Datasets:

J0nasW

/

paperswithcode

like 1

bedae3bf-97bb-4ee3-83a4-4ae995c45841

mtgflow-unsupervised-multivariate-time-series

2208.02108

https://arxiv.org/abs/2208.02108v3

https://arxiv.org/pdf/2208.02108v3.pdf

Detecting Multivariate Time Series Anomalies with Zero Known Label

Multivariate time series anomaly detection has been extensively studied under the semi-supervised setting, where a training dataset with all normal instances is required. However, preparing such a dataset is very laborious since each single data instance should be fully guaranteed to be normal. It is, therefore, desired to explore multivariate time series anomaly detection methods based on the dataset without any label knowledge. In this paper, we propose MTGFlow, an unsupervised anomaly detection approach for multivariate time series anomaly detection via dynamic graph and entity-aware normalizing flow, leaning only on a widely accepted hypothesis that abnormal instances exhibit sparse densities than the normal. However, the complex interdependencies among entities and the diverse inherent characteristics of each entity pose significant challenges on the density estimation, let alone to detect anomalies based on the estimated possibility distribution. To tackle these problems, we propose to learn the mutual and dynamic relations among entities via a graph structure learning model, which helps to model accurate distribution of multivariate time series. Moreover, taking account of distinct characteristics of the individual entities, an entity-aware normalizing flow is developed to describe each entity into a parameterized normal distribution, thereby producing fine-grained density estimation. Incorporating these two strategies, MTGFlow achieves superior anomaly detection performance. Experiments on five public datasets with seven baselines are conducted, MTGFlow outperforms the SOTA methods by up to 5.0 AUROC\%. Codes will be released at https://github.com/zqhang/Detecting-Multivariate-Time-Series-Anomalies-with-Zero-Known-Label.

['Wenchao Meng', 'Shibo He', 'Haoyu Liu', 'Jiming Chen', 'Qihang Zhou']

2022-08-03

['graph-structure-learning']

['graphs']

[-0.09098624 -0.06473374 0.0522949 -0.34641436 -0.44929737 -0.4874422 0.48225737 0.7311957 -0.01152911 0.38658112 -0.22248608 -0.3879252 -0.26564574 -0.9054872 -0.56953806 -0.89024776 -0.5621235 0.54276323 0.0421562 0.04174276 -0.03845495 0.389207 -1.3427432 -0.28999922 1.259354 1.1539416 -0.36026028 0.49627158 -0.3036324 0.5324986 -0.5840064 -0.3905841 0.13332269 -0.35983485 -0.31962803 0.21849403 0.1587733 -0.1777922 -0.2872286 1.2732714 0.12966257 0.17278038 0.96136534 -1.6724336 -0.3020852 0.52641326 -0.9111562 0.68057925 0.18327744 0.0495397 1.1646767 -0.6332961 0.07128046 0.8719095 0.4023267 0.1732298 -1.2435973 -0.67946076 0.46600437 0.2469016 -1.5776889 -0.10157215 0.9382021 -0.40278542 0.62522125 0.34116828 0.4227658 1.0840966 0.01734181 0.5985349 0.42388418 -0.08738092 0.2838933 -0.08095498 0.203937 0.49257636 0.5313842 -0.21723242 -0.21061379 -0.5868374 0.39355335 0.52817523 -0.19186614 -0.34719896 -0.8197954 0.6404516 0.2247759 0.44943383 -0.4613105 -0.29299703 0.6128722 0.3755517 0.79525447 0.08149368 -0.4792434 -0.14926524 -0.631478 0.13190043 0.67501307 0.9209609 0.67445064 0.3746071 -0.01485397 0.7405695 0.35914904 0.5352158 0.41668972 -0.13454752 0.63527715 0.9389365 -0.17660981 -1.48507 -0.43686622 -0.6325853 -1.3068806 -0.3449299 0.62780446 -0.21396956 -0.74427193 1.7926837 0.6240247 0.7337157 -0.16107136 0.55125576 0.36620018 0.6879646 0.20168875 -0.4007036 1.0940615 -0.3277568 -0.7175874 0.10540832 1.0151933 -0.3879872 0.8903766 0.31612948 -0.46010023 -0.02140946 -0.7583954 0.5933243 -0.41188383 -0.11150541 0.47284728 0.5695152 -0.5775954 0.481587 -1.1139315 -0.3295626 0.41377163 0.16309132 -0.46455055 -0.07735949 -1.1849363 0.34996885 0.618821 0.09331898 -0.5398055 -0.7735152 -1.0551466 0.13319725 0.53637505 -0.31535375 0.79723465 -0.6453327 -0.7131032 0.39507377 -0.16894193 -0.6197317 0.47821924 -0.04036773 -0.97374696 0.01706407 0.11303373 -0.39344642 0.8916991 -0.9134761 -0.6768568 -0.555095 -0.33431515 -0.12396032 -0.7434148 -0.31541052 -0.34774432 -0.9065228 0.15621755 -0.6663494 -0.21205825 -0.35649595 -0.7988028 -0.46905473 1.071134 -0.5489356 1.7131537 -2.3925893 -0.22384262 0.8092545 0.50651896 -0.03729048 0.04229549 0.4696722 -0.31103674 -0.06362896 -0.74290466 -0.3653602 -0.02680058 0.14140998 -0.4131572 0.75958216 0.35887694 0.45413557 -0.93407625 -0.3672621 0.08302029 0.15623026 -0.48821226 0.4031453 -0.07743395 0.53424186 -0.6731772 0.89396405 0.7997435 -0.35980126 -0.05620595 0.04405164 0.23186159 -0.30097803 -1.2981218 1.3596495 -0.14786628 0.10717073 -0.23259921 -1.4890085 1.0751711 0.34518096 0.86003375 -0.4194076 0.04341883 0.4509233 0.00650958 -0.36364433 0.13114902 0.11657662 -0.20029409 0.49018562 0.06864677 0.40525243 0.45756805 0.40968728 1.3618747 -0.27347568 0.29069272 -0.09536338 0.69809717 -0.28592777 0.7741229 0.5619512 -0.13262504 0.53579444 0.89780223 -0.35455963 -0.8174493 -1.2216463 -0.29313996 0.73393965 -0.02073103 -0.64815253 -0.4116555 -1.0533314 0.10754564 0.77002347 -0.5608062 -0.40670767 -0.29265657 -1.1594074 0.46849167 0.24022791 0.17532296 -0.7273318 -0.00527034 0.0595459 -0.14656603 -1.1467724 -0.5081712 0.00653717 -0.78158754 -1.2724664 -0.44467983 -0.21298 0.9204349 -0.10817115 0.9045871 0.12861122 -0.2670664 0.45877606 -0.4598433 -0.41627333 -0.18953381 0.03316217 0.28851584 0.68208486 0.6427705 -1.0323372 -0.5355193 0.21806109 -1.1896249 -0.61480385 0.39550284 0.5346536 0.66906774 0.47820085 0.82713956 -1.175335 0.4502126 -1.2165096 -0.6467284 0.09601197 -0.65529495 -0.16835326 0.89777267 -0.39727893 -0.8091702 -0.12314913 -0.0796975 -0.6267535 -0.54493624 0.6463538 -0.36458698 0.3395356 0.5748023 0.41047496 -0.07158152 -0.41132742 0.1085258 0.46875995 0.49734148 -0.55240303 1.2102998 0.36792365 0.1161689 -0.99973035 -0.67766917 -0.78639925 -0.51447207 -0.11449184 0.46109727 -0.83880675 -0.3073462 0.7123487 -0.71055496 0.06308348 -0.32613978 0.5033331 -0.1176855 0.6651638 -0.38971817 -0.9576377 -0.19952096 -0.5174253 0.8709921 0.08637686 -0.1215122 -1.3045034 0.25521648 -0.12183641 0.2978289 0.638685 0.9444553 -1.5179712 -0.30616644 -0.584501 -0.1519843 0.34363538 0.5293238 0.09523054 -0.83868843 -0.3081165 0.02623581 0.09374299 0.64764696 0.20160589 1.5230823 -0.36931136 -0.27304932 0.6210916 1.0871631 -0.05869334 0.42371032 0.09673486 1.0651585 0.5462477 0.58464736 0.9196316 0.5883184 0.20649618 0.49785691 0.21489424 0.61613876 -0.24085262 0.38298476 0.97196287 0.09968071 -0.4407592 -1.0198174 0.5778257 -1.8162957 -0.931193 -0.36026648 2.361137 0.5077038 0.22818983 0.37240818 0.40064436 0.93683624 0.288647 -0.7634972 0.07491618 -0.16548313 -0.23993625 0.1423116 -0.02611166 -1.2750806 0.44323954 4.3620048 0.92923194 -1.0701499 -0.20078157 0.67168367 0.12794505 -0.34138504 -0.16084242 -0.48909676 0.7758192 1.140724 -0.4856733 0.08794083 0.73950166 0.12069149 0.11008414 -0.8479034 0.9224402 -0.02318129 -0.58672905 0.12142985 0.12728155 0.5103794 -0.01644218 0.07077526 0.4289856 -0.02027258 -0.682631 0.2699298 0.6100881 0.5164012 -0.93205476 0.7808657 0.53399515 -1.4361206 -0.16112182 -0.1935913 0.27959007 0.07477231 1.2265745 -0.5967173 0.9681549 0.7362582 1.1032519 -0.6215244 1.1252656 0.05016679 1.0015644 -0.5395504 0.40172127 -0.04900596 -0.39810812 0.8842244 1.0339317 0.6520496 0.01801273 0.3676742 0.67267585 0.10198241 0.5081706 -0.73669 -0.2884094 0.35572177 1.5089501 -0.84887195 -0.07665595 -0.6047602 0.6220959 0.27247214 0.42734575 -0.79952556 -0.39791745 0.600952 0.26524302 0.07023171 -0.12827544 0.02722981 -1.5439614 0.32901713 -0.6343065 1.0236824 -0.05767419 -1.9097309 0.66161925 0.09396958 -1.7297921 -0.4035727 -0.35731894 -1.0007265 0.6374798 -1.3305504 -0.8999187 -0.31206965 0.8858367 0.1936281 -0.22221814 0.7290381 0.5166912 -1.0174763 0.7437374 0.0227528 0.33486667 0.64619225 -1.5067304 0.40207323 1.097876 0.29920462 0.36721423 0.60915184 -0.7515611 -1.1145757 -1.4499916 0.51891065 -0.38520974 1.1286186 -0.3617294 -1.6252372 0.56539166 -0.30645236 0.53632647 0.85928637 0.2755078 -0.34070632 -0.20006806 -1.092309 0.32822105 0.9402978 -0.43481934 -0.34409377 0.39648393 0.46942666 -0.18876332 -0.9300207 0.5747985 -0.06166585 -0.81618613 0.6023477 -0.55340296 0.03496457 -0.5052985 -0.04328706 -1.3854007 0.02241829 -0.50633925 -0.7670769 1.5134747 0.44659087 -1.1900853 0.55716056 0.555654 -0.06502936 -0.58405787 -0.9161035 -0.8114456 -0.26377362 -0.68681043 0.81408113 1.2456526 0.01476403 0.12164007 -0.27301562 0.6828901 0.7705591 -0.04821004 0.74856114 -1.5621364 -0.04767304 -0.3269968 -0.8384485 -0.50629175 0.2880829 -0.9313825 -0.3129872 -1.0335262 -0.14876877 -0.51490617 -0.72618383 0.31053543 -0.39712363 -0.04948711 -0.42802072 0.14016992 -0.6994981 0.8394736 0.690032 -0.01278426 -0.20682311 0.39510018 -0.50528985 0.75018805 0.93478024 -0.42230153 -0.6685361 0.06812024 0.07989256 -0.0966455 0.33854488 -0.94509184 0.30086106 -0.05919998 0.20667112 -0.6699216 -0.11000088 -0.9540469 0.05565588 -0.0083946 0.0611834 0.432769 0.09200902 1.1786517 -0.53733844 0.13477351 0.35998276 0.29729187 -0.6960046 0.9638407 -0.17157313 0.44965497 1.1405841 0.20589133 -0.03403381 -0.39728904 -0.80268246 0.47984555 0.18697557 0.43174344 0.5080196 -1.4731779 -0.77651554 0.40817362 0.58679765 0.34077096 0.67552656 1.081941 -0.21139273 -0.01695552 0.12646449 -0.8438767 -0.8250726 0.5519275 0.15049702 -0.2719686 -0.75617373 0.4188107 0.21656445 -0.44518453 0.15889019 -0.2171141 -0.40501437 0.17161182 0.4961063 0.31532624 0.09204116 -0.67280173 -0.3355978 0.29853073 -0.22927791 0.42062485 1.2047687 -0.18284784 -0.2411853 0.70197755 1.0224035 0.11506057 -0.90175843 -0.5430179 0.41130733 -0.5419905 -0.29737318 -0.14977884 -1.2838025 0.5547947 0.3566249 0.80749017 1.3948445 -0.05605501 0.6802882 0.26970333 0.12797621 -0.62844783 -0.0736894 0.3732067 0.5286119 -1.3479451 -0.14728546 -0.41888747 -0.5695238 0.9227423 0.8532444 -0.19173478 1.0503942 0.008409 -0.08934028 -0.26533356 -0.5649494 -0.10223074 0.63001525 0.45417786 0.23886728 0.11971289 -0.03082484 0.69146043 -0.04048832 -0.6125076 0.34460646 0.6179434 -0.08628462 -0.982463 -0.20413928 0.98714 -0.7075977 0.11688781 -0.06726016 0.67512715 -0.17381056 0.7939316 0.23628621 -0.26136997 0.4795234 0.2608057 -0.225624 -0.46778378 -0.14452627 0.08527276 -0.16621998 -0.4938554 -0.09750116 -0.83163995 -1.2097551 -0.24735115 -0.3652128 0.4268569 0.17877072 1.0078244 0.44358274 0.4625882 0.9186034 -0.30287126 -0.40030333 -0.99255574 -1.077862 0.5633249 0.56177956 -0.68949705 -0.85056925 -0.2512943 ]

[7.322405815124512, 2.7496509552001953]

f1eda516-f966-4720-9209-d2ff6430ceee

metaphysica-ood-robustness-in-physics

2303.03181

https://arxiv.org/abs/2303.03181v1

https://arxiv.org/pdf/2303.03181v1.pdf

MetaPhysiCa: OOD Robustness in Physics-informed Machine Learning

A fundamental challenge in physics-informed machine learning (PIML) is the design of robust PIML methods for out-of-distribution (OOD) forecasting tasks. These OOD tasks require learning-to-learn from observations of the same (ODE) dynamical system with different unknown ODE parameters, and demand accurate forecasts even under out-of-support initial conditions and out-of-support ODE parameters. In this work we propose a solution for such tasks, which we define as a meta-learning procedure for causal structure discovery (including invariant risk minimization). Using three different OOD tasks, we empirically observe that the proposed approach significantly outperforms existing state-of-the-art PIML and deep learning methods.

['Bruno Ribeiro', 'Muhammad Ashraful Alam', 'S Chandra Mouli']

2023-03-06

['physics-informed-machine-learning']

['graphs']

[-2.98522860e-01 2.53746778e-01 -3.27518016e-01 -2.58699924e-01 -8.94172132e-01 -4.27406251e-01 1.25901413e+00 1.87137559e-01 1.94454640e-01 8.24536204e-01 3.51171166e-01 -5.05739272e-01 -5.82969427e-01 -5.15802145e-01 -9.24915910e-01 -6.91881061e-01 -5.14903545e-01 9.82294679e-01 -1.21129796e-01 4.88275290e-02 3.02228481e-01 6.58611953e-01 -1.25797570e+00 -1.47055928e-02 9.97772872e-01 1.04386127e+00 -1.63647354e-01 6.46026611e-01 -1.12191647e-01 9.93585050e-01 -3.19069862e-01 -1.75128236e-01 2.64085710e-01 -1.66738480e-01 -4.99208421e-01 -4.21777993e-01 6.83799744e-01 -1.40364066e-01 -6.45627081e-01 8.51725757e-01 4.66975898e-01 1.93992004e-01 1.33174610e+00 -1.37113428e+00 -5.09200633e-01 3.69798124e-01 -3.64688158e-01 4.50176805e-01 -1.57680288e-01 2.06938207e-01 8.99002075e-01 -8.66707087e-01 5.61692715e-01 1.53359032e+00 6.67646408e-01 2.16433361e-01 -1.68929076e+00 -6.48581386e-01 2.98015237e-01 -4.61846497e-03 -9.75595355e-01 -2.06226036e-01 7.96521723e-01 -1.22276938e+00 7.32005000e-01 -1.26229182e-01 2.06342936e-01 1.49897039e+00 8.57859075e-01 4.52654481e-01 1.16943860e+00 -8.39306489e-02 7.60321915e-01 -7.12577179e-02 2.40713954e-01 5.91396153e-01 2.36950651e-01 6.75442398e-01 -8.16747606e-01 -6.95764005e-01 6.75340235e-01 -4.45533276e-01 2.01253608e-01 -4.97347683e-01 -1.01006305e+00 1.05026126e+00 -1.32111721e-02 -1.86120093e-01 -3.86581302e-01 4.38365668e-01 3.17017794e-01 2.56973267e-01 1.13689637e+00 5.73376298e-01 -7.87483871e-01 9.74374935e-02 -8.66467893e-01 7.97899127e-01 1.04985952e+00 5.60365379e-01 3.98120224e-01 2.26732597e-01 -2.89150804e-01 3.73955786e-01 4.98802245e-01 7.89358914e-01 -1.00322120e-01 -1.00579572e+00 4.02242571e-01 9.95985419e-02 3.69859040e-01 -7.26337373e-01 -7.31999457e-01 -6.71160877e-01 -1.18562484e+00 2.26861119e-01 5.48093915e-01 -4.76427615e-01 -8.47608984e-01 1.72159553e+00 5.66725016e-01 8.12479436e-01 -1.02763012e-01 7.00425446e-01 7.00084627e-01 9.27851617e-01 8.80991518e-02 -3.00485641e-01 9.26410437e-01 -5.86395562e-01 -4.90506738e-01 -1.89730704e-01 5.48537791e-01 -4.04538989e-01 7.70500302e-01 3.77916545e-01 -8.00533295e-01 -4.83137995e-01 -8.79441857e-01 2.07737550e-01 -1.28078312e-01 -1.47792980e-01 7.97856867e-01 1.95654646e-01 -5.55412769e-01 9.84248757e-01 -8.01269412e-01 1.19790301e-01 2.73254842e-01 2.23639887e-02 2.60903705e-02 3.22930813e-01 -1.26697505e+00 8.95654678e-01 2.07944170e-01 -1.53712984e-02 -1.84439301e+00 -1.66001487e+00 -4.41004187e-01 4.94398326e-02 3.93006980e-01 -8.54450405e-01 1.35719383e+00 -2.76064217e-01 -1.52957487e+00 4.76684779e-01 4.23986763e-02 -6.22603893e-01 7.08395064e-01 -5.11004806e-01 -3.23384345e-01 -2.93487430e-01 -1.22796752e-01 -3.14847333e-03 1.48823190e+00 -1.12655139e+00 -2.79892325e-01 -2.23760426e-01 -1.77544445e-01 -3.66775721e-01 2.73550332e-01 -9.88323912e-02 2.56105691e-01 -6.78609550e-01 -1.78630933e-01 -9.20816064e-01 -3.24385017e-01 -2.21959576e-01 -7.05757260e-01 -4.90658790e-01 6.45606875e-01 -7.03436017e-01 7.87010074e-01 -1.60928893e+00 4.52880979e-01 5.26960567e-02 5.24168730e-01 -7.64237866e-02 -7.07402006e-02 6.59239650e-01 -2.65313774e-01 4.03200760e-02 -1.32379040e-01 -4.78211910e-01 4.31067646e-01 3.78690921e-02 -1.02423608e+00 5.70742905e-01 2.86692321e-01 6.15231395e-01 -8.77944827e-01 -3.08512539e-01 3.37129116e-01 2.41351813e-01 -4.35149670e-01 6.51996374e-01 -8.48212361e-01 1.03093064e+00 -6.10051930e-01 2.74814725e-01 6.10558510e-01 -2.57743269e-01 -6.28231615e-02 5.74196912e-02 -2.45128632e-01 4.62187141e-01 -1.16759360e+00 1.39963770e+00 -6.76822007e-01 5.16214550e-01 -2.68139869e-01 -1.31024396e+00 7.93361604e-01 2.89336890e-01 6.15600526e-01 -2.78252810e-01 2.63865683e-02 1.59253195e-01 -2.42312737e-02 -3.89001101e-01 -9.20657963e-02 -2.53561735e-01 -1.96600616e-01 3.66940528e-01 3.87236983e-01 -2.46012434e-01 -1.57893356e-02 2.10595697e-01 1.22601748e+00 3.71821132e-03 2.31836945e-01 -8.39185596e-01 3.31481010e-01 -1.60332382e-01 5.38605571e-01 1.14137459e+00 1.00457564e-01 1.40388072e-01 9.20596838e-01 -6.58026099e-01 -1.09800935e+00 -1.25833380e+00 -4.75972414e-01 6.83725536e-01 -4.41391140e-01 -1.88796625e-01 -4.19022977e-01 -7.73173392e-01 5.09042919e-01 1.02198875e+00 -6.51456118e-01 -1.15965806e-01 -4.19083178e-01 -1.19569182e+00 1.73852861e-01 2.22718403e-01 1.19213993e-02 -6.41997039e-01 -3.99855003e-02 3.25237602e-01 3.84215504e-01 -9.75707591e-01 8.22015852e-03 1.21353030e-01 -8.17427516e-01 -1.02287126e+00 -5.47910690e-01 -3.71914245e-02 1.73142448e-01 -4.55877393e-01 1.27780879e+00 -6.55847549e-01 -3.44734937e-01 2.11472020e-01 2.79484302e-01 -6.42017007e-01 -7.31389821e-01 1.90221235e-01 5.37219167e-01 -1.78370089e-03 -2.28681803e-01 -8.82728577e-01 -2.08550081e-01 3.90928648e-02 -3.85900825e-01 -4.30252366e-02 4.36786979e-01 8.71376872e-01 5.34105897e-01 6.64434731e-02 7.50703394e-01 -7.02600896e-01 5.31734467e-01 -8.11663568e-01 -1.35682094e+00 1.88419864e-01 -5.83409846e-01 6.18499458e-01 8.16770315e-01 -5.19673467e-01 -1.33054316e+00 -2.29989529e-01 1.77636206e-01 -7.33057201e-01 -2.59154111e-01 7.47334361e-01 -3.93689647e-02 2.04627857e-01 5.10628045e-01 -2.78674275e-01 -3.73951852e-01 -9.52370942e-01 5.63941300e-01 2.24392489e-01 5.67689598e-01 -9.59412575e-01 1.06196129e+00 3.07889551e-01 6.13857508e-01 -8.85528564e-01 -1.56827497e+00 6.10276461e-02 -8.01024556e-01 -2.71363646e-01 9.20105398e-01 -1.16396713e+00 -7.70377278e-01 6.56614423e-01 -1.17310870e+00 -7.12759852e-01 -3.27036083e-01 6.69781387e-01 -6.33018911e-01 -2.36185908e-01 -4.52835709e-01 -1.11435461e+00 -2.59302650e-02 -7.97974288e-01 1.20040512e+00 6.97470978e-02 -1.17134057e-01 -1.36066544e+00 6.59343004e-01 -7.77575150e-02 2.52857506e-01 4.82205331e-01 1.26618874e+00 -5.77677071e-01 -5.79234838e-01 -1.23247966e-01 -6.67605847e-02 9.00007114e-02 -1.58080757e-01 1.40185133e-01 -1.17288506e+00 -1.36399299e-01 1.84231043e-01 -4.20707345e-01 1.11075628e+00 1.02443242e+00 1.33209515e+00 -3.68481219e-01 -4.73001361e-01 7.38829792e-01 8.97267580e-01 -2.97838330e-01 -9.41057783e-03 -1.22745961e-01 6.10371232e-01 7.09294796e-01 4.63942289e-01 7.39727259e-01 2.28267938e-01 6.12636209e-01 2.98591733e-01 1.53730348e-01 -6.59943968e-02 -5.37349880e-01 3.83911103e-01 7.65877426e-01 2.86227077e-01 -3.95462215e-01 -1.09586596e+00 2.78218359e-01 -2.13879347e+00 -9.10700440e-01 -3.86889875e-01 2.00880504e+00 8.21102440e-01 2.28176638e-01 -1.32175282e-01 -5.34691095e-01 5.17599106e-01 4.36089098e-01 -1.10008180e+00 -2.23100871e-01 9.67356041e-02 3.06688864e-02 4.75691080e-01 6.88170850e-01 -1.44465339e+00 6.21991754e-01 6.80000114e+00 9.81191993e-01 -9.10760343e-01 4.00550663e-01 8.21447492e-01 9.49038193e-03 -2.75223911e-01 1.78260148e-01 -1.09801185e+00 4.67472643e-01 1.32914686e+00 -3.06444466e-01 2.75534570e-01 7.66210616e-01 5.30119300e-01 3.51367593e-02 -1.42939889e+00 5.12087703e-01 -4.37793851e-01 -1.71250868e+00 -2.79672801e-01 2.62701929e-01 1.12788403e+00 3.58225197e-01 -7.06447065e-02 2.94660181e-01 8.33796144e-01 -1.03681076e+00 7.32368231e-01 1.03953052e+00 6.31096303e-01 -4.28553104e-01 4.99486476e-01 5.73240280e-01 -9.41696823e-01 -2.83989683e-02 -4.22234833e-01 -3.37002993e-01 3.32787186e-01 1.52601099e+00 -4.13147181e-01 6.08023465e-01 5.65961897e-01 8.83859277e-01 -5.58077618e-02 8.67525041e-01 -2.07068607e-01 1.13688028e+00 -5.44321656e-01 1.66775048e-01 -1.06974589e-02 -9.49898958e-02 9.55959380e-01 7.97570229e-01 3.10255229e-01 3.50464531e-03 3.35753143e-01 1.41356921e+00 -1.51099294e-01 -5.88614464e-01 -6.47396207e-01 -2.60203302e-01 4.68844324e-01 8.54702711e-01 -2.27784559e-01 -3.70522499e-01 -3.22035477e-02 2.26917103e-01 3.30603480e-01 2.34606519e-01 -9.33900833e-01 2.92724460e-01 1.11277258e+00 -1.70868840e-02 1.01800054e-01 -5.89602530e-01 -2.66335100e-01 -1.29954064e+00 -4.25646305e-01 -4.83326733e-01 4.52264249e-01 -5.94551504e-01 -2.11733103e+00 6.53962269e-02 5.46385288e-01 -8.84140015e-01 -4.55587566e-01 -6.39626801e-01 -9.32071209e-01 7.36292601e-01 -1.30525458e+00 -8.79224777e-01 1.43894643e-01 6.13731891e-02 4.20793086e-01 -4.11778063e-01 6.13178909e-01 6.21168911e-02 -7.40309894e-01 -2.57359296e-02 7.24783897e-01 -3.24188083e-01 6.38259649e-01 -1.31458378e+00 7.07316220e-01 6.75774932e-01 1.14022553e-01 1.21391900e-01 1.16999912e+00 -1.17798197e+00 -1.62806058e+00 -1.22290933e+00 9.05185938e-01 -5.98984003e-01 1.15809834e+00 -7.62831807e-01 -6.96852088e-01 5.80231786e-01 -1.11611180e-01 2.14450061e-01 1.39776185e-01 3.89313877e-01 -2.76598930e-01 -2.19618492e-02 -8.16220641e-01 2.76019782e-01 1.02244186e+00 -4.63492304e-01 -5.99923313e-01 8.42140913e-01 8.80686522e-01 -3.97134185e-01 -1.04627192e+00 3.98727655e-01 3.72205347e-01 -6.80893302e-01 1.16532016e+00 -1.14217329e+00 6.45539880e-01 -4.59421761e-02 6.37145899e-03 -1.61170578e+00 -1.88839227e-01 -8.94858599e-01 -7.19361067e-01 1.32943618e+00 2.37729222e-01 -7.80327797e-01 3.89369845e-01 5.71919978e-01 -1.53550133e-01 -5.78512847e-01 -1.15479517e+00 -1.08120489e+00 4.99426752e-01 -8.12347531e-01 3.79125506e-01 9.28065896e-01 -3.88498545e-01 3.15197527e-01 -6.18076205e-01 8.43098104e-01 1.31922519e+00 3.17113161e-01 7.38164902e-01 -1.91898298e+00 -3.64592314e-01 -3.77091855e-01 1.01372004e-01 -5.31585157e-01 7.13429153e-01 -8.22420001e-01 -1.38032138e-01 -1.14050162e+00 8.76505002e-02 -5.95477104e-01 -3.46650556e-02 2.57707089e-01 -6.46382803e-03 -4.70674634e-01 1.83879465e-01 3.01886737e-01 -1.51536554e-01 1.06056654e+00 6.65779352e-01 -1.70806363e-01 3.95599790e-02 1.18545584e-01 -1.13613598e-01 9.40232038e-01 6.35561943e-01 -1.05117297e+00 -3.42189789e-01 -6.56940416e-02 2.21317455e-01 3.85145992e-01 8.75223577e-01 -7.78324783e-01 -1.09995648e-01 -5.47073245e-01 3.61209750e-01 -8.70724797e-01 2.20705688e-01 -2.47373611e-01 1.23322554e-01 5.38854480e-01 -5.73278010e-01 -1.58485651e-01 3.45676690e-01 7.95828640e-01 1.93200499e-01 -2.60307878e-01 5.10151565e-01 1.41403284e-02 -2.69409597e-01 4.47150648e-01 -5.72915375e-01 5.47788382e-01 6.21851146e-01 1.14241552e+00 -5.25546014e-01 -4.16698188e-01 -5.52250981e-01 3.37144852e-01 -1.21973455e-01 3.86798769e-01 7.29651302e-02 -1.12879562e+00 -1.05515647e+00 4.04840112e-02 -1.23871528e-01 -2.53854487e-02 3.30419093e-01 7.29712903e-01 4.54260297e-02 6.55725658e-01 3.22559923e-01 -6.23903155e-01 -3.86590838e-01 4.44767833e-01 5.23558080e-01 -6.47549391e-01 -6.51127398e-01 8.33950102e-01 5.86332977e-01 -8.02225530e-01 2.95824349e-01 -4.01486218e-01 2.63664901e-01 1.24202594e-01 3.83847356e-01 5.35101473e-01 3.90245160e-03 -4.34148870e-02 -1.83495089e-01 2.93668151e-01 2.22184971e-01 -1.95916057e-01 1.44981647e+00 1.14014871e-01 8.14545900e-02 1.00277567e+00 8.90063584e-01 -1.59404680e-01 -1.72824061e+00 -1.46059483e-01 2.73394167e-01 -2.56710321e-01 5.24468899e-01 -1.02126801e+00 -7.79692471e-01 1.14974546e+00 6.15771592e-01 1.59610972e-01 2.50567228e-01 1.96828455e-01 5.21374106e-01 6.13011718e-01 2.20686555e-01 -9.55601633e-01 -5.68397865e-02 6.08976424e-01 1.17168319e+00 -1.14194977e+00 2.13618070e-01 3.82463783e-02 -2.84496725e-01 1.05583060e+00 3.15749496e-01 -4.28752065e-01 1.35466814e+00 4.98955399e-01 -4.67940122e-01 -3.99298072e-01 -1.26204884e+00 8.96075964e-02 8.53356004e-01 3.61839145e-01 5.91317080e-02 -6.79462925e-02 9.28513482e-02 6.88852906e-01 -2.33403109e-02 -1.01107873e-01 3.88493478e-01 4.52187747e-01 -2.84872651e-01 -8.18943918e-01 -2.54664153e-01 6.45089984e-01 -1.72674526e-02 -6.29249737e-02 -2.65707254e-01 7.82577515e-01 -1.94703370e-01 7.06176996e-01 1.27751753e-01 -1.40642002e-01 5.05577959e-02 3.66708994e-01 3.32208931e-01 -2.71077394e-01 -1.29292354e-01 2.87608057e-02 2.39184886e-01 -6.20611310e-01 -1.20484859e-01 -9.98349547e-01 -1.03006291e+00 -4.07627642e-01 -5.69166802e-02 -6.63048849e-02 5.93277216e-01 1.35298622e+00 6.02592409e-01 6.71360672e-01 5.95771551e-01 -1.38373733e+00 -1.03557920e+00 -1.03577769e+00 -6.78359747e-01 -1.21988349e-01 5.26108027e-01 -1.26360869e+00 -7.79225707e-01 -2.59871244e-01]

[6.912353038787842, 3.785682201385498]

1cd693f8-3fe3-4332-acaf-94568602fb1a

speech-reconstruction-with-reminiscent-sound

https://ieeexplore.ieee.org/document/9618777

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9618777

Speech Reconstruction with Reminiscent Sound via Visual Voice Memory

The goal of this work is to reconstruct speech from silent video, in both speaker dependent and independent ways. Unlike previous works that have been mostly restricted to a speaker dependent setting, we propose Visual Voice memory to restore essential auditory information to generate proper speech from different speakers and even unseen speakers. The proposed memory takes additional auditory information that corresponds to the input face movements and stores the auditory contexts that can be recalled by the given input visual features. Specifically, the Visual Voice memory contains value and key memory slots, where value memory slots are for saving the audio features, and key memory slots are for storing the visual features in the same location of the saved audio features. Guiding each memory to properly save each feature, the model can adequately produce the speech. Hence, our method employs both video and audio information during training time but does not require any additional auditory input during inference. Our key contributions are: (1) proposing the Visual Voice memory that brings rich information of audio that complements the visual features, thus producing high-quality speech from silent video, and (2) enabling multi-speaker and unseen speaker training by memorizing auditory features and the corresponding visual features. We validate the proposed framework on GRID and Lip2Wav datasets and show that our method surpasses the performance of previous works on both multi-speaker and speaker independent settings. We also demonstrate that the Visual Voice memory contains meaningful information to reconstruct speech.

['Yong Man Ro', 'Se Jin Park', 'Minsu Kim', 'Joanna Hong']

2021-11-17

ieee-acm-transactions-on-audio-speech-and-5

['speaker-specific-lip-to-speech-synthesis']

['computer-vision']

[ 1.35163367e-01 2.14561243e-02 -1.50198400e-01 -9.09176096e-02 -8.93071592e-01 -4.27288324e-01 4.72302407e-01 -5.96015453e-02 -9.79254320e-02 5.94192147e-01 3.22108954e-01 -1.48344517e-01 8.16319138e-02 -7.39059627e-01 -9.72935200e-01 -6.87271535e-01 1.80080935e-01 3.58090643e-03 3.20307910e-01 6.74216524e-02 2.54486620e-01 4.13962334e-01 -2.40997124e+00 5.96393168e-01 2.59296566e-01 1.03662395e+00 7.35263944e-01 1.02843308e+00 -3.96707237e-01 6.01744652e-01 -8.81622195e-01 4.72133718e-02 -2.10198224e-01 -4.94609535e-01 -7.03332901e-01 1.31715834e-01 5.01652837e-01 -3.82657886e-01 -4.12678421e-01 5.81415415e-01 9.10412371e-01 5.51336825e-01 4.30284142e-01 -1.21392345e+00 -9.02323186e-01 4.71044749e-01 -7.67187551e-02 3.78219277e-01 7.14446962e-01 1.51443824e-01 6.97885096e-01 -1.36954796e+00 6.08069301e-01 1.45860851e+00 2.09812880e-01 9.02412474e-01 -8.74479532e-01 -6.91631854e-01 1.56173363e-01 7.28146970e-01 -1.53095829e+00 -1.20646703e+00 7.94129431e-01 -2.12990791e-01 1.07952750e+00 5.60528755e-01 7.85394549e-01 1.24920511e+00 -1.12265721e-01 6.53461635e-01 7.41369009e-01 -4.81944621e-01 2.55130440e-01 2.62139469e-01 -1.17171355e-01 6.23693526e-01 -4.99854386e-01 2.70065308e-01 -1.32713723e+00 1.26703575e-01 6.18640661e-01 -1.61135048e-01 -5.78698039e-01 3.32804546e-02 -1.13937926e+00 6.24695897e-01 -1.08334012e-01 2.35928133e-01 -2.38170207e-01 2.85569206e-02 1.14596076e-01 3.04742754e-01 1.47266284e-01 -3.89822602e-01 -9.80125815e-02 -1.63617715e-01 -1.06631291e+00 -2.92587698e-01 4.60746258e-01 9.31655824e-01 7.49157608e-01 5.26734471e-01 -2.55662471e-01 9.49748993e-01 5.22120774e-01 9.25100863e-01 7.43795753e-01 -9.05437052e-01 3.32599699e-01 5.27248047e-02 1.04290228e-02 -9.43929374e-01 -1.05233788e-01 -1.73282608e-01 -5.93986332e-01 4.79539298e-02 -8.42043199e-03 2.07574844e-01 -9.13734198e-01 1.87579620e+00 4.22017574e-01 8.63528490e-01 3.03502709e-01 6.09584689e-01 1.64141798e+00 8.80196571e-01 -1.37972116e-01 -5.85260868e-01 1.37237227e+00 -8.31464827e-01 -9.43513811e-01 -2.32072666e-01 -1.04367815e-01 -8.22567582e-01 1.37640107e+00 2.65975863e-01 -1.34693646e+00 -9.83368993e-01 -1.09972250e+00 -4.32776622e-02 -9.81319100e-02 1.13268390e-01 8.43175650e-02 5.83928764e-01 -1.43262613e+00 2.24954814e-01 -6.03057086e-01 -1.31044403e-01 1.72415555e-01 2.28921056e-01 -5.32940984e-01 9.84680951e-02 -1.12726784e+00 5.45523107e-01 1.11417241e-01 1.25196919e-01 -1.32502043e+00 -6.76528037e-01 -1.19584298e+00 1.61475286e-01 3.50422978e-01 -7.23278940e-01 1.05467939e+00 -7.71598339e-01 -1.73362732e+00 4.44587022e-01 -8.29504073e-01 -2.76726812e-01 1.80795103e-01 1.96784735e-01 -6.22224748e-01 5.72301149e-01 -1.75352186e-01 8.79990935e-01 1.43577003e+00 -1.48550940e+00 -4.88949537e-01 -2.31330663e-01 -2.77168393e-01 1.99745566e-01 -4.52843964e-01 -1.89732656e-01 -7.63567150e-01 -7.44464278e-01 5.11377119e-02 -5.08318245e-01 5.20622194e-01 9.93385911e-02 -3.74680489e-01 5.62428720e-02 9.18205738e-01 -7.72711098e-01 1.17393196e+00 -2.51907945e+00 1.75391853e-01 2.69413292e-02 4.37067337e-02 2.00756267e-01 -4.13771003e-01 1.82548106e-01 5.19623682e-02 6.17427006e-03 4.15949039e-02 -7.91739881e-01 -1.65967196e-01 2.79702693e-01 -6.09971404e-01 1.35841563e-01 2.00240500e-02 7.03347802e-01 -5.37950039e-01 -7.84669042e-01 3.77087981e-01 1.18112040e+00 -4.36076373e-01 2.53082812e-01 1.21394426e-01 4.44888145e-01 2.46174727e-02 5.16830862e-01 6.70543015e-01 3.49312313e-02 6.31652772e-02 -1.61904663e-01 1.09942686e-02 3.62883300e-01 -1.44306052e+00 1.47273743e+00 -7.21280754e-01 6.07025445e-01 2.80263245e-01 -6.41736031e-01 9.04557288e-01 6.71395600e-01 1.59876719e-02 -8.58713925e-01 -1.07710510e-01 -1.97439581e-01 -5.46941578e-01 -7.06642568e-01 4.60623771e-01 -1.49921000e-01 4.00855958e-01 4.56874073e-01 2.73463190e-01 -2.98355911e-02 -1.37979716e-01 7.26607814e-02 6.32150471e-01 -1.69032723e-01 1.05151691e-01 2.73186266e-01 8.03251684e-01 -6.79051638e-01 3.64343256e-01 5.42084575e-01 -1.71593532e-01 6.48002326e-01 4.04010154e-02 1.10971279e-01 -5.58757484e-01 -1.49020576e+00 8.74770340e-04 1.41984487e+00 2.33701423e-01 -4.12017941e-01 -6.20233238e-01 -2.84072548e-01 -2.17811391e-01 7.60261536e-01 -5.74925482e-01 -2.68882841e-01 -4.27414507e-01 4.04782556e-02 4.90585595e-01 6.18743241e-01 1.78799927e-01 -1.39558160e+00 -5.08998930e-01 5.48403822e-02 -5.69909155e-01 -9.62647140e-01 -7.22941160e-01 -1.50986493e-01 -6.43072963e-01 -8.80278945e-01 -6.67531729e-01 -9.92214084e-01 4.63367552e-01 6.57686234e-01 8.42989087e-01 2.12977499e-01 -2.00467005e-01 7.74889648e-01 -1.44644380e-01 -5.10984249e-02 -5.78738987e-01 -3.76762390e-01 1.66455448e-01 2.10321859e-01 -1.98292598e-01 -7.85183012e-01 -3.95517707e-01 3.14286083e-01 -8.80349278e-01 1.48851261e-01 3.42066973e-01 8.52898717e-01 8.85127246e-01 -1.77451387e-01 9.01892960e-01 -2.79298961e-01 4.18533921e-01 -3.81961972e-01 -9.65103135e-02 2.40588084e-01 -9.08663049e-02 3.88539471e-02 5.33784270e-01 -6.69002473e-01 -1.05658090e+00 6.15399964e-02 -4.64926839e-01 -6.43199921e-01 -2.66987383e-01 2.04358578e-01 -4.77482110e-01 2.41102874e-01 3.69469225e-01 7.67100096e-01 1.75405636e-01 -5.09427488e-01 3.78907830e-01 7.91416109e-01 7.89017081e-01 -3.71540904e-01 3.91426742e-01 4.70061123e-01 -1.82441279e-01 -1.19200206e+00 -1.84336230e-01 -1.49205104e-01 -5.40629089e-01 -3.86052668e-01 7.49875307e-01 -8.82140279e-01 -9.52477276e-01 3.49368542e-01 -1.09747207e+00 -8.84397104e-02 -2.11638287e-01 4.01207805e-01 -6.24419630e-01 6.10483646e-01 -4.45876211e-01 -1.06644046e+00 -2.79470921e-01 -1.34208071e+00 1.07874608e+00 1.29156813e-01 -2.81398296e-02 -7.82009065e-01 -1.38202816e-01 4.14243102e-01 2.78178573e-01 -3.74330878e-01 8.81267071e-01 -3.15656066e-01 -5.63638568e-01 2.93076903e-01 1.44102693e-01 1.83693975e-01 2.45635137e-01 1.62222967e-01 -1.44300735e+00 -3.09168100e-01 -1.21451579e-01 -2.35750407e-01 1.07454753e+00 4.45124000e-01 9.52382505e-01 -4.68363881e-01 -2.89668292e-01 3.63209337e-01 8.24367106e-01 2.90215522e-01 6.68720007e-01 -1.84070259e-01 4.29352760e-01 6.48050964e-01 3.38460028e-01 4.33496475e-01 5.45197070e-01 7.52119303e-01 4.95247781e-01 9.58384350e-02 -5.78016937e-01 -4.65542763e-01 7.63603747e-01 1.07634914e+00 1.99427068e-01 -2.85988808e-01 -3.58029485e-01 8.23235810e-01 -1.48428094e+00 -1.23271954e+00 3.20693135e-01 2.39762402e+00 8.68577480e-01 -1.59149140e-01 1.68142840e-01 4.94630724e-01 9.75790441e-01 1.80614099e-01 -5.11427820e-01 -2.62361676e-01 -3.53501081e-01 4.92891967e-01 -4.20952171e-01 9.46482420e-01 -7.66473055e-01 8.10594499e-01 6.35151005e+00 8.86760056e-01 -1.21271420e+00 2.39510357e-01 2.37699836e-01 -4.88179684e-01 -6.01307273e-01 -4.06123698e-01 -7.44164526e-01 3.64921093e-01 1.14123082e+00 -2.03136116e-01 5.65471053e-01 7.30831206e-01 4.04618919e-01 5.95798832e-04 -1.00519967e+00 1.33921230e+00 4.53482062e-01 -1.37724018e+00 5.44939637e-01 -1.59566239e-01 -8.43167678e-03 -6.00534558e-01 5.48162520e-01 1.52070209e-01 -3.58592510e-01 -1.24628198e+00 1.08001411e+00 6.97892487e-01 1.03033042e+00 -8.76439691e-01 2.29221284e-01 2.96067983e-01 -1.64378488e+00 -8.66445303e-02 -2.66986579e-01 2.84631938e-01 1.80963978e-01 2.43522838e-01 -9.51000571e-01 3.55782062e-01 7.55219936e-01 5.28358757e-01 -5.69706380e-01 8.08393002e-01 -1.73127294e-01 6.46900833e-01 -1.27348959e-01 2.48472139e-01 -4.58363533e-01 5.24956226e-01 7.31360197e-01 9.48569179e-01 6.89401209e-01 -2.06769928e-01 -7.31171966e-02 6.01257026e-01 1.41149387e-01 3.05293202e-01 -5.63560903e-01 1.54697374e-01 1.04941154e+00 9.06644464e-01 -4.10003990e-01 -3.65255803e-01 -3.23995680e-01 1.01734829e+00 4.12594490e-02 3.30167592e-01 -7.70088613e-01 -2.51767397e-01 7.38790393e-01 7.69996420e-02 6.53200924e-01 -1.60118133e-01 5.45712672e-02 -7.77472794e-01 1.06798507e-01 -5.53722799e-01 2.92018205e-01 -1.04189646e+00 -6.78800046e-01 8.12010229e-01 -2.24062681e-01 -1.17655730e+00 -6.19898200e-01 -2.91791856e-01 -4.54567671e-01 8.60131085e-01 -1.51955140e+00 -1.27616560e+00 -3.22036803e-01 1.13877618e+00 8.16425920e-01 -3.06528181e-01 1.03939998e+00 3.22261035e-01 -4.39836532e-01 9.50675786e-01 -2.97672153e-01 -2.78023750e-01 8.11857402e-01 -6.93065345e-01 9.91129950e-02 6.95633650e-01 5.12841344e-01 5.99105775e-01 6.45348668e-01 -5.15688598e-01 -1.49389076e+00 -9.25197005e-01 9.88017380e-01 -2.48729885e-01 8.01984370e-02 -3.93776923e-01 -9.95270967e-01 5.36620200e-01 1.24650322e-01 6.22618571e-02 9.00217712e-01 -4.37197685e-01 -4.53356773e-01 -6.68167695e-02 -1.13012278e+00 3.98662359e-01 1.03304207e+00 -8.53620052e-01 -6.50832176e-01 -1.47092178e-01 1.01022220e+00 -3.62684369e-01 -4.06844735e-01 -3.79663743e-02 6.19585514e-01 -9.40561831e-01 1.27094984e+00 -4.47266281e-01 1.34680122e-01 -4.84508961e-01 -4.87225085e-01 -1.16943741e+00 -6.00804240e-02 -5.25458217e-01 -4.04483289e-01 1.58291781e+00 4.34850574e-01 -4.50263411e-01 3.47535461e-01 3.95413600e-02 -3.20690989e-01 -3.17446887e-01 -1.43986034e+00 -5.13965666e-01 -2.46513739e-01 -8.20212245e-01 7.55615950e-01 6.21191025e-01 -1.94956869e-01 2.33596429e-01 -6.53491735e-01 3.51685643e-01 3.43322635e-01 1.42888784e-01 6.35303020e-01 -1.12208319e+00 -3.55320334e-01 -1.22258611e-01 -3.02219868e-01 -1.13821721e+00 5.19208968e-01 -8.72772872e-01 1.15848117e-01 -1.49979699e+00 1.75080851e-01 -1.25300921e-02 -2.18131512e-01 7.39986837e-01 -9.23833102e-02 4.31555241e-01 3.17900151e-01 5.19809537e-02 -3.70798051e-01 6.26121521e-01 1.25656068e+00 -3.01996320e-01 -3.73137742e-01 6.27250075e-02 -6.58468246e-01 4.54112738e-01 4.10262406e-01 -2.69884676e-01 -6.73653483e-01 -2.99338043e-01 -9.86649394e-02 5.02689958e-01 7.74946809e-01 -9.50683832e-01 3.13389301e-01 -8.71990994e-02 5.33057332e-01 -6.72213614e-01 8.82610202e-01 -6.27348661e-01 2.21576974e-01 1.37215510e-01 -3.19916964e-01 -3.26928273e-02 5.21067739e-01 7.25383818e-01 -2.81068116e-01 -7.49129876e-02 7.10341811e-01 3.73503305e-02 -8.34154427e-01 4.37496379e-02 -6.57995522e-01 -3.03339902e-02 8.32430124e-01 -5.33814967e-01 -2.62371838e-01 -7.49080300e-01 -1.35442436e+00 -1.99080959e-01 1.43907592e-01 6.13371491e-01 1.25055897e+00 -1.48092663e+00 -4.94949818e-01 6.22974634e-01 -3.03401588e-03 -3.40078086e-01 8.14479709e-01 5.94077766e-01 7.20709935e-02 2.92171061e-01 -2.64642328e-01 -7.14309454e-01 -1.81336010e+00 7.72323370e-01 1.09976210e-01 5.93011022e-01 -7.33038306e-01 8.56503189e-01 2.30858684e-01 -7.91897029e-02 4.80276972e-01 -2.15844959e-01 -3.50316793e-01 3.11651289e-01 9.89929616e-01 3.99860054e-01 1.56202674e-01 -1.10211742e+00 -4.02178138e-01 6.52320027e-01 2.69114077e-01 -6.03842914e-01 1.20925891e+00 -5.48239112e-01 2.33496711e-01 5.82010329e-01 9.95415092e-01 5.26820242e-01 -1.14669752e+00 -1.29609808e-01 -6.42874002e-01 -5.37140131e-01 2.24436615e-02 -5.87243199e-01 -1.20018756e+00 1.28201139e+00 7.38175392e-01 -5.47884926e-02 1.24522972e+00 2.40387663e-01 7.73921788e-01 3.48608382e-02 2.14964271e-01 -8.51787448e-01 3.98189932e-01 2.86579341e-01 1.05765724e+00 -8.11567783e-01 -5.38545907e-01 -4.97715205e-01 -7.82145023e-01 1.09142089e+00 4.37572807e-01 5.82697809e-01 5.95236897e-01 4.44635868e-01 1.46658674e-01 2.54814208e-01 -9.56727564e-01 -2.82011271e-01 4.48200762e-01 1.04752696e+00 1.29132271e-01 -2.15539411e-01 3.43334079e-01 1.09845829e+00 -4.66581285e-01 -1.72200069e-01 2.99443513e-01 7.53284633e-01 -7.10236549e-01 -8.38734627e-01 -6.43624365e-01 -5.78130297e-02 -1.15309358e-01 -3.06651145e-01 -3.27203751e-01 4.76044983e-01 4.37478982e-02 1.41637576e+00 3.33212733e-01 -5.57009101e-01 2.67596126e-01 4.86469746e-01 5.07174313e-01 -5.38860202e-01 -3.75220269e-01 2.20883131e-01 -1.53336793e-01 -4.43040252e-01 -2.54876941e-01 -4.34269667e-01 -1.58645129e+00 -3.19936156e-01 -2.11824685e-01 8.61591026e-02 6.17864251e-01 8.43038917e-01 6.92625463e-01 7.02886760e-01 6.15196288e-01 -1.01201129e+00 -1.51932493e-01 -8.06788743e-01 -4.65395749e-01 1.94726110e-01 7.52510726e-01 -9.26785648e-01 -5.62644422e-01 2.90999472e-01]

[14.352807998657227, 5.060187816619873]

46b59dd4-3139-4392-bc20-4b5f9829f7e6

safe-exploration-incurs-nearly-no-additional

2206.14057

https://arxiv.org/abs/2206.14057v3

https://arxiv.org/pdf/2206.14057v3.pdf

Safe Exploration Incurs Nearly No Additional Sample Complexity for Reward-free RL

Reward-free reinforcement learning (RF-RL), a recently introduced RL paradigm, relies on random action-taking to explore the unknown environment without any reward feedback information. While the primary goal of the exploration phase in RF-RL is to reduce the uncertainty in the estimated model with minimum number of trajectories, in practice, the agent often needs to abide by certain safety constraint at the same time. It remains unclear how such safe exploration requirement would affect the corresponding sample complexity in order to achieve the desired optimality of the obtained policy in planning. In this work, we make a first attempt to answer this question. In particular, we consider the scenario where a safe baseline policy is known beforehand, and propose a unified Safe reWard-frEe ExploraTion (SWEET) framework. We then particularize the SWEET framework to the tabular and the low-rank MDP settings, and develop algorithms coined Tabular-SWEET and Low-rank-SWEET, respectively. Both algorithms leverage the concavity and continuity of the newly introduced truncated value functions, and are guaranteed to achieve zero constraint violation during exploration with high probability. Furthermore, both algorithms can provably find a near-optimal policy subject to any constraint in the planning phase. Remarkably, the sample complexities under both algorithms match or even outperform the state of the art in their constraint-free counterparts up to some constant factors, proving that safety constraint hardly increases the sample complexity for RF-RL.

['Yingbin Liang', 'Jing Yang', 'Ruiquan Huang']

2022-06-28

['safe-exploration']

['robots']

[ 1.18524820e-01 4.63846415e-01 -5.13760626e-01 2.01793656e-01 -1.03793991e+00 -8.79425347e-01 4.64752942e-01 2.08144754e-01 -6.47549272e-01 1.12169421e+00 5.33665344e-02 -4.76362079e-01 -8.14632952e-01 -9.00567770e-01 -8.95726204e-01 -9.47409332e-01 -4.38159674e-01 5.30409694e-01 2.90501455e-04 -2.12079078e-01 4.06090677e-01 3.69785458e-01 -1.14833283e+00 -5.26620746e-01 1.07740331e+00 1.01417017e+00 3.13936383e-01 4.55603331e-01 4.20120806e-01 5.61439216e-01 -1.05909228e-01 1.83290586e-01 6.55876338e-01 -3.00913453e-01 -6.75684392e-01 5.78254722e-02 -3.54903787e-01 -5.66769302e-01 -2.85464019e-01 1.22663414e+00 4.23056960e-01 5.69872439e-01 3.55178088e-01 -1.09018159e+00 -1.90962721e-02 6.98373139e-01 -6.08386517e-01 -1.17642984e-01 3.94164890e-01 2.53230691e-01 1.01644778e+00 -4.07123268e-01 6.67795599e-01 1.08645678e+00 7.41773332e-03 5.09504139e-01 -1.25707996e+00 -5.20299196e-01 5.98948598e-01 -3.34648043e-02 -1.14853716e+00 -1.99961156e-01 4.39496160e-01 -1.74063861e-01 4.64208394e-01 2.28549793e-01 6.91254973e-01 8.76263738e-01 1.20861478e-01 7.94126511e-01 1.36035681e+00 -1.86710060e-01 7.76539207e-01 -1.61033764e-01 -1.78038806e-01 4.91502643e-01 4.54296738e-01 7.69417465e-01 -4.36185300e-01 -2.05045491e-01 5.97804368e-01 -1.51277483e-01 -3.90590131e-01 -8.46222460e-01 -1.19554114e+00 9.99528587e-01 3.42138767e-01 3.31989229e-02 -4.31468278e-01 2.67104030e-01 2.21717894e-01 5.04605055e-01 -4.80587557e-02 6.14580035e-01 -2.56901115e-01 -3.89020115e-01 -6.83311224e-01 7.46353924e-01 7.20050395e-01 1.01272345e+00 5.04345596e-01 -1.25105847e-02 -3.25979829e-01 1.13253966e-01 1.40118927e-01 5.37650347e-01 -3.04154828e-02 -1.24031973e+00 8.16023290e-01 8.52694213e-02 9.16625977e-01 -7.19408572e-01 -3.56924891e-01 -6.80651665e-01 -4.66048777e-01 5.85108876e-01 6.48555219e-01 -2.83240587e-01 -6.80405617e-01 2.14273548e+00 5.10222673e-01 -1.02409750e-01 2.75735736e-01 9.37786400e-01 -2.87334263e-01 5.97404718e-01 -3.60305667e-01 -7.27561295e-01 8.19789648e-01 -6.66741610e-01 -6.19479179e-01 -2.26207390e-01 5.29577136e-01 -8.60073511e-03 1.14639366e+00 6.41243160e-01 -1.08607244e+00 5.09061664e-02 -1.10887361e+00 4.96799469e-01 1.95356786e-01 -3.81276250e-01 4.51404721e-01 5.44348419e-01 -6.23080730e-01 7.80913591e-01 -8.99018943e-01 -5.61010242e-02 4.14215416e-01 4.12814558e-01 -1.73221081e-01 -1.77381650e-01 -9.74337697e-01 8.68746638e-01 5.62351346e-01 1.01836286e-01 -1.44817150e+00 -4.47475374e-01 -5.72496355e-01 1.27775759e-01 1.37189233e+00 -3.46280247e-01 1.56396019e+00 -3.04577231e-01 -1.76312006e+00 -1.09926742e-02 2.22164467e-02 -7.61197329e-01 1.06136680e+00 -3.97802502e-01 2.97717333e-01 1.32220089e-01 1.80109322e-01 3.45598072e-01 6.78160369e-01 -1.28824711e+00 -7.16707289e-01 -3.17798227e-01 5.65936804e-01 4.36802566e-01 1.00459181e-01 -6.40265048e-01 -1.37094080e-01 -2.95341343e-01 7.48000816e-02 -1.08478606e+00 -8.03589582e-01 -1.89033180e-01 -3.46067101e-01 -7.90778771e-02 3.58728766e-01 -1.62956417e-01 1.07948685e+00 -1.81475508e+00 2.90821493e-01 4.76332247e-01 -1.73373118e-01 -7.67154694e-02 -2.25975677e-01 6.25440061e-01 3.93272072e-01 -3.64173367e-03 -5.30733407e-01 -2.47687399e-02 3.21913838e-01 5.55953205e-01 -7.80189037e-01 7.61223257e-01 -2.08370507e-01 7.62748182e-01 -1.33514357e+00 -1.82908744e-01 2.02957079e-01 -1.80630058e-01 -8.32898438e-01 1.76289737e-01 -5.57787359e-01 8.67867351e-01 -9.17487681e-01 3.62849534e-01 4.39066529e-01 1.53479338e-01 4.02766168e-01 5.38828194e-01 -2.57667392e-01 2.34004393e-01 -1.31409538e+00 1.58558977e+00 -4.59519863e-01 -3.53731541e-03 2.36865729e-01 -9.97422934e-01 5.83070993e-01 -4.13258225e-02 5.96189201e-01 -8.34237099e-01 2.50321925e-02 4.41516668e-01 -8.09010118e-03 -2.27301702e-01 4.17937070e-01 -1.03471264e-01 -3.14444155e-01 4.25162733e-01 -5.25290549e-01 7.14504644e-02 2.17537463e-01 -1.66416168e-02 1.08559465e+00 5.10812342e-01 6.25098169e-01 -3.08029652e-01 3.71606588e-01 -4.22610939e-02 9.10972178e-01 1.19060659e+00 -1.19102612e-01 -2.77442746e-02 9.98590708e-01 -7.52886292e-03 -6.45945668e-01 -9.70240474e-01 -7.11233914e-03 7.76809275e-01 4.60547656e-01 -1.58507213e-01 -4.52559739e-01 -9.03660893e-01 1.67656928e-01 1.01475203e+00 -7.79220521e-01 2.77174506e-02 -5.43534815e-01 -3.78844380e-01 4.16033342e-02 1.72954619e-01 3.91114980e-01 -8.25392783e-01 -1.33079743e+00 3.52212816e-01 6.71020374e-02 -8.16902161e-01 -4.11128223e-01 4.23633188e-01 -7.37941384e-01 -1.19522619e+00 -6.89409912e-01 -1.56624004e-01 7.72690296e-01 1.20396189e-01 5.37984431e-01 -3.75290334e-01 3.29764783e-01 3.18264008e-01 -4.60059434e-01 -6.34396225e-02 -7.34318644e-02 1.00028059e-02 1.89274088e-01 -1.99882552e-01 -2.58830547e-01 -6.48258507e-01 -6.13705099e-01 2.77607501e-01 -7.12438703e-01 -1.04082286e-01 6.30865812e-01 7.90543139e-01 7.77871966e-01 3.21473658e-01 6.38359368e-01 -5.35586894e-01 6.78485870e-01 -5.19584239e-01 -1.33121192e+00 9.74258557e-02 -7.65186548e-01 5.50507665e-01 7.48833895e-01 -4.43088621e-01 -6.85469389e-01 1.76263630e-01 1.43635422e-01 -3.94288957e-01 2.68512875e-01 5.61862111e-01 -2.19977885e-01 -1.63826168e-01 3.51263881e-01 3.42216313e-01 -1.53936937e-01 -2.61334747e-01 4.86508995e-01 1.71270713e-01 5.15670478e-01 -1.02910340e+00 1.09626806e+00 5.40449440e-01 5.11010885e-01 -3.10437799e-01 -9.36344802e-01 -1.37387738e-01 -2.77961791e-01 -2.22543880e-01 4.26543534e-01 -5.08178115e-01 -1.35182416e+00 -2.56947041e-01 -7.29448259e-01 -6.60304308e-01 -6.35647714e-01 4.10309017e-01 -1.17916131e+00 4.56545711e-01 1.26486436e-01 -1.58333921e+00 -6.92736916e-03 -1.16907799e+00 5.79517841e-01 1.32564679e-01 2.30050966e-01 -4.78674948e-01 1.44541740e-01 -1.22203626e-01 2.53106117e-01 7.12449253e-01 6.57896817e-01 -3.56272161e-01 -8.98652375e-01 2.15112284e-01 9.49843153e-02 -2.36292571e-01 -8.99792388e-02 -6.90178454e-01 -3.92670512e-01 -6.39514983e-01 1.47698089e-01 -4.47146535e-01 9.13040161e-01 3.64555687e-01 1.05397344e+00 -8.50195169e-01 -2.98279881e-01 3.96535069e-01 1.48145854e+00 5.05191445e-01 3.74152809e-01 5.79271317e-01 6.62107542e-02 7.35390723e-01 1.33929873e+00 9.69137192e-01 2.45915994e-01 8.20559263e-01 8.74110699e-01 5.50243795e-01 6.93780780e-01 -5.50972581e-01 5.55123031e-01 1.33193418e-01 5.59528060e-02 -1.34878770e-01 -6.01214647e-01 5.86611927e-01 -2.33133507e+00 -9.15588081e-01 4.28762019e-01 2.89646482e+00 9.04685140e-01 2.25517482e-01 2.82090992e-01 8.45431238e-02 3.13159466e-01 1.31582528e-01 -1.08321810e+00 -3.88938159e-01 2.05926716e-01 -9.24004475e-04 9.99552846e-01 7.56583929e-01 -8.36296737e-01 7.47270286e-01 5.59294271e+00 8.45004618e-01 -7.93640971e-01 3.86756174e-02 4.13919121e-01 -4.36990380e-01 -3.68911862e-01 3.16992760e-01 -6.56509697e-01 3.33701730e-01 8.57581675e-01 -5.77238083e-01 9.39798474e-01 1.11728239e+00 4.91352886e-01 -5.59624553e-01 -1.13664520e+00 6.30662799e-01 -5.71627617e-01 -1.13854933e+00 -4.35996115e-01 4.00642544e-01 6.79817080e-01 -1.76410139e-01 1.09782107e-01 4.36168045e-01 6.38883770e-01 -1.03750086e+00 9.20763016e-01 4.99195516e-01 7.37373233e-01 -1.28319693e+00 4.09527183e-01 6.96314335e-01 -1.12505960e+00 -6.28830612e-01 -3.03565174e-01 -1.30546302e-01 3.54680747e-01 3.53918821e-01 -5.77595472e-01 7.91846216e-01 2.31166154e-01 3.70051891e-01 1.13240987e-01 1.10076857e+00 -4.87071455e-01 2.56763637e-01 -4.37917948e-01 -1.56320676e-01 7.54214704e-01 -4.72091258e-01 9.24080729e-01 5.63288927e-01 3.22856873e-01 1.14066333e-01 6.04821086e-01 8.48880529e-01 3.37940812e-01 -1.94800913e-01 -5.37795305e-01 -1.50708705e-01 5.51083386e-01 8.67540956e-01 -6.29976034e-01 1.09392986e-01 4.35466086e-03 5.09721994e-01 3.50762844e-01 3.74465913e-01 -8.57624769e-01 -2.51417309e-01 6.27079964e-01 -2.12872609e-01 4.06032443e-01 -4.48218614e-01 -2.87269890e-01 -8.12815130e-01 2.16811180e-01 -8.30234647e-01 3.36870193e-01 -5.21446988e-02 -8.52880120e-01 1.93594575e-01 1.14720039e-01 -1.28918862e+00 -6.76356614e-01 -2.05788270e-01 -3.43906820e-01 5.90157211e-01 -1.84557259e+00 -4.81348991e-01 1.31781459e-01 4.60759372e-01 4.60783690e-01 1.16186395e-01 5.47356069e-01 -1.92453384e-01 -4.37832445e-01 3.86308193e-01 4.02164072e-01 -4.59607959e-01 2.77826905e-01 -1.35340726e+00 -8.41668323e-02 8.93411219e-01 -2.44033024e-01 4.36803758e-01 9.58280325e-01 -7.17636943e-01 -1.81340742e+00 -9.43067551e-01 6.61394969e-02 -9.15919840e-02 6.12373173e-01 -2.82586366e-01 -5.39206803e-01 4.58662242e-01 -2.66469151e-01 -8.72337669e-02 -2.59975884e-02 -4.73733991e-02 1.35860837e-03 -3.84621369e-03 -1.16851449e+00 8.81861508e-01 1.00667548e+00 -4.29962091e-02 -4.61011052e-01 2.91529268e-01 8.73336911e-01 -5.31392574e-01 -6.52843773e-01 5.23976445e-01 4.76044387e-01 -8.38476598e-01 7.43547201e-01 -6.17682695e-01 1.33773640e-01 -3.89270633e-01 -3.60290051e-01 -1.17682314e+00 -7.16110840e-02 -1.35695100e+00 -4.20749933e-01 6.34555221e-01 2.55942702e-01 -8.13949287e-01 8.49394083e-01 4.93316054e-01 -4.21046279e-02 -1.12403011e+00 -1.41589606e+00 -1.40763712e+00 3.43642920e-01 -5.66401780e-01 6.11145437e-01 4.37472105e-01 2.34755933e-01 -2.19532177e-01 -6.27597094e-01 3.80660683e-01 8.84751081e-01 5.00656247e-01 7.66876161e-01 -8.94507706e-01 -6.65906966e-01 -3.62317979e-01 3.59412998e-01 -1.26073945e+00 1.73823625e-01 -6.23588860e-01 2.63030171e-01 -1.49992585e+00 5.80277443e-02 -8.99517715e-01 -2.46963814e-01 4.25164014e-01 5.97583055e-02 -5.61320305e-01 4.53108042e-01 1.79428697e-01 -6.82211161e-01 1.05842280e+00 1.50939941e+00 1.65074185e-01 -6.31760418e-01 3.52056593e-01 -7.31295347e-01 4.92558479e-01 8.14644098e-01 -4.83758718e-01 -6.50883257e-01 6.72917813e-03 5.21770239e-01 7.20257699e-01 8.87608901e-02 -7.38556862e-01 1.02493137e-01 -8.79999816e-01 -2.52279639e-01 -6.33829892e-01 2.14322358e-01 -8.91105950e-01 5.78982895e-03 9.44766283e-01 -6.21648312e-01 -2.48431996e-01 -1.47577494e-01 1.09119058e+00 2.82273740e-01 -4.32119042e-01 7.82055259e-01 4.19448465e-02 -4.43314582e-01 5.29033959e-01 -2.48297632e-01 1.35463655e-01 1.25162125e+00 3.63471285e-02 -1.89658195e-01 -6.25979125e-01 -5.30216753e-01 7.16754377e-01 3.49536538e-01 9.12859440e-02 4.95106578e-01 -1.15196824e+00 -4.10915822e-01 -1.62919104e-01 -6.62547275e-02 1.84156045e-01 2.21571922e-02 1.08392382e+00 1.08177207e-01 6.87321305e-01 -6.62659705e-02 -1.29297569e-01 -5.94993055e-01 8.79706025e-01 3.54719520e-01 -7.13717520e-01 -7.62438297e-01 3.02169591e-01 2.17781112e-01 -2.98341483e-01 4.32707310e-01 -4.15409088e-01 -8.83919466e-03 -1.19958587e-01 4.68295723e-01 5.48078775e-01 -3.51512909e-01 3.91299278e-02 -1.77124441e-01 3.97925526e-01 2.86877491e-02 -6.28059089e-01 1.28691530e+00 -2.04440355e-01 4.30909753e-01 2.15796277e-01 6.00977600e-01 1.28099069e-01 -1.80489111e+00 -1.87801182e-01 2.70261139e-01 -6.92198992e-01 5.29147200e-02 -8.48312616e-01 -8.27016294e-01 5.24796963e-01 3.34260136e-01 1.71141297e-01 9.97625470e-01 -3.04356307e-01 6.51912808e-01 8.09081733e-01 9.72391665e-01 -1.16246378e+00 -4.01368216e-02 5.10637999e-01 9.50566828e-01 -8.73955429e-01 7.89700970e-02 -1.32623255e-01 -7.10460424e-01 9.11404312e-01 3.36453229e-01 -2.63751090e-01 2.36685369e-02 6.12410568e-02 -7.39250898e-01 2.20764935e-01 -8.73231351e-01 -6.05165601e-01 -1.12109870e-01 5.20122290e-01 -3.61839980e-01 1.20414428e-01 -5.89262366e-01 4.55859780e-01 -1.60928667e-01 -6.66922629e-02 6.18692636e-01 9.84264135e-01 -8.28165352e-01 -1.17161822e+00 -4.02991295e-01 3.30428839e-01 -2.10840061e-01 3.24699461e-01 -1.08354632e-02 8.82812202e-01 -3.40531468e-01 1.08074653e+00 -2.82885432e-01 -9.60670561e-02 1.61262393e-01 -3.57604057e-01 5.67296028e-01 -3.81256223e-01 -1.80889711e-01 2.10692942e-01 1.22064896e-01 -1.09062707e+00 -1.61154673e-01 -6.93989098e-01 -1.42106926e+00 -1.73501998e-01 -1.93978518e-01 3.28825623e-01 3.65669280e-01 1.09829497e+00 2.17970163e-01 3.19373310e-01 9.39080119e-01 -6.03639364e-01 -1.42754650e+00 -4.84835416e-01 -6.06362522e-01 -3.06918830e-01 5.43074787e-01 -9.34679449e-01 -4.60800767e-01 -8.35203707e-01]

[4.395223617553711, 2.504487991333008]

3db8a78f-29e2-450e-8515-493abdeaf2be

nonlinear-trend-removal-should-be-carefully

1605.05891

http://arxiv.org/abs/1605.05891v1

http://arxiv.org/pdf/1605.05891v1.pdf

Nonlinear trend removal should be carefully performed in heart rate variability analysis

$\bullet$ Background : In Heart rate variability analysis, the rate-rate time series suffer often from aperiodic non-stationarity, presence of ectopic beats etc. It would be hard to extract helpful information from the original signals. 10 $\bullet$ Problem : Trend removal methods are commonly practiced to reduce the influence of the low frequency and aperiodic non-stationary in RR data. This can unfortunately affect the signal and make the analysis on detrended data less appropriate. $\bullet$ Objective : Investigate the detrending effect (linear \& nonlinear) in temporal / nonliear analysis of heart rate variability of long-term RR data (in normal sinus rhythm, atrial fibrillation, 15 congestive heart failure and ventricular premature arrhythmia conditions). $\bullet$ Methods : Temporal method : standard measure SDNN; Nonlinear methods : multi-scale Fractal Dimension (FD), Detrended Fluctuation Analysis (DFA) \& Sample Entropy (Sam-pEn) analysis. $\bullet$ Results : The linear detrending affects little the global characteristics of the RR data, either 20 in temporal analysis or in nonlinear complexity analysis. After linear detrending, the SDNNs are just slightly shifted and all distributions are well preserved. The cross-scale complexity remained almost the same as the ones for original RR data or correlated. Nonlinear detrending changed not only the SDNNs distribution, but also the order among different types of RR data. After this processing, the SDNN became indistinguishable be-25 tween SDNN for normal sinus rhythm and ventricular premature beats. Different RR data has different complexity signature. Nonlinear detrending made the all RR data to be similar , in terms of complexity. It is thus impossible to distinguish them. The FD showed that nonlinearly detrended RR data has a dimension close to 2, the exponent from DFA is close to zero and SampEn is larger than 1.5 -- these complexity values are very close to those for 30 random signal. $\bullet$ Conclusions : Pre-processing by linear detrending can be performed on RR data, which has little influence on the corresponding analysis. Nonlinear detrending could be harmful and it is not advisable to use this type of pre-processing. Exceptions do exist, but only combined with other appropriate techniques to avoid complete change of the signal's intrinsic dynamics. 35 Keywords $\bullet$ heart rate variability $\bullet$ linear / nonlinear detrending $\bullet$ complexity analysis $\bullet$ mul-tiscale analysis $\bullet$ detrended fluctuation analysis $\bullet$ fractal dimension $\bullet$ sample entropy;

[]

2016-05-19

['heart-rate-variability']

['medical']

[-1.09268732e-01 -5.39239883e-01 2.79648483e-01 -3.39782871e-02 1.36784017e-01 -6.65301383e-01 2.16489464e-01 -2.27493137e-01 -1.84296697e-01 1.16352522e+00 -2.83225439e-02 -3.48902404e-01 -5.69872618e-01 -4.85663325e-01 8.70499387e-02 -8.68010879e-01 -8.38263631e-01 8.78901258e-02 -9.28565040e-02 -2.94504672e-01 2.71582127e-01 7.11211920e-01 -1.40325511e+00 -1.05579451e-01 8.48018527e-01 6.71026349e-01 -9.81602967e-02 9.52432513e-01 -1.94578916e-01 6.01758301e-01 -9.10993278e-01 3.95632207e-01 2.13547379e-01 -1.09458447e+00 -2.59328902e-01 -5.17569542e-01 -3.97935688e-01 1.00494802e-01 -1.70163941e-02 8.75225782e-01 8.84507418e-01 1.03789940e-01 8.49674582e-01 -9.17179406e-01 -2.83962727e-01 6.25384569e-01 -7.23919809e-01 1.05873132e+00 1.83411688e-01 1.53492123e-01 -1.38986884e-02 -5.04872203e-01 5.52022576e-01 9.59021211e-01 1.24462712e+00 3.69415522e-01 -1.37766528e+00 -6.11157000e-01 -5.03668189e-01 -1.47827148e-01 -1.36248016e+00 -2.22397581e-01 1.01301467e+00 -5.99145293e-01 7.09851623e-01 5.80871046e-01 1.03557539e+00 6.86017513e-01 8.85157049e-01 -3.52895379e-01 1.48137283e+00 -3.26449633e-01 -1.14215195e-01 -1.06066763e-01 6.78448439e-01 4.10971999e-01 4.13801968e-01 3.76892924e-01 6.52632490e-02 -3.99112135e-01 8.29522312e-01 1.87799856e-01 -4.67800051e-01 3.83294344e-01 -1.41539383e+00 2.87887454e-01 -1.91674791e-02 1.14609206e+00 -4.67412055e-01 7.81968459e-02 6.82842851e-01 8.77128243e-01 3.10030878e-01 4.21954095e-01 -4.25405204e-01 -6.49681687e-01 -9.96386766e-01 3.26331049e-01 6.92872643e-01 3.17530543e-01 3.89540195e-01 5.74257851e-01 -8.71119350e-02 7.68961251e-01 5.11856079e-02 9.32199717e-01 9.92319584e-01 -7.45703340e-01 -1.97395757e-02 2.38856465e-01 -2.13412233e-02 -1.19297767e+00 -7.22962737e-01 -5.96766651e-01 -1.31809890e+00 4.09677804e-01 8.00873220e-01 -3.05348814e-01 -5.72681546e-01 1.59566140e+00 -1.25043899e-01 1.27206355e-01 9.65826362e-02 5.97066402e-01 6.55705452e-01 5.94760060e-01 -6.98561296e-02 -1.23056185e+00 1.33629072e+00 -4.19966169e-02 -1.35822749e+00 4.57992971e-01 2.65859723e-01 -9.66404319e-01 9.41278577e-01 2.39928216e-01 -9.98893559e-01 -6.58750355e-01 -9.11391497e-01 6.06683075e-01 -1.55122310e-01 -2.53958613e-01 1.46461114e-01 6.95226669e-01 -8.31650674e-01 1.21181917e+00 -7.64856100e-01 -3.15838575e-01 -9.06432420e-02 1.13361320e-02 -5.25702983e-02 4.87545609e-01 -1.20589030e+00 9.47309494e-01 9.52197015e-02 2.76297480e-01 -3.67889949e-03 -5.90548456e-01 -3.84738028e-01 -1.18092604e-01 -4.34424043e-01 -5.71704149e-01 6.62498713e-01 -7.87925482e-01 -1.19954920e+00 4.69782889e-01 -3.46410334e-01 -3.38996202e-01 5.82103133e-01 3.15228283e-01 -1.08424735e+00 -2.28072181e-02 1.44906670e-01 -4.93506789e-01 1.02394915e+00 -8.82310212e-01 -1.24625266e-02 -3.80637884e-01 -9.56652522e-01 -1.44484639e-01 3.15477014e-01 -8.64674002e-02 5.60282290e-01 -9.35032606e-01 5.72705626e-01 -9.30991352e-01 -2.13914812e-02 -7.03102112e-01 -1.59205854e-01 8.53191838e-02 1.04278493e+00 -6.16504014e-01 1.82463408e+00 -2.42879438e+00 -4.82186466e-01 4.74848747e-01 3.25770080e-01 1.80229381e-01 3.54106724e-01 4.04957145e-01 -7.69600630e-01 5.15132904e-01 -5.08793116e-01 3.66173476e-01 -4.91820246e-01 1.85564071e-01 -3.10237229e-01 6.94131374e-01 -1.61233604e-01 5.18253207e-01 -7.06515431e-01 -3.73386681e-01 1.10306308e-01 3.97657067e-01 7.35378265e-02 -2.09229931e-01 5.50606608e-01 6.25864744e-01 -9.31192786e-02 2.94597864e-01 8.92075658e-01 1.79350793e-01 -2.43288696e-01 -1.51249409e-01 -5.73538065e-01 -3.11644882e-01 -1.12390542e+00 1.11464536e+00 -1.96998581e-01 9.65846658e-01 -2.33305156e-01 -8.67552638e-01 1.28958511e+00 6.15243435e-01 8.54351997e-01 -4.00529385e-01 5.09030484e-02 5.46780229e-01 6.15037918e-01 -7.72954583e-01 -2.88281552e-02 -6.99735999e-01 5.54725453e-02 5.29684722e-01 -4.93327618e-01 -1.56400174e-01 1.42373621e-01 -2.80708253e-01 1.00388312e+00 -2.27279235e-02 5.80141962e-01 -8.64773870e-01 5.63789010e-01 -1.60146862e-01 6.96421087e-01 6.40922725e-01 -3.84996474e-01 6.52554691e-01 7.09022343e-01 -8.13489318e-01 -7.88836956e-01 -1.15695238e+00 -6.44681633e-01 1.65044561e-01 -1.31514028e-01 -3.55361030e-02 -4.19680297e-01 1.40050910e-02 -5.42486757e-02 4.74791348e-01 -5.35167396e-01 -8.12141746e-02 -9.22380090e-01 -1.30233455e+00 6.70432091e-01 -9.20184031e-02 5.16489804e-01 -1.39466226e+00 -8.36536288e-01 6.59871399e-01 -3.23481560e-01 -6.53550208e-01 -2.01447234e-01 2.07334071e-01 -1.48665190e+00 -9.29620504e-01 -8.04506540e-01 -3.33102405e-01 3.55978578e-01 -1.88299671e-01 1.08227158e+00 -9.47775878e-03 -6.25272155e-01 -8.11383277e-02 -1.94550097e-01 -6.63835049e-01 -5.88736236e-01 -4.66878235e-01 3.21143746e-01 -3.71836841e-01 2.33875141e-01 -1.13883972e+00 -7.81351924e-01 4.18610036e-01 -5.74553609e-01 -7.91209340e-01 2.45699719e-01 6.67678773e-01 5.75221777e-01 3.88422549e-01 1.18864691e+00 -5.61960936e-01 1.19460964e+00 -4.14293945e-01 -2.16732740e-01 -4.50159967e-01 -9.89513576e-01 -1.30249590e-01 1.00387824e+00 -4.50589985e-01 -4.49638754e-01 -5.14132977e-01 2.11355805e-01 -4.94542450e-01 2.94926241e-02 3.54124725e-01 5.25981784e-01 2.81796455e-01 1.09804893e+00 4.63804543e-01 5.10209560e-01 -3.11079204e-01 -3.00425291e-01 4.46192145e-01 4.96367514e-01 -6.52578026e-02 6.01764619e-01 4.64739442e-01 1.87575981e-01 -1.15865743e+00 8.19842741e-02 -2.69013673e-01 -3.29729974e-01 -2.21853346e-01 9.54512239e-01 -4.20687646e-01 -6.62162006e-01 4.90708411e-01 -1.01921177e+00 7.75117576e-02 -8.74702811e-01 8.72543812e-01 -6.44607306e-01 5.17895520e-01 -5.81429005e-01 -1.05333829e+00 -6.02289081e-01 -5.84137857e-01 -1.06193900e-01 2.22771510e-01 -5.18899679e-01 -9.31046247e-01 5.05204558e-01 -3.13371122e-01 6.61791205e-01 1.03520024e+00 8.80976856e-01 -2.24518359e-01 2.62347817e-01 -2.87381232e-01 1.25703424e-01 4.49346930e-01 5.10690987e-01 5.29745579e-01 -8.51873398e-01 -1.56035632e-01 7.13782907e-01 7.60726929e-01 5.67163646e-01 1.03338897e+00 6.94260001e-01 -1.44669652e-01 -2.65414268e-01 5.35239458e-01 1.43645906e+00 8.43285084e-01 9.52841699e-01 8.03385377e-02 2.66754568e-01 3.76657963e-01 3.61928821e-01 4.74355042e-01 -3.50361854e-01 1.90875322e-01 -8.04256722e-02 -1.33118406e-01 -4.34877090e-02 4.05714810e-01 3.81349176e-01 1.08760059e+00 -9.35615301e-01 2.97744930e-01 -8.73448551e-01 4.28020358e-01 -1.41463697e+00 -1.78572869e+00 -8.34610879e-01 2.22226787e+00 7.21797109e-01 2.29159534e-01 3.53220940e-01 6.00367844e-01 8.95736516e-01 1.53940231e-01 -4.54386652e-01 -1.08945203e+00 -3.18495035e-01 3.01766247e-01 2.40037471e-01 3.58257234e-01 -7.06603527e-01 9.05376077e-02 6.16156864e+00 3.50672334e-01 -1.55202806e+00 -1.14967003e-01 4.19942677e-01 -5.96867576e-02 -1.36640668e-01 -1.11488268e-01 -3.12530935e-01 9.52857375e-01 1.35024595e+00 -6.76189542e-01 2.41052151e-01 4.52874720e-01 8.88265729e-01 -2.38714859e-01 -5.26893497e-01 1.51772249e+00 -3.92567337e-01 -1.04274082e+00 -5.58479488e-01 1.31046638e-01 3.74539912e-01 -2.01033689e-02 -1.88578770e-01 3.15181136e-01 -5.17632544e-01 -1.05057013e+00 1.64401412e-01 1.17839432e+00 1.02805555e+00 -8.66369724e-01 1.02896559e+00 3.44934404e-01 -1.46256936e+00 -1.37107402e-01 -2.87654996e-01 -2.36238793e-01 2.76162893e-01 1.21920145e+00 -3.85448545e-01 3.93250585e-01 7.74756432e-01 6.22068167e-01 -2.60143161e-01 7.90844560e-01 4.98016775e-01 6.50616527e-01 -4.29775655e-01 -2.63937116e-02 -3.30408603e-01 -5.20629406e-01 1.20948291e+00 1.11790097e+00 5.75563610e-01 3.81772071e-01 -5.30204296e-01 9.06301916e-01 6.66935802e-01 1.85229868e-01 -9.77766633e-01 -2.26124115e-02 4.12763625e-01 8.87969732e-01 -8.33409011e-01 -3.44710082e-01 -1.01509452e-01 4.63103235e-01 -7.94197321e-01 4.03180450e-01 -7.81322598e-01 -1.00113225e+00 4.84668732e-01 5.48197091e-01 -2.73611784e-01 -1.89153820e-01 -6.87202692e-01 -1.06946397e+00 6.68573231e-02 -7.91426361e-01 3.51834804e-01 -4.00337309e-01 -1.25360703e+00 1.05577874e+00 1.33821979e-01 -1.74759054e+00 -5.89571655e-01 -1.59487985e-02 -1.02644169e+00 1.43227959e+00 -8.65782261e-01 -2.21632104e-02 -1.64200068e-01 7.53270268e-01 1.82114035e-01 -1.55697331e-01 9.49371815e-01 1.58209875e-01 -2.96110123e-01 1.31416425e-01 2.26342201e-01 -1.14791304e-01 6.30144477e-01 -1.34978724e+00 -1.03549860e-01 7.26476848e-01 -3.90314221e-01 8.40193212e-01 1.07520950e+00 -6.96822464e-01 -6.49981856e-01 -5.20726562e-01 1.00817323e+00 -8.56007040e-02 4.15048718e-01 9.30758566e-02 -9.50071156e-01 -1.07503571e-01 1.33281589e-01 2.59472728e-01 7.61831999e-01 -1.43652201e-01 4.75463867e-01 -3.93177688e-01 -1.29156709e+00 4.38137710e-01 5.98632634e-01 -3.64097655e-01 -9.33124483e-01 -4.61712806e-03 3.43499929e-01 1.70297816e-01 -1.28779948e+00 5.06244302e-01 9.42310452e-01 -1.48331714e+00 8.14631820e-01 -8.01993310e-02 2.13215668e-02 -4.68432069e-01 3.67395401e-01 -1.01401186e+00 -5.40429950e-01 -1.13147485e+00 9.28592831e-02 9.58054900e-01 2.96390414e-01 -1.18781531e+00 1.71158150e-01 1.81185365e-01 -9.52503532e-02 -5.63310504e-01 -9.84614491e-01 -1.02071059e+00 -1.53186560e-01 -1.81363642e-01 1.68614745e-01 1.20574927e+00 3.17442775e-01 6.35890439e-02 -8.62945169e-02 -1.67735279e-01 6.68786049e-01 1.83773041e-01 3.09267312e-01 -1.53055286e+00 -2.64613181e-01 -5.27580738e-01 -5.55083275e-01 1.60870180e-02 -4.76337582e-01 -4.84216958e-01 -4.02014941e-01 -1.24470234e+00 -4.13554788e-01 -4.95585084e-01 -5.86714327e-01 4.97144535e-02 -1.10695973e-01 1.08866528e-01 6.17450178e-02 6.22403145e-01 6.02305472e-01 3.58569697e-02 1.43127632e+00 4.05491620e-01 -9.69670057e-01 4.03771132e-01 -3.32666159e-01 8.39934289e-01 9.33960140e-01 -6.67961717e-01 -5.98334551e-01 5.70376813e-01 6.60118759e-02 7.10455358e-01 1.96919411e-01 -1.35754323e+00 -3.72445524e-01 7.52171129e-02 5.31675577e-01 -7.28565454e-01 -1.83115482e-01 -7.41628349e-01 7.15791345e-01 8.71314526e-01 2.55419731e-01 7.17681110e-01 3.00708055e-01 3.92099023e-01 -3.86439323e-01 -2.02789992e-01 1.02893496e+00 -3.25997174e-01 -9.52913798e-03 3.50545868e-02 -7.70059466e-01 1.77993819e-01 8.84071231e-01 -7.39691675e-01 -1.71357334e-01 -3.10279816e-01 -1.45384693e+00 -4.31933969e-01 2.43013680e-01 -5.16678654e-02 3.60237777e-01 -1.20374548e+00 -6.63909256e-01 2.69672871e-01 -3.92976046e-01 -2.78945625e-01 7.83950686e-01 1.42747235e+00 -8.56718838e-01 7.70217031e-02 -5.98891318e-01 -6.96691513e-01 -1.21518493e+00 5.67653179e-01 6.14907444e-01 -2.38539979e-01 -7.79201686e-01 2.85165608e-01 -3.91541243e-01 3.10068339e-01 -3.67186695e-01 -9.12487626e-01 -6.34467661e-01 4.32636619e-01 4.52536404e-01 7.59144008e-01 -2.44993314e-01 -4.05323863e-01 -5.11394918e-01 1.05754876e+00 3.81192237e-01 -1.33557707e-01 1.12037897e+00 -3.68476272e-01 -4.64040339e-01 1.13555956e+00 1.22307980e+00 3.57851207e-01 -6.11506701e-01 4.93726224e-01 -1.79597184e-01 -1.36863083e-01 -3.98065925e-01 -5.44569194e-01 -6.93956494e-01 5.76610625e-01 1.07281077e+00 1.02591324e+00 1.45332849e+00 -5.98540008e-01 4.57967073e-01 3.97598743e-02 1.11803398e-01 -1.02451229e+00 -3.37478638e-01 4.91054654e-01 1.04928160e+00 -7.10126579e-01 1.13719448e-01 -8.77755135e-02 -4.38533723e-01 1.51086199e+00 -1.37900636e-01 -5.88419676e-01 1.22411108e+00 2.52454370e-01 2.41620287e-01 -1.11643031e-01 -5.23297071e-01 2.00961143e-01 -1.19507208e-01 4.55873549e-01 7.44634092e-01 -1.12042706e-02 -1.23227429e+00 4.37435687e-01 -3.62386107e-01 2.60228008e-01 8.57961774e-01 7.92447567e-01 -2.62257576e-01 -7.78134346e-01 -7.13429093e-01 6.46994591e-01 -7.34634519e-01 -9.18876156e-02 2.12674826e-01 9.05541658e-01 3.82589638e-01 8.57158184e-01 1.90610774e-02 -3.02307576e-01 4.49808359e-01 5.00436068e-01 1.69266954e-01 -7.99990222e-02 -7.45668471e-01 4.58496273e-01 -8.20258409e-02 -4.43420112e-01 -4.27445948e-01 -8.12054753e-01 -1.52923846e+00 -4.58782494e-01 -2.52540141e-01 4.28917646e-01 6.94991767e-01 5.13686240e-01 1.59432143e-01 7.24111974e-01 7.95238733e-01 -2.67033815e-01 -3.14543605e-01 -1.21556783e+00 -1.22908151e+00 4.06610489e-01 7.02592731e-01 -2.44944096e-01 -1.09179735e+00 3.99605125e-01]

[14.07344913482666, 3.1172266006469727]

797128e7-2bca-42c9-86e1-b925e6c0acbc

eel-efficiently-encoding-lattices-for

2306.00947

https://arxiv.org/abs/2306.00947v1

https://arxiv.org/pdf/2306.00947v1.pdf

EEL: Efficiently Encoding Lattices for Reranking

Standard decoding approaches for conditional text generation tasks typically search for an output hypothesis with high model probability, but this may not yield the best hypothesis according to human judgments of quality. Reranking to optimize for "downstream" metrics can better optimize for quality, but many metrics of interest are computed with pre-trained language models, which are slow to apply to large numbers of hypotheses. We explore an approach for reranking hypotheses by using Transformers to efficiently encode lattices of generated outputs, a method we call EEL. With a single Transformer pass over the entire lattice, we can approximately compute a contextualized representation of each token as if it were only part of a single hypothesis in isolation. We combine this approach with a new class of token-factored rerankers (TFRs) that allow for efficient extraction of high reranker-scoring hypotheses from the lattice. Empirically, our approach incurs minimal degradation error compared to the exponentially slower approach of encoding each hypothesis individually. When applying EEL with TFRs across three text generation tasks, our results show both substantial speedup compared to naive reranking and often better performance on downstream metrics than comparable approaches.

