,v)$ denotes the minimum weight of a $u-v$ path\nwith at most $h$ hops. Haeupler {\\em et al.} constructed embedding where $M$\ncontains $1-\\epsilon$ fraction of the vertices and $\\beta=t=O(\\frac{\\log^2\nn}{\\epsilon})$. They used their embedding to obtain multiple bicriteria\napproximation algorithms for hop-constrained network design problems.\n  In this paper, we first improve the Ramsey-type embedding to obtain\nparameters $t=\\beta=\\frac{\\tilde{O}(\\log n)}{\\epsilon}$, and generalize it to\narbitrary distortion parameter $t$ (in the cost of reducing the size of $M$).\nThis embedding immediately implies polynomial improvements for all the\napproximation algorithms from Haeupler {\\em et al.}. Further, we construct\nhop-constrained clan embeddings (where each vertex has multiple copies), and\nuse them to construct bicriteria approximation algorithms for the group Steiner\ntree problem, matching the state of the art of the non constrained version.\nFinally, we use our embedding results to construct hop constrained distance\noracles, distance labeling, and most prominently, the first hop constrained\ncompact routing scheme with provable guarantees.\n"}
{"abstract": "  Recent observations of young embedded clumpy clusters and statistical\nidentifications of binary star clusters have provided new insights into the\nformation process and subsequent dynamical evolution of star clusters. The\nearly dynamical evolution of clumpy stellar structures provides the conditions\nfor the origin of binary star clusters. Here, we carry out $N$-body simulations\nin order to investigate the formation of binary star clusters in the Milky Way\nand in the Large Magellanic Cloud (LMC). We find that binary star clusters can\nform from stellar aggregates with a variety of initial conditions. For a given\ninitial virial ratio, a higher degree of initial substructure results in a\nhigher fraction of binary star clusters. The number of binary star clusters\ndecreases over time due to merging or dissolution of the binary system.\nTypically, $\\sim 45\\%$ of the aggregates evolve into binary/multiple clusters\nwithin $t=20$~Myr in the Milky Way environment, while merely $\\sim30\\%$\nsurvives beyond $t=50$~Myr, with separations $\\lesssim 50$~pc. On the other\nhand, in the LMC, $\\sim 90\\%$ of the binary/multiple clusters survive beyond\n$t=20$~Myr and the fraction decreases to $\\sim 80\\%$ at $t=50$~Myr, with\nseparations $\\lesssim 35~$pc. Multiple clusters are also rapidly formed for\nhighly-substructured and expanding clusters. The additional components tend to\ndetach and the remaining binary star cluster merges. The merging process can\nproduce fast rotating star clusters with mostly flat rotation curves that speed\nup in the outskirts.\n"}
{"abstract": "  The measurement of the energy spectrum of cosmic ray helium nuclei from 70\nGeV to 80 TeV using 4.5 years of data recorded by the DArk Matter Particle\nExplorer (DAMPE) is reported in this work. A hardening of the spectrum is\nobserved at an energy of about 1.3 TeV, similar to previous observations. In\naddition, a spectral softening at about 34 TeV is revealed for the first time\nwith large statistics and well controlled systematic uncertainties, with an\noverall significance of $4.3\\sigma$. The DAMPE spectral measurements of both\ncosmic protons and helium nuclei suggest a particle charge dependent softening\nenergy, although with current uncertainties a dependence on the number of\nnucleons cannot be ruled out.\n"}
{"abstract": "  Drug membrane interaction is a very significant bioprocess to consider in\ndrug discovery. Here, we propose a novel deep learning framework coined DMInet\nto study drug-membrane interactions that leverages large-scale Martini\ncoarse-grained molecular simulations of permeation of drug-like molecules\nacross six different lipid membranes. The network of DMInet receives three\ninputs, viz, the drug-like molecule, membrane type and spatial distance across\nmembrane thickness, and predicts the potential of mean force with structural\nresolution across the lipid membrane and membrane selectivity. Inheriting from\ncoarse-grained Martini representation of organic molecules and combined with\ndeep learning, DMInet has the potential for more accelerated high throughput\nscreening in drug discovery across a much larger chemical space than that can\nbe explored by physics-based simulations alone. Moreover, DMInet is highly\nflexible in its nature and holds the possibilities for other properties\nprediction without significant change of the architecture. Last but not least,\nthe architecture of DMInet is general and can be applied to other membrane\nproblems involving permeation and selection.\n"}
{"abstract": "  Superconductivity is significantly enhanced in monolayer FeSe grown on\nSrTiO3, but not for multilayer films, in which large strength of nematicity\ndevelops. However, the link between the high-transition temperature\nsuperconductivity in monolayer and the correlation related nematicity in\nmultilayer FeSe films is not well understood. Here, we use low-temperature\nscanning tunneling microscopy to study few-layer FeSe thin films grown by\nmolecular beam epitaxy. We observe an incommensurate long-range smectic phase,\nwhich solely appears in bilayer FeSe films. The smectic order still locally\nexists and gradually fades away with increasing film thickness, while it\nsuddenly vanishes in monolayer FeSe, indicative of an abrupt smectic phase\ntransition. Surface alkali-metal doping can suppress the smectic phase and\ninduce high-Tc superconductivity in bilayer FeSe. Our observations provide\nevidence that the monolayer FeSe is in close proximity to the smectic phase,\nand its superconductivity is likely enhanced by this electronic instability as\nwell.\n"}
{"abstract": "  A large number of particle species allows to formulate quantum gravity in a\nspecial double-scaling limit, the species limit. In this regime, quantum\ngravitational amplitudes simplify substantially. An infinite set of\nperturbative corrections, that usually blur the picture, vanishes, whereas the\ncollective and non-perturbative effects can be cleanly extracted. Such are the\neffects that control physics of black holes and of de Sitter and their\nentanglement curves. In string theory example, we show that the entropy of open\nstrings matches the Gibbons-Hawking entropy of a would-be de Sitter state at\nthe point of saturation of the species bound. This shows, from yet another\nangle, why quantum gravity/string theory cannot tolerate a de Sitter vacuum.\nFinally, we discuss various observational implications.\n"}
{"abstract": "  We introduce SynSE, a novel syntactically guided generative approach for\nZero-Shot Learning (ZSL). Our end-to-end approach learns progressively refined\ngenerative embedding spaces constrained within and across the involved\nmodalities (visual, language). The inter-modal constraints are defined between\naction sequence embedding and embeddings of Parts of Speech (PoS) tagged words\nin the corresponding action description. We deploy SynSE for the task of\nskeleton-based action sequence recognition. Our design choices enable SynSE to\ngeneralize compositionally, i.e., recognize sequences whose action descriptions\ncontain words not encountered during training. We also extend our approach to\nthe more challenging Generalized Zero-Shot Learning (GZSL) problem via a\nconfidence-based gating mechanism. We are the first to present zero-shot\nskeleton action recognition results on the large-scale NTU-60 and NTU-120\nskeleton action datasets with multiple splits. Our results demonstrate SynSE's\nstate of the art performance in both ZSL and GZSL settings compared to strong\nbaselines on the NTU-60 and NTU-120 datasets. The code and pretrained models\nare available at https://github.com/skelemoa/synse-zsl\n"}
{"abstract": "  Multimodal sentiment analysis benefits various applications such as\nhuman-computer interaction and recommendation systems. It aims to infer the\nusers' bipolar ideas using visual, textual, and acoustic signals. Although\nresearchers affirm the association between cognitive cues and emotional\nmanifestations, most of the current multimodal approaches in sentiment analysis\ndisregard user-specific aspects. To tackle this issue, we devise a novel method\nto perform multimodal sentiment prediction using cognitive cues, such as\npersonality. Our framework constructs an adaptive tree by hierarchically\ndividing users and trains the LSTM-based submodels, utilizing an\nattention-based fusion to transfer cognitive-oriented knowledge within the\ntree. Subsequently, the framework consumes the conclusive agglomerative\nknowledge from the adaptive tree to predict final sentiments. We also devise a\ndynamic dropout method to facilitate data sharing between neighboring nodes,\nreducing data sparsity. The empirical results on real-world datasets determine\nthat our proposed model for sentiment prediction can surpass trending rivals.\nMoreover, compared to other ensemble approaches, the proposed transfer-based\nalgorithm can better utilize the latent cognitive cues and foster the\nprediction outcomes. Based on the given extrinsic and intrinsic analysis\nresults, we note that compared to other theoretical-based techniques, the\nproposed hierarchical clustering approach can better group the users within the\nadaptive tree.\n"}
{"abstract": "  Most evolutionary algorithms have parameters, which allow a great flexibility\nin controlling their behavior and adapting them to new problems. To achieve the\nbest performance, it is often needed to control some of the parameters during\noptimization, which gave rise to various parameter control methods. In recent\nworks, however, similar advantages have been shown, and even proven, for\nsampling parameter values from certain, often heavy-tailed, fixed\ndistributions. This produced a family of algorithms currently known as \"fast\nevolution strategies\" and \"fast genetic algorithms\".\n  However, only little is known so far about the influence of these\ndistributions on the performance of evolutionary algorithms, and about the\nrelationships between (dynamic) parameter control and (static) parameter\nsampling. We contribute to the body of knowledge by presenting, for the first\ntime, an algorithm that computes the optimal static distributions, which\ndescribe the mutation operator used in the well-known simple $(1+\\lambda)$\nevolutionary algorithm on a classic benchmark problem OneMax. We show that, for\nlarge enough population sizes, such optimal distributions may be surprisingly\ncomplicated and counter-intuitive. We investigate certain properties of these\ndistributions, and also evaluate the performance regrets of the $(1+\\lambda)$\nevolutionary algorithm using commonly used mutation distributions.\n"}
{"abstract": "  Unsupervised image-to-image translation methods have received a lot of\nattention in the last few years. Multiple techniques emerged tackling the\ninitial challenge from different perspectives. Some focus on learning as much\nas possible from several target style images for translations while other make\nuse of object detection in order to produce more realistic results on\ncontent-rich scenes. In this work, we assess how a method that has initially\nbeen developed for single object translation performs on more diverse and\ncontent-rich images. Our work is based on the FUNIT[1] framework and we train\nit with a more diverse dataset. This helps understanding how such method\nbehaves beyond their initial frame of application. We present a way to extend a\ndataset based on object detection. Moreover, we propose a way to adapt the\nFUNIT framework in order to leverage the power of object detection that one can\nsee in other methods.\n"}
{"abstract": "  The analysis of the compression effects in generative adversarial networks\n(GANs) after training, i.e. without any fine-tuning, remains an unstudied,\nalbeit important, topic with the increasing trend of their computation and\nmemory requirements. While existing works discuss the difficulty of compressing\nGANs during training, requiring novel methods designed with the instability of\nGANs training in mind, we show that existing compression methods (namely\nclipping and quantization) may be directly applied to compress GANs\npost-training, without any additional changes. High compression levels may\ndistort the generated set, likely leading to an increase of outliers that may\nnegatively affect the overall assessment of existing k-nearest neighbor (KNN)\nbased metrics. We propose two new precision and recall metrics based on\nlocality-sensitive hashing (LSH), which, on top of increasing the outlier\nrobustness, decrease the complexity of assessing an evaluation sample against\n$n$ reference samples from $O(n)$ to $O(\\log(n))$, if using LSH and KNN, and to\n$O(1)$, if only applying LSH. We show that low-bit compression of several\npre-trained GANs on multiple datasets induces a trade-off between precision and\nrecall, retaining sample quality while sacrificing sample diversity.\n"}
{"abstract": "  To tackle ever-increasing city traffic congestion problems, researchers have\nproposed deep learning models to aid decision-makers in the traffic control\ndomain. Although the proposed models have been remarkably improved in recent\nyears, there are still questions that need to be answered before deploying\nmodels. For example, it is difficult to figure out which models provide\nstate-of-the-art performance, as recently proposed models have often been\nevaluated with different datasets and experiment environments. It is also\ndifficult to determine which models would work when traffic conditions change\nabruptly (e.g., rush hour). In this work, we conduct two experiments to answer\nthe two questions. In the first experiment, we conduct an experiment with the\nstate-of-the-art models and the identical public datasets to compare model\nperformance under a consistent experiment environment. We then extract a set of\ntemporal regions in the datasets, whose speeds change abruptly and use these\nregions to explore model performance with difficult intervals. The experiment\nresults indicate that Graph-WaveNet and GMAN show better performance in\ngeneral. We also find that prediction models tend to have varying performances\nwith data and intervals, which calls for in-depth analysis of models on\ndifficult intervals for real-world deployment.\n"}
{"abstract": "  The notion of normal category was introduced by KSS Nambooripad in connection\nwith the study of the structure of regular semigroups using cross\nconnections\\cite{nambooripad1994theory}. It is an abstraction of the category\nof principal left ideals of a regular semigroup. A normal band is a semigroup\n$B$ satisfying $a^2=a$ and $abca=acba$ for all $a,b,c \\in B$. Since the normal\nbands are regular semigroups, the category $\\mathcal{L}(B)$ of principal left\nideals of a normal band $B$ is a normal category. One of the special properties\nof this category is that the morphism sets admit a partial order compatible\nwith the composition of morphisms. In this article we derive several properties\nof this partial order and obtain a new characterization of this partial order.\n"}
{"abstract": "  Fairness and accountability are two essential pillars for trustworthy\nArtificial Intelligence (AI) in healthcare. However, the existing AI model may\nbe biased in its decision marking. To tackle this issue, we propose an\nadversarial multi-task training strategy to simultaneously mitigate and detect\nbias in the deep learning-based medical image analysis system. Specifically, we\npropose to add a discrimination module against bias and a critical module that\npredicts unfairness within the base classification model. We further impose an\northogonality regularization to force the two modules to be independent during\ntraining. Hence, we can keep these deep learning tasks distinct from one\nanother, and avoid collapsing them into a singular point on the manifold.\nThrough this adversarial training method, the data from the underprivileged\ngroup, which is vulnerable to bias because of attributes such as sex and skin\ntone, are transferred into a domain that is neutral relative to these\nattributes. Furthermore, the critical module can predict fairness scores for\nthe data with unknown sensitive attributes. We evaluate our framework on a\nlarge-scale public-available skin lesion dataset under various fairness\nevaluation metrics. The experiments demonstrate the effectiveness of our\nproposed method for estimating and improving fairness in the deep\nlearning-based medical image analysis system.\n"}
{"abstract": "  The standard nonlinear Schr\\\"{o}dinger equation (NLSE) is one of the elegant\nequations to find the information about the modulational instability criteria\nof dust-ion-acoustic (DIA) waves (DIAWs) and associated DIA rogue waves\n(DIARWs) in a three-component dusty plasma medium having inertialess\nsuper-thermal kappa distributed electrons, and inertial warm positive ions and\nnegative dust grains. It can be seen that under the consideration of inertial\nwarm ions along with inertial negatively charged dust grains, the plasma system\nsupports both fast and slow DIA modes. The charge state and number density of\nthe ion and dust grain are responsible to change the instability conditions of\nthe DIAWs and the configuration of DIARWs. These results are to be considered\nthe cornerstone for explaining the real puzzles in space and laboratory dusty\nplasmas.\n"}
{"abstract": "  Users are demanding increased data security. As a result, security is rapidly\nbecoming a first-order design constraint in next generation computing systems.\nResearchers and practitioners are exploring various security technologies to\nmeet user demand such as trusted execution environments (e.g., Intel SGX, ARM\nTrustZone), homomorphic encryption, and differential privacy. Each technique\nprovides some degree of security, but differs with respect to threat coverage,\nperformance overheads, as well as implementation and deployment challenges. In\nthis paper, we present a systemization of knowledge (SoK) on these design\nconsiderations and trade-offs using several prominent security technologies.\nOur study exposes the need for \\textit{software-hardware-security} codesign to\nrealize efficient and effective solutions of securing user data. In particular,\nwe explore how design considerations across applications, hardware, and\nsecurity mechanisms must be combined to overcome fundamental limitations in\ncurrent technologies so that we can minimize performance overhead while\nachieving sufficient threat model coverage. Finally, we propose a set of\nguidelines to facilitate putting these secure computing technologies into\npractice.\n"}
{"abstract": "  Given entropy's central role in multiple areas of physics and science, one\nimportant task is to develop a systematic and unifying approach to defining\nentropy. Games of chance become a natural candidate for characterising the\nuncertainty of a physical system, as a system's performance in gambling games\ndepends solely on the uncertainty of its output. In this work, we construct\nfamilies of games which induce pre-orders corresponding to majorization,\nconditional majorization, and channel majorization. Finally, we provide\noperational interpretations for all pre-orders, show the relevance of these\nresults to dynamical resource theories, and find the only asymptotically\ncontinuous classical dynamic entropy.\n"}
{"abstract": "  Significant deviations from the Standard Model are observed in semileptonic\ncharged and neutral-current B-decays, the muon magnetic moment, and the\nextraction of the Cabibbo angle. We propose that these deviations point towards\na coherent pattern of New Physics effects induced by two scalar mediators, a\nleptoquark $S_1$ and a charged singlet $\\phi^+$. While $S_1$ can provide\nsolutions to charged-current $B$-decays and the muon magnetic moment, and\n$\\phi^+$ can accommodate the Cabibbo-angle anomaly independently, their\none-loop level synergy can also address neutral-current $B$-decays. This\nframework provides the most minimal explanation to the above-mentioned\nanomalies, while being consistent with all other phenomenological constraints.\n"}
{"abstract": "  In this work, we revisit the non-minimally coupled Higgs Inflation scenario\nand investigate its observational viability in light of the current Cosmic\nMicrowave Background, Baryon Acoustic Oscillation and type Ia Supernovae data.\nWe explore the effects of the Coleman-Weinberg approximation to the Higgs\npotential in the primordial universe, connecting the predictions for the\nLagrangian parameters at inflationary scales to the electroweak observables\nthrough Renormalization Group methods at two-loop order. As the main result, we\nfind that observations on the electroweak scale are in disagreement with the\nconstraints obtained from the cosmological data sets used in the analysis.\nSpecifically, an $\\approx 8\\sigma$-discrepancy between the inflationary\nparameters and the electroweak value of the top quark mass is found, which\nsuggests that a significant deviation from the scenario analysed is required by\nthe cosmological data.\n"}
{"abstract": "  In this paper, we propose CI-VI an efficient and scalable solver for\nsemi-implicit variational inference (SIVI). Our method, first, maps SIVI's\nevidence lower bound (ELBO) to a form involving a nonlinear functional nesting\nof expected values and then develops a rigorous optimiser capable of correctly\nhandling bias inherent to nonlinear nested expectations using an\nextrapolation-smoothing mechanism coupled with gradient sketching. Our\ntheoretical results demonstrate convergence to a stationary point of the ELBO\nin general non-convex settings typically arising when using deep network models\nand an order of $O(t^{-\\frac{4}{5}})$ gradient-bias-vanishing rate. We believe\nthese results generalise beyond the specific nesting arising from SIVI to other\nforms. Finally, in a set of experiments, we demonstrate the effectiveness of\nour algorithm in approximating complex posteriors on various data-sets\nincluding those from natural language processing.\n"}
{"abstract": "  Adversarial learning strategy has demonstrated remarkable performance in\ndealing with single-source Domain Adaptation (DA) problems, and it has recently\nbeen applied to Multi-source DA (MDA) problems. Although most existing MDA\nstrategies rely on a multiple domain discriminator setting, its effect on the\nlatent space representations has been poorly understood. Here we adopt an\ninformation-theoretic approach to identify and resolve the potential adverse\neffect of the multiple domain discriminators on MDA: disintegration of\ndomain-discriminative information, limited computational scalability, and a\nlarge variance in the gradient of the loss during training. We examine the\nabove issues by situating adversarial DA in the context of information\nregularization. This also provides a theoretical justification for using a\nsingle and unified domain discriminator. Based on this idea, we implement a\nnovel neural architecture called a Multi-source Information-regularized\nAdaptation Networks (MIAN). Large-scale experiments demonstrate that MIAN,\ndespite its structural simplicity, reliably and significantly outperforms other\nstate-of-the-art methods.\n"}
{"abstract": "  Since their appearance, Smart Cities have aimed at improving the daily life\nof people, helping to make public services smarter and more efficient. Several\nof these services are often intended to provide better security conditions for\ncitizens and drivers. In this vein, we present Heimdall, an AI-based video\nsurveillance system for traffic monitoring and anomalies detection. The\nproposed system features three main tiers: a ground level, consisting of a set\nof smart lampposts equipped with cameras and sensors, and an advanced AI unit\nfor detecting accidents and traffic anomalies in real time; a territorial\nlevel, which integrates and combines the information collected from the\ndifferent lampposts, and cross-correlates it with external data sources, in\norder to coordinate and handle warnings and alerts; a training level, in charge\nof continuously improving the accuracy of the modules that have to sense the\nenvironment. Finally, we propose and discuss an early experimental approach for\nthe detection of anomalies, based on a Faster R-CNN, and adopted in the\nproposed infrastructure.\n"}
{"abstract": "  We report on a process for the fiber-coupling of electrically driven\ncavity-enhanced quantum dot light emitting devices. The developed technique\nallows for the direct and permanent coupling of p-i-n-doped quantum dot\nmicropillar cavities to single-mode optical fibers. The coupling process, fully\ncarried out at room temperature, involves a spatial scanning technique, where\nthe fiber facet is positioned relative to a device with a diameter of 2 $\\mu$m\nusing the fiber-coupled electroluminescence of the cavity emission as a\nfeedback parameter. Subsequent gluing and UV curing enables a rigid and\npermanent coupling between micropillar and fiber core. Comparing our\nexperimental results with finite element method simulations indicates a\ncavity-to-fiber mode-coupling efficiency of ~46%. Furthermore, we demonstrate\npulsed current injection at a repetition rate exceeding 200 MHz as well as\nlow-temperature operation down to 77 K of the fiber-coupled micropillar device.\nThe technique presented in this work is an important step in the quest for\nefficient and practical quantum light sources for applications in quantum\ninformation.\n"}
{"abstract": "  We study the weakly stable hyperbolic boundary value problem with a large\nzero order oscillatory coefficient. This problem is related to linearized\nproblems in the study of Mach stem and vortex sheets. We wish to establish a\nuniform energy estimate with respect to {\\epsilon}, which is needed in\nmentioned applications and justification of geometric optics solutions, but the\nzero order oscillatory term gives rise to great obstacles. In this paper we\nobtain positive results in the small/medium frequency region by adapting the\napproach of Williams to a more general situation than the one treated there. We\nalso show that it is possible to construct high order approximate solutions by\nthe method of geometric optics for those systems without any restriction on\nfrequencies.\n"}
{"abstract": "  We introduce a framework for uncertainty estimation that both describes and\nextends many existing methods. We consider typical hyperparameters involved in\nclassical training as random variables and marginalise them out to capture\nvarious sources of uncertainty in the parameter space. We investigate which\nforms and combinations of marginalisation are most useful from a practical\npoint of view on standard benchmarking data sets. Moreover, we discuss how some\nmarginalisations may produce reliable estimates of uncertainty without the need\nfor extensive hyperparameter tuning and/or large-scale ensembling.\n"}
{"abstract": "  Embedding silicon nanoparticles in an intermetallic matrix is a promising\nstrategy to produce remarkable bulk anode materials for lithium-ion (Li-ion)\nbatteries with low potential, high electrochemical capacity and good cycling\nstability. These composite materials can be synthetized at a large scale using\nmechanical milling. However, for Si-Ni3Sn4 composites, milling also induces a\nchemical reaction between the two components leading to the formation of free\nSn and NiSi2, which is detrimental to the performance of the electrode. To\nprevent this reaction, a modification of the surface chemistry of the silicon\nhas been undertaken. Si nanoparticles coated with a surface layer of either\ncarbon or oxide were used instead of pure silicon. The influence of the coating\non the composition, (micro)structure and electrochemical properties of\nSi-Ni3Sn4 composites is studied and compared with that of pure Si. Si coating\nstrongly reduces the reaction between Si and Ni3Sn4 during milling. Moreover,\ncontrary to pure silicon, Si-coated composites have a plate-like mor-phology in\nwhich the surface-modified silicon particles are surrounded by a\nnanostructured, Ni3Sn4-based matrix leading to smooth potential profiles during\nelectrochemical cycling. The chemical homogeneity of the matrix is more uniform\nfor carbon-coated than for oxygen-coated silicon. As a consequence, different\nelectrochemical behaviors are obtained depending on the surface chemistry, with\nbetter lithiation properties for the carbon-covered silicon able to deliver\nover 500 mAh/g for at least 400 cycles.\n"}
{"abstract": "  Stylized models of the neurodynamics that underpin sensory motor control in\nanimals are proposed and studied. The voluntary motions of animals are\ntypically initiated by high level intentions created in the primary cortex\nthrough a combination of perceptions of the current state of the environment\nalong with memories of past reactions to similar states. Muscle movements are\nproduced as a result of neural processes in which the parallel activity of\nlarge multiplicities of neurons generate signals that collectively lead to\ndesired actions. Essential to coordinated muscle movement are intentionality,\nprediction, regions of the cortex dealing with misperceptions of sensory cues,\nand a significant level of resilience with respect to disruptions in the neural\npathways through which signals must propagate. While linear models of feedback\ncontrol systems have been well studied over decades, this paper proposes and\nanalyzes a class of models whose aims are to capture some of the essential\nfeatures of neural control of movement. Whereas most linear models of feedback\nsystems entail a state component whose dimension is higher than the number of\ninputs or outputs, the work that follows will treat models in which the numbers\nof input and output channels greatly exceed the state dimension. While we begin\nby considering continuous time systems, the aim will be to treat systems whose\nevolution involves classes of inputs that emulate neural spike trains. Within\nthe proposed class of models, the paper will study resilience to channel\ndropouts, the ways in which noise and uncertainty can be mitigated by an\nappropriate notion of consensus among noisy inputs, and finally, by a simple\nmodel in which binary activations of a multiplicity of input channels produce a\ndynamic response that closely approximates the dynamics of a prescribed linear\nsystem whose inputs are continuous functions of time.\n"}
{"abstract": "  In a recent paper by N. Santowsky et al. [Phys. Rev. D 102, 056014 (2020)],\ncovariant coupled equations were derived to describe a tetraquark in terms of a\nmix of four-quark states $2q 2\\bar{q}$ and two-quark states $q\\bar{q}$. These\nequations were expressed in terms of vertices describing the disintegration of\na tetraquark into identical two-meson states, into a diquark-antidiquark pair,\nand into a quark-antiquark pair. We show that these equations are inconsistent\nas they imply a $q\\bar{q}$ Bethe-Salpeter kernel that is $q\\bar{q}$-reducible.\n"}
{"abstract": "  In this work, we investigate the regularized solutions and their finite\nelement solutions to the inverse source problems governed by partial\ndifferential equations, and establish the stochastic convergence and optimal\nfinite element convergence rates of these solutions, under pointwise\nmeasurement data with random noise.\n  Unlike most existing regularization theories, the regularization error\nestimates are derived without any source conditions, while the error estimates\nof finite element solutions show their explicit dependence on the noise level,\nregularization parameter, mesh size, and time step size, which can guide\npractical choices among these key parameters in real applications. The error\nestimates also suggest an iterative algorithm for determining an optimal\nregularization parameter. Numerical experiments are presented to demonstrate\nthe effectiveness of the analytical results.\n"}
{"abstract": "  Matching markets are often organized in a multi-stage and decentralized\nmanner. Moreover, participants in real-world matching markets often have\nuncertain preferences. This article develops a framework for learning optimal\nstrategies in such settings, based on a nonparametric statistical approach and\nvariational analysis. We propose an efficient algorithm, built upon concepts of\n\"lower uncertainty bound\" and \"calibrated decentralized matching,\" for\nmaximizing the participants' expected payoff. We show that there exists a\nwelfare-versus-fairness trade-off that is characterized by the uncertainty\nlevel of acceptance. Participants will strategically act in favor of a low\nuncertainty level to reduce competition and increase expected payoff. We prove\nthat participants can be better off with multi-stage matching compared to\nsingle-stage matching. We demonstrate aspects of the theoretical predictions\nthrough simulations and an experiment using real data from college admissions.\n"}
{"abstract": "  The objective of this paper is to study the stability of a linear\none-dimensional thermoelastic Bresse system in a bounded domain, where the\ncoupling is given through the first component of the Bresse model with the heat\nconduction of Gurtin-Pipkin type. Two kinds of coupling are considered; the\nfirst coupling is of order one with respect to space variable, and the second\none is of order zero. We state the well-posedness and show the polynomial\nstability of the systems, where the decay rates depend on the smoothness of\ninitial data. Moreover, in case of coupling of order one, we prove the\nequivalence between the exponential stability and some new conditions on the\nparameters of the system. However, when the coupling is of order zero, we prove\nthe non-exponential stability independently of the parameters of the system.\nApplications to the corresponding particular Timoshenko models are also given,\nwhere we prove that both couplings lead to the exponential stability if and\nonly if some conditions on the parameters of the systems are satisfied, and\nboth couplings guarantee the polynomial stability independently of the\nparameters of the systems. The proof is based on the semigroup theory and a\ncombination of the energy method and the frequency domain approach.\n"}
{"abstract": "  Advanced communication protocols are critical to enable the coexistence of\nautonomous robots with humans. Thus, the development of explanatory\ncapabilities is an urgent first step toward autonomous robots. This survey\nprovides an overview of the various types of \"explainability\" discussed in\nmachine learning research. Then, we discuss the definition of \"explainability\"\nin the context of autonomous robots (i.e., explainable autonomous robots) by\nexploring the question \"what is an explanation?\" We further conduct a research\nsurvey based on this definition and present some relevant topics for future\nresearch.\n"}
{"abstract": "  Brain image analysis has advanced substantially in recent years with the\nproliferation of neuroimaging datasets acquired at different resolutions. While\nresearch on brain image super-resolution has undergone a rapid development in\nthe recent years, brain graph super-resolution is still poorly investigated\nbecause of the complex nature of non-Euclidean graph data. In this paper, we\npropose the first-ever deep graph super-resolution (GSR) framework that\nattempts to automatically generate high-resolution (HR) brain graphs with N'\nnodes (i.e., anatomical regions of interest (ROIs)) from low-resolution (LR)\ngraphs with N nodes where N < N'. First, we formalize our GSR problem as a node\nfeature embedding learning task. Once the HR nodes' embeddings are learned, the\npairwise connectivity strength between brain ROIs can be derived through an\naggregation rule based on a novel Graph U-Net architecture. While typically the\nGraph U-Net is a node-focused architecture where graph embedding depends mainly\non node attributes, we propose a graph-focused architecture where the node\nfeature embedding is based on the graph topology. Second, inspired by graph\nspectral theory, we break the symmetry of the U-Net architecture by\nsuper-resolving the low-resolution brain graph structure and node content with\na GSR layer and two graph convolutional network layers to further learn the\nnode embeddings in the HR graph. Third, to handle the domain shift between the\nground-truth and the predicted HR brain graphs, we incorporate adversarial\nregularization to align their respective distributions. Our proposed AGSR-Net\nframework outperformed its variants for predicting high-resolution functional\nbrain graphs from low-resolution ones. Our AGSR-Net code is available on GitHub\nat https://github.com/basiralab/AGSR-Net.\n"}
{"abstract": "  This article discusses the decoding of Gabidulin codes and shows how to\nextend the usual decoder to any supercode of a Gabidulin code at the cost of a\nsignificant decrease of the decoding radius. Using this decoder, we provide\npolynomial time attacks on the rank-metric encryption schemes RAMESSES and\nLIGA.\n"}
{"abstract": "  We present a method for path optimization for ground vehicles in off-road\nenvironments at high speeds. This path optimization considers the kinematic\nconstraints of the vehicle. By thinking in the actuator space we can represent\nthese constraints as limits in the space rather than derived properties of the\npath. In this paper we present a actuator space approach to path optimization\nfor off-road ground vehicles. This is done by representing and operation on the\npath as a list of steering angles over the path length. This transforms the set\nof kinematic constraints into constraints on the steering angle. We then put\nthis path into a gradient descent solver. This produced paths that are\nkinematically feasible and optimized in accordance with our cost function.\nFinally, we tested the system both in simulation and on an off-road vehicle at\nspeeds of 5 m/s.\n"}
{"abstract": "  Machine intelligence is very successful at standard recognition tasks when\nhaving high-quality training data. There is still a significant gap between\nmachine-level pattern recognition and human-level concept learning. Humans can\nlearn under uncertainty from only a few examples and generalize these concepts\nto solve new problems. The growing interest in explainable machine\nintelligence, requires experimental environments and diagnostic tests to\nanalyze weaknesses in existing approaches to drive progress in the field. In\nthis paper, we discuss existing diagnostic tests and test data sets such as\nCLEVR, CLEVERER, CLOSURE, CURI, Bongard-LOGO, V-PROM, and present our own\nexperimental environment: The KANDINSKYPatterns, named after the Russian artist\nWassily Kandinksy, who made theoretical contributions to compositivity, i.e.\nthat all perceptions consist of geometrically elementary individual components.\nThis was experimentally proven by Hubel &Wiesel in the 1960s and became the\nbasis for machine learning approaches such as the Neocognitron and the even\nlater Deep Learning. While KANDINSKYPatterns have computationally controllable\nproperties on the one hand, bringing ground truth, they are also easily\ndistinguishable by human observers, i.e., controlled patterns can be described\nby both humans and algorithms, making them another important contribution to\ninternational research in machine intelligence.\n"}
{"abstract": "  We introduce languages of higher-dimensional automata (HDAs) and develop some\nof their properties. To this end, we define a new category of precubical sets,\nuniquely naturally isomorphic to the standard one, and introduce a notion of\nevent consistency. HDAs are then finite, labeled, event-consistent precubical\nsets with distinguished subsets of initial and accepting cells. Their languages\nare sets of interval orders closed under subsumption; as a major technical step\nwe expose a bijection between interval orders and a subclass of HDAs. We show\nthat any finite subsumption-closed set of interval orders is the language of an\nHDA, that languages of HDAs are closed under binary unions and parallel\ncomposition, and that bisimilarity implies language equivalence.\n"}
{"abstract": "  In this paper, we attack a few-shot open-set recognition (FSOSR) problem,\nwhich is a combination of few-shot learning (FSL) and open-set recognition\n(OSR). It aims to quickly adapt a model to a given small set of labeled samples\nwhile rejecting unseen class samples. Since OSR requires rich data and FSL\nconsiders closed-set classification, existing OSR and FSL methods show poor\nperformances in solving FSOSR problems. The previous FSOSR method follows the\npseudo-unseen class sample-based methods, which collect pseudo-unseen samples\nfrom the other dataset or synthesize samples to model unseen class\nrepresentations. However, this approach is heavily dependent on the composition\nof the pseudo samples. In this paper, we propose a novel unknown class sample\ndetector, named SnaTCHer, that does not require pseudo-unseen samples. Based on\nthe transformation consistency, our method measures the difference between the\ntransformed prototypes and a modified prototype set. The modified set is\ncomposed by replacing a query feature and its predicted class prototype.\nSnaTCHer rejects samples with large differences to the transformed prototypes.\nOur method alters the unseen class distribution estimation problem to a\nrelative feature transformation problem, independent of pseudo-unseen class\nsamples. We investigate our SnaTCHer with various prototype transformation\nmethods and observe that our method consistently improves unseen class sample\ndetection performance without closed-set classification reduction.\n"}
{"abstract": "  We present a dead reckoning strategy for increased resilience to position\nestimation failures on multirotors, using only data from a low-cost IMU and\nnovel, bio-inspired airflow sensors. The goal is challenging, since low-cost\nIMUs are subject to large noise and drift, while 3D airflow sensing is made\ndifficult by the interference caused by the propellers and by the wind. Our\napproach relies on a deep-learning strategy to interpret the measurements of\nthe bio-inspired sensors, a map of the wind speed to compensate for\nposition-dependent wind, and a filter to fuse the information and generate a\npose and velocity estimate. Our results show that the approach reduces the\ndrift with respect to IMU-only dead reckoning by up to an order of magnitude\nover 30 seconds after a position sensor failure in non-windy environments, and\nit can compensate for the challenging effects of turbulent, and spatially\nvarying wind.\n"}
{"abstract": "  We present a SE(3)-equivariant graph neural network (GNN) approach that\ndirectly predicting the formation factor and effective permeability from\nmicro-CT images. FFT solvers are established to compute both the formation\nfactor and effective permeability, while the topology and geometry of the pore\nspace are represented by a persistence-based Morse graph. Together, they\nconstitute the database for training, validating, and testing the neural\nnetworks. While the graph and Euclidean convolutional approaches both employ\nneural networks to generate low-dimensional latent space to represent the\nfeatures of the micro-structures for forward predictions, the SE(3) equivariant\nneural network is found to generate more accurate predictions, especially when\nthe training data is limited. Numerical experiments have also shown that the\nnew SE(3) approach leads to predictions that fulfill the material frame\nindifference whereas the predictions from classical convolutional neural\nnetworks (CNN) may suffer from spurious dependence on the coordinate system of\nthe training data. Comparisons among predictions inferred from training the CNN\nand those from graph convolutional neural networks (GNN) with and without the\nequivariant constraint indicate that the equivariant graph neural network seems\nto perform better than the CNN and GNN without enforcing equivariant\nconstraints.\n"}
{"abstract": "  The non-detection of zero-metallicity stars in ultra-faint dwarf galaxies\n(UFDs) can be used to constrain the Initial Mass Function (IMF) of the first\n(PopIII) stars by means of a statistical comparison between available data and\npredictions from chemical evolution models. To this end we develop a model that\nfollows the formation of isolated UFDs, calibrated to best reproduce the\navailable data for the best studied system: Bo\\\"otes I. Our statistical\napproach shows that UFDs are the best suitable systems to study the\nimplications of the persisting non-detection of zero-metallicity stars on the\nPopIII IMF, i.e. its shape, the minimum mass ($m_{min}$), and the\ncharacteristic mass ($m_{ch}$). We show that accounting for the incomplete\nsampling of the IMF is essential to compute the expected number of long-lived\nPopIII stars in inefficiently star-forming UFDs. By simulating the Color\nMagnitude Diagram of Bo\\\"otes I, and thus take into account the mass-range of\nthe observed stars, we can obtain even tighter constrains on $m_{min}$. By\nexploiting the 96 stars with measured metallicities ($\\rm i < 19$) in the UFDs\nrepresented by our model, we demonstrate that: $m_{ch} > 1 \\rm M_{\\odot}$ or\n$m_{min} > 0.8 \\rm M_{\\odot}$ at $99\\%$ confidence level. This means that a\npresent day IMF for PopIII stars is excluded by our model, and a top-heavy\nPopIII IMF is strongly favoured. We can limit $m_{min} > 0.8 \\rm M_{\\odot}$\nindependent of the PopIII IMF shape by targeting the four UFDs Bo\\\"otes I,\nHercules, Leo IV and Eridanus II with future generation instruments, such as\nELT/MOSAIC ($\\rm i < 25$), which can provide samples of >10\\,000 stars.\n"}
{"abstract": "  Leveraging coherent light-matter interaction in solids is a promising new\ndirection towards control and functionalization of quantum materials, to\npotentially realize regimes inaccessible in equilibrium and stabilize new or\nuseful states of matter. We show how driving the strongly spin-orbit coupled\nproximal Kitaev magnet $\\alpha$-RuCl$_3$ with circularly-polarized light can\ngive rise to a novel ligand-mediated magneto-electric effect that both\nphoto-induces a large dynamical effective magnetic field and dramatically\nalters the interplay of competing isotropic and anisotropic exchange\ninteractions. We propose that tailored light pulses can nudge the material\ntowards the elusive Kitaev quantum spin liquid as well as probe competing\nmagnetic instabilities far from equilibrium, and predict that the transient\ncompetition of magnetic exchange processes can be readily observed via\npump-probe spectroscopy.\n"}
{"abstract": "  We present DaLAJ 1.0, a Dataset for Linguistic Acceptability Judgments for\nSwedish, comprising 9 596 sentences in its first version; and the initial\nexperiment using it for the binary classification task. DaLAJ is based on the\nSweLL second language learner data, consisting of essays at different levels of\nproficiency. To make sure the dataset can be freely available despite the GDPR\nregulations, we have sentence-scrambled learner essays and removed part of the\nmetadata about learners, keeping for each sentence only information about the\nmother tongue and the level of the course where the essay has been written. We\nuse the normalized version of learner language as the basis for the DaLAJ\nsentences, and keep only one error per sentence. We repeat the same sentence\nfor each individual correction tag used in the sentence. For DaLAJ 1.0 we have\nused four error categories (out of 35 available in SweLL), all connected to\nlexical or word-building choices. Our baseline results for the binary\nclassification show an accuracy of 58% for DaLAJ 1.0 using BERT embeddings. The\ndataset is included in the SwedishGlue (Swe. SuperLim) benchmark. Below, we\ndescribe the format of the dataset, first experiments, our insights and the\nmotivation for the chosen approach to data sharing.\n"}
{"abstract": "  Emission-line variability studies have the potential to provide key\ninformation about the structures of the near-nuclear outflow regions of AGN.\nHere we present a VLT/Xshooter spectrum of the nucleus of the ULIRG F01004-2237\nthat was taken in August 2018, about 8 yr after a major flare in its integrated\noptical emission. Compared with our WHT/ISIS spectrum from September 2015, the\nbroad, red wings of the emission lines most closely associated with the flaring\nevent, including HeII(4686), NIII(4100,4640) and HeI(5876), have substantially\ndeclined in flux. In contrast, the broad, blue wings that dominate the [OIII],\n[NeIII], [NeV] and [OI] forbidden lines have increased in flux by a factor 1.4\n- 4.4 (depending on the line). Moreover, the [FeVII](6087) line is detected in\nthe new spectrum for the first time. We interpret these results in terms of a\nlight echo from the outflowing warm gas: the direct emission from the flaring\nevent is continuing to fade, but due to light travel time effects we are only\nnow observing the impact of the flare on the emission from the extended outflow\nregion. Unless the outflow is confined to a small range of angles close to our\nline of sight, these observations imply that the outflow must be relatively\ncompact (r<50 pc). In terms of the nature of the flare event, we speculate that\nthe properties of the F01004-2237 flare may be the result of a tidal disruption\nevent (TDE) occurring in an object with pre-existing AGN activity.\n"}
{"abstract": "  Group testing allows saving chemical reagents, analysis time, and costs, by\ntesting pools of samples instead of individual samples. We introduce a class of\ngroup testing protocols with small dilution, suited to operate even at high\nprevalence ($5\\%-10\\%$), and maximizing the fraction of samples classified\npositive/negative within the first round of tests. Precisely, if the tested\ngroup has exactly one positive sample then the protocols identify it without\nfurther individual tests. The protocols also detect the presence of two or more\npositives in the group, in which case a second round could be applied to\nidentify the positive individuals. With a prevalence of $5\\%$ and maximum\ndilution 6, with 100 tests we classify 242 individuals, $92\\%$ of them in one\nround and $8\\%$ requiring a second individual test. In comparison, the\nDorfman's scheme can test 229 individuals with 100 tests, with a second round\nfor $18.5\\%$ of the individuals.\n"}
{"abstract": "  Cosmological models with an inhomogeneous viscous dark fluid, coupled with\ndark matter in the Friedmann- Robertson-Walker (FRW) flat universe, are\nconsidered. The influence of thermal effects caused by Hawking radiation on the\nvisible horizon is studied, in connection with the classified type I and type\nIII singularities which are known to occur within a finite amount of time.\nAllowance of thermal effects implies that a transition to a type II singularity\ncan take place, in a finite time. We take into account a bulk viscosity of the\ndark fluid, observing the equation of state in the case of radiation, and find\nthat there is a qualitative change in the singular universe of type I: it may\npass into a singularity of type III, or it may avoid the singularity at all.\n"}
{"abstract": "  Axisymmetric and nonaxisymmetric patterns in the cubic-quintic\nSwift-Hohenberg equation posed on a disk with Neumann boundary conditions are\nstudied via numerical continuation and bifurcation analysis. Axisymmetric\nlocalized solutions in the form of spots and rings known from earlier studies\npersist and snake in the usual fashion until they begin to interact with the\nboundary. Depending on parameters, including the disk radius, these states may\nor may not connect to the branch of domain-filling target states. Secondary\ninstabilities of localized axisymmetric states may create multi-arm localized\nstructures that grow and interact with the boundary before broadening into\ndomain filling states. High azimuthal wavenumber wall states referred to as\ndaisy states are also found. Secondary bifurcations from these states include\nlocalized daisies, i.e., wall states localized in both radius and angle.\nDepending on parameters, these states may snake much as in the one-dimensional\nSwift-Hohenberg equation, or invade the interior of the domain, yielding states\nreferred to as worms, or domain-filling stripes.\n"}
{"abstract": "  We propose a monocular depth estimator SC-Depth, which requires only\nunlabelled videos for training and enables the scale-consistent prediction at\ninference time. Our contributions include: (i) we propose a geometry\nconsistency loss, which penalizes the inconsistency of predicted depths between\nadjacent views; (ii) we propose a self-discovered mask to automatically\nlocalize moving objects that violate the underlying static scene assumption and\ncause noisy signals during training; (iii) we demonstrate the efficacy of each\ncomponent with a detailed ablation study and show high-quality depth estimation\nresults in both KITTI and NYUv2 datasets. Moreover, thanks to the capability of\nscale-consistent prediction, we show that our monocular-trained deep networks\nare readily integrated into the ORB-SLAM2 system for more robust and accurate\ntracking. The proposed hybrid Pseudo-RGBD SLAM shows compelling results in\nKITTI, and it generalizes well to the KAIST dataset without additional\ntraining. Finally, we provide several demos for qualitative evaluation.\n"}
{"abstract": "  The perfect fluid solutions admitting a group G$_3$ of isometries acting on\norbits S$_2$ whose curvature has a gradient which is tangent to the fluid flow\n(T-models) are studied from a thermodynamic approach. All the admissible\nthermodynamic schemes are obtained, and the solutions compatible with the\ngeneric ideal gas equation of state are studied in detail. The possible\nphysical interpretation of some previously known T-models is also analyzed.\n"}
{"abstract": "  A set of vertices in a hypergraph is called an independent set if no\nhyperedge is completely contained inside the set. Given a hypergraph, computing\nits largest size independent set is an NP-hard problem.\n  In this work, we study the independent set problem on hypergraphs in a\nnatural semi-random family of instances. Our semi-random model is inspired by\nthe Feige-Kilian model [FK01]. This popular model has also been studied in the\nworks of [FK01, Ste17, MMT20] etc. McKenzie, Mehta, and Trevisan [MMT20] gave\nalgorithms for computing independent sets in such a semi-random family of\ngraphs. The algorithms by McKenzie et al. [MMT20] are based on rounding a\n\"crude-SDP\". We generalize their results and techniques to hypergraphs for an\nanalogous family of hypergraph instances. Our algorithms are based on rounding\nthe \"crude-SDP\" of McKenzie et al. [MMT20], augmented with \"Lasserre/SoS like\"\nhierarchy of constraints. Analogous to the results of McKenzie et al. [MMT20],\nwe study the ranges of input parameters where we can recover the planted\nindependent set or a large independent set.\n"}
{"abstract": "  In this paper, we present QPack, a benchmark for NISQ era quantum computers\nusing QAOA algorithms. Unlike other evaluation metrics in the field, this\nbenchmark evaluates not only one, but multiple important aspects of quantum\ncomputing hardware: the maximum problem size a quantum computer can solve, the\nrequired run-time, as well as the achieved accuracy. This benchmark will also\nbe available for quantum simulators, where it is able to assess the run-time\nand required memory that a simulator needs to run the application. The\napplications MaxCut, dominating set and traveling salesman are included to\nprovide variation in resource requirements. We also discuss the design aspects\nthat are taken in consideration for an optimal benchmark, with critical quantum\nbenchmark requirements in mind. An implementation strategy is presented,\nproviding practical metrics. The measurement results show how QAOA can be\noptimized using classical optimization algorithms, suggesting the use of a\nglobal/local optimizer hybrid implementation. The implementation of the MaxCut,\ndominating set and traveling salesman problem show that different QAOA\napplications use a wide variety of quantum resources. This will allow for a\ndiverse benchmark that promotes optimal design considerations, avoiding\nhardware implementations for specific applications.\n"}
{"abstract": "  Weyl semimetals are playing a major role in condensed matter physics due to\nexotic topological properties, and their coexistence with ferromagnetism may\nlead to enhanced spin-related phenomena. Here, the inverse spin Hall effect\n(ISHE) in the ferromagnetic Weyl-semimetal Heusler alloy Co$_2$MnGa was\ninvestigated at room temperature by means of electrical spin injection in\nlateral spin valve structures. Spin transport properties such as spin\npolarization and spin diffusion length in this material were precisely\nextracted in order to estimate the spin Hall angle $\\theta_{\\textrm{SH}}$,\nwhich was found to be $-0.19\\pm0.04$ and is among the highest reported for a\nferromagnet. Although this value is on the same order of magnitude of known\nheavy metals, the significantly higher resistivity of Co$_2$MnGa implies an\nimprovement on the magnitude of detection voltages, while its ferromagnetic\nnature allows controlling the intensity of SHE through the magnetization\ndirection. It was also shown that Onsager's reciprocity does not hold for this\nsystem, which is in part attributable to a different spin-dependent Hall\nconductivity for spin-up and spin-down carriers.\n"}
{"abstract": "  Neural networks are prone to catastrophic forgetting when trained\nincrementally on different tasks. In order to prevent forgetting, most existing\nmethods retain a small subset of previously seen samples, which in turn can be\nused for joint training with new tasks. While this is indeed effective, it may\nnot always be possible to store such samples, e.g., due to data protection\nregulations. In these cases, one can instead employ generative models to create\nartificial samples or features representing memories from previous tasks.\nFollowing a similar direction, we propose GenIFeR (Generative Implicit Feature\nReplay) for class-incremental learning. The main idea is to train a generative\nadversarial network (GAN) to generate images that contain realistic features.\nWhile the generator creates images at full resolution, the discriminator only\nsees the corresponding features extracted by the continually trained\nclassifier. Since the classifier compresses raw images into features that are\nactually relevant for classification, the GAN can match this target\ndistribution more accurately. On the other hand, allowing the generator to\ncreate full resolution images has several benefits: In contrast to previous\napproaches, the feature extractor of the classifier does not have to be frozen.\nIn addition, we can employ augmentations on generated images, which not only\nboosts classification performance, but also mitigates discriminator overfitting\nduring GAN training. We empirically show that GenIFeR is superior to both\nconventional generative image and feature replay. In particular, we\nsignificantly outperform the state-of-the-art in generative replay for various\nsettings on the CIFAR-100 and CUB-200 datasets.\n"}
{"abstract": "  The recently discovered kagome metal series $A$V$_3$Sb$_5$ ($A$=K, Rb, Cs)\nexhibits topologically nontrivial band structures, chiral charge order and\nsuperconductivity, presenting a unique platform for realizing exotic electronic\nstates. The nature of the superconducting state and the corresponding pairing\nsymmetry are key questions that demand experimental clarification. Here, using\na technique based on the tunneling diode oscillator, the magnetic penetration\ndepth $\\Delta\\lambda(T)$ of CsV$_3$Sb$_5$ was measured down to 0.07 K. A clear\nexponential behavior in $\\Delta\\lambda(T)$ with marked deviations from a $T$ or\n$T^2$ temperature dependence is observed at low temperatures, indicating a\ndeficiency of nodal quasiparticles. Temperature dependence of the superfluid\ndensity and electronic specific heat can be described by two-gap $s$-wave\nsuperconductivity, consistent with the presence of multiple Fermi surfaces in\nCsV$_3$Sb$_5$. These results evidence nodeless superconductivity in\nCsV$_3$Sb$_5$ under ambient pressure, and constrain the allowed pairing\nsymmetry.\n"}
{"abstract": "  Remote sensing provides valuable information about objects or areas from a\ndistance in either active (e.g., RADAR and LiDAR) or passive (e.g.,\nmultispectral and hyperspectral) modes. The quality of data acquired by\nremotely sensed imaging sensors (both active and passive) is often degraded by\na variety of noise types and artifacts. Image restoration, which is a vibrant\nfield of research in the remote sensing community, is the task of recovering\nthe true unknown image from the degraded observed image. Each imaging sensor\ninduces unique noise types and artifacts into the observed image. This fact has\nled to the expansion of restoration techniques in different paths according to\neach sensor type. This review paper brings together the advances of image\nrestoration techniques with particular focuses on synthetic aperture radar and\nhyperspectral images as the most active sub-fields of image restoration in the\nremote sensing community. We, therefore, provide a comprehensive,\ndiscipline-specific starting point for researchers at different levels (i.e.,\nstudents, researchers, and senior researchers) willing to investigate the\nvibrant topic of data restoration by supplying sufficient detail and\nreferences. Additionally, this review paper accompanies a toolbox to provide a\nplatform to encourage interested students and researchers in the field to\nfurther explore the restoration techniques and fast-forward the community. The\ntoolboxes are provided in https://github.com/ImageRestorationToolbox.\n"}
{"abstract": "  The aim of this paper is to establish new pointwise regularity results for\nsolutions to degenerate second order partial differential equations with a\nKolmogorov-type operator of the form $$\\mathscr{L} :=\\sum_{i,j=1}^m\n\\partial^2_{x_i x_j } +\\sum_{i,j=1}^N b_{ij}x_j\\partial_{x_i}-\\partial_t, $$\nwhere $(x,t) \\in \\mathbb{R}^{N+1}$, $1 \\leq m \\le N$ and the matrix\n$B:=(b_{ij})_{i,j=1,\\ldots,N}$ has real constant entries. In particular, we\nshow that if the modulus of $L^p$-mean oscillation of $\\mathscr{L} u$ at the\norigin is Dini, then the origin is a Lebesgue point of continuity in $L^p$\naverage for the second order derivatives $\\partial^2_{x_i x_j} u$,\n$i,j=1,\\ldots,m$, and the Lie derivative $\\left(\\sum_{i,j=1}^N\nb_{ij}x_j\\partial_{x_i}-\\partial_t\\right)u$. Moreover, we are able to provide a\nTaylor-type expansion up to second order with estimate of the rest in $L^p$\nnorm. The proof is based on decay estimates, which we achieve by contradiction,\nblow-up and compactness results.\n"}
{"abstract": "  The goal of this paper is to develop a modeling framework that captures the\ninter-decision dynamics between mobility service providers (MSPs) and travelers\nthat can be used to optimize and analyze policies/regulations related to MSPs.\nTo meet this goal, the paper proposes a tri-level mathematical programming\nmodel with a public-sector decision maker (regulator) at the highest level, the\nMSP in the middle level, and travelers at the lowest level. The regulator aims\nto maximize social welfare via implementing regulations, policies, plans,\ntransit service designs, etc. The MSP aims to maximize profit by adjusting its\nservice designs. Travelers aim to maximize utility by changing their modes and\nroutes. The travelers' decisions depend on the regulator and MSP's decisions\nwhile the MSP decisions themselves depend on the regulator's decisions. To\nsolve the tri-level mathematical program, the study employs Bayesian\noptimization (BO) within a simulation-optimization solution approach. At the\nlowest level, the solution approach includes an agent-based transportation\nsystem simulation model to capture travelers' behavior subject to specific\ndecisions made by the regulator and MSP. The agent-based transportation\nsimulation model includes a mode choice model, a road network, a transit\nnetwork, and an MSP providing automated mobility-on-demand (AMOD) service with\nshared rides. The modeling and solution approaches are applied to Munich,\nGermany in order to validate the model. The case study investigates the tolls\nand parking costs the city administration should set, as well as changes in the\npublic transport budget and a limitation of the AMOD fleet size. Best policy\nsettings are derived for two social welfare definitions, in both of which the\nAMOD fleet size is not regulated as the shared-ride AMOD service provides\nsignificant value to travelers in Munich.\n"}
{"abstract": "  We consider MCMC methods for learning equivalence classes of sparse Gaussian\nDAG models when $p = e^{o(n)}$. The main contribution of this work is a rapid\nmixing result for a random walk Metropolis-Hastings algorithm, which we prove\nusing a canonical path method. It reveals that the complexity of Bayesian\nlearning of sparse equivalence classes grows only polynomially in $n$ and $p$,\nunder some common high-dimensional assumptions. Further, a series of\nhigh-dimensional consistency results is obtained by the path method, including\nthe strong selection consistency of an empirical Bayes model for structure\nlearning and the consistency of a greedy local search on the restricted search\nspace. Rapid mixing and slow mixing results for other structure-learning MCMC\nmethods are also derived. Our path method and mixing time results yield crucial\ninsights into the computational aspects of high-dimensional structure learning,\nwhich may be used to develop more efficient MCMC algorithms.\n"}
{"abstract": "  The present epoch of accelerated cosmic expansion is supposed to be driven by\nan unknown constituent called dark energy, which in the standard model takes\nthe form of a cosmological constant, characterized by a constant equation of\nstate w=-1. An interesting perspective over the role and nature of dark energy\ncan be achieved by drawing a parallel with a previous epoch of accelerated\nexpansion, inflation, which we assume to be driven by a single scalar field,\nthe inflaton. Since the Planck satellite has constrained the value of $n_s$\naway from 1, the inflaton cannot be identified with a pure cosmological\nconstant, as is also suggested by the fact that inflation ended. Thus, it is\ninteresting to verify whether a hypothetical observer would have been able to\nmeasure the deviation of the w of the inflaton from -1. To do so, we consider a\nclass of single-field slow-roll inflationary models dubbed HSR{i}, where the\nhierarchy of Hubble slow-roll parameters is truncated at the i-th order. The\nmodels are tested through a MCMC analysis based on combinations of the latest\nPlanck and BICEP2/Keck data sets, and the resulting chains are converted into\nsets of allowed evolution histories of w. HSR{1} is excluded observationally\nsince it would predict that $n_s=1$, in contrast with the recent Planck\nconstraints, while we find that HSR{2} would prefer w>-1, but is disfavoured by\nthe addition of the BICEP2/Keck data. The overall best description for the data\nis provided by HSR{3}, which yields a 68% upper bound of 1+w<0.0014. Therefore,\nif the current era of accelerated expansion happens to have the same equation\nof state as inflation during the observable epoch, then current and upcoming\ncosmological observations will not be able to detect that w$\\neq$-1. This\nprovides a cautionary tale for drawing conclusions about the nature of dark\nenergy on the basis of the non-observation of a deviation from w=-1.\n"}
{"abstract": "  The decision forest model is a recently proposed nonparametric choice model\nthat is capable of representing any discrete choice model and in particular,\ncan be used to represent non-rational customer behavior. In this paper, we\nstudy the problem of finding the assortment that maximizes expected revenue\nunder the decision forest model. This problem is of practical importance\nbecause it allows a firm to tailor its product offerings to profitably exploit\ndeviations from rational customer behavior, but at the same time is challenging\ndue to the extremely general nature of the decision forest model. We approach\nthis problem from a mixed-integer optimization perspective and propose three\ndifferent formulations. We theoretically compare these formulations in\nstrength, and analyze when these formulations are integral in the special case\nof a single tree. We propose a methodology for solving these problems at a\nlarge-scale based on Benders decomposition, and show that the Benders\nsubproblem can be solved efficiently by primal-dual greedy algorithms when the\nmaster solution is fractional for two of our formulations, and in closed form\nwhen the master solution is binary for all of our formulations. Using\nsynthetically generated instances, we demonstrate the practical tractability of\nour formulations and our Benders decomposition approach, and their edge over\nheuristic approaches.\n"}
{"abstract": "  Bell's theorem shows that no hidden-variable model can explain the\nmeasurement statistics of a quantum system shared between two parties, thus\nruling out a classical (local) understanding of nature. In this work we\ndemonstrate that by relaxing the positivity restriction in the hidden-variable\nprobability distribution it is possible to derive quasiprobabilistic Bell\ninequalities whose sharp upper bound is written in terms of a negativity\nwitness of said distribution. This provides an analytic solution for the amount\nof negativity necessary to violate the CHSH inequality by an arbitrary amount,\ntherefore revealing the amount of negativity required to emulate the quantum\nstatistics in a Bell test.\n"}
{"abstract": "  In a previous paper it was shown how to calculate the ground-state energy\ndensity $E$ and the $p$-point Green's functions $G_p(x_1,x_2,...,x_p)$ for the\n$PT$-symmetric quantum field theory defined by the Hamiltonian density\n$H=\\frac{1}{2}(\\nabla\\phi)^2+\\frac{1}{2}\\phi^2(i\\phi)^\\varepsilon$ in\n$D$-dimensional Euclidean spacetime, where $\\phi$ is a pseudoscalar field. In\nthis earlier paper $E$ and $G_p(x_1,x_2,...,x_p)$ were expressed as\nperturbation series in powers of $\\varepsilon$ and were calculated to first\norder in $\\varepsilon$. (The parameter $\\varepsilon$ is a measure of the\nnonlinearity of the interaction rather than a coupling constant.) This paper\nextends these perturbative calculations to the Euclidean Lagrangian $L=\n\\frac{1}{2}(\\nabla\\phi)^2+\\frac{1}{2}\\mu^2\\phi^2+\\frac{1}{2}\ng\\mu_0^2\\phi^2\\big(i\\mu_0^{1-D/2}\\phi\\big)^\\varepsilon-iv\\phi$, which now\nincludes renormalization counterterms that are linear and quadratic in the\nfield $\\phi$. The parameter $g$ is a dimensionless coupling strength and\n$\\mu_0$ is a scaling factor having dimensions of mass. Expressions are given\nfor the one-, two, and three-point Green's functions, and the renormalized\nmass, to higher-order in powers of $\\varepsilon$ in $D$ dimensions ($0\\leq\nD\\leq2$). Renormalization is performed perturbatively to second order in\n$\\varepsilon$ and the structure of the Green's functions is analyzed in the\nlimit $D\\to 2$. A sum of the most divergent terms is performed to {\\it all}\norders in $\\varepsilon$. Like the Cheng-Wu summation of leading logarithms in\nelectrodynamics, it is found here that leading logarithmic divergences combine\nto become mildly algebraic in form. Future work that must be done to complete\nthe perturbative renormalization procedure is discussed.\n"}
{"abstract": "  We propose a nanoscale rotor embedded between two ferromagnetic electrodes\nthat is driven by spin injection. The spin-rotation coupling allows this\nnanorotor to continuously receive angular momentum from an injected spin under\nsteady current flow between ferromagnetic electrodes in an antiparallel\nmagnetization configuration. We develop a quantum theory of this angular\nmomentum transfer and show that a relaxation process from a precession state\ninto a sleeping top state is crucial for the efficient driving of the nanorotor\nby solving the master equation. Our work clarifies a general strategy for\nefficient driving of a nanorotor.\n"}
{"abstract": "  The control of far-from-equilibrium physical systems, including active\nmaterials, has emerged as an important area for the application of\nreinforcement learning (RL) strategies to derive control policies for physical\nsystems. In active materials, non-linear dynamics and long-range interactions\nbetween particles prohibit closed-form descriptions of the system's dynamics\nand prevent explicit solutions to optimal control problems. Due to fundamental\nchallenges in solving for explicit control strategies, RL has emerged as an\napproach to derive control strategies for far-from-equilibrium active matter\nsystems. However, an important open question is how the mathematical structure\nand the physical properties of the active matter systems determine the\ntractability of RL for learning control policies. In this work, we show that RL\ncan only find good strategies to the canonical active matter task of mixing for\nsystems that combine attractive and repulsive particle interactions. Using\nmathematical results from dynamical systems theory, we relate the availability\nof both interaction types with the existence of hyperbolic dynamics and the\nability of RL to find homogeneous mixing strategies. In particular, we show\nthat for drag-dominated translational-invariant particle systems, hyperbolic\ndynamics and, therefore, mixing requires combining attractive and repulsive\ninteractions. Broadly, our work demonstrates how fundamental physical and\nmathematical properties of dynamical systems can enable or constrain\nreinforcement learning-based control.\n"}
{"abstract": "  Optical coherence tomography (OCT) is a prevalent imaging technique for\nretina. However, it is affected by multiplicative speckle noise that can\ndegrade the visibility of essential anatomical structures, including blood\nvessels and tissue layers. Although averaging repeated B-scan frames can\nsignificantly improve the signal-to-noise-ratio (SNR), this requires longer\nacquisition time, which can introduce motion artifacts and cause discomfort to\npatients. In this study, we propose a learning-based method that exploits\ninformation from the single-frame noisy B-scan and a pseudo-modality that is\ncreated with the aid of the self-fusion method. The pseudo-modality provides\ngood SNR for layers that are barely perceptible in the noisy B-scan but can\nover-smooth fine features such as small vessels. By using a fusion network,\ndesired features from each modality can be combined, and the weight of their\ncontribution is adjustable. Evaluated by intensity-based and structural\nmetrics, the result shows that our method can effectively suppress the speckle\nnoise and enhance the contrast between retina layers while the overall\nstructure and small blood vessels are preserved. Compared to the single\nmodality network, our method improves the structural similarity with low noise\nB-scan from 0.559 +\\- 0.033 to 0.576 +\\- 0.031.\n"}
{"abstract": "  Hyperspectral image (HSI) with narrow spectral bands can capture rich\nspectral information, but it sacrifices its spatial resolution in the process.\nMany machine-learning-based HSI super-resolution (SR) algorithms have been\nproposed recently. However, one of the fundamental limitations of these\napproaches is that they are highly dependent on image and camera settings and\ncan only learn to map an input HSI with one specific setting to an output HSI\nwith another. However, different cameras capture images with different spectral\nresponse functions and bands numbers due to the diversity of HSI cameras.\nConsequently, the existing machine-learning-based approaches fail to learn to\nsuper-resolve HSIs for a wide variety of input-output band settings. We propose\na single Meta-Learning-Based Super-Resolution (MLSR) model, which can take in\nHSI images at an arbitrary number of input bands' peak wavelengths and generate\nSR HSIs with an arbitrary number of output bands' peak wavelengths. We leverage\nNTIRE2020 and ICVL datasets to train and validate the performance of the MLSR\nmodel. The results show that the single proposed model can successfully\ngenerate super-resolved HSI bands at arbitrary input-output band settings. The\nresults are better or at least comparable to baselines that are separately\ntrained on a specific input-output band setting.\n"}
{"abstract": "  We study the surface elastic response of pure Ni, the random alloy FeNiCr and\nan average FeNiCr alloy in terms of the surface lattice Green's function. We\npropose a scheme for computing per-site Green's function and study their\nper-site variations. The average FeNiCr alloy accurately reproduces the mean\nGreen's function of the full random alloy. Variation around this mean is\nlargest near the edge of the surface Brillouin-zone and decays as $q^{-2}$ with\nwavevector $q$ towards the $\\Gamma$-point. We also present expressions for the\ncontinuum surface Green's function of anisotropic solids of finite and infinite\nthickness and show that the atomistic Green's function approaches continuum\nnear the $\\Gamma$-point. Our results are a first step towards efficient contact\ncalculations and Peierls-Nabarro type models for dislocations in high-entropy\nalloys.\n"}
{"abstract": "  We present a study of optical and near-infrared (NIR) spectra along with the\nlight curves of SN 2013ai. These data range from discovery until 380 days after\nexplosion. SN 2013ai is a fast declining type II supernova (SN II) with an\nunusually long rise time; $18.9\\pm2.7$d in $V$ band and a bright $V$ band peak\nabsolute magnitude of $-18.7\\pm0.06$ mag. The spectra are dominated by hydrogen\nfeatures in the optical and NIR. The spectral features of SN 2013ai are unique\nin their expansion velocities, which when compared to large samples of SNe II\nare more than 1,000 kms faster at 50 days past explosion. In addition, the long\nrise time of the light curve more closely resembles SNe IIb rather than SNe II.\nIf SN 2013ai is coeval with a nearby compact cluster we infer a progenitor ZAMS\nmass of $\\sim$17 M$_\\odot$. After performing light curve modeling we find that\nSN 2013ai could be the result of the explosion of a star with little hydrogen\nmass, a large amount of synthesized $^{56}$Ni, 0.3-0.4 M$_\\odot$, and an\nexplosion energy of $2.5-3.0\\times10^{51}$ ergs. The density structure and\nexpansion velocities of SN 2013ai are similar to that of the prototypical SN\nIIb, SN 1993J. However, SN 2013ai shows no strong helium features in the\noptical, likely due to the presence of a dense core that prevents the majority\nof $\\gamma$-rays from escaping to excite helium. Our analysis suggests that SN\n2013ai could be a link between SNe II and stripped envelope SNe.\n"}
{"abstract": "  The gravitational shock waves have provided crucial insights into\nentanglement structures of black holes in the AdS/CFT correspondence. Recent\nprogress on the soft hair physics suggests that these developments from\nholography may also be applicable to geometries beyond negatively curved\nspacetime. In this work, we derive a remarkably simple thermodynamic relation\nwhich relates the gravitational shock wave to a microscopic area deformation.\nOur treatment is based on the covariant phase space formalism and is applicable\nto any Killing horizon in generic static spacetime which is governed by\narbitrary covariant theory of gravity. The central idea is to probe the\ngravitational shock wave, which shifts the horizon in the $u$ direction, by the\nNoether charge constructed from a vector field which shifts the horizon in the\n$v$ direction. As an application, we illustrate its use for the Gauss-Bonnet\ngravity. We also derive a simplified form of the gravitational scattering\nunitary matrix and show that its leading-order contribution is nothing but the\nexponential of the horizon area: $\\mathcal{U}=\\exp(i \\text{Area})$.\n"}
{"abstract": "  We consider the popular and classical method of alternating projections for\nfinding a point in the intersection of two closed sets. By situating the\nalgorithm in a metric space, equipped only with well-behaved geodesics and\nangles (in the sense of Alexandrov), we are able to highlight the two key\ngeometric ingredients in a standard intuitive analysis of local linear\nconvergence. The first is a transversality-like condition on the intersection;\nthe second is a convexity-like condition on one set: \"uniform approximation by\ngeodesics\".\n"}
{"abstract": "  Image denoising is one of the most critical problems in mobile photo\nprocessing. While many solutions have been proposed for this task, they are\nusually working with synthetic data and are too computationally expensive to\nrun on mobile devices. To address this problem, we introduce the first Mobile\nAI challenge, where the target is to develop an end-to-end deep learning-based\nimage denoising solution that can demonstrate high efficiency on smartphone\nGPUs. For this, the participants were provided with a novel large-scale dataset\nconsisting of noisy-clean image pairs captured in the wild. The runtime of all\nmodels was evaluated on the Samsung Exynos 2100 chipset with a powerful Mali\nGPU capable of accelerating floating-point and quantized neural networks. The\nproposed solutions are fully compatible with any mobile GPU and are capable of\nprocessing 480p resolution images under 40-80 ms while achieving high fidelity\nresults. A detailed description of all models developed in the challenge is\nprovided in this paper.\n"}
{"abstract": "  We extend the theorem of Liouville on integration in finite terms to include\ndilogarithmic integrals. The results provide a necessary and sufficient\ncondition for an element of the base field to have an antiderivative in a field\nextension generated by transcendental elementary functions and dilogarithmic\nintegrals. We also study algebraic independence of certain dilogarithmic\nintegrals.\n"}
{"abstract": "  In this work, an efficient physics-constrained deep learning model is\ndeveloped for solving multiphase flow in 3D heterogeneous porous media. The\nmodel fully leverages the spatial topology predictive capability of\nconvolutional neural networks, and is coupled with an efficient\ncontinuity-based smoother to predict flow responses that need spatial\ncontinuity. Furthermore, the transient regions are penalized to steer the\ntraining process such that the model can accurately capture flow in these\nregions. The model takes inputs including properties of porous media, fluid\nproperties and well controls, and predicts the temporal-spatial evolution of\nthe state variables (pressure and saturation). While maintaining the continuity\nof fluid flow, the 3D spatial domain is decomposed into 2D images for reducing\ntraining cost, and the decomposition results in an increased number of training\ndata samples and better training efficiency. Additionally, a surrogate model is\nseparately constructed as a postprocessor to calculate well flow rate based on\nthe predictions of state variables from the deep learning model. We use the\nexample of CO2 injection into saline aquifers, and apply the\nphysics-constrained deep learning model that is trained from physics-based\nsimulation data and emulates the physics process. The model performs prediction\nwith a speedup of ~1400 times compared to physics-based simulations, and the\naverage temporal errors of predicted pressure and saturation plumes are 0.27%\nand 0.099% respectively. Furthermore, water production rate is efficiently\npredicted by a surrogate model for well flow rate, with a mean error less than\n5%. Therefore, with its unique scheme to cope with the fidelity in fluid flow\nin porous media, the physics-constrained deep learning model can become an\nefficient predictive model for computationally demanding inverse problems or\nother coupled processes.\n"}
{"abstract": "  A Simultaneous Localization and Mapping (SLAM) system must be robust to\nsupport long-term mobile vehicle and robot applications. However, camera and\nLiDAR based SLAM systems can be fragile when facing challenging illumination or\nweather conditions which degrade their imagery and point cloud data. Radar,\nwhose operating electromagnetic spectrum is less affected by environmental\nchanges, is promising although its distinct sensing geometry and noise\ncharacteristics bring open challenges when being exploited for SLAM. % However,\nthere are still open challenges since most existing visual and LiDAR SLAM\nsystems do not operate in bad weathers. This paper studies the use of a\nFrequency Modulated Continuous Wave radar for SLAM in large-scale outdoor\nenvironments. We propose a full radar SLAM system, including a novel radar\nmotion tracking algorithm that leverages radar geometry for reliable feature\ntracking. It also optimally compensates motion distortion and estimates pose by\njoint optimization. Its loop closure component is designed to be simple yet\nefficient for radar imagery by capturing and exploiting structural information\nof the surrounding environment. % while a scheme to reject ambiguous loop\nclosure candidates is also designed specifically for radar. Extensive\nexperiments on three public radar datasets, ranging from city streets and\nresidential areas to countryside and highways, show competitive accuracy and\nreliability performance of the proposed radar SLAM system compared to the\nstate-of-the-art LiDAR, vision and radar methods. The results show that our\nsystem is technically viable in achieving reliable SLAM in extreme weather\nconditions, e.g. heavy snow and dense fog, demonstrating the promising\npotential of using radar for all-weather localization and mapping.\n"}
{"abstract": "  In this technical report, we propose an algorithm, called Lex2vec that\nexploits lexical resources to inject information into word embeddings and name\nthe embedding dimensions by means of knowledge bases. We evaluate the optimal\nparameters to extract a number of informative labels that is readable and has a\ngood coverage for the embedding dimensions.\n"}
{"abstract": "  The dependence of the current density on the phase difference is investigated\nconsidering the layered superconducting structures of a SIS'IS type. To\nsimplify the calculations, the quasiclassical equations for the Green's\nfunctions in a \\textit{t}-representation are derived. An order parameter is\nconsidered as a piecewise constant function. To consider the general case, no\nrestrictions on the dielectric layer transparency and the thickness of the\nintermediate layer are imposed. It was found that a new analytical expression\nfor the current-phase relation can be used with the aim to obtain a number of\npreviously known results arising in particular cases.\n"}
{"abstract": "  It has been unknown in kinetic theory whether the linearized Boltzmann or\nLandau equation with soft potentials admits a spectral gap in the spatially\ninhomogeneous setting. Most of existing works indicate a negative answer\nbecause the spectrum of two linearized self-adjoint collision operators is\naccumulated to the origin in case of soft interactions. In the paper we rather\nprove it in an affirmative way when the space domain is bounded with an inflow\nboundary condition. The key strategy is to introduce a new Hilbert space with\nan exponential weight function that involves the inner product of space and\nvelocity variables and also has the strictly positive upper and lower bounds.\nThe action of the transport operator on such space-velocity dependent weight\nfunction induces an extra non-degenerate relaxation dissipation in large\nvelocity that can be employed to compensate the degenerate spectral gap and\nhence give the exponential decay for solutions in contrast with the\nsub-exponential decay in either the spatially homogeneous case or the case of\ntorus domain. The result reveals a new insight of hypocoercivity for kinetic\nequations with soft potentials in the specified situation.\n"}
{"abstract": "  We derive the lower bound on the absolute scale of lightest neutrino mass for\nnormal hierarchy and inverted hierarchy pattern of light neutrinos by studying\nthe new physics contributions to charged lepton flavour violations in the\nframework of a TeV scale left-right symmetric model. In the model, the fermion\nsector comprises the usual quarks and leptons plus a fermion singlet per\ngeneration and the scalar sector consists of isospin doublets, triplets and a\nbidoublet. The framework allows large light-heavy neutrino mixing where the\nlight neutrino mass formula is governed by a natural type-II seesaw mechanism,\nunlike the generic type-II seesaw dominance which assumes suppressed\nlight-heavy neutrino mixing. We demonstrate how sizeable loop-induced\ncontribution to light neutrino mass is kept under control such that the light\nneutrino mass formula is dominantly explained by the type-II seesaw mechanism.\nWe examine the heavy neutrino contributions with large light-heavy neutrino\nmixing to charged lepton flavour violating processes like $\\mu \\to e \\gamma$,\n$\\mu \\to 3 e$ and $\\mu \\to e$ conversion inside a nucleus. We present a\ncomplementary study between neutrinoless double beta decay and charged lepton\nflavour violation taking into account single beta decay bound, double beta\ndecay bound and cosmology bounds on the sum of light neutrino masses.\n"}
{"abstract": "  How effectively do we adhere to nudges and interventions that help us control\nour online browsing habits? If we have a temporary lapse and disable the\nbehavior change system, do we later resume our adherence, or has the dam\nbroken? In this paper, we investigate these questions through log analyses of\n8,000+ users on HabitLab, a behavior change platform that helps users reduce\ntheir time online. We find that, while users typically begin with\nhigh-challenge interventions, over time they allow themselves to slip into\neasier and easier interventions. Despite this, many still expect to return to\nthe harder interventions imminently: they repeatedly choose to be asked to\nchange difficulty again on the next visit, declining to have the system save\ntheir preference for easy interventions.\n"}
{"abstract": "  Reading text aloud is an important feature for modern computer applications.\nIt not only facilitates access to information for visually impaired people, but\nis also a pleasant convenience for non-impaired users. In this article, the\nstate of the art of speech synthesis is presented separately for\nmel-spectrogram generation and vocoders. It concludes with an overview of\navailable data sets for English and German with a discussion of the\ntransferability of the good speech synthesis results from English to German\nlanguage.\n"}
{"abstract": "  The FCC-ee offers powerful opportunities for direct or indirect evidence for\nphysics beyond the Standard Model, via a combination of high precision\nmeasurements and searches for forbidden and rare processes and feebly coupled\nparticles. A key element of FCC-ee physics program is the measurement of the Z\nlineshape from a total of $5\\times 10^{12}$ Z bosons and a beam-energy\ncalibration with relative uncertainty of $10^{-6}$. With this exceptionally\nlarge event sample, five orders of magnitude larger than that accumulated\nduring the whole LEP1 operation at the Z pole, the defining parameters -\n$m_{\\rm Z}$, $\\Gamma_{\\rm Z}$, $N_\\nu$, $\\sin^2\\theta_{\\rm W}^{\\rm eff}$,\n$\\alpha_{\\rm S}(m_{\\rm Z}^2)$, and $\\alpha_{\\rm QED}(m^2_{\\rm Z})$ - can be\nextracted with a leap in accuracy of up to two orders of magnitude with respect\nto the current state of the art. The ultimate goal that experimental and theory\nsystematic errors match the statistical accuracy (4\\,keV on the Z mass and\nwidth, $3\\times 10^{-6}$ on $\\sin^2\\theta_{\\rm W}^{\\rm eff}$, a relative\n$3\\times 10^{-5}$ on $\\alpha_{\\rm QED}$, and less than 0.0001 on $\\alpha_{\\rm\nS}$) leads to highly demanding requirements on collider operation, beam\ninstrumentation, detector design, computing facilities, theoretical\ncalculations, and Monte Carlo event generators. Such precise measurements also\ncall for innovative analysis methods, which require a joint effort and\nunderstanding between theorists, experimenters, and accelerator teams.\n"}
{"abstract": "  Reliable empirical models such as those used in software effort estimation or\ndefect prediction are inherently dependent on the data from which they are\nbuilt. As demands for process and product improvement continue to grow, the\nquality of the data used in measurement and prediction systems warrants\nincreasingly close scrutiny. In this paper we propose a taxonomy of data\nquality challenges in empirical software engineering, based on an extensive\nreview of prior research. We consider current assessment techniques for each\nquality issue and proposed mechanisms to address these issues, where available.\nOur taxonomy classifies data quality issues into three broad areas: first,\ncharacteristics of data that mean they are not fit for modeling; second, data\nset characteristics that lead to concerns about the suitability of applying a\ngiven model to another data set; and third, factors that prevent or limit data\naccessibility and trust. We identify this latter area as of particular need in\nterms of further research.\n"}
{"abstract": "  In this work, we studied the impact of galaxy morphology on photometric\nredshift (photo-$z$) probability density functions (PDFs). By including galaxy\nmorphological parameters like the radius, axis-ratio, surface brightness and\nthe S\\'ersic index in addition to the $ugriz$ broadbands as input parameters,\nwe used the machine learning photo-$z$ algorithm ANNz2 to train and test on\ngalaxies from the Canada-France-Hawaii Telescope Stripe-82 (CS82) Survey.\nMetrics like the continuous ranked probability score (CRPS), probability\nintegral transform (PIT), Bayesian odds parameter, and even the width and\nheight of the PDFs were evaluated, and the results were compared when different\nnumber of input parameters were used during the training process. We find\nimprovements in the CRPS and width of the PDFs when galaxy morphology has been\nadded to the training, and the improvement is larger especially when the number\nof broadband magnitudes are lacking.\n"}
{"abstract": "  We examine an intriguing possibility that a single field is responsible for\nboth inflation and dark matter, focussing on the minimal set-up where inflation\nis driven by a scalar coupling to curvature. We study in detail the reheating\nprocess in this framework, which amounts mainly to particle production in a\nquartic potential, and distinguish thermal and non-thermal dark matter options.\nIn the non-thermal case, the reheating is impeded by backreaction and\nrescattering, making this possibility unrealistic. On the other hand,\nthermalized dark matter is viable, yet the unitarity bound forces the inflaton\nmass into a narrow window close to half the Higgs mass.\n"}
{"abstract": "  We consider a nonnegative potential $W:\\mathbb{R}^2\\rightarrow\\mathbb{R}$\ninvariant under the action of the rotation group $C_N$ of the regular polygon\nwith $N$ sides, $N\\geq 3$. We assume that $W$ has $N$ nondegenerate zeros and\nprove the existence of a $N$-junction solution to the vector Allen-Cahn\nequation. The proof is variational and is based on sharp lower and upper bounds\nfor the energy and on a new pointwise estimate for vector minimizers.\n"}
{"abstract": "  Deep generative models are known to be able to model arbitrary probability\ndistributions. Among these, a recent deep generative model, dubbed sliceGAN,\nproposed a new way of using the generative adversarial network (GAN) to capture\nthe micro-structural characteristics of a two-dimensional (2D) slice and\ngenerate three-dimensional (3D) volumes with similar properties. While 3D\nmicrographs are largely beneficial in simulating diverse material behavior,\nthey are often much harder to obtain than their 2D counterparts. Hence,\nsliceGAN opens up many interesting directions of research by learning the\nrepresentative distribution from 2D slices, and transferring the learned\nknowledge to generate arbitrary 3D volumes. However, one limitation of sliceGAN\nis that latent space steering is not possible. Hence, we combine sliceGAN with\nAdaIN to endow the model with the ability to disentangle the features and\ncontrol the synthesis.\n"}
{"abstract": "  Starting with unitary quantum dynamics, we investigate how to add quantum\nmeasurements. Quantum measurements have four essential components: the\nfurcation, the witness production, an alignment projection, and the actual\nchoice decision. The first two components still lie in the domain of unitary\nquantum dynamics. The decoherence concept explains the third contribution. It\ncan be based on the requirement that witnesses reaching the end of time on the\nwave function side and the conjugate one have to be identical. In this way, it\nalso stays within the quantum dynamics domain. The surjection hypothesis\nexplains the actual choice decision. It is based on a two boundary\ninterpretation applied to the complete quantum universe. It offers a simple way\nto reduce these seemingly random projections to purely deterministic unitary\nquantum dynamics, eliminating the measurement problem.\n"}
{"abstract": "  It has been known for decades that a magnetic field can deflect phonons as\nthey flow in response to a thermal gradient, producing a thermal Hall effect.\nSeveral recent experiments have revealed ratios of the phonon Hall conductivity\nto the phonon longitudinal conductivity in oxide dielectrics that are larger\nthan $10^{-3}$ when phonon mean-free-paths exceed phonon wavelengths. At the\nsame time $\\kappa_{H}/\\kappa_{L}$ is not strongly temperature dependent. We\nargue that these two properties together imply a mechanism related to phonon\nscattering from defects that break time-reversal symmetry, and we show that\nLorentz forces acting on charged defects produce substantial skew-scattering\namplitudes, and related thermal Hall effects that are consistent with recent\nobservations.\n"}
{"abstract": "  The ongoing exponential rise in recording capacity calls for new approaches\nfor analysing and interpreting neural data. Effective dimensionality has\nemerged as an important property of neural activity across populations of\nneurons, yet different studies rely on different definitions and\ninterpretations of this quantity. Here we focus on intrinsic and embedding\ndimensionality, and discuss how they might reveal computational principles from\ndata. Reviewing recent works, we propose that the intrinsic dimensionality\nreflects information about the latent variables encoded in collective activity,\nwhile embedding dimensionality reveals the manner in which this information is\nprocessed. We conclude by highlighting the role of network models as an ideal\nsubstrate for testing more specifically various hypotheses on the computational\nprinciples reflected through intrinsic and embedding dimensionality.\n"}
{"abstract": "  We introduce three new analytical and semi-analytical tools that allow one to\ndetermine the topological character of impurity Shiba chains. The analytical\nmethods are based on calculating the effective Green's function of an infinite\nembedded chain using the T-matrix formalism and describing the chain as a {\\it\nline impurity}. We thus provide a solution to the longstanding size-effects\nproblem affecting the only general alternative method, the numerical\ntight-binding analysis. As an example we consider a chain of magnetic\nimpurities deposited on an s-wave superconducting substrate with Rashba\nspin-orbit and we calculate its topological phase diagram as a function of the\nmagnetic impurity strength and the chemical potential. We find a perfect\nagreement between all our new techniques and a numerical analysis.\n"}
{"abstract": "  We show that the $\\imath$Hall algebra of the Jordan quiver is a polynomial\nring in infinitely many generators and obtain transition relations among\nseveral generating sets. We establish a ring isomorphism from this $\\imath$Hall\nalgebra to the ring of symmetric functions in two parameters $t, \\theta$, which\nmaps the $\\imath$Hall basis to a class of (modified) inhomogeneous\nHall-Littlewood ($\\imath$HL) functions. The (modified) $\\imath$HL functions\nadmit a formulation via raising and lowering operators. We formulate and prove\nPieri rules for (modified) $\\imath$HL functions. The modified $\\imath$HL\nfunctions specialize at $\\theta=0$ to the modified HL functions; they\nspecialize at $\\theta=1$ to the deformed universal characters of type C, which\nfurther specialize at $(t=0, \\theta =1)$ to the universal characters of type C.\n"}
{"abstract": "  In this article we propose a distributed collision avoidance scheme for\nmulti-agent unmanned aerial vehicles(UAVs) based on nonlinear model predictive\ncontrol (NMPC),where other agents in the system are considered as dynamic\nobstacles with respect to the ego agent. Our control scheme operates at a low\nlevel and commands roll, pitch and thrust signals at a high frequency, each\nagent broadcasts its predicted trajectory to the other ones, and we propose an\nobstacle prioritization scheme based on the shared trajectories to allow\nup-scaling of the system. The NMPC problem is solved using an ad hoc solver\nwhere PANOC is combined with an augmented Lagrangian method to compute\ncollision-free trajectories. We evaluate the proposed scheme in several\nchallenging laboratory experiments for up to ten aerial agents, in dense aerial\nswarms.\n"}
{"abstract": "  Deep learning has shown great promise for CT image reconstruction, in\nparticular to enable low dose imaging and integrated diagnostics. These merits,\nhowever, stand at great odds with the low availability of diverse image data\nwhich are needed to train these neural networks. We propose to overcome this\nbottleneck via a deep reinforcement learning (DRL) approach that is integrated\nwith a style-transfer (ST) methodology, where the DRL generates the anatomical\nshapes and the ST synthesizes the texture detail. We show that our method bears\nhigh promise for generating novel and anatomically accurate high resolution CT\nimages at large and diverse quantities. Our approach is specifically designed\nto work with even small image datasets which is desirable given the often low\namount of image data many researchers have available to them.\n"}
{"abstract": "  While transient photoluminescence measurements are a very popular tool to\nmonitor the charge-carrier dynamics in the field of halide perovskite\nphotovoltaics, interpretation of data obtained on multilayer samples is highly\nchallenging due to the superposition of various effects that modulate the\ncharge-carrier concentration in the perovskite layer and thereby the measured\nPL. These effects include bulk and interfacial recombination, charge transfer\nto electron or hole transport layers and capacitive charging or discharging.\nHere, numerical simulations with Sentauraus TCAD, analytical solutions and\nexperimental data with a dynamic range of ~7 orders of magnitude on a variety\nof different sample geometries from perovskite films on glass to full devices\nare combined to present an improved understanding of this method. A\npresentation of the decay time of the TPL decay that follows from taking the\nderivative of the photoluminescence at every time is proposed. Plotting this\ndecay time as a function of the time-dependent quasi-Fermi level splitting\nenables distinguishing between the different contributions of radiative and\nnon-radiative recombination as well as charge extraction and capacitive effects\nto the decay.\n"}
{"abstract": "  Using the Large Binocular Telescope (LBT)/Multi-Object Dual Spectrograph\n(MODS), we have obtained optical spectroscopy of one of the most metal-poor\ndwarf star-forming galaxies (SFG) in the local Universe, J2229+2725. This\ngalaxy with a redshift z=0.0762 was selected from the Data Release 16 (DR16) of\nthe Sloan Digital Sky Survey (SDSS). Its properties derived from the LBT\nobservations are most extreme among SFGs in several ways. Its oxygen abundance\n12+logO/H=7.085+/-0.031 is among the lowest ever observed for a SFG. With its\nvery low metallicity, an absolute magnitude Mg=-16.39 mag, a low stellar mass\nMstar=9.1x10^6 Msun and a very low mass-to-light ratio Mstar/Lg~0.0166 (in\nsolar units), J2229+2725 deviates strongly from the luminosity-metallicity\nrelation defined by the bulk of the SFGs in the SDSS. J2229$+$2725 has a very\nhigh specific star-formation rate sSFR~75 Gyr^-1, indicating very active\nongoing star formation. Three other features of J2229+2725 are most striking,\nbeing the most extreme among lowest-metallicity SFGs: 1) a ratio\nO32=I([OIII]5007)/I([OII]3727)~53, 2) an equivalent width of the Hbeta emission\nline EW(Hbeta) of 577A, and 3) an electron number density of ~1000 cm^-3. These\nproperties imply that the starburst in J2229+2725 is very young. Using the\nextremely high O32 in J2229+2725, we have improved the strong-line calibration\nfor the determination of oxygen abundances in the most metal-deficient\ngalaxies, in the range 12 + logO/H<7.3.\n"}
{"abstract": "  Understanding the origins of unconventional superconductivity has been a\nmajor focus of condensed matter physics for many decades. While many questions\nremain unanswered, experiments have found that the systems with the highest\ncritical temperatures tend to be layered materials where superconductivity\noccurs in two-dimensional (2D) structures. However, to what extent the\nremarkable stability of these strongly correlated 2D superfluids is related to\ntheir reduced dimensionality is still an open question. In this work, we use\ndilute gases of ultracold fermionic atoms as a model system to directly observe\nthe influence of dimensionality on strongly interacting fermionic superfluids.\nWe achieve this by measuring the superfluid gap of a strongly correlated\nquasi-2D Fermi gas over a wide range of interaction strengths and comparing the\nresults to recent measurements in 3D Fermi gases. We find that the superfluid\ngap follows the same universal function of the interaction strength in both\nsystems, which suggests that there is no inherent difference in the stability\nof fermionic superfluidity between two- and three-dimensional quantum gases.\nFinally, we compare our data to results from solid state systems and find a\nsimilar relation between the interaction strength and the gap for a wide range\nof two- and three-dimensional superconductors.\n"}
{"abstract": "  Supervised methods for object delineation in remote sensing require labeled\nground-truth data. Gathering sufficient high quality ground-truth data is\ndifficult, especially when targets are of irregular shape or difficult to\ndistinguish from background or neighboring objects. Tree crown delineation\nprovides key information from remote sensing images for forestry, ecology, and\nmanagement. However, tree crowns in remote sensing imagery are often difficult\nto label and annotate due to irregular shape, overlapping canopies, shadowing,\nand indistinct edges. There are also multiple approaches to annotation in this\nfield (e.g., rectangular boxes vs. convex polygons) that further contribute to\nannotation imprecision. However, current evaluation methods do not account for\nthis uncertainty in annotations, and quantitative metrics for evaluation can\nvary across multiple annotators. In this paper, we address these limitations by\ndeveloping an adaptation of the Rand index for weakly-labeled crown delineation\nthat we call RandCrowns. Our new RandCrowns evaluation metric provides a method\nto appropriately evaluate delineated tree crowns while taking into account\nimprecision in the ground-truth delineations. The RandCrowns metric\nreformulates the Rand index by adjusting the areas over which each term of the\nindex is computed to account for uncertain and imprecise object delineation\nlabels. Quantitative comparisons to the commonly used intersection over union\nmethod shows a decrease in the variance generated by differences among multiple\nannotators. Combined with qualitative examples, our results suggest that the\nRandCrowns metric is more robust for scoring target delineations in the\npresence of uncertainty and imprecision in annotations that are inherent to\ntree crown delineation.\n"}
{"abstract": "  This report classifies emission inhomogeneities that manifest in InGaN\nquantum well blue light-emitting diodes grown by plasma-assisted molecular beam\nepitaxy on free-standing GaN substrates. By a combination of spatially resolved\nelectroluminescence and cathodoluminescence measurements, atomic force\nmicroscopy, scanning electron microscopy and hot wet KOH etching, the\nidentified inhomogeneities are found to fall in four categories. Labeled here\nas type I through IV, they are distinguishable by their size, density, energy,\nintensity, radiative and electronic characteristics and chemical etch pits\nwhich correlates them with dislocations. Type I exhibits a blueshift of about\n120 meV for the InGaN quantum well emission attributed to a perturbation of the\nactive region, which is related to indium droplets that form on the surface in\nthe metal-rich InGaN growth condition. Specifically, we attribute the blueshift\nto a decreased growth rate of and indium incorporation in the InGaN quantum\nwells underneath the droplet which is postulated to be the result of reduced\nincorporated N species due to increased N$_{2}$ formation. The location of\ndroplets are correlated with mixed type dislocations for type I defects. Types\nII through IV are due to screw dislocations, edge dislocations, and dislocation\nbunching, respectively, and form dark spots due to leakage current and\nnonradiative recombination.\n"}
{"abstract": "  We consider the problem of estimating a particular type of linear\nnon-Gaussian model. Without resorting to the overcomplete Independent Component\nAnalysis (ICA), we show that under some mild assumptions, the model is uniquely\nidentified by a hybrid method. Our method leverages the advantages of\nconstraint-based methods and independent noise-based methods to handle both\nconfounded and unconfounded situations. The first step of our method uses the\nFCI procedure, which allows confounders and is able to produce asymptotically\ncorrect results. The results, unfortunately, usually determine very few\nunconfounded direct causal relations, because whenever it is possible to have a\nconfounder, it will indicate it. The second step of our procedure finds the\nunconfounded causal edges between observed variables among only those adjacent\npairs informed by the FCI results. By making use of the so-called Triad\ncondition, the third step is able to find confounders and their causal\nrelations with other variables. Afterward, we apply ICA on a notably smaller\nset of graphs to identify remaining causal relationships if needed. Extensive\nexperiments on simulated data and real-world data validate the correctness and\neffectiveness of the proposed method.\n"}
{"abstract": "  Deceptive news posts shared in online communities can be detected with NLP\nmodels, and much recent research has focused on the development of such models.\nIn this work, we use characteristics of online communities and authors -- the\ncontext of how and where content is posted -- to explain the performance of a\nneural network deception detection model and identify sub-populations who are\ndisproportionately affected by model accuracy or failure. We examine who is\nposting the content, and where the content is posted to. We find that while\nauthor characteristics are better predictors of deceptive content than\ncommunity characteristics, both characteristics are strongly correlated with\nmodel performance. Traditional performance metrics such as F1 score may fail to\ncapture poor model performance on isolated sub-populations such as specific\nauthors, and as such, more nuanced evaluation of deception detection models is\ncritical.\n"}
{"abstract": "  We consider skew-product maps over circle rotations $x\\mapsto x+\\alpha$ (mod\n1) with factors that take values in SL(2,R). This includes maps of almost\nMathieu type. In numerical experiments, with $\\alpha$ the inverse golden mean,\nFibonacci iterates of maps from the almost Mathieu family exhibit asymptotic\nscaling behavior that is reminiscent of critical phase transitions. In a\nrestricted setup that is characterized by a symmetry, we prove that critical\nbehavior indeed occurs and is universal in an open neighborhood of the almost\nMathieu family. This behavior is governed by a periodic orbit of a\nrenormalization transformation. An extension of this transformation is shown to\nhave a second periodic orbit as well, and we present some evidence that this\norbit attracts supercritical almost Mathieu maps.\n"}
{"abstract": "  This paper proposes \\emph{knowledge-based paraonoia search} (KBPS) to find\nforced wins during trick-taking in the card game Skat; for some one of the most\ninteresting card games for three players. It combines efficient partial\ninformation game-tree search with knowledge representation and reasoning. This\nworst-case analysis, initiated after a small number of tricks, leads to a\nprioritized choice of cards. We provide variants of KBPS for the declarer and\nthe opponents, and an approximation to find a forced win against most worlds in\nthe belief space. Replaying thousands of expert games, our evaluation indicates\nthat the AIs with the new algorithms perform better than humans in their play,\nachieving an average score of over 1,000 points in the agreed standard for\nevaluating Skat tournaments, the extended Seeger system.\n"}
{"abstract": "  Learning the structure of dependence relations between variables is a\npervasive issue in the statistical literature. A directed acyclic graph (DAG)\ncan represent a set of conditional independences, but different DAGs may encode\nthe same set of relations and are indistinguishable using observational data.\nEquivalent DAGs can be collected into classes, each represented by a partially\ndirected graph known as essential graph (EG). Structure learning directly\nconducted on the EG space, rather than on the allied space of DAGs, leads to\ntheoretical and computational benefits. Still, the majority of efforts in the\nliterature has been dedicated to Gaussian data, with less attention to methods\ndesigned for multivariate categorical data. We then propose a Bayesian\nmethodology for structure learning of categorical EGs. Combining a constructive\nparameter prior elicitation with a graph-driven likelihood decomposition, we\nderive a closed-form expression for the marginal likelihood of a categorical EG\nmodel. Asymptotic properties are studied, and an MCMC sampler scheme developed\nfor approximate posterior inference. We evaluate our methodology on both\nsimulated scenarios and real data, with appreciable performance in comparison\nwith state-of-the-art methods.\n"}
{"abstract": "  We focus on a convolutional neural network (CNN), which has recently been\nutilized for fluid flow analyses, from the perspective on the influence of\nvarious operations inside it by considering some canonical regression problems\nwith fluid flow data. We consider two types of CNN-based fluid flow analyses;\n1. CNN metamodeling and 2. CNN autoencoder. For the first type of CNN with\nadditional scalar inputs, which is one of the common forms of CNN for fluid\nflow analysis, we investigate the influence of input placements in the CNN\ntraining pipeline. As an example, estimation of force coefficients of an\ninclined flat plate and two side-by-side cylinders in laminar flows is\nconsidered. We find that care should be taken for the placement of additional\nscalar inputs depending on the problem setting and the complexity of flows that\nusers handle. We then discuss the influence of various parameters and\noperations on the CNN performance, with the utilization of autoencoder (AE). A\ntwo-dimensional turbulence is considered for the demonstration of AE. The\nresults of AE highly rely on the decaying nature. Investigation on the\ninfluence of padding operation at a convolutional layer is also performed. The\nzero padding shows reasonable ability compared to other methods which account\nfor the boundary conditions assumed in the numerical data. Moreover, the effect\nof the dimensional reduction/extension methods inside CNN is also examined. The\nCNN model is robust against the difference in dimension reduction operations,\nwhile it is sensitive to the dimensional extension methods. The findings of\nthis paper will help us better design a CNN architecture for practical fluid\nflow analysis\n"}
{"abstract": "  Analytical formulas are presented for simplified but useful qubit geometries\nthat predict surface dielectric loss when its thickness is much less than the\nmetal thickness, the limiting case needed for real devices. These formulas can\nthus be used to precisely predict loss and optimize the qubit layout.\nSurprisingly, a significant fraction of surface loss comes from the small wire\nthat connects the Josephson junction to the qubit capacitor. Tapering this wire\nis shown to significantly lower its loss. Also predicted are the size and\ndensity of the two-level state (TLS) spectrum from individual surface\ndissipation sites.\n"}
{"abstract": "  As subjective artistic creations, artistic paintings carry emotion of their\ncreators. Emotions expressed in paintings and emotion aroused in spectators by\npaintings are two kinds of emotions that scholars have paid attention to.\nTraditional studies on emotions expressed by paintings are mainly conducted\nfrom qualitative perspectives, with neither quantitative output on the\nemotional values of a painting, nor exploration of trends in the expression of\nemotion in art history. In this research we threat facial expressions in\npaintings as an artistic characteristics of art history and employ cognitive\ncomputation technology to identify the facial emotions in paintings and to\ninvestigate the quantitative measures of paintings from three emotion-related\naspects: the spatial and temporal patterns of painting emotions in art history,\nthe gender difference on the emotion of paintings and the color preference\nassociated with emotions. We discovered that the emotion of happiness has a\ngrowing trend from ancient to modern times in paintings history, and men and\nwomen have different facial expressions patterns along time. As for color\npreference, artists with different culture backgrounds had similar association\npreferences between colors and emotions.\n"}
{"abstract": "  We present the first study on the amplification of magnetic fields by the\nturbulent dynamo in the highly subsonic regime, with Mach numbers ranging from\n$10^{-3}$ to $0.4$. We find that for the lower Mach numbers the saturation\nefficiency of the dynamo, $(E_{\\mathrm{mag}}/E_{\\mathrm{kin}})_{\\mathrm{sat}}$,\nincreases as the Mach number decreases. Even in the case when injection of\nenergy is purely through longitudinal forcing modes,\n$(E_{\\mathrm{mag}}/E_{\\mathrm{kin}})_{\\mathrm{sat}}$ $\\gtrsim 10^{-2}$ at a\nMach number of $10^{-3}$. We apply our results to magnetic field amplification\nin the early Universe and predict that a turbulent dynamo can amplify\nprimordial magnetic fields to $\\gtrsim$ $10^{-16}$ Gauss on scales up to 0.1 pc\nand $\\gtrsim$ $10^{-13}$ Gauss on scales up to 100 pc. This produces fields\ncompatible with lower limits of the intergalactic magnetic field inferred from\nblazar $\\gamma$-ray observations.\n"}
{"abstract": "  The theory of standard pentads is the theory aims to construct a graded Lie\nalgebra whose local part consists of a given Lie algebra and its\nrepresentation. In other words, using standard pentads, we can embed given Lie\nalgebra and its representation into a larger graded Lie algebra. As special\ncases of Lie algebras associated with standard pentads, we have the notion of\nPC Lie algebras. Our aim of this paper is to show that the class of PC Lie\nalgebras contains the class of Kac-Moody algebras, that is, to show that the\nnotion of PC Lie algebras is an extension of Kac-Moody algebras.\n"}
{"abstract": "  Power packet dispatching system has been proposed for smart power management\nin the form of discretized packet. In this paper, we discuss the routing\noptimization of power packets on the network of power routers. We propose a\ncost metric for the power packet delivery by circuit analysis of the router\nnetwork. Using the metric, we formulate the optimization problem as a general\nshortest path problem from a source node to a load node. The result of\nnumerical simulations shows that the proposed algorithm can allocate\ndistributed power sources to load demands and identify the optimal path for the\npower delivery.\n"}
{"abstract": "  Multidimensional arrays (NDArrays) are a central abstraction in modern\nscientific computing environments. Unfortunately, they can make reasoning about\nprograms harder as the number of different array shapes used in an execution of\na program is usually very large, and they rarely appear explicitly in program\ntext. To make things worse, many operators make implicit assumptions about the\nshapes of their inputs: array addition is commonly enriched with broadcasting\nsemantics, while matrix multiplication assumes that the lengths of contracted\ndimensions are equal. Because precise reasoning about shapes is crucial to\nwrite correct programs using NDArrays, and because shapes are often hard to\ninfer from a quick glance at the program, we developed Tensors Fitting\nPerfectly, a static analysis tool that reasons about NDArray shapes in Swift\nfor TensorFlow programs by synthesizing a set of shape constraints from an\nabstract interpretation of the program. It can both (1) check for possible\ninconsistencies, and (2) provide direct insights about the shapes of\nintermediate values appearing in the program, including via a mechanism called\nshape holes. The static analysis works in concert with optional runtime\nassertions to improve the productivity of program authors.\n"}
{"abstract": "  We say that the families $\\mathcal F_1,\\ldots, \\mathcal F_{s+1}$ of\n$k$-element subsets of $[n]$ are cross-dependent if there are no pairwise\ndisjoint sets $F_1,\\ldots, F_{s+1}$, where $F_i\\in \\mathcal F_i$ for each $i$.\nThe rainbow version of the Erd\\H os Matching Conjecture due to Aharoni and\nHoward and independently to Huang, Loh and Sudakov states that $\\min_{i}\n|\\mathcal F_i|\\le \\max\\big\\{{n\\choose k}-{n-s\\choose k}, {(s+1)k-1\\choose\nk}\\big\\}$. In this paper, we prove this conjecture for $n>3e(s+1)k$ and\n$s>10^7$. One of the main tools in the proof is a concentration inequality due\nto Frankl and the author.\n"}
{"abstract": "  The task of inducing, via continuous static state-feedback, an asymptotically\nstable heteroclinic orbit in a nonlinear control system is considered in this\npaper. The main motivation comes from the problem of ensuring convergence to a\nso-called point-to-point maneuver in an underactuated mechanical system.\nNamely, to a smooth curve in its state--control space, which is consistent with\nthe system dynamics and connects two (linearly) stabilizable equilibrium\npoints. The proposed method uses a particular parameterization, together with a\nstate projection onto the maneuver as to combine two linearization techniques\nfor this purpose: the Jacobian linearization at the equilibria on the\nboundaries and a transverse linearization along the orbit. This allows for the\ncomputation of stabilizing control gains offline by solving a semidefinite\nprogramming problem. The resulting nonlinear controller, which simultaneously\nasymptotically stabilizes both the orbit and the final equilibrium, is\ntime-invariant, locally Lipschitz continuous, requires no switching, and has a\nfamiliar feedforward plus feedback--like structure. The method is also\ncomplemented by synchronization function--based arguments for planning such\nmaneuvers for mechanical systems with one degree of underactuation. Numerical\nsimulations of the non-prehensile manipulation task of a ball rolling between\ntwo points upon the \"butterfly\" robot demonstrates the efficacy of the\nsynthesis.\n"}
{"abstract": "  Currently, the world is in the midst of a severe global pandemic, which has\naffected all aspects of people's lives. As a result, there is a deluge of\nCOVID-related digital media articles published in the United States, due to the\ndisparate effects of the pandemic. This large volume of information is\ndifficult to consume by the audience in a reasonable amount of time. In this\npaper, we develop a Natural Language Processing (NLP) pipeline that is capable\nof automatically distilling various digital articles into manageable pieces of\ninformation, while also modelling the progression topics discussed over time in\norder to aid readers in rapidly gaining holistic perspectives on pressing\nissues (i.e., the COVID-19 pandemic) from a diverse array of sources. We\nachieve these goals by first collecting a large corpus of COVID-related\narticles during the onset of the pandemic. After, we apply unsupervised and\nsemi-supervised learning procedures to summarize articles, then cluster them\nbased on their similarities using the community detection methods. Next, we\nidentify the topic of each cluster of articles using the BART algorithm.\nFinally, we provide a detailed digital media analysis based on the NLP-pipeline\noutputs and show how the conversation surrounding COVID-19 evolved over time.\n"}
{"abstract": "  Demand response (DR) refers to change in electricity consumption pattern of\ncustomers during on-peak hours in lieu of financial gains to reduce stress on\ndistribution systems. Existing dynamic price models have not provided adequate\nsuccess to price-based demand response (PBDR) programs. It happened as these\nmodels have raised typical socio-economic problems pertaining to cross-subsidy,\nfree-riders, social inequity, assured profit of utilities, financial gains and\ncomfort of customers, etc. This paper presents a new dynamic price model for\nPBDR in distribution systems which aims to overcome some of the above mentioned\nproblems of the existing price models. The main aim of the developed price\nmodel is to overcome the problems of cross-subsidy and free-riders of the\nexisting price models for widespread acceptance, deployment and efficient\nutilization of PBDR programs in contemporary distribution systems. Proposed\nprice model generates demand-linked price signal that imposes different price\nsignals to different customers during on-peak hours and remains static\notherwise. This makes proposed model a class apart from other existing models.\nThe novelty of the proposed model lies in the fact that the financial benefits\nand penalties pertaining to DR are self-adjusted among customers while\npreserving social equity and profit of the utility. Such an ideology has not\nbeen yet addressed in the literature. Detailed investigation of application\nresults on a standard test bench reveals that the proposed model equally cares\nregarding the interests of both customers and utility. For economic assessment,\na comparison of the proposed price model with the existing pricing models is\nalso performed.\n"}
{"abstract": "  A quantum internet connects remote quantum processors that need interact and\nexchange quantum signals over a long distance through photonic channels.\nHowever, these quantum nodes are usually composed of quantum systems with\nemitted photons unsuitable for long-distance transmission. Therefore, quantum\nwavelength conversion to telecom is crucial for long-distance quantum networks\nbased on optical fiber. Here we propose wavelength conversion devices for\nsingle-photon polarization qubits using continuous variable quantum\nteleportation, which can efficiently convert qubits between near-infrared\n(780/795 nm suitable for interacting with atomic quantum nodes) and telecom\nwavelength (1300-1500 nm suitable for long-distance transmission). The\nteleportation uses entangled photon sources (i.e., non-degenerate two-mode\nsqueezed state) that can be generated by four-wave mixing in rubidium atomic\nvapor cells, with a diamond configuration of atomic transitions. The entangled\nfields can be emitted in two orthogonal polarizations with locked relative\nphase, making them especially suitable for interfacing with single-photon\npolarization qubits. Our work paves the way for the realization of\nlong-distance quantum networks.\n"}
{"abstract": "  Sequence-to-sequence (seq2seq) models are prevalent in semantic parsing, but\nhave been found to struggle at out-of-distribution compositional\ngeneralization. While specialized model architectures and pre-training of\nseq2seq models have been proposed to address this issue, the former often comes\nat the cost of generality and the latter only shows limited success. In this\npaper, we study the impact of intermediate representations on compositional\ngeneralization in pre-trained seq2seq models, without changing the model\narchitecture at all, and identify key aspects for designing effective\nrepresentations. Instead of training to directly map natural language to an\nexecutable form, we map to a reversible or lossy intermediate representation\nthat has stronger structural correspondence with natural language. The\ncombination of our proposed intermediate representations and pre-trained models\nis surprisingly effective, where the best combinations obtain a new\nstate-of-the-art on CFQ (+14.8 accuracy points) and on the template-splits of\nthree text-to-SQL datasets (+15.0 to +19.4 accuracy points). This work\nhighlights that intermediate representations provide an important and\npotentially overlooked degree of freedom for improving the compositional\ngeneralization abilities of pre-trained seq2seq models.\n"}
{"abstract": "  This version removed by arXiv administrators because the submitter did not\nhave the right to agree to the license at the time of submission\n"}
{"abstract": "  An elimination tree for a connected graph $G$ is a rooted tree on the\nvertices of $G$ obtained by choosing a root $x$ and recursing on the connected\ncomponents of $G-x$ to produce the subtrees of $x$. Elimination trees appear in\nmany guises in computer science and discrete mathematics, and they encode many\ninteresting combinatorial objects, such as bitstrings, permutations and binary\ntrees. We apply the recent Hartung-Hoang-M\\\"utze-Williams combinatorial\ngeneration framework to elimination trees, and prove that all elimination trees\nfor a chordal graph $G$ can be generated by tree rotations using a simple\ngreedy algorithm. This yields a short proof for the existence of Hamilton paths\non graph associahedra of chordal graphs. Graph associahedra are a general class\nof high-dimensional polytopes introduced by Carr, Devadoss, and Postnikov,\nwhose vertices correspond to elimination trees and whose edges correspond to\ntree rotations. As special cases of our results, we recover several classical\nGray codes for bitstrings, permutations and binary trees, and we obtain a new\nGray code for partial permutations. Our algorithm for generating all\nelimination trees for a chordal graph $G$ can be implemented in time\n$\\mathcal{O}(m+n)$ per generated elimination tree, where $m$ and $n$ are the\nnumber of edges and vertices of $G$, respectively. If $G$ is a tree, we improve\nthis to a loopless algorithm running in time $\\mathcal{O}(1)$ per generated\nelimination tree. We also prove that our algorithm produces a Hamilton cycle on\nthe graph associahedron of $G$, rather than just Hamilton path, if the graph\n$G$ is chordal and 2-connected. Moreover, our algorithm characterizes\nchordality, i.e., it computes a Hamilton path on the graph associahedron of $G$\nif and only if $G$ is chordal.\n"}
{"abstract": "  We study how publicity, in the form of highlighting in the science press,\naffects the citations of research papers. After a brief review of prior work,\nwe analyze papers published in Physical Review Letters (PRL) that are\nhighlighted across eight different platforms. Using multiple linear regression\nwe identify how each platform contributes to citations. We also analyze how\nfrequently the highlighted papers end up in the top 1% cited papers in their\nfield. We find that the strongest predictors of medium-term citation impact --\nup to 7 years post-publication -- are Viewpoints in Physics, followed by\nResearch Highlights in Nature, Editors' Suggestions in PRL, and Research\nHighlights in Nature Physics. Our key conclusions are that (a) highlighting for\nimportance identifies a citation advantage, which (b) is stratified according\nto the degree of vetting during peer review (internal and external to the\njournal). This implies that we can view highlighting platforms as predictors of\ncitation accrual, with varying degrees of strength that mirror each platform's\nvetting level.\n"}
{"abstract": "  Self-supervised learning (especially contrastive learning) has attracted\ngreat interest due to its huge potential in learning discriminative\nrepresentations in an unsupervised manner. Despite the acknowledged successes,\nexisting contrastive learning methods suffer from very low learning efficiency,\ne.g., taking about ten times more training epochs than supervised learning for\ncomparable recognition accuracy. In this paper, we reveal two contradictory\nphenomena in contrastive learning that we call under-clustering and\nover-clustering problems, which are major obstacles to learning efficiency.\nUnder-clustering means that the model cannot efficiently learn to discover the\ndissimilarity between inter-class samples when the negative sample pairs for\ncontrastive learning are insufficient to differentiate all the actual object\nclasses. Over-clustering implies that the model cannot efficiently learn\nfeatures from excessive negative sample pairs, forcing the model to\nover-cluster samples of the same actual classes into different clusters. To\nsimultaneously overcome these two problems, we propose a novel self-supervised\nlearning framework using a truncated triplet loss. Precisely, we employ a\ntriplet loss tending to maximize the relative distance between the positive\npair and negative pairs to address the under-clustering problem; and we\nconstruct the negative pair by selecting a negative sample deputy from all\nnegative samples to avoid the over-clustering problem, guaranteed by the\nBernoulli Distribution model. We extensively evaluate our framework in several\nlarge-scale benchmarks (e.g., ImageNet, SYSU-30k, and COCO). The results\ndemonstrate our model's superiority (e.g., the learning efficiency) over the\nlatest state-of-the-art methods by a clear margin. Codes available at:\nhttps://github.com/wanggrun/triplet .\n"}
{"abstract": "  We present a novel construction of linear deformations for Lie algebras and\nuse it to prove the non-rigidity of several classes of Lie algebras in\ndifferent varieties. We consider the family of Lie algebras with an abelian\nfactor showing that, in general, they are not rigid even for the case of a\n1-dimensional abelian factor. We also address the problem of k-rigidity for\nk-step nilpotent Lie algebras. Using our construction and recent results in\n[BCC] we prove the that the k-step free nilpotent Lie algebras are k-rigid but\nnot (k + 1)-rigid and that Heisenberg Lie algebras are 2-rigid but not 3-rigid.\n"}
{"abstract": "  An interstellar communication network benefits from relay nodes placed in the\ngravitational lenses of stars. The signal gains are of order $10^{9}$ with\noptimal alignment, allowing for GBits connections at kW power levels with\nmeter-sized probes over parsec distances. If such a network exists, there might\nbe a node in our solar system: where is it? With some assumptions on the\nnetwork topology, candidate sky positions can be calculated. Apparent positions\nare influenced by the parallax motion from the Earth's orbit around the Sun,\nand the (slow) drifts caused by proper motions of nearby stars. With Gaia\nastrometry, instantaneous positions can be determined with arcsec accuracy.\nThese potential node locations can be observed in targeted\n"}
{"abstract": "  Since March 2020, companies nationwide have started work from home (WFH) due\nto the rapid increase of confirmed COVID-19 cases in an attempt to help prevent\nthe coronavirus from spreading and rescue the economy from the pandemic. Many\norganizations have conducted surveys to understand people's opinions towards\nWFH. However, the findings are limited due to small sample size and the dynamic\ntopics over time. This study aims to understand the U.S. public opinions on\nworking from home during the COVID-19 pandemic. We conduct a large-scale social\nmedia study using Twitter data to portrait different groups who have\npositive/negative opinions about WFH. We perform an ordinary least squares\nregression to investigate the relationship between the sentiment about WFH and\nuser characteristics including gender, age, ethnicity, median household income,\nand population density. To better understand public opinion, we use latent\nDirichlet allocation to extract topics and discover how tweet contents relate\nto people's attitudes. These findings provide evidence that sentiment about WFH\nvaries across user characteristics. Furthermore, the content analysis sheds\nlight on the nuanced differences in sentiment and reveals disparities relate to\nWFH.\n"}
{"abstract": "  This paper addresses the interesting problem of processing and analyzing data\nin geographic information systems (GIS) to achieve a clear perspective on urban\nsprawl. The term urban sprawl refers to overgrowth and expansion of low-density\nareas with issues such as car dependency and segregation between residential\nversus commercial use. Sprawl has impacts on the environment and public health.\nIn our work, spatiotemporal features related to real GIS data on urban sprawl\nsuch as population growth and demographics are mined to discover knowledge for\ndecision support. We adapt data mining algorithms, Apriori for association rule\nmining and J4.8 for decision tree classification to geospatial analysis,\ndeploying the ArcGIS tool for mapping. Knowledge discovered by mining this\nspatiotemporal data is used to implement a prototype spatial decision support\nsystem (SDSS). This SDSS predicts whether urban sprawl is likely to occur.\nFurther, it estimates the values of pertinent variables to understand how the\nvariables impact each other. The SDSS can help decision-makers identify\nproblems and create solutions for avoiding future sprawl occurrence and\nconducting urban planning where sprawl already occurs, thus aiding sustainable\ndevelopment. This work falls in the broad realm of geospatial intelligence and\nsets the stage for designing a large scale SDSS to process big data in complex\nenvironments, which constitutes part of our future work.\n"}
{"abstract": "  With parallel and distributed computing (PDC) now wide-spread, modern\ncomputing programs must incorporate PDC within the curriculum. ACM and IEEE\nComputer Society's Computer Science curricular guidelines have recommended\nexposure to PDC concepts since 2013. More recently, a variety of initiatives\nhave made PDC curricular content, lectures, and labs freely available for\nundergraduate computer science programs. Despite these efforts, progress in\nensuring computer science students graduate with sufficient PDC exposure has\nbeen uneven.\n  This paper discusses the impact of ABET's revised criteria that have required\nexposure to PDC to achieve accreditation for computer science programs since\n2018. The authors reviewed 20 top ABET-accredited computer science programs and\nanalyzed how they covered the required PDC components in their curricula. Using\ntheir own institutions as case studies, the authors examine in detail how three\ndifferent ABET-accredited computer science programs covered PDC using different\napproaches, yet meeting the PDC requirements of these ABET criteria. The paper\nalso shows how ACM/IEEE Computer Society curricular guidelines for computer\nengineering and software engineering programs, along with ABET accreditation\ncriteria, can cover PDC.\n"}
{"abstract": "  It is well-known that plug-in statistical estimation of optimal transport\nsuffers from the curse of dimensionality. Despite recent efforts to improve the\nrate of estimation with the smoothness of the problem, the computational\ncomplexity of these recently proposed methods still degrades exponentially with\nthe dimension. In this paper, thanks to an infinite-dimensional sum-of-squares\nrepresentation, we derive a statistical estimator of smooth optimal transport\nwhich achieves a precision $\\varepsilon$ from $\\tilde{O}(\\varepsilon^{-2})$\nindependent and identically distributed samples from the distributions, for a\ncomputational cost of $\\tilde{O}(\\varepsilon^{-4})$ when the smoothness\nincreases, hence yielding dimension-free statistical and computational rates,\nwith potentially exponentially dimension-dependent constants.\n"}
{"abstract": "  As self-driving systems become better, simulating scenarios where the\nautonomy stack may fail becomes more important. Traditionally, those scenarios\nare generated for a few scenes with respect to the planning module that takes\nground-truth actor states as input. This does not scale and cannot identify all\npossible autonomy failures, such as perception failures due to occlusion. In\nthis paper, we propose AdvSim, an adversarial framework to generate\nsafety-critical scenarios for any LiDAR-based autonomy system. Given an initial\ntraffic scenario, AdvSim modifies the actors' trajectories in a physically\nplausible manner and updates the LiDAR sensor data to match the perturbed\nworld. Importantly, by simulating directly from sensor data, we obtain\nadversarial scenarios that are safety-critical for the full autonomy stack. Our\nexperiments show that our approach is general and can identify thousands of\nsemantically meaningful safety-critical scenarios for a wide range of modern\nself-driving systems. Furthermore, we show that the robustness and safety of\nthese systems can be further improved by training them with scenarios generated\nby AdvSim.\n"}
{"abstract": "  Transformer is a ubiquitous model for natural language processing and has\nattracted wide attentions in computer vision. The attention maps are\nindispensable for a transformer model to encode the dependencies among input\ntokens. However, they are learned independently in each layer and sometimes\nfail to capture precise patterns. In this paper, we propose a novel and generic\nmechanism based on evolving attention to improve the performance of\ntransformers. On one hand, the attention maps in different layers share common\nknowledge, thus the ones in preceding layers can instruct the attention in\nsucceeding layers through residual connections. On the other hand, low-level\nand high-level attentions vary in the level of abstraction, so we adopt\nconvolutional layers to model the evolutionary process of attention maps. The\nproposed evolving attention mechanism achieves significant performance\nimprovement over various state-of-the-art models for multiple tasks, including\nimage classification, natural language understanding and machine translation.\n"}
{"abstract": "  We have run numerical simulations of Euclidean lattice quantum gravity for\nmetrics which are time-independent and spherically symmetric. The radial\nvariable is discretized as $r=hL_{Planck}$, with $h=0,1,...,N$ and $N$ up to\n$10^5$. The Lagrangian is of the form $\\sqrt{g}(R+\\alpha R^2)$ (in units\n$c=\\hbar=G=1$) and the action is positive-definite, allowing the use of a\nstandard Metropolis algorithm with update probability $\\exp(-\\beta \\Delta S)$.\nBy minimizing the $R+R^2$ action with respect to conformal modes, Bonanno and\nReuter have recently found analytical evidence of a non-trivial \"rippled\"\nground state resembling a kinetic condensate of QCD. Our simulations at low but\nfinite temperature ($T=\\beta^{-1}$) also display strong localized oscillations\nof the metric, whose total action $S$ remains $\\ll \\hbar$ thanks to the\nindefinite sign of $R$. The average metric $\\langle g_{rr} \\rangle$ is\nsignificantly different from flat space. The scaling properties of $S$ and\n$\\langle g_{rr} \\rangle$ are investigated in dependence on $N$ and $\\beta$.\n"}
{"abstract": "  The flow of grains through narrow apertures posses an extraordinary\nchallenge: clogging. Strategies to alleviate the effect of clogging, such as\nthe use of external vibration, are always part of the design of machinery for\nthe handling of bulk materials. It has recently been shown that one way to\nreduce clogging is to use a small fraction of small particles as an additive.\nBesides, several works reported that self-repelling magnetic grains can flow\nthrough narrow apertures with little clogging, which suggest these are\nexcellent candidates as \"lubricating\" additives for other granular materials.\nIn this work, we study the effect of adding self-repelling magnetic particles\nto a sample of grains in two-dimensions. We find that, in contrast with\nintuition, the added magnetic grains not necessarily aid the flow of the\noriginal species.\n"}
{"abstract": "  We present a novel theory and implementation for computing coupled electronic\nand quantal nuclear subsystems on a single potential energy surface, moving\nbeyond the standard Born-Oppenheimer (BO) separation of nuclei and electrons.\nWe formulate an exact self-consistent nucleus-electron embedding potential from\nthe single product molecular wavefunction, and demonstrate that the fundamental\nbehavior of correlated nucleus-electron can be computed for mean-field\nelectrons that are responsive to a quantal anharmonic vibration of selected\nnuclei in a discrete variable representation. Geometric gauge choices are\ndiscussed and necessary for formulating energy invariant biorthogonal\nelectronic equations. Our method is further applied to characterize\nvibrationally averaged molecular bonding properties of molecular energetics,\nbond length, protonic and electron density. Moreover, post-Hartree-Fock\nelectron correlation can be conveniently computed on the basis of\nnucleus-electron coupled molecular orbitals, as demonstrated to correlated\nmodels of second-order M{\\o}llet-Plesset perturbation and full configuration\ninteraction theories. Our approach not only accurately quantifies non-classical\nnucleus-electron couplings for revising molecular bonding properties, but also\nprovides an alternative time-independent approach for deploying non-BO\nmolecular quantum chemistry.\n"}
{"abstract": "  We characterize the extreme heartbeat star system MACHO 80.7443.1718 in the\nLMC using TESS photometry and spectroscopic observations from the Magellan\nInamori Kyocera Echelle (MIKE) and SOAR Goodman spectographs. MACHO\n80.7443.1718 was first identified as a heartbeat star system in the All-Sky\nAutomated Survey for SuperNovae (ASAS-SN) with $P_{\\rm\norb}=32.836\\pm0.008\\,{\\rm d}$. MACHO 80.7443.1718 is a young (${\\sim}6$~Myr),\nmassive binary, composed of a B0 Iae supergiant with $M_1 \\simeq 35 M_\\odot$\nand an O9.5V secondary with $M_2 \\simeq 16 M_\\odot$ on an eccentric\n($e=0.51\\pm0.03$) orbit. In addition to having the largest variability\namplitude amongst all known heartbeats stars, MACHO 80.7443.1718 is also one of\nthe most massive heartbeat stars yet discovered. The B[e] supergiant has Balmer\nemission lines and permitted/forbidden metallic emission lines associated with\na circumstellar disk. The disk rapidly dissipates at periastron which could\nindicate mass transfer to the secondary, but re-emerges immediately following\nperiastron passage. MACHO 80.7443.1718 also shows tidally excited oscillations\nat the $N=25$ and $N=41$ orbital harmonics and has a rotational period of 4.4\nd.\n"}
{"abstract": "  Digitalization has led to radical changes in the distribution of goods across\nvarious sectors. The tendency is to move from traditional buyer-seller markets\nto subscription-based on-demand \"smart\" matching platforms enabled by pervasive\nICTs. The driving force behind this lies in the fact that assets, which were\nscarce in the past, are readily abundant, approaching a regime of zero marginal\ncosts. This is also becoming a reality in electrified energy systems due to the\nsubstantial growth of distributed renewable energy sources such as solar and\nwind; the increasing number of small-scale storage units such as batteries and\nheat pumps; and the availability of flexible loads that enable demand-side\nmanagement. In this context, this article proposes a system architecture based\non a logical (cyber) association of spatially distributed (physical) elements\nas an approach to build a virtual microgrid operated as a software-defined\nenergy network (SDEN) that is enabled by packetized energy management. The\nproposed cyber-physical system presumes that electrical energy is shared among\nits members and that the energy sharing is enabled in the cyber domain by\nhandshakes inspired by resource allocation methods utilized in computer\nnetworks, wireless communications, and peer-to-peer Internet applications\n(e.g., BitTorrent). The proposal has twofold benefits: (i) reducing the\ncomplexity of current market-based solutions by removing unnecessary and costly\nmediations and (ii) guaranteeing energy access to all virtual microgrid members\naccording to their individual needs. This article concludes that the proposed\nsolution generally complies with the existing regulations but has highly\ndisruptive potential to organize a dominantly electrified energy system in the\nmid- to long-term, being a technical counterpart to the recently developed\nsocial-oriented microgrid proposals.\n"}
{"abstract": "  The lightcone singularity at the origin is resolved by blowing up the\nsingular point to $CP^1$. The Lorentz group acts on the resolved lightcone and\nhas $CP^1$ as a special orbit. Using Wigner's method of associating unitary\nirreducible representations of the Poincar\\'e group to particle states, we find\nthat the special orbit gives rise to new vacuum states. These vacuum states are\nlabelled by the principal series representations of $SL(2,C)$. Some remarks are\nincluded on the applications of these results to gauge theories and\nasymptotically flat spacetimes.\n"}
{"abstract": "  In this contribution we identify two scenarios for the evolutionary branch\ncut universe. In the first scenario, the universe evolves continuously from the\nnegative complex cosmological time sector, prior to a primordial singularity,\nto the positive one, circumventing continuously a branch cut, and no primordial\nsingularity occurs in the imaginary sector, only branch points. In the second\nscenario, the branch cut and branch point disappear after the realisation of\nthe imaginary component of the complex time by means of a Wick rotation, which\nis replaced by the thermal time. In the second scenario, the universe has its\norigin in the Big Bang, but the model contemplates simultaneously a mirrored\nparallel evolutionary universe going backwards in the cosmological thermal time\nnegative sector. A quantum formulation based on the WDW equation is sketched\nand preliminary conclusions are drawn.\n"}
{"abstract": "  In this work that continues the \"NGICS - New Italian City Geological Study\" a\nproject of the Department of Climatology and Geology of TS Corporation Srl, we\npresent the study relating to the Municipality of Montefiascone (VT). We\nanalyzed 25 years of astronomical, geological, meteorological and climatic\ndata, comparing them to verify the long-term trend of local variations in\ntemperatures, detections, solar radiation and geological events, with the\nultimate goal of understanding climate and geological changes a long term in\nthis geographical area. The analysis is performed using a statistical approach\nand attention is used to minimize any effects caused by the error in case of\nlack of data.\n"}
{"abstract": "  In the last decade many research efforts have been focused on understanding\nthe rheology of disordered materials, and several theoretical predictions have\nbeen put forward regarding their yielding behavior. Nevertheless, not many\nexperiments nor molecular dynamics simulations were dedicated to testing those\ntheoretical predictions. Here we use computer simulations to study the yielding\ntransition under two different loading schemes: standard simple shear dynamics,\nand self-propelled, dense active systems. In the active systems a yielding\ntransition is observed as expected, when the self-propulsion is increased.\nHowever, the range of self-propulsions in which a pure liquid regime exist\nappears to vanish upon approaching the so-called \"jamming point\" at which\nsolidity of soft-sphere packings is lost. Such an \"active yielding\" transition\nshares similarities with the generic yielding transition for shear flows. A\nHerschel-Bulkley law is observed in both loading scenarios, with a clear\ndifference in the critical scaling exponents between the two, suggesting the\nexistent of different universality classes for the yielding transition under\ndifferent driving conditions. In addition, we present direct measurements of\nlength and time scales for both driving scenarios. A comparison with\ntheoretical predictions from recent literature reveals poor agreement with our\nnumerical results.\n"}
{"abstract": "  Red lead and lead white are some of the most ancient and common pigments in\nmural paintings. However, they tend to blacken with time due to their oxidation\nto plattnerite (\\b{eta}-PbO2). The possibility to induce the reconversion\nreactions by CW laser heating is hereby discussed. A thermodynamic study by TGA\nshowed that direct cerussite or hydrocerussite formation from plattnerite are\nnot suitable reconversion routes, which was confirmed by laser irradiation\ntrials under CO2 and CO2/H2O fluxes. Minium (Pb3O4) and subsequent massicot\n(\\b{eta}-PbO) formation from plattnerite were achieved (confirmed by SEM-EDS,\nXRD and micro-Raman) under Ar+, 810 nm diode and Nd:YAG lasers. The latter\nappears to be the most suited for restauration purposes, given the broad minium\nreconversion irradiance range. This is confirmed by successful trials on\nmacroscopic areas of naturally darkened red lead containing samples.\n"}
{"abstract": "  In today's connected society, many people rely on mHealth and self-tracking\n(ST) technology to help them adopt healthier habits with a focus on breaking\ntheir sedentary lifestyle and staying fit. However, there is scarce evidence of\nsuch technological interventions' effectiveness, and there are no standardized\nmethods to evaluate their impact on people's physical activity (PA) and health.\nThis work aims to help ST and HCI practitioners and researchers by empowering\nthem with systematic guidelines and a framework for designing and evaluating\ntechnological interventions to facilitate health behavior change (HBC) and user\nengagement (UE), focusing on increasing PA and decreasing sedentariness. To\nthis end, we conduct a literature review of 117 papers between 2008 and 2020,\nwhich identifies the core ST HCI design methods and their efficacy, as well as\nand the most comprehensive list to date of UE evaluation metrics for ST. Based\non the review's findings, we propose PAST SELF, a framework to guide the design\nand evaluation of ST technology that has potential applications in industrial\nand scientific settings. Finally, to facilitate researchers and practitioners,\nwe complement this paper with an open corpus and an adaptive exploration tool\nfor the PAST SELF data.\n"}
{"abstract": "  Selective inference (post-selection inference) is a methodology that has\nattracted much attention in recent years in the fields of statistics and\nmachine learning. Naive inference based on data that are also used for model\nselection tends to show an overestimation, and so the selective inference\nconditions the event that the model was selected. In this paper, we develop\nselective inference in propensity score analysis with a semiparametric\napproach, which has become a standard tool in causal inference. Specifically,\nfor the most basic causal inference model in which the causal effect can be\nwritten as a linear sum of confounding variables, we conduct Lasso-type\nvariable selection by adding an $\\ell_1$ penalty term to the loss function that\ngives a semiparametric estimator. Confidence intervals are then given for the\ncoefficients of the selected confounding variables, conditional on the event of\nvariable selection, with asymptotic guarantees. An important property of this\nmethod is that it does not require modeling of nonparametric regression\nfunctions for the outcome variables, as is usually the case with semiparametric\npropensity score analysis.\n"}
{"abstract": "  Unmeasured covariates constitute one of the important problems in causal\ninference. Even if there are some unmeasured covariates, some instrumental\nvariable methods such as a two-stage residual inclusion (2SRI) estimator, or a\nlimited-information maximum likelihood (LIML) estimator can obtain an unbiased\nestimate for causal effects despite there being nonlinear outcomes such as\nbinary outcomes; however, it requires that we specify not only a correct\noutcome model but also a correct treatment model. Therefore, detecting correct\nmodels is an important process. In this paper, we propose two model selection\nprocedures: AIC-type and BIC-type, and confirm their properties. The proposed\nmodel selection procedures are based on a LIML estimator. We prove that a\nproposed BIC-type model selection procedure has model selection consistency,\nand confirm their properties of the proposed model selection procedures through\nsimulation datasets.\n"}
{"abstract": "  Row-column arrays have shown to be able to generate 3-D ultrafast ultrasound\nimages with an order of magnitude less independent electronic channels than\nclassic 2D matrix arrays. Unfortunately row-column array images suffer from\nmajor imaging artefacts due to the high side lobes. This paper proposes a\nrow-column specific beamforming technique that exploits the incoherent nature\nof certain row column array artefacts. The geometric mean of the data from each\nrow and column pair is taken prior to summation in beamforming, thus\ndrastically reducing incoherent imaging artefacts compared to traditional\ncoherent compounding. The effectiveness of this technique was demonstrated in\nsilico, and the results show an average fivefold reduction in side-lobe levels.\nSignificantly improved contrast was demonstrated with Tissue-to-noise ratio\nincreasing from $\\sim$10dB to $\\sim$30dB and Tissue Contrast Ratio increasing\nfrom $\\sim$21dB to $\\sim$42dB when using the proposed new method compared to\nDelay and Sum. These new techniques allowed for high quality 3D imaging whilst\nmaintaining high frame rate potential.\n"}
{"abstract": "  Inertial particles advected by a background flow can show complex structures.\nWe consider inertial particles in a 2D Taylor-Green (TG) flow and characterize\nparticle dynamics as a function of the particle's Stokes number using dynamic\nmode decomposition (DMD) method from particle image velocimetry (PIV)\nlike-data. We observe the formation of caustic structures and analyze them\nusing DMD to (a) determine the Stokes number of the particles, and (b) estimate\nthe particle Stokes number composition. Our analysis in this idealized flow\nwill provide useful insight to analyze inertial particles in more complex or\nturbulent flows. We propose that the DMD technique can be used to perform a\nsimilar analysis on an experimental system.\n"}
{"abstract": "  With the emergence of personality computing as a new research field related\nto artificial intelligence and personality psychology, we have witnessed an\nunprecedented proliferation of personality-aware recommendation systems. Unlike\nconventional recommendation systems, these new systems solve traditional\nproblems such as the cold start and data sparsity problems. This survey aims to\nstudy and systematically classify personality-aware recommendation systems. To\nthe best of our knowledge, this survey is the first that focuses on\npersonality-aware recommendation systems. We explore the different design\nchoices of personality-aware recommendation systems, by comparing their\npersonality modeling methods, as well as their recommendation techniques.\nFurthermore, we present the commonly used datasets and point out some of the\nchallenges of personality-aware recommendation systems.\n"}
{"abstract": "  This paper aims to provide understandings for the effect of an\nover-parameterized model, e.g. a deep neural network, memorizing\ninstance-dependent noisy labels. We first quantify the harms caused by\nmemorizing noisy instances, and show the disparate impacts of noisy labels for\nsample instances with different representation frequencies. We then analyze how\nseveral popular solutions for learning with noisy labels mitigate this harm at\nthe instance level. Our analysis reveals that existing approaches lead to\ndisparate treatments when handling noisy instances. While higher-frequency\ninstances often enjoy a high probability of an improvement by applying these\nsolutions, lower-frequency instances do not. Our analysis reveals new\nunderstandings for when these approaches work, and provides theoretical\njustifications for previously reported empirical observations. This observation\nrequires us to rethink the distribution of label noise across instances and\ncalls for different treatments for instances in different regimes.\n"}
{"abstract": "  Almost all statistical and machine learning methods in analyzing brain\nnetworks rely on distances and loss functions, which are mostly Euclidean or\nmatrix norms. The Euclidean or matrix distances may fail to capture underlying\nsubtle topological differences in brain networks. Further, Euclidean distances\nare sensitive to outliers. A few extreme edge weights may severely affect the\ndistance. Thus it is necessary to use distances and loss functions that\nrecognize topology of data. In this review paper, we survey various topological\ndistance and loss functions from topological data analysis (TDA) and persistent\nhomology that can be used in brain network analysis more effectively. Although\nthere are many recent brain imaging studies that are based on TDA methods,\npossibly due to the lack of method awareness, TDA has not taken as the\nmainstream tool in brain imaging field yet. The main purpose of this paper is\nprovide the relevant technical survey of these powerful tools that are\nimmediately applicable to brain network data.\n"}
{"abstract": "  A conjecture of Le says that the Deligne polytope $\\Delta_d$ is generically\nordinary if $p\\equiv 1\\ (\\!\\!\\bmod\\ D(\\Delta_d))$, where $D(\\Delta_d)$ is a\ncombinatorial constant determined by $\\Delta_d$. In this paper a counterexample\nis given to show that the conjecture is not true in general.\n"}
{"abstract": "  In event-based sensing, many sensors independently and asynchronously emit\nevents when there is a change in their input. Event-based sensing can present\nsignificant improvements in power efficiency when compared to traditional\nsampling, because (1) the output is a stream of events where the important\ninformation lies in the timing of the events, and (2) the sensor can easily be\ncontrolled to output information only when interesting activity occurs at the\ninput.\n  Moreover, event-based sampling can often provide better resolution than\nstandard uniform sampling. Not only does this occur because individual\nevent-based sensors have higher temporal resolution, it also occurs because the\nasynchrony of events allows for less redundant and more informative encoding.\nWe would like to explain how such curious results come about.\n  To do so, we use ideal time encoding machines as a proxy for event-based\nsensors. We explore time encoding of signals with low rank structure, and apply\nthe resulting theory to video. We then see how the asynchronous firing times of\nthe time encoding machines allow for better reconstruction than in the standard\nsampling case, if we have a high spatial density of time encoding machines that\nfire less frequently.\n"}
{"abstract": "  Our research explores the potential of a humanoid robot for work in\nunpredictable environments, but controlling a humanoid robot remains a very\ndifficult problem. In our previous work, we designed a prototype virtual\nreality (VR) interface to allow an operator to command a humanoid robot.\nHowever, while usable, the initial interface was not sufficient for commanding\nthe robot to perform the tasks; for example, in some cases, there was a lack of\nprecision available for robot control. The interface was overly cumbersome in\nsome areas as well. In this paper, we discuss numerous additions, inspired by\ntraditional interfaces and virtual reality video games, to our prior\nimplementation, providing additional ways to visualize and command a humanoid\nrobot to perform difficult tasks within a virtual world.\n"}
{"abstract": "  We present a nonperturbative quasi-classical theory of graphene\nphotoconductivity. We consider the influence of low-frequency (microwave,\nterahertz, mid-infrared) radiation on the static conductivity of a uniform\ngraphene layer and calculate its photoconductivity as a function of frequency,\npolarization and strength of the external ac electric field, as well as on the\nmaterial properties (electron density, scattering time) and temperature. The\ntheory is valid at frequencies $\\hbar\\omega\\lesssim 2E_F$ and at arbitrarily\nstrong ac electric fields. We compare our results with those of the third-order\nperturbation theory and determine the applicability range of the perturbative\nsolutions.\n"}
{"abstract": "  In spinor Bose-Einstein condensates, spin-changing collisions are a\nremarkable proxy to coherently realize macroscopic many-body quantum states.\nThese processes have been, e.g., exploited to generate entanglement, to study\ndynamical quantum phase transitions, and proposed for realizing nematic phases\nin atomic condensates. In the same systems dressed by Raman beams, the coupling\nbetween spin and momentum induces a spin dependence in the scattering processes\ntaking place in the gas. Here we show that, at weak couplings, such modulation\nof the collisions leads to an effective Hamiltonian which is equivalent to the\none of an artificial spinor gas with spin-changing collisions that are tunable\nwith the Raman intensity. By exploiting this dressed-basis description, we\npropose a robust protocol to coherently drive the spin-orbit coupled condensate\ninto the ferromagnetic stripe phase via crossing a quantum phase transition of\nthe effective low-energy model in an excited-state.\n"}
{"abstract": "  Scientific communities naturally tend to organize around data ecosystems\ncreated by the combination of their observational devices, their data\nrepositories, and the workflows essential to carry their research from\nobservation to discovery. However, these legacy data ecosystems are now\nbreaking down under the pressure of the exponential growth in the volume and\nvelocity of these workflows, which are further complicated by the need to\nintegrate the highly data intensive methods of the Artificial Intelligence\nrevolution. Enabling ground breaking science that makes full use of this new,\ndata saturated research environment will require distributed systems that\nsupport dramatically improved resource sharing, workflow portability and\ncomposability, and data ecosystem convergence.\n  The Cybercosm vision presented in this white paper describes a radically\ndifferent approach to the architecture of distributed systems for\ndata-intensive science and its application workflows. As opposed to traditional\nmodels that restrict interoperability by hiving off storage, networking, and\ncomputing resources in separate technology silos, Cybercosm defines a minimally\nsufficient hypervisor as a spanning layer for its data plane that virtualizes\nand converges the local resources of the system's nodes in a fully\ninteroperable manner. By building on a common, universal interface into which\nthe problems that infect today's data-intensive workflows can be decomposed and\nattacked, Cybercosm aims to support scalable, portable and composable workflows\nthat span and merge the distributed data ecosystems that characterize leading\nedge research communities today.\n"}
{"abstract": "  The refraction of space-time (ST) wave packets at planar interfaces between\nnon-dispersive, homogeneous, isotropic dielectrics exhibit fascinating\nphenomena, even at normal incidence. Examples of such refractive phenomena\ninclude group-velocity invariance across the interface, anomalous refraction,\nand group-velocity inversion. Crucial differences emerge at oblique incidence\nwith respect to the results established at normal incidence. For example, the\ngroup velocity of the refracted ST wave packet can be tuned simply by changing\nthe angle of incidence. In paper (III) of this sequence, we present\nexperimental verification of the refractive phenomena exhibited by ST wave\npackets at oblique incidence that were predicted in paper (I). We also examine\na proposal for 'blind synchronization' whereby identical ST wave packets arrive\nsimultaneously at different receivers without \\textit{a priori} knowledge of\ntheir locations except that they are all located at the same depth beyond an\ninterface between two media. A first proof-of-principle experimental\ndemonstration of this effect is provided.\n"}
{"abstract": "  In theories with higher time derivatives, the Hamiltonian analysis of\nOstrogradsky predicts an instability. However, this Hamiltonian treatment does\nnot correspond the way that these theories are treated in quantum field theory,\nand the instability may be avoided in at least some cases. We present a very\nsimple model which illustrates these features.\n"}
{"abstract": "  Conventional deep learning based methods for object detection require a large\namount of bounding box annotations for training, which is expensive to obtain\nsuch high quality annotated data. Few-shot object detection, which learns to\nadapt to novel classes with only a few annotated examples, is very challenging\nsince the fine-grained feature of novel object can be easily overlooked with\nonly a few data available. In this work, aiming to fully exploit features of\nannotated novel object and capture fine-grained features of query object, we\npropose Dense Relation Distillation with Context-aware Aggregation (DCNet) to\ntackle the few-shot detection problem. Built on the meta-learning based\nframework, Dense Relation Distillation module targets at fully exploiting\nsupport features, where support features and query feature are densely matched,\ncovering all spatial locations in a feed-forward fashion. The abundant usage of\nthe guidance information endows model the capability to handle common\nchallenges such as appearance changes and occlusions. Moreover, to better\ncapture scale-aware features, Context-aware Aggregation module adaptively\nharnesses features from different scales for a more comprehensive feature\nrepresentation. Extensive experiments illustrate that our proposed approach\nachieves state-of-the-art results on PASCAL VOC and MS COCO datasets. Code will\nbe made available at https://github.com/hzhupku/DCNet.\n"}
{"abstract": "  The Lawson criterion is a key concept in the pursuit of fusion energy,\nrelating the fuel density $n$, pulse duration $\\tau$ or energy confinement time\n$\\tau_E$, and fuel temperature $T$ to the energy gain $Q$ of a fusion plasma.\nThe purpose of this paper is to explain and review the Lawson criterion and to\nprovide a compilation of achieved parameters for a broad range of historical\nand contemporary fusion experiments. Although this paper focuses on the Lawson\ncriterion, it is only one of many equally important factors in assessing the\nprogress and ultimate likelihood of any fusion concept becoming a commercially\nviable fusion-energy system. Only experimentally measured or inferred values of\n$n$, $\\tau$ or $\\tau_E$, and $T$ that have been published in the peer-reviewed\nliterature are included in this paper, unless noted otherwise. For extracting\nthese parameters, we discuss methodologies that are necessarily specific to\ndifferent fusion approaches (including magnetic, inertial, and magneto-inertial\nfusion). This paper is intended to serve as a reference for fusion researchers\nand a tutorial for all others interested in fusion energy.\n"}
{"abstract": "  Inaccurate slow streaming models of acoustically-induced fluid flow remain in\nuse due to the lack of a generalized alternative. The Multiscale Articulated\nDifferentials Method (MADaM; [see co-article]) solves this problem with a\ncomplete, scale-sensitive spatiotemporal expansion. Applied to classic axial\nEckart streaming, the MADaM produces, in terms of a Burgers equation, a\nlong-sought transient solution able to accommodate frequencies and amplitudes\nfar beyond slow streaming models. Steady streaming is governed by a\ncorresponding Riccati equation that, when solved, produces simple expressions\nexplaining innate features of fast acoustic streaming.\n"}
{"abstract": "  The influence of Ga incorporation into cubic In$_2$O$_3$ on the electronic\nand vibrational properties is discussed for (In$_{1-x}$,Ga$_x$)$_2$O$_3$ alloy\nfilms grown by molecular beam epitaxy. Using spectroscopic ellipsometry, a\nlinear dependence of the absorption onset on the Ga content $x$ is found with a\nblueshift of up to 150 meV for $x = 0.1$. Consistently, the fundamental band\ngap exhibits a blueshift as determined by hard X-ray photoelectron\nspectroscopy. The dependence of the absorption onset and the effective electron\nmass on the electron concentration is derived from the infrared dielectric\nfunctions for a Sn doped alloy film. The influence of alloying on phonon modes\nis analyzed on the basis of Raman spectroscopic measurements. The frequencies\nof several phonon modes are identified as sensitive measures for the\nspectroscopic determination of the Ga content.\n"}
{"abstract": "  Simulating quantum many-body dynamics on classical computers is a challenging\nproblem due to the exponential growth of the Hilbert space. Artificial neural\nnetworks have recently been introduced as a new tool to approximate\nquantum-many body states. We benchmark the variational power of the restricted\nBoltzmann machine quantum states and different shallow and deep neural\nautoregressive quantum states to simulate global quench dynamics of a\nnon-integrable quantum Ising chain. We find that the number of parameters\nrequired to represent the quantum state at a given accuracy increases\nexponentially in time. The growth rate is only slightly affected by the network\narchitecture over a wide range of different design choices: shallow and deep\nnetworks, small and large filter sizes, dilated and normal convolutions, with\nand without shortcut connections.\n"}
{"abstract": "  We characterize lower growth estimates for subsolutions in halfspaces of\nfully nonlinear partial differential equations on the form $$ F(x,u,Du,D^2u) =\n0 $$ in terms of solutions to ordinary differential equations built solely upon\na growth assumption on $F$. Using this characterization we derive several sharp\nPhragmen-Lindel\\\"of-type theorems for certain classes of well known PDEs. The\nequation need not be uniformly elliptic nor homogeneous and we obtain results\nboth in case the subsolution is bounded or unbounded. Among our results we\nretrieve classical estimates in the halfspace for $p$-subharmonic functions and\nextend those to more general equations; we prove sharp growth estimates, in\nterms of $k$ and the asymptotic behaviour of $\\int_{0}^{R} C(s) ds$, for\nsubsolutions of equations allowing for sublinear growth in the gradient of the\nform $C(|x|)|Du|^k$ with $k\\geq 1$; we establish a Phragmen-Lindel\\\"of theorem\nfor weak subsolutions of the variable exponent $p$-Laplace equation in\nhalfspaces, $1 < p(x) < \\infty$, $p(x) \\in C^1$, of which we conclude sharpness\nby finding the \"slowest growing\" $p(x)$-harmonic function together with its\ncorresponding family of $p(x)$-exponents. The paper ends with a discussion of\nour results from the point of view of a spatially dependent diffusion problem.\n"}
{"abstract": "  Soft matter consists of meso-scale (nm~{\\mu}m) structures that are formed by\nweak interactions and reorganized under thermal fluctuations. The resulting\ncomplex relaxation phenomena may be probed with microrheology, by observing the\nmovement of embedded probe particles. Because of the softness of the material,\nhowever, perturbations to the probe that are inevitably added during\nmicrorheology experiments prevent direct translation of those movements to\nrheological properties. In this study, we conducted optical-trap-based\nmicrorheology with significantly reduced mechanical perturbations; dual\nfeedback technology allowed us to apply well-determined optical-trapping forces\nto a fluctuating embedded probe and precisely measure its response and\nfluctuations with high spatiotemporal resolution. We demonstrate the improved\nperformance of this technique by studying an reconstituted network of actin\ncytoskeletal filaments, observing their slow dynamics, homogeneous thermal\nfluctuations as well as activated hopping between mesoscale microenvironments.\n"}
{"abstract": "  Many extensions of the Standard Model predict the existence of new charged or\nneutral gauge bosons, with a wide variety of phenomenological implications\ndepending on the model adopted. The search for such particles is extensively\ncarried through at the Large Hadron Collider (LHC), and it is therefore of\ncrucial importance to have for each proposed scenario quantitative predictions\nthat can be matched to experiments. In this work we focus on the implications\nof one of these models, the TopFlavor Model, proposing a charged\n$\\text{W}^\\prime$ boson that has preferential couplings to the third generation\nfermions. We compare such predictions to the ones from the so called Sequential\nStandard Model (SSM), that is used as benchmark, being one of the simplest and\nmost commonly considered models for searches at the LHC. We identify the\nparameter space still open for searches at the LHC, and in particular we show\nthat the cross section for the processes $pp \\to \\text{W}^\\prime \\to \\tau \\nu$\nand $pp \\to \\text{W}^\\prime \\to tb$ can be up to two orders of magnitude\nsmaller with respect to the SSM, depending on the free parameters of the model,\nlike the particle mass and its width. This study makes the case for further\nsearches at the LHC, and shows how a complete and systematic model independent\nanalysis of $\\text{W}^\\prime$ boson phenomenology at colliders is essential to\nprovide guidance for future searches.\n"}
{"abstract": "  Aqueous proton transport plays a key role in acid-base neutralization, and\nenergy transport through biological membranes and hydrogen fuel cells.\nExtensive experimental and theoretical studies have resulted in a highly\ndetailed elucidation of one of the underlying microscopic mechanisms for\naqueous excess proton transport, known as the von Grotthuss mechanism,\ninvolving different hydrated proton configurations with associated high\nfluxional structural dynamics. Hydroxide transport, with approximately two-fold\nlower bulk diffusion rates than those of excess protons, has received much less\nattention. We present femtosecond UV/IR pump-probe experiments and ab initio\nmolecular dynamics simulations of different proton transport pathways of the\nbifunctional photoacid 7-hydroxyquinoline (7HQ) in water-methanol mixtures. For\n7HQ solvent-dependent photoacidity, free energy-reactivity correlation\nbehaviour and QM/MM trajectories point to a dominant OH-/CH3O- transport\npathway, for all water-methanol mixing ratios investigated. This provides\nconclusive evidence for the hydrolysis/methanolysis acid-base neutralization\npathway formulated by Manfred Eigen half a century ago.\n"}
{"abstract": "  Existing spatial object recommendation algorithms generally treat objects\nidentically when ranking them. However, spatial objects often cover different\nlevels of spatial granularity and thereby are heterogeneous. For example, one\nuser may prefer to be recommended a region (say Manhattan), while another user\nmight prefer a venue (say a restaurant). Even for the same user, preferences\ncan change at different stages of data exploration. In this paper, we study how\nto support top-k spatial object recommendations at varying levels of spatial\ngranularity, enabling spatial objects at varying granularity, such as a city,\nsuburb, or building, as a Point of Interest (POI). To solve this problem, we\npropose the use of a POI tree, which captures spatial containment relationships\nbetween POIs. We design a novel multi-task learning model called MPR (short for\nMulti-level POI Recommendation), where each task aims to return the top-k POIs\nat a certain spatial granularity level. Each task consists of two subtasks: (i)\nattribute-based representation learning; (ii) interaction-based representation\nlearning. The first subtask learns the feature representations for both users\nand POIs, capturing attributes directly from their profiles. The second subtask\nincorporates user-POI interactions into the model. Additionally, MPR can\nprovide insights into why certain recommendations are being made to a user\nbased on three types of hints: user-aspect, POI-aspect, and interaction-aspect.\nWe empirically validate our approach using two real-life datasets, and show\npromising performance improvements over several state-of-the-art methods.\n"}
{"abstract": "  We show that all resonances in Dirichlet obstacle scattering (in $\\mathbb{C}$\nin odd dimensions and in the logarithmic cover of $\\mathbb{C}\\setminus\\{0\\}$ in\neven dimensions) are generically simple in the class of obstacles with $C^k$\n(and $C^\\infty$) boundaries, $k \\geq 2$.\n"}
{"abstract": "  In this paper, we describe our participation in the TREC Health\nMisinformation Track 2020. We submitted $11$ runs to the Total Recall Task and\n13 runs to the Ad Hoc task. Our approach consists of 3 steps: (1) we create an\ninitial run with BM25 and RM3; (2) we estimate credibility and misinformation\nscores for the documents in the initial run; (3) we merge the relevance,\ncredibility and misinformation scores to re-rank documents in the initial run.\nTo estimate credibility scores, we implement a classifier which exploits\nfeatures based on the content and the popularity of a document. To compute the\nmisinformation score, we apply a stance detection approach with a pretrained\nTransformer language model. Finally, we use different approaches to merge\nscores: weighted average, the distance among score vectors and rank fusion.\n"}
{"abstract": "  Two-dimensional (2D) multiferroic materials with controllable magnetism have\npromising prospects in miniaturized quantum device applications, such as\nhigh-density data storage and spintronic devices. Here, using first-principles\ncalculations, we propose a coexistence of antiferromagnetism and\nferroelasticity in multiferroic $MnF_{4}$ monolayer. The $MnF_{4}$ monolayer is\nfound to be an intrinsic wide-gap semiconductor with large spin polarization ~3\n$\\mu_{B}$/Mn, in which the antiferromagnetic order originates from the\ncooperation and competition of the direct exchange and super exchange.\n$MnF_{4}$ monolayer is also characterized by strongly uniaxial magnetic\nanisotropic behavior, that can be manipulated by the reversible ferroelastic\nstrain and carrier doping. Remarkably, the carrier doping not only leads to an\nantiferromagnetic to ferromagnetic phase transformation, bult also could switch\nthe easy magnetization axis between the in-plane and out-of-plane directions.\nIn addition, the N\\'eel temperature was evaluated to be about 140 K from the\nMonte Carlo simulations based on the Heisenberg model. The combination of\nantiferromagnetic and ferroelastic properties in $MnF_{4}$ monolayer provides a\npromising platform for studying the magnetoelastic effects, and brings about\nnew concepts for next-generation nonvolatile memory and multi-stage storage.\n"}
{"abstract": "  The equilibrium states of one-dimensional proton conductors in the systems\nwith hydrogen bonds are investigated. Our extended hard-core boson lattice\nmodel includes short-range interactions between hydrogen ions, their transfer\nalong the hydrogen bonds with two-minima local anharmonic potential, as well as\ntheir inter-bond hopping, and the modulating field is taken into account. The\nexact diagonalization method for finite one-dimensional system with periodic\nboundary conditions is used. The existence of various phases of the system at\n$T = 0$, depending on the values of short-range interactions between particles\nand the modulating field strength, is established by analyzing the character of\nthe obtained frequency dependence of one-particle spectral density; the phase\ndiagrams are built.\n"}
{"abstract": "  Training Restricted Boltzmann Machines (RBMs) has been challenging for a long\ntime due to the difficulty of computing precisely the log-likelihood gradient.\nOver the past decades, many works have proposed more or less successful\ntraining recipes but without studying the crucial quantity of the problem: the\nmixing time, i.e. the number of Monte Carlo iterations needed to sample new\nconfigurations from a model. In this work, we show that this mixing time plays\na crucial role in the dynamics and stability of the trained model, and that\nRBMs operate in two well-defined regimes, namely equilibrium and\nout-of-equilibrium, depending on the interplay between this mixing time of the\nmodel and the number of steps, $k$, used to approximate the gradient. We\nfurther show empirically that this mixing time increases with the learning,\nwhich often implies a transition from one regime to another as soon as $k$\nbecomes smaller than this time. In particular, we show that using the popular\n$k$ (persistent) contrastive divergence approaches, with $k$ small, the\ndynamics of the learned model are extremely slow and often dominated by strong\nout-of-equilibrium effects. On the contrary, RBMs trained in equilibrium\ndisplay faster dynamics, and a smooth convergence to dataset-like\nconfigurations during the sampling. Finally we discuss how to exploit in\npractice both regimes depending on the task one aims to fulfill: (i) short $k$\ncan be used to generate convincing samples in short learning times, (ii) large\n$k$ (or increasingly large) is needed to learn the correct equilibrium\ndistribution of the RBM. Finally, the existence of these two operational\nregimes seems to be a general property of energy based models trained via\nlikelihood maximization.\n"}
{"abstract": "  We investigate radial oscillations of pure neutron stars and hybrid stars,\nemploying equations of state of nuclear matter from Brueckner-Hartree-Fock\ntheory, and of quark matter from the Dyson-Schwinger quark model, performing a\nGibbs construction for the mixed phase in hybrid stars. We calculate the\neigenfrequencies and corresponding oscillation functions. Our results for the\nzero points of the first-order radial oscillation frequencies give the maximum\nmass of stable neutron stars, consistent with the common criterion\n$dM/d\\rho_c=0$. Possible observations of the radial oscillation frequencies\ncould help to learn more about the equation of state, predict the maximum mass\nof neutron stars more precisely, and indicate the presence of quark matter.\n"}
{"abstract": "  This paper is a scoping study to identify current methods used in handling\nsparse data with contextual bandits in web settings. The area is highly current\nand state of the art methods are identified. The years 2017-2020 are\ninvestigated, and 19 method articles are identified, and two review articles.\nFive categories of methods are described, making it easy to choose how to\naddress sparse data using contextual bandits with a method available for\nmodification in the specific setting of concern. In addition, each method has\nmultiple techniques to choose from for future evaluation. The problem areas are\nalso mentioned that each article covers. An overall updated understanding of\nsparse data problems using contextual bandits in web settings is given. The\nidentified methods are policy evaluation (off-line and on-line) ,\nhybrid-method, model representation (clusters and deep neural networks),\ndimensionality reduction, and simulation.\n"}
{"abstract": "  Latent heat thermal energy storage (LHTES) has been recommended as an\neffective technology to the thermal management system of space exploration for\nits excellent ability of storing thermal energy. However, it is well known that\nthe low thermal conductivity of phase change material (PCM) seriously weakens\nthe heat charging and discharging rates of LHTES system. In present study, the\nelectrohydrodynamic (EHD), which is a popular active heat transfer enhancement\ntechnology, is introduced to enhance the PCM melting in a shell-tube LHTES\nsystem under microgravity. In our numerical simulations, we mainly focus on the\ncombined effects of the electric Rayleigh number $T$ and the eccentricity\n$\\Gamma$ on the melting performance under microgravity. Based on the numerical\nresults, it is found that in contrast to the case without the electric field,\nthe presence of the electric field causes the heat transfer efficiency and\nmelting behavior of LHTES system to be enhanced significantly. In addition, our\nresults indicates that the EHD technique always shows a good performance in\naccelerating the melting process even under microgravity, and specially, the\nmaximum time saving in some cases is more than $90\\%$. Furthermore, we note\nthat although the concentric annulus is always the optimal configuration under\nno-gravity condition, the optimal eccentric position of the internal tube\nstrongly depends on the electric Rayleigh number if the gravity effects are\ntaken into account.\n"}
{"abstract": "  We report non-invasive detection of spin coherence in a collection of\nRaman-driven cold atoms using dispersive Faraday rotation fluctuation\nmeasurements, which opens up new possibilities of probing spin correlations in\nquantum gases and other similar systems. We demonstrate five orders of\nmagnitude enhancement of the measured signal strength than the traditional spin\nnoise spectroscopy with thermal atoms in equilibrium. Our observations are in\ngood agreement with the comprehensive theoretical modeling of the driven atoms\nat various temperatures. The extracted spin relaxation rate of cold rubidium\natoms with atom number density $\\sim$10$^9/$cm$^3$ is of the order of\n2$\\pi\\times$0.5 kHz at 150 $\\mu$K, two orders of magnitude less than $\\sim$\n2$\\pi\\times$50 kHz of a thermal atomic vapor with atom number density\n$\\sim$10$^{12}/$cm$^3$ at 373 K.\n"}
{"abstract": "  Most recent works on sentiment analysis have exploited the text modality.\nHowever, millions of hours of video recordings posted on social media platforms\neveryday hold vital unstructured information that can be exploited to more\neffectively gauge public perception. Multimodal sentiment analysis offers an\ninnovative solution to computationally understand and harvest sentiments from\nvideos by contextually exploiting audio, visual and textual cues. In this\npaper, we, firstly, present a first of its kind Persian multimodal dataset\ncomprising more than 800 utterances, as a benchmark resource for researchers to\nevaluate multimodal sentiment analysis approaches in Persian language.\nSecondly, we present a novel context-aware multimodal sentiment analysis\nframework, that simultaneously exploits acoustic, visual and textual cues to\nmore accurately determine the expressed sentiment. We employ both\ndecision-level (late) and feature-level (early) fusion methods to integrate\naffective cross-modal information. Experimental results demonstrate that the\ncontextual integration of multimodal features such as textual, acoustic and\nvisual features deliver better performance (91.39%) compared to unimodal\nfeatures (89.24%).\n"}
{"abstract": "  Compressed sensing is a paradigm within signal processing that provides the\nmeans for recovering structured signals from linear measurements in a highly\nefficient manner. Originally devised for the recovery of sparse signals, it has\nbecome clear that a similar methodology would also carry over to a wealth of\nother classes of structured signals. In this work, we provide an overview over\nthe theory of compressed sensing for a particularly rich family of such\nsignals, namely those of hierarchically structured signals. Examples of such\nsignals are constituted by blocked vectors, with only few non-vanishing sparse\nblocks. We present recovery algorithms based on efficient hierarchical\nhard-thresholding. The algorithms are guaranteed to converge, in a stable\nfashion both with respect to measurement noise as well as to model mismatches,\nto the correct solution provided the measurement map acts isometrically\nrestricted to the signal class. We then provide a series of results\nestablishing the required condition for large classes of measurement ensembles.\nBuilding upon this machinery, we sketch practical applications of this\nframework in machine-type communications and quantum tomography.\n"}
{"abstract": "  Several social factors impact how people respond to AI explanations used to\njustify AI decisions affecting them personally. In this position paper, we\ndefine a framework called the \\textit{layers of explanation} (LEx), a lens\nthrough which we can assess the appropriateness of different types of\nexplanations. The framework uses the notions of \\textit{sensitivity} (emotional\nresponsiveness) of features and the level of \\textit{stakes} (decision's\nconsequence) in a domain to determine whether different types of explanations\nare \\textit{appropriate} in a given context. We demonstrate how to use the\nframework to assess the appropriateness of different types of explanations in\ndifferent domains.\n"}
{"abstract": "  Conversations on social media (SM) are increasingly being used to investigate\nsocial issues on the web, such as online harassment and rumor spread. For such\nissues, a common thread of research uses adversarial reactions, e.g., replies\npointing out factual inaccuracies in rumors. Though adversarial reactions are\nprevalent in online conversations, inferring those adverse views (or stance)\nfrom the text in replies is difficult and requires complex natural language\nprocessing (NLP) models. Moreover, conventional NLP models for stance mining\nneed labeled data for supervised learning. Getting labeled conversations can\nitself be challenging as conversations can be on any topic, and topics change\nover time. These challenges make learning the stance a difficult NLP problem.\n  In this research, we first create a new stance dataset comprised of three\ndifferent topics by labeling both users' opinions on the topics (as in pro/con)\nand users' stance while replying to others' posts (as in favor/oppose). As we\nfind limitations with supervised approaches, we propose a weakly-supervised\napproach to predict the stance in Twitter replies. Our novel method allows\nusing a smaller number of hashtags to generate weak labels for Twitter replies.\nCompared to supervised learning, our method improves the mean F1-macro by 8\\%\non the hand-labeled dataset without using any hand-labeled examples in the\ntraining set. We further show the applicability of our proposed method on COVID\n19 related conversations on Twitter.\n"}
{"abstract": "  The quantum Kibble-Zurek mechanism (QKZM) predicts universal dynamical\nbehavior in the vicinity of quantum phase transitions (QPTs). It is now well\nunderstood for one-dimensional quantum matter. Higher-dimensional systems,\nhowever, remain a challenge, complicated by fundamental differences of the\nassociated QPTs and their underlying conformal field theories. In this work, we\ntake the first steps towards exploring the QKZM in two dimensions. We study the\ndynamical crossing of the QPT in the paradigmatic Ising model by a joint effort\nof modern state-of-the-art numerical methods. As a central result, we quantify\nuniversal QKZM behavior close to the QPT. However, upon traversing further into\nthe ferromagnetic regime, we observe deviations from the QKZM prediction. We\nexplain the observed behavior by proposing an {\\it extended QKZM} taking into\naccount spectral information as well as phase ordering. Our work provides a\nstarting point towards the exploration of dynamical universality in\nhigher-dimensional quantum matter.\n"}
{"abstract": "  These brief lecture notes cover the basics of neural networks and deep\nlearning as well as their applications in the quantum domain, for physicists\nwithout prior knowledge. In the first part, we describe training using\nbackpropagation, image classification, convolutional networks and autoencoders.\nThe second part is about advanced techniques like reinforcement learning (for\ndiscovering control strategies), recurrent neural networks (for analyzing time\ntraces), and Boltzmann machines (for learning probability distributions). In\nthe third lecture, we discuss first recent applications to quantum physics,\nwith an emphasis on quantum information processing machines. Finally, the\nfourth lecture is devoted to the promise of using quantum effects to accelerate\nmachine learning.\n"}
{"abstract": "  This study is motivated by the question of how singularity formation and\nother forms of extreme behavior in nonlinear dissipative partial differential\nequations are affected by stochastic excitations. To address this question we\nconsider the 1D fractional Burgers equation with additive colored noise as a\nmodel problem. This system is interesting, because in the deterministic setting\nit exhibits finite-time blow-up or a globally well-posed behavior depending on\nthe value of the fractional dissipation exponent. The problem is studied by\nperforming a series of accurate numerical computations combining\nspectrally-accurate spatial discretization with a Monte-Carlo approach. First,\nwe carefully document the singularity formation in the deterministic system in\nthe supercritical regime where the blow-up time is shown to be a decreasing\nfunction of the fractional dissipation exponent. Our main result for the\nstochastic problem is that there is no evidence for the noise to regularize the\nevolution by suppressing blow-up in the supercritical regime, or for the noise\nto trigger blow-up in the subcritical regime. However, as the noise amplitude\nbecomes large, the blow-up times in the supercritical regime are shown to\nexhibit an increasingly non-Gaussian behavior. Analogous observations are also\nmade for the maximum attained values of the enstrophy and the times when the\nmaxima occur in the subcritical regime.\n"}
{"abstract": "  Pairs of photons entangled in their time-frequency degree of freedom are of\ngreat interest in quantum optics research and applications, due to their\nrelative ease of generation and their high capacity for encoding information.\nHere we analyze, both theoretically and experimentally, the behavior of\nphase-insensitive spectrally-resolved interferences arising from two pairs of\ntime-frequency entangled photons. At its core, this is a multimode entanglement\nswapping experiment, whereby a spectrally resolved joint measurement on the\nidler photons from both pairs results in projecting the signal photons onto a\nBell state whose form depends on the measurement outcome. Our analysis is a\nthorough exploration of what can be achieved using time-frequency entanglement\nand spectrally-resolved Bell-state measurements.\n"}
{"abstract": "  Public knowledge graphs such as DBpedia and Wikidata have been recognized as\ninteresting sources of background knowledge to build content-based recommender\nsystems. They can be used to add information about the items to be recommended\nand links between those. While quite a few approaches for exploiting knowledge\ngraphs have been proposed, most of them aim at optimizing the recommendation\nstrategy while using a fixed knowledge graph. In this paper, we take a\ndifferent approach, i.e., we fix the recommendation strategy and observe\nchanges when using different underlying knowledge graphs. Particularly, we use\ndifferent language editions of DBpedia. We show that the usage of different\nknowledge graphs does not only lead to differently biased recommender systems,\nbut also to recommender systems that differ in performance for particular\nfields of recommendations.\n"}
{"abstract": "  Neovascularisation is essential for tissue development and regeneration, in\naddition to playing a key role in pathological settings such as ischemia and\ntumour development. Experimental findings in the past two decades have led to\nthe identification of a new mechanism of neovascularisation, cluster-based\nvasculogenesis, during which endothelial progenitor cells (EPCs) mobilised from\nthe bone marrow are capable of bridging distant vascular beds in a variety of\nhypoxic settings in vivo. This process is characterised by the formation of EPC\nclusters during its early stages and, while much progress has been made in\nidentifying various mechanisms underlying cluster formation, we are still far\nfrom a comprehensive description of such spatio-temporal dynamics. In order to\nachieve this, we propose a novel mathematical model of the early stages of\ncluster-based vasculogenesis, comprising of a system of nonlocal partial\ndifferential equations including key mechanisms such as endogenous chemotaxis,\nmatrix degradation, cell proliferation and cell-to-cell adhesion. We conduct a\nlinear stability analysis on the system, solve the equations numerically,\nconduct a parametric analysis of the numerical solutions of the 1D problem to\ninvestigate the role of underlying dynamics on the speed of cluster formation\nand the size of clusters, and verify the key results of the parametric analysis\nwith simulations of the 2D problem. Our results, which qualitatively compare\nwith data from in vitro experiments, elucidate the complementary role played by\nendogenous chemotaxis and matrix degradation in the formation of clusters, and\nthey indicate that previous approaches to the nonlocal modelling of\ncell-to-cell adhesion, while they capture the aggregating effect of\ncell-to-cell adhesion, are not sufficient to capture its stabilising effect on\nclusters, and new continuum cell-adhesion modelling strategies are required.\n"}
{"abstract": "  In this work the robustness of a feedback optimization scheme is discussed.\nPreviously known results in literature, on the convergence to local optima of\nthe optimization problem of interest, are extended to the case where the\nsensitivities of the steady-state input-output map of the plant present bounded\nuncertainties. The application of the scheme to a biological setting, with the\ngoal of maximizing the concentration of products of interest in a bioreactor,\nunder a continuous perfusion framework, is suggested and the potential of the\napproach is exposed by means of a simple synthetic example. This extended\nversion contains also the Numerical data for example in \"Robustness of a\nfeedback optimization scheme with application to bioprocess manufacturing\"\nreport, for the sake of reproducibility of the results.\n"}
{"abstract": "  Using Planck polarization data, we search for and constrain spatial\nvariations of the polarized dust foreground for cosmic microwave background\n(CMB) observations, specifically in its spectral index, $\\beta_d$. Failure to\naccount for such variations will cause errors in the foreground cleaning that\npropagate into errors on cosmological parameter recovery from the cleaned CMB\nmap. It is unclear how robust prior studies of the Planck data which\nconstrained $\\beta_d$ variations are due to challenges with noise modeling,\nresidual systematics, and priors. To clarify constraints on $\\beta_d$ and its\nvariation, we employ two pixel space analyses of the polarized dust foreground\nat $>3.7^{\\circ}$ scales on $\\approx 60\\%$ of the sky at high Galactic\nlatitudes. A template fitting method, which measures $\\beta_d$ over three\nregions of $\\approx 20\\%$ of the sky, does not find significant deviations from\nan uniform $\\beta_d = 1.55$, consistent with prior Planck determinations. An\nadditional analysis in these regions, based on multifrequency fits to a dust\nand CMB model per pixel, puts limits on $\\sigma_{\\beta_d}$, the Gaussian\nspatial variation in $\\beta_d$. At the highest latitudes, the data support\n$\\sigma_{\\beta_d}$ up to $0.45$, $0.30$ at mid-latitudes, and $0.15$ at\nlow-latitudes. We also demonstrate that care must be taken when interpreting\nthe current Planck constraints, $\\beta_d$ maps, and noise simulations. Due to\nresidual systematics and low dust signal to noise at high latitudes, forecasts\nfor ongoing and future missions should include the possibility of large values\nof $\\sigma_{\\beta_d}$ as estimated in this paper, based on current polarization\ndata.\n"}
{"abstract": "  Dynamic recommendation is essential for modern recommender systems to provide\nreal-time predictions based on sequential data. In real-world scenarios, the\npopularity of items and interests of users change over time. Based on this\nassumption, many previous works focus on interaction sequences and learn\nevolutionary embeddings of users and items. However, we argue that\nsequence-based models are not able to capture collaborative information among\nusers and items directly. Here we propose Dynamic Graph Collaborative Filtering\n(DGCF), a novel framework leveraging dynamic graphs to capture collaborative\nand sequential relations of both items and users at the same time. We propose\nthree update mechanisms: zero-order 'inheritance', first-order 'propagation',\nand second-order 'aggregation', to represent the impact on a user or item when\na new interaction occurs. Based on them, we update related user and item\nembeddings simultaneously when interactions occur in turn, and then use the\nlatest embeddings to make recommendations. Extensive experiments conducted on\nthree public datasets show that DGCF significantly outperforms the\nstate-of-the-art dynamic recommendation methods up to 30. Our approach achieves\nhigher performance when the dataset contains less action repetition, indicating\nthe effectiveness of integrating dynamic collaborative information.\n"}
{"abstract": "  This paper proposes a novel active boundary loss for semantic segmentation.\nIt can progressively encourage the alignment between predicted boundaries and\nground-truth boundaries during end-to-end training, which is not explicitly\nenforced in commonly used cross-entropy loss. Based on the predicted boundaries\ndetected from the segmentation results using current network parameters, we\nformulate the boundary alignment problem as a differentiable direction vector\nprediction problem to guide the movement of predicted boundaries in each\niteration. Our loss is model-agnostic and can be plugged into the training of\nsegmentation networks to improve the boundary details. Experimental results\nshow that training with the active boundary loss can effectively improve the\nboundary F-score and mean Intersection-over-Union on challenging image and\nvideo object segmentation datasets.\n"}
{"abstract": "  Methyl-ammonium lead iodide perovskite crystallizes in different structures\ndepending on the temperature: orthorhombic, tetragonal and cubic. An important\npoint to be considered is the effect of the microscopic properties of the\ndifferent structures on the optical and photovoltaic properties. Using first\nprinciples we obtain the absorption coefficients that will determine the\nabsorption of solar radiation. In order to analyze the contributions of the\ndifferent atoms to the absorption coefficients we split them into a\nmany-species expansion. Using a similar methodology we also split the\nefficiencies as a many-species expansion. It allows the contribution of the\natomic species to be identified and quantified to the absorption coefficients\nand to the solar cell efficiency of the different phases. Additionally the\neffect of the cell thickness w is quantified.\n"}
{"abstract": "  This paper addresses the construction and the stability of self-similar\nsolutions to the isentropic compressible Euler equations. These solutions model\na gas that implodes isotropically, ending in a singularity formation in finite\ntime. The existence of smooth solutions that vanish at infinity and do not have\nvacuum regions was recently proved and, in this paper, we provide the first\nconstruction of such smooth profiles, the first characterization of their\nspectrum of radial perturbations as well as some endpoints of unstable\ndirections. Numerical simulations of the Euler equations provide evidence that\none of these endpoints is a shock formation that happens before the singularity\nat the origin, showing that the implosion process is unstable.\n"}
{"abstract": "  Visual change detection, aiming at segmentation of video frames into\nforeground and background regions, is one of the elementary tasks in computer\nvision and video analytics. The applications of change detection include\nanomaly detection, object tracking, traffic monitoring, human machine\ninteraction, behavior analysis, action recognition, and visual surveillance.\nSome of the challenges in change detection include background fluctuations,\nillumination variation, weather changes, intermittent object motion, shadow,\nfast/slow object motion, camera motion, heterogeneous object shapes and\nreal-time processing. Traditionally, this problem has been solved using\nhand-crafted features and background modelling techniques. In recent years,\ndeep learning frameworks have been successfully adopted for robust change\ndetection. This article aims to provide an empirical review of the\nstate-of-the-art deep learning methods for change detection. More specifically,\nwe present a detailed analysis of the technical characteristics of different\nmodel designs and experimental frameworks. We provide model design based\ncategorization of the existing approaches, including the 2D-CNN, 3D-CNN,\nConvLSTM, multi-scale features, residual connections, autoencoders and GAN\nbased methods. Moreover, an empirical analysis of the evaluation settings\nadopted by the existing deep learning methods is presented. To the best of our\nknowledge, this is a first attempt to comparatively analyze the different\nevaluation frameworks used in the existing deep change detection methods.\nFinally, we point out the research needs, future directions and draw our own\nconclusions.\n"}
{"abstract": "  General Relativity provides us with an extremely powerful tool to extract at\nthe same time astrophysical and cosmological information from the Stochastic\nGravitational Wave Backgrounds (SGWBs): the cross-correlation with other\ncosmological tracers, since their anisotropies share a common origin and the\nsame perturbed geodesics. In this letter we explore the cross-correlation of\nthe cosmological and astrophysical SGWBs with Cosmic Microwave Background (CMB)\nanisotropies, showing that future GW detectors, such as LISA or BBO, have the\nability to measure such cross-correlation signals. We also present, as a new\ntool in this context, constrained realization maps of the SGWBs extracted from\nthe high-resolution CMB {\\it Planck} maps. This technique allows, in the\nlow-noise regime, to faithfully reconstruct the expected SGWB map by starting\nfrom CMB measurements.\n"}
{"abstract": "  Principal Component Analysis (PCA) is the workhorse tool for dimensionality\nreduction in this era of big data. While often overlooked, the purpose of PCA\nis not only to reduce data dimensionality, but also to yield features that are\nuncorrelated. Furthermore, the ever-increasing volume of data in the modern\nworld often requires storage of data samples across multiple machines, which\nprecludes the use of centralized PCA algorithms. This paper focuses on the dual\nobjective of PCA, namely, dimensionality reduction and decorrelation of\nfeatures, but in a distributed setting. This requires estimating the\neigenvectors of the data covariance matrix, as opposed to only estimating the\nsubspace spanned by the eigenvectors, when data is distributed across a network\nof machines. Although a few distributed solutions to the PCA problem have been\nproposed recently, convergence guarantees and/or communications overhead of\nthese solutions remain a concern. With an eye towards communications\nefficiency, this paper introduces a feedforward neural network-based one\ntime-scale distributed PCA algorithm termed Distributed Sanger's Algorithm\n(DSA) that estimates the eigenvectors of the data covariance matrix when data\nis distributed across an undirected and arbitrarily connected network of\nmachines. Furthermore, the proposed algorithm is shown to converge linearly to\na neighborhood of the true solution. Numerical results are also provided to\ndemonstrate the efficacy of the proposed solution.\n"}
{"abstract": "  In autonomous applications for mobility and transport, a high-rate and highly\naccurate vehicle states estimation is achieved by fusing measurements of global\nnavigation satellite systems and inertial sensors. Since this kind of state\nestimation suffers from poor parameterization, environment disturbances, or\neven software and hardware failures, this paper introduces a novel scheme of\nmulti-sensor navigation system involving extended H$_\\infty$ filter for\nrobustness enhancement of the navigation solution and zonotope for protection\nlevel generation in combination with vehicle dynamic-model-aided fault\ndetection of the inertial sensor for reliable integrity monitoring. The\ninnovative scheme, applying extended H$_\\infty$ filter and zonotope, is shown\nas part of a tightly-coupled navigation system. Further, the consideration of\nredundant information, e.g., vehicle dynamic model, for fault detection purpose\nhas long been investigated and is systematically described and discussed using\ninterval analysis theory in current publication. The robustness of the designed\napproach is validated with real-world data in post-processing: decimeter\npositioning accuracy is maintained, while the solution of conventional extended\nKalman filter diverges from ground truth; the difference is also significant\nunder inertial sensor faults. A real-time implementation of the designed\napproach is promising and aimed in the future work.\n"}
{"abstract": "  CP violation in charm decay was observed in the decays $D^0\\rightarrow\nP^{\\pm}P^{\\mp}$ of a $D^0$ meson to two pseudoscalars. When interpreted within\nthe SM, the results imply that the ratio of the relevant rescattering\namplitudes has a magnitude and phase that are both of $O(1)$. We discuss ways\nto probe similar ratios in $D^0\\rightarrow V^{\\pm}P^{\\mp}$ decays, where $V$ is\na vector that decays to two pseudoscalars, from the Dalitz-plot analysis of\ntime-integrated three-body decays. Compared to two-body decays, three-body\ndecays have the advantage that the complete system can be solved without the\nneed for time-dependent CP violation measurements or use of correlated\n$D^0$--$\\overline{D}^0$ production. We discuss the decays $D^0\\rightarrow\n\\pi^+\\pi^-\\pi^0$ and $D^0\\rightarrow K^+K^-\\pi^0$ as examples by considering a\ntoy model of only two overlapping charged resonances, treating the underlying\npseudo two-body decays in full generality.\n"}
{"abstract": "  We consider the horocyclic flow corresponding to a (topologically mixing)\nAnosov flow or diffeomorphism, and establish the uniqueness of transverse\nquasi-invariant measures with H\\\"older Jacobians. In the same setting, we give\na precise characterization of the equilibrium states of the hyperbolic system,\nshowing that existence of a family of Radon measures on the horocyclic\nfoliation such that any probability (invariant or not) having conditionals\ngiven by this family, necessarily is the unique equilibrium state of the\nsystem.\n"}
{"abstract": "  Advancing the utility of social media data for research applications requires\nmethods for automatically detecting demographic information about social media\nstudy populations, including users' age. The objective of this study was to\ndevelop and evaluate a method that automatically identifies the exact age of\nusers based on self-reports in their tweets. Our end-to-end automatic natural\nlanguage processing (NLP) pipeline, ReportAGE, includes query patterns to\nretrieve tweets that potentially mention an age, a classifier to distinguish\nretrieved tweets that self-report the user's exact age (\"age\" tweets) and those\nthat do not (\"no age\" tweets), and rule-based extraction to identify the age.\nTo develop and evaluate ReportAGE, we manually annotated 11,000 tweets that\nmatched the query patterns. Based on 1000 tweets that were annotated by all\nfive annotators, inter-annotator agreement (Fleiss' kappa) was 0.80 for\ndistinguishing \"age\" and \"no age\" tweets, and 0.95 for identifying the exact\nage among the \"age\" tweets on which the annotators agreed. A deep neural\nnetwork classifier, based on a RoBERTa-Large pretrained model, achieved the\nhighest F1-score of 0.914 (precision = 0.905, recall = 0.942) for the \"age\"\nclass. When the age extraction was evaluated using the classifier's\npredictions, it achieved an F1-score of 0.855 (precision = 0.805, recall =\n0.914) for the \"age\" class. When it was evaluated directly on the held-out test\nset, it achieved an F1-score of 0.931 (precision = 0.873, recall = 0.998) for\nthe \"age\" class. We deployed ReportAGE on more than 1.2 billion tweets posted\nby 245,927 users, and predicted ages for 132,637 (54%) of them. Scaling the\ndetection of exact age to this large number of users can advance the utility of\nsocial media data for research applications that do not align with the\npredefined age groupings of extant binary or multi-class classification\napproaches.\n"}
{"abstract": "  Measurement error is a pervasive issue which renders the results of an\nanalysis unreliable. The measurement error literature contains numerous\ncorrection techniques, which can be broadly divided into those which aim to\nproduce exactly consistent estimators, and those which are only approximately\nconsistent. While consistency is a desirable property, it is typically attained\nonly under specific model assumptions. Two techniques, regression calibration\nand simulation extrapolation, are used frequently in a wide variety of\nparametric and semiparametric settings. However, in many settings these methods\nare only approximately consistent. We generalize these corrections, relaxing\nassumptions placed on replicate measurements. Under regularity conditions, the\nestimators are shown to be asymptotically normal, with a sandwich estimator for\nthe asymptotic variance. Through simulation, we demonstrate the improved\nperformance of the modified estimators, over the standard techniques, when\nthese assumptions are violated. We motivate these corrections using the\nFramingham Heart Study, and apply the generalized techniques to an analysis of\nthese data.\n"}
{"abstract": "  The Wigner function is a quantum analogue of the classical joined\ndistribution of position and momentum. As such is should be a good tool to\nstudy quantum-classical correspondence. In this paper, the classical limit of\nthe Wigner function is shown using the quantum harmonic oscillator as an\nexample. The Wigner function is found exactly for all states. The\nsemi-classical wavefunctions for highly excited states are used as the approach\nto the classical limit. Therefore, one can found the classical limit of the\nWigner function for highly excited states and shown that it gives the classical\nmicrocanonical ensemble.\n"}
{"abstract": "  This paper presents a Tracking-Error Learning Control (TELC) algorithm for\nprecise mobile robot path tracking in off-road terrain. In traditional tracking\nerror-based control approaches, feedback and feedforward controllers are\ndesigned based on the nominal model which cannot capture the uncertainties,\ndisturbances and changing working conditions so that they cannot ensure precise\npath tracking performance in the outdoor environment. In TELC algorithm, the\nfeedforward control actions are updated by using the tracking error dynamics\nand the plant-model mismatch problem is thus discarded. Therefore, the\nfeedforward controller gradually eliminates the feedback controller from the\ncontrol of the system once the mobile robot has been on-track. In addition to\nthe proof of the stability, it is proven that the cost functions do not have\nlocal minima so that the coefficients in TELC algorithm guarantee that the\nglobal minimum is reached. The experimental results show that the TELC\nalgorithm results in better path tracking performance than the traditional\ntracking error-based control method. The mobile robot controlled by TELC\nalgorithm can track a target path precisely with less than $10$ cm error in\noff-road terrain.\n"}
{"abstract": "  Discovered in the context of card shuffling by Aldous, Diaconis and\nShahshahani, the cutoff phenomenon has since then been established in a variety\nof Markov chains. However, proving cutoff remains a delicate affair, which\nrequires a detailed knowledge of the chain. Identifying the general mechanisms\nunderlying this phase transition -- without having to pinpoint its precise\nlocation -- remains one of the most fundamental open problems in the area of\nmixing times. In the present paper, we make a step in this direction by\nestablishing cutoff for Markov chains with non-negative curvature, under a\nsuitably refined product condition. The result applies, in particular, to\nrandom walks on abelian Cayley expanders satisfying a mild degree condition,\nhence in particular to \\emph{almost all} abelian Cayley graphs. Our proof\nrelies on a quantitative \\emph{entropic concentration principle}, which we\nbelieve to lie behind all cutoff phenomena.\n"}
{"abstract": "  We answer three questions posed by Bubeck and Linial on the limit densities\nof subtrees in trees. We prove there exist positive $\\varepsilon_1$ and\n$\\varepsilon_2$ such that every tree that is neither a path nor a star has\ninducibility at most $1-\\varepsilon_1$, where the inducibility of a tree $T$ is\ndefined as the maximum limit density of $T$, and that there are infinitely many\ntrees with inducibility at least $\\varepsilon_2$. Finally, we construct a\nuniversal sequence of trees; that is, a sequence in which the limit density of\nany tree is positive.\n"}
{"abstract": "  We present a knowledge-grounded dialog system developed for the ninth Dialog\nSystem Technology Challenge (DSTC9) Track 1 - Beyond Domain APIs: Task-oriented\nConversational Modeling with Unstructured Knowledge Access. We leverage\ntransfer learning with existing language models to accomplish the tasks in this\nchallenge track. Specifically, we divided the task into four sub-tasks and\nfine-tuned several Transformer models on each of the sub-tasks. We made\nadditional changes that yielded gains in both performance and efficiency,\nincluding the combination of the model with traditional entity-matching\ntechniques, and the addition of a pointer network to the output layer of the\nlanguage model.\n"}
{"abstract": "  The strong coupling constants are basic quantities that carry information of\nthe strong interactions among the baryon and meson multiplets as well as\ninformation on the natures and internal structures of the involved hadrons.\nThese parameters enter to the transition matrix elements of various decays as\nmain inputs and they play key roles in analyses of the experimental data\nincluding various hadrons. We determine the strong coupling constants among the\ndoubly heavy spin-$ 3/2 $ baryons, $\\Xi^*_{QQ'} $ and $\\Omega^*_{QQ'}$, and\nlight pseudoscalar mesons, $\\pi$, $K$ and $\\eta$, using the light-cone QCD. The\nvalues obtained for the strong coupling constants under study may be used in\nconstruction of the strong potentials among the doubly heavy spin-3/2 baryons\nand light pseudoscalar mesons.\n"}
{"abstract": "  We present new equilibrium solutions of stationary models of magnetized thick\ndisks (or tori) around Kerr black holes with synchronised scalar hair. The\nmodels reported here largely extend our previous results based on constant\nradial distributions of the specific angular momentum along the equatorial\nplane. We introduce a new way to prescribe the distribution of the disk's\nangular momentum based on a combination of two previous proposals and compute\nthe angular momentum distribution outside the equatorial plane by resorting to\nthe construction of von Zeipel cylinders. We find that the effect of the scalar\nhair on the black hole spacetime can yield significant differences in the disk\nmorphology and properties compared to what is found if the spacetime is purely\nKerr. Some of the tori built within the most extreme, background hairy black\nhole spacetime of our sample exhibit the appearance of two maxima in the\ngravitational energy density which impacts the radial profile distributions of\nthe disk's thermodynamical quantities. The models reported in this paper can be\nused as initial data for numerical evolutions with GRMHD codes to study their\nstability properties. Moreover, they can be employed as illuminating sources to\nbuild shadows of Kerr black holes with scalar hair which might help further\nconstrain the no-hair hypothesis as new observational data is collected.\n"}
{"abstract": "  Consider a random Schr\\\"odinger-type operator of the form $H:=-H_X+V+\\xi$\nacting on a general graph $\\mathscr G=(\\mathscr V,\\mathscr E)$, where $H_X$ is\nthe generator of a Markov process $X$ on $\\mathscr G$, $V$ is a deterministic\npotential with sufficient growth (so that $H$ has a purely discrete spectrum),\nand $\\xi$ is a random noise with at-most-exponential tails. We prove that $H$'s\neigenvalue point process is number rigid in the sense of Ghosh and Peres (Duke\nMath. J. 166 (2017), no. 10, 1789--1858); that is, the number of eigenvalues in\nany bounded domain $B\\subset\\mathbb C$ is determined by the configuration of\neigenvalues outside of $B$. Our general setting allows to treat cases where $X$\ncould be non-symmetric (hence $H$ is non-self-adjoint) and $\\xi$ has long-range\ndependence. Our strategy of proof consists of controlling the variance of the\ntrace of the semigroup $\\mathrm e^{-t H}$ using the Feynman-Kac formula.\n"}
{"abstract": "  Estimating the state of a deformable object is crucial for robotic\nmanipulation, yet accurate tracking is challenging when the object is\npartially-occluded. To address this problem, we propose an occlusion-robust\nRGBD sequence tracking framework based on Coherent Point Drift (CPD). To\nmitigate the effects of occlusion, our method 1) Uses a combination of locally\nlinear embedding and constrained optimization to regularize the output of CPD,\nthus enforcing topological consistency when occlusions create disconnected\npieces of the object; 2) Reasons about the free-space visible by an RGBD sensor\nto better estimate the prior on point location and to detect tracking failures\nduring occlusion; and 3) Uses shape descriptors to find the most relevant\nprevious state of the object to use for tracking after a severe occlusion. Our\nmethod does not rely on physics simulation or a physical model of the object,\nwhich can be difficult to obtain in unstructured environments. Despite having\nno physical model, our experiments demonstrate that our method achieves\nimproved accuracy in the presence of occlusion as compared to a physics-based\nCPD method while maintaining adequate run-time.\n"}
{"abstract": "  It has recently been demonstrated that black holes with spatially regular\nhorizons can support external scalar fields (scalar hairy configurations) which\nare non-minimally coupled to the Gauss-Bonnet invariant of the curved\nspacetime. The composed black-hole-scalar-field system is characterized by a\ncritical existence line $\\alpha=\\alpha(\\mu r_{\\text{H}})$ which, for a given\nmass of the supported scalar field, marks the threshold for the onset of the\nspontaneous scalarization phenomenon [here $\\{\\alpha,\\mu,r_{\\text{H}}\\}$ are\nrespectively the dimensionless non-minimal coupling parameter of the field\ntheory, the proper mass of the scalar field, and the horizon radius of the\ncentral supporting black hole]. In the present paper we use analytical\ntechniques in order to explore the physical and mathematical properties of the\nmarginally-stable composed black-hole-linearized-scalar-field configurations in\nthe eikonal regime $\\mu r_{\\text{H}}\\gg1$ of large field masses. In particular,\nwe derive a remarkably compact analytical formula for the critical\nexistence-line $\\alpha=\\alpha(\\mu r_{\\text{H}})$ of the system which separates\nbare Schwarzschild black-hole spacetimes from composed hairy (scalarized)\nblack-hole-field configurations.\n"}
{"abstract": "  \\emph{Uncertain Graph} (also known as \\emph{Probabilistic Graph}) is a\ngeneric model to represent many real\\mbox{-}world networks from social to\nbiological. In recent times analysis and mining of uncertain graphs have drawn\nsignificant attention from the researchers of the data management community.\nSeveral noble problems have been introduced and efficient methodologies have\nbeen developed to solve those problems. Hence, there is a need to summarize the\nexisting results on this topic in a self\\mbox{-}organized way. In this paper,\nwe present a comprehensive survey on uncertain graph mining focusing on mainly\nthree aspects: (i) different problems studied, (ii) computational challenges\nfor solving those problems, and (iii) proposed methodologies. Finally, we list\nout important future research directions.\n"}
{"abstract": "  The visibility of targets determines performance and even success rate of\nvarious applications, such as active slam, exploration, and target tracking.\nTherefore, it is crucial to take the visibility of targets into explicit\naccount in trajectory planning. In this paper, we propose a general metric for\ntarget visibility, considering observation distance and angle as well as\nocclusion effect. We formulate this metric into a differentiable visibility\ncost function, with which spatial trajectory and yaw can be jointly optimized.\nFurthermore, this visibility-aware trajectory optimization handles dynamic\nfeasibility of position and yaw simultaneously. To validate that our method is\npractical and generic, we integrate it into a customized quadrotor tracking\nsystem. The experimental results show that our visibility-aware planner\nperforms more robustly and observes targets better. In order to benefit related\nresearches, we release our code to the public.\n"}
{"abstract": "  Program SHELLSPEC is designed to calculate light curves, spectra, and images\nof interacting binaries and extra-solar planets immersed in a moving gaseous or\ndusty circumstellar matter. It solves simple radiative transfer along the line\nof sight in 3D moving media. The Roche model and synthetic spectra from the\nstellar atmosphere models such as TLUSTY from Ivan Hubeny can be used as a\nboundary condition for the radiative transfer. The latest publicly available\nversion is Shellspec39. The code has been combined with other methods such as\nDoppler tomography and interferometry and used to analyze spectroscopic,\nphotometric, and interferometric observations of binary stars and transiting\nexoplanets. A few examples are briefly mentioned.\n"}
{"abstract": "  In this article we study local constancy of the mod $p$ reduction of certain\n$2$-dimensional crystalline representations of\n$\\mathrm{Gal}\\left(\\bar{\\mathbb{Q}}_p/\\mathbb{Q}_p\\right)$ using the mod $p$\nlocal Langlands correspondence. We prove local constancy in the weight space by\ngiving an explicit lower bound on the local constancy radius centered around\nweights going up to $(p-1)^{2} +3$ and the slope fixed in $(0, \\ p-1)$\nsatisfying certain constraints. We establish the lower bound by determining\nexplicitly the mod $p$ reductions at nearby weights and applying a local\nconstancy result of Berger.\n"}
{"abstract": "  Performing reliable Bayesian inference on a big data scale is becoming a\nkeystone in the modern era of machine learning. A workhorse class of methods to\nachieve this task are Markov chain Monte Carlo (MCMC) algorithms and their\ndesign to handle distributed datasets has been the subject of many works.\nHowever, existing methods are not completely either reliable or computationally\nefficient. In this paper, we propose to fill this gap in the case where the\ndataset is partitioned and stored on computing nodes within a cluster under a\nmaster/slaves architecture. We derive a user-friendly centralised distributed\nMCMC algorithm with provable scaling in high-dimensional settings. We\nillustrate the relevance of the proposed methodology on both synthetic and real\ndata experiments.\n"}
{"abstract": "  A new search method for large polarization kernels is proposed. The algorithm\nproduces a kernel with given partial distances by employing depth-first search\ncombined with some methods which reduce the search space. Using the proposed\nmethod, we improved almost all existing lower bounds on the maximum rate of\npolarization for kernels of size from 17 to 27. We also obtained kernels which\nadmit low complexity processing by the recently proposed recursive trellis\nalgorithm. Numerical results demonstrate the advantage of polar codes with the\nproposed kernels compared with shortened polar codes and polar codes with small\nkernels.\n"}
{"abstract": "  Conventional breast cancer imaging techniques are nowadays based on the use\nof ionising radiations or ultrasound waves for the inspection of breast areas.\nNevertheless, these conventional techniques present some drawbacks related to\npatient safety, processing time and resolution issues. In this framework,\nmicrowave imaging can represent a valid alternative or a complementary\ntechnique compared to other conventional medical imaging modalities since it is\nsafe (using non-ionising radiations), relatively cheap and more comfortable\nfrom patient point of view. Unfortunately, it is slow and computationally\nexpensive, which strongly limit its use in clinical scenarios. In this paper,\nan artificial neural network for effective and almost real-time breast imaging\nis proposed. First, a realistic breast-like phantom generator was developed for\ntraining the network. Subsequently, numerical analyses have been conducted for\nthe optimisation and the performance evaluation of the approach. The results\nseem very promising in terms of recovery performance as well as for the\ncomputation burden.\n"}
{"abstract": "  Realtime model learning proves challenging for complex dynamical systems,\nsuch as drones flying in variable wind conditions. Machine learning technique\nsuch as deep neural networks have high representation power but is often too\nslow to update onboard. On the other hand, adaptive control relies on simple\nlinear parameter models can update as fast as the feedback control loop. We\npropose an online composite adaptation method that treats outputs from a deep\nneural network as a set of basis functions capable of representing different\nwind conditions. To help with training, meta-learning techniques are used to\noptimize the network output useful for adaptation. We validate our approach by\nflying a drone in an open air wind tunnel under varying wind conditions and\nalong challenging trajectories. We compare the result with other adaptive\ncontroller with different basis function sets and show improvement over\ntracking and prediction errors.\n"}
{"abstract": "  Large scale contextual representation models have significantly advanced NLP\nin recent years, understanding the semantics of text to a degree never seen\nbefore. However, they need to process large amounts of data to achieve\nhigh-quality results. Joining and accessing all these data from multiple\nsources can be extremely challenging due to privacy and regulatory reasons.\nFederated Learning can solve these limitations by training models in a\ndistributed fashion, taking advantage of the hardware of the devices that\ngenerate the data. We show the viability of training NLP models, specifically\nWord2Vec, with the Federated Learning protocol. In particular, we focus on a\nscenario in which a small number of organizations each hold a relatively large\ncorpus. The results show that neither the quality of the results nor the\nconvergence time in Federated Word2Vec deteriorates as compared to centralised\nWord2Vec.\n"}
{"abstract": "  The exponential rise of online social media has enabled the creation,\ndistribution, and consumption of information at an unprecedented rate. However,\nit has also led to the burgeoning of various forms of online abuse. Increasing\ncases of online antisemitism have become one of the major concerns because of\nits socio-political consequences. Unlike other major forms of online abuse like\nracism, sexism, etc., online antisemitism has not been studied much from a\nmachine learning perspective. To the best of our knowledge, we present the\nfirst work in the direction of automated multimodal detection of online\nantisemitism. The task poses multiple challenges that include extracting\nsignals across multiple modalities, contextual references, and handling\nmultiple aspects of antisemitism. Unfortunately, there does not exist any\npublicly available benchmark corpus for this critical task. Hence, we collect\nand label two datasets with 3,102 and 3,509 social media posts from Twitter and\nGab respectively. Further, we present a multimodal deep learning system that\ndetects the presence of antisemitic content and its specific antisemitism\ncategory using text and images from posts. We perform an extensive set of\nexperiments on the two datasets to evaluate the efficacy of the proposed\nsystem. Finally, we also present a qualitative analysis of our study.\n"}
{"abstract": "  The extremely large magnetoresistance (XMR) observed in many topologically\nnontrivial and trivial semimetals has attracted much attention in relation to\nits underlying physical mechanism. In this paper, by combining the band\nstructure and Fermi surface (FS) calculations with the Hall resistivity and de\nHaas-Van Alphen (dHvA) oscillation measurements, we studied the anisotropy of\nmagnetoresistance (MR) of ReO$_3$ with a simple cubic structure, an \"ordinary\"\nnonmagnetic metal considered previously. We found that ReO$_3$ exhibits almost\nall the characteristics of XMR semimetals: the nearly quadratic field\ndependence of MR, a field-induced upturn in resistivity followed by a plateau\nat low temperatures, high mobilities of charge carriers. It was found that for\nmagnetic field \\emph{H} applied along the \\emph{c} axis, the MR exhibits an\nunsaturated \\emph{H}$^{1.75}$ dependence, which was argued to arise from the\ncomplete carrier compensation supported by the Hall resistivity measurements.\nFor \\emph{H} applied along the direction of 15$^\\circ$ relative to the \\emph{c}\naxis, an unsaturated \\emph{H}$^{1.90}$ dependence of MR up to\n9.43~$\\times$~$10^3$$\\%$ at 10~K and 9~T was observed, which was explained by\nthe existence of electron open orbits extending along the $k_{x}$ direction.\nTwo mechanisms responsible for XMR observed usually in the semimetals occur\nalso in the simple metal ReO$_3$ due to its peculiar FS (two closed electron\npockets and one open electron pocket), once again indicating that the details\nof FS topology are a key factor for the observed XMR in materials.\n"}
{"abstract": "  Previous studies have shown significant differences in the enhancement of the\nstar-formation rate (SFR) and the star-formation efficiency (SFE=SFR/M_mol)\nbetween spiral-spiral and spiral-elliptical mergers. In order to shed light on\nthe physical mechanism of these differences, we present NOEMA observations of\nthe molecular gas distribution and kinematics (linear resolutions of ~ 2kpc) in\ntwo representative close major-merger star-forming pairs: the spiral-elliptical\npair Arp142 and the spiral-spiral pair Arp238. The CO in Arp142 is widely\ndistributed over a highly distorted disk without any nuclear concentration, and\nan off-centric ring-like structure is discovered in channel maps. The SFE\nvaries significantly within Arp142, with a starburst region (Region 1) near the\neastern tip of the distorted disk showing an SFE ~0.3 dex above the mean of the\ncontrol sample of isolated galaxies, and the SFE of the main disk (Region 4)\n0.43 dex lower than the mean of the control sample. In contrast, the CO\nemission in Arp238 is detected only in two compact sources at the galactic\ncenters. Compared to the control sample, Arp238-E shows an SFE enhancement of\nmore than 1 dex whereas Arp238-W has an enhancement of ~0.7 dex. We suggest\nthat the extended CO distribution and the large SFE variation in Arp142 are due\nto an expanding large-scale ring triggered by a recent high-speed head-on\ncollision between the spiral galaxy and the elliptical galaxy, and the compact\nCO sources with high SFEs in Arp238 are associated with nuclear starbursts\ninduced by gravitational tidal torques in a low-speed coplanar interaction.\n"}
{"abstract": "  We propose a complexity measure of a neural network mapping function based on\nthe diversity of the set of tangent spaces from different inputs. Treating each\ntangent space as a linear PAC concept we use an entropy-based measure of the\nbundle of concepts in order to estimate the conceptual capacity of the network.\nThe theoretical maximal capacity of a ReLU network is equivalent to the number\nof its neurons. In practice however, due to correlations between neuron\nactivities within the network, the actual capacity can be remarkably small,\neven for very big networks. Empirical evaluations show that this new measure is\ncorrelated with the complexity of the mapping function and thus the\ngeneralisation capabilities of the corresponding network. It captures the\neffective, as oppose to the theoretical, complexity of the network function. We\nalso showcase some uses of the proposed measure for analysis and comparison of\ntrained neural network models.\n"}
{"abstract": "  We have further extended our compartmental model describing the spread of the\ninfection in Italy. The model is based on the assumption that the time\nevolution of all of the observable quantities (number of people still positive\nto the infection, hospitalized and fatalities cases, healed people, and total\nnumber of people that has contracted the infection) depend on average\nparameters, namely people diffusion coefficient, infection cross-section, and\npopulation density. The model provides precious information on the tight\nrelationship between the variation of the reported infection cases and a well\ndefined observable physical quantity: the average number of people that lie\nwithin the daily displacement area of any single person. The extension of the\nmodel now includes self-consistent evaluation of the reproduction index, effect\nof immunization due to vaccination, and potential impact of virus variants on\nthe dynamical evolution of the outbreak. The model fits very well the epidemic\ndata, and allows us to strictly relate the time evolution of the number of\nhospitalized case and fatalities to the change of people mobility, vaccination\nrate, and appearance of an initial concentration of people positives for new\nvariants of the virus.\n"}
{"abstract": "  Recently discovered superconductivity in hole-doped nickelate\nNd$_{0.8}$Sr$_{0.2}$NiO$_2$ caught intensive attention in the field. An\nimmediate question is how to improve its superconducting properties. Guided by\nthe key characteristics of electronic structures of the cuprates and the\nnickelates, we propose that nickel chalcogenides with a similar lattice\nstructure should be a promising family of materials. Using NdNiS$_2$ as an\nexample, we find this particular crystal structure a stable one, through\nfirst-principle structural optimization and phonon calculation. We justify our\nproposal by comparing with CaCuO$_2$ and NdNiO$_2$ the strength of the\ncharge-transfer characteristics and the trend in their low-energy many-body\neffective Hamiltonians of doped hole carriers. These analysis indicates that\nnickel chalcogenides host low-energy physics closer to that of the cuprates,\nwith stronger magnetic interaction than the nickelates, and thus deserve\nfurther experimental exploration. Our proposal also opens up the possibility of\na wide range of parameter tuning through ligand substitution among\nchalcogenides, to further improve superconducting properties.\n"}
{"abstract": "  In this paper, we present a dataset of inter-language knowledge propagation\nin Wikipedia. Covering the entire 309 language editions and 33M articles, the\ndataset aims to track the full propagation history of Wikipedia concepts, and\nallow follow up research on building predictive models of them. For this\npurpose, we align all the Wikipedia articles in a language-agnostic manner\naccording to the concept they cover, which results in 13M propagation\ninstances. To the best of our knowledge, this dataset is the first to explore\nthe full inter-language propagation at a large scale. Together with the\ndataset, a holistic overview of the propagation and key insights about the\nunderlying structural factors are provided to aid future research. For example,\nwe find that although long cascades are unusual, the propagation tends to\ncontinue further once it reaches more than four language editions. We also find\nthat the size of language editions is associated with the speed of propagation.\nWe believe the dataset not only contributes to the prior literature on\nWikipedia growth but also enables new use cases such as edit recommendation for\naddressing knowledge gaps, detection of disinformation, and cultural\nrelationship analysis.\n"}
{"abstract": "  We investigate, within the framework of Quantum Field Theory in curved space,\nthe correlations across the horizon of a black hole in order to highlight the\npair creation mechanism at the origin of Hawking radiation. The analysis\nconcerns both acoustic black holes, formed by Bose-Einstein condensates, and\ngravitational black holes. We find a striking disagreement between the two\ncases. This is a consequence of the existence of a 'quantum atmosphere'\ndisplaced from the horizon as locus of the origin of Hawking radiation together\nwith the presence of a central singularity in the gravitational case swallowing\neverything is trapped inside the horizon.\n"}
{"abstract": "  The existence of quasinormal modes (QNMs) for waves propagating on pure de\nSitter space has been called into question in several works. We definitively\nprove the existence of quasinormal modes for massless and massive scalar fields\nin all dimensions and for all scalar field masses, and present a simple method\nfor the explicit calculation of QNMs and the corresponding mode solutions. By\npassing to coordinates which are regular at the cosmological horizon, we\ndemonstrate that certain QNMs only appear in the QNM expansion of the field\nwhen the initial data do not vanish near the cosmological horizon. The key\nobjects in the argument are dual resonant states. These are distributional mode\nsolutions of the adjoint field equation satisfying a generalized incoming\ncondition at the horizon, and they characterize the amplitudes with which QNMs\ncontribute to the QNM expansion of the field.\n"}
{"abstract": "  We analyze the properties of adversarial training for learning adversarially\nrobust halfspaces in the presence of agnostic label noise. Denoting\n$\\mathsf{OPT}_{p,r}$ as the best robust classification error achieved by a\nhalfspace that is robust to perturbations of $\\ell_{p}$ balls of radius $r$, we\nshow that adversarial training on the standard binary cross-entropy loss yields\nadversarially robust halfspaces up to (robust) classification error $\\tilde\nO(\\sqrt{\\mathsf{OPT}_{2,r}})$ for $p=2$, and $\\tilde O(d^{1/4}\n\\sqrt{\\mathsf{OPT}_{\\infty, r}} + d^{1/2} \\mathsf{OPT}_{\\infty,r})$ when\n$p=\\infty$. Our results hold for distributions satisfying anti-concentration\nproperties enjoyed by log-concave isotropic distributions among others. We\nadditionally show that if one instead uses a nonconvex sigmoidal loss,\nadversarial training yields halfspaces with an improved robust classification\nerror of $O(\\mathsf{OPT}_{2,r})$ for $p=2$, and $O(d^{1/4}\\mathsf{OPT}_{\\infty,\nr})$ when $p=\\infty$. To the best of our knowledge, this is the first work to\nshow that adversarial training provably yields robust classifiers in the\npresence of noise.\n"}
{"abstract": "  Modern systems on a chip (SoCs) utilize heterogeneous architectures where\nmultiple IP cores have concurrent access to on-chip shared resources. In\nsecurity-critical applications, IP cores have different privilege levels for\naccessing shared resources, which must be regulated by an access control\nsystem. AKER is a design and verification framework for SoC access control.\nAKER builds upon the Access Control Wrapper (ACW) -- a high performance and\neasy-to-integrate hardware module that dynamically manages access to shared\nresources. To build an SoC access control system, AKER distributes the ACWs\nthroughout the SoC, wrapping controller IP cores, and configuring the ACWs to\nperform local access control. To ensure the access control system is\nfunctioning correctly and securely, AKER provides a property-driven security\nverification using MITRE common weakness enumerations. AKER verifies the SoC\naccess control at the IP level to ensure the absence of bugs in the\nfunctionalities of the ACW module, at the firmware level to confirm the secure\noperation of the ACW when integrated with a hardware root-of-trust (HRoT), and\nat the system level to evaluate security threats due to the interactions among\nshared resources. The performance, resource usage, and security of access\ncontrol systems implemented through AKER is experimentally evaluated on a\nXilinx UltraScale+ programmable SoC, it is integrated with the OpenTitan\nhardware root-of-trust, and it is used to design an access control system for\nthe OpenPULP multicore architecture.\n"}
{"abstract": "  The binary of the pulsar PSRB1259$-$63 and the Be star LS 2883 has been\nobserved at the 2010 and 2014 periastron passages in the near-infrared (NIR)\nbands using the IRSF/SIRIUS and SIRPOL. The light curves in the J-,H-, and\nKs-bands are almost identical in these periastron passages. A flare starts no\nlater than 10 days before periastron and the maximum brightening of about 0.1\nmagnitude is observed 12--17 days after periastron. The rising part of the\nlight curve is steeper and reaches a peak slightly earlier in the Ks-band than\nin the other bands, thus a characteristic track appears on the NIR\ncolor-magnitude diagram. The time lag between the NIR light curves indicates\nthat the variation in the Be circumstellar disk first occurs in an outer\nregion. We propose that the initial rapid contraction followed by the gradual\nexpansion of the disk is evoked by the rapidly changing tidal torque around\nperiastron and the resultant change of the optically thick area causes the\nobserved NIR light curves.\n"}
{"abstract": "  We develop a word mechanism applied in knot and link diagrams for the\nillustration of a diagrammatic property. We also give a necessary condition for\ndetermining incompressible and pairwise incompressible surfaces, that are\nembedded in knot or link complements. Finally, we give a finiteness theorem and\nan upper bound on the Euler characteristic of such surfaces.\n"}
{"abstract": "  For solutions of ${\\rm div}\\,(DF(Du))=f$ we show that the quasiconformality\nof $z\\mapsto DF(z)$ is the key property leading to the Sobolev regularity of\nthe stress field $DF(Du)$, in relation with the summability of $f$. This class\nof nonlinearities encodes in a general way the notion of uniform ellipticity\nand encompasses all known instances where the stress field is known to be\nSobolev regular. We provide examples showing the optimality of this assumption\nand present three applications: the study of the strong locality of the\noperator ${\\rm div}\\,(DF(Du))$, a nonlinear Cordes condition for equations in\ndivergence form, and some partial results on the $C^{p'}$-conjecture.\n"}
{"abstract": "  Numerical experiments are the primary method of studying the evolution of\ncircumbinary disks due to the strong nonlinearities involved. Many circumbinary\nsimulations also require the use of numerical mass sinks: source terms which\nprevent gas from unphysically accumulating around the simulated point masses by\nremoving gas at a given rate. However, special care must be taken when drawing\nphysical conclusions from such simulations to ensure that results are not\nbiased by numerical artifacts. We demonstrate how the use of improved sink\nmethods reduces some of these potential biases in vertically-integrated\nsimulations of aspect ratio 0.1 accretion disks around binaries with mass\nratios between 0.1 and 1. Specifically, we show that sink terms that do not\nreduce the angular momentum of gas relative to the accreting object: 1) reduce\nthe dependence on the sink rate of physical quantities such as the torque on\nthe binary, distribution of accretion between binary components, and evolution\nof the binary semi-major axis; 2) reduce the degree to which the sink rate\naffects the structure of the accretion disks around each binary component; 3)\nalter the inferred variability of accretion onto the binary, making it more\nregular in time. We also investigate other potential sources of systematic\nerror, such as the precise from of gravitational softening and previously\nemployed simplifications to the viscous stress tensor. Because of the strong\ndependence of the orbital evolution of the binary on both the torque and\ndistribution of mass between binary components, the sink methods employed can\nhave a significant effect on the inferred orbital evolution of the binary.\n"}
{"abstract": "  The problem of long-tailed recognition, where the number of examples per\nclass is highly unbalanced, is considered. While training with class-balanced\nsampling has been shown effective for this problem, it is known to over-fit to\nfew-shot classes. It is hypothesized that this is due to the repeated sampling\nof examples and can be addressed by feature space augmentation. A new feature\naugmentation strategy, EMANATE, based on back-tracking of features across\nepochs during training, is proposed. It is shown that, unlike class-balanced\nsampling, this is an adversarial augmentation strategy. A new sampling\nprocedure, Breadcrumb, is then introduced to implement adversarial\nclass-balanced sampling without extra computation. Experiments on three popular\nlong-tailed recognition datasets show that Breadcrumb training produces\nclassifiers that outperform existing solutions to the problem.\n"}
{"abstract": "  This paper concerns the homogenization of Schrodinger equations for\nnon-crystalline matter, that is to say the coefficients are given by the\ncomposition of stationary functions with stochastic deformations. Two rigorous\nresults of so-called effective mass theorems in solid state physics are\nobtained: a general abstract result (beyond the classical stationary ergodic\nsetting), and one for quasi-perfect materials (i.e. the disorder in the\nnon-crystalline matter is limited). The former relies on the double-scale\nlimits and the wave function is spanned on the Bloch basis. Therefore, we have\nextended the Bloch Theory which was restrict until now to crystals (periodic\nsetting). The second result relies on the Perturbation Theory and a special\ncase of stochastic deformations, namely stochastic perturbation of the\nidentity.\n"}
{"abstract": "  Green's functions with continuum spectra are a way of avoiding the strong\nbounds on new physics from the absence of new narrow resonances in experimental\ndata. We model such a situation with a five-dimensional model with two branes\nalong the extra dimension $z$, the ultraviolet (UV) and the infrared (IR) one,\nsuch that the metric between the UV and the IR brane is AdS$_5$, thus solving\nthe hierarchy problem, and beyond the IR brane the metric is that of a linear\ndilaton model, which extends to $z\\to\\infty$. This simplified metric, which can\nbe considered as an approximation of a more complicated (and smooth) one, leads\nto analytical Green's functions (with a mass gap $m_g \\sim \\textrm{TeV}$ and a\ncontinuum for $s > m_g^2$) which could then be easily incorporated in the\nexperimental codes. The theory contains Standard Model gauge bosons in the bulk\nwith Neumann boundary conditions in the UV brane. To cope with electroweak\nobservables the theory is also endowed with an extra custodial gauge symmetry\nin the bulk, with gauge bosons with Dirichlet boundary conditions in the UV\nbrane, and without zero (massless) modes. All Green's functions have analytical\nexpressions and exhibit poles in the second Riemann sheet of the complex plane\nat $s=M_n^2-i M_n\\Gamma_n$, denoting a discrete (infinite) set of broad\nresonances with masses ($M_n$) and widths ($\\Gamma_n$). For gauge bosons with\nNeumann or Dirichlet boundary conditions, the masses and widths of resonances\nsatisfy the (approximate) equation $s= - 4 m_g^2 \\mathcal W_n^2[\\pm (1+i)/4]$,\nwhere $\\mathcal W_n$ is the $n$-th branch of the Lambert function.\n"}
{"abstract": "  We advance a novel explanation of similarity-based interference effects in\nsubject-verb and reflexive pronoun agreement processing, grounded in surprisal\nvalues computed from a pretrained large-scale Transformer model, GPT-2.\nSpecifically, we show that surprisal of the verb or reflexive pronoun predicts\nfacilitatory interference effects in ungrammatical sentences, where a\ndistractor noun that matches in number with the verb or pronoun leads to faster\nreading times, despite the distractor not participating in the agreement\nrelation. We review the human empirical evidence for such effects, including\nrecent meta-analyses and large-scale studies. We also show that attention\npatterns (indexed by entropy and other measures) in the Transformer show\npatterns of diffuse attention in the presence of similar distractors,\nconsistent with cue-based retrieval models of parsing. But in contrast to these\nmodels, the attentional cues and memory representations are learned entirely\nfrom the simple self-supervised task of predicting the next word.\n"}
{"abstract": "  We adapt the theory of anisotropic rubber elasticity to model cross-linked\ndouble-twist liquid crystal cylinders such as exhibited in biological systems.\nIn mechanical extension we recover strain-straightening, but with an exact\nexpression in the small twist-angle limit. In compression, we observe\ncoexistence between high and low twist phases. Coexistence begins at small\ncompressive strains and is robustly observed for any anisotropic cross-links\nand for general double-twist functions -- but disappears at large twist angles.\nWithin the coexistence region, significant compression of double-twist\ncylinders is allowed at constant stress. Our results are qualitatively\nconsistent with previous observations of swollen or compressed collagen\nfibrils, indicating that this phenomenon may be readily accessible\nexperimentally.\n"}
{"abstract": "  We utilize 3-D fully convolutional neural networks (CNN) to segment gliomas\nand its constituents from multimodal Magnetic Resonance Images (MRI). The\narchitecture uses dense connectivity patterns to reduce the number of weights\nand residual connections and is initialized with weights obtained from training\nthis model with BraTS 2018 dataset. Hard mining is done during training to\ntrain for the difficult cases of segmentation tasks by increasing the dice\nsimilarity coefficient (DSC) threshold to choose the hard cases as epoch\nincreases. On the BraTS2020 validation data (n = 125), this architecture\nachieved a tumor core, whole tumor, and active tumor dice of 0.744, 0.876,\n0.714,respectively. On the test dataset, we get an increment in DSC of tumor\ncore and active tumor by approximately 7%. In terms of DSC, our network\nperformances on the BraTS 2020 test data are 0.775, 0.815, and 0.85 for\nenhancing tumor, tumor core, and whole tumor, respectively. Overall survival of\na subject is determined using conventional machine learning from rediomics\nfeatures obtained using a generated segmentation mask. Our approach has\nachieved 0.448 and 0.452 as the accuracy on the validation and test dataset.\n"}
{"abstract": "  With the advances in the Internet of Things technology, electric vehicles\n(EVs) have become easier to schedule in daily life, which is reshaping the\nelectric load curve. It is important to design efficient charging algorithms to\nmitigate the negative impact of EV charging on the power grid. This paper\ninvestigates an EV charging scheduling problem to reduce the charging cost\nwhile shaving the peak charging load, under unknown future information about\nEVs, such as arrival time, departure time, and charging demand. First, we\nformulate an EV charging problem to minimize the electricity bill of the EV\nfleet and study the EV charging problem in an online setting without knowing\nfuture information. We develop an actor-critic learning-based smart charging\nalgorithm (SCA) to schedule the EV charging against the uncertainties in EV\ncharging behaviors. The SCA learns an optimal EV charging strategy with\ncontinuous charging actions instead of discrete approximation of charging. We\nfurther develop a more computationally efficient customized actor-critic\nlearning charging algorithm (CALC) by reducing the state dimension and thus\nimproving the computational efficiency. Finally, simulation results show that\nour proposed SCA can reduce EVs' expected cost by 24.03%, 21.49%, 13.80%,\ncompared with the Eagerly Charging Algorithm, Online Charging Algorithm,\nRL-based Adaptive Energy Management Algorithm, respectively. CALC is more\ncomputationally efficient, and its performance is close to that of SCA with\nonly a gap of 5.56% in the cost.\n"}
{"abstract": "  We give a brief overview how to couple general relativity to the Standard\nModel of elementary particles, within the higher gauge theory framework,\nsuitable for the spinfoam quantization procedure. We begin by providing a short\nreview of all relevant mathematical concepts, most notably the idea of a\ncategorical ladder, 3-groups and generalized parallel transport. Then, we give\nan explicit construction of the algebraic structure which describes the full\nStandard Model coupled to Einstein-Cartan gravity, along with the classical\naction, written in the form suitable for the spinfoam quantization procedure.\nWe emphasize the usefulness of the 3-group concept as a superior tool to\ndescribe gauge symmetry, compared to an ordinary Lie group, as well as the\npossibility to employ this new structure to classify matter fields and study\ntheir spectrum, including the origin of fermion families.\n"}
{"abstract": "  We consider a discrete Schr\\\"odinger operator $ H_\\varepsilon=\n-\\varepsilon^2\\Delta_\\varepsilon + V_\\varepsilon$ on $\\ell^2(\\varepsilon\n\\mathbb Z^d)$, where $\\varepsilon>0$ is a small parameter and the potential\n$V_\\varepsilon$ is defined in terms of a multiwell energy landscape $f$ on\n$\\mathbb R^d$. This operator can be seen as a discrete analog of the\nsemiclassical Witten Laplacian of $\\mathbb R^d$. It is unitarily equivalent to\nthe generator of a diffusion on $\\varepsilon \\mathbb Z^d$, satisfying the\ndetailed balance condition with respect to the Boltzmann weight\n$\\exp{(-f/\\varepsilon)}$. These type of diffusions exhibit metastable behaviour\nand arise in the context of disordered mean field models in Statistical\nMechanics. We analyze the bottom of the spectrum of $H_\\varepsilon$ in the\nsemiclassical regime $\\varepsilon\\ll1$ and show that there is a one-to-one\ncorrespondence between exponentially small eigenvalues and local minima of $f$.\nThen we analyze in more detail the bistable case and compute the precise\nasymptotic splitting between the two exponentially small eigenvalues. Through\nthis purely spectral-theoretical analysis of the discrete Witten Laplacian we\nrecover in a self-contained way the Eyring-Kramers formula for the metastable\ntunneling time of the underlying stochastic process.\n"}
{"abstract": "  Understanding the structure of loss landscape of deep neural networks\n(DNNs)is obviously important. In this work, we prove an embedding principle\nthat the loss landscape of a DNN \"contains\" all the critical points of all the\nnarrower DNNs. More precisely, we propose a critical embedding such that any\ncritical point, e.g., local or global minima, of a narrower DNN can be embedded\nto a critical point/hyperplane of the target DNN with higher degeneracy and\npreserving the DNN output function. The embedding structure of critical points\nis independent of loss function and training data, showing a stark difference\nfrom other nonconvex problems such as protein-folding. Empirically, we find\nthat a wide DNN is often attracted by highly-degenerate critical points that\nare embedded from narrow DNNs. The embedding principle provides an explanation\nfor the general easy optimization of wide DNNs and unravels a potential\nimplicit low-complexity regularization during the training. Overall, our work\nprovides a skeleton for the study of loss landscape of DNNs and its\nimplication, by which a more exact and comprehensive understanding can be\nanticipated in the near\n"}
{"abstract": "  We simulate a feasible experimental scenario that reconstruct the density\nmatrix of an electronic qubit via spin-polarized tunneling currents. Our system\nis composed of a single level quantum dot, attached to ferromagnetic leads. By\nproperly rotating the leads magnetizations we can select the spin component to\nbe detected. Counting the number of tunneling events taken in four different\nmagnetic alignments we find the Stokes parameters, that are used as a train\ndata set into a supervised machine learning algorithm. The predicted model\nallows the complete knowledge of the open system density matrix, including both\nprobability amplitudes and relative phases.\n"}
{"abstract": "  We present a formal geometric framework for the study of adiabatic quantum\nmechanics for arbitrary finite-dimensional non-degenerate Hamiltonians. This\nframework generalizes earlier holonomy interpretations of the geometric phase\nto non-cyclic states appearing for non-Hermitian Hamiltonians. We start with an\ninvestigation of the space of non-degenerate operators on a finite-dimensional\nstate space. We then show how the energy bands of a Hamiltonian family form a\ncovering space. Likewise, we show that the eigenrays form a bundle, a\ngeneralization of a principal bundle, which admits a natural connection\nyielding the (generalized) geometric phase. This bundle provides in addition a\nnatural generalization of the quantum geometric tensor and derived tensors, and\nwe show how it can incorporate the non-geometric dynamical phase as well. We\nfinish by demonstrating how the bundle can be recast as a principal bundle, so\nthat both the geometric phases and the permutations of eigenstates can be\nexpressed simultaneously by means of standard holonomy theory.\n"}
{"abstract": "  The present article is a part of the study of solvable Leibniz algebras with\na given nilradical. In this paper solvable Leibniz algebras, whose nilradicals\nis naturally graded quasi-filiform algebra and the complemented space to the\nnilradical has maximal dimension, are described up to isomorphism.\n"}
{"abstract": "  A unified system integrating a compact object detector and a surrounding\nenvironmental condition classifier for enhancing the robustness of object\ndetection scheme in advanced driver assistance systems (ADAS) is proposed in\nthis paper. ADAS are invented to improve traffic safety and effectiveness in\nautonomous driving systems where object detection plays an extremely important\nrole. However, modern object detectors integrated in ADAS are still unstable\ndue to high latency and the variation of the environmental contexts in the\ndeployment phase. Our system is proposed to address the aforementioned\nproblems. The proposed system includes two main components: (1) a compact\none-stage object detector which is expected to be able to perform at a\ncomparable accuracy compared to state-of-the-art object detectors, and (2) an\nenvironmental condition detector that helps to send a warning signal to the\ncloud in case the self-driving car needs human actions due to the significance\nof the situation. The empirical results prove the reliability and the\nscalability of the proposed system to realistic scenarios.\n"}
{"abstract": "  Autonomous vehicles face tremendous challenges while interacting with human\ndrivers in different kinds of scenarios. Developing control methods with safety\nguarantees while performing interactions with uncertainty is an ongoing\nresearch goal. In this paper, we present a real-time safe control framework\nusing bi-level optimization with Control Barrier Function (CBF) that enables an\nautonomous ego vehicle to interact with human-driven cars in ramp merging\nscenarios with a consistent safety guarantee. In order to explicitly address\nmotion uncertainty, we propose a novel extension of control barrier functions\nto a probabilistic setting with provable chance-constrained safety and analyze\nthe feasibility of our control design. The formulated bi-level optimization\nframework entails first choosing the ego vehicle's optimal driving style in\nterms of safety and primary objective, and then minimally modifying a nominal\ncontroller in the context of quadratic programming subject to the probabilistic\nsafety constraints. This allows for adaptation to different driving strategies\nwith a formally provable feasibility guarantee for the ego vehicle's safe\ncontroller. Experimental results are provided to demonstrate the effectiveness\nof our proposed approach.\n"}
{"abstract": "  In this paper, we generalize Witten's non-abelian bosonization in $(1+1)$-D\nto two and three spatial dimensions. Our theory applies to fermions with\nrelativistic dispersion. The bosonized theories are non-linear sigma models\nwith level-1 Wess-Zumino-Witten terms. We apply the bosonization results to the\n$SU(2)$ gauge theory of the $\\pi$-flux phase, critical spin liquids in 1,2,3\nspatial dimensions, and twisted bilayer graphene.\n"}
{"abstract": "  Vegetation, trees in particular, sequester carbon by absorbing carbon dioxide\nfrom the atmosphere. However, the lack of efficient quantification methods of\ncarbon stored in trees renders it difficult to track the process. We present an\napproach to estimate the carbon storage in trees based on fusing multi-spectral\naerial imagery and LiDAR data to identify tree coverage, geometric shape, and\ntree species -- key attributes to carbon storage quantification. We demonstrate\nthat tree species information and their three-dimensional geometric shapes can\nbe estimated from aerial imagery in order to determine the tree's biomass.\nSpecifically, we estimate a total of $52,000$ tons of carbon sequestered in\ntrees for New York City's borough Manhattan.\n"}
{"abstract": "  As collaborative learning allows joint training of a model using multiple\nsources of data, the security problem has been a central concern. Malicious\nusers can upload poisoned data to prevent the model's convergence or inject\nhidden backdoors. The so-called backdoor attacks are especially difficult to\ndetect since the model behaves normally on standard test data but gives wrong\noutputs when triggered by certain backdoor keys. Although Byzantine-tolerant\ntraining algorithms provide convergence guarantee, provable defense against\nbackdoor attacks remains largely unsolved. Methods based on randomized\nsmoothing can only correct a small number of corrupted pixels or labels;\nmethods based on subset aggregation cause a severe drop in classification\naccuracy due to low data utilization. We propose a novel framework that\ngeneralizes existing subset aggregation methods. The framework shows that the\nsubset selection process, a deciding factor for subset aggregation methods, can\nbe viewed as a code design problem. We derive the theoretical bound of data\nutilization ratio and provide optimal code construction. Experiments on non-IID\nversions of MNIST and CIFAR-10 show that our method with optimal codes\nsignificantly outperforms baselines using non-overlapping partition and random\nselection. Additionally, integration with existing coding theory results shows\nthat special codes can track the location of the attackers. Such capability\nprovides new countermeasures to backdoor attacks.\n"}
{"abstract": "  In this work, we extend the analyses devoted to Newtonian viscous fluids\npreviously reported by Ribe [Physical Review E 68, 036305 (2003)], by\ninvestigating shear thickening (dilatant) and shear thinning (pseudoplastic)\neffects on the development of folding instabilities in non-Newtonian viscous\nsheets of which viscosity is given by a power-law constitutive equation. Such\ninstabilities are trigged by compression stresses acting on viscous sheets that\nleave a channel at a very small initial velocity, fall, and then hit a solid\nsurface or a fluid substrate. Our study is conducted through a mixed approach\ncombining direct numerical simulations, energy budget analyses, scaling laws,\nand experiments. The numerical results are based on an adaptive variational\nmulti-scale method for multiphase flows, while Carpobol gel sheets are\nconsidered for the conducted experiments. Two folding regimes are observed: (1)\nthe viscous regime; and (2) the gravitational one. Interestingly, only the\nlatter is affected by shear thinning/thickening manifestations within the\nmaterial. In short, when gravity is balanced by viscous forces along the\nnon-Newtonian viscous sheet, both the folding amplitude and the folding\nfrequency are given by a power-law function of the sheet slenderness, the\nGalileo number (the ratio of the gravitational stress to the viscous one), and\nthe flow behaviour index. Highly shear thickening materials develop large\namplitude (and low frequency) instabilities, which, in contrast, tend to be\nsuppressed by shear thinning effects, and eventually cease. Lastly,\nnonNewtonian effects on folding onset/cessation are also carefully explored. As\na result, non-Newtonian folding onset and cessation criteria are presented.\n"}
{"abstract": "  Expressive reading, considered the defining attribute of oral reading\nfluency, comprises the prosodic realization of phrasing and prominence. In the\ncontext of evaluating oral reading, it helps to establish the speaker's\ncomprehension of the text. We consider a labeled dataset of children's reading\nrecordings for the speaker-independent detection of prominent words using\nacoustic-prosodic and lexico-syntactic features. A previous well-tuned random\nforest ensemble predictor is replaced by an RNN sequence classifier to exploit\npotential context dependency across the longer utterance. Further, deep\nlearning is applied to obtain word-level features from low-level acoustic\ncontours of fundamental frequency, intensity and spectral shape in an\nend-to-end fashion. Performance comparisons are presented across the different\nfeature types and across different feature learning architectures for prominent\nword prediction to draw insights wherever possible.\n"}
{"abstract": "  Although deep learning models have driven state-of-the-art performance on a\nwide array of tasks, they are prone to spurious correlations that should not be\nlearned as predictive clues. To mitigate this problem, we propose a\ncausality-based training framework to reduce the spurious correlations caused\nby observed confounders. We give theoretical analysis on the underlying general\nStructural Causal Model (SCM) and propose to perform Maximum Likelihood\nEstimation (MLE) on the interventional distribution instead of the\nobservational distribution, namely Counterfactual Maximum Likelihood Estimation\n(CMLE). As the interventional distribution, in general, is hidden from the\nobservational data, we then derive two different upper bounds of the expected\nnegative log-likelihood and propose two general algorithms, Implicit CMLE and\nExplicit CMLE, for causal predictions of deep learning models using\nobservational data. We conduct experiments on both simulated data and two\nreal-world tasks: Natural Language Inference (NLI) and Image Captioning. The\nresults show that CMLE methods outperform the regular MLE method in terms of\nout-of-domain generalization performance and reducing spurious correlations,\nwhile maintaining comparable performance on the regular evaluations.\n"}
{"abstract": "  Low excitation radio galaxies (LERGs) are weakly accreting active galactic\nnuclei (AGN) believed to be fuelled by radiatively inefficient accretion\nprocesses. Despite this, recent works have shown evidence for ionized and\nneutral hydrogen gas outflows in these galaxies. To investigate the potential\ndrivers of such outflows we select a sample of 802 LERGs using the Best &\nHeckman (2012) catalogue of radio galaxies. By modelling the [O III] $\\lambda\n5007$ profile in Sloan Digital Sky Survey spectra of a sample of 802 LERGs, we\ndetermine that the ionized outflows are present in $\\sim 1.5\\%$ of the\npopulation. Using $1.4~\\text{GHz}$ imaging from the Faint Images of the Radio\nSky at Twenty Centimeters survey we analyze the radio morphology of LERGs with\noutflows and find these to be consistent with the parent LERG population.\nHowever, we note that unlike the majority of the LERG population, those LERGs\nshowing outflows have Eddington scaled accretion rates close to $1\\%$. This is\nindicative that ionized outflows in LERGs are driven by the radiation pressure\nfrom the accretion disk of the AGN rather than the radio jets. We report\nspecific star formation rates in the range of $10^{-12} < \\text{sSFR} <\n10^{-9}~\\text{yr}^{-1}$. Moreover, we observe higher mass outflow rates of\n$7-150~M_{\\odot}~\\text{yr}^{-1}$ for these LERGs than luminous quasars for a\ngiven bolometric luminosity, which could possibly be due to the radio source in\nLERGs boosting the mass-loading. This scenario could indicate that these\noutflows could potentially drive feedback in LERGs.\n"}
{"abstract": "  We systematically investigate the ground-state properties of self-bound\ndroplets of quasi-two-dimensional binary Bose gases by using the Gaussian state\ntheory. We find that quantum droplets consists two macroscopic squeezed phases\nand a macroscopic coherent phase. We map out the phase diagram and determine\nall phase boundaries via both numerical and nearly analytical methods. In\nparticular, we find three easily accessible signatures for the quantum phases\nand the stablization mechanism of the self-bound droplets by precisely\nmeasuring their radial size. Our studies indicate that binary droplets\nrepresent an ideal platform for in-depth investigations of the quantum nature\nof the droplet state.\n"}
{"abstract": "  We utilize a recently proposed cubic nilpotent superfield to realize\ninflation in supergravity with the minimal degrees of freedom: the inflaton,\ngraviton, and massive gravitino. As an advantage, the resultant model is free\nfrom the catastrophic production of gravitinos due to its vanishing propagation\nspeed. However, the model suffers from the standard gravitino problem, and its\nviability depends on the mass spectrum and the thermal history of the universe.\n"}
{"abstract": "  Feedback-driven winds from star formation or active galactic nuclei might be\na relevant channel for the abrupt quenching star formation in massive galaxies.\nHowever, both observations and simulations support the idea that these\nprocesses are non-conflictingly co-evolving and self-regulating. Furthermore,\nevidence of disruptive events that are capable of fast quenching is rare, and\nconstraints on their statistical prevalence are lacking. Here we present a\nmassive starburst galaxy at z=1.4 which is ejecting $46 \\pm 13$\\% of its\nmolecular gas mass at a startling rate of $\\gtrsim 10,000$ M$_{\\odot}{\\rm\nyr}^{-1}$. A broad component that is red-shifted from the galaxy emission is\ndetected in four (low- and high-J) CO and [CI] transitions and in the ionized\nphase, which ensures a robust estimate of the expelled gas mass. The implied\nstatistics suggest that similar events are potentially a major star-formation\nquenching channel. However, our observations provide compelling evidence that\nthis is not a feedback-driven wind, but rather material from a merger that has\nbeen probably tidally ejected. This finding challenges some literature studies\nin which the role of feedback-driven winds might be overstated.\n"}
{"abstract": "  Objective: This study aims to develop an end-to-end natural language\nprocessing pipeline for triage and diagnosis of COVID-19 from patient-authored\nsocial media posts, in order to provide researchers and other interested\nparties with additional information on the symptoms, severity and prevalence of\nthe disease. Materials and Methods: The text processing pipeline first extracts\nCOVID-19 symptoms and related concepts such as severity, duration, negations,\nand body parts from patients posts using conditional random fields. An\nunsupervised rule-based algorithm is then applied to establish relations\nbetween concepts in the next step of the pipeline. The extracted concepts and\nrelations are subsequently used to construct two different vector\nrepresentations of each post. These vectors are applied separately to build\nsupport vector machine learning models to triage patients into three categories\nand diagnose them for COVID-19. Results: We report that macro- and\nmicro-averaged F1 scores in the range of 71-96% and 61-87%, respectively, for\nthe triage and diagnosis of COVID-19, when the models are trained on human\nlabelled data. Our experimental results indicate that similar performance can\nbe achieved when the models are trained using predicted labels from concept\nextraction and rule-based classifiers, thus yielding end-to-end machine\nlearning. Also, we highlight important features uncovered by our diagnostic\nmachine learning models and compare them with the most frequent symptoms\nrevealed in another COVID-19 dataset. In particular, we found that the most\nimportant features are not always the most frequent ones.\n"}
{"abstract": "  For quasiperiodic Schr\\\"odinger operators with one-frequency analytic\npotentials, from dynamical systems side, it has been proved that the\ncorresponding quasiperiodic Schr\\\"odinger cocycle is either rotations reducible\nor has positive Lyapunov exponent for all irrational frequency and almost every\nenergy. From spectral theory side, the \"Schr\\\"odinger conjecture\" and the\n\"Last's intersection spectrum conjecture\" have been verified. The proofs of\nabove results crucially depend on the analyticity of the potentials. People are\ncurious about if the analyticity is essential for those problems, see open\nproblems by Fayad-Krikorian and Jitomirskaya-Mar. In this paper, we prove the\nabove mentioned results for ultra-differentiable potentials.\n"}
{"abstract": "  Recent advances on Vision Transformer (ViT) and its improved variants have\nshown that self-attention-based networks surpass traditional Convolutional\nNeural Networks (CNNs) in most vision tasks. However, existing ViTs focus on\nthe standard accuracy and computation cost, lacking the investigation of the\nintrinsic influence on model robustness and generalization. In this work, we\nconduct systematic evaluation on components of ViTs in terms of their impact on\nrobustness to adversarial examples, common corruptions and distribution shifts.\nWe find some components can be harmful to robustness. By using and combining\nrobust components as building blocks of ViTs, we propose Robust Vision\nTransformer (RVT), which is a new vision transformer and has superior\nperformance with strong robustness. We further propose two new plug-and-play\ntechniques called position-aware attention scaling and patch-wise augmentation\nto augment our RVT, which we abbreviate as RVT*. The experimental results on\nImageNet and six robustness benchmarks show the advanced robustness and\ngeneralization ability of RVT compared with previous ViTs and state-of-the-art\nCNNs. Furthermore, RVT-S* also achieves Top-1 rank on multiple robustness\nleaderboards including ImageNet-C and ImageNet-Sketch. The code will be\navailable at \\url{https://git.io/Jswdk}.\n"}
{"abstract": "  We prove in this paper the Ax-Schanuel conjecture for all admissible\nvariations of mixed Hodge structures.\n"}
{"abstract": "  In this paper, we propose a quadruple-UPA architecture for realizing the\nbeamforming with wide coverage in terahertz (THz) communication. Considering\nthe severe path loss suffered in the THz wave propagation, the pilot signals of\nconventional channel estimation strategies may not be effectively detected. In\nsight of this, we propose a fast 3D beam training strategy to jointly realize\nthe angle estimation and the beamforming. Specifically, we first develop a\nnovel hierarchical codebook that predefines some codewords (including narrow\nbeams and wide beams) stage by stage. The narrow beams are designed as the\nbeamforming solution candidates, whereas the wide beams are designed for\nreducing the training complexity. Then, we propose a 3D beam training procedure\nto find the optimal beamforming solution, i.e., the optimal narrow-beam pair,\nin a fast manner. Numerical results are presented to validate the effectiveness\nof the proposed scheme.\n"}
{"abstract": "  Conditional Random Field (CRF) based neural models are among the most\nperformant methods for solving sequence labeling problems. Despite its great\nsuccess, CRF has the shortcoming of occasionally generating illegal sequences\nof tags, e.g. sequences containing an \"I-\" tag immediately after an \"O\" tag,\nwhich is forbidden by the underlying BIO tagging scheme. In this work, we\npropose Masked Conditional Random Field (MCRF), an easy to implement variant of\nCRF that impose restrictions on candidate paths during both training and\ndecoding phases. We show that the proposed method thoroughly resolves this\nissue and brings consistent improvement over existing CRF-based models with\nnear zero additional cost.\n"}
{"abstract": "  The presence of snow and ice on runway surfaces reduces the available\ntire-pavement friction needed for retardation and directional control and\ncauses potential economic and safety threats for the aviation industry during\nthe winter seasons. To activate appropriate safety procedures, pilots need\naccurate and timely information on the actual runway surface conditions. In\nthis study, XGBoost is used to create a combined runway assessment system,\nwhich includes a classifcation model to predict slippery conditions and a\nregression model to predict the level of slipperiness. The models are trained\non weather data and data from runway reports. The runway surface conditions are\nrepresented by the tire-pavement friction coefficient, which is estimated from\nflight sensor data from landing aircrafts. To evaluate the performance of the\nmodels, they are compared to several state-of-the-art runway assessment\nmethods. The XGBoost models identify slippery runway conditions with a ROC AUC\nof 0.95, predict the friction coefficient with a MAE of 0.0254, and outperforms\nall the previous methods. The results show the strong abilities of machine\nlearning methods to model complex, physical phenomena with a good accuracy when\ndomain knowledge is used in the variable extraction. The XGBoost models are\ncombined with SHAP (SHapley Additive exPlanations) approximations to provide a\ncomprehensible decision support system for airport operators and pilots, which\ncan contribute to safer and more economic operations of airport runways.\n"}
{"abstract": "  In this work, we investigate the problem of privacy-preserving supervisory\ncontrol against an external passive intruder via co-synthesis of dynamic mask,\nedit function, and supervisor for opacity enforcement and requirement\nsatisfaction. We attempt to achieve the following goals: 1) the system secret\ncannot be inferred by the intruder, i.e., opacity of secrets against the\nintruder, and the existence of the dynamic mask and the edit function should\nnot be discovered by the intruder, i.e., covertness of dynamic mask and edit\nfunction against the intruder; 2) the closed-loop system behaviors should\nsatisfy some safety and nonblockingness requirement. We assume the intruder can\neavesdrop both the sensing information generated by the sensors and the control\ncommands issued to the actuators, and we refer to such an intruder as a\nsensor-actuator eavesdropping intruder. Our approach is to model the\nco-synthesis problem as a distributed supervisor synthesis problem in the\nRamadge-Wonham supervisory control framework, and we propose an incremental\nsynthesis heuristic to incrementally synthesize a dynamic mask, an edit\nfunction, and a supervisor, which consists of three steps: 1) we first\nsynthesize an ensemble ME of dynamic mask and edit function to ensure the\nopacity and the covertness against a sensor eavesdropping but command\nnon-eavesdropping intruder, and marker-reachability; 2) we then decompose ME\ninto a dynamic mask and an edit function by using a constraint-based approach,\nwith the help of a Boolean satisfiability (SAT) solver; 3) finally, we\nsynthesize a supervisor such that opacity and covertness can be ensured against\nthe sensor-actuator eavesdropping intruder, and at the same time safety and\nnonblockingness requirement can be ensured. The effectiveness of our approach\nis illustrated on an example about the enforcement of location privacy for an\nautonomous vehicle.\n"}
{"abstract": "  We study single-field inflationary models with steep step-like features in\nthe potential that lead to the temporary violation of the slow-roll conditions\nduring the evolution of the inflaton. These features enhance the power spectrum\nof the curvature perturbations by several orders of magnitude at certain scales\nand also produce prominent oscillatory patterns. We study analytically and\nnumerically the inflationary dynamics. We describe quantitatively the size of\nthe enhancement, as well as the profile of the oscillations, which are shaped\nby the number and position of the features in the potential. The induced tensor\npower spectrum inherits the distinctive oscillatory profile of the curvature\nspectrum and is potentially detectable by near-future space interferometers.\nThe enhancement of the power specrtum by step-like features, though\nsignificant, may be insufficient to trigger the production of a sizeable number\nof primordial black holes if radiation dominates the energy density of the\nearly universe. However, it can result in sufficient black hole production if\nthe universe is dominated by non-relativistic matter. For the latter scenario,\nwe find that deviations from the standard monochromatic profile of the mass\nspectrum of primordial black holes are possible because of the multiple-peak\nstructure of the curvature power spectrum.\n"}
{"abstract": "  In this paper we prove two matrix weighted norm inequalities for the\ncommutator of a fractional integral operator and multiplication by a matrix\nsymbol. More precisely, we extend the recent results of the second author,\nPott, and Treil on two matrix weighted norm inequalities for commutators of\nCalderon-Zygmund operators and multiplication by a matrix symbol to the\nfractional integral operator setting. In particular, we completely extend the\nfractional Bloom theory of Holmes, Rahm, and Spencer to the two matrix weighted\nsetting with a matrix\n"}
{"abstract": "  We study d-minimal expansions of ordered fields, and dense pairs thereof. We\nalso consider other generalizations of o-minimality.\n"}
{"abstract": "  In this paper we introduce Stein's square function associated with bilinear\nBochner-Riesz means and develop a systematic study of its $L^p$ boundedness\nproperties. We also discuss applications of bilinear Bochner-Riesz square\nfunction in the context of bilinear fractional Schr\\\"{o}dinger multipliers,\ngeneralized bilinear spherical maximal function and more general bilinear\nmultipliers defined on $\\mathbb{R}^{2n}$ of the form $(\\xi,\\eta)\\rightarrow\nm\\left(|(\\xi,\\eta)|^2\\right)$.\n"}
{"abstract": "  For a given graph $G$, the metric and edge metric dimensions of $G$,\n$\\dim(G)$ and ${\\rm edim}(G)$, are the cardinalities of the smallest possible\nsubsets of vertices in $V(G)$ such that they uniquely identify the vertices and\nthe edges of $G$, respectively, by means of distances. It is already known that\nmetric and edge metric dimensions are not in general comparable. Infinite\nfamilies of graphs with pendant vertices in which the edge metric dimension is\nsmaller than the metric dimension are already known. In this article, we\nconstruct a 2-connected graph $G$ such that $\\dim(G)=a$ and ${\\rm edim}(G)=b$\nfor every pair of integers $a,b$, where $4\\le b<a$. For this we use\nsubdivisions of complete graphs, whose metric dimension is in some cases\nsmaller than the edge metric dimension. Along the way, we present an upper\nbound for the metric and edge metric dimensions of subdivision graphs under\nsome special conditions.\n"}
{"abstract": "  We establish long-time existence for a projected Sobolev gradient flow of\ngeneralized integral Menger curvature in the Hilbert case, and provide\n$C^{1,1}$-bounds in time for the solution that only depend on the initial\ncurve. The self-avoidance property of integral Menger curvature guarantees that\nthe knot class of the initial curve is preserved under the flow, and the\nprojection ensures that each curve along the flow is parametrized with the same\nspeed as the initial configuration. Finally, we describe how to simulate this\nflow numerically with substantially higher efficiency than in the corresponding\nnumerical $L^2$ gradient descent or other optimization methods.\n"}
{"abstract": "  A novel structure of sheet-pile groin consists of row piles and tied sheet\nhas been introduced to resist the impact of tidal bore in recent years.\nHowever, the dynamic responses of the structure under tidal bore have not been\ncarefully studied before. This paper proposed an analytical solution for the\ndynamic responses of the sheet-pile groin subjected to a transient lateral\nexcitation. In order to take the flexural and longitudinal vibrations into\nconsideration, the piles and the sheet are modeled by the Timoshenko beam (TB)\ntheory and a one-dimensional rod, respectively. The soil around the piles is\nsimulated by the dynamic Winkler foundation model. The excitation is considered\nby a lateral concentrated point load acting on the front pile. On this basis,\nthe governing equations are constructed in time domain and solved in frequency\ndomain, while the time-domain responses are finally obtained using the discrete\nInverse Fourier Transform. The accuracy of the presented analytical solution is\nvalidated by comparing with the results obtained from the FEM simulation and a\nmodel test. The effects of properties of the sheet, the piles and the soil on\nthe dynamic responses are investigated through a comprehensive parametric\nstudy.\n"}
{"abstract": "  It has often been conjectured that the effectiveness of line drawings can be\nexplained by the similarity of edge images to line drawings. This paper\npresents several problems with explaining line drawing perception in terms of\nedges, and how the recently-proposed Realism Hypothesis of Hertzmann (2020)\nresolves these problems. There is nonetheless existing evidence that edges are\noften the best features for predicting where people draw lines; this paper\ndescribes how the Realism Hypothesis can explain this evidence.\n"}
{"abstract": "  The generation of high-intensity optical fields from harmonic-wave photons,\ninteracting via a cross-phase modulation with dark solitons both propagating in\na Kerr nonlinear medium, is examined. The focus is on a pump consisting of\ntime-entangled dark-soliton patterns, forming a periodic waveguide along the\npath of the harmonic-wave probe. It is shown that an increase of the strength\nof cross-phase modulation respective to the self-phase modulation, favors\nsoliton-mode proliferation in the bound-state spectrum of the trapped\nharmonic-wave probe. The induced soliton modes, which display the structures of\nperiodic soliton lattices, are not just rich in numbers, they also form a great\ndiversity of population of soliton crystals with a high degree of degeneracy.\n"}
{"abstract": "  Here, we report both ac and dc magnetization, thermodynamic and electric\nproperties of hexagonal Ba$_3$NiIr$_2$O$_9$. The Ni$^{2+}$ (spin-1) forms\nlayered triangular-lattice and interacts antiferromagnetically while Ir$^{5+}$\nis believed to act as magnetic link between the layers. This complex magnetic\ninteraction results in magnetic frustration leading to a spin-glass transition\nat $T_f$ $\\sim$ 8.5 K. The observed magnetic relaxation and aging effect also\nconfirms the nonequilibrium ground state. The system further shows large\nexchange bias which is tunable with cooling field. Below the Curie-Weiss\ntemperature $\\theta_{CW}$ ($\\sim$ -29 K), the magnetic specific heat $C_m$\ndisplays a broad hump and at low temperature follows $C_m = \\gamma T^\\alpha$\ndependence where both $\\gamma$ and $\\alpha$ show dependence on temperature and\nmagnetic field. A sign change in magnetoresistace is observed which is due to\nan interplay among magnetic moment, field and spin-orbit coupling.\n"}
{"abstract": "  We study Alexandrov curvature in the tropical projective torus with respect\nto the tropical metric. Alexandrov curvature is a generalization of classical\nRiemannian sectional curvature to more general metric spaces; it is determined\nby a comparison of triangles in an arbitrary metric space to corresponding\ntriangles in Euclidean space. In the polyhedral setting of tropical geometry,\ntriangles are a combinatorial object, which adds a combinatorial dimension to\nour analysis. We study the effect that the triangle types have on curvature,\nand what can be revealed about these types from the curvature. We find that\npositive, negative, zero, and undefined Alexandrov curvature can exist\nconcurrently in tropical settings and that there is a tight connection between\ntriangle combinatorial type and curvature. Our results are established both by\nproof and numerical experiments, and shed light on the intricate geometry of\nthe tropical projective torus.\n  This paper is dedicated to Bernd Sturmfels on the occasion of his 60th\nbirthday.\n"}
{"abstract": "  I propose a new type of common envelope jets supernova (CEJSN) events where\ninstead of a single neutron star (NS; or a black hole; BH) a tight binary\nsystem of a NS and a main sequence star enters a common envelope evolution\n(CEE) with a red supergiant. The NS and the main sequence star of the tight\nbinary system merge inside the red supergiant envelope and enter a CEE of their\nown. The NS accretes some mass through an accretion disk and launches jets that\nexplodes the main sequence star. I estimate that the two jets that the NS\nlaunches at this phase carry an energy of ~10^{52} erg, about the same order of\nmagnitude as the energy that the jets will carry when the NS or its BH remnant\nwill enter the core in a later phase. For that, I term the entire event a\ndouble CEJSN. The outcome of the double CEJSN is a very long, months to years,\nand very energetic event, a total energy of ~10^{52} - 10^{53} erg, that will\nbe observationally classified as a peculiar super-energetic event. I crudely\nestimate that new transient surveys should detect about one CEJSN event from a\ntriple star system per year.\n"}
{"abstract": "  We consider a generalised oriented site percolation model (GOSP) on $\\mathbb\nZ^d$ with arbitrary neighbourhood. The key additional difficulties as compared\nto standard oriented percolation (OP) are the lack of symmetry and, in two\ndimensions, of planarity. We establish that, despite these deficiencies, in the\nsupercritical regime GOSP behaves qualitatively like OP.\n"}
{"abstract": "  Gaussian processes are widely employed as versatile modelling and predictive\ntools in spatial statistics, functional data analysis, computer modelling and\ndiverse applications of machine learning. They have been widely studied over\nEuclidean spaces, where they are specified using covariance functions or\ncovariograms for modelling complex dependencies. There is a growing literature\non Gaussian processes over Riemannian manifolds in order to develop richer and\nmore flexible inferential frameworks. While Gaussian processes have been\nextensively studied for asymptotic inference on Euclidean spaces using\nwell-defined (positive definite) covariograms, such results are relatively\nsparse on Riemannian manifolds. We undertake analogous developments for\nGaussian processes constructed over compact Riemannian manifolds. Building upon\nthe recently introduced Mat\\'ern covariograms on a compact Riemannian manifold,\nwe employ formal notions and conditions for the equivalence of two Mat\\'ern\nGaussian random measures on compact manifolds to derive the microergodic\nparameters and formally establish the consistency of their maximum likelihood\nestimates as well as asymptotic optimality of the best linear unbiased\npredictor. The circle and sphere are studied as two specific examples of\ncompact Riemannian manifolds with numerical experiments to illustrate and\ncorroborate the theory.\n"}
{"abstract": "  Two cavities in different size with the photonic crystal structure have been\ndeveloped for axion searches. In the cavities, the dispersion relation in the\nphotonic crystal is utilised, and so was named to \"DRiPC cavities\". The size of\nthe smaller one is 100 mm x 100 mm x 10 mm, where 16 cylindrical metal poles\nwith a diameter of 4 mm are introduced in a 4x4 grid at 20 mm intervals. In\nthis study, the grid interval in x direction in the small size cavity, Lx, was\nchanged to investigate resonance frequency, Q-value, and electric field profile\nat each Lx. The lowest three frequencies have been compared with the ones\nsimulated by the finite element method to be found in excellent agreement. The\nlowest frequency mode could be tuned from 5.10 GHz (Lx = 25.0 mm) to 6.72 GHz\n(13.9 mm), centering on 5.87 GHz at Lx = 20 mm. This wide range tunability,\n27.7%, was suitable for a search with a modest Q-value. By examining the\nelectric field distributions with the bead pull method, the lowest frequency\nmode at Lx = 16.0 - 25.0 mm were TM010-like. This mode was also obtained in a\nlarger size cavity (180 mm x 180 mm x 20 mm x 2) with the same photonic crystal\nstructure. These results led us to conclude a DRiPC cavity has the noble\nfeatures for future axion search experiments.\n"}
{"abstract": "  If dark matter has mass lower than around 1 GeV, it will not impart enough\nenergy to cause detectable nuclear recoils in many direct-detection\nexperiments. However, if dark matter is upscattered to high energy by\ncollisions with cosmic rays, it may be detectable in both direct-detection\nexperiments and neutrino experiments. We report the results of a dedicated\nsearch for boosted dark matter upscattered by cosmic rays using the PROSPECT\nreactor antineutrino experiment. We show that such a flux of upscattered dark\nmatter would display characteristic diurnal sidereal modulation, and use this\nto set new experimental constraints on sub-GeV dark matter exhibiting large\ninteraction cross-sections.\n"}
{"abstract": "  Polyelectrolyte-surfactant complexes (PESCs) have long been employed as\noil-in-water (o/w) emulsions stabilizers, but never in the structure of\ncolloidal complex coacervates providing a Pickering effect. The complexed state\nof PESCs could make them unsuitable o/w Pickering emulsifiers, which instead\nrequire a balance between colloidal structure and stability, amphiphilicity and\nwettability. Here we hypothesize that PESCs coacervates are efficient Pickering\nstabilizers. Instead of classical surfactants, we employ sophorolipid (SL)\nbiosurfactants, atypical anionic/neutral stimuli-responsive biosurfactants.\nDespite their tunable charge and mild amphiphilic character, they can be used\nin combination with cationic/neutral polyelectrolytes (chitosan, CHL, or\npoly-L-lysine, PLL) to form PESC coacervates for the development of biobased,\nbut also pH-switchable, Pickering emulsions.Aqueous solutions of SL-CHL (or\nSL-PLL) complex coacervates are emulsified with dodecane. Confocal laser\nscanning microscopy (CLSM) and scanning electron microscopy under cryogenic\nconditions (cryo-SEM) demonstrate the Pickering effect, while optical\nmicroscopy and oscillatory rheology respectively assess the emulsion formation\nand relative viscoelastic properties.Both SL-CHL and SL-PLL PESCs stabilize o/w\nemulsions up to $\\Phi$oil of 0.7 only in the pH region of complex coacervation\n(6 < pH < 9): outside this range, phase separation occurs. Rheology shows a\ntypical solid-like response and mechanical recovery upon applying large\ndeformations. CLSM and cryo-SEM highlight a colloidal structure, associated to\nthe complex coacervates, of the oil/water interface and suggest a Pickering\neffect. These findings demonstrate the Pickering effect from PESC coacervates\nand the possibility to use biobased and biocompatible components, with\napplication potential in cosmetics, food science, or oil recovery.\n"}
{"abstract": "  Autonomous trading robots have been studied in artificial intelligence area\nfor quite some time. Many AI techniques have been tested for building\nautonomous agents able to trade financial assets. These initiatives include\ntraditional neural networks, fuzzy logic, reinforcement learning but also more\nrecent approaches like deep neural networks and deep reinforcement learning.\nMany developers claim to be successful in creating robots with great\nperformance when simulating execution with historical price series, so called\nbacktesting. However, when these robots are used in real markets frequently\nthey present poor performance in terms of risks and return. In this paper, we\npropose an open source framework, called mt5se, that helps the development,\nbacktesting, live testing and real operation of autonomous traders. We built\nand tested several traders using mt5se. The results indicate that it may help\nthe development of better traders. Furthermore, we discuss the simple\narchitecture that is used in many studies and propose an alternative multiagent\narchitecture. Such architecture separates two main concerns for portfolio\nmanager (PM) : price prediction and capital allocation. More than achieve a\nhigh accuracy, a PM should increase profits when it is right and reduce loss\nwhen it is wrong. Furthermore, price prediction is highly dependent of asset's\nnature and history, while capital allocation is dependent only on analyst's\nprediction performance and assets' correlation. Finally, we discuss some\npromising technologies in the area.\n"}
{"abstract": "  Reinforcement learning (RL) is always the preferred embodiment to construct\nthe control strategy of complex tasks, like asymmetric assembly tasks. However,\nthe convergence speed of reinforcement learning severely restricts its\npractical application. In this paper, the convergence is first accelerated by\ncombining RL and compliance control. Then a completely innovative progressive\nextension of action dimension (PEAD) mechanism is proposed to optimize the\nconvergence of RL algorithms. The PEAD method is verified in DDPG and PPO. The\nresults demonstrate the PEAD method will enhance the data-efficiency and\ntime-efficiency of RL algorithms as well as increase the stable reward, which\nprovides more potential for the application of RL.\n"}
{"abstract": "  We numerically investigate the propagation of plane gravitational waves in\nthe form of an initial boundary value problem with de Sitter initial data. The\nfull non-linear Einstein equations with positive cosmological constant\n$\\lambda$ are written in the Friedrich-Nagy gauge which yields a wellposed\nsystem. The propagation of a single wave and the collision of two with colinear\npolarization are studied and contrasted with their Minkowskian analogues.\nUnlike with $\\lambda=0$, critical behaviours are found with $\\lambda>0$ and are\nbased on the relationship between the wave profile and $\\lambda$. We find that\nchoosing boundary data close to one of these bifurcations results in a \"false\"\nsteady state which violates the constraints. Simulations containing\n(approximate) impulsive wave profiles are run and general features are\ndiscussed. Analytic results of Woodard and Tsamis, which describe how\ngravitational waves could affect an expansion rate at an initial instance of\ntime, are explored and generalized to the entire space-time. Finally we put\nforward boundary conditions that, at least locally, slow down the expansion\nconsiderably for a time.\n"}
{"abstract": "  Atmospheric neutrinos are one of the most relevant natural neutrino sources\nthat can be exploited to infer properties about cosmic rays and neutrino\noscillations. The Jiangmen Underground Neutrino Observatory (JUNO) experiment,\na 20 kton liquid scintillator detector with excellent energy resolution is\ncurrently under construction in China. JUNO will be able to detect several\natmospheric neutrinos per day given the large volume. A study on the JUNO\ndetection and reconstruction capabilities of atmospheric $\\nu_e$ and $\\nu_\\mu$\nfluxes is presented in this paper. In this study, a sample of atmospheric\nneutrino Monte Carlo events has been generated, starting from theoretical\nmodels, and then processed by the detector simulation. The excellent timing\nresolution of the 3'' PMT light detection system of JUNO detector and the much\nhigher light yield for scintillation over Cherenkov allow to measure the time\nstructure of the scintillation light with very high precision. Since $\\nu_e$\nand $\\nu_\\mu$ interactions produce a slightly different light pattern, the\ndifferent time evolution of light allows to discriminate the flavor of primary\nneutrinos. A probabilistic unfolding method has been used, in order to infer\nthe primary neutrino energy spectrum from the detector experimental\nobservables. The simulated spectrum has been reconstructed between 100 MeV and\n10 GeV, showing a great potential of the detector in the atmospheric low energy\nregion.\n"}