['Greg Durrett', 'Xi Ye', 'Jiacheng Xu', 'Prasann Singhal']

2023-06-01

['conditional-text-generation']

['natural-language-processing']

[ 5.67095280e-01 5.59906900e-01 -1.30415335e-01 -2.46905267e-01 -1.55619347e+00 -7.16207683e-01 8.97441208e-01 4.47506011e-01 -4.75848258e-01 1.09101605e+00 8.73473227e-01 -4.08941031e-01 3.68181020e-01 -7.96665192e-01 -7.05784082e-01 -2.98846066e-01 -3.72547619e-02 8.91660631e-01 3.81937295e-01 -3.83813173e-01 4.17654097e-01 -5.70172332e-02 -1.53258622e+00 7.03307688e-01 9.36897278e-01 5.44763803e-01 4.09203947e-01 1.04015243e+00 -2.76578575e-01 5.84766984e-01 -9.65031445e-01 -6.61208749e-01 3.61945629e-02 -7.69471288e-01 -1.06419802e+00 -2.09446341e-01 3.55425596e-01 -2.89130777e-01 1.07788295e-01 6.99305177e-01 4.36304957e-01 -9.49743204e-03 7.78353333e-01 -8.08993876e-01 -3.06397736e-01 1.44088769e+00 -6.87155640e-03 1.69082582e-01 6.94283426e-01 -2.11953253e-01 1.72849345e+00 -9.40384686e-01 7.19292760e-01 1.27238500e+00 5.44229746e-01 5.05280554e-01 -1.45019901e+00 -4.22407120e-01 2.55697131e-01 -8.44110996e-02 -1.19927454e+00 -6.23538911e-01 -9.03661363e-03 -1.54994935e-01 1.46160805e+00 3.36090505e-01 6.61782742e-01 9.60086823e-01 1.39030412e-01 9.95568335e-01 8.75584304e-01 -7.06176221e-01 1.36100397e-01 3.03054657e-02 -1.53892651e-01 7.61585116e-01 2.18289346e-01 -1.60300821e-01 -8.70716512e-01 -5.37294507e-01 3.96147698e-01 -5.51757872e-01 -2.03165084e-01 2.09322080e-01 -1.67597854e+00 7.82301128e-01 -4.09498885e-02 3.57302785e-01 -1.94148690e-01 4.61366057e-01 2.92307436e-01 4.17268932e-01 5.83085179e-01 8.64191115e-01 -6.75972104e-01 -4.37981009e-01 -1.45942473e+00 6.58068359e-01 1.00269055e+00 1.09039259e+00 6.60751462e-01 -2.04337388e-01 -7.86180317e-01 7.59008527e-01 2.71696985e-01 2.13458344e-01 7.55500436e-01 -7.64806092e-01 6.14567161e-01 1.61741316e-01 3.45050633e-01 -2.98634708e-01 8.19254443e-02 -4.86627817e-01 -3.90775800e-01 -3.47668022e-01 4.30472553e-01 -9.60906669e-02 -1.02547789e+00 1.83846116e+00 -5.93848005e-02 -6.98606074e-02 1.60302818e-01 3.94456804e-01 1.24549486e-01 9.95542586e-01 1.64507087e-02 -3.01502109e-01 1.31905770e+00 -9.01448488e-01 -3.15791160e-01 -3.56878221e-01 9.45329487e-01 -9.82060254e-01 1.05201674e+00 5.87520063e-01 -1.33413780e+00 -1.37423053e-01 -1.13383138e+00 -1.26693189e-01 -1.03277050e-01 6.21229447e-02 4.96359110e-01 5.06607533e-01 -1.42581785e+00 8.89499903e-01 -6.23963118e-01 -1.80471197e-01 3.48509885e-02 2.43361667e-01 3.46603654e-02 1.12716191e-01 -1.38710153e+00 1.01452708e+00 4.61673349e-01 -3.17775548e-01 -9.88894224e-01 -4.20234054e-01 -6.64137959e-01 1.09737463e-01 7.09437504e-02 -7.60769963e-01 1.92843044e+00 -9.04678762e-01 -1.39087212e+00 4.13086951e-01 -7.82688558e-01 -5.83794773e-01 3.29936773e-01 -2.72559375e-01 9.43767373e-03 -7.41705596e-02 3.89802337e-01 7.63749421e-01 7.53490686e-01 -1.06055605e+00 -9.19525266e-01 1.90713853e-01 -1.02755412e-01 5.36567569e-01 -2.85655260e-01 2.29672134e-01 -2.83661455e-01 -6.02774858e-01 2.16983497e-01 -5.01839757e-01 -3.96448016e-01 -5.64033926e-01 -4.69253004e-01 -5.10361910e-01 5.21563180e-02 -6.26811862e-01 1.45258570e+00 -1.37478566e+00 1.63414761e-01 2.02175722e-01 2.49544084e-01 1.87235884e-02 -1.86482236e-01 6.15662992e-01 2.26249710e-01 5.04277050e-01 -1.82826370e-01 -3.64324838e-01 2.61491597e-01 3.77497822e-02 -5.93148947e-01 -1.32020697e-01 2.99753755e-01 1.00134313e+00 -1.25613618e+00 -6.83801293e-01 -3.95289838e-01 4.05350216e-02 -7.82563806e-01 2.20554382e-01 -7.28981674e-01 -3.21587861e-01 -3.72421205e-01 3.86322916e-01 -2.46148594e-02 -3.85686487e-01 3.03750098e-01 1.60329029e-01 -4.33010310e-02 1.08392632e+00 -1.06673062e+00 1.45394039e+00 -5.77035248e-01 5.32259762e-01 -3.14727515e-01 -5.55745840e-01 6.88946664e-01 6.33735001e-01 1.07695132e-01 -4.46404129e-01 -3.18063319e-01 5.81800640e-01 -6.87565580e-02 3.13488320e-02 9.75488484e-01 -5.08465886e-01 -2.99474448e-01 8.95839095e-01 1.48036927e-01 -3.16127449e-01 4.73634303e-01 6.29164755e-01 1.38737798e+00 2.25051790e-01 3.68331313e-01 -1.42587066e-01 1.08879171e-01 1.54525056e-01 4.26120579e-01 1.07131529e+00 3.91912550e-01 6.52627587e-01 5.63935518e-01 -5.69246970e-02 -1.39035773e+00 -1.04036713e+00 1.90308183e-01 1.38232005e+00 -3.40131342e-01 -1.02879465e+00 -8.01851988e-01 -6.86193049e-01 -2.14050069e-01 9.51077819e-01 -8.54960009e-02 8.17696154e-02 -7.18111396e-01 -9.21278417e-01 9.55448747e-01 4.43705648e-01 -9.74784419e-02 -1.01364982e+00 -4.03709978e-01 5.91079295e-01 -6.61021292e-01 -7.59987593e-01 -6.01097941e-01 4.96877730e-01 -9.90507841e-01 -5.70111394e-01 -8.32609892e-01 -8.30235839e-01 6.79879129e-01 -2.28647366e-01 1.54941165e+00 2.72784054e-01 1.64202914e-01 -1.06811538e-01 -4.77912307e-01 -1.07556343e-01 -8.34253252e-01 4.25925881e-01 -6.34464473e-02 -4.26191926e-01 1.33354083e-01 -2.73359716e-01 -2.07286239e-01 1.01872543e-02 -7.50714600e-01 3.57316256e-01 8.64377260e-01 1.11115348e+00 3.60905826e-01 -1.06117658e-01 5.77021241e-01 -8.87941301e-01 1.09389448e+00 -2.85930097e-01 -3.34082067e-01 4.01878566e-01 -8.21741819e-01 7.54796267e-01 8.01977992e-01 -2.55579561e-01 -9.25726175e-01 -1.58534721e-01 -2.68354028e-01 2.85697639e-01 9.70605835e-02 5.13888478e-01 1.11698605e-01 5.69703043e-01 8.21121573e-01 3.22506309e-01 -2.82984048e-01 -3.58688921e-01 4.92808789e-01 6.86982632e-01 1.01895183e-01 -9.35113966e-01 8.49236310e-01 -1.68520018e-01 -3.49700868e-01 -4.69257623e-01 -9.30160642e-01 -2.41499171e-01 -3.33476663e-01 -7.77792512e-03 5.15995622e-01 -9.39448416e-01 -1.24386773e-01 2.32741777e-02 -1.42739868e+00 -4.89899904e-01 -4.46871251e-01 3.22330266e-01 -5.21515012e-01 4.03170019e-01 -7.45829165e-01 -7.62225986e-01 -4.22234893e-01 -1.00496709e+00 1.48827732e+00 -1.93675935e-01 -5.65954328e-01 -8.19482088e-01 1.49470523e-01 9.13986042e-02 4.13990140e-01 -3.09816092e-01 1.32393014e+00 -8.68532896e-01 -6.80389285e-01 -1.27939567e-01 3.70556931e-03 2.37818867e-01 -2.55686700e-01 -1.56869039e-01 -8.38856280e-01 -6.12952746e-02 -4.64078188e-01 -5.62757313e-01 1.11177742e+00 -6.04083296e-03 7.10746706e-01 -6.53458297e-01 -5.38920581e-01 -4.42739986e-02 1.11493003e+00 -1.16500765e-01 6.36530161e-01 1.50577664e-01 5.17815351e-01 3.45130533e-01 6.65002108e-01 4.79458183e-01 4.37691331e-01 5.96712291e-01 -1.49269536e-01 1.37916639e-01 -5.28234281e-02 -7.24177659e-01 8.44571531e-01 1.13826466e+00 9.64212418e-03 -7.30406523e-01 -8.15792978e-01 6.62977457e-01 -1.70580471e+00 -1.08706105e+00 7.79543221e-02 2.37585902e+00 1.30870140e+00 3.37000519e-01 -6.24445174e-03 2.09213942e-01 6.60489857e-01 1.21203259e-01 -8.37704539e-02 -6.02239847e-01 7.83218257e-03 6.49583697e-01 5.97535729e-01 9.09544706e-01 -6.93317235e-01 1.24612057e+00 7.82563114e+00 1.03865480e+00 -6.18784308e-01 1.65308431e-01 5.75681865e-01 -1.38687760e-01 -1.01814878e+00 5.68345666e-01 -1.17948663e+00 3.18456590e-01 1.25757837e+00 -3.84537131e-01 5.06923914e-01 5.71136951e-01 1.49233177e-01 -7.90565088e-02 -1.37165844e+00 3.86690408e-01 1.42996475e-01 -1.27335787e+00 3.13610852e-01 1.23408005e-01 6.90441787e-01 -5.44106215e-02 -3.53658170e-01 4.52356517e-01 8.05296659e-01 -1.14954329e+00 1.10734606e+00 4.40673828e-01 6.21989310e-01 -6.73318446e-01 5.91216028e-01 4.76624459e-01 -1.15453625e+00 1.00974053e-01 -4.78595495e-01 -1.36192679e-01 3.79974097e-01 8.29223216e-01 -1.48708212e+00 2.25904360e-01 9.78547782e-02 3.82645041e-01 -4.38728333e-01 8.60171199e-01 -4.30501699e-01 8.72589111e-01 -3.13579619e-01 -5.47889292e-01 2.94021249e-01 2.11663857e-01 5.41222394e-01 1.51741683e+00 6.49772227e-01 -2.01519936e-01 2.25748003e-01 6.08696401e-01 -1.67027742e-01 2.20068634e-01 -3.83155733e-01 -2.28600696e-01 5.95125198e-01 9.81715083e-01 -6.40175223e-01 -8.56572568e-01 -2.85100602e-02 1.05507898e+00 4.34185147e-01 4.71497253e-02 -7.08284199e-01 -5.06604373e-01 4.41318393e-01 2.97753245e-01 4.16236550e-01 -2.92270482e-01 -5.88590614e-02 -1.31405592e+00 5.63854538e-02 -1.15064466e+00 2.34560639e-01 -6.54855669e-01 -9.76800978e-01 7.24038482e-01 1.64966196e-01 -9.11918700e-01 -8.59357893e-01 -3.49019021e-01 -4.93584633e-01 1.01028061e+00 -1.32365334e+00 -6.90470457e-01 3.69279772e-01 5.23218326e-02 7.60718405e-01 2.77826376e-02 9.78458285e-01 -8.49339142e-02 2.02531554e-02 7.03823864e-01 1.63706187e-02 -1.93334147e-01 6.82448864e-01 -1.69101095e+00 8.12541068e-01 1.04133153e+00 4.58232880e-01 6.97260618e-01 7.89275646e-01 -7.83579707e-01 -1.13638973e+00 -1.01074672e+00 1.95955098e+00 -4.75230098e-01 4.94156450e-01 -5.97566783e-01 -4.68844175e-01 6.61325634e-01 2.86138922e-01 -7.77053177e-01 5.98009706e-01 3.07491273e-01 -3.13808531e-01 2.36106098e-01 -6.37023151e-01 8.27389956e-01 1.06237888e+00 -4.33960348e-01 -7.06562936e-01 6.84784532e-01 8.73772502e-01 -2.98877329e-01 -7.15510011e-01 6.18059933e-02 6.51848495e-01 -6.79259419e-01 6.60254538e-01 -4.76564020e-01 4.63479638e-01 -4.09449071e-01 -2.10699901e-01 -1.48937941e+00 -2.74498284e-01 -9.38807964e-01 -4.66025770e-02 1.05418801e+00 1.10430562e+00 -4.31102306e-01 6.88987195e-01 3.39120597e-01 -2.72926629e-01 -7.85967052e-01 -7.02112436e-01 -6.02967083e-01 1.30096257e-01 -6.10276818e-01 7.11516261e-01 3.33187759e-01 3.25833738e-01 7.52600133e-01 -3.57160807e-01 -1.53495982e-01 4.60524112e-01 2.76786904e-03 4.66187775e-01 -1.00126696e+00 -6.04133725e-01 -5.52762568e-01 -2.09077045e-01 -1.48103857e+00 1.13593087e-01 -1.35964680e+00 6.64640248e-01 -1.82164264e+00 2.27852091e-01 -5.96590400e-01 -8.36721212e-02 6.33969069e-01 -3.11114728e-01 2.83796936e-01 4.45230454e-02 1.57546178e-01 -6.73815131e-01 3.59116614e-01 9.83371079e-01 -7.40571618e-02 -1.35414720e-01 -1.61307767e-01 -1.01555800e+00 3.65148097e-01 6.97068572e-01 -7.35736966e-01 -3.51971596e-01 -5.84551990e-01 5.76720715e-01 1.59721836e-01 -6.44568820e-03 -8.90412927e-01 1.95295230e-01 -5.99294044e-02 4.33946222e-01 -3.81229818e-01 2.49681473e-01 9.16304737e-02 -9.60640982e-03 2.68497378e-01 -9.47030306e-01 3.09338272e-01 -2.48421922e-01 4.15398985e-01 -8.38780105e-02 -5.85315585e-01 3.89361650e-01 -3.96360427e-01 -2.52190769e-01 -8.13514888e-02 -8.64371657e-01 3.19053978e-01 3.86511058e-01 -7.84502700e-02 -1.19109072e-01 -4.90843415e-01 -5.31779706e-01 5.60512859e-03 4.27053005e-01 1.47623911e-01 6.41033113e-01 -1.17642498e+00 -9.65052485e-01 4.26520146e-02 -1.73772313e-02 -1.20636515e-01 -7.23956406e-01 3.56446892e-01 -3.95635277e-01 6.51761234e-01 3.91566336e-01 -2.52699167e-01 -1.16836369e+00 -9.20866132e-02 1.52518138e-01 -9.11482334e-01 -4.57899511e-01 1.09646440e+00 -1.00065954e-01 -2.40399972e-01 1.64675221e-01 -5.62124193e-01 9.52892154e-02 1.17031008e-01 5.33256590e-01 6.38586879e-02 2.40405723e-01 -4.35712039e-01 -2.30274469e-01 7.50905201e-02 -3.36721063e-01 -9.42544758e-01 8.61489534e-01 1.03361040e-01 -1.66358098e-01 3.84829640e-01 9.20578241e-01 2.53135085e-01 -7.73926497e-01 -7.62537122e-02 2.80808270e-01 -2.59847403e-01 -1.48672760e-01 -8.13279569e-01 -2.53620565e-01 5.62468112e-01 -1.53640464e-01 1.54204562e-01 8.22269976e-01 2.29763631e-02 9.37227726e-01 6.67230666e-01 5.33941448e-01 -1.18993378e+00 2.26207927e-01 8.15285087e-01 5.96024871e-01 -6.11908376e-01 -4.39552926e-02 -7.75146484e-02 -7.54070163e-01 1.06111324e+00 2.56228507e-01 4.77323532e-02 1.95988387e-01 3.30691427e-01 -2.13863492e-01 1.91013560e-01 -1.53001881e+00 -2.78761178e-01 2.21116334e-01 2.31147870e-01 1.00316536e+00 2.07900405e-01 -4.31099325e-01 2.86180824e-01 -8.21303129e-01 -2.55191833e-01 4.95328128e-01 9.64428604e-01 -1.00647724e+00 -1.60030055e+00 -2.47763202e-01 9.57499146e-01 -5.34320235e-01 -7.17557847e-01 -3.80564988e-01 3.05601060e-01 1.10422950e-02 1.11275840e+00 6.71369359e-02 -4.71012563e-01 -8.21279436e-02 5.84636509e-01 6.60398543e-01 -1.06370699e+00 -7.75576651e-01 1.68767706e-01 5.32952368e-01 -3.96783412e-01 8.92480686e-02 -8.08167577e-01 -1.17204952e+00 6.43220730e-03 -5.00427306e-01 7.06056952e-01 5.74886560e-01 1.13175595e+00 2.04416052e-01 2.39701867e-01 4.88569081e-01 -7.25191057e-01 -8.44525874e-01 -1.00351322e+00 -2.31466651e-01 -1.77230835e-02 1.72500789e-01 -1.82405636e-01 -3.40424567e-01 4.89807464e-02]

[11.720924377441406, 9.086136817932129]

87e08da3-ac7d-44cc-8689-aad55b66de49

sa-text-simple-but-accurate-detector-for-text

1911.07046

https://arxiv.org/abs/1911.07046v3

https://arxiv.org/pdf/1911.07046v3.pdf

A method for detecting text of arbitrary shapes in natural scenes that improves text spotting

Understanding the meaning of text in images of natural scenes like highway signs or store front emblems is particularly challenging if the text is foreshortened in the image or the letters are artistically distorted. We introduce a pipeline-based text spotting framework that can both detect and recognize text in various fonts, shapes, and orientations in natural scene images with complicated backgrounds. The main contribution of our work is the text detection component, which we call UHT, short for UNet, Heatmap, and Textfill. UHT uses a UNet to compute heatmaps for candidate text regions and a textfill algorithm to produce tight polygonal boundaries around each word in the candidate text. Our method trains the UNet with groundtruth heatmaps that we obtain from text bounding polygons provided by groundtruth annotations. Our text spotting framework, called UHTA, combines UHT with the state-of-the-art text recognition system ASTER. Experiments on four challenging and public scene-text-detection datasets (Total-Text, SCUT-CTW1500, MSRA-TD500, and COCO-Text) show the effectiveness and generalization ability of UHT in detecting not only multilingual (potentially rotated) straight but also curved text in scripts of multiple languages. Our experimental results of UHTA on the Total-Text dataset show that UHTA outperforms four state-of-the-art text spotting frameworks by at least 9.1 percent points in the F-measure, which suggests that UHTA may be used as a complete text detection and recognition system in real applications.

['Margrit Betke', 'Yi Zheng', 'Qitong Wang']

2019-11-16

['text-spotting']

['computer-vision']

[ 4.66784239e-01 -3.57322454e-01 2.48699903e-01 -1.96693882e-01 -7.55427599e-01 -7.37479389e-01 8.87549400e-01 -2.34458193e-01 -1.02722801e-01 3.11183487e-03 2.19401047e-02 -5.65103590e-01 3.74314725e-01 -6.62425816e-01 -7.24828362e-01 -3.91630441e-01 6.61820233e-01 7.57895052e-01 6.34951711e-01 -2.54672378e-01 6.39188409e-01 4.42828357e-01 -1.42989087e+00 6.79222405e-01 1.00711703e+00 7.05218434e-01 4.20839757e-01 8.78926337e-01 -4.59320247e-01 3.43469709e-01 -5.32692850e-01 -5.65776110e-01 3.10022384e-01 -3.34343500e-02 -4.94607031e-01 5.33323407e-01 9.49652016e-01 -4.54508156e-01 -1.94470108e-01 7.32536256e-01 3.22466016e-01 -2.25788504e-01 8.35187793e-01 -8.87012661e-01 -4.36928898e-01 4.87732321e-01 -1.19007289e+00 -2.61708856e-01 4.90632474e-01 1.48525715e-01 7.74463296e-01 -1.54756474e+00 7.42257953e-01 1.41269422e+00 9.48901474e-01 -3.92145179e-02 -9.26556885e-01 -6.01623356e-01 8.86117145e-02 -1.94613293e-01 -1.65621245e+00 -2.52468884e-01 4.64683831e-01 -7.55320489e-01 1.01137900e+00 5.76191604e-01 3.40016752e-01 8.93054783e-01 3.13701123e-01 1.25389349e+00 9.08781946e-01 -7.92427778e-01 -1.13951251e-01 9.83740836e-02 -1.93609353e-02 1.00878882e+00 2.05457807e-01 -5.67923009e-01 -6.01135731e-01 -2.75384970e-02 6.37463212e-01 -2.98651129e-01 -9.35303122e-02 -1.92410067e-01 -1.54603064e+00 6.05523884e-01 -1.40552968e-01 6.33865818e-02 1.42233729e-01 -1.30465224e-01 3.78261000e-01 -1.91974994e-02 6.18268549e-01 1.85313806e-01 -2.44441535e-02 -1.24600537e-01 -1.29718018e+00 2.27355957e-01 5.77617645e-01 1.28985131e+00 4.85389262e-01 1.01719564e-02 -2.75818318e-01 1.04100478e+00 1.23813018e-01 1.10774755e+00 2.29325697e-01 4.56143990e-02 1.16527021e+00 8.50955129e-01 1.24912463e-01 -1.04894161e+00 -4.19528186e-01 9.79445949e-02 -5.13414025e-01 8.60553235e-02 4.84521776e-01 -1.72506303e-01 -1.26390827e+00 5.30923247e-01 3.65312845e-01 -4.66768324e-01 -4.34223533e-01 7.67496526e-01 5.79113424e-01 8.70268106e-01 -5.94138145e-01 4.76465493e-01 1.41738117e+00 -1.09676123e+00 -5.71936607e-01 -6.34331107e-01 8.85700703e-01 -1.44988501e+00 1.33162677e+00 5.29360354e-01 -4.58085388e-01 -2.31811568e-01 -1.07146072e+00 -3.02952886e-01 -6.32313073e-01 1.03422284e+00 1.59191817e-01 8.27222645e-01 -7.86045134e-01 -5.03278524e-02 -6.39275908e-01 -9.26769733e-01 2.84602344e-01 -2.49554478e-02 -1.17442131e-01 -1.60564080e-01 -5.64563572e-01 6.90031350e-01 2.61875480e-01 1.88790590e-01 -3.76740992e-01 -2.34355167e-01 -7.06247151e-01 -2.63295442e-01 6.19927347e-01 -1.95464343e-01 9.08155024e-01 -7.18682766e-01 -1.26741719e+00 1.04671288e+00 -1.37155265e-01 -5.77857308e-02 1.05566466e+00 -3.83896291e-01 -5.35806894e-01 1.94408983e-01 3.36338550e-01 5.83570600e-01 1.39576185e+00 -1.07243800e+00 -7.54750252e-01 -2.58414865e-01 -8.47378016e-01 4.03644115e-01 -1.43116742e-01 3.45817477e-01 -9.21639502e-01 -9.75095868e-01 1.58176601e-01 -9.08883691e-01 3.84712994e-01 3.69846493e-01 -1.09499216e+00 5.33909630e-03 1.62593842e+00 -1.05541265e+00 1.24051535e+00 -1.95288444e+00 -3.91877502e-01 3.52428555e-01 -3.85519341e-02 2.98994154e-01 -5.61353229e-02 5.85774601e-01 2.93128967e-01 1.86657175e-01 -1.85110450e-01 -1.74144432e-01 2.15844184e-01 -1.19900241e-01 -6.38012767e-01 5.31681776e-01 1.53581530e-01 7.86328137e-01 -3.12638551e-01 -7.00426877e-01 6.51246071e-01 3.15081537e-01 1.81564763e-02 -2.96609938e-01 -3.67740780e-01 -2.23821789e-01 -4.58989620e-01 1.07915711e+00 9.25940335e-01 -2.10314468e-02 1.24177113e-02 -2.71826820e-03 -5.07736802e-01 -1.06649347e-01 -1.31363678e+00 1.10365844e+00 1.17425159e-01 1.39805472e+00 -1.49124376e-02 -2.22037658e-01 1.06773651e+00 -1.80056900e-01 -8.00449774e-02 -5.99142849e-01 1.36972725e-01 2.19297528e-01 -6.50496960e-01 -7.11480975e-01 1.21921527e+00 6.24347985e-01 1.09235663e-02 5.14735401e-01 -5.64236701e-01 -5.04090190e-01 2.94434667e-01 2.69098788e-01 8.30568314e-01 2.77273417e-01 -9.64917019e-02 -3.47952634e-01 3.04267943e-01 2.68888980e-01 -9.66416597e-02 1.04002512e+00 1.87794730e-01 9.50091660e-01 6.11471415e-01 -5.14197528e-01 -1.49505067e+00 -6.68545723e-01 -4.04271007e-01 1.32050776e+00 7.05936402e-02 -4.35417444e-01 -1.00535452e+00 -5.45048892e-01 7.07742795e-02 7.24583268e-01 -6.77700937e-01 5.63499033e-01 -6.22775853e-01 -6.02080762e-01 7.95095026e-01 4.85719651e-01 7.94047773e-01 -6.34840190e-01 -4.70354557e-01 -2.43762016e-01 -2.29637519e-01 -1.45595670e+00 -9.39259887e-01 -1.18121900e-01 -4.12812233e-01 -9.51504290e-01 -9.80677664e-01 -9.19062436e-01 7.87744880e-01 5.43679178e-01 6.54910505e-01 -1.35337859e-01 -5.36245883e-01 3.58606696e-01 -4.60464716e-01 -3.56110275e-01 -4.87911016e-01 -4.47159708e-02 -3.69439691e-01 3.76179188e-01 1.92426413e-01 4.07067776e-01 -2.93936312e-01 9.20538723e-01 -8.63644361e-01 6.40873611e-01 5.28275609e-01 7.37789452e-01 4.82118994e-01 1.70499220e-01 -1.80051520e-01 -7.40019977e-01 4.84975010e-01 1.26569510e-01 -8.94127131e-01 6.20834470e-01 -4.09538388e-01 -1.65833607e-01 2.61106402e-01 -2.90578425e-01 -1.10567951e+00 2.38494173e-01 1.97400212e-01 -2.14456841e-01 -5.89764155e-02 2.29577824e-01 -2.48760134e-02 -1.66053772e-01 7.38279939e-01 5.69031119e-01 -3.58747661e-01 -2.88043171e-01 3.05711299e-01 1.25782919e+00 5.54840207e-01 -4.89597201e-01 9.52256620e-01 6.78748846e-01 -2.32569322e-01 -1.51262343e+00 -4.31629270e-01 -6.96664393e-01 -8.09586823e-01 -3.72146010e-01 7.81114519e-01 -6.73405647e-01 -1.27003238e-01 1.02591681e+00 -1.14670432e+00 -5.59623539e-01 2.96207935e-01 3.65480445e-02 -1.99231848e-01 7.43234098e-01 -3.92602175e-01 -8.60772610e-01 -5.74392378e-01 -1.12884057e+00 1.96414399e+00 -9.45870727e-02 1.18418887e-01 -7.35765219e-01 -3.44751775e-01 6.15437686e-01 4.92089950e-02 1.73720136e-01 7.28082895e-01 -4.33778346e-01 -5.11912823e-01 -4.54859376e-01 -6.33644998e-01 -6.48328960e-02 -2.10080847e-01 5.96104085e-01 -1.01247656e+00 -1.31278083e-01 -6.86838090e-01 -7.19335303e-02 9.56482947e-01 2.25789279e-01 8.67330253e-01 -1.71310514e-01 -5.64860940e-01 4.88504440e-01 1.16956031e+00 -2.87558115e-03 7.64730155e-01 5.21313310e-01 9.70773816e-01 4.83923614e-01 8.65264773e-01 5.35675168e-01 1.04555935e-01 8.27569485e-01 1.10326752e-01 -3.78834933e-01 -1.46887690e-01 -2.47322857e-01 4.76217419e-01 4.31990206e-01 2.48788640e-01 -6.21707618e-01 -1.45461786e+00 4.87551451e-01 -1.77849853e+00 -6.50508523e-01 -8.63935351e-01 2.06349516e+00 4.15806174e-01 1.66195542e-01 3.13577726e-02 1.15893014e-01 1.22579610e+00 -8.34283791e-03 -3.79661471e-01 -4.99101937e-01 -4.51955795e-01 -2.39762098e-01 8.37039709e-01 2.89056927e-01 -1.40323412e+00 1.36933017e+00 6.01667643e+00 1.15302813e+00 -1.19537461e+00 -3.46635640e-01 5.94132483e-01 2.70895958e-01 2.87768841e-01 -2.03826264e-01 -1.09032166e+00 3.13519746e-01 2.43399590e-01 2.32490405e-01 2.65293777e-01 8.21319699e-01 3.40250283e-01 -4.85728234e-01 -7.11168110e-01 1.06698632e+00 5.35977602e-01 -1.18725872e+00 1.67968333e-01 -1.63778245e-01 8.79418194e-01 2.33775958e-01 6.97771981e-02 -4.62234858e-03 2.52602726e-01 -8.30242276e-01 1.17989647e+00 2.22650126e-01 1.31167483e+00 -3.86496902e-01 4.80023563e-01 1.45634010e-01 -1.35190618e+00 2.11860120e-01 -3.26121211e-01 4.37369198e-01 -3.03391200e-02 6.28380716e-01 -1.57758749e+00 3.76302868e-01 5.73568344e-01 6.67313695e-01 -1.25118995e+00 1.09921610e+00 -2.64483511e-01 6.83677018e-01 -6.28455460e-01 -2.79724002e-01 3.60034287e-01 -1.42393008e-01 6.65305912e-01 1.84249139e+00 4.46551889e-01 -4.83815908e-01 2.97070563e-01 8.62482131e-01 7.05967098e-03 4.64410484e-01 -5.23872435e-01 -1.00622840e-01 3.87431562e-01 1.26141739e+00 -1.47332835e+00 -4.03756469e-01 -2.76898056e-01 1.18160081e+00 -3.22796255e-01 3.93900007e-01 -8.42674077e-01 -8.49549949e-01 -1.02000453e-01 3.52970332e-01 6.39448822e-01 -1.87233135e-01 -7.20195293e-01 -1.16409993e+00 4.04478580e-01 -8.13788593e-01 1.29900411e-01 -1.48613977e+00 -8.33384752e-01 4.02154624e-01 -2.55746543e-01 -1.33107901e+00 2.37826347e-01 -1.03135026e+00 -6.03565931e-01 7.34271646e-01 -9.92388248e-01 -1.87455094e+00 -5.37278831e-01 5.57175875e-01 1.07488418e+00 -1.29684210e-01 2.86880076e-01 -8.73317048e-02 -8.82125914e-01 6.34294033e-01 6.76096678e-01 6.30923867e-01 8.34722459e-01 -1.09451497e+00 1.01380110e+00 1.14979553e+00 1.06503308e-01 1.28904283e-01 6.36585712e-01 -1.08126950e+00 -1.60489714e+00 -1.32265067e+00 6.59183085e-01 -6.30503058e-01 8.80128145e-01 -1.07131052e+00 -7.30615556e-01 8.54374409e-01 4.54288349e-02 -4.77900565e-01 -1.42455742e-01 -1.74882084e-01 -5.17569602e-01 3.07777256e-01 -8.01488698e-01 8.14967155e-01 5.26398480e-01 -3.82369578e-01 -4.68925357e-01 7.67871916e-01 2.44448751e-01 -7.45883703e-01 -3.02354455e-01 -9.19673368e-02 8.75945926e-01 -6.93788111e-01 6.26093447e-01 2.22862244e-01 3.43011796e-01 -3.92287433e-01 8.17417167e-03 -8.22065890e-01 1.61788851e-01 -7.16576040e-01 5.11290967e-01 1.21264195e+00 4.95368004e-01 -6.03652656e-01 6.68190539e-01 3.74545068e-01 -3.39299798e-01 -2.39319369e-01 -9.15825665e-01 -7.67690182e-01 2.08146665e-02 -6.72019720e-01 4.25013244e-01 1.06278861e+00 5.92018664e-03 2.88971454e-01 -4.31440324e-01 4.45384234e-02 3.59952748e-01 8.24604779e-02 1.17196345e+00 -1.02473652e+00 2.17761129e-01 -5.83671570e-01 -3.96439821e-01 -1.19422865e+00 -3.44548672e-01 -7.17946887e-01 2.88153738e-01 -1.62658989e+00 1.61239758e-01 -1.60888314e-01 8.83720875e-01 7.35984623e-01 -1.95129048e-02 9.22704190e-02 9.80916321e-02 2.77117461e-01 -5.87275624e-01 1.93763375e-01 1.14653277e+00 -4.44797724e-01 -1.57520518e-01 -3.49367976e-01 -1.05763197e-01 8.36438715e-01 6.90487266e-01 -2.10130125e-01 6.50808588e-02 -6.88223183e-01 4.39641774e-01 -3.87563348e-01 3.45345169e-01 -8.94282222e-01 2.26126656e-01 -1.94046840e-01 5.38677931e-01 -1.49441600e+00 1.82494242e-02 -6.46460235e-01 -2.89448589e-01 -3.21678305e-03 -7.17022195e-02 -7.98139498e-02 3.73476714e-01 4.37418342e-01 4.29392755e-01 -1.75379440e-01 6.15156233e-01 3.81736159e-01 -6.85663879e-01 -3.48824233e-01 -7.87675738e-01 -6.36317059e-02 8.13376188e-01 -6.05677485e-01 -8.22575986e-01 -1.00090824e-01 2.64006690e-03 3.00270915e-01 6.88575506e-01 6.75929308e-01 6.94693625e-01 -8.42814445e-01 -8.77383769e-01 4.09356028e-01 4.65723217e-01 -1.72035862e-02 -2.09902253e-04 8.79603863e-01 -1.09698713e+00 6.27144754e-01 2.07833186e-01 -1.07260478e+00 -1.51049829e+00 2.98750341e-01 2.52614796e-01 -1.52094990e-01 -1.11924732e+00 4.42081720e-01 3.16282630e-01 -4.30485219e-01 1.17677450e-01 -6.78463578e-01 2.73993820e-01 -2.02655688e-01 5.49458683e-01 5.23796797e-01 3.72127891e-01 -7.82615006e-01 -1.42728701e-01 9.94784951e-01 -1.98137015e-01 -1.86675400e-01 8.02201331e-01 -1.92909881e-01 3.67050394e-02 1.95576623e-01 6.59788847e-01 3.10334802e-01 -1.05303538e+00 -1.41332462e-01 4.63318639e-02 -5.54506958e-01 4.83979732e-02 -1.12124169e+00 -6.52814567e-01 7.79045105e-01 5.32778263e-01 4.46979851e-02 7.85809398e-01 -1.76245660e-01 6.18230283e-01 7.50810623e-01 1.03053823e-01 -1.64979959e+00 7.99434036e-02 7.03808248e-01 1.13738120e+00 -1.15276551e+00 3.18517089e-01 -6.77272677e-01 -7.67341971e-01 1.54198492e+00 4.23986614e-01 3.78311545e-01 1.09633412e-02 5.46489716e-01 -3.21820714e-02 -2.79358357e-01 -3.53026807e-01 -1.19181141e-01 4.43229496e-01 4.13858742e-01 8.74122232e-02 1.47533894e-01 7.29669780e-02 1.13078291e-02 -3.05530101e-01 -3.92697662e-01 7.10877120e-01 1.10720706e+00 -8.12224209e-01 -6.30008161e-01 -1.19322348e+00 7.55878925e-01 -4.95608747e-02 -3.52157086e-01 -1.02561510e+00 9.90563393e-01 -2.10097656e-01 8.30627263e-01 7.87310749e-02 -4.49345142e-01 2.72827923e-01 2.40711465e-01 1.39464170e-01 -3.66808414e-01 -4.22035098e-01 6.37863338e-01 2.93234408e-01 -1.02322824e-01 1.81877166e-01 -7.59172440e-01 -1.20415699e+00 -3.48786980e-01 -7.01248288e-01 -5.39851725e-01 9.50242996e-01 8.35956097e-01 3.05923551e-01 1.96988031e-01 3.64201754e-01 -7.50897765e-01 -3.92135270e-02 -1.15166891e+00 -6.59500420e-01 1.31435081e-01 -1.81845278e-02 -5.21750927e-01 -1.50892720e-01 5.24666369e-01]

[12.009273529052734, 2.1955580711364746]

67cc16ca-7951-45c2-a8d8-1c6cabf5bb2c

enhancing-task-bot-engagement-with

2212.10008

https://arxiv.org/abs/2212.10008v1

https://arxiv.org/pdf/2212.10008v1.pdf

Enhancing Task Bot Engagement with Synthesized Open-Domain Dialog

Many efforts have been made to construct dialog systems for different types of conversations, such as task-oriented dialog (TOD) and open-domain dialog (ODD). To better mimic human-level conversations that usually fuse various dialog modes, it is essential to build a system that can effectively handle both TOD and ODD and access different knowledge sources. To address the lack of available data for the fused task, we propose a framework for automatically generating dialogues that combine knowledge-grounded ODDs and TODs in various settings. Additionally, we introduce a unified model PivotBot that is capable of appropriately adopting TOD and ODD modes and accessing different knowledge sources in order to effectively tackle the fused task. Evaluation results demonstrate the superior ability of the proposed model to switch seamlessly between TOD and ODD tasks.

['Zhu Zhang', 'Jianfeng Gao', 'Michel Galley', 'Baolin Peng', 'Miaoran Li']

2022-12-20

['open-domain-dialog']

['natural-language-processing']

[-3.97944182e-01 4.19102788e-01 5.33659607e-02 -4.33840871e-01 -5.70005476e-01 -8.97768617e-01 9.31697190e-01 -2.17474490e-01 -1.12315372e-01 1.06296086e+00 5.54945529e-01 -4.27968323e-01 -1.46388337e-01 -6.88400269e-01 2.73538768e-01 -1.02453426e-01 6.76108599e-01 9.05803442e-01 6.11498237e-01 -7.54976869e-01 -1.57984775e-02 1.59607932e-01 -1.21764100e+00 3.53916585e-01 1.17375815e+00 6.35672629e-01 6.06351674e-01 7.38018930e-01 -8.71758163e-01 8.75660479e-01 -7.61231005e-01 -7.50534534e-01 6.24216050e-02 -4.18790787e-01 -1.40969539e+00 6.53316155e-02 -1.87734097e-01 -5.73051929e-01 -1.79141551e-01 6.97202265e-01 5.46714127e-01 4.81781930e-01 7.31252253e-01 -1.52288067e+00 -5.57162046e-01 9.16434884e-01 3.07444185e-01 -5.19467108e-02 7.53369808e-01 3.80495101e-01 8.12377334e-01 -7.09277630e-01 4.35434759e-01 1.75418341e+00 3.14985782e-01 8.65439892e-01 -8.83020937e-01 -4.66808259e-01 1.36508271e-01 -2.13422075e-01 -8.98676038e-01 -4.76448447e-01 6.53133571e-01 -3.51213515e-01 7.98833966e-01 2.35940456e-01 1.79948196e-01 1.37854028e+00 -2.51400828e-01 7.91346490e-01 1.14406025e+00 -4.21541810e-01 -2.53977347e-02 7.61360645e-01 4.69519705e-01 3.02889287e-01 -2.24090144e-01 -5.22091687e-01 -5.61447561e-01 -3.29184175e-01 7.77595401e-01 -7.23327696e-02 -1.63815588e-01 8.34756717e-03 -1.43880844e+00 7.41311669e-01 -1.50181903e-02 5.07283390e-01 -3.02999914e-01 -6.79327488e-01 4.48875904e-01 4.58415270e-01 1.29110247e-01 7.11777210e-01 -5.06180465e-01 -5.08373320e-01 -1.25671551e-01 6.96535230e-01 1.56244183e+00 1.49742877e+00 5.75933456e-01 -3.26576173e-01 -7.00756967e-01 1.11840999e+00 2.89402217e-01 3.33269954e-01 4.62628901e-01 -1.05084491e+00 6.96810663e-01 1.01830018e+00 7.41766155e-01 -5.57737708e-01 -4.30923074e-01 4.35765892e-01 -6.23949051e-01 -4.05277789e-01 8.69419813e-01 -3.75258625e-01 -2.98617244e-01 1.75145328e+00 6.06837749e-01 -3.43302786e-01 6.30934775e-01 6.81289554e-01 1.31091928e+00 4.14379060e-01 1.14831850e-01 -7.66967610e-02 1.73173404e+00 -9.55442131e-01 -1.30898786e+00 -6.31824136e-02 4.80254650e-01 -7.76004851e-01 1.50225496e+00 -2.97276638e-02 -1.03149641e+00 -5.98555446e-01 -7.56027222e-01 -2.17599839e-01 -5.94583392e-01 -7.19665810e-02 4.56056923e-01 5.82225859e-01 -8.47027540e-01 -3.18046398e-02 -3.07856262e-01 -5.24656236e-01 -3.32825035e-01 7.24978819e-02 -2.36776322e-02 1.81026146e-01 -1.73173904e+00 1.24676192e+00 5.83184659e-01 -5.87260388e-02 -6.45751238e-01 -2.98702836e-01 -8.06109250e-01 -1.58133614e-03 7.15831399e-01 -7.47031450e-01 1.77168870e+00 -2.47130796e-01 -1.97079003e+00 4.66607302e-01 -5.33123128e-03 -1.37205973e-01 6.12660646e-01 -2.67819405e-01 -4.07047808e-01 2.09182967e-02 9.88861620e-02 5.29897928e-01 4.33645606e-01 -1.12089777e+00 -6.07042909e-01 -1.68317258e-01 7.07574189e-01 5.86828232e-01 -6.36615396e-01 2.03445241e-01 -3.14151436e-01 -2.92856038e-01 -3.32821637e-01 -8.82380903e-01 -1.26965940e-01 -3.16219181e-01 -6.74489677e-01 -6.04672849e-01 1.03477204e+00 -3.89109701e-01 1.28267455e+00 -1.83067191e+00 2.52982646e-01 -2.96638042e-01 3.64536345e-01 5.61666727e-01 1.60562292e-01 9.99009252e-01 7.67923474e-01 -8.68481249e-02 -1.62260085e-02 -5.01963437e-01 3.40587556e-01 5.28182089e-01 -3.29846263e-01 -6.48875594e-01 9.39092040e-02 7.51203835e-01 -8.78462195e-01 -7.28893697e-01 3.44323069e-01 1.88388061e-02 -3.91992688e-01 9.37581718e-01 -7.86119759e-01 1.03545845e+00 -9.08366621e-01 3.28469843e-01 4.87643808e-01 -2.68193692e-01 3.49488258e-01 7.72460504e-03 -1.62663937e-01 5.58393002e-01 -1.14969695e+00 1.63014162e+00 -8.00444365e-01 8.44951198e-02 4.37808514e-01 -1.78320631e-01 1.06920648e+00 7.90243983e-01 -5.42414822e-02 -1.01297617e-01 3.41472179e-01 3.93629968e-02 -1.82532743e-02 -7.52185345e-01 9.51122165e-01 -1.06387801e-01 -4.80655819e-01 7.15125442e-01 3.49569112e-01 -3.74011308e-01 9.72908735e-02 3.63993019e-01 8.20683360e-01 -1.72249004e-01 3.43566567e-01 7.20076868e-03 7.80518532e-01 3.58388871e-02 2.86328703e-01 7.32463121e-01 -4.35524821e-01 1.75288673e-02 4.73515570e-01 -1.91032842e-01 -6.08084977e-01 -1.00298440e+00 1.06894687e-01 1.44122624e+00 3.42416912e-01 -2.79101223e-01 -9.12742138e-01 -7.55745828e-01 -1.63480699e-01 8.42418790e-01 6.26455992e-02 9.53297466e-02 -3.42863262e-01 -2.69086272e-01 9.80966687e-01 3.01820576e-01 1.11686838e+00 -1.19432199e+00 -4.13014472e-01 3.01557362e-01 -9.12086368e-01 -1.61926067e+00 -5.01222610e-01 2.21580639e-02 -3.60824138e-01 -1.00384152e+00 -4.70232904e-01 -8.79689395e-01 1.03935242e-01 3.99068236e-01 1.00291169e+00 -1.03463560e-01 4.16996688e-01 4.63157415e-01 -7.14228690e-01 -2.80416608e-01 -1.08512008e+00 3.40570331e-01 4.39406931e-02 1.01919137e-02 3.86293054e-01 -4.75614905e-01 -2.15307206e-01 9.10318494e-01 -9.00836825e-01 3.09953630e-01 2.04877585e-01 8.74905825e-01 -4.39573497e-01 -1.54989526e-01 1.07117689e+00 -7.38953531e-01 1.52459335e+00 -5.33244848e-01 -2.33696103e-01 4.47178602e-01 -1.68162689e-01 8.14099982e-02 6.81369126e-01 -4.38881844e-01 -1.81680858e+00 -3.39662611e-01 -2.64760911e-01 -2.04415187e-01 -5.78523219e-01 4.02749419e-01 -4.27703083e-01 3.08814824e-01 4.97212708e-01 4.55850028e-02 2.17186168e-01 -6.70516074e-01 6.57429874e-01 1.55993438e+00 4.30216908e-01 -1.10987854e+00 4.60464954e-01 3.22185201e-03 -6.72055185e-01 -8.72488856e-01 -7.65772223e-01 -5.12256086e-01 -6.93041623e-01 -3.21765244e-01 1.20963645e+00 -6.75603330e-01 -1.00197840e+00 6.09566629e-01 -1.44562459e+00 -4.90817010e-01 -1.87658050e-04 1.74861297e-01 -4.62931275e-01 2.29292080e-01 -6.72511101e-01 -1.10080874e+00 -3.31152380e-01 -1.25493932e+00 9.64505255e-01 4.52967018e-01 -4.88562375e-01 -1.23711467e+00 8.38617887e-03 6.96589649e-01 5.95377445e-01 -2.70838708e-01 1.06703877e+00 -1.30605268e+00 -4.29631531e-01 8.45167413e-02 -1.20017871e-01 1.69416651e-01 6.43485725e-01 -2.87622809e-01 -1.06824684e+00 9.26238894e-02 7.67621472e-02 -9.01583910e-01 -6.48527518e-02 -3.13323915e-01 4.32336688e-01 -4.56055164e-01 -2.74289131e-01 -2.75585622e-01 5.22227347e-01 5.34898281e-01 3.25134009e-01 2.90433839e-02 3.70296627e-01 1.01437283e+00 8.44731569e-01 4.77081358e-01 1.04412973e+00 1.10487807e+00 -3.90454605e-02 2.25173637e-01 4.11781520e-02 -4.82467026e-01 1.95734888e-01 9.76643622e-01 2.92285383e-01 -3.54647666e-01 -7.98602462e-01 5.99368393e-01 -2.00887513e+00 -8.10806751e-01 1.16819613e-01 1.68985522e+00 1.20663154e+00 7.68587813e-02 4.05807555e-01 -1.79461464e-01 7.31671393e-01 8.68653879e-02 -2.95900166e-01 -5.25846183e-01 2.57358640e-01 -2.41108224e-01 -3.90007287e-01 6.44247234e-01 -8.26786757e-01 1.16085923e+00 6.21858168e+00 5.75869501e-01 -6.54489279e-01 1.96307510e-01 1.90383822e-01 5.42488515e-01 -2.58909553e-01 -4.20065485e-02 -9.62695718e-01 3.56483907e-01 7.88725317e-01 -3.88251930e-01 4.18724060e-01 7.56042361e-01 7.73853511e-02 1.54085429e-02 -1.23911667e+00 5.99828720e-01 -4.31592822e-01 -1.00290656e+00 1.17898703e-01 -1.89442024e-01 4.03339386e-01 -6.97031796e-01 -2.92052656e-01 7.37596154e-01 1.04035079e+00 -6.63582027e-01 3.03629428e-01 3.78975451e-01 4.83172894e-01 -1.55340299e-01 6.90222323e-01 9.75122750e-01 -1.10906494e+00 -8.07266682e-02 1.50292115e-02 -2.35316798e-01 3.03514868e-01 -1.07370414e-01 -1.53378487e+00 7.80171275e-01 4.56813574e-01 1.21445835e-01 -9.10794139e-02 4.06378359e-01 -8.23656470e-02 2.71428991e-02 -1.65741339e-01 -3.91290158e-01 1.75651014e-01 -1.80301189e-01 6.00288153e-01 9.48904276e-01 1.09681338e-01 1.83526725e-01 7.30185807e-01 1.02797651e+00 7.56503120e-02 6.82553602e-03 -9.62788522e-01 -3.37795019e-02 1.19557738e+00 1.25319493e+00 -3.98488820e-01 -4.24065709e-01 -4.71605331e-01 8.84361446e-01 3.44137251e-01 2.15523422e-01 -6.54791474e-01 -3.22658688e-01 7.69280970e-01 -2.42551491e-01 -4.46354635e-02 -2.62799233e-01 -5.73212691e-02 -1.06349969e+00 -1.50642559e-01 -1.02788067e+00 4.43262041e-01 -7.38627136e-01 -1.58242166e+00 9.02845562e-01 4.90923882e-01 -1.07402527e+00 -6.52101338e-01 -3.85278493e-01 -6.83155894e-01 1.01801848e+00 -1.12232673e+00 -1.29078543e+00 -4.61831838e-01 8.80584359e-01 8.97208512e-01 -1.90445602e-01 1.08595622e+00 2.02736512e-01 -5.06426871e-01 4.08403337e-01 -3.86659950e-01 1.05191268e-01 8.74623716e-01 -1.31421053e+00 1.44855529e-01 3.09978038e-01 -4.29612607e-01 8.89144003e-01 7.37526655e-01 -6.58074200e-01 -1.27647686e+00 -9.56833482e-01 7.95869470e-01 -7.47560680e-01 6.16364598e-01 -4.41779524e-01 -9.11408842e-01 7.58566618e-01 5.81141889e-01 -7.57001758e-01 6.71939433e-01 1.54928491e-01 -1.53206855e-01 2.13986114e-01 -1.41754901e+00 8.50987494e-01 9.19504881e-01 -6.77587569e-01 -1.21057999e+00 2.24749461e-01 1.11008263e+00 -7.38224864e-01 -9.64402199e-01 2.04609156e-01 1.94490448e-01 -9.82773960e-01 8.47145557e-01 -5.84082723e-01 1.60681203e-01 -5.97909912e-02 -1.37298822e-01 -1.44864643e+00 2.12766379e-01 -9.64342117e-01 4.81905527e-02 1.69677556e+00 2.38532171e-01 -8.63463402e-01 2.29288727e-01 1.01768923e+00 -3.01844478e-01 -2.82810956e-01 -8.43815506e-01 -5.80621719e-01 3.00107449e-02 -2.42236823e-01 9.93671834e-01 9.52854514e-01 6.37766123e-01 7.26799667e-01 -5.63719273e-01 9.73700061e-02 -4.06166762e-02 1.78665996e-01 1.33796966e+00 -1.29506969e+00 -1.48517385e-01 -2.39324659e-01 2.41230160e-01 -1.47744918e+00 2.37477601e-01 -6.16988540e-01 2.28614420e-01 -1.56230140e+00 -1.89144418e-01 -7.13528335e-01 1.71323031e-01 4.34506536e-01 -4.20315504e-01 -4.40417230e-01 2.06213892e-01 2.51183003e-01 -6.02970243e-01 9.04522300e-01 1.53474021e+00 1.05052032e-01 -7.45268703e-01 1.64503351e-01 -1.01378083e+00 7.88957238e-01 6.18547976e-01 4.55290824e-02 -7.22769558e-01 -1.73516527e-01 -4.46852565e-01 7.07327306e-01 1.82045728e-01 -7.09194303e-01 3.41269404e-01 -4.16889668e-01 -5.07500052e-01 -3.79827738e-01 5.90022027e-01 -8.07953894e-01 -7.16529340e-02 -7.93363973e-02 -7.21890509e-01 4.00548130e-02 1.41794711e-01 4.27507609e-01 -3.16745371e-01 -1.74248323e-01 5.46313047e-01 -2.93979973e-01 -5.80941260e-01 -1.68437645e-01 -6.64884031e-01 2.63649553e-01 1.14262164e+00 4.43536825e-02 -6.87440693e-01 -8.10866535e-01 -6.79866731e-01 7.61264443e-01 1.98740140e-01 7.83924103e-01 3.97330672e-01 -1.12937558e+00 -2.72888035e-01 9.49464832e-03 2.41408050e-01 1.04317563e-02 2.45135412e-01 5.67683578e-01 -1.40462846e-01 5.71560502e-01 -4.37677205e-01 -3.39625537e-01 -1.05441344e+00 3.80125016e-01 3.43660146e-01 -6.16067231e-01 -3.75312358e-01 5.04387379e-01 1.96439788e-01 -1.22986746e+00 4.58867222e-01 -2.69866884e-01 -5.12289107e-01 3.99914607e-02 5.67630529e-01 2.86359102e-01 -1.83125481e-01 -4.28656578e-01 -2.92903688e-02 -2.15705633e-01 -2.63619237e-02 -3.81574571e-01 6.65048838e-01 -5.39343953e-01 -6.67766705e-02 6.52280807e-01 3.81581843e-01 -1.48401037e-01 -9.39623117e-01 -6.05522156e-01 5.68889901e-02 -3.58824372e-01 -5.80159605e-01 -9.18730259e-01 -2.76947647e-01 7.64867246e-01 1.28367588e-01 8.72769773e-01 6.63728535e-01 -2.10832106e-03 9.88564909e-01 7.44585812e-01 7.39565492e-01 -9.89806533e-01 3.85740727e-01 8.98507595e-01 8.80868375e-01 -1.18740213e+00 -6.22643948e-01 -6.93256795e-01 -1.18714666e+00 8.99160147e-01 1.19224679e+00 6.68699801e-01 4.13173735e-01 3.46696049e-01 4.73110050e-01 -3.17926258e-01 -9.30154920e-01 -3.94518077e-01 -1.13881506e-01 7.52758682e-01 3.74841332e-01 -6.39341772e-02 -1.58969045e-01 9.16407704e-01 -1.33132756e-01 4.21036966e-02 4.20784146e-01 1.15531921e+00 -4.71099466e-01 -1.35348547e+00 -5.03285706e-01 2.33992472e-01 -8.36823657e-02 1.56548277e-01 -9.07907486e-01 6.71303689e-01 -3.19508970e-01 1.65058768e+00 -3.31291020e-01 -5.21847606e-01 5.89406729e-01 7.11985648e-01 -2.52582412e-03 -7.19039023e-01 -9.97888505e-01 -2.11680010e-01 7.32400656e-01 -8.13859925e-02 -5.48615336e-01 -5.82878664e-02 -1.04250014e+00 -3.79913151e-01 -4.42156166e-01 5.00134885e-01 1.21807970e-01 1.24033177e+00 3.31589401e-01 4.97906059e-01 4.85896587e-01 -8.27998281e-01 -8.51508439e-01 -1.36218441e+00 -4.13389087e-01 4.62167293e-01 8.90648216e-02 -1.06509268e+00 -1.48658101e-02 -2.55415887e-01]

[12.757184982299805, 8.0005521774292]

b0d4c104-f7ed-4ea5-9645-1f736ea66b41

boosting-high-level-vision-with-joint

2010.08919

https://arxiv.org/abs/2010.08919v2

https://arxiv.org/pdf/2010.08919v2.pdf

Boosting High-Level Vision with Joint Compression Artifacts Reduction and Super-Resolution

Due to the limits of bandwidth and storage space, digital images are usually down-scaled and compressed when transmitted over networks, resulting in loss of details and jarring artifacts that can lower the performance of high-level visual tasks. In this paper, we aim to generate an artifact-free high-resolution image from a low-resolution one compressed with an arbitrary quality factor by exploring joint compression artifacts reduction (CAR) and super-resolution (SR) tasks. First, we propose a context-aware joint CAR and SR neural network (CAJNN) that integrates both local and non-local features to solve CAR and SR in one-stage. Finally, a deep reconstruction network is adopted to predict high quality and high-resolution images. Evaluation on CAR and SR benchmark datasets shows that our CAJNN model outperforms previous methods and also takes 26.2% shorter runtime. Based on this model, we explore addressing two critical challenges in high-level computer vision: optical character recognition of low-resolution texts, and extremely tiny face detection. We demonstrate that CAJNN can serve as an effective image preprocessing method and improve the accuracy for real-scene text recognition (from 85.30% to 85.75%) and the average precision for tiny face detection (from 0.317 to 0.611).

['Jan P. Allebach', 'Qian Lin', 'Xiaoyu Xiang']

2020-10-18

['scene-text-recognition']

['computer-vision']

[ 7.39101887e-01 -5.04572451e-01 1.23736240e-01 -2.48322859e-01 -9.31186199e-01 -2.47369613e-02 4.50321823e-01 -4.35302049e-01 -3.93581241e-01 4.62753922e-01 1.48759067e-01 7.11031705e-02 1.00938819e-01 -9.18036938e-01 -8.78743052e-01 -5.57926357e-01 4.32297796e-01 -4.70345803e-02 2.29427859e-01 -3.94936986e-02 4.51000303e-01 7.94762194e-01 -1.88079190e+00 6.09854639e-01 8.83673072e-01 1.08271158e+00 5.19984663e-01 8.25295866e-01 -1.44289760e-02 8.44215572e-01 -7.26726830e-01 -3.75672102e-01 2.29855970e-01 -2.52308160e-01 -4.67093825e-01 1.53811350e-01 8.44796896e-01 -9.34333682e-01 -5.29794037e-01 1.27426779e+00 7.71917403e-01 2.09400933e-02 3.95347416e-01 -6.16397023e-01 -9.84703898e-01 4.85489309e-01 -1.24403977e+00 1.85467273e-01 4.64263320e-01 9.74187180e-02 3.57228011e-01 -1.21480811e+00 4.19383854e-01 1.47515154e+00 6.91940248e-01 6.30045533e-01 -1.18714702e+00 -8.05911899e-01 -1.81565970e-01 2.75670469e-01 -1.61392057e+00 -8.52193832e-01 5.30132234e-01 -5.20140193e-02 9.40654635e-01 3.34216893e-01 3.41330111e-01 1.02335560e+00 6.02943683e-03 5.79081476e-01 9.73481357e-01 -4.24454182e-01 -1.22529514e-01 -1.85882881e-01 -2.65439481e-01 6.82885408e-01 4.85914052e-01 -1.38148621e-01 -1.03246355e+00 2.21105009e-01 1.15468860e+00 4.33581434e-02 -3.87311250e-01 3.66771877e-01 -9.82431114e-01 4.16645527e-01 2.30859905e-01 2.50609517e-01 -3.24509650e-01 6.08469220e-03 1.76084548e-01 -1.79117045e-03 4.15809155e-01 1.27384484e-01 3.99668775e-02 1.08233877e-02 -1.22566319e+00 -6.53414726e-02 3.39836210e-01 8.98395240e-01 4.27778006e-01 4.66144800e-01 -3.61513317e-01 1.24328291e+00 1.13851875e-01 7.64144361e-01 5.75395525e-01 -9.01425600e-01 5.03571987e-01 3.52770835e-01 -1.66174024e-02 -1.18135083e+00 -2.14330792e-01 -2.14970276e-01 -1.35166264e+00 2.72007167e-01 1.96082890e-01 3.79706681e-01 -7.43321002e-01 1.33646750e+00 1.98303223e-01 1.15316398e-01 2.31137574e-02 1.19912016e+00 9.68403459e-01 8.37602913e-01 -2.54505008e-01 -4.54159558e-01 1.51574194e+00 -8.61657500e-01 -8.87818098e-01 -1.30618110e-01 4.74184826e-02 -1.02149308e+00 1.27538848e+00 7.41741955e-01 -1.52780902e+00 -7.41584063e-01 -1.32159472e+00 -5.50972760e-01 9.96278524e-02 4.24692124e-01 2.13065326e-01 5.20920992e-01 -1.15578353e+00 6.73582077e-01 -4.96384650e-01 -2.57516634e-02 6.23204589e-01 2.07520738e-01 -2.07127914e-01 -4.70791310e-01 -8.42463195e-01 6.16104364e-01 1.25435935e-02 7.02758878e-02 -6.46264195e-01 -6.23337030e-01 -5.85801065e-01 7.87259936e-02 3.23480517e-01 -3.49249840e-01 8.84720266e-01 -7.57391274e-01 -1.66966939e+00 7.69017279e-01 -2.08389536e-01 -3.09821814e-01 5.69923103e-01 -3.41382056e-01 -5.47827125e-01 4.43396330e-01 -9.70787778e-02 4.03746754e-01 1.34618127e+00 -9.78182793e-01 -5.73458016e-01 -6.02446318e-01 -5.78450441e-01 3.66320282e-01 -5.53642392e-01 4.36820269e-01 -9.61872041e-01 -7.21336722e-01 6.60755932e-02 -3.10179472e-01 2.89142072e-01 4.04835582e-01 -2.87662148e-01 -4.33035120e-02 1.03781164e+00 -1.04883432e+00 1.02964962e+00 -2.15187573e+00 -8.21537450e-02 -2.96072364e-01 2.84550518e-01 5.49714983e-01 -3.54598939e-01 -3.83304149e-01 1.64352432e-01 -2.62605958e-02 -1.14473894e-01 -5.00259817e-01 -3.51051807e-01 -1.92536995e-01 -3.35397065e-01 5.95684946e-01 2.67272592e-01 7.00854123e-01 -3.77759129e-01 -5.83650053e-01 3.87275726e-01 1.17706847e+00 -3.13722700e-01 1.58850268e-01 1.62280977e-01 1.09271042e-01 -1.14529364e-01 9.50452030e-01 1.19943249e+00 -2.40970343e-01 -2.98963720e-03 -6.33874178e-01 -2.14467883e-01 4.46904413e-02 -1.26400900e+00 1.49439240e+00 -4.44674015e-01 8.21660817e-01 3.61712515e-01 -6.82635844e-01 1.17563057e+00 -4.11844775e-02 3.07345569e-01 -1.39057302e+00 1.64710984e-01 3.07351518e-02 -6.08260870e-01 -3.33717972e-01 8.93217266e-01 2.48003229e-01 3.07775706e-01 4.27371562e-01 -3.60985875e-01 1.48046538e-01 2.92616598e-02 4.73526381e-02 9.11072135e-01 1.61706191e-03 4.67832461e-02 -8.19295794e-02 6.77053094e-01 -5.16511977e-01 3.73686016e-01 7.53201306e-01 -1.32587226e-03 9.97172356e-01 2.95528382e-01 -4.17528540e-01 -1.48896968e+00 -8.65537405e-01 -3.89686525e-01 1.12609255e+00 1.83872864e-01 -2.55738556e-01 -9.45855319e-01 -2.34246831e-02 -3.21994483e-01 3.17262948e-01 -1.23496056e-01 -7.19944239e-02 -9.60271597e-01 -8.69544089e-01 6.49815679e-01 4.66664165e-01 9.38899994e-01 -8.12708676e-01 -5.98245382e-01 -1.05504259e-01 -2.45596141e-01 -1.60903859e+00 -6.32338703e-01 -3.84090304e-01 -6.90770686e-01 -7.64082015e-01 -8.17039192e-01 -6.70396268e-01 5.77961445e-01 5.75237334e-01 9.33782578e-01 2.70576894e-01 -7.10434198e-01 7.47967931e-03 -1.73238546e-01 2.46595182e-02 -3.60735595e-01 -3.93367887e-01 1.72949210e-01 1.89953670e-01 2.13188544e-01 -4.35142100e-01 -7.17377484e-01 2.53033817e-01 -9.65642154e-01 3.29407901e-01 8.15398037e-01 8.13349903e-01 8.67043138e-01 9.00272578e-02 2.83552855e-01 -4.46173906e-01 5.71347594e-01 1.29869804e-01 -8.91932547e-01 1.08822398e-01 -7.00000346e-01 -2.11411677e-02 9.21973646e-01 -6.17040753e-01 -1.18212366e+00 8.10792521e-02 -6.61095530e-02 -5.69035470e-01 1.43455360e-02 -2.05539674e-01 -2.58930802e-01 -2.20896423e-01 6.81066215e-01 7.08951831e-01 7.61273429e-02 -6.17070735e-01 1.66077033e-01 9.67485547e-01 9.69330192e-01 -2.02789843e-01 6.90057516e-01 5.56586981e-01 1.16160974e-01 -1.21500647e+00 -6.89168870e-01 -1.74056944e-02 -4.77157891e-01 -6.81781992e-02 7.73472369e-01 -1.23485458e+00 -8.63454819e-01 7.74041891e-01 -1.15437031e+00 -1.62641525e-01 -1.52308956e-01 3.15871954e-01 -3.25439990e-01 7.91872680e-01 -9.12686229e-01 -8.37101221e-01 -8.52078557e-01 -1.09659326e+00 1.34058261e+00 2.72266656e-01 4.83817160e-01 -5.31075969e-02 -5.58800459e-01 5.29555798e-01 6.26067698e-01 -5.76072372e-02 4.07010585e-01 1.20576940e-01 -8.48124862e-01 1.05358928e-01 -1.00929260e+00 4.42013264e-01 -2.26272002e-01 -6.06781729e-02 -9.88553703e-01 -4.94744688e-01 1.31761327e-01 -4.51749533e-01 9.70981121e-01 3.62113535e-01 1.56759310e+00 -4.01686728e-01 8.73171091e-02 1.09742296e+00 1.58043623e+00 -1.05999410e-01 1.02103996e+00 1.72338098e-01 8.50747228e-01 2.29170427e-01 5.54064631e-01 7.63949633e-01 -1.31227812e-02 8.83186579e-01 2.90064961e-01 -3.00710723e-02 -6.22111559e-01 -9.85097885e-02 4.93891984e-01 6.77673042e-01 -1.31722584e-01 -6.92154616e-02 -5.24542689e-01 1.12068161e-01 -1.54852593e+00 -9.12710071e-01 -2.01813400e-01 2.25591087e+00 1.02852857e+00 -1.63920045e-01 -7.88784698e-02 2.57435858e-01 1.17602825e+00 1.89178988e-01 -7.41196990e-01 -2.53257781e-01 -5.18627763e-01 2.08279446e-01 5.26573300e-01 3.91599715e-01 -8.87721896e-01 8.97533476e-01 5.90893078e+00 1.18703127e+00 -1.15765584e+00 6.68422803e-02 7.78301179e-01 -3.11705172e-01 1.89406782e-01 -6.52055562e-01 -1.15136337e+00 5.01526713e-01 9.27153766e-01 1.11891784e-01 8.61158013e-01 7.76049316e-01 1.29005864e-01 -7.42034800e-03 -8.05646181e-01 1.56164527e+00 5.91690004e-01 -1.39685106e+00 3.41255844e-01 -4.81562652e-02 5.54665864e-01 -1.81813568e-01 3.10321540e-01 -7.59607553e-02 -1.03297800e-01 -1.38010919e+00 5.89889407e-01 6.34140968e-01 1.76705194e+00 -9.80525970e-01 3.62340659e-01 3.03078085e-01 -1.27528191e+00 -3.30454558e-02 -8.84238899e-01 2.64149934e-01 -8.45300779e-02 6.97470367e-01 -3.31667125e-01 1.36063665e-01 1.00732493e+00 6.23741269e-01 -4.08694506e-01 5.41814983e-01 6.33450747e-02 3.85307670e-02 -3.04313391e-01 8.99648964e-02 -3.51656854e-01 -1.61750183e-01 3.41720551e-01 1.30758429e+00 5.09671628e-01 3.61184388e-01 -2.71583587e-01 1.02313280e+00 -5.77403605e-01 -5.58161624e-02 -2.77371407e-01 3.60940486e-01 5.90054572e-01 1.29529822e+00 -3.93749535e-01 -2.59762287e-01 -4.67945069e-01 1.35366285e+00 1.91900611e-01 9.40195248e-02 -8.30499947e-01 -4.80566919e-01 3.87768209e-01 2.31113598e-01 3.28917086e-01 -5.11497222e-02 -3.60968918e-01 -1.33512020e+00 3.00136238e-01 -9.80078101e-01 3.46619003e-02 -9.95028973e-01 -9.69243228e-01 6.50228739e-01 -6.13188028e-01 -1.07018805e+00 8.09528157e-02 -6.70351446e-01 -1.19022511e-01 9.89390910e-01 -1.85869765e+00 -9.90330279e-01 -7.54557371e-01 8.39772046e-01 9.18464422e-01 -2.73168325e-01 5.61257184e-01 4.65606540e-01 -7.06021965e-01 9.09438729e-01 3.85297418e-01 1.76946133e-01 7.42974758e-01 -7.07390964e-01 3.06026757e-01 1.16390824e+00 -1.03280000e-01 1.61713973e-01 4.82792795e-01 -5.36755323e-01 -1.88055778e+00 -1.28456807e+00 5.23854792e-01 -7.56256357e-02 7.82344639e-02 -3.54259044e-01 -1.19753516e+00 6.82694465e-02 -1.36600539e-01 2.43175060e-01 8.39456096e-02 -5.93276322e-01 -5.82163393e-01 -4.03905869e-01 -1.42842579e+00 6.04724050e-01 1.03858364e+00 -5.65961063e-01 -1.25890970e-01 1.47530004e-01 7.97232687e-01 -3.82831484e-01 -9.04179335e-01 2.74017364e-01 6.29072726e-01 -1.25571752e+00 1.38290834e+00 2.38820612e-01 7.54393637e-01 -1.31634608e-01 -3.32762718e-01 -5.75802088e-01 -3.76770824e-01 -5.47596872e-01 -4.46173280e-01 1.14018881e+00 -1.11493379e-01 -1.89778134e-01 7.25190103e-01 3.39702845e-01 1.56161934e-01 -4.52458233e-01 -1.00957584e+00 -5.27102828e-01 -2.02262238e-01 -2.80353129e-01 5.55544376e-01 6.92607939e-01 -4.41914797e-01 3.75860125e-01 -8.09163511e-01 2.54633158e-01 1.11290765e+00 1.14559680e-01 6.68249846e-01 -1.01427889e+00 -1.36025682e-01 -5.07508695e-01 -1.51930779e-01 -1.42857409e+00 -1.44909292e-01 -3.34676504e-01 3.03168930e-02 -1.30450261e+00 5.46582699e-01 -6.16682805e-02 5.22022285e-02 5.64259738e-02 -3.60346809e-02 7.07270086e-01 2.65756696e-01 4.62839365e-01 -6.17796600e-01 4.93901193e-01 1.30302370e+00 -3.62541750e-02 5.78006245e-02 -3.77602220e-01 -6.92519486e-01 6.62034035e-01 5.69553614e-01 -4.15191092e-02 3.67831402e-02 -6.31694317e-01 1.23984210e-01 3.07843834e-01 3.17572922e-01 -1.00022864e+00 3.92160922e-01 2.92811845e-03 9.55495179e-01 -8.23527515e-01 5.34575105e-01 -6.21573031e-01 -2.14611083e-01 3.13183367e-01 -3.21606368e-01 -2.81101376e-01 5.72790690e-02 5.59263647e-01 -3.07059269e-02 -8.10598284e-02 1.38345516e+00 8.29553902e-02 -6.18178904e-01 2.09563106e-01 -1.59716755e-01 -2.33194441e-01 6.67248964e-01 -4.00435716e-01 -7.15303719e-01 -2.48749703e-01 -2.30411738e-01 -2.22870171e-01 4.47335958e-01 3.64671022e-01 1.31540215e+00 -1.23796201e+00 -1.02999806e+00 5.73917270e-01 -3.02188963e-01 1.81908339e-01 5.36475420e-01 6.50714636e-01 -6.75913274e-01 2.83264279e-01 -3.00819188e-01 -6.87317789e-01 -1.61938083e+00 5.49816072e-01 1.79394871e-01 -9.72631425e-02 -1.00589836e+00 7.11698830e-01 -3.31566408e-02 4.45174687e-02 5.29841304e-01 2.74428781e-02 -2.85950899e-01 -3.01098704e-01 1.26133704e+00 6.25055254e-01 2.88371332e-02 -7.84700394e-01 -9.81589109e-02 9.57437396e-01 -2.75104463e-01 2.15800941e-01 1.41326547e+00 -3.63601238e-01 -1.82447329e-01 -1.62302442e-02 1.16058838e+00 -3.05657592e-02 -1.46213305e+00 -4.72076923e-01 -4.78126168e-01 -9.13075864e-01 4.10261571e-01 -6.69428885e-01 -1.28123343e+00 8.41630340e-01 8.04778159e-01 -7.27421343e-02 1.54986179e+00 -2.20414907e-01 9.78175879e-01 2.96200991e-01 1.26684502e-01 -1.30522871e+00 4.49064642e-01 3.16544235e-01 1.12243485e+00 -1.19223976e+00 3.75807762e-01 -4.51836020e-01 -5.33052623e-01 1.24307764e+00 6.66215777e-01 -6.25612214e-02 -3.38958725e-02 6.97279692e-01 -3.52902800e-01 1.50760055e-01 -6.11398101e-01 1.95076659e-01 8.34165141e-02 5.87090313e-01 3.16963404e-01 -2.03456610e-01 -2.63847429e-02 5.54825485e-01 -7.42347687e-02 3.81115563e-02 7.98946083e-01 5.77874541e-01 -6.39307380e-01 -6.04352832e-01 -7.58756399e-01 5.32246947e-01 -7.09297240e-01 -2.93961376e-01 -2.87466981e-02 3.08410227e-01 -8.05385932e-02 8.14518273e-01 3.79102021e-01 -3.38213235e-01 2.76837647e-01 -4.09192830e-01 4.49565321e-01 -1.37895003e-01 -1.04395121e-01 4.06262010e-01 -2.16327518e-01 -7.82709181e-01 -2.57055402e-01 -3.45324904e-01 -1.05912113e+00 -8.17813098e-01 -3.51754516e-01 -4.94210750e-01 8.24808300e-01 4.43457305e-01 5.06187797e-01 5.99483252e-01 6.47976637e-01 -9.86353695e-01 -7.39100993e-01 -8.52309167e-01 -6.84170127e-01 3.41167986e-01 3.40587795e-01 -1.18601941e-01 -4.44263846e-01 9.01847854e-02]

[11.271931648254395, -1.9614120721817017]

55d53fa7-070c-4988-af56-97ccd968961c

proxyformer-proxy-alignment-assisted-point

2302.14435

https://arxiv.org/abs/2302.14435v1

https://arxiv.org/pdf/2302.14435v1.pdf

ProxyFormer: Proxy Alignment Assisted Point Cloud Completion with Missing Part Sensitive Transformer

Problems such as equipment defects or limited viewpoints will lead the captured point clouds to be incomplete. Therefore, recovering the complete point clouds from the partial ones plays an vital role in many practical tasks, and one of the keys lies in the prediction of the missing part. In this paper, we propose a novel point cloud completion approach namely ProxyFormer that divides point clouds into existing (input) and missing (to be predicted) parts and each part communicates information through its proxies. Specifically, we fuse information into point proxy via feature and position extractor, and generate features for missing point proxies from the features of existing point proxies. Then, in order to better perceive the position of missing points, we design a missing part sensitive transformer, which converts random normal distribution into reasonable position information, and uses proxy alignment to refine the missing proxies. It makes the predicted point proxies more sensitive to the features and positions of the missing part, and thus make these proxies more suitable for subsequent coarse-to-fine processes. Experimental results show that our method outperforms state-of-the-art completion networks on several benchmark datasets and has the fastest inference speed. Code is available at https://github.com/I2-Multimedia-Lab/ProxyFormer.