{"abstract": "  Recommender systems (RSs) have emerged as very useful tools to help customers\nwith their decision-making process, find items of their interest, and alleviate\nthe information overload problem. There are two different lines of approaches\nin RSs: (1) general recommenders with the main goal of discovering long-term\nusers' preferences, and (2) sequential recommenders with the main focus of\ncapturing short-term users' preferences in a session of user-item interaction\n(here, a session refers to a record of purchasing multiple items in one\nshopping event). While considering short-term users' preferences may satisfy\ntheir current needs and interests, long-term users' preferences provide users\nwith the items that they may interact with, eventually. In this thesis, we\nfirst focus on improving the performance of general RSs. Most of the existing\ngeneral RSs tend to exploit the users' rating patterns on common items to\ndetect similar users. The data sparsity problem (i.e. the lack of available\ninformation) is one of the major challenges for the current general RSs, and\nthey may fail to have any recommendations when there are no common items of\ninterest among users. We call this problem data sparsity with no feedback on\ncommon items (DSW-n-FCI). To overcome this problem, we propose a\npersonality-based RS in which similar users are identified based on the\nsimilarity of their personality traits.\n"}
{"abstract": "  This paper provides a general framework for Stein's density method for\nmultivariate continuous distributions. The approach associates to any\nprobability density function a canonical operator and Stein class, as well as\nan infinite collection of operators and classes which we call standardizations.\nThese in turn spawn an entire family of Stein identities and characterizations\nfor any continuous distribution on $\\mathbb{R}^d$, among which we highlight\nthose based on the score function and the Stein kernel. A feature of these\noperators is that they do not depend on normalizing constants. A new definition\nof Stein kernel is introduced and examined; integral formulas are obtained\nthrough a connection with mass transport, as well as ready-to-use explicit\nformulas for elliptical distributions. The flexibility of the kernels is used\nto compare in Stein discrepancy (and therefore 2-Wasserstein distance) between\ntwo normal distributions, Student and normal distributions, as well as two\nnormal-gamma distributions. Upper and lower bounds on the 1-Wasserstein\ndistance between continuous distributions are provided, and computed for a\nvariety of examples: comparison between different normal distributions\n(improving on existing bounds in some regimes), posterior distributions with\ndifferent priors in a Bayesian setting (including logistic regression), centred\nAzzalini-Dalla Valle distributions. Finally the notion of weak Stein equation\nand weak Stein factors is introduced, and new bounds are obtained for Lipschitz\ntest functions if the distribution admits a Poincar\\'e constant, which we use\nto compare in 1-Wasserstein distance between different copulas on the unit\nsquare.\n"}
{"abstract": "  Graph kernels are often used in bioinformatics and network applications to\nmeasure the similarity between graphs; therefore, they may be used to construct\nefficient graph classifiers. Many graph kernels have been developed thus far,\nbut to the best of our knowledge there is no existing graph kernel that\nconsiders all subgraphs to measure similarity. We propose a novel graph kernel\nthat applies a quantum computer to measure the graph similarity taking all\nsubgraphs into account by fully exploiting the power of quantum superposition\nto encode every subgraph into a feature. For the construction of the quantum\nkernel, we develop an efficient protocol that removes the index information of\nsubgraphs encoded in the quantum state. We also prove that the quantum computer\nrequires less query complexity to construct the feature vector than the\nclassical sampler used to approximate the same vector. A detailed numerical\nsimulation of a bioinformatics problem is presented to demonstrate that, in\nmany cases, the proposed quantum kernel achieves better classification accuracy\nthan existing graph kernels.\n"}
{"abstract": "  Matrix valued asymmetric truncated Toeplitz operators are compressions of\nmultiplication operators acting between two possibly different model spaces. In\nthis paper, we characterize matrix valued asymmetric truncated Toeplitz\noperators by using compressed shifts.\n"}
{"abstract": "  Let $R$ be a commutative ring with identity. In this paper, we introduce the\nconcept of quasi $J$-ideal which is a generalization of $J$-ideal. A proper\nideal of $R$ is called a quasi $J$-ideal if its radical is a $J$-ideal. Many\ncharacterizations of quasi $J$-ideals in some special rings are obtained. We\ncharacterize rings in which every proper ideal is quasi $J$-ideal. Further, as\na generalization of presimplifiable rings, we define the notion of quasi\npresimplifiable rings. We call a ring $R$ a quasi presimplifiable ring if\nwhenever $a,b\\in R$ and $a=ab$, then either $a$ is a nilpotent or $b$ is a\nunit. It is shown that a proper ideal $I$ that is contained in the Jacobson\nradical is a quasi $J$-ideal (resp. $J$-ideal) if and only if $R/I$ is a quasi\npresimplifiable (resp. presimplifiable) ring.\n"}
{"abstract": "  Rydberg atoms in optical tweezer arrays provide a playground for\nnonequilibrium quantum many-body physics. The PXP model describes the dynamics\nof such systems in the strongly interacting Rydberg blockade regime and notably\nexhibits weakly nonergodic dynamics due to quantum many-body scars. Here, we\nstudy the PXP model in a strong staggered external field, which has been\nproposed to manifest quasiparticle confinement in light of a mapping to a\nlattice gauge theory. We characterize this confining regime using both\nnumerical exact diagonalization and perturbation theory around the strong-field\nlimit. In addition to the expected emergent symmetry generated by the staggered\nfield, we find a second emergent symmetry that is special to the PXP model. The\ninterplay between these emergent symmetries and the Rydberg blockade constraint\ndramatically slows down the system's dynamics beyond naive expectations. We\ndevise a nested Schrieffer-Wolff perturbation theory to properly account for\nthe new emergent symmetry and show that this treatment is essential to\nunderstand the numerically observed relaxation time scales. We also discuss\nconnections to Hilbert space fragmentation and trace the origin of the new\nemergent symmetry to a \"nearly-$SU(2)$\" algebra discovered in the context of\nmany-body scarring.\n"}
{"abstract": "  New transitions in neutron rich $^{100}$Y have been identified in a\n$^9$Be+$^{238}$U experiment with mass- and Z- gates to provide full fragment\nidentification. These transitions and high spin levels of $^{100}$Y have been\ninvestigated by analyzing the high statistics $\\gamma$-$\\gamma$-$\\gamma$ and\n$\\gamma$-$\\gamma$-$\\gamma$-$\\gamma$ coincidence data from the spontaneous\nfission of $^{252}$Cf at the Gammasphere detector array. Two new bands, 14 new\nlevels and 23 new transitions have been identified. The $K^{\\pi}=4^+$ new band\ndecaying to an 1s isomeric state is assigned to be the high-$K$\nGallagher-Moszkowski (GM) partner of the known $K^{\\pi}=1^+$ band, with the\n$\\pi 5/2[522] \\otimes \\nu 3/2[411]$ configuration. This 4$^+$ band is also\nproposed to be the pseudo spin partner of the new $K^{\\pi}=5^+$ band with a\n5$^{+}$ $\\pi 5/2[422] \\otimes \\nu 5/2[413]$ configuration, to form a $\\pi\n5/2[422] \\otimes \\nu [312$ $5/2,3/2]$ neutron pseudospin doublet. Constrained\ntriaxial covariant density functional theory and quantal particle rotor model\ncalculations have been applied to interpret the band structure and available\nelectromagnetic transition probabilities and are found in good agreement with\nexperimental values.\n"}
{"abstract": "  Zero-shot learning (ZSL) aims to transfer knowledge from seen classes to\nsemantically related unseen classes, which are absent during training. The\npromising strategies for ZSL are to synthesize visual features of unseen\nclasses conditioned on semantic side information and to incorporate\nmeta-learning to eliminate the model's inherent bias towards seen classes.\nWhile existing meta generative approaches pursue a common model shared across\ntask distributions, we aim to construct a generative network adaptive to task\ncharacteristics. To this end, we propose an Attribute-Modulated generAtive\nmeta-model for Zero-shot learning (AMAZ). Our model consists of an\nattribute-aware modulation network, an attribute-augmented generative network,\nand an attribute-weighted classifier. Given unseen classes, the modulation\nnetwork adaptively modulates the generator by applying task-specific\ntransformations so that the generative network can adapt to highly diverse\ntasks. The weighted classifier utilizes the data quality to enhance the\ntraining procedure, further improving the model performance. Our empirical\nevaluations on four widely-used benchmarks show that AMAZ outperforms\nstate-of-the-art methods by 3.8% and 3.1% in ZSL and generalized ZSL settings,\nrespectively, demonstrating the superiority of our method. Our experiments on a\nzero-shot image retrieval task show AMAZ's ability to synthesize instances that\nportray real visual characteristics.\n"}
{"abstract": "  Echo-enabled harmonic generation free-electron lasers (EEHG FELs) are\npromising candidates to produce fully coherent soft x-ray pulses by virtue of\nefficient high harmonic frequency up-conversion from UV lasers. The ultimate\nspectral limit of EEHG, however, remains unclear, because of the broadening and\ndistortions induced in the output spectrum by residual broadband energy\nmodulations in the electron beam. We present a mathematical description of the\nimpact of incoherent (broadband) energy modulations on the bunching spectrum\nproduced by the microbunching instability through both the accelerator and the\nEEHG line. The model is in agreement with a systematic experimental\ncharacterization of the FERMI EEHG FEL in the photon energy range $130-210$ eV.\nWe find that amplification of electron beam energy distortions primarily in the\nEEHG dispersive sections explains an observed reduction of the FEL spectral\nbrightness that is proportional to the EEHG harmonic number. Local maxima of\nthe FEL spectral brightness and of the spectral stability are found for a\nsuitable balance of the dispersive sections' strength and the first seed laser\npulse energy. Such characterization provides a benchmark for user experiments\nand future EEHG implementations designed to reach shorter wavelengths.\n"}
{"abstract": "  Clinical diagnosis, which aims to assign diagnosis codes for a patient based\non the clinical note, plays an essential role in clinical decision-making.\nConsidering that manual diagnosis could be error-prone and time-consuming, many\nintelligent approaches based on clinical text mining have been proposed to\nperform automatic diagnosis. However, these methods may not achieve\nsatisfactory results due to the following challenges. First, most of the\ndiagnosis codes are rare, and the distribution is extremely unbalanced. Second,\nexisting methods are challenging to capture the correlation between diagnosis\ncodes. Third, the lengthy clinical note leads to the excessive dispersion of\nkey information related to codes. To tackle these challenges, we propose a\nnovel framework to combine the inheritance-guided hierarchical assignment and\nco-occurrence graph propagation for clinical automatic diagnosis. Specifically,\nwe propose a hierarchical joint prediction strategy to address the challenge of\nunbalanced codes distribution. Then, we utilize graph convolutional neural\nnetworks to obtain the correlation and semantic representations of medical\nontology. Furthermore, we introduce multi attention mechanisms to extract\ncrucial information. Finally, extensive experiments on MIMIC-III dataset\nclearly validate the effectiveness of our method.\n"}
{"abstract": "  Several speculative visions are conjecturing on what 6G services will be able\nto offer at the horizon of 2030. Nevertheless, the 6G design process is at its\npreliminary stages. The reality today is that hardware, technologies and new\nmaterials required to effectively meet the unprecedented performance targets\nrequired for future 6G services and network operation, have not been designed,\ntested or even do not exist yet. Today, a solid vision on the cost-benefit\ntrade-offs of machine learning and artificial intelligence support for 6G\nnetwork and services operation optimization is missing. This includes the\npossible support from hardware efficiency, operation effectiveness and, the\nimmeasurable cost due to data acquisition-transfer-processing. The contribution\nof this paper is three-fold. This is the first paper deriving crucial 6G key\nperformance indicators on hardware and technology design. Second, we present a\nnew hardware technologies design methodology conceived to enable the effective\nsoftware-hardware components integration required to meet the challenging\nperformance envisioned for future 6G networks. Third, we suggest a paradigm\nshift towards goal-oriented and semantic communications, in which a totally new\nopportunity of joint design of hardware, artificial intelligence and effective\ncommunication is offered. The proposed vision is consolidated by our recent\nresults on hardware, technology and machine learning performance.\n"}
{"abstract": "  In order to consider j-wise relative r-primality conditions that do not\nnecessarily require all j-tuples of elements in a Dedekind domain to be\nrelatively r-prime, we define the notion of j-wise relative r-primality with\nrespect to a fixed j-uniform hypergraph H. This allows us to provide further\ngeneralisations to several results on natural densities not only for a ring of\nalgebraic integers, but also for the ring F_q[x].\n"}
{"abstract": "  Humans leverage the dynamics of the environment and their own bodies to\naccomplish challenging tasks such as grasping an object while walking past it\nor pushing off a wall to turn a corner. Such tasks often involve switching\ndynamics as the robot makes and breaks contact. Learning these dynamics is a\nchallenging problem and prone to model inaccuracies, especially near contact\nregions. In this work, we present a framework for learning composite dynamical\nbehaviors from expert demonstrations. We learn a switching linear dynamical\nmodel with contacts encoded in switching conditions as a close approximation of\nour system dynamics. We then use discrete-time LQR as the differentiable policy\nclass for data-efficient learning of control to develop a control strategy that\noperates over multiple dynamical modes and takes into account discontinuities\ndue to contact. In addition to predicting interactions with the environment,\nour policy effectively reacts to inaccurate predictions such as unanticipated\ncontacts. Through simulation and real world experiments, we demonstrate\ngeneralization of learned behaviors to different scenarios and robustness to\nmodel inaccuracies during execution.\n"}
{"abstract": "  Molecular dynamics facilitates the simulation of a complex system to be\nanalyzed at molecular and atomic levels. Simulations can last a long period of\ntime, even months. Due to this cause the graphics processing units (GPUs) and\nmulti-core systems are used as solutions to overcome this impediment. The\ncurrent paper describes a comparison done between these two kinds of systems.\nThe first system used implies the graphics processing unit, respectively CUDA\nwith the OpenMM molecular dynamics package and OpenCL that allows the kernels\nto run on the GPU. This simulation is done on a new thermostat which mixes the\nBerendsen thermostat with the Langevin dynamics. The second comprises the\nmolecular dynamics simulation and energy minimization package GROMACS which is\nbased on a parallelization through MPI (Message Passing Interface) on\nmulti-core systems. The second simulation uses another new thermostat algorithm\nrelated respectively, dissipative particle dynamics - isotropic type (DPD-ISO).\nBoth thermostats are innovative, based on a new theory developed by us. Results\nshow that parallelization on multi-core systems has a performance up to 33\ntimes greater than the one performed on the graphics processing unit. In both\ncases temperature of the system was maintained close to the one taken as\nreference. For the simulation using the CUDA GPU, the faster runtime was\nobtained when the number of processors was equal to four, the simulation speed\nbeing 3.67 times faster compared to the case of only one processor.\n"}
{"abstract": "  We are proposing a novel fabrication method for single crystal diamond\nscanning probes, exploiting the method of Faraday cage angled etching (FCAE).\nElectron-beam lithography (EBL) and inductively coupled plasma (ICP) etching as\nstate-of-the-art method is often limiting the achievable diamond structure\ngeometries in aspect ratio and complexity. This poses challenges, e.g., for\nrealizing nitrogen vacancy (NV) centre magnetometry scanning probes. Combining\na planar design with FCAE offers several advantages over the established\nfabrication technology of nano-opto-mechanical diamond devices. Here, we report\non the direct comparison of both approaches and present first proof-of-concept\nFCAE-prototypes for scanning probe applications.\n"}
{"abstract": "  Polycyclic Aromatic Hydrocarbons (PAHs) have long been invoked in the study\nof interstellar and protostellar sources, but the unambiguous identification of\nany individual PAH has proven elusive until very recently. As a result, the\nformation mechanisms for this important class of molecules remain poorly\nconstrained. Here we report the first interstellar detection of a pure\nhydrocarbon PAH, indene (C$_9$H$_8$), as part of the GBT Observations of TMC-1:\nHunting for Aromatic Molecules (GOTHAM) survey. This detection provides a new\navenue for chemical inquiry, complementing the existing detections of\nCN-functionalized aromatic molecules. From fitting the GOTHAM observations,\nindene is found to be the most abundant organic ring detected in TMC-1 to date.\nAnd from astrochemical modeling with NAUTILUS, the observed abundance is\ngreater than the model's prediction by several orders of magnitude suggesting\nthat current formation pathways in astrochemical models are incomplete. The\ndetection of indene in relatively high abundance implies related species such\nas cyanoindene, cyclopentadiene, toluene, and styrene may be detectable in dark\nclouds.\n"}
{"abstract": "  We examine the use of time series data, derived from Electric Cell-substrate\nImpedance Sensing (ECIS), to differentiate between standard mammalian cell\ncultures and those infected with a mycoplasma organism. We perform\nfeature-based classification, extracting interpretable features from the ECIS\ntime courses. We can achieve high classification accuracy using only two\nfeatures, which depend on the cell line under examination. Initial results also\nshow the existence of experimental variation between plates and suggest types\nof features that may prove more robust to such variation. Our paper is the\nfirst to perform a broad examination of ECIS time course features in the\ncontext of detecting contamination, and to describe and suggest possibilities\nfor ameliorating plate-to-plate variation.\n"}
{"abstract": "  We investigated microscopic lattice states in the donor-acceptor ionic Mott\ninsulator, TTF-BA, by $^{79}$Br-NQR spectroscopy to explore cross-correlated\nfluctuations between spin, charge and lattice. A ferroelectric transition with\nlattice dimerization is captured by a NQR line splitting with the critical\nexponent $\\beta$ of 0.40, as expected in the 3D Ising universality class, and a\npeak formation in the spin-lattice relaxation rate $T_1^{-1}$ at the transition\ntemperature, $T_\\mathrm{c}$, of 53 K. Notably, $T_1^{-1}$ does not obey the\nconventional $T^2$ law expected for the Raman process of phonons even far above\n$T_\\mathrm{c}$, indicating the emergence of extraordinary lattice fluctuations.\nThey are very probably associated with polar fluctuations in the paraelectric\nand paramagnetic phase of TTF-BA and explain the previous observation of the\nanomalously suppressed paramagnetic spin susceptibility, which was conjectured\nto be due to the local spin-singlet pairing prior to the nonmagnetic\nferroelectric order [K. Sunami $et$ $al$., Phys. Rev. Res. 2, 043333 (2020)].\n"}
{"abstract": "  We launched a community platform for collecting the ATC speech world-wide in\nthe ATCO2 project. Filtering out unseen non-English speech is one of the main\ncomponents in the data processing pipeline. The proposed English Language\nDetection (ELD) system is based on the embeddings from Bayesian subspace\nmultinomial model. It is trained on the word confusion network from an ASR\nsystem. It is robust, easy to train, and light weighted. We achieved 0.0439\nequal-error-rate (EER), a 50% relative reduction as compared to the\nstate-of-the-art acoustic ELD system based on x-vectors, in the in-domain\nscenario. Further, we achieved an EER of 0.1352, a 33% relative reduction as\ncompared to the acoustic ELD, in the unseen language (out-of-domain) condition.\nWe plan to publish the evaluation dataset from the ATCO2 project.\n"}
{"abstract": "  Active aging technologies are increasingly designed to support an active\nlifestyle. However, the way in which they are designed can raise different\nbarriers to acceptance of and use by older adults. Their designers can adopt a\nnegative stereotype of aging. Thorough understanding of user requirements is\ncentral to this problem. This paper investigates user requirements for\ntechnologies that encourage an active lifestyle and provide older people with\nthe means to self-manage their physical, mental, and emotional health. This\nrequires consideration of the person and the sociotechnical context of use. We\ndescribe our work in collecting and analyzing older adults' requirements for a\ntechnology which enables an active lifestyle. The main contribution of the\npaper is a model of user requirements for inclusive technology for older\npeople.\n"}
{"abstract": "  Motivation: SBML is the most widespread language for the definition of\nbiochemical models. Although dozens of SBML simulators are available, there is\na general lack of support to the integration of SBML models within\nopen-standard general-purpose simulation ecosystems. This hinders co-simulation\nand integration of SBML models within larger model networks, in order to, e.g.\nenable in silico clinical trials of drugs, pharmacological protocols, or\nengineering artefacts such as biomedical devices against Virtual Physiological\nHuman models. Modelica is one of the most popular existing open-standard\ngeneral-purpose simulation languages, supported by many simulators. Modelica\nmodels are especially suited for the definition of complex networks of\nheterogeneous models from virtually all application domains. Models written in\nModelica (and in 100+ other languages) can be readily exported into black-box\nFunctional Mock-Up Units (FMUs), and seamlessly co-simulated and integrated\ninto larger model networks within open-standard language-independent simulation\necosystems.\n  Results: In order to enable SBML model integration within heterogeneous model\nnetworks, we present SBML2Modelica, a software system translating SBML models\ninto well-structured, user-intelligible, easily modifiable Modelica models.\nSBML2Modelica is SBML Level 3 Version 2-compliant and succeeds on 96.47% of the\nSBML Test Suite Core (with a few rare, intricate and easily avoidable\ncombinations of constructs unsupported and cleanly signalled to the user). Our\nexperimental campaign on 613 models from the BioModels database (with up to\n5438 variables) shows that the major open-source (general-purpose) Modelica and\nFMU simulators achieve performance comparable to state-of-the-art specialized\nSBML simulators.\n  Availability and implementation: https://bitbucket.org/mclab/sbml2modelica\n"}
{"abstract": "  Ensemble clustering is a fundamental problem in the machine learning field,\ncombining multiple base clusterings into a better clustering result. However,\nmost of the existing methods are unsuitable for large-scale ensemble clustering\ntasks due to the efficiency bottleneck. In this paper, we propose a large-scale\nspectral ensemble clustering (LSEC) method to strike a good balance between\nefficiency and effectiveness. In LSEC, a large-scale spectral clustering based\nefficient ensemble generation framework is designed to generate various base\nclusterings within a low computational complexity. Then all based clustering\nare combined through a bipartite graph partition based consensus function into\na better consensus clustering result. The LSEC method achieves a lower\ncomputational complexity than most existing ensemble clustering methods.\nExperiments conducted on ten large-scale datasets show the efficiency and\neffectiveness of the LSEC method. The MATLAB code of the proposed method and\nexperimental datasets are available at https://github.com/Li-\nHongmin/MyPaperWithCode.\n"}
{"abstract": "  The duality symmetry between electricity and magnetism hidden in classical\nMaxwell equations suggests the existence of dual charges, which have usually\nbeen interpreted as magnetic charges and have not been observed in experiments.\nIn quantum electrodynamics (QED), both the electric and magnetic fields have\nbeen unified into one gauge field $A_{\\mu}$, which makes this symmetry\ninconspicuous. Here, we recheck the duality symmetry of QED by introducing a\ndual gauge field. Within the framework of gauge-field theory, we show that the\nelectric-magnetic duality symmetry cannot give any new conservation law. By\nchecking charge-charge interaction and specifically the quantum Lorentz force\nequation, we find that the dual charges are electric charges, not magnetic\ncharges. More importantly, we show that true magnetic charges are not\ncompatible with the gauge-field theory of QED, because the interaction between\na magnetic charge and an electric charge can not be mediated by gauge photons.\n"}
{"abstract": "  We study the dynamics of an initially flat interface between two immiscible\nfluids, with a vortex situated on it. We show how surface tension causes\nvorticity generation at a general curved interface. This creates a velocity\njump across the interface which increases quadratically in time, and causes the\nKelvin-Helmholtz instability. Surface tension thus acts as a destabiliser by\nvorticity creation, winning over its own tendency to stabilize by smoothing out\ninterfacial perturbations to reduce surface energy. We further show that this\ninstability is manifested within the vortex core at times larger than $\\sim (k\nWe)^{1/4}$ for a Weber number $We$ and perturbation wavenumber $k$, destroying\nthe flow structure. The vorticity peels off into small-scale structures away\nfrom the interface. Using energy balance we provide the growth with time in\ntotal interface length. A density difference between the fluids produces\nadditional instabilities outside the vortex core due to centrifugal effects. We\ndemonstrate the importance of this mechanism in two-dimensional turbulence\nsimulations with a prescribed initial interface. We show an increase in kinetic\nenergy at large wave numbers due to the creation of small-scale vortices by the\naction of surface tension.\n"}
{"abstract": "  Linear mixed-effects models are commonly used to analyze clustered data\nstructures. There are numerous packages to fit these models in R and conduct\nlikelihood-based inference. The implementation of resampling-based procedures\nfor inference are more limited. In this paper, we introduce the lmeresampler\npackage for bootstrapping nested linear mixed-effects models fit via lme4 or\nnlme. Bootstrap estimation allows for bias correction, adjusted standard errors\nand confidence intervals for small samples sizes and when distributional\nassumptions break down. We will also illustrate how bootstrap resampling can be\nused to diagnose this model class. In addition, lmeresampler makes it easy to\nconstruct interval estimates of functions of model parameters.\n"}
{"abstract": "  In blind source separation of speech signals, the inherent imbalance in the\nsource spectrum poses a challenge for methods that rely on single-source\ndominance for the estimation of the mixing matrix. We propose an algorithm\nbased on the directional sparse filtering (DSF) framework that utilizes the\nLehmer mean with learnable weights to adaptively account for source imbalance.\nPerformance evaluation in multiple real acoustic environments show improvements\nin source separation compared to the baseline methods.\n"}
{"abstract": "  Three-dimensional metallic microstructures find applications as stents in\nmedicine, as ultrabroadband antennas in communications, in micromechanical\nparts or as structures of more fundamental interest in photonics like\nmetamaterials. Direct metal printing of such structures using three-dimensional\nlaser lithography is a promising approach, which is not extensively applied\nyet, as fabrication speed, surface quality, and stability of the resulting\nstructures are limited so far. In order to identify the limiting factors, we\ninvestigate the influence of light-particle interactions and varying scan speed\non heat generation and particle deposition in direct laser writing of silver.\nWe introduce a theoretical model which captures diffusion of particles and heat\nas well as the fluid dynamics of the photo-resist. Chemical reactions are\nexcluded from the model but particle production is calibrated using\nexperimental data. We find that optical forces generally surmount those due to\nconvection of the photo-resist. Simulations predict overheating of the\nphoto-resist at laser powers similar to those found in experiments. The thermal\nsensitivity of the system is essentially determined by the largest particles\npresent in the laser focus. Our results suggest that to improve particle\ndeposition and to achieve higher writing speeds in metal direct laser writing,\nstrong optical trapping of the emerging particles is desirable. Furthermore,\nprecise control of the particle size reduces the risk of spontaneous\noverheating.\n"}
{"abstract": "  We present the complete two-loop corrections in massless QCD for the\nproduction of two photons and a jet, taking into account all color structures.\nIn particular, we analytically compute all two-loop helicity amplitudes for the\nquark-antiquark, quark-gluon, and antiquark-gluon channel, and check them with\nan independent calculation of the polarization-summed interference with the\ntree amplitude. This is the first time that two-loop QCD corrections to a\nfive-point scattering process have been computed beyond the leading-color\napproximation for all helicity configurations.\n"}
{"abstract": "  Aspect sentiment triplet extraction (ASTE), which aims to identify aspects\nfrom review sentences along with their corresponding opinion expressions and\nsentiments, is an emerging task in fine-grained opinion mining. Since ASTE\nconsists of multiple subtasks, including opinion entity extraction, relation\ndetection, and sentiment classification, it is critical and challenging to\nappropriately capture and utilize the associations among them. In this paper,\nwe transform ASTE task into a multi-turn machine reading comprehension (MTMRC)\ntask and propose a bidirectional MRC (BMRC) framework to address this\nchallenge. Specifically, we devise three types of queries, including\nnon-restrictive extraction queries, restrictive extraction queries and\nsentiment classification queries, to build the associations among different\nsubtasks. Furthermore, considering that an aspect sentiment triplet can derive\nfrom either an aspect or an opinion expression, we design a bidirectional MRC\nstructure. One direction sequentially recognizes aspects, opinion expressions,\nand sentiments to obtain triplets, while the other direction identifies opinion\nexpressions first, then aspects, and at last sentiments. By making the two\ndirections complement each other, our framework can identify triplets more\ncomprehensively. To verify the effectiveness of our approach, we conduct\nextensive experiments on four benchmark datasets. The experimental results\ndemonstrate that BMRC achieves state-of-the-art performances.\n"}
{"abstract": "  In this paper, we find the existence of critical features hidden in Deep\nNeuralNetworks (DNNs), which are imperceptible but can actually dominate the\noutputof DNNs. We call these features dominant patterns. As the name suggests,\nfor a natural image, if we add the dominant pattern of a DNN to it, the output\nof this DNN is determined by the dominant pattern instead of the original\nimage, i.e., DNN's prediction is the same with the dominant pattern's. We\ndesign an algorithm to find such patterns by pursuing the insensitivity in the\nfeature space. A direct application of the dominant patterns is the Universal\nAdversarial Perturbations(UAPs). Numerical experiments show that the found\ndominant patterns defeat state-of-the-art UAP methods, especially in label-free\nsettings. In addition, dominant patterns are proved to have the potential to\nattack downstream tasks in which DNNs share the same backbone. We claim that\nDNN-specific dominant patterns reveal some essential properties of a DNN and\nare of great importance for its feature analysis and robustness enhancement.\n"}
{"abstract": "  Recently many first and second order variants of SGD have been proposed to\nfacilitate training of Deep Neural Networks (DNNs). A common limitation of\nthese works stem from the fact that they use the same learning rate across all\ninstances present in the dataset. This setting is widely adopted under the\nassumption that loss functions for each instance are similar in nature, and\nhence, a common learning rate can be used. In this work, we relax this\nassumption and propose an optimization framework which accounts for difference\nin loss function characteristics across instances. More specifically, our\noptimizer learns a dynamic learning rate for each instance present in the\ndataset. Learning a dynamic learning rate for each instance allows our\noptimization framework to focus on different modes of training data during\noptimization. When applied to an image classification task, across different\nCNN architectures, learning dynamic learning rates leads to consistent gains\nover standard optimizers. When applied to a dataset containing corrupt\ninstances, our framework reduces the learning rates on noisy instances, and\nimproves over the state-of-the-art. Finally, we show that our optimization\nframework can be used for personalization of a machine learning model towards a\nknown targeted data distribution.\n"}
{"abstract": "  Understanding the attitudes expressed in texts, also known as stance\ndetection, plays an important role in systems aiming to detect false\ninformation online, be it misinformation (unintentionally false) or\ndisinformation (intentionally false, spread deliberately with malicious\nintent). Stance detection has been framed in different ways in the literature,\nincluding (a) as a component of fact-checking, rumour detection, and detecting\npreviously fact-checked claims, or (b) as a task in its own right; and here we\nlook at both. While there have been prior efforts to contrast stance detection\nwith other related tasks such as argumentation mining and sentiment analysis,\nthere has been no survey examining the relationship between stance detection\nand mis- and disinformation detection. Here we aim to bridge this gap. In\nparticular, we review and analyse existing work in this area, with mis- and\ndisinformation in focus, and then we discuss lessons learnt and future\nchallenges.\n"}
{"abstract": "  Radio observations of tidal disruption events (TDEs) - when a star is tidally\ndisrupted by a supermassive black hole (SMBH) - provide a unique laboratory for\nstudying outflows in the vicinity of SMBHs and their connection to accretion\nonto the SMBH. Radio emission has been detected in only a handful of TDEs so\nfar. Here, we report the detection of delayed radio flares from an\noptically-discovered TDE. Our prompt radio observations of the TDE ASASSN-15oi\nshowed no radio emission until the detection of a flare six months later,\nfollowed by a second and brighter flare, years later. We find that the standard\nscenario, in which an outflow is launched briefly after the stellar disruption,\nis unable to explain the combined temporal and spectral properties of the\ndelayed flare. We suggest that the flare is due to the delayed ejection of an\noutflow, perhaps following a transition in accretion states. Our discovery\nmotivates observations of TDEs at various timescales and highlights a need for\nnew models.\n"}
{"abstract": "  An instrument and software algorithm is described for the purpose of\ncharacterization of large single crystals at the Alignment Facility (ALF) of\nthe ISIS spallation neutron source. We describe a method for both\ncharacterizing the quality of the sample and also aligning it in a particular\nscattering plane. We present a software package written for this instrument and\ndemonstrate its utility by way of an example of the structural characterization\nof large singles crystals of Pb(Mg$_{1/3}$Nb$_{2/3}$)O$_{3}$. We suggest\nextensions and modifications of characterization instruments for future\nimproved beamlines. It is hoped that this software will be used by the neutron\nuser community for pre characterizing large single crystals for spectroscopy\nexperiments and that future facilities will include such a facility as part of\nthe spectroscopy suite at spallation neutron sources.\n"}
{"abstract": "  If quantum mechanics is taken for granted the randomness derived from it may\nbe vacuous or even delusional, yet sufficient for many practical purposes.\n\"Random\" quantum events are intimately related to the emergence of both\nspace-time as well as the identification of physical properties through which\nso-called objects are aggregated. We also present a brief review of the\nmetaphysics of indeterminism.\n"}
{"abstract": "  We establish a connection between two areas of independent interest in\nrepresentation theory, namely Koszul duality and higher homological algebra.\nThis is done through a generalization of the notion of $T$-Koszul algebras, for\nwhich we obtain a higher version of classical Koszul duality. Our approach is\nmotivated by and has applications for $n$-hereditary algebras. In particular,\nwe characterize an important class of $n$-$T$-Koszul algebras of highest degree\n$a$ in terms of $(na-1)$-representation infinite algebras. As a consequence, we\nsee that an algebra is $n$-representation infinite if and only if its trivial\nextension is $(n+1)$-Koszul with respect to its degree $0$ part. Furthermore,\nwe show that when an $n$-representation infinite algebra is $n$-representation\ntame, then the bounded derived categories of graded modules over the trivial\nextension and over the associated $(n+1)$-preprojective algebra are equivalent.\nIn the $n$-representation finite case, we introduce the notion of almost\n$n$-$T$-Koszul algebras and obtain similar results.\n"}
{"abstract": "  In this work, we develop the theory of quasi-exact fault-tolerant quantum\n(QEQ) computation, which uses qubits encoded into quasi-exact quantum\nerror-correction codes (\"quasi codes\"). By definition, a quasi code is a\nparametric approximate code that can become exact by tuning its parameters. The\nmodel of QEQ computation lies in between the two well-known ones: the usual\nnoisy quantum computation without error correction and the usual fault-tolerant\nquantum computation, but closer to the later. Many notions of exact quantum\ncodes need to be adjusted for the quasi setting. Here we develop quasi\nerror-correction theory using quantum instrument, the notions of quasi\nuniversality, quasi code distances, and quasi thresholds, etc. We find a wide\nclass of quasi codes which are called valence-bond-solid codes, and we use them\nas concrete examples to demonstrate QEQ computation.\n"}
{"abstract": "  This work is an update of a previous paper on the same topic published a few\nyears ago. With the dramatic progress in generative modeling, a suite of new\nquantitative and qualitative techniques to evaluate models has emerged.\nAlthough some measures such as Inception Score, Frechet Inception Distance,\nPrecision-Recall, and Perceptual Path Length are relatively more popular, GAN\nevaluation is not a settled issue and there is still room for improvement.\nHere, I describe new dimensions that are becoming important in assessing models\n(e.g. bias and fairness) and discuss the connection between GAN evaluation and\ndeepfakes. These are important areas of concern in the machine learning\ncommunity today and progress in GAN evaluation can help mitigate them.\n"}
{"abstract": "  Hardware faults on the regular 2-D computing array of a typical deep learning\naccelerator (DLA) can lead to dramatic prediction accuracy loss. Prior\nredundancy design approaches typically have each homogeneous redundant\nprocessing element (PE) to mitigate faulty PEs for a limited region of the 2-D\ncomputing array rather than the entire computing array to avoid the excessive\nhardware overhead. However, they fail to recover the computing array when the\nnumber of faulty PEs in any region exceeds the number of redundant PEs in the\nsame region. The mismatch problem deteriorates when the fault injection rate\nrises and the faults are unevenly distributed. To address the problem, we\npropose a hybrid computing architecture (HyCA) for fault-tolerant DLAs. It has\na set of dot-production processing units (DPPUs) to recompute all the\noperations that are mapped to the faulty PEs despite the faulty PE locations.\nAccording to our experiments, HyCA shows significantly higher reliability,\nscalability, and performance with less chip area penalty when compared to the\nconventional redundancy approaches. Moreover, by taking advantage of the\nflexible recomputing, HyCA can also be utilized to scan the entire 2-D\ncomputing array and detect the faulty PEs effectively at runtime.\n"}
{"abstract": "  What we expect from radiology AI algorithms will shape the selection and\nimplementation of AI in the radiologic practice. In this paper I consider\nprevailing expectations of AI and compare them to expectations that we have of\nhuman readers. I observe that the expectations from AI and radiologists are\nfundamentally different. The expectations of AI are based on a strong and\njustified mistrust about the way that AI makes decisions. Because AI decisions\nare not well understood, it is difficult to know how the algorithms will behave\nin new, unexpected situations. However, this mistrust is not mirrored in our\nexpectations of human readers. Despite well-proven idiosyncrasies and biases in\nhuman decision making, we take comfort from the assumption that others make\ndecisions in a way as we do, and we trust our own decision making. Despite poor\nability to explain decision making processes in humans, we accept explanations\nof decisions given by other humans. Because the goal of radiology is the most\naccurate radiologic interpretation, our expectations of radiologists and AI\nshould be similar, and both should reflect a healthy mistrust of complicated\nand partially opaque decision processes undergoing in computer algorithms and\nhuman brains. This is generally not the case now.\n"}
{"abstract": "  Quantum error mitigation is essential for computing on the noisy quantum\ncomputer with a limited number of qubits. In this paper, we propose a practical\nprotocol of error mitigation by virtually purifying the quantum state without\nqubit overhead or requiring only one ancillary qubit. In dual-state\npurification, we effectively generate a purified state with increased fidelity\nusing the erroneous state and its dual state, respectively, prepared with the\nnoisy quantum circuit and the dual map of its inverse circuit. Combined with\ntomography purification, we can make sure that the final estimate of an\nobservable is obtained from a pure state. The numerical result suggests that\nour protocol reduces the error by a rescaling factor decreasing with the qubit\nnumber and circuit depth, i.e. the performance of purification is better for\nlarger circuits. On a cloud quantum computer, we successfully demonstrate the\nreduced error with a quantum variational eigensolver circuit.\n"}
{"abstract": "  One important challenge of applying deep learning to electronic health\nrecords (EHR) is the complexity of their multimodal structure. EHR usually\ncontains a mixture of structured (codes) and unstructured (free-text) data with\nsparse and irregular longitudinal features -- all of which doctors utilize when\nmaking decisions. In the deep learning regime, determining how different\nmodality representations should be fused together is a difficult problem, which\nis often addressed by handcrafted modeling and intuition. In this work, we\nextend state-of-the-art neural architecture search (NAS) methods and propose\nMUltimodal Fusion Architecture SeArch (MUFASA) to simultaneously search across\nmultimodal fusion strategies and modality-specific architectures for the first\ntime. We demonstrate empirically that our MUFASA method outperforms established\nunimodal NAS on public EHR data with comparable computation costs. In addition,\nMUFASA produces architectures that outperform Transformer and Evolved\nTransformer. Compared with these baselines on CCS diagnosis code prediction,\nour discovered models improve top-5 recall from 0.88 to 0.91 and demonstrate\nthe ability to generalize to other EHR tasks. Studying our top architecture in\ndepth, we provide empirical evidence that MUFASA's improvements are derived\nfrom its ability to both customize modeling for each data modality and find\neffective fusion strategies.\n"}
{"abstract": "  This paper proposes a method to estimate the pose of a sensor mounted on a\nvehicle as the vehicle moves through the world, an important topic for\nautonomous driving systems. Based on a set of commonly deployed vehicular\nodometric sensors, with outputs available on automotive communication buses\n(e.g. CAN or FlexRay), we describe a set of steps to combine a planar odometry\nbased on wheel sensors with a suspension model based on linear suspension\nsensors. The aim is to determine a more accurate estimate of the camera pose.\nWe outline its usage for applications in both visualisation and computer\nvision.\n"}
{"abstract": "  A formal cyber reasoning framework for automating the threat hunting process\nis described. The new cyber reasoning methodology introduces an operational\nsemantics that operates over three subspaces -- knowledge, hypothesis, and\naction -- to enable human-machine co-creation of threat hypotheses and\nprotective recommendations. An implementation of this framework shows that the\napproach is practical and can be used to generalize evidence-based\nmulti-criteria threat investigations.\n"}
{"abstract": "  In this paper, we investigate projection-based intrusive and data-driven\nnon-intrusive model order reduction methods in numerical simulation of rotating\nthermal shallow water equation (RTSWE) in parametric and non-parametric form.\nDiscretization of the RTSWE in space with centered finite differences leads to\nHamiltonian system of ordinary differential equations with linear and quadratic\nterms. The full-order model (FOM) is obtained by applying linearly implicit\nKahan's method in time. Applying proper orthogonal decomposition with Galerkin\nprojection (POD-G), we construct the intrusive reduced-order model (ROM). We\napply operator inference (OpInf) with re-projection for non-intrusive\nreduced-order modeling. In the parametric case, we make use of the parameter\ndependency at the level of the PDE without interpolating between the reduced\noperators. The least-squares problem of the OpInf is regularized with the\nminimum norm solution. Both ROMs behave similar and are able to accurately\npredict the test and training data and capture system behavior in the\nprediction phase with several orders of computational speedup over the FOM. The\npreservation of system physics such as the conserved quantities of the RTSWE by\nboth ROMs enables that the models fit better to data and stable solutions are\nobtained in long-term predictions, which are robust to parameter changes.\n"}
{"abstract": "  Spontaneous collapse models and Bohmian mechanics are two different solutions\nto the measurement problem plaguing orthodox quantum mechanics. They have, a\npriori nothing in common. At a formal level, collapse models add a non-linear\nnoise term to the Schr\\\"odinger equation, and extract definite measurement\noutcomes either from the wave function (e.g. mass density ontology) or the\nnoise itself (flash ontology). Bohmian mechanics keeps the Schr\\\"odinger\nequation intact but uses the wave function to guide particles (or fields),\nwhich comprise the primitive ontology. Collapse models modify the predictions\nof orthodox quantum mechanics, whilst Bohmian mechanics can be argued to\nreproduce them. However, it turns out that collapse models and their primitive\nontology can be exactly recast as Bohmian theories. More precisely, considering\n(i) a system described by a non-Markovian collapse model, and (ii) an extended\nsystem where a carefully tailored bath is added and described by Bohmian\nmechanics, the stochastic wave-function of the collapse model is exactly the\nwave-function of the original system conditioned on the Bohmian hidden\nvariables of the bath. Further, the noise driving the collapse model is a\nlinear functional of the Bohmian variables. The randomness that seems\nprogressively revealed in the collapse models lies entirely in the initial\nconditions in the Bohmian-like theory. Our construction of the appropriate bath\nis not trivial and exploits an old result from the theory of open quantum\nsystems. This reformulation of collapse models as Bohmian theories brings to\nthe fore the question of whether there exists `unromantic' realist\ninterpretations of quantum theory that cannot ultimately be rewritten this way,\nwith some guiding law. It also points to important foundational differences\nbetween `true' (Markovian) collapse models and non-Markovian models.\n"}
{"abstract": "  An increasing number of machine learning models have been deployed in domains\nwith high stakes such as finance and healthcare. Despite their superior\nperformances, many models are black boxes in nature which are hard to explain.\nThere are growing efforts for researchers to develop methods to interpret these\nblack-box models. Post hoc explanations based on perturbations, such as LIME,\nare widely used approaches to interpret a machine learning model after it has\nbeen built. This class of methods has been shown to exhibit large instability,\nposing serious challenges to the effectiveness of the method itself and harming\nuser trust. In this paper, we propose S-LIME, which utilizes a hypothesis\ntesting framework based on central limit theorem for determining the number of\nperturbation points needed to guarantee stability of the resulting explanation.\nExperiments on both simulated and real world data sets are provided to\ndemonstrate the effectiveness of our method.\n"}
{"abstract": "  We present a novel verification technique to prove interesting properties of\na class of array programs with a symbolic parameter N denoting the size of\narrays. The technique relies on constructing two slightly different versions of\nthe same program. It infers difference relations between the corresponding\nvariables at key control points of the joint control-flow graph of the two\nprogram versions. The desired post-condition is then proved by inducting on the\nprogram parameter $N$, wherein the difference invariants are crucially used in\nthe inductive step. This contrasts with classical techniques that rely on\nfinding potentially complex loop invaraints for each loop in the program. Our\nsynergistic combination of inductive reasoning and finding simple difference\ninvariants helps prove properties of programs that cannot be proved even by the\nwinner of Arrays sub-category from SV-COMP 2021. We have implemented a\nprototype tool called diffy to demonstrate these ideas. We present results\ncomparing the performance of diffy with that of state-of-the-art tools.\n"}
{"abstract": "  Descriptive code comments are essential for supporting code comprehension and\nmaintenance. We propose the task of automatically generating comments for\noverriding methods. We formulate a novel framework which accommodates the\nunique contextual and linguistic reasoning that is required for performing this\ntask. Our approach features: (1) incorporating context from the class\nhierarchy; (2) conditioning on learned, latent representations of specificity\nto generate comments that capture the more specialized behavior of the\noverriding method; and (3) unlikelihood training to discourage predictions\nwhich do not conform to invariant characteristics of the comment corresponding\nto the overridden method. Our experiments show that the proposed approach is\nable to generate comments for overriding methods of higher quality compared to\nprevailing comment generation techniques.\n"}
{"abstract": "  We studied the COVID-19 pandemic evolution in selected African countries. For\neach country considered, we modeled simultaneously the data of the active,\nrecovered and death cases. In this study, we used a year of data since the\nfirst cases were reported. We estimated the time-dependent basic reproduction\nnumbers, $R_0$, and the fractions of infected but unaffected populations, to\noffer insights into containment and vaccine strategies in African countries. We\nfound that $R_0\\leq 4$ at the start of the pandemic but has since fallen to\n$R_0 \\sim 1$. The unaffected fractions of the populations studied vary between\n$1-10$\\% of the recovered cases.\n"}
{"abstract": "  Quantum embedding methods have become a powerful tool to overcome\ndeficiencies of traditional quantum modelling in materials science. However\nwhile these are systematically improvable in principle, in practice it is\nrarely possible to achieve rigorous convergence and often necessary to employ\nempirical parameters. Here, we formulate a quantum embedding, building on the\nmethods of density-matrix embedding theory combined with local correlation\napproaches of quantum chemistry, to ensure the ability to systematically\nconverge properties of real materials with accurate correlated wave~function\nmethods, controlled by a single, rapidly convergent parameter. By expanding\nsupercell size, basis set, and the resolution of the fluctuation space of an\nembedded fragment, we show that the systematic improvability of the approach\nyields accurate structural and electronic properties of realistic solids\nwithout empirical parameters, even across changes in geometry. Results are\npresented in insulating, semi-metallic, and more strongly correlated regimes,\nfinding state of the art agreement to experimental data.\n"}
{"abstract": "  The discovery of infinite-layer nickelate superconductors has spurred\nenormous interest. While the Ni$^{1+}$ cations possess nominally the same\n3d$^9$ configuration as Cu$^{2+}$ in high-$T_C$ cuprates, the electronic\nstructure consistencies and variances remain elusive, due to the lack of direct\nexperimental probes. Here, we present a soft x-ray photoemission spectroscopy\nstudy on both parent and doped infinite-layer Pr-nickelate thin films with a\ndoped perovskite reference. By identifying the Ni character with resonant\nphotoemission and comparison to density function theory + U calculations, we\nestimate U ~ 5 eV, smaller than the charge transfer energy $\\Delta$ ~ 8 eV, in\ncontrast to the cuprates being charge transfer insulators. Near the Fermi level\n(EF), we observe a signature of rare-earth spectral intensity in the parent\ncompound, which is depleted upon doping. The parent compound, \"self-doped\" from\nrare-earth electrons, exhibits higher density of states at EF but manifests\nweaker superconducting instability than the Sr-doped case, demonstrating a\ncomplex interplay between the strongly-correlated Ni 3d and the\nweakly-interacting rare-earth 5d states in these \"oxide-intermetallic\"\nnickelates.\n"}
{"abstract": "  The density matrix renormalization group (DMRG) of White 1992 remains to this\nday an integral component of many state-of-the-art methods for efficiently\nsimulating strongly correlated quantum systems. In quantum chemistry, QC-DMRG\nbecame a powerful tool for ab initio calculations with the non-relativistic\nSchr\\\"odinger equation. An important issue in QC-DMRG is the so-called ordering\nproblem -- the optimal ordering of DMRG sites corresponding to electronic\norbitals that produces the most accurate results. To this end, a commonly used\nheuristic is the grouping of strongly correlated orbitals as measured via\nquantum mutual information. In this work, we show how such heuristics can be\ndirectly related to minimizing the entanglement entropy of matrix product\nstates and, consequently, to the truncation error of a fixed bond dimension\napproximation. Key to establishing this link is the strong subadditivity of\nentropy. This provides rigorous theoretical justification for the orbital\nordering methods and suggests alternate ordering criteria.\n"}
{"abstract": "  We study the structure of string theory flux compactification for a general\nfamily of elliptic CY 3-folds. We investigate the locus of the attractor points\nof the flux compactification in type IIB string theory on the boundary\ncomponents of period domains. Specifically we give equations describing this\nlocus through the asymptotic of nilpotent orbits on period domains. our\napproach is a mixture of techniques of asymptotic Hodge theory and the\nnumerical period vectors used in physics.\n"}
{"abstract": "  Marginalising over families of Gaussian Process kernels produces flexible\nmodel classes with well-calibrated uncertainty estimates. Existing approaches\nrequire likelihood evaluations of many kernels, rendering them prohibitively\nexpensive for larger datasets. We propose a Bayesian Quadrature scheme to make\nthis marginalisation more efficient and thereby more practical. Through use of\nthe maximum mean discrepancies between distributions, we define a kernel over\nkernels that captures invariances between Spectral Mixture (SM) Kernels. Kernel\nsamples are selected by generalising an information-theoretic acquisition\nfunction for warped Bayesian Quadrature. We show that our framework achieves\nmore accurate predictions with better calibrated uncertainty than\nstate-of-the-art baselines, especially when given limited (wall-clock) time\nbudgets.\n"}
{"abstract": "  In many areas of interest, modern risk assessment requires estimation of the\nextremal behaviour of sums of random variables. We derive the first order\nupper-tail behaviour of the weighted sum of bivariate random variables under\nweak assumptions on their marginal distributions and their copula. The extremal\nbehaviour of the marginal variables is characterised by the generalised Pareto\ndistribution and their extremal dependence through subclasses of the limiting\nrepresentations of Ledford and Tawn (1997) and Heffernan and Tawn (2004). We\nfind that the upper tail behaviour of the aggregate is driven by different\nfactors dependent on the signs of the marginal shape parameters; if they are\nboth negative, the extremal behaviour of the aggregate is determined by both\nmarginal shape parameters and the coefficient of asymptotic independence\n(Ledford and Tawn, 1996); if they are both positive or have different signs,\nthe upper-tail behaviour of the aggregate is given solely by the largest\nmarginal shape. We also derive the aggregate upper-tail behaviour for some well\nknown copulae which reveals further insight into the tail structure when the\ncopula falls outside the conditions for the subclasses of the limiting\ndependence representations.\n"}
{"abstract": "  We study the defect groups of $D_p^b(G)$ theories using geometric engineering\nand BPS quivers. In the simple case when $b=h^\\vee (G)$, we use the BPS quivers\nof the theory to see that the defect group is compatible with a known\nMaruyoshi-Song flow. To extend to the case where $b\\neq h^\\vee (G)$, we use a\nsimilar Maruyoshi-Song flow to conjecture that the defect groups of $D_p^b(G)$\ntheories are given by those of $G^{(b)}[k]$ theories. In the cases of $G=A_n,\n\\;E_6, \\;E_8$ we cross check our result by calculating the BPS quivers of the\n$G^{(b)}[k]$ theories and looking at the cokernel of their intersection matrix.\n"}
{"abstract": "  Despite tremendous investigations, a quantum spin liquid state realized in\nspin-1/2 kagome Heisenberg antiferromagnet remains largely elusive. In\nherbertsmithite ZnCu$_3$(OH)$_6$Cl$_2$, a quantum spin liquid candidate on the\nperfect kagome lattice, precisely characterizing the intrinsic physics of the\nkagome layers is extremely challenging due to the presence of interlayer Cu/Zn\nantisite disorder within its crystal structure. Here we measured the specific\nheat and thermal conductivity of single crystal herbertsmithite in magnetic\nfields with high resolution. Our results are highlighted by the excellent\nscaling collapse of the intrinsic magnetic specific heat contribution arising\nfrom the kagome layers as a function of $T/H$ (temperature/magnetic field). In\naddition, no residual linear term in the thermal conductivity\n$\\kappa/T(T\\rightarrow 0)$ is observed in zero and applied magnetic fields,\nindicating the absence of itinerant gapless excitations. These results suggest\na new picture for a quantum spin liquid state of the kagome layers of\nherbertsmithite, wherein localized orphan spins arise and interact with random\nexchanges in conjunction with a non-itinerant quantum spin liquid.\n"}
{"abstract": "  Aromatic organic compounds, because of their small excitation energies ~\nO(few eV) and scintillating properties, are promising targets for detecting\ndark matter of mass ~ O(few MeV). Additionally, their planar molecular\nstructures lead to large anisotropies in the electronic wavefunctions, yielding\na significant daily modulation in the event rate expected to be observed in\ncrystals of these molecules. We characterize the daily modulation rate of dark\nmatter interacting with an anisotropic scintillating organic crystal such as\ntrans-stilbene, and show that daily modulation is an ~ O(1) fraction of the\ntotal rate for small DM masses and comparable to, or larger than, the ~ 10%\nannual modulation fraction at large DM masses. As we discuss in detail, this\nmodulation provides significant leverage for detecting or excluding dark matter\nscattering, even in the presence of a non-negligible background rate. Assuming\na non-modulating background rate of 1/min/kg that scales with total exposure,\nwe find that a 100 kg yr experiment is sensitive to the cross section\ncorresponding to the correct relic density for dark matter masses between\n1.3-14 MeV (1.5-1000 MeV) if dark matter interacts via a heavy (light)\nmediator. This modulation can be understood using an effective velocity scale\nv* = Delta E/q*, where Delta E is the electronic transition energy and q* is a\ncharacteristic momentum scale of the electronic orbitals. We also characterize\npromising future directions for development of scintillating organic crystals\nas dark matter detectors.\n"}
{"abstract": "  In this work, we analyze the performance of a simple majority-rule protocol\nsolving a fundamental coordination problem in distributed systems -\n\\emph{binary majority consensus}, in the presence of probabilistic message\nloss. Using probabilistic analysis for a large scale, fully-connected, network\nof $2n$ agents, we prove that the Simple Majority Protocol (SMP) reaches\nconsensus in only three communication rounds with probability approaching $1$\nas $n$ grows to infinity. Moreover, if the difference between the numbers of\nagents that hold different opinions grows at a rate of $\\sqrt{n}$, then the SMP\nwith only two communication rounds attains consensus on the majority opinion of\nthe network, and if this difference grows faster than $\\sqrt{n}$, then the SMP\nreaches consensus on the majority opinion of the network in a single round,\nwith probability converging to $1$ exponentially fast as $n \\rightarrow\n\\infty$. We also provide some converse results, showing that these requirements\nare not only sufficient, but also necessary.\n"}
{"abstract": "  Sexism has become an increasingly major problem on social networks during the\nlast years. The first shared task on sEXism Identification in Social neTworks\n(EXIST) at IberLEF 2021 is an international competition in the field of Natural\nLanguage Processing (NLP) with the aim to automatically identify sexism in\nsocial media content by applying machine learning methods. Thereby sexism\ndetection is formulated as a coarse (binary) classification problem and a\nfine-grained classification task that distinguishes multiple types of sexist\ncontent (e.g., dominance, stereotyping, and objectification). This paper\npresents the contribution of the AIT_FHSTP team at the EXIST2021 benchmark for\nboth tasks. To solve the tasks we applied two multilingual transformer models,\none based on multilingual BERT and one based on XLM-R. Our approach uses two\ndifferent strategies to adapt the transformers to the detection of sexist\ncontent: first, unsupervised pre-training with additional data and second,\nsupervised fine-tuning with additional and augmented data. For both tasks our\nbest model is XLM-R with unsupervised pre-training on the EXIST data and\nadditional datasets and fine-tuning on the provided dataset. The best run for\nthe binary classification (task 1) achieves a macro F1-score of 0.7752 and\nscores 5th rank in the benchmark; for the multiclass classification (task 2)\nour best submission scores 6th rank with a macro F1-score of 0.5589.\n"}
{"abstract": "  The Advanced LIGO and Virgo gravitational wave observatories have opened a\nnew window with which to study the inspiral and mergers of binary compact\nobjects. These observations are most powerful when coordinated with\nmulti-messenger observations. This was underlined by the first observation of a\nbinary neutron star merger GW170817, coincident with a short Gamma-ray burst,\nGRB170817A, and the identification of the host galaxy NGC~4993 from the optical\ncounterpart AT~2017gfo. Finding the fast-fading optical counterpart critically\ndepends on the rapid production of a sky-map based on LIGO/Virgo data.\nCurrently, a rapid initial sky map is produced followed by a more accurate,\nhigh-latency, $\\gtrsim{12}{\\rm{hr}}$ sky map. We study optimization choices of\nthe Bayesian prior and signal model which can be used alongside other\napproaches such as reduced order quadrature. We find these yield up to a $60\\%$\nreduction in the time required to produce the high-latency localisation for\nbinary neutron star mergers.\n"}
{"abstract": "  We characterize the optimal control for a class of singular stochastic\ncontrol problems as the unique solution to a related Skorokhod reflection\nproblem. The considered optimization problems concern the minimization of a\ndiscounted cost functional over an infinite time-horizon through a process of\nbounded variation affecting an It\\^o-diffusion. The setting is\nmultidimensional, the drift of the state equation and the costs are convex, the\nvolatility matrix can be constant or linear in the state. We prove that the\noptimal control acts only when the underlying diffusion attempts to exit the\nso-called waiting region, and that the direction of this action is prescribed\nby the derivative of the value function. Our approach is based on the study of\na suitable monotonicity property of the derivative of the value function\nthrough its interpretation as the value of an optimal stopping game. Such a\nmonotonicity allows to construct nearly optimal policies which reflect the\nunderlying diffusion at the boundary of approximating waiting regions. The\nlimit of this approximation scheme then provides the desired characterization.\nOur result applies to a relevant class of linear-quadratic models, among\nothers. Furthermore, it allows to construct the optimal control in degenerate\nand non degenerate settings considered in the literature, where this important\naspect was only partially addressed.\n"}
{"abstract": "  The curse of dimensionality is a recognized challenge in nonparametric\nestimation. This paper develops a new L0-norm regularization approach to the\nconvex quantile and expectile regressions for subset variable selection. We\nshow how to use mixed integer programming to solve the proposed L0-norm\nregularization approach in practice and build a link to the commonly used\nL1-norm regularization approach. A Monte Carlo study is performed to compare\nthe finite sample performances of the proposed L0-penalized convex quantile and\nexpectile regression approaches with the L1-norm regularization approaches. The\nproposed approach is further applied to benchmark the sustainable development\nperformance of the OECD countries and empirically analyze the accuracy in the\ndimensionality reduction of variables. The results from the simulation and\napplication illustrate that the proposed L0-norm regularization approach can\nmore effectively address the curse of dimensionality than the L1-norm\nregularization approach in multidimensional spaces.\n"}
{"abstract": "  In recent work, we proposed a distributed Picard iteration (DPI) that allows\na set of agents, linked by a communication network, to find a fixed point of a\nlocally contractive (LC) map that is the average of individual maps held by\nsaid agents. In this work, we build upon the DPI and its local linear\nconvergence (LLC) guarantees to make several contributions. We show that\nSanger's algorithm for principal component analysis (PCA) corresponds to the\niteration of an LC map that can be written as the average of local maps, each\nmap known to each agent holding a subset of the data. Similarly, we show that a\nvariant of the expectation-maximization (EM) algorithm for parameter estimation\nfrom noisy and faulty measurements in a sensor network can be written as the\niteration of an LC map that is the average of local maps, each available at\njust one node. Consequently, via the DPI, we derive two distributed algorithms\n- distributed EM and distributed PCA - whose LLC guarantees follow from those\nthat we proved for the DPI. The verification of the LC condition for EM is\nchallenging, as the underlying operator depends on random samples, thus the LC\ncondition is of probabilistic nature.\n"}
{"abstract": "  Diffusion MRI measures of the human brain provide key insight into\nmicrostructural variations across individuals and into the impact of central\nnervous system diseases and disorders. One approach to extract information from\ndiffusion signals has been to use biologically relevant analytical models to\nlink millimetre scale diffusion MRI measures with microscale influences. The\nother approach has been to represent diffusion as an anomalous process and\ninfer information from the different anomalous diffusion equation parameters.\nHow parameters of various anomalous diffusion models vary with age in the human\nbrain white matter has not been investigated. We perform the tests within a\ngeneral continuous time random walk framework, in which temporal jump lengths\nand structural barriers to diffusion are considered. We outline the\nformulations for the super-diffusion (i.e., stretched exponential),\nsub-diffusion, quasi-diffusion and fractional Bloch-Torrey models with respect\nto the continuous time random walk expression. We also provide a direct link\nbetween diffusion kurtosis and the sub-diffusion model and demonstrate\nmathematical equivalence between parameters. 7T diffusion weighted MRI data in\nthe age range 19-67 was used in this study, particularly focusing on the corpus\ncallosum. Anomalous diffusion model parameters were found to show trends with\naging, and interestingly, our investigations led to a new robust technique for\ngenerating maps of diffusion kurtosis.\n"}
{"abstract": "  The evidence-based approach has increasingly been employed to synthesize\nempirical findings from the primary research in software engineering.\nNevertheless, the reproducibility of evidence-based software engineering (EBSE)\nstudies seems to be underemphasized. In our investigation into the automatic\nsearch of 311 sample studies, more than 50% of the search strings are not\nreusable; about 87.5% of the search activities (e.g., search field settings)\nare unrepeatable; and more than 95% of the whole automatic search\nimplementations are unreproducible. Considering that searching is a cornerstone\nof an EBSE study, we are afraid that the reproducibility of the current\nsecondary research could be worse than we can imagine. By analyzing and\nreporting the root causes of the aforementioned observations, we urge\ncollaboration and cooperation among all the stakeholders in our community to\nimprove the research reproducibility in EBSE.\n"}
{"abstract": "  For which graphs $F$ is there a sparse $F$-counting lemma in $C_4$-free\ngraphs? We are interested in identifying graphs $F$ with the property that,\nroughly speaking, if $G$ is an $n$-vertex $C_4$-free graph with on the order of\n$n^{3/2}$ edges, then the density of $F$ in $G$, after a suitable\nnormalization, is approximately at least the density of $F$ in an\n$\\epsilon$-regular approximation of $G$. In recent work, motivated by\napplications in extremal and additive combinatorics, we showed that $C_5$ has\nthis property. Here we construct a family of graphs with the property.\n"}
{"abstract": "  We consider a one-dimensional stationary stochastic process $x(\\tau)$ of\nduration $T$. We study the probability density function (PDF) $P(t_{\\rm m}|T)$\nof the time $t_{\\rm m}$ at which $x(\\tau)$ reaches its global maximum. By using\na path integral method, we compute $P(t_{\\rm m}|T)$ for a number of equilibrium\nand nonequilibrium stationary processes, including the Ornstein-Uhlenbeck\nprocess, Brownian motion with stochastic resetting and a single confined\nrun-and-tumble particle. For a large class of equilibrium stationary processes\nthat correspond to diffusion in a confining potential, we show that the scaled\ndistribution $P(t_{\\rm m}|T)$, for large $T$, has a universal form (independent\nof the details of the potential). This universal distribution is uniform in the\n``bulk'', i.e., for $0 \\ll t_{\\rm m} \\ll T$ and has a nontrivial edge scaling\nbehavior for $t_{\\rm m} \\to 0$ (and when $t_{\\rm m} \\to T$), that we compute\nexactly. Moreover, we show that for any equilibrium process the PDF $P(t_{\\rm\nm}|T)$ is symmetric around $t_{\\rm m}=T/2$, i.e., $P(t_{\\rm m}|T)=P(T-t_{\\rm\nm}|T)$. This symmetry provides a simple method to decide whether a given\nstationary time series $x(\\tau)$ is at equilibrium or not.\n"}
{"abstract": "  The article examines Nikolskii and Besov spaces with norms defined using\n\"$L_p$-averaged\" mixed moduli of continuity of functions of appropriate orders,\ninstead of mixed moduli of continuity of known orders for certain mixed\nderivative functions. For such spaces of functions defined in bounded domains\nof $m$-kind (in a broad sense), the author builds continuous linear mappings of\nthose spaces to ordinary Nikolskii and Besov spaces of mixed smoothness in\n$\\mathbb{R}^d$, that are function extension operators, thus incurring\ncoincidence of both categories of spaces defined in the said domains.\n"}
{"abstract": "  We consider the downlink of a cell-free massive multiple-input\nmultiple-output (MIMO) system with \\textcolor{red}{single}-antenna access\npoints (APs) and single-antenna users. An iterative robust minimum mean-square\nerror (RMMSE) precoder based on generalized loading is developed to mitigate\ninterference in the presence of imperfect channel state information (CSI). An\nachievable rate analysis is carried out and optimal and uniform power\nallocation schemes are developed based on the signal-to-interference-plus-noise\nratio. An analysis of the computational costs of the proposed RMMSE and\nexisting schemes is also presented. Numerical results show the improvement\nprovided by the proposed RMMSE precoder against linear minimum mean-square\nerror, zero-forcing and conjugate beamforming precoders in the presence of\nimperfect CSI.\n"}
{"abstract": "  PQ-trees and PC-trees are data structures that represent sets of linear and\ncircular orders, respectively, subject to constraints that specific subsets of\nelements have to be consecutive. While equivalent to each other, PC-trees are\nconceptually much simpler than PQ-trees; updating a PC-trees so that a set of\nelements becomes consecutive requires only a single operation, whereas PQ-trees\nuse an update procedure that is described in terms of nine transformation\ntemplates that have to be recursively matched and applied.\n  Despite these theoretical advantages, to date no practical PC-tree\nimplementation is available. This might be due to the original description by\nHsu and McConnell in some places only sketching the details of the\nimplementation. In this paper, we describe two alternative implementations of\nPC-trees. For the first one, we follow the approach by Hsu and McConnell,\nfilling in the necessary details and also proposing improvements on the\noriginal algorithm. For the second one, we use a different technique for\nefficiently representing the tree using a Union-Find data structure. In an\nextensive experimental evaluation we compare our implementations to a variety\nof other implementations of PQ-trees that are available on the web as part of\nacademic and other software libraries. Our results show that both PC-tree\nimplementations beat their closest fully correct competitor, the PQ-tree\nimplementation from the OGDF library, by a factor of 2 to 4, showing that\nPC-trees are not only conceptually simpler but also fast in practice. Moreover,\nwe find the Union-Find-based implementation, while having a slightly worse\nasymptotic runtime, to be twice as fast as the one based on the description by\nHsu and McConnell.\n"}
{"abstract": "  We consider a multi-dimensional continuum Schr\\\"odinger operator $H$ which is\ngiven by a perturbation of the negative Laplacian by a compactly supported\nbounded potential. We show that, for a fairly large class of test functions,\nthe second-order Szeg\\H{o}-type asymptotics for the spatially truncated Fermi\nprojection of $H$ is independent of the potential and, thus, identical to the\nknown asymptotics of the Laplacian.\n"}
{"abstract": "  The Bogoliubov energy functional proposed recently by Napi\\'orkowski, Reuvers\nand Solovej is revisited. We offer a direct proof of the existence of\nminimizers at zero temperature, which covers a significantly larger class of\ninteraction potentials. The ideas used in this proof also imply that in any\nground state, more than half of the particles are inside the Bose-Einstein\ncondensate.\n"}
{"abstract": "  Deep learning has become the gold standard for image processing over the past\ndecade. Simultaneously, we have seen growing interest in orbital activities\nsuch as satellite servicing and debris removal that depend on proximity\noperations between spacecraft. However, two key challenges currently pose a\nmajor barrier to the use of deep learning for vision-based on-orbit proximity\noperations. Firstly, efficient implementation of these techniques relies on an\neffective system for model development that streamlines data curation,\ntraining, and evaluation. Secondly, a scarcity of labeled training data (images\nof a target spacecraft) hinders creation of robust deep learning models. This\npaper presents an open-source deep learning pipeline, developed specifically\nfor on-orbit visual navigation applications, that addresses these challenges.\nThe core of our work consists of two custom software tools built on top of a\ncloud architecture that interconnects all stages of the model development\nprocess. The first tool leverages Blender, an open-source 3D graphics toolset,\nto generate labeled synthetic training data with configurable model poses\n(positions and orientations), lighting conditions, backgrounds, and commonly\nobserved in-space image aberrations. The second tool is a plugin-based\nframework for effective dataset curation and model training; it provides common\nfunctionality like metadata generation and remote storage access to all\nprojects while giving complete independence to project-specific code.\nTime-consuming, graphics-intensive processes such as synthetic image generation\nand model training run on cloud-based computational resources which scale to\nany scope and budget and allow development of even the largest datasets and\nmodels from any machine. The presented system has been used in the Texas\nSpacecraft Laboratory with marked benefits in development speed and quality.\n"}
{"abstract": "  The coherent population trapping (CPT) atomic clock is very promising for use\nin next-generation spaceborne applications owing to its compactness and high\nperformance. In this paper, we propose and implement a CPT atomic clock based\non the direct modulation of a large-modulation-bandwidth and narrow-linewidth\ndistributed Bragg reflector laser, which replaces the usually used external\nbulk modulator in the high-performance CPT clock. Our method retains the high\nperformance while significantly reducing the size. Using this highly compact\nbichromatic light source and simplest CPT configuration, in which a circularly\npolarized bichromatic laser interrogates the ^{87}Rb atom system, a CPT signal\nof clock transition with a narrow linewidth and high contrast is observed. We\nthen lock the local oscillator frequency to the CPT error signal and\ndemonstrate a short-term frequency stability of 3.6 \\times 10^{-13}\n{\\tau}^{-1/2} (4 s \\le {\\tau} \\le 200 s). We attribute it to the ultralow laser\nfrequency and intensity noise as well as to the high-quality-factor CPT signal.\nThis study can pave the way for the development of compact high-performance CPT\nclocks based on our scheme.\n"}
{"abstract": "  We consider capacity (fuzzy measure, non-additive probability) on a compactum\nas a monotone cooperative normed game. Then it is naturally to consider\nprobability measures as elements of core of such game. We prove an analogue of\nBondareva-Shapley theorem that non-emptiness of the core is equivalent to\nbalancedness of the capacity. We investigate categorical properties of balanced\ncapacities and give characterizations of some fuzzy integrals of balanced\ncapacities.\n"}
{"abstract": "  Graph Neural Networks (GNNs) have garnered a lot of recent interest because\nof their success in learning representations from graph-structured data across\nseveral critical applications in cloud and HPC. Owing to their unique compute\nand memory characteristics that come from an interplay between dense and sparse\nphases of computations, the emergence of reconfigurable dataflow (aka spatial)\naccelerators offers promise for acceleration by mapping optimized dataflows\n(i.e., computation order and parallelism) for both phases. The goal of this\nwork is to characterize and understand the design-space of dataflow choices for\nrunning GNNs on spatial accelerators in order for the compilers to optimize the\ndataflow based on the workload. Specifically, we propose a taxonomy to describe\nall possible choices for mapping the dense and sparse phases of GNNs spatially\nand temporally over a spatial accelerator, capturing both the intra-phase\ndataflow and the inter-phase (pipelined) dataflow. Using this taxonomy, we do\ndeep-dives into the cost and benefits of several dataflows and perform case\nstudies on implications of hardware parameters for dataflows and value of\nflexibility to support pipelined execution.\n"}
{"abstract": "  The interaction between a central qubit spin and a surrounding bath of spins\nis critical to spin-based solid state quantum sensing and quantum information\nprocessing. Spin-bath interactions are typically strongly anisotropic, and\nrapid physical rotation has long been used in solid-state nuclear magnetic\nresonance to simulate motional averaging of anisotropic interactions, such as\ndipolar coupling between nuclear spins. Here, we show that the interaction\nbetween electron spins of nitrogen-vacancy centers and a bath of $^{13}$C\nnuclear spins in a diamond rotated at up to 300,000rpm introduces decoherence\ninto the system via frequency-modulation of the nuclear spin Larmor precession.\nThe presence of an off-axis magnetic field necessary for averaging of the\ndipolar coupling leads to a rotational dependence of the electron-nuclear\nhyperfine interaction, which cannot be averaged out with experimentally\nachievable rotation speeds. Our findings offer new insights into the use of\nphysical rotation for quantum control with implications for quantum systems\nhaving motional and rotational degrees of freedom that are not fixed.\n"}
{"abstract": "  Bayesian inference is used to extract unknown parameters from gravitational\nwave signals. Detector noise is typically modelled as stationary, although data\nfrom the LIGO and Virgo detectors is not stationary. We demonstrate that the\nposterior of estimated waveform parameters is no longer valid under the\nassumption of stationarity. We show that while the posterior is unbiased, the\nerrors will be under- or overestimated compared to the true posterior. A\nformalism was developed to measure the effect of the mismodelling, and found\nthe effect of any form of non-stationarity has an effect on the results, but\nare not significant in certain circumstances. We demonstrate the effect of\nshort-duration Gaussian noise bursts and persistent oscillatory modulation of\nthe noise on binary-black-hole-like signals. In the case of short signals,\nnon-stationarity in the data does not have a large effect on the parameter\nestimation, but the errors from non-stationary data containing signals lasting\ntens of seconds or longer will be several times worse than if the noise was\nstationary. Accounting for this limiting factor in parameter sensitivity could\nbe very important for achieving accurate astronomical results, including an\nestimation of the Hubble parameter. This methodology for handling the\nnon-stationarity will also be invaluable for analysis of waveforms that last\nminutes or longer, such as those we expect to see with the Einstein Telescope.\n"}
{"abstract": "  Music is a fundamental human construct, and harmony provides the building\nblocks of musical language. Using the Kunstderfuge corpus of classical music,\nwe analyze the historical evolution of the richness of harmonic vocabulary of\n76 classical composers, covering almost 6 centuries. Such corpus comprises\nabout 9500 pieces, resulting in more than 5 million tokens of music codewords.\nThe fulfilment of Heaps' law for the relation between the size of the harmonic\nvocabulary of a composer (in codeword types) and the total length of his works\n(in codeword tokens), with an exponent around 0.35, allows us to define a\nrelative measure of vocabulary richness that has a transparent interpretation.\nWhen coupled with the considered corpus, this measure allows us to quantify\nharmony richness across centuries, unveiling a clear increasing linear trend.\nIn this way, we are able to rank the composers in terms of richness of\nvocabulary, in the same way as for other related metrics, such as entropy. We\nfind that the latter is particularly highly correlated with our measure of\nrichness. Our approach is not specific for music and can be applied to other\nsystems built by tokens of different types, as for instance natural language.\n"}
{"abstract": "  In this work, we present, analyze, and implement a class of Multi-Level\nMarkov chain Monte Carlo (ML-MCMC) algorithms based on independent\nMetropolis-Hastings proposals for Bayesian inverse problems. In this context,\nthe likelihood function involves solving a complex differential model, which is\nthen approximated on a sequence of increasingly accurate discretizations. The\nkey point of this algorithm is to construct highly coupled Markov chains\ntogether with the standard Multi-level Monte Carlo argument to obtain a better\ncost-tolerance complexity than a single-level MCMC algorithm. Our method\nextends the ideas of Dodwell, et al. \"A hierarchical multilevel Markov chain\nMonte Carlo algorithm with applications to uncertainty quantification in\nsubsurface flow,\" \\textit{SIAM/ASA Journal on Uncertainty Quantification 3.1\n(2015): 1075-1108,} to a wider range of proposal distributions. We present a\nthorough convergence analysis of the ML-MCMC method proposed, and show, in\nparticular, that (i) under some mild conditions on the (independent) proposals\nand the family of posteriors, there exists a unique invariant probability\nmeasure for the coupled chains generated by our method, and (ii) that such\ncoupled chains are uniformly ergodic. We also generalize the cost-tolerance\ntheorem of Dodwell et al., to our wider class of ML-MCMC algorithms. Finally,\nwe propose a self-tuning continuation-type ML-MCMC algorithm (C-ML-MCMC). The\npresented method is tested on an array of academic examples, where some of our\ntheoretical results are numerically verified. These numerical experiments\nevidence how our extended ML-MCMC method is robust when targeting some\n\\emph{pathological} posteriors, for which some of the previously proposed\nML-MCMC algorithms fail.\n"}
{"abstract": "  A method to design gratings in integrated photonics, is presented. The method\nis based on a transfer matrix formalism enhanced by Finite Element Method (FEM)\nparameter calculations. The main advantages of the proposed technique are the\neasy of use, the fast optimization time and the versatility of the approach.\nFew examples of optimized gratings to obtain various scattered light field\nprofiles for different applications are presented: a double-Gaussian profile, a\nflat top square profile, a spot profile on a chip surface, profiles suited to\nget efficient and selective coupling to single mode and multimode fibers. A\ndiscussion of the limits of the method and some insights on how to improve it\nare also discussed.\n"}
{"abstract": "  We find that spinning particles with suitable couplings propagating in\ncertain supersymmetric backgrounds of $D=4$, $N=2$ and $D=5$, $N=1$ minimal\nsupergravities are invariant under symmetries generated by the twisted\ncovariant form hierarchies of these theories. We also compare our results with\nthe symmetries of spinning particles generated by Killing-Yano forms which are\nresponsible for the separability properties of some gravitational backgrounds.\n"}
{"abstract": "  The slope gap distribution of a translation surface is a measure of how\nrandom the directions of the saddle connections on the surface are. It is known\nthat Veech surfaces, a highly symmetric type of translation surface, have gap\ndistributions that are piecewise real analytic. Beyond that, however, very\nlittle is currently known about the general behavior of the slope gap\ndistribution, including the number of points of non-analyticity or the tail.\n  We show that the limiting gap distribution of slopes of saddle connections on\na Veech translation surface is always piecewise real-analytic with\n\\emph{finitely} many points of non-analyticity. We do so by taking an explicit\nparameterization of a Poincar\\'{e} section to the horocycle flow on\n$\\text{SL}(2,\\mathbb{R})/\\text{SL}(X,\\omega)$ associated to an arbitrary Veech\nsurface $\\text{SL}(X,\\omega)$ and establishing a key finiteness result for the\nfirst return map under this flow. We use the finiteness result to show that the\ntail of the slope gap distribution of Veech surfaces always has quadratic\ndecay.\n"}
{"abstract": "  Early detection of skin cancers like melanoma is crucial to ensure high\nchances of survival for patients. Clinical application of Deep Learning\n(DL)-based Decision Support Systems (DSS) for skin cancer screening has the\npotential to improve the quality of patient care. The majority of work in the\nmedical AI community focuses on a diagnosis setting that is mainly relevant for\nautonomous operation. Practical decision support should, however, go beyond\nplain diagnosis and provide explanations. This paper provides an overview of\nworks towards explainable, DL-based decision support in medical applications\nwith the example of skin cancer diagnosis from clinical, dermoscopic and\nhistopathologic images. Analysis reveals that comparably little attention is\npayed to the explanation of histopathologic skin images and that current work\nis dominated by visual relevance maps as well as dermoscopic feature\nidentification. We conclude that future work should focus on meeting the\nstakeholder's cognitive concepts, providing exhaustive explanations that\ncombine global and local approaches and leverage diverse modalities. Moreover,\nthe possibility to intervene and guide models in case of misbehaviour is\nidentified as a major step towards successful deployment of AI as DL-based DSS\nand beyond.\n"}
{"abstract": "  We forecast the ability of future-generation experiments to detect the\nfine-structure lines of the carbon and oxygen ions, [CII] and [OIII] in\nintensity mapping (IM) from the Epoch of Reionization ($z \\sim 6-8$). Combining\nthe latest empirically derived constraints relating the luminosity of the\n[OIII] line to the ambient star-formation rate, and using them in conjunction\nwith previously derived estimates for the abundance of [CII] in haloes, we\npredict the expected auto-correlation IM signal to be observed using\nnext-generation facilities based on the Fred Young Submillimetre Telescope\n(FYST) and the balloon-borne facility, Experiment for Cryogenic Large-Aperture\nIntensity Mapping (EXCLAIM) over $z \\sim 5.3 - 7$. We describe how improvements\nto both the ground-based and balloon-based surveys in the future will enable a\ncross-correlation signal to be detected at $\\sim$ 10-40 $\\sigma$ over $z \\sim\n5.3 - 7$. Finally, we propose a space-based mission targeting the [OIII] 88 and\n52 $\\mu$m lines along with the [CII] 158 $\\mu$m line, configured to enhance the\nsignal-to-noise ratio of cross-correlation measurements. We find that such a\nconfiguration can achieve a high-significance detection (hundreds to thousands\nof $\\sigma$) in both auto- and cross-correlation modes.\n"}
{"abstract": "  This paper gives a new approach for the maximum likelihood estimation of the\njoint of the location and scale of the Cauchy distribution. We regard the joint\nas a single complex parameter and derive a new form of the likelihood equation\nof a complex variable. Based on the equation, we provide a new iterative scheme\napproximating the maximum likelihood estimate. We also handle the equation in\nan algebraic manner and derive a polynomial containing the maximum likelihood\nestimate as a root. This algebraic approach provides another scheme\napproximating the maximum likelihood estimate by root-finding algorithms for\npolynomials, and furthermore, gives non-existence of closed-form formulae for\nthe case that the sample size is five. We finally provide some numerical\nexamples to show our method is effective.\n"}
{"abstract": "  Freeform optical components offer significant compactization of multi-lens\nsystems, as well as advanced manipulation of light that is not possible with\ntraditional systems. However, their fabrication relies on machining processes\nthat are complex, time-consuming, and incompatible with rapid prototyping. This\nwork presents the ability to shape liquid volumes and solidify them into\ndesired freeform components, enabling rapid freeform prototyping with high\nsurface quality. The method is based on controlling the minimum energy state of\nthe interface between a curable optical liquid and an immersion liquid, by\ndictating a geometrical boundary constraint. The boundary shape is modeled as a\ncylinder whose arbitrary height is expressed as a Fourier series, allowing for\nan analytical solution of the resulting freeform surface as a sum of\nFourier-Bessel functions. Each of these functions represents a different basic\nmode, whose superposition creates complex topographies. This solution allows\ndeterministic design of freeform surfaces by controlling three key parameters -\nthe volume of the optical liquid, the density of the immersion liquid, and the\nshape of the bounding frame. The paper describes a complete workflow for rapid\nprototyping of such components, and demonstrates the fabrication of a 35 mm\ndiameter freeform component with sub-nanometer surface roughness within\nminutes.\n"}
{"abstract": "  We study fixed loci of antisymplectic involutions on projective hyperk\\\"ahler\nmanifolds. When the involution is induced by an ample class of square 2 in the\nBeauville-Bogomolov-Fujiki lattice, we show that the number of connected\ncomponents of the fixed locus is equal to the divisibility of the class, which\nis either 1 or 2.\n"}
{"abstract": "  We study the function series $\\sum_{n=1}^\\infty \\phi^{2m+2}\n\\text{cosch}^{2m+2}(n\\phi/2)$, and similar series, for integers $m$ and complex\n$\\phi$. This hyperbolic series is linearly related to the Lambert series. The\nLambert series is known to satisfy a functional equation which defines the\nRamanujan polynomials. By using residue theorem (summation theorem) we find the\nfunctional equation satisfied by this hyperbolic series. The functional\nequation identifies a class of polynomials which can be seen as a\ngeneralization of the Ramanujan polynomials. These polynomials coincide with\nthe asymptotic expansion of the hyperbolic series at the origin and they all\nvanish for $\\phi=\\pm 2\\pi i$. We furthermore derive several identities between\nHarmonic numbers and ordinary and generalized Bernoulli polynomials.\n"}
{"abstract": "  The multimessenger event GW170817/GRB 170817A confirmed that binary neutron\nstar (BNS) mergers can produce short gamma-ray burst (SGRB) jets. This evidence\npromoted new investigations on the mechanisms through which a BNS merger\nremnant can launch such a powerful relativistic outflow and on the propagation\nof the latter across the surrounding post-merger environment. In particular,\ngreat strides have been made in jet propagation models, establishing\nconnections between the initial jet launching conditions, including the\nincipient jet launching time (with respect to merger) and the injection\nparameters, and the observable SGRB prompt and afterglow emission. However,\npresent semi-analytical models and numerical simulations (with one notable\nexception) adopt simple hand-made prescriptions to account for the post-merger\nenvironment, lacking a direct association with any specific merging BNS system.\nHere, we present the first three-dimensional relativistic hydrodynamics\nsimulations of incipient SGRB jets propagating through a post-merger\nenvironment that is directly imported from the outcome of a previous general\nrelativistic BNS merger simulation. Our results show that the evolution and\nfinal properties of the jet can be largely affected by the anisotropies and the\ndeviations from axisymmetry and homologous expansion characterizing more\nrealistic BNS merger environments. In addition, we find that the inclusion of\nthe gravitational pull from the central compact object, often overlooked, can\nhave a major impact. Finally, we consider different jet launching times\nreferred to the same BNS merger model and discuss the consequences for the\nultimate jet properties.\n"}
{"abstract": "  Experimental results are presented on the efficiency limits for a quantum\ninterface between a matter-based qubit and a photonic qubit. Using a trapped\nion in an optical cavity, we obtain a single ion-entangled photon at the cavity\noutput with a probability of 0.69(3). The performance of our system is shown to\nsaturate the upper limit to photon-collection probability from a quantum\nemitter in a cavity, set by the emitter's electronic structure and by the\ncavity parameters. The probability for generating and detecting the\nion-entangled fiber-coupled photon is 0.462(3), a five-fold increase over the\nprevious best performance. Finally, the generation and detection of up to 15\nsequential polarised photons demonstrates the ability of a trapped ion to serve\nas a multi-photon source. The comparison between measured probabilities and\npredicted bounds is relevant for quantum emitters beyond trapped ions, in\nparticular, for the design of future systems optimising photon collection from,\nand absorption in, quantum matter.\n"}
{"abstract": "  In this article, we explore the relationship between cellular phone data and\nhousing prices in Budapest, Hungary. We determine mobility indicators from one\nmonths of Call Detail Records (CDR) data, while the property price data are\nused to characterize the socioeconomic status at the Capital of Hungary. First,\nwe validated the proposed methodology by comparing the Home and Work locations\nestimation and the commuting patterns derived from the cellular network dataset\nwith reports of the national mini census. We investigated the statistical\nrelationships between mobile phone indicators, such as Radius of Gyration, the\ndistance between Home and Work locations or the Entropy of visited cells, and\nmeasures of economic status based on housing prices. Our findings show that the\nmobility correlates significantly with the socioeconomic status. We performed\nPrincipal Component Analysis (PCA) on combined vectors of mobility indicators\nin order to characterize the dependence of mobility habits on socioeconomic\nstatus. The results of the PCA investigation showed remarkable correlation of\nhousing prices and mobility customs.\n"}
{"abstract": "  The independence polynomial originates in statistical physics as the\npartition function of the hard-core model. The location of the complex zeros of\nthe polynomial is related to phase transitions, and plays an important role in\nthe design of efficient algorithms to approximately compute evaluations of the\npolynomial.\n  In this paper we directly relate the location of the complex zeros of the\nindependence polynomial to computational hardness of approximating evaluations\nof the independence polynomial. We do this by moreover relating the location of\nzeros to chaotic behaviour of a naturally associated family of rational\nfunctions; the occupation ratios.\n"}
{"abstract": "  We develop a new equation of state (EoS) table involving thermal (anti)kaons,\nBose-Einstein condensate of $K^{-}$ mesons and $\\Lambda$-hyperons for\ncore-collapse supernova and neutron star merger simulations. This EoS table is\nbased on a finite temperature density-dependent relativistic hadron field\ntheory where baryon-baryon interaction is mediated by scalar $\\sigma$, vector\n$\\omega$ and $\\rho$ mesons, using the parameter set DD2 for nucleons. The\nrepulsive hyperon-hyperon interaction is mediated by an additional strange\n$\\phi$ meson. The EoS for the $K^-$ condensed matter is also calculated within\nthe framework of relativistic mean field model, whereas the low-density,\ninhomogeneous matter is calculated in the extended Nuclear Statistical\nEquilibrium model (NSE). The EoS table is generated for a wide range of values\nof three parameters - baryon density ($10^{-12}$ to $\\sim$ 1 fm$^{-3}$),\npositive charge fraction(0.01 to 0.60) and temperature(0.1 to 158.48 MeV).\n"}
{"abstract": "  In this paper, we propose a pseudo polynomial size LP formulation for finding\na payoff vector in the least core of a weighted voting game. The numbers of\nvariables and constraints in our formulation are both bounded by $\\mbox{O}(n\nW_+)$, where $n$ is the number of players and $W_+$ is the total sum of\n(integer) voting weights. When we employ our formulation, a commercial LP\nsolver calculates a payoff vector in the least core of practical weighted\nvoting games in a few seconds. We also extend our approach to vector weighted\nvoting games.\n"}
{"abstract": "  The motion of a vertically positioned bicycle is considered when a horizontal\ncontrol force, which maybe both internal and external in relation to the\nbicycle, is applied to its pedal. Tangential forces of dryfriction obeying the\nEuler { Coulomb law act at points of contact of the wheels with the horizontal\nsupportplane. The constraint at the points of contact of the wheels with the\nsupport is assumed to be unilateral.The problem of determining the acceleration\nof the centre of mass of the bicycle and the realized motionsof its wheels\n(with or without slip, and with or without detachment) with dierent values of\nthe design parameters and control force is solved. This approach is usefull for\ncontrol of vehicle and optimization of its dynamical parameters. Cases of\nnon-uniqueness of motion the Painleve paradox are found.\n"}
{"abstract": "  Attention is an effective mechanism to improve the deep model capability.\nSqueeze-and-Excite (SE) introduces a light-weight attention branch to enhance\nthe network's representational power. The attention branch is gated using the\nSigmoid function and multiplied by the feature map's trunk branch. It is too\nsensitive to coordinate and balance the trunk and attention branches'\ncontributions. To control the attention branch's influence, we propose a new\nattention method, called Shift-and-Balance (SB). Different from\nSqueeze-and-Excite, the attention branch is regulated by the learned control\nfactor to control the balance, then added into the feature map's trunk branch.\nExperiments show that Shift-and-Balance attention significantly improves the\naccuracy compared to Squeeze-and-Excite when applied in more layers, increasing\nmore size and capacity of a network. Moreover, Shift-and-Balance attention\nachieves better or close accuracy compared to the state-of-art Dynamic\nConvolution.\n"}
{"abstract": "  Memory-efficient continuous Sign Language Translation is a significant\nchallenge for the development of assisted technologies with real-time\napplicability for the deaf. In this work, we introduce a paradigm of designing\nrecurrent deep networks whereby the output of the recurrent layer is derived\nfrom appropriate arguments from nonparametric statistics. A novel variational\nBayesian sequence-to-sequence network architecture is proposed that consists of\na) a full Gaussian posterior distribution for data-driven memory compression\nand b) a nonparametric Indian Buffet Process prior for regularization applied\non the Gated Recurrent Unit non-gate weights. We dub our approach\nStick-Breaking Recurrent network and show that it can achieve a substantial\nweight compression without diminishing modeling performance.\n"}
{"abstract": "  The last 15 years have witnessed important progresses in our understanding of\nthe mechanism of superconductivity in the high-$T_{c}$ cuprates. There is now\nstrong evidence that the strange metal behavior is induced by the quantum\ncritical fluctuation at the pseudogap end point, where the Fermi surface\nchanges its topology from hole-like to electron-like. However, experiments show\nthat the quantum critical behavior in the high-$T_{c}$ cuprates is\nqualitatively different from that observed in the heavy Fermion systems and the\niron-based superconductors, in both of which the quantum critical behavior can\nbe attributed to the quantum phase transition toward a symmetry breaking phase.\nThe fact that the pseudogap exists as a spectral gap without a corresponding\nsymmetry breaking order, together with the fact that the strange metal behavior\noccurs as a quantum critical behavior without a corresponding symmetry breaking\nphase transition, exposes the central difficulty of the field: the lack of a\nuniversal low energy effective theory description of the high-$T_{c}$\nphenomenology beyond the Landau paradigm. Recent experiments imply that the\ndualism between the local moment and the itinerant quasiparticle character of\nthe electron in the high-$T_{c}$ cuprates may serve as an organizing principle\nto go beyond the Landau paradigm and may hold the key to the mystery of the\npseudogap phenomena and the strange metal behavior.\n"}
{"abstract": "  We ask the question of how small a self-assembling set of tiles can be yet\nhave interesting computational behaviour. We study this question in a model\nwhere supporting walls are provided as an input structure for tiles to grow\nalong: we call it the Maze-Walking Tile Assembly Model. The model has a number\nof implementation prospects, one being DNA strands that attach to a DNA origami\nsubstrate. Intuitively, the model suggests a separation of signal routing and\ncomputation: the input structure (maze) supplies a routing diagram, and the\nprogrammer's tile set provides the computational ability. We ask how simple the\ncomputational part can be.\n  We give two tiny tile sets that are computationally universal in the\nMaze-Walking Tile Assembly Model. The first has four tiles and simulates\nBoolean circuits by directly implementing NAND, NXOR and NOT gates. Our second\ntile set has 6 tiles and is called the Collatz tile set as it produces patterns\nfound in binary/ternary representations of iterations of the Collatz function.\nUsing computer search we find that the Collatz tile set is expressive enough to\nencode Boolean circuits using blocks of these patterns. These two tile sets\ngive two different methods to find simple universal tile sets, and provide\nmotivation for using pre-assembled maze structures as circuit wiring diagrams\nin molecular self-assembly based computing.\n"}
{"abstract": "  In this paper, we develop numerical methods based on the weighted Birkhoff\naverage for studying two-dimensional invariant tori for volume-preserving maps.\nThe methods do not rely on symmetries, such as time-reversal symmetry, nor on\napproximating tori by periodic orbits. The rate of convergence of the average\ngives a sharp distinction between chaotic and regular dynamics and allows\naccurate computation of rotation vectors for regular orbits. Resonant and\nrotational tori are distinguished by computing the resonance order of the\nrotation vector to a given precision. Critical parameter values, where tori are\ndestroyed, are computed by a sharp decrease in convergence rate of the Birkhoff\naverage. We apply these methods for a three-dimensional generalization of\nChirikov's standard map: an angle-action map with two angle variables.\nComputations on grids in frequency and perturbation amplitude allow estimates\nof the critical set. We also use continuation to follow tori with fixed\nrotation vectors. We test three conjectures for cubic fields that have been\nproposed to give locally robust invariant tori.\n"}
{"abstract": "  Statutory reasoning is the task of determining whether a legal statute,\nstated in natural language, applies to the text description of a case. Prior\nwork introduced a resource that approached statutory reasoning as a monolithic\ntextual entailment problem, with neural baselines performing nearly at-chance.\nTo address this challenge, we decompose statutory reasoning into four types of\nlanguage-understanding challenge problems, through the introduction of concepts\nand structure found in Prolog programs. Augmenting an existing benchmark, we\nprovide annotations for the four tasks, and baselines for three of them. Models\nfor statutory reasoning are shown to benefit from the additional structure,\nimproving on prior baselines. Further, the decomposition into subtasks\nfacilitates finer-grained model diagnostics and clearer incremental progress.\n"}
{"abstract": "  In the past few years, mobile deep-learning deployment progressed by leaps\nand bounds, but solutions still struggle to accommodate its severe and\nfluctuating operational restrictions, which include bandwidth, latency,\ncomputation, and energy. In this work, we help to bridge that gap, introducing\nthe first configurable solution for object detection that manages the triple\ncommunication-computation-accuracy trade-off with a single set of weights. Our\nsolution shows state-of-the-art results on COCO-2017, adding only a minor\npenalty on the base EfficientDet-D2 architecture. Our design is robust to the\nchoice of base architecture and compressor and should adapt well for future\narchitectures.\n"}
{"abstract": "  We construct algebra homomorphisms from affine Yangians to the current\nalgebras of rectangular $W$-algebras both in type A. The construction is given\nvia the coproduct and the evaluation map for the affine Yangians. As a\nconsequence, we show that parabolic inductions for representations of the\nrectangular $W$-algebras can be regarded as tensor product representations of\nthe affine Yangians under the homomorphisms. The same method is applicable also\nto the super setting.\n"}
{"abstract": "  Energy harvesting offers an attractive and promising mechanism to power\nlow-energy devices. However, it alone is insufficient to enable an\nenergy-neutral operation, which can eliminate tedious battery charging and\nreplacement requirements. Achieving an energy-neutral operation is challenging\nsince the uncertainties in harvested energy undermine the quality of service\nrequirements. To address this challenge, we present a runtime energy-allocation\nframework that optimizes the utility of the target device under energy\nconstraints using a rollout algorithm, which is a sequential approach to solve\ndynamic optimization problems. The proposed framework uses an efficient\niterative algorithm to compute initial energy allocations at the beginning of a\nday. The initial allocations are then corrected at every interval to compensate\nfor the deviations from the expected energy harvesting pattern. We evaluate\nthis framework using solar and motion energy harvesting modalities and American\nTime Use Survey data from 4772 different users. Compared to prior techniques,\nthe proposed framework achieves up to 35% higher utility even under\nenergy-limited scenarios. Moreover, measurements on a wearable device prototype\nshow that the proposed framework has 1000x smaller energy overhead than\niterative approaches with a negligible loss in utility.\n"}
{"abstract": "  We introduce the Software Heritage filesystem (SwhFS), a user-space\nfilesystem that integrates large-scale open source software archival with\ndevelopment workflows. SwhFS provides a POSIX filesystem view of Software\nHeritage, the largest public archive of software source code and version\ncontrol system (VCS) development history.Using SwhFS, developers can quickly\n\"checkout\" any of the 2 billion commits archived by Software Heritage, even\nafter they disappear from their previous known location and without incurring\nthe performance cost of repository cloning. SwhFS works across unrelated\nrepositories and different VCS technologies. Other source code artifacts\narchived by Software Heritage-individual source code files and trees, releases,\nand branches-can also be accessed using common programming tools and custom\nscripts, as if they were locally available.A screencast of SwhFS is available\nonline at dx.doi.org/10.5281/zenodo.4531411.\n"}
{"abstract": "  Developing human-like conversational agents is a prime area in HCI research\nand subsumes many tasks. Predicting listener backchannels is one such\nactively-researched task. While many studies have used different approaches for\nbackchannel prediction, they all have depended on manual annotations for a\nlarge dataset. This is a bottleneck impacting the scalability of development.\nTo this end, we propose using semi-supervised techniques to automate the\nprocess of identifying backchannels, thereby easing the annotation process. To\nanalyze our identification module's feasibility, we compared the backchannel\nprediction models trained on (a) manually-annotated and (b) semi-supervised\nlabels. Quantitative analysis revealed that the proposed semi-supervised\napproach could attain 95% of the former's performance. Our user-study findings\nrevealed that almost 60% of the participants found the backchannel responses\npredicted by the proposed model more natural. Finally, we also analyzed the\nimpact of personality on the type of backchannel signals and validated our\nfindings in the user-study.\n"}
{"abstract": "  We study the non-equilibrium dissipative dynamics of the center of mass of a\nparticle coupled to a field via its internal degrees of freedom. We model the\ninternal and external degrees of freedom of the particle as quantum harmonic\noscillators in 1+1 D, with the internal oscillator coupled to a scalar quantum\nfield at the center of mass position. Such a coupling results in a nonlinear\ninteraction between the three pertinent degrees of freedom -- the center of\nmass, internal degree of freedom, and the field. It is typically assumed that\nthe internal dynamics is decoupled from that of the center of mass owing to\ntheir disparate characteristic time scales. Here we use an influence functional\napproach that allows one to account for the self-consistent backaction of the\ndifferent degrees of freedom on each other, including the coupled\nnon-equilibrium dynamics of the internal degree of freedom and the field, and\ntheir influence on the dissipation and noise of the center of mass. Considering\na weak nonlinear interaction term, we employ a perturbative generating\nfunctional approach to derive a second order effective action and a\ncorresponding quantum Langevin equation describing the non-equilibrium dynamics\nof the center of mass. We analyze the resulting dissipation and noise arising\nfrom the field and the internal degree of freedom as a composite environment.\nFurthermore, we establish a generalized fluctuation-dissipation relation for\nthe late-time dissipation and noise kernels. Our results are pertinent to open\nquantum systems that possess intermediary degrees of freedom between system and\nenvironment, such as in the case of optomechanical interactions.\n"}
{"abstract": "  Multilayer-perceptrons (MLP) are known to struggle with learning functions of\nhigh-frequencies, and in particular cases with wide frequency bands. We present\na spatially adaptive progressive encoding (SAPE) scheme for input signals of\nMLP networks, which enables them to better fit a wide range of frequencies\nwithout sacrificing training stability or requiring any domain specific\npreprocessing. SAPE gradually unmasks signal components with increasing\nfrequencies as a function of time and space. The progressive exposure of\nfrequencies is monitored by a feedback loop throughout the neural optimization\nprocess, allowing changes to propagate at different rates among local spatial\nportions of the signal space. We demonstrate the advantage of SAPE on a variety\nof domains and applications, including regression of low dimensional signals\nand images, representation learning of occupancy networks, and a geometric task\nof mesh transfer between 3D shapes.\n"}
{"abstract": "  The short period variable star MG1-688432 has been discovered to exhibit\noccasional extremely high energy optical outbursts as high as 10^31 joules.\nOutbursts are typically of several hours duration. These events are often\nhighly structured, resembling sequential associated releases of energy. Twenty\nyears of time sequence photometry is presented, indicating a basic sinusoidal\nlight curve of mean period 6.65d, with some phase shifting and long-term\ntemporal trends in amplitude and mean brightness. Spectroscopy reveals a\npeculiar star, best resembling a K3 subgiant that has evolved off the main\nsequence moderately red-ward of the giant branch. Spectroscopic and radial\nvelocity analyses indicate a binary system orbiting its barycenter with an\nunseen companion to the K3IV primary. This is not an eclipsing system with the\ninclination of the orbit precluding eclipse by the secondary. The system is at\na distance of 1.5kpc and analysis of GAIA observations leads to the conclusion\nthat the HR diagram position of MG1-688432 is established by an intrinsic\nfeature of the system, most likely either the stellar evolutionary state of the\nobserved star or the presence of small (non-gray) dust within the system. Two\nmechanisms that might give rise to the system are 1) impacts with tidally\ndisrupted planetary debris, and 2) magnetically induced chromospheric activity.\nAn intriguing idea that requires further investigation suggests that the unseen\ncompanion is perhaps a white dwarf star which has encountered a planet and\ntidally shredded it to produce a debris and dust veil that modulates the\nbrightness of the primary.\n"}
{"abstract": "  Motivated by the Ax-Kochen/Ershov principle, a large number of questions\nabout henselian valued fields have been shown to reduce to analogous questions\nabout the value group and residue field. In this paper, we investigate the\nburden of henselian valued fields in the three-sorted Denef-Pas language. If\n$T$ is a theory of henselian valued fields admitting relative quantifier\nelimination (in any characteristic), we show that the burden of $T$ is equal to\nthe sum of the burdens of its value group and residue field. As a consequence,\n$T$ is NTP${}_2$ if and only if its residue field and value group are; the same\nis true for the statements \"$T$ is strong\" and \"$T$ has finite burden.\"\n"}
{"abstract": "  The discipline of process mining aims to study processes in a data-driven\nmanner by analyzing historical process executions, often employing Petri nets.\nEvent data, extracted from information systems (e.g. SAP), serve as the\nstarting point for process mining. Recently, novel types of event data have\ngathered interest among the process mining community, including uncertain event\ndata. Uncertain events, process traces and logs contain attributes that are\ncharacterized by quantified imprecisions, e.g., a set of possible attribute\nvalues. The PROVED tool helps to explore, navigate and analyze such uncertain\nevent data by abstracting the uncertain information using behavior graphs and\nnets, which have Petri nets semantics. Based on these constructs, the tool\nenables discovery and conformance checking.\n"}
{"abstract": "  Gold nanoparticles are unique electrocatalysts for oxygen reduction, carbon\ndioxide reduction, and alcohol oxidation. Electrocatalytic processes are\ninfluenced by the interaction with the solvent, yet the direct investigation of\nthe solvation of nanoparticles is scarce. Here we select gold nanoparticles as\n10 nm sized solutes of which we can control the charge. We perform mid-infrared\nand terahertz spectroscopy to compare the behavior of the water in solution\nwith micelles loaded with neutral and positive gold nanoparticles. We find\nindications that the hydration water around the gold-loaded micelles is\ncharacterized by weaker hydrogen bonds than bulk water. A positive nanoparticle\ncharge is observed to result in a larger blue shift of the OH stretch, quenches\nthe intensity of the collective translational mode, and increases the\nabsorption by librating water molecules. Water at the interface of a positive\ngold nanoparticle could experience a stiffer potential energy surface which, in\nturn, might unveil local thermodynamic properties.\n"}
{"abstract": "  Lossy compression algorithms are typically designed to achieve the lowest\npossible distortion at a given bit rate. However, recent studies show that\npursuing high perceptual quality would lead to increase of the lowest\nachievable distortion (e.g., MSE). This paper provides nontrivial results\ntheoretically revealing that, \\textit{1}) the cost of achieving perfect\nperception quality is exactly a doubling of the lowest achievable MSE\ndistortion, \\textit{2}) an optimal encoder for the \"classic\" rate-distortion\nproblem is also optimal for the perceptual compression problem, \\textit{3})\ndistortion loss is unnecessary for training a perceptual decoder. Further, we\npropose a novel training framework to achieve the lowest MSE distortion under\nperfect perception constraint at a given bit rate. This framework uses a GAN\nwith discriminator conditioned on an MSE-optimized encoder, which is superior\nover the traditional framework using distortion plus adversarial loss.\nExperiments are provided to verify the theoretical finding and demonstrate the\nsuperiority of the proposed training framework.\n"}
{"abstract": "  This paper describes energy distribution in a block medium simulated by a\none-dimensional chain of masses joined by springs and dampers. Equations\ndescribing the motion of masses are solved by the methods of the theory of\nordinary differential equations. The effect of the block medium parameters on\nenergy dissipation is investigated. An approximate analytical solution is\nobtained that describes the total energy of a block medium at large values of\ntime.\n"}
{"abstract": "  The ellipses model is a continuum percolation process in which ellipses with\nrandom orientation and eccentricity are placed in the plane according to a\nPoisson point process. A parameter $\\alpha$ controls the tail distribution of\nthe major axis' distribution and we focus on the regime $\\alpha \\in (1,2)$ for\nwhich there exists a unique infinite cluster of ellipses and this cluster\nfulfills the so called highway property. We prove that the distance within this\ninfinite cluster behaves asymptotically like the (unrestricted) Euclidean\ndistance in the plane. We also show that the chemical distance between points\n$x$ and $y$ behaves roughly as $c \\log\\log |x-y|$.\n"}
{"abstract": "  Automatic evaluations for natural language generation (NLG) conventionally\nrely on token-level or embedding-level comparisons with the text references.\nThis is different from human language processing, for which visual imaginations\noften improve comprehension. In this work, we propose ImaginE, an\nimagination-based automatic evaluation metric for natural language generation.\nWith the help of CLIP and DALL-E, two cross-modal models pre-trained on\nlarge-scale image-text pairs, we automatically generate an image as the\nembodied imagination for the text snippet and compute the imagination\nsimilarity using contextual embeddings. Experiments spanning several text\ngeneration tasks demonstrate that adding imagination with our ImaginE displays\ngreat potential in introducing multi-modal information into NLG evaluation, and\nimproves existing automatic metrics' correlations with human similarity\njudgments in many circumstances.\n"}
{"abstract": "  Black-hole (BH) accretion disks formed in compact-object mergers or\ncollapsars may be major sites of the rapid-neutron-capture (r-)process, but the\nconditions determining the electron fraction (Y_e) remain uncertain given the\ncomplexity of neutrino transfer and angular-momentum transport. After\ndiscussing relevant weak-interaction regimes, we study the role of neutrino\nabsorption for shaping Y_e using an extensive set of simulations performed with\ntwo-moment neutrino transport and again without neutrino absorption. We vary\nthe torus mass, BH mass and spin, and examine the impact of rest-mass and\nweak-magnetism corrections in the neutrino rates. We also test the dependence\non the angular-momentum transport treatment by comparing axisymmetric models\nusing the standard alpha-viscosity with viscous models assuming constant\nviscous length scales (l_t) and three-dimensional magnetohydrodynamic (MHD)\nsimulations. Finally, we discuss the nucleosynthesis yields and basic kilonova\nproperties. We find that absorption pushes Y_e towards ~0.5 outside the torus,\nwhile inside increasing the equilibrium value Y_e^eq by ~0.05-0.2.\nCorrespondingly, a substantial ejecta fraction is pushed above Y_e=0.25,\nleading to a reduced lanthanide fraction and a brighter, earlier, and bluer\nkilonova than without absorption. More compact tori with higher neutrino\noptical depth, tau, tend to have lower Y_e^eq up to tau~1-10, above which\nabsorption becomes strong enough to reverse this trend. Disk ejecta are less\n(more) neutron-rich when employing an l_t=const. viscosity (MHD treatment). The\nsolar-like abundance pattern found for our MHD model marginally supports\ncollapsar disks as major r-process sites, although a strong r-process may be\nlimited to phases of high mass-infall rates, Mdot>~ 2 x 10^(-2) Msun/s.\n"}
{"abstract": "  We present a general framework for the estimation of corporate default based\non a firm's capital structure, when its assets are assumed to follow a pure\njump L\\'evy processes; this setup provides a natural extension to usual default\nmetrics defined in diffusion (log-normal) models, and allows to capture extreme\nmarket events such as sudden drops in asset prices, which are closely linked to\ndefault occurrence. Within this framework, we introduce several pure jump\nprocesses featuring negative jumps only and derive practical closed formulas\nfor equity prices, which enable us to use a moment-based algorithm to calibrate\nthe parameters from real market data and to estimate the associated default\nmetrics. A notable feature of these models is the redistribution of credit risk\ntowards shorter maturity: this constitutes an interesting improvement to\ndiffusion models, which are known to underestimate short term default\nprobabilities. We also provide extensions to a model featuring both positive\nand negative jumps and discuss qualitative and quantitative features of the\nresults. For readers convenience, practical tools for model implementation and\nGitHub links are also included.\n"}
{"abstract": "  The widespread proliferation of data-driven decision-making has ushered in a\nrecent interest in the design of privacy-preserving algorithms. In this paper,\nwe consider the ubiquitous problem of gaussian process (GP) bandit optimization\nfrom the lens of privacy-preserving statistics. We propose a solution for\ndifferentially private GP bandit optimization that combines a uniform kernel\napproximator with random perturbations, providing a generic framework to create\ndifferentially-private (DP) Gaussian process bandit algorithms. For two\nspecific DP settings - joint and local differential privacy, we provide\nalgorithms based on efficient quadrature Fourier feature approximators, that\nare computationally efficient and provably no-regret for popular stationary\nkernel functions. Our algorithms maintain differential privacy throughout the\noptimization procedure and critically do not rely explicitly on the sample path\nfor prediction, making the parameters straightforward to release as well.\n"}
{"abstract": "  We report anomalous heating in a colloidal system, the first observation of\nthe inverse Mpemba effect, where an initially cold system heats up faster than\nan identical warm system coupled to the same thermal bath. For an overdamped,\nBrownian colloidal particle moving in a tilted double-well potential, we find a\nnon-monotonic dependence of the heating times on the initial temperature of the\nsystem, as predicted by an eigenfunction expansion of the associated\nFokker-Planck equation. By carefully tuning parameters, we also observe a\n\"strong\" version of anomalous heating, where a cold system heats up\nexponentially faster than systems prepared under slightly different conditions\n"}
{"abstract": "  We begin with proving a formula relating the Hilbert series of a graded\nalgebra $A$ and the Poincar\\'{e} series for $A$ in two variables. This gives\nthe Fr\\\"oberg formula in the case where the bigraded $Tor^A(k,k)$ is\nconcentrated on the diagonal, which we take as definition of $A$ being\n\"Koszul\". We look at a resolution in the commutative case obtained from the\nKoszul complex in the \"trivially Golod\" case. The algebra structure of\n$Ext_A(k,k)$ is introduced in different ways. Its subalgebra generated by the\none-dimensional elements is by definition the \"Koszul\" dual of $A$. We define\nthe\"generalized Koszul complex\" and construct a minimal resolution in the case\nwhere the cube of the augmentation ideal of $A$ is zero. The above results are\nat least 45 years old and most of it can be found in my thesis. In the second\npart graded Lie algebras are defined. Free Lie algebras and enveloping algebras\nare introduced. The Koszul dual is looked upon as the enveloping algebra of a\nLie algebra in the graded commutative case with examples. The\nPoincar\\'{e}-Birkhoff-Witt theorem is stated giving a formula for the Hilbert\nseries of the enveloping algebra. The homotopy Lie algebra is defined in\ndifferent ways, in particular the Lie subalgebra generated by the\none-dimensional elements. Examples of homotopy Lie algebras are given for\ncomplete intersections, Golod rings and rings with the cube of the augmentation\nideal equal to zero. A logarithmic formula for the dimensions of a Lie algebra\ngiven the Hilbert series for its enveloping algebra is given. Examples of\nperiodic Lie subalgebras are given, yielding irrational Poincar\\'{e} series.\nThe \"holonomy\" Lie algebra of a hyperplane arrangement is defined examplified\nby the graphical arrangement $K_4$.\n"}
{"abstract": "  This research is comparing learning environments to students dropout\nintentions. While using statistics I looked at data and the correlations\nbetween two articles to see how the two studies looked side to side. Learning\nenvironments and dropout intentions can both have vary effects on students.\nThey can both determine if a student does well, or bad in school especially\nmath.\n"}
{"abstract": "  We consider the probabilistic planning problem where the agent (called Player\n1, or P1) can jointly plan the control actions and sensor queries in a sensor\nnetwork and an attacker (called player 2, or P2) can carry out attacks on the\nsensors. We model such an adversarial interaction using a formal model -- a\nreachability game with partially controllable observation functions. The main\ncontribution of this paper is to assess the cost of P1's unawareness: Suppose\nP1 misinterprets the sensor failures as probabilistic node failures due to\nunreliable network communication, and P2 is aware of P1's misinterpretation in\naddition to her partial observability. Then, from which states can P2 carry out\nsensor attacks to ensure, with probability one, that P1 will not be able to\ncomplete her reachability task even though, due to misinterpretation, P1\nbelieves that she can almost-surely achieve her task. We develop an algorithm\nto solve the almost-sure winning sensor-attack strategy given P1's\nobservation-based strategy. Our attack analysis could be used for attack\ndetection in wireless communication networks and the design of provably secured\nattack-aware sensor allocation in decision-theoretic models for cyber-physical\nsystems.\n"}
{"abstract": "  This work is entirely devoted to compare the largest claims from two\nheterogeneous portfolios. It is assumed that the claim amounts in an insurance\nportfolio are nonnegative absolutely continuous random variables and belong to\na general family of distributions. The largest claims have been compared based\non various stochastic orderings. The established sufficient conditions are\nassociated with the matrices and vectors of model parameters. Applications of\nthe results are provided for the purpose of illustration.\n"}
{"abstract": "  In order to handle modern convolutional neural networks (CNNs) efficiently, a\nhardware architecture of CNN inference accelerator is proposed to handle\ndepthwise convolutions and regular convolutions, which are both essential\nbuilding blocks for embedded-computer-vision algorithms. Different from related\nworks, the proposed architecture can support filter kernels with different\nsizes with high flexibility since it does not require extra costs for\nintra-kernel parallelism, and it can generate convolution results faster than\nthe architecture of the related works. The experimental results show the\nimportance of supporting depthwise convolutions and dilated convolutions with\nthe proposed hardware architecture. In addition to depthwise convolutions with\nlarge-kernels, a new structure called DDC layer, which includes the combination\nof depthwise convolutions and dilated convolutions, is also analyzed in this\npaper. For face detection, the computational costs decrease by 30%, and the\nmodel size decreases by 20% when the DDC layers are applied to the network. For\nimage classification, the accuracy is increased by 1% by simply replacing $3\n\\times 3$ filters with $5 \\times 5$ filters in depthwise convolutions.\n"}
{"abstract": "  Binaries play a critical role in the formation, evolution, and fundamental\nproperties of planets, stars, and stellar associations. Observational studies\nin these areas often include a mix of observations aimed at detecting or ruling\nout the presence of stellar companions. Rarely can non-detections rule out all\npossible binary configurations. Here we present MOLUSC, our framework for\nconstraining the range of properties of unseen companions using astrometric,\nimaging, and velocity information. We showcase the use of MOLUSC on a number of\nsystems, ruling out stellar false positives in the signals of HIP67522b, and DS\nTuc Ab. We also demonstrate how MOLUSC could be used to predict the number of\nmissing companions in a stellar sample using the ZEIT sample of young planet\nhosts. Although our results are not significant, with a larger sample MOLUSC\ncould be used to see if close-in planets are less common in young binary\nsystems, as is seen for their older counterparts.\n"}
{"abstract": "  Thermostability is an important prerequisite for enzymes employed for\nindustrial applications. Several machine learning based models have thus been\nformulated for protein classification based on this particular trait. These\nmodels have employed features derived from sequences, structures or both\nresulting in a >93% accuracy based on a 10-fold cross-validation. Besides using\nvarious proteins from a wide range of organisms, such studies also rely on\nhundreds of features. In the present study, an enzyme specific classification\nmodel was created using significantly less number of features that provides a\nsimilar accuracy of classification for thermophilic and non-thermophilic enzyme\nserine proteases. For building the classifier, 219 thermophilic and 200\nmesophilic bacterial genomes were mined for their respective serine protease\nsequences. Features were extracted for 800 sequences followed by feature\nselection. We deployed a random forest based classifier that identified\nthermophilic and non-thermophilic serine proteases with an accuracy of 95.71%.\nKnowledge of thermostability along with amino acid positional shifts can be\nvital for downstream protein engineering techniques. Thus, to emphasize the\nreal time application of the enzyme specific classification model, a web\nplatform has been designed. Combining the sequence data and the classification\nmodel, this prototype can allow users to align their query serine protease\nsequence against the custom database and identify its thermophilic nature.\n"}
{"abstract": "  Network slicing to support multi-tenancy plays a key role in improving the\nperformance of 5G networks. In this paper, we propose a two time-scale\nframework for the reservation-based network slicing in the backhaul and Radio\nAccess Network (RAN). In the proposed two time-scale scheme, a subset of\nnetwork slices is activated via a novel sparse optimization framework in the\nlong time-scale with the goal of maximizing the expected utilities of tenants\nwhile in the short time-scale the activated slices are reconfigured according\nto the time-varying user traffic and channel states. Specifically, using the\nstatistics from users and channels and also considering the expected utility\nfrom serving users of a slice and the reconfiguration cost, we formulate a\nsparse optimization problem to update the configuration of a slice resources\nsuch that the maximum isolation of reserved resources is enforced. The\nformulated optimization problems for long and short time-scales are non-convex\nand difficult to solve. We use the $\\ell_q$-norm, $0<q<1$, and group LASSO\nregularizations to iteratively find convex approximations of the optimization\nproblems. We propose a Frank-Wolfe algorithm to iteratively solve approximated\nproblems in long time-scales. To cope with the dynamical nature of traffic\nvariations, we propose a fast, distributed algorithm to solve the approximated\noptimization problems in short time-scales. Simulation results demonstrate the\nperformance of our approaches relative to optimal solutions and the existing\nstate of the art method.\n"}
{"abstract": "  In the last decade, the advancement of the Internet of Things (IoT) has\ncaused unlicensed radio spectrum, especially the 2.4 GHz ISM band, to be\nimmensely crowded with smart wireless devices that are used in a wide range of\napplication domains. Due to their diversity in radio resource use and channel\naccess techniques, when collocated, these wireless devices create interference\nwith each other, known as Cross-Technology Interference (CTI), which can lead\nto increased packet losses and energy consumption. CTI is a significant problem\nfor low-power wireless networks, such as IEEE 802.15.4, as it decreases the\noverall dependability of the wireless network.\n  To improve the performance of low-power wireless networks under CTI\nconditions, we propose a data-driven proactive receiver-aware MAC protocol,\nLUCID, based on interference estimation and white space prediction. We leverage\nstatistical analysis of real-world traces from two indoor environments\ncharacterised by varying channel conditions to develop CTI prediction methods.\nThe CTI models that generate accurate predictions of interference behaviour are\nan intrinsic part of our solution. LUCID is thoroughly evaluated in realistic\nsimulations and we show that depending on the application data rate and the\nnetwork size, our solution achieves higher dependability, 1.2% increase in\npacket delivery ratio and 0.02% decrease in duty-cycle under bursty indoor\ninterference than state of the art alternative methods.\n"}
{"abstract": "  If $G$ is a group and $S$ a generating set, $G$ canonically embeds into the\nautomorphism group of its Cayley graph and it is natural to try to minimize,\nover all generating sets, the index of this inclusion. This infimum is called\nthe Cayley index of the group. In a recent series of works, we have\ncharacterized the infinite finitely generated groups with Cayley index $1$. We\ncomplement this characterization by showing that the Cayley index is $2$ in the\nremaining cases and is attained for a finite generating set.\n"}
{"abstract": "  For a given pseudo-Anosov homeomorphism $\\phi$ of a closed surface $S$, the\naction of $\\phi$ on the Teichm\\\"uller space $\\mathcal T(S)$ is an automorphism\nof the complex structure preserving the Weil-Petersson-Goldman symplectic form.\nWe give explicit formulae for two invariant functions $\\mathcal T(S)\\to \\mathbb\nR$ whose symplectic gradients generate autonomous Hamiltonian flows that\ncoincide with the action of $\\phi$ at time one. We compute the Poisson bracket\nof these two functions, which essentially amounts to computing the variation of\nlength of a H\\\"older cocyle on one lamination along the shear vector field\ndefined by another. For a measurably generic set of laminations, we prove that\nthe variation of length is expressed as the cosine of the angle between the two\nlaminations integrated against the product H\\\"older distribution, generalizing\na result of Kerckhoff. We also obtain rates of convergence for the supports of\ngerms of differentiable paths of measured laminations in the Hausdorff metric\non a hyperbolic surface, which may be of independent interest.\n"}
{"abstract": "  Various deep neural network architectures (DNNs) maintain massive vital\nrecords in computer vision. While drawing attention worldwide, the design of\nthe overall structure lacks general guidance. Based on the relationship between\nDNN design and numerical differential equations, we performed a fair comparison\nof the residual design with higher-order perspectives. We show that the widely\nused DNN design strategy, constantly stacking a small design (usually 2-3\nlayers), could be easily improved, supported by solid theoretical knowledge and\nwith no extra parameters needed. We reorganise the residual design in\nhigher-order ways, which is inspired by the observation that many effective\nnetworks can be interpreted as different numerical discretisations of\ndifferential equations. The design of ResNet follows a relatively simple\nscheme, which is Euler forward; however, the situation becomes complicated\nrapidly while stacking. We suppose that stacked ResNet is somehow equalled to a\nhigher-order scheme; then, the current method of forwarding propagation might\nbe relatively weak compared with a typical high-order method such as\nRunge-Kutta. We propose HO-ResNet to verify the hypothesis of widely used CV\nbenchmarks with sufficient experiments. Stable and noticeable increases in\nperformance are observed, and convergence and robustness are also improved. Our\nstacking strategy improved ResNet-30 by 2.15 per cent and ResNet-58 by 2.35 per\ncent on CIFAR-10, with the same settings and parameters. The proposed strategy\nis fundamental and theoretical and can therefore be applied to any network as a\ngeneral guideline.\n"}
{"abstract": "  In microbiome studies, one of the ways of studying bacterial abundances is to\nestimate bacterial composition based on the sequencing read counts. Various\ntransformations are then applied to such compositional data for downstream\nstatistical analysis, among which the centered log-ratio (clr) transformation\nis most commonly used.\n  Due to limited sequencing depth and DNA dropouts, many rare bacterial taxa\nmight not be captured in the final sequencing reads, which results in many zero\ncounts. Naive composition estimation using count normalization leads to many\nzero proportions, which makes clr transformation infeasible. This paper\nproposes a multi-sample approach to estimation of the clr matrix directly in\norder to borrow information across samples and across species. Empirical\nresults from real datasets suggest that the clr matrix over multiple samples is\napproximately low rank, which motivates a regularized maximum likelihood\nestimation with a nuclear norm penalty. An efficient optimization algorithm\nusing the generalized accelerated proximal gradient is developed. Theoretical\nupper bounds of the estimation errors and of its corresponding singular\nsubspace errors are established. Simulation studies demonstrate that the\nproposed estimator outperforms the naive estimators. The method is analyzed on\nGut Microbiome dataset and the American Gut project.\n"}
{"abstract": "  Hidden Markov models with observations in a Euclidean space play an important\nrole in signal and image processing. Previous work extending to models where\nobservations lie in Riemannian manifolds based on the Baum-Welch algorithm\nsuffered from high memory usage and slow speed. Here we present an algorithm\nthat is online, more accurate, and offers dramatic improvements in speed and\nefficiency.\n"}
{"abstract": "  Preprint is a version of a scientific paper that is publicly distributed\npreceding formal peer review. Since the launch of arXiv in 1991, preprints have\nbeen increasingly distributed over the Internet as opposed to paper copies. It\nallows open online access to disseminate the original research within a few\ndays, often at a very low operating cost. This work overviews how preprint has\nbeen evolving and impacting the research community over the past thirty years\nalongside the growth of the Web. In this work, we first report that the number\nof preprints has exponentially increased 63 times in 30 years, although it only\naccounts for 4% of research articles. Second, we quantify the benefits that\npreprints bring to authors: preprints reach an audience 14 months earlier on\naverage and associate with five times more citations compared with a\nnon-preprint counterpart. Last, to address the quality concern of preprints, we\ndiscover that 41% of preprints are ultimately published at a peer-reviewed\ndestination, and the published venues are as influential as papers without a\npreprint version. Additionally, we discuss the unprecedented role of preprints\nin communicating the latest research data during recent public health\nemergencies. In conclusion, we provide quantitative evidence to unveil the\npositive impact of preprints on individual researchers and the community.\nPreprints make scholarly communication more efficient by disseminating\nscientific discoveries more rapidly and widely with the aid of Web\ntechnologies. The measurements we present in this study can help researchers\nand policymakers make informed decisions about how to effectively use and\nresponsibly embrace a preprint culture.\n"}
{"abstract": "  We study the metallicity dependence of the H/H$_2$ and C$^+$/C/CO\ndistributions in a self-regulated interstellar medium (ISM) across a broad\nrange of metallicities ($0.1 < Z/Z_\\odot < 3$). To this end, we conduct\nhigh-resolution (particle mass of $1\\ {\\rm M_\\odot}$) hydrodynamical\nsimulations coupled with a time-dependent H$_2$ chemistry network. The results\nare then post-processed with an accurate chemistry network to model the\nassociated C$^+$/C/CO abundances, based on the time-dependent non-steady-state\n(``non-equilibrium'') H$_2$ abundances. We find that the time-averaged star\nformation rate and the ISM structure are insensitive to metallicity. The column\ndensities relevant for molecular shielding appear correlated with the volume\ndensities in gravitationally unstable gas. As metallicity decreases, H$_2$\nprogressively deviates from steady state (``equilibrium'') and shows shallow\nabundance profiles until they sharply truncate at the photodissociation fronts.\nIn contrast, the CO profile is sharp and controlled by photodissociation as CO\nquickly reaches steady state. We construct effective one-dimensional cloud\nmodels that successfully capture the time-averaged chemical distributions in\nsimulations. At low metallicities, the steady-state model significantly\noverestimates the abundance of H$_2$ in the diffuse medium. The overestimated\nH$_2$, however, has little impact on CO. Consequently, the mass fraction of\nCO-dark H$_2$ gas is significantly lower than what a fully steady-state model\npredicts. The mass ratios of H$_2$/C$^+$ and H$_2$/C both show a weaker\ndependence on $Z^{\\prime}$ than H$_2$/CO, which potentially indicates that\nC$^+$ and C could be alternative tracers for H$_2$ at low $Z^{\\prime}$ in terms\nof mass budget. Our chemistry code for post-processing is publicly available.\n"}
{"abstract": "  Several pathologies can alter the way people walk, i.e. their gait. Gait\nanalysis can therefore be used to detect impairments and help diagnose\nillnesses and assess patient recovery. Using vision-based systems, diagnoses\ncould be done at home or in a clinic, with the needed computation being done\nremotely. State-of-the-art vision-based gait analysis systems use deep\nlearning, requiring large datasets for training. However, to our best\nknowledge, the biggest publicly available pathological gait dataset contains\nonly 10 subjects, simulating 4 gait pathologies. This paper presents a new\ndataset called GAIT-IT, captured from 21 subjects simulating 4 gait\npathologies, with 2 severity levels, besides normal gait, being considerably\nlarger than publicly available gait pathology datasets, allowing to train a\ndeep learning model for gait pathology classification. Moreover, it was\nrecorded in a professional studio, making it possible to obtain nearly perfect\nsilhouettes, free of segmentation errors. Recognizing the importance of remote\nhealthcare, this paper proposes a prototype of a web application allowing to\nupload a walking person's video, possibly acquired using a smartphone camera,\nand execute a web service that classifies the person's gait as normal or across\ndifferent pathologies. The web application has a user friendly interface and\ncould be used by healthcare professionals or other end users. An automatic gait\nanalysis system is also developed and integrated with the web application for\npathology classification. Compared to state-of-the-art solutions, it achieves a\ndrastic reduction in the number of model parameters, which means significantly\nlower memory requirements, as well as lower training and execution times.\nClassification accuracy is on par with the state-of-the-art.\n"}
{"abstract": "  What is the minimum time required to take the temperature? In this paper, we\nsolve this question for any process where temperature is inferred by measuring\na probe (the thermometer) weakly coupled to the sample of interest, so that the\nprobe's evolution is well described by a quantum Markovian master equation.\nConsidering the most general control strategy on the probe (adaptive\nmeasurements, arbitrary control on the probe's state and Hamiltonian), we\nprovide bounds on the achievable measurement precision in a finite amount of\ntime, and show that in many scenarios these fundamental limits can be saturated\nwith a relatively simple experiment. We find that for a general class of\nsample-probe interactions the scaling of the measurement uncertainty is\ninversely proportional to the time of the process, a shot-noise like behaviour\nthat arises due to the dissipative nature of thermometry. As a side result, we\nshow that the Lamb shift induced by the probe-sample interaction can play a\nrelevant role in thermometry, allowing for finite measurement resolution in the\nlow-temperature regime (more precisely, the measurement uncertainty decays\npolynomially with the temperature as $T\\rightarrow 0$, in contrast to the usual\nexponential decay with $T^{-1}$). We illustrate these general results for (i) a\nqubit probe interacting with a bosonic sample, where the role of the Lamb shit\nis highlighted, and (ii) a collective superradiant coupling between a $N$-qubit\nprobe and a sample, which enables a quadratic decay with $N^2$ of the\nmeasurement uncertainty.\n"}
{"abstract": "  Instance-based interpretation methods have been widely studied for supervised\nlearning methods as they help explain how black box neural networks predict.\nHowever, instance-based interpretations remain ill-understood in the context of\nunsupervised learning. In this paper, we investigate influence functions [20],\na popular instance-based interpretation method, for a class of deep generative\nmodels called variational auto-encoders (VAE). We formally frame the\ncounter-factual question answered by influence functions in this setting, and\nthrough theoretical analysis, examine what they reveal about the impact of\ntraining samples on classical unsupervised learning methods. We then introduce\nVAE-TracIn, a computationally efficient and theoretically sound solution based\non Pruthi et al., for VAEs. Finally, we evaluate VAE-TracIn on several real\nworld datasets with extensive quantitative and qualitative analysis.\n"}
{"abstract": "  When an individual's DNA is sequenced, sensitive medical information becomes\navailable to the sequencing laboratory. A recently proposed way to hide an\nindividual's genetic information is to mix in DNA samples of other individuals.\nWe assume these samples are known to the individual but unknown to the\nsequencing laboratory. Thus, these DNA samples act as \"noise\" to the sequencing\nlaboratory, but still allow the individual to recover their own DNA samples\nafterward. Motivated by this idea, we study the problem of hiding a binary\nrandom variable X (a genetic marker) with the additive noise provided by mixing\nDNA samples, using mutual information as a privacy metric. This is equivalent\nto the problem of finding a worst-case noise distribution for recovering X from\nthe noisy observation among a set of feasible discrete distributions. We\ncharacterize upper and lower bounds to the solution of this problem, which are\nempirically shown to be very close. The lower bound is obtained through a\nconvex relaxation of the original discrete optimization problem, and yields a\nclosed-form expression. The upper bound is computed via a greedy algorithm for\nselecting the mixing proportions.\n"}
{"abstract": "  We calculate the deflection angle, as well as the positions and\nmagnifications of the lensed images, in the case of covariant $f(T)$ gravity.\nWe first extract the spherically symmetric solutions for both the pure-tetrad\nand the covariant formulation of the theory, since considering spherical\nsolutions the extension to the latter is crucial, in order for the results not\nto suffer from frame-dependent artifacts. Applying the weak-field, perturbative\napproximation we extract the deviations of the solutions comparing to General\nRelativity. Furthermore, we calculate the deflection angle and then the\ndifferences of the positions and magnifications in the lensing framework. This\neffect of consistent $f(T)$ gravity on the lensing features can serve as an\nobservable signature in the realistic cases where $f(T)$ is expected to deviate\nonly slightly from General Relativity, since lensing scales in general are not\nrestricted as in the case of Solar System data, and therefore deviations from\nGeneral Relativity could be observed more easily.\n"}
{"abstract": "  Let $D_N$ be the set of points around which a planar Brownian motion winds at\nleast $N$ times. We prove that the random measure on the plane with density $2\n\\pi N 1_{D_N}$ with respect to the Lebesgue measure converges almost surely\nweakly, as $N$ tends to infinity, towards the occupation measure of the\nBrownian motion.\n"}
{"abstract": "  This paper presents the characterization measurements and related uncertainty\nevaluation of a nonmagnetic material using the Vector Network Analyzer (VNA) at\nmicrowave frequencies. The permittivity of the material under test is computed\nfrom the scattering parameters (S-parameters). The aim of the work is to\nhighlight the different uncertainty contributions affecting the permittivity.\n"}
{"abstract": "  Interactions govern the flow of information and the formation of correlations\nin quantum systems, dictating the phases of matter found in nature and the\nforms of entanglement generated in the laboratory. Typical interactions decay\nwith distance and thus produce a network of connectivity governed by geometry,\ne.g., by the crystalline structure of a material or the trapping sites of atoms\nin a quantum simulator. However, many envisioned applications in quantum\nsimulation and computation require richer coupling graphs including nonlocal\ninteractions, which notably feature in mappings of hard optimization problems\nonto frustrated spin systems and in models of information scrambling in black\nholes. Here, we report on the realization of programmable nonlocal interactions\nin an array of atomic ensembles within an optical cavity, where photons carry\ninformation between distant atomic spins. By programming the\ndistance-dependence of interactions, we access effective geometries where the\ndimensionality, topology, and metric are entirely distinct from the physical\narrangement of atoms. As examples, we engineer an antiferromagnetic triangular\nladder, a Moebius strip with sign-changing interactions, and a treelike\ngeometry inspired by concepts of quantum gravity. The tree graph constitutes a\ntoy model of holographic duality, where the quantum system may be viewed as\nlying on the boundary of a higher-dimensional geometry that emerges from\nmeasured spin correlations. Our work opens broader prospects for simulating\nfrustrated magnets and topological phases, investigating quantum optimization\nalgorithms, and engineering new entangled resource states for sensing and\ncomputation.\n"}
{"abstract": "  Aggregating multiple sources of weak supervision (WS) can ease the\ndata-labeling bottleneck prevalent in many machine learning applications, by\nreplacing the tedious manual collection of ground truth labels. Current state\nof the art approaches that do not use any labeled training data, however,\nrequire two separate modeling steps: Learning a probabilistic latent variable\nmodel based on the WS sources -- making assumptions that rarely hold in\npractice -- followed by downstream model training. Importantly, the first step\nof modeling does not consider the performance of the downstream model. To\naddress these caveats we propose an end-to-end approach for directly learning\nthe downstream model by maximizing its agreement with probabilistic labels\ngenerated by reparameterizing previous probabilistic posteriors with a neural\nnetwork. Our results show improved performance over prior work in terms of end\nmodel performance on downstream test sets, as well as in terms of improved\nrobustness to dependencies among weak supervision sources.\n"}
{"abstract": "  The aim of this paper is to develop a Hamilton--Jacobi theory for contact\nHamiltonian systems. We find several forms for a suitable Hamilton-Jacobi\nequation accordingly to the Hamiltonian and the evolution vector fields for a\ngiven Hamiltonian function. We also analyze the corresponding formulation on\nthe symplectification of the contact Hamiltonian system, and establish the\nrelations between these two approaches. In the last section, some examples are\ndiscussed.\n"}
{"abstract": "  We establish the local Lipschitz continuity and the higher differentiability\nof vector-valued local minimizers of a class of energy integrals of the\nCalculus of Variations. The main novelty is that we deal with possibly\ndegenerate energy densities with respect to the x-variable.\n"}
{"abstract": "  Measurement of charge configurations in few-electron quantum dots is a vital\ntechnique for spin-based quantum information processing. While fast and\nhigh-fidelity measurement is possible by using proximal quantum dot charge\nsensors, their operating range is limited and prone to electrical disturbances.\nHere we demonstrate realtime operation of a charge sensor in a feedback loop to\nmaintain its sensitivity suitable for fast charge sensing in a Si/SiGe double\nquantum dot. Disturbances to the charge sensitivity, due to variation of gate\nvoltages for operating the quantum dot and $1/f$ charge fluctuation, are\ncompensated by a digital PID controller with the bandwidth of $\\approx\n100\\,{\\rm kHz}$. The rapid automated tuning of a charge sensor enables\nunobstructed charge stability diagram measurement facilitating realtime quantum\ndot tuning and submicrosecond single-shot spin readout without compromising the\nperformance of a charge sensor in time-consuming experiments for quantum\ninformation processing.\n"}
{"abstract": "  We address the problem of compiling defeasible theories to Datalog$^\\neg$\nprograms. We prove the correctness of this compilation, for the defeasible\nlogic $DL(\\partial_{||})$, but the techniques we use apply to many other\ndefeasible logics. Structural properties of $DL(\\partial_{||})$ are identified\nthat support efficient implementation and/or approximation of the conclusions\nof defeasible theories in the logic, compared with other defeasible logics. We\nalso use previously well-studied structural properties of logic programs to\nadapt to incomplete Datalog$^\\neg$ implementations.\n"}
{"abstract": "  Magnetic exchange interactions (MEIs) define networks of coupled magnetic\nmoments and lead to a surprisingly rich variety of their magnetic properties.\nTypically MEIs can be estimated by fitting experimental results. But how many\nMEIs need to be included in the fitting process for a material is not clear a\npriori, which limits the quality of results obtained by these conventional\nmethods. In this paper, based on linear spin-wave theory but without performing\nmatrix diagonalization, we show that for a general quadratic spin Hamiltonian,\nthere is a simple relation between the Fourier transform of MEIs and the sum of\nsquare of magnon energies (SSME). We further show that according to the\nreal-space distance range within which MEIs are considered relevant, one can\nobtain the corresponding relationships between SSME in momentum space. We also\ndevelop a theoretical tool for tabulating the rule about SSME. By directly\nutilizing these characteristics and the experimental magnon energies at only a\nfew high-symmetry k points in the Brillouin zone, one can obtain strong\nconstraints about the range of exchange path beyond which MEIs can be safely\nneglected. Our methodology is also general applicable for other Hamiltonian\nwith quadratic Fermi or Boson operators.\n"}
{"abstract": "  Galactic scale tests have proven to be powerful tools in constraining\nfundamental physics in previously under-explored regions of parameter space.\nThe astrophysical regime which they probe is inherently complicated, and the\ninference methods used to make these constraints should be robust to baryonic\neffects. Previous analyses have assumed simple empirical models for\nastrophysical noise without detailed calibration or justification. We outline a\nframework for assessing the reliability of such methods by constructing and\ntesting more advanced baryonic models using cosmological hydrodynamical\nsimulations. As a case study, we use the Horizon-AGN simulation to investigate\nwarping of stellar disks and offsets between gas and stars within galaxies,\nwhich are powerful probes of screened fifth forces. We show that the degree of\n`U'-shaped warping of galaxies is well modelled by Gaussian random noise, but\nthat the magnitude of the gas-star offset is correlated with the virial radius\nof the host halo. By incorporating this correlation we confirm recent results\nruling out astrophysically relevant Hu-Sawicki $f(R)$ gravity, and identify a\n$\\sim 30\\%$ systematic uncertainty due to baryonic physics. Such an analysis\nmust be performed case-by-case for future galactic tests of fundamental\nphysics.\n"}
{"abstract": "  The aim of the article is to study the Betti numbers of the tangent cone of\nGorenstein monomial curves in affine 4-space. If $C_S$ is a non-complete\nintersection Gorenstein monomial curve whose tangent cone is Cohen-Macaulay, we\nshow that the possible Betti sequences are (1,5,5,1), (1,5,6,2) and (1,6,8,3).\n"}
{"abstract": "  In this article we mainly aim to know what kind of asymptotic behavior of\ntypical orbits can display. For example, we show in any transitive system, the\nemprical measures of a typical orbit can cover all emprical measures of dense\norbits and can intersect some physical-like measures. In particular, if the\nunion set of emprical measures of all dense orbits is not singleton, then the\ntypical orbit will display historic behavior simultaneously for typical\ncontinuous functions and the limit set of ergodic average along every\ncontinuous function equals to a closed interval composed by the union of limit\nsets of ergodic average on all dense orbits. Moreover, if the union set of\nemprical measures of all dense orbits contains all ergodic measures, the above\ninterval equals to the rotation set. These results are not only suitable for\nsystems with specification-like properties or minimal systems, but also\nsuitable for many other systems including all general (not assumed uniformly\nhyperbolic) nontrivial homoclinic classes and Bowen eyes. Moreover, we\nintroduce a new property called m-g-product property weaker than classical\nspecification property and minimal property and nontrivial examples are\nconstructed.\n"}
{"abstract": "  We consider the genus-one curves which arise in the cuts of the sunrise and\nin the elliptic double-box Feynman integrals. We compute and compare invariants\nof these curves in a number of ways, including Feynman parametrization,\nlightcone and Baikov (in full and loop-by-loop variants). We find that the same\ngeometry for the genus-one curves arises in all cases, which lends support to\nthe idea that there exists an invariant notion of genus-one geometry,\nindependent on the way it is computed. We further indicate how to interpret\nsome previous results which found that these curves are related by isogenies\ninstead.\n"}
{"abstract": "  This and companion papers by Harrington et al. and Blecic et al. present the\nBayesian Atmospheric Radiative Transfer (BART) code, an open-source,\nopen-development package to characterize extrasolar-planet atmospheres. BART\ncombines a thermochemical equilibrium abundances (TEA), a radiative-transfer\n(Transit), and a Bayesian statistical (MC3) module to constrain atmospheric\ntemperatures and molecular abundances for given spectroscopic observations.\nHere, we describe the Transit radiative-transfer package, an efficient\nline-by-line radiative-transfer C code for one-dimensional atmospheres,\ndeveloped by P. Rojo and further modified by the UCF exoplanet group. This code\nproduces transmission and hemisphere-integrated emission spectra. Transit\nhandles line-by-line opacities from HITRAN, Partridge \\& Schwenke ({\\water}),\nSchwenke (TiO), and Plez (VO); and collision-induced absorption from Borysow,\nHITRAN, and ExoMol. Transit emission-spectra models agree with models from C.\nMorley (priv. comm.) within a few percent. We applied BART to the {\\Spitzer}\nand {\\Hubble} transit observations of the Neptune-sized planet HAT-P-11b. Our\nresults generally agree with those from previous studies, constraining the\n{\\water} abundance and finding an atmosphere enhanced in heavy elements.\nDifferent conclusions start to emerge when we make different assumptions from\nother studies. The BART source code and documentation are available at\nhttps://github.com/exosports/BART.\n"}
{"abstract": "  We construct a Darboux transformation for a class of two-dimensional Dirac\nsystems at zero energy. Our starting equation features a position-dependent\nmass, a matrix potential, and an additional degree of freedom that can be\ninterpreted either as a magnetic field perpendicular to the plane or a\ngeneralized Dirac oscillator interaction. We obtain a number of\nDarbouxtransformed Dirac equations for which the zero energy solutions are\nexactly known.\n"}
{"abstract": "  One of the most influential results in neural network theory is the universal\napproximation theorem [1, 2, 3] which states that continuous functions can be\napproximated to within arbitrary accuracy by single-hidden-layer feedforward\nneural networks. The purpose of this paper is to establish a result in this\nspirit for the approximation of general discrete-time linear dynamical systems\n- including time-varying systems - by recurrent neural networks (RNNs). For the\nsubclass of linear time-invariant (LTI) systems, we devise a quantitative\nversion of this statement. Specifically, measuring the complexity of the\nconsidered class of LTI systems through metric entropy according to [4], we\nshow that RNNs can optimally learn - or identify in system-theory parlance -\nstable LTI systems. For LTI systems whose input-output relation is\ncharacterized through a difference equation, this means that RNNs can learn the\ndifference equation from input-output traces in a metric-entropy optimal\nmanner.\n"}
{"abstract": "  The propagation of acoustic or elastic waves in artificial crystals,\nincluding the case of phononic and sonic crystals, is inherently anisotropic.\nAs is known from the theory of periodic composites, anisotropy is directly\ndictated by the space group of the unit cell of the crystal and the rank of the\nelastic tensor. Here, we examine effective velocities in the long wavelength\nlimit of periodic acoustic and elastic composites as a function of the\ndirection of propagation. We derive explicit and efficient formulas for\nestimating the effective velocity surfaces, based on second-order perturbation\ntheory, generalizing the Christofell equation for elastic waves in solids. We\nidentify strongly anisotropic sonic crystals for scalar acoustic waves and\nstrongly anisotropic phononic crystals for vector elastic waves. Furthermore,\nwe observe that under specific conditions, quasi-longitudinal waves can be made\nmuch slower than shear waves propagating in the same direction.\n"}
{"abstract": "  Confined nanoscale spaces, electric fields and tunneling currents make the\nmolecular electronic junction an experimental device for the discovery of new,\nout-of-equilibrium chemical reactions. Reaction-rate theory for\ncurrent-activated chemical reactions is developed by combining a Keldysh\nnonequilibrium Green's functions treatment of electrons, Fokker-Planck\ndescription of the reaction coordinate, and Kramers' first-passage time\ncalculations. The NEGF provide an adiabatic potential as well as a diffusion\ncoefficient and temperature with local dependence on the reaction coordinate.\nVan Kampen's Fokker-Planck equation, which describes a Brownian particle moving\nin an external potential in an inhomogeneous medium with a position-dependent\nfriction and diffusion coefficient, is used to obtain an analytic expression\nfor the first-passage time. The theory is applied to several transport\nscenarios: a molecular junction with a single, reaction coordinate dependent\nmolecular orbital, and a model diatomic molecular junction. We demonstrate the\nnatural emergence of Landauer's blowtorch effect as a result of the interplay\nbetween the configuration dependent viscosity and diffusion coefficients. The\nresultant localized heating in conjunction with the bond-deformation due to\ncurrent-induced forces are shown to be the determining factors when considering\nchemical reaction rates; each of which result from highly tunable parameters\nwithin the system.\n"}
{"abstract": "  In recent years, there has been an increasing awareness of both the public\nand scientific community that algorithmic systems can reproduce, amplify, or\neven introduce unfairness in our societies. These lecture notes provide an\nintroduction to some of the core concepts in algorithmic fairness research. We\nlist different types of fairness-related harms, explain two main notions of\nalgorithmic fairness, and map the biases that underlie these harms upon the\nmachine learning development process.\n"}
{"abstract": "  In this paper, using $I$-method, we establish $H^s_x\\times H^s_x$ scattering\ntheories for the following Cauchy problem \\begin{equation*} \\left\\{\n\\begin{array}{lll} iu_t+\\Delta u=\\lambda |v|^2u,\\quad iv_t+\\Delta\nv=\\mu|u|^2v,\\quad x\\in \\mathbb{R}^3,\\ t>0,\\\\ u(x,0)=u_0(x),\\quad\nv(x,0)=v_0(x),\\quad x\\in \\mathbb{R}^3. \\end{array}\\right. \\end{equation*} Here\n$\\lambda>0$, $\\mu>0$, $(u_0,v_0)\\in H^s_x(\\mathbb{R}^3)\\times\nH^s_x(\\mathbb{R}^3)$ and $\\frac{1}{2}<s<1$.\n"}
{"abstract": "  Hand segmentation is a crucial task in first-person vision. Since\nfirst-person images exhibit strong bias in appearance among different\nenvironments, adapting a pre-trained segmentation model to a new domain is\nrequired in hand segmentation. Here, we focus on appearance gaps for hand\nregions and backgrounds separately. We propose (i) foreground-aware image\nstylization and (ii) consensus pseudo-labeling for domain adaptation of hand\nsegmentation. We stylize source images independently for the foreground and\nbackground using target images as style. To resolve the domain shift that the\nstylization has not addressed, we apply careful pseudo-labeling by taking a\nconsensus between the models trained on the source and stylized source images.\nWe validated our method on domain adaptation of hand segmentation from real and\nsimulation images. Our method achieved state-of-the-art performance in both\nsettings. We also demonstrated promising results in challenging multi-target\ndomain adaptation and domain generalization settings. Code is available at\nhttps://github.com/ut-vision/FgSty-CPL.\n"}
{"abstract": "  Embedded solitons are exceptional modes in nonlinear-wave systems with the\npropagation constant falling in the system's propagation band. An especially\nchallenging topic is seeking for such modes in nonlinear dynamical lattices\n(discrete systems). We address this problem for a system of coupled discrete\nequations modeling the light propagation in an array of tunnel-coupled\nwaveguides with a combination of intrinsic quadratic\n(second-harmonic-generating) and cubic nonlinearities. Solutions for discrete\nembedded solitons (DESs) are constructed by means of two analytical\napproximations, adjusted, severally, for broad and narrow DESs, and in a\nsystematic form with the help of numerical calculations. DESs of several types,\nincluding ones with identical and opposite signs of their fundamental-frequency\nand second-harmonic components, are produced. In the most relevant case of\nnarrow DESs, the analytical approximation produces very accurate results, in\ncomparison with the numerical findings. In this case, the DES branch extends\nfrom the propagation band into a semi-infinite gap as a family of regular\ndiscrete solitons. The study of stability of DESs demonstrates that, in\naddition to ones featuring the well-known property of semi-stability, there are\nlinearly stable DESs which are genuinely robust modes.\n"}
{"abstract": "  We revisit several aspects of the interaction of self-gravitating, slowly\nvarying sources with their own emitted radiation within the context of\npost-Newtonian approximation to General Relativity. We discuss and clarify the\nchoice of boundary conditions of Green's functions used to determine\nconservative potentials, and the interplay between the so-called near and far\nzones, as well as the relation between far zone ultra-violet divergences and\nemitted power. Both near and far zone contributions are required for the\ncomputation of the conservative dynamics. Within a field-theory approach we\nrederive far-zone self-energy processes, known as tail and memory effects,\ngeneralising the calculation of their divergent part to arbitrary order in the\npost-Newtonian expansion.\n"}
{"abstract": "  In this work we explore the information processing inside neural networks\nusing logistic regression probes \\cite{probes} and the saturation metric\n\\cite{featurespace_saturation}. We show that problem difficulty and neural\nnetwork capacity affect the predictive performance in an antagonistic manner,\nopening the possibility of detecting over- and under-parameterization of neural\nnetworks for a given task. We further show that the observed effects are\nindependent from previously reported pathological patterns like the ``tail\npattern'' described in \\cite{featurespace_saturation}. Finally we are able to\nshow that saturation patterns converge early during training, allowing for a\nquicker cycle time during analysis\n"}
{"abstract": "  Millimeter wave (mmWave) is a practical solution to provide high data rate\nfor vehicle-to-everything (V2X) communications. This enables the future\nautonomous vehicles to exchange big data with the base stations (BSs) such as\nthe velocity, and the location to enhance the safety for the advanced driving\nassistance system (ADAS). To achieve this goal, we propose to develop a\nsituational rate map to characterize the distribution of the rates achieved\nbetween the BSs and a uniform grid of vehicles. In this context, we consider a\nmmWave 5G cellular system with two physical structures which are the\nanalog-only beamforming and hybrid precoding with limited feedback to\ninvestigate the rate distribution for single and multiuser scenarios. We will\nuse the Ray-Tracing tool to construct the simulation environment and generate\nthe channels between the BSs and the grid of vehicles. Finally, we will study\nthe effects of the carrier frequency, the bandwidth, the BSs deployment, the\nblockage, the codebook, the physical architectures and the number of served\nusers on the rate distribution maps. Moreover, we will present the rate\nstatistics to evaluate the coverage of served users for certain services\nrequirements and for various road shapes such as corners, intersections and\nstraightways.\n"}
{"abstract": "  Voice-driven services (VDS) are being used in a variety of applications\nranging from smart home control to payments using digital assistants. The input\nto such services is often captured via an open voice channel, e.g., using a\nmicrophone, in an unsupervised setting. One of the key operational security\nrequirements in such setting is the freshness of the input speech. We present\nAEOLUS, a security overlay that proactively embeds a dynamic acoustic nonce at\nthe time of user interaction, and detects the presence of the embedded nonce in\nthe recorded speech to ensure freshness. We demonstrate that acoustic nonce can\n(i) be reliably embedded and retrieved, and (ii) be non-disruptive (and even\nimperceptible) to a VDS user. Optimal parameters (acoustic nonce's operating\nfrequency, amplitude, and bitrate) are determined for (i) and (ii) from a\npractical perspective. Experimental results show that AEOLUS yields 0.5% FRR at\n0% FAR for speech re-use prevention upto a distance of 4 meters in three\nreal-world environments with different background noise levels. We also conduct\na user study with 120 participants, which shows that the acoustic nonce does\nnot degrade overall user experience for 94.16% of speech samples, on average,\nin these environments. AEOLUS can therefore be used in practice to prevent\nspeech re-use and ensure the freshness of speech input.\n"}
{"abstract": "  This paper proposes a neural network that performs audio transformations to\nuser-specified sources (e.g., vocals) of a given audio track according to a\ngiven description while preserving other sources not mentioned in the\ndescription. Audio Manipulation on a Specific Source (AMSS) is challenging\nbecause a sound object (i.e., a waveform sample or frequency bin) is\n`transparent'; it usually carries information from multiple sources, in\ncontrast to a pixel in an image. To address this challenging problem, we\npropose AMSS-Net, which extracts latent sources and selectively manipulates\nthem while preserving irrelevant sources. We also propose an evaluation\nbenchmark for several AMSS tasks, and we show that AMSS-Net outperforms\nbaselines on several AMSS tasks via objective metrics and empirical\nverification.\n"}
{"abstract": "  We study gravitational wave (GW) emission in the strong-field regime by a\nhierarchical triple system composed of a binary system placed in the vicinity\nof a supermassive black hole (SMBH). The LIGO-Virgo collaboration recently\nreported evidence for coalescences with this dynamical origin. These systems\nare common in galactic centers and thus are a target for the space-based LISA\nmission as well as other advanced detectors. Doppler shifts, aberration,\nlensing and strong amplitude modulations are features present in the GW signal\nfrom these systems, built into our framework and with no need for\nphenomenological patches. We find that the binary can resonantly excite the\nquasinormal modes of the SMBH, as in the resonant excitation of two tuning\nforks with matching frequencies. The flux of energy crossing the SMBH horizon\ncan be significant, when compared with that from standard extreme-mass-ratio\ninspirals. Therefore, these triple systems are an outstanding probe of\nstrong-field physics and of the BH nature of compact objects.\n"}
{"abstract": "  Consider the elastic scattering of an incident wave by a rigid obstacle in\nthree dimensions, which is formulated as an exterior problem for the Navier\nequation. By constructing a Dirichlet-to-Neumann (DtN) operator and introducing\na transparent boundary condition, the scattering problem is reduced\nequivalently to a boundary value problem in a bounded domain. The discrete\nproblem with the truncated DtN operator is solved by using the a posteriori\nerror estimate based adaptive finite element method. The estimate takes account\nof both the finite element approximation error and the truncation error of the\nDtN operator, where the latter is shown to converge exponentially with respect\nto the truncation parameter. Moreover, the generalized Woodbury matrix identity\nis utilized to solve the resulting linear system efficiently. Numerical\nexperiments are presented to demonstrate the superior performance of the\nproposed method.\n"}
{"abstract": "  We compute the leading-logarithmic correction to the neutrino mass matrix in\nthe Standard Model effective field theory (SMEFT) to dimension seven. In the\nlimit of negligible lepton and down-type quark Yukawa couplings, it receives\ncontributions from the Weinberg dimension-five operator as well as from 11\ndimension-six and five dimension-seven independent interactions. Two of the\nmain implications we derive from this result are the following. First, we find\ndimension-seven operators which, despite violating lepton number, do not\nrenormalise neutrino masses at one loop. And second, we demonstrate that the\npresence of dimension-six operators around the TeV scale can modify the\nStandard Model prediction by up to $\\mathcal{O}(50\\,\\%)$. Our result comprises\nalso one step forward towards the renormalisation of the SMEFT to order\n$v^3/\\Lambda^3$.\n"}
{"abstract": "  Accurate prediction of epileptic seizures allows patients to take preventive\nmeasures in advance to avoid possible injuries. In this work, a novel\nconvolutional neural network (CNN) is proposed to analyze time, frequency, and\nchannel information of electroencephalography (EEG) signals. The model uses\nthree-dimensional (3D) kernels to facilitate the feature extraction over the\nthree dimensions. The application of multiscale dilated convolution enables the\n3D kernel to have more flexible receptive fields. The proposed CNN model is\nevaluated with the CHB-MIT EEG database, the experimental results indicate that\nour model outperforms the existing state-of-the-art, achieves 80.5% accuracy,\n85.8% sensitivity and 75.1% specificity.\n"}
{"abstract": "  The phenomenon of adversarial examples illustrates one of the most basic\nvulnerabilities of deep neural networks. Among the variety of techniques\nintroduced to surmount this inherent weakness, adversarial training has emerged\nas the most effective strategy to achieve robustness. Typically, this is\nachieved by balancing robust and natural objectives. In this work, we aim to\nfurther optimize the trade-off between robust and standard accuracy by\nenforcing a domain-invariant feature representation. We present a new\nadversarial training method, Domain Invariant Adversarial Learning (DIAL),\nwhich learns a feature representation that is both robust and domain invariant.\nDIAL uses a variant of Domain Adversarial Neural Network (DANN) on the natural\ndomain and its corresponding adversarial domain. In the case where the source\ndomain consists of natural examples and the target domain is the adversarially\nperturbed examples, our method learns a feature representation constrained not\nto discriminate between the natural and adversarial examples, and can therefore\nachieve a more robust representation. Our experiments indicate that our method\nimproves both robustness and standard accuracy, when compared to other\nstate-of-the-art adversarial training methods.\n"}
{"abstract": "  In this paper we develop a stochastic analysis for marked binomial processes,\nthat can be viewed as the discrete analogues of marked Poisson processes. The\nstarting point is the statement of a chaotic expansion for square-integrable\n(marked binomial) functionals, prior to the elaboration of a Markov-Malliavin\nstructure within this framework. We take advantage of the new formalism to deal\nwith two main applications. First, we revisit the Chen-Stein method for the\n(compound) Poisson approximation which we perform in the paradigm of the built\nMarkov-Malliavin structure, before studying in the second one the problem of\nportfolio optimisation in the trinomial model.\n"}
{"abstract": "  Ehrenfest urns with interaction that are connected in a ring is considered as\na paradigm model for non-equilibrium thermodynamics and is shown to exhibit two\ndistinct non-equilibrium steady states (NESS) of uniform and non-uniform\nparticle distributions. As the inter-particle attraction varies, a first order\nnon-equilibrium phase transition occurs between these two NESSs characterized\nby a coexistence regime. The phase boundaries, the NESS particle distributions\nnear saddle points and the associated particle fluxes, average urn population\nfractions, and the relaxational dynamics to the NESSs are obtained analytically\nand verified numerically. A generalized non-equilibrium thermodynamics law is\nalso obtained, which explicitly identifies the heat, work, energy and entropy\nof the system.\n"}
{"abstract": "  In this article, we provide a review of the current state of the research of\nthe black hole shadow, focusing on analytical (as opposed to numerical and\nobservational) studies. We start with particular attention to the definition of\nthe shadow and its relation to the often used concepts of escape cone, critical\nimpact parameter and particle cross-section. For methodological purposes, we\npresent the derivation of the angular size of the shadow for an arbitrary\nspherically symmetric and static space-time, which allows one to calculate the\nshadow for an observer at arbitrary distance from the center. Then we discuss\nthe calculation of the shadow of a Kerr black hole, for an observer anywhere\noutside of the black hole. For observers at large distances we present and\ncompare two methods used in the literature. Special attention is given to\ncalculating the shadow in space-times which are not asymptotically flat.\nShadows of wormholes and other black-hole impostors are reviewed. Then we\ndiscuss the calculation of the black hole shadow in an expanding universe as\nseen by a comoving observer. The influence of a plasma on the shadow of a black\nhole is also considered.\n"}
{"abstract": "  For indications where only unstable reference treatments are available and\nuse of placebo is ethically justified, three-arm `gold standard' designs with\nan experimental, reference and placebo arm are recommended for non-inferiority\ntrials. In such designs, the demonstration of efficacy of the reference or\nexperimental treatment is a requirement. They have the disadvantage that only\nlittle can be concluded from the trial if the reference fails to be\nefficacious. To overcome this, we investigate novel single-stage, adaptive test\nstrategies where non-inferiority is tested only if the reference shows\nsufficient efficacy and otherwise $\\delta$-superiority of the experimental\ntreatment over placebo is tested. With a properly chosen superiority margin,\n$\\delta$-superiority indirectly shows non-inferiority. We optimize the sample\nsize for several decision rules and find that the natural, data driven test\nstrategy, which tests non-inferiority if the reference's efficacy test is\nsignificant, leads to the smallest overall and placebo sample sizes. We proof\nthat under specific constraints on the sample sizes, this procedure controls\nthe family-wise error rate. All optimal sample sizes are found to meet this\nconstraint. We finally show how to account for a relevant placebo drop-out rate\nin an efficient way and apply the new test strategy to a real life data set.\n"}
{"abstract": "  Camera trapping is increasingly used to monitor wildlife, but this technology\ntypically requires extensive data annotation. Recently, deep learning has\nsignificantly advanced automatic wildlife recognition. However, current methods\nare hampered by a dependence on large static data sets when wildlife data is\nintrinsically dynamic and involves long-tailed distributions. These two\ndrawbacks can be overcome through a hybrid combination of machine learning and\nhumans in the loop. Our proposed iterative human and automated identification\napproach is capable of learning from wildlife imagery data with a long-tailed\ndistribution. Additionally, it includes self-updating learning that facilitates\ncapturing the community dynamics of rapidly changing natural systems. Extensive\nexperiments show that our approach can achieve a ~90% accuracy employing only\n~20% of the human annotations of existing approaches. Our synergistic\ncollaboration of humans and machines transforms deep learning from a relatively\ninefficient post-annotation tool to a collaborative on-going annotation tool\nthat vastly relieves the burden of human annotation and enables efficient and\nconstant model updates.\n"}
{"abstract": "  Model compression techniques allow to significantly reduce the computational\ncost associated with data processing by deep neural networks with only a minor\ndecrease in average accuracy. Simultaneously, reducing the model size may have\na large effect on noisy cases or objects belonging to less frequent classes. It\nis a crucial problem from the perspective of the models' safety, especially for\nobject detection in the autonomous driving setting, which is considered in this\nwork. It was shown in the paper that the sensitivity of compressed models to\ndifferent distortion types is nuanced, and some of the corruptions are heavily\nimpacted by the compression methods (i.e., additive noise), while others (blur\neffect) are only slightly affected. A common way to improve the robustness of\nmodels is to use data augmentation, which was confirmed to positively affect\nmodels' robustness, also for highly compressed models. It was further shown\nthat while data imbalance methods brought only a slight increase in accuracy\nfor the baseline model (without compression), the impact was more striking at\nhigher compression rates for the structured pruning. Finally, methods for\nhandling data imbalance brought a significant improvement of the pruned models'\nworst-detected class accuracy.\n"}
{"abstract": "  In this paper, we consider mixed-integer nonsmooth constrained optimization\nproblems whose objective/constraint functions are available only as the output\nof a black-box zeroth-order oracle (i.e., an oracle that does not provide\nderivative information) and we propose a new derivative-free linesearch-based\nalgorithmic framework to suitably handle those problems. We first describe a\nscheme for bound constrained problems that combines a dense sequence of\ndirections (to handle the nonsmoothness of the objective function) with\nprimitive directions (to handle discrete variables). Then, we embed an exact\npenalty approach in the scheme to suitably manage nonlinear (possibly\nnonsmooth) constraints. We analyze the global convergence properties of the\nproposed algorithms toward stationary points and we report the results of an\nextensive numerical experience on a set of mixed-integer test problems.\n"}
{"abstract": "  As the use of spectral/$hp$ element methods, and high-order finite element\nmethods in general, continues to spread, community efforts to create efficient,\noptimized algorithms associated with fundamental high-order operations have\ngrown. Core tasks such as solution expansion evaluation at quadrature points,\nstiffness and mass matrix generation, and matrix assembly have received\ntremendousattention. With the expansion of the types of problems to which\nhigh-order methods are applied, and correspondingly the growth in types of\nnumerical tasks accomplished through high-order methods, the number and types\nof these core operations broaden. This work focuses on solution expansion\nevaluation at arbitrary points within an element. This operation is core to\nmany postprocessing applications such as evaluation of streamlines and\npathlines, as well as to field projection techniques such as mortaring. We\nexpand barycentric interpolation techniques developed on an interval to 2D\n(triangles and quadrilaterals) and 3D (tetrahedra, prisms, pyramids, and\nhexahedra) spectral/$hp$ element methods. We provide efficient algorithms for\ntheir implementations, and demonstrate their effectiveness using the\nspectral/$hp$ element library Nektar++.\n"}
{"abstract": "  Infrared absorption spectroscopy study of endohedral water molecule in a\nsolid mixture of H$_2$O@C$_{60}$ and C$_{60}$ was carried out at liquid helium\ntemperature. From the evolution of the spectra during the ortho-para conversion\nprocess, the spectral lines were identified as para- and ortho-water\ntransitions. Eight vibrational transitions with rotational side peaks were\nobserved in the mid-infrared: $\\omega_1$, $\\omega_2$, $\\omega_3$, $2\\omega_1$,\n$2\\omega_2$, $\\omega_1 +\\omega_3$, $\\omega_2 +\\omega_3$, and\n$2\\omega_2+\\omega_3$. The vibrational frequencies $\\omega_2$ and 2$\\omega_2$\nare lower by 1.6\\% and the rest by 2.4\\%, as compared to free \\water/. A model\nconsisting of a rovibrational Hamiltonian with the dipole and quadrupole\nmoments of water interacting with the crystal field was used to fit the\ninfrared absorption spectra. The electric quadrupole interaction with the\ncrystal field lifts the degeneracy of the rotational levels. The finite\namplitudes of the pure $v_1$ and $v_2$ vibrational transitions are consistent\nwith the interaction of the water molecule dipole moment with a lattice-induced\nelectric field. The permanent dipole moment of encapsulated \\water/ is found to\nbe $0.5\\pm 0.1$ D as determined from the far-infrared rotational line\nintensities. The translational mode of the quantized center of mass motion of\n\\water/ in the molecular cage of C$_{60}$ was observed at 110cm$^{-1}$\n(13.6meV).\n"}
{"abstract": "  We systematically develop a theory of stratification in the context of tensor\ntriangular geometry and apply it to classify the localizing tensor-ideals of\ncertain categories of spectral $G$-Mackey functors for all finite groups $G$.\nOur theory of stratification is based on the approach of Stevenson which uses\nthe Balmer-Favi notion of big support for tensor-triangulated categories whose\nBalmer spectrum is weakly noetherian. We clarify the role of the\nlocal-to-global principle and establish that the Balmer-Favi notion of support\nprovides the universal approach to weakly noetherian stratification. This\nprovides a uniform new perspective on existing classifications in the\nliterature and clarifies the relation with the theory of Benson-Iyengar-Krause.\nOur systematic development of this approach to stratification, involving a\nreduction to local categories and the ability to pass through finite \\'{e}tale\nextensions, may be of independent interest. Moreover, we strengthen the\nrelationship between stratification and the telescope conjecture. The starting\npoint for our equivariant applications is the recent computation by\nPatchkoria-Sanders-Wimmer of the Balmer spectrum of the category of derived\nMackey functors, which was found to capture precisely the height $0$ and height\n$\\infty$ chromatic layers of the spectrum of the equivariant stable homotopy\ncategory. We similarly study the Balmer spectrum of the category of\n$E(n)$-local spectral Mackey functors noting that it bijects onto the height\n$\\le n$ chromatic layers of the spectrum of the equivariant stable homotopy\ncategory; conjecturally the topologies coincide. Despite our incomplete\nknowledge of the topology of the Balmer spectrum, we are able to completely\nclassify the localizing tensor-ideals of these categories of spectral Mackey\nfunctors.\n"}
{"abstract": "  In this paper, we use large scale structure observations to test the cosmic\ndistance duality relation (CDDR), $D_{\\rm L}(1+z)^{-2}/D_{\\rm A}=\\eta=1 $, with\n$D_{\\rm L}$ and $D_{\\rm A}$, being the luminosity and angular diameter\ndistances, respectively. In order to perform the test, the following data set\nare considered: strong lensing systems and galaxy cluster measurements (gas\nmass fractions). No specific cosmological model is adopted, only a flat\nuniverse is assumed. By considering two $\\eta(z)$ parametrizations, we obtained\nthe validity of the CDDR within $1.5\\sigma$ which is in full agreement with\nother recent tests involving cosmological data. It is worth to comment that our\nresults are independent of the baryon budget of galaxy clusters.\n"}
{"abstract": "  Target networks are at the core of recent success in Reinforcement Learning.\nThey stabilize the training by using old parameters to estimate the $Q$-values,\nbut this also limits the propagation of newly-encountered rewards which could\nultimately slow down the training. In this work, we propose an alternative\ntraining method based on functional regularization which does not have this\ndeficiency. Unlike target networks, our method uses up-to-date parameters to\nestimate the target $Q$-values, thereby speeding up training while maintaining\nstability. Surprisingly, in some cases, we can show that target networks are a\nspecial, restricted type of functional regularizers. Using this approach, we\nshow empirical improvements in sample efficiency and performance across a range\nof Atari and simulated robotics environments.\n"}
{"abstract": "  Intermediate-mass black holes (IMBHs) span the approximate mass range\n$100$--$10^5\\,M_\\odot$, between black holes (BHs) formed by stellar collapse\nand the supermassive BHs at the centers of galaxies. Mergers of IMBH binaries\nare the most energetic gravitational-wave sources accessible by the terrestrial\ndetector network. Searches of the first two observing runs of Advanced LIGO and\nAdvanced Virgo did not yield any significant IMBH binary signals. In the third\nobserving run (O3), the increased network sensitivity enabled the detection of\nGW190521, a signal consistent with a binary merger of mass $\\sim\n150\\,M_\\odot\\,$ providing direct evidence of IMBH formation. Here we report on\na dedicated search of O3 data for further IMBH binary mergers, combining both\nmodelled (matched filter) and model independent search methods. We find some\nmarginal candidates, but none are sufficiently significant to indicate\ndetection of further IMBH mergers. We quantify the sensitivity of the\nindividual search methods and of the combined search using a suite of IMBH\nbinary signals obtained via numerical relativity, including the effects of\nspins misaligned with the binary orbital axis, and present the resulting upper\nlimits on astrophysical merger rates. Our most stringent limit is for equal\nmass and aligned spin BH binary of total mass $200\\,M_\\odot$ and effective\naligned spin 0.8 at $0.056\\,Gpc^{-3} yr^{-1}$ (90 $\\%$ confidence), a factor of\n3.5 more constraining than previous LIGO-Virgo limits. We also update the\nestimated rate of mergers similar to GW190521 to $0.08\\, Gpc^{-3}yr^{-1}$.\n"}
{"abstract": "  User response prediction, which aims to predict the probability that a user\nwill provide a predefined positive response in a given context such as clicking\non an ad or purchasing an item, is crucial to many industrial applications such\nas online advertising, recommender systems, and search ranking. However, due to\nthe high dimensionality and super sparsity of the data collected in these\ntasks, handcrafting cross features is inevitably time expensive. Prior studies\nin predicting user response leveraged the feature interactions by enhancing\nfeature vectors with products of features to model second-order or high-order\ncross features, either explicitly or implicitly. Nevertheless, these existing\nmethods can be hindered by not learning sufficient cross features due to model\narchitecture limitations or modeling all high-order feature interactions with\nequal weights. This work aims to fill this gap by proposing a novel\narchitecture Deep Cross Attentional Product Network (DCAP), which keeps cross\nnetwork's benefits in modeling high-order feature interactions explicitly at\nthe vector-wise level. Beyond that, it can differentiate the importance of\ndifferent cross features in each network layer inspired by the multi-head\nattention mechanism and Product Neural Network (PNN), allowing practitioners to\nperform a more in-depth analysis of user behaviors. Additionally, our proposed\nmodel can be easily implemented and train in parallel. We conduct comprehensive\nexperiments on three real-world datasets. The results have robustly\ndemonstrated that our proposed model DCAP achieves superior prediction\nperformance compared with the state-of-the-art models. Public codes are\navailable at https://github.com/zachstarkk/DCAP.\n"}
{"abstract": "  Continuous integration of renewable energy sources into power networks is\ncausing a paradigm shift in energy generation and distribution with regards to\ntrading and control; the intermittent nature of renewable sources affects\npricing of energy sold or purchased; the networks are subject to operational\nconstraints, voltage limits at each node, rated capacities for the power\nelectronic devices, current bounds for distribution lines. These economic and\ntechnical constraints coupled with intermittent renewable injection may pose a\nthreat to system stability and performance. We propose a novel holistic\napproach to energy trading composed of a distributed predictive control\nframework to handle physical interactions, i,e., voltage constraints and power\ndispatch, together with a negotiation framework to determine pricing policies\nfor energy transactions. We study the effect of forecasting generation and\nconsumption on the overall network's performance and market behaviours. We\nprovide a rigorous convergence analysis for both the negotiation framework and\nthe distributed control. Lastly, we assess the impact of forecasting in the\nproposed system with the aid of testing scenarios.\n"}
{"abstract": "  Word embeddings trained on large corpora have shown to encode high levels of\nunfair discriminatory gender, racial, religious and ethnic biases.\n  In contrast, human-written dictionaries describe the meanings of words in a\nconcise, objective and an unbiased manner.\n  We propose a method for debiasing pre-trained word embeddings using\ndictionaries, without requiring access to the original training resources or\nany knowledge regarding the word embedding algorithms used.\n  Unlike prior work, our proposed method does not require the types of biases\nto be pre-defined in the form of word lists, and learns the constraints that\nmust be satisfied by unbiased word embeddings automatically from dictionary\ndefinitions of the words.\n  Specifically, we learn an encoder to generate a debiased version of an input\nword embedding such that it\n  (a) retains the semantics of the pre-trained word embeddings,\n  (b) agrees with the unbiased definition of the word according to the\ndictionary, and\n  (c) remains orthogonal to the vector space spanned by any biased basis\nvectors in the pre-trained word embedding space.\n  Experimental results on standard benchmark datasets show that the proposed\nmethod can accurately remove unfair biases encoded in pre-trained word\nembeddings, while preserving useful semantics.\n"}
{"abstract": "  Design erosion is a persistent problem within the software engineering\ndiscipline. Software designs tend to deteriorate over time and there is a need\nfor tools and techniques that support software architects when dealing with\nlegacy systems. This paper presents an evaluation of a Search Based Software\nEngineering (SBSE) approach intended to recover high-level architecture designs\nof software systems by structuring low-level artefacts into high-level\narchitecture artefact configurations. In particular , this paper describes the\nperformance evaluation of a number of metaheuristic search algorithms applied\nto architecture reconstruction problems with high dimensionality in terms of\nobjectives. These problems have been selected as representative of the typical\nchallenges faced by software architects dealing with legacy systems and the\nresults inform the ongoing developed of a software tool that supports the\nanalysis of trade-offs between different reconstructed architectures.\n"}
{"abstract": "  This paper develops a generalization of the Hansen-Jagannathan bound that\nincorporates information beyond the mean and variance of returns. The resulting\nbound compares the physical and risk neutral distribution for every\n$\\tau$-quantile, where $\\tau \\in (0,1)$. An empirical application with S\\&P500\nreturn data shows that the new bound is stronger than the Hansen-Jagannathan\nbound for small values of $\\tau$. The long run risk model cannot reconcile this\nfeature of the data, due to the absence of disaster risk. I extend this finding\nusing conditioning information and document that disaster risk is time-varying,\nusing a semiparametric approach. I also propose a new measure of quantile\nforecastability and show that many stylized facts about the equity premium\ncarry over to the quantile setting.\n"}
{"abstract": "  The use of an infinity of fluctuating paths of least time that are compatible\nwith the quantum mechanics indeterminacy provides a new interpretation in\ngeometrical optic of the interference pattern of Young's double slit\nexperiment, which suggests that the wave behavior of matter and radiation is\ndictated by the space-time geodesics. Moreover, the association of a wave\nfunction to each path of least time as a probability amplitude together with an\nuncertainty for momentum and position allows to derive the Schrodinger's\nequation starting from the geodesic's characteristics. A new insight is\nobtained regarding the Van Der Waals torque as well as Casimir\nattraction/repulsion mechanism.\n"}
{"abstract": "  Bayesian nonparametric estimates of Australian mental health distributions\nare obtained to assess how the mental health status of the population has\nchanged over time and to compare the mental health status of female/male and\nindigenous/non-indigenous population subgroups. First- and second-order\nstochastic dominance are used to compare distributions, with results presented\nin terms of the posterior probability of dominance and the posterior\nprobability of no dominance. Our results suggest mental health has deteriorated\nin recent years, that males mental health status is better than that of\nfemales, and non-indigenous health status is better than that of the indigenous\npopulation.\n"}
{"abstract": "  We study the possibility to use line-intensity mapping (LIM) to seek photons\nfrom the radiative decay of neutrinos in the cosmic neutrino background. The\nStandard Model prediction for the rate for these decays is extremely small, but\nit can be enhanced if new physics increases the neutrino electromagnetic\nmoments. The decay photons will appear as an interloper of astrophysical\nspectral lines. We propose that the neutrino-decay line can be identified with\nanisotropies in LIM clustering and also with the voxel intensity distribution.\nOngoing and future LIM experiments will have -- depending on the neutrino\nhierarchy, transition and experiment considered -- a sensitivity to an\neffective electromagnetic transition moment $\\sim 10^{-12}\\, -\\,10^{-8}\\,\n(m_ic^2/{0.1 \\rm eV})^{3/2}\\mu_{\\rm B}$, where $m_i$ is the mass of the\ndecaying neutrino and $\\mu_{\\rm B}$ is the Bohr magneton. This will be\nsignificantly more sensitive than cosmic microwave background spectral\ndistortions, and it will be competitive with stellar cooling studies. As a\nbyproduct, we also report an analytic form of the one-point probability\ndistribution function for neutrino-density fluctuations, obtained from the\nQuijote simulations using symbolic regression.\n"}
{"abstract": "  Traditionally, the main focus of image super-resolution techniques is on\nrecovering the most likely high-quality images from low-quality images, using a\none-to-one low- to high-resolution mapping. Proceeding that way, we ignore the\nfact that there are generally many valid versions of high-resolution images\nthat map to a given low-resolution image. We are tackling in this work the\nproblem of obtaining different high-resolution versions from the same\nlow-resolution image using Generative Adversarial Models. Our learning approach\nmakes use of high frequencies available in the training high-resolution images\nfor preserving and exploring in an unsupervised manner the structural\ninformation available within these images. Experimental results on the CelebA\ndataset confirm the effectiveness of the proposed method, which allows the\ngeneration of both realistic and diverse high-resolution images from\nlow-resolution images.\n"}
{"abstract": "  Magnetic Resonance Imaging (MRI) is a valuable clinical diagnostic modality\nfor spine pathologies with excellent characterization for infection, tumor,\ndegenerations, fractures and herniations. However in surgery, image-guided\nspinal procedures continue to rely on CT and fluoroscopy, as MRI slice\nresolutions are typically insufficient. Building upon state-of-the-art single\nimage super-resolution, we propose a reference-based, unpaired multi-contrast\ntexture-transfer strategy for deep learning based in-plane and across-plane MRI\nsuper-resolution. We use the scattering transform to relate the texture\nfeatures of image patches to unpaired reference image patches, and additionally\na loss term for multi-contrast texture. We apply our scheme in different\nsuper-resolution architectures, observing improvement in PSNR and SSIM for 4x\nsuper-resolution in most of the cases.\n"}
{"abstract": "  The Riccati equation method is used to establish an oscillatory and a non\noscillatory criteria for nonhomogeneous linear systems of two first-order\nordinary differential equations. It is shown that the obtained oscillatory\ncriterion is a generalization of J. S. W. Wong's oscillatory criterion.\n"}
{"abstract": "  We investigate the quantum dynamics of a spin chain that implements a quantum\nanalog of Conway's game of life. We solve the time-dependent Schr\\\"odinger\nequation starting with initial separable states and analyse the evolution of\nquantum correlations across the lattice. We report examples of evolutions\nleading to all-entangled chains and/or to time oscillating entangling\nstructures and characterize them by means of entanglement and network measures.\nThe quantum patterns result to be quite different from the classical ones, even\nin the dynamics of local observables. A peculiar instance is a structure\nbehaving as the quantum analog of a blinker, but that has no classical\ncounterpart.\n"}
{"abstract": "  Many complex activities of production cycles, such as quality control or\nfault analysis, require highly experienced specialists to perform various\noperations on (semi)finished products using different tools. In practical\nscenarios, the selection of a next operation is complicated, since each expert\nhas only a local view on the total set of operations to be performed. As a\nresult, decisions made by the specialists are suboptimal and might cause\nsignificant costs. In this paper, we consider a Multi-resource Partial-ordering\nFlexible Job-shop Scheduling (MPF-JSS) problem where partially-ordered\nsequences of operations must be scheduled on multiple required resources, such\nas tools and specialists. The resources are flexible and can perform one or\nmore operations depending on their properties. The problem is modeled using\nAnswer Set Programming (ASP) in which the time assignments are efficiently done\nusing Difference Logic. Moreover, we suggest two multi-shot solving strategies\naiming at the identification of the time bounds allowing for a solution of the\nschedule optimization problem. Experiments conducted on a set of instances\nextracted from a medium-sized semiconductor fault analysis lab indicate that\nour approach can find schedules for 87 out of 91 considered real-world\ninstances.\n"}
{"abstract": "  Electrons and ions trapped with electromagnetic fields have long served as\nimportant high-precision metrological instruments, and more recently have also\nbeen proposed as a platform for quantum information processing. Here we point\nout that these systems can also be used as highly sensitive detectors of\npassing charged particles, due to the combination of their extreme\ncharge-to-mass ratio and low-noise quantum readout and control. In particular,\nthese systems can be used to detect energy depositions many orders of magnitude\nbelow typical ionization scales. As illustrations, we suggest some applications\nin particle physics. We outline a non-destructive time-of-flight measurement\ncapable of sub-eV energy resolution for slowly moving, collimated particles. We\nalso show that current devices can be used to provide competitive sensitivity\nto models where ambient dark matter particles carry small electric millicharges\n$\\ll e$. Our calculations may also be useful in the characterization of noise\nin quantum computers coming from backgrounds of charged particles.\n"}
{"abstract": "  Turbulence generally arises in shear flows if velocities and hence inertial\nforces are sufficiently large. In striking contrast, viscoelastic fluids can\nexhibit disordered motion even at vanishing inertia. Intermediate between these\ncases, a novel state of chaotic motion, `elasto-inertial turbulence' (EIT), has\nbeen observed in a narrow Reynolds number interval. We here determine the\norigin of EIT in experiments and show that characteristic EIT structures can be\ndetected across an unexpectedly wide range of parameters. Close to onset a\npattern of chevron shaped streaks emerges in excellent agreement with linear\ntheory. However, the instability can be traced to far lower Reynolds numbers\nthan permitted by theory. For increasing inertia, a secondary instability gives\nrise to a wall mode composed of inclined near wall streaks and shear layers.\nThis mode persists to what is known as the `maximum drag reduction limit' and\noverall EIT is found to dominate viscoelastic flows across more than three\norders of magnitude in Reynolds number.\n"}
{"abstract": "  Edge computing is emerging as a new paradigm to allow processing data at the\nedge of the network, where data is typically generated and collected, by\nexploiting multiple devices at the edge collectively. However, exploiting the\npotential of edge computing is challenging mainly due to the heterogeneous and\ntime-varying nature of edge devices. Coded computation, which advocates mixing\ndata in sub-tasks by employing erasure codes and offloading these sub-tasks to\nother devices for computation, is recently gaining interest, thanks to its\nhigher reliability, smaller delay, and lower communication cost. In this paper,\nour focus is on characterizing the cost-benefit trade-offs of coded computation\nfor practical edge computing systems, and develop an adaptive coded computation\nframework. In particular, we focus on matrix multiplication as a\ncomputationally intensive task, and develop an adaptive coding for matrix\nmultiplication (ACM^2) algorithm by taking into account the heterogeneous and\ntime varying nature of edge devices. ACM^2 dynamically selects the best coding\npolicy by taking into account the computing time, storage requirements as well\nas successful decoding probability. We show that ACM^2 improves the task\ncompletion delay significantly as compared to existing coded matrix\nmultiplication algorithms.\n"}
{"abstract": "  We construct stable Adams operations on the cohomology theory Tmf of\ntopological modular forms. This is done by extending the sheaf defining Tmf to\nbe functorial with respect to isogenies of generalised elliptic curves of\ninvertible degree. These Adams operations are then calculated on the homotopy\ngroups of Tmf using a combination of descent spectral sequences, the theory of\nsynthetic spectra, and Anderson duality.\n"}
{"abstract": "  We develop a sophisticated model of FRB observations, accounting for the\nintrinsic cosmological gas distribution and host galaxy contributions, and give\nthe most detailed account yet of observational biases due to burst width,\ndispersion measure, and the exact telescope beamshape. Our results offer a\nsignificant increase in both accuracy and precision beyond those previously\nobtained. Using results from ASKAP and Parkes, we present our best-fit FRB\npopulation parameters in a companion paper. Here, we consider in detail the\nexpected and fitted distributions in redshift, dispersion measure, and\nsignal-to-noise. We estimate that the unlocalised ASKAP FRBs arise from\n$z<0.5$, with between a third and a half within $z<0.1$. Our predicted\nsource-counts (\"logN--logS\") distribution confirms previous indications of a\nsteepening index near the Parkes detection threshold of $1$\\,Jy\\,ms. We find no\nevidence for a minimum FRB energy, and rule out $E_{\\rm min} > 10^{38.5}$\\,erg\nat 90\\% C.L. Importantly, we find that above a certain DM, observational biases\ncause the Macquart (DM--z) relation to become inverted, implying that the\nhighest-DM events detected in the unlocalised Parkes and ASKAP samples are\nunlikely to be the most distant. We do not expect our quantitative estimates in\nthis region to be accurate until it is directly probed with localised FRBs.\nSince the cause of this effect is a well-understood observational bias however,\nit is guaranteed to be present to some degree. Works assuming a 1--1 DM--z\nrelation may therefore derive erroneous results.\n"}
{"abstract": "  Background: Studies on intravoxel incoherent motion (IVIM) imaging are\ncarried out with different acquisition protocols. Purpose: Investigate the\ndependence of IVIM parameters on the B_0field strength when using a bi- or\ntriexponential model. Field Strength/Sequence: 20 volunteers were examined at\ntwo field strengths (1.5 and 3T). Diffusion-weighted images of the abdomen were\nacquired at 24 b-values ranging from 0.2 to 500 s/mm2. Assessment: ROIs were\nmanually drawn in the liver. Data were fitted with a bi- and a triexponential\nIVIM model. Resulting parameters were compared between both field strengths.\nResults: At b-values below 6s/mm2, the triexponential model provided better\nagreement with the data than the biexponential model. The average tissue\ndiffusivity was D=1.22/1.00 um/ms at 1.5/3T. The average pseudo-diffusion\ncoefficients for the biexponential model were D*=308/260um/ms at 1.5/3T; and\nfor the triexponential model D1*=81.3/65.9um/ms , D2*=2453/2333um/ms at 1.5/3T.\nThe average perfusion fractions for the biexponential model were f=0.286/0.303\nat 1.5/3T; and for the triexponential model f1=0.161/0.174 and f2=0.152/0.159\nat 1.5/3T. A significant B0 dependence was only found for the biexponential\npseudo-diffusion coefficient (ANOVA/KW p=0.037/0.0453) and tissue diffusivity\n(ANOVA/KW: p<0.001). Conclusion: Our experimental results suggest that\ntriexponential pseudo-diffusion coefficients and perfusion fractions obtained\nat different field strengths could be compared across different studies using\ndifferent B0. However, it is recommendable to take the field strength into\naccount when comparing tissue diffusivities or using the biexponential IVIM\nmodel. Considering published values for oxygenation-dependent transversal\nrelaxation times of blood, it is unlikely that the two blood compartments of\nthe triexponential model represent venous and arterial blood.\n"}
{"abstract": "  Over the last few decades, deforestation and climate change have caused\nincreasing number of forest fires. In Southeast Asia, Indonesia has been the\nmost affected country by tropical peatland forest fires. These fires have a\nsignificant impact on the climate resulting in extensive health, social and\neconomic issues. Existing forest fire prediction systems, such as the Canadian\nForest Fire Danger Rating System, are based on handcrafted features and require\ninstallation and maintenance of expensive instruments on the ground, which can\nbe a challenge for developing countries such as Indonesia. We propose a novel,\ncost-effective, machine-learning based approach that uses remote sensing data\nto predict forest fires in Indonesia. Our prediction model achieves more than\n0.81 area under the receiver operator characteristic (ROC) curve, performing\nsignificantly better than the baseline approach which never exceeds 0.70 area\nunder ROC curve on the same tasks. Our model's performance remained above 0.81\narea under ROC curve even when evaluated with reduced data. The results support\nour claim that machine-learning based approaches can lead to reliable and\ncost-effective forest fire prediction systems.\n"}
{"abstract": "  Recent high pressure experiments discovered abnormal double-dome\nsuperconductivities in the newly-synthesized kagome materials $A$V$_3$Sb$_5$\n($A$ = K, Rb, Cs), which also host abundant emergent quantum phenomena such as\ncharge density wave (CDW), anomalous Hall effect, nontrivial topological\nproperty, etc. In this work, by using first-principles electronic structure\ncalculations, we have studied the CDW state, superconductivity, and topological\nproperty in CsV$_3$Sb$_5$ under pressures ($<$ 50 GPa). Based on the\nelectron-phonon coupling theory, our calculated superconducting $T_\\text{c}$s\nare consistent with the observed ones in the second superconducting dome at\nhigh pressure, but are much higher than the measured values at low pressure.\nThe further calculations including the Hubbard U indicate that with modest\nelectron-electron correlation the magnetism on the V atoms exists at low\npressure and diminishes gradually at high pressure. We thus propose that the\nexperimentally observed superconductivity in CsV$_3$Sb$_5$ at ambient/low\npressures may still belong to the conventional Bardeen-Cooper-Schrieffer (BCS)\ntype but is partially suppressed by the V magnetism, while the\nsuperconductivity under high pressure is fully conventional without invoking\nthe magnetism. We also predict that there are a second weak CDW state and\ntopological phase transitions in CsV$_3$Sb$_5$ under pressures. Our theoretical\nassertion calls for future experimental examination.\n"}
{"abstract": "  In the last decades, the interest to understand the connection between brain\nand body has grown notably. For example, in psychoneuroimmunology many studies\nassociate stress, arising from many different sources and situations, to\nchanges in the immune system from the medical or immunological point of view as\nwell as from the biochemical one. In this paper we identify important\nbehaviours of this interplay between the immune system and stress from medical\nstudies and seek to represent them qualitatively in a paradigmatic, yet simple,\nmathematical model. To that end we develop a differential equation model with\ntwo equations for infection level and immune system, which integrates the\neffects of stress as an additional parameter. We are able to reproduce a stable\nhealthy state for little stress, an oscillatory state between healthy and\ninfected states for high stress, and a \"burn-out\" or stable sick state for\nextremely high stress. The mechanism between the different dynamics is\ncontrolled by two saddle-node in cycle (SNIC) bifurcations. Furthermore, our\nmodel is able to capture an induced infection upon dropping from moderate to\nlow stress, and it predicts increasing infection periods upon increasing before\neventually reaching a burn-out state.\n"}
{"abstract": "  For any storage ring-based large-scale scientific facility, one of the most\nimportant performance parameters is the dynamic aperture (DA), which measures\nthe motion stability of charged particles in a global manner. To date,\nlong-term tracking-based simulation is regarded as the most reliable method to\ncalculate DA. However, numerical tracking may become a significant issue,\nespecially when lots of candidate designs of a storage ring need to be\nevaluated. In this paper, we present a novel machine learning-based method,\nwhich can reduce the computation cost of DA tracking by approximately one order\nof magnitude, while keeping sufficiently high evaluation accuracy. Moreover, we\ndemonstrate that this method is independent of concrete physical models of a\nstorage ring. This method has the potential to be applied to similar problems\nof identifying irregular motions in other complex dynamical systems.\n"}
{"abstract": "  The effect of asymmetric functionally graded material on the edge resonance\nand the Fano resonance in semi-infinite FGM plates are reported in this work.\nThe edge resonance is weakened by the material perturbation and the complete\nmode conversion is illustrated with incident $S_0$ mode. The Fano resonance\noccurs on the reflected $A_0$ mode as a result of interference between the\nresonance and direct scattering with incident $A_0$ mode. A hybrid analytical\nmodel based on the collocation discretization and the modal decomposition of\nthe elastic field is developed to analyze the scattering properties of the\nsemi-infinite plates. The Fano line-shape is discussed in detail. The results\nshow that the Fano line shape is sensitive to the volume fraction, which is\nbeneficial for the quantitative guided wave application.\n"}
{"abstract": "  Temperature-dependent angle-resolved photoemission spectroscopy (ARPES) was\ncarried out on single-crystalline EuB6 samples. By measuring ARPES spectra in\nan extended Brillouin zone, a B 2p hole pocket centered at the X point is\nclearly observed, thus proving the semimetallic nature of EuB6. Below the Curie\ntemperature TC, ARPES spectra show two B 2p bands of which separation is due to\nan exchange interaction between local Eu 4f and itinerant B 2p electrons. The\nexchange splitting becomes smaller as the temperature increases and disappears\nwell above TC. Additionally, a diffuse structure near the Fermi level survives\njust above TC. Such behavior is well described by Monte Carlo simulations of a\nKondo lattice model, thus supporting the formation of magnetic polarons in\nEuB6, which accounts for the resistivity upturn near above TC when lowering the\ntemperature.\n"}
{"abstract": "  A Snyder model generated by the noncommutative coordinates and Lorentz\ngenerators close a Lie algebra. The application of the Heisenberg double\nconstruction is investigated for the Snyder coordinates and momenta generators.\nIt leads to the phase space of the Snyder model. Further, the extended Snyder\nalgebra is constructed by using the Lorentz algebra, in one dimension higher.\nThe dual pair of extended Snyder algebra and extended Snyder group is then\nformulated. Two Heisenberg doubles are considered, one with the conjugate\ntensorial momenta and another with the Lorentz matrices. Explicit formulae for\nall Heisenberg doubles are given.\n"}
{"abstract": "  In the field of network security, the concept of honeypots is well\nestablished in research as well as in production. Honeypots are used to imitate\na legitimate target on the network and to raise an alert on any interaction.\nThis does not only help learning about a breach, but also allows researchers to\nstudy the techniques of an attacker. With the rise of cloud computing,\ncontainer-based virtualization gained popularity for application deployment.\nThis paper investigates the possibilities of container-based honeypots and\nintroduces the concept of simulating container escapes as a deception\ntechnique.\n"}
{"abstract": "  Spontaneous parametric down conversion (PDC), in the perturbative limit, can\nbe considered as a probabilistic splitting of one input photon into two output\nphotons. Conversely, sum-frequency generation (SFG) implements the reverse\nprocess of combining two input photons into one. Here we show that a\nsingle-photon projective measurement in the temporal-mode basis of the output\nphoton of a two-photon SFG process effects a generalized measurement on the\ninput two-photon state. We describe the positive-operator-valued measure (POVM)\nassociated with such a measurement, and show that its elements are proportional\nto the two-photon states produced by the time-reversed PDC process. Such a\ndetection acts as a joint measurement on two photons, and is thus an important\ncomponent of many quantum information processing protocols relying on photonic\nentanglement. Using the retrodictive approach, we analyze the properties of the\ntwo-photon POVM that are relevant for quantum protocols exploiting two-photon\nstates and measurements.\n"}
{"abstract": "  Polar code lattices are formed from binary polar codes using Construction D.\nIn this paper, we propose a design technique for finite-dimension polar code\nlattices. The dimension $n$ and target probability of decoding error are\nparameters for this design. To select the rates of the Construction D component\ncodes, rather than using the capacity as in past work, we use the explicit\nfinite-length properties of the polar code. Under successive cancellation\ndecoding, density evolution allows choosing code rates that satisfy the equal\nerror probability rule. At an error-rate of $10^{-4}$, a dimension $n=128$\npolar code lattice achieves a VNR of 2.5 dB, within 0.2 dB of the best-known\nBCH code lattice, but with significantly lower decoding complexity.\n"}
{"abstract": "  We present Pylot, a platform for autonomous vehicle (AV) research and\ndevelopment, built with the goal to allow researchers to study the effects of\nthe latency and accuracy of their models and algorithms on the end-to-end\ndriving behavior of an AV. This is achieved through a modular structure enabled\nby our high-performance dataflow system that represents AV software pipeline\ncomponents (object detectors, motion planners, etc.) as a dataflow graph of\noperators which communicate on data streams using timestamped messages. Pylot\nreadily interfaces with popular AV simulators like CARLA, and is easily\ndeployable to real-world vehicles with minimal code changes.\n  To reduce the burden of developing an entire pipeline for evaluating a single\ncomponent, Pylot provides several state-of-the-art reference implementations\nfor the various components of an AV pipeline. Using these reference\nimplementations, a Pylot-based AV pipeline is able to drive a real vehicle, and\nattains a high score on the CARLA Autonomous Driving Challenge. We also present\nseveral case studies enabled by Pylot, including evidence of a need for\ncontext-dependent components, and per-component time allocation. Pylot is open\nsource, with the code available at https://github.com/erdos-project/pylot.\n"}
{"abstract": "  We summarize the results of temporal and spectral analysis of the X-ray\npulsar 2S 1553-542 using the Nuclear Spectroscopic Telescope Array (NuSTAR) and\nSwift during the outburst in January-February 2021. During the outburst, the\nspin period of the neutron star was $P = 9.2822\\pm 0.0001$ s based on NuSTAR\ndata. The temporal evolution of the spin period, pulse profile, and pulse\nfraction is studied during the outburst. The spectra of the source are studied\nfor different days of the outburst and can be well described by a model\nconsisting of -- a black body emission or a power law. We have investigated the\ninter-day evolution of different timing and spectral parameters during the\noutburst. The energy dependence of the pulse profile was studied to investigate\nthe evolution of the individual peaks and emission geometry of the pulsar with\na different energy. The pulse profile of the source shows strong single peak\nnature with a hump-like feature of relatively lower intensity and it evolves\nsignificantly with different energy ranges. The evolution of the pulse profile\nis studied during different phases of the outburst and the pulse fraction shows\na positive correlation with energy.\n"}
{"abstract": "  Since the renaissance of deep learning (DL), facial expression recognition\n(FER) has received a lot of interest, with continual improvement in the\nperformance. Hand-in-hand with performance, new challenges have come up. Modern\nFER systems deal with face images captured under uncontrolled conditions (also\ncalled in-the-wild scenario) including occlusions and pose variations. They\nsuccessfully handle such conditions using deep networks that come with various\ncomponents like transfer learning, attention mechanism and local-global context\nextractor. However, these deep networks are highly complex with large number of\nparameters, making them unfit to be deployed in real scenarios. Is it possible\nto build a light-weight network that can still show significantly good\nperformance on FER under in-the-wild scenario? In this work, we methodically\nbuild such a network and call it as Imponderous Net. We leverage on the\naforementioned components of deep networks for FER, and analyse, carefully\nchoose and fit them to arrive at Imponderous Net. Our Imponderous Net is a low\ncalorie net with only 1.45M parameters, which is almost 50x less than that of a\nstate-of-the-art (SOTA) architecture. Further, during inference, it can process\nat the real time rate of 40 frames per second (fps) in an intel-i7 cpu. Though\nit is low calorie, it is still power packed in its performance, overpowering\nother light-weight architectures and even few high capacity architectures.\nSpecifically, Imponderous Net reports 87.09\\%, 88.17\\% and 62.06\\% accuracies\non in-the-wild datasets RAFDB, FERPlus and AffectNet respectively. It also\nexhibits superior robustness under occlusions and pose variations in comparison\nto other light-weight architectures from the literature.\n"}
{"abstract": "  We investigate the operation of dual-gate reconfigurable field-effect\ntransistor (RFET) in the programgate at drain (PGAD) and program-gate at source\n(PGAS) configurations. To this end, dual-gate silicon nanowire (SiNW) FETs are\nfabricated based on anisotropic wet chemical silicon etching and nickel\nsilicidation yielding silicide-SiNW Schottky junctions at source and drain.\nWhereas in PGAD-configuration ambipolar operation is suppressed, switching is\ndeteriorated due to the injection through a Schottky-barrier. Operating the\nRFET in PGAS configuration yields a switching behavior close to a conventional\nMOSFET. This, howewer, needs to be traded off against strongly non-linear\noutput characteristics for small bias.\n"}
{"abstract": "  We present a brief overview of some of the analytic perturbative techniques\nfor the computation of the Floquet Hamiltonian for a periodically driven, or\nFloquet, quantum many-body system. The key technical points about each of the\nmethods discussed are presented in a pedagogical manner. They are followed by a\nbrief account of some chosen phenomena where these methods have provided useful\ninsights. We provide an extensive discussion of the Floquet-Magnus expansion,\nthe adiabatic-impulse approximation, and the Floquet perturbation theory. This\nis followed by a relatively short discourse on the rotating wave approximation,\na Floquet-Magnus resummation technique and the Hamiltonian flow method. We also\nprovide a discussion of some open problems which may possibly be addressed\nusing these methods.\n"}
{"abstract": "  We prove that, under some generic non-degeneracy assumptions, real analytic,\ncentrally symmetric plane domains are determined by their Dirichlet (resp.\nNeumann) spectra. We prove that the conditions are open-dense for real analytic\nconvex domains. The proof is parallel to the proof that up/down symmetric\ndomains are spectrally determined. One step is to use a Maslov index\ncalculation to show that the second derivative of the defining function of a\ncentrally symmetric domain at the endpoints of a bouncing ball orbit is a\nspectral invariant. This is also true for up/down symmetric domains, removing\nan assumption from the proof in that case.\n"}
{"abstract": "  We study random surfaces with a uniformly convex gradient interaction in the\npresence of quenched disorder taking the form of a random independent external\nfield. Previous work on the model has focused on proving existence and\nuniqueness of infinite-volume gradient Gibbs measures with a given tilt and on\nstudying the fluctuations of the surface and its discrete gradient.\n  In this work we focus on the convergence of the thermodynamic limit,\nestablishing convergence of the finite-volume distributions with Dirichlet\nboundary conditions to translation-covariant (gradient) Gibbs measures.\nSpecifically, it is shown that, when the law of the random field has finite\nsecond moment and is symmetric, the distribution of the gradient of the surface\nconverges in dimensions $d\\geq4$ while the distribution of the surface itself\nconverges in dimensions $d\\geq 5$. Moreover, a power-law upper bound on the\nrate of convergence in Wasserstein distance is obtained. The results partially\nanswer a question discussed by Cotar and K\\\"ulske\n"}
{"abstract": "  How do inter-organizational networks emerge? Accounting for interdependence\namong ties while studying tie formation is one of the key challenges in this\narea of research. We address this challenge using an equilibrium framework\nwhere firms' decisions to form links with other firms are modeled as a\nstrategic game. In this game, firms weigh the costs and benefits of\nestablishing a relationship with other firms and form ties if their net payoffs\nare positive. We characterize the equilibrium networks as exponential random\ngraphs (ERGM), and we estimate the firms' payoffs using a Bayesian approach. To\ndemonstrate the usefulness of our approach, we apply the framework to a\nco-investment network of venture capital firms in the medical device industry.\nThe equilibrium framework allows researchers to draw economic interpretation\nfrom parameter estimates of the ERGM Model. We learn that firms rely on their\njoint partners (transitivity) and prefer to form ties with firms similar to\nthemselves (homophily). These results hold after controlling for the\ninterdependence among ties. Another, critical advantage of a structural\napproach is that it allows us to simulate the effects of economic shocks or\npolicy counterfactuals. We test two such policy shocks, namely, firm entry and\nregulatory change. We show how new firms' entry or a regulatory shock of\nminimum capital requirements increase the co-investment network's density and\nclustering.\n"}
{"abstract": "  In this paper we propose coupling conditions for a relaxation model for\nvehicular traffic on networks. We present a matched asymptotic expansion\nprocedure to derive a LWR- network with well-known classical coupling\nconditions from the relaxation network in the macroscopic limit. Similar to the\nasymptotic limit of boundary value problems, we perform an asymptotic analysis\nof the interface layers at the nodes and a matching procedure using\nhalf-Riemann problems for the limit conservation law. Moreover, we present\nnumerical experiments comparing the relaxation network with the LWR network for\na broader range of coupling conditions.\n"}
{"abstract": "  Nuclear spins of noble gases can maintain coherence for hours at ambient\nconditions owing to their extraordinary isolation by the enclosing, complete\nelectronic shells. This isolation, however, impedes the ability to manipulate\nand control them by optical means or by physical coupling to other spin gases.\nHere we experimentally achieve strong coherent coupling between noble-gas spins\nand the optically-accessible spins of alkali-metal vapor. Stochastic\nspin-exchange collisions, underlying the coupling, accumulate to a coherent\nperiodic exchange of spin excitations between the two gases. We obtain a\ncoupling rate 10 times higher than the decay rate, observe the resultant\navoided crossing in the spectral response of the spins, and demonstrate the\nexternal control over the coupling by magnetic fields. These results open a\nroute for efficient and rapid interfacing with noble-gas spins for applications\nin quantum sensing and information.\n"}
{"abstract": "  In this article, we propose and investigate ML group explainers in a general\ngame-theoretic setting with the focus on coalitional game values and games\nbased on the conditional and marginal expectation of an ML model. The\nconditional game takes into account the joint distribution of the predictors,\nwhile the marginal game depends on the structure of the model. The objective of\nthe article is to unify the two points of view under predictor dependencies and\nto reduce the complexity of group explanations. To achieve this, we propose a\nfeature grouping technique that employs an information-theoretic measure of\ndependence and design appropriate groups explainers. Furthermore, in the\ncontext of coalitional game values with a two-step formulation, we introduce a\ntheoretical scheme that generates recursive coalitional game values under a\npartition tree structure and investigate the properties of the corresponding\ngroup explainers.\n"}
{"abstract": "  With the growing popularity of cloud storage, removing duplicated data across\nusers is getting more critical for service providers to reduce costs. Recently,\nData resemblance detection is a novel technology to detect redundancy among\nsimilarity. It extracts feature from each chunk content and treat chunks with\nhigh similarity as candidates for removing redundancy. However, popular\nresemblance methods such as \"N-transform\" and \"Finesse\" use only the chunk data\nfor feature extraction. A minor modification on the data chunk could seriously\ndeteriorate its capability for resemblance detection. In this paper, we\nproposes a novel chunk-context aware resemblance detection algorithm, called\nCARD, to mitigate this issue. CARD introduces a BP-Neural network-based\nchunk-context aware model, and uses N-sub-chunk shingles-based initial feature\nextraction strategy. It effectively integrates each data chunk content's\ninternal structure with the context information for feature extraction, the\nimpact of small changes in data chunks is significantly reduced. To evaluate\nits performance, we implement a CARD prototype and conduct extensive\nexperiments using real-world data sets. The results show that CARD can detect\nup to 75.03% more redundant data and accelerate the resemblance detection\noperations by 5.6 to 17.8 times faster compared with the state-of-the-art\nresemblance detection approaches.\n"}
{"abstract": "  Artificial Intelligence techniques powered by deep neural nets have achieved\nmuch success in several application domains, most significantly and notably in\nthe Computer Vision applications and Natural Language Processing tasks.\nSurpassing human-level performance propelled the research in the applications\nwhere different modalities amongst language, vision, sensory, text play an\nimportant role in accurate predictions and identification. Several multimodal\nfusion methods employing deep learning models are proposed in the literature.\nDespite their outstanding performance, the complex, opaque and black-box nature\nof the deep neural nets limits their social acceptance and usability. This has\ngiven rise to the quest for model interpretability and explainability, more so\nin the complex tasks involving multimodal AI methods. This paper extensively\nreviews the present literature to present a comprehensive survey and commentary\non the explainability in multimodal deep neural nets, especially for the vision\nand language tasks. Several topics on multimodal AI and its applications for\ngeneric domains have been covered in this paper, including the significance,\ndatasets, fundamental building blocks of the methods and techniques,\nchallenges, applications, and future trends in this domain\n"}
{"abstract": "  We present results from a resolved stellar population search for dwarf\nsatellite galaxies of six nearby (D $<5$ Mpc), sub-Milky-Way mass hosts using\ndeep ($m\\sim27$ mag) optical imaging from the Large Binocular Telescope. We\nperform image simulations to quantify our detection efficiency for dwarfs over\na large range in luminosity and size, and develop a fast catalog-based emulator\nthat includes a treatment of unresolved photometric blending. We discover no\nnew dwarf satellites, but we recover two previously known dwarfs (DDO 113 and\nLV J1228+4358) with $M_{\\text{V}}<-12$ that lie in our survey volume. We\npreview a new theoretical framework to predict satellite luminosity functions\nusing analytic probability distribution functions and apply it to our sample,\nfinding that we predict one fewer classical dwarf and one more faint dwarf\n($M_{\\text{V}}\\sim-7.5$) than we find in our observational sample (i.e., the\nobservational sample is slightly top-heavy). However, the overall number of\ndwarfs in the observational sample (2) is in good agreement with the\ntheoretical expectations. Interestingly, DDO 113 shows signs of environmental\nquenching and LV J1228+4358 is tidally disrupting, suggesting that low-mass\nhosts may affect their satellites more severely than previously believed.\n"}
{"abstract": "  A parsec-scale dusty torus is thought to be the cause of Active Galactic\nNuclei (AGN) dichotomy in the 1/2 types, narrow/broad emission lines. In a\nprevious work, on the basis of parsec-scale resolution infrared / optical dust\nmaps it was found that dust filaments, few parsecs wide, several hundred\nparsecs long, were ubiquitous features crossing the centre of type 2 AGN, their\noptical thickness being sufficient to fully obscure the optical nucleus. This\nwork presents the complementary view for type 1 and intermediate-type AGN. The\nsame type of narrow, collimated, dust filaments are equally found at the centre\nof these AGN. The difference now resides in their location with respect to the\nnucleus, next to it but not crossing it, as it is the case in type 2, and their\nreduced optical thickness towards the centre, $A_V \\lesssim 1.5 \\rm{mag}$,\ninsufficient to obscure at UV nucleus wavelengths. It is concluded that large\nscale, hundred pc to kpc long, dust filaments and lanes, reminiscent of those\nseen in the Milky Way, are a common ingredient to the central parsec of\ngalaxies. Their optical thickness changes along their structure, in type 2\nreaching optical depths high enough to obscure the nucleus in full. Their\nlocation with respect to the nucleus and increasing gradient in optical depth\ntowards the centre could naturally lead to the canonical type 1/2 AGN\nclassification, making these filaments to play the role of the torus. Dust\nfilaments and lanes show equivalent morphologies in molecular gas. Available\ngas kinematic indicates mass inflows at rates $ < ~ 1 M\\odot~ yr^{-1}$.\n"}
{"abstract": "  Capillaries are the smallest vessels in the body responsible for the delivery\nof oxygen and nutrients to the surrounding cells. Various diseases have been\nshown to alter the density of nutritive capillaries and the flow velocity of\nerythrocytes. In previous studies, capillary density and flow velocity have\nbeen assessed manually by trained specialists. Manual analysis of a standard\n20-second long microvascular video takes on average 20 minutes and requires\nextensive training. Several studies have reported that manual analysis hinders\nthe application of microvascular microscopy in a clinical setting. In this\npaper, we present a fully automated state-of-the-art system, called\nCapillaryNet, that can quantify skin nutritive capillary density and red blood\ncell velocity from handheld microscopy videos. Moreover, CapillaryNet measures\nseveral novel microvascular parameters that researchers were previously unable\nto quantify, i.e. capillary hematocrit and Intra-capillary flow velocity\nheterogeneity. Our system has been used to analyze skin microcirculation videos\nfrom various patient groups (COVID-19, pancreatitis, and acute heart diseases).\nOur proposed system excels from existing capillary detection systems as it\ncombines the speed of traditional computer vision algorithms and the accuracy\nof convolutional neural networks.\n"}
{"abstract": "  Directional tests to compare nested parametric models are developed in the\ngeneral context of covariance selection for Gaussian graphical models. The\nexactness of the underlying saddlepoint approximation leads to exceptional\naccuracy of the proposed approach. This is verified by simulation experiments\nwith high-dimensional parameters of interest, where the accuracy of standard\nasymptotic approximations to the likelihood ratio test and some of its\nhigher-order modifications fails. The directional p-value isused to illustrate\nthe assessment of Markovian dependencies in a dataset from a veterinary trial\non cattle. A second example with microarray data shows how to select the graph\nstructure related to genetic anomalies due to acute lymphocytic leukemia.\n"}
{"abstract": "  Two-wave with diffuse power (TWDP) is one of the most promising models for\ndescription of small-scale fading effects in emerging wireless networks.\nHowever, its current statistical characterization has several fundamental\nissues. Primarily, conventional TWDP parameterization is not in accordance with\nthe model's underlying physical mechanisms. In addition, available TWDP\nexpressions for PDF, CDF, and MGF are given either in integral or approximate\nforms, or as mathematically untractable closed-form expressions. Consequently,\nthe existing TWDP statistical characterization does not allow accurate\nevaluation of system performance (such as error and outage probability) in all\nfading conditions for most modulation and diversity techniques. In this paper,\nthe existing statistical characterization of the TWDP fading model is improved\nby overcoming some of the noticed issues. In this regard, physically justified\nTWDP parameterization is proposed and used for further calculations.\nAdditionally, exact infinite-series PDF and CDF are introduced. Based on these\nexpressions, the exact MGF of the SNR is derived in form suitable for\nmathematical manipulations. The applicability of the proposed MGF for\nderivation of the exact average symbol error probability (ASEP) is demonstrated\nwith the example of M-ary PSK modulation. Therefore, in this paper, M-ary PSK\nASEP is derived as an explicit expression for the first time in the literature.\nThe derived expression is further simplified for large SNR values in order to\nobtain a closed-form asymptotic ASEP, which is shown to be applicable for SNR >\n20 dB. All proposed expressions are verified by Monte Carlo simulation in a\nvariety of TWDP fading conditions.\n"}
{"abstract": "  Model discovery aims at autonomously discovering differential equations\nunderlying a dataset. Approaches based on Physics Informed Neural Networks\n(PINNs) have shown great promise, but a fully-differentiable model which\nexplicitly learns the equation has remained elusive. In this paper we propose\nsuch an approach by integrating neural network-based surrogates with Sparse\nBayesian Learning (SBL). This combination yields a robust model discovery\nalgorithm, which we showcase on various datasets. We then identify a connection\nwith multitask learning, and build on it to construct a Physics Informed\nNormalizing Flow (PINF). We present a proof-of-concept using a PINF to directly\nlearn a density model from single particle data. Our work expands PINNs to\nvarious types of neural network architectures, and connects neural\nnetwork-based surrogates to the rich field of Bayesian parameter inference.\n"}
{"abstract": "  Being a state-of-the-art network, Software Defined Networking (SDN) decouples\ncontrol and management planes from data plane of the forwarding devices by\nimplementing both the control and management planes at logically centralized\nentity, called controller. This helps to make simple and easy both the network\ncontrol and management. Failure of links occurs frequently in a computer\nnetwork. To deal with the link failures, the existing approaches computes and\ninstalls multiple paths for a flow at the switches in SDN without considering\nthe reliability value of the primary path. This incurs extra computation to\ncompute multiple paths, and both increased computation time and traffic to\ninstall extra flow rules in the network. In this research work we propose a new\napproach that calculates the link reliability and then installs the number of\nmultiple paths based on the reliability value of the primary path. More\nspecifically, if a primary path has higher reliability then a smaller number of\nalternative paths should be installed. This shall decrease the path\ncomputational time and flow rule installation load at controller. Resultantly\nthere shall be less flow rule entries in switch flow table which in turn will\navoid the overflow of the flow table. Through simulation results, our proposed\napproach performs better as compared to the existing approach in term of\ncomputational overhead at controller, end-to-end delay for packet deliver and\nthe traffic overhead for flow rule installation.\n"}
{"abstract": "  We investigate four-body nuclear reactions in stellar environments\ncontributing to creation of light nuclei, exemplified by $^9$Be and $^{12}$C.\nThe originally assumed process is radiative capture, where nuclear clusters\ncombine into the excited final nucleus and photon emission populates the stable\nnuclear ground states. Instead, we consider nuclear four-body recombination\nreactions where a spectator nuclear particle replaces the photon. We first\ndevelop the elaborate formalism for both, direct and sequential capture\nprocesses, where the decaying three-body resonance is formed without and with\npopulation of an intermediate two-body resonance, respectively. To facilitate\nboth calculations and practical applications we parameterize the involved cross\nsections as done successfully in previous computations of reaction rates. We\nconsider the lowest-lying nuclear states with their dominant contributions at\nlow stellar temperatures. We calculate and compare reaction and production\nrates for different processes. The direct reaction mechanism dominates by many\norders of magnitude at low temperature, where the sequential stepping stones\nare energetically too expensive to use. At somewhat higher temperatures these\ntwo different nuclear four-body mechanisms become comparable. Comparison to\nradiative three-body capture reveals already formally, but also numerically,\nthat four-body nuclear recombination must dominate for sufficiently high\nnuclear densities. Numerical values are given for all these rates as function\nof temperature and density. The relative importance is exhibited.\n"}
{"abstract": "  We experimentally demonstrate a polarization-entangled photon source at 810\nnm using a type-II phase-matched PPKTP crystal pumped by a low-cost, broadband\nlaser diode with a central wavelength of 405 nm and a typical bandwidth of 0.53\nnm. The PPKTP crystal is placed in a Sagnac-loop to realize the compact size\nand high stability. The downconverted biphotons, the signal and the idler, have\ntypical bandwidths of 5.57 nm and 7.32 nm. We prepare two Bell states |Psi+>\nand |Psi-> with the fidelities of 0.948+-0.004 and 0.963+-0.002. In\npolarization correlation measurement, the visibilities are all higher than\n96.2%, and in the Bell inequality test, the S value can achieve 2.78+-0.01. To\nour knowledge, this experiment is the first to combine a multi-mode pump laser\nwith a Sagnac-loop configuration. This high-quality and low-cost entangled\nphoton source may have many practical applications in quantum information\nprocessing.\n"}
{"abstract": "  Impending catastrophic failure of granular earth slopes manifests distinct\nkinematic patterns in space and time. While risk assessments of slope failure\nhazards have routinely relied on the monitoring of ground motion, such\nprecursory failure patterns remain poorly understood. A key challenge is the\nmultiplicity of spatiotemporal scales and dynamical regimes. In particular,\nthere exist a precursory failure regime where two mesoscale mechanisms\ncoevolve, namely, the preferred transmission paths for force and damage.\nDespite extensive studies, a formulation which can address their coevolution\nnot just in laboratory tests but also in large, uncontrolled field environments\nhas proved elusive. Here we address this problem by developing a slope\nstability analytics framework which uses network flow theory and mesoscience to\nmodel this coevolution and predict emergent kinematic clusters solely from\nsurface ground motion data. We test this framework on four data sets: one at\nthe laboratory scale using individual grain displacement data; three at the\nfield scale using line-of-sight displacement of a slope surface, from\nground-based radar in two mines and from space-borne radar for the 2017 Xinmo\nlandslide. The dynamics of the kinematic clusters deliver an early prediction\nof the geometry, location and time of failure.\n"}
{"abstract": "  The paper is concerned with identification of the key mechanisms controlling\ndeflagration-to-detonation transition in stellar medium. The issue of thermal\nrunaway triggered by positive feedback between the advancing flame and the\nflame-driven precompression is discussed in the framework of a one-dimensional\nflame-folding model. The paper is an extension of the authors' previous study\ndealing with the non-stoichiometric fusion, $fuel \\to products$, kinetics\n(Phys.Rev.E, 103(2021)) over physically more relevant, $fuel+fuel \\to\nproducts$, kinetics. Despite this change the runaway effect endures. The\ntransition occurs prior to merging of the flame with the flame-supported\nprecursor shock, i.e. the pretransition flame does not reach the threshold of\nChapman-Jouguet deflagration.\n"}
{"abstract": "  Synchronized movement of (both unicellular and multicellular) systems can be\nobserved almost everywhere. Understanding of how organisms are regulated to\nsynchronized behavior is one of the challenging issues in the field of\ncollective motion. It is hypothesized that one or a few agents in a group\nregulate(s) the dynamics of the whole collective, known as leader(s). The\nidentification of the leader (influential) agent(s) is very crucial. This\narticle reviews different mathematical models that represent different types of\nleadership. We focus on the improvement of the leader-follower classification\nproblem. It was found using a simulation model that the use of interaction\ndomain information significantly improves the leader-follower classification\nability using both linear schemes and information-theoretic schemes for\nquantifying influence. This article also reviews different schemes that can be\nused to identify the interaction domain using the motion data of agents.\n"}
{"abstract": "  Quark-gluon plasma (QGP) droplet formation is re-considered with the addition\nof three loop correction to the earlier loop factors in the mean field\npotential. The correction of the three loop factor increases stability in the\ndroplet formations of QGP at different parametrization factors of the QGP fluid\nand it is in better agreement in comparison to the lattice results of pressure,\nenergy density and other thermodynamic relations. This implies that the\ncontribution of the three loop enhances in showing the characteristic features\nof the QGP fluid. It shows that increasing the loop increased the strength of\nparametrization value which we defined earlier as a number parameter of fluid\ndynamics. It indicates that the model with the loop correction boosts in\nexplaining about the formation of QGP droplet in the expansion of early\nuniverse\n"}
{"abstract": "  The distributions and abundances of molecules in protoplanetary disks are\npowerful tracers of the physical and chemical disk structures. The abundance\nratios of HCN and its isomer HNC are known to be sensitive to gas temperature.\nTheir line ratios might therefore offer a unique opportunity to probe the\nproperties of the emitting gas. Using the 2D thermochemical code DALI, we ran a\nset of disk models accounting for different stellar and disk properties, with\nan updated chemical network for the nitrogen chemistry. These modeling results\nwere then compared with observations, including new ALMA observations of HNC\n$J=3-2$ for the TW Hya disk and HNC $J=1-0$ for 29 disks in Lupus. Similar to\nCN, HCN and HNC have brighter line emission in models with larger disk flaring\nangles and higher UV fluxes. HNC and HCN are predicted to be abundant in the\nwarm surface layer and outer midplane region, which results in ring-shaped\nemission patterns. However, the precise emitting regions and emission\nmorphology depend on the probed transition, as well as on other parameters such\nas C and O abundances. The modeled HNC-to-HCN line intensity ratio increases\nfrom $<0.1$ in the inner disk to up to 0.8 in the outer disk regions, which can\nbe explained by efficient HNC destruction at high temperatures. Disk-integrated\nHNC line fluxes from current scarce observations and its radial distribution in\nthe TW Hya disk are broadly consistent with our model predictions. The\nHNC-to-HCN flux ratio robustly increases with radius (decreasing temperature),\nbut its use as a chemical thermometer in disks is affected by other factors,\nincluding UV flux and C and O abundances. High-spatial resolution ALMA disk\nobservations of HNC and HCN that can locate the emitting layers would have the\ngreat potential to constrain both the disk thermal and UV radiation structures,\nand also to verify our understanding of the nitrogen chemistry.\n"}
{"abstract": "  We present a method to measure the spectral density of in-band optical\ntransmission impairments without coherent electrical reception and digital\nsignal processing at the receiver. We determine the method's accuracy by\nnumerical simulations and show experimentally its feasibility, including the\nmeasure of in-band nonlinear distortions power densities.\n"}
{"abstract": "  We present a detailed comparison of the rheological behaviour of sheared\nsediment beds in a pressure-driven, straight channel configuration based on\ndata that was generated by means of fully coupled, grain-resolved direct\nnumerical simulations and experimental measurements reviously published by\nAussillous {\\it et al.} (J. Fluid Mech., vol. 736, 2013, pp. 594-615). The\nhighly-resolved simulation data allows to compute the stress balance of the\nsuspension in the streamwise and vertical directions and the stress exchange\nbetween the fluid and particle phase, which is information needed to infer the\nrheology, but has so far been unreachable in experiments. Applying this\nknowledge to the experimental and numerical data, we obtain the\nstatistically-stationary, depth-resolved profiles of the relevant rheological\nquantities. The scaling behavior of rheological quantities such as the shear\nand normal viscosities and the effective friction coefficient are examined and\ncompared to data coming from rheometry experiments and from widely-used\nrheological correlations. We show that rheological properties that have\npreviously been inferred for annular Couette-type shear flows with neutrally\nbuoyant particles still hold for our setup of sediment transport in a\nPoiseuille flow and in the dense regime we found good agreement with empirical\nrelationships derived therefrom. Subdividing the total stress into parts from\nparticle contact and hydrodynamics suggests a critical particle volume fraction\nof 0.3 to separate the dense from the dilute regime. In the dilute regime,\ni.e., the sediment transport layer, long-range hydrodynamic interactions are\nscreened by the porous media and the effective viscosity obeys the Einstein\nrelation.\n"}
{"abstract": "  This is a motivating tutorial introduction to a semantic analysis of\nprogramming languages using a graphical language as the representation of\nterms, and graph rewriting as a representation of reduction rules. We show how\nthe graphical language automatically incorporates desirable features, such as\nalpha-equivalence and how it can describe pure computation, imperative store,\nand control features in a uniform framework. The graph semantics combines some\nof the best features of structural operational semantics and abstract machines,\nwhile offering powerful new methods for reasoning about contextual equivalence.\n  All technical details are available in an extended technical report by Muroya\nand the author and in Muroya's doctoral dissertation.\n"}
{"abstract": "  The world is increasingly urbanizing and the building industry accounts for\nmore than 40% of energy consumption in the United States. To improve urban\nsustainability, many cities adopt ambitious energy-saving strategies through\nretrofitting existing buildings and constructing new communities. In this\nsituation, an accurate urban building energy model (UBEM) is the foundation to\nsupport the design of energy-efficient communities. However, current UBEM are\nlimited in their abilities to capture the inter-building interdependency due to\ntheir dynamic and non-linear characteristics. Those models either ignored or\noversimplified these building interdependencies, which can substantially affect\nthe accuracy of urban energy modeling. To fill the research gap, this study\nproposes a novel data-driven UBEM synthesizing the solar-based building\ninterdependency and spatial-temporal graph convolutional network (ST-GCN)\nalgorithm. Especially, we took a university campus located in downtown Atlanta\nas an example to predict the hourly energy consumption. Furthermore, we tested\nthe feasibility of the proposed model by comparing the performance of the\nST-GCN model with other common time-series machine learning models. The results\nindicate that the ST-GCN model overall outperforms all others. In addition, the\nphysical knowledge embedded in the model is well interpreted. After discussion,\nit is found that data-driven models integrated engineering or physical\nknowledge can significantly improve the urban building energy simulation.\n"}
{"abstract": "  Recently, free-space optical (FSO) communication systems have obtained a wide\nrange of applications for communications between unmanned aerial vehicles\n(UAVs) because they offer several advantages with respect to the traditional RF\ncommunication systems. To investigate the performance of UAV-based FSO systems\nin terms of achievable data rates, we derive the exact closed-form expression\nfor the mean achievable rates of UAV-to-UAV transmission. Our results may serve\nto investigate the interaction between different FSO system parameters and the\naverage data rates achieved by UAV-based optical wireless communications.\n"}
{"abstract": "  Floods are the most common and among the most severe natural disasters in\nmany countries around the world. As global warming continues to exacerbate sea\nlevel rise and extreme weather, governmental authorities and environmental\nagencies are facing the pressing need of timely and accurate evaluations and\npredictions of flood risks. Current flood forecasts are generally based on\nhistorical measurements of environmental variables at monitoring stations. In\nrecent years, in addition to traditional data sources, large amounts of\ninformation related to floods have been made available via social media.\nMembers of the public are constantly and promptly posting information and\nupdates on local environmental phenomena on social media platforms. Despite the\ngrowing interest of scholars towards the usage of online data during natural\ndisasters, the majority of studies focus exclusively on social media as a\nstand-alone data source, while its joint use with other type of information is\nstill unexplored. In this paper we propose to fill this gap by integrating\ntraditional historical information on floods with data extracted by Twitter and\nGoogle Trends. Our methodology is based on vine copulas, that allow us to\ncapture the dependence structure among the marginals, which are modelled via\nappropriate time series methods, in a very flexible way. We apply our\nmethodology to data related to three different coastal locations on the South\ncoast of the United Kingdom (UK). The results show that our approach, based on\nthe integration of social media data, outperforms traditional methods in terms\nof evaluation and prediction of flood events.\n"}
{"abstract": "  The effect of boundaries and how these can be used to influence the bulk\nbehaviour in geometrically frustrated systems are both long-standing puzzles,\noften relegated to secondary role. Here we use numerical simulations and \"proof\nof concept\" experiments to demonstrate that boundaries can be engineered to\ncontrol the bulk behavior in a colloidal artificial ice. We show that an\nantiferromagnetic frontier forces the system to rapidly reach the ground state\n(GS), as opposed to the commonly implemented open or periodic boundary\nconditions. We also show that strategically placing defects at the corners\ngenerates novel bistable states, or topological strings which result from\ncompeting GS regions in the bulk. Our results could be generalized to other\nfrustrated micro and nanostructures where boundary conditions may be engineered\nwith lithographic techniques.\n"}
{"abstract": "  With more than 4,300 confirmed exoplanets and counting, the next milestone in\nexoplanet research is to determine which of these newly found worlds could\nharbor life. Coronal Mass Ejections (CMEs), spawn by magnetically active,\nsuperflare-triggering dwarf stars, pose a direct threat to the habitability of\nterrestrial exoplanets as they can deprive them from their atmospheres. Here we\ndevelop a readily implementable atmosphere sustainability constraint for\nterrestrial exoplanets orbiting active dwarfs, relying on the magnetospheric\ncompression caused by CME impacts. Our constraint focuses on a systems\nunderstanding of CMEs in our own heliosphere that, applying to a given\nexoplanet, requires as key input the observed bolometric energy of flares\nemitted by its host star. Application of our constraint to six famous\nexoplanets, (Kepler-438b, Proxima-Centauri b, and Trappist-1d, -1e, -1f and\n-1g), within or in the immediate proximity of their stellar host's habitable\nzones, showed that only for Kepler-438b might atmospheric sustainability\nagainst stellar CMEs be likely. This seems to align with some recent studies\nthat, however, may require far more demanding computational resources and\nobservational inputs. Our physically intuitive constraint can be readily and en\nmasse applied, as is or generalized, to large-scale exoplanet surveys to detect\nplanets that could be sieved for atmospheres and, perhaps, possible\nbiosignatures at higher priority by current and future instrumentation.\n"}
{"abstract": "  Structural breaks have been commonly seen in applications. Specifically for\ndetection of change points in time, research gap still remains on the setting\nin ultra high dimension, where the covariates may bear spurious correlations.\nIn this paper, we propose a two-stage approach to detect change points in ultra\nhigh dimension, by firstly proposing the dynamic titled current correlation\nscreening method to reduce the input dimension, and then detecting possible\nchange points in the framework of group variable selection. Not only the\nspurious correlation between ultra-high dimensional covariates is taken into\nconsideration in variable screening, but non-convex penalties are studied in\nchange point detection in the ultra high dimension. Asymptotic properties are\nderived to guarantee the asymptotic consistency of the selection procedure, and\nthe numerical investigations show the promising performance of the proposed\napproach.\n"}
{"abstract": "  The involutory subalgebra K(E$_9$) of the affine Kac-Moody algebra E$_9$ was\nrecently shown to admit an infinite sequence of unfaithful representations of\never increasing dimensions arXiv:2102.00870. We revisit these representations\nand describe their associated ideals in more detail, with particular emphasis\non two chiral versions that can be constructed for each such representation.\nFor every such unfaithful representation we show that the action of K(E$_9$)\ndecomposes into a direct sum of two mutually commuting (`chiral' and\n`anti-chiral') parabolic algebras with Levi subalgebra\n$\\mathfrak{so}(16)_+\\,\\oplus\\,\\mathfrak{so}(16)_-$. We also spell out the\nconsistency conditions for uplifting such representations to unfaithful\nrepresentations of K(E$_{10}$). From these results it is evident that the\nholonomy groups so far discussed in the literature are mere shadows (in a\nPlatonic sense) of a much larger structure.\n"}
{"abstract": "  We fill in a gap in the proof of the main theorem in our earlier paper [Ol].\nAt the same time, we prove a slightly stronger version of the theorem needed\nfor another paper.\n"}
{"abstract": "  Recent theories and experiments have suggested hydrodynamic phonon transport\nfeatures in graphite at unusually high temperatures. Here, we report a\npico-second pump-probe thermal reflectance measurement of heat pulse\npropagation in graphite. The measurement results reveal transient lattice\ncooling near the adiabatic center of a 15 $\\mu$m diameter ring-shape pump beam\nat temperatures between 80 and 120 K. While such lattice cooling has not been\nreported in recent diffraction measurements of second sound in graphite, the\nobservation here is consistent with both hydrodynamic phonon transport theory\nand prior heat pulse measurements of second sound in bulk sodium fluoride.\n"}
{"abstract": "  Cerebrospinal fluid (CSF) plays a pivotal role in normal functioning of\nBrain. Intracranial compartments such as blood, brain and CSF are\nincompressible in nature. Therefore, if a volume imbalance in one of the\naforenoted compartments is observed, the other reaches out to maintain net\nchange to zero. Whereas, CSF has higher compliance over long term. However, if\nthe CSF flow is obstructed in the ventricles, this compliance may get exhausted\nearly. Brain tumor on the other hand poses a similar challenge towards\ndestabilization of CSF flow by compressing any section of ventricles thereby\nensuing obstruction. To avoid invasive procedures to study effects of tumor on\nCSF flow, numerical-based methods such as Finite element modeling (FEM) are\nused which provide excellent description of underlying pathological\ninteraction. A 3D fluid-structure interaction (FSI) model is developed to study\nthe effect of tumor growth on the flow of cerebrospinal fluid in ventricle\nsystem. The FSI model encapsulates all the physiological parameters which may\nbe necessary in analyzing intraventricular CSF flow behavior. Findings of the\nmodel show that brain tumor affects CSF flow parameters by deforming the walls\nof ventricles in this case accompanied by a mean rise of 74.23% in CSF flow\nvelocity and considerable deformation on the walls of ventricles.\n"}
{"abstract": "  Extreme mass-ratio inspirals~(EMRIs) detectable by the Laser Inteferometric\nSpace Antenna~(LISA) are unique probes of astrophysics and fundamental physics.\nParameter estimation for these sources is challenging, especially because the\nwaveforms are long, complicated, known only numerically, and slow to compute in\nthe most relevant regime, where the dynamics is relativistic. We perform a\ntime-consuming Fisher-matrix error analysis of the EMRI parameters using\nfully-relativistic numerical waveforms to leading order in an adiabatic\nexpansion on a Kerr background, taking into account the motion of the LISA\nconstellation, higher harmonics, and also including the leading correction from\nthe spin of the secondary in the post-adiabatic approximation. We pay\nparticular attention to the convergence of the numerical derivatives in the\nFisher matrix and to the numerical stability of the covariance matrix, which\nfor some systems requires computing the numerical waveforms with approximately\n$90$-digit precision. Our analysis confirms previous results (obtained with\napproximated but much more computationally efficient waveforms) for the\nmeasurement errors on the binary's parameters. We also show that the inclusion\nof higher harmonics improves the errors on the luminosity distance and on the\norbital angular momentum angles by one order and two orders of magnitude,\nrespectively, which might be useful to identify the environments where EMRIs\nlive. We particularly focus on the measurability of the spin of the secondary,\nconfirming that it cannot be measured with sufficient accuracy. However, due to\ncorrelations, its inclusion in the waveform model can deteriorate the accuracy\non the measurements of other parameters by orders of magnitude, unless a\nphysically-motivated prior on the secondary spin is imposed.\n"}
{"abstract": "  Coronavirus disease 2019 (CoViD-19) is an infectious disease caused by severe\nacute respiratory syndrome coronavirus 2 (SARS-CoV-2). Among many symptoms,\ncough, fever and tiredness are the most common. People over 60 years old and\nwith associated comorbidities are most likely to develop a worsening health\ncondition. This paper proposes a non-integer order model to describe the\ndynamics of CoViD-19 in a standard population. The model incorporates the\nreinfection rate in the individuals recovered from the disease. Numerical\nsimulations are performed for different values of the order of the fractional\nderivative and of reinfection rate. The results are discussed from a biological\npoint of view.\n"}
{"abstract": "  The actual mechanism(s) powering galactic outflows in active galactic nuclei\n(AGN) is still a matter of debate. At least two physical models have been\nconsidered in the literature: wind shocks and radiation pressure on dust. Here\nwe provide a first quantitative comparison of the AGN radiative feedback\nscenario with observations of galactic outflows. We directly compare our\nradiation pressure-driven shell models with the observational data from the\nmost recent compilation of molecular outflows on galactic scales. We show that\nthe observed dynamics and energetics of galactic outflows can be reproduced by\nAGN radiative feedback, with the inclusion of radiation trapping and/or\nluminosity evolution. The predicted scalings of the outflow energetics with AGN\nluminosity can also quantitatively account for the observational scaling\nrelations. Furthermore, sources with both ultra-fast and molecular outflow\ndetections are found to be located in the `forbidden' region of the\n$N_\\mathrm{H} - \\lambda$ plane. Overall, an encouraging agreement is obtained\nover a wide range of AGN and host galaxy parameters. We discuss our results in\nthe context of recent observational findings and numerical simulations. In\nconclusion, AGN radiative feedback is a promising mechanism for driving\ngalactic outflows that should be considered, alongside wind feedback, in the\ninterpretation of future observational data.\n"}
{"abstract": "  Stars form by accreting material from their surrounding disks. There is a\nconsensus that matter flowing through the disk is channelled onto the stellar\nsurface by the stellar magnetic field. This is thought to be strong enough to\ntruncate the disk close to the so-called corotation radius where the disk\nrotates at the same rate as the star. Spectro-interferometric studies in young\nstellar objects show that Hydrogen is mostly emitted in a region of a few\nmilliarcseconds across, usually located within the dust sublimation radius. Its\norigin is still a matter of debate and it can be interpreted as coming from the\nstellar magnetosphere, a rotating wind or a disk. In the case of\nintermediate-mass Herbig AeBe stars, the fact that the Br gamma emission is\nspatially resolved rules out that most of the emission comes from the\nmagnetosphere. This is due to the weak magnetic fields (some tenths of G)\ndetected in these sources, resulting in very compact magnetospheres. In the\ncase of T Tauri sources, their larger magnetospheres should make them easier to\nresolve. However, the small angular size of the magnetosphere (a few tenths of\nmilliarcseconds), along with the presence of winds emitting in Hydrogen make\nthe observations interpretation challenging. Here, we present direct evidence\nof magnetospheric accretion by spatially resolving the inner disk of the 60 pc\nT Tauri star TW Hydrae through optical long baseline interferometry. We find\nthat the hydrogen near-infrared emission comes from a region approximately 3.5\nstellar radii (R*) across. This region is within the continuum dusty disk\nemitting region (Rcont = 7 R*) and smaller than the corotation radius which is\ntwice as big. This indicates that the hydrogen emission originates at the\naccretion columns, as expected in magnetospheric accretion models, rather than\nin a wind emitted at much larger distance (>1au).\n"}
{"abstract": "  We prove that if a compact set E in complex Euclidean space is contained in\nan arc J, then there is a choice of J whose polynomial hull is the union of J\nand the polynomial hull of E. This strengthens an earlier result of the author.\nWe also correct an inaccuracy in the statement, and fill a gap in the proof, of\nthat earlier result.\n"}
{"abstract": "  We propose an unsupervised solution to the Authorship Verification task that\nutilizes pre-trained deep language models to compute a new metric called\nDV-Distance. The proposed metric is a measure of the difference between the two\nauthors comparing against pre-trained language models. Our design addresses the\nproblem of non-comparability in authorship verification, frequently encountered\nin small or cross-domain corpora. To the best of our knowledge, this paper is\nthe first one to introduce a method designed with non-comparability in mind\nfrom the ground up, rather than indirectly. It is also one of the first to use\nDeep Language Models in this setting. The approach is intuitive, and it is easy\nto understand and interpret through visualization. Experiments on four datasets\nshow our methods matching or surpassing current state-of-the-art and strong\nbaselines in most tasks.\n"}
{"abstract": "  We prove explicit lower bounds for the smallest singular value and upper\nbounds for the condition number of rectangular, multivariate Vandermonde\nmatrices with scattered nodes on the complex unit circle. Analogously to the\nShannon-Nyquist criterion, the nodes are assumed to be separated by a constant\ndivided by the used polynomial degree. If this constant grows linearly with the\nspatial dimension, the condition number is uniformly bounded. If it grows only\nlogarithmically with the spatial dimension, the condition number grows slightly\nstronger than exponentially with the spatial dimension. Both results are quasi\noptimal and improve over all previously known results of such type.\n"}
{"abstract": "  Ship detection in remote sensing images plays a crucial role in various\napplications and has drawn increasing attention in recent years. However,\nexisting arbitrary-oriented ship detection methods are generally developed on a\nset of predefined rotated anchor boxes. These predefined boxes not only lead to\ninaccurate angle predictions but also introduce extra hyper-parameters and high\ncomputational cost. Moreover, the prior knowledge of ship size has not been\nfully exploited by existing methods, which hinders the improvement of their\ndetection accuracy. Aiming at solving the above issues, in this paper, we\npropose a center-head point extraction based detector (named CHPDet) to achieve\narbitrary-oriented ship detection in remote sensing images. Our CHPDet\nformulates arbitrary-oriented ships as rotated boxes with head points which are\nused to determine the direction. And rotated Gaussian kernel is used to map the\nannotations into target heatmaps. Keypoint estimation is performed to find the\ncenter of ships. Then, the size and head point of the ships are regressed. The\norientation-invariant model (OIM) is also used to produce orientation-invariant\nfeature maps. Finally, we use the target size as prior to finetune the results.\nMoreover, we introduce a new dataset for multi-class arbitrary-oriented ship\ndetection in remote sensing images at a fixed ground sample distance (GSD)\nwhich is named FGSD2021. Experimental results on FGSD2021 and two other widely\nused data sets, i.e., HRSC2016, and UCAS-AOD demonstrate that our CHPDet\nachieves state-of-the-art performance and can well distinguish between bow and\nstern. Code and FGSD2021 dataset are available at\nhttps://github.com/zf020114/CHPDet.\n"}
{"abstract": "  The $H^m$-conforming virtual elements of any degree $k$ on any shape of\npolytope in $\\mathbb R^n$ with $m, n\\geq1$ and $k\\geq m$ are recursively\nconstructed by gluing conforming virtual elements on faces in a universal way.\nFor the lowest degree case $k=m$, the set of degrees of freedom only involves\nfunction values and derivatives up to order $m-1$ at the vertices of the\npolytope. The inverse inequality and several norm equivalences for the\n$H^m$-conforming virtual elements are rigorously proved. The $H^m$-conforming\nvirtual elements are then applied to discretize a polyharmonic equation with a\nlower order term. With the help of the interpolation error estimate and norm\nequivalences, the optimal error estimates are derived for the $H^m$-conforming\nvirtual element method.\n"}
{"abstract": "  Imitation learning and instruction-following are two common approaches to\ncommunicate a user's intent to a learning agent. However, as the complexity of\ntasks grows, it could be beneficial to use both demonstrations and language to\ncommunicate with an agent. In this work, we propose a novel setting where an\nagent is given both a demonstration and a description, and must combine\ninformation from both the modalities. Specifically, given a demonstration for a\ntask (the source task), and a natural language description of the differences\nbetween the demonstrated task and a related but different task (the target\ntask), our goal is to train an agent to complete the target task in a zero-shot\nsetting, that is, without any demonstrations for the target task. To this end,\nwe introduce Language-Aided Reward and Value Adaptation (LARVA) which, given a\nsource demonstration and a linguistic description of how the target task\ndiffers, learns to output a reward / value function that accurately describes\nthe target task. Our experiments show that on a diverse set of adaptations, our\napproach is able to complete more than 95% of target tasks when using\ntemplate-based descriptions, and more than 70% when using free-form natural\nlanguage.\n"}
{"abstract": "  We propose a novel distance-based regularization method for deep metric\nlearning called Multi-level Distance Regularization (MDR). MDR explicitly\ndisturbs a learning procedure by regularizing pairwise distances between\nembedding vectors into multiple levels that represents a degree of similarity\nbetween a pair. In the training stage, the model is trained with both MDR and\nan existing loss function of deep metric learning, simultaneously; the two\nlosses interfere with the objective of each other, and it makes the learning\nprocess difficult. Moreover, MDR prevents some examples from being ignored or\noverly influenced in the learning process. These allow the parameters of the\nembedding network to be settle on a local optima with better generalization.\nWithout bells and whistles, MDR with simple Triplet loss achieves\nthe-state-of-the-art performance in various benchmark datasets: CUB-200-2011,\nCars-196, Stanford Online Products, and In-Shop Clothes Retrieval. We\nextensively perform ablation studies on its behaviors to show the effectiveness\nof MDR. By easily adopting our MDR, the previous approaches can be improved in\nperformance and generalization ability.\n"}
{"abstract": "  Implicitly filtered large-eddy simulation (LES) is by nature numerically\nunder-resolved. With the sole exception of Fourier-spectral methods, discrete\nnumerical derivative operators cannot accurately represent the dynamics of all\nof the represented scales. Since the resolution scale in an LES usually lies in\nthe inertial range, these poorly represented scales are dynamically significant\nand errors in their dynamics can affect all resolved scales. This Letter is\nfocused on characterizing the effects of numerical dispersion error by studying\nthe energy cascade in LES of convecting homogeneous isotropic turbulence.\nNumerical energy and transfer spectra reveal that energy is not transferred at\nthe appropriate rate to wavemodes where significant dispersion error is\npresent. This leads to a deficiency of energy in highly dispersive modes and an\naccompanying pile up of energy in the well resolved modes, since dissipation by\nthe subgrid model is diminished. An asymptotic analysis indicates that\ndispersion error causes a phase decoherence between triad interacting\nwavemodes, leading to a reduction in the mean energy transfer rate for these\nscales. These findings are relevant to a wide range of LES, since turbulence\ncommonly convects through the grid in practical simulations. Further, these\nresults indicate that the resolved scales should be defined to not include the\ndispersive modes.\n"}
{"abstract": "  Brazil is the 5th largest country in the world, despite of having a ``High\nHuman Development'' it is the 9th most unequal country. The existing Brazilian\nmicro pension programme is one of the safety nets for poor people. To become\neligible for this benefit, each person must have an income that is less than a\nquarter of the Brazilian minimum monthly wage and be either over 65 or\nconsidered disabled. That minimum income corresponds to approximately $2$\ndollars per day. This paper analyses quantitatively some aspects of this\nprogramme in the Public Pension System of Brazil. We look for the impact of\nsome particular economic variables on the number of people receiving the\nbenefit, and seek if that impact significantly differs among the 27 Brazilian\nFederal Units. We search for heterogeneity. We perform regression and spatial\ncluster analysis for detection of geographical grouping. We use a database that\nincludes the entire population that receives the benefit. Afterwards, we\ncalculate the amount that the system spends with the beneficiaries, estimate\nvalues \\textit{per capita} and the weight of each UF, searching for\nheterogeneity reflected on the amount spent \\textit{per capita}. In this latter\ncalculation we use a more comprehensive database, by individual, that includes\nall people that started receiving a benefit under the programme in the period\nfrom 2nd of January 2018 to 6th of April 2018. We compute the expected\ndiscounted benefit and confirm a high heterogeneity among UF's as well as\ngender. We propose achieving a more equitable system by introducing `age\nadjusting factors' to change the benefit age.\n"}
{"abstract": "  By constructing the configuration of D3-branes with D(-1)-branes as\nD-instantons, we study the three-dimensional Yang-Mills Chern-Simons theory in\nholography. Due to the presence of the D-instantons, the D7-branes with\ndiscrepant embedding functions are able to be introduced in order to include\nthe fundamental fermions (as flavors) and the Chern-Simons term (at very low\nenergy) in the dual theory. The vacuum structure at zero temperature is studied\nin the soliton background and it illustrates the topological phase transition\nin the presence of instantons. Moreover, since the confinement/deconfinement\nphase transition could be holographically identified as the Hawking-Page\ntransition in the bulk, we accordingly calculate the critical temperature of\nthe deconfinement phase transition by collecting the bulk onshell action as the\nthermodynamical free energy. On the other hand, we evaluate the difference of\nthe entanglement entropy in slab configuration by using the RT formula since\nthe confinement may also be characterized by the entanglement entropy.\nAltogether we find the behavior of the critical temperature is in qualitative\nagreement with the behavior of the critical length determined by the\nentanglement entropy which implies the entanglement entropy could indeed be a\ncharacter of the confinement in our setup and the D3-D(-1) system would be a\nremarkable approach to study the three-dimensional gauge theory.\n"}
{"abstract": "  Which samples should be labelled in a large data set is one of the most\nimportant problems for trainingof deep learning. So far, a variety of active\nsample selection strategies related to deep learning havebeen proposed in many\nliteratures. We defined them as Active Deep Learning (ADL) only if\ntheirpredictor is deep model, where the basic learner is called as predictor\nand the labeling schemes iscalled selector. In this survey, three fundamental\nfactors in selector designation were summarized. Wecategory ADL into\nmodel-driven ADL and data-driven ADL, by whether its selector is model-drivenor\ndata-driven. The different characteristics of the two major type of ADL were\naddressed in indetail respectively. Furthermore, different sub-classes of\ndata-driven and model-driven ADL are alsosummarized and discussed emphatically.\nThe advantages and disadvantages between data-driven ADLand model-driven ADL\nare thoroughly analyzed. We pointed out that, with the development of\ndeeplearning, the selector in ADL also is experiencing the stage from\nmodel-driven to data-driven. Finally,we make discussion on ADL about its\nuncertainty, explanatory, foundations of cognitive science etc.and survey on\nthe trend of ADL from model-driven to data-driven.\n"}
{"abstract": "  In this paper, we employ Machine Learning algorithms on multi-mission\nobservations for the classification of accretion states of outbursting Black\nhole X-ray binaries for the first time. Archival data from RXTE, Swift, MAXI\nand AstroSat observatories are used to generate the hardness intensity diagrams\n(HIDs) for outbursts of the sources XTE J1859+226 (1999 outburst), GX 339-4\n(2002, 2004, 2007 and 2010 outbursts), IGR J17091-3624 (2016 outburst), and\nMAXI J1535-571 (2017 outburst). Based on variation of X-ray flux, hardness\nratios, presence of various types of Quasi-periodic Oscillations (QPOs), photon\nindices and disk temperature, we apply clustering algorithms like K-Means\nclustering and Hierarchical clustering to classify the accretion states\n(clusters) of each outburst. As multiple parameters are involved in the\nclassification process, we show that clustering algorithms club together the\nobservations of similar characteristics more efficiently than the `standard'\nmethod of classification. We also infer that K-Means clustering provides more\nreliable results than Hierarchical clustering. We demonstrate the importance of\nthe classification based on machine learning by comparing it with results from\n`standard' classification.\n"}
{"abstract": "  We consider the possible existence of gravitationally bound stringlike\nobjects in the framework of the generalized hybrid metric-Palatini gravity\ntheory, in which the gravitational action is represented by an arbitrary\nfunction of the Ricci and of the Palatini scalars, respectively. The theory\nadmits an equivalent scalar-tensor representation in terms of two independent\nscalar fields. Assuming cylindrical symmetry, and the boost invariance of the\nmetric, we obtain the gravitational field equations that describe cosmic\nstringlike structures in the theory. The physical and geometrical properties of\nthe cosmic strings are determined by the two scalar fields, as well by an\neffective field potential, functionally dependent on both scalar fields. The\nfield equations can be exactly solved for a vanishing, and a constant\npotential, respectively, with the corresponding string tension taking both\nnegative and positive values. Furthermore, for more general classes of\npotentials, having an additive and a multiplicative algebraic structure in the\ntwo scalar fields, the gravitational field equations are solved numerically.\nFor each potential we investigate the effects of the variations of the\npotential parameters and of the boundary conditions on the structure of the\ncosmic string. In this way, we obtain a large class of stable stringlike\nastrophysical configurations, whose basic parameters (string tension and\nradius) depend essentially on the effective field potential, and on the\nboundary conditions.\n"}
{"abstract": "  We consider the problem of partitioning a spectrum band into M channels of\nequal bandwidth, and then further assigning these M channels into P licensed\nchannels and M-P unlicensed channels. Licensed channels can be accessed both\nfor licensed and opportunistic use following a tiered structure which has a\nhigher priority for licensed use. Unlicensed channels can be accessed only for\nopportunistic use. We address the following question in this paper. Given a\nmarket setup, what values of M and P maximize the net spectrum utilization of\nthe spectrum band? While this problem is of fundamental nature, it is highly\nrelevant practically, e.g., in the context of partitioning the recently\nproposed Citizens Broadband Radio Service band. If M is too high or too low, it\nmay decrease spectrum utilization due to limited channel capacity or due to\nwastage of channel capacity, respectively. If P is too high (low), it will not\nincentivize the wireless operators who are primarily interested in unlicensed\nchannels (licensed channels) to join the market. These tradeoffs are captured\nin our optimization problem which manifests itself as a two-stage Stackelberg\ngame. We design an algorithm to solve the Stackelberg game and hence find the\noptimal M and P. The algorithm design also involves an efficient Monte Carlo\nintegrator to evaluate the expected value of the involved random variables like\nspectrum utilization and operators' revenue. We also benchmark our algorithms\nusing numerical simulations.\n"}
{"abstract": "  Software developers are social creatures: they communicate, collaborate, and\npromote their work in a variety of channels. Twitter, GitHub, Stack Overflow,\nand other platforms offer developers opportunities to network and exchange\nideas. Researchers analyze content on these sites to learn about trends and\ntopics in software engineering. However, insight mined from the text of Stack\nOverflow questions or GitHub issues is highly focused on detailed and technical\naspects of software development. In this paper, we present a relatively new\nonline community for software developers called DEV. On DEV users write\nlong-form posts about their experiences, preferences, and working life in\nsoftware, zooming out from specific issues and files to reflect on broader\ntopics. About 50,000 users have posted over 140,000 articles related to\nsoftware development. In this work, we describe the content of posts on DEV\nusing a topic model, showing that developers discuss a rich variety and mixture\nof social and technical aspects of software development. We show that\ndevelopers use DEV to promote themselves and their work: 83% link their\nprofiles to their GitHub profiles and 56% to their Twitter profiles. 14% of\nusers pin specific GitHub repos in their profiles. We argue that DEV is\nemerging as an important hub for software developers, and a valuable source of\ninsight for researchers to complement data from platforms like GitHub and Stack\nOverflow.\n"}
{"abstract": "  First-generation long-distance quantum repeater networks require quantum\nmemories capable of interfacing with telecom photons to perform\nquantum-interference-mediated entanglement generation operations. The ability\nto demonstrate these interconnections using real-life fiber connections in a\nlong-distance setting is paramount to realize a scalable quantum internet. Here\nwe address these significant challenges by observing Hong-Ou-Mandel (HOM)\ninterference between indistinguishable telecom photons produced in two\nindependent room temperature quantum memories, separated by a distance of 158\nkm. We obtained interference visibilities after long-distance propagation of\n$\\rm \\boldsymbol{V=(38\\pm2)\\%}$ for single-photon level experimental inputs.\nThis first-of-its-kind quantum network prototype connecting quantum\nlaboratories in Stony Brook University and Brookhaven National Laboratory is\nenvisioned to evolve into a large-scale memory-assisted entanglement\ndistribution quantum network, the basis for inter-city quantum communication.\n"}
{"abstract": "  As environments involving both robots and humans become increasingly common,\nso does the need to account for people during planning. To plan effectively,\nrobots must be able to respond to and sometimes influence what humans do. This\nrequires a human model which predicts future human actions. A simple model may\nassume the human will continue what they did previously; a more complex one\nmight predict that the human will act optimally, disregarding the robot;\nwhereas an even more complex one might capture the robot's ability to influence\nthe human. These models make different trade-offs between computational time\nand performance of the resulting robot plan. Using only one model of the human\neither wastes computational resources or is unable to handle critical\nsituations. In this work, we give the robot access to a suite of human models\nand enable it to assess the performance-computation trade-off online. By\nestimating how an alternate model could improve human prediction and how that\nmay translate to performance gain, the robot can dynamically switch human\nmodels whenever the additional computation is justified. Our experiments in a\ndriving simulator showcase how the robot can achieve performance comparable to\nalways using the best human model, but with greatly reduced computation.\n"}
{"abstract": "  Continual learning has become increasingly important as it enables NLP models\nto constantly learn and gain knowledge over time. Previous continual learning\nmethods are mainly designed to preserve knowledge from previous tasks, without\nmuch emphasis on how to well generalize models to new tasks. In this work, we\npropose an information disentanglement based regularization method for\ncontinual learning on text classification. Our proposed method first\ndisentangles text hidden spaces into representations that are generic to all\ntasks and representations specific to each individual task, and further\nregularizes these representations differently to better constrain the knowledge\nrequired to generalize. We also introduce two simple auxiliary tasks: next\nsentence prediction and task-id prediction, for learning better generic and\nspecific representation spaces. Experiments conducted on large-scale benchmarks\ndemonstrate the effectiveness of our method in continual text classification\ntasks with various sequences and lengths over state-of-the-art baselines. We\nhave publicly released our code at https://github.com/GT-SALT/IDBR.\n"}
{"abstract": "  Partially Observable Monte-Carlo Planning (POMCP) is a powerful online\nalgorithm able to generate approximate policies for large Partially Observable\nMarkov Decision Processes. The online nature of this method supports\nscalability by avoiding complete policy representation. The lack of an explicit\nrepresentation however hinders policy interpretability and makes policy\nverification very complex. In this work, we propose two contributions. The\nfirst is a method for identifying unexpected actions selected by POMCP with\nrespect to expert prior knowledge of the task. The second is a shielding\napproach that prevents POMCP from selecting unexpected actions. The first\nmethod is based on Satisfiability Modulo Theory (SMT). It inspects traces\n(i.e., sequences of belief-action-observation triplets) generated by POMCP to\ncompute the parameters of logical formulas about policy properties defined by\nthe expert. The second contribution is a module that uses online the logical\nformulas to identify anomalous actions selected by POMCP and substitutes those\nactions with actions that satisfy the logical formulas fulfilling expert\nknowledge. We evaluate our approach on Tiger, a standard benchmark for POMDPs,\nand a real-world problem related to velocity regulation in mobile robot\nnavigation. Results show that the shielded POMCP outperforms the standard POMCP\nin a case study in which a wrong parameter of POMCP makes it select wrong\nactions from time to time. Moreover, we show that the approach keeps good\nperformance also if the parameters of the logical formula are optimized using\ntrajectories containing some wrong actions.\n"}
{"abstract": "  In this paper we study the electromagnetic scattering, absorption, and\nperformance bounds for short time modulated pulses that impinge on a\ntime-varying lossy layer that is sandwiched between vacuum and a perfect\nelectric conductor. The electric characteristics of the layer, namely, the\nconductivity, permittivity, and permeability are assumed to change abruptly or\ngradually in time. We demonstrate numerically that a time-varying absorbing\nlayer that undergoes temporal switching of its permittivity and conductance can\nabsorb the power of a modulated, ultra-wideband, as well as a\nquasi-monochromatic, pulsed wave beyond what is dictated by the time invariant\nRozanov bound when integrating over the whole frequency spectrum. We suggest\nand simulate a practical metamaterial realization that is constructed as a\nthree-dimensional array of resistor loaded dipole. By switching only the\ndipole's load resistance, desired effective media properties are obtained.\nFurthermore, we show that Rozanov's bound can be bypassed with abrupt and a\nmore practical gradual, soft, switching thus overcoming some possible causality\nissue in abrupt switching.\n"}
{"abstract": "  A novel algorithm named Perturbed Prox-Preconditioned SPIDER (3P-SPIDER) is\nintroduced. It is a stochastic variancereduced proximal-gradient type algorithm\nbuilt on Stochastic Path Integral Differential EstimatoR (SPIDER), an algorithm\nknown to achieve near-optimal first-order oracle inequality for nonconvex and\nnonsmooth optimization. Compared to the vanilla prox-SPIDER, 3P-SPIDER uses\npreconditioned gradient estimators. Preconditioning can either be applied\n\"explicitly\" to a gradient estimator or be introduced \"implicitly\" as in\napplications to the EM algorithm. 3P-SPIDER also assumes that the\npreconditioned gradients may (possibly) be not known in closed analytical form\nand therefore must be approximated which adds an additional degree of\nperturbation. Studying the convergence in expectation, we show that 3P-SPIDER\nachieves a near-optimal oracle inequality O(n^(1/2) /epsilon) where n is the\nnumber of observations and epsilon the target precision even when the gradient\nis estimated by Monte Carlo methods. We illustrate the algorithm on an\napplication to the minimization of a penalized empirical loss.\n"}