['Mingqiang Wei', 'Xiaoyang Tan', 'Pan Gao', 'Shanshan Li']

2023-02-28

http://openaccess.thecvf.com//content/CVPR2023/html/Li_ProxyFormer_Proxy_Alignment_Assisted_Point_Cloud_Completion_With_Missing_Part_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Li_ProxyFormer_Proxy_Alignment_Assisted_Point_Cloud_Completion_With_Missing_Part_CVPR_2023_paper.pdf

cvpr-2023-1

['point-cloud-completion']

['computer-vision']

[-6.56252950e-02 -1.10186785e-01 -1.23680644e-02 -3.96018296e-01 -8.23466301e-01 -4.51765627e-01 2.34917611e-01 -4.61945646e-02 8.69487002e-02 4.39143807e-01 2.23485783e-01 1.50716260e-01 -7.21645132e-02 -9.37697887e-01 -1.07958376e+00 -6.20142281e-01 3.36089671e-01 9.19496357e-01 5.20577312e-01 -1.29299834e-02 2.45064527e-01 8.00845206e-01 -1.45864105e+00 2.42007092e-01 8.37302029e-01 1.07703102e+00 7.89538443e-01 8.38077664e-02 -2.52277434e-01 2.33671173e-01 -3.55711043e-01 -2.25416217e-02 3.75972241e-01 4.79899138e-01 -2.41795138e-01 2.14397013e-01 2.28078857e-01 -7.29072809e-01 -5.34828424e-01 1.11369061e+00 1.17430702e-01 1.78681128e-02 3.30229044e-01 -1.14366186e+00 -4.86763299e-01 2.78364211e-01 -6.86064601e-01 -3.80884916e-01 2.18879089e-01 1.36284113e-01 6.71167493e-01 -1.38537717e+00 3.65735263e-01 1.45041513e+00 5.89514554e-01 1.80377975e-01 -9.18850482e-01 -9.91281152e-01 3.61713946e-01 9.25605372e-03 -1.53092670e+00 -2.85099030e-01 9.96931911e-01 -3.40010494e-01 2.19962463e-01 1.64183080e-01 4.96411651e-01 7.45980859e-01 -3.22980076e-01 8.55377913e-01 5.70722222e-01 9.68239680e-02 -5.71926916e-03 -1.45909846e-01 -2.50268042e-01 5.16443729e-01 2.33081996e-01 -1.00055285e-01 -1.44540265e-01 -3.40412527e-01 1.20738792e+00 8.07498276e-01 -5.97215831e-01 -4.53135133e-01 -1.52095425e+00 4.39648896e-01 8.34153891e-01 -9.67652872e-02 -5.75515568e-01 2.12310821e-01 -1.89917609e-01 -1.64547235e-01 4.08171088e-01 -9.22342837e-02 -5.84702730e-01 8.74499083e-02 -7.87144721e-01 2.27587193e-01 2.80281961e-01 1.39942646e+00 1.15841019e+00 -3.59160781e-01 3.46175730e-02 7.91463196e-01 6.73397183e-01 6.93315685e-01 3.11080143e-02 -1.15854120e+00 1.13039625e+00 8.56815934e-01 4.65064168e-01 -1.13035309e+00 -1.38537869e-01 -2.63278544e-01 -9.93407547e-01 8.84614065e-02 1.00041367e-01 1.00473814e-01 -7.88686454e-01 1.18460059e+00 6.27464533e-01 6.53436482e-01 -3.47939640e-01 1.08059454e+00 8.52226853e-01 1.00189114e+00 -4.59415674e-01 4.52102721e-02 1.16387379e+00 -6.47673488e-01 -2.33869746e-01 -2.63780802e-01 -1.93008687e-02 -9.70837653e-01 8.91525984e-01 4.10008639e-01 -1.00160146e+00 -8.78181696e-01 -9.14631844e-01 -2.89640844e-01 1.42668739e-01 5.47488511e-01 3.95646930e-01 -2.68089950e-01 -6.18028104e-01 7.65533090e-01 -8.30051064e-01 2.29922131e-01 5.94332874e-01 3.28959435e-01 -2.40878925e-01 -5.44554830e-01 -7.48096764e-01 2.10023880e-01 1.33621782e-01 3.85138631e-01 -7.82318234e-01 -9.82379854e-01 -5.81773579e-01 1.19348392e-01 4.16335046e-01 -7.82431841e-01 1.00148571e+00 -3.99206817e-01 -9.20685291e-01 2.48471394e-01 -2.18373850e-01 2.60816246e-01 5.96330822e-01 -2.51457840e-01 -1.12308167e-01 6.98269457e-02 3.85858983e-01 7.44280875e-01 8.49343717e-01 -1.72896659e+00 -9.54032302e-01 -6.64462745e-01 -1.58843502e-01 1.67283967e-01 1.41548887e-01 -4.74720538e-01 -1.07787454e+00 -4.32468295e-01 8.06971490e-01 -8.07328820e-01 -1.90211311e-01 6.55896842e-01 -4.84214604e-01 -3.07927340e-01 9.30567443e-01 -8.22163641e-01 4.44840312e-01 -2.43016505e+00 1.57333887e-03 3.43089521e-01 3.84808183e-01 2.26042476e-02 -1.13732554e-01 4.49984968e-01 -2.64051463e-02 -1.93573356e-01 -1.26679003e-01 -7.03127265e-01 -9.08133760e-02 3.85125667e-01 -7.20848322e-01 4.61424053e-01 4.37462050e-03 3.82374287e-01 -7.43925810e-01 -3.19964468e-01 5.39781034e-01 7.49255598e-01 -3.67811292e-01 2.34807953e-01 -2.59839714e-01 5.89630961e-01 -9.34298933e-01 9.33865368e-01 1.31520391e+00 -1.47637382e-01 -5.08995831e-01 -5.03739178e-01 -2.58660406e-01 8.33887383e-02 -1.48724079e+00 1.93307054e+00 -3.58838052e-01 1.94749281e-01 2.33630016e-01 -4.10196990e-01 1.26336002e+00 1.93632007e-01 6.15828574e-01 -1.53168991e-01 -9.34347212e-02 2.18830928e-01 -5.87656200e-01 -2.50578105e-01 3.96919549e-01 2.91551322e-01 2.61848301e-01 -2.12428167e-01 -4.62472111e-01 -4.22626555e-01 -3.49859238e-01 7.39745647e-02 8.28759611e-01 3.40283304e-01 -2.75451094e-01 4.65935320e-01 5.52679420e-01 7.78492866e-03 1.05710006e+00 2.41946056e-01 1.74999207e-01 1.28084433e+00 2.23268822e-01 -4.91155833e-01 -1.08670461e+00 -1.18110263e+00 -4.54249792e-03 2.22193733e-01 6.22155905e-01 -1.71244636e-01 -2.53090352e-01 -3.73110503e-01 2.73707271e-01 5.37721813e-01 -2.44965807e-01 7.60719478e-02 -6.49682522e-01 -1.79935336e-01 -2.83450987e-02 7.81523764e-01 5.38448751e-01 -8.15706968e-01 4.36045267e-02 2.92454928e-01 -4.63060468e-01 -1.08358598e+00 -5.32430768e-01 -3.73970896e-01 -1.19457245e+00 -1.10225785e+00 -6.82623863e-01 -7.14496672e-01 9.65922594e-01 7.54132926e-01 8.32496524e-01 4.54048455e-01 1.95267633e-01 7.46951029e-02 -2.17920288e-01 -4.01699811e-01 1.27006367e-01 -2.96094447e-01 -1.40734827e-02 9.15669128e-02 1.07910387e-01 -8.72420192e-01 -7.93157756e-01 5.39268672e-01 -7.36973941e-01 1.22206934e-01 9.05462205e-01 5.74613333e-01 1.18220568e+00 1.10486776e-01 -6.89864904e-02 -4.82700348e-01 1.35775194e-01 -5.13940215e-01 -9.19854879e-01 -1.69948697e-01 -1.93793699e-01 -3.08495671e-01 6.50562882e-01 -2.85192758e-01 -8.77905011e-01 4.47291285e-01 -1.40782520e-01 -1.14240277e+00 -1.79713696e-01 2.94161469e-01 -7.20897257e-01 6.85212687e-02 4.72523123e-02 2.93853015e-01 -1.40247658e-01 -9.76458132e-01 1.48669645e-01 5.11659682e-01 5.88267028e-01 -6.27191603e-01 1.24260437e+00 8.49146783e-01 8.63540396e-02 -6.72889352e-01 -5.96012414e-01 -7.04084873e-01 -6.56050503e-01 -1.59329474e-01 4.99276012e-01 -1.21923125e+00 -6.19339883e-01 2.19626874e-01 -1.66346323e+00 3.08916420e-01 -3.37253034e-01 4.70433623e-01 -2.97985613e-01 4.96103078e-01 -5.55654109e-01 -3.95060360e-01 -1.86954945e-01 -1.29358411e+00 1.53234768e+00 2.94344693e-01 3.12910050e-01 -3.02019894e-01 -6.13172501e-02 3.82471144e-01 -2.37160698e-01 1.11495517e-01 5.99510849e-01 -2.06153691e-01 -1.40745807e+00 -4.07107532e-01 -4.70400482e-01 4.02483195e-01 1.38376523e-02 2.31392965e-01 -8.83596778e-01 -1.66117147e-01 1.35295227e-01 2.39594474e-01 6.07410789e-01 3.42095912e-01 1.44684649e+00 -9.16020349e-02 -6.54502392e-01 9.10263538e-01 1.42989504e+00 -4.12873039e-03 7.21577585e-01 1.05251335e-01 1.11952651e+00 5.32253802e-01 1.03732014e+00 6.26879513e-01 6.90673470e-01 5.30928373e-01 1.00503302e+00 1.99323166e-02 -7.34875873e-02 -7.02531457e-01 -1.47298500e-02 1.04537749e+00 -1.86690256e-01 4.92112786e-02 -8.16902995e-01 5.48993409e-01 -1.92955601e+00 -7.42793858e-01 -4.96219248e-01 2.15872049e+00 3.83766592e-01 3.47288512e-02 -3.73904765e-01 2.99945120e-02 9.68772054e-01 1.54585809e-01 -5.31556129e-01 6.52027428e-01 2.38161534e-01 -1.31949827e-01 4.11793172e-01 2.63622582e-01 -7.39671052e-01 5.45780599e-01 3.68089223e+00 8.38226497e-01 -6.67039931e-01 -4.91030887e-02 2.80409366e-01 1.81221530e-01 -3.88285309e-01 3.62399995e-01 -8.02004993e-01 7.98814535e-01 -2.40150224e-02 2.55885541e-01 3.36658329e-01 1.09794188e+00 2.44280666e-01 1.32693365e-01 -1.08219469e+00 1.24714220e+00 -2.99275279e-01 -1.40354919e+00 1.03508413e-01 2.14005038e-01 6.10526085e-01 3.16104382e-01 -1.46474183e-01 1.24079473e-01 1.27862886e-01 -4.38456208e-01 7.47137785e-01 9.69027936e-01 5.94175398e-01 -8.01290989e-01 7.69086361e-01 7.36821234e-01 -1.48280501e+00 4.12975699e-02 -1.05138671e+00 3.58610973e-02 1.26504794e-01 1.01466942e+00 -7.14605033e-01 6.25083387e-01 8.49216282e-01 8.17219853e-01 -2.58156359e-01 1.42103601e+00 -3.96820486e-01 2.80030340e-01 -5.69939137e-01 3.50161105e-01 -8.14169794e-02 -4.86725986e-01 6.73635185e-01 4.03285325e-01 7.14176834e-01 3.33054870e-01 4.95044768e-01 1.03683031e+00 -2.06216380e-01 -2.38312438e-01 -3.86653125e-01 6.08299255e-01 1.15358055e+00 1.31469429e+00 -5.12764096e-01 -3.18309903e-01 -4.27043736e-01 7.18355715e-01 2.08574638e-01 3.05158854e-01 -7.45887160e-01 -2.24054918e-01 8.67190540e-01 2.62929410e-01 4.11558181e-01 -2.98424393e-01 -3.71776402e-01 -1.22055364e+00 5.75070679e-01 -3.93840611e-01 -5.19139506e-02 -1.28145957e+00 -1.23775542e+00 3.25794995e-01 -8.71033743e-02 -1.88093662e+00 2.70514339e-01 -2.59444684e-01 -7.73302495e-01 1.06309712e+00 -1.43835926e+00 -1.33844817e+00 -9.20177042e-01 6.39491379e-01 6.61075115e-01 2.48416722e-01 3.33718717e-01 4.02147442e-01 -2.93640792e-01 -1.49737433e-01 2.14149743e-01 1.70915276e-01 5.87171137e-01 -8.32044005e-01 6.27304912e-01 7.07747102e-01 -8.46625492e-02 4.87007320e-01 3.81488860e-01 -9.51048911e-01 -1.46565008e+00 -1.31601894e+00 4.90757048e-01 -4.70240533e-01 1.96741208e-01 -2.21863449e-01 -1.04364789e+00 6.91501677e-01 -6.30844355e-01 2.71650523e-01 -7.91435316e-02 -2.90928572e-01 -8.23112652e-02 -4.85537320e-01 -1.03037250e+00 4.91416037e-01 8.49061310e-01 -3.41485232e-01 -5.78119457e-01 4.42341894e-01 9.91259038e-01 -7.14332938e-01 -6.97219670e-01 5.66605449e-01 1.73441827e-01 -8.75144362e-01 1.27087963e+00 2.60759920e-01 4.70802963e-01 -7.41225302e-01 -2.36715093e-01 -1.27232993e+00 -4.23143387e-01 -1.09909602e-01 1.44565821e-01 1.50982308e+00 -2.00105123e-02 -6.47288084e-01 1.03837442e+00 5.85770845e-01 -6.46378815e-01 -7.17536569e-01 -8.52469385e-01 -4.06939685e-01 -4.05634254e-01 -5.45592368e-01 1.28845990e+00 7.57060349e-01 -7.50292003e-01 1.49135850e-02 -2.33573824e-01 8.80763471e-01 7.20470190e-01 5.21198332e-01 1.12982106e+00 -1.47785318e+00 2.85937693e-02 9.52589735e-02 -3.50633889e-01 -1.60691726e+00 -1.99773207e-01 -4.82832521e-01 1.62290841e-01 -1.82145572e+00 -1.20043948e-01 -9.35548306e-01 1.45432293e-01 3.95208061e-01 -3.70238721e-02 -5.58842868e-02 2.84401149e-01 7.37816215e-01 -2.34851286e-01 8.16503346e-01 1.39626396e+00 -2.98343122e-01 -5.31605259e-02 5.45743823e-01 -4.86771673e-01 9.61937666e-01 6.23085380e-01 -5.29528439e-01 -2.14890987e-01 -8.82985294e-01 -1.92351476e-03 4.29260373e-01 5.51076472e-01 -1.35071230e+00 4.01221335e-01 -1.77577689e-01 8.32065761e-01 -1.38829064e+00 8.66139472e-01 -1.39339387e+00 3.90913159e-01 1.36220708e-01 2.78273135e-01 1.25000805e-01 -1.10153407e-01 6.97502196e-01 -1.90945879e-01 -1.75479949e-01 2.42452174e-01 -2.15694532e-01 -5.08430123e-01 9.98695731e-01 4.18163329e-01 -4.04156893e-01 8.51419210e-01 -2.33448163e-01 -2.83140361e-01 -2.50340074e-01 -4.24877584e-01 6.06319606e-01 6.74905062e-01 5.40271819e-01 1.13951039e+00 -1.43931508e+00 -7.08653867e-01 3.48231256e-01 1.17239937e-01 1.11194754e+00 4.03418392e-01 4.93541539e-01 -7.59859800e-01 1.19101875e-01 1.53277114e-01 -9.57362592e-01 -1.09146607e+00 5.75071156e-01 -1.04635656e-01 2.40671724e-01 -8.29044044e-01 6.97958946e-01 3.77583653e-01 -6.20985150e-01 2.25710586e-01 -6.61291838e-01 7.22148567e-02 -1.73481673e-01 4.82045740e-01 5.21442115e-01 7.35016465e-02 -6.20463371e-01 -1.37439117e-01 8.83478045e-01 -4.40104445e-03 2.82321930e-01 1.50670016e+00 -1.99879959e-01 -1.96276546e-01 2.48294711e-01 1.01357734e+00 2.23781794e-01 -1.68645704e+00 -3.78308266e-01 -5.46254992e-01 -8.68193030e-01 9.75952670e-03 -3.22806031e-01 -1.35715115e+00 1.00452459e+00 2.91375071e-01 -2.25375056e-01 8.90066326e-01 1.03058614e-01 9.57195640e-01 2.29041234e-01 7.82788873e-01 -6.24672413e-01 -2.13537037e-01 2.50841796e-01 1.06135404e+00 -1.02115881e+00 3.10183913e-01 -9.08011317e-01 -2.88067847e-01 1.09221876e+00 7.91042030e-01 -3.84134352e-01 6.07367039e-01 1.69370193e-02 -1.44031167e-01 -2.90361971e-01 -4.74551886e-01 2.67559052e-01 1.80220962e-01 7.03660250e-01 -3.59402746e-01 2.10924461e-01 3.51291299e-01 6.14941835e-01 -1.38583720e-01 -3.06235161e-02 2.96347529e-01 7.10470080e-01 -4.57938671e-01 -9.36871171e-01 -9.10692990e-01 4.96144235e-01 1.23314053e-01 8.29004869e-02 -1.33784354e-01 6.90427125e-01 3.90074909e-01 7.16318369e-01 2.25049436e-01 -4.17161971e-01 5.58711648e-01 -6.05375111e-01 2.25402862e-02 -6.54428601e-01 -4.69259769e-02 2.31164485e-01 -3.93472224e-01 -8.23746085e-01 -1.30306512e-01 -7.10103571e-01 -1.51768064e+00 -3.37544888e-01 -3.70372862e-01 1.07402451e-01 7.07537174e-01 6.34375930e-01 4.30889547e-01 4.10424709e-01 7.99457490e-01 -1.44080245e+00 -5.24874747e-01 -8.61609519e-01 -3.92149955e-01 2.80647516e-01 3.71665657e-01 -6.74520791e-01 -2.43187875e-01 -2.78701633e-01]

[8.31652545928955, -3.551234722137451]

54621054-1352-4394-a675-ac983e03f03f

score-refinement-for-confidence-based-3d

2107.04327

https://arxiv.org/abs/2107.04327v1

https://arxiv.org/pdf/2107.04327v1.pdf

Score refinement for confidence-based 3D multi-object tracking

Multi-object tracking is a critical component in autonomous navigation, as it provides valuable information for decision-making. Many researchers tackled the 3D multi-object tracking task by filtering out the frame-by-frame 3D detections; however, their focus was mainly on finding useful features or proper matching metrics. Our work focuses on a neglected part of the tracking system: score refinement and tracklet termination. We show that manipulating the scores depending on time consistency while terminating the tracklets depending on the tracklet score improves tracking results. We do this by increasing the matched tracklets' score with score update functions and decreasing the unmatched tracklets' score. Compared to count-based methods, our method consistently produces better AMOTA and MOTA scores when utilizing various detectors and filtering algorithms on different datasets. The improvements in AMOTA score went up to 1.83 and 2.96 in MOTA. We also used our method as a late-fusion ensembling method, and it performed better than voting-based ensemble methods by a solid margin. It achieved an AMOTA score of 67.6 on nuScenes test evaluation, which is comparable to other state-of-the-art trackers. Code is publicly available at: \url{https://github.com/cogsys-tuebingen/CBMOT}.

['Andreas Zell', 'Jona Schröder', 'Nuri Benbarka']

2021-07-09

['3d-multi-object-tracking']

['computer-vision']

[-4.13254410e-01 -5.16151428e-01 -6.66152015e-02 7.60124698e-02 -8.66823912e-01 -7.51180053e-01 5.51371038e-01 3.71125974e-02 -8.73049438e-01 6.29800797e-01 -2.02942744e-01 -6.09628446e-02 -1.03037573e-01 -5.63429654e-01 -5.97828090e-01 -8.25776279e-01 2.05749571e-02 5.33828616e-01 1.14438963e+00 -2.74246305e-01 3.28642391e-02 5.89504063e-01 -1.99663353e+00 -6.96663409e-02 5.52350163e-01 9.27335083e-01 2.74793863e-01 8.00476313e-01 1.61982253e-01 3.63920927e-01 -7.68853903e-01 -3.07416171e-01 4.69396770e-01 -2.75704980e-01 -8.71727429e-03 -7.07867682e-01 9.50420201e-01 1.87032241e-02 -2.31892899e-01 1.20693505e+00 6.69078648e-01 1.86990395e-01 4.01317358e-01 -1.28468919e+00 -9.20222141e-03 2.08742917e-01 -7.07535565e-01 4.84647840e-01 2.90972650e-01 3.26275751e-02 8.52165043e-01 -6.71231449e-01 6.72055781e-01 1.08873653e+00 1.02883244e+00 6.08221292e-01 -1.00141716e+00 -1.07834113e+00 1.49201795e-01 2.92301178e-01 -1.28723598e+00 -4.35993791e-01 1.06106341e-01 -6.77995086e-01 6.27059698e-01 5.26119888e-01 7.65484214e-01 6.11007690e-01 5.94761431e-01 5.74418902e-01 9.02463138e-01 -3.22112262e-01 -1.82543471e-02 -8.44018534e-02 3.13434929e-01 8.29674482e-01 7.78808534e-01 6.10992193e-01 -4.85536158e-01 -1.39960885e-01 5.66279113e-01 -1.56150103e-01 1.39746685e-02 -3.94101739e-01 -1.43760872e+00 7.21946299e-01 4.28448498e-01 2.59398669e-01 -2.91316569e-01 3.28646779e-01 2.49695152e-01 2.89251059e-01 3.69708061e-01 2.51597673e-01 -3.68481666e-01 -5.84072992e-02 -8.70120287e-01 5.87677658e-01 3.52270097e-01 8.29827964e-01 4.91426438e-01 -1.12259895e-01 -4.95185465e-01 4.37481403e-01 4.73443568e-01 9.25814927e-01 1.54079691e-01 -9.67381120e-01 5.34002110e-02 4.93319631e-01 4.47318852e-01 -5.88539481e-01 -8.08892250e-01 -6.17917061e-01 -2.80123621e-01 9.54785526e-01 8.50854874e-01 -2.51378238e-01 -1.07719862e+00 1.71888196e+00 6.97852969e-01 -1.78880966e-03 -2.78305084e-01 1.11517036e+00 8.98043215e-01 2.30292141e-01 5.85778356e-02 -3.93671393e-02 1.65814185e+00 -8.46990287e-01 -7.52729177e-01 -8.47658366e-02 6.61775470e-01 -1.08834219e+00 3.24600697e-01 2.46145502e-01 -8.34499836e-01 -8.44882131e-01 -1.11662364e+00 3.87886763e-01 -2.59590715e-01 4.01586831e-01 4.09890980e-01 8.97173882e-01 -1.00193429e+00 3.82004499e-01 -1.02112055e+00 -5.06168783e-01 1.16224512e-01 3.87423187e-01 -2.42991135e-01 3.01032096e-01 -7.89270937e-01 1.37572217e+00 4.10373449e-01 -2.66750544e-01 -5.77404618e-01 -5.29811323e-01 -5.90721011e-01 -3.47507805e-01 3.95369112e-01 -8.27145517e-01 1.32022274e+00 -1.58250898e-01 -1.26531911e+00 8.25302005e-01 -1.18163154e-01 -5.97335696e-01 6.69082105e-01 -4.90082055e-01 -5.83612740e-01 -3.29393625e-01 2.54438460e-01 6.22847736e-01 4.98933971e-01 -9.93299603e-01 -1.11092782e+00 -4.26983476e-01 -1.18911259e-01 1.88289776e-01 2.51681149e-01 2.96357721e-01 -5.82297146e-01 -4.75282431e-01 2.85855800e-01 -1.13796282e+00 -1.75762817e-01 2.99777657e-01 1.12810217e-01 -3.09691429e-01 9.04813528e-01 -4.40976441e-01 1.05873930e+00 -1.89090633e+00 9.30567905e-02 -1.46051124e-01 1.79248214e-01 4.43455696e-01 8.67270455e-02 1.47744983e-01 3.87911111e-01 -4.65889931e-01 3.86219025e-01 -2.55889565e-01 -7.53373131e-02 -1.25441536e-01 9.83503982e-02 8.12126338e-01 -2.88159817e-01 5.68653405e-01 -9.11381304e-01 -5.74190915e-01 6.64611638e-01 3.49487752e-01 -4.14407879e-01 -2.10430264e-01 -1.42087474e-01 5.82747996e-01 -4.04499263e-01 7.49666214e-01 6.86082840e-01 -8.85333214e-03 -2.54540175e-01 -1.81394801e-01 -8.12578082e-01 -1.26628935e-01 -1.48089552e+00 1.69196725e+00 1.70255508e-02 8.21448922e-01 -7.83262700e-02 -2.60552675e-01 8.52066815e-01 7.63769224e-02 7.58009374e-01 -5.81345797e-01 3.32188845e-01 2.66084224e-01 3.25608850e-01 -4.30242009e-02 7.71843195e-01 7.41245747e-02 -1.23614810e-01 4.10987958e-02 1.28942013e-01 1.50168985e-01 5.14772654e-01 9.89652574e-02 1.24374974e+00 4.64653671e-01 3.18988532e-01 -2.92835563e-01 4.63407218e-01 4.15046185e-01 5.95174372e-01 9.72398043e-01 -5.34545243e-01 4.26913381e-01 -2.22397432e-01 -2.94438899e-01 -8.57622981e-01 -1.05860615e+00 -3.55945438e-01 1.18738115e+00 6.83847070e-01 -6.18902504e-01 -3.61363828e-01 -4.91018772e-01 2.64040291e-01 3.75004023e-01 -6.06768966e-01 9.55357042e-04 -4.85085636e-01 -6.85147762e-01 5.25420845e-01 4.35460299e-01 3.26640368e-01 -7.19119668e-01 -1.03955615e+00 2.54080951e-01 -1.43450752e-01 -9.16484237e-01 -3.29650760e-01 2.18115374e-01 -6.09203041e-01 -1.24189496e+00 -8.11083138e-01 -2.32042283e-01 3.71630132e-01 5.41035831e-01 7.96982229e-01 7.16743618e-02 -2.04740137e-01 2.32011437e-01 -4.06195700e-01 -8.41717541e-01 -2.28401214e-01 -1.07725918e-01 3.28033954e-01 -4.97826397e-01 3.28907907e-01 9.06187296e-02 -5.77683449e-01 9.64853942e-01 -2.69226730e-01 2.89715063e-02 5.39880633e-01 5.86141527e-01 7.33707488e-01 -4.01414961e-01 1.10830583e-01 -3.30012470e-01 6.29255250e-02 -1.20927230e-01 -1.27074087e+00 1.05132006e-01 -3.75024706e-01 1.13142438e-01 5.39700352e-02 -5.85413218e-01 -7.64571905e-01 2.86868304e-01 -2.28707030e-01 -4.31259215e-01 3.56557332e-02 -1.26678556e-01 1.84345886e-01 -5.97567677e-01 8.73108625e-01 -2.41897643e-01 5.17173521e-02 -4.49982971e-01 1.16251282e-01 3.01208615e-01 4.38329607e-01 -2.70654380e-01 9.10285890e-01 6.26328528e-01 5.77449575e-02 -5.51272869e-01 -8.77903283e-01 -1.08327007e+00 -5.24821162e-01 -7.91965187e-01 9.44622219e-01 -1.01328921e+00 -9.83480573e-01 3.79546851e-01 -9.21699345e-01 -6.23390228e-02 -1.83392346e-01 9.03634548e-01 -1.94048971e-01 1.95312157e-01 -1.49398938e-01 -9.82079566e-01 -2.78737873e-01 -1.00082195e+00 1.12516034e+00 4.85228390e-01 -1.72534913e-01 -5.15652835e-01 5.86346507e-01 1.26795873e-01 3.94816846e-01 4.16392624e-01 -1.88513502e-01 -4.26765412e-01 -5.62047958e-01 -5.65823138e-01 -1.57686546e-01 -2.36713469e-01 -1.44951090e-01 -2.79030632e-02 -9.62866545e-01 -4.60424453e-01 -3.98792475e-01 1.97457597e-01 1.18461740e+00 6.97878420e-01 4.41177011e-01 4.99490172e-01 -8.37236106e-01 3.89698625e-01 1.31335342e+00 2.89568484e-01 3.32890660e-01 7.17437625e-01 4.23311591e-01 9.60923880e-02 1.04283988e+00 3.58403087e-01 2.21845269e-01 1.30616915e+00 5.87253273e-01 3.03957343e-01 -5.50144315e-01 8.14255476e-02 4.53962386e-01 3.63172591e-01 -4.82158870e-01 -2.23325059e-01 -7.83768892e-01 3.07303131e-01 -2.23507428e+00 -1.22528338e+00 -7.08703399e-01 2.37042952e+00 2.56106526e-01 4.20789123e-01 4.70974028e-01 -1.88997939e-01 9.21695352e-01 -3.95036191e-02 -4.88918513e-01 4.30498421e-01 -8.55037495e-02 -6.83858842e-02 1.07755995e+00 3.44804436e-01 -1.38995969e+00 8.01544249e-01 5.91651630e+00 7.51994610e-01 -7.67163157e-01 3.64278734e-01 -6.05010033e-01 -3.21279883e-01 3.67476970e-01 -9.66575146e-02 -1.68834543e+00 5.72402179e-01 9.00709391e-01 -9.13526565e-02 -1.45683154e-01 7.10464120e-01 2.21576765e-02 -5.43618321e-01 -5.99827290e-01 8.69920790e-01 -5.43802381e-02 -1.19337213e+00 -5.34531653e-01 1.90360934e-01 6.27808154e-01 5.41722715e-01 -3.71497601e-01 3.69653881e-01 5.73206663e-01 -3.45906854e-01 1.26164794e+00 5.91784835e-01 3.70459884e-01 -3.50294739e-01 8.44015121e-01 3.25938582e-01 -1.67026329e+00 8.20466578e-02 -3.82292747e-01 -3.38224173e-02 4.01408166e-01 5.36114812e-01 -5.30393422e-01 8.10262859e-01 1.04742658e+00 4.44504529e-01 -7.71890163e-01 2.00755334e+00 -1.23556674e-01 2.47409910e-01 -7.06338406e-01 -2.62855649e-01 -1.94367357e-02 1.35127783e-01 8.87459338e-01 1.10003662e+00 5.27408421e-01 -8.76414776e-02 2.67971486e-01 3.76534998e-01 3.23277682e-01 -2.38299549e-01 -3.64280611e-01 5.56962967e-01 4.55546379e-01 1.39266276e+00 -1.03278971e+00 -2.78996229e-01 -3.44582856e-01 3.40983361e-01 -2.11558258e-03 -1.30948171e-01 -1.31669700e+00 -2.95230478e-01 8.71513724e-01 9.45473015e-02 5.10571659e-01 -3.93012017e-01 -1.22107036e-01 -9.40486729e-01 1.12552028e-02 -3.06981564e-01 6.83895171e-01 -6.12302482e-01 -8.50706637e-01 5.31409383e-01 2.06149355e-01 -1.77856672e+00 -1.14679001e-02 -7.90053785e-01 -3.48574042e-01 6.69744909e-01 -1.13791049e+00 -1.06451619e+00 -6.08807027e-01 2.45419621e-01 3.36118311e-01 -5.50334454e-02 5.44245720e-01 6.52783990e-01 -3.19052011e-01 5.18968046e-01 1.49199694e-01 -2.32265294e-02 9.25714791e-01 -9.48520601e-01 4.52604145e-01 1.11565101e+00 2.84718186e-01 2.93644875e-01 1.04248333e+00 -8.68758619e-01 -1.20713007e+00 -8.60462368e-01 5.41472733e-01 -9.66978967e-01 6.13578141e-01 -1.30995438e-01 -6.04949057e-01 5.93895197e-01 -1.45682637e-02 4.39621061e-02 3.31700981e-01 2.06627667e-01 6.98850751e-02 6.22657500e-02 -9.26280141e-01 2.85108984e-01 1.30307221e+00 1.76142812e-01 -4.58518088e-01 2.15377912e-01 4.67736572e-01 -9.84599233e-01 -7.16602564e-01 6.95785761e-01 1.05637646e+00 -9.73817706e-01 1.09684944e+00 -1.04422323e-01 -5.94181538e-01 -1.06116891e+00 2.74899090e-03 -1.08169174e+00 -7.23678589e-01 -1.59780666e-01 -2.89112300e-01 7.64629126e-01 2.38869384e-01 -6.23917818e-01 7.75453448e-01 -1.02471868e-02 -5.28050959e-01 -1.32462904e-01 -1.17662644e+00 -1.19116795e+00 -4.83112663e-01 -4.33077216e-01 1.98909774e-01 4.10760581e-01 -5.50229490e-01 2.54588090e-02 -3.63294750e-01 3.58239770e-01 1.14844692e+00 1.20853767e-01 1.00820148e+00 -1.72649825e+00 -3.83228846e-02 -6.47924423e-01 -6.94049120e-01 -7.90481091e-01 -3.55964839e-01 -7.79258072e-01 2.52187401e-01 -1.52260280e+00 2.56930105e-02 -4.43224758e-01 -3.98977131e-01 4.82358515e-01 -2.10476786e-01 5.44535041e-01 3.85611504e-01 3.24251324e-01 -1.03504789e+00 2.57796586e-01 1.13190758e+00 2.57431380e-02 -8.39454383e-02 4.23463404e-01 -2.38781750e-01 5.71048021e-01 7.35821247e-01 -7.77294636e-01 3.13702971e-01 -2.67309129e-01 4.32687104e-02 -1.69256017e-01 6.46570802e-01 -1.73171008e+00 6.72009647e-01 1.52468726e-01 6.03843570e-01 -1.26622057e+00 4.92059857e-01 -7.24789798e-01 7.43156433e-01 9.09222782e-01 1.81644663e-01 2.09281310e-01 6.41003728e-01 4.79548693e-01 9.89315882e-02 -2.49620631e-01 8.44428480e-01 2.66580023e-02 -1.14357817e+00 5.05567864e-02 -3.21766764e-01 -1.86302468e-01 1.24467850e+00 -4.51578081e-01 -5.21799624e-01 1.52835533e-01 -6.59207463e-01 3.97513568e-01 5.75597227e-01 5.78618824e-01 2.55340368e-01 -1.54769242e+00 -7.35242784e-01 1.48651481e-01 3.31719607e-01 -3.33557904e-01 1.36255100e-01 1.25488031e+00 -3.33293796e-01 4.84178454e-01 -4.31721836e-01 -1.07203007e+00 -1.82026136e+00 3.59531082e-02 3.50208849e-01 -1.18167795e-01 -6.34392977e-01 8.25880289e-01 -1.88099295e-01 -2.85376847e-01 4.44668323e-01 -2.72623390e-01 -2.31702611e-01 2.85966635e-01 6.40381217e-01 6.50547624e-01 2.35881701e-01 -6.23507738e-01 -7.01398849e-01 9.57492650e-01 9.25954059e-02 -1.52252451e-01 9.83958125e-01 4.89135599e-03 3.77071947e-01 3.10360789e-01 3.91092956e-01 1.14078909e-01 -1.30799198e+00 -8.18905595e-04 7.22289160e-02 -5.49448192e-01 1.53806806e-01 -8.14933479e-01 -7.61752903e-01 1.04001068e-01 1.14418066e+00 1.33106530e-01 6.18232310e-01 2.09095418e-01 4.26259071e-01 2.52462029e-01 8.65978539e-01 -7.94575870e-01 -1.71954170e-01 7.62983918e-01 5.52544534e-01 -1.51344323e+00 3.67606282e-01 -6.26984239e-02 -2.68532068e-01 8.17310154e-01 7.48693466e-01 -2.26679947e-02 4.39243168e-01 4.25439537e-01 1.98081478e-01 -3.12083542e-01 -4.59803104e-01 -8.34353745e-01 6.24388993e-01 4.78119403e-01 3.32780272e-01 1.29782772e-02 -3.96004677e-01 1.98967144e-01 -2.21097812e-01 -8.25504139e-02 4.18939888e-02 1.01106882e+00 -1.01003790e+00 -9.83246744e-01 -8.86104405e-01 4.34452176e-01 -2.11066291e-01 3.70673031e-01 -1.07177280e-01 1.05207634e+00 2.88171589e-01 7.92274058e-01 -1.03588440e-01 -6.39739692e-01 6.66968644e-01 -1.19902208e-01 7.99370348e-01 -4.15307701e-01 -7.12307155e-01 1.27648070e-01 1.54984578e-01 -5.53126574e-01 -7.66503274e-01 -1.13692772e+00 -1.25568950e+00 -3.42674583e-01 -9.70576167e-01 1.37054190e-01 7.84397483e-01 6.24562025e-01 1.40034303e-01 7.68463433e-01 -3.34623680e-02 -1.13917184e+00 -1.97800547e-01 -9.89894688e-01 -2.02416137e-01 4.33966666e-02 3.30899388e-01 -1.32279623e+00 -2.02407390e-01 -2.84825414e-01]

[6.586888313293457, -2.093747854232788]

1180af78-7231-4203-be33-af27ce9b6e91

ontology-matching-through-absolute

2204.0404

https://arxiv.org/abs/2204.04040v1

https://arxiv.org/pdf/2204.04040v1.pdf

Ontology Matching Through Absolute Orientation of Embedding Spaces

Ontology matching is a core task when creating interoperable and linked open datasets. In this paper, we explore a novel structure-based mapping approach which is based on knowledge graph embeddings: The ontologies to be matched are embedded, and an approach known as absolute orientation is used to align the two embedding spaces. Next to the approach, the paper presents a first, preliminary evaluation using synthetic and real-world datasets. We find in experiments with synthetic data, that the approach works very well on similarly structured graphs; it handles alignment noise better than size and structural differences in the ontologies.

['Heiko Paulheim', 'Michael Hladik', 'Katharina Kreplin', 'Karolin Stefani', 'Guilherme Costa', 'Jan Portisch']

2022-04-08

['knowledge-graph-embeddings', 'ontology-matching', 'knowledge-graph-embeddings']

['graphs', 'knowledge-base', 'methodology']

[-2.01789383e-02 6.00271106e-01 -6.04527406e-02 -2.90325940e-01 -1.25727654e-01 -5.74785352e-01 6.53035760e-01 6.33301675e-01 -4.38451231e-01 5.71679235e-01 4.93804544e-01 -2.24068433e-01 -7.72402823e-01 -1.03227007e+00 -4.32842702e-01 7.44460523e-02 -4.41811740e-01 8.30342650e-01 6.19761527e-01 -5.76478541e-01 9.61255208e-02 3.98941785e-01 -1.75286758e+00 2.78019309e-02 7.41338253e-01 5.93044162e-01 -2.41472706e-01 3.09499145e-01 -6.92975819e-01 7.84483016e-01 -2.68971682e-01 -7.52735078e-01 3.69139075e-01 -1.81728214e-01 -1.33314991e+00 -3.77796054e-01 5.86423874e-01 4.90808606e-01 -3.34975243e-01 1.36231148e+00 4.99753088e-01 -6.04042187e-02 3.52176011e-01 -1.58993673e+00 -8.85006845e-01 5.48652351e-01 1.48551151e-01 1.81938440e-01 1.05158854e+00 -4.98651624e-01 1.16090083e+00 -6.78357244e-01 1.08572388e+00 1.30206013e+00 1.09282982e+00 1.35414973e-01 -1.21445179e+00 -1.99787498e-01 -2.30178937e-01 5.23444235e-01 -1.54494727e+00 -1.94659367e-01 6.44808650e-01 -5.14792323e-01 1.05573058e+00 3.99545342e-01 6.86981857e-01 6.69702709e-01 1.91487148e-01 -1.61101937e-01 9.34365630e-01 -8.28645647e-01 2.33244047e-01 4.74986196e-01 2.99093187e-01 6.89008355e-01 7.38635242e-01 1.39208743e-02 -2.15653062e-01 -5.41230142e-01 1.75388649e-01 -3.47377896e-01 -2.18897507e-01 -1.06048286e+00 -1.22759736e+00 7.49883056e-01 6.05010092e-01 8.59138429e-01 -2.77588278e-01 -6.08396865e-02 4.02851313e-01 7.07630038e-01 2.97860384e-01 6.14043593e-01 -5.75455278e-02 4.27475460e-02 -4.31357116e-01 2.54107416e-01 1.27751064e+00 1.14997244e+00 9.22524333e-01 -2.91069090e-01 3.99195880e-01 6.68515563e-01 5.21775663e-01 -4.78056148e-02 6.22983992e-01 -6.05258405e-01 4.38898593e-01 1.15637624e+00 3.30016203e-02 -1.52157581e+00 -5.28192699e-01 1.59282520e-01 -1.63690615e-02 1.81414187e-01 1.68531358e-01 3.36383700e-01 -5.78045487e-01 1.56236768e+00 4.88918096e-01 2.01264292e-01 4.79205698e-01 5.27295649e-01 1.13899326e+00 1.90118358e-01 7.80815110e-02 2.14650571e-01 1.52205145e+00 -6.39114141e-01 -1.01841152e+00 7.25441799e-02 7.28805006e-01 -7.39118576e-01 7.74116635e-01 -2.04237640e-01 -8.99594545e-01 -2.78683156e-01 -1.39080215e+00 -1.42439052e-01 -1.31516182e+00 -7.49542654e-01 3.99192750e-01 7.84054279e-01 -1.38592947e+00 7.35683203e-01 -2.98620909e-01 -1.18918741e+00 -4.10051011e-02 3.23588550e-01 -9.91321743e-01 1.71911512e-02 -1.51644850e+00 1.42107630e+00 9.98279393e-01 -3.67957085e-01 1.27786979e-01 -5.24717152e-01 -1.16104817e+00 -2.14982443e-02 2.74217188e-01 -7.07417309e-01 5.61993659e-01 -7.03032732e-01 -1.02008808e+00 1.21879709e+00 3.34742725e-01 -5.70456922e-01 3.24216783e-01 1.09106362e-01 -1.12595415e+00 -7.91033879e-02 1.31460717e-02 3.62855494e-01 1.93437800e-01 -1.32409310e+00 -5.58974445e-01 -5.75470626e-01 2.37210765e-01 -1.19022168e-02 -8.28351617e-01 3.05737317e-01 -2.21573398e-01 -4.40269589e-01 1.05072990e-01 -9.01002526e-01 -5.76172285e-02 2.99695600e-02 8.17226544e-02 -2.63711035e-01 7.46120036e-01 -7.38335431e-01 1.38872361e+00 -1.87220407e+00 3.75636011e-01 6.83437824e-01 2.94141650e-01 1.19592324e-01 -1.83525354e-01 1.09611213e+00 -3.21801543e-01 2.22918406e-01 -3.43515068e-01 1.87940925e-01 4.90709692e-01 4.95945990e-01 1.61582217e-01 3.30526680e-01 -1.19146273e-01 8.99337769e-01 -9.58346605e-01 -7.36313760e-01 9.81690660e-02 3.93427730e-01 -2.92187780e-01 5.25476597e-02 1.46176308e-01 -2.83647865e-01 -1.32306606e-01 4.56485480e-01 4.27726686e-01 -1.04243346e-02 7.08305836e-01 -4.89571899e-01 -3.26305740e-02 -4.33742180e-02 -1.56001163e+00 1.78610075e+00 -3.48960370e-01 6.28017306e-01 -2.48327196e-01 -9.16220725e-01 1.16821110e+00 5.16005039e-01 6.05406880e-01 -7.86249220e-01 -2.27340460e-02 4.72068340e-01 -8.60761628e-02 -7.37678051e-01 6.78366423e-01 2.07491726e-01 -1.32033061e-02 1.75662786e-01 3.31433803e-01 4.76317070e-02 4.59059358e-01 1.21679917e-01 1.13990450e+00 1.21654183e-01 7.67625093e-01 -6.76598728e-01 7.18748093e-01 1.04121506e-01 3.49427462e-01 7.77890831e-02 -8.58353898e-02 1.44402698e-01 3.30870450e-01 -8.67911100e-01 -1.19775403e+00 -1.00104415e+00 -2.34861895e-01 6.20582938e-01 3.57564598e-01 -8.23775113e-01 -7.54133403e-01 -6.83282256e-01 3.52305502e-01 5.59904516e-01 -7.46992290e-01 -2.36642107e-01 -5.06952643e-01 -3.28067243e-01 6.26877725e-01 4.35363889e-01 -5.24719022e-02 -8.57707739e-01 -4.66430455e-01 4.05173212e-01 7.54299313e-02 -1.14903605e+00 -7.01584145e-02 -1.67630419e-01 -6.95158601e-01 -1.45858204e+00 -6.32373393e-02 -1.09003794e+00 4.47150320e-01 -1.50429308e-01 1.35196376e+00 1.70839876e-01 -4.74460870e-01 5.08620799e-01 -5.13061583e-01 -4.08299327e-01 -6.19924188e-01 1.11899083e-03 7.28637129e-02 2.60240454e-02 7.10897803e-01 -7.97146201e-01 1.51152583e-02 2.82828093e-01 -1.18871486e+00 -4.79747653e-01 2.06524864e-01 5.95665574e-01 4.27379340e-01 6.06321134e-02 2.48086229e-01 -8.66928101e-01 9.25715387e-01 -5.31015754e-01 -7.01416254e-01 6.35330796e-01 -9.49409544e-01 2.50742882e-01 3.07281077e-01 2.04419047e-02 -4.30930167e-01 -4.33721058e-02 -6.05349652e-02 -2.72363335e-01 2.49454807e-02 6.56151354e-01 -3.54914963e-01 -6.95798934e-01 7.34582365e-01 -4.27464247e-01 2.25855365e-01 -6.32861495e-01 5.24721622e-01 8.53775561e-01 3.08569998e-01 -3.28049362e-01 9.18213010e-01 4.23555940e-01 -1.93441119e-02 -7.07248807e-01 -4.25292477e-02 -3.68426353e-01 -9.05621350e-01 5.00238761e-02 7.42412567e-01 -5.61591744e-01 -3.10706407e-01 -1.29250661e-01 -9.78950500e-01 1.89541772e-01 -6.02022529e-01 4.82580841e-01 -5.22817671e-01 3.76422107e-01 -2.32502213e-03 -2.43185088e-01 -1.54022858e-01 -8.17750216e-01 5.66722631e-01 -2.20638700e-02 -4.32705373e-01 -1.50369465e+00 8.46588671e-01 5.80790825e-02 4.25905108e-01 5.03096342e-01 1.13762748e+00 -1.21863687e+00 -2.19062954e-01 -5.31602025e-01 -1.53520852e-01 7.44137838e-02 3.18550110e-01 1.07299522e-01 -6.87743485e-01 -4.19474185e-01 -5.72279692e-01 1.84638470e-01 1.25595361e-01 -5.91583908e-01 5.43805242e-01 -1.66500121e-01 -5.07555783e-01 4.71751243e-01 1.84239876e+00 9.61863101e-02 9.58076119e-01 8.63323689e-01 7.09394276e-01 8.31767678e-01 4.91345078e-01 7.46285841e-02 7.40664959e-01 1.08074999e+00 5.31510353e-01 -3.21440622e-02 -2.43944854e-01 -2.77996421e-01 -6.48045018e-02 1.12626386e+00 -4.02827598e-02 -2.16577634e-01 -1.21793139e+00 7.18848169e-01 -2.01248407e+00 -1.07299721e+00 -1.27018481e-01 2.26612949e+00 4.05629098e-01 -1.14262991e-01 9.93110240e-02 2.08175763e-01 7.34513521e-01 4.95142937e-02 2.41399869e-01 -6.15301073e-01 -2.64646292e-01 3.65025908e-01 6.55454993e-01 7.53316164e-01 -9.21564937e-01 6.39393687e-01 7.09636450e+00 8.75459537e-02 -6.19799197e-01 3.81695509e-01 -6.05130911e-01 4.85989153e-01 -7.03099489e-01 2.84633577e-01 -3.38091135e-01 5.35336912e-01 1.01558185e+00 -5.33862174e-01 3.83367836e-01 5.44050515e-01 -5.13849974e-01 4.80752230e-01 -1.10052967e+00 8.74317288e-01 2.30322897e-01 -1.35262382e+00 2.21356690e-01 1.50527894e-01 4.12623286e-01 -5.51507948e-03 -5.90993941e-01 -5.18354960e-02 4.50289309e-01 -8.82408679e-01 5.76970339e-01 6.56270862e-01 4.74843770e-01 -7.55345106e-01 1.14105618e+00 -4.09785509e-01 -1.53927243e+00 -1.83125839e-01 -4.61352110e-01 2.28968799e-01 8.36603343e-02 1.27523810e-01 -5.56735635e-01 1.35097039e+00 8.44295025e-01 5.56626022e-01 -9.18521166e-01 1.27782094e+00 -9.84490439e-02 -1.60984024e-01 -1.57414019e-01 6.12693056e-02 -1.08535506e-03 -4.89551097e-01 6.11301839e-01 1.10867262e+00 4.37973589e-01 -4.40415561e-01 1.17394067e-01 5.19605696e-01 5.54797612e-02 5.73993266e-01 -1.18400204e+00 -1.86057612e-01 6.76423967e-01 1.07420111e+00 -4.55669999e-01 -2.25676343e-01 -7.70445943e-01 8.21643293e-01 4.86206353e-01 4.85130474e-02 -7.43237376e-01 -8.69173408e-01 7.98983812e-01 7.46460110e-02 2.71772772e-01 2.08706222e-02 3.41076493e-01 -9.83383536e-01 3.29609275e-01 -9.74450707e-01 8.58764410e-01 -5.99291980e-01 -1.26381016e+00 9.57823932e-01 5.68760708e-02 -1.26979041e+00 -1.26025140e-01 -6.93612516e-01 -3.36340487e-01 7.76240706e-01 -1.47936428e+00 -1.10230947e+00 -4.55283195e-01 4.87764746e-01 -2.64667660e-01 -3.04150462e-01 1.27387822e+00 7.47177184e-01 -2.66206801e-01 5.48645377e-01 2.24437371e-01 7.14514777e-02 6.06877565e-01 -1.37135339e+00 6.26788437e-01 6.54336989e-01 5.77224672e-01 6.56843424e-01 8.32804799e-01 -5.81171811e-01 -1.48345971e+00 -9.15240169e-01 1.29328847e+00 -4.32502568e-01 1.05723011e+00 -1.99047580e-01 -1.07192469e+00 9.08559859e-01 6.37147188e-01 4.01837170e-01 9.62854207e-01 1.23441376e-01 -6.10071540e-01 -1.99629933e-01 -1.43396735e+00 3.54741633e-01 1.41308546e+00 -6.72441661e-01 -1.20552170e+00 3.95892650e-01 7.69940078e-01 -2.33270362e-01 -1.65947092e+00 5.86768508e-01 5.91469049e-01 -8.25568616e-01 1.07253778e+00 -1.01325917e+00 -2.79405713e-01 -6.33557677e-01 -4.47992206e-01 -1.41596913e+00 -3.80512953e-01 -5.33504307e-01 1.30467499e-02 1.23912895e+00 4.04826730e-01 -1.19089425e+00 4.40979600e-01 3.44216228e-01 1.28490120e-01 -2.07282871e-01 -1.08630395e+00 -1.28217268e+00 -1.97441190e-01 9.37401503e-02 1.26063800e+00 1.52650368e+00 3.71608526e-01 -2.29571182e-02 5.19957803e-02 2.93270200e-01 4.72056568e-01 -5.93307018e-02 7.96239018e-01 -1.82508957e+00 1.98571980e-01 -3.37192774e-01 -1.57750356e+00 2.27955058e-01 1.66069075e-01 -1.08109832e+00 -5.72018385e-01 -1.84635699e+00 -2.96455264e-01 -4.18340951e-01 -6.06671035e-01 3.62985343e-01 1.18842408e-01 1.41813874e-01 1.38945773e-01 -5.78063242e-02 -4.72888410e-01 2.90447503e-01 5.19395113e-01 -2.10183144e-01 6.97105527e-02 -8.38069081e-01 -4.31418806e-01 3.99485618e-01 5.61852992e-01 -6.29478216e-01 -4.77974683e-01 -3.79595518e-01 4.93364215e-01 -4.52175379e-01 3.28162134e-01 -1.24201620e+00 3.30349028e-01 1.54866815e-01 -2.94209987e-01 -9.35027078e-02 5.73169217e-02 -1.38105083e+00 1.00932539e+00 4.32863563e-01 -1.86708823e-01 6.56928897e-01 1.72360808e-01 5.06310046e-01 -4.13265765e-01 -4.66957450e-01 5.31729102e-01 8.92719720e-03 -1.17180860e+00 1.58083528e-01 2.56386042e-01 6.29976764e-02 1.41803992e+00 -4.99281466e-01 -2.23431692e-01 6.68263063e-02 -8.45516145e-01 1.97518796e-01 1.06133831e+00 9.66688693e-01 3.13750714e-01 -1.83828723e+00 -5.12367725e-01 6.41217157e-02 7.89137065e-01 -7.14994371e-01 -2.60448903e-01 5.11426568e-01 -8.90298247e-01 3.81489694e-01 -6.56072438e-01 -2.18689486e-01 -1.42176139e+00 1.00574505e+00 4.49196994e-01 4.73807100e-03 -6.19113326e-01 2.62043834e-01 -7.41406024e-01 -1.05352545e+00 1.42706335e-01 6.16577780e-03 -5.38234591e-01 3.50444138e-01 4.39409584e-01 5.54501474e-01 5.23490071e-01 -1.10467553e+00 -7.01206744e-01 8.58902633e-01 3.06697786e-01 -5.23332786e-03 1.25049460e+00 -1.27482498e-02 -5.80328763e-01 5.08020401e-01 1.37692654e+00 1.58540919e-01 -1.55041218e-01 -3.63666445e-01 7.63477564e-01 -7.76522994e-01 -2.96918988e-01 -2.35761374e-01 -8.06465864e-01 3.30481201e-01 1.00864279e+00 8.54678333e-01 6.27868414e-01 -4.41435911e-02 4.06924605e-01 3.94872278e-01 5.72149515e-01 -1.17493176e+00 -4.75539714e-01 2.01287493e-01 8.52079988e-01 -9.60308611e-01 8.72607529e-02 -5.40027797e-01 -1.09287575e-01 1.23148215e+00 4.01016563e-01 -1.89866200e-02 6.71301246e-01 7.60451704e-02 3.31405997e-01 -7.08987951e-01 -2.92459249e-01 -4.59598422e-01 2.65215039e-01 7.64744937e-01 3.68633032e-01 -8.94187167e-02 -7.03446150e-01 4.74780016e-02 -2.56061256e-01 8.95450823e-03 4.22705740e-01 1.28776550e+00 -3.08170766e-01 -1.52297139e+00 -3.34019840e-01 2.19408095e-01 -8.66108090e-02 1.46502599e-01 -6.81508303e-01 1.24103403e+00 2.02848241e-01 6.40791118e-01 2.03689933e-01 -4.36331630e-01 9.37810004e-01 2.94518113e-01 5.19247115e-01 -3.87240916e-01 -5.34501970e-01 -9.36366022e-01 5.35818398e-01 -9.58689094e-01 -6.91753387e-01 -4.37660545e-01 -1.09534633e+00 -2.56269753e-01 -3.94743741e-01 4.18501109e-01 7.33144999e-01 5.78276277e-01 6.39351606e-01 4.44537789e-01 3.76286596e-01 -2.18262300e-01 -3.41184348e-01 -5.61772645e-01 -5.84564328e-01 9.03432667e-01 -1.43871143e-01 -9.07115817e-01 -1.83664039e-01 -1.54814392e-01]

[9.210307121276855, 8.093515396118164]

5700c4bd-f637-40e1-8493-e68f5c0b3208

show-attend-and-interact-perceivable-human

1702.08626

http://arxiv.org/abs/1702.08626v1

http://arxiv.org/pdf/1702.08626v1.pdf

Show, Attend and Interact: Perceivable Human-Robot Social Interaction through Neural Attention Q-Network

For a safe, natural and effective human-robot social interaction, it is essential to develop a system that allows a robot to demonstrate the perceivable responsive behaviors to complex human behaviors. We introduce the Multimodal Deep Attention Recurrent Q-Network using which the robot exhibits human-like social interaction skills after 14 days of interacting with people in an uncontrolled real world. Each and every day during the 14 days, the system gathered robot interaction experiences with people through a hit-and-trial method and then trained the MDARQN on these experiences using end-to-end reinforcement learning approach. The results of interaction based learning indicate that the robot has learned to respond to complex human behaviors in a perceivable and socially acceptable manner.

['Hiroshi Ishiguro', 'Ahmed Hussain Qureshi', 'Yutaka Nakamura', 'Yuichiro Yoshikawa']

2017-02-28

['deep-attention', 'deep-attention']

['computer-vision', 'natural-language-processing']

[-3.88963044e-01 6.81385159e-01 5.38178027e-01 -3.28046620e-01 -6.04562089e-03 1.61546282e-02 3.80468607e-01 -2.85820752e-01 -6.28163159e-01 1.00260627e+00 2.66050629e-04 4.86230105e-01 1.22161798e-01 -3.88837785e-01 -4.68082160e-01 -4.80803192e-01 -5.70399880e-01 7.93147087e-01 1.32164173e-02 -8.02979350e-01 4.15954329e-02 2.93814778e-01 -1.69117820e+00 -2.38721460e-01 6.93293869e-01 3.05306435e-01 3.59367788e-01 1.07328701e+00 6.44213319e-01 1.42760766e+00 -3.33165795e-01 -8.22921544e-02 1.23496920e-01 -7.13269770e-01 -8.94567072e-01 1.22093566e-01 -6.20701432e-01 -7.27969885e-01 -3.95460427e-01 6.93159342e-01 3.90149742e-01 8.24549437e-01 8.41276169e-01 -1.76447618e+00 -5.18512309e-01 4.44977850e-01 -1.71820760e-01 -4.47190166e-01 1.09894300e+00 7.31481850e-01 5.01801610e-01 -5.24018884e-01 6.45058095e-01 1.71310318e+00 3.16206574e-01 1.16573381e+00 -7.79763997e-01 -5.32142401e-01 -2.85834759e-01 1.65387824e-01 -8.06187212e-01 -1.92231268e-01 3.44559610e-01 -2.59589612e-01 1.38349319e+00 -4.55129743e-01 1.02358699e+00 1.39272141e+00 6.20759964e-01 7.34292686e-01 5.75063288e-01 -2.76128352e-01 3.81628782e-01 2.09993869e-01 4.65224534e-02 7.72279501e-01 -8.88997093e-02 4.83278513e-01 -3.18955362e-01 8.11185464e-02 7.92676926e-01 3.30172241e-01 2.96343327e-01 -4.42339867e-01 -1.36956930e+00 7.88607955e-01 7.62236893e-01 1.51278704e-01 -9.90117550e-01 5.91661096e-01 4.87280965e-01 6.45383775e-01 -2.76308507e-01 5.93638539e-01 -3.77967507e-01 -5.74420631e-01 5.44917703e-01 1.72620386e-01 8.83520544e-01 8.60876858e-01 7.42444158e-01 5.38990349e-02 1.11380415e-02 6.80940509e-01 4.57613587e-01 8.99035752e-01 2.73925573e-01 -1.60881507e+00 -1.77851409e-01 7.44807720e-01 5.73915482e-01 -9.57524121e-01 -7.20232189e-01 5.88481843e-01 -6.55938327e-01 6.58139110e-01 3.61921079e-02 -7.61592269e-01 -4.78601366e-01 1.73287964e+00 4.07417983e-01 -2.19818458e-01 8.58045399e-01 9.62076783e-01 6.90654695e-01 7.86960959e-01 4.96093482e-01 -1.04541175e-01 1.14810956e+00 -9.38939035e-01 -8.85653019e-01 -1.29365653e-01 6.52747393e-01 7.06319138e-02 1.24743903e+00 4.59146202e-01 -8.09787750e-01 -4.99903679e-01 -8.66868436e-01 2.42126226e-01 -1.14853635e-01 -2.59036839e-01 6.02727711e-01 -4.68444824e-03 -1.20035517e+00 4.30539966e-01 -7.57246733e-01 -1.42067814e+00 -8.21743682e-02 7.37321675e-01 -6.94797575e-01 2.20605046e-01 -1.13204217e+00 1.21014178e+00 3.52677137e-01 2.50738356e-02 -1.55316174e+00 5.28794646e-01 -1.10733080e+00 -7.83239529e-02 2.34116182e-01 -5.97844362e-01 1.56587863e+00 -1.16305935e+00 -2.18500113e+00 6.53366387e-01 4.79306042e-01 -4.37069714e-01 1.28845409e-01 -4.28967506e-01 -2.16794595e-01 5.14630556e-01 1.54531136e-01 1.26107895e+00 5.35678804e-01 -1.49469554e+00 -6.17995977e-01 -1.57570511e-01 9.36710685e-02 7.86948204e-01 -1.77961767e-01 1.55455038e-01 1.86432041e-02 7.30061755e-02 -6.18561029e-01 -1.47657001e+00 -4.69001293e-01 -2.77570426e-01 -3.95500287e-02 -6.57335818e-01 7.70554006e-01 -8.90730768e-02 -3.58332396e-02 -2.28090835e+00 3.82516325e-01 9.41167846e-02 1.38120636e-01 4.28323541e-03 -3.65511417e-01 6.95219994e-01 4.13111508e-01 -5.22002518e-01 1.84260860e-01 -4.07661468e-01 2.34963983e-01 2.15495169e-01 2.02220142e-01 3.87843609e-01 3.61647122e-02 8.48018467e-01 -1.39852107e+00 -4.59236473e-01 3.78577501e-01 4.31096852e-01 -5.04907131e-01 1.33640409e+00 2.68156156e-02 9.57861304e-01 -5.80401182e-01 2.99475640e-01 -1.00400910e-01 6.18373267e-02 1.36398179e-02 7.26231992e-01 2.76237607e-01 -4.71705288e-01 -3.70595127e-01 1.33547592e+00 -2.56529719e-01 6.06531262e-01 1.23228595e-01 -4.39722985e-01 8.71326387e-01 5.48000455e-01 5.30146837e-01 -7.87051022e-01 6.46206141e-01 -6.42374754e-02 -1.10686058e-02 -1.17812729e+00 3.76945436e-01 -7.64946342e-02 -6.39225662e-01 7.38128960e-01 1.72055826e-01 -2.70980954e-01 -2.47925431e-01 4.32368726e-01 1.39611602e+00 1.04066744e-01 2.48080939e-01 1.34870693e-01 2.25162342e-01 2.13352088e-02 2.60541886e-01 7.32718706e-01 -7.94469833e-01 -4.18941826e-02 1.42055988e-01 -4.72368687e-01 -8.33089352e-01 -1.07746994e+00 8.98304343e-01 1.54182446e+00 4.60656881e-01 5.19681931e-01 -8.91148627e-01 -4.64907289e-01 -2.96861410e-01 9.86962199e-01 -7.36165106e-01 -7.20917642e-01 -4.63560551e-01 1.40817866e-01 3.11251163e-01 3.18605214e-01 8.38891208e-01 -2.66910219e+00 -1.37329996e+00 1.98681980e-01 2.42812634e-02 -8.34077895e-01 -1.94592103e-01 2.57596493e-01 -3.41837883e-01 -8.43475461e-01 -5.83162904e-01 -1.23438275e+00 7.79948175e-01 2.10708454e-01 7.36635625e-01 1.25360042e-01 -5.70446029e-02 1.12168217e+00 -8.42669487e-01 -2.52690226e-01 -4.09057349e-01 -2.13279545e-01 6.75184250e-01 -3.93563002e-01 5.82235575e-01 -5.13731956e-01 -6.44345343e-01 2.25465015e-01 -4.37861949e-01 -2.70251036e-01 5.83323359e-01 6.78043604e-01 -4.77961600e-01 -1.09440878e-01 8.41318965e-01 -1.27018631e-01 1.19446337e+00 -7.42547393e-01 -1.55536085e-01 2.84173880e-02 -1.43506490e-02 -1.85848296e-01 4.33398038e-01 -6.55187607e-01 -1.02025390e+00 4.61701810e-01 3.11076075e-01 -9.86462459e-02 -5.66088498e-01 9.56099033e-02 2.22167242e-02 2.17892617e-01 6.62522078e-01 -5.96710630e-02 5.50826132e-01 4.73788321e-01 3.60998809e-01 9.73895848e-01 6.02413416e-01 -3.73542547e-01 4.81124043e-01 2.13623643e-02 -4.06156301e-01 -9.42690074e-01 1.00432694e-01 -6.02619499e-02 -4.32325095e-01 -9.61006880e-01 1.44026756e+00 -8.41411114e-01 -2.03480268e+00 7.83171117e-01 -1.21013749e+00 -1.14811313e+00 -1.35914773e-01 5.42524457e-01 -1.05117071e+00 1.84696317e-02 -7.55961776e-01 -1.37210548e+00 -3.25526029e-01 -9.94599760e-01 7.87164092e-01 5.75346828e-01 -7.07336247e-01 -5.35075784e-01 4.02134389e-01 1.83025748e-01 3.76996279e-01 1.02979466e-01 4.08889621e-01 -5.62724411e-01 -2.64136106e-01 -9.76469070e-02 1.09920716e-02 5.29304938e-03 1.47687390e-01 -2.33920053e-01 -4.76454645e-01 -4.72257644e-01 -1.36995032e-01 -1.30479383e+00 -6.12199605e-02 7.30681792e-02 6.25570342e-02 -3.58902395e-01 -1.77618027e-01 -3.55451882e-01 8.39889228e-01 8.34562361e-01 8.23729038e-01 1.97822675e-01 3.15538526e-01 1.17324102e+00 9.14700150e-01 4.87838626e-01 8.55997443e-01 3.22209954e-01 6.45115912e-01 -4.01700214e-02 5.87867439e-01 -2.58429110e-01 1.03774726e+00 3.45799893e-01 5.46223298e-02 -3.90045941e-01 -8.44422698e-01 5.57565451e-01 -2.37169218e+00 -9.75984871e-01 2.27975354e-01 1.78592563e+00 4.93278086e-01 -9.00368989e-02 4.87544656e-01 -4.01733190e-01 6.45473838e-01 -8.12990427e-01 -7.72378385e-01 -7.90078580e-01 3.67148101e-01 -6.12662256e-01 6.34957105e-02 5.35661697e-01 -7.01940119e-01 1.40305209e+00 6.64804888e+00 -2.47845218e-01 -6.44869685e-01 -6.80100843e-02 5.40441632e-01 1.33031905e-01 4.29525763e-01 -4.21987325e-01 -1.19970977e-01 6.36900440e-02 1.17894924e+00 1.46097422e-01 9.53521490e-01 1.02246046e+00 5.06882131e-01 -6.02954388e-01 -1.11643684e+00 9.29683626e-01 7.46254176e-02 -1.10940114e-01 -5.69435775e-01 -1.38391167e-01 3.96344393e-01 -7.52729923e-02 -6.28456250e-02 9.80256915e-01 1.26523709e+00 -1.30804682e+00 3.64235103e-01 9.56333399e-01 3.73805255e-01 -9.20217037e-01 9.62929726e-01 6.05445921e-01 -7.84270465e-01 -5.85763514e-01 -1.70866668e-01 -5.39031446e-01 3.78428906e-01 -5.77444851e-01 -1.19959128e+00 -3.09970796e-01 9.95016873e-01 4.84195769e-01 -1.60487264e-01 3.77756447e-01 -1.30567372e-01 6.75209565e-04 3.50116119e-02 -9.75330174e-01 2.58624434e-01 -4.45320964e-01 4.68208522e-01 6.37820542e-01 2.55055338e-01 7.14276850e-01 2.92481244e-01 4.61069405e-01 2.20352754e-01 -1.86347201e-01 -1.44823503e+00 -7.19494596e-02 2.45872557e-01 1.04843748e+00 -7.36045957e-01 -2.15290725e-01 5.78501374e-02 1.48297369e+00 6.72346532e-01 4.66535479e-01 -7.31754482e-01 -6.03100121e-01 3.42539370e-01 -4.80893284e-01 3.38438270e-03 -2.65646577e-01 2.29923040e-01 -5.89673519e-01 -4.22179461e-01 -9.28149700e-01 6.26524389e-02 -1.31753957e+00 -1.29792345e+00 9.25752103e-01 6.21976005e-03 -1.12419510e+00 -7.62604177e-01 -1.55380771e-01 -5.47356129e-01 2.32295021e-02 -4.30120736e-01 -9.49754059e-01 -5.28729916e-01 7.95781195e-01 4.31083858e-01 -7.49031723e-01 9.19254065e-01 -3.05648595e-01 -4.63756621e-01 2.00246781e-01 -5.29436111e-01 -6.77915066e-02 7.04497516e-01 -1.07950461e+00 -2.73784161e-01 9.22499523e-02 -7.89290428e-01 5.47535300e-01 1.17382789e+00 -8.33369851e-01 -1.56069148e+00 -7.94242382e-01 4.34414387e-01 -4.27270591e-01 3.02743316e-01 -3.25083166e-01 -5.76623678e-01 8.66339982e-01 1.03798461e+00 -5.24991512e-01 4.73005056e-01 -3.36026549e-01 4.94391978e-01 -1.06369711e-01 -1.62941504e+00 1.11825442e+00 8.21935177e-01 -3.02826464e-01 -8.96373510e-01 2.45311245e-01 1.04810596e+00 2.68475622e-01 -3.15710813e-01 1.38156191e-01 4.32883024e-01 -7.38780141e-01 4.53654259e-01 -6.03636682e-01 3.79301310e-01 -2.12996811e-01 -7.84455240e-02 -1.46768999e+00 -1.45730928e-01 -9.76930797e-01 2.95076251e-01 8.02822769e-01 1.93513349e-01 -7.23959923e-01 5.34176946e-01 1.02175570e+00 1.07109748e-01 -1.81544110e-01 -5.67495763e-01 -4.64259624e-01 -2.00483158e-01 1.00504868e-01 2.15960652e-01 4.70958829e-01 9.97166693e-01 6.48856819e-01 -7.03946531e-01 -6.40244782e-02 4.12890911e-01 -8.01750839e-01 1.34075105e+00 -1.09077835e+00 1.15954503e-01 1.48952613e-02 -3.29526156e-01 -7.76438177e-01 5.03365695e-01 -3.05504382e-01 1.13199711e+00 -1.56230927e+00 4.81678903e-01 4.89446670e-02 -9.06567846e-04 4.80594099e-01 5.33895567e-02 -7.46253654e-02 2.10830241e-01 1.00529633e-01 -1.49035180e+00 1.17129958e+00 1.09171283e+00 1.26971617e-01 -6.18907690e-01 -2.21967533e-01 -3.41442198e-01 6.29564464e-01 8.75760257e-01 -3.93225282e-01 -5.76358795e-01 1.35266572e-01 2.19046429e-01 6.94939077e-01 3.64498466e-01 -1.23705590e+00 4.23339516e-01 -2.94379860e-01 6.17983239e-03 -3.83802414e-01 5.62773943e-01 -1.26935172e+00 -2.20215395e-02 1.09817624e+00 -4.16939229e-01 2.15664670e-01 -1.09367877e-01 6.80880368e-01 1.70286313e-01 1.92627966e-01 7.58987665e-01 -2.33802855e-01 -7.56883144e-01 -1.14844702e-01 -1.33970511e+00 -4.48366702e-01 1.82623708e+00 -1.97150901e-01 1.96838342e-02 -1.31125331e+00 -8.19247305e-01 9.24450397e-01 5.27001917e-01 4.31350976e-01 1.03172338e+00 -1.28213799e+00 -5.52015960e-01 -7.73865581e-02 1.33536547e-01 -3.69891256e-01 2.71215141e-01 2.29416177e-01 -6.58920705e-01 6.41837940e-02 -9.15233970e-01 -3.46328110e-01 -1.03026927e+00 5.36341548e-01 4.28289592e-01 6.42480180e-02 -6.03067815e-01 5.68162441e-01 2.60393452e-02 -1.03368831e+00 6.74659610e-01 2.90095568e-01 -7.56461501e-01 -4.08439398e-01 2.28075787e-01 4.27546501e-01 -9.08257484e-01 -7.67023861e-01 -2.42945999e-01 3.10376137e-01 1.98278546e-01 -6.26547813e-01 1.47158682e+00 -3.88317466e-01 -1.78520605e-02 6.37307465e-01 9.36136842e-01 -8.34201753e-01 -1.62212336e+00 2.72761971e-01 -1.14927776e-01 1.07771121e-02 -7.41298854e-01 -7.91314662e-01 -7.49790430e-01 5.41351020e-01 8.87701690e-01 3.02370399e-01 9.25702631e-01 -9.09614265e-02 8.18796754e-01 1.30080485e+00 9.08297002e-01 -1.42928064e+00 1.17251515e+00 9.11509395e-01 1.22263718e+00 -1.74948955e+00 -5.16628504e-01 2.38002196e-01 -1.58254778e+00 7.14824319e-01 1.18606818e+00 -5.59315681e-01 5.91053426e-01 -8.67222995e-02 3.47733945e-01 -4.94087011e-01 -9.48456109e-01 1.44162457e-02 -4.79615152e-01 1.13529944e+00 -1.46528304e-01 1.99229196e-01 3.33592832e-01 2.56000847e-01 -2.75024414e-01 2.48886403e-02 6.96012080e-01 9.08820271e-01 -8.21409106e-01 -4.01762187e-01 -1.96136668e-01 -1.32666633e-01 2.86722124e-01 6.76406682e-01 -7.02605724e-01 7.66698539e-01 -3.68716240e-01 1.58910799e+00 -8.71984288e-03 -6.08149827e-01 4.20599997e-01 -2.42293611e-01 6.46360219e-02 -7.08886445e-01 -7.13377118e-01 -3.61979693e-01 6.75471276e-02 -7.91921198e-01 -3.66211474e-01 -7.23595083e-01 -2.08316922e+00 -3.26268852e-01 3.22118402e-01 2.98868790e-02 1.93169504e-01 8.70171130e-01 2.83960223e-01 2.73530632e-01 1.07621706e+00 -1.32057631e+00 -2.11420178e-01 -1.46058607e+00 -4.60919529e-01 6.25109553e-01 4.24891680e-01 -6.25537872e-01 -3.94051969e-01 -1.76896289e-01]

[4.826992511749268, 1.0419059991836548]

c47c5c1b-9cfd-4a32-977b-3e66865b5b18

cascade-transformers-for-end-to-end-person

2203.09642

https://arxiv.org/abs/2203.09642v1

https://arxiv.org/pdf/2203.09642v1.pdf

Cascade Transformers for End-to-End Person Search

The goal of person search is to localize a target person from a gallery set of scene images, which is extremely challenging due to large scale variations, pose/viewpoint changes, and occlusions. In this paper, we propose the Cascade Occluded Attention Transformer (COAT) for end-to-end person search. Our three-stage cascade design focuses on detecting people in the first stage, while later stages simultaneously and progressively refine the representation for person detection and re-identification. At each stage the occluded attention transformer applies tighter intersection over union thresholds, forcing the network to learn coarse-to-fine pose/scale invariant features. Meanwhile, we calculate each detection's occluded attention to differentiate a person's tokens from other people or the background. In this way, we simulate the effect of other objects occluding a person of interest at the token-level. Through comprehensive experiments, we demonstrate the benefits of our method by achieving state-of-the-art performance on two benchmark datasets.