{"abstract": "  Detection in large-scale scenes is a challenging problem due to small objects\nand extreme scale variation. It is essential to focus on the image regions of\nsmall objects. In this paper, we propose a novel Adaptive Zoom (AdaZoom)\nnetwork as a selective magnifier with flexible shape and focal length to\nadaptively zoom the focus regions for object detection. Based on policy\ngradient, we construct a reinforcement learning framework for focus region\ngeneration, with the reward formulated by object distributions. The scales and\naspect ratios of the generated regions are adaptive to the scales and\ndistribution of objects inside. We apply variable magnification according to\nthe scale of the region for adaptive multi-scale detection. We further propose\ncollaborative training to complementarily promote the performance of AdaZoom\nand the detection network. To validate the effectiveness, we conduct extensive\nexperiments on VisDrone2019, UAVDT, and DOTA datasets. The experiments show\nAdaZoom brings a consistent and significant improvement over different\ndetection networks, achieving state-of-the-art performance on these datasets,\nespecially outperforming the existing methods by AP of 4.64% on Vis-Drone2019.\n"}
{"abstract": "  We consider the structure of the point-line incidence matrix of the\nprojective space $\\mathrm{PG}(3,q)$ connected with orbits of points and lines\nunder the stabilizer group of the twisted cubic. Structures of submatrices with\nincidences between a union of line orbits and an orbit of points are\ninvestigated. For the unions consisting of two or three line orbits, the\noriginal submatrices are split into new ones, in which the incidences are also\nconsidered. For each submatrix (apart from the ones corresponding to a special\ntype of lines), the numbers of lines through every point and of points lying on\nevery line are obtained. This corresponds to the numbers of ones in columns and\nrows of the submatrices.\n"}
{"abstract": "  We address the problem of weakly-supervised semantic segmentation (WSSS)\nusing bounding box annotations. Although object bounding boxes are good\nindicators to segment corresponding objects, they do not specify object\nboundaries, making it hard to train convolutional neural networks (CNNs) for\nsemantic segmentation. We find that background regions are perceptually\nconsistent in part within an image, and this can be leveraged to discriminate\nforeground and background regions inside object bounding boxes. To implement\nthis idea, we propose a novel pooling method, dubbed background-aware pooling\n(BAP), that focuses more on aggregating foreground features inside the bounding\nboxes using attention maps. This allows to extract high-quality pseudo\nsegmentation labels to train CNNs for semantic segmentation, but the labels\nstill contain noise especially at object boundaries. To address this problem,\nwe also introduce a noise-aware loss (NAL) that makes the networks less\nsusceptible to incorrect labels. Experimental results demonstrate that learning\nwith our pseudo labels already outperforms state-of-the-art weakly- and\nsemi-supervised methods on the PASCAL VOC 2012 dataset, and the NAL further\nboosts the performance.\n"}
{"abstract": "  This paper introduces a family of valid inequalities, that we term\nconsistency cuts, to be applied to a Dantzig-Wolfe reformulation (or\ndecomposition) with linking variables. We prove that these cuts ensure an\ninteger solution to the corresponding Dantzig-Wolfe relaxation when certain\ncriteria to the structure of the decomposition are met. We implement the cuts\nand use them to solve a commonly used test set of 200 instances of the temporal\nknapsack problem. We assess the performance with and without the cuts and\ncompare further to CPLEX and other solution methods that have historically been\nused to solve the test set. By separating consistency cuts we show that we can\nobtain optimal integer solutions much faster than the other methods and even\nsolve the remaining unsolved problems in the test set. We also perform a second\ntest on instances from the MIPLIB 2017 online library of mixed-integer\nprograms, showing the potential of the cuts on a wider range of problems.\n"}
{"abstract": "  We study homogenization of a boundary obstacle problem on $ C^{1,\\alpha} $\ndomain $D$ for some elliptic equations with uniformly elliptic coefficient\nmatrices $\\gamma$. For any $ \\epsilon\\in\\mathbb{R}_+$, $\\partial D=\\Gamma \\cup\n\\Sigma$, $\\Gamma \\cap \\Sigma=\\emptyset $ and $ S_{\\epsilon}\\subset \\Sigma $\nwith suitable assumptions,\\ we prove that as $\\epsilon$ tends to zero, the\nenergy minimizer $ u^{\\epsilon} $ of $ \\int_{D} |\\gamma\\nabla u|^{2} dx $,\nsubject to $ u\\geq \\varphi $ on $ S_{\\varepsilon} $, up to a subsequence,\nconverges weakly in $ H^{1}(D) $ to $ \\widetilde{u} $ which minimizes the\nenergy functional $\\int_{D}|\\gamma\\nabla u|^{2}+\\int_{\\Sigma}\n(u-\\varphi)^{2}_{-}\\mu(x) dS_{x}$, where $\\mu(x)$ depends on the structure of\n$S_{\\epsilon}$ and $ \\varphi $ is any given function in\n$C^{\\infty}(\\overline{D})$.\n"}
{"abstract": "  To improve the system performance towards the Shannon limit, advanced radio\nresource management mechanisms play a fundamental role. In particular,\nscheduling should receive much attention, because it allocates radio resources\namong different users in terms of their channel conditions and QoS\nrequirements. The difficulties of scheduling algorithms are the tradeoffs need\nto be made among multiple objectives, such as throughput, fairness and packet\ndrop rate. We propose a smart scheduling scheme based on deep reinforcement\nlearning (DRL). We not only verify the performance gain achieved, but also\nprovide implementation-friend designs, i.e., a scalable neural network design\nfor the agent and a virtual environment training framework. With the scalable\nneural network design, the DRL agent can easily handle the cases when the\nnumber of active users is time-varying without the need to redesign and retrain\nthe DRL agent. Training the DRL agent in a virtual environment offline first\nand using it as the initial version in the practical usage helps to prevent the\nsystem from suffering from performance and robustness degradation due to the\ntime-consuming training. Through both simulations and field tests, we show that\nthe DRL-based smart scheduling outperforms the conventional scheduling method\nand can be adopted in practical systems.\n"}
{"abstract": "  First principles investigations of the high temperature superconducting\nsystem Ba$_2$CuO$_{3+\\delta}$, recently discovered at $\\delta\\approx0.2$ at\n$T_c=70$ K, are applied to demonstrate the effects of oxygen ordering on the\nelectronic and magnetic properties. The observed `highly over-doped'\nsuperconducting phase displays stretched Cu-planar oxygen O$_{\\rm P}$ distances\nand anomalously shortened Cu-apical O$_{\\rm A}$ separations compared with other\ncuprates. The stoichiometric system $\\delta=0$, with its strongly\none-dimensional (1D) Cu-O$_{\\rm P}$ chain structure, when nonmagnetic shows 1D\nFermi surfaces that lead, within density functional theory, to\nantiferromagnetic Cu-O$_{\\rm P}$ chains (a spin-Peierls instability).\nAccounting for 1D fluctuations and small interchain coupling according to the\ntheory of Schulz indicates this system, like Sr$_2$CuO$_3$, is near the 1D\nLuttinger-liquid quantum critical phase. The unusual Cu-O bond lengths per se\nhave limited effects on other properties for $\\delta$=0. We find that a\n`doubled bilayer' structure of alternating Cu-O$_{\\rm P}$ chains and wide rung\nCu$_3$O$_4$ ladders is the energetically preferred one of three possibilities\nwhere the additional oxygen ions bridge Cu-O$_{\\rm P}$ chains in the\nsuperconducting phase $\\delta=1/4$. Nominal formal valences of the three Cu\nsites are discussed. The six-fold (octahedral) site is the most highly\noxidized, accepting somewhat more holes in the $d_{z^2}$ orbital than in the\n$d_{x^2-y^2}$ orbital. The implication is that two-band physics is involved in\nthe pairing mechanism and the superconducting carriers. The Fermi surfaces of\nthis metallic bilayer structure show both 1D and 2D strong (incipient) nesting\ninstabilities, possibly accounting for the lack of clean single-phase samples\nbased on this structure and suggesting importance for the pairing mechanism.\n"}
{"abstract": "  In this paper, we investigate the problem of coordination between economic\ndispatch (ED) and demand response (DR) in multi-energy systems (MESs), aiming\nto improve the economic utility and reduce the waste of energy in MESs. Since\nmultiple energy sources are coupled through energy hubs (EHs), the\nsupply-demand constraints are nonconvex. To deal with this issue, we propose a\nlinearization method to transform the coordination problem to a convex social\nwelfare optimization one. Then a decentralized algorithm based on parallel\nAlternating Direction Method of Multipliers (ADMM) and dynamic average tracking\nprotocol is developed, where each agent could only make decisions based on\ninformation from their neighbors. Moreover, by using variational inequality and\nLyapunov-based techniques, we show that our algorithm could always converge to\nthe global optimal solution. Finally, a case study on the modified IEEE 14-bus\nnetwork verifies the feasibility and effectiveness of our algorithm.\n"}
{"abstract": "  In this paper, we propose a fully automatic system for generating comic books\nfrom videos without any human intervention. Given an input video along with its\nsubtitles, our approach first extracts informative keyframes by analyzing the\nsubtitles, and stylizes keyframes into comic-style images. Then, we propose a\nnovel automatic multi-page layout framework, which can allocate the images\nacross multiple pages and synthesize visually interesting layouts based on the\nrich semantics of the images (e.g., importance and inter-image relation).\nFinally, as opposed to using the same type of balloon as in previous works, we\npropose an emotion-aware balloon generation method to create different types of\nword balloons by analyzing the emotion of subtitles and audios. Our method is\nable to vary balloon shapes and word sizes in balloons in response to different\nemotions, leading to more enriched reading experience. Once the balloons are\ngenerated, they are placed adjacent to their corresponding speakers via speaker\ndetection. Our results show that our method, without requiring any user inputs,\ncan generate high-quality comic pages with visually rich layouts and balloons.\nOur user studies also demonstrate that users prefer our generated results over\nthose by state-of-the-art comic generation systems.\n"}
{"abstract": "  We introduce a state-of-the-art audio-visual on-screen sound separation\nsystem which is capable of learning to separate sounds and associate them with\non-screen objects by looking at in-the-wild videos. We identify limitations of\nprevious work on audio-visual on-screen sound separation, including the\nsimplicity and coarse resolution of spatio-temporal attention, and poor\nconvergence of the audio separation model. Our proposed model addresses these\nissues using cross-modal and self-attention modules that capture audio-visual\ndependencies at a finer resolution over time, and by unsupervised pre-training\nof audio separation model. These improvements allow the model to generalize to\na much wider set of unseen videos. We also show a robust way to further improve\nthe generalization capability of our models by calibrating the probabilities of\nour audio-visual on-screen classifier, using only a small amount of in-domain\nvideos labeled for their on-screen presence. For evaluation and semi-supervised\ntraining, we collected human annotations of on-screen audio from a large\ndatabase of in-the-wild videos (YFCC100m). Our results show marked improvements\nin on-screen separation performance, in more general conditions than previous\nmethods.\n"}
{"abstract": "  We show that if a sequence $M_n$ of closed aspherical $d$-dimensional\nRiemannian manifolds with Ricci curvature uniformly bounded below and diameter\nuniformly bounded above collapses, then for all large enough $n$, the\nfundamental groups $\\pi_1(M_n)$ have non-trivial finitely generated abelian\nnormal subgroups. In particular, the groups $\\pi_1(M_n)$ cannot be\nnon-elementary hyperbolic.\n"}
{"abstract": "  Cancer diseases constitute one of the most significant societal challenges.\nIn this paper we introduce a novel histopathological dataset for prostate\ncancer detection. The proposed dataset, consisting of over 2.6 million tissue\npatches extracted from 430 fully annotated scans, 4675 scans with assigned\nbinary diagnosis, and 46 scans with diagnosis given independently by a group of\nhistopathologists, can be found at https://ai-econsilio.diag.pl. Furthermore,\nwe propose a machine learning framework for detection of cancerous tissue\nregions and prediction of scan-level diagnosis, utilizing thresholding and\nstatistical analysis to abstain from the decision in uncertain cases. During\nthe experimental evaluation we identify several factors negatively affecting\nthe performance of considered models, such as presence of label noise, data\nimbalance, and quantity of data, that can serve as a basis for further\nresearch. The proposed approach, composed of ensembles of deep neural networks\noperating on the histopathological scans at different scales, achieves 94.6%\naccuracy in patch-level recognition, and is compared in a scan-level diagnosis\nwith 9 human histopathologists.\n"}
{"abstract": "  This paper is concerned with the output feedback boundary stabilization of\ngeneral 1-D reaction diffusion PDEs in the presence of an arbitrarily large\ninput delay. We consider the cases of Dirichlet/Neumann/Robin boundary\nconditions for the both boundary control and boundary condition. The boundary\nmeasurement takes the form of a either Dirichlet or Neumann trace. The adopted\ncontrol strategy is composed of a finite-dimensional observer estimating the\nfirst modes of the PDE coupled with a predictor to compensate the input delay.\nIn this context, we show for any arbitrary value of the input delay that the\ncontrol strategy achieves the exponential stabilization of the closed-loop\nsystem, for system trajectories evaluated in $H^1$ norm (also in $L^2$ norm in\nthe case of a Dirichlet boundary measurement), provided the dimension of the\nobserver is selected large enough. The reported proof of this result requires\nto perform both control design and stability analysis using simultaneously the\n(non-homogeneous) original version of the PDE and one of its equivalent\nhomogeneous representations.\n"}
{"abstract": "  For graphene nanoribbons with Rashba spin-orbit coupling, the peculiar\nmagnetic response due to the presence of a magnetization and geometric\nconfinement are analyzed within a tight-binding model. We observe a sizable\ntransverse susceptibility that can be considered as a gate voltage-induced\nmagnetoelectric torque without the need of a bias voltage, with different\ndirections for zigzag and armchair ribbons. The local torque generates\nnon-collinear spin polarization between the two edges and/or along the ribbon,\nand the net torque averages to zero if the magnetization is homogeneous.\nNevertheless, a nonzero net torque can appear in partially magnetized\nnanoribbons or in nanoflakes of irregular shapes. The equilibrium spin current\nproduced by the spin-orbit coupling also appears in nanoribbons, but the\ncomponent flowing in the direction of confinement is strongly suppressed. Even\nwithout the magnetization, an out-of-plane polarized chiral edge spin current\nis produced, resembling that in the quantum spin Hall effect. Moreover, a\nmagnetization pointing perpendicular to the edge produces a laminar flow of\nedge charge currents, whose flow direction is symmetric (non chiral) or\nantisymmetric (chiral) between the two edges depends on whether the\nmagnetization points in-plane or out-of-plane.\n"}
{"abstract": "  In many applications, the governing PDE to be solved numerically contains a\nstiff component. When this component is linear, an implicit time stepping\nmethod that is unencumbered by stability restrictions is often preferred. On\nthe other hand, if the stiff component is nonlinear, the complexity and cost\nper step of using an implicit method is heightened, and explicit methods may be\npreferred for their simplicity and ease of implementation. In this article, we\nanalyze new and existing linearly stabilized schemes for the purpose of\nintegrating stiff nonlinear PDEs in time. These schemes compute the nonlinear\nterm explicitly and, at the cost of solving a linear system with a matrix that\nis fixed throughout, are unconditionally stable, thus combining the advantages\nof explicit and implicit methods. Applications are presented to illustrate the\nuse of these methods.\n"}
{"abstract": "  With the rapid evolution of social media, fake news has become a significant\nsocial problem, which cannot be addressed in a timely manner using manual\ninvestigation. This has motivated numerous studies on automating fake news\ndetection. Most studies explore supervised training models with different\nmodalities (e.g., text, images, and propagation networks) of news records to\nidentify fake news. However, the performance of such techniques generally drops\nif news records are coming from different domains (e.g., politics,\nentertainment), especially for domains that are unseen or rarely-seen during\ntraining. As motivation, we empirically show that news records from different\ndomains have significantly different word usage and propagation patterns.\nFurthermore, due to the sheer volume of unlabelled news records, it is\nchallenging to select news records for manual labelling so that the\ndomain-coverage of the labelled dataset is maximized. Hence, this work: (1)\nproposes a novel framework that jointly preserves domain-specific and\ncross-domain knowledge in news records to detect fake news from different\ndomains; and (2) introduces an unsupervised technique to select a set of\nunlabelled informative news records for manual labelling, which can be\nultimately used to train a fake news detection model that performs well for\nmany domains while minimizing the labelling cost. Our experiments show that the\nintegration of the proposed fake news model and the selective annotation\napproach achieves state-of-the-art performance for cross-domain news datasets,\nwhile yielding notable improvements for rarely-appearing domains in news\ndatasets.\n"}
{"abstract": "  Variational quantum algorithms (VQAs) promise efficient use of near-term\nquantum computers. However, training VQAs often requires an extensive amount of\ntime and suffers from the barren plateau problem where the magnitude of the\ngradients vanishes with increasing number of qubits. Here, we show how to\noptimally train VQAs for learning quantum states. Parameterized quantum\ncircuits can form Gaussian kernels, which we use to derive adaptive learning\nrates for gradient ascent. We introduce the generalized quantum natural\ngradient that features stability and optimized movement in parameter space.\nBoth methods together outperform other optimization routines in training VQAs.\nOur methods also excel at numerically optimizing driving protocols for quantum\ncontrol problems. The gradients of the VQA do not vanish when the fidelity\nbetween the initial state and the state to be learned is bounded from below. We\nidentify a VQA for quantum simulation with such a constraint that thus can be\ntrained free of barren plateaus. Finally, we propose the application of\nGaussian kernels for quantum machine learning.\n"}
{"abstract": "  NASA's Kepler, K2 and TESS missions employ Simple Aperture Photometry (SAP)\nto derive time-series photometry, where an aperture is estimated for each star,\nand pixels containing each star are summed to create a single light curve. This\nmethod is simple, but in crowded fields the derived time-series can be highly\ncontaminated. The alternate method of fitting a Point Spread Function (PSF) to\nthe data is able to account for crowding, but is computationally expensive. In\nthis paper, we present a new approach to extracting photometry from these\ntime-series missions, which fits the PSF directly, but makes simplifying\nassumptions in order to greatly reduce the computation expense. Our method\nfixes the scene of the field in each image, estimates the PSF shape of the\ninstrument with a linear model, and allows only source flux and position to\nvary. We demonstrate that our method is able to separate the photometry from\nblended targets in the Kepler dataset that are separated by less than a pixel.\nOur method is fast to compute, and fully accounts for uncertainties from\ndegeneracies due to crowded fields. We name the method described in this work\nLinearized Field Deblending (LFD). We demonstrate our method on the false\npositive Kepler target \\koi. We are able to separate the photometry of the two\nsources in the data, and demonstrate the contaminating transiting signal is\nconsistent with a small, sub-stellar companion with a radius of $2.67R_{jup}$\n($0.27R_{sol}$). Our method is equally applicable to extracting photometry from\nNASA's TESS mission.\n"}
{"abstract": "  The main purpose of this paper is to describe the category of isotropic\ncellular spectra over flexible fields. Guided by [9], we show that it is\nequivalent as a stable $\\infty$-category equipped with a $t$-structure to the\nderived category of left comodules over the dual of the classical topological\nSteenrod algebra. In order to obtain this result, the category of isotropic\ncellular modules over the motivic Brown-Peterson spectrum is also studied and\nisotropic Adams and Adams-Novikov spectral sequences are developed. As a\nconsequence, we also compute hom-sets in the category of isotropic Tate motives\nbetween motives of isotropic cellular spectra.\n"}
{"abstract": "  As neural networks gain widespread adoption in embedded devices, there is a\nneed for model compression techniques to facilitate deployment in\nresource-constrained environments. Quantization is one of the go-to methods\nyielding state-of-the-art model compression. Most approaches take a fully\ntrained model, apply different heuristics to determine the optimal\nbit-precision for different layers of the network, and retrain the network to\nregain any drop in accuracy. Based on Activation Density (AD)-the proportion of\nnon-zero activations in a layer-we propose an in-training quantization method.\nOur method calculates bit-width for each layer during training yielding a mixed\nprecision model with competitive accuracy. Since we train lower precision\nmodels during training, our approach yields the final quantized model at lower\ntraining complexity and also eliminates the need for re-training. We run\nexperiments on benchmark datasets like CIFAR-10, CIFAR-100, TinyImagenet on\nVGG19/ResNet18 architectures and report the accuracy and energy estimates for\nthe same. We achieve ~4.5x benefit in terms of estimated\nmultiply-and-accumulate (MAC) reduction while reducing the training complexity\nby 50% in our experiments. To further evaluate the energy benefits of our\nproposed method, we develop a mixed-precision scalable Process In Memory (PIM)\nhardware accelerator platform. The hardware platform incorporates shift-add\nfunctionality for handling multi-bit precision neural network models.\nEvaluating the quantized models obtained with our proposed method on the PIM\nplatform yields ~5x energy reduction compared to 16-bit models. Additionally,\nwe find that integrating AD based quantization with AD based pruning (both\nconducted during training) yields up to ~198x and ~44x energy reductions for\nVGG19 and ResNet18 architectures respectively on PIM platform compared to\nbaseline 16-bit precision, unpruned models.\n"}
{"abstract": "  The lanthanide elements have well-documented similarities in their chemical\nbehavior, which makes the valuable trivalent lanthanide cations (Ln(III))\nparticularly difficult to separate from each other in water. In this work, we\napply ab initio molecular dynamics simulations to compare the free energies\n(Delta G(ads)) associated with the adsorption of lanthanide cations to silica\nsurfaces at a pH condition where SiO- groups are present. The predicted Delta\nG(ads) for lutetium (Lu(III)) and europium (Eu(III)) are similar within\nstatistical uncertainties; this is in qualitative agreement with our batch\nadsorption measurements on silica. This finding is remarkable because the two\ncations exhibit hydration free energies (Delta G(hyd}) that differ by >2 eV,\ndifferent hydration numbers, and different hydrolysis behavior far from silica\nsurfaces. We observe that the similarity in Lu(III) and Eu(III) Delta G(ads) is\nthe result of a delicate cancellation between the difference in Eu(III) and\nLu(III) hydration (Delta G(hyd})), and their difference in binding energies to\nsilica. We propose that disrupting this cancellation at the two end points,\neither for adsorbed or completely desorbed lanthanides (e.g., via\nnanoconfinment or mixed solvents), will lead to effective Ln separation.\n"}
{"abstract": "  The main objective of the M5 competition, which focused on forecasting the\nhierarchical unit sales of Walmart, was to evaluate the accuracy and\nuncertainty of forecasting methods in the field in order to identify best\npractices and highlight their practical implications. However, whether the\nfindings of the M5 competition can be generalized and exploited by retail firms\nto better support their decisions and operation depends on the extent to which\nthe M5 data is sufficiently similar to unit sales data of retailers that\noperate in different regions, sell different types of products, and consider\ndifferent marketing strategies. To answer this question, we analyze the\ncharacteristics of the M5 time series and compare them with those of two\ngrocery retailers, namely Corporaci\\'on Favorita and a major Greek supermarket\nchain, using feature spaces. Our results suggest that there are only small\ndiscrepancies between the examined data sets, supporting the representativeness\nof the M5 data.\n"}
{"abstract": "  We derive a general expression for the threshold resummation of transverse\nmomentum distributions for processes with a colorless final state, by suitably\ngeneralizing the renormalization-group based approach to threshold resummation\npreviously pursued by two of us. The ensuing expression holds to all\nlogarithmic orders, and it can be used to extend available results in the\nliterature, which only hold up to the next-to-leading log (NLL) level. We check\nagreement of our result with the existing NLL result, as well as against the\nknown fixed next-to-leading order results for the Higgs transverse momentum\ndistribution in gluon fusion, and we provide explicit expressions at the\nnext-to-next-to-leading log level.\n"}
{"abstract": "  We utilise an isobar model to investigate the $K^+ \\Sigma^-$ photoproduction\noff a neutron in the resonance region. Except for the Born terms, we include\nhigh-spin (spin-3/2 and spin-5/2) nucleon resonances in the consistent\nformalism together with a few $\\Delta$ and kaon resonances to achieve an\nacceptable agreement with data. Interestingly, we reveal that no hyperon\nresonances are needed to achieve a reasonable description of data. On the other\nhand the $N(1720)3/2^+$ resonance was found to be very important for correct\ndescription of data. The free parameters of the model were fitted to\nexperimental data from the LEPS and CLAS Collaborations on either differential\ncross sections or photon beam asymmetry. The novel feature of the fitting\nprocedure is the use of a regularization method, the Least Absolute Shrinkage\nSelection Operator, and information criteria in order to choose the best fit.\n"}
{"abstract": "  We show two sphere theorems for the Riemannian manifolds with scalar\ncurvature bounded below and the non-collapsed $\\mathrm{RCD}(n-1,n)$ spaces with\nmean distance close to $\\frac{\\pi}{2}$.\n"}
{"abstract": "  The two-dimensional transition metal dichalcogenides (TMDs) have been\nproposed as candidates for the channel material in future field effect\ntransistor designs. The heterophase design which utilizes the metallic T- or T'\nphase of the TMD as contacts to the semiconducting H phase channel has shown\npromising results in terms of bringing down the contact resistance of the\ndevice. In this work, we use ab-initio calculations to demonstrate how\natomic-scale and quantum effects influence the ballistic transport properties\nin such heterophase transistors with channel lengths up to 20 nm. We\ninvestigate how the charge transfer depends on the carrier density both in T'-H\nMoTe$_2$ Schottky contacts and planar T'-H-T' MoTe$_2$ transistors. We find\nthat the size of the Schottky barrier and the charge transfer is dominated by\nthe local atomic arrangements at the interface and the doping level.\nFurthermore, two types of quantum states have a large influence on the charge\ntransport; interface states and standing waves in the semiconductor due to\nquantum confinement. We find that the latter can be associated with rises in\nthe current by more than an order of magnitude due to resonant tunneling. Our\nresults demonstrate the quantum mechanical nature of these 2D transistors and\nhighlight several challenges and possible solutions for achieving a competitive\nperformance of such devices.\n"}
{"abstract": "  Graph representation learning has become a ubiquitous component in many\nscenarios, ranging from social network analysis to energy forecasting in smart\ngrids. In several applications, ensuring the fairness of the node (or graph)\nrepresentations with respect to some protected attributes is crucial for their\ncorrect deployment. Yet, fairness in graph deep learning remains\nunder-explored, with few solutions available. In particular, the tendency of\nsimilar nodes to cluster on several real-world graphs (i.e., homophily) can\ndramatically worsen the fairness of these procedures. In this paper, we propose\na novel biased edge dropout algorithm (FairDrop) to counter-act homophily and\nimprove fairness in graph representation learning. FairDrop can be plugged in\neasily on many existing algorithms, is efficient, adaptable, and can be\ncombined with other fairness-inducing solutions. After describing the general\nalgorithm, we demonstrate its application on two benchmark tasks, specifically,\nas a random walk model for producing node embeddings, and to a graph\nconvolutional network for link prediction. We prove that the proposed algorithm\ncan successfully improve the fairness of all models up to a small or negligible\ndrop in accuracy, and compares favourably with existing state-of-the-art\nsolutions. In an ablation study, we demonstrate that our algorithm can flexibly\ninterpolate between biasing towards fairness and an unbiased edge dropout.\nFurthermore, to better evaluate the gains, we propose a new dyadic group\ndefinition to measure the bias of a link prediction task when paired with\ngroup-based fairness metrics. In particular, we extend the metric used to\nmeasure the bias in the node embeddings to take into account the graph\nstructure.\n"}
{"abstract": "  Most heat transfer models for bulk crystal growth rely on the classical\nStefan formulation to evaluate interface motion during phase change. However,\nwhen the interface is non-smooth the use of the classical Stefan formulation\nmay lead to singularities. To address this problem, we propose a simulation\nmodel based on the weak formulation of the Stefan problem. Numerical solutions\nof the weak Stefan formulation are obtained using the finite volume method.\nThis approach provides an energy conserving discretization scheme that\naccurately evaluates heat transfer around non-smooth interfaces. We apply the\nweak formulation to numerically simulate the solidification of silicon in the\nhorizontal ribbon growth process. Results exhibit a limitation on the ribbon's\npull speed, which previous classical Stefan models have failed to demonstrate.\nA comparison of heat transfer between radiation and gas cooling shows that gas\ncooling increases the pull speed limit for the same amount of heat removed.\n"}
{"abstract": "  Despite the rapid progress in understanding the first intrinsic magnetic\ntopological insulator MnBi$_2$Te$_4$, its electronic structure remains a topic\nunder debates. Here we perform a thorough spectroscopic investigation into the\nelectronic structure of MnBi$_2$Te$_4$ via laser-based angle-resolved\nphotoemission spectroscopy. Through quantitative analysis, we estimate an upper\nbound of 3 meV for the gap size of the topological surface state. Furthermore,\nour circular dichroism measurements reveal band chiralities for both the\ntopological surface state and quasi-2D bands, which can be well reproduced in a\nband hybridization model. A numerical simulation of energy-momentum dispersions\nbased on a four-band model with an additional step potential near the surface\nprovides a promising explanation for the origin of the quasi-2D bands. Our\nstudy represents a solid step forward in reconciling the existing controversies\nin the electronic structure of MnBi$_2$Te$_4$, and provides an important\nframework to understand the electronic structures of other relevant topological\nmaterials MnBi$_{2n}$Te$_{3n+1}$.\n"}
{"abstract": "  Recently, referring image segmentation has aroused widespread interest.\nPrevious methods perform the multi-modal fusion between language and vision at\nthe decoding side of the network. And, linguistic feature interacts with visual\nfeature of each scale separately, which ignores the continuous guidance of\nlanguage to multi-scale visual features. In this work, we propose an encoder\nfusion network (EFN), which transforms the visual encoder into a multi-modal\nfeature learning network, and uses language to refine the multi-modal features\nprogressively. Moreover, a co-attention mechanism is embedded in the EFN to\nrealize the parallel update of multi-modal features, which can promote the\nconsistent of the cross-modal information representation in the semantic space.\nFinally, we propose a boundary enhancement module (BEM) to make the network pay\nmore attention to the fine structure. The experiment results on four benchmark\ndatasets demonstrate that the proposed approach achieves the state-of-the-art\nperformance under different evaluation metrics without any post-processing.\n"}
{"abstract": "  This article presents a study on the quality and execution of research code\nfrom publicly-available replication datasets at the Harvard Dataverse\nrepository. Research code is typically created by a group of scientists and\npublished together with academic papers to facilitate research transparency and\nreproducibility. For this study, we define ten questions to address aspects\nimpacting research reproducibility and reuse. First, we retrieve and analyze\nmore than 2000 replication datasets with over 9000 unique R files published\nfrom 2010 to 2020. Second, we execute the code in a clean runtime environment\nto assess its ease of reuse. Common coding errors were identified, and some of\nthem were solved with automatic code cleaning to aid code execution. We find\nthat 74\\% of R files crashed in the initial execution, while 56\\% crashed when\ncode cleaning was applied, showing that many errors can be prevented with good\ncoding practices. We also analyze the replication datasets from journals'\ncollections and discuss the impact of the journal policy strictness on the code\nre-execution rate. Finally, based on our results, we propose a set of\nrecommendations for code dissemination aimed at researchers, journals, and\nrepositories.\n"}
{"abstract": "  We study covariant equations in quantum cosmology of an extended\nminisuperspace obtained by the Eisenhart--Duval lift. We find that a Dirac-type\nequation is naturally introduced in the extended minisuperspace. Explicit forms\nof the fundamental solutions are yielded for specific models. The possible\nfurther development in this direction is also discussed.\n"}
{"abstract": "  We show that if a hyperbolic 3-manifold admits a partially hyperbolic\ndiffeomorphism then it also admits an Anosov flow. Moreover, we give a complete\nclassification of partially hyperbolic diffeomorphism in hyperbolic 3-manifolds\nas well as partially hyperbolic diffeomorphisms in Seifert manifolds inducing\npseudo-Anosov dynamics in the base. This classification is given in terms of\nthe structure of their center (branching) foliations and the notion of\ncollapsed Anosov flows.\n"}
{"abstract": "  Understanding the Bondi-Metzner-Sachs (BMS) frame of the gravitational waves\nproduced by numerical relativity is crucial for ensuring that analyses on such\nwaveforms are performed properly. It is also important that models are built\nfrom waveforms in the same BMS frame. Up until now, however, the BMS frame of\nnumerical waveforms has not been thoroughly examined, largely because the\nnecessary tools have not existed. In this paper, we show how to analyze and map\nto a suitable BMS frame for numerical waveforms calculated with the Spectral\nEinstein Code (SpEC). However, the methods and tools that we present are\ngeneral and can be applied to any numerical waveforms. We present an extensive\nstudy of 13 binary black hole systems that broadly span parameter space. From\nthese simulations, we extract the strain and also the Weyl scalars using both\nSpECTRE's Cauchy-characteristic extraction module and also the standard\nextrapolation procedure with a displacement memory correction applied during\npostprocessing. First, we show that the current center-of-mass correction used\nto map these waveforms to the center-of-mass frame is not as effective as\npreviously thought. Consequently, we also develop an improved correction that\nutilizes asymptotic Poincar\\'e charges instead of a Newtonian center-of-mass\ntrajectory. Next, we map our waveforms to the post-Newtonian (PN) BMS frame\nusing a PN strain waveform. This helps us find the unique BMS transformation\nthat minimizes the $L^{2}$ norm of the difference between the numerical and PN\nstrain waveforms during the early inspiral phase. We find that once the\nwaveforms are mapped to the PN BMS frame, they can be hybridized with a PN\nstrain waveform much more effectively than if one used any of the previous\nalignment schemes, which only utilize the Poincar\\'e transformations.\n"}
{"abstract": "  We consider the approximability of constraint satisfaction problems in the\nstreaming setting. For every constraint satisfaction problem (CSP) on $n$\nvariables taking values in $\\{0,\\ldots,q-1\\}$, we prove that improving over the\ntrivial approximability by a factor of $q$ requires $\\Omega(n)$ space even on\ninstances with $O(n)$ constraints. We also identify a broad subclass of\nproblems for which any improvement over the trivial approximability requires\n$\\Omega(n)$ space. The key technical core is an optimal,\n$q^{-(k-1)}$-inapproximability for the case where every constraint is given by\na system of $k-1$ linear equations $\\bmod\\; q$ over $k$ variables. Prior to our\nwork, no such hardness was known for an approximation factor less than $1/2$\nfor any CSP. Our work builds on and extends the work of Kapralov and Krachun\n(Proc. STOC 2019) who showed a linear lower bound on any non-trivial\napproximation of the max cut in graphs. This corresponds roughly to the case of\nMax $k$-LIN-$\\bmod\\; q$ with $k=q=2$. Each one of the extensions provides\nnon-trivial technical challenges that we overcome in this work.\n"}
{"abstract": "  Depth estimation is an important computer vision problem with many practical\napplications to mobile devices. While many solutions have been proposed for\nthis task, they are usually very computationally expensive and thus are not\napplicable for on-device inference. To address this problem, we introduce the\nfirst Mobile AI challenge, where the target is to develop an end-to-end deep\nlearning-based depth estimation solutions that can demonstrate a nearly\nreal-time performance on smartphones and IoT platforms. For this, the\nparticipants were provided with a new large-scale dataset containing RGB-depth\nimage pairs obtained with a dedicated stereo ZED camera producing\nhigh-resolution depth maps for objects located at up to 50 meters. The runtime\nof all models was evaluated on the popular Raspberry Pi 4 platform with a\nmobile ARM-based Broadcom chipset. The proposed solutions can generate VGA\nresolution depth maps at up to 10 FPS on the Raspberry Pi 4 while achieving\nhigh fidelity results, and are compatible with any Android or Linux-based\nmobile devices. A detailed description of all models developed in the challenge\nis provided in this paper.\n"}
{"abstract": "  Aerial navigation in GPS-denied, indoor environments, is still an open\nchallenge. Drones can perceive the environment from a richer set of viewpoints,\nwhile having more stringent compute and energy constraints than other\nautonomous platforms. To tackle that problem, this research displays a\nbiologically inspired deep-learning algorithm for simultaneous localization and\nmapping (SLAM) and its application in a drone navigation system. We propose an\nunsupervised representation learning method that yields low-dimensional latent\nstate descriptors, that mitigates the sensitivity to perceptual aliasing, and\nworks on power-efficient, embedded hardware. The designed algorithm is\nevaluated on a dataset collected in an indoor warehouse environment, and\ninitial results show the feasibility for robust indoor aerial navigation.\n"}
{"abstract": "  We extend upper bounds on the quantum independence number and the quantum\nShannon capacity of graphs to their counterparts in the commuting operator\nmodel. We introduce a von Neumann algebraic generalization of the fractional\nHaemers bound (over $\\mathbb{C}$) and prove that the generalization upper\nbounds the commuting quantum independence number. We call our bound the tracial\nHaemers bound, and we prove that it is multiplicative with respect to the\nstrong product. In particular, this makes it an upper bound on the Shannon\ncapacity. The tracial Haemers bound is incomparable with the Lov\\'asz theta\nfunction, another well-known upper bound on the Shannon capacity. We show that\nseparating the tracial and fractional Haemers bounds would refute Connes'\nembedding conjecture.\n  Along the way, we prove that the tracial rank and tracial Haemers bound are\nelements of the (commuting quantum) asymptotic spectrum of graphs (Zuiddam,\nCombinatorica, 2019). We also show that the inertia bound (an upper bound on\nthe quantum independence number) upper bounds the commuting quantum\nindependence number.\n"}
{"abstract": "  High-temperature conventional superconductivity in hydrogen-rich materials\nunder high pressure has been reportedly found in twelve different compounds in\nrecent years. However, the experimental evidence on which these claims are\nbased has recently been called into question. Here we discuss the measurement\nof trapped magnetic flux, that should establish definitively that these\nmaterials are indeed high-temperature superconductors. Its absence should\nconfirm claims to the contrary.\n"}
{"abstract": "  We present a novel framework to learn to convert the perpixel photometric\ninformation at each view into spatially distinctive and view-invariant\nlow-level features, which can be plugged into existing multi-view stereo\npipeline for enhanced 3D reconstruction. Both the illumination conditions\nduring acquisition and the subsequent per-pixel feature transform can be\njointly optimized in a differentiable fashion. Our framework automatically\nadapts to and makes efficient use of the geometric information available in\ndifferent forms of input data. High-quality 3D reconstructions of a variety of\nchallenging objects are demonstrated on the data captured with an illumination\nmultiplexing device, as well as a point light. Our results compare favorably\nwith state-of-the-art techniques.\n"}
{"abstract": "  Scalable data science requires access to metadata, which is increasingly\nmanaged by databases called data catalogs. With today's data catalogs, users\nchoose between designs that make it easy to store or retrieve metadata, but not\nboth. We find this problem arises because catalogs lack an easy to understand\nmental model.\n  In this paper, we present 5W1H+R, a new catalog mental model that is\ncomprehensive in the metadata it represents, and comprehensible in that it\npermits users to locate metadata easily. We demonstrate these properties via a\nuser study. We then study different schema designs for the new mental model\nimplementation and evaluate them on different backends to understand their\nrelative merits. We conclude mental models are important to make data catalogs\nmore useful and to boost metadata management efforts.\n"}
{"abstract": "  Existing 3D human pose estimators suffer poor generalization performance to\nnew datasets, largely due to the limited diversity of 2D-3D pose pairs in the\ntraining data. To address this problem, we present PoseAug, a new\nauto-augmentation framework that learns to augment the available training poses\ntowards a greater diversity and thus improve generalization of the trained\n2D-to-3D pose estimator. Specifically, PoseAug introduces a novel pose\naugmentor that learns to adjust various geometry factors (e.g., posture, body\nsize, view point and position) of a pose through differentiable operations.\nWith such differentiable capacity, the augmentor can be jointly optimized with\nthe 3D pose estimator and take the estimation error as feedback to generate\nmore diverse and harder poses in an online manner. Moreover, PoseAug introduces\na novel part-aware Kinematic Chain Space for evaluating local joint-angle\nplausibility and develops a discriminative module accordingly to ensure the\nplausibility of the augmented poses. These elaborate designs enable PoseAug to\ngenerate more diverse yet plausible poses than existing offline augmentation\nmethods, and thus yield better generalization of the pose estimator. PoseAug is\ngeneric and easy to be applied to various 3D pose estimators. Extensive\nexperiments demonstrate that PoseAug brings clear improvements on both\nintra-scenario and cross-scenario datasets. Notably, it achieves 88.6% 3D PCK\non MPI-INF-3DHP under cross-dataset evaluation setup, improving upon the\nprevious best data augmentation based method by 9.1%. Code can be found at:\nhttps://github.com/jfzhang95/PoseAug.\n"}
{"abstract": "  The design of recurrent neural networks (RNNs) to accurately process\nsequential inputs with long-time dependencies is very challenging on account of\nthe exploding and vanishing gradient problem. To overcome this, we propose a\nnovel RNN architecture which is based on a structure preserving discretization\nof a Hamiltonian system of second-order ordinary differential equations that\nmodels networks of oscillators. The resulting RNN is fast, invertible (in\ntime), memory efficient and we derive rigorous bounds on the hidden state\ngradients to prove the mitigation of the exploding and vanishing gradient\nproblem. A suite of experiments are presented to demonstrate that the proposed\nRNN provides state of the art performance on a variety of learning tasks with\n(very) long-time dependencies.\n"}
{"abstract": "  Machine Learning (ML) is a fundamental part of modern perception systems. In\nthe last decade, the performance of computer vision using trained deep neural\nnetworks has outperformed previous approaches based on careful feature\nengineering. However, the opaqueness of large ML models is a substantial\nimpediment for critical applications such as in the automotive context. As a\nremedy, Gradient-weighted Class Activation Mapping (Grad-CAM) has been proposed\nto provide visual explanations of model internals. In this paper, we\ndemonstrate how Grad-CAM heatmaps can be used to increase the explainability of\nan image recognition model trained for a pedestrian underpass. We argue how the\nheatmaps support compliance to the EU's seven key requirements for Trustworthy\nAI. Finally, we propose adding automated heatmap analysis as a pipe segment in\nan MLOps pipeline. We believe that such a building block can be used to\nautomatically detect if a trained ML-model is activated based on invalid pixels\nin test images, suggesting biased models.\n"}
{"abstract": "  Recent work on mid-infrared (MIR) detection through the process of\nnon-degenerate two-photon absorption (NTA) in semiconducting materials has\nshown that wide-field MIR imaging can be achieved with standard Si cameras.\nWhile this approach enables MIR imaging at high pixel densities, the low\nnonlinear absorption coefficient of Si prevents fast NTA-based imaging at lower\nillumination doses. Here we overcome this limitation by using InGaAs as the\nphotosensor. Taking advantage of the much higher nonlinear absorption\ncoefficient of this direct bandgap semiconductor, we demonstrate high-speed MIR\nimaging up to 500 fps with under 1 ms exposure per frame, enabling 2D or 3D\nmapping without pre- or post-processing of the image.\n"}
{"abstract": "  Winner-take-all competitions in forecasting and machine-learning suffer from\ndistorted incentives. Witkowski et al. 2018 identified this problem and\nproposed ELF, a truthful mechanism to select a winner. We show that, from a\npool of $n$ forecasters, ELF requires $\\Theta(n\\log n)$ events or test data\npoints to select a near-optimal forecaster with high probability. We then show\nthat standard online learning algorithms select an $\\epsilon$-optimal\nforecaster using only $O(\\log(n) / \\epsilon^2)$ events, by way of a strong\napproximate-truthfulness guarantee. This bound matches the best possible even\nin the nonstrategic setting. We then apply these mechanisms to obtain the first\nno-regret guarantee for non-myopic strategic experts.\n"}
{"abstract": "  We propose a subvolume method to study the $\\theta$ dependence of the free\nenergy density of the four-dimensional SU($N$) Yang-Mills theory on the\nlattice. As an attempt, the method is first applied to SU(2) Yang-Mills theory\nat $T=1.2\\,T_c$ to understand the systematics of the method. We then proceed to\nthe calculation of the vacuum energy density and obtain the $\\theta$ dependence\nqualitatively different from the high temperature case. The numerical results\ncombined with the theoretical requirements provide the evidence for the\nspontaneous CP violation at $\\theta = \\pi$, which is in accordance with the\nlarge $N$ prediction and indicates that the similarity between 4d SU($N$) and\n2d CP$^{N-1}$ theories does not hold for $N$=2.\n"}
{"abstract": "  While work in fields of CSCW (Computer Supported Collaborative Work),\nPsychology and Social Sciences have progressed our understanding of team\nprocesses and their effect performance and effectiveness, current methods rely\non observations or self-report, with little work directed towards studying team\nprocesses with quantifiable measures based on behavioral data. In this report\nwe discuss work tackling this open problem with a focus on understanding\nindividual differences and its effect on team adaptation, and further explore\nthe effect of these factors on team performance as both an outcome and a\nprocess. We specifically discuss our contribution in terms of methods that\naugment survey data and behavioral data that allow us to gain more insight on\nteam performance as well as develop a method to evaluate adaptation and\nperformance across and within a group. To make this problem more tractable we\nchose to focus on specific types of environments, Alternate Reality Games\n(ARGs), and for several reasons. First, these types of games involve setups\nthat are similar to a real-world setup, e.g., communication through slack or\nemail. Second, they are more controllable than real environments allowing us to\nembed stimuli if needed. Lastly, they allow us to collect data needed to\nunderstand decisions and communications made through the entire duration of the\nexperience, which makes team processes more transparent than otherwise\npossible. In this report we discuss the work we did so far and demonstrate the\nefficacy of the approach.\n"}
{"abstract": "  For a set W of vertices and a vertex v in a graph G, the k-vector r2(v|W) =\n(aG(v,w1),...,aG(v,wk)) is the adjacency representation of v with respect to W,\nwhere W = {w1,...,wk} and aG(x,y) is the minimum of 2 and the distance between\nthe vertices x and y. The set W is an adjacency resolving set for G if distinct\nvertices of G have distinct adjacency representations with respect to W. The\nminimum cardinality of an adjacency resolving set for G is its adjacency\ndimension. It is clear that the adjacency dimension of an n-vertex graph G is\nbetween 1 and n-1. The graphs with adjacency dimension 1 and n-1 are known. All\ngraphs with adjacency dimension 2, and all n-vertex graphs with adjacency\ndimension n-2 are studied in this paper. In terms of the diameter and order of\nG, a sharp upper bound is found for adjacency dimension of G. Also, a sharp\nlower bound for adjacency dimension of G is obtained in terms of order of G.\nUsing these two bounds, all graphs with adjacency dimension 2, and all n-vertex\ngraphs with adjacency dimension n-2 are characterized.\n"}
{"abstract": "  The purpose of the present paper is to derive a partial differential equation\n(PDE) for the single-time single-point probability density function (PDF) of\nthe velocity field of a turbulent flow. The PDF PDE is a highly non-linear\nparabolic-transport equation, which depends on two conditional statistical\nnumerics of important physical significance. The PDF PDE is a general form of\nthe classical Reynolds mean flow equation, and is a precise formulation of the\nPDF transport equation. The PDF PDE provides us with a new method for modelling\nturbulence. An explicit example is constructed, though the example is seemingly\nartificial, but it demonstrates the PDF method based on the new PDF PDE.\n"}
{"abstract": "  Individual investors are now massively using online brokers to trade stocks\nwith convenient interfaces and low fees, albeit losing the advice and\npersonalization traditionally provided by full-service brokers. We frame the\nproblem faced by online brokers of replicating this level of service in a\nlow-cost and automated manner for a very large number of users. Because of the\ncare required in recommending financial products, we focus on a risk-management\napproach tailored to each user's portfolio and risk profile. We show that our\nhybrid approach, based on Modern Portfolio Theory and Collaborative Filtering,\nprovides a sound and effective solution. The method is applicable to stocks as\nwell as other financial assets, and can be easily combined with various\nfinancial forecasting models. We validate our proposal by comparing it with\nseveral baselines in a domain expert-based study.\n"}
{"abstract": "  Artificial neural networks are a valuable tool for radio-frequency (RF)\nsignal classification in many applications, but digitization of analog signals\nand the use of general purpose hardware non-optimized for training make the\nprocess slow and energetically costly. Recent theoretical work has proposed to\nuse nano-devices called magnetic tunnel junctions, which exhibit intrinsic RF\ndynamics, to implement in hardware the Multiply and Accumulate (MAC) operation,\na key building block of neural networks, directly using analogue RF signals. In\nthis article, we experimentally demonstrate that a magnetic tunnel junction can\nperform multiplication of RF powers, with tunable positive and negative\nsynaptic weights. Using two magnetic tunnel junctions connected in series we\ndemonstrate the MAC operation and use it for classification of RF signals.\nThese results open the path to embedded systems capable of analyzing RF signals\nwith neural networks directly after the antenna, at low power cost and high\nspeed.\n"}
{"abstract": "  Capsule networks (see e.g. Hinton et al., 2018) aim to encode knowledge and\nreason about the relationship between an object and its parts. % In this paper\nwe focus on a clean version of this problem, where data is generated from\nmultiple geometric objects (e.g. triangles, squares) at arbitrary translations,\nrotations and scales, and the observed datapoints (parts) come from the corners\nof all objects, without any labelling of the objects.\n  We specify a generative model for this data, and derive a variational\nalgorithm for inferring the transformation of each object and the assignments\nof points to parts of the objects.\n  Recent work by Kosiorek et al. [2019] has used amortized inference via\nstacked capsule autoencoders (SCA) to tackle this problem -- our results show\nthat we significantly outperform them.\n  We also investigate inference for this problem using a RANSAC-type algorithm.\n"}
{"abstract": "  Two-dimensional materials on metallic surfaces or stacked one on top of the\nother can form a variety of moir\\'e superstructures depending on the possible\nparameter and symmetry mismatch and misorientation angle. In most cases, such\nas incommensurate lattices or identical lattices but with a small twist angle,\nthe common periodicity may be very large, thus making numerical simulations\nprohibitive. We propose here a general procedure to determine the minimal\nsimulation cell which approximates, within a certain tolerance and a certain\nsize, the primitive cell of the common superlattice, given the two interfacing\nlattices and the relative orientation angle. As case studies to validate our\nprocedure, we report two applications of particular interest: the case of\nmisaligned hexagonal/hexagonal identical lattices, describing a twisted\ngraphene bilayer or a graphene monolayer grown on Ni(111), and the case of\nhexagonal/square lattices, describing for instance a graphene monolayer grown\non Ni(100) surface. The first one, which has also analytic solutions,\nconstitutes a solid benchmark for the algorithm; the second one shows that a\nvery nice description of the experimental observations can be obtained also\nusing the resulting relatively small coincidence cells.\n"}
{"abstract": "  We consider the regression problem where the dependence of the response Y on\na set of predictors X is fully captured by the regression function E(Y |\nX)=g(B'X), for an unknown function g and low rank parameter B matrix. We\ncombine neural networks with sufficient dimension reduction in order to remove\nthe limitation of small p and n of the latter. We show in simulations that the\nproposed estimator is on par with competing sufficient dimension reduction\nmethods in small p and n settings, such as minimum average variance estimation\nand conditional variance estimation. Among those, it is the only\ncomputationally applicable in large p and n problems.\n"}
{"abstract": "  We present a class of new explicit and stable numerical algorithms to solve\nthe spatially discretized linear heat or diffusion equation. After discretizing\nthe space and the time variables like conventional finite difference methods,\nwe do not approximate the time derivatives by finite differences, but use\nconstant neighbor and linear neighbour approximations to decouple the ordinary\ndifferential equations and solve them analytically. During this process, the\ntimestep-size appears not in polynomial, but in exponential form with negative\nexponents, which guarantees stability. We compare the performance of the new\nmethods with analytical and numerical solutions. According to our results, the\nmethods are first and second order in time and can be much faster than the\ncommonly used explicit or implicit methods, especially in the case of extremely\nlarge stiff systems.\n"}
{"abstract": "  Many organic chemical reactions are governed by potential energy surfaces\nthat have a region with the topographical features of a caldera. If the caldera\nhas a symmetry then trajectories transiting the caldera region are observed to\nexhibit a phenomenon that is referred to as dynamical matching. Dynamical\nmatching is a constraint that restricts the way in which a trajectory can exit\nthe caldera based solely on how it enters the caldera. In this paper we show\nthat bifurcations of the critical points of the caldera potential energy\nsurface can destroy dynamical matching even when the symmetry of the caldera is\nnot affected by the bifurcation.\n"}
{"abstract": "  The fundamental interaction between free electrons and light stands at the\nbase of both classical and quantum physics, with applications in free-electron\nacceleration, radiation sources, and electron microscopy. Yet, to this day, all\nexperiments involving free-electron light interactions are fully explained by\ndescribing the light as a classical wave, disregarding its quantum nature.\nHere, we observe quantum statistics effects of photons on free-electron-light\ninteractions. We demonstrate interactions passing continuously from Poissonian\nto super-Poissonian and up to thermal statistics, unveiling a surprising\nmanifestation of Bohr's Correspondence Principle: the transition from quantum\nwalk to classical random walk on the free-electron energy ladder. The electron\nwalker serves as the probe in non-destructive quantum detection, measuring the\nphoton-correlation ${g^{(2)} (0)}$ and higher-orders ${g^{(n)} (0)}$. Unlike\nconventional quantum-optical detectors, the electron can perform both quantum\nweak measurements and projective measurements by evolving into an entangled\njoint-state with the photons. Our findings suggest free-electron-based\nnon-destructive quantum tomography of light, and constitute an important step\ntowards combined attosecond-temporal and sub-A-spatial resolution microscopy.\n"}
{"abstract": "  The asymptotic decision theory by Le Cam and Hajek has been given a lucid\nperspective by the Ibragimov-Hasminskii theory on convergence of the likelihood\nrandom field. Their scheme has been applied to stochastic processes by\nKutoyants, and today this plot is called the IHK program. This scheme ensures\nthat asymptotic properties of an estimator follow directly from the convergence\nof the random field if a large deviation estimate exists. The quasi-likelihood\nanalysis (QLA) proved a polynomial type large deviation (PLD) inequality to go\nthrough a bottleneck of the program. A conclusion of the QLA is that if the\nquasi-likelihood random field is asymptotically quadratic and if a key index\nreflecting identifiability the random field has is non-degenerate, then the PLD\ninequality is always valid, and as a result, the IHK program can run. Many\nstudies already took advantage of the QLA theory. However, not a few of them\nare using it in an inefficient way yet. The aim of this paper is to provide a\nreformed and simplified version of the QLA and to improve accessibility to the\ntheory. As an example of the effects of the theory based on the PLD, the user\ncan obtain asymptotic properties of the quasi-Bayesian estimator by only\nverifying non-degeneracy of the key index.\n"}
{"abstract": "  Reliable and efficient trajectory generation methods are a fundamental need\nfor autonomous dynamical systems of tomorrow. The goal of this article is to\nprovide a comprehensive tutorial of three major convex optimization-based\ntrajectory generation methods: lossless convexification (LCvx), and two\nsequential convex programming algorithms known as SCvx and GuSTO. In this\narticle, trajectory generation is the computation of a dynamically feasible\nstate and control signal that satisfies a set of constraints while optimizing\nkey mission objectives. The trajectory generation problem is almost always\nnonconvex, which typically means that it is not readily amenable to efficient\nand reliable solution onboard an autonomous vehicle. The three algorithms that\nwe discuss use problem reformulation and a systematic algorithmic strategy to\nnonetheless solve nonconvex trajectory generation tasks through the use of a\nconvex optimizer. The theoretical guarantees and computational speed offered by\nconvex optimization have made the algorithms popular in both research and\nindustry circles. To date, the list of applications includes rocket landing,\nspacecraft hypersonic reentry, spacecraft rendezvous and docking, aerial motion\nplanning for fixed-wing and quadrotor vehicles, robot motion planning, and\nmore. Among these applications are high-profile rocket flights conducted by\norganizations like NASA, Masten Space Systems, SpaceX, and Blue Origin. This\narticle aims to give the reader the tools and understanding necessary to work\nwith each algorithm, and to know what each method can and cannot do. A publicly\navailable source code repository supports the provided numerical examples. By\nthe end of the article, the reader should be ready to use the methods, to\nextend them, and to contribute to their many exciting modern applications.\n"}
{"abstract": "  We propose a general and tunable platform to realize high-density arrays of\nquantum spin-valley Hall kink (QSVHK) states with spin-valley-momentum locking\nbased on a two-dimensional hexagonal topological insulator. Through the\nanalysis of Berry curvature and topological charge, the QSVHK states are found\nto be topologically protected by the valley-inversion and time-reversal\nsymmetries. Remarkably, the conductance of QSVHK states remains quantized\nagainst either nonmagnetic or long-range magnetic disorder, verified by the\nGreen function calculations. Based on first-principles results, we show that\nQSVHK states, protected with a gap up to 287 meV, can be realized in bismuthene\nby alloy engineering, surface functionalization, or electric field, supporting\nnon-volatile applications of spin-valley filters, valves, and waveguides even\nat room temperature.\n"}
{"abstract": "  We propose new variants of the sketch-and-project method for solving large\nscale ridge regression problems. Firstly, we propose a new momentum alternative\nand provide a theorem showing it can speed up the convergence of\nsketch-and-project, through a fast $\\textit{sublinear}$ convergence rate. We\ncarefully delimit under what settings this new sublinear rate is faster than\nthe previously known linear rate of convergence of sketch-and-project without\nmomentum. Secondly, we consider combining the sketch-and-project method with\nnew modern sketching methods such as the count sketch, subcount sketch (a new\nmethod we propose), and subsampled Hadamard transforms. We show experimentally\nthat when combined with the sketch-and-project method, the (sub)count sketch is\nvery effective on sparse data and the standard subsample sketch is effective on\ndense data. Indeed, we show that these sketching methods, combined with our new\nmomentum scheme, result in methods that are competitive even when compared to\nthe Conjugate Gradient method on real large scale data. On the contrary, we\nshow the subsampled Hadamard transform does not perform well in this setting,\ndespite the use of fast Hadamard transforms, and nor do recently proposed\nacceleration schemes work well in practice. To support all of our experimental\nfindings, and invite the community to validate and extend our results, with\nthis paper we are also releasing an open source software package:\n$\\texttt{RidgeSketch}$. We designed this object-oriented package in Python for\ntesting sketch-and-project methods and benchmarking ridge regression solvers.\n$\\texttt{RidgeSketch}$ is highly modular, and new sketching methods can easily\nbe added as subclasses. We provide code snippets of our package in the\nappendix.\n"}
{"abstract": "  Recent inclusive and differential cross section measurements of the\nassociated production of top quark pairs with gauge bosons or heavy-flavor jets\nare reported. A search for physics beyond the standard model in the top quark\nsector is also presented. All measurements are based on data samples of\nproton-proton collisions at $\\sqrt{s}=13$ TeV collected by the ATLAS and CMS\nexperiments at the CERN LHC. No significant deviation from the standard model\npredictions is observed.\n"}
{"abstract": "  We prove a structure theorem for the isometry group Iso(M, g) of a compact\nLorentz manifold, under the assumption that a closed subgroup has exponential\ngrowth. We don't assume anything about the identity component of Iso(M, g), so\nthat our results apply for discrete isometry groups. We infer a full\nclassification of lattices that can act isometrically on compact Lorentz\nmanifolds. Moreover, without any growth hypothesis, we prove a Tits alternative\nfor discrete subgroups of Iso(M, g).\n"}
{"abstract": "  Given $r>n$ general hyperplanes in $\\mathbb P^n,$ a star configuration of\npoints is the set of all the $n$-wise intersection of them. We introduce {\\it\ncontact star configurations}, which are star configurations where all the\nhyperplanes are osculating to the same rational normal curve. In this paper we\nfind a relation between this construction and Hadamard products of linear\nvarieties. Moreover, we study the union of contact star configurations on a\nsame conic in $\\mathbb P^2$, we prove that the union of two contact star\nconfigurations has a special $h$-vector and, in some cases, this is a complete\nintersection.\n"}
{"abstract": "  Networks have been widely used to represent the relations between objects\nsuch as academic networks and social networks, and learning embedding for\nnetworks has thus garnered plenty of research attention. Self-supervised\nnetwork representation learning aims at extracting node embedding without\nexternal supervision. Recently, maximizing the mutual information between the\nlocal node embedding and the global summary (e.g. Deep Graph Infomax, or DGI\nfor short) has shown promising results on many downstream tasks such as node\nclassification. However, there are two major limitations of DGI. Firstly, DGI\nmerely considers the extrinsic supervision signal (i.e., the mutual information\nbetween node embedding and global summary) while ignores the intrinsic signal\n(i.e., the mutual dependence between node embedding and node attributes).\nSecondly, nodes in a real-world network are usually connected by multiple edges\nwith different relations, while DGI does not fully explore the various\nrelations among nodes. To address the above-mentioned problems, we propose a\nnovel framework, called High-order Deep Multiplex Infomax (HDMI), for learning\nnode embedding on multiplex networks in a self-supervised way. To be more\nspecific, we first design a joint supervision signal containing both extrinsic\nand intrinsic mutual information by high-order mutual information, and we\npropose a High-order Deep Infomax (HDI) to optimize the proposed supervision\nsignal. Then we propose an attention based fusion module to combine node\nembedding from different layers of the multiplex network. Finally, we evaluate\nthe proposed HDMI on various downstream tasks such as unsupervised clustering\nand supervised classification. The experimental results show that HDMI achieves\nstate-of-the-art performance on these tasks.\n"}
{"abstract": "  Correlation clustering is a central topic in unsupervised learning, with many\napplications in ML and data mining. In correlation clustering, one receives as\ninput a signed graph and the goal is to partition it to minimize the number of\ndisagreements. In this work we propose a massively parallel computation (MPC)\nalgorithm for this problem that is considerably faster than prior work. In\nparticular, our algorithm uses machines with memory sublinear in the number of\nnodes in the graph and returns a constant approximation while running only for\na constant number of rounds. To the best of our knowledge, our algorithm is the\nfirst that can provably approximate a clustering problem on graphs using only a\nconstant number of MPC rounds in the sublinear memory regime. We complement our\nanalysis with an experimental analysis of our techniques.\n"}
{"abstract": "  We explore the relationships between scattering states and bound states of\ndifferent non-analytic segments (depending on $|x|$) of the exponential\npotential, and elucidate the status of the special scattering states found in\nan earlier publication by Ahmed et al. A similar analysis can be made of\nnon-analytic segments of power potentials such as $x^3$.\n"}
{"abstract": "  This paper is concerned with an inverse source problem for the stochastic\nbiharmonic operator wave equation. The driven source is assumed to be a\nmicrolocally isotropic Gaussian random field with its covariance operator being\na classical pseudo-differential operator. The well-posedness of the direct\nproblem is examined in the distribution sense and the regularity of the\nsolution is discussed for the given rough source. For the inverse problem, the\nstrength of the random source, involved in the principal symbol of its\ncovariance operator, is shown to be uniquely determined by a single realization\nof the magnitude of the wave field averaged over the frequency band with\nprobability one. Numerical experiments are presented to illustrate the validity\nand effectiveness of the proposed method for the case that the random source is\nthe white noise.\n"}
{"abstract": "  In this paper, we compute the distribution of the first letter statistic on\nnine avoidance classes of permutations corresponding to two pairs of patterns\nof length four. In particular, we show that the distribution is the same for\neach class and is given by the entries of a new Schr\\\"oder number triangle.\nThis answers in the affirmative a recent conjecture of Lin and Kim. We employ a\nvariety of techniques to prove our results, including generating trees, direct\nbijections and the kernel method. For the latter, we make use of in a creative\nway what we are trying to show in three cases to aid in solving a system of\nfunctional equations satisfied by the associated generating functions.\n"}
{"abstract": "  The big rip, the little rip and the little sibling of the big rip are\ncosmological doomsdays predicted by some phantom dark energy models that could\ndescribe the future evolution of our Universe. When the universe evolves\ntowards either of these future cosmic events all bounded structures and,\nultimately, space-time itself are ripped apart. Nevertheless, it is commonly\nbelief that quantum gravity effects may smooth or even avoid these classically\npredicted singularities. In this review, we discuss the classical and quantum\noccurrence of these riplike events in the scheme of metric $f(R)$ theories of\ngravity. The quantum analysis is performed in the framework of $f(R)$ quantum\ngeometrodynamics. In this context, we analyse the fulfilment of the DeWitt\ncriterion for the avoidance of these singular fates.\n"}
{"abstract": "  Using a large sample of sub-L$_*$ galaxies, with similar UV magnitudes,\nM$_{\\rm UV}\\simeq -19$ at $z\\simeq6$, extracted from the FirstLight\nsimulations, we show the diversity of galaxies at the end of the reionization\nepoch. We find a factor $\\sim$40 variation in the specific star-formation rate\n(sSFR). This drives a $\\sim$1 dex range in equivalent width of the\n[OIII]$\\lambda$5007 line. Variations in nebular metallicity and ionization\nparameter within HII regions lead to a scatter in the equivalent widths and\n[OIII]/H$\\alpha$ line ratio at a fixed sSFR. [OIII]-bright emitters\n([OIII]/H$\\alpha$>1) have higher ionization parameters and/or higher\nmetallicities than H$\\alpha$-bright ([OIII]/H$\\alpha$<1) galaxies. According to\nthe surface brightness maps in both [OIII] and H$\\alpha$, [OIII]-bright\nemitters are more compact than H$\\alpha$-bright galaxies. H$\\alpha$ luminosity\nis higher than [OIII] if star formation is distributed over extended regions.\nOIII dominates if it is concentrated in compact clumps. In both cases, the\nH$\\alpha$-emitting gas is significantly more extended than [OIII].\n"}
{"abstract": "  Within the intersection of Ehrhart theory, commutative algebra, and algebraic\ngeometry lie lattice polytopes. Ehrhart theory is concerned with lattice point\nenumeration in dilates of polytopes; lattice polytopes provide a sandbox in\nwhich to test many conjectures in commutative algebra; and many properties of\nprojectively normal toric varieties in algebraic geometry are encoded through\ncorresponding lattice polytopes. In this article we focus on reflexive\nsimplices and work to identify when these have the integer decomposition\nproperty (IDP), or equivalently, when certain weighted projective spaces are\nprojectively normal. We characterize the reflexive, IDP simplices whose\nassociated weighted projective spaces have one projective coordinate with\nweight fixed to unity and for which the remaining coordinates can assume one of\nthree distinct weights. We show that several subfamilies of such reflexive\nsimplices have unimodal $h^\\ast$-polynomials, thereby making progress towards\nconjectures and questions of Stanley, Hibi-Ohsugi, and others regarding the\nunimodality of their $h^\\ast$-polynomials. We also provide computational\nresults and introduce the notion of reflexive stabilizations to explore the\n(non-)ubiquity of reflexive simplices that are simultaneously IDP and\n$h^\\ast$-unimodal.\n"}
{"abstract": "  The research into insulating ferrimagnetic garnets has gained enormous\nmomentum in the past decade. This is partly due to the improvement in the\ntechniques to grow high-quality ultrathin films with desirable properties and\nthe advances in understanding the spin transport within the ferrimagnetic\ngarnets and through their interfaces with conducting materials. In recent\nyears, we have seen remarkable progress in controlling the magnetization state\nof ferrimagnetic garnets by electrical means in suitable heterostructures and\ndevice architectures. These advances have readily placed ferrimagnetic garnets\nin a favorable position for the future development of insulating spintronic\nconcepts. The purpose of this article is to review recent experimental results\nof the current-induced magnetization control and associated phenomena in\nferrimagnetic garnets, as well as to discuss future directions in this rapidly\nevolving area of spintronics.\n"}
{"abstract": "  Interference alignment (IA) is a promising scheme to increase the throughput\nof the system with shared mediums. In this paper, we propose a novel IA scheme\nin molecular interference channels (IFCs) based on the choice of\nreleasing/sampling times. To cancel the aligned interference signals and reduce\nthe signal dependent noise, we use molecular reaction in the proposed IA\nscheme. We obtain the feasible region for the releasing/sampling times in the\nIA scheme using reaction. Further, we investigate the error performance of the\nproposed scheme. Our results show that the proposed IA scheme using reaction\nimproves the performance significantly.\n"}
{"abstract": "  The $t$-$J$-$U$ model of high-$T_c$ copper-oxide superconductors incorporates\nboth the on-site Coulomb repulsion and kinetic exchange interaction and yields\na semi-quantitative description of the static properties of those materials. We\nextend this analysis to dynamic quantities and address collective spin- and\ncharge excitations in the correlated metallic state of the $t$-$J$-$U$ model.\nWe employ VWF+$1/\\mathcal{N}_f$ approach that combines the variational wave\nfunction (VWF) approach with the expansion in the inverse number of fermionic\nflavors ($1/\\mathcal{N}_f$). It is shown that the resonant (paramagnon)\ncontribution to the dynamic magnetic susceptibility remains robust as one\ninterpolates between the Hubbard- and $t$-$J$-model limits, whereas the\nincoherent continuum undergoes substantial renormalization. Energy of the\ncollective charge mode diminishes as the strong-coupling limit is approached.\nWe also introduce the concept of effective kinetic exchange interaction that\nallows for a unified interpretation of magnetic dynamics in the Hubbard,\n$t$-$J$, and $t$-$J$-$U$ models. The results are discussed in the context of\nrecent resonant inelastic $x$-ray scattering experiments for the high-$T_c$\ncuprates.\n"}
{"abstract": "  We present a catalog of 536 fast radio bursts (FRBs) detected by the Canadian\nHydrogen Intensity Mapping Experiment Fast Radio Burst (CHIME/FRB) Project\nbetween 400 and 800 MHz from 2018 July 25 to 2019 July 1, including 62 bursts\nfrom 18 previously reported repeating sources. The catalog represents the first\nlarge sample, including bursts from repeaters and non-repeaters, observed in a\nsingle survey with uniform selection effects. This facilitates comparative and\nabsolute studies of the FRB population. We show that repeaters and apparent\nnon-repeaters have sky locations and dispersion measures (DMs) that are\nconsistent with being drawn from the same distribution. However, bursts from\nrepeating sources differ from apparent non-repeaters in intrinsic temporal\nwidth and spectral bandwidth. Through injection of simulated events into our\ndetection pipeline, we perform an absolute calibration of selection effects to\naccount for systematic biases. We find evidence for a population of FRBs -\ncomprising a large fraction of the overall population - with a scattering time\nat 600 MHz in excess of 10 ms, of which only a small fraction are observed by\nCHIME/FRB. We infer a power-law index for the cumulative fluence distribution\nof $\\alpha=-1.40\\pm0.11(\\textrm{stat.})^{+0.06}_{-0.09}(\\textrm{sys.})$,\nconsistent with the $-3/2$ expectation for a non-evolving population in\nEuclidean space. We find $\\alpha$ is steeper for high-DM events and shallower\nfor low-DM events, which is what would be expected when DM is correlated with\ndistance. We infer a sky rate of\n$[820\\pm60(\\textrm{stat.})^{+220}_{-200}({\\textrm{sys.}})]/\\textrm{sky}/\\textrm{day}$\nabove a fluence of 5 Jy ms at 600 MHz, with scattering time at $600$ MHz under\n10 ms, and DM above 100 pc cm$^{-3}$.\n"}
{"abstract": "  We consider the problem of embedding a dynamic network, to obtain\ntime-evolving vector representations of each node, which can then be used to\ndescribe the changes in behaviour of a single node, one or more communities, or\nthe entire graph. Given this open-ended remit, we wish to guarantee stability\nin the spatio-temporal positioning of the nodes: assigning the same position,\nup to noise, to nodes behaving similarly at a given time (cross-sectional\nstability) and a constant position, up to noise, to a single node behaving\nsimilarly across different times (longitudinal stability). These properties are\ndefined formally within a generic dynamic latent position model. By showing how\nthis model can be recast as a multilayer random dot product graph, we\ndemonstrate that unfolded adjacency spectral embedding satisfies both stability\nconditions, allowing, for example, spatio-temporal clustering under the dynamic\nstochastic block model. We also show how alternative methods, such as omnibus,\nindependent or time-averaged spectral embedding, lack one or the other form of\nstability.\n"}
{"abstract": "  We prove that the grades of simple modules indexed by boolean permutations,\nover the incidence algebra of the symmetric group with respect to the Bruhat\norder, are given by Lusztig's a-function. Our arguments are combinatorial, and\ninclude a description of the intersection of two principal order ideals when at\nleast one permutation is boolean. An important object in our work is a reduced\nword written as minimally many runs of consecutive integers, and one step of\nour argument shows that this minimal quantity is equal to the length of the\nsecond row in the permutation's shape under the Robinson-Schensted\ncorrespondence. We also prove that a simple module over the above-mentioned\nincidence algebra is perfect if and only if its index is the longest element of\na parabolic subgroup.\n"}
{"abstract": "  Motivated by recent Parker Solar Probe (PSP) observations of \"switchbacks\"\n(abrupt, large-amplitude reversals in the radial magnetic field, which exhibit\nAlfv\\'enic correlations) we examine the dynamics of large-amplitude Alfv\\'en\nwaves in the expanding solar wind. We develop an analytic model which makes\nseveral predictions: switchbacks should preferentially occur in regions where\nthe solar wind plasma has undergone a greater expansion, the switchback\nfraction at radii comparable to PSP should be an increasing function of radius,\nand switchbacks should have their gradients preferentially perpendicular to the\nmean magnetic field direction. The expansion of the plasma generates small\ncompressive components as part of the wave's nonlinear evolution: these are\nmaximized when the normalized fluctuation amplitude is comparable to\n$\\sin\\theta$, where $\\theta$ is the angle between the propagation direction and\nthe mean magnetic field. These compressive components steepen the primary\nAlfv\\'enic waveform, keeping the solution in a state of nearly constant\nmagnetic field strength as its normalized amplitude $\\delta B/B$ grows due to\nexpansion. The small fluctuations in the magnetic-field-strength are minimized\nat a particular $\\theta$-dependent value of $\\beta$, usually of order unity,\nand the density and magnetic-field-strength fluctuations can be correlated or\nanticorrelated depending on $\\beta$ and $\\theta$. Example solutions of our\ndynamical equation are presented; some do indeed form magnetic-field reversals.\nOur predictions appear to match some previously unexplained phenomena in\nobservations and numerical simulations, providing evidence that the observed\nswitchbacks result from the nonlinear evolution of the initially\nsmall-amplitude Alfv\\'en waves already known to be present at the coronal base.\n"}
{"abstract": "  Many ionic surfactants with wide applications in personal-care and house-hold\ndetergency show limited water solubility at lower temperatures (Krafft point).\nThis drawback can be overcome by using mixed solutions, where the ionic\nsurfactant is incorporated in mixed micelles with another surfactant, which is\nsoluble at lower temperatures. The solubility and electrolytic conductivity for\na binary surfactant mixture of anionic methyl ester sulfonates (MES) with\nnonionic alkyl polyglucoside and alkyl polyoxyethylene ether at 5 deg C during\nlong-term storage were measured. Phase diagrams were established; a general\ntheoretical model for their explanation was developed and checked\nexperimentally. The binary and ternary phase diagrams for studied surfactant\nmixtures include phase domains: mixed micelles; micelles and crystallites;\ncrystallites, and molecular solution. The proposed general methodology, which\nutilizes the equations of molecular thermodynamics at minimum number of\nexperimental measurements, is convenient for construction of such phase\ndiagrams. The results could increase the range of applicability of\nMES-surfactants with relatively high Krafft temperature, but with various\nuseful properties such as excellent biodegradability and skin compatibility;\nstability in hard water; good wetting and cleaning performance.\n"}
{"abstract": "  The long-term variations of the orbit of the Earth govern the insolation on\nits surface and hence its climate. The use of the astronomical signal, whose\nimprint has been recovered in the geological records, has revolutionized the\ndetermination of the geological time scales (e.g. Gradstein & Ogg 2020).\nHowever, the orbital variations beyond 60 Myr cannot be reliably predicted\nbecause of the chaotic dynamics of the planetary orbits in the Solar System\n(Laskar 1989). Taking into account this dynamical uncertainty is necessary for\na complete astronomical calibration of geological records. Our work addresses\nthis problem with a statistical analysis of 120 000 orbital solutions of the\nsecular model of the Solar System ranging from 500 Myr to 5 Gyr. We obtain the\nmarginal probability density functions of the fundamental secular frequencies\nusing kernel density estimation. The uncertainty of the density estimation is\nalso obtained here in the form of confidence intervals determined by the moving\nblock bootstrap method. The results of the secular model are shown to be in\ngood agreement with those of the direct integrations of a comprehensive model\nof the Solar System. Application of our work is illustrated on two geological\ndata: the Newark-Hartford records and the Libsack core.\n"}
{"abstract": "  It has become increasingly common to collect high-dimensional binary data;\nfor example, with the emergence of new sampling techniques in ecology. In\nsmaller dimensions, multivariate probit (MVP) models are routinely used for\ninferences. However, algorithms for fitting such models face issues in scaling\nup to high dimensions due to the intractability of the likelihood, involving an\nintegral over a multivariate normal distribution having no analytic form.\nAlthough a variety of algorithms have been proposed to approximate this\nintractable integral, these approaches are difficult to implement and/or\ninaccurate in high dimensions. We propose a two-stage Bayesian approach for\ninference on model parameters while taking care of the uncertainty propagation\nbetween the stages. We use the special structure of latent Gaussian models to\nreduce the highly expensive computation involved in joint parameter estimation\nto focus inference on marginal distributions of model parameters. This\nessentially makes the method embarrassingly parallel for both stages. We\nillustrate performance in simulations and applications to joint species\ndistribution modeling in ecology.\n"}
{"abstract": "  This paper presents an automatic finite element simulation scheme accounting\nfor high geometric nonlinearity and the difference between linear and nonlinear\nbuckling of composite thin-walled lenticular tubes (CTLTs). Parameterizing of\ncross-section shapes and generation of design space for CTLTs with both\ncircular and parabolic arcs were accomplished, and several key factors were\nidentified, in particular the contrary effect of lumbus length and parabolic\ncoefficient on the bending stiffness anisotropy. The first quantitative\ncomparison of triangular rollable and collapsible (TRAC) booms and CTLTs is\ngiven in terms of bending performance in two directions, showing that the\noptimal CTLT carefully selected from the design space demonstrates a comparable\nor even better performance than the TRAC boom. This is of great importance from\nboth academic and engineering perspectives. Our efforts enhance the\nunderstanding of nonlinear buckling and post-buckling behavior of CTLTs, and\nprovide guidelines for future design of CTLTs with desirable performance.\n"}
{"abstract": "  We demonstrate high fidelity enhancement of planetary digital elevation\nmodels (DEMs) using optical images and deep learning with convolutional neural\nnetworks. Enhancement can be applied recursively to the limit of available\noptical data, representing a 90x resolution improvement in global Mars DEMs.\nDeep learning-based photoclinometry robustly recovers features obscured by\nnon-ideal lighting conditions. Method can be automated at global scale.\nAnalysis shows enhanced DEM slope errors are comparable with high resolution\nmaps using conventional, labor intensive methods.\n"}
{"abstract": "  We present mGENRE, a sequence-to-sequence system for the Multilingual Entity\nLinking (MEL) problem -- the task of resolving language-specific mentions to a\nmultilingual Knowledge Base (KB). For a mention in a given language, mGENRE\npredicts the name of the target entity left-to-right, token-by-token in an\nautoregressive fashion. The autoregressive formulation allows us to effectively\ncross-encode mention string and entity names to capture more interactions than\nthe standard dot product between mention and entity vectors. It also enables\nfast search within a large KB even for mentions that do not appear in mention\ntables and with no need for large-scale vector indices. While prior MEL works\nuse a single representation for each entity, we match against entity names of\nas many languages as possible, which allows exploiting language connections\nbetween source input and target name. Moreover, in a zero-shot setting on\nlanguages with no training data at all, mGENRE treats the target language as a\nlatent variable that is marginalized at prediction time. This leads to over 50%\nimprovements in average accuracy. We show the efficacy of our approach through\nextensive evaluation including experiments on three popular MEL benchmarks\nwhere mGENRE establishes new state-of-the-art results. Code and pre-trained\nmodels at https://github.com/facebookresearch/GENRE.\n"}
{"abstract": "  We propose Anticipative Video Transformer (AVT), an end-to-end\nattention-based video modeling architecture that attends to the previously\nobserved video in order to anticipate future actions. We train the model\njointly to predict the next action in a video sequence, while also learning\nframe feature encoders that are predictive of successive future frames'\nfeatures. Compared to existing temporal aggregation strategies, AVT has the\nadvantage of both maintaining the sequential progression of observed actions\nwhile still capturing long-range dependencies--both critical for the\nanticipation task. Through extensive experiments, we show that AVT obtains the\nbest reported performance on four popular action anticipation benchmarks:\nEpicKitchens-55, EpicKitchens-100, EGTEA Gaze+, and 50-Salads; and it wins\nfirst place in the EpicKitchens-100 CVPR'21 challenge.\n"}
{"abstract": "  Generalization is a central problem in machine learning, especially when data\nis limited. Using prior information to enforce constraints is the principled\nway of encouraging generalization. In this work, we propose to leverage the\nprior information embedded in pretrained language models (LM) to improve\ngeneralization for intent classification and slot labeling tasks with limited\ntraining data. Specifically, we extract prior knowledge from pretrained LM in\nthe form of synthetic data, which encode the prior implicitly. We fine-tune the\nLM to generate an augmented language, which contains not only text but also\nencodes both intent labels and slot labels. The generated synthetic data can be\nused to train a classifier later. Since the generated data may contain noise,\nwe rephrase the learning from generated data as learning with noisy labels. We\nthen utilize the mixout regularization for the classifier and prove its\neffectiveness to resist label noise in generated data. Empirically, our method\ndemonstrates superior performance and outperforms the baseline by a large\nmargin.\n"}
{"abstract": "  Second-harmonic generation (SHG) is a direct measure of the strength of\nsecond-order nonlinear optical effects, which also include frequency mixing and\nparametric oscillations. Natural and artificial materials with broken\ncenter-of-inversion symmetry in their unit cell display high SHG efficiency,\nhowever the silicon-foundry compatible group-IV semiconductors (Si, Ge) are\ncentrosymmetric, thereby preventing full integration of second-order\nnonlinearity in silicon photonics platforms. Here we demonstrate strong SHG in\nGe-rich quantum wells grown on Si wafers. The symmetry breaking is artificially\nrealized with a pair of asymmetric coupled quantum wells (ACQW), in which three\nof the quantum-confined states are equidistant in energy, resulting in a double\nresonance for SHG. Laser spectroscopy experiments demonstrate a giant\nsecond-order nonlinearity at mid-infrared pump wavelengths between 9 and 12\nmicrons. Leveraging on the strong intersubband dipoles, the nonlinear\nsusceptibility almost reaches 10^5 pm/V\n"}
{"abstract": "  Meta learning with multiple objectives can be formulated as a Multi-Objective\nBi-Level optimization Problem (MOBLP) where the upper-level subproblem is to\nsolve several possible conflicting targets for the meta learner. However,\nexisting studies either apply an inefficient evolutionary algorithm or linearly\ncombine multiple objectives as a single-objective problem with the need to tune\ncombination weights. In this paper, we propose a unified gradient-based\nMulti-Objective Meta Learning (MOML) framework and devise the first\ngradient-based optimization algorithm to solve the MOBLP by alternatively\nsolving the lower-level and upper-level subproblems via the gradient descent\nmethod and the gradient-based multi-objective optimization method,\nrespectively. Theoretically, we prove the convergence properties of the\nproposed gradient-based optimization algorithm. Empirically, we show the\neffectiveness of the proposed MOML framework in several meta learning problems,\nincluding few-shot learning, neural architecture search, domain adaptation, and\nmulti-task learning.\n"}
{"abstract": "  Terahertz frequency bands will likely be used for the next-generation\nwireless communication systems to provide data rates of hundreds of Gbps or\neven Tbps because of the wide swaths of unused and unexplored spectrum. This\npaper presents two outdoor wideband measurement campaigns in downtown Brooklyn\n(urban microcell environment) in the sub-THz band of 140 GHz with TX-RX\nseparation distance up to 100 m: i) terrestrial urban microcell measurement\ncampaign, and ii) rooftop surrogate satellite and backhaul measurement\ncampaign. Outdoor omnidirectional and directional path loss models for both\nline-of-sight and non-line-of-sight scenarios, as well as foliage loss (signal\nattenuation through foliage), are provided at 140 GHz for urban microcell\nenvironments. These measurements and models provide an understanding of both\nthe outdoor terrestrial (e.g., 6G cellular and backhaul) and non-terrestrial\n(e.g., satellite and unmanned aerial vehicle communications) wireless channels,\nand prove the feasibility of using THz frequency bands for outdoor fixed and\nmobile cellular communications. This paper can be used for future outdoor\nwireless system design at frequencies above 100 GHz.\n"}
{"abstract": "  This paper examines a class of barrier options-multi-step barrier options,\nwhich can have any finite number of barriers of any level. We obtain a general,\nexplicit expression of option prices of this type under the Black-Scholes\nmodel. Multi-step barrier options are not only useful in that they can handle\nbarriers of different levels and time steps, but can also approximate options\nwith arbitrary barriers. Moreover, they can be embedded in financial products\nsuch as deposit insurances based on jump models with simple barriers. Along the\nway, we derive multi-step reflection principle, which generalizes the\nreflection principle of Brownian motion.\n"}
{"abstract": "  We analyze the effective field equations of the Randall-Sundrum braneworld\ncoupled with a Klein-Gordon scalar field through the minimal geometric\ndeformation decoupling method (MGD-decoupling). We introduce two different ways\nto apply the MGD-decoupling method to obtain new solutions for this enlarged\nsystem. We also compare the behavior of the new solutions with those obtained\ndirectly from the Randall-Sumdrum braneworld without coupling to the scalar\nfield.\n"}
{"abstract": "  The next technological breakthrough in millimeter-submillimeter astronomy is\n3D imaging spectrometry with wide instantaneous spectral bandwidths and wide\nfields of view. The total optimization of the focal-plane instrument, the\ntelescope, the observing strategy, and the signal-processing software must\nenable efficient removal of foreground emission from the Earth's atmosphere,\nwhich is time-dependent and highly nonlinear in frequency. Here we present\nTiEMPO: Time-Dependent End-to-End Model for Post-process Optimization of the\nDESHIMA Spectrometer. TiEMPO utilizes a dynamical model of the atmosphere and\nparametrized models of the astronomical source, the telescope, the instrument,\nand the detector. The output of TiEMPO is a time-stream of sky brightness\ntemperature and detected power, which can be analyzed by standard\nsignal-processing software. We first compare TiEMPO simulations with an on-sky\nmeasurement by the wideband DESHIMA spectrometer and find good agreement in the\nnoise power spectral density and sensitivity. We then use TiEMPO to simulate\nthe detection of a line emission spectrum of a high-redshift galaxy using the\nDESHIMA 2.0 spectrometer in development. The TiEMPO model is open source. Its\nmodular and parametrized design enables users to adapt it to design and\noptimize the end-to-end performance of spectroscopic and photometric\ninstruments on existing and future telescopes.\n"}
{"abstract": "  Magnetic droplets are dissipative magnetodynamical solitons that can form\nunder current driven nanocontacts in magnetic layers with large perpendicular\nmagnetic anisotropy. Here, we extend the original droplet theory by studying\nthe impact of the dipolar interactions on the dynamics of droplet solitons. By\nvarying the thickness of the free layer of a spin torque nano-oscillator, we\nsystematically tune the internal field of the free layer to investigate the\ndynamics of droplet solitons. Our numerical results show that increasing the\nfree layer thickness increases the droplet threshold current, decreases the\ndroplet frequency and diameter, enlarges the current hysteresis and also\nmodifies the structure of the droplet. The Oersted field of the current breaks\nthe phase coherency and deteriorates the stability of the droplet in free\nlayers with larger thicknesses. Moreover, our findings show a simple relation\nto determine the impact of the free layer thickness on the droplet nucleation\nboundaries. Our study presents the missing brick on the physics behind magnetic\ndroplet solitons, and further illustrates that magnetic droplets in thinner\nlayers possess more promising characteristics for spintronic applications and\nenable devices with higher speed of operation.\n"}
{"abstract": "  In this paper, we propose MINE to perform novel view synthesis and depth\nestimation via dense 3D reconstruction from a single image. Our approach is a\ncontinuous depth generalization of the Multiplane Images (MPI) by introducing\nthe NEural radiance fields (NeRF). Given a single image as input, MINE predicts\na 4-channel image (RGB and volume density) at arbitrary depth values to jointly\nreconstruct the camera frustum and fill in occluded contents. The reconstructed\nand inpainted frustum can then be easily rendered into novel RGB or depth views\nusing differentiable rendering. Extensive experiments on RealEstate10K, KITTI\nand Flowers Light Fields show that our MINE outperforms state-of-the-art by a\nlarge margin in novel view synthesis. We also achieve competitive results in\ndepth estimation on iBims-1 and NYU-v2 without annotated depth supervision. Our\nsource code is available at https://github.com/vincentfung13/MINE\n"}
{"abstract": "  We compute the tautological ring for a Hilbert modular variety at an\nunramified prime. The method generalises that of van der Geer from the Siegel\ncase.\n"}
{"abstract": "  In this document, some general results in approximation theory and matrix\nanalysis with applications to sparse identification of time series models and\nnonlinear discrete-time dynamical systems are presented. The aforementioned\ntheoretical methods are translated into algorithms that can be used for sparse\nmodel identification of discrete-time dynamical systems, based on structured\ndata measured from the systems. The approximation of the state-transition\noperators that are determined primarily by matrices of parameters to be\nidentified based on data measured from a given system, is approached by\nidentifying conditions for the existence of low-rank approximations of\nsubmatrices of the trajectory matrices corresponding to the measured data, that\ncan be used to compute approximate sparse representations of the matrices of\nparameters. One of the main advantages of the low-rank approximation approach\npresented in this document, concerns the parameter estimation for linear and\nnonlinear models where numerical or measurement noise could affect the\nestimates significantly. Prototypical algorithms based on the aforementioned\ntechniques together with some applications to approximate identification and\npredictive simulation of time series models with symmetries and nonlinear\nstructured dynamical systems in theoretical physics, fluid dynamics and weather\nforecasting are presented.\n"}
{"abstract": "  We construct the first ever multimodal sarcasm dataset for Spanish. The\naudiovisual dataset consists of sarcasm annotated text that is aligned with\nvideo and audio. The dataset represents two varieties of Spanish, a Latin\nAmerican variety and a Peninsular Spanish variety, which ensures a wider\ndialectal coverage for this global language. We present several models for\nsarcasm detection that will serve as baselines in the future research. Our\nresults show that results with text only (89%) are worse than when combining\ntext with audio (91.9%). Finally, the best results are obtained when combining\nall the modalities: text, audio and video (93.1%).\n"}
{"abstract": "  A common assumption in the social learning literature is that agents exchange\ninformation in an unselfish manner. In this work, we consider the scenario\nwhere a subset of agents aims at deceiving the network, meaning they aim at\ndriving the network beliefs to the wrong hypothesis. The adversaries are\nunaware of the true hypothesis. However, they will \"blend in\" by behaving\nsimilarly to the other agents and will manipulate the likelihood functions used\nin the belief update process to launch inferential attacks. We will\ncharacterize the conditions under which the network is misled. Then, we will\nexplain that it is possible for such attacks to succeed by showing that\nstrategies exist that can be adopted by the malicious agents for this purpose.\nWe examine both situations in which the agents have access to information about\nthe network model as well as the case in which they do not. For the first case,\nwe show that there always exists a way to construct fake likelihood functions\nsuch that the network is deceived regardless of the true hypothesis. For the\nlatter case, we formulate an optimization problem and investigate the\nperformance of the derived attack strategy by establishing conditions under\nwhich the network is deceived. We illustrate the learning performance of the\nnetwork in the aforementioned adversarial setting via simulations. In a\nnutshell, we clarify when and how a network is deceived in the context of\nnon-Bayesian social learning.\n"}
{"abstract": "  We study the polyhedral convex hull structure of a mixed-integer set which\narises in a class of cardinality-constrained concave submodular minimization\nproblems. This class of problems has an objective function in the form of\n$f(a^\\top x)$, where $f$ is a univariate concave function, $a$ is a\nnon-negative vector, and $x$ is a binary vector of appropriate dimension. Such\nminimization problems frequently appear in applications that involve\nrisk-aversion or economies of scale. We propose three classes of strong valid\nlinear inequalities for this convex hull and specify their facet conditions\nwhen $a$ has two distinct values. We show how to use these inequalities to\nobtain valid inequalities for general $a$ that contains multiple values. We\nfurther provide a complete linear convex hull description for this\nmixed-integer set when $a$ contains two distinct values and the cardinality\nconstraint upper bound is two. Our computational experiments on the mean-risk\noptimization problem demonstrate the effectiveness of the proposed inequalities\nin a branch-and-cut framework.\n"}
{"abstract": "  In order to run Computational Fluid Dynamics (CFD) codes on large scale\ninfrastructures, parallel computing has to be used because of the computational\nintensive nature of the problems. In this paper we investigate the ADAPT\nplatform where we couple flow Partial Differential Equations and a Poisson\nequation. This leads to a linear system which we solve using direct methods.\nThe implementation deals with the MUMPS parallel multi-frontal direct solver\nand mesh partitioning methods using METIS to improve the performance of the\nframework. We also investigate, in this paper, how the mesh partitioning\nmethods are able to optimize the mesh cell distribution for the ADAPT solver.\nThe experience gained in this paper facilitates the move to the 3D version of\nADAPT and the move to a Service Oriented view of ADAPT as future work.\n"}
{"abstract": "  We show that if we suppose n>3 and the (2n-1)-stem in the stable homotopy\ngroups of spheres has no 2-torsion, then the Whitehead squares of the identity\nmaps of (2n+1) and (4n+3)-spheres are divisible by 2. By applying the result of\nG. Wang and Z. Xu on the 61-stem in the stable homotopy groups of spheres, we\nfind that the Kervaire invariant one elements in dimensions 62 and 126 exist.\n"}
{"abstract": "  When documenting oral-languages, Unsupervised Word Segmentation (UWS) from\nspeech is a useful, yet challenging, task. It can be performed from phonetic\ntranscriptions, or in the absence of these, from the output of unsupervised\nspeech discretization models. These discretization models are trained using raw\nspeech only, producing discrete speech units which can be applied for\ndownstream (text-based) tasks. In this paper we compare five of these models:\nthree Bayesian and two neural approaches, with regards to the exploitability of\nthe produced units for UWS. Two UWS models are experimented with and we report\nresults for Finnish, Hungarian, Mboshi, Romanian and Russian in a low-resource\nsetting (using only 5k sentences). Our results suggest that neural models for\nspeech discretization are difficult to exploit in our setting, and that it\nmight be necessary to adapt them to limit sequence length. We obtain our best\nUWS results by using the SHMM and H-SHMM Bayesian models, which produce high\nquality, yet compressed, discrete representations of the input speech signal.\n"}
{"abstract": "  In odd dimensions the integrated conformal anomaly is entirely due to the\nboundary terms \\cite{Solodukhin:2015eca}. In this paper we present a detailed\nanalysis of the anomaly in five dimensions. We give the complete list of the\nboundary conformal invariants that exist in five dimensions. Additionally to 8\ninvariants known before we find a new conformal invariant that contains the\nderivatives of the extrinsic curvature along the boundary. Then, for a\nconformal scalar field satisfying either the Dirichlet or the conformal\ninvariant Robin boundary conditions we use the available general results for\nthe heat kernel coefficient $a_5$, compute the conformal anomaly and identify\nthe corresponding values of all boundary conformal charges.\n"}
{"abstract": "  Extensions to the Standard Model that use strictly off-shell degrees of\nfreedom - the fakeons - allow for new measurable interactions at energy scales\nusually precluded by the constraints that target the on-shell propagation of\nnew particles. Here we employ the interactions between a new fake scalar\ndoublet and the muon to explain the recent Fermilab measurement of its\nanomalous magnetic moment. Remarkably, unlike in the case of usual particles,\nthe experimental result can be matched for fakeon masses below the electroweak\nscale without contradicting the stringent precision data and collider bounds on\nnew light degrees of freedom. Our analysis, therefore, demonstrates that the\nfakeon approach offers unexpected viable possibilities to model new physics\nnaturally at low scales.\n"}
{"abstract": "  Microresonator-based platforms with $\\chi^{(2)}$ nonlinearities have the\npotential to perform frequency conversion at high efficiencies and ultralow\npowers with small footprints. The standard doctrine for achieving high\nconversion efficiency in cavity-based devices requires \"perfect matching\", that\nis, zero phase mismatch while all relevant frequencies are precisely at a\ncavity resonance, which is difficult to achieve in integrated platforms due to\nfabrication errors and limited tunabilities. In this Letter, we show that the\nviolation of perfect matching does not necessitate a reduction in conversion\nefficiency. On the contrary, in many cases, mismatches should be intentionally\nintroduced to improve the efficiency or tunability of conversion. We identify\nthe universal conditions for maximizing the efficiency of cavity-based\nfrequency conversion and show a straightforward approach to fully compensate\nfor parasitic processes such as thermorefractive and photorefractive effects\nthat, typically, can limit the conversion efficiency. We also show rigorously\nthat these high-efficiency states are stable.\n"}
{"abstract": "  Neutrino oscillations constitute the first experimental confirmation of new\nphysics, as they require a new piece that is absent in the Standard Model,\nnon-zero neutrino masses. Questions such as how these masses are generated or\nwhy they are so small compared to other masses, are still under investigation.\nIn this thesis, we study radiative neutrino mass models, a class of models\nwhere small neutrino masses are generated naturally, while the energy scale at\nwhich new physics appears is still low enough to be tested by current and\nfuture experiments, offering a falsifiable window to test beyond the Standard\nModel physics. In these models, neutrino masses are forbidden at the\ntree-level, but allowed at a certain order in loops. Consequently, we get\nmasses suppressed by loop factors without requiring a large new physics scale\nand/or small couplings. In particular, we study the systematic classifications\nof the $d=7$ Majorana neutrino mass operator at one-loop order, the Weinberg\noperator at three-loop order and the $d=4$ Dirac neutrino mass models at the\ntwo-loop level. We then move to more specific models and mechanisms, analysing\nthe connection between the generation of radiative neutrino masses and the\nstability of dark matter from a model-independent point of view, and examine a\nparticular realisation of the type-I seesaw which relies on the radiative\ngeneration of Dirac Yukawa coupling. Finally, we analyse the phenomenology of\nsome of the models discussed in the classifications, as well as a very\nparticular case of neutrinoless double-$\\beta$ decay where electrons are\nemitted with opposite chiralities along with a light scalar (Majoron).\n"}
{"abstract": "  The multi-color passband CCD light curves of ASAS J124343+1531.7 and LINEAR\n2323566 were first obtained by the 0.84-m Ritchey-Chr\\'{e}tien telescope with\nfollow up observations by the WIYN 0.90m Cassegrain telescope. The data from\nthe $Transiting\\quad Exoplanet\\quad Survey\\quad Satellite\\quad (TESS)$ of ASAS\nJ124343+1531.7 was also applied for subsequent analysis. By analyzing the data\nthrough the W-D program, their mass ratios and fill-out factors were determined\nas 3.758, 1.438 and 31.8$\\%$, 14.9$\\%$, respectively. ASAS J124343+1531.7 is a\nW-subtype median contact binary, while LINEAR 2323566 is a W-subtype shallow\ncontact binary, and the asymmetric light curves prove that they both have the\nO'Connell effect, which is generally explained by magnetic activity. The\nequivalent widths (EWs) of H$_\\alpha$ lines were calculated, which show they\ncertainly have magnetic activity. Moreover, LINEAR 2323566 has a stronger\nmagnetic activity. The analysis of orbital period changes shows that ASAS\nJ124343+1531.7 has a trend of secular period increase, which is generally\nexplained by the mass transfer from the less massive to the more massive star.\nAccording to the estimated absolute parameters, their evolutionary states are\ndiscussed. The two components of ASAS J124343+1531.7 are both main sequence\nstars. While for LINEAR 2323566, the more massive star is a main sequence star,\nthe less massive star has evolved out of main sequence and is over-luminous and\nover-sized.\n"}
{"abstract": "  We demonstrate that a conditional wavefunction theory enables a unified and\nefficient treatment of the equilibrium structure and nonadiabatic dynamics of\ncorrelated electron-ion systems. The conditional decomposition of the many-body\nwavefunction formally recasts the full interacting wavefunction of a closed\nsystem as a set of lower dimensional (conditional) coupled `slices'. We\nformulate a variational wavefunction ansatz based on a set of conditional\nwavefunction slices, and demonstrate its accuracy by determining the structural\nand time-dependent response properties of the hydrogen molecule. We then extend\nthis approach to include time-dependent conditional wavefunctions, and address\nparadigmatic nonequilibrium processes including strong-field molecular\nionization, laser driven proton transfer, and Berry phase effects induced by a\nconical intersection. This work paves the road for the application of\nconditional wavefunction theory in equilibrium and out of equilibrium ab-initio\nmolecular simulations of finite and extended systems.\n"}
{"abstract": "  We consider symmetric positive definite preconditioners for multiple\nsaddle-point systems of block tridiagonal form, which can be applied within the\nMINRES algorithm. We describe such a preconditioner for which the\npreconditioned matrix has only two distinct eigenvalues, 1 and -1, when the\npreconditioner is applied exactly. We discuss the relative merits of such an\napproach compared to a more widely studied block diagonal preconditioner, and\nspecify the computational work associated with applying the new preconditioner\ninexactly. Numerical results validate our theoretical findings.\n"}
{"abstract": "  Transient non-gaussian noise in gravitational wave detectors, commonly\nreferred to as glitches, pose challenges for inference of the astrophysical\nproperties of detected signals when the two are coincident in time. Current\nanalyses aim towards modeling and subtracting the glitches from the data using\na flexible, morphology-independent model in terms of sine-gaussian wavelets\nbefore the signal source properties are inferred using templates for the\ncompact binary signal. We present a new analysis of gravitational wave data\nthat contain both a signal and glitches by simultaneously modeling the compact\nbinary signal in terms of templates and the instrumental glitches using\nsine-gaussian wavelets. The model for the glitches is generic and can thus be\napplied to a wide range of glitch morphologies without any special tuning. The\nsimultaneous modeling of the astrophysical signal with templates allows us to\nefficiently separate the signal from the glitches, as we demonstrate using\nsimulated signals injected around real O2 glitches in the two LIGO detectors.\nWe show that our new proposed analysis can separate overlapping glitches and\nsignals, estimate the compact binary parameters, and provide ready-to-use\nglitch-subtracted data for downstream inference analyses.\n"}
{"abstract": "  We investigate the gravitational perturbations of the Schwarzschild black\nhole in the nonlocal gravity model recently proposed by Deser and Woodard\n(DW-II). The analysis is performed in the localized version in which the\nnonlocal corrections are represented by some auxiliary scalar fields. We find\nthat the nonlocal corrections do not affect the axial gravitational\nperturbations, and hence the axial modes are completely identical to those in\nGeneral Relativity (GR). However, the polar modes get different from their GR\ncounterparts when the scalar fields are excited at the background level. In\nsuch a case the polar modes are sourced by an additional massless scalar mode\nand, as a result, the isospectrality between the axial and the polar modes\nbreaks down. We also perform a similar analysis for the predecessor of this\nmodel (DW-I) and arrive at the same conclusion for it.\n"}
{"abstract": "  Human enterprise often suffers from direct negative effects caused by\njellyfish blooms. The investigation of a prior jellyfish monitoring system\nshowed that it was unable to reliably perform in a cross validation setting,\ni.e. in new underwater environments. In this paper, a number of enhancements\nare proposed to the part of the system that is responsible for object\nclassification. First, the training set is augmented by adding synthetic data,\nmaking the deep learning classifier able to generalise better. Then, the\nframework is enhanced by employing a new second stage model, which analyzes the\noutputs of the first network to make the final prediction. Finally, weighted\nloss and confidence threshold are added to balance out true and false\npositives. With all the upgrades in place, the system can correctly classify\n30.16% (comparing to the initial 11.52%) of all spotted jellyfish, keep the\namount of false positives as low as 0.91% (comparing to the initial 2.26%) and\noperate in real-time within the computational constraints of an autonomous\nembedded platform.\n"}
{"abstract": "  We propose three constructions of classically verifiable non-interactive\nzero-knowledge proofs and arguments (CV-NIZK) for QMA in various preprocessing\nmodels.\n  - We construct a CV-NIZK for QMA in the quantum secret parameter model where\na trusted setup sends a quantum proving key to the prover and a classical\nverification key to the verifier. It is information theoretically sound and\nzero-knowledge.\n  - Assuming the quantum hardness of the learning with errors problem, we\nconstruct a CV-NIZK for QMA in a model where a trusted party generates a CRS\nand the verifier sends an instance-independent quantum message to the prover as\npreprocessing. This model is the same as one considered in the recent work by\nColadangelo, Vidick, and Zhang (CRYPTO '20).\n  Our construction has the so-called dual-mode property, which means that there\nare two computationally indistinguishable modes of generating CRS, and we have\ninformation theoretical soundness in one mode and information theoretical\nzero-knowledge property in the other. This answers an open problem left by\nColadangelo et al, which is to achieve either of soundness or zero-knowledge\ninformation theoretically. To the best of our knowledge, ours is the first\ndual-mode NIZK for QMA in any kind of model.\n  - We construct a CV-NIZK for QMA with quantum preprocessing in the quantum\nrandom oracle model. This quantum preprocessing is the one where the verifier\nsends a random Pauli-basis states to the prover. Our construction uses the\nFiat-Shamir transformation. The quantum preprocessing can be replaced with the\nsetup that distributes Bell pairs among the prover and the verifier, and\ntherefore we solve the open problem by Broadbent and Grilo (FOCS '20) about the\npossibility of NIZK for QMA in the shared Bell pair model via the Fiat-Shamir\ntransformation.\n"}
{"abstract": "  We characterize the equality between ultradifferentiable function classes\ndefined in terms of abstractly given weight matrices and in terms of the\ncorresponding matrix of associated weight functions by using new growth\nindices. These indices, defined by means of weight sequences and (associated)\nweight functions, are extending the notion of O-regular variation to a mixed\nsetting. Hence we are extending the known comparison results concerning classes\ndefined in terms of a single weight sequence and of a single weight function\nand give also these statements an interpretation expressed in O-regular\nvariation.\n"}
{"abstract": "  Causal discovery methods seek to identify causal relations between random\nvariables from purely observational data, as opposed to actively collected\nexperimental data where an experimenter intervenes on a subset of correlates.\nOne of the seminal works in this area is the Inferred Causation algorithm,\nwhich guarantees successful causal discovery under the assumption of a\nconditional independence (CI) oracle: an oracle that can states whether two\nrandom variables are conditionally independent given another set of random\nvariables. Practical implementations of this algorithm incorporate statistical\ntests for conditional independence, in place of a CI oracle. In this paper, we\nanalyze the sample complexity of causal discovery algorithms without a CI\noracle: given a certain level of confidence, how many data points are needed\nfor a causal discovery algorithm to identify a causal structure? Furthermore,\nour methods allow us to quantify the value of domain expertise in terms of data\nsamples. Finally, we demonstrate the accuracy of these sample rates with\nnumerical examples, and quantify the benefits of sparsity priors and known\ncausal directions.\n"}
{"abstract": "  We prove a result which provides a link between the decomposition of\nparabolically induced representations and the Bushnell--Kutzko theory of\ntypical representations. As an application, we show that there exists a\nwell-defined inertial Langlands correspondence which respects the monodromy\naction of L-parameters, under some standard conjectures regarding the local\nLanglands correspondence. To allow for potential applications of this inertial\nLanglands correspondence, we also provide a complete construction of the set of\ntypical representations, giving a parametrization of these in terms of the\nstructure of the Bruhat--Tits building of $G$.\n"}
{"abstract": "  Boolean Satisfiability (SAT) is a well-known NP-complete problem. Despite\nthis theoretical hardness, SAT solvers based on Conflict Driven Clause Learning\n(CDCL) can solve large SAT instances from many important domains. CDCL learns\nclauses from conflicts, a technique that allows a solver to prune its search\nspace. The selection heuristics in CDCL prioritize variables that are involved\nin recent conflicts. While only a fraction of decisions generate any conflicts,\nmany generate multiple conflicts.\n  In this paper, we study conflict-generating decisions in CDCL in detail. We\ninvestigate the impact of single conflict (sc) decisions, which generate only\none conflict, and multi-conflict (mc) decisions which generate two or more. We\nempirically characterize these two types of decisions based on the quality of\nthe learned clauses produced by each type of decision. We also show an\nimportant connection between consecutive clauses learned within the same mc\ndecision, where one learned clause triggers the learning of the next one\nforming a chain of clauses. This leads to the consideration of similarity\nbetween conflicts, for which we formulate the notion of conflictsproximity as a\nsimilarity measure. We show that conflicts in mc decisions are more closely\nrelated than consecutive conflicts generated from sc decisions. Finally, we\ndevelop Common Reason Variable Reduction (CRVR) as a new decision strategy that\nreduces the selection priority of some variables from the learned clauses of mc\ndecisions. Our empirical evaluation of CRVR implemented in three leading\nsolvers demonstrates performance gains in benchmarks from the main track of SAT\nCompetition-2020.\n"}
{"abstract": "  Resolving a linear combination of point sources from their band-limited\nFourier data is a fundamental problem in imaging and signal processing. With\nthe incomplete Fourier data and the inevitable noise in the measurement, there\nis a fundamental limit on the separation distance between point sources that\ncan be resolved. This is the so-called resolution limit problem.\nCharacterization of this resolution limit is still a long-standing puzzle\ndespite the prevalent use of the classic Rayleigh limit. It is well-known that\nRayleigh limit is heuristic and its drawbacks become prominent when dealing\nwith data that is subjected to delicate processing, as is what modern\ncomputational imaging methods do. Therefore, more precise characterization of\nthe resolution limit becomes increasingly necessary with the development of\ndata processing methods. For this purpose, we developed a theory of\n\"computational resolution limit\" for both number detection and support recovery\nin one dimension in [arXiv:2003.02917[cs.IT], arXiv:1912.05430[eess.IV]]. In\nthis paper, we extend the one-dimensional theory to multi-dimensions. More\nprecisely, we define and quantitatively characterize the \"computational\nresolution limit\" for the number detection and support recovery problems in a\ngeneral k-dimensional space. Our results indicate that there exists a phase\ntransition phenomenon regarding to the super-resolution factor and the\nsignal-to-noise ratio in each of the two recovery problems. Our main results\nare derived using a subspace projection strategy. Finally, to verify the\ntheory, we proposed deterministic subspace projection based algorithms for the\nnumber detection and support recovery problems in dimension two and three. The\nnumerical results confirm the phase transition phenomenon predicted by the\ntheory.\n"}
{"abstract": "  The relaxation of out-of-time-ordered correlators (OTOCs) has been studied as\na mean to characterize the scrambling properties of a quantum system. We show\nthat the presence of local conserved quantities typically results in, at the\nfastest, an algebraic relaxation of the OTOC provided (i) the dynamics is local\nand (ii) the system follows the eigenstate thermalization hypothesis. Our\nresult relies on the algebraic scaling of the infinite-time value of OTOCs with\nsystem size, which is typical in thermalizing systems with local conserved\nquantities, and on the existence of finite speed of propagation of correlations\nfor finite-range-interaction systems. We show that time-independence of the\nHamiltonian is not necessary as the above conditions (i) and (ii) can occur in\ntime-dependent systems, both periodic or aperiodic. We also remark that our\nresult can be extended to systems with power-law interactions.\n"}
{"abstract": "  This paper introduces a new video-and-language dataset with human actions for\nmultimodal logical inference, which focuses on intentional and aspectual\nexpressions that describe dynamic human actions. The dataset consists of 200\nvideos, 5,554 action labels, and 1,942 action triplets of the form <subject,\npredicate, object> that can be translated into logical semantic\nrepresentations. The dataset is expected to be useful for evaluating multimodal\ninference systems between videos and semantically complicated sentences\nincluding negation and quantification.\n"}
{"abstract": "  We explicitly construct global attractors of fully nonlinear parabolic\nequations. The attractors are decomposed as equilibria (time independent\nsolutions) and heteroclinic orbits (solutions that converge to distinct\nequilibria backwards and forwards in time). In particular, we state necessary\nand sufficient conditions for the occurrence of heteroclinics between\nhyperbolic equilibria, which is accompanied by a method that compute such\nconditions.\n"}
{"abstract": "  In the Milky Way, $\\sim$18 Wolf-Rayet+O (WR+O) binaries are known with\nestimates of their stellar and orbital parameters. Whereas black hole+O (BH+O)\nbinaries are thought to evolve from the former, only one such system is known\nin the Milky Way. To resolve this disparity, it was suggested that upon core\ncollapse, the WR stars receive large kicks such that most of the binaries are\ndisrupted. We reassess this issue, with emphasis on the uncertainty in the\nformation of an accretion disk around wind-accreting BHs in BH+O binaries,\nwhich is key to identifying such systems. We follow the methodology of previous\nwork and apply an improved analytic criterion for the formation of an accretion\ndisk around wind accreting BHs. We then use stellar models to predict the\nproperties of the BH+O binaries which are expected to descend from the observed\nWR+O binaries, if the WR stars would form BHs without a natal kick. We find\nthat disk formation depends sensitively on the O stars' wind velocity, the\nspecific angular momentum carried by the wind, the efficiency of angular\nmomentum accretion by the BH, and the spin of the BH. We show that the\nassumption of a low wind velocity may lead to predicting that most of the BH+O\nstar binaries will have an extended X-ray bright period. However, this is not\nthe case when typical wind velocities of O stars are considered. We find that a\nhigh spin of the BH can boost the duration of the X-ray active phase as well as\nthe X-ray brightness during this phase, producing a strong bias for detecting\nhigh mass BH binaries in X-rays with high BH spin parameters. We conclude that\nlarge BH formation kicks are not required to understand the sparsity of X-ray\nbright BH+O stars in the Milky Way. Probing for a population of X-ray silent\nBH+O systems with alternative methods can inform us about BH kicks and the\nconditions for high energy emission from high mass BH binaries. (Abridged)\n"}
{"abstract": "  Achieving accurate and robust global situational awareness of a complex\ntime-evolving field from a limited number of sensors has been a longstanding\nchallenge. This reconstruction problem is especially difficult when sensors are\nsparsely positioned in a seemingly random or unorganized manner, which is often\nencountered in a range of scientific and engineering problems. Moreover, these\nsensors can be in motion and can become online or offline over time. The key\nleverage in addressing this scientific issue is the wealth of data accumulated\nfrom the sensors. As a solution to this problem, we propose a data-driven\nspatial field recovery technique founded on a structured grid-based\ndeep-learning approach for arbitrary positioned sensors of any numbers. It\nshould be noted that the na\\\"ive use of machine learning becomes prohibitively\nexpensive for global field reconstruction and is furthermore not adaptable to\nan arbitrary number of sensors. In the present work, we consider the use of\nVoronoi tessellation to obtain a structured-grid representation from sensor\nlocations enabling the computationally tractable use of convolutional neural\nnetworks. One of the central features of the present method is its\ncompatibility with deep-learning based super-resolution reconstruction\ntechniques for structured sensor data that are established for image\nprocessing. The proposed reconstruction technique is demonstrated for unsteady\nwake flow, geophysical data, and three-dimensional turbulence. The current\nframework is able to handle an arbitrary number of moving sensors, and thereby\novercomes a major limitation with existing reconstruction methods. The\npresented technique opens a new pathway towards the practical use of neural\nnetworks for real-time global field estimation.\n"}
{"abstract": "  This work aims at intensifying text-independent speaker identification\nperformance in real application situations such as noisy and emotional talking\nconditions. This is achieved by incorporating two different modules: a\nComputational Auditory Scene Analysis CASA based pre-processing module for\nnoise reduction and cascaded Gaussian Mixture Model Convolutional Neural\nNetwork GMM-CNN classifier for speaker identification followed by emotion\nrecognition. This research proposes and evaluates a novel algorithm to improve\nthe accuracy of speaker identification in emotional and highly-noise\nsusceptible conditions. Experiments demonstrate that the proposed model yields\npromising results in comparison with other classifiers when Speech Under\nSimulated and Actual Stress SUSAS database, Emirati Speech Database ESD, the\nRyerson Audio-Visual Database of Emotional Speech and Song RAVDESS database and\nthe Fluent Speech Commands database are used in a noisy environment.\n"}
{"abstract": "  While nanoscale color generations have been studied for years, high\nperformance transmission structural colors, simultaneously equipped with large\ngamut, high resolution, low loss and optical multiplexing abilities, still\nremain as a hanging issue. Here, beneficial from metasurfaces, we demonstrate a\nsilicon metasurface embedded Fabry-P\\'erot cavity (meta-FP cavity), with\npolydimethylsiloxanes (PDMS) surrounding media and silver film mirrors. By\nchanging the planar geometries of the embedded nanopillars, the meta-FP cavity\nprovides transmission colors with ultra large gamut of 194% sRGB and ultrahigh\nresolution of 141111 DPI, along with considerably average transmittance of 43%\nand more than 300% enhanced angular tolerance. Such high density allows\ntwo-dimensional color mixing at diffraction limit scale. The color gamut and\nthe resolution can be flexibly tuned and improved by modifying the silver film\nthickness and the lattice period. The polarization manipulation ability of the\nmetasurface also enables arbitrary color arrangement between cyan and red for\ntwo orthogonal linear polarization states, at deep subwavelength scale. Our\nproposed cavities can be used in filters, printings, optical storages and many\nother applications in need of high quality and density colors.\n"}
{"abstract": "  Astronomers require efficient automated detection and classification\npipelines when conducting large-scale surveys of the (optical) sky for variable\nand transient sources. Such pipelines are fundamentally important, as they\npermit rapid follow-up and analysis of those detections most likely to be of\nscientific value. We therefore present a deep learning pipeline based on the\nconvolutional neural network architecture called $\\texttt{MeerCRAB}$. It is\ndesigned to filter out the so called 'bogus' detections from true astrophysical\nsources in the transient detection pipeline of the MeerLICHT telescope. Optical\ncandidates are described using a variety of 2D images and numerical features\nextracted from those images. The relationship between the input images and the\ntarget classes is unclear, since the ground truth is poorly defined and often\nthe subject of debate. This makes it difficult to determine which source of\ninformation should be used to train a classification algorithm. We therefore\nused two methods for labelling our data (i) thresholding and (ii) latent class\nmodel approaches. We deployed variants of $\\texttt{MeerCRAB}$ that employed\ndifferent network architectures trained using different combinations of input\nimages and training set choices, based on classification labels provided by\nvolunteers. The deepest network worked best with an accuracy of 99.5$\\%$ and\nMatthews correlation coefficient (MCC) value of 0.989. The best model was\nintegrated to the MeerLICHT transient vetting pipeline, enabling the accurate\nand efficient classification of detected transients that allows researchers to\nselect the most promising candidates for their research goals.\n"}
{"abstract": "  We introduce and study $\\ell_p$-norm-multiway-cut: the input here is an\nundirected graph with non-negative edge weights along with $k$ terminals and\nthe goal is to find a partition of the vertex set into $k$ parts each\ncontaining exactly one terminal so as to minimize the $\\ell_p$-norm of the cut\nvalues of the parts. This is a unified generalization of min-sum multiway cut\n(when $p=1$) and min-max multiway cut (when $p=\\infty$), both of which are\nwell-studied classic problems in the graph partitioning literature. We show\nthat $\\ell_p$-norm-multiway-cut is NP-hard for constant number of terminals and\nis NP-hard in planar graphs. On the algorithmic side, we design an $O(\\log^2\nn)$-approximation for all $p\\ge 1$. We also show an integrality gap of\n$\\Omega(k^{1-1/p})$ for a natural convex program and an\n$O(k^{1-1/p-\\epsilon})$-inapproximability for any constant $\\epsilon>0$\nassuming the small set expansion hypothesis.\n"}
{"abstract": "  In this paper, we propose a unifying framework incorporating several\nmomentum-related search directions for solving strongly monotone variational\ninequalities. The specific combinations of the search directions in the\nframework are made to guarantee the optimal iteration complexity bound of\n$\\mathcal{O}\\left(\\kappa\\ln(1/\\epsilon)\\right)$ to reach an\n$\\epsilon$-solution, where $\\kappa$ is the condition number. This framework\nprovides the flexibility for algorithm designers to train -- among different\nparameter combinations -- the one that best suits the structure of the problem\nclass at hand. The proposed framework includes the following iterative points\nand directions as its constituents: the extra-gradient, the optimistic gradient\ndescent ascent (OGDA) direction (aka \"optimism\"), the \"heavy-ball\" direction,\nand Nesterov's extrapolation points. As a result, all the afore-mentioned\nmethods become the special cases under the general scheme of extra points. We\nalso specialize this approach to strongly convex minimization, and show that a\nsimilar extra-point approach achieves the optimal iteration complexity bound of\n$\\mathcal{O}(\\sqrt{\\kappa}\\ln(1/\\epsilon))$ for this class of problems.\n"}
{"abstract": "  We study the transport properties of generic out-of-equilibrium quantum\nsystems connected to fermionic reservoirs. We develop a new method, based on an\nexpansion of the current in terms of the inverse system size and out of\nequilibrium formulations such as the Keldysh technique and the Meir-Wingreen\nformula. Our method allows a simple and compact derivation of the current for a\nlarge class of systems showing diffusive/ohmic behavior. In addition, we obtain\nexact solutions for a large class of quantum stochastic Hamiltonians (QSHs)\nwith time and space dependent noise, using a self consistent Born diagrammatic\nmethod in the Keldysh representation. We show that these QSHs exhibit diffusive\nregimes which are encoded in the Keldysh component of the single particle\nGreen's function. The exact solution for these QSHs models confirms the\nvalidity of our system size expansion ansatz, and its efficiency in capturing\nthe transport properties. We consider in particular three fermionic models: i)\na model with local dephasing ii) the quantum simple symmetric exclusion process\nmodel iii) a model with long-range stochastic hopping. For i) and ii) we\ncompute the full temperature and dephasing dependence of the conductance of the\nsystem, both for two- and four-points measurements. Our solution gives access\nto the regime of finite temperature of the reservoirs which could not be\nobtained by previous approaches. For iii), we unveil a novel\nballistic-to-diffusive transition governed by the range and the nature (quantum\nor classical) of the hopping. As a by-product, our approach equally describes\nthe mean behavior of quantum systems under continuous measurement.\n"}
{"abstract": "  Although lattice gases composed by $k$NN particles, forbidding up to their\n$k$th nearest neighbors of being occupied, have been widely investigated in\nliterature, the location and the universality class of the fluid-columnar\ntransition in the 2NN model on the square lattice is still a topic of debate.\nHere, we present grand-canonical solutions of this model on Husimi lattices\nbuilt with diagonal square lattices, with $2L(L+1)$ sites, for $L \\leqslant 7$.\nThe systematic sequence of mean-field solutions confirms the existence of a\ncontinuous transition in this system and extrapolations of the critical\nchemical potential $\\mu_{2,c}(L)$ and particle density $\\rho_{2,c}(L)$ to $L\n\\rightarrow \\infty$ yield estimates of these quantities in close agreement with\nprevious results for the 2NN model on the square lattice. To confirm the\nreliability of this approach we employ it also for the 1NN model, where very\naccurate estimates for the critical parameters $\\mu_{1,c}$ and $\\rho_{1,c}$ --\nfor the fluid-solid transition in this model on the square lattice -- are found\nfrom extrapolations of data for $L \\leqslant 6$. The non-classical critical\nexponents for these transitions are investigated through the coherent anomaly\nmethod (CAM), which in the 1NN case yields $\\beta$ and $\\nu$ differing by at\nmost 6\\% from the expected Ising exponents. For the 2NN model, the CAM analysis\nis somewhat inconclusive, because the exponents sensibly depend on the value of\n$\\mu_{2,c}$ used to calculate them. Notwithstanding, our results suggest that\n$\\beta$ and $\\nu$ are considerably larger than the Ashkin-Teller exponents\nreported in numerical studies of the 2NN system.\n"}
{"abstract": "  In Standard Model (SM) Higgs Boson pair production initiated by photons\n($\\gamma \\gamma \\to h h$) is loop-generated process and thereby very sensitive\nto any new couplings and particles that may come in loops. The Composite Higgs\nModels provide an alternate mechanism to address the hierarchy problem of SM\nwhere Higgs instead of being an elementary field could be a bound state of a\nstrongly interacting sector. These set of models apart from modifying the SM\nHiggs couplings could also introduce new effective couplings that can have\nsubstantial impact on the loop processes. In this work we have studied the\nimpact of such modifications by Composite Higgs models in $\\gamma\\gamma \\to h\nh$ production process.\n"}
{"abstract": "  We shift the paradigm of feedback control from the control of quantum states\nto the control of phase transitions in quantum systems. We show that feedback\nallows tuning the universality class of phase transitions via modifying its\ncritical exponent. We expand our previous treatment [D. A. Ivanov, T. Yu.\nIvanova, S. F. Caballero-Benitez, and I. B. Mekhov, Phys. Rev. Lett. 124,\n010603 (2020)] of Dicke model and go beyond the approximation of adiabatically\neliminated light field. Both linearized and nonlinear models of spin ensembles\nare considered. The tunability of quantum fluctuations near the critical point\nby the feedbacks of nontrivial shapes is explained by considering the\nfluctuation spectra and the system behavior at single quantum trajectories.\n"}
{"abstract": "  This paper addresses the problem of 3D human body shape and pose estimation\nfrom RGB images. Recent progress in this field has focused on single images,\nvideo or multi-view images as inputs. In contrast, we propose a new task: shape\nand pose estimation from a group of multiple images of a human subject, without\nconstraints on subject pose, camera viewpoint or background conditions between\nimages in the group. Our solution to this task predicts distributions over SMPL\nbody shape and pose parameters conditioned on the input images in the group. We\nprobabilistically combine predicted body shape distributions from each image to\nobtain a final multi-image shape prediction. We show that the additional body\nshape information present in multi-image input groups improves 3D human shape\nestimation metrics compared to single-image inputs on the SSP-3D dataset and a\nprivate dataset of tape-measured humans. In addition, predicting distributions\nover 3D bodies allows us to quantify pose prediction uncertainty, which is\nuseful when faced with challenging input images with significant occlusion. Our\nmethod demonstrates meaningful pose uncertainty on the 3DPW dataset and is\ncompetitive with the state-of-the-art in terms of pose estimation metrics.\n"}
{"abstract": "  Probability theory is far from being the most general mathematical theory of\nuncertainty. A number of arguments point at its inability to describe\nsecond-order ('Knightian') uncertainty. In response, a wide array of theories\nof uncertainty have been proposed, many of them generalisations of classical\nprobability. As we show here, such frameworks can be organised into clusters\nsharing a common rationale, exhibit complex links, and are characterised by\ndifferent levels of generality. Our goal is a critical appraisal of the current\nlandscape in uncertainty theory.\n"}
{"abstract": "  In the metallic pyrochlore Nd$_2$Mo$_2$O$_7$, the conducting Molybdenum\nsublattice adopts canted, yet nearly collinear ferromagnetic order with nonzero\nscalar spin chirality. The chemical potential may be controlled by replacing\nNd$^{3+}$ with Ca$^{2+}$, while introducing only minimal additional disorder to\nthe conducting states. Here, we demonstrate the stability of the canted\nferromagnetic state, including the tilting angle of Molybdenum spins, in\n(Nd$_{1-x}$Ca$_{x}$)$_2$Mo$_2$O$_7$ (NCMO) with $x\\le 0.15$ using magnetic\nsusceptibility measurements. Mo-Mo and Mo-Nd magnetic couplings both change\nsign above $x=0.22$, where the canted ferromagnetic state gives way to a\nspin-glass metallic region. Contributions to the Curie-Weiss law from two\nmagnetic sublattices are separated systematically.\n"}
{"abstract": "  The traditional notion of capacity studied in the context of memoryless\nnetwork communication builds on the concept of block-codes and requires that,\nfor sufficiently large blocklength n, all receiver nodes simultaneously decode\ntheir required information after n channel uses. In this work, we generalize\nthe traditional capacity region by exploring communication rates achievable\nwhen some receivers are required to decode their information before others, at\ndifferent predetermined times; referred here as the \"time-rate\" region. Through\na reduction to the standard notion of capacity, we present an inner-bound on\nthe time-rate region. The time-rate region has been previously studied and\ncharacterized for the memoryless broadcast channel (with a sole common message)\nunder the name \"static broadcasting\".\n"}
{"abstract": "  The chemical and physical evolution of starless and pre-stellar cores are of\nparamount importance to understanding the process of star formation. The Taurus\nMolecular Cloud cores TMC 1-C and TMC 1-CP share similar initial conditions and\nprovide an excellent opportunity to understand the evolution of the pre-stellar\ncore phase. We investigated the evolutionary stage of starless cores based on\nobservations towards the prototypical dark cores TMC 1-C and TMC 1-CP, mapping\nthem in the CS $3\\rightarrow 2$, C$^{34}$S $3\\rightarrow 2$, $^{13}$CS\n$2\\rightarrow 1$, DCN $1\\rightarrow 0$, DCN $2\\rightarrow 1$, DNC $1\\rightarrow\n0$, DNC $2\\rightarrow 1$, DN$^{13}$C $1\\rightarrow 0$, DN$^{13}$C $2\\rightarrow\n1$, N$_2$H$^+$ $1\\rightarrow 0$, and N$_2$D$^+$ $1\\rightarrow 0$ transitions.\nWe performed a multi-transitional study of CS and its isotopologs, DCN, and DNC\nlines to characterize the physical and chemical properties of these cores. We\nstudied their chemistry using the state-of-the-art gas-grain chemical code\nNautilus and pseudo time-dependent models to determine their evolutionary\nstage. Observational diagnostics seem to indicate that TMC 1-C is in a later\nevolutionary stage than TMC 1-CP, with a chemical age $\\sim$1 Myr. TMC 1-C\nshows signs of being an evolved core at the onset of star formation, while TMC\n1-CP appears to be in an earlier evolutionary stage due to a more recent\nformation or, alternatively, a collapse slowed down by a magnetic support.\n"}
{"abstract": "  Measures of face identification proficiency are essential to ensure accurate\nand consistent performance by professional forensic face examiners and others\nwho perform face identification tasks in applied scenarios. Current proficiency\ntests rely on static sets of stimulus items, and so, cannot be administered\nvalidly to the same individual multiple times. To create a proficiency test, a\nlarge number of items of \"known\" difficulty must be assembled. Multiple tests\nof equal difficulty can be constructed then using subsets of items. Here, we\nintroduce a proficiency test, the Triad Identity Matching (TIM) test, based on\nstimulus difficulty measures based on Item Response Theory (IRT). Participants\nview face-image \"triads\" (N=225) (two images of one identity and one image of a\ndifferent identity) and select the different identity. In Experiment 1,\nuniversity students (N=197) showed wide-ranging accuracy on the TIM test.\nFurthermore, IRT modeling demonstrated that the TIM test produces items of\nvarious difficulty levels. In Experiment 2, IRT-based item difficulty measures\nwere used to partition the TIM test into three equally \"easy\" and three equally\n\"difficult\" subsets. Simulation results indicated that the full set, as well as\ncurated subsets, of the TIM items yielded reliable estimates of subject\nability. In summary, the TIM test can provide a starting point for developing a\nframework that is flexible, calibrated, and adaptive to measure proficiency\nacross various ability levels (e.g., professionals or populations with face\nprocessing deficits)\n"}
{"abstract": "  Deep Spiking Neural Networks (SNNs) present optimization difficulties for\ngradient-based approaches due to discrete binary activation and complex\nspatial-temporal dynamics. Considering the huge success of ResNet in deep\nlearning, it would be natural to train deep SNNs with residual learning.\nPrevious Spiking ResNet mimics the standard residual block in ANNs and simply\nreplaces ReLU activation layers with spiking neurons, which suffers the\ndegradation problem and can hardly implement residual learning. In this paper,\nwe propose the spike-element-wise (SEW) ResNet to realize residual learning in\ndeep SNNs. We prove that the SEW ResNet can easily implement identity mapping\nand overcome the vanishing/exploding gradient problems of Spiking ResNet. We\nevaluate our SEW ResNet on ImageNet, DVS Gesture, and CIFAR10-DVS datasets, and\nshow that SEW ResNet outperforms the state-of-the-art directly trained SNNs in\nboth accuracy and time-steps. Moreover, SEW ResNet can achieve higher\nperformance by simply adding more layers, providing a simple method to train\ndeep SNNs. To our best knowledge, this is the first time that directly training\ndeep SNNs with more than 100 layers becomes possible. Our codes are available\nat https://github.com/fangwei123456/Spike-Element-Wise-ResNet.\n"}
{"abstract": "  We generalize leading-order asymptotics of a form of the heat content of a\nsubmanifold (van den Berg & Gilkey 2015) to the setting of time-dependent\ndiffusion processes in the limit of vanishing diffusivity. Such diffusion\nprocesses arise naturally when advection-diffusion processes are viewed in\nLagrangian coordinates. We prove that as diffusivity $\\varepsilon$ goes to\nzero, the diffusive transport out of a material set $S$ under the\ntime-dependent, mass-preserving advection-diffusion equation with initial\ncondition given by the characteristic function $\\mathds{1}_S$, is\n$\\sqrt{\\varepsilon/\\pi} d\\overline{A}(\\partial S) + o(\\sqrt{\\varepsilon})$. The\nsurface measure $d\\overline A$ is that of the so-called geometry of mixing, as\nintroduced in (Karrasch & Keller, 2020). We apply our result to the\ncharacterisation of coherent structures in time-dependent dynamical systems.\n"}
{"abstract": "  Instance segmentation of planar regions in indoor scenes benefits visual SLAM\nand other applications such as augmented reality (AR) where scene understanding\nis required. Existing methods built upon two-stage frameworks show satisfactory\naccuracy but are limited by low frame rates. In this work, we propose a\nreal-time deep neural architecture that estimates piece-wise planar regions\nfrom a single RGB image. Our model employs a variant of a fast single-stage CNN\narchitecture to segment plane instances. Considering the particularity of the\ntarget detected, we propose Fast Feature Non-maximum Suppression (FF-NMS) to\nreduce the suppression errors resulted from overlapping bounding boxes of\nplanes. We also utilize a Residual Feature Augmentation module in the Feature\nPyramid Network (FPN). Our method achieves significantly higher frame-rates and\ncomparable segmentation accuracy against two-stage methods. We automatically\nlabel over 70,000 images as ground truth from the Stanford 2D-3D-Semantics\ndataset. Moreover, we incorporate our method with a state-of-the-art planar\nSLAM and validate its benefits.\n"}
{"abstract": "  To date, formal models of collective intelligence have lacked a plausible\nmathematical description of the relationship between local-scale interactions\nbetween highly autonomous sub-system components (individuals) and global-scale\nbehavior of the composite system (the collective). In this paper we use the\nActive Inference Formulation (AIF), a framework for explaining the behavior of\nany non-equilibrium steady state system at any scale, to posit a minimal\nagent-based model that simulates the relationship between local\nindividual-level interaction and collective intelligence (operationalized as\nsystem-level performance). We explore the effects of providing baseline AIF\nagents (Model 1) with specific cognitive capabilities: Theory of Mind (Model\n2); Goal Alignment (Model 3), and Theory of Mind with Goal Alignment (Model 4).\nThese stepwise transitions in sophistication of cognitive ability are motivated\nby the types of advancements plausibly required for an AIF agent to persist and\nflourish in an environment populated by other AIF agents, and have also\nrecently been shown to map naturally to canonical steps in human cognitive\nability. Illustrative results show that stepwise cognitive transitions increase\nsystem performance by providing complementary mechanisms for alignment between\nagents' local and global optima. Alignment emerges endogenously from the\ndynamics of interacting AIF agents themselves, rather than being imposed\nexogenously by incentives to agents' behaviors (contra existing computational\nmodels of collective intelligence) or top-down priors for collective behavior\n(contra existing multiscale simulations of AIF). These results shed light on\nthe types of generic information-theoretic patterns conducive to collective\nintelligence in human and other complex adaptive systems.\n"}
{"abstract": "  Tunnelling lies at the heart of quantum mechanics and is a fundamental\nprocess in attosecond science, molecular biology, and quantum devices. Whether\ntunnelling takes time and how a microscopic particle transits through a barrier\nhave been debated since the early days of quantum mechanics. The time required\nfor an electron to tunnel through an atomic potential barrier has been measured\nwith attosecond angular streaking (attoclock), and a recent work on the\nhydrogen atom claimed that electron tunnelling is instantaneous. However, the\ntime required for Rb atoms to tunnel through an optical potential barrier has\nbeen measured to be on the order of milliseconds with a recent Larmor clock\nmeasurement. The essence of electron and atom tunnelling is identical, but the\nreason for the contradictory conclusions remains unknown. Here, we demonstrate\nthat the sub-barrier potential interaction is the root of the nonzero\ntunnelling delay time. We reveal that the wave-particle duality of a tunnelling\nelectron must be fully taken into account when decoding the tunnelling delay\ntime from the attoclock measurements. Based on energy-resolved attoclock\nmeasurements, we show that the tunnelling delay time of an electron ranges from\n24 to 58 attoseconds, and the counterintuitive result that an electron with a\nlower energy may spend less time in the barrier is consistent with the\nvelocity-dependent tunnelling time of atoms in Larmor clock measurements. Our\nresults unify the tunnelling time of microscopic particles by highlighting the\nclassically forbidden sub-barrier potential interactions of matter waves.\n"}
{"abstract": "  After examining the Feynman diagrams corresponding to the $\\bar D^{(*)}\n\\Sigma_c^{(*)}$, $\\bar D^{(*)} \\Lambda_c$, $D^{(*)} \\bar K^{*}$, and $D^{(*)}\n\\bar D^{(*)}$ hadronic molecular states, we propose a possible binding\nmechanism induced by shared light quarks. This mechanism is similar to the\ncovalent bond in chemical molecules induced by shared electrons. We use the\nmethod of QCD sum rules to calculate its corresponding light-quark-exchange\ndiagrams, and the obtained results indicate a model-independent hypothesis: the\nlight-quark-exchange interaction is attractive when the shared light quarks are\ntotally antisymmetric so that obey the Pauli principle. We build a toy model\nwith four parameters to formulize this picture, and estimate binding energies\nof some possibly-existing covalent hadronic molecules. A unique feature of this\npicture is that binding energies of the $(I)J^P = (0)1^+$ $D\\bar B^*/D^* \\bar\nB$ hadronic molecules are much larger than those of the $(I)J^P = (0)1^+$\n$DD^*/\\bar B \\bar B^*$ ones, while the $(I)J^P = (1/2)1/2^+$ $\\bar D\n\\Sigma_c/\\bar D \\Sigma_b/B \\Sigma_c/B \\Sigma_b$ hadronic molecules have similar\nbinding energies.\n"}
{"abstract": "  We generally expect quantum systems to thermalize and satisfy the eigenstate\nthermalization hypothesis (ETH), which states that finite energy density\neigenstates are thermal. However, some systems, such as many-body localized\nsystems and systems with quantum many-body scars, violate ETH and have\nhigh-energy athermal eigenstates. In systems with scars, most eigenstates\nthermalize, but a few atypical scar states do not. Scar states can give rise to\na periodic revival when time-evolving particular initial product states, which\ncan be detected experimentally. Recently, a family of spin Hamiltonians was\nfound with magnetically ordered 3-colored eigenstates that are quantum\nmany-body scars [Lee et al. Phys. Rev. B 101, 241111(2020)]. These models can\nbe realized in any lattice that can be tiled by triangles, such as the\ntriangular or kagome lattices, and have been shown to have close connections to\nthe physics of quantum spin liquids in the Heisenberg kagome antiferromagnet.\nIn this work, we introduce a generalized family of $n$-colored Hamiltonians\nwith \"spiral colored\" eigenstates made from $n$-spin motifs such as polygons or\npolyhedra. We show how these models can be realized in many different lattice\ngeometries and provide numerical evidence that they can exhibit quantum\nmany-body scars with periodic revivals that can be observed by time-evolving\nsimple product states. The simple structure of these Hamiltonians makes them\npromising candidates for future experimental studies of quantum many-body\nscars.\n"}
{"abstract": "  We propose a novel mechanism for the photo-induced superconducting-like\nresponse recently reported in cuprates and other strongly correlated materials.\nThis mechanism relies on quantum-fluctuating bosons consisting of electron\npairs. With periodic drive, the electron pairs and vacancies of pairs form a\ncoherent non-equilibrium condensate, different from conventional\nsuperconductors, yet showing superconducting-like response in some regimes even\nwith dissipation. Unlike the case of driven fermionic bands which results in\nthe familiar Floquet bands with hybridization gaps, for driven bosons the \"gap\"\nopens up in the momentum direction, resulting in a resonant region in momentum\nspace where the eigenvalues are complex. We give a simple physical argument why\nthis picture leads to a \"perfect conductor\" which exhibits superconducting-like\nfrequency-dependent conductivity but no Meissner response. While our model is\nquite general, in the case of cuprates, quantum-fluctuating pair density wave\nin the pseudogap region may serve as the origin of the quantum-fluctuating\nelectron pairs.\n"}
{"abstract": "  A playtest is the process in which human testers are recruited to play video\ngames and to reveal software bugs. Manual testing is expensive and\ntime-consuming, especially when there are many mobile games to test and every\nsoftware version requires for extensive testing before being released. Existing\ntesting frameworks (e.g., Android Monkey) are limited because they adopt no\ndomain knowledge to play games. Learning-based tools (e.g., Wuji) involve a\nhuge amount of training data and computation before testing any game. This\npaper presents LIT -- our lightweight approach to generalize playtesting\ntactics from manual testing, and to adopt the generalized tactics to automate\ngame testing. LIT consists of two phases. In Phase I, while a human plays an\nAndroid game app G for a short period of time (e.g., eight minutes), \\tool\nrecords the user's actions (e.g., swipe) and the scene before each action.\nBased on the collected data, LIT generalizes a set of \\emph{context-aware,\nabstract playtesting tactics} which describe under what circumstances, what\nactions can be taken to play the game. In Phase II, LIT tests G based on the\ngeneralized tactics. Namely, given a randomly generated game scene, LIT\nsearches match for the abstract context of any inferred tactic; if there is a\nmatch, LIT customizes the tactic and generates a feasible event to play the\ngame. Our evaluation with nine games shows LIT to outperform two\nstate-of-the-art tools. This implies that by automating playtest, LIT will\nsignificantly reduce manual testing and boost the quality of game apps.\n"}
{"abstract": "  This article is devoted to studying individual ergodic theorems for\nsubsequential weighted ergodic averages on the noncommutative Lp-spaces\nassociated to a semifinite von Neumann algebra M. In particular, we establish\nthe convergence of these averages along sequences with density one and certain\ntypes of block sequences with positive lower density, and we extend known\nresults along uniform sequences in the sense of Brunel and Keane.\n"}
{"abstract": "  We design a fast algorithm that computes, for a given linear differential\noperator with coefficients in $Z[x ]$, all the characteristic polynomials of\nits p-curvatures, for all primes $p < N$ , in asymptotically quasi-linear bit\ncomplexity in N. We discuss implementations and applications of our algorithm.\nWe shall see in particular that the good performances of our algorithm are\nquickly visible.\n"}
{"abstract": "  Few-shot learning aims to learn a classifier using a few labelled instances\nfor each class. Metric-learning approaches for few-shot learning embed\ninstances into a high-dimensional space and conduct classification based on\ndistances among instance embeddings. However, such instance embeddings are\nusually shared across all episodes and thus lack the discriminative power to\ngeneralize classifiers according to episode-specific features. In this paper,\nwe propose a novel approach, namely \\emph{Episode Adaptive Embedding Network}\n(EAEN), to learn episode-specific embeddings of instances. By leveraging the\nprobability distributions of all instances in an episode at each channel-pixel\nembedding dimension, EAEN can not only alleviate the overfitting issue\nencountered in few-shot learning tasks, but also capture discriminative\nfeatures specific to an episode. To empirically verify the effectiveness and\nrobustness of EAEN, we have conducted extensive experiments on three widely\nused benchmark datasets, under various combinations of different generic\nembedding backbones and different classifiers. The results show that EAEN\nsignificantly improves classification accuracy about $10\\%$ to $20\\%$ in\ndifferent settings over the state-of-the-art methods.\n"}
{"abstract": "  We investigate the properties of foliated gauge fields and construct several\nfoliated field theories in 3+1d that describe foliated fracton orders both with\nand without matter, including the recent hybrid fracton models. These field\ntheories describe Abelian or non-Abelian gauge theories coupled to foliated\ngauge fields, and they fall into two classes of models that we call the\nelectric models and the magnetic models. We show that these two classes of\nfoliated field theories enjoy a duality. We also construct a model (using\nfoliated gauge fields and an exactly solvable lattice Hamiltonian model) for a\nsubsystem-symmetry protected topological (SSPT) phase, which is analogous to a\none-form symmetry protected topological phase, with the subsystem symmetry\nacting on codimension-two subregions. We construct the corresponding gauged\nSSPT phase as a foliated two-form gauge theory. Some instances of the gauged\nSSPT phase are a variant of the X-cube model with the same ground state\ndegeneracy and the same fusion, but different particle statistics.\n"}
{"abstract": "  Security is one of the main challenges of the Internet of Things (IoT). IoT\ndevices are mainly powered by low-cost microcontrollers (MCUs) that typically\nlack basic hardware security mechanisms to separate security-critical\napplications from less critical components. Recently, Arm has started to\nrelease Cortex-M MCUs enhanced with TrustZone technology (i.e., TrustZone-M), a\nsystem-wide security solution aiming at providing robust protection for IoT\ndevices. Trusted Execution Environments (TEEs) relying on TrustZone hardware\nhave been perceived as safe havens for securing mobile devices. However, for\nthe past few years, considerable effort has gone into unveiling hundreds of\nvulnerabilities and proposing a collection of relevant defense techniques to\naddress several issues. While new TEE solutions built on TrustZone-M start\nflourishing, the lessons gathered from the research community appear to be\nfalling short, as these new systems are trapping into the d\\'ej\\`a vu pitfalls\nof the past. In this paper, we present uTango, the first multi-world TEE for\nmodern IoT devices. uTango proposes a novel architecture aiming at tackling the\nmajor architectural deficiencies currently affecting TrustZone(-M)-assisted\nTEEs. In particular, we leverage the very same TrustZone hardware primitives\nused by dual-world implementations to create multiple, equally-secure execution\nenvironments within the normal world. We demonstrate the benefits of uTango by\nconducting an extensive evaluation on a real TrustZone-M hardware platform,\ni.e., Arm Musca-B1. uTango will be open-sourced and freely available on GitHub\nin hopes of engaging academia and industry on securing the foreseeable trillion\nIoT devices.\n"}
{"abstract": "  We address the exploitation of an optical parametric oscillator (OPO) in the\ntask of mitigating, at least partially, phase noise produced by phase\ndiffusion. In particular, we analyze two scenarios where phase diffusion is\ntypically present. The first one is the estimation of the phase of an optical\nfield, while the second involves a quantum communication protocol based on\nphase shift keying. In both cases, we prove that an OPO may lead to a partial\nor full compensation of the noise.\n"}
{"abstract": "  Recent advances in the high sensitivity spectroscopy have made it possible,\nin combination with accurate theoretical predictions, to observe for the first\ntime very weak electric quadrupole transitions in a polar polyatomic molecule\nof water. Here we present accurate theoretical predictions of the complete\nquadrupole ro-vibrational spectrum of a non-polar molecule CO$_2$, important in\natmospheric and astrophysical applications. Our predictions are validated by\nrecent cavity enhanced absorption spectroscopy measurements and are used to\nassign few weak features in the recent ExoMars ACS MIR spectroscopic\nobservations of the martian atmosphere. Predicted quadrupole transitions appear\nin some of the mid-infrared CO$_2$ and water vapor transparency regions, making\nthem important for detection and characterization of the minor absorbers in\nwater- and CO$_2$-rich environments, such as present in the atmospheres of\nEarth, Venus and Mars.\n"}
{"abstract": "  The kinematic features of a centaur-type humanoid platform, combined with a\npowerful actuation, enable the experimentation of a variety of agile and\ndynamic motions. However, the higher number of degrees-of-freedom and the\nincreased weight of the system, compared to the bipedal and quadrupedal\ncounterparts, pose significant research challenges in terms of computational\nload and real implementation. To this end, this work presents a control\narchitecture to perform agile actions, conceived for torque-controlled\nplatforms, which decouples for computational purposes offline optimal control\nplanning of lower-body primitives, based on a template kinematic model, and\nonline control of the upper-body motion to maintain balance. Three stabilizing\nstrategies are presented, whose performance is compared in two types of\nsimulated jumps, while experimental validation is performed on a half-squat\njump using the CENTAURO robot.\n"}
{"abstract": "  We extend the calculation of dark matter direct detection rates via\nelectronic transitions in general dielectric crystal targets, combining\nstate-of-the-art density functional theory calculations of electronic band\nstructures and wave functions near the band gap, with semi-analytic\napproximations to include additional states farther away from the band gap. We\nshow, in particular, the importance of all-electron reconstruction for\nrecovering large momentum components of electronic wave functions, which,\ntogether with the inclusion of additional states, has a significant impact on\ndirect detection rates, especially for heavy mediator models and at\n$\\mathcal{O}(10\\,\\text{eV})$ and higher energy depositions. Applying our\nframework to silicon and germanium (that have been established already as\nsensitive dark matter detectors), we find that our extended calculations can\nappreciably change the detection prospects. Our calculational framework is\nimplemented in an open-source program $\\texttt{EXCEED-DM}$ (EXtended\nCalculation of Electronic Excitations for Direct detection of Dark Matter), to\nbe released in an upcoming publication.\n"}
{"abstract": "  The concept of spin ice can be extended to a general graph. We study the\ndegeneracy of spin ice graph on arbitrary interaction structures via graph\ntheory. Via the mapping of spin ices to the Ising model, we clarify whether the\ninverse mapping is possible via a modified Krausz construction. From the gauge\nfreedom of frustrated Ising systems, we derive exact, general results about\nfrustration and degeneracy. We demonstrate for the first time that every spin\nice graph, with the exception of the 1D Ising model, is degenerate. We then\nstudy how degeneracy scales in size, using the mapping between Eulerian trails\nand spin ice manifolds, and a permanental identity for the number of Eulerian\norientations. We show that the Bethe permanent technique provides both an\nestimate and a lower bound to the frustration of spin ices on arbitrary graphs\nof even degree. While such technique can be used also to obtain an upper bound,\nwe find that in all the examples we studied but one, another upper bound based\non Schrijver inequality is tighter.\n"}
{"abstract": "  In this paper, we focus on category-level 6D pose and size estimation from\nmonocular RGB-D image. Previous methods suffer from inefficient category-level\npose feature extraction which leads to low accuracy and inference speed. To\ntackle this problem, we propose a fast shape-based network (FS-Net) with\nefficient category-level feature extraction for 6D pose estimation. First, we\ndesign an orientation aware autoencoder with 3D graph convolution for latent\nfeature extraction. The learned latent feature is insensitive to point shift\nand object size thanks to the shift and scale-invariance properties of the 3D\ngraph convolution. Then, to efficiently decode category-level rotation\ninformation from the latent feature, we propose a novel decoupled rotation\nmechanism that employs two decoders to complementarily access the rotation\ninformation. Meanwhile, we estimate translation and size by two residuals,\nwhich are the difference between the mean of object points and ground truth\ntranslation, and the difference between the mean size of the category and\nground truth size, respectively. Finally, to increase the generalization\nability of FS-Net, we propose an online box-cage based 3D deformation mechanism\nto augment the training data. Extensive experiments on two benchmark datasets\nshow that the proposed method achieves state-of-the-art performance in both\ncategory- and instance-level 6D object pose estimation. Especially in\ncategory-level pose estimation, without extra synthetic data, our method\noutperforms existing methods by 6.3% on the NOCS-REAL dataset.\n"}
{"abstract": "  We study past-directed extendibility of\nFriedmann-Lema\\^{i}tre-Robertson-Walker (FLRW) and Bianchi type I spacetimes\nwith a scale factor vanishing in the past. We give criteria for determining\nwhether a boundary for past-directed incomplete geodesics is a parallelly\npropagated curvature singularity, which cannot necessarily be read off from\nscalar curvature invariants. It is clarified that, for incomplete FLRW\nspacetime to avoid the singularity, the spacetime necessarily reduces to the\nMilne universe or flat de Sitter universe toward the boundary. For incomplete\nBianchi type I spacetime to be free of singularity, it is necessary that the\nspacetime asymptotically fits into the product of the extendible isotropic\ngeometry (Milne or flat de Sitter) and flat space, or, anisotropic spacetime\nwith specific power law scale factors. Furthermore, we investigate in detail\nthe time-dependence of the scale factor compatible with the extendibility in\nboth spacetimes beyond the leading order.\n"}
{"abstract": "  This paper proposes a learning reference governor (LRG) approach to enforce\nstate and control constraints in systems for which an accurate model is\nunavailable, and this approach enables the reference governor to gradually\nimprove command tracking performance through learning while enforcing the\nconstraints during learning and after learning is completed. The learning can\nbe performed either on a black-box type model of the system or directly on the\nhardware. After introducing the LRG algorithm and outlining its theoretical\nproperties, this paper investigates LRG application to fuel truck (tank truck)\nrollover avoidance. Through simulations based on a fuel truck model that\naccounts for liquid fuel sloshing effects, we show that the proposed LRG can\neffectively protect fuel trucks from rollover accidents under various operating\nconditions.\n"}
{"abstract": "  Insider threats are the cyber attacks from within the trusted entities of an\norganization. Lack of real-world data and issue of data imbalance leave insider\nthreat analysis an understudied research area. To mitigate the effect of skewed\nclass distribution and prove the potential of multinomial classification\nalgorithms for insider threat detection, we propose an approach that combines\ngenerative model with supervised learning to perform multi-class classification\nusing deep learning. The generative adversarial network (GAN) based insider\ndetection model introduces Conditional Generative Adversarial Network (CGAN) to\nenrich minority class samples to provide data for multi-class anomaly\ndetection. The comprehensive experiments performed on the benchmark dataset\ndemonstrates the effectiveness of introducing GAN derived synthetic data and\nthe capability of multi-class anomaly detection in insider activity analysis.\nMoreover, the method is compared with other existing methods against different\nparameters and performance metrics.\n"}
{"abstract": "  Ambient backscatter communication (BackCom) is faced with the challenge that\na single BackCom device can occupy multiple orthogonal resource blocks\nunintentionally. As a result, in order to avoid co-channel interference, a\nconventional approach is to serve multiple BackCom devices in different time\nslots, which reduces both spectral efficiency and connectivity. This letter\ndemonstrates that the use of non-orthogonal multiple access (NOMA) can\nefficiently improve the system throughput and support massive connectivity in\nambient BackCom networks. In particular, two transceiver design approaches are\ndeveloped in the letter to realize different tradeoffs between system\nperformance and complexity.\n"}
{"abstract": "  In this letter, we performed molecular dynamics simulations of ice nucleation\non a rigid surface model of cubic zinc blende structure with different surface\ndipole strength and orientation. It follows that, despite the excellent lattice\nmatch between cubic ice and substrates, the ice nucleation happened only when\nthe interfacial water molecules (IWs) have the same or similar orientations as\nthat of the water molecules in cubic ice. The free energy landscapes revealed\nthat, for substrates with improper dipole strength/orientation, large free\nenergy barriers arose to prevent the dipole of IWs rotating to the right\norientation to trigger ice formation. Our results suggest that the traditional\nconcept of lattice match, the similarity of lattice length between a substrate\nand the new-formed crystalline, should be extended to a broader match include\nthe similarity between the molecular orientations of the interfacial component\nand the component in the specific new-formed crystalline face.\n"}
{"abstract": "  Based on symmetry consideration, quasi-one-dimensional (1D) objects, relevant\nto numerous observables or phenomena, can be classified into eight different\ntypes. We provide various examples of each 1D type, and discuss their Symmetry\nOperational Similarity (SOS) relationships, which are often permutable. A\nnumber of recent experimental observations, including current-induced\nmagnetization in polar or chiral conductors, non-linear Hall effect in polar\nconductors, spin-polarization of tunneling current to chiral conductors, and\nferro-rotational domain imaging with linear gyration are discussed in terms of\n(permutable) SOS. In addition, based on (permutable) SOS, we predict a large\nnumber of new phenomena in low symmetry materials that can be experimentally\nverified in the future.\n"}
{"abstract": "  Since March 2020, the environment surrounding football has changed\ndramatically because of the COVID-19 pandemic. After a few months' break,\nre-scheduled matches were held behind closed doors without spectators. The main\nobjective of this study is a quantitative evaluation of ``crowd effects'' on\nhome advantage, using the results of these closed matches. The proposed\nanalysis uses pairwise comparison method to reduce the effects caused by the\nunbalanced schedule. The following conclusions were drawn from the statistical\nhypothesis tests conducted in this study: In four major European leagues, the\nhome advantage is reduced in closed matches compared to than in the normal\nsituation, i.e., with spectators. The reduction amounts among leagues were\ndifferent. For example, in Germany, the home advantage was negative during the\nclosed-match period. On the other hand,in England,statistically significant\ndifferences in home advantage were not observed between closed matches and\nnormal situation.\n"}
{"abstract": "  Random scattering of photons in disordered one-dimensional solids gives rise\nto an exponential suppression of transmission, which is known as Anderson\nlocalization. Here, we experimentally study Anderson localization in a\nsuperconducting waveguide quantum electrodynamics system comprising eight\nindividually tunable qubits coupled to a photonic continuum of a waveguide.\nEmploying the qubit frequency control, we artificially introduce frequency\ndisorder to the system and observe an exponential suppression of the\ntransmission coefficient in the vicinity of its subradiant dark modes. The\nlocalization length decreases with the disorder strength, which we control\nin-situ by varying individual qubit frequencies. Employing a one-dimensional\nnon-interacting model of coupled qubits and photons, we are able to support and\ncomplement the experimental results. The difference between our investigation\nand previous studies of localization in qubit arrays is the coupling via a\ncommon waveguide, allowing us to explore the localization of mediating photons\nin an intrinsically open system. The experiment opens the door to the study of\nvarious localization phenomena on a new platform, which offers a high degree of\ncontrol and readout possibilities.\n"}
{"abstract": "  This paper studies the estimation of network connectedness with focally\nsparse structure. We try to uncover the network effect with a flexible sparse\ndeviation from a predetermined adjacency matrix. To be more specific, the\nsparse deviation structure can be regarded as latent or misspecified linkages.\nTo obtain high-quality estimator for parameters of interest, we propose to use\na double regularized high-dimensional generalized method of moments (GMM)\nframework. Moreover, this framework also facilitates us to conduct the\ninference. Theoretical results on consistency and asymptotic normality are\nprovided with accounting for general spatial and temporal dependency of the\nunderlying data generating processes. Simulations demonstrate good performance\nof our proposed procedure. Finally, we apply the methodology to study the\nspatial network effect of stock returns.\n"}
{"abstract": "  Globally, 7-10 billion tonnes of waste are produced annually, including\n300-500 million tonnes of hazardous wastes (HW)--explosive, flammable, toxic,\ncorrosive, and infective ones. About 10 % of these HW are traded through a\nworld-wide waste web (W4). The volume of HW traded through the W4 in the last\n30 years has grown by 500 % and will continue to grow, creating serious legal,\neconomic, environmental and health problems at global scale. Here we\ninvestigate the tip of the iceberg of the W4 by studying networks of 108\ncategories of wastes traded among 163 countries in the period 2003-2009.\nAlthough, most of the HW were traded between developed nations, a\ndisproportionate asymmetry existed in the flow of waste from developed to\ndeveloping countries. Using a dynamical model we simulate how waste congestion\npropagates through the W4. We identify 32 countries with poor environmental\nperformance which are at high risk of waste congestion. Therefore, they are a\nthreat of improper handling and disposal of HW. We found contamination by heavy\nmetals (HM), by volatile organic compounds (VOC) and/or by persistent organic\npollutants (POP), which were used as chemical fingerprints (CF) of the improper\nhandling of HW in 94 % of these countries.\n"}
{"abstract": "  In this work we define some map-germs, called elementary joins, for the\npurpose of producing new ${\\mathcal A}$-finite map-germs from them. In\nparticular, we describe a general form of an ${\\mathcal A}$-finite monomial map\nfrom $(\\mathbb{C}^n,0)$ to $(\\mathbb{C}^{p},0)$ for $p\\geq 2n$ of any corank in\nterms of elementary join maps. Our main tools are the delta invariant and some\ninvariants of curves.\n"}
{"abstract": "  We present a quantum kernel method for high-dimensional data analysis using\nGoogle's universal quantum processor, Sycamore. This method is successfully\napplied to the cosmological benchmark of supernova classification using real\nspectral features with no dimensionality reduction and without vanishing kernel\nelements. Instead of using a synthetic dataset of low dimension or\npre-processing the data with a classical machine learning algorithm to reduce\nthe data dimension, this experiment demonstrates that machine learning with\nreal, high dimensional data is possible using a quantum processor; but it\nrequires careful attention to shot statistics and mean kernel element size when\nconstructing a circuit ansatz. Our experiment utilizes 17 qubits to classify 67\ndimensional data - significantly higher dimensionality than the largest prior\nquantum kernel experiments - resulting in classification accuracy that is\ncompetitive with noiseless simulation and comparable classical techniques.\n"}
{"abstract": "  Previous X-ray studies of the Perseus Cluster, consisting of 85 Suzaku\npointings along eight azimuthal directions, revealed a particularly steep\ndecrease in the projected temperature profile near the virial radius (~r200)\ntowards the northwest (NW). To further explore this shock candidate, another 4\nSuzaku observations on the NW edge of the Perseus Cluster have been obtained.\nThese deeper data were designed to provide the best possible control of\nsystematic uncertainties in the spectral analysis. Using the combined Suzaku\nobservations, we have carefully investigated this interesting region by\nanalyzing the spectra of various annuli and extracting projected thermodynamic\nprofiles. We find that the projected temperature profile shows a break near\nr200, indicating a shock with M = 1.9+-0.3. Corresponding discontinuities are\nalso found in the projected emission measure and the density profiles at the\nsame location. This evidence of a shock front so far away from the cluster\ncenter is unprecedented, and may provide a first insight into the properties of\nlarge-scale virial shocks which shape the process of galaxy cluster growth.\n"}
{"abstract": "  On any metric space, I provide an intrinsic characterization of those\ncomplex-valued functions which are uniform limits of Lipschitz functions. There\nare applications to function theory on complete Riemannian manifolds and, in\nparticular, on bounded symmetric domains in complex Euclidean space.\n"}
{"abstract": "  Sparse regression and classification estimators capable of group selection\nhave application to an assortment of statistical problems, from multitask\nlearning to sparse additive modeling to hierarchical selection. This work\nintroduces a class of group-sparse estimators that combine group subset\nselection with group lasso or ridge shrinkage. We develop an optimization\nframework for fitting the nonconvex regularization surface and present\nfinite-sample error bounds for estimation of the regression function. Our\nmethods and analyses accommodate the general setting where groups overlap. As\nan application of group selection, we study sparse semiparametric modeling, a\nprocedure that allows the effect of each predictor to be zero, linear, or\nnonlinear. For this task, the new estimators improve across several metrics on\nsynthetic data compared to alternatives. Finally, we demonstrate their efficacy\nin modeling supermarket foot traffic and economic recessions using many\npredictors. All of our proposals are made available in the scalable\nimplementation grpsel.\n"}
{"abstract": "  Electrical Impedance Tomography gives rise to the severely ill-posed\nCalder\\'on problem of determining the electrical conductivity distribution in a\nbounded domain from knowledge of the associated Dirichlet-to-Neumann map for\nthe governing equation. The uniqueness and stability questions for the\nthree-dimensional problem were largely answered in the affirmative in the\n1980's using complex geometrical optics solutions, and this led further to a\ndirect reconstruction method relying on a non-physical scattering transform. In\nthis paper, the reconstruction problem is taken one step further towards\npractical applications by considering data contaminated by noise. Indeed, a\nregularization strategy for the three-dimensional Calder\\'on problem is\npresented based on a suitable and explicit truncation of the scattering\ntransform. This gives a certified, stable and direct reconstruction method that\nis robust to small perturbations of the data. Numerical tests on simulated\nnoisy data illustrate the feasibility and regularizing effect of the method,\nand suggest that the numerical implementation performs better than predicted by\ntheory.\n"}
{"abstract": "  We present PyTorch Geometric Temporal a deep learning framework combining\nstate-of-the-art machine learning algorithms for neural spatiotemporal signal\nprocessing. The main goal of the library is to make temporal geometric deep\nlearning available for researchers and machine learning practitioners in a\nunified easy-to-use framework. PyTorch Geometric Temporal was created with\nfoundations on existing libraries in the PyTorch eco-system, streamlined neural\nnetwork layer definitions, temporal snapshot generators for batching, and\nintegrated benchmark datasets. These features are illustrated with a\ntutorial-like case study. Experiments demonstrate the predictive performance of\nthe models implemented in the library on real world problems such as\nepidemiological forecasting, ridehail demand prediction and web-traffic\nmanagement. Our sensitivity analysis of runtime shows that the framework can\npotentially operate on web-scale datasets with rich temporal features and\nspatial structure.\n"}
{"abstract": "  In this paper, we use the homogeneous $q$-operators [J. Difference Equ. Appl.\n{\\bf20 } (2014), 837--851.] to derive Rogers formulas, extended Rogers formulas\nand Srivastava-Agarwal type bilinear generating functions for Cigler's\npolynomials [J. Difference Equ. Appl. {\\bf 24} (2018), 479--502.]. Finally, we\nalso derive two interesting transformation formulas between ${}_2\\Phi_1, \\,\n{}_2\\Phi_2$ and ${}_3\\Phi_2$.\n"}
{"abstract": "  The possibility of a non-gravitational interaction between the dark matter\nand the dark energy has been reconstructed using some recent datasets. The\ncrucial aspect is that the interaction is not parametrized at the outset, but\nrather reconstructed directly from the data in a non-parametric way. The Cosmic\nChronometer Hubble data, the Pantheon Supernova compilation of CANDELS and\nCLASH Multy-Cycle Treasury programs obtained by the HST, and the Baryon\nAcoustic Oscillation Hubble data have been considered in this work. The widely\naccepted Gaussian Process is used for the reconstruction. The results clearly\nindicate that a no interaction scenario is quite a possibility. Also, the\ninteraction, if any, is not really significant at the present epoch. The\ndirection of the flow of energy is clearly from the dark energy to the dark\nmatter which is consistent with the thermodynamic requirement.\n"}
{"abstract": "  We introduce Attention Free Transformer (AFT), an efficient variant of\nTransformers that eliminates the need for dot product self attention. In an AFT\nlayer, the key and value are first combined with a set of learned position\nbiases, the result of which is multiplied with the query in an element-wise\nfashion. This new operation has a memory complexity linear w.r.t. both the\ncontext size and the dimension of features, making it compatible to both large\ninput and model sizes. We also introduce AFT-local and AFT-conv, two model\nvariants that take advantage of the idea of locality and spatial weight sharing\nwhile maintaining global connectivity. We conduct extensive experiments on two\nautoregressive modeling tasks (CIFAR10 and Enwik8) as well as an image\nrecognition task (ImageNet-1K classification). We show that AFT demonstrates\ncompetitive performance on all the benchmarks, while providing excellent\nefficiency at the same time.\n"}
{"abstract": "  We develop a numerical method based on matrix product states for simulating\nquantum many-body systems at finite temperatures without importance sampling\nand evaluate its performance in spin 1/2 systems. Our method is an extension of\nthe random phase product state (RPPS) approach introduced recently [T. Iitaka,\narXiv:2006.14459]. We show that the original RPPS approach often gives\nunphysical values for thermodynamic quantities even in the Heisenberg chain. We\nfind that by adding the operation of Trotter gates to the RPPS, the sampling\nefficiency of the approach significantly increases and its results are\nconsistent with those of the purification approach. We also apply our method to\na frustrated spin 1/2 system to exemplify that it can simulate a system in\nwhich the purification approach fails.\n"}
{"abstract": "  We consider a distributionally robust second-order stochastic dominance\nconstrained optimization problem. We require the dominance constraints hold\nwith respect to all probability distributions in a Wasserstein ball centered at\nthe empirical distribution. We adopt the sample approximation approach to\ndevelop a linear programming formulation that provides a lower bound. We\npropose a novel split-and-dual decomposition framework which provides an upper\nbound. We establish quantitative convergency for both lower and upper\napproximations given some constraint qualification conditions. To efficiently\nsolve the non-convex upper bound problem, we use a sequential convex\napproximation algorithm. Numerical evidences on a portfolio selection problem\nvalid the convergency and effectiveness of the proposed two approximation\nmethods.\n"}
{"abstract": "  This work studies the spectral convergence of graph Laplacian to the\nLaplace-Beltrami operator when the graph affinity matrix is constructed from\n$N$ random samples on a $d$-dimensional manifold embedded in a possibly high\ndimensional space. By analyzing Dirichlet form convergence and constructing\ncandidate approximate eigenfunctions via convolution with manifold heat kernel,\nwe prove that, with Gaussian kernel, one can set the kernel bandwidth parameter\n$\\epsilon \\sim (\\log N/ N)^{1/(d/2+2)}$ such that the eigenvalue convergence\nrate is $N^{-1/(d/2+2)}$ and the eigenvector convergence in 2-norm has rate\n$N^{-1/(d+4)}$; When $\\epsilon \\sim N^{-1/(d/2+3)}$, both eigenvalue and\neigenvector rates are $N^{-1/(d/2+3)}$. These rates are up to a $\\log N$ factor\nand proved for finitely many low-lying eigenvalues. The result holds for\nun-normalized and random-walk graph Laplacians when data are uniformly sampled\non the manifold, as well as the density-corrected graph Laplacian (where the\naffinity matrix is normalized by the degree matrix from both sides) with\nnon-uniformly sampled data. As an intermediate result, we prove new point-wise\nand Dirichlet form convergence rates for the density-corrected graph Laplacian.\nNumerical results are provided to verify the theory.\n"}
{"abstract": "  We discuss the construction of a nuclear Energy Density Functional (EDF) from\nab initio calculations, and we advocate the need of a methodical approach that\nis free from ad hoc assumptions. The equations of state (EoS) of symmetric\nnuclear and pure neutron matter are computed using the chiral NNLO$_{\\rm sat}$\nand the phenomenological AV4$^\\prime$+UIX$_{c}$ Hamiltonians as inputs in the\nSelf-consistent Green's Function (SCGF) and Auxiliary Field Diffusion Monte\nCarlo (AFDMC) methods, respectively. We propose a convenient parametrization of\nthe EoS as a function of the Fermi momentum and fit it on the SCGF and AFDMC\ncalculations. We apply the ab initio-based EDF to carry out an analysis of the\nbinding energies and charge radii of different nuclei in the local density\napproximation. The NNLO$_{\\rm sat}$-based EDF produces encouraging results,\nwhereas the AV4$^\\prime$+UIX$_{c}$-based one is farther from experiment.\nPossible explanations of these different behaviors are suggested, and the\nimportance of gradient and spin-orbit terms is analyzed. Our work paves the way\nfor a practical and systematic way to merge ab initio nuclear theory and DFT,\nwhile at the same time it sheds light on some of the critical aspects of this\nprocedure.\n"}
{"abstract": "  Quantum mechanics has so far not been tested for mechanical objects at the\nscale of the Planck mass $\\sqrt{\\hbar c/ G} \\simeq 22\\,\\mu\\mathrm{g}$. We\npresent an experiment where a 1 mm quartz micropillar resonating at 3.6 MHz\nwith an effective mass of 30 $\\mu$g is cooled to 500 mK with a dilution\nrefrigerator, and further optomechanically sideband-cooled to an effective\ntemperature of 3 mK, corresponding to a mode thermal occupancy of 20 phonons.\nThis nearly 1000-fold increase in the mass of an optomechanical system with\nrespect to previous experiments near the quantum ground state opens new\nperspectives in the exploration of the quantum/classical border.\n"}
{"abstract": "  The hypothesis of self-organized criticality explains the existence of\nlong-range `space-time' correlations, observed inseparably in many natural\ndynamical systems. A simple link between these correlations is yet unclear,\nparticularly in fluctuations at `external drive' time scales. As an example, we\nconsider a class of sandpile models displaying non-trivial correlations.\nEmploying the scaling methods, we demonstrate the computation of spatial\ncorrelation by establishing a link between local and global temporal\ncorrelations.\n"}
{"abstract": "  Aspect Sentiment Triplet Extraction (ASTE) aims to extract triplets from\nsentences, where each triplet includes an entity, its associated sentiment, and\nthe opinion span explaining the reason for the sentiment. Most existing\nresearch addresses this problem in a multi-stage pipeline manner, which\nneglects the mutual information between such three elements and has the problem\nof error propagation. In this paper, we propose a Semantic and Syntactic\nEnhanced aspect Sentiment triplet Extraction model (S3E2) to fully exploit the\nsyntactic and semantic relationships between the triplet elements and jointly\nextract them. Specifically, we design a Graph-Sequence duel representation and\nmodeling paradigm for the task of ASTE: we represent the semantic and syntactic\nrelationships between word pairs in a sentence by graph and encode it by Graph\nNeural Networks (GNNs), as well as modeling the original sentence by LSTM to\npreserve the sequential information. Under this setting, we further apply a\nmore efficient inference strategy for the extraction of triplets. Extensive\nevaluations on four benchmark datasets show that S3E2 significantly outperforms\nexisting approaches, which proves our S3E2's superiority and flexibility in an\nend-to-end fashion.\n"}
{"abstract": "  The use of Z-pinch facilities makes it possible to carry out well-controlled\nand diagnosable laboratory experiments to study laboratory jets with scaling\nparameters close to those of the jets from young stars. This makes it possible\nto observe processes that are inaccessible to astronomical observations. Such\nexperiments are carried out at the PF-3 facility (\"plasma focus\", Kurchatov\nInstitute), in which the emitted plasma emission propagates along the drift\nchamber through the environment at a distance of one meter. The paper presents\nthe results of experiments with helium, in which a successive release of two\nejections was observed. An analysis of these results suggests that after the\npassage of the first supersonic ejection, a region with a low concentration is\nformed behind it, the so-called vacuum trace, due to which the subsequent\nejection practically does not experience environmental resistance and\npropagates being collimated. The numerical modeling of the propagation of two\nejections presented in the paper confirms this point of view. Using scaling\nlaws and appropriate numerical simulations of astrophysical ejections, it is\nshown that this effect can also be significant for the jets of young stars.\n"}
{"abstract": "  In this master thesis, we extend results from classical simple homotopy\ntheory to the world of stratified homotopy theory. To obtain a well-established\nframework to work in, we prove a series of results on two model categories of\nsimplicial sets and topological spaces, both equipped with a notion of\nfiltration, introduced by Sylvain Douteau in his PHD thesis. In particular, we\nshow that there is a fully faithful embedding of homotopy categories from the\n(finite) filtered simplicial into the filtered topological setting. We also use\nthese results to characterize the morphisms in the topological filtered\nhomotopy category between filtered spaces that are triangulable and stratified\nin some very general sense as stratified homotopy classes of stratum preserving\nmaps. Then, moving on to simple homotopy theory, we propose a class of\ncombinatorial elementary expansions for filtered simplicial sets that\ngeneralize both the classical ones as well as a class of stratified expansions\nsuggested by Banagl et al. We show that they fulfill a series of axioms,\nsuggested by Eckmann and Siebenmann for the construction of simple homotopy\ntheory. In doing so, we obtain a combinatorially defined Whitehead group and\ntorsion for filtered simplicial sets (and hence also for triangulable\nstratified spaces). We then begin a detailed investigation of their formal\nproperties, proving for example that a Mayer-Vietoris formula holds. Next, we\napply our results on stratified homotopy theory to obtain a more topological\ndescription of the Whitehead group and to generalize the Whitehead torsion to\narbitrary stratum preserving maps of triangulated filtered spaces. Finally, we\nprove that our simple homotopy theory is a generalization of the classical one,\nin the sense that it agrees with the latter when one considers trivially\nfiltered simplicial sets as CW-complexes.\n"}
{"abstract": "  The Thue number $\\pi(G)$ of a graph $G$ is the minimum number of colors\nneeded to color $G$ without creating a square on a path of $G$. For a graph\nclass $C$, $\\pi(C)$ is the supremum of $\\pi(G)$ over the graphs $G\\in C$. The\nThue number has been investigated for famous minor-closed classes:\n$\\pi(tree)=4$, $7\\le\\pi(outerplanar)\\le12$, and $11\\le\\pi(planar)\\le768$.\nFollowing a suggestion of Grytczuk, we consider the generalized parameters\n$\\pi_k(C)$ such that only squares of period at least $k$ must be avoided. Thus,\n$\\pi(C)=\\pi_1(C)$. We show that $\\pi_5(tree)=2$, $\\pi_2(tree)=3$, and\n$\\pi_k(planar)\\ge11$ for every fixed $k$.\n"}
{"abstract": "  Pedestrian trajectory prediction in urban scenarios is essential for\nautomated driving. This task is challenging because the behavior of pedestrians\nis influenced by both their own history paths and the interactions with others.\nPrevious research modeled these interactions with pooling mechanisms or\naggregating with hand-crafted attention weights. In this paper, we present the\nSocial Interaction-Weighted Spatio-Temporal Convolutional Neural Network\n(Social-IWSTCNN), which includes both the spatial and the temporal features. We\npropose a novel design, namely the Social Interaction Extractor, to learn the\nspatial and social interaction features of pedestrians. Most previous works\nused ETH and UCY datasets which include five scenes but do not cover urban\ntraffic scenarios extensively for training and evaluation. In this paper, we\nuse the recently released large-scale Waymo Open Dataset in urban traffic\nscenarios, which includes 374 urban training scenes and 76 urban testing scenes\nto analyze the performance of our proposed algorithm in comparison to the\nstate-of-the-art (SOTA) models. The results show that our algorithm outperforms\nSOTA algorithms such as Social-LSTM, Social-GAN, and Social-STGCNN on both\nAverage Displacement Error (ADE) and Final Displacement Error (FDE).\nFurthermore, our Social-IWSTCNN is 54.8 times faster in data pre-processing\nspeed, and 4.7 times faster in total test speed than the current best SOTA\nalgorithm Social-STGCNN.\n"}
{"abstract": "  Mobile phone operators have begun the roll-out of 5G networks, deploying\nmassive MIMO base stations. Commercial product ranges start with 16 independent\nradio chains connected to a large-scale antenna system to exploit both channel\nhardening and favourable propagation in order to obtain increased spectral\nefficiency. In this work, the cumulative distribution function describing the\ngain for large-scale antenna systems considering spatial and spectral diversity\nis evaluated empirically in terms of a fading margin and compared to an\nanalytical maximum diversity reference system. This allows for a simple\ninvestigation of the trade-off between deployment size and exploitation of\nchannel hardening. For the considered site-specific measurement data, little\nadditional diversity is harvested with systems larger than 32 antenna elements.\n"}
{"abstract": "  Transmission Electron Microscopy (TEM) is a powerful tool for imaging\nmaterial structure and characterizing material chemistry. Recent advances in\ndata collection technology for TEM have enabled high-volume and high-resolution\ndata collection at a microsecond frame rate. Taking advantage of these advances\nin data collection rates requires the development and application of data\nprocessing tools, including image analysis, feature extraction, and streaming\ndata processing techniques. In this paper, we highlight a few areas in\nmaterials science that have benefited from combining signal processing and\nstatistical analysis with data collection capabilities in TEM and present a\nfuture outlook on opportunities of integrating signal processing with automated\nTEM data analysis.\n"}
{"abstract": "  Under the global pandemic of COVID-19, building an automated framework that\nquantifies the severity of COVID-19 and localizes the relevant lesion on chest\nX-ray images has become increasingly important. Although pixel-level lesion\nseverity labels, e.g. lesion segmentation, can be the most excellent target to\nbuild a robust model, collecting enough data with such labels is difficult due\nto time and labor-intensive annotation tasks. Instead, array-based severity\nlabeling that assigns integer scores on six subdivisions of lungs can be an\nalternative choice enabling the quick labeling. Several groups proposed deep\nlearning algorithms that quantify the severity of COVID-19 using the\narray-based COVID-19 labels and localize the lesions with explainability maps.\nTo further improve the accuracy and interpretability, here we propose a novel\nVision Transformer tailored for both quantification of the severity and\nclinically applicable localization of the COVID-19 related lesions. Our model\nis trained in a weakly-supervised manner to generate the full probability maps\nfrom weak array-based labels. Furthermore, a novel progressive self-training\nmethod enables us to build a model with a small labeled dataset. The\nquantitative and qualitative analysis on the external testset demonstrates that\nour method shows comparable performance with radiologists for both tasks with\nstability in a real-world application.\n"}
{"abstract": "  For proper planning, operation, control, and protection of the power system,\nthe development of a suitable steady-state mathematical model of FACTS devices\nis a key issue. The Fast and Flexible Holomorphic Embedding (FFHE) method\nconverges faster and provides the flexibility to use any state as an initial\nguess. But to investigate the effect and ability of FACTS devices using FFHE\ntechnique, it is necessary to develop an embedded system for these devices.\nTherefore, this paper presents an FFHE-based embedded system for VSC-based\nFACTS controllers, such as SSSC and IPFC. The embedded system is also proposed\nfor their controlling modes.\n  The introduced embedded system is flexible which allows to take any state as\nan initial guess instead of fixed state, which leads towards the reduced\nruntime and decreases the required number of terms, as compared to standard\nHELM. To demonstrate the effectiveness and practicability, the proposed\nFFHE-based models of FACTS devices have been tested for several cases. Further,\nthe developed recursive formulas for power balance equations, devices' physical\nconstraints, and their controlling modes are thoroughly investigated and\nexamined. From several tests, it is found that the proposed FFHE-based FACTS\nmodels require less execution time and reduce the error at higher rate.\n"}
{"abstract": "  VR/AR technology is a key enabler for new ways of immersively experiencing\ncultural heritage artifacts based on their virtual counterparts obtained from a\ndigitization process. In this paper, we focus on enriching VR-based object\ninspection by additional haptic feedback, thereby creating tangible cultural\nheritage experiences. For this purpose, we present an approach for interactive\nand collaborative VR-based object inspection and annotation. Our system\nsupports high-quality 3D models with accurate reflectance characteristics while\nadditionally providing haptic feedback regarding the object shape features\nbased on a 3D printed replica. The digital object model in terms of a printable\nrepresentation of the geometry as well as reflectance characteristics are\nstored in a compact and streamable representation on a central server, which\nstreams the data to remotely connected users/clients. The latter can jointly\nperform an interactive inspection of the object in VR with additional haptic\nfeedback through the 3D printed replica. Evaluations regarding system\nperformance, visual quality of the considered models as well as insights from a\nuser study indicate an improved interaction, assessment and experience of the\nconsidered objects.\n"}
{"abstract": "  A Lagrangian experimental study of an axisymmetric turbulent water jet is\nperformed to investigate the highly anisotropic and inhomogeneous flow field.\nThe measurements were conducted within a Lagrangian exploration module, an\nicosahedron apparatus, to facilitate optical access of three cameras. The\nstereoscopic particle tracking velocimetry results in three component tracks of\nposition, velocity and acceleration of the tracer particles within the\nvertically-oriented jet with a Taylor-based Reynolds number $\\mathcal R_\\lambda\n\\simeq 230$. Analysis is performed at seven locations from 15 diameters up to\n45 diameters downstream. Eulerian analysis is first carried out to obtain\ncritical parameters of the jet and relevant scales, namely the Kolmogorov and\nlarge turnover (integral) scales as well as the energy dissipation rate.\nLagrangian statistical analysis is then performed on velocity components\nstationarised following methods inspired by Batchelor (\\textit{J. Fluid Mech.},\nvol. 3, 1957, pp. 67-80) which aim to extend stationary Lagrangian theory of\nturbulent diffusion by Taylor to the case of self-similar flows. The evolution\nof typical Lagrangian scaling parameters as a function of the developing jet is\nexplored and results show validation of the proposed stationarisation. The\nuniversal scaling constant $C_0$ (for the Lagrangian second-order structure\nfunction), as well as Eulerian and Lagrangian integral time scales are\ndiscussed in this context. $C_0$ is found to converge to a constant value (of\nthe order of $C_0 = 3$) within 30 diameters downstream of the nozzle. Finally,\nthe existence of finite particle size effects are investigated through\nconsideration of acceleration dependent quantities.\n"}
{"abstract": "  The article presents and evaluates a scalable algorithm for validating\nsolutions of linear programming problems on cluster computing systems. The main\nidea of the method is to generate a regular set of points (validation set) on a\nsmall-radius hypersphere centered at the point of the solution under\nvalidation. The objective function is calculated for each point of the\nvalidation set that belongs to the feasible region. If all these values are\nless than or equal to the value of the objective function at the point under\nvalidation, then this point is the correct solution. The parallel\nimplementation of the VaLiPro algorithm is performed in C++ through the\nparallel BSF-skeleton, which encapsulates all aspects related to the MPI-based\nparallelization of the program. We provide the results of large-scale\ncomputational experiments on a cluster computing system to study the\nscalability of the VaLiPro algorithm.\n"}
{"abstract": "  We present measurements of partial branching fractions of inclusive\nsemileptonic $B \\to X_u \\, \\ell^+\\, \\nu_{\\ell}$ decays using the full Belle\ndata set of 711 fb$^{-1}$ of integrated luminosity at the $\\Upsilon(4S)$\nresonance and for $\\ell = e, \\mu$. Inclusive semileptonic $B \\to X_u \\,\n\\ell^+\\, \\nu_{\\ell}$ decays are CKM suppressed and measurements are complicated\nby the large background from CKM-favored $B \\to X_c \\, \\ell^+\\, \\nu_{\\ell}$\ntransitions, which have a similar signature. Using machine learning techniques,\nwe reduce this and other backgrounds effectively, whilst retaining access to a\nlarge fraction of the $B \\to X_u \\, \\ell^+\\, \\nu_{\\ell}$ phase space and high\nsignal efficiency. We measure partial branching fractions in three phase-space\nregions covering about $31\\%$ to $86\\%$ of the accessible $B \\to X_u \\,\n\\ell^+\\, \\nu_{\\ell}$ phase space. The most inclusive measurement corresponds to\nthe phase space with lepton energies of $E_\\ell^B > 1 $ GeV, and we obtain\n$\\Delta \\mathcal{B}(B \\to X_u \\ell^+ \\, \\nu_\\ell) = \\left( 1.59 \\pm 0.07 \\pm\n0.16 \\right) \\times 10^{-3}$ from a two-dimensional fit of the hadronic mass\nspectrum and the four-momentum-transfer squared distribution, with the\nuncertainties denoting the statistical and systematic error. We find $\\left|\nV_{ub} \\right| = \\left( 4.10 \\pm 0.09 \\pm 0.22 \\pm 0.15 \\right) \\times 10^{-3}$\nfrom an average of four calculations for the partial decay rate with the third\nuncertainty denoting the average theory error. This value is higher but\ncompatible with the determination from exclusive semileptonic decays within 1.3\nstandard deviations. In addition, we report charmless inclusive partial\nbranching fractions separately for $B^+$ and $B^0$ mesons as well as for\nelectron and muon final states. No isospin breaking or lepton flavor\nuniversality violating effects are observed.\n"}
{"abstract": "  In this work, we are interested in the detailed analysis of complexity\naspects of both time and space that arises from the implementation of a quantum\nalgorithm on a quantum based hardware. In particular, some steps of the\nimplementation, as state preparation and readout processes, in most of the\ncases can surpass the complexity aspects of the algorithm itself. We present\nthe complexity involved in the full implementation of quantum algorithms for\nsolving linear systems of equations and linear system of differential\nequations, from state preparation to the number of measurements needed to\nobtain good statistics from the final states of the quantum system, in order to\nassess the overall complexity of the processes.\n"}
{"abstract": "  Reinforcement learning (RL) is empirically successful in complex nonlinear\nMarkov decision processes (MDPs) with continuous state spaces. By contrast, the\nmajority of theoretical RL literature requires the MDP to satisfy some form of\nlinear structure, in order to guarantee sample efficient RL. Such efforts\ntypically assume the transition dynamics or value function of the MDP are\ndescribed by linear functions of the state features. To resolve this\ndiscrepancy between theory and practice, we introduce the Effective Planning\nWindow (EPW) condition, a structural condition on MDPs that makes no linearity\nassumptions. We demonstrate that the EPW condition permits sample efficient RL,\nby providing an algorithm which provably solves MDPs satisfying this condition.\nOur algorithm requires minimal assumptions on the policy class, which can\ninclude multi-layer neural networks with nonlinear activation functions.\nNotably, the EPW condition is directly motivated by popular gaming benchmarks,\nand we show that many classic Atari games satisfy this condition. We\nadditionally show the necessity of conditions like EPW, by demonstrating that\nsimple MDPs with slight nonlinearities cannot be solved sample efficiently.\n"}
{"abstract": "  Speaker counting is the task of estimating the number of people that are\nsimultaneously speaking in an audio recording. For several audio processing\ntasks such as speaker diarization, separation, localization and tracking,\nknowing the number of speakers at each timestep is a prerequisite, or at least\nit can be a strong advantage, in addition to enabling a low latency processing.\nIn a previous work, we addressed the speaker counting problem with a\nmultichannel convolutional recurrent neural network which produces an\nestimation at a short-term frame resolution. In this work, we show that, for a\ngiven frame, there is an optimal position in the input sequence for best\nprediction accuracy. We empirically demonstrate the link between that optimal\nposition, the length of the input sequence and the size of the convolutional\nfilters.\n"}
{"abstract": "  We consider the p-ordered characteristic function and its Fourier transform,\nthe quasidistribution function, of squeezed coherent photons in a thermal state\nof photons and calculate the mean number and number variance of squeezed\ncoherent photons. All the properties calculated from the previous\ncharacteristic function can be used to calculate said properties for photons in\na thermal state of squeezed coherent photons via a parametric transformation.\nIn particular, one can obtain the mean number and number variance via the\nparametric transformation.\n"}
{"abstract": "  The spread of the coronavirus disease 2019 (COVID-19) caused by severe acute\nrespiratory syndrome coronavirus-2 (SARS-CoV-2) has become a global health\ncrisis. The binding affinity of SARS-CoV-2 (in particular the receptor binding\ndomain, RBD) to its receptor angiotensin converting enzyme 2 (ACE2) and the\nantibodies is of great importance in understanding the infectivity of COVID-19\nand evaluating the candidate therapeutic for COVID-19. In this work, we propose\na new method based on molecular mechanics/Poisson-Boltzmann surface area\n(MM/PBSA) to accurately calculate the free energy of SARS-CoV-2 RBD binding to\nACE2 and antibodies. The calculated binding free energy of SARS-CoV-2 RBD to\nACE2 is -13.3 kcal/mol, and that of SARS-CoV RBD to ACE2 is -11.4 kcal/mol,\nwhich agrees well with experimental result (-11.3 kcal/mol and -10.1 kcal/mol,\nrespectively). Moreover, we take two recently reported antibodies as the\nexample, and calculate the free energy of antibodies binding to SARS-CoV-2 RBD,\nwhich is also consistent with the experimental findings. Further, within the\nframework of the modified MM/PBSA, we determine the key residues and the main\ndriving forces for the SARS-CoV-2 RBD/CB6 interaction by the computational\nalanine scanning method. The present study offers a computationally efficient\nand numerically reliable method to evaluate the free energy of SARS-CoV-2\nbinding to other proteins, which may stimulate the development of the\ntherapeutics against the COVID-19 disease in real applications.\n"}
{"abstract": "  When the users in a MIMO broadcast channel experience different spatial\ntransmit correlation matrices, a class of gains is produced that is denoted\ntransmit correlation diversity. This idea was conceived for channels in which\ntransmit correlation matrices have mutually exclusive eigenspaces, allowing\nnon-interfering training and transmission. This paper broadens the scope of\ntransmit correlation diversity to the case of partially and fully overlapping\neigenspaces and introduces techniques to harvest these generalized gains. For\nthe two-user MIMO broadcast channel, we derive achievable degrees of freedom\n(DoF) and achievable rate regions with/without channel state information at the\nreceiver (CSIR). When CSIR is available, the proposed achievable DoF region is\ntight in some configurations of the number of receive antennas and the channel\ncorrelation ranks. We then extend the DoF results to the $K$-user case by\nanalyzing the interference graph that characterizes the overlapping structure\nof the eigenspaces. Our achievability results employ a combination of product\nsuperposition in the common part of the eigenspaces, and pre-beamforming (rate\nsplitting) to create multiple data streams in non-overlapping parts of the\neigenspaces. Massive MIMO is a natural example in which spatially correlated\nlink gains are likely to occur. We study the achievable downlink sum rate for a\nfrequency-division duplex massive MIMO system under transmit correlation\ndiversity.\n"}
{"abstract": "  Constructing good representations is critical for learning complex tasks in a\nsample efficient manner. In the context of meta-learning, representations can\nbe constructed from common patterns of previously seen tasks so that a future\ntask can be learned quickly. While recent works show the benefit of\nsubspace-based representations, such results are limited to linear-regression\ntasks. This work explores a more general class of nonlinear tasks with\napplications ranging from binary classification, generalized linear models and\nneural nets. We prove that subspace-based representations can be learned in a\nsample-efficient manner and provably benefit future tasks in terms of sample\ncomplexity. Numerical results verify the theoretical predictions in\nclassification and neural-network regression tasks.\n"}
{"abstract": "  One of the top remaining science challenges in astronomical optics is the\ndirect imaging and characterization of extrasolar planets and planetary\nsystems. Directly imaging exoplanets from ground-based observatories requires\ncombining high-order adaptive optics with a stellar coronagraph observing at\nwavelengths ranging from the visible to the mid-IR. A limiting factor in\nachieving the required contrast (planet-to-star intensity ratio) is\nquasi-static speckles, caused largely by non-common path aberrations (NCPA) in\nthe coronagraph. Starting with a realistic simulator of a telescope with an AO\nsystem and a coronagraph, this article provides simulations of several closely\nrelated millisecond regression models requiring inputs of the measured\nwavefronts and science camera images. The simplest regression model, called the\nnaive estimator, does not treat the noise and other sources of information loss\nin the WFS. The naive estimator provided a useful estimate of the NCPA of\n$\\sim$ 0.5 radian RMS, with an accuracy of $\\sim$ 0.06 radian RMS in one minute\nof simulated sky time on a magnitude 8 star. The bias-corrected estimator\ngeneralizes the regression model to account for the noise and information loss\nin the WFS. A simulation of the bias-corrected estimator with four minutes of\nsky time included an NCPA of $\\sim 0.05 \\,$ radian RMS and an extended\nexoplanet scene. The joint regression of the bias-corrected estimator\nsimultaneously achieved an NCPA estimate with an accuracy of $\\sim\n5\\times10^{-3} \\,$radian and contrast of $\\sim 10^{-5}$ on the exoplanet scene.\nIn addition, the estimate of the exoplanet image was completely free of the\nsubtraction artifacts that always plague differential imaging. The estimate of\nthe exoplanet image obtained by the joint regression was nearly identical to\nthe image obtained by subtraction of a perfectly known point-spread function.\n"}
{"abstract": "  Learning from streaming tasks leads a model to catastrophically erase unique\nexperiences it absorbs from previous episodes. While regularization techniques\nsuch as LWF, SI, EWC have proven themselves as an effective avenue to overcome\nthis issue by constraining important parameters of old tasks from changing when\naccepting new concepts, these approaches do not exploit common information of\neach task which can be shared to existing neurons. As a result, they do not\nscale well to large-scale problems since the parameter importance variables\nquickly explode. An Inter-Task Synaptic Mapping (ISYANA) is proposed here to\nunderpin knowledge retention for continual learning. ISYANA combines\ntask-to-neuron relationship as well as concept-to-concept relationship such\nthat it prevents a neuron to embrace distinct concepts while merely accepting\nrelevant concept. Numerical study in the benchmark continual learning problems\nhas been carried out followed by comparison against prominent continual\nlearning algorithms. ISYANA exhibits competitive performance compared to state\nof the arts. Codes of ISYANA is made available in\n\\url{https://github.com/ContinualAL/ISYANAKBS}.\n"}
{"abstract": "  Let $k$ be an infinite field of characteristic 0, and $X$ a del Pezzo surface\nof degree $d$ with at least one $k$-rational point. Various methods from\nalgebraic geometry and arithmetic statistics have shown the Zariski density of\nthe set $X(k)$ of $k$-rational points in $X$ for $d\\geq2$ (under an extra\ncondition for $d=2$), but fail to work in generality when the degree of $X$ is\n1, leaving a large class of del Pezzo surfaces for which the question of\ndensity of rational points is still open. In this paper, we prove the Zariski\ndensity of $X(k)$ when $X$ has degree 1 and is represented in the weighted\nprojective space $\\mathbb{P}(2,3,1,1)$ with coordinates $x,y,z,w$ by an\nequation of the form $y^2=x^3+az^6+bz^3w^3+cw^6$ for $a,b,c\\in k$ with $a,c$\nnon-zero, under the condition that the elliptic surface obtained by blowing up\nthe base point of the anticanonical linear system $|-K_X|$ contains a smooth\nfiber above a point in $\\mathbb{P}^1\\setminus\\{(1:0),(0:1)\\}$ with positive\nrank over $k$. When $k$ is of finite type over $\\mathbb{Q}$, this condition is\nsufficient and necessary.\n"}
{"abstract": "  We consider decentralized stochastic optimization problems where a network of\nagents each owns a local cost function cooperate to find a minimizer of the\nglobal-averaged cost. A widely studied decentralized algorithm for this problem\nis D-SGD in which each node applies a stochastic gradient descent step, then\naverages its estimate with its neighbors. D-SGD is attractive due to its\nefficient single-iteration communication and can achieve linear speedup in\nconvergence (in terms of the network size). However, D-SGD is very sensitive to\nthe network topology. For smooth objective functions, the transient stage\n(which measures how fast the algorithm can reach the linear speedup stage) of\nD-SGD is on the order of $O(n/(1-\\beta)^2)$ and $O(n^3/(1-\\beta)^4)$ for\nstrongly convex and generally convex cost functions, respectively, where\n$1-\\beta \\in (0,1)$ is a topology-dependent quantity that approaches $0$ for a\nlarge and sparse network. Hence, D-SGD suffers from slow convergence for large\nand sparse networks.\n  In this work, we study the non-asymptotic convergence property of the\nD$^2$/Exact-diffusion algorithm. By eliminating the influence of data\nheterogeneity between nodes, D$^2$/Exact-diffusion is shown to have an enhanced\ntransient stage that are on the order of $O(n/(1-\\beta))$ and\n$O(n^3/(1-\\beta)^2)$ for strongly convex and generally convex cost functions,\nrespectively. Moreover, we provide a lower bound of the transient stage of\nD-SGD under homogeneous data distributions, which coincides with the transient\nstage of D$^2$/Exact-diffusion in the strongly-convex setting. These results\nshow that removing the influence of data heterogeneity can ameliorate the\nnetwork topology dependence of D-SGD. Compared with existing decentralized\nalgorithms bounds, D$^2$/Exact-diffusion is least sensitive to network\ntopology.\n"}
{"abstract": "  We give an example of a semisimple symmetric space $G/H$ and an irreducible\nrepresentation of $G$ which has multiplicity 1 in $L^2(G/H)$ and multiplicity 2\nin $C^\\infty(G/H)$.\n"}
{"abstract": "  This paper studies video inpainting detection, which localizes an inpainted\nregion in a video both spatially and temporally. In particular, we introduce\nVIDNet, Video Inpainting Detection Network, which contains a two-stream\nencoder-decoder architecture with attention module. To reveal artifacts encoded\nin compression, VIDNet additionally takes in Error Level Analysis frames to\naugment RGB frames, producing multimodal features at different levels with an\nencoder. Exploring spatial and temporal relationships, these features are\nfurther decoded by a Convolutional LSTM to predict masks of inpainted regions.\nIn addition, when detecting whether a pixel is inpainted or not, we present a\nquad-directional local attention module that borrows information from its\nsurrounding pixels from four directions. Extensive experiments are conducted to\nvalidate our approach. We demonstrate, among other things, that VIDNet not only\noutperforms by clear margins alternative inpainting detection methods but also\ngeneralizes well on novel videos that are unseen during training.\n"}
{"abstract": "  Recent DFT calculations for Ba2CoO4 (BCO) and neutron scattering experiments\nfor SrRuO3 (SRO) have shown that oxygen develops a magnetic polarization.\nMoreover, DFT calculations for these compounds also unveiled unexpected nodes\nin the spin density, both along Co-O and Ru-O. For BCO, the overall\nantiferromagnetic state in its triangular lattice contains unusual zigzag spin\npatterns. Here, using simple model calculations supplemented by DFT we explain\nand extend these results. We predict that ligands that in principle should be\nspinless, such as O$^{2-}$, will develop a net polarization when they act as\nelectronic bridges between transition metal (TM) spins ferromagnetically\nordered, regardless of the number of intermediate ligand atoms. The reason is\nthe hybridization between atoms and mobility of the electrons with spins\nopposite to those of the closest TM atoms. Moreover, for bonds with TMs\nantiferromagnetically ordered, counterintuitively our calculations show that\noxygens should also have a net magnetization for the super-super-exchange cases\nTM-O-O-TM while for only one oxygen, as in Cu-O-Cu, the O-polarization should\ncancel. Our simple model also allows us to explain the presence of nodes based\non the antibonding character of the dominant singly occupied molecular orbitals\nalong the TM-O bonds. Finally, the zigzag pattern order becomes the ground\nstate mainly due to the influence of the Hubbard $U$, that creates the moments,\nin combination with a robust easy-axis anisotropy that suppresses the competing\n120$^{\\circ}$ degree antiferromagnetic order of a triangular lattice. Our\npredictions are generic and should be applicable to any other compound with\ncharacteristics similar to those of BCO and SRO.\n"}
{"abstract": "  Assuming that the free energy of a gas depends non-locally on the logarithm\nof its mass density, the body force in the resulting equation of motion\nconsists of the sum of density gradient terms. Truncating this series after the\nsecond term, Bohm's quantum potential and the Madelung equation are identically\nobtained, showing explicitly that some of the hypotheses that led to the\nformulation of quantum mechanics admit a classical interpretation based on\nnon-locality.\n"}
{"abstract": "  For research results to be comparable, it is important to have common\ndatasets for experimentation and evaluation. The size of such datasets,\nhowever, can be an obstacle to their use. The Vimeo Creative Commons Collection\n(V3C) is a video dataset designed to be representative of video content found\non the web, containing roughly 3800 hours of video in total, split into three\nshards. In this paper, we present insights on the second of these shards (V3C2)\nand discuss their implications for research areas, such as video retrieval, for\nwhich the dataset might be particularly useful. We also provide all the\nextracted data in order to simplify the use of the dataset.\n"}
{"abstract": "  In comparative effectiveness research, treated and control patients might\nhave a different start of follow up, e.g. when treatment is started later in\nthe disease trajectory. When the follow up period of control patients starts\nearlier in the disease trajectory than follow up of the treated, estimation of\nthe treatment effect suffers from different types of survival/survivorship\nbias. We study unobserved heterogeneity and illustrate how failing to account\nfor the time difference in recruitment between treated and controls leads to\nbias in the estimated treatment effect. We explore five methods to adjust for\nthis survivorship bias by including the time between diagnosis and treatment\ninitiation (wait time) in the analysis in different ways. We first conducted a\nsimulation study on whether these methods reduce survivorship bias, then\napplied our methods to fertility data on insemination. The five methods were:\nfirst, to regress on wait time as an additional covariate in the analysis\nmodel. Second, to match on wait time. Third, to select treated who started\ntreatment immediately. Fourth, to select controls who survived up to the median\ntime of treatment initiation. Fifth, to consider the wait time as the time of\nleft truncation. All methods reduced survivorship bias in the simulation. In\nthe application to fertility data, failing to adjust for survivorship bias led\nto a hazard ratio (HR) of 1.63 (95%CI: 1.13-1.65) whereas a HR of 2.15\n(1.71-2.69) was expected when using a time-varying covariate for treatment,\nwhich in this prospective cohort coincided with the left truncation approach.\nIn agreement with our simulations, the method in which adjustment corrected the\nHR upwards the most was left truncation. We conclude that the wait time between\ndiagnosis and treatment initiation should be taken into account in the analysis\nto respect the chronology of the disease and treatment trajectory.\n"}
{"abstract": "  An open challenge in physics is to expand the frontiers of the validity of\nquantum mechanics by evidencing nonclassicality of the centre of mass state of\na macroscopic object. Yet another equally important task is to evidence the\nessential nonclassicality of the interactions which act between macroscopic\nobjects. Here we introduce a new tool to meet these challenges: massive spatial\nqubits. In particular, we show that if two distinct localized states of a mass\nare used as the $|0\\rangle$ and $|1\\rangle$ states of a qubit, then we can\nmeasure this encoded spatial qubit with a high fidelity in the $\\sigma_x,\n\\sigma_y$ and $\\sigma_z$ bases simply by measuring its position after different\ndurations of free evolution. We show how this technique can be used to reveal\nan irreducible nonclassicality through a Bell-inequality violation arising from\nthe entanglement of the centre of mass of a nano-crystal with its spin in a\nStern-Gerlach setup. Secondly, we show how our methodology, in conjuction with\nthe Casimir interaction, offers a powerful method to create and certify\nnon-Gaussian entanglement between two neutral nano-objects. Fundamentally, the\ngeneration of such an entanglement provides an empirical means for\ndemonstrating an inherent quantumness of the Casimir interaction.\n"}
{"abstract": "  In this paper, we survey some recent results on statistical inference\n(parametric and nonparametric statistical estimation, hypotheses testing) about\nthe spectrum of stationary models with tapered data, as well as, a question\nconcerning robustness of inferences, carried out on a linear stationary process\ncontaminated by a small trend. We also discuss some question concerning tapered\nToeplitz matrices and operators, central limit theorems for tapered Toeplitz\ntype quadratic functionals, and tapered Fej\\'er-type kernels and singular\nintegrals. These are the main tools for obtaining the corresponding results,\nand also are of interest in themselves. The processes considered will be\ndiscrete-time and continuous-time Gaussian, linear or L\\'evy-driven linear\nprocesses with memory.\n"}
{"abstract": "  Machine learning-based malware detection systems are often vulnerable to\nevasion attacks, in which a malware developer manipulates their malicious\nsoftware such that it is misclassified as benign. Such software hides some\nproperties of the real class or adopts some properties of a different class by\napplying small perturbations. A special case of evasive malware hides by\nrepackaging a bonafide benign mobile app to contain malware in addition to the\noriginal functionality of the app, thus retaining most of the benign properties\nof the original app. We present a novel malware detection system based on\nmetamorphic testing principles that can detect such benign-seeming malware\napps. We apply metamorphic testing to the feature representation of the mobile\napp rather than to the app itself. That is, the source input is the original\nfeature vector for the app and the derived input is that vector with selected\nfeatures removed. If the app was originally classified benign and is indeed\nbenign, the output for the source and derived inputs should be the same class,\ni.e., benign, but if they differ, then the app is exposed as likely malware.\nMalware apps originally classified as malware should retain that classification\nsince only features prevalent in benign apps are removed. This approach enables\nthe machine learning model to classify repackaged malware with reasonably few\nfalse negatives and false positives. Our training pipeline is simpler than many\nexisting ML-based malware detection methods, as the network is trained\nend-to-end to learn appropriate features and perform classification. We\npre-trained our classifier model on 3 million apps collected from the\nwidely-used AndroZoo dataset. We perform an extensive study on other publicly\navailable datasets to show our approach's effectiveness in detecting repackaged\nmalware with more than94% accuracy, 0.98 precision, 0.95 recall, and 0.96 F1\nscore.\n"}
{"abstract": "  Consider the compressed sensing setup where the support $s^*$ of an\n$m$-sparse $d$-dimensional signal $x$ is to be recovered from $n$ linear\nmeasurements with a given algorithm. Suppose that the measurements are such\nthat the algorithm does not guarantee perfect support recovery and that true\nfeatures may be missed. Can they efficiently be retrieved? This paper addresses\nthis question through a simple error-correction module referred to as LiRE.\nLiRE takes as input an estimate $s_{in}$ of the true support $s^*$, and outputs\na refined support estimate $s_{out}$. In the noiseless measurement setup,\nsufficient conditions are established under which LiRE is guaranteed to recover\nthe entire support, that is $s_{out}$ contains $s^*$. These conditions imply,\nfor instance, that in the high-dimensional regime LiRE can correct a sublinear\nin $m$ number of errors made by Orthogonal Matching Pursuit (OMP). The\ncomputational complexity of LiRE is $O(mnd)$. Experimental results with random\nGaussian design matrices show that LiRE substantially reduces the number of\nmeasurements needed for perfect support recovery via Compressive Sampling\nMatching Pursuit, Basis Pursuit (BP), and OMP. Interestingly, adding LiRE to\nOMP yields a support recovery procedure that is more accurate and significantly\nfaster than BP. This observation carries over in the noisy measurement setup.\nFinally, as a standalone support recovery algorithm with a random\ninitialization, experiments show that LiRE's reconstruction performance lies\nbetween OMP and BP. These results suggest that LiRE may be used generically, on\ntop of any suboptimal baseline support recovery algorithm, to improve support\nrecovery or to operate with a smaller number of measurements, at the cost of a\nrelatively small computational overhead. Alternatively, LiRE may be used as a\nstandalone support recovery algorithm that is competitive with respect to OMP.\n"}
{"abstract": "  Clinical images are vital for diagnosing and monitoring skin diseases, and\ntheir importance has increased with the growing popularity of machine learning.\nLack of standards has stifled innovation in dermatological imaging, unlike\nother image-intensive specialties such as radiology. We investigate the\nmeta-requirements for utilizing the popular DICOM standard for metadata\nmanagement of images in dermatology. We propose practical design solutions and\nprovide open-source tools to integrate dermatologists' workflow with enterprise\nimaging systems. Using the tool, dermatologists can tag, search, organize and\nconvert clinical images to the DICOM format. We believe that our less\ndisruptive approach will improve the adoption of standards in the specialty.\n"}
{"abstract": "  Drug-target interaction (DTI) prediction plays a crucial role in drug\ndiscovery, and deep learning approaches have achieved state-of-the-art\nperformance in this field. We introduce an ensemble of deep learning models\n(EnsembleDLM) for DTI prediction. EnsembleDLM only uses the sequence\ninformation of chemical compounds and proteins, and it aggregates the\npredictions from multiple deep neural networks. This approach not only achieves\nstate-of-the-art performance in Davis and KIBA datasets but also reaches\ncutting-edge performance in the cross-domain applications across different\nbio-activity types and different protein classes. We also demonstrate that\nEnsembleDLM achieves a good performance (Pearson correlation coefficient and\nconcordance index > 0.8) in the new domain with approximately 50% transfer\nlearning data, i.e., the training set has twice as much data as the test set.\n"}
{"abstract": "  This article presents a new way to classify geodesics on a cone in the\nEuclidean 3-space. This proof is obtained considering our main result, which\nestablishes the necessary and sufficient conditions that a curve in space must\nsatisfy: to be rectifying curve or that its trace is contained in the sphere\ncentered at the origin.\n"}
{"abstract": "  Generative adversarial networks (GANs) are often billed as \"universal\ndistribution learners\", but precisely what distributions they can represent and\nlearn is still an open question. Heavy-tailed distributions are prevalent in\nmany different domains such as financial risk-assessment, physics, and\nepidemiology. We observe that existing GAN architectures do a poor job of\nmatching the asymptotic behavior of heavy-tailed distributions, a problem that\nwe show stems from their construction. Additionally, when faced with the\ninfinite moments and large distances between outlier points that are\ncharacteristic of heavy-tailed distributions, common loss functions produce\nunstable or near-zero gradients. We address these problems with the Pareto GAN.\nA Pareto GAN leverages extreme value theory and the functional properties of\nneural networks to learn a distribution that matches the asymptotic behavior of\nthe marginal distributions of the features. We identify issues with standard\nloss functions and propose the use of alternative metric spaces that enable\nstable and efficient learning. Finally, we evaluate our proposed approach on a\nvariety of heavy-tailed datasets.\n"}
{"abstract": "  Photonic bound states in the continuum (BICs) are special localized and\nnon-decaying states of a photonic system with a frequency embedded into the\nspectrum of scattered states. The simplest photonic structure displaying a\nsingle BIC is provided by two waveguides side-coupled to a common waveguide\nlattice, where the BIC is protected by symmetry. Here we consider such a simple\nphotonic structure and show that, breaking mirror symmetry and allowing for\nnon-nearest neighbor couplings, a doublet of quasi-BIC states can be sustained,\nenabling weakly-damped embedded Rabi oscillations of photons between the\nwaveguides.\n"}
{"abstract": "  This paper proposes a dynamic process of portfolio risk measurement to\naddress potential information loss. The proposed model takes advantage of\nfinancial big data to incorporate out-of-target-portfolio information that may\nbe missed when one considers the Value at Risk (VaR) measures only from certain\nassets of the portfolio. We investigate how the curse of dimensionality can be\novercome in the use of financial big data and discuss where and when benefits\noccur from a large number of assets. In this regard, the proposed approach is\nthe first to suggest the use of financial big data to improve the accuracy of\nrisk analysis. We compare the proposed model with benchmark approaches and\nempirically show that the use of financial big data improves small portfolio\nrisk analysis. Our findings are useful for portfolio managers and financial\nregulators, who may seek for an innovation to improve the accuracy of portfolio\nrisk estimation.\n"}
{"abstract": "  Axion-like particles (ALPs) appear in various new physics models with\nspontaneous global symmetry breaking. When the ALP mass is in the range of MeV\nto GeV, the cosmology and astrophysics bounds are so far quite weak. In this\nwork, we investigate such light ALPs through the ALP-strahlung production\nprocesses $pp \\to W^\\pm a, Z a$ with the sequential decay $a \\to \\gamma\\gamma$\nat the 14 TeV LHC with an integrated luminosity of 3000 fb$^{-1}$ (HL-LHC).\nBuilding on the concept of jet image which uses calorimeter towers as the\npixels of the image and measures a jet as an image, we investigate the\npotential of machine learning techniques based on convolutional neural network\n(CNN) to identify the highly boosted ALPs which decay to a pair of highly\ncollimated photons. With the CNN tagging algorithm, we demonstrate that our\napproach can extend current LHC sensitivity and probe the ALP mass range from\n0.3~GeV to 5~GeV. The obtained bounds are stronger than the existing limits on\nthe ALP-photon coupling.\n"}
{"abstract": "  We consider an exponentially growing population of cells undergoing mutations\nand ask about the effect of reproductive fluctuations (genetic drift) on its\nlong-term evolution. We combine first step analysis with the stochastic\ndynamics of a birth-death process to analytically calculate the probability\nthat the parent of a given genotype will go extinct. We compare the results\nwith numerical simulations and show how this turnover of genetic clones can be\nused to infer the rates underlying the population dynamics. Our work is\nmotivated by growing populations of tumour cells, the epidemic spread of\nviruses, and bacterial growth.\n"}
{"abstract": "  The MOS devices are the basic building block of any digital and analog\ncircuits, where silicon (Si) is the most commonly used material. The\nInternational Technology Roadmap Semiconductor (ITRS) report predicts the gate\nlength of the MOS device will shrink to 4.5 nm up to 2023, this may create\nsevere short channel effects (SCEs). Therefore, new channel materials have been\nrealized, which have shown their potential to maintain the proper balance\nbetween device performance and SECs. Among them, graphene has shown its strong\npresence as an alternative channel material in terms of its fascinating\nelectrical and mechanical properties. It has ultra-high carrier mobility\n(77,000 cm2V-1s-1) and saturation velocity, which makes it compatible with\nhigh-speed circuit applications. This review paper has a detailed report on the\nseveral analytical modeling approaches for graphene-based FET device, which\nincludes the drift-diffusion, gradual channel approximation, virtual source,\nand ballistic approaches. The device modeling plays an important role to\npredict the device performance and used to reveal the physics behind it. In\naddition, the compact model of the device will use in the development of\nelectronic design automation (EDA) tools, which are used for the circuits\nsimulation.\n"}
{"abstract": "  Quantum error correction is essential for robust quantum information\nprocessing with noisy devices. As bosonic quantum systems play a crucial role\nin quantum sensing, communication, and computation, it is important to design\nerror correction codes suitable for these systems against various different\ntypes of noises. While most efforts aim at protecting qubits encoded into the\ninfinite dimensional Hilbert space of a bosonic mode, [Phys. Rev. Lett. 125,\n080503 (2020)] proposed an error correction code to maintain the\ninfinite-dimensional-Hilbert-space nature of bosonic systems by encoding a\nsingle bosonic mode into multiple bosonic modes. Enabled by\nGottesman-Kitaev-Preskill states as ancilla, the code overcomes the no-go\ntheorem of Gaussian error correction. In this work, we generalize the error\ncorrection code to the scenario with general correlated and heterogeneous\nGaussian noises, including memory effects. We introduce Gaussian pre-processing\nand post-processing to convert the general noise model to an independent but\nheterogeneous collection of additive white Gaussian noise channels and then\napply concatenated codes in an optimized manner. To evaluate the performance,\nwe develop a theory framework to enable the efficient calculation of the noise\nstandard deviation after the error correction, despite the non-Gaussian nature\nof the codes. Our code provides the optimal scaling of the residue noise\nstandard deviation with the number of modes and can be widely applied to\ndistributed sensor-networks, network communication and composite quantum memory\nsystems.\n"}
{"abstract": "  An iteration sequence based on the BLUES (beyond linear use of equation\nsuperposition) function method is presented for calculating analytic\napproximants to solutions of nonlinear partial differential equations. This\nextends previous work using this method for nonlinear ordinary differential\nequations with an external source term. Now, the initial condition plays the\nrole of the source. The method is tested on three examples: a\nreaction-diffusion-convection equation, the porous medium equation with growth\nor decay, and the nonlinear Black-Scholes equation. A comparison is made with\nthree other methods: the Adomian decomposition method (ADM), the variational\niteration method (VIM), and the variational iteration method with Green\nfunction (GVIM). As a physical application, a deterministic differential\nequation is proposed for interface growth under shear, combining Burgers and\nKardar- Parisi-Zhang nonlinearities. Thermal noise is neglected. This model is\nstudied with Gaussian and space-periodic initial conditions. A detailed Fourier\nanalysis is performed and the analytic coefficients are compared with those of\nADM, VIM, GVIM, and standard perturbation theory. The BLUES method turns out to\nbe a worthwhile alternative to the other methods. The advantages that it offers\nensue from the freedom of choosing judiciously the linear part, with associated\nGreen function, and the residual containing the nonlinear part of the\ndifferential operator at hand.\n"}
{"abstract": "  Recently, intensive studies have revealed fascinating physics, such as charge\ndensity wave and superconducting states, in the newly synthesized\nkagome-lattice materials $A$V$_3$Sb$_5$ ($A$=K, Rb, Cs). Despite the rapid\nprogress, fundamental aspects like the magnetic properties and electronic\ncorrelations in these materials have not been clearly understood yet. Here,\nbased on the density functional theory plus the single-site dynamical\nmean-field theory calculations, we investigate the correlated electronic\nstructure and the magnetic properties of the KV$_3$Sb$_5$ family materials in\nthe normal state. We show that these materials are good metals with weak local\ncorrelations. The obtained Pauli-like paramagnetism and the absence of local\nmoments are consistent with recent experiment. We reveal that the band\ncrossings around the Fermi level form three groups of nodal lines protected by\nthe spacetime inversion symmetry, each carrying a quantized $\\pi$ Berry phase.\nOur result suggests that the local correlation strength in these materials\nappears to be too weak to generate unconventional superconductivity, and\nnon-local electronic correlation might be crucial in this kagome system.\n"}
{"abstract": "  On the dual space of \\textit{extended structure}, equations governing the\ncollective motion of two mutually interacting Lie-Poisson systems are derived.\nBy including a twisted 2-cocycle term, this novel construction is providing the\nmost general realization of (de)coupling of Lie-Poisson systems. A double cross\nsum (matched pair) of 2-cocycle extensions are constructed. The conditions are\ndetermined for this double cross sum to be a 2-cocycle extension by itself. On\nthe dual spaces, Lie-Poisson equations are computed. We complement the\ndiscussion by presenting a double cross sum of some symmetric brackets, such as\ndouble bracket, Cartan-Killing bracket, Casimir dissipation bracket, and\nHamilton dissipation bracket. Accordingly, the collective motion of two\nmutually interacting irreversible dynamics, as well as mutually interacting\nmetriplectic flows, are obtained. The theoretical results are illustrated in\nthree examples. As an infinite-dimensional physical model, decompositions of\nthe BBGKY hierarchy are presented. As finite-dimensional examples, the coupling\nof two Heisenberg algebras and coupling of two copies of $3D$ dynamics are\nstudied.\n"}
{"abstract": "  By associating frequencies to larger scales, we provide a simpler way to\nderive local uniformity of multiplicative functions on average from the results\nof Matom\\\"aki-Radziwill.\n"}
{"abstract": "  We propose PeriodNet, a non-autoregressive (non-AR) waveform generation model\nwith a new model structure for modeling periodic and aperiodic components in\nspeech waveforms. The non-AR waveform generation models can generate speech\nwaveforms parallelly and can be used as a speech vocoder by conditioning an\nacoustic feature. Since a speech waveform contains periodic and aperiodic\ncomponents, both components should be appropriately modeled to generate a\nhigh-quality speech waveform. However, it is difficult to decompose the\ncomponents from a natural speech waveform in advance. To address this issue, we\npropose a parallel model and a series model structure separating periodic and\naperiodic components. The features of our proposed models are that explicit\nperiodic and aperiodic signals are taken as input, and external\nperiodic/aperiodic decomposition is not needed in training. Experiments using a\nsinging voice corpus show that our proposed structure improves the naturalness\nof the generated waveform. We also show that the speech waveforms with a pitch\noutside of the training data range can be generated with more naturalness.\n"}
{"abstract": "  Rapid proliferation of hyperspectral imaging in scanning probe microscopies\ncreates unique opportunities to systematically capture and categorize higher\ndimensional datasets, toward new insights into electronic, mechanical and\nchemical properties of materials with nano- and atomic-scale resolution. Here\nwe demonstrate similarity learning for tunneling spectroscopy acquired on\nsuperconducting material (FeSe) with sparse density of imperfections (Fe\nvacancies). Popular methods for unsupervised learning and discrete\nrepresentation of the data in terms of clusters of characteristic behaviors\nwere found to produce inconsistencies with respect to capturing the location\nand tunneling characteristics of the vacancy sites. To this end, we applied a\nmore general, non-linear similarity learning. This approach was found to\noutperform several widely used methods for dimensionality reduction and produce\na clear differentiation of the type of tunneling spectra. In particular,\nsignificant spectral weight transfer likely associated with the electronic\nreconstruction by the vacancy sites, is systematically captured, as is the\nspatial extent of the vacancy region. Given that a great variety of electronic\nmaterials will exhibit similarly smooth variation of the spectral responses due\nto random or engineered inhomogeneities in their structure, we believe our\napproach will be useful for systematic analysis of hyperspectral imaging with\nminimal prior knowledge, as well as prospective comparison of experimental\nmeasurements to theoretical calculations with explicit consideration of\ndisorder.\n"}
{"abstract": "  Background: Advanced biological techniques have helped produce more\ninsightful findings on the genetic etiology of infertility that may lead to\nbetter management of the condition. This review provides an update on genes\npredisposing to syndromic and nonsyndromic infertility.\n  Main body: The review identified 65 genes linked with infertility and\ninfertility-related disorders. These genes regulate fertility. However,\nmutational loss of the functions of the genes predisposes to infertility.\nTwenty-three (23) genes representing 35% were linked with syndromic\ninfertility, while 42 genes (65%) cause nonsyndromic infertility. Of the 42\nnonsyndromic genes, 26 predispose to spermatogenic failure and sperm\nmorphological abnormalities, 11 cause ovarian failures, and 5 cause sex\nreversal and puberty delay. Overall, 31 genes (48%) predispose to male\ninfertility, 15 genes (23%) cause female infertility, and 19 genes (29%)\npredispose to both. The common feature of male infertility was spermatogenic\nfailure and sperm morphology abnormalities, while ovarian failure has been the\nmost frequently reported among infertile females. The mechanisms leading to\nthese pathologies are gene-specific, which, if targeted in the affected, may\nlead to improved treatment.\n  Conclusions: Mutational loss of the functions of some genes involved in the\ndevelopment and maintenance of fertility may predispose to syndromic or\nnonsyndromic infertility via gene-specific mechanisms. A treatment procedure\nthat targets the affected gene(s) in individuals expressing infertility may\nlead to improved treatment.\n"}
{"abstract": "  Semantic segmentation of breast cancer metastases in histopathological slides\nis a challenging task. In fact, significant variation in data characteristics\nof histopathology images (domain shift) make generalization of deep learning to\nunseen data difficult. Our goal is to address this challenge by using a\nconditional Fully Convolutional Network (co-FCN) whose output can be\nconditioned at run time, and which can improve its performance when a properly\nselected set of reference slides are used to condition the output. We adapted\nto our task a co-FCN originally applied to organs segmentation in volumetric\nmedical images and we trained it on the Whole Slide Images (WSIs) from three\nout of five medical centers present in the CAMELYON17 dataset. We tested the\nperformance of the network on the WSIs of the remaining centers. We also\ndeveloped an automated selection strategy for selecting the conditioning\nsubset, based on an unsupervised clustering process applied to a\ntarget-specific set of reference patches, followed by a selection policy that\nrelies on the cluster similarities with the input patch. We benchmarked our\nproposed method against a U-Net trained on the same dataset with no\nconditioning. The conditioned network shows better performance that the U-Net\non the WSIs with Isolated Tumor Cells and micro-metastases from the medical\ncenters used as test. Our contributions are an architecture which can be\napplied to the histopathology domain and an automated procedure for the\nselection of conditioning data.\n"}
{"abstract": "  We address the task of multi-view image-to-image translation for person image\ngeneration. The goal is to synthesize photo-realistic multi-view images with\npose-consistency across all views. Our proposed end-to-end framework is based\non a joint learning of multiple unpaired image-to-image translation models, one\nper camera viewpoint. The joint learning is imposed by constraints on the\nshared 3D human pose in order to encourage the 2D pose projections in all views\nto be consistent. Experimental results on the CMU-Panoptic dataset demonstrate\nthe effectiveness of the suggested framework in generating photo-realistic\nimages of persons with new poses that are more consistent across all views in\ncomparison to a standard Image-to-Image baseline. The code is available at:\nhttps://github.com/sony-si/MultiView-Img2Img\n"}
{"abstract": "  The first repeating fast radio burst source, FRB 121102, is observed to emit\nbursts periodically. We show that FRB 121102 can be interpreted as an\ninteracting neutron star binary system with an orbital period of 159 days. We\ndevelop a binary comb model by introducing an eccentricity in the orbit.\nBesides the original funnel mode of the binary comb model, which was applied to\nFRB 180916.J0158+65 by Ioka and Zhang 2020, we also identify two new modes of\nthe binary comb model, i.e. the tau-crossing mode and the inverse funnel mode,\nand apply them to interpret FRB 121102. These new developments expand the\napplicable parameter space, allowing the companion star to be a massive star, a\nmassive black hole, or a supermassive black hole, with the latter two having\nlarger parameter spaces. These models are also consistent with other\nobservations, such as the persistent bright radio counterpart associated with\nthe source. We also argue that the observed frequency dependence of the active\nwindow does not disfavor the binary comb model, in contrast to recent claims,\nand propose two possible scenarios to interpret the data.\n"}
{"abstract": "  In order to interact with the world, agents must be able to predict the\nresults of the world's dynamics. A natural approach to learn about these\ndynamics is through video prediction, as cameras are ubiquitous and powerful\nsensors. Direct pixel-to-pixel video prediction is difficult, does not take\nadvantage of known priors, and does not provide an easy interface to utilize\nthe learned dynamics. Object-centric video prediction offers a solution to\nthese problems by taking advantage of the simple prior that the world is made\nof objects and by providing a more natural interface for control. However,\nexisting object-centric video prediction pipelines require dense object\nannotations in training video sequences. In this work, we present\nObject-centric Prediction without Annotation (OPA), an object-centric video\nprediction method that takes advantage of priors from powerful computer vision\nmodels. We validate our method on a dataset comprised of video sequences of\nstacked objects falling, and demonstrate how to adapt a perception model in an\nenvironment through end-to-end video prediction training.\n"}
{"abstract": "  Context.Coherence in the characteristics of neighboring sources in 2D and 3D\nspace may suggest the existence of large-scale cosmic structures, which are\nuseful for cosmological studies. Numerous works have been conducted to detect\nsuch features in global scalesas well as in confined areas of the sky. However,\nresults are often contradictory and their interpretation remains controversial.\nAims.We investigate the potential alignment of parsec-scale radio jets in\nlocalized regions of the coordinates-redshift space. Methods.We use data from\nthe Astrogeo VLBI FITS image database to deduce jet directions of radio\nsources. We perform the search for statistical alignments between nearby\nsources and explore the impact of instrumental biases. Results.We unveil four\nregions for which the alignment between jet directions deviates from randomness\nat a significance level of more than 5 sigma and is unlikely due to\ninstrumental systematics. Intriguingly, their locations coincide with other\nknown large-scale cosmic structures and/or regions of alignments.\nConclusions.If the alignments found are the result of physical processes, the\ndiscovered regions may designate some of the largest structures known to date.\n"}
{"abstract": "  {\\alpha}-Sn with a diamond-type crystal structure provides an ideal avenue to\ninvestigate novel topological properties owing to its rich diagram of\ntopological phases and simple elemental material structure. Thus far, however,\nrealisation of high-quality {\\alpha}-Sn remains a challenge, which limits our\nunderstanding of its quantum transport properties and device applications.\nHere, we present epitaxial growth of {\\alpha}-Sn on InSb (001) with the highest\nquality thus far and reveal that it is a topological Dirac semimetal (TDS) by\nquantum transport investigations together with first-principles calculations.\nWe realise unprecedentedly high quantum mobilities of both the surface state\n(30000 cm^2/Vs), which is ten times higher than the previously reported values,\nand the bulk heavy-hole (HH) state (1700 cm^2/Vs), which has never been\nobtained experimentally. These excellent features allow us, for the first time,\nto quantitatively characterise the nontrivial interfacial and bulk band\nstructure of {\\alpha}-Sn via Shubnikov-de Haas oscillations at various\ntemperatures and under various magnetic field directions. These results reveal\nthe existence of a topological surface state (TSS) and a bulk HH band, both\nwith nontrivial phase shifts, indicating that the TDS phase of {\\alpha}-Sn is\nestablished. Furthermore, we demonstrate a crossover from the TDS to a\ntwo-dimensional topological insulator (2D-TI) and a subsequent phase transition\nto a trivial insulator when varying the thickness of {\\alpha}-Sn. Our work\nindicates that {\\alpha}-Sn is an excellent model system to study novel\ntopological phases and a prominent material candidate for topological devices.\n"}
{"abstract": "  Video-based person re-identification (Re-ID) aims to retrieve video sequences\nof the same person under non-overlapping cameras. Previous methods usually\nfocus on limited views, such as spatial, temporal or spatial-temporal view,\nwhich lack of the observations in different feature domains. To capture richer\nperceptions and extract more comprehensive video representations, in this paper\nwe propose a novel framework named Trigeminal Transformers (TMT) for\nvideo-based person Re-ID. More specifically, we design a trigeminal feature\nextractor to jointly transform raw video data into spatial, temporal and\nspatial-temporal domain. Besides, inspired by the great success of vision\ntransformer, we introduce the transformer structure for video-based person\nRe-ID. In our work, three self-view transformers are proposed to exploit the\nrelationships between local features for information enhancement in spatial,\ntemporal and spatial-temporal domains. Moreover, a cross-view transformer is\nproposed to aggregate the multi-view features for comprehensive video\nrepresentations. The experimental results indicate that our approach can\nachieve better performance than other state-of-the-art approaches on public\nRe-ID benchmarks. We will release the code for model reproduction.\n"}
{"abstract": "  Recent years have witnessed the rapid advance in neural machine translation\n(NMT), the core of which lies in the encoder-decoder architecture. Inspired by\nthe recent progress of large-scale pre-trained language models on machine\ntranslation in a limited scenario, we firstly demonstrate that a single\nlanguage model (LM4MT) can achieve comparable performance with strong\nencoder-decoder NMT models on standard machine translation benchmarks, using\nthe same training data and similar amount of model parameters. LM4MT can also\neasily utilize source-side texts as additional supervision. Though modeling the\nsource- and target-language texts with the same mechanism, LM4MT can provide\nunified representations for both source and target sentences, which can better\ntransfer knowledge across languages. Extensive experiments on pivot-based and\nzero-shot translation tasks show that LM4MT can outperform the encoder-decoder\nNMT model by a large margin.\n"}
{"abstract": "  A local artificial neural network (LANN) framework is developed for\nturbulence modeling. The Reynolds-averaged Navier-Stokes (RANS) unclosed terms\nare reconstructed by artificial neural network (ANN) based on the local\ncoordinate system which is orthogonal to the curved walls. We verify the\nproposed model for the flows over periodic hills. The correlation coefficients\nof the RANS unclosed terms predicted by the LANN model can be made larger than\n0.96 in an a priori analysis, and the relative error of the unclosed terms can\nbe made smaller than 18%. In an a posteriori analysis, detailed comparisons are\nmade on the results of RANS simulations using the LANN and Spalart-Allmaras\n(SA) models. It is shown that the LANN model performs better than the SA model\nin the prediction of the average velocity, wall-shear stress and average\npressure, which gives the results that are essentially indistinguishable from\nthe direct numerical simulation (DNS) data. The LANN model trained in low\nReynolds number Re = 2800 can be directly applied in the cases of high Reynolds\nnumbers Re = 5600, 10595, 19000, 37000 with accurate predictions. Furthermore,\nthe LANN model is verified for flows over periodic hills with varying slopes.\nThese results suggest that the LANN framework has a great potential to be\napplied to complex turbulent flows with curved walls.\n"}
{"abstract": "  Predicting the future motion of actors in a traffic scene is a crucial part\nof any autonomous driving system. Recent research in this area has focused on\ntrajectory prediction approaches that optimize standard trajectory error\nmetrics. In this work, we describe three important properties -- physical\nrealism guarantees, system maintainability, and sample efficiency -- which we\nbelieve are equally important for developing a self-driving system that can\noperate safely and practically in the real world. Furthermore, we introduce\nPTNet (PathTrackingNet), a novel approach for vehicle trajectory prediction\nthat is a hybrid of the classical pure pursuit path tracking algorithm and\nmodern graph-based neural networks. By combining a structured robotics\ntechnique with a flexible learning approach, we are able to produce a system\nthat not only achieves the same level of performance as other state-of-the-art\nmethods on traditional trajectory error metrics, but also provides strong\nguarantees about the physical realism of the predicted trajectories while\nrequiring half the amount of data. We believe focusing on this new class of\nhybrid approaches is an useful direction for developing and maintaining a\nsafety-critical autonomous driving system.\n"}
{"abstract": "  Surgical skill directly affects surgical procedure outcomes; thus, effective\ntraining is needed to ensure satisfactory results. Many objective assessment\nmetrics have been developed and some are widely used in surgical training\nsimulators. These objective metrics provide the trainee with descriptive\nfeedback about their performance however, often lack feedback on how to proceed\nto improve performance. The most effective training method is one that is\nintuitive, easy to understand, personalized to the user and provided in a\ntimely manner. We propose a framework to enable user-adaptive training using\nnear-real-time detection of performance, based on intuitive styles of surgical\nmovements (e.g., fluidity, smoothness, crispness, etc.), and propose a haptic\nfeedback framework to assist with correcting styles of movement. We evaluate\nthe ability of three types of force feedback (spring, damping, and spring plus\ndamping feedback), computed based on prior user positions, to improve different\nstylistic behaviors of the user during kinematically constrained reaching\nmovement tasks. The results indicate that four out of the six styles studied\nhere were statistically significantly improved (p<0.05) using spring guidance\nforce feedback and a significant reduction in task time was also found using\nspring feedback. The path straightness and targeting error in the task were\nother task performance metrics studied which were improved significantly using\nthe spring-damping feedback. This study presents a groundwork for adaptive\ntraining in robotic surgery based on near-real-time human-centric models of\nsurgical behavior.\n"}
{"abstract": "  In dense star clusters, such as globular and open clusters, dynamical\ninteractions between stars and black holes (BHs) can be extremely frequent,\nleading to various astrophysical transients. Close encounters between a star\nand a stellar mass BH make it possible for the star to be tidally disrupted by\nthe BH. Due to the relative low mass of the BH and the small cross section of\nthe tidal disruption event (TDE) for cases with high penetration, disruptions\ncaused by close encounters are usually partial disruptions. The existence of\nthe remnant stellar core and its non-negligible mass compared to the stellar\nmass BH alters the accretion process significantly. We study this problem with\nSPH simulations using the code {\\tt Phantom}, with the inclusion of radiation\npressure, which is important for small mass BHs. Additionally, we develop a\nnew, more general method of computing the fallback rate which does not rely on\nany approximation. Our study shows that the powerlaw slope of the fallback rate\nhas a strong dependence on the mass of the BH in the stellar mass BH regime.\nFurthermore, in this regime, self-gravity of the fallback stream and local\ninstabilities become more significant, and cause the disrupted material to\ncollapse into small clumps before returning to the BH. This results in an\nabrupt increase of the fallback rate, which can significantly deviate from a\npowerlaw. Our results will help in the identification of TDEs by stellar mass\nBHs in dense clusters.\n"}
{"abstract": "  We consider Gaussian measures $\\mu, \\tilde{\\mu}$ on a separable Hilbert\nspace, with fractional-order covariance operators $A^{-2\\beta}$ resp.\n$\\tilde{A}^{-2\\tilde{\\beta}}$, and derive necessary and sufficient conditions\non $A, \\tilde{A}$ and $\\beta, \\tilde{\\beta} > 0$ for I. equivalence of the\nmeasures $\\mu$ and $\\tilde{\\mu}$, and II. uniform asymptotic optimality of\nlinear predictions for $\\mu$ based on the misspecified measure $\\tilde{\\mu}$.\nThese results hold, e.g., for Gaussian processes on compact metric spaces. As\nan important special case, we consider the class of generalized\nWhittle-Mat\\'ern Gaussian random fields, where $A$ and $\\tilde{A}$ are elliptic\nsecond-order differential operators, formulated on a bounded Euclidean domain\n$\\mathcal{D}\\subset\\mathbb{R}^d$ and augmented with homogeneous Dirichlet\nboundary conditions. Our outcomes explain why the predictive performances of\nstationary and non-stationary models in spatial statistics often are\ncomparable, and provide a crucial first step in deriving consistency results\nfor parameter estimation of generalized Whittle-Mat\\'ern fields.\n"}
{"abstract": "  The primordial heavy or superheavy dark matter that could be created during\nthe reheating or preheating stage of the Universe can undergo QCD cascade decay\nprocess to produce leptons or $\\gamma$ as end products. Although these could be\nrare decays, the energy involved in such decay process can influence 21-cm\nsignal of hyperfine transition of neutral hydrogen during the reionization era.\nWe explore in this work, possible multimessenger signals of such heavy dark\nmatter decays. One of which could be the source of ultra high energy neutrino\n(of $\\sim$ PeV energy regime) signals at IceCube detector whereas the other\nsignal attributes to the cooling/heating of the baryons by the exchange of\nenergy involved in this decay process and its consequent influence on 21-cm\nsignal. The effect of evaporation of primordial black holes and baryon\nscattering with light cold dark matter are also included in relation to the\nevolution of the 21-cm signal temperature and their influence are also\ndiscussed.\n"}
{"abstract": "  Street network analysis holds appeal as a tool for the assessment of\npedestrian connectivity and its relation to the intensity and mix of land-uses;\nhowever, application within urban-design triggers a range of questions on\nimplementary specifics due to a variety of theories, methods, and\nconsiderations and it is not immediately clear which of these might be the most\napplicable at the pedestrian scale in relation to land-uses. It is,\nfurthermore, difficult to directly evaluate differing approaches on a\nlike-for-like basis without recourse to the underlying algorithms and\ncomputational workflows. To this end, the cityseer-api Python package is here\nused to develop, compute, and compare a range of centrality methods which are\nthen applied to the Ordnance Survey Open Roads dataset for Greater London. The\ncentralities are correlated to high-resolution land-use and mixed-used measures\ncomputed from the Ordnance Survey Points of Interest dataset for the same\npoints of analysis using a spatially precise methodology based on network\ndistances to premise locations. The comparisons show that mixed-uses correlate\nmore strongly against closeness than betweenness centralities; segmented\nmeasures tend to offer slightly stronger correlations than node-based\nequivalents; weighted variants offer correlations similar to unweighted\nversions, but with a greater degree of spatial specificity; simplest-path\nmethods confer an advantage in the context of local high-street mixed-uses but\nnot necessarily for district-wide mixed-uses or land-use accessibilities; and\nthe application of centrality measures to the dual network does not offer\ntangible benefits over the primal network.\n"}
{"abstract": "  Synthesis of face images from visual attributes is an important problem in\ncomputer vision and biometrics due to its applications in law enforcement and\nentertainment. Recent advances in deep generative networks have made it\npossible to synthesize high-quality face images from visual attributes.\nHowever, existing methods are specifically designed for generating unimodal\nimages (i.e visible faces) from attributes. In this paper, we propose a novel\ngenerative adversarial network that simultaneously synthesizes identity\npreserving multimodal face images (i.e. visible, sketch, thermal, etc.) from\nvisual attributes without requiring paired data in different domains for\ntraining the network. We introduce a novel generator with multimodal\nstretch-out modules to simultaneously synthesize multimodal face images.\nAdditionally, multimodal stretch-in modules are introduced in the discriminator\nwhich discriminates between real and fake images. Extensive experiments and\ncomparisons with several state-of-the-art methods are performed to verify the\neffectiveness of the proposed attribute-based multimodal synthesis method.\n"}
{"abstract": "  We present a case study of using a novel method to identify red supergiant\n(RSG) candidates in NGC 6822, based on their 1.6 $\\mu$m \"H-bump\". We collected\n32 bands of photometric data for NGC 6822 ranging from optical to MIR. By using\nthe theoretical spectra from MARCS, we demonstrate that there is a prominent\ndifference around 1.6 $\\mu$m (\"H-bump\") between low-surface-gravity (LSG) and\nhigh-surface-gravity (HSG) targets. Taking advantage of this feature, we\nidentify efficient color-color diagrams (CCDs) of rzH and rzK to separate HSG\nand LSG targets from crossmatching of optical and NIR data. Moreover, synthetic\nphotometry from ATLAS9 also give similar results. Further separating RSG\ncandidates from the rest of the LSG candidates is done by using semi-empirical\ncriteria on NIR CMDs and resulted in 323 RSG candidates. Meanwhile, the\nsimulation of foreground stars from Besan\\c{c}on models also indicates that our\nselection criteria is largely free from the contamination of Galactic giants.\nIn addition to the \"H-bump\" method, we also use the traditional BVR method as a\ncomparison and/or supplement, by applying a slightly aggressive cut to select\nas much as possible RSG candidates (358 targets). Furthermore, the Gaia\nastrometric solution is used to constrain the sample, where 181 and 193 targets\nwere selected from the \"H-bump\" and BVR method, respectively. The percentages\nof selected targets in both methods are similar as $\\sim$60\\%, indicating the\ncomparable accuracy of the two methods. In total, there are 234 RSG candidates\nafter combining targets from both methods with 140 ($\\sim$60\\%) of them in\ncommon. The final RSG candidates are in the expected locations on the MIR CMDs,\nwhile the spatial distribution is also coincident with the FUV-selected star\nformation regions, suggesting the selection is reasonable and reliable.\n"}
{"abstract": "  We compared high-contrast near-infrared images of the core of R136 taken by\nVLT/SPHERE, in two epochs separated by 3.06 years. For the first time we\nmonitored the dynamics of the detected sources in the core of R136 from a\nground-based telescope with adaptive optics. The aim of these observations was\nto search for High prOper Motion cAndidates (HOMAs) in the central region of\nR136 (r<6\") where it has been challenging for other instruments. Two bright\nsources (K<15mag and V<16mag) are located near R136a1 and R136c (massive WR\nstars) and have been identified as potential HOMAs. These sources have\nsignificantly shifted in the images with respect to the mean shift of all\nreliable detected sources and their neighbours, and six times their own\nastrometric errors. We calculate their proper motions to be 1.36\\pm0.22 mas/yr\n(321\\pm52 km/s) and 1.15\\pm0.11 mas/yr (273\\pm26 km/s). We discuss different\npossible scenarios to explain the magnitude of such extreme proper motions, and\nargue for the necessity to conduct future observations to conclude on the\nnature of HOMAs in the core of R136.\n"}
{"abstract": "  Most methods for dimensionality reduction are based on either tensor\nrepresentation or local geometry learning. However, the tensor-based methods\nseverely rely on the assumption of global and multilinear structures in\nhigh-dimensional data; and the manifold learning methods suffer from the\nout-of-sample problem. In this paper, bridging the tensor decomposition and\nmanifold learning, we propose a novel method, called Hypergraph Regularized\nNonnegative Tensor Factorization (HyperNTF). HyperNTF can preserve\nnonnegativity in tensor factorization, and uncover the higher-order\nrelationship among the nearest neighborhoods. Clustering analysis with HyperNTF\nhas low computation and storage costs. The experiments on four synthetic data\nshow a desirable property of hypergraph in uncovering the high-order\ncorrelation to unfold the curved manifolds. Moreover, the numerical experiments\non six real datasets suggest that HyperNTF robustly outperforms\nstate-of-the-art algorithms in clustering analysis.\n"}
{"abstract": "  A chiral photonic interface is a quantum system that has different\nprobabilities for emitting photons to the left and right. An on-chip compatible\nchiral interface is attractive for both fundamental studies of light-matter\ninteractions and applications to quantum information processing. We propose\nsuch a chiral interface based on superconducting circuits, which has wide\nbandwidth, rich tunability, and high tolerance to fabrication variations. The\nproposed interface consists of a core that uses Cooper-pair boxes (CPBs) to\nbreak time-reversal symmetry, and two superconducting transmons that connect\nthe core to a waveguide in the manner reminiscent of a ``giant atom.'' The\ntransmons form a state decoupled from the core, akin to dark states of atomic\nphysics, rendering the whole interface insensitive to the CPB charge noise. The\nproposed interface can be extended to realize a broadband fully passive on-chip\ncirculator for microwave photons.\n"}
{"abstract": "  Learning-from-demonstrations is an emerging paradigm to obtain effective\nrobot control policies for complex tasks via reinforcement learning without the\nneed to explicitly design reward functions. However, it is susceptible to\nimperfections in demonstrations and also raises concerns of safety and\ninterpretability in the learned control policies. To address these issues, we\nuse Signal Temporal Logic to evaluate and rank the quality of demonstrations.\nTemporal logic-based specifications allow us to create non-Markovian rewards,\nand also define interesting causal dependencies between tasks such as\nsequential task specifications. We validate our approach through experiments on\ndiscrete-world and OpenAI Gym environments, and show that our approach\noutperforms the state-of-the-art Maximum Causal Entropy Inverse Reinforcement\nLearning.\n"}
{"abstract": "  Apollonius of Perga, showed that for two given points $A,B$ in the Euclidean\nplane and a positive real number $k\\neq 1$, geometric locus of the points $X$\nthat satisfies the equation $|XA|=k|XB|$ is a circle. This circle is called\nApollonius circle. In this paper we generalize the definition of the Apollonius\ncircle for two given circles $\\Gamma_1,\\Gamma_2$ and we show that geometric\nlocus of the points $X$ with the ratio of the power with respect to the circles\n$\\Gamma_1,\\Gamma_2$ is constant, is also a circle. Using this we generalize the\ndefinition of Apollonius Circle, and generalize some results about Apollonius\nCircle.\n"}
{"abstract": "  Polymer-based hydrogels have been widely applied for chronic wound\ntherapeutics, due to their well-acclaimed wound exudate management capability.\nAt the same time, there is still an unmet clinical need for simple wound\ndiagnostic tools to assist clinical decision-making at the point of care and\ndeliver on the vision of patient-personalised wound management. To explore this\nchallenge, we present a one-step synthetic strategy to realise a\nredox-responsive, hyaluronic acid (HA)-based hydrogel that is sensitive to\nwound environment-related variations in glutathione (GSH) concentration. By\nselecting aminoethyl disulfide (AED) as a GSH-sensitive crosslinker and\nconsidering GSH concentration variations in active and non-self-healing wounds,\nwe investigated the impact of GSH-induced AED cleavage on hydrogel dimensions,\naiming to build GSH-size relationships for potential point-of-care wound\ndiagnosis. The hydrogel was also found to be non-cytotoxic and aided L929\nfibroblast growth and proliferation over seven days in vitro. Such a material\noffers a very low-cost tool for the visual detection of a target analyte that\nvaries dependent on the status of the cells and tissues (wound detection) and\nmay be further exploited as an implant for fibroblast growth and tissue\nregeneration (wound repair).\n"}
{"abstract": "  Face parsing aims to predict pixel-wise labels for facial components of a\ntarget face in an image. Existing approaches usually crop the target face from\nthe input image with respect to a bounding box calculated during\npre-processing, and thus can only parse inner facial Regions of\nInterest~(RoIs). Peripheral regions like hair are ignored and nearby faces that\nare partially included in the bounding box can cause distractions. Moreover,\nthese methods are only trained and evaluated on near-frontal portrait images\nand thus their performance for in-the-wild cases has been unexplored. To\naddress these issues, this paper makes three contributions. First, we introduce\niBugMask dataset for face parsing in the wild, which consists of 21,866\ntraining images and 1,000 testing images. The training images are obtained by\naugmenting an existing dataset with large face poses. The testing images are\nmanually annotated with $11$ facial regions and there are large variations in\nsizes, poses, expressions and background. Second, we propose RoI Tanh-polar\ntransform that warps the whole image to a Tanh-polar representation with a\nfixed ratio between the face area and the context, guided by the target\nbounding box. The new representation contains all information in the original\nimage, and allows for rotation equivariance in the convolutional neural\nnetworks~(CNNs). Third, we propose a hybrid residual representation learning\nblock, coined HybridBlock, that contains convolutional layers in both the\nTanh-polar space and the Tanh-Cartesian space, allowing for receptive fields of\ndifferent shapes in CNNs. Through extensive experiments, we show that the\nproposed method improves the state-of-the-art for face parsing in the wild and\ndoes not require facial landmarks for alignment.\n"}
{"abstract": "  Spin waves in magnetic insulators are low-damping signal carriers that could\nenable a new generation of spintronic devices. The excitation, control, and\ndetection of spin waves by metal electrodes is crucial for interfacing these\ndevices to electrical circuits. It is therefore important to understand\nmetal-induced damping of spin-wave transport, but characterizing this process\nrequires access to the underlying magnetic films. Here we show that spins in\ndiamond enable imaging of spin waves that propagate underneath metals in\nmagnetic insulators, and then use this capability to reveal a 100-fold increase\nin spin-wave damping. By analyzing spin-wave-induced currents in the metal, we\nderive an effective damping parameter that matches these observations well. We\nfurthermore detect buried scattering centers, highlighting the technique's\npower for assessing spintronic device quality. Our results open new avenues for\nstudying metal - spin-wave interaction and provide access to interfacial\nprocesses such as spin-wave injection via the spin-Hall effect.\n"}
{"abstract": "  Photodetector nonlinearity, the main limiting factor in terms of optical\npower in the detection chain, is corrected to improve the signal-to-noise ratio\nof a short-time measurement in dual-comb spectroscopy. An iterative correction\nalgorithm minimizing out-of-band spectral artifacts based on nonlinearity\ncorrection methods used in classical Fourier-transform spectrometers is\npresented. The exactitude of the nonlinearity correction is validated using a\nlow power linear measurement. Spectroscopic lines of H$^{12}$CN are provided\nand the error caused by the saturation of the detector is corrected yielding\nresiduals limited by the measurement noise.\n"}
{"abstract": "  The presence of hierarchy in many real-world networks is not yet fully\nexplained. Complex interaction networks are often coarse-grain models of vast\nmodular networks, where tightly connected subgraphs are agglomerated into nodes\nfor simplicity of representation and feasibility of analysis. The emergence of\nhierarchy in growing complex networks may stem from one particular property of\nthese ignored subgraphs: their graph conductance. Being a quantification of the\nmain bottleneck of flow on the subgraph, all such subgraphs will then have a\nspecific structural limitation associated with this scalar value. This supports\nthe consideration of heterogeneous degree restrictions on a randomly growing\nnetwork for which a hidden variable model is proposed based on the basic\n\\textit{rich-get-richer} scheme. Such node degree restrictions are drawn from\nvarious probability distributions, and it is shown that restriction generally\nleads to increased measures of hierarchy, while altering the tail of the degree\ndistribution. Thus, a general mechanism is provided whereby inherent\nlimitations lead to hierarchical self-organization.\n"}
{"abstract": "  In this article, we have described the Rao distance (due to C.R. Rao) and\nideas of conformal mappings on 3D objects with angle preservations. Three\npropositions help us to construct distances between the points within the 3D\nobjects in \\mathbb{R}^{3} and line integrals within complex planes. We\nhighlight the application of these concepts to virtual tourism.\n"}
{"abstract": "  The question of how a pure fiat currency is enforced and comes to have a\nnon-zero value has been much debated \\cite{10.2307/2077948}. What is less often\naddressed is, in the case where the enforcement is taken for granted and we ask\nwhat value (in terms of goods and services) the currency will end up taking.\nEstablishing a decentralised mechanism for price formation has proven a\nchallenge for economists: \"Since no decentralized out-of-equilibrium adjustment\nmechanism has been discovered, we currently have no acceptable dynamical model\nof the Walrasian system\" (Gintis 2006). In his paper, Gintis put forward a\nmodel for price discovery based on the evolution of the model's agents, i.e.\n\"poorly performing agents dying and being replaced by copies of the well\nperforming agents.\" It seems improbable that this mechanism is the driving\nforce behind price discovery in the real world. This paper proposes a more\nrealistic mechanism and presents results from a corresponding agent based\nmodel.\n"}
{"abstract": "  We consider the Newton stratification on Iwahori double cosets for a\nconnected reductive group. We prove the existence of Newton strata whose\nclosures cannot be expressed as a union of strata, and show how this is implied\nby the existence of non-equidimensional affine Deligne-Lusztig varieties. We\nalso give an explicit example for a group of type $A_4$.\n"}
{"abstract": "  We study Shimura (special) subvarieties in the moduli space $A_{p,D}$ of\ncomplex abelian varieties of dimension $p$ and polarization type $D$. These\nsubvarieties arise from families of covers compatible with a fixed group action\non the base curve such that the quotient of the base curve by the group is\nisomorphic to $\\mathbb{P}^1$. We give a criterion for the image of these\nfamilies under the Prym map to be a special subvariety and, using computer\nalgebra, obtain 210 Shimura subvarieties contained in the Prym locus.\n"}
{"abstract": "  The Meissner effect is one of the defining properties of superconductivity,\nwith a conventional superconductor completely repelling an external magnetic\nfield. In contrast to this diamagnetic behavior, odd-frequency superconducting\npairing has often been seen to produce a paramagnetic Meissner effect, which\ninstead makes the superconductor unstable due to the attraction of magnetic\nfield. In this work we study how both even- and odd-frequency superconducting\npairing contributes to the Meissner effect in a generic two-orbital\nsuperconductor with a tunable odd-frequency pairing component. By dividing the\ncontributions to the Meissner effect into intra- and inter-band processes, we\nfind that the odd-frequency pairing actually generates both dia- and\nparamagnetic Meissner responses, determined by the normal-state band structure.\nMore specifically, for materials with two electron-like (hole-like) low-energy\nbands we find that the odd-frequency inter-band contribution is paramagnetic\nbut nearly canceled by a diamagnetic odd-frequency intra-band contribution.\nCombined with a diamagnetic even-frequency contribution, such superconductors\nthus always display a large diamagnetic Meissner response to an external\nmagnetic field, even in the presence of large odd-frequency pairing. For\nmaterials with an inverted, or topological, band structure, we find the\nodd-frequency inter-band contribution to instead be diamagnetic and even the\ndominating contribution to the Meissner effect in the near-metallic regime.\nTaken together, our results show that odd-frequency pairing in multi-orbital\nsuperconductors does not generate a destabilizing paramagnetic Meissner effect\nand can even generate a diamagnetic response in topological materials.\n"}
{"abstract": "  Here I investigate some mathematical aspects of the maximum entropy theory of\necology (METE). In particular I address the geometrical structure of METE\nendowed by information geometry. As novel results, the macrostate entropy is\ncalculated analytically by the Legendre transformation of the log-normalizer in\nMETE. This result allows for the calculation of the metric terms in the\ninformation geometry arising from METE and, by consequence, the covariance\nmatrix between METE variables.\n"}
{"abstract": "  In this work, we study the convergence of the normalized Yamabe flow with\npositive Yamabe constant on a class of pseudo-manifolds that includes\nstratified spaces with iterated cone-edge metrics. We establish convergence\nunder a low energy condition. We also prove a concentration - compactness\ndichotomy, and investigate what the alternatives to convergence are. We end by\ninvestigating a non-convergent example due to Viaclovsky in more detail.\n"}
{"abstract": "  Significant progress has been made in boundary detection with the help of\nconvolutional neural networks. Recent boundary detection models not only focus\non real object boundary detection but also \"crisp\" boundaries (precisely\nlocalized along the object's contour). There are two methods to evaluate crisp\nboundary performance. One uses more strict tolerance to measure the distance\nbetween the ground truth and the detected contour. The other focuses on\nevaluating the contour map without any postprocessing. In this study, we\nanalyze both methods and conclude that both methods are two aspects of crisp\ncontour evaluation. Accordingly, we propose a novel network named deep\nrefinement network (DRNet) that stacks multiple refinement modules to achieve\nricher feature representation and a novel loss function, which combines\ncross-entropy and dice loss through effective adaptive fusion. Experimental\nresults demonstrated that we achieve state-of-the-art performance for several\navailable datasets.\n"}
{"abstract": "  We investigate the pure annihilation type radiative $B$ meson decays $B^0 \\to\n\\phi \\gamma$ and $B_s \\to \\rho^0(\\omega)\\gamma$ in the soft-collinear effective\ntheory. We consider three types of contributions to the decay amplitudes,\nincluding the direct annihilation topology, the contribution from the\nelectro-magnetic penguin operator and the contribution of the neutral vector\nmeson mixings. The numerical analysis shows that the decay amplitudes are\ndominated by the $\\omega-\\phi$ mixing effect in the $B^0 \\to \\phi\\gamma$ and\n$B_s \\to \\omega\\gamma$ modes. The corresponding decay branching ratios are\nenhanced about three orders of magnitudes relative to the pure annihilation\ntype contribution in these two decay channels. The decay rate of $B_s \\to\n\\rho^0\\gamma$ is much smaller than that of $B_s \\to \\omega\\gamma$ because of\nthe smaller $\\rho^0-\\phi$ mixing. The predicted branching ratios\n$B(B^{0}\\rightarrow\\phi\\gamma)=(3.99^{+1.67}_{-1.46}\n)\\times10^{-9},\\,B(B_s\\rightarrow\\omega\\gamma)=(2.01^{+0.81}_{-0.71}\n)\\times10^{-7}$ are to be tested by the Belle-II and LHC-b experiments.\n"}
{"abstract": "  MCBTE solves the linearized Boltzmann transport equation for phonons in three\ndimensions using a variance-reduced Monte Carlo solution approach. The\nalgorithm is suited for both transient and steady-state analysis of thermal\ntransport in structured materials with size features in the nanometer to\nhundreds of microns range. The code is portable and integrated with both\nfirst-principles density functional theory calculations and empirical relations\nfor the input of phonon frequency, group velocity, and mean free path required\nfor calculating the thermal properties. The program outputs space- and\ntime-resolved temperature and heat flux for the transient study. For the\nsteady-state simulations, the frequency-resolved contribution of phonons to\ntemperature and heat flux is written to the output files, thus allowing the\nstudy of cumulative thermal conductivity as a function of phonon frequency or\nmean free path. We provide several illustrative examples, including ballistic\nand quasi-ballistic thermal transport, the thermal conductivity of thin films\nand periodic nanostructures, to demonstrate the functionality and to benchmark\nour code against available theoretical/analytical/computational results from\nthe literature. Moreover, we parallelize the code using the Matlab Distributed\nComputing Server, providing near-linear scaling with the number of processors.\n"}
{"abstract": "  In the paper the relation of homology groups of the direct limit of cochain\ncomplexes and the inverse limit of homology groups of corresponding chain\ncomplexes is studied. As a corollary, the tautness property of a homology\ntheory is obtained. Moreover, for the defined exact homology theory the\ncontinuous property is obtained on the category of compact pairs.\n"}
{"abstract": "  Semiconductor superlattice multipliers have emerged as a nonlinear medium\ncapable to generate radiation in a wide frequency range. This property\nfacilitates the potential of sources suitable for sensing and spectroscopy\napplications. In this study, we further investigate the consequences on\nharmonic generation in a superlattice multiplier after excitation by an input\nsignal oscillating at different frequencies. Here we provide a rigorous\ndescription of our theoretical model including a semiclassical Boltzmann\napproach to nonlinear miniband transport and non-equilibrium Greens functions\ncalculations treating scattering processes under forward and reverse bias. To\nfully exploit the features of this radiation source, we focus on the effects of\nelastic scattering and systemat-ic imperfections in the superlattice structure\nwhich lead to asymmetric current flow.\n"}
{"abstract": "  We study the zeroes of a family of random holomorphic functions on the unit\ndisc, distinguished by their invariance with respect to the hyperbolic\ngeometry. Our main finding is a transition in the limiting behaviour of the\nnumber of zeroes in a large hyperbolic disc.\n  We find a normal distribution if the covariance decays faster than a certain\ncritical value. In contrast, in the regime of 'long-range dependence' when the\ncovariance decays slowly, the limiting distribution is skewed. For a closely\nrelated model we emphasise a link with Gaussian multiplicative chaos.\n"}
{"abstract": "  The stability of the solution to the equation $(*)\\dot{u} = F(t,u)+f(t)$,\n$t\\ge 0$, $u(0)=u_0$ is studied. Here $F(t,u)$ is a nonlinear operator in a\nBanach space $\\mathcal{X}$ for any fixed $t\\ge 0$ and $F(t,0)=0$, $\\forall t\\ge\n0$. We assume that the Fr\\'echet derivative of $F(t,u)$ is H\\\"{o}lder\ncontinuous of order $q>0$ with respect to $u$ for any fixed $t\\ge 0$, i.e.,\n$\\|F'_u(t,w) - F'_u(t,v)\\|\\le \\alpha(t)\\|v - w\\|^{q}$, $q>0$. We proved that\nthe equilibrium solution $v=0$ to the equation $\\dot{v} = F(t,v)$ is Lyapunov\nstable under persistently acting perturbation $f(t)$ if $\\sup_{t\\ge 0}\\int_0^t\n\\alpha(\\xi)\\|U(t,\\xi)\\|\\, d\\xi<\\infty$ and $\\sup_{t\\ge 0}\\|U(t)\\|<\\infty$.\nHere, $U(t):=U(t,0)$ and $U(t,\\xi)$ is the solution to the equation\n$\\frac{d}{dt}{U}(t,\\xi) = F'_u(t,0)U(t,\\xi)$, $t\\ge \\xi$, $U(\\xi,\\xi)=I$, where\n$I$ is the identity operator in $\\mathcal{X}$. Sufficient conditions for the\nsolution $u(t)$ to equation (*) to be bounded and for $\\lim_{t\\to\\infty}u(t) =\n0$ are proposed and justified. Stability of solutions to equations with\nunbounded operators in Hilbert spaces is also studied.\n"}
{"abstract": "  We prove $L^p\\to L^q$ estimates for local maximal operators associated with\ndilates of codimension two spheres in Heisenberg groups; these are sharp up to\ntwo endpoints. The results can be applied to improve currently known bounds on\nsparse domination for global maximal operators. We also consider lacunary\nvariants, and extensions to M\\'etivier groups.\n"}
{"abstract": "  In this paper, the role of secret key with finite rate is studied to enhance\nthe secrecy performance of the system when users are operating in interference\nlimited scenarios. To address this problem, a 2-user Gaussian Z-IC with secrecy\nconstraint at the receiver is considered. One of the fundamental problems here\nis how to use the secret key as a part of the encoding process. The paper\nproposes novel achievable schemes, where the schemes differ from each other\nbased on how the key has been used in the encoding process. The first\nachievable scheme uses one part of the key for one-time pad and remaining part\nof the key for wiretap coding. The encoding is performed such that the receiver\nexperiencing interference can decode some part of the interference without\nviolating the secrecy constraint. As a special case of the derived result, one\ncan obtain the secrecy rate region when the key is completely used for one-time\npad or part of the wiretap coding. The second scheme uses the shared key to\nencrypt the message using one-time pad and in contrast to the previous case no\ninterference is decoded at the receiver. The paper also derives an outer bound\non the sum rate and secrecy rate of the transmitter which causes interference.\nThe main novelty of deriving outer bound lies in the selection of side\ninformation provided to the receiver and using the secrecy constraint at the\nreceiver. The derived outer bounds are found to be tight depending on the\nchannel conditions and rate of the key. The scaling behaviour of key rate is\nalso explored for different schemes using the notion of secure GDOF. The\noptimality of different schemes is characterized for some specific cases. The\ndeveloped results show the importance of key rate splitting in enhancing the\nsecrecy performance of the system when users are operating under interference\nlimited environment.\n"}
{"abstract": "  We introduce a general framework for robust data-enabled predictive control\n(DeePC) for linear time-invariant (LTI) systems. The proposed framework enables\nus to obtain model-free optimal control for LTI systems based on noisy\ninput/output data. More specifically, robust DeePC solves a min-max\noptimization problem to compute the optimal control sequence that is resilient\nto all possible realizations of the uncertainties in the input/output data\nwithin a prescribed uncertainty set. We present computationally tractable\nreformulations of the min-max problem with various uncertainty sets.\nFurthermore, we show that even though an accurate prediction of the future\nbehavior is unattainable in practice due to inaccessibility of the perfect\ninput/output data, the obtained robust optimal control sequence provides\nperformance guarantees for the actually realized input/output cost. We further\nshow that the robust DeePC generalizes and robustifies the regularized DeePC\n(with quadratic regularization or 1-norm regularization) proposed in the\nliterature. Finally, we demonstrate the performance of the proposed robust\nDeePC algorithm on high-fidelity, nonlinear, and noisy simulations of a\ngrid-connected power converter system.\n"}
{"abstract": "  We present a new discretization for advection-diffusion problems with Robin\nboundary conditions on complex time-dependent domains. The method is based on\nsecond order cut cell finite volume methods introduced by Bochkov et al. to\ndiscretize the Laplace operator and Robin boundary condition. To overcome the\nsmall cell problem, we use a splitting scheme that uses a semi-Lagrangian\nmethod to treat advection. We demonstrate second order accuracy in the $L^1$,\n$L^2$, and $L^\\infty$ norms for both analytic test problems and numerical\nconvergence studies. We also demonstrate the ability of the scheme to handle\nconversion of one concentration field to another across a moving boundary.\n"}
{"abstract": "  Optimal Transport (OT) metrics allow for defining discrepancies between two\nprobability measures. Wasserstein distance is for longer the celebrated\nOT-distance frequently-used in the literature, which seeks probability\ndistributions to be supported on the $\\textit{same}$ metric space. Because of\nits high computational complexity, several approximate Wasserstein distances\nhave been proposed based on entropy regularization or on slicing, and\none-dimensional Wassserstein computation. In this paper, we propose a novel\nextension of Wasserstein distance to compare two incomparable distributions,\nthat hinges on the idea of $\\textit{distributional slicing}$, embeddings, and\non computing the closed-form Wassertein distance between the sliced\ndistributions. We provide a theoretical analysis of this new divergence, called\n$\\textit{heterogeneous Wasserstein discrepancy (HWD)}$, and we show that it\npreserves several interesting properties including rotation-invariance. We show\nthat the embeddings involved in HWD can be efficiently learned. Finally, we\nprovide a large set of experiments illustrating the behavior of HWD as a\ndivergence in the context of generative modeling and in query framework.\n"}
{"abstract": "  Using Levi-Civita's theory of ideal fluids, we derive the complex Korteweg-de\nVries (KdV) equation, describing the complex velocity of a shallow fluid up to\nfirst order. We use perturbation theory, and the long wave, slowly varying\nvelocity approximations for shallow water. The complex KdV equation describes\nthe nontrivial dynamics of all water particles from the surface to the bottom\nof the water layer. A crucial new step made in our work is the proof that a\nnatural consequence of the complex KdV theory is that the wave elevation is\ndescribed by the real KdV equation. The complex KdV approach in the theory of\nshallow fluids is thus more fundamental than the one based on the real KdV\nequation. We demonstrate how it allows direct calculation of the particle\ntrajectories at any point of the fluid, and that these results agree well with\nnumerical simulations of other authors.\n"}