['Brian Clipp', 'Anthony Hoogs', 'Christopher Funk', 'Daniel Davila', 'Rodney LaLonde', 'Dawei Du', 'Rui Yu']

2022-03-17

http://openaccess.thecvf.com//content/CVPR2022/html/Yu_Cascade_Transformers_for_End-to-End_Person_Search_CVPR_2022_paper.html

http://openaccess.thecvf.com//content/CVPR2022/papers/Yu_Cascade_Transformers_for_End-to-End_Person_Search_CVPR_2022_paper.pdf

cvpr-2022-1

['person-search']

['computer-vision']

[-1.59679711e-01 -5.58230102e-01 3.80994558e-01 -2.38970980e-01 -6.46053255e-01 -4.62813795e-01 5.73608220e-01 -1.47575766e-01 -7.24328279e-01 3.77763987e-01 2.93786824e-01 4.89054650e-01 2.71791369e-01 -5.51032007e-01 -3.17334354e-01 -4.72856075e-01 1.26974255e-01 6.66306973e-01 2.61913329e-01 4.56312858e-02 -4.75887954e-02 4.42938924e-01 -1.37924230e+00 1.89172387e-01 5.71501851e-01 7.50921249e-01 1.05427705e-01 5.57590008e-01 2.61752099e-01 3.58039171e-01 -8.95729542e-01 -7.88617730e-01 3.80897969e-01 -1.30534127e-01 -6.88724220e-01 3.24553460e-01 8.55236828e-01 -5.60753167e-01 -6.36877775e-01 1.14276254e+00 8.82109463e-01 2.53563583e-01 3.31724167e-01 -1.14763951e+00 -8.47053230e-01 3.01864862e-01 -9.70206916e-01 5.58786929e-01 6.68361485e-01 5.00695586e-01 8.00833046e-01 -1.14845955e+00 1.00644872e-01 1.73844004e+00 5.00419915e-01 6.73391104e-01 -1.01490664e+00 -8.14463973e-01 7.50102997e-01 3.08010548e-01 -1.70445919e+00 -4.24770594e-01 6.13792360e-01 -4.28894132e-01 6.24068618e-01 2.54653752e-01 8.36280048e-01 1.08609462e+00 -3.90037924e-01 1.05720460e+00 6.61350727e-01 -1.47896454e-01 -2.09331766e-01 9.44022164e-02 3.02006423e-01 8.35405052e-01 3.18162799e-01 9.03409123e-02 -6.04574800e-01 -1.38516769e-01 6.79380894e-01 2.91690856e-01 -1.95396990e-01 -1.16526239e-01 -1.00176692e+00 4.73587841e-01 7.62213647e-01 6.79820105e-02 -3.83972436e-01 2.30067328e-01 3.32831264e-01 -7.24396110e-02 3.24035496e-01 -4.89222705e-02 -2.26536870e-01 2.90808171e-01 -7.88212597e-01 3.46305698e-01 2.77673632e-01 8.67239892e-01 4.39496398e-01 -3.08824629e-01 -9.92872000e-01 7.72596180e-01 2.02669680e-01 4.89923328e-01 3.39282304e-01 -3.60806644e-01 4.38298255e-01 8.25871289e-01 4.09914553e-01 -9.26751792e-01 -2.20416456e-01 -9.82418597e-01 -8.21294725e-01 -2.20001638e-01 4.86569107e-01 -6.39842227e-02 -1.14136958e+00 1.76747131e+00 5.96692681e-01 3.99696946e-01 -4.09077525e-01 1.40706909e+00 9.14793611e-01 1.97914183e-01 3.57720107e-01 1.71653479e-01 1.95033228e+00 -1.20184815e+00 -3.18133205e-01 -4.89123195e-01 -7.62186050e-02 -4.93568420e-01 9.16130006e-01 -1.58652980e-02 -1.06820917e+00 -1.01434791e+00 -4.02626961e-01 -2.55258679e-01 -2.82887816e-01 6.45202816e-01 3.91958266e-01 6.54283762e-01 -8.03208947e-01 6.73481524e-02 -5.83343089e-01 -4.28163618e-01 6.34365499e-01 3.73895913e-01 -2.05379292e-01 -1.18949033e-01 -1.19306576e+00 5.09492874e-01 -1.70164675e-01 3.92703205e-01 -1.01208484e+00 -5.85890174e-01 -6.66108072e-01 3.57436210e-01 3.41175795e-01 -1.13119209e+00 9.60223556e-01 -7.25132585e-01 -9.54770446e-01 1.18350446e+00 -5.07366240e-01 -2.97321916e-01 1.02273655e+00 -5.78361213e-01 -1.68371156e-01 -2.22011376e-02 6.26677752e-01 7.04680324e-01 9.12815273e-01 -9.77799833e-01 -9.99057174e-01 -7.79994488e-01 9.39786136e-02 3.68929595e-01 -3.63995105e-01 4.99407351e-01 -1.31755722e+00 -7.09321737e-01 -9.56566408e-02 -7.38555908e-01 -2.72995532e-01 2.03496173e-01 -6.09954357e-01 -6.45770550e-01 5.91297328e-01 -7.02459097e-01 9.28202510e-01 -2.01197600e+00 1.17210582e-01 1.16137445e-01 4.15044963e-01 2.13899866e-01 -1.79589540e-01 -1.15515344e-01 1.50801942e-01 -4.68938351e-02 3.45977902e-01 -9.35086846e-01 -3.50982673e-03 -3.46271902e-01 -7.03954557e-03 5.94057679e-01 1.96009174e-01 1.04027081e+00 -8.40867996e-01 -5.52584589e-01 2.16246456e-01 5.64093351e-01 -4.88111883e-01 2.09917903e-01 3.28940511e-01 5.26472151e-01 -6.13434494e-01 8.95326853e-01 5.86261213e-01 -2.86988884e-01 -3.87810826e-01 -2.05981448e-01 -1.92306116e-02 -7.09728226e-02 -1.23052359e+00 1.42869151e+00 -7.02004656e-02 3.99741977e-01 8.14959630e-02 -4.48423386e-01 6.23810530e-01 -2.83892229e-02 2.12655976e-01 -6.18635535e-01 1.16038702e-01 -3.29046845e-01 -2.02010214e-01 -3.16926092e-01 3.56550306e-01 4.49061215e-01 -2.92376280e-01 2.53085732e-01 -1.50956690e-01 9.25680101e-01 2.27463111e-01 1.78084776e-01 8.17657232e-01 -1.82441920e-01 -2.21710261e-02 -1.21520184e-01 8.28849792e-01 -4.23121452e-01 6.11788392e-01 1.14826357e+00 -5.73145211e-01 6.05220616e-01 -8.14967602e-03 -8.06738913e-01 -7.02915907e-01 -1.09893095e+00 2.11463630e-01 1.47082353e+00 4.94902998e-01 -2.01426804e-01 -8.70100021e-01 -6.55263662e-01 1.33878142e-01 4.71303090e-02 -7.65761554e-01 -1.01110026e-01 -6.29286885e-01 -7.36383915e-01 6.72739863e-01 5.62582612e-01 9.80415642e-01 -9.88284290e-01 -6.55120075e-01 6.50943294e-02 -4.64230537e-01 -1.09928453e+00 -1.26393735e+00 -5.47052860e-01 1.32395625e-01 -1.17795682e+00 -1.24756789e+00 -9.55406010e-01 8.68354619e-01 5.01589894e-01 1.01120591e+00 3.17637056e-01 -6.80240691e-01 2.45686725e-01 3.67083438e-02 -3.34223926e-01 4.03471589e-01 7.63955787e-02 2.40841478e-01 5.25036395e-01 5.76200068e-01 -9.17114317e-02 -1.11920309e+00 5.77788532e-01 -1.49614424e-01 -2.04607472e-01 3.61044794e-01 6.26819491e-01 4.24466521e-01 3.18532914e-01 2.09183872e-01 -3.18395883e-01 6.08755052e-01 -1.71728141e-04 -5.30752778e-01 4.00299430e-01 2.99944095e-02 -3.56118381e-01 2.84531027e-01 -6.44065261e-01 -8.89459252e-01 2.78752387e-01 5.21884300e-02 -6.56192243e-01 -2.49442667e-01 -2.27663189e-01 -4.41541076e-01 -3.44926454e-02 4.63804483e-01 3.29139262e-01 -5.55754006e-01 -5.74983299e-01 7.99862519e-02 4.71980751e-01 9.09804523e-01 -5.18533766e-01 9.57513750e-01 6.52552187e-01 -6.06578052e-01 -4.06689793e-01 -9.92878437e-01 -7.70163894e-01 -7.42340147e-01 -3.77406925e-01 8.19345176e-01 -1.28352034e+00 -1.08948910e+00 7.48532474e-01 -1.18130350e+00 -3.41736972e-02 -4.64567840e-02 1.16703659e-01 2.65536308e-01 2.61694938e-01 -6.05376959e-01 -9.61028814e-01 -5.73448896e-01 -1.07951176e+00 1.43324113e+00 5.79795480e-01 -1.04451329e-01 -5.36706567e-01 -2.20359117e-01 4.72705752e-01 1.20933764e-01 4.41046767e-02 9.95698050e-02 -5.15817225e-01 -7.04372585e-01 -5.06689131e-01 -4.71071392e-01 -2.12103218e-01 1.36289179e-01 -4.60817605e-01 -9.97531176e-01 -7.07081139e-01 -3.01015764e-01 -7.97884073e-03 1.22376800e+00 2.73127407e-01 1.18905270e+00 -3.38928461e-01 -6.94212794e-01 7.04423547e-01 9.73812878e-01 -2.78508008e-01 2.23661274e-01 1.21607266e-01 9.17442501e-01 4.98805881e-01 4.30301964e-01 4.52629924e-01 4.93071675e-01 8.55065763e-01 4.05499712e-02 -4.13424611e-01 -2.83856273e-01 -4.47976679e-01 2.12484956e-01 -3.10423672e-01 -1.93582043e-01 -2.57668346e-01 -6.96038067e-01 7.01058805e-01 -1.71475494e+00 -1.12454236e+00 1.72383264e-01 2.29796839e+00 4.76165324e-01 1.79538637e-01 5.15995085e-01 -1.87113225e-01 1.29210997e+00 -2.60724202e-02 -6.86694562e-01 5.47304392e-01 -8.21685195e-02 -1.68250546e-01 3.32062632e-01 4.49507684e-01 -1.42110717e+00 1.24059355e+00 5.73026085e+00 7.46958911e-01 -7.81524181e-01 8.22300911e-02 7.45073259e-01 -4.04854506e-01 3.12615007e-01 -3.26062083e-01 -1.30385828e+00 7.01379299e-01 -7.39555806e-02 -2.82836128e-02 4.47845608e-01 6.65933609e-01 1.83525100e-01 9.59860682e-02 -1.04203320e+00 1.29931426e+00 2.31439814e-01 -7.48672247e-01 1.16786910e-02 3.89395654e-02 4.70437557e-01 -2.32656896e-01 2.19269082e-01 4.34508443e-01 3.70465249e-01 -8.65000367e-01 8.36640000e-01 4.63102490e-01 6.60600126e-01 -8.00930500e-01 6.23159468e-01 3.35845172e-01 -1.71172678e+00 -2.99946815e-01 -3.07082355e-01 -7.17774779e-02 3.04738820e-01 3.55454266e-01 -4.95619804e-01 5.01518138e-02 1.11130452e+00 3.69511724e-01 -8.51940334e-01 1.25713611e+00 -2.25098148e-01 7.06790239e-02 -4.21748668e-01 6.51158243e-02 1.03111314e-02 1.93416521e-01 6.10170066e-01 1.38845456e+00 -1.14104701e-02 1.51691213e-01 6.42651200e-01 1.03074563e+00 -1.11827299e-01 -2.65443057e-01 -4.81473356e-02 6.64518356e-01 5.56089044e-01 1.21925068e+00 -6.10357523e-01 -4.23432350e-01 -2.73310125e-01 1.45319152e+00 5.00612617e-01 5.53234398e-01 -1.04630399e+00 1.12679759e-02 7.87967205e-01 1.89664230e-01 3.76515269e-01 1.30007759e-01 1.17062151e-01 -1.27922261e+00 3.54560316e-01 -7.41876662e-01 7.90523708e-01 -5.65238178e-01 -1.52071142e+00 5.30926645e-01 -2.99207717e-01 -7.56189644e-01 2.43946761e-01 -2.64241576e-01 -7.49719441e-01 1.21952498e+00 -1.29602122e+00 -1.64641953e+00 -7.04290926e-01 9.50570047e-01 7.00614631e-01 -2.16580465e-01 4.56638336e-01 5.48676670e-01 -9.81019795e-01 1.15183330e+00 -5.37532985e-01 9.24392998e-01 6.86867535e-01 -1.05258346e+00 7.33235598e-01 1.12874305e+00 -2.55984906e-02 8.27125251e-01 4.77465242e-01 -7.77269721e-01 -8.59294951e-01 -1.23022783e+00 8.87248993e-01 -6.20568871e-01 2.29165569e-01 -6.01332843e-01 -5.50705969e-01 5.91719985e-01 -2.28343293e-01 2.58553207e-01 3.48183632e-01 2.90910393e-01 -4.58168268e-01 -1.04598038e-01 -1.07742512e+00 6.50479138e-01 1.36110449e+00 -5.15549660e-01 -2.89345741e-01 3.83208424e-01 3.99659276e-01 -4.66193974e-01 -3.44905943e-01 1.19641155e-01 6.73091412e-01 -8.85721564e-01 1.52594495e+00 -6.21824801e-01 -9.71622020e-02 -5.23150563e-01 2.17613921e-01 -9.53265071e-01 -8.62220287e-01 -5.92640281e-01 2.23338008e-02 1.37776875e+00 -1.28725037e-01 -4.11255330e-01 8.56001139e-01 7.77681351e-01 2.55718380e-01 -4.32419360e-01 -8.89993310e-01 -4.79206383e-01 -3.64799947e-01 3.73221263e-02 6.75808847e-01 5.07593334e-01 -3.73435467e-01 5.12980521e-01 -6.03800535e-01 6.24527454e-01 1.02171350e+00 1.71563923e-01 9.87018108e-01 -1.09054911e+00 -3.29039812e-01 -4.24020618e-01 -4.10377115e-01 -1.45406818e+00 4.01033927e-03 -5.09364784e-01 -5.12483306e-02 -1.31978989e+00 7.84966946e-01 -3.82037848e-01 -4.58237916e-01 4.78317171e-01 -8.27378213e-01 3.82627100e-01 3.98380011e-01 3.81308913e-01 -1.00271678e+00 4.75122064e-01 1.16295278e+00 -5.13421476e-01 -2.24849135e-01 3.34407210e-01 -9.46100175e-01 6.79825306e-01 5.33129871e-01 -2.55185515e-01 6.61474466e-03 -7.01858759e-01 -3.98753613e-01 -2.75141895e-01 1.11730671e+00 -1.02468312e+00 5.25807798e-01 1.90055013e-01 1.10678875e+00 -7.47261047e-01 4.56236184e-01 -4.72455204e-01 -1.57239586e-01 6.33752704e-01 -3.09143424e-01 1.14410497e-01 8.46445262e-02 6.04429781e-01 1.15475923e-01 2.86926985e-01 8.29892755e-01 -3.77427459e-01 -7.23051608e-01 8.47139597e-01 6.72708824e-02 7.23347589e-02 8.24454486e-01 -1.97320819e-01 -1.09362833e-01 -2.47081563e-01 -8.21755469e-01 6.92730486e-01 3.14477950e-01 5.63719392e-01 5.79691410e-01 -1.21430731e+00 -1.04049158e+00 2.85678804e-01 4.49282043e-02 6.43693432e-02 4.84041065e-01 5.15359819e-01 -6.16272464e-02 2.92407125e-01 1.01829030e-01 -6.47561550e-01 -1.63933945e+00 6.76127911e-01 8.13801885e-01 -3.19384605e-01 -6.30062878e-01 1.30795944e+00 7.47946024e-01 -9.28309187e-02 5.72552085e-01 1.54787406e-01 -2.45860726e-01 -5.71608171e-02 8.88068557e-01 2.22800821e-01 -5.01506865e-01 -8.14559639e-01 -7.02862859e-01 6.97674334e-01 -4.12779242e-01 4.16985229e-02 7.08920836e-01 -3.30685794e-01 1.44446462e-01 -2.47505486e-01 8.43365014e-01 1.09220721e-01 -1.38536632e+00 -4.69777524e-01 -4.06622171e-01 -7.65305161e-01 -2.97890902e-01 -6.54419422e-01 -1.17384338e+00 6.34962559e-01 8.35007012e-01 -1.40951738e-01 9.14427817e-01 2.89869070e-01 7.64648259e-01 2.16174349e-01 3.40283990e-01 -1.03517640e+00 2.84243584e-01 2.61699468e-01 8.42882335e-01 -1.18937647e+00 2.98342090e-02 -4.87237543e-01 -3.85967851e-01 4.38495636e-01 9.51133788e-01 -9.20302793e-02 2.79482156e-01 -6.07237369e-02 -4.80688959e-02 -6.77520931e-02 -3.70505720e-01 -7.07269728e-01 3.09955478e-01 6.51252508e-01 -8.87660235e-02 -7.32179545e-03 3.38287264e-01 7.81592548e-01 -2.61420101e-01 -1.65073052e-01 -2.82087237e-01 4.33113635e-01 -3.79916072e-01 -5.96111298e-01 -7.67221928e-01 1.86390683e-01 -4.82698321e-01 -1.37301147e-01 -7.21246779e-01 5.96194625e-01 5.59373021e-01 9.79257107e-01 2.70418972e-01 -1.01678997e-01 5.20778179e-01 -3.42239678e-01 3.35228115e-01 -3.92718196e-01 -8.12660456e-01 2.30814055e-01 -1.85086101e-01 -4.31404978e-01 -1.69130370e-01 -8.59642267e-01 -8.43570948e-01 -3.53442460e-01 -2.78502464e-01 3.39133553e-02 -6.70657456e-02 8.11013341e-01 3.60854685e-01 5.13830423e-01 4.52927321e-01 -9.06155348e-01 -3.37127388e-01 -9.63434041e-01 -1.73310056e-01 7.64543176e-01 4.05097276e-01 -7.81621575e-01 2.11670771e-02 3.65662836e-02]

[14.80836009979248, 0.8138939738273621]

a11c0e89-8880-4f3d-9714-e95b16d59160

quantifying-model-uncertainty-for-semantic

2211.01999

https://arxiv.org/abs/2211.01999v1

https://arxiv.org/pdf/2211.01999v1.pdf

Quantifying Model Uncertainty for Semantic Segmentation using Operators in the RKHS

Deep learning models for semantic segmentation are prone to poor performance in real-world applications due to the highly challenging nature of the task. Model uncertainty quantification (UQ) is one way to address this issue of lack of model trustworthiness by enabling the practitioner to know how much to trust a segmentation output. Current UQ methods in this application domain are mainly restricted to Bayesian based methods which are computationally expensive and are only able to extract central moments of uncertainty thereby limiting the quality of their uncertainty estimates. We present a simple framework for high-resolution predictive uncertainty quantification of semantic segmentation models that leverages a multi-moment functional definition of uncertainty associated with the model's feature space in the reproducing kernel Hilbert space (RKHS). The multiple uncertainty functionals extracted from this framework are defined by the local density dynamics of the model's feature space and hence automatically align themselves at the tail-regions of the intrinsic probability density function of the feature space (where uncertainty is the highest) in such a way that the successively higher order moments quantify the more uncertain regions. This leads to a significantly more accurate view of model uncertainty than conventional Bayesian methods. Moreover, the extraction of such moments is done in a single-shot computation making it much faster than Bayesian and ensemble approaches (that involve a high number of forward stochastic passes of the model to quantify its uncertainty). We demonstrate these advantages through experimental evaluations of our framework implemented over four different state-of-the-art model architectures that are trained and evaluated on two benchmark road-scene segmentation datasets (Camvid and Cityscapes).

['Jose C. Principe', 'Rishabh Singh']

2022-11-03

['scene-segmentation']

['computer-vision']

[ 2.71642543e-02 2.75100648e-01 -3.30360159e-02 -3.89596283e-01 -1.23990846e+00 -4.42988575e-01 8.26600075e-01 1.58363655e-01 -5.16009927e-01 7.14929938e-01 -1.54993027e-01 -2.08126992e-01 -3.26890588e-01 -8.76607358e-01 -9.19194400e-01 -8.11361969e-01 1.67075336e-01 9.05257046e-01 4.05103654e-01 1.60890043e-01 1.90632150e-01 2.91416168e-01 -1.62031782e+00 -1.51323721e-01 1.13746989e+00 1.22693074e+00 -1.49503872e-01 5.23538291e-01 -1.01957031e-01 3.15764308e-01 -3.30169111e-01 -3.00385356e-01 4.90453690e-02 -6.14075102e-02 -9.61602688e-01 -3.48064266e-02 2.35924497e-01 -2.33347401e-01 9.99342427e-02 1.36080909e+00 4.22470234e-02 2.60090500e-01 1.20170045e+00 -1.01533699e+00 4.32928763e-02 4.36931461e-01 -3.34233969e-01 6.88628703e-02 6.25408068e-02 1.78020503e-02 1.08015597e+00 -8.26375067e-01 4.62791890e-01 1.25451148e+00 7.23144472e-01 2.09290519e-01 -1.58290303e+00 -2.63673574e-01 -6.07735850e-02 1.08196303e-01 -1.56488740e+00 -2.86449760e-01 4.56902266e-01 -7.17392743e-01 7.77689457e-01 9.07339230e-02 5.00376523e-01 9.26663816e-01 2.74378002e-01 8.05586994e-01 1.28662395e+00 -8.01265314e-02 7.95048475e-01 2.05221400e-01 1.65704817e-01 5.05525589e-01 1.43384516e-01 2.42576554e-01 -4.90602612e-01 -1.15186788e-01 5.87536216e-01 -3.78750175e-01 -1.28714994e-01 -7.76560426e-01 -8.47730339e-01 1.05883276e+00 4.72314268e-01 1.43463045e-01 -3.16625327e-01 4.64398533e-01 2.16890693e-01 -3.87978673e-01 7.71615505e-01 7.82388449e-02 -4.42204148e-01 -3.17470163e-01 -1.26319790e+00 4.97563511e-01 8.99147689e-01 6.42862082e-01 8.80068898e-01 -2.39380270e-01 -1.04318544e-01 6.02523267e-01 7.56982028e-01 4.45623964e-01 -8.78833979e-02 -1.12873912e+00 1.74013555e-01 2.44141400e-01 4.09387320e-01 -7.35656023e-01 -2.12234035e-01 -4.93376046e-01 -6.53959930e-01 3.88117045e-01 6.86700642e-01 5.28670996e-02 -1.10769522e+00 1.66155934e+00 6.07030153e-01 4.65317339e-01 -3.80898416e-02 7.21240640e-01 3.97528201e-01 5.55323184e-01 1.62532389e-01 1.76708221e-01 1.25242555e+00 -3.86854619e-01 -3.77581060e-01 -6.02503307e-02 4.32701796e-01 -3.91178757e-01 8.83841574e-01 3.44160289e-01 -7.26401210e-01 -4.88424242e-01 -9.90187526e-01 1.49904713e-01 -4.74414766e-01 -2.53609329e-01 3.60479683e-01 8.12935948e-01 -7.76624382e-01 1.03846991e+00 -1.08920228e+00 8.20879415e-02 7.84424007e-01 1.20811127e-01 -1.44958034e-01 1.64168477e-01 -1.25422132e+00 1.04283166e+00 6.11201882e-01 2.81363130e-01 -8.09219718e-01 -9.33789432e-01 -1.12548709e+00 3.36684324e-02 3.62286538e-01 -5.05045831e-01 1.33818889e+00 -5.89005888e-01 -1.56829739e+00 6.33199394e-01 2.32951678e-02 -6.51169717e-01 9.65196073e-01 -3.65723878e-01 8.36448893e-02 3.07797231e-02 2.12730408e-01 6.67539716e-01 1.18818533e+00 -1.33380651e+00 -4.89805281e-01 -3.47572684e-01 -2.13174909e-01 -2.47429982e-02 4.57297742e-01 -4.08327907e-01 -1.90938026e-01 -1.81738406e-01 3.52539510e-01 -8.84675443e-01 -4.54634368e-01 -1.51132941e-01 -4.17964399e-01 -9.25939530e-02 6.38552248e-01 -5.93616128e-01 9.63032901e-01 -1.85465777e+00 8.77533332e-02 4.42924410e-01 1.79050282e-01 1.94310948e-01 4.30878073e-01 1.69905752e-01 3.24036658e-01 3.41833800e-01 -9.29089487e-01 -3.16833466e-01 4.28777188e-01 3.85938764e-01 -3.70633125e-01 5.88112354e-01 4.00768310e-01 8.67779970e-01 -9.90577579e-01 -5.98506093e-01 6.69995725e-01 6.56275690e-01 -3.21427554e-01 -5.12405150e-02 -5.08661568e-01 5.44741511e-01 -5.82083106e-01 2.40169480e-01 8.83956313e-01 -1.34413421e-01 -3.68981183e-01 -1.00997731e-01 -5.38752507e-03 1.41502813e-01 -1.39777958e+00 1.64667058e+00 -4.81642902e-01 4.58001524e-01 -1.80003464e-01 -7.16182888e-01 6.97983623e-01 3.27716097e-02 4.70998496e-01 -5.50739206e-02 2.23946363e-01 2.46202633e-01 -2.73843884e-01 -6.82763606e-02 4.71595466e-01 -2.84114271e-01 -1.96260333e-01 2.40179032e-01 2.10315749e-01 -8.47896039e-01 -1.21497200e-03 1.55119836e-01 7.92795539e-01 6.44931912e-01 1.65617839e-01 -5.45033157e-01 4.18134809e-01 -1.55041739e-01 5.01241863e-01 8.88838947e-01 -3.87124985e-01 6.46065414e-01 6.30684257e-01 -1.23053335e-01 -9.83692288e-01 -1.38991034e+00 -7.66294420e-01 3.35703313e-01 1.67002171e-01 -2.56593347e-01 -1.17681754e+00 -7.18855202e-01 1.12683065e-01 1.25838363e+00 -7.86907554e-01 -1.18275605e-01 1.62272286e-02 -6.93388820e-01 3.47341865e-01 5.00349581e-01 7.12898374e-01 -7.05498338e-01 -9.51721847e-01 3.54159474e-01 -1.45420209e-01 -1.20768511e+00 -1.20059829e-02 7.84272999e-02 -8.76505196e-01 -9.35254753e-01 -4.81528163e-01 1.65305167e-01 2.79832006e-01 -5.18349171e-01 1.11957324e+00 -4.95405883e-01 -2.29977444e-01 4.45417851e-01 -3.36899832e-02 -5.39535642e-01 -4.78668541e-01 -2.93430686e-01 -1.91843778e-01 7.46726319e-02 3.66607577e-01 -3.97912651e-01 -3.25710356e-01 1.61647275e-01 -8.88428390e-01 -1.65704250e-01 2.27313772e-01 8.23488057e-01 7.69811392e-01 3.73120040e-01 2.46980637e-01 -7.73459554e-01 3.49545985e-01 -4.76594120e-01 -1.07082820e+00 1.13513462e-01 -5.84380627e-01 4.04988199e-01 5.25511354e-02 -1.90168992e-02 -1.24941802e+00 7.95825869e-02 -2.40398813e-02 -2.47329950e-01 -3.33163649e-01 6.78533912e-01 9.65285227e-02 6.63596541e-02 5.79661608e-01 -1.55738443e-01 -1.65991336e-01 -3.21810246e-01 5.16603112e-01 4.13487792e-01 5.44077039e-01 -7.92568743e-01 6.66152775e-01 6.48779035e-01 3.83173585e-01 -8.52947950e-01 -8.75276089e-01 -6.05175376e-01 -8.88553262e-01 -4.69089746e-01 1.00768697e+00 -6.08919799e-01 -5.11215389e-01 6.03440285e-01 -1.10528946e+00 -1.37710720e-01 -5.41326106e-01 5.32156825e-01 -9.00996447e-01 4.22614276e-01 -2.64171064e-01 -1.32069993e+00 -1.60574336e-02 -1.28904450e+00 1.51905739e+00 3.13137591e-01 -1.89087406e-01 -1.11658967e+00 6.70206621e-02 3.53090793e-01 2.69122243e-01 5.03811657e-01 9.35372055e-01 -4.61773366e-01 -5.96964896e-01 -3.17004055e-01 -4.12119329e-01 6.11317337e-01 -3.62829924e-01 3.94955873e-01 -1.35085833e+00 2.44746625e-01 1.44130990e-01 -1.20942958e-01 1.16853082e+00 8.21879327e-01 8.71745408e-01 1.13675348e-01 -2.71362692e-01 2.36832023e-01 1.50299883e+00 -3.74924511e-01 6.30195916e-01 -1.10800765e-01 5.73726177e-01 6.98064983e-01 6.89650357e-01 3.46462935e-01 3.45124125e-01 6.11959457e-01 6.59397781e-01 3.14267695e-01 2.57422715e-01 -2.39948034e-01 1.93003818e-01 3.31536353e-01 6.15629442e-02 -2.24283859e-02 -1.18355393e+00 5.78933895e-01 -2.13153815e+00 -7.62138009e-01 -2.97354698e-01 2.50630975e+00 6.86591029e-01 4.54499632e-01 -8.07208791e-02 1.38302550e-01 4.73187327e-01 2.09018901e-01 -5.68197012e-01 -4.25672501e-01 2.66123027e-01 2.53066272e-01 6.18321359e-01 7.67871201e-01 -1.28717375e+00 9.29094553e-01 6.01880836e+00 1.11189604e+00 -5.83365679e-01 8.93643964e-03 7.04456389e-01 2.75669187e-01 -3.84126693e-01 2.55693644e-01 -6.71524167e-01 5.12153029e-01 1.16590810e+00 1.64417818e-01 1.62664533e-01 8.74611378e-01 6.14516474e-02 -8.68480623e-01 -1.16057515e+00 7.75989175e-01 -6.11251950e-01 -1.31435513e+00 -2.92535722e-01 2.11882815e-01 7.26394176e-01 3.06839526e-01 1.30894125e-01 2.07391277e-01 4.58691746e-01 -1.24460292e+00 8.62504184e-01 9.57116604e-01 4.47914749e-01 -9.00733650e-01 8.40741992e-01 5.41358352e-01 -8.72285604e-01 2.02503666e-01 -3.45874965e-01 2.08919480e-01 3.32012564e-01 1.15076113e+00 -9.45519865e-01 5.12413919e-01 8.30081105e-01 5.38841307e-01 -2.80903399e-01 8.52221727e-01 -1.62532285e-01 6.87726021e-01 -7.37765849e-01 1.06329739e-01 4.75683719e-01 -5.09332657e-01 8.30978692e-01 1.07353437e+00 1.58705458e-01 -2.03653991e-01 -1.65778816e-01 1.66416526e+00 2.57932335e-01 -2.80415922e-01 -4.94005322e-01 7.59270461e-03 2.67502964e-01 1.06099439e+00 -9.39940453e-01 -1.74001887e-01 -3.05154137e-02 6.25174582e-01 1.38487414e-01 3.16381276e-01 -7.67544329e-01 2.62834243e-02 6.63393974e-01 -1.62271410e-01 3.56271803e-01 -3.07444960e-01 -5.01215756e-01 -9.74164069e-01 1.72941592e-02 -3.96593541e-01 1.84448719e-01 -7.09406435e-01 -1.21103895e+00 2.92279929e-01 4.70725387e-01 -8.32972467e-01 -6.28475547e-01 -7.36098528e-01 -3.68598551e-01 1.07332039e+00 -1.38604522e+00 -1.13333559e+00 1.02467790e-01 2.84227043e-01 2.96339244e-01 4.43693340e-01 8.72651100e-01 -2.96825707e-01 -2.83498913e-01 7.99776707e-03 2.69796312e-01 -1.76657796e-01 6.99210614e-02 -1.61436737e+00 5.11496186e-01 8.92488420e-01 2.73503125e-01 2.51473039e-01 9.60254788e-01 -6.98695242e-01 -8.26272130e-01 -8.82402718e-01 6.56724036e-01 -9.85648751e-01 7.42881417e-01 -2.99214572e-01 -9.49781597e-01 2.91402131e-01 -4.58471566e-01 2.84752697e-01 3.18956792e-01 8.51209164e-02 -3.99383485e-01 1.52546629e-01 -1.39078605e+00 3.24745953e-01 4.71819341e-01 -7.31471121e-01 -4.64984119e-01 1.37975097e-01 5.49408972e-01 -3.78773957e-01 -9.93481874e-01 7.46160209e-01 5.40950239e-01 -1.33955324e+00 7.42974997e-01 -3.73545259e-01 2.50233591e-01 -4.07019019e-01 -4.32754904e-01 -1.14200997e+00 1.20383821e-01 -4.69405800e-01 -1.04929104e-01 1.09337509e+00 3.44547510e-01 -6.48190796e-01 6.48147166e-01 9.12633896e-01 2.53073201e-02 -8.04579079e-01 -1.43624425e+00 -6.61294699e-01 2.07115903e-01 -1.11003959e+00 6.38448954e-01 3.59081775e-01 -5.31048894e-01 -9.64834392e-02 1.16352431e-01 3.86436075e-01 1.04658484e+00 -1.25939205e-01 4.80860829e-01 -1.59152555e+00 -1.42284676e-01 -4.59734857e-01 -7.75063932e-01 -5.92855930e-01 2.78699577e-01 -5.27985275e-01 4.71263260e-01 -1.33979499e+00 -2.84060836e-02 -4.32277292e-01 -2.78308243e-01 -1.00878388e-01 1.46425748e-03 -1.20859027e-01 -1.42207980e-01 8.73620659e-02 -3.95646453e-01 6.25024915e-01 7.89565384e-01 2.68302485e-02 3.67175415e-02 2.10141823e-01 -1.15773976e-01 1.10074639e+00 5.21274328e-01 -5.40822268e-01 -3.54832172e-01 -3.41031477e-02 2.79754460e-01 -8.45755339e-02 9.05530930e-01 -1.05046785e+00 -2.97915731e-02 -1.10576802e-03 2.44835526e-01 -7.26472378e-01 3.62376124e-01 -8.37364376e-01 2.08349496e-01 2.05138966e-01 -1.92962840e-01 -6.00139856e-01 1.91084892e-01 9.00655448e-01 -1.91536531e-01 -5.07442713e-01 1.00159717e+00 -2.81437114e-02 -7.34107256e-01 2.48740420e-01 -2.09094137e-01 5.12136742e-02 8.37765634e-01 -1.24917649e-01 2.52637804e-01 -4.25811857e-01 -6.94483876e-01 3.49343382e-02 4.12981838e-01 6.98743016e-02 3.68311435e-01 -1.04602456e+00 -6.31199718e-01 4.11647968e-02 8.02582055e-02 4.76896644e-01 4.47847098e-01 8.14803123e-01 -3.25979203e-01 3.88527453e-01 1.78465128e-01 -1.22807896e+00 -6.07358158e-01 1.95234656e-01 6.56755984e-01 -4.59773779e-01 -4.71612871e-01 9.54833806e-01 4.88199964e-02 -5.70812821e-01 -4.91569228e-02 -6.59848154e-01 -2.19290014e-02 1.31505027e-01 1.18372545e-01 5.59488833e-01 1.67446807e-01 -9.93382514e-01 -3.88014525e-01 5.56631923e-01 1.28851324e-01 -5.20212710e-01 1.03871250e+00 -1.74627468e-01 4.40154783e-02 7.51308978e-01 1.02255750e+00 -3.35273445e-01 -1.83524644e+00 -2.48872444e-01 2.96160430e-01 -3.76572520e-01 5.03293216e-01 -7.74400353e-01 -5.86546063e-01 1.05100262e+00 6.06361747e-01 1.74948707e-01 5.98435581e-01 6.98515326e-02 4.48964238e-01 5.80660887e-02 4.00696367e-01 -1.51527929e+00 -3.79963160e-01 4.80160385e-01 7.38337696e-01 -1.40259433e+00 7.81011432e-02 -3.59672010e-01 -6.12046599e-01 1.00442171e+00 7.38457888e-02 -9.03191343e-02 1.08777237e+00 3.48422490e-02 -1.03311442e-01 -2.80900002e-01 -2.52500623e-01 -3.61047924e-01 6.91251993e-01 6.08286977e-01 -3.23859206e-03 3.17513585e-01 2.04838499e-01 5.47319114e-01 -2.25073993e-01 -1.21451907e-01 2.97840208e-01 5.73064029e-01 -4.51580375e-01 -7.40083337e-01 -2.44727552e-01 4.43442762e-01 -2.82684326e-01 7.98565820e-02 1.53742516e-02 7.33413219e-01 1.05018854e-01 1.09781134e+00 -7.27824345e-02 -9.51395035e-02 -8.95451102e-03 2.32741654e-01 5.14309287e-01 -5.18857956e-01 6.87952861e-02 6.66467473e-02 3.29092294e-02 -9.09927368e-01 -3.68896067e-01 -9.58438635e-01 -1.34537697e+00 4.50046593e-03 -5.26029468e-01 8.98177773e-02 1.16630793e+00 1.38234580e+00 5.97714670e-02 3.40356827e-01 2.30743885e-01 -1.06533587e+00 -9.65140939e-01 -8.39594126e-01 -7.75263608e-01 1.25548288e-01 3.39128852e-01 -9.48078811e-01 -4.49645817e-01 -2.71602750e-01]

[7.267406940460205, 3.4886274337768555]

57905948-686e-4bb0-adab-33234eee0052

jacobian-norm-for-unsupervised-source-free

2204.03467

https://arxiv.org/abs/2204.03467v1

https://arxiv.org/pdf/2204.03467v1.pdf

Jacobian Norm for Unsupervised Source-Free Domain Adaptation

Unsupervised Source (data) Free domain adaptation (USFDA) aims to transfer knowledge from a well-trained source model to a related but unlabeled target domain. In such a scenario, all conventional adaptation methods that require source data fail. To combat this challenge, existing USFDAs turn to transfer knowledge by aligning the target feature to the latent distribution hidden in the source model. However, such information is naturally limited. Thus, the alignment in such a scenario is not only difficult but also insufficient, which degrades the target generalization performance. To relieve this dilemma in current USFDAs, we are motivated to explore a new perspective to boost their performance. For this purpose and gaining necessary insight, we look back upon the origin of the domain adaptation and first theoretically derive a new-brand target generalization error bound based on the model smoothness. Then, following the theoretical insight, a general and model-smoothness-guided Jacobian norm (JN) regularizer is designed and imposed on the target domain to mitigate this dilemma. Extensive experiments are conducted to validate its effectiveness. In its implementation, just with a few lines of codes added to the existing USFDAs, we achieve superior results on various benchmark datasets.

['Songcan Chen', 'Meng Cao', 'Weikai Li']

2022-04-07

['source-free-domain-adaptation']

['computer-vision']

[ 3.17943126e-01 3.25554498e-02 -5.39673984e-01 -4.56368715e-01 -6.26474977e-01 -4.85134155e-01 5.04772246e-01 -1.37304083e-01 -9.28905308e-02 9.03174281e-01 1.21872425e-01 -8.91708136e-02 -1.62746787e-01 -6.70092165e-01 -6.16141737e-01 -9.15828824e-01 4.21287745e-01 1.90351013e-04 9.51682106e-02 -2.89061934e-01 8.57958421e-02 8.93944800e-02 -9.62768912e-01 -1.90695867e-01 1.38034427e+00 8.72293890e-01 3.14391911e-01 -1.75101504e-01 -1.65893108e-01 3.92077446e-01 -3.44953090e-01 -3.21042567e-01 3.77648205e-01 -6.84757829e-01 -7.26643443e-01 3.21155012e-01 5.73455915e-02 -1.60214886e-01 -8.60348642e-02 1.33487284e+00 3.21952850e-01 2.16982156e-01 8.78451526e-01 -1.17042994e+00 -8.23678792e-01 3.02080601e-01 -4.97782230e-01 1.79206431e-01 9.60998535e-02 -4.20958363e-02 7.38138556e-01 -9.74007607e-01 4.96402204e-01 9.73467827e-01 5.78035533e-01 6.50528848e-01 -1.12203884e+00 -6.12804890e-01 5.51576316e-01 4.07016128e-02 -1.30606031e+00 -3.88781279e-01 1.15517950e+00 -4.13530409e-01 3.13962191e-01 -8.12984034e-02 2.90371984e-01 1.29239714e+00 -1.06151663e-01 6.75585508e-01 1.11152399e+00 -4.60833460e-01 2.69943267e-01 5.72493851e-01 2.40407754e-02 4.08656597e-01 2.19794154e-01 9.56674069e-02 -3.18052948e-01 3.43065485e-02 8.04842114e-01 -3.10772955e-02 -4.25074160e-01 -7.54169405e-01 -1.15871549e+00 7.68246472e-01 6.47159517e-01 3.44627678e-01 -2.06780761e-01 -5.92872441e-01 2.43176386e-01 3.50564867e-01 4.68480915e-01 2.68358707e-01 -5.65474749e-01 1.78367451e-01 -6.54605091e-01 -8.61540660e-02 5.69528878e-01 1.04557300e+00 8.69462371e-01 2.70285517e-01 4.63301800e-02 1.00018501e+00 4.53844696e-01 5.13573885e-01 7.68593550e-01 -5.59743404e-01 6.50507867e-01 7.96481192e-01 1.78692102e-01 -1.10228074e+00 -1.74848244e-01 -8.70390236e-01 -1.05604887e+00 -6.97234869e-02 4.96988058e-01 -1.61092266e-01 -7.56394148e-01 1.99826884e+00 5.91286421e-01 1.55850440e-01 2.66155899e-01 1.11460042e+00 3.41181457e-01 7.31302619e-01 6.43812642e-02 -4.49476451e-01 9.76393223e-01 -9.42184687e-01 -6.41719699e-01 -4.16563988e-01 5.91418803e-01 -6.78453267e-01 1.34178984e+00 2.34131172e-01 -5.66149116e-01 -7.37677634e-01 -1.19769561e+00 2.24178165e-01 -2.76629686e-01 2.11947709e-01 4.40284938e-01 6.05711401e-01 -4.76348042e-01 3.33310783e-01 -6.56762719e-01 -5.33993781e-01 3.06221992e-01 1.30705923e-01 -4.05816853e-01 -7.00916871e-02 -1.38249242e+00 8.17881644e-01 6.34966195e-01 1.04778938e-01 -6.55774891e-01 -6.38998568e-01 -6.10086203e-01 -1.61690414e-01 3.85369956e-01 -5.71507514e-01 1.09721208e+00 -1.33758211e+00 -1.79644525e+00 6.22845531e-01 -1.20820925e-01 -2.25602746e-01 4.40484762e-01 -1.71461374e-01 -5.69404423e-01 -1.31259456e-01 2.12226868e-01 2.18954489e-01 1.00392187e+00 -1.37635434e+00 -5.34088314e-01 -3.31878543e-01 -5.70832007e-02 3.47299367e-01 -9.69781458e-01 -3.00165832e-01 -2.50328690e-01 -8.87367725e-01 2.36386076e-01 -7.14588642e-01 -2.76193947e-01 -1.49477884e-01 -2.80498773e-01 -1.26209781e-01 6.83618367e-01 -5.81627190e-01 1.38981795e+00 -2.32247376e+00 2.96287715e-01 2.48304978e-01 -1.09064810e-01 5.04924774e-01 -7.48176575e-02 4.35105503e-01 -2.16375768e-01 -1.26951575e-01 -6.33388817e-01 -7.28144348e-02 -7.65910000e-03 1.54315755e-01 -6.07580304e-01 4.22805101e-01 1.65403157e-01 5.76506853e-01 -9.26703334e-01 -4.10022974e-01 -4.81926044e-03 2.80821413e-01 -6.41608059e-01 3.16974938e-01 -2.10550994e-01 8.31716537e-01 -8.73114765e-01 5.32754719e-01 8.42152357e-01 -3.52036715e-01 1.26384139e-01 -2.16093257e-01 1.97130106e-02 1.63471803e-01 -1.30050611e+00 1.89340603e+00 -4.10084307e-01 -5.22616506e-03 3.85233178e-03 -1.50016749e+00 1.25426662e+00 1.90172151e-01 4.86039728e-01 -4.94605929e-01 8.16053897e-03 5.17346978e-01 -1.14318825e-01 -4.29858744e-01 3.47816981e-02 -4.38790530e-01 -1.90538503e-02 8.66108760e-02 8.54344144e-02 1.67044222e-01 -1.84522644e-01 -2.33576372e-02 6.68287992e-01 4.99871224e-01 5.47441721e-01 -4.31147784e-01 9.12429094e-01 7.78326914e-02 8.75899971e-01 2.58222908e-01 -3.40503812e-01 4.22429472e-01 1.50222585e-01 -2.39784747e-01 -8.43385398e-01 -1.12451875e+00 -2.15633199e-01 8.79763007e-01 2.65777200e-01 -1.71699747e-01 -7.79214084e-01 -8.81599605e-01 -2.15799615e-01 5.58805525e-01 -4.53202158e-01 -4.94736731e-01 -4.16003287e-01 -7.59752095e-01 2.17644572e-01 5.22737324e-01 8.32313657e-01 -6.43727303e-01 -1.08345747e-01 2.46464387e-01 -2.32818231e-01 -1.00764859e+00 -6.59647584e-01 3.63696627e-02 -9.96920645e-01 -8.10046434e-01 -1.04649377e+00 -1.04878116e+00 8.88598800e-01 2.68926084e-01 6.70217693e-01 -3.25173527e-01 4.68978554e-01 5.07544130e-02 -4.64683443e-01 -3.50544602e-01 -4.21073675e-01 3.94379020e-01 3.29910368e-01 3.27210367e-01 5.22503138e-01 -7.96076894e-01 -3.89617532e-01 4.97337043e-01 -9.55993891e-01 3.28032002e-02 8.01846743e-01 9.91593063e-01 4.15312260e-01 9.33099613e-02 1.12307382e+00 -6.97610676e-01 5.54136455e-01 -7.27061808e-01 -7.36412346e-01 2.49769419e-01 -8.95895183e-01 -6.34634048e-02 1.02504539e+00 -6.65682435e-01 -1.38507974e+00 2.46751934e-01 2.38652285e-02 -4.51520473e-01 -1.71554059e-01 8.30510795e-01 -6.35905266e-01 1.15837485e-01 8.68761539e-01 4.31271315e-01 -2.47703958e-02 -6.16566539e-01 3.24944407e-01 8.34521472e-01 4.43044662e-01 -6.31538630e-01 1.17843807e+00 3.23567182e-01 -1.99155599e-01 -6.25536263e-01 -1.05897105e+00 -3.66491914e-01 -8.02390456e-01 7.81570077e-02 5.45165122e-01 -8.76175523e-01 5.02544222e-03 3.58044147e-01 -9.32901442e-01 -2.61283219e-01 -1.97761998e-01 6.62947297e-01 -5.06564617e-01 3.98231655e-01 -1.14179194e-01 -6.01508856e-01 -9.31560900e-03 -8.33456993e-01 5.63913167e-01 3.56656671e-01 1.22475691e-01 -1.23827672e+00 1.75484687e-01 9.61764976e-02 5.13238609e-01 9.87035334e-02 8.38508248e-01 -7.82486558e-01 -3.80276829e-01 -1.06159605e-01 -2.49874324e-01 6.46724701e-01 5.85697651e-01 -4.61276442e-01 -8.44373882e-01 -3.53253067e-01 4.15207326e-01 -5.20020835e-02 5.06228805e-01 1.13771856e-01 8.88160467e-01 -4.04087901e-01 -3.70837390e-01 6.97291315e-01 1.39097452e+00 2.21136823e-01 3.06721658e-01 4.87968862e-01 6.64997399e-01 6.10124350e-01 7.89274991e-01 3.70868236e-01 3.73582929e-01 7.49417245e-01 1.28427073e-01 -9.26244780e-02 -2.19283085e-02 -6.15051389e-01 5.31014800e-01 1.08571005e+00 1.05509460e-01 -3.19075733e-02 -8.51102948e-01 5.95966399e-01 -1.67848051e+00 -4.59554374e-01 1.63437456e-01 2.30413866e+00 1.15253246e+00 5.12594581e-02 1.32473066e-01 -1.67065337e-02 8.50386262e-01 -5.39452918e-02 -8.63310277e-01 8.02627206e-02 -1.50053725e-01 -2.58652687e-01 1.80300951e-01 3.68305892e-01 -1.05697620e+00 8.87472570e-01 5.47315264e+00 1.03922510e+00 -1.37661409e+00 7.87529200e-02 3.89190376e-01 3.27209413e-01 -3.51157039e-01 2.10537866e-01 -7.73654997e-01 5.47173738e-01 7.35012412e-01 -4.09786612e-01 3.67004097e-01 1.00430107e+00 1.99344173e-01 3.11681777e-01 -1.04337192e+00 9.00486529e-01 -5.41988797e-02 -8.33779216e-01 -2.00946834e-02 5.18204533e-02 7.67258763e-01 -3.01564604e-01 1.55152222e-02 4.93996918e-01 2.25098059e-02 -6.05649590e-01 4.20089692e-01 3.26648623e-01 7.83357561e-01 -6.56572104e-01 5.88831186e-01 6.19376421e-01 -9.92095470e-01 -6.35893717e-02 -6.25144005e-01 -8.63771215e-02 7.56568089e-02 7.07120895e-01 -7.42329836e-01 8.58716547e-01 4.34281975e-01 9.85739291e-01 -4.28982049e-01 9.07452226e-01 -2.59349376e-01 7.35701859e-01 -2.39437789e-01 3.00005019e-01 2.72940956e-02 -3.54435176e-01 6.95871174e-01 8.68681729e-01 4.76880759e-01 1.28570989e-01 2.12179288e-01 9.41500127e-01 -3.02348714e-02 4.25865263e-01 -7.50938237e-01 -2.16520447e-02 4.37857091e-01 1.01371706e+00 -4.10319209e-01 -2.79200435e-01 -4.44090158e-01 9.82792079e-01 2.65646189e-01 6.83322728e-01 -8.68000031e-01 -3.62051815e-01 5.01974165e-01 5.04049025e-02 1.54523954e-01 -2.19249055e-01 -3.23174328e-01 -1.58850098e+00 2.88661003e-01 -9.28652883e-01 3.32962006e-01 -4.30747688e-01 -1.66831493e+00 4.98336256e-01 8.70973915e-02 -1.84535670e+00 -9.36903208e-02 -3.94577235e-01 -3.89771223e-01 9.33640122e-01 -1.81444883e+00 -1.15816069e+00 -2.03188106e-01 7.90724576e-01 4.85186905e-01 -3.47119212e-01 8.05939078e-01 5.06375015e-01 -7.04982817e-01 9.28396404e-01 3.97506267e-01 8.57913271e-02 1.03216696e+00 -9.17716026e-01 6.71189278e-02 9.05255735e-01 -1.11038253e-01 8.86508644e-01 5.53407013e-01 -5.67099035e-01 -1.30693448e+00 -1.09269929e+00 6.66801095e-01 -3.69381815e-01 7.49107599e-01 -2.76993781e-01 -1.29894531e+00 6.85284197e-01 2.06291117e-02 3.71513143e-02 5.97671926e-01 -4.72963937e-02 -3.93201470e-01 -3.80304635e-01 -1.04797292e+00 5.26137054e-01 9.58155215e-01 -3.72849703e-01 -7.93820500e-01 2.86059022e-01 7.13256836e-01 -2.86607534e-01 -9.42218959e-01 5.20235419e-01 2.34628960e-01 -8.39407504e-01 8.92123878e-01 -6.82201684e-01 4.03129101e-01 -5.41987896e-01 -2.54977822e-01 -1.52172053e+00 -2.38598973e-01 -6.02374911e-01 6.86818808e-02 1.53726113e+00 2.86851734e-01 -9.44028974e-01 6.70154631e-01 4.14908767e-01 -2.83391684e-01 -6.39671624e-01 -9.00379837e-01 -1.21479940e+00 4.17965919e-01 -1.40385032e-01 6.45578206e-01 1.34195340e+00 1.61065727e-01 4.47564155e-01 -4.21672076e-01 3.88014257e-01 5.36857426e-01 2.22017184e-01 8.50578725e-01 -1.16763318e+00 -1.43019348e-01 -2.79522598e-01 -1.17042005e-01 -1.40030169e+00 8.22910592e-02 -8.52420032e-01 1.59331283e-03 -1.21696985e+00 4.83286045e-02 -6.52183831e-01 -5.62343121e-01 4.77097958e-01 -3.35979432e-01 -1.38079181e-01 5.72713092e-02 5.34508944e-01 -2.95474291e-01 9.04503345e-01 1.35081601e+00 1.82725508e-02 -2.65681416e-01 1.26883522e-01 -9.69919562e-01 7.48870492e-01 8.67941320e-01 -4.11303699e-01 -7.94322014e-01 -4.27269131e-01 -3.64114195e-02 -4.31686305e-02 2.19641060e-01 -9.71820235e-01 6.62276894e-02 -4.94796813e-01 1.57335341e-01 -1.32589310e-01 1.33683756e-02 -9.64265823e-01 -1.89842179e-01 2.19087362e-01 -2.00412855e-01 -3.91132474e-01 -3.45241651e-02 7.04322338e-01 -4.46174562e-01 -1.91147357e-01 1.01297784e+00 2.14715093e-01 -8.11319530e-01 2.55047709e-01 1.12441383e-01 8.26636180e-02 1.04229748e+00 -2.62374133e-01 -8.11897218e-02 -3.19357455e-01 -5.07269084e-01 1.30953863e-01 5.81489682e-01 4.31203634e-01 4.13655937e-01 -1.62456501e+00 -6.15279138e-01 5.40613234e-01 3.04902732e-01 1.74270138e-01 2.31970534e-01 8.88443828e-01 9.86695066e-02 4.66317415e-01 -2.89706290e-01 -5.18894374e-01 -5.34758806e-01 8.71324480e-01 2.83743501e-01 -2.83636060e-02 -4.49270219e-01 5.64371169e-01 4.67003971e-01 -6.27736211e-01 6.77953959e-02 -2.75866330e-01 -2.87770122e-01 -5.61095737e-02 3.23822260e-01 1.09502450e-01 -1.13938740e-02 -5.76289296e-01 -3.46615165e-01 7.46643424e-01 3.43217365e-02 9.93904620e-02 1.13443840e+00 -5.44174373e-01 2.63886690e-01 4.23538715e-01 1.16724575e+00 1.66201502e-01 -1.50095069e+00 -6.97051883e-01 1.81081891e-02 -4.83619094e-01 -2.08186805e-01 -6.35180473e-01 -9.29349482e-01 9.61856186e-01 5.55998266e-01 2.78759897e-01 1.48039758e+00 -1.97714448e-01 7.82637715e-01 3.26772124e-01 3.77051204e-01 -1.07818246e+00 1.21613242e-01 3.76552105e-01 8.05621624e-01 -1.43798506e+00 -1.13282651e-01 -4.94662613e-01 -7.23777354e-01 9.96276617e-01 7.85528958e-01 -1.32388890e-01 6.28073156e-01 -3.32367003e-01 2.53225893e-01 2.25856930e-01 -3.30986470e-01 6.08143955e-03 3.02600741e-01 7.37848938e-01 3.04274499e-01 -2.12429866e-01 -3.29822391e-01 9.93378043e-01 -1.59475524e-02 2.09293038e-01 2.17718899e-01 7.16873765e-01 -4.31985289e-01 -1.30541492e+00 -3.16316843e-01 4.23374176e-02 -2.33587787e-01 6.30592182e-02 -5.74920252e-02 7.20963478e-01 -3.36424336e-02 7.93777883e-01 -5.10154307e-01 -2.44540840e-01 4.28880811e-01 1.18808355e-02 6.92737475e-02 -6.10557079e-01 7.85769895e-02 2.93158919e-01 -3.99474531e-01 -2.08115250e-01 -5.54043174e-01 -5.45146346e-01 -1.18231630e+00 9.40281078e-02 -3.72453660e-01 3.42954636e-01 4.24444497e-01 1.03458726e+00 3.72542709e-01 3.16488773e-01 8.72117400e-01 -5.13456404e-01 -9.77362037e-01 -1.05261672e+00 -5.82006276e-01 4.70659107e-01 3.47022980e-01 -8.32080245e-01 -4.38165188e-01 2.88033068e-01]

[10.349784851074219, 3.208911895751953]

ea239d62-9239-48bf-9cdf-cd78183888d0

learning-bilingual-word-embeddings-with

https://aclanthology.org/P17-1042

https://aclanthology.org/P17-1042.pdf

Learning bilingual word embeddings with (almost) no bilingual data

Most methods to learn bilingual word embeddings rely on large parallel corpora, which is difficult to obtain for most language pairs. This has motivated an active research line to relax this requirement, with methods that use document-aligned corpora or bilingual dictionaries of a few thousand words instead. In this work, we further reduce the need of bilingual resources using a very simple self-learning approach that can be combined with any dictionary-based mapping technique. Our method exploits the structural similarity of embedding spaces, and works with as little bilingual evidence as a 25 word dictionary or even an automatically generated list of numerals, obtaining results comparable to those of systems that use richer resources.

['Gorka Labaka', 'Mikel Artetxe', 'Eneko Agirre']

2017-07-01

acl-2017-7

['multilingual-word-embeddings']

['methodology']

[-3.93232286e-01 -1.70071289e-01 -5.30010641e-01 -2.38020808e-01 -3.96052361e-01 -8.98689926e-01 9.87160683e-01 5.91889679e-01 -1.07969594e+00 8.14583063e-01 5.45588434e-01 -7.00998485e-01 2.19892979e-01 -9.26886499e-01 -3.33517671e-01 -3.07990760e-01 3.75910550e-01 7.09004164e-01 1.75663531e-01 -8.00731301e-01 2.61948138e-01 3.66712809e-01 -1.22000146e+00 -6.07323647e-02 8.74008954e-01 9.89515036e-02 3.28774333e-01 2.57888317e-01 -5.68609536e-01 2.82186151e-01 -3.20302486e-01 -7.64934063e-01 4.51857209e-01 -5.15173376e-01 -7.44753718e-01 -3.98234397e-01 4.50692356e-01 -1.08366378e-01 -3.86696488e-01 1.15117836e+00 5.09401619e-01 -1.13376185e-01 4.69357520e-01 -6.13895416e-01 -8.99186611e-01 8.50675702e-01 -1.99423179e-01 4.82871264e-01 6.12164736e-01 -1.74214676e-01 1.24157488e+00 -1.12226188e+00 7.47469902e-01 9.32934761e-01 7.11592436e-01 2.87599981e-01 -1.42839396e+00 -6.24708354e-01 -1.11842431e-01 3.18574250e-01 -1.69742858e+00 -3.47843587e-01 7.48763204e-01 -4.78728354e-01 1.36102176e+00 2.69331243e-02 9.89035964e-01 1.06602347e+00 1.48235440e-01 1.07679171e-02 1.46099734e+00 -1.17198253e+00 -1.54635191e-01 7.61358440e-01 1.54977769e-01 6.21857524e-01 5.92632771e-01 2.13078469e-01 -3.80894989e-01 9.77568235e-03 7.86914885e-01 -2.52057821e-01 -9.51027572e-02 -5.98691583e-01 -1.54670346e+00 1.19704187e+00 9.13494006e-02 1.14518678e+00 -7.35442713e-02 -3.55878949e-01 6.10949397e-01 8.54681492e-01 4.24343169e-01 8.66712391e-01 -3.46767336e-01 -3.07195541e-02 -9.99419332e-01 1.42627567e-01 9.51957226e-01 8.17424893e-01 9.53006566e-01 -5.57830296e-02 4.33356822e-01 1.01936150e+00 1.86493158e-01 4.64478642e-01 9.02265131e-01 -3.42368037e-01 3.43175262e-01 4.37176347e-01 -5.29407635e-02 -1.16233754e+00 -2.42002338e-01 -1.38505161e-01 -5.15440941e-01 -8.88395682e-02 5.30021191e-01 4.69900258e-02 -4.50381517e-01 1.71814656e+00 1.89197332e-01 -3.80476892e-01 2.02510655e-01 6.71321511e-01 4.70822126e-01 5.17973602e-01 -1.76416904e-01 -1.28092214e-01 1.49702156e+00 -1.00923026e+00 -7.51360655e-01 -1.21979252e-01 7.72866666e-01 -1.21016037e+00 1.25693488e+00 1.66705474e-01 -7.71547616e-01 -6.13957405e-01 -1.23083127e+00 -1.89737067e-01 -9.23333943e-01 -1.91816568e-01 8.42224121e-01 1.12414694e+00 -1.31002295e+00 6.16992950e-01 -5.55507243e-01 -7.43822634e-01 -1.78426951e-01 3.73691499e-01 -7.07037747e-01 -1.33766755e-01 -1.37036872e+00 1.58161342e+00 7.62523711e-01 -2.45096326e-01 -2.99345881e-01 -5.83857238e-01 -1.01434040e+00 -3.54037195e-01 9.18351412e-02 -5.34679949e-01 5.63028634e-01 -1.01083601e+00 -1.41075635e+00 1.26227105e+00 9.91714075e-02 -2.90588289e-01 3.86380643e-01 -3.78824435e-02 -6.53744638e-01 -1.74724638e-01 1.28104195e-01 6.24798954e-01 5.23982108e-01 -9.37822044e-01 -3.40663165e-01 -2.83732861e-01 3.04465950e-01 1.21208183e-01 -9.47661400e-01 3.69933695e-01 -2.69590735e-01 -9.98827517e-01 -2.13584155e-01 -9.37530935e-01 -4.57967997e-01 -3.46794814e-01 1.04660586e-01 -2.39044309e-01 3.44453417e-02 -8.19156468e-01 1.24304307e+00 -1.95352936e+00 3.36254239e-01 2.27618426e-01 2.30449494e-02 3.62344146e-01 -5.21978140e-01 8.99140120e-01 -1.13320388e-01 2.53242441e-02 5.80508402e-03 -4.04300690e-02 8.60212669e-02 4.15433347e-01 -2.02674806e-01 4.94667739e-01 8.87392610e-02 6.47621036e-01 -1.23278236e+00 -6.70483589e-01 2.75179327e-01 4.04364794e-01 -6.72315300e-01 1.76120028e-02 2.11712107e-01 1.70340478e-01 -6.01067347e-03 3.32073390e-01 3.61415505e-01 2.66212761e-01 8.01167309e-01 -2.48550653e-01 -4.29618478e-01 8.80005002e-01 -1.14311433e+00 1.98355746e+00 -9.91868198e-01 7.00615048e-01 -4.07476574e-01 -1.11327016e+00 1.11998844e+00 3.89455795e-01 3.25090587e-01 -6.74191773e-01 1.64940730e-01 6.97167397e-01 3.74890804e-01 -2.56998867e-01 7.62679100e-01 -4.58340406e-01 -1.15717866e-01 5.85649729e-01 5.07706940e-01 -2.41461098e-01 6.16217315e-01 -2.05628444e-02 8.89204443e-01 9.28138793e-02 6.68723404e-01 -7.90722609e-01 5.87884247e-01 2.02676564e-01 4.03948128e-01 4.07100320e-01 2.30758697e-01 2.16954663e-01 1.44781172e-01 -5.93697548e-01 -1.57465315e+00 -9.98608470e-01 -4.29261535e-01 9.40895319e-01 7.00365901e-02 -8.21342111e-01 -5.50199211e-01 -6.71340227e-01 -1.01487651e-01 5.82872629e-01 -5.10717928e-01 3.10467949e-05 -8.92719865e-01 -6.68344259e-01 5.10772467e-01 4.72529143e-01 -1.23518161e-01 -9.68591332e-01 -2.31487244e-01 3.56014282e-01 8.51057842e-02 -9.54661310e-01 -4.80478764e-01 2.95650125e-01 -9.32498574e-01 -7.92478979e-01 -6.19665682e-01 -1.15468431e+00 5.75222850e-01 2.40672380e-02 1.37982249e+00 -1.36363553e-02 -1.02001868e-01 1.36123508e-01 -6.64794326e-01 -2.09315598e-01 -6.07212663e-01 2.94266969e-01 4.95239496e-01 -3.44127148e-01 9.38377440e-01 -8.60205472e-01 5.90338595e-02 4.89188246e-02 -7.96343148e-01 -2.15110809e-01 6.94204748e-01 1.12761211e+00 2.16583163e-01 -3.92909080e-01 5.71576357e-01 -8.56384218e-01 7.66303957e-01 -2.97992200e-01 -6.52505934e-01 2.03629777e-01 -8.92664254e-01 4.19035703e-01 7.44891703e-01 -6.23548090e-01 -4.18658108e-01 -1.02670483e-01 -3.53746504e-01 -1.01510599e-01 -3.13535333e-03 5.30569851e-01 -1.11649960e-01 -2.22159475e-01 8.01570177e-01 3.17765892e-01 -7.23214447e-02 -6.57863855e-01 6.82825029e-01 7.78419375e-01 3.00346315e-01 -5.58412611e-01 1.06717193e+00 9.74304751e-02 -4.35031056e-01 -7.82069027e-01 -4.67144340e-01 -4.50332165e-01 -1.09712863e+00 2.17790425e-01 6.12771690e-01 -9.15522754e-01 1.26491606e-01 -1.42357424e-01 -1.25798917e+00 -8.96738470e-02 -4.06252265e-01 1.06981337e+00 -2.27024794e-01 5.21675169e-01 -5.78230977e-01 -3.10933590e-02 -3.41205887e-04 -9.44531739e-01 5.33192277e-01 -2.79048622e-01 -6.07394576e-01 -1.43894660e+00 7.48262286e-01 1.76949993e-01 4.64463830e-01 -1.98489591e-01 9.74832475e-01 -8.54825020e-01 -1.66470185e-01 -1.90411106e-01 -7.71611705e-02 5.33465624e-01 4.41460848e-01 -3.92922610e-01 -5.66169202e-01 -5.17103076e-01 2.87590418e-02 -2.78742880e-01 5.55980682e-01 -3.98208678e-01 3.67808133e-01 -1.12984844e-01 -1.98313087e-01 4.33611155e-01 1.63944781e+00 7.62844235e-02 4.31401819e-01 5.92809856e-01 7.68094957e-01 5.64205468e-01 2.99323738e-01 6.27099499e-02 5.74500144e-01 8.66238773e-01 -3.87500942e-01 -1.21673062e-01 -3.41638803e-01 -3.20247412e-01 4.49643314e-01 1.80760193e+00 -1.77154735e-01 2.74649948e-01 -1.17291307e+00 1.13084626e+00 -1.28828156e+00 -7.80309558e-01 3.99328433e-02 2.18451571e+00 1.28093350e+00 -1.08502181e-02 8.28141347e-02 2.00728625e-01 4.68560129e-01 2.32816279e-01 2.02439025e-01 -7.60128975e-01 -2.82051355e-01 7.13764071e-01 5.88031769e-01 7.15277076e-01 -7.87527084e-01 1.19609618e+00 7.32071686e+00 6.48711920e-01 -1.08688772e+00 5.27458072e-01 -2.73876369e-01 1.71384901e-01 -7.99958885e-01 2.27337256e-01 -6.01090491e-01 4.44181979e-01 9.76365507e-01 -3.83497626e-01 5.92936754e-01 5.51972866e-01 -1.72589332e-01 5.37110195e-02 -1.45774531e+00 1.10165823e+00 5.30807555e-01 -1.14966929e+00 2.09996015e-01 1.85046464e-01 7.73716927e-01 1.61582708e-01 -1.63293675e-01 2.06145853e-01 3.89398634e-01 -1.07295835e+00 5.30312836e-01 1.49462730e-01 8.50531399e-01 -6.63767517e-01 8.01750779e-01 8.10509175e-02 -1.00440109e+00 2.55667597e-01 -6.39745712e-01 -2.30944619e-01 8.25742260e-02 5.66196442e-01 -6.35016382e-01 6.04487598e-01 4.39079285e-01 6.30000293e-01 -7.24071980e-01 7.51540542e-01 -3.08413357e-01 3.88994336e-01 -2.79539853e-01 -1.35144323e-01 2.50689328e-01 -5.07979035e-01 4.30986345e-01 1.46259952e+00 4.14989471e-01 -4.36122447e-01 1.08456254e-01 4.21816766e-01 -9.84442979e-03 8.92826557e-01 -1.13309264e+00 -2.11256862e-01 4.19491529e-01 1.08241117e+00 -6.82217658e-01 -3.51241380e-01 -9.08257902e-01 1.11644697e+00 5.54517806e-01 -2.16968298e-01 -4.44459289e-01 -5.28085291e-01 4.33662325e-01 8.87238011e-02 3.37649614e-01 -6.68412626e-01 -1.31231621e-01 -1.47032988e+00 9.47842598e-02 -1.29335189e+00 1.49752617e-01 -2.60085583e-01 -1.42840087e+00 1.00467467e+00 3.31243724e-02 -1.16659665e+00 -4.19569045e-01 -7.67250419e-01 -2.43250012e-01 1.05362487e+00 -1.45744348e+00 -1.10948610e+00 2.52214789e-01 6.74564123e-01 2.62955636e-01 -5.66549242e-01 1.20632803e+00 8.28062177e-01 -2.07311511e-01 8.18340361e-01 9.02093649e-02 2.53207743e-01 1.02829552e+00 -1.27529120e+00 3.62926722e-01 7.37876177e-01 9.28493977e-01 9.93632913e-01 6.85712337e-01 -3.78651023e-01 -1.31419492e+00 -5.89956522e-01 1.83111477e+00 -6.95882559e-01 1.10585439e+00 -6.06391788e-01 -8.28195155e-01 5.44283211e-01 7.90897489e-01 -9.16084275e-02 1.04108620e+00 6.32258236e-01 -7.93873310e-01 -2.35282376e-01 -7.16146648e-01 6.75993621e-01 1.14321113e+00 -9.62493896e-01 -1.25314081e+00 3.89344245e-01 5.01042545e-01 -2.69684419e-02 -1.02786446e+00 1.30908906e-01 4.78823304e-01 -6.91854775e-01 8.71222079e-01 -5.80979049e-01 2.03337833e-01 -2.95126468e-01 -2.83879846e-01 -1.59680772e+00 -2.10662439e-01 -3.08524221e-01 4.36723620e-01 1.24546540e+00 6.47590160e-01 -7.49538481e-01 3.61862868e-01 -2.86653787e-02 2.62801182e-02 -2.62339860e-01 -8.54343176e-01 -1.20057511e+00 4.58992690e-01 -1.08463518e-01 5.74613631e-01 1.63270986e+00 3.90556961e-01 6.20958626e-01 -3.64750743e-01 -3.45363319e-01 2.09815502e-01 1.27379239e-01 6.61393523e-01 -1.19254422e+00 -3.46230358e-01 -5.79043031e-01 -5.82652807e-01 -6.82902336e-01 5.90732932e-01 -1.38969803e+00 -2.38205075e-01 -1.20214081e+00 6.63472787e-02 -6.88813269e-01 -4.13726240e-01 3.80936593e-01 -2.23668534e-02 5.67375124e-01 -1.96315888e-02 9.82252881e-02 -2.46464059e-01 4.01152879e-01 7.83720911e-01 -1.11086830e-01 -1.62863880e-01 -7.47470379e-01 -6.69400632e-01 6.42923057e-01 7.89923668e-01 -6.64406717e-01 -2.52969503e-01 -7.46390462e-01 5.19999683e-01 -4.87394392e-01 -1.61573701e-02 -8.53459537e-01 1.62921706e-03 -6.37309998e-02 1.39984623e-01 2.15473752e-02 1.21329077e-01 -8.74350011e-01 2.03272417e-01 4.32285666e-01 -7.26857334e-02 7.04321682e-01 2.65181750e-01 1.98519006e-01 -5.63626349e-01 -5.21170318e-01 6.45564854e-01 -2.42754102e-01 -6.06125236e-01 -1.29158854e-01 -4.79236037e-01 -3.08416765e-02 8.63703489e-01 -1.30887806e-01 2.04197675e-01 1.44385576e-01 -3.80861461e-01 -2.11300895e-01 9.13587153e-01 7.45028734e-01 2.78299600e-01 -1.74779987e+00 -8.48010957e-01 3.21523547e-01 2.93017149e-01 -7.72186100e-01 -4.54473317e-01 8.18360209e-01 -7.69907176e-01 5.95493793e-01 -7.54407823e-01 -7.50238076e-02 -1.03102612e+00 8.88821542e-01 1.77990049e-02 -4.27199423e-01 -5.66932380e-01 3.20279330e-01 -4.68241334e-01 -8.06993663e-01 -2.76381165e-01 -1.37202162e-02 -4.09032822e-01 4.45394844e-01 3.31037104e-01 5.87758869e-02 1.46044269e-01 -9.28548753e-01 -4.56132829e-01 6.72270715e-01 -6.62376955e-02 -5.01641035e-01 1.31363869e+00 4.75395881e-02 -4.52458888e-01 6.50344253e-01 1.05878139e+00 8.30796599e-01 -2.27095053e-01 -4.43488002e-01 2.23371372e-01 -7.74382770e-01 -4.20769863e-02 -3.81134361e-01 -7.55255520e-01 6.97293997e-01 5.22383511e-01 1.59068599e-01 8.30302596e-01 -1.10327704e-02 7.23261058e-01 4.98109758e-01 5.91020525e-01 -1.16099107e+00 -1.30885750e-01 6.24363661e-01 7.14356005e-01 -1.23157251e+00 -5.68638509e-03 -1.86229452e-01 -2.10195839e-01 1.03323507e+00 2.47785076e-01 -3.72604430e-01 7.43044436e-01 1.64986923e-01 3.71923387e-01 1.16287723e-01 -3.74049902e-01 -4.05905813e-01 2.70429373e-01 6.74179435e-01 8.38995218e-01 8.64053965e-02 -1.29336107e+00 3.20491970e-01 -4.64872479e-01 -2.55432636e-01 3.99033219e-01 7.50222027e-01 -4.71465960e-02 -2.30127025e+00 -3.56447965e-01 2.69092202e-01 -3.85297716e-01 -7.08214641e-01 -3.93696576e-01 1.00221300e+00 3.70215505e-01 7.36135185e-01 9.53394175e-02 -3.56897771e-01 1.91510081e-01 2.45071009e-01 8.31249714e-01 -8.45178127e-01 -6.93260312e-01 -1.61236525e-01 1.87102988e-01 -3.13606203e-01 -7.18511939e-01 -7.09170341e-01 -5.66383839e-01 -4.57160473e-01 -3.15334648e-01 3.31849843e-01 6.71742320e-01 9.01274264e-01 -5.52560426e-02 1.27883675e-02 5.06482840e-01 -5.63550770e-01 -2.74778485e-01 -1.07044542e+00 -4.55853969e-01 3.82206857e-01 -5.58789372e-02 -6.01382256e-01 5.74139005e-04 4.84952554e-02]

[10.968047142028809, 10.04788875579834]

37e7cb69-b67c-4132-9b55-ab21162f1a72

anomaly-detection-in-image-or-latent-space-of

2307.02495

https://arxiv.org/abs/2307.02495v1

https://arxiv.org/pdf/2307.02495v1.pdf

Anomaly detection in image or latent space of patch-based auto-encoders for industrial image analysis

We study several methods for detecting anomalies in color images, constructed on patch-based auto-encoders. Wecompare the performance of three types of methods based, first, on the error between the original image and its reconstruction,second, on the support estimation of the normal image distribution in the latent space, and third, on the error between the originalimage and a restored version of the reconstructed image. These methods are evaluated on the industrial image database MVTecADand compared to two competitive state-of-the-art methods.

['Carole Lartizien', 'Robin Trombetta', 'Nicolas Pinon']

2023-07-04

['anomaly-detection']

['methodology']

[ 3.10857415e-01 -1.40633807e-01 3.33173543e-01 -5.83927631e-02 -7.04427302e-01 -2.77023584e-01 6.06299222e-01 -2.42704246e-02 3.66411619e-02 3.63855213e-01 -3.85828167e-01 -1.25131667e-01 -1.20414168e-01 -7.61415958e-01 -8.40276241e-01 -9.68231618e-01 -6.69929981e-02 1.97500616e-01 3.51211280e-01 -1.55748557e-02 4.40869749e-01 5.15594900e-01 -1.83718336e+00 5.42036831e-01 5.77803254e-01 1.46617270e+00 -9.47201326e-02 8.92953038e-01 1.05000988e-01 7.93184936e-01 -7.35751569e-01 -3.02406371e-01 9.22299325e-02 -8.20788801e-01 -6.09581769e-01 7.63795733e-01 5.49095869e-01 -3.08998466e-01 -4.24113512e-01 1.54409587e+00 -7.68792555e-02 -2.21566662e-01 7.53400922e-01 -1.30612719e+00 -7.49813557e-01 3.09161580e-04 -5.76771080e-01 2.48591810e-01 4.37340945e-01 -1.23518534e-01 6.63588762e-01 -8.20880532e-01 6.78739786e-01 1.07150543e+00 6.42914176e-01 -5.99339269e-02 -1.73008859e+00 -2.40867794e-01 -2.16896087e-01 5.60859978e-01 -1.10636818e+00 -4.58974332e-01 8.61961603e-01 -5.97262502e-01 5.56208730e-01 1.04586124e-01 3.84103030e-01 1.05507171e+00 6.46970987e-01 4.79303926e-01 1.35637653e+00 -7.46151507e-01 2.46464953e-01 1.32722780e-01 -2.82346457e-01 8.73220444e-01 1.81756154e-01 4.43956345e-01 -3.26667637e-01 -3.42195779e-01 8.02244663e-01 -5.15575074e-02 -3.62820745e-01 -3.25058192e-01 -9.08155322e-01 7.50747561e-01 7.88591511e-04 5.21234632e-01 -8.00915241e-01 6.66197836e-02 1.65480793e-01 5.15423000e-01 7.15087771e-01 4.77324575e-02 -1.28168598e-01 -5.58214001e-02 -1.02265108e+00 -2.22906753e-01 5.59870183e-01 5.82881808e-01 8.53627086e-01 3.07503164e-01 2.10954070e-01 5.71144521e-01 4.79721278e-01 7.90745616e-01 5.38434565e-01 -9.76693630e-01 9.53186005e-02 4.12595242e-01 -8.92198905e-02 -1.28308332e+00 1.06501400e-01 -1.09292269e-01 -8.26285779e-01 8.54253709e-01 2.26584643e-01 4.72138710e-02 -1.05976820e+00 1.27221429e+00 -7.30659366e-02 4.24319029e-01 -1.28593640e-02 4.91972148e-01 1.23096786e-01 8.64453435e-01 -6.85098529e-01 -2.98761427e-01 6.99668050e-01 -6.63068295e-01 -8.19255054e-01 1.33849718e-02 -1.91950630e-02 -1.05792534e+00 6.33215427e-01 8.98023129e-01 -1.09601104e+00 -9.51422572e-01 -1.29069066e+00 6.13587081e-01 -1.27508163e-01 4.87143874e-01 -6.66202884e-03 6.27383351e-01 -1.10948205e+00 8.64627361e-01 -8.18446815e-01 -3.48979801e-01 7.64353201e-02 -2.96522677e-01 -6.57424927e-01 -1.11298608e-02 -6.09371603e-01 6.28752828e-01 2.20352739e-01 -1.83964856e-02 -9.17530775e-01 -3.04185301e-01 -7.80313790e-01 -5.47538064e-02 -1.36473402e-01 7.83075467e-02 7.29854822e-01 -1.50616026e+00 -1.54950345e+00 8.56211245e-01 -8.86788871e-03 -4.09500897e-01 6.43105745e-01 -1.30052403e-01 -8.47709298e-01 5.55379212e-01 1.47208095e-01 -7.14315102e-02 1.45511305e+00 -1.51386368e+00 -6.72340691e-01 -3.54331374e-01 -4.81522948e-01 -4.59435791e-01 3.72804552e-02 -2.73443818e-01 -3.15058738e-01 -7.57234216e-01 5.02193511e-01 -8.88306499e-01 -3.69119011e-02 2.04031020e-01 -7.53192902e-01 1.50487334e-01 1.05215824e+00 -1.00138938e+00 6.79284036e-01 -2.78121471e+00 -3.23344246e-02 6.06168747e-01 -3.33737910e-01 1.02904700e-01 -2.97058284e-01 4.74609643e-01 -7.30753422e-01 -1.48444936e-01 -4.49416757e-01 -1.74150780e-01 -1.11087926e-01 1.42900288e-01 -6.11445904e-01 1.03465259e+00 3.04475933e-01 1.69810012e-01 -6.88856900e-01 -3.90632451e-01 3.56495082e-01 2.93963850e-01 -3.60593721e-02 5.55121720e-01 1.81945920e-01 3.37675810e-01 1.61129728e-01 4.91789430e-01 7.82072365e-01 -9.85636190e-02 5.95474504e-02 -1.70631766e-01 -6.73861876e-02 -8.01458731e-02 -1.20232928e+00 1.46726024e+00 -1.58298373e-01 7.37016201e-01 -3.00092340e-01 -1.11775017e+00 9.56077516e-01 6.70119703e-01 6.96652472e-01 -7.55185723e-01 -1.49625465e-01 2.75146276e-01 -2.57072479e-01 -5.06581068e-01 2.73905188e-01 7.44657144e-02 3.16874057e-01 4.91928518e-01 2.71870285e-01 4.48259339e-02 3.46093059e-01 -7.01627433e-02 1.01603925e+00 4.62515950e-01 1.77318037e-01 -2.40276828e-01 6.35825992e-01 -2.59157866e-01 2.96692669e-01 7.52470076e-01 -6.96730316e-02 8.13184202e-01 9.11115885e-01 -2.49755114e-01 -1.25852942e+00 -1.47556281e+00 -2.87630320e-01 4.58313495e-01 1.01131313e-01 -1.17592841e-01 -7.75766730e-01 -7.34234869e-01 4.37628888e-02 7.26030111e-01 -7.15891600e-01 -2.70977706e-01 -1.29823551e-01 -5.75012922e-01 4.49030042e-01 3.97193849e-01 6.36870563e-01 -8.15815806e-01 -5.61754465e-01 5.36191883e-03 -2.67501712e-01 -1.13632751e+00 -4.00767326e-02 1.12295248e-01 -8.71888101e-01 -1.33631921e+00 -4.10651952e-01 -5.67570984e-01 7.20774710e-01 2.37601716e-02 1.24169791e+00 1.58963799e-01 -4.14742470e-01 4.63513941e-01 -2.60723710e-01 -1.36881992e-01 -7.54846215e-01 -7.77271986e-01 -1.38517302e-02 5.86127698e-01 8.76050293e-02 -5.42351544e-01 -2.60452807e-01 4.96907443e-01 -9.21132803e-01 -4.08075780e-01 6.59296751e-01 8.90733480e-01 8.71313751e-01 6.58672988e-01 -9.19820741e-02 -7.69633532e-01 6.25629351e-02 -3.61313850e-01 -1.07835019e+00 4.19036269e-01 -8.36671829e-01 2.14485556e-01 3.96336019e-01 -2.92060584e-01 -1.01369023e+00 2.50674516e-01 -1.15807727e-01 -6.94483697e-01 -3.28391969e-01 1.73182070e-01 6.75495267e-02 1.51717454e-01 6.06531560e-01 5.22878528e-01 1.01061158e-01 -4.41953838e-01 3.99110734e-01 5.48847675e-01 1.07417381e+00 -3.87513608e-01 9.60062385e-01 6.17187083e-01 4.25847284e-02 -1.10504115e+00 -2.65745789e-01 -4.82418686e-01 -7.65577197e-01 -2.60915607e-01 1.07783043e+00 -5.89080572e-01 -2.05696359e-01 8.85696948e-01 -1.37713373e+00 -8.60631838e-02 -6.18330479e-01 4.17087913e-01 -7.22206175e-01 7.09701061e-01 -6.10515475e-01 -7.41308749e-01 2.64962256e-01 -1.03831005e+00 1.06684935e+00 -2.87643284e-01 3.41896445e-01 -1.01292360e+00 4.36461002e-01 -3.69216919e-01 7.41682947e-02 5.41000605e-01 1.10721958e+00 -3.36354643e-01 -4.46756184e-01 -6.56495273e-01 -2.71774709e-01 9.72821712e-01 3.56693298e-01 4.12451386e-01 -1.01108134e+00 -3.04210693e-01 3.88615370e-01 5.51036326e-03 6.82756484e-01 3.70070994e-01 1.04043019e+00 -1.80551454e-01 -2.39182174e-01 5.91053903e-01 1.81447935e+00 2.77972937e-01 1.08413970e+00 3.14660311e-01 2.35470653e-01 3.23817849e-01 4.34022367e-01 3.60595256e-01 -3.11339259e-01 7.24840522e-01 6.54706597e-01 -2.25133330e-01 -1.81888610e-01 -3.35348584e-02 4.14747745e-01 7.21780121e-01 -1.54761955e-01 -2.57179588e-01 -4.67234969e-01 4.63938087e-01 -1.69914019e+00 -1.00262082e+00 -3.70666564e-01 2.23314166e+00 1.98510453e-01 1.92814127e-01 -9.38796401e-02 5.46415031e-01 8.43497217e-01 2.92138100e-01 -3.95388454e-01 -4.50243205e-01 -1.95344254e-01 1.49574563e-01 3.68165433e-01 1.19519286e-01 -1.21698463e+00 1.61526784e-01 7.91646624e+00 4.53978568e-01 -1.02414358e+00 4.01993915e-02 5.82261562e-01 8.72290373e-01 1.18527502e-01 4.57093976e-02 -1.31681979e-01 5.12893379e-01 1.09806728e+00 -2.25717444e-02 3.91133517e-01 9.26578224e-01 -3.32535416e-01 -3.00968856e-01 -1.13294733e+00 9.27612841e-01 5.33974946e-01 -8.07068288e-01 -1.60259992e-01 2.03112572e-01 8.62587512e-01 -2.29424089e-01 2.71394700e-01 -3.96268249e-01 1.18592881e-01 -6.20845497e-01 8.78865421e-01 8.57373536e-01 6.90433145e-01 -6.75641596e-01 1.09998035e+00 -1.97698876e-01 -8.68631244e-01 1.49182543e-01 -5.99371254e-01 3.74612987e-01 -9.58110243e-02 8.30140591e-01 -2.97291577e-01 6.23512805e-01 1.09364045e+00 8.50988626e-01 -6.65103436e-01 9.45482254e-01 -5.52253067e-01 6.67055607e-01 9.13745314e-02 7.92446434e-01 2.02269517e-02 -4.65754062e-01 6.95899248e-01 1.02521074e+00 5.47737420e-01 -4.22288716e-01 1.14696980e-01 1.03491056e+00 4.09748137e-01 -2.56358087e-01 -7.50461876e-01 -1.24946795e-01 -9.03290659e-02 9.61215854e-01 -6.49484396e-01 -4.18791950e-01 -4.95387673e-01 1.48007488e+00 -2.96165138e-01 5.06511569e-01 -6.11122608e-01 -6.06381655e-01 4.26343024e-01 -8.43447074e-02 8.30753624e-01 -3.98889035e-02 9.02791619e-02 -9.64087665e-01 5.67239113e-02 -9.42744911e-01 4.36955839e-01 -1.11790466e+00 -1.46076989e+00 1.00532162e+00 -6.85735643e-02 -1.64261949e+00 -5.99535882e-01 -8.49030316e-01 -4.96874273e-01 9.29683447e-01 -1.24366510e+00 -6.36810601e-01 -1.59070596e-01 6.75170481e-01 3.90244514e-01 -5.02267420e-01 9.96164858e-01 1.13735266e-01 -4.14893776e-01 2.44076058e-01 5.79703867e-01 2.16183931e-01 5.77463388e-01 -1.44605267e+00 3.55908811e-01 1.35076809e+00 3.67567450e-01 -1.22187845e-01 8.92697394e-01 -5.16032755e-01 -9.90085483e-01 -7.91025639e-01 3.56195271e-01 -2.55114347e-01 6.26236022e-01 4.69231084e-02 -9.86423492e-01 7.94910550e-01 4.64977413e-01 2.63632387e-01 3.18607360e-01 -2.18407914e-01 -4.69437927e-01 -1.56882197e-01 -1.12119806e+00 -7.07972944e-02 2.29930192e-01 -7.57823944e-01 -4.56784815e-01 3.17749590e-01 2.10468262e-01 -1.87620342e-01 -7.41600037e-01 9.01429579e-02 5.17980516e-01 -1.44846070e+00 8.96025181e-01 -4.31403965e-01 7.52129734e-01 -3.49195927e-01 -5.08825243e-01 -1.47746027e+00 -5.69822073e-01 -2.31029481e-01 -2.40238346e-02 9.19523478e-01 1.52072325e-01 -6.23071730e-01 4.93829221e-01 -2.73703426e-01 1.10541493e-01 -2.64870495e-01 -8.62232208e-01 -1.06464934e+00 -2.89614797e-01 -2.53807366e-01 4.15632665e-01 6.18036330e-01 -3.20905119e-01 -7.22582117e-02 -4.39324349e-01 5.68641305e-01 7.84273922e-01 2.36840472e-01 5.75787783e-01 -1.25703835e+00 -5.50770342e-01 -1.96883410e-01 -1.18826568e+00 -4.10834819e-01 -5.20638041e-02 -3.04250300e-01 3.96849811e-01 -1.23778236e+00 -8.46140310e-02 2.67801851e-01 -5.82470357e-01 2.92955726e-01 1.38998061e-01 6.45374775e-01 -1.89519882e-01 2.10220322e-01 -3.74251306e-01 3.43443990e-01 6.04227841e-01 -2.01770574e-01 2.92905420e-01 -8.07356164e-02 1.45129673e-02 8.73237967e-01 4.16022748e-01 -6.51493371e-01 -2.17507005e-01 -2.73231398e-02 3.65907326e-02 1.54463276e-01 6.27620995e-01 -1.38670444e+00 -1.98307678e-01 1.07929192e-01 5.83164752e-01 -6.87635720e-01 1.63370192e-01 -1.03825152e+00 1.63117439e-01 8.10066521e-01 -1.08359791e-02 6.75910234e-01 7.31431991e-02 8.01445246e-01 -5.18576741e-01 -4.47977275e-01 1.04697526e+00 -6.12119995e-02 -6.67137444e-01 -8.85431692e-02 -5.71053207e-01 -3.47741723e-01 8.42652380e-01 -1.41332760e-01 -3.25494081e-01 -5.61557829e-01 -5.99977970e-01 -7.98906803e-01 6.15930200e-01 2.06234440e-01 8.20415378e-01 -1.50151598e+00 -7.83397138e-01 7.55412519e-01 2.03537434e-01 -7.30392218e-01 5.42342477e-02 9.17533398e-01 -7.22219229e-01 1.11507066e-01 -4.90443915e-01 -8.55145752e-01 -1.31344831e+00 8.49639595e-01 4.66247529e-01 -1.42423779e-01 -8.06164086e-01 3.40329140e-01 1.27493277e-01 9.50760394e-02 -1.07181206e-01 -1.30101144e-01 2.50560697e-03 -4.06982273e-01 4.91487235e-01 5.02100229e-01 1.95358068e-01 -9.17072177e-01 -3.13447326e-01 6.15624845e-01 1.68633536e-01 -2.54100174e-01 1.16000533e+00 -1.00472450e-01 -4.73036706e-01 8.48581195e-01 1.36140513e+00 3.28305140e-02 -1.30826247e+00 -1.96355149e-01 4.49785776e-02 -9.91976023e-01 2.76486129e-01 -6.76744699e-01 -1.29950047e+00 6.18443549e-01 1.44496858e+00 7.64025390e-01 1.44444728e+00 -2.65286833e-01 3.01639140e-01 2.94586360e-01 1.30467758e-01 -1.04840767e+00 4.41642076e-01 8.50078985e-02 7.43475020e-01 -1.13264132e+00 2.05814496e-01 -2.77947992e-01 -3.94924134e-01 1.42666233e+00 -5.59843369e-02 -4.92609054e-01 9.78756249e-01 -4.83058281e-02 3.10972989e-01 -2.41290689e-01 -5.57951808e-01 -9.21478122e-02 4.49500680e-01 7.12029755e-01 1.83809668e-01 -2.62231678e-01 2.44023070e-01 -1.89601049e-01 1.32085279e-01 -1.50282040e-01 5.44876516e-01 8.74833405e-01 -1.82994545e-01 -9.73369300e-01 -7.40841210e-01 2.19409391e-01 -5.10124028e-01 2.02303663e-01 -1.44933656e-01 7.31861353e-01 2.44626641e-01 1.15402138e+00 3.88326287e-01 -3.66396785e-01 4.66228038e-01 -5.21673448e-02 4.89897966e-01 -1.61957100e-01 2.57573456e-01 7.86291137e-02 -2.36514509e-01 -8.29142869e-01 -5.71905971e-01 -1.04696715e+00 -5.02970994e-01 6.95022866e-02 -5.48356235e-01 2.10936323e-01 9.73065674e-01 9.02901828e-01 5.34394085e-02 5.99566877e-01 1.09463322e+00 -7.95263171e-01 -4.97791350e-01 -8.66101742e-01 -9.87789094e-01 6.35017514e-01 4.09035474e-01 -4.39473599e-01 -5.49886048e-01 4.38575357e-01]

[11.7147216796875, -1.9319080114364624]

636fadf4-671e-4cba-bd6e-c489d084fa22

regression-or-classification-automated-essay

https://aclanthology.org/W19-4409

https://aclanthology.org/W19-4409.pdf

Regression or classification? Automated Essay Scoring for Norwegian

In this paper we present first results for the task of Automated Essay Scoring for Norwegian learner language. We analyze a number of properties of this task experimentally and assess (i) the formulation of the task as either regression or classification, (ii) the use of various non-neural and neural machine learning architectures with various types of input representations, and (iii) applying multi-task learning for joint prediction of essay scoring and native language identification. We find that a GRU-based attention model trained in a single-task setting performs best at the AES task.