{"abstract": "  A graph $G$ is semilinear of complexity $t$ if the vertices of $G$ are\nelements of $\\mathbb{R}^{d}$ for some $d\\in\\mathbb{Z}^{+}$, and the edges of\n$G$ are defined by the sign patterns of $t$ linear functions\n$f_1,\\dots,f_t:\\mathbb{R}^{d}\\times \\mathbb{R}^{d}\\rightarrow\\mathbb{R}$. We\nshow that semilinear graphs of constant complexity have very tame Ramsey\nproperties. More precisely, we prove that if $G$ is a semilinear graph of\ncomplexity $t$ which contains no clique of size $s$ and no independent set of\nsize $n$, then $G$ has at most $O_{s,t}(n)\\cdot(\\log n)^{O_t(1)}$ vertices. We\nalso show that the logarithmic term cannot be omitted. In particular, this\nimplies that if $G$ is a semilinear graph of constant complexity on $n$\nvertices, and $G$ contains no clique of size $s$, then $G$ can be properly\ncolored with $\\mbox{polylog}(n)$ colors. In the past 60 years, this coloring\nquestion was extensively studied for several special instances of semilinear\ngraphs, e.g. shift graphs, intersection and disjointness graphs of certain\ngeometric objects, and overlap graphs. Our main result provides a general upper\nbound on the chromatic number of all such, seemingly unrelated, graphs.\n  Furthermore, we consider the symmetric Ramsey problem for semilinear graphs\nas well. It is known that if there exists an intersection graph of $N$ boxes in\n$\\mathbb{R}^{d}$ (such graphs are semilinear of complexity $2d$) that contains\nno clique or independent set of size $n$, then $N=O_d(n^2(\\log n)^{d-1})$. That\nis, the exponent of $n$ does not grow with the dimension. We prove a result\nabout the symmetric Ramsey properties of semilinear graphs, which puts this\nphenomenon in a more general context.\n"}
{"abstract": "  This paper proves Lee-Lee's conjecture that establishes a coincidence between\nthe set of associated roots of non-self-intersecting curves in a $n$-punctured\ndisc and the set of real Schur roots of acyclic (valued) quivers with $n$\nvertices.\n"}
{"abstract": "  We explore a simple extension to the Standard Model containing two gauge\nsinglets: a Dirac fermion and a real pseudoscalar. In some regions of the\nparameter space both singlets are stable without the necessity of additional\nsymmetries, then becoming a possible two-component dark matter model. We study\nthe relic abundance production via freeze-out, with the latter determined by\nannihilations, conversions and semi-annihilations. Experimental constraints\nfrom invisible Higgs decay, dark matter relic abundance and direct/indirect\ndetection are studied. We found three viable regions of the parameter space,\nand the model is sensitive to indirect searches.\n"}
{"abstract": "  We give the first examples of collapsing Ricci limit spaces on which the\nHausdorff dimension of the singular set exceeds that of the regular set;\nmoreover, the Hausdorff dimension of these spaces can be non-integers. This\nanswers a question of Cheeger-Colding about collapsing Ricci limit spaces.\n"}
{"abstract": "  Proper function of a wastewater treatment plant (WWTP) relies on maintaining\na delicate balance between a multitude of competing microorganisms. Gaining a\ndetailed understanding of the complex network of interactions therein is\nessential to maximising not only current operational efficiencies, but also for\nthe effective design of new treatment technologies. Metagenomics offers an\ninsight into these dynamic systems through the analysis of the microbial DNA\nsequences present. Unique taxa are inferred through sequence clustering to form\noperational taxonomic units (OTUs), with per-taxa abundance estimates obtained\nfrom corresponding sequence counts. The data in this study comprise weekly OTU\ncounts from an activated sludge (AS) tank of a WWTP. To model the OTU dynamics,\nwe develop a Bayesian hierarchical vector autoregressive model, which is a\nlinear approximation to the commonly used generalised Lotka-Volterra (gLV)\nmodel. To tackle the high dimensionality and sparsity of the data, they are\nfirst clustered into 12 \"bins\" using a seasonal phase-based approach. The\nautoregressive coefficient matrix is assumed to be sparse, so we explore\ndifferent shrinkage priors by analysing simulated data sets before selecting\nthe regularised horseshoe prior for the biological application. We find that\nammonia and chemical oxygen demand have a positive relationship with several\nbins and pH has a positive relationship with one bin. These results are\nsupported by findings in the biological literature. We identify several\nnegative interactions, which suggests OTUs in different bins may be competing\nfor resources and that these relationships are complex. We also identify two\npositive interactions. Although simpler than a gLV model, our vector\nautoregression offers valuable insight into the microbial dynamics of the WWTP.\n"}
{"abstract": "  We investigate the potential added value of running three limited-area\nensemble systems (with the WRF, Meso-NH and MOLOCH models and a grid spacing of\napproximately 2.5 km) for two heavy-precipitation events in Italy. Such\nhigh-resolution ensembles include an explicit treatment of convective processes\nand dynamically downscale the ECMWF global data, which have a grid spacing of\napproximately 18 km. The predictions are verified against rain-gauge data and\ntheir accuracy is evaluated over that of the driving coarser-resolution\nensemble system. Furthermore, we compare the simulation speed (defined as the\nratio of simulation length to wall-clock time) of the three limited-area models\nto estimate the computational effort for operational convection-permitting\nensemble forecasting. We also study how the simulation wall-clock time scales\nwith increasing numbers of computing elements (from 36 to 1152 cores).\nObjective verification methods generally show that convection-permitting\nforecasts outperform global forecasts for both events, although precipitation\npeaks remain largely underestimated for one of the two events. Comparing\nsimulation speeds, the MOLOCH model is the fastest and the Meso-NH is the\nslowest one. The WRF model attains efficient scalability, whereas it is limited\nfor the Meso-NH and MOLOCH models when using more than 288 cores. We finally\ndemonstrate how the model simulation speed has the largest impact on a joint\nevaluation with the model performance because the accuracy of the three\nlimited-area ensembles, amplifying the forecasting capability of the global\npredictions, does not differ substantially.\n"}
{"abstract": "  We consider a fluid model including viscoelastic and viscoplastic effects.\nThe state is given by the fluid velocity and an internal stress tensor that is\ntransported along the flow with the Zaremba-Jaumann derivative. Moreover, the\nstress tensor obeys a nonlinear and nonsmooth dissipation law as well as stress\ndiffusion. We prove the existence of global-in-time weak solutions satisfying\nan energy inequality under general Dirichlet conditions for the velocity field\nand Neumann conditions for the stress tensor.\n"}
{"abstract": "  Food recommendation has become an important means to help guide users to\nadopt healthy dietary habits. Previous works on food recommendation either i)\nfail to consider users' explicit requirements, ii) ignore crucial health\nfactors (e.g., allergies and nutrition needs), or iii) do not utilize the rich\nfood knowledge for recommending healthy recipes. To address these limitations,\nwe propose a novel problem formulation for food recommendation, modeling this\ntask as constrained question answering over a large-scale food knowledge\nbase/graph (KBQA). Besides the requirements from the user query, personalized\nrequirements from the user's dietary preferences and health guidelines are\nhandled in a unified way as additional constraints to the QA system. To\nvalidate this idea, we create a QA style dataset for personalized food\nrecommendation based on a large-scale food knowledge graph and health\nguidelines. Furthermore, we propose a KBQA-based personalized food\nrecommendation framework which is equipped with novel techniques for handling\nnegations and numerical comparisons in the queries. Experimental results on the\nbenchmark show that our approach significantly outperforms non-personalized\ncounterparts (average 59.7% absolute improvement across various evaluation\nmetrics), and is able to recommend more relevant and healthier recipes.\n"}
{"abstract": "  One of the most distinctive hallmarks of many-body systems far from\nequilibrium is the spontaneous emergence of order under conditions where\ndisorder would be plausibly expected. Here, we report on the self-transition\nbetween ordered and disordered emulsions in divergent microfluidic channels,\ni.e. from monodisperse assemblies to heterogeneous polydisperse foam-like\nstructures, and back again to ordered ones. The transition is driven by the\nnonlinear competition between viscous dissipation and surface tension forces as\ncontrolled by the device geometry, particularly the aperture angle of the\ndivergent microfluidic channel. An unexpected route back to order is observed\nin the regime of large opening angles, where a trend towards increasing\ndisorder would be intuitively expected.\n"}
{"abstract": "  Dynamics of jammed packings of soft athermal disks under finite-rate shear\nare studied by means of molecular dynamics simulations. Particularly, we\ninvestigate the spatial structures of stress drop events, which are expected to\nprovide information about plasticity. Investigating the displacement fields\nduring stress drop events, we show that there are qualitatively different two\ntypes of events in the low rate limit: localized ones and extended ones. We\nfurther investigate the time evolution of events and clarify that both types of\nevents are due to oscillatory motion of the stress, which is unique for systems\nunder finite-rate shear. The difference between two types of events is the\nregime that events reside in: while localized events take place during plastic\nevents, extended ones occur in the elastic branch.\n"}
{"abstract": "  The Boltzmann distribution of an ideal gas is determined by the Hamiltonian\nfunction generating single particle dynamics. Systems with higher complexity\noften exhibit topological constraints, which are independent of the Hamiltonian\nand may affect the shape of the distribution function as well. Here, we study a\nfurther source of heterogeneity, the curvature of spacetime arising from the\ngeneral theory of relativity. The present construction relies on three\nassumptions: first, the statistical ensemble is made of particles obeying\ngeodesic equations, which define the phase space of the system. Next, the\nmetric coefficients are time-symmetric, implying that, if thermodynamic\nequilibrium is achieved, all physical observables are independent of coordinate\ntime. Finally, ergodicity is enforced with respect to proper time, so that\nambiguity in the choice of a time variable for the statistical ensemble is\nremoved. Under these hypothesis, we derive the distribution function of\nthermodynamic equilibrium, and verify that it reduces to the Boltzmann\ndistribution in the classical limit. We further show that spacetime curvature\naffects physical observables, even far from the source of the metric. Two\nexamples are analyzed: an ideal gas in Schwarzschild spacetime and a charged\ngas in Kerr-Newman spacetime. In the Schwarzschild case, conservation of\nmacroscopic constraints, such as angular momentum, combined with relativistic\ndistortion of the distribution function can produce configurations with\ndecreasing density and growing azimuthal rotation velocity far from the event\nhorizon of the central mass. In the Kerr-Newman case, it is found that kinetic\nenergy associated with azimuthal rotations is an increasing function of the\nradial coordinate, and it eventually approaches a constant value corresponding\nto classical equipartition, even though spatial particle density decreases.\n"}
{"abstract": "  There are many deterministic mathematical operations (e.g. compression,\nclipping, downsampling) that degrade speech quality considerably. In this paper\nwe introduce a neural network architecture, based on a modification of the\nDiffWave model, that aims to restore the original speech signal. DiffWave, a\nrecently published diffusion-based vocoder, has shown state-of-the-art\nsynthesized speech quality and relatively shorter waveform generation times,\nwith only a small set of parameters. We replace the mel-spectrum upsampler in\nDiffWave with a deep CNN upsampler, which is trained to alter the degraded\nspeech mel-spectrum to match that of the original speech. The model is trained\nusing the original speech waveform, but conditioned on the degraded speech\nmel-spectrum. Post-training, only the degraded mel-spectrum is used as input\nand the model generates an estimate of the original speech. Our model results\nin improved speech quality (original DiffWave model as baseline) on several\ndifferent experiments. These include improving the quality of speech degraded\nby LPC-10 compression, AMR-NB compression, and signal clipping. Compared to the\noriginal DiffWave architecture, our scheme achieves better performance on\nseveral objective perceptual metrics and in subjective comparisons.\nImprovements over baseline are further amplified in a out-of-corpus evaluation\nsetting.\n"}
{"abstract": "  Ganymede's atmosphere is produced by charged particle sputtering and\nsublimation of its icy surface. Previous far-ultraviolet observations of the\nO{\\small I\\,}1356-{\\AA} and O{\\small I\\,}1304-{\\AA} oxygen emissions were used\nto infer sputtered molecular oxygen (O$_2$) as an atmospheric constituent, but\nan expected sublimated water (H$_2$O) component remained undetected. Here we\npresent an analysis of high-sensitivity spectra and spectral images acquired by\nthe Hubble Space Telescope revealing H$_2$O in Ganymede's atmosphere. The\nrelative intensity of the oxygen emissions requires contributions from\ndissociative excitation of water vapor, indicating that H$_2$O is more abundant\nthan O$_2$ around the sub-solar point. Away from the sub-solar region, the\nemissions are consistent with a pure O$_2$ atmosphere. Eclipse observations\nconstrain atomic oxygen to be at least two orders of magnitude less abundant\nthan these other species. The higher H$_2$O/O$_2$ ratio above the warmer\ntrailing hemisphere compared to the colder leading hemisphere, the spatial\nconcentration to the sub-solar region, and the estimated abundance of\n$\\sim$10$^{15}$ H$_2$O/cm$^{2}$ are consistent with sublimation of the icy\nsurface as source.\n"}
{"abstract": "  Conventional works that learn grasping affordance from demonstrations need to\nexplicitly predict grasping configurations, such as gripper approaching angles\nor grasping preshapes. Classic motion planners could then sample trajectories\nby using such predicted configurations. In this work, our goal is instead to\nfill the gap between affordance discovery and affordance-based policy learning\nby integrating the two objectives in an end-to-end imitation learning framework\nbased on deep neural networks. From a psychological perspective, there is a\nclose association between attention and affordance. Therefore, with an\nend-to-end neural network, we propose to learn affordance cues as visual\nattention that serves as a useful indicating signal of how a demonstrator\naccomplishes tasks, instead of explicitly modeling affordances. To achieve\nthis, we propose a contrastive learning framework that consists of a Siamese\nencoder and a trajectory decoder. We further introduce a coupled triplet loss\nto encourage the discovered affordance cues to be more affordance-relevant. Our\nexperimental results demonstrate that our model with the coupled triplet loss\nachieves the highest grasping success rate in a simulated robot environment.\nOur project website can be accessed at\nhttps://sites.google.com/asu.edu/affordance-aware-imitation/project.\n"}
{"abstract": "  The friction f is the property of wall-bounded flows that sets the pumping\ncost of a pipeline, the draining capacity of a river, and other variables of\npractical relevance. For highly turbulent rough-walled pipe flows, f depends\nsolely on the roughness length scale r, and the f -- r relation may be\nexpressed by the Strickler empirical scaling f $\\propto$ r$^{\\frac{1}{3}}$.\nHere, we show experimentally that for soap film flows that are the\ntwo-dimensional (2D) equivalent of highly turbulent rough-walled pipe flows, f\n$\\propto$ r and the f -- r relation is not the same in 2D as in 3D. Our\nfindings are beyond the purview of the standard theory of friction but\nconsistent with a competing theory in which f is linked to the turbulent\nspectrum via the spectral exponent $\\alpha$: In 3D, $\\alpha$ = 5/3 and the\ntheory yields f $\\propto$ r$^{\\frac{1}{3}}$; in 2D, $\\alpha$ = 3 and the theory\nyields f $\\propto$ r.\n"}
{"abstract": "  In the last few years, the notion of optimal polynomial approximant has\nappeared in the mathematics literature in connection with Hilbert spaces of\nanalytic functions of one or more variables. In the 70s, researchers in\nengineering and applied mathematics introduced least-squares inverses in the\ncontext of digital filters in signal processing. It turns out that in the Hardy\nspace $H^2$ these objects are identical. This paper is a survey of known\nresults about optimal polynomial approximants. In particular, we will examine\ntheir connections with orthogonal polynomials and reproducing kernels in\nweighted spaces and digital filter design. We will also describe what is known\nabout the zeros of optimal polynomial approximants, their rates of decay, and\nconvergence results. Throughout the paper, we state many open questions that\nmay be of interest.\n"}
{"abstract": "  To address the issue of inaccurate distributions in practical stochastic\nsystems, a minimax linear-quadratic control method is proposed using the\nWasserstein metric. Our method aims to construct a control policy that is\nrobust against errors in an empirical distribution of underlying uncertainty,\nby adopting an adversary that selects the worst-case distribution. The opponent\nreceives a Wasserstein penalty proportional to the amount of deviation from the\nempirical distribution. A closed-form expression of the finite-horizon optimal\npolicy pair is derived using a Riccati equation. The result is then extended to\nthe infinite-horizon average cost setting by identifying conditions under which\nthe Riccati recursion converges to the unique positive semi-definite solution\nto an algebraic Riccati equation. Our method is shown to possess several\nsalient features including closed-loop stability, and an out-of-sample\nperformance guarantee. We also discuss how to optimize the penalty parameter\nfor enhancing the distributional robustness of our control policy. Last but not\nleast, a theoretical connection to the classical $H_\\infty$-method is\nidentified from the perspective of distributional robustness.\n"}
{"abstract": "  We consider the differential entropy of probability measures absolutely\ncontinuous with respect to a given $\\sigma$-finite reference measure on an\narbitrary measurable space. We state the asymptotic equipartition property in\nthis general case; the result is part of the folklore but our presentation is\nto some extent novel. Then we study a general framework under which such\nentropies satisfy a chain rule: disintegrations of measures. We give an\nasymptotic interpretation for conditional entropies in this case. Finally, we\napply our result to Haar measures in canonical relation.\n"}
{"abstract": "  We consider parabolic equations on bounded smooth open sets $\\Om\\subset \\R^N$\n($N\\ge 1$) with mixed Dirichlet type boundary-exterior conditions associated\nwith the elliptic operator $\\mathscr{L} \\coloneqq - \\Delta + (-\\Delta)^{s}$\n($0<s<1$). Firstly, we prove several well-posedness and regularity results of\nthe associated elliptic and parabolic problems with smooth, and then with\nsingular boundary-exterior data. Secondly, we show the existence of optimal\nsolutions of associated optimal control problems, and we characterize the\noptimality conditions. This is the first time that such topics have been\npresented and studied in a unified fashion for mixed local-nonlocal PDEs with\nsingular data.\n"}
{"abstract": "  We present the timing results of out-of-eclipse observations of Centaurus X-3\nspanning half a binary orbit, performed on 12-13 December, 2016 with the Large\nArea X-ray Proportional Counter (LAXPC) on-board AstroSat. The pulse profile\nwas confirmed to exhibit a prominent pulse peak with a secondary inter-pulse.\nThe systemic spin period of the pulsar was found to be $4.80188 \\pm 0.000085$ s\nin agreement with its spin up trend. The spin up timescale seems to have\nincreased to $7709 \\pm 58$ yr that points to negative torque effects in the\ninner accretion disk. We also report the derived values of projected semi-major\naxis and orbital velocity of the neutron star.\n"}
{"abstract": "  Fully homomorphic encryption (FHE) is an encryption scheme which enables\ncomputation on encrypted data without revealing the underlying data. While\nthere have been many advances in the field of FHE, developing programs using\nFHE still requires expertise in cryptography. In this white paper, we present a\nfully homomorphic encryption transpiler that allows developers to convert\nhigh-level code (e.g., C++) that works on unencrypted data into high-level code\nthat operates on encrypted data. Thus, our transpiler makes transformations\npossible on encrypted data.\n  Our transpiler builds on Google's open-source XLS SDK\n(https://github.com/google/xls) and uses an off-the-shelf FHE library, TFHE\n(https://tfhe.github.io/tfhe/), to perform low-level FHE operations. The\ntranspiler design is modular, which means the underlying FHE library as well as\nthe high-level input and output languages can vary. This modularity will help\naccelerate FHE research by providing an easy way to compare arbitrary programs\nin different FHE schemes side-by-side. We hope this lays the groundwork for\neventual easy adoption of FHE by software developers. As a proof-of-concept, we\nare releasing an experimental transpiler\n(https://github.com/google/fully-homomorphic-encryption/tree/main/transpiler)\nas open-source software.\n"}
{"abstract": "  We investigate macroscopic and microscopic quantum correlations in a heavy\nquark-gluon medium holographically. In order to represent heavy quarks, we\nconsider the string cloud geometry which involves uniformly distributed open\nstrings. In this system, the macroscopic quantum correlation is described by\nthe entanglement entropy and increases when the gluon's excitation and the\nquark's density increase. We also investigate a microscopic two-point\ncorrelation of local operators in the quark-gluon medium. Our calculation shows\nthat the two-point function of local operators decreases as the background\nentanglement entropy increases. This is because the strong quantum correlation\nof the background medium gives rise to a strong screening effect. Furthermore,\nwe survey the effect of the quark-gluon medium on the quark-antiquark pair by\nemploying the holographic Wilson loop.\n"}
{"abstract": "  Deep generative models are becoming widely used across science and industry\nfor a variety of purposes. A common challenge is achieving a precise implicit\nor explicit representation of the data probability density. Recent proposals\nhave suggested using classifier weights to refine the learned density of deep\ngenerative models. We extend this idea to all types of generative models and\nshow how latent space refinement via iterated generative modeling can\ncircumvent topological obstructions and improve precision. This methodology\nalso applies to cases were the target model is non-differentiable and has many\ninternal latent dimensions which must be marginalized over before refinement.\nWe demonstrate our Latent Space Refinement (LaSeR) protocol on a variety of\nexamples, focusing on the combinations of Normalizing Flows and Generative\nAdversarial Networks.\n"}
{"abstract": "  The hot nucleus $^{171}$Yb is investigated by the covariant density\nfunctional theory with the PC-PK1 effective interaction. The thermodynamic\nquantities are evaluated with the canonical ensemble theory. The pairing\ncorrelations is treated by the shell-model-like approach, in which the particle\nnumbers are conserved strictly and in which the blocking effect is handled\nexactly. An S-shaped heat capacity versus temperature of $^{171}$Yb appears. It\nhas been studied in terms of the blocking effect, the single-particle levels,\nthe pairing gap, and defined seniority components, and compared to the heat\ncapacity of $^{172}$Yb. The pairing transition from the superfluid state to the\nnormal state can result in the S-shaped heat capacity of $^{172}$Yb where the\none-pair-broken and two-pair-broken states dominate, while the single-particle\nlevel structure near the Fermi surface is associated with the S-shaped heat\ncapacity of $^{171}$Yb. For odd-A nuclei, although the one-pair-broken and\ntwo-pair-broken states still contribute, the pairing gap and the pairing\ntransition is relatively weak. The S-shaped heat capacity could be affected due\nto the blocking of the single-particle level near the Fermi surface.\n"}
{"abstract": "  We consider extrapolation of the Arnoldi algorithm to accelerate computation\nof the dominant eigenvalue/eigenvector pair. The basic algorithm uses sequences\nof Krylov vectors to form a small eigenproblem which is solved exactly. The two\ndominant eigenvectors output from consecutive Arnoldi steps are then recombined\nto form an extrapolated iterate, and this accelerated iterate is used to\nrestart the next Arnoldi process. We present numerical results testing the\nalgorithm on a variety of cases and find on most examples it substantially\nimproves the performance of restarted Arnoldi. The extrapolation is a simple\npost-processing step which has minimal computational cost.\n"}
{"abstract": "  We define and study the existence of very stable Higgs bundles on Riemann\nsurfaces, how it implies a precise formula for the multiplicity of the very\nstable components of the global nilpotent cone and its relationship to mirror\nsymmetry. The main ingredients are the Bialynicki-Birula theory of ${\\mathbb\nC}^*$-actions on semiprojective varieties, ${\\mathbb C}^*$ characters of\nindices of ${\\mathbb C}^*$-equivariant coherent sheaves, Hecke transformation\nfor Higgs bundles, relative Fourier-Mukai transform along the Hitchin\nfibration, hyperholomorphic structures on universal bundles and cominuscule\nHiggs bundles.\n"}
{"abstract": "  In this paper we carry out numerical analysis for a family of simplified gas\ntransport models with hydrate formation and dissociation in subsurface, in\nequilibrium and non-equilibrium conditions. These models are adequate for\nsimulation of hydrate phase change at basin and at shorter time scales, but the\nanalysis does not account directly for the related effects of evolving\nhydraulic properties.\n  To our knowledge this is the first analysis of such a model. It is carried\nout for the transport steps while keeping the pressure solution fixed. We frame\nthe transport model as conservation law with a non-smooth space-dependent flux\nfunction; the kinetic model approximates this equilibrium. We prove weak\nstability of the upwind scheme applied to the regularized conservation law. We\nillustrate the model, confirm convergence with numerical simulations, and\nillustrate its use for some relevant equilibrium and non-equilibrium scenarios.\n"}
{"abstract": "  Emotions are usually evoked in humans by images. Recently, extensive research\nefforts have been dedicated to understanding the emotions of images. In this\nchapter, we aim to introduce image emotion analysis (IEA) from a computational\nperspective with the focus on summarizing recent advances and suggesting future\ndirections. We begin with commonly used emotion representation models from\npsychology. We then define the key computational problems that the researchers\nhave been trying to solve and provide supervised frameworks that are generally\nused for different IEA tasks. After the introduction of major challenges in\nIEA, we present some representative methods on emotion feature extraction,\nsupervised classifier learning, and domain adaptation. Furthermore, we\nintroduce available datasets for evaluation and summarize some main results.\nFinally, we discuss some open questions and future directions that researchers\ncan pursue.\n"}
{"abstract": "  Many multi-agent systems in nature are comprised of agents that interact\nwith, and respond to, the dynamics of their environment. In this paper, we\napproach the study of such agent-environment interactions through the study of\npassively compliant vehicles coupled to their environment via simple\nnonholonomic constraints. We first consider a single passively compliant\nChaplygin beanie atop a platform having translational compliance, introduce the\nreduced equations for the system using the notion of nonholonomic momentum, and\nprovide proof for its stability under arbitrary deformations of the elastic\nelement modeling its compliance. We then direct our focus to results concerning\nthe frequency response and control of passive Chaplygin beanies under actuation\nof the platform, discuss rich dynamical features arising from periodic\nactuation, and develop rules by which control can be exerted to collect and\ndisperse multiple passive vehicles. We then discuss how the latter of these\nresults clarifies the extent to which stable behavior can be excited in the\nsystem through exogenous control.\n"}
{"abstract": "  Many experiments have shown large flow enhancement ratios (up to 10^5) in\ncarbon nanotubes (CNT) with diameters larger than 5nm. However, molecular\ndynamics simulations have never replicated these results maintaining a\nthree-order-of-magnitude gap with measurements. Our study provides a generic\nmodel of nanofluidics for continuum slip flow (diameter>3nm) that fills this\nsignificant gap and sheds light on its origin. Compared to 140 literature\ncases, the model explains the entire range of experimental flow enhancements by\nchanges of nanotube diameters and finite variations of interfacial energies.\nDespite large variations of flow enhancement ratios spanning 5 orders of\nmagnitude in experimental results, the ratio between these data and\ncorresponding model predictions approaches unity for the majority of\nexperiments. The role of viscous entrance effects is discussed. The model\nprovides insight into puzzling observations such as differences of CNTs and\nboron nitride nanotubes, the slip on low-contact-angle surfaces and massive\nfunctionalization effects. This study could advance our understanding of\nnano-scale transport mechanisms and aid the design of tailored nanomembranes.\n"}
{"abstract": "  Using the short-cadence data (1-min interval) of the Kepler space telescope,\nwe conducted a statistical analysis for the characteristic time of stellar\nflares on Sun-like stars (SLS). Akin to solar flares, stellar flares show rise\nand decay light-curve profiles, which reflect the two distinct phases (rise\nphase and decay phase) of the flare process. We derived characteristic times of\nthe two phases for the stellar flares of SLS, resulting in a median rise time\nof about 5.9 min and a median decay time of 22.6 min. It is found that both the\nrise time and the decay time of the stellar flares follow a lognormal\ndistribution. The peak positions of the lognormal distributions for flare rise\ntime and decay time are 3.5 min and 14.8 min, respectively. These time values\nfor stellar flares are similar to the time-scale of solar flares, which\nsupports the idea that stellar flares and solar flares have the same physical\nmechanism. The statistical results obtained in this work for SLS can be a\nbenchmark of flare characteristic times when comparing with other types of\nstars.\n"}
{"abstract": "  Optical imaging of genetically encoded calcium indicators is a powerful tool\nto record the activity of a large number of neurons simultaneously over a long\nperiod of time from freely behaving animals. However, determining the exact\ntime at which a neuron spikes and estimating the underlying firing rate from\ncalcium fluorescence data remains challenging, especially for calcium imaging\ndata obtained from a longitudinal study. We propose a multi-trial time-varying\n$\\ell_0$ penalized method to jointly detect spikes and estimate firing rates by\nrobustly integrating evolving neural dynamics across trials. Our simulation\nstudy shows that the proposed method performs well in both spike detection and\nfiring rate estimation. We demonstrate the usefulness of our method on calcium\nfluorescence trace data from two studies, with the first study showing\ndifferential firing rate functions between two behaviors and the second study\nshowing evolving firing rate function across trials due to learning.\n"}
{"abstract": "  The performance of a binary classifier (\"predictor\") depends heavily upon the\ncontext (\"workflow\") in which it operates. Classic measures of predictor\nperformance do not reflect the realized utility of predictors unless certain\nimplied workflow assumptions are met. Failure to meet these implied assumptions\nresults in suboptimal classifier implementations and a mismatch between\npredicted or assessed performance and the actual performance obtained in\nreal-world deployments. The mismatch commonly arises when multiple predictions\ncan be made for the same event, the event is relatively rare, and redundant\ntrue positive predictions for the same event add little value, e.g., a system\nthat makes a prediction each minute, repeatedly issuing interruptive alarms for\na predicted event that may never occur.\n  We explain why classic metrics do not correctly represent the performance of\npredictors in such contexts, and introduce an improved performance assessment\ntechnique (\"u-metrics\") using utility functions to score each prediction.\nU-metrics explicitly account for variability in prediction utility arising from\ntemporal relationships. Compared to traditional performance measures, u-metrics\nmore accurately reflect the real-world benefits and costs of a predictor\noperating in a workflow context. The difference can be significant.\n  We also describe the use of \"snoozing,\" a method whereby predictions are\nsuppressed for a period of time, commonly improving predictor performance by\nreducing false positives while retaining the capture of events. Snoozing is\nespecially useful when predictors generate interruptive alerts, as so often\nhappens in clinical practice. Utility-based performance metrics correctly\npredict and track the performance benefits of snoozing, whereas traditional\nperformance metrics do not.\n"}
{"abstract": "  Collecting a large number of reliable training images annotated by multiple\nland-cover class labels in the framework of multi-label classification is\ntime-consuming and costly in remote sensing (RS). To address this problem,\npublicly available thematic products are often used for annotating RS images\nwith zero-labeling-cost. However, such an approach may result in constructing a\ntraining set with noisy multi-labels, distorting the learning process. To\naddress this problem, we propose a Consensual Collaborative Multi-Label\nLearning (CCML) method. The proposed CCML identifies, ranks and corrects\ntraining images with noisy multi-labels through four main modules: 1)\ndiscrepancy module; 2) group lasso module; 3) flipping module; and 4) swap\nmodule. The discrepancy module ensures that the two networks learn diverse\nfeatures, while obtaining the same predictions. The group lasso module detects\nthe potentially noisy labels by estimating the label uncertainty based on the\naggregation of two collaborative networks. The flipping module corrects the\nidentified noisy labels, whereas the swap module exchanges the ranking\ninformation between the two networks. The experimental results confirm the\nsuccess of the proposed CCML under high (synthetically added) multi-label noise\nrates. The code of the proposed method is publicly available at\nhttps://noisy-labels-in-rs.org\n"}
{"abstract": "  Automatic sequences are not suitable sequences for cryptographic applications\nsince both their subword complexity and their expansion complexity are small,\nand their correlation measure of order 2 is large. These sequences are highly\npredictable despite having a large maximum order complexity. However, recent\nresults show that polynomial subsequences of automatic sequences, such as the\nThue--Morse sequence, are better candidates for pseudorandom sequences. A\nnatural generalization of automatic sequences are morphic sequences, given by a\nfixed point of a prolongeable morphism that is not necessarily uniform. In this\npaper we prove a lower bound for the maximum order complexity of the sum of\ndigits function in Zeckendorf base which is an example of a morphic sequence.\nWe also prove that the polynomial subsequences of this sequence keep large\nmaximum order complexity, such as the Thue--Morse sequence.\n"}
{"abstract": "  We revisit recent methods that employ graph neural networks for decoding\nerror correcting codes and employ messages that are computed in an\nautoregressive manner. The outgoing messages of the variable nodes are\nconditioned not only on the incoming messages, but also on an estimation of the\nSNR and on the inferred codeword and on two downstream computations: (i) an\nextended vector of parity check outcomes, (ii) the mismatch between the\ninferred codeword and the re-encoding of the information bits of this codeword.\nUnlike most learned methods in the field, our method violates the symmetry\nconditions that enable the other methods to train exclusively with the\nzero-word. Despite not having the luxury of training on a single word, and the\ninability to train on more than a small fraction of the relevant sample space,\nwe demonstrate effective training. The new method obtains a bit error rate that\noutperforms the latest methods by a sizable margin.\n"}
{"abstract": "  We introduce Activity Graph Transformer, an end-to-end learnable model for\ntemporal action localization, that receives a video as input and directly\npredicts a set of action instances that appear in the video. Detecting and\nlocalizing action instances in untrimmed videos requires reasoning over\nmultiple action instances in a video. The dominant paradigms in the literature\nprocess videos temporally to either propose action regions or directly produce\nframe-level detections. However, sequential processing of videos is problematic\nwhen the action instances have non-sequential dependencies and/or non-linear\ntemporal ordering, such as overlapping action instances or re-occurrence of\naction instances over the course of the video. In this work, we capture this\nnon-linear temporal structure by reasoning over the videos as non-sequential\nentities in the form of graphs. We evaluate our model on challenging datasets:\nTHUMOS14, Charades, and EPIC-Kitchens-100. Our results show that our proposed\nmodel outperforms the state-of-the-art by a considerable margin.\n"}
{"abstract": "  Suspended microparticles subjected to AC electrical fields collectively\norganize into band patterns perpendicular to the field direction. The bands\nfurther develop into zigzag shaped patterns, in which the particles are\nobserved to circulate. We demonstrate that this phenomenon can be observed\nquite generically by generating such patterns with a wide range of particles:\nsilica spheres, fatty acid, oil, and coacervate droplets, bacteria, and ground\ncoffee. We show that the phenomenon can be well understood in terms of second\norder electrokinetic flow, which correctly predicts the hydrodynamic\ninteractions required for the pattern formation process.Brownian particle\nsimulations based on these interactions accurately recapitulate all of the\nobserved pattern formation and symmetry-breaking events, starting from a\nhomogeneous particle suspension. The emergence of the formed patterns can be\npredicted quantitatively within a parameter-free theory.\n"}
{"abstract": "  For $\\mathcal{O}$ a reduced operad, a generalized divergence from the\nderivations of a free $\\mathcal{O}$-algebra to a suitable trace space is\nconstructed. In the case of the Lie operad, this corresponds to Satoh's trace\nmap and, for the associative operad, to the double divergence of Alekseev,\nKawazumi, Kuno and Naef. The generalized divergence is shown to be a\n$1$-cocycle for the usual Lie algebra structure on derivations. These results\nplace the previous constructions into a unified framework; moreover, they are\nnatural with respect to the operad.\n  An important new ingredient is the use of naturality with respect to the\ncategory of finite-rank free modules and split monomorphisms over a commutative\nring $R$. This allows the notion of torsion for such functors to be exploited.\n  Supposing that the ring $R$ is a PID and that the operad $\\mathcal{O}$ is\nbinary, the main result relates the kernel of the generalized divergence to the\nsub Lie algebra of the Lie algebra of derivations that is generated by the\nelements of degree one with respect to the grading induced by arity.\n"}
{"abstract": "  We establish a version of Kashiwara's theorem for twisted sheaves of\nBerthelot's arithmetic differential operators for a closed immersion between\nsmooth p-adic formal schemes. As an application, we construct simple modules\nfor crystalline distribution algebras of reductive groups using the twisted\nversion of the crystalline localisation theorem.\n"}
{"abstract": "  We derive exact analytic results for several four-point correlation functions\nfor statistical models exhibiting phase separation in two-dimensions. Our\ntheoretical results are then specialized to the Ising model on the\ntwo-dimensional strip and found to be in excellent agreement with\nhigh-precision Monte Carlo simulations.\n"}
{"abstract": "  Serverless computing, or Function-as-a-Service (FaaS), enables a new way of\nbuilding and scaling applications by allowing users to deploy fine-grained\nfunctions while providing fully-managed resource provisioning and auto-scaling.\nCustom FaaS container support is gaining traction as it enables better control\nover OSes, versioning, and tooling for modernizing FaaS applications. However,\nproviding rapid container provisioning introduces non-trivial challenges for\nFaaS providers, since container provisioning is costly, and real-world FaaS\nworkloads exhibit highly dynamic patterns. In this paper, we design FaaSNet, a\nhighly-scalable middleware system for accelerating FaaS container provisioning.\nFaaSNet is driven by the workload and infrastructure requirements of the FaaS\nplatform at one of the world's largest cloud providers, Alibaba Cloud Function\nCompute. FaaSNet enables scalable container provisioning via a lightweight,\nadaptive function tree (FT) structure. FaaSNet uses an I/O efficient, on-demand\nfetching mechanism to further reduce provisioning costs at scale. We implement\nand integrate FaaSNet in Alibaba Cloud Function Compute. Evaluation results\nshow that FaaSNet: (1) finishes provisioning 2500 function containers on 1000\nvirtual machines in 8.3 seconds, (2) scales 13.4x and 16.3x faster than Alibaba\nCloud's current FaaS platform and a state-of-the-art P2P container registry\n(Kraken), respectively, and (3) sustains a bursty workload using 75.2% less\ntime than an optimized baseline.\n"}
{"abstract": "  The practical application requests both accuracy and efficiency on\nmulti-person pose estimation algorithms. But the high accuracy and fast\ninference speed are dominated by top-down methods and bottom-up methods\nrespectively. To make a better trade-off between accuracy and efficiency, we\npropose a novel multi-person pose estimation framework, SIngle-network with\nMimicking and Point Learning for Bottom-up Human Pose Estimation (SIMPLE).\nSpecifically, in the training process, we enable SIMPLE to mimic the pose\nknowledge from the high-performance top-down pipeline, which significantly\npromotes SIMPLE's accuracy while maintaining its high efficiency during\ninference. Besides, SIMPLE formulates human detection and pose estimation as a\nunified point learning framework to complement each other in single-network.\nThis is quite different from previous works where the two tasks may interfere\nwith each other. To the best of our knowledge, both mimicking strategy between\ndifferent method types and unified point learning are firstly proposed in pose\nestimation. In experiments, our approach achieves the new state-of-the-art\nperformance among bottom-up methods on the COCO, MPII and PoseTrack datasets.\nCompared with the top-down approaches, SIMPLE has comparable accuracy and\nfaster inference speed.\n"}
{"abstract": "  This work considers predicting the relational structure of a hypergraph for a\ngiven set of vertices, as common for applications in particle physics,\nbiological systems and other complex combinatorial problems. A problem arises\nfrom the number of possible multi-way relationships, or hyperedges, scaling in\n$\\mathcal{O}(2^n)$ for a set of $n$ elements. Simply storing an indicator\ntensor for all relationships is already intractable for moderately sized $n$,\nprompting previous approaches to restrict the number of vertices a hyperedge\nconnects. Instead, we propose a recurrent hypergraph neural network that\npredicts the incidence matrix by iteratively refining an initial guess of the\nsolution. We leverage the property that most hypergraphs of interest are\nsparsely connected and reduce the memory requirement to $\\mathcal{O}(nk)$,\nwhere $k$ is the maximum number of positive edges, i.e., edges that actually\nexist. In order to counteract the linearly growing memory cost from training a\nlengthening sequence of refinement steps, we further propose an algorithm that\napplies backpropagation through time on randomly sampled subsequences. We\nempirically show that our method can match an increase in the intrinsic\ncomplexity without a performance decrease and demonstrate superior performance\ncompared to state-of-the-art models.\n"}
{"abstract": "  In this work we improve the runtime of recent classical algorithms for strong\nsimulation of quantum circuits composed of Clifford and T gates. The\nimprovement is obtained by establishing a new upper bound on the stabilizer\nrank of $m$ copies of the magic state\n$|T\\rangle=\\sqrt{2}^{-1}(|0\\rangle+e^{i\\pi/4}|1\\rangle)$ in the limit of large\n$m$. In particular, we show that $|T\\rangle^{\\otimes m}$ can be exactly\nexpressed as a superposition of at most $O(2^{\\alpha m})$ stabilizer states,\nwhere $\\alpha\\leq 0.3963$, improving on the best previously known bound $\\alpha\n\\leq 0.463$. This furnishes, via known techniques, a classical algorithm which\napproximates output probabilities of an $n$-qubit Clifford + T circuit $U$ with\n$m$ uses of the T gate to within a given inverse polynomial relative error\nusing a runtime $\\mathrm{poly}(n,m)2^{\\alpha m}$. We also provide improved\nupper bounds on the stabilizer rank of symmetric product states\n$|\\psi\\rangle^{\\otimes m}$ more generally; as a consequence we obtain a strong\nsimulation algorithm for circuits consisting of Clifford gates and $m$\ninstances of any (fixed) single-qubit $Z$-rotation gate with runtime\n$\\text{poly}(n,m) 2^{m/2}$. We suggest a method to further improve the upper\nbounds by constructing linear codes with certain properties.\n"}
{"abstract": "  Continual learning is a machine learning sub-field specialized in settings\nwith non-iid data. Hence, the training data distribution is not static and\ndrifts through time. Those drifts might cause interferences in the trained\nmodel and knowledge learned on previous states of the data distribution might\nbe forgotten. Continual learning's challenge is to create algorithms able to\nlearn an ever-growing amount of knowledge while dealing with data distribution\ndrifts.\n  One implementation difficulty in these field is to create data loaders that\nsimulate non-iid scenarios. Indeed, data loaders are a key component for\ncontinual algorithms. They should be carefully designed and reproducible. Small\nerrors in data loaders have a critical impact on algorithm results, e.g. with\nbad preprocessing, wrong order of data or bad test set. Continuum is a simple\nand efficient framework with numerous data loaders that avoid researcher to\nspend time on designing data loader and eliminate time-consuming errors. Using\nour proposed framework, it is possible to directly focus on the model design by\nusing the multiple scenarios and evaluation metrics implemented. Furthermore\nthe framework is easily extendable to add novel settings for specific needs.\n"}
{"abstract": "  Convection and turbulence in core-collapse supernovae (CCSNe) are inherently\nthree-dimensional in nature. However, 3D simulations of CCSNe are\ncomputationally demanding. Thus, it is valuable to modify simulations in\nspherical symmetry to incorporate 3D effects using some parametric model. In\nthis paper, we report on the formulation and implementation of general\nrelativistic (GR) neutrino-driven turbulent convection in the spherically\nsymmetric core-collapse supernova code \\texttt{GR1D}. This is based upon the\nrecently proposed method of Supernova Turbulence in Reduced-dimensionality\n(\\textit{STIR}) in Newtonian simulations from \\cite{Couch2020_STIR}. When the\nparameters of this model are calibrated to 3D simulations, we find that our GR\nformulation of \\textit{STIR} requires larger turbulent eddies to achieve a\nshock evolution similar to the original \\textit{STIR} model. We also find that\ngeneral relativity may alter the correspondence between progenitor mass and\nsuccessful vs.~failed explosions.\n"}
{"abstract": "  We present some properties of the Weierstrass $\\wp$-function associated to\nthe hexagonal (or triangular) lattice. In particular, with the help of an old\ntheorem of I.N. Baker \\cite{B} on the characterization of meromorphic solutions\nof the equation $X^3 + Y^3 = 1$ we determine the zeros of the function $\\wp'(z)\n\\pm \\sqrt{3}$.\n"}
{"abstract": "  In a recent MNRAS article, Raposo-Pulido and Pelaez (RPP) designed a scheme\nfor obtaining very close seeds for solving the elliptic Kepler Equation with\nthe classical and the modified Newton-Rapshon methods. This implied an\nimportant reduction in the number of iterations needed to reach a given\naccuracy. However, RPP also made strong claims about the errors of their method\nthat are incorrect. In particular, they claim that their accuracy can always\nreach the level $\\sim5\\varepsilon$, where $\\varepsilon$ is the machine epsilon\n(e.g. $\\varepsilon=2.2\\times10^{-16} $ in double precision), and that this\nresult is attained for all values of the eccentricity $e<1$ and the mean\nanomaly $M\\in[0,\\pi]$, including for $e$ and $M$ that are arbitrarily close to\n$1$ and $0$, respectively. However, we demonstrate both numerically and\nanalytically that any implementation of the classical or modified\nNewton-Raphson methods for Kepler's equation, including those described by RPP,\nhave a limiting accuracy of the order $\\sim\\varepsilon/\\sqrt{2(1-e)}$.\nTherefore the errors of these implementations diverge in the limit $e\\to1$, and\ndiffer dramatically from the incorrect results given by RPP. Despite these\nshortcomings, the RPP method can provide a very efficient option for reaching\nsuch limiting accuracy. We also provide a limit that is valid for the accuracy\nof any algorithm for solving Kepler equation, including schemes like bisection\nthat do not use derivatives. Moreover, similar results are also demonstrated\nfor the hyperbolic Kepler Equation. The methods described in this work can\nprovide guidelines for designing more accurate solutions of the elliptic and\nhyperbolic Kepler equations.\n"}
{"abstract": "  The deep crustal heating, associated with exothermal nuclear reactions, is\nbelieved to be a key parameter for describing the thermal evolution of\naccreting neutron stars. In this paper, we present the first thermodynamically\nconsistent calculations of the crustal heating for realistic compositions of\nthermonuclear ashes. In contrast to previous studies based on the traditional\napproach, we account for neutron hydrostatic/diffusion (nHD) equilibrium\ncondition imposed by superfluidity of neutrons in a major part of the inner\ncrust and rapid diffusion in the remaining part of the inner crust. We apply a\nsimplified reaction network to model nuclear evolution of various\nmulti-component thermonuclear burning ashes (superburst, KEPLER, and extreme\nrp-process ashes) in the outer crust and calculate the deep crustal heating\nenergy release Q, parametrized by the pressure at the outer-inner crust\ninterface, P_{oi}. Using the general thermodynamic arguments, we set a lower\nlimit on Q, Q>0.13-0.2 MeV per baryon (an actual value depends on the ash\ncomposition and the employed mass model).\n"}
{"abstract": "  Studying the statistical properties of the large-scale structure in the\nUniverse with weak gravitational lensing is a prime goal of several current and\nforthcoming galaxy surveys. The power that weak lensing has to constrain\ncosmological parameters can be enhanced by considering statistics beyond\nsecond-order shear correlation functions or power spectra. One such\nhigher-order probe that has proven successful in observational data is the\ndensity split statistics (DSS), in which one analyses the mean shear profiles\naround points that are classified according to their foreground galaxy density.\nIn this paper, we generalise the most accurate DSS model to allow for a broad\nclass of angular filter functions used for the classification of the different\nlocal density regions. This approach is motivated by earlier findings showing\nthat an optimised filter can provide tighter constraints on model parameters\ncompared to the standard top-hat case. We build on large deviation theory\napproaches and approximations thereof to model the matter density PDF, and on\nperturbative calculations of higher-order moments of the density field. The\nnovel addition relies on the generalisation of these previously employed\ncalculations to allow for general filter functions and is validated on several\nsets of numerical simulations. The revised model fits well the simulation\nmeasurements, with a residual systematic offset that is small compared to the\nstatistical accuracy of current weak lensing surveys. The accuracy of the model\nis slightly lower for a compensated filter than for a non-negative filter\nfunction, and that it increases with the filter size. Using a Fisher matrix\napproach, we find constraints comparable to the commonly used two-point cosmic\nshear measures. Hence, our DSS model can be used in competitive analyses of\ncurrent cosmic shear data, while it may need refinements for forthcoming\nlensing surveys.\n"}
{"abstract": "  The software Randentropy is designed to estimate inequality in a random\nsystem where several individuals interact moving among many communities and\nproducing dependent random quantities of an attribute. The overall inequality\nis assessed by computing the Random Theil's Entropy. Firstly, the software\nestimates a piecewise homogeneous Markov chain by identifying the\nchanging-points and the relative transition probability matrices. Secondly, it\nestimates the multivariate distribution function of the attribute using a\ncopula function approach and finally, through a Monte Carlo algorithm,\nevaluates the expected value of the Random Theil's Entropy. Possible\napplications are discussed as related to the fields of finance and human\nmobility\n"}
{"abstract": "  Hierarchical computational methods for multiscale mechanics such as the\nFE$^2$ and FE-FFT methods are generally accompanied by high computational\ncosts. Data-driven approaches are able to speed the process up significantly by\nenabling to incorporate the effective micromechanical response in macroscale\nsimulations without the need of performing additional computations at each\nGauss point explicitly. Traditionally artificial neural networks (ANNs) have\nbeen the surrogate modeling technique of choice in the solid mechanics\ncommunity. However they suffer from severe drawbacks due to their parametric\nnature and suboptimal training and inference properties for the investigated\ndatasets in a three dimensional setting. These problems can be avoided using\nlocal approximate Gaussian process regression (laGPR). This method can allow\nthe prediction of stress outputs at particular strain space locations by\ntraining local regression models based on Gaussian processes, using only a\nsubset of the data for each local model, offering better and more reliable\naccuracy than ANNs. A modified Newton-Raphson approach is proposed to\naccommodate for the local nature of the laGPR approximation when solving the\nglobal structural problem in a FE setting. Hence, the presented work offers a\ncomplete and general framework enabling multiscale calculations combining a\ndata-driven constitutive prediction using laGPR, and macroscopic calculations\nusing an FE scheme that we test for finite-strain three-dimensional\nhyperelastic problems.\n"}
{"abstract": "  The recent advances in cancer immunotherapy boosted the development of\ntumor-immune system models aiming to provide mechanistic understanding and\nindicate more efficient treatment regimes. However, the complexity of such\nmodels, their multi-scale dynamics and their overparameterized character\nrenders them inaccessible for wide utilization. In this work, the dynamics of a\nfundamental model formulating the interactions of tumor cells with natural\nkiller cells, CD8$^+$ T cells and circulating lymphocytes is examined. It is\nfirst shown that the long-term evolution of the system towards high-tumor or\ntumor-free equilibria is determined by the dynamics of an initial\n\\emph{explosive stage} of tumor progression. Focusing on this stage, the\nalgorithmic Computational Singular Perturbation methodology is employed to\nidentify the underlying mechanisms confining the system's evolution towards the\nequilibrium and the governing slow dynamics along them. It is shown that these\ninsights are preserved along different tumor-immune system and\npatient-dependent realizations. Utilizing the obtained mechanistic\nunderstanding, a novel reduced model is constructed in an algorithmic fashion,\nwhich accurately predicts the dynamics of the system during the explosive stage\nand includes half of the parameters of the detailed model. This present\nanalysis demonstrates the potential of algorithmic asymptotic analysis to\nsimplify the complex, overeparameterized and multi-scale nature of cancer\nimmunology models and to indicate the interactions and cell types to target for\nmore effective treatment development.\n"}
{"abstract": "  Several studies point to the antimicrobial effects of ELF electromagnetic\nfields. Such fields have accompanied life from the very beginning, and it is\npossible that they played a significant role in its emergence and evolution.\nHowever, the literature on the biological effects of ELF electromagnetic fields\nis controversial, and we still lack an understanding of the complex mechanisms\nthat make such effects, observed in many experiments, possible. The Covid-19\npandemic has shown how fragile we are in the face of powerful processes\noperating in the biosphere. We believe that understanding the role of ELF\nelectromagnetic fields in regulating the biosphere is important in our fight\nagainst Covid-19, and research in this direction should be intensified.\n"}
{"abstract": "  A simple, self-calibrating, rotating-waveplate polarimeter is largely\ninsensitive to light intensity fluctuations and is shown to be useful for\ndetermining the Stokes parameters of light. This study shows how to minimize\nthe in situ self-calibration time, the measurement time and the measurement\nuncertainty. The suggested methods are applied to measurements of spatial\nvariations in the linear and circular polarizations of laser light passing\nthrough glass plates with a laser intensity dependent birefringence. These are\ncrucial measurements for the ACME electron electric dipole measurements,\nrequiring accuracies in circular and linear polarization fraction of about 0.1%\nand 0.4%, with laser intensities up to 100 $\\text{mW/mm}^2$ incident into the\npolarimeter.\n"}
{"abstract": "  Transverse spin angular momentum of light is a key concept in recent\nnanophotonics to realize unidirectional light transport in waveguides by\nspin-momentum locking. Herein we theoretically propose subwavelength\nnanoparticle chain waveguides that efficiently sort optical spins with\nengineerable spin density distributions. By arranging high-refractive-index\nnanospheres of different sizes in a zigzag manner, directional optical spin\npropagation is realized. The origin of the efficient spin transport is revealed\nby analyzing the dispersion relation and spin angular momentum density\ndistributions. In contrast to conventional waveguides, the proposed asymmetric\nwaveguide can spatially separate up- and down-spins and locate one parity\ninside and the other outside the structure. Moreover, robustness against\nbending the waveguide and its application as an optical spin sorter are\npresented. Compared to previous reports on spatial engineering of local spins\nin photonic crystal waveguides, we achieved substantial miniaturization of the\nentire footprint down to subwavelength scale.\n"}
{"abstract": "  We investigate a correspondence between two formalisms for discrete\nprobabilistic modeling: probabilistic graphical models (PGMs) and tensor\nnetworks (TNs), a powerful modeling framework for simulating complex quantum\nsystems. The graphical calculus of PGMs and TNs exhibits many similarities,\nwith discrete undirected graphical models (UGMs) being a special case of TNs.\nHowever, more general probabilistic TN models such as Born machines (BMs)\nemploy complex-valued hidden states to produce novel forms of correlation among\nthe probabilities. While representing a new modeling resource for capturing\nstructure in discrete probability distributions, this behavior also renders the\ndirect application of standard PGM tools impossible. We aim to bridge this gap\nby introducing a hybrid PGM-TN formalism that integrates quantum-like\ncorrelations into PGM models in a principled manner, using the\nphysically-motivated concept of decoherence. We first prove that applying\ndecoherence to the entirety of a BM model converts it into a discrete UGM, and\nconversely, that any subgraph of a discrete UGM can be represented as a\ndecohered BM. This method allows a broad family of probabilistic TN models to\nbe encoded as partially decohered BMs, a fact we leverage to combine the\nrepresentational strengths of both model families. We experimentally verify the\nperformance of such hybrid models in a sequential modeling task, and identify\npromising uses of our method within the context of existing applications of\ngraphical models.\n"}
{"abstract": "  Many researchers have studied false data injection (FDI) attacks in power\nstate estimation, but existing state estimation approaches are still highly\nvulnerable to FDI attacks. In this paper, we investigate the problem of the\nabove three FDI attacks against dynamic power state estimation (DSE). Although\nthe three attacks were discovered in SSE several years ago, none of them has\nbeen well addressed in static power state systems. In this research, we propose\ntwo robust defense approaches against the above three efficient FDI attacks on\nDSE. Compared to existing approaches, our proposed approaches have three major\ndifferences and significant strengths: (1) they defend against the three FDI\nattacks on dynamic power state estimation rather than static power state\nestimation, (2) they give a robust estimator that can accurately extract a\nsubset of attack-free sensors for power state estimation, and (3) they adopt\nthe little-known Mahalanobis distance in the consistency check of power sensor\nmeasurements, which is different from the Euclidean distance used in all the\nexisting studies on power state estimation. We mathematically prove that the\nMahalanobis distance is not only useful but also much better than the Euclidean\ndistance in the consistency check of power sensor measurements. Our time\ncomplexity analysis shows that the two proposed robust defense approaches are\nefficient. Moreover, in order to demonstrate the effectiveness of the proposed\napproaches, we compare them with the three well-known approaches: the least\nsquare approach, the Imhotep-SMT approach, and the MEE-UKF approach. Our\nextensive experiments show that the proposed approaches further reduce the\nestimation error by two orders of magnitude and four orders of magnitude\ncompared to the Imhotep-SMT approach and the least square approach,\nrespectively. Moreover, our approach is more stable than the MEE-UKF approach.\n"}
{"abstract": "  We compare three populations commonly used in experiments by economists and\nother social scientists: undergraduate students at a physical location (lab),\nAmazon's Mechanical Turk (MTurk), and Prolific. The comparison is made along\nthree dimensions: the noise in the data due to inattention, the cost per\nobservation, and the elasticity of response. We draw samples from each\npopulation, examining decisions in four one-shot games with varying tensions\nbetween the individual and socially efficient choices. When there is no\ntension, where individual and pro-social incentives coincide, noisy behavior\naccounts for 60% of the observations on MTurk, 19% on Prolific, and 14% for the\nlab. Taking costs into account, if noisy data is the only concern Prolific\ndominates from an inferential power point of view, combining relatively low\nnoise with a cost per observation one fifth of the lab's. However, because the\nlab population is more sensitive to treatment, across our main PD game\ncomparison the lab still outperforms both Prolific and MTurk.\n"}
{"abstract": "  In this paper we derive a combinatorial formula for mixed Eulerian numbers in\ntype $A$ from Peterson Schubert calculus. We also provide a simple computation\nfor mixed $\\Phi$-Eulerian numbers in arbitrary Lie types.\n"}
{"abstract": "  Entanglement underpins a variety of quantum-enhanced communication, sensing,\nand computing capabilities. Entanglement-assisted communication (EACOMM)\nleverages entanglement pre-shared by communication parties to boost the rate of\nclassical information transmission. Pioneering theory works showed that EACOMM\ncan enable a communication rate well beyond the ultimate classical capacity of\noptical communications, but an experimental demonstration of any EACOMM\nadvantage remains elusive. Here, we report the implementation of EACOMM\nsurpassing the classical capacity over lossy and noisy bosonic channels. We\nconstruct a high-efficiency entanglement source and a phase-conjugate quantum\nreceiver to reap the benefit of pre-shared entanglement, despite entanglement\nbeing broken by channel loss and noise. We show that EACOMM beats the\nHolevo-Schumacher-Westmoreland capacity of classical communication by up to\n14.6%, when both protocols are subject to the same power constraint at the\ntransmitter. As a practical performance benchmark, a classical communication\nprotocol without entanglement assistance is implemented, showing that EACOMM\ncan reduce the bit-error rate by up to 69% over the same bosonic channel. Our\nwork opens a route to provable quantum advantages in a wide range of quantum\ninformation processing tasks.\n"}
{"abstract": "  Pseudo-relevance feedback mechanisms, from Rocchio to the relevance models,\nhave shown the usefulness of expanding and reweighting the users' initial\nqueries using information occurring in an initial set of retrieved documents,\nknown as the pseudo-relevant set. Recently, dense retrieval -- through the use\nof neural contextual language models such as BERT for analysing the documents'\nand queries' contents and computing their relevance scores -- has shown a\npromising performance on several information retrieval tasks still relying on\nthe traditional inverted index for identifying documents relevant to a query.\nTwo different dense retrieval families have emerged: the use of single embedded\nrepresentations for each passage and query (e.g. using BERT's [CLS] token), or\nvia multiple representations (e.g. using an embedding for each token of the\nquery and document). In this work, we conduct the first study into the\npotential for multiple representation dense retrieval to be enhanced using\npseudo-relevance feedback. In particular, based on the pseudo-relevant set of\ndocuments identified using a first-pass dense retrieval, we extract\nrepresentative feedback embeddings (using KMeans clustering) -- while ensuring\nthat these embeddings discriminate among passages (based on IDF) -- which are\nthen added to the query representation. These additional feedback embeddings\nare shown to both enhance the effectiveness of a reranking as well as an\nadditional dense retrieval operation. Indeed, experiments on the MSMARCO\npassage ranking dataset show that MAP can be improved by upto 26% on the TREC\n2019 query set and 10% on the TREC 2020 query set by the application of our\nproposed ColBERT-PRF method on a ColBERT dense retrieval approach.\n"}
{"abstract": "  Increasingly large datasets are rapidly driving up the computational costs of\nmachine learning. Prototype generation methods aim to create a small set of\nsynthetic observations that accurately represent a training dataset but greatly\nreduce the computational cost of learning from it. Assigning soft labels to\nprototypes can allow increasingly small sets of prototypes to accurately\nrepresent the original training dataset. Although foundational work on `less\nthan one'-shot learning has proven the theoretical plausibility of learning\nwith fewer than one observation per class, developing practical algorithms for\ngenerating such prototypes remains an unexplored territory. We propose a novel,\nmodular method for generating soft-label prototypical lines that still\nmaintains representational accuracy even when there are fewer prototypes than\nthe number of classes in the data. In addition, we propose the Hierarchical\nSoft-Label Prototype k-Nearest Neighbor classification algorithm based on these\nprototypical lines. We show that our method maintains high classification\naccuracy while greatly reducing the number of prototypes required to represent\na dataset, even when working with severely imbalanced and difficult data. Our\ncode is available at https://github.com/ilia10000/SLkNN.\n"}
{"abstract": "  Many objects do not appear frequently enough in complex scenes (e.g., certain\nhandbags in living rooms) for training an accurate object detector, but are\noften found frequently by themselves (e.g., in product images). Yet, these\nobject-centric images are not effectively leveraged for improving object\ndetection in scene-centric images. In this paper, we propose Mosaic of\nObject-centric images as Scene-centric images (MosaicOS), a simple and novel\nframework that is surprisingly effective at tackling the challenges of\nlong-tailed object detection. Keys to our approach are three-fold: (i) pseudo\nscene-centric image construction from object-centric images for mitigating\ndomain differences, (ii) high-quality bounding box imputation using the\nobject-centric images' class labels, and (iii) a multi-stage training\nprocedure. On LVIS object detection (and instance segmentation), MosaicOS leads\nto a massive 60% (and 23%) relative improvement in average precision for rare\nobject categories. We also show that our framework can be compatibly used with\nother existing approaches to achieve even further gains. Our pre-trained models\nare publicly available at https://github.com/czhang0528/MosaicOS/.\n"}
{"abstract": "  Self-organized criticality (SOC) refers to the ability of complex systems to\nevolve towards a 2nd-order phase transition at which interactions between\nsystem components lead to scale-invariant events beneficial for system\nperformance. For the last two decades, considerable experimental evidence\naccumulated that the mammalian cortex with its diversity in cell types and\nconnections might exhibit SOC. Here we review experimental findings of\nisolated, layered cortex preparations to self-organize towards four dynamical\nmotifs identified in the cortex in vivo: up-states, oscillations, neuronal\navalanches, and coherence potentials. During up-states, the synchronization\nobserved for nested theta/gamma-oscillations embeds scale-invariant neuronal\navalanches that exhibit robust power law scaling in size with a slope of -3/2\nand a critical branching parameter of 1. This dynamical coordination, tracked\nin the local field potential (nLFP) and pyramidal neuron activity using\n2-photon imaging, emerges autonomously in superficial layers of organotypic\ncortex cultures and acute cortex slices, is homeostatically regulated, displays\nseparation of time scales, and reveals unique size vs. quiet time dependencies.\nA threshold operation identifies coherence potentials; avalanches that in\naddition maintain the precise time course of propagated synchrony. Avalanches\nemerge under conditions of external driving. Control parameters are established\nby the balance of excitation and inhibition (E/I) and the neuromodulator\ndopamine. This rich dynamical repertoire is not observed in dissociated cortex\ncultures, which lack cortical layers and exhibit dynamics similar to a\n1st-order phase transition. The precise interactions between up-states, nested\noscillations, avalanches, and coherence potentials in superficial cortical\nlayers provide compelling evidence for SOC in the brain.\n"}
{"abstract": "  Information flow measures, over the duration of a game, the audience's belief\nof who will win, and thus can reflect the amount of surprise in a game. To\nquantify the relationship between information flow and audiences' perceived\nquality, we conduct a case study where subjects watch one of the world's\nbiggest esports events, LOL S10. In addition to eliciting information flow, we\nalso ask subjects to report their rating for each game. We find that the amount\nof surprise in the end of the game plays a dominant role in predicting the\nrating. This suggests the importance of incorporating when the surprise occurs,\nin addition to the amount of surprise, in perceived quality models. For content\nproviders, it implies that everything else being equal, it is better for twists\nto be more likely to happen toward the end of a show rather than uniformly\nthroughout.\n"}
{"abstract": "  We study the one parameter family of potential functions $q\\varphi^u$\nassociated with the geometric potential $\\varphi^u$ for the geodesic flow of a\ncompact rank 1 surface of nonpositive curvature. For $q<1$ it is known that\nthere is a unique equilibrium state associated with $q\\varphi^u$, and it has\nfull support. For $q > 1$ it is known that an invariant measure is an\nequilibrium state if and only if it is supported on the singular set. We study\nthe critical value $q=1$ and show that the ergodic equilibrium states are\neither the restriction to the regular set of the Liouville measure, or measures\nsupported on the singular set. In particular, when~$q = 1$, there is a unique\nergodic equilibrium state that gives positive measure to the regular set.\n"}
{"abstract": "  Reconfigurable intelligent surfaces (RISs) are envisioned to be a disruptive\nwireless communication technique that is capable of reconfiguring the wireless\npropagation environment. In this paper, we study a far-field RIS-assisted\nmultiple-input single-output (MISO) communication system operating in free\nspace. To maximize the received power of the receiver from the physics and\nelectromagnetic nature point of view, an optimization, including beamforming of\nthe transmitter, phase shifts of the RIS, orientation and position of the RIS\nis formulated and solved. After exploiting the property of line-of-sight (LoS),\nwe derive closed-form solutions of beamforming and phase shifts. For the\nnon-trivial RIS position optimization problem in arbitrary three-dimensional\nspace, a dimensional-reducing theory is proved. The simulation results show\nthat the proposed closed-form beamforming and phase shifts are near-optimal\nsolutions. Besides, the RIS significantly enhances the performance of the\ncommunication system when it is deployed at the optimal position.\n"}
{"abstract": "  We investigate the geometrical structure of Vaidya's spacetime in the case of\na white hole with decreasing mass, stabilising to a black hole in finite or\ninfinite time or evaporating completely. Our approach relies on a detailed\nanalysis of the ordinary differential equation describing the incoming\nprincipal null geodesics, among which are the generators of the past horizon.\nWe devote special attention to the case of a complete evaporation in infinite\ntime and establish the existence of an asymptotic light-like singularity of the\nconformal curvature, touching both the past space-like singularity and future\ntime-like infinity. This singularity is present independently of the decay rate\nof the mass. We derive an explicit formula that relates directly the strength\nof this null singularity to the asymptotic behaviour of the mass function.\n"}
{"abstract": "  We define topological time crystals, a dynamical phase of periodically driven\nquantum many-body systems capturing the coexistence of topological order with\nthe spontaneous breaking of discrete time-translation symmetry. We show that\nmany-body localization can stabilize this phase against generic perturbations\nand establish some of its key features and signatures. We link topological and\nordinary time crystals through three complementary perspectives: higher-form\nsymmetries, quantum error-correcting codes, and a holographic correspondence.\nWe also propose an experimental realization of a surface-code-based topological\ntime crystal for the Google Sycamore processor.\n"}
{"abstract": "  The paper concerns a new statistical method for assessing dissimilarity of\ntwo random sets based on one realisation of each of them. The method focuses on\nshapes of the components of the random sets, namely on the curvature of their\nboundaries together with the ratios of their perimeters and areas. Theoretical\nbackground is introduced and then, the method is described, justified by a\nsimulation study and applied to real data of two different types of tissue -\nmammary cancer and mastopathy.\n"}
{"abstract": "  In the events of peripheral dissociation of relativistic nuclei in the\nnuclear track emulsion, it is possible to study the emerging ensembles of He\nand H nuclei, including those from decays of unstable $^{8}$Be and $^{9}$B\nnuclei, as well as the Hoyle state. These extremely short-lived states are\nidentified by invariant masses calculated from the angles in 2$\\alpha$-pairs,\n2$\\alpha p$- and 3$\\alpha$-triplets in the approximation of conservation of\nmomentum per nucleon of the primary nucleus. In the same approach, it is\npossible to search for more complex states. This paper explores the correlation\nbetween the formation of $^{8}$Be nuclei and the multiplicity of accompanying\n$\\alpha$-particles in the dissociation of relativistic $^{16}$O, $^{22}$Ne,\n$^{28}$Si, and $^{197}$Au nuclei. On the above basis, estimates of this\ncorrelation are presented for the unstable $^{9}$B nucleus and the Hoyle state.\nThe enhancement in the $^{8}$Be contribution to dissociation with the\n$\\alpha$-particle multiplicity has been found. Decays of $^{9}$B nuclei and\nHoyle states follow the same trend.\n"}
{"abstract": "  Online federated learning (OFL) becomes an emerging learning framework, in\nwhich edge nodes perform online learning with continuous streaming local data\nand a server constructs a global model from the aggregated local models. Online\nmultiple kernel learning (OMKL), using a preselected set of P kernels, can be a\ngood candidate for OFL framework as it has provided an outstanding performance\nwith a low-complexity and scalability. Yet, an naive extension of OMKL into OFL\nframework suffers from a heavy communication overhead that grows linearly with\nP. In this paper, we propose a novel multiple kernel-based OFL (MK-OFL) as a\nnon-trivial extension of OMKL, which yields the same performance of the naive\nextension with 1/P communication overhead reduction. We theoretically prove\nthat MK-OFL achieves the optimal sublinear regret bound when compared with the\nbest function in hindsight. Finally, we provide the numerical tests of our\napproach on real-world datasets, which suggests its practicality.\n"}
{"abstract": "  We develop a far-from-equilibrium hydrodynamic model to evolve\nultrarelativistic heavy-ion collisions in event-by-event simulations.\nAnisotropic hydrodynamics is designed to better handle the strong and highly\nanisotropic expansion during the early stages of the collision. The large\ngradients cause conventional second-order viscous hydrodynamic approaches to\nbreak down at early times. Anisotropic hydrodynamics evolves the large pressure\nanisotropies present in the quark-gluon plasma non-perturbatively, which\nprevents negative longitudinal pressures from developing even under extreme\nconditions. This increased stability allows us to start anisotropic\nhydrodynamics already at a very early longitudinal proper time to evolve the\npre-hydrodynamic stage. In current pre-hydrodynamic models, the equation of\nstate is not consistent with the QCD equation of state used in the subsequent\nfluid dynamic stage. Since our approach avoids this inconsistency, we are able\nto achieve a smooth transition to non-conformal viscous hydrodynamics as the\ngradients decrease over time. For our first phenomenological application, we\napply our new simulation to model fluctuating Pb+Pb collisions at LHC energies\n($\\sqrt{s_\\text{NN}} = 2.76$ TeV) and find that our preliminary calculations\nfor the hadronic observables are in excellent agreement with the experimental\ndata.\n"}
{"abstract": "  A direct formulation of linear elasticity of cell complexes based on discrete\nexterior calculus is presented. The primary unknown are displacements,\nrepresented by primal vector-valued 0-cochain. Displacement differences and\ninternal forces are represented by primal vector-valued 1-cochain and dual\nvector-valued 2-cochain, respectively. The macroscopic constitutive relation is\nenforced at primal 0-cells with the help of musical isomorphisms mapping\ncochains to smooth fields and vice versa. The balance of linear momentum is\nestablished at primal 0-cells. The governing equations are solved as a Laplace\nequation with a non-local and non-diagonal material Hodge star. Numerical\nsimulations of several classical problems with analytic solutions are presented\nto validate the formulation. Good agreement with known solutions is obtained.\nThe formulation provides a method to calculate the relations between\ndisplacement differences and internal forces for any lattice structure, when\nthe structure is required to follow a prescribed macroscopic elastic behaviour.\nThis is also the first and critical step in developing formulations for\ndissipative processes in cell complexes.\n"}
{"abstract": "  Magic-angle twisted bilayer graphene exhibits a panoply of many-body\nphenomena that are intimately tied to the appearance of narrow and\nwell-isolated electronic bands. The microscopic ingredients that are\nresponsible for the complex experimental phenomenology include\nelectron-electron (phonon) interactions and non-trivial Bloch wavefunctions\nassociated with the narrow bands. Inspired by recent experiments, we focus on\ntwo independent quantities that are considerably modified by Coulomb\ninteraction driven band renormalization, namely the density of states and the\nminimal spatial extent associated with the Wannier functions. First, we show\nthat a filling-dependent enhancement of the density of states, caused by band\nflattening, in combination with phonon-mediated attraction due to\nelectron-phonon umklapp processes, increases the tendency towards\nsuperconducting pairing in a range of angles around magic-angle. Second, we\ndemonstrate that the minimal spatial extent associated with the Wannier\nfunctions, which contributes towards increasing the superconducting phase\nstiffness, also develops a non-trivial enhancement due to the interaction\ninduced renormalization of the Bloch wavefunctions. While our modeling of\nsuperconductivity assumes a weak electron-phonon coupling and does not consider\nmany of the likely relevant correlation effects, it explains simply the\nexperimentally observed robustness of superconductivity in the wide range of\nangles that occurs in the relevant range of fillings.\n"}
{"abstract": "  We develop simple methods for constructing parameter priors for model choice\namong Directed Acyclic Graphical (DAG) models. In particular, we introduce\nseveral assumptions that permit the construction of parameter priors for a\nlarge number of DAG models from a small set of assessments. We then present a\nmethod for directly computing the marginal likelihood of every DAG model given\na random sample with no missing observations. We apply this methodology to\nGaussian DAG models which consist of a recursive set of linear regression\nmodels. We show that the only parameter prior for complete Gaussian DAG models\nthat satisfies our assumptions is the normal-Wishart distribution. Our analysis\nis based on the following new characterization of the Wishart distribution: let\n$W$ be an $n \\times n$, $n \\ge 3$, positive-definite symmetric matrix of random\nvariables and $f(W)$ be a pdf of $W$. Then, f$(W)$ is a Wishart distribution if\nand only if $W_{11} - W_{12} W_{22}^{-1} W'_{12}$ is independent of\n$\\{W_{12},W_{22}\\}$ for every block partitioning $W_{11},W_{12}, W'_{12},\nW_{22}$ of $W$. Similar characterizations of the normal and normal-Wishart\ndistributions are provided as well.\n"}
{"abstract": "  We compute corrections to the hard thermal (or dense) loop photon\npolarization tensor associated to a small mass $m$ of the fermions of an\nelectromagnetic plasma at high temperature $T$ (or chemical potential $\\mu$).\nTo this aim we use the on-shell effective field theory, amended with mass\ncorrections. We also carry out the computation using transport theory, reaching\nto the same result. Interemediate steps in the computations reveal the presence\nof potential infrared divergencies. We use dimensional regularization, as it is\nrespectful with the gauge symmetry, and then show that all infrared\ndivergencies cancel in the final result. We compare the mass corrections with\nboth the power and two-loop corrections, and claim that they are equally\nimportant if the mass is soft, that is, of order $e T$ (or $e \\mu$), where $e$\nis the gauge coupling constant, but are dominat if the mass obeys $ e T < m \\ll\nT $ (or $e \\mu < m \\ll \\mu)$.\n"}
{"abstract": "  The fractional quantum Hall effect has been considered as a puzzling quantum\nmany-body phenomenon that has yet to be fully explained. The plateau width and\nexcitation energy gap are particularly problematic. We report here that those\ntwo are determined by degrees of multi-particle correlations among the skipping\nelectrons forming the edge current flowing in incompressible strips (ISs).\nConsideration of the total angular momentum of correlated skipping electrons\nand their images, which are introduced to eliminate the confining potential\nwithin the IS, yields additional Zeeman energies that hierarchically split the\nLandau levels (LLs) by correlation order. This level splitting produces all the\nodd-denominator plateaus and explains the occurrence of fractional charges,\nwhile the split distances representing the correlation strengths determine both\nplateau widths and excitation energy gaps. With such a scheme, we explicitly\nreproduce an experimental Hall resistivity curve for the lowest LL and reveal\nthe characteristics of the half-filling state.\n"}
{"abstract": "  Kirigami, art of paper cutting, enables two-dimensional sheets transforming\ninto unique shapes which are also hard to reshape once with prescribed cutting\npatterns. Rare kirigami designs manipulate cuts on three-dimensional objects to\ncompose periodic structures with programmability and/or re-programmability.\nHere, we propose a new class of three-dimensional modular kirigami by\nintroducing cuts on cuboid-shaped objects, based on which constructing two\nquasi-three-15 dimensional architected kirigamis with even-flat structural\nform. We demonstrate the proposed architected kirigamis are with rich\nmobilities triggered by kinematic bifurcations inherited from their composed\nmodular kirigami, and can undergo living-matter-like metamorphosis evolving\ninto miscellaneous transformable three-dimensional architectures and even a\npluripotent platform capable of being re-programmed into curvature different\nsurfaces through inverse design. Such 20 metamorphic structures could find\nbroad applications in reconfigurable metamaterials, transformable robots and\narchitectures.\n"}
{"abstract": "  A set $\\mathbb{Y}$ of elements (vertices or edges) in space is said to be a\n$generator$ of a metric space if each element of the space is recognized by its\ndistances from the elements of $\\mathbb{Y}$, uniquely. The generator with\nminimum cardinality is known as the $basis$ of the metric space, and this\ncardinality is the $dimension$ of the given space. In this article, we further\ndiscuss these notions with respect to a heptagonal circular ladder. We show\nthat for a heptagonal circular ladder $\\Gamma_{n}$, the edge metric dimension\nis three and find that it equals its metric dimension. We also introduce a new\nfamily of the convex polytope graph (denoted by $\\Delta_{n}$) from a heptagonal\ncircular ladder and find its metric dimension. Furthermore, we prove that the\nminimum generator (metric and edge metric) are independent for all of these\nfamilies of the convex polytopes.\n"}
{"abstract": "  In theoretical computer science, conferences play an important role in the\nscientific process. The decisions whether to accept or reject articles is taken\nby the program committee (PC) members. Serving on a PC for the first time can\nbe a daunting experience. This guide will help new program-committee members to\nunderstand how the system works, and provide useful tips and guidelines. It\ndiscusses every phase of the paper-selection process, and the tasks associated\nto it.\n"}
{"abstract": "  We prove an $\\Omega(n^2)$ lower bound on the query complexity of local\nproportionality in the Robertson-Webb cake-cutting model. Local proportionality\nrequires that each agent prefer their allocation to the average of their\nneighbors' allocations in some undirected social network. It is a weaker\nfairness notion than envy-freeness, which also has query complexity\n$\\Omega(n^2)$, and generally incomparable to proportionality, which has query\ncomplexity $\\Theta(n \\log n)$. This result separates the complexity of local\nproportionality from that of ordinary proportionality, confirming the intuition\nthat finding a locally proportional allocation is a more difficult\ncomputational problem.\n"}
{"abstract": "  Real-time adaptation is imperative to the control of robots operating in\ncomplex, dynamic environments. Adaptive control laws can endow even nonlinear\nsystems with good trajectory tracking performance, provided that any uncertain\ndynamics terms are linearly parameterizable with known nonlinear features.\nHowever, it is often difficult to specify such features a priori, such as for\naerodynamic disturbances on rotorcraft or interaction forces between a\nmanipulator arm and various objects. In this paper, we turn to data-driven\nmodeling with neural networks to learn, offline from past data, an adaptive\ncontroller with an internal parametric model of these nonlinear features. Our\nkey insight is that we can better prepare the controller for deployment with\ncontrol-oriented meta-learning of features in closed-loop simulation, rather\nthan regression-oriented meta-learning of features to fit input-output data.\nSpecifically, we meta-learn the adaptive controller with closed-loop tracking\nsimulation as the base-learner and the average tracking error as the\nmeta-objective. With a nonlinear planar rotorcraft subject to wind, we\ndemonstrate that our adaptive controller outperforms other controllers trained\nwith regression-oriented meta-learning when deployed in closed-loop for\ntrajectory tracking control.\n"}
{"abstract": "  High cost of training time caused by multi-step adversarial example\ngeneration is a major challenge in adversarial training. Previous methods try\nto reduce the computational burden of adversarial training using single-step\nadversarial example generation schemes, which can effectively improve the\nefficiency but also introduce the problem of catastrophic overfitting, where\nthe robust accuracy against Fast Gradient Sign Method (FGSM) can achieve nearby\n100\\% whereas the robust accuracy against Projected Gradient Descent (PGD)\nsuddenly drops to 0\\% over a single epoch. To address this problem, we propose\na novel Fast Gradient Sign Method with PGD Regularization (FGSMPR) to boost the\nefficiency of adversarial training without catastrophic overfitting. Our core\nidea is that single-step adversarial training can not learn robust internal\nrepresentations of FGSM and PGD adversarial examples. Therefore, we design a\nPGD regularization term to encourage similar embeddings of FGSM and PGD\nadversarial examples. The experiments demonstrate that our proposed method can\ntrain a robust deep network for L$_\\infty$-perturbations with FGSM adversarial\ntraining and reduce the gap to multi-step adversarial training.\n"}
{"abstract": "  Quantum conference is a process of securely exchanging messages between three\nor more parties, using quantum resources. A Measurement Device Independent\nQuantum Dialogue (MDI-QD) protocol, which is secure against information\nleakage, has been proposed (Quantum Information Processing 16.12 (2017): 305)\nin 2017, is proven to be insecure against intercept-and-resend attack strategy.\nWe first modify this protocol and generalize this MDI-QD to a three-party\nquantum conference and then to a multi-party quantum conference. We also\npropose a protocol for quantum multi-party XOR computation. None of these three\nprotocols proposed here use entanglement as a resource and we prove the\ncorrectness and security of our proposed protocols.\n"}
{"abstract": "  Carmichael showed for sufficiently large $L$, that $F_L$ has at least one\nprime divisor that is $\\pm 1({\\rm mod}\\, L)$. For a given $F_L$, we will show\nthat a product of distinct odd prime divisors with that congruence condition is\na Fibonacci pseudoprime. Such pseudoprimes can be used in an attempt, here\nunsuccessful, to find an example of a Baillie-PSW pseudoprime, i.e.\\ an odd\nFibonacci pseudoprime that is congruent to $\\pm 2({\\rm mod}\\, 5)$ and is also a\nbase-2 pseudoprime.\n"}
{"abstract": "  Recent work has demonstrated the effectiveness of cross-lingual language\nmodel pretraining for cross-lingual understanding. In this study, we present\nthe results of two larger multilingual masked language models, with 3.5B and\n10.7B parameters. Our two new models dubbed XLM-R XL and XLM-R XXL outperform\nXLM-R by 1.8% and 2.4% average accuracy on XNLI. Our model also outperforms the\nRoBERTa-Large model on several English tasks of the GLUE benchmark by 0.3% on\naverage while handling 99 more languages. This suggests pretrained models with\nlarger capacity may obtain both strong performance on high-resource languages\nwhile greatly improving low-resource languages. We make our code and models\npublicly available.\n"}
{"abstract": "  Dense retrieval has been shown to be effective for retrieving relevant\ndocuments for Open Domain QA, surpassing popular sparse retrieval methods like\nBM25. REALM (Guu et al., 2020) is an end-to-end dense retrieval system that\nrelies on MLM based pretraining for improved downstream QA efficiency across\nmultiple datasets. We study the finetuning of REALM on various QA tasks and\nexplore the limits of various hyperparameter and supervision choices. We find\nthat REALM was significantly undertrained when finetuning and simple\nimprovements in the training, supervision, and inference setups can\nsignificantly benefit QA results and exceed the performance of other models\npublished post it. Our best model, REALM++, incorporates all the best working\nfindings and achieves significant QA accuracy improvements over baselines\n(~5.5% absolute accuracy) without any model design changes. Additionally,\nREALM++ matches the performance of large Open Domain QA models which have 3x\nmore parameters demonstrating the efficiency of the setup.\n"}
{"abstract": "  Yellowballs (YBs) were first discovered during the Milky Way Project\ncitizen-science initiative (MWP; Simpson et al. 2012). MWP users noticed\ncompact, yellow regions in Spitzer Space Telescope mid-infrared (MIR) images of\nthe Milky Way plane and asked professional astronomers to explain these \"yellow\nballs.\" Follow-up work by Kerton et al. (2015) determined that YBs likely trace\ncompact photo-dissociation regions associated with massive and\nintermediate-mass star formation. YBs were included as target objects in a\nversion of the Milky Way Project launched in 2016 (Jayasinghe et al. 2016),\nwhich produced a listing of over 6000 YB locations. We have measured distances,\ncross-match associations, physical properties, and MIR colors of ~500 YBs\nwithin a pilot region covering the l= 30 - 40 degrees, b= +/- 1 degree region\nof the Galactic plane. We find 20-30% of YBs in our pilot region contain\nhigh-mass star formation capable of becoming expanding H II regions that\nproduce MIR bubbles. A majority of YBs represent intermediate-mass star-forming\nregions whose placement in evolutionary diagrams suggest they are still\nactively accreting, and may be precursors to optically-revealed Herbig Ae/Be\nnebulae. Many of these intermediate-mass YBs were missed by surveys of massive\nstar-formation tracers and thus this catalog provides information for many new\nsites of star formation. Future work will expand this pilot region analysis to\nthe entire YB catalog.\n"}
{"abstract": "  In this article, we propose a method to realize the \"delayed choice\nexperiment\" using ultra-cold atoms. Here we attempt to probe the \"welcher-Weg\"\ninformation without collapsing the wavefunction of the atom. This experiment\nconsists of components built around proven techniques that are put together in\nnovel configuration to preserve the coherence of the system during the\nmeasurement. The Ramsey interference is used to establish the wave nature of\nthe atom and the particle nature of the atom is probed by detecting its\ninternal state by performing a nondemolition measurement using an ultra-high\nfinesse cavity. The coherence of the atom is preserved by adjusting the\natom-cavity interaction time such that the state of the atom is unchanged when\nit emerges out of the cavity.\n"}
{"abstract": "  A light-weight high-performance Deepfake detection method, called DefakeHop,\nis proposed in this work. State-of-the-art Deepfake detection methods are built\nupon deep neural networks. DefakeHop extracts features automatically using the\nsuccessive subspace learning (SSL) principle from various parts of face images.\nThe features are extracted by c/w Saab transform and further processed by our\nfeature distillation module using spatial dimension reduction and soft\nclassification for each channel to get a more concise description of the face.\nExtensive experiments are conducted to demonstrate the effectiveness of the\nproposed DefakeHop method. With a small model size of 42,845 parameters,\nDefakeHop achieves state-of-the-art performance with the area under the ROC\ncurve (AUC) of 100%, 94.95%, and 90.56% on UADFV, Celeb-DF v1 and Celeb-DF v2\ndatasets, respectively.\n"}
{"abstract": "  Neural networks allow us to model complex relationships between variables. We\nshow how to efficiently find extrema of a trained neural network in regression\nproblems. Finding the extremizing input of an approximated model is formulated\nas the training of an additional neural network with a loss function that\nminimizes when the extremizing input is achieved. We further show how to\nincorporate additional constraints on the input vector such as limiting the\nextrapolation of the extremizing input vector from the original training data\nset. An instructional example of this approach using TensorFlow is included.\n"}
{"abstract": "  In this work, we consider two-stage polynomial optimization problems under\nuncertainty. In the first stage, one needs to decide upon the values of a\nsubset of optimization variables (control variables). In the second stage, the\nuncertainty is revealed and the rest of optimization variables (state\nvariables) are set up as a solution to a known system of possibly non-linear\nequations. This type of problem occurs, for instance, in optimization for\ndynamical systems, such as electric power systems. We combine tools from\npolynomial and robust optimization to provide a framework for general\nadjustable robust polynomial optimization problems. In particular, we propose\nan iterative algorithm to build a sequence of (approximately) robustly feasible\nsolutions with an improving objective value and verify robust feasibility or\ninfeasibility of the resulting solution under a semialgebraic uncertainty set.\nAt each iteration, the algorithm optimizes over a subset of the feasible set\nand uses affine approximations of the second-stage equations while preserving\nthe non-linearity of other constraints. The algorithm allows for additional\nsimplifications in case of possibly non-convex quadratic problems under\nellipsoidal uncertainty. We implement our approach for AC Optimal Power Flow\nand demonstrate the performance of our proposed method on Matpower instances.\n"}
{"abstract": "  The inspiral and merger of two black holes produces a remnant black hole with\nmass and spin determined by the properties of its parent black holes. Using the\ninferred population properties of component black holes from the first two and\na half observing runs of Advanced LIGO and Virgo, we calculate the population\nproperties of the leftover remnant black holes. By integrating their rate of\nformation over the age of the universe, we estimate the number density of\nremnant black holes today. Using simple prescriptions for the cosmic star\nformation rate and black hole inspiral delay times, we determine the number\ndensity of this leftover black hole population to be $660_{-240}^{+440}\n\\mathrm{Mpc}^{-3}$, corresponding to $\\sim 60,000$ black hole remnants per\nMilky-Way-equivalent galaxy. The mass spectrum of these remnants starts at\n$\\sim 10 M_\\odot$ and can be approximated by a decreasing exponential with\ncharacteristic length $\\sim 15 M_\\odot$, the final spin distribution is sharply\npeaked at $\\chi_f\\sim0.7$, and the kick velocities range from tens to thousands\nof km/s. These kick velocities suggest that globular clusters and nuclear star\nclusters may retain up to $3_{-2}^{+3}\\%$ and $46_{-15}^{+17}\\%$ of their\nremnant black holes, respectively, while young star clusters would only retain\na few tenths of a percent. The estimates in this work assume that none of the\nremnants participate in subsequent hierarchical mergers. If hierarchical\nmergers occur, the overall number density would drop accordingly and the\nremnant mass distribution shape would evolve over time. This population of\nleftover black holes is an inescapable result from gravitational-wave\nobservations of binary black-hole mergers.\n"}
{"abstract": "  Hall instability in electron magnetohydrodynamics is interpreted as the\nshear-Hall instability driven jointly by helicoidal oscillations and shear in\nthe electron current velocity. This explanation suggests an antiparallel\norientation of the background magnetic field and vorticity of the current\nvelocity as the necessary condition for Hall instability. The condition is\ntested and generally confirmed by numerical computations in plane slab\ngeometry. Unstable eigenmodes are localized in the spatial regions of the\nantiparallel field and vorticity. Computations of the tearing-type mode of the\ninstability are complemented by (and generally agree with) asymptotic\nanalytical estimations for large Hall numbers. The stabilizing effect of\nperfect conductor boundary conditions is found and explained. For large Hall\nnumbers, the growth rates approach the power law dependence $\\sigma \\propto\nB^\\alpha\\eta^{1-\\alpha}$ on the magnetic field ($B$) and diffusivity ($\\eta$).\nAlmost all computations give the power index $\\alpha = 3/4$ with one exception\nof the tearing-type mode with vacuum boundary conditions for which case $\\alpha\n= 2/3$.\n"}
{"abstract": "  We prove that for any integer $k\\geq 2$ and $\\varepsilon>0$, there is an\ninteger $\\ell_0\\geq 1$ such that any $k$-uniform hypergraph on $n$ vertices\nwith minimum codegree at least $(1/2+\\varepsilon)n$ has a fractional\ndecomposition into tight cycles of length $\\ell$ ($\\ell$-cycles for short)\nwhenever $\\ell\\geq \\ell_0$ and $n$ is large in terms of $\\ell$. This is\nessentially tight.\n  This immediately yields also approximate integral decompositions for these\nhypergraphs into $\\ell$-cycles. Moreover, for graphs this even guarantees\nintegral decompositions into $\\ell$-cycles and solves a problem posed by Glock,\nK\\\"uhn and Osthus. For our proof, we introduce a new method for finding a set\nof $\\ell$-cycles such that every edge is contained in roughly the same number\nof $\\ell$-cycles from this set by exploiting that certain Markov chains are\nrapidly mixing.\n"}
{"abstract": "  We show that viewing graphs as sets of node features and incorporating\nstructural and positional information into a transformer architecture is able\nto outperform representations learned with classical graph neural networks\n(GNNs). Our model, GraphiT, encodes such information by (i) leveraging relative\npositional encoding strategies in self-attention scores based on positive\ndefinite kernels on graphs, and (ii) enumerating and encoding local\nsub-structures such as paths of short length. We thoroughly evaluate these two\nideas on many classification and regression tasks, demonstrating the\neffectiveness of each of them independently, as well as their combination. In\naddition to performing well on standard benchmarks, our model also admits\nnatural visualization mechanisms for interpreting graph motifs explaining the\npredictions, making it a potentially strong candidate for scientific\napplications where interpretation is important. Code available at\nhttps://github.com/inria-thoth/GraphiT.\n"}
{"abstract": "  Let $n\\ge 2$ and $k\\ge 1$ be two integers. Let $M$ be an isometrically\nimmersed closed $n$-submanifold of co-dimension $k$ that is homotopic to a\npoint in a complete manifold $N$, where the sectional curvature of $N$ is no\nmore than $\\delta<0$. We prove that the total squared mean curvature of $M$ in\n$N$ and the first non-zero eigenvalue $\\lambda_1(M)$ of $M$ satisfies\n$$\\lambda_1(M)\\le n\\left(\\delta +\\frac{1}{\\operatorname{Vol} M}\\int_M |H|^2\n\\operatorname{dvol}\\right).$$ The equality implies that $M$ is minimally\nimmersed in a metric sphere after lifted to the universal cover of $N$. This\ncompletely settles an open problem raised by E. Heintze in 1988.\n"}
{"abstract": "  The development of Industrial Internet of Things systems (IIoT) requires\ntools robust enough to cope with the complexity and heterogeneity of such\nsystems, which are supposed to work in safety-critical conditions. The\navailability of methodologies to support early analysis, verification, and\nvalidation is still an open issue in the research community. The early\nreal-time schedulability analysis can help quantify to what extent the desired\nsystem's timing performance can eventually be achieved. In this paper, we\npresent CHESSIoT, a model-driven environment to support the design and analysis\nof industrial IoT systems. CHESSIoT follows a multi-view, component-based\nmodelling approach with a comprehensive way to perform event-based modelling on\nsystem components for code generation purposes employing an intermediate\nThingML model. To showcase the capability of the extension, we have designed\nand analysed an Industrial real-time safety use case.\n"}
{"abstract": "  Generalization is the ability of a model to predict on unseen domains and is\na fundamental task in machine learning. Several generalization bounds, both\ntheoretical and empirical have been proposed but they do not provide tight\nbounds .In this work, we propose a simple yet effective method to predict the\ngeneralization performance of a model by using the concept that models that are\nrobust to augmentations are more generalizable than those which are not. We\nexperiment with several augmentations and composition of augmentations to check\nthe generalization capacity of a model. We also provide a detailed motivation\nbehind the proposed method. The proposed generalization metric is calculated\nbased on the change in the output of the model after augmenting the input. The\nproposed method was the first runner up solution for the NeurIPS competition on\nPredicting Generalization in Deep Learning.\n"}
{"abstract": "  A phase slip is a localized disturbance in the coherence of a superconductor\nallowing an abrupt 2$\\pi$ phase shift. Phase slips are a ubiquitous feature of\none-dimensional superconductors and also have an analogue in two-dimensions.\nHere we present electrical transport measurements on boron-doped\nnanocrystalline diamond (BNCD) microbridges where, despite their\nthree-dimensional macroscopic geometry, we find clear evidence of phase\nslippage in both the resistance-temperature and voltage-current\ncharacteristics. We attribute this behavior to the unusual microstructure of\nBNCD. We argue that the columnar crystal structure of BNCD forms an intrinsic\nJosephson junction array that supports a line of phase slippage across the\nmicrobridge. The voltage-state in these bridges is metastable and we\ndemonstrate the ability to switch deterministically between different\nsuperconducting states by applying electromagnetic noise pulses. This\nmetastability is remarkably similar to that observed in $\\delta$-MoN nanowires,\nbut with a vastly greater response voltage.\n"}
{"abstract": "  Despite the vast literature on Human Activity Recognition (HAR) with wearable\ninertial sensor data, it is perhaps surprising that there are few studies\ninvestigating semisupervised learning for HAR, particularly in a challenging\nscenario with class imbalance problem. In this work, we present a new\nbenchmark, called A*HAR, towards semisupervised learning for class-imbalanced\nHAR. We evaluate state-of-the-art semi-supervised learning method on A*HAR, by\ncombining Mean Teacher and Convolutional Neural Network. Interestingly, we find\nthat Mean Teacher boosts the overall performance when training the classifier\nwith fewer labelled samples and a large amount of unlabeled samples, but the\nclassifier falls short in handling unbalanced activities. These findings lead\nto an interesting open problem, i.e., development of semi-supervised HAR\nalgorithms that are class-imbalance aware without any prior knowledge on the\nclass distribution for unlabeled samples. The dataset and benchmark evaluation\nare released at https://github.com/I2RDL2/ASTAR-HAR for future research.\n"}
{"abstract": "  We extend previous work on symbolic self-triggered control for\nnon-deterministic continuous-time nonlinear systems without stability\nassumptions to a larger class of specifications. Our goal is to synthesise a\ncontroller for two objectives: the first one is modelled as a right-recursive\nLTL formula, and the second one is to ensure that the average communication\nrate between the controller and the system stays below a given threshold. We\ntranslate the control problem to solving a mean-payoff parity game played on a\ndiscrete graph. Apart from extending the class of specifications, we propose a\nheuristic method to shorten the computation time. Finally, we illustrate our\nresults on the example of a navigating nonholonomic robot with several\nspecifications.\n"}
{"abstract": "  This paper formalizes connections between stability of polynomials and\nconvergence rates of Markov Chain Monte Carlo (MCMC) algorithms. We prove that\nif a (multivariate) partition function is nonzero in a region around a real\npoint $\\lambda$ then spectral independence holds at $\\lambda$. As a\nconsequence, for Holant-type problems (e.g., spin systems) on bounded-degree\ngraphs, we obtain optimal $O(n\\log n)$ mixing time bounds for the single-site\nupdate Markov chain known as the Glauber dynamics. Our result significantly\nimproves the running time guarantees obtained via the polynomial interpolation\nmethod of Barvinok (2017), refined by Patel and Regts (2017).\n  There are a variety of applications of our results. In this paper, we focus\non Holant-type (i.e., edge-coloring) problems, including weighted edge covers\nand weighted even subgraphs. For the weighted edge cover problem (and several\nnatural generalizations) we obtain an $O(n\\log{n})$ sampling algorithm on\nbounded-degree graphs. The even subgraphs problem corresponds to the\nhigh-temperature expansion of the ferromagnetic Ising model. We obtain an\n$O(n\\log{n})$ sampling algorithm for the ferromagnetic Ising model with a\nnonzero external field on bounded-degree graphs, which improves upon the\nclassical result of Jerrum and Sinclair (1993) for this class of graphs. We\nobtain further applications to antiferromagnetic two-spin models on line\ngraphs, weighted graph homomorphisms, tensor networks, and more.\n"}
{"abstract": "  We address an issue of how to accurately include the self energy effect of\nthe screened electron-electron Coulomb interaction in the phonon-mediated\nsuperconductors from first principles. In the Eliashberg theory for\nsuperconductors, self energy is usually decomposed using the $2\\times 2$ Pauli\nmatrices in the electron-hole space. We examine how the diagonal ($\\sigma_{0}$\nand $\\sigma_{3}$) components, which results in the quasiparticle correction to\nthe normal state, behave in the homogeneous electron gas in order to establish\na norm of treating those components in real metallic systems. Within the\n$G_{0}W_{0}$ approximation, we point out that these components are non-analytic\nnear the Fermi surface but their directional derivatives and resulting\ncorrections to the quasiparticle velocity are nevertheless well defined. In the\nlow-energy spectrum, we observe large cancellation between effects of these\ncomponents and, without the numerically more tedious $\\sigma_{3}$ component,\nthe effective mass is incorrectly increased. Feasible paths to manage this\ncancellation in the ab initio Eliashberg calculations are discussed.\n"}
{"abstract": "  In March 2021, Turkey withdrew from The Istanbul Convention, a human-rights\ntreaty that addresses violence against women, citing issues with the\nconvention's implicit recognition of sexual and gender minorities. In this\nwork, we trace disinformation campaigns related to the Istanbul Convention and\nits associated Turkish law that circulate on divorced men's rights Facebook\ngroups. We find that these groups adjusted the narrative and focus of the\ncampaigns to appeal to a larger audience, which we refer to as \"tactical\nreframing.\" Initially, the men organized in a grass-roots manner to campaign\nagainst the Turkish law that was passed to codify the convention, focusing on\none-sided custody of children and indefinite alimony. Later, they reframed\ntheir campaign and began attacking the Istanbul Convention, highlighting its\nacknowledgment of homosexuality. This case study highlights how disinformation\ncampaigns can be used to weaponize homophobia in order to limit the rights of\nwomen. To the best of our knowledge, this is the first case study that analyzes\na narrative reframing in the context of a disinformation campaign on social\nmedia.\n"}
{"abstract": "  Age of Incorrect Information (AoII) is a newly introduced performance metric\nthat considers communication goals. Therefore, comparing with traditional\nperformance metrics and the recently introduced metric - Age of Information\n(AoI), AoII achieves better performance in many real-life applications.\nHowever, the fundamental nature of AoII has been elusive so far. In this paper,\nwe consider the AoII in a system where a transmitter sends updates about a\nmulti-state Markovian source to a remote receiver through an unreliable\nchannel. The communication goal is to minimize AoII subject to a power\nconstraint. We cast the problem into a Constrained Markov Decision Process\n(CMDP) and prove that the optimal policy is a mixture of two deterministic\nthreshold policies. Afterward, by leveraging the notion of Relative Value\nIteration (RVI) and the structural properties of threshold policy, we propose\nan efficient algorithm to find the threshold policies as well as the mixing\ncoefficient. Lastly, numerical results are laid out to highlight the\nperformance of AoII-optimal policy.\n"}
{"abstract": "  We show that compact Riemannian manifolds, regarded as metric spaces with\ntheir global geodesic distance, cannot contain a number of rigid structures\nsuch as (a) arbitrarily large regular simplices or (b) arbitrarily long\nsequences of points equidistant from pairs of points preceding them in the\nsequence. All of this provides evidence that Riemannian metric spaces admit\nwhat we term loose embeddings into finite-dimensional Euclidean spaces:\ncontinuous maps that preserve both equality as well as inequality.\n  We also prove a local-to-global principle for Riemannian-metric-space loose\nembeddability: if every finite subspace thereof is loosely embeddable into a\ncommon $\\mathbb{R}^N$, then the metric space as a whole is loosely embeddable\ninto $\\mathbb{R}^N$ in a weakened sense.\n"}
{"abstract": "  We show that gapless spin liquids, which are potential candidates to describe\nthe ground state of frustrated Heisenberg models in two dimensions, become\ntrivial insulators on cylindrical geometries with an even number of legs. In\nparticular, we report calculations for Gutzwiller-projected fermionic states on\nstrips of square and kagome lattices. By choosing different boundary conditions\nfor the fermionic degrees of freedom, both gapless and gapped states may be\nrealized, the latter ones having a lower variational energy. The direct\nevaluation of static and dynamical correlation functions, as well as overlaps\nbetween different states, allows us to demonstrate the sharp difference between\nthe ground-state properties obtained within cylinders or directly in the\ntwo-dimensional lattice. Our results shed light on the difficulty to detect\nbona fide gapless spin liquids in such cylindrical geometries.\n"}
{"abstract": "  We prove that any $n$-dimensional closed mean convex $\\lambda$-hypersurface\nis convex if $\\lambda\\le 0.$ This generalizes Guang's work on $2$-dimensional\nstrictly mean convex $\\lambda$-hypersurfaces. As a corollary, we obtain a gap\ntheorem for closed $\\lambda$-hypersurfaces with $\\lambda\\le 0.$\n"}
{"abstract": "  We propose a Bayesian inference approach for static Origin-Destination\n(OD)-estimation in large-scale networked transit systems. The approach finds\nposterior distribution estimates of the OD-coefficients, which describe the\nrelative proportions of passengers travelling between origin and destination\nlocations, via a Hamiltonian Monte Carlo sampling procedure. We suggest two\ndifferent inference model formulations: the instantaneous-balance and\naverage-delay model. We discuss both models' sensitivity to various count\nobservation properties, and establish that the average-delay model is generally\nmore robust in determining the coefficient posteriors. The\ninstantaneous-balance model, however, requires lower resolution count\nobservations and produces comparably accurate estimates as the average-delay\nmodel, pending that count observations are only moderately interfered by trend\nfluctuations or the truncation of the observation window, and sufficient number\nof dispersed data records are available. We demonstrate that the Bayesian\nposterior distribution estimates provide quantifiable measures of the\nestimation uncertainty and prediction quality of the model, whereas the point\nestimates obtained from an alternative constrained quadratic programming\noptimisation approach only provide the residual errors between the predictions\nand observations. Moreover, the Bayesian approach proves more robust in scaling\nto high-dimensional underdetermined problems. The Bayesian\ninstantaneous-balance OD-coefficient posteriors are determined for the New York\nCity (NYC) subway network, based on several years of entry and exit count\nobservations recorded at station turnstiles across the network. The\naverage-delay model proves intractable on the real-world test scenario, given\nits computational time complexity and the incompleteness as well as coarseness\nof the turnstile records.\n"}
{"abstract": "  Thin films of topological insulators (TI) attract large attention because of\nexpected topological effects from the inter-surface hybridization of Dirac\npoints. However, these effects may be depleted by unexpectedly large energy\nsmearing $\\Gamma$ of surface Dirac points by the random potential of abundant\nCoulomb impurities. We show that in a typical TI film with large dielectric\nconstant $\\sim 50$ sandwiched between two low dielectric constant layers, the\nRytova-Chaplik-Entin-Keldysh modification of the Coulomb potential of a charge\nimpurity allows a larger number of the film impurities to contribute to\n$\\Gamma$. As a result, $\\Gamma$ is large and independent of the TI film\nthickness $d$ for $d > 5$ nm. In thinner films $\\Gamma$ grows with decreasing\n$d$ due to reduction of screening by the hybridization gap. We study the\nsurface conductivity away from the neutrality point and at the neutrality\npoint. In the latter case, we find the maximum TI film thickness at which the\nhybridization gap is still able to make a TI film insulating and allow\nobservation of the quantum spin Hall effect, $d_{\\max} \\sim 7$ nm.\n"}
{"abstract": "  This work proposes a time-efficient Natural Gradient Descent method, called\nTENGraD, with linear convergence guarantees. Computing the inverse of the\nneural network's Fisher information matrix is expensive in NGD because the\nFisher matrix is large. Approximate NGD methods such as KFAC attempt to improve\nNGD's running time and practical application by reducing the Fisher matrix\ninversion cost with approximation. However, the approximations do not reduce\nthe overall time significantly and lead to less accurate parameter updates and\nloss of curvature information. TENGraD improves the time efficiency of NGD by\ncomputing Fisher block inverses with a computationally efficient covariance\nfactorization and reuse method. It computes the inverse of each block exactly\nusing the Woodbury matrix identity to preserve curvature information while\nadmitting (linear) fast convergence rates. Our experiments on image\nclassification tasks for state-of-the-art deep neural architecture on CIFAR-10,\nCIFAR-100, and Fashion-MNIST show that TENGraD significantly outperforms\nstate-of-the-art NGD methods and often stochastic gradient descent in\nwall-clock time.\n"}
{"abstract": "  The end-to-end architecture has made promising progress in speech translation\n(ST). However, the ST task is still challenging under low-resource conditions.\nMost ST models have shown unsatisfactory results, especially in the absence of\nword information from the source speech utterance. In this study, we survey\nmethods to improve ST performance without using source transcription, and\npropose a learning framework that utilizes a language-independent universal\nphone recognizer. The framework is based on an attention-based\nsequence-to-sequence model, where the encoder generates the phonetic embeddings\nand phone-aware acoustic representations, and the decoder controls the fusion\nof the two embedding streams to produce the target token sequence. In addition\nto investigating different fusion strategies, we explore the specific usage of\nbyte pair encoding (BPE), which compresses a phone sequence into a\nsyllable-like segmented sequence. Due to the conversion of symbols, a segmented\nsequence represents not only pronunciation but also language-dependent\ninformation lacking in phones. Experiments conducted on the Fisher\nSpanish-English and Taigi-Mandarin drama corpora show that our method\noutperforms the conformer-based baseline, and the performance is close to that\nof the existing best method using source transcription.\n"}
{"abstract": "  In this paper, when the order of $\\theta$ is even, we prove that there exists\nno central difference sets in $A_2(m,\\theta)$ and establish some non-existence\nresults of central partial difference sets in $A_p(m,\\theta)$ with $p>2$. When\nthe order of $\\theta$ is odd, we construct central difference sets in\n$A_2(m,\\theta)$. Furthermore, we give some reduced linking systems of\ndifference sets in $A_2(m,\\theta)$ by using the difference sets we constructed.\nIn the case $p>2$, we construct Latin square type central partial difference\nsets in $A_p(m,\\theta)$ by a similar method.\n"}
{"abstract": "  Using the Laughlin's argument on a torus with two pin-holes, we numerically\ndemonstrate the bulk-edge correspondence of the fractional quantum Hall states\nof anyons, where the many-body Chern number predicts the discontinuities of the\ncenter-of-mass in the pumping phenomenon with edges. Through the process of\ntrading the magnetic flux for the statistical one (flux-attachment), the\ngeneral feature of the edge states remains unchanged as predicted by the Chern\nnumber while the ground state degeneracy is discretely changed. This supports\nthe stability of the quantum Hall edge states in the adiabatic heuristic\nprinciple.\n"}
{"abstract": "  Inertial measurement units (IMUs) increasingly function as a basic component\nof wearable sensor network (WSN)systems. IMU-based joint angle estimation (JAE)\nis a relatively typical usage of IMUs, with extensive applications. However,\nthe issue that IMUs move with respect to their original placement during JAE is\nstill a research gap, and limits the robustness of deploying the technique in\nreal-world application scenarios. In this study, we propose to detect and\ncorrect the IMU movement online in a relatively computationally lightweight\nmanner. Particularly, we first experimentally investigate the influence of IMU\nmovements. Second, we design the metrics for detecting IMU movements by\nmathematically formulating how the IMU movement affects the IMU measurements.\nThird, we determine the optimal thresholds of metrics by synthetic IMU data\nfrom a significantly amended simulation model. Finally, a correction method is\nproposed to correct the effects of IMU movements. We demonstrate our method on\nboth synthetic data and real-user data. The results demonstrate our method is a\npromising solution to detecting and correcting IMU movements during JAE.\n"}
{"abstract": "  It is non-trivial to recognize phase transitions and track dynamics inside a\nstochastic process because of its intrinsic stochasticity. In this paper, we\nemploy the deep learning method to classify the phase orders and predict the\ndamping coefficient of fluctuating systems under Langevin's description. As a\nconcrete set-up, we demonstrate this paradigm for the scalar condensation in\nQCD matter near the critical point, in which the order parameter of chiral\nphase transition can be characterized in a $1+1$-dimensional Langevin equation\nfor $\\sigma$ field. In a supervised learning manner, the Convolutional Neural\nNetworks(CNNs) accurately classify the first-order phase transition and\ncrossover based on $\\sigma$ field configurations with fluctuations. Noise in\nthe stochastic process does not significantly hinder the performance of the\nwell-trained neural network for phase order recognition. For mixed dynamics\nwith diverse dynamical parameters, we further devise and train the machine to\npredict the damping coefficients $\\eta$ in a broad range. The results show that\nit is robust to extract the dynamics from the bumpy field configurations.\n"}
{"abstract": "  We propose RPSRNet - a novel end-to-end trainable deep neural network for\nrigid point set registration. For this task, we use a novel $2^D$-tree\nrepresentation for the input point sets and a hierarchical deep feature\nembedding in the neural network. An iterative transformation refinement module\nin our network boosts the feature matching accuracy in the intermediate stages.\nWe achieve an inference speed of 12-15ms to register a pair of input point\nclouds as large as 250K. Extensive evaluation on (i) KITTI LiDAR odometry and\n(ii) ModelNet-40 datasets shows that our method outperforms prior\nstate-of-the-art methods - e.g., on the KITTI data set, DCP-v2 by1.3 and 1.5\ntimes, and PointNetLK by 1.8 and 1.9 times better rotational and translational\naccuracy respectively. Evaluation on ModelNet40 shows that RPSRNet is more\nrobust than other benchmark methods when the samples contain a significant\namount of noise and other disturbances. RPSRNet accurately registers point\nclouds with non-uniform sampling densities, e.g., LiDAR data, which cannot be\nprocessed by many existing deep-learning-based registration methods.\n"}
{"abstract": "  Different from previous modelings of self-propelled particles, we develop a\nmethod to propel the particles with a constant average velocity instead of a\nconstant force. This constant propulsion velocity (CPV) approach is validated\nby its agreement with the conventional constant propulsion force (CPF) approach\nin the flowing regime. However, the CPV approach shows its advantage of\naccessing quasistatic flows of yield stress fluids with a vanishing propulsion\nvelocity, while the CPF approach is usually unable to because of finite system\nsize. Taking this advantage, we realize the cyclic self-propulsion and study\nthe evolution of the propulsion force with propelled particle displacement,\nboth in the quasistatic flow regime. By mapping shear stress and shear rate to\npropulsion force and propulsion velocity, we find similar rheological behaviors\nof self-propelled systems to sheared systems, including the yield force gap\nbetween the CPF and CPV approaches, propulsion force overshoot,\nreversible-irreversible transition under cyclic propulsion, and propulsion\nbands in plastic flows. These similarities suggest the underlying connections\nbetween self-propulsion and shear, although they act on systems in different\nways.\n"}
{"abstract": "  In previous work, we constructed Fleming--Viot-type measure-valued diffusions\n(and diffusions on a space of interval partitions of the unit interval $[0,1]$)\nthat are stationary with the Poisson--Dirichlet laws with parameters\n$\\alpha\\in(0,1)$ and $\\theta\\geq 0$. In this paper, we complete the proof that\nthese processes resolve a conjecture by Feng and Sun (2010) by showing that the\nprocesses of ranked atom sizes (or of ranked interval lengths) of these\ndiffusions are members of a two-parameter family of diffusions introduced by\nPetrov (2009), extending a model by Ethier and Kurtz (1981) in the case\n$\\alpha=0$. The latter diffusions are continuum limits of up-down Chinese\nrestaurant processes.\n"}
{"abstract": "  In this article, the definition on the observe ability and its relation to\nthe signal detecting performance are studied systematically for the linear\ndiscrete-time(LDT) systems. Firstly, to define and analyze the observe ability\nfor the practical systems with the measured noise, six kinds of bounded noise\nmodels are classified. For the noise energy bounded case, the observability\nellipsoid and the image observability ellipsoid are defined by the state\nobserved error and then a novel concept on the LDT systems, called as the\nobserve ability, is proposed. Based on that, some theorems and properties about\nthe observe ability and the signal detecting performances are given and proven,\nand then the reason that to maximize the observe ability is to optimize the\nsignal detecting performances is established. Secondly, a dual relation between\nthe observability ellipsoid and the controllability ellipsoid, which volumes\nand radii are respectively with some inverse relations, is stated and proven.\nAccordingly, the analytical computing equations for the volume of the two\nobservability ellipsoids are got and some analytical shape factors of these\nellipsoids are deconstructed. Based on these effective compting for the\nvolumes, radii, and shape factors, analyzing and optimizing for the observe\nability can be carried out. Thirdly, to compare rationally the state observe\nability between the different systems or different system parameters, the\nnormalization of the output variables, the state variables, and the system\nmodels are discussed. Finally, some numerical experiments and their results\nshow the effectiveness of the computing and comparing methods for the observe\nability.\n"}
{"abstract": "  Personal and home sensors generate valuable information that could be used in\nSmart Cities. Unfortunately, typically, this data is locked out and used only\nby application/system developer. While vendors are to blame, one should\nconsider also the \"binary nature\" of data access. Specifically, either owner\nhas full control over her data (e.g. in a \"closed system\"), or she completely\nlooses control, when the data is \"opened\". In this context, we propose, a\nsemantic technologies-based, authorization and privacy control framework that\nenables user to maintain flexible, yet manageable data access control policies.\nThe proposed approach is described in detail, including implementation and\ntesting.\n"}
{"abstract": "  The Boltzmann equation is a fundamental kinetic equation that describes the\ndynamics of dilute gas. In this paper we study the regularity of both dynamical\nand steady Boltzmann equation in strictly convex domain with the\nCercignani-Lampis (C-L) boundary condition. The C-L boundary condition\ndescribes the intermediate reflection law between diffuse reflection and\nspecular reflection via two accommodation coefficients. We construct local\nweighted $C^1$ dynamical solution using repeated interaction through the\ncharacteristic. When we assume small fluctuation to the wall temperature and\naccommodation coefficients, we construct weighted $C^1$ steady solution.\n"}
{"abstract": "  Anomalous heat transport in one-dimensional nanostructures, such as nanotubes\nand nanowires, is a widely debated problem in condensed matter and statistical\nphysics, with contradicting pieces of evidence from experiments and\nsimulations. Using a comprehensive modeling approach, comprised of lattice\ndynamics and molecular dynamics simulations, we proved that the infinite length\nlimit of the thermal conductivity of a (10,0) single-wall carbon nanotube is\nfinite but this limit is reached only for macroscopic lengths due to thermal\nphonon mean free path of several millimeters. Our calculations showed that the\nextremely high thermal conductivity of this system at room temperature is\ndictated by quantum effects. Modal analysis showed that the divergent nature of\nthermal conductivity, observed in one-dimensional model systems, is suppressed\nin carbon nanotubes by anharmonic scattering channels provided by the flexural\nand optical modes with polarization in the plane orthogonal to the transport\ndirection.\n"}
{"abstract": "  Veracity of data posted on the microblog platforms has in recent years been a\nsubject of intensive study by professionals specializing in various fields of\ninformatics as well as sociology, particularly in the light of increasing\nimportance of online tools for news spreading. On Twitter and similar sites, it\nis possible to report on ongoing situations globally with minimal delay, while\nthe cost of such reporting remains negligible. One of the most important\nfeatures of this social network is that content delivery can be customized to\nallow users to focus only on news items covering subject matters they find\ninteresting. With this in mind, it becomes necessary to create verification\nmechanisms that can ascertain whether the claims made on Twitter can be taken\nseriously and prevent false content from spreading too far. This study\ndemonstrates an innovative System for verification of information that can\nfulfill the role described above. The System is comprised of four mutually\nconnected modules: a legacy module, a trustworthiness classifier; a module\nmanaging user authority, and a ranking procedure. All of the modules function\nwithin an integrated framework and jointly contribute to an accurate\nclassification of messages and authors. Effectiveness of the solution was\nevaluated empirically on a sample of Twitter users, with a strict 10-fold\nevaluation procedure applied for each module. The findings indicate that the\nsolution successfully meets the primary objectives of the study and performs\nits function as expected.\n"}
{"abstract": "  The detection of low-energy deposition in the range of sub-eV through\nionization using germanium (Ge) with a bandgap of $\\sim$0.7 eV requires\ninternal amplification of charge signal. This can be achieved through high\nelectric field which accelerates charge carriers to generate more charge\ncarriers. The minimum electric field required to generate internal charge\namplification is derived for different temperatures. A point contact Ge\ndetector provides extremely high electric field in proximity to the point\ncontact. We show the development of a planar point contact detector and its\nperformance. The field distribution is calculated for this planar point contact\ndetector. We demonstrate the required electric field can be achieved with a\npoint contact detector.\n"}
{"abstract": "  There has been significant recent interest in devising verification\ntechniques for learning-enabled controllers (LECs) that manage safety-critical\nsystems. Given the opacity and lack of interpretability of the neural policies\nthat govern the behavior of such controllers, many existing approaches enforce\nsafety properties through the use of shields, a dynamic monitoring and repair\nmechanism that ensures a LEC does not emit actions that would violate desired\nsafety conditions. These methods, however, have shown to have significant\nscalability limitations because verification costs grow as problem\ndimensionality and objective complexity increase. In this paper, we propose a\nnew automated verification pipeline capable of synthesizing high-quality safety\nshields even when the problem domain involves hundreds of dimensions, or when\nthe desired objective involves stochastic perturbations, liveness\nconsiderations, and other complex non-functional properties. Our key insight\ninvolves separating safety verification from neural controller, using\npre-computed verified safety shields to constrain neural controller training\nwhich does not only focus on safety. Experimental results over a range of\nrealistic high-dimensional deep RL benchmarks demonstrate the effectiveness of\nour approach.\n"}
{"abstract": "  The quantum approximate optimization algorithm (QAOA) is a hybrid\nquantum-classical variational algorithm that offers the potential to handle\ncombinatorial optimization problems. Introducing constraints in such\ncombinatorial optimization problems poses a significant challenge in the\nextensions of QAOA to support relevant larger-scale problems. This paper\nintroduces a quantum machine learning approach to learn the mixer Hamiltonian\nrequired to hard constrain the search subspace. We show that this method can be\nused for encoding any general form of constraints. One can directly plug the\nlearnt unitary into the QAOA framework using an adaptable ansatz. This\nprocedure gives the flexibility to control the depth of the circuit at the cost\nof the accuracy of enforcing the constraint, thus having immediate application\nin the Noisy Intermediate Scale Quantum (NISQ) era. We also develop an\nintuitive metric that uses Wasserstein distance to assess the performance of\ngeneral approximate optimization algorithms with/without constraints. Finally,\nusing this metric, we evaluate the performance of the proposed algorithm.\n"}
{"abstract": "  Recent approaches have exploited weaknesses in monolingual question answering\n(QA) models by adding adversarial statements to the passage. These attacks\ncaused a reduction in state-of-the-art performance by almost 50%. In this\npaper, we are the first to explore and successfully attack a multilingual QA\n(MLQA) system pre-trained on multilingual BERT using several attack strategies\nfor the adversarial statement reducing performance by as much as 85%. We show\nthat the model gives priority to English and the language of the question\nregardless of the other languages in the QA pair. Further, we also show that\nadding our attack strategies during training helps alleviate the attacks.\n"}
{"abstract": "  Ever-increasing interests to more accurate thermodynamic predictions of phase\ndiagrams motivate more reliable thermodynamic models to be developed. The\nModified Quasichemical Model within the two-sublattice Quadruplet Approximation\n(MQMQA) was thus established in well response to these interests and\nmotivations. However, the model still needs to be further improved in order to\nhave better thermodynamic predictions of arbitrary reciprocal solutions. The\npresent paper proposes a new and generic formalism to characterize the Gibbs\nenergy of the ternary reciprocal quadruplet within the framework of the MQMQA.\nThe new formalism circumvents the problem spawned from solving the singular\nmatrix of mass equations. Resultantly, energy landscapes of reciprocal\nsolutions can be better defined everywhere in reciprocal composition spaces.\nWith the current improvement, the MQMQA is believed to be one of the most\nreliable thermodynamic models for various types of solutions with or without\nshort-range ordering.\n"}
{"abstract": "  We apply variational autoencoders (VAE) to X-ray diffraction (XRD) data\nanalysis on both simulated and experimental thin-film data. We show that\ncrystal structure representations learned by a VAE reveal latent information,\nsuch as the structural similarity of textured diffraction patterns. While other\nartificial intelligence (AI) agents are effective at classifying XRD data into\nknown phases, a similarly conditioned VAE is uniquely effective at knowing what\nit does not know, rapidly identifying novel phases and mixtures. These\ncapabilities demonstrate that a VAE is a valuable AI agent for materials\ndiscovery and understanding XRD measurements both on-the-fly and during post\nhoc analysis.\n"}
{"abstract": "  Let $f$ be a polynomial over a global field $K$. For each $\\alpha$ in $K$ and\n$N$ in $\\mathbb{Z}_{\\geq 0}$ denote by $K_N(f,\\alpha)$ the arboreal field\n$K(f^{-N}(\\alpha))$ and by $D_N(f,\\alpha)$ its degree over $K$.\n  It is conjectured that $D_N(f,\\alpha)$ should grow as a double exponential\nfunction of $N$, unless $f$ is post-critically finite (PCF), in which case\nthere are examples like $D_N(x^2,\\alpha) \\leq 4^{N}$. There is evidence\nconditionally on Vojta's conjecture. However, before the present work, no\nunconditional non-trivial lower bound was known for post-critically infinite\n$f$. In the case $f$ is PCF, no non-trivial lower bound was known, not even\nunder Vojta's conjecture.\n  In this paper we give two simple methods that turn the finiteness of the\ncritical orbit into an exploitable feature, also in the post-critically\ninfinite case. First, assuming GRH for number fields, we establish for all PCF\npolynomials $f$ of degree at least $2$ and all $\\alpha$ outside of the critical\norbits of $f$, the existence of a positive constant $c(f,\\alpha)$ such that\n$$D_N(f,\\alpha) \\geq \\text{exp}(c(f,\\alpha)N), $$ for all $N$, which is sharp\nup to, possibly, improve the constant $c(f,\\alpha)$.\n  Second, we show unconditionally that if $f$ is post-critically infinite over\nany number field $K$ and unicritical, then, for each $\\alpha$ in $K$, there\nexists a positive constant $c(f,\\alpha)$ such that $$D_N(f,\\alpha) \\geq\n\\text{exp}(c(f,\\alpha)N), $$ for all $N$. The main input here is to work modulo\na suitably chosen prime and use a construction available for PCF unicritical\npolynomials with periodic critical orbit.\n"}
{"abstract": "  Beam quality optimization in mammography traditionally considers detection of\na target obscured by quantum noise on a homogenous background. It can be argued\nthat this scheme does not correspond well to the clinical imaging task because\nreal mammographic images contain a complex superposition of anatomical\nstructures, resulting in anatomical noise that may dominate over quantum noise.\nUsing a newly developed spectral mammography system, we measured the\ncorrelation and magnitude of the anatomical noise in a set of mammograms. The\nresults from these measurements were used as input to an observer-model\noptimization that included quantum noise as well as anatomical noise. We found\nthat, within this framework, the detectability of tumors and\nmicrocalcifications behaved very differently with respect to beam quality and\ndose. The results for small microcalcifications were similar to what\ntraditional optimization methods would yield, which is to be expected since\nquantum noise dominates over anatomical noise at high spatial frequencies. For\nlarger tumors, however, low-frequency anatomical noise was the limiting factor.\nBecause anatomical structure has similar energy dependence as tumor contrast,\noptimal x-ray energy was significantly higher and the useful energy region\nwider than traditional methods suggest. Measurements on a tissue phantom\nconfirmed these theoretical results. Furthermore, since quantum noise\nconstitutes only a small fraction of the noise, the dose could be reduced\nsubstantially without sacrificing tumor detectability. Exposure settings used\nclinically are therefore not necessarily optimal for this imaging task. The\nimpact of these findings on the mammographic imaging task as a whole is,\nhowever, at this stage unclear.\n"}
{"abstract": "  We demonstrate that the time evolution of the van Hove dynamical pair\ncorrelation function is governed by adiabatic forces that arise from the free\nenergy and by superadiabatic forces that are induced by the flow of the van\nHove function. The superadiabatic forces consist of drag, viscous, and\nstructural contributions, as occur in active Brownian particles, in liquids\nunder shear and in lane forming mixtures. For hard sphere liquids we present a\npower functional theory that predicts these universal force fields in\nquantitative agreement with our Brownian dynamics simulation results.\n"}
{"abstract": "  We present our current progress in the holographic computation of the\nscattering amplitude for Deeply Virtual Compton Scattering (DVCS) processe, as\na function of the Mandelstam invariant $t$. We show that it is possible to\ndescribe simultaneously the differential cross-section and total cross-section\nof DVCS data with a single holographic model for the pomeron. Using data from\nH1-ZEUS we obtained a $\\chi^2_{dof} \\sim 1.5$ for the best fit to the data.\n"}
{"abstract": "  We develop a low scale left-right symmetric model based on $SU(3)_C\\times\nSU(2)_L\\times SU(2)_R\\times U(1)_{B-L}\\times Z_2$ with a simplified Higgs\nsector consisting of only one bidoublet and one $SU(2)_R$ doublet. In this\nmodel, the tiny values of light neutrino masses are generated through an\ninverse-seesaw mechanism. We emphasize that in this setup, the tree-level\nflavor changing neutral current can be strongly suppressed, consistent with the\ncurrent experimental constraints. We show that the lightest $CP$-even Higgs\nboson, which is like the standard model Higgs boson, and the next lightest\nHiggs boson, $h'$, are generated from the neutral components of the bidoublet.\nWe show that the mass of the next lightest Higgs boson can be of an order a few\nhundred~\\text{GeVs}. We analyze the detection of $h'$ at the Large Hadron\nCollider (LHC) for a center-of-mass energy $\\sqrt{s}=14~\\text{TeV}$ and\nintegrated luminosity $L_{\\text{int}}=300~{\\text{fb}}^{-1}$ via di-Higgs\nchannel: $h'\\to hh\\to b\\bar{b}\\gamma\\gamma$ and also in the $ZZ$ channel:\n$h'\\to ZZ\\to 4\\ell~(\\ell=e,~\\mu)$ at an integrated luminosity\n$L_{\\text{int}}=3000~{\\text{fb}}^{-1}$. We consider three benchmark points for\nthis analysis with $m_{h'}=250~\\text{GeV},~400~\\text{GeV}$, and\n$600~\\text{GeV}$. We show that promising signals with good statistical\nsignificances can be obtained in di-Higgs channel, with $2\\gamma+2b$-jets final\nstates.\n"}
{"abstract": "  Broadcast consensus protocols (BCPs) are a model of computation, in which\nanonymous, identical, finite-state agents compute by sending/receiving global\nbroadcasts. BCPs are known to compute all number predicates in\n$\\mathsf{NL}=\\mathsf{NSPACE}(\\log n)$ where $n$ is the number of agents. They\ncan be considered an extension of the well-established model of population\nprotocols. This paper investigates execution time characteristics of BCPs. We\nshow that every predicate computable by population protocols is computable by a\nBCP with expected $\\mathcal{O}(n \\log n)$ interactions, which is asymptotically\noptimal. We further show that every log-space, randomized Turing machine can be\nsimulated by a BCP with $\\mathcal{O}(n \\log n \\cdot T)$ interactions in\nexpectation, where $T$ is the expected runtime of the Turing machine. This\nallows us to characterise polynomial-time BCPs as computing exactly the number\npredicates in $\\mathsf{ZPL}$, i.e. predicates decidable by log-space bounded\nrandomised Turing machine with zero-error in expected polynomial time where the\ninput is encoded as unary.\n"}
{"abstract": "  In this paper, we study an interesting combination of sleeping and\ncombinatorial stochastic bandits. In the mixed model studied here, at each\ndiscrete time instant, an arbitrary \\emph{availability set} is generated from a\nfixed set of \\emph{base} arms. An algorithm can select a subset of arms from\nthe \\emph{availability set} (sleeping bandits) and receive the corresponding\nreward along with semi-bandit feedback (combinatorial bandits).\n  We adapt the well-known CUCB algorithm in the sleeping combinatorial bandits\nsetting and refer to it as \\CSUCB. We prove -- under mild smoothness conditions\n-- that the \\CSUCB\\ algorithm achieves an $O(\\log (T))$ instance-dependent\nregret guarantee. We further prove that (i) when the range of the rewards is\nbounded, the regret guarantee of \\CSUCB\\ algorithm is $O(\\sqrt{T \\log (T)})$\nand (ii) the instance-independent regret is $O(\\sqrt[3]{T^2 \\log(T)})$ in a\ngeneral setting. Our results are quite general and hold under general\nenvironments -- such as non-additive reward functions, volatile arm\navailability, a variable number of base-arms to be pulled -- arising in\npractical applications. We validate the proven theoretical guarantees through\nexperiments.\n"}
{"abstract": "  We show that if we factor the long exact sequence in cohomology of a cofiber\nsequence of spectra into short exact sequences, then the $d_2$-differential in\nthe Adams spectral sequence of any one term is related in a precise way to\nYoneda composition with the 2-extension given by the complementary terms in the\nlong exact sequence. We use this to give a complete analysis of the Adams\nspectral sequence for the connective image-of-$J$ spectrum, finishing a\ncalculation that was begun by D. Davis in 1975.\n"}
{"abstract": "  Machine learning based data-driven technologies have shown impressive\nperformances in a variety of application domains. Most enterprises use data\nfrom multiple sources to provide quality applications. The reliability of the\nexternal data sources raises concerns for the security of the machine learning\ntechniques adopted. An attacker can tamper the training or test datasets to\nsubvert the predictions of models generated by these techniques. Data poisoning\nis one such attack wherein the attacker tries to degrade the performance of a\nclassifier by manipulating the training data.\n  In this work, we focus on label contamination attack in which an attacker\npoisons the labels of data to compromise the functionality of the system. We\ndevelop Gradient-based Data Subversion strategies to achieve model degradation\nunder the assumption that the attacker has limited-knowledge of the victim\nmodel. We exploit the gradients of a differentiable convex loss function\n(residual errors) with respect to the predicted label as a warm-start and\nformulate different strategies to find a set of data instances to contaminate.\nFurther, we analyze the transferability of attacks and the susceptibility of\nbinary classifiers. Our experiments show that the proposed approach outperforms\nthe baselines and is computationally efficient.\n"}
{"abstract": "  Two-component coupled Bose gas in a 1D optical lattice is examined. In\naddition to the postulated Mott insulator and superfluid phases, multiple\nbosonic components manifest spin degrees of freedom. Coupling of the components\nin the Bose gas leads to substantial change in the previously observed spin\nphases, giving rise to new effective spin Hamiltonian and unraveling remarkable\nspin correlations. The system exhibiting ferromagnetic and non-ferromagnetic\nspin phases for on-site intra-component interaction stronger than\ninter-component interaction switches from first-order to second-order phase\ntransition between the spin phases upon introduction of coupling, on which is\ndependent the transition width. For comparable on-site inter- and intra-\ncomponent interaction, with coupling, instead of one, two spin phases emerge\nwith a second-order phase transition. Exact diagonalization and Variational\nMonte Carlo (VMC) with stochastic minimization on Entangled Plaquette State\n(EPS) bestow a unique and enhanced perspective into the system beyond the scope\nof a mean-field treatment.\n"}
{"abstract": "  We define R-equivalence for group schemes over a semilocal ring and relate\nthis with rational properties. Two main cases are investigated: tori and\nisotropic semisimple simply connected group schemes where we show in certain\ncases that R-equivalence coincide with Karoubi-Villamayor equivalence and is\nalso related to the Kneser-Tits problem in this setting. Finally we construct\nspecialization maps for R-equivalence in the case of regular algebras\ncontaining a field.\n"}
{"abstract": "  Time-series representation learning is a fundamental task for time-series\nanalysis. While significant progress has been made to achieve accurate\nrepresentations for downstream applications, the learned representations often\nlack interpretability and do not expose semantic meanings. Different from\nprevious efforts on the entangled feature space, we aim to extract the\nsemantic-rich temporal correlations in the latent interpretable factorized\nrepresentation of the data. Motivated by the success of disentangled\nrepresentation learning in computer vision, we study the possibility of\nlearning semantic-rich time-series representations, which remains unexplored\ndue to three main challenges: 1) sequential data structure introduces complex\ntemporal correlations and makes the latent representations hard to interpret,\n2) sequential models suffer from KL vanishing problem, and 3) interpretable\nsemantic concepts for time-series often rely on multiple factors instead of\nindividuals. To bridge the gap, we propose Disentangle Time Series (DTS), a\nnovel disentanglement enhancement framework for sequential data. Specifically,\nto generate hierarchical semantic concepts as the interpretable and\ndisentangled representation of time-series, DTS introduces multi-level\ndisentanglement strategies by covering both individual latent factors and group\nsemantic segments. We further theoretically show how to alleviate the KL\nvanishing problem: DTS introduces a mutual information maximization term, while\npreserving a heavier penalty on the total correlation and the dimension-wise KL\nto keep the disentanglement property. Experimental results on various\nreal-world benchmark datasets demonstrate that the representations learned by\nDTS achieve superior performance in downstream applications, with high\ninterpretability of semantic concepts.\n"}
{"abstract": "  We consider the algebras generated by observables in quantum field theory\nlocalized in regions in the null plane. For a scalar free field theory, we show\nthat the one-particle structure can be decomposed into a continuous direct\nintegral of lightlike fibres, and the modular operator decomposes accordingly.\nThis implies that a certain form of QNEC is valid in free fields involving the\ncausal completions of half-spaces on the null plane (null cuts). We also\ncompute the relative entropy of null cut algebras with respect to the vacuum\nand some coherent states.\n"}
{"abstract": "  We have come to rely on industrial-scale cyber-physical systems more and more\nto manage tasks and machinery in safety-critical situations. Efficient,\nreliable fault identification and management has become a critical factor in\nthe design of these increasingly sophisticated and complex devices. Teams of\nco-operating software agents are one way to coordinate the flow of diagnostic\ninformation gathered during fault-finding. By wielding domain knowledge of the\nsoftware architecture used to construct the system, agents build and refine\ntheir beliefs about the location and root cause of faults. This paper examines\nhow agents constructed within the GORITE Multi-Agent Framework create and\nrefine their beliefs. We demonstrate three different belief structures\nimplemented within our Fault Diagnostic Engine, showing how each supports a\ndistinct aspect of the agent's reasoning. Using domain knowledge of the IEC\n61499 Function Block architecture, agents are able to examine and rigorously\nevaluate both individual components and entire subsystems.\n"}
{"abstract": "  Providing high-speed Internet service using satellite network has attracted\nresearchers from both academia and industry mainly due to the characteristics\nof Low Earth Orbit (LEO) satellite networks such as global coverage,\nscalability, and lower transmission delay. With the recent advancements in the\nSoftware-Defined Network (SDN), implementation of SDN in Non-Terrestrial\nNetworks (NTN) can help to achieve the set goals for 5G and beyond networks.\nSince satellite networks have a distinct architecture, some of the traditional\nprotocols no longer remain useful. Therefore, to satisfy the diverse Quality of\nService (QoS) requirements for a variety of applications, we propose a novel\nand centralized QoS-aware routing algorithm, called fybrrLink in which the\nglobal view of the network in SDN is utilized. We implement a modified\nBresenham's algorithm and Dijkstra's algorithm to find the optimal path in a\nsignificantly reduced computation time. Also, taking advantage of the\ndeterministic satellite constellation, we propose a flow rule transfer\nalgorithm and a topology monitoring algorithm. Further, fybrrLink is evaluated\nwith multiple NS3 simulations, and results confirm its supremacy over other\nstate-of-the-art algorithms.\n"}
{"abstract": "  The progress in Query-focused Multi-Document Summarization (QMDS) has been\nlimited by the lack of sufficient largescale high-quality training datasets. We\npresent two QMDS training datasets, which we construct using two data\naugmentation methods: (1) transferring the commonly used single-document\nCNN/Daily Mail summarization dataset to create the QMDSCNN dataset, and (2)\nmining search-query logs to create the QMDSIR dataset. These two datasets have\ncomplementary properties, i.e., QMDSCNN has real summaries but queries are\nsimulated, while QMDSIR has real queries but simulated summaries. To cover both\nthese real summary and query aspects, we build abstractive end-to-end neural\nnetwork models on the combined datasets that yield new state-of-the-art\ntransfer results on DUC datasets. We also introduce new hierarchical encoders\nthat enable a more efficient encoding of the query together with multiple\ndocuments. Empirical results demonstrate that our data augmentation and\nencoding methods outperform baseline models on automatic metrics, as well as on\nhuman evaluations along multiple attributes.\n"}
{"abstract": "  Let $G$ be a Wheeler graph and $r$ be the number of runs in a Burrows-Wheeler\nTransform of $G$, and suppose $G$ can be decomposed into $\\upsilon$\nedge-disjoint directed paths whose internal vertices each have in- and\nout-degree exactly 1. We show how to store $G$ in $O (r + \\upsilon)$ space such\nthat later, given a pattern $P$, in $O (|P| \\log \\log |G|)$ time we can count\nthe vertices of $G$ reachable by directed paths labelled $P$, and then report\nthose vertices in $O (\\log \\log |G|)$ time per vertex.\n"}
{"abstract": "  The Griddy Gibbs sampling was proposed by Ritter and Tanner (1992) as a\ncomputationally efficient approximation of the well-known Gibbs sampling\nmethod. The algorithm is simple and effective and has been used successfully to\naddress problems in various fields of applied science. However, the approximate\nnature of the algorithm has prevented it from being widely used: the Markov\nchains generated by the Griddy Gibbs sampling method are not reversible in\ngeneral, so the existence and uniqueness of its invariant measure is not\nguaranteed. Even when such an invariant measure uniquely exists, there was no\nestimate of the distance between it and the probability distribution of\ninterest, hence no means to ensure the validity of the algorithm as a means to\nsample from the true distribution.\n  In this paper, we show, subject to some fairly natural conditions, that the\nGriddy Gibbs method has a unique, invariant measure. Moreover, we provide $L^p$\nestimates on the distance between this invariant measure and the corresponding\nmeasure obtained from Gibbs sampling. These results provide a theoretical\nfoundation for the use of the Griddy Gibbs sampling method. We also address a\nmore general result about the sensitivity of invariant measures under small\nperturbations on the transition probability. That is, if we replace the\ntransition probability $P$ of any Monte Carlo Markov chain by another\ntransition probability $Q$ where $Q$ is close to $P$, we can still estimate the\ndistance between the two invariant measures. The distinguishing feature between\nour approach and previous work on convergence of perturbed Markov chain is that\nby considering the invariant measures as fixed points of linear operators on\nfunction spaces, we don't need to impose any further conditions on the rate of\nconvergence of the Markov chain.\n"}
{"abstract": "  Quantum detector tomography is a fundamental technique for calibrating\nquantum devices and performing quantum engineering tasks. In this paper, we\ndesign optimal probe states for detector estimation based on the minimum upper\nbound of the mean squared error (UMSE) and the maximum robustness. We establish\nthe minimum UMSE and the minimum condition number for quantum detectors and\nprovide concrete examples that can achieve optimal detector tomography. In\norder to enhance estimation precision, we also propose a two-step adaptive\ndetector tomography algorithm and investigate how this adaptive strategy can be\nused to achieve efficient estimation of quantum detectors. Moreover, the\nsuperposition of coherent states are used as probe states for quantum detector\ntomography and the estimation error is analyzed. Numerical results demonstrate\nthe effectiveness of both the proposed optimal and adaptive quantum detector\ntomography methods.\n"}
{"abstract": "  Nothing can physically travel faster than light in vacuum. This is why it has\nbeen considered that there is no \\v{C}erenkov radiation (\\v{C}R) without an\neffective refractive index due to some background field. In this Letter, we\ntheoretically predict \\v{C}R in vacuum from a spatiotemporally modulated\nboundary. We consider the modulation of traveling wave type and apply a uniform\nelectrostatic field on the boundary to generate electric dipoles. Since the\ninduced dipoles stick to the interface, they travel at the modulation speed.\nWhen the grating travels faster than light, it emits \\v{C}R. In order to\nquantitatively examine this argument, we need to calculate the field scattered\nat the boundary. We utilise a dynamical differential method, which we developed\nin the previous paper, to quantitatively evaluate the field distribution in\nsuch a situation. We can confirm that all scattered fields are evanescent if\nthe modulation speed is slower than light while some become propagating if the\nmodulation is faster than light.\n"}
{"abstract": "  We show that Hertzian particle contacts are the underlying cause of the\nas-yet-unexplained noninteger power laws in weakly nonlinear rheology. In the\nmedium amplitude oscillatory shear (MAOS) region, the cubic scaling of the\nleading order nonlinear shear stress ($\\sigma_\\mathrm{3} \\sim\n\\gamma_\\mathrm{0}^{m_\\mathrm{3}}$, $m_\\mathrm{3}=3$) is the standard\nexpectation. Expanding on the work by Natalia et al. [J. Rheol. 64 625-635\n(2020)], we report an extensive data set of noncubical, noninteger power law\nscalings $m_\\mathrm{3}$ for particle suspensions in two immiscible fluids with\na capillary attractive interaction, known as capillary suspensions. Here, we\nshow that distinct power law exponents are found for the storage and loss\nmoduli and these noninteger scalings occur at every secondary fluid\nconcentration for two different contact angles. These compelling results\nindicate that the noninteger scalings are related to the underlying\nmicrostructure of capillary suspensions. We show that the magnitude of the\nthird harmonic elastic stress scaling $m_\\mathrm{3,elastic}$ originates from\nHertzian-like contacts in combination with the attractive capillary force. The\nrelated third harmonic viscous stress scaling $m_\\mathrm{3,viscous}$ is, found\nto be associated with adhesive-controlled friction. These observations,\nconducted for a wide range of compositions, can help explain previous reports\nof noninteger scaling for materials involving particle contacts and offers a\nnew opportunity using the variable power law exponent of MAOS rheology to\nreveal the physics of particle bonds and friction in the rheological response\nunder low deformation instead of at very high shear rates.\n"}
{"abstract": "  Dot product latent space embedding is a common form of representation\nlearning in undirected graphs (e.g. social networks, co-occurrence networks).\nWe show that such models have problems dealing with 'intransitive' situations\nwhere A is linked to B, B is linked to C but A is not linked to C. Such\nsituations occur in social networks when opposites attract (heterophily) and in\nco-occurrence networks when there are substitute nodes (e.g. the presence of\nPepsi or Coke, but rarely both, in otherwise similar purchase baskets). We\npresent a simple expansion which we call the attract-repel (AR) decomposition:\na set of latent attributes on which similar nodes attract and another set of\nlatent attributes on which similar nodes repel. We demonstrate the AR\ndecomposition in real social networks and show that it can be used to measure\nthe amount of latent homophily and heterophily. In addition, it can be applied\nto co-occurrence networks to discover roles in teams and find substitutable\ningredients in recipes.\n"}
{"abstract": "  We present analysis of two type-I X-ray bursts observed by NuSTAR originating\nfrom the very faint transient neutron star low-mass X-ray binary GRS\n1741.9-2853 during a period of outburst in May 2020. We show that the\npersistent emission can be modeled as an absorbed, Comptonized blackbody in\naddition to Fe K$\\alpha$ emission which can be attributed to relativistic disk\nreflection. We measure a persistent bolometric, unabsorbed luminosity of\n$L_{\\mathrm{bol}}=7.03^{+0.04}_{-0.05}\\times10^{36}\\,\\mathrm{erg\\,s^{-1}}$,\nassuming a distance of 7 kpc, corresponding to an Eddington ratio of $4.5\\%$.\nThis persistent luminosity combined with light curve analysis leads us to infer\nthat the bursts were the result of pure He burning rather than mixed H/He\nburning. Time-resolved spectroscopy reveals that the bolometric flux of the\nfirst burst exhibits a double-peaked structure, placing the source within a\nsmall population of accreting neutron stars which exhibit multiple-peaked\ntype-I X-ray bursts. We find that the second, brighter burst shows evidence for\nphotospheric radius expansion (PRE) and that at its peak, this PRE event had an\nunabsorbed bolometric flux of\n$F_{\\mathrm{peak}}=2.94^{+0.28}_{-0.26}\\times10^{-8}\\,\\mathrm{erg\\,cm^{-2}\\,s^{-1}}$.\nThis yields a new distance estimate of $d=9.0\\pm0.5$ kpc, assuming that this\ncorresponds to the Eddington limit for pure He burning on the surface of a\ncanonical neutron star. Additionally, we performed a detailed timing analysis\nwhich failed to find evidence for quasiperiodic oscillations or burst\noscillations, and we place an upper limit of $16\\%$ on the rms variability\naround 589 Hz, the frequency at which oscillations have previously been\nreported.\n"}
{"abstract": "  A low-energy magnetic dipole $(M1)$ spin-scissors resonance (SSR) located\njust below the ordinary orbital scissors resonance (OSR) was recently predicted\nin deformed nuclei within the Wigner Function Moments (WFM) approach. We\nanalyze this prediction using fully self-consistent Skyrme Quasiparticle Random\nPhase Approximation (QRPA) method. Skyrme forces SkM*, SVbas and SG2 are\nimplemented to explore SSR and OSR in $^{160,162,164}$Dy and $^{232}$Th.\nAccuracy of the method is justified by a good description of M1 spin-flip giant\nresonance. The calculations show that isotopes $^{160,162,164}$Dy indeed have\nat 1.5-2.4 MeV (below OSR) $I^{\\pi}K=1^+1$ states with a large $M1$ spin\nstrength ($K$ is the projection of the total nuclear moment to the symmetry\nz-axis). These states are almost fully exhausted by $pp[411\\uparrow,\n411\\downarrow]$ and $nn[521\\uparrow, 521\\downarrow]$ spin-flip configurations\ncorresponding to $pp[2d_{3/2}, 2d_{5/2}]$ and $nn[2f_{5/2}, 2f_{7/2}]$\nstructures in the spherical limit. So the predicted SSR is actually reduced to\nlow-orbital (l=2,3) spin-flip states. Following our analysis and in\ncontradiction with WFM spin-scissors picture, deformation is not the principle\norigin of the low-energy spin $M1$ states but only a factor affecting their\nfeatures. The spin and orbital strengths are generally mixed and exhibit the\ninterference: weak destructive in SSR range and strong constructive in OSR\nrange. In $^{232}$Th, the $M1$ spin strength is found very small. Two groups of\n$I^{\\pi}=1^+$ states observed experimentally at 2.4-4 MeV in $^{160,162,164}$Dy\nand at 2-4 MeV in $^{232}$Th are mainly explained by fragmentation of the\norbital strength. Distributions of nuclear currents in QRPA states partly\ncorrespond to the isovector orbital-scissors flow but not to spin-scissors one.\n"}
{"abstract": "  Automatic identification of plant specimens from amateur photographs could\nimprove species range maps, thus supporting ecosystems research as well as\nconservation efforts. However, classifying plant specimens based on image data\nalone is challenging: some species exhibit large variations in visual\nappearance, while at the same time different species are often visually\nsimilar; additionally, species observations follow a highly imbalanced,\nlong-tailed distribution due to differences in abundance as well as observer\nbiases. On the other hand, most species observations are accompanied by side\ninformation about the spatial, temporal and ecological context. Moreover,\nbiological species are not an unordered list of classes but embedded in a\nhierarchical taxonomic structure. We propose a multimodal deep learning model\nthat takes into account these additional cues in a unified framework. Our\nDigital Taxonomist is able to identify plant species in photographs better than\na classifier trained on the image content alone, the performance gained is over\n6 percent points in terms of accuracy.\n"}
{"abstract": "  We investigate how well BERT performs on predicting factuality in several\nexisting English datasets, encompassing various linguistic constructions.\nAlthough BERT obtains a strong performance on most datasets, it does so by\nexploiting common surface patterns that correlate with certain factuality\nlabels, and it fails on instances where pragmatic reasoning is necessary.\nContrary to what the high performance suggests, we are still far from having a\nrobust system for factuality prediction.\n"}
{"abstract": "  In \\cite[Theorem 2.5]{Bac16} Bachiller proved that if $(G, \\cdot, \\circ)$ is\na brace of order the power of a prime $p$ and the rank of $(G,\\cdot)$ is\nsmaller than $p-1$, then the order of any element is the same in the additive\nand multiplicative group. This means that in this case the isomorphism type of\n$(G,\\circ)$ determines the isomorphism type of $(G,\\cdot)$. In this paper we\ncomplement Bachiller's result in two directions. In Theorem 2.2 we prove that\nif $(G, \\cdot, \\circ)$ is a brace of order the power of a prime $p$, then\n$(G,\\cdot)$ has small rank (i.e. $< p-1$) if and only if $(G,\\circ)$ has small\nrank. We also provide examples of groups of rank $p-1$ in which elements of\narbitrarily large order in the additive group become of prime order in the\nmultiplicative group. When the rank is larger, orders may increase.\n"}
{"abstract": "  In traffic management, it is a very important issue to shorten the response\ntime by detecting the incidents (accident, vehicle breakdown, an object falling\non the road, etc.) and informing the corresponding personnel. In this study, an\nanomaly detection framework for road junctions is proposed. The final judgment\nis based on the trajectories followed by the vehicles. Trajectory information\nis provided by vehicle detection and tracking algorithms on visual data\nstreamed from a fisheye camera. Deep learning algorithms are used for vehicle\ndetection, and Kalman Filter is used for tracking. To observe the trajectories\nmore accurately, the detected vehicle coordinates are transferred to the bird's\neye view coordinates using the lens distortion model prediction algorithm. The\nsystem determines whether there is an abnormality in trajectories by comparing\nhistorical trajectory data and instantaneous incoming data. The proposed system\nhas achieved 84.6% success in vehicle detection and 96.8% success in\nabnormality detection on synthetic data. The system also works with a 97.3%\nsuccess rate in detecting abnormalities on real data.\n"}
{"abstract": "  An \\emph{equitable $(q, r)$-tree-coloring} of a graph $G$ is a $q$-coloring\nof $G$ such that the subgraph induced by each color class is a forest of\nmaximum degree at most $r$ and the sizes of any two color classes differ by at\nmost $1.$ Let the \\emph{strong equitable vertex $r$-arboricity} of a graph $G,$\ndenoted by $va^\\equiv_r (G)$, be the minimum $p$ such that $G$ has an equitable\n$(q, r)$-tree-coloring for every $q\\geq p.$\n  The values of $va^\\equiv_1 (K_{n,n})$ were investigated by Tao and Lin and\nWu, Zhang, and Li where exact values of $va^\\equiv_1 (K_{n,n})$ were found in\nsome special cases. In this paper, we extend their results by giving the exact\nvalues of $va^\\equiv_1 (K_{n,n})$ for all cases. In the process, we introduce a\nnew function related to an equitable coloring and obtain a more general result\nby determining the exact value of each $va^\\equiv_1 (K_{m,n})$ and $va^\\equiv_1\n(G)$ where $G$ is a balanced complete $k$-partite graph $K_{n,\\ldots,n}.$\n"}
{"abstract": "  We propose a near-term quantum simulator based on the fluxonium qubits\ninductively coupled to form a chain. This system provides long coherence time,\nlarge anharmonicity, and strong coupling, making it suitable to study Ising\nspin models. At the half-flux quantum sweet spot, the system is described by\nthe transverse field Ising model (TFIM). We evaluate the propagation of qubit\nexcitations through the system. As disorder increases, the excitations become\nlocalized. A single qubit measurement using the circuit QED methods is\nsufficient to identify localization transition without introducing tunable\ncouplers. We argue that inductively coupled fluxoniums provide opportunities to\nstudy localization and many-body effects in highly coherent quantum systems.\n"}
{"abstract": "  First-principles calculations using density functional theory and two methods\nin comparison, Quantum ESPRESSO and Siesta, are done on large supercells which\ndescribe different placements of two identical adsorbed alkali metal atoms (of\neither Na, or K species) on the monolayer of boron carbide BC3. The energy of\nsingle-atom adsorption over the center of C6 ring, of the C4B2 hexagon and over\na boron atom have been preliminarily estimated, the effect of applying the\nGrimme D2 correction on the adsorption characteristics evaluated, and the\ncomparison of these results with available data discussed. The interaction of\ntwo identical Na or K atoms adsorbed at close enough distances (less than 10\nAngstrom) is negligible if the adsorption occurs at the opposite sides of the\nBC3 layer, but creates a steep repulsive potential at distances less than 8\nAngstrom if both atoms are adsorbed on the same side of the monolayer.\nRelaxation patterns resulting from the two K atoms being trapped at adjacent\nadsorption sites in the lattice are explained. The results suggest that the\ndensity of adsorbed K atoms on BC3 can be interestingly high.\n"}
{"abstract": "  Image denoising is often empowered by accurate prior information. In recent\nyears, data-driven neural network priors have shown promising performance for\nRGB natural image denoising. Compared to classic handcrafted priors (e.g.,\nsparsity and total variation), the \"deep priors\" are learned using a large\nnumber of training samples -- which can accurately model the complex image\ngenerating process. However, data-driven priors are hard to acquire for\nhyperspectral images (HSIs) due to the lack of training data. A remedy is to\nuse the so-called unsupervised deep image prior (DIP). Under the unsupervised\nDIP framework, it is hypothesized and empirically demonstrated that proper\nneural network structures are reasonable priors of certain types of images, and\nthe network weights can be learned without training data. Nonetheless, the most\neffective unsupervised DIP structures were proposed for natural images instead\nof HSIs. The performance of unsupervised DIP-based HSI denoising is limited by\na couple of serious challenges, namely, network structure design and network\ncomplexity. This work puts forth an unsupervised DIP framework that is based on\nthe classic spatio-spectral decomposition of HSIs. Utilizing the so-called\nlinear mixture model of HSIs, two types of unsupervised DIPs, i.e., U-Net-like\nnetwork and fully-connected networks, are employed to model the abundance maps\nand endmembers contained in the HSIs, respectively. This way, empirically\nvalidated unsupervised DIP structures for natural images can be easily\nincorporated for HSI denoising. Besides, the decomposition also substantially\nreduces network complexity. An efficient alternating optimization algorithm is\nproposed to handle the formulated denoising problem. Semi-real and real data\nexperiments are employed to showcase the effectiveness of the proposed\napproach.\n"}
{"abstract": "  We present experiments revealing the competing effect of quantum fluctuations\nand of a coherent seed in the dynamics of a spin-1 Bose-Einstein condensate,\nand discuss the relevance of a mean-field description of our system. We first\nexplore a near-equilibrium situation, where the mean-field equations can be\nlinearized around a fixed point corresponding to all atoms in the same Zeeman\nstate $m=0$. Preparing the system at this classical fixed point, we observe a\nreversible dynamics triggered by quantum fluctuations, which cannot be\nunderstood within a classical framework. We demonstrate that the classical\ndescription becomes accurate provided a coherent seed of a few atoms only is\npresent in the other Zeeman states $m= \\pm 1$. In a second regime characterized\nby a strong non-linearity of the mean-field equations, we observe a collapse\ndynamics driven by quantum fluctuations. This behavior cannot be accounted for\nby a classical description and persists for a large range of initial states. We\nshow that all our experimental results can be explained with a semi-classical\ndescription (truncated Wigner approximation), using stochastic classical\nvariables to model the quantum noise.\n"}
{"abstract": "  TypeShift is a tool for visualizing linguistic patterns in the timing of\ntyping production. Language production is a complex process which draws on\nlinguistic, cognitive and motor skills. By visualizing holistic trends in the\ntyping process, TypeShift aims to elucidate the often noisy information signals\nthat are used to represent typing patterns, both at the word-level and\ncharacter-level. It accomplishes this by enabling a researcher to compare and\ncontrast specific linguistic phenomena, and compare an individual typing\nsession to multiple group averages. Finally, although TypeShift was originally\ndesigned for typing data, it can easy be adapted to accommodate speech data, as\nwell. A web demo is available at https://angoodkind.shinyapps.io/TypeShift/.\nThe source code can be accessed at https://github.com/angoodkind/TypeShift.\n"}
{"abstract": "  This paper considers the use of instruments to identify and estimate private\nand social returns to education within a model of employer learning. What an\ninstrument identifies depends on whether it is hidden from, or transparent\n(i.e., observed) to, the employers. A hidden instrument identifies private\nreturns to education, and a transparent instrument identifies social returns to\neducation. We use variation in compulsory schooling laws across non-central and\ncentral municipalities in Norway to, respectively, construct hidden and\ntransparent instruments. We estimate a private return of 7.9%, of which 70% is\ndue to increased productivity and the remaining 30% is due to signaling.\n"}
{"abstract": "  We study X-ray and soft gamma-ray spectra from the hard state of the\naccreting black-hole binary MAXI J1820+070. We perform analysis of two joint\nspectra from NuSTAR and INTEGRAL, covering the range of 3--650 keV, and of an\naverage joint spectrum over the rise of the hard state, covering the 3--2200\nkeV range. The spectra are well modelled by Comptonization of soft seed\nphotons. However, the distributions of the scattering electrons are not purely\nthermal; we find they have substantial high-energy tails, well modelled as\npower laws. The photon tail in the average spectrum is detected well beyond the\nthreshold for electron-positron pair production, 511 keV. This allows us to\ncalculate the rate of the electron-positron pair production and put a lower\nlimit on the size of the source from pair equilibrium. At the fitted Thomson\noptical depth of the Comptonizing plasma, the limit is about 4 gravitational\nradii. If we adopt the sizes estimated by us from the reflection spectroscopy\nof $>$20 gravitational radii, the fractional pair abundance becomes much less\nthan unity. The low pair abundance is confirmed by the lack of both an\nannihilation feature and of a pair absorption cutoff above 511 keV in the\naverage spectrum.\n"}
{"abstract": "  We establish a connection between federated learning, a concept from machine\nlearning, and mean-field games, a concept from game theory and control theory.\nIn this analogy, the local federated learners are considered as the players and\nthe aggregation of the gradients in a central server is the mean-field effect.\nWe present federated learning as a differential game and discuss the properties\nof the equilibrium of this game. We hope this novel view to federated learning\nbrings together researchers from these two distinct areas to work on\nfundamental problems of large-scale distributed and privacy-preserving learning\nalgorithms.\n"}
{"abstract": "  It is known that nuclear deformation plays an important role in inducing the\nhalo structure in neutron-rich nuclei by mixing several angular momentum\ncomponents. While previous theoretical studies on this problem in the\nliterature assume axially symmetric deformation, we here consider non-axially\nsymmetric deformations. With triaxial deformation, the $\\Omega$ quantum number\nis admixed in a single-particle wave function, where $\\Omega$ is the projection\nof the single-particle angular momentum on the symmetric axis, and the halo\nstructure may arise even when it is absent with the axially symmetric\ndeformation. In this way, the area of halo nuclei may be extended when triaxial\ndeformation is considered. We demonstrate this idea using a deformed\nWoods-Saxon potential for nuclei with neutron number N=13 and 43.\n"}
{"abstract": "  We report the first quasioptical simulations of wave beams in a hot plasma\nusing the quasioptical code PARADE (PAraxial RAy DEscription) [Phys. Plasmas\n26, 072112 (2019)]. This code is unique in that it accounts for inhomogeneity\nof the dissipation-rate across the beam and mode conversion simultaneously. We\nshow that the dissipation-rate inhomogeneity shifts beams relative to their\ntrajectories in cold plasma and that the two electromagnetic modes are coupled\nvia this process, an effect that was ignored in the past. We also propose a\nsimplified approach to accounting for the dissipation-rate inhomogeneity. This\napproach is computationally inexpensive and simplifies analysis of actual\nexperiments.\n"}
{"abstract": "  Autonomous vehicles are conceived to provide safe and secure services by\nvalidating the safety standards as indicated by SOTIF-ISO/PAS-21448 (Safety of\nthe intended functionality). Keeping in this context, the perception of the\nenvironment plays an instrumental role in conjunction with localization,\nplanning and control modules. As a pivotal algorithm in the perception stack,\nobject detection provides extensive insights into the autonomous vehicle's\nsurroundings. Camera and Lidar are extensively utilized for object detection\namong different sensor modalities, but these exteroceptive sensors have\nlimitations in resolution and adverse weather conditions. In this work,\nradar-based object detection is explored provides a counterpart sensor modality\nto be deployed and used in adverse weather conditions. The radar gives complex\ndata; for this purpose, a channel boosting feature ensemble method with\ntransformer encoder-decoder network is proposed. The object detection task\nusing radar is formulated as a set prediction problem and evaluated on the\npublicly available dataset in both good and good-bad weather conditions. The\nproposed method's efficacy is extensively evaluated using the COCO evaluation\nmetric, and the best-proposed model surpasses its state-of-the-art counterpart\nmethod by $12.55\\%$ and $12.48\\%$ in both good and good-bad weather conditions.\n"}
{"abstract": "  Recent theoretical work has shown that the competition between coherent\nunitary dynamics and stochastic measurements, performed by the environment,\nalong wavefunction trajectories can give rise to transitions in the\nentanglement scaling. In this work, complementary to these previous studies, we\nanalyze a situation where the role of Hamiltonian and dissipative dynamics is\nreversed. We consider an engineered dissipation, which stabilizes an entangled\nphase of a quantum spin$-1/2$ chain, while competing single-particle or\ninteracting Hamiltonian dynamics induce a disentangled phase. Focusing on the\nsingle-particle unitary dynamics, we find that the system undergoes an\nentanglement transition from a logarithmic growth to an area law when the\ncompetition ratio between the unitary evolution and the non-unitary dynamics\nincreases. We evidence that the transition manifests itself in state-dependent\nobservables at a finite competition ratio for Hamiltonian and measurement\ndynamics. On the other hand, it is absent in trajectory-averaged steady-state\ndynamics, governed by a Lindblad master equation: although purely dissipative\ndynamics stabilizes an entangled state, for any non-vanishing Hamiltonian\ncontribution the system ends up irremediably in a disordered phase. In\naddition, a single trajectory analysis reveals that the distribution of the\nentanglement entropy constitutes an efficient indicator of the transition.\nComplementarily, we explore the competition of the dissipation with coherent\ndynamics generated by an interacting Hamiltonian, and demonstrate that the\nentanglement transition also occurs in this second model. Our results suggest\nthat this type of transition takes place for a broader class of Hamiltonians,\nunderlining its robustness in monitored open quantum many-body systems.\n"}
{"abstract": "  We establish connectedness criteria for graphs associated to monomials in\ncertain quotients of the mod 2 dual Steenrod algebra. We also investigate\nquestions about trees and Hamilton cycles in the context of these graphs.\nFinally, we improve upon a known connection between the graph theoretic\ninterpretation of the mod 2 dual Steenrod algebra and its structure as a Hopf\nalgebra.\n"}
{"abstract": "  Natural Language Inference (NLI) datasets contain examples with highly\nambiguous labels. While many research works do not pay much attention to this\nfact, several recent efforts have been made to acknowledge and embrace the\nexistence of ambiguity, such as UNLI and ChaosNLI. In this paper, we explore\nthe option of training directly on the estimated label distribution of the\nannotators in the NLI task, using a learning loss based on this ambiguity\ndistribution instead of the gold-labels. We prepare AmbiNLI, a trial dataset\nobtained from readily available sources, and show it is possible to reduce\nChaosNLI divergence scores when finetuning on this data, a promising first step\ntowards learning how to capture linguistic ambiguity. Additionally, we show\nthat training on the same amount of data but targeting the ambiguity\ndistribution instead of gold-labels can result in models that achieve higher\nperformance and learn better representations for downstream tasks.\n"}
{"abstract": "  Our work aimed at experimentally assessing the benefits of model ensembling\nwithin the context of neural methods for passage reranking. Starting from\nrelatively standard neural models, we use a previous technique named Fast\nGeometric Ensembling to generate multiple model instances from particular\ntraining schedules, then focusing or attention on different types of approaches\nfor combining the results from the multiple model instances (e.g., averaging\nthe ranking scores, using fusion methods from the IR literature, or using\nsupervised learning-to-rank). Tests with the MS-MARCO dataset show that model\nensembling can indeed benefit the ranking quality, particularly with supervised\nlearning-to-rank although also with unsupervised rank aggregation.\n"}
{"abstract": "  Present theories of irreversible energy losses and heat generation within\nLi-ion cells are unsatisfactory because they are not compatible with energy\nconservation. This work aims to provide a consistent theoretical treatment of\nenergy transport and losses in such devices. An energy conservation law is\nderived from the Doyle-Fuller-Newman (DFN) model of a Li-ion cell using a\nrigorous mathematical approach. The resulting law allows irreversible chemical\nenergy losses to be located to seven different regions of the cell, namely: (i)\nthe electrolyte, (ii) the anode particles, (iii) the cathode particles, (iv)\nthe solid parts of the anode (ohmic losses), (v) the solid parts of the cathode\n(ohmic losses), (vi) the surfaces of the anode particles (polarisation losses),\nand (vii) the surfaces of the cathode particles (polarisation losses).\nNumerical solutions to the DFN model are used to validate the conservation law\nin the cases of a drive cycle and constant current discharges, and to compare\nthe energy losses occurring in different locations. It is indicated how cell\ndesign can be improved, for a specified set of operating conditions, by\ncomparing the magnitude of energy losses in the different regions of the cell.\n"}
{"abstract": "  Propagation effects in the interstellar medium and intrinsic profile changes\ncan cause variability in the timing of pulsars, which limits the accuracy of\nfundamental science done via pulsar timing. One of the best timing pulsars, PSR\nJ1713+0747, has gone through two `dip' events in its dispersion measure (DM)\ntime series. If these events reflect real changes in electron column density,\nthey should lead to multiple imaging. We show that the events are are well\nfitted by an underdense corrugated sheet model, and look for associated\nvariability in the pulse profile using principal component analysis. We find\nthat there are transient pulse profile variations, but they vary in concert\nwith the dispersion measure, unlike what is expected from lensing due to a\ncorrugated sheet. The change is consistent in shape across profiles from both\nthe Greenbank and Arecibo radio observatories, and its amplitude appears to be\nachromatic across the 820-MHz, 1.4-GHz, and 2.3-GHz bands, again unlike\nexpected from interference between lensed images. This result is puzzling. We\nnote that some of the predicted lensing effects would need higher time and\nfrequency resolution data than used in this analysis. Future events appear\nlikely, and storing baseband data or keeping multiple time-frequency\nresolutions will allow more in-depth study of propagation effects and hence\nimprovements to pulsar timing accuracy.\n"}
{"abstract": "  Imitation Learning (IL) is an effective framework to learn visuomotor skills\nfrom offline demonstration data. However, IL methods often fail to generalize\nto new scene configurations not covered by training data. On the other hand,\nhumans can manipulate objects in varying conditions. Key to such capability is\nhand-eye coordination, a cognitive ability that enables humans to adaptively\ndirect their movements at task-relevant objects and be invariant to the\nobjects' absolute spatial location. In this work, we present a learnable action\nspace, Hand-eye Action Networks (HAN), that can approximate human's hand-eye\ncoordination behaviors by learning from human teleoperated demonstrations.\nThrough a set of challenging multi-stage manipulation tasks, we show that a\nvisuomotor policy equipped with HAN is able to inherit the key spatial\ninvariance property of hand-eye coordination and achieve zero-shot\ngeneralization to new scene configurations. Additional materials available at\nhttps://sites.google.com/stanford.edu/han\n"}
{"abstract": "  As the applications of deep learning models on edge devices increase at an\naccelerating pace, fast adaptation to various scenarios with varying resource\nconstraints has become a crucial aspect of model deployment. As a result, model\noptimization strategies with adaptive configuration are becoming increasingly\npopular. While single-shot quantized neural architecture search enjoys\nflexibility in both model architecture and quantization policy, the combined\nsearch space comes with many challenges, including instability when training\nthe weight-sharing supernet and difficulty in navigating the exponentially\ngrowing search space. Existing methods tend to either limit the architecture\nsearch space to a small set of options or limit the quantization policy search\nspace to fixed precision policies. To this end, we propose BatchQuant, a robust\nquantizer formulation that allows fast and stable training of a compact,\nsingle-shot, mixed-precision, weight-sharing supernet. We employ BatchQuant to\ntrain a compact supernet (offering over $10^{76}$ quantized subnets) within\nsubstantially fewer GPU hours than previous methods. Our approach,\nQuantized-for-all (QFA), is the first to seamlessly extend one-shot\nweight-sharing NAS supernet to support subnets with arbitrary ultra-low\nbitwidth mixed-precision quantization policies without retraining. QFA opens up\nnew possibilities in joint hardware-aware neural architecture search and\nquantization. We demonstrate the effectiveness of our method on ImageNet and\nachieve SOTA Top-1 accuracy under a low complexity constraint ($<20$ MFLOPs).\nThe code and models will be made publicly available at\nhttps://github.com/bhpfelix/QFA.\n"}
{"abstract": "  We show that two families of equations, the generalized inviscid\nProudman-Johnson equation, and the $r$-Hunter-Saxton equation (recently\nintroduced by Cotter et al.) coincide for a certain range of parameters. This\ngives a new geometric interpretation of these Proudman-Johnson equations as\ngeodesic equations of right invariant homogeneous $W^{1,r}$-Finsler metrics on\nthe diffeomorphism group. Generalizing a construction of Lenells for the\nHunter-Saxton equation, we analyze these equations using an isometry from the\ndiffeomorphism group to an appropriate subset of real-valued functions. Thereby\nwe show that the periodic case is equivalent to the geodesic equations on the\n$L^r$-sphere in the space of functions, and the non-periodic case is equivalent\nto a geodesic flow on a flat space. This allows us to give explicit solutions\nto these equations in the non-periodic case, and answer several questions of\nCotter et al. regarding their limiting behavior.\n"}
{"abstract": "  We work out the interplay between lowest-order perturbative computations in\nthe 't Hooft coupling, $g^2=g^2_{YM} N$, operator mixing, renormalization-group\n(RG) improved ultraviolet (UV) asymptotics of leading-order (LO)\nnonplanar/planar contributions to $2$-point correlators, and nonperturbative\nlarge-$N$ expansion of perturbatively massless QCD-like theories. As concrete\nexamples, we compute to the lowest perturbative order in $SU(N)$ YM theory the\nratios, $r_i$, of LO-nonplanar to planar contributions to the $2$-point\ncorrelators in the orthogonal basis in the coordinate representation of the\ngauge-invariant dimension-$8$ scalar operators and all the twist-$2$ operators.\nWe demonstrate that -- if $\\frac{\\gamma_0}{\\beta_0}$ has no LO-nonplanar\ncontribution, with $\\gamma_0$ and $\\beta_0$ the one-loop coefficients of the\nanomalous-dimension matrix and beta function respectively -- $r_i$ actually\ncoincides with the corresponding ratio in the large-$N$ expansion of the\nRG-improved UV asymptotics of the $2$-point correlators, provided that a\ncertain canonical nonresonant diagonal renormalization scheme exists for the\ncorresponding operators. Contrary to the aforementioned scalar operators, for\nthe first $10^3$ twist-$2$ operators we actually verify the above conditions,\nand we get the universal value $r_i=-\\frac{1}{N^2}$. Hence, nonperturbatively\nsuch $r_i$ must coincide with the UV asymptotics of the ratio of the glueball\nself-energy loop to the glueball tree contribution to the $2$-point correlators\nabove. As a consequence, the universality of $r_i$ reflects the universality of\nthe effective coupling in the nonperturbative large-$N$ YM theory for the\ntwist-$2$ operators in the coordinate representation.\n"}
{"abstract": "  We consider two variants of private stochastic online learning. The first\nvariant is differentially private stochastic bandits. Previously, Sajed and\nSheffet (2019) devised the DP Successive Elimination (DP-SE) algorithm that\nachieves the optimal $ O \\biggl(\\sum\\limits_{1\\le j \\le K: \\Delta_j >0} \\frac{\n\\log T}{ \\Delta_j} + \\frac{ K\\log T}{\\epsilon} \\biggr)$ problem-dependent\nregret bound, where $K$ is the number of arms, $\\Delta_j$ is the mean reward\ngap of arm $j$, $T$ is the time horizon, and $\\epsilon$ is the required privacy\nparameter. However, like other elimination style algorithms, it is not an\nanytime algorithm. Until now, it was not known whether UCB-based algorithms\ncould achieve this optimal regret bound. We present an anytime, UCB-based\nalgorithm that achieves optimality. Our experiments show that the UCB-based\nalgorithm is competitive with DP-SE. The second variant is the full information\nversion of private stochastic online learning. Specifically, for the problems\nof decision-theoretic online learning with stochastic rewards, we present the\nfirst algorithm that achieves an $ O \\left( \\frac{ \\log K}{ \\Delta_{\\min}} +\n\\frac{ \\log K}{\\epsilon} \\right)$ regret bound, where $\\Delta_{\\min}$ is the\nminimum mean reward gap. The key idea behind our good theoretical guarantees in\nboth settings is the forgetfulness, i.e., decisions are made based on a certain\namount of newly obtained observations instead of all the observations obtained\nfrom the very beginning.\n"}
{"abstract": "  We consider moderately interacting particle systems with singular interaction\nkernel and environmental noise. It is shown that the mollified empirical\nmeasures converge in strong norms to the unique (local) solutions of nonlinear\nFokker-Planck equations. The approach works for the Biot-Savart and repulsive\nPoisson kernels.\n"}
{"abstract": "  The variational method is a powerful approach to solve many-body quantum\nproblems non perturbatively. However, in the context of relativistic quantum\nfield theory (QFT), it needs to meet 3 seemingly incompatible requirements\noutlined by Feynman: extensivity, computability, and lack of UV sensitivity. In\npractice, variational methods break one of the 3, which translates into the\nneed to have an IR or UV cutoff. In this letter, I introduce a relativistic\nmodification of continuous matrix product states that satisfies the 3\nrequirements jointly in 1+1 dimensions. I apply it to the self-interacting\nscalar field, without UV cutoff and directly in the thermodynamic limit.\nNumerical evidence suggests the error decreases faster than any power law in\nthe number of parameters, while the cost remains only polynomial.\n"}
{"abstract": "  Results of Koebe (1936), Schramm (1992), and Springborn (2005) yield\nrealizations of $3$-polytopes with edges tangent to the unit sphere. Here we\nstudy the algebraic degrees of such realizations. This initiates the research\non constrained realization spaces of polytopes.\n"}
{"abstract": "  The superconducting gap structure of a topological crystalline insulator\n(TCI) candidate ZrRuAs ($T^{\\rm on}_{\\rm c}$ = 7.9(1) K) with a\nnoncentrosymmetric crystal structure has been investigated using muon spin\nrotation/relaxation ($\\mu$SR) measurements in transverse-field (TF) and\nzero-field (ZF) geometries. We also present the results of magnetization,\nelectrical resistivity and heat capacity measurements on ZrRuAs, which reveal\nbulk superconductivity below 7.9~K. The temperature dependence of the effective\npenetration depth obtained from the analysis of the TF-$\\mu$SR spectra below\n$T_{\\rm c}$ is well described by an isotropic $s$-wave gap model as also\ninferred from an analysis of the heat capacity in the superconducting state. ZF\n$\\mu$SR data do not show any significant change in the muon spin relaxation\nrate above and below the superconducting transition temperature indicating that\ntime-reversal symmetry is preserved in the superconducting state of this\nmaterial.\n"}
{"abstract": "  Despite their success in massive engineering applications, deep neural\nnetworks are vulnerable to various perturbations due to their black-box nature.\nRecent study has shown that a deep neural network can misclassify the data even\nif the input data is perturbed by an imperceptible amount. In this paper, we\naddress the robustness issue of neural networks by a novel close-loop control\nmethod from the perspective of dynamic systems. Instead of modifying the\nparameters in a fixed neural network architecture, a close-loop control process\nis added to generate control signals adaptively for the perturbed or corrupted\ndata. We connect the robustness of neural networks with optimal control using\nthe geometrical information of underlying data to design the control objective.\nThe detailed analysis shows how the embedding manifolds of state trajectory\naffect error estimation of the proposed method. Our approach can simultaneously\nmaintain the performance on clean data and improve the robustness against many\ntypes of data perturbations. It can also further improve the performance of\nrobustly trained neural networks against different perturbations. To the best\nof our knowledge, this is the first work that improves the robustness of neural\nnetworks with close-loop control.\n"}
{"abstract": "  Despite commercial airlines mandating masks, there have been multiple\ndocumented events of COVID-19 superspreading on flights. We used available data\nfrom three flights, including cabin layout and seat locations of infected and\nuninfected passengers, to suggest interventions to mitigate COVID-19\nsuperspreading events during air travel. Specifically, we studied: 1) London to\nHanoi with 201 passengers, including 13 secondary infections among passengers;\n2) Singapore to Hangzhou with 321 passengers, including 12 to 14 secondary\ninfections; 3) a non-superspreading event on a private jet in Japan with 9\npassengers and no secondary infections. We show that inclusion of passenger\nmovement better explains the infection spread patterns than conventional models\ndo. We also found that FFP2/N95 mask usage would have reduced infection by\n95-100%, while cloth masks would have reduced it by only 40-80%. This suggests\nthe need for more stringent guidelines to reduce aviation-related\nsuperspreading events of COVID-19.\n"}
{"abstract": "  We study the structure of Gromov-Hausdorff limits of sequences of Riemannian\nmanifolds $\\{(M_\\alpha^n,g_\\alpha)\\}_{\\alpha \\in A}$ whose Ricci curvature\nsatisfies a uniform Kato bound. We first obtain Mosco convergence of the\nDirichlet energies to the Cheeger energy and show that tangent cones of such\nlimits satisfy the $\\mathrm{RCD}(0,n)$ condition. When assuming a\nnon-collapsing assumption, we introduce a new family of monotone quantities,\nwhich allows us to prove that tangent cones are also metric cones. We then show\nthe existence of a well-defined stratification in terms of splittings of\ntangent cones. We finally prove volume convergence to the Hausdorff\n$n$-measure.\n"}
{"abstract": "  Many modern nanofabrication and imaging techniques require an ultra-quiet\nenvironment to reach optimal resolution. Isolation from ambient vibrations is\noften achieved by placing the sensitive instrument atop a massive block that\nfloats on air springs and is surrounded by acoustic barriers. Because typical\nbuilding noise drops off above 120 Hz, it is advantageous to raise the flexural\nresonance frequencies of the inertia block and instrument far above 120 Hz.\nHowever, it can be challenging to obtain a high fundamental frequency of the\nfloating block using a simple rectangular design. Here we design, construct,\nand characterize a vibration isolation system with a cylindrical inertia block,\nwhose lowest resonance frequency of 249 Hz shows good agreement between finite\nelement analysis simulation and directly measured modes. Our simulations show\nthat a cylindrical design can achieve higher fundamental resonance frequency\nthan a rectangular design of the same mass.\n"}
{"abstract": "  We consider the problem of predicting the covariance of a zero mean Gaussian\nvector, based on another feature vector. We describe a covariance predictor\nthat has the form of a generalized linear model, i.e., an affine function of\nthe features followed by an inverse link function that maps vectors to\nsymmetric positive definite matrices. The log-likelihood is a concave function\nof the predictor parameters, so fitting the predictor involves convex\noptimization. Such predictors can be combined with others, or recursively\napplied to improve performance.\n"}
{"abstract": "  In this paper, we propose a Riemannian smoothing steepest descent method to\nminimize a nonconvex and non-Lipschitz function on submanifolds. The\ngeneralized subdifferentials on Riemannian manifold and the Riemannian gradient\nsub-consistency are defined and discussed. We prove that any accumulation point\nof the sequence generated by the Riemannian smoothing steepest descent method\nis a stationary point associated with the smoothing function employed in the\nmethod, which is necessary for the local optimality of the original\nnon-Lipschitz problem. Under the Riemannian gradient sub-consistency condition,\nwe also prove that any accumulation point is a Riemannian limiting stationary\npoint of the original non-Lipschitz problem. Numerical experiments are\nconducted to demonstrate the efficiency of the proposed method.\n"}
{"abstract": "  The transport layer is ossified. With most of the research and deployment\nefforts in the past decade focussing on the Transmission Control Protocol (TCP)\nand its extensions, the QUIC standardization by the Internet Engineering Task\nForce (IETF) is to be finalized in early 2021. In addition to addressing the\nmost urgent issues of TCP, QUIC ensures its future extendibility and is\ndestined to drastically change the transport protocol landscape. In this work,\nwe present a first look at emerging protocols and their IETF standardization\nefforts beyond QUIC v1. While multiple proposed extensions improve on QUIC\nitself, Multiplexed Application Substrate over QUIC Encryption (MASQUE) as well\nas WebTransport present different approaches to address long-standing problems,\nand their interplay extends on QUIC's take to address transport layer\nossification challenges.\n"}
{"abstract": "  The formation and evolution of galaxies is known to be sensitive to tidal\nprocesses leading to intrinsic correlations between their shapes and\norientations. Such correlations can be measured to high significance today,\nsuggesting that cosmological information can be extracted from them. Among the\nmost pressing questions in particle physics and cosmology is the nature of dark\nmatter. If dark matter is self-interacting, it can leave an imprint on galaxy\nshapes. In this work, we investigate whether self-interactions can produce a\nlong-lasting imprint on intrinsic galaxy shape correlations. We investigate\nthis observable at low redshift ($z<0.4$) using a state-of-the-art suite of\ncosmological hydro-dynamical simulations where the dark matter model is varied.\nWe find that dark matter self-interactions induce a mass dependent suppression\nin the intrinsic alignment signal by up to 50\\% out to ten's of mega-parsecs,\nshowing that self-interactions can impact structure outside the very core of\nclusters. We find evidence that self-interactions have a scale-dependent impact\non the intrinsic alignment signal that is sufficiently different from\nsignatures introduced by differing baryonic physics prescriptions, suggesting\nthat it is detectable with up-coming all-sky surveys.\n"}
{"abstract": "  The problem of answering questions using knowledge from pre-trained language\nmodels (LMs) and knowledge graphs (KGs) presents two challenges: given a QA\ncontext (question and answer choice), methods need to (i) identify relevant\nknowledge from large KGs, and (ii) perform joint reasoning over the QA context\nand KG. In this work, we propose a new model, QA-GNN, which addresses the above\nchallenges through two key innovations: (i) relevance scoring, where we use LMs\nto estimate the importance of KG nodes relative to the given QA context, and\n(ii) joint reasoning, where we connect the QA context and KG to form a joint\ngraph, and mutually update their representations through graph neural networks.\nWe evaluate our model on QA benchmarks in the commonsense (CommonsenseQA,\nOpenBookQA) and biomedical (MedQA-USMLE) domains. QA-GNN outperforms existing\nLM and LM+KG models, and exhibits capabilities to perform interpretable and\nstructured reasoning, e.g., correctly handling negation in questions.\n"}
{"abstract": "  Sleep disorders can be a negative factor both for learning as for the mental\nand physical development of adolescents. It has been shown that, in many\npopulations, adolescents tend to have a poor sleep quality, and a very late\nchronotype. Furthermore, these features peak at adolescence, in the sense that\nadults tend to sleep better and have an earlier chronotype. But what happens\nwhen we consider adolescents in a population where already adults have poor\nsleep quality and a very late chronotype? We have conducted two non-clinical\nstudies in the city of Bariloche, Argentina aimed at measuring sleep quality,\nchronotype, and social jet lag, using the Pittsburgh and Munich questionnaires.\nThese were administered individually to groups of high school students, as well\nas to smaller samples of adults and preadolescents, in order to study\ndifferences between adolescents and these groups. The results show that in this\npopulation sleep quality is much poorer than in most other healthy populations\nrecorded elsewhere. Furthermore, sleep quality is consistently worse for\nadolescents than for the other groups. The difference with adults seems to be\ndue mainly to increased daytime sleepiness and sleep latency, whereas the\ndifference with preadolescents seems to be due mainly to shorter sleep\nduration. We also found that the chronotypes of all the groups are very late,\nwith a peak at an age between 18 and 24 ys. Social jet lag and sleep onset\nlatency are also large, and they peak at adolescence, which suggests that they\nmight be closely related to the large prevalence of poor sleep quality that we\nfind in adolescents.\n"}
{"abstract": "  Accurate 7Li(d,n)24He thermonuclear reaction rates are crucial for precise\nprediction of the primordial abundances of Lithium and Beryllium and to probe\nthe mysteries beyond fundamental physics and the standard cosmological model.\nHowever, uncertainties still exist in current reaction rates of 7Li(d,n)24He\nwidely used in Big Bang Nucleosynthesis (BBN) simulations. In this work, we\nreevaluate the 7Li(d,n)24He reaction rate using the latest data on the three\nnear-threshold 9Be excited states from experimental measurements. We present\nfor the first time uncertainties that are directly constrained by experiments.\nAdditionally, we take into account for the first time the contribution from the\nsubthreshold resonance at 16.671 MeV of 9Be. We obtain a 7Li(d,n)24He rate that\nis overall smaller than the previous estimation by about a factor of 60 at the\ntypical temperature of the onset of primordial nucleosynthesis. We implemented\nour new rate in BBN nucleosynthesis calculations, and we show that the new\nrates have a very limited impact on the final light element abundances in\nuniform density models. Typical abundance variations are in the order of\n0.002%. For nonuniform density BBN models, the predicted 7Li production can be\nincreased by 10% and the primordial production of light nuclides with mass\nnumber A>7 can be increased by about 40%. Our results confirm that the\ncosmological lithium problem remains a long-standing unresolved puzzle from the\nstandpoint of nuclear physics.\n"}
{"abstract": "  We investigate the performance of wall-modeled LES for external aerodynamics\nin the NASA Juncture Flow. We characterize the errors in the prediction of mean\nvelocity profiles and pressure coefficient for three different locations over\nthe aircraft: the upstream region of the fuselage, the wing-body juncture, and\nthe wing-body juncture close to the trailing-edge. The last two locations are\ncharacterized by strong mean-flow three-dimensionality and separation. The\nmessage conveyed by our error analysis is that WMLES performs as expected in\nregions where the flow resembles a zero-pressure-gradient flat plate boundary\nlayer. However, there is a clear decline of the current models in the presence\nof wing-body junctions and, more acutely, in separated zones. The slow\nconvergence to the solution in these regions renders the brute-force\ngrid-refinement approach to improve the accuracy of the solution unfeasible.\nThe results reported above pertain to the mean velocity profile predicted using\nthe typical grid resolution for external aerodynamics applications, i.e., 5--20\npoints per boundary-layer thickness. The prediction of the pressure coefficient\nis below 5\\% error for all grid sizes considered, even when boundary layers\nwere marginally resolved. The latter accuracy can be attributed to the\nouter-layer nature of the mean pressure, which becomes less sensitive to flow\ndetails within the turbulent boundary layer. Finally, we show that\nboundary-layer-conforming grids (i.e., grids maintaining a constant number of\npoints per boundary-layer thickness) allow for a more efficient distribution of\ngrid points and smaller errors. Our results suggest that novel modeling venues\nencompassing physical insights, together with numerical and gridding\nadvancements, must be exploited to attain predictions within the tolerance\nrequired for Certification by Analysis.\n"}
{"abstract": "  We show that there is an operator space notion of Lipschitz embeddability\nbetween operator spaces which is strictly weaker than its linear counterpart\nbut which is still strong enough to impose linear restrictions on operator\nspace structures. This shows that there is a nontrivial theory of nonlinear\ngeometry for operator spaces and it answers a question in [BCD21]. For that, we\nintroduce the operator space version of Lipschitz-free Banach spaces and prove\nseveral properties of it. In particular, we show that separable operator spaces\nsatisfy a sort of isometric Lipschitz-lifting property in the sense of G.\nGodefroy and N. Kalton. Gateaux differentiability of Lipschitz maps in the\noperator space category is also studied.\n"}
{"abstract": "  The intracluster medium (ICM) is expected to experience on average about\nthree passages of weak shocks with low sonic Mach numbers, $M\\lesssim 3$,\nduring the formation of galaxy clusters. Both protons and electrons could be\naccelerated to become high energy cosmic rays (CRs) at such ICM shocks via\ndiffusive shock acceleration (DSA). We examine the effects of DSA by multiple\nshocks on the spectrum of accelerated CRs by including {\\it in situ}\ninjection/acceleration at each shock, followed by repeated re-acceleration at\nsuccessive shocks in the test-particle regime. For simplicity, the accelerated\nparticles are assumed to undergo adiabatic decompression without energy loss\nand escape from the system, before they encounter subsequent shocks. We show\nthat in general the CR spectrum is flattened by multiple shock passages,\ncompared to a single episode of DSA, and that the acceleration efficiency\nincreases with successive shock passages. However, the decompression due to the\nexpansion of shocks into the cluster outskirts may reduce the amplification and\nflattening of the CR spectrum by multiple shock passages. The final CR spectrum\nbehind the last shock is determined by the accumulated effects of repeated\nre-acceleration by all previous shocks, but it is relatively insensitive to the\nordering of the shock Mach numbers. Thus multiple passages of shocks may cause\nthe slope of the CR spectrum to deviate from the canonical DSA power-law slope\nof the current shock.\n"}
{"abstract": "  The thermal properties of materials are critically important to various\ntechnologies and are increasingly the target of materials design efforts.\nHowever, it is only relatively recent advances in first-principles\ncomputational techniques that have enabled researchers to explore the\nmicroscopic mechanisms of thermal properties, such as thermal expansion. We use\nthe Gr\\\"{u}neisen theory of thermal expansion in combination with density\nfunctional calculations and the quasiharmonic approximation to uncover\nmechanisms of thermal expansion in PbTiO$_3$ thin-films in terms of elastic and\nvibrational contributions to the free energy. Surprisingly, we find that\nalthough the structural parameters of PbTiO$_3$ thin-films evolve with\ntemperature as if they are dominated by linear elasticity, PbTiO$_3$ thin-films\nare strongly anharmonic, with large changes in the elastic constants and\nGr\\\"{u}neisen parameters with both misfit strain and temperature. We show that\na fortuitous near-cancellation between different types of anharmonicity gives\nrise to this behavior. Our results illustrate the importance of high-order\nphonon-strain anharmonicity in determining the temperature-dependent structural\nparameters of PbTiO$_3$ thin-films, and highlight the complex manner in which\nthermal expansion, misfit strain and elastic and vibrational properties are\nintertwined.\n"}
{"abstract": "  Deep learning has become the dominant approach in coping with various tasks\nin Natural LanguageProcessing (NLP). Although text inputs are typically\nrepresented as a sequence of tokens, there isa rich variety of NLP problems\nthat can be best expressed with a graph structure. As a result, thereis a surge\nof interests in developing new deep learning techniques on graphs for a large\nnumberof NLP tasks. In this survey, we present a comprehensive overview onGraph\nNeural Networks(GNNs) for Natural Language Processing. We propose a new\ntaxonomy of GNNs for NLP, whichsystematically organizes existing research of\nGNNs for NLP along three axes: graph construction,graph representation\nlearning, and graph based encoder-decoder models. We further introducea large\nnumber of NLP applications that are exploiting the power of GNNs and summarize\nthecorresponding benchmark datasets, evaluation metrics, and open-source codes.\nFinally, we discussvarious outstanding challenges for making the full use of\nGNNs for NLP as well as future researchdirections. To the best of our\nknowledge, this is the first comprehensive overview of Graph NeuralNetworks for\nNatural Language Processing.\n"}
{"abstract": "  Two-dimensional angular correlation of annihilation radiation (2D-ACAR) and\nCompton scattering are both powerful techniques to investigate the bulk\nelectronic structure of crystalline solids through the momentum density of the\nelectrons. Here we apply both methods to a single crystal of Pd to study the\nelectron momentum density and the occupancy in the first Brillouin zone, and to\npoint out the complementary nature of the two techniques. To retrieve the 2D\nspectra from 1D Compton profiles, a new direct inversion method (DIM) is\nimplemented and benchmarked against the well-established Cormack's method. The\ncomparison of experimental spectra with first principles density functional\ntheory calculations of the electron momentum density and the two photon\nmomentum density clearly reveals the importance of positron probing effects on\nthe determination of the electronic structure. While the calculations are in\ngood agreement with the experimental data, our results highlight some\nsignificant discrepancies.\n"}
{"abstract": "  The prospects of using a reconfigurable intelligent surface (RIS) to aid\nwireless communication systems have recently received much attention. Among the\ndifferent use cases, the most popular one is where each element of the RIS\nscatters the incoming signal with a controllable phase-shift, without\nincreasing its power. In prior literature, this setup has been analyzed by\nneglecting the electromagnetic interference, consisting of the inevitable\nincoming waves from external sources. In this letter, we provide a physically\nmeaningful model for the electromagnetic interference that can be used as a\nbaseline when evaluating RIS-aided communications. The model is used to show\nthat electromagnetic interference has a non-negligible impact on communication\nperformance, especially when the size of the RIS grows large. When the direct\nlink is present (though with a relatively weak gain), the RIS can even reduce\nthe communication performance. Importantly, it turns out that the SNR grows\nquadratically with the number of RIS elements only when the spatial correlation\nmatrix of the electromagnetic interference is asymptotically orthogonal to that\nof the effective channel (including RIS phase-shifts) towards the intended\nreceiver. Otherwise, the SNR only increases linearly.\n"}
{"abstract": "  A dynamical symmetry is employed to determine the structure of the quantum\nnon-Markovian time-local master equation. Such a structure is composed from two\ncomponents: scalar kinetic coefficients and the standard quantum Markovian\noperator form. The kinetic coefficients are generally time-dependent and\nincorporate information on the kinematics and memory effects, while the\noperators manifest the dynamical symmetry. Specifically, we focus on\ntime-translation symmetric dynamics, where the Lindblad jump operators\nconstitute the eigenoperators of the free dynamics. This symmetry is motivated\nby thermodynamic microscopic considerations, where strict energy conservation\nbetween system and environment imposes the time-translation symmetry. The\nconstruction is generalized to other symmetries, and to driven quantum systems.\nThe formalism is illustrated by three exactly solvable non-Markovian models,\nwhere the exact reduced description exhibits a dynamical symmetric structure.\nThe formal structure of the master equation leads to a first principle\ncalculation of the exact kinetic coefficients. This opens the possibility to\nsimulate in a modular fashion non-Markovian dynamics.\n"}
{"abstract": "  We prove three theorems about the exact solutions of a generalized or\ninteracting Black-Scholes equation that explicitly includes arbitrage bubbles.\nThese arbitrage bubbles can be characterized by an arbitrage number $A_N(T)$.\nThe first theorem states that if $A_N(T) = 0$, then the solution at the\nmaturity of the interacting equation is identical to the solution of the free\nBlack-Scholes equation with the same initial interest rate $r$. The second\ntheorem states that if $A_N(T) \\ne 0$, the solution can be expressed in terms\nof all higher derivatives of solutions to the free Black-Scholes equation with\nthe initial interest rate $r$. The third theorem states that whatever the\narbitrage number is, the solution is a solution to the free Black-Scholes\nequation with a variable interest rate $r(\\tau) = r + (1/\\tau) A_N(\\tau)$.\nAlso, we show, by using the Feynman-Kac theorem, that for the special case of a\nCall contract, the exact solution for a Call with strike price $K$ is\nequivalent to the usual Call solution to the Black-Scholes equation with strike\nprice $\\tilde{K} = K e^{-A_N(T)}$.\n"}
{"abstract": "  In this paper, we study the bandits with knapsacks (BwK) problem and develop\na primal-dual based algorithm that achieves a problem-dependent logarithmic\nregret bound. The BwK problem extends the multi-arm bandit (MAB) problem to\nmodel the resource consumption associated with playing each arm, and the\nexisting BwK literature has been mainly focused on deriving asymptotically\noptimal distribution-free regret bounds. We first study the primal and dual\nlinear programs underlying the BwK problem. From this primal-dual perspective,\nwe discover symmetry between arms and knapsacks, and then propose a new notion\nof sub-optimality measure for the BwK problem. The sub-optimality measure\nhighlights the important role of knapsacks in determining algorithm regret and\ninspires the design of our two-phase algorithm. In the first phase, the\nalgorithm identifies the optimal arms and the binding knapsacks, and in the\nsecond phase, it exhausts the binding knapsacks via playing the optimal arms\nthrough an adaptive procedure. Our regret upper bound involves the proposed\nsub-optimality measure and it has a logarithmic dependence on length of horizon\n$T$ and a polynomial dependence on $m$ (the numbers of arms) and $d$ (the\nnumber of knapsacks). To the best of our knowledge, this is the first\nproblem-dependent logarithmic regret bound for solving the general BwK problem.\n"}
{"abstract": "  Transferability estimation has been an essential tool in selecting a\npre-trained model and the layers of it to transfer, so as to maximize the\nperformance on a target task and prevent negative transfer. Existing estimation\nalgorithms either require intensive training on target tasks or have\ndifficulties in evaluating the transferability between layers. We propose a\nsimple, efficient, and effective transferability measure named TransRate. With\nsingle pass through the target data, TransRate measures the transferability as\nthe mutual information between the features of target examples extracted by a\npre-trained model and labels of them. We overcome the challenge of efficient\nmutual information estimation by resorting to coding rate that serves as an\neffective alternative to entropy. TransRate is theoretically analyzed to be\nclosely related to the performance after transfer learning. Despite its\nextraordinary simplicity in 10 lines of codes, TransRate performs remarkably\nwell in extensive evaluations on 22 pre-trained models and 16 downstream tasks.\n"}
{"abstract": "  Leveraging the advances of natural language processing, most recent scene\ntext recognizers adopt an encoder-decoder architecture where text images are\nfirst converted to representative features and then a sequence of characters\nvia `direct decoding'. However, scene text images suffer from rich noises of\ndifferent sources such as complex background and geometric distortions which\noften confuse the decoder and lead to incorrect alignment of visual features at\nnoisy decoding time steps. This paper presents I2C2W, a novel scene text\nrecognizer that is accurate and tolerant to various noises in scenes. I2C2W\nconsists of an image-to-character module (I2C) and a character-to-word module\n(C2W) which are complementary and can be trained end-to-end. I2C detects\ncharacters and predicts their relative positions in a word. It strives to\ndetect all possible characters including incorrect and redundant ones based on\ndifferent alignments of visual features without the restriction of time steps.\nTaking the detected characters as input, C2W learns from character semantics\nand their positions to filter out incorrect and redundant detection and produce\nthe final word recognition. Extensive experiments over seven public datasets\nshow that I2C2W achieves superior recognition performances and outperforms the\nstate-of-the-art by large margins on challenging irregular scene text datasets.\n"}
{"abstract": "  Even though superconductivity has been studied intensively for more than a\ncentury, the vast majority of superconductivity research today is carried out\nin nearly the same manner as decades ago. That is, each study tends to focus on\nonly a single material or small subset of materials, and discoveries are made\nmore or less serendipitously. Recent increases in computing power, novel\nmachine learning algorithms, and improved experimental capabilities offer new\nopportunities to revolutionize superconductor discovery. These will enable the\nrapid prediction of structures and properties of novel materials in an\nautomated, high-throughput fashion and the efficient experimental testing of\nthese predictions. Here, we review efforts to use machine learning to attain\nthis goal.\n"}
{"abstract": "  A numerical study of laminar flow through symmetric and slightly asymmetric\nsudden expansion, of expansion ratio 1:3, in channels with increasing cross\nsection, is carried out using two different approaches - Conventional CFD and\nLattice Boltzmann Method. The effect of divergence of walls of the channel,\nafter a sudden expansion, on the symmetry of flow and recirculation is studied\nfor various Reynolds numbers. It is seen that the angles of the walls play an\nactive role in disrupting the symmetry of flow. For non-parallel walls, the\nsymmetry breaking bifurcation phenomenon no longer exists and the loss of\nsymmetry is a gradual process. The effect of asymmetry of geometry on flow is\nalso studied by considering two types of asymmetry - First type is at the plane\nof expansion where the steps on either side are of unequal heights, while the\nsecond one deals the walls of the channel are at a different inclination with\nthe direction of inflow. The present study reveals that small asymmetry impedes\nthe sharp transition and thus smoothens the pitchfork bifurcation. Increased\nasymmetry finally leads to the unique solution of the governing Equations,\nindicating complete loss of bifurcation pattern.\n"}
{"abstract": "  We study $\\bar{Q}Q\\bar{q}q$ and $\\bar{Q}qQ\\bar{q}$ states as mixed states in\nQCD sum rules. By calculating the two-point correlation functions of pure\nstates of their corresponding currents, we review the mass and coupling\nconstant predictions of $J^{PC}=1^{++}$, $1^{--}$, $1^{-+}$ states. By\ncalculating the two-point mixed correlation functions of $\\bar{Q}Q\\bar{q}q$ and\n$\\bar{Q}qQ\\bar{q}$ currents, and we estimate the mass and coupling constants of\nthe corresponding ``physical state\" that couples to both $\\bar{Q}Q\\bar{q}q$ and\n$\\bar{Q}qQ\\bar{q}$ currents. Our results suggest that $1^{++}$ states are more\nlikely mixing from $\\bar{Q}Q\\bar{q}q$ and $\\bar{Q}qQ\\bar{q}$ components, while\nfor $1^{--}$ and $1^{-+}$ states, there is less mixing between\n$\\bar{Q}Q\\bar{q}q$ and $\\bar{Q}qQ\\bar{q}$. Our results suggest the $Y$ series\nof states have more complicated components.\n"}
{"abstract": "  The possibility of detection of 5.5 MeV and 14.4 keV solar axions by\nobserving axion-induced nuclear and atomic transitions is investigated. The\npresence of nuclear transitions between spin-orbit partners can be manifested\nby the subsequent de-excitation via gamma ray emissions. The transition rates\ncan also be studied in the context of radiative axion absorption by a nucleus.\nThe elementary interaction is obtained in the context of the axion-quark\ncouplings predicted by existing axion models. Then these couplings will be\ntransformed to the nucleon level utilizing reasonable existing models, which\nlead to effective transition operators. Using these operators we calculate the\nneeded nuclear matrix elements employing wave functions obtained in the context\nof the nuclear shell model. With these ingredients, we discuss possibilities of\nexperimental observation of the axion-induced nuclear gamma-rays. In the second\npart, we will examine the axion-induced production of X-rays (axion-photon\nconversion) or ionization from deeply bound electron orbits. In this case, the\naxion electron coupling is predicted by existing axion models, no\nrenormalization is needed. The experimental signal is the observation of\ndirectly produced electrons and/or the emission of hard X-rays and Auger\nelectrons, following the de-excitation of the final atom. Critical discussion\nis made on the experimental feasibility of detecting the solar axions by using\nmulti-ton scale NaI detectors.\n"}
{"abstract": "  Earth's magnetosphere is vital for today's technologically dependent society.\nTo date, numerous design studies have been conducted and over a dozen science\nmissions have own to study the magnetosphere. However, a majority of these\nsolutions relied on large monolithic satellites, which limited the spatial\nresolution of these investigations, as did the technological limitations of the\npast. To counter these limitations, we propose the use of a satellite swarm\ncarrying numerous and distributed payloads for magnetospheric measurements. Our\nmission is named APIS (Applications and Potentials of Intelligent Swarms),\nwhich aims to characterize fundamental plasma processes in the Earth's\nmagnetosphere and measure the effect of the solar wind on our magnetosphere. We\npropose a swarm of 40 CubeSats in two highly-elliptical orbits around the\nEarth, which perform radio tomography in the magnetotail at 8-12 Earth Radii\n(RE) downstream, and the subsolar magnetosphere at 8-12RE upstream. In\naddition, in-situ measurements of the magnetic and electric fields, plasma\ndensity will be performed by on-board instruments.\n  In this article, we present an outline of previous missions and designs for\nmagnetospheric studies, along with the science drivers and motivation for the\nAPIS mission. Furthermore, preliminary design results are included to show the\nfeasibility of such a mission. The science requirements drive the APIS mission\ndesign, the mission operation and the system requirements. In addition to the\nvarious science payloads, critical subsystems of the satellites are\ninvestigated e.g., navigation, communication, processing and power systems. We\nsummarize our findings, along with the potential next steps to strengthen our\ndesign study.\n"}
{"abstract": "  Decline in gait features is common in older adults and an indicator of\ndisability and mortality. Cortical control of gait, specifically in the\npre-frontal cortex as measured by functional near infrared spectroscopy\n(fNIRS), during dual task walking has shown to be moderated by age, gender,\ncognitive status, and various age-related disease conditions. In this study, we\ndevelop classification models using machine learning methods to classify active\nwalking tasks in older adults based on fNIRS signals into either\nSingle-Task-Walk (STW) or Dual-Task-Walk (DTW) conditions. In this study, we\ndevelop classification models using machine learning methods to classify active\nwalking tasks in older adults based on fNIRS signals into either single-task\nwalking (STW) or dual-task walking (DTW). The fNIRS measurements included\noxyhemoglobin (HbO2) and deoxyhemoglobin (Hb) signals obtained from prefrontal\ncortex (PFC) of the subject performing on the ground active walking tasks with\nor without a secondary cognitive task. We extract the fNIRS-related features by\ncalculating the minimum, maximum, mean, skewness and kurtosis values of Hb and\nHbo2 signals. We then use feature encoding to map the values into binary space.\nUsing these features, we apply and evaluate various machine learning methods\nincluding logistic regression (LR), decision tree (DT), support vector machine\n(SVM), k-nearest neighbors (kNN), multilayer perceptron (MLP), and Random\nForest (RF). Results showed that the machine learning models can achieve around\n97\\% classification accuracy.\n"}
{"abstract": "  As data-driven methods are deployed in real-world settings, the processes\nthat generate the observed data will often react to the decisions of the\nlearner. For example, a data source may have some incentive for the algorithm\nto provide a particular label (e.g. approve a bank loan), and manipulate their\nfeatures accordingly. Work in strategic classification and decision-dependent\ndistributions seeks to characterize the closed-loop behavior of deploying\nlearning algorithms by explicitly considering the effect of the classifier on\nthe underlying data distribution. More recently, works in performative\nprediction seek to classify the closed-loop behavior by considering general\nproperties of the mapping from classifier to data distribution, rather than an\nexplicit form. Building on this notion, we analyze repeated risk minimization\nas the perturbed trajectories of the gradient flows of performative risk\nminimization. We consider the case where there may be multiple local minimizers\nof performative risk, motivated by situations where the initial conditions may\nhave significant impact on the long-term behavior of the system. We provide\nsufficient conditions to characterize the region of attraction for the various\nequilibria in this settings. Additionally, we introduce the notion of\nperformative alignment, which provides a geometric condition on the convergence\nof repeated risk minimization to performative risk minimizers.\n"}
{"abstract": "  For $1<p\\leq q$ we show that the Poisson semigroup $e^{-t\\sqrt{-\\Delta}}$ on\nthe $n$-sphere is hypercontractive from $L^{p}$ to $L^{q}$ in dimensions $n\n\\leq 3$ if and only if $e^{-t\\sqrt{n}} \\leq \\sqrt{\\frac{p-1}{q-1}}$. We also\nshow that the equivalence fails in large dimensions.\n"}
{"abstract": "  Binary millisecond pulsars (MSPs) are believed to have descended from\nlow-mass X-ray binaries (LMXBs), which have experienced substantial mass\ntransfer and tidal circularization. Therefore, they should have very circular\norbits. However, the discovery of several eccentric binary MSPs (with\neccentricity $e\\sim 0.01-0.1$) challenges this standard picture. Three models\nhave been proposed thus far based on accretion-induced collapse of massive\nwhite dwarfs (WDs), neutron star-strange star transition, and formation of\ncircumbinary disks. All of them are subject to various uncertainties, and are\nnot entirely consistent with observations. Here we propose an alternative model\ntaking into account the influence of thermonuclear flashes on proto-WDs. We\nassume that the flashes lead to asymmetrical mass ejection, which imparts a\nmild kick on the proto-WDs. By simulating orbital changes of binary MSPs with\nmultiple shell flashes, we show that it is possible to reproduce the observed\neccentricities, provided that the kick velocities are around a few kms$^{-1}$.\n"}
{"abstract": "  Message passing neural networks have become a method of choice for learning\non graphs, in particular the prediction of chemical properties and the\nacceleration of molecular dynamics studies. While they readily scale to large\ntraining data sets, previous approaches have proven to be less data efficient\nthan kernel methods. We identify limitations of invariant representations as a\nmajor reason and extend the message passing formulation to rotationally\nequivariant representations. On this basis, we propose the polarizable atom\ninteraction neural network (PaiNN) and improve on common molecule benchmarks\nover previous networks, while reducing model size and inference time. We\nleverage the equivariant atomwise representations obtained by PaiNN for the\nprediction of tensorial properties. Finally, we apply this to the simulation of\nmolecular spectra, achieving speedups of 4-5 orders of magnitude compared to\nthe electronic structure reference.\n"}
{"abstract": "  This work consists in the theorical development on the analysis of the\nThermodynamic Laws and thermodynamic systems in relative motion, according to\nthe laws of Classical Mechanics. The difference of this work for many of the\nliterature is the methodology used. We follow a strictly mathematical line,\ninstead of a phenomenological line. We proved that the First and Second Laws of\nThermodynamics in non-inertial frames remain invariant under any spatial\ntranslation between reference frames in motion. From the results obtained, we\nshow the transformation of the main thermodynamic quantities as temperature,\npressure, chemical potential, internal energy, heat, state equation for an\nideal gas and efficiency of Carnot's machine are also invariant under any\ntransformation between non-inertial reference frames.\n"}
{"abstract": "  We introduce a simple and concrete way of visualizing in three dimensions a\n\"flat\" Klein bottle -- one whose local intrinsic geometry is the same as that\nof a flat plane -- which preserves most its topological and geometric\nstructure. Concretely, the flatness property means that a small patch of the\nsurface around any point can be flattened to a patch of the plane without\nstretching or compressing. Thus we can use the medium of curved-crease origami\nwith inelastic film to make a model which, except for its self-intersections,\nnecessarily has the flatness property, even along its folded edges. As such,\nthe sculpture presented here illustrates both the flatness, and, through its\ncoloring, the non-orientability of a Klein bottle.\n"}
{"abstract": "  In this work we investigate the short-term variations in air quality\nemissions, attributed to the prevention measures, applied in different cities,\nto mitigate the COVID-19 spread. In particular, we emphasize on the\nconcentration effects regarding specific pollutant gases, such as carbon\nmonoxide (CO), ozone (O3), nitrogen dioxide (NO2) and sulphur dioxide (SO2).\nThe assessment of the impact of lockdown on air quality focused on four\nEuropean Cities (Athens, Gladsaxe, Lodz and Rome). Available data on pollutant\nfactors were obtained using global satellite observations. The level of the\nemployed prevention measures is employed using the Oxford COVID-19 Government\nResponse Tracker. The second part of the analysis employed a variety of machine\nlearning tools, utilized for estimating the concentration of each pollutant,\ntwo days ahead. The results showed that a weak to moderate correlation exists\nbetween the corresponding measures and the pollutant factors and that it is\npossible to create models which can predict the behaviour of the pollutant\ngases under daily human activities.\n"}
{"abstract": "  In pulsar astronomy, detecting effective pulsar signals among numerous pulsar\ncandidates is an important research topic. Starting from space X-ray pulsar\nsignals, the two-dimensional autocorrelation profile map (2D-APM) feature\nmodelling method, which utilizes epoch folding of the autocorrelation function\nof X-ray signals and expands the time-domain information of the periodic axis,\nis proposed. A uniform setting criterion regarding the time resolution of the\nperiodic axis addresses pulsar signals without any prior information. Compared\nwith the traditional profile, the model has a strong anti-noise ability, a\ngreater abundance of information and consistent characteristics. The new\nfeature is simulated with double Gaussian components, and the characteristic\ndistribution of the model is revealed to be closely related to the distance\nbetween the double peaks of the profile. Next, a deep convolutional neural\nnetwork (DCNN) is built, named Inception-ResNet. According to the order of the\npeak separation and number of arriving photons, 30 data sets based on the\nPoisson process are simulated to construct the training set, and the\nobservation data of PSRs B0531+21, B0540-69 and B1509-58 from the Rossi X-ray\nTiming Explorer (RXTE) are selected to generate the test set. The number of\ntraining sets and the number of test sets are 30,000 and 5,400, respectively.\nAfter achieving convergence stability, more than 99$\\%$ of the pulsar signals\nare recognized, and more than 99$\\%$ of the interference is successfully\nrejected, which verifies the high degree of agreement between the network and\nthe feature model and the high potential of the proposed method in searching\nfor pulsars.\n"}
{"abstract": "  Due to the continuously changing environment of service-based applications\n(SBAs), the ability to adapt to environmental and contextual changes has become\na crucial characteristic of such applications. Providing SBAs with this ability\nis a complex task, usually prepared in an unsystematic way and interwoven with\napplication logic. As a result, developing and maintaining adaptive SBAs has\nbecome a costly and hardly repeatable process. The objective of this paper is\nto present a model-based approach to developing adaptive SBAs, which separates\ndevelopment of adaptation concerns from development of SBAs behaviors. This\napproach aims to facilitate and automate development of adaptive behaviors. In\nthis paper, the process of developing an adaptive SBA is defined as specifying\nadaptive SBA models based on a metamodel and reusable adaptation tactics. These\nmodels are then transformed into runtime model artifacts and running system\nunits performing runtime adaptive behaviors. The approach introduces a\nsystematic method to derive adaptation behaviors from adaptation models, which\nfacilitates the development of adaptive behaviors. The results show that using\nour approach can enhance the development process of adaptive behaviors in terms\nof development time and code quality, particularly for scenarios that are more\ncomplex in terms of adaptation rather than business logic.\n"}
{"abstract": "  In an earlier study, we reported that the ram gauge of the COmet Pressure\nSensor (COPS), one of the three instruments of the Rosetta Orbiter Spectrometer\nfor Ion and Neutral Analysis (ROSINA), could be used to obtain information\nabout the sublimating content of icy particles, made up of volatiles and\nconceivably refractories coming from comet 67P/Churyumov-Gerasimenko.\n  In this work, we extend the investigation to the second COPS gauge, the nude\ngauge. In particular, we analyse the volume of the volatile content of coma\nparticles, along with a search for possible dependencies between the nude gauge\ndetection rate (i.e. the rate at which icy particles are detected by the nude\ngauge) and the position of the Rosetta spacecraft. We also investigate the\ncorrelations of the nude gauge detection rate with the quantities associated\nwith cometary activity.\n  Although it was not originally designed for such a purpose, the COPS nude\ngauge has been able to detect $\\sim$67000 features generated by the sublimation\nof the volatile content of icy particles. The nude gauge detection rate follows\na trend that is inversely proportional to the heliocentric distance. This\nresult is interpreted as a confirmation of a possible relation between the nude\ngauge detection rate and cometary activity. Thus, we compared the former with\nparameters related to cometary activity and obtained significant correlations,\nindicating that the frequency of icy particle detection is driven by cometary\nactivity. Furthermore, by representing the volatile part of the icy particles\nas equivalent spheres with a density of 1 g cm$^{-3}$, we obtained a range of\ndiameters between 60 and 793 nanometres, with the smaller ones\n($<390\\,\\mathrm{nm}$ in diameter) having a size distribution power index of\n$-4.79\\pm 0.26$.\n"}
{"abstract": "  Given a social network of users with selection cost, the \\textsc{Budgeted\nInfluence Maximization Problem} (\\emph{BIM Problem} in short) asks for\nselecting a subset of the nodes (known as \\emph{seed nodes}) within an\nallocated budget for initial activation to maximize the influence in the\nnetwork. In this paper, we study this problem under the\n\\emph{co\\mbox{-}operative game theoretic} framework. We model this problem as a\nco\\mbox{-}operative game where the users of the network are the players and for\na group of users, the expected influence by them under the \\emph{Maximum\nInfluence Arborences} diffusion model is its utility. We call this game as\n\\emph{BIM Game} and show this is `non-convex' and `sub-additive'. Based on the\nproposed game\\mbox{-}theoretic model and using the solution concept called\n`Shapley Value', we propose an iterative algorithm for finding seed nodes. The\nproposed methodology is divided into mainly two broad steps: the first one is\ncomputing the approximate marginal gain in \\emph{Shapley Value} for all the\nnodes of the network, and the second one is selecting seed nodes from the\nsorted list until the budget is exhausted. We also show that the proposed\nmethodology can even be more effective when the community structure of the\nnetwork is exploited. The proposed methodologies have been implemented, and an\nextensive set of experiments have been conducted with three publicly available\nsocial network datasets. From the experiments, we observe that the seed set\nselected by the proposed methodologies lead to more number of influence nodes\ncompared to many standard and baseline methods from the literature with a\nreasonable computational overhead. In particular, if the community structure of\nthe network is exploited then there is an increase upto $2 \\%$ in number of\ninfluenced nodes.\n"}
{"abstract": "  We develop a product functional quantization of rough volatility. Since the\nquantizers can be computed offline, this new technique, built on the insightful\nworks by Luschgy and Pages, becomes a strong competitor in the new arena of\nnumerical tools for rough volatility. We concentrate our numerical analysis to\npricing VIX Futures in the rough Bergomi model and compare our results to other\nrecently suggested benchmarks.\n"}
{"abstract": "  Time series forecasting is a crucial task in machine learning, as it has a\nwide range of applications including but not limited to forecasting electricity\nconsumption, traffic, and air quality. Traditional forecasting models rely on\nrolling averages, vector auto-regression and auto-regressive integrated moving\naverages. On the other hand, deep learning and matrix factorization models have\nbeen recently proposed to tackle the same problem with more competitive\nperformance. However, one major drawback of such models is that they tend to be\noverly complex in comparison to traditional techniques. In this paper, we\nreport the results of prominent deep learning models with respect to a\nwell-known machine learning baseline, a Gradient Boosting Regression Tree\n(GBRT) model. Similar to the deep neural network (DNN) models, we transform the\ntime series forecasting task into a window-based regression problem.\nFurthermore, we feature-engineered the input and output structure of the GBRT\nmodel, such that, for each training window, the target values are concatenated\nwith external features, and then flattened to form one input instance for a\nmulti-output GBRT model. We conducted a comparative study on nine datasets for\neight state-of-the-art deep-learning models that were presented at top-level\nconferences in the last years. The results demonstrate that the window-based\ninput transformation boosts the performance of a simple GBRT model to levels\nthat outperform all state-of-the-art DNN models evaluated in this paper.\n"}
{"abstract": "  This document shows how Z specifications can be translated into $\\{log\\}$\nand, later, on how $\\{log\\}$ can be used to run simulations and automated\nproofs. This can help users of other specification languages such as B and VDM\nto use $\\{log\\}$ along the same lines. The presentation is rather informal and\nuser-oriented. More technical and formal presentations can be found in the\npapers published by the authors. We also assume the reader has at least a basic\nknowledge of the Z notation.\n"}
{"abstract": "  In this paper we develop the theory of presentations for globular operads and\nconstruct presentations for the globular operads corresponding to several key\ntheories of $n$-category for $n \\leqslant 4$.\n"}
{"abstract": "  This paper presents a stereo object matching method that exploits both 2D\ncontextual information from images as well as 3D object-level information.\nUnlike existing stereo matching methods that exclusively focus on the\npixel-level correspondence between stereo images within a volumetric space\n(i.e., cost volume), we exploit this volumetric structure in a different\nmanner. The cost volume explicitly encompasses 3D information along its\ndisparity axis, therefore it is a privileged structure that can encapsulate the\n3D contextual information from objects. However, it is not straightforward\nsince the disparity values map the 3D metric space in a non-linear fashion.\nThus, we present two novel strategies to handle 3D objectness in the cost\nvolume space: selective sampling (RoISelect) and 2D-3D fusion\n(fusion-by-occupancy), which allow us to seamlessly incorporate 3D object-level\ninformation and achieve accurate depth performance near the object boundary\nregions. Our depth estimation achieves competitive performance in the KITTI\ndataset and the Virtual-KITTI 2.0 dataset.\n"}
{"abstract": "  Generative adversarial networks (GANs) are among the most successful models\nfor learning high-complexity, real-world distributions. However, in theory, due\nto the highly non-convex, non-concave landscape of the minmax training\nobjective, GAN remains one of the least understood deep learning models. In\nthis work, we formally study how GANs can efficiently learn certain\nhierarchically generated distributions that are close to the distribution of\nimages in practice. We prove that when a distribution has a structure that we\nrefer to as Forward Super-Resolution, then simply training generative\nadversarial networks using gradient descent ascent (GDA) can indeed learn this\ndistribution efficiently, both in terms of sample and time complexities. We\nalso provide concrete empirical evidence that not only our assumption \"forward\nsuper-resolution\" is very natural in practice, but also the underlying learning\nmechanisms that we study in this paper (to allow us efficiently train GAN via\nGDA in theory) simulates the actual learning process of GANs in practice on\nreal-world problems.\n"}
{"abstract": "  Recently, object detection in aerial images has gained much attention in\ncomputer vision. Different from objects in natural images, aerial objects are\noften distributed with arbitrary orientation. Therefore, the detector requires\nmore parameters to encode the orientation information, which are often highly\nredundant and inefficient. Moreover, as ordinary CNNs do not explicitly model\nthe orientation variation, large amounts of rotation augmented data is needed\nto train an accurate object detector. In this paper, we propose a\nRotation-equivariant Detector (ReDet) to address these issues, which explicitly\nencodes rotation equivariance and rotation invariance. More precisely, we\nincorporate rotation-equivariant networks into the detector to extract\nrotation-equivariant features, which can accurately predict the orientation and\nlead to a huge reduction of model size. Based on the rotation-equivariant\nfeatures, we also present Rotation-invariant RoI Align (RiRoI Align), which\nadaptively extracts rotation-invariant features from equivariant features\naccording to the orientation of RoI. Extensive experiments on several\nchallenging aerial image datasets DOTA-v1.0, DOTA-v1.5 and HRSC2016, show that\nour method can achieve state-of-the-art performance on the task of aerial\nobject detection. Compared with previous best results, our ReDet gains 1.2, 3.5\nand 2.6 mAP on DOTA-v1.0, DOTA-v1.5 and HRSC2016 respectively while reducing\nthe number of parameters by 60\\% (313 Mb vs. 121 Mb). The code is available at:\n\\url{https://github.com/csuhan/ReDet}.\n"}
{"abstract": "  We describe conformal defects of $p$ dimensions in a free scalar theory on a\n$d$-dimensional flat space as boundary conditions on the conformally flat space\n$\\mathbb{H}^{p+1}\\times \\mathbb{S}^{d-p-1}$. We classify two types of boundary\nconditions, Dirichlet type and Neumann type, on the boundary of the subspace\n$\\mathbb{H}^{p+1}$ which correspond to the types of conformal defects in the\nfree scalar theory. We find Dirichlet boundary conditions always exist while\nNeumann boundary conditions are allowed only for defects of lower codimensions.\nOur results match with a recent classification of the non-monodromy defects,\nshowing Neumann boundary conditions are associated with non-trivial defects. We\ncheck this observation by calculating the difference of the free energies on\n$\\mathbb{H}^{p+1}\\times \\mathbb{S}^{d-p-1}$ between Dirichlet and Neumann\nboundary conditions. We also examine the defect RG flows from Neumann to\nDirichlet boundary conditions and provide more support for a conjectured\n$C$-theorem in defect CFTs.\n"}
{"abstract": "  Self-supervised learning (SSL) of energy based models has an intuitive\nrelation to equilibrium thermodynamics because the softmax layer, mapping\nenergies to probabilities, is a Gibbs distribution. However, in what way SSL is\na thermodynamic process? We show that some SSL paradigms behave as a\nthermodynamic composite system formed by representations and self-labels in\ncontact with a nonequilibrium reservoir. Moreover, this system is subjected to\nusual thermodynamic cycles, such as adiabatic expansion and isochoric heating,\nresulting in a generalized Gibbs ensemble (GGE). In this picture, we show that\nlearning is seen as a demon that operates in cycles using feedback measurements\nto extract negative work from the system. As applications, we examine some SSL\nalgorithms using this idea.\n"}
{"abstract": "  Differentially private algorithms protect individuals in data analysis\nscenarios by ensuring that there is only a weak correlation between the\nexistence of the user in the data and the result of the analysis. Dynamic graph\nalgorithms maintain the solution to a problem (e.g., a matching) on an evolving\ninput, i.e., a graph where nodes or edges are inserted or deleted over time.\nThey output the value of the solution after each update operation, i.e.,\ncontinuously. We study (event-level and user-level) differentially private\nalgorithms for graph problems under continual observation, i.e., differentially\nprivate dynamic graph algorithms. We present event-level private algorithms for\npartially dynamic counting-based problems such as triangle count that improve\nthe additive error by a polynomial factor (in the length $T$ of the update\nsequence) on the state of the art, resulting in the first algorithms with\nadditive error polylogarithmic in $T$.\n  We also give $\\varepsilon$-differentially private and partially dynamic\nalgorithms for minimum spanning tree, minimum cut, densest subgraph, and\nmaximum matching. The additive error of our improved MST algorithm is $O(W\n\\log^{3/2}T / \\varepsilon)$, where $W$ is the maximum weight of any edge,\nwhich, as we show, is tight up to a $(\\sqrt{\\log T} / \\varepsilon)$-factor. For\nthe other problems, we present a partially-dynamic algorithm with\nmultiplicative error $(1+\\beta)$ for any constant $\\beta > 0$ and additive\nerror $O(W \\log(nW) \\log(T) / (\\varepsilon \\beta) )$. Finally, we show that the\nadditive error for a broad class of dynamic graph algorithms with user-level\nprivacy must be linear in the value of the output solution's range.\n"}
{"abstract": "  Modeling nuclear reaction networks for nuclear science applications and for\nsimulations of astrophysical environments relies on cross section data for a\nvast number of reactions, many of which have never been measured.\n  In this paper we examine the validity of the Weisskopf-Ewing approximation\nfor determining unknown (n,n') and (n, 2n) cross sections from surrogate data.\nUsing statistical reaction calculations with realistic parameterizations, we\ninvestigate first whether the assumptions underlying the Weisskopf-Ewing\napproximation are valid for (n,n') and (n, 2n) reactions on representative\ntarget nuclei. We then produce simulated surrogate reaction data and assess the\nimpact of applying the Weisskopf-Ewing approximation when extracting (n,n') and\n(n, 2n) cross sections in situations where the approximation is not strictly\njustified.\n  We find that peak cross sections can be estimated using the Weisskopf-Ewing\napproximation, but the shape of the (n,n') and (n,2n) cross sections,\nespecially for low neutron energies, cannot be reliably determined without\naccounting for the angular-momentum differences between the neutron-induced and\nthe surrogate reaction.\n"}
{"abstract": "  Despite having promising results, style transfer, which requires preparing\nstyle images in advance, may result in lack of creativity and accessibility.\nFollowing human instruction, on the other hand, is the most natural way to\nperform artistic style transfer that can significantly improve controllability\nfor visual effect applications. We introduce a new task -- language-driven\nimage style transfer (\\texttt{LDIST}) -- to manipulate the style of a content\nimage, guided by a text. We propose contrastive language visual artist (CLVA)\nthat learns to extract visual semantics from style instructions and accomplish\n\\texttt{LDIST} by the patch-wise style discriminator. The discriminator\nconsiders the correlation between language and patches of style images or\ntransferred results to jointly embed style instructions. CLVA further compares\ncontrastive pairs of content image and style instruction to improve the mutual\nrelativeness between transfer results. The transferred results from the same\ncontent image can preserve consistent content structures. Besides, they should\npresent analogous style patterns from style instructions that contain similar\nvisual semantics. The experiments show that our CLVA is effective and achieves\nsuperb transferred results on \\texttt{LDIST}.\n"}
{"abstract": "  This work focuses on the analysis that whether 3D face models can be learned\nfrom only the speech inputs of speakers. Previous works for cross-modal face\nsynthesis study image generation from voices. However, image synthesis includes\nvariations such as hairstyles, backgrounds, and facial textures, that are\narguably irrelevant to voice or without direct studies to show correlations. We\ninstead investigate the ability to reconstruct 3D faces to concentrate on only\ngeometry, which is more physiologically grounded. We propose both the\nsupervised learning and unsupervised learning frameworks. Especially we\ndemonstrate how unsupervised learning is possible in the absence of a direct\nvoice-to-3D-face dataset under limited availability of 3D face scans when the\nmodel is equipped with knowledge distillation. To evaluate the performance, we\nalso propose several metrics to measure the geometric fitness of two 3D faces\nbased on points, lines, and regions. We find that 3D face shapes can be\nreconstructed from voices. Experimental results suggest that 3D faces can be\nreconstructed from voices, and our method can improve the performance over the\nbaseline. The best performance gains (15% - 20%) on ear-to-ear distance ratio\nmetric (ER) coincides with the intuition that one can roughly envision whether\na speaker's face is overall wider or thinner only from a person's voice. See\nour project page for codes and data.\n"}
{"abstract": "  We argue that in quantum gravity there is no Born rule. The quantum-gravity\nregime, described by a non-normalisable Wheeler-DeWitt wave functional $\\Psi$,\nmust be in quantum nonequilibrium with a probability distribution\n$P\\neq\\left\\vert \\Psi\\right\\vert ^{2}$ (initially and always). A Born rule can\nemerge only in the semiclassical regime of quantum systems on a classical\nspacetime background, with normalisable Schr\\\"{o}dinger wave functions $\\psi$.\nConditioning on the underlying quantum-gravitational ensemble yields a\nnonequilibrium distribution $\\rho\\neq\\left\\vert \\psi\\right\\vert ^{2}$ at the\nbeginning of the semiclassical regime, with quantum relaxation\n$\\rho\\rightarrow\\left\\vert \\psi\\right\\vert ^{2}$ taking place only afterwards.\nQuantum gravity naturally creates an early nonequilibrium universe. We also\nshow how small corrections to the Schr\\\"{o}dinger equation yield an\nintermediate regime in which the Born rule is unstable: an initial distribution\n$\\rho=\\left\\vert \\psi\\right\\vert ^{2}$ can evolve to a final distribution\n$\\rho\\neq\\left\\vert \\psi\\right\\vert ^{2}$. These results arise naturally in the\nde Broglie-Bohm pilot-wave formulation of quantum gravity. We show that quantum\ninstability during inflation generates a large-scale deficit $\\sim1/k^{3}$ in\nthe primordial power spectrum at wavenumber $k$, though the effect is too small\nto observe. Similarly we find an unobservably large timescale for quantum\ninstability in a radiation-dominated universe. Quantum instability may be\nimportant in black-hole evaporation, with a final burst of Hawking radiation\nthat violates the Born rule. Deviations from the Born rule can also be\ngenerated for atomic systems in the gravitational field of the earth, though\nthe effects are unlikely to be observable. The most promising scenario for the\ndetection of Born-rule violations appears to be in radiation from exploding\nprimordial black holes.\n"}
{"abstract": "  We present a model predictive controller (MPC) that automatically discovers\ncollision-free locomotion while simultaneously taking into account the system\ndynamics, friction constraints, and kinematic limitations. A relaxed barrier\nfunction is added to the optimization's cost function, leading to collision\navoidance behavior without increasing the problem's computational complexity.\nOur holistic approach does not require any heuristics and enables legged robots\nto find whole-body motions in the presence of static and dynamic obstacles. We\nuse a dynamically generated euclidean signed distance field for static\ncollision checking. Collision checking for dynamic obstacles is modeled with\nmoving cylinders, increasing the responsiveness to fast-moving agents.\nFurthermore, we include a Kalman filter motion prediction for moving obstacles\ninto our receding horizon planning, enabling the robot to anticipate possible\nfuture collisions. Our experiments demonstrate collision-free motions on a\nquadrupedal robot in challenging indoor environments. The robot handles complex\nscenes like overhanging obstacles and dynamic agents by exploring motions at\nthe robot's dynamic and kinematic limits.\n"}
{"abstract": "  This is the second article of a sequence of research on deformations of\nQ-curvature. In the previous one, we studied local stability and rigidity\nphenomena of Q-curvature. In this article, we mainly investigate the volume\ncomparison with respect to Q-curvature. In particular, we show that volume\ncomparison theorem holds for metrics close to strictly stable positive Einstein\nmetrics. This result shows that Q-curvature can still control the volume of\nmanifolds under certain conditions, which provides a fundamental geometric\ncharacterization of Q-curvature. Applying the same technique, we derive the\nlocal rigidity of strictly stable Ricci-flat manifolds with respect to\nQ-curvature, which shows the non-existence of metrics with positive Q-curvature\nnear the reference metric.\n"}
{"abstract": "  We report on atom probe tomography analyses of Pd and Pd@Au nanoparticles\nembedded in a Ni matrix and the effects of local evaporation field variations\non the atom probe data. In order to assess the integrity of the reconstructed\natom maps, we performed numerical simulations of the field evaporation\nprocesses and compared the simulated datasets with experimentally acquired\ndata. The distortions seen in the atom maps for both Pd and Pd@Au nanoparticles\ncould be mostly ascribed to local variations in chemical composition and\nelemental evaporation fields. The evaporation field values for Pd and Ni, taken\nfrom the image hump model and assumed in the simulations, yielded a good\nagreement between experimental and simulation results. In contrast, the\nevaporation field for Au, as predicted from the image hump model, appeared to\nbe substantially overestimated and resulted in a large discrepancy between\nexperiments and simulations.\n"}
{"abstract": "  The multi-reference alignment (MRA) problem entails estimating an image from\nmultiple noisy and rotated copies of itself. If the noise level is low, one can\nreconstruct the image by estimating the missing rotations, aligning the images,\nand averaging out the noise. While accurate rotation estimation is impossible\nif the noise level is high, the rotations can still be approximated, and thus\ncan provide indispensable information. In particular, learning the\napproximation error can be harnessed for efficient image estimation. In this\npaper, we propose a new computational framework, called Synch-EM, that consists\nof angular synchronization followed by expectation-maximization (EM). The\nsynchronization step results in a concentrated distribution of rotations; this\ndistribution is learned and then incorporated into the EM as a Bayesian prior.\nThe learned distribution also dramatically reduces the search space, and thus\nthe computational load, of the EM iterations. We show by extensive numerical\nexperiments that the proposed framework can significantly accelerate EM for MRA\nin high noise levels, occasionally by a few orders of magnitude, without\ndegrading the reconstruction quality.\n"}
{"abstract": "  We propose automatic speech recognition (ASR) models inspired by echo state\nnetwork (ESN), in which a subset of recurrent neural networks (RNN) layers in\nthe models are randomly initialized and untrained. Our study focuses on RNN-T\nand Conformer models, and we show that model quality does not drop even when\nthe decoder is fully randomized. Furthermore, such models can be trained more\nefficiently as the decoders do not require to be updated. By contrast,\nrandomizing encoders hurts model quality, indicating that optimizing encoders\nand learn proper representations for acoustic inputs are more vital for speech\nrecognition. Overall, we challenge the common practice of training ASR models\nfor all components, and demonstrate that ESN-based models can perform equally\nwell but enable more efficient training and storage than fully-trainable\ncounterparts.\n"}
{"abstract": "  This report describes our winning submission to the Real Robot Challenge\n(https://real-robot-challenge.com/). The Real Robot Challenge is a three-phase\ndexterous manipulation competition that involves manipulating various\nrectangular objects with the TriFinger Platform. Our approach combines motion\nplanning with several motion primitives to manipulate the object. For Phases 1\nand 2, we additionally learn a residual policy in simulation that applies\ncorrective actions on top of our controller. Our approach won first place in\nPhase 2 and Phase 3 of the competition. We were anonymously known as\n`ardentstork' on the competition leaderboard\n(https://real-robot-challenge.com/leader-board). Videos and our code can be\nfound at https://github.com/ripl-ttic/real-robot-challenge.\n"}
{"abstract": "  Highly reliable Monte-Carlo event generators and detector simulation programs\nare important for the precision measurement in the high energy physics. Huge\namounts of computing resources are required to produce a sufficient number of\nsimulated events. Moreover, simulation parameters have to be fine-tuned to\nreproduce situations in the high energy particle interactions which is not\ntrivial in some phase spaces in physics interests. In this paper, we suggest a\nnew method based on the Wasserstein Generative Adversarial Network (WGAN) that\ncan learn the probability distribution of the real data. Our method is capable\nof event generation at a very short computing time compared to the traditional\nMC generators. The trained WGAN is able to reproduce the shape of the real data\nwith high fidelity.\n"}
{"abstract": "  Chiral spin-triplet superconductivity is a topologically nontrivial pairing\nstate with broken time-reversal symmetry, which can host Majorana\nquasiparticles. The recently discovered heavy-fermion superconductor UTe$_2$\nexhibits peculiar properties of spin-triplet pairing, and the possible chiral\nstate has been actively discussed. However, the symmetry and nodal structure of\nits order parameter in the bulk, which determine the Majorana surface states,\nremains controversial. Here we focus on the number and positions of\nsuperconducting gap nodes in the ground state of UTe$_2$. Our magnetic\npenetration depth measurements for three field orientations in the Meissner\nstate reveal the power-law temperature dependence with exponents less than 2,\nwhich excludes single-component spin-triplet states. The anisotropy of\nlow-energy quasiparticle excitations indicates multiple point nodes near the\n$k_y$- and $k_z$-axes, evidencing that the order parameter has multiple\ncomponents in a chiral complex form. We find that most consistent is a chiral\n$B_{3u}+iA_u$ non-unitary state, which provides fundamentals of the topological\nproperties in UTe$_2$.\n"}
{"abstract": "  We study the distribution of a general class of asymptoticallylinear\nstatistics which are symmetric functions of $N$ independent observations. The\ndistribution functions of these statistics are approximated by an Edgeworth\nexpansion with a remainder of order $o(N^{-1})$. The Edgeworth expansion is\nbased on Hoeffding's decomposition which provides a stochastic expansion into a\nlinear part, a quadratic part as well as smaller higher order parts. The\nvalidity of this Edgeworth expansion is proved under Cram\\'er's condition on\nthe linear part, moment assumptions for all parts of the statistic and an\noptimal dimensionality requirement for the non linear part.\n"}
{"abstract": "  In a hybrid speech model, both voiced and unvoiced components can coexist in\na segment. Often, the voiced speech is regarded as the deterministic component,\nand the unvoiced speech and additive noise are the stochastic components.\nTypically, the speech signal is considered stationary within fixed segments of\n20-40 ms, but the degree of stationarity varies over time. For decomposing\nnoisy speech into its voiced and unvoiced components, a fixed segmentation may\nbe too crude, and we here propose to adapt the segment length according to the\nsignal local characteristics. The segmentation relies on parameter estimates of\na hybrid speech model and the maximum a posteriori (MAP) and log-likelihood\ncriteria as rules for model selection among the possible segment lengths, for\nvoiced and unvoiced speech, respectively. Given the optimal segmentation\nmarkers and the estimated statistics, both components are estimated using\nlinear filtering. A codebook-based approach differentiates between unvoiced\nspeech and noise. A better extraction of the components is possible by taking\ninto account the adaptive segmentation, compared to a fixed one. Also, a lower\ndistortion for voiced speech and higher segSNR for both components is possible,\nas compared to other decomposition methods.\n"}
{"abstract": "  Prediction of a molecule's 3D conformer ensemble from the molecular graph\nholds a key role in areas of cheminformatics and drug discovery. Existing\ngenerative models have several drawbacks including lack of modeling important\nmolecular geometry elements (e.g. torsion angles), separate optimization stages\nprone to error accumulation, and the need for structure fine-tuning based on\napproximate classical force-fields or computationally expensive methods such as\nmetadynamics with approximate quantum mechanics calculations at each geometry.\nWe propose GeoMol--an end-to-end, non-autoregressive and SE(3)-invariant\nmachine learning approach to generate distributions of low-energy molecular 3D\nconformers. Leveraging the power of message passing neural networks (MPNNs) to\ncapture local and global graph information, we predict local atomic 3D\nstructures and torsion angles, avoiding unnecessary over-parameterization of\nthe geometric degrees of freedom (e.g. one angle per non-terminal bond). Such\nlocal predictions suffice both for the training loss computation, as well as\nfor the full deterministic conformer assembly (at test time). We devise a\nnon-adversarial optimal transport based loss function to promote diverse\nconformer generation. GeoMol predominantly outperforms popular open-source,\ncommercial, or state-of-the-art machine learning (ML) models, while achieving\nsignificant speed-ups. We expect such differentiable 3D structure generators to\nsignificantly impact molecular modeling and related applications.\n"}
{"abstract": "  Spike-timing-dependent plasticity(STDP) is a biological process in which the\nprecise order and timing of neuronal spikes affect the degree of synaptic\nmodification. While there have been numerous research focusing on the role of\nSTDP in neural coding, the functional implications of STDP at the macroscopic\nlevel in the brain have not been fully explored yet. In this work, we propose a\nneurodynamical model based on STDP that renders storage and retrieval of a\ngroup of associative memories. We showed that the function of STDP at the\nmacroscopic level is to form a \"memory plane\" in the neural state space which\ndynamically encodes high dimensional data. We derived the analytic relation\nbetween the input, the memory plane, and the induced macroscopic neural\noscillations around the memory plane. Such plane produces a limit cycle in\nreaction to a similar memory cue, which can be used for retrieval of the\noriginal input.\n"}
{"abstract": "  With the advent of 21st century and increasing advancements in the field of\ntechnology and connectivity, inter-networking in real-time has achieved great\nimportance. Distributed control and multi-agent paradigm has groped rapidly\nwith history of big time failures of centralized systems in the past. The\nconcepts of synchronization and network control systems have been used\nextensively in the near past to map, analyze and solve defined set of\nobjectives. In this thesis, a diverse set of applications from power flow point\nof view are taken into consideration and modelled/analyzed using\nsynchronization as the central theme. These systems are proposed (or assumed)\nto be network connected and its control has been devised accordingly. It has\nbeen shown how some examples from nature can help recreate similar dynamics\nsynthetically and help achieve system objectives. Few of the applications of\nthe smart world have been ascribed in the thesis and distributed control of\nthese seen from a multi-agent perspective have been devised in order to better\ndesign and operate such systems in real-time. Synchronization happens to be the\nheart of all the networks, as all the agents work in tandem in order to provide\nto a common objective as well abide by the defined constraints. As an\ninflection to the work, I set up a platform for development of new distributed\nand fast acting control strategies for conventional as well as futuristic\ncontrol systems existing/non-existing in the literature. As a major\ncontribution of this dissertation, by combining the ideas from physics and\npower systems, I open up various interesting phenomena already existing in\neither fields to be explored for one another.\n"}
{"abstract": "  Solar wind stream interaction regions (SIRs) are often characterized by\nenergetic ion enhancements. The mechanisms accelerating these particles, as\nwell as the locations where the acceleration occurs, remain debated. Here, we\nreport the findings of a simulation of a SIR event observed by Parker Solar\nProbe at ~0.56 au and the Solar Terrestrial Relations Observatory-Ahead at\n~0.95 au in 2019 September when both spacecraft were approximately radially\naligned with the Sun. The simulation reproduces the solar wind configuration\nand the energetic particle enhancements observed by both spacecraft. Our\nresults show that the energetic particles are produced at the compression waves\nassociated with the SIR and that the suprathermal tail of the solar wind is a\ngood candidate to provide the seed population for particle acceleration. The\nsimulation confirms that the acceleration process does not require shock waves\nand can already commence within Earth's orbit, with an energy dependence on the\nprecise location where particles are accelerated. The three-dimensional\nconfiguration of the solar wind streams strongly modulates the energetic\nparticle distributions, illustrating the necessity of advanced models to\nunderstand these particle events.\n"}
{"abstract": "  The existence of a fundamental scale is expected to be a key feature of\nquantum gravity. Many approaches take this property as a starting assumption.\nHere, instead, we take a less conventional viewpoint based on a critical\ninspection of both fundamental principles and kinematic laws. We point out that\nrigorous arguments suggest a more urgent need to revise known theories to\nincorporate a fundamental acceleration scale already in flat space. The\nreciprocity principle can naturally do so. In addition to noticing links with\nstring theory, we argue that the reciprocity principle implies an\ninfinite-derivative generalization of the Einstein-Hilbert action that makes\nthe gravitational interaction fundamentally nonlocal, thus providing a guiding\nprinciple that could lead us towards the formulation of a consistent theory of\nquantum gravity.\n"}
{"abstract": "  As an alternative to the popular parametrisations of the dark energy equation\nof state, we construct a quintessence model where the scalar field has a linear\ndependence on the number of e-folds. Constraints on more complex models are\ntypically limited by the degeneracies that increase with the number of\nparameters. The proposed parametrisation conveniently constrains the evolution\nof the dark energy equation of state as it allows for a wide variety of time\nevolutions. We also consider a non-minimal coupling to cold dark matter. We fit\nthe model with Planck and KiDS observational data. The CMB favours a\nnon-vanishing coupling with energy transfer from dark energy to dark matter.\nConversely, gravitational weak lensing measurements slightly favour energy\ntransfer from dark matter to dark energy, with a substantial departure of the\ndark energy equation of state from -1.\n"}
{"abstract": "  We propose a novel and theoretical model, blocked and hierarchical\nvariational autoencoder (BHiVAE), to get better-disentangled representation. It\nis well known that information theory has an excellent explanatory meaning for\nthe network, so we start to solve the disentanglement problem from the\nperspective of information theory. BHiVAE mainly comes from the information\nbottleneck theory and information maximization principle. Our main idea is that\n(1) Neurons block not only one neuron node is used to represent attribute,\nwhich can contain enough information; (2) Create a hierarchical structure with\ndifferent attributes on different layers, so that we can segment the\ninformation within each layer to ensure that the final representation is\ndisentangled. Furthermore, we present supervised and unsupervised BHiVAE,\nrespectively, where the difference is mainly reflected in the separation of\ninformation between different blocks. In supervised BHiVAE, we utilize the\nlabel information as the standard to separate blocks. In unsupervised BHiVAE,\nwithout extra information, we use the Total Correlation (TC) measure to achieve\nindependence, and we design a new prior distribution of the latent space to\nguide the representation learning. It also exhibits excellent disentanglement\nresults in experiments and superior classification accuracy in representation\nlearning.\n"}
{"abstract": "  In this paper we study a dynamical optimal transport problem on a network\nthat allows for transport of mass between different edges if a penalty $\\kappa$\nis paid. We show existence of minimisers using duality and discuss the\nrelationships of the distance-functional to other metrics such as the\nFisher-Rao and the classical Wasserstein metric and analyse the resulting\ndistance functional in the limiting case $\\kappa \\rightarrow \\infty$.\n"}
{"abstract": "  In this paper we continue the discussion about relations between exponential\npolynomials and generalized moment generating functions on a commutative\nhypergroup. We are interested in the following problem: is it true that every\nfinite dimensional variety is spanned by moment functions? Let $m$ be an\nexponential on $X$. In our former paper we have proved that if the linear space\nof all $m$-sine functions in the variety of an $m$-exponential monomial is (at\nmost) one dimensional, then this variety is spanned by moment functions\ngenerated by $m$. In this paper we show that this may happen also in cases\nwhere the $m$-sine functions span a more than one dimensional subspace in the\nvariety. We recall the notion of a polynomial hypergroup in $d$ variables,\ndescribe exponentials on it and give the characterization of the so called\n$m$-sine functions. Next we show that the Fourier algebra of a polynomial\nhypergroup in $d$ variables is the polynomial ring in $d$ variables. Finally,\nusing Ehrenpreis--Palamodov Theorem we show that every exponential polynomial\non the polynomial hypergroup in $d$ variables is a linear combination of moment\nfunctions contained in its variety.\n"}
{"abstract": "  Among the multitude of software development processes available, hardly any\nis used by the book. Regardless of company size or industry sector, a majority\nof project teams and companies use customized processes that combine different\ndevelopment methods -- so-called hybrid development methods. Even though such\nhybrid development methods are highly individualized, a common understanding of\nhow to systematically construct synergetic practices is missing. In this paper,\nwe make a first step towards devising such guidelines. Grounded in 1,467 data\npoints from a large-scale online survey among practitioners, we study the\ncurrent state of practice in process use to answer the question: What are\nhybrid development methods made of? Our findings reveal that only eight methods\nand few practices build the core of modern software development. This small set\nallows for statistically constructing hybrid development methods. Using an 85%\nagreement level in the participants' selections, we provide two examples\nillustrating how hybrid development methods are characterized by the practices\nthey are made of. Our evidence-based analysis approach lays the foundation for\ndevising hybrid development methods.\n"}
{"abstract": "  Mixed-precision quantization is a powerful tool to enable memory and compute\nsavings of neural network workloads by deploying different sets of bit-width\nprecisions on separate compute operations. Recent research has shown\nsignificant progress in applying mixed-precision quantization techniques to\nreduce the memory footprint of various workloads, while also preserving task\nperformance. Prior work, however, has often ignored additional objectives, such\nas bit-operations, that are important for deployment of workloads on hardware.\nHere we present a flexible and scalable framework for automated mixed-precision\nquantization that optimizes multiple objectives. Our framework relies on\nNeuroevolution-Enhanced Multi-Objective Optimization (NEMO), a novel search\nmethod, to find Pareto optimal mixed-precision configurations for memory and\nbit-operations objectives. Within NEMO, a population is divided into\nstructurally distinct sub-populations (species) which jointly form the Pareto\nfrontier of solutions for the multi-objective problem. At each generation,\nspecies are re-sized in proportion to the goodness of their contribution to the\nPareto frontier. This allows NEMO to leverage established search techniques and\nneuroevolution methods to continually improve the goodness of the Pareto\nfrontier. In our experiments we apply a graph-based representation to describe\nthe underlying workload, enabling us to deploy graph neural networks trained by\nNEMO to find Pareto optimal configurations for various workloads trained on\nImageNet. Compared to the state-of-the-art, we achieve competitive results on\nmemory compression and superior results for compute compression for\nMobileNet-V2, ResNet50 and ResNeXt-101-32x8d. A deeper analysis of the results\nobtained by NEMO also shows that both the graph representation and the\nspecies-based approach are critical in finding effective configurations for all\nworkloads.\n"}
{"abstract": "  In this paper we argue that data science is a coherent and novel approach to\nempirical problems that, in its most general form, does not build understanding\nabout phenomena. Within the new type of mathematization at work in data\nscience, mathematical methods are not selected because of any relevance for a\nproblem at hand; mathematical methods are applied to a specific problem only by\n'forcing', i.e. on the basis of their ability to reorganize the data for\nfurther analysis and the intrinsic richness of their mathematical structure. In\nparticular, we argue that deep learning neural networks are best understood\nwithin the context of forcing optimization methods. We finally explore the\nbroader question of the appropriateness of data science methods in solving\nproblems. We argue that this question should not be interpreted as a search for\na correspondence between phenomena and specific solutions found by data science\nmethods; rather, it is the internal structure of data science methods that is\nopen to precise forms of understanding.\n"}
{"abstract": "  This paper provides a new solution approach to a multi-player perimeter\ndefense game, in which the intruders' team tries to enter the territory, and a\nteam of defenders protects the territory by capturing intruders on the\nperimeter of the territory. The objective of the defenders is to detect and\ncapture the intruders before the intruders enter the territory. Each defender\nindependently senses the intruder and computes his trajectory to capture the\nassigned intruders in a cooperative fashion. The intruder is estimated to reach\na specific location on the perimeter at a specific time. Each intruder is\nviewed as a spatio-temporal task, and the defenders are assigned to execute\nthese spatio-temporal tasks. At any given time, the perimeter defense problem\nis converted into a Decentralized Multi-UAV Spatio-Temporal Multi-Task\nAllocation (DMUST-MTA) problem. The cost of executing a task for a trajectory\nis defined by a composite cost function of both the spatial and temporal\ncomponents. In this paper, a decentralized consensus-based bundle algorithm has\nbeen modified to solve the spatio-temporal multi-task allocation problem, and\nthe performance evaluation of the proposed approach is carried out based on\nMonte-Carlo simulations. The simulation results show the effectiveness of the\nproposed approach to solve the perimeter defense game under different\nscenarios. Performance comparison with a state-of-the-art centralized approach\nwith full observability, clearly indicates that DMUST-MTA achieves similar\nperformance in a decentralized way with partial observability conditions with a\nlesser computational time and easy scaling up.\n"}
{"abstract": "  Bose-Einstein condensation, observed in either strongly interacting liquid\nhelium or weakly interacting atomic Bose gases, is widely known to be a\nsecond-order phase transition. Here, we predict a first-order Bose-Einstein\ncondensation in a cloud of harmonically trapped bosons interacting with both\nattractive two-body interaction and repulsive three-body interaction,\ncharacterized respectively by an $s$-wave scattering length $a<0$ and a\nthree-body scattering hypervolume $D>0$. It happens when the harmonic trapping\npotential is weak, so with increasing temperature the system changes from a\nlow-temperature liquid-like quantum droplet to a normal gas, and therefore\nexperiences a first-order liquid-to-gas transition. At large trapping\npotential, however, the quantum droplet can first turn into a superfluid gas,\nrendering the condensation transition occurred later from a superfluid gas to a\nnormal gas smooth. We determine a rich phase diagram and show the existence of\na tri-critical point, where the three phases - quantum droplet, superfluid gas\nand normal gas - meet together. We argue that an ensemble of spin-polarized\ntritium atoms could be a promising candidate to observe the predicted\nfirst-order Bose-Einstein condensation, across which the condensate fraction or\ncentral condensate density jumps to zero and the surface-mode frequencies\ndiverge.\n"}
{"abstract": "  A central goal of neuroscience is to understand the representations formed by\nbrain activity patterns and their connection to behavior. The classical\napproach is to investigate how individual neurons encode the stimuli and how\ntheir tuning determines the fidelity of the neural representation. Tuning\nanalyses often use the Fisher information to characterize the sensitivity of\nneural responses to small changes of the stimulus. In recent decades,\nmeasurements of large populations of neurons have motivated a complementary\napproach, which focuses on the information available to linear decoders. The\ndecodable information is captured by the geometry of the representational\npatterns in the multivariate response space. Here we review neural tuning and\nrepresentational geometry with the goal of clarifying the relationship between\nthem. The tuning induces the geometry, but different sets of tuned neurons can\ninduce the same geometry. The geometry determines the Fisher information, the\nmutual information, and the behavioral performance of an ideal observer in a\nrange of psychophysical tasks. We argue that future studies can benefit from\nconsidering both tuning and geometry to understand neural codes and reveal the\nconnections between stimulus, brain activity, and behavior.\n"}
{"abstract": "  Wireless indoor coverage and data capacity are important aspects of cellular\nnetworks. With the ever-increasing data traffic, demand for more data capacity\nindoors is also growing. The lower frequencies of the legacy frequency bands of\nmacro outdoor cells manage to provide coverage inside buildings, however, new\nfrequencies foreseen for the 5th generation (5G) of mobile communications in\nthe millimeter wave (mmWave) spectrum penetrate very poorly into buildings.\nTherefore, a massive densification of the network would require to deploy a\nlarge number of indoor small cells, which would lead to high deployment costs\nto install the necessary wired/optical backhaul. Hence, other methods are\nneeded that allow an increase of the data capacity indoors, bearing a lower\ncost than a fiber deployment. We propose a cost-efficient out-of-band repeater\narchitecture that provides more data capacity indoors than an outdoor\nmacro/micro network can provide to indoor, without adversely affecting a legacy\nnetwork, and which readily works with the established cellular infrastructure\nas well as standard handsets/smartphones. This proposal is compared to\nconventional in- and out-of-band repeaters and relay nodes in order to\nhighlight the advantages of our solution. While the data capacity for a single\nlink is similar to that of repeaters and relays, a macro cell can be\neffectively offloaded. Cell capacities corresponding to at least 3--4 times\nthat of a repeater or relay solution can be provided, depending on the number\nof parallel installed links and the bandwidth in the mmWave spectrum.\n"}
{"abstract": "  We are developing tools for working with arbitrary left-exact localizations\nof $\\infty$-topoi. We introduce a notion of higher sheaf with respect to an\narbitrary set of maps $\\Sigma$ in an $\\infty$-topos $\\mathscr{E}$. We show that\nthe full subcategory of higher sheaves $\\mathrm{Sh}(\\mathscr{E},\\Sigma)$ is an\n$\\infty$-topos, and that the sheaf reflection $\\mathscr{E}\\to\n\\mathrm{Sh}(\\mathscr{E},\\Sigma)$ is the left-exact localization generated by\n$\\Sigma$. The proof depends on the notion of congruence, which is a substitute\nfor the notion of Grothendieck topology in 1-topos theory.\n"}
{"abstract": "  The FCC-ee physics program will deliver two complementary top-notch precision\ndeterminations of the W boson mass, and width. The first and main measurement\nrelies on the rapid rise of the W-pair production cross section near its\nkinematic threshold. This method is extremely simple and clean, involving only\nthe selection and counting of events, in all different decay channels. An\noptimal threshold-scan strategy with a total integrated luminosity of $12\\,{\\rm\nab}^{-1}$ shared on energy points between 157 and 163 GeV will provide a\nstatistical uncertainty on the W mass of 0.5 MeV and on the W width of 1.2 MeV.\nFor these measurements, the goal of keeping the impact of systematic\nuncertainties below the statistical precision will be demanding, but feasible.\nThe second method exploits the W-pair final state reconstruction and kinematic\nfit, making use of events with either four jets or two jets, one lepton and\nmissing energy. The projected statistical precision of the second method is\nsimilar to the first method's, with uncertainties of $\\sim 0.5$ ($1$) MeV for\nthe W mass (width), employing W-pair data collected at the production threshold\nand at 240-365 GeV. For the kinematic reconstruction method, the final impact\nof systematic uncertainties is currently less clear, in particular\nuncertainties connected to the modelling of the W hadronic decays. The use and\ninterplay of Z$\\gamma$ and ZZ events, reconstructed and fitted with the same\ntechniques as the WW events, will be important for the extraction of W mass\nmeasurements with data at the higher 240 and 365 GeV energies.\n"}
{"abstract": "  When studying the evolution of complex systems one refers to model\nrepresentations comprising various descriptive parameters. There is hardly\nresearch where system evolution is described on the base of information flows\nin the system. The paper focuses on the link between the dynamics of\ninformation and system evolution. Information, exchanged between different\nsystem's parts, before being processed is first provided with meaning by the\nsystem. Meanings are generated from the perspective of hindsight, i.e. against\nthe arrow of time. The same information can be differently interpreted by\ndifferent system's parts (i,e,provided with different meanings) so that the\nnumber of options for possible system development is proliferated. Some options\neventually turn into observable system states. So that system evolutionary\ndynamics can be considered as due to information processing within the system.\nThis process is considered here in a model representation. The model under\nstudy is Triple Helix (TH) model, which was earlier used to describe\ninteractions between university, industry and government to foster innovations.\nIn TH model the system is comprised of three interacting parts where each part\nprocess information ina different way. The model is not limited to the sphere\nof innovation and can be used in a broader perspective. Here TH is\nconceptualized in the framework of three compertment model used to describe\ninfectious disease. The paper demonstrates how the dynamics of information and\nmeaning can be incorporated in the description of Covid-19 infectious\npropagation. The results show correspondence of model predictions with\nobservable infection dynamics.\n"}
{"abstract": "  In the Maximum Minimal Vertex Cover (MMVC) problem, we are given a graph $G$\nand a positive integer $k$, and the objective is to decide whether $G$ contains\na minimal vertex cover of size at least $k$. Motivated by the kernelization of\nMMVC with parameter $k$, our main contribution is to introduce a simple general\nframework to obtain kernelization lower bounds for a certain type of kernels\nfor optimization problems, which we call lop-kernels. Informally, this type of\nkernels is required to preserve large optimal solutions in the reduced\ninstance, and captures the vast majority of existing kernels in the literature.\n  As a consequence of this framework, we show that the trivial quadratic kernel\nfor MMVC is essentially optimal, answering a question of Boria et al. [Discret.\nAppl. Math. 2015], and that the known cubic kernel for Maximum Minimal Feedback\nVertex Set is also essentially optimal. We present further applications for\nTree Deletion Set and for Maximum Independent Set on $K_t$-free graphs.\n  Back to the MMVC problem, given the (plausible) non-existence of subquadratic\nkernels for MMVC on general graphs, we provide subquadratic kernels on $H$-free\ngraphs for several graphs $H$, such as the bull, the paw, or the complete\ngraphs, by making use of the Erd\\\"os-Hajnal property. Finally, we prove that\nMMVC does not admit polynomial kernels parameterized by the size of a minimum\nvertex cover of the input graph, even on bipartite graphs, unless ${\\sf NP}\n\\subseteq {\\sf coNP} / {\\sf poly}$.\n"}
{"abstract": "  This paper proposes parametric and non-parametric hypothesis testing\nalgorithms for detecting anisotropy -- rotational variance of the covariance\nfunction in random fields. Both algorithms are based on resampling mechanisms,\nwhich enable avoiding relying on asymptotic assumptions, as is common in\nprevious algorithms. The algorithms' performance is illustrated numerically in\nsimulation experiments and on real datasets representing a variety of potential\nchallenges.\n"}
{"abstract": "  In this paper, we consider a multi-agent resilient consensus problem, where\nsome of the nodes may behave maliciously. The approach is to equip all nodes\nwith a scheme to detect neighboring nodes when they behave in an abnormal\nfashion. To this end, the nodes exchange not only their own states but also\ninformation regarding their neighbor nodes. Such two-hop communication has long\nbeen studied in fault-tolerant algorithms in computer science. We propose two\ndistributed schemes for detection of malicious nodes and resilient consensus\nwith different requirements on resources for communication and the structures\nof the networks. In particular, the detection schemes become effective under\ncertain connectivity properties in the network so that the non-malicious nodes\ncan share enough information about their neighbors. It is shown that the\nrequirements are however less stringent than those for conventional algorithms.\nA numerical example is presented to demonstrate the performance of the proposed\nmethods in wireless sensor networks.\n"}
{"abstract": "  Recently, the European Commission supported by many European countries has\nannounced large investments towards the commercialization of quantum technology\n(QT) to address and mitigate some of the biggest challenges facing today's\ndigital era - e.g. secure communication and computing power. For more than two\ndecades the QT community has been working on the development of QTs, which\npromise landmark breakthroughs leading to commercialization in various areas.\nThe ambitious goals of the QT community and expectations of EU authorities\ncannot be met solely by individual initiatives of single countries, and\ntherefore, require a combined European effort of large and unprecedented\ndimensions comparable only to the Galileo or Copernicus programs. Strong\ninternational competition calls for a coordinated European effort towards the\ndevelopment of QT in and for space, including research and development of\ntechnology in the areas of communication and sensing. Here, we aim at\nsummarizing the state of the art in the development of quantum technologies\nwhich have an impact in the field of space applications. Our goal is to outline\na complete framework for the design, development, implementation, and\nexploitation of quantum technology in space.\n"}
{"abstract": "  Traditionally, practitioners use formal methods pre-dominately for one half\nof the quality-assurance process: verification (do we build the software\nright?). The other half -- validation (do we build the right software?) -- has\nbeen given comparatively little attention. While verification is the core of\nrefinement-based formal methods, where each new refinement step must preserve\nall properties of its abstract model, validation is usually postponed until the\nlatest stages of the development, when models can be automatically executed.\nThus mistakes in requirements or in their interpretation are caught too late:\nusually at the end of the development process. In this paper, we present a\nnovel approach to check compliance between requirements and their formal\nrefinement-based specification during the earlier stages of development. Our\nproposed approach -- \"validation obligations\" -- is based on the simple idea\nthat both verification and validation are an integral part of all refinement\nsteps of a system.\n"}
{"abstract": "  We study the problem of recovering the common $k$-sized support of a set of\n$n$ samples of dimension $d$, using $m$ noisy linear measurements per sample.\nMost prior work has focused on the case when $m$ exceeds $k$, in which case $n$\nof the order $(k/m)\\log(d/k)$ is both necessary and sufficient. Thus, in this\nregime, only the total number of measurements across the samples matter, and\nthere is not much benefit in getting more than $k$ measurements per sample. In\nthe measurement-constrained regime where we have access to fewer than $k$\nmeasurements per sample, we show an upper bound of $O((k^{2}/m^{2})\\log d)$ on\nthe sample complexity for successful support recovery when $m\\ge 2\\log d$.\nAlong with the lower bound from our previous work, this shows a phase\ntransition for the sample complexity of this problem around $k/m=1$. In fact,\nour proposed algorithm is sample-optimal in both the regimes. It follows that,\nin the $m\\ll k$ regime, multiple measurements from the same sample are more\nvaluable than measurements from different samples.\n"}
{"abstract": "  The 3rd edition of the Montreal AI Ethics Institute's The State of AI Ethics\ncaptures the most relevant developments in AI Ethics since October 2020. It\naims to help anyone, from machine learning experts to human rights activists\nand policymakers, quickly digest and understand the field's ever-changing\ndevelopments. Through research and article summaries, as well as expert\ncommentary, this report distills the research and reporting surrounding various\ndomains related to the ethics of AI, including: algorithmic injustice,\ndiscrimination, ethical AI, labor impacts, misinformation, privacy, risk and\nsecurity, social media, and more.\n  In addition, The State of AI Ethics includes exclusive content written by\nworld-class AI Ethics experts from universities, research institutes,\nconsulting firms, and governments. Unique to this report is \"The Abuse and\nMisogynoir Playbook,\" written by Dr. Katlyn Tuner (Research Scientist, Space\nEnabled Research Group, MIT), Dr. Danielle Wood (Assistant Professor, Program\nin Media Arts and Sciences; Assistant Professor, Aeronautics and Astronautics;\nLead, Space Enabled Research Group, MIT) and Dr. Catherine D'Ignazio (Assistant\nProfessor, Urban Science and Planning; Director, Data + Feminism Lab, MIT). The\npiece (and accompanying infographic), is a deep-dive into the historical and\nsystematic silencing, erasure, and revision of Black women's contributions to\nknowledge and scholarship in the United Stations, and globally. Exposing and\ncountering this Playbook has become increasingly important following the firing\nof AI Ethics expert Dr. Timnit Gebru (and several of her supporters) at Google.\n  This report should be used not only as a point of reference and insight on\nthe latest thinking in the field of AI Ethics, but should also be used as a\ntool for introspection as we aim to foster a more nuanced conversation\nregarding the impacts of AI on the world.\n"}
{"abstract": "  We consider the random $\\beta$-transformation $K_{\\beta}$, defined on\n$\\{0,1\\}^{\\mathbb N}\\times[0, \\frac{\\lfloor\\beta\\rfloor}{\\beta-1}]$, that\ngenerates all possible expansions of the form\n$x=\\sum_{i=0}^{\\infty}\\frac{a_i}{\\beta^i}$, where $a_i\\in\n\\{0,1,\\cdots,\\lfloor\\beta\\rfloor\\}$. This transformation was first introduced\nby Dajani and Kraaikamp, and later studied by Dajani and de Vries, where two\nnatural invariant ergodic measures were found. The first is the unique measure\nof maximal entropy, and the second is a measure of the form $m_p\\times\n\\mu_{\\beta}$, with $m_p$ the Bernoulli $(p,1-p)$ product measure and\n$\\mu_{\\beta}$ is a measure equivalent to Lebesgue measure. In this paper, we\ngive an uncountable family of $K_{\\beta}$-invariant exact $g$-measures for a\ncertain collection of algebraic $\\beta$'s.\n"}
{"abstract": "  This paper describes the solution of Shanda Innovations team to Task 1 of\nKDD-Cup 2012. A novel approach called Multifaceted Factorization Models is\nproposed to incorporate a great variety of features in social networks. Social\nrelationships and actions between users are integrated as implicit feedbacks to\nimprove the recommendation accuracy. Keywords, tags, profiles, time and some\nother features are also utilized for modeling user interests. In addition, user\nbehaviors are modeled from the durations of recommendation records. A\ncontext-aware ensemble framework is then applied to combine multiple predictors\nand produce final recommendation results. The proposed approach obtained\n0.43959 (public score) / 0.41874 (private score) on the testing dataset, which\nachieved the 2nd place in the KDD-Cup competition.\n"}
{"abstract": "  Since human-labeled samples are free for the target set, unsupervised person\nre-identification (Re-ID) has attracted much attention in recent years, by\nadditionally exploiting the source set. However, due to the differences on\ncamera styles, illumination and backgrounds, there exists a large gap between\nsource domain and target domain, introducing a great challenge on cross-domain\nmatching. To tackle this problem, in this paper we propose a novel method named\nDual-stream Reciprocal Disentanglement Learning (DRDL), which is quite\nefficient in learning domain-invariant features. In DRDL, two encoders are\nfirst constructed for id-related and id-unrelated feature extractions, which\nare respectively measured by their associated classifiers. Furthermore,\nfollowed by an adversarial learning strategy, both streams reciprocally and\npositively effect each other, so that the id-related features and id-unrelated\nfeatures are completely disentangled from a given image, allowing the encoder\nto be powerful enough to obtain the discriminative but domain-invariant\nfeatures. In contrast to existing approaches, our proposed method is free from\nimage generation, which not only reduces the computational complexity\nremarkably, but also removes redundant information from id-related features.\nExtensive experiments substantiate the superiority of our proposed method\ncompared with the state-of-the-arts. The source code has been released in\nhttps://github.com/lhf12278/DRDL.\n"}
{"abstract": "  A boost in the development of flexible and wearable electronics facilitates\nthe design of new materials to be applied as transparent conducting films\n(TCFs). Although single-walled carbon nanotube (SWCNT) films are the most\npromising candidates for flexible TCFs, they still do not meet optoelectronic\nrequirements demanded their successful industrial integration. In this study,\nwe proposed and thoroughly investigated a new approach that comprises\nsimultaneous bilateral (outer and inner surfaces) SWCNT doping after their\nopening by thermal treatment at 400 C under an ambient air atmosphere. Doping\nby a chloroauric acid (HAuCl$_{4}$) ethanol solution allowed us to achieve the\nrecord value of sheet resistance of 31 $\\pm$ 4 $\\Omega$/sq at a transmittance\nof 90% in the middle of visible spectra (550 nm). The strong p-doping was\nexamined by open-circuit potential (OCP) measurements and confirmed by ab\ninitio calculations demonstrating a downshift of Fermi level around 1 eV for\nthe case of bilateral doping.\n"}
{"abstract": "  In this paper, we consider the Cauchy problem for semi-linear wave equations\nwith structural damping term $\\nu (-\\Delta)^2 u_t$, where $\\nu >0$ is a\nconstant. As being mentioned in [8,10], the linear principal part brings both\nthe diffusion phenomenon and the regularity loss of solutions. This implies\nthat, for the nonlinear problems, the choice of solution spaces plays an\nimportant role to obtain global solutions with sharp decay properties in time.\nOur main purpose of this paper is to prove the global (in time) existence of\nsolutions for the small data and their decay properties for the supercritical\nnonlinearities.\n"}
{"abstract": "  We investigate several aspects of BPS latitude Wilson loops in gauge theories\nin three dimensions with $\\mathcal{N}\\ge 4$ supersymmetry. We derive a matrix\nmodel for the bosonic latitude Wilson loop in ABJM using supersymmetric\nlocalization, and show how to extend the computation to more general\nChern-Simons-matter theories. We then define latitude type Wilson and vortex\nloop operators in theories without Chern-Simons terms, and explore a connection\nto the recently derived superalgebra defining local Higgs and Coulomb branch\noperators in these theories. Finally, we identify a BPS loop operator dual to\nthe bosonic latitude Wilson loop which is a novel bound state of Wilson and\nvortex loops, defined using a worldvolume supersymmetric quantum mechanics.\n"}
{"abstract": "  In this paper we derive closed-form formulas of feedback capacity and\nnonfeedback achievable rates, for Additive Gaussian Noise (AGN) channels driven\nby nonstationary autoregressive moving average (ARMA) noise (with unstable one\npoles and zeros), based on time-invariant feedback codes and channel input\ndistributions. From the analysis and simulations follows the surprising\nobservations, (i) the use of time-invariant channel input distributions gives\nrise to multiple regimes of capacity that depend on the parameters of the ARMA\nnoise, which may or may not use feedback, (ii) the more unstable the pole\n(resp. zero) of the ARMA noise the higher (resp. lower) the feedback capacity,\n(iii) certain conditions, known as detectability and stabilizability are\nnecessary and sufficient to ensure the feedback capacity formulas and\nnonfeedback achievable rates {\\it are independent of the initial state of the\nARMA noise}. Another surprizing observation is that Kim's \\cite{kim2010}\ncharacterization of feedback capacity which is developed for stable ARMA noise,\nif applied to the unstable ARMA noise, gives a lower value of feedback capacity\ncompared to our feedback capacity formula.\n"}
{"abstract": "  Let $X_{1,n}\\le\\cdots\\le X_{n,n}$ be the order statistics of $n$ independent\nrandom variables with a common distribution function $F$ having right heavy\ntail with tail index $\\gamma$. Given known constants $d_{i,n}$, $1\\le i\\le n$,\nconsider the weighted power sums\n$\\sum^{k_n}_{i=1}d_{n+1-i,n}\\log^pX_{n+1-i,n}$, where $p>0$ and the $k_n$ are\npositive integers such that $k_n\\to\\infty$ and $k_n/n\\to0$ as $n\\to\\infty$.\nUnder some constraints on the weights $d_{i,n}$, we prove asymptotic normality\nfor the power sums over the whole heavy-tail model. We apply the obtained\nresult to construct a new class of estimators for the parameter $\\gamma$.\n"}
{"abstract": "  The magnetic damping constant is a critical parameter for magnetization\ndynamics and the efficiency of memory devices and magnon transport. Therefore,\nits manipulation by electric fields is crucial in spintronics. Here, we\ntheoretically demonstrate the voltage-control of magnetic damping in ferro- and\nferrimagnetic-insulator (FI)/topological-insulator (TI) bilayers. Assuming a\ncapacitor-like setup, we formulate an effective dissipation torque induced by\nspin-charge pumping at the FI/TI interface as a function of an applied voltage.\nBy using realistic material parameters, we find that the effective damping for\na FI with 10nm thickness can be tuned by one order of magnitude under the\nvoltage with 0.25V. Also, we provide perspectives on the voltage-induced\nmodulation of the magnon spin transport on proximity-coupled FIs.\n"}
{"abstract": "  The gradient flow exact renormalization group (GFERG) is an exact\nrenormalization group motivated by the Yang--Mills gradient flow and its\nsalient feature is a manifest gauge invariance. We generalize this GFERG,\noriginally formulated for the pure Yang--Mills theory, to vector-like gauge\ntheories containing fermion fields, keeping the manifest gauge invariance. For\nthe chiral symmetry we have two options: one possible formulation preserves the\nconventional form of the chiral symmetry and the other simpler formulation\nrealizes the chiral symmetry in a modified form \\`a la Ginsparg--Wilson. We\nwork out a gauge-invariant local Wilson action in quantum electrodynamics to\nthe lowest nontrivial order of perturbation theory. This Wilson action\nreproduces the correct axial anomaly in~$D=2$.\n"}
{"abstract": "  A detailed theoretical analysis of the 1s photoionization of neon is\npresented. It is found that the most significant many-electron correlation in\ncomputing photoionization of inner shells is the rearrangement of the outer\nshells caused by the inner vacancy. Further noticeable effects are: (i) the\npolarization of the ion core by the outgoing photoelectron and (ii) the\ncoherent effect of double excitation/ionization. The core polarization\nincreases the photoionization cross section by about 10% at the 1s threshold,\nand the coherent excitation results in further increases by about 5%.\nIncoherent excitation of the satellite channel leads to an additional 10%\nincrease in the photoabsorption cross section in the double-ionization\nthreshold region.\n"}
{"abstract": "  We study the self-tuning subclass of kinetic gravity braiding and obtain\nrobust predictions on self-tuning and dynamics in the tadpole-free shift\nsymmetric sector of the theory. In particular, we show inevitability of cosmic\nacceleration, prove the dynamical stability of this late-time asymptotic state,\nand derive ghost and gradient stability constraints on the self-tuning vacuum.\nWe discuss the results concretely in the context of generalized cubic covariant\nGalileon theory and an exponential kinetic gravity braiding.\n"}
{"abstract": "  On April 7, 2001 the Muon g-2 Experiment at Fermilab presented its first\nresults which lead to a 4.2$\\sigma$ discrepancy between the world average of\nthe muon anomalous magnetic dipole moment and the standard model prediction of\nthis quantity. I take issue with the statistical meaning of this finding.\n"}
{"abstract": "  In this paper we suggest NICE: a new algorithm to generate counterfactual\nexplanations for heterogeneous tabular data. The design of our algorithm\nspecifically takes into account algorithmic requirements that often emerge in\nreal-life deployments: the ability to provide an explanation for all\npredictions, being efficient in run-time, and being able to handle any\nclassification model (also non-differentiable ones). More specifically, our\napproach exploits information from a nearest instance tospeed up the search\nprocess. We propose four versions of NICE, where three of them optimize the\nexplanations for one of the following properties: sparsity, proximity or\nplausibility. An extensive empirical comparison on 10 datasets shows that our\nalgorithm performs better on all properties than the current state-of-the-art.\nThese analyses show a trade-off between on the one hand plausiblity and on the\nother hand proximity or sparsity, with our different optimization methods\noffering the choice to select the preferred trade-off. An open-source\nimplementation of NICE can be found at https://github.com/ADMAntwerp/NICE.\n"}
{"abstract": "  Responsible development of technology involves applications being inclusive\nof the diverse set of users they hope to support. An important part of this is\nunderstanding the many ways to refer to a person and being able to fluently\nchange between the different forms as needed. We perform a case study on the\nsingular they, a common way to promote gender inclusion in English. We define a\nre-writing task, create an evaluation benchmark, and show how a model can be\ntrained to produce gender-neutral English with <1% word error rate with no\nhuman-labeled data. We discuss the practical applications and ethical\nconsiderations of the task, providing direction for future work into inclusive\nnatural language systems.\n"}
{"abstract": "  Physics of self-propelled objects at the nanoscale is a rapidly developing\nresearch field where recent experiments focused on motion of individual\ncatalytic enzymes. Contrary to the experimental advancements, theoretical\nunderstanding of possible self-propulsion mechanisms at these scales is\nlimited. A particularly puzzling question concerns origins of reportedly high\ndiffusivities of the individual enzymes. Here we start with the fundamental\nprinciple of microscopic reversibility (MR) of chemical reactions powering the\nself-propulsion and demonstrate that MR can lead to increase of the particle\nmobility and of short- and long-time diffusion coefficients as compared to\ndynamics where MR is neglected. Moreover, the diffusion coefficients can be\nenhanced by a constant external force. We propose a way to use these effects in\nexperimental investigations of active propulsion mechanisms at the nanoscale.\nOur results can shed new light on interpretation of the measured diffusivities\nand help to test and to evaluate relevance of MR for the active motion of\nindividual nanoswimmers.\n"}
{"abstract": "  The current event rate estimates of long gamma-ray bursts based on distinct\nmethods or samples especially at lower redshift are largely debated, which\nmotivates us to re-study the dependence of luminosity function and event rates\nfor different burst samples on the criteria of sample selection and threshold\neffect in this letter. To ensure the sample completeness as possible, we have\nchosen two samples including 88 and 118 long bright bursts with known redshift\nand peak flux over 2.6 ph cm$^{-2}$ s$^{-1}$. It is found that the evolution of\nluminosity with redshift can be expressed by $L\\propto(1+z)^k$ with a diverse\n$k$ relied more on the sample selection. Interestingly, the cumulative\ndistributions of either non-evolving luminosities or redshifts are found to be\nalso determined by the sample selection rather the instrumental sensitivity.\nNevertheless, the non-evolving luminosities of our samples are similarly\ndistributed with a comparable break luminosity of $L_0\\sim10^{51}$ erg\ns$^{-1}$. Importantly, we verify with a K-S test that three cases of event\nrates for the two burst samples evolve with redshift similarly except a small\ndiscrepancy due to sampling differences at low-redshift of $z<1$, in which all\nevent rates show an excess of gaussian profile instead of monotonous decline.\nMost importantly, it is found that the low-redshift burst event rates violate\nthe star formation rates, while both of them are good in agreement with each\nother in the higher-redshift regions as many authors discovered previously.\nConsequently, we predict that two types of long gamma-ray bursts should be\nexpected on the basis of whether they match the star formation or not.\n"}
{"abstract": "  Voltage control of magnetism and spintronics have been highly desirable, but\nrarely realized. In this work, we show voltage-controlled spin-orbit torque\n(SOT) switching in W/CoFeB/MgO films with perpendicular magnetic anisotropy\n(PMA) with voltage administered through SrTiO3 with a high dielectric constant.\nWe show that a DC voltage can significantly lower PMA by 45%, reduce switching\ncurrent by 23%, and increase the damping-like torque as revealed by the first\nand second-harmonic measurements. These are characteristics that are\nprerequisites for voltage-controlled and voltage-select SOT switching\nspintronic devices.\n"}
{"abstract": "  We present a link between the theory of deep water waves and that of bubble\nsurface perturbations. Theory correspondence is shown analytically for small\nwavelengths in the linear regime and investigated numerically in the nonlinear\nregime. To do so, we develop the second-order spatial perturbation equations\nfor the Rayleigh-Plesset equation and solve them numerically. Our code is\npublicly available. Studying capillary waves on stable bubbles, we recreate the\nKolmogorov-Zakharov spectrum predicted by weak turbulence theory, putting wave\nturbulence theory to use for bubbles. In this investigation, it seems that\ncurvature does not affect turbulent properties. Calculated bubble surface\nqualitatively responds to low gravity experiments. The link demonstrated opens\nnew possibilities for studying several bubble phenomena, including\nsonoluminescence and cavitation, using the extensive tools developed in the\nwave turbulence framework.\n"}
{"abstract": "  We consider one-dimensional infinite chains of harmonic oscillators with\nrandom exchanges of momenta and long-range interaction potentials which have\npolynomial decay rate $|x|^{-\\theta}, x \\to \\infty, \\theta > 1$ where $x \\in\n\\mathbb{Z}$ is the interaction range. The dynamics conserve total momentum,\ntotal length and total energy. We prove that the systems evolve macroscopically\non superballistic space-time scale $(y \\varepsilon^{-1}, t \\varepsilon^{-\n\\frac{\\theta - 1}{2}})$ when $1 < \\theta < 3$, $(y \\varepsilon^{-1}, t\n\\varepsilon^{-1} \\sqrt{- \\log(\\varepsilon^{-1})}^{-1})$ when $\\theta = 3$, and\nballistic space-time scale $(y \\varepsilon^{-1}, t \\varepsilon^{-1})$ when\n$\\theta > 3$. Combining our results and the results in [10], we show the\nexistence of two different space-time scales on which the systems evolve. In\naddition, we prove scaling limits of recentered normal modes of superballistic\nwave equations, which are analogues of Riemann invariants and capture\nfluctuations around characteristics. The space-time scale is superdiffusive\nwhen $2 < \\theta \\le 4$ and diffusive when $\\theta > 4$.\n"}
{"abstract": "  This dissertation presents a semiclassical analysis of conical topology\nchange in $1+1$ spacetime dimensions wherein, to lowest order, the ambient\nspacetime is classical and fixed while the scalar field coupled to it is\nquantized. The vacuum expectation value of the scalar field stress-energy\ntensor is calculated via two different approaches. The first of these involves\nthe explicit determination of the so called Sorkin-Johnston state on the cone\nand an original regularization scheme, while the latter employs the conformal\nvacuum and the more conventional point-splitting renormalization. It is found\nthat conical topology change seems not to suffer from the same pathologies that\ntrousers-type topology change does. This provides tentative agreement with\nconjectures due to Sorkin and Borde, which attempt to classify topology\nchanging spacetimes with respect to their Morse critical points and in\nparticular, that the cone and yarmulke in $1+1$ dimensions lack critical points\nof unit Morse index.\n"}
{"abstract": "  Reservoir computing is a best-in-class machine learning algorithm for\nprocessing information generated by dynamical systems using observed\ntime-series data. Importantly, it requires very small training data sets, uses\nlinear optimization, and thus requires minimal computing resources. However,\nthe algorithm uses randomly sampled matrices to define the underlying recurrent\nneural network and has a multitude of metaparameters that must be optimized.\nRecent results demonstrate the equivalence of reservoir computing to nonlinear\nvector autoregression, which requires no random matrices, fewer metaparameters,\nand provides interpretable results. Here, we demonstrate that nonlinear vector\nautoregression excels at reservoir computing benchmark tasks and requires even\nshorter training data sets and training time, heralding the next generation of\nreservoir computing.\n"}
{"abstract": "  Fireballs created in relativistic heavy-ion collisions at different beam\nenergies have been argued to follow different trajectories in the QCD phase\ndiagram in which the QCD critical point serves as a landmark. Using a\n(1+1)-dimensional model setting with transverse homogeneity, we study the\ncomplexities introduced by the fact that the evolution history of each fireball\ncannot be characterized by a single trajectory but rather covers an entire\nswath of the phase diagram, with the finally emitted hadron spectra integrating\nover contributions from many different trajectories. Studying the phase diagram\ntrajectories of fluid cells at different space-time rapidities, we explore how\nbaryon diffusion shuffles them around, and how they are affected by critical\ndynamics near the QCD critical point. We find a striking insensitivity of\nbaryon diffusion to critical effects. Its origins are analyzed and possible\nimplications discussed.\n"}
{"abstract": "  In many computer vision classification tasks, class priors at test time often\ndiffer from priors on the training set. In the case of such prior shift,\nclassifiers must be adapted correspondingly to maintain close to optimal\nperformance. This paper analyzes methods for adaptation of probabilistic\nclassifiers to new priors and for estimating new priors on an unlabeled test\nset. We propose a novel method to address a known issue of prior estimation\nmethods based on confusion matrices, where inconsistent estimates of decision\nprobabilities and confusion matrices lead to negative values in the estimated\npriors. Experiments on fine-grained image classification datasets provide\ninsight into the best practice of prior shift estimation and classifier\nadaptation, and show that the proposed method achieves state-of-the-art results\nin prior adaptation. Applying the best practice to two tasks with naturally\nimbalanced priors, learning from web-crawled images and plant species\nclassification, increased the recognition accuracy by 1.1% and 3.4%\nrespectively.\n"}
{"abstract": "  We introduce a hybrid method to couple continuous Galerkin finite element\nmethods and high-order finite difference methods in a nonconforming multiblock\nfashion. The aim is to optimize computational efficiency when complex\ngeometries are present. The proposed coupling technique requires minimal\nchanges in the existing schemes while maintaining strict stability, accuracy,\nand energy conservation. Results are demonstrated on linear and nonlinear\nscalar conservation laws in two spatial dimensions.\n"}
{"abstract": "  We calculate the collinear anomalous dimensions in massless four-loop QCD and\n$\\mathcal{N} = 4$ supersymmetric Yang-Mills theory from the infrared poles of\nvertex form factors. We give very precise numerical approximations and a\nconjecture for the complete analytic results in both models we consider.\n"}
{"abstract": "  Communication technologies like voice over IP operate under constrained\nreal-time conditions, with voice packets being subject to delays and losses\nfrom the network. In such cases, the packet loss concealment (PLC) algorithm\nreconstructs missing frames until a new real packet is received. Recently,\nautoregressive deep neural networks have been shown to surpass the quality of\nsignal processing methods for PLC, specially for long-term predictions beyond\n60 ms. In this work, we propose a non-autoregressive adversarial auto-encoder,\nnamed PLAAE, to perform real-time PLC in the waveform domain. PLAAE has a\ncausal convolutional structure, and it learns in an auto-encoder fashion to\nreconstruct signals with gaps, with the help of an adversarial loss. During\ninference, it is able to predict smooth and coherent continuations of such gaps\nin a single feed-forward step, as opposed to autoregressive models. Our\nevaluation highlights the superiority of PLAAE over two classic PLCs and two\ndeep autoregressive models in terms of spectral and intonation reconstruction,\nperceptual quality, and intelligibility.\n"}
{"abstract": "  Informal lecture notes with examples on sheaf theory and the derived category\nof sheaves; sheaves and Morse theory; perverse sheaves, and some applications\nto representation theory. Added Oct 2021: cellular perverse sheaves. Proofs are\noutlined with references for further details.\n"}
{"abstract": "  Facial expression spotting is the preliminary step for micro- and\nmacro-expression analysis. The task of reliably spotting such expressions in\nvideo sequences is currently unsolved. The current best systems depend upon\noptical flow methods to extract regional motion features, before categorisation\nof that motion into a specific class of facial movement. Optical flow is\nsusceptible to drift error, which introduces a serious problem for motions with\nlong-term dependencies, such as high frame-rate macro-expression. We propose a\npurely deep learning solution which, rather than tracking frame differential\nmotion, compares via a convolutional model, each frame with two temporally\nlocal reference frames. Reference frames are sampled according to calculated\nmicro- and macro-expression durations. We show that our solution achieves\nstate-of-the-art performance (F1-score of 0.105) in a dataset of high\nframe-rate (200 fps) long video sequences (SAMM-LV) and is competitive in a low\nframe-rate (30 fps) dataset (CAS(ME)2). In this paper, we document our deep\nlearning model and parameters, including how we use local contrast\nnormalisation, which we show is critical for optimal results. We surpass a\nlimitation in existing methods, and advance the state of deep learning in the\ndomain of facial expression spotting.\n"}
{"abstract": "  This paper presents new \\emph{variance-aware} confidence sets for linear\nbandits and linear mixture Markov Decision Processes (MDPs). With the new\nconfidence sets, we obtain the follow regret bounds: For linear bandits, we\nobtain an $\\tilde{O}(poly(d)\\sqrt{1 + \\sum_{k=1}^{K}\\sigma_k^2})$\ndata-dependent regret bound, where $d$ is the feature dimension, $K$ is the\nnumber of rounds, and $\\sigma_k^2$ is the \\emph{unknown} variance of the reward\nat the $k$-th round. This is the first regret bound that only scales with the\nvariance and the dimension but \\emph{no explicit polynomial dependency on $K$}.\nWhen variances are small, this bound can be significantly smaller than the\n$\\tilde{\\Theta}\\left(d\\sqrt{K}\\right)$ worst-case regret bound. For linear\nmixture MDPs, we obtain an $\\tilde{O}(poly(d, \\log H)\\sqrt{K})$ regret bound,\nwhere $d$ is the number of base models, $K$ is the number of episodes, and $H$\nis the planning horizon. This is the first regret bound that only scales\n\\emph{logarithmically} with $H$ in the reinforcement learning with linear\nfunction approximation setting, thus \\emph{exponentially improving} existing\nresults, and resolving an open problem in \\citep{zhou2020nearly}. We develop\nthree technical ideas that may be of independent interest: 1) applications of\nthe peeling technique to both the input norm and the variance magnitude, 2) a\nrecursion-based estimator for the variance, and 3) a new convex potential lemma\nthat generalizes the seminal elliptical potential lemma.\n"}
{"abstract": "  A theoretical framework and numerical simulations quantifying the impact of\natmospheric group velocity dispersion on wireless terahertz communication link\nerror rate were developed based upon experimental work. We present, for the\nfirst time, predictions of symbol error rate as a function of link distance,\nsignal bandwidth, signal-to-noise ratio, and atmospheric conditions, revealing\nthat long-distance, broadband terahertz communication systems may be limited by\ninter-symbol interference stemming from group velocity dispersion, rather than\nattenuation. In such dispersion limited links, increasing signal strength does\nnot improve the symbol error rate and, consequently, theoretical predictions of\nsymbol error rate based only on signal-to-noise ratio are invalid for the\nbroadband case. This work establishes a new and necessary foundation for link\nbudget analysis in future long-distance terahertz communication systems that\naccounts for the non-negligible effects of both attenuation and dispersion.\n"}
{"abstract": "  We define a spatially-dependent fragmentation process, which involves\nrectangles breaking up into progressively smaller pieces at rates that depend\non their shape. Long, thin rectangles are more likely to break quickly, and are\nalso more likely to split along their longest side. We are interested in how\nthe system evolves over time: how many fragments are there of different shapes\nand sizes, and how did they reach that state? Our theorem gives an almost sure\ngrowth rate along paths, which does not match the growth rate in expectation -\nthere are paths where the expected number of fragments of that shape and size\nis exponentially large, but in reality no such fragments exist at large times\nalmost surely.\n"}
{"abstract": "  WhatsApp is the most popular messaging app in the world. Due to its\npopularity, WhatsApp has become a powerful and cheap tool for political\ncampaigning being widely used during the 2019 Indian general election, where it\nwas used to connect to the voters on a large scale. Along with the campaigning,\nthere have been reports that WhatsApp has also become a breeding ground for\nharmful speech against various protected groups and religious minorities. Many\nsuch messages attempt to instil fear among the population about a specific\n(minority) community. According to research on inter-group conflict, such `fear\nspeech' messages could have a lasting impact and might lead to real offline\nviolence. In this paper, we perform the first large scale study on fear speech\nacross thousands of public WhatsApp groups discussing politics in India. We\ncurate a new dataset and try to characterize fear speech from this dataset. We\nobserve that users writing fear speech messages use various events and symbols\nto create the illusion of fear among the reader about a target community. We\nbuild models to classify fear speech and observe that current state-of-the-art\nNLP models do not perform well at this task. Fear speech messages tend to\nspread faster and could potentially go undetected by classifiers built to\ndetect traditional toxic speech due to their low toxic nature. Finally, using a\nnovel methodology to target users with Facebook ads, we conduct a survey among\nthe users of these WhatsApp groups to understand the types of users who consume\nand share fear speech. We believe that this work opens up new research\nquestions that are very different from tackling hate speech which the research\ncommunity has been traditionally involved in.\n"}
{"abstract": "  The quantification of uncertainties of computer simulations due to input\nparameter uncertainties is paramount to assess a model's credibility. For\ncomputationally expensive simulations, this is often feasible only via\nsurrogate models that are learned from a small set of simulation samples. The\nsurrogate models are commonly chosen and deemed trustworthy based on heuristic\nmeasures, and substituted for the simulation in order to approximately\npropagate the simulation input uncertainties to the simulation output. In the\nprocess, the contribution of the uncertainties of the surrogate itself to the\nsimulation output uncertainties are usually neglected. In this work, we\nspecifically address the case of doubtful surrogate trustworthiness, i.e.\nnon-negligible surrogate uncertainties. We find that Bayesian probability\ntheory yields a natural measure of surrogate trustworthiness, and that\nsurrogate uncertainties can easily be included in simulation output\nuncertainties. For a Gaussian likelihood for the simulation data, with unknown\nsurrogate variance and given a generalized linear surrogate model, the\nresulting formulas reduce to simple matrix multiplications. The framework\ncontains Polynomial Chaos Expansions as a special case, and is easily extended\nto Gaussian Process Regression. Additionally, we show a simple way to\nimplicitly include spatio-temporal correlations. Lastly, we demonstrate a\nnumerical example where surrogate uncertainties are in part negligible and in\npart non-negligible.\n"}
{"abstract": "  Classical regression has a simple geometric description in terms of a\nprojection of the training labels onto the column space of the design matrix.\nHowever, for over-parameterized models -- where the number of fit parameters is\nlarge enough to perfectly fit the training data -- this picture becomes\nuninformative. Here, we present an alternative geometric interpretation of\nregression that applies to both under- and over-parameterized models. Unlike\nthe classical picture which takes place in the space of training labels, our\nnew picture resides in the space of input features. This new feature-based\nperspective provides a natural geometric interpretation of the double-descent\nphenomenon in the context of bias and variance, explaining why it can occur\neven in the absence of label noise. Furthermore, we show that adversarial\nperturbations -- small perturbations to the input features that result in large\nchanges in label values -- are a generic feature of biased models, arising from\nthe underlying geometry. We demonstrate these ideas by analyzing three minimal\nmodels for over-parameterized linear least squares regression: without basis\nfunctions (input features equal model features) and with linear or nonlinear\nbasis functions (two-layer neural networks with linear or nonlinear activation\nfunctions, respectively).\n"}
{"abstract": "  Clustering artworks is difficult for several reasons. On the one hand,\nrecognizing meaningful patterns based on domain knowledge and visual perception\nis extremely hard. On the other hand, applying traditional clustering and\nfeature reduction techniques to the highly dimensional pixel space can be\nineffective. To address these issues, in this paper we propose DELIUS: a DEep\nlearning approach to cLustering vIsUal artS. The method uses a pre-trained\nconvolutional network to extract features and then feeds these features into a\ndeep embedded clustering model, where the task of mapping the raw input data to\na latent space is jointly optimized with the task of finding a set of cluster\ncentroids in this latent space. Quantitative and qualitative experimental\nresults show the effectiveness of the proposed method. DELIUS can be useful for\nseveral tasks related to art analysis, in particular visual link retrieval and\nhistorical knowledge discovery in painting datasets.\n"}
{"abstract": "  This paper describes the operation of the Coherent CAPTAIN-Mills (CCM)\ndetector located in the Lujan Neutron Science Center at Los Alamos National\nLaboratory. CCM is a 10-ton liquid argon (LAr) detector located 20 meters from\na high flux neutron/neutrino source and is designed to search for sterile\nneutrinos and light dark matter. An engineering run was performed in Fall 2019\nto study the characteristics of the CCM120 detector by searching for signals\nconsistent with sterile neutrinos and light dark matter resulting from $\\pi^+$\nand $\\pi^0$ decays in the tungsten target. New parameter space in a leptophobic\ndark matter model was excluded for dark matter masses between $\\sim2.5$ and 60\nMeV. The lessons learned from this run have guided the development and\nconstruction of the new CCM200 detector that will begin operations in 2021 and\nsignificantly improve on these searches.\n"}
{"abstract": "  Let $F:\\mathcal{A}\\to \\mathcal{B}$ be a left adjoint between abelian\ncategories and let $Ch(F)$ be the induced left adjoint on chain complexes. If\nthe abelian categories $\\mathcal{A}$ and $\\mathcal{B}$ are equipped with\nsufficiently nice cotorsion pairs, then we find model structures on their\ncategories of chain complexes. This paper gives novel criteria under which the\ninduced adjunction becomes a Quillen adjunction. In particular, our criteria\ncan be checked on the level of the underlying abelian categories instead of\nchain complexes. We do the same for adjunctions of $n$ variables and we also\nshow how our results imply that the flat model structure can be used to compute\nthe derived tensor product.\n"}
{"abstract": "  Existing algorithms with iterations as the principle for 3D inverse heat\nconduction problems (IHCPs) are usually time-consuming. With the recent\nadvancements in deep learning techniques, it is possible to apply the neural\nnetwork to compute IHCPs. In this paper, a new framework based on\nConvolutional-LSTM is introduced to predict the transient heat flux via\nmeasured temperature. The inverse heat conduction models concerned in this work\nhave 3D complex structures with non-linear boundary conditions and\nthermophysical parameters. In order to reach high precision, a forward solver\nbased on the finite element method is utilized to generate sufficient data for\ntraining. The fully trained framework can provide accurate predictions\nefficiently once the measured temperature and models are acquired. It is\nbelieved that the proposed framework offers a new pattern for real-time heat\nflux inversion.\n"}
{"abstract": "  The origin of supermassive black holes (with $\\gtrsim\\!10^9\\,M_{\\odot}$) in\nthe early universe (redshift $z \\sim 7$) remains poorly understood.\nGravitational collapse of a massive primordial gas cloud is a promising initial\nprocess, but theoretical studies have difficulty growing the black hole fast\nenough. We focus on the magnetic effects on star formation that occurs in an\natomic-cooling gas cloud. Using a set of three-dimensional magnetohydrodynamic\n(MHD) simulations, we investigate the star formation process in the magnetized\natomic-cooling gas cloud with different initial magnetic field strengths. Our\nsimulations show that the primordial magnetic seed field can be quickly\namplified during the early accretion phase after the first protostar formation.\nThe strong magnetic field efficiently extracts angular momentum from accreting\ngas and increases the accretion rate, which results in the high fragmentation\nrate in the gravitationally unstable disk region. On the other hand, the\ncoalescence rate of fragments is also enhanced by the angular momentum transfer\ndue to the magnetic effects. Almost all the fragments coalesce to the primary\nstar, so the mass growth rate of the massive star increases due to the magnetic\neffects. We conclude that the magnetic effects support the direct collapse\nscenario of supermassive star formation.\n"}
{"abstract": "  It has been recently shown how the tensorial nature of real-time path\nintegrals involving the Feynman-Vernon influence functional can be utilized\nusing matrix product states, taking advantage of the finite length of the\nnon-Markovian memory. Tensor networks promise to provide a new, unified\nlanguage to express the structure of path integral. Here, a generalized tensor\nnetwork is derived and implemented specifically incorporating the pairwise\ninteraction structure of the influence functional, allowing for a compact\nrepresentation and efficient evaluation. This pairwise connected tensor network\npath integral (PCTNPI) is illustrated through applications to typical\nspin-boson problems and explorations of the differences caused by the exact\nform of the spectral density. The storage requirements and performance are\ncompared with iterative quasi-adiabatic propagator path integral and iterative\nblip-summed path integral. Finally, the viability of using PCTNPI for\nsimulating multistate problems is demonstrated taking advantage of the\ncompressed representation.\n"}
{"abstract": "  Deep neural networks often suffer from overconfidence which can be partly\nremedied by improved out-of-distribution detection. For this purpose, we\npropose a novel approach that allows for the generation of out-of-distribution\ndatasets based on a given in-distribution dataset. This new dataset can then be\nused to improve out-of-distribution detection for the given dataset and machine\nlearning task at hand. The samples in this dataset are with respect to the\nfeature space close to the in-distribution dataset and therefore realistic and\nplausible. Hence, this dataset can also be used to safeguard neural networks,\ni.e., to validate the generalization performance. Our approach first generates\nsuitable representations of an in-distribution dataset using an autoencoder and\nthen transforms them using our novel proposed Soft Brownian Offset method.\nAfter transformation, the decoder part of the autoencoder allows for the\ngeneration of these implicit out-of-distribution samples. This newly generated\ndataset then allows for mixing with other datasets and thus improved training\nof an out-of-distribution classifier, increasing its performance.\nExperimentally, we show that our approach is promising for time series using\nsynthetic data. Using our new method, we also show in a quantitative case study\nthat we can improve the out-of-distribution detection for the MNIST dataset.\nFinally, we provide another case study on the synthetic generation of\nout-of-distribution trajectories, which can be used to validate trajectory\nprediction algorithms for automated driving.\n"}
{"abstract": "  We establish mean convergence for multiple ergodic averages with iterates\ngiven by distinct fractional powers of primes and related multiple recurrence\nresults. A consequence of our main result is that every set of integers with\npositive upper density contains patterns of the form $\\{m,m+[p_n^a],\nm+[p_n^b]\\}$, where $a,b$ are positive non-integers and $p_n$ denotes the\n$n$-th prime, a property that fails if $a$ or $b$ is a natural number. Our\napproach is based on a recent criterion for joint ergodicity of collections of\nsequences and the bulk of the proof is devoted to obtaining good seminorm\nestimates for the related multiple ergodic averages. The input needed from\nnumber theory are upper bounds for the number of prime $k$-tuples that follow\nfrom elementary sieve theory estimates and equidistribution results of\nfractional powers of primes in the circle.\n"}
{"abstract": "  We construct Schwinger functions as the superposition of Schwinger functions\nwhich correspond to those of free fields with sharp masses $ m $. We prove that\nall axioms of Osterwalder and Schrader are satisfied. This construction works\nindependently of the space-time dimension $ d $. The Schwinger functions under\nconsideration are the moments of a non-Gaussian measure on the space of\ntempered distributions $ S'(\\mathbb R^d) $.\n"}
{"abstract": "  Given an unknown dynamic system such as a coupled harmonic oscillator with\n$n$ springs and point masses. We are often interested in gaining insights into\nits physical parameters, i.e. stiffnesses and masses, by observing trajectories\nof motion. How do we achieve this from video frames or time-series data and\nwithout the knowledge of the dynamics model? We present a neural framework for\nestimating physical parameters in a manner consistent with the underlying\nphysics. The neural framework uses a deep latent variable model to disentangle\nthe system physical parameters from canonical coordinate observations. It then\nreturns a Hamiltonian parameterization that generalizes well with respect to\nthe discovered physical parameters. We tested our framework with simple\nharmonic oscillators, $n=1$, and noisy observations and show that it discovers\nthe underlying system parameters and generalizes well with respect to these\ndiscovered parameters. Our model also extrapolates the dynamics of the system\nbeyond the training interval and outperforms a non-physically constrained\nbaseline model. Our source code and datasets can be found at this URL:\nhttps://github.com/gbarber94/ConSciNet.\n"}
{"abstract": "  Traffic event cognition and reasoning in videos is an important task that has\na wide range of applications in intelligent transportation, assisted driving,\nand autonomous vehicles. In this paper, we create a novel dataset,\nSUTD-TrafficQA (Traffic Question Answering), which takes the form of video QA\nbased on the collected 10,080 in-the-wild videos and annotated 62,535 QA pairs,\nfor benchmarking the cognitive capability of causal inference and event\nunderstanding models in complex traffic scenarios. Specifically, we propose 6\nchallenging reasoning tasks corresponding to various traffic scenarios, so as\nto evaluate the reasoning capability over different kinds of complex yet\npractical traffic events. Moreover, we propose Eclipse, a novel Efficient\nglimpse network via dynamic inference, in order to achieve\ncomputation-efficient and reliable video reasoning. The experiments show that\nour method achieves superior performance while reducing the computation cost\nsignificantly. The project page: https://github.com/SUTDCV/SUTD-TrafficQA.\n"}
{"abstract": "  The one-shot Person Re-ID scenario faces two kinds of uncertainties when\nconstructing the prediction model from $X$ to $Y$. The first is model\nuncertainty, which captures the noise of the parameters in DNNs due to a lack\nof training data. The second is data uncertainty, which can be divided into two\nsub-types: one is image noise, where severe occlusion and the complex\nbackground contain irrelevant information about the identity; the other is\nlabel noise, where mislabeled affects visual appearance learning. In this\npaper, to tackle these issues, we propose a novel Self-Paced Uncertainty\nEstimation Network (SPUE-Net) for one-shot Person Re-ID. By introducing a\nself-paced sampling strategy, our method can estimate the pseudo-labels of\nunlabeled samples iteratively to expand the labeled samples gradually and\nremove model uncertainty without extra supervision. We divide the pseudo-label\nsamples into two subsets to make the use of training samples more reasonable\nand effective. In addition, we apply a Co-operative learning method of local\nuncertainty estimation combined with determinacy estimation to achieve better\nhidden space feature mining and to improve the precision of selected\npseudo-labeled samples, which reduces data uncertainty. Extensive comparative\nevaluation experiments on video-based and image-based datasets show that\nSPUE-Net has significant advantages over the state-of-the-art methods.\n"}
{"abstract": "  High spin (S=3/2) organic triradicals may offer enhanced properties with\nrespect to several emerging technologies, but those synthesized to date\ntypically exhibit small doublet quartet energy gaps and/or possess limited\nthermal stability and processability. We report a quartet ground state\ntriradical 3, synthesized by a Pd(0)-catalyzed radical-radical cross-coupling\nreaction, which possesses two doublet-quartet energy gaps, DeltaEDQ ~ 0.2-0.3\nkcal mol-1 and DealtaEDQ2 ~ 1.2-1.8 kcal mol-1. The triradical has a 70+%\npopulation of the quartet ground state at room temperature, and good thermal\nstability with onset of decomposition at > 160 {\\deg}C under inert atmosphere.\nMagnetic properties of 3 are characterized by SQUID magnetometry in polystyrene\nglass and by quantitative EPR spectroscopy. Triradical 3 is evaporated under\nultra-high vacuum to form thin films of intact triradicals on silicon\nsubstrate, as confirmed by high resolution X-ray photoelectron spectroscopy.\nAFM and SEM images of the around 1 nm thick films indicate that the triradical\nmolecules form islands on the substrate. The films are stable under ultra-high\nvacuum for at least 17 h but show onset of decomposition after 4 h at ambient\nconditions. The drop-cast films are less prone to degradation in air and have\nlonger lifetime.\n"}
{"abstract": "  Exemplar-based colourisation aims to add plausible colours to a grayscale\nimage using the guidance of a colour reference image. Most of the existing\nmethods tackle the task as a style transfer problem, using a convolutional\nneural network (CNN) to obtain deep representations of the content of both\ninputs. Stylised outputs are then obtained by computing similarities between\nboth feature representations in order to transfer the style of the reference to\nthe content of the target input. However, in order to gain robustness towards\ndissimilar references, the stylised outputs need to be refined with a second\ncolourisation network, which significantly increases the overall system\ncomplexity. This work reformulates the existing methodology introducing a novel\nend-to-end colourisation network that unifies the feature matching with the\ncolourisation process. The proposed architecture integrates attention modules\nat different resolutions that learn how to perform the style transfer task in\nan unsupervised way towards decoding realistic colour predictions. Moreover,\naxial attention is proposed to simplify the attention operations and to obtain\na fast but robust cost-effective architecture. Experimental validations\ndemonstrate efficiency of the proposed methodology which generates high quality\nand visual appealing colourisation. Furthermore, the complexity of the proposed\nmethodology is reduced compared to the state-of-the-art methods.\n"}
{"abstract": "  Total entropy generated by the Unruh effect is calculated within the\nframework of information theory. In contrast to previous studies, here the\ncalculations are done for the finite time of existence of the non-inertial\nreference frame. In this case only the finite number of particles is produced.\nDependence on mass of the emitted particles is taken into account. Analytic\nexpression for the entropy of radiated boson and fermion spectra is derived. We\nstudy also its asymptotics corresponding to limiting cases of low and high\nacceleration. The obtained results can be further generalized to other\nintrinsic degrees of freedom of the emitted particles, such as spin and\nelectric charge.\n"}
{"abstract": "  Channel estimation has long been deemed as one of the most critical problems\nin three-dimensional (3D) massive multiple-input multiple-output (MIMO), which\nis recognized as the leading technology that enables 3D spatial signal\nprocessing in the fifth-generation (5G) wireless communications and beyond.\nRecently, by exploring the angular channel model and tensor decompositions, the\naccuracy of single-user channel estimation for 3D massive MIMO communications\nhas been significantly improved given a limited number of pilot signals.\nHowever, these existing approaches cannot be straightforwardly extended to the\nmulti-user channel estimation task, where the base station (BS) aims at\nacquiring the channels of multiple users at the same time. The difficulty is\nthat the coupling among multiple users' channels makes the channel estimation\ndeviate from widely-used tensor decompositions. It gives a non-standard tensor\ndecomposition format that has not been well tackled. To overcome this\nchallenge, besides directly fitting the new tensor model for channel estimation\nto the wireless data via block coordinate descent (BCD) method, which is prone\nto the overfitting of noises or requires regularization parameter tuning, we\nfurther propose a novel tuning-free channel estimation algorithm that can\nautomatically control the channel model complexity and thus effectively avoid\nthe overfitting. Numerical results are presented to demonstrate the excellent\nperformance of the proposed algorithm in terms of both estimation accuracy and\noverfitting avoidance.\n"}
{"abstract": "  Time plays a fundamental role in our ability to make sense of the physical\nlaws in the world around us. The nature of time has puzzled people -- from the\nancient Greeks to the present day -- resulting in a long running debate between\nphilosophers and physicists alike to whether time needs change to exist (the\nso-called relatival theory), or whether time flows regardless of change (the\nso-called substantival theory). One way to decide between the two is to attempt\nto measure the flow of time with a stationary clock, since if time were\nsubstantival, the flow of time would manifest itself in the experiment. Alas,\nconventional wisdom suggests that in order for a clock to function, it cannot\nbe a static object, thus rendering this experiment seemingly impossible. Here\nwe show, with the aid of counterfactual measurements, the surprising result\nthat a quantum clock can measure the passage of time even while being switched\noff, thus lending constructive support for the substantival theory of time.\n"}
{"abstract": "  Let $G$ be a group. A function $l:G\\rightarrow \\lbrack 0,\\infty )$ is called\na length function if\n  (1) $l(g^{n})=|n|l(g)$ for any $g\\in G$ and $n\\in \\mathbb{Z};$\n  (2) $l(hgh^{-1})=l(g)$ for any $h,g\\in G;$ and\n  (3) $l(ab)\\leq l(a)+l(b)$ for commuting elements $a,b.$\n  Such length functions exist in many branches of mathematics, mainly as stable\nword lengths, stable norms, smooth measure-theoretic entropy, translation\nlengths on $\\mathrm{CAT}(0)$ spaces and Gromov $\\delta $% -hyperbolic spaces,\nstable norms of quasi-cocycles, rotation numbers of circle homeomorphisms,\ndynamical degrees of birational maps and so on. We study length functions on\nLie groups, Gromov hyperbolic groups, arithmetic subgroups, matrix groups over\nrings and Cremona groups. As applications, we prove that every group\nhomomorphism from an arithmetic subgroup of a simple algebraic\n$\\mathbb{Q}$-group of $\\mathbb{Q}$-rank at least $2,$ or a finite-index\nsubgroup of the elementary group $E_{n}(R)$ $(n\\geq 3)$ over an associative\nring, or the Cremona group $\\mathrm{Bir}(P_{\\mathbb{C}}^{2})$ to any group $G$\nhaving a purely positive length function must have its image finite. Here $G$\ncan be outer automorphism group $\\mathrm{Out}(F_{n})$ of free groups, mapping\nclasses group $\\mathrm{MCG}(\\Sigma_{g})$, $\\mathrm{CAT}% (0)$ groups or Gromov\nhyperbolic groups, or the group $\\mathrm{Diff}(\\Sigma ,\\omega )$ of\ndiffeomorphisms of a hyperbolic closed surface preserving an area form $\\omega\n.$\n"}
{"abstract": "  We propose the novel numerical scheme for solution of the multidimensional\nFokker-Planck equation, which is based on the Chebyshev interpolation and the\nspectral differentiation techniques as well as low rank tensor approximations,\nnamely, the tensor train decomposition and the multidimensional cross\napproximation method, which in combination makes it possible to drastically\nreduce the number of degrees of freedom required to maintain accuracy as\ndimensionality increases. We demonstrate the effectiveness of the proposed\napproach on a number of multidimensional problems, including Ornstein-Uhlenbeck\nprocess and the dumbbell model. The developed computationally efficient solver\ncan be used in a wide range of practically significant problems, including\ndensity estimation in machine learning applications.\n"}
{"abstract": "  We present a set of idealised numerical experiments of a solstitial\naquaplanet ocean and examine the thermodynamic and dynamic implications of\nsurface gravity waves (SGWs) upon its mean state. The aquaplanet's oceanic\ncirculation is dominated by an equatorial zonal jet and four Ekman driven\nmeridional overturning circulation (MOC) cells aligned with the westerly\natmospheric jet streams and easterly trade winds in both hemispheres. Including\nSGW parameterization (representing modulations of air-sea momentum fluxes,\nLangmuir circulation and Stokes-Coriolis force) increases mixed layer vertical\nmomentum diffusivity by approx. 40% and dampens surface momentum fluxes by\napprox. 4%. The correspondingly dampened MOC impacts the oceanic density\nstructure to 1 km depth by lessening the large-scale advective transports of\nheat and salt, freshening the equatorial latitudes (where evaporation minus\nprecipitation [E-P] is negative) and increasing salinity in the subtropics\n(where E-P is positive) by approx. 1%. The midlatitude pycnocline in both\nhemispheres is deepened by the inclusion of SGWs. Including SGWs into the\naquaplanet ocean model acts to increase mixed layer depth by approx. 10% (up to\n20% in the wintertime in midlatitudes), decrease vertical shear in the upper\n200 m and alter local midlatitude buoyancy frequency. Generally, the impacts of\nSGWs upon the aquaplanet ocean are found to be consistent across cooler and\nwarmer climates. We suggest that the implications of these simulations could be\nrelevant to understanding future projections of SGW climate, exoplanetary\noceans, and the dynamics of the Southern Ocean mixed layer.\n"}
{"abstract": "  Network programmability is an area of research both defined by its potential\nand its current limitations. While programmable hardware enables customization\nof device operation, tailoring processing to finely tuned objectives, limited\nresources stifle much of the capability and scalability desired for future\ntechnologies. Current solutions to overcome these limitations simply shift the\nproblem, temporarily offloading memory needs or processing to other systems\nwhile incurring both round-trip time and complexity costs. To overcome these\nunnecessary costs, we introduce Infinity, a resource disaggregation method to\nmove processing to capable devices while continuing to forward as the original\nowner, limiting unnecessary buffering and round-trip processing. By forwarding\nboth the processing need and associated data simultaneously we are able to\nscale operation with minimal overhead and delay, improving both capability and\nperformance objectives for in-network processing.\n"}
{"abstract": "  The statistical significance that characterizes a discrepancy between a\nmeasurement and theoretical prediction is usually calculated assuming that the\nstatistical and systematic uncertainties are known. Many types of systematic\nuncertainties are, however, estimated on the basis of approximate procedures\nand thus the values of the assigned errors are themselves uncertain. Here the\nimpact of the uncertainty {\\it on the assigned uncertainty} is investigated in\nthe context of the muon $g-2$ anomaly. The significance of the observed\ndiscrepancy between the Standard Model prediction of the muon's anomalous\nmagnetic moment and measured values are shown to decrease substantially if the\nrelative uncertainty in the uncertainty assigned to the Standard Model\nprediction exceeds around 30\\%. The reduction in sensitivity increases for\nhigher significance, so that establishing a $5\\sigma$ effect will require not\nonly small uncertainties but the uncertainties themselves must be estimated\naccurately to correspond to one standard deviation.\n"}
{"abstract": "  First principles simulations are conducted to shed light on the question of\nwhether a two-dimensional topological insulator (2DTI) phase may be obtained at\nthe interface between InAs and GaSb. To this end, the InAs/GaSb interface is\ncompared and contrasted with the HgTe/CdTe interface. Density functional theory\n(DFT) simulations of these interfaces are performed using a machine-learned\nHubbard U correction [npj Comput. Mater. 6, 180 (2020)]. For the HgTe/CdTe\ninterface our simulations show that band crossing is achieved and an inverted\ngap is obtained at a critical thickness of 5.1 nm of HgTe, in agreement with\nexperiment and previous DFT calculations. In contrast, for InAs/GaSb the gap\nnarrows with increasing thickness of InAs; however the gap does not close for\ninterfaces with up to 50 layers (about 15 nm) of each material. When an\nexternal electric field is applied across the InAs/GaSb interface, the\nGaSb-derived valence band maximum is shifted up in energy with respect to the\nInAs-derived conduction band minimum until eventually the bands cross and an\ninverted gap opens. Our results show that it may be possible to reach the\ntopological regime at the InAs/GaSb interface under the right conditions.\nHowever, it may be challenging to realize these conditions experimentally,\nwhich explains the difficulty of experimentally demonstrating an inverted gap\nin InAs/GaSb.\n"}
{"abstract": "  Graphical models are a key class of probabilistic models for studying the\nconditional independence structure of a set of random variables. Circular\nvariables are special variables, characterized by periodicity, arising in\nseveral contexts and fields. However, models for studying the\ndependence/independence structure of circular variables are under-explored.\nThis paper analyses three multivariate circular distributions, the von Mises,\nthe Wrapped Normal and the Inverse Stereographic distributions, focusing on\ntheir properties concerning conditional independence. For each one of these\ndistributions, we discuss the main properties related to conditional\nindependence and introduce suitable classes of graphical models. The usefulness\nof the proposed models is shown by modelling the conditional independence among\ndihedral angles characterizing the three-dimensional structure of some\nproteins.\n"}
{"abstract": "  This paper develops an integer programming approach on two-sided many-to-one\nmatching by investigating stable integral matchings of a fictitious continuum\nmarket induced from the original matching market. Each stable integral matching\nof the continuum market corresponds to a stable matching of the original\nmatching market. We show that a stable matching exists in the original matching\nmarket when firms' preference profile satisfies a unimodularity condition. Our\nresult indicates that a stable matching is guaranteed to exist with various\nforms of complementary preferences.\n"}
{"abstract": "  We examine the effect of boson-fermion interaction in a one-dimensional\nBose-Fermi mixture by using the density matrix renormalization group method. We\nshow that the boson superfluidity is enhanced by fermions for a weak\nboson-fermion coupling at an approximate integer boson filling factor (e.g.,\n$0.935\\le \\rho_b \\le 1.0$), and this enhancement is produced both in a fermion\nmetallic state and in a fermion insulating state. A metal-insulator phase\ntransition of fermions induced by boson-fermion interaction is observed even\nthough there is no fermion-fermion interaction in the parent Hamiltonian.\nFurthermore, we find that the boson superfluid order and density wave order can\ncoexist in a deep fermion Mott region. All these features could be measured in\nfuture experiments and open up the possibility of detecting the new physical\neffect in the Bose-Fermi mixture.\n"}
{"abstract": "  We build on recent work of Yeats, Courtiel, and others involving connected\nchord diagrams. We first derive from a Hopf-algebraic foundation a class of\ntree-like functional equations and prove that they are solved by weighted\ngenerating functions of two different subsets of weighted connected chord\ndiagrams: arbitrary diagrams and diagrams forbidding so-called top cycle\nsubdiagrams. These equations generalize the classic specification for\nincreasing ordered trees and their solution uses a novel decomposition,\nsimplifying and generalizing previous results. The resulting tree perspective\non chord diagrams leads to new enumerative insights through the study of novel\ndiagram classes. We present a recursive bijection between connected\ntop-cycle-free diagrams with $n$ chords and triangulations of a disk with $n+1$\nvertices, thereby counting the former. This connects to combinatorial maps,\nCatalan intervals, and uniquely sorted permutations, leading to new conjectured\nbijective relationships between diagram classes defined by forbidding graphical\nsubdiagrams and imposing connectedness properties and a variety of other rich\ncombinatorial objects. We conclude by exhibiting and studying a direct\nbijection between diagrams of size $n$ with a single terminal chord and\ndiagrams of size $n-1$.\n"}
{"abstract": "  We formulate a comprehensive hydrodynamic theory of two-dimensional liquid\ncrystals with generic $p-$fold rotational symmetry, also known as $p-$atics, of\nwhich mematics $(p=2)$ and hexatics $(p=6)$ are the two best known examples.\nPrevious hydrodynamic theories of $p-$atics are characrerized by continuous\n${\\rm O}(2)$ rotational symmetry, which is higher than the discrete rotational\nsymmetry of $p-$atic phases. By contrast, here we demonstrate that the discrete\nrotational symmetry allows the inclusion of additional terms in the\nhydrodynamic equations, which, in turn, lead to novel phenomena, such as the\npossibility of flow alignment at high shear rates, even for $p>2$. Furthermore,\nwe show that any finite imposed shear will induce long-ranged orientational\norder in any $p-$atic liquid crystal, in contrast to the quasi-long-ranged\norder that occurs in the absence of shear. The induced order parameter scales\nlike a non-universal power of the applied shear rate at small shear rates.\n"}
{"abstract": "  This paper focuses on the analysis of a free energy functional, that models a\ndilute suspension of magnetic nanoparticles in a two-dimensional nematic well.\nThe {\\it first part} of the article is devoted to the asymptotic analysis of\nglobal energy minimizers in the limit of vanishing elastic constant, $\\ell\n\\rightarrow 0$ where the re-scaled elastic constant $\\ell$ is inversely\nproportional to the domain area. The first results concern the strong\n$H^1$-convergence and a $\\ell$-independent $H^2$-bound for the global\nminimizers on smooth bounded 2D domains, with smooth boundary and topologically\ntrivial Dirichlet conditions. The {\\it second part} focuses on the discrete\napproximation of regular solutions of the corresponding non-linear system of\npartial differential equations with cubic non-linearity and non-homogeneous\nDirichlet boundary conditions. We establish (i) the existence and local\nuniqueness of the discrete solutions using fixed point argument, (ii) a best\napproximation result in energy norm, (iii) error estimates in the energy and\n$L^2$ norms with $\\ell $- discretization parameter dependency for the\nconforming finite element method. Finally, the theoretical results are\ncomplemented by numerical experiments on the discrete solution profiles, the\nnumerical convergence rates that corroborates the theoretical estimates,\nfollowed by plots that illustrate the dependence of the discretization\nparameter on $\\ell$.\n"}
{"abstract": "  We consider the coded caching problem with shared caches where several users\nshare a cache, but each user has access to only a single cache. For this\nnetwork, the fundamental limits of coded caching are known for centralized and\ndecentralized settings under uncoded placement. In the centralized case, to\nachieve the gains offered by coded caching, one requires a sub-packetization\nwhich increases exponentially with the number of caches. The dedicated cache\nnetworks had a similar issue, and placement delivery arrays (PDAs) were\nintroduced as a solution to it. Using the PDA framework, we propose a procedure\nto obtain new coded caching schemes for shared caches with lower\nsub-packetization requirements. The advantage of this procedure is that we can\ntransform all the existing PDA structures into coded caching schemes for shared\ncaches, thus resulting in low sub-packetization schemes. We also show that the\noptimal scheme given by Parrinello, Unsal and Elia (Fundamental Limits of Coded\nCaching with Multiple Antennas, Shared Caches and Uncoded Prefetching) can be\nrecovered using a Maddah-Ali Niesen PDA.\n"}
{"abstract": "  We introduce a machine-learning density-functional-theory formalism for the\nspinless Hubbard model in one dimension at both zero and finite temperature. In\nthe zero-temperature case this establishes a one-to-one relation between the\nsite occupation and the total energy, which is then minimised at the\nground-state occupation. In contrast, at finite temperature the same relation\nis defined between the Helmholtz free energy and the equilibrium site\noccupation. Most importantly, both functionals are semi-local, so that they are\nindependent from the size of the system under investigation and can be\nconstructed over exact data for small systems. These 'exact' functionals are\nnumerically defined by neural networks. We also define additional neural\nnetworks for finite-temperature thermodynamical quantities, such as the entropy\nand heat capacity. These can be either a functional of the ground-state site\noccupation or of the finite-temperature equilibrium site occupation. In the\nfirst case their equilibrium value does not correspond to an extremal point of\nthe functional, while it does in the second case. Our work gives us access to\nfinite-temperature properties of many-body systems in the thermodynamic limit.\n"}
{"abstract": "  Let $\\phi(x, y)\\colon \\mathbb{R}^d\\times \\mathbb{R}^d\\to \\mathbb{R}$ be a\nfunction. We say $\\phi$ is a Mattila--Sj\\\"{o}lin type function of index\n$\\gamma$ if $\\gamma$ is the smallest number satisfying the property that for\nany compact set $E\\subset \\mathbb{R}^d$, $\\phi(E, E)$ has a non-empty interior\nwhenever $\\dim_H(E)>\\gamma$. The usual distance function, $\\phi(x, y)=|x-y|$,\nis conjectured to be a Mattila--Sj\\\"{o}lin type function of index\n$\\frac{d}{2}$. In the setting of finite fields $\\mathbb{F}_q$, this definition\nis equivalent to the statement that $\\phi(E, E)=\\mathbb{F}_q$ whenever $|E|\\gg\nq^{\\gamma}$. The main purpose of this paper is to prove the existence of such\nfunctions with index $\\frac{d}{2}$ in the vector space $\\mathbb{F}_q^d$.\n"}
{"abstract": "  We perform a detailed study of the phase transitions and mechanisms of\nelectron localization in the extended Hubbard model using the dynamical cluster\napproximation on a $2\\times 2$ cluster. We explore the interplay of charge\norder and Mott physics. We find that a nearest-neighbor Coulomb interaction $V$\ncauses \"screening\" effects close to the Mott phase transition, pushing the\nphase boundary to larger values of $U$. We also demonstrate the different\neffects of $V$ on correlations in metallic and insulating regimes and document\nthe different correlation aspects of charge order and Mott states.\n"}
{"abstract": "  The malfunction or interruption of wireless coverage services has been shown\nto increase the mortality rate during natural disasters. Wireless coverage by\nan unmanned aerial vehicle (UAV) provides network coverage to ground user\ndevices during and post-disaster events. The relay hops receive wireless\ncoverage and can be forwarded to user devices that are out of coverage allowing\nreliable connectivity for large-scale user devices. This work evaluates the\noptimal relay hops performance to improve wireless coverage services and\nestablish connectivity in post-disaster scenarios. The results demonstrate the\nUAV line of sights understanding to select an optimal relay for improving\nwireless coverage services. The path loss probability and system capacity were\nall affected by the user device distance and relay densities. The optimal relay\nhop distance and the UAV positions static are also investigated to improve\ncoverage likelihood which could be especially useful for UAV deployment design.\nIt is found that the dense relays node in UAV systems enhances the capacity\ncoverage area and energy efficiency by decentralized connectivity through a\nmultihop device to device wireless network.\n"}
{"abstract": "  Using the Carlson-Simpson theorem, we give a new general condition for a\nstructure in a finite binary relational language to have finite big Ramsey\ndegrees\n"}
{"abstract": "  A promising and cost-effective method for numerical simulation of high Re\nwall-bounded flows is wall-modeled large-eddy simulation. Most wall models are\nformulated from the Reynolds-averaged Navier-Stokes equations (RANS). These\nRANS-based wall models are calibrated using mean turbulence data and make no\nuse of the current vast knowledge on turbulent flow structure. Moreover,\nRANS-based models are limited to predicting the mean velocity profile, and the\nmean wall shear stress. Using the knowledge of the wall-normal self-similarity\nof high-$Re_\\tau$ wall-bounded turbulent flows, we present a coupling between a\nnear-wall patch of DNS resolution fixed in inner-units and an outer LES flow\nfield. The near-wall patch captures the near-wall self-sustaining cycle as well\nas the lower portion of the self-similar hierarchy of eddies. Given that both\nthe near-wall patch and the LES capture separate portions of the self-similar\nhierarchy of eddies an instantaneous top boundary condition for the patch is\nformulated. The near-wall model is capable of predicting subgrid-scale\nquantities such as the wall stress, velocity fluctuations, kinetic energy\nspectra, and flow structure across the entire near-wall layer.\n"}
{"abstract": "  Asset health monitoring continues to be of increasing importance on\nproductivity, reliability, and cost reduction. Early Fault detection is a\nkeystone of health management as part of the emerging Prognostics and Health\nManagement (PHM) philosophy. This paper proposes a Hidden Markov Model (HMM) to\nassess the machine health degradation. using Principal Component Analysis (PCA)\nto enhance features extracted from vibration signals is considered. The\nenhanced features capture the second order structure of the data. The\nexperimental results based on a bearing test bed show the plausibility of the\nproposed method.\n"}
{"abstract": "  Quasi one-dimensional (1D) vanadium tetrasulfide ($VS_4$) nanowires (NWs) are\nsynthetic semiconductors which combine with each other through Van der Waals\ninteractions to form bulk phases. However, the properties of these individual\nnanowires remain unknown. Nevertheless, our calculations of their stability\nindicate that $VS_4$) NWs can be separated from their bulk structures.\nAccordingly, we theoretically investigated the geometrical, electronic, and\nmagnetic properties of bulk phase and isolated $VS_4$ NWs. Our results indicate\nthat both bulk phase and isolated $VS_4$ NWs are semiconductors with band gaps\nof 2.24 and 2.64 eV, respectively, and that they prefer the antiferromagnetic\n(AFM) ground state based on DFT calculations. These calculations also suggested\nthat isolated $VS_4$ NWs show half-metallic antiferromagnetism upon electron\nand hole doping because carrier doping splits the spin degeneracy to induce\nlocal spin polarisation. As a result, spin polarisation currents in isolated\n$VS_4$ NWs can be manipulated with locally applied gate voltage. Therefore,\nthese 1D AFM materials have a high potential for advancing both fundamental\nresearch and spintronic applications because they are more resistant to\nmagnetic perturbation than their 1D ferromagnetic counterparts.\n"}
{"abstract": "  Two-view structure-from-motion (SfM) is the cornerstone of 3D reconstruction\nand visual SLAM. Existing deep learning-based approaches formulate the problem\nby either recovering absolute pose scales from two consecutive frames or\npredicting a depth map from a single image, both of which are ill-posed\nproblems. In contrast, we propose to revisit the problem of deep two-view SfM\nby leveraging the well-posedness of the classic pipeline. Our method consists\nof 1) an optical flow estimation network that predicts dense correspondences\nbetween two frames; 2) a normalized pose estimation module that computes\nrelative camera poses from the 2D optical flow correspondences, and 3) a\nscale-invariant depth estimation network that leverages epipolar geometry to\nreduce the search space, refine the dense correspondences, and estimate\nrelative depth maps. Extensive experiments show that our method outperforms all\nstate-of-the-art two-view SfM methods by a clear margin on KITTI depth, KITTI\nVO, MVS, Scenes11, and SUN3D datasets in both relative pose and depth\nestimation.\n"}
{"abstract": "  In this paper, we define both the upper and lower order of a sense-preserving\nharmonic mapping in $\\mathbb{D}$. We generalize to the harmonic case some known\nresults about holomorphic functions with positive lower order and we show some\nconsequences of a function having finite upper order. In addition, we improve a\nrelated result in the case when there is equality in a known distortion theorem\nfor harmonic mappings with finite upper order. Some examples are provided to\nillustrate the developed theory.\n"}
{"abstract": "  Lithium is an important element for the understanding of ultracool dwarfs\nbecause it is lost to fusion at masses above $\\sim 68\\, M_{\\rm J}$. Hence, the\npresence or absence of atomic Li has served as an indicator of the nearby\nH-burning boundary at about $75\\,M_{\\rm J}$ between brown-dwarfs and very\nlow-mass stars. Historically the \"Lithium test\", a search for the presence and\nstrength of the Li line at 670.8 nm, has been a marker if an object has a\nsubstellar mass with stellar-like spectral energy distribution (e.g., a\nlate-type M dwarf). While the Li test could in principle also be used to\ndistinguish masses of later-type L-T dwarfs, Li is predominantly no longer\nfound as an atomic gas, but rather a molecular species such as LiH, LiF, LiOH,\nand LiCl in their cooler atmospheres. L- and T-type brown dwarfs are also quite\nfaint at 670 nm and thus challenging targets for high resolution spectroscopy.\nBut only recently have experimental molecular line lists become available for\nthe molecular Li species, allowing molecular Li mass discrimination. In this\nstudy, we generated the latest opacity of each of these Li-bearing molecules\nand performed thermochemical equilibrium atmospheric composition calculation of\nthe abundance of these molecules. Finally, we computed thermal emission spectra\nfor a series of radiative-convective equilibrium models of cloudy and cloudless\nbrown dwarf atmospheres (with $T_{\\rm eff}=$ 500--2400~K, and $\\log g$=4.0,\n4.5, 5.0) to understand where the presence or absence of atmospheric\nlithium-bearing species is most easily detected as a function of brown dwarf\nmass and age. After atomic Li, the best spectral signatures were found to be\nLiF at $10.5-12.5$~\\micron and LiCl at $14.5-18.5$ $\\micron$. LiH also shows a\nnarrow feature at $\\sim 9.38$ $\\micron$.\n"}
{"abstract": "  In complex industrial settings, it is common practice to monitor the\noperation of machines in order to detect undesired states, adjust maintenance\nschedules, optimize system performance or collect usage statistics of\nindividual machines. In this work, we focus on estimating the power output of a\nCombined Heat and Power (CHP) machine of a medium-sized company facility by\nanalyzing the total facility power consumption. We formulate the problem as a\ntime-series classification problem where the class label represents the CHP\npower output. As the facility is fully instrumented and sensor measurements\nfrom the CHP are available, we generate the training labels in an automated\nfashion from the CHP sensor readings. However, sensor failures result in\nmislabeled training data samples which are hard to detect and remove from the\ndataset. Therefore, we propose a novel multi-task deep learning approach that\njointly trains a classifier and an autoencoder with a shared embedding\nrepresentation. The proposed approach targets to gradually correct the\nmislabelled data samples during training in a self-supervised fashion, without\nany prior assumption on the amount of label noise. We benchmark our approach on\nseveral time-series classification datasets and find it to be comparable and\nsometimes better than state-of-the-art methods. On the real-world use-case of\npredicting the CHP power output, we thoroughly evaluate the architectural\ndesign choices and show that the final architecture considerably increases the\nrobustness of the learning process and consistently beats other recent\nstate-of-the-art algorithms in the presence of unstructured as well as\nstructured label noise.\n"}
{"abstract": "  Controlling false discovery rate (FDR) while leveraging the side information\nof multiple hypothesis testing is an emerging research topic in modern data\nscience. Existing methods rely on the test-level covariates while ignoring\npossible hierarchy among the covariates. This strategy may not be optimal for\ncomplex large-scale problems, where hierarchical information often exists among\nthose test-level covariates. We propose NeurT-FDR which boosts statistical\npower and controls FDR for multiple hypothesis testing while leveraging the\nhierarchy among test-level covariates. Our method parametrizes the test-level\ncovariates as a neural network and adjusts the feature hierarchy through a\nregression framework, which enables flexible handling of high-dimensional\nfeatures as well as efficient end-to-end optimization. We show that NeurT-FDR\nhas strong FDR guarantees and makes substantially more discoveries in synthetic\nand real datasets compared to competitive baselines.\n"}
{"abstract": "  Ensemble and auxiliary tasks are both well known to improve the performance\nof machine learning models when data is limited. However, the interaction\nbetween these two methods is not well studied, particularly in the context of\ndeep reinforcement learning. In this paper, we study the effects of ensemble\nand auxiliary tasks when combined with the deep Q-learning algorithm. We\nperform a case study on ATARI games under limited data constraint. Moreover, we\nderive a refined bias-variance-covariance decomposition to analyze the\ndifferent ways of learning ensembles and using auxiliary tasks, and use the\nanalysis to help provide some understanding of the case study. Our code is open\nsource and available at https://github.com/NUS-LID/RENAULT.\n"}
{"abstract": "  Attention is a key component of the now ubiquitous pre-trained language\nmodels. By learning to focus on relevant pieces of information, these\nTransformer-based architectures have proven capable of tackling several tasks\nat once and sometimes even surpass their single-task counterparts. To better\nunderstand this phenomenon, we conduct a structural analysis of a new\nall-purpose question answering model that we introduce. Surprisingly, this\nmodel retains single-task performance even in the absence of a strong transfer\neffect between tasks. Through attention head importance scoring, we observe\nthat attention heads specialize in a particular task and that some heads are\nmore conducive to learning than others in both the multi-task and single-task\nsettings.\n"}
{"abstract": "  We report the spatial beam self-cleaning in bi-tapered conventional multimode\nfibers (MMFs) with different tapered lengths. Through the introduction of the\nbi-tapered structure in MMFs, the input beam with poor beam quality from a\nhigh-power fiber laser can be converted to a centered, bell-shaped beam in a\nshort length, due to the strengthened nonlinear modes coupling. It is found\nthat the bi-tapered MMF with longer tapered length at the same waist diameter\nshows better beam self-cleaning effect and larger spectral broadening. The\nobtained results offer a new method to improve the beam quality of high-power\nlaser at low cost. Besides, it may be interesting for manufacturing bi-tapered\nMMF-based devices to obtain the quasi-fundamental mode beam in spatiotemporal\nmode-locked fiber lasers.\n"}
{"abstract": "  This paper presents the application of a meta model and single underlying\nmodel on an applied avionics system design use case. System models, safety\nassurance cases and safety requirements are maintained in a central repository.\nThis enables to link these data which are originally developed in unrelated\ntools. By having such a central repository, traceability can be established,\nand consistency can be ensured, which leads to less errors and a shorter\ndevelopment time. A meta model was constructed which matches the central\nrepository to enable bidirectional synchronization with an external authoring\ntool.\n"}
{"abstract": "  The massive screened expansion for pure SU(3) Yang-Mills theory is extended\nto finite temperature in the Landau gauge. All thermal integrals are evaluated\nanalytically up to an external one-dimensional integration, yielding explicit\nintegral representations of analytic functions which can be continued to the\nwhole complex plane. The gluon propagator is first explored in the Euclidean\nspace by making use of parameters obtained from first principles, which were\nalready found to accurately reproduce the lattice data at zero temperature.\nWithin such a scheme, the agreement with the lattice at $T\\neq 0$ turns out to\nbe only qualitative. The description improves provided that the parameters are\ntuned in a temperature-dependent way by a fit to the data, carried out\nseparately for each component of the propagator; in particular, the transverse\ncomponent closely follows the lattice data, while the agreement of the\nlongitudinal component with the data is poor at small momenta and moderately\nhigh temperatures. The dispersion relations of the quasi-gluon are then\nextracted from the pole trajectory in the complex plane using the fitted\nparameters. A crossover is found for the mass, suppressed by temperature like\nan order parameter in the confined phase, while increasing like an ordinary\nthermal mass in the deconfined phase.\n"}
{"abstract": "  As the Covid-19 outbreaks rapidly all over the world day by day and also\naffects the lives of million, a number of countries declared complete lock-down\nto check its intensity. During this lockdown period, social media plat-forms\nhave played an important role to spread information about this pandemic across\nthe world, as people used to express their feelings through the social\nnetworks. Considering this catastrophic situation, we developed an experimental\napproach to analyze the reactions of people on Twitter taking into ac-count the\npopular words either directly or indirectly based on this pandemic. This paper\nrepresents the sentiment analysis on collected large number of tweets on\nCoronavirus or Covid-19. At first, we analyze the trend of public sentiment on\nthe topics related to Covid-19 epidemic using an evolutionary classification\nfollowed by the n-gram analysis. Then we calculated the sentiment ratings on\ncollected tweet based on their class. Finally, we trained the long-short term\nnetwork using two types of rated tweets to predict sentiment on Covid-19 data\nand obtained an overall accuracy of 84.46%.\n"}
{"abstract": "  Sentiment analysis as a sub-field of natural language processing has received\nincreased attention in the past decade enabling organisations to more\neffectively manage their reputation through online media monitoring. Many\ndrivers impact reputation, however, this thesis focuses only the aspect of\nfinancial performance and explores the gap with regards to financial sentiment\nanalysis in a South African context. Results showed that pre-trained sentiment\nanalysers are least effective for this task and that traditional lexicon-based\nand machine learning approaches are best suited to predict financial sentiment\nof news articles. The evaluated methods produced accuracies of 84\\%-94\\%. The\npredicted sentiments correlated quite well with share price and highlighted the\npotential use of sentiment as an indicator of financial performance. A main\ncontribution of the study was updating an existing sentiment dictionary for\nfinancial sentiment analysis. Model generalisation was less acceptable due to\nthe limited amount of training data used. Future work includes expanding the\ndata set to improve general usability and contribute to an open-source\nfinancial sentiment analyser for South African data.\n"}
{"abstract": "  The isospin splitting of the in-medium $NN\\rightarrow N\\Delta$ cross sections\nin asymmetric nuclear medium are investigated in the framework of the one-boson\nexchange model by including $\\delta$ and $\\rho$ mesons. Our results show that\nthe medium correction factors $R=\\sigma_{ NN\\rightarrow\nN\\Delta}^*/\\sigma_{NN\\rightarrow N\\Delta}^{\\text{free}}$ have $R_{pp \\to\nn\\Delta ^{++}} < R_{nn \\to p\\Delta ^{-}}$ and $R_{NN \\to N\\Delta ^{+}} <R_{NN\n\\to N\\Delta ^{0}}$ by using the without-$\\delta$ sets. By including the\n$\\delta$ meson, it appears the totally opposite results in the $R$ for\ndifferent channels, i.e., $R_{pp \\to n\\Delta ^{++}} > R_{nn \\to p\\Delta ^{-}}$\nand $R_{NN \\to N\\Delta ^{+}} >R_{NN \\to N\\Delta ^{0}}$.\n"}
{"abstract": "  Structural flexibility and/or dynamic interactions with other molecules is a\ncritical aspect of protein function. CryoEM provides direct visualization of\nindividual macromolecules sampling different conformational and compositional\nstates. While numerous methods are available for computational classification\nof discrete states, characterization of continuous conformational changes or\nlarge numbers of discrete state without human supervision remains challenging.\nHere we present e2gmm, a machine learning algorithm to determine a\nconformational landscape for proteins or complexes using a 3-D Gaussian mixture\nmodel mapped onto 2-D particle images in known orientations. Using a deep\nneural network architecture, e2gmm can automatically resolve the structural\nheterogeneity within the protein complex and map particles onto a small latent\nspace describing conformational and compositional changes. This system presents\na more intuitive and flexible representation than other manifold methods\ncurrently in use. We demonstrate this method on both simulated data as well as\nthree biological systems, to explore compositional and conformational changes\nat a range of scales. The software is distributed as part of EMAN2.\n"}
{"abstract": "  Many modern time-series datasets contain large numbers of output response\nvariables sampled for prolonged periods of time. For example, in neuroscience,\nthe activities of 100s-1000's of neurons are recorded during behaviors and in\nresponse to sensory stimuli. Multi-output Gaussian process models leverage the\nnonparametric nature of Gaussian processes to capture structure across multiple\noutputs. However, this class of models typically assumes that the correlations\nbetween the output response variables are invariant in the input space.\nStochastic linear mixing models (SLMM) assume the mixture coefficients depend\non input, making them more flexible and effective to capture complex output\ndependence. However, currently, the inference for SLMMs is intractable for\nlarge datasets, making them inapplicable to several modern time-series\nproblems. In this paper, we propose a new regression framework, the orthogonal\nstochastic linear mixing model (OSLMM) that introduces an orthogonal constraint\namongst the mixing coefficients. This constraint reduces the computational\nburden of inference while retaining the capability to handle complex output\ndependence. We provide Markov chain Monte Carlo inference procedures for both\nSLMM and OSLMM and demonstrate superior model scalability and reduced\nprediction error of OSLMM compared with state-of-the-art methods on several\nreal-world applications. In neurophysiology recordings, we use the inferred\nlatent functions for compact visualization of population responses to auditory\nstimuli, and demonstrate superior results compared to a competing method\n(GPFA). Together, these results demonstrate that OSLMM will be useful for the\nanalysis of diverse, large-scale time-series datasets.\n"}
{"abstract": "  Compared to abundant classical statistics-based literature, to date, very\nlittle Bayesian literature exists on Procrustes shape analysis in Geometric\nMorphometrics, probably because of being a relatively new branch of statistical\nresearch and because of inherent computational difficulty associated with\nBayesian analysis. Moreover, we may obtain a plethora of novel inferences from\nBayesian Procrustes analysis of shape parameter distributions. In this paper,\nwe propose to regard the posterior of Procrustes shape variance as\nmorphological variability indicators. Here we propose novel Bayesian\nmethodologies for Procrustes shape analysis based on landmark data's isotropic\nvariance assumption and propose a Bayesian statistical test for model\nvalidation of new species discovery using morphological variation reflected in\nthe posterior distribution of landmark-variance of objects studied under\nGeometric Morphometrics. We will consider Gaussian distribution-based and\nheavy-tailed t distribution-based models for Procrustes analysis.\n  To date, we are not aware of any direct R package for Bayesian Procrustes\nanalysis for landmark-based Geometric Morphometrics. Hence, we introduce a\nnovel, simple R package \\textbf{BPviGM1} (\"Bayesian Procrustes Variance-based\ninferences in Geometric Morphometrics 1\"), which essentially contains the R\ncode implementations of the computations for proposed models and methodologies,\nsuch as R function for Markov Chain Monte Carlo (MCMC) run for drawing samples\nfrom posterior of parameters of concern and R function for the proposed\nBayesian test of model validation based on significance morphological\nvariation. As an application, we can quantitatively show that primate male-face\nmay be genetically viable to more shape-variation than the same for females.\n"}
{"abstract": "  Since 2012, in a case-study in Bucaramanga-Colombia, 179 pedestrians died in\ncar accidents, and another 2873 pedestrians were injured. Each day, at least\none passerby is involved in a tragedy. Knowing the causes to decrease accidents\nis crucial, and using system-dynamics to reproduce the collisions' events is\ncritical to prevent further accidents. This work implements simulations to save\nlives by reducing the city's accidental rate and suggesting new safety policies\nto implement. Simulation's inputs are video recordings in some areas of the\ncity. Deep Learning analysis of the images results in the segmentation of the\ndifferent objects in the scene, and an interaction model identifies the primary\nreasons which prevail in the pedestrians or vehicles' behaviours. The first and\nmost efficient safety policy to implement-validated by our simulations-would be\nto build speed bumps in specific places before the crossings reducing the\naccident rate by 80%.\n"}
{"abstract": "  We explore the decays of $B\\to V_1V_2$ ($V_{1,2}= (\\rho, \\omega,K^*, \\phi)$\nand $B= (B^0, B^+,B_s)$) with transverse polarizations. We explicitly evaluate\nthe eigenstates of T-odd scalar operators involving spins for the first time,\nwhich offer physical insight among the T violating observables. Based on the\nhelicity suppression of tree operators for transverse polarizations in the\nstandard model (SM), we deduce that $\\Delta \\phi _p= \\Delta \\phi_\\parallel -\n\\Delta \\phi_\\perp=0$ with $\\Delta\\phi_{\\perp,\\parallel}$ the weak phases of the\ntransverse amplitudes. In contrast, the experiments show that $\\Delta \\phi _p\n(B^0 \\to K^{*0} \\omega)= -0.84\\pm 0.54$, which could be a signal of new\nphysics. There is also a discrepancy between our result in the SM and the\nexperimental data for the transverse polarized branching ratio in $B^0 \\to\nK^{*0} \\omega$. In addition, by counting the helicity flips, we argue that\n$\\sin(\\phi_\\parallel - \\phi_\\perp ) \\approx 0$, which meets well with most of\nthe experiments except for the ones with $B^0\\to K^{*0}\\omega$, where\n$\\phi_{\\parallel,\\perp}$ are the strong phases in the transverse amplitudes.\n"}
{"abstract": "  In view of the great contribution of neutrino-electron scattering to the deep\nunderstanding of electroweak interactions, we focus in this paper on the study\nof elastic scattering of a muon neutrino by an electron\n$(e^{-}\\nu_{\\mu}\\rightarrow e^{-}\\nu_{\\mu})$ in the presence of a circularly\npolarized electromagnetic field. We perform our theoretical calculation within\nthe framework of Fermi theory using the exact wave functions of charged\nparticles in an electromagnetic field. The expression of the differential cross\nsection (DCS) for this process is obtained analytically in the absence and\npresence of the laser field. The effect of the field strength and frequency on\nthe exchange of photons as well as on the DCS is presented and analyzed.\nMassive neutrino effects are also included and discussed. This study, added to\nthe previous ones, will significantly enrich our knowledge in fundamental\nphysics.\n"}
{"abstract": "  Image-to-image (I2I) translation has matured in recent years and is able to\ngenerate high-quality realistic images. However, despite current success, it\nstill faces important challenges when applied to small domains. Existing\nmethods use transfer learning for I2I translation, but they still require the\nlearning of millions of parameters from scratch. This drawback severely limits\nits application on small domains. In this paper, we propose a new transfer\nlearning for I2I translation (TransferI2I). We decouple our learning process\ninto the image generation step and the I2I translation step. In the first step\nwe propose two novel techniques: source-target initialization and\nself-initialization of the adaptor layer. The former finetunes the pretrained\ngenerative model (e.g., StyleGAN) on source and target data. The latter allows\nto initialize all non-pretrained network parameters without the need of any\ndata. These techniques provide a better initialization for the I2I translation\nstep. In addition, we introduce an auxiliary GAN that further facilitates the\ntraining of deep I2I systems even from small datasets. In extensive experiments\non three datasets, (Animal faces, Birds, and Foods), we show that we outperform\nexisting methods and that mFID improves on several datasets with over 25\npoints.\n"}
{"abstract": "  We consider various integrable two-parameter deformations of the $AdS_3\n\\times S^3 \\times T^4$ superstring with quantum group symmetry. Working on the\nstring worldsheet in light-cone gauge and to quadratic order in fermions, we\nobtain their common massive tree-level two-body S matrix, which matches the\nexpansion of the conjectured exact $q$-deformed S matrix. We then analyze the\nbehavior of the exact S matrix under mirror transformation -- a double Wick\nrotation on the worldsheet -- and find that it satisfies a mirror duality\nrelation analogous to the distinguished $q$-deformed $AdS_5 \\times S^5$ S\nmatrix in the one parameter deformation limit. Finally, we show that the\nfermionic $q$-deformed $AdS_5 \\times S^5$ S matrix also satisfies such a\nrelation.\n"}
{"abstract": "  In data-driven SHM, the signals recorded from systems in operation can be\nnoisy and incomplete. Data corresponding to each of the operational,\nenvironmental, and damage states are rarely available a priori; furthermore,\nlabelling to describe the measurements is often unavailable. In consequence,\nthe algorithms used to implement SHM should be robust and adaptive, while\naccommodating for missing information in the training-data -- such that new\ninformation can be included if it becomes available. By reviewing novel\ntechniques for statistical learning (introduced in previous work), it is argued\nthat probabilistic algorithms offer a natural solution to the modelling of SHM\ndata in practice. In three case-studies, probabilistic methods are adapted for\napplications to SHM signals -- including semi-supervised learning, active\nlearning, and multi-task learning.\n"}
{"abstract": "  Stability and dynamical behavior of binary Bose-Einstein condensed mixtures\ntrapped on the surface of a rigid spherical shell are investigated in the\nmean-field level, exploring the miscibility with and without vortex charges,\nconsidering repulsive and attractive interactions. In order to compute the\ncritical points for the stability, we follow the Bogoliubov-de Gennes method\nfor the analysis of perturbed solutions, with the constraint that initially the\nstationary states are in a complete miscible configuration. For the perturbed\nequal density mixture, of a homogeneous uniform gas and when hidden vorticity\nis verified, with the species having opposite azimuthal circulation, we\nconsider small perturbation analysis for each unstable mode, providing a\ncomplete diagram with the intra- and inter-species interaction role on the\nstability of the miscible system. Finally, beyond small perturbation analysis,\nwe explore the dynamics of some repulsive and attractive inter-species states\nby full numerical solutions of the time-dependent Gross-Pitaevskii equation.\n"}
{"abstract": "  We present optical and near-infrared (NIR, $YJH$-band) observations of 42\nType Ia supernovae (SNe Ia) discovered by the untargeted intermediate Palomar\nTransient Factory (iPTF) survey. This new data-set covers a broad range of\nredshifts and host galaxy stellar masses, compared to previous SN Ia efforts in\nthe NIR. We construct a sample, using also literature data at optical and NIR\nwavelengths, to examine claimed correlations between the host stellar masses\nand the Hubble diagram residuals. The SN magnitudes are corrected for host\ngalaxy extinction using either a global total-to-selective extinction ratio,\n$R_V$=2.0 for all SNe, or a best-fit $R_V$ for each SN individually. Unlike\nprevious studies which were based on a narrower range in host stellar mass, we\ndo not find evidence for a \"mass-step\", between the color- and\nstretch-corrected peak $J$ and $H$ magnitudes for galaxies below and above\n$\\log(M_{*}/M_{\\odot}) = 10$. However, the mass-step remains significant\n($3\\sigma$) at optical wavelengths ($g,r,i$) when using a global $R_V$, but\nvanishes when each SN is corrected using their individual best-fit $R_V$. Our\nstudy confirms the benefits of the NIR SN Ia distance estimates, as these are\nlargely exempted from the empirical corrections dominating the systematic\nuncertainties in the optical.\n"}
{"abstract": "  The heterogeneous computing paradigm represents a real programming challenge\ndue to the proliferation of devices with different hardware characteristics.\nRecently Intel introduced oneAPI, a new programming environment that allows\ncode developed in DPC++ to be run on different devices such as CPUs, GPUs,\nFPGAs, among others. This paper presents our first experiences in porting two\nCUDA applications to DPC++ using the oneAPI dpct tool. From the experimental\nwork, it was possible to verify that dpct does not achieve 100% of the\nmigration task; however, it performs most of the work, reporting the programmer\nof possible pending adaptations. Additionally, it was possible to verify the\nfunctional portability of the DPC++ code obtained, having successfully executed\nit on different CPU and GPU architectures.\n"}
{"abstract": "  We prove that an infinite field interpretable in a $p$-adically closed field\n$K$ is definably isomorphic to a finite extension of $K$. The result remains\ntrue in any $P$-minimal field where definable functions are generically\ndifferentiable.\n"}
{"abstract": "  We present a Flexible Ansatz for N-body Configuration Interaction (FANCI)\nthat includes any multideterminant wavefunction. This ansatz is a\ngeneralization of the Configuration Interaction (CI) wavefunction, where the\ncoefficients are replaced by a specified function of certain parameters. By\nmaking an appropriate choice for this function, we can reproduce popular\nwavefunction structures like CI, Coupled-Cluster, Tensor Network States, and\ngeminal-product wavefunctions. The universality of this framework suggests a\nprogramming structure that allows for the easy construction and optimization of\narbitrary wavefunctions. Here, we will discuss the structures of the FANCI\nframework and its implications for wavefunction properties, particularly\naccuracy, cost, and size-consistency. We demonstrate the flexibility of this\nframework by reconstructing popular wavefunction ans\\\"{a}tze and modifying them\nto construct novel wavefunction forms. FANCI provides a powerful framework for\nexploring, developing, and testing new wavefunction forms.\n"}
{"abstract": "  In this paper, we address the problem of estimating a covariance matrix of a\nmultivariate Gaussian distribution, relative to a Stein loss function, from a\ndecision theoretic point of view. We investigate the case where the covariance\nmatrix is invertible and the case when it is non--invertible in a unified\napproach.\n"}
{"abstract": "  A physical theory is presented for polarized light from an aspect of\npolarization singularity.This is carried out by analyzing the evolution\nequation of the Stokes parameters that is derived from the nonlinear\nSchrodinger type equation. The problem is explored from two aspects:The first\nis concerning the single mode Stokes parameter that is described by the\npropagation distance $ z $. The trajectory of the polarization singularities is\nsimply given by the points satisfying a rule: $ S_3 (z) = +(-)S_0 $ and $ 0 $\n($ S_3 $ represents the third component of the Stokes parameters), which form a\ntube like surface centered with the C-line surrounded by L-surfaces in the\npresence of the linear as well as the nonlinear birefringence. It is pointed\nout that the optical activity (or chirality) plays an \"obstacle\" to reveal the\npolarization singularity. As the other aspect, we consider the field theoretic\naspect of the Stokes parameters, for which a special solution, called an\noptical skyrmion, is constructed to explicate the polarization singularity.The\nskyrmion forms a {tube}, and the dynamical content of which is briefly\ndiscussed. On the basis of the work presented here further analysis would be\nexpected to reveal hidden aspects of polarization optics.\n"}
{"abstract": "  The KATRIN experiment has recently reported a new upper limit to the mass of\nthe electron neutrino of $m<0.8 eV$ (90\\%CL), and a best value of $m^2= 0.26\n\\pm 0.34 eV^2/c^4$ based on a fit to the observed tritium beta decay spectrum.\nHere another interpretation of their results is discussed.\n"}
{"abstract": "  Let $\\xi_1$, $\\xi_2,\\ldots$ be i.i.d. random variables of zero mean and\nfinite variance and $\\eta_1$, $\\eta_2,\\ldots$ positive i.i.d. random variables\nwhose distribution belongs to the domain of attraction of an $\\alpha$-stable\ndistribution, $\\alpha\\in (0,1)$. The two collections are assumed independent.\nWe consider a Markov chain with jumps of two types. If the present position of\nthe Markov chain is positive, then the jump $\\xi_k$ occurs; if the present\nposition of the Markov chain is nonpositive, then its next position is\n$\\eta_j$. We prove a functional limit theorem for this Markov chain under\nDonsker's scaling. The weak limit is a nonnegative process $(X(t))_{t\\geq 0}$\nsatisfying a stochastic equation ${\\rm d}X(t)={\\rm d}W(t)+ {\\rm\nd}U_\\alpha(L_X^{(0)}(t))$, where $W$ is a Brownian motion, $U_\\alpha$ is an\n$\\alpha$-stable subordinator which is independent of $W$, and $L_X^{(0)}$ is a\nlocal time of $X$ at $0$. Also, we explain that $X$ is a Feller Brownian motion\nwith a `jump-type' exit from $0$.\n"}
{"abstract": "  In this paper, we propose the use of two Passive Optical Network (PON)-based\nnetwork architectures to connect free-space Optical Wireless Communication\n(OWC) Access Points (APs) within a room with multiple users. We optimize\nthrough a Mixed Linear Integer Programming (MILP) model the assignment of\nmobile OWC users to more than one AP to improve the resilience of the fronthaul\nnetwork, i.e the OWC system and the wired network linked to APs, and study the\nimpact of users distribution and channel characteristics.\n"}
{"abstract": "  In this study, we present both data mining and information visualization\ntechniques to identify accident-prone areas, most accident-prone time, day, and\nmonth. Also, we surveyed among volunteers to understand which visualization\ntechniques help non-expert users to understand the findings better. Findings of\nthis study suggest that most accidents occur in the dusk (i.e., between 6 to 7\npm), and on Fridays. Results also suggest that most accidents occurred in\nOctober, which is a popular month for tourism. These findings are consistent\nwith social information and can help policymakers, residents, tourists, and\nother law enforcement agencies. This study can be extended to draw broader\nimplications.\n"}
{"abstract": "  A \\textit{functional $k$-batch} code of dimension $s$ consists of $n$ servers\nstoring linear combinations of $s$ linearly independent information bits. Any\nmultiset request of size $k$ of linear combinations (or requests) of the\ninformation bits can be recovered by $k$ disjoint subsets of the servers. The\ngoal under this paradigm is to find the minimum number of servers for given\nvalues of $s$ and $k$. A recent conjecture states that for any $k=2^{s-1}$\nrequests the optimal solution requires $2^s-1$ servers. This conjecture is\nverified for $s\\leq 5$ but previous work could only show that codes with\n$n=2^s-1$ servers can support a solution for $k=2^{s-2} + 2^{s-4} +\n\\left\\lfloor \\frac{ 2^{s/2}}{\\sqrt{24}} \\right\\rfloor$ requests. This paper\nreduces this gap and shows the existence of codes for $k=\\lfloor\n\\frac{5}{6}2^{s-1} \\rfloor - s$ requests with the same number of servers.\nAnother construction in the paper provides a code with $n=2^{s+1}-2$ servers\nand $k=2^{s}$ requests, which is an optimal result.These constructions are\nmainly based on Hadamard codes and equivalently provide constructions for\n\\textit{parallel Random I/O (RIO)} codes.\n"}
{"abstract": "  Given a singular modulus $j_0$ and a set of primes $S$, we study the problem\nof finding effective bounds for the cardinality of the set of singular moduli\n$j$ such that $j-j_0$ is an $S$-unit. For every $j_0 \\neq 0$, we provide a way\nof finding these bounds for infinitely many sets $S$. The same is true if\n$j_0=0$ and we assume the Generalized Riemann Hypothesis. Certain numerical\nexperiments will also lead to the formulation of a \"uniformity conjecture\" for\nsingular $S$-units.\n"}
{"abstract": "  Mixing halides in metal halide perovskites (MHPs) is an effective approach to\nadjust MHPs bandgap for applications in tandem solar cells. However,\nmixed-halide (MH-) MHPs undergo light-induced-phase-segregation (LIPS) under\ncontinuous illumination. Therefore, understanding the mechanism of LIPS is\nnecessary for developing stable MH-MHPs. In this work, we investigated LIPS in\nlayered (L) MHPs and discovered a critical role of spacer cations in LIPS.\nThrough probing chemical changes of LIPS, we unveil\nlight-induced-iodide-repulsion and the formation of Br-rich-phase in\nilluminated regions during LIPS. This discovery also gives insight into LIPS\nprocess in three dimensional (3D) MHPs. By further investigating LIPS in 3D\nMHPs, we reveal that LIPS induces not only the formation of Br-rich and I-rich\ndomains but also an overall change of halide distribution along the film\nthickness direction, which can affect the electronic energy alignment and\nconsequently MHPs devices performance. Moreover, LIPS is more significant in\nthe bulk due to larger population of photogenerated charge carriers. Overall,\nthis study reveals the chemical mechanism of LIPS in MHPs and its potential\neffect on device performance, offering insight into understanding LIPS\nmechanism and improving the stability of MHPs.\n"}
{"abstract": "  How to effectively incorporate cross-utterance information cues into a neural\nlanguage model (LM) has emerged as one of the intriguing issues for automatic\nspeech recognition (ASR). Existing research efforts on improving\ncontextualization of an LM typically regard previous utterances as a sequence\nof additional input and may fail to capture complex global structural\ndependencies among these utterances. In view of this, we in this paper seek to\nrepresent the historical context information of an utterance as\ngraph-structured data so as to distill cross-utterances, global word\ninteraction relationships. To this end, we apply a graph convolutional network\n(GCN) on the resulting graph to obtain the corresponding GCN embeddings of\nhistorical words. GCN has recently found its versatile applications on\nsocial-network analysis, text summarization, and among others due mainly to its\nability of effectively capturing rich relational information among elements.\nHowever, GCN remains largely underexplored in the context of ASR, especially\nfor dealing with conversational speech. In addition, we frame ASR N-best\nreranking as a prediction problem, leveraging bidirectional encoder\nrepresentations from transformers (BERT) as the vehicle to not only seize the\nlocal intrinsic word regularity patterns inherent in a candidate hypothesis but\nalso incorporate the cross-utterance, historical word interaction cues\ndistilled by GCN for promoting performance. Extensive experiments conducted on\nthe AMI benchmark dataset seem to confirm the pragmatic utility of our methods,\nin relation to some current top-of-the-line methods.\n"}
{"abstract": "  Gas phase Elemental abundances in Molecular CloudS (GEMS) is an IRAM 30m\nLarge Program designed to estimate the S, C, N, and O depletions and gas\nionization degree, X(e-), in a set of star-forming filaments of Taurus, Perseus\nand Orion. Our immediate goal is to build up a complete database of molecular\nabundances that can serve as an observational basis for estimating X(e-) and\nthe C, O, N, and S depletions through chemical modeling. We observed and\nderived the abundances of 14 species (13CO, C18O, HCO+, H13CO+, HC18O+, HCN,\nH13CN, HNC, HCS+, CS, SO, 34SO, H2S, and OCS) in 244 positions, covering the AV\n3 to 100 mag, n(H2) a few 10$^{3}$ to 10$^6$ cm$^{-3}$, and Tk 10 to 30 K\nranges in these clouds, avoiding protostars, HII regions, and outflows. A\nstatistical analysis is carried out to identify general trends between\ndifferent species and with physical parameters. Relations between molecules\nreveal strong linear correlations which define three different families: (1)\n13CO and C18O; (2) H13CO+, HC18O+, H13CN, and HNC; and (3) the S-bearing\nmolecules. The abundances of the CO isotopologs increase with the gas kinetic\ntemperature until TK 15 K. For higher temperatures, the abundance remains\nconstant with a scatter of a factor of 3. The abundances of H13CO+, HC18O+,\nH13CN, and HNC are well correlated with each other, and all of them decrease\nwith molecular hydrogen density, following the law n(H2)$^{-0.8\\pm0.2}$. The\nabundances of S-bearing species also decrease with n(H2) at a rate of\n(S-bearing/H)gas n(H2)$^{-0.6\\pm0.1}$. The abundances of molecules belonging to\ngroups 2 and 3 do not present any clear trend with gas temperature. At scales\nof molecular clouds, the C18O abundance is the quantity that better correlates\nwith the cloud mass. We discuss the utility of the 13CO/C18O, HCO+/H13CO+, and\nH13CO+/H13CN abundance ratios as chemical diagnostics of star formation in\nexternal galaxies.\n"}
{"abstract": "  We give an abstract approach to approximations with a wide range of\nregularity classes $X$ in spaces of pseudocontinuable functions\n$K^p_\\vartheta$, where $\\vartheta$ is an inner function and $p>0$. More\nprecisely, we demonstrate a general principle, attributed to A. B. Aleksandrov,\nwhich asserts that if a certain linear manifold $X$ is dense in the space of\npseudocontinuable functions $K^{p_0}_\\vartheta$, for some $p_0>0$, then $X$ is\nin fact dense in $K^p_{\\vartheta}$, for all $p>0$. %This allows for\ngeneralizations of the recent result on density by functions with smooth\nboundary extensions. Moreover, for a rich class of Banach spaces of analytic\nfunctions $X$, we describe the precise mechanism that determines when $X$ is\ndense in a certain space of pseudocontinuable functions. As a consequence, we\nobtain an extension of Aleksandrov's density theorem to the class of analytic\nfunctions with uniformly convergent Taylor series.\n"}
{"abstract": "  Memory-compute disaggregation promises transparent elasticity, high\nutilization and balanced usage for resources in data centers by physically\nseparating memory and compute into network-attached resource \"blades\". However,\nexisting designs achieve performance at the cost of resource elasticity,\nrestricting memory sharing to a single compute blade to avoid costly memory\ncoherence traffic over the network.\n  In this work, we show that emerging programmable network switches can enable\nan efficient shared memory abstraction for disaggregated architectures by\nplacing memory management logic in the network fabric. We find that\ncentralizing memory management in the network permits bandwidth and\nlatency-efficient realization of in-network cache coherence protocols, while\nprogrammable switch ASICs support other memory management logic at line-rate.\nWe realize these insights into MIND, an in-network memory management unit for\nrack-scale memory disaggregation. MIND enables transparent resource elasticity\nwhile matching the performance of prior memory disaggregation proposals for\nreal-world workloads.\n"}
{"abstract": "  Action segmentation refers to inferring boundaries of semantically consistent\nvisual concepts in videos and is an important requirement for many video\nunderstanding tasks. For this and other video understanding tasks, supervised\napproaches have achieved encouraging performance but require a high volume of\ndetailed frame-level annotations. We present a fully automatic and unsupervised\napproach for segmenting actions in a video that does not require any training.\nOur proposal is an effective temporally-weighted hierarchical clustering\nalgorithm that can group semantically consistent frames of the video. Our main\nfinding is that representing a video with a 1-nearest neighbor graph by taking\ninto account the time progression is sufficient to form semantically and\ntemporally consistent clusters of frames where each cluster may represent some\naction in the video. Additionally, we establish strong unsupervised baselines\nfor action segmentation and show significant performance improvements over\npublished unsupervised methods on five challenging action segmentation\ndatasets. Our code is available at\nhttps://github.com/ssarfraz/FINCH-Clustering/tree/master/TW-FINCH\n"}
{"abstract": "  We study the problem of out-of-distribution dynamics (OODD) detection, which\ninvolves detecting when the dynamics of a temporal process change compared to\nthe training-distribution dynamics. This is relevant to applications in\ncontrol, reinforcement learning (RL), and multi-variate time-series, where\nchanges to test time dynamics can impact the performance of learning\ncontrollers/predictors in unknown ways. This problem is particularly important\nin the context of deep RL, where learned controllers often overfit to the\ntraining environment. Currently, however, there is a lack of established OODD\nbenchmarks for the types of environments commonly used in RL research. Our\nfirst contribution is to design a set of OODD benchmarks derived from common RL\nenvironments with varying types and intensities of OODD. Our second\ncontribution is to design a strong OODD baseline approach based on recurrent\nimplicit quantile networks (RIQNs), which monitors autoregressive prediction\nerrors for OODD detection. Our final contribution is to evaluate the RIQN\napproach on the benchmarks to provide baseline results for future comparison.\n"}
{"abstract": "  Boundary layer turbulence in coastal regions differs from that in deep ocean\nbecause of bottom interactions. In this paper, we focus on the merging of\nsurface and bottom boundary layers in a finite-depth coastal ocean by\nnumerically solving the wave-averaged equations using a large eddy simulation\nmethod. The ocean fluid is driven by combined effects of wind stress, surface\nwave, and a steady current in the presence of stable vertical stratification.\nThe resulting flow consists of two overlapping boundary layers, i.e. surface\nand bottom boundary layers, separated by an interior stratification. The\noverlapping boundary layers evolve through three phases, i.e. a rapid\ndeepening, an oscillatory equilibrium and a prompt merger, separated by two\ntransitions. Before the merger, internal waves are observed in the stratified\nlayer, and they are excited mainly by Langmuir turbulence in the surface\nboundary layer. These waves induce a clear modulation on the bottom-generated\nturbulence, facilitating the interaction between the surface and bottom\nboundary layers. After the merger, the Langmuir circulations originally\nconfined to the surface layer are found to grow in size and extend down to the\nsea bottom (even though the surface waves do not feel the bottom), reminiscent\nof the well-organized Langmuir supercells. These full-depth Langmuir\ncirculations promote the vertical mixing and enhance the bottom shear, leading\nto a significant enhancement of turbulence levels in the vertical column.\n"}
{"abstract": "  Deep neural networks have become a pervasive tool in science and engineering.\nHowever, modern deep neural networks' growing energy requirements now\nincreasingly limit their scaling and broader use. We propose a radical\nalternative for implementing deep neural network models: Physical Neural\nNetworks. We introduce a hybrid physical-digital algorithm called Physics-Aware\nTraining to efficiently train sequences of controllable physical systems to act\nas deep neural networks. This method automatically trains the functionality of\nany sequence of real physical systems, directly, using backpropagation, the\nsame technique used for modern deep neural networks. To illustrate their\ngenerality, we demonstrate physical neural networks with three diverse physical\nsystems-optical, mechanical, and electrical. Physical neural networks may\nfacilitate unconventional machine learning hardware that is orders of magnitude\nfaster and more energy efficient than conventional electronic processors.\n"}
{"abstract": "  Correlated band theory implemented as a combination of density functional\ntheory with exact diagonalization [DFT+U(ED)] of the Anderson impurity term\nwith Coulomb repulsion $U$ in the open 14-orbital $5f$ shell is applied to\nUTe$_2$. The small gap for $U$=0, evidence of the half-filled $j=\\frac{5}{2}$\nsubshell of $5f^3$ uranium, is converted for $U$=3 eV to a flat band semimetal\nwith small heavy-carrier Fermi surfaces that will make properties sensitive to\npressure, magnetic field, and off-stoichiometry, as observed experimentally.\nThe predicted Kondo temperature around 100 K matches the experimental values\nfrom resistivity. The electric field gradients for the two Te sites are\ncalculated by DFT+U(ED) to differ by a factor of seven, indicating a strong\nsite distinction, while the anisotropy factor $\\eta=0.18$ is similar for all\nthree sites. The calculated uranium moment $<M^2>^{1/2}$ of 3.5$\\mu_B$ is\nroughly consistent with the published experimental Curie-Weiss values of\n2.8$\\mu_B$ and 3.3$\\mu_B$ (which are field-direction dependent), and the\ncalculated separate spin and orbital moments are remarkably similar to Hund's\nrule values for an $f^3$ ion. The $U$=3 eV spectral density is compared with\nangle-integrated and angle-resolved photoemission spectra, with agreement that\nthere is strong $5f$ character at, and for several hundred meV below, the Fermi\nenergy. Our results support the picture that the underlying ground state of\nUTe$_2$ is that of a half-filled $j=\\frac{5}{2}$ subshell with two half-filled\n$m_j=\\pm\\frac{1}{2}$ orbitals forming a narrow gap by hybridization, then\ndriven to a conducting state by configuration mixing (spin-charge\nfluctuations). UTe$_2$ displays similarities to UPt$_3$ with its $5f$ dominated\nFermi surfaces rather than a strongly localized Kondo lattice system.\n"}
{"abstract": "  The matrix functions can be defined by Cauchy's integral formula and can be\napproximated by the linear combination of inverses of shifted matrices using a\nquadrature formula. In this paper, we show a concrete construction of a\nframework to implement the linear combination of the inverses on quantum\ncomputers and propose a quantum algorithm for matrix functions based on the\nframework. Compared with the previous study [S. Takahira, A. Ohashi, T. Sogabe,\nand T.S. Usuda, Quant. Inf. Comput., 20, 1&2, 14--36, (Feb. 2020)] that\nproposed a quantum algorithm to compute a quantum state for the matrix function\nbased on the circular contour centered at the origin, the quantum algorithm in\nthe present paper can be applied to a more general contour. Moreover, the\nalgorithm is described by the block-encoding framework. Similarly to the\nprevious study, the algorithm can be applied even if the input matrix is not a\nHermitian or normal matrix.\n"}
{"abstract": "  The AWAKE experiment relies on the self-modulation instability of a long\nproton bunch to effectively drive wakefields and accelerate an electron bunch\nto GeV-level energies. During the first experimental run (2016-2018) the\ninstability was made phase reproducible by means of a seeding process: a short\nlaser pulse co-propagates within the proton bunch in a rubidium vapor. Thus,\nthe fast creation of plasma and the onset of beam-plasma interaction within the\nbunch drives seed wakefields. However, this seeding method leaves the front of\nthe bunch not modulated. The bunch front could self-modulate in a second,\npreformed plasma and drive wakefields that would interfere with those driven by\nthe (already self-modulated) back of the bunch and with the acceleration\nprocess. We present studies of the seeded the self-modulation (SSM) of a long\nproton bunch using a short electron bunch. The short seed bunch is placed ahead\nof the proton bunch leading to self-modulation of the entire bunch. Numerical\nsimulations show that this method have other advantages when compared to the\nionization front method. We discuss the requirements for the electron bunch\nparameters (charge, emittance, transverse size at the focal point, length), to\neffectively seed the self-modulation process. We also present preliminary\nexperimental studies on the electron bunch seed wakefields generation.\n"}
{"abstract": "  We consider the problem of providing users of deep Reinforcement Learning\n(RL) based systems with a better understanding of when their output can be\ntrusted. We offer an explainable artificial intelligence (XAI) framework that\nprovides a three-fold explanation: a graphical depiction of the systems\ngeneralization and performance in the current game state, how well the agent\nwould play in semantically similar environments, and a narrative explanation of\nwhat the graphical information implies. We created a user-interface for our XAI\nframework and evaluated its efficacy via a human-user experiment. The results\ndemonstrate a statistically significant increase in user trust and acceptance\nof the AI system with explanation, versus the AI system without explanation.\n"}
{"abstract": "  Understanding the effects of atomic structure modification in hematite\nphotoanodes is essential for the rational design of high-efficiency\nfunctionalizations. Recently it was found that interface modification with\nSn/Sb segregates considerably increases hematite photocatalytic efficiency.\nHowever, the understanding of the different electronic effects of these\nmodifications at the atomic level is still lacking. This letter describes the\nsegregation effects of two different dopants-Sn and Sb-on both the solid-solid\n(grain-boundaries) and solid-liquid interfaces (surfaces) of hematite. Within\nan ab-initio approach, we quantitatively extract the potential barrier\nreduction on polycrystalline interfaces due to the dopant, which causes an\nincrease in the inter-grain electron transport. Concomitantly, the dopants'\nsegregation on hematite surfaces results in a decrease of the oxygen vacancy\nformation energy. Such vacancies lead to the experimentally observed rise of\nthe flat-band potential. The comprehension of the electronic effects of dopants\non both types of interfaces explains the experimental peak efficiency of\ninterface-modified hematite with dopant segregates, also enabling the control\nand design of interfaces for different higher-efficiency applications.\n"}
{"abstract": "  Due to the use of huge number of sensors and the increasing use of\ncommunication networks, cyber-physical systems (CPS) are becoming vulnerable to\ncyber-attacks. The ever-increasing complexity of CPS bring up the need for\ndata-driven machine learning applications to fill in the need of model creation\nto describe the system behavior. In this paper, a novel stability condition\npredictor based on cascaded feedforward neural network is proposed. The\nproposed method aims to identify anomaly due to cyber or physical disturbances\nas an early sign of instability. The proposed neural network utilizes cascaded\nconnections in order to increase accuracy of the prediction. The conjugate\ngradient backpropagation and Polak-Ribi\\`ere formula are utilized for training\nprocess. This method also can predict the critical generators to mitigate the\neffect of the cascading failure and consequent blackout in the system.\nSimulations results on the IEEE 39-bus system indicate the superiority of the\nproposed method in terms of accuracy, speed, and robustness.\n"}
{"abstract": "  Computer simulations generate microscopic trajectories of complex systems at\na single thermodynamic state point. We recently introduced a Maximum Caliber\n(MaxCal) approach for dynamical reweighting. Our approach mapped these\ntrajectories to a Markovian description on the configurational coordinates, and\nreweighted path probabilities as a function of external forces. Trajectory\nprobabilities can be dynamically reweighted both from and to equilibrium or\nnon-equilibrium steady states. As the system's dimensionality increases, an\nexhaustive description of the microtrajectories becomes prohibitive--even with\na Markovian assumption. Instead we reduce the dimensionality of the\nconfigurational space to collective variables (CVs). Going from configurational\nto CV space, we define local entropy productions derived from configurationally\naveraged mean forces. The entropy production is shown to be a suitable\nconstraint on MaxCal for non-equilibrium steady states expressed as a function\nof CVs. We test the reweighting procedure on two systems: a particle subject to\na two-dimensional potential and a coarse-grained peptide. Our CV-based MaxCal\napproach expands dynamical reweighting to larger systems, for both static and\ndynamical properties, and across a large range of driving forces.\n"}
{"abstract": "  Motivation: Traditional image attribution methods struggle to satisfactorily\nexplain predictions of neural networks. Prediction explanation is important,\nespecially in medical imaging, for avoiding the unintended consequences of\ndeploying AI systems when false positive predictions can impact patient care.\nThus, there is a pressing need to develop improved models for model\nexplainability and introspection. Specific problem: A new approach is to\ntransform input images to increase or decrease features which cause the\nprediction. However, current approaches are difficult to implement as they are\nmonolithic or rely on GANs. These hurdles prevent wide adoption. Our approach:\nGiven an arbitrary classifier, we propose a simple autoencoder and gradient\nupdate (Latent Shift) that can transform the latent representation of a\nspecific input image to exaggerate or curtail the features used for prediction.\nWe use this method to study chest X-ray classifiers and evaluate their\nperformance. We conduct a reader study with two radiologists assessing 240\nchest X-ray predictions to identify which ones are false positives (half are)\nusing traditional attribution maps or our proposed method. Results: We found\nlow overlap with ground truth pathology masks for models with reasonably high\naccuracy. However, the results from our reader study indicate that these models\nare generally looking at the correct features. We also found that the Latent\nShift explanation allows a user to have more confidence in true positive\npredictions compared to traditional approaches (0.15$\\pm$0.95 in a 5 point\nscale with p=0.01) with only a small increase in false positive predictions\n(0.04$\\pm$1.06 with p=0.57).\n  Accompanying webpage: https://mlmed.org/gifsplanation\n  Source code: https://github.com/mlmed/gifsplanation\n"}
{"abstract": "  Tropical geometry with the max-plus algebra has been applied to statistical\nlearning models over tree spaces because geometry with the tropical metric over\ntree spaces has some nice properties such as convexity in terms of the tropical\nmetric. One of the challenges in applications of tropical geometry to tree\nspaces is the difficulty interpreting outcomes of statistical models with the\ntropical metric. This paper focuses on combinatorics of tree topologies along a\ntropical line segment, an intrinsic geodesic with the tropical metric, between\ntwo phylogenetic trees over the tree space and we show some properties of a\ntropical line segment between two trees. Specifically we show that a\nprobability of a tropical line segment of two randomly chosen trees going\nthrough the origin (the star tree) is zero if the number of leave is greater\nthan four, and we also show that if two given trees differ only one nearest\nneighbor interchange (NNI) move, then the tree topology of a tree in the\ntropical line segment between them is the same tree topology of one of these\ngiven two trees with possible zero branch lengths.\n"}
{"abstract": "  The ORBIT code has been widely used for beam dynamics simulations including\ninjection and acceleration in high-intensity hadron synchrotrons. When the\nORBIT's 1D longitudinal tracking was employed for the acceleration process in\nCSNS/RCS, the longitudinal emittance in eV-s was found decreasing substantially\nduring acceleration, though the adiabatic condition is still met during this\nprocess. This is against the Liouville theorem that predicts the preservation\nof the emittance during acceleration. The recent machine study in the\naccelerator and the simulations with a self-made code demonstrate that the\nlongitudinal emittance is almost invariant, which further indicates that the\nORBIT longitudinal tracking might be incorrect. A detailed check-over in the\nORBIT code source finds that the longitudinal finite difference equation used\nin the code is erroneous when applied to an acceleration process. The new code\nformat PyORBIT has the same problem. After the small secondary factor is\nincluded in the code, ORBIT can produce results keeping the longitudinal\nemittance invariant. This paper presents some details about the study.\n"}
{"abstract": "  The Everett interpretation of quantum mechanics divides naturally into two\nparts: first, the interpretation of the structure of the quantum state, in\nterms of branching, and second, the interpretation of this branching structure\nin terms of probability. This is the first of two reviews of the Everett\ninterpretation, and focuses on structure, with particular attention to the role\nof decoherence theory. Written in terms of the quantum histories formalism,\ndecoherence theory just is the theory of branching structure, in Everett's\nsense.\n"}
{"abstract": "  The literature on awareness modeling includes both syntax-free and\nsyntax-based frameworks. Heifetz, Meier \\& Schipper (HMS) propose a lattice\nmodel of awareness that is syntax-free. While their lattice approach is elegant\nand intuitive, it precludes the simple option of relying on formal language to\ninduce lattices, and does not explicitly distinguish uncertainty from\nunawareness. Contra this, the most prominent syntax-based solution, the\nFagin-Halpern (FH) model, accounts for this distinction and offers a simple\nrepresentation of awareness, but lacks the intuitiveness of the lattice\nstructure. Here, we combine these two approaches by providing a lattice of\nKripke models, induced by atom subset inclusion, in which uncertainty and\nunawareness are separate. We show our model equivalent to both HMS and FH\nmodels by defining transformations between them which preserve satisfaction of\nformulas of a language for explicit knowledge, and obtain completeness through\nour and HMS' results. Lastly, we prove that the Kripke lattice model can be\nshown equivalent to the FH model (when awareness is propositionally determined)\nalso with respect to the language of the Logic of General Awareness, for which\nthe FH model where originally proposed.\n"}
{"abstract": "  Transfer learning, also referred as knowledge transfer, aims at reusing\nknowledge from a source dataset to a similar target one. While many empirical\nstudies illustrate the benefits of transfer learning, few theoretical results\nare established especially for regression problems. In this paper a theoretical\nframework for the problem of parameter transfer for the linear model is\nproposed. It is shown that the quality of transfer for a new input vector $x$\ndepends on its representation in an eigenbasis involving the parameters of the\nproblem. Furthermore a statistical test is constructed to predict whether a\nfine-tuned model has a lower prediction quadratic risk than the base target\nmodel for an unobserved sample. Efficiency of the test is illustrated on\nsynthetic data as well as real electricity consumption data.\n"}
{"abstract": "  New mass-produced, wide-field, small-aperture telescopes have the potential\nto revolutionize ground-based astronomy by greatly reducing the cost of\ncollecting area. In this paper, we introduce a new class of large telescope\nbased on these advances: an all-sky, arcsecond-resolution, 1000-telescope array\nwhich builds a simultaneously high-cadence and deep survey by observing the\nentire sky all night. As a concrete example, we describe the Argus Array, a\n5m-class telescope with an all-sky field of view and the ability to reach\nextremely high cadences using low-noise CMOS detectors. Each 55 GPix Argus\nexposure covers 20% of the entire sky to g=19.6 each minute and g=21.9 each\nhour; a high-speed mode will allow sub-second survey cadences for short times.\nDeep coadds will reach g=23.6 every five nights over 47% of the sky; a\nlarger-aperture array telescope, with an \\'etendue close to the Rubin\nObservatory, could reach g=24.3 in five nights. These arrays can build\ntwo-color, million-epoch movies of the sky, enabling sensitive and rapid\nsearches for high-speed transients, fast-radio-burst counterparts,\ngravitational-wave counterparts, exoplanet microlensing events, occultations by\ndistant solar system bodies, and myriad other phenomena. An array of O(1,000)\ntelescopes, however, would be one of the most complex astronomical instruments\nyet built. Standard arrays with hundreds of tracking mounts entail thousands of\nmoving parts and exposed optics, and maintenance costs would rapidly outpace\nthe mass-produced-hardware cost savings compared to a monolithic large\ntelescope. We discuss how to greatly reduce operations costs by placing all\noptics in a thermally controlled, sealed dome with a single moving part.\nCoupled with careful software scope control and use of existing pipelines, we\nshow that the Argus Array could become the deepest and fastest Northern sky\nsurvey, with total costs below $20M.\n"}
{"abstract": "  The advent of the Internet and a large number of digital technologies has\nbrought with it many different challenges. A large amount of data is found on\nthe web, which in most cases is unstructured and unorganized, and this\ncontributes to the fact that the use and manipulation of this data is quite a\ndifficult process. Due to this fact, the usage of different machine and deep\nlearning techniques for Text Classification has gained its importance, which\nimproved this discipline and made it more interesting for scientists and\nresearchers for further study. This paper aims to classify the pedagogical\ncontent using two different models, the K-Nearest Neighbor (KNN) from the\nconventional models and the Long short-term memory (LSTM) recurrent neural\nnetwork from the deep learning models. The result indicates that the accuracy\nof classifying the pedagogical content reaches 92.52 % using KNN model and\n87.71 % using LSTM model.\n"}
{"abstract": "  In this work, we investigate if the wav2vec 2.0 self-supervised pretraining\nhelps mitigate the overfitting issues with connectionist temporal\nclassification (CTC) training to reduce its performance gap with flat-start\nlattice-free MMI (E2E-LFMMI) for automatic speech recognition with limited\ntraining data. Towards that objective, we use the pretrained wav2vec 2.0 BASE\nmodel and fine-tune it on three different datasets including out-of-domain\n(Switchboard) and cross-lingual (Babel) scenarios. Our results show that for\nsupervised adaptation of the wav2vec 2.0 model, both E2E-LFMMI and CTC achieve\nsimilar results; significantly outperforming the baselines trained only with\nsupervised data. Fine-tuning the wav2vec 2.0 model with E2E-LFMMI and CTC we\nobtain the following relative WER improvements over the supervised baseline\ntrained with E2E-LFMMI. We get relative improvements of 40% and 44% on the\nclean-set and 64% and 58% on the test set of Librispeech (100h) respectively.\nOn Switchboard (300h) we obtain relative improvements of 33% and 35%\nrespectively. Finally, for Babel languages, we obtain relative improvements of\n26% and 23% on Swahili (38h) and 18% and 17% on Tagalog (84h) respectively.\n"}
{"abstract": "  Scattered polynomials over a finite field $\\mathbb{F}_{q^n}$ have been\nintroduced by Sheekey in 2016, and a central open problem regards the\nclassification of those that are exceptional. So far, only two families of\nexceptional scattered polynomials are known. Very recently, Longobardi and\nZanella weakened the property of being scattered by introducing the notion of\nL-$q^t$-partially scattered and R-$q^t$-partially scattered polynomials, for\n$t$ a divisor of $n$. Indeed, a polynomial is scattered if and only if it is\nboth L-$q^t$-partially scattered and R-$q^t$-partially scattered. In this\npaper, by using techniques from algebraic geometry over finite fields and\nfunction fields theory, we show that the property which is is the hardest to be\npreserved is the L-$q^t$-partially scattered one. On the one hand, we are able\nto extend the classification results of exceptional scattered polynomials to\nexceptional L-$q^t$-partially scattered polynomials. On the other hand, the\nR-$q^t$-partially scattered property seems more stable. We present a large\nfamily of R-$q^t$-partially scattered polynomials, containing examples of\nexceptional R-$q^t$-partially scattered polynomials, which turn out to be\nconnected with linear sets of so-called pseudoregulus type. In order to detect\nnew examples of polynomials which are R-$q^t$-partially scattered, we introduce\ntwo different notions of equivalence preserving this property and concerning\nnatural actions of the groups ${\\rm \\Gamma L}(2,q^n)$ and ${\\rm \\Gamma\nL}(2n/t,q^t)$. In particular, our family contains many examples of inequivalent\npolynomials, and geometric arguments are used to determine the equivalence\nclasses under the action of ${\\rm \\Gamma L}(2n/t,q^t)$.\n"}
{"abstract": "  This paper aims to solve a distributed learning problem under Byzantine\nattacks. In the underlying distributed system, a number of unknown but\nmalicious workers (termed as Byzantine workers) can send arbitrary messages to\nthe master and bias the learning process, due to data corruptions, computation\nerrors or malicious attacks. Prior work has considered a total variation (TV)\nnorm-penalized approximation formulation to handle the Byzantine attacks, where\nthe TV norm penalty forces the regular workers' local variables to be close,\nand meanwhile, tolerates the outliers sent by the Byzantine workers. To solve\nthe TV norm-penalized approximation formulation, we propose a Byzantine-robust\nstochastic alternating direction method of multipliers (ADMM) that fully\nutilizes the separable problem structure. Theoretically, we prove that the\nproposed method converges to a bounded neighborhood of the optimal solution at\na rate of O(1/k) under mild assumptions, where k is the number of iterations\nand the size of neighborhood is determined by the number of Byzantine workers.\nNumerical experiments on the MNIST and COVERTYPE datasets demonstrate the\neffectiveness of the proposed method to various Byzantine attacks.\n"}
{"abstract": "  The propagators for the Faddeev-Popov (FP) ghosts in Yang-Mills theory and\nperturbative gravity in the covariant gauge are infrared (IR) divergent in de\nSitter spacetime. An IR cutoff in the momentum space to regularize these\ndivergences breaks the de Sitter invariance. These IR divergences are due to\nthe spatially constant modes in the Yang-Mills case and the modes proportional\nto the Killing vectors in the case of perturbative gravity. It has been\nproposed that these IR divergences can be removed, with the de Sitter\ninvariance preserved, by first regularizing them with an additional mass term\nfor the FP ghosts and then taking the massless limit. In the Yang-Mills case,\nthis procedure has been shown to correspond to requiring that the physical\nstates, and the vacuum state in particular, be annihilated by some conserved\ncharges in the Landau gauge. In this paper we show that there are similar\nconserved charges in perturbative gravity in the covariant Landau gauge in de\nSitter spacetime and that the IR-regularization procedure described above also\ncorrespond to requiring that the vacuum state be annihilated by these charges\nwith a natural definition of the interacting vacuum state.\n"}
{"abstract": "  In this paper, the asymptotic behavior of the solutions of a monotone problem\nposed in a locally periodic oscillating domain is studied. Nonlinear monotone\nboundary conditions are imposed on the oscillating part of the boundary whereas\nthe Dirichlet condition is considered on the smooth separate part. Using the\nunfolding method, under natural hypothesis on the regularity of the domain, we\nprove the weak $L^2$-convergence of the zero-extended solutions of the\nnonlinear problem and their flows to the solutions of a limit distributional\nproblem.\n"}
{"abstract": "  We compute Witt groups of maximal isotropic Grassmannians, aka. spinor\nvarieties. They are examples of type D homogenuous varieties. Our method relies\non the Blow-up setup of Balmer-Calm\\`es, and we investigate the connecting\nhomomorphism via the projective bundle formula of Walter-Nenashev, the\nprojection formula of Calm\\`es-Hornbostel and the excess intersection formula\nof Fasel. The computation in the Type D case can be presented by so called\n\"even shifted young diagrams\".\n"}
{"abstract": "  Electromagnetic waves in vacuum and most materials have transverse\npolarization. Longitudinal electromagnetic waves with electric field parallel\nto wave vector are very rare and appear under special conditions in a limited\nclass of media, for example in plasma. In this work, we study the dispersion\nproperties of an easy-to-manufacture metamaterial consisting of two\nthree-dimensional cubic lattices of connected metallic wires inserted one into\nanother, also known as an interlaced wire medium. It is shown that the\nmetamaterial supports longitudinal waves at extremely wide frequency band from\nvery low frequencies up to the Bragg resonances of the structure. The waves\nfeature unprecedentedly short wavelengths comparable to the period of the\nmaterial. The revealed effects highlight spatially dispersive response of\ninterlaced wire medium and provide a route toward generating electromagnetic\nfields with strong spatial variation.\n"}
{"abstract": "  The mobile communication system has transformed to be the fundamental\ninfrastructure to support digital demands from all industry sectors, and 6G is\nenvisioned to go far beyond the communication-only purpose. There is coming to\na consensus that 6G will treat Artificial Intelligence (AI) as the cornerstone\nand has a potential capability to provide \"intelligence inclusion\", which\nimplies to enable the access of AI services at anytime and anywhere by anyone.\nApparently, the intelligent inclusion vision produces far-reaching influence on\nthe corresponding network architecture design in 6G and deserves a clean-slate\nrethink. In this article, we propose an end-to-end system architecture design\nscope for 6G, and talk about the necessity to incorporate an independent data\nplane and a novel intelligent plane with particular emphasis on end-to-end AI\nworkflow orchestration, management and operation. We also highlight the\nadvantages to provision converged connectivity and computing services at the\nnetwork function plane. Benefiting from these approaches, we believe that 6G\nwill turn to an \"everything as a service\" (XaaS) platform with significantly\nenhanced business merits.\n"}
{"abstract": "  When multiple parties that deal with private data aim for a collaborative\nprediction task such as medical image classification, they are often\nconstrained by data protection regulations and lack of trust among\ncollaborating parties. If done in a privacy-preserving manner, predictive\nanalytics can benefit from the collective prediction capability of multiple\nparties holding complementary datasets on the same machine learning task. This\npaper presents PRICURE, a system that combines complementary strengths of\nsecure multi-party computation (SMPC) and differential privacy (DP) to enable\nprivacy-preserving collaborative prediction among multiple model owners. SMPC\nenables secret-sharing of private models and client inputs with non-colluding\nsecure servers to compute predictions without leaking model parameters and\ninputs. DP masks true prediction results via noisy aggregation so as to deter a\nsemi-honest client who may mount membership inference attacks. We evaluate\nPRICURE on neural networks across four datasets including benchmark medical\nimage classification datasets. Our results suggest PRICURE guarantees privacy\nfor tens of model owners and clients with acceptable accuracy loss. We also\nshow that DP reduces membership inference attack exposure without hurting\naccuracy.\n"}
{"abstract": "  For a knot $K$, the doubly slice genus $g_{ds}(K)$ is the minimal $g$ such\nthat $K$ divides a closed, orientable, and unknotted surface of genus $g$\nembedded in $S^4$. In this paper, we identify the doubly slice genera of 2909\nof the 2977 prime knots which have a crossing number of 12 or fewer.\n"}
{"abstract": "  Let $\\Omega \\subset \\mathbb{R}^{n+1}$, $n\\ge 2$, be a 1-sided\nnon-tangentially accessible domain (i.e., quantitatively open and\npath-connected) satisfiying the capacity density condition. Let $L_0\nu=-\\mathrm{div}(A_0 \\nabla u)$, $Lu=-\\mathrm{div}(A\\nabla u)$ be two real\nuniformly elliptic operators in $\\Omega$, with $\\omega_{L_0}, \\omega_L$ the\nassociated elliptic measures. We establish the equivalence between the\nfollowing properties: (i) $\\omega_L \\in A_{\\infty}(\\omega_{L_0})$, (ii) $L$ is\n$L^p(\\omega_{L_0})$-solvable for some $p\\in (1,\\infty)$, (iii) bounded null\nsolutions of $L$ satisfy Carleson measure estimates with respect to\n$\\omega_{L_0}$, (iv) the conical square function is controlled by the\nnon-tangential maximal function in $L^q(\\omega_{L_0})$ for some (or for all)\n$q\\in (0,\\infty)$ for any null solution of $L$, and (v) $L$ is\n$\\mathrm{BMO}(\\omega_{L_0})$-solvable. Moreover, in each of the properties\n(ii)-(v) it is enough to consider the class of solutions $u(X)=\\omega_L^X(S)$\nwith arbitrary Borel sets $S\\subset\\partial\\Omega$.\n  Also, we characterize the absolute continuity of $\\omega_{L_0}$ with respect\nto $\\omega_L$ in terms of some qualitative local $L^2(\\omega_{L_0})$ estimates\nfor the truncated conical square function for any bounded null solution of $L$.\nThis is also equivalent to the finiteness $\\omega_{L_0}$-a.e. of the truncated\nconical square function for any bounded null solution of $L$. As applications,\nwe show that $\\omega_{L_0}\\ll\\omega_L$ if the disagreement of the coefficients\nsatisfies some qualitative quadratic estimate in truncated cones for\n$\\omega_{L_0}$-a.e. vertex. Finally, when $L_0$ is either the transpose of $L$\nor its symmetric part, we obtain the corresponding absolute continuity when the\nantisymmetric part of the coefficients has some controlled oscillation in\ntruncated cones for $\\omega_{L_0}$-a.e. vertex.\n"}
{"abstract": "  We propose a model for hierarchical structured data as an extension to the\nstochastic temporal convolutional network. The proposed model combines an\nautoregressive model with a hierarchical variational autoencoder and\ndownsampling to achieve superior computational complexity. We evaluate the\nproposed model on two different types of sequential data: speech and\nhandwritten text. The results are promising with the proposed model achieving\nstate-of-the-art performance.\n"}
{"abstract": "  We present a determination of the ratio of kaon and pion leptonic decay\nconstants in isosymmetric QCD (isoQCD), $f_K / f_\\pi$, making use of the gauge\nensembles produced by the Extended Twisted Mass Collaboration (ETMC) with $N_f\n= 2 + 1 + 1$ flavors of Wilson-clover twisted-mass quarks, including\nconfigurations close to the physical point for all dynamical flavors. The\nsimulations are carried out at three values of the lattice spacing ranging from\n$\\sim 0.068$ to $\\sim 0.092$ fm with linear lattice size up to $L \\sim 5.5$~fm.\nThe scale is set by the PDG value of the pion decay constant, $f_\\pi^{isoQCD} =\n130.4~(2)$ MeV, at the isoQCD pion point, $M_\\pi^{isoQCD} = 135.0~(2)$ MeV,\nobtaining for the gradient-flow (GF) scales the values $w_0 = 0.17383~(63)$ fm,\n$\\sqrt{t_0} = 0.14436~(61)$ fm and $t_0 / w_0 = 0.11969~(62)$ fm. The data are\nanalyzed within the framework of SU(2) Chiral Perturbation Theory (ChPT)\nwithout resorting to the use of renormalized quark masses. At the isoQCD kaon\npoint $M_K^{isoQCD} = 494.2~(4)$ MeV we get $(f_K / f_\\pi)^{isoQCD} =\n1.1995~(44)$, where the error includes both statistical and systematic\nuncertainties. Implications for the Cabibbo-Kobayashi-Maskawa (CKM) matrix\nelement $|V_{us}|$ and for the first-row CKM unitarity are discussed.\n"}
{"abstract": "  The Mullineux map is a combinatorial function on partitions which describes\nthe effect of tensoring a simple module for the symmetric group in\ncharacteristic $p$ with the one-dimensional sign representation. It can also be\ninterpreted as an isomorphism between crystal graphs for\n$\\widehat{\\mathfrak{sl}}_p$. We give a new combinatorial description of the\nMullineux map by expressing this crystal isomorphism as a composition of\nisomorphisms between different crystals. These isomorphisms are defined in\nterms of new generalised regularisation maps introduced by Millan Berdasco.\n  We then given two applications of our new realisation of the Mullineux map,\nby providing purely combinatorial proofs of a conjecture of Lyle relating the\nMullineux map with regularisation, and a theorem of Paget describing the\nMullineux map in RoCK blocks of symmetric groups.\n"}
{"abstract": "  We present a network formulation for a traffic flow model with nonlocal\nvelocity in the flux function. The modeling framework includes suitable\ncoupling conditions at intersections to either ensure maximum flux or\ndistribution parameters. Based on an upwind type numerical scheme, we prove the\nmaximum principle and the existence of weak solutions on networks. We also\ninvestigate the limiting behavior of the proposed models when the nonlocal\ninfluence tends to infinity. Numerical examples show the difference between the\nproposed coupling conditions and a comparison to the Lighthill-Whitham-Richards\nnetwork model.\n"}
{"abstract": "  Knowledge Graph (KG) completion has been excessively studied with a massive\nnumber of models proposed for the Link Prediction (LP) task. The main\nlimitation of such models is their insensitivity to time. Indeed, the temporal\naspect of stored facts is often ignored. To this end, more and more works\nconsider time as a parameter to complete KGs. In this paper, we first\ndemonstrate that, by simply increasing the number of negative samples, the\nrecent AttH model can achieve competitive or even better performance than the\nstate-of-the-art on Temporal KGs (TKGs), albeit its nontemporality. We further\npropose Hercules, a time-aware extension of AttH model, which defines the\ncurvature of a Riemannian manifold as the product of both relation and time.\nOur experiments show that both Hercules and AttH achieve competitive or new\nstate-of-the-art performances on ICEWS04 and ICEWS05-15 datasets. Therefore,\none should raise awareness when learning TKGs representations to identify\nwhether time truly boosts performances.\n"}
{"abstract": "  Current work in explainable reinforcement learning generally produces\npolicies in the form of a decision tree over the state space. Such policies can\nbe used for formal safety verification, agent behavior prediction, and manual\ninspection of important features. However, existing approaches fit a decision\ntree after training or use a custom learning procedure which is not compatible\nwith new learning techniques, such as those which use neural networks. To\naddress this limitation, we propose a novel Markov Decision Process (MDP) type\nfor learning decision tree policies: Iterative Bounding MDPs (IBMDPs). An IBMDP\nis constructed around a base MDP so each IBMDP policy is guaranteed to\ncorrespond to a decision tree policy for the base MDP when using a\nmethod-agnostic masking procedure. Because of this decision tree equivalence,\nany function approximator can be used during training, including a neural\nnetwork, while yielding a decision tree policy for the base MDP. We present the\nrequired masking procedure as well as a modified value update step which allows\nIBMDPs to be solved using existing algorithms. We apply this procedure to\nproduce IBMDP variants of recent reinforcement learning methods. We empirically\nshow the benefits of our approach by solving IBMDPs to produce decision tree\npolicies for the base MDPs.\n"}
{"abstract": "  In the near future, robots are expected to work with humans or operate alone\nand may replace human workers in various fields such as homes and factories. In\na previous study, we proposed bilateral control-based imitation learning that\nenables robots to utilize force information and operate almost simultaneously\nwith an expert's demonstration. In addition, we recently proposed an\nautoregressive neural network model (SM2SM) for bilateral control-based\nimitation learning to obtain long-term inferences. In the SM2SM model, both\nmaster and slave states must be input, but the master states are obtained from\nthe previous outputs of the SM2SM model, resulting in destabilized estimation\nunder large environmental variations. Hence, a new autoregressive neural\nnetwork model (S2SM) is proposed in this study. This model requires only the\nslave state as input and its outputs are the next slave and master states,\nthereby improving the task success rates. In addition, a new feedback\ncontroller that utilizes the error between the responses and estimates of the\nslave is proposed, which shows better reproducibility.\n"}
{"abstract": "  The primary objective of this paper is to investigate the modified\nLeray-alpha equation on the two-dimensional sphere $\\mathbb{S}^2$, the square\ntorus $\\mathbb{T}^2$ and the three-torus $\\mathbb{T}^3$. In the strategy, we\nprove the existence and the uniqueness of the weak solutions and also the\nexistence of the global attractor for the equation. Then we establish the upper\nand lower bounds of the Hausdorff and fractal dimensions of the global\nattractor on both $\\mathbb{S}^2$ and $\\mathbb{T}^2$. Our method is based on the\nestimates for the vorticity scalar equations and the stationary solutions\naround the invariant manifold that are constructed by using the Kolmogorov\nflows. Finally, we will use the results on $\\mathbb{T}^2$ to study the lower\nbound for attractor's dimensions on the case of $\\mathbb{T}^3$.\n"}
{"abstract": "  We extend recent results on the exact hydrodynamics of a system of diffusive\nactive particles displaying a motility-induced phase separation to account for\ntypical fluctuations of the dynamical fields. By calculating correlation\nfunctions exactly in the homogeneous phase, we find that two macroscopic length\nscales develop in the system. The first is related to the diffusive length of\nthe particles and the other to the collective behavior of the particles. The\nlatter diverges as the critical point is approached. Our results show that the\ncritical behavior of the model in one dimension belongs to the universality\nclass of a mean-field Ising model, both for static and dynamic properties, when\nthe thermodynamic limit is taken in a specified manner. The results are\ncompared to the critical behavior exhibited by the ABC model. In particular, we\nfind that in contrast to the ABC model the density large deviation function, at\nits Gaussian approximation, does not contain algebraically decaying\ninteractions but is of a finite, macroscopic, extent which is dictated by the\ndiverging correlation length.\n"}
{"abstract": "  In this paper, we consider a closed Riemannian manifold $M^{n+1}$ with\ndimension $3\\leq n+1\\leq 7$, and a compact Lie group $G$ acting as isometries\non $M$ with cohomogeneity at least $3$. Suppose the union of non-principal\norbits $M\\setminus M^{reg}$ is a smooth embedded submanifold of $M$ without\nboundary and ${\\rm dim}(M\\setminus M^{reg})\\leq n-2 $. Then for any\n$c\\in\\mathbb{R}$, we show the existence of a nontrivial, smooth, closed,\n$G$-equivariant almost embedded $G$-invariant hypersurface $\\Sigma^n$ of\nconstant mean curvature $c$.\n"}
{"abstract": "  A comprehensive study of magnetocrystalline anisotropy of a layered\nvan-der-Waals ferromagnet VI3 was performed. We measured angular dependences of\nthe torque and magnetization with respect to the direction of the applied\nmagnetic field within the \"ac\" plane perpendicular to and within the basal ab\nplane, respectively. A two-fold butterfly-like signal was detected by\nmagnetization in the perpendicular \"ac\" plane. This signal symmetry remains\nconserved throughout all magnetic regimes as well as through the known\nstructural transition down to the lowest temperatures. The maximum of the\nmagnetization signal and the resulting magnetization easy axis is significantly\ntilted from the principal c axis by ~40{\\deg}. The close relation of the\nmagnetocrystalline anisotropy to the crystal structure was documented. In\ncontrast, a two-fold-like angular signal was detected in the paramagnetic\nregion within the ab plane in the monoclinic phase, which transforms into a\nsix-fold-like signal below the Curie temperature TC. With further cooling,\nanother six-fold-like signal with an angular shift of ~30{\\deg} grows\napproaching TFM. Below TFM, in the triclinic phase, the original six-fold-like\nsignal vanishes, being replaced by a secondary six-fold-like signal with an\nangular shift of ~30{\\deg}.\n"}
{"abstract": "  Context. The [CII] 158micron far-infrared fine-structure line is one of the\ndominant cooling lines of the star-forming interstellar medium (ISM). Hence\n[CII] emission originates in and thus can be used to trace a range of ISM\nprocesses. Velocity-resolved large-scale mapping of [CII] in star-forming\nregions provides a unique perspective of the kinematics of these regions and\ntheir interactions with the exciting source of radiation.\n  Aims. We explore the scientific applications of large-scale mapping of\nvelocity-resolved [CII] observations. With the [CII] observations, we\ninvestigate the effect of stellar feedback on the ISM. We present the details\nof observation, calibration, and data reduction using a heterodyne array\nreceiver mounted on an airborne observatory.\n  Results. A square-degree [CII] map with a spectral resolution of 0.3 km/s is\npresented. The scientific potential of this data is summarized with discussion\nof mechanical and radiative stellar feedback, filament tracing using [CII],\n[CII] opacity effects, [CII] and carbon recombination lines, and [CII]\ninteraction with the large molecular cloud. The data quality and calibration is\ndiscussed in detail, and new techniques are presented to mitigate the effects\nof unavoidable instrument deficiencies (e.g. baseline stability) and thus to\nimprove the data quality. A comparison with a smaller [CII] map taken with the\nHerschel/Heterodyne Instrument for the Far-Infrared (HIFI) spectrometer is\npresented.\n"}
{"abstract": "  This work aims to study the dynamic between research in the industry and\nacademia in computer vision. The results are demonstrated on a set of top-5\nvision conferences that are representative of the field. Since data for such\nanalysis was not readily available, significant effort was spent on gathering\nand processing meta-data from the original publications. First, this study\nquantifies the share of industry-sponsored research. Specifically, it shows\nthat the proportion of papers published by industry-affiliated researchers is\nincreasing and that more academics join companies or collaborate with them.\nNext, the possible impact of industry presence is further explored, namely in\nthe distribution of research topics and citation patterns. The results indicate\nthat the distribution of the research topics is similar in industry and\nacademic papers. However, there is a strong preference towards citing industry\npapers. Finally, possible reasons for citation bias, such as code availability\nand influence, are investigated.\n"}
{"abstract": "  Data pre-treatment plays a significant role in improving data quality, thus\nallowing extraction of accurate information from raw data. One of the data\npre-treatment techniques commonly used is outliers detection. The so-called\n3${\\sigma}$ method is a common practice to identify the outliers. As shown in\nthe manuscript, it does not identify all outliers, resulting in possible\ndistortion of the overall statistics of the data. This problem can have a\nsignificant impact on further data analysis and can lead to reduction in the\naccuracy of predictive models. There is a plethora of various techniques for\noutliers detection, however, aside from theoretical work, they all require case\nstudy work. Two types of outliers were considered: short-term (erroneous data,\nnoise) and long-term outliers (e.g. malfunctioning for longer periods). The\ndata used were taken from the vacuum distillation unit (VDU) of an Asian\nrefinery and included 40 physical sensors (temperature, pressure and flow\nrate). We used a modified method for 3${\\sigma}$ thresholds to identify the\nshort-term outliers, i.e. ensors data are divided into chunks determined by\nchange points and 3${\\sigma}$ thresholds are calculated within each chunk\nrepresenting near-normal distribution. We have shown that piecewise 3${\\sigma}$\nmethod offers a better approach to short-term outliers detection than\n3${\\sigma}$ method applied to the entire time series. Nevertheless, this does\nnot perform well for long-term outliers (which can represent another state in\nthe data). In this case, we used principal component analysis (PCA) with\nHotelling's $T^2$ statistics to identify the long-term outliers. The results\nobtained with PCA were subject to DBSCAN clustering method. The outliers (which\nwere visually obvious and correctly detected by the PCA method) were also\ncorrectly identified by DBSCAN which supported the consistency and accuracy of\nthe PCA method.\n"}
{"abstract": "  We study geometric set cover problems in dynamic settings, allowing\ninsertions and deletions of points and objects. We present the first dynamic\ndata structure that can maintain an $O(1)$-approximation in sublinear update\ntime for set cover for axis-aligned squares in 2D. More precisely, we obtain\nrandomized update time $O(n^{2/3+\\delta})$ for an arbitrarily small constant\n$\\delta>0$. Previously, a dynamic geometric set cover data structure with\nsublinear update time was known only for unit squares by Agarwal, Chang, Suri,\nXiao, and Xue [SoCG 2020]. If only an approximate size of the solution is\nneeded, then we can also obtain sublinear amortized update time for disks in 2D\nand halfspaces in 3D. As a byproduct, our techniques for dynamic set cover also\nyield an optimal randomized $O(n\\log n)$-time algorithm for static set cover\nfor 2D disks and 3D halfspaces, improving our earlier $O(n\\log n(\\log\\log\nn)^{O(1)})$ result [SoCG 2020].\n"}
{"abstract": "  Mutually unbiased measurements are a generalization of mutually unbiased\nbases in which the measurement operators need not to be rank one projectors. In\na $d$-dimension space, the purity of measurement elements ranges from $1/d$ for\nthe measurement operators corresponding to maximally mixed states to $1$ for\nthe rank one projectors. In this contribution, we provide a class of MUM that\nencompasses the full range of purity. Similar to the MUB in which the operators\ncorresponding to different outcomes of the same measurement commute mutually,\nour class of MUM possesses this sense of compatibility within each measurement.\nThis makes the provided class more similar to the MUB, so that the main\ndifference between them and MUB is due to the purity of the measurement\noperators. The spectra of these MUMs provides a way to construct a class of\n$d$-dimensional orthogonal matrices which leave the vector of equal components\ninvariant. Based on this property, and by using the MUM-based entanglement\nwitnesses, we investigate the role of purity to detect entanglement of\nbipartite states.\n"}
{"abstract": "  With graphs rapidly growing in size and deeper graph neural networks (GNNs)\nemerging, the training and inference of GNNs become increasingly expensive.\nExisting network weight pruning algorithms cannot address the main space and\ncomputational bottleneck in GNNs, caused by the size and connectivity of the\ngraph. To this end, this paper first presents a unified GNN sparsification\n(UGS) framework that simultaneously prunes the graph adjacency matrix and the\nmodel weights, for effectively accelerating GNN inference on large-scale\ngraphs. Leveraging this new tool, we further generalize the recently popular\nlottery ticket hypothesis to GNNs for the first time, by defining a graph\nlottery ticket (GLT) as a pair of core sub-dataset and sparse sub-network,\nwhich can be jointly identified from the original GNN and the full dense graph\nby iteratively applying UGS. Like its counterpart in convolutional neural\nnetworks, GLT can be trained in isolation to match the performance of training\nwith the full model and graph, and can be drawn from both randomly initialized\nand self-supervised pre-trained GNNs. Our proposal has been experimentally\nverified across various GNN architectures and diverse tasks, on both\nsmall-scale graph datasets (Cora, Citeseer and PubMed), and large-scale\ndatasets from the challenging Open Graph Benchmark (OGB). Specifically, for\nnode classification, our found GLTs achieve the same accuracies with 20%~98%\nMACs saving on small graphs and 25%~85% MACs saving on large ones. For link\nprediction, GLTs lead to 48%~97% and 70% MACs saving on small and large graph\ndatasets, respectively, without compromising predictive performance. Codes\navailable at https://github.com/VITA-Group/Unified-LTH-GNN.\n"}
{"abstract": "  Our goal in this work is to better understand the relationship between\nreplica symmetry breaking, shattering, and metastability. To this end, we study\nthe static and dynamic behaviour of spherical pure $p$-spin glasses above the\nreplica symmetry breaking temperature $T_{s}$. In this regime, we find that\nthere are at least two distinct temperatures related to non-trivial behaviour.\nFirst we prove that there is a regime of temperatures in which the spherical\n$p$-spin model exhibits a shattering phase. Our results holds in a regime above\nbut near $T_s$. We then find that metastable states exist up to an even higher\ntemperature $T_{BBM}$ as predicted by Barrat--Burioni--M\\'ezard which is\nexpected to be higher than the phase boundary for the shattering phase $T_d\n<T_{BBM}$. We develop this work by first developing a\nThouless--Anderson--Palmer decomposition which builds on the work of Subag. We\nthen present a series of questions and conjectures regarding the sharp phase\nboundaries for shattering and slow mixing.\n"}
{"abstract": "  We propose a method for building a squeezed vacuum state laser with zero\ndiffusion, which results from the introduction of the reservoir engineering\ntechnique into the laser theory. As well as the reservoir engineering, our\nsqueezed vacuum laser demands the construction of an effective atom-field\ninteraction. And by building an isomorphism between the cavity field operators\nin the effective and the Jaynes-Cummings Hamiltonians, we derive the equations\nof our effective laser directly from the conventional laser theory. Our method,\nwhich is less susceptible to errors than reservoir engineering, can be extended\nfor the construction of other nonclassical state lasers, and our squeezed\nvacuum laser can contribute to the newly emerging field of gravitational\ninterferometry.\n"}
{"abstract": "  Trust in human-robot interactions (HRI) is measured in two main ways: through\nsubjective questionnaires and through behavioral tasks. To optimize\nmeasurements of trust through questionnaires, the field of HRI faces two\nchallenges: the development of standardized measures that apply to a variety of\nrobots with different capabilities, and the exploration of social and\nrelational dimensions of trust in robots (e.g., benevolence). In this paper we\nlook at how different trust questionnaires fare given these challenges that\npull in different directions (being general vs. being exploratory) by studying\nwhether people think the items in these questionnaires are applicable to\ndifferent kinds of robots and interactions. In Study 1 we show that after being\npresented with a robot (non-humanoid) and an interaction scenario (fire\nevacuation), participants rated multiple questionnaire items such as \"This\nrobot is principled\" as \"Non-applicable to robots in general\" or\n\"Non-applicable to this robot\". In Study 2 we show that the frequency of these\nratings change (indeed, even for items rated as N/A to robots in general) when\na new scenario is presented (game playing with a humanoid robot). Finally,\nwhile overall trust scores remained robust to N/A ratings, our results revealed\npotential fallacies in the way these scores are commonly interpreted. We\nconclude with recommendations for the development, use and results-reporting of\ntrust questionnaires for future studies, as well as theoretical implications\nfor the field of HRI.\n"}
{"abstract": "  We demonstrate that the general model of a linearly time-dependent crossing\nof two energy bands is integrable. Namely, the Hamiltonian of this model has a\nquadratically time-dependent commuting operator. We apply this property to\nfour-state Landau-Zener (LZ) models that have previously been used to describe\nthe Landau-St\\\"uckelberg interferometry experiments with an electron shuttling\nbetween two semiconductor quantum dots. The integrability then leads to simple\nbut nontrivial exact relations for the transition probabilities. In addition,\nthe integrability leads to a semiclassical theory that provides analytical\napproximation for the transition probabilities in these models for all\nparameter values. The results predict a dynamic phase transition, and show that\nsimilarly-looking models belong to different topological classes.\n"}
{"abstract": "  In this paper, we consider the optimal control problem for a class of\nevolution inclusions with Volterra type operators, which can be\nhistory-dependent. We establish the existence of a solution to the stated\noptimal control problem under some hypothesis on data-in. The motivation for\nthis work comes from the optimal control problems for variational inequalities\narising in the study of injection molding processes, contact mechanics,\nprinciples of electro-wetting on dielectric, and others.\n"}
{"abstract": "  Over the past decade, wind energy has gained more attention in the world.\nHowever, owing to its indirectness and volatility properties, wind power\npenetration has increased the difficulty and complexity in dispatching and\nplanning of electric power systems. Therefore, it is needed to make the\nhigh-precision wind power prediction in order to balance the electrical power.\nFor this purpose, in this study, the prediction performance of linear\nregression, k-nearest neighbor regression and decision tree regression\nalgorithms is compared in detail. k-nearest neighbor regression algorithm\nprovides lower coefficient of determination values, while decision tree\nregression algorithm produces lower mean absolute error values. In addition,\nthe meteorological parameters of wind speed, wind direction, barometric\npressure and air temperature are evaluated in terms of their importance on the\nwind power parameter. The biggest importance factor is achieved by wind speed\nparameter. In consequence, many useful assessments are made for wind power\npredictions.\n"}
{"abstract": "  The manifold Helmholtzian (1-Laplacian) operator $\\Delta_1$ elegantly\ngeneralizes the Laplace-Beltrami operator to vector fields on a manifold\n$\\mathcal M$. In this work, we propose the estimation of the manifold\nHelmholtzian from point cloud data by a weighted 1-Laplacian $\\mathbf{\\mathcal\nL}_1$. While higher order Laplacians ave been introduced and studied, this work\nis the first to present a graph Helmholtzian constructed from a simplicial\ncomplex as an estimator for the continuous operator in a non-parametric\nsetting. Equipped with the geometric and topological information about\n$\\mathcal M$, the Helmholtzian is a useful tool for the analysis of flows and\nvector fields on $\\mathcal M$ via the Helmholtz-Hodge theorem. In addition, the\n$\\mathbf{\\mathcal L}_1$ allows the smoothing, prediction, and feature\nextraction of the flows. We demonstrate these possibilities on substantial sets\nof synthetic and real point cloud datasets with non-trivial topological\nstructures; and provide theoretical results on the limit of $\\mathbf{\\mathcal\nL}_1$ to $\\Delta_1$.\n"}
{"abstract": "  It has been well known for a long time that the height function of random\nlozenge tilings of large domains follow a law of large number and possible\nlimits called dimer limit shapes are well understood. For the next order, it is\nexpected that fluctuations behave like version of a Gaussian Free field, at\nleast away from some special \"frozen\" regions. However despite being one of the\nmain questions in the domain for 20 years, only special cases have been\nobtained. In this paper we show that for any specified limit shape with no\nfrozen region, one can construct a sequence of domains whose height functions\nconverge to that limit shape and where the height fluctuation converge to a\nvariant of the Gaussian Free Field.\n"}
{"abstract": "  The widespread application of deep neural network (DNN) techniques is being\nchallenged by adversarial examples, the legitimate input added with\nimperceptible and well-designed perturbations that can fool DNNs easily in the\nDNN testing/deploying stage. Previous adversarial example generation algorithms\nfor adversarial white-box attacks used Jacobian gradient information to add\nperturbations. This information is too imprecise and inexplicit, which will\ncause unnecessary perturbations when generating adversarial examples. This\npaper aims to address this issue. We first propose to apply a more informative\nand distilled gradient information, namely integrated gradient, to generate\nadversarial examples. To further make the perturbations more imperceptible, we\npropose to employ the restriction combination of $L_0$ and $L_1/L_2$ secondly,\nwhich can restrict the total perturbations and perturbation points\nsimultaneously. Meanwhile, to address the non-differentiable problem of $L_1$,\nwe explore a proximal operation of $L_1$ thirdly. Based on these three works,\nwe propose two Integrated gradient based White-box Adversarial example\ngeneration algorithms (IWA): IFPA and IUA. IFPA is suitable for situations\nwhere there are a determined number of points to be perturbed. IUA is suitable\nfor situations where no perturbation point number is preset in order to obtain\nmore adversarial examples. We verify the effectiveness of the proposed\nalgorithms on both structured and unstructured datasets, and we compare them\nwith five baseline generation algorithms. The results show that our proposed\nalgorithms do craft adversarial examples with more imperceptible perturbations\nand satisfactory crafting rate. $L_2$ restriction is more suitable for\nunstructured dataset and $L_1$ restriction performs better in structured\ndataset.\n"}
{"abstract": "  The Vehicle Routing Problem (VRP) is one of the most intensively studied\ncombinatorial optimisation problems for which numerous models and algorithms\nhave been proposed. To tackle the complexities, uncertainties and dynamics\ninvolved in real-world VRP applications, Machine Learning (ML) methods have\nbeen used in combination with analytical approaches to enhance problem\nformulations and algorithmic performance across different problem solving\nscenarios. However, the relevant papers are scattered in several traditional\nresearch fields with very different, sometimes confusing, terminologies. This\npaper presents a first, comprehensive review of hybrid methods that combine\nanalytical techniques with ML tools in addressing VRP problems. Specifically,\nwe review the emerging research streams on ML-assisted VRP modelling and\nML-assisted VRP optimisation. We conclude that ML can be beneficial in\nenhancing VRP modelling, and improving the performance of algorithms for both\nonline and offline VRP optimisations. Finally, challenges and future\nopportunities of VRP research are discussed.\n"}
{"abstract": "  Scaling and structural evolutions are contemplated in a new perspective for\nturbulent channel flows. The total integrated turbulence kinetic energy remains\nconstant when normalized by the friction velocity squared, while the total\ndissipation increases linearly with respect to the Reynolds number. This serves\nas a global constraint on the turbulence structure. Motivated by the flux\nbalances in the root turbulence variables, we also discover dissipative scaling\nfor u2 and v2, respectively through its first and second gradients. This\nself-similarity allows for profile reconstructions at any Reynolds numbers\nbased on a common template, through a simple multiplicative operation. Using\nthese scaled variables in the Lagrangian transport equation derives the\nReynolds shear stress, which in turn computes the mean velocity profile. The\nself-similarities along with the transport equations render possible succinct\nviews of the turbulence dynamics and computability of the full structure in\nchannel flows.\n"}
{"abstract": "  In general, most of the substances in nature exist in mixtures, and the\nnoninvasive identification of mixture composition with high speed and accuracy\nremains a difficult task. However, the development of Raman spectroscopy,\nmachine learning, and deep learning techniques have paved the way for achieving\nefficient analytical tools capable of identifying mixture components, making an\napparent breakthrough in the identification of mixtures beyond the traditional\nchemical analysis methods. This article summarizes the work of Raman\nspectroscopy in identifying the composition of substances as well as provides\ndetailed reviews on the preprocessing process of Raman spectroscopy, the\nanalysis methods and applications of artificial intelligence. This review\nsummarizes the work of Raman spectroscopy in identifying the composition of\nsubstances and reviews the preprocessing process of Raman spectroscopy, the\nanalysis methods and applications of artificial intelligence. Finally, the\nadvantages and disadvantages and development prospects of Raman spectroscopy\nare discussed in detail.\n"}
{"abstract": "  Deep Neural Network (DNN) frameworks use distributed training to enable\nfaster time to convergence and alleviate memory capacity limitations when\ntraining large models and/or using high dimension inputs. With the steady\nincrease in datasets and model sizes, model/hybrid parallelism is deemed to\nhave an important role in the future of distributed training of DNNs. We\nanalyze the compute, communication, and memory requirements of Convolutional\nNeural Networks (CNNs) to understand the trade-offs between different\nparallelism approaches on performance and scalability. We leverage our\nmodel-driven analysis to be the basis for an oracle utility which can help in\ndetecting the limitations and bottlenecks of different parallelism approaches\nat scale. We evaluate the oracle on six parallelization strategies, with four\nCNN models and multiple datasets (2D and 3D), on up to 1024 GPUs. The results\ndemonstrate that the oracle has an average accuracy of about 86.74% when\ncompared to empirical results, and as high as 97.57% for data parallelism.\n"}
{"abstract": "  We present EDIpack, an exact diagonalization package to solve generic quantum\nimpurity problems. The algorithm includes a generalization of the look-up\nmethod introduced in [Lin, Gubernatis Comput. Phys., 7 (4) (1993), 400] and\nenables a massively parallel execution of the matrix-vector linear operations\nrequired by Lanczos and Arnoldi algorithms. We show that a suitable Fock basis\norganization is crucial to optimize the inter-processors communication in a\ndistributed memory setup and to reach sub-linear scaling in sufficiently large\nsystems. We discuss the algorithm in details indicating how to deal with\nmultiple orbitals and electron-phonon coupling. Finally, we outline the\ndownload, installation and functioning of the package.\n"}
{"abstract": "  Using the notions of Topological dynamics, H. Furstenberg defined central\nsets and proved the Central Sets Theorem. Later V. Bergelson and N. Hindman\ncharacterized central sets in terms of algebra of the Stone-\\v{C}ech\nCompactification of discrete semigroup. They found that central sets are the\nmembers of the minimal idempotents of $\\beta S$, the Stone-\\v{C}ech\nCompactification of a semigroup $\\left(S,\\cdot\\right)$. We know that any closed\nsubsemigroup of $\\beta S$ is generated by a filter. We call a set $A$ to be a\n$\\mathcal{F}$-central set if it is a member of a minimal idempotent of a closed\nsubsemigroup of $\\beta S$, generated by the filter $\\mathcal{F}$. In this\narticle we will characterize the $\\mathcal{F}$-central sets dynamically.\n"}
{"abstract": "  To understand the behavior of large dynamical systems like transportation\nnetworks, one must often rely on measurements transmitted by a set of sensors,\nfor instance individual vehicles. Such measurements are likely to be incomplete\nand imprecise, which makes it hard to recover the underlying signal of\ninterest.Hoping to quantify this phenomenon, we study the properties of a\npartially-observed state-space model. In our setting, the latent state $X$\nfollows a high-dimensional Vector AutoRegressive process $X_t = \\theta X_{t-1}\n+ \\varepsilon_t$. Meanwhile, the observations $Y$ are given by a\nnoise-corrupted random sample from the state $Y_t = \\Pi_t X_t + \\eta_t$.\nSeveral random sampling mechanisms are studied, allowing us to investigate the\neffect of spatial and temporal correlations in the distribution of the sampling\nmatrices $\\Pi_t$.We first prove a lower bound on the minimax estimation error\nfor the transition matrix $\\theta$. We then describe a sparse estimator based\non the Dantzig selector and upper bound its non-asymptotic error, showing that\nit achieves the optimal convergence rate for most of our sampling mechanisms.\nNumerical experiments on simulated time series validate our theoretical\nfindings, while an application to open railway data highlights the relevance of\nthis model for public transport traffic analysis.\n"}
{"abstract": "  Introducing an interpolation method we estimate the spectral gap for Brownian\nmotion on general domains with sticky-reflecting boundary diffusion associated\nto the first nontrivial eigenvalue for the Laplace operator with corresponding\nWentzell-type boundary condition. In the manifold case our proofs involve novel\napplications of the celebrated Reilly formula.\n"}
{"abstract": "  Many characteristics of dwarf carbon stars are broadly consistent with a\nbinary origin, including mass transfer from an evolved companion. While the\npopulation overall appears to have old-disc or halo kinematics, roughly\n2$\\,$per cent of these stars exhibit H$\\alpha$ emission, which in low-mass\nmain-sequence stars is generally associated with rotation and relative youth.\nIts presence in an older population therefore suggests either irradiation or\nspin-up. This study presents time-series analyses of photometric and\nradial-velocity data for seven dwarf carbon stars with H$\\alpha$ emission. All\nare shown to have photometric periods in the range 0.2--5.2$\\,$d, and orbital\nperiods of similar length, consistent with tidal synchronisation. It is\nhypothesised that dwarf carbon stars with emission lines are the result of\nclose-binary evolution, indicating that low-mass, metal-weak or metal-poor\nstars can accrete substantial material prior to entering a common-envelope\nphase.\n"}
{"abstract": "  Many super-resolution (SR) models are optimized for high performance only and\ntherefore lack efficiency due to large model complexity. As large models are\noften not practical in real-world applications, we investigate and propose\nnovel loss functions, to enable SR with high perceptual quality from much more\nefficient models. The representative power for a given low-complexity generator\nnetwork can only be fully leveraged by strong guidance towards the optimal set\nof parameters. We show that it is possible to improve the performance of a\nrecently introduced efficient generator architecture solely with the\napplication of our proposed loss functions. In particular, we use a Fourier\nspace supervision loss for improved restoration of missing high-frequency (HF)\ncontent from the ground truth image and design a discriminator architecture\nworking directly in the Fourier domain to better match the target HF\ndistribution. We show that our losses' direct emphasis on the frequencies in\nFourier-space significantly boosts the perceptual image quality, while at the\nsame time retaining high restoration quality in comparison to previously\nproposed loss functions for this task. The performance is further improved by\nutilizing a combination of spatial and frequency domain losses, as both\nrepresentations provide complementary information during training. On top of\nthat, the trained generator achieves comparable results with and is 2.4x and\n48x faster than state-of-the-art perceptual SR methods RankSRGAN and SRFlow\nrespectively.\n"}
{"abstract": "  In this article, we investigate Maximum Likelihood Estimation with tools from\nTropical Geometry and Bernstein--Sato theory. We investigate the critical\npoints of very affine varieties and study their asymptotic behavior. We relate\nthese asymptotics to particular rays in the tropical variety as well as to\nBernstein--Sato ideals and give a connection to Maximum Likelihood Estimation\nin Statistics.\n"}
{"abstract": "  We introduce a novel method for discerning optical telescope images of stars\nfrom those of galaxies using Gaussian processes (GPs). Although applications of\nGPs often struggle in high-dimensional data modalities such as optical image\nclassification, we show that a low-dimensional embedding of images into a\nmetric space defined by the principal components of the data suffices to\nproduce high-quality predictions from real large-scale survey data. We develop\na novel method of GP classification hyperparameter training that scales\napproximately linearly in the number of image observations, which allows for\napplication of GP models to large-size Hyper Suprime-Cam (HSC) Subaru Strategic\nProgram data. In our experiments we evaluate the performance of a principal\ncomponent analysis (PCA) embedded GP predictive model against other machine\nlearning algorithms including a convolutional neural network and an image\nphotometric morphology discriminator. Our analysis shows that our methods\ncompare favorably with current methods in optical image classification while\nproducing posterior distributions from the GP regression that can be used to\nquantify object classification uncertainty. We further describe how\nclassification uncertainty can be used to efficiently parse large-scale survey\nimaging data to produce high-confidence object catalogs.\n"}
{"abstract": "  In this work we investigate greedy energy sequences on the unit circle for\nthe logarithmic and Riesz potentials. By definition, if $(a_n)_{n=0}^{\\infty}$\nis a greedy $s$-energy sequence on the unit circle, the Riesz potential\n$U_{N,s}(x):=\\sum_{k=0}^{N-1}|a_k-x|^{-s}$, $s>0$, generated by the first $N$\npoints of the sequence attains its minimum value at the point $a_{N}$, for\nevery $N\\geq 1$. In the case $s=0$ we minimize instead the logarithmic\npotential $U_{N,0}(x):=-\\sum_{k=0}^{N-1}\\log |a_{k}-x|$. We analyze the\nasymptotic properties of these extremal values $U_{N,s}(a_N)$, studying\nseparately the cases $s=0$, $0<s<1$, $s=1$, and $s>1$. We obtain second-order\nasymptotic formulas for $U_{N,s}(a_N)$ in the cases $s=0$, $0<s<1$, and $s=1$\n(the corresponding first-order formulas are well known). A first-order result\nfor $s>1$ is proved, and it is shown that the normalized sequence\n$U_{N,s}(a_N)/N^s$ is bounded and divergent in this case. We also consider,\nbriefly, greedy energy sequences in which the minimization condition is\nrequired starting from the point $a_{p+1}$ (instead of the point $a_{1}$ as\npreviously stated), for some $p\\geq 1$. For this more general class of greedy\nsequences, we prove a first-order asymptotic result for $0\\leq s<1$.\n"}
{"abstract": "  We present here a self-consistent cosmological zoom-in simulation of a triple\nsupermassive black hole (SMBH) system forming in a complex multiple galaxy\nmerger. The simulation is run with an updated version of our code KETJU, which\nis able to follow the motion of SMBHs down to separations of tens of\nSchwarzschild radii while simultaneously modeling the large-scale astrophysical\nprocesses in the surrounding galaxies, such as gas cooling, star formation, and\nstellar and AGN feedback. Our simulation produces initially a SMBH binary\nsystem for which the hardening process is interrupted by the late arrival of a\nthird SMBH. The KETJU code is able to accurately model the complex behavior\noccurring in such a triple SMBH system, including the ejection of one SMBH to a\nkiloparsec-scale orbit in the galaxy due to strong three-body interactions as\nwell as Lidov-Kozai oscillations suppressed by relativistic precession when the\nSMBHs are in a hierarchical configuration. One pair of SMBHs merges $\\sim\n3\\,\\mathrm{Gyr}$ after the initial galaxy merger, while the remaining binary is\nat a parsec-scale separation when the simulation ends at redshift $z=0$. We\nalso show that KETJU can capture the effects of the SMBH binaries and triplets\non the surrounding stellar population, which can affect the binary merger\ntimescales as the stellar density in the system evolves. Our results\ndemonstrate the importance of dynamically resolving the complex behavior of\nmultiple SMBHs in galactic mergers, as such systems cannot be readily modeled\nusing simple orbit-averaged semi-analytic models.\n"}
{"abstract": "  Distribution grid agents are obliged to exchange and disclose their states\nexplicitly to neighboring regions to enable distributed optimal power flow\ndispatch. However, the states contain sensitive information of individual\nagents, such as voltage and current measurements. These measurements can be\ninferred by adversaries, such as other participating agents or eavesdroppers.\nTo address the issue, we propose a privacy-preserving distributed optimal power\nflow (OPF) algorithm based on partially homomorphic encryption (PHE). First of\nall, we exploit the alternating direction method of multipliers (ADMM) to solve\nthe OPF in a distributed fashion. In this way, the dual update of ADMM can be\nencrypted by PHE. We further relax the augmented term of the primal update of\nADMM with the $\\ell_1$-norm regularization. In addition, we transform the\nrelaxed ADMM with the $\\ell_1$-norm regularization to a semidefinite program\n(SDP), and prove that this transformation is exact. The SDP can be solved\nlocally with only the sign messages from neighboring agents, which preserves\nthe privacy of the primal update. At last, we strictly prove the privacy\npreservation guarantee of the proposed algorithm. Numerical case studies\nvalidate the effectiveness and exactness of the proposed approach.\n"}
{"abstract": "  Remote state preparation (RSP) provides a useful way of transferring quantum\ninformation between two distant nodes based on the previously shared\nentanglement. In this paper, we study RSP of an arbitrary single-photon state\nin two degrees of freedom (DoFs). Using hyper-entanglement as a shared\nresource, our first goal is to remotely prepare the single-photon state in\npolarization and frequency DoFs and the second one is to reconstruct the\nsingle-photon state in polarization and time-bin DoFs. In the RSP process, the\nsender will rotate the quantum state in each DoF of the photon according to the\nknowledge of the state to be communicated. By performing a projective\nmeasurement on the polarization of the sender's photon, the original\nsingle-photon state in two DoFs can be remotely reconstructed at the receiver's\nquantum systems. This work demonstrates a novel capability for long-distance\nquantum communication.\n"}
{"abstract": "  In this present work, we have obtained a singularity-free spherically\nsymmetric stellar model with anisotropic pressure in the background of\nEinstein's general theory of relativity. The Einstein's field equations have\nbeen solved by exploiting Tolman {\\em ansatz} [Richard C Tolman, Phys. Rev.\n55:364, 1939] in $(3+1)$-dimensional space-time. Using observed values of mass\nand radius of the compact star PSR J1903+327, we have calculated the numerical\nvalues of all the constants from the boundary conditions. All the physical\ncharacteristics of the proposed model have been discussed both analytically and\ngraphically. The new exact solution satisfies all the physical criteria for a\nrealistic compact star. The matter variables are regular and well behaved\nthroughout the stellar structure. Constraints on model parameters have been\nobtained. All the energy conditions are verified with the help of graphical\nrepresentation. The stability condition of the present model has been described\nthrough different testings.\n"}
{"abstract": "  This paper describes the system submitted by our team, KBCNMUJAL, for Task 2\nof the shared task Hate Speech and Offensive Content Identification in\nIndo-European Languages (HASOC), at Forum for Information Retrieval Evaluation,\nDecember 16-20, 2020, Hyderabad, India. The datasets of two Dravidian languages\nViz. Malayalam and Tamil of size 4000 observations, each were shared by the\nHASOC organizers. These datasets are used to train the machine using different\nmachine learning algorithms, based on classification and regression models. The\ndatasets consist of tweets or YouTube comments with two class labels offensive\nand not offensive. The machine is trained to classify such social media\nmessages in these two categories. Appropriate n-gram feature sets are extracted\nto learn the specific characteristics of the Hate Speech text messages. These\nfeature models are based on TFIDF weights of n-gram. The referred work and\nrespective experiments show that the features such as word, character and\ncombined model of word and character n-grams could be used to identify the term\npatterns of offensive text contents. As a part of the HASOC shared task, the\ntest data sets are made available by the HASOC track organizers. The best\nperforming classification models developed for both languages are applied on\ntest datasets. The model which gives the highest accuracy result on training\ndataset for Malayalam language was experimented to predict the categories of\nrespective test data. This system has obtained an F1 score of 0.77. Similarly\nthe best performing model for Tamil language has obtained an F1 score of 0.87.\nThis work has received 2nd and 3rd rank in this shared Task 2 for Malayalam and\nTamil language respectively. The proposed system is named HASOC_kbcnmujal.\n"}
{"abstract": "  The use of external background knowledge can be beneficial for the task of\nmatching schemas or ontologies automatically. In this paper, we exploit six\ngeneral-purpose knowledge graphs as sources of background knowledge for the\nmatching task. The background sources are evaluated by applying three different\nexploitation strategies. We find that explicit strategies still outperform\nlatent ones and that the choice of the strategy has a greater impact on the\nfinal alignment than the actual background dataset on which the strategy is\napplied. While we could not identify a universally superior resource, BabelNet\nachieved consistently good results. Our best matcher configuration with\nBabelNet performs very competitively when compared to other matching systems\neven though no dataset-specific optimizations were made.\n"}
{"abstract": "  The recent development of metasurfaces has motivated their potential use for\nimproving the performance of wireless communication networks by manipulating\nthe propagation environment through nearly-passive sub-wavelength scattering\nelements arranged on a surface. However, most studies of this technology focus\non reflective metasurfaces, i.e., the surface reflects the incident signals\ntowards receivers located on the same side of the transmitter, which restricts\nthe coverage to one side of the surface. In this article, we introduce the\nconcept of intelligent omni-surface (IOS), which is able to serve mobile users\non both sides of the surface to achieve full-dimensional communications by\njointly engineering its reflective and refractive properties. The working\nprinciple of the IOS is introduced and a novel hybrid beamforming scheme is\nproposed for IOS-based wireless communications. Moreover, we present a\nprototype of IOS-based wireless communications and report experimental results.\nFurthermore, potential applications of the IOS to wireless communications\ntogether with relevant research challenges are discussed.\n"}
{"abstract": "  Diverse domains of science and engineering use parameterised mechanistic\nmodels. Engineers and scientists can often hypothesise several rival models to\nexplain a specific process or phenomenon. Consider a model discrimination\nsetting where we wish to find the best mechanistic, dynamic model candidate and\nthe best model parameter estimates. Typically, several rival mechanistic models\ncan explain the available data, so design of dynamic experiments for model\ndiscrimination helps optimally collect additional data by finding experimental\nsettings that maximise model prediction divergence. We argue there are two main\napproaches in the literature for solving the optimal design problem: (i) the\nanalytical approach, using linear and Gaussian approximations to find\nclosed-form expressions for the design objective, and (ii) the data-driven\napproach, which often relies on computationally intensive Monte Carlo\ntechniques. Olofsson et al. (ICML 35, 2018) introduced Gaussian process (GP)\nsurrogate models to hybridise the analytical and data-driven approaches, which\nallowed for computationally efficient design of experiments for discriminating\nbetween black-box models. In this study, we demonstrate that we can extend\nexisting methods for optimal design of dynamic experiments to incorporate a\nwider range of problem uncertainty. We also extend the Olofsson et al. (2018)\nmethod of using GP surrogate models for discriminating between dynamic\nblack-box models. We evaluate our approach on a well-known case study from\nliterature, and explore the consequences of using GP surrogates to approximate\ngradient-based methods.\n"}
{"abstract": "  Machine learning has achieved much success on supervised learning tasks with\nlarge sets of well-annotated training samples. However, in many practical\nsituations, such strong and high-quality supervision provided by training data\nis unavailable due to the expensive and labor-intensive labeling process.\nAutomatically identifying and recognizing object categories in a large volume\nof unlabeled images with weak supervision remains an important, yet unsolved\nchallenge in computer vision. In this paper, we propose a novel machine\nlearning framework, artificial perceptual learning (APL), to tackle the problem\nof weakly supervised image categorization. The proposed APL framework is\nconstructed using state-of-the-art machine learning algorithms as building\nblocks to mimic the cognitive development process known as infant\ncategorization. We develop and illustrate the proposed framework by\nimplementing a wide-field fine-grain ecological survey of tree species over an\n8,000-hectare area of the El Yunque rainforest in Puerto Rico. It is based on\nunlabeled high-resolution aerial images of the tree canopy. Misplaced\nground-based labels were available for less than 1% of these images, which\nserve as the only weak supervision for this learning framework. We validate the\nproposed framework using a small set of images with high quality human\nannotations and show that the proposed framework attains human-level cognitive\neconomy.\n"}
{"abstract": "  Lee waves generated by stratified flow over rough bottom topography in the\nocean extract momentum and energy from the geostrophic flow, causing drag and\nenhancing turbulence and mixing in the interior ocean when they break. Inviscid\nlinear theory is generally used to predict the generation rate of lee waves,\nbut the location and mechanism of wave breaking leading to eventual dissipation\nof energy and irreversible mixing are poorly constrained. In this study, a\nlinear model with viscosity, diffusivity, and an upper boundary is used to\ndemonstrate the potential importance of the surface in reflecting lee wave\nenergy back into the interior, making the case for treating lee waves as a full\nwater column process. In the absence of critical levels, it is shown that lee\nwaves can be expected to interact with the upper ocean, resulting in enhanced\nvertical velocities and dissipation and mixing near the surface. The impact of\nthe typical oceanic conditions of increasing background velocity and\nstratification with height above bottom are investigated and shown to\ncontribute to enhanced upper ocean vertical velocities and mixing.\n"}
{"abstract": "  Recently, several weakly supervised learning methods have been devoted to\nutilize bounding box supervision for training deep semantic segmentation\nmodels. Most existing methods usually leverage the generic proposal generators\n(\\eg, dense CRF and MCG) to produce enhanced segmentation masks for further\ntraining segmentation models. These proposal generators, however, are generic\nand not specifically designed for box-supervised semantic segmentation, thereby\nleaving some leeway for improving segmentation performance. In this paper, we\naim at seeking for a more accurate learning-based class-agnostic pseudo mask\ngenerator tailored to box-supervised semantic segmentation. To this end, we\nresort to a pixel-level annotated auxiliary dataset where the class labels are\nnon-overlapped with those of the box-annotated dataset. For learning pseudo\nmask generator from the auxiliary dataset, we present a bi-level optimization\nformulation. In particular, the lower subproblem is used to learn\nbox-supervised semantic segmentation, while the upper subproblem is used to\nlearn an optimal class-agnostic pseudo mask generator. The learned pseudo\nsegmentation mask generator can then be deployed to the box-annotated dataset\nfor improving weakly supervised semantic segmentation. Experiments on PASCAL\nVOC 2012 dataset show that the learned pseudo mask generator is effective in\nboosting segmentation performance, and our method can further close the\nperformance gap between box-supervised and fully-supervised models. Our code\nwill be made publicly available at\nhttps://github.com/Vious/LPG_BBox_Segmentation .\n"}
{"abstract": "  As interest in autonomous driving increases, efforts are being made to meet\nrequirements for the high-level automation of vehicles. In this context, the\nfunctionality inside the vehicle cabin plays a key role in ensuring a safe and\npleasant journey for driver and passenger alike. At the same time, recent\nadvances in the field of artificial intelligence (AI) have enabled a whole\nrange of new applications and assistance systems to solve automated problems in\nthe vehicle cabin. This paper presents a thorough survey on existing work that\nutilizes AI methods for use-cases inside the driving cabin, focusing, in\nparticular, on application scenarios related to (1) driving safety and (2)\ndriving comfort. Results from the surveyed works show that AI technology has a\npromising future in tackling in-cabin tasks within the autonomous driving\naspect.\n"}
{"abstract": "  The widespread adoption of nonlinear Receding Horizon Control (RHC)\nstrategies by industry has led to more than 30 years of intense research\nefforts to provide stability guarantees for these methods. However, current\ntheoretical guarantees require that each (generally nonconvex) planning problem\ncan be solved to (approximate) global optimality, which is an unrealistic\nrequirement for the derivative-based local optimization methods generally used\nin practical implementations of RHC. This paper takes the first step towards\nunderstanding stability guarantees for nonlinear RHC when the inner planning\nproblem is solved to first-order stationary points, but not necessarily global\noptima. Special attention is given to feedback linearizable systems, and a\nmixture of positive and negative results are provided. We establish that, under\ncertain strong conditions, first-order solutions to RHC exponentially stabilize\nlinearizable systems. Crucially, this guarantee requires that state costs\napplied to the planning problems are in a certain sense `compatible' with the\nglobal geometry of the system, and a simple counter-example demonstrates the\nnecessity of this condition. These results highlight the need to rethink the\nrole of global geometry in the context of optimization-based control.\n"}
{"abstract": "  The third annual International Applied Category Theory Conference (ACT2020)\nwas planned to take place at MIT in Cambridge, Massachusetts USA. However, the\nglobal COVID-19 pandemic made the prospect of holding a large in-person meeting\nimpossible, and the event was thus held completely online. Holding the talks\nonline had the new benefits of reducing carbon footprint, being inclusive of\npeople from more parts of the world, and producing higher-quality video talks,\nwhich have been posted online for posterity.\n  The ACT2020 contributions spanned a broad spectrum of application areas,\nincluding databases, dynamical systems, functional programming, game theory,\nlenses, neuroscience, probabilistic programming, natural language processing,\nquantum mechanics, and cyberphysical systems. Papers featured a broad range of\ncategorical techniques.\n  Papers in this Proceedings volume represents about half of the talks\npresented at ACT2020. Being included in the proceedings vs. not is not an\nindication of talk quality, but instead almost exclusively the choice of the\nauthors, e.g. to present work already published elsewhere.\n"}
{"abstract": "  X-ray emission generated through solar-wind charge exchange (SWCX) is known\nto contaminate X-ray observation data, the amount of which is often significant\nor even dominant, particularly in the soft X-ray band, when the main target is\ncomparatively weak diffuse sources, depending on the space weather during the\nobservation. In particular, SWCX events caused by interplanetary coronal mass\nejections (ICMEs) tend to be spectrally rich and to provide critical\ninformation about the metal abundance in the ICME plasma. We analyzed the\nSN1006 background data observed with Suzaku on 2005 September 11 shortly after\nan X6-class solar flare, signatures of which were separately detected together\nwith an associated ICME. We found that the data include emission lines from a\nvariety of highly ionized ions generated through SWCX. The relative abundances\nof the detected ions were found to be consistent with those in past ICME-driven\nSWCX events. Thus, we conclude that this event was ICME-driven. In addition, we\ndetected a sulfur XVI line for the first time as one from the SWCX emission,\nthe fact of which suggests that it is the most spectrally-rich SWCX event ever\nobserved. We suggest that observations of ICME-driven SWCX events can provide a\nunique probe to study the population of highly-ionized ions in the plasma,\nwhich is difficult to measure in currently-available in-situ observations.\n"}
{"abstract": "  A key way to construct complex distributed systems is through modular\ncomposition of linearizable concurrent objects. A prominent example is shared\nregisters, which have crash-tolerant implementations on top of message-passing\nsystems, allowing the advantages of shared memory to carry over to\nmessage-passing. Yet linearizable registers do not always behave properly when\nused inside randomized programs. A strengthening of linearizability, called\nstrong linearizability, has been shown to preserve probabilistic behavior, as\nwell as other hypersafety properties. In order to exploit composition and\nabstraction in message-passing systems, it is crucial to know whether there\nexist strongly-linearizable implementations of registers in message-passing.\nThis paper answers the question in the negative: there are no\nstrongly-linearizable fault-tolerant message-passing implementations of\nmulti-writer registers, max-registers, snapshots or counters. This result is\nproved by reduction from the corresponding result by Helmi et al. The reduction\nis a novel extension of the BG simulation that connects shared-memory and\nmessage-passing, supports long-lived objects, and preserves strong\nlinearizability. The main technical challenge arises from the discrepancy\nbetween the potentially minuscule fraction of failures to be tolerated in the\nsimulated message-passing algorithm and the large fraction of failures that can\nafflict the simulating shared-memory system. The reduction is general and can\nbe viewed as the inverse of the ABD simulation of shared memory in\nmessage-passing.\n"}
{"abstract": "  Motivated by the recent observation of anomalous electron-positron angular\ncorrelations in the decay of the 18.15 MeV 1+ excited states in 8Be, we\nreexamine in detail the Standard Model expectations for these angular\ncorrelations. The 18.15 MeV state is above particle threshold, and several\nmultipoles can contribute to its $e^+e^-$ decay. We present the general\ntheoretical expressions for $e^+e^-$ angular distributions for nuclear decay by\nC0, C1, C2 M1, E1, and E2 multipoles, and we examine their relative\ncontribution to the $e^+e^-$ decay of 8Be at 18.15 MeV. We find that this\nresonance is dominated by M1 and E1 decay, and that the ratio of M1 to E1\nstrength is a strong function of energy. This is in contract to the original\nanalysis of the $e^+e^-$ angular distributions, where the M1/E1 ratio was\nassumed to be a constant over the energy region Ep = 0:8-1:2 MeV. We find that\nthe existence of a `bump' in the measured angular distribution is strongly\ndependent on the assumed M1/E1 ratio, with the present analysis finding the\nmeasured large-angle contributions to the $e^+e^-$ angular distribution to be\nlower than expectation. Thus, in the current analysis we find no evidence for\naxion decay in the 18.15 MeV resonance region of 8Be.\n"}
{"abstract": "  The use of magneto-optical techniques to tune the plasmonic response of\nnanostructures is a hot topic in active plasmonics, with fascinating\nimplications for several plasmon-based applications and devices. For this\nemerging field, called magnetoplasmonics, plasmonic nanomaterials with strong\noptical response to magnetic field are desired, which is generally challenging\nto achieve with pure noble metals. To overcome this issue, several efforts have\nbeen carried out to design and tailor the magneto-optical response of metal\nnanostructures, mainly by combining plasmonic and magnetic materials in a\nsingle nanostructure. In this tutorial we focus our attention on\nmagnetoplasmonic effects in purely plasmonic nanostructures, as they are a\nvaluable model system allowing for an easier rationalization of\nmagnetoplasmonic effects. The most common magneto-optical experimental methods\nemployed to measure these effects are introduced, followed by a review of the\nmajor experimental observations that are discussed within the framework of an\nanalytical model developed for the rationalization of magnetoplasmonic effects.\nDifferent materials are discussed, from noble metals to novel plasmonic\nmaterials, such as heavily doped semiconductors.\n"}
{"abstract": "  This paper addresses the problem of assessing trajectory quality in\nconditions when no ground truth poses are available or when their accuracy is\nnot enough for the specific task - for example, small-scale mapping in outdoor\nscenes. In our work, we propose a no-reference metric, Mutually Orthogonal\nMetric (MOM), that estimates the quality of the map from registered point\nclouds via the trajectory poses. MOM strongly correlates with full-reference\ntrajectory metric Relative Pose Error, making it a trajectory benchmarking tool\non setups where 3D sensing technologies are employed. We provide a mathematical\nfoundation for such correlation and confirm it statistically in synthetic\nenvironments. Furthermore, since our metric uses a subset of points from\nmutually orthogonal surfaces, we provide an algorithm for the extraction of\nsuch subset and evaluate its performance in synthetic CARLA environment and on\nKITTI dataset. The code of the proposed metric is publicly available as\npip-package.\n"}
{"abstract": "  Object detection plays an important role in self-driving cars for security\ndevelopment. However, mobile systems on self-driving cars with limited\ncomputation resources lead to difficulties for object detection. To facilitate\nthis, we propose a compiler-aware neural pruning search framework to achieve\nhigh-speed inference on autonomous vehicles for 2D and 3D object detection. The\nframework automatically searches the pruning scheme and rate for each layer to\nfind a best-suited pruning for optimizing detection accuracy and speed\nperformance under compiler optimization. Our experiments demonstrate that for\nthe first time, the proposed method achieves (close-to) real-time, 55ms and\n99ms inference times for YOLOv4 based 2D object detection and PointPillars\nbased 3D detection, respectively, on an off-the-shelf mobile phone with minor\n(or no) accuracy loss.\n"}
{"abstract": "  Spatio-temporal counts of infectious disease cases often contain an excess of\nzeros. With existing zero inflated count models applied to such data it is\ndifficult to quantify space-time heterogeneity in the effects of disease spread\nbetween areas. Also, existing methods do not allow for separate dynamics to\naffect the reemergence and persistence of the disease. As an alternative, we\ndevelop a new zero-state coupled Markov switching negative binomial model,\nunder which the disease switches between periods of presence and absence in\neach area through a series of partially hidden nonhomogeneous Markov chains\ncoupled between neighboring locations. When the disease is present, an\nautoregressive negative binomial model generates the cases with a possible 0\nrepresenting the disease being undetected. Bayesian inference and prediction is\nillustrated using spatio-temporal counts of dengue fever cases in Rio de\nJaneiro, Brazil.\n"}
{"abstract": "  Verifying quantum systems has attracted a lot of interests in the last\ndecades. In this paper, we initialised the model checking of quantum\ncontinuous-time Markov chain (QCTMC). As a real-time system, we specify the\ntemporal properties on QCTMC by signal temporal logic (STL). To effectively\ncheck the atomic propositions in STL, we develop a state-of-art real root\nisolation algorithm under Schanuel's conjecture; further, we check the general\nSTL formula by interval operations with a bottom-up fashion, whose query\ncomplexity turns out to be linear in the size of the input formula by calling\nthe real root isolation algorithm. A running example of an open quantum walk is\nprovided to demonstrate our method.\n"}
{"abstract": "  We employ the inductive structure of determinantal varieties to calculate the\nweight filtration on local cohomology modules with determinantal support. We\nshow that the weight of a simple composition factor is uniquely determined by\nits support and cohomological degree. As a consequence, we obtain the\nequivariant structure of the Hodge filtration on each local cohomology module,\nand we provide a formula for its generation level. In the case of square\nmatrices, we express the Hodge filtration in terms of the Hodge ideals for the\ndeterminant hypersurface. As an application, we describe a recipe for\ncalculating the mixed Hodge module structure on any iteration of local\ncohomology functors with determinantal support.\n"}
{"abstract": "  Human motion prediction aims to forecast future human poses given a prior\npose sequence. The discontinuity of the predicted motion and the performance\ndeterioration in long-term horizons are still the main challenges encountered\nin current literature. In this work, we tackle these issues by using a compact\nmanifold-valued representation of human motion. Specifically, we model the\ntemporal evolution of the 3D human poses as trajectory, what allows us to map\nhuman motions to single points on a sphere manifold. To learn these\nnon-Euclidean representations, we build a manifold-aware Wasserstein generative\nadversarial model that captures the temporal and spatial dependencies of human\nmotion through different losses. Extensive experiments show that our approach\noutperforms the state-of-the-art on CMU MoCap and Human 3.6M datasets. Our\nqualitative results show the smoothness of the predicted motions.\n"}
{"abstract": "  Mixture-of-Experts (MoE) models can achieve promising results with outrageous\nlarge amount of parameters but constant computation cost, and thus it has\nbecome a trend in model scaling. Still it is a mystery how MoE layers bring\nquality gains by leveraging the parameters with sparse activation. In this\nwork, we investigate several key factors in sparse expert models. We observe\nthat load imbalance may not be a significant problem affecting model quality,\ncontrary to the perspectives of recent studies, while the number of sparsely\nactivated experts $k$ and expert capacity $C$ in top-$k$ routing can\nsignificantly make a difference in this context. Furthermore, we take a step\nforward to propose a simple method called expert prototyping that splits\nexperts into different prototypes and applies $k$ top-$1$ routing. This\nstrategy improves the model quality but maintains constant computational costs,\nand our further exploration on extremely large-scale models reflects that it is\nmore effective in training larger models. We push the model scale to over $1$\ntrillion parameters and implement it on solely $480$ NVIDIA V100-32GB GPUs, in\ncomparison with the recent SOTAs on $2048$ TPU cores. The proposed giant model\nachieves substantial speedup in convergence over the same-size baseline.\n"}
{"abstract": "  We present a method of defining projectors in the virtual Temperley-Lieb\nalgebra, that generalizes the Jones-Wenzl projectors in Temperley-Lieb algebra.\nWe show that the projectors have similar properties with the Jones-Wenzl\nprojectors, and contain an extra property which is associated with the virtual\ngenerator elements, that is, the product of a projector with a virtual\ngenerator is unchanged. We also show the uniqueness of the projector $f_n$ in\nterms of its axiomatic properties in the virtual Temperley-Lieba algebra\n$VTL_n(d)$. Finally, we find the coefficients of $f_n$ and give an explicit\nformula for the projector $f_n$.\n"}
{"abstract": "  We develop a novel Monte Carlo algorithm for the vector consisting of the\nsupremum, the time at which the supremum is attained and the position of an\nexponentially tempered L\\'{e}vy process. The algorithm, based on the increments\nof the process without tempering, converges geometrically fast (as a function\nof the computational cost) for discontinuous and locally Lipschitz functions of\nthe vector. We prove that the corresponding multilevel Monte Carlo estimator\nhas optimal computational complexity (i.e. of order $\\epsilon^{-2}$ if the mean\nsquared error is at most $\\epsilon^{2}$) and provide its central limit theorem\n(CLT). Using the CLT we construct confidence intervals for barrier option\nprices and various risk measures based on drawdown under the tempered stable\n(CGMY) model calibrated/estimated on real-world data. We provide non-asymptotic\nand asymptotic comparisons of our algorithm with existing approximations,\nleading to rule-of-thumb guidelines for users to the best method for a given\nset of parameters, and illustrate its performance with numerical examples.\n"}
{"abstract": "  With recent progress in the field of Explainable Artificial Intelligence\n(XAI) and increasing use in practice, the need for an evaluation of different\nXAI methods and their explanation quality in practical usage scenarios arises.\nFor this purpose, we present VitrAI, which is a web-based service with the goal\nof uniformly demonstrating four different XAI algorithms in the context of\nthree real life scenarios and evaluating their performance and\ncomprehensibility for humans. This work reveals practical obstacles when\nadopting XAI methods and gives qualitative estimates on how well different\napproaches perform in said scenarios.\n"}
{"abstract": "  Extra dimensions can be utilized to simplify problems in classical mechanics,\noffering new insights. Here we show a simple example of how the motion of a\ntest particle under the influence of an inverse-quadratic potential in 1D is\nequivalent to that of another test particle moving freely in 2D Euclidean space\nand 3D Minkowskian space.\n"}
{"abstract": "  We design a neural network to extract and process features from absorption\nimages taken of one-dimensional Bose gases in the quasi-condensate regime.\nSpecifically, the network is trained to predict both the temperature of single\nrealizations of the system and the uncertainty thereof. For multiple\nrealizations, the individual predictions can be combined in an estimate of the\nmean temperature, improving precision. We benchmark our model on both simulated\nand experimentally measured data and compare it to the established method of\ndensity ripples thermometry. We find the predictions of the two methods\ncompatible, although the neural network reaches similar precision needing much\nfewer realizations, thus highlighting the efficiency gain achievable when\nincorporating neural networks into analysis of data from cold gas experiments.\nFurther, we study feature maps to reveal which local features of the condensate\nare extracted by the network and how said features correlate with properties of\nthe system. A similar analysis could be employed to uncover physical relations\nin more complex systems.\n"}
{"abstract": "  Art is a variety of human activities that include the production of visual,\nauditory, or performing objects that express the creativity, creative concepts,\nor technological abilities of the artist, intended primarily for their beauty\nor emotional power to be appreciated. The renaissance of historic and forgotten\nart has been made possible by modern developments in Artificial Intelligence.\nTechniques for Computer Vision have long been related to such arts. Style\nTransfer using Neural Networks refers to optimization techniques, where a\ncontent image and a style image are taken and blended such that it feels like\nthe content image is reconstructed in the style image color palette. This paper\nimplements the Style Transfer using three different Neural Networks in form of\nan application that is accessible to the general population thereby reviving\ninterest in lost art styles.\n"}
{"abstract": "  We study the dynamics of the one-dimensional $\\varepsilon$-dependent\nCahn-Hilliard / Allen-Cahn equation within a neighborhood of an equilibrium of\n$N$ transition layers, that in general does not conserve mass. Two different\nsettings are considered which differ in that, for the second, we impose a\nmass-conservation constraint in place of one of the zero-mass flux boundary\nconditions at $x=1$. Motivated by the study of Carr and Pego on the layered\nmetastable patterns of Allen-Cahn in [10], and by this of Bates and Xun in [5]\nfor the Cahn-Hilliard equation, we implement an $N$-dimensional, and a\nmass-conservative $ N-1$-dimensional manifold respectively; therein, a\nmetastable state with $N$ transition layers is approximated. We then determine,\nfor both cases, the essential dynamics of the layers (ode systems with the\nequations of motion), expressed in terms of local coordinates relative to the\nmanifold used. In particular, we estimate the spectrum of the linearized\nCahn-Hilliard / Allen-Cahn operator, and specify wide families of\n$\\varepsilon$-dependent weights $\\delta(\\varepsilon)$, $\\mu(\\varepsilon)$,\nacting at each part of the operator, for which the dynamics are stable and rest\nexponentially small in $\\varepsilon$. Our analysis enlightens the role of mass\nconservation in the classification of the general mixed problem into two main\ncategories where the solution has a profile close to Allen-Cahn, or, when the\nmass is conserved, close to the Cahn-Hilliard solution.\n"}
{"abstract": "  We study actions of pointed Hopf algebras in the $\\ZZ$-graded setting. Our\nmain result classifies inner-faithful actions of generalized Taft algebras on\nquantum generalized Weyl algebras which respect the $\\ZZ$-grading. We also show\nthat generically the invariant rings of Taft actions on quantum generalized\nWeyl algebras are commutative Kleinian singularities.\n"}
{"abstract": "  Quantum computing is among the most significant technologies to emerge in\nrecent decades, offering the promise of paradigm-shifting computational\ncapacity with significant ethical consequences. On a technical level, the\nunique features of quantum computation have consequences for the imposition of\nfairness and ethical constraints on computation. Despite its significance,\nlittle if no structured research has been undertaken into the ethical\nimplications of quantum technologies. In this paper, we fill this gap in the\nliterature by presenting a roadmap for ethical quantum computing setting out\nprospective research programmes. We situate quantum ethics at the\ncross-disciplinary intersection of quantum information science, technology\nethics and moral philosophy. We summarise the key elements of quantum\ninformation processing relevant to ethical analysis and set-out taxonomies for\nuse by researchers considering the ethics of quantum technologies. In doing so,\nwe inaugurate the cross-disciplinary field of the ethics of quantum computing.\n"}
{"abstract": "  In the case that user profiles are not available, the recommendation based on\nanonymous session is particularly important, which aims to predict the items\nthat the user may click at the next moment based on the user's access sequence\nover a while. In recent years, with the development of recurrent neural\nnetwork, attention mechanism, and graph neural network, the performance of\nsession-based recommendation has been greatly improved. However, the previous\nmethods did not comprehensively consider the context dependencies and\nshort-term interest first of the session. Therefore, we propose a context-aware\nshort-term interest first model (CASIF).The aim of this paper is improve the\naccuracy of recommendations by combining context and short-term interest. In\nCASIF, we dynamically construct a graph structure for session sequences and\ncapture rich context dependencies via graph neural network (GNN), latent\nfeature vectors are captured as inputs of the next step. Then we build the\nshort-term interest first module, which can to capture the user's general\ninterest from the session in the context of long-term memory, at the same time\nget the user's current interest from the item of the last click. In the end,\nthe short-term and long-term interest are combined as the final interest and\nmultiplied by the candidate vector to obtain the recommendation probability.\nFinally, a large number of experiments on two real-world datasets demonstrate\nthe effectiveness of our proposed method.\n"}
{"abstract": "  Subspace codes are collections of subspaces of a projective space such that\nany two subspaces satisfy a pairwise minimum distance criterion. Recent results\nhave shown that it is possible to construct optimal $(5,3)$ subspace codes from\npairs of partial spreads in the projective space $\\mathrm{PG}(4,q)$ over the\nfinite field $ \\mathbb{F}_q $, termed doubling codes. We have utilized a\ncomplete classification of maximal partial line spreads in $\\mathrm{PG}(4,2)$\nin literature to establish the types of the spreads in the doubling code\ninstances obtained from two recent constructions of optimum $(5,3)_q$ codes,\nrestricted to $ \\mathbb{F}_2 $. Further we present a new characterization of a\nsubclass of binary doubling codes based on the intersection patterns of key\nsubspaces in the pair of constituent spreads.\n"}
{"abstract": "  Generalised linear models for multi-class classification problems are one of\nthe fundamental building blocks of modern machine learning tasks. In this\nmanuscript, we characterise the learning of a mixture of $K$ Gaussians with\ngeneric means and covariances via empirical risk minimisation (ERM) with any\nconvex loss and regularisation. In particular, we prove exact asymptotics\ncharacterising the ERM estimator in high-dimensions, extending several previous\nresults about Gaussian mixture classification in the literature. We exemplify\nour result in two tasks of interest in statistical learning: a) classification\nfor a mixture with sparse means, where we study the efficiency of $\\ell_1$\npenalty with respect to $\\ell_2$; b) max-margin multi-class classification,\nwhere we characterise the phase transition on the existence of the multi-class\nlogistic maximum likelihood estimator for $K>2$. Finally, we discuss how our\ntheory can be applied beyond the scope of synthetic data, showing that in\ndifferent cases Gaussian mixtures capture closely the learning curve of\nclassification tasks in real data sets.\n"}
{"abstract": "  In its observation band, the Laser Interferometer Space Antenna (LISA) will\nsimultaneously observe stochastic gravitational-wave background (SGWB) signals\nof different origins; orbitally modulated waveforms from galactic white dwarf\nbinaries, a binary black hole produced background, and possibly a\ncosmologically produced SGWB. We simulate the emission of gravitational waves\nfrom galactic white dwarf binaries based on the Lamberts \\cite{Lamberts}\ndistributions and determine a complex waveform from the galactic foreground. We\ngenerate the modulated galactic signal detected by LISA due to its orbital\nmotion, and present a data analysis strategy to address it. The Fisher\nInformation and Markov Chain Monte Carlo methods give an estimate of the LISA\nnoise and parameters for the different signal sources. We simultaneously\nestimate the galactic foreground, the astrophysical and cosmological\nbackgrounds, and estimate detection limits for the future LISA observation of\nthe SGWB in the spectral domain with the 3 LISA channels $ A $, $ E $ and $ T\n$. In the context of the expected astrophysical background and a galactic\nforeground, a cosmological background energy density of about $\n\\Omega_{GW,Cosmo} \\approx 8 \\times 10^{-13} $ could be detected by LISA with\nour spectral separation strategy.\n"}
{"abstract": "  Recommending cold-start items is a long-standing and fundamental challenge in\nrecommender systems. Without any historical interaction on cold-start items, CF\nscheme fails to use collaborative signals to infer user preference on these\nitems. To solve this problem, extensive studies have been conducted to\nincorporate side information into the CF scheme. Specifically, they employ\nmodern neural network techniques (e.g., dropout, consistency constraint) to\ndiscover and exploit the coalition effect of content features and collaborative\nrepresentations. However, we argue that these works less explore the mutual\ndependencies between content features and collaborative representations and\nlack sufficient theoretical supports, thus resulting in unsatisfactory\nperformance. In this work, we reformulate the cold-start item representation\nlearning from an information-theoretic standpoint. It aims to maximize the\nmutual dependencies between item content and collaborative signals.\nSpecifically, the representation learning is theoretically lower-bounded by the\nintegration of two terms: mutual information between collaborative embeddings\nof users and items, and mutual information between collaborative embeddings and\nfeature representations of items. To model such a learning process, we devise a\nnew objective function founded upon contrastive learning and develop a simple\nyet effective Contrastive Learning-based Cold-start Recommendation\nframework(CLCRec). In particular, CLCRec consists of three components:\ncontrastive pair organization, contrastive embedding, and contrastive\noptimization modules. It allows us to preserve collaborative signals in the\ncontent representations for both warm and cold-start items. Through extensive\nexperiments on four publicly accessible datasets, we observe that CLCRec\nachieves significant improvements over state-of-the-art approaches in both\nwarm- and cold-start scenarios.\n"}
{"abstract": "  Existing scene understanding systems mainly focus on recognizing the visible\nparts of a scene, ignoring the intact appearance of physical objects in the\nreal-world. Concurrently, image completion has aimed to create plausible\nappearance for the invisible regions, but requires a manual mask as input. In\nthis work, we propose a higher-level scene understanding system to tackle both\nvisible and invisible parts of objects and backgrounds in a given scene.\nParticularly, we built a system to decompose a scene into individual objects,\ninfer their underlying occlusion relationships, and even automatically learn\nwhich parts of the objects are occluded that need to be completed. In order to\ndisentangle the occluded relationships of all objects in a complex scene, we\nuse the fact that the front object without being occluded is easy to be\nidentified, detected, and segmented. Our system interleaves the two tasks of\ninstance segmentation and scene completion through multiple iterations, solving\nfor objects layer-by-layer. We first provide a thorough experiment using a new\nrealistically rendered dataset with ground-truths for all invisible regions. To\nbridge the domain gap to real imagery where ground-truths are unavailable, we\nthen train another model with the pseudo-ground-truths generated from our\ntrained synthesis model. We demonstrate results on a wide variety of datasets\nand show significant improvement over the state-of-the-art.\n"}
{"abstract": "  As predictive models are deployed into the real world, they must increasingly\ncontend with strategic behavior. A growing body of work on strategic\nclassification treats this problem as a Stackelberg game: the decision-maker\n\"leads\" in the game by deploying a model, and the strategic agents \"follow\" by\nplaying their best response to the deployed model. Importantly, in this\nframing, the burden of learning is placed solely on the decision-maker, while\nthe agents' best responses are implicitly treated as instantaneous. In this\nwork, we argue that the order of play in strategic classification is\nfundamentally determined by the relative frequencies at which the\ndecision-maker and the agents adapt to each other's actions. In particular, by\ngeneralizing the standard model to allow both players to learn over time, we\nshow that a decision-maker that makes updates faster than the agents can\nreverse the order of play, meaning that the agents lead and the decision-maker\nfollows. We observe in standard learning settings that such a role reversal can\nbe desirable for both the decision-maker and the strategic agents. Finally, we\nshow that a decision-maker with the freedom to choose their update frequency\ncan induce learning dynamics that converge to Stackelberg equilibria with\neither order of play.\n"}
{"abstract": "  Monte Carlo simulation has been employed to investigate the magnetic\nproperties and phase diagrams of ferrimagnetic mixed-spin (1/2, 1) triangular\nIsing nanotube with core-shell structure. In particular, the effect of the\nexchange couplings and the temperature on the magnetic and thermodynamic\nproperties, hysteresis loops as well as the compensation temperature have been\ndiscussed in detail. Moreover, the effects of the single-ion anisotropy, as\nwell as external magnetic field, have been examined. The threshold values of\nthe exchange couplings and single-ion anisotropy have been found, determining\nwhether the system exhibits a compensation temperature. We have found that the\nappearance of the compensation temperature is strongly linked with the system\nparameters. Moreover, we have obtained the double and triple hysteresis loops\nfor certain physical parameters in the considered magnetic system.\n"}
{"abstract": "  Label smoothing (LS) is an arising learning paradigm that uses the positively\nweighted average of both the hard training labels and uniformly distributed\nsoft labels. It was shown that LS serves as a regularizer for training data\nwith hard labels and therefore improves the generalization of the model. Later\nit was reported LS even helps with improving robustness when learning with\nnoisy labels. However, we observe that the advantage of LS vanishes when we\noperate in a high label noise regime. Puzzled by the observation, we proceeded\nto discover that several proposed learning-with-noisy-labels solutions in the\nliterature instead relate more closely to negative label smoothing (NLS), which\ndefines as using a negative weight to combine the hard and soft labels! We show\nthat NLS differs substantially from LS in their achieved model confidence. To\ndifferentiate the two cases, we will call LS the positive label smoothing\n(PLS), and this paper unifies PLS and NLS into generalized label smoothing\n(GLS). We provide understandings for the properties of GLS when learning with\nnoisy labels. Among other established properties, we theoretically show NLS is\nconsidered more beneficial when the label noise rates are high. We provide\nextensive experimental results on multiple benchmarks to support our findings\ntoo.\n"}
{"abstract": "  Fake news detection is an important task for increasing the credibility of\ninformation on the media since fake news is constantly spreading on social\nmedia every day and it is a very serious concern in our society. Fake news is\nusually created by manipulating images, texts, and videos. In this paper, we\npresent a novel method for detecting fake news by fusing multimodal features\nderived from textual and visual data. Specifically, we used a pre-trained BERT\nmodel to learn text features and a VGG-19 model pre-trained on the ImageNet\ndataset to extract image features. We proposed a scale-dot product attention\nmechanism to capture the relationship between text features and visual\nfeatures. Experimental results showed that our approach performs better than\nthe current state-of-the-art method on a public Twitter dataset by 3.1%\naccuracy.\n"}
{"abstract": "  Given the prevalence of missing data in modern statistical research, a broad\nrange of methods is available for any given imputation task. How does one\nchoose the 'best' imputation method in a given application? The standard\napproach is to select some observations, set their status to missing, and\ncompare prediction accuracy of the methods under consideration of these\nobservations. Besides having to somewhat artificially mask observations, a\nshortcoming of this approach is that imputations based on the conditional mean\nwill rank highest if predictive accuracy is measured with quadratic loss. In\ncontrast, we want to rank highest an imputation that can sample from the true\nconditional distributions. In this paper, we develop a framework called\n\"Imputation Scores\" (I-Scores) for assessing missing value imputations. We\nprovide a specific I-Score based on density ratios and projections, that is\napplicable to discrete and continuous data. It does not require to mask\nadditional observations for evaluations and is also applicable if there are no\ncomplete observations. The population version is shown to be proper in the\nsense that the highest rank is assigned to an imputation method that samples\nfrom the correct conditional distribution. The propriety is shown under the\nmissing completely at random (MCAR) assumption but is also shown to be valid\nunder missing at random (MAR) with slightly more restrictive assumptions. We\nshow empirically on a range of data sets and imputation methods that our score\nconsistently ranks true data high(est) and is able to avoid pitfalls usually\nassociated with performance measures such as RMSE. Finally, we provide the\nR-package Iscores available on CRAN with an implementation of our method.\n"}
{"abstract": "  We consider the sub-Riemannian $3$-sphere $(\\mathbb{S}^3,g_h)$ obtained by\nrestriction of the Riemannian metric of constant curvature $1$ to the planar\ndistribution orthogonal to the vertical Hopf vector field. It is known that\n$(\\mathbb{S}^3,g_h)$ contains a family of spherical surfaces\n$\\{\\mathcal{S}_\\lambda\\}_{\\lambda\\geq 0}$ with constant mean curvature\n$\\lambda$. In this work we first prove that the two closed half-spheres of\n$\\mathcal{S}_0$ with boundary $C_0=\\{0\\}\\times\\mathbb{S}^1$ minimize the\nsub-Riemannian area among compact $C^1$ surfaces with the same boundary. We\nalso see that the only $C^2$ solutions to this Plateau problem are vertical\ntranslations of such half-spheres. Second, we establish that the closed\n$3$-ball enclosed by a sphere $\\mathcal{S}_\\lambda$ with $\\lambda>0$ uniquely\nsolves the isoperimetric problem in $(\\mathbb{S}^3,g_h)$ for $C^1$ sets inside\na vertical solid tube and containing a horizontal section of the tube. The\nproofs mainly rely on calibration arguments.\n"}
{"abstract": "  In this article we present a functional central limit theorem for the paths\nof the directed random polymer in space dimension two in the subcritical\nintermediate disorder regime. More precisely, diffusively rescaled polymer\npaths tested against bounded continuous functions converge in probability to\ntheir Brownian motion counterparts, when taking the weak disorder limit. So far\nanalogous results have only been established for $d\\neq 2$. Moreover, we show\nconvergence in probability of the finite-dimensional marginal distributions of\nthe polymer measure to the ones of Brownian motion. Along the way, we prove a\nlocal limit theorem which allows us to factorise the point-to-point partition\nfunction of the directed polymer into a product of two point-to-plane partition\nfunctions.\n"}
{"abstract": "  High Altitude Platform Station (HAPS) has the potential to provide global\nwireless connectivity and data services such as high-speed wireless backhaul,\nindustrial Internet of things (IoT), and public safety for large areas not\nserved by terrestrial networks. A unified HAPS design is desired to support\nvarious use cases and a wide range of requirements. In this paper, we present\ntwo architecture designs of the HAPS system: i) repeater based HAPS, and ii)\nbase station based HAPS, which are both viable technical solutions. The energy\nefficiency is analyzed and compared between the two architectures using\nconsumption factor theory. The system performance of these two architectures is\nevaluated through Monte Carlo simulations and is characterized in metrics of\nspectral efficiency using LTE band 1 for both single-cell and multi-cell cases.\nBoth designs can provide good downlink spectral efficiency and coverage, while\nthe uplink coverage is significantly limited by UE transmit power and antenna\ngain. Using directional antennas at the UEs can improve the system performance\nfor both downlink and uplink.\n"}
{"abstract": "  Unsourced random access is a novel communication paradigm designed for\nhandling a large number of uncoordinated users that sporadically transmit very\nshort messages. Under this model, coded compressed sensing (CCS) has emerged as\na low-complexity scheme that exhibits good error performance. Yet, one of the\nchallenges faced by CCS pertains to disentangling a large number of codewords\npresent on a single factor graph. To mitigate this issue, this article\nintroduces a modified CCS scheme whereby active devices stochastically\npartition themselves into groups that utilize separate sampling matrices with\nlow cross-coherence for message transmission. At the receiver, ideas from the\nfield of compressed demixing are employed for support recovery, and separate\nfactor graphs are created for message disambiguation in each cluster. This\nreduces the number of active users on a factor graph, which improves\nperformance significantly in typical scenarios. Indeed, coded demixing reduces\nthe probability of error as the number of groups increases, up to a point.\nFindings are supported with numerical simulations.\n"}
{"abstract": "  Cloud applications are increasingly shifting from large monolithic services,\nto large numbers of loosely-coupled, specialized microservices. Despite their\nadvantages in terms of facilitating development, deployment, modularity, and\nisolation, microservices complicate resource management, as dependencies\nbetween them introduce backpressure effects and cascading QoS violations.\n  We present Sinan, a data-driven cluster manager for interactive cloud\nmicroservices that is online and QoS-aware. Sinan leverages a set of scalable\nand validated machine learning models to determine the performance impact of\ndependencies between microservices, and allocate appropriate resources per tier\nin a way that preserves the end-to-end tail latency target. We evaluate Sinan\nboth on dedicated local clusters and large-scale deployments on Google Compute\nEngine (GCE) across representative end-to-end applications built with\nmicroservices, such as social networks and hotel reservation sites. We show\nthat Sinan always meets QoS, while also maintaining cluster utilization high,\nin contrast to prior work which leads to unpredictable performance or\nsacrifices resource efficiency. Furthermore, the techniques in Sinan are\nexplainable, meaning that cloud operators can yield insights from the ML models\non how to better deploy and design their applications to reduce unpredictable\nperformance.\n"}
{"abstract": "  I present a scheme to obtain a 2 to 40 GeV low emittance muon beam, not\nrequiring cooling and within today's technological resources, to be used for\nearly commissioning of muon accelerator projects, or alternatively dedicated\nmuon and neutrino parameter measurements. In particular, a muon rate of 5x10^4\nmu/s in a normalized transverse emittance of 5 um at 22 GeV, and energy spread\nof 1 GeV obtained from O(10^11) e+/s on target at 44 GeV. This emittance is\nbelow the expected results of advanced emittance cooling techniques for muons\nproduced from protons-on-target, and represents an alternative for the duration\nof complete muon cooling studies. The scheme has beam designed to adjust the\nmuon beam energy in the GeV energy range to the needs for precise parameter\nmeasurements of muons and neutrinos. Although the rate is small compared to\nother muon sources, it does not seem to represent a big limitation for its\nusage. Furthermore, the muon rate could be in principle increased\nproportionally to the availability of higher positron rates, already foreseen\nfor future collider projects.\n"}
{"abstract": "  This article is motivated by the objective of providing a new analytically\ntractable and fully frequentist framework to characterize and implement\nregression trees while also allowing a multivariate (potentially high\ndimensional) response. The connection to regression trees is made by a high\ndimensional model with dynamic mean vectors over multi-dimensional change axes.\nOur theoretical analysis is carried out under a single two dimensional change\npoint setting. An optimal rate of convergence of the proposed estimator is\nobtained, which in turn allows existence of limiting distributions.\nDistributional behavior of change point estimates are split into two distinct\nregimes, the limiting distributions under each regime is then characterized, in\nturn allowing construction of asymptotically valid confidence intervals for\n$2d$-location of change. All results are obtained under a high dimensional\nscaling $s\\log^2 p=o(T_wT_h),$ where $p$ is the response dimension, $s$ is a\nsparsity parameter, and $T_w,T_h$ are sampling periods along change axes. We\ncharacterize full regression trees by defining a multiple multi-dimensional\nchange point model. Natural extensions of the single $2d$-change point\nestimation methodology are provided. Two applications, first on segmentation of\n{\\it Infra-red astronomy satellite (IRAS)} data and second to segmentation of\ndigital images are provided. Methodology and theoretical results are supported\nwith monte-carlo simulations.\n"}
{"abstract": "  New hardware architectures open up immense opportunities for supercomputer\nsimulations. However, programming techniques for different architectures vary\nsignificantly, which leads to the necessity of developing and supporting\nmultiple code versions, each being optimized for specific hardware features.\nThe oneAPI framework, recently introduced by Intel, contains a set of\nprogramming tools for the development of portable codes that can be compiled\nand fine-tuned for CPUs, GPUs, FPGAs, and accelerators. In this paper, we\nreport on the experience of porting the implementation of Boris particle pusher\nto oneAPI. Boris particle pusher is one of the most demanding computational\nstages of the Particle-in-Cell method, which, in particular, is used for\nsupercomputer simulations of laser-plasma interactions. We show how to adapt\nthe C++ implementation of the particle push algorithm from the Hi-Chi project\nto the DPC++ programming language and report the performance of the code on\nhigh-end Intel CPUs (Xeon Platinum 8260L) and Intel GPUs (P630 and Iris Xe\nMax). It turned out that our C++ code can be easily ported to DPC++. We found\nthat on CPUs the resulting DPC++ code is only ~10% on average inferior to the\noptimized C++ code. Moreover, the code is compiled and run on new Intel GPUs\nwithout any specific optimizations and shows the expected performance, taking\ninto account the parameters of the hardware.\n"}
{"abstract": "  In this article, we show that in the usual type-I seesaw framework, augmented\nsolely by a neutrino portal interaction, the dark matter (DM) relic density can\nbe created through freeze-in in a manner fully determined by the seesaw\ninteractions and the DM particle mass. This simple freeze-in scenario, where\ndark matter is not a seesaw state, proceeds through slow, seesaw-induced decays\nof Higgs, $W$ and $Z$ bosons. We identify two scenarios, one of which predicts\nthe existence of an observable neutrino line.\n"}
{"abstract": "  In this note, approaches towards local $p$-adic and global adelic quantum\ngates are proposed.\n"}
{"abstract": "  We present a framework to generate synthetic historical documents with\nprecise ground truth using nothing more than a collection of unlabeled\nhistorical images. Obtaining large labeled datasets is often the limiting\nfactor to effectively use supervised deep learning methods for Document Image\nAnalysis (DIA). Prior approaches towards synthetic data generation either\nrequire expertise or result in poor accuracy in the synthetic documents. To\nachieve high precision transformations without requiring expertise, we tackle\nthe problem in two steps. First, we create template documents with\nuser-specified content and structure. Second, we transfer the style of a\ncollection of unlabeled historical images to these template documents while\npreserving their text and layout. We evaluate the use of our synthetic\nhistorical documents in a pre-training setting and find that we outperform the\nbaselines (randomly initialized and pre-trained). Additionally, with visual\nexamples, we demonstrate a high-quality synthesis that makes it possible to\ngenerate large labeled historical document datasets with precise ground truth.\n"}
{"abstract": "  Multi-source-extractors are functions that extract uniform randomness from\nmultiple (weak) sources of randomness. Quantum multi-source-extractors were\nconsidered by Kasher and Kempe (for the quantum-independent-adversary and the\nquantum-bounded-storage-adversary), Chung, Li and Wu (for the\ngeneral-entangled-adversary) and Arnon-Friedman, Portmann and Scholz (for the\nquantum-Markov-adversary). One of the main objectives of this work is to unify\nall the existing quantum multi-source adversary models. We propose two new\nmodels of adversaries: 1) the quantum-measurement-adversary (qm-adv), which\ngenerates side-information using entanglement and on post-measurement and 2)\nthe quantum-communication-adversary (qc-adv), which generates side-information\nusing entanglement and communication between multiple sources. We show that, 1.\nqm-adv is the strongest adversary among all the known adversaries, in the sense\nthat the side-information of all other adversaries can be generated by qm-adv.\n2. The (generalized) inner-product function (in fact a general class of\ntwo-wise independent functions) continues to work as a good extractor against\nqm-adv with matching parameters as that of Chor and Goldreich. 3. A\nnon-malleable-extractor proposed by Li (against classical-adversaries)\ncontinues to be secure against quantum side-information. This result implies a\nnon-malleable-extractor result of Aggarwal, Chung, Lin and Vidick with uniform\nseed. We strengthen their result via a completely different proof to make the\nnon-malleable-extractor of Li secure against quantum side-information even when\nthe seed is not uniform. 4. A modification (working with weak sources instead\nof uniform sources) of the Dodis and Wichs protocol for privacy-amplification\nis secure against active quantum adversaries. This strengthens on a recent\nresult due to Aggarwal, Chung, Lin and Vidick which uses uniform sources.\n"}
{"abstract": "  Even when neural networks are widely used in a large number of applications,\nthey are still considered as black boxes and present some difficulties for\ndimensioning or evaluating their prediction error. This has led to an\nincreasing interest in the overlapping area between neural networks and more\ntraditional statistical methods, which can help overcome those problems. In\nthis article, a mathematical framework relating neural networks and polynomial\nregression is explored by building an explicit expression for the coefficients\nof a polynomial regression from the weights of a given neural network, using a\nTaylor expansion approach. This is achieved for single hidden layer neural\nnetworks in regression problems. The validity of the proposed method depends on\ndifferent factors like the distribution of the synaptic potentials or the\nchosen activation function. The performance of this method is empirically\ntested via simulation of synthetic data generated from polynomials to train\nneural networks with different structures and hyperparameters, showing that\nalmost identical predictions can be obtained when certain conditions are met.\nLastly, when learning from polynomial generated data, the proposed method\nproduces polynomials that approximate correctly the data locally.\n"}
{"abstract": "  Solar flares convert magnetic energy into thermal and non-thermal plasma\nenergy, the latter implying particle acceleration of charged particles such as\nprotons. Protons are injected out of the coronal acceleration region and can\ninteract with dense plasma in the lower solar atmosphere, producing mesons that\nsubsequently decay into gamma rays and neutrinos at O(MeV-GeV) energies. We\npresent the results of the first search for GeV neutrinos emitted during solar\nflares carried out with the IceCube Neutrino Observatory. While the experiment\nwas originally designed to detect neutrinos with energies between 10 GeV and a\nfew PeV, a new approach allowing for a O(GeV) energy threshold will be\npresented. The resulting limits allow us to constrain some of the theoretical\nestimates of the expected neutrino flux.\n"}
{"abstract": "  A large majority of American adults get at least some of their news from the\nInternet. Even though many online news products have the goal of informing\ntheir users about the news, they lack scalable and reliable tools for measuring\nhow well they are achieving this goal, and therefore have to resort to noisy\nproxy metrics (e.g., click-through rates or reading time) to track their\nperformance.\n  As a first step towards measuring news informedness at a scale, we study the\nproblem of quiz-style multiple-choice question generation, which may be used to\nsurvey users about their knowledge of recent news. In particular, we formulate\nthe problem as two sequence-to-sequence tasks: question-answer generation (QAG)\nand distractor, or incorrect answer, generation (DG). We introduce NewsQuizQA,\nthe first dataset intended for quiz-style question-answer generation,\ncontaining 20K human written question-answer pairs from 5K news article\nsummaries. Using this dataset, we propose a series of novel techniques for\napplying large pre-trained Transformer encoder-decoder models, namely PEGASUS\nand T5, to the tasks of question-answer generation and distractor generation.\n  We show that our models outperform strong baselines using both automated\nmetrics and human raters. We provide a case study of running weekly quizzes on\nreal-world users via the Google Surveys platform over the course of two months.\nWe found that users generally found the automatically generated questions to be\neducational and enjoyable. Finally, to serve the research community, we are\nreleasing the NewsQuizQA dataset.\n"}
{"abstract": "  We present Contingency Model Predictive Control (CMPC), a motion planning and\ncontrol framework that optimizes performance objectives while simultaneously\nmaintaining a contingency plan -- an alternate trajectory that avoids a\npotential hazard. By preserving the existence of a feasible avoidance\ntrajectory, CMPC anticipates emergency and keeps the controlled system in a\nsafe state that is selectively robust to the identified hazard. We accomplish\nthis by adding an additional prediction horizon in parallel to the typical\nModel Predictive Control (MPC) horizon. This extra horizon is constrained to\nguarantee safety from the contingent threat and is coupled to the nominal\nhorizon at its first command. Thus, the two horizons negotiate to compute\ncommands that are both optimized for performance and robust to the contingent\nevent. This article presents a linear formulation for CMPC, illustrates its key\nfeatures on a toy problem, and then demonstrates its efficacy experimentally on\na full-size automated road vehicle that encounters a realistic pop-out\nobstacle. Contingency MPC approaches potential emergencies with safe,\nintuitive, and interpretable behavior that balances conservatism with incentive\nfor high performance operation.\n"}
{"abstract": "  We formulate a general, signature-independent form of the law of the excluded\nmiddle and prove that a logic is semisimple if and only if it enjoys this law,\nprovided that it satisfies a weak form of the so-called inconsistency lemma of\nRaftery. We then show that this equivalence can be used to provide simple\nsyntactic proofs of the theorems of Kowalski and Kracht characterizing the\nsemisimple varieties of FLew-algebras and Boolean algebras with operators, and\nto extend them to FLe-algebras and Heyting algebras with operators. Moreover,\nunder stronger assumptions this correspondence works at the level of individual\nmodels: the semisimple models of such a logic are precisely those which satisfy\nan axiomatic form of the law of the excluded middle, and a Glivenko-like\nconnection obtains between the logic and its extension by the axiom of the\nexcluded middle. This in particular subsumes the well-known Glivenko theorems\nrelating intuitionistic and classical logic and the modal logics S4 and S5. As\na consequence, we also obtain a description of the subclassical substructural\nlogics which are Glivenko related to classical logic.\n"}
{"abstract": "  Novelty Detection methods identify samples that are not representative of a\nmodel's training set thereby flagging misleading predictions and bringing a\ngreater flexibility and transparency at deployment time. However, research in\nthis area has only considered Novelty Detection in the offline setting.\nRecently, there has been a growing realization in the computer vision community\nthat applications demand a more flexible framework - Continual Learning - where\nnew batches of data representing new domains, new classes or new tasks become\navailable at different points in time. In this setting, Novelty Detection\nbecomes more important, interesting and challenging. This work identifies the\ncrucial link between the two problems and investigates the Novelty Detection\nproblem under the Continual Learning setting. We formulate the Continual\nNovelty Detection problem and present a benchmark, where we compare several\nNovelty Detection methods under different Continual Learning settings.\n  We show that Continual Learning affects the behaviour of novelty detection\nalgorithms, while novelty detection can pinpoint insights in the behaviour of a\ncontinual learner. We further propose baselines and discuss possible research\ndirections. We believe that the coupling of the two problems is a promising\ndirection to bring vision models into practice.\n"}
{"abstract": "  In this paper, we prove that the class of all special Gelfand--Dorfman\nalgebras (GD-algebras) is closed with respect to homomorphisms and thus forms a\nvariety. We also prove that every 2-dimensional GD-algebra is special. For the\nlatter, we give a technical method to find all special identities of\nGD-algebras and compute the degree 6 component of the Gr\\\"obner basis for the\nshuffle operad constructed on the symmetric operad governing the class of\nGD-algebras.\n"}
{"abstract": "  This paper uses new and recently introduced methodologies to study the\nsimilarity in the dynamics and behaviours of cryptocurrencies and equities\nsurrounding the COVID-19 pandemic. We study two collections; 45\ncryptocurrencies and 72 equities, both independently and in conjunction. First,\nwe examine the evolution of cryptocurrency and equity market dynamics, with a\nparticular focus on their change during the COVID-19 pandemic. We demonstrate\nmarkedly more similar dynamics during times of crisis. Next, we apply recently\nintroduced methods to contrast trajectories, erratic behaviours, and extreme\nvalues among the two multivariate time series. Finally, we introduce a new\nframework for determining the persistence of market anomalies over time.\nSurprisingly, we find that although cryptocurrencies exhibit stronger\ncollective dynamics and correlation in all market conditions, equities behave\nmore similarly in their trajectories, extremes, and show greater persistence in\nanomalies over time.\n"}
{"abstract": "  Intrinsic tame filling functions are quasi-isometry invariants that are\nrefinements of the intrinsic diameter function of a group. The main purpose of\nthis paper is to show that every finite presentation of a group has an\nintrinsic tame filling function that is equivalent to its intrinsic diameter\nfunction. We also introduce some alternative filling functions--based on\nconcepts similar to those used to define intrinsic tame filling functions--that\nare potential proper refinements of the intrinsic diameter function.\n"}
{"abstract": "  With the goal of making deep learning more label-efficient, a growing number\nof papers have been studying active learning (AL) for deep models. However,\nthere are a number of issues in the prevalent experimental settings, mainly\nstemming from a lack of unified implementation and benchmarking. Issues in the\ncurrent literature include sometimes contradictory observations on the\nperformance of different AL algorithms, unintended exclusion of important\ngeneralization approaches such as data augmentation and SGD for optimization, a\nlack of study of evaluation facets like the labeling efficiency of AL, and\nlittle or no clarity on the scenarios in which AL outperforms random sampling\n(RS). In this work, we present a unified re-implementation of state-of-the-art\nAL algorithms in the context of image classification via our new open-source AL\ntoolkit DISTIL, and we carefully study these issues as facets of effective\nevaluation. On the positive side, we show that AL techniques are $2\\times$ to\n$4\\times$ more label-efficient compared to RS with the use of data\naugmentation. Surprisingly, when data augmentation is included, there is no\nlonger a consistent gain in using BADGE, a state-of-the-art approach, over\nsimple uncertainty sampling. We then do a careful analysis of how existing\napproaches perform with varying amounts of redundancy and number of examples\nper class. Finally, we provide several insights for AL practitioners to\nconsider in future work, such as the effect of the AL batch size, the effect of\ninitialization, the importance of retraining the model at every round, and\nother insights.\n"}
{"abstract": "  We present an analysis of the dispersed spectral data from 11 epochs (March\n2011 to September 2018) of supernova remnant (SNR) 1987A observations performed\nwith Chandra. These observations were performed with the High Energy\nTransmission Grating (HETG) as part of our ongoing Chandra monitoring campaign\nof SNR 1987A, whose 1st-order dispersed spectrum provides a significantly\ngreater energy resolution than the previously-published 0th-order spectrum. Our\ndata sets with moderate exposure times of ~50-70ks per epoch cover the time\nperiod between deep Chandra HETG observations (with individual exposures\n>200ks) taken in March 2011 and March 2018. These data have a much higher\ncadence than the widely-spaced deep high-resolution spectra, at the expense of\ntotal exposure time. While statistical uncertainties are large due to low\nphoton count statistics in the observed 1st-order spectra, we find that\nspectral model parameters are generally in line with the shock wave propagating\ninto the medium beyond the dense inner ring, as suggested by Frank et al. 2016.\nAs the reverse shock begins ionizing the heavier elements of the supernova\nejecta interior to the equatorial ring, spectral fit parameters are expected to\nchange as the chemical makeup and physical properties of the shocked gas\nevolve. Based on our broadband spectral model fits, we find that abundance\nvalues appear to be constant in this time period. While our results are\nsomewhat limited due to photon statistics, we demonstrate the utility of the\ndispersed HETG spectral analysis that can be performed with our regular Chandra\nmonitoring observations of SNR 1987A.\n"}
{"abstract": "  In this paper we study sharp pointwise inequalities for maximal operators. In\nparticular, we strengthen DeVore's inequality for the moduli of smoothness and\na logarithmic variant of Bennett--DeVore--Sharpley's inequality for\nrearrangements. As a consequence, we improve the classical Stein--Zygmund\nembedding deriving $\\dot{B}^{d/p}_\\infty L_{p,\\infty}(\\mathbb{R}^d)\n\\hookrightarrow \\text{BMO}(\\mathbb{R}^d)$ for $1 < p < \\infty$. Moreover, these\nresults are also applied to establish new Fefferman--Stein inequalities,\nCalder\\'on--Scott type inequalities, and extrapolation estimates. Our approach\nis based on the limiting interpolation techniques.\n"}
{"abstract": "  Sparse-view computed tomography (CT) is known as a widely used approach to\nreduce radiation dose while accelerating imaging through lowered projection\nviews and correlated calculations. However, its severe imaging noise and\nstreaking artifacts turn out to be a major issue in the low dose protocol. In\nthis paper, we propose a dual-domain deep learning-based method that breaks\nthrough the limitations of currently prevailing algorithms that merely process\nsingle image slices. Since the scanned object usually contains a high degree of\nspatial continuity, the obtained consecutive imaging slices embody rich\ninformation that is largely unexplored. Therefore, we establish a cascade model\nnamed LS-AAE which aims to tackle the above problem. In addition, in order to\nadapt to the social trend of lightweight medical care, our model adopts the\ninverted residual with linear bottleneck in the module design to make it mobile\nand lightweight (reduce model parameters to one-eighth of its original) without\nsacrificing its performance. In our experiments, sparse sampling is conducted\nat intervals of 4{\\deg}, 8{\\deg} and 16{\\deg}, which appears to be a\nchallenging sparsity that few scholars have attempted before. Nevertheless, our\nmethod still exhibits its robustness and achieves the state-of-the-art\nperformance by reaching the PSNR of 40.305 and the SSIM of 0.948, while\nensuring high model mobility. Particularly, it still exceeds other current\nmethods when the sampling rate is one-fourth of them, thereby demonstrating its\nremarkable superiority.\n"}
{"abstract": "  The Boolean Hidden Matching (BHM) problem, introduced in a seminal paper of\nGavinsky et. al. [STOC'07], has played an important role in the streaming lower\nbounds for graph problems such as triangle and subgraph counting, maximum\nmatching, MAX-CUT, Schatten $p$-norm approximation, maximum acyclic subgraph,\ntesting bipartiteness, $k$-connectivity, and cycle-freeness. The one-way\ncommunication complexity of the Boolean Hidden Matching problem on a universe\nof size $n$ is $\\Theta(\\sqrt{n})$, resulting in $\\Omega(\\sqrt{n})$ lower bounds\nfor constant factor approximations to several of the aforementioned graph\nproblems. The related (and, in fact, more general) Boolean Hidden Hypermatching\n(BHH) problem introduced by Verbin and Yu [SODA'11] provides an approach to\nproving higher lower bounds of $\\Omega(n^{1-1/t})$ for integer $t\\geq 2$.\nReductions based on Boolean Hidden Hypermatching generate distributions on\ngraphs with connected components of diameter about $t$, and basically show that\nlong range exploration is hard in the streaming model of computation with\nadversarial arrivals.\n  In this paper we introduce a natural variant of the BHM problem, called noisy\nBHM (and its natural noisy BHH variant), that we use to obtain higher than\n$\\Omega(\\sqrt{n})$ lower bounds for approximating several of the aforementioned\nproblems in graph streams when the input graphs consist only of components of\ndiameter bounded by a fixed constant. We also use the noisy BHM problem to show\nthat the problem of classifying whether an underlying graph is isomorphic to a\ncomplete binary tree in insertion-only streams requires $\\Omega(n)$ space,\nwhich seems challenging to show using BHM or BHH alone.\n"}
{"abstract": "  The aim of this paper is to introduce a translation operator associated to\nthe canonical Fourier Bessel transform $\\mathcal{F}_{\\nu}^{\\mathbf{m}}$ and\nstudy some of the important properties. We derive a convolution product for\nthis transform and as application we study the heat equation and the heat\nsemigroup related to $\\Delta_{\\nu}^{\\mathbf{m}}.$\n"}
{"abstract": "  Naturally occurring radionuclide $^{210}$Pb ($T_{1/2}$=22.2 y) is an\nimportant source of background in rare event searches, such as neutrinoless\ndouble-$\\beta$ decay and dark matter direct detection experiments. When a\nsample mass of hundreds of grams is available, $\\gamma$-counting measurements\ncan be performed. However, there are other cases where only grams of sample can\nbe used. For these cases, better sensitivities are required.\n  In this paper, in collaboration with the Astroparticle Physics group at\nCarleton University, the capabilities of the A.E. Lalonde AMS Laboratory at the\nUniversity of Ottawa for $^{210}$Pb measurements are discussed. PbF$_{2}$ and\nPbO targets were used, selecting in the low energy sector, respectively,\n(PbF$_{3}$)$^{-}$ or (PbO$_{2}$)$^{-}$ ions.\n  For fluoride targets, the blank $^{210}$Pb/$^{206}$Pb ratio was in the\n10$^{-14}$ to 10$^{-13}$ range, but current output was lower and less stable.\nFor oxide targets, current output showed better stability, despite a\nsignificant difference in current output for commercial PbO and processed\nsamples, and background studies suggested a background not much higher than\nthat of the fluoride targets. Both target materials showed, therefore, good\nperformance for $^{210}$Pb AMS assay.\n  Measurements of Kapton films, an ultra-thin polymer material, where masses\navailable are typically just several grams, were performed. 90% C.L. upper\nlimits for the $^{210}$Pb specific activity in the range of 0.74-2.8 Bq/kg were\nestablished for several Kapton HN films.\n"}
{"abstract": "  This paper builds the clustering model of measures of market microstructure\nfeatures which are popular in predicting stock returns. In a 10-second\ntime-frequency, we study the clustering structure of different measures to find\nout the best ones for predicting. In this way, we can predict more accurately\nwith a limited number of predictors, which removes the noise and makes the\nmodel more interpretable.\n"}
{"abstract": "  ARES (Accelerator Research Experiment at SINBAD) is a linear accelerator at\nthe SINBAD (Short INnovative Bunches and Accelerators at DESY) facility at\nDESY. ARES was designed to combine reproducible beams from conventional\nRF-based accelerator technology with novel but still experimental acceleration\ntechniques. It aims to produce high brightness ultra-short electron bunches in\nthe range of sub fs to few fs, at a beam energy of 100-150 MeV, suitable for\ninjection into novel acceleration experiments like Dielectric Laser\nAcceleration (DLA) and Laser driven Plasma Acceleration (LPA). This paper\nreports the conceptual design and simulations of a final focus system for\ninjecting into a LPA experiment at ARES, including permanent magnetic\nquadrupoles (PMQ), sufficient longitudinal space for collinear laser and\nelectron transport, space for required diagnostics and a LPA setup.\nSpace-charge effects play a significant role and are included. Simulation\nresults on the focusing of the ARES electron bunches into and their transport\nthrough the laser-driven plasma are presented. The effects of several errors\nhave been simulated and are reported.\n"}
{"abstract": "  We propose UCBMQ, Upper Confidence Bound Momentum Q-learning, a new algorithm\nfor reinforcement learning in tabular and possibly stage-dependent, episodic\nMarkov decision process. UCBMQ is based on Q-learning where we add a momentum\nterm and rely on the principle of optimism in face of uncertainty to deal with\nexploration. Our new technical ingredient of UCBMQ is the use of momentum to\ncorrect the bias that Q-learning suffers while, at the same time, limiting the\nimpact it has on the second-order term of the regret. For UCBMQ , we are able\nto guarantee a regret of at most $O(\\sqrt{H^3SAT}+ H^4 S A )$ where $H$ is the\nlength of an episode, $S$ the number of states, $A$ the number of actions, $T$\nthe number of episodes and ignoring terms in poly$log(SAHT)$. Notably, UCBMQ is\nthe first algorithm that simultaneously matches the lower bound of\n$\\Omega(\\sqrt{H^3SAT})$ for large enough $T$ and has a second-order term (with\nrespect to the horizon $T$) that scales only linearly with the number of states\n$S$.\n"}
{"abstract": "  In this work we present the design decisions and advantages for accomplishing\ncross mesh format checkpoint-restart in community code FLASH-X. AMReX and\nParamesh are the two AMR mesh formats developed and supported by FLASH-X. We\nalso highlight strong and weak scaling study of existing HDF5 I/O checkpoint\nwriting along with new ideas and results (presented during talk) for utilizing\nheterogeneous compute architectures for improved I/O performance.\n"}
{"abstract": "  The ability of an autonomous vehicle to perform 3D tracking is essential for\nsafe planing and navigation in cluttered environments. The main challenges for\nmulti-object tracking (MOT) in autonomous driving applications reside in the\ninherent uncertainties regarding the number of objects, when and where the\nobjects may appear and disappear, and uncertainties regarding objects' states.\nRandom finite set (RFS) based approaches can naturally model these\nuncertainties accurately and elegantly, and they have been widely used in\nradar-based tracking applications. In this work, we developed an RFS-based MOT\nframework for 3D LiDAR data. In partiuclar, we propose a Poisson\nmulti-Bernoulli mixture (PMBM) filter to solve the amodal MOT problem for\nautonomous driving applications. To the best of our knowledge, this represents\na first attempt for employing an RFS-based approach in conjunction with 3D\nLiDAR data for MOT applications with comprehensive validation using challenging\ndatasets made available by industry leaders. The superior experimental results\nof our PMBM tracker on public Waymo and Argoverse datasets clearly illustrate\nthat an RFS-based tracker outperforms many state-of-the-art deep learning-based\nand Kalman filter-based methods, and consequently, these results indicate a\ngreat potential for further exploration of RFS-based frameworks for 3D MOT\napplications.\n"}
{"abstract": "  A toroidal dipole moment appears independent of the electric and magnetic\ndipole moment in the multipole expansion of electrodynamics. It arises\nnaturally from vortex-like arrangements of spins. Observing and controlling\nspontaneous long-range orders of toroidal moments are highly promising for\nspintronics but remain challenging. Here we demonstrate that a vortex-like spin\nconfiguration with a staggered arrangement of toroidal moments, a ferritoroidal\nstate, is realized in a chiral triangular-lattice magnet BaCoSiO4. Upon\napplying a magnetic field, we observe multi-stair toroidal transitions\ncorrelating directly with metamagnetic transitions. We establish a\nfirst-principles microscopic Hamiltonian that explains both the formation of\ntoroidal states and the metamagnetic toroidal transition as a combined effect\nof the magnetic frustration and the Dzyaloshinskii-Moriya interactions allowed\nby the crystallographic chirality in BaCoSiO4.\n"}
{"abstract": "  In this report, we introduce the technical details of our submission to the\nVIPriors object detection challenge. Our solution is based on mmdetction of a\nstrong baseline open-source detection toolbox. Firstly, we introduce an\neffective data augmentation method to address the lack of data problem, which\ncontains bbox-jitter, grid-mask, and mix-up. Secondly, we present a robust\nregion of interest (ROI) extraction method to learn more significant ROI\nfeatures via embedding global context features. Thirdly, we propose a\nmulti-model integration strategy to refinement the prediction box, which\nweighted boxes fusion (WBF). Experimental results demonstrate that our approach\ncan significantly improve the average precision (AP) of object detection on the\nsubset of the COCO2017 dataset.\n"}
{"abstract": "  In this work, we present an in-depth and systematic analysis using tools such\nas local interpretable model-agnostic explanations (LIME) (arXiv:1602.04938)\nand divergence measures to analyze what changes lead to improvement in\nperformance in fine tuned models for synthetic aperture sonar (SAS) data. We\nexamine the sensitivity to factors in the fine tuning process such as class\nimbalance. Our findings show not only an improvement in seafloor texture\nclassification, but also provide greater insight into what features play\ncritical roles in improving performance as well as a knowledge of the\nimportance of balanced data for fine tuning deep learning models for seafloor\nclassification in SAS imagery.\n"}
{"abstract": "  Improvement of the classical gravity with the running gravitational coupling\nobtained from asymptotically safe gravity, is a good way of considering the\neffects of quantum gravity. This is usually done for metric theories of\ngravity. Here we investigate the effects of such an improvement for pure affine\ntheories of gravity. To motivate the approach, we first consider the effects of\nquantum improvement on the connection using metric theory and investigate the\neffects on the causal structure of black hole solution. Next in the framework\nof Schr\\\"odinger-Eddington affine theory, the general way of affine improvement\nis presented and a spherically symmetric solution is obtained and compared with\nother ways of improvement.\n"}
{"abstract": "  Recently, there has been a rising surge of momentum for deep representation\nlearning in hyperbolic spaces due to theirhigh capacity of modeling data like\nknowledge graphs or synonym hierarchies, possessing hierarchical structure. We\nrefer to the model as hyperbolic deep neural network in this paper. Such a\nhyperbolic neural architecture potentially leads to drastically compact model\nwithmuch more physical interpretability than its counterpart in Euclidean\nspace. To stimulate future research, this paper presents acoherent and\ncomprehensive review of the literature around the neural components in the\nconstruction of hyperbolic deep neuralnetworks, as well as the generalization\nof the leading deep approaches to the Hyperbolic space. It also presents\ncurrent applicationsaround various machine learning tasks on several publicly\navailable datasets, together with insightful observations and identifying\nopenquestions and promising future directions.\n"}
{"abstract": "  If a binary liquid mixture, composed of two alternative species with equal\namounts, is quenched from a high temperature to a low temperature, below the\ncritical point of demixing, then the mixture will phase separate through a\nprocess known as spinodal decomposition. However, if the two alternative\nspecies are allowed to interconvert, either naturally (e.g. the equilibrium\ninterconversion of enantiomers) or forcefully (e.g. via an external source of\nenergy or matter), then the process of phase separation may drastically change.\nIn this case, depending on the nature of interconversion, two phenomena could\nbe observed: either phase amplification, the growth of one phase at the expense\nof another stable phase, or microphase separation, the formation of nongrowing\n(steady-state) microphase domains. In this work, we generalize the\nCahn-Hilliard theory of spinodal decomposition to include molecular\ninterconversion of species and describe the physical properties of systems\nundergoing either phase amplification or microphase separation. We apply the\ndeveloped theory to describe the simulation results of three atomistic models\nwhich demonstrate phase amplification and/or microphase separation. We also\ndiscuss the application of our approach to phase transitions in polyamorphic\nliquids. Lastly, we describe the effects of fluctuations of the order parameter\nin the critical region on phase amplification and microphase separation.\n"}
{"abstract": "  I consider higher-order terms of the Seeley expansion of the heat kernel,\nwhich for smooth metrics are suppressed as inverse powers of the UV cutoff\n$\\Lambda$, and demonstrate how they result in an anomalous contribution to the\nstring effective action after doing uncertainties $\\Lambda^{-2}\\times\n\\Lambda^2$. For the Polyakov string these anomalies precisely reproduce at one\nloop the result of KPZ-DDK obtained for the Liouville theory by the conformal\nfield theory technique. For the Nambu-Goto string I find a deviation from this\nresult which shows that the two string formulations may differ.\n"}
{"abstract": "  Proposal of large-scale datasets has facilitated research on deep neural\nmodels for news summarization. Deep learning can also be potentially useful for\nspoken dialogue summarization, which can benefit a range of real-life scenarios\nincluding customer service management and medication tracking. To this end, we\npropose DialogSum, a large-scale labeled dialogue summarization dataset. We\nconduct empirical analysis on DialogSum using state-of-the-art neural\nsummarizers. Experimental results show unique challenges in dialogue\nsummarization, such as spoken terms, special discourse structures, coreferences\nand ellipsis, pragmatics and social common sense, which require specific\nrepresentation learning technologies to better deal with.\n"}
{"abstract": "  In this work we analyse the stability of self gravitating spheres in the\ncontext of gravitational cracking. Besides exploring the role played by the\nanisotropy in the occurrence of cracking, we also study the effect of the\ncomplexity factor recently introduced by L. Herrera in Phys. Rev. D 97, 044010\n(2018). The models under study correspond to anisotropic solutions obtained in\nthe framework of the Gravitational Decoupling. The effect that the variation of\nthe decoupling parameter and the compactness of the source have on the\nbehaviour of the radial force is studied in detail.\n"}
{"abstract": "  Let $R$ be a commutative ring with identity and $M$ be a unitary $R$-module.\nBy $J(R),$ we denote the Jacobson radical of $R$. The purpose of this paper is\nto introduce the concept of weakly $J$-submodules generalizing $J$-submodules.\nWe call a proper submodule $N$ of $M$ a weakly $J$-submodule if whenever $0\\neq\nrm\\in N$ for $r\\in R$ and $m\\in M$, then $r\\in(J(R)M:M)$ or $m\\in N.$ Various\nproperties and characterizations of weakly $J$-submodules are investigated\nespecially in the case of multiplication modules.\n"}
{"abstract": "  The BFV-BRST Hamiltonian quantization method is presented for the theories\nwhere the gauge parameters are restricted by differential equations. The\ngeneral formalism is exemplified by the Maxwell-like theory of symmetric tensor\nfield.\n"}
{"abstract": "  Synchronization in neural system plays an important role in many brain\nfunctions. Synchronization in the gamma frequency band (30Hz-100Hz) is involved\nin a variety of cognitive phenomena; abnormalities of the gamma synchronization\nare found in schizophrenia and autism spectrum disorder. Frequently, the\nstrength of synchronization is not very high and is intermittent even on short\ntime scales (a few cycles of oscillations). That is, the network exhibits\nintervals of synchronization followed by intervals of desynchronization. Neural\ncircuits dynamics may show different distributions of desynchronization\ndurations even if the synchronization strength is fixed. In this study, we use\na conductance-based neural network exhibiting pyramidal-interneuron (PING)\ngamma rhythm to study the temporal patterning of synchronized neural\noscillations. We found that changes in the synaptic strength (as well as\nchanges in the membrane kinetics) can alter the temporal patterning of\nsynchrony. Moreover, we found that the changes in the temporal pattern of\nsynchrony may be independent of the changes in the average synchrony strength.\nEven though the temporal patterning may vary, there is a tendency for dynamics\nwith short (although potentially numerous) desynchronizations, similar to what\nwas observed in experimental studies of neural activity synchronization in the\nbrain. Recent studies suggested that the short desynchronizations dynamics may\nfacilitate the formation and the break-up of transient neural assemblies. Thus,\nthe results of this study suggest that changes of synaptic strength may alter\nthe temporal patterning of the gamma synchronization as to make the neural\nnetworks more efficient in the formation of neural assemblies and the\nfacilitation of cognitive phenomena.\n"}
{"abstract": "  We propose a pseudo-Goldstone boson dark matter (pGDM) particle in $SO(10)$\ngrand unified theory (GUT). Due to its Goldstone nature, this pGDM evades the\ndirect DM detection experiments which, otherwise, severely constrain the\nparameter space of DM models. In $SO(10)$, the pGDM is embedded as a linear\ncombination of the Standard Model (SM) singlet scalars in ${\\bf 16_H}$ and\n${\\bf 126_H}$ representations. We consider two scenarios for the intermediate\nroute of $SO(10)$ symmetry breaking (SB) to the SM: $SU(5) \\times U(1)_X$ and\nPati-Salam the $SU(4)_c \\times SU(2)_L \\times SU(2)_R$ (4-2-2) gauge groups.\nThe vacuum expectation value of ${\\bf 126_H}$, which triggers the breaking of\n$U(1)_X$ and 4-2-2 symmetry in the two scenarios, respectively, determines the\npGDM lifetime whose astrophysical lower bound provides one of the most\nstringent constraints. For the 4-2-2 route to $SO(10)$, the successful SM gauge\ncoupling unification requires the 4-2-2 breaking scale to be ${\\cal O}\n(10^{11})$ GeV, and most of the parameter space is excluded. For the $SU(5)\n\\times U(1)_X$ route, on the other hand, the $U(1)_X$ breaking scale can be\nsignificantly higher, and a wide range of the parameter space is allowed.\nFurthermore, the proton lifetime in the $SU(5)$ case is predicted to be $4.53\n\\times 10^{34}$ years, which lies well within the sensitivity reach of the\nHyper-Kamiokande experiment. We also examine the constraints on the model\nparameter space from the Large Hadron Collider and the indirect DM search by\nFermi-LAT and MAGIC experiments.\n"}
{"abstract": "  We derive simple relations which express 2D light front force distributions\nin terms of 3D Breit frame pressure and shear force distributions.\nMathematically the relations correspond to invertible Abel transformation and\nthey establish one-to-one mathematical equivalence of 3D Breit frame force\ndistributions and 2D light front ones.\n  Any knowledge (model calculation, experimental measurement, etc.) about\npressure and shear force distributions in Breit frame can be unambiguously\ntransformed into light front force distributions with the help of Abel\ntransformation. It is important that the transformation ensures 2D stability\nconditions if the 3D stability conditions are satisfied.\n  As an illustration of how the relations work, we calculated the light front\nforce distributions for a large nucleus as a liquid drop, and for large $N_c$\nnucleon as a chiral soliton.\n"}
{"abstract": "  In the standard Black-Scholes-Merton framework, dividends are represented as\na continuous dividend yield and the pricing of Vanilla options on a stock is\nachieved through the well-known Black-Scholes formula. In reality however,\nstocks pay a discrete fixed cash dividend at each dividend ex-date. This leads\nto the so-called piecewise lognormal model, where the asset jumps from a fixed\nknown amount at each dividend date. There is however no exact closed-form\nformula for the pricing of Vanilla options under this model. Approximations\nmust be used. While there exists many approximations taylored to this specific\nproblem in the litterature, this paper explores the use of existing well-known\nbasket option formulas for the pricing of European options on a single asset\nwith cash dividends in the piecewise lognormal model.\n"}
{"abstract": "  Presidential speeches indicate the government's intentions and justifications\nsupported by a dedicated style and rhetoric oscillating between explanation and\ncontroversy. Over a period of sixty years, can we observe stylistic variations\nby the different French presidents of the Fifth Republic (1958-2018)? Based on\nofficial transcripts of all their allocution, this paper illustrates the\nstylistic evolution and presents the underlying main trends. This study shows\nthat de Gaulle's rhetoric is not mainly dedicated to his own person, or that\nthe two terms of J. Chirac are not fully similar. According to several overall\nstylistic indicators, Macron's style does not appear as complex compared to his\npredecessors (F. Hollande or N. Sarkozy) but a more careful analysis clearly\ndemonstrates his noticeable new style. Compared to the recent US presidents,\nthe French ones present some similarities (e.g., similar mean sentence length)\nand dissimilarities (more I-words, less we-words). In this comparative\nanalysis, Macron's style is also clearly distinctive from both the US and\nformer French presidents. Opting for a more abstract discourse, less anchored\nin space, using less numbers, E. Macron tends to use long sentences. These\nvarious stylistic and rhetorical features could explain his being misunderstood\nby the French people and his recurrent low approval ratings.\n"}
{"abstract": "  Background: Cardiovascular diseases (CVDs) are among the leading causes of\ndeath worldwide. Predictive scores providing personalised risk of developing\nCVD are increasingly used in clinical practice. Most scores, however, utilise a\nhomogenous set of features and require the presence of a physician.\n  Objective: The aim was to develop a new risk model (DiCAVA) using statistical\nand machine learning techniques that could be applied in a remote setting. A\nsecondary goal was to identify new patient-centric variables that could be\nincorporated into CVD risk assessments.\n  Methods: Across 466,052 participants, Cox proportional hazards (CPH) and\nDeepSurv models were trained using 608 variables derived from the UK Biobank to\ninvestigate the 10-year risk of developing a CVD. Data-driven feature selection\nreduced the number of features to 47, after which reduced models were trained.\nBoth models were compared to the Framingham score.\n  Results: The reduced CPH model achieved a c-index of 0.7443, whereas DeepSurv\nachieved a c-index of 0.7446. Both CPH and DeepSurv were superior in\ndetermining the CVD risk compared to Framingham score. Minimal difference was\nobserved when cholesterol and blood pressure were excluded from the models\n(CPH: 0.741, DeepSurv: 0.739). The models show very good calibration and\ndiscrimination on the test data.\n  Conclusion: We developed a cardiovascular risk model that has very good\npredictive capacity and encompasses new variables. The score could be\nincorporated into clinical practice and utilised in a remote setting, without\nthe need of including cholesterol. Future studies will focus on external\nvalidation across heterogeneous samples.\n"}
{"abstract": "  There is a growing set of observational data demonstrating that cosmic rays\nexhibit small-scale anisotropies (5-30 deg) with amplitude deviations lying\nbetween 0.01-0.1 percent that of the average cosmic ray flux. A broad range of\nmodels have been proposed to explain these anisotropies ranging from\nfinite-scale magnetic field structures to dark matter annihilation. The\nstandard diffusion transport methods used in cosmic ray propagation do not\ncapture the transport physics in a medium with finite-scale or coherent\nmagnetic field structures. Here, we present a Monte Carlo transport method,\napplying it to a series of finite-scale magnetic field structures to determine\nthe requirements of such fields in explaining the observed cosmic\nray,small-scale anisotropies.\n"}
{"abstract": "  As a representative device exploiting both the solar energy and the radiative\ncooling of deep-sky, the radiative temperature regulator (RTR) could switch\nbetween heating and cooling modes self-adaptively at different temperatures.\nHowever, the concept of RTR is challenging to be implemented due to the intense\nparasitic absorption in phase-changing layers. Here, based on the theoretical\nframework of energy conservation, we quantitatively reveal the intrinsic\nrelationships between solar heating and radiative cooling, especially\naddressing the fundamental limiting factors, including the parasitic absorption\nand the spectral emission selectivity, as well as the dynamic responses of the\nphase-changing device under various operating conditions. The investigation\npresents more insight into the underlying physics of RTRs and provides feasible\narchitectures for realizing such a kind of new functional device.\n"}
{"abstract": "  We develop numerical methods for computing statistics of stochastic processes\non surfaces of general shape with drift-diffusion dynamics $d\\mathbf{X}_t =\na(\\mathbf{X}_t)dt + \\mathbf{b}(\\mathbf{X}_t)d\\mathbf{W}_t$. We formulate\ndescriptions of Brownian motion and general drift-diffusion processes on\nsurfaces. We consider statistics of the form $u(\\mathbf{x}) =\n\\mathbb{E}^{\\mathbf{x}}\\left[\\int_0^\\tau g(\\mathbf{X}_t)dt \\right] +\n\\mathbb{E}^{\\mathbf{x}}\\left[f(\\mathbf{X}_\\tau)\\right]$ for a domain $\\Omega$\nand the exit stopping time $\\tau = \\inf_t \\{t > 0 \\; |\\; \\mathbf{X}_t \\not\\in\n\\Omega\\}$, where $f,g$ are general smooth functions. For computing these\nstatistics, we develop high-order Generalized Moving Least Squares (GMLS)\nsolvers for associated surface PDE boundary-value problems based on\nBackward-Kolmogorov equations. We focus particularly on the mean First Passage\nTimes (FPTs) given by the case $f = 0,\\, g = 1$ where $u(\\mathbf{x}) =\n\\mathbb{E}^{\\mathbf{x}}\\left[\\tau\\right]$. We perform studies for a variety of\nshapes showing our methods converge with high-order accuracy both in capturing\nthe geometry and the surface PDE solutions. We then perform studies showing how\nstatistics are influenced by the surface geometry, drift dynamics, and\nspatially dependent diffusivities.\n"}
{"abstract": "  Starting from some results regarding the form of the Ricci scalar at a point\n$P$ in a spacetime endowed with a minimum distance, we investigate how they\nmight be accommodated, specifically for the case of null separations, in a\nas-simple-as-possible quantum structure for spacetime at $P$, and we try to\naccomplish this in terms of potentially operationally-defined concepts. In so\ndoing, we provide a possible explicit form for the operator expressing the\nRicci scalar as a quantum observable, and give quantum-informational support,\nthus regardless of or before field equations, to associating with a patch of\nhorizon an entropy proportional to its area.\n"}
{"abstract": "  The importance of a node in a social network is identified through a set of\nmeasures called centrality. Degree centrality, closeness centrality,\nbetweenness centrality and clustering coefficient are the most frequently used\nmetrics to compute node centrality. Their computational complexity in some\ncases makes unfeasible, when not practically impossible, their computations.\nFor this reason we focused on two alternative measures, WERW-Kpath and Game of\nThieves, which are at the same time highly descriptive and computationally\naffordable. Our experiments show that a strong correlation exists between\nWERW-Kpath and Game of Thieves and the classical centrality measures. This may\nsuggest the possibility of using them as useful and more economic replacements\nof the classical centrality measures.\n"}
{"abstract": "  We present a new method to assess the properties of transiting planet\ncandidates by multicolor photometry. By analyzing multicolor transit/eclipse\nlight curves and apparent magnitudes of the target in parallel, this method\nattempts to identify the nature of the system and provide a quantitative\nconstraint on the properties of unresolved companion(s). We demonstrate our\nmethod by observing the six systems hosting candidate transiting planets,\nidentified by the K2 mission (EPIC 206036749, EPIC 206500801, EPIC 210513446,\nEPIC 211800191, EPIC 220621087, and EPIC 220696233). Applying our analysis code\nto the six targets, we find that EPIC 206036749, EPIC 210513446, and EPIC\n211800191 are likely to be triple-star systems including eclipsing binaries,\nand EPIC 220696233 is likely a planetary system, albeit further observations\nare required to confirm the nature. Additionally, we confirm that the\nsystematic errors in the derived system parameters arising from adopting\nspecific isochrone models and observing instruments (passbands) are relatively\nsmall. While this approach alone is not powerful enough to validate or refute\nplanet candidates, the technique allows us to constrain the properties of\nresolved/unresolved companions, and prioritize the planet candidates for\nfurther follow-up observations (e.g., radial-velocity measurements).\n"}
{"abstract": "  We introduce a new topological quantum code, the three-dimensional subsystem\ntoric code (3D STC), which is a generalization of the stabilizer toric code.\nThe 3D STC can be realized by measuring geometrically-local parity checks of\nweight at most three on the cubic lattice with open boundary conditions. We\nprove that single-shot quantum error correction (QEC) is possible with the 3D\nSTC, i.e., one round of local parity-check measurements suffices to perform\nreliable QEC even in the presence of measurement errors. We also propose an\nefficient single-shot QEC strategy for the 3D STC and investigate its\nperformance. In particular, we numerically estimate the resulting storage\nthreshold against independent bit-flip, phase-flip and measurement errors to be\n$p_\\text{STC} \\approx 1.045\\%$. Such a high threshold together with local\nparity-check measurements of small weight make the 3D STC particularly\nappealing for realizing fault-tolerant quantum computing.\n"}
{"abstract": "  We study the problem of designing systems in order to minimize cost while\nmeeting a given flexibility target. Flexibility is attained by enforcing a\njoint chance constraint, which ensures that the system will exhibit feasible\noperation with a given target probability level. Unfortunately, joint chance\nconstraints are complicated mathematical objects that often need to be\nreformulated using mixed-integer programming (MIP) techniques. In this work, we\ncast the design problem as a conflict resolution problem that seeks to minimize\ncost while maximizing flexibility. We propose a purely continuous relaxation of\nthis problem that provides a significantly more scalable approach relative to\nMIP methods and show that the formulation delivers solutions that closely\napproximate the Pareto set of the original joint chance-constrained problem.\n"}
{"abstract": "  In this work, we investigate the topological phase transitions in an\neffective model for a topological thin film with high-frequency pumping. In\nparticular, our results show that the circularly polarized light can break the\ntime-reversal symmetry and induce the quantum anomalous Hall insulator (QAHI)\nphase. Meanwhile, the bulk magnetic moment can also break the time-reversal\nsymmetry. Therefore, it shows rich phase diagram by tunning the intensity of\nthe light and the thickness of the thin film. Using the parameters fitted by\nexperimental data, we give the topological phase diagram of the Cr-doped\nBi$_{2}$Se$_{3}$ thin film, showing that by modulating the strength of the\npolarized optical field in an experimentally accessible range, there are four\ndifferent phases: the normal insulator phase, the time-reversal-symmetry-broken\nquantum spin Hall insulator phase, and two different QAHI phases with opposite\nChern numbers. Comparing with the non-doped Bi$_{2}$Se$_{3}$, it is found that\nthe interplay between the light and bulk magnetic moment separates the two\ndifferent QAHI phases with opposite Chern numbers. The results show that an\nintrinsic magnetic topological insulator with high-frequency pumping is an\nideal platform for further exploring various topological phenomena with a\nspontaneously broken time-reversal symmetry.\n"}
{"abstract": "  Consistent inference of the electron density and temperature has been carried\nout with multiple heterogeneous plasma diagnostic data sets at Wendelstein 7-X.\nThe predictive models of the interferometer, Thomson scattering and helium beam\nemission spectroscopy systems have been developed in the Minerva framework and\ncombined to a single joint model. The electron density and temperature profiles\nare modelled by Gaussian processes with their hyperparameters. The model\nparameters such as the calibration factor of the Thomson scattering system and\nthe model predictive uncertainties are regarded as additional unknown\nparameters. The joint posterior probability distribution of the electron\ndensity and temperature profiles, hyperparameters of the Gaussian processes and\nmodel parameters is explored by Markov chain Monte Carlo algorithms. The\nposterior samples drawn from the joint posterior distribution are numerically\nmarginalised over the hyperparameters and model parameters to obtain the\nmarginal posterior distributions of the electron density and temperature\nprofiles. The inference of these profiles is performed with different\ncombinations of the interferometer and Thomson scattering data as well as\neither the empirical electron density and temperature constraints at the\nlimiter/divertor positions introduced by \\textit{virtual observations} or the\nedge density and temperature from the helium beam emission data. Furthermore,\nthe addition of the X-ray imaging crystal spectrometers to the joint model for\nthe ion temperature profiles is demonstrated. All these profiles presented in\nthis work are inferred with the optimal hyperparameters and model parameters by\nexploring the full joint posterior distribution which intrinsically embodies\nBayesian Occam's razor.\n"}
{"abstract": "  In this paper, a deep collocation method (DCM) for thin plate bending\nproblems is proposed. This method takes advantage of computational graphs and\nbackpropagation algorithms involved in deep learning. Besides, the proposed DCM\nis based on a feedforward deep neural network (DNN) and differs from most\nprevious applications of deep learning for mechanical problems. First, batches\nof randomly distributed collocation points are initially generated inside the\ndomain and along the boundaries. A loss function is built with the aim that the\ngoverning partial differential equations (PDEs) of Kirchhoff plate bending\nproblems, and the boundary/initial conditions are minimised at those\ncollocation points. A combination of optimizers is adopted in the\nbackpropagation process to minimize the loss function so as to obtain the\noptimal hyperparameters. In Kirchhoff plate bending problems, the C1 continuity\nrequirement poses significant difficulties in traditional mesh-based methods.\nThis can be solved by the proposed DCM, which uses a deep neural network to\napproximate the continuous transversal deflection, and is proved to be suitable\nto the bending analysis of Kirchhoff plate of various geometries.\n"}