['Lilja {\\O}vrelid', 'Stig Johan Berggren', 'Taraka Rama']

2019-08-01

ws-2019-8

['automated-essay-scoring', 'native-language-identification']

['natural-language-processing', 'natural-language-processing']

[ 5.13557419e-02 -9.56469476e-02 -9.81609076e-02 -3.61693859e-01 -1.17423642e+00 -5.76422215e-01 4.80651408e-01 1.55408949e-01 -7.72668123e-01 8.83701682e-01 3.24660718e-01 -8.28929245e-01 -2.93118268e-01 -4.12324876e-01 -4.20139760e-01 -2.09974870e-01 5.11723280e-01 7.92073369e-01 2.08464190e-01 -4.44979638e-01 5.95364869e-01 1.51664853e-01 -1.43201125e+00 3.74724418e-01 8.05102050e-01 8.16983402e-01 1.14358328e-01 1.48822856e+00 -3.30853432e-01 1.36012173e+00 -9.44672167e-01 -7.74577737e-01 3.02394945e-03 -5.01003623e-01 -1.18713057e+00 -5.70001662e-01 8.05899858e-01 -2.38354027e-01 -1.94265351e-01 8.42697144e-01 4.63142127e-01 2.95605868e-01 1.11671960e+00 -7.53209233e-01 -1.20230746e+00 4.57928002e-01 -2.39790708e-01 5.17256081e-01 7.62216568e-01 -2.58922517e-01 1.40667760e+00 -7.56688893e-01 2.86827803e-01 8.68294120e-01 9.05306697e-01 7.86594212e-01 -9.31807101e-01 -4.63005394e-01 -1.15458578e-01 2.47641772e-01 -9.88566518e-01 -4.86402303e-01 4.68443334e-01 -6.46387279e-01 1.10720038e+00 2.36594751e-01 1.81310028e-01 1.21923280e+00 3.97516876e-01 1.14023471e+00 1.44797742e+00 -8.26422215e-01 -1.40351728e-01 2.55180269e-01 9.91606593e-01 9.47381794e-01 -3.08581498e-02 -5.39846113e-03 -8.01259577e-01 -3.13716859e-01 3.80279034e-01 -4.26949233e-01 -1.45171180e-01 3.91016245e-01 -9.61910963e-01 1.11610234e+00 -2.80480504e-01 3.89672041e-01 -1.07417703e-01 1.94657043e-01 6.30430698e-01 1.05914509e+00 6.94000900e-01 7.65684009e-01 -7.70031154e-01 -1.94940537e-01 -1.10550916e+00 2.51317859e-01 1.10993099e+00 6.80882275e-01 3.50234240e-01 3.04183483e-01 -5.00977993e-01 1.09759450e+00 1.50393471e-01 1.06680669e-01 1.21184874e+00 -5.79377592e-01 6.40016735e-01 3.02480847e-01 -3.06717902e-02 -3.27138215e-01 -6.55455351e-01 -1.21969037e-01 -1.36690244e-01 4.05241311e-01 9.38390255e-01 -4.64671165e-01 -5.57233155e-01 1.63972485e+00 -5.36085069e-01 -1.13401815e-01 1.70172900e-01 5.96229553e-01 1.20028055e+00 5.06046891e-01 1.69263765e-01 -1.18700005e-01 1.17837250e+00 -1.19884920e+00 -7.66232848e-01 -8.55661482e-02 9.59862649e-01 -5.28020024e-01 1.24259555e+00 5.16416907e-01 -1.48490536e+00 -8.85270655e-01 -1.02168214e+00 -5.37249267e-01 -5.18410146e-01 3.32295984e-01 3.94219190e-01 1.07080853e+00 -1.41517377e+00 4.62670714e-01 -3.25644106e-01 -2.73025893e-02 -9.24635679e-02 5.84051371e-01 -2.70965278e-01 3.70193243e-01 -1.03224409e+00 1.32154620e+00 3.85856198e-04 -4.63775218e-01 -3.73706311e-01 -5.53090036e-01 -8.34060550e-01 4.02613878e-01 -2.80776531e-01 -4.35011327e-01 1.68401921e+00 -1.37574244e+00 -2.02226353e+00 1.21877992e+00 -1.88997149e-01 -2.89576620e-01 3.79450291e-01 -2.60969073e-01 -1.36402398e-01 -1.90364867e-01 2.83808410e-02 1.00581765e-01 4.80552882e-01 -5.67270100e-01 -5.89102387e-01 -4.48303938e-01 -1.50003105e-01 3.24520618e-01 -6.56497777e-01 5.01681983e-01 2.77761161e-01 -6.03300333e-01 -3.76271218e-01 -6.19992435e-01 1.05707102e-01 -6.94074929e-01 -1.15974480e-02 -9.47774947e-01 4.57303077e-01 -1.03481936e+00 1.40921760e+00 -1.52774775e+00 3.02165419e-01 -1.09637372e-01 1.96260437e-01 1.66756928e-01 -3.38256270e-01 2.18738958e-01 -8.25322270e-02 1.54278025e-01 2.42567390e-01 -6.55211389e-01 2.95139272e-02 -1.96005017e-01 -2.12350801e-01 2.85329670e-01 1.97119880e-02 1.27507353e+00 -6.71370625e-01 -3.46714109e-01 -2.33689204e-01 -8.84016603e-02 -3.01234901e-01 3.94506335e-01 -5.09282947e-02 9.26740915e-02 -2.56984115e-01 4.33511317e-01 7.76569247e-02 -1.00666545e-01 -9.02520027e-03 6.14803195e-01 -2.95761466e-01 9.18754399e-01 -1.02591920e+00 1.60725093e+00 -6.82161391e-01 1.01692688e+00 4.96193953e-02 -7.86380947e-01 1.08034706e+00 5.07051110e-01 3.10124364e-02 -6.43077195e-01 1.55910075e-01 4.76050675e-01 2.53496021e-01 -4.70576227e-01 7.52832353e-01 -1.56792462e-01 -3.42578530e-01 1.13671792e+00 7.45604277e-01 9.36958790e-02 1.34077862e-01 -3.81340422e-02 1.37312281e+00 6.81903362e-02 5.73998213e-01 -6.33659005e-01 7.45066404e-01 -2.82367498e-01 1.15912994e-02 1.07750762e+00 -2.47741058e-01 5.50876677e-01 4.58991587e-01 -5.68889439e-01 -9.13153291e-01 -6.80107415e-01 -1.78055480e-01 2.22841239e+00 -5.68278909e-01 -5.48763722e-02 -5.27310014e-01 -6.99463546e-01 8.71958286e-02 8.67242873e-01 -7.27381349e-01 -1.68792561e-01 -6.93545938e-01 -8.10419679e-01 9.10386622e-01 6.44338429e-01 -3.71696427e-02 -1.13460958e+00 -6.57417774e-01 3.32309484e-01 1.25987485e-01 -6.25597179e-01 -5.96593916e-01 8.08488131e-01 -6.38295829e-01 -9.59555864e-01 -8.18564296e-01 -1.06278956e+00 4.76229489e-02 3.33422348e-02 1.36399424e+00 6.15704060e-01 1.62087068e-01 6.93748593e-01 -1.84971243e-01 -4.94665712e-01 -7.00117648e-01 5.20468533e-01 7.11712465e-02 -3.24234277e-01 8.31015825e-01 -1.81077749e-01 2.48941034e-01 -1.27012208e-01 -4.36549157e-01 -2.68064409e-01 4.21932071e-01 1.32474756e+00 -2.31782764e-01 -8.19599330e-01 9.80677247e-01 -1.09347510e+00 1.54884744e+00 -3.84781837e-01 -2.43711978e-01 6.30295157e-01 -6.50425434e-01 1.67452738e-01 6.41970813e-01 -4.75536615e-01 -8.12269747e-01 -6.23403341e-02 -4.24864441e-01 -3.36583965e-02 -2.82476872e-01 5.16879082e-01 4.21032131e-01 -3.70217174e-01 7.95064092e-01 2.86980480e-01 8.86746570e-02 -4.37667578e-01 -1.37854218e-01 9.39114749e-01 3.51108193e-01 -7.95121133e-01 3.64882708e-01 -5.00310481e-01 -1.75075665e-01 -6.90984786e-01 -1.15982914e+00 -5.32522261e-01 -1.08170033e+00 -2.41089612e-01 8.38420689e-01 -8.33589733e-01 -8.21281910e-01 4.69188988e-01 -1.28053248e+00 -7.02891827e-01 1.94138303e-01 4.01087940e-01 -5.05593777e-01 2.96888977e-01 -1.11616826e+00 -9.50866342e-01 -5.16305387e-01 -1.15536392e+00 7.51065671e-01 3.11961412e-01 -4.80711669e-01 -1.38522744e+00 5.84447026e-01 5.68859756e-01 3.25914472e-01 -5.74913323e-01 1.08060753e+00 -1.43203008e+00 1.77193865e-01 -1.33795112e-01 -2.21979603e-01 2.51971632e-01 -3.99947912e-01 -6.85868561e-02 -1.33352256e+00 -1.40438169e-01 -5.04868180e-02 -1.04100585e+00 1.09531343e+00 4.09307122e-01 1.09263420e+00 -1.61642805e-01 3.44100356e-01 4.83466238e-01 1.16705406e+00 -6.00199625e-02 2.18569800e-01 4.62695271e-01 5.32240927e-01 5.02116323e-01 -3.01937573e-02 2.39398461e-02 6.44665599e-01 6.30003691e-01 -9.42372456e-02 2.07246438e-01 -9.35842097e-02 1.39968485e-01 5.36878347e-01 1.18276513e+00 -2.28178754e-01 -3.76289040e-01 -1.11819959e+00 5.67193031e-01 -1.87788785e+00 -9.91785526e-01 -5.13205349e-01 2.04127574e+00 8.89484167e-01 3.95834120e-03 4.67380702e-01 1.24532230e-01 5.58179379e-01 1.36919826e-01 -6.74356669e-02 -1.31406391e+00 -1.39612570e-01 8.44279170e-01 4.99342084e-01 8.41924369e-01 -1.17172444e+00 8.26278985e-01 7.82247734e+00 4.91481125e-01 -6.88113570e-01 6.47124827e-01 5.38506806e-01 -1.66547410e-02 -5.33805192e-02 -7.36268163e-01 -1.14435029e+00 3.20817828e-01 1.44020665e+00 -2.46912539e-01 3.22816312e-01 6.65969968e-01 -4.17050719e-01 6.62107989e-02 -1.10373521e+00 3.76422912e-01 5.37100315e-01 -8.68816614e-01 7.60850012e-02 -1.78757474e-01 7.42846847e-01 1.25964686e-01 8.53898153e-02 9.04626489e-01 6.25365853e-01 -1.35839748e+00 8.82344723e-01 4.89811033e-01 7.73687184e-01 -6.22603178e-01 7.99999535e-01 4.72088367e-01 -6.45849764e-01 -4.73899990e-01 -5.29870093e-01 -6.84373438e-01 -4.86053795e-01 -1.04337908e-01 -5.47092497e-01 1.47651702e-01 2.75723547e-01 4.13952708e-01 -7.36263692e-01 9.79494274e-01 -4.44636732e-01 8.43302190e-01 6.87177479e-02 -7.69463241e-01 3.52344066e-01 -8.53376910e-02 2.72164375e-01 1.72670817e+00 2.06822887e-01 -8.49596411e-02 1.33146524e-01 5.85048854e-01 -2.47654110e-01 3.68380725e-01 -4.39339638e-01 2.93032616e-01 1.16633274e-01 1.16429710e+00 -1.61331922e-01 -1.23421922e-01 -8.41480017e-01 1.00301278e+00 9.94073927e-01 1.80128470e-01 -3.75464618e-01 -4.77898806e-01 4.53452826e-01 -2.95291960e-01 -3.22092026e-02 -2.05327451e-01 -8.81783187e-01 -1.05783582e+00 -3.44757229e-01 -6.51485205e-01 6.42776906e-01 -7.06047773e-01 -1.24060082e+00 4.72730726e-01 -5.69396019e-01 -5.66791475e-01 -6.35147989e-01 -1.28831065e+00 -1.22530556e+00 1.26169550e+00 -1.64327681e+00 -9.32993412e-01 5.64711578e-02 4.50941801e-01 7.56721973e-01 -9.98290300e-01 1.34102607e+00 9.62338299e-02 -6.40184402e-01 1.06845844e+00 4.91626590e-01 2.99188584e-01 8.47470701e-01 -1.97654438e+00 1.30656317e-01 5.17056584e-01 3.49677026e-01 3.67742598e-01 3.43244791e-01 -2.66893625e-01 -1.01041806e+00 -6.62775040e-01 1.38742483e+00 -7.31983125e-01 9.58412707e-01 -4.21701521e-01 -8.99446547e-01 7.77804852e-01 4.78248566e-01 -5.64994633e-01 1.02275431e+00 7.37450838e-01 -3.45903099e-01 3.47160339e-01 -8.27514350e-01 3.84737104e-01 4.91641670e-01 -8.39449883e-01 -9.23269570e-01 5.11389017e-01 3.27964306e-01 -4.51753259e-01 -8.95110428e-01 -6.04523011e-02 7.21344650e-01 -7.63356030e-01 7.64956772e-01 -1.27676392e+00 1.01808381e+00 7.17687845e-01 2.32541934e-01 -1.52326119e+00 -7.88380563e-01 -3.87248427e-01 -1.93942755e-01 8.49143386e-01 7.24056959e-01 -3.49004328e-01 6.61389530e-01 6.17554605e-01 -3.96152139e-01 -9.04028714e-01 -1.14077330e+00 -3.35856825e-01 7.96274304e-01 -1.26977921e-01 -2.89778155e-03 9.46137607e-01 3.61301839e-01 6.93021357e-01 -1.99860349e-01 -2.75332749e-01 7.99960718e-02 -5.56402579e-02 3.90428692e-01 -1.60369194e+00 -3.85086000e-01 -1.14132476e+00 -1.73600674e-01 -9.12887156e-01 8.41604054e-01 -1.02925527e+00 -1.30231887e-01 -1.21227145e+00 2.53491908e-01 -5.08919433e-02 -6.05241597e-01 3.69195729e-01 -5.70055008e-01 2.35489234e-01 -5.02363294e-02 1.34942085e-01 -8.14586699e-01 1.06724650e-01 7.29437351e-01 -6.45885766e-02 -3.29187922e-02 3.84048134e-01 -7.82886565e-01 4.42336291e-01 8.42355251e-01 -5.69656670e-01 6.99422359e-02 -5.74368954e-01 4.96049404e-01 2.82478660e-01 4.73860018e-02 -7.42815018e-01 3.71675283e-01 6.78805038e-02 4.10432726e-01 -1.42354861e-01 1.88044250e-01 -9.38808694e-02 -9.23107207e-01 4.58100796e-01 -7.85367012e-01 4.58125919e-01 1.41988024e-01 8.37775469e-02 -1.03039920e-01 -1.27203047e+00 7.59586394e-01 -4.37236100e-01 -4.27655369e-01 -1.62079871e-01 -7.58740067e-01 1.50686160e-01 5.33291817e-01 -1.66926742e-01 -3.69303703e-01 -2.99710035e-01 -4.76605296e-01 1.88040406e-01 9.04936939e-02 4.55397189e-01 2.72404999e-01 -1.06846309e+00 -1.18480265e+00 1.60042480e-01 -1.08117819e-01 -7.30892360e-01 -3.74924570e-01 7.33583570e-01 -4.78734851e-01 8.30220580e-01 -2.73619980e-01 -5.27770109e-02 -1.55991900e+00 -1.00351311e-02 5.13746500e-01 -8.66835475e-01 -1.72316760e-01 1.07333446e+00 -1.65913388e-01 -9.19398427e-01 2.24919483e-01 1.34033665e-01 -7.54920363e-01 2.16549769e-01 5.73336720e-01 5.26145458e-01 2.98927486e-01 -3.96625608e-01 1.73471197e-01 1.35393396e-01 -3.14752012e-01 -1.46942511e-01 1.31401980e+00 3.27628404e-01 3.66174392e-02 8.79833519e-01 9.66541886e-01 4.44004349e-02 -6.00032210e-01 -3.95455062e-02 4.72979397e-01 4.61211428e-02 3.80222052e-01 -1.07892323e+00 -5.08779764e-01 9.76733565e-01 3.55668413e-03 2.92941600e-01 5.73089063e-01 -4.11825746e-01 3.98044616e-01 7.02789724e-01 7.31365085e-02 -1.39172649e+00 1.75460547e-01 1.17697847e+00 8.12824667e-01 -1.19041002e+00 2.75990181e-02 4.43825990e-01 -6.22726321e-01 1.73450208e+00 7.94167936e-01 -2.68734902e-01 4.08902794e-01 3.40912223e-01 2.23530471e-01 -1.26562491e-01 -1.08275008e+00 -7.69482227e-03 7.48389006e-01 3.53494465e-01 1.19893479e+00 -5.28174490e-02 -5.74270189e-01 1.27582479e+00 -4.40119982e-01 -2.10012704e-01 8.78277004e-01 5.16941369e-01 -5.84401608e-01 -8.57624173e-01 -3.75210851e-01 9.82544422e-01 -9.25844610e-01 -1.45835102e-01 -8.37782443e-01 5.69525182e-01 -3.79764527e-01 8.48594248e-01 -1.58441082e-01 -2.96557724e-01 9.27328244e-02 9.37315404e-01 5.72748780e-01 -8.68629396e-01 -1.53790057e+00 -5.72367787e-01 2.58967042e-01 1.06112659e-01 -1.55257294e-02 -1.06032836e+00 -6.34031475e-01 -1.45253614e-01 -6.91913843e-01 1.20900735e-01 6.80896878e-01 1.20201850e+00 -2.81198055e-01 7.10336208e-01 3.40669483e-01 -6.24710917e-01 -1.18318069e+00 -1.53638506e+00 -7.31352448e-01 5.86882047e-02 2.77331322e-01 -3.94509822e-01 -4.33856964e-01 -3.62656623e-01]

[11.312026023864746, 9.371820449829102]

932dcdae-f9d1-4ee0-86ba-1f69f2d372f0

prefix-tuning-optimizing-continuous-prompts

2101.0019

https://arxiv.org/abs/2101.00190v1

https://arxiv.org/pdf/2101.00190v1.pdf

Prefix-Tuning: Optimizing Continuous Prompts for Generation

Fine-tuning is the de facto way to leverage large pretrained language models to perform downstream tasks. However, it modifies all the language model parameters and therefore necessitates storing a full copy for each task. In this paper, we propose prefix-tuning, a lightweight alternative to fine-tuning for natural language generation tasks, which keeps language model parameters frozen, but optimizes a small continuous task-specific vector (called the prefix). Prefix-tuning draws inspiration from prompting, allowing subsequent tokens to attend to this prefix as if it were "virtual tokens". We apply prefix-tuning to GPT-2 for table-to-text generation and to BART for summarization. We find that by learning only 0.1\% of the parameters, prefix-tuning obtains comparable performance in the full data setting, outperforms fine-tuning in low-data settings, and extrapolates better to examples with topics unseen during training.

['Percy Liang', 'Xiang Lisa Li']

2021-01-01

https://aclanthology.org/2021.acl-long.353

https://aclanthology.org/2021.acl-long.353.pdf

acl-2021-5

['table-to-text-generation']

['natural-language-processing']

[ 2.78724134e-01 4.96823519e-01 -4.98600036e-01 -2.32762992e-01 -1.20180106e+00 -8.29063118e-01 1.00211072e+00 1.03980750e-01 -4.74262506e-01 9.39654112e-01 7.24629760e-01 -5.47384381e-01 3.23295265e-01 -7.07453132e-01 -8.84123862e-01 -4.88000244e-01 1.02205269e-01 8.06462348e-01 4.47987393e-02 -3.47736716e-01 2.46606529e-01 4.66059633e-02 -1.00536656e+00 6.36888742e-01 8.63931417e-01 3.91882151e-01 2.76130944e-01 6.97154164e-01 -5.16932905e-01 4.70285594e-01 -9.26437676e-01 -4.37947035e-01 1.67382032e-01 -2.27726519e-01 -1.09869671e+00 -8.86195526e-02 4.65747178e-01 -2.55228400e-01 -2.77478784e-01 3.87587547e-01 6.82134867e-01 3.22314858e-01 7.16348350e-01 -8.48600745e-01 -8.44640136e-01 1.24237669e+00 -3.27652991e-01 2.93775290e-01 1.55433714e-01 3.42812985e-01 1.02134335e+00 -8.04292202e-01 6.73845649e-01 1.39350712e+00 6.72036052e-01 8.63316417e-01 -1.38098526e+00 -5.74968696e-01 3.83351743e-01 -4.74940568e-01 -1.03865647e+00 -7.99864054e-01 3.54390770e-01 -3.65365237e-01 1.56424272e+00 -4.90755476e-02 2.25611299e-01 1.50083470e+00 1.91156775e-01 8.03443253e-01 6.50650442e-01 -5.13969183e-01 1.75874084e-01 -1.55865895e-02 -2.13007350e-02 4.25205231e-01 2.38263458e-01 5.10571189e-02 -6.10295832e-01 -4.72071707e-01 5.48482776e-01 -4.03108090e-01 -9.10935849e-02 5.38900085e-02 -1.47481620e+00 8.74968350e-01 2.07426637e-01 1.87447414e-01 -3.49004418e-01 6.26039207e-01 5.73855519e-01 3.40436608e-01 6.61529124e-01 8.98481905e-01 -7.76147723e-01 -3.96674633e-01 -9.89815652e-01 5.51371872e-01 9.80450451e-01 1.24409640e+00 7.22485125e-01 2.26015151e-01 -9.93363380e-01 9.17190969e-01 -3.35515626e-02 2.79964328e-01 9.08130348e-01 -8.47190320e-01 9.29083884e-01 3.12453538e-01 2.25516289e-01 -1.88361526e-01 -2.64581680e-01 -6.27785265e-01 -7.41528630e-01 -4.81585175e-01 4.00245428e-01 -6.72366023e-01 -1.13491440e+00 2.03684545e+00 2.45346203e-02 4.49397452e-02 1.59470141e-01 2.94217974e-01 5.90621710e-01 1.00862634e+00 2.63885140e-01 -3.10061201e-02 1.19977021e+00 -1.32775283e+00 -4.27511513e-01 -7.11655319e-01 9.92875457e-01 -7.43090153e-01 1.57485223e+00 2.03176439e-01 -1.25882554e+00 -4.67416853e-01 -6.12579882e-01 -3.16478908e-01 -4.02361006e-01 1.79825306e-01 7.19855070e-01 4.31578398e-01 -1.29622018e+00 7.04865575e-01 -7.04194546e-01 -3.55048478e-01 3.14994335e-01 2.07442641e-01 -6.66609332e-02 6.33573458e-02 -1.30806661e+00 7.96372592e-01 6.89018011e-01 -2.27111340e-01 -9.72779214e-01 -9.47857738e-01 -7.89145350e-01 3.02786171e-01 2.94059753e-01 -1.25222838e+00 1.75041616e+00 -3.99803877e-01 -1.50776684e+00 7.03892648e-01 -4.90199268e-01 -7.51730323e-01 4.58465725e-01 -3.75076056e-01 -1.18827298e-01 -2.61586934e-01 2.30096191e-01 1.05911732e+00 8.31179321e-01 -1.03828299e+00 -3.38098079e-01 1.47430524e-01 5.95498979e-02 2.47456625e-01 -3.41119796e-01 -1.32087946e-01 -5.44518232e-01 -9.98177111e-01 -3.79250824e-01 -8.64372551e-01 -3.80818814e-01 -5.31070650e-01 -6.53882921e-01 -5.56137502e-01 2.06077814e-01 -3.38322222e-01 1.61528373e+00 -1.80223966e+00 -1.48747221e-01 -7.71386325e-02 5.34342602e-02 3.83864552e-01 -5.24183035e-01 6.85117245e-01 9.72188264e-02 4.61701632e-01 -3.02046955e-01 -5.93959033e-01 1.81936488e-01 3.03933769e-01 -7.00641334e-01 -2.64446050e-01 3.56133163e-01 1.26085985e+00 -1.04764080e+00 -4.66114819e-01 -2.45718509e-01 1.16935737e-01 -9.08383191e-01 4.96011227e-03 -8.13056231e-01 6.07517734e-02 -5.47346950e-01 1.86164901e-01 3.48295510e-01 -3.63000661e-01 -1.04189618e-02 2.20158830e-01 -6.46729534e-03 7.95631349e-01 -7.83763826e-01 2.00243139e+00 -7.54062116e-01 3.94025296e-01 -1.57222942e-01 -6.19544208e-01 8.29977095e-01 4.51994449e-01 8.55304487e-03 -5.10527074e-01 -1.43804938e-01 1.35685042e-01 -3.04881096e-01 -3.92241061e-01 9.55733120e-01 -2.94157624e-01 -4.40303355e-01 7.93829978e-01 1.92842215e-01 -2.59367853e-01 4.73008603e-01 5.04477978e-01 1.18196499e+00 1.86402306e-01 1.95850253e-01 -2.66099989e-01 2.41636515e-01 -1.18305506e-02 3.87524098e-01 1.16704452e+00 2.22618699e-01 4.70896751e-01 5.50131381e-01 -1.83739170e-01 -1.04442656e+00 -1.13334596e+00 1.83243603e-01 1.56489205e+00 -5.01211882e-01 -7.60781646e-01 -9.02096629e-01 -6.93814635e-01 1.53831556e-01 1.11406791e+00 -5.77122033e-01 -3.24469626e-01 -8.18528533e-01 -8.82819235e-01 7.62985647e-01 5.22111416e-01 1.88125432e-01 -1.33526480e+00 -6.60017654e-02 4.63818431e-01 -3.73872459e-01 -8.26917946e-01 -1.02598190e+00 3.45314026e-01 -1.09759188e+00 -3.82717252e-01 -7.12867677e-01 -6.80251420e-01 5.01764536e-01 8.78188666e-03 1.69870830e+00 -6.90719411e-02 1.54854760e-01 -8.94739851e-02 -1.39689431e-01 -4.79193509e-01 -8.10919046e-01 9.45349932e-01 -2.26072490e-01 -3.24355483e-01 2.37876147e-01 -5.81381321e-01 -3.01099360e-01 -1.12486690e-01 -8.58074129e-01 1.54294642e-02 8.37817252e-01 9.85141218e-01 5.24292767e-01 -3.72248590e-01 9.06111240e-01 -1.35691881e+00 1.16718328e+00 -5.07835686e-01 -3.17119509e-01 3.47316414e-01 -6.02456510e-01 7.12986588e-01 8.69432092e-01 -5.71796715e-01 -1.10341263e+00 -3.48740935e-01 -1.13007845e-02 -2.39821985e-01 -7.32329637e-02 4.29180175e-01 -2.62309448e-03 6.41447425e-01 1.04919994e+00 3.29968035e-01 -3.77246499e-01 -7.37187624e-01 7.43648171e-01 5.25590777e-01 6.23007298e-01 -9.43756402e-01 9.02122676e-01 5.89152128e-02 -4.87924427e-01 -4.52845335e-01 -1.15155554e+00 -2.35545844e-01 -2.70479679e-01 5.70424616e-01 5.67697406e-01 -1.02239931e+00 -2.60341287e-01 2.64630079e-01 -1.30419075e+00 -9.14295554e-01 -6.73929155e-01 2.40268912e-02 -5.62859893e-01 1.22186981e-01 -6.87971175e-01 -4.30520386e-01 -8.33493114e-01 -7.18290150e-01 1.30369437e+00 -5.39036505e-02 -5.19002855e-01 -1.21520233e+00 2.58902103e-01 1.63329333e-01 6.05993748e-01 -1.80442989e-01 1.15764868e+00 -7.43644297e-01 -3.95685822e-01 2.04350892e-02 6.43385500e-02 2.21670344e-01 9.13009420e-02 -1.95549577e-01 -9.52584684e-01 -3.34004462e-01 -2.58748025e-01 -3.99022222e-01 1.23816037e+00 3.01096678e-01 1.25799787e+00 -7.12622881e-01 -4.73310530e-01 7.42000520e-01 9.64259982e-01 -1.23337552e-01 3.39698225e-01 2.75732487e-01 5.65570354e-01 2.89364129e-01 3.63623202e-01 4.71861959e-01 5.48711658e-01 5.50675154e-01 -9.67542231e-02 4.11247090e-02 -1.44347057e-01 -8.02504838e-01 6.03985429e-01 6.63407326e-01 4.05271679e-01 -5.53491831e-01 -7.91118085e-01 7.61586010e-01 -1.61488843e+00 -1.00646365e+00 1.44648999e-01 2.12190962e+00 1.47546196e+00 3.30696911e-01 4.96327057e-02 -3.97263169e-01 5.40229619e-01 3.31517249e-01 -6.83339059e-01 -7.17995882e-01 -2.00867623e-01 4.83016014e-01 4.63943958e-01 6.81438148e-01 -9.12190020e-01 1.52039695e+00 7.17556620e+00 1.14883983e+00 -1.09600914e+00 1.43057510e-01 6.37461841e-01 -4.59392011e-01 -7.48272777e-01 1.21254720e-01 -1.17456055e+00 5.75927198e-01 1.23182166e+00 -5.56345224e-01 5.66221237e-01 6.14411116e-01 3.67609352e-01 1.27778083e-01 -1.28489327e+00 4.77035671e-01 -1.00734949e-01 -1.49239826e+00 5.89618385e-01 -1.07740704e-02 9.69724476e-01 3.47904593e-01 4.73587625e-02 7.85977662e-01 6.93783224e-01 -1.10022950e+00 6.58764243e-01 1.74543917e-01 7.60145903e-01 -6.06370091e-01 2.94666886e-01 6.56556308e-01 -7.18325913e-01 -4.35536653e-02 -5.49545288e-01 5.75797632e-02 3.68532270e-01 6.40936911e-01 -1.42613947e+00 1.80572435e-01 1.03469059e-01 4.08813328e-01 -6.31175101e-01 6.31610990e-01 -3.73995155e-01 9.49518859e-01 -2.53299326e-01 1.28652230e-01 5.19227207e-01 7.73725137e-02 4.44235027e-01 1.58254158e+00 3.84087890e-01 -2.12529108e-01 2.59250522e-01 9.73467350e-01 -4.99648720e-01 -3.42505537e-02 -5.37529945e-01 -1.79820210e-01 7.62112319e-01 1.07947338e+00 -7.75013268e-02 -7.27157772e-01 -6.58528507e-02 9.18438852e-01 4.69192743e-01 4.64963049e-01 -5.88046193e-01 -5.23836493e-01 6.24324620e-01 2.53846586e-01 3.70688826e-01 -1.82829306e-01 -3.20883006e-01 -1.23927677e+00 -1.08729247e-02 -7.98393965e-01 4.11972851e-01 -7.36363947e-01 -1.23493648e+00 6.84437573e-01 1.71406209e-01 -8.07644665e-01 -9.64044511e-01 -3.04462582e-01 -9.04384315e-01 1.15913904e+00 -1.67660832e+00 -1.04911423e+00 8.33657011e-02 4.40845400e-01 7.10210085e-01 -1.52817473e-01 9.29299533e-01 -1.56146675e-01 -5.64429820e-01 9.94237781e-01 2.75830358e-01 8.15407932e-02 1.02947736e+00 -1.42904449e+00 1.27475584e+00 7.25753009e-01 1.62782252e-01 8.54869902e-01 7.42913842e-01 -5.95239162e-01 -9.90701616e-01 -1.34014654e+00 1.53817749e+00 -6.23567462e-01 7.65050769e-01 -5.98531127e-01 -8.76993120e-01 9.36071515e-01 3.85031939e-01 -3.39576691e-01 6.75666094e-01 2.66171396e-01 -3.84126931e-01 8.99195597e-02 -8.79384577e-01 8.10179114e-01 1.20739758e+00 -4.22060728e-01 -7.95977294e-01 5.82492411e-01 1.21204126e+00 -5.21098256e-01 -8.02134275e-01 -1.77961051e-01 1.80055052e-01 -4.51911569e-01 8.99976850e-01 -9.31713879e-01 5.43886006e-01 7.20132962e-02 2.42621779e-01 -1.84997141e+00 -3.59693587e-01 -1.19443214e+00 -3.97170693e-01 1.34265685e+00 8.99424314e-01 -7.49331832e-01 8.66689920e-01 4.09206182e-01 -4.92914140e-01 -7.56745994e-01 -5.67132592e-01 -8.99588466e-01 7.81103015e-01 -2.33981356e-01 9.50153232e-01 7.47299850e-01 -1.29881620e-01 6.99706554e-01 -2.09358573e-01 -2.24913865e-01 4.33401972e-01 9.75190550e-02 8.12444031e-01 -9.23620760e-01 -6.91413462e-01 -4.74872917e-01 6.55145288e-01 -1.44954836e+00 4.24537331e-01 -1.25748789e+00 3.71875055e-02 -1.56585240e+00 2.91104540e-02 -6.91330135e-01 -1.58876181e-01 7.64482200e-01 -5.59818447e-01 -1.18590750e-01 1.94201633e-01 1.28952593e-01 -3.62943560e-01 5.65443814e-01 1.25858796e+00 -1.25117719e-01 -4.93565440e-01 8.35059583e-02 -1.30875134e+00 3.01625907e-01 1.01924443e+00 -4.84348983e-01 -6.33097291e-01 -1.02211678e+00 2.21735537e-01 -5.84303103e-02 4.67823818e-02 -6.07930541e-01 1.04537964e-01 -3.20827097e-01 2.21485376e-01 -4.99266982e-01 2.46507719e-01 5.11601083e-02 -1.04493834e-01 1.41216412e-01 -9.03983772e-01 2.06634447e-01 3.42117399e-01 4.14703041e-01 -3.67002711e-02 -3.76116693e-01 5.21688104e-01 -4.73266810e-01 -3.81880343e-01 3.11683029e-01 -4.53871489e-01 7.33569503e-01 4.21881974e-01 -1.77377760e-01 -5.28860807e-01 -2.96119303e-01 -6.24960959e-01 4.86873597e-01 3.79211903e-01 5.16342044e-01 2.24626988e-01 -1.20769286e+00 -9.78686213e-01 2.54348189e-01 2.26537302e-01 5.72945654e-01 1.19737543e-01 3.34253013e-01 -7.83184916e-02 8.74264598e-01 3.54473710e-01 -2.60004938e-01 -6.86952770e-01 4.66936082e-01 1.90586314e-01 -8.59650314e-01 -4.26443279e-01 1.04296732e+00 3.54255825e-01 -6.65831149e-01 7.33601451e-02 -6.07834280e-01 1.28163308e-01 -1.55248633e-02 4.99674022e-01 8.99273753e-02 2.06748962e-01 -9.59407985e-02 3.79387140e-02 1.98294863e-01 -4.77183700e-01 -4.10258234e-01 1.15273666e+00 2.58441903e-02 1.00890554e-01 3.08384538e-01 1.05124819e+00 1.15353530e-02 -1.29303086e+00 -6.00716531e-01 9.26845148e-02 -1.51200563e-01 -2.86392607e-02 -1.06665862e+00 -7.04651356e-01 9.25071001e-01 -3.08423817e-01 -1.15297355e-01 8.27876389e-01 3.96619327e-02 1.04979992e+00 8.57417643e-01 2.34673351e-01 -1.09883094e+00 1.30856007e-01 9.81648684e-01 8.52299750e-01 -8.07797074e-01 -2.94577062e-01 -2.55544037e-02 -7.45862305e-01 7.75722146e-01 6.83090448e-01 1.16047427e-01 2.75479615e-01 3.52421016e-01 -1.16558917e-01 1.56757504e-01 -1.63619542e+00 2.13823900e-01 7.90904984e-02 6.12953782e-01 6.88040495e-01 1.12569720e-01 1.00256868e-01 5.04893959e-01 -8.23744655e-01 1.38478324e-01 4.67736602e-01 8.06108952e-01 -5.35704255e-01 -1.39364207e+00 -2.37570986e-01 7.49325395e-01 -4.83851492e-01 -7.31114209e-01 -3.40508431e-01 5.94435275e-01 -8.36002547e-03 7.41086781e-01 2.44001970e-01 -4.67499951e-03 3.33561778e-01 3.89588922e-01 3.32130879e-01 -1.09179091e+00 -1.10462844e+00 -8.61868635e-02 2.89634109e-01 -3.10689956e-01 3.20350736e-01 -4.87098604e-01 -1.16067791e+00 -4.30035174e-01 -6.86295182e-02 4.15623486e-01 4.58692491e-01 7.03349888e-01 7.86888421e-01 5.10653198e-01 5.23628831e-01 -8.38588059e-01 -1.20444012e+00 -1.25330055e+00 -1.81536078e-01 2.18509197e-01 4.01097059e-01 -1.80984452e-01 -3.29111189e-01 7.00100064e-02]

[11.549721717834473, 8.944610595703125]

78e6637c-3ebf-4397-ae4a-0b694de6dfe4

melon-nerf-with-unposed-images-using

2303.08096

https://arxiv.org/abs/2303.08096v1

https://arxiv.org/pdf/2303.08096v1.pdf

MELON: NeRF with Unposed Images Using Equivalence Class Estimation

Neural radiance fields enable novel-view synthesis and scene reconstruction with photorealistic quality from a few images, but require known and accurate camera poses. Conventional pose estimation algorithms fail on smooth or self-similar scenes, while methods performing inverse rendering from unposed views require a rough initialization of the camera orientations. The main difficulty of pose estimation lies in real-life objects being almost invariant under certain transformations, making the photometric distance between rendered views non-convex with respect to the camera parameters. Using an equivalence relation that matches the distribution of local minima in camera space, we reduce this space to its quotient set, in which pose estimation becomes a more convex problem. Using a neural-network to regularize pose estimation, we demonstrate that our method - MELON - can reconstruct a neural radiance field from unposed images with state-of-the-art accuracy while requiring ten times fewer views than adversarial approaches.

['Dmitry Lagun', 'Gordon Wetzstein', 'Matan Sela', 'Mark Matthews', 'Axel Levy']

2023-03-14

['inverse-rendering']

['computer-vision']

[ 5.94765306e-01 1.89245448e-01 3.07198763e-01 -3.22703868e-01 -7.09301174e-01 -1.02465177e+00 5.35436988e-01 -6.72182500e-01 -2.48214826e-01 4.92488950e-01 5.27315140e-02 7.61813149e-02 1.08777188e-01 -7.35565960e-01 -1.22708094e+00 -6.45215511e-01 4.79445666e-01 3.89937371e-01 -1.87730372e-01 -2.24813998e-01 2.22464204e-01 7.81050086e-01 -1.36805189e+00 -1.29395723e-01 3.66476715e-01 8.68127584e-01 -2.52471827e-02 8.85151923e-01 4.77796376e-01 6.58621311e-01 -4.77248579e-01 -7.48653769e-01 9.27345276e-01 -3.57351303e-01 -4.90271538e-01 4.34580773e-01 1.33598256e+00 -7.99668193e-01 -5.79014838e-01 1.19703555e+00 3.26225191e-01 1.81894660e-01 6.08155727e-01 -9.26724494e-01 -6.05987310e-01 -2.40345538e-01 -5.13703585e-01 -2.14754775e-01 6.38344109e-01 -1.39692456e-01 7.08271921e-01 -8.62928212e-01 7.31699884e-01 1.08157933e+00 7.97478497e-01 5.75018644e-01 -1.55531180e+00 -3.09025913e-01 -4.61591333e-02 -1.84027568e-01 -1.48874199e+00 -6.11428201e-01 1.02258778e+00 -2.98666924e-01 6.38558984e-01 6.45643055e-01 6.61694705e-01 8.93240631e-01 2.57001281e-01 1.35738492e-01 1.10339272e+00 -3.80095482e-01 1.66251898e-01 5.81434593e-02 -6.94167614e-01 7.58852482e-01 7.05761164e-02 2.57489502e-01 -4.05861855e-01 7.17523545e-02 1.36332607e+00 1.26419172e-01 -7.22023904e-01 -9.08134460e-01 -1.25292516e+00 5.96936226e-01 4.70117390e-01 -2.61078089e-01 -1.86537281e-01 1.92941785e-01 -3.36085558e-01 1.71679854e-01 4.58841771e-01 7.08768249e-01 -3.36812764e-01 2.21127138e-01 -5.75365543e-01 1.04463972e-01 7.77556539e-01 1.04172468e+00 9.18781459e-01 3.18214178e-01 6.56687438e-01 4.65398610e-01 -4.93737683e-03 8.32389355e-01 1.02374211e-01 -1.60392332e+00 4.31810826e-01 2.06795603e-01 2.58728534e-01 -1.13981664e+00 -1.35937884e-01 -2.92039096e-01 -8.26679289e-01 6.79250360e-01 5.98103881e-01 -9.19453502e-02 -7.46772766e-01 1.72924936e+00 3.94963026e-01 -2.93976665e-02 4.62758401e-03 9.54527974e-01 3.63023430e-01 6.00547373e-01 -8.79958272e-01 -3.11872363e-01 1.10058546e+00 -5.03542066e-01 -4.26698267e-01 -4.44141150e-01 -2.55809754e-01 -8.99647534e-01 9.34883773e-01 6.09470308e-01 -1.35251439e+00 -3.45082581e-01 -1.19600606e+00 -2.58732229e-01 -4.89559397e-03 -2.55081385e-01 4.26658392e-01 8.17261875e-01 -1.00981140e+00 6.05161369e-01 -6.61915421e-01 9.56921354e-02 8.26757178e-02 4.47702169e-01 -8.07501256e-01 -6.04186878e-02 -6.00721538e-01 9.70755696e-01 -5.49977869e-02 1.04431294e-01 -7.55913138e-01 -9.78190541e-01 -1.01627910e+00 -2.11631253e-01 4.80876088e-01 -8.48790884e-01 1.02238750e+00 -1.11037731e+00 -1.93758011e+00 1.14265430e+00 -2.66809110e-02 -4.90634963e-02 6.91158891e-01 -4.31320012e-01 -2.12844476e-01 3.14247757e-01 -2.07213610e-01 5.99036098e-01 1.38350844e+00 -1.57857549e+00 -6.50269017e-02 -5.40833652e-01 4.04191792e-01 6.92841589e-01 7.29380082e-03 -4.21922028e-01 -5.89528799e-01 -5.41986823e-01 4.64317977e-01 -1.00955844e+00 -2.35798299e-01 4.50637400e-01 -3.99978548e-01 6.74343109e-01 8.16323817e-01 -7.78984427e-01 3.76217961e-01 -1.95567489e+00 5.07829964e-01 2.47694999e-02 2.58169383e-01 -1.73947066e-01 -1.01605937e-01 1.08181521e-01 -4.05926615e-01 -3.19023818e-01 -5.51047996e-02 -2.53499418e-01 -3.01895440e-01 -1.34719433e-02 -4.51022029e-01 9.71585453e-01 -1.65474221e-01 7.03820825e-01 -5.52948654e-01 6.32957295e-02 6.58443809e-01 7.21881390e-01 -6.97928548e-01 4.36648160e-01 -1.02863744e-01 6.98801637e-01 -1.57675177e-01 3.62670571e-01 8.90625775e-01 -7.55051970e-02 2.98478035e-03 -5.72067261e-01 2.29212660e-02 -1.47467136e-01 -1.24393010e+00 1.87162781e+00 -7.48801172e-01 8.76155198e-01 3.06643367e-01 -5.70362389e-01 8.06879818e-01 3.05797700e-02 6.14287078e-01 -5.17321944e-01 1.50607303e-01 -1.61795199e-01 -3.89564514e-01 -1.92511201e-01 4.08929080e-01 -4.90222305e-01 1.61928937e-01 2.59020984e-01 -1.30807817e-01 -9.26469028e-01 -3.74570727e-01 -1.32394433e-01 7.27550983e-01 4.60280657e-01 2.47741371e-01 1.72459111e-01 4.24715489e-01 -3.87422323e-01 3.94053429e-01 2.04018101e-01 3.62998545e-01 1.32311583e+00 2.11133227e-01 -8.23390305e-01 -1.32250714e+00 -1.40808535e+00 -2.11324900e-01 3.44078422e-01 3.57228458e-01 3.01189534e-02 -1.05734503e+00 -3.10276419e-01 -2.08702505e-01 7.42080688e-01 -6.56422496e-01 -1.04463674e-01 -7.46789753e-01 -3.65448177e-01 4.40736860e-02 2.30783552e-01 4.71491784e-01 -6.26407325e-01 -7.44997442e-01 -1.64180040e-01 -1.73734546e-01 -1.32906795e+00 -7.14697123e-01 1.09369699e-02 -7.73697376e-01 -1.08837998e+00 -9.54923689e-01 -5.51887453e-01 1.02037048e+00 4.07417625e-01 1.34396994e+00 -4.73043114e-01 -3.75473768e-01 6.17619455e-01 3.73863310e-01 -1.69376150e-01 -3.20221245e-01 -5.08190930e-01 2.19109327e-01 2.14092791e-01 -2.47063354e-01 -9.47725058e-01 -7.27414846e-01 6.34264052e-01 -9.19784367e-01 2.78524905e-01 2.05554262e-01 6.49812937e-01 7.57186890e-01 -1.11263707e-01 -3.00398856e-01 -7.11927354e-01 -8.36867467e-03 2.71462262e-01 -1.20371640e+00 2.21984401e-01 -2.38389343e-01 8.68842378e-02 8.59468400e-01 -5.47109425e-01 -1.23266292e+00 4.57514256e-01 2.75645912e-01 -8.61346841e-01 2.70366929e-02 -2.66983896e-01 -4.07301277e-01 -6.46524787e-01 9.30675030e-01 1.91823214e-01 3.03281751e-02 -1.70972288e-01 5.84438741e-01 -5.19458055e-02 9.81498480e-01 -3.48912328e-01 1.26658177e+00 8.87714863e-01 4.86265510e-01 -9.47427511e-01 -1.00189793e+00 -1.29617810e-01 -8.33893657e-01 -3.09335053e-01 9.42136347e-01 -1.05987000e+00 -9.63474929e-01 3.92146736e-01 -1.12961459e+00 -2.26112410e-01 -4.31867599e-01 4.76968765e-01 -9.43834722e-01 4.22755271e-01 -3.08096319e-01 -5.17518878e-01 6.05463870e-02 -1.12431920e+00 1.28739214e+00 4.25790288e-02 3.21651399e-02 -9.98667777e-01 2.11089462e-01 7.29635417e-01 4.42346744e-02 5.35344541e-01 5.23175120e-01 4.51678723e-01 -1.17979860e+00 -3.13079745e-01 5.23506962e-02 5.09796202e-01 1.35851130e-01 -1.62335709e-01 -1.19512093e+00 -4.04527098e-01 5.62500358e-01 -2.54310489e-01 3.85031313e-01 5.45809031e-01 1.05600703e+00 -3.66944879e-01 1.49716094e-01 1.49172091e+00 1.71556270e+00 6.74718171e-02 8.40640545e-01 1.45502418e-01 1.07162225e+00 5.96216381e-01 2.17193857e-01 2.03194812e-01 -2.23765522e-02 8.98324311e-01 8.70590448e-01 -1.96900278e-01 4.00338210e-02 -2.41000846e-01 2.99340039e-01 5.45538127e-01 -2.90206134e-01 -1.28652349e-01 -5.36547244e-01 3.97733413e-02 -1.29240263e+00 -9.41598952e-01 7.26294070e-02 2.66985297e+00 5.61983347e-01 -9.15820077e-02 -2.53100306e-01 -2.03892037e-01 4.55328971e-01 3.98460180e-01 -8.19038749e-01 -1.59398496e-01 -1.78367361e-01 1.90241251e-03 9.51304138e-01 7.81641245e-01 -9.51395214e-01 6.46692812e-01 6.79685497e+00 2.43361264e-01 -1.11739743e+00 -1.55526161e-01 6.57295644e-01 -2.78436750e-01 -5.55863082e-01 5.86926565e-03 -4.10387456e-01 1.95729751e-02 4.59448814e-01 -3.99445407e-02 1.05531287e+00 7.20819056e-01 -2.44134530e-01 -6.95182756e-02 -1.32901895e+00 1.35051954e+00 6.46725416e-01 -1.43141186e+00 2.60138158e-02 3.16517800e-01 1.09261167e+00 -5.12932688e-02 4.30708677e-01 -4.29735422e-01 1.79747492e-01 -1.09165001e+00 7.62423158e-01 7.28807509e-01 1.10173786e+00 -9.34638321e-01 1.02999762e-01 3.53926688e-01 -7.69715846e-01 1.17156565e-01 -5.81027627e-01 1.59839511e-01 2.71285146e-01 3.00817311e-01 -5.05171001e-01 3.98110926e-01 5.13970554e-01 5.59865892e-01 -3.10333908e-01 5.30398071e-01 -1.06942341e-01 -2.11540848e-01 -4.32393044e-01 4.56414729e-01 -2.36397788e-01 -7.49395013e-01 6.57017171e-01 3.12779397e-01 3.32445651e-01 2.68250763e-01 -2.16669276e-01 9.41669464e-01 -3.44658226e-01 -2.71060318e-01 -9.53970611e-01 5.27968109e-01 -1.71832331e-02 1.39201307e+00 -4.98721004e-01 1.47337779e-01 -3.83074194e-01 1.40370047e+00 1.52063847e-01 5.94941080e-01 -6.78888381e-01 -6.63185343e-02 7.43649006e-01 3.32906693e-01 1.89557984e-01 -3.37347537e-01 -1.38963535e-01 -1.56944478e+00 3.61105621e-01 -9.65889871e-01 -2.25156158e-01 -1.35993278e+00 -8.44165742e-01 6.92059040e-01 1.03821881e-01 -1.56133533e+00 -5.73001087e-01 -9.38620269e-01 -3.81121188e-01 8.88086319e-01 -1.02756763e+00 -1.16285157e+00 -4.26460266e-01 6.56918883e-01 4.94325459e-01 -1.19897321e-01 9.89072144e-01 -8.44900683e-02 -1.53351635e-01 3.96598995e-01 2.86739528e-01 -6.56097978e-02 4.33562666e-01 -1.32104564e+00 5.71808517e-01 9.57637131e-01 3.97605330e-01 4.35463935e-01 7.17529118e-01 -1.83414549e-01 -1.83022809e+00 -7.17996180e-01 1.70031607e-01 -9.03817534e-01 2.66971260e-01 -5.86323917e-01 -4.80231732e-01 8.14196885e-01 1.87205881e-01 3.07409585e-01 2.19361320e-01 -2.71861732e-01 -4.64842826e-01 -2.49305770e-01 -1.09563243e+00 8.38489056e-01 1.01233578e+00 -9.60300028e-01 -2.61290878e-01 5.06933212e-01 5.69173515e-01 -8.14821005e-01 -8.32015514e-01 1.13000475e-01 7.06070423e-01 -1.24247062e+00 1.48818707e+00 -2.53261119e-01 4.82407749e-01 -3.59652281e-01 -3.49474132e-01 -1.36931407e+00 -2.11652949e-01 -9.62869823e-01 1.29234672e-01 6.93651974e-01 7.28155747e-02 -5.46730101e-01 1.04626477e+00 7.00204909e-01 1.81306079e-01 -4.32481855e-01 -8.74863803e-01 -6.28151417e-01 -1.24659777e-01 -6.78687617e-02 2.33003199e-01 8.65082443e-01 -7.02959239e-01 4.87217903e-01 -5.97673178e-01 4.87614125e-01 1.03621745e+00 2.47538939e-01 1.06776881e+00 -9.97030437e-01 -6.22478604e-01 -5.39238863e-02 -4.97204989e-01 -1.24109554e+00 3.09474766e-01 -4.17932749e-01 4.62159738e-02 -1.01431155e+00 -2.43853014e-02 4.92207706e-02 5.30740261e-01 -9.22892839e-02 2.10967734e-01 5.85177779e-01 1.59402460e-01 1.37440935e-01 -7.51059055e-02 4.90547061e-01 1.51019788e+00 3.04068327e-02 2.07780935e-02 2.07760066e-01 -4.72360641e-01 1.17608142e+00 2.84323990e-01 -2.28236243e-01 -6.59795344e-01 -7.20109999e-01 6.87288463e-01 4.36105102e-01 5.11035919e-01 -1.07050264e+00 4.33408432e-02 -2.06073761e-01 7.07823396e-01 -5.28793156e-01 1.03514683e+00 -1.14736676e+00 8.09795558e-01 3.19671519e-02 -2.01384023e-01 2.13012204e-01 1.50097813e-02 6.38850749e-01 5.99947125e-02 -2.27595285e-01 8.69534791e-01 -3.77295941e-01 -3.45877469e-01 4.08011556e-01 9.28981751e-02 1.62563756e-01 9.27306652e-01 -4.90006655e-01 -1.93401083e-01 -9.40297484e-01 -5.25972545e-01 -5.44059694e-01 1.11496782e+00 1.82867885e-01 7.19005585e-01 -1.35081494e+00 -5.50913990e-01 5.75530648e-01 -8.67525712e-02 3.97437632e-01 3.27789187e-01 2.56366223e-01 -1.17302370e+00 2.51268689e-02 -2.90390342e-01 -7.38647819e-01 -1.31153607e+00 7.12429881e-01 8.08757544e-01 1.38768837e-01 -6.75948858e-01 8.31050515e-01 7.86023676e-01 -7.12653100e-01 4.00556438e-02 -5.98052219e-02 2.32687905e-01 -3.93106133e-01 3.94196600e-01 1.98065802e-01 6.04924858e-02 -8.37901056e-01 -8.79994854e-02 1.10691202e+00 2.10215613e-01 -4.35385913e-01 1.34705043e+00 -1.26528755e-01 5.94810508e-02 1.13490656e-01 1.52803242e+00 4.15366054e-01 -2.06385946e+00 -4.05754559e-02 -8.60350490e-01 -8.39733303e-01 2.53173441e-01 -4.56796318e-01 -1.29215169e+00 7.66276300e-01 5.29075563e-01 -2.12771874e-02 1.17351902e+00 -2.22499520e-01 4.42877114e-01 6.67938828e-01 3.73728424e-01 -7.66456604e-01 8.22095126e-02 2.48153195e-01 1.22075558e+00 -1.12829626e+00 3.36101800e-01 -6.25555396e-01 -4.12899673e-01 1.18056500e+00 5.62239408e-01 -3.99355382e-01 5.09875238e-01 2.56970644e-01 3.14041018e-01 -4.26377617e-02 -2.12042496e-01 4.89975959e-01 6.04584515e-01 6.51902080e-01 4.17249314e-02 -5.38209155e-02 6.67414308e-01 -5.60204446e-01 -5.19506156e-01 -6.40306532e-01 6.20018303e-01 4.20969307e-01 1.27700418e-01 -8.87999654e-01 -6.47789180e-01 -2.41157860e-02 -2.93872178e-01 -9.93727893e-02 -3.12477350e-01 8.22891414e-01 -9.38685909e-02 4.89705831e-01 3.58353078e-01 -1.57293841e-01 5.11760473e-01 -3.99508238e-01 1.11258221e+00 -3.89558494e-01 -1.83889881e-01 1.93715721e-01 -2.55577210e-02 -8.76863480e-01 -4.48978603e-01 -6.61800027e-01 -6.08755529e-01 -4.31228876e-01 -2.76326507e-01 -4.79649156e-01 6.85211182e-01 7.35645056e-01 -1.21667860e-02 2.33047053e-01 1.08445835e+00 -1.36896551e+00 -6.39917314e-01 -3.49455804e-01 -5.72397530e-01 5.18267095e-01 5.51984251e-01 -3.02813441e-01 -4.28618044e-01 3.24519724e-01]

[9.25918197631836, -2.9649791717529297]

cc1f65c6-b9af-4c21-bfa5-c05d1bebaef7

video-text-as-game-players-hierarchical

2303.14369

https://arxiv.org/abs/2303.14369v1

https://arxiv.org/pdf/2303.14369v1.pdf

Video-Text as Game Players: Hierarchical Banzhaf Interaction for Cross-Modal Representation Learning

Contrastive learning-based video-language representation learning approaches, e.g., CLIP, have achieved outstanding performance, which pursue semantic interaction upon pre-defined video-text pairs. To clarify this coarse-grained global interaction and move a step further, we have to encounter challenging shell-breaking interactions for fine-grained cross-modal learning. In this paper, we creatively model video-text as game players with multivariate cooperative game theory to wisely handle the uncertainty during fine-grained semantic interaction with diverse granularity, flexible combination, and vague intensity. Concretely, we propose Hierarchical Banzhaf Interaction (HBI) to value possible correspondence between video frames and text words for sensitive and explainable cross-modal contrast. To efficiently realize the cooperative game of multiple video frames and multiple text words, the proposed method clusters the original video frames (text words) and computes the Banzhaf Interaction between the merged tokens. By stacking token merge modules, we achieve cooperative games at different semantic levels. Extensive experiments on commonly used text-video retrieval and video-question answering benchmarks with superior performances justify the efficacy of our HBI. More encouragingly, it can also serve as a visualization tool to promote the understanding of cross-modal interaction, which have a far-reaching impact on the community. Project page is available at https://jpthu17.github.io/HBI/.

['Jie Chen', 'Li Yuan', 'Xiangyang Ji', 'Chang Liu', 'Shangxuan Tian', 'Pengfei Xiong', 'Jinfa Huang', 'Peng Jin']

2023-03-25

http://openaccess.thecvf.com//content/CVPR2023/html/Jin_Video-Text_As_Game_Players_Hierarchical_Banzhaf_Interaction_for_Cross-Modal_Representation_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Jin_Video-Text_As_Game_Players_Hierarchical_Banzhaf_Interaction_for_Cross-Modal_Representation_CVPR_2023_paper.pdf

cvpr-2023-1

['video-question-answering', 'video-retrieval']

['computer-vision', 'computer-vision']

[-2.47621492e-01 -2.26136908e-01 -8.66484195e-02 -9.39595327e-02 -9.68055248e-01 -4.83812183e-01 5.97398937e-01 -1.10248074e-01 -2.54538298e-01 3.27982843e-01 3.82912189e-01 -1.18528731e-01 -6.09930873e-01 -4.83760774e-01 -6.29653811e-01 -7.87474334e-01 -1.71604306e-01 3.53623122e-01 2.04950169e-01 -3.20526272e-01 3.12220126e-01 -2.24698320e-01 -1.84232509e+00 9.55099165e-01 7.56433547e-01 8.67127717e-01 5.55057764e-01 5.95842659e-01 -6.63793385e-01 1.21285212e+00 -4.78099316e-01 -5.04311025e-01 -5.92611381e-04 -4.29144472e-01 -9.20985639e-01 2.29727238e-01 1.69762090e-01 -4.49201576e-02 -3.52090061e-01 1.11224926e+00 3.19778591e-01 4.32960153e-01 5.74208498e-01 -1.55270672e+00 -7.12056756e-01 9.98607695e-01 -7.81397939e-01 2.67303348e-01 5.91330230e-01 2.48511955e-02 1.33171284e+00 -1.09880078e+00 4.75049108e-01 1.51293993e+00 2.78020978e-01 3.89623284e-01 -6.89445674e-01 -8.08138847e-01 5.14320254e-01 9.04274106e-01 -1.70028269e+00 -7.42070749e-02 8.51568878e-01 -5.73353231e-01 7.15247810e-01 4.79875743e-01 6.67320967e-01 1.00539279e+00 6.23453930e-02 1.30245292e+00 6.61325395e-01 -3.10294598e-01 -3.52060087e-02 -2.13495359e-01 6.27810881e-02 7.35112309e-01 -1.50992841e-01 -2.73850769e-01 -7.52273321e-01 1.59435689e-01 7.22334564e-01 3.12775105e-01 -3.20561916e-01 -1.95337012e-01 -1.34358442e+00 7.67953992e-01 2.06171632e-01 5.17125428e-01 -2.04683781e-01 2.94453055e-01 6.24637008e-01 2.93950737e-01 3.91837507e-01 5.50695397e-02 -1.94088683e-01 -4.10257429e-01 -9.66858566e-01 1.87142193e-01 4.04244602e-01 1.02832210e+00 7.46529698e-01 -1.75168380e-01 -4.29458201e-01 8.42052579e-01 4.07437265e-01 9.89579260e-02 4.24397111e-01 -1.11361706e+00 4.51011777e-01 5.77454984e-01 -2.21331105e-01 -1.22851264e+00 -2.04698384e-01 -2.24009119e-02 -1.02556181e+00 -1.36220500e-01 3.32636923e-01 1.60659969e-01 -5.26063383e-01 1.54932702e+00 1.70893684e-01 6.33834243e-01 -1.39748648e-01 1.10610402e+00 1.10000980e+00 1.03118539e+00 3.22284669e-01 -3.37845296e-01 1.62402213e+00 -1.11703146e+00 -7.93358266e-01 1.73807606e-01 7.09887028e-01 -8.00197363e-01 1.41505265e+00 5.23714542e-01 -1.26246989e+00 -6.36541486e-01 -7.13798940e-01 -2.11944476e-01 -3.72096270e-01 -1.17583618e-01 4.92303878e-01 1.73530191e-01 -1.13216221e+00 1.56074017e-01 -4.39742237e-01 -1.27074182e-01 2.96023130e-01 1.37003139e-01 -1.51996300e-01 -2.86059618e-01 -1.57457864e+00 1.18663512e-01 5.74411869e-01 6.59303591e-02 -5.70091605e-01 -7.25516796e-01 -6.96462870e-01 1.52743801e-01 9.08301353e-01 -7.02031136e-01 9.53832328e-01 -1.02383161e+00 -1.19180787e+00 8.56943309e-01 7.06544146e-02 -9.12609696e-02 4.49882597e-01 -1.75638810e-01 -2.16360077e-01 4.34969693e-01 1.06775746e-01 7.16416717e-01 9.36539710e-01 -1.23466694e+00 -7.72333205e-01 -2.10222855e-01 5.17530799e-01 7.33331978e-01 -3.22702676e-01 2.71391720e-01 -1.05928504e+00 -8.52502227e-01 -1.11469835e-01 -5.90563118e-01 -8.32203124e-03 -9.88109857e-02 -7.07371160e-02 -5.45711696e-01 6.22437775e-01 -3.98382753e-01 1.45267236e+00 -2.41681981e+00 4.63986188e-01 -8.15021694e-02 6.21954501e-01 -7.03447387e-02 -1.72586739e-01 3.45607996e-01 2.61678379e-02 2.33347923e-01 3.38894784e-01 -2.50220329e-01 1.57720089e-01 1.41224027e-01 -2.16811225e-01 1.88669100e-01 -9.36772004e-02 1.05984211e+00 -9.76652980e-01 -1.08322573e+00 4.68347222e-01 4.31479603e-01 -7.16982663e-01 1.64685234e-01 -2.80347288e-01 2.26678610e-01 -7.18781173e-01 4.98473793e-01 5.54539084e-01 -4.30829585e-01 7.94103835e-03 -5.69540203e-01 2.79887170e-02 -4.02774721e-01 -1.37997091e+00 1.83971739e+00 -1.46482199e-01 6.15639806e-01 9.25540477e-02 -1.18809509e+00 6.37662232e-01 8.42255130e-02 7.74411857e-01 -7.80384362e-01 2.17788696e-01 -1.89316168e-01 -3.14385504e-01 -7.70835400e-01 6.76711142e-01 -2.49128137e-02 -3.18167239e-01 2.24583402e-01 1.74157858e-01 -1.79155190e-02 2.46234611e-01 6.08256400e-01 5.76248884e-01 7.78053626e-02 3.62703741e-01 -3.36095572e-01 7.08608389e-01 -2.44375095e-01 1.73961967e-01 8.20912540e-01 -2.41035193e-01 6.98500037e-01 4.99620706e-01 -3.21042895e-01 -5.61879218e-01 -9.91553962e-01 2.28327408e-01 1.78332710e+00 8.00458848e-01 -8.68186951e-01 -6.46272779e-01 -2.14254498e-01 -4.21617687e-01 2.37669736e-01 -4.91479576e-01 -1.13141187e-01 -3.77338320e-01 -4.78471756e-01 4.84175950e-01 3.07158679e-01 4.97073084e-01 -1.07375097e+00 -2.03643009e-01 4.24354561e-02 -8.33235562e-01 -1.03389978e+00 -7.99367487e-01 -1.70877486e-01 -1.81724101e-01 -1.05985165e+00 -5.86689472e-01 -9.99932885e-01 1.52535364e-01 7.45860696e-01 1.25756085e+00 4.34866160e-01 -3.47699106e-01 8.29054534e-01 -9.80028927e-01 4.04420011e-02 9.46513563e-02 -2.25446805e-01 -9.74721089e-02 9.59417745e-02 4.12394255e-01 -3.23867172e-01 -5.91349840e-01 4.68456417e-01 -1.24786603e+00 4.74893808e-01 1.34688959e-01 9.94453013e-01 6.75029635e-01 3.17445964e-01 1.57259628e-01 -3.57138425e-01 6.40385389e-01 -7.53221929e-01 -2.08092168e-01 5.31060815e-01 -5.59715517e-02 -6.34901747e-02 5.42068124e-01 -5.97016990e-01 -1.18157983e+00 -5.88829041e-01 1.07653067e-01 -8.28507721e-01 -6.13854155e-02 6.83714449e-01 -3.35975468e-01 2.64431506e-01 3.41872185e-01 2.81931609e-01 -2.23509684e-01 1.55736050e-02 6.39314532e-01 5.71369708e-01 3.75861764e-01 -1.10310161e+00 4.84391004e-01 3.85697782e-01 -4.38784033e-01 -8.04512143e-01 -6.14476442e-01 -6.42137825e-01 -3.11169207e-01 -7.07885504e-01 1.15306318e+00 -1.26475751e+00 -1.09014452e+00 4.98118848e-01 -9.84974086e-01 -3.20120245e-01 -1.15157746e-01 4.55382973e-01 -6.20522857e-01 7.11359084e-01 -6.60253882e-01 -6.17928565e-01 -1.65030025e-02 -1.09236193e+00 1.24478102e+00 3.47116411e-01 4.95842621e-02 -1.07154071e+00 -2.80586749e-01 8.07617664e-01 4.36173193e-02 -2.30237186e-01 6.22267902e-01 -4.77231383e-01 -8.66539657e-01 3.76860231e-01 -3.69248927e-01 -4.52630781e-02 -1.16002180e-01 2.06567511e-01 -5.54012895e-01 -2.62076199e-01 -1.38929605e-01 -4.98150706e-01 9.29410160e-01 6.06757402e-01 1.51893234e+00 -1.33103743e-01 -4.41936664e-02 4.76467490e-01 1.19805670e+00 1.75786227e-01 6.86344564e-01 4.33577061e-01 9.52392340e-01 6.15827203e-01 9.10647213e-01 8.75818610e-01 6.70016825e-01 6.56563282e-01 3.30597788e-01 1.04809172e-01 -5.84848635e-02 -1.63551152e-01 4.13674861e-01 1.12524855e+00 -2.86315501e-01 -5.83493590e-01 -8.69584560e-01 3.93373191e-01 -2.20994902e+00 -1.44532204e+00 -6.04039915e-02 1.57502389e+00 5.61199009e-01 -3.67607698e-02 -8.11599866e-02 -7.62869567e-02 8.80465388e-01 5.51412225e-01 -2.36143962e-01 1.29248193e-02 -2.36912191e-01 -1.31175190e-01 -1.02812694e-02 5.03090322e-01 -9.55742478e-01 1.16268253e+00 4.71574354e+00 1.81230950e+00 -8.27188969e-01 2.28873700e-01 6.34988844e-01 -2.49675632e-01 -5.76273143e-01 -2.31854424e-01 -4.57863629e-01 5.14306903e-01 4.35238481e-01 -4.55218375e-01 6.24638975e-01 4.90421146e-01 2.83627450e-01 -5.84051870e-02 -7.30675578e-01 1.65246177e+00 2.35973611e-01 -1.62095726e+00 4.43851709e-01 -2.86132127e-01 5.63196361e-01 -2.91898340e-01 1.80745780e-01 6.05753899e-01 2.32789546e-01 -8.71487319e-01 8.55512440e-01 6.28456473e-01 7.32481837e-01 -7.11813927e-01 5.18938839e-01 2.52801627e-01 -1.87307727e+00 -5.75793870e-02 -2.62038112e-01 -1.18057944e-01 1.52460873e-01 2.69108653e-01 -5.78172319e-02 8.65011632e-01 1.02202415e+00 9.08225775e-01 -4.68531847e-01 6.58134997e-01 2.67625004e-01 3.35359633e-01 -7.14915842e-02 -4.15617637e-02 3.85657310e-01 -3.09016436e-01 4.49943036e-01 1.35328722e+00 3.08781236e-01 5.83260119e-01 4.96000975e-01 5.87108254e-01 -2.41961721e-02 3.22696000e-01 -2.53222466e-01 1.27449036e-02 5.83338737e-01 1.21634960e+00 -8.24112713e-01 -4.96662527e-01 -5.66958547e-01 9.19866800e-01 2.86323041e-01 4.67112482e-01 -1.21360755e+00 -8.61885622e-02 7.85268247e-01 -1.10023536e-01 2.55833417e-01 -1.09172910e-01 1.20143080e-02 -1.57017148e+00 3.32935080e-02 -9.74490941e-01 8.39080274e-01 -1.10964572e+00 -1.33033264e+00 4.48007405e-01 2.36748517e-01 -1.41908360e+00 -8.01353753e-02 -5.80830455e-01 -4.54492003e-01 3.15206081e-01 -1.13303053e+00 -1.21395051e+00 -5.46563208e-01 1.31272995e+00 1.08845651e+00 -1.76373452e-01 3.11866254e-01 5.31825483e-01 -3.11467886e-01 5.79629719e-01 1.13983445e-01 -1.77902058e-02 7.10780978e-01 -9.89248872e-01 -2.41762161e-01 6.41915381e-01 4.46033061e-01 2.33947784e-01 5.91308951e-01 -4.35117394e-01 -1.38788188e+00 -1.05626893e+00 4.10888165e-01 -2.94968307e-01 9.43013370e-01 -2.68362820e-01 -9.31556106e-01 5.40549695e-01 3.41338366e-01 -1.01916648e-01 6.15235567e-01 9.48186219e-03 -3.28252614e-01 5.25989644e-02 -6.03482842e-01 8.27237606e-01 1.15380836e+00 -8.57084453e-01 -6.84078932e-01 2.75559634e-01 1.01038575e+00 -4.46392894e-01 -6.48768842e-01 1.37801573e-01 4.46308851e-01 -1.09912753e+00 1.23049974e+00 -5.40429711e-01 5.52753866e-01 -3.38529736e-01 -3.94891918e-01 -9.12265897e-01 -4.67071742e-01 -6.16615832e-01 6.76558390e-02 1.32510793e+00 -9.70730409e-02 -6.47714734e-02 5.74552357e-01 4.06558245e-01 -1.81780666e-01 -6.33844137e-01 -8.38209391e-01 -5.52492917e-01 1.25730053e-01 -8.66836905e-01 3.36788535e-01 1.05392230e+00 4.13094699e-01 1.55172184e-01 -7.31997907e-01 5.53361401e-02 5.04897714e-01 4.90969941e-02 5.31725883e-01 -1.02396190e+00 -3.31718057e-01 -8.06547701e-01 -4.84563321e-01 -1.30818498e+00 3.51928890e-01 -8.77599895e-01 -1.15947826e-02 -1.34344053e+00 6.22937918e-01 -9.06502232e-02 -4.87158597e-01 2.29131371e-01 -4.22498971e-01 3.54244262e-01 5.05881429e-01 3.41695487e-01 -1.48017216e+00 6.04547560e-01 1.43186545e+00 -3.18049073e-01 -9.30954814e-02 -4.81471211e-01 -7.29733825e-01 5.89695811e-01 6.82583094e-01 -1.16400234e-01 -7.00921655e-01 -4.43701923e-01 4.08005506e-01 2.20716968e-01 4.02757972e-01 -7.31440187e-01 5.33212721e-01 -3.75859112e-01 -4.03885432e-02 -5.49453020e-01 3.64100754e-01 -6.46323681e-01 2.95054108e-01 6.82488456e-03 -4.95505154e-01 -1.37994885e-01 6.42983019e-02 6.11717880e-01 -6.20338261e-01 6.02254160e-02 4.21287835e-01 -2.01058149e-01 -1.18704343e+00 4.81888711e-01 -5.89072764e-01 4.76531386e-01 1.11740065e+00 -3.97941232e-01 -1.95293620e-01 -6.28627539e-01 -8.62655640e-01 6.29664063e-01 1.05167143e-01 7.24230766e-01 7.37088323e-01 -1.46654427e+00 -6.73603773e-01 -8.06563068e-03 2.08532751e-01 -4.43255343e-02 1.09554803e+00 7.98864365e-01 -1.98775560e-01 2.10988432e-01 -1.58620343e-01 -7.77375638e-01 -1.41759849e+00 6.04710281e-01 1.62464976e-01 -3.56895685e-01 -5.20267248e-01 1.06811607e+00 8.64404798e-01 -1.25045422e-02 3.79190564e-01 -3.38293314e-01 -3.74966532e-01 4.62245494e-01 6.77752435e-01 2.74515390e-01 -3.81966889e-01 -7.34721839e-01 -2.95861304e-01 7.80108094e-01 -1.42576769e-01 1.40504122e-01 9.95477676e-01 -7.47464478e-01 -1.04230233e-01 5.09840906e-01 1.23191810e+00 -2.84259915e-01 -1.18278897e+00 -4.04227972e-01 -1.57936424e-01 -5.29877186e-01 -6.11243695e-02 -3.11453253e-01 -1.16946805e+00 7.02475548e-01 3.93384993e-01 3.76660615e-01 1.18732131e+00 6.01680756e-01 4.90959466e-01 2.24585325e-01 3.27940196e-01 -1.08615315e+00 4.42175627e-01 4.95669812e-01 8.18295240e-01 -1.24379563e+00 -2.19473451e-01 -3.34161997e-01 -8.62157702e-01 1.10925519e+00 7.16089308e-01 8.95266235e-02 6.81634426e-01 1.35529608e-01 -7.48495292e-03 -4.20472056e-01 -9.84616458e-01 -3.47895235e-01 3.28053266e-01 5.07992983e-01 3.54297429e-01 2.10873589e-01 -3.07268590e-01 9.56516623e-01 2.29518302e-02 -8.56937841e-02 3.95159930e-01 4.99758631e-01 -4.91463363e-01 -7.74570286e-01 -4.36511010e-01 3.23002815e-01 -2.97568530e-01 -3.39084387e-01 7.17666671e-02 8.07989120e-01 2.60559797e-01 9.87226844e-01 1.68724284e-01 -5.46651125e-01 7.64497221e-02 -1.49161488e-01 3.25817794e-01 -2.08471432e-01 -3.05182368e-01 3.66897374e-01 -2.31052652e-01 -8.06199849e-01 -7.66193926e-01 -5.93273759e-01 -1.48424649e+00 -5.39381385e-01 -8.38937536e-02 3.88055146e-01 5.03798053e-02 1.10135388e+00 3.31121951e-01 5.53052962e-01 4.11295712e-01 -1.01300073e+00 1.25604291e-02 -5.34447193e-01 -7.71148801e-01 5.67178011e-01 1.24229431e-01 -8.40424657e-01 -5.37474275e-01 3.02888513e-01]

[10.327431678771973, 1.003305435180664]

945db439-7130-433c-a82c-b3596f17d6e7

swin-deformable-attention-hybrid-u-net-for

2302.1445

https://arxiv.org/abs/2302.14450v1

https://arxiv.org/pdf/2302.14450v1.pdf

Swin Deformable Attention Hybrid U-Net for Medical Image Segmentation

How to harmonize convolution and multi-head self-attention mechanisms has recently emerged as a significant area of research in the field of medical image segmentation. Various combination methods have been proposed. However, there is a common flaw in these works: failed to provide a direct explanation for their hybrid model, which is crucial in clinical scenarios. Deformable Attention can improve the segmentation performance and provide an explanation based on the deformation field. Incorporating Deformable Attention into a hybrid model could result in a synergistic effect to boost segmentation performance while enhancing the explainability. In this study, we propose the incorporation of Swin Deformable Attention with hybrid architecture to improve the segmentation performance while establishing explainability. In the experiment section, our proposed Swin Deformable Attention Hybrid UNet (SDAH-UNet) demonstrates state-of-the-art performance on both anatomical and lesion segmentation tasks.