{"abstract": "  Our recent work (Ayral et al., 2020 IEEE Computer Society Annual Symposium on\nVLSI (ISVLSI)) showed the first implementation of the Quantum Divide and\nCompute (QDC) method, which allows to break quantum circuits into smaller\nfragments with fewer qubits and shallower depth. QDC can thus deal with the\nlimited number of qubits and short coherence times of noisy, intermediate-scale\nquantum processors. This article investigates the impact of different noise\nsources -- readout error, gate error and decoherence -- on the success\nprobability of the QDC procedure. We perform detailed noise modeling on the\nAtos Quantum Learning Machine, allowing us to understand tradeoffs and\nformulate recommendations about which hardware noise sources should be\npreferentially optimized. We describe in detail the noise models we used to\nreproduce experimental runs on IBM's Johannesburg processor. This work also\nincludes a detailed derivation of the equations used in the QDC procedure to\ncompute the output distribution of the original quantum circuit from the output\ndistribution of its fragments. Finally, we analyze the computational complexity\nof the QDC method for the circuit under study via tensor-network\nconsiderations, and elaborate on the relation the QDC method with\ntensor-network simulation methods.\n"}
{"abstract": "  Graph signal processing (GSP) is a key tool for satisfying the growing demand\nfor information processing over networks. However, the success of GSP in\ndownstream learning and inference tasks is heavily dependent on the prior\nidentification of the relational structures. Graphs are natural descriptors of\nthe relationships between entities of complex environments. The underlying\ngraph is not readily detectable in many cases and one has to infer the topology\nfrom the observed signals. Firstly, we address the problem of graph signal\nclassification by proposing a novel framework for discriminative graph\nlearning. To learn discriminative graphs, we invoke the assumption that signals\nbelonging to each class are smooth with respect to the corresponding graph\nwhile maintaining non-smoothness with respect to the graphs corresponding to\nother classes. Secondly, we extend our work to tackle increasingly dynamic\nenvironments and real-time topology inference. We develop a proximal gradient\n(PG) method which can be adapted to situations where the data are acquired\non-the-fly. Beyond discrimination, this is the first work that addresses the\nproblem of dynamic graph learning from smooth signals where the sought network\nalters slowly. The validation of the proposed frameworks is comprehensively\ninvestigated using both synthetic and real data.\n"}
{"abstract": "  The rapid growth in Mobile application users has made the researchers and\npractitioners to think of intelligent tools that can help the users and\napplications in delivering quality of services. Therefore, intelligent agent is\nexpected to become the tool for development of mobile outcome based education\n(OBE) particularly in higher learning Institutions (HLI). In this context,\nthere is a lacking of OBE intelligent agent in assisting the academicians to\nuse in OBE management for mobile application system environment. This paper\npresents the conceptual design and development of a mobile intelligent agent\nbased on mobile OBE called as i-MOBE. Since that, i-MOBE that we developed is\nconsidered very important for academicians and students to facilitate them in\nusing for academic purpose in HLI particularly in helping them to monitor the\nperformance in teaching and learning (T&L). The system architecture will be\ncovering the conceptual design and its interaction as well as the system\nconfiguration in helping the academicians to use the system in their T&L toward\neffective and efficiency also can be applied in monitoring based on scenarios\nsuch as test, assignment and projects and so on.\n"}
{"abstract": "  The continuum-fitting and the iron-line methods are currently the two leading\ntechniques for measuring the spins of accreting black holes. In the past few\nyears, these two methods have been developed for testing fundamental physics.\nIn the present work, we employ state-of-the-art models to test black holes\nthrough the continuum-fitting and the iron-line methods and we analyze three\nNuSTAR observations of the black hole binary GRS 1716-249 during its outburst\nin 2016-2017. In these three observations, the source was in a\nhard-intermediate state and the spectra show both a strong thermal component\nand prominent relativistic reflection features. Our analysis confirms the Kerr\nnature of the black hole in GRS 1716-249 and provides quite stringent\nconstraints on possible deviations from the predictions of general relativity.\n"}
{"abstract": "  This paper presents the first general (supervised) statistical learning\nframework for point processes in general spaces. Our approach is based on the\ncombination of two new concepts, which we define in the paper: i) bivariate\ninnovations, which are measures of discrepancy/prediction-accuracy between two\npoint processes, and ii) point process cross-validation (CV), which we here\ndefine through point process thinning. The general idea is to carry out the\nfitting by predicting CV-generated validation sets using the corresponding\ntraining sets; the prediction error, which we minimise, is measured by means of\nbivariate innovations. Having established various theoretical properties of our\nbivariate innovations, we study in detail the case where the CV procedure is\nobtained through independent thinning and we apply our statistical learning\nmethodology to three typical spatial statistical settings, namely parametric\nintensity estimation, non-parametric intensity estimation and Papangelou\nconditional intensity fitting. Aside from deriving theoretical properties\nrelated to these cases, in each of them we numerically show that our\nstatistical learning approach outperforms the state of the art in terms of mean\n(integrated) squared error.\n"}
{"abstract": "  Periodically poled thin-film lithium niobate (TFLN) waveguides have emerged\nas a leading platform for highly efficient frequency conversion in the\nnear-infrared. However, the commonly used silica bottom-cladding results in\nhigh absorption loss at wavelengths beyond 2.5 $\\mu$m. In this work, we\ndemonstrate efficient frequency conversion in a TFLN-on-sapphire platform,\nwhich features high transparency up to 4.5 $\\mu$m. In particular, we report\ngenerating mid-infrared light up to 3.66 $\\mu$m via difference-frequency\ngeneration of a fixed 1-$\\mu$m source and a tunable telecom source, with\nnormalized efficiencies up to 200%/W-cm$^2$. These results show\nTFLN-on-sapphire to be a promising platform for integrated nonlinear\nnanophotonics in the mid-infrared.\n"}
{"abstract": "  Neutrino-matter interactions play an important role in core-collapse\nsupernova (CCSN) explosions as they contribute to both lepton number and/or\nfour-momentum exchange between neutrinos and matter, and thus act as the agent\nfor neutrino-driven explosions. Due to the multiscale nature of neutrino\ntransport in CCSN simulations, an implicit treatment of neutrino-matter\ninteractions is desired, which requires solutions of coupled nonlinear systems\nin each step of the time integration scheme. In this paper we design and\ncompare nonlinear iterative solvers for implicit systems with energy coupling\nneutrino-matter interactions commonly used in CCSN simulations. Specifically,\nwe consider electron neutrinos and antineutrinos, which interact with static\nmatter configurations through the Bruenn~85 opacity set. The implicit systems\narise from the discretization of a non-relativistic two-moment model for\nneutrino transport, which employs the discontinuous Galerkin (DG) method for\nphase-space discretization and an implicit-explicit (IMEX) time integration\nscheme. In the context of this DG-IMEX scheme, we propose two approaches to\nformulate the nonlinear systems -- a coupled approach and a nested approach.\nFor each approach, the resulting systems are solved with Anderson-accelerated\nfixed-point iteration and Newton's method. The performance of these four\niterative solvers has been compared on relaxation problems with various degree\nof collisionality, as well as proto-neutron star deleptonization problems with\nseveral matter profiles adopted from spherically symmetric CCSN simulations.\nNumerical results suggest that the nested Anderson-accelerated fixed-point\nsolver is more efficient than other tested solvers for solving implicit\nnonlinear systems with energy coupling neutrino-matter interactions.\n"}
{"abstract": "  Let $F$ be a moduli space of lattice-polarized K3 surfaces. Suppose that one\nhas chosen a canonical effective ample divisor $R$ on a general K3 in $F$. We\ncall this divisor \"recognizable\" if its flat limit on Kulikov surfaces is well\ndefined. We prove that the normalization of the stable pair compactification\n$\\overline{F}^R$ for a recognizable divisor is a Looijenga semitoroidal\ncompactification.\n  For polarized K3 surfaces $(X,L)$ of degree $2d$, we show that the sum of\nrational curves in the linear system $|L|$ is a recognizable divisor, giving a\nmodular semitoroidal compactification of $F_{2d}$ for all $d$.\n"}
{"abstract": "  This paper aims to assess the effects of industrial pollution on infant\nmortality between the years 1850-1940 using full count decennial censuses. In\nthis period, US economy experienced a tremendous rise in industrial activity\nwith significant variation among different counties in absorbing manufacturing\nindustries. Since manufacturing industries are shown to be the main source of\npollution, we use the share of employment at the county level in this industry\nto proxy for space-time variation in industrial pollution. Since male embryos\nare more vulnerable to external stressors like pollution during prenatal\ndevelopment, they will face higher likelihood of fetal death. Therefore, we\nproxy infant mortality with different measures of gender ratio. We show that\nthe upswing in industrial pollution during late nineteenth century and early\ntwentieth century has led to an increase in infant mortality. The results are\nconsistent and robust across different scenarios, measures for our proxies, and\naggregation levels. We find that infants and more specifically male infants had\npaid the price of pollution during upswing in industrial growth at the dawn of\nthe 20th century. Contemporary datasets are used to verify the validity of the\nproxies. Some policy implications are discussed.\n"}
{"abstract": "  Recent studies indicate that in many situations service times are affected by\nthe experienced queueing delay of the particular customer. This effect has been\ndetected in different areas, such as health care, call centers and\ntelecommunication networks. In this paper we present a methodology to analyze a\nmodel having this property. The specific model is an M/M/c queue in which any\ncustomer may be tagged at her arrival time if her queueing time will be above a\ncertain fixed threshold. All tagged customers are then served at a given rate\nthat may differ from the rate used for the non-tagged customers. We show how it\nis possible to model the virtual queueing time of this queueing system by a\nspecific Markov chain. Then, solving the corresponding balance equations, we\ngive a recursive solution to compute the stationary distribution, which\ninvolves a mixture of exponential terms. Using numerical experiments, we\ndemonstrate that the differences in service rates can have a crucial impact on\nqueueing time performance.\n"}
{"abstract": "  E-commerce provides an efficient and effective way to exchange goods between\nsellers and customers. E-commerce has been a popular method for doing business,\nbecause of its simplicity of having commerce activity transparently available,\nincluding customer voice and opinion about their own experience. Those\nexperiences can be a great benefit to understand customer experience\ncomprehensively, both for sellers and future customers. This paper applies to\ne-commerces and customers in Indonesia. Many Indonesian customers expressed\ntheir voice to open social network services such as Twitter and Facebook, where\na large proportion of data is in the form of conversational data. By\nunderstanding customer behavior through open social network service, we can\nhave descriptions about the e-commerce services level in Indonesia. Thus, it is\nrelated to the government's effort to improve the Indonesian digital economy\necosystem. A method for finding core topics in large-scale internet\nunstructured text data is needed, where the method should be fast but\nsufficiently accurate. Processing large-scale data is not a straightforward\njob, it often needs special skills of people and complex software and hardware\ncomputer system. We propose a fast methodology of text mining methods based on\nfrequently appeared words and their word association to form network text\nmethodology. This method is adapted from Social Network Analysis by the model\nrelationships between words instead of actors.\n"}
{"abstract": "  We report a simple and compact design of a dispersion compensated mode-locked\nYb:fiber oscillator based on a nonlinear amplifying loop mirror (NALM). The\nfully polarization maintaining (PM) fiber integrated laser features a chirped\nfiber Bragg grating (CFBG) for dispersion compensation and a fiber integrated\ncompact non-reciprocal phase bias device, which is alignment-free. The main\ndesign parameters were determined by numerically simulating the pulse evolution\nin the oscillator and by analyzing their impact on the laser performance.\nExperimentally, we achieved an 88 fs compressed pulse duration with sub-fs\ntiming jitter at 54 MHz repetition rate and 51 mW of output power with 5.5 *\n10-5 [20 Hz, 1 MHz] integrated relative intensity noise (RIN). Furthermore, we\ndemonstrate tight phase-locking of the laser's carrier-envelope offset\nfrequency (fceo) to a stable radio frequency (RF) reference and of one\nfrequency comb tooth to a stable optical reference at 291 THz.\n"}
{"abstract": "  HotStuff is a state-of-the-art Byzantine fault-tolerant consensus protocol.\nIt can be pipelined to build large-scale blockchains. One of its variants\ncalled LibraBFT is adopted in Facebook's Libra blockchain. Although it is well\nknown that pipelined HotStuff is secure against up to $1/3$ of Byzantine nodes,\nits performance in terms of throughput and delay is still under-explored. In\nthis paper, we develop a multi-metric evaluation framework to quantitatively\nanalyze pipelined \\mbox{HotStuff's performance} with respect to its chain\ngrowth rate, chain quality, and latency. We then propose two attack strategies\nand evaluate their effects on the performance of pipelined HotStuff. Our\nanalysis shows that the chain growth rate (resp, chain quality) of pipelined\nHotStuff under our attacks can drop to as low as 4/9 (resp, 12/17) of that\nwithout attacks when $1/3$ nodes are Byzantine. As another application, we use\nour framework to evaluate certain engineering optimizations adopted by\nLibraBFT. We find that these optimizations make the system more vulnerable to\nour attacks than the original pipelined HotStuff. Finally, we provide two\ncountermeasures to thwart these attacks. We hope that our studies can shed\nlight on the rigorous understanding of the state-of-the-art pipelined HotStuff\nprotocol as well as its variants.\n"}
{"abstract": "  We consider tensor product random fields $Y_d$, $d\\in\\mathbb{N}$, whose\ncovariance funtions are Gaussian kernels. The average case approximation\ncomplexity $n^{Y_d}(\\varepsilon)$ is defined as the minimal number of\nevaluations of arbitrary linear functionals needed to approximate $Y_d$, with\nrelative $2$-average error not exceeding a given threshold\n$\\varepsilon\\in(0,1)$. We investigate the growth of $n^{Y_d}(\\varepsilon)$ for\narbitrary fixed $\\varepsilon\\in(0,1)$ and $d\\to\\infty$.\n  Namely, we find criteria of boundedness for $n^{Y_d}(\\varepsilon)$ on $d$ and\nof tending $n^{Y_d}(\\varepsilon)\\to\\infty$, $d\\to\\infty$, for any fixed\n$\\varepsilon\\in(0,1)$. In the latter case we obtain necessary and sufficient\nconditions for the following logarithmic asymptotics\n  \\begin{eqnarray*}\n  \\ln n^{Y_d}(\\varepsilon)= a_d+q(\\varepsilon)b_d+o(b_d),\\quad d\\to\\infty,\n  \\end{eqnarray*}\n  with any $\\varepsilon\\in(0,1)$. Here $q\\colon (0,1)\\to\\mathbb{R}$ is a\nnon-decreasing function, $(a_d)_{d\\in\\mathbb{N}}$ is a sequence and\n$(b_d)_{d\\in\\mathbb{N}}$ is a positive sequence such that $b_d\\to\\infty$,\n$d\\to\\infty$.\n  We show that only special quantiles of self-decomposable distribution\nfunctions appear as functions $q$ in a given asymptotics. These general results\napply to $n^{Y_d}(\\varepsilon)$ under particular assumptions on the length\nscale parameters.\n"}
{"abstract": "  In this work, we study the problem of user preference learning on the example\nof parameter setting for a hearing aid (HA). We propose to use an agent that\ninteracts with a HA user, in order to collect the most informative data, and\nlearns user preferences for HA parameter settings, based on these data. We\nmodel the HA system as two interacting sub-systems, one representing a user\nwith his/her preferences and another one representing an agent. In this system,\nthe user responses to HA settings, proposed by the agent. In our user model,\nthe responses are driven by a parametric user preference function. The agent\ncomprises the sequential mechanisms for user model inference and HA parameter\nproposal generation. To infer the user model (preference function), Bayesian\napproximate inference is used in the agent. Here we propose the normalized\nweighted Kullback-Leibler (KL) divergence between true and agent-assigned\npredictive user response distributions as a metric to assess the quality of\nlearned preferences. Moreover, our agent strategy for generating HA parameter\nproposals is to generate HA settings, responses to which help resolving\nuncertainty associated with prediction of the user responses the most. The\nresulting data, consequently, allows for efficient user model learning. The\nnormalized weighted KL-divergence plays an important role here as well, since\nit characterizes the informativeness of the data to be used for probing the\nuser. The efficiency of our approach is validated by numerical simulations.\n"}
{"abstract": "  Empirical work often uses treatment assigned following geographic boundaries.\nWhen the effects of treatment cross over borders, classical\ndifference-in-differences estimation produces biased estimates for the average\ntreatment effect. In this paper, I introduce a potential outcomes framework to\nmodel spillover effects and decompose the estimate's bias in two parts: (1) the\ncontrol group no longer identifies the counterfactual trend because their\noutcomes are affected by treatment and (2) changes in treated units' outcomes\nreflect the effect of their own treatment status and the effect from the\ntreatment status of \"close\" units. I propose estimation strategies that can\nremove both sources of bias and semi-parametrically estimate the spillover\neffects themselves including in settings with staggered treatment timing. To\nhighlight the importance of spillover effects, I revisit analyses of three\nplace-based interventions.\n"}
{"abstract": "  Exploiting short packets for communications is one of the key technologies\nfor realizing emerging application scenarios such as massive machine type\ncommunications (mMTC) and ultra-reliable low-latency communications (uRLLC). In\nthis paper, we investigate short-packet communications to provide both\nreliability and security guarantees simultaneously with an eavesdropper. In\nparticular, an outage probability considering both reliability and secrecy is\ndefined according to the characteristics of short-packet transmission, while\nthe effective throughput in the sense of outage is established as the\nperformance metric. Specifically, a general analytical framework is proposed to\napproximate the outage probability and effective throughput. Furthermore,\nclosed-form expressions for these quantities are derived for the high\nsignal-to-noise ratio (SNR) regime. Both effective throughput obtained via a\ngeneral analytical framework and a high-SNR approximation are maximized under\nan outage-probability constraint by searching for the optimal blocklength.\nNumerical results verify the feasibility and accuracy of the proposed\nanalytical framework, and illustrate the influence of the main system\nparameters on the blocklength and system performance under the\noutage-probability constraint.\n"}
{"abstract": "  We present the detections of shocked molecular hydrogen (H2) gas in near- and\nmid-infrared and broad CO in millimeter from the mixed-morphology supernova\nremnant (SNR) HB~3 (G132.7+1.3) using Palomar WIRC, the Spitzer GLIMPSE360 and\nWISE surveys, and HHSMT. Our near-infrared narrow-band filter H2 2.12 micron\nimages of HB~3 show that both Spitzer IRAC and WISE 4.6 micron emission\noriginates from shocked H2 gas. The morphology of H2 exhibits thin filamentary\nstructures and a large scale of interaction sites between the HB~3 and nearby\nmolecular clouds. Half of HB~3, the southern and eastern shell of the SNR,\nemits H2 in a shape of a \"butterfly\" or \"W\", indicating the interaction sites\nbetween the SNR and dense molecular clouds. Interestingly, the H2 emitting\nregion in the southeast is also co-spatial to the interacting area between HB~3\nand the H~II regions of the W3 complex, where we identified star-forming\nactivity.\n  We further explore the interaction between HB~3 and dense molecular clouds\nwith detections of broad CO(3-2) and CO(2-1) molecular lines from the southern\nand southeastern shells along the H2 emitting region. The widths of the broad\nlines are 8-20 km/s; the detection of such broad lines is unambiguous, dynamic\nevidence of the interactions between the SNR and clouds. The CO broad lines are\nfrom two branches of the bright, southern H2 shell. We apply the Paris-Durham\nshock model to the CO line profiles, which infer the shock velocities of 20 -\n40 km/s, relatively low densities of 10^{3-4} cm^{-3} and strong (>200 micro\nGauss) magnetic fields.\n"}
{"abstract": "  The recoil of atoms in arrays due to the emission or absorption of photons is\nstudied for sub-wavelength interatomic spacing. The atoms in the array interact\nwith each other through collective dipole-dipole interactions and with the\nincident laser field in the low intensity limit. Shining uniform light on the\narray gives rise to patterns of excitation and recoil in the array. These arise\ndue to the interference of different eigenmodes of excitation. The relation\nbetween the recoil and the decay dynamics is studied when the array is in its\nexcitation eigenstates. The recoil experienced by a subradiant collective decay\nis substantially larger than from independent atom decay. A method to calculate\nthe rate of recoil when steady state has been achieved with a constant influx\nof photons is also described.\n"}
{"abstract": "  Recent theoretical works on over-parameterized neural nets have focused on\ntwo aspects: optimization and generalization. Many existing works that study\noptimization and generalization together are based on neural tangent kernel and\nrequire a very large width. In this work, we are interested in the following\nquestion: for a binary classification problem with two-layer mildly\nover-parameterized ReLU network, can we find a point with small test error in\npolynomial time? We first show that the landscape of loss functions with\nexplicit regularization has the following property: all local minima and\ncertain other points which are only stationary in certain directions achieve\nsmall test error. We then prove that for convolutional neural nets, there is an\nalgorithm which finds one of these points in polynomial time (in the input\ndimension and the number of data points). In addition, we prove that for a\nfully connected neural net, with an additional assumption on the data\ndistribution, there is a polynomial time algorithm.\n"}
{"abstract": "  A new framework of thermodynamic modeling is proposed by introducing the\nconcept of differentiable programming, where all the thermodynamic observables\nincluding both thermochemical quantities and phase equilibria can be\ndifferentiated with respect to the underlying model parameters, thus allowing\nthe models learned by gradient-based optimization. It is shown that\nthermodynamic modeling and deep learning can be seamlessly integrated and\nunified within this framework. A preliminary successful application is\ndemonstrated for the Cu-Rh system. It is expected that thermodynamic modeling\nin a deep learning style can increase prediction power of models, and provide\nmore effective guidance for design, synthesis and optimization of\nmulti-component materials with complex chemistry via learning various types of\ndata.\n"}
{"abstract": "  Recent study reported that an aerosolised virus (COVID-19) can survive in the\nair for a few hours. It is highly possible that people get infected with the\ndisease by breathing and contact with items contaminated by the aerosolised\nvirus. However, the aerosolised virus transmission and trajectories in various\nmeteorological environments remain unclear. This paper has investigated the\nmovement of aerosolised viruses from a high concentration source across a dense\nurban area. The case study looks at the highly air polluted areas of London:\nUniversity College Hospital (UCH) and King Cross and St Pancras International\nStation (KCSPI). We explored the spread and decay of COVID-19 released from the\nhospital and railway stations with the prescribed meteorological conditions.\nThe study has three key findings: the primary result is that it is possible for\nthe virus to travel from meters up to hundred meters from the source location.\nThe secondary finding shows viruses released into the atmosphere from entry and\nexit points at KCSPI remain trapped within a small radial distance of < 50m.\nThis strengthens the case for the use of face coverings to reduce the infection\nrate. The final finding shows that there are different levels of risk at\nvarious door locations for UCH, depending on which door is used there can be a\nhigher concentration of COVID-19. Although our results are based on London,\nsince the fundamental knowledge processes are the same, our study can be\nfurther extended to other locations (especially the highly air polluted areas)\nin the world.\n"}
{"abstract": "  Despite the explosion of interest in healthcare AI research, the\nreproducibility and benchmarking of those research works are often limited due\nto the lack of standard benchmark datasets and diverse evaluation metrics. To\naddress this reproducibility challenge, we develop PyHealth, an open-source\nPython toolbox for developing various predictive models on healthcare data.\n  PyHealth consists of data preprocessing module, predictive modeling module,\nand evaluation module. The target users of PyHealth are both computer science\nresearchers and healthcare data scientists. With PyHealth, they can conduct\ncomplex machine learning pipelines on healthcare datasets with fewer than ten\nlines of code. The data preprocessing module enables the transformation of\ncomplex healthcare datasets such as longitudinal electronic health records,\nmedical images, continuous signals (e.g., electrocardiogram), and clinical\nnotes into machine learning friendly formats. The predictive modeling module\nprovides more than 30 machine learning models, including established ensemble\ntrees and deep neural network-based approaches, via a unified but extendable\nAPI designed for both researchers and practitioners. The evaluation module\nprovides various evaluation strategies (e.g., cross-validation and\ntrain-validation-test split) and predictive model metrics.\n  With robustness and scalability in mind, best practices such as unit testing,\ncontinuous integration, code coverage, and interactive examples are introduced\nin the library's development. PyHealth can be installed through the Python\nPackage Index (PyPI) or https://github.com/yzhao062/PyHealth .\n"}
{"abstract": "  We present results of a detailed investigation of the poorly studied X-ray\npulsar XTE J1858+034 based on the data obtained with the NuSTAR observatory\nduring the outburst of the source in 2019. The spectral analysis resulted in\nthe discovery of a cyclotron absorption feature in the source spectrum at ~48\nkeV both in the pulse phase averaged and resolved spectra. Accurate X-ray\nlocalization of the source using the NuSTAR and Chandra observatories allowed\nus to accurately determine the position of the X-ray source and identify the\noptical companion of the pulsar. The analysis of the counterpart properties\nsuggested that the system is likely a symbiotic binary hosting an X-ray pulsar\nand a late type companion star of K-M classes rather than Be X-ray binary as\npreviously suggested.\n"}
{"abstract": "  Queuing systems with an unlimited number of devices with an incoming\nnonstationary Poisson flow and a random flow controlled by a Markov chain are\ninvestigated. The inexpediency of ap-proximation of the birth process by\nPoisson flows in long-term forecasting is shown.\n"}
{"abstract": "  We consider plain vanilla European options written on an underlying asset\nthat follows a continuous time semi-Markov multiplicative process. We derive a\nformula and a renewal type equation for the martingale option price. In the\ncase in which intertrade times follow the Mittag-Leffler distribution, under\nappropriate scaling, we prove that these option prices converge to the price of\nan option written on geometric Brownian motion time-changed with the inverse\nstable subordinator. For geometric Brownian motion time changed with an inverse\nsubordinator, in the more general case when the subordinator's Laplace exponent\nis a special Bernstein function, we derive a time-fractional generalization of\nthe equation of Black and Scholes.\n"}
{"abstract": "  Event extraction is challenging due to the complex structure of event records\nand the semantic gap between text and event. Traditional methods usually\nextract event records by decomposing the complex structure prediction task into\nmultiple subtasks. In this paper, we propose Text2Event, a\nsequence-to-structure generation paradigm that can directly extract events from\nthe text in an end-to-end manner. Specifically, we design a\nsequence-to-structure network for unified event extraction, a constrained\ndecoding algorithm for event knowledge injection during inference, and a\ncurriculum learning algorithm for efficient model learning. Experimental\nresults show that, by uniformly modeling all tasks in a single model and\nuniversally predicting different labels, our method can achieve competitive\nperformance using only record-level annotations in both supervised learning and\ntransfer learning settings.\n"}
{"abstract": "  This paper aims at introducing the formal foundations of the application of\nreduced density-matrix theory and Green's function theory to the analysis of\nmolecular electronic transitions. For this sake, their mechanics, applied to\nspecific objects containing information related to the passage and the\ninterference between electronic states - the difference and the transition\ndensity operators - are rigorously introduced in a self-contained way: After\nreducing the corresponding N-body operators (where N is the number of electrons\nof the system) using an operator partial-trace procedure, we derive the kernel\nof the reduced one-body difference and transition density operators, as well as\nthe matrix representation of these operators in a finite-dimensional\none-particle-state basis. These derivations are done in first and second\nquantization for the sake of completeness - the two formulations are\nequivalently present in the literature -, and because second quantization is\nextensively used in a second part of the paper. Natural orbitals are introduced\nas appropriate bases for reducing the dimensionality of the problem and the\ncomplexity of the analysis of the transition phenomenon: Natural-orbital\nrepresentation of density operators are often used as a tool to characterize\nthe nature of molecular electronic transitions, so we suggested with this\ncontribution to revisit their theoretical foundations in order to better\nunderstand the origin and nature of these tools.\n"}
{"abstract": "  In order to compute with $l$--adic sheaves or crystals on a line over\n$\\mathbb{F} _q$ a low-technology alternative to the traditional computation\nwith the Hecke operators on the automorphic side could be helpful. A program\nwhich has evolved over the years in our discussions with M. Kontsevich centers\naround the concept that, in the geometric case, there must exist certain\nmultiplication laws on the Galois--representation side that could be thought of\nas precursors of the automorphic lifts: non-abelian Abel's theorems, and their\nrestrictions to diagonal, Clausen identities. To a varying extent, they can\ndetermine the trace functions of $l$--adic sheaves or crystals with prescribed\nramification without directly appealing to the Hecke--eigen property on the\nautomorphic side.\n"}
{"abstract": "  Neural Machine Translation (NMT) approaches employing monolingual data are\nshowing steady improvements in resource rich conditions. However, evaluations\nusing real-world low-resource languages still result in unsatisfactory\nperformance. This work proposes a novel zero-shot NMT modeling approach that\nlearns without the now-standard assumption of a pivot language sharing parallel\ndata with the zero-shot source and target languages. Our approach is based on\nthree stages: initialization from any pre-trained NMT model observing at least\nthe target language, augmentation of source sides leveraging target monolingual\ndata, and learning to optimize the initial model to the zero-shot pair, where\nthe latter two constitute a self-learning cycle. Empirical findings involving\nfour diverse (in terms of a language family, script and relatedness) zero-shot\npairs show the effectiveness of our approach with up to +5.93 BLEU improvement\nagainst a supervised bilingual baseline. Compared to unsupervised NMT,\nconsistent improvements are observed even in a domain-mismatch setting,\nattesting to the usability of our method.\n"}
{"abstract": "  As users increasingly rely on cloud-based computing services, it is important\nto ensure that uploaded speech data remains private. Existing solutions rely\neither on server-side methods or focus on hiding speaker identity. While these\napproaches reduce certain security concerns, they do not give users client-side\ncontrol over whether their biometric information is sent to the server. In this\npaper, we formally define client-side privacy and discuss its three unique\ntechnical challenges: (1) direct manipulation of raw data on client devices,\n(2) adaptability with a broad range of server-side processing models, and (3)\nlow time and space complexity for compatibility with limited-bandwidth devices.\nSolving these challenges requires new models that achieve high-fidelity\nreconstruction, privacy preservation of sensitive personal attributes, and\nefficiency during training and inference. As a step towards client-side privacy\nfor speech recognition, we investigate three techniques spanning signal\nprocessing, disentangled representation learning, and adversarial training.\nThrough a series of gender and accent masking tasks, we observe that each\nmethod has its unique strengths, but none manage to effectively balance the\ntrade-offs between performance, privacy, and complexity. These insights call\nfor more research in client-side privacy to ensure a safer deployment of\ncloud-based speech processing services.\n"}
{"abstract": "  We construct characteristic identities for the split (polarized) Casimir\noperators of the simple Lie algebras in adjoint representation. By means of\nthese characteristic identities, for all simple Lie algebras we derive explicit\nformulae for invariant projectors onto irreducible subrepresentations in\nT^{\\otimes 2} in the case when T is the adjoint representation. These\nprojectors and characteristic identities are considered from the viewpoint of\nthe universal description of the simple Lie algebras in terms of the Vogel\nparameters.\n"}
{"abstract": "  This paper investigates the phenomenon of support and resistance levels (SR\nlevels) in financial time series, which act as temporary price barriers that\nreverses price trends. We develop a heuristic discovery algorithm for this\npurpose, to discover and evaluate SR levels for intraday price series. Our\nsimple approach discovers SR levels which are able to reverse price trends\nstatistically significantly. Asset price entering SR levels with higher number\nof price bounces before are more likely to bounce on such SR levels again. We\nalso show that the decay aspect of the discovered SR levels as decreasing\nprobability of price bounce over time. We conclude SR levels are features in\nfinancial time series are not explained simply by AR(1) processes, stationary\nor otherwise; and that they contribute to the temporary predictability and\nstationarity of the investigated price series.\n"}
{"abstract": "  Prediction is the making of statements, usually probabilistic, about future\nevents based on current information. Retrodiction is the making of statements\nabout past events based on current information. We present the foundations of\nquantum retrodiction and highlight its intimate connection with the Bayesian\ninterpretation of probability. The close link with Bayesian methods enables us\nto explore controversies and misunderstandings about retrodiction that have\nappeared in the literature. To be clear, quantum retrodiction is universally\napplicable and draws its validity directly from conventional predictive quantum\ntheory coupled with Bayes' theorem.\n"}
{"abstract": "  We report a spin-free formulation of the multireference (MR) driven\nsimilarity renormalization group (DSRG) by employing the ensemble normal\nordering of Mukherjee and Kutzelnigg [W. Kutzelnigg and D. Mukherjee, J. Chem.\nPhys. 107, 432 (1997)]. This ensemble averages over all microstates for a given\ntotal spin quantum number and, therefore, it is invariant with respect to SU(2)\ntransformations. As such, all equations may be reformulated in terms of\nspin-free quantities and they closely resemble those of spin-adapted\nclosed-shell coupled cluster (CC) theory. The current implementation is used to\nassess the accuracy of various truncated MR-DSRG methods (perturbation theory\nup to third order and iterative methods with single and double excitations) in\ncomputing the constants of thirty-three first-row diatomic molecules. The\naccuracy trends for these first-row diatomics are consistent with our previous\nbenchmark on a small subset of closed-shell diatomic molecules. We then present\nthe first MR-DSRG application on transition-metal complexes by computing the\nspin splittings of the [Fe(H$_2$O)$_6$]$^{2+}$ and [Fe(NH$_3$)$_6$]$^{2+}$\nmolecules. Focal point analysis (FPA) shows that third-order perturbative\ncorrections are essential to achieve reasonably converged energetics. A FPA\nbased on the linearized MR-DSRG theory with one- and two-body operators and up\nto a quintuple-$\\zeta$ basis set predicts the spin splittings of\n[Fe(H$_2$O)$_6$]$^{2+}$ and [Fe(NH$_3$)$_6$]$^{2+}$ to be $-35.7$ and $-17.1$\nkcal mol$^{-1}$, respectively, showing good agreement with results of local CC\ntheory with singles, doubles, and perturbative triples.\n"}
{"abstract": "  We revisit two well-established verification techniques, $k$-induction and\nbounded model checking (BMC), in the more general setting of fixed point theory\nover complete lattices. Our main theoretical contribution is latticed\n$k$-induction, which (i) generalizes classical $k$-induction for verifying\ntransition systems, (ii) generalizes Park induction for bounding fixed points\nof monotonic maps on complete lattices, and (iii) extends from naturals $k$ to\ntransfinite ordinals $\\kappa$, thus yielding $\\kappa$-induction. The\nlattice-theoretic understanding of $k$-induction and BMC enables us to apply\nboth techniques to the fully automatic verification of infinite-state\nprobabilistic programs. Our prototypical implementation manages to\nautomatically verify non-trivial specifications for probabilistic programs\ntaken from the literature that - using existing techniques - cannot be verified\nwithout synthesizing a stronger inductive invariant first.\n"}
{"abstract": "  Transferring human motion and appearance between videos of human actors\nremains one of the key challenges in Computer Vision. Despite the advances from\nrecent image-to-image translation approaches, there are several transferring\ncontexts where most end-to-end learning-based retargeting methods still perform\npoorly. Transferring human appearance from one actor to another is only ensured\nwhen a strict setup has been complied, which is generally built considering\ntheir training regime's specificities. In this work, we propose a shape-aware\napproach based on a hybrid image-based rendering technique that exhibits\ncompetitive visual retargeting quality compared to state-of-the-art neural\nrendering approaches. The formulation leverages the user body shape into the\nretargeting while considering physical constraints of the motion in 3D and the\n2D image domain. We also present a new video retargeting benchmark dataset\ncomposed of different videos with annotated human motions to evaluate the task\nof synthesizing people's videos, which can be used as a common base to improve\ntracking the progress in the field. The dataset and its evaluation protocols\nare designed to evaluate retargeting methods in more general and challenging\nconditions. Our method is validated in several experiments, comprising publicly\navailable videos of actors with different shapes, motion types, and camera\nsetups. The dataset and retargeting code are publicly available to the\ncommunity at: https://www.verlab.dcc.ufmg.br/retargeting-motion.\n"}
{"abstract": "  We study the problem of distributed cooperative learning, where a group of\nagents seeks to agree on a set of hypotheses that best describes a sequence of\nprivate observations. In the scenario where the set of hypotheses is large, we\npropose a belief update rule where agents share compressed (either sparse or\nquantized) beliefs with an arbitrary positive compression rate. Our algorithm\nleverages a unified communication rule that enables agents to access\nwide-ranging compression operators as black-box modules. We prove the almost\nsure asymptotic exponential convergence of beliefs around the set of optimal\nhypotheses. Additionally, we show a non-asymptotic, explicit, and linear\nconcentration rate in probability of the beliefs on the optimal hypothesis set.\nWe provide numerical experiments to illustrate the communication benefits of\nour method. The simulation results show that the number of transmitted bits can\nbe reduced to 5-10% of the non-compressed method in the studied scenarios.\n"}
{"abstract": "  In the context of a geodesically complete Riemannian manifold $M$, we study\nthe self-adjointness of $\\nabla^{\\dagger}\\nabla+V$ where $\\nabla$ is a metric\ncovariant derivative (with formal adjoint $\\nabla^{\\dagger}$) on a Hermitian\nvector bundle $\\mathcal{V}$ over $M$, and $V$ is a locally square integrable\nsection of $\\textrm{End }\\mathcal{V}$ such that the (fiberwise) norm of the\n\"negative\" part $V^{-}$ belongs to the local Kato class (or, more generally,\nlocal contractive Dynkin class). Instead of the lower semiboundedness\nhypothesis, we assume that there exists a number $\\varepsilon \\in [0,1]$ and a\npositive function $q$ on $M$ satisfying certain growth conditions, such that\n$\\varepsilon \\nabla^{\\dagger}\\nabla+V\\geq -q$, the inequality being understood\nin the quadratic form sense over $C_{c}^{\\infty}(\\mathcal{V})$. In the first\nresult, which pertains to the case $\\epsilon \\in [0,1)$, we use the elliptic\nequation method. In the second result, which pertains to the case\n$\\varepsilon=1$, we use the hyperbolic equation method.\n"}
{"abstract": "  In this paper, we conduct one of the very first studies for cross-corpora\nperformance evaluation in the spoken language identification (LID) problem.\nCross-corpora evaluation was not explored much in LID research, especially for\nthe Indian languages. We have selected three Indian spoken language corpora:\nIIITH-ILSC, LDC South Asian, and IITKGP-MLILSC. For each of the corpus, LID\nsystems are trained on the state-of-the-art time-delay neural network (TDNN)\nbased architecture with MFCC features. We observe that the LID performance\ndegrades drastically for cross-corpora evaluation. For example, the system\ntrained on the IIITH-ILSC corpus shows an average EER of 11.80 % and 43.34 %\nwhen evaluated with the same corpora and LDC South Asian corpora, respectively.\nOur preliminary analysis shows the significant differences among these corpora\nin terms of mismatch in the long-term average spectrum (LTAS) and\nsignal-to-noise ratio (SNR). Subsequently, we apply different feature level\ncompensation methods to reduce the cross-corpora acoustic mismatch. Our results\nindicate that these feature normalization schemes can help to achieve promising\nLID performance on cross-corpora experiments.\n"}
{"abstract": "  The ability of bacteria to form biofilms hinders any conventional treatment\nagainst chronic infections and has serious socio-economic implications. In this\nsense, a nanocarrier capable of overcoming the barrier of the\nmucopolysaccharide matrix of the biofilm and releasing its loaded-antibiotic\nwithin would be desirable. Herein, a new nanosystem based on levofloxacin\n(LEVO)-loaded mesoporous silica nanoparticles (MSNs) decorated with lectin\nConcanavalin A (ConA) has been developed. The presence of ConA promotes its\ninternalization into the biofilm matrix, which increases the antimicrobial\nefficacy of the antibiotic hosted within the mesopores. This nanodevice is\nenvisioned as a promising alternative to conventional infection treatments by\nimproving the antimicrobial efficacy and reducing side effects.\n"}
{"abstract": "  This paper presents a review of ideas that interconnect Astrochemistry and\nGalactic Dynamics. Since these two areas are vast and not recent, each one has\nalready been covered separately by several reviews. After a general historical\nintroduction, and a needed quick review of processes like the stellar\nnucleosynthesis which gives the base to understand the interstellar formation\nof simple chemical compounds (H2, CO, NH3 and H2O), we focus on a number of\ntopics which are at the crossing of the two areas, Dynamics and Astrochemistry.\nAstrochemistry is a flourishing field which intends to study the presence and\nformation of molecules as well as the influence of them into the structure,\nevolution and dynamics of astronomical objects. The progress in the knowledge\non the existence of new complex molecules and of their process of formation\noriginates from the observational, experimental and theoretical areas which\ncompose the field. The interfacing areas include star formation, protoplanetary\ndisks, the role of the spiral arms and the chemical abundance gradients in the\ngalactic disk. It often happens that the physical conditions in some regions of\nthe ISM are only revealed by means of molecular observations. To organise a\nclassification of chemical evolution processes, we discuss about how\nastrochemistry can act in three different contexts: i. the chemistry of the\nearly universe, including external galaxies, ii. star forming regions, and iii.\nAGB stars and circumstellar envelopes. We mention that our research is\nstimulated by plans for instruments and projects, such as the on-going LLAMA,\nwhich consists in the construction of a 12m sub-mm radio telescope in the\nAndes. Thus, modern and new facilities can play a key role in new discoveries\nnot only in astrochemistry but also in radio astronomy and related areas.\nFurthermore, the research of the origin of life is also a stimulating\nperspective.\n"}
{"abstract": "  This article encloses some results on nonncommutative analogue of nonabelian\nequations of Langmuir oscillations. One of the main contributions of this work\nis to construct the Darbboux transformation for the solution of that equation\nin noncommutative framework incorporating associated discrete Lax system.\nFurther the standard Darboux transformation on arbitrary eigenfunctions of the\nLax system are presented in quasideterminants for few index values. Moreover,\nthese computations include the derivation of noncommutative version of\nnonabelian discrete nonlinear Schr$\\ddot{o}$dinger which coincides with its\nclassical model under commutative limit. The end portion of this article\nreveals the identity of noncommutative formalism incorporating a derivation of\nan equation of motion which coincides with its existing commutative form in\nbackground zero value of spectral parameter.\n"}
{"abstract": "  The ubiquity of mobile phones makes mobile GUI understanding an important\ntask. Most previous works in this domain require human-created metadata of\nscreens (e.g. View Hierarchy) during inference, which unfortunately is often\nnot available or reliable enough for GUI understanding. Inspired by the\nimpressive success of Transformers in NLP tasks, targeting for purely\nvision-based GUI understanding, we extend the concepts of Words/Sentence to\nPixel-Words/Screen-Sentence, and propose a mobile GUI understanding\narchitecture: Pixel-Words to Screen-Sentence (PW2SS). In analogy to the\nindividual Words, we define the Pixel-Words as atomic visual components (text\nand graphic components), which are visually consistent and semantically clear\nacross screenshots of a large variety of design styles. The Pixel-Words\nextracted from a screenshot are aggregated into Screen-Sentence with a Screen\nTransformer proposed to model their relations. Since the Pixel-Words are\ndefined as atomic visual components, the ambiguity between their visual\nappearance and semantics is dramatically reduced. We are able to make use of\nmetadata available in training data to auto-generate high-quality annotations\nfor Pixel-Words. A dataset, RICO-PW, of screenshots with Pixel-Words\nannotations is built based on the public RICO dataset, which will be released\nto help to address the lack of high-quality training data in this area. We\ntrain a detector to extract Pixel-Words from screenshots on this dataset and\nachieve metadata-free GUI understanding during inference. We conduct\nexperiments and show that Pixel-Words can be well extracted on RICO-PW and well\ngeneralized to a new dataset, P2S-UI, collected by ourselves. The effectiveness\nof PW2SS is further verified in the GUI understanding tasks including relation\nprediction, clickability prediction, screen retrieval, and app type\nclassification.\n"}
{"abstract": "  The study aims to design and develop a virtual structural design that\nsimulates the campus and its buildings of a community college in Bukidnon,\nPhilippines through Virtual Reality. With the immersion of technology, this\nproject represents the architectural design of the establishment with the use\nof Virtual Reality Technology. The project uses a modified Iterative\nDevelopment Model which is a guide for the design and development of the 3D\nModels and VR Application. TinkerCAD which is a web-based application has been\nused to design buildings on the other hand Unity is used to develop the\nstructural designs of the buildings. The respondents of this study are the\nGrade 12 Senior High students from the 4 schools which are geographically near\nto the college. With this study, the researchers were able to showcase its VR\nApplication to the students and later evaluated using a System Usability Scale,\na 10 item questionnaire measuring usability with an overall average of 90% or\nPoint Score of 4.5 which is interpreted as excellent in a Likert table for\ndescriptive interpretation. With the use of the VR application potential\nstudents of the college will be able to visualize and experience the present\nstructures of the college without being physically present in the area. In this\npaper, the buildings and structures of NBCC were designed and developed through\na Virtual Reality Platform allowing students from different secondary schools\nthat are geographically near to the college to experience the feeling to be in\nthe school without being able to set a step in physically.\n"}
{"abstract": "  Cells within living soft biological tissues seem to promote the maintenance\nof a mechanical state within a defined range near a so-called set-point. This\nmechanobiological process is often referred to as mechanical homeostasis.\nDuring this process, cells intimately interact with the fibers of the\nsurrounding extracellular matrix (ECM). It remains poorly understood, however,\nwhat individual cells actually regulate during these interactions, and how\nthese micromechanical regulations are translated to tissue level to lead to\nwhat we macroscopically call mechanical homeostasis. Herein, we examine this\nquestion by a combination of experiments, theoretical analysis and\ncomputational modeling. We demonstrate that on short time scales (hours) -\nduring which deposition and degradation of ECM fibers can largely be neglected\n- cells appear to regulate neither the stress / strain in the ECM nor their own\nshape, but rather only the contractile forces that they exert on the\nsurrounding ECM.\n"}
{"abstract": "  We present EuroCrops, a dataset based on self-declared field annotations for\ntraining and evaluating methods for crop type classification and mapping,\ntogether with its process of acquisition and harmonisation. By this, we aim to\nenrich the research efforts and discussion for data-driven land cover\nclassification via Earth observation and remote sensing. Additionally, through\ninclusion of self-declarations gathered in the scope of subsidy control from\nall countries of the European Union (EU), this dataset highlights the\ndifficulties and pitfalls one comes across when operating on a transnational\nlevel. We, therefore, also introduce a new taxonomy scheme, HCAT-ID, that\naspires to capture all the aspects of reference data originating from\nadministrative and agency databases. To address researchers from both the\nremote sensing and the computer vision and machine learning communities, we\npublish the dataset in different formats and processing levels.\n"}
{"abstract": "  We use two non-Riemannian curvature tensors, the $\\chi$-curvature and the\nmean Berwald curvature to characterise a class of Finsler metrics admitting\nfirst integrals.\n"}
{"abstract": "  The development of modern technology extends human presence beyond cislunar\nspace and onto other planets, which presents an urgent need for high-capacity,\nlong-distance and interplanetary communication. Communication using photons as\ncarriers has a high channel capacity, but the optical diffraction limit in deep\nspace leads to inevitable huge geometric loss, setting an insurmountable\ntransmission distance for existing optical communication technologies. Here, we\npropose and experimentally demonstrate a photon-inter-correlation optical\ncommunication (PICOC) against an ultra-high channel loss. We treat light as a\nstream of photons, and retrieve the additional information of internal\ncorrelation and photon statistics globally from extremely weak pulse sequences.\nWe successfully manage to build high-fidelity communication channel with a loss\nup to 160dB by separating a single-photon signal embedded in a noise ten times\nhigher. With only commercially available telescopes, PICOC allows establishment\nof communication links from Mars to Earth communication using a milliwatt\nlaser, and from the edge of the solar system to Earth using a few watts laser.\n"}
{"abstract": "  The crux of knowledge distillation is to effectively train a resource-limited\nstudent model with the guide of a pre-trained larger teacher model. However,\nwhen there is a large difference between the model complexities of teacher and\nstudent (i.e., capacity gap), knowledge distillation loses its strength in\ntransferring knowledge from the teacher to the student, thus training a weaker\nstudent. To mitigate the impact of the capacity gap, we introduce knowledge\ndistillation via intermediate heads. By extending the intermediate layers of\nthe teacher (at various depths) with classifier heads, we cheaply acquire a\ncohort of heterogeneous pre-trained teachers. The intermediate classifier heads\ncan all together be efficiently learned while freezing the backbone of the\npre-trained teacher. The cohort of teachers (including the original teacher)\nco-teach the student simultaneously. Our experiments on various teacher-student\npairs and datasets have demonstrated that the proposed approach outperforms the\ncanonical knowledge distillation approach and its extensions.\n"}
{"abstract": "  We describe an efficient method for extracting the parts of $D$-dimensional\nloop integrals that are needed to derive observables in classical general\nrelativity from scattering amplitudes. Our approach simplifies the soft-region\nmethod of integration by judiciously combining terms before the final\nintegrations. We demonstrate the method by computing the required integrals for\nblack-hole scattering to the second Post-Minkowskian order in Einstein gravity\ncoupled to scalars. We also confirm recent results at the third\nPost-Minkowskian order regarding universality and high-energy behavior of\ngravitational interactions in maximal supergravity.\n"}
{"abstract": "  Non-linear effects in the early Universe generate non-zero bispectra of the\ncosmic microwave background (CMB) temperature and polarization, even in the\nabsence of primordial non-Gaussianity. In this paper, we compute the\ncontributions from isocurvature modes to the CMB bispectra using a modified\nversion of the second-order Boltzmann solver SONG. We investigate the ability\nof current and future CMB experiments to constrain these modes with\nobservations of the bispectrum. Our results show that the enhancement due to\nsingle isocurvature modes mixed with the adiabatic mode is negligible for the\nparameter ranges currently allowed by the most recent Planck results. However,\nwe find that a large compensated isocurvature mode can produce a detectable\nbispectrum when its correlation with the adiabatic mode is appreciable. The\nnon-observation of this contribution in searches for the lensing bispectrum\nfrom Planck allows us to place a new constraint on the relative amplitude of\nthe correlated part of the compensated isocurvature mode of $f_{\\rm\nCIP}=1\\pm100$. We compute forecasts for future observations by COrE, SO, CMB-S4\nand an ideal experiment and conclude that a dedicated search for the bispectrum\nfrom compensated modes could rule out a number of scenarios realised in the\ncurvaton model. In addition, the CMB-S4 experiment could detect the most\nextreme of those scenarios ($f_{\\rm CIP}=16.5$) at 2 to 3-$\\sigma$\nsignificance.\n"}
{"abstract": "  Abstract: An induced-transmission filter (ITF) uses an ultrathin layer of\nmetal positioned at an electric-field node within a dielectric thin-film\nbandpass filter to select one transmission band while suppressing other\ntransmission bands that would have been present without the metal layer. Here,\nwe introduce a switchable mid-infrared ITF where the metal film can be\n\"switched on and off\", enabling the modulation of the filter response from\nsingle-band to multiband. The switching is enabled by a deeply subwavelength\nfilm of vanadium dioxide (VO2), which undergoes a reversible insulator-to-metal\nphase transition. We designed and experimentally demonstrated an ITF that can\nswitch between two states: one broad passband across the long-wave infrared\n(LWIR, 8 - 12 um) and one narrow passband at ~8.8 um. Our work generalizes the\nITF -- previously a niche type of bandpass filter -- into a new class of\ntunable devices. Furthermore, our unique fabrication process -- which begins\nwith thin-film VO2 on a suspended membrane -- enables the integration of VO2\ninto any thin-film assembly that is compatible with physical vapor deposition\n(PVD) processes, and is thus a new platform for realizing tunable thin-film\nfilters.\n"}
{"abstract": "  Aims.We aim to locate the origin of a stealth coronal mass ejection (CME)\ndetected in situ by the MAG instrument on board Solar Orbiter, and make\nconnections between the CME observed at the Sun, and the interplanetary CME\n(ICME) measured in situ. Methods. Remote sensing data are analysed using\nadvanced image processing techniques to identify the source region of the\nstealth CME, and the global magnetic field at the time of the eruption is\nexamined using Potential Field Source Surface (PFSS) models. The observations\nof the stealth CME at the Sun are compared with the magnetic field measured by\nthe Solar Orbiter spacecraft, and plasma properties measured by the Wind\nspacecraft. Results. The source of the CME is found to be a quiet Sun cavity in\nthe northern hemisphere. We find that the stealth CME has a strong magnetic\nfield in situ, despite originating from a quiet Sun region with an extremely\nweak magnetic field. Conclusions. The interaction of the ICME with its\nsurrounding environment is the likely cause of a higher magnetic field strength\nmeasured in situ. Stealth CMEs require multi-wavelength and multi-viewpoint\nobservations in order to confidently locate the source region, however their\nelusive signatures still pose many problems for space weather forecasting. The\nfindings have implications for Solar Orbiter observing sequences with\ninstruments such as EUI that are designed to capture stealth CMEs\n"}
{"abstract": "  The development of the A64FX processor by Fujitsu has been a massive\ninnovation in vectorized processors and led to Fugaku: the current world's\nfastest supercomputer. We use a variety of tools to analyze the behavior and\nperformance of several OpenMP applications with different compilers, and how\nthese applications scale on the different A64FX processors on clusters at Stony\nBrook University and RIKEN.\n"}
{"abstract": "  We present here the discovery of a new class of super-slow rotating asteroids\n(P>1000 hours) in data extracted from the Asteroid Terrestrial-impact Last\nAlert System (ATLAS) and Zwicky Transient Facility (ZTF) all-sky surveys. Of\nthe 39 rotation periods we report here, 32 have periods longer than any\npreviously reported unambiguous rotation periods currently in the Asteroid\nLight Curve Database. In our sample, 7 objects have a rotation period > 4000\nhours and the longest period we report here is 4812 hours (~200 days). We do\nnot observe any correlation between taxonomy, albedo, or orbital properties\nwith super-slow rotating status. The most plausible mechanism for the creation\nof these very slow rotators is if their rotations were slowed by YORP\nspin-down. Super-slow rotating asteroids may be common, with at least 0.4% of\nthe main-belt asteroid population with a size range between 2 and 20 km in\ndiameter rotating with periods longer than 1000 hours.\n"}
{"abstract": "  We introduce a new sinc-type molecular beam epitaxy model which is derived\nfrom a cosine-type energy functional. The landscape of the new functional is\nremarkably similar to the classical MBE model with double well potential but\nhas the additional advantage that all its derivatives are uniformly bounded. We\nconsider first order IMEX and second order BDF2 discretization schemes. For\nboth cases we quantify explicit time step constraints for the energy\ndissipation which is in good accord with the practical numerical simulations.\nFurthermore we introduce a new theoretical framework and prove unconditional\nuniform energy boundedness with no size restrictions on the time step. This is\nthe first unconditional (i.e. independent of the time step size) result for\nsemi-implicit methods applied to the phase field models without introducing any\nartificial stabilization terms or fictitious variables.\n"}
{"abstract": "  The next generation of bolometric experiments searching for rave events, in\nparticular for the neutrino-less double beta decay, needs fast,\nhigh-sensitivity and easy-to-scale cryogenic light detectors. The CALDER\nproject (2014-2020) developed a new technology for light detection at cryogenic\ntemperature. In this paper we describe the achievements and the final prototype\nof this project, consisting of a $5\\times5$ cm$^2$, 650 $\\mu$m thick silicon\nsubstrate coupled to a single kinetic inductance detector made of a three-layer\naluminum-titanium-aluminum. The baseline energy resolution is\n34$\\pm$1(stat)$\\pm$2(syst) eV RMS and the response time is 120 $\\mu$s. These\nfeatures, along with the natural multiplexing capability of kinetic inductance\ndetectors, meet the requirements of future large-scale experiments.\n"}
{"abstract": "  Knowledge retrieval and reasoning are two key stages in multi-hop question\nanswering (QA) at web scale. Existing approaches suffer from low confidence\nwhen retrieving evidence facts to fill the knowledge gap and lack transparent\nreasoning process. In this paper, we propose a new framework to exploit more\nvalid facts while obtaining explainability for multi-hop QA by dynamically\nconstructing a semantic graph and reasoning over it. We employ Abstract Meaning\nRepresentation (AMR) as semantic graph representation. Our framework contains\nthree new ideas: (a) {\\tt AMR-SG}, an AMR-based Semantic Graph, constructed by\ncandidate fact AMRs to uncover any hop relations among question, answer and\nmultiple facts. (b) A novel path-based fact analytics approach exploiting {\\tt\nAMR-SG} to extract active facts from a large fact pool to answer questions. (c)\nA fact-level relation modeling leveraging graph convolution network (GCN) to\nguide the reasoning process. Results on two scientific multi-hop QA datasets\nshow that we can surpass recent approaches including those using additional\nknowledge graphs while maintaining high explainability on OpenBookQA and\nachieve a new state-of-the-art result on ARC-Challenge in a computationally\npracticable setting.\n"}
{"abstract": "  Text style transfer has gained increasing attention from the research\ncommunity over the recent years. However, the proposed approaches vary in many\nways, which makes it hard to assess the individual contribution of the model\ncomponents. In style transfer, the most important component is the optimization\ntechnique used to guide the learning in the absence of parallel training data.\nIn this work we empirically compare the dominant optimization paradigms which\nprovide supervision signals during training: backtranslation, adversarial\ntraining and reinforcement learning. We find that backtranslation has\nmodel-specific limitations, which inhibits training style transfer models.\nReinforcement learning shows the best performance gains, while adversarial\ntraining, despite its popularity, does not offer an advantage over the latter\nalternative. In this work we also experiment with Minimum Risk Training, a\npopular technique in the machine translation community, which, to our\nknowledge, has not been empirically evaluated in the task of style transfer. We\nfill this research gap and empirically show its efficacy.\n"}
{"abstract": "  We extend semi-supervised learning to the problem of domain adaptation to\nlearn significantly higher-accuracy models that train on one data distribution\nand test on a different one. With the goal of generality, we introduce\nAdaMatch, a method that unifies the tasks of unsupervised domain adaptation\n(UDA), semi-supervised learning (SSL), and semi-supervised domain adaptation\n(SSDA). In an extensive experimental study, we compare its behavior with\nrespective state-of-the-art techniques from SSL, SSDA, and UDA on vision\nclassification tasks. We find AdaMatch either matches or significantly exceeds\nthe state-of-the-art in each case using the same hyper-parameters regardless of\nthe dataset or task. For example, AdaMatch nearly doubles the accuracy compared\nto that of the prior state-of-the-art on the UDA task for DomainNet and even\nexceeds the accuracy of the prior state-of-the-art obtained with pre-training\nby 6.4% when AdaMatch is trained completely from scratch. Furthermore, by\nproviding AdaMatch with just one labeled example per class from the target\ndomain (i.e., the SSDA setting), we increase the target accuracy by an\nadditional 6.1%, and with 5 labeled examples, by 13.6%.\n"}
{"abstract": "  In complex, expensive optimization domains we often narrowly focus on finding\nhigh performing solutions, instead of expanding our understanding of the domain\nitself. But what if we could quickly understand the complex behaviors that can\nemerge in said domains instead? We introduce surrogate-assisted phenotypic\nniching, a quality diversity algorithm which allows to discover a large,\ndiverse set of behaviors by using computationally expensive phenotypic\nfeatures. In this work we discover the types of air flow in a 2D fluid dynamics\noptimization problem. A fast GPU-based fluid dynamics solver is used in\nconjunction with surrogate models to accurately predict fluid characteristics\nfrom the shapes that produce the air flow. We show that these features can be\nmodeled in a data-driven way while sampling to improve performance, rather than\nexplicitly sampling to improve feature models. Our method can reduce the need\nto run an infeasibly large set of simulations while still being able to design\na large diversity of air flows and the shapes that cause them. Discovering\ndiversity of behaviors helps engineers to better understand expensive domains\nand their solutions.\n"}
{"abstract": "  Density estimation is an important technique for characterizing distributions\ngiven observations. Much existing research on density estimation has focused on\ncases wherein the data lies in a Euclidean space. However, some kinds of data\nare not well-modeled by supposing that their underlying geometry is Euclidean.\nInstead, it can be useful to model such data as lying on a {\\it manifold} with\nsome known structure. For instance, some kinds of data may be known to lie on\nthe surface of a sphere. We study the problem of estimating densities on\nmanifolds. We propose a method, inspired by the literature on \"dequantization,\"\nwhich we interpret through the lens of a coordinate transformation of an\nambient Euclidean space and a smooth manifold of interest. Using methods from\nnormalizing flows, we apply this method to the dequantization of smooth\nmanifold structures in order to model densities on the sphere, tori, and the\northogonal group.\n"}
{"abstract": "  The mining and utilization of features directly affect the classification\nperformance of models used in the classification and recognition of\nhyperspectral remote sensing images. Traditional models usually conduct feature\nmining from a single perspective, with the features mined being limited and the\ninternal relationships between them being ignored. Consequently, useful\nfeatures are lost and classification results are unsatisfactory. To fully mine\nand utilize image features, a new multi-scale feature-mining learning algorithm\n(MGRNet) is proposed. The model uses principal component analysis to reduce the\ndimensionality of the original hyperspectral image (HSI) to retain 99.99% of\nits semantic information and extract dimensionality reduction features. Using a\nmulti-scale convolution algorithm, the input dimensionality reduction features\nwere mined to obtain shallow features, which then served as inputs into a\nmulti-scale graph convolution algorithm to construct the internal relationships\nbetween eigenvalues at different scales. We then carried out cross fusion of\nmulti-scale information obtained by graph convolution, before inputting the new\ninformation obtained into the residual network algorithm for deep feature\nmining. Finally, a flexible maximum transfer function classifier was used to\npredict the final features and complete the classification. Experiments on\nthree common hyperspectral datasets showed the MGRNet algorithm proposed in\nthis paper to be superior to traditional methods in recognition accuracy.\n"}
{"abstract": "  Simulation of time dynamical physical problems has been a challenge for\nclassical computers due to their time-complexity. To demonstrate the dominance\nof quantum computers over classical computers in this regime, here we simulate\na semi-empirical model where two neutral particles interact through\nLennard-Jones potential in a one-dimensional system. We implement the above\nscenario on the IBM quantum experience platform using a 5-qubit real device. We\nconstruct the Hamiltonian and then efficiently map it to quantum operators onto\nquantum gates using the time-evolutionary unitary matrix obtained from the\nHamiltonian. We verify the results collected from the QASM-simulator and\ncompare it with that of the 5-qubit real chip ibmqx2.\n"}
{"abstract": "  We characterize the Lie groups with finitely many connected components that\nare $O(u)$-bilipschitz equivalent (almost quasiisometric in the sense that the\nsublinear function $u$ replaces the additive bounds of quasiisometry) to the\nreal hyperbolic space, or to the complex hyperbolic plane. The\ncharacterizations are expressed in terms of deformations of Lie algebras and in\nterms of pinching of sectional curvature of left-invariant Riemannian metrics\nin the real case. We also compare sublinear bilipschitz equivalence and coarse\nequivalence, and prove that every coarse equivalence between the logarithmic\ncoarse structures of geodesic spaces is a $O(\\log)$-bilipschitz equivalence.\nThe Lie groups characterized are exactly those whose logarithmic coarse\nstructure is equivalent to that of a real hyperbolic space or the complex\nhyperbolic plane.\n"}
{"abstract": "  Because of the pervasive usage of Neural Networks in human sensitive\napplications, their interpretability is becoming an increasingly important\ntopic in machine learning. In this work we introduce a simple way to interpret\nthe output function of a neural network classifier that take as input\ncategorical variables. By exploiting a mapping between a neural network\nclassifier and a physical energy model, we show that in these cases each layer\nof the network, and the logits layer in particular, can be expanded as a sum of\nterms that account for the contribution to the classification of each input\npattern. For instance, at the first order, the expansion considers just the\nlinear relation between input features and output while at the second order\npairwise dependencies between input features are also accounted for. The\nanalysis of the contributions of each pattern, after an appropriate gauge\ntransformation, is presented in two cases where the effectiveness of the method\ncan be appreciated.\n"}
{"abstract": "  Legged robots need to be capable of walking on diverse terrain conditions. In\nthis paper, we present a novel reinforcement learning framework for learning\nlocomotion on non-rigid dynamic terrains. Specifically, our framework can\ngenerate quadruped locomotion on flat elastic terrain that consists of a matrix\nof tiles moving up and down passively when pushed by the robot's feet. A\ntrained robot with 55cm base length can walk on terrain that can sink up to\n5cm. We propose a set of observation and reward terms that enable this\nlocomotion; in which we found that it is crucial to include the end-effector\nhistory and end-effector velocity terms into observation. We show the\neffectiveness of our method by training the robot with various terrain\nconditions.\n"}
{"abstract": "  We demonstrate that black holes and stars in general relativity can be\ndestabilized by perturbations of non-minimally coupled vector fields. Focusing\non static and spherically symmetric backgrounds, our analysis shows that black\nholes with sufficiently small mass and stars with sufficiently high densities\nare subject to ghost or gradient-type instabilities. This holds for a large\nclass of vector-tensor theories whenever non-minimal couplings contribute to\nlinearized dynamics about a state with vanishing vector field and applies to\ngeneralized Proca models that have sparked attention for their potential role\nin cosmology and astrophysics. The stability criteria translate into bounds of\nrelevance for low-scale theories of dark energy and for ultra-light dark matter\nscenarios.\n"}
{"abstract": "  For 6-DOF (degrees of freedom) interactive virtual acoustic environments\n(VAEs), the spatial rendering of diffuse late reverberation in addition to\nearly (specular) reflections is important. In the interest of computational\nefficiency, the acoustic simulation of the late reverberation can be simplified\nby using a limited number of spatially distributed virtual reverb sources (VRS)\neach radiating incoherent signals. A sufficient number of VRS is needed to\napproximate spatially anisotropic late reverberation, e.g., in a room with\ninhomogeneous distribution of absorption at the boundaries. Here, a highly\nefficient and perceptually plausible method to generate and spatially render\nlate reverberation is suggested, extending the room acoustics simulator RAZR\n[Wendt et al., J. Audio Eng. Soc., 62, 11 (2014)]. The room dimensions and\nfrequency-dependent absorption coefficients at the wall boundaries are used to\ndetermine the parameters of a physically-based feedback delay network (FDN) to\ngenerate the incoherent VRS signals. The VRS are spatially distributed around\nthe listener with weighting factors representing the spatially subsampled\ndistribution of absorption coefficients on the wall boundaries. The minimum\nnumber of VRS required to be perceptually distinguishable from the maximum\n(reference) number of 96 VRS was assessed in a listening test conducted with a\nspherical loudspeaker array within an anechoic room. For the resulting low\nnumbers of VRS suited for spatial rendering, optimal physically-based parameter\nchoices for the FDN are discussed.\n"}
{"abstract": "  We present the results from an observing campaign to confirm the peculiar\nmotion of the supermassive black hole (SMBH) in J0437+2456 first reported in\nPesce et al. (2018). Deep observations with the Arecibo Observatory have\nyielded a detection of neutral hydrogen (HI) emission, from which we measure a\nrecession velocity of 4910 km s$^{-1}$ for the galaxy as a whole. We have also\nobtained near-infrared integral field spectroscopic observations of the\ngalactic nucleus with the Gemini North telescope, yielding spatially resolved\nstellar and gas kinematics with a central velocity at the innermost radii\n($0.1^{\\prime \\prime} \\approx 34$ pc) of 4860 km s$^{-1}$. Both measurements\ndiffer significantly from the $\\sim$4810 km s$^{-1}$ H$_2$O megamaser velocity\nof the SMBH, supporting the prior indications of a velocity offset between the\nSMBH and its host galaxy. However, the two measurements also differ\nsignificantly from one another, and the galaxy as a whole exhibits a complex\nvelocity structure that implies the system has recently been dynamically\ndisturbed. These results make it clear that the SMBH is not at rest with\nrespect to the systemic velocity of the galaxy, though the specific nature of\nthe mobile SMBH -- i.e., whether it traces an ongoing galaxy merger, a binary\nblack hole system, or a gravitational wave recoil event -- remains unclear.\n"}
{"abstract": "  We define standardized constructions of finite fields, and standardized\ngenerators of (multiplicative) cyclic subgroups in these fields.\n  The motivation is to provide a substitute for Conway polynomials which can be\nused by various software packages and in collections of mathematical data which\ninvolve finite fields.\n"}
{"abstract": "  Very recently, the construction of twist actuators from magnetorheological\ngels and elastomers has been suggested. These materials consist of magnetizable\ncolloidal particles embedded in a soft elastic polymeric environment. The twist\nactuation is enabled by a net chirality of the internal particle arrangement.\nUpon magnetization by a homogeneous external magnetic field, the systems\nfeature an overall torsional deformation around the magnetization direction.\nStarting from a discrete minimal mesoscopic model set-up we work towards a\nmacroscopic characterization. The two scales are linked by identifying\nexpressions for the macroscopic system parameters as functions of the\nmesoscopic model parameters. In this way, the observed behavior of a\nmacroscopic system can in principle be mapped to and illustratively be\nunderstood from an appropriate mesoscopic picture. Our results apply equally\nwell to corresponding soft electrorheological gels and elastomers.\n"}
{"abstract": "  Radiative feedback can influence subsequent star formation. We quantify the\nheating from OB stars in the local star-forming regions in the JCMT Gould Belt\nsurvey. Dust temperatures are calculated from 450/850 micron flux ratios from\nSCUBA-2 observations at the JCMT assuming a fixed dust opacity spectral index\n$\\beta=1.8$. Mean dust temperatures are calculated for each submillimetre clump\nalong with projected distances from the main OB star in the region. Temperature\nvs. distance is fit with a simple model of dust heating by the OB star\nradiation plus the interstellar radiation field and dust cooling through\noptically thin radiation. Classifying the heating sources by spectral type,\nO-type stars produce the greatest clump average temperature rises and largest\nheating extent, with temperatures over 40 K and significant heating out to at\nleast 2.4 pc. Early-type B stars (B4 and above) produce temperatures of over 20\nK and significant heating over 0.4 pc. Late-type B stars show a marginal\nheating effect within 0.2 pc. For a given projected distance, there is a\nsignificant scatter in clump temperatures that is due to local heating by other\nluminous stars in the region, projection effects, or shadowing effects. Even in\nthese local, `low-mass' star-forming regions, radiative feedback is having an\neffect on parsec scales, with 24% of the clumps heated to at least 3 K above\nthe 15 K base temperature expected from heating by only the interstellar\nradiation field, and a mean dust temperature for heated clumps of 24 K.\n"}
{"abstract": "  The software ecosystem is a trust-rich part of the world. Collaboratively,\nsoftware engineers trust major hubs in the ecosystem, such as package managers,\nrepository services, and programming language ecosystems. This trust, however,\nis often broken by vulnerabilities, ransomware, and abuse from malignant\nactors.\n  But what is trust? In this paper we explore, through twelve in-depth\ninterviews with software engineers, how they perceive trust in their daily\nwork. From the interviews we conclude three things. First, software engineers\nmake a distinction between an adoption factor and a trust factor when selecting\na package. Secondly, while in literature mostly technical factors are\nconsidered as the main trust factors, the software engineers in this study\nconclude that organizational factors are more important. Finally, we find that\ndifferent kinds of software engineers require different views on trust, and\nthat it is impossible to create one unified perception of trust.\n  Keywords: software ecosystem trust, empirical software engineering,\nTrustSECO, external software adoption, cross-sectional exploratory interview\nanalysis, trust perception.\n"}
{"abstract": "  In this paper, we formulate black hole solutions through extended\ngravitational decoupling scheme in the framework of self-interacting\nBrans-Dicke theory. The addition of a new source in the matter distribution\nincreases the degrees of freedom in the system of field equations.\nTransformations in radial as well as temporal metric functions split the system\ninto two arrays. Each array includes the effects of only one source (either\nseed or additional). The seed source is assumed to be a vacuum and the\ncorresponding system is specified through the Schwarzschild metric. In order to\nconstruct a suitable solution of the second system, constraints are applied on\nthe metric potentials and energy-momentum tensor of the additional source. We\nobtain three solutions corresponding to different values of the decoupling\nparameter in the presence of a massive scalar field. The extra source is\nclassified as normal or exotic through energy conditions. It is found that two\nsolutions agree with the energy bounds and thus have normal matter as their\nsource.\n"}
{"abstract": "  A famous result by R\\\"odl, Ruci\\'nski, and Szemer\\'edi guarantees a (tight)\nHamilton cycle in $k$-uniform hypergraphs $H$ on $n$ vertices with minimum\n$(k-1)$-degree $\\delta_{k-1}(H)\\geq (1/2+o(1))n$, thereby extending Dirac's\nresult from graphs to hypergraphs. For graphs, much more is known; each graph\non $n$ vertices with $\\delta(G)\\geq (1/2+o(1))n$ contains $(1-o(1))r$\nedge-disjoint Hamilton cycles where $r$ is the largest integer such that $G$\ncontains a spanning $2r$-regular subgraph, which is clearly asymptotically\noptimal. This was proved by Ferber, Krivelevich, and Sudakov answering a\nquestion raised by K\\\"uhn, Lapinskas, and Osthus.\n  We extend this result to hypergraphs; every $k$-uniform hypergraph $H$ on $n$\nvertices with $\\delta_{k-1}(H)\\geq (1/2+o(1))n$ contains $(1-o(1))r$\nedge-disjoint (tight) Hamilton cycles where $r$ is the largest integer such\nthat $H$ contains a spanning subgraph with each vertex belonging to $kr$ edges.\nIn particular, this yields an asymptotic solution to a question of Glock,\nK\\\"uhn, and Osthus.\n  In fact, our main result applies to approximately vertex-regular $k$-uniform\nhypergraphs with a weak quasirandom property and provides approximate\ndecompositions into cycle factors without too short cycles.\n"}
{"abstract": "  Neural networks trained on visual data are well-known to be vulnerable to\noften imperceptible adversarial perturbations. The reasons for this\nvulnerability are still being debated in the literature. Recently Ilyas et al.\n(2019) showed that this vulnerability arises, in part, because neural network\nclassifiers rely on highly predictive but brittle \"non-robust\" features. In\nthis paper we extend the work of Ilyas et al. by investigating the nature of\nthe input patterns that give rise to these features. In particular, we\nhypothesize that in a neural network trained in a standard way, non-robust\nfeatures respond to small, \"non-semantic\" patterns that are typically entangled\nwith larger, robust patterns, known to be more human-interpretable, as opposed\nto solely responding to statistical artifacts in a dataset. Thus, adversarial\nexamples can be formed via minimal perturbations to these small, entangled\npatterns. In addition, we demonstrate a corollary of our hypothesis: robust\nclassifiers are more effective than standard (non-robust) ones as a source for\ngenerating transferable adversarial examples in both the untargeted and\ntargeted settings. The results we present in this paper provide new insight\ninto the nature of the non-robust features responsible for adversarial\nvulnerability of neural network classifiers.\n"}
{"abstract": "  This paper focuses on building semantic maps, containing object poses and\nshapes, using a monocular camera. This is an important problem because robots\nneed rich understanding of geometry and context if they are to shape the future\nof transportation, construction, and agriculture. Our contribution is an\ninstance-specific mesh model of object shape that can be optimized online based\non semantic information extracted from camera images. Multi-view constraints on\nthe object shape are obtained by detecting objects and extracting\ncategory-specific keypoints and segmentation masks. We show that the errors\nbetween projections of the mesh model and the observed keypoints and masks can\nbe differentiated in order to obtain accurate instance-specific object shapes.\nWe evaluate the performance of the proposed approach in simulation and on the\nKITTI dataset by building maps of car poses and shapes.\n"}
{"abstract": "  Unsupervised domain adaptation (UDA) enables cross-domain learning without\ntarget domain labels by transferring knowledge from a labeled source domain\nwhose distribution differs from that of the target. However, UDA is not always\nsuccessful and several accounts of `negative transfer' have been reported in\nthe literature. In this work, we prove a simple lower bound on the target\ndomain error that complements the existing upper bound. Our bound shows the\ninsufficiency of minimizing source domain error and marginal distribution\nmismatch for a guaranteed reduction in the target domain error, due to the\npossible increase of induced labeling function mismatch. This insufficiency is\nfurther illustrated through simple distributions for which the same UDA\napproach succeeds, fails, and may succeed or fail with an equal chance.\nMotivated from this, we propose novel data poisoning attacks to fool UDA\nmethods into learning representations that produce large target domain errors.\nWe evaluate the effect of these attacks on popular UDA methods using benchmark\ndatasets where they have been previously shown to be successful. Our results\nshow that poisoning can significantly decrease the target domain accuracy,\ndropping it to almost 0% in some cases, with the addition of only 10% poisoned\ndata in the source domain. The failure of these UDA methods demonstrates their\nlimitations at guaranteeing cross-domain generalization consistent with our\nlower bound. Thus, evaluating UDA methods in adversarial settings such as data\npoisoning provides a better sense of their robustness to data distributions\nunfavorable for UDA.\n"}
{"abstract": "  We study the problem of labelling effort for semantic segmentation of\nlarge-scale 3D point clouds. Existing works usually rely on densely annotated\npoint-level semantic labels to provide supervision for network training.\nHowever, in real-world scenarios that contain billions of points, it is\nimpractical and extremely costly to manually annotate every single point. In\nthis paper, we first investigate whether dense 3D labels are truly required for\nlearning meaningful semantic representations. Interestingly, we find that the\nsegmentation performance of existing works only drops slightly given as few as\n1% of the annotations. However, beyond this point (e.g. 1 per thousand and\nbelow) existing techniques fail catastrophically. To this end, we propose a new\nweak supervision method to implicitly augment the total amount of available\nsupervision signals, by leveraging the semantic similarity between neighboring\npoints. Extensive experiments demonstrate that the proposed Semantic Query\nNetwork (SQN) achieves state-of-the-art performance on six large-scale open\ndatasets under weak supervision schemes, while requiring only 1000x fewer\nlabeled points for training. The code is available at\nhttps://github.com/QingyongHu/SQN.\n"}
{"abstract": "  Elastic metamaterials are often designed for a single permanent function. We\nexplore the possibility of altering a material's function repeatedly through a\nself-organization, \"training\" process, controlled by applied strains. We show\nthat the elastic function can be altered numerous times, though each new\ntrained task imprints a memory. This ultimately leads to material degradation\nthrough the gradual reduction of the frequency gap in the density of states. We\nalso show that retraining adapts previously trained low energy modes to a new\nfunction. As a result consecutive trained responses are realized similarly. We\nshow how retraining can be exploited to attain a response that would otherwise\nbe difficult.\n"}
{"abstract": "  Spreadsheet formula prediction has been an important program synthesis\nproblem with many real-world applications. Previous works typically utilize\ninput-output examples as the specification for spreadsheet formula synthesis,\nwhere each input-output pair simulates a separate row in the spreadsheet.\nHowever, this formulation does not fully capture the rich context in real-world\nspreadsheets. First, spreadsheet data entries are organized as tables, thus\nrows and columns are not necessarily independent from each other. In addition,\nmany spreadsheet tables include headers, which provide high-level descriptions\nof the cell data. However, previous synthesis approaches do not consider\nheaders as part of the specification. In this work, we present the first\napproach for synthesizing spreadsheet formulas from tabular context, which\nincludes both headers and semi-structured tabular data. In particular, we\npropose SpreadsheetCoder, a BERT-based model architecture to represent the\ntabular context in both row-based and column-based formats. We train our model\non a large dataset of spreadsheets, and demonstrate that SpreadsheetCoder\nachieves top-1 prediction accuracy of 42.51%, which is a considerable\nimprovement over baselines that do not employ rich tabular context. Compared to\nthe rule-based system, SpreadsheetCoder assists 82% more users in composing\nformulas on Google Sheets.\n"}
{"abstract": "  We investigate the Poincar\\'e approach to computing 3d gravity partition\nfunctions dual to Rational CFT. For a single genus-1 boundary, we show that for\ncertain infinite sets of levels, the SU(2)$_k$ WZW models provide unitary\nexamples for which the Poincare series is a positive linear combination of two\nmodular-invariant partition functions. This supports the interpretation that\nthe bulk gravity theory (a topological Chern-Simons theory in this case) is\ndual to an average of distinct CFT's sharing the same Kac-Moody algebra. We\ncompute the weights of this average for all seed primaries and all relevant\nvalues of k. We then study other WZW models, notably SU($N$)$_1$ and SU(3)$_k$,\nand find that each class presents rather different features. Finally we\nconsider multiple genus-1 boundaries, where we find a class of seed functions\nfor the Poincar\\'e sum that reproduces both disconnected and connected\ncontributions -- the latter corresponding to analogues of 3-manifold\n\"wormholes\" -- such that the expected average is correctly reproduced.\n"}
{"abstract": "  We study in detail the evaluation of the gauge invariant overlap for analytic\nsolutions constructed out of elements in the $KBc$ algebra in open string field\ntheory. We compute this gauge invariant observable using analytical and\nnumerical techniques based on the sliver frame $\\mathcal{L}_0$ and traditional\nVirasoro $L_0$ level expansions of the solutions.\n"}
{"abstract": "  Three different elements, Silicon, Selenium, and Tellurium, are ion-implanted\nin Gallium arsenide to form a conducting layer that serves as a back-gate to a\nmolecular beam epitaxy (MBE) overgrown two-dimensional electron gas (2DEG).\nWhile the heavy ion Tellurium creates too many damages in the gallium arsenide\nto form a conducting layer, both Silicon and Selenium show promising results\ncombined with MBE-grown high-quality 2DEGs. Similar 2DEG mobility compared to\nnon-implanted reference samples is achieved for both Silicon and Selenium\nimplanted structures. Individual contacts to the back-gate are challenging.\nHowever, Silicon implanted structures, annealed before the MBE growth, result\nin a function back-gate, and the electron density of the 2DEG is tuned via the\nback-gate.\n"}
{"abstract": "  We report world record high data transmission over standard optical fiber\nfrom a single optical source. We achieve a line rate of 44.2 Terabits per\nsecond (Tb/s) employing only the C-band at 1550nm, resulting in a spectral\nefficiency of 10.4 bits/s/Hz. We use a new and powerful class of micro-comb\ncalled soliton crystals that exhibit robust operation and stable generation as\nwell as a high intrinsic efficiency that, together with an extremely low\nspacing of 48.9 GHz enables a very high coherent data modulation format of 64\nQAM. We achieve error free transmission across 75 km of standard optical fiber\nin the lab and over a field trial with a metropolitan optical fiber network.\nThis work demonstrates the ability of optical micro-combs to exceed other\napproaches in performance for the most demanding practical optical\ncommunications applications.\n"}
{"abstract": "  Distributed systems become more and more important to our life. Especially in\nareas like Smart Home and the Internet of Things (IoT) reliable low-power\nsensor networks become increasingly important. For ensuring this there are a\nlot of trust metrics. In this paper we compare a model of a distributed\nlow-power sensor network including one root node and the corresponding Simple\nTrust Metric to the requirements from \"Representation of Trust and Reputation\nin Self-Managed Computing Systems\" [1], the Weighted Trust Metric and the\nWeighted Simple Exponential Smoothing\n"}
{"abstract": "  Interactions with either environments or expert policies during training are\nneeded for most of the current imitation learning (IL) algorithms. For IL\nproblems with no interactions, a typical approach is Behavior Cloning (BC).\nHowever, BC-like methods tend to be affected by distribution shift. To mitigate\nthis problem, we come up with a Robust Model-Based Imitation Learning (RMBIL)\nframework that casts imitation learning as an end-to-end differentiable\nnonlinear closed-loop tracking problem. RMBIL applies Neural ODE to learn a\nprecise multi-step dynamics and a robust tracking controller via Nonlinear\nDynamics Inversion (NDI) algorithm. Then, the learned NDI controller will be\ncombined with a trajectory generator, a conditional VAE, to imitate an expert's\nbehavior. Theoretical derivation shows that the controller network can\napproximate an NDI when minimizing the training loss of Neural ODE. Experiments\non Mujoco tasks also demonstrate that RMBIL is competitive to the\nstate-of-the-art generative adversarial method (GAIL) and achieves at least 30%\nperformance gain over BC in uneven surfaces.\n"}
{"abstract": "  The mean field theory is revisited in the classical and quantum mechanical\nlimits. Taking into account the boundary conditions at the phase transition and\nthe third law of the thermodynamics the physical properties of the ordered and\ndisordered phases were reported. The equation for the order parameter predicts\nthe occurrence of a saturation of $\\Psi^2$ = 1 near $\\Theta_S$, the temperature\nbelow the quantum mechanical ground state is reached. The theoretical\npredictions are also compared with high resolution thermal expansion data of\nSrTiO$_{\\text{3}}$ monocrystalline samples and other some previous results. An\nexcellent agreement has been found suggesting a universal behavior of the\ntheoretical model to describe continuous structural phase transitions.\n"}
{"abstract": "  This paper deals with the dynamic factor analysis problem for an ARMA\nprocess. To robustly estimate the number of factors, we construct a confidence\nregion centered in a finite sample estimate of the underlying model which\ncontains the true model with a prescribed probability. In this confidence\nregion, the problem, formulated as a rank minimization of a suitable spectral\ndensity, is efficiently approximated via a trace norm convex relaxation. The\nlatter is addressed by resorting to the Lagrange duality theory, which allows\nto prove the existence of solutions. Finally, a numerical algorithm to solve\nthe dual problem is presented. The effectiveness of the proposed estimator is\nassessed through simulation studies both with synthetic and real data.\n"}
{"abstract": "  The shapes of galaxy N-point correlation functions can be used as standard\nrulers to constrain the distance-redshift relationship and thence the expansion\nrate of the Universe. The cosmological density fields traced by late-time\ngalaxy formation are initially nearly Gaussian, and hence all the cosmological\ninformation can be extracted from their 2-Point Correlation Function (2PCF) or\nits Fourier-space analog the power spectrum. Subsequent nonlinear evolution\nunder gravity, as well as halo and then galaxy formation, generate higher-order\ncorrelation functions. Since the mapping of the initial to the final density\nfield is, on large scales, invertible, it is often claimed that the information\ncontent of the initial field's power spectrum is equal to that of all the\nhigher-order functions of the final, nonlinear field. This claim implies that\nreconstruction of the initial density field from the nonlinear field renders\nanalysis of higher-order correlation functions of the latter superfluous. We\nhere show that this claim is false when the N-point functions are used as\nstandard rulers. Constraints available from joint analysis of the galaxy power\nspectrum and bispectrum (Fourier-space analog of the 3-Point Correlation\nFunction) can, in some cases, exceed those offered by the initial power\nspectrum even when the reconstruction is perfect. We provide a mathematical\njustification for this claim and also demonstrate it using a large suite of\nN-body simulations. In particular, we show that for the z = 0 real-space matter\nfield in the limit of vanishing shot noise, taking modes up to k_max = 0.2\nh/Mpc, using the bispectrum alone offers a factor of two reduction in the\nvariance on the cosmic distance scale relative to that available from the power\nspectrum.\n"}
{"abstract": "  Lithium-ion batteries are ubiquitous in modern day applications ranging from\nportable electronics to electric vehicles. Irrespective of the application,\nreliable real-time estimation of battery state of health (SOH) by on-board\ncomputers is crucial to the safe operation of the battery, ultimately\nsafeguarding asset integrity. In this paper, we design and evaluate a machine\nlearning pipeline for estimation of battery capacity fade - a metric of battery\nhealth - on 179 cells cycled under various conditions. The pipeline estimates\nbattery SOH with an associated confidence interval by using two parametric and\ntwo non-parametric algorithms. Using segments of charge voltage and current\ncurves, the pipeline engineers 30 features, performs automatic feature\nselection and calibrates the algorithms. When deployed on cells operated under\nthe fast-charging protocol, the best model achieves a root mean squared percent\nerror of 0.45\\%. This work provides insights into the design of scalable\ndata-driven models for battery SOH estimation, emphasising the value of\nconfidence bounds around the prediction. The pipeline methodology combines\nexperimental data with machine learning modelling and can be generalized to\nother critical components that require real-time estimation of SOH.\n"}
{"abstract": "  The memory wall bottleneck is a key challenge across many data-intensive\napplications. Multi-level FeFET-based embedded non-volatile memories are a\npromising solution for denser and more energy-efficient on-chip memory.\nHowever, reliable multi-level cell storage requires careful optimizations to\nminimize the design overhead costs. In this work, we investigate the interplay\nbetween FeFET device characteristics, programming schemes, and memory array\narchitecture, and explore different design choices to optimize performance,\nenergy, area, and accuracy metrics for critical data-intensive workloads. From\nour cross-stack design exploration, we find that we can store DNN weights and\nsocial network graphs at a density of over 8MB/mm^2 and sub-2ns read access\nlatency without loss in application accuracy.\n"}
{"abstract": "  In this research, we present a comprehensive, longitudinal empirical summary\nof the R package ecosystem, including not just CRAN, but also Bioconductor and\nGitHub. We analyze more than 25,000 packages, 150,000 releases, and 15 million\nfiles across two decades, providing comprehensive counts and trends for common\nmetrics across packages, releases, authors, licenses, and other important\nmetadata. We find that the historical growth of the ecosystem has been robust\nunder all measures, with a compound annual growth rate of 29% for active\npackages, 28% for new releases, and 26% for active maintainers. As with many\nsimilar social systems, we find a number of highly right-skewed distributions\nwith practical implications, including the distribution of releases per\npackage, packages and releases per author or maintainer, package and maintainer\ndependency in-degree, and size per package and release. For example, the top\nfive packages are imported by nearly 25% of all packages, and the top ten\nmaintainers support packages that are imported by over half of all packages. We\nalso highlight the dynamic nature of the ecosystem, recording both dramatic\nacceleration and notable deceleration in the growth of R. From a licensing\nperspective, we find a notable majority of packages are distributed under\ncopyleft licensing or omit licensing information entirely. The data, methods,\nand calculations herein provide an anchor for public discourse and industry\ndecisions related to R and CRAN, serving as a foundation for future research on\nthe R software ecosystem and \"data science\" more broadly.\n"}
{"abstract": "  Matched Field Processing (MFP) locates the underwater sources by matching the\nreceived data with the replica vectors, which could be regarded as a\ngeneralized beamformer. In this paper, the MFP method is combined with a\nrecently developed framework -- Graph Signal Processing (GSP) method. Following\nthe paradigm of GSP, a spatial adjacency matrix is constructed for the\narbitrary distributed sensors based on the Green's function, then the source is\nlocated by utilizing the graph Fourier transform. The simulation results\nillustrate that the Graph-based MFP outperforms the the conventional MFP\nprocessors -- the Bartlett processor and the Minimum Variance processor -- for\nits good accuracy and robustness.\n"}
{"abstract": "  We explore whether assumptions about dust grain shape affect resulting\nestimates of the composition and grain size distribution of the AU Microscopii\n(AU Mic) debris disk from scattered light data collected by Lomax et al.\n(2018). The near edge-on orientation of the AU Mic debris disk makes it ideal\nfor studying the effect of the scattering phase function (SPF) on the measured\nflux ratios as a function of wavelength and projected distance. Previous\nefforts to model the AU Mic debris disk have invoked a variety of dust grain\ncompositions and explored the effect of porosity, but did not undertake a\nsystematic effort to explore a full range of size distributions and\ncompositions to understand possible degeneracies in fitting the data. The\ndegree to which modelling dust grains with more realistic shapes compounds\nthese degeneracies has also not previously been explored. We find differences\nin the grain properties retrieved depending on the grain shape model used. We\nalso present here our calculations of porous grains of size parameters x = 0.1\nto 48 and complex refractive indices (m = n+ik) ranging from n = 1.1 to 2.43\nand k = 0 to 1.0, covering multiple compositions at visible and near infrared\nwavelengths such as ice, silicates, amorphous carbon, and tholins.\n"}
{"abstract": "  Existing table question answering datasets contain abundant factual questions\nthat primarily evaluate the query and schema comprehension capability of a\nsystem, but they fail to include questions that require complex reasoning and\nintegration of information due to the constraint of the associated short-form\nanswers. To address these issues and to demonstrate the full challenge of table\nquestion answering, we introduce FeTaQA, a new dataset with 10K Wikipedia-based\n{table, question, free-form answer, supporting table cells} pairs. FeTaQA\nyields a more challenging table question answering setting because it requires\ngenerating free-form text answers after retrieval, inference, and integration\nof multiple discontinuous facts from a structured knowledge source. Unlike\ndatasets of generative QA over text in which answers are prevalent with copies\nof short text spans from the source, answers in our dataset are human-generated\nexplanations involving entities and their high-level relations. We provide two\nbenchmark methods for the proposed task: a pipeline method based on\nsemantic-parsing-based QA systems and an end-to-end method based on large\npretrained text generation models, and show that FeTaQA poses a challenge for\nboth methods.\n"}
{"abstract": "  In the context of Wilsonian Renormalization, renormalization group (RG) flows\nare a set of differential equations that defines how the coupling constants of\na theory depend on an energy scale. These equations closely resemble\nthermodynamical equations and how thermodynamical systems are related to\ntemperature. In this sense, it is natural to look for structures in the flows\nthat show a thermodynamics-like behaviour. The mathematical theory to study\nthese equations is called Dynamical Systems, and applications of that have been\nused to study RG flows. For example, the classical Zamolodchikov's C-Theorem\nand its higher-dimensional counterparts, that show that there is a\nmonotonically decreasing function along the flow and it is a property that\nresembles the second-law of thermodynamics, is related to the Lyapunov function\nin the context of Dynamical Systems. It can be used to rule out exotic\nasymptotic behaviours like periodic flows (also known as limit cycles). We also\nstudy bifurcation theory and index theories, which have been proposed to be\nuseful in the study of RG flows, the former can be used to explain couplings\ncrossing through marginality and the latter to extract global information about\nthe space the flows lives in. In this dissertation, we also look for\napplications in holographic RG flows and we try to see if the structural\nbehaviours in holographic theories are the same as the ones in the dual field\ntheory side.\n"}
{"abstract": "  We study the dynamics of black holes in Einstein-scalar-Gauss-Bonnet theories\nthat exhibit spontaneous black hole scalarization using recently introduced\nmethods for solving the full, non-perturbative equations of motion. For one\nsign of the coupling parameter, non-spinning vacuum black holes are unstable to\ndeveloping scalar hair, while for the other, instability only sets in for black\nholes with sufficiently large spin. We study scalarization in both cases,\ndemonstrating that there is a range of parameter space where the theory\nmaintains hyperbolic evolution and for which the instability saturates in a\nscalarized black hole that is stable without symmetry assumptions. However,\nthis parameter space range is significantly smaller than the range for which\nstationary scalarized black hole solutions exist. We show how different choices\nfor the subleading behavior of the Gauss-Bonnet coupling affect the dynamics of\nthe instability and the final state, or lack thereof. Finally, we present\nmergers of binary black holes and demonstrate the imprint of the scalar hair in\nthe gravitational radiation.\n"}
{"abstract": "  We study here the conditions to perform the distribution of a pure state on a\nquantum network using quantum operations which can succeed with a non-zero\nprobability, the Stochastic Local Operation and Classical Communication (SLOCC)\noperations.\n  In their pioneering 2010 work, Kobayashi et al. showed how to convert any\nclassical network coding protocol into a quantum network coding protocol.\nHowever, they left open whether the existence of a quantum network coding\nprotocol implied the existence of a classical one. Motivated by this question,\nwe characterize the set of distribution tasks achievable with non zero\nprobability for both classical and quantum networks. We develop a formalism\nwhich encompasses both types of distribution protocols by reducing the solving\nof a distribution task to the factorization of a tensor with complex\ncoefficients or real positive ones. Using this formalism, we examine the\nequivalences and differences between both types of distribution protocols\nexhibiting several elementary and fundamental relations between them as well as\nconcrete examples of both convergence and divergence. We answer by the negative\nto the issue previously left open: some tasks are achievable in the quantum\nsetting, but not in the classical one. We believe this formalism to be a useful\ntool for studying the extent of quantum network ability to perform multipartite\ndistribution tasks.\n"}
{"abstract": "  We reexpress the superfluid density of a disordered superconductor obtained\nby two of us earlier [Phys. Rev. B 102, 024514 (2020)] in a new highly\nconvergent form, and use the results to make an extensive and successful\ncomparison with experiment in the dirty limit for all temperatures. We point\nout that there is a regime (conventional superconductor with low, but\nincreasing disorder) where theoretical predictions need to be confronted with\naccurate experiment.\n"}
{"abstract": "  A large number of consensus algorithms have been proposed. However, the\nrequirement of strict consistency limits their wide adoption, especially in\nhigh-performance required systems. In this paper, we propose a weak consensus\nalgorithm that only maintains the consistency of relative positions between the\nmessages. We apply this consensus algorithm to construct a high-performance\nblockchain system, called \\textit{Sphinx}. We implement the system with 32k+\nlines of code including all components like consensus/P2P/ledger/etc. The\nevaluations show that Sphinx can reach a peak throughput of 43k TPS (with 8\nfull nodes), which is significantly faster than current blockchain systems such\nas Ethereum given the same experimental environment. To the best of our\nknowledge, we present the first weak consensus algorithm with a fully\nimplemented blockchain system.\n"}
{"abstract": "  Hate speech on social media is a growing concern, and automated methods have\nso far been sub-par at reliably detecting it. A major challenge lies in the\npotentially evasive nature of hate speech due to the ambiguity and fast\nevolution of natural language. To tackle this, we introduce a vectorisation\nbased on a crowd-sourced and continuously updated dictionary of hate words and\npropose fusing this approach with standard word embedding in order to improve\nthe classification performance of a CNN model. To train and test our model we\nuse a merge of two established datasets (110,748 tweets in total). By adding\nthe dictionary-enhanced input, we are able to increase the CNN model's\npredictive power and increase the F1 macro score by seven percentage points.\n"}
{"abstract": "  This work focuses on the numerical solution of hyperbolic conservations laws\n(possibly endowed with a source term) using the Active Flux method. This method\nis an extension of the finite volume method. Instead of solving a Riemann\nProblem, the Active Flux method uses actively evolved point values along the\ncell boundary in order to compute the numerical flux. Early applications of the\nmethod were linear equations with an available exact solution operator, and\nActive Flux was shown to be structure preserving in such cases. For nonlinear\nPDEs or balance laws, exact evolution operators generally are unavailable.\nHere, strategies are shown how sufficiently accurate approximate evolution\noperators can be designed which allow to make Active Flux structure preserving\n/ well-balanced for nonlinear problems.\n"}
{"abstract": "  For the minimization of state-based systems (i.e. the reduction of the number\nof states while retaining the system's semantics), there are two obvious\naspects: removing unnecessary states of the system and merging redundant states\nin the system. In the present article, we relate the two aspects on coalgebras\nby defining an abstract notion of minimality. The abstract notion minimality\nand minimization live in a general category with a factorization system. We\nwill find criteria on the category that ensure uniqueness, existence, and\nfunctoriality of the minimization aspects. The proofs of these results\ninstantiate to those for reachability and observability minimization in the\nstandard coalgebra literature. Finally, we will see how the two aspects of\nminimization interact and under which criteria they can be sequenced in any\norder, like in automata minimization.\n"}
{"abstract": "  A problem that is frequently encountered in a variety of mathematical\ncontexts, is to find the common invariant subspaces of a single, or set of\nmatrices. In a recent paper, a new method was proposed that gives an answer to\nthat problem. In the present article conditions are given for the more general\nproblem of putting set of matrices into block-diagonal form simultaneously. In\naddition, two algorithms for simultaneous block triangularization and block\ndiagonalization of sets of matrices are presented and are applied to several\nconcrete examples using the symbolic manipulation system Maple.\n"}
{"abstract": "  This paper presents our solution for ICDAR 2021 competition on scientific\nliterature parsing taskB: table recognition to HTML. In our method, we divide\nthe table content recognition task into foursub-tasks: table structure\nrecognition, text line detection, text line recognition, and box assignment.Our\ntable structure recognition algorithm is customized based on MASTER [1], a\nrobust image textrecognition algorithm. PSENet [2] is used to detect each text\nline in the table image. For text linerecognition, our model is also built on\nMASTER. Finally, in the box assignment phase, we associatedthe text boxes\ndetected by PSENet with the structure item reconstructed by table structure\nprediction,and fill the recognized content of the text line into the\ncorresponding item. Our proposed methodachieves a 96.84% TEDS score on 9,115\nvalidation samples in the development phase, and a 96.32%TEDS score on 9,064\nsamples in the final evaluation phase.\n"}
{"abstract": "  Recently it was suggested that certain perturbations of integrable spin\nchains lead to a weak breaking of integrability in the sense that integrability\nis preserved at the first order in the coupling. Here we examine this claim\nusing level spacing distribution. We find that the volume dependent crossover\nbetween integrable and chaotic level spacing statistics which marks the onset\nof quantum chaotic behaviour, is markedly different for weak vs. strong\nbreaking of integrability. In particular, for the gapless case we find that the\ncrossover coupling as a function of the volume $L$ scales with a $1/L^2$ law\nfor weak breaking as opposed to the $1/L^3$ law previously found for the strong\ncase.\n"}
{"abstract": "  The objective of person re-identification (re-ID) is to retrieve a person's\nimages from an image gallery, given a single instance of the person of\ninterest. Despite several advancements, learning discriminative\nidentity-sensitive and viewpoint invariant features for robust Person\nRe-identification is a major challenge owing to the large pose variation of\nhumans. This paper proposes a re-ID pipeline that utilizes the image generation\ncapability of Generative Adversarial Networks combined with pose clustering and\nfeature fusion to achieve pose invariant feature learning. The objective is to\nmodel a given person under different viewpoints and large pose changes and\nextract the most discriminative features from all the appearances. The pose\ntransformational GAN (pt-GAN) module is trained to generate a person's image in\nany given pose. In order to identify the most significant poses for\ndiscriminative feature extraction, a Pose Clustering module is proposed. The\ngiven instance of the person is modelled in varying poses and these features\nare effectively combined through the Feature Fusion Network. The final re-ID\nmodel consisting of these 3 sub-blocks, alleviates the pose dependence in\nperson re-ID. Also, The proposed model is robust to occlusion, scale, rotation\nand illumination, providing a framework for viewpoint invariant feature\nlearning. The proposed method outperforms the state-of-the-art GAN based models\nin 4 benchmark datasets. It also surpasses the state-of-the-art models that\nreport higher re-ID accuracy in terms of improvement over baseline.\n"}
{"abstract": "  In this paper, we investigate qualitative integrals (generalizations of\nSugeno integral) acting on recently introduced Dragonfly algebras. These\nalgebras are designed for applications in data analysis (based on fuzzy\nrelational compositions) when some data are unknown (e.g., missing). Unknown\ndata are represented by the additional dummy value. Definitions of operations\non Dragonfly algebras follow lower estimation strategy, i.e., results of\noperations can be interpreted as lower estimations of results obtained when\nunknown values are replaced by known ones. We define qualitative integrals on\nDragonfly algebras and show their monotonicity. We also prove results\ncharacterizing when these integrals return known (or unknown) values. We also\npresent an illustrative example and hint directions of further research.\n"}
{"abstract": "  The laborious process of labeling data often bottlenecks projects that aim to\nleverage the power of supervised machine learning. Active Learning (AL) has\nbeen established as a technique to ameliorate this condition through an\niterative framework that queries a human annotator for labels of instances with\nthe most uncertain class assignment. Via this mechanism, AL produces a binary\nclassifier trained on less labeled data but with little, if any, loss in\npredictive performance. Despite its advantages, AL can have difficulty with\nclass-imbalanced datasets and results in an inefficient labeling process. To\naddress these drawbacks, we investigate our unsupervised instance selection\n(UNISEL) technique followed by a Random Forest (RF) classifier on 10 outlier\ndetection datasets under low-label conditions. These results are compared to AL\nperformed on the same datasets. Further, we investigate the combination of\nUNISEL and AL. Results indicate that UNISEL followed by an RF performs\ncomparably to AL with an RF and that the combination of UNISEL and AL\ndemonstrates superior performance. The practical implications of these findings\nin terms of time savings and generalizability afforded by UNISEL are discussed.\n"}
{"abstract": "  We construct a class of knot solutions of the gravitoelectromagnetic (GEM)\nequations in vacuum in the linearized gravity approximation by analogy with the\nRa\\~{n}ada-Hopf fields. For these solutions, the dual metric tensors of the\nbi-metric geometry of the gravitational vacuum with knot perturbations are\ngiven and the geodesic equation as a function of two complex parameters of the\nGEM knots are calculated. Finally, the Landau--Lifshitz pseudo-tensor and a\nscalar invariant of the GEM knots are computed.\n"}
{"abstract": "  Sound event localization and detection consists of two subtasks which are\nsound event detection and direction-of-arrival estimation. While sound event\ndetection mainly relies on time-frequency patterns to distinguish different\nsound classes, direction-of-arrival estimation uses magnitude or phase\ndifferences between microphones to estimate source directions. Therefore, it is\noften difficult to jointly train these two subtasks simultaneously. We propose\na novel feature called spatial cue-augmented log-spectrogram (SALSA) with exact\ntime-frequency mapping between the signal power and the source\ndirection-of-arrival. The feature includes multichannel log-spectrograms\nstacked along with the estimated direct-to-reverberant ratio and a normalized\nversion of the principal eigenvector of the spatial covariance matrix at each\ntime-frequency bin on the spectrograms. Experimental results on the DCASE 2021\ndataset for sound event localization and detection with directional\ninterference showed that the deep learning-based models trained on this new\nfeature outperformed the DCASE challenge baseline by a large margin. We\ncombined several models with slightly different architectures that were trained\non the new feature to further improve the system performances for the DCASE\nsound event localization and detection challenge.\n"}
{"abstract": "  Image registration as an important basis in signal processing task often\nencounter the problem of stability and efficiency. Non-learning registration\napproaches rely on the optimization of the similarity metrics between the fix\nand moving images. Yet, those approaches are usually costly in both time and\nspace complexity. The problem can be worse when the size of the image is large\nor the deformations between the images are severe. Recently, deep learning, or\nprecisely saying, the convolutional neural network (CNN) based image\nregistration methods have been widely investigated in the research community\nand show promising effectiveness to overcome the weakness of non-learning based\nmethods. To explore the advanced learning approaches in image registration\nproblem for solving practical issues, we present in this paper a method of\nintroducing attention mechanism in deformable image registration problem. The\nproposed approach is based on learning the deformation field with a Transformer\nframework (AiR) that does not rely on the CNN but can be efficiently trained on\nGPGPU devices also. In a more vivid interpretation: we treat the image\nregistration problem as the same as a language translation task and introducing\na Transformer to tackle the problem. Our method learns an unsupervised\ngenerated deformation map and is tested on two benchmark datasets. The source\ncode of the AiR will be released at Gitlab.\n"}
{"abstract": "  Random feature methods have been successful in various machine learning\ntasks, are easy to compute, and come with theoretical accuracy bounds. They\nserve as an alternative approach to standard neural networks since they can\nrepresent similar function spaces without a costly training phase. However, for\naccuracy, random feature methods require more measurements than trainable\nparameters, limiting their use for data-scarce applications or problems in\nscientific machine learning. This paper introduces the sparse random feature\nexpansion to obtain parsimonious random feature models. Specifically, we\nleverage ideas from compressive sensing to generate random feature expansions\nwith theoretical guarantees even in the data-scarce setting. In particular, we\nprovide generalization bounds for functions in a certain class (that is dense\nin a reproducing kernel Hilbert space) depending on the number of samples and\nthe distribution of features. The generalization bounds improve with additional\nstructural conditions, such as coordinate sparsity, compact clusters of the\nspectrum, or rapid spectral decay. In particular, by introducing sparse\nfeatures, i.e. features with random sparse weights, we provide improved bounds\nfor low order functions. We show that the sparse random feature expansions\noutperforms shallow networks in several scientific machine learning tasks.\n"}
{"abstract": "  A framework for the generation of synthetic time-series transmission-level\nload data is presented. Conditional generative adversarial networks are used to\nlearn the patterns of a real dataset of hourly-sampled week-long load profiles\nand generate unique synthetic profiles on demand, based on the season and type\nof load required. Extensive testing of the generative model is performed to\nverify that the synthetic data fully captures the characteristics of real loads\nand that it can be used for downstream power system and/or machine learning\napplications.\n"}
{"abstract": "  The weights of a deep neural network model are optimized in conjunction with\nthe governing flow equations to provide a model for sub-grid-scale stresses in\na temporally developing plane turbulent jet at Reynolds number $Re_0=6\\,000$.\nThe objective function for training is first based on the instantaneous\nfiltered velocity fields from a corresponding direct numerical simulation, and\nthe training is by a stochastic gradient descent method, which uses the adjoint\nNavier--Stokes equations to provide the end-to-end sensitivities of the model\nweights to the velocity fields. In-sample and out-of-sample testing on multiple\ndual-jet configurations show that its required mesh density in each coordinate\ndirection for prediction of mean flow, Reynolds stresses, and spectra is half\nthat needed by the dynamic Smagorinsky model for comparable accuracy. The same\nneural-network model trained directly to match filtered sub-grid-scale stresses\n-- without the constraint of being embedded within the flow equations during\nthe training -- fails to provide a qualitatively correct prediction. The\ncoupled formulation is generalized to train based only on mean-flow and\nReynolds stresses, which are more readily available in experiments. The\nmean-flow training provides a robust model, which is important, though a\nsomewhat less accurate prediction for the same coarse meshes, as might be\nanticipated due to the reduced information available for training in this case.\nThe anticipated advantage of the formulation is that the inclusion of resolved\nphysics in the training increases its capacity to extrapolate. This is assessed\nfor the case of passive scalar transport, for which it outperforms established\nmodels due to improved mixing predictions.\n"}
{"abstract": "  Track radii Re induced by swift heavy ions are studied. The experimental\ntrack data are analyzed in Gd2Ti2O7 pyrochlore in the range Se=6-54 keV/nm (Se\n- electronic stopping power) including irradiations by low and high velocity\nions. The Analytical Thermal Spike Model (ATSM) is applied whose main features\nare reviewed. The track data of Gd2Ti2O7 exhibit scaling features which mean\nsimple quantitative relationships with track sizes of other insulators\ncontrolled only by a single materials parameter (MP), the melting temperature\nTm. The Re2-Se track evolution curve is described in the whole range of Se\nwithout the application of any individual fitting parameter and MPs apart Tm;\nthe threshold for track formation Set=6.1/14.4 keV/nm for E<2 MeV/nucleon and\nE>8 MeV/nucleon, respectively. The unique role of Tm is a highly important\nlimitation for the mechanism of track formation. The practical consequences of\nthe results are discussed with respect of the estimation of irradiation\nparameters\n"}
{"abstract": "  What role does culture play in determining institutions in a country? This\npaper argues that the establishment of institutions is a process originating\npredominantly in a nation's culture and tries to discern the role of a cultural\nbackground in the governance of countries. We use the six Hofstede's Cultural\nDimensions and the six Worldwide Governance Indicators to test the strength of\nthe relationship on 94 countries between 1996 and 2019. We find that the\nstrongest cultural characteristics are Power Distance with negative effect on\ngovernance and Long-Term Orientation with positive effect. We also determine\nhow well countries transform their cultural characteristics into institutions\nusing stochastic frontier analysis.\n"}
{"abstract": "  We introduce the notion of adaptive synchronisation for pushdown automata, in\nwhich there is an external observer who has no knowledge about the current\nstate of the pushdown automaton, but can observe the contents of the stack. The\nobserver would then like to decide if it is possible to bring the automaton\nfrom any state into some predetermined state by giving inputs to it in an\n\\emph{adaptive} manner, i.e., the next input letter to be given can depend on\nhow the contents of the stack changed after the current input letter. We show\nthat for non-deterministic pushdown automata, this problem is\n2-EXPTIME-complete and for deterministic pushdown automata, we show\nEXPTIME-completeness.\n  To prove the lower bounds, we first introduce (different variants of)\nsubset-synchronisation and show that these problems are polynomial-time\nequivalent with the adaptive synchronisation problem. We then prove hardness\nresults for the subset-synchronisation problems. For proving the upper bounds,\nwe consider the problem of deciding if a given alternating pushdown system has\nan accepting run with at most $k$ leaves and we provide an $n^{O(k^2)}$ time\nalgorithm for this problem.\n"}
{"abstract": "  We provide the largest and most homogeneous sample of $\\alpha$-element (Mg,\nCa, Ti) and iron abundances for field RR Lyrae (RRLs, 162 variables) by using\nhigh-resolution spectra. The current measurements were complemented with\nsimilar abundances available in the literature for 46 field RRLs brought to our\nmetallicity scale. We ended up with a sample of old (t$\\ge$ 10 Gyr), low-mass\nstellar tracers (208 RRLs: 169 fundamental, 38 first overtone, 1 mixed mode)\ncovering three dex in iron abundance (-3.00$\\le$[Fe/H]$\\le$0.24). We found that\nfield RRLs are $\\sim$0.3 dex more $\\alpha$-poor than typical Halo tracers in\nthe metal-rich regime, ([Fe/H]$\\ge$-1.2) while in the metal-poor regime\n([Fe/H]$\\le$-2.2) they seem to be on average $\\sim$0.1 dex more\n$\\alpha$-enhanced. This is the first time that the depletion in\n$\\alpha$-elements for solar iron abundances is detected on the basis of a\nlarge, homogeneous and coeval sample of old stellar tracers. Interestingly, we\nalso detected a close similarity in the [$\\alpha$/Fe] trend between\n$\\alpha$-poor, metal-rich RRLs and red giants (RGs) in the Sagittarius dwarf\ngalaxy as well as between $\\alpha$-enhanced, metal-poor RRLs and RGs in ultra\nfaint dwarf galaxies. These results are supported by similar elemental\nabundances for 46 field Horizontal Branch (HB) stars. These stars share with\nRRLs the same evolutionary phase and the same progenitors. This evidence\nfurther supports the key role that old stellar tracers play in constraining the\nearly chemical enrichment of the Halo and, in particular, in investigating the\nimpact that dwarf galaxies have had in the mass assembly of the Galaxy.\n"}
{"abstract": "  Spatio-temporal action detection is an important and challenging problem in\nvideo understanding. The existing action detection benchmarks are limited in\naspects of small numbers of instances in a trimmed video or low-level atomic\nactions. This paper aims to present a new multi-person dataset of\nspatio-temporal localized sports actions, coined as MultiSports. We first\nanalyze the important ingredients of constructing a realistic and challenging\ndataset for spatio-temporal action detection by proposing three criteria: (1)\nmulti-person scenes and motion dependent identification, (2) with well-defined\nboundaries, (3) relatively fine-grained classes of high complexity. Based on\nthese guide-lines, we build the dataset of MultiSports v1.0 by selecting 4\nsports classes, collecting 3200 video clips, and annotating 37701 action\ninstances with 902k bounding boxes. Our datasets are characterized with\nimportant properties of high diversity, dense annotation, and high quality. Our\nMulti-Sports, with its realistic setting and detailed annotations, exposes the\nintrinsic challenges of spatio-temporal action detection. To benchmark this, we\nadapt several baseline methods to our dataset and give an in-depth analysis on\nthe action detection results in our dataset. We hope our MultiSports can serve\nas a standard benchmark for spatio-temporal action detection in the future. Our\ndataset website is at https://deeperaction.github.io/multisports/.\n"}
{"abstract": "  We use a 1030 nm laser with 7 ps pulse duration and average power up to 100 W\nto ablate pyramid-shape subwavelength structures (SWS) on alumina and sapphire.\nThe SWS give an effective and cryogenically robust anti-reflection coating in\nthe millimeter-wave band. We demonstrate average ablation rate of up to 34\nmm$^3$/min and 20 mm$^3$/min for structure heights of 900 $\\mu$m and 750 $\\mu$m\non alumina and sapphire, respectively. These rates are a factor of 34 and 9\nhigher than reported previously on similar structures. We propose a model that\nrelates structure height to cumulative laser fluence. The model depends on the\nabsorption length $\\delta$, which is assumed to depend on peak fluence, and on\nthe threshold fluence $\\phi_{th}$. Using a best-fit procedure we find an\naverage $\\delta = 630$ nm and 650 nm, and $\\phi_{th} = 2.0^{+0.5}_{-0.5}$\nJ/cm$^2$ and $2.3^{+0.1}_{-0.1}$ J/cm$^2$ for alumina and sapphire,\nrespectively, for peak fluence values between 30 and 70 J/cm$^{2}$. With the\nbest fit values, the model and data values for cumulative fluence agree to\nwithin 10%. Given inputs for $\\delta$ and $\\phi_{th}$ the model is used to\npredict average ablation rates as a function of SWS height and average laser\npower.\n"}
{"abstract": "  A co-polycondensation reaction is discussed analytically and by Monte-Carlo\nsimulations where two reactive units compete for reactions with an alternating\nthird reactive unit, whereby irreversible reactions replace bonds which are\nable to undergo exchange reactions. The resulting number average molar mass,\n$M_{\\text{n}}$, exhibits only one distinct peak at the stoichiometric condition\nof both competitors with the alternating partner. The weight average molar\nmass, $M_{\\text{w}}$, reaches an additional second peak at the stoichiometric\ncondition between the dominating competitor and the alternating partner. Both\npeaks of $M_{\\text{w}}$ surround a range of compositions where a rather high\nand approximately constant $M_{\\text{w}}$ is obtained. The degree of\npolymerization of the dominating and alternating reaction partners is rather\ninsensitive towards composition fluctuations if the reaction mixture remains\nwithin this composition window. This promotes high molecular weight species and\nmore homogeneous weight distributions at incomplete mixing conditions. An ideal\nreference case (identical reaction rates for all reactions) is solved\nanalytically to describe these reactions. The position of the stable\ncomposition window and the average molar masses inside this window can be tuned\nby choosing appropriate precursor molecules, reaction mixtures or post-tuning\nsteps at later times.\n"}
{"abstract": "  The training of Generative Adversarial Networks (GANs) requires a large\namount of data, stimulating the development of new augmentation methods to\nalleviate the challenge. Oftentimes, these methods either fail to produce\nenough new data or expand the dataset beyond the original manifold. In this\npaper, we propose a new augmentation method that guarantees to keep the new\ndata within the original data manifold thanks to the optimal transport theory.\nThe proposed algorithm finds cliques in the nearest-neighbors graph and, at\neach sampling iteration, randomly draws one clique to compute the Wasserstein\nbarycenter with random uniform weights. These barycenters then become the new\nnatural-looking elements that one could add to the dataset. We apply this\napproach to the problem of landmarks detection and augment the available\nannotation in both unpaired and in semi-supervised scenarios. Additionally, the\nidea is validated on cardiac data for the task of medical segmentation. Our\napproach reduces the overfitting and improves the quality metrics beyond the\noriginal data outcome and beyond the result obtained with popular modern\naugmentation methods.\n"}
{"abstract": "  RGB-infrared person re-identification is a challenging task due to the\nintra-class variations and cross-modality discrepancy. Existing works mainly\nfocus on learning modality-shared global representations by aligning image\nstyles or feature distributions across modalities, while local feature from\nbody part and relationships between person images are largely neglected. In\nthis paper, we propose a Dual-level (i.e., local and global) Feature Fusion\n(DF^2) module by learning attention for discriminative feature from local to\nglobal manner. In particular, the attention for a local feature is determined\nlocally, i.e., applying a learned transformation function on itself. Meanwhile,\nto further mining the relationships between global features from person images,\nwe propose an Affinities Modeling (AM) module to obtain the optimal intra- and\ninter-modality image matching. Specifically, AM employes intra-class\ncompactness and inter-class separability in the sample similarities as\nsupervised information to model the affinities between intra- and\ninter-modality samples. Experimental results show that our proposed method\noutperforms state-of-the-arts by large margins on two widely used\ncross-modality re-ID datasets SYSU-MM01 and RegDB, respectively.\n"}
{"abstract": "  The coded caching problem with secrecy constraint i.e., the users should not\nbe able to gain any information about the content of the files that they did\nnot demand, is known as the secretive coded caching problem. This was proposed\nby Ravindrakumar et al. in the paper titled ``Private Coded Caching'' that\nappeared in \\emph{ IEEE Transactions on Information Forensics and Security},\n2018 and is characterised by subpacketization levels growing exponentially with\nthe number of users. In the context of coded caching without secrecy, coded\ncaching schemes at subexponential subpacketization levels are feasible by\nrepresenting the caching system in the form of a Placement Delivery Array (PDA)\nand designing placement and delivery policies from it. Motivated by this, we\npropose a secretive coded caching scheme with low subpacketization using PDA,\nfor users with dedicated caches in the centralized setting. When our scheme is\napplied to a special class of PDA known as MN PDA, the scheme proposed by\nRavindrakumar et al. is recovered.\n"}
{"abstract": "  Andromeda XXI (And XXI) has been proposed as a dwarf spheroidal galaxy with a\ncentral dark matter density that is lower than expected in the Standard\n$\\Lambda$ Cold Dark Matter ($\\Lambda$CDM) cosmology. In this work, we present\ndynamical observations for 77 member stars in this system, more than doubling\nprevious studies to determine whether this galaxy is truly a low density\noutlier. We measure a systemic velocity of $v_r=-363.4\\pm1.0\\,{\\rm kms}^{-1}$\nand a velocity dispersion of $\\sigma_v=6.1^{+1.0}_{-0.9}\\,{\\rm kms}^{-1}$,\nconsistent with previous work and within $1\\sigma$ of predictions made within\nthe modified Newtonian dynamics framework. We also measure the metallicity of\nour member stars from their spectra, finding a mean value of ${\\rm\n[Fe/H]}=-1.7\\pm0.1$~dex. We model the dark matter density profile of And~XXI\nusing an improved version of \\GravSphere, finding a central density of\n$\\rho_{\\rm DM}({\\rm 150 pc})=2.7_{-1.7}^{+2.7} \\times 10^7 \\,{\\rm\nM_\\odot\\,kpc^{-3}}$ at 68\\% confidence, and a density at two half light radii\nof $\\rho_{\\rm DM}({\\rm 1.75 kpc})=0.9_{-0.2}^{+0.3} \\times 10^5 \\,{\\rm\nM_\\odot\\,kpc^{-3}}$ at 68\\% confidence. These are both a factor ${\\sim}3-5$\nlower than the densities expected from abundance matching in $\\Lambda$CDM. We\nshow that this cannot be explained by `dark matter heating' since And~XXI had\ntoo little star formation to significantly lower its inner dark matter density,\nwhile dark matter heating only acts on the profile inside the half light\nradius. However, And~XXI's low density can be accommodated within $\\Lambda$CDM\nif it experienced extreme tidal stripping (losing $>95\\%$ of its mass), or if\nit inhabits a low concentration halo on a plunging orbit that experienced\nrepeated tidal shocks.\n"}
{"abstract": "  A new non-ergodic ground-motion model (GMM) for effective amplitude spectral\n($EAS$) values for California is presented in this study. $EAS$, which is\ndefined in Goulet et al. (2018), is a smoothed rotation-independent Fourier\namplitude spectrum of the two horizontal components of an acceleration time\nhistory. The main motivation for developing a non-ergodic $EAS$ GMM, rather\nthan a spectral acceleration GMM, is that the scaling of $EAS$ does not depend\non spectral shape, and therefore, the more frequent small magnitude events can\nbe used in the estimation of the non-ergodic terms.\n  The model is developed using the California subset of the NGAWest2 dataset\nAncheta et al. (2013). The Bayless and Abrahamson (2019b) (BA18) ergodic $EAS$\nGMM was used as backbone to constrain the average source, path, and site\nscaling. The non-ergodic GMM is formulated as a Bayesian hierarchical model:\nthe non-ergodic source and site terms are modeled as spatially varying\ncoefficients following the approach of Landwehr et al. (2016), and the\nnon-ergodic path effects are captured by the cell-specific anelastic\nattenuation attenuation following the approach of Dawood and Rodriguez-Marek\n(2013). Close to stations and past events, the mean values of the non-ergodic\nterms deviate from zero to capture the systematic effects and their epistemic\nuncertainty is small. In areas with sparse data, the epistemic uncertainty of\nthe non-ergodic terms is large, as the systematic effects cannot be determined.\n  The non-ergodic total aleatory standard deviation is approximately $30$ to\n$40\\%$ smaller than the total aleatory standard deviation of BA18. This\nreduction in the aleatory variability has a significant impact on hazard\ncalculations at large return periods. The epistemic uncertainty of the ground\nmotion predictions is small in areas close to stations and past events.\n"}
{"abstract": "  Purpose. We present an approach for forecasting mental health conditions and\nemotions of a given population during the COVID-19 pandemic in Argentina based\non language expressions used in social media. This approach permits\nanticipating high prevalence periods in short- to medium-term time horizons.\nDesign. Mental health conditions and emotions are captured via markers, which\nlink social media contents with lexicons. First, we build descriptive timelines\nfor decision makers to monitor the evolution of markers, and their correlation\nwith crisis events. Second, we model the timelines as time series, and support\ntheir forecasting, which in turn serve to identify high prevalence points for\nthe estimated markers. Findings. Results showed that different time series\nforecasting strategies offer different capabilities. In the best scenario, the\nemergence of high prevalence periods of emotions and mental health disorders\ncan be satisfactorily predicted with a neural network strategy, even when\nlimited data is available in early stages of a crisis (e.g., 7 days).\nOriginality. Although there have been efforts in the literature to predict\nmental states of individuals, the analysis of mental health at the collective\nlevel has received scarce attention. We take a step forward by proposing a\nforecasting approach for analyzing the mental health of a given population (or\ngroup of individuals) at a larger scale. Practical implications. We believe\nthat this work contributes to a better understanding of how psychological\nprocesses related to crisis manifest in social media, being a valuable asset\nfor the design, implementation and monitoring of health prevention and\ncommunication policies.\n"}
{"abstract": "  In this work we introduce a new measure of dispersion of the eigenvalues of a\nHermitian (self-adjoint) compact operator, that we call spectral spread. Then,\nwe obtain several inequalities for unitarily invariant norms involving the\nspectral spread of self-adjoint compact operators, that are related with\nBhatia-Davis's and Corach-Porta-Recht's work on Arithmetic-Geometric mean\ninequalities, Zhan's inequality for the difference of positive compact\noperators, Tao's inequalities for anti-diagonal blocks of positive compact\noperators and Kittaneh's commutator inequalities for positive operators.\n"}
{"abstract": "  Accurate state and uncertainty estimation is imperative for mobile robots and\nself driving vehicles to achieve safe navigation in pedestrian rich\nenvironments. A critical component of state and uncertainty estimation for\nrobot navigation is to perform robustly under out-of-distribution noise.\nTraditional methods of state estimation decouple perception and state\nestimation making it difficult to operate on noisy, high dimensional data.\nHere, we describe an approach that combines the expressiveness of deep neural\nnetworks with principled approaches to uncertainty estimation found in\nrecursive filters. We particularly focus on techniques that provide better\nrobustness to out-of-distribution noise and demonstrate applicability of our\napproach on two scenarios: a simple noisy pendulum state estimation problem and\nreal world pedestrian localization using the nuScenes dataset. We show that our\napproach improves state and uncertainty estimation compared to baselines while\nachieving approximately 3x improvement in computational efficiency.\n"}
{"abstract": "  We show that the Adaptive Greedy algorithm of Golovin and Krause (2011)\nachieves an approximation bound of $(\\ln (Q/\\eta)+1)$ for Stochastic Submodular\nCover: here $Q$ is the \"goal value\" and $\\eta$ is the smallest non-zero\nmarginal increase in utility deliverable by an item. (For integer-valued\nutility functions, we show a bound of $H(Q)$, where $H(Q)$ is the $Q^{th}$\nHarmonic number.) Although this bound was claimed by Golovin and Krause in the\noriginal version of their paper, the proof was later shown to be incorrect by\nNan and Saligrama (2017). The subsequent corrected proof of Golovin and Krause\n(2017) gives a quadratic bound of $(\\ln(Q/\\eta) + 1)^2$. Other previous bounds\nfor the problem are $56(\\ln(Q/\\eta) + 1)$, implied by work of Im et al. (2016)\non a related problem, and $k(\\ln (Q/\\eta)+1)$, due to Deshpande et al. (2016)\nand Hellerstein and Kletenik (2018), where $k$ is the number of states. Our\nbound generalizes the well-known $(\\ln~m + 1)$ approximation bound on the\ngreedy algorithm for the classical Set Cover problem, where $m$ is the size of\nthe ground set.\n"}
{"abstract": "  We prove the equivalence between two explicit expressions for two-point\nWitten-Kontsevich correlators obtained by M. Bertola, B. Dubrovin, D. Yang and\nby P. Zograf.\n"}
{"abstract": "  Recent developments in the field of model-based RL have proven successful in\na range of environments, especially ones where planning is essential. However,\nsuch successes have been limited to deterministic fully-observed environments.\nWe present a new approach that handles stochastic and partially-observable\nenvironments. Our key insight is to use discrete autoencoders to capture the\nmultiple possible effects of an action in a stochastic environment. We use a\nstochastic variant of Monte Carlo tree search to plan over both the agent's\nactions and the discrete latent variables representing the environment's\nresponse. Our approach significantly outperforms an offline version of MuZero\non a stochastic interpretation of chess where the opponent is considered part\nof the environment. We also show that our approach scales to DeepMind Lab, a\nfirst-person 3D environment with large visual observations and partial\nobservability.\n"}
{"abstract": "  A randomized version of the recently developed barycenter method for\nderivative--free optimization has desirable properties of a gradient search. We\ndeveloped a complex version to avoid evaluations at high--gradient points. The\nmethod, which is also applicable to non--convex and to non--smooth functions,\nis parallelizable in a natural way and shown to be robust under noisy\nmeasurements. The goal of this paper is to present an overview of the method,\nwhose properties make it particularly useful in control applications.\n"}
{"abstract": "  This article considers the application of Langevin dynamics to sampling and\ninvestigates how to choose the damping parameter in Langevin dynamics for the\npurpose of maximizing thoroughness of sampling. Also, it considers the\ncomputation of measures of sampling thoroughness.\n"}
{"abstract": "  While structural colors are ubiquitous in nature, saturated reds are\nmysteriously absent. Hence, a longstanding problem is in fabricating\nnanostructured surfaces that exhibit reflectance approaching the theoretical\nlimit. This limit is termed the Schrodinger red and demands sharp spectral\ntransitions from \"stopband\" to a high reflectance \"passband\" with total\nsuppression of higher-order resonances at blue and green wavelengths. Current\napproaches based on metallic or dielectric nanoantennas are insufficient to\nsimultaneously meet these conditions. Here, for the 1st time, we designed and\nfabricated tall Si nanoantenna arrays on quartz substrate to support two\npartially overlapping y polarized quasi bound-state-in-the-continuum (q-BIC)\nmodes in the red wavelengths with sharp spectral edges. These structures\nproduce possibly the most saturated and brightest reds with ~80% reflectance,\nexceeding the red vertex in sRGB and even the cadmium red pigment. We employed\na gradient descent algorithm with structures supporting q BIC as the starting\npoint. Although the current design is polarization dependent, the proposed\nparadigm has enabled us to achieve the elusive structural red and the design\nprinciple could be generalized to Schrodinger's pixels of other colors. The\ndesign is suitable for scale up using other nanofabrication techniques for\nlarger area applications, such as red pixels in displays, decorative coatings,\nand miniaturized spectrometers with high wavelength selectivity.\n"}
{"abstract": "  Graph Neural Networks (GNNs) have achieved promising results for\nsemi-supervised learning tasks on graphs such as node classification. Despite\nthe great success of GNNs, many real-world graphs are often sparsely and\nnoisily labeled, which could significantly degrade the performance of GNNs, as\nthe noisy information could propagate to unlabeled nodes via graph structure.\nThus, it is important to develop a label noise-resistant GNN for\nsemi-supervised node classification. Though extensive studies have been\nconducted to learn neural networks with noisy labels, they mostly focus on\nindependent and identically distributed data and assume a large number of noisy\nlabels are available, which are not directly applicable for GNNs. Thus, we\ninvestigate a novel problem of learning a robust GNN with noisy and limited\nlabels. To alleviate the negative effects of label noise, we propose to link\nthe unlabeled nodes with labeled nodes of high feature similarity to bring more\nclean label information. Furthermore, accurate pseudo labels could be obtained\nby this strategy to provide more supervision and further reduce the effects of\nlabel noise. Our theoretical and empirical analysis verify the effectiveness of\nthese two strategies under mild conditions. Extensive experiments on real-world\ndatasets demonstrate the effectiveness of the proposed method in learning a\nrobust GNN with noisy and limited labels.\n"}
{"abstract": "  OBJECTIVE: Our objective is to evaluate the possibility of using cough audio\nrecordings (spontaneous or simulated) to detect sound patterns in people who\nare diagnosed with COVID-19. The research question that led our work was: what\nis the sensitivity and specificity of a machine learning based COVID-19 cough\nclassifier, using RT-PCR tests as gold standard?\n  SETTING: The audio samples that were collected for this study belong to\nindividuals who were swabbed in the City of Buenos Aires in 20 public and 1\nprivate facilities where RT-PCR studies were carried out on patients suspected\nof COVID, and 14 out-of-hospital isolation units for patients with confirmed\nCOVID mild cases. The audios were collected through the Buenos Aires city\ngovernment WhatsApp chatbot that was specifically designed to address citizen\ninquiries related to the coronavirus pandemic (COVID-19).\n  PARTICIPANTS: The data collected corresponds to 2821 individuals who were\nswabbed in the City of Buenos Aires, between August 11 and December 2, 2020.\nIndividuals were divided into 1409 that tested positive for COVID-19 and 1412\nthat tested negative. From this sample group, 52.6% of the individuals were\nfemale and 47.4% were male. 2.5% were between the age of 0 and 20 , 61.1%\nbetween the age of 21 and 40 , 30.3% between the age of 41 and 60 and 6.1% were\nover 61 years of age.\n  RESULTS: Using the dataset of 2821 individuals our results showed that the\nneural network classifier was able to discriminate between the COVID-19\npositive and the healthy coughs with an accuracy of 86%. This accuracy obtained\nduring the training process was later tested and confirmed with a second\ndataset corresponding to 492 individuals.\n"}
{"abstract": "  The renormalization constant $Z_J$ of the flavor-singlet axial-vector current\nwith a non-anticommuting $\\gamma_5$ in dimensional regularization is determined\nto order $\\alpha_s^3$ in QCD with massless quarks. The result is obtained by\ncomputing the matrix elements of the operators appearing in the axial-anomaly\nequation $\\big[\\partial_{\\mu} J^{\\mu}_{5} \\big]_{R} = \\frac{\\alpha_s}{4 \\pi}\\,\nn_f\\, {\\displaystyle \\mathrm{T}_{F}} \\, \\big[F \\tilde{F} \\big]_{R}$ between the\nvacuum and a state of two (off-shell) gluons to 4-loop order. Furthermore,\nthrough this computation, the conjectured equality between the\n$\\overline{\\mathrm{MS}}$ renormalization constant $Z_{F\\tilde{F}}$ associated\nwith the operator $\\big[F \\tilde{F} \\big]_{R}$ and that of $\\alpha_s$ is\nverified explicitly to hold true at 4-loop order. This equality automatically\nensures a relation between the respective anomalous dimensions,\n$\\gamma_{{\\scriptscriptstyle J}} = \\frac{\\alpha_s}{4 \\pi}\\, n_f\\,\n{\\displaystyle \\mathrm{T}_{F}} \\, \\gamma_{{\\scriptscriptstyle FJ}} $, at order\n$\\alpha_s^4$ given the validity of the axial-anomaly equation which was used to\ndetermine the non-$\\overline{\\mathrm{MS}}$ piece of $Z_J$ for the particular\n$\\gamma_5$ prescription in use.\n"}
{"abstract": "  Microstructure characterization is of great value to understanding nuclear\ngraphite's properties and irradiation behavior. However, graphite is soft and\ncould be easily damaged during sample preparation. A three-step polishing\nmethod involving mechanical polishing, ion milling and rapid oxidation is\nproposed for graphite. Ion milling is adopted to remove the broken graphite\npieces produced by mechanical polishing. Rapid oxidation is then adopted to\nremove irradiation-induced damage layer during ion milling. The Raman spectra\nshow very low G peak width and ID/IG ratio after rapid oxidation, indicating a\nsurface completely free from artificial defects. The micro-cracks which were\nconventionally observed via a transmission electron microscope can be observed\non rapid-oxidized surface in a scanning electron microscope. By digital image\nprocessing, the micro-cracks along with the gas-escape pores in nuclear\ngraphite IG-110 are statistically analyzed. Porosity's distributions on crack\n(pore) size (spanning from 10 nm to 100 um) are given, which could help to\nunderstand and simulate graphite's performances in reactors.\n"}
{"abstract": "  Speakers of non-English languages often adopt loanwords from English to\nexpress new or unusual concepts. While these loanwords may be borrowed\nunchanged, speakers may also integrate the words to fit the constraints of\ntheir native language, e.g. creating Spanish \"tuitear\" from English \"tweet.\"\nLinguists have often considered the process of loanword integration to be more\ndependent on language-internal constraints, but sociolinguistic constraints\nsuch as speaker background remain only qualitatively understood. We investigate\nthe role of social context and speaker background in Spanish speakers' use of\nintegrated loanwords on social media. We find first that newspaper authors use\nthe integrated forms of loanwords and native words more often than social media\nauthors, showing that integration is associated with formal domains. In social\nmedia, we find that speaker background and expectations of formality explain\nloanword and native word integration, such that authors who use more Spanish\nand who write to a wider audience tend to use integrated verb forms more often.\nThis study shows that loanword integration reflects not only language-internal\nconstraints but also social expectations that vary by conversation and speaker.\n"}
{"abstract": "  Small-sized systems exhibit a finite number of routes to chaos. However, in\nextended systems, not all routes to complex spatiotemporal behavior have been\nfully explored. Starting from the sine-Gordon model of parametrically driven\nchain of damped nonlinear oscillators, we investigate a route to spatiotemporal\nchaos emerging from standing waves. The route from the stationary to the\nchaotic state proceeds through quasiperiodic dynamics. The standing wave\nundergoes the onset of oscillatory instability, which subsequently exhibits a\ndifferent critical frequency, from which the complexity originates. A suitable\namplitude equation, valid close to the parametric resonance, makes it possible\nto produce universe results. The respective phase-space structure and\nbifurcation diagrams are produced in a numerical form. We characterize the\nrelevant dynamical regimes by means of the largest Lyapunov exponent, the power\nspectrum, and the evolution of the total intensity of the wave field.\n"}
{"abstract": "  Humans are highly efficient learners, with the ability to grasp the meaning\nof a new concept from just a few examples. Unlike popular computer vision\nsystems, humans can flexibly leverage the compositional structure of the visual\nworld, understanding new concepts as combinations of existing concepts. In the\ncurrent paper, we study how people learn different types of visual\ncompositions, using abstract visual forms with rich relational structure. We\nfind that people can make meaningful compositional generalizations from just a\nfew examples in a variety of scenarios, and we develop a Bayesian program\ninduction model that provides a close fit to the behavioral data. Unlike past\nwork examining special cases of compositionality, our work shows how a single\ncomputational approach can account for many distinct types of compositional\ngeneralization.\n"}
{"abstract": "  The Kurth solution is a particular non-isotropic steady state solution to the\ngravitational Vlasov-Poisson system. It has the property that by means of a\nsuitable time-dependent transformation it can be turned into a family of\ntime-dependent solutions. Therefore, for a general steady state $Q(x,\nv)=\\tilde{Q}(e_Q, \\beta)$, depending upon the particle energy $e_Q$ and\n$\\beta=\\ell^2=|x\\wedge v|^2$, the question arises if solutions $f$ could be\ngenerated that are of the form \\[ f(t)=\\tilde{Q}\\Big(e_Q(R(t), P(t), B(t)),\nB(t)\\Big) \\] for suitable functions $R$, $P$ and $B$, all depending on $(t, r,\np_r, \\beta)$ for $r=|x|$ and $p_r=\\frac{x\\cdot v}{|x|}$. We are going to show\nthat, under some mild assumptions, basically if $R$ and $P$ are independent of\n$\\beta$, and if $B=\\beta$ is constant, then $Q$ already has to be the Kurth\nsolution.\n  This paper is dedicated to the memory of Professor Robert Glassey.\n"}
{"abstract": "  Most software maintenance and evolution tasks require developers to\nunderstand the source code of their software systems. Software developers\nusually inspect class comments to gain knowledge about program behavior,\nregardless of the programming language they are using. Unfortunately, (i)\ndifferent programming languages present language-specific code commenting\nnotations/guidelines; and (ii) the source code of software projects often lacks\ncomments that adequately describe the class behavior, which complicates program\ncomprehension and evolution activities.\n  To handle these challenges, this paper investigates the different\nlanguage-specific class commenting practices of three programming languages:\nPython, Java, and Smalltalk. In particular, we systematically analyze the\nsimilarities and differences of the information types found in class comments\nof projects developed in these languages.\n  We propose an approach that leverages two techniques, namely Natural Language\nProcessing and Text Analysis, to automatically identify various types of\ninformation from class comments i.e., the specific types of semantic\ninformation found in class comments. To the best of our knowledge, no previous\nwork has provided a comprehensive taxonomy of class comment types for these\nthree programming languages with the help of a common automated approach. Our\nresults confirm that our approach can classify frequent class comment\ninformation types with high accuracy for Python, Java, and Smalltalk\nprogramming languages. We believe this work can help to monitor and assess the\nquality and evolution of code comments in different program languages, and thus\nsupport maintenance and evolution tasks.\n"}
{"abstract": "  We report the first isotopic composition measurements of trinitite, nuclear\ndetonation debris from the Trinity test, using the novel forensics technique of\ndecay energy spectroscopy (DES). DES measures the unique total decay energy (Q\nvalue) of each alpha-decaying isotope in a small radioactive sample embedded in\na microcalorimeter detector. We find that DES can measure the major\nalpha-decaying isotopes in small particles of trinitite with no dissolution or\nchemical processing. These first measurements demonstrate the potential of DES\nto provide a radiometric isotopic characterization method with sensitivity and\nprecision to complement traditional forensics techniques.\n"}
{"abstract": "  We advocate the development of a discipline of interacting with and\nextracting information from models, both mathematical (e.g. game-theoretic\nones) and computational (e.g. agent-based models). We outline some directions\nfor the development of a such a discipline:\n  - the development of logical frameworks for the systematic formal\nspecification of stylized facts and social mechanisms in (mathematical and\ncomputational) social science. Such frameworks would bring to attention new\nissues, such as phase transitions, i.e. dramatical changes in the validity of\nthe stylized facts beyond some critical values in parameter space. We argue\nthat such statements are useful for those logical frameworks describing\nproperties of ABM.\n  - the adaptation of tools from the theory of reactive systems (such as\nbisimulation) to obtain practically relevant notions of two systems \"having the\nsame behavior\".\n  - the systematic development of an adversarial theory of model perturbations,\nthat investigates the robustness of conclusions derived from models of social\nbehavior to variations in several features of the social dynamics. These may\ninclude: activation order, the underlying social network, individual agent\nbehavior.\n"}
{"abstract": "  Transformations for enhancing sparsity in the approximation of color images\nby 2D atomic decomposition are discussed. The sparsity is firstly considered\nwith respect to the most significant coefficients in the wavelet decomposition\nof the color image. The discrete cosine transform is singled out as an\neffective transformation for this purpose. The enhanced feature is further\nexploited by approximating the transformed arrays using an effective greedy\nstrategy with a separable highly redundant dictionary. The relevance of the\nachieved sparsity is illustrated by a simple encoding procedure. On a set of\ntypical test images the compression at high quality recovery is shown to\nsignificantly improve upon JPEG and WebP formats. The results are competitive\nwith those produced by the JPEG2000 standard.\n"}
{"abstract": "  As job markets worldwide have become more competitive and applicant selection\ncriteria have become more opaque, and different (and sometimes contradictory)\ninformation and advice is available for job seekers wishing to progress in\ntheir careers, it has never been more difficult to determine which factors in a\nr\\'esum\\'e most effectively help career progression. In this work we present a\nnovel, large scale dataset of over half a million r\\'esum\\'es with preliminary\nanalysis to begin to answer empirically which factors help or hurt people\nwishing to transition to more senior roles as they progress in their career. We\nfind that previous experience forms the most important factor, outweighing\nother aspects of human capital, and find which language factors in a r\\'esum\\'e\nhave significant effects. This lays the groundwork for future inquiry in career\ntrajectories using large scale data analysis and natural language processing\ntechniques.\n"}
{"abstract": "  In this note, we consider the homogenization of the compressible\nNavier-Stokes equations in a periodically perforated domain in $\\mathbb{R}^3$.\nAssuming that the particle size scales like $\\varepsilon^3$, where\n$\\varepsilon>0$ is their mutual distance, and that the Mach number decreases\nfast enough, we show that in the limit $\\varepsilon\\to 0$, the velocity and\ndensity converge to a solution of the incompressible Navier-Stokes equations\nwith Brinkman term. We strongly follow the methods of H\\\"ofer, Kowalczik and\nSchwarzacher [arXiv:2007.09031], where they proved convergence to Darcy's law\nfor the particle size scaling like $\\varepsilon^\\alpha$ with $\\alpha\\in (1,3)$.\n"}
{"abstract": "  We present the results of a quantitative study of the phase behavior of a\nmodel polymer chain with side spheres using two independent computer simulation\ntechniques. We find that the mere addition of side spheres results in key\nmodifications of standard polymer behavior. One obtains a novel marginally\ncompact phase at low temperatures, the structures in this phase are reduced in\ndimensionality and are ordered, they include strands assembled into sheets and\na variety of helices, and at least one of the transitions on lowering the\ntemperature to access these ordered states is found to be first order. Our\nmodel serves to partially bridge conventional polymer phases with biomolecular\nphases.\n"}
{"abstract": "  We study the light rings and shadows of an uniparametric family of\nspherically symmetric geometries interpolating between the Schwarzschild\nsolution, a regular black hole, and a traversable wormhole, and dubbed as black\nbounces, all of them sharing the same critical impact parameter. We consider\nthe ray-tracing method in order to study the impact parameter regions\ncorresponding to the direct, lensed, and photon ring emission, finding a\nbroadening of all these regions for black bounce solutions as compared to the\nSchwarzschild one. Using this, we determine the optical appearance of black\nbounces when illuminated by three standard toy models of optically and\ngeometrically thin accretion disks viewed in face-on orientation.\n"}
{"abstract": "  This paper describes computer models of three interventions used for treating\nrefractory pulmonary hypertension (RPH). These procedures create either an\natrial septal defect, a ventricular septal defect, or, in the case of a Potts\nshunt, a patent ductus arteriosus. The aim in all three cases is to generate a\nright-to-left shunt, allowing for either pressure or volume unloading of the\nright side of the heart in the setting of right ventricular failure, while\nmaintaining cardiac output. These shunts are created, however, at the expense\nof introducing de-oxygenated blood into the systemic circulation, thereby\nlowering the systemic arterial oxygen saturation. The models developed in this\npaper are based on compartmental descriptions of human hemodynamics and oxygen\ntransport. An important parameter included in our models is the cross-sectional\narea of the surgically created defect. Numerical simulations are performed to\ncompare different interventions and various shunt sizes and to assess their\nimpact on hemodynamic variables and oxygen saturations. We also create a model\nfor exercise and use it to study exercise tolerance in simulated\npre-intervention and post-intervention RPH patients.\n"}
{"abstract": "  In this work, we consider the asynchronous distributed optimization problem\nin which each node has its own convex cost function and can communicate\ndirectly only with its neighbors, as determined by a directed communication\ntopology (directed graph or digraph). First, we reformulate the optimization\nproblem so that Alternating Direction Method of Multipliers (ADMM) can be\nutilized. Then, we propose an algorithm, herein called Asynchronous Approximate\nDistributed Alternating Direction Method of Multipliers (AsyAD-ADMM), using\nfinite-time asynchronous approximate ratio consensus, to solve the multi-node\nconvex optimization problem, in which every node performs iterative\ncomputations and exchanges information with its neighbors asynchronously. More\nspecifically, at every iteration of AsyAD-ADMM, each node solves a local convex\noptimization problem for one of the primal variables and utilizes a finite-time\nasynchronous approximate consensus protocol to obtain the value of the other\nvariable which is close to the optimal value, since the cost function for the\nsecond primal variable is not decomposable. If the individual cost functions\nare convex but not necessarily differentiable, the proposed algorithm converges\nat a rate of $\\mathcal{O}(1/k)$, where $k$ is the iteration counter. The\nefficacy of AsyAD-ADMM is exemplified via a proof-of-concept distributed\nleast-square optimization problem with different performance-influencing\nfactors investigated.\n"}
{"abstract": "  We are motivated by the goal of generalist robots that can complete a wide\nrange of tasks across many environments. Critical to this is the robot's\nability to acquire some metric of task success or reward, which is necessary\nfor reinforcement learning, planning, or knowing when to ask for help. For a\ngeneral-purpose robot operating in the real world, this reward function must\nalso be able to generalize broadly across environments, tasks, and objects,\nwhile depending only on on-board sensor observations (e.g. RGB images). While\ndeep learning on large and diverse datasets has shown promise as a path towards\nsuch generalization in computer vision and natural language, collecting high\nquality datasets of robotic interaction at scale remains an open challenge. In\ncontrast, \"in-the-wild\" videos of humans (e.g. YouTube) contain an extensive\ncollection of people doing interesting tasks across a diverse range of\nsettings. In this work, we propose a simple approach, Domain-agnostic Video\nDiscriminator (DVD), that learns multitask reward functions by training a\ndiscriminator to classify whether two videos are performing the same task, and\ncan generalize by virtue of learning from a small amount of robot data with a\nbroad dataset of human videos. We find that by leveraging diverse human\ndatasets, this reward function (a) can generalize zero shot to unseen\nenvironments, (b) generalize zero shot to unseen tasks, and (c) can be combined\nwith visual model predictive control to solve robotic manipulation tasks on a\nreal WidowX200 robot in an unseen environment from a single human demo.\n"}
{"abstract": "  Quantum self-testing is the task of certifying quantum states and\nmeasurements using the output statistics solely, with minimal assumptions about\nthe underlying quantum system. It is based on the observation that some\nextremal points in the set of quantum correlations can only be achieved, up to\nisometries, with specific states and measurements. Here, we present a new\napproach for quantum self-testing in Bell non-locality scenarios, motivated by\nthe following observation: the quantum maximum of a given Bell inequality is,\nin general, difficult to characterize. However, it is strictly contained in an\neasy-to-characterize set: the \\emph{theta body} of a vertex-weighted induced\nsubgraph $(G,w)$ of the graph in which vertices represent the events and edges\njoin mutually exclusive events. This implies that, for the cases where the\nquantum maximum and the maximum within the theta body (known as the Lov\\'asz\ntheta number) of $(G,w)$ coincide, self-testing can be demonstrated by just\nproving self-testability with the theta body of $G$. This graph-theoretic\nframework allows us to (i) recover the self-testability of several quantum\ncorrelations that are known to permit self-testing (like those violating the\nClauser-Horne-Shimony-Holt (CHSH) and three-party Mermin Bell inequalities for\nprojective measurements of arbitrary rank, and chained Bell inequalities for\nrank-one projective measurements), (ii) prove the self-testability of quantum\ncorrelations that were not known using existing self-testing techniques (e.g.,\nthose violating the Abner Shimony Bell inequality for rank-one projective\nmeasurements). Additionally, the analysis of the chained Bell inequalities\ngives us a closed-form expression of the Lov\\'asz theta number for a family of\nwell-studied graphs known as the M\\\"obius ladders, which might be of\nindependent interest in the community of discrete mathematics.\n"}
{"abstract": "  We give an algorithm to determine whether a kernel sheaf over a smooth\nprojective curve over an algebraically closed field is semistable. The\nalgorithm uses symmetric powers to make destabilizing subbundles visible as\nglobal sections.\n"}
{"abstract": "  Neural natural language generation (NLG) and understanding (NLU) models are\ndata-hungry and require massive amounts of annotated data to be competitive.\nRecent frameworks address this bottleneck with generative models that\nsynthesize weak labels at scale, where a small amount of training labels are\nexpert-curated and the rest of the data is automatically annotated. We follow\nthat approach, by automatically constructing a large-scale weakly-labeled data\nwith a fine-tuned GPT-2, and employ a semi-supervised framework to jointly\ntrain the NLG and NLU models. The proposed framework adapts the parameter\nupdates to the models according to the estimated label-quality. On both the E2E\nand Weather benchmarks, we show that this weakly supervised training paradigm\nis an effective approach under low resource scenarios and outperforming\nbenchmark systems on both datasets when 100% of training data is used.\n"}
{"abstract": "  We present Integral Field Spectroscopic (IFS) observations of the nearby\n($z\\sim0.03$) dual Active Galactic Nuclei (AGN) Mrk 739, whose projected\nnuclear separation is $\\sim$3.4~kpc, obtained with the Multi Unit Spectroscopic\nExplorer (MUSE) at the Very Large Telescope (VLT). We find that the galaxy has\nan extended AGN-ionized emission-line region extending up to $\\sim 20$ kpc away\nfrom the nuclei, while star-forming regions are more centrally concentrated\nwithin 2 - 3 kpc. We model the kinematics of the ionized gas surrounding the\nEast nucleus using a circular disk profile, resulting in a peak velocity of\n$237^{+26}_{-28}$ km s$^{-1}$ at a distance of $\\sim 1.2$ kpc. The enclosed\ndynamical mass within 1.2 kpc is $\\log M(M_{\\odot})=10.20\\pm0.06$, $\\sim$1,000\ntimes larger than the estimated supermassive black hole (SMBH) mass of Mrk\n739E. The morphology and dynamics of the system are consistent with an early\nstage of the collision, where the foreground galaxy (Mrk 739W) is a young\nstar-forming galaxy in an ongoing first passage with its background companion\n(Mrk 739E). Since the SMBH in Mrk 739W does not show evidence of being rapidly\naccreting, we claim that the northern spiral arms of Mrk 739W are ionized by\nthe nuclear activity of Mrk 739E.\n"}
{"abstract": "  The spin-memory effect in the GaAs / InGaAs heterostructures with\n$\\delta$<Mn> layer in GaAs barrier have been investigated. The effect consists\nin spin polarization of Mn atoms due to interaction with photogenerated\nspin-polarized holes. The investigation of the effect was carried out by\nanalyzing the polarization of the probe photoluminescence pulse in the\npump-probe technique. It was shown that the circular polarization degree of\nprobe pulse generated photoluminescence is strongly affected by the interaction\nof hole spins with spins of Mn atoms polarized by the pump pulse. The latter\nleads to decrease of circular polarization degree as compared with single pulse\nexcitation ($\\delta P$ effect). The amplitude of $\\delta P$-effect is most\nstrongly affected by the concentration of resident electrons in the quantum\nwell which is believed to be due the specific compliance with selection rules\nfor optical transition with the participation of unpolarized resident\nelectrons. The rest of the sample's parameters including the spatial separation\nbetween $\\delta$<Mn> layer and InGaAs quantum well ($d_s$) have a minor effect\non the $\\delta P$ value which leads to a paradoxical situation of decreasing\n$\\delta P$-effect with the decrease of $d_s$. The proposed experimental\ntechnique consisting in creating the significant concentration of resident\nelectrons in the QW may serve as a reliable photoluminescence method\ndetermining the strength of this effect as well as the Mn spin relaxation time\nin a particular nanostructure.\n"}
{"abstract": "  We discuss the spectral energy distributions and physical properties of six\ngalaxies whose photometric redshifts suggest they lie beyond a redshift\n$z\\simeq$9. Each was selected on account of a prominent excess seen in the\nSpitzer/IRAC 4.5$\\mu$m band which, for a redshift above $z=9.0$, likely\nindicates the presence of a rest-frame Balmer break and a stellar component\nthat formed earlier than a redshift $z\\simeq10$. In addition to constraining\nthe earlier star formation activity on the basis of fits using stellar\npopulation models with BAGPIPES, we have undertaken the necessary, but\nchallenging, follow-up spectroscopy for each candidate using various\ncombinations of Keck/MOSFIRE, VLT/X-shooter, Gemini/FLAMINGOS2 and ALMA. Based\non either Lyman-$\\alpha$ or [OIII] 88 $\\mu$m emission, we determine a\nconvincing redshift of $z$=8.78 for GN-z-10-3 and a likely redshift of $z$=9.28\nfor the lensed galaxy MACS0416-JD. For GN-z9-1, we conclude the case remains\npromising for a source beyond $z\\simeq$9. Together with earlier spectroscopic\ndata for MACS1149-JD1, our analysis of this enlarged sample provides further\nsupport for a cosmic star formation history extending beyond redshifts\n$z\\simeq$10. We use our best-fit stellar population models to reconstruct the\npast rest-frame UV luminosities of our sources and discuss the implications for\ntracing earlier progenitors of such systems with the James Webb Space\nTelescope.\n"}
{"abstract": "  The optimal time for the controllability of linear hyperbolic systems in one\ndimensional space with one-side controls has been obtained recently for\ntime-independent coefficients in our previous works. In this paper, we consider\nlinear hyperbolic systems with time-varying zero-order terms. We show the\npossibility that the optimal time for the null-controllability becomes\nsignificantly larger than the one of the time-invariant setting even when the\nzero-order term is indefinitely differentiable. When the analyticity with\nrespect to time is imposed for the zero-order term, we also establish that the\noptimal time is the same as in the time-independent setting.\n"}
{"abstract": "  Spatiotemporal imaging is common in medical imaging, with applications in\ne.g. cardiac diagnostics, surgical guidance and radiotherapy monitoring. In\nthis paper, we present an unsupervised model that identifies the underlying\ndynamics of the system, only based on the sequential images. The model maps the\ninput to a low-dimensional latent space wherein a linear relationship holds\nbetween a hidden state process and the observed latent process. Knowledge of\nthe system dynamics enables denoising, imputation of missing values and\nextrapolation of future image frames. We use a Variational Auto-Encoder (VAE)\nfor the dimensionality reduction and a Linear Gaussian State Space Model\n(LGSSM) for the latent dynamics. The model, known as a Kalman Variational\nAuto-Encoder, is end-to-end trainable and the weights, both in the VAE and\nLGSSM, are simultaneously updated by maximizing the evidence lower bound of the\nmarginal log likelihood. Our experiment, on cardiac ultrasound time series,\nshows that the dynamical model provide better reconstructions than a similar\nmodel without dynamics. And also possibility to impute and extrapolate for\nmissing samples.\n"}
{"abstract": "  Calculation of dynamical diffraction patterns for X-ray topography and\nsimilar x-ray scattering-imaging techniques require the numerical integration\nof the Takagi-Taupin equations. This is usually done with a simple second order\nfinite difference scheme on a sheared computational grid with two of the axes\naligned with the wave vectors of the incident and scattered beams respectively.\nHere we present a finite difference scheme that carries out this integration on\nan arbitrary orthogonal grid by implicitly utilizing Fourier interpolation. The\nscheme achieves the expected second order convergence and a similar error to\nthe traditional approach on similarly dense grids but is more computationally\nexpensive due to the use of FFT operations.\n"}
{"abstract": "  Real-time video analytics on the edge is challenging as the computationally\nconstrained resources typically cannot analyse video streams at full fidelity\nand frame rate, which results in loss of accuracy. This paper proposes a\nTransprecise Object Detector (TOD) which maximises the real-time object\ndetection accuracy on an edge device by selecting an appropriate Deep Neural\nNetwork (DNN) on the fly with negligible computational overhead. TOD makes two\nkey contributions over the state of the art: (1) TOD leverages characteristics\nof the video stream such as object size and speed of movement to identify\nnetworks with high prediction accuracy for the current frames; (2) it selects\nthe best-performing network based on projected accuracy and computational\ndemand using an effective and low-overhead decision mechanism. Experimental\nevaluation on a Jetson Nano demonstrates that TOD improves the average object\ndetection precision by 34.7 % over the YOLOv4-tiny-288 model on average over\nthe MOT17Det dataset. In the MOT17-05 test dataset, TOD utilises only 45.1 % of\nGPU resource and 62.7 % of the GPU board power without losing accuracy,\ncompared to YOLOv4-416 model. We expect that TOD will maximise the application\nof edge devices to real-time object detection, since TOD maximises real-time\nobject detection accuracy given edge devices according to dynamic input\nfeatures without increasing inference latency in practice.\n"}
{"abstract": "  Quality-Diversity algorithms search for large collections of diverse and\nhigh-performing solutions, rather than just for a single solution like typical\noptimisation methods. They are specially adapted for multi-modal problems that\ncan be solved in many different ways, such as complex reinforcement learning or\nrobotics tasks. However, these approaches are highly dependent on the choice of\nfeature descriptors (FDs) quantifying the similarity in behaviour of the\nsolutions. While FDs usually needs to be hand-designed, recent studies have\nproposed ways to define them automatically by using feature extraction\ntechniques, such as PCA or Auto-Encoders, to learn a representation of the\nproblem from previously explored solutions. Here, we extend these approaches to\nmore complex problems which cannot be efficiently explored by relying only on a\nsingle representation but require instead a set of diverse and complementary\nrepresentations. We describe MC-AURORA, a Quality-Diversity approach that\noptimises simultaneously several collections of solutions, each with a\ndifferent set of FDs, which are, in turn, defined automatically by an ensemble\nof modular auto-encoders. We show that this approach produces solutions that\nare more diverse than those produced by single-representation approaches.\n"}
{"abstract": "  Estimating 3D hand and object pose from a single image is an extremely\nchallenging problem: hands and objects are often self-occluded during\ninteractions, and the 3D annotations are scarce as even humans cannot directly\nlabel the ground-truths from a single image perfectly. To tackle these\nchallenges, we propose a unified framework for estimating the 3D hand and\nobject poses with semi-supervised learning. We build a joint learning framework\nwhere we perform explicit contextual reasoning between hand and object\nrepresentations by a Transformer. Going beyond limited 3D annotations in a\nsingle image, we leverage the spatial-temporal consistency in large-scale\nhand-object videos as a constraint for generating pseudo labels in\nsemi-supervised learning. Our method not only improves hand pose estimation in\nchallenging real-world dataset, but also substantially improve the object pose\nwhich has fewer ground-truths per instance. By training with large-scale\ndiverse videos, our model also generalizes better across multiple out-of-domain\ndatasets. Project page and code: https://stevenlsw.github.io/Semi-Hand-Object\n"}
{"abstract": "  Bayesian experimental design (BED) is to answer the question that how to\nchoose designs that maximize the information gathering. For implicit models,\nwhere the likelihood is intractable but sampling is possible, conventional BED\nmethods have difficulties in efficiently estimating the posterior distribution\nand maximizing the mutual information (MI) between data and parameters. Recent\nwork proposed the use of gradient ascent to maximize a lower bound on MI to\ndeal with these issues. However, the approach requires a sampling path to\ncompute the pathwise gradient of the MI lower bound with respect to the design\nvariables, and such a pathwise gradient is usually inaccessible for implicit\nmodels. In this paper, we propose a novel approach that leverages recent\nadvances in stochastic approximate gradient ascent incorporated with a smoothed\nvariational MI estimator for efficient and robust BED. Without the necessity of\npathwise gradients, our approach allows the design process to be achieved\nthrough a unified procedure with an approximate gradient for implicit models.\nSeveral experiments show that our approach outperforms baseline methods, and\nsignificantly improves the scalability of BED in high-dimensional problems.\n"}
{"abstract": "  This paper proposes a novel training model based on shape and appearance\nfeatures for object segmentation in images and videos. Whereas most such models\nrely on two-dimensional appearance templates or a finite set of descriptors,\nour appearance-based feature is a one-dimensional function, which is\nefficiently coupled with the object's shape by integrating intensities along\nthe object's iso-contours. Joint PCA training on these shape and appearance\nfeatures further exploits shape-appearance correlations and the resulting\ntraining model is incorporated in an active-contour-type energy functional for\nrecognition-segmentation tasks. Experiments on synthetic and infrared images\ndemonstrate how this shape and appearance training model improves accuracy\ncompared to methods based on the Chan-Vese energy.\n"}
{"abstract": "  Context. The diffusion of volatile species on amorphous solid water ice\naffects the chemistry on dust grains in the interstellar medium as well as the\ntrapping of gases enriching planetary atmospheres or present in cometary\nmaterial.\n  Aims. The aim of the work is to provide diffusion coefficients of CH$_4$ on\namorphous solid water (ASW), and to understand how they are affected by the ASW\nstructure.\n  Methods. Ice mixtures of H$_2$O and CH$_4$ were grown in different conditions\nand the sublimation of CH$_4$ was monitored via infrared spectroscopy or via\nthe mass loss of a cryogenic quartz crystal microbalance. Diffusion\ncoefficients were obtained from the experimental data assuming the systems obey\nFick's law of diffusion. Monte Carlo simulations modeled the different\namorphous solid water ice structures investigated and were used to reproduce\nand interpret the experimental results.\n  Results. Diffusion coefficients of methane on amorphous solid water have been\nmeasured to be between 10$^{-12}$ and 10$^{-13}$ cm$^2$ s$^{-1}$ for\ntemperatures ranging between 42 K and 60 K. We showed that diffusion can differ\nby one order of magnitude depending on the morphology of amorphous solid water.\nThe porosity within water ice, and the network created by pore coalescence,\nenhance the diffusion of species within the pores.The diffusion rates derived\nexperimentally cannot be used in our Monte Carlo simulations to reproduce the\nmeasurements.\n  Conclusions. We conclude that Fick's law can be used to describe diffusion at\nthe macroscopic scale, while Monte Carlo simulations describe the microscopic\nscale where trapping of species in the ices (and their movement) is considered.\n"}
{"abstract": "  Let $p$ be prime, and $n,m \\in \\mathbb{N}$. When $K/F$ is a cyclic extension\nof degree $p^n$, we determine the $\\mathbb{F}_p[\\text{Gal}(K/F)]$-module\nstructure of $K^\\times/K^{\\times p^m}$. With at most one exception, each\nindecomposable summand is cyclic and free over some quotient group of\n$\\text{Gal}(K/F)$. For fixed values of $m$ and $n$, there are only finitely\nmany possible isomorphism classes for the non-free indecomposable summand.\n  This result seems to be a significant breakthrough in the study of Galois\nmodules. For one, these Galois modules act as parameterizing spaces for\nsolutions to certain inverse Galois problems, and therefore this module\ncomputation should provide key insight into absolute Galois groups. More\nimmediately, however, these results show that Galois cohomology is a context in\nwhich seemingly impossible module decompositions can practically be achieved:\nwhen $m,n>1$ the modular representation theory allows for an infinite number of\nindecomposable summands, with no known classification of indecomposable types.\nThis paper marks the first time a Galois module decomposition has been computed\nunder these circumstances over an infinite family of modules.\n"}
{"abstract": "  In this paper, we consider a second-order scalar auxiliary variable (SAV)\nFourier spectral method to solve the nonlinear fractional generalized wave\nequation. Unconditional energy conservation or dissipation properties of the\nfully discrete scheme are first established. Next, we utilize the\ntemporal-spatial error splitting argument to obtain unconditional optimal error\nestimate of the fully discrete scheme, which overcomes time-step restrictions\ncaused by strongly nonlinear system, or the restrictions that the nonlinear\nterm needs to satisfy the assumption of global Lipschitz condition in all\nprevious works for fractional undamped or damped wave equations. Finally, some\nnumerical experiments are presented to confirm our theoretical analysis.\n"}
{"abstract": "  Automatic draping of carbon-fiber prepreg plies for the aerospace industry is\na promising technique for lowering the manufacturing costs and to this end, a\nthorough in-process quality control is crucial. In this paper, out-of-plane\ndefects in the layup are investigated. After draping, air pockets are\noccasionally encountered. The question is, if such apparent defects can be\nmitigated sufficiently during vacuum debulking. The 3D topology is measured by\nmeans of a structured-light 3D scanner and air pockets are segmented. An\napproximate mass-spring ply model is used to study the behavior of the air\npockets during application of vacuum pressure. The model is computationally\nfast and will indicate online whether the air pocket will be removed or manual\nintervention is required. Upon comparing the model predictions with\nexperimental data, it is shown that the system is capable of correctly\npredicting 13 out of 14 air pockets in a test layup.\n"}
{"abstract": "  Mixing transformations in quantum field theory are non-trivial, since they\nare intimately related to the unitary inequivalence between Fock spaces for\nfields with definite mass and fields with definite flavor. Considering the\nsuperposition of two neutral scalar (spin-0) bosonic fields, we investigate\nsome features of the emerging condensate structure of the flavor vacuum. In\nparticular, we quantify the flavor vacuum entanglement in terms of the von\nNeumann entanglement entropy of the reduced state. Furthermore, in a suitable\nlimit, we show that the flavor vacuum has a structure akin to the thermal\nvacuum of Thermo Field Dynamics, with a temperature dependent on both the\nmixing angle and the particle mass difference.\n"}
{"abstract": "  Let $S(t) =\n\\frac{1}{\\pi}\\operatorname{Im}\\log\\zeta\\left(\\frac{1}{2}+it\\right)$. Using\nSoundararajan's resonance method we prove an unconditional lower bound on the\nsize of the tails of the distribution of $S(t)$. In particular we reproduce the\nbest unconditional $\\Omega$ result for $S(t)$ which is due to Tsang, \\[ S(t) =\n\\Omega_\\pm\\left(\\left(\\frac{\\log t}{\\log \\log t}\\right)^{1/3}\\right), \\] and\nget a bound on how often large values of $S(t)$ occur. We also give a\nprobabilistic argument for why this $\\Omega$ result may be the best possible\ngiven our current knowledge of the zeros of the zeta function.\n"}
{"abstract": "  We present a search for radio afterglows from long gamma-ray bursts using the\nAustralian Square Kilometre Array Pathfinder (ASKAP). Our search used the Rapid\nASKAP Continuum Survey, covering the entire celestial sphere south of\ndeclination $+41^\\circ$, and three epochs of the Variables and Slow Transients\nPilot Survey (Phase 1), covering $\\sim 5,000$ square degrees per epoch. The\nobservations we used from these surveys spanned a nine-month period from 2019\nApril 21 to 2020 January 11. We crossmatched radio sources found in these\nsurveys with 779 well-localised (to $\\leq 15''$) long gamma-ray bursts\noccurring after 2004 and determined whether the associations were more likely\nafterglow- or host-related through the analysis of optical images. In our\nsearch, we detected one radio afterglow candidate associated with GRB 171205A,\na local low-luminosity gamma-ray burst with a supernova counterpart SN 2017iuk,\nin an ASKAP observation 511 days post-burst. We confirmed this detection with\nfurther observations of the radio afterglow using the Australia Telescope\nCompact Array at 859 days and 884 days post-burst. Combining this data with\narchival data from early-time radio observations, we showed the evolution of\nthe radio spectral energy distribution alone could reveal clear signatures of a\nwind-like circumburst medium for the burst. Finally, we derived semi-analytical\nestimates for the microphysical shock parameters of the burst: electron\npower-law index $p = 2.84$, normalised wind-density parameter $A_* = 3$,\nfractional energy in electrons $\\epsilon_{e} = 0.3$, and fractional energy in\nmagnetic fields $\\epsilon_{B} = 0.0002$.\n"}
{"abstract": "  The scarcity of labeled data often impedes the application of deep learning\nto the segmentation of medical images. Semi-supervised learning seeks to\novercome this limitation by exploiting unlabeled examples in the learning\nprocess. In this paper, we present a novel semi-supervised segmentation method\nthat leverages mutual information (MI) on categorical distributions to achieve\nboth global representation invariance and local smoothness. In this method, we\nmaximize the MI for intermediate feature embeddings that are taken from both\nthe encoder and decoder of a segmentation network. We first propose a global MI\nloss constraining the encoder to learn an image representation that is\ninvariant to geometric transformations. Instead of resorting to\ncomputationally-expensive techniques for estimating the MI on continuous\nfeature embeddings, we use projection heads to map them to a discrete cluster\nassignment where MI can be computed efficiently. Our method also includes a\nlocal MI loss to promote spatial consistency in the feature maps of the decoder\nand provide a smoother segmentation. Since mutual information does not require\na strict ordering of clusters in two different assignments, we incorporate a\nfinal consistency regularization loss on the output which helps align the\ncluster labels throughout the network. We evaluate the method on four\nchallenging publicly-available datasets for medical image segmentation.\nExperimental results show our method to outperform recently-proposed approaches\nfor semi-supervised segmentation and provide an accuracy near to full\nsupervision while training with very few annotated images.\n"}
{"abstract": "  Since the day of its explosion, SN 1987A (SN87A) was closely monitored with\nthe aim to study its evolution and to detect its central compact relic. The\ndetection of neutrinos from the supernova strongly supports the formation of a\nneutron star (NS). However, the constant and fruitless search for this object\nhas led to different hypotheses on its nature. Up to date, the detection in the\nALMA data of a feature somehow compatible with the emission arising from a\nproto Pulsar Wind Nebula (PWN) is the only hint of the existence of such\nelusive compact object. Here we tackle this 33-years old issue by analyzing\narchived observations of SN87A performed Chandra and NuSTAR in different years.\nWe firmly detect nonthermal emission in the $10-20$ kev energy band, due to\nsynchrotron radiation. The possible physical mechanism powering such emission\nis twofold: diffusive shock acceleration (DSA) or emission arising from an\nabsorbed PWN. By relating a state-of-the-art magneto-hydrodynamic simulation of\nSN87A to the actual data, we reconstruct the absorption pattern of the PWN\nembedded in the remnant and surrounded by cold ejecta. We found that, even\nthough the DSA scenario cannot be firmly excluded, the most likely scenario\nthat well explains the data is the PWN emission.\n"}
{"abstract": "  Closed form expressions to calculate the exponential of a general multivector\n(MV) in Clifford geometric algebras (GAs) Cl(p,q) are presented for n=p+q=3.\nThe obtained exponential formulas were applied to find exact GA trigonometric\nand hyperbolic functions of MV argument. We have verified that the presented\nexact formulas are in accord with series expansion of MV hyperbolic and\ntrigonometric functions. The exponentials may be applied to solve GA\ndifferential equations, in signal and image processing, automatic control and\nrobotics.\n"}
{"abstract": "  While predicting robot grasps with parallel jaw grippers have been well\nstudied and widely applied in robot manipulation tasks, the study on natural\nhuman grasp generation with a multi-finger hand remains a very challenging\nproblem. In this paper, we propose to generate human grasps given a 3D object\nin the world. Our key observation is that it is crucial to model the\nconsistency between the hand contact points and object contact regions. That\nis, we encourage the prior hand contact points to be close to the object\nsurface and the object common contact regions to be touched by the hand at the\nsame time. Based on the hand-object contact consistency, we design novel\nobjectives in training the human grasp generation model and also a new\nself-supervised task which allows the grasp generation network to be adjusted\neven during test time. Our experiments show significant improvement in human\ngrasp generation over state-of-the-art approaches by a large margin. More\ninterestingly, by optimizing the model during test time with the\nself-supervised task, it helps achieve larger gain on unseen and out-of-domain\nobjects. Project page: https://hwjiang1510.github.io/GraspTTA/\n"}
{"abstract": "  Haustral folds are colon wall protrusions implicated for high polyp miss rate\nduring optical colonoscopy procedures. If segmented accurately, haustral folds\ncan allow for better estimation of missed surface and can also serve as\nvaluable landmarks for registering pre-treatment virtual (CT) and optical\ncolonoscopies, to guide navigation towards the anomalies found in pre-treatment\nscans. We present a novel generative adversarial network, FoldIt, for\nfeature-consistent image translation of optical colonoscopy videos to virtual\ncolonoscopy renderings with haustral fold overlays. A new transitive loss is\nintroduced in order to leverage ground truth information between haustral fold\nannotations and virtual colonoscopy renderings. We demonstrate the\neffectiveness of our model on real challenging optical colonoscopy videos as\nwell as on textured virtual colonoscopy videos with clinician-verified haustral\nfold annotations. All code and scripts to reproduce the experiments of this\npaper will be made available via our Computational Endoscopy Platform at\nhttps://github.com/nadeemlab/CEP.\n"}
{"abstract": "  In this paper, the author gives a complete set of simple Yetter-Drinfeld\nmodules over Suzuki algebra $A_{N\\,2n+1}^{\\mu\\lambda}$ and investigates the\nNichols algebras over those irreducible Yetter-Drinfeld modules. The finite\ndimensional Nichols algebras of diagonal type are of Cartan type $A_1$,\n$A_1\\times A_1$, $A_2$, Super type ${A}_{2}(q;I_2)$ and the Nichols algebra\n$\\mathfrak{ufo}(8)$. And the involved finite dimensional Nichols algebras of\nnon-diagonal type are $12$, $4m$ and $m^2$ dimensional. The left three unsolved\ncases are set as open problems.\n"}
{"abstract": "  The study is devoted to geometrically non-linear modelling of viscoplastic\nstructures with residual stresses. We advocate and develop a special approach\nto residual stresses based on the transition between reference configurations.\nThe finite strain kinematics of the viscoplastic material is modelled by the\nmultiplicative decomposition of the deformation gradient tensor. Numerical\nalgorithms originally developed for unstressed materials are extended to\nmaterials with pre-stresses. Owing to the weak invariance of constitutive\nequations, the incorporation of pre-stresses happens without additional costs.\nThus, the advocated approach is especially efficient. A novel\nexperimental/theoretical method for assessment of residual stresses in welded\nstructures is presented; the method combines advantages of purely experimental\nand theoretical approaches. To demonstrate the applicability of the proposed\nprocedure, we simulate plate welding. As an example we show that the procedure\nallows to extrapolate the filed of residual stresses away from the measurement\npoints. As another example, we address the reduction of weldment-related\nresidual stresses by mechanical treatment.\n"}
{"abstract": "  Visual place recognition is one of the essential and challenging problems in\nthe fields of robotics. In this letter, we for the first time explore the use\nof multi-modal fusion of semantic and visual modalities in dynamics-invariant\nspace to improve place recognition in dynamic environments. We achieve this by\nfirst designing a novel deep learning architecture to generate the static\nsemantic segmentation and recover the static image directly from the\ncorresponding dynamic image. We then innovatively leverage the\nspatial-pyramid-matching model to encode the static semantic segmentation into\nfeature vectors. In parallel, the static image is encoded using the popular\nBag-of-words model. On the basis of the above multi-modal features, we finally\nmeasure the similarity between the query image and target landmark by the joint\nsimilarity of their semantic and visual codes. Extensive experiments\ndemonstrate the effectiveness and robustness of the proposed approach for place\nrecognition in dynamic environments.\n"}
{"abstract": "  Many operators have been bullish on the role of millimeter-wave (mmWave)\ncommunications in fifth-generation (5G) mobile broadband because of its\ncapability of delivering extreme data speeds and capacity. However, mmWave\ncomes with challenges related to significantly high path loss and\nsusceptibility to blockage. Particularly, when mmWave communication is applied\nto a mobile terminal device, communication can be frequently broken because of\nrampant hand blockage. Although a number of mobile phone companies have\nsuggested configuring multiple sets of antenna modules at different locations\non a mobile phone to circumvent this problem, identifying an optimal antenna\nmodule and a beam pair by simultaneously opening multiple sets of antenna\nmodules causes the problem of excessive power consumption and device costs. In\nthis study, a fast antenna and beam switching method termed Fast-ABS is\nproposed. In this method, only one antenna module is used for the reception to\npredict the best beam of other antenna modules. As such, unmasked antenna\nmodules and their corresponding beam pairs can be rapidly selected for\nswitching to avoid the problem of poor quality or disconnection of\ncommunications caused by hand blockage. Thorough analysis and extensive\nsimulations, which include the derivation of relevant Cram\\'{e}r-Rao lower\nbounds, show that the performance of Fast-ABS is close to that of an oracle\nsolution that can instantaneously identify the best beam of other antenna\nmodules even in complex multipath scenarios. Furthermore, Fast-ABS is\nimplemented on a software defined radio and integrated into a 5G New Radio\nphysical layer. Over-the-air experiments reveal that Fast-ABS can achieve\nefficient and seamless connectivity despite hand blockage.\n"}
{"abstract": "  The Yang-Baxter (YB) deformations of Wess-Zumino-Witten (WZW) model on the\nHeisenberg Lie group ($H_4$) are examined. We proceed to obtain the\nnonequivalent solutions of (modified) classical Yang-Baxter equation ((m)CYBE)\nfor the $ h_4$ Lie algebra by using its corresponding automorphism\ntransformation. Then we show that YB deformations of $H_4$ WZW model are split\ninto ten nonequivalent backgrounds including metric and $B$-field such that\nsome of the metrics of these backgrounds can be transformed to the metric of\n$H_{4}$ WZW model while the antisymmetric $B$-fields are changed. The rest of\nthe deformed metrics have a different isometric group structure than the\n$H_{4}$ WZW model metric. As an interesting result, it is shown that all new\nintegrable backgrounds of the YB deformed $ H_4$ WZW model are conformally\ninvariant up to two-loop order. In this way, we obtain the general form of the\ndilaton fields satisfying the vanishing beta-function equations of the\ncorresponding $\\sigma$-models.\n"}
{"abstract": "  We present the statistical methods that have been developed to analyse the\nOzDES reverberation mapping sample. To perform this statistical analysis we\nhave created a suite of customisable simulations that mimic the characteristics\nof each source in the OzDES sample. These characteristics include: the\nvariability in the photometric and spectroscopic lightcurves, the measurement\nuncertainties, and the observational cadence. By simulating the sources in the\nOzDES sample that contain the CIV emission line, we developed a set of criteria\nthat rank the reliability of a recovered time lag depending on the agreement\nbetween different recovery methods, the magnitude of the uncertainties, and the\nrate at which false positives were found in the simulations. These criteria\nwere applied to simulated light curves and these results used to estimate the\nquality of the resulting Radius-Luminosity relation.We grade the results using\nthree quality levels (gold, silver and bronze). The input slope of the R-L\nrelation was recovered within $1\\sigma$ for each of the three quality samples,\nwith the gold standard having the lowest dispersion with a recovered a R-L\nrelation slope of $0.454\\pm 0.016$ with an input slope of 0.47. Future work\nwill apply these methods to the entire OzDES sample of 771 AGN.\n"}
{"abstract": "  Optimization for sensor placement has been intensely studied to monitor\ncomplex, large scale systems, whereas one needs to overcome its intractable\nnature of the objective function for the optimization. In this study, a fast\nalgorithm for greedy sensor selection is presented for a linear reduced-ordered\nreconstruction under the assumption of correlated noise on the sensor signals.\nThe presented algorithm accomplishes the maximization of the determinant of the\nFisher information matrix in the linear inverse problem, while this study\nfirstly shows that the objective function with correlated noise is neither\nsubmodular nor supermodular. Efficient one-rank computations in the greedy\nselection procedure are introduced in both of the underdetermined and\noversampled problem. Several numerical experiments show the effectiveness of\nthe selection algorithm for its accuracy in the estimation of the states of\nlarge dimensional measurement data.\n"}
{"abstract": "  We study the magnetic moments and transition magnetic moments of $P_c$ and\n$P_{cs}$ states in the molecular picture. We first revisit the magnetic moments\nof $P_c$ states as the $S$ wave molecular states without coupled channel\neffects. The coupled channel effects and the $D$ wave contributions are then\ninvestigated carefully. The coupled channel effects contribute to the change of\n$0.1\\sim 0.4$ nuclear magneton $\\mu_N$ for most cases while the $D$ wave only\ninduces the variation of less than $0.03 ~\\mu_N$. In addition, we obtain the\ntransition magnetic moments between different $P_c$ states and the related\nelectromagnetic decay widths of $P_c'\\to P_c\\gamma$. The magnetic moments of\n$P_{cs}$ are much different for the assumption of spin being 1/2 and 3/2. The\nstudy of electromagnetic properties will help us disclose further the structure\nof these unconventional states.\n"}
{"abstract": "  Motivated by questions about $\\mathbb{C}_p$-valued Fourier transform on the\nlocally compact group $(\\mathbb{Q}_p^d,+)$, we study invariant norms on the\n$p$-adic Schr\\\"odinger representation of the Heisenberg group. Our main result\nis a minimality and rigidity property for norms in a family of invariant norms\nparameterized by a Grassmannian. This family is the orbit of the sup norm under\nthe action of the symplectic group, acting via intertwining operators. We also\nprove general fundamental properties of quotients of the universal unitary\ncompletion of cyclic algebraic representations. Combined with the rigidity\nproperty, we are able to show that the completion of the Schr\\\"odinger\nrepresentation in any of the norms in that family satisfies a strong notion of\nirreducibility and a strong version of Schur's lemma. Norms that can be formed\nas the maximum of a finite number of norms from that family are also studied.\nWe conclude this paper with a list of open questions.\n"}
{"abstract": "  We study the photon spectrum line-shape of the $J/\\psi \\to \\gamma\\eta_c\\to\n\\gamma X$ decay process in a nonrelativistic effective field theory (EFT)\nframework of quantum chromodynamics (QCD). We take into account the finite\nwidth of the $\\eta_c$ and include $\\mathcal{O}(v^2)$ corrections. We observe\nthat the photon spectrum line-shape is divergent at large energies due to\npolynomially and logarithmically divergent terms, that upon integration over\nthe photon energies in Dimensional Regularization (DR) produce no contribution\nor can be renormalized. We propose to subtract these divergences at the\nline-shape level in a manner consistent with the calculation of the width in DR\nand $\\overline{\\hbox{MS}}$ scheme. We analyze CLEO's data with the proposed\nsubtracted line-shape and find good agreement between the theoretical\nprediction and the experimental result.\n"}
{"abstract": "  In this paper, we examine the impact of cyberattacks in an integrated\ntransmission and distribution (T&D) power grid model with distributed energy\nresource (DER) integration. We adopt the OCTAVE Allegro methodology to identify\ncritical system assets, enumerate potential threats, analyze, and prioritize\nrisks for threat scenarios. Based on the analysis, attack strategies and\nexploitation scenarios are identified which could lead to system compromise.\nSpecifically, we investigate the impact of data integrity attacks in\ninverted-based solar PV controllers, control signal blocking attacks in\nprotective switches and breakers, and coordinated monitoring and switching\ntime-delay attacks.\n"}
{"abstract": "  We study to what extent the Bitcoin blockchain security permanently depends\non the underlying distribution of cryptocurrency market outcomes. We use daily\nblockchain and Bitcoin data for 2014-2019 and employ the ARDL approach. We test\nthree equilibrium hypotheses: (i) sensitivity of the Bitcoin blockchain to\nmining reward; (ii) security outcomes of the Bitcoin blockchain and the\nproof-of-work cost; and (iii) the speed of adjustment of the Bitcoin blockchain\nsecurity to deviations from the equilibrium path. Our results suggest that the\nBitcoin price and mining rewards are intrinsically linked to Bitcoin security\noutcomes. The Bitcoin blockchain security's dependency on mining costs is\ngeographically differenced - it is more significant for the global mining\nleader China than for other world regions. After input or output price shocks,\nthe Bitcoin blockchain security reverts to its equilibrium security level.\n"}
{"abstract": "  We use conformal supergravity techniques to study four-derivative corrections\nin four-dimensional gauged supergravity. We show that the four-derivative\nLagrangian for the propagating degrees of freedom of the $\\mathcal{N}=2$\ngravity multiplet is determined by two real dimensionless constants. We\ndemonstrate that all solutions of the two-derivative equations of motion in the\nsupergravity theory also solve the four-derivative equations of motion. These\nresults are then applied to explicitly calculate the regularized on-shell\naction for any asymptotically locally AdS$_4$ solution of the two-derivative\nequations of motion. The four-derivative terms in the supergravity Lagrangian\nmodify the entropy and other thermodynamic observables for the black hole\nsolutions of the theory. We calculate these corrections explicitly and\ndemonstrate that the quantum statistical relation holds for general stationary\nblack holes in the presence of the four-derivative corrections. Employing an\nembedding of this supergravity model in M-theory we show how to use\nsupersymmetric localization results in the holographically dual\nthree-dimensional SCFT to determine the unknown coefficients in the\nfour-derivative supergravity action. This in turn leads to new detailed results\nfor the first subleading $N^{\\frac{1}{2}}$ correction to the large $N$\npartition function of a class of three-dimensional SCFTs on compact Euclidean\nmanifolds. In addition, we calculate explicitly the first subleading correction\nto the Bekenstein-Hawking entropy of asymptotically AdS$_4$ black holes in\nM-theory. We also discuss how to add matter multiplets to the supergravity\ntheory in the presence of four-derivative terms and to generalize some of these\nresults to six- and higher-derivative supergravity.\n"}
{"abstract": "  Armchair phosphorene nanoribbons (APNRs) are known to be semiconductors with\nan indirect bandgap. Here, we propose to introduce new states in the gap of\nAPNRs by creating a periodic structure of vacancies (antidots). Based on the\ntight-binding model, we show that a periodic array of vacancies or nanopores\nleads to the formation of an impurity band inside the gap region. We first\npresent an analytical expression for the dispersion relation of an impurity\nband induced by hybridization of bound states associated with each single\nvacancy defect. Then, we increase the size of vacancy defects to include a\nbunch of atoms and theoretically investigate the effect of nanopores size and\ntheir spacing on electronic band structure, carrier transmission function, and\nthermoelectric properties. Our analysis of the power generation rate and\nthermoelectric efficiency of these structures reveals that an ANPR can be used\nas a superb thermoelectric power generation module.\n"}
{"abstract": "  Though Adams and Hardy-Adams inequalities can be extended to general\nsymmetric spaces of noncompact type fairly straightforwardly by following\nclosely the systematic approach developed in our early works on real and\ncomplex hyperbolic spaces, higher order Poincar\\'e-Sobolev and\nHardy-Sobolev-Maz'ya inequalities are more difficult to establish. The main\npurpose of this goal is to establish the Poincar\\'e-Sobolev and\nHardy-Sobolev-Maz'ya inequalities on quaternionic hyperbolic spaces and the\nCayley hyperbolic plane. A crucial part of our work is to establish appropriate\nfactorization theorems on these spaces which are of their independent\ninterests. To this end, we need to identify and introduce the ``Quaternionic\nGeller's operators\" and ``Octonionic Geller's operators\" which have been absent\non these spaces. Combining the factorization theorems and the Geller type\noperators with the Helgason-Fourier analysis on symmetric spaces, the precise\nheat and Bessel-Green-Riesz kernel estimates and the Kunze-Stein phenomenon for\nconnected real simple groups of real rank one with finite center, we succeed to\nestablish the higher order Poincar\\'e-Sobolev and Hardy-Sobolev-Maz'ya\ninequalities on quaternionic hyperbolic spaces and the Cayley hyperbolic plane.\nThe kernel estimates required to prove these inequalities are also sufficient\nfor us to establish, as a byproduct, the Adams and Hardy-Adams inequalities on\nthese spaces. This paper, together with our earlier works, completes our study\nof the factorization theorems, higher order Poincar\\'e-Sobolev,\nHardy-Sobolev-Maz'ya, Adams and Hardy-Adams inequalities on all rank one\nsymmetric spaces of noncompact type.\n"}
{"abstract": "  Background: There is growing evidence that social and behavioral determinants\nof health (SBDH) play a substantial effect in a wide range of health outcomes.\nElectronic health records (EHRs) have been widely employed to conduct\nobservational studies in the age of artificial intelligence (AI). However,\nthere has been little research into how to make the most of SBDH information\nfrom EHRs. Methods: A systematic search was conducted in six databases to find\nrelevant peer-reviewed publications that had recently been published. Relevance\nwas determined by screening and evaluating the articles. Based on selected\nrelevant studies, a methodological analysis of AI algorithms leveraging SBDH\ninformation in EHR data was provided. Results: Our synthesis was driven by an\nanalysis of SBDH categories, the relationship between SBDH and\nhealthcare-related statuses, and several NLP approaches for extracting SDOH\nfrom clinical literature. Discussion: The associations between SBDH and health\noutcomes are complicated and diverse; several pathways may be involved. Using\nNatural Language Processing (NLP) technology to support the extraction of SBDH\nand other clinical ideas simplifies the identification and extraction of\nessential concepts from clinical data, efficiently unlocks unstructured data,\nand aids in the resolution of unstructured data-related issues. Conclusion:\nDespite known associations between SBDH and disease, SBDH factors are rarely\ninvestigated as interventions to improve patient outcomes. Gaining knowledge\nabout SBDH and how SBDH data can be collected from EHRs using NLP approaches\nand predictive models improves the chances of influencing health policy change\nfor patient wellness, and ultimately promoting health and health equity.\n  Keywords: Social and Behavioral Determinants of Health, Artificial\nIntelligence, Electronic Health Records, Natural Language Processing,\nPredictive Model\n"}
{"abstract": "  In this paper we introduce Besov-type spaces with variable smoothness and\nintegrability. We show that these spaces are characterized by the $\\varphi\n$-transforms in appropriate sequence spaces and we obtain atomic decompositions\nfor these spaces. Moreover the Sobolev embeddings for these function spaces are\nobtained.\n"}
{"abstract": "  The professional services industry (legal, accounting, consulting,\narchitectural services{\\ldots}) employs an important share of the active\npopulation in mature countries. However, after decades of undisputed growth,\nthe sector appears to be at a turning point in certain geographies. This\narticle therefore proposes a simple framework to help diagnose where growth\nopportunities (if any) may lie.When applied to the US economic context, the\nmodel indicates that at a macro-economic national level the sector should\nstall, which concurs with the trend observed over the past decade. However, it\nalso highlights that a few industrial sectors (e.g. the US beverage industry)\nstill offer pockets of growth for a variety of professional expertises.\nReplicating and fine-tuning those findings could be interesting for\npractitioners to steer their marketing and business development efforts. On the\nother hand, the quantitative framework presented in this study could pave the\nway for future research in the academic community.\n"}
{"abstract": "  Two novel properties, unique for semiconductors: a negative electron affinity\n[1-2], and a high p-type surface electrical conductivity [3-4], were discovered\nin diamond at the end of the last century. Both properties appear when the\ndiamond surface is hydrogenated. A natural question arises: is the influence of\nthe surface hydrogen on diamond limited only to the electrical properties?\nHere, we report the first observation of a transparency peak at 1328 cm-1 in IR\nabsorption of hydrogen-terminated pure (undoped) nanodiamonds. This new optical\nproperty is ascribed to Fano-type destructive interference between zone-center\nphonons and free carriers (holes) appearing in the near-surface layer of\nhydrogenated nanodiamond. Our work opens the way to exploring the physics of\nelectron-phonon coupling in undoped diamonds and promises the application of\nthe H-terminated nanodiamonds as a new optical material with an induced\ntransparency in IR optical range.\n"}
{"abstract": "  In this paper homological ideals associated to some Nakayama algebras are\ncharacterized and enumerated via integer specializations of some suitable\nBrauer configuration algebras. Besides, it is shown how the number of such\nhomological ideals can be connected with the categorification process of\nFibonacci numbers defined by Ringel and Fahr.\n"}
{"abstract": "  Topology optimization offers optimal material layouts, enabling automation in\nthe design of devices. Given the recent advances in computer technology and\nadditive manufacturing, topology optimization is increasingly being used to\ndesign complex porous structures, for example, microfluidic devices. For the\nflow of fluids in such miniature-sized porous structures, viscous shearing\nstress will be significant. But the Darcy model -- the most popular\nmathematical model describing the flow of a single-phase incompressible fluid\nin rigid porous media -- neglects the internal friction arising from viscous\nshearing stress. We will therefore develop a material design framework under\nthe topology optimization based on the Darcy-Brinkman model -- a mathematical\nmodel for the flow of fluids through porous media that accounts for internal\nfriction besides the drag considered in the Darcy model. The proposed framework\nuses the total rate of mechanical dissipation -- a physical quantity -- for the\nobjective function. To understand the effect of viscous shearing stress on the\ndesign, we will compare the optimal material layouts provided by the Darcy and\nDarcy-Brinkman models under topology optimization. In particular, we show,\nusing analytical solutions corroborated by numerical simulations, that the\noptimal material layouts are identical for the class of problems exhibiting\naxisymmetry, for which viscous shearing stress vanishes. These analytical\nsolutions will be valuable to check the veracity of numerical simulators. For\nthose problems with dominant viscous shearing stresses (e.g., flow in a\nbackward-facing step domain), we show, using numerical solutions, that the\noptimal material layouts under the Darcy and Darcy-Brinkman models are very\ndifferent. Also, the associated solution fields (i.e., velocity and pressure)\nare qualitatively different under their respective optimal layouts for these\ntwo models.\n"}
{"abstract": "  We establish upper bounds for the $L^p$-quantization error, p in (1, 2+d),\ninduced by the recursive Markovian quantization of a d-dimensional diffusion\ndiscretized via the Euler scheme. We introduce a hybrid recursive quantization\nscheme, easier to implement in the high-dimensional framework, and establish\nupper bounds to the corresponding $L^p$-quantization error. To take advantage\nof these extensions, we propose a time discretization scheme and a recursive\nquantization-based discretization scheme associated to a reflected Backward\nStochastic Differential Equation and estimate $L^p$-error bounds induced by the\nspace approximation. We will explain how to numerically compute the solution of\nthe reflected BSDE relying on the recursive quantization and compare it to\nother types of quantization.\n"}
{"abstract": "  We present Calyx, a new intermediate language (IL) for compiling high-level\nprograms into hardware designs. Calyx combines a hardware-like structural\nlanguage with a software-like control flow representation with loops and\nconditionals. This split representation enables a new class of hardware-focused\noptimizations that require both structural and control flow information which\nare crucial for high-level programming models for hardware design. The Calyx\ncompiler lowers control flow constructs using finite-state machines and\ngenerates synthesizable hardware descriptions.\n  We have implemented Calyx in an optimizing compiler that translates\nhigh-level programs to hardware. We demonstrate Calyx using two DSL-to-RTL\ncompilers, a systolic array generator and one for a recent imperative\naccelerator language, and compare them to equivalent designs generated using\nhigh-level synthesis (HLS). The systolic arrays are $4.6\\times$ faster and\n$1.1\\times$ larger on average than HLS implementations, and the HLS-like\nimperative language compiler is within a few factors of a highly optimized\ncommercial HLS toolchain. We also describe three optimizations implemented in\nthe Calyx compiler.\n"}
{"abstract": "  The k-distance strong shortest path union cover of a graph is the minimum\ncardinality among all strong shortest path union cover at distance k of G. In\nthis paper we determine the 2-strong shortest path union cover for certain\ngraphs, also we prove that the k-strong shortest path union cover problem, in\ngeneral, is NP-complete.\n"}
{"abstract": "  This paper aims at investigating the influence of space-time curvature on the\nuncertainty relation. In particular, relying on previous findings, we assume\nthe quantum wave function to be confined to a geodesic ball on a given\nspace-like hypersurface whose radius is a measure of the position uncertainty.\nOn the other hand, we concurrently work out a viable physical definition of the\nmomentum operator and its standard deviation in the non-relativistic limit of\nthe 3+1 formalism. Finally, we evaluate the uncertainty relation which to\nsecond order depends on the Ricci scalar of the effective 3-metric and the\ncorresponding covariant derivative of the shift vector. For the sake of\nillustration, we apply our general result to a number of examples arising in\nthe context of both general relativity and extended theories of gravity.\n"}
{"abstract": "  Due to their unique persuasive power, language-capable robots must be able to\nboth act in line with human moral norms and clearly and appropriately\ncommunicate those norms. These requirements are complicated by the possibility\nthat humans may ascribe blame differently to humans and robots. In this work,\nwe explore how robots should communicate in moral advising scenarios, in which\nthe norms they are expected to follow (in a moral dilemma scenario) may be\ndifferent from those their advisees are expected to follow. Our results suggest\nthat, in fact, both humans and robots are judged more positively when they\nprovide the advice that favors the common good over an individual's life. These\nresults raise critical new questions regarding people's moral responses to\nrobots and the design of autonomous moral agents.\n"}
{"abstract": "  $\\textit{RigL}$, a sparse training algorithm, claims to directly train sparse\nnetworks that match or exceed the performance of existing dense-to-sparse\ntraining techniques (such as pruning) for a fixed parameter count and compute\nbudget. We implement $\\textit{RigL}$ from scratch in Pytorch and reproduce its\nperformance on CIFAR-10 within 0.1% of the reported value. On both\nCIFAR-10/100, the central claim holds -- given a fixed training budget,\n$\\textit{RigL}$ surpasses existing dynamic-sparse training methods over a range\nof target sparsities. By training longer, the performance can match or exceed\niterative pruning, while consuming constant FLOPs throughout training. We also\nshow that there is little benefit in tuning $\\textit{RigL}$'s hyper-parameters\nfor every sparsity, initialization pair -- the reference choice of\nhyperparameters is often close to optimal performance. Going beyond the\noriginal paper, we find that the optimal initialization scheme depends on the\ntraining constraint. While the Erdos-Renyi-Kernel distribution outperforms the\nUniform distribution for a fixed parameter count, for a fixed FLOP count, the\nlatter performs better. Finally, redistributing layer-wise sparsity while\ntraining can bridge the performance gap between the two initialization schemes,\nbut increases computational cost.\n"}
{"abstract": "  Current formulations of nonequilibrium thermodynamics of open chemical\nreaction networks only consider chemostats as free-energy sources sustaining\nnonequilibrium behaviours. Here, we extend the theory to include incoherent\nlight as a source of free energy. We do so by relying on a local equilibrium\nassumption to derive the chemical potential of photons relative to the system\nthey interact with. This allows us to identify the thermodynamic potential and\nthe thermodynamic forces driving light-reacting chemical systems out of\nequilibrium. We use this framework to treat two paradigmatic photochemical\nmechanisms describing light-induced unimolecular reactions -- namely the\nadiabatic and diabatic mechanisms -- and highlight the different thermodynamics\nthey lead to. Furthermore, using a thermodynamic coarse-graining procedure, we\nexpress our findings in terms of commonly measured experimental quantities such\nas quantum yields.\n"}
{"abstract": "  Combining the $b\\to s\\mu^+\\mu^-$ anomaly and dark matter observables, we\nstudy the capability of LHC to test dark matter, $Z^{\\prime}$, and vector-like\nquark. We focus on a local $U(1)_{L_\\mu-L_\\tau}$ model with a vector-like\n$SU(2)_L$ doublet quark $Q$ and a complex singlet scalar whose lightest\ncomponent $X_I$ is a candidate of dark matter. After imposing relevant\nconstraints, we find that the $b\\to s\\mu^+\\mu^-$ anomaly and the relic\nabundance of dark matter favor $m_{X_I}< 350$ GeV and $m_{Z^{\\prime}}< 450$ GeV\nfor $m_Q<$ 2 TeV and $m_{X_R}<$ 2 TeV (the heavy partner of $m_{X_I}$). The\ncurrent searches for jets and missing transverse momentum at the LHC sizably\nreduce the mass ranges of the vector-like quark, and $m_Q$ is required to be\nlarger than 1.7 TeV. Finally, we discuss the possibility of probing these new\nparticles at the high luminosity LHC via the QCD process $pp \\to D\\bar{D}$ or\n$pp \\to U\\bar{U}$ followed by the decay $D\\to s (b) Z'X_I$ or $U \\to u (c) Z'\nX_I$ and then $Z'\\to\\mu^+\\mu^-$. Taking a benchmark point of $m_Q$=1.93 TeV,\n$m_{Z^\\prime}=170$ GeV, and $m_{X_I}=$ 145 GeV, we perform a detailed Monte\nCarlo simulation, and find that such benchmark point can be accessible at the\n14 TeV LHC with an integrated luminosity 3000 fb$^{-1}$.\n"}
{"abstract": "  We prove a proposition that connects constant-PAF sequences and the\ncorresponding Legendre pairs with integer PSD values. We show how to determine\nexplicitly the complete spectrum of the $(\\ell/3)$-rd value of the discrete\nFourier transform for Legendre pairs of lengths $\\ell \\equiv 0 \\, (\\mod 3)$.\nThis is accomplished by two new algorithms based on number-theoretic arguments.\nAs an application, we prove that Legendre pairs of the open lengths 117, 129,\n133, and 147 exist by finding Legendre pairs of these lengths with a multiplier\ngroup of order at least 3. As a consequence, 85, 87, 115, 145, 159, 161, 169,\n175, 177, 185, 187, 195 are the twelve integers in the range < 200 for which\nthe question of existence of Legendre pairs remains unsolved.\n"}
{"abstract": "  This note is an erratum to the paper \"Tautological classes on moduli spaces\nof hyper-K\\\"ahler manifolds.\" Thorsten Beckman and Mirko Mauri have pointed to\nus a gap in the proof of \\cite[Theorem 8.2.1]{Duke}. We do not know how to\ncorrect the proof. We can only recover a partial statement. This gap affects\nthe proof of one of the two main results of \\cite{Duke}, we explain how to\ncorrect it.\n"}
{"abstract": "  The determination of absolute and relative distances of molecular clouds\nalong the line-of-sight towards the central molecular zone (CMZ) is crucial to\ninfer its orbital structure, dynamics, and to understand star formation in the\nclouds. Recent results by Zoccali et al. 2021 suggest that the G0.253 + 0.016\ncloud (the Brick) does not belong to the CMZ. This motivated us to cross check\ntheir results computing the absolute and relative distance to the Brick and\nalso to other two molecular clouds (the 50 km/s, and the 20 km/s clouds), and\ndiscuss their CMZ membership. We used the colour magnitude diagrams $K_s$ vs.\n$H-K_s$ to compare stars detected towards the target clouds with stars detected\ntowards three reference regions in the nuclear stellar disc (NSD) and the\nGalactic bulge. We used red clump (RC) stars to estimate the distance to each\nregion. We obtained that all the clouds present a double RC feature. Such a\ndouble RC has been reported in previous work for the NSD, but not for the bulge\nadjacent to it. We exclude the possibility that the different RC features are\nlocated at significantly different distances. The obtained absolute and\nrelative distances are compatible with the Galactic centre distance ($\\sim8$\nkpc).\n"}
{"abstract": "  We introduce the Python package, FDApy, as an implementation of functional\ndata. This package provide modules for the analysis of such data. It includes\nclasses for different dimensional data as well as irregularly sampled\nfunctional data. A simulation toolbox is also provided. It might be used to\nsimulate different clusters of functional data. Some methodologies to handle\nthese data are implemented, such as dimension reduction and clustering. New\nmethods can be easily added. The package is publicly available on the Python\nPackage Index and Github.\n"}
{"abstract": "  In MRI, deep neural networks have been proposed to reconstruct diffusion\nmodel parameters. However, the inputs of the networks were designed for a\nspecific diffusion gradient scheme (i.e., diffusion gradient directions and\nnumbers) and a specific b-value that are the same as the training data. In this\nstudy, a new deep neural network, referred to as DIFFnet, is developed to\nfunction as a generalized reconstruction tool of the diffusion-weighted signals\nfor various gradient schemes and b-values. For generalization, diffusion\nsignals are normalized in a q-space and then projected and quantized, producing\na matrix (Qmatrix) as an input for the network. To demonstrate the validity of\nthis approach, DIFFnet is evaluated for diffusion tensor imaging (DIFFnetDTI)\nand for neurite orientation dispersion and density imaging (DIFFnetNODDI). In\neach model, two datasets with different gradient schemes and b-values are\ntested. The results demonstrate accurate reconstruction of the diffusion\nparameters at substantially reduced processing time (approximately 8.7 times\nand 2240 times faster processing time than conventional methods in DTI and\nNODDI, respectively; less than 4% mean normalized root-mean-square errors\n(NRMSE) in DTI and less than 8% in NODDI). The generalization capability of the\nnetworks was further validated using reduced numbers of diffusion signals from\nthe datasets. Different from previously proposed deep neural networks, DIFFnet\ndoes not require any specific gradient scheme and b-value for its input. As a\nresult, it can be adopted as an online reconstruction tool for various complex\ndiffusion imaging.\n"}
{"abstract": "  The construction of HNN-extensions of involutive Hom-associative algebras and\ninvolutive Hom-Lie algebras is described. Then, as an application of\nHNN-extension, by using the validity of Poincar\\'e-Birkhoff-Witt theorem for\ninvolutive Hom-Lie algebras, we provide an embedding theorem.\n"}
{"abstract": "  The optical binding of many particles has great potential to achieve the\nwide-area formation of a \"crystal\" of small materials. Unlike conventional\noptical binding, where the whole assembly of targeted particles is irradiated\nwith light, if one can indirectly manipulate remote particles using a single\ntrapped particle through optical binding, the degrees of freedom to create\nordered structures will be greatly enhanced. In this Letter, we theoretically\ninvestigate the dynamics of the assembly of gold nanoparticles that is\nmanipulated using a single particle trapped by a focused laser. As a result, we\ndemonstrate that the spin--orbit coupling and angular momentum generation of\nlight via scattering induce the assembly and rotational motion of particles\nthrough indirect optical force. This result opens the possibility of creating\nordered structures with a wide area and manipulating them, controlling local\nproperties using scanning laser beams.\n"}
{"abstract": "  Adaptive charging can charge electric vehicles (EVs) at scale cost\neffectively, despite the uncertainty in EV arrivals. We formulate adaptive EV\ncharging as a feasibility problem that meets all EVs' energy demands before\ntheir deadlines while satisfying constraints in charging rate and total\ncharging power. We propose an online algorithm, smoothed least-laxity-first\n(sLLF), that decides the current charging rates without the knowledge of future\narrivals and demands. We characterize the performance of the sLLF algorithm\nanalytically and numerically. Numerical experiments with real-world data show\nthat it has a significantly higher rate of feasible EV charging than several\nother existing EV charging algorithms. Resource augmentation framework is\nemployed to assess the feasibility condition of the algorithm. The assessment\nshows that the sLLF algorithm achieves perfect feasibility with only a 0.07\nincrease in resources.\n"}
{"abstract": "  Nanoelectromechanical resonators have been successfully used for a variety of\nsensing applications. Their extreme resolution comes from their small size at\nthe cost of low capture area, making the \"needle in a haystack\" issue acute.\nThis leads to poor instrument sensitivity and long analysis time. Moreover,\nelectrical transductions are limited in frequency, which limits the achievable\nmechanical bandwidth again limiting throughput. Multiplexing a large number of\nhigh-frequency resonators appears as a solution, but this is complex with\nelectrical transductions. We propose here a route to solve these issues, with a\nmultiplexing scheme for very high frequency optomechanical resonators. We\ndemonstrate the simultaneous frequency measurement of three silicon microdisks\nresonators fabricated through a Very Large Scale Integration process. The\nreadout architecture is simple and does not degrade the sensing resolutions.\nThis paves the way towards the realization of sensors for multi-parametric\nanalysis, extremely low limit of detection and response time.\n"}
{"abstract": "  A surprising and unexpected phenomenon observed during Cassini's Grand Finale\nwas the spacecraft charging to positive potentials in Saturn's ionosphere.\nHere, the ionospheric plasma was depleted of free electrons with negatively\ncharged ions and dust accumulating up to over 95 % of the negative charge\ndensity. To further understand the spacecraft-plasma interaction, we perform a\nthree dimensional Particle-In-Cell study of a model Cassini spacecraft immersed\nin plasma representative of Saturn's ionosphere. The simulations reveal complex\ninteraction features such as electron wings and a highly structured wake\ncontaining spacecraft-scale vortices. The results show how a large negative ion\nconcentration combined with a large negative to positive ion mass ratio is able\nto drive the spacecraft to the observed positive potentials. Despite the high\nelectron depletions, the electron properties are found as a significant\ncontrolling factor for the spacecraft potential together with the magnetic\nfield orientation which induces a potential gradient directed across Cassini's\nasymmetric body. This study reveals the global spacecraft interaction\nexperienced by Cassini during the Grand Finale and how this is influenced by\nthe unexpected negative ion and dust populations.\n"}
{"abstract": "  The Vertical Shear Instability is an axisymmetric effect suggested to drive\nturbulence in the magnetically inactive zones of protoplanetary accretion\ndisks. Here we examine its physical mechanism in analytically tractable\n``minimal models\" in three settings that include a uniform density fluid, a\nstratified atmosphere, and a shearing-box section of a protoplanetary disk.\nEach of these analyses show that the vertical shear instability's essence is\nsimilar to the slantwise convective symmetric instability in the mid-latitude\nEarth atmosphere, in the presence of vertical shear of the baroclinic jet\nstream, as well as mixing in the top layers of the Gulf Stream. We show that in\norder to obtain instability the fluid parcels' slope should exceed the slope of\nthe mean absolute momentum in the disk radial-vertical plane. We provide a\ndetailed and mutually self-consistent physical explanation from three\nperspectives: in terms of angular momentum conservation, as a dynamical\ninterplay between a fluid's radial and azimuthal vorticity components, and from\nan energy perspective involving a generalized Solberg-H{\\o}iland Rayleigh\ncondition. Furthermore, we explain why anelastic dynamics yield oscillatory\nunstable modes and isolate the oscillation mechanism from the instability one.\n"}
{"abstract": "  We propose a novel, efficient, modular and scalable framework for content\nbased visual media retrieval systems by leveraging the power of Deep Learning\nwhich is flexible to work both for images and videos conjointly and we also\nintroduce an efficient comparison and filtering metric for retrieval. We put\nforward our findings from critical performance tests comparing our method to\nthe predominant conventional approach to demonstrate the feasibility and\nefficiency of the proposed solution with best practices, possible improvements\nthat may further augment the ability of retrieval architectures.\n"}
{"abstract": "  Recently, the action growth rate of a variety of four-dimensional regular\nmagnetic black holes in F frame is obtained in [1]. Here, we study the action\ngrowth rate of a four-dimensional regular electric black hole in P frame that\nis the Legendre transformation of F frame. We also investigate the action\ngrowth rates of the Wheeler-De Witt patch for such black hole configurations at\nthe late time and examine the Lloyd bound on the rate of quantum computation.\nWe show that although the form of the Lloyd bound formula remains unaltered,\nthe energy modifies due to a non-vanishing trace of the energy-momentum tensor\nand some extra terms may appear in the total growth action. We also investigate\nthe asymptotic behavior of complexity in two conjectures for static and\nrotating regular black holes.\n"}
{"abstract": "  The research field of adversarial machine learning witnessed a significant\ninterest in the last few years. A machine learner or model is secure if it can\ndeliver main objectives with acceptable accuracy, efficiency, etc. while at the\nsame time, it can resist different types and/or attempts of adversarial\nattacks. This paper focuses on studying aspects and research trends in\nadversarial machine learning specifically in text analysis and generation. The\npaper summarizes main research trends in the field such as GAN algorithms,\nmodels, types of attacks, and defense against those attacks.\n"}
{"abstract": "  This paper introduces our position on the critical issue of bias that\nrecently appeared in AI applications. Specifically, we discuss the combination\nof current technologies used in AI applications i.e., Machine Learning and\nKnowledge Graphs, and point to their involvement in (de)biased applications of\nthe C4I domain. Although this is a wider problem that currently emerges from\ndifferent application domains, bias appears more critical in C4I than in others\ndue to its security-related nature. While proposing certain actions to be taken\ntowards debiasing C4I applications, we acknowledge the immature aspect of this\ntopic within the Knowledge Graph and Semantic Web communities.\n"}
{"abstract": "  Machine learning has proved to be a useful tool for extracting knowledge from\nscientific data in numerous research fields, including astrophysics, genomics,\nand molecular dynamics. Often, data sets from these research areas need to be\nprocessed in distributed platforms due to their magnitude. This can be done\nusing one of the various distributed machine learning libraries available. One\nof these libraries is dislib, a distributed machine learning library for Python\nespecially designed to process large scale data sets on HPC clusters, which\nmakes dislib an ideal candidate for analyzing scientific data. However,\ndislib's main distributed data structure, called Dataset, has some limitations,\nincluding poor performance in certain operations and low flexibility and\nusability. In this paper, we propose a novel distributed data structure for\ndislib, called ds-array, that addresses dislib's main limitations in data\nmanagement. Ds-arrays simplify distributed data management in dislib by\nexposing a NumPy-like API, provide more flexibility, and reduce the\ncomputational complexity of some operations. This results in performance\nimprovements of up to two orders of magnitude over Datasets, while also greatly\nimproving scalability and usability.\n"}
{"abstract": "  Derivatives of equations of motion describing the rigid body dynamics are\nbecoming increasingly relevant for the robotics community and find many\napplications in design and control of robotic systems. Controlling robots, and\nmultibody systems comprising elastic components in particular, not only\nrequires smooth trajectories but also the time derivatives of the control\nforces/torques, hence of the equations of motion (EOM). This paper presents\nnovel nth order time derivatives of the EOM in both closed and recursive forms.\nWhile the former provides a direct insight into the structure of these\nderivatives,the latter leads to their highly efficient implementation for large\ndegree of freedom robotic system.\n"}
{"abstract": "  The integer quantum Hall transition (IQHT) is one of the most mysterious\nmembers of the family of Anderson transitions. Since the 1980s, the scaling\nbehavior near the IQHT has been vigorously studied in experiments and numerical\nsimulations. Despite all efforts, it is notoriously difficult to pin down the\nprecise values of critical exponents, which seem to vary with model details and\nthus challenge the principle of universality. Recently, M.\nZirnbauer\\citep{Zirnbauer2019} [Nucl. Phys. B \\textbf{941}, 458 (2019)] has\nconjectured a conformal field theory for the transition, in which linear terms\nin the beta-functions vanish, leading to a very slow flow in the fixed point's\nvicinity which we term marginal scaling. In this work, we provide numerical\nevidence for such a scenario by using extensive simulations of various network\nmodels of the IQHT at unprecedented length scales. At criticality, we show that\nthe finite-size scaling of the disorder averaged longitudinal Landauer\nconductance is consistent with its recently predicted fixed-point value and a\nthird-order expansion of RG beta functions. In the future, our numerical\nfindings can be checked with analytical results from the conformal field\ntheory. Away from criticality we describe a mechanism that could account for\nthe emergence of an \\emph{effective} critical exponents $\\nu_\\mathrm{eff}$,\nwhich is necessarily dependent on the parameters of the model. We further\nsupport this idea by numerical determination of $\\nu_\\mathrm{eff}$ in suitably\nchosen models.\n"}
{"abstract": "  We expand upon the simulation study of Setodji et al. (2017) which compared\nthree promising balancing methods when assessing the average treatment effect\non the treated for binary treatments: generalized boosted models (GBM),\ncovariate-balancing propensity scores (CBPS), and entropy balance (EB). The\nstudy showed that GBM can outperform CBPS and EB when there are likely to be\nnon-linear associations in both the treatment assignment and outcome models and\nCBPS and EB are fine-tuned to obtain balance only on first order moments. We\nexplore the potential benefit of using higher-order moments in the balancing\nconditions for CBPS and EB. Our findings showcase that CBPS and EB should, by\ndefault, include higher order moments and that focusing only on first moments\ncan result in substantial bias in both CBPS and EB estimated treatment effect\nestimates that could be avoided by the use of higher moments.\n"}
{"abstract": "  Simultaneous Localization and Mapping (SLAM) is one of the key robotics tasks\nas it tackles simultaneous mapping of the unknown environment defined by\nmultiple landmark positions and localization of the unknown pose (i.e.,\nattitude and position) of the robot in three-dimensional (3D) space. The true\nSLAM problem is modeled on the Lie group of $\\mathbb{SLAM}_{n}\\left(3\\right)$,\nand its true dynamics rely on angular and translational velocities. This paper\nproposes a novel geometric nonlinear stochastic estimator algorithm for SLAM on\n$\\mathbb{SLAM}_{n}\\left(3\\right)$ that precisely mimics the nonlinear motion\ndynamics of the true SLAM problem. Unlike existing solutions, the proposed\nstochastic filter takes into account unknown constant bias and noise attached\nto the velocity measurements. The proposed nonlinear stochastic estimator on\nmanifold is guaranteed to produce good results provided with the measurements\nof angular velocities, translational velocities, landmarks, and inertial\nmeasurement unit (IMU). Simulation and experimental results reflect the ability\nof the proposed filter to successfully estimate the six-degrees-of-freedom (6\nDoF) robot's pose and landmark positions. Keywords: Simultaneous Localization\nand Mapping, nonlinear stochastic observer for SLAM, stochastic differential\nequations, pose estimator, position, attitude, Brownian motion process,\ninertial measurement unit, landmarks, features, SDE, SO(3), SE(3), SLAM.\n"}
{"abstract": "  In this paper, thanks to the generalizations of the dual spaces of the\nHardy-amalgam spaces $\\mathcal H^{(q,p)}$ and\n$\\mathcal{H}_{\\mathrm{loc}}^{(q,p)}$ for $0<q\\leq1$ and $q\\leq p<\\infty$,\nobtained in our earlier paper, we prove that the inclusion of $\\mathcal\nH^{(1,p)}$ in $(L^1,\\ell^p)$ for $1\\leq p<\\infty$ is strict, and more\ngenerally, the one of $\\mathcal H^{(q,p)}$ in\n$\\mathcal{H}_{\\mathrm{loc}}^{(q,p)}$ for $0<q\\leq1$ and $q\\leq p<\\infty$.\nMoreover, as other applications, we obtain results of boundedness of\nCalder\\'on-Zygmund and convolution operators, generalizing those known in the\ncontext of the spaces $\\mathcal H^1$ and $BMO(\\mathbb{R}^d)$.\n"}
{"abstract": "  We analyse backward Euler time stepping schemes for the primal DPG\nformulation of a class of parabolic problems. Optimal error estimates are shown\nin the natural norm and in the $L^2$ norm of the field variable. For the heat\nequation the solution of our primal DPG formulation equals the solution of a\nstandard Galerkin scheme and, thus, optimal error bounds are found in the\nliterature. In the presence of advection and reaction terms, however, the\nlatter identity is not valid anymore and the analysis of optimal error bounds\nrequires to resort to elliptic projection operators. It is essential that these\noperators be projections with respect to the spatial part of the PDE, as in\nstandard Galerkin schemes, and not with respect to the full PDE at a time step,\nas done previously.\n"}
{"abstract": "  Visual attention in Visual Question Answering (VQA) targets at locating the\nright image regions regarding the answer prediction, offering a powerful\ntechnique to promote multi-modal understanding. However, recent studies have\npointed out that the highlighted image regions from the visual attention are\noften irrelevant to the given question and answer, leading to model confusion\nfor correct visual reasoning. To tackle this problem, existing methods mostly\nresort to aligning the visual attention weights with human attentions.\nNevertheless, gathering such human data is laborious and expensive, making it\nburdensome to adapt well-developed models across datasets. To address this\nissue, in this paper, we devise a novel visual attention regularization\napproach, namely AttReg, for better visual grounding in VQA. Specifically,\nAttReg firstly identifies the image regions which are essential for question\nanswering yet unexpectedly ignored (i.e., assigned with low attention weights)\nby the backbone model. And then a mask-guided learning scheme is leveraged to\nregularize the visual attention to focus more on these ignored key regions. The\nproposed method is very flexible and model-agnostic, which can be integrated\ninto most visual attention-based VQA models and require no human attention\nsupervision. Extensive experiments over three benchmark datasets, i.e., VQA-CP\nv2, VQA-CP v1, and VQA v2, have been conducted to evaluate the effectiveness of\nAttReg. As a by-product, when incorporating AttReg into the strong baseline\nLMH, our approach can achieve a new state-of-the-art accuracy of 60.00% with an\nabsolute performance gain of 7.01% on the VQA-CP v2 benchmark dataset...\n"}
{"abstract": "  Understanding assembly instruction has the potential to enhance the robot s\ntask planning ability and enables advanced robotic applications. To recognize\nthe key components from the 2D assembly instruction image, We mainly focus on\nsegmenting the speech bubble area, which contains lots of information about\ninstructions. For this, We applied Cascade Mask R-CNN and developed a\ncontext-aware data augmentation scheme for speech bubble segmentation, which\nrandomly combines images cuts by considering the context of assembly\ninstructions. We showed that the proposed augmentation scheme achieves a better\nsegmentation performance compared to the existing augmentation algorithm by\nincreasing the diversity of trainable data while considering the distribution\nof components locations. Also, we showed that deep learning can be useful to\nunderstand assembly instruction by detecting the essential objects in the\nassembly instruction, such as tools and parts.\n"}
{"abstract": "  We formulate an alternative approach to describing Ehresmann semigroups by\nmeans of left and right \\'etale actions of a meet semilattice on a category. We\nalso characterize the Ehresmann semigroups that arise as the set of all subsets\nof a finite category. As applications, we prove that every restriction\nsemigroup can be nicely embedded into a restriction semigroup constructed from\na category, and we describe when a restriction semigroup can be nicely embedded\ninto an inverse semigroup.\n"}
{"abstract": "  We follow up on the idea of Lars Arge to rephrase the Reduce and Apply\nprocedures of Binary Decision Diagrams (BDDs) as iterative I/O-efficient\nalgorithms. We identify multiple avenues to simplify and improve the\nperformance of his proposed algorithms. Furthermore, we extend the technique to\nother common BDD operations, many of which are not derivable using Apply\noperations alone, and we provide asymptotic improvements for the procedures\nthat can be derived using Apply.\n  These algorithms are implemented in a new BDD package, named Adiar. We see\nvery promising results when comparing the performance of Adiar with\nconventional BDD package that use recursive depth-first algorithms. For\ninstances larger than 9.5 GiB, our algorithms, in parts using the disk, are\n1.47 to 3.69 times slower compared to CUDD and Sylvan, exclusively using main\nmemory. Yet, our proposed techniques are able to obtain this performance at a\nfraction of the main memory needed by conventional BDD packages to function.\nFurthermore, with Adiar we are able to manipulate BDDs that outgrow main memory\nand so surpass the limits of other BDD packages.\n"}
{"abstract": "  This paper addresses the problem of optimal control using search trees. We\nstart by considering multi-armed bandit problems with continuous action spaces\nand propose LD-HOO, a limited depth variant of the hierarchical optimistic\noptimization (HOO) algorithm. We provide a regret analysis for LD-HOO and show\nthat, asymptotically, our algorithm exhibits the same cumulative regret as the\noriginal HOO while being faster and more memory efficient. We then propose a\nMonte Carlo tree search algorithm based on LD-HOO for optimal control problems\nand illustrate the resulting approach's application in several optimal control\nproblems.\n"}
{"abstract": "  Trustworthy deployment of ML models requires a proper measure of uncertainty,\nespecially in safety-critical applications. We focus on uncertainty\nquantification (UQ) for classification problems via two avenues -- prediction\nsets using conformal prediction and calibration of probabilistic predictors by\npost-hoc binning -- since these possess distribution-free guarantees for i.i.d.\ndata. Two common ways of generalizing beyond the i.i.d. setting include\nhandling covariate and label shift. Within the context of distribution-free UQ,\nthe former has already received attention, but not the latter. It is known that\nlabel shift hurts prediction, and we first argue that it also hurts UQ, by\nshowing degradation in coverage and calibration. Piggybacking on recent\nprogress in addressing label shift (for better prediction), we examine the\nright way to achieve UQ by reweighting the aforementioned conformal and\ncalibration procedures whenever some unlabeled data from the target\ndistribution is available. We examine these techniques theoretically in a\ndistribution-free framework and demonstrate their excellent practical\nperformance.\n"}
{"abstract": "  TeraHertz (THz) communications are envisioned as a promising technology,\nowing to its unprecedented multi-GHz bandwidth. One fundamental challenge when\nmoving to new spectrum is to understand the science of radio propagation and\ndevelop an accurate channel model. In this paper, a wideband channel\nmeasurement campaign between 130 GHz and 143 GHz is investigated in a typical\nmeeting room. Directional antennas are utilized and rotated for resolving the\nmulti-path components (MPCs) in the angular domain. With careful system\ncalibration that eliminates system errors and antenna effects, a realistic\npower delay profile is developed. Furthermore, a combined MPC clustering and\nmatching procedure with ray-tracing techniques is proposed to investigate the\ncluster behavior and wave propagation of THz signals. In light of the\nmeasurement results, physical parameters and insights in the THz indoor channel\nare comprehensively analyzed, including the line-of-sight path loss, power\ndistributions, temporal and spatial features, and correlations among THz\nmulti-path characteristics. Finally, a hybrid channel model that combines\nray-tracing and statistical methods is developed for THz indoor communications.\nNumerical results demonstrate that the proposed hybrid channel model shows good\nagreement with the measurement and outperforms the conventional statistical and\ngeometric-based stochastic channel model in terms of the temporal-spatial\ncharacteristics.\n"}
{"abstract": "  In this paper, we introduce a Liouville action for a harmonic diffeomorphism\nfrom a compact Riemann surface to a compact hyperbolic Riemann surface of genus\n$g\\ge 2$. We derive the variational formula of this Liouville action for\nharmonic diffeomorphisms when the source Riemann surfaces vary with a fixed\ntarget Riemann surface.\n"}
{"abstract": "  Streaming end-to-end automatic speech recognition (ASR) systems are widely\nused in everyday applications that require transcribing speech to text in\nreal-time. Their minimal latency makes them suitable for such tasks. Unlike\ntheir non-streaming counterparts, streaming models are constrained to be causal\nwith no future context and suffer from higher word error rates (WER). To\nimprove streaming models, a recent study [1] proposed to distill a\nnon-streaming teacher model on unsupervised utterances, and then train a\nstreaming student using the teachers' predictions. However, the performance gap\nbetween teacher and student WERs remains high. In this paper, we aim to close\nthis gap by using a diversified set of non-streaming teacher models and\ncombining them using Recognizer Output Voting Error Reduction (ROVER). In\nparticular, we show that, despite being weaker than RNN-T models, CTC models\nare remarkable teachers. Further, by fusing RNN-T and CTC models together, we\nbuild the strongest teachers. The resulting student models drastically improve\nupon streaming models of previous work [1]: the WER decreases by 41% on\nSpanish, 27% on Portuguese, and 13% on French.\n"}
{"abstract": "  Vanadium dioxide (VO2) is a phase transition material that undergoes a\nreversible insulator-metal phase transition at ~ 68 C. Atmospheric pressure\nthermal oxidation (APTO) of vanadium (V) is a simple VO2 synthesis method in\nwhich V thin film is oxidized in open air. For an optimum oxidation duration,\nVO2 films are obtained with good phase transition properties. We recently\nreported a modified APTO process using a step temperature profile for oxidation\n(Thin Solid Films 706, 138003 (2020)). We demonstrated an ultra-low thermal\nbudget synthesis of VO2 thin films with good electrical and optical phase\ntransition properties. For a 130 nm room-temperature RF sputtered V thin film,\nan optimum oxidation duration of ~ 30 s was obtained. In this work, we study\nhow the starting V film thickness and deposition temperature affects the\noptimum oxidation duration. V thin films of varying thickness (15-212 nm) and\n120 nm thick V films with varying deposition temperature (~27-450 C) are\nprepared using RF magnetron sputtering. These films are oxidized for different\noxidation durations and characterized using Raman and four-probe measurements\nto find the optimum oxidation duration for each deposition condition. We find\nthat the optimum oxidation duration increases with the increase in V film\nthickness and V deposition temperature. We model the effect of V film thickness\nand deposition temperature on the optimal oxidation time using a parabolic law\nwhich can be used to obtain the optimal oxidation times for intermediate V\nthicknesses/deposition temperatures.\n"}
{"abstract": "  Because of the explosive growth of face photos as well as their widespread\ndissemination and easy accessibility in social media, the security and privacy\nof personal identity information becomes an unprecedented challenge. Meanwhile,\nthe convenience brought by advanced identity-agnostic computer vision\ntechnologies is attractive. Therefore, it is important to use face images while\ntaking careful consideration in protecting people's identities. Given a face\nimage, face de-identification, also known as face anonymization, refers to\ngenerating another image with similar appearance and the same background, while\nthe real identity is hidden. Although extensive efforts have been made,\nexisting face de-identification techniques are either insufficient in\nphoto-reality or incapable of well-balancing privacy and utility. In this\npaper, we focus on tackling these challenges to improve face de-identification.\nWe propose IdentityDP, a face anonymization framework that combines a\ndata-driven deep neural network with a differential privacy (DP) mechanism.\nThis framework encompasses three stages: facial representations\ndisentanglement, $\\epsilon$-IdentityDP perturbation and image reconstruction.\nOur model can effectively obfuscate the identity-related information of faces,\npreserve significant visual similarity, and generate high-quality images that\ncan be used for identity-agnostic computer vision tasks, such as detection,\ntracking, etc. Different from the previous methods, we can adjust the balance\nof privacy and utility through the privacy budget according to pratical demands\nand provide a diversity of results without pre-annotations. Extensive\nexperiments demonstrate the effectiveness and generalization ability of our\nproposed anonymization framework.\n"}
{"abstract": "  We study the bilinear and higher-order fermion condensates in $4$-dimensional\n$SU(N)$ gauge theories with a single Dirac fermion in a general representation.\nAugmented with a mixed anomaly between the $0$-form discrete chiral, $1$-form\ncenter, and $0$-form baryon number symmetries (BC anomaly), we sort out\ntheories that admit higher-order condensates and vanishing fermion bilinears.\nThen, the BC anomaly is utilized to prove, in the absence of a topological\nquantum field theory, that nonvanishing fermion bilinears are inevitable in\ninfrared-gapped theories with $2$-index (anti)symmetric fermions. We also\ncontrast the BC anomaly with the $0$-form anomalies and show that it is the\nformer anomaly that determines the infrared physics; we argue that the BC\nanomaly lurks deep to the infrared while the $0$-form anomalies are just\nvariations of local terms. We provide evidence of this assertion by studying\nthe BC anomaly in vector-like theories compactified on a small spacial circle.\nThese theories are weakly-coupled, under analytical control, and they admit a\ndual description in terms of abelian photons that determine the deep infrared\ndynamics. We show that the dual photons talk directly to the $1$-form center\nsymmetry in order to match the BC anomaly, while the $0$-form anomalies are\nvariations of local terms and are matched by fiat. Finally, we study the fate\nof the BC anomaly in the compactified theories when they are held at a finite\ntemperature. The effective field theory that describes the low-energy physics\nis $2$-dimensional. We show that the BC anomaly cascades from $4$ to $2$\ndimensions.\n"}
{"abstract": "  The rapidly expanding nature of the Internet of Things (IoT) networks is\nbeginning to attract interest across a range of applications, including smart\nhomes, smart transportation, smart health, and industrial contexts. This\ncutting-edge technology enables individuals to track and control their\nintegrated environment in real-time and remotely via a thousand IoT devices\ncomprised of sensors and actuators that actively participate in sensing,\nprocessing, storing, and sharing information. Nonetheless, IoT devices are\nfrequently deployed in hostile environments, wherein adversaries attempt to\ncapture and breach them in order to seize control of the entire network. One\nsuch example of potentially malicious behaviour is the cloning of IoT devices,\nin which an attacker can physically capture the devices, obtain some sensitive\ninformation, duplicate the devices, and intelligently deploy them in desired\nlocations to conduct various insider attacks. A device cloning attack on IoT\nnetworks is a significant security concern since it allows for selective\nforwarding, sink-hole, and black-hole attacks. To address this issue, this\npaper provides an efficient scheme for detecting clone node attacks on IoT\nnetworks that makes use of semantic information about IoT devices known as\ncontext information sensed from the deployed environment to locate them\nsecurely. We design a location proof mechanism by combining location proofs and\nbatch verification of the extended elliptic curve digital signature technique\nto accelerate the verification process at selected trusted nodes. We\ndemonstrate the security of our scheme and its resilience to secure clone node\nattack detection by conducting a comprehensive security analysis. The\nperformance of our proposed scheme provides a high degree of detection accuracy\nwith minimal detection time and significantly reduces the computation,\ncommunication and storage overhead.\n"}
{"abstract": "  In 1968, V.I. Oseledets formulated the question of convergence in the\nBirkhoff theorem and the multiplicative ergodic theorem for measurable cocycles\nover flows under the condition of integrability for each individual t. A.M.\nStepin and the author established (2016) the convergence along subsets of time\nof density 1. In this note, we show that moreover the convergence is fulfilled\nmodulo time subsets of finite measure.\n"}
{"abstract": "  We formulate both Markov chain Monte Carlo (MCMC) sampling algorithms and\nbasic statistical physics in terms of elementary symmetries. This perspective\non sampling yields derivations of well-known MCMC algorithms and a new parallel\nalgorithm that appears to converge more quickly than current state of the art\nmethods. The symmetry perspective also yields a parsimonious framework for\nstatistical physics and a practical approach to constructing meaningful notions\nof effective temperature and energy directly from time series data. We apply\nthese latter ideas to Anosov systems.\n"}
{"abstract": "  We consider a general (beyond $T\\bar T$) deformation of the 2D $O(N+1)$\n$\\sigma$-model by the irrelevant dimension-four operators. The theory deformed\nin this most general way is not integrable, and the $S$-matrix loses its\nfactorization properties. We perform the all-order summation of the leading\ninfrared logs for the $2 \\to 2$ scattering amplitude and provide the exact\nresult for the $2 \\to 2$ $S$-matrix in the leading logarithmic approximation.\nThese results can provide us with new insights into the properties of the\ntheories deformed by irrelevant operators more general than the $T\\bar T$\ndeformation.\n"}
{"abstract": "  Natal kicks and spins are characteristic properties of neutron stars (NSs)\nand black holes (BHs). Both offer valuable clues to dynamical processes during\nstellar core collapse and explosion. Moreover, they influence the evolution of\nstellar multiple systems and the gravitational-wave signals from their inspiral\nand merger. Observational evidence of possibly generic spin-kick alignment has\nbeen interpreted as indication that NS spins are either induced with the NS\nkicks or inherited from progenitor rotation, which thus might play a\ndynamically important role during stellar collapse. Current three-dimensional\nsupernova simulations suggest that NS kicks are transferred in the first\nseconds of the explosion, mainly by anisotropic mass ejection and, on a\nsecondary level, anisotropic neutrino emission. In contrast, the NS spins are\nonly determined minutes to hours later by angular momentum associated with\nfallback of matter that does not become gravitationally unbound in the\nsupernova. Here, we propose a novel scenario to explain spin-kick alignment as\na consequence of tangential vortex flows in the fallback matter that is\naccreted mostly from the direction of the NS's motion. For this effect the\ninitial NS kick is crucial, because it produces a growing offset of the NS away\nfrom the explosion center, thus promoting onesided accretion. In this new\nscenario conclusions based on traditional concepts are reversed. For example,\npre-kick NS spins are not required, and rapid progenitor-core rotation can\nhamper spin-kick alignment. We also discuss implications for natal BH kicks and\nthe possibility of tossing the BH's spin axis during its formation.\n"}
{"abstract": "  To reliably model real robot characteristics, interval linear systems of\nequations allow to describe families of problems that consider sets of values.\nThis allows to easily account for typical complexities such as sets of joint\nstates and design parameter uncertainties. Inner approximations of the\nsolutions to the interval linear systems can be used to describe the common\ncapabilities of a robotic manipulator corresponding to the considered sets of\nvalues. In this work, several classes of problems are considered. For each\nclass, reliable and efficient polytope, n-cube, and n-ball inner approximations\nare presented. The interval approaches usually proposed are inefficient because\nthey are too computationally heavy for certain applications, such as control.\nWe propose efficient new inner approximation theorems for the considered\nclasses of problems. This allows for usage with real-time applications as well\nas rapid analysis of potential designs. Several applications are presented for\na redundant planar manipulator including locally evaluating the manipulator's\nvelocity, acceleration, and static force capabilities, and evaluating its\nfuture acceleration capabilities over a given time horizon.\n"}
{"abstract": "  Increasing the batch size of a deep learning model is a challenging task.\nAlthough it might help in utilizing full available system memory during\ntraining phase of a model, it results in significant loss of test accuracy most\noften. LARS solved this issue by introducing an adaptive learning rate for each\nlayer of a deep learning model. However, there are doubts on how popular\ndistributed machine learning systems such as SystemML or MLlib will perform\nwith this optimizer. In this work, we apply LARS optimizer to a deep learning\nmodel implemented using SystemML.We perform experiments with various batch\nsizes and compare the performance of LARS optimizer with \\textit{Stochastic\nGradient Descent}. Our experimental results show that LARS optimizer performs\nsignificantly better than Stochastic Gradient Descent for large batch sizes\neven with the distributed machine learning framework, SystemML.\n"}
{"abstract": "  Multiview representation learning of data can help construct coherent and\ncontextualized users' representations on social media. This paper suggests a\njoint embedding model, incorporating users' social and textual information to\nlearn contextualized user representations used for understanding their\nlifestyle choices. We apply our model to tweets related to two lifestyle\nactivities, `Yoga' and `Keto diet' and use it to analyze users' activity type\nand motivation. We explain the data collection and annotation process in detail\nand provide an in-depth analysis of users from different classes based on their\nTwitter content. Our experiments show that our model results in performance\nimprovements in both domains.\n"}
{"abstract": "  Analyzing structural properties of social networks, such as identifying their\nclusters or finding their most central nodes, has many applications. However,\nthese applications are not supported by federated social networks that allow\nusers to store their social links locally on their end devices. In the\nfederated regime, users want access to personalized services while also keeping\ntheir social links private. In this paper, we take a step towards enabling\nanalytics on federated networks with differential privacy guarantees about\nprotecting the user links or contacts in the network. Specifically, we present\nthe first work to compute hierarchical cluster trees using local differential\nprivacy. Our algorithms for computing them are novel and come with theoretical\nbounds on the quality of the trees learned. The private hierarchical cluster\ntrees enable a service provider to query the community structure around a user\nat various granularities without the users having to share their raw contacts\nwith the provider. We demonstrate the utility of such queries by redesigning\nthe state-of-the-art social recommendation algorithms for the federated setup.\nOur recommendation algorithms significantly outperform the baselines which do\nnot use social contacts and are on par with the non-private algorithms that use\ncontacts.\n"}
{"abstract": "  We present a photometric and kinematical analysis of two and poorly studied\nopen clusters; Koposov 12 (FSR 802) and Koposov 43 (FSR 848) by using\ncross-matched data from PPMXL and Gaia DR2 catalog. We use astrometric\nparameters to identify 285 and 310 cluster members for Koposov 12 and Koposov\n43, respectively. Using the extracted member candidates and isochrone fitting\nto near-infrared (J, H, Ks) and Gaia DR2 bands (G, GBP, GRP), and Color\nMagnitude Diagrams (CMDs), we have estimated ages: log (age/yr) = 9.00 +/- 0.20\nand 9.50 +/- 0.20, and distances d = 1850 +/- 43 pc and 2500 +/- 50 pc for\nKoposov 12 and Koposov 43, respectively, assuming Solar metallicity (Z=0.019).\nThe estimated masses of the cluster derived using initial mass function and\nsynthetic CMD are 364 +/- 19 M_sun and 352 +/- 19 M_sun. We have also computed\ntheir velocity ellipsoid parameters based on (3x3) matrix elements (mu_ij).\n"}
{"abstract": "  We present here a novel cantilever based apparatus to perform translational\nstress or strain controlled rheology in very soft solids, and obtain\nsimultaneous confocal imaging of the 3 dimensional microstructure. The stress\nis measured using eddy based sensors. Both the stress and strain are controlled\nby applying PID control loops on measured quantities and changing position\nusing a micromanipulator. To get rid of surface tension forces, the sample and\ncantilever are immersed. This enables stress measurement and control down to\n3mPa. With this apparatus, we can independently apply shear and normal stress,\nor strain, with same precision. We demonstrate the technical capability of the\nsetup with steady shear strain or stress experiments on a soft protein gel\nsystem. The simultaneous confocal imaging offers insight into the macroscopic\nbreaking observed in an increasing shear strain experiment.\n"}
{"abstract": "  The massive machine-type communications (mMTC) paradigm based on media\nmodulation in conjunction with massive MIMO base stations (BSs) is emerging as\na viable solution to support the massive connectivity for the future\nInternet-of-Things, in which the inherent massive access at the BSs poses\nsignificant challenges for device activity and data detection (DADD). This\npaper considers the DADD problem for both uncoded and coded media modulation\nbased mMTC with a slotted access frame structure, where the device activity\nremains unchanged within one frame. Specifically, due to the slotted access\nframe structure and the adopted media modulated symbols, the access signals\nexhibit a doubly structured sparsity in both the time domain and the modulation\ndomain. Inspired by this, a doubly structured approximate message passing\n(DS-AMP) algorithm is proposed for reliable DADD in the uncoded case. Also, we\nderive the state evolution of the DS-AMP algorithm to theoretically\ncharacterize its performance. As for the coded case, we develop a\nbit-interleaved coded media modulation scheme and propose an iterative DS-AMP\n(IDS-AMP) algorithm based on successive inference cancellation (SIC), where the\nsignal components associated with the detected active devices are successively\nsubtracted to improve the data decoding performance. In addition, the channel\nestimation problem for media modulation based mMTC is discussed and an\nefficient data-aided channel state information (CSI) update strategy is\ndeveloped to reduce the training overhead in block fading channels. Finally,\nsimulation results and computational complexity analysis verify the superiority\nof the proposed algorithms in both uncoded and coded cases. Also, our results\nverify the validity of the proposed data-aided CSI update strategy.\n"}
{"abstract": "  In this work, we approach the minimization of a continuously differentiable\nconvex function under linear equality constraints by a second-order dynamical\nsystem with asymptotically vanishing damping term. The system is formulated in\nterms of the augmented Lagrangian associated to the minimization problem. We\nshow fast convergence of the primal-dual gap, the feasibility measure, and the\nobjective function value along the generated trajectories. In case the\nobjective function has Lipschitz continuous gradient, we show that the\nprimal-dual trajectory asymptotically weakly converges to a primal-dual optimal\nsolution of the underlying minimization problem. To the best of our knowledge,\nthis is the first result which guarantees the convergence of the trajectory\ngenerated by a primal-dual dynamical system with asymptotic vanishing damping.\nMoreover, we will rediscover in case of the unconstrained minimization of a\nconvex differentiable function with Lipschitz continuous gradient all\nconvergence statements obtained in the literature for Nesterov's accelerated\ngradient method.\n"}
{"abstract": "  Vector-boson scattering (VBS) processes probe the innermost structure of\nelectroweak interactions in the Standard Model, and provide a unique\nsensitivity for new physics phenomena affecting the gauge sector. In this\nreview, we report on the salient aspects of this class of processes, both from\nthe theory and experimental point of view. We start by discussing recent\nachievements relevant for their theoretical description, some of which have set\nimportant milestones in improving the precision and accuracy of the\ncorresponding simulations. We continue by covering the development of\nexperimental techniques aimed at detecting these rare processes and improving\nthe signal sensitivity over large backgrounds. We then summarise the details of\nthe most relevant VBS signatures and review the related measurements available\nto date, along with their comparison with Standard-Model predictions. We\nconclude by discussing the perspective at the upcoming Large Hadron Collider\nruns and at future hadron facilities.\n"}
{"abstract": "  We develop a self-consistent approach to study the spectral properties of a\nclass of quantum mechanical operators by using the knowledge about monodromies\nof $2\\times 2$ linear systems (Riemann-Hilbert correspondence). Our technique\napplies to a variety of problems, though in this paper we only analyse in\ndetail two examples. First we review the case of the (modified) Mathieu\noperator, which corresponds to a certain linear system on the sphere and makes\ncontact with the Painlev\\'e $\\mathrm{III}_3$ equation. Then we extend the\nanalysis to the 2-particle elliptic Calogero-Moser operator, which corresponds\nto a linear system on the torus. By using the Kiev formula for the\nisomonodromic tau functions, we obtain the spectrum of such operators in terms\nof self-dual Nekrasov functions ($\\epsilon_1+\\epsilon_2=0$). Through blowup\nrelations, we also find Nekrasov-Shatashvili type of quantizations\n($\\epsilon_2=0$).\n  In the case of the torus with one regular singularity we obtain certain\nresults which are interesting by themselves. Namely, we derive blowup equations\n(filling some gaps in the literature) and we relate them to the bilinear form\nof the isomonodromic deformation equations. In addition, we extract the\n$\\epsilon_2\\to 0$ limit of the blowup relations from the regularized action\nfunctional and CFT arguments.\n"}
{"abstract": "  Ultra-luminous infrared galaxies (ULIRGs) have infrared luminosities\n$L_{\\mathrm{IR}} \\geq 10^{12} L_{\\odot}$, making them the most luminous objects\nin the infrared sky. These dusty objects are generally powered by starbursts\nwith star-formation rates that exceed $100~ M_{\\odot}~ \\mathrm{yr}^{-1}$,\npossibly combined with a contribution from an active galactic nucleus. Such\nenvironments make ULIRGs plausible sources of astrophysical high-energy\nneutrinos, which can be observed by the IceCube Neutrino Observatory at the\nSouth Pole. We present a stacking search for high-energy neutrinos from a\nrepresentative sample of 75 ULIRGs with redshift $z \\leq 0.13$ using 7.5 years\nof IceCube data. The results are consistent with a background-only observation,\nyielding upper limits on the neutrino flux from these 75 ULIRGs. For an\nunbroken $E^{-2.5}$ power-law spectrum, we report an upper limit on the stacked\nflux $\\Phi_{\\nu_\\mu + \\bar{\\nu}_\\mu}^{90\\%} = 3.24 \\times 10^{-14}~\n\\mathrm{TeV^{-1}~ cm^{-2}~ s^{-1}}~ (E/10~ \\mathrm{TeV})^{-2.5}$ at 90%\nconfidence level. In addition, we constrain the contribution of the ULIRG\nsource population to the observed diffuse astrophysical neutrino flux as well\nas model predictions.\n"}
{"abstract": "  We have studied the Co layer thickness dependences of magnetocrystalline\nanisotropy (MCA), Dzyaloshinskii-Moriya interaction (DMI), and orbital moment\nanisotropy (OMA) in W/Co/Pt trilayers. We find the MCA favors magnetization\nalong the film normal and monotonically increases with decreasing effective\nmagnetic layer thickness ($t_\\mathrm{eff}$). The magnitude of the\nDzyaloshinskii-Moriya exchange constant ($|D|$) increases with decreasing\n$t_\\mathrm{eff}$ until $t_\\mathrm{eff} \\sim$1 nm, below which $|D|$ decreases.\nMCA and $|D|$ scale with $1/t_\\mathrm{eff}$ for $t_\\mathrm{eff}$ larger than\n$\\sim$1.0 nm, indicating an interfacial origin. To clarify the cause of the\n$t_\\mathrm{eff}$ dependences of MCA and DMI, the OMA of Co in W/Co/Pt trilayers\nis studied using x-ray magnetic circular dichroism (XMCD). We find non-zero OMA\nwhen $t_\\mathrm{eff}$ is smaller than $\\sim$0.8 nm. The OMA increases with\ndecreasing $t_\\mathrm{eff}$ at a rate that is larger than what is expected from\nthe MCA and Bruno's formula, indicating that other factors contribute to the\nMCA at small $t_\\mathrm{eff}$ to break the $1/t_\\mathrm{eff}$ scaling. The\n$t_\\mathrm{eff}$ dependence of the OMA also suggests that $|D|$ at\n$t_\\mathrm{eff}$ smaller than $\\sim$1 nm is independent of the OMA at the\ninterface. We consider the growth of Co on W results in a strain that reduces\nthe interfacial MCA and DMI at small $t_\\mathrm{eff}$, indicating the\nimportance of lattice structure to control their properties.\n"}
{"abstract": "  Curvature in form of the Hessian or its generalized Gauss-Newton (GGN)\napproximation is valuable for algorithms that rely on a local model for the\nloss to train, compress, or explain deep networks. Existing methods based on\nimplicit multiplication via automatic differentiation or Kronecker-factored\nblock diagonal approximations do not consider noise in the mini-batch. We\npresent ViViT, a curvature model that leverages the GGN's low-rank structure\nwithout further approximations. It allows for efficient computation of\neigenvalues, eigenvectors, as well as per-sample first- and second-order\ndirectional derivatives. The representation is computed in parallel with\ngradients in one backward pass and offers a fine-grained cost-accuracy\ntrade-off, which allows it to scale. As examples for ViViT's usefulness, we\ninvestigate the directional gradients and curvatures during training, and how\nnoise information can be used to improve the stability of second-order methods.\n"}
{"abstract": "  This paper is interested in independent sets (or equivalently, cliques) in\nuniform random cographs. We also study their permutation analogs, namely,\nincreasing subsequences in uniform random separable permutations.\n  First, we prove that, with high probability as $n$ gets large, the largest\nindependent set in a uniform random cograph with $n$ vertices has size $o(n)$.\nThis answers a question of Kang, McDiarmid, Reed and Scott. Using the\nconnection between graphs and permutations via inversion graphs, we also give a\nsimilar result for the longest increasing subsequence in separable\npermutations. These results are proved using the self-similarity of the\nBrownian limits of random cographs and random separable permutations, and\nactually apply more generally to all families of graphs and permutations with\nthe same limit.\n  Second, and unexpectedly given the above results, we show that for $\\beta >0$\nsufficiently small, the expected number of independent sets of size $\\beta n$\nin a uniform random cograph with $n$ vertices grows exponentially fast with\n$n$. We also prove a permutation analog of this result. This time the proofs\nrely on singularity analysis of the associated bivariate generating functions.\n"}
{"abstract": "  We investigate the formation of Free Trade Agreement (FTA) in a competing\nimporters framework with $n$ countries. We show that (i) FTA formation causes a\nnegative externality to non-participants, (ii) a non-participant is willing to\njoin an FTA, and (iii) new participation may decrease the welfare of incumbent\nparticipants. A unique subgame perfect equilibrium of a sequential FTA\nformation game does not achieve global free trade under an open-access rule\nwhere a new applicant needs consent of members for accession, currently\nemployed by many open regionalism agreements including APEC. We further show\nthat global FTA is a unique subgame perfect equilibrium under an open-access\nrule without consent.\n"}
{"abstract": "  This project discusses the control schemes for distribution grids with a\nlarge amount of wind penetration. Microgrids are constantly gaining popularity,\nespecially in the countries, where there is energy crisis. Various systems,\nincluding synchronous generators, grid and loads, have been investigated in\nthis project. Major focus is placed on active and reactive power sharing. Droop\ncontrol for multiple synchronous generators is explored. The phenomenon of load\ntransients has also been reviewed and associated simulations have been carried\nout on SimPower Systems. Constant wind power has been introduced and behaviour\nof the electrical system is observed. Behaviour of the system under variable\nwind has also been analysed. Moreover, recent development projects and previous\nworks, regarding microgrids and distributed generation, have been discussed.\n"}
{"abstract": "  Deep learning models can enable accurate and efficient disease diagnosis, but\nhave thus far been hampered by the data scarcity present in the medical world.\nAutomated diagnosis studies have been constrained by underpowered single-center\ndatasets, and although some results have shown promise, their generalizability\nto other institutions remains questionable as the data heterogeneity between\ninstitutions is not taken into account. By allowing models to be trained in a\ndistributed manner that preserves patients' privacy, federated learning\npromises to alleviate these issues, by enabling diligent multi-center studies.\nWe present the first federated learning study on the modality of cardiovascular\nmagnetic resonance (CMR) and use four centers derived from subsets of the M\\&M\nand ACDC datasets, focusing on the diagnosis of hypertrophic cardiomyopathy\n(HCM). We adapt a 3D-CNN network pretrained on action recognition and explore\ntwo different ways of incorporating shape prior information to the model, and\nfour different data augmentation set-ups, systematically analyzing their impact\non the different collaborative learning choices. We show that despite the small\nsize of data (180 subjects derived from four centers), the privacy preserving\nfederated learning achieves promising results that are competitive with\ntraditional centralized learning. We further find that federatively trained\nmodels exhibit increased robustness and are more sensitive to domain shift\neffects.\n"}
{"abstract": "  Recently, we have determined the spectrum and the wave functions of the\nHamiltonian of a Landau particle with time-dependent mass and frequency\nundergoing the influence of a uniform time-dependent electric field[J. Math.\nPhys. 56, 072104 (2018)]. In the present paper we extend the study of this\nmodel that we name the time-dependent Landau problem into the context of\ncoherent states. By means of the traditional factorization method of the\neigenfunctions of this system expressed in terms of the generalized Laguerre\npolynomials, we derive the generators of the su(1,1) Lie algebra and we\nconstruct the coherent states a la Barut-Girardello. These states are shown to\nsatisfy the Klauder's mathematical requirement to build coherent states and\nsome of their statistical properties are calculated and analyzed. We find that\nthese states are sub-Poissonian in nature. We show that, addition of photons\nfrom these coherent states, increases the statistical properties and changes\nthe mathematical properties of these states.\n"}
{"abstract": "  Many types of distributional word embeddings (weakly) encode linguistic\nregularities as directions (the difference between \"jump\" and \"jumped\" will be\nin a similar direction to that of \"walk\" and \"walked,\" and so on). Several\nattempts have been made to explain this fact. We respond to Allen and\nHospedales' recent (ICML, 2019) theoretical explanation, which claims that\nword2vec and GloVe will encode linguistic regularities whenever a specific\nrelation of paraphrase holds between the four words involved in the regularity.\nWe demonstrate that the explanation does not go through: the paraphrase\nrelations needed under this explanation do not hold empirically.\n"}
{"abstract": "  High-fidelity large-eddy simulations (LES) are performed to investigate\naeroacoustic characteristics of jets issuing from twin rectangular nozzles with\nan aspect ratio of 2:1 at two over-expanded conditions and the design\ncondition. For all three jet conditions simulated, LES predicts qualitatively\nsimilar near-field flow statistics to those measured at the University of\nCincinnati. Using the Ffowcs Williams-Hawkings method, LES captures the\nfundamental screech tone and its harmonics fairly well at multiple observer\nlocations in the far-field. Intense jet flapping motions in the near-field\nalong the minor axis, which are enhanced by jet-to-jet interactions, are found\nto correspond to those frequencies. Moreover, the overall sound pressure levels\nare within 1-2 dB difference from the experimental measurements. However, the\nscreech tones appear to be intermittent, as the twin-jet interaction pattern\nvaries irregularly. To extract dominant flow structures at the screech\nfrequencies and identify the twin-jet coupling modes, spectral proper\northogonal decomposition (SPOD) analysis is used. SPOD analysis recovers\nenergetic peaks at the screech frequencies, and the corresponding leading modes\nindicate strong upstream radiation originating from the fifth/sixth\nshock-cells. For the two over-expanded conditions, the leading modes show\nanti-symmetric coupling in the minor axis at the fundamental screech\nfrequencies. In contrast, the two jets behave symmetrically with respect to\neach other in the major axis, in line with the absence of jet flapping in this\ndirection. Furthermore, the leading SPOD eigenvalues turn out to be, at least,\ntwo orders of magnitude larger than higher-order eigenvalues, suggesting\npotential of reduced-order models for the twin-jet screech.\n"}
{"abstract": "  We research relations between optimal transport theory (OTT) and approximate\nBayesian computation (ABC) possibly connected to relevant metrics defined on\nprobability measures. Those of ABC are computational methods based on Bayesian\nstatistics and applicable to a given generative model to estimate its a\nposteriori distribution in case the likelihood function is intractable. The\nidea is therefore to simulate sets of synthetic data from the model with\nrespect to assigned parameters and, rather than comparing prospects of these\ndata with the corresponding observed values as typically ABC requires, to\nemploy just a distance between a chosen distribution associated to the\nsynthetic data and another of the observed values. Our focus lies in\ntheoretical and methodological aspects, although there would exist a remarkable\npart of algorithmic implementation, and more precisely issues regarding\nmathematical foundation and asymptotic properties are carefully analysed,\ninspired by an in-depth study of what is then our main bibliographic reference,\nthat is Bernton et al. (2019), carrying out what follows: a rigorous\nformulation of the set-up for the ABC rejection algorithm, also to regain a\ntransparent and general result of convergence as the ABC threshold goes to zero\nwhereas the number n of samples from the prior stays fixed; general technical\nproposals about distances leaning on OTT; weak assumptions which lead to lower\nbounds for small values of threshold and as n goes to infinity, ultimately\nshowing a reasonable possibility of lack of concentration which is contrary to\nwhat is proposed in Bernton et al. (2019) itself.\n"}
{"abstract": "  The objective of this paper is to explore how financial big data and machine\nlearning methods can be applied to model and understand complex financial\nproducts. We focus on residential mortgage backed securities, resMBS, that were\nat the heart of the 2008 US financial crisis. The securities are contained\nwithin a prospectus and have a complex payoff structure. Multiple financial\ninstitutions form a supply chain to create the prospectuses. We provide insight\ninto the performance of the resMBS securities through a series of increasingly\ncomplex models. First, models at the security level directly identify salient\nfeatures of resMBS securities that impact their performance. Second, we extend\nthe model to include prospectus level features. We are the first to demonstrate\nthat the composition of the prospectus is associated with the performance of\nsecurities. Finally, to develop a deeper understanding of the role of the\nsupply chain, we use unsupervised probabilistic methods, in particular, dynamic\ntopics models (DTM), to understand community formation and temporal evolution\nalong the chain. A comprehensive model provides insight into the impact of DTM\ncommunities on the issuance and evolution of prospectuses, and eventually the\nperformance of resMBS securities.\n"}
{"abstract": "  Phase-locked loops (PLLs) are now widely used in communication systems and\nhave been a classic system for more than 60 years. Well-known mathematical\nmodels of such systems are constructed in a number of approximations, so\nquestions about how they describe the experimental dynamics qualitatively and\nquantitatively, and how the accuracy of the model description depends on the\nbehavior mode, remain open. One of the most direct approaches to the\nverification of any model is its reconstruction from the time series obtained\nin experiment. If it is possible to fit the model to experimental data and the\nresulting parameter values are close to the expected values (calculated from\nthe first principles), the quantitative correspondence between the model and\nthe physical object is nearly proved. In this paper, for the first time, the\nequations of the PLL model with a bandpass filter are reconstructed from the\nexperimental signals of the generator in various modes. The reconstruction\nshowed that the model known in the literature generally describes the\nexperimental dynamics in regular and chaotic regimes. The relative error of\nparameter estimation is between 2% and 50% for different regimes and\nparameters. The reconstructed nonlinear function of phase is not harmonic and\nhighly asymmetric in contrary to the model one.\n"}
{"abstract": "  We report polarization properties for eight narrowband bursts from FRB 121102\nthat have been re-detected in a high-frequency (4-8 GHz) Breakthrough Listen\nobservation with the Green Bank Telescope, originally taken on 2017 August 26.\nThe bursts were found to exhibit nearly 100% linear polarization, Faraday\nrotation measures (RM) bordering 9.3$\\times$10$^4$ rad-m$^{-2}$, and stable\npolarization position angles (PA), all of which agree with burst properties\npreviously reported for FRB 121102 at the same epoch. We confirm that these\ndetections are indeed physical bursts with limited spectral occupancies and\nfurther support the use of sub-banded search techniques in FRB detection.\n"}
{"abstract": "  Rockfall detection is a crucial procedure in the field of geology, which\nhelps to reduce the associated risks. Currently, geologists identify rockfall\nevents almost manually utilizing point cloud and imagery data obtained from\ndifferent caption devices such as Terrestrial Laser Scanner or digital cameras.\nMulti-temporal comparison of the point clouds obtained with these techniques\nrequires a tedious visual inspection to identify rockfall events which implies\ninaccuracies that depend on several factors such as human expertise and the\nsensibility of the sensors. This paper addresses this issue and provides an\nintelligent framework for rockfall event detection for any individual working\nin the intersection of the geology domain and decision support systems. The\ndevelopment of such an analysis framework poses significant research challenges\nand justifies intensive experimental analysis. In particular, we propose an\nintelligent system that utilizes multiple machine learning algorithms to detect\nrockfall clusters of point cloud data. Due to the extremely imbalanced nature\nof the problem, a plethora of state-of-the-art resampling techniques\naccompanied by multiple models and feature selection procedures are being\ninvestigated. Various machine learning pipeline combinations have been\nbenchmarked and compared applying well-known metrics to be incorporated into\nour system. Specifically, we developed statistical and machine learning\ntechniques and applied them to analyze point cloud data extracted from\nTerrestrial Laser Scanner in two distinct case studies, involving different\ngeological contexts: the basaltic cliff of Castellfollit de la Roca and the\nconglomerate Montserrat Massif, both located in Spain. Our experimental data\nsuggest that some of the above-mentioned machine learning pipelines can be\nutilized to detect rockfall incidents on mountain walls, with experimentally\nproven accuracy.\n"}
{"abstract": "  Given a periodic quotient of a torsion-free hyperbolic group, we provide a\nfine lower estimate of the growth function of any sub-semi-group. This\ngeneralizes results of Razborov and Safin for free groups.\n"}
{"abstract": "  We prove an existence and uniqueness theorem for exact WKB solutions of\ngeneral singularly perturbed linear second-order ODEs in the complex domain.\nThese include the one-dimensional time-independent complex Schr\\\"odinger\nequation. Notably, our results are valid both in the case of generic WKB\ntrajectories as well as closed WKB trajectories. We also explain in what sense\nexact and formal WKB solutions form a basis. As a corollary of the proof, we\nestablish the Borel summability of formal WKB solutions for a large class of\nproblems, and derive an explicit formula for the Borel transform.\n"}
{"abstract": "  We give upper bounds for the bottom of the essential spectrum of properly\nimmersed minimal submanifolds of $\\mathbb{R}^{n}$ in terms of their volume\ngrowth. Our result improves the extrinsic version of Brook's essential spectrum\nestimate given by Ilias-Nelli-Soret in \\cite[Cor.3]{ins}.\n"}
{"abstract": "  Structured light are custom light fields where the phase, polarization, and\nintensity vary with position. It has been used for nanotweezers, nanoscale\nimaging, and quantum information technology, but its role in exciting optical\ntransitions in materials has been little examined so far. Here we use group\ntheory to derive the optical selection rules for nanosystems that get excited\nby structured light. If the size of the nanostructure is comparable to the\nlight wavelength, it will sample the full beam profile during excitation with\nprofound consequences on optical excitations. Using nanooligomers as model\nnanosystems, we show that structured light excites optical transitions that are\nforbidden for linearly polarized or unpolarized light. Such dipole forbidden\nmodes have longer lifetimes and narrower resonances than dipole allowed\ntransitions. We derive symmetry-adapted eigenmodes for nanooligomers containing\nup to six monomers. Our study includes tables with selection rules for\ncylindrical vector beams, for beams with orbital angular momentum, and for\nfield retardation along the propagation direction. We discuss multi-photon\nprocesses of nonlinear optics in addition to one-photon absorption. Structured\nlight will unlock a broad range of excitations in nanooligomers and other\nnanostructures that are currently inaccessible to optical studies.\n"}
{"abstract": "  In a recent work [B. Koc et al., arXiv:2010.03750, SIAM J. Numer. Anal., to\nappear], the authors showed that including difference quotients (DQs) is\nnecessary in order to prove optimal pointwise in time error bounds for proper\northogonal decomposition (POD) reduced order models of the heat equation. In\nthis work, we introduce a new approach to including DQs in the POD procedure.\nInstead of computing the POD modes using all of the snapshot data and DQs, we\nonly use the first snapshot along with all of the DQs and special POD weights.\nWe show that this approach retains all of the numerical analysis benefits of\nthe standard POD DQ approach, while using a POD data set that has half the\nnumber of snapshots as the standard POD DQ approach, i.e., the new approach is\nmore computationally efficient. We illustrate our theoretical results with\nnumerical experiments.\n"}
{"abstract": "  In this paper we construct short time classical solutions to a class of\nmaster equations in the presence of non-degenerate individual noise arising in\nthe theory of mean field games. The considered Hamiltonians are non-separable\nand $local$ functions of the measure variable, therefore the equation is\nrestricted to absolutely continuous measures whose densities lie in suitable\nSobolev spaces. Our results hold for smooth enough Hamiltonians, without any\nadditional structural conditions as convexity or monotonicity.\n"}
{"abstract": "  Inspection of ageing, reinforced concrete structures is a world-wide\nchallenge. Existing non-destructive evaluation techniques in civil and\nstructural engineering have limited penetration depth and don't allow to\nprecisely ascertain the configuration of reinforcement within large concrete\nobjects. The big challenge for critical infrastructure (bridges, dams, dry\ndocks, nuclear bioshields etc.) is understanding the internal condition of the\nconcrete and steel, not just the location of the reinforcement. In most new\nconstructions the location should be known and recorded in the as-built\ndrawings, where these might not exist due to poor record keeping for older\nstructures. Muon scattering tomography is a non-destructive and non-invasive\ntechnique which shows great promise for high-depth 3D concrete imaging.\nPreviously, we have demonstrated that individual bars with a diameter of 33.7\n+- 7.3 mm at 50 cm depth can be located using muon scattering tomography. Here\nwe present an improved method that exploits the periodicity of bar structures.\nWith this new method, reinforcement with bars down to 6 mm thickness can be\ndetected and imaged.\n"}