['Guang Yang', 'Jiahao Huang', 'Lichao Wang']

2023-02-28

['lesion-segmentation']

['medical']

[ 1.92326438e-02 4.07527953e-01 -4.30687070e-02 -5.55356443e-01 -7.44390845e-01 -2.40171850e-01 2.50168145e-01 -3.35138179e-02 -2.77316570e-01 5.00008583e-01 1.86798707e-01 -2.01098919e-01 -3.09155166e-01 -5.54944456e-01 -6.60906553e-01 -7.62685239e-01 3.93544108e-01 3.44270587e-01 5.71155727e-01 -4.02272046e-01 1.22112706e-01 3.20686787e-01 -1.03481972e+00 4.25700456e-01 1.28181493e+00 6.91145897e-01 4.34094340e-01 5.60095012e-01 -3.24558944e-01 3.78593534e-01 -3.13530713e-01 -3.18389028e-01 1.11111119e-01 -5.13483465e-01 -1.05935395e+00 -1.66312769e-01 4.43814933e-01 -2.81324029e-01 -8.58563185e-02 8.92304182e-01 7.45304823e-01 -1.52254375e-02 4.85816956e-01 -9.75713789e-01 -1.11429965e+00 9.05148923e-01 -8.19435716e-01 7.05692828e-01 -1.13854520e-01 1.80121124e-01 5.83258569e-01 -5.09995401e-01 4.13271040e-01 1.00102603e+00 7.74103165e-01 7.90098369e-01 -8.72949421e-01 -5.92130661e-01 1.86184272e-01 3.58064145e-01 -9.45025444e-01 3.63237634e-02 6.41562164e-01 -1.96524844e-01 9.61995125e-01 5.29915452e-01 8.45910668e-01 8.30401838e-01 5.04649580e-01 9.22065794e-01 1.10635984e+00 -2.99373507e-01 -2.42187724e-01 -7.05380440e-02 5.91933906e-01 6.63101017e-01 3.26997966e-01 -1.54402450e-01 3.25931213e-03 2.57918656e-01 1.03673816e+00 8.75251219e-02 -4.06400472e-01 1.75290659e-01 -1.01853597e+00 7.92511761e-01 1.15490425e+00 8.84256542e-01 -4.13755774e-01 2.79198021e-01 2.61326224e-01 -4.46618557e-01 6.48863018e-01 6.40924990e-01 -3.36074799e-01 1.37977317e-01 -1.02499223e+00 1.14351064e-01 3.33753005e-02 3.41206223e-01 2.63474554e-01 1.28171489e-01 -6.06547773e-01 5.67694664e-01 3.97833228e-01 2.53797323e-01 8.48928273e-01 -4.96594727e-01 1.96018174e-01 7.81813562e-01 -4.93652612e-01 -7.42248297e-01 -8.25782597e-01 -6.71908557e-01 -8.31231058e-01 2.34001353e-01 4.47550088e-01 -2.81492203e-01 -1.43105614e+00 1.63315165e+00 4.42905307e-01 5.84346414e-01 -1.30450204e-01 1.32305443e+00 1.34500706e+00 2.21508071e-01 3.71072203e-01 1.43550351e-01 1.35089219e+00 -1.35486162e+00 -1.00978851e+00 -1.83831453e-01 5.00564516e-01 -7.76443839e-01 1.02438617e+00 -2.13641211e-01 -1.27755344e+00 -6.57353759e-01 -9.12537813e-01 -2.21364245e-01 -1.99337438e-01 -7.65056908e-02 9.28682268e-01 6.77774966e-01 -1.17530143e+00 5.26391327e-01 -1.27218628e+00 -3.45188946e-01 7.38494873e-01 7.34351099e-01 -1.63913697e-01 3.23001355e-01 -1.13719666e+00 1.11408973e+00 3.03744823e-01 3.59001130e-01 -4.40973788e-01 -9.90310252e-01 -6.65132761e-01 2.78599206e-02 1.63149133e-01 -1.08832061e+00 1.14183545e+00 -9.92365241e-01 -1.30457723e+00 8.93978179e-01 -7.16254190e-02 -5.71485341e-01 6.80453897e-01 -3.20389390e-01 -1.82012618e-01 8.17015469e-02 8.11422318e-02 8.58603060e-01 4.42143947e-01 -1.14543438e+00 -2.55849630e-01 -6.57768190e-01 -7.11509436e-02 1.50121912e-01 -1.13657139e-01 -1.28479078e-01 -5.05991340e-01 -7.05827534e-01 1.20593436e-01 -1.07533371e+00 -5.02313554e-01 -1.21534772e-01 -6.60405517e-01 -1.09784625e-01 9.57597017e-01 -8.06958258e-01 1.23834026e+00 -1.82276738e+00 2.21275136e-01 -3.26922312e-02 5.45384228e-01 6.56942546e-01 7.87285045e-02 -4.01145846e-01 -2.54506052e-01 5.75915992e-01 -4.31270421e-01 -1.70675695e-01 -4.11721498e-01 2.32589349e-01 1.89299405e-01 2.64049977e-01 2.53500432e-01 1.46140850e+00 -5.65119386e-01 -6.43115401e-01 4.66391474e-01 7.87172318e-01 -6.55266047e-01 1.07481092e-01 6.51200935e-02 8.41189206e-01 -7.46855915e-01 3.78887862e-01 8.34672570e-01 -3.84159952e-01 -2.48670191e-01 -3.96688014e-01 -7.27206767e-02 -3.91981214e-01 -7.07289338e-01 1.83506286e+00 -2.18917459e-01 4.56077218e-01 -1.12582915e-01 -9.70201910e-01 4.01871294e-01 4.49977100e-01 6.94517374e-01 -5.31791210e-01 5.44518828e-01 3.13928753e-01 6.09014750e-01 -8.18612337e-01 2.58604050e-01 -3.73452783e-01 3.42021644e-01 8.87703225e-02 8.34775120e-02 -2.01236364e-02 -2.59780139e-01 -2.56578270e-02 8.29037905e-01 7.82966092e-02 1.95174828e-01 -4.08680618e-01 5.11179388e-01 5.81494672e-03 2.54638702e-01 6.04170740e-01 -4.14737314e-01 1.01200247e+00 1.58538446e-01 -4.62555110e-01 -8.64951193e-01 -7.67946661e-01 -4.09145325e-01 6.02687120e-01 4.44935709e-01 2.36125886e-01 -1.19312584e+00 -8.75851214e-01 -1.90925449e-01 4.55210954e-01 -1.19538879e+00 -1.88929603e-01 -6.69184744e-01 -1.24528229e+00 6.27535880e-01 9.26256120e-01 9.37316537e-01 -1.19623101e+00 -6.68153703e-01 3.30835909e-01 -3.91658783e-01 -1.00724649e+00 -6.75037801e-01 9.51673687e-02 -9.19409573e-01 -8.46248388e-01 -1.02720022e+00 -6.18647516e-01 7.37962365e-01 1.66730657e-01 7.28956401e-01 4.78316486e-01 -5.11034489e-01 1.07961982e-01 -2.76227027e-01 -5.45158923e-01 -2.84494996e-01 3.41691762e-01 -5.85357249e-01 -9.38558504e-02 -1.71766564e-01 -2.45305061e-01 -9.15747464e-01 2.10993424e-01 -1.04547501e+00 3.70723456e-01 5.79981446e-01 7.72535324e-01 7.23635614e-01 -4.58832920e-01 6.42270386e-01 -1.01004875e+00 5.99898696e-01 -4.85791951e-01 1.10233377e-03 4.58665341e-01 -5.41501641e-01 1.84196219e-01 2.26132393e-01 -4.00907010e-01 -1.26285946e+00 -1.49742618e-01 -7.19891608e-01 -3.48828584e-01 -2.38218263e-01 4.50294584e-01 1.45007133e-01 -3.43554556e-01 5.09620428e-01 -6.12043366e-02 1.15027577e-01 -3.07366580e-01 2.74209529e-01 3.59494478e-01 4.82943088e-01 -2.94032276e-01 3.99097353e-01 4.84201163e-01 -1.80619061e-01 -4.10852939e-01 -1.00669563e+00 -3.34862173e-01 -7.03488588e-01 -1.34594202e-01 1.58991861e+00 -4.25640583e-01 -6.59860432e-01 5.14471948e-01 -1.17846978e+00 -4.07157898e-01 -2.54619986e-01 4.89524156e-01 -2.51408190e-01 2.24141166e-01 -6.28705025e-01 -3.09977084e-01 -9.30667520e-01 -1.74507713e+00 1.07155240e+00 7.67786026e-01 -1.56902373e-01 -1.33814776e+00 -1.64031029e-01 6.27269745e-01 9.03746426e-01 5.25103450e-01 7.52206266e-01 -7.68966496e-01 -4.27540004e-01 4.51705605e-02 -5.99678338e-01 2.76812725e-02 2.46462569e-01 -1.55088276e-01 -1.00069225e+00 4.40189987e-03 -3.80457491e-02 1.57797843e-01 9.20157075e-01 1.25153875e+00 1.42373776e+00 8.09515417e-02 -4.55705464e-01 8.25221062e-01 1.48013878e+00 2.68595517e-01 7.26841390e-01 3.73820484e-01 8.63169611e-01 1.63148582e-01 1.75153017e-01 2.93075815e-02 4.16853458e-01 6.78111136e-01 6.40159965e-01 -9.82303441e-01 -7.12060750e-01 2.19513074e-01 -2.43163258e-01 5.73197186e-01 -5.54619074e-01 -1.87141985e-01 -1.00632036e+00 4.15001690e-01 -1.88348150e+00 -7.32496202e-01 -7.17937708e-01 1.66106379e+00 6.11578166e-01 3.19793373e-02 3.06810215e-02 -1.01411410e-01 7.71960855e-01 -2.37584934e-01 -6.51783764e-01 -3.51142317e-01 -1.32662281e-01 3.65761846e-01 4.72836763e-01 6.07234180e-01 -1.05225766e+00 8.48537445e-01 6.17769146e+00 7.33498216e-01 -1.56123006e+00 5.30965209e-01 7.54493952e-01 2.62584239e-01 -4.22052741e-01 -3.41740459e-01 -6.36126459e-01 3.41047406e-01 6.25189960e-01 -6.40245602e-02 -1.59578174e-01 4.86659318e-01 2.47632563e-01 7.28153512e-02 -5.32988191e-01 5.67863464e-01 9.40619707e-02 -1.41990423e+00 2.70664375e-02 1.17241986e-01 7.40110397e-01 6.57774732e-02 2.78264999e-01 1.50015846e-01 -6.29031435e-02 -1.08480692e+00 4.72581685e-01 4.40946311e-01 6.33627295e-01 -3.27231109e-01 1.11170876e+00 3.30585465e-02 -1.04876876e+00 2.05054581e-01 1.28500462e-01 3.26903045e-01 3.02061111e-01 1.94185108e-01 -8.63135099e-01 8.41784537e-01 7.42784500e-01 5.65402210e-01 -7.51439869e-01 1.48409891e+00 -6.96747378e-02 7.11793900e-01 -1.91945851e-01 2.31228203e-01 5.69631040e-01 9.77807492e-02 5.15155256e-01 1.14457333e+00 1.92862988e-01 3.33379716e-01 1.69811681e-01 1.11506224e+00 1.69823289e-01 2.94147313e-01 -7.49498606e-02 4.52073067e-01 -2.82334894e-01 1.31730032e+00 -1.24260819e+00 -3.99536818e-01 -3.27779710e-01 9.10115421e-01 6.56696111e-02 1.50183320e-01 -1.31667554e+00 6.19554296e-02 3.50061923e-01 3.83023620e-01 6.57244995e-02 1.98850855e-01 -8.18442702e-01 -9.91731107e-01 -2.36304998e-01 -5.26676238e-01 4.08895493e-01 -7.16686666e-01 -1.07471037e+00 1.02514470e+00 6.74901977e-02 -7.06326485e-01 3.51625830e-01 -2.90726006e-01 -7.58367121e-01 8.38249445e-01 -1.65725851e+00 -1.59775233e+00 -7.15458930e-01 4.97606754e-01 8.90117705e-01 3.61465305e-01 6.27067208e-01 4.12887365e-01 -7.43039191e-01 7.10764945e-01 -3.76540869e-01 -7.57450163e-02 6.22363091e-01 -1.32130218e+00 2.25328118e-01 7.75544405e-01 -1.40293613e-01 5.90880632e-01 6.87511206e-01 -7.94764280e-01 -7.96380818e-01 -1.00244701e+00 2.07727179e-01 -3.84796232e-01 2.95727313e-01 3.79283220e-01 -1.23941386e+00 6.15371287e-01 5.67274868e-01 3.98237824e-01 5.72008491e-01 -2.01788902e-01 2.09679469e-01 2.98575848e-01 -1.41903675e+00 4.53031093e-01 7.83206880e-01 7.14407861e-02 -6.21557593e-01 1.93258286e-01 9.13688123e-01 -1.03312802e+00 -9.82011199e-01 6.54134512e-01 4.58267808e-01 -8.11359048e-01 9.92059648e-01 -4.70089644e-01 7.43293166e-01 -1.73539877e-01 3.62003714e-01 -1.24146283e+00 -4.62955981e-01 -2.81980515e-01 2.60699719e-01 1.09213817e+00 5.33288062e-01 -7.12689221e-01 7.25265861e-01 8.75856340e-01 -7.51181006e-01 -1.19287038e+00 -7.87640393e-01 -3.24216813e-01 4.66211677e-01 -3.39916199e-01 6.49653614e-01 9.18134451e-01 -3.32719922e-01 5.13534695e-02 -1.02699868e-01 9.82477739e-02 4.24348712e-01 -1.15163393e-01 2.05096543e-01 -9.67439830e-01 -8.41913000e-02 -7.28654325e-01 -2.70456731e-01 -6.82113349e-01 9.89057124e-02 -1.22194374e+00 -2.22305119e-01 -2.08802390e+00 5.96538246e-01 -4.14049834e-01 -4.46541280e-01 8.48543346e-01 -7.12881446e-01 3.79024416e-01 2.67713368e-01 4.40307967e-02 -3.45496982e-01 2.83068508e-01 1.95057499e+00 -6.01401292e-02 -2.03157917e-01 -1.67025793e-02 -9.58009005e-01 6.97933555e-01 7.31146812e-01 -2.19482943e-01 -3.16310555e-01 -7.50843763e-01 -5.35489857e-01 1.99638546e-01 3.98390383e-01 -8.60353231e-01 1.02013879e-01 7.72604421e-02 4.11124676e-01 -4.07015502e-01 -4.05191444e-02 -8.46298754e-01 1.48988947e-01 6.33153200e-01 -3.16534907e-01 9.40322578e-02 6.83389902e-01 2.88397729e-01 -1.81141809e-01 -4.52846428e-03 1.01043832e+00 -1.79431528e-01 -4.92431164e-01 4.99387711e-01 -5.05032875e-02 2.55149063e-02 1.19706833e+00 -4.71970767e-01 -2.99797475e-01 7.00828210e-02 -9.40476835e-01 2.61635900e-01 5.84194884e-02 5.15994012e-01 6.26437545e-01 -1.05954611e+00 -8.16142261e-01 7.27334693e-02 -3.81900877e-01 2.81559415e-02 8.83138478e-01 1.29967320e+00 -6.01150155e-01 5.04496753e-01 -4.29072380e-01 -7.09821761e-01 -1.35878766e+00 3.12655091e-01 8.34509313e-01 -5.85807860e-01 -7.97122419e-01 9.91975725e-01 5.10076284e-01 -1.94616705e-01 -1.36563718e-01 -7.43047655e-01 -4.28198606e-01 -4.03038621e-01 2.72844762e-01 1.39439523e-01 1.45608172e-01 -6.51495337e-01 -5.05542755e-01 8.98074329e-01 -2.25448698e-01 1.97401568e-01 1.36317050e+00 -9.86718610e-02 7.81094357e-02 -9.16196629e-02 9.01240408e-01 -3.48290831e-01 -9.72522736e-01 1.06862828e-01 -2.66134471e-01 -2.80175954e-01 4.00343984e-01 -1.23888588e+00 -1.67207491e+00 9.91035461e-01 1.05242145e+00 1.61272272e-01 1.08957231e+00 1.11755013e-01 1.24213696e+00 -5.85833728e-01 -2.88493801e-02 -4.73478705e-01 9.11295190e-02 5.33836931e-02 1.03076661e+00 -1.59206009e+00 -2.46574104e-01 -6.80851340e-01 -7.55469084e-01 1.07410455e+00 1.05395710e+00 -3.22710425e-02 7.15875089e-01 5.17456949e-01 1.36865765e-01 -5.03407836e-01 -1.25009820e-01 -4.27114278e-01 7.17224479e-01 3.38280976e-01 9.13612723e-01 -2.07691193e-02 -7.26731718e-01 9.34366763e-01 1.94529239e-02 -6.07412234e-02 3.06661963e-01 4.62879479e-01 -2.85333991e-01 -1.02299571e+00 -3.64592165e-01 4.51699167e-01 -9.63831306e-01 -1.17119789e-01 -1.29475787e-01 8.84091139e-01 4.12196934e-01 6.57895148e-01 -1.46137744e-01 -8.08717683e-02 4.47343320e-01 -2.03371987e-01 6.33975267e-01 -5.55613518e-01 -1.25021756e+00 2.71027684e-01 -3.97415251e-01 -3.89598936e-01 -4.88690555e-01 -5.21601498e-01 -1.89446259e+00 -3.63035947e-02 -7.37229705e-01 -2.62543485e-02 6.74285889e-01 1.23482692e+00 2.71264374e-01 1.28651500e+00 -1.08616743e-02 -7.48921812e-01 -4.79916930e-02 -9.52613652e-01 -1.67157948e-01 6.10771477e-01 2.19822422e-01 -6.82232261e-01 -6.48201331e-02 2.21517961e-03]

[14.613574981689453, -2.5345144271850586]

399a919c-f1a9-40d8-9e5e-b5047c0372a7

text-with-knowledge-graph-augmented

2303.12423

https://arxiv.org/abs/2303.12423v2

https://arxiv.org/pdf/2303.12423v2.pdf

Text with Knowledge Graph Augmented Transformer for Video Captioning

Video captioning aims to describe the content of videos using natural language. Although significant progress has been made, there is still much room to improve the performance for real-world applications, mainly due to the long-tail words challenge. In this paper, we propose a text with knowledge graph augmented transformer (TextKG) for video captioning. Notably, TextKG is a two-stream transformer, formed by the external stream and internal stream. The external stream is designed to absorb additional knowledge, which models the interactions between the additional knowledge, e.g., pre-built knowledge graph, and the built-in information of videos, e.g., the salient object regions, speech transcripts, and video captions, to mitigate the long-tail words challenge. Meanwhile, the internal stream is designed to exploit the multi-modality information in videos (e.g., the appearance of video frames, speech transcripts, and video captions) to ensure the quality of caption results. In addition, the cross attention mechanism is also used in between the two streams for sharing information. In this way, the two streams can help each other for more accurate results. Extensive experiments conducted on four challenging video captioning datasets, i.e., YouCookII, ActivityNet Captions, MSRVTT, and MSVD, demonstrate that the proposed method performs favorably against the state-of-the-art methods. Specifically, the proposed TextKG method outperforms the best published results by improving 18.7% absolute CIDEr scores on the YouCookII dataset.

['Longyin Wen', 'Tiejian Luo', 'Libo Zhang', 'YuFei Wang', 'Guang Chen', 'Xin Gu']

2023-03-22

http://openaccess.thecvf.com//content/CVPR2023/html/Gu_Text_With_Knowledge_Graph_Augmented_Transformer_for_Video_Captioning_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Gu_Text_With_Knowledge_Graph_Augmented_Transformer_for_Video_Captioning_CVPR_2023_paper.pdf

cvpr-2023-1

['video-captioning']

['computer-vision']

[ 3.09008900e-02 3.26080644e-03 -3.83051813e-01 -9.93884653e-02 -7.35781252e-01 -3.51886064e-01 5.46488583e-01 -1.19464375e-01 -9.33508426e-02 5.53124905e-01 5.79929709e-01 1.53334156e-01 3.14669579e-01 -2.56539583e-01 -1.00264609e+00 -6.51733398e-01 9.44632739e-02 1.19672917e-01 3.83112252e-01 -2.15024669e-02 -7.08555728e-02 -2.00506151e-01 -1.35433769e+00 4.59004879e-01 6.95342422e-01 1.34082162e+00 5.47058225e-01 2.67777592e-01 -3.42714041e-01 1.04400396e+00 -2.21444815e-01 -4.86295044e-01 -1.67663321e-01 -4.50908691e-01 -4.46100891e-01 2.88116693e-01 3.72069687e-01 -4.79260802e-01 -8.76424491e-01 9.59963202e-01 3.94810349e-01 1.49243116e-01 2.21947029e-01 -1.62308419e+00 -6.96331263e-01 5.73145330e-01 -6.41038716e-01 2.85434216e-01 3.95720124e-01 3.15083981e-01 8.68592918e-01 -9.14461613e-01 6.29177988e-01 1.23004401e+00 8.40906501e-02 5.58181882e-01 -4.94191766e-01 -8.63594592e-01 5.40390432e-01 5.93691230e-01 -1.21228194e+00 -4.32792842e-01 7.40818977e-01 -4.73931015e-01 5.32723069e-01 1.05936036e-01 6.06046021e-01 1.29040837e+00 -1.65590838e-01 1.11143517e+00 5.14677644e-01 1.25348698e-02 -1.94821116e-02 2.07932934e-01 2.72096903e-03 5.98346829e-01 6.25849068e-02 -1.81789368e-01 -7.84909189e-01 6.11863099e-02 6.75364614e-01 1.61209151e-01 -7.55533636e-01 -4.49060768e-01 -1.31933272e+00 5.49952090e-01 3.94122809e-01 1.60104007e-01 -5.32892168e-01 2.54830956e-01 6.00823939e-01 -2.14631647e-01 4.26202416e-01 -6.32728636e-02 -2.73160040e-01 -3.01404923e-01 -6.12077296e-01 -1.03477739e-01 4.61739242e-01 1.38198161e+00 5.26501119e-01 -1.39144696e-02 -5.32908022e-01 7.24812925e-01 6.71201706e-01 7.29118168e-01 4.69874024e-01 -5.83606541e-01 1.11422467e+00 3.73831481e-01 1.94082633e-01 -1.18805552e+00 9.07613114e-02 -3.18437040e-01 -6.92725718e-01 -7.05081940e-01 -1.20341661e-03 -1.09440036e-01 -1.07115984e+00 1.95465159e+00 2.29117796e-01 8.63355756e-01 1.20345101e-01 1.29529703e+00 1.23165607e+00 1.20086133e+00 3.42633396e-01 -2.74479538e-01 1.56927550e+00 -1.38081455e+00 -9.71358240e-01 -3.54756474e-01 1.70606241e-01 -6.91530824e-01 7.45804310e-01 -4.95374501e-02 -9.41890001e-01 -5.84206820e-01 -8.24797451e-01 2.69843549e-01 1.94056425e-02 1.67871162e-01 3.30662757e-01 8.05837512e-02 -8.15129995e-01 9.05829202e-03 -5.83151162e-01 -3.21935117e-01 4.67162043e-01 2.74008643e-02 -3.18491966e-01 -4.92068559e-01 -1.41907775e+00 3.89660656e-01 5.01079082e-01 1.27151459e-01 -1.14075983e+00 -5.61583042e-01 -8.78492653e-01 2.43155301e-01 7.14536965e-01 -5.32025337e-01 1.17433608e+00 -1.20703673e+00 -1.09682000e+00 4.25894260e-01 -3.06589603e-01 -3.08301032e-01 4.08680320e-01 -2.36884668e-01 -5.44182420e-01 4.63902146e-01 9.43658277e-02 8.79757047e-01 8.70132565e-01 -1.43721271e+00 -6.86022818e-01 -1.19179338e-01 1.62784427e-01 5.20482421e-01 -5.61920285e-01 8.07806551e-02 -1.44457352e+00 -7.00263262e-01 -2.07894459e-01 -9.95590329e-01 3.25239539e-01 5.52111538e-04 -3.01172614e-01 -1.73424169e-01 1.07491863e+00 -9.93540108e-01 1.20814538e+00 -2.47570777e+00 1.70933530e-01 -7.95301571e-02 1.60353541e-01 4.39477473e-01 -4.43882912e-01 3.57017040e-01 -9.27979201e-02 8.60122964e-02 5.69465421e-02 -4.18817639e-01 -1.95832014e-01 2.90081859e-01 -2.78813720e-01 2.09482282e-01 2.16763154e-01 9.71327007e-01 -1.23312998e+00 -6.61840796e-01 1.23468615e-01 6.25179350e-01 -1.83795094e-01 5.01305282e-01 -4.12190109e-01 3.20267051e-01 -6.25936389e-01 4.84349132e-01 5.82879603e-01 -6.39529228e-01 -5.76269180e-02 -5.30799270e-01 1.59767464e-01 -5.72476909e-02 -7.83794701e-01 1.69698417e+00 -3.10114950e-01 7.38337576e-01 9.27536041e-02 -7.69227147e-01 5.91286302e-01 8.59851360e-01 4.61840719e-01 -7.67167211e-01 2.15820357e-01 -5.03654703e-02 -4.18135464e-01 -7.97928214e-01 3.00730258e-01 1.93754658e-01 1.44792467e-01 1.69905797e-01 2.29235888e-01 5.57362974e-01 1.79343089e-01 7.50062823e-01 7.90230930e-01 1.48044825e-01 -5.06306291e-02 3.59884739e-01 4.99196321e-01 -1.88005015e-01 6.51592195e-01 3.88596624e-01 -3.39899927e-01 7.36853182e-01 4.18100953e-01 -2.73252986e-02 -9.14551079e-01 -8.55071068e-01 6.35329425e-01 9.21637654e-01 8.31981778e-01 -4.15498823e-01 -6.94541812e-01 -7.29438782e-01 -2.65119553e-01 4.90075111e-01 -5.81189871e-01 -3.67128462e-01 -3.88098806e-01 -2.57820368e-01 2.58109391e-01 6.34944856e-01 6.25128031e-01 -1.16553366e+00 -6.35765269e-02 2.27563515e-01 -8.62066031e-01 -1.69229579e+00 -1.08211970e+00 -5.23555398e-01 -5.24214268e-01 -1.07917893e+00 -1.01598096e+00 -8.72314811e-01 5.76041579e-01 8.38727653e-01 8.60817850e-01 1.85718179e-01 2.53220826e-01 5.88880777e-01 -8.67989123e-01 -1.58806741e-01 -1.44557029e-01 -3.11244518e-01 -2.16939658e-01 5.14638901e-01 1.75314158e-01 -1.26796320e-01 -5.63977242e-01 4.83787388e-01 -1.08476579e+00 5.56895852e-01 6.69502079e-01 8.21227372e-01 5.46771288e-01 -1.44860715e-01 5.71434617e-01 -3.75580788e-01 3.09900135e-01 -1.05893433e+00 -2.49499783e-01 3.14038962e-01 -2.39983127e-01 -1.62881836e-01 6.20689273e-01 -6.28464878e-01 -1.16687405e+00 -2.07034815e-02 1.47472143e-01 -9.56614316e-01 5.69362342e-02 5.92031419e-01 -4.99482244e-01 1.82438076e-01 -4.79904935e-02 5.16266227e-01 -7.17054009e-02 -3.48842561e-01 2.79628783e-01 8.08607757e-01 7.23614573e-01 -3.02564442e-01 7.49522984e-01 3.17577720e-01 -4.85999554e-01 -6.13442302e-01 -8.28037262e-01 -7.27528095e-01 1.59476519e-01 -5.72982252e-01 1.06486142e+00 -1.27649188e+00 -4.17872548e-01 5.56991696e-01 -1.40623486e+00 5.60628390e-03 1.34816512e-01 5.81141412e-01 -4.62629944e-01 4.59985495e-01 -5.72997928e-01 -6.88210726e-01 -4.26661581e-01 -1.24902141e+00 1.05164254e+00 5.82678497e-01 4.26590770e-01 -6.82352781e-01 -3.49070281e-01 8.23362470e-01 1.17053561e-01 -4.70828526e-02 6.04119778e-01 -7.44934916e-01 -9.33569133e-01 6.63552955e-02 -6.87565386e-01 4.15348798e-01 1.45357594e-01 -2.21568659e-01 -8.25440228e-01 -2.77467519e-01 -3.53804410e-01 -2.97732383e-01 6.86100543e-01 1.38986304e-01 1.08442712e+00 -3.61069381e-01 -4.10035640e-01 3.20496887e-01 1.19473708e+00 5.25245309e-01 7.11939096e-01 1.31316140e-01 9.57404673e-01 4.26073670e-01 8.69417548e-01 4.44552779e-01 6.49069965e-01 8.38841021e-01 7.34993398e-01 -6.25935644e-02 -4.33067560e-01 -5.83889306e-01 5.02552688e-01 1.07820427e+00 9.37621444e-02 -7.60040402e-01 -6.83908403e-01 7.26158619e-01 -2.25380588e+00 -9.31510866e-01 6.34035990e-02 1.89442658e+00 5.05196810e-01 -3.63471918e-02 -1.44864172e-01 -3.07885677e-01 1.05548394e+00 3.14177513e-01 -6.47298276e-01 1.58123925e-01 -1.37994394e-01 -5.62945843e-01 2.88063705e-01 -1.05046220e-02 -1.05732882e+00 7.97499120e-01 4.36468887e+00 9.63898182e-01 -1.00671303e+00 2.49523655e-01 4.79812771e-01 1.71421524e-02 -3.16534102e-01 -9.67807993e-02 -5.17051041e-01 1.07859457e+00 7.79887497e-01 -2.92588800e-01 4.57637727e-01 5.94038725e-01 3.73030394e-01 1.23948872e-01 -8.63833606e-01 1.24316800e+00 4.90714520e-01 -1.33074057e+00 2.13823065e-01 -2.56031096e-01 5.71536779e-01 1.32773370e-01 -7.21798092e-02 3.08695555e-01 -2.11259127e-01 -5.13019741e-01 8.52507412e-01 4.88442600e-01 7.39770949e-01 -5.57740748e-01 1.03914249e+00 1.42124265e-01 -1.53073990e+00 4.19303924e-02 8.72661266e-03 4.47257102e-01 5.06494820e-01 3.12150210e-01 -6.13998652e-01 8.82978082e-01 7.82690942e-01 1.02435660e+00 -3.14356178e-01 1.29036939e+00 -2.42361739e-01 8.39737236e-01 -1.77667305e-01 4.53391708e-02 5.44671953e-01 -6.32766038e-02 5.60002327e-01 1.21770966e+00 3.32375556e-01 2.78159887e-01 2.09220812e-01 5.05092919e-01 -4.09411073e-01 1.92303911e-01 -3.69092137e-01 -3.77569497e-01 4.42142218e-01 1.10171223e+00 -3.86812717e-01 -5.54270208e-01 -8.05439234e-01 1.00802481e+00 7.93738887e-02 6.30082309e-01 -1.27281618e+00 -2.12065846e-01 6.83505714e-01 -8.24113488e-02 6.22658134e-01 6.84495568e-02 4.51030463e-01 -1.39642787e+00 3.30889702e-01 -8.27160180e-01 4.12559271e-01 -1.40517581e+00 -1.23925531e+00 6.19071424e-01 2.30121296e-02 -1.33526719e+00 5.51125780e-02 -2.02327967e-01 -5.23377597e-01 5.93956470e-01 -1.71536422e+00 -1.11895096e+00 -7.96195626e-01 7.06862867e-01 7.94866204e-01 -7.69143775e-02 2.40965635e-01 7.22369254e-01 -6.91920280e-01 4.83767539e-01 9.84198526e-02 3.06610197e-01 8.16155076e-01 -5.43570340e-01 1.24856673e-01 9.12284493e-01 7.86700547e-02 1.45518169e-01 5.70622683e-01 -8.09395075e-01 -1.49378109e+00 -1.36542344e+00 5.31060517e-01 1.35730579e-01 5.99394262e-01 -2.55862981e-01 -1.06181073e+00 5.76184094e-01 2.81417072e-01 1.31088823e-01 3.75526339e-01 -5.35667479e-01 -5.48818648e-01 -1.65164068e-01 -7.69244254e-01 4.69024420e-01 1.00155759e+00 -4.13595408e-01 -4.93823588e-01 3.44965339e-01 1.19301283e+00 -4.12881166e-01 -5.76550663e-01 2.33087689e-01 3.62548828e-01 -4.62924272e-01 7.98645318e-01 -4.54173416e-01 5.87556064e-01 -4.73815113e-01 -1.10127680e-01 -1.31837249e+00 -4.08261791e-02 -6.32369578e-01 -4.21980292e-01 1.56425452e+00 2.41534024e-01 -2.51778215e-01 5.54963470e-01 4.98874843e-01 -3.64174426e-01 -6.72516167e-01 -7.98458695e-01 -6.92796111e-01 -6.82137787e-01 -3.20439160e-01 5.75586438e-01 8.61064672e-01 1.22802621e-02 5.45110822e-01 -9.81661201e-01 1.79075614e-01 4.79267567e-01 1.55340070e-02 5.90912044e-01 -7.84555078e-01 -1.15278482e-01 -4.78993915e-02 -4.19222713e-01 -1.42560780e+00 1.87611774e-01 -6.86021328e-01 2.69879222e-01 -1.78320289e+00 6.95138037e-01 -2.33855039e-01 -4.09353971e-01 5.33819914e-01 -5.32196164e-01 1.04860283e-01 5.72502255e-01 2.87706703e-01 -1.14720321e+00 7.60668755e-01 1.47497761e+00 -2.77395248e-01 -2.70274710e-02 -4.21526074e-01 -7.44477749e-01 4.61529553e-01 6.90397084e-01 -4.09672469e-01 -6.89590156e-01 -7.08145320e-01 -5.06483503e-02 3.54818732e-01 3.41304481e-01 -8.65963936e-01 4.51289386e-01 -2.21413374e-01 6.04857206e-02 -5.60613990e-01 5.90651870e-01 -8.83901060e-01 1.83814213e-01 2.67744064e-02 -1.71584249e-01 5.47959656e-03 2.11979598e-01 9.86131012e-01 -5.21747351e-01 1.17131911e-01 5.02750635e-01 6.36418015e-02 -9.98594224e-01 8.14324260e-01 -1.21774070e-01 2.56924510e-01 1.15465307e+00 -9.20841396e-02 -6.63698852e-01 -8.61405253e-01 -4.45457608e-01 8.51271629e-01 2.69779533e-01 9.31960225e-01 9.13633525e-01 -1.47653866e+00 -8.07693601e-01 -7.81740472e-02 6.12328768e-01 -6.98762611e-02 6.30754709e-01 8.92275512e-01 -3.78197022e-02 4.51238066e-01 -1.28658608e-01 -4.21941906e-01 -1.36820531e+00 6.86569810e-01 -6.87551033e-03 -8.51561576e-02 -5.85061371e-01 7.45667219e-01 5.99628985e-01 3.56616765e-01 4.38479155e-01 1.35432296e-02 -3.63153070e-01 9.39464420e-02 7.53932178e-01 7.57574290e-02 -3.59806865e-01 -9.70483720e-01 -4.42110896e-01 4.39855725e-01 -1.50558159e-01 1.28580108e-01 1.04805529e+00 -4.80368942e-01 1.31318167e-01 2.62595415e-02 1.28058004e+00 -3.55823576e-01 -1.31104290e+00 -5.34155309e-01 -4.12122786e-01 -5.19591272e-01 3.42660099e-02 -7.74470448e-01 -1.50585365e+00 7.95107245e-01 4.08394098e-01 -1.54734731e-01 1.20246387e+00 2.83274204e-01 1.24036682e+00 1.21001713e-02 2.93467492e-01 -7.90831566e-01 3.68237734e-01 3.23953390e-01 7.98257887e-01 -1.21808815e+00 -3.16590577e-01 -5.72550535e-01 -1.00469112e+00 7.49874175e-01 8.30359399e-01 3.56013000e-01 2.59081453e-01 -2.62677312e-01 9.64481607e-02 2.97169238e-02 -8.84994984e-01 -1.49894625e-01 4.08751726e-01 6.14163220e-01 2.94446945e-03 -6.72611892e-02 -7.03559816e-02 7.42640078e-01 5.24736762e-01 1.73421174e-01 4.60109204e-01 6.08798862e-01 -3.11666727e-01 -6.92120373e-01 -3.03978413e-01 4.63344872e-01 -3.23176742e-01 -1.87242866e-01 -1.61619395e-01 5.38765788e-01 5.16323708e-02 1.24284816e+00 -2.35745162e-02 -5.60919523e-01 2.95590997e-01 -1.81512967e-01 -1.13300588e-02 -4.41087961e-01 -3.11657012e-01 7.12596253e-02 1.79656908e-01 -7.74312019e-01 -5.70412040e-01 -3.81950408e-01 -1.25513136e+00 -3.14385556e-02 -3.83901656e-01 5.19620121e-01 6.73482478e-01 9.23477829e-01 5.83230615e-01 5.74007034e-01 5.57227969e-01 -6.41038299e-01 -9.00712162e-02 -7.85793304e-01 -3.37103575e-01 6.60239816e-01 3.66838217e-01 -7.52219141e-01 -3.93990904e-01 2.66246051e-01]

[10.473132133483887, 0.7882286310195923]

4d12865c-fc80-427c-bcd8-fead10aa9618

limitr-leveraging-local-information-for

2303.11755

https://arxiv.org/abs/2303.11755v1

https://arxiv.org/pdf/2303.11755v1.pdf

LIMITR: Leveraging Local Information for Medical Image-Text Representation

Medical imaging analysis plays a critical role in the diagnosis and treatment of various medical conditions. This paper focuses on chest X-ray images and their corresponding radiological reports. It presents a new model that learns a joint X-ray image & report representation. The model is based on a novel alignment scheme between the visual data and the text, which takes into account both local and global information. Furthermore, the model integrates domain-specific information of two types -- lateral images and the consistent visual structure of chest images. Our representation is shown to benefit three types of retrieval tasks: text-image retrieval, class-based retrieval, and phrase-grounding.

['Ayellet Tal', 'Elad Hirsch', 'Gefen Dawidowicz']

2023-03-21

['phrase-grounding']

['natural-language-processing']

[ 1.92993656e-01 8.95230025e-02 -6.46340072e-01 -6.73996091e-01 -1.51249945e+00 -3.70787978e-01 5.06956041e-01 7.74035752e-01 -2.22521871e-01 3.18757951e-01 5.57322025e-01 -2.73984909e-01 -4.72255200e-01 -5.00109494e-01 -3.54938805e-01 -5.11751652e-01 -1.51706515e-02 6.96974874e-01 2.85966754e-01 2.02933952e-01 1.94201946e-01 4.95892137e-01 -1.00466406e+00 8.26285660e-01 4.35401164e-02 7.88473487e-01 5.90174973e-01 8.19224834e-01 -1.66568071e-01 1.10954332e+00 -5.19221127e-01 -1.42043546e-01 -2.89364439e-02 -5.43338120e-01 -9.88964081e-01 5.22703588e-01 6.15737617e-01 -4.10111606e-01 -6.12603962e-01 9.74465013e-01 6.54079437e-01 -4.27813977e-02 1.07324827e+00 -7.64715195e-01 -7.31079042e-01 2.49680087e-01 -9.00812149e-01 7.56819487e-01 3.94570649e-01 -1.78633183e-01 9.70638037e-01 -8.38399947e-01 1.11446869e+00 9.27369654e-01 2.76342750e-01 3.56817275e-01 -7.06651509e-01 -2.20741957e-01 4.75185402e-02 3.96061450e-01 -1.30676830e+00 2.04227000e-01 6.29352391e-01 -5.70969880e-01 8.72541487e-01 5.41276991e-01 7.46849477e-01 7.42719293e-01 8.49763155e-01 8.44512761e-01 1.02124357e+00 -6.67008817e-01 -2.39793379e-02 7.63885155e-02 1.90040767e-01 1.18058038e+00 8.15114658e-03 -1.11436583e-01 -6.25457287e-01 -3.69861394e-01 9.40603256e-01 4.31567967e-01 -4.80817318e-01 -5.18855095e-01 -1.19548392e+00 8.56372952e-01 6.07526898e-01 6.82976186e-01 -5.67531466e-01 1.22445479e-01 4.90641713e-01 1.85130965e-02 3.12403321e-01 1.26249164e-01 1.85667783e-01 4.34419870e-01 -1.08624911e+00 -6.60326555e-02 3.23476940e-01 9.38179910e-01 3.68427411e-02 -4.38972652e-01 -3.87429684e-01 8.76025081e-01 4.53419626e-01 6.39110327e-01 7.90121973e-01 -3.28491509e-01 7.31101215e-01 4.72667813e-01 -3.85172486e-01 -1.14490616e+00 -6.74705327e-01 -2.54674613e-01 -8.48773062e-01 6.26894757e-02 -2.00927436e-01 6.54856980e-01 -1.49503982e+00 1.34324217e+00 1.84737951e-01 -5.71819484e-01 -3.17652225e-01 1.08904731e+00 1.39132738e+00 6.17912471e-01 3.34761977e-01 -4.49075043e-01 1.80964470e+00 -9.90338981e-01 -1.07491183e+00 -9.63452905e-02 3.86265576e-01 -8.56543243e-01 7.29337275e-01 -4.92210174e-03 -1.56410205e+00 -4.67850864e-01 -1.05807817e+00 -3.05745840e-01 -2.48030603e-01 5.18776834e-01 2.25950658e-01 1.76933050e-01 -1.06145477e+00 7.95522481e-02 -8.76312494e-01 -6.26858532e-01 3.29124182e-01 3.09859723e-01 -7.72150755e-01 -3.38476300e-01 -6.77839458e-01 1.03544533e+00 3.66651177e-01 -1.94746330e-01 -5.62600732e-01 -5.63772976e-01 -8.60285878e-01 9.21427608e-02 3.45192820e-01 -1.08948660e+00 1.31853545e+00 -3.13067853e-01 -9.51008260e-01 1.44856167e+00 -4.49580662e-02 4.23799604e-02 3.76143187e-01 -3.94163430e-02 -4.45327699e-01 1.04097748e+00 1.50683135e-01 5.33938110e-01 1.02598715e+00 -1.23281312e+00 -4.61663812e-01 -5.60704589e-01 -3.22324783e-01 4.69084293e-01 1.34196812e-02 2.09883928e-01 -1.05917156e+00 -1.14784348e+00 5.14495492e-01 -7.28254497e-01 -1.54297054e-01 3.65655929e-01 -3.56333196e-01 -3.15718092e-02 6.66578829e-01 -9.70731199e-01 1.32750607e+00 -2.10683870e+00 3.63678455e-01 6.33376956e-01 6.24627113e-01 -2.38997266e-01 6.52598888e-02 5.21977305e-01 -4.36492980e-01 4.76308167e-02 -6.66065663e-02 -4.10942346e-01 -3.08385134e-01 3.06470335e-01 -3.13130587e-01 5.15242696e-01 -1.27425224e-01 1.05792022e+00 -6.32089853e-01 -1.21766305e+00 3.07578176e-01 2.81610191e-01 -3.22748348e-02 3.53525162e-01 1.54817104e-01 3.74430746e-01 -5.87039053e-01 7.72765279e-01 1.84417561e-01 -9.23884749e-01 4.20825005e-01 -5.93058586e-01 2.28787839e-01 1.27807319e-01 -5.89538455e-01 1.99544334e+00 -3.40511411e-01 4.75017697e-01 -1.56778350e-01 -5.50649643e-01 3.65891963e-01 5.94938815e-01 6.74972534e-01 -1.00632942e+00 8.49133879e-02 -1.28017783e-01 -4.98931646e-01 -8.35966468e-01 3.87696296e-01 -2.87307799e-01 7.37235248e-02 9.49098110e-01 1.86287165e-02 -3.76941085e-01 4.67447452e-02 7.03714907e-01 1.00525486e+00 -3.84224981e-01 8.39425623e-01 1.11850955e-01 3.47135395e-01 7.76555762e-02 -1.45277068e-01 9.30233896e-01 6.08049482e-02 1.21986699e+00 1.43961951e-01 -4.71341252e-01 -1.03323615e+00 -1.44702613e+00 -1.57669768e-01 8.43505740e-01 2.63578653e-01 -7.61327744e-01 -2.43803188e-01 -9.84890282e-01 -2.19610661e-01 1.25128388e-01 -9.37356591e-01 -1.34622395e-01 -6.70521498e-01 -5.51801741e-01 1.69052056e-03 9.26668823e-01 6.92088157e-02 -8.82458091e-01 -7.14493692e-01 -2.29032680e-01 -2.06882849e-01 -8.11818361e-01 -7.48023987e-01 1.96667425e-02 -1.06581056e+00 -1.19896805e+00 -1.11210155e+00 -8.88286591e-01 9.60463226e-01 2.24650294e-01 1.16997278e+00 4.27531570e-01 -9.20412004e-01 1.23311722e+00 -2.81533629e-01 -9.97371376e-02 -5.36327362e-01 -2.14881867e-01 -4.58214968e-01 -4.33574468e-01 -2.91386366e-01 2.75425147e-02 -7.14479983e-01 -4.28666398e-02 -1.26048625e+00 1.58195868e-01 9.33451295e-01 9.60561216e-01 1.11383307e+00 -1.98414728e-01 -1.35207191e-01 -1.06857252e+00 5.49785614e-01 -3.57069880e-01 1.03818187e-02 9.56234157e-01 -4.15205270e-01 -6.86819153e-03 -1.11182444e-01 -1.66880190e-01 -1.02933705e+00 -1.90470159e-01 -2.71096476e-03 -6.76098883e-01 -6.43280968e-02 6.53116405e-01 4.73114759e-01 9.80855227e-02 4.22650367e-01 2.10729286e-01 -4.30647284e-02 -5.58224678e-01 3.18336040e-01 3.77680898e-01 6.60443842e-01 -1.62896946e-01 4.85412300e-01 6.28775001e-01 2.63976514e-01 -6.52887583e-01 -9.93997335e-01 -9.02900457e-01 -7.07037985e-01 -3.73214900e-01 1.09608722e+00 -6.54449582e-01 -1.03595071e-01 -3.34255189e-01 -1.11419928e+00 3.65309089e-01 -2.96869576e-01 7.87666082e-01 -6.06232464e-01 6.90834403e-01 -8.42722178e-01 -1.35013238e-01 -5.54243207e-01 -1.28358591e+00 1.32469869e+00 -1.51378289e-01 -2.13112548e-01 -1.06676996e+00 2.69528061e-01 2.25688621e-01 -5.54788217e-04 1.93209305e-01 1.42762256e+00 -4.77626324e-01 -5.23249745e-01 -3.17715645e-01 -5.44818878e-01 6.29753992e-02 1.90425470e-01 -1.79363430e-01 -5.56959271e-01 -4.45217580e-01 2.34093547e-01 -2.62533128e-01 9.02114630e-01 5.44133961e-01 1.20689476e+00 -2.71811672e-02 -6.38539612e-01 3.55893075e-01 1.44645166e+00 2.35810295e-01 2.71977663e-01 1.24244630e-01 7.53049076e-01 4.83826905e-01 5.53432405e-01 2.03182653e-01 3.74900907e-01 6.89186156e-01 3.70324701e-01 -4.67760533e-01 -5.30797720e-01 -1.22330353e-01 -4.00610894e-01 1.20601535e+00 -7.92202950e-02 -2.76395082e-01 -1.24658322e+00 4.18358207e-01 -1.80503690e+00 -7.67979324e-01 -1.85680091e-02 1.79534841e+00 6.22618496e-01 -1.62266448e-01 -8.05983767e-02 -9.03490782e-02 7.49324799e-01 2.50016958e-01 -8.76159221e-02 1.12666972e-01 2.25087270e-01 4.21329379e-01 2.42308438e-01 2.24796221e-01 -1.21061838e+00 2.87545249e-02 7.74142504e+00 8.13004136e-01 -1.17057967e+00 3.52741212e-01 3.71782124e-01 -1.16732776e-01 -1.70613468e-01 -4.62572455e-01 -2.37819642e-01 -1.25036091e-02 3.05785775e-01 -1.96712032e-01 -5.92484713e-01 7.08133280e-01 -3.21564347e-01 -3.49375069e-01 -1.29128683e+00 1.18263090e+00 6.91835701e-01 -1.65239429e+00 6.80255890e-01 1.27714276e-02 4.27260518e-01 -1.50891289e-01 4.06836599e-01 -1.88635260e-01 -1.87003300e-01 -9.45338309e-01 8.10249627e-01 7.86434174e-01 1.05274713e+00 -4.34180379e-01 7.03176677e-01 -8.52836743e-02 -1.12294221e+00 1.62733212e-01 -8.27233717e-02 7.90470243e-01 2.39547864e-01 -4.48149256e-02 -1.00343990e+00 1.01668048e+00 8.22131693e-01 7.62620509e-01 -7.87060499e-01 1.20486951e+00 -1.54785827e-01 2.65238971e-01 2.57053170e-02 4.42850113e-01 2.32659087e-01 2.87057132e-01 4.91675526e-01 1.38560152e+00 1.91214845e-01 4.60162848e-01 3.92204732e-01 4.19747233e-01 4.56187390e-02 3.68431330e-01 -6.59054458e-01 1.71246409e-01 -1.47061378e-01 1.04791009e+00 -1.12050700e+00 -6.10909760e-01 -4.99895096e-01 5.67836642e-01 1.04408644e-01 2.16697663e-01 -6.06391430e-01 7.72381425e-02 -3.71899664e-01 1.99123263e-01 2.40676105e-01 -5.66686876e-02 -1.80352598e-01 -9.82408226e-01 -3.61768045e-02 -6.65700972e-01 1.16868997e+00 -1.20746553e+00 -1.27833223e+00 8.27180028e-01 4.90871459e-01 -1.49335837e+00 -4.41451550e-01 -5.16076207e-01 -2.94171244e-01 6.24566674e-01 -1.35671246e+00 -1.32511425e+00 -2.07286432e-01 7.16844261e-01 5.68157852e-01 -1.77624539e-01 8.80859971e-01 2.51657605e-01 -1.40296176e-01 2.83162773e-01 -4.47227657e-02 1.17657945e-01 7.59814799e-01 -1.37370491e+00 -2.92396963e-01 4.08223569e-01 2.38131315e-01 6.97149217e-01 4.02671307e-01 -6.06749892e-01 -1.04060721e+00 -7.92749524e-01 6.63672507e-01 -4.71427292e-01 4.35296744e-01 2.19921097e-01 -9.16449010e-01 7.10967898e-01 4.24243301e-01 7.48320073e-02 1.04887128e+00 -3.60215753e-01 -4.63903427e-01 1.55148923e-01 -9.28054333e-01 3.46680611e-01 4.24601167e-01 -9.35869277e-01 -1.05861342e+00 6.93226099e-01 4.23875809e-01 -8.57019901e-01 -9.90100861e-01 4.20198232e-01 5.50608516e-01 -5.31572998e-01 1.18972611e+00 -6.97573841e-01 5.27308226e-01 -9.16395411e-02 -1.98906064e-01 -7.82538295e-01 -3.77836734e-01 1.86087057e-01 3.61494832e-02 5.93170047e-01 1.47913426e-01 5.05039543e-02 3.46759558e-01 2.87900388e-01 -1.67169794e-02 -8.44140410e-01 -9.04483795e-01 -3.77476931e-01 -2.96364933e-01 -2.73077816e-01 7.33333528e-02 8.94714475e-01 8.54588598e-02 2.59526134e-01 -7.20329136e-02 2.77688235e-01 4.54080582e-01 4.61218983e-01 2.71276623e-01 -8.26730728e-01 -3.86661083e-01 -3.60152036e-01 -5.95405579e-01 -5.92730403e-01 -3.53549510e-01 -1.13920987e+00 -2.84441918e-01 -2.06008792e+00 7.67283916e-01 -5.15086800e-02 -4.46915209e-01 4.91281301e-01 -5.36257140e-02 6.32720232e-01 4.02427495e-01 5.81078947e-01 -8.94379258e-01 2.08240777e-01 1.41649663e+00 -3.15670788e-01 2.34192729e-01 -2.23143771e-01 -1.51308626e-01 7.81915605e-01 4.76586163e-01 -8.18024516e-01 -4.38158631e-01 -4.69198942e-01 3.58085573e-01 6.16926849e-01 4.16522145e-01 -7.18133450e-01 4.19305652e-01 -6.02271631e-02 5.88832617e-01 -1.08032179e+00 5.39072931e-01 -9.50832367e-01 -1.06192917e-01 5.50725698e-01 -6.07207656e-01 6.17059231e-01 1.81280300e-01 8.70006442e-01 -5.77654302e-01 -3.84877264e-01 6.89896882e-01 -5.92336297e-01 -4.71919447e-01 2.83851564e-01 -2.82440335e-01 -1.58132792e-01 1.01376379e+00 5.91603592e-02 -4.67642635e-01 -3.32755774e-01 -1.20414674e+00 5.93075790e-02 1.65706977e-01 3.23298007e-01 1.05405092e+00 -1.33224368e+00 -5.24207532e-01 -1.11999661e-02 6.53699338e-01 -2.14087620e-01 4.78834897e-01 1.00450397e+00 -7.02057660e-01 5.95733404e-01 -1.26359135e-01 -1.03368413e+00 -1.84443736e+00 6.32119477e-01 4.84131910e-02 -8.25763941e-01 -1.03026807e+00 5.19581199e-01 7.01635897e-01 1.51842654e-01 3.14340591e-01 -4.69703257e-01 -3.74687165e-01 -8.97769704e-02 7.01518893e-01 -1.92555040e-01 3.46306950e-01 -7.04297364e-01 -3.86992842e-01 9.39269722e-01 -6.65059686e-01 -2.70871758e-01 1.22475874e+00 -2.93952644e-01 -1.30017117e-01 5.69387674e-01 1.28903127e+00 -1.76964805e-01 -1.65942162e-01 -5.35175323e-01 -3.96311320e-02 -5.18420279e-01 4.15822208e-01 -9.50572252e-01 -1.00399435e+00 9.03258920e-01 7.36675024e-01 7.68113881e-02 1.15965915e+00 7.25954711e-01 4.54683393e-01 3.97051454e-01 8.12101886e-02 -8.33772480e-01 5.72876453e-01 -1.16051354e-01 1.22101068e+00 -1.25960207e+00 7.85338819e-01 -3.83784622e-01 -6.98644042e-01 1.24229228e+00 1.75151363e-01 2.57143825e-01 6.68104649e-01 4.32397127e-02 2.85591722e-01 -9.25322413e-01 -8.28088582e-01 -1.32626042e-01 1.03286695e+00 4.19232398e-01 4.14818048e-01 -5.88383200e-03 -3.67033154e-01 2.43204236e-01 4.00163770e-01 -2.80686826e-01 1.09901987e-01 1.58408403e+00 -3.66585135e-01 -9.34604049e-01 -6.84957922e-01 4.90763366e-01 -8.11034620e-01 1.52241401e-02 -3.19749802e-01 1.18160546e+00 -2.76439577e-01 3.60178083e-01 7.46033490e-02 -1.20841458e-01 3.39699209e-01 -8.52855481e-03 8.39944184e-01 -8.83864939e-01 -6.04360819e-01 4.10273552e-01 -3.27878118e-01 -5.51035523e-01 -7.21244872e-01 -4.35009509e-01 -1.25426769e+00 3.03753406e-01 -1.84440479e-01 2.66658049e-02 7.70780504e-01 8.61924589e-01 -6.17599860e-02 9.41710174e-01 3.14280391e-01 -3.91265333e-01 -2.55671620e-01 -6.38209581e-01 -7.64488995e-01 5.09210289e-01 4.86553460e-01 -6.81698322e-01 3.76136079e-02 3.51265013e-01]

[14.93877124786377, -1.6190842390060425]

4edc7d0c-fa65-414f-9d0a-73eaa46db9dd

exploring-better-text-image-translation-with

2305.17415

https://arxiv.org/abs/2305.17415v2

https://arxiv.org/pdf/2305.17415v2.pdf

Exploring Better Text Image Translation with Multimodal Codebook

Text image translation (TIT) aims to translate the source texts embedded in the image to target translations, which has a wide range of applications and thus has important research value. However, current studies on TIT are confronted with two main bottlenecks: 1) this task lacks a publicly available TIT dataset, 2) dominant models are constructed in a cascaded manner, which tends to suffer from the error propagation of optical character recognition (OCR). In this work, we first annotate a Chinese-English TIT dataset named OCRMT30K, providing convenience for subsequent studies. Then, we propose a TIT model with a multimodal codebook, which is able to associate the image with relevant texts, providing useful supplementary information for translation. Moreover, we present a multi-stage training framework involving text machine translation, image-text alignment, and TIT tasks, which fully exploits additional bilingual texts, OCR dataset and our OCRMT30K dataset to train our model. Extensive experiments and in-depth analyses strongly demonstrate the effectiveness of our proposed model and training framework.

['Jinsong Su', 'Degen Huang', 'Bin Wang', 'Jian Luan', 'Wen Zhang', 'Xiang Li', 'Jiawei Yu', 'Zhibin Lan']

2023-05-27

['optical-character-recognition']

['computer-vision']

[ 6.88714206e-01 -4.68369186e-01 -1.96816817e-01 -1.89160496e-01 -1.02872217e+00 -6.39768660e-01 7.77635515e-01 -4.66140389e-01 -5.38025856e-01 5.25683880e-01 1.97217062e-01 -3.12750101e-01 3.97267789e-01 -1.73410326e-01 -7.12715447e-01 -6.55097902e-01 8.53583872e-01 5.60408950e-01 8.14383104e-02 3.13035101e-02 3.35030586e-01 3.52371037e-02 -8.98410141e-01 4.57682818e-01 1.23131800e+00 9.37673688e-01 5.10241508e-01 4.35434401e-01 -1.18271090e-01 6.33947551e-01 -4.08229858e-01 -9.67207730e-01 7.00982139e-02 -6.85024500e-01 -6.21975183e-01 3.00210536e-01 4.53501552e-01 -3.43969762e-01 -3.11484307e-01 1.18732214e+00 5.37065744e-01 -2.79833168e-01 7.04078913e-01 -1.13547134e+00 -1.02461767e+00 5.03975391e-01 -5.96511781e-01 -1.85940608e-01 1.54717982e-01 1.76467955e-01 8.60360920e-01 -1.17561114e+00 6.82346106e-01 1.08219337e+00 3.08093041e-01 5.53796411e-01 -1.00684702e+00 -6.94458663e-01 -1.19480453e-01 2.38683075e-01 -1.38199139e+00 -5.29349983e-01 5.50122201e-01 -3.62098724e-01 6.17295325e-01 3.00021201e-01 5.41442811e-01 1.34159935e+00 1.35709032e-01 1.21479273e+00 1.32328129e+00 -5.78760147e-01 -2.12929696e-01 3.88046622e-01 -5.17282784e-01 4.08520520e-01 -1.29720226e-01 -2.44296640e-01 -5.28056443e-01 2.98293412e-01 6.94640517e-01 -1.25806659e-01 -2.98857898e-01 -2.19017401e-01 -1.49749732e+00 5.35294771e-01 2.08740413e-01 3.28907251e-01 -1.82028878e-02 -1.58459127e-01 3.97149533e-01 2.66194701e-01 4.02429432e-01 1.19966149e-01 -1.04695737e-01 -1.36675090e-01 -9.17791426e-01 -2.50868827e-01 3.16423178e-01 1.34772444e+00 6.54780328e-01 -1.51973918e-01 -2.50056773e-01 1.11427295e+00 3.57392341e-01 1.04234970e+00 6.76411211e-01 -4.01101679e-01 1.20318735e+00 6.47434294e-01 -9.98121276e-02 -9.85363245e-01 1.22498974e-01 -1.66886643e-01 -9.28848267e-01 -5.56517839e-01 4.91920263e-02 1.10440135e-01 -6.55550957e-01 1.33505201e+00 1.35258839e-01 -1.56057939e-01 2.71233201e-01 1.06071103e+00 5.96066117e-01 7.27821350e-01 -9.36739296e-02 -2.18291342e-01 1.32051587e+00 -1.38345313e+00 -8.52320731e-01 -2.85570979e-01 5.77507675e-01 -1.39588547e+00 1.25064671e+00 5.14159352e-02 -9.28183019e-01 -6.05838239e-01 -8.02025080e-01 -3.21776271e-01 -6.92347512e-02 8.22687566e-01 2.53459692e-01 4.41980034e-01 -1.01890767e+00 -1.30480111e-01 -7.13441491e-01 -5.20234108e-01 3.15086722e-01 2.42605060e-01 -3.66916060e-01 -3.63017142e-01 -1.04491663e+00 8.57385159e-01 3.59252334e-01 5.46674073e-01 -8.06572616e-01 4.92207371e-02 -8.42639387e-01 -2.21253231e-01 4.09466922e-01 -4.44953501e-01 1.12767661e+00 -1.28818059e+00 -1.52770185e+00 9.36559916e-01 -4.34466422e-01 -1.06410496e-01 7.92933941e-01 -2.49977544e-01 -4.27076370e-01 3.30054253e-01 2.38842830e-01 7.21075952e-01 1.04703271e+00 -1.13066292e+00 -5.45847416e-01 -3.71412903e-01 -3.68350446e-01 5.55228949e-01 -7.76346624e-01 3.70773703e-01 -1.23968637e+00 -8.85407507e-01 -8.39865953e-03 -1.07308900e+00 -2.40733959e-02 -2.31053397e-01 -6.11811042e-01 9.32744667e-02 7.33393610e-01 -8.41019988e-01 1.31144130e+00 -2.10363793e+00 4.26701069e-01 -2.27827013e-01 -1.05928525e-01 3.06414813e-01 -2.87405372e-01 6.61395371e-01 1.91262037e-01 -1.28007997e-02 -3.90607595e-01 -6.71393573e-01 2.38477383e-02 3.78412716e-02 -5.34357429e-01 2.92766750e-01 3.63069057e-01 1.26433599e+00 -5.53265274e-01 -1.02036309e+00 1.99549019e-01 2.77146637e-01 -2.27665037e-01 3.19458008e-01 -2.23929256e-01 7.01005042e-01 -6.05312109e-01 7.56571472e-01 6.71326280e-01 -1.71212986e-01 2.67896980e-01 -2.67647624e-01 -2.06104606e-01 1.43763572e-02 -6.39923513e-01 1.78158832e+00 -3.83944631e-01 6.81035459e-01 -1.84484392e-01 -9.31149065e-01 8.04469049e-01 3.64272565e-01 2.19635919e-01 -1.02479208e+00 2.67699987e-01 4.58590925e-01 -3.21349531e-01 -7.07886457e-01 6.85965538e-01 1.68717399e-01 -1.06778361e-01 5.26766717e-01 -1.01555102e-01 -1.96260706e-01 1.77789152e-01 1.99061200e-01 5.15750945e-01 4.93070185e-01 -1.34540528e-01 1.25838935e-01 6.49039388e-01 2.76303530e-01 4.15435135e-01 3.04423332e-01 -1.45914346e-01 7.94012666e-01 3.33454818e-01 -1.83514938e-01 -1.25806308e+00 -5.78815341e-01 -6.30906150e-02 7.85074949e-01 2.99533516e-01 -2.26440355e-01 -1.08464003e+00 -6.63011372e-01 -6.23163939e-01 4.37186986e-01 -4.46730137e-01 3.48694399e-02 -7.43127584e-01 -7.78485715e-01 6.58397079e-01 4.06748265e-01 7.51462162e-01 -1.08216047e+00 -1.24069020e-01 1.93434209e-02 -9.14722502e-01 -1.79602969e+00 -1.10292983e+00 -2.75668263e-01 -8.37291121e-01 -9.15694058e-01 -1.03967857e+00 -1.18093956e+00 9.56350565e-01 5.81640303e-01 7.32343674e-01 1.90737024e-02 -2.11660415e-02 2.37756073e-01 -5.24400592e-01 -9.50819179e-02 -6.27208114e-01 1.76722452e-01 -8.48841444e-02 3.01815778e-01 3.54192227e-01 -7.51173198e-02 -5.04744351e-01 6.66544259e-01 -1.15896475e+00 6.54939294e-01 1.13949943e+00 8.88253689e-01 6.13735676e-01 -2.19539195e-01 1.79382220e-01 -6.36626542e-01 4.16995168e-01 4.46757637e-02 -6.59062684e-01 7.06374705e-01 -5.24883151e-01 -2.12001964e-01 6.87266290e-01 -4.91084874e-01 -1.07721734e+00 2.07704425e-01 -1.74030047e-02 -4.37466174e-01 5.38386358e-03 4.60571975e-01 -3.88511598e-01 -3.52814160e-02 1.97782397e-01 9.00759280e-01 -1.95599258e-01 -2.99922466e-01 2.48023614e-01 1.21894598e+00 7.00404346e-01 -6.35449290e-01 1.04437590e+00 3.44259948e-01 -4.41845477e-01 -6.83390260e-01 -7.19899058e-01 -2.97022074e-01 -9.96935427e-01 -1.86097160e-01 1.04444218e+00 -1.10925496e+00 -3.36719543e-01 7.44512379e-01 -1.19617379e+00 -2.96506256e-01 4.40332860e-01 6.89638376e-01 -5.17993271e-01 6.84281409e-01 -6.86576486e-01 -5.63711584e-01 -4.85209167e-01 -1.48697340e+00 1.38305688e+00 1.19418979e-01 3.21319222e-01 -7.74550557e-01 -1.42688066e-01 1.09074998e+00 2.79117227e-01 -2.03251809e-01 8.37198317e-01 -3.89547974e-01 -7.68840730e-01 -7.21374825e-02 -5.59372544e-01 3.80321115e-01 7.28338212e-02 -6.87843561e-02 -7.27934062e-01 -3.74233723e-01 -9.46957096e-02 -6.68987393e-01 5.85360885e-01 -5.25165349e-02 8.93949151e-01 -2.11993247e-01 -2.05959037e-01 5.16560495e-01 1.26413846e+00 1.09808862e-01 7.58871853e-01 4.38429207e-01 9.38523173e-01 5.27372718e-01 8.33577454e-01 -3.29997800e-02 6.32123172e-01 9.74764228e-01 1.86973646e-01 -4.21792567e-01 -1.18908107e-01 -4.29743946e-01 6.62511230e-01 1.50942492e+00 1.17464639e-01 -3.60367686e-01 -8.60196114e-01 5.03811121e-01 -1.96522963e+00 -5.88194132e-01 -1.43778041e-01 2.06302810e+00 1.08032596e+00 -5.50375804e-02 -1.87649310e-01 -2.87873566e-01 9.23851490e-01 -1.23512909e-01 -4.97397840e-01 1.09935198e-02 -2.98564881e-01 -1.33111417e-01 2.69940138e-01 1.11456089e-01 -1.07092512e+00 1.43752313e+00 5.71176529e+00 1.11799598e+00 -1.29308212e+00 8.32024515e-02 6.32800758e-01 3.51172358e-01 -1.49233788e-01 3.51620354e-02 -7.67612040e-01 6.17760718e-01 6.08812928e-01 8.04522559e-02 4.42520708e-01 4.80044752e-01 2.27833331e-01 1.28787816e-01 -9.34692740e-01 1.01403427e+00 4.56400424e-01 -9.80975211e-01 4.57951427e-01 8.10960606e-02 8.19653690e-01 -5.27495407e-02 2.18134269e-01 2.56416231e-01 1.25215878e-03 -9.01521623e-01 8.91966224e-01 1.91496238e-01 1.23172569e+00 -4.37910497e-01 7.16022015e-01 3.82580489e-01 -1.14965737e+00 1.61042005e-01 -5.17546117e-01 4.53916758e-01 1.73333654e-04 1.04243159e-01 -7.69218445e-01 8.43978763e-01 4.81120735e-01 8.81887257e-01 -8.17219257e-01 7.16275513e-01 -4.52938110e-01 5.45947015e-01 8.01351517e-02 -3.69030572e-02 2.89054185e-01 -5.31014919e-01 2.06073493e-01 1.32533276e+00 3.48319948e-01 -2.38578901e-01 2.16512948e-01 7.80053258e-01 -2.41880745e-01 5.41070759e-01 -3.32465470e-01 -4.64169711e-01 3.60297114e-01 1.45130026e+00 -6.94267035e-01 -3.76280814e-01 -6.99662328e-01 1.51720452e+00 2.28165656e-01 5.40678740e-01 -9.34115946e-01 -2.47359306e-01 8.04112032e-02 -2.18752712e-01 2.68825620e-01 -2.55249143e-01 -7.32302517e-02 -1.71355450e+00 2.66040862e-01 -1.27614331e+00 2.03795172e-02 -1.01648510e+00 -1.00845313e+00 6.50465548e-01 -3.15489084e-01 -1.57511270e+00 6.01315722e-02 -5.32327652e-01 -2.83778250e-01 8.52334142e-01 -1.59165597e+00 -1.73575854e+00 -3.21053237e-01 7.85231888e-01 7.89582789e-01 -1.60076320e-01 5.88829160e-01 5.47983229e-01 -9.44453955e-01 7.97482193e-01 4.00292665e-01 6.15919471e-01 1.17573094e+00 -8.24818432e-01 5.53215683e-01 1.03491580e+00 3.15551192e-01 6.10377848e-01 2.41707906e-01 -6.14080131e-01 -1.73358357e+00 -1.24336553e+00 9.47918773e-01 -4.31781292e-01 7.47475505e-01 -5.47807336e-01 -6.22275233e-01 7.71326065e-01 4.29235011e-01 -4.00125742e-01 5.61218739e-01 -4.90666389e-01 -2.08030164e-01 -3.01906746e-02 -5.11964738e-01 9.59437490e-01 7.97468364e-01 -6.96668267e-01 -4.23616081e-01 4.53969449e-01 6.97474897e-01 -5.19529700e-01 -6.98936939e-01 1.63874194e-01 5.92157662e-01 -7.10964620e-01 6.16809547e-01 -1.84178323e-01 8.18904400e-01 -4.32042003e-01 -2.42089644e-01 -1.01159310e+00 1.16378203e-01 -5.69622993e-01 3.91142279e-01 1.43366659e+00 4.61746722e-01 -3.86034757e-01 4.26633686e-01 3.51435155e-01 -2.85924673e-01 -4.92707700e-01 -8.50184858e-01 -4.61827725e-01 -1.94927417e-02 -3.14508587e-01 3.37249279e-01 1.01787877e+00 -1.93066284e-01 6.33745849e-01 -8.19207311e-01 5.57603128e-02 4.44083989e-01 4.14003283e-01 9.91993248e-01 -5.29444516e-01 -9.60126296e-02 -3.88110816e-01 -9.71152559e-02 -1.51493478e+00 5.64384088e-02 -8.56244624e-01 2.83588231e-01 -1.37187290e+00 8.61934006e-01 -3.06090087e-01 9.17051286e-02 4.49074775e-01 -5.18724978e-01 5.69721103e-01 1.35753632e-01 8.98557663e-01 -8.58788908e-01 8.53811264e-01 1.55063498e+00 -3.17737550e-01 1.80056199e-01 -2.53834069e-01 -5.91635942e-01 3.84175509e-01 4.72583085e-01 -4.01669085e-01 -2.83805370e-01 -1.03477049e+00 1.38860643e-01 1.05926551e-01 1.94439039e-01 -6.27520680e-01 3.15380692e-01 -2.79856116e-01 4.49044138e-01 -6.55104756e-01 2.12869093e-01 -8.91509652e-01 -6.75543845e-02 2.70732939e-01 -3.41085643e-01 1.35317877e-01 -2.80817803e-02 4.17717695e-01 -4.40803200e-01 -2.52550274e-01 5.97954094e-01 7.70136416e-02 -3.99481505e-01 3.51780981e-01 -2.60116905e-01 -1.51425108e-01 7.52619207e-01 -1.59521714e-01 -2.28934526e-01 -3.39317709e-01 -1.54442534e-01 4.77087468e-01 7.20257342e-01 6.27047360e-01 5.16596496e-01 -1.46891117e+00 -8.61745596e-01 7.53357410e-02 4.43286300e-01 -1.23886824e-01 1.28398895e-01 1.15670538e+00 -5.22138715e-01 6.04181230e-01 -7.16491789e-02 -8.40924323e-01 -1.21240544e+00 5.45582592e-01 -1.39843384e-02 -1.94165200e-01 -5.48838258e-01 4.07570601e-01 3.40469360e-01 -5.63278258e-01 7.62389377e-02 1.74583343e-03 -1.44267634e-01 -1.68899462e-01 3.97380829e-01 -1.99654520e-01 -1.51847051e-02 -1.09751379e+00 -1.12981133e-01 9.64094222e-01 -2.39134416e-01 -3.51083279e-01 9.37584698e-01 -5.99081755e-01 -2.49996990e-01 1.31051436e-01 1.05528021e+00 -3.64812054e-02 -1.05707383e+00 -5.70724130e-01 -5.84271774e-02 -4.64447230e-01 -3.15845102e-01 -6.75382555e-01 -9.00992930e-01 1.18315899e+00 4.44109470e-01 -4.61301565e-01 1.27647245e+00 -1.57146916e-01 1.17192292e+00 4.28581953e-01 2.35785797e-01 -1.19806588e+00 3.49437177e-01 3.69726449e-01 7.71238089e-01 -1.40699744e+00 -1.61174580e-01 -3.50882828e-01 -9.87807810e-01 1.15755987e+00 6.30590022e-01 4.80358362e-01 -1.65445328e-01 -3.49061936e-02 5.33126116e-01 2.37665981e-01 -5.49080670e-01 -9.54920352e-02 3.77114177e-01 2.91108578e-01 3.59190583e-01 -1.88178271e-01 -3.19578469e-01 4.77907389e-01 -1.25753894e-01 -7.81751722e-02 2.72378087e-01 7.01169431e-01 -2.10043252e-01 -1.32531953e+00 -4.22260106e-01 -9.03017074e-02 -4.44856733e-01 -4.72209692e-01 -6.29843354e-01 5.76508224e-01 -1.67193651e-01 9.32926655e-01 -2.76402920e-01 -3.94726634e-01 1.64993644e-01 -2.44166497e-02 3.55181962e-01 -3.88981462e-01 -4.10040945e-01 6.64821923e-01 -3.64272565e-01 -2.32823059e-01 -6.26240373e-01 -6.64940357e-01 -8.21225882e-01 -5.35174385e-02 -5.17071426e-01 1.46387145e-01 7.93265820e-01 1.16654873e+00 3.37354779e-01 2.11349979e-01 7.92111099e-01 -5.13497233e-01 -3.87592137e-01 -1.14429843e+00 -1.20663404e-01 3.77285004e-01 -8.53349920e-03 -1.68862924e-01 -7.23326346e-04 5.81118226e-01]

[11.530354499816895, 1.5865225791931152]

a93c78fd-5997-4cc5-8845-47ca1513f2eb

are-pretrained-multilingual-models-equally

https://openreview.net/forum?id=GnlD4Dzr1t

https://openreview.net/pdf?id=GnlD4Dzr1t

Are Pretrained Multilingual Models Equally Fair Across Languages?

Pretrained multilingual language models can help bridge the digital language divide, enabling high-quality NLP models for lower-resourced languages. Studies of multilingual models have so far focused on performance, consistency, and cross-lingual generalization. However, with their wide-spread application in the wild and downstream societal impact, it is important to put multilingual models under the same scrutiny as monolingual models. This work investigates the group fairness of multilingual models, asking whether these models are equally fair across languages. To this end, we create a new four-way multilingual dataset of parallel cloze test examples (MozArt), equipped with demographic information (balanced with regard to gender and native tongue) about the test participants. We evaluate three multilingual models on MozArt -- mBERT, XLM-R, and mT5 -- and show that across the four target languages, the three models exhibit different levels of group disparity, e.g., exhibiting near-equal risk for Spanish, but high levels of disparity for German.