{"abstract": "  In this article, we deal with the global regularity of weak solutions to a\nclass of problems involving the fractional $(p,q)$-Laplacian, denoted by\n$(-\\Delta)^{s_1}_{p}+(-\\Delta)^{s_2}_{q}$, for $s_2, s_1\\in (0,1)$ and\n$1<p,q<\\infty$. We establish completely new H\\\"older continuity results, up to\nthe boundary, for the weak solutions to fractional $(p,q)$-problems involving\nsingular as well as regular nonlinearities. Moreover, as applications to\nboundary estimates, we establish new Hopf type maximum principle and strong\ncomparison principle in both situations.\n"}
{"abstract": "  Viewing optimization methods as numerical integrators for ordinary\ndifferential equations (ODEs) provides a thought-provoking modern framework for\nstudying accelerated first-order optimizers. In this literature, acceleration\nis often supposed to be linked to the quality of the integrator (accuracy,\nenergy preservation, symplecticity). In this work, we propose a novel ordinary\ndifferential equation that questions this connection: both the explicit and the\nsemi-implicit (a.k.a symplectic) Euler discretizations on this ODE lead to an\naccelerated algorithm for convex programming. Although semi-implicit methods\nare well-known in numerical analysis to enjoy many desirable features for the\nintegration of physical systems, our findings show that these properties do not\nnecessarily relate to acceleration.\n"}
{"abstract": "  Our work focuses on extra gradient learning algorithms for finding Nash\nequilibria in bilinear zero-sum games. The proposed method, which can be\nformally considered as a variant of Optimistic Mirror Descent\n\\cite{DBLP:conf/iclr/MertikopoulosLZ19}, uses a large learning rate for the\nintermediate gradient step which essentially leads to computing (approximate)\nbest response strategies against the profile of the previous iteration.\nAlthough counter-intuitive at first sight due to the irrationally large, for an\niterative algorithm, intermediate learning step, we prove that the method\nguarantees last-iterate convergence to an equilibrium. Particularly, we show\nthat the algorithm reaches first an $\\eta^{1/\\rho}$-approximate Nash\nequilibrium, with $\\rho > 1$, by decreasing the Kullback-Leibler divergence of\neach iterate by at least $\\Omega(\\eta^{1+\\frac{1}{\\rho}})$, for sufficiently\nsmall learning rate, $\\eta$, until the method becomes a contracting map, and\nconverges to the exact equilibrium. Furthermore, we perform experimental\ncomparisons with the optimistic variant of the multiplicative weights update\nmethod, by \\cite{Daskalakis2019LastIterateCZ} and show that our algorithm has\nsignificant practical potential since it offers substantial gains in terms of\naccelerated convergence.\n"}
{"abstract": "  Latency-insensitive design mitigates increasing interconnect delay and\nenables productive component reuse in complex digital systems. This design\nstyle has been adopted in high-level design flows because untimed functional\nblocks connected through latency-insensitive interfaces provide a natural\ncommunication abstraction. However, latency-insensitive design with high-level\nlanguages also introduces a unique set of verification challenges that\njeopardize functional correctness. In particular, bugs due to invalid\nconsumption of inputs and deadlocks can be difficult to detect and debug with\ndynamic simulation methods. To tackle these two classes of bugs, we propose\nformal model checking methods to guarantee that a high-level\nlatency-insensitive design is unaffected by invalid input data and is free of\ndeadlock. We develop a well-structured verification wrapper for each property\nto automatically construct the corresponding formal model for checking. Our\nexperiments demonstrate that the formal checks are effective in realistic bug\nscenarios from high-level designs.\n"}
{"abstract": "  In this short note we present several infinite dimensional theorems which\ngeneralize corresponding facts from the finite dimensional differential\ninclusions theory.\n"}
{"abstract": "  Curvilinear, multiblock summation-by-parts finite difference methods with the\nsimultaneous approximation term method provide a stable and accurate method for\nsolving the wave equation in second order form. That said, the standard method\ncan become arbitrarily stiff when characteristic boundary conditions and\nnonlinear interface conditions are used. Here we propose a new technique that\navoids this stiffness by using characteristic variables to \"upwind\" the\nboundary and interface treatment. This is done through the introduction of an\nadditional block boundary displacement variable. Using a unified energy, which\nexpresses both the standard as well as characteristic boundary and interface\ntreatment, we show that the resulting scheme has semidiscrete energy stability\nfor the anistropic wave equation. The theoretical stability results are\nconfirmed with numerical experiments that also demonstrate the accuracy and\nrobustness of the proposed scheme. The numerical results also show that the\ncharacteristic scheme has a time step restriction based on standard wave\npropagation considerations and not the boundary closure.\n"}
{"abstract": "  Life pattern clustering is essential for abstracting the groups'\ncharacteristics of daily mobility patterns and activity regularity. Based on\nmillions of GPS records, this paper proposed a framework on the life pattern\nclustering which can efficiently identify the groups have similar life pattern.\nThe proposed method can retain original features of individual life pattern\ndata without aggregation. Metagraph-based data structure is proposed for\npresenting the diverse life pattern. Spatial-temporal similarity includes\nsignificant places semantics, time sequential properties and frequency are\nintegrated into this data structure, which captures the uncertainty of an\nindividual and the diversities between individuals.\nNon-negative-factorization-based method was utilized for reducing the\ndimension. The results show that our proposed method can effectively identify\nthe groups have similar life pattern and takes advantages in computation\nefficiency and robustness comparing with the traditional method. We revealed\nthe representative life pattern groups and analyzed the group characteristics\nof human life patterns during different periods and different regions. We\nbelieve our work will help in future infrastructure planning, services\nimprovement and policies making related to urban and transportation, thus\npromoting a humanized and sustainable city.\n"}
{"abstract": "  Fanaroff-Riley (FR) 0 radio galaxies compose a new class of radio galaxies,\nwhich are usually weaker but much more numerous than the well-established class\nof FR 1 and FR 2 galaxies. The latter classes have been proposed as sources of\nthe ultra-high-energy cosmic rays (UHECRs) with energies reaching up to\n${\\sim}10^{20}$ eV. Based on this conjecture, the possibility of UHECR\nacceleration and survival in an FR 0 source environment is examined in this\nwork.\n  In doing so, an average spectral energy distribution (SED) based on data from\nthe FR 0 catalog (FR0CAT) is compiled. The resulting photon fields are used as\ntargets for UHECRs, which suffer from electromagnetic pair production,\nphoto-disintegration, photo-meson production losses, and synchrotron radiation.\nMultiple mechanisms are discussed to assess the UHECR acceleration probability,\nincluding Fermi-I order and gradual shear accelerations, and particle escape\nfrom the source region.\n  This work shows that in a hybrid scenario, combining Fermi and shear\naccelerations, FR 0 galaxies can contribute to the observed UHECR flux, as long\nas $\\Gamma_\\mathrm{j}\\gtrsim 1.6$, where shear acceleration starts to dominate\nover escape. Even in less optimistic scenarios, FR 0s can be expected to\ncontribute to the cosmic-ray flux between the knee and the ankle. Our results\nare relatively robust with respect to the realized magnetic turbulence model\nand the speed of the accelerating shocks.\n"}
{"abstract": "  In this analysis, we consider the effects of non-quiescent initial conditions\ndriven by pre-impact air-water interactions on the classical Wagner model of\nimpact theory. We consider the problem of a rigid, solid impactor moving\nvertically towards a liquid pool. Prior to impact, viscous forces in the air\nact to deform the liquid free surface, inducing a flow in the pool. These\ninteractions are then incorporated as initial conditions in the post-impact\nanalysis. We derive expressions for the size of the effective contact set, the\nleading-order pressure and force on the impactor, and the speed and thickness\nof the jet at its base. In all cases, we show that the effect of the pre-impact\nbehaviour is to cushion the impactor, reducing the size of the effective\ncontact set and, hence, the force on the impactor. Small- and large-time\nasymptotic solutions are derived for general power-law impactors, and we show\nthat the effects of the air die away as the impact progresses, so that we\napproach the classical Wagner solution.\n"}
{"abstract": "  In this paper, we establish a local regularity result for\n$W^{2,p}_{\\mathrm{loc}}$ solutions to complex degenerate nonlinear elliptic\nequations $F(D^2_{\\mathbb{C}} u)=f$ when they are comparable to the\nMonge-Amp\\`ere equation. Notably, we apply our result to the so-called\n$k$-Monge-Amp\\`ere equation.\n"}
{"abstract": "  In this paper, we develop a drift homotopy implicit particle filter method.\nThe methodology of our approach is to adopt the concept of drift homotopy in\nthe resampling procedure of the particle filter method for solving the\nnonlinear filtering problem, and we introduce an implicit particle filter\nmethod to improve the efficiency of the drift homotopy resampling procedure.\nNumerical experiments are carried out to demonstrate the effectiveness and\nefficiency of our drift homotopy implicit particle filter.\n"}
{"abstract": "  We present IMRPhenomTPHM, a phenomenological model for the gravitational wave\nsignals emitted by the coalescence of quasi-circular precessing binary black\nholes systems. The model is based on the \"twisting up\" approximation, which\nmaps non-precessing signals to precessing ones in terms of a time dependent\nrotation described by three Euler angles, and which has been utilized in\nseveral frequency domain waveform models that have become standard tools in\ngravitational wave data analysis. Our model is however constructed in the time\ndomain, which allows several improvements over the frequency domain models: we\ndo not use the stationary phase approximation, we employ a simple approximation\nfor the precessing Euler angles for the ringdown signal, and we implement a new\nmethod for computing the Euler angles through the evolution of the spin\ndynamics of the system, which is more accurate and also computationally\nefficient.\n"}
{"abstract": "  Semantic hashing represents documents as compact binary vectors (hash codes)\nand allows both efficient and effective similarity search in large-scale\ninformation retrieval. The state of the art has primarily focused on learning\nhash codes that improve similarity search effectiveness, while assuming a\nbrute-force linear scan strategy for searching over all the hash codes, even\nthough much faster alternatives exist. One such alternative is multi-index\nhashing, an approach that constructs a smaller candidate set to search over,\nwhich depending on the distribution of the hash codes can lead to sub-linear\nsearch time. In this work, we propose Multi-Index Semantic Hashing (MISH), an\nunsupervised hashing model that learns hash codes that are both effective and\nhighly efficient by being optimized for multi-index hashing. We derive novel\ntraining objectives, which enable to learn hash codes that reduce the candidate\nsets produced by multi-index hashing, while being end-to-end trainable. In\nfact, our proposed training objectives are model agnostic, i.e., not tied to\nhow the hash codes are generated specifically in MISH, and are straight-forward\nto include in existing and future semantic hashing models. We experimentally\ncompare MISH to state-of-the-art semantic hashing baselines in the task of\ndocument similarity search. We find that even though multi-index hashing also\nimproves the efficiency of the baselines compared to a linear scan, they are\nstill upwards of 33% slower than MISH, while MISH is still able to obtain\nstate-of-the-art effectiveness.\n"}
{"abstract": "  Does the Heisenberg uncertainty principle (HUP) apply along the time\ndimension in the same way it applies along the three space dimensions?\nRelativity says it should; current practice says no. With recent advances in\nmeasurement at the attosecond scale it is now possible to decide this question\nexperimentally.\n  The most direct test is to measure the time-of-arrival of a quantum particle:\nif the HUP applies in time, then the dispersion in the time-of-arrival will be\nmeasurably increased.\n  We develop an appropriate metric of time-of-arrival in the standard case;\nextend this to include the case where there is uncertainty in time; then\ncompare. There is -- as expected -- increased uncertainty in the\ntime-of-arrival if the HUP applies along the time axis. The results are fully\nconstrained by Lorentz covariance, therefore uniquely defined, therefore\nfalsifiable.\n  So we have an experimental question on our hands. Any definite resolution\nwould have significant implications with respect to the role of time in quantum\nmechanics and relativity. A positive result would also have significant\npractical applications in the areas of quantum communication, attosecond\nphysics (e.g. protein folding), and quantum computing.\n"}
{"abstract": "  We present a novel artificial cognitive mapping system using generative deep\nneural networks, called variational autoencoder/generative adversarial network\n(VAE/GAN), which can map input images to latent vectors and generate temporal\nsequences internally. The results show that the distance of the predicted image\nis reflected in the distance of the corresponding latent vector after training.\nThis indicates that the latent space is self-organized to reflect the proximity\nstructure of the dataset and may provide a mechanism through which many aspects\nof cognition are spatially represented. The present study allows the network to\ninternally generate temporal sequences that are analogous to the hippocampal\nreplay/pre-play ability, where VAE produces only near-accurate replays of past\nexperiences, but by introducing GANs, the generated sequences are coupled with\ninstability and novelty.\n"}
{"abstract": "  Bilevel optimization has recently attracted growing interests due to its wide\napplications in modern machine learning problems. Although recent studies have\ncharacterized the convergence rate for several such popular algorithms, it is\nstill unclear how much further these convergence rates can be improved. In this\npaper, we address this fundamental question from two perspectives. First, we\nprovide the first-known lower complexity bounds of\n$\\widetilde{\\Omega}(\\frac{1}{\\sqrt{\\mu_x}\\mu_y})$ and $\\widetilde\n\\Omega\\big(\\frac{1}{\\sqrt{\\epsilon}}\\min\\{\\frac{1}{\\mu_y},\\frac{1}{\\sqrt{\\epsilon^{3}}}\\}\\big)$\nrespectively for strongly-convex-strongly-convex and convex-strongly-convex\nbilevel optimizations. Second, we propose an accelerated bilevel optimizer\nnamed AccBiO, for which we provide the first-known complexity bounds without\nthe gradient boundedness assumption (which was made in existing analyses) under\nthe two aforementioned geometries. We also provide significantly tighter upper\nbounds than the existing complexity when the bounded gradient assumption does\nhold. We show that AccBiO achieves the optimal results (i.e., the upper and\nlower bounds match up to logarithmic factors) when the inner-level problem\ntakes a quadratic form with a constant-level condition number. Interestingly,\nour lower bounds under both geometries are larger than the corresponding\noptimal complexities of minimax optimization, establishing that bilevel\noptimization is provably more challenging than minimax optimization.\n"}
{"abstract": "  We present a novel analogue for finite exchangeable sequences of the de\nFinetti, Hewitt and Savage theorem and investigate its implications for\nmulti-marginal optimal transport (MMOT) and Bayesian statistics. If\n$(Z_1,...,Z_N)$ is a finitely exchangeable sequence of $N$ random variables\ntaking values in some Polish space $X$, we show that the law $\\mu_k$ of the\nfirst $k$ components has a representation of the form\n  $\\mu_k=\\int_{{\\mathcal P}_{\\frac{1}{N}}(X)} F_{N,k}(\\lambda) \\, \\mbox{d}\n\\alpha(\\lambda)$\n  for some probability measure $\\alpha$ on the set of $1/N$-quantized\nprobability measures on $X$ and certain universal polynomials $F_{N,k}$. The\nlatter consist of a leading term $N^{k-1}\\! /{\\small \\prod_{j=1}^{k-1}(N\\! -\\!\nj)\\, \\lambda^{\\otimes k}}$ and a finite, exponentially decaying series of\ncorrelated corrections of order $N^{-j}$ ($j=1,...,k$). The $F_{N,k}(\\lambda)$\nare precisely the extremal such laws, expressed via an explicit polynomial\nformula in terms of their one-point marginals $\\lambda$. Applications include\nnovel approximations of MMOT via polynomial convexification and the\nidentification of the remainder which is estimated in the celebrated error\nbound of Diaconis-Freedman between finite and infinite exchangeable laws.\n"}
{"abstract": "  The Aria project consists of a plant, hosting a 350 m cryogenic isotopic\ndistillation column, the tallest ever built, which is currently in the\ninstallation phase in a mine shaft at Carbosulcis S.p.A., Nuraxi-Figus (SU),\nItaly. Aria is one of the pillars of the argon dark-matter search experimental\nprogram, lead by the Global Argon Dark Matter Collaboration. Aria was designed\nto reduce the isotopic abundance of $^{39}$Ar, a $\\beta$-emitter of cosmogenic\norigin, whose activity poses background and pile-up concerns in the detectors,\nin the argon used for the dark-matter searches, the so-called Underground Argon\n(UAr). In this paper, we discuss the requirements, design, construction, tests,\nand projected performance of the plant for the isotopic cryogenic distillation\nof argon. We also present the successful results of isotopic cryogenic\ndistillation of nitrogen with a prototype plant, operating the column at total\nreflux.\n"}
{"abstract": "  The sharpest known high probability generalization bounds for uniformly\nstable algorithms (Feldman, Vondr\\'{a}k, 2018, 2019), (Bousquet, Klochkov,\nZhivotovskiy, 2020) contain a generally inevitable sampling error term of order\n$\\Theta(1/\\sqrt{n})$. When applied to excess risk bounds, this leads to\nsuboptimal results in several standard stochastic convex optimization problems.\nWe show that if the so-called Bernstein condition is satisfied, the term\n$\\Theta(1/\\sqrt{n})$ can be avoided, and high probability excess risk bounds of\norder up to $O(1/n)$ are possible via uniform stability. Using this result, we\nshow a high probability excess risk bound with the rate $O(\\log n/n)$ for\nstrongly convex and Lipschitz losses valid for \\emph{any} empirical risk\nminimization method. This resolves a question of Shalev-Shwartz, Shamir,\nSrebro, and Sridharan (2009). We discuss how $O(\\log n/n)$ high probability\nexcess risk bounds are possible for projected gradient descent in the case of\nstrongly convex and Lipschitz losses without the usual smoothness assumption.\n"}
{"abstract": "  This paper revisits two classical distributed problems in anonymous networks,\nnamely spanning tree construction and topology recognition, from the point of\nview of graph covering theory. For both problems, we characterize necessary and\nsufficient conditions on the communication graph in terms of directed symmetric\ncoverings. These characterizations answer along-standing open question posed by\nYamashita and Kameda [YK96], and shed new light on the connection between\ncoverings and the concepts of views and quotient graphs developed by the same\nauthors. Characterizing conditions in terms of coverings is significant because\nit connects the field with a vast body of classical literature in graph theory\nand algebraic topology. In particular, it gives access to powerful tools such\nas Reidemeister's theorem and Mazurkiewicz's algorithm. Combined together,\nthese tools allow us to present elegant proofs of otherwise intricate results,\nand their constructive nature makes them effectively usable in the algorithms.\nThis paper also gives us the opportunity to present the field of covering\ntheory in a pedagogical way, with a focus on the two aforementioned tools,\nwhose potential impact goes beyond the specific problems considered in this\nwork.\n"}
{"abstract": "  With the unprecedented shift towards automated urban environments in recent\nyears, a new paradigm is required to study pedestrian behaviour. Studying\npedestrian behaviour in futuristic scenarios requires modern data sources that\nconsider both the Automated Vehicle (AV) and pedestrian perspectives. Current\nopen datasets on AVs predominantly fail to account for the latter, as they do\nnot include an adequate number of events and associated details that involve\npedestrian and vehicle interactions. To address this issue, we propose using\nVirtual Reality (VR) data as a complementary resource to current datasets,\nwhich can be designed to measure pedestrian behaviour under specific\nconditions. In this research, we focus on the context-aware pedestrian\ntrajectory prediction framework for automated vehicles at mid-block\nunsignalized crossings. For this purpose, we develop a novel multi-input\nnetwork of Long Short-Term Memory (LSTM) and fully connected dense layers. In\naddition to past trajectories, the proposed framework incorporates pedestrian\nhead orientations and distance to the upcoming vehicles as sequential input\ndata. By merging the sequential data with contextual information of the\nenvironment, we train a model to predict the future pedestrian trajectory. Our\nresults show that the prediction error and overfitting to the training data are\nreduced by considering contextual information in the model. To analyze the\napplication of the methods to real AV data, the proposed framework is trained\nand applied to pedestrian trajectory extracted from an open-access video\ndataset. Finally, by implementing a game theory-based model interpretability\nmethod, we provide detailed insights and propose recommendations to improve the\ncurrent automated vehicle sensing systems from a pedestrian-oriented point of\nview.\n"}
{"abstract": "  The present paper deals with the generalization of Midpoint Ellipse Drawing\nAlgorithm (MPEDA) to minimize the error in the existing MPEDA in cartesian\nform. In this method, we consider three different values of h, i.e., 1, 0.5 and\n0.1. For h = 1, all the results of MPEDA have been verified. For other values\nof h it is observed that as the value of h decreases, the number of iteration\nincreases but the error between the points generated and the original ellipse\npoints decreases and vice-versa.\n"}
{"abstract": "  The solidification behavior of a eutectic AlCu specimen is investigated via\nin situ scanning transmission electron microscope (STEM) experiments.\nSolidification conditions are varied by imposing various cooling rates via a\nmicro-electro-mechanical system (MEMS) based membrane. The methodology allows\nthe use of material processed by a melting and casting route close to\nindustrial metallurgically fabricated material for in situ STEM solidification\nstudies. Several rapid solidification morphologies could be obtained solely on\na single specimen by the demonstrated strategy. A change from nanometer scaled\nlamellar to dendritic morphology is observed by differing cooling from medium\nto the highest tested rates. Cooling rates of 100, 10000 and 30000 K/s are\nused. For the comparatively slow cooling rate of 100 K/s, a coarser\nthree-dimensional morphology is gained. Additional post-solidification heat\ntreatments are investigated in terms of observation of spheroidization of\nlamellas during annealing at elevated temperatures.\n"}
{"abstract": "  Schizophrenia (SZ) is a mental disorder that typically emerges in late\nadolescence or early adulthood. It reduces the life expectancy of patients by\n15 years. Abnormal behavior, perception of emotions, social relationships, and\nreality perception are among its most significant symptoms. Past studies have\nrevealed the temporal and anterior lobes of hippocampus regions of brain get\naffected by SZ. Also, increased volume of cerebrospinal fluid (CSF) and\ndecreased volume of white and gray matter can be observed due to this disease.\nThe magnetic resonance imaging (MRI) is the popular neuroimaging technique used\nto explore structural/functional brain abnormalities in SZ disorder owing to\nits high spatial resolution. Various artificial intelligence (AI) techniques\nhave been employed with advanced image/signal processing methods to obtain\naccurate diagnosis of SZ. This paper presents a comprehensive overview of\nstudies conducted on automated diagnosis of SZ using MRI modalities. Main\nfindings, various challenges, and future works in developing the automated SZ\ndetection are described in this paper.\n"}
{"abstract": "  Nielsen's theorem states that any triangle can be inscribed in a planar\nJordan curve. We prove a generalisation of this theorem, extending to any\nJordan curve $J$ embedded in $\\mathbb{R}^{n}$, for a restricted set of\ntriangles. We then conclude by investigating a condition under which a given\npoint of $J$ inscribes an equilateral triangle in particular.\n"}
{"abstract": "  One-shot learning can be achieved by algorithms and animals, but how the\nlatter do it is poorly understood as most of the algorithms are not\nbiologically plausible. Experiments studying one-shot learning in rodents have\nshown that after initial gradual learning of associations between cues and\nlocations, new associations can be learned with just a single exposure to each\nnew cue-location pair. Foster, Morris and Dayan (2000) developed a hybrid\ntemporal difference - symbolic model that exhibited one-shot learning for dead\nreckoning to displaced single locations. While the temporal difference rule for\nlearning the agent's actual coordinates was biologically plausible, the model's\nsymbolic mechanism for learning target coordinates was not, and one-shot\nlearning for multiple target locations was not addressed. Here we extend the\nmodel by replacing the symbolic mechanism with a reservoir of recurrently\nconnected neurons resembling cortical microcircuitry. Biologically plausible\nlearning of target coordinates was achieved by subjecting the reservoir's\noutput weights to synaptic plasticity governed by a novel 4-factor variant of\nthe exploratory Hebbian (EH) rule. As with rodents, the reservoir model\nexhibited one-shot learning for multiple paired associations.\n"}
{"abstract": "  We formulate a Boolean algebra in the set of idempotents of Clifford algebra\nCl($R^{n,n}$) and within this frame we examine different formulations of the\nBoolean Satisfiability Problem in Clifford algebra. Exploiting the isomorphism\nbetween null subspaces of $R^{n,n}$ associated to simple spinors and the\northogonal group O(n) we ultimately give a continuous formulation of the\nBoolean Satisfiability Problem within this group that opens unexplored\nperspectives.\n"}
{"abstract": "  We consider the edge-triangle model, a two-parameter family of exponential\nrandom graphs in which dependence between edges is introduced through\ntriangles. In the so-called replica symmetric regime, the limiting free energy\nexists together with a complete characterization of the phase diagram of the\nmodel. We borrow tools from statistical mechanics to obtain limit theorems for\nthe edge density. First, we determine the asymptotic distribution of this\nquantity, as the graph size tends to infinity, in the various phases. Then we\nstudy the fluctuations of the edge density around its average value off the\ncritical curve and formulate conjectures about the behavior at criticality\nbased on the analysis of a mean-field approximation of the model. Some of our\nresults can be extended with no substantial changes to more general classes of\nexponential random graphs.\n"}
{"abstract": "  Objective: To compare non-contact acoustic micro-tapping optical coherence\nelastography (AuT-OCE) with destructive mechanical tests to confirm corneal\nelastic anisotropy.\n  Design: Ex vivo, laboratory study with non-contact AuT-OCE followed by\nmechanical rheometry and extensometry.\n  Subjects: Inflated cornea of whole-globe porcine eyes.\n  Methods: A non-contact transducer was used to launch mechanical waves in the\ncornea that were imaged with phase-sensitive OCT at physiologically relevant\npressures. Reconstruction of both Young's modulus (E) and out-of-plane shear\nmodulus (G) in the cornea from experimental data was performed using a model of\na nearly incompressible transversally isotropic (NITI) medium. Samples were\nthen excised and parallel plate rheometry was performed to measure the shear\nmodulus G. Corneal samples were then subjected to strip extensomety to measure\nthe Young's modulus.\n  Main Outcome Measures: Strong corneal anisotropy was confirmed with both\nAuT-OCE and mechanical tests, with the Young's and shear moduli differing by\nover an order of magnitude. These results show that AuT-OCE can quantify both\nmoduli with a non-contact, non-invasive, clinically translatable technique.\n"}
{"abstract": "  We present a novel mixed reality (MR) telepresence system enabling a local\nuser to interact with a remote user through full-body avatars in their own\nrooms. If the remote rooms have different sizes and furniture arrangements,\ndirectly applying a user's motion to an avatar leads to a mismatch of placement\nand deictic gesture. To overcome this problem, we retarget the placement, arm\ngesture, and head movement of a local user to an avatar in a remote room to\npreserve a local user's environment and interaction context. This allows\navatars to utilize real furniture and interact with a local user and shared\nobjects as if they were in the same room. This paper describes our system's\ndesign and implementation in detail and a set of example scenarios in the\nliving room and office room. A qualitative user study delves into a user\nexperience, challenges, and possible extensions of the proposed system.\n"}
{"abstract": "  New particles $\\phi$ in the MeV-GeV range produced at colliders and escaping\ndetection can be searched for at operating $b-$ and $\\tau-$factories such as\nBelle II. A typical search topology involves pair-produced $\\tau$s (or mesons),\none of which decaying to visibles plus the $\\phi$, and the other providing a\ntag. One crucial impediment of these searches is the limited ability to\nreconstruct the parents' separate boosts. This is the case in the 'typical'\ntopology where both decay branches include escaping particles. We observe that\nsuch topology lends itself to the use of kinematic variables such as $M_2$,\ndesigned for pairwise decays to visibles plus escaping particles, and endowed\nwith a built-in ('MAOS') way to efficiently guess the parents' separate boosts.\nStarting from this observation, we construct several kinematic quantities able\nto discriminate signal from background, and apply them to a benchmark search,\n$\\tau \\to e + \\phi$, where $\\phi$ can be either an axion-like particle or a\nhidden photon. Our considered variables can be applied to a wider range of\ntopologies than the current reference technique, based on the event thrust,\nwith which they are nearly uncorrelated. Application of our strategy leads to\nan improvement by a factor close to 3 in the branching-ratio upper limit for\n$\\tau \\to e \\phi$, with respect to the currently expected limit, assuming\n$m_\\phi \\lesssim 1$ MeV. For example, we anticipate a sensitivity of $1.7\n\\times 10^{-5}$ with the data collected before the 2022 shutdown.\n"}
{"abstract": "  We calculate the strength of the effective onsite Coulomb interaction\n(Hubbard $U$) in two-dimensional (2D) transition-metal (TM) dihalides MX$_2$\nand trihalides MX$_3$ (M=Ti, V, Cr, Mn, Fe, Co, Ni; X=Cl, Br, I) from first\nprinciples using the constrained random-phase approximation. The correlated\nsubspaces are formed from $t_{2g}$ or $e_g$ bands at the Fermi energy.\nElimination of the efficient screening taking place in these narrow bands gives\nrise to sizable interaction parameters U between the localized $t_{2g}$ ($e_g$)\nelectrons. Due to this large Coulomb interaction, we find $U/W >1$ (with the\nband width $W$) in most TM halides, making them strongly correlated materials.\nAmong the metallic TM halides in paramagnetic state, the correlation strength\n$U/W$ reaches a maximum in NiX$_2$ and CrX$_3$ with values much larger than the\ncorresponding values in elementary TMs and other TM compounds. Based on the\nStoner model and the calculated $U$ and $J$ values, we discuss the tendency of\nthe electron spins to order ferromagnetically.\n"}
{"abstract": "  The cold dark matter paradigm has been posited as the standard explanation\nfor the non-Keplerian behavior of galaxy rotation curves, where for galaxies\nsatisfying the Tully-Fisher relation, the mass of the dark matter halo from a\nlarge class of universal dark matter profiles ought to roughly increase\nlinearly with radial distance at large distances, $m(r) \\sim r/nG$ ($G$ is the\ngravitational constant and $n$ is a dimensionless parameter which depends on\nthe amount of baryonic matter $M$ within the galaxy). Despite numerous advances\nin modeling galaxy formation and evolution, a scientific consensus on the\norigin of the observed dependence of the dimensionless parameter $n =\n(GMa_{0})^{-1/2}$ on the mass of baryonic matter $M$ within the galaxy (the\nTully-Fisher relation), and the connection of the cosmological constant\n$\\Lambda$ to the parameter $a_{0} \\sim (\\Lambda/3)^{1/2}$ remains elusive.\nHere, we show that Einstein Field Equations can be remolded into\n$\\nabla_{\\nu}\\mathcal{K}^{\\nu}_{\\,\\,\\mu} = 8\\pi\nGM\\Psi^{*}\\mathcal{D}_{\\mu}\\Psi$, where $\\mathcal{K}_{\\mu\\nu}$ is a complex\nHermitian tensor, $\\mathcal{D}_{\\mu}$ is a covariant derivative and $\\Psi$ is a\ncomplex-valued function. This avails a novel constraint,\n$\\nabla_{\\mu}\\nabla_{\\nu}\\mathcal{K}^{\\mu\\nu} = 0$ not necessarily available in\nEinstein's General Relativity. In the weak-field regime, we can readily\nreproduce the Tully-Fisher relation using the usual charge-less pressure-less\nfluid. Moreover, our approach is equivalent to a Ginzburg-Landau theory of $n$\nbosons, where the order parameter is normalized as $\\int_{0}^{1/a_{0}} dr\\,4\\pi\nr^2\\Psi^*\\Psi = n$ and $1/a_{0} \\sim (\\Lambda/3)^{-1/2}$ is the cut-off length\nscale comparable to the size of the de Sitter universe. Our investigations\nprovide a framework that reproduces the mass-asymptotic speed relation in\ngalaxies within the cold dark matter paradigm.\n"}
{"abstract": "  Deep learning is a rapidly-evolving technology with possibility to\nsignificantly improve physics reach of collider experiments. In this study we\ndeveloped a novel algorithm of vertex finding for future lepton colliders such\nas the International Linear Collider. We deploy two networks; one is simple\nfully-connected layers to look for vertex seeds from track pairs, and the other\nis a customized Recurrent Neural Network with an attention mechanism and an\nencoder-decoder structure to associate tracks to the vertex seeds. The\nperformance of the vertex finder is compared with the standard ILC\nreconstruction algorithm.\n"}
{"abstract": "  The Extragalactic Distance Database (EDD) was created as a repository for\nhigh quality, redshift-independent distances. A key component of EDD is the\nColor Magnitude Diagrams/Tip of the Red Giant Branch (CMDs/TRGB) catalog, which\nprovides information on the stellar content of nearby galaxies observed with\nthe Hubble Space Telescope (HST). Here we provide a decadal update to this\ncatalog, which has now doubled in size to over 500 galaxies. We highlight the\nadditions to our data reduction and analysis techniques, and provide examples\nof the science that has been made possible with this large data set. We find\nthe TRGB to be a reliable measure for distance, and we aim to extend its\ndistance coverage with HST to every galaxy within 10 Mpc. In the near-future,\nthe combination of the James Webb Space Telescope and the Nancy Grace Roman\nSpace Telescope will dramatically increase the number of targets within our\ngrasp.\n"}
{"abstract": "  The peculiar morphology of Head-Tail (HT) radio galaxies indicates strong\ninteractions between the radio jets and their intra-cluster medium. We\nsystematically search for HT radio galaxies from LOFAR Two-metre Sky Survey\nfirst data release (LoTSS DR1) at 144 MHz frequency. We present here a\ncatalogue of fifty new HT radio sources, among them, five are Narrow-Angle\nTailed sources (NATs) and forty-five are Wide Angle Tailed sources (WATs). NATs\nare characterized by tails bent in a narrow V like shape with less than a\nninety-degree opening angle. For WAT radio galaxies, the opening angle between\njets is more than ninety degrees which exhibit wide C like morphologies. We\nfound that thirty-one out of fifty HT sources are associated with known galaxy\nclusters. The various physical properties and statistical studies of these HT\nsources are also presented in this paper.\n"}
{"abstract": "  In this paper, we propose an end-to-end lifelong learning mixture of experts.\nEach expert is implemented by a Variational Autoencoder (VAE). The experts in\nthe mixture system are jointly trained by maximizing a mixture of individual\ncomponent evidence lower bounds (MELBO) on the log-likelihood of the given\ntraining samples. The mixing coefficients in the mixture, control the\ncontributions of each expert in the goal representation. These are sampled from\na Dirichlet distribution whose parameters are determined through non-parametric\nestimation during lifelong learning. The model can learn new tasks fast when\nthese are similar to those previously learnt. The proposed Lifelong mixture of\nVAE (L-MVAE) expands its architecture with new components when learning a\ncompletely new task. After the training, our model can automatically determine\nthe relevant expert to be used when fed with new data samples. This mechanism\nbenefits both the memory efficiency and the required computational cost as only\none expert is used during the inference. The L-MVAE inference model is able to\nperform interpolation in the joint latent space across the data domains\nassociated with different tasks and is shown to be efficient for disentangled\nlearning representation.\n"}
{"abstract": "  Let $A$ be a Noetherian flat $K[t]$-algebra, $h$ an integer and let $N$ be a\ngraded $K[t]$-module, we introduce and study \"$N$-fiber-full up to $h$\"\n$A$-modules. We prove that an $A$-module $M$ is $N$-fiber-full up to $h$ if and\nonly if $\\mathrm{Ext}^i_A(M, N)$ is flat over $K[t]$ for all $i\\le h-1$. And we\nshow some applications of this result extending the recent result on squarefree\nGr\\\"obner degenerations by Conca and Varbaro.\n"}
{"abstract": "  Considered in this paper is the generalized Camassa-Holm-Novikov equation\nwith high order nonlinearity, which unifies the Camassa-Holm and Novikov\nequations as special cases. We show that the solution map of generalized\nCamassa-Holm-Novikov equation is not uniformly continuous on the initial data\nin Besov spaces $B_{p, r}^s(\\mathbb{R})$ with $s>\\max\\{1+\\frac{1}{p},\n\\frac{3}{2}\\}$, $1\\leq p, r< \\infty$ as well as in critical space $B_{2,\n1}^{\\frac{3}{2}}(\\mathbb{R}).$ Our result covers and improves the previous work\ngiven by Li et al. \\cite{Li 2020, 1Li 2020, Li 2021}(J. Differ. Equ. 269 (2020)\n8686-8700; J. Math. Fluid Mech. 22 (2020) 4:50; J. Math. Fluid Mech., (2021)\n23:36).\n"}
{"abstract": "  We study phase-imprinted solitons of ultracold bosons in an optical lattice\nwith a harmonic trap, which shows the superfluid (SF) and Mott-insulator (MI)\nshell structures. The earlier study [Konstantin V. Krutitsky, J. Larson, and M.\nLewenstein, Phys. Rev. A 82, 033618 (2010).] reported three types of\nphase-imprinted solitons in the Bose-Hubbard model: in-phase soliton,\nout-of-phase soliton, and wavelet. In this paper, we uncover the dynamical\nphase diagram of these phase-imprinted solitons, and find another type of the\nphase-imprinted soliton namely, the hybrid soliton. In the harmonically trapped\nsystem, the solitonic excitations created at the SF core cannot penetrate into\nthe outer SF shell. This repulsion at the surface of the outer SF shell can be\ncured by inposing a repulsive potential at the center of the trap. These\nresults can be interpreted as a kind of the impedance matching of excitations\nin BECs in terms of the effective chemical potentials or the local particle\nnumbers in the shell, and the analogous results can be observed also in the\nsound wave created by the local single-shot pulse potential.\n"}
{"abstract": "  We introduce a novel class of quantum circuits that are unitary along three\ndistinct \"arrows of time\". These dynamics share some of the analytical\ntractability of \"dual-unitary\" circuits, while exhibiting distinctive and\nricher phenomenology. We find that two-point correlations in these dynamics are\nstrictly confined to three directions in $(1+1)$-dimensional spacetime -- the\ntwo light cone edges, $\\delta x=\\pm v\\delta t$, and the static worldline\n$\\delta x=0$. Along these directions, correlation functions are obtained\nexactly in terms of quantum channels built from the individual gates that make\nup the circuit. We prove that, for a class of initial states, entanglement\ngrows at the maximum allowed speed up to an entropy density of at least one\nhalf of the thermal value, at which point it becomes model-dependent. Finally,\nwe extend our circuit construction to $2+1$ dimensions, where two-point\ncorrelation functions are confined to the one-dimensional edges of a\ntetrahedral light cone -- a subdimensional propagation of information\nreminiscent of \"fractonic\" physics.\n"}
{"abstract": "  Early prediction of cerebral palsy is essential as it leads to early\ntreatment and monitoring. Deep learning has shown promising results in\nbiomedical engineering thanks to its capacity of modelling complicated data\nwith its non-linear architecture. However, due to their complex structure, deep\nlearning models are generally not interpretable by humans, making it difficult\nfor clinicians to rely on the findings. In this paper, we propose a channel\nattention module for deep learning models to predict cerebral palsy from\ninfants' body movements, which highlights the key features (i.e. body joints)\nthe model identifies as important, thereby indicating why certain diagnostic\nresults are found. To highlight the capacity of the deep network in modelling\ninput features, we utilize raw joint positions instead of hand-crafted\nfeatures. We validate our system with a real-world infant movement dataset. Our\nproposed channel attention module enables the visualization of the vital joints\nto this disease that the network considers. Our system achieves 91.67%\naccuracy, suppressing other state-of-the-art deep learning methods.\n"}
{"abstract": "  Spiking neural networks (SNNs) have emerged as energy-efficient neural\nnetworks with temporal information. SNNs have shown a superior efficiency on\nneuromorphic devices, but the devices are susceptible to noise, which hinders\nthem from being applied in real-world applications. Several studies have\nincreased noise robustness, but most of them considered neither deep SNNs nor\ntemporal information. In this paper, we investigate the effect of noise on deep\nSNNs with various neural coding methods and present a noise-robust deep SNN\nwith temporal information. With the proposed methods, we have achieved a deep\nSNN that is efficient and robust to spike deletion and jitter.\n"}
{"abstract": "  We propose and study the single-frame anisoplanatic deconvolution problem\nassociated with image classification using machine learning algorithms, named\nthe nonuniform defocus removal (NDR) problem. Mathematical analysis of the NDR\nproblem is done and the so-called defocus removal (DR) algorithm for solving it\nis proposed. Global convergence of the DR algorithm is established without\nimposing any unverifiable assumption. Numerical results on simulation data show\nsignificant features of DR including solvability, noise robustness,\nconvergence, model insensitivity and computational efficiency. Physical\nrelevance of the NDR problem and practicability of the DR algorithm are tested\non experimental data. Back to the application that originally motivated the\ninvestigation of the NDR problem, we show that the DR algorithm can improve the\naccuracy of classifying distorted images using convolutional neural networks.\nThe key difference of this paper compared to most existing works on\nsingle-frame anisoplanatic deconvolution is that the new method does not\nrequire the data image to be decomposable into isoplanatic subregions.\nTherefore, solution approaches partitioning the image into isoplanatic zones\nare not applicable to the NDR problem and those handling the entire image such\nas the DR algorithm need to be developed and analyzed.\n"}
{"abstract": "  The study investigates the relationship between bank profitability and a\ncomprehensive list of bank specific, industry specific and macroeconomic\nvariables using unique panel data from 23 Bangladeshi banks with large market\nshares from 2005 to 2019 employing the Pooled Ordinary Least Square (POLS)\nMethod for regression estimation. The random Effect model has been used to\ncheck for robustness. Three variables, namely, Return on Asset (ROA), Return on\nEquity (ROE), and Net Interest Margin (NIM), have been used as profitability\nproxies. Non-interest income, capital ratio, and GDP growth have been found to\nhave a significant relationship with ROA. In addition to non-interest income,\nmarket share, bank size, and real exchange rates are significant explaining\nvariables if profitability is measured as NIM. The only significant determinant\nof profitability measured by ROE is market share. The primary contribution of\nthis study to the existing knowledge base is an extensive empirical analysis by\ncovering the entire gamut of independent variables (bank specific, industry\nrelated, and macroeconomic) to explain the profitability of the banks in\nBangladesh. It also covers an extensive and recent data set. Banking sector\nstakeholders may find great value from the outputs of this paper. Regulators\nand policymakers may find this useful in undertaking analyses in setting policy\nrates, banking industry stability, and impact assessment of critical policy\nmeasures before and after the enactment, etc. Investors and the bank management\nare to use the findings of this paper in analyzing the real drivers of\nprofitability of the banks they are contemplating to invest and managing on a\ndaily basis.\n"}
{"abstract": "  Proper identification of grade levels of children's reading materials is an\nimportant step towards effective learning. Recent studies in readability\nassessment for the English domain applied modern approaches in natural language\nprocessing (NLP) such as machine learning (ML) techniques to automate the\nprocess. There is also a need to extract the correct linguistic features when\nmodeling readability formulas. In the context of the Filipino language, limited\nwork has been done [1, 2], especially in considering the language's lexical\ncomplexity as main features. In this paper, we explore the use of lexical\nfeatures towards improving the development of readability identification of\nchildren's books written in Filipino. Results show that combining lexical\nfeatures (LEX) consisting of type-token ratio, lexical density, lexical\nvariation, foreign word count with traditional features (TRAD) used by previous\nworks such as sentence length, average syllable length, polysyllabic words,\nword, sentence, and phrase counts increased the performance of readability\nmodels by almost a 5% margin (from 42% to 47.2%). Further analysis and ranking\nof the most important features were shown to identify which features contribute\nthe most in terms of reading complexity.\n"}
{"abstract": "  In this paper we tackle the problem of pose guided person image generation,\nwhich aims to transfer a person image from the source pose to a novel target\npose while maintaining the source appearance. Given the inefficiency of\nstandard CNNs in handling large spatial transformation, we propose a\nstructure-aware flow based method for high-quality person image generation.\nSpecifically, instead of learning the complex overall pose changes of human\nbody, we decompose the human body into different semantic parts (e.g., head,\ntorso, and legs) and apply different networks to predict the flow fields for\nthese parts separately. Moreover, we carefully design the network modules to\neffectively capture the local and global semantic correlations of features\nwithin and among the human parts respectively. Extensive experimental results\nshow that our method can generate high-quality results under large pose\ndiscrepancy and outperforms state-of-the-art methods in both qualitative and\nquantitative comparisons.\n"}
{"abstract": "  On a closed manifold, consider the space of all Riemannian metrics for which\n-\\Delta + kR is positive (nonnegative) definite, where k > 0 and R is the\nscalar curvature. This spectral generalization of positive (nonnegative) scalar\ncurvature arises naturally for different values of k in the study of scalar\ncurvature in dimension via minimal hypersurfaces, the Yamabe problem, and\nPerelman's Ricci flow with surgery. When k=1/2, the space models apparent\nhorizons in time-symmetric initial data to the Einstein equations. We study\nthese spaces in unison and generalize Coda Marques's path-connectedness\ntheorem. Applying this with k=1/2, we compute the Bartnik mass of 3-dimensional\napparent horizons and the Bartnik--Bray mass of their outer-minimizing\ngeneralizations in all dimensions. Our methods also yield efficient\nconstructions for the scalar-nonnegative fill-in problem.\n"}
{"abstract": "  Accelerated particles are ubiquitous in the Cosmos and play a fundamental\nrole in many processes governing the evolution of the Universe at all scales,\nfrom the sub-AU ones relevant for the formation and evolution of stars and\nplanets to the Mpc ones involved in Galaxy assembly. We reveal the presence of\nenergetic particles in many classes of astrophysical sources thanks to their\nproduction of non-thermal radiation, and we detect them directly at Earth as\nCosmic Rays. In the last two decades both direct and indirect observations have\nprovided us a wealth of new, high quality data about Cosmic Rays and their\ninteractions both in sources and during propagation, in the Galaxy and in the\nsolar system. Some of the new data have confirmed existing theories about\nparticle acceleration and propagation and their interplay with the environment\nin which they occur. Some others have brought about interesting surprises,\nwhose interpretation is not straightforward within the standard framework and\nmay require a change of paradigm in terms of our ideas about the origin of\ncosmic rays of different species or in different energy ranges. In this\narticle, we will focus on Cosmic Rays of galactic origin, namely with energies\nbelow a few PeV, where a steepening is observed in the spectrum of energetic\nparticles detected at the Earth. We review the recent observational findings\nand the current status of the theory about the origin and propagation of\nGalactic Cosmic Rays.\n"}
{"abstract": "  Rare earth intermetallic compounds have been fascinating scientists due to\nrich phenomena induced by the interplay between localized $f$-orbitals and\nconduction electrons. However, since the energy scale of the\ncrystal-electric-field (CEF) splitting, which defines $f$-orbitals, is very\nsmall only in a few meV, the nature of mobile electrons accompanied by\nCEF-excitations has not been unveiled so far. It thus leaves these systems as\nfrontiers for discovering exotic quasiparticles not yet captured in condensed\nmatter physics. Here, we examined very low-energy electronic structures of CeSb\ngoing through the anomalous magnetostructural transitions below the N{\\'{e}}el\ntemperature ($T_{\\rm{N}}$) $\\sim$17 K, called devil's staircase, by a\ncombination of laser angle-resolved photoemission, Raman and neutron scattering\nspectroscopies. We found a new type of electron-boson coupling between the\nmobile electrons and quadrupole CEF-excitations of the 4$f$-orbitals, which\nrenormalizes the Sb 5$p$ band prominently, yielding a remarkable kink at very\nlow-energy ($\\sim$7 meV). This coupling strength is exceedingly strong and\nexhibits anomalous step-like enhancement during the devil's staircase\ntransition, unveiling a new type of quasiparticle, named multipole polaron,\nthat is a mobile electron largely dressed with a cloud of the quadrupole\nCEF-polarization.\n"}
{"abstract": "  As algorithms are increasingly applied to screen applicants for high-stakes\ndecisions in employment, lending, and other domains, concerns have been raised\nabout the effects of algorithmic monoculture, in which many decision-makers all\nrely on the same algorithm. This concern invokes analogies to agriculture,\nwhere a monocultural system runs the risk of severe harm from unexpected\nshocks. Here we show that the dangers of algorithmic monoculture run much\ndeeper, in that monocultural convergence on a single algorithm by a group of\ndecision-making agents, even when the algorithm is more accurate for any one\nagent in isolation, can reduce the overall quality of the decisions being made\nby the full collection of agents. Unexpected shocks are therefore not needed to\nexpose the risks of monoculture; it can hurt accuracy even under \"normal\"\noperations, and even for algorithms that are more accurate when used by only a\nsingle decision-maker. Our results rely on minimal assumptions, and involve the\ndevelopment of a probabilistic framework for analyzing systems that use\nmultiple noisy estimates of a set of alternatives.\n"}
{"abstract": "  Reverb plays a critical role in music production, where it provides listeners\nwith spatial realization, timbre, and texture of the music. Yet, it is\nchallenging to reproduce the musical reverb of a reference music track even by\nskilled engineers. In response, we propose an end-to-end system capable of\nswitching the musical reverb factor of two different mixed vocal tracks. This\nmethod enables us to apply the reverb of the reference track to the source\ntrack to which the effect is desired. Further, our model can perform\nde-reverberation when the reference track is used as a dry vocal source. The\nproposed model is trained in combination with an adversarial objective, which\nmakes it possible to handle high-resolution audio samples. The perceptual\nevaluation confirmed that the proposed model can convert the reverb factor with\nthe preferred rate of 64.8%. To the best of our knowledge, this is the first\nattempt to apply deep neural networks to converting music reverb of vocal\ntracks.\n"}
{"abstract": "  An MRI-only adaptive radiotherapy (ART) workflow is desirable for managing\ninterfractional changes in anatomy, but producing synthetic CT (sCT) data\nthrough paired data-driven deep learning (DL) for abdominal dose calculations\nremains a challenge due to the highly variable presence of intestinal gas. We\npresent the preliminary dosimetric evaluation of our novel approach to sCT\nreconstruction that is well suited to handling intestinal gas in abdominal\nMRI-only ART.\n  We utilize a paired data DL approach enabled by the intensity projection\nprior, in which well-matching training pairs are created by propagating air\nfrom MRI to corresponding CT scans. Evaluations focus on two classes: patients\nwith (1) little involvement of intestinal gas, and (2) notable differences in\nintestinal gas presence between corresponding scans. Comparisons between\nsCT-based plans and CT-based clinical plans for both classes are made at the\nfirst treatment fraction to highlight the dosimetric impact of the variable\npresence of intestinal gas.\n  Class 1 patients ($n=13$) demonstrate differences in prescribed dose coverage\nof the PTV of $1.3 \\pm 2.1\\%$ between clinical plans and sCT-based plans. Mean\nDVH differences in all structures for Class 1 patients are found to be\nstatistically insignificant. In Class 2 ($n=20$), target coverage is $13.3 \\pm\n11.0\\%$ higher in the clinical plans and mean DVH differences are found to be\nstatistically significant.\n  Significant deviations in calculated doses arising from the variable presence\nof intestinal gas in corresponding CT and MRI scans may limit the effectiveness\nof adaptive dose escalation efforts. We have proposed a paired data-driven DL\napproach to sCT reconstruction for accurate dose calculations in abdominal ART\nenabled by the creation of a clinically unavailable training data set with\nwell-matching representations of intestinal gas.\n"}
{"abstract": "  This paper computes the Fadell-Husseini index of Stiefel manifolds in the\ncase where the group acts via permutations of the orthogonal vectors. The\ncomputations are carried out in the case of elementary Abelian $p$-groups. The\nresults are shown to imply certain generalizations of the\nKakutani-Yamabe-Yujobo theorem.\n"}
{"abstract": "  We are interested in the dynamics of a population structured by a phenotypic\ntrait. Individuals reproduce sexually, which is represented by a non-linear\nintegral operator. This operator is combined to a multiplicative operator\nrepresenting selection. When the strength of selection is small, we show that\nthe dynamics of the population is governed by a simple macroscopic differential\nequation, and that solutions converge exponentially to steady-states that are\nlocally unique. The analysis is based on Wasserstein distance inequalities\nusing a uniform lower bound on distributions. These inequalities are coupled to\ntail estimates to show the stability of the steady-states.\n"}
{"abstract": "  Context: Many if not most planetary nebulae (PNe) are now thought to be the\noutcome of binary evolutionary scenarios. However only a few percent of PNe in\nthe Milky Way are known to host binary systems. The high precision repeated\nobserving and long time baseline of Gaia make it well suited to detect new\nclose binaries through photometric variability.\n  Aims: We aim to find new close binary central stars of PNe (CSPNe) using data\nfrom the Gaia mission, building towards a statistically significant sample of\npost common envelope, close binary CSPNe.\n  Methods: As the vast majority of Gaia sources do not have published epoch\nphotometry, we use the uncertainty in the mean photometry as a proxy for\ndetermining the variability of our CSPN sample in the second Gaia data release.\nWe derive a quantity that expresses the significance of the variability, and\nconsider what is necessary to build a clean sample of genuine variable sources.\n  Results: Our selection recovers a large fraction of the known close binary\nCSPN population, while other CSPNe lying in the same region of the parameter\nspace likely represent low-hanging fruit for ground-based confirmatory followup\nobservations. Gaia epoch photometry for four of the newly identified variable\nsources confirms that the variability is genuine and consistent with binarity\n"}
{"abstract": "  Since the past few decades, several designs and control methods have been\ndeveloped for the Spherical Robots (SRs) with thoroughly analyzed mechanics on\ngeneralized 3D terrains. But the vertical motion and the steep inclination\nmaneuver has been an unsolved problem with the existing SR's driving\nmechanisms. Also, the possibilities of wind-powered or air-propelled SRs have\nnot been fully explored. This paper introduces the new Omnidirectional\nSpherical Robot mechanism named SPIROS: The Six Propeller and Intermeshed Rotor\nbased Omnidirectional Spherical Robot. The SPIROS is driven by a novel\noctahedral arrangement of six intermeshed rotary air thrusters placed in the\nGoldberg Polyhedral shaped spherical gridshell. The advantage of the proposed\ndesign lies in its air-powered propulsion, which improves on the obstacle\navoidance and the slope climbing abilities. The robot's dynamic models are\nderived using the existing kinematical models of the Continuous Rolling\nSpherical Robots (CR-SR), with subcategories, triple axes rolling (3R-SR), dual\naxes rolling (2R-SR) and rolling and turning (RT-SR) spherical robots. The path\ntracking control scheme based on the pure pursuit algorithm is presented.\nSimulations are carried out in MATLAB and Simulink to validate the developed\nmodels and the effectiveness of the proposed control schemes.\n"}
{"abstract": "  ECH capacities are rich obstructions to symplectic embeddings in 4-dimensions\nthat have also been seen to arise in the context of algebraic positivity for\n(possibly singular) projective surfaces. We extend this connection to relate\ngeneral convex toric domains on the symplectic side with towers of polarised\ntoric surfaces on the algebraic side, and then use this perspective to show\nthat the sub-leading asymptotics of ECH capacities for all convex and concave\ntoric domains are $O(1)$. We obtain sufficient criteria for when the\nsub-leading asymptotics converge in this context, generalising results of\nHutchings and of the author, and derive new obstructions to embeddings between\ntoric domains of the same volume. We also propose two invariants to more\nprecisely describe when convergence occurs in the toric case. Our methods are\nlargely non-toric in nature, and apply more widely to towers of polarised\nLooijenga pairs.\n"}
{"abstract": "  Multi-modal MRIs are widely used in neuroimaging applications since different\nMR sequences provide complementary information about brain structures. Recent\nworks have suggested that multi-modal deep learning analysis can benefit from\nexplicitly disentangling anatomical (shape) and modality (appearance)\ninformation into separate image presentations. In this work, we challenge\nmainstream strategies by showing that they do not naturally lead to\nrepresentation disentanglement both in theory and in practice. To address this\nissue, we propose a margin loss that regularizes the similarity in\nrelationships of the representations across subjects and modalities. To enable\nrobust training, we further use a conditional convolution to design a single\nmodel for encoding images of all modalities. Lastly, we propose a fusion\nfunction to combine the disentangled anatomical representations as a set of\nmodality-invariant features for downstream tasks. We evaluate the proposed\nmethod on three multi-modal neuroimaging datasets. Experiments show that our\nproposed method can achieve superior disentangled representations compared to\nexisting disentanglement strategies. Results also indicate that the fused\nanatomical representation has potential in the downstream task of zero-dose PET\nreconstruction and brain tumor segmentation. The code is available at\n\\url{https://github.com/ouyangjiahong/representation-disentanglement}.\n"}
{"abstract": "  Recently, word enhancement has become very popular for Chinese Named Entity\nRecognition (NER), reducing segmentation errors and increasing the semantic and\nboundary information of Chinese words. However, these methods tend to ignore\nthe information of the Chinese character structure after integrating the\nlexical information. Chinese characters have evolved from pictographs since\nancient times, and their structure often reflects more information about the\ncharacters. This paper presents a novel Multi-metadata Embedding based\nCross-Transformer (MECT) to improve the performance of Chinese NER by fusing\nthe structural information of Chinese characters. Specifically, we use\nmulti-metadata embedding in a two-stream Transformer to integrate Chinese\ncharacter features with the radical-level embedding. With the structural\ncharacteristics of Chinese characters, MECT can better capture the semantic\ninformation of Chinese characters for NER. The experimental results obtained on\nseveral well-known benchmarking datasets demonstrate the merits and superiority\nof the proposed MECT method.\\footnote{The source code of the proposed method is\npublicly available at https://github.com/CoderMusou/MECT4CNER.\n"}
{"abstract": "  Let $\\mathcal{I}_{d,g,R}$ be the union of irreducible components of the\nHilbert scheme whose general points parametrize smooth, irreducible, curves of\ndegree $d$, genus $g$, which are non--degenerate in the projective space\n$\\mathbb{P}^R$. Under some numerical assumptions on $d$, $g$ and $R$, we\nconstruct irreducible components of $\\mathcal{I}_{d,g,R}$ other than the\nso-called {\\em distinguished component}, dominating the moduli space\n$\\mathcal{M}_g$ of smooth genus--$g$ curves, which are generically smooth and\nturn out to be of dimension higher than the expected one. The general point of\nany such a component corresponds to a curve $X \\subset \\mathbb{P}^R$ which is a\nsuitable ramified $m$--cover of an irrational curve $Y \\subset\n\\mathbb{P}^{R-1}$, $m \\geqslant 2$, lying in a surface cone over $Y$. The paper\nextends some of the results in previous papers of Y. Choi, H. Iliev, S. Kim\n(cf. [12,13] in Bibliography).\n"}
{"abstract": "  For $k\\ge1$, a $k$-almost prime is a positive integer with exactly $k$ prime\nfactors, counted with multiplicity. In this article we give elementary proofs\nof precise asymptotics for the reciprocal sum of $k$-almost primes. Our results\nmatch the strength of those of classical analytic methods. We also study the\nlimiting behavior of the constants appearing in these estimates, which may be\nviewed as higher analogues of the Mertens constant $\\beta=0.2614..$. Further,\nin the case $k=2$ of semiprimes we give yet finer-scale and explicit estimates,\nas well as a conjecture.\n"}
{"abstract": "  Learning vector representations for words is one of the most fundamental\ntopics in NLP, capable of capturing syntactic and semantic relationships useful\nin a variety of downstream NLP tasks. Vector representations can be limiting,\nhowever, in that typical scoring such as dot product similarity intertwines\nposition and magnitude of the vector in space. Exciting innovations in the\nspace of representation learning have proposed alternative fundamental\nrepresentations, such as distributions, hyperbolic vectors, or regions. Our\nmodel, Word2Box, takes a region-based approach to the problem of word\nrepresentation, representing words as $n$-dimensional rectangles. These\nrepresentations encode position and breadth independently and provide\nadditional geometric operations such as intersection and containment which\nallow them to model co-occurrence patterns vectors struggle with. We\ndemonstrate improved performance on various word similarity tasks, particularly\non less common words, and perform a qualitative analysis exploring the\nadditional unique expressivity provided by Word2Box.\n"}
{"abstract": "  The calcium superhydrides are synthesized at high pressure of 180 GPa and\n1000 degree high temperatures. Superconductivity with Tc onset about 210 K is\nrealized in thus obtained hydrogen rich calcium compounds at high pressure. The\ncritical magnetic field Hc2(T) is estimated to around 166 T in the Ginzburg\nLandau model.\n"}
{"abstract": "  In the field of light-matter interactions, it is often assumed that a\nclassical light field that interacts with a quantum particle remains almost\nunchanged and thus contains nearly no information about the manipulated\nparticles. To investigate the validity of this assumption, we develop and\ntheoretically analyze a simple Gedankenexperiment which involves the\ninteraction of a coherent state with a quantum particle in an optical cavity.\nWe quantify the resulting alteration of the light field by measuring the\nfidelity of its initial and equilibrium states. Using Bayesian inference, we\ndemonstrate the information transfer through photon measurements. In addition,\nwe employ the concepts of quantum entropy and mutual information to quantify\nthe entropy transfer from the particle to the light field. In the weak coupling\nlimit, we validate the usually assumed negligible alteration of the light field\nand entropy transfer. In the strong coupling limit, however, we observe that\nthe information of the initial particle state can be fully encoded in the light\nfield, even for large photon numbers. Nevertheless, we show that spontaneous\nemission is a sufficient mechanism for removing the entropy initially stored in\nthe particle. Our analysis provides a deeper understanding of the entropy\nexchange between quantum matter and classical light.\n"}
{"abstract": "  In this article, we study a divisor function in an arbitrary number field\nakin to Koshliakov's work on Vorono\\\"{\\dotlessi} summation formula. More\nprecisely, we generalize Koshliakov's kernel and Koshliakov's transform over\nany number field to obtain identities for the Lambert series associated to the\ndivisor function in an arbitrary number field.\n"}
{"abstract": "  Domain adaptation methods face performance degradation in object detection,\nas the complexity of tasks require more about the transferability of the model.\nWe propose a new perspective on how CNN models gain the transferability,\nviewing the weights of a model as a series of motion patterns. The directions\nof weights, and the gradients, can be divided into domain-specific and\ndomain-invariant parts, and the goal of domain adaptation is to concentrate on\nthe domain-invariant direction while eliminating the disturbance from\ndomain-specific one. Current UDA object detection methods view the two\ndirections as a whole while optimizing, which will cause domain-invariant\ndirection mismatch even if the output features are perfectly aligned. In this\npaper, we propose the domain-specific suppression, an exemplary and\ngeneralizable constraint to the original convolution gradients in\nbackpropagation to detach the two parts of directions and suppress the\ndomain-specific one. We further validate our theoretical analysis and methods\non several domain adaptive object detection tasks, including weather, camera\nconfiguration, and synthetic to real-world adaptation. Our experiment results\nshow significant advance over the state-of-the-art methods in the UDA object\ndetection field, performing a promotion of $10.2\\sim12.2\\%$ mAP on all these\ndomain adaptation scenarios.\n"}
{"abstract": "  In laser metal deposition (LMD) the height of the deposited track can vary\nwithin and between layers, causing significant deviations during the process\nevolution. Previous works have shown that in certain conditions a\nself-stabilizing mechanism occurs, maintaining a regular height growth and a\nconstant standoff distance between the part and the deposition nozzle. Here we\nanalyze the link between the powder catchment efficiency and the deposition\nheight stability. To this purpose, a monitoring system was developed to study\nthe deposition in different process conditions, using inline measurements of\nthe specimen weight in combination with the layer height information obtained\nwith coaxial optical triangulation. An analytical model was used to estimate\nthe deposition efficiency in real-time from the height monitoring and the\nprocess parameters, which was verified by the direct mass measurements. The\nresults show that the track height stabilization is associated to a reduction\nof the powder catchment efficiency, which is governed by the melt pool relative\nposition with respect to the powder cone and the laser beam. For a given set of\nparameters, the standoff distance can be estimated to achieve the highest\npowder catchment efficiency and a regular height through the build direction.\n"}
{"abstract": "  Circular colliders have the advantage of delivering collisions to multiple\ninteraction points, which allow different detector designs to be studied and\noptimized - up to four for FCC-ee. On the one hand, the detectors must satisfy\nthe constraints imposed by the invasive interaction region layout. On the other\nhand, the performance of heavy-flavour tagging, of particle identification, of\ntracking and particle-flow reconstruction, and of lepton, jet, missing energy\nand angular resolution, need to match the physics programme and the exquisite\nstatistical precision offered by FCC-ee. During the FCC feasibility study\n(2021-2025), benchmark physics processes will be used to determine, via\nappropriate simulations, the requirements on the detector performance or design\nthat must be satisfied to ensure that the systematic uncertainties of the\nmeasurements are commensurate with their statistical precision. The usage of\nthe data themselves, in order to reach the challenging goals on the stability\nand on the alignment of the detector, in particular for the programme at and\naround the Z peak, will also be studied. In addition, the potential for\ndiscovering very weakly coupled new particles, in decays of Z or Higgs bosons,\ncould motivate dedicated detector designs that would increase the efficiency\nfor reconstructing the unusual signatures of such processes. These studies are\na crucial input to the further optimization of the two concepts described in\nthe Conceptual Design Report, CLD, and IDEA, and to the development of new\nconcepts which might actually prove to be better adapted to the FCC-ee physics\nprogramme, or parts thereof.\n"}
{"abstract": "  We calculate the electric conductivity $\\sigma$ in deconfined QCD matter\nusing a holographic QCD model, i.e., the Sakai-Sugimoto Model with varying\nmagnetic field $B$ and chiral anomaly strength. After confirming that our\nestimated $\\sigma$ for $B=0$ is consistent with the lattice-QCD results, we\nstudy the case with $B\\neq 0$ in which the coefficient $\\alpha$ in the\nChern-Simons term controls the chiral anomaly strength. Our results imply that\nthe transverse conductivity, $\\sigma_\\perp$, is suppressed to be $\\lesssim\n70\\%$ at $B\\sim 1\\,\\mathrm{GeV}^2$ as compared to the $B=0$ case when the\ntemperature is fixed as $T= 0.2\\,\\mathrm{GeV}$. Since the Sakai-Sugimoto Model\nhas massless fermions, the longitudinal conductivity, $\\sigma_\\parallel$, with\n$B\\neq 0$ should diverge due to production of the matter chirality. Yet, it is\npossible to extract a regulated part out from $\\sigma_\\parallel$ with an extra\ncondition to neutralize the matter chirality. This regulated quantity is\ninterpreted as an Ohmic part of $\\sigma_\\parallel$. We show that the\nlongitudinal Ohmic conductivity increases with increasing $B$ for small\n$\\alpha$, while it is suppressed with larger $B$ for physical $\\alpha=3/4$ due\nto anomaly induced interactions.\n"}
{"abstract": "  We present systematic large-scale shell-model calculations for Rn isotopes\nwith $A=$ 207 to 216. For the $^{207-212}$Rn isotopes, we perform calculations\nwith KHH7B interaction, while for $^{213-216}$Rn isotopes with KHPE and KHH7B\ninteractions. The calculated energies and electromagnetic properties are\ncompared with the available experimental data and predicted where experimental\ndata are not available. We also suggest spins and parities of several\nunconfirmed states available from the recent experimental data. Comprehensive\nstudy of several isomeric states from the calculated shell-model configurations\nand half-lives is also reported.\n"}
{"abstract": "  Among non-Hermitian systems, pseudo-Hermitian phases represent a special\nclass of physical models characterized by real energy spectra and by the\nabsence of non-Hermitian skin effects. Here, we show that several\npseudo-Hermitian phases in two and three dimensions can be built by employing\n$q$-deformed matrices, which are related to the representation of deformed\nalgebras. Through this algebraic approach we present and study the\npseudo-Hermitian version of well known Hermitian topological phases, raging\nfrom two-dimensional Chern insulators and time-reversal-invariant topological\ninsulators to three-dimensional Weyl semimetals and chiral topological\ninsulators. We analyze their topological bulk states through non-Hermitian\ngeneralizations of Abelian and non-Abelian tensor Berry connections and quantum\nmetric. Although our pseudo-Hermitian models and their Hermitian counterparts\nshare the same topological invariants, their band geometries are different. We\nindeed show that some of our pseudo-Hermitian phases naturally support\nnearly-flat topological bands, opening the route to the study of\npseudo-Hermitian strongly-interacting systems. Finally, we provide an\nexperimental protocol to realize our models and measure the full non-Hermitian\nquantum geometric tensor in synthetic matter.\n"}
{"abstract": "  At present, people usually use some methods based on convolutional neural\nnetworks (CNNs) for Electroencephalograph (EEG) decoding. However, CNNs have\nlimitations in perceiving global dependencies, which is not adequate for common\nEEG paradigms with a strong overall relationship. Regarding this issue, we\npropose a novel EEG decoding method that mainly relies on the attention\nmechanism. The EEG data is firstly preprocessed and spatially filtered. And\nthen, we apply attention transforming on the feature-channel dimension so that\nthe model can enhance more relevant spatial features. The most crucial step is\nto slice the data in the time dimension for attention transforming, and finally\nobtain a highly distinguishable representation. At this time, global averaging\npooling and a simple fully-connected layer are used to classify different\ncategories of EEG data. Experiments on two public datasets indicate that the\nstrategy of attention transforming effectively utilizes spatial and temporal\nfeatures. And we have reached the level of the state-of-the-art in\nmulti-classification of EEG, with fewer parameters. As far as we know, it is\nthe first time that a detailed and complete method based on the transformer\nidea has been proposed in this field. It has good potential to promote the\npracticality of brain-computer interface (BCI). The source code can be found\nat: \\textit{https://github.com/anranknight/EEG-Transformer}.\n"}
{"abstract": "  Scene text detection and recognition have been well explored in the past few\nyears. Despite the progress, efficient and accurate end-to-end spotting of\narbitrarily-shaped text remains challenging. In this work, we propose an\nend-to-end text spotting framework, termed PAN++, which can efficiently detect\nand recognize text of arbitrary shapes in natural scenes. PAN++ is based on the\nkernel representation that reformulates a text line as a text kernel (central\nregion) surrounded by peripheral pixels. By systematically comparing with\nexisting scene text representations, we show that our kernel representation can\nnot only describe arbitrarily-shaped text but also well distinguish adjacent\ntext. Moreover, as a pixel-based representation, the kernel representation can\nbe predicted by a single fully convolutional network, which is very friendly to\nreal-time applications. Taking the advantages of the kernel representation, we\ndesign a series of components as follows: 1) a computationally efficient\nfeature enhancement network composed of stacked Feature Pyramid Enhancement\nModules (FPEMs); 2) a lightweight detection head cooperating with Pixel\nAggregation (PA); and 3) an efficient attention-based recognition head with\nMasked RoI. Benefiting from the kernel representation and the tailored\ncomponents, our method achieves high inference speed while maintaining\ncompetitive accuracy. Extensive experiments show the superiority of our method.\nFor example, the proposed PAN++ achieves an end-to-end text spotting F-measure\nof 64.9 at 29.2 FPS on the Total-Text dataset, which significantly outperforms\nthe previous best method. Code will be available at: https://git.io/PAN.\n"}
{"abstract": "  Detecting and segmenting salient objects from given image scenes has received\ngreat attention in recent years. A fundamental challenge in training the\nexisting deep saliency detection models is the requirement of large amounts of\nannotated data. While gathering large quantities of training data becomes cheap\nand easy, annotating the data is an expensive process in terms of time, labor\nand human expertise. To address this problem, this paper proposes to learn the\neffective salient object detection model based on the manual annotation on a\nfew training images only, thus dramatically alleviating human labor in training\nmodels. To this end, we name this task as the few-cost salient object detection\nand propose an adversarial-paced learning (APL)-based framework to facilitate\nthe few-cost learning scenario. Essentially, APL is derived from the self-paced\nlearning (SPL) regime but it infers the robust learning pace through the\ndata-driven adversarial learning mechanism rather than the heuristic design of\nthe learning regularizer. Comprehensive experiments on four widely-used\nbenchmark datasets demonstrate that the proposed method can effectively\napproach to the existing supervised deep salient object detection models with\nonly 1k human-annotated training images. The project page is available at\nhttps://github.com/hb-stone/FC-SOD.\n"}
{"abstract": "  In a recent paper it was shown how a matrix S-lemma can be applied to\nconstruct controllers from noisy data. The current paper complements these\nresults by proving a matrix version of the classical Finsler's lemma. This\nmatrix Finsler's lemma provides a tractable condition under which all matrix\nsolutions to a quadratic equality also satisfy a quadratic inequality. We will\napply this result to bridge known data-driven control design techniques for\nboth exact and noisy data, thereby revealing a more general theory. The result\nis also applied to data-driven control of Lur'e systems.\n"}
{"abstract": "  We consider the single eigenvalue fluctuations of random matrices of general\nWigner-type, under a one-cut assumption on the density of states. For\neigenvalues in the bulk, we prove that the asymptotic fluctuations of a single\neigenvalue around its classical location are Gaussian with a universal variance\nwhich agrees with the GOE and GUE cases.\n  Our method is based on a dynamical approach to mesoscopic linear spectral\nstatistics which reduces their behavior on short scales to that on larger\nscales. We prove a central limit theorem for linear spectral statistics on\nlarger scales via resolvent techniques and show that for certain classes of\ntest functions, the leading order contribution to the variance is universal,\nagreeing with the GOE/GUE cases.\n"}
{"abstract": "  Due to the vulnerability of deep neural networks to adversarial examples,\nnumerous works on adversarial attacks and defenses have been burgeoning over\nthe past several years. However, there seem to be some conventional views\nregarding adversarial attacks and object detection approaches that most\nresearchers take for granted. In this work, we bring a fresh perspective on\nthose procedures by evaluating the impact of universal perturbations on object\ndetection at a class-level. We apply it to a carefully curated data set related\nto autonomous driving. We use Faster-RCNN object detector on images of five\ndifferent categories: person, car, truck, stop sign and traffic light from the\nCOCO data set, while carefully perturbing the images using Universal Dense\nObject Suppression algorithm. Our results indicate that person, car, traffic\nlight, truck and stop sign are resilient in that order (most to least) to\nuniversal perturbations. To the best of our knowledge, this is the first time\nsuch a ranking has been established which is significant for the security of\nthe data sets pertaining to autonomous vehicles and object detection in\ngeneral.\n"}
{"abstract": "  Suppose that $(T,\\star)$ is a groupoid with a left identity such that each\nelement $a\\in T$ has a left inverse. Then $T$ is called a \\textit{gyrogroup} if\nand only if $(i)$ there exists a function $gyr:T\\times T\\longrightarrow Aut(T)$\nsuch that for all $a,b,c\\in T$, $a\\star(b\\star c)= (a\\star b)\\star gyr[a,b]c$,\nwhere $gyr[a,b]c=gyr(a,b)(c)$; and $(ii)$ for all $a,b\\in T$,\n$gyr[a,b]=gyr[a\\star b,b]$. In this paper, the structure of normal\nsubgyrogroups of certain gyrogroups are investigated.\n"}
{"abstract": "  Attention-based encoder-decoder (AED) models learn an implicit internal\nlanguage model (ILM) from the training transcriptions. The integration with an\nexternal LM trained on much more unpaired text usually leads to better\nperformance. A Bayesian interpretation as in the hybrid autoregressive\ntransducer (HAT) suggests dividing by the prior of the discriminative acoustic\nmodel, which corresponds to this implicit LM, similarly as in the hybrid hidden\nMarkov model approach. The implicit LM cannot be calculated efficiently in\ngeneral and it is yet unclear what are the best methods to estimate it. In this\nwork, we compare different approaches from the literature and propose several\nnovel methods to estimate the ILM directly from the AED model. Our proposed\nmethods outperform all previous approaches. We also investigate other methods\nto suppress the ILM mainly by decreasing the capacity of the AED model,\nlimiting the label context, and also by training the AED model together with a\npre-existing LM.\n"}
{"abstract": "  As announced \"Intrinsic mirror symmetry and punctured invariants\" in 2016, we\nconstruct and prove consistency of the canonical wall structure. This\nconstruction starts with a log Calabi-Yau pair (X,D) and produces a wall\nstructure, as defined by Gross-Hacking-Siebert. Roughly put, the canonical wall\nstructure is a data structure which encodes an algebro-geometric analogue of\ncounts of Maslov index zero disks. These enumerative invariants are defined in\nterms of the punctured invariants of Abramovich-Chen-Gross-Siebert. There are\nthen two main theorems of the paper. First, we prove consistency of the\ncanonical wall structure, so that the canonical wall structure gives rise to a\nmirror family. Second, we prove that this mirror family coincides with the\nintrinsic mirror constructed in our paper \"Intrinsic mirror symmetry\". While\nthe setup of this paper is narrower than that of the latter paper, it gives a\nmore detailed description of the mirror.\n"}
{"abstract": "  We study an interesting and challenging problem, learning any part of a\nBayesian network (BN) structure. In this challenge, it will be computationally\ninefficient using existing global BN structure learning algorithms to find an\nentire BN structure to achieve the part of a BN structure in which we are\ninterested. And local BN structure learning algorithms encounter the false edge\norientation problem when they are directly used to tackle this challenging\nproblem. In this paper, we first present a new concept of Expand-Backtracking\nto explain why local BN structure learning methods have the false edge\norientation problem, then propose APSL, an efficient and accurate Any Part of\nBN Structure Learning algorithm. Specifically, APSL divides the V-structures in\na Markov blanket (MB) into two types: collider V-structure and non-collider\nV-structure, then it starts from a node of interest and recursively finds both\ncollider V-structures and non-collider V-structures in the found MBs, until the\npart of a BN structure in which we are interested are oriented. To improve the\nefficiency of APSL, we further design the APSL-FS algorithm using Feature\nSelection, APSL-FS. Using six benchmark BNs, the extensive experiments have\nvalidated the efficiency and accuracy of our methods.\n"}
{"abstract": "  The impact parameter dependence of color charge correlators in the proton is\nobtained from the light front formalism in light cone gauge. We include NLO\ncorrections due to the $|qqqg\\rangle$ Fock state via light-cone perturbation\ntheory. Near the center of the proton, the $b$-dependence of the correlations\nis very different from a \"transverse profile function\". The resulting\n$t$-dependence of exclusive $J/\\Psi$ photoproduction transitions from\nexponential to power law at $|t| \\approx 1$ GeV$^2$. This prediction could be\ntested at upcoming DIS facilities or in nucleus-proton ultraperipheral\ncollisions (UPCs).\n"}
{"abstract": "  Today, human security analysts collapse under the sheer volume of alerts they\nhave to triage during investigations. The inability to cope with this load,\ncoupled with a high false positive rate of alerts, creates alert fatigue. This\nresults in failure to detect complex attacks, such as advanced persistent\nthreats (APTs), because they manifest over long time frames and attackers tread\ncarefully to evade detection mechanisms. In this paper, we contribute a new\nmethod to synthesize attack graphs from state machines. We use the network\ndirection to derive potential attack stages from single and meta-alerts and\nmodel resulting attack scenarios in a kill chain state machine (KCSM). Our\nalgorithm yields a graphical summary of the attack, APT scenario graphs, where\nnodes represent involved hosts and edges infection activity. We evaluate the\nfeasibility of our approach in multiple experiments based on the\nCSE-CIC-IDS2018 data set. We obtain up to 446 458 singleton alerts that our\nalgorithm condenses into 700 APT scenario graphs resulting in a reduction of up\nto three orders of magnitude. This reduction makes it feasible for human\nanalysts to effectively triage potential incidents. An evaluation on the same\ndata set, in which we embedded a synthetic yet realistic APT campaign, supports\nthe applicability of our approach of detecting and contextualizing complex\nattacks. The APT scenario graphs constructed by our algorithm correctly link\nlarge parts of the APT campaign and present a coherent view to support the\nhuman analyst in further analyses.\n"}
{"abstract": "  We present a systematic study of higher-order exchange interactions beyond\nthe pair-wise Heisenberg exchange in transition-metal trilayers based on\ndensity functional theory calculations. We show that these terms can play an\nimportant role in magnetic trilayers composed of a single hexagonal Fe or Co\natomic layer sandwiched between $4d$ and $5d$ transition-metal layers. We study\nthe dependence of the biquadratic and the three-site and four-site four spin\ninteraction on the band filling of the $4d$ and $5d$ layers as well as the\nstacking sequence, i.e. fcc vs. hcp stacking. Our calculations reveal\nrelatively small higher-order interactions for Co based trilayers. For Fe based\ntrilayers with a Rh or Ir layer the higher-order terms can be on the same order\nof magnitude as pair-wise Heisenberg exchange. The trends obtained for\nfreestanding trilayers are used to understand the higher-order interactions in\nultrathin film systems on surfaces that are experimentally accessible. It is\nshown that hcp Rh/Fe/Ir(111) and hcp-Rh/Fe/Rh(111) exhibit the largest values\nfor the biquadratic and the three-site four spin interaction of all systems\nunder study. We further demonstrate that the three-site four spin interaction\nis responsible for the experimentally observed change of the magnetic ground\nstate of Rh/Fe/Ir(111) from a spin spiral (single-Q) for fcc-Rh to a 2Q state\nfor hcp-Rh. We find similar trends for Rh/Fe/Rh(111), i.e. replacing the Ir\nsurface by the isoelectronic Rh surface. For Rh/Co/Ir(111), we obtain a\nnegative value for the four-site four spin interaction which will lead to a\nreduced stability of magnetic skyrmions which are metastable in this film at\nzero magnetic field. In contrast, for Pd/Fe/Ir(111), the four-site four spin\ninteraction is positive which leads to an enhanced stability of skyrmions.\n"}
{"abstract": "  This paper presents a new vision Transformer, called Swin Transformer, that\ncapably serves as a general-purpose backbone for computer vision. Challenges in\nadapting Transformer from language to vision arise from differences between the\ntwo domains, such as large variations in the scale of visual entities and the\nhigh resolution of pixels in images compared to words in text. To address these\ndifferences, we propose a hierarchical Transformer whose representation is\ncomputed with \\textbf{S}hifted \\textbf{win}dows. The shifted windowing scheme\nbrings greater efficiency by limiting self-attention computation to\nnon-overlapping local windows while also allowing for cross-window connection.\nThis hierarchical architecture has the flexibility to model at various scales\nand has linear computational complexity with respect to image size. These\nqualities of Swin Transformer make it compatible with a broad range of vision\ntasks, including image classification (87.3 top-1 accuracy on ImageNet-1K) and\ndense prediction tasks such as object detection (58.7 box AP and 51.1 mask AP\non COCO test-dev) and semantic segmentation (53.5 mIoU on ADE20K val). Its\nperformance surpasses the previous state-of-the-art by a large margin of +2.7\nbox AP and +2.6 mask AP on COCO, and +3.2 mIoU on ADE20K, demonstrating the\npotential of Transformer-based models as vision backbones. The hierarchical\ndesign and the shifted window approach also prove beneficial for all-MLP\narchitectures. The code and models are publicly available\nat~\\url{https://github.com/microsoft/Swin-Transformer}.\n"}
{"abstract": "  The modification of the hard core of jets in a dense QCD medium is studied.\nIn particular, we consider partons which possess a virtuality somewhat larger\nthan the multiple scattering scale of the medium ($\\hat{q} \\tau$, where\n$\\hat{q}$ is the transverse broadening jet transport coefficient, and $\\tau$ is\nthe formation length of a particular emission). We delineate the region of\nparameter space where the higher-twist approach is applicable, and derive the\nin-medium DGLAP evolution equation. We study a region in parameter space where\nthis is the dominant mechanism of energy loss. We argue that such a regime is\npervasive in most cases of jets in $A$-$A$ and future $e$-$A$ collisions, and\ncontrols the modification of the hard core of jets and the leading single\nparticle spectrum at high transverse momentum ($p_\\mathrm{T}$).\n"}
{"abstract": "  We present a search for continuous gravitational-wave emission due to r-modes\nin the pulsar PSR J0537-6910 using data from the LIGO-Virgo Collaboration\nobserving run O3. PSR J0537-6910 is a young energetic X-ray pulsar and is the\nmost frequent glitcher known. The inter-glitch braking index of the pulsar\nsuggests that gravitational-wave emission due to r-mode oscillations may play\nan important role in the spin evolution of this pulsar. Theoretical models\nconfirm this possibility and predict emission at a level that can be probed by\nground-based detectors. In order to explore this scenario, we search for r-mode\nemission in the epochs between glitches by using a contemporaneous timing\nephemeris obtained from NICER data. We do not detect any signals in the\ntheoretically expected band of 86-97 Hz, and report upper limits on the\namplitude of the gravitational waves. Our results improve on previous amplitude\nupper limits from r-modes in J0537-6910 by a factor of up to 3 and place\nstringent constraints on theoretical models for r-mode driven spin-down in PSR\nJ0537-6910, especially for higher frequencies at which our results reach below\nthe spin-down limit defined by energy conservation.\n"}
{"abstract": "  Maritime autonomous transportation has played a crucial role in the\nglobalization of the world economy. Deep Reinforcement Learning (DRL) has been\napplied to automatic path planning to simulate vessel collision avoidance\nsituations in open seas. End-to-end approaches that learn complex mappings\ndirectly from the input have poor generalization to reach the targets in\ndifferent environments. In this work, we present a new strategy called\nstate-action rotation to improve agent's performance in unseen situations by\nrotating the obtained experience (state-action-state) and preserving them in\nthe replay buffer. We designed our model based on Deep Deterministic Policy\nGradient, local view maker, and planner. Our agent uses two deep Convolutional\nNeural Networks to estimate the policy and action-value functions. The proposed\nmodel was exhaustively trained and tested in maritime scenarios with real maps\nfrom cities such as Montreal and Halifax. Experimental results show that the\nstate-action rotation on top of the CVN consistently improves the rate of\narrival to a destination (RATD) by up 11.96% with respect to the Vessel\nNavigator with Planner and Local View (VNPLV), as well as it achieves superior\nperformance in unseen mappings by up 30.82%. Our proposed approach exhibits\nadvantages in terms of robustness when tested in a new environment, supporting\nthe idea that generalization can be achieved by using state-action rotation.\n"}
{"abstract": "  In this article, we give explicit calculations for the family Floer mirrors\nof some non-compact Calabi-Yau surfaces. We compare it with the mirror\nconstruction of Gross-Hacking-Keel for suitably chosen log Calabi-Yau pairs and\nthe rank two cluster varieties of finite type. In particular, the\nanalytifications of the later two give partial compactifications of the family\nFloer mirrors that we computed.\n"}
{"abstract": "  Neural style transfer, allowing to apply the artistic style of one image to\nanother, has become one of the most widely showcased computer vision\napplications shortly after its introduction. In contrast, related tasks in the\nmusic audio domain remained, until recently, largely untackled. While several\nstyle conversion methods tailored to musical signals have been proposed, most\nlack the 'one-shot' capability of classical image style transfer algorithms. On\nthe other hand, the results of existing one-shot audio style transfer methods\non musical inputs are not as compelling. In this work, we are specifically\ninterested in the problem of one-shot timbre transfer. We present a novel\nmethod for this task, based on an extension of the vector-quantized variational\nautoencoder (VQ-VAE), along with a simple self-supervised learning strategy\ndesigned to obtain disentangled representations of timbre and pitch. We\nevaluate the method using a set of objective metrics and show that it is able\nto outperform selected baselines.\n"}
{"abstract": "  We present the first multi-task learning model -- named PhoNLP -- for joint\nVietnamese part-of-speech (POS) tagging, named entity recognition (NER) and\ndependency parsing. Experiments on Vietnamese benchmark datasets show that\nPhoNLP produces state-of-the-art results, outperforming a single-task learning\napproach that fine-tunes the pre-trained Vietnamese language model PhoBERT\n(Nguyen and Nguyen, 2020) for each task independently. We publicly release\nPhoNLP as an open-source toolkit under the Apache License 2.0. Although we\nspecify PhoNLP for Vietnamese, our PhoNLP training and evaluation command\nscripts in fact can directly work for other languages that have a pre-trained\nBERT-based language model and gold annotated corpora available for the three\ntasks of POS tagging, NER and dependency parsing. We hope that PhoNLP can serve\nas a strong baseline and useful toolkit for future NLP research and\napplications to not only Vietnamese but also the other languages. Our PhoNLP is\navailable at: https://github.com/VinAIResearch/PhoNLP\n"}
{"abstract": "  We develop a novel method for carrying out model selection for Bayesian\nautoencoders (BAEs) by means of prior hyper-parameter optimization. Inspired by\nthe common practice of type-II maximum likelihood optimization and its\nequivalence to Kullback-Leibler divergence minimization, we propose to optimize\nthe distributional sliced-Wasserstein distance (DSWD) between the output of the\nautoencoder and the empirical data distribution. The advantages of this\nformulation are that we can estimate the DSWD based on samples and handle\nhigh-dimensional problems. We carry out posterior estimation of the BAE\nparameters via stochastic gradient Hamiltonian Monte Carlo and turn our BAE\ninto a generative model by fitting a flexible Dirichlet mixture model in the\nlatent space. Consequently, we obtain a powerful alternative to variational\nautoencoders, which are the preferred choice in modern applications of\nautoencoders for representation learning with uncertainty. We evaluate our\napproach qualitatively and quantitatively using a vast experimental campaign on\na number of unsupervised learning tasks and show that, in small-data regimes\nwhere priors matter, our approach provides state-of-the-art results,\noutperforming multiple competitive baselines.\n"}
{"abstract": "  We present ARCH, a computational pathology (CP) multiple instance captioning\ndataset to facilitate dense supervision of CP tasks. Existing CP datasets focus\non narrow tasks; ARCH on the other hand contains dense diagnostic and\nmorphological descriptions for a range of stains, tissue types and pathologies.\nUsing intrinsic dimensionality estimation, we show that ARCH is the only CP\ndataset to (ARCH-)rival its computer vision analog MS-COCO Captions. We\nconjecture that an encoder pre-trained on dense image captions learns\ntransferable representations for most CP tasks. We support the conjecture with\nevidence that ARCH representation transfers to a variety of pathology sub-tasks\nbetter than ImageNet features or representations obtained via self-supervised\nor multi-task learning on pathology images alone. We release our best model and\ninvite other researchers to test it on their CP tasks.\n"}
{"abstract": "  Using the state-of-the-art SKA precursor, the MeerKAT radio telescope, we\nexplore the limits to precision pulsar timing of millisecond pulsars achievable\ndue to pulse stochasticity (jitter). We report new jitter measurements in 15 of\nthe 29 pulsars in our sample and find that the levels of jitter can vary\ndramatically between them. For some, like the 2.2~ms pulsar PSR J2241--5236, we\nmeasure an implied jitter of just $\\sim$ 4~ns/hr, while others like the 3.9~ms\nPSR J0636--3044 are limited to $\\sim$ 100 ns/hr. While it is well known that\njitter plays a central role to limiting the precision measurements of arrival\ntimes for high signal-to-noise ratio observations, its role in the measurement\nof dispersion measure (DM) has not been reported, particularly in broad-band\nobservations. Using the exceptional sensitivity of MeerKAT, we explored this on\nthe bright millisecond pulsar PSR J0437--4715 by exploring the DM of literally\nevery pulse. We found that the derived single pulse DMs vary by typically\n0.0085 cm$^{-3}$ pc from the mean, and that the best DM estimate is limited by\nthe differential pulse jitter across the band. We postulate that all\nmillisecond pulsars will have their own limit on DM precision which can only be\novercome with longer integrations. Using high-time resolution filterbank data\nof 9 $\\mu$s, we also present a statistical analysis of single pulse\nphenomenology. Finally, we discuss optimization strategies for the MeerKAT\npulsar timing program and its role in the context of the International Pulsar\nTiming Array (IPTA).\n"}
{"abstract": "  Highly accurate profiles of consumers daily energy usage are reported to\npower grid via smart meters which enables smart grid to effectively regulate\npower demand and supply. However, consumers energy consumption pattern can\nreveal personal and sensitive information regarding their lifestyle. Therefore,\nto ensure users privacy, differentially distributed noise is added to the\noriginal data. This technique comes with a trade off between privacy of the\nconsumer versus utility of the data in terms of providing services like\nbilling, Demand Response schemes, and Load Monitoring. In this paper, we\npropose a technique - Differential Privacy with Noise Cancellation Technique\n(DPNCT) - to maximize utility in aggregated load monitoring and fair billing\nwhile preserving users privacy by using noise cancellation mechanism on\ndifferentially private data. We introduce noise to the sensitive data stream\nbefore it leaves smart meters in order to guarantee privacy at individual\nlevel. Further, we evaluate the effects of different periodic noise cancelling\nschemes on privacy and utility i.e., billing and load monitoring. Our proposed\nscheme outperforms the existing scheme in terms of preserving the privacy while\naccurately calculating the bill.\n"}
{"abstract": "  In point of fact the Indian tradition in mathematics is long and glorious. It\ndates back to earliest times, and indeed many of the Indian discoveries from\n5000 years ago correspond rather naturally to modern mathematical results.\n"}
{"abstract": "  Over half of disk galaxies are barred, yet the mechanisms for bar formation\nand the life-time of bar buckling remain poorly understood. In simulations, a\nthin bar undergoes a rapid (<1 Gyr) event called \"buckling,\" during which the\ninner part of the bar is asymmetrically bent out of the galaxy plane and\neventually thickens, developing a peanut/X-shaped profile when viewed side-on.\nThrough analyzing stellar kinematics of N-body model snapshots of a galaxy\nbefore, during, and after the buckling phase, we confirm a distinct quadrupolar\npattern of out-of-plane stellar velocities in nearly face-on galaxies. This\nkinematic signature of buckling allows us to identify five candidates of\ncurrently buckling bars among 434 barred galaxies in the Mapping Nearby\nGalaxies at Apache Point Observatory (MaNGA) Survey, an integral field unit\n(IFU) spectroscopic survey that measures the composition and kinematic\nstructure of nearby galaxies. The frequency of buckling events detected is\nconsistent with the 0.5-1 Gyr timescale predicted by simulations. The five\ncandidates we present more than double the total number of candidate buckling\nbars, and are the only ones found using the kinematic signature.\n"}
{"abstract": "  The difficulty of solving a multi-objective optimization problem is impacted\nby the number of objectives to be optimized. The presence of many objectives\ntypically introduces a number of challenges that affect the choice/design of\noptimization algorithms. This paper investigates the drivers of these\nchallenges from two angles: (i) the influence of the number of objectives on\nproblem characteristics and (ii) the practical behavior of commonly used\nprocedures and algorithms for coping with many objectives. In addition to\nreviewing various drivers, the paper makes theoretical contributions by\nquantifying some drivers and/or verifying these drivers empirically by carrying\nout experiments on multi-objective NK landscapes and other typical benchmarks.\nWe then make use of our theoretical and empirical findings to derive practical\nrecommendations to support algorithm design. Finally, we discuss remaining\ntheoretical gaps and opportunities for future research in the area of multi-\nand many-objective optimization.\n"}
{"abstract": "  Searching large digital repositories can be extremely frustrating, as common\nlist-based formats encourage users to adopt a convenience-sampling approach\nthat favours chance discovery and random search, over meaningful exploration.\nWe have designed a methodology that allows users to visually and thematically\nexplore corpora, while developing personalised holistic reading strategies. We\ndescribe the results of a three-phase qualitative study, in which experienced\nresearchers used our interactive visualisation approach to analyse a set of\npublications and select relevant themes and papers. Using in-depth\nsemi-structured interviews and stimulated recall, we found that users: (i)\nselected papers that they otherwise would not have read, (ii) developed a more\ncoherent reading strategy, and (iii) understood the thematic structure and\nrelationships between papers more effectively. Finally, we make six design\nrecommendations to enhance current digital repositories that we have shown\nencourage users to adopt a more holistic and thematic research approach.\n"}
{"abstract": "  Detection of semantic contradictory sentences is one of the most challenging\nand fundamental issues for NLP applications such as recognition of textual\nentailments. Contradiction in this study includes different types of semantic\nconfrontation, such as conflict and antonymy. Due to lack of sufficient data to\napply precise machine learning and specifically deep learning methods to\nPersian and other low resource languages, rule-based approaches that can\nfunction similarly to these systems will be of a great interest. Also recently,\nemergence of new methods such as transfer learning, has opened up the\npossibility of deep learning for low-resource languages. Considering two above\npoints, in this study, along with a simple rule-base baseline, a novel\nrule-base system for identifying semantic contradiction along with a Bert base\ndeep contradiction detection system for Persian texts have been introduced. The\nrule base system has used frequent rule mining method to extract appropriate\ncontradiction rules using a development set. Extracted rules are tested for\ndifferent categories of contradictory sentences. In this system the maximum\nf-measure among contradiction categories is obtained for negation about 90% and\nthe average F-measure of system for all classes is about 76% which outperforms\nother algorithms on Persian texts. On the other hand, because of medium\nperformance of rule base system for some categories of contradiction, we use a\nBert base deep learning system using our translated dataset; with average\nF-measure of 73. Our hybrid system has f-measure of about 80.\n"}
{"abstract": "  In this work, we study thermal conductivity (k) of BC5, an ultra-hard\ndiamondlike semiconductor material, using first-principles computations and\nanalyze the effect of both isotopic disorder as well as length scale\ndependence. k of isotopically pure BC5 is computed to be 169 Wm- 1K-1 (along\na-axis) at 300K; this high k is found to be due to the high frequencies and\nphonon group velocities of both acoustic and optical phonons owing to the light\natomic mass of Carbon (C) and Boron (B) atoms and strong C-C and B-C bonds. We\nalso observe a dominance of optical phonons (~ 54%) over acoustic phonons in\nheat conduction at higher temperatures (~500 K). This unusually high\ncontribution of optical phonons is found to be due to a unique effect in BC5\nrelated to a weaker temperature dependence of optical phonon scattering rates\nrelative to acoustic phonons. The effect is explained in terms of high\nfrequencies of optical phonons causing decay into other high frequency phonons,\nwhere low phonon populations cause the decay term to become insensitive to\ntemperature. The effect further leads to high nanoscale thermal conductivity of\n77 Wm-1K-1 at 100 nm length scale due to optical phonon meanfreepaths being in\nnanometer regime. These results provide avenues for application of BC5 in\nnanoscale thermal management.\n"}
{"abstract": "  In this paper, a novel adaptive smooth disturbance observer-based fast\nfinite-time adaptive backstepping control scheme is presented for the attitude\ntracking of the 3-DOF helicopter system subject to compound disturbances.\nFirst, an adaptive smooth disturbance observer (ASDO) is proposed to estimate\nthe composite disturbance, which owns the characteristics of smooth output,\nfast finite-time convergence, and adaptability to the disturbance of unknown\nderivative boundary. Then, a finite-time backstepping control protocol is\nconstruct to drive the elevation and pitch angles to track reference\ntrajectories. To tackle the \"explosion of complexity\" and \"singularity\"\nproblems in the conventional backstepping design framework, a fast finite-time\ncommand filter (FFTCF) is utilized to estimate the virtual control signal and\nits derivative. Moreover, a fractional power-based auxiliary dynamic system is\nintroduced to compensate the error caused by the FFTCF estimation. Furthermore,\nan improved fractional power-based adaptive law with the $\\sigma $-modification\nterm is designed to attenuate the observer approximation error, such that the\ntracking performance is further enhanced. In terms of the fast finite-time\nstability theory, the signals of the closed-loop system are all fast\nfinite-time bounded while the attitude tracking errors can fast converge to a\nsufficiently small region of the origin in finite time. Finally, a contrastive\nnumerical simulation is carried out to validate the effectiveness and\nsuperiority of the designed control scheme.\n"}
{"abstract": "  THz pulses are generated from femtosecond pulse-excited\nferromagnetic/nonmagnetic spintronic heterostructures via inverse spin Hall\neffect. The contribution from ultrafast demagnetization/remagnetization is\nextremely weak, in the comparison. The highest possible THz signal strength\nfrom spintronic THz emitters is limited by the optical damage threshold of the\ncorresponding heterostructures. The THz generation efficiency does not saturate\nwith the excitation fluence even up till the damage threshold. Bilayer (Fe,\nCoFeB)/(Pt, Ta) based FM/NM spintronic heterostructures have been studied for\nan optimized performance for THz generation when pumped by sub-50 fs amplified\nlaser pulses at 800 nm. Among them, CoFeB/Pt is the best combination for an\nefficient THz source. The optimized FM/NM spintronic heterostructure on a\nquartz substrate, having alpha-phase Ta as the nonmagnetic layer, show the\nhighest damage threshold as compared to those with Pt, irrespective of their\ngeneration efficiency. The damage threshold of the Fe/Ta heterostructure on\nquartz substrate is ~85 GW/cm2.\n"}
{"abstract": "  A dark matter search project needs and extremely low background radiation\ndetector since the expected event rate of dark matter is less than a few events\nin one year in one tonne of the detector mass. The authors developed a highly\nradiopure NaI(Tl) crystal to search for dark matter. The best combination of\nthe purification methods was developed, resulting $^{\\mathrm{nat}}$K and\n$^{210}$Pb were less than 20 ppb and 5.7 $\\mu$Bq/kg, respectively.\n  The authors will construct a large volume detector system with high-purity\nNaI(Tl) crystals. The design and the performance of the prototype detector\nmodule will be reported in this article.\n"}
{"abstract": "  In Phys. Rev. Lett. 120, 200603 (2018), a segmented XXZ spin chain with zero\nanisotropy in one half and a large anisotropy on the other half gave rise to a\nspin current rectification which is perfect in the thermodynamic limit. Here we\nextend the previous study to segmented chains with interacting integrable as\nwell as non-integrable halves, considering even cases in which no ballistic\ntransport can emerge in either half. We demonstrate that, also in this more\ngeneral case, it is possible to obtain giant rectification when the two\ninteracting half chains are sufficiently different. We also show that the\nmechanism causing this effect is the emergence of an energy gap in the\nexcitation spectrum of the out-of-equilibrium insulating steady state in one of\nthe two biases. Finally we demonstrate that in the thermodynamic limit there is\nno perfect rectification when each of the two half chains is interacting.\n"}
{"abstract": "  PdCoO2 , belonging to a family of triangular oxides called delafossite, is\none of the most conducting oxides. Its in-plane conductivity is comparable to\nthose of the best metals, and exhibits hydrodynamic electronic transport with\nextremely long mean free path at cryogenic temperatures. Nonetheless, it is\nnonmagnetic despite the presence of the cobalt ion. Here, we show that a mild\nhydrogenation process reduces PdCoO2 thin films to an atomically-mixed alloy of\nPdCo with strong out-of-plane ferromagnetism and sign-tunable anomalous Hall\neffect. Considering that many other compounds remain little affected under a\nsimilar hydrogenation condition, this discovery may provide a route to creating\nnovel spintronic heterostructures combining strong ferromagnetism, involving\noxides and other functional materials.\n"}
{"abstract": "  In the present work, NIR, LIBS, Raman and XRD techniques have been\ncomplementarily used to carry out a comprehensive characterization of a\nterrestrial analogue selected from the Chesapeake Bay Impact Structure (CBIS).\nThe obtained data clearly highlight the key role of Raman spectroscopy in the\ndetection of minor and trace compounds, through which inferences about\ngeological processes occurred in the CBIS can be extrapolated. Beside the use\nof commercial systems, further Raman analyses were performed by the Raman Laser\nSpectrometer (RLS) ExoMars Simulator. This instrument represents the most\nreliable tool to effectively predict the scientific capabilities of the\nExoMars/Raman system that will be deployed on Mars in 2021. By emulating the\nanalytical procedures and operational restrictions established by the ExoMars\nmission rover design, it was proved that the RLS ExoMars Simulator is able to\ndetect the amorphization of quartz, which constitutes an analytical clue of the\nimpact origin of craters. On the other hand, the detection of barite and\nsiderite, compounds crystallizing under hydrothermal conditions, helps to\nindirectly confirm the presence of water in impact targets. Furthermore, the\nRLS ExoMars Simulator capability of performing smart molecular mappings was\nalso evaluated. According to the obtained results, the algorithms developed for\nits operation provide a great analytical advantage over most of the automatic\nanalysis systems employed by commercial Raman instruments, encouraging its\napplication for many additional scientific and commercial purposes.\n"}
{"abstract": "  We propose a field-based design for dielectric antennas to interface diamond\ncolor centers with a Gaussian propagating far field. This antenna design\nenables an efficient spin-photon interface with a Purcell factor exceeding 400\nand a 93% mode overlap to a 0.4 numerical aperture far-field Gaussian mode. The\nantenna design is robust to fabrication imperfections, such as variations in\nthe dimensions of the dielectric perturbations and the emitter dipole location.\nThe field-based dielectric antenna design provides an efficient free-space\ninterface to closely packed arrays of quantum memories for multiplexed quantum\nrepeaters, arrayed quantum sensors, and modular quantum computers.\n"}
{"abstract": "  In a wireless network that conveys status updates from sources (i.e.,\nsensors) to destinations, one of the key issues studied by existing literature\nis how to design an optimal source sampling strategy on account of the\ncommunication constraints which are often modeled as queues. In this paper, an\nalternative perspective is presented -- a novel status-aware communication\nscheme, namely \\emph{parallel communications}, is proposed which allows sensors\nto be communication-agnostic. Specifically, the proposed scheme can determine,\nbased on an online prediction functionality, whether a status packet is worth\ntransmitting considering both the network condition and status prediction, such\nthat sensors can generate status packets without communication constraints. We\nevaluate the proposed scheme on a Software-Defined-Radio (SDR) test platform,\nwhich is integrated with a collaborative autonomous driving simulator, i.e.,\nSimulation-of-Urban-Mobility (SUMO), to produce realistic vehicle control\nmodels and road conditions. The results show that with online status\npredictions, the channel occupancy is significantly reduced, while guaranteeing\nlow status recovery error. Then the framework is applied to two scenarios: a\nmulti-density platooning scenario, and a flight formation control scenario.\nSimulation results show that the scheme achieves better performance on the\nnetwork level, in terms of keeping the minimum safe distance in both vehicle\nplatooning and flight control.\n"}
{"abstract": "  The recently proposed action spotting task consists in finding the exact\ntimestamp in which an event occurs. This task fits particularly well for soccer\nvideos, where events correspond to salient actions strictly defined by soccer\nrules (a goal occurs when the ball crosses the goal line). In this paper, we\ndevise a lightweight and modular network for action spotting, which can\nsimultaneously predict the event label and its temporal offset using the same\nunderlying features. We enrich our model with two training strategies: the\nfirst one for data balancing and uniform sampling, the second for masking\nambiguous frames and keeping the most discriminative visual cues. When tested\non the SoccerNet dataset and using standard features, our full proposal exceeds\nthe current state of the art by 3 Average-mAP points. Additionally, it reaches\na gain of more than 10 Average-mAP points on the test set when fine-tuned in\ncombination with a strong 2D backbone.\n"}
{"abstract": "  Biomedical research is intensive in processing information in the previously\npublished papers. This motivated a lot of efforts to provide tools for text\nmining and information extraction from PDF documents over the past decade. The\n*nix (Unix/Linux) operating systems offer many tools for working with text\nfiles, however, very few such tools are available for processing the contents\nof PDF files. This paper reports our effort to develop shell script utilities\nfor *nix systems with the core functionality focused on viewing and searching\nmultiple PDF documents combining logical and regular expressions, and enabling\nmore reliable text extraction from PDF documents with subsequent manipulation\nof the resulting blocks of text. Furthermore, a procedure for extracting the\nmost frequently occurring multi-word phrases was devised and then demonstrated\non several scientific papers in life sciences. Our experiments revealed that\nthe procedure is surprisingly robust to deficiencies in text extraction and the\nactual scoring function used to rank the phrases in terms of their importance\nor relevance. The keyword relevance is strongly context dependent, the word\nstemming did not provide any recognizable advantage, and the stop-words should\nonly be removed from the beginning and the end of phrases. In addition, the\ndeveloped utilities were used to convert the list of acronyms and the index\nfrom a PDF e-book into a large list of biochemical terms which can be exploited\nin other text mining tasks. All shell scripts and data files are available in a\npublic repository named \\pp\\ on the Github. The key lesson learned in this work\nis that semi-automated methods combining the power of algorithms with the\ncapabilities of research experience are the most promising for improving the\nresearch efficiency.\n"}
{"abstract": "  Sampling-based model-predictive control (MPC) is a promising tool for\nfeedback control of robots with complex, non-smooth dynamics, and cost\nfunctions. However, the computationally demanding nature of sampling-based MPC\nalgorithms has been a key bottleneck in their application to high-dimensional\nrobotic manipulation problems in the real world. Previous methods have\naddressed this issue by running MPC in the task space while relying on a\nlow-level operational space controller for joint control. However, by not using\nthe joint space of the robot in the MPC formulation, existing methods cannot\ndirectly account for non-task space related constraints such as avoiding joint\nlimits, singular configurations, and link collisions. In this paper, we develop\na system for fast, joint space sampling-based MPC for manipulators that is\nefficiently parallelized using GPUs. Our approach can handle task and joint\nspace constraints while taking less than 8ms~(125Hz) to compute the next\ncontrol command. Further, our method can tightly integrate perception into the\ncontrol problem by utilizing learned cost functions from raw sensor data. We\nvalidate our approach by deploying it on a Franka Panda robot for a variety of\ndynamic manipulation tasks. We study the effect of different cost formulations\nand MPC parameters on the synthesized behavior and provide key insights that\npave the way for the application of sampling-based MPC for manipulators in a\nprincipled manner. We also provide highly optimized, open-source code to be\nused by the wider robot learning and control community. Videos of experiments\ncan be found at: https://sites.google.com/view/manipulation-mpc\n"}
{"abstract": "  Three state-of-the-art statistical ranking methods for forecasting football\nmatches are combined with several other predictors in a hybrid machine learning\nmodel. Namely an ability estimate for every team based on historic matches; an\nability estimate for every team based on bookmaker consensus; average\nplus-minus player ratings based on their individual performances in their home\nclubs and national teams; and further team covariates (e.g., market value, team\nstructure) and country-specific socio-economic factors (population, GDP). The\nproposed combined approach is used for learning the number of goals scored in\nthe matches from the four previous UEFA EUROs 2004-2016 and then applied to\ncurrent information to forecast the upcoming UEFA EURO 2020. Based on the\nresulting estimates, the tournament is simulated repeatedly and winning\nprobabilities are obtained for all teams. A random forest model favors the\ncurrent World Champion France with a winning probability of 14.8% before\nEngland (13.5%) and Spain (12.3%). Additionally, we provide survival\nprobabilities for all teams and at all tournament stages.\n"}
{"abstract": "  Deep neural networks have received considerable attention in clinical\nimaging, particularly with respect to the reduction of radiation risk. Lowering\nthe radiation dose by reducing the photon flux inevitably results in the\ndegradation of the scanned image quality. Thus, researchers have sought to\nexploit deep convolutional neural networks (DCNNs) to map low-quality, low-dose\nimages to higher-dose, higher-quality images thereby minimizing the associated\nradiation hazard. Conversely, computed tomography (CT) measurements of\ngeomaterials are not limited by the radiation dose. In contrast to the human\nbody, however, geomaterials may be comprised of high-density constituents\ncausing increased attenuation of the X-Rays. Consequently, higher dosage images\nare required to obtain an acceptable scan quality. The problem of prolonged\nacquisition times is particularly severe for micro-CT based scanning\ntechnologies. Depending on the sample size and exposure time settings, a single\nscan may require several hours to complete. This is of particular concern if\nphenomena with an exponential temperature dependency are to be elucidated. A\nprocess may happen too fast to be adequately captured by CT scanning. To\naddress the aforementioned issues, we apply DCNNs to improve the quality of\nrock CT images and reduce exposure times by more than 60\\%, simultaneously. We\nhighlight current results based on micro-CT derived datasets and apply transfer\nlearning to improve DCNN results without increasing training time. The approach\nis applicable to any computed tomography technology. Furthermore, we contrast\nthe performance of the DCNN trained by minimizing different loss functions such\nas mean squared error and structural similarity index.\n"}
{"abstract": "  We study theoretically spin current generation from a band insulator with PT\nsymmetry, which is associated with Zener tunneling in strong dc electric\nfields. Each band in this system is doubly degenerate with opposite spins, but\nspin rotational symmetry is not preserved in general. We consider the condition\nfor spin current generation in connection with the nature of the wave function,\nwhich ultimately depends on a geometric quantity known as the shift vector.\nFrom an analysis of a two-band model, we find that the shift vector is\nnecessary for spin current generation in a PT symmetric system. We also present\nzigzag chain models that have shift vectors, and confirm from numerical\ncalculations that a nonzero tunneling spin current occurs in spin-degenerate\nsystems.\n"}
{"abstract": "  We analyze the average fidelity (say, F) and the fidelity deviation (say, D)\nin noisy-channel quantum teleportation. Here, F represents how well\nteleportation is performed on average and D quantifies whether the\nteleportation is performed impartially on the given inputs, that is, the\ncondition of universality. Our analysis results prove that the achievable\nmaximum average fidelity ensures zero fidelity deviation, that is, perfect\nuniversality. This structural trait of teleportation is distinct from those of\nother limited-fidelity probabilistic quantum operations, for instance,\nuniversal-NOT or quantum cloning. This feature is confirmed again based on a\ntighter relationship between F and D in the qubit case. We then consider\nanother realistic noise model where F decreases and D increases due to\nimperfect control. To alleviate such deterioration, we propose a\nmachine-learning-based algorithm. We demonstrate by means of numerical\nsimulations that the proposed algorithm can stabilize the system. Notably, the\nrecovery process consists solely of the maximization of F, which reduces the\ncontrol time, thus leading to a faster cure cycle.\n"}
{"abstract": "  We address an outstanding problem that represents a critical roadblock in the\ndevelopment of the Majorana-based topological qubit using\nsemiconductor-superconductor hybrid structures: the quantitative\ncharacterization of disorder effects generated by the unintentional presence of\ncharge impurities within the hybrid device. Given that disorder can have\nfar-reaching consequences for the Majorana physics, but is intrinsically\ndifficult to probe experimentally in a hybrid structure, providing a\nquantitative theoretical description of disorder effects becomes essential. To\naccomplish this task, we develop a microscopic theory that (i) provides a\nquantitative characterization of the effective potential generated by a charge\nimpurity embedded inside a semiconductor wire proximity-coupled to a\nsuperconductor layer by solving self-consistently the associated\nthree-dimensional Schr\\\"odinger-Poisson problem, (ii) describes the low-energy\nphysics of the hybrid structure in the presence of s-wave superconductivity,\nspin-orbit coupling, Zeeman splitting, and disorder arising from multiple\ncharge impurities by using the results of (i) within a standard free fermion\napproach, and (iii) links the microscopic results to experimentally observable\nfeatures by generating tunneling differential conductance maps as function of\nthe control parameters (e.g., Zeeman field and chemical potential). We find\nthat charge impurities lead to serious complications regarding the realization\nand observation of Majorana zero modes, which have direct implications for the\ndevelopment of Majorana-based qubits. More importantly, our work provides a\nclear direction regarding what needs to be done for progress in the field,\nincluding specific materials quality and semiconductor purity targets that must\nbe achieved to create a topological qubit.\n"}
{"abstract": "  The Minkowski question-mark function $?(x)$ is a continuous strictly\nincreasing function defined on $[0,1]$ interval. It is well known fact that the\nderivative of this function, if exists, can take only two values: $0$ and\n$+\\infty$. It is also known that the value of the derivative $?'(x)$ at the\npoint $x=[0;a_1,a_2,\\ldots,a_t,\\ldots]$ is connected with the limit behavior of\nthe arithmetic mean $(a_1+a_2+\\ldots+a_t)/t$. Particularly, N. Moshchevitin and\nA. Dushistova showed that if $a_1+a_2+\\ldots+a_t<\\kappa_1 t$, where $\\kappa_1 =\n2\\log\\bigl({\\frac{1+\\sqrt{5}}{2}}\\bigr)/\\log{2}= 1.3884\\ldots$, then\n$?'(x)=+\\infty$. They also proved that the constant $\\kappa_1$ is\nnon-improvable. We consider a dual problem: how small can be the quantity\n$a_1+a_2+\\ldots+a_t-\\kappa_1 t$ if $?'(x)=0$? We obtain the non-improvable\nestimates of this quantity.\n"}
{"abstract": "  We sharpen to nearly optimal the known asymptotic and explicit bounds for the\nnumber of $\\mathbb{F}_q$-rational points on a geometrically irreducible\nhypersurface over a (large) finite field. The proof involves a Bertini-type\nprobabilistic combinatorial technique. Namely, we study the number of\n$\\mathbb{F}_q$-points on the intersection of the given hypersurface with a\nrandom plane.\n"}
{"abstract": "  Computational notebooks have become the tool of choice for many data\nscientists and practitioners for performing analyses and disseminating results.\nDespite their increasing popularity, the research community cannot yet count on\na large, curated dataset of computational notebooks. In this paper, we fill\nthis gap by introducing KGTorrent, a dataset of Python Jupyter notebooks with\nrich metadata retrieved from Kaggle, a platform hosting data science\ncompetitions for learners and practitioners with any levels of expertise. We\ndescribe how we built KGTorrent, and provide instructions on how to use it and\nrefresh the collection to keep it up to date. Our vision is that the research\ncommunity will use KGTorrent to study how data scientists, especially\npractitioners, use Jupyter Notebook in the wild and identify potential\nshortcomings to inform the design of its future extensions.\n"}
{"abstract": "  This work studies the role of inter-user device-to-device (D2D) cooperation\nfor improving physical-layer secret communication in multi-user downlink\nsystems. It is assumed that there are out-of-band D2D channels, on each of\nwhich a selected legitimate user transmits an amplified version of the received\ndownlink signal to other legitimate users. A key technical challenge for\ndesigning such systems is that eavesdroppers can overhear downlink as well as\nD2D cooperation signals. We tackle the problem of jointly optimizing the\ndownlink precoding, artificial noise covariance, and amplification coefficients\nthat maximize the minimum rate. An iterative alternating optimization algorithm\nis proposed based on the matrix fractional programming. Numerical results\nconfirm the performance gains of the proposed D2D cooperation scheme compared\nto benchmark secret communication schemes.\n"}
{"abstract": "  In this paper, we extend the reinterpreted discrete fracture model for flow\nsimulation of fractured porous media containing flow blocking barriers on\nnon-conforming meshes. The methodology of the approach is to modify the\ntraditional Darcy's law into the hybrid-dimensional Darcy's law where fractures\nand barriers are represented as Dirac-delta functions contained in the\npermeability tensor and resistance tensor, respectively. As a natural extension\nof the reinterpreted discrete fracture model for highly conductive fractures,\nthis model is able to account for the influence of both highly conductive\nfractures and blocking barriers accurately on non-conforming meshes. The local\ndiscontinuous Galerkin (LDG) method is employed to accommodate the form of the\nhybrid-dimensional Darcy's law and the nature of the pressure/flux\ndiscontinuity. The performance of the model is demonstrated by several\nnumerical tests.\n"}
{"abstract": "  In this paper, we consider several possible ways to set up Heterogeneous\nMultiscale Methods for the Landau-Lifshitz equation with a highly oscillatory\ndiffusion coefficient, which can be seen as a means to modeling rapidly varying\nferromagnetic materials. We then prove estimates for the errors introduced when\napproximating the relevant quantity in each of the models given a periodic\nproblem, using averaging in time and space of the solution to a corresponding\nmicro problem. In our setup, the Landau-Lifshitz equation with highly\noscillatory coefficient is chosen as the micro problem for all models. We then\nshow that the averaging errors only depend on $\\varepsilon$, the size of the\nmicroscopic oscillations, as well as the size of the averaging domain in time\nand space and the choice of averaging kernels.\n"}
{"abstract": "  Reusable decoys offer a cost-effective alternative to the single-use hardware\ncommonly applied to protect surface assets from threats. Such decoys portray\nfake assets to lure threats away from the true asset. To deceive a threat, a\ndecoy first has to position itself such that it can break the radar lock.\nConsidering multiple simultaneous threats, this paper introduces an approach\nfor controlling multiple decoys to minimise the time required to break the\nlocks of all the threats. The method includes the optimal allocation of one\ndecoy to every threat with an assignment procedure that provides local position\nconstraints to guarantee collision avoidance and thereby decouples the control\nof the decoys. A crude model of a decoy with uncertainty is considered for\nmotion planning. The task of a decoy reaching a state in which the lock of the\nassigned threat can be broken is formulated as a temporal logic specification.\nTo this end, the requirements to complete the task are modelled as time-varying\nset-membership constraints. The temporal and logical combination of the\nconstraints is encoded in a mixed-integer optimisation problem. To demonstrate\nthe results a simulated case study is provided.\n"}
{"abstract": "  In this paper, we present Talaria, a novel permissioned blockchain simulator\nthat supports numerous protocols and use cases, most notably in supply chain\nmanagement. Talaria extends the capability of BlockSim, an existing blockchain\nsimulator, to include permissioned blockchains and serves as a foundation for\nfurther private blockchain assessment. Talaria is designed with both practical\nByzantine Fault Tolerance (pBFT) and simplified version of Proof-of-Authority\nconsensus protocols, but can be revised to include other permissioned protocols\nwithin its modular framework. Moreover, Talaria is able to simulate different\ntypes of malicious authorities and a variable daily transaction load at each\nnode. In using Talaria, business practitioners and policy planners have an\nopportunity to measure, evaluate, and adapt a range of blockchain solutions for\ncommercial operations.\n"}
{"abstract": "  Let L be a general length function for modules over a Noetherian ring R. We\nuse L to define Hilbert series and polynomials for R[X]-modules. The leading\nterm of any such polynomial is an invariant of R[X]-modules, which refines the\nalgebraic entropy.\n"}
{"abstract": "  We study quasi periodic and frequency locked states that can occur in a\nsinusoidally driven linear harmonic oscillator in the special relativistic\nregime. We show how the shift in natural frequency of the oscillator with\nincreasing relativistic effects leads to frequency locking or quasi periodicity\nand the chaotic states that arise due to the increasing non linearity. We find\nthe same system can have multi-stable states in the presence of small damping.\nWe also report an enhancement of chaos in the relativistic Henon-Heiles system.\n"}
{"abstract": "  High-resolution manometry (HRM) is the primary procedure used to diagnose\nesophageal motility disorders. Its interpretation and classification includes\nan initial evaluation of swallow-level outcomes and then derivation of a\nstudy-level diagnosis based on Chicago Classification (CC), using a tree-like\nalgorithm. This diagnostic approach on motility disordered using HRM was\nmirrored using a multi-stage modeling framework developed using a combination\nof various machine learning approaches. Specifically, the framework includes\ndeep-learning models at the swallow-level stage and feature-based machine\nlearning models at the study-level stage. In the swallow-level stage, three\nmodels based on convolutional neural networks (CNNs) were developed to predict\nswallow type, swallow pressurization, and integrated relaxation pressure (IRP).\nAt the study-level stage, model selection from families of the\nexpert-knowledge-based rule models, xgboost models and artificial neural\nnetwork(ANN) models were conducted, with the latter two model designed and\naugmented with motivation from the export knowledge. A simple model-agnostic\nstrategy of model balancing motivated by Bayesian principles was utilized,\nwhich gave rise to model averaging weighted by precision scores. The averaged\n(blended) models and individual models were compared and evaluated, of which\nthe best performance on test dataset is 0.81 in top-1 prediction, 0.92 in top-2\npredictions. This is the first artificial-intelligence-style model to\nautomatically predict CC diagnosis of HRM study from raw multi-swallow data.\nMoreover, the proposed modeling framework could be easily extended to\nmulti-modal tasks, such as diagnosis of esophageal patients based on clinical\ndata from both HRM and functional luminal imaging probe panometry (FLIP).\n"}
{"abstract": "  Open clusters are central elements of our understanding of the Galactic disk\nevolution, as an accurate determination of their parameters leads to an\nunbiased picture of our Galaxy's structure. Extending the analysis towards\nfainter magnitudes in cluster sequences has a significant impact on the derived\nfundamental parameters, such as extinction and total mass. We perform a\nhomogeneous analysis of six open stellar clusters in the Galactic disk using\nkinematic and photometric information from the Gaia DR2 and VVV surveys:\nNGC6067, NGC6259, NGC4815, Pismis18, Trumpler23, and Trumpler20. We implement\ntwo coarse-to-fine characterization methods: first, we employ Gaussian mixture\nmodels to tag fields around each open cluster in the proper motion space, and\nthen we apply an unsupervised machine learning method to make the membership\nassignment to each cluster. For the studied clusters, with ages in the\n$\\sim$120-1900 Myr range, we report an increase of $\\sim$45 % new member\ncandidates on average in our sample. The data-driven selection approach of\ncluster members makes our catalog a valuable resource for testing stellar\nevolutionary models and for assessing the cluster low-to-intermediate mass\npopulations. This study is the first of a series intended to homogeneously\nreveal open cluster near-infrared sequences.\n"}
{"abstract": "  In this article, we propose the fractional lower order covariance method\n(FLOC) for estimating the parameters of vector autoregressive process (VAR) of\norder $p$, $p\\geq 1$ with symmetric stable noise. Further, we show the\nefficiency, accuracy, and simplicity of our methods through Monte-Carlo\nsimulation.\n"}
{"abstract": "  With the rise of AI in SE, researchers have shown how AI can be applied to\nassist software developers in a wide variety of activities. However, it has not\nbeen accompanied by a complementary increase in labelled datasets, which is\nrequired in many supervised learning methods. Several studies have been using\ncrowdsourcing platforms to collect labelled training data in recent years.\nHowever, research has shown that the quality of labelled data is unstable due\nto participant bias, knowledge variance, and task difficulty. Thus, we present\nCodeLabeller, a web-based tool that aims to provide a more efficient approach\nin handling the process of labelling Java source files at scale by improving\nthe data collection process throughout, and improving the degree of reliability\nof responses by requiring each labeller to attach a confidence rating to each\nof their responses. We test CodeLabeller by constructing a corpus of over a\nthousand source files obtained from a large collection of opensource Java\nprojects, and labelling each Java source file with their respective design\npatterns and summaries. Apart from assisting researchers to crowdsource a\nlabelled dataset, the tool has practical applicability in software engineering\neducation and assists in building expert ratings for software artefacts. This\npaper discusses the motivation behind the creation of CodeLabeller, the\nintended users, a tool demonstration and its UI, its implementation, benefits,\nand lastly, the evaluation through a user study and in-practice usage.\n"}
{"abstract": "  We report the identification of symmetry-enforced nodal planes (NPs) in CoSi\nproviding the missing topological charges in an entire network of\nband-crossings comprising in addition multifold degeneracies and Weyl points,\nsuch that the fermion doubling theorem is satisfied. In our study we have\ncombined measurements of Shubnikov-de Haas (SdH) oscillations in CoSi with\nmaterial-specific calculations of the electronic structure and Berry curvature,\nas well as a general analysis of the band topology of space group (SG) 198. The\nobservation of two nearly dispersionless SdH frequency branches provides\nunambiguous evidence of four Fermi surface sheets at the R point that reflect\nthe symmetry-enforced orthogonality of the underlying wave functions at the\nintersections with the NPs. Hence, irrespective of the spin-orbit coupling\nstrength, SG198 features always six- and fourfold degenerate crossings at R and\n$\\Gamma$ that are intimately connected to the topological charges distributed\nacross the network.\n"}
{"abstract": "  Context: Software startups are an essential source of innovation and\nsoftware-intensive products. The need to understand product development in\nstartups and to provide relevant support are highlighted in software research.\nWhile state-of-the-art literature reveals how startups develop their software,\nthe reasons why they adopt these activities are underexplored. Objective: This\nstudy investigates the tactics behind software engineering (SE) activities by\nanalyzing key engineering events during startup journeys. We explore how\nentrepreneurial mindsets may be associated with SE knowledge areas and with\neach startup case. Method: Our theoretical foundation is based on causation and\neffectuation models. We conducted semi-structured interviews with 40 software\nstartups. We used two-round open coding and thematic analysis to describe and\nidentify entrepreneurial software development patterns. Additionally, we\ncalculated an effectuation index for each startup case. Results: We identified\n621 events merged into 32 codes of entrepreneurial logic in SE from the sample.\nWe found a systemic occurrence of the logic in all areas of SE activities.\nMinimum Viable Product (MVP), Technical Debt (TD), and Customer Involvement\n(CI) tend to be associated with effectual logic, while testing activities at\ndifferent levels are associated with causal logic. The effectuation index\nrevealed that startups are either effectuation-driven or mixed-logics-driven.\nConclusions: Software startups fall into two types that differentiate between\nhow traditional SE approaches may apply to them. Effectuation seems the most\nrelevant and essential model for explaining and developing suitable SE\npractices for software startups.\n"}
{"abstract": "  Deep neural networks (DNNs) are known vulnerable to backdoor attacks, a\ntraining time attack that injects a trigger pattern into a small proportion of\ntraining data so as to control the model's prediction at the test time.\nBackdoor attacks are notably dangerous since they do not affect the model's\nperformance on clean examples, yet can fool the model to make incorrect\nprediction whenever the trigger pattern appears during testing. In this paper,\nwe propose a novel defense framework Neural Attention Distillation (NAD) to\nerase backdoor triggers from backdoored DNNs. NAD utilizes a teacher network to\nguide the finetuning of the backdoored student network on a small clean subset\nof data such that the intermediate-layer attention of the student network\naligns with that of the teacher network. The teacher network can be obtained by\nan independent finetuning process on the same clean subset. We empirically\nshow, against 6 state-of-the-art backdoor attacks, NAD can effectively erase\nthe backdoor triggers using only 5\\% clean training data without causing\nobvious performance degradation on clean examples. Code is available in\nhttps://github.com/bboylyg/NAD.\n"}
{"abstract": "  The rapid deployment of distributed energy resources (DERs) in distribution\nnetworks has brought challenges to balance the system and stabilize frequency.\nDERs have the ability to provide frequency regulation; however, existing\ndynamic frequency simulation tools-which were developed mainly for the\ntransmission system-lack the capability to simulate distribution network\ndynamics with high penetrations of DERs. Although electromagnetic transient\n(EMT) simulation tools can simulate distribution network dynamics, the\ncomputation efficiency limits their use for large-scale\ntransmission-and-distribution (T&D) simulations. This paper presents an\nefficient T&D dynamic frequency co-simulation framework for DER frequency\nresponse based on the HELICS platform and existing off-the-shelf simulators.\nThe challenge of synchronizing frequency between the transmission network and\nDERs hosted in the distribution network is approached by detailed modeling of\nDERs in frequency dynamic models while DER phasor models are also preserved in\nthe distribution networks. Thereby, local voltage constraints can be respected\nwhen dispatching the DER power for frequency response. The DER frequency\nresponses (primary and secondary)-are simulated in case studies to validate the\nproposed framework. Lastly, fault-induced delayed voltage recovery (FIDVR)\nevent of a large system is presented to demonstrate the efficiency and\neffectiveness of the overall framework.\n"}
{"abstract": "  This paper describes a technique for the autonomous mission planning of\nunmanned aerial system swarms. Given a swarm operating in a known area, a\ncentral command system generates measurements from the swarm. If those\nmeasurements indicate changes to the mission situation such as target movement,\nthe swarm planning is updated to reflect the new situation and guidance updates\nare broadcast to the swarm. The primary algorithms featured in this work are A*\npathfinding and the Generalized Labeled Multi-Bernoulli multi-target tracking\nmethod.\n"}
{"abstract": "  Recently, experiments have been reported where researchers were able to\nperform high dynamic range (HDR) tomography in a heuristic fashion, by fusing\nmultiple tomographic projections. This approach to HDR tomography has been\ninspired by HDR photography and inherits the same disadvantages. Taking a\ncomputational imaging approach to the HDR tomography problem, we here suggest a\nnew model based on the Modulo Radon Transform (MRT), which we rigorously\nintroduce and analyze. By harnessing a joint design between hardware and\nalgorithms, we present a single-shot HDR tomography approach, which to our\nknowledge, is the only approach that is backed by mathematical guarantees.\n  On the hardware front, instead of recording the Radon Transform projections\nthat my potentially saturate, we propose to measure modulo values of the same.\nThis ensures that the HDR measurements are folded into a lower dynamic range.\nOn the algorithmic front, our recovery algorithms reconstruct the HDR images\nfrom folded measurements. Beyond mathematical aspects such as injectivity and\ninversion of the MRT for different scenarios including band-limited and\napproximately compactly supported images, we also provide a first\nproof-of-concept demonstration. To do so, we implement MRT by experimentally\nfolding tomographic measurements available as an open source data set using our\ncustom designed modulo hardware. Our reconstruction clearly shows the\nadvantages of our approach for experimental data. In this way, our MRT based\nsolution paves a path for HDR acquisition in a number of related imaging\nproblems.\n"}
{"abstract": "  Due to the different photosensitive properties of infrared and visible light,\nthe registered RGB-T image pairs shot in the same scene exhibit quite different\ncharacteristics. This paper proposes a siamese infrared and visible light\nfusion Network (SiamIVFN) for RBG-T image-based tracking. SiamIVFN contains two\nmain subnetworks: a complementary-feature-fusion network (CFFN) and a\ncontribution-aggregation network (CAN). CFFN utilizes a two-stream multilayer\nconvolutional structure whose filters for each layer are partially coupled to\nfuse the features extracted from infrared images and visible light images. CFFN\nis a feature-level fusion network, which can cope with the misalignment of the\nRGB-T image pairs. Through adaptively calculating the contributions of infrared\nand visible light features obtained from CFFN, CAN makes the tracker robust\nunder various light conditions. Experiments on two RGB-T tracking benchmark\ndatasets demonstrate that the proposed SiamIVFN has achieved state-of-the-art\nperformance. The tracking speed of SiamIVFN is 147.6FPS, the current fastest\nRGB-T fusion tracker.\n"}
{"abstract": "  Deep-learning (DL) has emerged as a powerful machine-learning technique for\nseveral classic problems encountered in generic wireless communications.\nSpecifically, random Fourier Features (RFF) based deep-learning has emerged as\nan attractive solution for several machine-learning problems; yet there is a\nlacuna of rigorous results to justify the viability of RFF based DL-algorithms\nin general. To address this gap, we attempt to analytically quantify the\nviability of RFF based DL. Precisely, in this paper, analytical proofs are\npresented demonstrating that RFF based DL architectures have lower\napproximation-error and probability of misclassification as compared to\nclassical DL architectures. In addition, a new distribution-dependent RFF is\nproposed to facilitate DL architectures with low training-complexity. Through\ncomputer simulations, the practical application of the presented analytical\nresults and the proposed distribution-dependent RFF, are depicted for various\nmachine-learning problems encountered in next-generation communication systems\nsuch as: a) line of sight (LOS)/non-line of sight (NLOS) classification, and b)\nmessage-passing based detection of low-density parity check codes (LDPC) codes\nover nonlinear visible light communication (VLC) channels. Especially in the\nlow training-data regime, the presented simulations show that significant\nperformance gains are achieved when utilizing RFF maps of observations. Lastly,\nin all the presented simulations, it is observed that the proposed\ndistribution-dependent RFFs significantly outperform RFFs, which make them\nuseful for potential machine-learning/DL based applications in the context of\nnext-generation communication systems.\n"}
{"abstract": "  We present a projection-based framework for solving a\nthermodynamically-consistent Cahn-Hilliard Navier-Stokes system that models\ntwo-phase flows. In this work we extend the fully implicit method presented in\nKhanwale et al. [{\\it A fully-coupled framework for solving Cahn-Hilliard\nNavier-Stokes equations: Second-order, energy-stable numerical methods on\nadaptive octree based meshes.}, arXiv:2009.06628 (2020)], to a block iterative\nhybrid method. We use a projection-based semi-implicit time discretization for\nthe Navier-Stokes and a fully-implicit time discretization for the\nCahn-Hilliard equation. We use a conforming continuous Galerkin (cG) finite\nelement method in space equipped with a residual-based variational multiscale\n(RBVMS) formulation. Pressure is decoupled using a projection step, which\nresults in two linear positive semi-definite systems for velocity and pressure,\ninstead of the saddle point system of a pressure-stabilized method. All the\nlinear systems are solved using an efficient and scalable algebraic multigrid\n(AMG) method. We deploy this approach on a massively parallel numerical\nimplementation using parallel octree-based adaptive meshes. The overall\napproach allows the use of relatively large time steps with much faster\ntime-to-solve. We present comprehensive numerical experiments showing detailed\ncomparisons with results from the literature for canonical cases, including the\nsingle bubble rise and Rayleigh-Taylor instability.\n"}
{"abstract": "  We present a novel task scheduling scheme for accelerating computational\napplications involving distributed iterative processes that are executed on\nnetworked computing resources. Such an application consists of multiple tasks,\neach of which outputs data at each iteration to be processed by neighboring\ntasks; these dependencies between the tasks can be represented as a directed\ngraph. We first mathematically formulate the problem as a Binary Quadratic\nProgram (BQP), accounting for both computation and communication costs. We show\nthat the problem is NP-hard. We then relax the problem as a Semi-Definite\nProgram (SDP) and utilize a randomized rounding technique based on sampling\nfrom a suitably-formulated multi-variate Gaussian distribution. Furthermore, we\nderive the expected value of bottleneck time. Finally, we apply our proposed\nscheme on gossip-based federated learning as an application of iterative\nprocesses. Through numerical evaluations on the MNIST and CIFAR-10 datasets, we\nshow that our proposed approach outperforms well-known scheduling techniques\nfrom distributed computing. In particular, for arbitrary settings, we show that\nit reduces bottleneck time by $91\\%$ compared to HEFT and $84\\%$ compared to\nthroughput HEFT.\n"}
{"abstract": "  This work focuses on complete 3D facial geometry prediction, including 3D\nfacial alignment via 3D face modeling and face orientation estimation using the\nproposed multi-task, multi-modal, and multi-representation landmark refinement\nnetwork (M$^3$-LRN). Our focus is on the important facial attributes, 3D\nlandmarks, and we fully utilize their embedded information to guide 3D facial\ngeometry learning. We first propose a multi-modal and multi-representation\nfeature aggregation for landmark refinement. Next, we are the first to study\n3DMM regression from sparse 3D landmarks and utilize multi-representation\nadvantage to attain better geometry prediction. We attain the state of the art\nfrom extensive experiments on all tasks of learning 3D facial geometry. We\nclosely validate contributions of each modality and representation. Our results\nare robust across cropped faces, underwater scenarios, and extreme poses.\nSpecially we adopt only simple and widely used network operations in M$^3$-LRN\nand attain a near 20\\% improvement on face orientation estimation over the\ncurrent best performance. See our project page here.\n"}
{"abstract": "  Major cause of midvehicle collision is due to the distraction of drivers in\nboth the Front and rear-end vehicle witnessed in dense traffic and high speed\nroad conditions. In view of this scenario, a crash detection and collision\navoidance algorithm coined as Midvehicle Collision Detection and Avoidance\nSystem (MCDAS) is proposed to evade the possible crash at both ends of the host\nvehicle. The method based upon Constant Velocity (CV) model specifically,\naddresses two scenarios, the first scenario encompasses two sub-scenario\nnamely, a) A rear-end collision avoidance mechanism that accelerates the host\nvehicle under no front-end vehicle condition and b) Curvilinear motion based on\nfront and host vehicle offset (position), whilst, the other scenario deals with\nparallel parking issues. The offset based curvilinear motion of the host\nvehicle plays a vital role in threat avoidance from the front-end vehicle. A\ndesired curvilinear strategy on left and right sides is achieved by the host\nvehicle with concern of possible CV to avoid both end collisions. In this\nmethodology, path constraint is applicable for both scenarios with required\ndirection. Monte Carlo analysis of MCDAS covering vehicle kinematics\ndemonstrated acute discrimination with consistent performance for the collision\nvalidated on simulated with real-time data.\n"}
{"abstract": "  This paper presents abstract harmonic analysis foundations for structure of\ncovariant function algebras of invariant characters of normal subgroups.\nSuppose that $G$ is a locally compact group and $N$ is a closed normal subgroup\nof $G$. Let $\\xi:N\\to\\mathbb{T}$ be a continuous $G$-invariant character, $1\\le\np<\\infty$, and $L_\\xi^p(G,N)$ be the $L^p$-space of all covariant functions of\n$\\xi$ on $G$. We study structure of covariant convolution in $L^p_\\xi(G,N)$. It\nis proved that $L^1_\\xi(G,N)$ is a Banach $*$-algebra and $L^p_\\xi(G,N)$ is a\nBanach $L^1_\\xi(G,N)$-module. We then investigate the theory of covariant\nconvolutions for the case of characters of canonical normal subgroups in\nsemi-direct product groups. The paper is concluded by realization of the theory\nin the case of different examples.\n"}
{"abstract": "  The Martian Moons Exploration (MMX) spacecraft is a JAXA mission to Mars and\nits moons Phobos and Deimos. MMX will carry the Circum-Martian Dust Monitor\n(CMDM) which is a newly developed light-weight ($\\mathrm{650\\,g}$) large area\n($\\mathrm{1\\,m^2}$) dust impact detector. Cometary meteoroid streams (also\nreferred to as trails) exist along the orbits of comets, forming fine\nstructures of the interplanetary dust cloud. The streams consist predominantly\nof the largest cometary particles (with sizes of approximately\n$\\mathrm{100\\,\\mu m}$ to 1~cm) which are ejected at low speeds and remain very\nclose to the comet orbit for several revolutions around the Sun. The\nInterplanetary Meteoroid Environment for eXploration (IMEX) dust streams in\nspace model is a new and recently published universal model for cometary\nmeteoroid streams in the inner Solar System. We use IMEX to study the detection\nconditions of cometary dust stream particles with CMDM during the MMX mission\nin the time period 2024 to 2028. The model predicts traverses of 12 cometary\nmeteoroid streams with fluxes of $\\mathrm{100\\,\\mu m}$ and bigger particles of\nat least $\\mathrm{10^{-3}\\,m^{-2}\\,day^{-1}}$ during a total time period of\napproximately 90~days. The highest flux of $\\mathrm{0.15\\,m^{-2}\\,day^{-1}}$ is\npredicted for comet 114P/Wiseman-Skiff in October 2026. With its large\ndetection area and high sensitivity CMDM will be able to detect cometary\nmeteoroid streams en route to Phobos. Our simulation results for the Mars\norbital phase of MMX also predict the occurrence of meteor showers in the\nMartian atmosphere which may be observable from the Martian surface with\ncameras on board landers or rovers. Finally, the IMEX model can be used to\nstudy the impact hazards imposed by meteoroid impacts on to large-area\nspacecraft structures that will be particularly necessary for crewed deep space\nmissions.\n"}
{"abstract": "  Deep learning model (primarily convolutional networks and LSTM) for time\nseries classification has been studied broadly by the community with the wide\napplications in different domains like healthcare, finance, industrial\nengineering and IoT. Meanwhile, Transformer Networks recently achieved frontier\nperformance on various natural language processing and computer vision tasks.\nIn this work, we explored a simple extension of the current Transformer\nNetworks with gating, named Gated Transformer Networks (GTN) for the\nmultivariate time series classification problem. With the gating that merges\ntwo towers of Transformer which model the channel-wise and step-wise\ncorrelations respectively, we show how GTN is naturally and effectively\nsuitable for the multivariate time series classification task. We conduct\ncomprehensive experiments on thirteen dataset with full ablation study. Our\nresults show that GTN is able to achieve competing results with current\nstate-of-the-art deep learning models. We also explored the attention map for\nthe natural interpretability of GTN on time series modeling. Our preliminary\nresults provide a strong baseline for the Transformer Networks on multivariate\ntime series classification task and grounds the foundation for future research.\n"}
{"abstract": "  Depression is one of the most common mental illness problems, and the\nsymptoms shown by patients are not consistent, making it difficult to diagnose\nin the process of clinical practice and pathological research. Although\nresearchers hope that artificial intelligence can contribute to the diagnosis\nand treatment of depression, the traditional centralized machine learning needs\nto aggregate patient data, and the data privacy of patients with mental illness\nneeds to be strictly confidential, which hinders machine learning algorithms\nclinical application. To solve the problem of privacy of the medical history of\npatients with depression, we implement federated learning to analyze and\ndiagnose depression. First, we propose a general multi-view federated learning\nframework using multi-source data, which can extend any traditional machine\nlearning model to support federated learning across different institutions or\nparties. Secondly, we adopt late fusion methods to solve the problem of\ninconsistent time series of multi-view data. Finally, we compare the federated\nframework with other cooperative learning frameworks in performance and discuss\nthe related results.\n"}
{"abstract": "  While the theory of labeled well-quasi-order has received significant\nattention in the graph setting, it has not yet been considered in the context\nof permutation patterns. We initiate this study here, and using labeled\nwell-quasi-order are able to subsume and extend all of the well-quasi-order\nresults in the permutation patterns literature. Connections to the graph\nsetting are emphasized throughout. In particular, we establish that a\npermutation class is labeled well-quasi-ordered if and only if its\ncorresponding graph class is also labeled well-quasi-ordered.\n"}
{"abstract": "  This paper proposes a novel framework for addressing the challenge of\nautonomous overtaking and obstacle avoidance, which incorporates the overtaking\npath planning into Gaussian Process-based model predictive control (GPMPC).\nCompared with the conventional control strategies, this approach has two main\nadvantages. Firstly, combining Gaussian Process (GP) regression with a nominal\nmodel allows for learning from model mismatch and unmodeled dynamics, which\nenhances a simple model and delivers significantly better results. Due to the\napproximation for propagating uncertainties, we can furthermore satisfy the\nconstraints and thereby safety of the vehicle is ensured. Secondly, we convert\nthe geometric relationship between the ego vehicle and other obstacle vehicles\ninto the constraints. Without relying on a higherlevel path planner, this\napproach substantially reduces the computational burden. In addition, we\ntransform the state constraints under the model predictive control (MPC)\nframework into a soft constraint and incorporate it as relaxed barrier function\ninto the cost function, which makes the optimizer more efficient. Simulation\nresults reveal the usefulness of the proposed approach.\n"}
{"abstract": "  Given a positive integer $ r $, the $ r $-color size-Ramsey number of a graph\n$ H $, denoted by $ \\hat{R}(H, r) $, is the smallest integer $ m $ for which\nthere exists a graph $ G $ with $ m $ edges such that, in any edge coloring of\n$ G $ with $ r $ colors, $G$ contains a monochromatic copy of $ H $. Haxell,\nKohayakawa and \\L uczak showed that the size-Ramsey number of a cycle $ C_n $\nis linear in $ n $ i.e. $ \\hat{R}(C_n, r) \\leq c_rn $, for some constant $ c_r\n$. Their proof, however, is based on the Szemer\\'edi's regularity lemma and so\nno specific constant $ c_r $ is known. Javadi, Khoeini, Omidi and Pokrovskiy\ngave an alternative proof for this result which avoids using of the regularity\nlemma. Indeed, they proved that if $ n $ is even, then $ c_r $ is exponential\nin $ r $ and if $ n $ is odd, then $ c_r $ is doubly exponential in $ r $.\n  \\noindent In this paper, we improve the bound $c_r$ and prove that $c_r$ is\npolynomial in $r$ when $n$ is even and is exponential in $r$ when $n$ is odd.\nWe also prove that in the latter case, it cannot be improved to a polynomial\nbound in $r$. More precisely, we prove that there are some positive constants\n$c_1,c_2$ such that for every even integer $n$, we have $c_1r^2n\\leq\n\\hat{R}(C_n,r)\\leq c_2r^{120}(\\log^2 r)n$ and for every odd integer $n$, we\nhave\n  $c_1 2^{r}n \\leq \\hat{R}(C_n, r)\\leq c_22^{16 r^2+2\\log r}n $.\n"}
{"abstract": "  Tearing modes in tokamaks typically rotate while small and then lock at a\nfixed location when larger. Research on present-day devices has focused almost\nexclusively on stabilisation of rotating modes, as it has been considered\nimperative to avoid locked modes. However, in larger devices, such as those\ncontemplated for tokamak reactors, the locking occurs at a smaller island size,\nand the island can be safely stabilised after locking. The stabilisation of\nsmall locked modes can be performed at lower wave power and broader deposition\ncompared to rotating islands. On large devices, it thus becomes surprisingly\nadvantageous to allow the mode to grow and lock naturally before stabilising\nit. Calculations indicate that the ITER international megaproject would be best\nstabilised through this approach.\n"}
{"abstract": "  We present results from high signal-to-noise optical spectropolarimetric\nobservations of the Seyfert 1 galaxies NGC783 and Mrk509 in the 3500-7000 A\nrange. We find complex structure in the polarized emission for both objects. In\nparticular, Position Angle (PA) changes across the Balmer lines show a\ndistinctive 'M'-shaped profile that had not been observed in this detail\nbefore, but could represent a common trait in Seyfert 1 galaxies. In fact,\nwhile this shape is observed in all Balmer lines in NGC3783, Mrk509 transitions\ninto a 'M'-shaped PA profile for higher transitions lines. We have modeled the\nobserved profiles using the STOKES radiative transfer code and assuming that\nthe scattering region is co-spatial with the BLR and outflowing. The results\ngive compelling new evidence for the presence of nuclear winds in these two\nSeyfert 1 galaxies.\n"}
{"abstract": "  A core goal of functional neuroimaging is to study how the environment is\nprocessed in the brain. The mainstream paradigm involves concurrently measuring\na broad spectrum of brain responses to a small set of environmental features\npreselected with reference to previous studies or a theoretical framework. As a\ncomplement, we invert this approach by allowing the investigator to record the\nmodulation of a preselected brain response by a broad spectrum of environmental\nfeatures. Our approach is optimal when theoretical frameworks or previous\nempirical data are impoverished. By using a prespecified closed-loop design,\nthe approach addresses fundamental challenges of reproducibility and\ngeneralisability in brain research. These conditions are particularly acute\nwhen studying the developing brain, where our theories based on adult brain\nfunction may fundamentally misrepresent the topography of infant cognition and\nwhere there are substantial practical challenges to data acquisition. Our\nmethodology employs machine learning to map modulation of a neural feature\nacross a space of experimental stimuli. Our method collects, processes and\nanalyses EEG brain data in real-time; and uses a neuro-adaptive Bayesian\noptimisation algorithm to adjust the stimulus presented depending on the prior\nsamples of a given participant. Unsampled stimuli can be interpolated by\nfitting a Gaussian process regression along the dataset. We show that our\nmethod can automatically identify the face of the infant's mother through\nonline recording of their Nc brain response to a face continuum. We can\nretrieve model statistics of individualised responses for each participant,\nopening the door for early identification of atypical development. This\napproach has substantial potential in infancy research and beyond for improving\npower and generalisability of mapping the individual cognitive topography of\nbrain function.\n"}
{"abstract": "  In this comment on the article by Alonso-Serrano and Liska (arXiv:\n2008.04805) a formal resemblance between their expression of curvature scale\nand the scale dependent on the matter energy-momentum ambiguity of Finkelstein\net al (JMP, 2001) is pointed out. Physical significance of this observation is\nalso discussed.\n"}
{"abstract": "  We numerically and theoretically investigate how the softness of particles\naffects the rheology of sheared dilute granular gases. We find that the kinetic\ntheory predicts the deviation of the flow curve from the Bagnold scaling, and\nit works well below a certain shear rate when we compare with the simulation\nresults. It is also found that there is no theoretical solution above this\nshear rate, which is because the energy loss due to inelastic collisions cannot\nbe balanced with the energy injection by the shear.\n"}
{"abstract": "  We present an intimate connection among the following fields:\n  (a) distributed local algorithms: coming from the area of computer science,\n  (b) finitary factors of iid processes: coming from the area of analysis of\nrandomized processes,\n  (c) descriptive combinatorics: coming from the area of combinatorics and\nmeasure theory.\n  In particular, we study locally checkable labellings in grid graphs from all\nthree perspectives. Most of our results are for the perspective (b) where we\nprove time hierarchy theorems akin to those known in the field (a) [Chang,\nPettie FOCS 2017]. This approach that borrows techniques from the fields (a)\nand (c) implies a number of results about possible complexities of finitary\nfactor solutions. Among others, it answers three open questions of [Holroyd et\nal. Annals of Prob. 2017] or the more general question of [Brandt et al. PODC\n2017] who asked for a formal connection between the fields (a) and (b). In\ngeneral, we hope that our treatment will help to view all three perspectives as\na part of a common theory of locality, in which we follow the insightful paper\nof [Bernshteyn 2020+] .\n"}
{"abstract": "  We compute the low-energy excitation spectrum and the dynamical spin\nstructure factor of the Kitaev-Heisenberg-Gamma model through a variational\napproach based on the exact fractionalized excitations of the pure Kitaev\nhoneycomb model. This novel approach reveals the physical reason for the\nasymmetric stability of the Kitaev spin liquid phases around the ferromagnetic\nand antiferromagnetic Kitaev limits. Moreover, we demonstrate that the\nfractionalized excitations form bound states in specific regions of each Kitaev\nspin liquid phase and that certain phase transitions induced by Heisenberg and\nGamma interactions are driven by the condensation of such a bound state.\nRemarkably, this bound state appears as a sharp mode in the dynamical spin\nstructure factor, while its condensation patterns at the appropriate phase\ntransitions provide a simple explanation for the magnetically ordered phases\nsurrounding each Kitaev spin liquid phase.\n"}
{"abstract": "  Finding Bertram's optimal trading strategy for a pair of cointegrated assets\nfollowing the Ornstein--Uhlenbeck price difference process can be formulated as\nan unconstrained convex optimization problem for maximization of expected\nprofit per unit of time. This model is generalized to the form where the\nriskiness of profit, measured by its per-time-unit volatility, is controlled\n(e.g. in case of existence of limits on riskiness of trading strategies imposed\nby regulatory bodies). The resulting optimization problem need not be convex.\nIn spite of this undesirable fact, it is demonstrated that the problem is still\nefficiently solvable. In addition, the problem that parameters of the price\ndifference process are never known exactly and are imprecisely estimated from\nan observed finite sample is investigated (recalling that this problem is\ncritical for practice). It is shown how the imprecision affects the optimal\ntrading strategy by quantification of the loss caused by the imprecise estimate\ncompared to a theoretical trader knowing the parameters exactly. The main\nresults focus on the geometric and optimization-theoretic viewpoint of the\nrisk-bounded trading strategy and the imprecision resulting from the\nstatistical estimates.\n"}
{"abstract": "  In this note, we show that certain dual canonical basis elements of\n$\\mathbb{C}[SL_m]$ are positive when evaluated on $k$-positive matrices,\nmatrices whose minors of size $k \\times k$ and smaller are positive. Skandera\nshowed that all dual canonical basis elements of $\\mathbb{C}[SL_m]$ can be\nwritten in terms of Kazhdan-Lusztig immanants, which were introduced by Rhoades\nand Skandera. We focus on the basis elements which are expressed in terms of\nKazhdan-Lusztig immanants indexed by 1324- and 2143-avoiding permutations. This\nextends previous work of the authors on Kazhdan-Lusztig immanants and uses\nsimilar tools, namely Lewis Carroll's identity (also known as the\nDesnanot-Jacobi identity).\n"}
{"abstract": "  The information describing the conditions of a system or a person is\nconstantly evolving and may become obsolete and contradict other information. A\ndatabase, therefore, must be consistently updated upon the acquisition of new\nvalid observations that contradict obsolete ones contained in the database. In\nthis paper, we propose a novel approach for dealing with the information\nobsolescence problem. Our approach aims to detect, in real-time, contradictions\nbetween observations and then identify the obsolete ones, given a\nrepresentation model. Since we work within an uncertain environment\ncharacterized by the lack of information, we choose to use a Bayesian network\nas our representation model and propose a new approximate concept,\n$\\epsilon$-Contradiction. The new concept is parameterised by a confidence\nlevel of having a contradiction in a set of observations. We propose a\npolynomial-time algorithm for detecting obsolete information. We show that the\nresulting obsolete information is better represented by an AND-OR tree than a\nsimple set of observations. Finally, we demonstrate the effectiveness of our\napproach on a real elderly fall-prevention database and showcase how this tree\ncan be used to give reliable recommendations to doctors. Our experiments give\nsystematically and substantially very good results.\n"}
{"abstract": "  Exploratory analysis of time series data can yield a better understanding of\ncomplex dynamical systems. Granger causality is a practical framework for\nanalysing interactions in sequential data, applied in a wide range of domains.\nIn this paper, we propose a novel framework for inferring multivariate Granger\ncausality under nonlinear dynamics based on an extension of self-explaining\nneural networks. This framework is more interpretable than other\nneural-network-based techniques for inferring Granger causality, since in\naddition to relational inference, it also allows detecting signs of\nGranger-causal effects and inspecting their variability over time. In\ncomprehensive experiments on simulated data, we show that our framework\nperforms on par with several powerful baseline methods at inferring Granger\ncausality and that it achieves better performance at inferring interaction\nsigns. The results suggest that our framework is a viable and more\ninterpretable alternative to sparse-input neural networks for inferring Granger\ncausality.\n"}
{"abstract": "  We consider an elliptic boundary value problem with unilateral constraints\nand subdifferential boundary conditions. The problem describes the heat\ntransfer in a domain $D\\subset\\R^d$ and its weak formulation is in the form of\na hemivariational inequality for the temperature field, denoted by $\\cP$. We\nassociate to Problem $\\cP$ an optimal control problem, denoted by $\\cQ$. Then,\nusing appropriate Tykhonov triples, governed by a nonlinear operator $G$ and a\nconvex $\\wK$, we provide results concerning the well-posedness of problems\n$\\cP$ and $\\cQ$. Our main results are Theorems 14 and 18, together with their\ncorollaries. Their proofs are based on arguments of compactness, lower\nsemicontinuity and pseudomonotonicity. Moreover, we consider three relevant\nperturbations of the heat transfer boundary valued problem which lead to\npenalty versions of Problem $\\cP$, constructed with particular choices of $G$\nand $\\wK$. We prove that Theorems 14 and 18 as well as their corollaries can be\napplied in the study of these problems, in order to obtain various convergence\nresults.\n"}
{"abstract": "  We study a quantum network that distributes entangled quantum states to\nmultiple sets of users that are connected to the network. Each user is\nconnected to a switch of the network via a link. All the links of the network\ngenerate bipartite Bell-state entangled states in each time-slot with certain\nprobabilities, and each end node stores one qubit of the entanglement generated\nby the link. To create shared entanglements for a set of users, measurement\noperations are performed on qubits of link-level entanglements on a set of\nrelated links, and these operations are probabilistic in nature and are\nsuccessful with certain probabilities. Requests arrive to the system seeking\nshared entanglements for different sets of users. Each request is for the\ncreation of shared entanglements for a fixed set of users using link-level\nentanglements on a fixed set of links. Requests are processed according to\nFirst-Come-First-Served service discipline and unserved requests are stored in\nbuffers. Once a request is selected for service, measurement operations are\nperformed on qubits of link-level entanglements on related links to create a\nshared entanglement. For given set of request arrival rates and link-level\nentanglement generation rates, we obtain necessary conditions for the stability\nof queues of requests. In each time-slot, the scheduler has to schedule\nentanglement swapping operations for different sets of users to stabilize the\nnetwork. Next, we propose a Max-Weight scheduling policy and show that this\npolicy stabilizes the network for all feasible arrival rates. We also provide\nnumerical results to support our analysis. The analysis of a single quantum\nswitch that creates multipartite entanglements for different sets of users is a\nspecial case of our work.\n"}
{"abstract": "  In many applications free surface flow through rigid porous media has to be\nmodeled. Examples refer to coastal engineering applications as well as\ngeotechnical or biomedical applications. Albeit the frequent applications,\nslight inconsistencies in the formulation of the governing equations can be\nfound in the literature. The main goal of this paper is to identify these\ndifferences and provide a quantitative assessment of different approaches.\nFollowing a review of the different formulations, simulation results obtained\nfrom three alternative formulations are compared with experimental and\nnumerical data. Results obtained by 2D and 3D test cases indicate that the\npredictive differences returned by the different formulations remain small for\nmost applications, in particular for small porous Reynolds number ReP < 5000.\nThus it seems justified to select a formulation that supports an efficient\nalgorithm and coding structure.\n"}
{"abstract": "  This work is motivated by an application in an industrial context, where the\nactivity of sensors is recorded at a high frequency. The objective is to\nautomatically detect abnormal measurement behaviour. Considering the sensor\nmeasures as functional data, we are formally interested in detecting outliers\nin a multivariate functional data set. Due to the heterogeneity of this data\nset, the proposed contaminated mixture model both clusters the multivariate\nfunctional data into homogeneous groups and detects outliers. The main\nadvantage of this procedure over its competitors is that it does not require us\nto specify the proportion of outliers. Model inference is performed through an\nExpectation-Conditional Maximization algorithm, and the BIC criterion is used\nto select the number of clusters. Numerical experiments on simulated data\ndemonstrate the high performance achieved by the inference algorithm. In\nparticular, the proposed model outperforms competitors. Its application on the\nreal data which motivated this study allows us to correctly detect abnormal\nbehaviours.\n"}
{"abstract": "  The present level of proliferation of fake, biased, and propagandistic\ncontent online has made it impossible to fact-check every single suspicious\nclaim or article, either manually or automatically. Thus, many researchers are\nshifting their attention to higher granularity, aiming to profile entire news\noutlets, which makes it possible to detect likely \"fake news\" the moment it is\npublished, by simply checking the reliability of its source. Source factuality\nis also an important element of systems for automatic fact-checking and \"fake\nnews\" detection, as they need to assess the reliability of the evidence they\nretrieve online. Political bias detection, which in the Western political\nlandscape is about predicting left-center-right bias, is an equally important\ntopic, which has experienced a similar shift towards profiling entire news\noutlets. Moreover, there is a clear connection between the two, as highly\nbiased media are less likely to be factual; yet, the two problems have been\naddressed separately. In this survey, we review the state of the art on media\nprofiling for factuality and bias, arguing for the need to model them jointly.\nWe further discuss interesting recent advances in using different information\nsources and modalities, which go beyond the text of the articles the target\nnews outlet has published. Finally, we discuss current challenges and outline\nfuture research directions.\n"}
{"abstract": "  It is well known that deep learning models have a propensity for fitting the\nentire training set even with random labels, which requires memorization of\nevery training sample. In this paper, we investigate the memorization effect in\nadversarial training (AT) for promoting a deeper understanding of capacity,\nconvergence, generalization, and especially robust overfitting of adversarially\ntrained classifiers. We first demonstrate that deep networks have sufficient\ncapacity to memorize adversarial examples of training data with completely\nrandom labels, but not all AT algorithms can converge under the extreme\ncircumstance. Our study of AT with random labels motivates further analyses on\nthe convergence and generalization of AT. We find that some AT methods suffer\nfrom a gradient instability issue, and the recently suggested complexity\nmeasures cannot explain robust generalization by considering models trained on\nrandom labels. Furthermore, we identify a significant drawback of memorization\nin AT that it could result in robust overfitting. We then propose a new\nmitigation algorithm motivated by detailed memorization analyses. Extensive\nexperiments on various datasets validate the effectiveness of the proposed\nmethod.\n"}
{"abstract": "  The design, integration, testing, and launch of the first Finnish satellite\nAalto-1 is briefly presented in this paper. Aalto-1, a three-unit CubeSat,\nlaunched into Sun-synchronous polar orbit at an altitude of approximately 500\nkm, is operational since June 2017. It carries three experimental payloads:\nAalto Spectral Imager (AaSI), Radiation Monitor (RADMON), and Electrostatic\nPlasma Brake (EPB). AaSI is a hyperspectral imager in visible and near-infrared\n(NIR) wavelength bands, RADMON is an energetic particle detector and EPB is a\nde-orbiting technology demonstration payload. The platform was designed to\naccommodate multiple payloads while ensuring sufficient data, power, radio,\nmechanical and electrical interfaces. The design strategy of platform and\npayload subsystems consists of in-house development and commercial subsystems.\nThe CubeSat Assembly, Integration & Test (AIT) followed Flatsat --\nEngineering-Qualification Model (EQM) -- Flight Model (FM) model philosophy for\nqualification and acceptance.\n  The paper briefly describes the design approach of platform and payload\nsubsystems, their integration and test campaigns, and spacecraft launch. The\npaper also describes the ground segment & services that were developed by the\nAalto-1 team.\n"}
{"abstract": "  The Zeeman effect is of limited utility for probing the magnetism of the\nquiet solar chromosphere. The Hanle effect in some spectral lines is sensitive\nto such magnetism, but the interpretation of the scattering polarization\nsignals requires taking into account that the chromospheric plasma is highly\ninhomogeneous and dynamic (i.e., that the magnetic field is not the only cause\nof symmetry breaking). Here we investigate the reliability of a well-known\nformula for mapping the azimuth of chromospheric magnetic fields directly from\nthe scattering polarization observed in the \\ion{Ca}{2}~8542~\\AA\\, line, which\nis typically in the saturation regime of the Hanle effect. To this end, we use\nthe Stokes profiles of the \\ion{Ca}{2}~8542~\\AA\\, line computed with the PORTA\nradiative transfer code in a three-dimensional (3D) model of the solar\nchromosphere, degrading them to mimic spectropolarimetric observations for a\nrange of telescope apertures and noise levels. The simulated observations are\nused to obtain the magnetic field azimuth at each point of the field of view,\nwhich we compare with the actual values within the 3D model. We show that,\napart from intrinsic ambiguities, the method provides solid results. Their\naccuracy depends more on the noise level than on the telescope diameter.\nLarge-aperture solar telescopes, like DKIST and EST, are needed to achieve the\nrequired noise-to-signal ratios using reasonable exposure times.\n"}
{"abstract": "  I propose a practical procedure based on bias correction and sample splitting\nto calculate confidence intervals for functionals of generic kernel methods,\ni.e. nonparametric estimators learned in a reproducing kernel Hilbert space\n(RKHS). For example, an analyst may desire confidence intervals for functionals\nof kernel ridge regression. I propose a bias correction that mirrors kernel\nridge regression. The framework encompasses (i) evaluations over discrete\ndomains, (ii) derivatives over continuous domains, (iii) treatment effects of\ndiscrete treatments, and (iv) incremental treatment effects of continuous\ntreatments. For the target quantity, whether it is (i)-(iv), I prove root-n\nconsistency, Gaussian approximation, and semiparametric efficiency by finite\nsample arguments. I show that the classic assumptions of RKHS learning theory\nalso imply inference.\n"}
{"abstract": "  Let $X$ be a symmetric, isotropic random vector in $\\mathbb{R}^m$ and let\n$X_1...,X_n$ be independent copies of $X$. We show that under mild assumptions\non $\\|X\\|_2$ (a suitable thin-shell bound) and on the tail-decay of the\nmarginals $\\langle X,u\\rangle$, the random matrix $A$, whose columns are\n$X_i/\\sqrt{m}$ exhibits a Gaussian-like behaviour in the following sense: for\nan arbitrary subset of $T\\subset \\mathbb{R}^n$, the distortion $\\sup_{t \\in T}\n| \\|At\\|_2^2 - \\|t\\|_2^2 |$ is almost the same as if $A$ were a Gaussian\nmatrix.\n  A simple outcome of our result is that if $X$ is a symmetric, isotropic,\nlog-concave random vector and $n \\leq m \\leq c_1(\\alpha)n^\\alpha$ for some\n$\\alpha>1$, then with high probability, the extremal singular values of $A$\nsatisfy the optimal estimate: $1-c_2(\\alpha) \\sqrt{n/m} \\leq \\lambda_{\\rm min}\n\\leq \\lambda_{\\rm max} \\leq 1+c_2(\\alpha) \\sqrt{n/m}$.\n"}
{"abstract": "  Modern, torque-controlled service robots can regulate contact forces when\ninteracting with their environment. Model Predictive Control (MPC) is a\npowerful method to solve the underlying control problem, allowing to plan for\nwhole-body motions while including different constraints imposed by the robot\ndynamics or its environment. However, an accurate model of the\nrobot-environment is needed to achieve a satisfying closed-loop performance.\nCurrently, this necessity undermines the performance and generality of MPC in\nmanipulation tasks. In this work, we combine an MPC-based whole-body controller\nwith two adaptive schemes, derived from online system identification and\nadaptive control. As a result, we enable a general mobile manipulator to\ninteract with unknown environments, without any need for re-tuning parameters\nor pre-modeling the interacting objects. In combination with the MPC\ncontroller, the two adaptive approaches are validated and benchmarked with a\nball-balancing manipulator in door opening and object lifting tasks.\n"}
{"abstract": "  Modern neural text generation systems can produce remarkably fluent and\ngrammatical texts. While earlier language models suffered from repetition and\nsyntactic errors, the errors made by contemporary models are often semantic,\nnarrative, or discourse failures.\n  To facilitate research of these complex error types, we introduce a new\nstructured, crowdsourced error annotation schema called Scarecrow. The error\ncategories used in Scarecrow -- such as redundancy, commonsense errors, and\nincoherence -- were identified by combining expert analysis with several pilot\nrounds of ontology-free crowd annotation to arrive at a schema which covers the\nerror phenomena found in real machine generated text.\n  We use Scarecrow to collect 13k annotations of 1.3k human and machine\ngenerate paragraphs of English language news text, amounting to over 41k spans\neach labeled with its error category, severity, a natural language explanation,\nand antecedent span (where relevant). We collect annotations for text generated\nby state-of-the-art systems with varying known performance levels, from GPT-2\nSmall through the largest GPT-3. We isolate several factors for detailed\nanalysis, including parameter count, training data, and decoding technique. Our\nresults show both expected and surprising differences across these settings.\nThese findings demonstrate the value of Scarecrow annotations in the assessment\nof current and future text generation systems. We release our complete\nannotation toolkit and dataset at https://yao-dou.github.io/scarecrow/.\n"}
{"abstract": "  Automated event extraction in social science applications often requires\ncorpus-level evaluations: for example, aggregating text predictions across\nmetadata and unbiased estimates of recall. We combine corpus-level evaluation\nrequirements with a real-world, social science setting and introduce the\nIndiaPoliceEvents corpus--all 21,391 sentences from 1,257 English-language\nTimes of India articles about events in the state of Gujarat during March 2002.\nOur trained annotators read and label every document for mentions of police\nactivity events, allowing for unbiased recall evaluations. In contrast to other\ndatasets with structured event representations, we gather annotations by posing\nnatural questions, and evaluate off-the-shelf models for three different tasks:\nsentence classification, document ranking, and temporal aggregation of target\nevents. We present baseline results from zero-shot BERT-based models fine-tuned\non natural language inference and passage retrieval tasks. Our novel\ncorpus-level evaluations and annotation approach can guide creation of similar\nsocial-science-oriented resources in the future.\n"}
{"abstract": "  We study the properties of homogeneous and isotropic turbulence in higher\nspatial dimensions through the lens of chaos and predictability using numerical\nsimulations. We employ both direct numerical simulations (DNS) and numerical\ncalculations of the Eddy Damped Quasi Normal Markovian (EDQNM) closure\napproximation. Our closure results show a remarkable transition to a\nnon-chaotic regime above critical dimension $d_c\\approx 5.88$. We relate these\nresults to the properties of the energy cascade as a function of spatial\ndimension in the context of the idea of a critical dimension for turbulence\nwhere Kolmogorov's 1941 theory becomes exact.\n"}
{"abstract": "  We develop a novel approach for confidently accelerating inference in the\nlarge and expensive multilayer Transformers that are now ubiquitous in natural\nlanguage processing (NLP). Amortized or approximate computational methods\nincrease efficiency, but can come with unpredictable performance costs. In this\nwork, we present CATs -- Confident Adaptive Transformers -- in which we\nsimultaneously increase computational efficiency, while guaranteeing a\nspecifiable degree of consistency with the original model with high confidence.\nOur method trains additional prediction heads on top of intermediate layers,\nand dynamically decides when to stop allocating computational effort to each\ninput using a meta consistency classifier. To calibrate our early prediction\nstopping rule, we formulate a unique extension of conformal prediction. We\ndemonstrate the effectiveness of this approach on four classification and\nregression tasks.\n"}
{"abstract": "  The Merino-Welsh conjectures say that subject to conditions, there is an\ninequality among the Tutte-polynomial evaluations $T(M;2,0)$, $T(M;0,2)$, and\n$T(M;1,1)$. We present three results on a Merino-Welsh conjecture. These\nresults are \"inconsequential\" in the sense that although they imply a version\nof the conjecture for many matroids, they seem to be dead ends.\n"}
{"abstract": "  Let $G$ be the Klein Four-group and let $k$ be an arbitrary field of\ncharacteristic 2. A classification of indecomposable $kG$-modules is known. We\ncalculate the relative cohomology groups $H_\\{chi}^i(G,N)$ for every\nindecomposable $kG$-module $N$, where $\\{chi}$ is the set of proper subgroups\nin $G$. This extends work of Pamuk and Yalcin to cohomology with non-trivial\ncoefficients. We also show that all cup products in strictly positive degree in\n$H_\\{chi}^*(G,k)$ are trivial.\n"}
{"abstract": "  In this paper, we propose in a Hilbertian setting a second-order\ntime-continuous dynamic system with fast convergence guarantees to solve\nstructured convex minimization problems with an affine constraint. The system\nis associated with the augmented Lagrangian formulation of the minimization\nproblem. The corresponding dynamics brings into play three general time-varying\nparameters, each with specific properties, and which are respectively\nassociated with viscous damping, extrapolation and temporal scaling. By\nappropriately adjusting these parameters, we develop a Lyapunov analysis which\nprovides fast convergence properties of the values and of the feasibility gap.\nThese results will naturally pave the way for developing corresponding\naccelerated ADMM algorithms, obtained by temporal discretization.\n"}
{"abstract": "  The underlying theme of this article is a class of sequences in metric\nstructures satisfying a much weaker kind of Cauchy condition, namely\nquasi-Cauchy sequences (introduced in \\cite{bc}) that has been used to define\nseveral new concepts in recent articles \\cite{PDSPNA2, PDSPNA1}. We first\nconsider a weaker notion of precompactness based on the idea of quasi-Cauchy\nsequences and establish several results including a new characterization of\ncompactness in metric spaces. Next we consider associated idea of continuity,\nnamely, ward continuous functions \\cite{caka}, as this class of functions\nstrictly lies between the classes of continuous and uniformly continuous\nfunctions and mainly establish certain coincidence results. Finally a new class\nof Lipschitz functions called \"quasi-Cauchy Lipschitz functions\" is introduced\nfollowing the line of investigations in \\cite{Beer1,Beer2,Beer3,g1} and again\nseveral coincidence results are proved. The motivation behind such kind of\nLipschitz functions is ascertained by the observation that every real valued\nward continuous function defined on a metric space can be uniformly\napproximated by real valued quasi-Cauchy Lipschitz functions.\n"}
{"abstract": "  We use numerical relativity to study the merger and ringdown stages of\n\"superkick\" binary black hole systems (those with equal mass and anti-parallel\nspins). We find a universal way to describe the mass and current quadrupole\ngravitational waves emitted by these systems during the merger and ringdown\nstage: (i) The time evolutions of these waves are insensitive to the\nprogenitor's parameters (spins) after being normalized by their own peak\nvalues. (ii) The peak values, which encode all the spin information of the\nprogenitor, can be consistently fitted to formulas inspired by post-Newtonian\ntheory. We find that the universal evolution of the mass quadrupole wave can be\naccurately modeled by the so-called Backwards One-Body (BOB) model. However,\nthe BOB model, in its present form, leads to a lower waveform match and a\nsignificant parameter-estimation bias for the current quadrupole wave. We also\ndecompose the ringdown signal into seven overtones, and study the dependence of\nmode amplitudes on the progenitor's parameters. Such dependence is found to be\ninsensitive to the overtone index (up to a scaling factor). Finally, we use the\nFisher matrix technique to investigate how the ringdown waveform can be at\nleast as important for parameter estimation as the inspiral stage. Assuming the\nCosmic Explorer, we find the contribution of ringdown portion dominates as the\ntotal mass exceeds ~ 250 solar mass. For massive BBH systems, the accuracy of\nparameter measurement is improved by incorporating the information of ringdown\n-- the ringdown sector gives rise to a different parameter correlation from\ninspiral stage, hence the overall parameter correlation is reduced in the full\nsignal.\n"}
{"abstract": "  We compute the quasinormal frequencies of $d$-dimensional spherically\nsymmetric black holes with leading string $\\alpha'$ corrections for tensorial\ngravitational perturbations in the highly damped regime. We solve\nperturbatively the master differential equation and we compute the monodromies\nof the master perturbation variable (analytically continued to the complex\nplane) in different contours, in order to obtain the quasinormal mode spectra.\nWe proceed analogously for the quasinormal modes of test scalar fields.\nDifferently than in Einstein gravity, we obtain distinct results for the two\ncases.\n"}
{"abstract": "  In this work, a new model of effective lift-off threshold of oxide/metal\ntarget is presented. The influence of nonlinear processes in the oxide layer on\nits removal from the metallic samples using picosecond laser was investigated.\nNonlinear and saturable absorption in the layer was incorporated into modeling\nfor prediction of effective laser lift-off threshold fluence change with\nvarying peak intensities in z-scan type experiment for the first time. The new\nmodel coincides well with the experimental results.\n"}
{"abstract": "  This article introduces new acceleration methods for fixed-point iterations.\nExtrapolations are computed using two or three mappings alternately and a new\ntype of step length is proposed with good properties for nonlinear\napplications. The methods require no problem-specific adaptation and are\nespecially efficient in high-dimensional contexts. Their computation uses few\nobjective function evaluations, no matrix inversion and little extra memory. A\nconvergence analysis is followed by eight applications including gradient\ndescent acceleration for constrained and unconstrained optimization.\nPerformances are on par with or better than competitive alternatives. The\nalgorithm is available as the Julia package SpeedMapping.jl.\n"}
{"abstract": "  We report electron transport studies on InSb-Al hybrid\nsemiconductor-superconductor nanowire devices. Tunnelling spectroscopy is used\nto measure the evolution of subgap states while varying magnetic field and\nvoltages applied to various nearby gates. At magnetic fields between 0.7 and\n0.9 T, the differential conductance contains large zero bias peaks (ZBPs) whose\nheight reaches values on the order 2e2/h. We investigate these ZBPs for large\nranges of gate voltages in different devices. We discuss possible\ninterpretations in terms of disorder-induced subgap states, Andreev bound\nstates and Majorana zero modes.\n"}
{"abstract": "  For the Stokes system in the half space, Kang [Math.~Ann.~2005] showed that a\nsolution generated by a compactly supported, H\\\"older continuous boundary flux\nmay have unbounded normal derivatives near the boundary. In this paper we first\nprove explicit global pointwise estimates of the above solution, showing in\nparticular that it has finite global energy and its derivatives blow up\neverywhere on the boundary away from the flux. We then use the above solution\nas a profile to construct solutions of the Navier-Stokes equations which also\nhave finite global energy and unbounded normal derivatives due to the flux. Our\nmain tool is the pointwise estimates of the Green tensor of the Stokes system\nproved by us in \\cite{Green} (arXiv:2011.00134). We also examine the Stokes\nflows generated by dipole bumps boundary flux, and identify the regions where\nthe normal derivatives of the solutions tend to positive or negative infinity\nnear the boundary.\n"}
{"abstract": "  In this paper we show a deterministic parallel all-pairs shortest paths\nalgorithm for real-weighted directed graphs. The algorithm has\n$\\tilde{O}(nm+(n/d)^3)$ work and $\\tilde{O}(d)$ depth for any depth parameter\n$d\\in [1,n]$. To the best of our knowledge, such a trade-off has only been\npreviously described for the real-weighted single-source shortest paths problem\nusing randomization [Bringmann et al., ICALP'17]. Moreover, our result improves\nupon the parallelism of the state-of-the-art randomized parallel algorithm for\ncomputing transitive closure, which has $\\tilde{O}(nm+n^3/d^2)$ work and\n$\\tilde{O}(d)$ depth [Ullman and Yannakakis, SIAM J. Comput. '91]. Our APSP\nalgorithm turns out to be a powerful tool for designing efficient planar graph\nalgorithms in both parallel and sequential regimes.\n  One notable ingredient of our parallel APSP algorithm is a simple\ndeterministic $\\tilde{O}(nm)$-work $\\tilde{O}(d)$-depth procedure for computing\n$\\tilde{O}(n/d)$-size hitting sets of shortest $d$-hop paths between all pairs\nof vertices of a real-weighted digraph. Such hitting sets have also been called\n$d$-hub sets. Hub sets have previously proved especially useful in designing\nparallel or dynamic shortest paths algorithms and are typically obtained via\nrandom sampling. Our procedure implies, for example, an $\\tilde{O}(nm)$-time\ndeterministic algorithm for finding a shortest negative cycle of a\nreal-weighted digraph. Such a near-optimal bound for this problem has been so\nfar only achieved using a randomized algorithm [Orlin et al., Discret. Appl.\nMath. '18].\n"}
{"abstract": "  The interaction of Older adults with robots requires effective feedback to\nkeep them aware of the state of the interaction for optimum interaction\nquality. This study examines the effect of different feedback modalities in a\ntable setting robot assistant for elder care. Two different feedback modalities\n(visual and auditory) and their combination were evaluated for three complexity\nlevels. The visual feedback included the use of LEDs and a GUI screen. The\nauditory feedback included alerts (beeps) and verbal commands. The results\nrevealed that the quality of interaction was influenced mainly by the feedback\nmodality, and complexity had less influence. The verbal feedback was\nsignificantly preferable and increased the involvement of the participants\nduring the experiment. The combination of LED lights and verbal commands\nincreased participants' understanding contributing to the quality of\ninteraction.\n"}
{"abstract": "  This paper shows that all immersed totally geodesic suborbifolds of\narithmetic hyperbolic orbifolds correspond to finite order subgroups of the\ncommensurator. We call such totally geodesic suborbifolds fc-subspaces (finite\ncentraliser subspaces) and use them to produce an arithmeticity criterion for\nhyperbolic lattices. We show that a hyperbolic orbifold $M$ is arithmetic if\nand only if it has infinitely many fc-subspaces, and provide examples of\nnon-arithmetic orbifolds that contain non-fc subspaces. Finally, we study\narithmetic properties of orbifolds that descend to their totally geodesic\nsuborbifolds and deduce some information about their geodesic embeddings. It\nfollows that certain arithmetic hyperbolic orbifolds cannot be geodesically\nimmersed in any other arithmetic ones.\n"}
{"abstract": "  We study the low temperature quench dynamics of the two-dimensional Potts\nmodel in the limit of large number of states, q >> 1. We identify a\nq-independent crossover temperature (the pseudo spinodal) below which no\nhigh-temperature metastability stops the curvature driven coarsening process.\nAt short length scales, the latter is decorated by freezing for some lattice\ngeometries, notably the square one. With simple analytic arguments we evaluate\nthe relevant time-scale in the coarsening regime, which turns out to be of\nArrhenius form and independent of q for large q. Once taken into account\ndynamic scaling is universal.\n"}
{"abstract": "  A new analytic approach to investigate the zero-temperature time evolution of\nthe Jaynes-Cummings system with cavity losses is developed. With the realistic\ncoupling between the cavity and the environment assumed, a simple master\nequation is derived, leading to the explicit analytic solution for the resonant\ncase. This solution is suitable for the analyses not only on the single\nexcitation states but also on many excitation states, which enables us to\ninvestigate the photon coherent state and to observe sharp collapses and\nrevivals under dissipation. For the off-resonant case, on the other hand, the\npresent study presents an analytic, systematic method instead. We examine the\nsmall and large detuning limits and discuss the condition where the widely-used\nphenomenological treatment is justified. Explicit evaluations of the time\nevolutions for various initial states with finite detuning are also presented.\n"}
{"abstract": "  The goal of group testing is to efficiently identify a few specific items,\ncalled positives, in a large population of items via tests. A test is an action\non a subset of items which returns positive if the subset contains at least one\npositive and negative otherwise. In non-adaptive group testing, all tests are\nfixed in advance and can be performed in parallel. In this work, we consider\nnon-adaptive group testing with consecutive positives in which the items are\nlinearly ordered and the positives are consecutive in that order.\n  We present two contributions here. The first is the direct use of a binary\ncode to construct measurement matrices compared to the use of Gray code in the\nstate-of-the-art work, which is a rearrangement of the binary code, when the\nmaximum number of consecutive positives is known. This leads to a reduction in\ndecoding time in practice. The second one is efficient designs to identify\npositives when the number of consecutive positives is known. To the best of our\nknowledge, this setting has not been surveyed yet. Our simulations verify the\nefficiency of our proposed designs. In particular, it only requires up to $300$\ntests to identify up to $100$ positives in a set of $2^{32} \\approx\n4.3\\mathrm{B}$ items in less than $300$ nanoseconds. When the maximum number of\nconsecutive positives is known, the simulations validate the superiority of our\nproposed design in decoding compared to the state-of-the-art work. Moreover,\nwhen the number of consecutive positives is known, the number of tests and the\ndecoding time are almost reduced half.\n"}
{"abstract": "  To remain aware of the fast-evolving cyber threat landscape, open-source\nCyber Threat Intelligence (OSCTI) has received growing attention from the\ncommunity. Commonly, knowledge about threats is presented in a vast number of\nOSCTI reports. Despite the pressing need for high-quality OSCTI, existing OSCTI\ngathering and management platforms, however, have primarily focused on\nisolated, low-level Indicators of Compromise. On the other hand, higher-level\nconcepts (e.g., adversary tactics, techniques, and procedures) and their\nrelationships have been overlooked, which contain essential knowledge about\nthreat behaviors that is critical to uncovering the complete threat scenario.\nTo bridge the gap, we propose SecurityKG, a system for automated OSCTI\ngathering and management. SecurityKG collects OSCTI reports from various\nsources, uses a combination of AI and NLP techniques to extract high-fidelity\nknowledge about threat behaviors, and constructs a security knowledge graph.\nSecurityKG also provides a UI that supports various types of interactivity to\nfacilitate knowledge graph exploration.\n"}
{"abstract": "  We study Einstein-Gauss-Bonnet gravity coupled to a bumblebee field which\nleads to a spontaneous Lorentz symmetry breaking in the gravitational sector.\nWe obtain an exact black hole solution and a cosmological solution in four\ndimensional spacetime by a regularization scheme. We also obtain a\nSchwarzschild-like bumblebee black hole solution in $D$-dimensional spacetime.\nWe find that the bumblebee field doesn't affect the locations of the black hole\nhorizon, but only affects the gravitational potential. That is, its\ngravitational potential has a minimum value(negative) in the black hole\ninterior and has a positive value $1+\\ell$ at short distance $r\\rightarrow0$.\nIf the constant $\\ell$ is large enough, then this kind of black hole is\npractically free from the singularity problem. The thermodynamics and phase\ntransition are also studied. In a cosmological context, it is interesting that\nthe Gauss-Bonnet term has no effect on the conservation of energy equation. A\nlate-time expansion of de Sitter universe can be replicated in an empty space.\nThe Gauss-Bonnet term and the bumblebee field can both actually act as a form\nof dark energy.\n"}
{"abstract": "  This article discusses cointegration tests for nonlinear cointegration in the\npresence of variance breaks in the errors. We build on approaches of Cavaliere\nand Taylor (2006, Journal of Time Series Analysis) for heteroskedastic\ncointegration tests and of Choi and Saikkonen (2010, Econometric Theory) for\nnonlinear cointegration tests. We propose a bootstrap test and prove its\nconsistency. A Monte Carlo study shows the approach to have appealing finite\nsample properties and to work better than an approach using subresiduals. We\nprovide an empirical application to the environmental Kuznets curves (EKC),\nfinding that the cointegration tests do not reject the EKC hypothesis in most\ncases.\n"}
{"abstract": "  Training the large deep neural networks that dominate NLP requires large\ndatasets. Many of these are collected automatically or via crowdsourcing, and\nmay exhibit systematic biases or annotation artifacts. By the latter, we mean\ncorrelations between inputs and outputs that are spurious, insofar as they do\nnot represent a generally held causal relationship between features and\nclasses; models that exploit such correlations may appear to perform a given\ntask well, but fail on out of sample data. In this paper we propose methods to\nfacilitate identification of training data artifacts, using new hybrid\napproaches that combine saliency maps (which highlight important input\nfeatures) with instance attribution methods (which retrieve training samples\ninfluential to a given prediction). We show that this proposed training-feature\nattribution approach can be used to uncover artifacts in training data, and use\nit to identify previously unreported artifacts in a few standard NLP datasets.\nWe execute a small user study to evaluate whether these methods are useful to\nNLP researchers in practice, with promising results. We make code for all\nmethods and experiments in this paper available.\n"}
{"abstract": "  We study the problem of approximating the cone of positive semidefinite (PSD)\nmatrices with a cone that can be described by smaller-sized PSD constraints.\nSpecifically, we ask the question: \"how closely can we approximate the set of\nunit-trace $n \\times n$ PSD matrices, denoted by $D$, using at most $N$ number\nof $k \\times k$ PSD constraints?\" In this paper, we prove lower bounds on $N$\nto achieve a good approximation of $D$ by considering two constructions of an\napproximating set. First, we consider the unit-trace $n \\times n$ symmetric\nmatrices that are PSD when restricted to a fixed set of $k$-dimensional\nsubspaces in $\\mathbb{RR}^n$. We prove that if this set is a good approximation\nof $D$, then the number of subspaces must be at least exponentially large in\n$n$ for any $k = o(n)$. % Second, we show that any set $S$ that approximates\n$D$ within a constant approximation ratio must have superpolynomial\n$\\mathbf{S}_+^k$-extension complexity. To be more precise, if $S$ is a constant\nfactor approximation of $D$, then $S$ must have $\\mathbf{S}_+^k$-extension\ncomplexity at least $\\exp( C \\cdot \\min \\{ \\sqrt{n}, n/k \\})$ where $C$ is some\nabsolute constant. In addition, we show that any set $S$ such that $D \\subseteq\nS$ and the Gaussian width of $D$ is at most a constant times larger than the\nGaussian width of $D$ must have $\\mathbf{S}_+^k$-extension complexity at least\n$\\exp( C \\cdot \\min \\{ n^{1/3}, \\sqrt{n/k} \\})$. These results imply that the\ncone of $n \\times n$ PSD matrices cannot be approximated by a polynomial number\nof $k \\times k$ PSD constraints for any $k = o(n / \\log^2 n)$. These results\ngeneralize the recent work of Fawzi on the hardness of polyhedral\napproximations of $\\mathbf{S}_+^n$, which corresponds to the special case with\n$k=1$.\n"}
{"abstract": "  Following Johnsen and Verdure (2013), we can associate to any linear code $C$\nan abstract simplicial complex and in turn, a Stanley-Reisner ring $R_C$. The\nring $R_C$ is a standard graded algebra over a field and its projective\ndimension is precisely the dimension of $C$. Thus $R_C$ admits a graded minimal\nfree resolution and the resulting graded Betti numbers are known to determine\nthe generalized Hamming weights of $C$. The question of purity of the minimal\nfree resolution of $R_C$ was considered by Ghorpade and Singh (2020) when $C$\nis the generalized Reed-Muller code. They showed that the resolution is pure in\nsome cases and it is not pure in many other cases. Here we give a complete\ncharacterization of the purity of graded minimal free resolutions of\nStanley-Reisner rings associated to generalized Reed-Muller codes of an\narbitrary order.\n"}
{"abstract": "  In response to the Kaggle's COVID-19 Open Research Dataset (CORD-19)\nchallenge, we have proposed three transformer-based question-answering systems\nusing BERT, ALBERT, and T5 models. Since the CORD-19 dataset is unlabeled, we\nhave evaluated the question-answering models' performance on two labeled\nquestions answers datasets \\textemdash CovidQA and CovidGQA. The BERT-based QA\nsystem achieved the highest F1 score (26.32), while the ALBERT-based QA system\nachieved the highest Exact Match (13.04). However, numerous challenges are\nassociated with developing high-performance question-answering systems for the\nongoing COVID-19 pandemic and future pandemics. At the end of this paper, we\ndiscuss these challenges and suggest potential solutions to address them.\n"}
{"abstract": "  In this review we discuss recent advances in the self-assembly of\nself-propelled colloidal particles and highlight some of the most exciting\nresults in this field with a specific focus on dry active matter. We explore\nthis phenomenology through the lens of the complexity of the colloidal building\nblocks. We begin by considering the behavior of isotropic spherical particles.\nWe then discuss the case of amphiphilic and dipolar Janus particles. Finally,\nwe show how the geometry of the colloids and/or the directionality of their\ninteractions can be used to control the physical properties of the assembled\nactive aggregates, and suggest possible strategies on how to exploit activity\nas a tunable driving force for self-assembly. The unique properties of active\ncolloids lend promise for the design of the next generation of functional,\nenvironment-sensing microstructures able to perform specific tasks in an\nautonomous and targeted manner.\n"}
{"abstract": "  The characteristics of high loss rate, resource constraint, being eager for\ngood security haven't been fully considered in the existing key establishment\nprotocols of wireless sensor networks. Analyzing the key establishing problem\nfrom the MAC and physical layers, existing protocols are not practicable enough\ndue to their overlong agreement packets and single round key establishment. To\nmitigate the impact of these problems, a group of design principles for secure\nsensor networks has been presented and TinyAKE, an authenticated key transport\nprotocol based on lightweight certificate, is proposed in this paper. The\nsecurity of TinyAKE are proved with the theory of indistinguishability,\nmeanwhile, the correctness is also proved, the performance is analyzed and\ncompared with the existing similar protocols. Finally TinyAKE is implemented in\nthe TinyOS with TinyECC. Our evaluation shows that TinyAKE is a more\npracticable and trustable authenticated key establishment protocol than\nexisting protocols. The experimental result shows that the key transport with\ncertificate mechanism is feasible in WSNs. Moreover, the simulation results\nshow that the optimal number of repeated negotiation is one when the secure\nconnectivity rate of TinyAKE is improved by using the repeated key negotiation.\n"}
{"abstract": "  Generating natural speech with diverse and smooth prosody pattern is a\nchallenging task. Although random sampling with phone-level prosody\ndistribution has been investigated to generate different prosody patterns, the\ndiversity of the generated speech is still very limited and far from what can\nbe achieved by human. This is largely due to the use of uni-modal distribution,\nsuch as single Gaussian, in the prior works of phone-level prosody modelling.\nIn this work, we propose a novel approach that models phone-level prosodies\nwith GMM based mixture density network (GMM-MDN). Experiments on the LJSpeech\ndataset demonstrate that phone-level prosodies can precisely control the\nsynthetic speech and GMM-MDN can generate more natural and smooth prosody\npattern than a single Gaussian. Subjective evaluations further show that the\nproposed approach not only achieves better naturalness, but also significantly\nimproves the prosody diversity in synthetic speech without the need of manual\ncontrol.\n"}
{"abstract": "  In a spatial-temporal model, structural change and/or spatial heterogeneity\ncan easily affect estimation of parameters. Following the spatial-temporal\nmodel in [1], we develop a nonparametric procedure for test-ing the presence of\nstructural change and spatial heterogeneity using bootstrap techniques and the\nforward search algorithm. The time series bootstrap can filter the effect of\ntemporary structural change in the con-struction of a confidence interval for\nthe temporal parameter. The forward search will also facilitate the\nconstruction of a robust confidence interval for the spatial parameter. These\nconfidence intervals are then used in deciding on the null hypothesis that\nthere is no structural change/spatial heterogeneity. Simulation studies\nillustrate the ability of the proposed test procedure in detecting presence of\nstructural change and spatial heterogeneity under certain conditions.\n"}
{"abstract": "  In this paper we show that a $\\rho$-Einstein solitons conformal to a\npseudo-Euclidean space, invariant under the action of the pseudo-orthogonal\ngroup with zero scalar curvature is stady and consequently flat. How\napplication of the results obtained we present an explicit example for a the\nquestion proposed by Kazdan in [17].\n"}
{"abstract": "  We calculate the production of hypothetical millicharged particles (MCPs) of\nsub-GeV masses by the J-PARC proton beam in the framework of T2K and future\nT2HK neutrino oscillation experiments. Concentrating on the region of model\nparameter space, where an MCP can hit the near neutrino detector twice, we\nadopt this background-free signature to estimate the sensitivity of T2K and\nT2HK experiments to MCPs. We find that a previously inaccessible in direct\nsearches region of charges 5$\\times$$10^{-4}$-$10^{-2}$ $e$ for MCP masses\n0.1-0.5 GeV can be probed.\n"}
{"abstract": "  Consider a weighted branching process generated by a point process on\n$[0,1]$, whose atoms sum up to one. Then the weights of all individuals in any\ngiven generation sum up to one, as well. We define a nested occupancy scheme in\nrandom environment as the sequence of balls-in-boxes schemes (with random\nprobabilities) in which boxes of the $j$th level, $j=1,2,\\ldots$ are identified\nwith the $j$th generation individuals and the hitting probabilities of boxes\nare identified with the corresponding weights. The collection of balls is the\nsame for all generations, and each ball starts at the root and moves along the\ntree of the weighted branching process according to the following rule:\ntransition from a mother box to a daughter box occurs with probability given by\nthe ratio of the daughter and mother weights.\n  Assuming that there are $n$ balls, we give a full classification of regimes\nof the a.s.\\ convergence for the number of occupied (ever hit) boxes in the\n$j$th level, properly normalized, as $n$ and $j=j_n$ grow to $\\infty$. Here,\n$(j_n)_{n\\in\\mathbb N}$ is a sequence of positive numbers growing\nproportionally to $\\log n$. We call such levels late, for the nested occupancy\nscheme gets extinct in the levels which grow super-logarithmically in $n$ in\nthe sense that each occupied box contains one ball. Also, in some regimes we\nprove the strong laws of large numbers (a) for the number of the $j$th level\nboxes which contain at least $k$ balls, $k\\geq 2$ and (b) under the assumption\nthat the mean number of the first level boxes is finite, for the number of\nempty boxes in the $j$th level.\n"}
{"abstract": "  This paper presents a novel method for obtaining the probability wave of\nbreaking ($P_b$) of deep water, dominant wind-sea waves (that is, waves made of\nthe energy within $\\pm$30\\% of the peak wave frequency) derived from Gaussian\nwave field theory. For a given input wave spectrum we demonstrate how it is\npossible to derive a joint probability density function between wave phase\nspeed ($c$) and horizontal orbital velocity at wave crest ($u$) from which a\nmodel for $P_b$ can be obtained. A non-linear kinematic wave breaking criterion\nconsistent with the Gaussian framework is further proposed. Our model would\nallow, therefore, for application of the classical wave breaking criterion\n(that is, wave breaking occurs if $u/c > 1$) in spectral wave models which, to\nthe authors' knowledge, has not been done to date. Our results show that the\nproposed theoretical model has errors in the same order of magnitude as six\nother historical models when assessed using three field datasets. With\noptimization of the proposed model's single free parameter, it can become the\nbest performing model for specific datasets. Although our results are\npromising, additional, more complete wave breaking datasets collected in the\nfield are needed to comprehensively assess the present model, especially in\nregards to the dependence on phenomena such as direct wind forcing, long wave\nmodulation and wave directionality.\n"}
{"abstract": "  Various proxy metrics for test quality have been defined in order to guide\ndevelopers when writing tests. Code coverage is particularly well established\nin practice, even though the question of how coverage relates to test quality\nis a matter of ongoing debate. Mutation testing offers a promising alternative:\nArtificial defects can identify holes in a test suite, and thus provide\nconcrete suggestions for additional tests. Despite the obvious advantages of\nmutation testing, it is not yet well established in practice. Until recently,\nmutation testing tools and techniques simply did not scale to complex systems.\nAlthough they now do scale, a remaining obstacle is lack of evidence that\nwriting tests for mutants actually improves test quality. In this paper we aim\nto fill this gap: By analyzing a large dataset of almost 15 million mutants, we\ninvestigate how these mutants influenced developers over time, and how these\nmutants relate to real faults. Our analyses suggest that developers using\nmutation testing write more tests, and actively improve their test suites with\nhigh quality tests such that fewer mutants remain. By analyzing a dataset of\npast fixes of real high-priority faults, our analyses further provide evidence\nthat mutants are indeed coupled with real faults. In other words, had mutation\ntesting been used for the changes introducing the faults, it would have\nreported a live mutant that could have prevented the bug.\n"}
{"abstract": "  NLP systems rarely give special consideration to numbers found in text. This\nstarkly contrasts with the consensus in neuroscience that, in the brain,\nnumbers are represented differently from words. We arrange recent NLP work on\nnumeracy into a comprehensive taxonomy of tasks and methods. We break down the\nsubjective notion of numeracy into 7 subtasks, arranged along two dimensions:\ngranularity (exact vs approximate) and units (abstract vs grounded). We analyze\nthe myriad representational choices made by 18 previously published number\nencoders and decoders. We synthesize best practices for representing numbers in\ntext and articulate a vision for holistic numeracy in NLP, comprised of design\ntrade-offs and a unified evaluation.\n"}
{"abstract": "  We investigate the dust attenuation in both stellar populations and ionized\ngas in kpc-scale regions in nearby galaxies, using integral field spectroscopy\ndata from MaNGA MPL-9. We identify star-forming (HII) and diffuse ionized gas\n(DIG) regions from MaNGA datacubes. From the stacked spectrum of each region,\nwe measure the stellar attenuation, $E(B-V)_{\\rm star}$, using the technique\ndeveloped by Li et al.(2020), as well as the gas attenuation, $E(B-V)_{\\rm\ngas}$, from the Balmer decrement. We then examine the correlation of\n$E(B-V)_{\\rm star}$, $E(B-V)_{\\rm gas}$, $E(B-V)_{\\rm gas}-E(B-V)_{\\rm star}$\nand $E(B-V)_{\\rm star}/E(B-V)_{\\rm gas}$ with 16 regional/global properties,\nand for regions with different $\\rm H{\\alpha}$ surface brightnesses\n($\\Sigma_{\\rm H\\alpha}$). We find a stronger correlation between $E(B-V)_{\\rm\nstar}$ and $E(B-V)_{\\rm gas}$ in regions of higher $\\Sigma_{\\rm H\\alpha}$.\nLuminosity-weighted age ($t_L$) is found to be the property that is the most\nstrongly correlated with $E(B-V)_{\\rm star}$, and consequently with\n$E(B-V)_{\\rm gas}-E(B-V)_{\\rm star}$ and $E(B-V)_{\\rm star}/E(B-V)_{\\rm gas}$.\nAt fixed $\\Sigma_{\\rm H\\alpha}$, $\\log_{10}t_L$ is linearly and negatively\ncorrelated with $E(B-V)_{\\rm star}/E(B-V)_{\\rm gas}$ at all ages. Gas-phase\nmetallicity and ionization level are important for the attenuation in the gas.\nOur results indicate that the ionizing source for DIG regions is likely\ndistributed in the outer-skirt of galaxies, while for HII regions our results\ncan be well explained by the two-component dust model of Charlot & Fall (2000).\n"}
{"abstract": "  We study the principal core g-mode oscillation in hybrid stars containing\nquark matter and find that they have an unusually large frequency range\n($\\approx$ 200 - 600 Hz) compared to ordinary neutron stars or self-bound quark\nstars of the same mass. Theoretical arguments and numerical calculations that\ntrace this effect to the difference in the behaviour of the equilibrium and\nadiabatic sound speeds in the mixed phase of quarks and nucleons are provided.\nWe propose that the sensitivity of core g-mode oscillations to non-nucleonic\nmatter in neutron stars could be due to the presence of a mixed quark-nucleon\nphase. Based on our analysis, we conclude that for binary mergers where one or\nboth components may be a hybrid star, the fraction of tidal energy pumped into\nresonant g-modes in hybrid stars can exceed that of a normal neutron star by a\nfactor of 2-3, although resonance occurs during the last stages of inspiral. A\nself-bound star, on the other hand, has a much weaker tidal overlap with the\ng-mode. The cumulative tidal phase error in hybrid stars, $\\Delta\\phi\\cong$ 0.5\nrad, is comparable to that from tides in ordinary neutron stars, presenting a\nchallenge in distinguishing between the two cases. However, should the\nprincipal g-mode be excited to sufficient amplitude for detection in a\npost-merger remnant with quark matter in its interior, its frequency would be a\npossible indication for the existence of non-nucleonic matter in neutron stars.\n"}
{"abstract": "  In this work we present the Scaled QUantum IDentifier (SQUID), an open-source\nframework for exploring hybrid Quantum-Classical algorithms for classification\nproblems. The classical infrastructure is based on PyTorch and we provide a\nstandardized design to implement a variety of quantum models with the\ncapability of back-propagation for efficient training. We present the structure\nof our framework and provide examples of using SQUID in a standard binary\nclassification problem from the popular MNIST dataset. In particular, we\nhighlight the implications for scalability for gradient-based optimization of\nquantum models on the choice of output for variational quantum models.\n"}
{"abstract": "  We study reactive synthesis of systems interacting with environments using\ninfinite alphabets. Register automata and transducers are popular formalisms\nfor specifying and modelling such systems. They extend finite-state automata by\nadding registers to store data values and to compare the incoming data values\nagainst stored ones. Synthesis from nondeterministic or universal register\nautomata is undecidable in general. Its register-bounded variant, where\nadditionally a bound on the number of registers in a sought transducer is\ngiven, is however known to be decidable for universal register automata which\ncan compare data for equality. In this paper, we generalise this result. We\nintroduce the notion of $\\omega$-regularly approximable data domains, and show\nthat register-bounded synthesis from universal register automata on such\ndomains is decidable. Importantly, the data domain ($\\mathbb{N}$, =, <) with\nnatural order is $\\omega$-regularly approximable, and its closer examination\nreveals that the synthesis problem is decidable in time doubly exponential in\nthe number of registers, matching the known complexity of the equality-only\ncase ($\\mathbb{N}$, =). We then introduce a notion of reducibility between data\ndomains which we exploit to show decidability of synthesis over, e.g., the\ndomains ($\\mathbb{N}^d$, $=^d$, $<^d$) of tuples of numbers equipped with the\ncomponent-wise partial order and ($\\Sigma^*$, =, $\\prec$) of finite strings\nwith the prefix relation.\n"}
{"abstract": "  With the development of end-to-end control based on deep learning, it is\nimportant to study new system modeling techniques to realize dynamics modeling\nwith high-dimensional inputs. In this paper, a novel Koopman-based deep\nconvolutional network, called CKNet, is proposed to identify latent dynamics\nfrom raw pixels. CKNet learns an encoder and decoder to play the role of the\nKoopman eigenfunctions and modes, respectively. The Koopman eigenvalues can be\napproximated by eigenvalues of the learned state transition matrix. The\ndeterministic convolutional Koopman network (DCKNet) and the variational\nconvolutional Koopman network (VCKNet) are proposed to span some subspace for\napproximating the Koopman operator respectively. Because CKNet is trained under\nthe constraints of the Koopman theory, the identified latent dynamics is in a\nlinear form and has good interpretability. Besides, the state transition and\ncontrol matrices are trained as trainable tensors so that the identified\ndynamics is also time-invariant. We also design an auxiliary weight term for\nreducing multi-step linearity and prediction losses. Experiments were conducted\non two offline trained and four online trained nonlinear forced dynamical\nsystems with continuous action spaces in Gym and Mujoco environment\nrespectively, and the results show that identified dynamics are adequate for\napproximating the latent dynamics and generating clear images. Especially for\noffline trained cases, this work confirms CKNet from a novel perspective that\nwe visualize the evolutionary processes of the latent states and the Koopman\neigenfunctions with DCKNet and VCKNet separately to each task based on the same\nepisode and results demonstrate that different approaches learn similar\nfeatures in shapes.\n"}
{"abstract": "  The availability of Open Government Data (OGD) provides means for citizens to\nunderstand and follow governmental policies and decisions, showing evidence of\nhow the latter have contributed to both the place they live in and their lives.\nIn such a scenario, one of the proposals is the use of visualizations to\nsupport the process of data analysis and interpretation. Herein, we present the\nuse of three different visualization tools, a commercial one and two academic\nones, applied to two specific Brazilian cases: the implementation of the Drink\nDriving Law and the construction of a new overpass in an important city avenue.\nOur focus was on the analysis of how visualization could help in the\nidentification of the effects of such traffic public policies. As our main\ncontributions, we present details on the effects of the observed policies, as\nwell as new cases showing how visualization tools can assist users to interpret\nOGD.\n"}
{"abstract": "  Probabilistic logic programming is a major part of statistical relational\nartificial intelligence, where approaches from logic and probability are\nbrought together to reason about and learn from relational domains in a setting\nof uncertainty. However, the behaviour of statistical relational\nrepresentations across variable domain sizes is complex, and scaling inference\nand learning to large domains remains a significant challenge. In recent years,\nconnections have emerged between domain size dependence, lifted inference and\nlearning from sampled subpopulations. The asymptotic behaviour of statistical\nrelational representations has come under scrutiny, and projectivity was\ninvestigated as the strongest form of domain-size dependence, in which query\nmarginals are completely independent of the domain size.\n  In this contribution we show that every probabilistic logic program under the\ndistribution semantics is asymptotically equivalent to an acyclic probabilistic\nlogic program consisting only of determinate clauses over probabilistic facts.\nWe conclude that every probabilistic logic program inducing a projective family\nof distributions is in fact everywhere equivalent to a program from this\nfragment, and we investigate the consequences for the projective families of\ndistributions expressible by probabilistic logic programs.\n  To facilitate the application of classical results from finite model theory,\nwe introduce the abstract distribution semantics, defined as an arbitrary\nlogical theory over probabilistic facts. This bridges the gap to the\ndistribution semantics underlying probabilistic logic programming. In this\nrepresentation, determinate logic programs correspond to quantifier-free\ntheories, making asymptotic quantifier elimination results available for the\nsetting of probabilistic logic programming.\n  This paper is under consideration for acceptance in TPLP.\n"}
{"abstract": "  We theoretically show that the three-dimensional (3D) topological insulator\n(TI)/thin-film ferromagnetic metal (FMM) bilayer structure is possible to be a\nquantum anomalous Hall (QAH) insulator with a wide global band gap. Studying\nthe band structure and the weight distributions of eigenstates, we demonstrate\nthat the attachment of a metallic thin-film on the 3DTI can shift the\ntopologically non-trivial state into the metal layers due to the hybridization\nof bands around the original Dirac point. By introducing the magnetic exchange\ninteraction in the thin-film layers, we compute the anomalous Hall conductivity\nand magnetic anisotropy of the bilayer structure to suggest the appearance of\nwider gap realizing QAH effect than usual materials, such as magnetically doped\nthin-films of 3DTI and 3DTI/ferromagnetic insulator heterostructures. Our\nresults indicate that the 3DTI/thin-film FMM bilayer structure may implement\nthe QAH effect even at room temperature, which will pave a way to the\nexperimental realization of other exotic topological quantum phenomena.\n"}
{"abstract": "  The physical picture of interacting magnetic islands provides a useful\nparadigm for certain plasma dynamics in a variety of physical environments,\nsuch as the solar corona, the heliosheath, and the Earth's magnetosphere. In\nthis work, we derive an island kinetic equation to describe the evolution of\nthe island distribution function (in area and in flux of islands) subject to a\ncollisional integral designed to account for the role of magnetic reconnection\nduring island mergers. This equation is used to study the inverse transfer of\nmagnetic energy through the coalescence of magnetic islands in 2D. We solve our\nisland kinetic equation numerically for three different types of initial\ndistribution: delta-distribution, Gaussian and power-law distribution. The time\nevolution of several key quantities is found to agree well with our analytical\npredictions: magnetic energy decays as $\\tilde t^{-1}$, the number of islands\ndecreases as $\\tilde t^{-1}$, and the averaged area of islands grows as $\\tilde\nt$, where $\\tilde t$ is the time normalized to the characteristic reconnection\ntime scale of islands. General properties of the distribution function and the\nmagnetic energy spectrum are also studied.\n"}
{"abstract": "  Time-dependent driving of quantum systems has emerged as a powerful tool to\nengineer exotic phases far from thermal equilibrium; when the drive is periodic\nthis is called Floquet engineering. The presence of many-body interactions can\nlead to runaway heating, so that generic systems are believed to heat up until\nthey reach a featureless infinite-temperature state. Finding mechanisms to slow\ndown or even avoid this heat death is a major goal -- one such mechanism is to\ndrive toward an even distribution of electrons in momentum space. Here we show\nhow such a mechanism avoids the heat death for a charge-density-wave chain in a\nstrong dc electric field; minibands with nontrivial distribution functions\ndevelop as the current is prematurely driven to zero. We also show how the\nfield strength tunes between positive, negative, or close-to-infinite effective\ntemperatures for each miniband. These results suggest that nontrivial\nmetastable distribution functions should be realized in the prethermal regime\nof quantum systems coupled to slow bosonic modes.\n"}
{"abstract": "  This paper introduces an efficient algorithm for the sequential positioning\n(or nested dissection) of two planar interfaces in an arbitrary polyhedron,\nsuch that, after each truncation, the respectively remaining polyhedron admits\na prescribed volume. This task, among others, is frequently encountered in the\nnumerical simulation of three-phase flows when resorting to the geometric\nVolume-of-Fluid method. For two-phase flows, the recent work of Kromer & Bothe\n(doi.org/10.1016/j.jcp.2021.110776) addresses the positioning of a single plane\nby combining an implicit bracketing of the sought position with up to\nthird-order derivatives of the volume fraction. An analogous application of\ntheir highly efficient root-finding scheme to three-material configurations\nrequires computing the volume of a twice truncated arbitrary polyhedron. The\npresent manuscript achieves this by recursive application of the Gaussian\ndivergence theorem in appropriate form, which allows to compute the volume as a\nsum of quantities associated to the faces of the original polyhedron. With a\nsuitable choice of the coordinate origin, accounting for the sequential\ncharacter of the truncation, the volume parametrization becomes co-moving with\nrespect to the planes. This eliminates the necessity to establish topological\nconnectivity and tetrahedron decomposition after each truncation. After a\ndetailed mathematical description of the concept, we conduct a series of\ncarefully designed numerical experiments to assess the performance in terms of\npolyhedron truncations. The high efficiency of the two-phase positioning\npersists for sequential application, thereby being robust with respect to input\ndata and possible intersection topologies. In comparison to an existing\ndecomposition-based approach, the number of truncations was reduced by up to an\norder of magnitude.\n"}
{"abstract": "  In this paper, we present a method of clothes retargeting; generating the\npotential poses and deformations of a given 3D clothing template model to fit\nonto a person in a single RGB image. The problem is fundamentally ill-posed as\nattaining the ground truth data is impossible, i.e., images of people wearing\nthe different 3D clothing template model at exact same pose. We address this\nchallenge by utilizing large-scale synthetic data generated from physical\nsimulation, allowing us to map 2D dense body pose to 3D clothing deformation.\nWith the simulated data, we propose a semi-supervised learning framework that\nvalidates the physical plausibility of the 3D deformation by matching with the\nprescribed body-to-cloth contact points and clothing silhouette to fit onto the\nunlabeled real images. A new neural clothes retargeting network (CRNet) is\ndesigned to integrate the semi-supervised retargeting task in an end-to-end\nfashion. In our evaluation, we show that our method can predict the realistic\n3D pose and deformation field needed for retargeting clothes models in\nreal-world examples.\n"}
{"abstract": "  Laser-induced breakdown spectroscopy (LIBS) is a popular, fast elemental\nanalysis technique used to determine the chemical composition of target\nsamples, such as in industrial analysis of metals or in space exploration.\nRecently, there has been a rise in the use of machine learning (ML) techniques\nfor LIBS data processing. However, ML for LIBS is challenging as: (i) the\npredictive models must be lightweight since they need to be deployed in highly\nresource-constrained and battery-operated portable LIBS systems; and (ii) since\nthese systems can be remote, the models must be able to self-adapt to any\ndomain shift in input distributions which could be due to the lack of different\ntypes of inputs in training data or dynamic environmental/sensor noise. This\non-device retraining of model should not only be fast but also unsupervised due\nto the absence of new labeled data in remote LIBS systems. We introduce a\nlightweight multi-layer perceptron (MLP) model for LIBS that can be adapted\non-device without requiring labels for new input data. It shows 89.3% average\naccuracy during data streaming, and up to 2.1% better accuracy compared to an\nMLP model that does not support adaptation. Finally, we also characterize the\ninference and retraining performance of our model on Google Pixel2 phone.\n"}
{"abstract": "  Hosono obtained sharper estimates of the Ohsawa--Takegoshi $L^2$-extention\ntheorem by allowing the constant depending on the weight function for a domain\nin $\\mathbb{C}$. In this article, we show the higher dimensional case of\nsharper estimates of the Ohsawa--Takegoshi $L^2$-extention theorem. To prove\nthe higher dimensional case of them, we establish an analogue of\nBerndtsson--Lempert type $L^2$-extension theorem by using the pluricomplex\nGreen functions with poles along subvarieties. As a special case, we consider\nthe sharper estimates in terms of the Azukawa pseudometric and show that the\nhigher dimensional case of sharper estimate provides the $L^2$-minimum\nextension for radial case.\n"}
{"abstract": "  The notion of quasi $f$-ideals was first presented in $[14]$ which generalize\nthe idea of $f$-ideals. In this paper, we give the complete characterization of\nquasi $f$-ideals of degree greater or equal to $2$. Additionally, we show that\nthe property of being quasi $f$-ideals remains the same after taking the Newton\ncomplementary dual of a squarefree monomial ideal $I$ provided that the minimal\ngenerating set of $I$ is perfect.\n"}
{"abstract": "  We propose a novel procedure, which combines agglomerative hierarchical\nclustering and a test of overidentifying restrictions for selecting valid\ninstrumental variables (IV) from a large set of candidate IVs. Some of these\nIVs may be invalid in the sense that they fail the exclusion restriction. We\nshow that if the largest group of IVs is valid, our method achieves oracle\nproperties. Compared with existing methods, our method can deal with weak\ninstruments, multiple endogenous regressors and heterogeneous treatment\neffects. In simulations, we show that our method outperforms the two closest\nmethods, the Hard Thresholding method and the Confidence Interval method.l the\ninstruments are strong. Also, our method works well when some of the candidate\ninstruments are weak, outperforming HT and CIM. We apply our method to the\nestimation of the effect of immigration on wages in the US.\n"}
{"abstract": "  Let $X\\subset \\mathbb{P}^r$ be an integral and non-degenerate variety. Let\n$\\sigma _{a,b}(X)\\subseteq \\mathbb{P}^r$, $(a,b)\\in \\mathbb{N}^2$, be the join\nof $a$ copies of $X$ and $b$ copies of the tangential variety of $X$. Using the\nclassical Alexander-Hirschowitz theorem (case $b=0$) and a recent paper by H.\nAbo and N. Vannieuwenhoven (case $a=0$) we compute $\\dim \\sigma _{a,b}(X)$ in\nmany cases when $X$ is the $d$-Veronese embedding of $\\mathbb{P}^n$. This is\nrelated to certain additive decompositions of homogeneous polynomials. We give\na general theorem proving that $\\dim \\sigma _{0,b}(X)$ is the expected one when\n$X=Y\\times \\mathbb{P}^1$ has a suitable Segre-Veronese style embedding in\n$\\mathbb{P}^r$. As a corollary we prove that if $d_i\\ge 3$, $1\\le i \\le n$, and\n$(d_1+1)(d_2+1)\\ge 38$ the tangential variety of $(\\mathbb{P}^1)^n$ embedded by\n$|\\mathcal{O} _{(\\mathbb{P} ^1)^n}(d_1,\\dots ,d_n)|$ is not defective and a\nsimilar statement for $\\mathbb{P}^n\\times \\mathbb{P}^1$. For an arbitrary $X$\nand an ample line bundle $L$ on $X$ we prove the existence of an integer $k_0$\nsuch that for all $t\\ge k_0$ the tangential variety of $X$ with respect to\n$|L^{\\otimes t}|$ is not defective.\n"}
{"abstract": "  Establishing cause-effect relationships from observational data often relies\non untestable assumptions. It is crucial to know whether, and to what extent,\nthe conclusions drawn from non-experimental studies are robust to potential\nunmeasured confounding. In this paper, we focus on the average causal effect\n(ACE) as our target of inference. We build on the work of Franks et al.\n(2019)and Robins (2000) by specifying non-identified sensitivity parameters\nthat govern a contrast between the conditional (on measured covariates)\ndistributions of the outcome under treatment (control) between treated and\nuntreated individuals. We use semiparametric theory to derive the\nnon-parametric efficient influence function of the ACE, for fixed sensitivity\nparameters. We use this influence function to construct a one-step\nbias-corrected estimator of the ACE. Our estimator depends on semiparametric\nmodels for the distribution of the observed data; importantly, these models do\nnot impose any restrictions on the values of sensitivity analysis parameters.\nWe establish sufficient conditions ensuring that our estimator has root-n\nasymptotics. We use our methodology to evaluate the causal effect of smoking\nduring pregnancy on birth weight. We also evaluate the performance of\nestimation procedure in a simulation study.\n"}
{"abstract": "  We introduce a semi-parametric method to simultaneously infer both the drift\nand volatility functions of a discretely observed scalar diffusion. We\nintroduce spline bases to represent these functions and develop a Markov chain\nMonte Carlo algorithm to infer, a posteriori, the coefficients of these\nfunctions in the spline basis. A key innovation is that we use spline bases to\nmodel transformed versions of the drift and volatility functions rather than\nthe functions themselves. The output of the algorithm is a posterior sample of\nplausible drift and volatility functions that are not constrained to any\nparticular parametric family. The flexibility of this approach provides\npractitioners a powerful investigative tool, allowing them to posit parametric\nmodels to better capture the underlying dynamics of their processes of\ninterest. We illustrate the versatility of our method by applying it to\nchallenging datasets from finance, paleoclimatology, and astrophysics. In view\nof the parametric diffusion models widely employed in the literature for those\nexamples, some of our results are surprising since they call into question some\naspects of these models.\n"}
{"abstract": "  We present GIGANTES, the most extensive and realistic void catalog suite ever\nreleased -- containing over 1 billion cosmic voids covering a volume larger\nthan the observable Universe, more than 20 TB of data, and created by running\nthe void finder VIDE on QUIJOTE's halo simulations. The expansive and detailed\nGIGANTES suite, spanning thousands of cosmological models, opens up the study\nof voids, answering compelling questions: Do voids carry unique cosmological\ninformation? How is this information correlated with galaxy information?\nLeveraging the large number of voids in the GIGANTES suite, our Fisher\nconstraints demonstrate voids contain additional information, critically\ntightening constraints on cosmological parameters. We use traditional void\nsummary statistics (void size function, void density profile) and the void\nauto-correlation function, which independently yields an error of\n$0.13\\,\\mathrm{eV}$ on $\\sum\\,m_{\\nu}$ for a 1 $h^{-3}\\mathrm{Gpc}^3$\nsimulation, without CMB priors. Combining halos and voids we forecast an error\nof $0.09\\,\\mathrm{eV}$ from the same volume. Extrapolating to next generation\nmulti-Gpc$^3$ surveys such as DESI, Euclid, SPHEREx, and the Roman Space\nTelescope, we expect voids should yield an independent determination of\nneutrino mass. Crucially, GIGANTES is the first void catalog suite expressly\nbuilt for intensive machine learning exploration. We illustrate this by\ntraining a neural network to perform likelihood-free inference on the void size\nfunction. Cosmology problems provide an impetus to develop novel deep learning\ntechniques, leveraging the symmetries embedded throughout the universe from\nphysical laws, interpreting models, and accurately predicting errors. With\nGIGANTES, machine learning gains an impressive dataset, offering unique\nproblems that will stimulate new techniques.\n"}
{"abstract": "  The Large Hadron Collider (LHC) at CERN is undergoing a major upgrade with\nthe goal of increasing the luminosity as more statistics are needed for\nprecision measurements. The presented work pertains to the corresponding\nupgrade of the ALICE Muon Trigger (MTR) Detector, now named the Muon Identifier\n(MID). Previously operated in a triggered readout manner, this detector has\ntransitioned to continuous readout with time-delimited data payloads. However,\nthis results in data rates much higher than the previous operation and hence a\nnew Online-Offline (O2) computing system is also being developed for real-time\ndata processing to reduce the storage requirements. Part of the O2 System is\nbased on FPGA technology and is known as the Common Readout Unit (CRU). Being\ncommon to many detectors necessitates the development of custom user logic per\ndetector. This work concerns the development of the ALICE MID user logic which\nwill interface to the core CRU firmware and perform the required data\nprocessing. It presents the development of a conceptual design and a prototype\nfor the user logic. The resulting prototype shows the ability to meet the\nestablished requirements in an effective and optimized manner. Additionally,\nthe modular design approach employed, allows for more features to be easily\nintroduced.\n"}
{"abstract": "  An extension of Riemmann's geometry into a direction dependent geometric\nstructure is usually described by Finsler's geometry. Historically, this\nconstruction was motivated by the well-known Riemann's quartic length element\nexample. Quite surprisingly, the same quartic expression emerges in solid-state\nelectrodynamics as a basic dispersion relation -- covariant Fresnel equation.\nConsequently, Riemann's quartic length expression can be interpreted as a\nmathematical model of a well-established physics phenomena. In this paper, we\npresent various examples of Riemann's quartic that demonstrate that Finsler's\ngeometry is too restrictive even in the case of a positive definite Euclidean\nsignature space. In the case of the spaces endowed with an indefinite\n(Minkowski) signature, there are much more singular hypersurfaces where the\nstrong axioms of Finsler's geometry are broken down. We propose a weaker\ndefinition of Finsler's structure that is required to be satisfied only on open\nsubsets of the tangent bundle. We exhibit the characteristic singular\nhypersurfaces related to Riemann's quartic and briefly discuss their possible\nphysical interpretation.\n"}
{"abstract": "  We present an algorithm for data-driven reachability analysis that estimates\nfinite-horizon forward reachable sets for general nonlinear systems using level\nsets of a certain class of polynomials known as Christoffel functions. The\nlevel sets of Christoffel functions are known empirically to provide good\napproximations to the support of probability distributions: the algorithm uses\nthis property for reachability analysis by solving a probabilistic relaxation\nof the reachable set computation problem. We also provide a guarantee that the\noutput of the algorithm is an accurate reachable set approximation in a\nprobabilistic sense, provided that a certain sample size is attained. We also\ninvestigate three numerical examples to demonstrate the algorithm's\ncapabilities, such as providing non-convex reachable set approximations and\ndetecting holes in the reachable set.\n"}
{"abstract": "  We consider the cubic and quintic nonlinear Schr\\\"{o}dinger equations (NLS)\nunder the $\\mathbb{R}^{d}$ and $\\mathbb{T}^{d}$ energy-supercritical setting.\nVia a newly developed unified scheme, we prove the unconditional uniqueness for\nsolutions to NLS at critical regularity for all dimensions. Thus, together with\n[18,19], the unconditional uniqueness problems for $H^{1}$-critical and\n$H^{1}$-supercritical cubic and quintic NLS are completely and uniformly\nresolved at critical regularity for these domains. One application of our\ntheorem is to prove that defocusing blowup solutions of the type in [54] is the\nonly possible $C([0,T);\\dot{H}^{s_{c}})$ solution if exist in these domains.\n"}
{"abstract": "  We present TNCR, a new table dataset with varying image quality collected\nfrom free websites. The TNCR dataset can be used for table detection in scanned\ndocument images and their classification into 5 different classes. TNCR\ncontains 9428 high-quality labeled images. In this paper, we have implemented\nstate-of-the-art deep learning-based methods for table detection to create\nseveral strong baselines. Cascade Mask R-CNN with ResNeXt-101-64x4d Backbone\nNetwork achieves the highest performance compared to other methods with a\nprecision of 79.7%, recall of 89.8%, and f1 score of 84.4% on the TNCR dataset.\nWe have made TNCR open source in the hope of encouraging more deep learning\napproaches to table detection, classification, and structure recognition. The\ndataset and trained model checkpoints are available at\nhttps://github.com/abdoelsayed2016/TNCR_Dataset.\n"}
{"abstract": "  The existence of charged black holes has suggested that wormholes may also be\ncharged. The purpose of this paper is to construct a general model of a charged\nwormhole that proves to be a natural extension of the original Morris-Thorne\nwormhole. This goal is achieved by means of the classical embedding theory that\nhas played a major role in the general theory of relativity.\n"}
{"abstract": "  Herbarium contains treasures of millions of specimens which have been\npreserved for several years for scientific studies. To speed up more scientific\ndiscoveries, a digitization of these specimens is currently on going to\nfacilitate easy access and sharing of its data to a wider scientific community.\nOnline digital repositories such as IDigBio and GBIF have already accumulated\nmillions of specimen images yet to be explored. This presents a perfect time to\nautomate and speed up more novel discoveries using machine learning and\ncomputer vision. In this study, a thorough analysis and comparison of more than\n50 peer-reviewed studies which focus on application of computer vision and\nmachine learning techniques to digitized herbarium specimen have been examined.\nThe study categorizes different techniques and applications which have been\ncommonly used and it also highlights existing challenges together with their\npossible solutions. It is our hope that the outcome of this study will serve as\na strong foundation for beginners of the relevant field and will also shed more\nlight for both computer science and ecology experts.\n"}
{"abstract": "  We study the contributions to the relativistic Fe $K_{\\alpha}$ line profile\nfrom higher order images (HOIs) produced by strongly deflected rays from the\ndisk which cross the plunging region, located between the innermost stable\ncircular orbit (ISCO) radius and the event horizon of a Kerr black hole. We\ninvestigate the characteristics features imprinted by the HOIs in the line\nprofile for different black hole spins, disk emissivity laws and inclinations.\nWe find that they extend from the red wing of the profile up to energies\nslightly lower than those of the blue peak, adding $\\sim 0.4 - 1.3$\\% to the\ntotal line flux. The contribution to the specific flux is often in the $\\sim\n1$\\% to 7\\% range, with the highest values attained for low and negative spin\n($a\\lesssim 0.3$) black holes surrounded by intermediate inclination angle\n($i\\sim40^{\\circ}$) disks. We simulate future observations of a black hole\nX-ray binary system with the Large Area Detector of the planned X-ray astronomy\n\\emph{enhanced X-ray Timing and Polarimetry Mission} (eXTP) and find that the\n\\fekal\\ of systems accreting at $\\lesssim 1 $\\% the Eddington rate are affected\nby the HOI features for a range of parameters. This would provide evidence of\nthe extreme gravitational lensing of HOI rays. Our simulations show also that\nnot accounting for HOI contributions to the Fe $K_{\\alpha}$ line profile may\nsystematically bias measurements of the black hole spin parameter towards\nvalues higher by up to $\\sim 0.3$ than the inputted ones.\n"}
{"abstract": "  Stress-induced martensitic transformations enable metastable alloys to\nexhibit enhanced strain hardening capacity, leading to improved formability and\ntoughness. As is well-known from transformation-induced plasticity (TRIP)\nsteels, however, the resulting martensite can limit ductility and fatigue life\ndue to its intrinsic brittleness. In this work, we explore an alloy design\nstrategy that utilizes stress-induced martensitic transformations but does not\nretain the martensite phase. This strategy is based on the introduction of\nsuperelastic nano-precipitates, which exhibit reverse transformation after\ninitial stress-induced forward transformation. To this end, utilizing ab-initio\nsimulations and thermodynamic calculations we designed and produced a\nV45Ti30Ni25 (at%) alloy. In this alloy, TiNi is present as nano-precipitates\nuniformly distributed within a ductile V-rich base-centered cubic (bcc) beta\nmatrix, as well as being present as a larger matrix phase. We characterized the\nmicrostructure of the produced alloy using various scanning electron microscopy\n(SEM) and transmission electron microscopy (TEM) methods. The bulk mechanical\nproperties of the alloy are demonstrated through tensile tests, and the\nreversible transformation in each of the TiNi morphologies were confirmed by\nin-situ TEM micro-pillar compression experiments, in-situ high-energy\ndiffraction synchrotron cyclic tensile tests, indentation experiments, and\ndifferential scanning calorimetry experiments. The observed transformation\npathways and variables impacting phase stability are critically discussed\n"}
{"abstract": "  In a wide range of applications, the estimate of droplet evaporation time is\nbased on the classical d2-law, which, assuming a fast mixing and fixed\nenvironmental properties, states that the droplet surface decreases linearly\nwith time at a determined rate. However, in many cases the predicted\nevaporation rate is overestimated. In this Letter, we propose a revision of the\nd2-law capable to accurately determine droplet evaporation rate in dilute\nconditions by a proper estimate of the asymptotic droplet properties. Besides a\ndiscussion of the main assumptions, we tested the proposed model against data\nfrom direct numerical simulations finding an excellent agreement for predicted\ndroplet evaporation time in dilute turbulent jet-sprays.\n"}
{"abstract": "  We prove the Gross order of vanishing conjecture in special cases where the\nvanishing order of the character in question can be arbitrarily large. In\nalmost all previously known cases the vanishing order is zero or one. One major\ningredient of our proofs is the equivalence of this conjecture to the\nGross-Kuz'min conjecture. We present here a direct proof of this equivalence,\nusing only the known validity of the Iwasawa Main Conjecture over totally real\nfields.\n"}
{"abstract": "  Within the framework of the Lindblad master equation, we propose a general\nmethodology to describe the effects of the environment on a system in dilute\ngas phase. The phenomenological parameters characterizing the transitions\nbetween rovibrational states of the system induced by collisions can be\nextracted from experimental transition kinetic constants, relying on Energy Gap\nfitting laws. As the availability of this kind of experimental data can be\nlimited, the present work relied on experimental line broadening coefficients,\nhowever still using Energy Gap fitting laws. The 3 $\\mu$m infrared spectral\nrange of acetylene was chosen to illustrate the proposed approach. The method\nshows fair agreement with available experimental data while being\ncomputationally inexpensive. The results are discussed in the context of state\nlaser quantum control.\n"}
{"abstract": "  We consider the near-critical Erd\\H{o}s-R\\'{e}nyi random graph $G(n,p)$ and\nprovide a new probabilistic proof of the fact that, when $p$ is of the form\n$p=p(n)=1/n+\\lambda/n^{4/3}$ and $A$ is large,\n\\[\\mathbb{P}(|\\mathcal{C}_{\\max}|>An^{2/3})\\asymp\nA^{-3/2}e^{-\\frac{A^3}{8}+\\frac{\\lambda A^2}{2}-\\frac{\\lambda^2A}{2}}\\] where\n$\\mathcal{C}_{\\max}$ is the largest connected component of the graph. Our\nresult allows $A$ and $\\lambda$ to depend on $n$. While this result is already\nknown, our proof relies only on conceptual and adaptable tools such as ballot\ntheorems, whereas the existing proof relies on a combinatorial formula specific\nto Erd\\H{o}s-R\\'{e}nyi graphs, together with analytic estimates.\n"}
{"abstract": "  For the general parametric regression models with covariates contaminated\nwith normal measurement errors, this paper proposes an accelerated version of\nthe classical simulation extrapolation algorithm to estimate the unknown\nparameters in the regression function. By applying the conditional expectation\ndirectly to the target function, the proposed algorithm successfully removes\nthe simulation step, by generating an estimation equation either for immediate\nuse or for extrapolating, thus significantly reducing the computational time.\nLarge sample properties of the resulting estimator, including the consistency\nand the asymptotic normality, are thoroughly discussed. Potential wide\napplications of the proposed estimation procedure are illustrated by examples,\nsimulation studies, as well as a real data analysis.\n"}
{"abstract": "  Microstructural materials design is one of the most important applications of\ninverse modeling in materials science. Generally speaking, there are two broad\nmodeling paradigms in scientific applications: forward and inverse. While the\nforward modeling estimates the observations based on known parameters, the\ninverse modeling attempts to infer the parameters given the observations.\nInverse problems are usually more critical as well as difficult in scientific\napplications as they seek to explore the parameters that cannot be directly\nobserved. Inverse problems are used extensively in various scientific fields,\nsuch as geophysics, healthcare and materials science. However, it is\nchallenging to solve inverse problems, because they usually need to learn a\none-to-many non-linear mapping, and also require significant computing time,\nespecially for high-dimensional parameter space. Further, inverse problems\nbecome even more difficult to solve when the dimension of input (i.e.\nobservation) is much lower than that of output (i.e. parameters). In this work,\nwe propose a framework consisting of generative adversarial networks and\nmixture density networks for inverse modeling, and it is evaluated on a\nmaterials science dataset for microstructural materials design. Compared with\nbaseline methods, the results demonstrate that the proposed framework can\novercome the above-mentioned challenges and produce multiple promising\nsolutions in an efficient manner.\n"}
{"abstract": "  Morales-Mendoza et al. present in 2013 a new class of discrete orthogonal\npolynomials. They use these polynomials to design an unbiased FIR filter. In\ntheir paper they make the statement that a representation of the polynomials\nvia hypergeometric functions is unknown. However Shakibaei Asli et al. found in\n2017 a $_3F_2$ hypergeometric function representation. In this paper it is\nshown that the \"new\" orthogonal polynomials belong to the class of the Hahn\npolynomials. The transfer function of the unbiased FIR filter has been\ndetermined. In the second part, an orthogonal unbiased FIR filter is designed\nfollowing the method of the orthogonal derivative described in the thesis of\nthe author. In the Appendix a table with a number of possible transformations\nof the Hahn polynomials is given.\n"}
{"abstract": "  Abductive reasoning starts from some observations and aims at finding the\nmost plausible explanation for these observations. To perform abduction, humans\noften make use of temporal and causal inferences, and knowledge about how some\nhypothetical situation can result in different outcomes. This work offers the\nfirst study of how such knowledge impacts the Abductive NLI task -- which\nconsists in choosing the more likely explanation for given observations. We\ntrain a specialized language model LMI that is tasked to generate what could\nhappen next from a hypothetical scenario that evolves from a given event. We\nthen propose a multi-task model MTL to solve the Abductive NLI task, which\npredicts a plausible explanation by a) considering different possible events\nemerging from candidate hypotheses -- events generated by LMI -- and b)\nselecting the one that is most similar to the observed outcome. We show that\nour MTL model improves over prior vanilla pre-trained LMs fine-tuned on\nAbductive NLI. Our manual evaluation and analysis suggest that learning about\npossible next events from different hypothetical scenarios supports abductive\ninference.\n"}
{"abstract": "  Fooling deep neural networks (DNNs) with the black-box optimization has\nbecome a popular adversarial attack fashion, as the structural prior knowledge\nof DNNs is always unknown. Nevertheless, recent black-box adversarial attacks\nmay struggle to balance their attack ability and visual quality of the\ngenerated adversarial examples (AEs) in tackling high-resolution images. In\nthis paper, we propose an attention-guided black-box adversarial attack based\non the large-scale multiobjective evolutionary optimization, termed as LMOA. By\nconsidering the spatial semantic information of images, we firstly take\nadvantage of the attention map to determine the perturbed pixels. Instead of\nattacking the entire image, reducing the perturbed pixels with the attention\nmechanism can help to avoid the notorious curse of dimensionality and thereby\nimproves the performance of attacking. Secondly, a large-scale multiobjective\nevolutionary algorithm is employed to traverse the reduced pixels in the\nsalient region. Benefiting from its characteristics, the generated AEs have the\npotential to fool target DNNs while being imperceptible by the human vision.\nExtensive experimental results have verified the effectiveness of the proposed\nLMOA on the ImageNet dataset. More importantly, it is more competitive to\ngenerate high-resolution AEs with better visual quality compared with the\nexisting black-box adversarial attacks.\n"}
{"abstract": "  Hysteretic magnetoresistance (MR) is often used as a signature of\nferromagnetism in conducting oxide thin films and heterostructures. Here,\nmagnetotransport is investigated in a non-magnetic uniformly La-doped SrSnO3\nfilm grown using hybrid molecular beam epitaxy. A 12 nm La:SrSnO3/2 nm\nSrSnO3/GdScO3 (110) film with insulating behavior exhibited a robust hysteresis\nloop in the MR at T < 5 K accompanied by an anomaly at ~ +/- 3 T at T < 2.5 K.\nFurthermore, MR with the field in-plane yielded a value exceeded 100% at 1.8 K.\nUsing detailed temperature-, angle- and magnetic field-dependent resistance\nmeasurements, we illustrate the origin of hysteresis is not due to magnetism in\nthe film but rather is associated with the magnetocaloric effect of the GdScO3\nsubstrate. Given GdScO3 and similar substrates are commonly used in complex\noxide research, this work highlights the importance of thermal coupling to\nprocesses in the substrates which must be carefully accounted for in the data\ninterpretation for thin films and heterostructures utilizing these substrates.\n"}
{"abstract": "  An alternative scheme is described to estimate the layer 1 LAXPC 20\nbackground for faint sources where the source contribution to the 50-80 keV\ncount rate is less than 0.25 counts/sec (15 milli-crabs or $6 \\times 10^{-11}$\nergs/s/cm$^2$). We consider 12 blank sky observations and based on their 50-80\nkeV count rate in 100 second time-bins, generate four template spectra which\nare then used to estimate the background spectrum and lightcurve for a given\nfaint source observation. The variance of the estimated background subtracted\nspectra for the 12 blank sky observations is taken as the energy dependent\nsystematic uncertainty which will dominate over the statistical one for\nexposures longer than 5 ksecs. The estimated 100 second time bin background\nlightcurve in the 4-20 keV band with a 3\\% systematic error matches with the\nblank sky ones. The 4-20 keV spectrum can be constrained for a source with flux\n$\\gtrapprox 1$ milli-crab. Fractional r.m.s variability of 10\\% can be\ndetermined for a $\\sim 5$ milli-crab source lightcurve binned at 100 seconds.\nTo illustrate the scheme, the lightcurves, and spectra of three different blank\nsky observations, three AGN sources (Mrk 0926, Mrk 110, NGC 4593), and LMC X-1\nare shown.\n"}
{"abstract": "  We propose that Hawking radiation-like phenomena may be observed in systems\nthat show butterfly effects. Suppose that a classical dynamical system has a\nLyapunov exponent $\\lambda_L$, and is deterministic and non-thermal ($T=0$). We\nargue that, if we quantize this system, the quantum fluctuations may imitate\nthermal fluctuations with temperature $T \\sim \\hbar \\lambda_L/2 \\pi $ in a\nsemi-classical regime, and it may cause analogous Hawking radiation. We also\ndiscuss that our proposal may provide an intuitive explanation of the existence\nof the bound of chaos proposed by Maldacena, Shenker and Stanford.\n"}
{"abstract": "  Determining the factors that have an influence on software systems\ndevelopment and deployment project outcomes has been the focus of extensive and\nongoing research for more than 30 years. We provide here a survey of the\nresearch literature that has addressed this topic in the period 1996-2006, with\na particular focus on empirical analyses. On the basis of this survey we\npresent a new classification framework that represents an abstracted and\nsynthesized view of the types of factors that have been asserted as influencing\nproject outcomes.\n"}
{"abstract": "  National and local governments have implemented a large number of policies,\nparticularly non-pharmaceutical interventions, in response to the Covid-19\npandemic. Evaluating the effects of these policies, both on the number of\nCovid-19 cases as well on other economic outcomes is a key ingredient for\npolicymakers to be able to determine which policies are most effective as well\nas the relative costs and benefits of particular policies. In this paper, we\nconsider the relative merits of common identification strategies exploiting\nvariation in policy choices made across different locations by checking whether\nthe identification strategies are compatible with leading epidemic models in\nthe epidemiology literature. We argue that unconfoundedness type approaches are\nlikely to be more useful for evaluating policies than difference in differences\ntype approaches due to the highly nonlinear spread of cases during a pandemic.\nFor difference in differences, we further show that a version of this problem\ncontinues to exist even when one is interested in understanding the effect of a\npolicy on other economic outcomes when those outcomes also depend on the number\nof Covid-19 cases. We propose alternative approaches that are able to\ncircumvent these issues. We apply our proposed approach to study the effect of\nstate level shelter-in-place orders early in the pandemic.\n"}
{"abstract": "  The impact of potential future measurements of beam charge asymmetries on the\ncurrent knowledge of Compton form factors is evaluated. Elaborating on the\nresults of a global neural network fit to deeply virtual Compton scattering\ndata, a Bayesian reweighting analysis quantifies the improved determination of\nthe real part of the Compton form factor $\\mathcal{H}$. Such an improvement is\nparticularly relevant for the experimental determination of the proton\nmechanical properties or for universality studies of generalized parton\ndistributions through analyses of virtual Compton scattering in both spacelike\nand timelike regions.\n"}
{"abstract": "  This paper identifies and addresses a serious design bias of existing salient\nobject detection (SOD) datasets, which unrealistically assume that each image\nshould contain at least one clear and uncluttered salient object. This design\nbias has led to a saturation in performance for state-of-the-art SOD models\nwhen evaluated on existing datasets. However, these models are still far from\nsatisfactory when applied to real-world scenes. Based on our analyses, we\npropose a new high-quality dataset and update the previous saliency benchmark.\nSpecifically, our dataset, called Salient Objects in Clutter (SOC), includes\nimages with both salient and non-salient objects from several common object\ncategories. In addition to object category annotations, each salient image is\naccompanied by attributes that reflect common challenges in real-world scenes,\nwhich can help provide deeper insight into the SOD problem. Further, with a\ngiven saliency encoder, e.g., the backbone network, existing saliency models\nare designed to achieve mapping from the training image set to the training\nground-truth set. We, therefore, argue that improving the dataset can yield\nhigher performance gains than focusing only on the decoder design. With this in\nmind, we investigate several dataset-enhancement strategies, including label\nsmoothing to implicitly emphasize salient boundaries, random image augmentation\nto adapt saliency models to various scenarios, and self-supervised learning as\na regularization strategy to learn from small datasets. Our extensive results\ndemonstrate the effectiveness of these tricks. We also provide a comprehensive\nbenchmark for SOD, which can be found in our repository:\nhttp://dpfan.net/SOCBenchmark.\n"}
{"abstract": "  We derive an effective field theory for a type-II fracton starting from the\nHaah code on the lattice. The effective topological theory is not given\nexclusively in terms of an action; it must be supplemented with a condition\nthat selects physical states. Without the constraint, the action only describes\na type-I fracton. The constraint emerges from a condition that cube operators\nmultiply to the identity, and it cannot be consistently implemented in the\ncontinuum theory at the operator level, but only in a weaker form, in terms of\nmatrix elements of physical states. Informed by these studies and starting from\nthe opposite end, i.e., the continuum, we discuss a Chern-Simons-like theory\nthat does not need a constraint or projector, and yet has no mobile\nexcitations. Whether this continuum theory admits a lattice counterpart remains\nunanswered.\n"}
{"abstract": "  As a wide variety of quantum computing platforms become available, methods\nfor assessing and comparing the performance of these devices are of increasing\ninterest and importance. Inspired by the success of single-qubit error rate\ncomputations for tracking the progress of gate-based quantum computers, this\nwork proposes a Quantum Annealing Single-qubit Assessment (QASA) protocol for\nquantifying the performance of individual qubits in quantum annealing\ncomputers. The proposed protocol scales to large quantum annealers with\nthousands of qubits and provides unique insights into the distribution of qubit\nproperties within a particular hardware device. The efficacy of the QASA\nprotocol is demonstrated by analyzing the properties of a D-Wave 2000Q system,\nrevealing unanticipated correlations in the qubit performance of that device. A\nstudy repeating the QASA protocol at different annealing times highlights how\nthe method can be utilized to understand the impact of annealing parameters on\nqubit performance. Overall, the proposed QASA protocol provides a useful tool\nfor assessing the performance of current and emerging quantum annealing\ndevices.\n"}
{"abstract": "  Data-driven Reynolds-Averaged Navier-Stokes (RANS) turbulence closures are\nincreasing seen as a viable alternative to general-purpose RANS closures, when\nLES reference data is available -- also in wind-energy. Parsimonious closures\nwith few, simple terms have advantages in terms of stability,\ninterpret-ability, and execution speed. However experience suggests that\nclosure model corrections need be made only in limited regions -- e.g.\\ in the\nnear-wake of wind turbines and not in the majority of the flow. A parsimonious\nmodel therefore must find a middle ground between precise corrections in the\nwake, and zero corrections elsewhere. We attempt to resolve this impasse by\nintroducing a classifier to identify regions needing correction, and only fit\nand apply our model correction there. We observe that such a classifier (which\nmust be computed only from RANS-available quantities) is straightforward to\nconstruct, and accurate in operation. We further observe that the correction\nmodels are significantly simpler (with fewer terms) when limited to the\nidentified regions than otherwise, but have similar or better accuracy. We\napply our framework to three flows consisting of multiple wind-turbines in\nneutral conditions with interacting wakes.\n"}
{"abstract": "  In this paper we present an analysis of the spin behavior of electrons\npropagating through a laser field. We present an experimentally realizable\nscenario in which spin-dependent effects of the interaction between the laser\nand the electrons are dominant. The laser interaction strength and incident\nelectron velocity are in the nonrelativistic domain. This analysis may thus\nlead to novel methods of creating and characterizing spin-polarized\nnonrelativistic femtosecond electron pulses.\n"}
{"abstract": "  We suggest a generalization of nonlinear $\\sigma$-model for diffusive\nsuperconducting systems to account for magnetoelectric effects due to\nspin-orbit scattering. In the leading orders of spin-orbit strength and\ngradient expansion it includes two additional terms responsible for the\nspin-Hall effect and the spin-current swapping. First, assuming a\ndelta-correlated disorder we derive the new terms from the Keldysh path\nintegral representation of the generating functional. Then we argue\nphenomenologically that they exhaust all invariants allowed in the effective\naction to the leading order in the spin-orbit coupling (SOC). Finally, the\nresults are confirmed by a direct derivation of the saddle-point (Usadel)\nequation from the quantum kinetic equations in the presence of randomly\ndistributed impurities with SOC. At this point we correct a recent derivation\nof the Usadel equation that includes magneto-electric effects and does not\nresort to the Born approximation.\n"}
{"abstract": "  Graph Neural Networks (GNNs) perform learned message passing over an input\ngraph, but conventional wisdom says performing more than handful of steps makes\ntraining difficult and does not yield improved performance. Here we show the\ncontrary. We train a deep GNN with up to 100 message passing steps and achieve\nseveral state-of-the-art results on two challenging molecular property\nprediction benchmarks, Open Catalyst 2020 IS2RE and QM9. Our approach depends\ncrucially on a novel but simple regularisation method, which we call ``Noisy\nNodes'', in which we corrupt the input graph with noise and add an auxiliary\nnode autoencoder loss if the task is graph property prediction. Our results\nshow this regularisation method allows the model to monotonically improve in\nperformance with increased message passing steps. Our work opens new\nopportunities for reaping the benefits of deep neural networks in the space of\ngraph and other structured prediction problems.\n"}
{"abstract": "  Quantum spin liquids are topological states of matter that arise in\nfrustrated quantum magnets at low temperatures. At low energies, such states\nexhibit emergent gauge fields and fractionalized quasiparticles and can also\npossess enhanced global symmetries compared to their parent microscopic\nHamiltonians. We study the consequences of this emergent gauge and symmetry\nstructure for the hydrodynamics of quantum spin liquids. Specifically, we\nanalyze two cases, the $U(1)$ spin liquid with a Fermi surface and the\n$SU(4)$-symmetric \"algebraic\" spin liquid. We show that the emergent degrees of\nfreedom in the spin liquid phase lead to a variety of additional hydrodynamic\nmodes compared to the high-temperature paramagnetic phase. We identify a\nhydrodynamic regime for the internal $U(1)$ gauge field common to both states,\ncharacterized by slow diffusion of the internal transverse photon.\n"}
{"abstract": "  Medical data sets are usually corrupted by noise and missing data. These\nmissing patterns are commonly assumed to be completely random, but in medical\nscenarios, the reality is that these patterns occur in bursts due to sensors\nthat are off for some time or data collected in a misaligned uneven fashion,\namong other causes. This paper proposes to model medical data records with\nheterogeneous data types and bursty missing data using sequential variational\nautoencoders (VAEs). In particular, we propose a new methodology, the Shi-VAE,\nwhich extends the capabilities of VAEs to sequential streams of data with\nmissing observations. We compare our model against state-of-the-art solutions\nin an intensive care unit database (ICU) and a dataset of passive human\nmonitoring. Furthermore, we find that standard error metrics such as RMSE are\nnot conclusive enough to assess temporal models and include in our analysis the\ncross-correlation between the ground truth and the imputed signal. We show that\nShi-VAE achieves the best performance in terms of using both metrics, with\nlower computational complexity than the GP-VAE model, which is the\nstate-of-the-art method for medical records.\n"}
{"abstract": "  Deep generative models, such as Variational Autoencoders (VAEs), have been\nemployed widely in computational creativity research. However, such models\ndiscourage out-of-distribution generation to avoid spurious sample generation,\nlimiting their creativity. Thus, incorporating research on human creativity\ninto generative deep learning techniques presents an opportunity to make their\noutputs more compelling and human-like. As we see the emergence of generative\nmodels directed to creativity research, a need for machine learning-based\nsurrogate metrics to characterize creative output from these models is\nimperative. We propose group-based subset scanning to quantify, detect, and\ncharacterize creative processes by detecting a subset of anomalous\nnode-activations in the hidden layers of generative models. Our experiments on\noriginal, typically decoded, and \"creatively decoded\" (Das et al 2020) image\ndatasets reveal that the proposed subset scores distribution is more useful for\ndetecting creative processes in the activation space rather than the pixel\nspace. Further, we found that creative samples generate larger subsets of\nanomalies than normal or non-creative samples across datasets. The node\nactivations highlighted during the creative decoding process are different from\nthose responsible for normal sample generation.\n"}
{"abstract": "  Basing on some recently proposed methods for solving variational inequalities\nwith non-smooth operators, we propose an analogue of the Mirror Prox method for\nthe corresponding class of problems under the assumption of relative smoothness\nand relative strong monotonicity of the operator.\n"}
{"abstract": "  In this paper, we propose deep architectures for learning instance specific\nabstain (reject option) binary classifiers. The proposed approach uses double\nsigmoid loss function as described by Kulin Shah and Naresh Manwani in (\"Online\nActive Learning of Reject Option Classifiers\", AAAI, 2020), as a performance\nmeasure. We show that the double sigmoid loss is classification calibrated. We\nalso show that the excess risk of 0-d-1 loss is upper bounded by the excess\nrisk of double sigmoid loss. We derive the generalization error bounds for the\nproposed architecture for reject option classifiers. To show the effectiveness\nof the proposed approach, we experiment with several real world datasets. We\nobserve that the proposed approach not only performs comparable to the\nstate-of-the-art approaches, it is also robust against label noise. We also\nprovide visualizations to observe the important features learned by the network\ncorresponding to the abstaining decision.\n"}
{"abstract": "  The Convolutional Neural Network (CNN) is a state-of-the-art architecture for\na wide range of deep learning problems, the quintessential example of which is\ncomputer vision. CNNs principally employ the convolution operation, which can\nbe accelerated using the Fourier transform. In this paper, we present an\noptical hardware accelerator that combines silicon photonics and free-space\noptics, leveraging the use of the optical Fourier transform within several CNN\narchitectures. The hardware presented is a proof of concept, demonstrating that\nthis technology can be applied to artificial intelligence problems with a large\nefficiency boost with respect to canonical methods.\n"}
{"abstract": "  Anti-ThCr$_2$Si$_2$ type $RE_2$O$_2$Sb ($RE$ = rare earth) with Sb square net\nhas shown insulating conduction so far. Here we report the synthesis of\nLa$_2$O$_2$Sb epitaxial thin films for the first time by multilayer solid-phase\nepitaxy. The valence state of Sb was about -2 evaluated from X-ray\nphotoemission spectroscopy measurement, and the indirect band gap of 0.17 eV\nwas observed. The La$_2$O$_2$Sb epitaxial thin film showed unexpectedly high\nelectrical conduction as a narrow gap semiconductor, whose resistivity at room\ntemperature was approximately ten-thousand-fold lower than that of\nLa$_2$O$_2$Sb bulk polycrystal, attributed to increased carrier mobility\nprobably due to suppressed Sb dimerization.\n"}
{"abstract": "  The Beijing-Arizona Sky Survey (BASS) Data Release 3 (DR3) catalogue was\nreleased in 2019, which contains the data from all BASS and the Mosaic z-band\nLegacy Survey (MzLS) observations during 2015 January and 2019 March, about 200\nmillion sources. We cross-match BASS DR3 with spectral databases from the Sloan\nDigital Sky Survey (SDSS) and the Large Sky Area Multi-object Fiber\nSpectroscopic Telescope (LAMOST) to obtain the spectroscopic classes of known\nsamples. Then, the samples are cross-matched with ALLWISE database. Based on\noptical and infrared information of the samples, we use the XGBoost algorithm\nto construct different classifiers, including binary classification and\nmulticlass classification. The accuracy of these classifiers with the best\ninput pattern is larger than 90.0 per cent. Finally, all selected sources in\nthe BASS DR3 catalogue are classified by these classifiers. The classification\nlabel and probabilities for individual sources are assigned by different\nclassifiers. When the predicted results by binary classification are the same\nas multiclass classification with optical and infrared information, the number\nof star, galaxy and quasar candidates is separately 12 375 838 (P_S>0.95), 18\n606 073 (P_G>0.95) and 798 928 (P_Q>0.95). For these sources without infrared\ninformation, the predicted results can be as a reference. Those candidates may\nbe taken as input catalogue of LAMOST, DESI or other projects for follow up\nobservation. The classified result will be of great help and reference for\nfuture research of the BASS DR3 sources.\n"}
{"abstract": "  Visualization recommendation (VisRec) systems provide users with suggestions\nfor potentially interesting and useful next steps during exploratory data\nanalysis. These recommendations are typically organized into categories based\non their analytical actions, i.e., operations employed to transition from the\ncurrent exploration state to a recommended visualization. However, despite the\nemergence of a plethora of VisRec systems in recent work, the utility of the\ncategories employed by these systems in analytical workflows has not been\nsystematically investigated. Our paper explores the efficacy of recommendation\ncategories by formalizing a taxonomy of common categories and developing a\nsystem, Frontier, that implements these categories. Using Frontier, we evaluate\nworkflow strategies adopted by users and how categories influence those\nstrategies. Participants found recommendations that add attributes to enhance\nthe current visualization and recommendations that filter to sub-populations to\nbe comparatively most useful during data exploration. Our findings pave the way\nfor next-generation VisRec systems that are adaptive and personalized via\ncarefully chosen, effective recommendation categories.\n"}
{"abstract": "  Cycle sets are algebraic structures introduced by Rump to study set theoretic\nsolutions to the Yang-Baxter equation. While studying cycle sets Rump also\nintroduced braces, which have since overtaken cycle sets as a tool for studying\nsolutions. This survey paper is primarily an introduction to cycle sets,\nmotivating their study and relating them to key results of brace theory and\nYang-Baxter theory. It is aimed at anyone from those already very familiar with\nbraces but less familiar with cycle sets, to those with only a basic level of\nbackground in ring theory and group theory. We introduce cycle sets following\nRump's original results - giving more detailed, easy to follow versions of his\nproofs - and then relate them back to left braces. We also go on to discuss\ninteresting constructions of cycle sets which do not necessarily correspond\ndirectly to braces.\n"}
{"abstract": "  Individuals infected with SARS-CoV-2, the virus that causes COVID-19, may\nshed the virus in stool before developing symptoms, suggesting that\nmeasurements of SARS-CoV-2 concentrations in wastewater could be a \"leading\nindicator\" of COVID-19 prevalence. Multiple studies have corroborated the\nleading indicator concept by showing that the correlation between wastewater\nmeasurements and COVID-19 case counts is maximized when case counts are lagged.\nHowever, the meaning of \"leading indicator\" will depend on the specific\napplication of wastewater-based epidemiology, and the correlation analysis is\nnot relevant for all applications. In fact, the quantification of a leading\nindicator will depend on epidemiological, biological, and health systems\nfactors. Thus, there is no single \"lead time\" for wastewater-based COVID-19\nmonitoring. To illustrate this complexity, we enumerate three different\napplications of wastewater-based epidemiology for COVID-19: a qualitative\n\"early warning\" system; an independent, quantitative estimate of disease\nprevalence; and a quantitative alert of bursts of disease incidence. The\nleading indicator concept has different definitions and utility in each\napplication.\n"}
{"abstract": "  Measuring sentence similarity is a key research area nowadays as it allows\nmachines to better understand human languages. In this paper, we proposed a\nCross-Attention Siamese Network (CATsNet) to carry out the task of learning the\nsemantic meanings of Chinese sentences and comparing the similarity between two\nsentences. This novel model is capable of catching non-local features.\nAdditionally, we also tried to apply the long short-term memory (LSTM) network\nin the model to improve its performance. The experiments were conducted on the\nLCQMC dataset and the results showed that our model could achieve a higher\naccuracy than previous work.\n"}
{"abstract": "  The speech act of complaining is used by humans to communicate a negative\nmismatch between reality and expectations as a reaction to an unfavorable\nsituation. Linguistic theory of pragmatics categorizes complaints into various\nseverity levels based on the face-threat that the complainer is willing to\nundertake. This is particularly useful for understanding the intent of\ncomplainers and how humans develop suitable apology strategies. In this paper,\nwe study the severity level of complaints for the first time in computational\nlinguistics. To facilitate this, we enrich a publicly available data set of\ncomplaints with four severity categories and train different transformer-based\nnetworks combined with linguistic information achieving 55.7 macro F1. We also\njointly model binary complaint classification and complaint severity in a\nmulti-task setting achieving new state-of-the-art results on binary complaint\ndetection reaching up to 88.2 macro F1. Finally, we present a qualitative\nanalysis of the behavior of our models in predicting complaint severity levels.\n"}
{"abstract": "  In this article, we present our methodologies for SemEval-2021 Task-4:\nReading Comprehension of Abstract Meaning. Given a fill-in-the-blank-type\nquestion and a corresponding context, the task is to predict the most suitable\nword from a list of 5 options. There are three sub-tasks within this task:\nImperceptibility (subtask-I), Non-Specificity (subtask-II), and Intersection\n(subtask-III). We use encoders of transformers-based models pre-trained on the\nmasked language modelling (MLM) task to build our Fill-in-the-blank (FitB)\nmodels. Moreover, to model imperceptibility, we define certain linguistic\nfeatures, and to model non-specificity, we leverage information from hypernyms\nand hyponyms provided by a lexical database. Specifically, for non-specificity,\nwe try out augmentation techniques, and other statistical techniques. We also\npropose variants, namely Chunk Voting and Max Context, to take care of input\nlength restrictions for BERT, etc. Additionally, we perform a thorough ablation\nstudy, and use Integrated Gradients to explain our predictions on a few\nsamples. Our best submissions achieve accuracies of 75.31% and 77.84%, on the\ntest sets for subtask-I and subtask-II, respectively. For subtask-III, we\nachieve accuracies of 65.64% and 62.27%.\n"}
{"abstract": "  This paper considers the problem of nonparametric quantile regression under\nthe assumption that the target conditional quantile function is a composition\nof a sequence of low-dimensional functions. We study the nonparametric quantile\nregression estimator using deep neural networks to approximate the target\nconditional quantile function. For convenience, we shall refer to such an\nestimator as a deep quantile regression (DQR) estimator. We show that the DQR\nestimator achieves the nonparametric optimal convergence rate up to a\nlogarithmic factor determined by the intrinsic dimension of the underlying\ncompositional structure of the conditional quantile function, not the ambient\ndimension of the predictor. Therefore, DQR is able to mitigate the curse of\ndimensionality under the assumption that the conditional quantile function has\na compositional structure. To establish these results, we analyze the\napproximation error of a composite function by neural networks and show that\nthe error rate only depends on the dimensions of the component functions. We\napply our general results to several important statistical models often used in\nmitigating the curse of dimensionality, including the single index, the\nadditive, the projection pursuit, the univariate composite, and the generalized\nhierarchical interaction models. We explicitly describe the prefactors in the\nerror bounds in terms of the dimensionality of the data and show that the\nprefactors depends on the dimensionality linearly or quadratically in these\nmodels. We also conduct extensive numerical experiments to evaluate the\neffectiveness of DQR and demonstrate that it outperforms a kernel-based method\nfor nonparametric quantile regression.\n"}
{"abstract": "  The mobile impurity in a Bose-Einstein condensate (BEC) is a paradigmatic\nmany-body problem. For weak interaction between the impurity and the BEC, the\nimpurity deforms the BEC only slightly and it is well described within the\nFr\\\"ohlich model and the Bogoliubov approximation. For strong local attraction\nthis standard approach, however, fails to balance the local attraction with the\nweak repulsion between the BEC particles and predicts an instability where an\ninfinite number of bosons is attracted toward the impurity. Here we present a\nsolution of the Bose polaron problem beyond the Bogoliubov approximation which\nincludes the local repulsion between bosons and thereby stabilizes the Bose\npolaron even near and beyond the scattering resonance. We show that the Bose\npolaron energy remains bounded from below across the resonance and the size of\nthe polaron dressing cloud stays finite. Our results demonstrate how the\ndressing cloud replaces the attractive impurity potential with an effective\nmany-body potential that excludes binding. We find that at resonance, including\nthe effects of boson repulsion, the polaron energy depends universally on the\neffective range. Moreover, while the impurity contact is strongly peaked at\npositive scattering length, it remains always finite. Our solution highlights\nhow Bose polarons are self-stabilized by repulsion, providing a mechanism to\nunderstand quench dynamics and nonequilibrium time evolution at strong\ncoupling.\n"}
{"abstract": "  Precision studies at electron-positron colliders with center-of-mass energies\nin the charm-tau region and below have strongly contributed to our\nunderstanding of light-meson interactions at low energies. We focus on the\nprocesses involving two or three light mesons with invariant masses below\nnucleon-antinucleon threshold. A prominent role is given to the interactions of\nthe nine lightest pseudoscalar mesons (pions, kaons, $\\eta$, and $\\eta'$) and\nthe two narrow neutral isoscalar vector mesons $\\omega$ and $\\phi$.\nExperimental methods used to produce the mesons are reviewed as well as theory\ntools to extract properties of the meson-meson interactions. Examples of recent\nresults from the DA$\\Phi$NE, BEPCII, and VEPP-2000 colliders are presented. In\nthe outlook we briefly discuss prospects for further studies at future\nsuper-charm-tau factories.\n"}
{"abstract": "  Given a natural number $n\\geq3$ and two points $a$ and $b$ in the unit disk\n$\\mathbb D$ in the complex plane, it is known that there exists a unique\nelliptical disk having $a$ and $b$ as foci that can also be realized as the\nintersection of a collection of convex cyclic $n$-gons whose vertices fill the\nwhole unit circle $\\mathbb T$. What is less clear is how to find a convenient\nformula or expression for such an elliptical disk. Our main results reveal how\northogonal polynomials on the unit circle provide a useful tool for finding\nsuch a formula for some values of $n$. The main idea is to realize the\nelliptical disk as the numerical range of a matrix and the problem reduces to\nfinding the eigenvalues of that matrix.\n"}
{"abstract": "  Polar codes can theoretically achieve very competitive Frame Error Rates. In\npractice, their performance may depend on the chosen decoding procedure, as\nwell as other parameters of the communication system they are deployed upon. As\na consequence, designing efficient polar codes for a specific context can\nquickly become challenging. In this paper, we introduce a methodology that\nconsists in training deep neural networks to predict the frame error rate of\npolar codes based on their frozen bit construction sequence. We introduce an\nalgorithm based on Projected Gradient Descent that leverages the gradient of\nthe neural network function to generate promising frozen bit sequences. We\nshowcase on generated datasets the ability of the proposed methodology to\nproduce codes more efficient than those used to train the neural networks, even\nwhen the latter are selected among the most efficient ones.\n"}
{"abstract": "  We propose and study two DC (difference of convex functions) algorithms based\non exact penalty functions for solving nonsmooth DC optimization problems with\nnonsmooth DC equality and inequality constraints. Both methods employ adaptive\npenalty updating strategies to improve their performance. The first method is\nbased on exact penalty functions with individual penalty parameter for each\nconstraint (i.e. multidimensional penalty parameter) and utilizes a primal-dual\napproach to penalty updates. The second method is based on the so-called\nsteering exact penalty methodology and relies on solving some auxiliary convex\nsubproblems to determine a suitable value of the penalty parameter. We present\na detailed convergence analysis of both methods and give several simple\nnumerical examples highlighting peculiarites of two different penalty updating\nstrategies studied in this paper.\n"}
{"abstract": "  We call internal-wave billiard the dynamical system of a point particle that\nmoves freely inside a planar domain (the table) and is reflected by its\nboundary according to this rule: reflections are standard Fresnel reflections\nbut with the pretense that the boundary at any collision point is either\nhorizontal or vertical (relative to a predetermined direction representing\ngravity). These systems are point particle approximations for the motion of\ninternal gravity waves in closed containers, hence the name. For a class of\ntables similar to rectangular trapezoids, but with the slanted leg replaced by\na general curve with downward concavity, we prove that the dynamics has only\nthree asymptotic regimes: (1) there exist a global attractor and a global\nrepellor, which are periodic and might coincide; (2) there exists a beam of\nperiodic trajectories, whose boundary (if any) comprises an attractor and a\nrepellor for all the other trajectories; (3) all trajectories are dense (that\nis, the system is minimal). Furthermore, in the prominent case where the table\nis an actual trapezoid, we study the sets in parameter space relative to the\nthree regimes. We prove in particular that the set for (1) has positive measure\n(giving a rigorous proof of the existence of Arnol'd tongues for internal-wave\nbilliards), whereas the sets for (2) and (3) are non-empty but have measure\nzero.\n"}
{"abstract": "  Darwinian evolution is driven by random mutations, genetic recombination\n(gene shuffling) and selection that favors genotypes with high fitness. For\nsystems where each genotype can be represented as a bitstring of length $L$, an\noverview of possible evolutionary trajectories is provided by the $L$-cube\ngraph with nodes labeled by genotypes and edges directed toward the genotype\nwith higher fitness. Peaks (sinks in the graphs) are important since a\npopulation can get stranded at a suboptimal peak. The fitness landscape is\ndefined by the fitness values of all genotypes in the system. Some notion of\ncurvature is necessary for a more complete analysis of the landscapes,\nincluding the effect of recombination. The shape approach uses triangulations\n(shapes) induced by fitness landscapes. The main topic for this work is the\ninterplay between peak patterns and shapes. Because of constraints on the\nshapes for $L=3$ imposed by peaks, there are in total 25 possible combinations\nof peak patterns and shapes. Similar constraints exist for higher $L$.\nSpecifically, we show that the constraints induced by the staircase\ntriangulation can be formulated as a condition of {\\emph{universal positive\nepistasis}}, an order relation on the fitness effects of arbitrary sets of\nmutations that respects the inclusion relation between the corresponding\ngenetic backgrounds. We apply the concept to a large protein fitness landscape\nfor an immunoglobulin-binding protein expressed in Streptococcal bacteria.\n"}