['Anonymous']

2022-01-16

acl-arr-january-2022-1

['cloze-test', 'pretrained-multilingual-language-models']

['natural-language-processing', 'natural-language-processing']

[-6.80243015e-01 -1.20832950e-01 -8.01223397e-01 -2.79799223e-01 -1.17628419e+00 -9.91026640e-01 7.91795075e-01 2.88993239e-01 -7.88763821e-01 8.65950823e-01 5.13964057e-01 -7.72588789e-01 -6.87643811e-02 -3.63847852e-01 -5.33089101e-01 -1.77108228e-01 1.11239418e-01 5.71420968e-01 -3.12610477e-01 -1.92170277e-01 -1.22962140e-01 6.08172230e-02 -1.02960026e+00 2.38612860e-01 1.54449034e+00 3.49891305e-01 -3.07147894e-02 2.24658579e-01 -8.68693441e-02 6.15589142e-01 -3.25049847e-01 -1.24311721e+00 2.34405041e-01 -2.04219460e-01 -5.73130071e-01 -4.94418830e-01 8.88357997e-01 3.74540803e-03 -1.71374351e-01 1.18748415e+00 6.28021657e-01 -4.25411016e-01 8.42771828e-01 -1.01559734e+00 -1.17912340e+00 1.08932042e+00 -5.83732724e-01 9.78795364e-02 5.54305553e-01 2.69564122e-01 1.36641943e+00 -8.27320933e-01 1.08385146e+00 1.64627755e+00 8.82580101e-01 3.37523431e-01 -1.55597210e+00 -1.04722428e+00 3.97994190e-01 1.45366877e-01 -1.58294570e+00 -4.76740509e-01 1.81275085e-01 -6.37050867e-01 7.59653986e-01 7.74239823e-02 4.47856128e-01 1.49519289e+00 3.57603431e-01 5.59912622e-01 1.80210781e+00 -3.94483805e-01 -1.81470305e-01 5.65092325e-01 1.52953327e-01 3.22674155e-01 5.03396869e-01 2.48110667e-01 -5.57196379e-01 -4.90611345e-02 1.39748901e-01 -6.98153734e-01 -2.30667233e-01 -1.31443471e-01 -1.23711145e+00 8.57069254e-01 2.69164234e-01 5.86721480e-01 6.99479654e-02 -4.35187370e-01 4.16286618e-01 5.42635322e-01 7.79798090e-01 4.22537416e-01 -7.41183758e-01 -7.19708437e-03 -8.26752067e-01 5.64951777e-01 7.57980645e-01 8.05925846e-01 4.69453692e-01 7.90136233e-02 -6.53928816e-02 1.17691135e+00 3.11202109e-01 8.48388672e-01 6.72321618e-01 -4.18733954e-01 9.55824316e-01 3.85766149e-01 -1.13135971e-01 -9.04516995e-01 -4.13441479e-01 -4.43956882e-01 -5.66389561e-01 3.79214436e-02 7.74721563e-01 -1.55025393e-01 -5.22283614e-01 2.23250246e+00 -1.70981716e-02 -3.77078474e-01 1.76566228e-01 6.04492068e-01 6.39254272e-01 4.17573035e-01 7.31441736e-01 -8.57666656e-02 1.32803035e+00 -5.26914358e-01 -4.74097103e-01 -5.38358152e-01 1.02623188e+00 -8.14096689e-01 1.58518660e+00 3.42966795e-01 -1.04105997e+00 -4.12393004e-01 -9.52339172e-01 -4.24790919e-01 -7.27527320e-01 2.97379848e-02 5.43521881e-01 1.11945343e+00 -9.69860852e-01 2.24451989e-01 -2.82137126e-01 -4.38929617e-01 1.53003156e-01 4.94253002e-02 -6.45508945e-01 -4.90270883e-01 -1.46544468e+00 1.26826906e+00 2.87580043e-01 -2.90986866e-01 -8.30349505e-01 -1.11577511e+00 -9.36201870e-01 -2.03596860e-01 -4.63615246e-02 -2.53314912e-01 1.00768292e+00 -9.39967096e-01 -9.92804468e-01 1.30245519e+00 1.21585478e-03 -3.91214043e-02 8.58958721e-01 -1.12979792e-01 -7.76175618e-01 -6.13642871e-01 6.50756478e-01 6.32083058e-01 1.65923655e-01 -1.19276917e+00 -6.72809184e-01 -5.51106036e-01 -1.31738797e-01 3.92399907e-01 -4.29174572e-01 5.04042566e-01 -6.37667105e-02 -7.59380639e-01 -2.23828912e-01 -8.43862832e-01 9.95978247e-03 -4.41468567e-01 -3.65595639e-01 -2.56323099e-01 -1.25856381e-02 -1.24498308e+00 1.41923046e+00 -2.14869285e+00 2.55211703e-02 1.25379950e-01 1.41543582e-01 -6.11854829e-02 -2.49393776e-01 2.32188612e-01 -1.77513584e-01 7.38492668e-01 1.83658093e-01 -3.72144133e-01 3.63751560e-01 2.40137111e-02 -1.49021924e-01 6.48079932e-01 5.55854104e-02 9.79195178e-01 -8.54333818e-01 -5.24373472e-01 -2.12527350e-01 2.81276464e-01 -8.13745916e-01 -3.64703029e-01 -8.19668267e-03 5.25154114e-01 9.26899090e-02 6.96479917e-01 6.77228689e-01 2.84261376e-01 4.27692384e-01 1.96932226e-01 -3.79019499e-01 5.34076929e-01 -7.80084193e-01 1.42816603e+00 -6.58708692e-01 7.34719872e-01 1.69876948e-01 -1.98591456e-01 5.80863118e-01 1.32965267e-01 -5.71541302e-02 -1.24988735e+00 2.34077610e-02 8.71941030e-01 5.69354832e-01 -2.73120016e-01 5.49243689e-01 -4.53834981e-01 -4.85807419e-01 4.19840544e-01 4.53398749e-02 -9.63984355e-02 4.52880859e-01 -3.45571227e-02 3.22120100e-01 -1.05330616e-01 3.94593477e-01 -7.83583045e-01 3.16092461e-01 -3.71898636e-02 7.08540440e-01 5.50914824e-01 -2.14141190e-01 1.88031122e-01 4.40072596e-01 2.25032177e-02 -1.02582824e+00 -1.40010667e+00 -6.70860827e-01 1.29145026e+00 -3.13514262e-01 -3.37177426e-01 -6.65678680e-01 -6.28569663e-01 2.66709775e-01 1.11294389e+00 -3.74581695e-01 -1.64611917e-02 -4.42805350e-01 -9.88862216e-01 8.44735980e-01 1.67100161e-01 1.55058607e-01 -7.78618991e-01 2.30567247e-01 -9.44466218e-02 -2.84395844e-01 -1.19134092e+00 -4.03512418e-01 -3.17566395e-02 -5.03051996e-01 -9.09075141e-01 -7.99121559e-01 -9.32549059e-01 2.22657979e-01 -2.22890541e-01 1.56261384e+00 -3.54690403e-01 1.12535633e-01 1.04389377e-01 -1.37857452e-01 -4.58112389e-01 -6.63245142e-01 3.95015836e-01 2.99419999e-01 -3.90783668e-01 6.56452835e-01 -3.58787656e-01 4.56011742e-02 2.32169852e-01 -4.90207404e-01 -1.54789641e-01 4.18583900e-01 4.69982415e-01 3.70635837e-01 -3.11854810e-01 8.53797376e-01 -8.27060997e-01 8.78078043e-01 -8.02795529e-01 -4.20443565e-01 4.44199234e-01 -7.51416683e-01 -2.53687590e-01 6.28849328e-01 -6.16671264e-01 -1.02761865e+00 -8.27181518e-01 -3.46290767e-02 2.25355849e-01 -7.39410147e-03 8.13685536e-01 -4.51233745e-01 1.13953203e-01 8.17385554e-01 -2.44347885e-01 -4.79080796e-01 -6.61741734e-01 5.82801700e-01 9.23981488e-01 4.97057736e-01 -1.01077282e+00 6.73703253e-01 -1.97730977e-02 -7.29798555e-01 -7.14741290e-01 -6.81893528e-01 1.20300613e-01 -5.54213107e-01 -1.29506988e-02 8.56117368e-01 -1.34420121e+00 -4.04971510e-01 5.26397109e-01 -8.21687043e-01 -5.48680961e-01 8.20908099e-02 7.57598042e-01 -2.08908349e-01 -3.09810955e-02 -7.03136861e-01 -4.31517273e-01 -8.47752439e-04 -1.33373940e+00 6.53045952e-01 -1.47327438e-01 -6.26478434e-01 -1.49087417e+00 2.31493130e-01 3.55655313e-01 1.13007478e-01 9.65241566e-02 1.56716561e+00 -9.21719313e-01 -2.40895480e-01 -6.05108328e-02 -3.00098568e-01 2.19797105e-01 4.88487212e-03 -1.60450727e-01 -8.10681760e-01 -3.82372320e-01 -3.00092280e-01 -5.79296768e-01 4.66015041e-01 3.38277489e-01 3.65466714e-01 -1.74535185e-01 -2.27593735e-01 6.12748384e-01 1.33313298e+00 -8.70053843e-02 2.72720784e-01 4.83978838e-01 7.93021798e-01 7.98016012e-01 3.26043159e-01 -1.31020173e-01 9.41723943e-01 6.03661060e-01 -3.18746120e-01 1.56668663e-01 -1.88988790e-01 -6.90364420e-01 8.02403331e-01 1.26381588e+00 3.84203345e-01 -5.45341969e-02 -1.29412127e+00 7.89994538e-01 -1.25687611e+00 -6.66267693e-01 -2.10564315e-01 2.38478398e+00 1.21470416e+00 7.57344216e-02 1.96485430e-01 -1.91611111e-01 6.08505249e-01 1.98797032e-01 -4.02524233e-01 -5.82829356e-01 -6.67788923e-01 -2.60919165e-02 3.05363685e-01 7.73231626e-01 -7.24320292e-01 1.20209503e+00 6.87536287e+00 1.01004517e+00 -1.12905324e+00 3.12955469e-01 8.23600769e-01 -1.23090617e-01 -6.42604947e-01 -8.55826065e-02 -1.03633177e+00 4.44053441e-01 1.10935307e+00 -6.89151227e-01 6.29570544e-01 5.24184525e-01 1.53011993e-01 3.17288935e-02 -1.27177405e+00 7.89776325e-01 2.04884946e-01 -6.12003088e-01 6.25981775e-04 7.96963349e-02 1.10430825e+00 3.94242674e-01 3.66737157e-01 7.16455400e-01 6.40636146e-01 -1.25145888e+00 1.25056517e+00 -3.39780785e-02 1.26028585e+00 -7.90201962e-01 4.31302577e-01 4.13192749e-01 -8.52249026e-01 -1.71270624e-01 -2.61038065e-01 -1.00718506e-01 1.02983683e-01 3.87626320e-01 -4.18630838e-01 4.27650303e-01 6.65120423e-01 5.63588440e-01 -9.37284589e-01 6.47846341e-01 -3.77104312e-01 5.64978063e-01 -1.68048546e-01 2.62844741e-01 1.74543485e-01 -6.07698485e-02 1.94390714e-01 1.31223297e+00 4.75131780e-01 -6.81010067e-01 2.07280263e-01 7.55056024e-01 -3.46271157e-01 8.24381530e-01 -7.25002050e-01 -2.66077191e-01 5.55579126e-01 9.81270730e-01 -3.14952672e-01 -4.52894717e-02 -8.28710079e-01 4.86337721e-01 6.59828603e-01 3.23944032e-01 -6.52856946e-01 2.68429727e-03 6.82254076e-01 1.66633919e-01 -4.40768361e-01 -2.21774131e-01 -5.92449963e-01 -1.31236780e+00 3.84379807e-03 -1.49537098e+00 4.38422740e-01 -5.84427536e-01 -1.82526481e+00 3.87737155e-01 -8.27447772e-02 -8.38038981e-01 -3.35400343e-01 -9.04259622e-01 -1.13901421e-01 1.37052071e+00 -1.58238006e+00 -1.43996418e+00 2.81300306e-01 5.81184924e-01 7.77693242e-02 -4.34037149e-01 7.12731540e-01 7.69641101e-01 -5.17775834e-01 1.12058640e+00 1.49154201e-01 1.57235995e-01 1.27146041e+00 -1.18411231e+00 4.05015409e-01 6.70099616e-01 2.47674868e-01 8.92217994e-01 4.65191454e-01 -9.31445360e-01 -7.84942687e-01 -9.56957102e-01 1.55792749e+00 -8.51962745e-01 1.04022670e+00 -5.83856940e-01 -7.89903581e-01 1.07794237e+00 2.97051221e-01 -5.71807563e-01 8.55068862e-01 8.01439643e-01 -8.64385128e-01 -1.27063855e-03 -1.03143907e+00 1.12148297e+00 1.08963621e+00 -9.66869354e-01 -3.98701280e-01 1.93625554e-01 6.83309793e-01 -2.23386213e-01 -1.09538352e+00 2.61260569e-01 6.88234448e-01 -9.11151230e-01 7.65189588e-01 -7.52666891e-01 5.62974036e-01 1.21765919e-01 -3.60334009e-01 -1.77278888e+00 -2.87540108e-01 -2.96241164e-01 7.27156460e-01 1.65495598e+00 8.62762332e-01 -8.08352172e-01 1.95577696e-01 4.66108054e-01 -4.06745858e-02 -4.10743713e-01 -1.08000469e+00 -9.01833594e-01 1.11334479e+00 -7.28211164e-01 4.53846544e-01 1.47199106e+00 1.23329401e-01 4.40691590e-01 -1.35446772e-01 5.78245670e-02 3.69026065e-01 -3.10417622e-01 6.50293827e-01 -1.31256962e+00 -2.40401998e-02 -7.86585629e-01 -9.77582857e-02 -3.31097841e-01 8.28999996e-01 -1.58879030e+00 -4.22462314e-01 -1.15989566e+00 5.14246404e-01 -5.51823437e-01 9.01880767e-03 4.00066733e-01 -3.55458885e-01 1.75248817e-01 4.09407824e-01 5.10586426e-03 -3.02464724e-01 2.46405870e-01 1.06800771e+00 -3.10536940e-02 -1.67248040e-01 -4.30326194e-01 -1.16626334e+00 7.46280551e-01 7.62584567e-01 -4.85984862e-01 -2.52801538e-01 -8.37639093e-01 5.56049585e-01 -2.97295392e-01 2.13663101e-01 -7.93425679e-01 6.60866406e-03 -1.22432992e-01 3.69190484e-01 -7.02316361e-03 -9.27658975e-02 -6.49708807e-01 7.15922415e-02 2.65453696e-01 -4.28378165e-01 5.76241791e-01 4.51838166e-01 -8.18991289e-02 -2.25853384e-01 -1.53416187e-01 6.13484621e-01 -4.27599698e-02 -3.71346295e-01 1.81663603e-01 -2.49019027e-01 7.49210238e-01 6.65436506e-01 6.51706681e-02 -6.37262225e-01 -2.28866443e-01 -4.74447757e-01 3.66415292e-01 6.42406762e-01 9.14719880e-01 -2.33332768e-01 -1.55918062e+00 -1.19347310e+00 2.06554353e-01 5.33422589e-01 -7.11352050e-01 2.19618201e-01 7.83124089e-01 -2.39592955e-01 6.29149258e-01 -2.04149708e-01 -2.30769694e-01 -8.85612905e-01 4.72324520e-01 4.49295074e-01 -5.20218730e-01 8.46961662e-02 5.06963670e-01 4.03440267e-01 -1.12433636e+00 -3.89616713e-02 -1.40896261e-01 -5.44461701e-03 5.53868353e-01 2.70522654e-01 4.56122935e-01 -6.58550635e-02 -1.03009570e+00 -2.35056981e-01 6.55818582e-01 -2.16079965e-01 -3.82206172e-01 1.00083733e+00 -1.55480862e-01 -2.88799137e-01 7.91460812e-01 1.13979053e+00 8.80766809e-01 -6.79898083e-01 -1.20601207e-01 3.75680715e-01 -3.48833144e-01 -2.59225160e-01 -1.18601787e+00 -6.50174439e-01 7.87065446e-01 4.34668094e-01 -1.33291453e-01 5.57128966e-01 -6.24798518e-03 2.53976405e-01 -7.68720210e-02 5.53381920e-01 -1.20373821e+00 -4.37617749e-01 6.64123833e-01 8.47438693e-01 -1.18384171e+00 -2.18877509e-01 -1.32104099e-01 -6.68881178e-01 2.89906383e-01 7.05938220e-01 2.56344855e-01 4.65241194e-01 2.11505681e-01 4.34344292e-01 1.89995870e-01 -6.01031423e-01 -3.90465893e-02 3.13396871e-01 7.18942523e-01 9.81848657e-01 4.17706430e-01 -7.25591004e-01 9.44285333e-01 -8.63113701e-01 -2.93971896e-01 9.66934264e-02 1.62087724e-01 1.78949103e-01 -1.21507263e+00 -4.91749913e-01 2.50611961e-01 -8.44010949e-01 -5.32788754e-01 -5.73447287e-01 1.41965401e+00 3.89720768e-01 8.38020623e-01 -4.27262597e-02 -1.87077731e-01 4.32519913e-01 4.96559858e-01 4.08773184e-01 -4.65364814e-01 -8.71118069e-01 -1.39066339e-01 4.37162906e-01 -1.42989561e-01 1.16179697e-01 -9.01080966e-01 -6.13373935e-01 -7.46099710e-01 5.96479401e-02 -1.21647894e-01 5.07875919e-01 7.63744354e-01 4.75125760e-02 1.45341992e-01 1.59262091e-01 -5.46971142e-01 -6.10435188e-01 -1.12054968e+00 -6.47557914e-01 5.17174542e-01 1.06589444e-01 -3.78993005e-01 -5.21855772e-01 -3.97615790e-01]

[10.44095230102539, 10.089821815490723]

da16a271-cc81-4ab4-b197-ce916b076f18

comparison-of-synthetic-dataset-generation

2209.11493

https://arxiv.org/abs/2209.11493v1

https://arxiv.org/pdf/2209.11493v1.pdf

Comparison of synthetic dataset generation methods for medical intervention rooms using medical clothing detection as an example

The availability of real data from areas with high privacy requirements, such as the medical intervention space, is low and the acquisition legally complex. Therefore, this work presents a way to create a synthetic dataset for the medical context, using medical clothing as an example. The goal is to close the reality gap between the synthetic and real data. For this purpose, methods of 3D-scanned clothing and designed clothing are compared in a Domain-Randomization and Structured-Domain-Randomization scenario using an Unreal-Engine plugin or Unity. Additionally a Mixed-Reality dataset in front of a greenscreen and a target domain dataset were used. Our experiments show, that Structured-Domain-Randomization of designed clothing together with Mixed-Reality data provide a baseline achieving 72.0% mAP on a test dataset of the clinical target domain. When additionally using 15% of available target domain train data, the gap towards 100% (660 images) target domain train data could be nearly closed 80.05% mAP (81.95% mAP). Finally we show that when additionally using 100% target domain train data the accuracy could be increased to 83.35% mAP.

['Marcus Vetter', 'Steffen Diehl', 'Nils Rathmann', 'Anke Siebert', 'Marcus Pfister', 'Yannick Bukschat', 'Indira Emter', 'Ronja Vorpahl', 'Hannah Teufel', 'Patrick Schülein']

2022-09-23

['mixed-reality']

['computer-vision']

[ 2.72999346e-01 7.92092264e-01 1.69443175e-01 -5.12291789e-01 -9.97242093e-01 -7.44845688e-01 4.23913598e-01 7.54082575e-02 -6.08202994e-01 8.55089724e-01 6.12906814e-02 -3.96554917e-01 2.33873744e-02 -7.27320254e-01 -1.05627906e+00 -3.10573667e-01 1.73876718e-01 6.21702492e-01 2.29099348e-01 -2.33337507e-01 -3.16182643e-01 4.98382211e-01 -1.13077009e+00 8.13007534e-01 6.47205949e-01 7.78281331e-01 2.68904716e-02 6.06380701e-01 4.25015390e-01 6.36756599e-01 -6.74823523e-01 -5.90801895e-01 8.82887602e-01 -1.77036524e-01 -4.95449513e-01 1.47184851e-02 6.92606151e-01 -6.08502388e-01 -1.19504789e-02 8.72684181e-01 9.72535551e-01 -2.51319647e-01 2.80040979e-01 -8.98394167e-01 -2.81972170e-01 1.74144745e-01 -4.00384635e-01 -3.99276018e-01 8.96529853e-01 2.41599649e-01 1.06877714e-01 -5.18782258e-01 1.36900234e+00 7.37048149e-01 8.88563693e-01 7.10326254e-01 -1.32187450e+00 -5.89489043e-01 -3.97741854e-01 -4.64901239e-01 -1.25452471e+00 -2.23552391e-01 5.32802284e-01 -6.06169343e-01 4.48927909e-01 7.62797356e-01 6.97281361e-01 1.62719941e+00 1.17462158e-01 3.92436832e-01 1.65419817e+00 -3.90912622e-01 3.58804107e-01 1.04217374e+00 -6.00130297e-02 5.04607081e-01 4.35262203e-01 4.83900964e-01 -7.05077201e-02 -3.23125422e-01 9.18098807e-01 -2.70837903e-01 -2.94461757e-01 -6.15792990e-01 -1.32733333e+00 5.89670062e-01 4.10620868e-01 2.68023580e-01 -3.06785017e-01 -2.82123864e-01 6.08684242e-01 3.49052191e-01 3.01578701e-01 8.60708892e-01 -4.33262974e-01 9.23996121e-02 -1.02182150e+00 3.20240945e-01 8.64385784e-01 1.19669282e+00 2.11256817e-01 -1.60613388e-01 -1.12786517e-01 3.33850026e-01 -7.18981847e-02 5.31628072e-01 2.20066860e-01 -6.09251857e-01 5.60187459e-01 4.20234829e-01 4.80142593e-01 -9.21444952e-01 -5.55269182e-01 -3.09902668e-01 -7.04174757e-01 5.27220488e-01 7.62391508e-01 -3.66232812e-01 -1.12667549e+00 1.33302116e+00 7.49025404e-01 -1.73024868e-03 2.58319110e-01 1.14123273e+00 1.11447299e+00 1.37075454e-01 -1.88394204e-01 -1.57786354e-01 1.47637796e+00 -4.76465434e-01 -6.58178627e-01 2.21333858e-02 5.90372086e-01 -8.58262420e-01 1.23999131e+00 8.03084135e-01 -1.16730523e+00 -6.98097467e-01 -9.72847283e-01 1.51282057e-01 -4.65095371e-01 1.62078023e-01 2.60533839e-01 1.13351476e+00 -1.00068676e+00 5.22715509e-01 -6.60425127e-01 -3.79417688e-01 5.44795871e-01 3.02473843e-01 -9.22067940e-01 -4.29112524e-01 -1.13554132e+00 1.08311749e+00 3.56711864e-01 -2.59149045e-01 -8.04635882e-01 -9.95433390e-01 -8.08507681e-01 -7.75549710e-01 1.09169498e-01 -6.38448179e-01 8.64857316e-01 -8.70927811e-01 -1.06521237e+00 1.52481568e+00 8.09086442e-01 -4.99872088e-01 1.47107625e+00 -1.56425938e-01 -6.60992146e-01 2.15239555e-01 7.10005611e-02 4.94687855e-01 4.85911608e-01 -1.56523192e+00 -3.08135450e-01 -4.31494653e-01 8.34094808e-02 -5.41036129e-02 7.37189949e-02 -3.04371752e-02 -1.95186168e-01 -6.65537775e-01 -4.68372524e-01 -1.18440902e+00 -4.93704289e-01 1.72138363e-01 -4.87450719e-01 9.36171353e-01 3.23733419e-01 -1.20881414e+00 8.19425762e-01 -2.21772408e+00 -3.31089884e-01 3.78075749e-01 1.58988789e-01 2.93408453e-01 1.33936211e-01 2.14745060e-01 -4.24395829e-01 5.22116432e-03 -1.43036559e-01 -3.86150442e-02 -2.46241748e-01 7.17399418e-02 -9.84729268e-03 6.27696812e-01 -1.00566514e-01 7.04423368e-01 -9.14241016e-01 -5.94786406e-01 5.24377108e-01 5.12573838e-01 -4.15984869e-01 1.96846932e-01 1.46144524e-01 8.37796330e-01 -3.10523242e-01 6.08337402e-01 1.04094839e+00 2.84609228e-01 4.01910841e-01 -3.05065423e-01 6.28141984e-02 -4.21835393e-01 -1.32417071e+00 2.03722358e+00 -3.89534205e-01 3.39309484e-01 9.03840140e-02 -4.85803992e-01 1.08871889e+00 4.16255713e-01 6.12649500e-01 -1.06636536e+00 1.80758759e-01 2.87779301e-01 -1.43034026e-01 -7.98288047e-01 4.62586313e-01 -3.60868245e-01 -3.28806520e-01 -1.16931498e-01 -3.47918063e-01 -2.58199304e-01 -4.41161335e-01 -7.44458362e-02 1.28409350e+00 3.25873703e-01 2.68387228e-01 -3.02730173e-01 -3.05122919e-02 5.56691825e-01 1.02721997e-01 4.96900022e-01 -2.39532411e-01 1.17911947e+00 3.63549203e-01 -7.17735469e-01 -1.22590077e+00 -1.18877852e+00 -2.69480109e-01 1.79820210e-01 4.10182215e-02 2.75067259e-02 -9.61159289e-01 -1.06905186e+00 -1.33463591e-01 9.11641777e-01 -9.01081681e-01 -1.68019429e-01 -5.10417938e-01 -5.01962006e-01 6.11183226e-01 2.88511544e-01 3.23686123e-01 -7.63771415e-01 -1.20886993e+00 4.32187282e-02 -4.48709907e-04 -1.28750515e+00 -4.87451116e-03 -1.29988313e-01 -7.28839397e-01 -1.17674756e+00 -9.27227914e-01 -4.30139095e-01 7.65114188e-01 -3.09372663e-01 1.28272915e+00 -5.25425076e-01 -7.33711839e-01 4.16785091e-01 -5.08566201e-01 -6.39937103e-01 -6.80339217e-01 -4.48857605e-01 -1.24667808e-01 -2.11357504e-01 8.36996660e-02 -2.28543550e-01 -8.58114898e-01 5.63985407e-01 -8.76837075e-01 2.01226994e-01 4.32821006e-01 7.36003101e-01 6.31147861e-01 -4.86400425e-01 2.99260080e-01 -1.33050501e+00 3.61600161e-01 -4.46302235e-01 -5.26888490e-01 3.12492460e-01 -2.70154208e-01 -4.03256536e-01 4.78173852e-01 -7.00926185e-01 -7.71049023e-01 5.74918449e-01 -2.45752912e-02 -8.36088240e-01 -3.27753365e-01 5.05283214e-02 -1.37947217e-01 6.50173649e-02 1.41468728e+00 -3.12537134e-01 1.53232783e-01 -4.69139636e-01 2.98090458e-01 6.11905456e-01 4.33383733e-01 -3.85536492e-01 5.16643286e-01 4.98888850e-01 -1.98467880e-01 -3.36035401e-01 -1.92888111e-01 -1.52058586e-01 -3.64233166e-01 -3.77623796e-01 9.94821548e-01 -1.01141405e+00 -2.89635748e-01 7.63257742e-02 -8.45776081e-01 -4.75558817e-01 -8.69679511e-01 7.66435623e-01 -6.00807607e-01 9.03339535e-02 -3.12584229e-02 -8.17692816e-01 -9.74643230e-02 -1.15760195e+00 1.30663037e+00 -1.45954579e-01 -4.34998244e-01 -7.56436169e-01 1.25660151e-01 6.29754603e-01 2.11135820e-01 1.40384853e+00 2.37856761e-01 -8.43169093e-01 -1.04803145e-01 -6.19510770e-01 -2.02974677e-02 3.00980955e-01 1.19256832e-01 -3.10236305e-01 -1.15966046e+00 -3.55943531e-01 1.78275511e-01 -2.70286500e-01 7.97194093e-02 3.11224759e-01 8.71209860e-01 -1.38100281e-01 -3.50243241e-01 2.61447698e-01 1.53612888e+00 3.11380088e-01 1.06617546e+00 3.69528592e-01 4.89907056e-01 8.42353284e-01 1.40069926e+00 4.82258350e-01 1.22078322e-01 8.12029243e-01 4.78843063e-01 -7.64790535e-01 -1.29955903e-01 -3.32583487e-02 2.20095348e-02 1.50672719e-01 -4.26918082e-02 8.93635675e-02 -1.18749869e+00 6.07825279e-01 -1.51998103e+00 -6.91482425e-01 -3.51700872e-01 2.55392027e+00 7.87738264e-01 1.20937116e-01 4.65169042e-01 3.81443352e-02 6.97455287e-01 -4.33667034e-01 -3.55796516e-01 -5.86042047e-01 1.39072463e-01 3.14786285e-01 9.54099178e-01 2.09258899e-01 -1.15891194e+00 2.32196704e-01 6.37763596e+00 6.07649028e-01 -1.08640099e+00 2.50857115e-01 7.54623473e-01 -2.75814831e-01 -1.04714237e-01 -5.51587164e-01 -2.84462541e-01 5.56360841e-01 1.06932247e+00 2.52370536e-01 7.47002438e-02 1.04575169e+00 2.82446414e-01 -1.23745963e-01 -1.15219474e+00 1.21749854e+00 7.22289830e-02 -1.22741795e+00 -4.57081616e-01 3.20565462e-01 5.82689285e-01 -1.52271882e-01 1.52133062e-01 2.03571692e-01 3.61746848e-01 -1.17049026e+00 6.70553029e-01 6.48055136e-01 1.58031785e+00 -4.94073033e-01 9.11941290e-01 4.22733605e-01 -5.53141534e-01 3.41757149e-01 -9.06417072e-02 4.43644315e-01 -4.69247252e-02 5.11010766e-01 -1.49908268e+00 1.07665646e+00 7.41853774e-01 1.22312292e-01 -5.20247042e-01 9.64918256e-01 4.65483516e-01 1.84978113e-01 -3.44338894e-01 2.76701361e-01 -1.14737943e-01 -4.16887216e-02 3.06416601e-01 1.32778311e+00 2.05119595e-01 6.68608490e-03 7.87902772e-02 3.99215043e-01 2.77891666e-01 2.23992318e-01 -1.18658388e+00 3.54799926e-01 -5.51165752e-02 1.08985007e+00 -5.10649085e-01 -1.01642266e-01 -1.68960720e-01 1.03802776e+00 -2.05480978e-01 -9.13365409e-02 -1.37366271e+00 -1.64296731e-01 2.44679034e-01 6.43922269e-01 -5.33181317e-02 2.74419636e-01 -3.01379144e-01 -9.13847029e-01 2.24521488e-01 -1.22315192e+00 3.86593133e-01 -1.12870669e+00 -1.03695238e+00 9.50244308e-01 2.05485329e-01 -1.90733230e+00 -1.35529399e-01 -7.13775575e-01 2.03804404e-01 8.34380865e-01 -9.32260633e-01 -1.44406247e+00 -5.58067083e-01 5.93251348e-01 2.51537293e-01 -2.44715303e-01 1.19637156e+00 6.33923829e-01 -2.54265033e-02 8.14756572e-01 2.85074234e-01 -7.92750996e-03 8.61069560e-01 -1.29752111e+00 1.82747766e-01 3.19511443e-01 -9.31731611e-02 3.07423025e-01 8.44944119e-01 -7.52569854e-01 -1.21441543e+00 -1.21551323e+00 9.70540717e-02 -9.87940371e-01 2.18510196e-01 -7.22142637e-01 -5.81560254e-01 5.03743231e-01 2.75233266e-04 3.16243470e-01 7.76438475e-01 -5.14376342e-01 -1.25377193e-01 -1.27326980e-01 -2.24104810e+00 4.86403048e-01 9.16069984e-01 -2.75347501e-01 -5.13172388e-01 6.26711071e-01 6.42243624e-01 -1.05534673e+00 -1.44930017e+00 4.68758017e-01 8.22514772e-01 -1.00975227e+00 9.90169942e-01 -5.69603801e-01 4.13594395e-01 -1.68128610e-01 -3.71076137e-01 -1.13288975e+00 2.48502418e-01 -5.21960080e-01 2.27796793e-01 8.94707203e-01 5.32375336e-01 -4.99924779e-01 9.94366765e-01 1.08439863e+00 6.24146536e-02 -7.26985157e-01 -1.04052854e+00 -7.33629823e-01 -4.11707908e-02 -3.51164013e-01 4.04875934e-01 1.35444164e+00 -2.73181200e-01 -1.60609260e-01 -4.15480256e-01 1.35796696e-01 3.87300164e-01 -2.81434655e-01 1.06208014e+00 -8.07941198e-01 -3.15307558e-01 4.22067940e-01 -6.94838583e-01 -7.92507529e-02 -6.13798499e-01 -6.36874199e-01 -3.56127888e-01 -1.70895433e+00 -5.27302213e-02 -6.40011907e-01 3.05176969e-03 2.57649988e-01 3.37367624e-01 4.19724941e-01 1.52047902e-01 -2.28551582e-01 -9.38524082e-02 -2.35451639e-01 1.35929692e+00 2.09112931e-02 -5.61328791e-02 -8.85816813e-02 -6.01582944e-01 4.54823941e-01 7.06023216e-01 -5.07722080e-01 -5.59091687e-01 -5.96815683e-02 3.70642915e-02 2.17227265e-01 5.59643924e-01 -1.10293674e+00 -2.19457597e-01 6.46757558e-02 5.73466182e-01 -4.15652871e-01 5.30396998e-01 -1.60355377e+00 1.08750618e+00 7.45491922e-01 -7.47604668e-02 -2.00629562e-01 6.37078583e-01 5.39308071e-01 8.66707265e-02 1.92226488e-02 7.10746586e-01 -3.73688519e-01 -3.58444422e-01 -2.86170483e-01 2.10471645e-01 6.41263500e-02 1.73240411e+00 -8.76532733e-01 -2.68870145e-01 -3.51597853e-02 -1.27993751e+00 -3.30851674e-02 7.93537557e-01 4.24298972e-01 6.86968625e-01 -1.19379401e+00 -8.75653684e-01 1.51970863e-01 2.51212984e-01 5.66774309e-02 5.22816658e-01 6.96395159e-01 -9.75313783e-01 -1.07270643e-01 -5.66126883e-01 -5.85052609e-01 -1.60761476e+00 9.61871982e-01 4.95940506e-01 -4.30581868e-01 -6.06150687e-01 4.62341636e-01 2.64687061e-01 -8.57690096e-01 -3.88807282e-02 -4.19216305e-01 2.41946410e-02 -5.48873134e-02 4.29912120e-01 2.99090654e-01 4.44185197e-01 -3.39137465e-01 -5.45370579e-01 4.15968299e-01 1.28436148e-01 -3.05546105e-01 1.22779346e+00 3.84665191e-01 5.52149951e-01 2.33301878e-01 1.05335784e+00 2.84085840e-01 -9.91256952e-01 4.18088853e-01 -3.56853247e-01 -8.87718916e-01 -3.65477294e-01 -1.48721504e+00 -9.21602190e-01 6.29899144e-01 1.29610193e+00 8.74228552e-02 1.33210075e+00 -1.74341232e-01 2.36008614e-01 -1.87996626e-01 8.49502265e-01 -9.00195181e-01 -1.70285746e-01 -4.86164033e-01 1.15594637e+00 -1.26244020e+00 1.75463989e-01 -6.04023993e-01 -1.21965075e+00 7.52873659e-01 5.06183982e-01 -2.18542114e-01 4.33301896e-01 4.91680682e-01 3.01955968e-01 -3.84967446e-01 -2.50749458e-02 1.89040840e-01 1.30757049e-01 1.11302769e+00 1.71310157e-02 2.19309732e-01 -2.05418080e-01 6.68046296e-01 -2.60949552e-01 4.00648445e-01 5.28181255e-01 9.91817713e-01 3.66842598e-01 -9.53753769e-01 -6.91969514e-01 5.57939410e-01 -6.63002610e-01 1.80920303e-01 -4.89667624e-01 1.34862864e+00 4.80373740e-01 6.87081158e-01 -8.68245587e-02 -4.85227317e-01 1.11483097e+00 -4.06924784e-01 5.86822629e-01 -5.55240333e-01 -1.29697430e+00 -9.95514318e-02 6.32805705e-01 -5.47964215e-01 -2.06076100e-01 -5.05649567e-01 -7.75939405e-01 -1.93653181e-01 -1.89732701e-01 -1.47244275e-01 9.38151717e-01 7.70188272e-02 4.23662215e-01 6.56234860e-01 4.62792993e-01 -4.34435576e-01 -3.86049211e-01 -6.77991986e-01 -5.29774606e-01 5.47771454e-01 1.96393088e-01 -4.73003536e-01 1.19997077e-01 1.22017287e-01]

[14.346242904663086, -1.9897027015686035]

9b953d6c-c0bc-4113-81d9-7c6a0089fefb

camera-aware-proxies-for-unsupervised-person

2012.10674

https://arxiv.org/abs/2012.10674v2

https://arxiv.org/pdf/2012.10674v2.pdf

Camera-aware Proxies for Unsupervised Person Re-Identification

This paper tackles the purely unsupervised person re-identification (Re-ID) problem that requires no annotations. Some previous methods adopt clustering techniques to generate pseudo labels and use the produced labels to train Re-ID models progressively. These methods are relatively simple but effective. However, most clustering-based methods take each cluster as a pseudo identity class, neglecting the large intra-ID variance caused mainly by the change of camera views. To address this issue, we propose to split each single cluster into multiple proxies and each proxy represents the instances coming from the same camera. These camera-aware proxies enable us to deal with large intra-ID variance and generate more reliable pseudo labels for learning. Based on the camera-aware proxies, we design both intra- and inter-camera contrastive learning components for our Re-ID model to effectively learn the ID discrimination ability within and across cameras. Meanwhile, a proxy-balanced sampling strategy is also designed, which facilitates our learning further. Extensive experiments on three large-scale Re-ID datasets show that our proposed approach outperforms most unsupervised methods by a significant margin. Especially, on the challenging MSMT17 dataset, we gain $14.3\%$ Rank-1 and $10.2\%$ mAP improvements when compared to the second place. Code is available at: \texttt{https://github.com/Terminator8758/CAP-master}.

['Xian-Sheng Hua', 'Xiaojin Gong', 'Jianqiang Huang', 'Baisheng Lai', 'Menglin Wang']

2020-12-19

['unsupervised-person-re-identification']

['computer-vision']

[-1.14179747e-02 -3.19265902e-01 -2.19597653e-01 -5.86621284e-01 -8.97224188e-01 -5.99030852e-01 6.58849537e-01 -1.06030531e-01 -5.22709191e-01 6.47832930e-01 1.88082635e-01 2.19385535e-01 -7.65567347e-02 -4.34123874e-01 -4.94089186e-01 -6.98388755e-01 2.65258163e-01 7.12345660e-01 2.67724134e-02 4.69377398e-01 1.55552998e-01 8.73043612e-02 -1.54987741e+00 9.36166197e-02 1.03904200e+00 6.85764730e-01 1.57497704e-01 3.35324019e-01 -3.40431258e-02 5.75913668e-01 -6.09601557e-01 -7.06832588e-01 3.88161540e-01 -4.95433480e-01 -6.73889160e-01 2.65944004e-01 5.37171483e-01 -3.01714927e-01 -2.50893921e-01 1.25684643e+00 6.48948431e-01 1.91892534e-01 6.64074242e-01 -1.35731530e+00 -6.15581632e-01 6.29883647e-01 -8.82053375e-01 -4.23562787e-02 3.73618126e-01 -8.13991949e-02 8.09157073e-01 -7.98365593e-01 3.30471158e-01 1.16535139e+00 7.76169956e-01 7.92161465e-01 -1.28880715e+00 -1.16667604e+00 1.84835494e-01 3.77311021e-01 -1.83502448e+00 -4.88226056e-01 8.01823854e-01 -4.43855494e-01 1.40956163e-01 3.09580207e-01 1.95279971e-01 1.17865622e+00 -7.58186221e-01 8.29842746e-01 1.26661110e+00 -4.16907489e-01 3.87822581e-03 4.14663583e-01 1.97861195e-01 3.64579618e-01 2.65746057e-01 -9.54384431e-02 -2.74081826e-01 -1.11278862e-01 5.76527834e-01 3.48535210e-01 -1.88144848e-01 -2.47475490e-01 -1.35500002e+00 5.12677073e-01 3.83917361e-01 1.23604923e-01 3.64525132e-02 -6.11248687e-02 2.81828493e-01 -1.49707019e-01 2.10201070e-01 1.02983505e-01 -1.14774190e-01 -1.85308903e-01 -9.69169319e-01 -6.08393997e-02 4.93407041e-01 1.19565666e+00 8.78848493e-01 -3.65892559e-01 -1.36079624e-01 1.30686033e+00 2.20600426e-01 5.41550636e-01 4.93318975e-01 -1.04939055e+00 5.17321527e-01 5.61071932e-01 4.36671615e-01 -1.01016378e+00 -2.95205176e-01 -5.12290359e-01 -1.07097995e+00 -2.33755529e-01 5.41484118e-01 -1.74832210e-01 -7.78163612e-01 1.81582999e+00 3.07065755e-01 5.49892843e-01 -2.23594800e-01 8.19579840e-01 6.63756132e-01 3.92120153e-01 1.63397953e-01 -1.97604835e-01 1.36511827e+00 -1.05024648e+00 -4.59034592e-01 -1.90624759e-01 5.08080900e-01 -5.98656476e-01 7.90030777e-01 2.91259646e-01 -6.83445275e-01 -9.63631094e-01 -9.43351507e-01 2.45571136e-01 -1.97706804e-01 5.92020869e-01 1.92574322e-01 7.46603489e-01 -1.04505563e+00 2.81952649e-01 -4.27876890e-01 -4.85913545e-01 3.37408423e-01 4.61602330e-01 -3.40463787e-01 -2.93671846e-01 -9.88592982e-01 3.31495106e-01 4.24247533e-01 -7.58230388e-02 -7.11101830e-01 -4.37562317e-01 -4.35630113e-01 -1.60978764e-01 4.02250916e-01 -3.30826074e-01 9.91291761e-01 -9.61495519e-01 -1.16226232e+00 1.08301139e+00 -4.11018789e-01 -1.41603053e-02 6.22807205e-01 -8.59760419e-02 -6.66127622e-01 7.37653896e-02 5.07522166e-01 7.02750623e-01 6.96760356e-01 -1.79922843e+00 -8.89120519e-01 -4.54357982e-01 -1.15282446e-01 2.16904894e-01 -5.73183477e-01 -7.78292120e-02 -1.03065979e+00 -7.48619795e-01 1.37089137e-02 -1.30277848e+00 -2.37516314e-03 -3.79012465e-01 -5.03671527e-01 -2.82580048e-01 3.80960405e-01 -6.28498018e-01 1.33698952e+00 -2.19232512e+00 -6.61580786e-02 2.01735824e-01 2.94292301e-01 3.82227540e-01 -4.22630757e-02 2.26184264e-01 -1.60404235e-01 1.57515883e-01 -1.25387043e-01 -7.47654378e-01 2.48529445e-02 -9.54571068e-02 1.74954578e-01 3.64246488e-01 -3.78462672e-01 4.83575195e-01 -8.80853534e-01 -6.68272555e-01 2.42645577e-01 4.13279057e-01 -2.87707120e-01 3.81257564e-01 3.51978630e-01 6.99999034e-01 -2.95686394e-01 6.25498891e-01 9.34465647e-01 -3.52380037e-01 3.76475513e-01 -3.24765533e-01 2.33837962e-02 -2.35304430e-01 -1.55138445e+00 1.48741674e+00 -2.44221389e-01 1.77320227e-01 -6.99193776e-02 -9.71467137e-01 9.88310277e-01 1.60936967e-01 5.61535299e-01 -5.46535373e-01 4.43994738e-02 1.71341240e-01 -2.76005775e-01 -1.79761097e-01 3.16785336e-01 5.59305102e-02 -1.47200167e-01 5.88969886e-01 -1.00118384e-01 7.37709820e-01 2.06391096e-01 1.58486292e-01 6.03543997e-01 1.56733468e-02 1.20378390e-01 -1.18362956e-01 7.68074989e-01 -2.27896869e-01 8.32642555e-01 8.81189644e-01 -4.92236018e-01 8.25201273e-01 4.78851683e-02 -3.10893744e-01 -9.72735941e-01 -1.08287215e+00 -1.86415732e-01 1.08333826e+00 5.52529097e-01 -4.01447982e-01 -9.38803792e-01 -7.80654430e-01 -2.89055705e-02 4.69812125e-01 -4.98122543e-01 2.95685399e-02 -4.08904433e-01 -9.91409719e-01 6.32751882e-01 5.11098564e-01 9.42561686e-01 -5.84112108e-01 1.65988728e-01 -6.48549851e-03 -6.60931706e-01 -1.16348815e+00 -7.24112332e-01 -2.82832980e-01 -6.14671052e-01 -1.08744979e+00 -1.00804901e+00 -9.74082828e-01 9.47145700e-01 7.14385867e-01 7.56166935e-01 1.08552724e-01 -7.91998804e-02 4.61918622e-01 -4.96541977e-01 3.65563296e-02 -1.16759472e-01 1.81264549e-01 4.46716368e-01 5.23401499e-01 8.52843821e-01 -5.32346010e-01 -7.53587067e-01 8.26932669e-01 -5.11928260e-01 -8.95944331e-03 4.25177455e-01 7.92721093e-01 6.10214770e-01 2.55824238e-01 5.38507879e-01 -9.34437454e-01 3.96246105e-01 -5.34016371e-01 -3.74078900e-01 4.45189893e-01 -7.56962776e-01 -1.81097552e-01 6.59429789e-01 -5.60762107e-01 -1.18230677e+00 1.85875833e-01 1.01941213e-01 -3.43036950e-01 -4.69885409e-01 -9.90218595e-02 -5.18002748e-01 1.67603806e-01 4.31389600e-01 2.90279299e-01 -1.94331482e-01 -6.92537069e-01 2.53369480e-01 1.20536852e+00 6.93062365e-01 -7.43679464e-01 1.00322688e+00 5.58705628e-01 -4.96938914e-01 -5.11620283e-01 -7.67279327e-01 -8.85347009e-01 -9.28012431e-01 -2.52204329e-01 8.88136923e-01 -1.31169319e+00 -8.24176729e-01 6.73025250e-01 -7.20102429e-01 -2.03069896e-01 1.43552989e-01 5.39759636e-01 -1.85043991e-01 6.51985049e-01 -4.25755382e-01 -7.85208702e-01 -1.84315577e-01 -1.14120996e+00 8.64655912e-01 5.12342215e-01 -1.54115319e-01 -8.51652920e-01 -1.64622795e-02 7.17216015e-01 3.95606682e-02 -2.70204386e-03 3.37796748e-01 -7.36692011e-01 -4.31502700e-01 -3.64924073e-01 -5.62017679e-01 3.52928162e-01 4.21868742e-01 -3.33163321e-01 -1.17762041e+00 -5.08469880e-01 -3.28532100e-01 -1.72528878e-01 7.05100179e-01 -7.34066516e-02 1.27701569e+00 -2.40599632e-01 -6.13402069e-01 7.37602890e-01 1.39527345e+00 2.58012891e-01 4.97293264e-01 3.97659451e-01 1.21190786e+00 5.41618645e-01 5.22148132e-01 5.97659767e-01 8.03629041e-01 9.14704978e-01 -3.22635844e-02 3.85121442e-02 -1.42265156e-01 -4.23840135e-01 2.24921808e-01 8.52921069e-01 -3.19387704e-01 -1.49533376e-01 -9.45903778e-01 5.27850330e-01 -1.83435309e+00 -1.05205393e+00 -9.78011936e-02 2.57260394e+00 7.56062388e-01 -1.07111134e-01 5.38994074e-01 7.51255974e-02 1.36681819e+00 -8.59315768e-02 -5.85161507e-01 3.58486056e-01 8.61985609e-02 -2.73454249e-01 5.93798399e-01 4.26460475e-01 -1.26160192e+00 7.75692821e-01 5.23763227e+00 9.15347517e-01 -6.11416399e-01 1.71469584e-01 7.64386833e-01 7.88001716e-02 1.05890527e-01 -5.81624657e-02 -1.13029528e+00 1.01964796e+00 7.64629900e-01 -3.44537422e-02 5.45433342e-01 7.76334286e-01 4.08809632e-03 -7.71153644e-02 -1.01455390e+00 1.59812605e+00 2.95879483e-01 -7.20386744e-01 -2.02466864e-02 1.23475209e-01 8.48675430e-01 -2.79195070e-01 -4.68253717e-02 1.89434499e-01 4.41208571e-01 -7.34721839e-01 5.77080607e-01 4.66426104e-01 1.02575660e+00 -8.71814013e-01 7.84742117e-01 2.95381874e-01 -1.45737898e+00 -3.47232074e-01 -3.89536619e-01 2.29684561e-01 -9.72654670e-03 4.99119073e-01 -5.43550372e-01 6.83415651e-01 1.04766500e+00 8.93726230e-01 -1.00712693e+00 1.06549239e+00 -1.90003946e-01 4.38264579e-01 -7.46044964e-02 3.07456642e-01 -2.72728473e-01 -2.96297193e-01 2.08107516e-01 1.25106788e+00 4.30048317e-01 7.45385513e-02 3.44530612e-01 4.86206591e-01 -2.35832959e-01 -1.01837605e-01 -1.48541853e-01 3.94784808e-01 9.16800797e-01 1.26735353e+00 -7.00833142e-01 -5.14983714e-01 -4.90457892e-01 1.32343376e+00 3.92424941e-01 4.85155642e-01 -1.08074713e+00 -3.13952893e-01 5.72688639e-01 2.09347501e-01 2.57030785e-01 -9.54034179e-02 -1.53413098e-02 -1.37160456e+00 7.01081455e-02 -8.87806654e-01 6.97106242e-01 -4.93884712e-01 -1.59157467e+00 5.01874685e-01 2.34820843e-01 -1.60005867e+00 -8.26523602e-02 -3.50141078e-01 -3.50488812e-01 6.89676702e-01 -1.30761230e+00 -1.25070381e+00 -7.24672854e-01 7.03939021e-01 2.57677674e-01 -2.89030880e-01 6.09915257e-01 7.52630949e-01 -9.45812225e-01 1.17446089e+00 4.48784620e-01 5.62374949e-01 1.24468422e+00 -1.16644096e+00 1.37042075e-01 8.78291845e-01 1.75426677e-02 8.14449191e-01 1.89529583e-01 -5.78658640e-01 -9.57605779e-01 -1.14691687e+00 7.06711531e-01 -6.55964077e-01 2.34282598e-01 -5.24322510e-01 -6.97172880e-01 7.02352285e-01 -8.63596126e-02 -7.43850097e-02 9.60716128e-01 1.89365208e-01 -5.22670090e-01 -5.84009826e-01 -1.15086257e+00 4.95140821e-01 1.33264005e+00 -5.24366915e-01 -3.28475296e-01 1.41769111e-01 2.90553331e-01 8.82169232e-03 -9.76924598e-01 2.34069481e-01 5.36061406e-01 -1.00475276e+00 1.16140497e+00 2.32307501e-02 -3.62865403e-02 -6.87523723e-01 -3.51399258e-02 -1.00097847e+00 -6.45344198e-01 -4.06208634e-01 8.72783214e-02 1.99031961e+00 8.27228352e-02 -7.17903316e-01 7.71301091e-01 8.32119584e-01 2.50614762e-01 -8.20102692e-02 -7.36012757e-01 -9.04300570e-01 -1.84344158e-01 -1.11765042e-01 7.83119738e-01 1.18624187e+00 -1.62024632e-01 2.21574679e-01 -7.13820696e-01 2.72091448e-01 1.12374520e+00 -1.95888635e-02 1.04446065e+00 -1.35884726e+00 -2.22338706e-01 -2.33161867e-01 -2.53279150e-01 -1.16646278e+00 -3.18022296e-02 -8.90868843e-01 5.86544536e-02 -1.32537091e+00 7.10406601e-01 -9.44982767e-01 -4.11250353e-01 4.27660882e-01 -5.49016654e-01 5.61914742e-01 3.50641608e-01 8.72150719e-01 -9.54142451e-01 1.79368809e-01 8.61268103e-01 -1.45852491e-01 -1.37050509e-01 1.95929762e-02 -9.08509970e-01 5.92959404e-01 9.17810678e-01 -4.72427845e-01 -4.20450360e-01 -4.80894297e-01 -2.95471162e-01 -4.47261125e-01 3.83872777e-01 -1.35577726e+00 4.52774316e-01 1.68924201e-02 7.13784277e-01 -4.67110097e-01 1.72579378e-01 -8.16380203e-01 4.97346938e-01 1.50330484e-01 -2.61399388e-01 -4.38425876e-02 -2.12952837e-01 6.77254915e-01 -1.18705511e-01 -1.62663803e-01 8.15473557e-01 -3.02523047e-01 -8.25785518e-01 4.92844731e-01 7.80027211e-02 1.32654026e-01 1.06837308e+00 -2.94046462e-01 -2.21236631e-01 -3.63057017e-01 -6.13032043e-01 3.99027616e-01 7.17957675e-01 4.38739181e-01 2.50764787e-01 -1.55174220e+00 -7.08101511e-01 3.83480713e-02 3.96206468e-01 -7.55246580e-02 5.01108646e-01 6.42574549e-01 -1.41812727e-01 2.20628321e-01 -8.27742666e-02 -7.24254191e-01 -1.24574685e+00 7.12311983e-01 1.96804494e-01 -1.61992550e-01 -2.65580893e-01 7.86335409e-01 3.04606974e-01 -6.11915231e-01 4.11769390e-01 5.24976790e-01 -3.13846052e-01 2.49209777e-01 7.94691086e-01 6.43346250e-01 -4.05567914e-01 -1.00377786e+00 -5.25088966e-01 8.73235703e-01 -2.46908635e-01 5.27207106e-02 9.95563745e-01 -4.94316369e-01 -4.12118882e-02 2.95558989e-01 1.30454719e+00 2.11450458e-02 -1.32355177e+00 -4.10914838e-01 1.19222082e-01 -6.58457518e-01 -5.70548117e-01 -6.57081842e-01 -1.02087188e+00 6.44167781e-01 9.06857073e-01 -1.13311652e-02 1.11741686e+00 1.70824816e-03 7.74166822e-01 8.32561702e-02 4.91220415e-01 -1.33921742e+00 2.00137958e-01 6.11557551e-02 3.07339549e-01 -1.51785517e+00 1.36941612e-01 -3.39802325e-01 -7.17773318e-01 7.08462715e-01 8.42852175e-01 1.40824258e-01 4.30380195e-01 -2.29702204e-01 1.71055391e-01 3.11278015e-01 6.72180876e-02 -3.05733770e-01 1.07768983e-01 8.50422144e-01 2.03807116e-01 3.02876830e-01 -1.21380053e-01 6.99807286e-01 1.21303815e-02 -7.26903155e-02 2.57894933e-01 5.54537237e-01 -1.17792405e-01 -1.52267230e+00 -6.85138583e-01 2.81720221e-01 -3.01038653e-01 8.33542421e-02 -3.25789243e-01 6.33271277e-01 5.33627987e-01 1.12113237e+00 3.51216458e-02 -8.46835017e-01 1.75181523e-01 -2.38474039e-03 2.37018272e-01 -4.58680898e-01 -3.23078156e-01 -2.36111716e-03 -2.33469978e-02 -2.71380007e-01 -7.98662782e-01 -8.30496550e-01 -9.54631567e-01 -4.89710599e-01 -1.50112927e-01 2.34249696e-01 2.73795456e-01 6.04009330e-01 4.81203496e-01 1.23929366e-01 9.87374842e-01 -7.58214116e-01 -2.89473385e-01 -9.24305379e-01 -5.48522949e-01 8.02926481e-01 9.27224532e-02 -7.12290645e-01 -5.33061385e-01 2.83281803e-01]

[14.814281463623047, 1.0823616981506348]

23600fb0-9bf7-4383-a97d-bd517147a820

accelerating-self-play-learning-in-go

1902.10565

https://arxiv.org/abs/1902.10565v5

https://arxiv.org/pdf/1902.10565v5.pdf

Accelerating Self-Play Learning in Go

By introducing several improvements to the AlphaZero process and architecture, we greatly accelerate self-play learning in Go, achieving a 50x reduction in computation over comparable methods. Like AlphaZero and replications such as ELF OpenGo and Leela Zero, our bot KataGo only learns from neural-net-guided Monte Carlo tree search self-play. But whereas AlphaZero required thousands of TPUs over several days and ELF required thousands of GPUs over two weeks, KataGo surpasses ELF's final model after only 19 days on fewer than 30 GPUs. Much of the speedup involves non-domain-specific improvements that might directly transfer to other problems. Further gains from domain-specific techniques reveal the remaining efficiency gap between the best methods and purely general methods such as AlphaZero. Our work is a step towards making learning in state spaces as large as Go possible without large-scale computational resources.

['David J. Wu']

2019-02-27

['game-of-go']

['playing-games']

[-4.76631373e-01 1.94001049e-02 -5.19728899e-01 1.79414839e-01 -9.49858963e-01 -7.76610911e-01 3.80899251e-01 -4.27121930e-02 -6.21051371e-01 1.04572630e+00 -1.36698514e-01 -8.31050634e-01 -7.85615157e-06 -1.01269305e+00 -8.68757188e-01 -4.34499502e-01 -3.95444125e-01 7.58829892e-01 7.46577442e-01 -3.49213928e-01 2.20950663e-01 6.72996566e-02 -1.43307114e+00 2.63052702e-01 6.23902678e-01 5.91962576e-01 -1.15876034e-01 1.04209745e+00 -2.65021296e-03 9.88438487e-01 -4.60328430e-01 1.57971354e-03 3.50199133e-01 -2.68591911e-01 -1.07172000e+00 -4.38995600e-01 3.98922086e-01 -6.00143194e-01 -5.21648288e-01 8.54411304e-01 4.44148630e-01 3.45443308e-01 1.06159627e-01 -1.22236240e+00 9.57392007e-02 8.66320252e-01 -4.76138026e-01 3.50257874e-01 4.10111854e-03 9.34133768e-01 9.73661184e-01 -3.76371853e-02 4.34822857e-01 1.34745359e+00 1.00586653e+00 5.61402857e-01 -1.47142124e+00 -8.67755234e-01 6.24183044e-02 -1.39967026e-02 -8.96416962e-01 -2.02178791e-01 -1.05721373e-02 -1.82630002e-01 1.38746011e+00 1.09615929e-01 1.01327682e+00 1.20532000e+00 2.63754964e-01 1.01663518e+00 1.25599682e+00 -2.39840224e-01 5.76840341e-01 -4.61402804e-01 2.53174126e-01 1.05948305e+00 3.70852858e-01 4.61757898e-01 -3.97439390e-01 -6.16536915e-01 1.15613186e+00 -3.68084610e-01 3.07183355e-01 -2.91881323e-01 -1.20838976e+00 9.24415350e-01 2.45284453e-01 -6.37180433e-02 -1.94220528e-01 1.14502835e+00 7.73361683e-01 4.43519592e-01 2.87669480e-01 6.62153840e-01 -8.01792145e-01 -1.00823247e+00 -1.00684345e+00 7.64415741e-01 1.30549312e+00 6.97948277e-01 9.38394606e-01 4.50941175e-01 2.45737165e-01 1.16835386e-01 -1.38682619e-01 3.42761427e-01 3.90812963e-01 -1.72552037e+00 3.22899461e-01 1.35723978e-01 1.97334573e-01 -4.70894843e-01 -4.47049856e-01 -6.67083979e-01 -5.28454244e-01 5.74780643e-01 7.84056127e-01 -6.26089096e-01 -7.31762826e-01 1.64186394e+00 3.91065836e-01 4.78236318e-01 4.80152527e-03 5.28329909e-01 1.43435434e-01 6.88240826e-01 -5.47703728e-02 1.99552864e-01 1.24530566e+00 -1.28405654e+00 5.10986261e-02 -4.89609510e-01 1.03821659e+00 -2.64854550e-01 1.36689413e+00 5.96921921e-01 -1.35944581e+00 -6.99908659e-02 -8.56233180e-01 3.93152162e-02 3.19893658e-02 -1.59254864e-01 1.48178530e+00 7.40648746e-01 -1.08518636e+00 9.95671153e-01 -1.51150215e+00 -3.34859073e-01 4.08580393e-01 6.38440967e-01 1.11778744e-01 2.63160378e-01 -1.00276411e+00 7.18172431e-01 6.98144138e-01 -4.95193809e-01 -1.37129843e+00 -9.62597072e-01 -3.50782722e-01 3.00495535e-01 8.66945744e-01 -9.79637623e-01 1.82415104e+00 -6.13945067e-01 -1.78681791e+00 4.97558683e-01 -3.03966533e-02 -9.84488964e-01 6.62943006e-01 -2.62026668e-01 3.51016015e-01 -8.53126571e-02 3.78623419e-02 7.54765391e-01 4.06736374e-01 -7.34136581e-01 -8.59182715e-01 -1.39942728e-02 5.81606865e-01 2.36922558e-02 -4.38953079e-02 -6.71964288e-02 -3.25652957e-01 -2.04406992e-01 -1.78818971e-01 -1.08896363e+00 -6.57467127e-01 -2.33627856e-01 9.69586894e-02 -4.74313170e-01 5.71889341e-01 -2.62616396e-01 9.21191335e-01 -1.59075677e+00 -1.59916386e-01 7.08062993e-03 3.83169055e-01 3.25700641e-01 -2.77455539e-01 5.00113010e-01 3.68375063e-01 9.72912386e-02 1.24082774e-01 1.54835314e-01 8.25762227e-02 5.16795337e-01 -4.08877283e-01 1.64138108e-01 -1.80277884e-01 8.96214187e-01 -1.39738393e+00 -3.05870742e-01 7.04115033e-02 -1.81453243e-01 -1.16625118e+00 -2.15295225e-01 -7.69213974e-01 2.56401300e-01 -6.13522887e-01 2.70762324e-01 2.41869181e-01 -4.76383388e-01 1.83458224e-01 4.05941218e-01 -2.15714976e-01 6.98461771e-01 -9.78176951e-01 1.83406460e+00 -6.48787320e-01 4.66445357e-01 1.65885836e-01 -8.22206259e-01 3.79571706e-01 7.90217426e-03 4.29680288e-01 -2.64867127e-01 2.23505348e-01 1.50997043e-01 6.59817010e-02 -1.73495412e-01 5.82828641e-01 1.80145144e-01 -1.97879106e-01 7.11393774e-01 9.94066596e-02 -3.59389395e-01 5.79399288e-01 2.18266472e-01 1.54944944e+00 4.32652056e-01 2.08739474e-01 -3.53420079e-01 -1.26447558e-01 7.62706101e-01 5.70441961e-01 1.24129450e+00 -2.35584959e-01 -2.12038621e-01 9.89160478e-01 -7.05015659e-01 -1.21165752e+00 -7.98165500e-01 2.69791991e-01 1.48558164e+00 -6.00909814e-02 -9.32772040e-01 -9.68623877e-01 -6.91186666e-01 9.28470641e-02 7.83224285e-01 -3.59958261e-01 3.55441943e-02 -6.06546998e-01 -7.14837968e-01 1.15627444e+00 6.14092171e-01 9.88748968e-01 -7.99680948e-01 -7.32539117e-01 4.75847155e-01 -1.66923199e-02 -5.93049109e-01 -2.19709888e-01 4.97701079e-01 -1.38299286e+00 -9.63187039e-01 -3.44440848e-01 -4.05006081e-01 2.01217197e-02 2.05399603e-01 1.24522936e+00 -8.13231021e-02 -2.23910436e-01 -8.45418945e-02 2.18200106e-02 -1.51452333e-01 -4.66702998e-01 7.11097956e-01 3.86443697e-02 -1.20110142e+00 3.49309742e-01 -9.18969750e-01 -4.82616097e-01 5.28733954e-02 -3.58072817e-01 1.27351001e-01 3.76910299e-01 1.10848558e+00 7.55955726e-02 1.11139439e-01 2.04214290e-01 -8.74672294e-01 7.48593688e-01 -5.57213068e-01 -9.80839550e-01 -1.51148379e-01 -6.40522420e-01 3.56275976e-01 5.59989274e-01 -6.96438849e-01 -7.69108832e-01 -7.76697174e-02 -1.48340076e-01 -4.64617342e-01 4.79473025e-02 3.35200787e-01 5.50861418e-01 -8.55349228e-02 1.27733994e+00 1.38366610e-01 -2.82159634e-03 -3.22418243e-01 3.94615173e-01 1.81113288e-01 5.01421809e-01 -1.17728853e+00 5.17461956e-01 1.74406469e-01 4.30314423e-04 -4.33102220e-01 -6.33372486e-01 -2.10353225e-01 -5.14120534e-02 1.85652420e-01 5.71180403e-01 -1.00718045e+00 -1.67252910e+00 4.15251911e-01 -9.92730916e-01 -1.10219860e+00 -3.71905565e-01 3.05081546e-01 -7.56775081e-01 2.59396106e-01 -1.05923593e+00 -6.87921703e-01 -3.40015799e-01 -9.88634348e-01 8.18287969e-01 5.34588881e-02 -3.17329884e-01 -8.50748956e-01 3.58430952e-01 4.16815549e-01 4.17540938e-01 -1.87080771e-01 7.32451379e-01 -4.06814665e-01 -8.37847173e-01 -1.03558358e-02 -6.44004941e-02 5.27216755e-02 -3.11127037e-01 -9.32403132e-02 -6.03214860e-01 -5.97813666e-01 -3.46376479e-01 -8.86031628e-01 9.50396478e-01 1.81989834e-01 1.08577394e+00 -3.81885350e-01 -3.29186976e-01 6.95443928e-01 1.28453219e+00 1.89544559e-01 3.49504054e-01 8.18270981e-01 4.40886527e-01 2.15692520e-02 6.20516658e-01 4.45849121e-01 4.30950671e-01 3.17769349e-01 5.87853074e-01 2.95488108e-02 2.05846697e-01 -5.19712448e-01 6.09305143e-01 5.04911125e-01 -3.62012357e-01 -6.12055473e-02 -1.28890312e+00 5.49797475e-01 -2.06582570e+00 -1.08157992e+00 -1.02775261e-01 2.03659177e+00 1.14311504e+00 4.05439705e-01 4.92109805e-01 -1.56386524e-01 2.12673396e-01 1.75118729e-01 -9.68017399e-01 -5.22159398e-01 3.98499876e-01 5.24783075e-01 1.01329052e+00 7.45382726e-01 -9.86248910e-01 1.67755127e+00 7.75654650e+00 1.35344541e+00 -1.14943981e+00 3.04621071e-01 4.33159620e-01 -4.21104223e-01 7.44944662e-02 5.10697901e-01 -1.00407088e+00 2.99427599e-01 1.35729122e+00 -3.02181005e-01 8.86754215e-01 1.35665905e+00 1.67237803e-01 -4.39325660e-01 -8.34651291e-01 6.16605401e-01 -5.82712531e-01 -1.62197614e+00 -4.08141345e-01 3.42193782e-01 9.84683633e-01 6.28366351e-01 -2.41557866e-01 6.89329147e-01 1.69353318e+00 -8.81162047e-01 3.83543193e-01 -5.70619367e-02 5.70190609e-01 -1.01632082e+00 3.80982697e-01 8.79259706e-01 -9.27070439e-01 -1.91912070e-01 -6.09795690e-01 -7.60932565e-01 -2.34345719e-01 2.96280056e-01 -9.22608554e-01 5.82534857e-02 7.00121343e-01 4.15821910e-01 -2.60021538e-01 8.29043865e-01 -3.95035625e-01 1.25927627e+00 -7.80211568e-01 -2.25163504e-01 8.90239894e-01 2.49396497e-03 4.37098503e-01 1.02617955e+00 5.67386970e-02 -1.19409924e-02 5.48106909e-01 7.54475653e-01 1.38023913e-01 -5.14998436e-01 -4.21233356e-01 -1.06564105e-01 5.72788417e-01 1.09556723e+00 -5.80111980e-01 -6.78699434e-01 2.67655011e-02 7.17620492e-01 4.50417221e-01 1.34925485e-01 -9.69197333e-01 -3.55136842e-01 8.96960735e-01 1.33524030e-01 2.11355209e-01 -6.43827677e-01 -3.90457243e-01 -1.33904970e+00 -4.62449998e-01 -1.14376605e+00 1.35177970e-01 -7.19030440e-01 -7.98277915e-01 2.65623271e-01 8.57839659e-02 -7.72891104e-01 -7.37034500e-01 -4.64772016e-01 -6.56643569e-01 5.98483622e-01 -1.12827992e+00 -9.90283310e-01 7.07489848e-02 5.23302197e-01 5.45721531e-01 -9.73662585e-02 7.81023085e-01 -1.14462808e-01 -4.85046059e-01 5.95861852e-01 6.21149659e-01 -2.61624902e-01 4.48129207e-01 -1.38677287e+00 1.04488146e+00 5.98315299e-01 -1.55850708e-01 6.76906824e-01 8.04514349e-01 -7.15240002e-01 -1.38381803e+00 -8.37866247e-01 5.26260316e-01 -3.37228417e-01 1.25985563e+00 -3.27199221e-01 -5.63333035e-01 1.01455390e+00 1.30578801e-01 -1.04130372e-01 8.51441696e-02 6.12635493e-01 -3.89616251e-01 3.83457765e-02 -8.33166122e-01 7.87464976e-01 1.32916760e+00 -3.55717957e-01 -4.19074565e-01 4.67415184e-01 5.27576983e-01 -7.22439289e-01 -6.83511376e-01 -1.27630487e-01 5.38415432e-01 -9.27846670e-01 9.84529257e-01 -9.61188257e-01 4.09387708e-01 -5.55602796e-02 1.67799428e-01 -1.23009229e+00 -4.01284635e-01 -1.02156675e+00 -3.32462341e-01 7.24105358e-01 -2.66596414e-02 -9.46108520e-01 1.40298307e+00 2.12281644e-01 5.97911365e-02 -6.71724260e-01 -7.94816375e-01 -1.07997644e+00 7.42972255e-01 -5.80608308e-01 5.21680295e-01 6.70956433e-01 1.36805117e-01 2.50660688e-01 -3.97159576e-01 -1.39771834e-01 6.53106570e-01 -8.92598033e-02 1.19443977e+00 -9.68594253e-01 -1.04274201e+00 -2.78227240e-01 -8.35295767e-02 -1.43549991e+00 2.44409889e-01 -7.93509305e-01 -2.61881966e-02 -1.04315102e+00 3.03750247e-01 -5.82691371e-01 -4.70668916e-03 1.12612414e+00 1.27064854e-01 1.68765441e-01 2.84299761e-01 2.96195328e-01 -7.10977793e-01 2.74627477e-01 1.28022099e+00 2.71016676e-02 -1.80982068e-01 -5.59164770e-03 -5.91079831e-01 1.10902357e+00 8.60900044e-01 -7.27299571e-01 -4.89799529e-01 -5.26976645e-01 4.36549157e-01 5.24379849e-01 4.34988648e-01 -1.30141675e+00 3.81221682e-01 -5.02925813e-01 -9.69741270e-02 -3.74164015e-01 2.09801346e-01 -2.15950638e-01 -1.26949355e-01 9.36806202e-01 -2.36202151e-01 3.70721556e-02 5.44484198e-01 3.52312535e-01 1.14082463e-01 -2.08286270e-01 7.30197668e-01 -5.33938408e-01 -4.39429581e-01 7.01696277e-02 -1.11450088e+00 3.16993147e-01 9.11601365e-01 -2.76336432e-01 -4.65342075e-01 -4.83412355e-01 -7.36154139e-01 4.31131780e-01 6.15246356e-01 -1.53416857e-01 -1.97216600e-01 -6.42213464e-01 -3.38922799e-01 -1.14567749e-01 -5.19023716e-01 9.07691866e-02 1.16804622e-01 6.83345854e-01 -9.30728257e-01 6.43222332e-01 -3.07110369e-01 -5.16058207e-01 -1.35382128e+00 1.87627420e-01 4.72450912e-01 -9.61547852e-01 -7.31214523e-01 8.53239119e-01 3.53652611e-02 -6.77071929e-01 1.32060990e-01 -4.91035640e-01 6.38443410e-01 -4.65903103e-01 2.67602712e-01 8.00331593e-01 -2.28950024e-01 3.88267428e-01 -1.22754090e-01 1.69289351e-01 -1.39364570e-01 -3.75282943e-01 1.35369229e+00 3.82996351e-01 1.76986098e-01 2.14430854e-01 5.39362311e-01 -2.34528393e-01 -1.89540005e+00 -4.19687405e-02 1.47264209e-02 -2.48502851e-01 3.57825041e-01 -9.69039738e-01 -7.97406077e-01 1.07188582e+00 3.30423743e-01 -2.42375493e-01 7.48095155e-01 -1.72003210e-01 9.78654087e-01 1.25754893e+00 1.00753748e+00 -9.09720361e-01 8.29754248e-02 1.08153915e+00 1.56228259e-01 -8.69961441e-01 9.67286248e-03 1.58821017e-01 -3.18565875e-01 9.65979695e-01 7.25410938e-01 -7.03240752e-01 6.18034042e-02 5.68467557e-01 -4.04425412e-01 3.37759056e-03 -1.39009821e+00 -2.21506953e-01 -6.44695461e-01 5.06284535e-01 -1.09271146e-02 -9.09880847e-02 -2.80703336e-01 3.88320684e-01 -5.85937142e-01 5.43735743e-01 5.27499318e-01 1.12001812e+00 -7.57972777e-01 -1.27940440e+00 -2.27816582e-01 3.38988096e-01 -4.91113752e-01 -1.53158695e-01 8.81269481e-03 1.12835205e+00 -6.19927794e-02 6.51119173e-01 3.06707859e-01 -3.03450853e-01 -1.32134661e-01 -1.02181248e-01 5.87252438e-01 -5.45179188e-01 -1.09702289e+00 -2.69748420e-02 4.59569186e-01 -8.59587133e-01 2.33173952e-01 -5.38440287e-01 -1.43436134e+00 -1.31790042e+00 -1.74189508e-01 2.77326465e-01 5.23547530e-01 7.21508503e-01 4.60818738e-01 3.30973536e-01 1.60336271e-01 -1.04390395e+00 -9.87995505e-01 -6.83018863e-01 -2.99780309e-01 -4.01164025e-01 3.90120037e-02 -4.45872337e-01 -3.25396627e-01 -3.86497825e-01]

[3.5497488975524902, 1.435394048690796]

d77e6273-071a-4a37-906f-645cf724f978

spot-evasion-attacks-adversarial-examples-for

1911.00927

https://arxiv.org/abs/1911.00927v2

https://arxiv.org/pdf/1911.00927v2.pdf

Spot Evasion Attacks: Adversarial Examples for License Plate Recognition Systems with Convolutional Neural Networks

Recent studies have shown convolution neural networks (CNNs) for image recognition are vulnerable to evasion attacks with carefully manipulated adversarial examples. Previous work primarily focused on how to generate adversarial examples closed to source images, by introducing pixel-level perturbations into the whole or specific part of images. In this paper, we propose an evasion attack on CNN classifiers in the context of License Plate Recognition (LPR), which adds predetermined perturbations to specific regions of license plate images, simulating some sort of naturally formed spots (such as sludge, etc.). Therefore, the problem is modeled as an optimization process searching for optimal perturbation positions, which is different from previous work that consider pixel values as decision variables. Notice that this is a complex nonlinear optimization problem, and we use a genetic-algorithm based approach to obtain optimal perturbation positions. In experiments, we use the proposed algorithm to generate various adversarial examples in the form of rectangle, circle, ellipse and spots cluster. Experimental results show that these adversarial examples are almost ignored by human eyes, but can fool HyperLPR with high attack success rate over 93%. Therefore, we believe that this kind of spot evasion attacks would pose a great threat to current LPR systems, and needs to be investigated further by the security community.

['Jing-sheng Lei', 'Wu-jie Zhou', 'Jian-hai Chen', 'Jia-min Wang', 'Dan-feng Ma', 'Ya-guan Qian', 'Jun Pan', 'Bin Wang']

2019-10-27

['license-plate-recognition']

['computer-vision']

[ 4.55231190e-01 -4.92627583e-02 4.52633142e-01 6.13845885e-02 -3.14315230e-01 -1.09175777e+00 6.07502997e-01 -6.78884208e-01 -3.87461990e-01 6.87817752e-01 -4.29572701e-01 -5.18589079e-01 3.78324896e-01 -1.04129231e+00 -1.17240691e+00 -8.70689631e-01 1.56556249e-01 -9.65896249e-02 2.69333869e-01 -4.68060911e-01 5.46847224e-01 1.01356506e+00 -1.02675056e+00 1.39938310e-01 6.28137827e-01 6.28029764e-01 -3.66707832e-01 7.60461152e-01 1.11759752e-01 6.52987301e-01 -1.30994689e+00 -7.93543279e-01 8.02438498e-01 -3.03451061e-01 -3.31234694e-01 9.53279808e-02 2.79019207e-01 -2.55687952e-01 -5.28944731e-01 1.75738788e+00 4.36465949e-01 1.11260533e-01 6.59135520e-01 -1.51169872e+00 -1.16594470e+00 5.35328865e-01 -6.40163183e-01 2.39966437e-01 1.31028429e-01 6.20027423e-01 6.21907376e-02 -4.80605900e-01 3.22428107e-01 1.21485662e+00 6.98251069e-01 8.90410304e-01 -8.42145026e-01 -1.05911589e+00 -1.18998729e-01 2.16218103e-02 -1.64969957e+00 -1.13775328e-01 9.49116230e-01 -1.53478742e-01 4.61885065e-01 6.24443293e-01 1.64003819e-01 1.32787192e+00 3.78062248e-01 4.76597250e-01 1.23399103e+00 -4.04380918e-01 2.12589413e-01 4.06545758e-01 -1.93381533e-01 3.31869811e-01 4.11722511e-01 3.66621345e-01 2.21192360e-01 -2.03650653e-01 9.63805616e-01 3.96095170e-03 -4.99938488e-01 2.77810186e-01 -8.42242181e-01 8.92134607e-01 7.28803635e-01 9.11502093e-02 -1.36041164e-01 1.09238006e-01 5.88849261e-02 1.80961072e-01 -4.47085090e-02 7.40590632e-01 -4.62408438e-02 3.61144602e-01 -3.82254213e-01 2.98220605e-01 6.06629014e-01 8.70670438e-01 5.18687904e-01 5.76157212e-01 -9.26672146e-02 5.27221739e-01 3.06796700e-01 8.06434751e-01 4.84503359e-01 -4.18378502e-01 5.46894908e-01 4.22506720e-01 1.77600607e-01 -1.52753639e+00 8.79369327e-04 -3.13717544e-01 -8.45900655e-01 8.03674698e-01 4.74685252e-01 -6.43984020e-01 -1.11581874e+00 1.31998074e+00 3.78919393e-02 5.64923823e-01 4.58314449e-01 1.11087692e+00 7.41876423e-01 8.56044352e-01 -7.94192404e-02 1.77095965e-01 1.27921963e+00 -7.20980525e-01 -5.26159704e-01 -1.98356226e-01 2.32034534e-01 -8.06724906e-01 9.48148310e-01 3.56468618e-01 -6.69204235e-01 -3.20517153e-01 -1.33085442e+00 7.12531090e-01 -7.70954490e-01 1.03724105e-02 7.29769021e-02 1.16863585e+00 -6.82735860e-01 3.13792765e-01 -2.35901356e-01 6.96889609e-02 6.48004711e-01 4.34750825e-01 -3.12182337e-01 -1.45009711e-01 -1.50020719e+00 8.39822769e-01 3.48631650e-01 4.40197408e-01 -8.67656171e-01 -5.00056207e-01 -6.76358759e-01 -9.76014733e-02 3.50126892e-01 2.32079439e-02 7.30995953e-01 -1.29536736e+00 -1.42565751e+00 6.97996259e-01 5.07190228e-01 -6.15469038e-01 7.20623493e-01 4.53107096e-02 -6.94129169e-01 -2.54174341e-02 -5.59924603e-01 3.90114218e-01 1.08631492e+00 -1.65448678e+00 -1.32680789e-01 -2.42959753e-01 4.52918857e-01 6.15056194e-02 -2.27634728e-01 4.69821036e-01 -2.44548917e-01 -1.01025462e+00 -2.89500743e-01 -1.12285066e+00 -4.90337282e-01 -1.74141392e-01 -8.26777756e-01 3.07657570e-01 1.09627032e+00 -4.98149335e-01 7.86113203e-01 -2.28689957e+00 -5.54233551e-01 4.52539295e-01 8.56520515e-03 8.26841772e-01 -2.39789076e-02 7.82797635e-02 -2.83972502e-01 6.99680269e-01 -3.41692656e-01 3.07011694e-01 -1.21189587e-01 -8.40491951e-02 -7.81137168e-01 7.59236515e-01 3.81361574e-01 9.66978550e-01 -4.60909069e-01 -6.08381741e-02 1.37119204e-01 4.85526204e-01 -2.25919798e-01 9.53682885e-02 -4.67049368e-02 3.35312709e-02 -6.53829694e-01 8.19068313e-01 1.20320332e+00 3.45250607e-01 -4.89133745e-01 -1.52663961e-01 1.19050741e-01 -7.23909259e-01 -1.11629140e+00 3.75903904e-01 8.95690247e-02 8.40422511e-01 -1.48269311e-01 -8.85146260e-01 1.29386175e+00 1.55324712e-01 -1.94743097e-01 -2.83446819e-01 5.44440806e-01 -3.51034403e-02 1.76998794e-01 -3.66432339e-01 3.53835523e-01 -7.54242912e-02 -1.60173133e-01 2.60355324e-01 -6.97818339e-01 -2.78925411e-02 -3.26294065e-01 -1.62744537e-01 1.06984019e+00 -2.93730110e-01 -1.03999242e-01 3.76336314e-02 7.14822829e-01 2.15983048e-01 4.13329452e-01 9.81072664e-01 -2.68869251e-01 7.92715669e-01 5.49449503e-01 -5.50438762e-01 -1.15497947e+00 -8.65937650e-01 5.57569265e-02 1.18239947e-01 5.57532191e-01 4.13579971e-01 -1.03365457e+00 -8.35299373e-01 -1.32917359e-01 7.39553750e-01 -6.45983517e-01 -4.88111675e-01 -7.49484360e-01 -9.82806504e-01 1.32261074e+00 3.18016201e-01 1.02083111e+00 -1.40562522e+00 -3.70715618e-01 -4.67312224e-02 3.60147625e-01 -1.19806588e+00 -3.76066864e-01 -2.63225704e-01 -3.38256419e-01 -1.12051296e+00 -9.27827120e-01 -8.43396723e-01 1.08904040e+00 2.07153052e-01 5.77296436e-01 2.27658808e-01 -4.32130247e-01 3.51949409e-02 -5.32740772e-01 -7.18571365e-01 -7.74915993e-01 -5.66693485e-01 5.62934652e-02 3.38332444e-01 3.31151873e-01 -2.44838953e-01 -4.05848026e-01 7.28157103e-01 -1.34599578e+00 -5.16881943e-01 6.57715440e-01 4.79264617e-01 2.03180611e-01 6.55294299e-01 4.19243366e-01 -8.72107863e-01 1.02337742e+00 -2.93301404e-01 -8.44978571e-01 3.23828459e-01 5.91230914e-02 -3.45417261e-01 1.16123116e+00 -9.75750506e-01 -8.52388918e-01 7.46331587e-02 3.55918135e-04 -9.98231173e-01 -5.84198058e-01 1.54753000e-01 -5.40064573e-01 -8.10336709e-01 9.90890265e-01 5.99058867e-01 5.91612756e-02 3.59316245e-02 1.62922353e-01 6.91844046e-01 5.16039550e-01 -3.55118960e-01 1.62403238e+00 3.55885088e-01 -1.27248727e-02 -8.65610421e-01 -5.62166497e-02 2.65352279e-01 1.26451449e-02 -4.33178425e-01 7.22059369e-01 -4.58143860e-01 -8.13213766e-01 1.03685701e+00 -1.15536010e+00 -2.06719711e-01 1.83731288e-01 2.10150063e-01 -1.94939151e-01 4.23135966e-01 -4.27870154e-01 -7.75267780e-01 -6.05630614e-02 -1.33975601e+00 5.27494133e-01 5.59023440e-01 3.87843668e-01 -6.83985472e-01 -3.09774160e-01 3.17469597e-01 5.72446823e-01 7.77181923e-01 5.38234293e-01 -9.87956822e-01 -8.23105335e-01 -7.35840321e-01 -1.73880428e-01 5.01271188e-01 -6.55842200e-02 3.77873719e-01 -7.49965847e-01 -2.24367991e-01 1.59488648e-01 -2.07268178e-01 4.60069090e-01 9.30733159e-02 1.26042020e+00 -8.19255412e-01 -2.57263333e-01 7.41145313e-01 1.55221176e+00 7.75082052e-01 1.37351251e+00 6.44962490e-01 6.55852854e-01 2.30633959e-01 5.55900633e-01 1.84412330e-01 -4.94994640e-01 5.15541911e-01 7.80009151e-01 -2.98965722e-01 2.53292531e-01 6.32982180e-02 4.56491500e-01 -2.60777861e-01 -4.22792733e-02 -6.96384072e-01 -9.00189817e-01 1.52722076e-02 -1.36341250e+00 -1.12019598e+00 -6.78531975e-02 1.88681650e+00 5.44569194e-01 3.56527895e-01 -1.35078460e-01 8.99860486e-02 1.35699558e+00 1.23422094e-01 -5.70808709e-01 -4.99663711e-01 -2.82509804e-01 -2.75588874e-02 1.09320986e+00 1.91703975e-01 -1.23076057e+00 1.11830771e+00 6.09838963e+00 9.83574033e-01 -1.46824694e+00 -4.46205854e-01 8.62783968e-01 2.82702059e-01 -8.59324932e-02 -2.85742998e-01 -8.06204915e-01 6.16329908e-01 3.93089205e-01 -2.14868575e-01 5.36743581e-01 8.07511926e-01 1.10101670e-01 4.32133257e-01 -4.94539350e-01 7.55968988e-01 3.11383665e-01 -1.30574548e+00 2.92052388e-01 1.66303739e-01 6.69303060e-01 -3.57533842e-01 5.21644354e-01 2.32782349e-01 5.94369888e-01 -1.41577780e+00 4.00595218e-01 4.42122519e-01 5.19953966e-01 -1.09028435e+00 9.42190468e-01 2.72038788e-01 -6.36409938e-01 -5.72761819e-02 -6.77432239e-01 2.61535525e-01 -2.11478204e-01 5.26182167e-02 -8.57493103e-01 1.84213579e-01 5.37428498e-01 3.12826596e-02 -6.27366304e-01 1.03114200e+00 -2.72105575e-01 6.92455351e-01 -2.04708830e-01 -4.18619424e-01 4.83096033e-01 -7.55426139e-02 8.04190159e-01 9.97796893e-01 3.12926471e-01 1.94123954e-01 -2.91699290e-01 1.22741902e+00 -2.58949250e-01 -2.40309760e-02 -1.03532171e+00 1.10589318e-01 5.00378788e-01 1.01796091e+00 -8.76443148e-01 1.46576075e-03 -1.41691685e-01 8.23043466e-01 -2.14413837e-01 5.74463129e-01 -1.40514719e+00 -7.52714634e-01 5.89164197e-01 3.23403962e-02 3.38167220e-01 -1.80363227e-02 -1.74011603e-01 -9.35375392e-01 1.18806139e-02 -1.17926264e+00 5.22326455e-02 -8.83372903e-01 -1.20602691e+00 6.78041458e-01 -2.44973481e-01 -1.54363942e+00 2.89845914e-01 -8.88462842e-01 -1.25515318e+00 8.61832201e-01 -1.26989031e+00 -8.54829550e-01 -1.59719050e-01 8.48333955e-01 4.61021125e-01 -6.11244202e-01 4.45934951e-01 -1.15347557e-01 -7.93681622e-01 1.04386854e+00 9.00264308e-02 8.88203204e-01 3.45378041e-01 -7.47074902e-01 5.05162358e-01 1.25936115e+00 -8.78421217e-02 5.87160408e-01 8.02238226e-01 -6.15618765e-01 -1.33367074e+00 -1.50512671e+00 -4.80860565e-03 -3.39085728e-01 4.53636259e-01 -2.79949367e-01 -8.90854001e-01 5.54406404e-01 1.89312652e-01 2.18850777e-01 4.71394509e-01 -7.36355424e-01 -2.43087053e-01 5.84893264e-02 -1.67581642e+00 1.17819166e+00 4.64384168e-01 -5.86205125e-02 -4.82873589e-01 3.32836509e-01 8.44187021e-01 -3.75181288e-01 -5.26234925e-01 4.04687256e-01 1.04988642e-01 -6.77055418e-01 1.18427646e+00 -6.02580726e-01 5.59535205e-01 -7.00076520e-01 -1.12629030e-02 -1.24479806e+00 -1.30832950e-02 -5.97150683e-01 3.48800600e-01 1.20870113e+00 5.77729583e-01 -8.41375351e-01 8.48647594e-01 6.07492685e-01 -3.82753983e-02 -4.70156580e-01 -7.74038911e-01 -8.97808492e-01 1.74152046e-01 -2.54554331e-01 7.02213109e-01 1.04303408e+00 -5.04599869e-01 -3.37325096e-01 -5.26858151e-01 1.02665639e+00 5.89458942e-01 -3.77306491e-01 9.75895822e-01 -4.94373798e-01 -1.77672923e-01 -5.27968049e-01 -9.24036443e-01 -4.08570528e-01 7.70928264e-02 -4.54777360e-01 1.90844521e-01 -7.65088558e-01 -3.84312749e-01 -3.27738672e-01 -1.66061759e-01 4.14786905e-01 -2.46288911e-01 7.10338235e-01 3.73872429e-01 1.42306060e-01 -1.30458772e-01 2.78756291e-01 1.21489024e+00 -4.71209437e-01 2.42181569e-02 2.94620037e-01 -7.72764027e-01 7.75040507e-01 1.12680852e+00 -6.00591421e-01 -2.16527730e-01 -1.28034502e-01 5.69854453e-02 -1.27283886e-01 5.84546268e-01 -1.17290151e+00 2.02804446e-01 -4.27516997e-01 5.58513820e-01 -2.42529646e-01 8.08665752e-02 -1.08192372e+00 2.24022523e-01 5.96962333e-01 -8.75268951e-02 -7.66332746e-02 4.83149618e-01 4.31055069e-01 -1.75006986e-01 -6.44288063e-01 1.02379596e+00 -3.14242840e-01 -7.15493977e-01 2.09738195e-01 -5.53796768e-01 -1.75243855e-01 1.61840379e+00 -5.65246940e-01 -5.88737905e-01 -2.97997981e-01 -4.97526556e-01 -1.11559510e-01 5.03120959e-01 4.96905625e-01 8.62004101e-01 -1.31183612e+00 -8.03909600e-01 3.08713138e-01 -1.32835656e-01 -4.64484058e-02 1.47987872e-01 7.57015049e-02 -1.08284354e+00 2.32912861e-02 -4.39921111e-01 -2.78858662e-01 -1.35921276e+00 8.49418938e-01 6.11397624e-01 6.04337975e-02 -4.76292372e-01 7.70369828e-01 2.21112609e-01 -3.07175577e-01 1.37655303e-01 1.72997788e-01 -4.18284774e-01 -4.74174380e-01 5.38987815e-01 1.17861032e-01 -2.37107873e-01 -5.62001526e-01 -2.37027720e-01 6.49588585e-01 -1.20339237e-01 1.82732105e-01 9.71458912e-01 4.40970898e-01 8.70394930e-02 -3.89734238e-01 1.09557819e+00 1.53019369e-01 -1.20775735e+00 1.27071608e-02 -4.47486252e-01 -7.64454365e-01 -3.36490691e-01 -7.73965478e-01 -1.29913592e+00 4.74149227e-01 5.21992564e-01 4.03744698e-01 1.07405281e+00 -3.84504825e-01 5.57335496e-01 5.92914164e-01 1.90416887e-01 -6.30938470e-01 1.88460067e-01 3.67048740e-01 1.11271548e+00 -1.01902413e+00 -3.46787959e-01 -3.78060877e-01 -7.19493628e-01 1.18800306e+00 1.04540944e+00 -7.50333607e-01 4.46782619e-01 6.02057576e-01 3.52341294e-01 7.81580582e-02 -1.75781369e-01 4.07789558e-01 6.28065839e-02 7.82620847e-01 -4.16047037e-01 -1.06422499e-01 -3.78065929e-02 5.79889476e-01 -3.25189978e-01 -3.68160546e-01 1.00040662e+00 9.24584329e-01 -4.89130408e-01 -7.06711173e-01 -1.21331322e+00 1.46868691e-01 -6.70164526e-01 -9.77086127e-02 -5.88024378e-01 1.07209432e+00 1.97650880e-01 7.99995184e-01 -2.74996576e-03 -5.59119225e-01 4.84907478e-01 -2.48596743e-01 1.01028001e-02 -1.34778708e-01 -5.80189526e-01 -4.04789567e-01 -4.26618695e-01 -1.01833977e-01 4.23406549e-02 -3.36377650e-01 -9.32791591e-01 -4.88491178e-01 -4.87956703e-01 -6.18279390e-02 6.54305458e-01 7.35980928e-01 -3.14257629e-02 5.11358261e-01 1.14166880e+00 -6.87700272e-01 -7.80048549e-01 -7.82460153e-01 -5.02913594e-01 4.50351864e-01 1.63925663e-01 -3.81468117e-01 -8.73920262e-01 9.44805667e-02]

[5.49254035949707, 7.952006816864014]

66ef6e0d-13bc-4e64-87e3-1cbb72e4f68d

deep-stereo-video-inpainting

http://openaccess.thecvf.com//content/CVPR2023/html/Wu_Deep_Stereo_Video_Inpainting_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Wu_Deep_Stereo_Video_Inpainting_CVPR_2023_paper.pdf

Deep Stereo Video Inpainting

Stereo video inpainting aims to fill the missing regions on the left and right views of the stereo video with plausible content simultaneously. Compared with the single video inpainting that has achieved promising results using deep convolutional neural networks, inpainting the missing regions of stereo video has not been thoroughly explored. In essence, apart from the spatial and temporal consistency that single video inpainting needs to achieve, another key challenge for stereo video inpainting is to maintain the stereo consistency between left and right views and hence alleviate the 3D fatigue for viewers. In this paper, we propose a novel deep stereo video inpainting network named SVINet, which is the first attempt for stereo video inpainting task utilizing deep convolutional neural networks. SVINet first utilizes a self-supervised flow-guided deformable temporal alignment module to align the features on the left and right view branches, respectively. Then, the aligned features are fed into a shared adaptive feature aggregation module to generate missing contents of their respective branches. Finally, the parallax attention module (PAM) that uses the cross-view information to consider the significant stereo correlation is introduced to fuse the completed features of left and right views. Furthermore, we develop a stereo consistency loss to regularize the trained parameters, so that our model is able to yield high-quality stereo video inpainting results with better stereo consistency. Experimental results demonstrate that our SVINet outperforms state-of-the-art single video inpainting models.