{"abstract": "  Given an integral hyperplane arrangement, Kamiya-Takemura-Terao (2008 & 2011)\nintroduced the notion of characteristic quasi-polynomial, which enumerates the\ncardinality of the complement of the arrangement modulo a positive integer. The\nmost popular candidate for period of the characteristic quasi-polynomials is\nthe lcm period. In this paper, we initiate a study of period collapse in\ncharacteristic quasi-polynomials stemming from the concept of period collapse\nin the theory of Ehrhart quasi-polynomials. We say that period collapse occurs\nin a characteristic quasi-polynomial when the minimum period is strictly less\nthan the lcm period. Our first main result is that in the non-central case,\nwith regard to period collapse anything is possible: period collapse occurs in\nany dimension $\\ge 1$, occurs for any value of the lcm period $\\ge 2$, and the\nminimum period when it is not the lcm period can be any proper divisor of the\nlcm period. Our second main result states that in the central case, however, no\nperiod collapse is possible in any dimension, that is, the lcm period is always\nthe minimum period.\n"}
{"abstract": "  Is bias amplified when neural machine translation (NMT) models are optimized\nfor speed and evaluated on generic test sets using BLEU? We investigate\narchitectures and techniques commonly used to speed up decoding in\nTransformer-based models, such as greedy search, quantization, average\nattention networks (AANs) and shallow decoder models and show their effect on\ngendered noun translation. We construct a new gender bias test set, SimpleGEN,\nbased on gendered noun phrases in which there is a single, unambiguous, correct\nanswer. While we find minimal overall BLEU degradation as we apply speed\noptimizations, we observe that gendered noun translation performance degrades\nat a much faster rate.\n"}
{"abstract": "  This essay builds on the idea of grouping the polar curves of 2-variable\nfunction germs into polar clusters. In the topological category, one obtains a\nbijective correspondence between certain partitions of the polar quotients of\ntwo topologically equivalent function germs. We explain how this bijective\ncorrespondence may be refined in the Lipschitz category in terms of the\nassociated gradient canyons.\n"}
{"abstract": "  Involuntary subject motion is the main source of artifacts in weight-bearing\ncone-beam CT of the knee. To achieve image quality for clinical diagnosis, the\nmotion needs to be compensated. We propose to use inertial measurement units\n(IMUs) attached to the leg for motion estimation. We perform a simulation study\nusing real motion recorded with an optical tracking system. Three IMU-based\ncorrection approaches are evaluated, namely rigid motion correction, non-rigid\n2D projection deformation and non-rigid 3D dynamic reconstruction. We present\nan initialization process based on the system geometry. With an IMU noise\nsimulation, we investigate the applicability of the proposed methods in real\napplications. All proposed IMU-based approaches correct motion at least as good\nas a state-of-the-art marker-based approach. The structural similarity index\nand the root mean squared error between motion-free and motion corrected\nvolumes are improved by 24-35% and 78-85%, respectively, compared with the\nuncorrected case. The noise analysis shows that the noise levels of\ncommercially available IMUs need to be improved by a factor of $10^5$ which is\ncurrently only achieved by specialized hardware not robust enough for the\napplication. The presented study confirms the feasibility of this novel\napproach and defines improvements necessary for a real application.\n"}
{"abstract": "  Metasurface lenses, namely metalenses, are ultrathin planar nanostructures\nthat are capable of manipulating the properties of incoming light and imparting\nlens-like wavefront to the output. Although they have shown promising\npotentials for the future miniaturization of optics, the chromatic aberration\ninherited from their diffractive nature plagues them towards many practical\napplications. Current solutions for creating achromatic metalenses usually\nrequire searching through a large number of meta-atoms to find designs that\nfulfill not only phase but phase dispersion requirements, which leads to\nintensive design efforts. Besides, most designs are based on regular-shaped\nantennas driven by the designers' intuition and experience, hence only cover a\nlimited design space. Here, we present an inverse design approach that\nefficiently produces meta-atoms with unintuitive geometries required for\nbroadband achromatic metalenses. We restricted the generated shapes to hold\nfour-fold reflectional symmetry so that the resulting metalenses are\npolarization insensitive. In addition, meta-atoms generated by our method\ninheritably have round edges and corners, which make them\nnanofabrication-friendly. Our experimental characterization shows that our\nmetalenses exhibit superior performance over a broad bandwidth of 465 nm in the\nnear-infrared regime. Our method offers a fast and efficient way of designing\nhigh-performance achromatic metalenses and sheds new insights for unintuitive\ndesign of other metaphotonic devices.\n"}
{"abstract": "  Ketterer and Rajala showed an example of metric measure space, satisfying the\nmeasure contraction property $MCP(0,3)$, that has different topological\ndimensions at different regions of the space. In this article I propose a\nrefinement of that example, which satisfies the $CD(0,\\infty)$ condition,\nproving the non-constancy of topological dimension for CD spaces. This example\nalso shows that the weak curvature dimension bound, in the sense of\nLott-Sturm-Villani, is not sufficient to deduce any reasonable non-branching\ncondition. Moreover, it allows to answer to some open question proposed by\nSchultz, about strict curvature dimension bounds and their stability with\nrespect to the measured Gromov Hausdorff convergence.\n"}
{"abstract": "  Gravitational wave observation has the potential of probing ultralight\nbosonic fields such as axion. Axion forms a cloud around a rotating black hole\n(BH) by superradiant instability and should affect the gravitational waveform\nfrom binary BHs. On the other hand, considering the cloud associated with a BH\nin a binary system, tidal interaction depletes the cloud in some cases during\nthe inspiral phase. We made the exhaustive study of cloud depletion numerically\nin a wide parameter range for equal mass binaries, assuming only the\nquadrupolar tidal perturbation is at work. We found that clouds can avoid\ndisappearing due to the tidal effect only when $l=1$ mode is the fastest\ngrowing mode and when the binary orbit is counter-rotating in the\nnon-relativistic parameter region.\n"}
{"abstract": "  Federated learning (FL) has emerged as a promising master/slave learning\nparadigm to alleviate systemic privacy risks and communication costs incurred\nby cloud-centric machine learning methods. However, it is very challenging to\nresist the single point of failure of the master aggregator and attacks from\nmalicious participants while guaranteeing model convergence speed and accuracy.\nRecently, blockchain has been brought into FL systems transforming the paradigm\nto a decentralized manner thus further improve the system security and learning\nreliability. Unfortunately, the traditional consensus mechanism and\narchitecture of blockchain systems can hardly handle the large-scale FL task\ndue to the huge resource consumption, limited transaction throughput, and high\ncommunication complexity. To address these issues, this paper proposes a\ntwo-layer blockchaindriven FL framework, called as ChainsFL, which is composed\nof multiple subchain networks (subchain layer) and a direct acyclic graph\n(DAG)-based mainchain (mainchain layer). In ChainsFL, the subchain layer limits\nthe scale of each shard for a small range of information exchange, and the\nmainchain layer allows each shard to share and validate the learning model in\nparallel and asynchronously to improve the efficiency of cross-shard\nvalidation. Furthermore, the FL procedure is customized to deeply integrate\nwith blockchain technology, and the modified DAG consensus mechanism is\nproposed to mitigate the distortion caused by abnormal models. In order to\nprovide a proof-ofconcept implementation and evaluation, multiple subchains\nbase on Hyperledger Fabric are deployed as the subchain layer, and the\nself-developed DAG-based mainchain is deployed as the mainchain layer. The\nexperimental results show that ChainsFL provides acceptable and sometimes\nbetter training efficiency and stronger robustness compared with the typical\nexisting FL systems.\n"}
{"abstract": "  Fifth-generation (5G) systems will extensively employ radio access network\n(RAN) softwarization. This key innovation enables the instantiation of \"virtual\ncellular networks\" running on different slices of the shared physical\ninfrastructure. In this paper, we propose the concept of Private Cellular\nConnectivity as a Service (PCCaaS), where infrastructure providers deploy\ncovert network slices known only to a subset of users. We then present SteaLTE\nas the first realization of a PCCaaS-enabling system for cellular networks. At\nits core, SteaLTE utilizes wireless steganography to disguise data as noise to\nadversarial receivers. Differently from previous work, however, it takes a\nfull-stack approach to steganography, contributing an LTE-compliant\nsteganographic protocol stack for PCCaaS-based communications, and packet\nschedulers and operations to embed covert data streams on top of traditional\ncellular traffic (primary traffic). SteaLTE balances undetectability and\nperformance by mimicking channel impairments so that covert data waveforms are\nalmost indistinguishable from noise. We evaluate the performance of SteaLTE on\nan indoor LTE-compliant testbed under different traffic profiles, distance and\nmobility patterns. We further test it on the outdoor PAWR POWDER platform over\nlong-range cellular links. Results show that in most experiments SteaLTE\nimposes little loss of primary traffic throughput in presence of covert data\ntransmissions (< 6%), making it suitable for undetectable PCCaaS networking.\n"}
{"abstract": "  Relaxor ferrolectrics are important in technological applications due to a\nstrong electromechanical response, energy storage capacity, electrocaloric\neffect, and pyroelectric energy conversion properties. Current efforts to\ndiscover and design new materials in this class generally rely on\nsubstitutional doping of known ferroelectrics, as slight changes to local\ncompositional order can significantly affect the Curie temperature,\nmorphotropic phase boundary, and electromechanical responses. In this work, we\ndemonstrate that moving to the strong limit of compositional complexity in an\nABO3 perovskite allows stabilization of novel relaxor responses that do not\nrely on a single narrow phase transition region. Entropy-assisted synthesis\napproaches are used to create single crystal Ba(Ti0.2Sn0.2Zr0.2Hf0.2Nb0.2)O3\n[Ba(5B)O] films. The high levels of configurational disorder present in this\nsystem is found to influence dielectric relaxation, phase transitions,\nnano-polar domain formation, and Curie temperature. Temperature-dependent\ndielectric, Raman spectroscopy and second-harmonic generation measurements\nreveal multiple phase transitions, a high Curie temperature of 570 K, and the\nrelaxor ferroelectric nature of Ba(5B)O films. The first principles theory\ncalculations are used to predict possible combinations of cations to quantify\nthe relative feasibility of formation of highly disordered single-phase\nperovskite systems. The ability to stabilize single-phase perovskites with such\na large number of different cations on the B-sites offers new possibilities for\ndesigning high-performance materials for piezoelectric, pyroelectric and\ntunable dielectric applications.\n"}
{"abstract": "  The advent of acoustic metamaterials opened up a new frontier in the control\nof sound transmission. A key limitation, however, is that an acoustic\nmetamaterial based on a single local resonator in the unit cell produces a\nrestricted narrow-band attenuation peak; and when multiple local resonators are\nused the emerging attenuation peaks -- while numerous -- are each still narrow\nand separated by pass bands. Here, we present a new acoustic metamaterial\nconcept that yields a sound transmission loss through two antiresonances -- in\na single band gap -- that are fully coupled and hence provide a broadband\nattenuation range; this is in addition to delivering an isolation intensity\nthat exceeds 90 decibels for both peaks. The underlying coupled resonance\nmechanism is realized in the form of a single-panel, single-material pillared\nplate structure with internal contiguous holes$-$a practical configuration that\nlends itself to design adjustments and optimization for a frequency range of\ninterest, down to sub-kilohertz, and to mass fabrication.\n"}
{"abstract": "  We introduce a one-dimensional lattice model whose hopping amplitudes are\nmodulated for equally spaced sites. Such mosaic lattice exhibits many\ninteresting topological and localization phenomena that do not exist in the\nregular off-diagonal lattices. When the mosaic modulation is commensurate with\nthe underlying lattice, topologically nontrivial phases with zero- and\nnonzero-energy edge modes are observed as we tune the modulation, where the\nnontrivial regimes are characterized by quantized Berry phases. If the mosaic\nlattice becomes incommensurate, Anderson localization will be induced purely by\nthe quasiperiodic off-diagonal modulations. The localized eigenstate is found\nto be centered on two neighboring sites connected by the quasiperiodic hopping\nterms. Furthermore, both the commensurate and incommensurate off-diagonal\nmosaic lattices can host Chern insulators in their two-dimensional\ngeneralizations. Our work provides a platform for exploring topological phases\nand Anderson localization in low-dimensional systems.\n"}
{"abstract": "  Three novel precipitation strengthened bcc alloys which exhibit a smooth\nmicrostructural gradient with composition have been fabricated in bulk form by\ninduction casting. All three alloys are comprised of a mixture of disordered\nA2-(Fe, Cr) and L2$_1$-ordered (Ni, Fe)$_{2}$AlTi type phases both as-cast and\nafter long-term annealing at 900 $^{\\circ}$C. The ratio of disordered to\nordered phase, primary dendrite fraction, and overall microstructural\ncoarseness all decrease as Cr is replaced by Al and Ti. Differences in phase\ncomposition are quantified through domain averaged principal component analysis\nof energy dispersive spectroscopy data obtained during scanning transmission\nelectron microscopy. Bulk tensile testing reveals retained strengths of nearly\n250 MPa up to 900 $^{\\circ}$C for the alloys which contain a nanoscale\nmaze-like arrangement of ordered and disordered phases. One alloy, containing a\nduplex microstructure with ductile dendritic regions and highly creep resistant\ninterdendritic regions, shows a promising balance between high temperature\nductility and strength. For this alloy, tension creep testing was carried out\nat 700, 750, and 800 $^{\\circ}$C for a broad range of loading conditions and\nrevealed upper bound creep rates which surpass similar ferritic superalloys and\nrival those of several conventionally employed high temperature structural\nalloys, including Inconel 617 and 718, at much lower density and raw material\ncost.\n"}
{"abstract": "  Entity linking (EL), the task of disambiguating mentions in text by linking\nthem to entities in a knowledge graph, is crucial for text understanding,\nquestion answering or conversational systems. Entity linking on short text\n(e.g., single sentence or question) poses particular challenges due to limited\ncontext. While prior approaches use either heuristics or black-box neural\nmethods, here we propose LNN-EL, a neuro-symbolic approach that combines the\nadvantages of using interpretable rules based on first-order logic with the\nperformance of neural learning. Even though constrained to using rules, LNN-EL\nperforms competitively against SotA black-box neural approaches, with the added\nbenefits of extensibility and transferability. In particular, we show that we\ncan easily blend existing rule templates given by a human expert, with multiple\ntypes of features (priors, BERT encodings, box embeddings, etc), and even\nscores resulting from previous EL methods, thus improving on such methods. For\ninstance, on the LC-QuAD-1.0 dataset, we show more than $4$\\% increase in F1\nscore over previous SotA. Finally, we show that the inductive bias offered by\nusing logic results in learned rules that transfer well across datasets, even\nwithout fine tuning, while maintaining high accuracy.\n"}
{"abstract": "  Mandarin-English code-switching (CS) is frequently used among East and\nSoutheast Asian people. However, the intra-sentence language switching of the\ntwo very different languages makes recognizing CS speech challenging.\nMeanwhile, the recent successful non-autoregressive (NAR) ASR models remove the\nneed for left-to-right beam decoding in autoregressive (AR) models and achieved\noutstanding performance and fast inference speed, but it has not been applied\nto Mandarin-English CS speech recognition. This paper takes advantage of the\nMask-CTC NAR ASR framework to tackle the CS speech recognition issue. We\nfurther propose to change the Mandarin output target of the encoder to Pinyin\nfor faster encoder training and introduce the Pinyin-to-Mandarin decoder to\nlearn contextualized information. Moreover, we use word embedding label\nsmoothing to regularize the decoder with contextualized information and\nprojection matrix regularization to bridge that gap between the encoder and\ndecoder. We evaluate these methods on the SEAME corpus and achieved exciting\nresults.\n"}
{"abstract": "  How and where proteins interface with one another can ultimately impact the\nproteins' functions along with a range of other biological processes. As such,\nprecise computational methods for protein interface prediction (PIP) come\nhighly sought after as they could yield significant advances in drug discovery\nand design as well as protein function analysis. However, the traditional\nbenchmark dataset for this task, Docking Benchmark 5 (DB5), contains only a\nmodest 230 complexes for training, validating, and testing different machine\nlearning algorithms. In this work, we expand on a dataset recently introduced\nfor this task, the Database of Interacting Protein Structures (DIPS), to\npresent DIPS-Plus, an enhanced, feature-rich dataset of 42,112 complexes for\ngeometric deep learning of protein interfaces. The previous version of DIPS\ncontains only the Cartesian coordinates and types of the atoms comprising a\ngiven protein complex, whereas DIPS-Plus now includes a plethora of new\nresidue-level features including protrusion indices, half-sphere amino acid\ncompositions, and new profile hidden Markov model (HMM)-based sequence features\nfor each amino acid, giving researchers a large, well-curated feature bank for\ntraining protein interface prediction methods. We demonstrate through rigorous\nbenchmarks that training an existing state-of-the-art (SOTA) model for PIP on\nDIPS-Plus yields SOTA results, surpassing the performance of all other models\ntrained on residue-level and atom-level encodings of protein complexes to date.\n"}
{"abstract": "  We present a population of 19 radio-luminous supernovae (SNe) with emission\nreaching $L_{\\nu}{\\sim}10^{26}-10^{29}\\,\\rm{erg\\,s^{-1}Hz^{-1}}$ in the first\nepoch of the Very Large Array Sky Survey (VLASS) at $2-4$GHz. Our sample\nincludes one long Gamma-Ray Burst, SN 2017iuk/GRB171205A, and 18 core-collapse\nSNe detected at $\\approx (1-60)$years after explosion. No thermonuclear\nexplosion shows evidence for bright radio emission, and hydrogen-poor\nprogenitors dominate the sub-sample of core-collapse events with spectroscopic\nclassification at the time of explosion (79\\%). We interpret these findings\ninto the context of the expected radio emission from the forward shock\ninteraction with the circumstellar medium (CSM). We conclude that these\nobservations require a departure from the single wind-like density profile\n(i.e., $\\rho_{\\rm{CSM}}\\propto r^{-2}$) that is expected around massive stars\nand/or a departure from a spherical Newtonian shock. Viable alternatives\ninclude the shock interaction with a detached, dense shell of CSM formed by a\nlarge effective progenitor mass-loss rate $\\dot M \\sim (10^{-4}-10^{-1})$\nM$_{\\odot}$ yr$^{-1}$ (for an assumed wind velocity of\n$1000\\,\\rm{km\\,s^{-1}}$); emission from an off-axis relativistic jet entering\nour line of sight; or the emergence of emission from a newly-born pulsar-wind\nnebula. The relativistic SN 2012ap that is detected 5.7 and 8.5 years after\nexplosion with $L_{\\nu}{\\sim}10^{28}$ erg s$^{-1}$ Hz$^{-1}$ might constitute\nthe first detections of an off-axis jet+cocoon system in a massive star.\nHowever, none of the VLASS-SNe with archival data points are consistent with\nour model off-axis jet light curves. Future multi-wavelength observations will\ndistinguish among these scenarios.Our VLASS source catalogs, which were used to\nperform the VLASS cross matching, are publicly available at\nhttps://doi.org/10.5281/zenodo.4895112.\n"}
{"abstract": "  In our earlier publication we have shown how to compute by iteration a\nrational number ${u_{2,k}}$ in the two-term Machin-like formula for pi of kind\n$$\\frac{\\pi}{4}=2^{k-1}\\arctan\\left(\\frac{1}{u_{1,k}}\\right)+\\arctan\\left(\\frac{1}{u_{2,k}}\\right),\\qquad\nk\\in \\mathbb{Z},\\quad k\\ge 1,$$ where ${u_{1,k}}$ can be chosen as an integer\n${u_{1,k}} = \\left\\lfloor{{a_k}/\\sqrt{2-a_{k-1}}}\\right\\rfloor$ with nested\nradicals defined as ${a_k}=\\sqrt{2+a_{k-1}}$ and $a_0 = 0$. In this work we\nreport an alternative method for determination of the integer $u_{1,k}$. This\napproach is based on a simple iteration and does not require any irrational\n(surd) numbers from the set $\\left\\{a_k\\right\\}$ in computation of the integer\n$u_{1,k}$. Mathematica programs validating these results are presented.\n"}
{"abstract": "  We report observations of nanosecond nanometer scale heterogeneous dynamics\nin a free flowing colloidal jet revealed by ultrafast x-ray speckle visibility\nspectroscopy. The nanosecond double-bunch mode of the Linac Coherent Light\nSource free electron laser enabled the production of pairs of femtosecond\ncoherent hard x-ray pulses. By exploring the anisotropic summed speckle\nvisibility which relates to the correlation functions, we are able to evaluate\nnot only the average particle flow rate in a colloidal nanoparticle jet, but\nalso the heterogeneous flow field within. The reported methodology presented\nhere establishes the foundation for the study of nano- and atomic-scale\nheterogeneous fluctuations in complex matter using x-ray free electron laser\nsources.\n"}
{"abstract": "  Deformable templates are essential to large-scale medical image registration,\nsegmentation, and population analysis. Current conventional and deep\nnetwork-based methods for template construction use only regularized\nregistration objectives and often yield templates with blurry and/or\nanatomically implausible appearance, confounding downstream biomedical\ninterpretation. We reformulate deformable registration and conditional template\nestimation as an adversarial game wherein we encourage realism in the moved\ntemplates with a generative adversarial registration framework conditioned on\nflexible image covariates. The resulting templates exhibit significant gain in\nspecificity to attributes such as age and disease, better fit underlying\ngroup-wise spatiotemporal trends, and achieve improved sharpness and\ncentrality. These improvements enable more accurate population modeling with\ndiverse covariates for standardized downstream analyses and easier anatomical\ndelineation for structures of interest.\n"}
{"abstract": "  Biological membranes undergo noticeable thermal fluctuations at physiological\ntemperatures. When two membranes approach each other, they hinder the out of\nplane fluctuations of the other. This hindrance leads to an entropic repulsive\nforce between membranes which, in an interplay with attractive and repulsive\nforces due to other sources, impacts a range of biological functions: cell\nadhesion, membrane fusion, self-assembly, binding-unbinding transition among\nothers. In this work, we take cognizance of the fact that biological membranes\nare not purely mechanical entities and, due to the phenomenon of\nflexoelectricity, exhibit a coupling between deformation and electric\npolarization. The ensuing coupled mechanics-electrostatics-statistical\nmechanics problem is analytically intractable. We use a variational\nperturbation method to analyze, in closed-form, the contribution of\nflexoelectricity to the entropic force between two fluctuating membranes and\ndiscuss its possible physical implications. We find that flexoelectricity leads\nto a correction that switches from an enhanced attraction at close membrane\nseparations and an enhanced repulsion when the membranes are further apart.\n"}
{"abstract": "  Reconfigurable intelligent surfaces (RISs) are one of the foremost\ntechnological enablers of future wireless systems. They improve communication\nand localization by providing a strong non-line-of-sight path to the receiver.\nIn this paper, we propose a pilot transmission method to enable the receiver to\nseparate signals arriving from different RISs and from the uncontrolled\nmultipath. This facilitates channel estimation and localization, as the channel\nor its geometric parameters can be estimated for each path separately. Our\nmethod is based on designing temporal phase profiles that are orthogonal across\nRISs without affecting the RIS beamforming capabilities. We take into\nconsideration the limited resolution of the RIS phase shifters and show that in\nthe presence of this practical limitation, orthogonal phase profiles can be\ndesigned based on Butson-type Hadamard matrices. For a localization scenario,\nwe show that with our proposed method the estimator can attain the theoretical\nlower bound even with one-bit RIS phase resolution.\n"}
{"abstract": "  This paper develops a planner that plans the action sequences and motion for\na dual-arm robot to lift up and flip heavy plates using crane pulley blocks.\nThe problem is motivated by the low payload of modern collaborative robots.\nInstead of directly manipulating heavy plates that collaborative robots cannot\nafford, the paper develops a planner for collaborative robots to operate crane\npulley blocks. The planner assumes a target plate is pre-attached to the crane\nhook. It optimizes dual-arm action sequences and plans the robot's dual-arm\nmotion that pulls the rope of the crane pulley blocks to lift up the plate. The\ncrane pulley blocks reduce the payload that each robotic arm needs to bear.\nWhen the plate is lifted up to a satisfying pose, the planner plans a pushing\nmotion for one of the robot arms to tumble over the plate while considering\nforce and moment constraints. The article presents the technical details of the\nplanner and several experiments and analysis carried out using a dual-arm robot\nmade by two Universal Robots UR3 arms. The influence of various parameters and\noptimization goals are investigated and compared in depth. The results show\nthat the proposed planner is flexible and efficient.\n"}
{"abstract": "  We design and validate new block propagation protocols for the peer-to-peer\n(P2P) network of the Bitcoin blockchain. Despite its strong protection for\nsecurity and privacy, the current Bitcoin blockchain can only support a low\nnumber of transactions per second (TPS). In this work, we redesign the current\nBitcoin's networking protocol to increase TPS without changing vital components\nin its consensus-building protocol. In particular, we improve the compact-block\nrelaying protocol to enable the propagation of blocks containing a massive\nnumber of transactions without inducing extra propagation latencies. Our\nimprovements consist of (i) replacing the existing store-and-forward\ncompact-block relaying scheme with a cut-through compact-block relaying scheme;\n(ii) exploiting rateless erasure codes for P2P networks to increase\nblock-propagation efficiency. Since our protocols only need to rework the\ncurrent Bitcoin's networking protocol and does not modify the data structures\nand crypto-functional components, they can be seamlessly incorporated into the\nexisting Bitcoin blockchain. To validate our designs, we perform analysis on\nour protocols and implement a Bitcoin network simulator on NS3 to run different\nblock propagation protocols. The analysis and experimental results confirm that\nour new block propagation protocols could increase the TPS of the Bitcoin\nblockchain by 100x without compromising security and consensus-building.\n"}
{"abstract": "  This study proposes the construction of a transmit signal for large-scale\nantenna systems with cost-effective 1-bit digital-to-analog converters in the\ndownlink. Under quadrature-amplitude-modulation constellations, it is still an\nopen problem to overcome a severe error floor problem caused by its nature\nproperty. To this end, we first present a feasibility condition which\nguarantees that each user's noiseless signal is placed in the desired decision\nregion. For robustness to additive noise, we formulate an optimization problem,\nwe then transform the feasibility conditions to cascaded matrix form. We\npropose a low-complexity algorithm to generate a 1-bit transmit signal based on\nthe proposed optimization problem formulated as a well-defined\nmixed-integer-linear-programming. Numerical results validate the superiority of\nthe proposed method in terms of detection performance and computational\ncomplexity.\n"}
{"abstract": "  For a harmonically trapped system consisting of two bosons in one spatial\ndimension with infinite contact repulsion (hard core bosons), we derive an\nexpression for the one-body density matrix $\\rho_B$ in terms of centre of mass\nand relative coordinates of the particles. The deviation from $\\rho_F$, the\ndensity matrix for the two fermions case, can be clearly identified. Moreover,\nthe obtained $\\rho_B$ allows us to derive a closed form expression of the\ncorresponding momentum distribution $n_{B}(p)$. We show how the result deviates\nfrom the noninteracting fermionic case, the deviation being associated to the\nshort range character of the interaction. Mathematically, our analytical\nmomentum distribution is expressed in terms of one and two variables confluent\nhypergeometric functions. Our formula satisfies the correct normalization and\npossesses the expected behavior at zero momentum. It also exhibits the high\nmomentum $1/p^4 $ tail with the appropriate Tan's coefficient. Numerical\nresults support our findings.\n"}
{"abstract": "  Making predictions in a robust way is not easy for nonlinear systems. In this\nwork, a neural network computing framework, i.e., a spatiotemporal\nconvolutional network (STCN), was developed to efficiently and accurately\nrender a multistep-ahead prediction of a time series by employing a\nspatial-temporal information (STI) transformation. The STCN combines the\nadvantages of both the temporal convolutional network (TCN) and the STI\nequation, which maps the high-dimensional/spatial data to the future temporal\nvalues of a target variable, thus naturally providing the prediction of the\ntarget variable. From the observed variables, the STCN also infers the causal\nfactors of the target variable in the sense of Granger causality, which are in\nturn selected as effective spatial information to improve the prediction\nrobustness. The STCN was successfully applied to both benchmark systems and\nreal-world datasets, all of which show superior and robust performance in\nmultistep-ahead prediction, even when the data were perturbed by noise. From\nboth theoretical and computational viewpoints, the STCN has great potential in\npractical applications in artificial intelligence (AI) or machine learning\nfields as a model-free method based only on the observed data, and also opens a\nnew way to explore the observed high-dimensional data in a dynamical manner for\nmachine learning.\n"}
{"abstract": "  This paper proposes a hierarchical Bayesian model based on spatial concepts\nthat enables a robot to transfer the knowledge of places from experienced\nenvironments to a new environment. The transfer of knowledge based on spatial\nconcepts is modeled as the calculation process of the posterior distribution\nbased on the observations obtained in each environment with the parameters of\nspatial concepts generalized to environments as prior knowledge. We conducted\nexperiments to evaluate the generalization performance of spatial knowledge for\ngeneral places such as kitchens and the adaptive performance of spatial\nknowledge for unique places such as `Emma's room' in a new environment. In the\nexperiments, the accuracies of the proposed method and conventional methods\nwere compared in the prediction task of location names from an image and a\nposition, and the prediction task of positions from a location name. The\nexperimental results demonstrated that the proposed method has a higher\nprediction accuracy of location names and positions than the conventional\nmethod owing to the transfer of knowledge.\n"}
{"abstract": "  We study an analogue of Serre's modularity conjecture for projective\nrepresentations $\\overline{\\rho}: \\operatorname{Gal}(\\overline{K} / K)\n\\rightarrow \\operatorname{PGL}_2(k)$, where $K$ is a totally real number field.\nWe prove new cases of this conjecture when $k = \\mathbb{F}_5$ by using the\nautomorphy lifting theorems over CM fields established in previous work of the\nauthors.\n"}
{"abstract": "  Distributed linearly separable computation, where a user asks some\ndistributed servers to compute a linearly separable function, was recently\nformulated by the same authors and aims to alleviate the bottlenecks of\nstragglers and communication cost in distributed computation. For this purpose,\nthe data center assigns a subset of input datasets to each server, and each\nserver computes some coded packets on the assigned datasets, which are then\nsent to the user. The user should recover the task function from the answers of\na subset of servers, such the effect of stragglers could be tolerated. In this\npaper, we formulate a novel secure framework for this distributed linearly\nseparable computation, where we aim to let the user only retrieve the desired\ntask function without obtaining any other information about the input datasets,\neven if it receives the answers of all servers. In order to preserve the\nsecurity of the input datasets, some common randomness variable independent of\nthe datasets should be introduced into the transmission. We show that any\nnon-secure linear-coding based computing scheme for the original distributed\nlinearly separable computation problem, can be made secure without increasing\nthe communication cost. Then we focus on the case where the computation cost of\neach server is minimum and aim to minimize the size of the randomness variable\nintroduced in the system while achieving the optimal communication cost. We\nfirst propose an information theoretic converse bound on the randomness size.\nWe then propose secure computing schemes based on two well-known data\nassignments, namely fractional repetition assignment and cyclic assignment. We\nthen propose a computing scheme with novel assignment, which strictly\noutperforms the above two schemes. Some additional optimality results are also\nobtained.\n"}
{"abstract": "  We revisit the gradient descent dynamics of the spherical\nSherrington-Kirkpatrick ($p=2$) model with finite number of degrees of freedom.\nFor fully random initial conditions we confirm that the relaxation takes place\nin three time regimes: a first algebraic one controlled by the decay of the\neigenvalue distribution of the random exchange interaction matrix at its edge\nin the infinite size limit; a faster algebraic one determined by the\ndistribution of the gap between the two extreme eigenvalues; and a final\nexponential one determined by the minimal gap sampled in the disorder average.\nWe also analyse the finite size effects on the relaxation from initial states\nwhich are almost projected on the saddles of the potential energy landscape,\nand we show that for deviations scaling as $N^{-\\nu}$ from perfect alignment\nthe system escapes the initial configuration in a time-scale scaling as $\\ln N$\nafter which the dynamics no longer \"self-averages\" with respect to the initial\nconditions. We prove these statements with a combination of analytic and\nnumerical methods.\n"}
{"abstract": "  The paper focuses on the kinematics control of a compliant serial manipulator\ncomposed of a new type of dualtriangle elastic segments. Some useful\noptimization techniques were applied to solve the geometric redundancy problem,\nensure the stability of the manipulator configurations with respect to the\nexternal forces/torques applied to the endeffector. The efficiency of the\ndeveloped control algorisms is confirmed by simulation.\n"}
{"abstract": "  Single domain generalization is a challenging case of model generalization,\nwhere the models are trained on a single domain and tested on other unseen\ndomains. A promising solution is to learn cross-domain invariant\nrepresentations by expanding the coverage of the training domain. These methods\nhave limited generalization performance gains in practical applications due to\nthe lack of appropriate safety and effectiveness constraints. In this paper, we\npropose a novel learning framework called progressive domain expansion network\n(PDEN) for single domain generalization. The domain expansion subnetwork and\nrepresentation learning subnetwork in PDEN mutually benefit from each other by\njoint learning. For the domain expansion subnetwork, multiple domains are\nprogressively generated in order to simulate various photometric and geometric\ntransforms in unseen domains. A series of strategies are introduced to\nguarantee the safety and effectiveness of the expanded domains. For the domain\ninvariant representation learning subnetwork, contrastive learning is\nintroduced to learn the domain invariant representation in which each class is\nwell clustered so that a better decision boundary can be learned to improve\nit's generalization. Extensive experiments on classification and segmentation\nhave shown that PDEN can achieve up to 15.28% improvement compared with the\nstate-of-the-art single-domain generalization methods.\n"}
{"abstract": "  Many near-term quantum computing algorithms are conceived as variational\nquantum algorithms, in which parameterized quantum circuits are optimized in a\nhybrid quantum-classical setup. Examples are variational quantum eigensolvers,\nquantum approximate optimization algorithms as well as various algorithms in\nthe context of quantum-assisted machine learning. A common bottleneck of any\nsuch algorithm is constituted by the optimization of the variational\nparameters. A popular set of optimization methods work on the estimate of the\ngradient, obtained by means of circuit evaluations. We will refer to the way in\nwhich one can combine these circuit evaluations as gradient rules. This work\nprovides a comprehensive picture of the family of gradient rules that vary\nparameters of quantum gates individually. The most prominent known members of\nthis family are the parameter shift rule and the finite differences method. To\nunite this family, we propose a generalized parameter shift rule that expresses\nall members of the aforementioned family as special cases, and discuss how all\nof these can be seen as providing access to a linear combination of exact\nfirst- and second-order derivatives. We further prove that a parameter shift\nrule with one non-shifted evaluation and only one shifted circuit evaluation\ncan not exist does not exist, and introduce a novel perspective for approaching\nnew gradient rules.\n"}
{"abstract": "  The CW RF test of 1.3 GHz 9-cell cavity in liquid helium bath at 2 K is a\nvery important key point in the cavity procurement. Some problems can be found\nthrough the test, according which to optimized and improve the process of\ncavity. Recently, Medium temperature (mid-T) furnace bake of 1.3 GHz 9-cell\ncavities have been carried out at IHEP. Through the proceed of mid-T bake, the\nquality factor of cavity has been greatly improved. While the excitation of the\nparasitic modes in the high Q cavities CW cold test has been encountered, which\nimplies an error source for the cavity gradient and quality factor\ndetermination. In order to ensure the testing accuracy of superconducting\ncavity, we have improved the testing system. Finally, the parasitic mode is\ncompletely suppressed and the CW RF cold test of high Q cavity is guaranteed.\n"}
{"abstract": "  We present THUNDR, a transformer-based deep neural network methodology to\nreconstruct the 3d pose and shape of people, given monocular RGB images. Key to\nour methodology is an intermediate 3d marker representation, where we aim to\ncombine the predictive power of model-free-output architectures and the\nregularizing, anthropometrically-preserving properties of a statistical human\nsurface model like GHUM -- a recently introduced, expressive full body\nstatistical 3d human model, trained end-to-end. Our novel transformer-based\nprediction pipeline can focus on image regions relevant to the task, supports\nself-supervised regimes, and ensures that solutions are consistent with human\nanthropometry. We show state-of-the-art results on Human3.6M and 3DPW, for both\nthe fully-supervised and the self-supervised models, for the task of inferring\n3d human shape, joint positions, and global translation. Moreover, we observe\nvery solid 3d reconstruction performance for difficult human poses collected in\nthe wild.\n"}
{"abstract": "  This paper is a review of a particular approach to the method of maximum\nentropy as a general framework for inference. The discussion emphasizes the\npragmatic elements in the derivation. An epistemic notion of information is\ndefined in terms of its relation to the Bayesian beliefs of ideally rational\nagents. The method of updating from a prior to a posterior probability\ndistribution is designed through an eliminative induction process. The\nlogarithmic relative entropy is singled out as the unique tool for updating\nthat (a) is of universal applicability; (b) that recognizes the value of prior\ninformation; and (c) that recognizes the privileged role played by the notion\nof independence in science. The resulting framework -- the ME method -- can\nhandle arbitrary priors and arbitrary constraints. It includes MaxEnt and\nBayes' rule as special cases and, therefore, it unifies entropic and Bayesian\nmethods into a single general inference scheme. The ME method goes beyond the\nmere selection of a single posterior, but also addresses the question of how\nmuch less probable other distributions might be, which provides a direct bridge\nto the theories of fluctuations and large deviations.\n"}
{"abstract": "  Sanctioned by its constitution, India is home to the world's most\ncomprehensive affirmative action program, where historically discriminated\ngroups are protected with vertical reservations implemented as \"set asides,\"\nand other disadvantaged groups are protected with horizontal reservations\nimplemented as \"minimum guarantees.\" A mechanism mandated by the Supreme Court\nin 1995 suffers from important anomalies, triggering countless litigations in\nIndia. Foretelling a recent reform correcting the flawed mechanism, we propose\nthe 2SMG mechanism that resolves all anomalies, and characterize it with\ndesiderata reflecting laws of India. Subsequently rediscovered with a high\ncourt judgment and enforced in Gujarat, 2SMG is also endorsed by Saurav Yadav\nv. State of UP (2020), in a Supreme Court ruling that rescinded the flawed\nmechanism. While not explicitly enforced, 2SMG is indirectly enforced for an\nimportant subclass of applications in India, because no other mechanism\nsatisfies the new mandates of the Supreme Court.\n"}
{"abstract": "  In 1994, P.-G. Becker and W. Bergweiler listed all the differentially\nalgebraic solutions of three famous functional equations: the Schr{\\\"o}der's,\nB{\\\"o}ttcher's and Abel's equations. The proof of this theorem combines various\ndomains of mathematics. This goes from the theory of iteration, which gave\nbirth to these equations, to the algebro-differential notion of coherent\nfamilies developed by M. Boshernitzan and L. A. Rubel. This survey is an\nexcursion into the history of these equations, in order to enlighten the\ndifferent pieces of mathematics they bring together and how these parts fit\ninto the result of P.-G. Becker and W. Bergweiler.\n"}
{"abstract": "  Over-parameterization of neural networks benefits the optimization and\ngeneralization yet brings cost in practice. Pruning is adopted as a\npost-processing solution to this problem, which aims to remove unnecessary\nparameters in a neural network with little performance compromised. It has been\nbroadly believed the resulted sparse neural network cannot be trained from\nscratch to comparable accuracy. However, several recent works (e.g., [Frankle\nand Carbin, 2019a]) challenge this belief by discovering random sparse networks\nwhich can be trained to match the performance with their dense counterpart.\nThis new pruning paradigm later inspires more new methods of pruning at\ninitialization. In spite of the encouraging progress, how to coordinate these\nnew pruning fashions with the traditional pruning has not been explored yet.\nThis survey seeks to bridge the gap by proposing a general pruning framework so\nthat the emerging pruning paradigms can be accommodated well with the\ntraditional one. With it, we systematically reflect the major differences and\nnew insights brought by these new pruning fashions, with representative works\ndiscussed at length. Finally, we summarize the open questions as worthy future\ndirections.\n"}
{"abstract": "  Deuteration has been used as a tracer of the evolutionary phases of low- and\nhigh-mass star formation. The APEX Telescope Large Area Survey (ATLASGAL)\nprovides an important repository for a detailed statistical study of massive\nstar-forming clumps in the inner Galactic disc at different evolutionary\nphases. We study the amount of deuteration using NH2D in a representative\nsample of high-mass clumps discovered by the ATLASGAL survey covering various\nevolutionary phases of massive star formation. Unbiased spectral line surveys\nat 3 mm were thus conducted towards ATLASGAL clumps between 85 and 93 GHz with\nthe Mopra telescope and from 84 to 115 GHz using the IRAM 30m telescope. A\nsubsample was followed up in the NH2D transition at 74 GHz with the IRAM 30m\ntelescope. We determined the deuterium fractionation from the column density\nratio of NH2D and NH3 and measured the NH2D excitation temperature for the\nfirst time from the simultaneous modelling of the 74 and 110 GHz line using\nMCWeeds. We find a large range of the NH2D to NH3 column density ratio up to\n1.6+/-0.7 indicating a high degree of NH3 deuteration in a subsample of the\nclumps. Our analysis yields a clear difference between NH3 and NH2D rotational\ntemperatures for a fraction of the sources. We therefore advocate observation\nof the NH2D transitions at 74 and 110 GHz simultaneously to determine the NH2D\ntemperature directly. We determine a median ortho-to-para column density ratio\nof 3.7+/-1.2. The high detection rate of NH2D confirms a high deuteration\npreviously found in massive star-forming clumps. Using the excitation\ntemperature of NH2D instead of NH3 is needed to avoid an overestimation of\ndeuteration. We measure a higher detection rate of NH2D in sources at early\nevolutionary stages. The deuterium fractionation shows no correlation with\nevolutionary tracers such as the NH3 (1,1) line width, or rotational\ntemperature.\n"}
{"abstract": "  High dimensional data can contain multiple scales of variance. Analysis tools\nthat preferentially operate at one scale can be ineffective at capturing all\nthe information present in this cross-scale complexity. We propose a multiscale\njoint characterization approach designed to exploit synergies between global\nand local approaches to dimensionality reduction. We illustrate this approach\nusing Principal Components Analysis (PCA) to characterize global variance\nstructure and t-stochastic neighbor embedding (t-sne) to characterize local\nvariance structure. Using both synthetic images and real-world imaging\nspectroscopy data, we show that joint characterization is capable of detecting\nand isolating signals which are not evident from either PCA or t-sne alone.\nBroadly, t-sne is effective at rendering a randomly oriented low-dimensional\nmap of local clusters, and PCA renders this map interpretable by providing\nglobal, physically meaningful structure. This approach is illustrated using\nimaging spectroscopy data, and may prove particularly useful for other\ngeospatial data given robust local variance structure due to spatial\nautocorrelation and physical interpretability of global variance structure due\nto spectral properties of Earth surface materials. However, the fundamental\npremise could easily be extended to other high dimensional datasets, including\nimage time series and non-image data.\n"}
{"abstract": "  The Hall effects on Casson fluid flow along a vertical plate has been\ninvestigated numerically. The governing equations have been derived from\nNavier-Stokes' equation and boundary layer approximation has been employed. By\nusing usual transformations, the obtained non-linear coupled partial\ndifferential equations have been transformed into dimensionless governing\nequations. These equations have been solved by applying the explicit finite\ndifference method. The MATLAB R2015a tool has been used for numerical\nsimulation. The stability and convergence criteria have been analyzed. The\neffect of some important parameters on the primary velocity, secondary\nvelocity, temperature and concentration distributions as well as local shear\nstress, Nusselt number and Sherwood number have been shown graphically.\n"}
{"abstract": "  This paper investigates the optimal signal detection problem with a\nparticular interest in large-scale multiple-input multiple-output (MIMO)\nsystems. The problem is NP-hard and can be solved optimally by searching the\nshortest path on the decision tree. Unfortunately, the existing optimal search\nalgorithms often involve prohibitively high complexities, which indicates that\nthey are infeasible in large-scale MIMO systems. To address this issue, we\npropose a general heuristic search algorithm, namely, hyperaccelerated tree\nsearch (HATS) algorithm. The proposed algorithm employs a deep neural network\n(DNN) to estimate the optimal heuristic, and then use the estimated heuristic\nto speed up the underlying memory-bounded search algorithm. This idea is\ninspired by the fact that the underlying heuristic search algorithm reaches the\noptimal efficiency with the optimal heuristic function. Simulation results show\nthat the proposed algorithm reaches almost the optimal bit error rate (BER)\nperformance in large-scale systems, while the memory size can be bounded. In\nthe meanwhile, it visits nearly the fewest tree nodes. This indicates that the\nproposed algorithm reaches almost the optimal efficiency in practical\nscenarios, and thereby it is applicable for large-scale systems. Besides, the\ncode for this paper is available at https://github.com/skypitcher/hats.\n"}
{"abstract": "  We study the connection between the X-ray and UV properties of the broad\nabsorption line (BAL) wind in the highly X-ray variable quasar PG 2112+059 by\ncomparing Chandra-ACIS data with contemporaneous UV HST/STIS spectra in three\ndifferent epochs. We observe a correlation whereby an increase in the\nequivalent-widths (EWs) of the BALs is accompanied by a redder UV spectrum. The\ngrowth in the BALs EWs is also accompanied by a significant dimming in soft\nX-ray emission (<2 keV), consistent with increased absorption. Variations in\nthe hard X-ray emission (>2 keV) are only accompanied by minor spectral\nvariations of the UV-BALs and do not show significant changes in the EW of\nBALs. These trends suggest a wind-shield scenario where the outflow inclination\nwith respect to the line of sight is decreasing and/or the wind mass is\nincreasing. These changes elevate the covering fraction and/or column densities\nof the BALs and are likely accompanied by a nearly contemporaneous increase in\nthe column density of the shield.\n"}
{"abstract": "  Sub-nanosecond precision clock synchronization over the packet network has\nbeen achieved by the White Rabbit protocol for a decade. However, few computer\nsystems utilize such a technique. We try to attract more interest in the clock\nsynchronization problem. We first introduce the basics of clock and\nsynchronization in the time and frequency discipline. Then we revisit several\nrelated works, such as Google's Spanner, Huygens, FARMv2, DTP, and Sundial,\nexplain why these works could be improved. Finally, we briefly discuss an\nindependent time network approach towards low-cost and high-precision\nsynchronization.\n"}
{"abstract": "  A chain of evolution algebras (CEA) is an uncountable family (depending on\ntime) of evolution algebras on the field of real numbers. The matrix of\nstructural constants of a CEA satisfies Kolmogorov-Chapman equation. In this\npaper, we consider three CEAs of three-dimensional real evolution algebras.\nThese CEAs depend on several (non-zero) functions defined on the set of time.\nFor each chain we give full classification (up to isomorphism) of the algebras\ndepending on the time-parameter. We find concrete functions ensuring that the\ncorresponding CEA contains all possible three-dimensional evolution algebras.\n"}
{"abstract": "  The mixing of N2 with H2 leads to very different plasmas from pure N2 and H2\nplasma discharges. Numerous issues are therefore raised involving the processes\nleading to ammonia (NH3) formation. The aim of this work is to better\ncharacterize capacitively-coupled radiofrequency plasma discharges in N2 with\nfew percents of H2 (up to 5 pct), at low pressure (0.3 to 1 mbar) and low\ncoupled power (3 to 13 W). Both experimental measurements and numerical\nsimulations are performed. For clarity, we separated the results in two\ncomplementary parts. The actual one (first part), presents the details on the\nexperimental measurements, while the second focuses on the simulation, a hybrid\nmodel combining a 2D fluid module and a 0D kinetic module. Electron density is\nmeasured by a resonant cavity method. It varies from 0.4 to 5e9 cm-3,\ncorresponding to ionization degrees from 2e-8 to 4e-7. Ammonia density is\nquantified by combining IR absorption and mass spectrometry. It increases\nlinearly with the amount of H2 (up to 3e13 cm-3 at 5 pct H2). On the contrary,\nit is constant with pressure, which suggests the dominance of surface processes\non the formation of ammonia. Positive ions are measured by mass spectrometry.\nNitrogen-bearing ions are hydrogenated by the injection of H2, N2H+ being the\nmajor ion as soon as the amount of H2 is larger than 1 pct. The increase of\npressure leads to an increase of secondary ions formed by ion (or radical) -\nneutral collisions (ex: N2H+, NH4+, H3+), while an increase of the coupled\npower favours ions formed by direct ionization (ex: N2+, NH3+, H2+).\n"}
{"abstract": "  We prove that the rank of the cohomology of a closed symplectic manifold with\ncoefficients in a field of characteristic $p$ is smaller than the number of\nperiodic orbits of any non-degenerate Hamiltonian flow. Following Floer, the\nproof relies on constructing a homology group associated to each such flow, and\ncomparing it with the homology of the ambient symplectic manifold. The proof\ndoes not proceed by constructing a version of Floer's complex with\ncharacteristic $p$ coefficients, but uses instead the canonical (stable)\ncomplex orientations of moduli spaces of Floer trajectories to construct a\nversion of Floer homology with coefficients in Morava's $K$-theories, and can\nthus be seen as an implementation of Cohen, Jones, and Segal's vision for a\nFloer homotopy theory. The key feature of Morava K-theory that allows the\nconstruction to be carried out is the fact that the corresponding homology and\ncohomology groups of classifying spaces of finite groups satisfy Poincar\\'e\nduality.\n"}
{"abstract": "  In this work, in the context of modified gravity, a curved spacetime\nanalogous to the \"canonical acoustic black hole\" is constructed. The source is\na self-interacting scalar field which is non-minimally coupled to gravity\nthrough the Gauss-Bonnet invariant. The scalar-Gauss-Bonnet coupling function\nis characterized by three positive parameters: $\\sigma$ with units of\n$(length)$, $\\mu$ with units of $(length)^{4}$, and a dimensionless parameter\n$s$, thus defining a three-parameter model for which the line element of\ncanonical acoustic black hole is a solution. The spacetime is equipped with\nspherical and static symmetry and has a single horizon determined in\nSchwarzschild coordinates by the region $r=\\mu^{1/4}$. The solution admits a\nphoton sphere at $r=(3\\mu)^{1/4}$, and it is shown that in the region\n$(3\\mu)^{1/4}\\leq r<\\infty$ the scalar field satisfies the null, weak, and\nstrong energy conditions. Nonetheless, the model with $s=1$ has major physical\nrelevance since for this case the scalar field is well defined in the entire\nregion $r\\geq\\mu^{1/4}$, while for $s\\neq1$ the scalar field blows up on the\nhorizon.\n"}
{"abstract": "  The idea of nondeterministic typical hesitant fuzzy automata is a\ngeneralization of the fuzzy automata presented by Costa and Bedregal. This\npaper, presents the sufficient and necessary conditions for a typical hesitant\nfuzzy language to be computed by nondeterministic typical hesitant fuzzy\nautomata. Besides, the paper introduces a new class of Typical Hesitant Fuzzy\nAutomata with crisp transitions, and we will show that this new class is\nequivalent to the original class introduced by Costa and Bedregal\n"}
{"abstract": "  The Hamiltonian Monte Carlo (HMC) method allows sampling from continuous\ndensities. Favorable scaling with dimension has led to wide adoption of HMC by\nthe statistics community. Modern auto-differentiating software should allow\nmore widespread usage in Bayesian inverse problems. This paper analyzes the two\nmajor difficulties encountered using HMC for inverse problems: poor\nconditioning and multi-modality. Novel results on preconditioning and replica\nexchange Monte Carlo parameter selection are presented in the context of\nspectroscopy. Recommendations are analyzed rigorously in the Gaussian case, and\nshown to generalize in a fusion plasma reconstruction.\n"}
{"abstract": "  A planner aims to target individuals who exceed a threshold in a\ncharacteristic, such as wealth or ability. The individuals can rank their\nfriends according to the characteristic. We study a strategy-proof mechanism\nfor the planner to use the rankings for targeting. We discuss how the mechanism\nworks in practice, when the rankings may contain errors.\n"}
{"abstract": "  The most common tools for word-alignment rely on a large amount of parallel\nsentences, which are then usually processed according to one of the IBM model\nalgorithms. The training data is, however, the same as for machine translation\n(MT) systems, especially for neural MT (NMT), which itself is able to produce\nword-alignments using the trained attention heads. This is convenient because\nword-alignment is theoretically a viable byproduct of any attention-based NMT,\nwhich is also able to provide decoder scores for a translated sentence pair.\n  We summarize different approaches on how word-alignment can be extracted from\nalignment scores and then explore ways in which scores can be extracted from\nNMT, focusing on inferring the word-alignment scores based on output sentence\nand token probabilities. We compare this to the extraction of alignment scores\nfrom attention. We conclude with aggregating all of the sources of alignment\nscores into a simple feed-forward network which achieves the best results when\ncombined alignment extractors are used.\n"}
{"abstract": "  We propose subsampling as a unified algorithmic technique for submodular\nmaximization in centralized and online settings. The idea is simple:\nindependently sample elements from the ground set, and use simple combinatorial\ntechniques (such as greedy or local search) on these sampled elements. We show\nthat this approach leads to optimal/state-of-the-art results despite being much\nsimpler than existing methods. In the usual offline setting, we present\nSampleGreedy, which obtains a $(p + 2 + o(1))$-approximation for maximizing a\nsubmodular function subject to a $p$-extendible system using $O(n + nk/p)$\nevaluation and feasibility queries, where $k$ is the size of the largest\nfeasible set. The approximation ratio improves to $p+1$ and $p$ for monotone\nsubmodular and linear objectives, respectively. In the streaming setting, we\npresent SampleStreaming, which obtains a $(4p +2 - o(1))$-approximation for\nmaximizing a submodular function subject to a $p$-matchoid using $O(k)$ memory\nand $O(km/p)$ evaluation and feasibility queries per element, where $m$ is the\nnumber of matroids defining the $p$-matchoid. The approximation ratio improves\nto $4p$ for monotone submodular objectives. We empirically demonstrate the\neffectiveness of our algorithms on video summarization, location summarization,\nand movie recommendation tasks.\n"}
{"abstract": "  We present GNNAutoScale (GAS), a framework for scaling arbitrary\nmessage-passing GNNs to large graphs. GAS prunes entire sub-trees of the\ncomputation graph by utilizing historical embeddings from prior training\niterations, leading to constant GPU memory consumption in respect to input node\nsize without dropping any data. While existing solutions weaken the expressive\npower of message passing due to sub-sampling of edges or non-trainable\npropagations, our approach is provably able to maintain the expressive power of\nthe original GNN. We achieve this by providing approximation error bounds of\nhistorical embeddings and show how to tighten them in practice. Empirically, we\nshow that the practical realization of our framework, PyGAS, an easy-to-use\nextension for PyTorch Geometric, is both fast and memory-efficient, learns\nexpressive node representations, closely resembles the performance of their\nnon-scaling counterparts, and reaches state-of-the-art performance on\nlarge-scale graphs.\n"}
{"abstract": "  Synthesizing graceful and life-like behaviors for physically simulated\ncharacters has been a fundamental challenge in computer animation. Data-driven\nmethods that leverage motion tracking are a prominent class of techniques for\nproducing high fidelity motions for a wide range of behaviors. However, the\neffectiveness of these tracking-based methods often hinges on carefully\ndesigned objective functions, and when applied to large and diverse motion\ndatasets, these methods require significant additional machinery to select the\nappropriate motion for the character to track in a given scenario. In this\nwork, we propose to obviate the need to manually design imitation objectives\nand mechanisms for motion selection by utilizing a fully automated approach\nbased on adversarial imitation learning. High-level task objectives that the\ncharacter should perform can be specified by relatively simple reward\nfunctions, while the low-level style of the character's behaviors can be\nspecified by a dataset of unstructured motion clips, without any explicit clip\nselection or sequencing. These motion clips are used to train an adversarial\nmotion prior, which specifies style-rewards for training the character through\nreinforcement learning (RL). The adversarial RL procedure automatically selects\nwhich motion to perform, dynamically interpolating and generalizing from the\ndataset. Our system produces high-quality motions that are comparable to those\nachieved by state-of-the-art tracking-based techniques, while also being able\nto easily accommodate large datasets of unstructured motion clips. Composition\nof disparate skills emerges automatically from the motion prior, without\nrequiring a high-level motion planner or other task-specific annotations of the\nmotion clips. We demonstrate the effectiveness of our framework on a diverse\ncast of complex simulated characters and a challenging suite of motor control\ntasks.\n"}
{"abstract": "  The classical Banach space $L_1(L_p)$ consists of measurable scalar functions\n$f$ on the unit square for which $$\\|f\\| = \\int_0^1\\Big(\\int_0^1 |f(x,y)|^p\ndy\\Big)^{1/p}dx < \\infty.$$ We show that $L_1(L_p)$ $(1 < p < \\infty)$ is\nprimary, meaning that, whenever $L_1(L_p) = E\\oplus F$ then either $E$ or $F$\nis isomorphic to $L_1(L_p)$. More generally we show that $L_1(X)$ is primary,\nfor a large class of rearrangement invariant Banach function spaces.\n"}
{"abstract": "  Inspired by recent developments in the study of chaos in many-body systems,\nwe construct a measure of local information spreading for a stochastic Cellular\nAutomaton in the form of a spatiotemporally resolved Hamming distance. This\ndecorrelator is a classical version of an Out-of-Time-Order Correlator studied\nin the context of quantum many-body systems. Focusing on the one-dimensional\nKauffman Cellular Automaton, we extract the scaling form of our decorrelator\nwith an associated butterfly velocity $v_b$ and a velocity-dependent Lyapunov\nexponent $\\lambda(v)$. The existence of the latter is not a given in a discrete\nclassical system. Second, we account for the behaviour of the decorrelator in a\nframework based solely on the boundary of the information spreading, including\nan effective boundary random walk model yielding the full functional form of\nthe decorrelator. In particular, we obtain analytic results for $v_b$ and the\nexponent $\\beta$ in the scaling ansatz $\\lambda(v) \\sim \\mu (v - v_b)^\\beta$,\nwhich is usually only obtained numerically. Finally, a full scaling collapse\nestablishes the decorrelator as a unifying diagnostic of information spreading.\n"}
{"abstract": "  The article presents and evaluates a scalable FRaGenLP algorithm for\ngenerating random linear programming problems of large dimension $n$ on cluster\ncomputing systems. To ensure the consistency of the problem and the boundedness\nof the feasible region, the constraint system includes $2n+1$ standard\ninequalities, called support inequalities. New random inequalities are\ngenerated and added to the system in a manner that ensures the consistency of\nthe constraints. Furthermore, the algorithm uses two likeness metrics to\nprevent the addition of a new random inequality that is similar to one already\npresent in the constraint system. The algorithm also rejects random\ninequalities that cannot affect the solution of the linear programming problem\nbounded by the support inequalities. The parallel implementation of the\nFRaGenLP algorithm is performed in C++ through the parallel BSF-skeleton, which\nencapsulates all aspects related to the MPI-based parallelization of the\nprogram. We provide the results of large-scale computational experiments on a\ncluster computing system to study the scalability of the FRaGenLP algorithm.\n"}
{"abstract": "  We theoretically study the fundamental problem of learning a single neuron\nwith a bias term ($\\mathbf{x} \\mapsto \\sigma(<\\mathbf{w},\\mathbf{x}> + b)$) in\nthe realizable setting with the ReLU activation, using gradient descent.\nPerhaps surprisingly, we show that this is a significantly different and more\nchallenging problem than the bias-less case (which was the focus of previous\nworks on single neurons), both in terms of the optimization geometry as well as\nthe ability of gradient methods to succeed in some scenarios. We provide a\ndetailed study of this problem, characterizing the critical points of the\nobjective, demonstrating failure cases, and providing positive convergence\nguarantees under different sets of assumptions. To prove our results, we\ndevelop some tools which may be of independent interest, and improve previous\nresults on learning single neurons.\n"}
{"abstract": "  In this pilot study, we investigate the use of a deep learning (DL) model to\ntemporally evolve the dynamics of gas accreting onto a black hole in the form\nof a radiatively inefficient accretion flow (RIAF). We have trained a machine\nto forecast such a spatiotemporally chaotic system -- i.e. black hole weather\nforecasting -- using a convolutional neural network (CNN) and a training\ndataset which consists of numerical solutions of the hydrodynamical equations,\nfor a range of initial conditions. We find that deep neural networks seem to\nlearn well black hole accretion physics and evolve the accretion flow orders of\nmagnitude faster than traditional numerical solvers, while maintaining a\nreasonable accuracy for a long time. For instance, CNNs predict well the\ntemporal evolution of a RIAF over a long duration of $8\\times 10^4 GM/c^3$,\nwhich corresponds to 80 dynamical times at $r=100 GM/c^2$. The DL model is able\nto evolve flows from initial conditions not present in the training dataset\nwith good accuracy. Our approach thus seems to generalize well. Once trained,\nthe DL model evolves a turbulent RIAF on a single GPU four orders of magnitude\nfaster than usual fluid dynamics integrators running in parallel on 200 CPU\ncores. We speculate that a data-driven machine learning approach should be very\npromising for accelerating not only fluid dynamics simulations, but also\ngeneral relativistic magnetohydrodynamic ones.\n"}
{"abstract": "  We analyze the magnetic excitations and the spin-mediated superconductivity\nin iron-based superconductors within a low-energy model that operates in the\nband basis but fully incorporates the orbital character of the spin\nexcitations. We show how the orbital selectivity, encoded in our low-energy\ndescription, simplifies substantially the analysis and allows for analytical\ntreatments, while retaining all the main features of both spin-excitations and\ngap functions computed using multiorbital models. Importantly, our analysis\nunveils the orbital matching between the hole and electron pockets as the key\nparameter to determine the momentum-dependence and the hierarchy of the\nsuperconducting gaps, instead of the Fermi surface matching as in the common\nnesting scenario.\n"}
{"abstract": "  Let $S$ be a symmetric linear relation in the Pontyagin space $\\left(\n\\mathcal{K},\\, \\left[ .,. \\right] \\right)$ and let $\\Pi=\\left(\n\\mathcal{\\mathcal{H}}, \\Gamma_{0}, \\Gamma_{1} \\right)$ be the corresponding\nboundary triple. We prove that the corresponding Weyl function $Q$ satisfies\n$Q\\in N_{\\kappa }(\\mathcal{H})$. Conversely, for regular $Q\\in N_{\\kappa\n}(\\mathcal{H})$, we find linear relation $S \\subsetneq A$, where $A$ is\nrepresenting self-adjoint linear relation of $Q$, and we prove that $Q$ is the\nWeyl function of the relation $S$. We also prove $\\hat{A}=\\ker \\Gamma_{1}$,\nwhere $\\hat{A}$ is the representing relation of the $\\hat{Q}:=-Q^{-1}$. In\naddition, if we assume that the derivative at infinity $Q^{'}\\left( \\infty\n\\right):=\\lim \\limits_{z \\to \\infty}{zQ(z)}$ is a boundedly invertible operator\nthen we are able to decompose $A$, $\\hat{A}$ and $S^{+}$ in terms of $S$, i.e.\nwe express relation matrices of $A$, $\\hat{A}$ and $S^{+}$ in terms of $S$,\nwhich is a bounded operator in this case.\n"}
{"abstract": "  We consider the robust single-source capacitated facility location problem\nwith uncertainty in customer demands. A cardinality-constrained uncertainty set\nis assumed for the robust problem. To solve it efficiently, we propose an\nallocation-based formulation derived by Dantzig-Wolfe decomposition and a\nbranch-and-price algorithm. The computational experiments show that our\nbranch-and-price algorithm outperforms CPLEX in many cases, which solves the\nordinary robust reformulation. We also examine the trade-off relationship\nbetween the empirical probability of infeasibility and the additional costs\nincurred and observe that the robustness of solutions can be improved\nsignificantly with small additional costs.\n"}
{"abstract": "  We propose using deep reinforcement learning to solve dynamic stochastic\ngeneral equilibrium models. Agents are represented by deep artificial neural\nnetworks and learn to solve their dynamic optimisation problem by interacting\nwith the model environment, of which they have no a priori knowledge. Deep\nreinforcement learning offers a flexible yet principled way to model bounded\nrationality within this general class of models. We apply our proposed approach\nto a classical model from the adaptive learning literature in macroeconomics\nwhich looks at the interaction of monetary and fiscal policy. We find that,\ncontrary to adaptive learning, the artificially intelligent household can solve\nthe model in all policy regimes.\n"}
{"abstract": "  Given a hermitian line bundle $L\\to M$ on a closed Riemannian manifold\n$(M^n,g)$, the self-dual Yang-Mills-Higgs energies are a natural family of\nfunctionals \\begin{align*}\n  &E_\\epsilon(u,\\nabla):=\\int_M\\Big(|\\nabla\nu|^2+\\epsilon^2|F_\\nabla|^2+\\frac{(1-|u|^2)^2}{4\\epsilon^2}\\Big) \\end{align*}\ndefined for couples $(u,\\nabla)$ consisting of a section $u\\in\\Gamma(L)$ and a\nhermitian connection $\\nabla$ with curvature $F_\\nabla$. While the critical\npoints of these functionals have been well-studied in dimension two by the\ngauge theory community, it was shown in previous work of the second- and\nthird-named authors that critical points in higher dimension converge as\n$\\epsilon\\to 0$ (in an appropriate sense) to minimal submanifolds of\ncodimension two, with strong parallels to the correspondence between the\nAllen-Cahn equations and minimal hypersurfaces.\n  In this paper, we complement this idea by showing the $\\Gamma$-convergence of\n$E_\\epsilon$ to ($2\\pi$ times) the codimension two area: more precisely, given\na family of couples $(u_\\epsilon,\\nabla_\\epsilon)$ with $\\sup_\\epsilon\nE_\\epsilon(u_\\epsilon,\\nabla_\\epsilon)<\\infty$, we prove that a suitable gauge\ninvariant Jacobian $J(u_\\epsilon,\\nabla_\\epsilon)$ converges to an integral\n$(n-2)$-cycle $\\Gamma$, in the homology class dual to the Euler class $c_1(L)$,\nwith mass $2 \\pi \\mathbb{M}(\\Gamma)\\le\\liminf_{\\epsilon \\rightarrow\n0}E_\\epsilon(u_\\epsilon,\\nabla_\\epsilon)$. We also obtain a recovery sequence\nfor any integral cycle in this homology class.\n  Finally, we apply these techniques to compare min-max values for the\n$(n-2)$-area from the Almgren-Pitts theory with those obtained from the\nYang-Mills-Higgs framework, showing that the former values always provide a\nlower bound for the latter. As an ingredient, we also establish a Huisken-type\nmonotonicity result along the gradient flow of $E_{\\epsilon}$.\n"}
{"abstract": "  We study the problem of polygonal curve simplification under uncertainty,\nwhere instead of a sequence of exact points, each uncertain point is\nrepresented by a region, which contains the (unknown) true location of the\nvertex. The regions we consider are disks, line segments, convex polygons, and\ndiscrete sets of points. We are interested in finding the shortest subsequence\nof uncertain points such that no matter what the true location of each\nuncertain point is, the resulting polygonal curve is a valid simplification of\nthe original polygonal curve under the Hausdorff or the Fr\\'echet distance. For\nboth these distance measures, we present polynomial-time algorithms for this\nproblem.\n"}
{"abstract": "  We show for the first time that a multilayer perceptron (MLP) can serve as\nthe only scene representation in a real-time SLAM system for a handheld RGB-D\ncamera. Our network is trained in live operation without prior data, building a\ndense, scene-specific implicit 3D model of occupancy and colour which is also\nimmediately used for tracking.\n  Achieving real-time SLAM via continual training of a neural network against a\nlive image stream requires significant innovation. Our iMAP algorithm uses a\nkeyframe structure and multi-processing computation flow, with dynamic\ninformation-guided pixel sampling for speed, with tracking at 10 Hz and global\nmap updating at 2 Hz. The advantages of an implicit MLP over standard dense\nSLAM techniques include efficient geometry representation with automatic detail\ncontrol and smooth, plausible filling-in of unobserved regions such as the back\nsurfaces of objects.\n"}
{"abstract": "  Context. Magnetic helicity is a physical quantity of great importance in the\nstudy of astrophysical and natural plasmas. Although a density for helicity\ncannot be defined, a good proxy for it is field line helicity. The appropriate\nquantity for use in solar conditions is relative field line helicity (RFLH).\nAims. This work aims to study in detail the behaviour of RFLH, for the first\ntime, in a solar active region (AR). Methods. The target active region is the\nlarge, eruptive AR 11158. In order to compute RFLH and all other quantities of\ninterest we use a non-linear force-free reconstruction of the AR coronal\nmagnetic field of excelent quality. Results. We find that the photospheric\nmorphology of RFLH is quite different than that of the magnetic field or of the\nelectrical current, and this is not sensitive to the chosen gauge in the\ncomputation of RFLH. The value of helicity experiences a large decrease, 25% of\nits pre-flare value, during an X-class flare of the AR, a change that is also\ndepicted in the photospheric morphology of RFLH. Moreover, the area of this\nchange coincides with the area that encompasses the flux rope, the magnetic\nstructure that later erupted. Conclusions. The use of RFLH can provide\nimportant information about the value and location of the magnetic helicity\nexpelled from the solar atmosphere during eruptive events.\n"}
{"abstract": "  We propose a new algorithm for Promise Constraint Satisfaction Problems\n(PCSPs). It is a combination of the Constraint Basic LP relaxation and the\nAffine IP relaxation (CLAP). We give a characterisation of the power of CLAP in\nterms of a minion homomorphism. Using this characterisation, we identify a\ncertain weak notion of symmetry which, if satisfied by infinitely many\npolymorphisms of PCSPs, guarantees tractability.\n  We demonstrate that there are PCSPs solved by CLAP that are not solved by any\nof the existing algorithms for PCSPs; in particular, not by the BLP+AIP\nalgorithm of Brakensiek and Guruswami [SODA'20] and not by a reduction to\ntractable finite-domain CSPs.\n"}
{"abstract": "  We provide a systematic analysis of the multipolar gravitational waveform,\nenergy and angular momentum fluxes emitted by a nonspinning test particle\norbiting a Kerr black hole along equatorial, eccentric orbits. These quantities\nare computed by numerically solving the Teukolsky equation in the time domain\nand are then used to test and improve the radiation reaction (and waveform) of\nan effective-one-body (EOB) model. Eccentricity is incorporated into EOB by\nreplacing the quasi-circular Newtonian (or leading-order) prefactors in the\nEOB-factorized multipolar waveform (and fluxes) with their generic\ncounterparts. The comparison between numerical and analytical quantities is\ncarried out over a large portion of the parameter space, notably for orbits\nclose to the separatrix and with high eccentricities. The analytical model\nagrees to $\\sim 1\\%$ with the numerical data for orbits with moderate\neccentricities ($e\\lesssim 0.3$) and moderate spins ($\\hat{a}\\lesssim 0.5$),\nalthough this increases up to $\\sim 33\\%$ for large, positive, black hole spins\n($\\sim 0.9$) and large eccentricities ($\\sim 0.9$). For moderate\neccentricities, the new EOB fluxes can be used to drive the dynamics through\nthe nonadiabatic transition from eccentric inspiral to plunge, merger and\nringdown, thus providing accurate an description of the merger. We also show\nthat the same technique can be applied to hyperbolic encounters. In general,\nour approach to radiation reaction for eccentric inspirals should be seen as a\nfirst step toward EOB modelization of extreme-mass-ratio-inspirals waveforms\nfor LISA.\n"}
{"abstract": "  There is an increasing scientific interest in automatically analysing and\nunderstanding human behavior, with particular reference to the evolution of\nfacial expressions and the recognition of the corresponding emotions. In this\npaper we propose a technique based on Functional ANOVA to extract significant\npatterns of face muscles movements, in order to identify the emotions expressed\nby actors in recorded videos. We determine if there are time-related\ndifferences on expressions among emotional groups by using a functional F-test.\nSuch results are the first step towards the construction of a reliable\nautomatic emotion recognition system\n"}
{"abstract": "  Multimedia content is of predominance in the modern Web era. Investigating\nhow users interact with multimodal items is a continuing concern within the\nrapid development of recommender systems. The majority of previous work focuses\non modeling user-item interactions with multimodal features included as side\ninformation. However, this scheme is not well-designed for multimedia\nrecommendation. Specifically, only collaborative item-item relationships are\nimplicitly modeled through high-order item-user-item relations. Considering\nthat items are associated with rich contents in multiple modalities, we argue\nthat the latent semantic item-item structures underlying these multimodal\ncontents could be beneficial for learning better item representations and\nfurther boosting recommendation. To this end, we propose a LATent sTructure\nmining method for multImodal reCommEndation, which we term LATTICE for brevity.\nTo be specific, in the proposed LATTICE model, we devise a novel modality-aware\nstructure learning layer, which learns item-item structures for each modality\nand aggregates multiple modalities to obtain latent item graphs. Based on the\nlearned latent graphs, we perform graph convolutions to explicitly inject\nhigh-order item affinities into item representations. These enriched item\nrepresentations can then be plugged into existing collaborative filtering\nmethods to make more accurate recommendations. Extensive experiments on three\nreal-world datasets demonstrate the superiority of our method over\nstate-of-the-art multimedia recommendation methods and validate the efficacy of\nmining latent item-item relationships from multimodal features.\n"}
{"abstract": "  Over the last years, machine learning techniques have been applied to more\nand more application domains, including software engineering and, especially,\nsoftware quality assurance. Important application domains have been, e.g.,\nsoftware defect prediction or test case selection and prioritization. The\nability to predict which components in a large software system are most likely\nto contain the largest numbers of faults in the next release helps to better\nmanage projects, including early estimation of possible release delays, and\naffordably guide corrective actions to improve the quality of the software.\nHowever, developing robust fault prediction models is a challenging task and\nmany techniques have been proposed in the literature. Closely related to\nestimating defect-prone parts of a software system is the question of how to\nselect and prioritize test cases, and indeed test case prioritization has been\nextensively researched as a means for reducing the time taken to discover\nregressions in software. In this survey, we discuss various approaches in both\nfault prediction and test case prioritization, also explaining how in recent\nstudies deep learning algorithms for fault prediction help to bridge the gap\nbetween programs' semantics and fault prediction features. We also review\nrecently proposed machine learning methods for test case prioritization (TCP),\nand their ability to reduce the cost of regression testing without negatively\naffecting fault detection capabilities.\n"}
{"abstract": "  We present our recent study of cross sections and angular distributions of\nprojectile fragments from heavy-ion reactions at beam energy of 15 MeV/nucleon.\nWe studied the production cross sections and the angular distributions of\nneutron-rich nuclides from collisions of a 86 Kr (15 MeV/nucleon) beam with\nheavy targets ( 64 Ni, 124 Sn and 238 U). Experimental data from our previous\nwork at Texas A & M were compared with model calculations. Our calculations\nwere based on a two-step approach: the dynamical stage of the collision was\ndescribed with, first, the phenomenological Deep-Inelastic Transfer model (DIT)\nand, alternatively, with the microscopic Constrained Molecular Dynamics model\n(CoMD). The de-excitation of the hot heavy projectile fragments was performed\nwith the Statistical Multifragmentation Model (SMM). An overall good\ndiscription of the available data was obtained with the models employed.\nFurthermore, we performed calculations with a radioactive beam of 92 Kr (15\nMeV/nucleon) interacting with a target of 238 U. We observed that the\nmultinucleon transfer mechanism leads to extremely neutron-rich nuclides toward\nand beyond the astrophysical r-process path.\n"}
{"abstract": "  For scattering amplitudes in strong background fields, it is -- at least in\nprinciple -- possible to perturbatively expand the background to obtain\nhigher-point vacuum amplitudes. In the case of self-dual plane wave backgrounds\nwe consider this expansion for two-point, one-loop amplitudes in pure\nYang-Mills, QED and QCD. This enables us to obtain multicollinear limits of\n1-loop vacuum amplitudes; the resulting helicity configurations are\nsurprisingly restricted, with only the all-plus helicity amplitude surviving.\nThese results are shown to be consistent with well-known vacuum amplitudes. We\nalso show that for both abelian and non-abelian supersymmetric gauge theories,\nthere is no helicity flip (and hence no vacuum birefringence) on any plane wave\nbackground, generalising a result previously known in the Euler-Heisenberg\nlimit of super-QED.\n"}
{"abstract": "  Reinforcement learning is an active research area with a vast number of\napplications in robotics, and the RoboCup competition is an interesting\nenvironment for studying and evaluating reinforcement learning methods. A known\ndifficulty in applying reinforcement learning to robotics is the high number of\nexperience samples required, being the use of simulated environments for\ntraining the agents followed by transfer learning to real-world (sim-to-real) a\nviable path. This article introduces an open-source simulator for the IEEE Very\nSmall Size Soccer and the Small Size League optimized for reinforcement\nlearning experiments. We also propose a framework for creating OpenAI Gym\nenvironments with a set of benchmarks tasks for evaluating single-agent and\nmulti-agent robot soccer skills. We then demonstrate the learning capabilities\nof two state-of-the-art reinforcement learning methods as well as their\nlimitations in certain scenarios introduced in this framework. We believe this\nwill make it easier for more teams to compete in these categories using\nend-to-end reinforcement learning approaches and further develop this research\narea.\n"}
{"abstract": "  We show that the polygamous nature of multi-party quantum correlations can be\ncharacterized in a {\\em stronger} form based on Tsallis $q$-entropy and\n$q$-expectation value. By considering the amount of entanglement that can be\ndistributed in multi-party systems, we establish a class of strong polygamy\ninequalities of multi-party entanglement in terms of Tsallis $q$-entropy and\n$q$-expectation for $q \\geq 1$. Our new class of inequalities is in fact\ntighter than the usual polygamy inequalities of multi-party entanglement, and\nthe tightness is explicitly illustrated by an example. Moreover, our new class\nof inequalities is concerned with the $q$-expected entanglement distributed\nbetween a single party and any possible subsets of the rest parties whereas the\nusual polygamy inequality only considers the entanglement between a single\nparty and another. We further establish the equivalence between strong polygamy\ninequalities of quantum entanglement and quantum discord distributed in\nmulti-party quantum systems.\n"}
{"abstract": "  Despite the great efforts to find an effective way for COVID-19 prediction,\nthe virus nature and mutation represent a critical challenge to diagnose the\ncovered cases. However, developing a model to predict COVID-19 via Chest X-Ray\n(CXR) images with accurate performance is necessary to help in early diagnosis.\nIn this paper, a hybrid quantum-classical convolutional Neural Networks (HQCNN)\nmodel used the random quantum circuits (RQCs) as a base to detect COVID-19\npatients with CXR images. A collection of 6952 CXR images, including 1161\nCOVID-19, 1575 normal, and 5216 pneumonia images, were used as a dataset in\nthis work. The proposed HQCNN model achieved higher performance with an\naccuracy of 98.4\\% and a sensitivity of 99.3\\% on the first dataset cases.\nBesides, it obtained an accuracy of 99\\% and a sensitivity of 99.7\\% on the\nsecond dataset cases. Also, it achieved accuracy, and sensitivity of 88.6\\%,\nand 88.7\\%, respectively, on the third multi-class dataset cases. Furthermore,\nthe HQCNN model outperforms various models in balanced accuracy, precision,\nF1-measure, and AUC-ROC score. The experimental results are achieved by the\nproposed model prove its ability in predicting positive COVID-19 cases.\n"}
{"abstract": "  In this paper, we consider decentralized federated learning (FL) over\nwireless networks, where over-the-air computation (AirComp) is adopted to\nfacilitate the local model consensus in a device-to-device (D2D) communication\nmanner. However, the AirComp-based consensus phase brings the additive noise in\neach algorithm iterate and the consensus needs to be robust to wireless network\ntopology changes, which introduce a coupled and novel challenge of establishing\nthe convergence for wireless decentralized FL algorithm. To facilitate\nconsensus phase, we propose an AirComp-based DSGD with gradient tracking and\nvariance reduction (DSGT-VR) algorithm, where both precoding and decoding\nstrategies are developed for D2D communication. Furthermore, we prove that the\nproposed algorithm converges linearly and establish the optimality gap for\nstrongly convex and smooth loss functions, taking into account the channel\nfading and noise. The theoretical result shows that the additional error bound\nin the optimality gap depends on the number of devices. Extensive simulations\nverify the theoretical results and show that the proposed algorithm outperforms\nother benchmark decentralized FL algorithms over wireless networks.\n"}
{"abstract": "  Fels-Kaup (Acta Mathematica 2008) classified homogeneous $\\mathfrak{C}_{2,1}$\nhypersurfaces $M^5 \\subset \\mathbb{C}^3$ and discovered that they are all\nbiholomorphic to tubes $S^2 \\times i \\mathbb{R}^3$ over some affinely\nhomogeneous surface $S^2 \\subset \\mathbb{R}^3$. The second and third authors in\n2003.08166, by performing highly non-straightforward calculations, conducted\nthe Cartan method of equivalence to classify homogeneous models of PDE systems\nrelated to such $\\mathfrak{C}^{2,1}$ hypersurfaces $M^5 \\subset \\mathbb{C}^3$.\nKolar-Kossovskiy 1905.05629 and the authors 2003.01952 constructed a formal and\na convergent Poincar\\'e-Moser normal form for $\\mathfrak{C}_{2,1}$\nhypersurfaces $M^5 \\subset \\mathbb{C}^3$. But this was only a first,\npreliminary step. Indeed, the invariant branching tree underlying Fels-Kaup's\nclassification was still missing in the literature, due to computational\nobstacles. The present work applies the power series method of equivalence,\nconfirms Fels-Kaup 2008, and finds a differential-invariant tree. To terminate\nthe middle (thickest) branch, it is necessary to compute up to order $10$ with\n$5$ variables. Again, calculations, done by hand, are non-straightforward.\n"}
{"abstract": "  Pulsar searches are computationally demanding efforts to discover dispersed\nperiodic signals in time- and frequency-resolved data from radio telescopes.\nThe complexity and computational expense of simultaneously determining the\nfrequency-dependent delay (dispersion) and the periodicity of the signal is\nfurther exacerbated by the presence of various types of radio-frequency\ninterference (RFI) and observing-system effects. New observing systems with\nwider bandwidths, higher bit rates and greater overall sensitivity (also to\nRFI) further enhance these challenges. We present a novel approach to the\nanalysis of pulsar search data. Specifically, we present a neural-network-based\npipeline that efficiently suppresses a wide range of RFI signals and\ninstrumental instabilities and furthermore corrects for (a priori unknown)\ninterstellar dispersion. After initial training of the network, our analysis\ncan be run in real time on a standard desktop computer with a commonly\navailable, consumer-grade GPU. We complement our neural network with standard\nalgorithms for periodicity searches. In particular with the Fast Fourier\nTransform (FFT) and the Fast Folding Algorithm (FFA) and demonstrate that with\nthese straightforward extensions, our method is capable of identifying even\nfaint pulsars, while maintaining an extremely low number of false positives. We\nfurthermore apply our analysis to a subset of the PALFA survey and demonstrate\nthat in most cases the automated dispersion removal of our network produces a\ntime series of similar quality as dedispersing using the actual dispersion\nmeasure of the pulsar in question. On our test data we are able to make\npredictions whether a pulsar is present in the data or not 200 times faster\nthan real time.\n"}
{"abstract": "  Continuum mechanics simulators, numerically solving one or more partial\ndifferential equations, are essential tools in many areas of science and\nengineering, but their performance often limits application in practice. Recent\nmodern machine learning approaches have demonstrated their ability to\naccelerate spatio-temporal predictions, although, with only moderate accuracy\nin comparison. Here we introduce MultiScaleGNN, a novel multi-scale graph\nneural network model for learning to infer unsteady continuum mechanics.\nMultiScaleGNN represents the physical domain as an unstructured set of nodes,\nand it constructs one or more graphs, each of them encoding different scales of\nspatial resolution. Successive learnt message passing between these graphs\nimproves the ability of GNNs to capture and forecast the system state in\nproblems encompassing a range of length scales. Using graph representations,\nMultiScaleGNN can impose periodic boundary conditions as an inductive bias on\nthe edges in the graphs, and achieve independence to the nodes' positions. We\ndemonstrate this method on advection problems and incompressible fluid\ndynamics. Our results show that the proposed model can generalise from uniform\nadvection fields to high-gradient fields on complex domains at test time and\ninfer long-term Navier-Stokes solutions within a range of Reynolds numbers.\nSimulations obtained with MultiScaleGNN are between two and four orders of\nmagnitude faster than the ones on which it was trained.\n"}
{"abstract": "  In robot sensing scenarios, instead of passively utilizing human captured\nviews, an agent should be able to actively choose informative viewpoints of a\n3D object as discriminative evidence to boost the recognition accuracy. This\ntask is referred to as active object recognition. Recent works on this task\nrely on a massive amount of training examples to train an optimal view\nselection policy. But in realistic robot sensing scenarios, the large-scale\ntraining data may not exist and whether the intelligent view selection policy\ncan be still learned from few object samples remains unclear. In this paper, we\nstudy this new problem which is extremely challenging but very meaningful in\nrobot sensing -- Few-shot Active Object Recognition, i.e., to learn view\nselection policies from few object samples, which has not been considered and\naddressed before. We solve the proposed problem by adopting the framework of\nmeta learning and name our method \"MetaView\". Extensive experiments on both\ncategory-level and instance-level classification tasks demonstrate that the\nproposed method can efficiently resolve issues that are hard for\nstate-of-the-art active object recognition methods to handle, and outperform\nseveral baselines by large margins.\n"}
{"abstract": "  Bitcoin has become the leading cryptocurrency system, but the limit on its\ntransaction processing capacity has resulted in increased transaction fees and\ndelayed transaction confirmation. As such, it is pertinent to understand and\nprobably predict how transactions are handled by Bitcoin such that a user may\nadapt the transaction requests and a miner may adjust the block generation\nstrategy and/or the mining pool to join. To this aim, the present paper\nintroduces results from an analysis of transaction handling in Bitcoin.\n  Specifically, the analysis consists of two-part. The first part is an\nexploratory data analysis revealing key characteristics in Bitcoin transaction\nhandling. The second part is a predictability analysis intended to provide\ninsights on transaction handling such as (i) transaction confirmation time,\n(ii) block attributes, and (iii) who has created the block. The result shows\nthat some models do reasonably well for (ii), but surprisingly not for (i) or\n(iii).\n"}
{"abstract": "  As new classes of transients and variable stars are discovered, and\ntheoretical models are established to work or not to work for a few members of\nthe class, it is often the case that some researchers will make arguments on\nthe basis of Occam's razor that all members of the class must be produced by\nwhichever mechanism first successfully explained one of the objects. It is also\nfrequent that this assumption will be more more implicitly. Retrospective\nanalysis shows rather clearly that this argument fails a large fraction of the\ntime, and in many cases, this search for false consistency has led to more\nfundamental astrophysical errors, a few of which are quite prominent in the\nhistory of astronomy. A corollary of this is that on numerous occasions,\ntheoretical models to explain transients have turned out to be models that\ndescribe real (but often not yet discovered) phenomenona other than the ones to\nwhich they have first been applied, albeit with minor errors that caused the\nmodel to appear to fit to a known phenomenon it did not describe. A set of\nexamples of such events is presented here (some of which will be quite familiar\nto most astronomers), along with a discussion of why this phenomenon occurs,\nhow it may be manifesting itself at the present time. Some discussion will also\nbe made of why and when survey designs have led to immediate separation of\nvarious transient mechanisms, generally by being overpowered in some way\nrelative to what is needed to {\\it detect} a new class of objects.\n"}
{"abstract": "  A novel multifunctional nanodevice based in doxorubicin (DOX)- loaded\nmesoporous silica nanoparticles (MSNs) as nanoplatforms for the assembly of\ndifferent building blocks has been developed for bone cancer treatment. These\nbuilding blocks consists of: i) a polyacrylic acid (PAA) capping layer grafted\nto MSNs via an acid-cleavable acetal linker, to minimize premature cargo\nrelease and provide the nanosystem of pHresponsive drug delivery ability; and\nii) a targeting ligand, the plant lectin concanavalin A (ConA), able to\nselectively recognize, bind and internalize owing to certain cell-surface\nglycans, such as sialic acids (SA), overexpressed in given tumor cells. This\nmultifunctional nanosystem exhibits a noticeable higher internalization degree\ninto human osteosarcoma cells (HOS), overexpressing SA, compared to healthy\npreosteoblast cells (MC3T3-E1). Moreover, the results indicate that small DOX\nloading leads to almost 100% of osteosarcoma cell death in comparison with\nhealthy bone cells, which significantly preserve their viability. Besides, this\nnanodevice has a cytotoxicity on tumor cells 8- fold higher than that caused by\nthe free drug. These findings demonstrate that the synergistic combination of\ndifferent building blocks into a unique nanoplatform increases antitumor\neffectiveness and decreases toxicity towards normal cells. This line of attack\nopens up new insights in targeted bone cancer therapy.\n"}
{"abstract": "  We generalise the correspondence between $\\aleph 0$-categorical theories and\ntheir automorphism groups to arbitrary complete theories in classical logic,\nand to some theories (including, in particular, all $\\aleph 0$-categorical\nones) in continuous logic.\n"}
{"abstract": "  In this paper, we consider the explicit bound for the second-order\napproximation of the quadratic variation of a general fractional Gaussian\nprocess $(G_t)_{t\\ge 0}$. The second order mixed partial derivative of the\ncovariance function $ R(t,\\, s)=\\mathbb{E}[G_t G_s]$ can be decomposed into two\nparts, one of which coincides with that of fractional Brownian motion and the\nother of which is bounded by $(ts)^{H-1}$ up to a constant factor. This\ncondition is valid for a class of continuous Gaussian processes that fails to\nbe self-similar or have stationary increments. %Some examples include the\nsubfractional Brownian motion and the bi-fractional Brownian motion.\n  Under this assumption, we obtain the optimal Berry-Ess\\'{e}en bounds when\n$H\\in (0,\\,\\frac23]$ and the upper Berry-Ess\\'{e}en bounds when $H\\in\n(\\frac23,\\,\\frac34]$. As a by-product, we also show the almost sure central\nlimit theorem (ASCLT) for the quadratic variation when $H\\in (0,\\,\\frac34]$.\nThe results extend that of \\cite{NP 09} to the case of general Gaussian\nprocesses, unify and improve the Berry-Ess\\'{e}en bounds in \\cite{Tu 11},\n\\cite{AE 12} and \\cite{KL 21} for respectively the sub-fractional Brownian\nmotion, the bi-fractional Brownian motion and the sub-bifractional Brownian\nmotion.\n"}
{"abstract": "  We show that if one starts with a Universe with some matter and a\ncosmological constant, then quantum mechanics naturally induces an attractive\ngravitational potential and an effective Newton's coupling. Thus gravity is an\nemergent phenomenon and what should be quantized are the fundamental degrees of\nfreedom from which it emerges.\n"}
{"abstract": "  R 144 is the visually brightest WR star in the Large Magellanic Cloud (LMC).\nR 144 was reported to be a binary, making it potentially the most massive\nbinary thus observed. We perform a comprehensive spectral, photometric,\norbital, and polarimetric analysis of R 144.\n  R 144 is an eccentric (e=0.51) 74.2-d binary comprising two relatively\nevolved (age~2 Myr), H-rich WR stars. The hotter primary (WN5/6h, T=50 kK) and\nthe cooler secondary (WN6/7h,T=45kK) have nearly equal masses. The combination\nof low rotation and H-depletion observed in the system is well reproduced by\ncontemporary evolution models that include boosted mass-loss at the upper-mass\nend. The systemic velocity of R 144 and its relative isolation suggest that it\nwas ejected as a runaway from the neighbouring R 136 cluster. The optical\nlight-curve shows a clear orbital modulation that can be well explained as a\ncombination of two processes: excess emission stemming from wind-wind\ncollisions and double wind eclipses. Our light-curve model implies an orbital\ninclination of i=60.4+-1.5deg, resulting in accurately constrained dynamical\nmasses of 74+-4 and 69+-4 Msun. Assuming that both binary components are core\nH-burning, these masses are difficult to reconcile with the derived\nluminosities (logL1,2 = 6.44, 6.39 [Lsun]), which correspond to evolutionary\nmasses of the order of 110 and 100Msun, respectively. Taken at face value, our\nresults imply that both stars have high classical Eddington factors of Gamma_e\n= 0.78+-0.1. If the stars are on the main sequence, their derived radii\n(~25Rsun) suggest that they are only slightly inflated, even at this high\nEddington factor. Alternatively, the stars could be core-He burning, strongly\ninflated from the regular size of classical Wolf-Rayet stars (~1Rsun), a\nscenario that could help resolve the observed mass discrepancy.\n"}
{"abstract": "  The Eco-Driving control problem seeks to perform fuel efficient speed\nplanning for a Connected and Autonomous Vehicle (CAV) that can exploit\ninformation available from advanced mapping, and from Vehicle-to-Everything\n(V2X) communication. The ability of an Eco-Driving strategy to adapt in real\ntime to variable traffic scenarios where surrounding vehicles can be either\nconnected or unconnected is critical for further development and deployment of\nthis technology in the transportation sector. In this work, the Eco-Driving\nstrategy, formulated as a receding-horizon optimal control problem, is\nintegrated with a target vehicle speed prediction model and solved via Dynamic\nProgramming (DP) to determine the optimal speed trajectory in the presence of a\nhuman-driven target vehicle. An encoder-decoder architecture analyzes the\npatterns in the target vehicle velocity recorded over a historic window using a\nGated-Recurrent-Unit (GRU) based encoder and generates an estimate of the\nfuture velocity trajectory using the GRU based decoder. A sensitivity study is\ndone to analyze the effect of the historical and prediction windows on the\naccuracy of the velocity predictor. The proposed Eco-Driving controller is\nevaluated through microscopic simulations using a traffic simulator.\n"}
{"abstract": "  Observing the significance of spectrally-efficient secure non-orthogonal\nmultiple access (NOMA), this paper proposes a novel quality of service (QoS)\naware secure NOMA protocol that maximizes secrecy fairness among untrusted\nusers. Considering a base station (BS) and two users, a novel decoding order is\ndesigned that provides security to both users. With the objective of ensuring\nsecrecy fairness between users, while satisfying their QoS requirements under\nBS transmit power budget constraint, we explore the problem of minimizing the\nmaximum secrecy outage probability (SOP). Closed-form expression of pair outage\nprobability (POP) and optimal power allocation (PA) minimizing POP are\nobtained. To analyze secrecy performance, analytical expressions of SOP for\nboth users are derived, and individual SOP minimization problems are solved\nusing concept of generalized-convexity. High signal-to-noise ratio\napproximation of SOP and asymptotically optimized solution minimizing this\napproximation is also found. Furthermore, a global-optimal solution from\nsecrecy fairness standpoint is obtained at low computational complexity, and\ntight approximation is derived to get analytical insights. Numerical results\npresent useful insights on globally optimized PA which ensure secrecy fairness\nand provide performance gain of about 55.12%, 69.30%, and 19.11%, respectively,\ncompared to fixed PA and individual users' optimal PAs. Finally, a tradeoff\nbetween secrecy fairness performance and QoS demands is presented.\n"}
{"abstract": "  This work utilizes the plethora of work on verification of sequential\nprograms for the purpose of verifying concurrent programs. We reduce the\nverification of a concurrent program to a series of verification tasks of\nsequential programs. Our approach is modular in the sense that each sequential\nverification task roughly corresponds to the verification of a single thread,\nwith some additional information about the environment in which it operates.\nInformation regarding the environment is gathered during the run of the\nalgorithm, by need.\n  While our approach is general, it specializes on concurrent programs where\nthe threads are structured hierarchically. The idea is to exploit the hierarchy\nin order to minimize the amount of information that needs to be transferred\nbetween threads. To that end, we verify one of the threads, considered \"main\",\nas a sequential program. Its verification process initiates queries to its\n\"environment\" (which may contain multiple threads). Those queries are answered\nby sequential verification, if the environment consists of a single thread, or,\notherwise, by applying the same hierarchical algorithm on the environment.\n  Our technique is fully automatic, and allows us to use any off-the-shelf\nsequential model checker. We implemented our technique in a tool called CoMuS\nand evaluated it against established tools for concurrent verification. Our\nexperiments show that it works particularly well on hierarchically structured\nprograms.\n"}
{"abstract": "  Autoencoders are a widespread tool in machine learning to transform\nhigh-dimensional data into a lowerdimensional representation which still\nexhibits the essential characteristics of the input. The encoder provides an\nembedding from the input data manifold into a latent space which may then be\nused for further processing. For instance, learning interpolation on the\nmanifold may be simplified via the new manifold representation in latent space.\nThe efficiency of such further processing heavily depends on the regularity and\nstructure of the embedding. In this article, the embedding into latent space is\nregularized via a loss function that promotes an as isometric and as flat\nembedding as possible. The required training data comprises pairs of nearby\npoints on the input manifold together with their local distance and their local\nFrechet average. This regularity loss functional even allows to train the\nencoder on its own. The loss functional is computed via a Monte Carlo\nintegration which is shown to be consistent with a geometric loss functional\ndefined directly on the embedding map. Numerical tests are performed using\nimage data that encodes different data manifolds. The results show that smooth\nmanifold embeddings in latent space are obtained. These embeddings are regular\nenough such that interpolation between not too distant points on the manifold\nis well approximated by linear interpolation in latent space.\n"}
{"abstract": "  This paper studies the fish growth trajectory tracking via reinforcement\nlearning under a representative bioenergetic growth model. Due to the complex\naquaculture condition and uncertain environmental factors such as temperature,\ndissolved oxygen, un-ionized ammonia, and strong nonlinear couplings, including\nmulti-inputs of the fish growth model, the growth trajectory tracking problem\ncan not be efficiently solved by the model-based control approaches in\nprecision aquaculture. To this purpose, we formulate the growth trajectory\ntracking problem as sampled-data optimal control using discrete state-action\npairs Markov decision process. We propose two Q-learning algorithms that learn\nthe optimal control policy from the sampled data of the fish growth\ntrajectories at every stage of the fish life cycle from juveniles to the\ndesired market weight in the aquaculture environment. The Q-learning scheme\nlearns the optimal feeding control policy to fish growth rate cultured in cages\nand the optimal feeding rate control policy with an optimal temperature profile\nfor the aquaculture fish growth rate in tanks. The simulation results\ndemonstrate that both Q-learning strategies achieve high trajectory tracking\nperformance with less amount feeding rates.\n"}
{"abstract": "  Heterogeneous catalysis is an example of a complex materials function,\ngoverned by an intricate interplay of several processes, e.g., the different\nsurface chemical reactions, and the dynamic re-structuring of the catalyst\nmaterial at reaction conditions. Modelling the full catalytic progression via\nfirst-principles statistical mechanics is impractical, if not impossible.\nInstead, we show here how a tailored artificial-intelligence approach can be\napplied, even to a small number of materials, to model catalysis and determine\nthe key descriptive parameters (\"materials genes\") reflecting the processes\nthat trigger, facilitate, or hinder catalyst performance. We start from a\nconsistent experimental set of \"clean data\", containing nine vanadium-based\noxidation catalysts. These materials were synthesized, fully characterized, and\ntested according to standardized protocols. By applying the symbolic-regression\nSISSO approach, we identify correlations between the few most relevant\nmaterials properties and their reactivity. This approach highlights the\nunderlying physicochemical processes, and accelerates catalyst design.\n"}
{"abstract": "  Software development methods are usually not applied by the book. Companies\nare under pressure to continuously deploy software products that meet market\nneeds and stakeholders' requests. To implement efficient and effective\ndevelopment processes, companies utilize multiple frameworks, methods and\npractices, and combine these into hybrid methods. A common combination contains\na rich management framework to organize and steer projects complemented with a\nnumber of smaller practices providing the development teams with tools to\ncomplete their tasks. In this paper, based on 732 data points collected through\nan international survey, we study the software development process use in\npractice. Our results show that 76.8% of the companies implement hybrid\nmethods. Company size as well as the strategy in devising and evolving hybrid\nmethods affect the suitability of the chosen process to reach company or\nproject goals. Our findings show that companies that combine planned\nimprovement programs with process evolution can increase their process'\nsuitability by up to 5%.\n"}
{"abstract": "  Blind face restoration usually relies on facial priors, such as facial\ngeometry prior or reference prior, to restore realistic and faithful details.\nHowever, very low-quality inputs cannot offer accurate geometric prior while\nhigh-quality references are inaccessible, limiting the applicability in\nreal-world scenarios. In this work, we propose GFP-GAN that leverages rich and\ndiverse priors encapsulated in a pretrained face GAN for blind face\nrestoration. This Generative Facial Prior (GFP) is incorporated into the face\nrestoration process via novel channel-split spatial feature transform layers,\nwhich allow our method to achieve a good balance of realness and fidelity.\nThanks to the powerful generative facial prior and delicate designs, our\nGFP-GAN could jointly restore facial details and enhance colors with just a\nsingle forward pass, while GAN inversion methods require expensive\nimage-specific optimization at inference. Extensive experiments show that our\nmethod achieves superior performance to prior art on both synthetic and\nreal-world datasets.\n"}
{"abstract": "  We present RePOSE, a fast iterative refinement method for 6D object pose\nestimation. Prior methods perform refinement by feeding zoomed-in input and\nrendered RGB images into a CNN and directly regressing an update of a refined\npose. Their runtime is slow due to the computational cost of CNN, which is\nespecially prominent in multiple-object pose refinement. To overcome this\nproblem, RePOSE leverages image rendering for fast feature extraction using a\n3D model with a learnable texture. We call this deep texture rendering, which\nuses a shallow multi-layer perceptron to directly regress a view-invariant\nimage representation of an object. Furthermore, we utilize differentiable\nLevenberg-Marquardt (LM) optimization to refine a pose fast and accurately by\nminimizing the feature-metric error between the input and rendered image\nrepresentations without the need of zooming in. These image representations are\ntrained such that differentiable LM optimization converges within few\niterations. Consequently, RePOSE runs at 92 FPS and achieves state-of-the-art\naccuracy of 51.6% on the Occlusion LineMOD dataset - a 4.1% absolute\nimprovement over the prior art, and comparable result on the YCB-Video dataset\nwith a much faster runtime. The code is available at\nhttps://github.com/sh8/repose.\n"}
{"abstract": "  Logical relations widely exist in human activities. Human use them for making\njudgement and decision according to various conditions, which are embodied in\nthe form of \\emph{if-then} rules. As an important kind of cognitive\nintelligence, it is prerequisite of representing and storing logical relations\nrightly into computer systems so as to make automatic judgement and decision,\nespecially for high-risk domains like medical diagnosis. However, current\nnumeric ANN (Artificial Neural Network) models are good at perceptual\nintelligence such as image recognition while they are not good at cognitive\nintelligence such as logical representation, blocking the further application\nof ANN. To solve it, researchers have tried to design logical ANN models to\nrepresent and store logical relations. Although there are some advances in this\nresearch area, recent works still have disadvantages because the structures of\nthese logical ANN models still don't map more directly with logical relations\nwhich will cause the corresponding logical relations cannot be read out from\ntheir network structures. Therefore, in order to represent logical relations\nmore clearly by the neural network structure and to read out logical relations\nfrom it, this paper proposes a novel logical ANN model by designing the new\nlogical neurons and links in demand of logical representation. Compared with\nthe recent works on logical ANN models, this logical ANN model has more clear\ncorresponding with logical relations using the more direct mapping method\nherein, thus logical relations can be read out following the connection\npatterns of the network structure. Additionally, less neurons are used.\n"}
{"abstract": "  The analyticity properties of the scattering amplitude for a massive scalar\nfield is reviewed in this article where the spacetime geometry is\n$R^{3,1}\\otimes S^1$ i.e. one spatial dimension is compact. Khuri investigated\nthe analyticity of scattering amplitude in a nonrelativitstic potential model\nin three dimensions with an additional compact dimension. He showed that, under\ncertain circumstances, the forward amplitude is nonanalytic. He argued that in\nhigh energy scattering if such a behaviour persists it would be in conflicts\nwith the established results of quantum field theory and LHC might observe such\nbehaviors. We envisage a real scalar massive field in flat Minkowski spacetime\nin five dimensions. The Kaluza-Klein (KK) compactification is implemented on a\ncircle. The resulting four dimensional manifold is $R^{3,1}\\otimes S^1$. The\nLSZ formalism is adopted to study the analyticity of the scattering amplitude.\nThe nonforward dispersion relation is proved. In addition the Jin-Martin bound\nand an analog of the Froissart-Martin bound are proved. A novel proposal is\npresented to look for evidence of the large-radius-compactification scenario. A\nseemingly violation of Froissart-Martin bound at LHC energy might hint that an\nextra dimension might be decompactified. However, we find no evidence for\nviolation of the bound in our analysis.\n"}
{"abstract": "  Graph convolutional networks (GCNs) have received considerable research\nattention recently. Most GCNs learn the node representations in Euclidean\ngeometry, but that could have a high distortion in the case of embedding graphs\nwith scale-free or hierarchical structure. Recently, some GCNs are proposed to\ndeal with this problem in non-Euclidean geometry, e.g., hyperbolic geometry.\nAlthough hyperbolic GCNs achieve promising performance, existing hyperbolic\ngraph operations actually cannot rigorously follow the hyperbolic geometry,\nwhich may limit the ability of hyperbolic geometry and thus hurt the\nperformance of hyperbolic GCNs. In this paper, we propose a novel hyperbolic\nGCN named Lorentzian graph convolutional network (LGCN), which rigorously\nguarantees the learned node features follow the hyperbolic geometry.\nSpecifically, we rebuild the graph operations of hyperbolic GCNs with\nLorentzian version, e.g., the feature transformation and non-linear activation.\nAlso, an elegant neighborhood aggregation method is designed based on the\ncentroid of Lorentzian distance. Moreover, we prove some proposed graph\noperations are equivalent in different types of hyperbolic geometry, which\nfundamentally indicates their correctness. Experiments on six datasets show\nthat LGCN performs better than the state-of-the-art methods. LGCN has lower\ndistortion to learn the representation of tree-likeness graphs compared with\nexisting hyperbolic GCNs. We also find that the performance of some hyperbolic\nGCNs can be improved by simply replacing the graph operations with those we\ndefined in this paper.\n"}
{"abstract": "  We study the potential of future Parity-Violating Electron Scattering (PVES)\ndata to probe the parameter space of the Standard Model Effective Field Theory\n(SMEFT). We contrast the constraints derived from Drell-Yan data taken at the\nLarge Hadron Collider (LHC) with projections of the planned PVES experiments\nSoLID and P2. We show that the PVES data can complement the bounds set by the\nLHC data in the dimension-6 operator space since it probes different\ncombinations of operators than Drell-Yan. The lower characteristic energy of P2\nand SoLID also helps disentangle effects of dimension-6 and dimension-8\noperators that are difficult to resolve with LHC Drell-Yan data alone.\n"}
{"abstract": "  The Egalitarian Allocation (EA) is a well-known profit sharing method for\ncooperative games which attempts to distribute profit among participants in a\nmost equal way while respecting the individual contributions to the obtained\nprofit. Despite having desirable properties from the viewpoint of game theory\nlike being contained in the core, the EA is in general hard to compute. Another\nwell-known method is given by Vickrey Payments (VP). Again, the VP have\ndesirable properties like coalitional rationality, the VP do not fulfill budget\nbalance in general and, thus, are not contained in the core in general. One\nattempt to overcome this shortcoming is to scale down the VP. This can be done\nby a unique scaling factor, or, by individual scaling factors. Now, the\nindividual scaled Vickrey Payments (ISV) are computed by maximizing the scaling\nfactors lexicographically. In this paper we show that the ISV payments are in\nfact identical to a weighted EA, thus exhibiting an interesting connection\nbetween EA and VP. With this, we conclude the uniqueness of the ISV payments\nand provide a polynomial time algorithm for computing a special weighted EA.\n"}
{"abstract": "  A novel theoretical convergence rate estimate for a Balancing Domain\nDecomposition by Constraints algorithm is proven for the solution of the\ncardiac Bidomain model, describing the propagation of the electric impulse in\nthe cardiac tissue. The non-linear system arises from a fully implicit time\ndiscretization and a monolithic solution approach. The preconditioned\nnon-symmetric operator is constructed from the linearized system arising within\nthe Newton-Krylov approach for the solution of the non-linear problem; we\ntheoretically analyze and prove a convergence rate bound for the Generalised\nMinimal Residual iterations' residual. The theory is confirmed by extensive\nparallel numerical tests, widening the class of robust and efficient solvers\nfor implicit time discretizations of the Bidomain model.\n"}
{"abstract": "  We use the superspace formulation of supergravity in eleven and ten\ndimensions to compute fermion couplings on the M2-brane and on D$p$-branes. In\nthis formulation fermionic couplings arise naturally from the\n$\\theta$-expansion of the superfields from which the brane actions are\nconstructed. The techniques we use and develop can in principle be applied to\ndetermine the fermionic couplings to general background fields up to arbitrary\norder. Starting with the superspace formulation of 11-dimensional supergravity,\nwe use a geometric technique known as the `normal coordinate' method to obtain\nthe $\\theta$-expansion of the M2-brane action. We then present a method which\nallows us to translate the knowledge of fermionic couplings on the M2-brane to\nknowledge of such couplings on the D2-brane, and then to any D$p$-brane. This\nmethod is based on superspace generalizations of both the compactification\ntaking 11-dimensional supergravity to type IIA supergravity and the T-duality\nrules connecting the type IIA and type IIB supergravities.\n"}
{"abstract": "  We propose a cascade of neural models that performs sentence classification,\nphrase recognition, and triple extraction to automatically structure the\nscholarly contributions of NLP publications. To identify the most important\ncontribution sentences in a paper, we used a BERT-based classifier with\npositional features (Subtask 1). A BERT-CRF model was used to recognize and\ncharacterize relevant phrases in contribution sentences (Subtask 2). We\ncategorized the triples into several types based on whether and how their\nelements were expressed in text, and addressed each type using separate\nBERT-based classifiers as well as rules (Subtask 3). Our system was officially\nranked second in Phase 1 evaluation and first in both parts of Phase 2\nevaluation. After fixing a submission error in Pharse 1, our approach yields\nthe best results overall. In this paper, in addition to a system description,\nwe also provide further analysis of our results, highlighting its strengths and\nlimitations. We make our code publicly available at\nhttps://github.com/Liu-Hy/nlp-contrib-graph.\n"}
{"abstract": "  Networks of genetic expression can be modelled by hypergraphs with the\nadditional structure that real coefficients are given to each vertex-edge\nincidence. The spectra, i.e. the multiset of the eigenvalues, of such\nhypergraphs, are known to encode structural information of the data. We show\nhow these spectra can be used, in particular, in order to give an estimation of\ncellular redundancy, a novel measure of gene expression heterogeneity, of the\nnetwork. We analyze some simulated and real data sets of gene expression for\nillustrating the new method proposed here.\n"}
{"abstract": "  This PhD thesis is the conclusion of some of my works carried out at the\nCentre de Physique Th\\'eorique, under the supervision of Serge Lazzarini. Two\nsubjects are presented in the manuscript, both aiming at studying the effect of\nthe spin of elementary particles on otherwise known theories.\n  First is a study of the L\\'evy-Leblond-Newton (LLN) equation, which is used\nto describe the evolution of a quantum system with spin 1/2 that is coupled to\nits own gravitational potential. After reviewing some symmetries of non\nrelativistic Quantum Mechanics, and how to geometrize them with the help of\nBargmann structures, we recall what is the L\\'evy-Leblond equation. In this\nchapter, the LLN equation is described in a fully covariant way on Bargmann\nstructures. This covariant formulation then helps to derive the dynamical\nsymmetries of the equation, and conserved quantities. The symmetry group of\nthis equation turns out to be the already known Schr\\\"odinger-Newton group.\n  The second part deals with the trajectory of particles with spin in General\nRelativity. We first highlight Souriau's geometric method to derive the\nMathisson-Papapetrou-Dixon (MPD) equations, and then discuss the different\npossible Spin Supplementary Conditions (SSC) that exist to close the system of\nMPD equations. In this chapter, we assume the Tulczyjew SSC to describe the\ntrajectory of photons, leading to the Souriau-Saturnini equations. We then\npresent some works where we have applied these equations, in a Schwarzschild\nspacetime, and in a spacetime deformed by a gravitational wave. In the first\nwork, we found gravitational birefringence: the trajectory deviates from the\ngeodesic plane, depending on the helicity of the photon and its wavelength. The\nsecond work presented also found birefringence in gravitational interferometry\nexperiments, however the effect is many orders of magnitude lower than what we\ncan detect.\n"}
{"abstract": "  Automation of excavation tasks requires real-time trajectory planning\nsatisfying various constraints. To guarantee both constraint feasibility and\nreal-time trajectory re-plannability, we present an integrated framework for\nreal-time optimization-based trajectory planning of a hydraulic excavator. The\nproposed framework is composed of two main modules: a global planner and a\nreal-time local planner. The global planner computes the entire global\ntrajectory considering excavation volume and energy minimization while the\nlocal counterpart tracks the global trajectory in a receding horizon manner,\nsatisfying dynamic feasibility, physical constraints, and\ndisturbance-awareness. We validate the proposed planning algorithm in a\nsimulation environment where two types of operations are conducted in the\npresence of emulated disturbance from hydraulic friction and soil-bucket\ninteraction: shallow and deep excavation. The optimized global trajectories are\nobtained in an order of a second, which is tracked by the local planner at\nfaster than 30 Hz. To the best of our knowledge, this work presents the first\nreal-time motion planning framework that satisfies constraints of a hydraulic\nexcavator, such as force/torque, power, cylinder displacement, and flow rate\nlimits.\n"}
{"abstract": "  This paper uses Covasim, an agent-based model (ABM) of COVID-19, to evaluate\nand scenarios of epidemic spread in New York State (USA), the UK, and the\nNovosibirsk region (Russia). Epidemiological parameters such as contagiousness\n(virus transmission rate), initial number of infected people, and probability\nof being tested depend on the region's demographic and geographical features,\nthe containment measures introduced; they are calibrated to data about COVID-19\nspread in the region of interest. At the first stage of our study,\nepidemiological data (numbers of people tested, diagnoses, critical cases,\nhospitalizations, and deaths) for each of the mentioned regions were analyzed.\nThe data were characterized in terms of seasonality, stationarity, and\ndependency spaces, and were extrapolated using machine learning techniques to\nspecify unknown epidemiological parameters of the model. At the second stage,\nthe Optuna optimizer based on the tree Parzen estimation method for objective\nfunction minimization was applied to determine the model's unknown parameters.\nThe model was validated with the historical data of 2020. The modeled results\nof COVID-19 spread in New York State, the UK and the Novosibirsk region have\ndemonstrated that if the level of testing and containment measures is\npreserved, the number of positive cases in New York State and the Novosibirsk\nregion will remain the same during March of 2021, while in the UK it will\nreduce. Due to the features of the data for the Novosibirsk region (two\ndatasets are characterized as stationary series with probability of 1), the\nforecast precision is relatively high for the number of hospitalizations, but\nis lower for new cases of COVID-19.\n"}
{"abstract": "  Movement tracks of wild animals frequently fit models of anomalous rather\nthan simple diffusion, mostly reported as ergodic superdiffusive motion\ncombining area-restricted search within a local patch and larger-scale\ncommuting between patches, as highlighted by the L\\'evy walk paradigm. Since\nL\\'evy walks are scale invariant, superdiffusive motion is also expected within\npatches, yet investigation of such local movements has been precluded by the\nlack of accurate high-resolution data at this scale. Here, using rich\nhigh-resolution movement datasets ($>\\! 7 \\times 10^7$ localizations) from 70\nindividuals and continuous-time random walk modeling, we found subdiffusive\nbehavior and ergodicity breaking in the localized movement of three species of\navian predators. Small-scale, within-patch movement was qualitatively\ndifferent, not inferrable and separated from large-scale inter-patch movement\nvia a clear phase transition. Local search is characterized by long\npower-law-distributed waiting times with diverging mean, giving rise to\nergodicity breaking in the form of considerable variability uniquely observed\nat this scale. This implies that wild animal movement is scale specific rather\nthan scale free, with no typical waiting time at the local scale. Placing these\nfindings in the context of the static-ambush to mobile-cruise foraging\ncontinuum, we verify predictions based on the hunting behavior of the study\nspecies and the constraints imposed by their prey.\n"}
{"abstract": "  Data aggregation is an efficient approach to handle the congestion introduced\nby a massive number of machine type devices (MTDs). The aggregators not only\ncollect data but also implement scheduling mechanisms to cope with scarce\nnetwork resources. We use the concept of meta distribution (MD) of the\nsignal-to-interference ratio (SIR) to gain a complete understanding of the\nper-link reliability and describe the performance of two scheduling methods for\ndata aggregation of machine type communication (MTC): random resource\nscheduling (RRS) and channel-aware resource scheduling (CRS). The results show\nthe fraction of users in the network that achieves a target reliability, which\nis an important aspect to consider when designing wireless systems with\nstringent service requirements.\n"}
{"abstract": "  We study three exceptional theta correspondences for p-adic groups, where one\nmember of the dual pair is the exceptional group G2. We prove the Howe duality\nconjecture for these dual pairs and a dichotomy theorem, and determine\nexplicitly the theta lifts of all non-cuspidal representations.\n"}
{"abstract": "  Several attacks have been proposed against Proof-of-Work blockchains, which\nmay increase the attacker's share of mining rewards (e.g., selfish mining,\nblock withholding). A further impact of such attacks, which has not been\nconsidered in prior work, is that decreasing the profitability of mining for\nhonest nodes incentivizes them to stop mining or to leave the attacked chain\nfor a more profitable one. The departure of honest nodes exacerbates the attack\nand may further decrease profitability and incentivize more honest nodes to\nleave. In this paper, we first present an empirical analysis showing that there\nis a statistically significant correlation between the profitability of mining\nand the total hash rate, confirming that miners indeed respond to changing\nprofitability. Second, we present a theoretical analysis showing that selfish\nmining under such elastic hash supply leads either to the collapse of a chain,\ni.e., all honest nodes leaving, or to a stable equilibrium depending on the\nattacker's initial share.\n"}
{"abstract": "  When we use End-to-end automatic speech recognition (E2E-ASR) system for\nreal-world applications, a voice activity detection (VAD) system is usually\nneeded to improve the performance and to reduce the computational cost by\ndiscarding non-speech parts in the audio. This paper presents a novel\nend-to-end (E2E), multi-task learning (MTL) framework that integrates ASR and\nVAD into one model. The proposed system, which we refer to as Long-Running\nSpeech Recognizer (LR-SR), learns ASR and VAD jointly from two seperate\ntask-specific datasets in the training stage. With the assistance of VAD, the\nASR performance improves as its connectionist temporal classification (CTC)\nloss function can leverage the VAD alignment information. In the inference\nstage, the LR-SR system removes non-speech parts at low computational cost and\nrecognizes speech parts with high robustness. Experimental results on segmented\nspeech data show that the proposed MTL framework outperforms the baseline\nsingle-task learning (STL) framework in ASR task. On unsegmented speech data,\nwe find that the LR-SR system outperforms the baseline ASR systems that build\nan extra GMM-based or DNN-based voice activity detector.\n"}
{"abstract": "  We tackle a 3D scene stylization problem - generating stylized images of a\nscene from arbitrary novel views given a set of images of the same scene and a\nreference image of the desired style as inputs. Direct solution of combining\nnovel view synthesis and stylization approaches lead to results that are blurry\nor not consistent across different views. We propose a point cloud-based method\nfor consistent 3D scene stylization. First, we construct the point cloud by\nback-projecting the image features to the 3D space. Second, we develop point\ncloud aggregation modules to gather the style information of the 3D scene, and\nthen modulate the features in the point cloud with a linear transformation\nmatrix. Finally, we project the transformed features to 2D space to obtain the\nnovel views. Experimental results on two diverse datasets of real-world scenes\nvalidate that our method generates consistent stylized novel view synthesis\nresults against other alternative approaches.\n"}
{"abstract": "  Pure Yang-Mills theory in 2 spacetime dimensions shows exact Casimir scaling.\nThus there are infinitely many string tensions, and this has been understood as\na result of non-propagating gluons in 2 dimensions. From ordinary symmetry\nconsiderations, however, this richness in the spectrum of string tensions seems\nmysterious. Conventional wisdom has it that it is the center symmetry that\nclassifies string tensions, but being finite it cannot explain infinitely many\nconfining strings. In this note, we resolve this discrepancy between dynamics\nand kinematics by pointing out the existence of a non-invertible 1-form\nsymmetry, which is able to distinguish Wilson loops in different\nrepresentations. We speculate on possible implications for Yang-Mills theories\nin 3 and 4 dimensions.\n"}
{"abstract": "  Field Programmable Gate Arrays (FPGAs) have the potential to accelerate\nspecific HPC codes. However even with the advent of High Level Synthesis (HLS),\nwhich enables FPGA programmers to write code in C or C++, programming such\ndevices still requires considerable expertise. Much of this is due to the fact\nthat these architectures are founded on dataflow rather than the Von Neumann\nabstraction of CPUs or GPUs. Thus programming FPGAs via imperative languages is\nnot optimal and can result in very significant performance differences between\nthe first and final versions of algorithms on dataflow architectures with the\nsteps in between often not obvious and requiring considerable expertise.\n  In this position paper we argue that languages built upon dataflow principals\nshould be exploited to enable fast by construction codes for FPGAs, and this is\nakin to the programmer adopting the abstraction of developing a bespoke\ndataflow machine specialised for their application. It is our belief that much\ncan be learnt from the generation of dataflow languages that gained popularity\nin the 1970s and 1980s around programming general purpose dataflow machines,\nand we introduce Lucent which is a modern derivative of Lucid, and used as a\nvehicle to explore this hypothesis. The idea behind Lucent is to provide high\nprogrammer productivity and performance for FPGAs by giving developers the most\nsuitable language level abstractions. The focus of Lucent is very much to\nsupport the acceleration of HPC kernels, rather than the embedded electronics\nand circuit level, and we provide a brief overview of the language driven by\nexamples.\n"}
{"abstract": "  Machine learning is playing an increasingly important role in medical image\nanalysis, spawning new advances in the clinical application of neuroimaging.\nThere have been some reviews on machine learning and epilepsy before, and they\nmainly focused on electrophysiological signals such as electroencephalography\n(EEG) and stereo electroencephalography (SEEG), while neglecting the potential\nof neuroimaging in epilepsy research. Neuroimaging has its important advantages\nin confirming the range of the epileptic region, which is essential in\npresurgical evaluation and assessment after surgery. However, it is difficult\nfor EEG to locate the accurate epilepsy lesion region in the brain. In this\nreview, we emphasize the interaction between neuroimaging and machine learning\nin the context of epilepsy diagnosis and prognosis. We start with an overview\nof epilepsy and typical neuroimaging modalities used in epilepsy clinics, MRI,\nDWI, fMRI, and PET. Then, we elaborate two approaches in applying machine\nlearning methods to neuroimaging data: i) the conventional machine learning\napproach combining manual feature engineering and classifiers, ii) the deep\nlearning approach, such as the convolutional neural networks and autoencoders.\nSubsequently, the application of machine learning on epilepsy neuroimaging,\nsuch as segmentation, localization, and lateralization tasks, as well as tasks\ndirectly related to diagnosis and prognosis are looked into in detail. Finally,\nwe discuss the current achievements, challenges, and potential future\ndirections in this field, hoping to pave the way for computer-aided diagnosis\nand prognosis of epilepsy.\n"}
{"abstract": "  In this research-expository paper we recall the basic results of reduction\ntheory of positive definite quadratic forms. Using the result of Ryskov on\nadmissible centerings and the result of Tammela about the determination of a\nMinkowski-reduced form, we prove that the absolute values of coordinates of a\nminimum vector in a six-dimensional Minkowski-reduced basis are less or equal\nto three. To get this little sharpening of the result which can be deduced\nautomatically from Tammela's works we combine some elementary geometric\nreasonings with the mentioned theoretical results.\n"}
{"abstract": "  Event detection (ED) aims at detecting event trigger words in sentences and\nclassifying them into specific event types. In real-world applications, ED\ntypically does not have sufficient labelled data, thus can be formulated as a\nfew-shot learning problem. To tackle the issue of low sample diversity in\nfew-shot ED, we propose a novel knowledge-based few-shot event detection method\nwhich uses a definition-based encoder to introduce external event knowledge as\nthe knowledge prior of event types. Furthermore, as external knowledge\ntypically provides limited and imperfect coverage of event types, we introduce\nan adaptive knowledge-enhanced Bayesian meta-learning method to dynamically\nadjust the knowledge prior of event types. Experiments show our method\nconsistently and substantially outperforms a number of baselines by at least 15\nabsolute F1 points under the same few-shot settings.\n"}
{"abstract": "  Modern data lakes are deeply heterogeneous in the vocabulary that is used to\ndescribe data. We study a problem of disambiguation in data lakes: how can we\ndetermine if a data value occurring more than once in the lake has different\nmeanings and is therefore a homograph? While word and entity disambiguation\nhave been well studied in computational linguistics, data management and data\nscience, we show that data lakes provide a new opportunity for disambiguation\nof data values since they represent a massive network of interconnected values.\nWe investigate to what extent this network can be used to disambiguate values.\nDomainNet uses network-centrality measures on a bipartite graph whose nodes\nrepresent values and attributes to determine, without supervision, if a value\nis a homograph. A thorough experimental evaluation demonstrates that\nstate-of-the-art techniques in domain discovery cannot be re-purposed to\ncompete with our method. Specifically, using a domain discovery method to\nidentify homographs has a precision and a recall of 38% versus 69% with our\nmethod on a synthetic benchmark. By applying a network-centrality measure to\nour graph representation, DomainNet achieves a good separation between\nhomographs and data values with a unique meaning. On a real data lake our\ntop-200 precision is 89%.\n"}
{"abstract": "  For a subset $S$ of nonnegative integers and a vector\n$\\mathbf{a}=(a_1,\\dots,a_k)$ of positive integers, let $V'_S(\\mathbf{a})=\\{\na_1s_1+\\cdots+a_ks_k : s_i\\in S\\} \\setminus \\{0\\}$. For a positive integer $n$,\nlet $\\mathcal T(n)$ be the set of integers greater than or equal to $n$. In\nthis paper, we consider the problem of finding all vectors $\\mathbf{a}$\nsatisfying $V'_S(\\mathbf{a})=\\mathcal T(n)$, when $S$ is the set of\n(generalized) $m$-gonal numbers and $n$ is a positive integer. In particular,\nwe completely resolve the case when $S$ is the set of triangular numbers.\n"}
{"abstract": "  In this paper we prove that the only blowup solutions to the focusing,\nquintic nonlinear Schr{\\\"o}dinger equation with mass equal to the mass of the\nsoliton are rescaled solitons or the pseudoconformal transformation of those\nsolitons.\n"}
{"abstract": "  Smart healthcare systems (SHSs) are providing fast and efficient disease\ntreatment leveraging wireless body sensor networks (WBSNs) and implantable\nmedical devices (IMDs)-based internet of medical things (IoMT). In addition,\nIoMT-based SHSs are enabling automated medication, allowing communication among\nmyriad healthcare sensor devices. However, adversaries can launch various\nattacks on the communication network and the hardware/firmware to introduce\nfalse data or cause data unavailability to the automatic medication system\nendangering the patient's life. In this paper, we propose SHChecker, a novel\nthreat analysis framework that integrates machine learning and formal analysis\ncapabilities to identify potential attacks and corresponding effects on an\nIoMT-based SHS. Our framework can provide us with all potential attack vectors,\neach representing a set of sensor measurements to be altered, for an SHS given\na specific set of attack attributes, allowing us to realize the system's\nresiliency, thus the insight to enhance the robustness of the model. We\nimplement SHChecker on a synthetic and a real dataset, which affirms that our\nframework can reveal potential attack vectors in an IoMT system. This is a\nnovel effort to formally analyze supervised and unsupervised machine learning\nmodels for black-box SHS threat analysis.\n"}
{"abstract": "  This report aims to study adaptive synchronization between a general class of\nhyperchaotic complex-valued systems with unknown parameters, which is motivated\nby extensive application areas of this topic in nonlinear sciences (e.g.,\nsecure communications, encryption techniques, etc.). Based on the complexity of\nhyperchaotic dynamical systems, which may be beneficial in secure\ncommunications, a scheme to achieve adaptive synchronization of a general class\nof hyperchaotic complex-valued systems with unknown parameters is proposed. To\nverify scheme's consistency, the control functions based on adaptive laws of\nparameters are derived analytically, and the related numerical simulations are\nperformed. The complex-valued Rabinovich system describing the parametric\nexcitation of waves in a magneto-active plasma is considered as an interesting\nexample to study this kind of synchronization. A scheme for secure\ncommunication and improving the cryptosystem is proposed; the sketch is\nconstructed to split the message and inject some bit of information signal into\nparameters modulation and the other bit into the transmitter system's states,\nwhich, in turn, complicates decryption task by intruders. Meanwhile, the\ninformation signal can be accurately retrieved at the receiver side by adaptive\ntechniques and decryption function. Different types of encrypted messages are\nconsidered for testing the robustness of the proposed scheme (e.g., plain text\nand gray images with diverse scales of white Gaussian noise).\n"}
{"abstract": "  Inspection and maintenance are two crucial aspects of industrial pipeline\nplants. While robotics has made tremendous progress in the mechanic design of\nin-pipe inspection robots, the autonomous control of such robots is still a big\nopen challenge due to the high number of actuators and the complex manoeuvres\nrequired. To address this problem, we investigate the usage of Deep\nReinforcement Learning for achieving autonomous navigation of in-pipe robots in\npipeline networks with complex topologies. Moreover, we introduce a\nhierarchical policy decomposition based on Hierarchical Reinforcement Learning\nto learn robust high-level navigation skills. We show that the hierarchical\nstructure introduced in the policy is fundamental for solving the navigation\ntask through pipes and necessary for achieving navigation performances superior\nto human-level control.\n"}
{"abstract": "  In this paper, disturbance reconstruction and robust trajectory tracking\ncontrol of biped robots with hybrid dynamics in the port-Hamiltonian form is\ninvestigated. A new type of Hamiltonian function is introduced, which ensures\nthe finite-time stability of the closed-loop system. The proposed control\nsystem consists of two loops: an inner and an outer loop. A fractional\nproportional-integral-derivative filter is used to achieve finite-time\nconvergence for position tracking errors at the outer loop. A fractional-order\nsliding mode controller acts as a centralized controller at the inner-loop,\nensuring the finite-time stability of the velocity tracking error. In this\nloop, the undesired effects of unknown external disturbance and parameter\nuncertainties are compensated using estimators. Two disturbance estimators are\nenvisioned. The former is designed using fractional calculus. The latter is an\nadaptive estimator, and it is constructed using the general dynamic of biped\nrobots. Stability analysis shows that the closed-loop system is finite-time\nstable in both contact-less and impact phases. Simulation studies on two types\nof biped robots (i.e., two-link walker and RABBIT biped robot) demonstrate the\nproposed controller's tracking performance and disturbance rejection\ncapability.\n"}
{"abstract": "  Aim The pharmacokinetic (PK) properties of cyclosporine (CsA) in renal\ntransplant recipients are patient- and time-dependent. Knowledge of this\ntime-related variability is necessary to maintain or achieve CsA target\nexposure. Here, we aimed to identify factors explaining variabilities in CsA PK\nproperties and characterise time-dependent clearance (CL/F) by performing a\ncomprehensive analysis of CsA PK factors using population PK (popPK) modelling\nof long-term follow-up data from our institution. Methods In total, 3,674\nwhole-blood CsA concentrations from 183 patients who underwent initial renal\ntransplantation were analysed using nonlinear mixed-effects modelling. The\neffects of potential covariates were selected according to a previous report\nand well-accepted theoretical mechanisms. Model-informed individualised\ntherapeutic regimens were also conducted. Results A two-compartment model\nadequately described the data and the estimated mean CsA CL/F was 32.6 L h-1\n(5%). Allometrically scaled body size, haematocrit (HCT) level, CGC haplotype\ncarrier status, and postoperative time may contribute to CsA PK variability.\nThe CsA bioavailability in patients receiving a prednisolone dose (PD) of 80 mg\nwas 20.6% lower than that in patients receiving 20 mg. A significant decrease\n(52.6%) in CL/F was observed as the HCT increased from 10.5% to 60.5%. The CL/F\nof the non-CGC haplotype carrier was 14.4% lower than that of the CGC haplotype\ncarrier at 3 months post operation. CsA dose adjustments should be considered\nin different postoperative periods. Conclusions By monitoring body size, HCT,\nPD, and CGC haplotype, changes in CsA CL/F over time could be predicted. Such\ninformation could be used to optimise CsA therapy.\n"}
{"abstract": "  We develop a \"hybrid\" model of the proton inelastic structure functions\napplicable in a wide region of invariant mass of produced states $W$ and\nmomentum transfer $Q$ including deep inelastic scattering (DIS), nucleon\nresonance production as well as the region close to inelastic threshold. DIS is\ndescribed in terms of the parton distributions together with higher-twist\ncorrections from available global QCD fit. The resonant part is addressed in\nterms of the Breit-Wiegner contributions from five states: $\\Delta(1232)$\nresonance, $N(1440)$ Roper resonance, and three effective resonances describing\nthe second and third resonance regions. The couplings of the nucleon resonances\nto photon are described in terms of helicity amplitudes. The nonresonant\nbackground is addressed in terms of a smooth extrapolation of DIS structure\nfunctions to low-$W$ and low-$Q$ values with the proper behavior at the real\nphoton limit $Q^2=0$ as well as near inelastic threshold. We independently\ntreat the transverse, $F_T$, and the longitudinal, $F_L$, structure function\nand fix the model parameters from a global analysis of the world hydrogen\nelectroproduction and photoproduction inclusive cross-section data. We\ndemonstrate a very good performance of the model by comparing our predictions\nwith data on differential cross sections and the structure functions $F_2$ and\n$R=F_L/F_T$.\n"}
{"abstract": "  Recently, end-to-end Korean singing voice systems have been designed to\ngenerate realistic singing voices. However, these systems still suffer from a\nlack of robustness in terms of pronunciation accuracy. In this paper, we\npropose N-Singer, a non-autoregressive Korean singing voice system, to\nsynthesize accurate and pronounced Korean singing voices in parallel. N-Singer\nconsists of a Transformer-based mel-generator, a convolutional network-based\npostnet, and voicing-aware discriminators. It can contribute in the following\nways. First, for accurate pronunciation, N-Singer separately models linguistic\nand pitch information without other acoustic features. Second, to achieve\nimproved mel-spectrograms, N-Singer uses a combination of Transformer-based\nmodules and convolutional network-based modules. Third, in adversarial\ntraining, voicing-aware conditional discriminators are used to capture the\nharmonic features of voiced segments and noise components of unvoiced segments.\nThe experimental results prove that N-Singer can synthesize a natural singing\nvoice in parallel with a more accurate pronunciation than the baseline model.\n"}
{"abstract": "  We consider the quantum dynamics of $N$ interacting fermions in the large $N$\nlimit. The particles in the system interact with each other via repulsive\ninteraction that is regularized Coulomb potential with a polynomial cutoff with\nrespect to $N$.From the quantum system, we derive the Vlasov-Poisson system by\nsimultaneously estimating the semiclassical and mean-field residues in terms of\nthe Husimi measure.\n"}
{"abstract": "  The solution of linear systems of equations is a very frequent operation and\nthus important in many fields. The complexity using classical methods increases\nlinearly with the size of equations. The HHL algorithm proposed by Harrow et\nal. achieves exponential acceleration compared with the best classical\nalgorithm. However, it has a relatively high demand for qubit resources and the\nsolution $\\left| x \\right\\rangle $ is in a normalized form. Assuming that the\neigenvalues of the coefficient matrix of the linear systems of equations can be\nrepresented perfectly by finite binary number strings, three hybrid iterative\nphase estimation algorithms (HIPEA) are designed based on the iterative phase\nestimation algorithm in this paper. The complexity is transferred to the\nmeasurement operation in an iterative way, and thus the demand of qubit\nresources is reduced in our hybrid algorithms. Moreover, the solution is stored\nin a classical register instead of a quantum register, so the exact\nunnormalized solution can be obtained. The required qubit resources in the\nthree HIPEA algorithms are different. HIPEA-1 only needs one single ancillary\nqubit. The number of ancillary qubits in HIPEA-2 is equal to the number of\nnondegenerate eigenvalues of the coefficient matrix of linear systems of\nequations. HIPEA-3 is designed with a flexible number of ancillary qubits. The\nHIPEA algorithms proposed in this paper broadens the application range of\nquantum computation in solving linear systems of equations by avoiding the\nproblem that quantum programs may not be used to solve linear systems of\nequations due to the lack of qubit resources.\n"}
{"abstract": "  In this work, we analyse the links between ghost penalty stabilisation and\naggregation-based discrete extension operators for the numerical approximation\nof elliptic partial differential equations on unfitted meshes. We explore the\nbehavior of ghost penalty methods in the limit as the penalty parameter goes to\ninfinity, which returns a strong version of these methods. We observe that\nthese methods suffer locking in that limit. On the contrary, aggregated finite\nelement spaces are locking-free because they can be expressed as an extension\noperator from well-posed to ill-posed degrees of freedom. Next, we propose\nnovel ghost penalty methods that penalise the distance between the solution and\nits aggregation-based discrete extension. These methods are locking-free and\nconverge to aggregated finite element methods in the infinite penalty parameter\nlimit. We include an exhaustive set of numerical experiments in which we\ncompare weak (ghost penalty) and strong (aggregated finite elements) schemes in\nterms of error quantities, condition numbers and sensitivity with respect to\npenalty coefficients on different geometries, intersection locations and mesh\ntopologies.\n"}
{"abstract": "  Edge & Fog computing have received considerable attention as promising\ncandidates for the evolution of robotic systems. In this letter, we propose\nCOTORRA, an Edge & Fog driven robotic testbed that combines context information\nwith robot sensor data to validate innovative concepts for robotic systems\nprior to being applied in a production environment. In lab/university, we\nestablished COTORRA as an easy applicable and modular testbed on top of\nheterogeneous network infrastructure. COTORRA is open for pluggable robotic\napplications. To verify its feasibility and assess its performance, we ran set\nof experiments that show how autonomous navigation applications can achieve\ntarget latencies bellow 15ms or perform an inter-domain (DLT) federation within\n19 seconds.\n"}
{"abstract": "  Contact tracing is the techno-choice of reference to address the COVID-19\npandemic. Many of the current approaches have severe privacy and security\nissues and fail to offer a sustainable contact tracing infrastructure. We\naddress these issues introducing an innovative, privacy-preserving,\nsustainable, and experimentally tested architecture that leverages batteryless\nBLE beacons.\n"}
{"abstract": "  Deep learning implementations on CPUs (Central Processing Units) are gaining\nmore traction. Enhanced AI capabilities on commodity x86 architectures are\ncommercially appealing due to the reuse of existing hardware and virtualization\nease. A notable work in this direction is the SLIDE system. SLIDE is a C++\nimplementation of a sparse hash table based back-propagation, which was shown\nto be significantly faster than GPUs in training hundreds of million parameter\nneural models. In this paper, we argue that SLIDE's current implementation is\nsub-optimal and does not exploit several opportunities available in modern\nCPUs. In particular, we show how SLIDE's computations allow for a unique\npossibility of vectorization via AVX (Advanced Vector Extensions)-512.\nFurthermore, we highlight opportunities for different kinds of memory\noptimization and quantizations. Combining all of them, we obtain up to 7x\nspeedup in the computations on the same hardware. Our experiments are focused\non large (hundreds of millions of parameters) recommendation and NLP models.\nOur work highlights several novel perspectives and opportunities for\nimplementing randomized algorithms for deep learning on modern CPUs. We provide\nthe code and benchmark scripts at https://github.com/RUSH-LAB/SLIDE\n"}
{"abstract": "  The principal limitation in many areas of astronomy, especially for directly\nimaging exoplanets, arises from instability in the point spread function (PSF)\ndelivered by the telescope and instrument. To understand the transfer function,\nit is often necessary to infer a set of optical aberrations given only the\nintensity distribution on the sensor - the problem of phase retrieval. This can\nbe important for post-processing of existing data, or for the design of optical\nphase masks to engineer PSFs optimized to achieve high contrast, angular\nresolution, or astrometric stability. By exploiting newly efficient and\nflexible technology for automatic differentiation, which in recent years has\nundergone rapid development driven by machine learning, we can perform both\nphase retrieval and design in a way that is systematic, user-friendly, fast,\nand effective. By using modern gradient descent techniques, this converges\nefficiently and is easily extended to incorporate constraints and\nregularization. We illustrate the wide-ranging potential for this approach\nusing our new package, Morphine. Challenging applications performed with this\ncode include precise phase retrieval for both discrete and continuous phase\ndistributions, even where information has been censored such as\nheavily-saturated sensor data. We also show that the same algorithms can\noptimize continuous or binary phase masks that are competitive with existing\nbest solutions for two example problems: an Apodizing Phase Plate (APP)\ncoronagraph for exoplanet direct imaging, and a diffractive pupil for\nnarrow-angle astrometry. The Morphine source code and examples are available\nopen-source, with a similar interface to the popular physical optics package\nPoppy.\n"}
{"abstract": "  Codebooks with small maximum cross-correlation amplitudes are used to\ndistinguish the signals from different users in CDMA communication systems. In\nthis paper, we first study the Jacobi sums over Galois rings of arbitrary\ncharacteristics and completely determine their absolute values, which extends\nthe work in [34], where the Jacobi sums over Galois rings with characteristics\nof a square of a prime number were discussed. Then, the deterministic\nconstruction of codebooks based on the Jacobi sums over Galois rings of\narbitrary characteristics is presented, which produces asymptotically optimal\ncodebooks with respect to the Welch bound. In addition, the parameters of the\ncodebooks provided in this paper are new.\n"}
{"abstract": "  We report a $\\gamma$-ray enhancement event detected from Tycho's supernova\nremnant (SNR), the outcome of a type Ia supernova explosion that occurred in\nyear 1572. The event lasted for 1.5 years and showed a factor of 3.6 flux\nincrease mainly in the energy range of 4--100 GeV. While several young SNRs\n(including Tycho's SNR) were previously found to show peculiar X-ray structures\nwith flux variations in one- or several-year timescales, such an event at\n$\\gamma$-ray energies is for the first time seen. The hard $\\gamma$-ray\nemission and year-long timescale of the event necessitate a synchrotron\nradiation process, although the required conditions are either ultra-high\nenergies for the electrons in the process, upto $\\sim$10 PeV (well above the\ncosmic-ray \"knee\" energy), or high inhomogeneity of the magnetic field in the\nSNR. This event in Tycho's SNR is likely analogous to the $\\gamma$-ray flares\nobserved in the Crab nebula, the comparably short timescales of them both\nrequiring a synchrotron process, and similar magnetohydrodynamic processes such\nas magnetic reconnection would be at work as well in the SNR to accelerate\nparticles to ultra-relativistic energies. The event and its implications thus\nreveal the more complicated side of the physical processes that can occur in\nyoung SNRs.\n"}
{"abstract": "  We prove that the only non-Archimedean strictly convex spaces are the zero\nspace and the one-dimensional linear space over $\\, \\mathbb{Z}/3\\mathbb{Z}$,\nwith any of its trivial norms.\n"}
{"abstract": "  Early identification of college dropouts can provide tremendous value for\nimproving student success and institutional effectiveness, and predictive\nanalytics are increasingly used for this purpose. However, ethical concerns\nhave emerged about whether including protected attributes in the prediction\nmodels discriminates against underrepresented student groups and exacerbates\nexisting inequities. We examine this issue in the context of a large U.S.\nresearch university with both residential and fully online degree-seeking\nstudents. Based on comprehensive institutional records for this entire student\npopulation across multiple years, we build machine learning models to predict\nstudent dropout after one academic year of study, and compare the overall\nperformance and fairness of model predictions with or without four protected\nattributes (gender, URM, first-generation student, and high financial need). We\nfind that including protected attributes does not impact the overall prediction\nperformance and it only marginally improves algorithmic fairness of\npredictions. While these findings suggest that including protected attributes\nis preferred, our analysis also offers guidance on how to evaluate the impact\nin a local context, where institutional stakeholders seek to leverage\npredictive analytics to support student success.\n"}
{"abstract": "  We introduce the NCA and NCA-Markov dynamical maps for open quantum systems,\nwhich go beyond weak-coupling master equations replacing the Born approximation\nwith a non-crossing approximation (NCA). We apply these NCA maps on a prototype\nmodel of open quantum systems, the Ohmic spin-boson model at zero temperature,\nshowing that they can qualitatively reproduce its non-perturbative behaviour,\nincluding its crossover from a coherent to an incoherent dynamics and its\nlocalization quantum phase transition, which are entirely missed by\nperturbative master equations. Remarkably, the time-local NCA-Markov map\nqualitatively captures this physics entirely, opening up the possibility to\naddress strong-coupling effects at a numerical cost comparable to standard\nMarkovian master equations.\n"}
{"abstract": "  Learning dynamics governed by differential equations is crucial for\npredicting and controlling the systems in science and engineering. Neural\nOrdinary Differential Equation (NODE), a deep learning model integrated with\ndifferential equations, is popular in learning dynamics recently due to its\nrobustness to irregular samples and its flexibility to high-dimensional input.\nHowever, the training of NODE is sensitive to the precision of the numerical\nsolver, which makes the convergence of NODE unstable, especially for\nill-conditioned dynamical systems. In this paper, to reduce the reliance on the\nnumerical solver, we propose to enhance the supervised signal in the training\nof NODE. Specifically, we pre-train a neural differential operator (NDO) to\noutput an estimation of the derivatives to serve as an additional supervised\nsignal. The NDO is pre-trained on a class of basis functions and learns the\nmapping between the trajectory samples of these functions to their derivatives.\nTo leverage both the trajectory signal and the estimated derivatives from NDO,\nwe propose an algorithm called NDO-NODE, in which the loss function contains\ntwo terms: the fitness on the true trajectory samples and the fitness on the\nestimated derivatives that are outputted by the pre-trained NDO. Experiments on\nvarious kinds of dynamics show that our proposed NDO-NODE can consistently\nimprove the forecasting accuracy with one pre-trained NDO. Especially for the\nstiff ODEs, we observe that NDO-NODE can capture the transitions in the\ndynamics more accurately compared with other regularization methods.\n"}
{"abstract": "  Semi-supervised methods of anomaly detection have seen substantial\nadvancement in recent years. Of particular interest are applications of such\nmethods to diverse, real-world anomaly detection problems where anomalous\nvariations can vary from the visually obvious to the very subtle. In this work,\nwe propose a novel fine-grained VAE-GAN architecture trained in a\nsemi-supervised manner in order to detect both visually distinct and subtle\nanomalies. With the use of a residually connected dual-feature extractor, a\nfine-grained discriminator and a perceptual loss function, we are able to\ndetect subtle, low inter-class (anomaly vs. normal) variant anomalies with\ngreater detection capability and smaller margins of deviation in AUC value\nduring inference compared to prior work whilst also remaining time-efficient\nduring inference. We achieve state of-the-art anomaly detection results when\ncompared extensively with prior semi-supervised approaches across a multitude\nof anomaly detection benchmark tasks including trivial leave-one out tasks\n(CIFAR-10 - AUPRCavg: 0.91; MNIST - AUPRCavg: 0.90) in addition to challenging\nreal-world anomaly detection tasks (plant leaf disease - AUC: 0.776; threat\nitem X-ray - AUC: 0.51), video frame-level anomaly detection (UCSDPed1 - AUC:\n0.95) and high frequency texture with object anomalous defect detection (MVTEC\n- AUCavg: 0.83).\n"}
{"abstract": "  In this paper questions of the width in verbal subgroups of $A_n$ is\nconsidered. Verbal width by set of squares of alternating group is found, it is\nhappened equals to 2. The criterion of squareness in $A_n$ is presented. The\nnecessary and sufficient conditions when an element of alternating group can be\npresented as a squares of one element are also found by us. Some necessary\nconditions to an element $g\\in A_n$ being the square in $A_n$ are investigated.\n"}
{"abstract": "  This paper introduces the Feasibility Governor (FG): an add-on unit that\nenlarges the region of attraction of Model Predictive Control by manipulating\nthe reference to ensure that the underlying optimal control problem remains\nfeasible. The FG is developed for linear systems subject to polyhedral state\nand input constraints. Offline computations using polyhedral projection\nalgorithms are used to construct the feasibility set. Online implementation\nrelies on the solution of a convex quadratic program that guarantees recursive\nfeasibility. The closed-loop system is shown to satisfy constraints, achieve\nasymptotic stability, and exhibit zero-offset tracking.\n"}
{"abstract": "  We present near-IR photometry and spectroscopy of 30 extremely luminous radio\nand mid-IR selected galaxies. With bolometric luminosities exceeding\n$\\sim10^{13}$ $\\rm{L_{\\odot}}$ and redshifts ranging from $z = 0.880-2.853$, we\nuse VLT instruments X-shooter and ISAAC to investigate this unique population\nof galaxies. Broad multi-component emission lines are detected in 18 galaxies\nand we measure the near-IR lines $\\rm{H\\rm{\\beta}}$,\n$\\text{[OIII]}\\rm{\\lambda}\\rm{\\lambda}4959,5007$ and $\\rm{H\\rm{\\alpha}}$ in\nsix, 15 and 13 galaxies respectively, with 10 $\\rm{Ly\\alpha}$ and five CIV\nlines additionally detected in the UVB arm. We use the broad\n$\\text{[OIII]}\\rm{\\lambda}5007$ emission lines as a proxy for the bolometric\nAGN luminosity, and derive lower limits to supermassive black hole masses of\n$10^{7.9}$-$10^{9.4}$ $\\text{M}_{\\odot}$ with expectations of corresponding\nhost masses of $10^{10.4}$-$10^{12.0}$ $\\text{M}_{\\odot}$. We measure\n$\\rm{\\lambda}_{Edd}$ > 1 for eight of these sources at a $2\\sigma$\nsignificance. Near-IR photometry and SED fitting are used to compare stellar\nmasses directly. We detect both Balmer lines in five galaxies and use these to\ninfer a mean visual extinction of $A_{V}$ = 2.68 mag. Due to non-detections and\nuncertainties in our $\\rm{H\\rm{\\beta}}$ emission line measurements, we simulate\na broad $\\rm{H\\rm{\\beta}}$ line of FWHM = 1480 $\\rm{kms^{-1}}$ to estimate\nextinction for all sources with measured $\\rm{H\\rm{\\alpha}}$ emission. We then\nuse this to infer a mean $A_{V}=3.62$ mag, demonstrating the highly-obscured\nnature of these galaxies, with the consequence of increasing our estimates of\nblack-hole masses by an 0.5 orders of magnitude in the most extreme and\nobscured cases.\n"}
{"abstract": "  The Kerr geometry admits the Carter symmetry, which ensures that the geodesic\nequations are integrable. It is shown that gravitational waveforms associated\nwith extreme-mass-ratio inspirals involving a non-integrable compact object\ndisplay `glitch' phenomena, where the frequencies of gravitational waves\nincrease abruptly, when the orbit crosses certain spacetime regions known as\nBirkhoff islands. The presence or absence of these features in data from\nupcoming space-borne detectors will therefore allow not only for tests of\ngeneral relativity, but also of fundamental spacetime symmetries.\n"}
{"abstract": "  The quantum vacuum shows an intensity-dependent nonlinear index of\nrefraction. Consequently, we may consider vacuum as a non-linear medium\nenabling, at sufficiently high field strengths, a direct interaction of light\nwith light. In this context, we advocate an experiment to measure\nlight-by-light scattering by the means of self-focussing of an x-ray beam in\nvacuum. Although hard to measure, we argue that for laser beams consisting of\nhighly energetic photons the beams' power distribution carries a distinctive\nenough signature of self-focussing such that an observation at modern\nfree-electron laser facilities appears possible. Self-focussing also imposes a\nlimit on the most powerful x-ray beams that may propagate in vacuum being on\nthe order of $10$ PW for $10$ keV x-rays.\n"}
{"abstract": "  Diffusive molecular communications (DiMC) have recently gained attention as a\ncandidate for nano- to micro- and macro-scale communications due to its\nsimplicity and energy efficiency. As signal propagation is solely enabled by\nBrownian motion mechanics, DiMC faces severe inter-symbol interference (ISI),\nwhich limits reliable and high data-rate communications. Herein, recent\nliterature on DiMC performance enhancement strategies is surveyed; key research\ndirections are identified. Signaling design and associated design constraints\nare presented. Classical and novel transceiver designs are reviewed with an\nemphasis on methods for ISI mitigation and performance-complexity tradeoffs.\nKey parameter estimation strategies such as synchronization and channel\nestimation are considered in conjunction with asynchronous and timing error\nrobust receiver methods. Finally, source and channel coding in the context of\nDiMC is presented.\n"}
{"abstract": "  The problem of Multi-Agent Path Finding (MAPF) calls for finding a set of\nconflict-free paths for a fleet of agents operating in a given environment.\nArguably, the state-of-the-art approach to computing optimal solutions is\nConflict-Based Search (CBS). In this work we revisit the complexity analysis of\nCBS to provide tighter bounds on the algorithm's run-time in the worst-case.\nOur analysis paves the way to better pinpoint the parameters that govern (in\nthe worst case) the algorithm's computational complexity.\n  Our analysis is based on two complementary approaches: In the first approach\nwe bound the run-time using the size of a Multi-valued Decision Diagram (MDD)\n-- a layered graph which compactly contains all possible single-agent paths\nbetween two given vertices for a specific path length.\n  In the second approach we express the running time by a novel recurrence\nrelation which bounds the algorithm's complexity. We use generating\nfunctions-based analysis in order to tightly bound the recurrence.\n  Using these technique we provide several new upper-bounds on CBS's\ncomplexity. The results allow us to improve the existing bound on the running\ntime of CBS for many cases. For example, on a set of common benchmarks we\nimprove the upper-bound by a factor of at least $2^{10^{7}}$.\n"}
{"abstract": "  In this paper we give some results concerning Frechet differentiable mappings\nbetween domains in normed spaces with controlled growth. The results are mainly\nmotivated by Pavlovic's equality for the Bloch semi-norm of continuously\ndifferentiable mappings in the Bloch class on the unit ball of the Euclidean\nspace as well as the very recent Jocic's generalization of this result.\n"}
{"abstract": "  An energetic muon beam is an attractive key to unlock new physics beyond the\nStandard Model: the lepton flavor violation or the anomalous magnetic moment,\nand also is a competitive candidate for the expected neutrino factory. Lots of\nthe muon scientific applications are limited by low flux cosmic-ray muons, low\nenergy muon sources or extremely expensive muon accelerators. An prompt\nacceleration of the low-energy muon beam is found in the beam-driven plasma\nwakefield up to $\\mathrm{TV/m}$. The muon beam is accelerated from\n$275\\mathrm{MeV}$ to more than $10\\mathrm{GeV}$ within $22.5\\mathrm{ps}$.\nChoosing the injection time of the muon beam in a proper range, the\nlongitudinal spatial distribution and the energy distribution of the\naccelerated muon beam are compressed. The efficiency of the energy transfer\nfrom the driven electron beam to the muon beam can reach $20\\%$. The prompt\nacceleration scheme is a promising avenue to bring the expected neutrino\nfactory and the muon collider into reality and to catch new physics beyond the\nStandard Model.\n"}
{"abstract": "  Using the generalized extreme value theory to characterize tail\ndistributions, we address liquidation, leverage, and optimal margins for\nbitcoin long and short futures positions. The empirical analysis of perpetual\nbitcoin futures on BitMEX shows that (1) daily forced liquidations to out-\nstanding futures are substantial at 3.51%, and 1.89% for long and short; (2)\ninvestors got forced liquidation do trade aggressively with average leverage of\n60X; and (3) exchanges should elevate current 1% margin requirement to 33% (3X\nleverage) for long and 20% (5X leverage) for short to reduce the daily margin\ncall probability to 1%. Our results further suggest normality assumption on\nreturn significantly underestimates optimal margins. Policy implications are\nalso discussed.\n"}
{"abstract": "  The power management techniques to extend battery lifespan is becoming\nincreasingly important due to longer user applications' running time in mobile\ndevices. Even when users do not use any applications, battery lifespan\ndecreases continually. It occurs because of service daemons of mobile platform\nand network-based data synchronization operations. In this paper, we propose a\nnew power management system that recognizes the idle time of the device to\nreduce the battery consumption of mobile devices.\n"}
{"abstract": "  Cybersecurity risk analysis plays an essential role in supporting\norganizations make effective decision about how to manage and control\ncybersecurity risk. Cybersecurity risk is a function of the interplay between\nthe defender, i.e., the organisation, and the attacker: decisions and actions\nmade by the defender second guess the decisions and actions taken by the\nattacker and vice versa. Insight into this game between these two agents\nprovides a means for the defender to identify and make optimal decisions. To\ndate, the adversarial risk analysis framework has provided a\ndecision-analytical approach to solve such game problems in the presence of\nuncertainty and uses Monte Carlo simulation to calculate and identify optimal\ndecisions. We propose an alternative framework to construct and solve a serial\nof sequential Defend-Attack models, that incorporates the adversarial risk\nanalysis approach, but uses a new class of influence diagrams algorithm, called\nhybrid Bayesian network inference, to identify optimal decision strategies.\nCompared to Monte Carlo simulation the proposed hybrid Bayesian network\ninference is more versatile because it provides an automated way to compute\nhybrid Defend-Attack models and extends their use to involve mixtures of\ncontinuous and discrete variables, of any kind. More importantly, the hybrid\nBayesian network approach is novel in that it supports dynamic decision making\nwhereby new real-time observations can update the Defend-Attack model in\npractice. We also extend the Defend-Attack model to support cases involving\nextra variables and longer decision sequence. Examples are presented,\nillustrating how the proposed framework can be adjusted for more complicated\nscenarios, including dynamic decision making.\n"}
{"abstract": "  Observations indicate that a continuous supply of gas is needed to maintain\nobserved star formation rates in large, disky galaxies. To fuel star formation,\ngas must reach the inner regions of such galaxies. Despite its crucial\nimportance for galaxy evolution, how and where gas joins galaxies is poorly\nconstrained observationally and is rarely explored in fully cosmological\nsimulations. To investigate gas accretion in the vicinity of galaxies, we\nanalyze the FIRE-2 cosmological zoom-in simulations for 4 Milky Way mass\ngalaxies (M_halo ~ 10E12 solar masses), focusing on simulations with cosmic ray\nphysics. We find that at z~0, gas approaches the disk with angular momentum\nsimilar to the gaseous disk edge and low radial velocities, piling-up near the\nedge and settling into full rotational support. Accreting gas moves\npredominantly parallel to the disk with small but nonzero vertical velocity\ncomponents, and joins the disk largely in the outskirts as opposed to \"raining\"\ndown onto the disk. Once in the disk, gas trajectories are complex, being\ndominated by spiral arm induced oscillations and feedback. However, time and\nazimuthal averages show clear but slow net radial infall with transport speeds\nof 1-3 km/s and net mass fluxes through the disk on the order of one solar mass\nper year, comparable to the star formation rates of the galaxies and decreasing\ntowards galactic center as gas is sunk into star formation. These rates are\nslightly higher in simulations without cosmic rays (1-7 km/s, ~4-5 solar masses\nper year). We find overall consistency of our results with observational\nconstraints and discuss prospects of future observations of gas flows in and\naround galaxies.\n"}
{"abstract": "  In this paper we present a decoding algorithm for algebraic geometry codes\nwith error-correcting capacity beyond half the designed distance of the code.\nThis algorithm comes as a fusion of the Power Error Locating Pairs algorithm\nfor algebraic geometry codes and the technique used by Ehrhard in order to\ncorrect these codes up to half the designed distance. The decoding radius of\nthis algorithm reaches that of Sudan algorithm, without any penalty given by\nthe genus of the curve.\n"}
{"abstract": "  We define a probability measure preserving and r-discrete groupoid that is\nassociated to every invariant point process on a locally compact and second\ncountable group. This groupoid governs certain factor processes of the point\nprocess, in particular the existence of Cayley factor graphs. With this method\nwe are able to show that point processes on amenable groups admit all (and only\nadmit) Cayley factor graphs of amenable groups, and that the Poisson point\nprocess on groups with Kazhdan's Property (T) admits no Cayley factor graphs.\nThis gives examples of pmp countable Borel equivalence relations that cannot be\ngenerated by any free action of a countable group.\n"}
{"abstract": "  Let $G$ be a connected, complex reductive group. In this paper, we classify\n$G\\times G$-equivariant normal $\\mathbb R$-test configurations of a polarized\n$G$-compactification. Then for $\\mathbb Q$-Fano $G$-compactifications, we\nexpress the H-invariant of its equivariant normal $\\mathbb R$-test\nconfigurations in terms of the combinatory data. Based on \\cite{Han-Li}, we\ncompute the semistable limit of a K-unstable Fano $G$-compactification. As an\napplication, we show that for the two K-unstable Fano $SO_4(\\mathbb\nC)$-compactifications, the corresponding semistable limits are indeed the limit\nspaces of the normalized K\\\"ahler-Ricci flow.\n"}
{"abstract": "  In this paper Modified gravity is studied over the linearized metric\nperturbation in post-Minkowskian theory. This is a different aspect for two\nbody dynamics without taking usual self force originated from the radiative\nbackscattering of gravitational waves. The new multiplicative approach to\ndetermine the background metric of curved space-time for two different massive\nsources is computed in post-Newtonian theory. The theoretical result is checked\nfor galactic flat rotation curve and is very good agreement with solar\nrotational speed.\n"}
{"abstract": "  The LHC may produce light, weakly-interacting particles that decay to dark\nmatter, creating an intense and highly collimated beam of dark matter particles\nin the far-forward direction. We investigate the prospects for detecting this\ndark matter in two far-forward detectors proposed for a future Forward Physics\nFacility: FASER$\\nu$2, a 10-tonne emulsion detector, and FLArE, a 10- to\n100-tonne LArTPC. We focus here on nuclear scattering, including elastic\nscattering, resonant pion production, and deep inelastic scattering, and devise\ncuts that efficiently remove the neutrino-induced background. In the\ninvisibly-decaying dark photon scenario, DM-nuclear scattering probes new\nparameter space for dark matter masses 5 MeV $\\lesssim m_{\\chi} \\lesssim$ 500\nMeV. When combined with the DM-electron scattering studied previously,\nFASER$\\nu$2 and FLArE will be able to discover dark matter in a large swath of\nthe cosmologically-favored parameter space with MeV $\\lesssim m_{\\chi} \\lesssim\n$ GeV.\n"}
{"abstract": "  We demonstrate an electro-optic wide-field method to enable fluorescence\nlifetime microscopy (FLIM) with high throughput and single-molecule\nsensitivity. Resonantly driven Pockels cells are used to efficiently gate\nimages at 39 MHz, allowing fluorescence lifetime to be captured on standard\ncamera sensors. Lifetime imaging of single molecules is enabled in wide-field\nwith exposure times of less than 100 milliseconds. This capability allows\ncombination of wide-field FLIM with single-molecule super-resolution\nlocalization microscopy. Fast single-molecule dynamics such as FRET and\nmolecular binding events are captured from wide-field images without prior\nspatial knowledge. A lifetime sensitivity of 1.9 times the photon shot-noise\nlimit is achieved, and high throughput is shown by acquiring wide-field FLIM\nimages with millisecond exposure and $>10^8$ photons per frame. Resonant\nelectro-optic FLIM allows lifetime contrast in any wide-field microscopy\nmethod.\n"}
{"abstract": "  Scaling of the behavior of a nanodevice means that the device function\n(selectivity) is a unique smooth and monotonic function of a scaling parameter\nthat is an appropriate combination of the system's parameters. For the\nuniformly charged cylindrical nanopore studied here these parameters are the\nelectrolyte concentration, $c$, voltage, $U$, the radius and the length of the\nnanopore, $R$ and $H$, and the surface charge density on the nanopore's\nsurface, $\\sigma$. Due to the non-linear dependence of selectivites on these\nparameters, scaling can only be applied in certain limits. We show that the\nDukhin number, $\\mathrm{Du}=|\\sigma|/eRc\\sim\n|\\sigma|\\lambda_{\\mathrm{D}}^{2}/eR$ ($\\lambda_{\\mathrm{D}}$ is the Debye\nlength), is an appropriate scaling parameter in the nanotube limit\n($H\\rightarrow\\infty$). Decreasing the length of the nanopore, namely,\napproaching the nanohole limit ($H\\rightarrow 0$), an alternative scaling\nparameter has been obtained that contains the pore length and is called the\nmodified Dukhin number: $\\mathrm{mDu}\\sim \\mathrm{Du}\\,\nH/\\lambda_{\\mathrm{D}}\\sim |\\sigma|\\lambda_{\\mathrm{D}}H/eR$. We found that the\nreason of non-linearity is that the double layers accumulating at the pore wall\nin the radial dimension correlate with the double layers accumulating at the\nentrances of the pore near the membrane on the two sides. Our modeling study\nusing the Local Equilibrium Monte Carlo method and the Poisson-Nernst-Planck\ntheory provides concentration, flux, and selectivity profiles, that show\nwhether the surface or the volume conduction dominates in a given region of the\nnanopore for a given combination of the variables. We propose that the\ninflection point of the scaling curve may be used to characterize the\ntransition point between the surface and volume conductions.\n"}
{"abstract": "  We provide a new simple and transparent proof of the version of Kummer's test\ngiven in [Tong, J. (1994). Amer. Math. Monthly. 101(5): 450--452]. Our proof is\nbased on an application of a Hardy--Littlewood Tauberian theorem.\n"}
{"abstract": "  Transformer models are permutation equivariant. To supply the order and type\ninformation of the input tokens, position and segment embeddings are usually\nadded to the input. Recent works proposed variations of positional encodings\nwith relative position encodings achieving better performance. Our analysis\nshows that the gain actually comes from moving positional information to\nattention layer from the input. Motivated by this, we introduce Decoupled\nPositional Attention for Transformers (DIET), a simple yet effective mechanism\nto encode position and segment information into the Transformer models. The\nproposed method has faster training and inference time, while achieving\ncompetitive performance on GLUE, XTREME and WMT benchmarks. We further\ngeneralize our method to long-range transformers and show performance gain.\n"}
{"abstract": "  We study possible superconducting states in transition metal dichalcogenide\n(TMD) monolayers, assuming an on-site pairing potential that includes both\nintra- and inter-orbital terms. We find that if the mirror symmetry with\nrespect to the system's plane is broken (e.g., by a substrate), this type of\npairing can give rise to unconventional superconductivity, including\ntime-reversal-invariant nodal and fully gapped topological phases. Using a\nmulti-orbital renormalization group procedure, we show how these phases may\nresult from the interplay between the local Coulomb repulsion, Hund's rule\ncoupling, and phonon-mediated attraction. In particular, for a range of\ninteraction parameters, the system transitions from a trivial phase to a nodal\nphase and finally to a gapped topological phase upon increasing the strength of\nthe mirror symmetry breaking term.\n"}
{"abstract": "  The reionization epoch concludes when ionizing photons reach every corner of\nthe Universe. Reionization has generally been assumed to be limited primarily\nby the rate at which galaxies produce ionizing photons, but the recent\nmeasurement of a surprisingly short ionizing photon mean free path of\n$0.75^{+0.65}_{-0.45}$ proper Mpc at $z = 6$ by Becker et al. (2021) suggests\nthat absorption by residual neutral hydrogen in the otherwise ionized\nintergalactic medium may play a much larger role than previously expected. Here\nwe show that consistency between this short mean free path and the coeval dark\npixel fraction in the Ly$\\alpha$ forest requires a cumulative output of\n$6.1^{+11}_{-2.4}$ ionizing photons per baryon by reionization's end. This\nrepresents a dramatic increase in the ionizing photon budget over previous\nestimates, greatly exacerbating the tension with measurements of the ionizing\noutput from galaxies at later times. Translating this constraint into the\ninstantaneous ionizing production from galaxies in our model, we find\n$\\log_{10}f_{\\rm esc}\\xi_{\\rm ion}/\\text{(erg/Hz)}^{-1} =25.02_{-0.21}^{+0.45}$\nat $z\\sim6$. Even with optimistic assumptions about the ionizing production\nefficiency of early stellar populations, and assuming the galaxy luminosity\nfunction extends to extremely faint sources ($M_{\\text{UV}}\\leq-11$), complete\nreionization requires the escape fraction of ionizing photons to exceed $20\\%$\nacross the galaxy population. This is far larger than observed in any galaxy\npopulation at lower redshifts, requiring rapid evolution in galaxy properties\nafter the first billion years of cosmic time. This tension cannot be completely\nrelieved within existing observational constraints on the hydrogen neutral\nfraction and mean free path.\n"}
{"abstract": "  Social robots able to continually learn facial expressions could\nprogressively improve their emotion recognition capability towards people\ninteracting with them. Semi-supervised learning through ensemble predictions is\nan efficient strategy to leverage the high exposure of unlabelled facial\nexpressions during human-robot interactions. Traditional ensemble-based\nsystems, however, are composed of several independent classifiers leading to a\nhigh degree of redundancy, and unnecessary allocation of computational\nresources. In this paper, we proposed an ensemble based on convolutional\nnetworks where the early layers are strong low-level feature extractors, and\ntheir representations shared with an ensemble of convolutional branches. This\nresults in a significant drop in redundancy of low-level features processing.\nTraining in a semi-supervised setting, we show that our approach is able to\ncontinually learn facial expressions through ensemble predictions using\nunlabelled samples from different data distributions.\n"}
{"abstract": "  We study both averaging and maximal averaging problems for Product\n$j$-varieties defined by $\\Pi_j=\\{x\\in \\mathbb F_q^d: \\prod_{k=1}^d x_k=j\\}$\nfor $j\\in \\mathbb F_q^*,$ where $\\mathbb F_q^d$ denotes a $d$-dimensional\nvector space over the finite field $\\mathbb F_q$ with $q$ elements. We prove\nthe sharp $L^p\\to L^r$ boundedness of averaging operators associated to Product\n$j$-varieties. We also obtain the optimal $L^p$ estimate for a maximal\naveraging operator related to a family of Product $j$-varieties\n$\\{\\Pi_j\\}_{j\\in \\mathbb F_q^*}.$\n"}
{"abstract": "  Heavy-tailed random samples, as well as their sum or average, are encountered\nin a number of signal processing applications in radar, communications,\nfinance, and natural sciences. Modeling such data through the Pareto\ndistribution is particularly attractive due to its simple analytical form, but\nmay lead to infinite variance and/or mean, which is not physically plausible:\nin fact, samples are always bounded in practice, namely because of clipping\nduring the signal acquisition or deliberate censoring or trimming (truncation)\nat the processing stage. Based on this motivation, the paper derives and\nanalyzes the distribution of the sum of right-censored Pareto Type-II\nvariables, which generalizes the conventional Pareto (Type-I) and Lomax\ndistributions. The distribution of the sum of truncated Pareto is also\nobtained, and an analytical connection is drawn with the unbounded case. A\nnumerical analysis illustrates the findings, providing insights on several\naspects including the intimate mixture structure of the obtained expressions.\n"}
{"abstract": "  The rapid growth of Decentralized Finance (DeFi) boosts the Ethereum\necosystem. At the same time, attacks towards DeFi applications (apps) are\nincreasing. However, to the best of our knowledge, existing smart contract\nvulnerability detection tools cannot be directly used to detect DeFi attacks.\nThat's because they lack the capability to recover and understand high-level\nDeFi semantics, e.g., a user trades a token pair X and Y in a Decentralized\nEXchange (DEX).\n  In this work, we focus on the detection of two types of new attacks on DeFi\napps, including direct and indirect price manipulation attacks. The former one\nmeans that an attacker directly manipulates the token price in DEX by\nperforming an unwanted trade in the same DEX by attacking the vulnerable DeFi\napp. The latter one means that an attacker indirectly manipulates the token\nprice of the vulnerable DeFi app (e.g., a lending app). To this end, we propose\na platform-independent way to recover high-level DeFi semantics by first\nconstructing the cash flow tree from raw Ethereum transactions and then lifting\nthe low-level semantics to high-level ones, including token trade, liquidity\nmining, and liquidity cancel. Finally, we detect price manipulation attacks\nusing the patterns expressed with the recovered DeFi semantics.\n  We have implemented a prototype named \\tool{} and applied it to more than 350\nmillion transactions. It successfully detected 432 real-world attacks in the\nwild. We confirm that they belong to four known security incidents and five\nzero-day ones. We reported our findings. Two CVEs have been assigned. We\nfurther performed an attack analysis to reveal the root cause of the\nvulnerability, the attack footprint, and the impact of the attack. Our work\nurges the need to secure the DeFi ecosystem.\n"}
{"abstract": "  We present a method for the direct measurement of the Wigner characteristic\nfunction of a thermalizing harmonic oscillator that is completely inaccessible\nfor control or measurement. The strategy employs a recently proposed\nprobe-measurement-based scheme [Phys. Rev. Lett. 122, 110406 (2019)] which\nrelies on the pulsed control of a two-level probe. We generalize this scheme to\nthe case of a nonunitary time evolution of the target harmonic oscillator,\ndescribing its thermalization through contact to a finite-temperature\nenvironment, given in the form of a Lindblad master equation. This\ngeneralization is achieved using a superoperator formalism and yields\nanalytical expressions for the direct measurement of the characteristic\nfunction, accounting for the decoherence during the measurement process.\n"}
{"abstract": "  We discuss theoretically static and dynamical properties of $XY$ and\nHeisenberg antiferromagnets on triangular lattice with random vacancies and\nsubstitutional spins. It is shown that the distortion of $120^\\circ$ magnetic\norder produced by a single defect is described by electrostatic equations for a\nfield of an electrically neutral complex of six charges located around the\nimpurity. The first finite term in the multipole expansion of this field is the\noctupole moment which decays as $1/r^3$ with the distance $r$. The linearity of\nequations allows to describe analytically the distortion of the long-range\nmagnetic order at a small concentration $c$ of defects. We obtain analytically\nrenormalization of the elastic neutron scattering cross section and the magnon\nspectrum $\\epsilon_{\\bf k}$ in the leading order in $c$. We find that the\nscattering on impurities renormalizes weakly the bare spectrum $\\epsilon_{\\bf\nk}\\propto k$ at $k\\gg\\sqrt c$. However the renormalization is substantial of\nthe long-wavelength magnon spectrum at $k\\ll\\sqrt c$: $\\epsilon_{\\bf k}\\propto\n\\sqrt{c /\\ln(1/k)}$ at $k\\to0$ and there is a parametrically large region in\nwhich magnons with not too small momenta are overdamped and localized. This\nstrong modification of the long-wavelength spectrum leads to the stabilization\nof the slightly distorted magnetic long-range order at $T<T_N\\sim\nS^2J/\\ln(1/c)$ and to the considerable change in the density of states and in\nthe specific heat. The overdamped modes arise also in quasi-2D spin systems on\na stacked triangular lattice.\n"}
{"abstract": "  The ability to independently deploy parts of a software system is one of the\ncornerstones of modern software development, and allows for these parts to\nevolve independently and at different speeds.\n  A major challenge of such independent deployment, however, is to ensure that\ndespite their individual evolution, the interfaces between interacting parts\nremain compatible. This is especially important for enterprise software\nsystems, which are often highly integrated and based on heterogenous IT\ninfrastructures.\n  Although several approaches for interface evolution have been proposed, many\nof these rely on the developer to adhere to certain rules, but provide little\nguidance for doing so. In this paper, we present an approach for interface\nevolution that is easy to use for developers, and also addresses typical\nchallenges of heterogenous enterprise software, especially legacy system\nintegration.\n"}
{"abstract": "  Electrocatalysts for bifunctional oxygen reduction (ORR) and oxygen evolution\nreaction (OER) are commonly studied under hydrodynamic conditions, rendering\nthe use of binders necessary to ensure the mechanical stability of the\nelectrode films. The presence of a binder, however, may influence the\nproperties of the materials under examination to an unknown extent. Herein, we\ninvestigate the impact of Nafion on a highly active ORR/OER catalyst consisting\nof MnFeNi oxide nanoparticles supported on multi-walled carbon nanotubes.\nElectrochemical studies revealed that, in addition to enhancing the mechanical\nstability and particle connectivity, Nafion poses a major impact on the ORR\nselectivity, which correlates with a decrease in the valence state of Mn\naccording to X-ray absorption spectroscopy. These findings call for awareness\nregarding the use of electrode additives, since in some cases the extent of\ntheir impact on the properties of electrode films cannot be regarded as\nnegligible.\n"}
{"abstract": "  Using the official data and aware of the uncertain source and insufficient\nnumber of samples, we present a first and (for the moment) unique attempt to\nstudy the first two months spread of COVID-19 in Madagascar. The approach has\nbeen tested by predicting the number of contaminated persons for the next week\nafter fitting the inputs data collected within 7 or 15 days using standard\nleast $\\chi^2$-fit method. Encouraged by this first test, we study\nsystematically during 67 days , 1-2 weeks new data and predict the contaminated\npersons for the coming week. We find that the first month data are well\ndescribed by a linear or quadratic polynomial with an increase of about (4-5)\ninfected persons per day. Pursuing the analysis, one note that data until 46\ndays favour a cubic polynomial behaviour which signals an eventual near future\nstronger growth as confirmed by the new data on the 48th day. We complete the\nanalysis until 67 days and find that the data until 77 days confirm the cubic\npolynomial behaviour which is a remarkable feature of the pandemic spread in\nMadagascar. We expect that these results will be useful for some new model\nbuildings. A comparison with some other SI-like models predictions is\ndone.These results may also be interpreted as the lowest values of the real\ncase due to the insufficient number of samples (12907 for 27 million habitants\non 05/06/20). The data analysis of the absolute number of cured persons until\n67 days shows an approximate linear behaviour with about 3 cured persons per\nday. However, the number of percentage number of cured persons decreases above\n42-46 days indicating the limits of the hospital equipment and care to face the\n2nd phase of the pandemic for the 67th first days. Some comments on the social,\neconomical and political impacts of COVID-19 and confinement for Madagascar\nand, in general, for Worldwide are shortly discussed.\n"}
{"abstract": "  It has been experimentally observed that the efficiency of distributed\ntraining with stochastic gradient (SGD) depends decisively on the batch size\nand -- in asynchronous implementations -- on the gradient staleness.\nEspecially, it has been observed that the speedup saturates beyond a certain\nbatch size and/or when the delays grow too large. We identify a data-dependent\nparameter that explains the speedup saturation in both these settings. Our\ncomprehensive theoretical analysis, for strongly convex, convex and non-convex\nsettings, unifies and generalized prior work directions that often focused on\nonly one of these two aspects. In particular, our approach allows us to derive\nimproved speedup results under frequently considered sparsity assumptions. Our\ninsights give rise to theoretically based guidelines on how the learning rates\ncan be adjusted in practice. We show that our results are tight and illustrate\nkey findings in numerical experiments.\n"}
{"abstract": "  Biomarkers play an important role in early detection and intervention in\nAlzheimer's disease (AD). However, obtaining effective biomarkers for AD is\nstill a big challenge. In this work, we propose to use the worst transportation\ncost as a univariate biomarker to index cortical morphometry for tracking AD\nprogression. The worst transportation (WT) aims to find the least economical\nway to transport one measure to the other, which contrasts to the optimal\ntransportation (OT) that finds the most economical way between measures. To\ncompute the WT cost, we generalize the Brenier theorem for the OT map to the WT\nmap, and show that the WT map is the gradient of a concave function satisfying\nthe Monge-Ampere equation. We also develop an efficient algorithm to compute\nthe WT map based on computational geometry. We apply the algorithm to analyze\ncortical shape difference between dementia due to AD and normal aging\nindividuals. The experimental results reveal the effectiveness of our proposed\nmethod which yields better statistical performance than other competiting\nmethods including the OT.\n"}
{"abstract": "  The origin semantics for transducers was proposed in 2014, and led to various\ncharacterizations and decidability results that are in contrast with the\nclassical semantics. In this paper we add a further decidability result for\ncharacterizing transducers that are close to one-way transducers in the origin\nsemantics. We show that it is decidable whether a non-deterministic two-way\nword transducer can be resynchronized by a bounded, regular resynchronizer into\nan origin-equivalent one-way transducer. The result is in contrast with the\nusual semantics, where it is undecidable to know if a non-deterministic two-way\ntransducer is equivalent to some one-way transducer.\n"}
{"abstract": "  The interplay of disorder and strong correlations in quantum many-body\nsystems remains an open question. That is despite much progress made in recent\nyears with ultracold atoms in optical lattices to better understand phenomena\nsuch as many-body localization or the effect of disorder on Mott\nmetal-insulator transitions. Here, we utilize the numerical linked-cluster\nexpansion technique, extended to treat disordered quantum lattice models, and\nstudy exact thermodynamic properties of the disordered Fermi-Hubbard model on\nthe square and cubic geometries. We consider box distributions for the disorder\nin the onsite energy, the interaction strength, as well as the hopping\namplitude and explore how energy, double occupancy, entropy, heat capacity and\nmagnetic correlations of the system in the thermodynamic limit evolve as the\nstrength of disorder changes. We compare our findings with those obtained from\ndeterminant quantum Monte Carlo simulations and discuss the relevance of our\nresults to experiments with cold fermionic atoms in optical lattices.\n"}
{"abstract": "  The power-law distribution is ubiquitous and its mechanism seems to be\nvarious. We find a general mechanism for the distribution. The distribution of\na geometrically growing system can be approximated by a log - completely\nsquared chi distribution with 1 degree of freedom (log-CS$\\chi_1$), which\nreaches asymptotically a power-law distribution, or by a log-normal\ndistribution, which has an infinite asymptotic slope, at the upper limit. For\nthe log-CS$\\chi_1$, the asymptotic exponent of the power-law or the slope in a\nlog-log diagram seems to be related only to the variances of the system\nparameters and their mutual correlation but independent of an initial\ndistribution of the system or any mean value of parameters. We can take the\nlog-CS$\\chi_1$ as a unique approximation when the system should have a singular\ninitial distribution. The mechanism shows a comprehensiveness to be applicable\nto wide practice. We derive a simple formula for the Zipf's exponent, which\nwill probably demand that the exponent should be near -1 rather than exactly\n-1. We show that this approach can explain statistics of the COVID-19 pandemic.\n"}
{"abstract": "  We prove that any one-relator group $G$ is the fundamental group of a compact\nSasakian manifold if and only if $G$ is either finite cyclic or isomorphic to\nthe fundamental group of a compact Riemann surface of genus g > 0 with at most\none orbifold point of order $n \\geq 1$. We also classify all groups of\ndeficiency at least two that are also the fundamental group of some compact\nSasakian manifold.\n"}
{"abstract": "  It is desirable, and often a necessity, for machine learning models to be\nrobust against adversarial attacks. This is particularly true for Bayesian\nmodels, as they are well-suited for safety-critical applications, in which\nadversarial attacks can have catastrophic outcomes. In this work, we take a\ndeeper look at the adversarial robustness of Bayesian Neural Networks (BNNs).\nIn particular, we consider whether the adversarial robustness of a BNN can be\nincreased by model choices, particularly the Lipschitz continuity induced by\nthe prior. Conducting in-depth analysis on the case of i.i.d., zero-mean\nGaussian priors and posteriors approximated via mean-field variational\ninference, we find evidence that adversarial robustness is indeed sensitive to\nthe prior variance.\n"}
{"abstract": "  Propagation Phase Contrast Synchrotron Microtomography (PPC-SR${\\mu}$CT) is\nthe gold standard for non-invasive and non-destructive access to internal\nstructures of archaeological remains. In this analysis, the virtual specimen\nneeds to be segmented to separate different parts or materials, a process that\nnormally requires considerable human effort. In the Automated SEgmentation of\nMicrotomography Imaging (ASEMI) project, we developed a tool to automatically\nsegment these volumetric images, using manually segmented samples to tune and\ntrain a machine learning model. For a set of four specimens of ancient Egyptian\nanimal mummies we achieve an overall accuracy of 94-98% when compared with\nmanually segmented slices, approaching the results of off-the-shelf commercial\nsoftware using deep learning (97-99%) at much lower complexity. A qualitative\nanalysis of the segmented output shows that our results are close in terms of\nusability to those from deep learning, justifying the use of these techniques.\n"}
{"abstract": "  We introduce differential primary decompositions for ideals in a commutative\nring. Ideal membership is characterized by differential conditions. The minimal\nnumber of conditions needed is the arithmetic multiplicity. Minimal\ndifferential primary decompositions are unique up to change of bases. Our\nresults generalize the construction of Noetherian operators for primary ideals\nin the analytic theory of Ehrenpreis-Palamodov, and they offer a concise method\nfor representing affine schemes. The case of modules is also addressed. We\nimplemented an algorithm in Macaulay2 that computes the minimal decomposition\nfor an ideal in a polynomial ring.\n"}
{"abstract": "  Axial light field resolution refers to the ability to distinguish features at\ndifferent depths by refocusing. The axial refocusing precision corresponds to\nthe minimum distance in the axial direction between two distinguishable\nrefocusing planes. High refocusing precision can be essential for some light\nfield applications like microscopy. In this paper, we propose a learning-based\nmethod to extrapolate novel views from axial volumes of sheared epipolar plane\nimages (EPIs). As extended numerical aperture (NA) in classical imaging, the\nextrapolated light field gives re-focused images with a shallower depth of\nfield (DOF), leading to more accurate refocusing results. Most importantly, the\nproposed approach does not need accurate depth estimation. Experimental results\nwith both synthetic and real light fields show that the method not only works\nwell for light fields with small baselines as those captured by plenoptic\ncameras (especially for the plenoptic 1.0 cameras), but also applies to light\nfields with larger baselines.\n"}
{"abstract": "  We determine the probability that a random polynomial of degree $n$ over\n$\\mathbb{Z}_p$ has exactly $r$ roots in $\\mathbb{Q}_p$, and show that it is\ngiven by a rational function of $p$ that is invariant under replacing $p$ by\n$1/p$.\n"}
{"abstract": "  We investigate the two-dimensional energy-momentum-tensor (EMT) distributions\nof the nucleon on the light front, using the Abel transforms of the\nthree-dimensional EMT ones. We explicitly show that the main features of all\nEMT distributions are kept intact in the course of the Abel transform. We also\nexamine the equivalence between the global and local conditions for the nucleon\nstability in the three-dimensional Breit frame and in the two-dimensional\ntransverse plane on the light front. We also discuss the two-dimensional force\nfields inside a nucleon on the light front.\n"}
{"abstract": "  In recent years, single modality based gait recognition has been extensively\nexplored in the analysis of medical images or other sensory data, and it is\nrecognised that each of the established approaches has different strengths and\nweaknesses. As an important motor symptom, gait disturbance is usually used for\ndiagnosis and evaluation of diseases; moreover, the use of multi-modality\nanalysis of the patient's walking pattern compensates for the one-sidedness of\nsingle modality gait recognition methods that only learn gait changes in a\nsingle measurement dimension. The fusion of multiple measurement resources has\ndemonstrated promising performance in the identification of gait patterns\nassociated with individual diseases. In this paper, as a useful tool, we\npropose a novel hybrid model to learn the gait differences between three\nneurodegenerative diseases, between patients with different severity levels of\nParkinson's disease and between healthy individuals and patients, by fusing and\naggregating data from multiple sensors. A spatial feature extractor (SFE) is\napplied to generating representative features of images or signals. In order to\ncapture temporal information from the two modality data, a new correlative\nmemory neural network (CorrMNN) architecture is designed for extracting\ntemporal features. Afterwards, we embed a multi-switch discriminator to\nassociate the observations with individual state estimations. Compared with\nseveral state-of-the-art techniques, our proposed framework shows more accurate\nclassification results.\n"}
{"abstract": "  On-the-job learning consists in continuously learning while being used in\nproduction, in an open environment, meaning that the system has to deal on its\nown with situations and elements never seen before. The kind of systems that\nseem to be especially adapted to on-the-job learning are dialogue systems,\nsince they can take advantage of their interactions with users to collect\nfeedback to adapt and improve their components over time. Some dialogue systems\nperforming on-the-job learning have been built and evaluated but no general\nmethodology has yet been defined. Thus in this paper, we propose a first\ngeneral methodology for evaluating on-the-job learning dialogue systems. We\nalso describe a task-oriented dialogue system which improves on-the-job its\nnatural language component through its user interactions. We finally evaluate\nour system with the described methodology.\n"}
{"abstract": "  Reconfigurable intelligent surface (RIS) assisted radio is considered as an\nenabling technology with great potential for the sixth-generation (6G) wireless\ncommunications standard. The achievable secrecy rate (ASR) is one of the most\nfundamental metrics to evaluate the capability of facilitating secure\ncommunication for RIS-assisted systems. However, the definition of ASR is based\non Shannon's information theory, which generally requires long codewords and\nthus fails to quantify the secrecy of emerging delay-critical services.\nMotivated by this, in this paper we investigate the problem of maximizing the\nsecrecy rate under a delay-limited quality-of-service (QoS) constraint, termed\nas the effective secrecy rate (ESR), for an RIS-assisted multiple-input\nsingle-output (MISO) wiretap channel subject to a transmit power constraint. We\npropose an iterative method to find a stationary solution to the formulated\nnon-convex optimization problem using a block coordinate ascent method (BCAM),\nwhere both the beamforming vector at the transmitter as well as the phase\nshifts at the RIS are obtained in closed forms in each iteration. We also\npresent a convergence proof, an efficient implementation, and the associated\ncomplexity analysis for the proposed method. Our numerical results demonstrate\nthat the proposed optimization algorithm converges significantly faster that an\nexisting solution. The simulation results also confirm that the secrecy rate\nperformance of the system with stringent delay requirements reduce\nsignificantly compared to the system without any delay constraints, and that\nthis reduction can be significantly mitigated by an appropriately placed\nlarge-size RIS.\n"}
{"abstract": "  The fitting of physical models is often done only using a single target\nobservable. However, when multiple targets are considered, the fitting\nprocedure becomes cumbersome, there being no easy way to quantify the\nrobustness of the model for all different observables. Here, we illustrate that\none can jointly search for the best model for each desired observable through\nmulti-objective optimization. To do so we construct the Pareto front to study\nif there exists a set of parameters of the model that can jointly describe\nmultiple, or all, observables. To alleviate the computational cost, the\npredicted error for each targeted objective is approximated with a Gaussian\nprocess model, as it is commonly done in the Bayesian optimization framework.\nWe applied this methodology to improve three different models used in the\nsimulation of stationary state $cis-trans$ photoisomerization of retinal in\nrhodopsin. Optimization was done with respect to different experimental\nmeasurements, including emission spectra, peak absorption frequencies for the\n$cis$ and $trans$ conformers, and the energy storage.\n"}
{"abstract": "  Using VLTI/GRAVITY and SINFONI data, we investigate the sub-pc gas and dust\nstructure around the nearby type 1 AGN hosted by NGC 3783. The K-band coverage\nof GRAVITY uniquely allows a simultaneous analysis of the size and kinematics\nof the broad line region (BLR), the size and structure of the near-IR continuum\nemitting hot dust, and the size of the coronal line region (CLR). We find the\nBLR probed through broad Br$\\gamma$ emission is well described by a rotating,\nthick disk with a radial distribution of clouds peaking in the inner region. In\nour BLR model the physical mean radius of 16 light days is nearly twice the 10\nday time lag that would be measured, which matches very well the 10 day time\nlag that has been measured by reverberation mapping. We measure a hot dust FWHM\nsize of 0.74 mas (0.14 pc) and further reconstruct an image of the hot dust\nwhich reveals a faint (5% of the total flux) offset cloud which we interpret as\nan accreting cloud heated by the central AGN. Finally, we directly measure the\nFWHM size of the nuclear CLR as traced by the [CaVIII] and narrow Br$\\gamma$\nline. We find a FWHM size of 2.2 mas (0.4 pc), fully in line with the\nexpectation of the CLR located between the BLR and narrow line region.\nCombining all of these measurements together with larger scale near-IR integral\nfield unit and mid-IR interferometry data, we are able to comprehensively map\nthe structure and dynamics of gas and dust from 0.01--100 pc.\n"}
{"abstract": "  In this paper, we employ the novel design of the metamaterial half-wave plate\nby using the multiple layers of the metamaterials with some specific rotation\nangles. The rotation angles are given by composite pulse control which is the\nwell-known quantum control technique. A big advantage of this method can\nanalytically calculate the rotation angles and can be easily extended to the\nmultiple layers. The more layers of the metamaterials can present more\nbandwidth of half-wave plate. In this paper, we present 1, 3, 5, 7\nconfigurations of layers of a metamaterial half-wave plate and we demonstrate\nthat our device increasingly enhance the bandwidth of performance.\n"}
{"abstract": "  Explanation techniques are commonly evaluated using human-grounded methods,\nlimiting the possibilities for large-scale evaluations and rapid progress in\nthe development of new techniques. We propose a functionally-grounded\nevaluation procedure for local model-agnostic explanation techniques. In our\napproach, we generate ground truth for explanations when the black-box model is\nLogistic Regression and Gaussian Naive Bayes and compare how similar each\nexplanation is to the extracted ground truth. In our empirical study,\nexplanations of Local Interpretable Model-agnostic Explanations (LIME), SHapley\nAdditive exPlanations (SHAP), and Local Permutation Importance (LPI) are\ncompared in terms of how similar they are to the extracted ground truth. In the\ncase of Logistic Regression, we find that the performance of the explanation\ntechniques is highly dependent on the normalization of the data. In contrast,\nLocal Permutation Importance outperforms the other techniques on Naive Bayes,\nirrespective of normalization. We hope that this work lays the foundation for\nfurther research into functionally-grounded evaluation methods for explanation\ntechniques.\n"}
{"abstract": "  Active learning has been increasingly applied to screening functional\nmaterials from existing materials databases with desired properties. However,\nthe number of known materials deposited in the popular materials databases such\nas ICSD and Materials Project is extremely limited and consists of just a tiny\nportion of the vast chemical design space. Herein we present an active\ngenerative inverse design method that combines active learning with a deep\nvariational autoencoder neural network and a generative adversarial deep neural\nnetwork model to discover new materials with a target property in the whole\nchemical design space. The application of this method has allowed us to\ndiscover new thermodynamically stable materials with high band gap (SrYF$_5$)\nand semiconductors with specified band gap ranges (SrClF$_3$, CaClF$_5$,\nYCl$_3$, SrC$_2$F$_3$, AlSCl, As$_2$O$_3$), all of which are verified by the\nfirst principle DFT calculations. Our experiments show that while active\nlearning itself may sample chemically infeasible candidates, these samples help\nto train effective screening models for filtering out materials with desired\nproperties from the hypothetical materials created by the generative model. The\nexperiments show the effectiveness of our active generative inverse design\napproach.\n"}
{"abstract": "  We present the discovery of the fading radio transient FIRST\nJ153350.8+272729. The source had a maximum observed 5-GHz radio luminosity of\n$8\\times10^{39}$ erg s$^{-1}$ in 1986, but by 2019 had faded by a factor of\nnearly 400. It is located 0.15 arcsec from the center of a galaxy (SDSS\nJ153350.89+272729) at 147 Mpc, which shows weak Type II Seyfert activity. We\nshow that a tidal disruption event (TDE) is the preferred scenario for FIRST\nJ153350.8+272729, although it could plausibly be interpreted as the afterglow\nof a long-duration gamma-ray burst. This is only the second TDE candidate to be\nfirst discovered at radio wavelengths. Its luminosity fills a gap between the\nradio afterglows of sub-relativistic TDEs in the local universe, and\nrelativistic TDEs at high redshifts. The unusual properties of FIRST\nJ153350.8+272729 (ongoing nuclear activity in the host galaxy, high radio\nluminosity) motivate more extensive TDE searches in untargeted radio surveys.\n"}
{"abstract": "  We search some splitting (resp. finiteness) criteria by applying certain\nsplitting (resp. finiteness) property of the cohomological functors. We present\na connection from our approach to the realization problem of Nunke. This\nequipped with several applications. For instance, we recover some results of\nJensen (and others) by applying simple methods. Additional applications include\na computation of some numerical invariants such as the projective dimension of\nsome injective modules and et cetera.\n"}
{"abstract": "  We find the optimal investment strategy in a Black-Scholes market to minimize\nthe probability of so-called {\\it lifetime exponential Parisian ruin}, that is,\nthe probability that wealth exhibits an excursion below zero of an\nexponentially distributed time before the individual dies. We find that\nleveraging the risky asset is worse for negative wealth when minimizing the\nprobability of lifetime exponential Parisian ruin than when minimizing the\nprobability of lifetime ruin. Moreover, when wealth is negative, the optimal\namount invested in the risky asset increases as the hazard rate of the\nexponential ``excursion clock'' increases. In view of the heavy leveraging when\nwealth is negative, we also compute the minimum probability of lifetime\nexponential Parisian ruin under a constraint on investment. Finally, we derive\nan asymptotic expansion of the minimum probability of lifetime exponential\nParisian ruin for small values of the hazard rate of the excursion clock. It is\ninteresting to find that, for small values of this hazard rate, the minimum\nprobability of lifetime exponential Parisian ruin is proportional to the\nminimum occupation time studied in Bayraktar and Young, and the proportion\nequals the hazard rate. To the best of our knowledge, our work is the first to\n{\\it control} the probability of Parisian ruin.\n"}
{"abstract": "  Previous math word problem solvers following the encoder-decoder paradigm\nfail to explicitly incorporate essential math symbolic constraints, leading to\nunexplainable and unreasonable predictions. Herein, we propose Neural-Symbolic\nSolver (NS-Solver) to explicitly and seamlessly incorporate different levels of\nsymbolic constraints by auxiliary tasks. Our NS-Solver consists of a problem\nreader to encode problems, a programmer to generate symbolic equations, and a\nsymbolic executor to obtain answers. Along with target expression supervision,\nour solver is also optimized via 4 new auxiliary objectives to enforce\ndifferent symbolic reasoning: a) self-supervised number prediction task\npredicting both number quantity and number locations; b) commonsense constant\nprediction task predicting what prior knowledge (e.g. how many legs a chicken\nhas) is required; c) program consistency checker computing the semantic loss\nbetween predicted equation and target equation to ensure reasonable equation\nmapping; d) duality exploiting task exploiting the quasi duality between\nsymbolic equation generation and problem's part-of-speech generation to enhance\nthe understanding ability of a solver. Besides, to provide a more realistic and\nchallenging benchmark for developing a universal and scalable solver, we also\nconstruct a new large-scale MWP benchmark CM17K consisting of 4 kinds of MWPs\n(arithmetic, one-unknown linear, one-unknown non-linear, equation set) with\nmore than 17K samples. Extensive experiments on Math23K and our CM17k\ndemonstrate the superiority of our NS-Solver compared to state-of-the-art\nmethods.\n"}
{"abstract": "  Today, power waste is one of the most crucial problems which power stations\nacross the world are facing. One of the recent trends of the energy system is\nthe lean management technique. When lean management is indicated by the system,\ncustomer value is increased and the waste process in industry or in a power\nstation is reduced. In this paper, first of all, we propose mathematical\nmodeling to reduce the cost of production, and then the simulation technique\napplies to electricity transmission distribution systems. Furthermore, we\nconsider two criteria for comparison including the different costs of the\nsystem and the rate of energy waste during the transmission. The primary\napproach is to use both of the models in order to draw a comparison between\nsimulation results and mathematical models. Finally, the analysis of the test\nresults done by the CPLEX toolbox of MATLAB Software 2019 leads to a remarkable\ndecrease in the costs of energy demand in the electricity transmitting network\ndistribution system.\n"}
{"abstract": "  In the Equal Maximum Flow Problem (EMFP), we aim for a maximum flow where we\nrequire the same flow value on all edges in some given subsets of the edge set.\nIn this paper, we study the closely related Almost Equal Maximum Flow Problems\n(AEMFP) where the flow values on edges of one homologous edge set differ at\nmost by the valuation of a so called deviation function~$\\Delta$. We prove that\nthe integer almost equal maximum flow problem (integer AEMFP) is in general\n$\\mathcal{NP}$-complete, and show that even the problem of finding a fractional\nmaximum flow in the case of convex deviation functions is also\n$\\mathcal{NP}$-complete. This is in contrast to the EMFP, which is polynomial\ntime solvable in the fractional case. We provide inapproximability results for\nthe integral AEMFP. For the integer AEMFP we state a polynomial algorithm for\nthe constant deviation and concave case for a fixed number of homologous sets.\n"}
{"abstract": "  Administering standardized examinations is a challenging task, especially for\nthose universities for which colleges affiliated to it are geographically\ndistributed over a wide area. Some of the challenges include maintaining\nintegrity and confidentiality of examination records, preventing mal-practices,\nissuing unique identification numbers to a large student population and\nmanaging assets required for the smooth conduct of examinations. These\nchallenges aggravate when colleges affiliated to universities demand academic\nand administrative autonomy by demonstrating best practices consistently over a\nlong period.\n  In this chapter, we describe a model for decentralized and autonomous\nexamination system to provide the necessary administrative support. The model\nis based on two emerging technologies of Blockchain Technology and Internet of\nThings (IoT). We adopt a software architecture approach to describe the model.\nThe prescriptive architecture consists of {\\em architectural mappings} which\nmap functional and non-functional requirements to architectural elements of\nblockchain technology and IoT. In architectural mappings, first, we identify\ncommon use-cases in administering standardized examinations. Then we map these\nuse-cases to the core elements of blockchain, i.e. distributed ledgers,\ncryptography, consensus protocols and smart-contracts and IoT. Such kind of\nprescriptive architecture guide downstream software engineering processes of\nimplementation and testing\n"}
{"abstract": "  A hot Markovian system can cool down faster than a colder one: this is known\nas the Mpemba effect. Here, we show that a non-equilibrium driving via\nstochastic reset can induce this phenomenon, when absent. Moreover, we derive\nan optimal driving protocol simultaneously optimizing the appearance time of\nthe Mpemba effect, and the total energy dissipation into the environment,\nrevealing the existence of a Pareto front. Building upon previous experimental\nresults, our findings open up the avenue of possible experimental realizations\nof optimal cooling protocols in Markovian systems.\n"}
{"abstract": "  The study of the prophet inequality problem in the limited information regime\nwas initiated by Azar et al. [SODA'14] in the pursuit of prior-independent\nposted-price mechanisms. As they show, $O(1)$-competitive policies are\nachievable using only a single sample from the distribution of each agent. A\nnotable portion of their results relies on reducing the design of single-sample\nprophet inequalities (SSPIs) to that of order-oblivious secretary (OOS)\npolicies. The above reduction comes at the cost of not fully utilizing the\navailable samples. However, to date, this is essentially the only method for\nproving SSPIs for many combinatorial sets. Very recently, Rubinstein et al.\n[ITCS'20] give a surprisingly simple algorithm which achieves the optimal\ncompetitive ratio for the single-choice SSPI problem $-$ a result which is\nunobtainable going through the reduction to secretary problems. Motivated by\nthis discrepancy, we study the competitiveness of simple SSPI policies\ndirectly, without appealing to results from OOS literature. In this direction,\nwe first develop a framework for analyzing policies against a greedy-like\nprophet solution. Using this framework, we obtain the first SSPI for general\n(non-bipartite) matching environments, as well as improved competitive ratios\nfor transversal and truncated partition matroids. Second, motivated by the\nobservation that many OOS policies for matroids decompose the problem into\nindependent rank-$1$ instances, we provide a meta-theorem which applies to any\nmatroid satisfying this partition property. Leveraging the recent results by\nRubinstein et al., we obtain improved competitive guarantees (most by a factor\nof $2$) for a number of matroids captured by the reduction of Azar et al.\nFinally, we discuss applications of our SSPIs to the design of mechanisms for\nmulti-dimensional limited information settings with improved revenue and\nwelfare guarantees.\n"}
{"abstract": "  Pion and kaon structural properties provide insights into the emergence of\nmass within the Standard Model and attendant modulations by the Higgs boson.\nNovel expressions of these effects, in impact parameter space and in mass and\npressure profiles, are exposed via $\\pi$ and $K$ generalised parton\ndistributions, built using the overlap representation from light-front wave\nfunctions constrained by one-dimensional valence distribution functions that\ndescribe available data. Notably, e.g. $K$ pressure profiles are spatially more\ncompact than $\\pi$ profiles and both achieve near-core pressures of similar\nmagnitude to that found in neutron stars.\n"}
{"abstract": "  Epileptic seizure prediction has gained considerable interest in the\ncomputational Epilepsy research community. This paper presents a Machine\nLearning based method for epileptic seizure prediction which outperforms\nstate-of-the art methods. We compute a probability for a given epoch, of being\npre-ictal against interictal using the Random Forest classifier and introduce\nnew concepts to enhance the robustness of the algorithm to false alarms. We\nassessed our method on 20 patients of the benchmark scalp EEG CHB-MIT dataset\nfor a seizure prediction horizon (SPH) of 5 minutes and a seizure occurrence\nperiod (SOP) of 30 minutes. Our approach achieves a sensitivity of 82.07 % and\na low false positive rate (FPR) of 0.0799 /h. We also tested our approach on\nintracranial EEG recordings.\n"}
{"abstract": "  This study investigates the electronic states and physical quantities of an\norganic charge-transfer complex HMTSF-TCNQ, which undergoes a\ncharge-density-wave (CDW) phase transition at temperature $T_c\\simeq 30$ K. A\nfirst-principles calculation is utilized to determine that the normal state is\na topological semimetal with open nodal lines. Besed on the first-principles\ncalculation, we develop a tight-binding model to investigate the electronic\nstate in detail. Below $T_c$, the CDW phase is examined in the tight-binding\nscheme using the mean-field approximation. It is shown that the open nodal\nlines are deformed into closed ones, and their shapes are sensitive to the\norder parameter. Using this tight-binding model, we theoretically evaluate the\ntemperature dependencies of two physical quantities: the spin-lattice\nrelaxation time $T_1$ and the orbital magnetic susceptibility. In particular,\nan anomalous plateau is obtained at low temperatures in the orbital\ndiamagnetism. We presume that this anomalous plateau originates owing to the\nconflict between the interband diamagnetism, impurity scattering, and the nodal\nline deformation. We also conduct an experiment to investigate the orbital\nmagnetism, and the results are in excellent quantitative agreement with the\ntheory.\n"}
{"abstract": "  Dialog State Tracking (DST), an integral part of modern dialog systems, aims\nto track user preferences and constraints (slots) in task-oriented dialogs. In\nreal-world settings with constantly changing services, DST systems must\ngeneralize to new domains and unseen slot types. Existing methods for DST do\nnot generalize well to new slot names and many require known ontologies of slot\ntypes and values for inference. We introduce a novel ontology-free framework\nthat supports natural language queries for unseen constraints and slots in\nmulti-domain task-oriented dialogs. Our approach is based on generative\nquestion-answering using a conditional language model pre-trained on\nsubstantive English sentences. Our model improves joint goal accuracy in\nzero-shot domain adaptation settings by up to 9% (absolute) over the previous\nstate-of-the-art on the MultiWOZ 2.1 dataset.\n"}
{"abstract": "  Face attribute editing aims to generate faces with one or multiple desired\nface attributes manipulated while other details are preserved. Unlike prior\nworks such as GAN inversion, which has an expensive reverse mapping process, we\npropose a simple feed-forward network to generate high-fidelity manipulated\nfaces. By simply employing some existing and easy-obtainable prior information,\nour method can control, transfer, and edit diverse attributes of faces in the\nwild. The proposed method can consequently be applied to various applications\nsuch as face swapping, face relighting, and makeup transfer. In our method, we\ndecouple identity, expression, pose, and illumination using 3D priors; separate\ntexture and colors by using region-wise style codes. All the information is\nembedded into adversarial learning by our identity-style normalization module.\nDisentanglement losses are proposed to enhance the generator to extract\ninformation independently from each attribute. Comprehensive quantitative and\nqualitative evaluations have been conducted. In a single framework, our method\nachieves the best or competitive scores on a variety of face applications.\n"}
{"abstract": "  Consider the class of optimal partition problems with long range interactions\n\\[ \\inf \\left\\{ \\sum_{i=1}^k \\lambda_1(\\omega_i):\\ (\\omega_1,\\ldots, \\omega_k)\n\\in \\mathcal{P}_r(\\Omega) \\right\\}, \\] where $\\lambda_1(\\cdot)$ denotes the\nfirst Dirichlet eigenvalue, and $\\mathcal{P}_r(\\Omega)$ is the set of open\n$k$-partitions of $\\Omega$ whose elements are at distance at least $r$:\n$\\textrm{dist}(\\omega_i,\\omega_j)\\geq r$ for every $i\\neq j$. In this paper we\nprove optimal uniform bounds (as $r\\to 0^+$) in $\\mathrm{Lip}$-norm for the\nassociated $L^2$-normalized eigenfunctions, connecting in particular the\nnonlocal case $r>0$ with the local one $r \\to 0^+$.\n  The proof uses new pointwise estimates for eigenfunctions, a one-phase\nAlt-Caffarelli-Friedman and the Caffarelli-Jerison-Kenig monotonicity formulas,\ncombined with elliptic and energy estimates. Our result extends to other\ncontexts, such as singularly perturbed harmonic maps with distance constraints.\n"}
{"abstract": "  We consider a bounded domain $\\Omega$ of $\\mathbb{R}^N$, $N\\ge3$, $h$ and $b$\ncontinuous functions on $\\Omega$. Let $\\Gamma$ be a closed curve contained in\n$\\Omega$. We study existence of positive solutions $u \\in\nH^1_0\\left(\\Omega\\right)$ to the perturbed Hardy-Sobolev equation: $$ -\\Delta\nu+h u+bu^{1+\\delta}=\\rho^{-\\sigma}_\\Gamma u^{2^*_\\sigma-1} \\qquad \\textrm{ in }\n\\Omega, $$ where $2^*_\\sigma:=\\frac{2(N-\\sigma)}{N-2}$ is the critical\nHardy-Sobolev exponent, $\\sigma\\in [0,2)$, $0< \\delta<\\frac{4}{N-2}$ and\n$\\rho_\\Gamma$ is the distance function to $\\Gamma$. We show that the existence\nof minimizers does not depend on the local geometry of $\\Gamma$ nor on the\npotential $h$. For $N=3$, the existence of ground-state solution may depends on\nthe trace of the regular part of the Green function of $-\\Delta+h$ and or on\n$b$. This is due to the perturbative term of order ${1+\\delta}$.\n"}
{"abstract": "  In the sense of distributions, the derivative of the Heaviside unit step\nfunction $H(t)$ is a generalized Dirac-$\\delta$ distribution. If the velocity\n$V(t)$ of a flat plate is impulsive, as $V(t)=H(t)$ (i.e., it is suddenly set\ninto motion with unit velocity at $t=0^+$), then its acceleration is\n$V'(t)=\\delta(t)$. The Dirac-$\\delta$ distribution has no point values.\nHowever, when the Dirac-$\\delta$ is the forcing term of an ODE (in $t$), it\ncontributes to the solution. The recently published paper [Chaos Solitons\nFractals 117 (2018) 68] incorrectly treats the Dirac-$\\delta$ function as being\nidentically 0. This Comment analyzes the source of this error, and provides\nguidance on how to correct it (based on the established literature). The\nmathematical error identified is in addition to some issues about rheological\nmodels with fractional derivatives, which are also noted. That is to say,\nwhether or not an \"Atangana--Baleanu fractional derivative\" is used in [Chaos\nSolitons Fractals 117 (2018) 68], the solution to Stokes' first problem\nprovided therein is not correct.\n"}
{"abstract": "  Bin picking is a core problem in industrial environments and robotics, with\nits main module as 6D pose estimation. However, industrial depth sensors have a\nlack of accuracy when it comes to small objects. Therefore, we propose a\nframework for pose estimation in highly cluttered scenes with small objects,\nwhich mainly relies on RGB data and makes use of depth information only for\npose refinement. In this work, we compare synthetic data generation approaches\nfor object detection and pose estimation and introduce a pose filtering\nalgorithm that determines the most accurate estimated poses. We will make our\n"}
{"abstract": "  As open source software (OSS) becomes increasingly mature and popular, there\nare significant challenges with properly accounting for usability concerns for\nthe diverse end users. Participatory design, where multiple stakeholders\ncollaborate on iterating the design, can be an efficient way to address the\nusability concerns for OSS projects. However, barriers such as a code-centric\nmindset and insufficient tool support often prevent OSS teams from effectively\nincluding end users in participatory design methods. This paper proposes\npreliminary contributions to this problem through the user-centered exploration\nof (1) a set of design guidelines that capture the needs of OSS participatory\ndesign tools, (2) two personas that represent the characteristics of OSS\ndesigners and end users, and (3) a low-fidelity prototype tool for end user\ninvolvement in OSS projects. This work paves the road for future studies about\ntool design that would eventually help improve OSS usability.\n"}
{"abstract": "  Intent Recognition and Slot Identification are crucial components in spoken\nlanguage understanding (SLU) systems. In this paper, we present a novel\napproach towards both these tasks in the context of low resourced and unwritten\nlanguages. We present an acoustic based SLU system that converts speech to its\nphonetic transcription using a universal phone recognition system. We build a\nword-free natural language understanding module that does intent recognition\nand slot identification from these phonetic transcription. Our proposed SLU\nsystem performs competitively for resource rich scenarios and significantly\noutperforms existing approaches as the amount of available data reduces. We\nobserve more than 10% improvement for intent classification in Tamil and more\nthan 5% improvement for intent classification in Sinhala. We also present a\nnovel approach towards unsupervised slot identification using normalized\nattention scores. This approach can be used for unsupervised slot labelling,\ndata augmentation and to generate data for a new slot in a one-shot way with\nonly one speech recording\n"}
{"abstract": "  Previous works on formally studying mobile robotic swarms consider necessary\nand sufficient system hypotheses enabling to solve theoretical benchmark\nproblems (geometric pattern formation, gathering, scattering, etc.). We argue\nthat formal methods can also help in the early stage of mobile robotic swarms\nprotocol design, to obtain protocols that are correct-by-design, even for\nproblems arising from real-world use cases, not previously studied\ntheoretically. Our position is supported by a concrete case study. Starting\nfrom a real-world case scenario, we jointly design the formal problem\nspecification, a family of protocols that are able to solve the problem, and\ntheir corresponding proof of correctness, all expressed with the same formal\nframework. The concrete framework we use for our development is the PACTOLE\nlibrary based on the COQ proof assistant.\n"}
{"abstract": "  In recent years, transformer-based models have shown state-of-the-art results\nfor Natural Language Processing (NLP). In particular, the introduction of the\nBERT language model brought with it breakthroughs in tasks such as question\nanswering and natural language inference, advancing applications that allow\nhumans to interact naturally with embedded devices. FPGA-based overlay\nprocessors have been shown as effective solutions for edge image and video\nprocessing applications, which mostly rely on low precision linear matrix\noperations. In contrast, transformer-based NLP techniques employ a variety of\nhigher precision nonlinear operations with significantly higher frequency. We\npresent NPE, an FPGA-based overlay processor that can efficiently execute a\nvariety of NLP models. NPE offers software-like programmability to the end user\nand, unlike FPGA designs that implement specialized accelerators for each\nnonlinear function, can be upgraded for future NLP models without requiring\nreconfiguration. We demonstrate that NPE can meet real-time conversational AI\nlatency targets for the BERT language model with $4\\times$ lower power than\nCPUs and $6\\times$ lower power than GPUs. We also show NPE uses $3\\times$ fewer\nFPGA resources relative to comparable BERT network-specific accelerators in the\nliterature. NPE provides a cost-effective and power-efficient FPGA-based\nsolution for Natural Language Processing at the edge.\n"}
{"abstract": "  A Heron triangle is a triangle whose side lengths and area are integers. Two\nHeron triangles are amicable if the perimeter of one is the area of the other.\nWe show, using elementary techniques, that there is only one pair of amicable\nHeron triangles.\n"}
{"abstract": "  Today, the Internet of Things (IoT) is one of the emerging technologies that\nenable the connection and transfer of information through communication\nnetworks. The main idea of the IoT is the widespread presence of objects such\nas mobile devices, sensors, and RFID. With the increase in traffic volume in\nurban areas, the existing intelligent urban traffic management system based on\nIoT can be vital. Therefore, this paper focused on security in urban traffic\nbased on using RFID. In our scheme, RFID tags chose as the purpose of this\narticle. We, in this paper, present a mutual authentication protocol that leads\nto privacy based on hybrid cryptography. Also, an authentication process with\nRFID tags is proposed that can be read at high speed. The protocol has\nattempted to reduce the complexity of computing. At the same time, the proposed\nmethod can withstand attacks such as spoofing of tag and reader, tag tracking,\nand replay attack.\n"}
{"abstract": "  Color-kinematics duality in the adjoint has proven key to the relationship\nbetween gauge and gravity theory scattering amplitude predictions. In recent\nwork, we demonstrated that at four-point tree-level, a small number of\ncolor-dual EFT building blocks could encode all higher-derivative single-trace\nmassless corrections to gauge and gravity theories compatible with adjoint\ndouble-copy. One critical aspect was the trivialization of building\nhigher-derivative color-weights -- indeed, it is the mixing of kinematics with\nnon-adjoint-type color-weights (like the permutation-invariant d^4) which\npermits description via adjoint double-copy. Here we find that such ideas\nclarify the predictions of local five-point higher-dimensional operators as\nwell. We demonstrate how a single scalar building block can be combined with\ncolor structures to build higher-derivative color factors that generate,\nthrough double copy, the predictions of higher-derivative gauge-theory\noperators. These may then be suitably mapped, through another double-copy, to\nhigher-derivative corrections in gravity.\n"}
{"abstract": "  We consider semilinear elliptic second-order partial differential\ninequalities of the form Lu +|u|q-1u < and = Lv +|v|q-1v (*) in the whole space\nRn, where n > and = 2, q > 0 and L is a linear elliptic second-order partial\ndifferential operator in divergence form. We assume that the coefficients of\nthe operator L are defined, measurable and locally bounded in Rn, and that the\nquadratic form associated with the operator L is symmetric and non-negative\ndefinite. We obtain a Liouville comparison principle in terms of a capacity\nassociated with the operator L for solutions of (*), which are defined and\nmeasurable in Rn and which belong locally to a Sobolev-type function space also\nassociated with the operator L.\n"}
{"abstract": "  We count the ordered sum-free triplets of subsets in the group\n$\\mathbb{Z}/p\\mathbb{Z}$, i.e., the triplets $(A,B,C)$ of sets $A,B,C \\subset\n\\mathbb{Z}/p\\mathbb{Z}$ for which the equation $a+b=c$ has no solution with\n$a\\in A$, $b \\in B$ and $c \\in C$. Our main theorem improves on a recent result\nby Semchankau, Shabanov, and Shkredov using a different and simpler method. Our\nproof relates previous results on the number of independent sets of regular\ngraphs by Kahn, Perarnau and Perkins, and Csikv\\'ari to produce explicit\nestimates on smaller order terms. We also obtain estimates for the number of\nsum-free triplets of subsets in a general abelian group.\n"}
{"abstract": "  In this paper, we are interested in the edge intersection graphs of paths of\na grid where each path has at most one bend, called B1-EPG graphs and first\nintroduced by Golumbic et al (2009). We also consider a proper subclass of\nB1-EPG, the L-EPG graphs, which allows paths only in ``L'' shape. We show that\ntwo superclasses of trees are B1-EPG (one of them being the cactus graphs). On\nthe other hand, we show that the block graphs are L-EPG and provide a linear\ntime algorithm to produce L-EPG representations of generalization of trees.\nThese proofs employed a new technique from previous results in the area based\non block-cutpoint trees of the respective graphs.\n"}
{"abstract": "  In this paper, inspired by the successes of visionlanguage pre-trained models\nand the benefits from training with adversarial attacks, we present a novel\ntransformerbased cross-modal fusion modeling by incorporating the both notions\nfor VQA challenge 2021. Specifically, the proposed model is on top of the\narchitecture of VinVL model [19], and the adversarial training strategy [4] is\napplied to make the model robust and generalized. Moreover, two implementation\ntricks are also used in our system to obtain better results. The experiments\ndemonstrate that the novel framework can achieve 76.72% on VQAv2 test-std set.\n"}
{"abstract": "  We report on depinning of nearly-commensurate charge-density waves in 1T-TaS2\nthin-films at room temperature. A combination of the differential\ncurrent-voltage measurements with the low-frequency noise spectroscopy provide\nunambiguous means for detecting the depinning threshold field in quasi-2D\nmaterials. The depinning process in 1T-TaS2 is not accompanied by an observable\nabrupt increase in electric current - in striking contrast to depinning in the\nconventional charge-density-wave materials with quasi-1D crystal structure. We\nexplained it by the fact that the current density from the charge-density waves\nin the 1T-TaS2 devices is orders of magnitude smaller than the current density\nof the free carriers available in the discommensuration network surrounding the\ncommensurate charge-density-wave islands. The depinning fields in 1T-TaS2\nthin-film devices are several orders of magnitude larger than those in quasi-1D\nvan der Waals materials. Obtained results are important for the proposed\napplications of the charge-density-wave devices in electronics.\n"}
{"abstract": "  Abstractive summarization for long-document or multi-document remains\nchallenging for the Seq2Seq architecture, as Seq2Seq is not good at analyzing\nlong-distance relations in text. In this paper, we present BASS, a novel\nframework for Boosting Abstractive Summarization based on a unified Semantic\ngraph, which aggregates co-referent phrases distributing across a long range of\ncontext and conveys rich relations between phrases. Further, a graph-based\nencoder-decoder model is proposed to improve both the document representation\nand summary generation process by leveraging the graph structure. Specifically,\nseveral graph augmentation methods are designed to encode both the explicit and\nimplicit relations in the text while the graph-propagation attention mechanism\nis developed in the decoder to select salient content into the summary.\nEmpirical results show that the proposed architecture brings substantial\nimprovements for both long-document and multi-document summarization tasks.\n"}
{"abstract": "  The excessive search for parking, known as cruising, generates pollution and\ncongestion. Cities are looking for approaches that will reduce the negative\nimpact associated with searching for parking. However, adequately measuring the\nnumber of vehicles in search of parking is difficult and requires sensing\ntechnologies. In this paper, we develop an approach that eliminates the need\nfor sensing technology by using parking meter payment transactions to estimate\nparking occupancy and the number of cars searching for parking. The estimation\nscheme is based on Particle Markov Chain Monte Carlo. We validate the\nperformance of the Particle Markov Chain Monte Carlo approach using data\nsimulated from a GI/GI/s queue. We show that the approach generates\nasymptotically unbiased Bayesian estimates of the parking occupancy and\nunderlying model parameters such as arrival rates, average parking time, and\nthe payment compliance rate. Finally, we estimate parking occupancy and\ncruising using parking meter data from SFpark, a large scale parking experiment\nand subsequently, compare the Particle Markov Chain Monte Carlo parking\noccupancy estimates against the ground truth data from the parking sensors. Our\napproach is easily replicated and scalable given that it only requires using\ndata that cities already possess, namely historical parking payment\ntransactions.\n"}
{"abstract": "  Recommender system recommends interesting items to users based on users' past\ninformation history. Researchers have been paying attention to improvement of\nalgorithmic performance such as MAE and precision@K. Major techniques such as\nmatrix factorization and learning to rank are optimized based on such\nevaluation metrics. However, the intrinsic Matthew Effect problem poses great\nthreat to the fairness of the recommender system, and the unfairness problem\ncannot be resolved by optimization of traditional metrics. In this paper, we\npropose a novel algorithm that incorporates Matthew Effect reduction with the\nmatrix factorization framework. We demonstrate that our approach can boost the\nfairness of the algorithm and enhances performance evaluated by traditional\nmetrics.\n"}
{"abstract": "  One of the key signatures of the Quark Gluon Plasma (QGP) is the energy loss\nof high momentum particles as they traverse the strongly interacting medium.\nThe energy loss of these particles is governed by the jet transport coefficient\n$\\hat{q}/T^3$, wherein it has been thought that there is a large jump as the\nsystem transitions between the hadron gas and Quark Gluon Plasma phases. Here\nwe calculate $\\hat{q}/T^3$ within the Hadron Resonance Gas (HRG) model with the\nparticle list PDG16+ and find that, if one incorporates the experimental error\nin the hadronic calculation of $\\hat{q}/T^3$ and assumes a higher\npseudo-critical temperature, then a smooth transition from the hadron gas phase\ninto the Quark Gluon Plasma phase is possible. We also find a significant\nenhancement in $\\hat{q}/T^3$ at finite baryon chemical potential and find\nissues with the relationship between the shear viscosity and the jet transport\ncoefficient within a hadron gas phase.\n"}
{"abstract": "  In recent times, object detection and pose estimation have gained significant\nattention in the context of robotic vision applications. Both the\nidentification of objects of interest as well as the estimation of their pose\nremain important capabilities in order for robots to provide effective\nassistance for numerous robotic applications ranging from household tasks to\nindustrial manipulation. This problem is particularly challenging because of\nthe heterogeneity of objects having different and potentially complex shapes,\nand the difficulties arising due to background clutter and partial occlusions\nbetween objects. As the main contribution of this work, we propose a system\nthat performs real-time object detection and pose estimation, for the purpose\nof dynamic robot grasping. The robot has been pre-trained to perform a small\nset of canonical grasps from a few fixed poses for each object. When presented\nwith an unknown object in an arbitrary pose, the proposed approach allows the\nrobot to detect the object identity and its actual pose, and then adapt a\ncanonical grasp in order to be used with the new pose. For training, the system\ndefines a canonical grasp by capturing the relative pose of an object with\nrespect to the gripper attached to the robot's wrist. During testing, once a\nnew pose is detected, a canonical grasp for the object is identified and then\ndynamically adapted by adjusting the robot arm's joint angles, so that the\ngripper can grasp the object in its new pose. We conducted experiments using a\nhumanoid PR2 robot and showed that the proposed framework can detect\nwell-textured objects, and provide accurate pose estimation in the presence of\ntolerable amounts of out-of-plane rotation. The performance is also illustrated\nby the robot successfully grasping objects from a wide range of arbitrary\nposes.\n"}
{"abstract": "  Modern deep learning applications require increasingly more compute to train\nstate-of-the-art models. To address this demand, large corporations and\ninstitutions use dedicated High-Performance Computing clusters, whose\nconstruction and maintenance are both environmentally costly and well beyond\nthe budget of most organizations. As a result, some research directions become\nthe exclusive domain of a few large industrial and even fewer academic actors.\nTo alleviate this disparity, smaller groups may pool their computational\nresources and run collaborative experiments that benefit all participants. This\nparadigm, known as grid- or volunteer computing, has seen successful\napplications in numerous scientific areas. However, using this approach for\nmachine learning is difficult due to high latency, asymmetric bandwidth, and\nseveral challenges unique to volunteer computing. In this work, we carefully\nanalyze these constraints and propose a novel algorithmic framework designed\nspecifically for collaborative training. We demonstrate the effectiveness of\nour approach for SwAV and ALBERT pretraining in realistic conditions and\nachieve performance comparable to traditional setups at a fraction of the cost.\nFinally, we provide a detailed report of successful collaborative language\nmodel pretraining with 40 participants.\n"}
{"abstract": "  The boundary element method is an efficient algorithm for simulating acoustic\npropagation through homogeneous objects embedded in free space. The\nconditioning of the system matrix strongly depends on physical parameters such\nas density, wavespeed and frequency. In particular, high contrast in density\nand wavespeed across a material interface leads to an ill-conditioned\ndiscretisation matrix. Therefore, the convergence of Krylov methods to solve\nthe linear system is slow. Here, specialised boundary integral formulations are\ndesigned for the case of acoustic scattering at high-contrast media. The\neigenvalues of the resulting system matrix accumulate at two points in the\ncomplex plane that depend on the density ratio and stay away from zero. The\nspectral analysis of the Calder\\'on preconditioned PMCHWT formulation yields a\nsingle accumulation point. Benchmark simulations demonstrate the computational\nefficiency of the high-contrast Neumann formulation for scattering at\nhigh-contrast media.\n"}
{"abstract": "  We theoretically investigated the Floquet states of bilayer graphene\nheterostructure under the irradiation by horizontally incident circular\npolarized light. The in-plane and out-of-plane electric field of the light\nperiodically perturbs the intra-layer and inter-layer hopping, respectively.\nFor circular polarized light, the two components of the electric field has\n$\\pi/2$ phase difference, so that the two types of hopping are periodically\nperturbed with the $\\pi/2$ phase difference, which modify the effective\ninter-layer hopping. We focus on the model of bilayer graphene in the\nheterostructure of antiferromagnetic van der Walls spin valve. The amplitude of\nthe irradiation can tune the band gap and topological properties of the bulk\nstate. The spin-polarized quantum anomalous Hall phase with Chern number being\none is predicted. The incident angle of the irradiation can tune the band gap\nand dispersion of the edge states in zigzag nanoribbons.\n"}
{"abstract": "  Tensor decomposition is a popular technique for tensor completion, However\nmost of the existing methods are based on linear or shallow model, when the\ndata tensor becomes large and the observation data is very small, it is prone\nto over fitting and the performance decreases significantly. To address this\nproblem, the completion method for a tensor based on a Biased Deep Tensor\nFactorization Network (BDTFN) is proposed. This method can not only overcome\nthe shortcomings of traditional tensor factorization, but also deal with\ncomplex non-linear data. Firstly, the horizontal and lateral tensors\ncorresponding to the observed values of the input tensors are used as inputs\nand projected to obtain their horizontal (lateral) potential feature tensors.\nSecondly, the horizontal (lateral) potential feature tensors are respectively\nconstructed into a multilayer perceptron network. Finally, the horizontal and\nlateral output tensors are fused by constructing a bilinear pooling layer.\nTensor forward-propagation is composed of those three step, and its parameters\nare updated by tensor back-propagation using the multivariable chain rule. In\nthis paper, we consider the large-scale 5-minute traffic speed data set and use\nit to address the missing data imputation problem for large-scale\nspatiotemporal traffic data. In addition, we compare the numerical performance\nof the proposed algorithm with those for state-of-the-art approaches on video\nrecovery and color image recovery. Numerical experimental results illustrate\nthat our approach is not only much more accurate than those state-of-the-art\nmethods, but it also has high speed.\n"}
{"abstract": "  Explaining to users why some items are recommended is critical, as it can\nhelp users to make better decisions, increase their satisfaction, and gain\ntheir trust in recommender systems (RS). However, existing explainable RS\nusually consider explanation as a side output of the recommendation model,\nwhich has two problems: (1) it is difficult to evaluate the produced\nexplanations because they are usually model-dependent, and (2) as a result, how\nthe explanations impact the recommendation performance is less investigated.\n  In this paper, explaining recommendations is formulated as a ranking task,\nand learned from data, similar to item ranking for recommendation. This makes\nit possible for standard evaluation of explanations via ranking metrics (e.g.,\nNDCG). Furthermore, we extend traditional item ranking to an item-explanation\njoint-ranking formalization to study if purposely selecting explanations could\nachieve certain learning goals, e.g., improving recommendation performance. A\ngreat challenge, however, is that the sparsity issue in the\nuser-item-explanation data would be inevitably severer than that in traditional\nuser-item interaction data, since not every user-item pair can be associated\nwith all explanations. To mitigate this issue, we propose to perform two sets\nof matrix factorization by considering the ternary relationship as two groups\nof binary relationships. Experiments on three large datasets verify our\nsolution's effectiveness on both item recommendation and explanation ranking.\nTo facilitate the development of explainable RS, we plan to make our code\npublicly available.\n"}
{"abstract": "  Short-lived radioactive nuclei (SLR) with mean lives below 100 Myr provide us\nwith unique insights into current galactic nucleosynthetic events, as well as\nevents that contributed to the material of our Solar System more that 4.6 Gyr\nago. Here we present a statistical analysis of the ratios of these radioactive\nnuclei at the time of early Solar System (ESS) using both analytical\nderivations and Monte Carlo methods. We aim to understand the interplay between\nthe production frequency and the mean lives of these isotopes, and its impact\non their theoretically predicted ratios in the interstellar medium (ISM). We\nfind that when the ratio of two SRLs, instead of the ratios of each single SLR\nrelative to its stable or long-lived isotope, is considered, not only the\nuncertainties related to the galactic chemical evolution of the stable isotope\nare completely eliminated, but also the statistical uncertainties are much\nlower. We identify four ratios, 247Cm/129I, 107Pd/182Hf, 97Tc/98Tc, and\n53Mn/97Tc, that have the potential to provide us with new insights into the r-,\ns-, and p-process nucleosynthesis at the time of the formation of the Sun, and\nneed to be studied using variable stellar yields. Additionally, the latter two\nratios need to be better determined in the ESS to allow us to fully exploit\nthem to investigate the galactic sites of the p process.\n"}
{"abstract": "  We consider the Gaussian free field $\\varphi$ on $\\mathbb{Z}^d$, for\n$d\\geq3$, and give sharp bounds on the probability that the radius of a finite\ncluster in the excursion set $\\{\\varphi \\geq h\\}$ exceeds a large value $N$,\nfor any height $h \\neq h_*$, where $h_*$ refers to the corresponding\npercolation critical parameter. In dimension $d=3$, we prove that this\nprobability is sub-exponential in $N$ and decays as\n$\\exp\\{-\\frac{\\pi}{6}(h-h_*)^2 \\frac{N}{\\log N} \\}$ as $N \\to \\infty$ to\nprincipal exponential order. When $d\\geq 4$, we prove that these tails decay\nexponentially in $N$. Our results extend to other quantities of interest, such\nas truncated two-point functions and the two-arms probability for annuli\ncrossings at scale N.\n"}
{"abstract": "  In this Letter, we introduce a paradigm to realize magneto-mechanical\nmetastructures inspired by multi-layer 2D materials, such as graphene bilayers.\nThe metastructures are intended to capture two aspects of their nanoscale\ncounterparts. One is the multi-layer geometry, which is implemented by stacking\nhexagonal lattice sheets. The other is the landscape of weak inter-layer\nforces, which is mimicked by the interactions between pairs of magnets located\nat corresponding lattice sites on adjacent layers. We illustrate the potential\nof this paradigm through a three-layer prototype. The two rigid outer lattices\nserve as control layers, while the thin inner layer is free to experience\nflexural motion under the confining action of the magnetic forces exchanged\nwith the outer ones, thus behaving as a lattice on elastic foundation. The\ninner layer is free to rotate relatively to the others, giving rise to a rich\nspectrum of inter-layer interaction patterns. Our objective is to determine how\nthe dynamical response can be tuned by changing the twist angle between the\nlayers. Specifically, we demonstrate experimentally that switching between\ndifferent stacking patterns has profound consequences on the phonon landscape,\nopening and closing bandgaps in different frequency regimes.\n"}
{"abstract": "  The infinite affine Lie algebras of type ABCD, also called\n$\\widehat{\\mathfrak{gl}}(\\infty)$, $\\widehat{\\mathfrak{o}}(\\infty)$,\n$\\widehat{\\mathfrak{sp}}(\\infty)$, are equivalent to subalgebras of the quantum\n$W_{1+\\infty}$ algebras. They have well-known representations on the Fock space\nof either a Dirac fermion ($\\hat A_\\infty$), a Majorana fermion ($\\hat\nB_\\infty$ and $\\hat D_\\infty$) or a symplectic boson ($\\hat C_\\infty$).\nExplicit formulas for the action of the quantum $W_{1+\\infty}$ subalgebras on\nthe Fock states are proposed for each representation. These formulas are the\nequivalent of the \\textit{vertical presentation} of the quantum toroidal\n$\\mathfrak{gl}(1)$ algebra Fock representation. They provide an alternative to\nthe fermionic and bosonic expressions of the \\textit{horizontal presentation}.\nFurthermore, these algebras are known to have a deep connection with symmetric\npolynomials. The action of the quantum $W_{1+\\infty}$ generators leads to the\nderivation of Pieri-like rules and q-difference equations for these\npolynomials. In the specific case of $\\hat B_\\infty$, a q-difference equation\nis obtained for $Q$-Schur polynomials indexed by strict partitions.\n"}
{"abstract": "  Similar to the case of solar system planets, reflected starlight from\nexoplanets is expected to be polarized due to atmospheric scattering and the\nnet disk integrated polarization should be non-zero owing to the asymmetrical\nillumination of the planetary disk. The computation of the disk-integrated\nreflected flux and its state of polarization involves techniques for the\ncalculation of the local reflection matrices as well as the numerical recipes\nfor integration over the planetary disks. In this paper, we present a novel\napproach to calculate the azimuth-dependent reflected intensity vectors at each\nlocation on the planetary disk divided into grids. We achieve this by solving\nthe vector radiative transfer equations that describe linear polarization. Our\ncalculations incorporate self-consistent atmospheric models of exoplanets over\na wide range of equilibrium temperature, surface gravity, atmospheric\ncomposition, and cloud structure. A comparison of the flux and the amount of\npolarization calculated by considering both single and multiple scattering\nexhibits the effect of depolarization due to multiple scattering of light\ndepending on the scattering albedo of the atmosphere. We have benchmarked our\nbasic calculations against some of the existing models. We have also presented\nour models for the hot Jupiter HD 189733 b, indicating the level of precision\nrequired by future observations to detect the polarization of this planet in\nthe optical and near-infrared wavelength region. The generic nature and the\naccuracy offered by our models make them an effective tool for modeling the\nfuture observations of the polarized light reflected from exoplanets.\n"}
{"abstract": "  The origins of thermodynamics from the microscopic properties of matter have\nnot been satisfactorily accounted for. This work presents a formulation that\nconnects Lagrangian mechanics to thermodynamics. By using such a formulation\nand following similar steps to those performed in the laboratory, the heat\ncapacities, energetic and mechanical response coefficients, absorbed and\nemitted heats, entropy changes, and thermodynamic energies of a prototype water\nsystem, employed as an illustrative example, are found to be close to the\nexperimental results on water bulk. The present formulation is realistic. After\nconnecting Lagrangian mechanics to Langevin stochastic dynamics, the first law\nof thermodynamics is derived. The formulation not only exhibits the\ntransformation of time-reversible equations onto time-irreversible equations,\nand the molecular origins of heat as well, but also reveals new pathways to\ninvestigate the thermodynamics of systems of nanometric sizes.\n"}
{"abstract": "  Image-text matching plays a critical role in bridging the vision and\nlanguage, and great progress has been made by exploiting the global alignment\nbetween image and sentence, or local alignments between regions and words.\nHowever, how to make the most of these alignments to infer more accurate\nmatching scores is still underexplored. In this paper, we propose a novel\nSimilarity Graph Reasoning and Attention Filtration (SGRAF) network for\nimage-text matching. Specifically, the vector-based similarity representations\nare firstly learned to characterize the local and global alignments in a more\ncomprehensive manner, and then the Similarity Graph Reasoning (SGR) module\nrelying on one graph convolutional neural network is introduced to infer\nrelation-aware similarities with both the local and global alignments. The\nSimilarity Attention Filtration (SAF) module is further developed to integrate\nthese alignments effectively by selectively attending on the significant and\nrepresentative alignments and meanwhile casting aside the interferences of\nnon-meaningful alignments. We demonstrate the superiority of the proposed\nmethod with achieving state-of-the-art performances on the Flickr30K and MSCOCO\ndatasets, and the good interpretability of SGR and SAF modules with extensive\nqualitative experiments and analyses.\n"}
{"abstract": "  Lightning casualties cause tremendous loss to life and property. However,\nvery lately lightning has been considered as one of the major natural\ncalamities which is now studied or monitored with proper instrumentation. The\nlightning characteristics over India have been studying by using daily data low\nresolution time series and monthly data high resolution monthly climatology. We\nhave used ANN time series method (a neural network) to analyze the time series\nand defined which one will be the best predictor of lightning over India. The\ntime series of lightning is output(dependent) and input (independent) are\nk-index, AOD, Cape etc. The Gaussian process regression, support vector\nmachine, regression trees and linear regression defined the input variables.\nWhich show approximately linear relation.\n"}
{"abstract": "  In this paper, we introduce a new concept, namely $\\epsilon$-arithmetics, for\nreal vectors of any fixed dimension. The basic idea is to use vectors of\nrational values (called rational vectors) to approximate vectors of real values\nof the same dimension within $\\epsilon$ range. For rational vectors of a fixed\ndimension $m$, they can form a field that is an $m$th order extension\n$\\mathbf{Q}(\\alpha)$ of the rational field $\\mathbf{Q}$ where $\\alpha$ has its\nminimum polynomial of degree $m$ over $\\mathbf{Q}$. Then, the arithmetics, such\nas addition, subtraction, multiplication, and division, of real vectors can be\ndefined by using that of their approximated rational vectors within $\\epsilon$\nrange. We also define complex conjugate of a real vector and then inner\nproducts for two real vectors and two real vector sequences of the same length.\n"}
{"abstract": "  An increasing share of captured images and videos are transmitted for storage\nand remote analysis by computer vision algorithms, rather than to be viewed by\nhumans. Contrary to traditional standard codecs with engineered tools, neural\nnetwork based codecs can be trained end-to-end to optimally compress images\nwith respect to a target rate and any given differentiable performance metric.\nAlthough it is possible to train such compression tools to achieve better\nrate-accuracy performance for a particular computer vision task, it could be\npractical and relevant to re-use the compressed bit-stream for multiple machine\ntasks. For this purpose, we introduce 'Connectors' that are inserted between\nthe decoder and the task algorithms to enable a direct transformation of the\ncompressed content, which was previously optimized for a specific task, to\nmultiple other machine tasks. We demonstrate the effectiveness of the proposed\nmethod by achieving significant rate-accuracy performance improvement for both\nimage classification and object segmentation, using the same bit-stream,\noriginally optimized for object detection.\n"}
{"abstract": "  We prove upper bounds on the one-arm exponent $\\eta_1$ for dependent\npercolation models; while our main interest is level set percolation of smooth\nGaussian fields, the arguments apply to other models in the Bernoulli\npercolation universality class, including Poisson-Voronoi and Poisson-Boolean\npercolation. More precisely, in dimension $d=2$ we prove $\\eta_1 \\le 1/3$ for\nGaussian fields with rapid correlation decay (e.g.\\ the Bargmann-Fock field),\nand in general dimensions we prove $\\eta_1 \\le d/3$ for finite-range fields and\n$\\eta_1 \\le d-2$ for fields with rapid correlation decay. Although these\nresults are classical for Bernoulli percolation (indeed they are best-known in\ngeneral), existing proofs do not extend to dependent percolation models, and we\ndevelop a new approach based on exploration and relative entropy arguments. We\nalso establish a new Russo-type inequality for smooth Gaussian fields which we\nuse to prove the sharpness of the phase transition for finite-range fields.\n"}
{"abstract": "  Due to real-time image semantic segmentation needs on power constrained edge\ndevices, there has been an increasing desire to design lightweight semantic\nsegmentation neural network, to simultaneously reduce computational cost and\nincrease inference speed. In this paper, we propose an efficient asymmetric\ndilated semantic segmentation network, named EADNet, which consists of multiple\ndeveloped asymmetric convolution branches with different dilation rates to\ncapture the variable shapes and scales information of an image. Specially, a\nmulti-scale multi-shape receptive field convolution (MMRFC) block with only a\nfew parameters is designed to capture such information. Experimental results on\nthe Cityscapes dataset demonstrate that our proposed EADNet achieves\nsegmentation mIoU of 67.1 with smallest number of parameters (only 0.35M) among\nmainstream lightweight semantic segmentation networks.\n"}
{"abstract": "  We characterize the optical coherence and energy-level properties of the 795\nnm $^3$H$_6$ to $^3$H$_4$ transition of Tm$^{3+}$ in a Ti$^{4+}$:LiNbO$_{3}$\nwaveguide at temperatures as low as 0.65 K. Coherence properties are measured\nwith varied temperature, magnetic field, optical excitation power and\nwavelength, and measurement time-scale. We also investigate nuclear\nspin-induced hyperfine structure and population dynamics with varying magnetic\nfield and laser excitation power. Except for accountable differences due to\ndifference Ti$^{4+}$ and Tm$^{3+}$-doping concentrations, we find that the\nproperties of Tm$^{3+}$:Ti$^{4+}$:LiNbO$_{3}$ produced by indiffusion doping\nare consistent with those of a bulk-doped Tm$^{3+}$:LiNbO$_{3}$ crystal\nmeasured under similar conditions. Our results, which complement previous work\nin a narrower parameter space, support using rare-earth-ions for integrated\noptical and quantum signal processing.\n"}
{"abstract": "  We present a scattering model for nuclei with similar masses. In this\nthree-body model, the projectile has a core+valence structure, whereas the\ntarget is identical to the core nucleus. The three-body wave functions must be\nsymmetrized for the exchange of the cores. This property gives rise to\nnon-local potentials, which are computed without approximation. The present\nmodel is an extension of the Continuum Discretized Coupled Channel (CDCC)\nformalism, with an additional treatment of core exchange. We solve the\ncoupled-channel system, including non-local terms, by the $R$-matrix method\nusing Lagrange functions. This model is applied to the $^{13}{\\rm C}+^{12}$C,\n$^{13}{\\rm N}+^{12}$C and $^{16}{\\rm O}+^{12}$C systems. Experimental\nscattering cross sections are fairly well reproduced without any parameter\nfitting. The backward-angle enhancement of the elastic cross sections is due to\nthe non-local potential. We discuss in more detail the various non-local\ncontributions and present effective local potentials.\n"}
{"abstract": "  Privacy-constrained source coding problems have become increasingly important\nrecently, and the utility-privacy trade-off has been investigated for various\nsystems. As pioneering work, Yamamoto (1983) found theoretical limits of the\ncoding rate, privacy and utility in two cases; (i) both public and private\ninformation is encoded and (ii) only public information is encoded. However,\nthe theoretical limit has not been characterized in a more general case; (iii)\nencoded messages consist of public information and a part of private\ninformation. Then in this paper, we characterize the trade-off relation in case\n(iii). The obtained expression of the achievable region is a \"unified\nexpression\" because it includes the ones in case (i) and (ii) as special cases.\nThrough the numerical results, we clarify the significance of the choice of the\nencoded set.\n"}
{"abstract": "  We survey some recent developments in the analytic theory of multiple\nDirichlet series with arithmetical coefficients on the numerators.\n"}
{"abstract": "  The non-equilibrium densities of nonlocal mass-energy are self-governed by\nkinetic stresses toward quasi-equilibrium sub-configurations. System energy\nintegral of continuous matter-extension coordinates its adaptive densities on\neach hierarchic level of structural sub-organizations. The logarithmic\npotential for metric fields, kinetic densities, and local stresses in the\nnonlocal distribution of mass-energy corresponds to the Shannon fundamental\nlimit for information rates. The phenomenological pulls of Newtonian gravity\ncan be quantitatively replaced by volume kinetic pushes from local stresses of\ncontinuous space-matter with moving inertial densities. Thus, the observable\nphenomenon of gravitation between volumetric parts of distributed mass-energy\nis interpreted in this Cartesian nonlocality of self-governed material space\nthrough correlated accelerations of continuous kinetic densities with local\ninertia. The gravitational palliative of negative potentials is replaced with\nthe monistic approach to Cartesian matter through always positive kinetic\nmass-energy.\n"}
{"abstract": "  Understanding voluminous historical records provides clues on the past in\nvarious aspects, such as social and political issues and even natural science\nfacts. However, it is generally difficult to fully utilize the historical\nrecords, since most of the documents are not written in a modern language and\npart of the contents are damaged over time. As a result, restoring the damaged\nor unrecognizable parts as well as translating the records into modern\nlanguages are crucial tasks. In response, we present a multi-task learning\napproach to restore and translate historical documents based on a\nself-attention mechanism, specifically utilizing two Korean historical records,\nones of the most voluminous historical records in the world. Experimental\nresults show that our approach significantly improves the accuracy of the\ntranslation task than baselines without multi-task learning. In addition, we\npresent an in-depth exploratory analysis on our translated results via topic\nmodeling, uncovering several significant historical events.\n"}
{"abstract": "  Unsupervised Domain adaptation is an effective method in addressing the\ndomain shift issue when transferring knowledge from an existing richly labeled\ndomain to a new domain. Existing manifold-based methods either are based on\ntraditional models or largely rely on Grassmannian manifold via minimizing\ndifferences of single covariance matrices of two domains. In addition, existing\npseudo-labeling algorithms inadequately consider the quality of pseudo labels\nin aligning the conditional distribution between two domains. In this work, a\ndeep spherical manifold Gaussian kernel (DSGK) framework is proposed to map the\nsource and target subspaces into a spherical manifold and reduce the\ndiscrepancy between them by embedding both extracted features and a Gaussian\nkernel. To align the conditional distributions, we further develop an\neasy-to-hard pseudo label refinement process to improve the quality of the\npseudo labels and then reduce categorical spherical manifold Gaussian kernel\ngeodesic loss. Extensive experimental results show that DSGK outperforms\nstate-of-the-art methods, especially on challenging cross-domain learning\ntasks.\n"}
{"abstract": "  The stratified proportional hazards model represents a simple solution to\naccount for heterogeneity within the data while keeping the multiplicative\neffect on the hazard function. Strata are typically defined a priori by\nresorting to the values taken by a categorical covariate. A general framework\nis proposed, which allows for the stratification of a generic accelerated life\ntime model, including as a special case the Weibull proportional hazard model.\nThe stratification is determined a posteriori by taking into account that\nstrata might be characterized by different baseline survivals as well as\ndifferent effects of the predictors. This is achieved by considering a Bayesian\nnonparametric mixture model and the posterior distribution it induces on the\nspace of data partitions. The optimal stratification is then identified by\nmeans of the variation of information criterion and, in turn, stratum-specific\ninference is carried out. The performance of the proposed method and its\nrobustness to the presence of right-censored observations are investigated by\nmeans of an extensive simulation study. A further illustration is provided by\nthe analysis of a data set extracted from the University of Massachusetts AIDS\nResearch Unit IMPACT Study.\n"}
{"abstract": "  For a given K-polystable Fano manifold X and a natural number l, we show that\nthere exists a rational number 0 < c < 1 depending only on the dimension of X,\nsuch that $D\\in |-lK_X|$ is GIT-(semi/poly)stable under the action of Aut(X) if\nand only if the pair $(X, \\frac{\\epsilon}{l} D)$ is K-(semi/poly)stable for\nsome rational $0 < {\\epsilon} < c$.\n"}
{"abstract": "  Neural network pruning is an essential approach for reducing the\ncomputational complexity of deep models so that they can be well deployed on\nresource-limited devices. Compared with conventional methods, the recently\ndeveloped dynamic pruning methods determine redundant filters variant to each\ninput instance which achieves higher acceleration. Most of the existing methods\ndiscover effective sub-networks for each instance independently and do not\nutilize the relationship between different inputs. To maximally excavate\nredundancy in the given network architecture, this paper proposes a new\nparadigm that dynamically removes redundant filters by embedding the manifold\ninformation of all instances into the space of pruned networks (dubbed as\nManiDP). We first investigate the recognition complexity and feature similarity\nbetween images in the training set. Then, the manifold relationship between\ninstances and the pruned sub-networks will be aligned in the training\nprocedure. The effectiveness of the proposed method is verified on several\nbenchmarks, which shows better performance in terms of both accuracy and\ncomputational cost compared to the state-of-the-art methods. For example, our\nmethod can reduce 55.3% FLOPs of ResNet-34 with only 0.57% top-1 accuracy\ndegradation on ImageNet.\n"}
{"abstract": "  To realize autonomous shipping, autonomous berthing and unberthing are some\nof the technical challenges. In the past, numerous research have been done on\nthe optimization of trajectory planning of berthing problems. However, these\nstudies assumed only a simple berth and did not consider obstacles.\nOptimization of trajectory planning on berthing and unberthing in actual ports\nmust consider the spatial constraints and maintain sufficient distance to\nobstacles. The main contributions of this study are as follows: (i) a collision\navoidance algorithm based on the ship domain which has variable size by the\nship speed is proposed, to include the spatial constraints to optimization;\n(ii) the effect of wind disturbance is taken into account to the trajectory\nplanning to make a feasible trajectory based on the capacity limit of\nactuators; (iii) showing that the optimization method for berthing is also\neligible for the unberthing, which has been almost neglected; (iv) waypoints\nare included to the optimization process, to make optimization easier on\npractical applications. The authors tested the proposed method on two existing\nports. The proposed method performed well on both the berthing and the\nunberthing problem and optimized the control input and the trajectory while\navoiding collision with the complex obstacles.\n"}
{"abstract": "  We report a systematic study of the superconducting (SC) and normal-state\nanisotropy of Fe$_{1+y}$Te$_{0.6}$Se$_{0.4}$ single crystals with controlled\namounts of excess Fe ($y$ = 0, 0.07, and 0.14). The SC state anisotropy\n$\\gamma_{H}$ was obtained by measuring the upper critical fields under high\nmagnetic fields over 50 T for both $H\\parallel ab$ and $H\\parallel c$. On the\nother hand, the normal state anisotropy $\\gamma_{\\rho}$ was obtained by\nmeasuring the resistivity with current flowing in the $ab$ plane ($\\rho_{ab}$)\nand along the $c$ axis ($\\rho_c$). To precisely measure $\\rho_{ab}$ and\n$\\rho_c$ in the same part of a specimen avoiding the variation dependent on\npieces or parts, we adopt a new method using a micro-fabricated bridge with an\nadditional neck part along $c$ axis. The $\\gamma_{H}$ decreases from a value\ndependent on the amount of excess Fe at $T_{\\rm{c}}$ to a common value $\\sim$ 1\nat 2 K. The different $\\gamma_{H}$ at $T_{\\rm{c}}$ ($\\sim$1.5 for $y$ = 0, and\n2.5 for $y$ = 0.14) suggests that the anisotropy of effective mass\n$m_c^*/m_{ab}^*$ increases from $\\sim$ 2.25 ($y$ = 0) to 6.25 ($y$ = 0.14) with\nthe excess Fe. The almost isotropic $\\gamma_{H}$ at low temperatures is due to\nthe strong spin paramagnetic effect at $H\\parallel ab$. By contrast, the\n$\\gamma_{\\rho}$ shows a much larger value of $\\sim$ 17 ($y$ = 0) to $\\sim$ 50\n($y$ = 0.14) at the temperature just above $T_{\\rm{c}}$. Combined the results\nof $\\gamma_{H}$ and $\\gamma_{\\rho}$ near $T_{\\rm{c}}$, we found out that the\ndiscrepant anisotropies between the SC and normal states originates from a\nlarge anisotropy of scattering time $\\tau_{ab}$/$\\tau_c$ $\\sim$ 7.8. The\n$\\tau_{ab}$/$\\tau_c$ is found to be independent of the excess Fe.\n"}
{"abstract": "  We look for solutions to the Schr\\\"{o}dinger-Poisson-Slater equation $$-\n\\Delta u + \\lambda u - \\gamma (|x|^{-1} * |u|^2) u - a |u|^{p-2}u = 0 \\quad\n\\text{in} \\quad \\mathbb{R}^3, $$ which satisfy \\begin{equation*}\n\\int_{\\mathbb{R}^3}|u|^2 \\, dx = c \\end{equation*} for some prescribed $c>0$.\nHere $ u \\in H^1(\\mathbb{R}^3)$, $\\gamma \\in \\mathbb{R},$ $ a \\in \\mathbb{R}$\nand $p \\in (\\frac{10}{3}, 6]$. When $\\gamma >0$ and $a > 0$, both in the\nSobolev subcritical case $p \\in (\\frac{10}{3}, 6)$ and in the Sobolev critical\ncase $p=6$, we show that there exists a $c_1>0$ such that, for any $c \\in\n(0,c_1)$, the equation admits two solutions $u_c^+$ and $u_c^-$ which can be\ncharacterized respectively as a local minima and as a mountain pass critical\npoint of the associated {\\it Energy} functional restricted to the norm\nconstraint. In the case $\\gamma >0$ and $a < 0$, we show that, for any $p \\in\n(\\frac{10}{3},6]$ and any $c>0$, the equation admits a solution which is a\nglobal minimizer. Finally, in the case $\\gamma <0$, $a >0$ and $p=6$ we show\nthat it does not admit positive solutions.\n"}
{"abstract": "  Graph is an usual representation of relational data, which are ubiquitous in\nmanydomains such as molecules, biological and social networks. A popular\napproach to learningwith graph structured data is to make use of graph kernels,\nwhich measure the similaritybetween graphs and are plugged into a kernel\nmachine such as a support vector machine.Weisfeiler-Lehman (WL) based graph\nkernels, which employ WL labeling scheme to extract subtree patterns and\nperform node embedding, are demonstrated to achieve great performance while\nbeing efficiently computable. However, one of the main drawbacks of ageneral\nkernel is the decoupling of kernel construction and learning process. For\nmoleculargraphs, usual kernels such as WL subtree, based on substructures of\nthe molecules, consider all available substructures having the same importance,\nwhich might not be suitable inpractice. In this paper, we propose a method to\nlearn the weights of subtree patterns in the framework of WWL kernels, the\nstate of the art method for graph classification task [14]. To overcome the\ncomputational issue on large scale data sets, we present an efficient learning\nalgorithm and also derive a generalization gap bound to show its convergence.\nFinally, through experiments on synthetic and real-world data sets, we\ndemonstrate the effectiveness of our proposed method for learning the weights\nof subtree patterns.\n"}
{"abstract": "  We study the entanglement islands and subsystem volume complexity\ncorresponding to the left/ right entanglement of a conformal defect in\n$d$-dimensions in Randall-Sundrum (RS) braneworld model with subcritical\ntension brane. The left and right modes of the defect mimic the eternal black\nhole and radiation system respectively. Hence the entanglement entropy between\nthe two follows an eternal black hole Page curve which is unitarity compatible.\nWe compute the volumes corresponding to the left and right branes with\npreferred Ryu-Takanayagi (RT) surfaces at different times, which provide a\nprobe of the subregion complexity of the black hole and the radiation states\nrespectively. An interesting jump in volume is found at Page time, where the\nentanglement curve is saturated due to the inclusion of the island surfaces. We\nexplain various possibilities of this phase transition in complexity at Page\ntime and argue how these results match with a covariant proposal qualitatively.\n"}
{"abstract": "  Frequency Response Functions (FRFs) are one of the cornerstones of musical\nacoustic experimental research. They describe the way in which musical\ninstruments vibrate in a wide range of frequencies and are used to predict and\nunderstand the acoustic differences between them. In the specific case of\nstringed musical instruments such as violins, FRFs evaluated at the bridge are\nknown to capture the overall body vibration. These indicators, also called\nbridge admittances, are widely used in the literature for comparative analyses.\nHowever, due to their complex structure they are rather difficult to\nquantitatively compare and study. In this manuscript we present a way to\nquantify differences between FRFs, in particular violin bridge admittances,\nthat separates the effects in frequency, amplitude and quality factor of the\nfirst resonance peaks characterizing the responses. This approach allows us to\ndefine a distance between FRFs and clusterise measurements according to this\ndistance. We use two case studies, one based on Finite Element Analysis and\nanother exploiting measurements on real violins, to prove the effectiveness of\nsuch representation. In particular, for simulated bridge admittances the\nproposed distance is able to highlight the different impact of consecutive\nsimulation `steps' on specific vibrational properties and, for real violins,\ngives a first insight on similar styles of making, as well as opposite ones.\n"}
{"abstract": "  Deep reinforcement learning (DRL) algorithms have been demonstrated to be\neffective in a wide range of challenging decision making and control tasks.\nHowever, these methods typically suffer from severe action oscillations in\nparticular in discrete action setting, which means that agents select different\nactions within consecutive steps even though states only slightly differ. This\nissue is often neglected since the policy is usually evaluated by its\ncumulative rewards only. Action oscillation strongly affects the user\nexperience and can even cause serious potential security menace especially in\nreal-world domains with the main concern of safety, such as autonomous driving.\nTo this end, we introduce Policy Inertia Controller (PIC) which serves as a\ngeneric plug-in framework to off-the-shelf DRL algorithms, to enables adaptive\ntrade-off between the optimality and smoothness of the learned policy in a\nformal way. We propose Nested Policy Iteration as a general training algorithm\nfor PIC-augmented policy which ensures monotonically non-decreasing updates\nunder some mild conditions. Further, we derive a practical DRL algorithm,\nnamely Nested Soft Actor-Critic. Experiments on a collection of autonomous\ndriving tasks and several Atari games suggest that our approach demonstrates\nsubstantial oscillation reduction in comparison to a range of commonly adopted\nbaselines with almost no performance degradation.\n"}
{"abstract": "  Building a dialogue system that can communicate naturally with humans is a\nchallenging yet interesting problem of agent-based computing. The rapid growth\nin this area is usually hindered by the long-standing problem of data scarcity\nas these systems are expected to learn syntax, grammar, decision making, and\nreasoning from insufficient amounts of task-specific dataset. The recently\nintroduced pre-trained language models have the potential to address the issue\nof data scarcity and bring considerable advantages by generating contextualized\nword embeddings. These models are considered counterpart of ImageNet in NLP and\nhave demonstrated to capture different facets of language such as hierarchical\nrelations, long-term dependency, and sentiment. In this short survey paper, we\ndiscuss the recent progress made in the field of pre-trained language models.\nWe also deliberate that how the strengths of these language models can be\nleveraged in designing more engaging and more eloquent conversational agents.\nThis paper, therefore, intends to establish whether these pre-trained models\ncan overcome the challenges pertinent to dialogue systems, and how their\narchitecture could be exploited in order to overcome these challenges. Open\nchallenges in the field of dialogue systems have also been deliberated.\n"}
{"abstract": "  Motivated by the recent successful formation of the MoSi2N4 monolayer [Hong\net al., Sci. 369, 670 (2020)], the structural, electronic and magnetic\nproperties of MoSi2N4 nanoribbons (NRs) is investigated for the first time .\nThe band structure calculations showed spin-polarization in zigzag edges and a\nnon-magnetic semiconducting character in armchair edges. For armchair-edges, we\nidentify an indirect to direct band gap shift compared to the MoSi2N4\nmonolayer, and its energy gap increases with increasing NR width. Anisotropic\nelectrical and magnetic behavior is observed via band structure calculations in\nthe zigzag and armchair edges, where, surprisingly, for the one type of\nzigzag-edges configuration, we identify a Dirac-semimetal character. The\nappearance of magnetism and Dirac-semimetal in MoSi2N4 ribbon can give rise to\nnovel physical properties, which could be useful in applications for\nnext-generation electronic devices.\n"}
{"abstract": "  With the advent of next generation high resolution telescopes, our\nunderstanding of how the magnetic field is organized in the internetwork (IN)\nphotosphere is likely to advance.We aim to evaluate the extent to which we can\nretrieve information about the magnetic vector in the IN photosphere using\ninversions. We use snapshots produced from high resolution 3D\nmagnetohydrodynamic (MHD) simulations and employ the Stokes Inversions based on\nResponse functions (SIR) code to produce synthetic observables in the near\ninfrared spectral window observed by the GREGOR Infrared Spectrograph (GRIS),\nwhich contains the highly magnetically sensitive photospheric Fe I line pair at\n15648.52 A and 15652.87 A. We perform nearly 14 million inversions to test how\nwell the true MHD atmospheric parameters can be constrained. Finally, we\ndegrade the synthetic Stokes vectors spectrally and spatially to GREGOR\nresolutions and examine how this influences observations, considering the\nimpact of stray light, spatial resolution and signal-to-noise (S-to-N). We find\nthe depth-averaged parameters can be recovered by the inversions of the\nundegraded profiles, and by adding gradients to magnetic field strength,\ninclination and line of sight velocity we show an improvement in the chi\nsquared value is achieved. We evaluate the extent to which we can constrain\nthese parameters at various optical depths, with the kinematic and\nthermodynamic parameters sensitive deeper in the atmosphere than the magnetic\nparameters. We find the S-to-N and spatial resolution play a significant role\nin determining how the atmosphere appears and the magnetic and kinematic\nparameters are invariant upon inclusion of unpolarized stray light. We studied\na linear polarization feature which resembles those recently observed by GRIS,\nappearing as loop-like structures with similar magnetic flux density.\n"}
{"abstract": "  We deduce a microstructure inspired model for humidity and temperature\neffects on the mechanical response of spider silks, modelled as a composite\nmaterial with a hard crystalline and a soft amorphous region. Water molecules\ndecrease the percentage of crosslinks in the softer region inducing a variation\nof natural configuration of the macromolecules. The resulting kinematic\nincompatibility between the regions crucially influences the final mechanical\nresponse. We demonstrate the predictivity of the model by quantitatively\nreproducing the experimentally observed behavior\n"}
{"abstract": "  This paper investigates whether the power of the models pre-trained on text\ndata, such as BERT, can be transferred to general token sequence classification\napplications. To verify pre-trained models' transferability, we test the\npre-trained models on text classification tasks with meanings of tokens\nmismatches, and real-world non-text token sequence classification data,\nincluding amino acid, DNA, and music. We find that even on non-text data, the\nmodels pre-trained on text converge faster, perform better than the randomly\ninitialized models, and only slightly worse than the models using task-specific\nknowledge. We also find that the representations of the text and non-text\npre-trained models share non-trivial similarities.\n"}
{"abstract": "  The ability to train randomly initialised deep neural networks is known to\ndepend strongly on the variance of the weight matrices and biases as well as\nthe choice of nonlinear activation. Here we complement the existing geometric\nanalysis of this phenomenon with an information theoretic alternative. Lower\nbounds are derived for the mutual information between an input and hidden layer\noutputs. Using a mean field analysis we are able to provide analytic lower\nbounds as functions of network weight and bias variances as well as the choice\nof nonlinear activation. These results show that initialisations known to be\noptimal from a training point of view are also superior from a mutual\ninformation perspective.\n"}
{"abstract": "  The levels of heavy elements in stars are the product of enhancement by\nprevious stellar generations, and the distribution of this metallicity among\nthe population contains clues to the process by which a galaxy formed. Most\nfamously, the \"G-dwarf problem\" highlighted the small number of low-metallicity\nG-dwarf stars in the Milky Way, which is inconsistent with the simplest picture\nof a galaxy formed from a \"closed box\" of gas. It can be resolved by treating\nthe Galaxy as an open system that accretes gas throughout its life. This\nobservation has classically only been made in the Milky Way, but the\navailability of high-quality spectral data from SDSS-IV MaNGA and the\ndevelopment of new analysis techniques mean that we can now make equivalent\nmeasurements for a large sample of spiral galaxies. Our analysis shows that\nhigh-mass spirals generically show a similar deficit of low-metallicity stars,\nimplying that the Milky Way's history of gas accretion is common. By contrast,\nlow-mass spirals show little sign of a G-dwarf problem, presenting the\nmetallicity distribution that would be expected if such systems evolved as\npretty much closed boxes. This distinction can be understood from the differing\ntimescales for star formation in galaxies of differing masses.\n"}
{"abstract": "  Raychaudhuri equation is derived by assuming geometric flow in spacetime M of\nn+1 dimensions. The equation turns into a harmonic oscillator form under\nsuitable transformations.Thereby a relation between geometrical entropy and\nmean geodesic deviation is established. This has a connection to chaos theory\nwhere the trajectories diverge exponentially. We discuss its application to\ncosmology and black holes. Thus we present a connection between chaos theory\nand general relativity.\n"}
{"abstract": "  We identify a class of \"semi-modular\" forms invariant on special subgroups of\n$GL_2(\\mathbb Z)$, which includes classical modular forms together with\ncomplementary classes of functions that are also nice in a specific sense. We\ndefine an Eisenstein-like series summed over integer partitions, and use it to\nconstruct families of semi-modular forms.\n"}
{"abstract": "  A finite difference algorithm based on the integral Laguerre transform in\ntime for solving a three-dimensional one-way wave equation is proposed. This\nallows achieving high accuracy of calculation results. In contrast to the\nFourier method, the approach does not need to solve systems of linear algebraic\nequations with indefinite matrices. To filter the unstable components of a wave\nfield, Richardson extrapolation or spline approximation can be used. However,\nthese methods impose additional limitations on the integration step in depth.\nThis problem can be solved if the filtering is performed not in the direction\nof extrapolation of the wave field, but in a horizontal plane. This approach\ncalled for fast methods of converting the Laguerre series coefficients into the\nFourier series coefficients and vice versa. The high stability of the new\nalgorithm allows calculations with a large depth step without loss of accuracy\nand, in combination with Marchuk-Strang splitting, this can significantly\nreduce the calculation time. Computational experiments are performed. The\nresults have shown that this algorithm is highly accurate and efficient in\nsolving the problems of seismic migration.\n"}
{"abstract": "  Permutation statistics $\\wnm$ and $\\rlm$ are both arising from permutation\ntableaux. $\\wnm$ was introduced by Chen and Zhou, which was proved equally\ndistributed with the number of unrestricted rows of a permutation tableau.\nWhile $\\rlm$ is showed by Nadeau equally distributed with the number of $1$'s\nin the first row of a permutation tableau.\n  In this paper, we investigate the joint distribution of $\\wnm$ and $\\rlm$.\nStatistic $(\\rlm,\\wnm,\\rlmin,\\des,(\\underline{321}))$ is shown equally\ndistributed with $(\\rlm,\\rlmin,\\wnm,\\des,(\\underline{321}))$ on $S_n$. Then the\ngenerating function of $(\\rlm,\\wnm)$ follows. An involution is constructed to\nexplain the symmetric property of the generating function. Also, we study the\ntriple statistic $(\\wnm,\\rlm,\\asc)$, which is shown to be equally distributed\nwith $(\\rlmax-1,\\rlmin,\\asc)$ as studied by Josuat-Verg$\\grave{e}$s. The main\nmethod we adopt throughout the paper is constructing bijections based on a\nblock decomposition of permutations.\n"}
{"abstract": "  The paper presents a novel model-based method for intelligent tutoring, with\nparticular emphasis on the problem of selecting teaching interventions in\ninteraction with humans. Whereas previous work has focused on either\npersonalization of teaching or optimization of teaching intervention sequences,\nthe proposed individualized model-based planning approach represents\nconvergence of these two lines of research. Model-based planning picks the best\ninterventions via interactive learning of a user memory model's parameters. The\napproach is novel in its use of a cognitive model that can account for several\nkey individual- and material-specific characteristics related to\nrecall/forgetting, along with a planning technique that considers users'\npractice schedules. Taking a rule-based approach as a baseline, the authors\nevaluated the method's benefits in a controlled study of artificial teaching in\nsecond-language vocabulary learning (N=53).\n"}
{"abstract": "  An indispensable component in task-oriented dialogue systems is the dialogue\nstate tracker, which keeps track of users' intentions in the course of\nconversation. The typical approach towards this goal is to fill in multiple\npre-defined slots that are essential to complete the task. Although various\ndialogue state tracking methods have been proposed in recent years, most of\nthem predict the value of each slot separately and fail to consider the\ncorrelations among slots. In this paper, we propose a slot self-attention\nmechanism that can learn the slot correlations automatically. Specifically, a\nslot-token attention is first utilized to obtain slot-specific features from\nthe dialogue context. Then a stacked slot self-attention is applied on these\nfeatures to learn the correlations among slots. We conduct comprehensive\nexperiments on two multi-domain task-oriented dialogue datasets, including\nMultiWOZ 2.0 and MultiWOZ 2.1. The experimental results demonstrate that our\napproach achieves state-of-the-art performance on both datasets, verifying the\nnecessity and effectiveness of taking slot correlations into consideration.\n"}
{"abstract": "  This paper is concerned with weak solutions {e,h} in L^2 x L^2 of the\ntime-dependent Maxwell equations. We show that these solutions obey an energy\nequality. Our method of proof is based on the approximation of {e,h} by its\nSteklov mean with respect to time t. This approximation technique is well-known\nfor establishing integral estimates for weak solutions of parabolic equations.\nIn addition we prove the uniqueness of {e,h}.\n"}