['Yan Yan', 'Hanyu Xuan', 'Changchang Sun', 'Zhiliang Wu']

2023-01-01

cvpr-2023-1

['video-inpainting']

['computer-vision']

[ 2.96367332e-02 -1.00025080e-01 -5.17829955e-02 -2.04738736e-01 -5.14792323e-01 -2.15889201e-01 2.22960919e-01 -5.02779067e-01 2.81051230e-02 7.68117249e-01 4.50750709e-01 1.04186058e-01 1.81481451e-01 -5.64616978e-01 -9.64653611e-01 -5.59100032e-01 4.71147329e-01 -1.97250377e-02 2.64722496e-01 -6.73617572e-02 1.95845380e-01 4.10958439e-01 -1.42758858e+00 5.00935018e-01 1.11569059e+00 1.09746432e+00 5.95865250e-01 4.63228881e-01 -2.23802906e-02 1.00823355e+00 -1.63865685e-01 -2.03411192e-01 5.98945260e-01 -5.84956050e-01 -6.25953853e-01 4.11364257e-01 1.05220246e+00 -1.02229452e+00 -7.79044390e-01 9.18149173e-01 3.68604720e-01 1.53993502e-01 1.79460183e-01 -1.07834959e+00 -3.89682591e-01 2.02338934e-01 -8.62277925e-01 1.90876767e-01 5.51537335e-01 4.09448922e-01 7.80038178e-01 -8.43209267e-01 9.12813246e-01 1.29626667e+00 4.91343886e-01 4.76468801e-01 -1.22541595e+00 -6.39897287e-01 2.06431761e-01 3.18128318e-01 -1.02261925e+00 -4.19060349e-01 1.19537485e+00 -3.97337407e-01 6.78265691e-01 1.04539216e-01 9.28748310e-01 9.76703227e-01 4.33285773e-01 7.35562027e-01 8.30138803e-01 -1.68020308e-01 -5.04924841e-02 -2.95305878e-01 -4.80729371e-01 5.98765194e-01 -1.67422235e-01 2.71344304e-01 -7.28908598e-01 2.79876679e-01 1.26981759e+00 3.20251197e-01 -6.95100844e-01 -4.44419116e-01 -1.23696625e+00 5.62313437e-01 3.32195878e-01 2.15127498e-01 -6.01331949e-01 3.11200395e-02 3.88188332e-01 2.79462576e-01 6.15142286e-01 2.34956026e-01 -3.54107618e-01 1.16470950e-02 -1.36306226e+00 3.47468287e-01 2.66035110e-01 9.35685992e-01 8.78654540e-01 1.73056558e-01 -2.69016325e-01 8.68438542e-01 2.42988080e-01 1.70128480e-01 2.96759009e-01 -1.60748518e+00 8.66352677e-01 5.34128249e-01 1.84457719e-01 -1.20674539e+00 1.40364021e-02 -1.73117742e-01 -9.94487822e-01 3.53205383e-01 1.27592131e-01 4.37280014e-02 -6.26264572e-01 1.62664592e+00 3.05046022e-01 3.62411350e-01 -1.21856801e-01 1.05256152e+00 7.73462296e-01 8.40136766e-01 -3.63397270e-01 -4.78291988e-01 1.01681507e+00 -1.36472750e+00 -8.33491027e-01 -2.77390510e-01 1.22741900e-01 -1.06334567e+00 7.35465348e-01 2.34968737e-01 -1.55427337e+00 -8.59246194e-01 -9.39311683e-01 -5.99153340e-01 4.16918099e-01 -2.32201010e-01 2.93252438e-01 -1.12843901e-01 -1.05981076e+00 7.17119455e-01 -8.12543869e-01 1.47623904e-02 3.97223324e-01 6.58091679e-02 -6.27881944e-01 -3.13115448e-01 -1.06439078e+00 5.99475086e-01 2.03298375e-01 1.35707155e-01 -8.91196847e-01 -1.07252562e+00 -1.12553191e+00 1.95965827e-01 2.81916916e-01 -1.02609611e+00 9.50928152e-01 -1.30574143e+00 -1.41049516e+00 6.89969957e-01 -3.66333723e-01 -2.67521590e-01 8.97577286e-01 -4.29339975e-01 -1.04201250e-01 4.14671868e-01 4.05974060e-01 1.14944899e+00 1.11569917e+00 -1.31024063e+00 -7.35441208e-01 -8.25263709e-02 -5.44588976e-02 5.40863931e-01 -1.10720836e-01 -1.97966605e-01 -8.41332972e-01 -1.01946652e+00 2.43543848e-01 -5.51578522e-01 -2.98785176e-02 3.89461398e-01 -3.17466557e-01 2.78710872e-01 1.21962547e+00 -1.10349536e+00 1.14745653e+00 -2.25337052e+00 5.56127071e-01 -2.00822055e-01 3.73676538e-01 1.80260926e-01 -2.24733129e-01 3.18742007e-01 -3.23608339e-01 -4.70682770e-01 -1.58964679e-01 -6.05041623e-01 -5.96615255e-01 3.17575485e-01 -4.79769796e-01 2.74312675e-01 1.95415989e-01 6.80936694e-01 -9.02081966e-01 -5.08082390e-01 6.68088078e-01 6.61101043e-01 -9.70962346e-01 6.54081941e-01 -1.66518316e-01 9.24324393e-01 -2.10473791e-01 4.60296720e-01 1.15271187e+00 -1.65489279e-02 -3.70960188e-05 -5.67340910e-01 -3.22092175e-01 2.16894507e-01 -1.05202067e+00 2.04689765e+00 -5.09988904e-01 7.37815738e-01 2.70970434e-01 -6.94198668e-01 7.28554130e-01 1.76256984e-01 8.61467421e-01 -9.09178674e-01 -4.29214037e-04 1.96942836e-01 -3.91199797e-01 -5.91770947e-01 5.27402937e-01 -4.60713245e-02 4.70591575e-01 1.80967629e-01 -1.71814952e-02 -1.93914741e-01 1.70532867e-01 2.13604569e-01 6.80221379e-01 4.61858124e-01 6.06928952e-03 7.73431882e-02 7.19827890e-01 -4.65438277e-01 8.45502198e-01 1.51688740e-01 -8.98964778e-02 1.38317120e+00 4.62399274e-01 -7.32937396e-01 -1.35227120e+00 -9.57692921e-01 1.75014377e-01 4.63419944e-01 5.26289821e-01 -3.97557765e-01 -7.66217887e-01 -4.54018474e-01 -2.07577854e-01 4.73098099e-01 -4.48804051e-01 -6.86109960e-02 -8.32616925e-01 1.54477894e-01 -1.65763423e-01 4.89417940e-01 8.64433646e-01 -1.13900006e+00 -5.67157447e-01 3.30614150e-01 -7.46093631e-01 -1.18844569e+00 -1.04944348e+00 -1.86805621e-01 -1.05500448e+00 -1.19878149e+00 -1.00707972e+00 -9.14519370e-01 6.42088294e-01 7.63671637e-01 9.53709722e-01 3.80510129e-02 -1.09211402e-02 6.30609039e-03 -2.03056157e-01 3.05579245e-01 -4.09724563e-01 -2.20542058e-01 -1.99510962e-01 1.88692629e-01 -2.85060495e-01 -1.02497947e+00 -9.90308166e-01 4.04567003e-01 -1.20535100e+00 7.84849524e-01 4.03297901e-01 1.10264087e+00 5.82643509e-01 -1.31615371e-01 -8.05238038e-02 -4.92051303e-01 1.09308109e-01 -2.67605484e-01 -4.65516984e-01 2.32814681e-02 -1.26153186e-01 -3.09467036e-02 8.69799376e-01 -2.17255354e-01 -1.18502378e+00 -1.77551266e-02 -2.32673466e-01 -1.11004710e+00 1.00937225e-01 1.57175556e-01 -2.86449254e-01 1.29298540e-02 1.09657876e-01 4.66521919e-01 2.42925763e-01 -4.24841791e-01 1.23321943e-01 2.13905126e-01 5.87452054e-01 -3.18524390e-01 6.95912361e-01 6.33062720e-01 -6.40503392e-02 -5.94388723e-01 -9.02811229e-01 -3.14121962e-01 -6.68191314e-01 -3.58360171e-01 9.86803114e-01 -1.15616572e+00 -4.03002799e-01 6.37317359e-01 -1.40387499e+00 -1.75395966e-01 -3.19836408e-01 3.96180719e-01 -9.44774210e-01 9.19707835e-01 -8.30398202e-01 -2.50861377e-01 -3.34046394e-01 -1.35159743e+00 1.28119445e+00 2.89042026e-01 -1.89164460e-01 -7.54441977e-01 6.11790046e-02 6.48121834e-01 1.69949338e-01 1.39441147e-01 6.65374160e-01 2.29001939e-01 -9.46610391e-01 3.28206688e-01 -3.10075372e-01 6.43488228e-01 3.16699713e-01 -1.52131065e-03 -6.49462879e-01 -4.32032853e-01 3.35880131e-01 -2.14697361e-01 9.25141633e-01 6.77014649e-01 1.10396481e+00 -3.14198345e-01 -1.14900365e-01 1.13604987e+00 1.25182617e+00 2.09487081e-01 1.01820266e+00 4.13939267e-01 9.73363280e-01 5.96961617e-01 7.24158883e-01 4.23231512e-01 2.95286864e-01 8.65423620e-01 5.95505118e-01 -2.96027184e-01 -4.64219868e-01 -5.15587389e-01 4.41918582e-01 7.57021368e-01 -1.14102662e-01 -8.99247155e-02 -2.42217794e-01 3.96965861e-01 -2.00026131e+00 -1.29911900e+00 8.54798928e-02 2.11327267e+00 8.28305423e-01 -6.35705665e-02 -1.98569179e-01 4.78469878e-02 7.93799102e-01 5.19008458e-01 -5.33513725e-01 -2.02807277e-01 -1.58090234e-01 -2.77757775e-02 4.96746376e-02 6.70647740e-01 -8.89923811e-01 7.82606125e-01 5.13581705e+00 7.96738923e-01 -1.20510566e+00 -5.28402738e-02 8.45654070e-01 -2.26836339e-01 -3.79465103e-01 1.99539140e-01 -4.35960561e-01 8.54387581e-01 2.27742881e-01 2.20079526e-01 5.50165892e-01 5.51059604e-01 6.95255160e-01 -1.87013373e-01 -1.11771739e+00 1.14858115e+00 7.91008472e-02 -1.57822955e+00 3.13974857e-01 -2.23784111e-02 1.11809075e+00 -2.73777992e-01 5.11205159e-02 -1.31883755e-01 -2.74076402e-01 -6.94476724e-01 9.13885415e-01 6.80266082e-01 8.19394529e-01 -8.29473078e-01 5.26507318e-01 1.73101604e-01 -1.10345495e+00 -1.97216809e-01 -2.98133969e-01 6.05604388e-02 5.76183736e-01 5.71371913e-01 -1.60252035e-01 7.96916962e-01 8.00135970e-01 1.16924262e+00 -2.05515832e-01 9.65661168e-01 -1.23976395e-01 5.87432757e-02 8.31558555e-03 9.84002054e-01 2.88532466e-01 -4.75789607e-01 6.67935252e-01 6.64643943e-01 5.41291535e-01 -1.03852376e-01 4.03555296e-02 8.17159712e-01 -7.59516209e-02 -7.71881565e-02 -6.40736759e-01 3.96966875e-01 2.46291324e-01 1.16222680e+00 -2.42956027e-01 -3.56319249e-01 -6.09320879e-01 1.37256789e+00 2.46989727e-01 4.08580869e-01 -8.97621393e-01 2.60324813e-02 7.60703444e-01 4.46191281e-01 5.34043252e-01 -2.94268616e-02 -1.62637413e-01 -1.52084529e+00 3.31892133e-01 -8.77848268e-01 1.92101166e-01 -1.13288379e+00 -1.11034584e+00 4.76243228e-01 -2.91618526e-01 -1.83376229e+00 -3.33512872e-01 -9.15116668e-02 -8.29939544e-01 9.22128022e-01 -1.60812593e+00 -1.20760834e+00 -5.92546463e-01 7.19009578e-01 1.03817809e+00 -1.14502810e-01 2.67393708e-01 5.01027465e-01 -4.47887242e-01 3.74202043e-01 8.26178491e-02 -8.67733881e-02 9.69259620e-01 -6.81197107e-01 1.79736689e-01 1.04977083e+00 -3.97160619e-01 3.81514162e-01 7.12737679e-01 -7.25899160e-01 -1.16447377e+00 -1.24092388e+00 8.29024076e-01 1.17056496e-01 1.45660371e-01 9.36685950e-02 -9.19248044e-01 6.97926939e-01 5.64191580e-01 2.26619169e-01 9.09970999e-02 -6.48623466e-01 -3.45382839e-01 -4.31496918e-01 -9.82607007e-01 6.83233321e-01 1.10544026e+00 -3.53302449e-01 -3.65711123e-01 1.00122295e-01 7.37065554e-01 -6.17645085e-01 -5.71887910e-01 3.40411723e-01 5.32135308e-01 -1.66783917e+00 1.04472780e+00 -2.55618691e-01 1.25158596e+00 -5.03342807e-01 -1.01768551e-02 -1.24934936e+00 -3.80093485e-01 -8.14812064e-01 -1.06676705e-01 1.27548683e+00 -2.38053918e-01 -1.98973730e-01 7.95278907e-01 4.84282881e-01 -4.95198935e-01 -7.34232545e-01 -9.82527614e-01 -3.89759839e-01 -2.36957461e-01 -1.23663530e-01 2.44581729e-01 9.22616720e-01 -2.95536041e-01 1.45628631e-01 -8.25243831e-01 -9.84464884e-02 5.92775345e-01 4.42965806e-01 8.08150053e-01 -8.27059567e-01 -3.75406712e-01 -3.49706173e-01 -2.16825664e-01 -1.47153461e+00 1.65340647e-01 -5.14780939e-01 -1.86564356e-01 -1.47818601e+00 3.72233063e-01 -8.64482448e-02 1.11102931e-01 1.69806316e-01 -2.09064171e-01 2.76356101e-01 3.18217367e-01 2.76985586e-01 -3.32556486e-01 9.63800848e-01 1.84104002e+00 -1.34377167e-01 -3.22064847e-01 -2.49435022e-01 -5.56603551e-01 6.54313862e-01 4.48140979e-01 -1.72869295e-01 -4.51465607e-01 -6.33925498e-01 -7.34656537e-03 7.27949560e-01 5.18490911e-01 -1.03195441e+00 1.79501250e-01 -3.40636581e-01 6.93946481e-01 -8.62812281e-01 5.29284716e-01 -9.69175935e-01 3.97614866e-01 3.49783063e-01 -1.58968911e-01 2.69059211e-01 2.33580489e-02 4.41300064e-01 -6.62173867e-01 5.93201518e-02 1.06764293e+00 -1.52658135e-01 -5.60326338e-01 6.06982768e-01 -1.56163454e-01 1.76074021e-02 1.05627549e+00 -4.68775302e-01 2.50721797e-02 -5.59588432e-01 -4.63441163e-01 3.65449995e-01 8.28892112e-01 5.13468087e-01 9.82405245e-01 -1.43561316e+00 -6.29120171e-01 5.60552061e-01 -1.96710408e-01 3.36971462e-01 9.59092438e-01 8.73916447e-01 -8.99042428e-01 2.11835787e-01 -7.68488705e-01 -7.43867338e-01 -1.22241056e+00 5.79241991e-01 3.01547825e-01 -3.45943779e-01 -8.67546558e-01 5.23407280e-01 7.32100368e-01 4.17603850e-02 2.11330518e-01 -2.19189107e-01 -1.01036109e-01 -6.67397901e-02 5.66617548e-01 2.85508811e-01 -1.49935573e-01 -5.91073573e-01 -3.18052918e-02 7.11849332e-01 -2.63914675e-01 -2.01381035e-02 1.42636204e+00 -4.39083457e-01 -2.98945040e-01 -1.98091324e-02 1.23574090e+00 1.31678820e-01 -2.08899260e+00 -2.87128855e-02 -8.80486250e-01 -8.99282694e-01 -6.00626580e-02 -2.70120680e-01 -1.54448545e+00 9.48436916e-01 2.83419549e-01 -2.54464835e-01 1.39375520e+00 -3.88248831e-01 1.27559328e+00 -2.95938134e-01 1.15106180e-01 -9.61407900e-01 2.46506497e-01 4.66759771e-01 1.14088094e+00 -1.10883558e+00 -2.05918960e-02 -6.75107121e-01 -6.17264986e-01 1.39573276e+00 1.02988625e+00 -2.34476656e-01 3.00119877e-01 1.08000740e-01 -1.10453449e-01 1.80673629e-01 -7.52365589e-01 3.22446883e-01 2.43368149e-01 5.75715244e-01 3.63288283e-01 -5.24696887e-01 -3.11029494e-01 1.02642380e-01 5.56857251e-02 1.26578346e-01 4.69334275e-01 6.18122876e-01 -1.91365480e-01 -1.11225963e+00 -3.20083112e-01 7.71994069e-02 -2.23315716e-01 -1.96499750e-01 -4.27389704e-02 5.64271331e-01 2.21644416e-01 7.04898894e-01 2.21952707e-01 -3.14966470e-01 2.15508685e-01 -3.64254951e-01 5.13689101e-01 -3.83998781e-01 -4.01081681e-01 3.98566216e-01 -1.54315948e-01 -1.00908172e+00 -5.32287598e-01 -4.29608703e-01 -9.76879299e-01 -4.22141582e-01 1.71697944e-01 -1.95222974e-01 -1.95493177e-02 8.03086519e-01 5.44663012e-01 4.69005823e-01 8.98684084e-01 -1.35922372e+00 -1.33665204e-01 -6.93728626e-01 -4.90015656e-01 6.78431869e-01 4.69423950e-01 -6.28685236e-01 -2.63885438e-01 2.05796137e-01]

[10.826117515563965, -1.4238476753234863]

3313b5a5-bee8-4a28-90ca-7b6effb1b855

towards-feature-selection-for-ranking-and

2205.04346

https://arxiv.org/abs/2205.04346v1

https://arxiv.org/pdf/2205.04346v1.pdf

Towards Feature Selection for Ranking and Classification Exploiting Quantum Annealers

Feature selection is a common step in many ranking, classification, or prediction tasks and serves many purposes. By removing redundant or noisy features, the accuracy of ranking or classification can be improved and the computational cost of the subsequent learning steps can be reduced. However, feature selection can be itself a computationally expensive process. While for decades confined to theoretical algorithmic papers, quantum computing is now becoming a viable tool to tackle realistic problems, in particular special-purpose solvers based on the Quantum Annealing paradigm. This paper aims to explore the feasibility of using currently available quantum computing architectures to solve some quadratic feature selection algorithms for both ranking and classification. The experimental analysis includes 15 state-of-the-art datasets. The effectiveness obtained with quantum computing hardware is comparable to that of classical solvers, indicating that quantum computers are now reliable enough to tackle interesting problems. In terms of scalability, current generation quantum computers are able to provide a limited speedup over certain classical algorithms and hybrid quantum-classical strategies show lower computational cost for problems of more than a thousand features.

['Paolo Cremonesi', 'Guglielmo Faggioli', 'Nicola Ferro', 'Riccardo Nembrini', 'Fabio Moroni', 'Maurizio Ferrari Dacrema']

2022-05-09

['classification']

['methodology']

[ 3.36665630e-01 -2.18809605e-01 -1.12633921e-01 -4.13506150e-01 -1.05199349e+00 -2.64937550e-01 6.07441008e-01 5.95263004e-01 -6.56576037e-01 9.67743337e-01 -5.04919171e-01 8.12800881e-03 -4.18163091e-01 -1.17621648e+00 -1.63217410e-01 -1.00883079e+00 -2.60804713e-01 9.71894324e-01 2.55356729e-01 -5.94798684e-01 5.49269915e-01 5.90682387e-01 -2.30225420e+00 2.43841708e-01 6.93803072e-01 9.86581922e-01 -8.62443522e-02 4.44152772e-01 3.02655250e-02 -9.39013530e-03 -2.02712759e-01 -5.09062171e-01 3.56817156e-01 -6.17500961e-01 -9.09521461e-01 -6.08252645e-01 8.95457119e-02 2.71800518e-01 -1.47307545e-01 1.30972850e+00 6.28274739e-01 3.66277605e-01 3.50759685e-01 -8.00574005e-01 -1.15722992e-01 4.65071231e-01 7.31806979e-02 3.95665392e-02 4.97827858e-01 -1.04915574e-01 1.44422042e+00 -3.78368884e-01 6.36853576e-01 9.08149064e-01 2.11525932e-01 5.43925583e-01 -1.28498673e+00 -4.59625840e-01 -5.07196128e-01 6.92861736e-01 -1.29711831e+00 -1.98868155e-01 4.35527831e-01 8.94287601e-02 1.17590833e+00 6.31136417e-01 8.36719871e-01 4.49043006e-01 3.13009739e-01 3.05820704e-01 1.26451194e+00 -5.80493331e-01 5.62788606e-01 3.04895222e-01 4.07251239e-01 7.63892472e-01 4.81570184e-01 3.94954234e-01 -7.54793584e-01 -4.92783606e-01 -7.07406923e-02 -2.38330916e-01 -6.37849048e-02 -2.62377828e-01 -1.08718884e+00 1.21489131e+00 5.12214065e-01 3.36991847e-01 -4.03261572e-01 1.42087042e-01 5.31500816e-01 5.39895713e-01 3.68629158e-01 9.16218877e-01 -4.93630379e-01 -2.70659655e-01 -7.90651143e-01 4.87310767e-01 7.46348381e-01 4.45851237e-01 9.79369462e-01 -4.45610344e-01 8.59752372e-02 2.63579845e-01 -7.85194784e-02 5.93675315e-01 4.13619190e-01 -7.01656580e-01 -8.75087008e-02 6.42622054e-01 8.24574605e-02 -5.31236768e-01 -7.52025723e-01 -5.30547380e-01 -7.44962394e-01 2.83407003e-01 2.84466952e-01 2.11484894e-01 -5.68643093e-01 1.28361142e+00 4.45619613e-01 -2.79977471e-01 7.42866844e-02 9.88148272e-01 6.13387465e-01 6.26709878e-01 -2.79931158e-01 -5.40525019e-01 1.45791483e+00 -4.61132020e-01 -4.49426532e-01 1.15011662e-01 9.03969467e-01 -7.42339790e-01 5.32771409e-01 8.48029733e-01 -7.73193300e-01 -2.16804594e-01 -1.19834697e+00 9.75849181e-02 -5.45479774e-01 -1.36765972e-01 1.33524513e+00 1.13947427e+00 -6.93332434e-01 1.27670884e+00 -8.70864511e-01 -3.18846107e-01 1.81098536e-01 1.02945852e+00 -4.93816555e-01 -1.16285674e-01 -1.48318589e+00 1.11079729e+00 5.61146319e-01 7.66474381e-02 -3.13627899e-01 -2.18708009e-01 -4.41460282e-01 1.31651297e-01 2.18897417e-01 -8.51031542e-01 8.07903230e-01 -6.25964046e-01 -1.74147630e+00 8.30274105e-01 -1.78146183e-01 -5.97100198e-01 2.10815594e-01 1.30188512e-02 -2.30291754e-01 3.48480381e-02 -2.12561592e-01 1.47561580e-01 6.21996164e-01 -2.25402206e-01 -7.09409297e-01 -5.96239388e-01 -1.41925991e-01 2.86396891e-01 -1.67897567e-01 1.72801420e-01 2.88539648e-01 2.79598951e-01 4.61892903e-01 -1.32681942e+00 -5.25841415e-01 -7.11372018e-01 -9.48365405e-02 -4.66527849e-01 1.98035687e-01 1.59626186e-01 9.62435901e-01 -1.83123124e+00 4.72923785e-01 4.52900857e-01 -2.30300456e-01 2.21922457e-01 6.47020265e-02 6.23859584e-01 5.43081239e-02 -4.42844145e-02 -2.74191387e-02 1.10096700e-01 8.20124000e-02 1.43335499e-02 1.79894064e-02 8.53873849e-01 6.95639476e-02 6.22273922e-01 -8.85767162e-01 -1.94910601e-01 1.98913440e-01 3.67734224e-01 -6.37405932e-01 -4.03980821e-01 -6.51097223e-02 2.83422530e-01 -7.45990336e-01 3.25190932e-01 5.10902643e-01 -1.82033837e-01 1.75352097e-01 1.45858258e-01 -2.95709670e-01 5.32425761e-01 -1.44203925e+00 1.51774549e+00 -2.46597543e-01 4.28671837e-01 -4.29016173e-01 -1.02488136e+00 7.40348577e-01 -1.26951607e-02 3.73202771e-01 -7.74988353e-01 2.22690180e-01 6.09588921e-01 3.23031962e-01 -7.10027292e-02 6.46082878e-01 -4.52197731e-01 -2.46834353e-01 3.10561627e-01 5.70937507e-02 -4.81964618e-01 4.79044855e-01 -1.54753849e-01 1.19476461e+00 -1.70368448e-01 3.46336156e-01 -2.91922092e-01 7.17493474e-01 3.50882411e-01 2.91553527e-01 7.00821936e-01 2.64378991e-02 4.12185848e-01 3.27770293e-01 -8.18377018e-01 -8.74522209e-01 -5.78735888e-01 -5.67300141e-01 9.84571576e-01 4.18603182e-01 -7.23646164e-01 -5.79709888e-01 -1.15868233e-01 -1.84591278e-01 5.81975698e-01 -4.38418925e-01 -4.56592828e-01 -4.04712975e-01 -1.48890579e+00 -3.73401009e-02 -2.36734703e-01 2.48940498e-01 -9.65761423e-01 -9.18823242e-01 3.06762099e-01 2.91492164e-01 -7.51885653e-01 6.18372619e-01 7.60762215e-01 -1.12225640e+00 -9.16226864e-01 -1.53989807e-01 -2.80937940e-01 3.03826451e-01 1.55541614e-01 9.38308477e-01 3.07264656e-01 -6.82578981e-01 -3.35128188e-01 -6.42710865e-01 -1.56838655e-01 -3.42850387e-01 4.40751076e-01 2.64375776e-01 -2.16891572e-01 5.63625515e-01 -2.49429420e-01 -5.25598407e-01 -3.21990997e-02 -7.24593699e-01 -2.77297050e-01 7.30222225e-01 1.22950339e+00 5.92045009e-01 4.57194567e-01 1.65729851e-01 -9.04486477e-01 2.85092950e-01 5.79570569e-02 -8.91269863e-01 1.35724157e-01 -8.09589624e-01 5.27852237e-01 7.56758690e-01 2.75552012e-02 -6.74399614e-01 4.85447347e-01 -2.09507316e-01 5.86784065e-01 -7.87393823e-02 3.83915305e-01 1.74224094e-01 -5.04815400e-01 7.80160844e-01 9.27340686e-02 -2.35212937e-01 -1.44644573e-01 2.16235429e-01 5.32704711e-01 -2.78465867e-01 -2.74705708e-01 6.65480554e-01 5.81067860e-01 8.06431770e-01 -1.04140174e+00 -7.36117899e-01 -3.69444728e-01 -6.91519439e-01 2.58628130e-01 6.83027685e-01 -4.81887519e-01 -8.77220750e-01 2.15717163e-02 -8.25519800e-01 3.37697893e-01 -2.95206964e-01 7.28397131e-01 -4.41828817e-01 2.58417010e-01 -4.38016772e-01 -9.46953475e-01 -3.30277741e-01 -1.38244426e+00 1.05553281e+00 4.62345481e-01 7.82369748e-02 -5.92787683e-01 1.05342127e-01 3.28495890e-01 2.42455304e-01 -1.63711999e-02 9.37716067e-01 -7.38555014e-01 -7.02276111e-01 -8.60162675e-01 1.19519182e-01 -7.86472633e-02 -3.79395187e-01 -3.40868570e-02 -9.73377287e-01 -3.85329574e-01 -1.56602100e-01 -3.34989220e-01 1.03664887e+00 4.82951663e-02 7.27828085e-01 5.32339513e-01 -3.37232381e-01 3.61847579e-01 1.43867123e+00 -8.76369178e-02 7.17232108e-01 5.97239733e-01 -8.17137957e-02 5.74397326e-01 9.75381255e-01 5.29394925e-01 -1.28690749e-01 8.75471175e-01 4.44076985e-01 4.93777812e-01 3.43214601e-01 3.84661198e-01 1.85374573e-01 6.83427513e-01 -3.92171144e-01 2.41816178e-01 -7.82130718e-01 7.27189258e-02 -1.68093455e+00 -1.13768613e+00 -6.03860557e-01 2.74887371e+00 5.20143509e-01 3.69770229e-01 -1.63807228e-01 3.61528933e-01 6.44087791e-01 -2.24323943e-01 -2.42282122e-01 -6.62617445e-01 -1.12509385e-01 8.48425806e-01 5.02943635e-01 2.20897272e-01 -1.10222149e+00 8.80477488e-01 6.14015198e+00 9.63478088e-01 -9.96915936e-01 1.08142205e-01 2.13000551e-01 -3.41780812e-01 -5.89533225e-02 3.97631377e-01 -8.83652270e-01 1.95006996e-01 1.24417031e+00 -2.95323223e-01 6.55913889e-01 6.06917441e-01 -2.51948833e-01 -6.75052404e-01 -1.10220098e+00 1.22222495e+00 -1.72712266e-01 -1.17596495e+00 -1.10819593e-01 2.35769734e-01 7.08258092e-01 3.14364105e-01 4.70006391e-02 3.31295311e-01 -3.80187690e-01 -9.83196795e-01 2.56388605e-01 3.28297883e-01 4.89279747e-01 -1.12086129e+00 1.19880533e+00 3.30504119e-01 -7.37882435e-01 -1.23553403e-01 -7.88945496e-01 -4.67398554e-01 -5.89266457e-02 8.41513574e-01 -2.30271354e-01 8.90001476e-01 6.24152899e-01 2.11195439e-01 -6.96970105e-01 1.32028520e+00 -1.04526103e-01 2.67110467e-01 -7.57033587e-01 -7.17074215e-01 3.76617223e-01 -6.90881729e-01 4.12035167e-01 6.66173100e-01 3.42887998e-01 2.65465587e-01 -2.13259496e-02 2.61938363e-01 1.32031783e-01 4.30080384e-01 -5.24056315e-01 -1.03690006e-01 1.09538980e-01 1.46930373e+00 -1.05992746e+00 -2.15323538e-01 -1.16377644e-01 1.03356993e+00 2.95365930e-01 -4.37742263e-01 -6.16515219e-01 -7.16240287e-01 4.04139519e-01 -3.74324232e-01 1.40462354e-01 -8.87049809e-02 -2.25760669e-01 -1.19944620e+00 -1.75473616e-01 -6.48071826e-01 3.30029935e-01 -7.87273720e-02 -6.99649572e-01 7.21903801e-01 -3.63529086e-01 -1.16331518e+00 -2.34142557e-01 -8.66668344e-01 -1.44286618e-01 6.38640821e-01 -1.35635781e+00 -5.87412775e-01 8.83150771e-02 6.77380636e-02 -2.97348686e-02 -2.86440738e-02 1.42257226e+00 1.36499301e-01 -4.34019059e-01 1.95005208e-01 7.74419665e-01 -6.11121833e-01 6.27098918e-01 -1.25307322e+00 -1.01445422e-01 4.54062849e-01 4.08919066e-01 5.15744209e-01 1.14221644e+00 -2.86472440e-01 -1.99138129e+00 -2.83898622e-01 1.03455269e+00 -3.29020083e-01 5.95881999e-01 -3.35742384e-01 -5.85703492e-01 -1.96269810e-01 -2.20314726e-01 1.68350816e-01 8.67802143e-01 6.12069488e-01 -4.63646501e-02 -2.98218906e-01 -1.06451213e+00 1.86827421e-01 8.25668991e-01 -6.34060025e-01 -3.61163348e-01 8.04288745e-01 8.89289975e-02 -2.62447149e-01 -6.79035664e-01 3.25570256e-01 7.14995503e-01 -1.30738890e+00 7.09356964e-01 -6.37733996e-01 1.71892028e-02 -1.55174956e-01 -6.79081306e-02 -1.23251939e+00 -2.09784776e-01 -7.90775061e-01 2.56722212e-01 5.80015779e-01 5.60207248e-01 -8.27026010e-01 9.00819600e-01 6.06028557e-01 2.37126410e-01 -6.38828933e-01 -1.53615832e+00 -8.08642924e-01 7.96167254e-02 -4.30032998e-01 4.11906540e-01 6.45842791e-01 3.25363517e-01 5.30864477e-01 -2.53356308e-01 -8.58815759e-03 6.62649930e-01 7.17733502e-01 4.64558542e-01 -1.64118052e+00 -3.93237740e-01 -5.70546806e-01 -1.06871843e+00 -1.64471105e-01 1.04191601e-01 -1.11032486e+00 -1.29191488e-01 -1.14249694e+00 4.51514989e-01 -3.70253354e-01 -4.47103173e-01 4.18772139e-02 -1.42895877e-01 5.04811645e-01 8.56020022e-03 -1.82807562e-03 -8.45388770e-01 5.19395113e-01 8.73523355e-01 1.81215536e-02 -1.01919897e-01 1.19438283e-01 -2.30048388e-01 4.43730205e-01 6.89267635e-01 -8.26041222e-01 1.80343047e-01 1.05660513e-01 8.87022972e-01 1.26530165e-02 -5.65155409e-02 -1.18310559e+00 3.63029599e-01 1.22332238e-01 2.14341253e-01 -2.93856591e-01 4.99712318e-01 -5.53464711e-01 1.90524891e-01 8.87914002e-01 -1.53006181e-01 -2.68735796e-01 -2.07013473e-01 4.67446864e-01 -3.24678481e-01 -8.23911369e-01 8.76821995e-01 -7.67952651e-02 -5.25135636e-01 6.33072387e-03 -3.76210421e-01 -3.81191969e-01 1.15696454e+00 2.64568459e-02 -1.92469627e-01 -1.18007198e-01 -6.39812946e-01 2.01458856e-02 5.37597418e-01 -1.11470267e-01 3.92298132e-01 -9.58444297e-01 -5.31923652e-01 -1.03072904e-01 2.56497741e-01 -4.27865267e-01 2.23423317e-01 9.69159484e-01 -7.83504248e-01 7.77206302e-01 -2.06766725e-01 -4.37929690e-01 -1.41197228e+00 5.29345036e-01 2.06136018e-01 -4.60895211e-01 -4.64422137e-01 9.20920670e-01 -4.69225496e-01 -2.15177745e-01 -1.91771641e-01 8.12561065e-02 1.28009645e-02 2.15615839e-01 5.42662859e-01 6.08528435e-01 6.95657670e-01 -4.87911850e-01 -7.21228123e-01 6.57847822e-01 -6.51584491e-02 6.44019619e-02 1.48726737e+00 3.73896837e-01 -5.01925945e-01 3.70736063e-01 1.15787375e+00 -2.07432687e-01 -2.55882919e-01 1.91112682e-01 3.30005378e-01 -2.77957410e-01 4.92585033e-01 -5.49373984e-01 -4.13867027e-01 1.00957131e+00 9.58335936e-01 5.21093249e-01 1.07494056e+00 -1.72927539e-04 6.38589442e-01 1.15710378e+00 1.15550160e+00 -1.27753460e+00 -5.86310208e-01 5.55077493e-01 2.62546897e-01 -1.57561755e+00 5.86214483e-01 -5.63126028e-01 -3.12166482e-01 1.50608456e+00 -2.37972345e-02 -4.20121491e-01 3.57577413e-01 -2.18449458e-01 -4.81691241e-01 -3.48671764e-01 -8.91424954e-01 -8.16881955e-01 3.19074452e-01 -1.31024659e-01 6.18155420e-01 3.40797782e-01 -9.65051770e-01 1.51177526e-01 -3.03708166e-01 -2.82148778e-01 5.62244117e-01 9.72785592e-01 -6.23766124e-01 -1.76569653e+00 -2.93560386e-01 7.68759727e-01 -5.71430504e-01 -8.67392793e-02 -3.16147804e-01 6.19523823e-01 8.75596330e-02 9.06548083e-01 -2.97232181e-01 -4.74567682e-01 1.57195821e-01 1.80622086e-01 9.94525671e-01 -7.49454021e-01 -8.60721111e-01 -2.78817713e-01 2.74115622e-01 -7.73551643e-01 -4.67715502e-01 -9.07713652e-01 -1.48540294e+00 -1.97818533e-01 -1.09733117e+00 7.68825889e-01 1.08882940e+00 1.00087714e+00 2.79289186e-01 2.07868949e-01 6.22759461e-01 -9.82338071e-01 -1.05423665e+00 -5.17914712e-01 -8.76503348e-01 2.68109322e-01 -1.73814986e-02 -8.88311207e-01 -4.04703766e-01 -7.44997919e-01]

[5.612372875213623, 4.905333042144775]

dea562a1-d455-441f-ac61-49b272860312

automatic-noisy-label-correction-for-fine

2205.03011

https://arxiv.org/abs/2205.03011v2

https://arxiv.org/pdf/2205.03011v2.pdf

Automatic Noisy Label Correction for Fine-Grained Entity Typing

Fine-grained entity typing (FET) aims to assign proper semantic types to entity mentions according to their context, which is a fundamental task in various entity-leveraging applications. Current FET systems usually establish on large-scale weakly-supervised/distantly annotation data, which may contain abundant noise and thus severely hinder the performance of the FET task. Although previous studies have made great success in automatically identifying the noisy labels in FET, they usually rely on some auxiliary resources which may be unavailable in real-world applications (e.g. pre-defined hierarchical type structures, human-annotated subsets). In this paper, we propose a novel approach to automatically correct noisy labels for FET without external resources. Specifically, it first identifies the potentially noisy labels by estimating the posterior probability of a label being positive or negative according to the logits output by the model, and then relabel candidate noisy labels by training a robust model over the remaining clean labels. Experiments on two popular benchmarks prove the effectiveness of our method. Our source code can be obtained from https://github.com/CCIIPLab/DenoiseFET.

['Feida Zhu', 'Wei Wei', 'Weiran Pan']

2022-05-06

['entity-typing']

['natural-language-processing']

[-1.01778610e-02 8.03792700e-02 -3.12522382e-01 -5.37809610e-01 -1.12188983e+00 -7.03460813e-01 4.21892434e-01 3.75308961e-01 -5.97621679e-01 1.01822913e+00 2.61013269e-01 -9.39573646e-02 1.74348488e-01 -7.18231678e-01 -6.16774082e-01 -6.64428771e-01 3.99632782e-01 5.33018112e-01 2.89885104e-01 2.26720080e-01 2.02373415e-02 -1.13547944e-01 -1.44670117e+00 2.15021536e-01 1.04637718e+00 8.98435771e-01 1.35695666e-01 2.43256822e-01 -3.23253512e-01 6.49787307e-01 -5.86770475e-01 -6.35223627e-01 -1.39647499e-02 -1.45984083e-01 -1.04687405e+00 -3.45192224e-01 7.99654946e-02 -4.38378146e-03 1.05161354e-01 1.54069686e+00 3.86401534e-01 7.13093653e-02 6.46894753e-01 -1.11922264e+00 -4.76410955e-01 9.03848529e-01 -4.10834879e-01 1.09923363e-01 2.44110703e-01 -1.04761101e-01 1.42966866e+00 -1.10410726e+00 7.65848160e-01 1.04276371e+00 8.86066496e-01 6.33791625e-01 -1.31516910e+00 -7.64312506e-01 2.14388579e-01 2.16639087e-01 -1.69482410e+00 -5.49066782e-01 5.77915609e-01 -4.24299896e-01 5.35537481e-01 4.22161192e-01 1.03380168e-02 1.04539979e+00 -4.35947180e-01 7.31626153e-01 1.17087626e+00 -3.23823601e-01 3.17194223e-01 3.26589435e-01 5.24690449e-01 3.81921411e-01 4.37783808e-01 -3.56077015e-01 -3.63211155e-01 -4.51627851e-01 1.92952290e-01 -2.10313812e-01 -1.32221907e-01 -6.98147416e-02 -1.20200646e+00 4.77237612e-01 3.92581314e-01 4.91425902e-01 -4.04898793e-01 -1.11721821e-01 5.06296396e-01 3.98498364e-02 7.66017377e-01 3.81072342e-01 -7.64392555e-01 -1.58836037e-01 -7.89737165e-01 1.20992094e-01 7.72298574e-01 9.98445034e-01 9.85828280e-01 -6.43715858e-01 -2.55078703e-01 1.04849243e+00 2.87060946e-01 4.32372779e-01 3.37906241e-01 -6.97232068e-01 4.85798389e-01 7.39032805e-01 4.53339756e-01 -7.21904337e-01 -4.91244972e-01 -1.79936975e-01 -8.70867014e-01 -3.67229134e-01 5.38250029e-01 -3.19631308e-01 -8.49862576e-01 1.70893764e+00 7.54285753e-01 4.15171623e-01 -1.41130596e-01 8.68378997e-01 8.57045114e-01 4.58853811e-01 5.33309281e-01 -6.61141053e-02 1.48018467e+00 -7.93336272e-01 -7.51220584e-01 -8.05733874e-02 9.06879604e-01 -7.51181960e-01 1.04909623e+00 4.70896997e-02 -4.59420592e-01 -2.26180628e-01 -5.37871540e-01 -1.62486985e-01 -4.48362976e-01 3.24232370e-01 3.78878117e-01 5.98386943e-01 -5.85254848e-01 5.42668641e-01 -8.77273440e-01 -2.24694863e-01 4.56808418e-01 2.00753391e-01 -3.89148295e-01 -1.14630722e-01 -1.39665663e+00 6.85886145e-01 5.92002809e-01 2.90795743e-01 -5.06918192e-01 -5.53731322e-01 -7.78294027e-01 1.30294695e-01 7.55689144e-01 -4.63633865e-01 1.39934289e+00 -6.38395965e-01 -9.20712233e-01 7.67704368e-01 -4.82986450e-01 -1.70053899e-01 4.86412972e-01 -2.21706554e-01 -3.34158212e-01 -3.55778903e-01 5.97455204e-01 3.22730422e-01 4.07894969e-01 -1.34228718e+00 -1.03949916e+00 -2.16742337e-01 1.54400572e-01 9.69401374e-02 -2.78142005e-01 4.97095510e-02 -4.47700500e-01 -6.04540765e-01 7.25305900e-02 -8.72354031e-01 -3.31797779e-01 -4.75583494e-01 -7.21333385e-01 -7.18189776e-01 4.54900205e-01 -6.93329096e-01 1.48842609e+00 -2.04258037e+00 -1.24607444e-01 2.89583057e-01 3.23057562e-01 2.79717267e-01 1.83427662e-01 1.67012185e-01 2.15854838e-01 4.99519110e-01 -2.73328304e-01 -3.93714637e-01 3.10288638e-01 2.06517085e-01 -4.55389917e-02 3.73175710e-01 7.66872093e-02 7.21166611e-01 -1.24623144e+00 -6.48733377e-01 -1.35957256e-01 2.56578803e-01 -3.29065144e-01 2.74470866e-01 -3.19607615e-01 3.99090856e-01 -6.40125573e-01 7.95843303e-01 6.16379499e-01 -3.46604615e-01 3.62376362e-01 -3.86744201e-01 7.46686906e-02 6.71772182e-01 -1.46648574e+00 1.21590328e+00 -3.43271345e-01 3.88615280e-02 1.00520805e-01 -7.98242450e-01 5.72494745e-01 3.88047338e-01 3.08406800e-01 -3.18282485e-01 9.02320594e-02 4.92255211e-01 -4.29349184e-01 -5.31122565e-01 6.65785134e-01 -2.35627249e-01 -3.82853478e-01 2.95701534e-01 -1.50976740e-02 4.55549389e-01 2.97958285e-01 1.88958287e-01 1.11620188e+00 1.10072948e-01 4.21961457e-01 -1.04675241e-01 5.01345277e-01 -4.04443964e-02 1.11226225e+00 7.23863661e-01 -2.96051323e-01 4.73180354e-01 4.74514186e-01 -1.03218295e-01 -9.43619013e-01 -6.66345000e-01 -3.33814561e-01 1.29665816e+00 3.44543457e-01 -6.17568851e-01 -7.63822615e-01 -1.26378226e+00 -6.54039830e-02 6.40147209e-01 -4.91830975e-01 1.51633382e-01 -3.20810795e-01 -8.96821737e-01 7.65397966e-01 4.17233795e-01 1.52170002e-01 -1.12788975e+00 3.00325006e-02 2.97593117e-01 -6.70715213e-01 -1.08764732e+00 -5.01265168e-01 4.13605869e-01 -2.89833486e-01 -1.14233410e+00 -4.02505159e-01 -7.41593182e-01 8.60643327e-01 9.02410895e-02 1.10418653e+00 3.59709293e-01 2.31155589e-01 -2.47295842e-01 -5.07215023e-01 -5.21645993e-02 -2.94547409e-01 4.82058257e-01 1.49811476e-01 1.31010292e-02 8.50176036e-01 -2.64949232e-01 -3.56868267e-01 4.35256094e-01 -8.19587529e-01 3.85169499e-02 3.72413903e-01 1.06932259e+00 8.30886185e-01 2.57186711e-01 8.67898881e-01 -1.67106581e+00 4.12093252e-01 -8.46813858e-01 -4.82613713e-01 3.70085955e-01 -6.39780819e-01 1.17868371e-01 6.08533382e-01 -2.69280076e-01 -1.31293356e+00 1.91501945e-01 -4.83158320e-01 3.99525836e-02 -4.72182155e-01 6.30422235e-01 -5.51717520e-01 3.73651922e-01 5.74007511e-01 -2.36918718e-01 -9.27728951e-01 -7.43423700e-01 3.60774696e-01 1.03920484e+00 4.92286921e-01 -9.15118873e-01 8.52494597e-01 1.01475410e-01 -5.68587303e-01 -1.96279705e-01 -1.37697613e+00 -9.26241279e-01 -6.84560001e-01 8.11505988e-02 5.47395408e-01 -9.20968354e-01 -3.95506829e-01 3.38841051e-01 -9.41864252e-01 -2.14144483e-01 -7.32083470e-02 2.29832381e-01 8.31489712e-02 3.56432915e-01 -6.72124207e-01 -7.85987377e-01 -3.49347442e-01 -1.02123690e+00 1.17756379e+00 4.32835132e-01 -3.01506937e-01 -8.35511863e-01 -2.78724413e-02 4.79506016e-01 1.24379992e-01 1.91403721e-02 8.36714864e-01 -9.82755899e-01 -3.31518561e-01 -2.24218220e-01 -3.70408505e-01 2.67630160e-01 1.87554598e-01 -2.29331583e-01 -1.04229975e+00 6.99910894e-02 -4.76937801e-01 -3.86435926e-01 8.18579078e-01 -4.87942696e-02 8.97240937e-01 -4.22833830e-01 -5.93029916e-01 2.87704557e-01 1.35033703e+00 -2.12952211e-01 1.94826722e-01 3.49556893e-01 1.02841091e+00 4.45715994e-01 9.39143360e-01 4.03521299e-01 7.07794726e-01 6.35656178e-01 1.14537574e-01 4.79885004e-02 1.19741045e-01 -4.85846072e-01 1.15856238e-01 8.97209108e-01 8.80507603e-02 -2.62005359e-01 -9.47568476e-01 7.96552956e-01 -2.00305820e+00 -7.18456209e-01 -2.58362085e-01 2.03236461e+00 1.48312509e+00 1.89430058e-01 -5.29397465e-02 7.95594156e-02 1.10203683e+00 -1.59784436e-01 -5.32694876e-01 1.00031696e-01 2.90419068e-02 3.59832272e-02 5.30521095e-01 4.26923990e-01 -1.34547973e+00 1.03055859e+00 4.53287745e+00 1.11243916e+00 -6.90479696e-01 3.40993077e-01 6.85757637e-01 1.94549561e-01 -4.51861739e-01 4.65030521e-02 -1.14275134e+00 7.72528291e-01 8.43735158e-01 -3.49465609e-01 1.81236461e-01 8.97574067e-01 6.02391213e-02 -7.35079199e-02 -9.52316523e-01 6.14170909e-01 -3.85731369e-01 -9.88599420e-01 -4.70142096e-01 -1.39846593e-01 9.22902405e-01 6.89846501e-02 -3.77848595e-01 6.47802651e-01 6.31415963e-01 -7.65810549e-01 7.50099540e-01 3.35597932e-01 8.89117897e-01 -6.30152941e-01 1.19806683e+00 6.46881104e-01 -1.22546756e+00 1.28297687e-01 -4.59176034e-01 2.06193477e-01 8.94922838e-02 1.06316578e+00 -8.78709853e-01 5.49066186e-01 7.99510717e-01 5.63590884e-01 -5.65721571e-01 1.08327162e+00 -5.29807925e-01 9.67007637e-01 -5.10300875e-01 -6.78582415e-02 -4.60821912e-02 1.06894687e-01 3.18345696e-01 1.47791541e+00 1.88326463e-01 1.62929535e-01 3.28137815e-01 6.83760405e-01 -6.89114928e-01 2.75252730e-01 -1.47224963e-01 3.68065909e-02 8.75925243e-01 1.64090681e+00 -7.93649614e-01 -4.66830850e-01 -4.24966574e-01 6.59395278e-01 6.18998826e-01 1.91359252e-01 -7.32438207e-01 -3.30465168e-01 5.32417536e-01 -2.80123148e-02 1.84891105e-01 1.44559026e-01 -4.01499063e-01 -1.45415246e+00 1.71999395e-01 -6.89548135e-01 6.31593347e-01 -3.80612999e-01 -1.70476580e+00 4.80950952e-01 -3.67473930e-01 -1.07457078e+00 -3.58403772e-02 -1.41802013e-01 -2.56858557e-01 8.49286497e-01 -1.52358902e+00 -1.09774303e+00 -3.07503819e-01 1.85850665e-01 2.92750150e-01 4.23235208e-01 8.64751637e-01 7.40784228e-01 -8.64620030e-01 8.17562759e-01 2.18895227e-01 3.99563938e-01 1.05736589e+00 -1.59863067e+00 4.59491551e-01 9.43625748e-01 1.33858457e-01 6.25377715e-01 7.07202554e-01 -7.91430950e-01 -8.48300874e-01 -1.45598173e+00 1.34922528e+00 -4.94817793e-01 7.30890274e-01 -4.43424404e-01 -1.24803388e+00 7.79545486e-01 -2.79436976e-01 2.37411395e-01 5.99138737e-01 5.42814791e-01 -5.57842016e-01 1.02023937e-01 -1.22760582e+00 3.65395397e-01 1.09646881e+00 -4.39651132e-01 -4.94937122e-01 3.71245623e-01 5.55329800e-01 -5.50523579e-01 -9.43397582e-01 3.48331869e-01 3.42118979e-01 -4.60087240e-01 6.60282433e-01 -5.05284607e-01 1.73504278e-01 -6.30289674e-01 -1.52950482e-02 -1.39685142e+00 -2.95282811e-01 -1.91282228e-01 1.45698506e-02 1.95158291e+00 7.36563802e-01 -4.64077502e-01 6.63713157e-01 8.08753252e-01 4.28232411e-03 -5.04497707e-01 -9.40515161e-01 -4.90916699e-01 -2.15752736e-01 -3.77324343e-01 6.73211515e-01 1.19252741e+00 1.11065283e-01 4.57346320e-01 -3.94559830e-01 5.91627300e-01 5.38212836e-01 2.09809970e-02 5.38217008e-01 -1.30316186e+00 4.25763614e-02 -1.70987174e-01 -1.61604270e-01 -7.19231308e-01 4.93263960e-01 -1.02123272e+00 5.98651826e-01 -1.46583891e+00 4.14016217e-01 -1.22186959e+00 -5.86110711e-01 9.83389258e-01 -7.96230793e-01 3.93320948e-01 -2.22098038e-01 4.10328329e-01 -1.15952516e+00 3.71123344e-01 6.38755441e-01 -2.03144066e-02 -4.09561880e-02 9.04704705e-02 -9.52764153e-01 8.41360688e-01 7.54906416e-01 -9.72639859e-01 3.27442512e-02 -3.57380480e-01 4.14501101e-01 -3.04300666e-01 1.92048937e-01 -6.49451613e-01 3.18685472e-01 -1.78729653e-01 5.38825765e-02 -3.20744604e-01 -1.32123724e-01 -8.09588373e-01 2.20308036e-01 -8.29320624e-02 -4.57270205e-01 -3.48400056e-01 -1.90233916e-01 6.39153957e-01 -2.78437793e-01 -5.81307530e-01 6.24526262e-01 1.18806232e-02 -8.27974260e-01 2.02794716e-01 5.22196619e-03 4.26162779e-01 6.90776169e-01 4.64668870e-01 -4.69166875e-01 4.00308706e-02 -6.80082083e-01 3.13349187e-01 6.31122947e-01 2.95585930e-01 1.18419472e-02 -1.36116862e+00 -6.97780907e-01 -1.43201113e-01 4.28606838e-01 2.83616960e-01 1.71658397e-01 7.02775657e-01 -5.61750010e-02 2.09623486e-01 3.86981875e-01 -4.00520533e-01 -1.40436757e+00 4.08221662e-01 2.31568530e-01 -4.82734472e-01 -5.55969477e-01 8.88092697e-01 1.15623981e-01 -5.72794914e-01 2.40020260e-01 -1.09321363e-01 -3.26372474e-01 4.75424491e-02 4.56157893e-01 2.43744478e-01 3.31146181e-01 -7.88161993e-01 -4.83132750e-01 1.16113000e-01 -1.84907600e-01 2.11422935e-01 1.32718933e+00 -3.65746021e-01 -1.86151758e-01 4.06281114e-01 8.79424095e-01 3.33079785e-01 -1.02545094e+00 -7.81359613e-01 6.92551732e-01 -3.33252847e-01 -6.41166046e-02 -7.94130802e-01 -8.74113679e-01 4.16097164e-01 1.70651540e-01 3.39142054e-01 9.39602137e-01 1.75546184e-01 9.19220209e-01 3.50476593e-01 4.46474552e-01 -1.11616778e+00 -5.80867946e-01 4.64951068e-01 2.29807377e-01 -1.38718557e+00 -2.05086797e-01 -6.95435643e-01 -6.38431966e-01 4.91837680e-01 5.95077515e-01 2.87095159e-01 5.54580212e-01 3.89052153e-01 2.21083432e-01 4.00181524e-02 -7.66053975e-01 -4.13783163e-01 2.29948640e-01 4.24600482e-01 5.82699955e-01 3.59472007e-01 -4.99280900e-01 1.19420588e+00 -1.13747656e-01 -8.48793089e-02 3.40763718e-01 7.18139291e-01 -4.29064274e-01 -1.27861094e+00 -2.76236087e-01 6.51871681e-01 -8.48429382e-01 -2.21777216e-01 -1.85648903e-01 2.64756352e-01 5.31742454e-01 1.01553667e+00 -4.79691416e-01 -2.17883706e-01 4.42658335e-01 2.64594555e-01 -1.28152579e-01 -8.99014175e-01 -6.86519980e-01 9.78570152e-03 5.03063083e-01 -1.75434843e-01 -4.64076906e-01 -8.11447382e-01 -1.29699755e+00 -2.12060660e-01 -6.93029702e-01 4.76005644e-01 3.84116501e-01 1.10505891e+00 2.98319936e-01 2.79627204e-01 4.86514777e-01 -4.61496979e-01 -4.66916710e-01 -1.14824903e+00 -5.20907402e-01 7.86348939e-01 1.20221592e-01 -8.61410677e-01 -3.82691503e-01 2.35819310e-01]

[9.535276412963867, 8.910539627075195]

0e5ec53b-9ad9-43d6-abe0-ae69e57e0c26

edog-adversarial-edge-detection-for-graph

2212.13607

https://arxiv.org/abs/2212.13607v1

https://arxiv.org/pdf/2212.13607v1.pdf

EDoG: Adversarial Edge Detection For Graph Neural Networks

Graph Neural Networks (GNNs) have been widely applied to different tasks such as bioinformatics, drug design, and social networks. However, recent studies have shown that GNNs are vulnerable to adversarial attacks which aim to mislead the node or subgraph classification prediction by adding subtle perturbations. Detecting these attacks is challenging due to the small magnitude of perturbation and the discrete nature of graph data. In this paper, we propose a general adversarial edge detection pipeline EDoG without requiring knowledge of the attack strategies based on graph generation. Specifically, we propose a novel graph generation approach combined with link prediction to detect suspicious adversarial edges. To effectively train the graph generative model, we sample several sub-graphs from the given graph data. We show that since the number of adversarial edges is usually low in practice, with low probability the sampled sub-graphs will contain adversarial edges based on the union bound. In addition, considering the strong attacks which perturb a large number of edges, we propose a set of novel features to perform outlier detection as the preprocessing for our detection. Extensive experimental results on three real-world graph datasets including a private transaction rule dataset from a major company and two types of synthetic graphs with controlled properties show that EDoG can achieve above 0.8 AUC against four state-of-the-art unseen attack strategies without requiring any knowledge about the attack type; and around 0.85 with knowledge of the attack type. EDoG significantly outperforms traditional malicious edge detection baselines. We also show that an adaptive attack with full knowledge of our detection pipeline is difficult to bypass it.

['Bo Li', 'Carl A. Gunter', 'Alok Lal', 'Hanzhang Wang', 'Yue Yu', 'Xiaojun Xu']

2022-12-27

['edge-detection']

['computer-vision']

[ 3.89880419e-01 4.45597231e-01 -1.06876761e-01 2.55578756e-01 -3.64864647e-01 -9.84697819e-01 4.25351590e-01 4.52761531e-01 1.06944472e-01 6.57561004e-01 -4.16506261e-01 -7.95635700e-01 6.82175020e-03 -1.32896280e+00 -1.30578554e+00 -4.94470268e-01 -5.22090137e-01 4.73763794e-01 5.99816501e-01 -2.75369614e-01 -7.44563527e-03 6.08968079e-01 -6.34495616e-01 2.20377017e-02 9.43059981e-01 5.78814745e-01 -6.53403878e-01 7.75556445e-01 3.51790816e-01 6.45707905e-01 -7.70714402e-01 -7.05700576e-01 7.13169694e-01 -4.47417796e-01 -4.10676390e-01 -7.13348240e-02 3.22027206e-01 -2.79243797e-01 -7.77034044e-01 1.58101892e+00 5.86907744e-01 -1.79249585e-01 4.78498727e-01 -1.68157399e+00 -4.32811379e-01 8.73590350e-01 -5.99684298e-01 1.73025712e-01 4.19396400e-01 4.85769212e-01 9.83017266e-01 -1.54451773e-01 7.79462337e-01 1.05129361e+00 8.45780671e-01 5.08953035e-01 -1.31721723e+00 -8.62111807e-01 2.02533171e-01 1.68482754e-02 -1.29778230e+00 6.44103214e-02 9.54173923e-01 -2.29045317e-01 7.29013979e-01 3.80365878e-01 2.92754620e-01 1.46614993e+00 3.41603845e-01 2.66842961e-01 6.90137744e-01 -1.21147178e-01 4.02740180e-01 -6.92534521e-02 1.38559323e-02 8.26793849e-01 9.67385948e-01 2.57689476e-01 1.54222231e-02 -8.09042275e-01 4.16312128e-01 9.91933867e-02 -4.11518931e-01 -2.72465408e-01 -6.29101813e-01 1.04541957e+00 7.33898640e-01 6.00468963e-02 -1.69427350e-01 2.85767972e-01 7.20207810e-01 5.05755246e-01 2.80719817e-01 6.47563934e-01 -4.19603348e-01 4.10026938e-01 -3.39618176e-01 -1.89056173e-02 1.24084353e+00 8.49594712e-01 5.32116830e-01 2.85024136e-01 -4.00164314e-02 1.26405194e-01 -1.96801629e-02 2.52275288e-01 8.94875675e-02 -2.11223960e-01 5.15266657e-01 7.48039901e-01 -2.82869041e-01 -1.49781394e+00 -3.23237836e-01 -5.69249928e-01 -9.79460299e-01 -4.15043160e-02 6.47695124e-01 -3.91232163e-01 -1.14018989e+00 1.67146194e+00 4.67751473e-01 7.26439238e-01 -1.23436235e-01 4.21685457e-01 5.48403502e-01 2.76388198e-01 -8.45612437e-02 -9.32093933e-02 1.02338374e+00 -5.21650612e-01 -3.37500602e-01 -3.29023540e-01 8.75572562e-01 -3.04879427e-01 8.35152268e-01 4.31424320e-01 -4.85528052e-01 8.09095725e-02 -1.24430120e+00 6.14618719e-01 -5.91004252e-01 -6.97028100e-01 6.88607395e-01 1.13853776e+00 -6.38293505e-01 8.42204034e-01 -8.39970827e-01 -2.57678211e-01 4.82972115e-01 5.03664196e-01 -4.79146510e-01 -1.74017116e-01 -1.39172733e+00 3.88040900e-01 5.20717025e-01 -1.46417722e-01 -1.08629715e+00 -5.34918547e-01 -9.09389377e-01 6.18143082e-02 8.43659222e-01 -5.13794065e-01 6.61804020e-01 -8.57018530e-01 -9.38204765e-01 4.53275323e-01 3.80429059e-01 -8.29849243e-01 7.70004630e-01 1.81451559e-01 -6.22681320e-01 2.79807806e-01 -1.47005692e-01 -3.13267857e-01 9.33021009e-01 -1.05543244e+00 -3.57783288e-02 -2.97744066e-01 3.52803171e-01 -3.44120055e-01 -4.16370898e-01 -1.27862185e-01 -2.95976847e-01 -7.52018631e-01 -1.91500202e-01 -1.08508205e+00 -5.81902087e-01 -3.79161656e-01 -1.17294633e+00 1.02868803e-01 9.29979622e-01 -3.33966613e-01 1.24154198e+00 -1.93098307e+00 -2.92254955e-01 9.40332770e-01 6.15015388e-01 4.03755128e-01 -1.93215057e-01 6.48925662e-01 -3.10597152e-01 5.40784061e-01 -2.47240603e-01 2.45168716e-01 -5.64292222e-02 4.99927960e-02 -4.51950252e-01 6.89326823e-01 1.47084668e-01 9.88002300e-01 -1.10854685e+00 6.15616702e-02 -1.87708184e-01 8.55787843e-02 -6.42367184e-01 1.47450000e-01 -4.04951602e-01 2.12505490e-01 -5.48982799e-01 7.14386106e-01 7.24829853e-01 -4.02027994e-01 5.39930165e-01 -3.15068550e-02 9.69310284e-01 6.56447038e-02 -1.29807270e+00 8.62553835e-01 1.48284912e-01 1.12476856e-01 -5.93208522e-02 -1.02441430e+00 7.24638760e-01 1.04499698e-01 1.80646867e-01 -9.82933939e-02 2.67948896e-01 1.92594543e-01 5.07312596e-01 -2.60502081e-02 -2.24215597e-01 1.52572289e-01 -2.18622074e-01 5.12217283e-01 -8.86804610e-02 1.83795214e-01 2.97991008e-01 6.93840563e-01 2.11524272e+00 -5.95102906e-01 4.01246399e-01 9.76143852e-02 3.04241478e-01 -1.14283748e-01 7.68671095e-01 1.10003281e+00 -1.50065988e-01 3.78309488e-01 1.16507554e+00 -3.81209821e-01 -7.41084576e-01 -1.07522893e+00 2.98595607e-01 5.49238205e-01 9.33737084e-02 -7.15723932e-01 -7.65130222e-01 -1.45708561e+00 2.37519145e-01 5.39214909e-01 -6.86526358e-01 -7.19895482e-01 -4.56057578e-01 -9.60359275e-01 9.30795252e-01 3.01322520e-01 1.99735731e-01 -8.54476154e-01 3.76239330e-01 1.98650926e-01 3.51314962e-01 -1.46963108e+00 -8.19034338e-01 1.08464487e-01 -4.91293907e-01 -1.72274566e+00 1.77191511e-01 -4.59625542e-01 1.06728756e+00 -4.47191373e-02 9.98301744e-01 4.06730890e-01 -3.69016618e-01 1.90658540e-01 -3.45891327e-01 -4.55826283e-01 -9.33836758e-01 1.83041275e-01 1.46569714e-01 1.10563830e-01 2.71165580e-01 -9.67567503e-01 -3.32049549e-01 1.93033636e-01 -1.12025130e+00 -6.16447389e-01 5.09121656e-01 6.88185215e-01 3.04485232e-01 4.85735476e-01 6.19532049e-01 -1.65748453e+00 8.41410875e-01 -7.96057940e-01 -7.45168865e-01 1.22707896e-01 -6.57698333e-01 2.47360282e-02 1.23347604e+00 -8.39508057e-01 -1.26324043e-01 3.37350629e-02 4.57007065e-02 -6.37736022e-01 -7.75736105e-03 5.26357234e-01 -5.63917875e-01 -4.39122796e-01 9.93198633e-01 6.43761363e-03 -5.99123649e-02 5.50781712e-02 1.95630237e-01 1.19515821e-01 2.52034605e-01 -3.18139970e-01 1.51146746e+00 2.53220499e-01 5.35916626e-01 -5.14602065e-01 -4.30969536e-01 -8.60284194e-02 -8.51517469e-02 4.83960956e-02 2.17848346e-01 -5.03491163e-01 -8.56846571e-01 4.51884806e-01 -9.13114607e-01 -2.44875535e-01 -7.09113991e-03 8.71646702e-02 -4.43290211e-02 9.03206766e-01 -8.05211484e-01 -5.45834839e-01 -6.38117731e-01 -1.09809399e+00 5.04367530e-01 -1.69091523e-01 -6.81098029e-02 -1.13707864e+00 -1.48724005e-01 -4.46416773e-02 1.51044369e-01 1.16098928e+00 9.25731540e-01 -1.55536520e+00 -6.36079490e-01 -7.85230219e-01 2.88251415e-02 3.66210639e-01 3.43355745e-01 7.73506910e-02 -6.91432238e-01 -6.78962946e-01 -3.02119642e-01 -1.21315449e-01 6.80228174e-01 -4.33170795e-03 1.25324595e+00 -7.31158555e-01 -5.57229578e-01 6.91990316e-01 1.34899282e+00 3.79487649e-02 6.20389819e-01 5.21734692e-02 1.06927717e+00 1.22513153e-01 2.07531065e-01 2.11308792e-01 -1.52397946e-01 3.04627895e-01 8.51653278e-01 -1.01109177e-01 2.05539063e-01 -5.92924476e-01 6.07999146e-01 2.02348724e-01 2.21643075e-01 -6.53877854e-01 -9.10239339e-01 2.12105229e-01 -1.68902588e+00 -8.96062791e-01 -3.00414801e-01 2.35312057e+00 7.12078452e-01 7.38439500e-01 2.44531542e-01 2.38337100e-01 9.83989775e-01 6.89464360e-02 -7.73713529e-01 -4.42280889e-01 -2.56536882e-02 3.63950431e-01 1.06803727e+00 4.52152938e-01 -1.14059198e+00 9.16087031e-01 5.47429848e+00 9.33768988e-01 -9.07451630e-01 -2.06809282e-01 5.82032382e-01 2.32446805e-01 -2.23201171e-01 1.47796124e-01 -6.90194786e-01 5.98804951e-01 1.03927100e+00 -5.25107265e-01 5.89881182e-01 8.78574491e-01 -1.15369476e-01 4.11738187e-01 -1.14063776e+00 4.54727173e-01 4.91959043e-02 -1.20009530e+00 2.55401045e-01 3.95215720e-01 7.42765248e-01 5.79221593e-03 -1.47668064e-01 3.48837137e-01 9.38959658e-01 -1.14693511e+00 -3.52138914e-02 1.14167772e-01 5.86525202e-01 -1.02452600e+00 6.75586939e-01 4.73526388e-01 -1.07194662e+00 -1.33297741e-01 -1.59956932e-01 2.10510328e-01 -1.61210403e-01 8.31441641e-01 -1.36614549e+00 9.27435458e-01 3.12525153e-01 4.04125094e-01 -6.51697397e-01 8.62462819e-01 -5.39125264e-01 1.10670912e+00 -5.99559307e-01 1.17287017e-01 1.62226468e-01 -9.42420810e-02 9.35920656e-01 8.84686470e-01 1.22282661e-01 -2.05666736e-01 4.77667421e-01 6.69857860e-01 -6.45774245e-01 -6.22070730e-02 -1.09556997e+00 -3.37954640e-01 6.44144118e-01 1.22622442e+00 -8.40043187e-01 -8.74889921e-03 -3.21529448e-01 8.67883623e-01 3.61294478e-01 3.63012880e-01 -1.04009533e+00 -6.99360013e-01 5.43368995e-01 4.39592034e-01 2.49863878e-01 1.31494805e-01 8.64905268e-02 -1.10120738e+00 1.17296606e-01 -1.26214349e+00 7.50670910e-01 -1.14406019e-01 -1.61683702e+00 5.92424333e-01 -2.61350513e-01 -1.17066634e+00 -2.19991714e-01 -6.18649006e-01 -1.16612101e+00 5.39683998e-01 -1.01537061e+00 -1.02691901e+00 -1.42930776e-01 6.62000358e-01 -1.59473583e-01 -1.94686383e-01 6.78740799e-01 1.11751594e-01 -7.75571465e-01 1.05436420e+00 -1.52072504e-01 7.57904768e-01 7.34448314e-01 -1.29815924e+00 8.72250617e-01 1.40738738e+00 3.67134996e-02 5.63162148e-01 8.68405700e-01 -1.20386457e+00 -1.50366628e+00 -1.60072720e+00 1.43857732e-01 -3.76161397e-01 1.05387425e+00 -6.57643259e-01 -1.04703939e+00 1.03509879e+00 -2.34034747e-01 6.38454020e-01 4.93992537e-01 -7.28149861e-02 -5.27978003e-01 -2.92257760e-02 -1.50571597e+00 7.03039110e-01 1.27212107e+00 -3.06069732e-01 -3.69019955e-02 7.15218723e-01 8.89236569e-01 -4.59232211e-01 -7.49620438e-01 6.86690032e-01 -1.49412570e-03 -6.79821730e-01 9.29111481e-01 -9.96799231e-01 -2.58767437e-02 -3.42099726e-01 1.96692407e-01 -1.39488244e+00 -1.68148577e-01 -1.14128804e+00 -6.37657464e-01 1.07615232e+00 5.45789063e-01 -1.22901845e+00 9.57897425e-01 3.51644784e-01 2.01748013e-01 -7.33523607e-01 -8.50497961e-01 -1.13465893e+00 -1.10190839e-01 -2.67540276e-01 6.02904141e-01 1.20696270e+00 -7.75152296e-02 2.63532490e-01 -2.93259889e-01 8.01624894e-01 7.81895816e-01 -2.51283944e-01 1.09556377e+00 -1.20416975e+00 -6.33493602e-01 -2.58810669e-01 -9.53752458e-01 -3.49957138e-01 1.97283059e-01 -1.06780815e+00 -4.66385752e-01 -1.11979246e+00 5.73336706e-02 -1.36332527e-01 -3.17408264e-01 5.91528475e-01 -5.06606102e-01 2.83609122e-01 -2.51027137e-01 -2.11344332e-01 -2.73302615e-01 1.09752424e-01 9.29980993e-01 -3.50866020e-01 -1.47210434e-01 2.20593944e-01 -8.22813570e-01 7.56001770e-01 9.36180651e-01 -7.77864397e-01 -5.51553428e-01 3.66101772e-01 3.74493062e-01 -2.85730809e-01 3.50693256e-01 -7.48622656e-01 1.23673424e-01 3.15590799e-02 1.41172975e-01 -1.27121255e-01 -1.98814452e-01 -7.71958590e-01 2.76092678e-01 7.84471512e-01 7.74479434e-02 3.07123009e-02 1.74808174e-01 1.12537777e+00 1.80748016e-01 -1.59176439e-02 7.41431832e-01 6.41318038e-02 -1.94201380e-01 7.60599792e-01 5.88861853e-02 3.23517591e-01 1.47227550e+00 -7.79405236e-02 -6.21574283e-01 -5.84654510e-01 -7.22304761e-01 2.42416963e-01 6.25178218e-01 2.12921783e-01 5.39630175e-01 -1.12220216e+00 -6.91159427e-01 2.67801523e-01 1.10277213e-01 -1.41404822e-01 -1.26860231e-01 6.88868701e-01 -5.30633152e-01 -2.47872502e-01 1.70350239e-01 -2.45428652e-01 -1.35904968e+00 1.02416909e+00 5.00861824e-01 -7.47305930e-01 -5.13667405e-01 5.18820882e-01 6.19304106e-02 -4.03999388e-01 1.17057391e-01 2.09307810e-03 2.61285424e-01 -4.44788218e-01 1.71458483e-01 3.71841788e-01 1.42551452e-01 -2.78170466e-01 -4.30998117e-01 -9.56280064e-03 -3.90254080e-01 7.39773870e-01 1.14296353e+00 5.39830327e-01 -7.28802234e-02 -1.23051144e-01 1.05171311e+00 4.68835741e-01 -8.48470032e-01 -1.37727529e-01 7.27031380e-02 -3.87353778e-01 -3.23803216e-01 -5.70191145e-01 -1.22062910e+00 3.83924484e-01 9.97283906e-02 7.10647643e-01 1.16000199e+00 -1.48619235e-01 8.38251472e-01 3.93539488e-01 3.49254876e-01 -5.25704563e-01 1.23147890e-02 2.93356210e-01 5.62238395e-01 -1.02826834e+00 9.58148688e-02 -9.38441455e-01 -2.84690946e-01 8.55905235e-01 7.97185183e-01 -4.60282147e-01 6.07939601e-01 3.55293453e-01 -3.32710832e-01 -2.21291214e-01 -6.78118050e-01 1.08828634e-01 1.67936116e-01 6.13651395e-01 -1.75636694e-01 5.04349805e-02 -1.33919537e-01 6.04582012e-01 -8.40165094e-02 -5.56719601e-01 8.20393205e-01 6.85139298e-01 -1.97316363e-01 -1.22213054e+00 -2.47904941e-01 8.37301195e-01 -8.16431761e-01 -2.30128348e-01 -9.25622165e-01 7.83493340e-01 -2.20584318e-01 9.37876463e-01 -4.83975828e-01 -6.85610235e-01 4.58354652e-01 -2.04458758e-01 1.76732719e-01 -6.81800246e-01 -6.59868777e-01 -3.49732131e-01 3.18035036e-01 -7.14490712e-01 4.13488179e-01 -2.69514501e-01 -1.10437930e+00 -5.62682867e-01 -6.10796273e-01 2.23678067e-01 5.30106053e-02 5.92464328e-01 6.52456105e-01 4.83889848e-01 9.36804414e-01 -2.92394966e-01 -8.52137506e-01 -9.14252043e-01 -7.71293521e-01 6.95673406e-01 1.13694727e-01 -4.38935488e-01 -9.84266639e-01 -4.60494071e-01]

[6.117924213409424, 7.350695610046387]

06948395-95cd-4d6a-8c24-07e0887b0e8d

exploring-representation-level-augmentation

2210.12285

https://arxiv.org/abs/2210.12285v1

https://arxiv.org/pdf/2210.12285v1.pdf

Exploring Representation-Level Augmentation for Code Search

Code search, which aims at retrieving the most relevant code fragment for a given natural language query, is a common activity in software development practice. Recently, contrastive learning is widely used in code search research, where many data augmentation approaches for source code (e.g., semantic-preserving program transformation) are proposed to learn better representations. However, these augmentations are at the raw-data level, which requires additional code analysis in the preprocessing stage and additional training costs in the training stage. In this paper, we explore augmentation methods that augment data (both code and query) at representation level which does not require additional data processing and training, and based on this we propose a general format of representation-level augmentation that unifies existing methods. Then, we propose three new augmentation methods (linear extrapolation, binary interpolation, and Gaussian scaling) based on the general format. Furthermore, we theoretically analyze the advantages of the proposed augmentation methods over traditional contrastive learning methods on code search. We experimentally evaluate the proposed representation-level augmentation methods with state-of-the-art code search models on a large-scale public dataset consisting of six programming languages. The experimental results show that our approach can consistently boost the performance of the studied code search models. Our source code is available at https://github.com/Alex-HaochenLi/RACS.

['Yanlin Wang', 'Hongyu Zhang', 'Yuan Huang', 'Yanxian Huang', 'Cyril Leung', 'Chunyan Miao', 'Haochen Li']

2022-10-21

['code-search', 'code-search']

['computer-code', 'computer-vision']

[ 1.36693031e-01 -4.09792125e-01 -7.89439619e-01 -1.75814882e-01 -9.66296017e-01 -3.36756527e-01 4.60470825e-01 5.66901684e-01 -2.79141456e-01 1.62132412e-01 4.92850132e-02 -6.40000165e-01 3.51982787e-02 -7.62332499e-01 -7.76815891e-01 -2.82621980e-01 5.73292673e-02 1.30628929e-01 2.78289735e-01 -2.91914701e-01 6.86359346e-01 1.11944929e-01 -1.72374129e+00 2.34799027e-01 1.41856480e+00 8.08793068e-01 4.72264916e-01 4.02890712e-01 -6.68184936e-01 9.31476712e-01 -2.64312029e-01 -2.57024616e-01 1.43160820e-01 -2.65540123e-01 -9.27047014e-01 -3.97017896e-01 2.03163669e-01 -6.57511130e-02 -3.59011769e-01 1.32782543e+00 2.16576338e-01 1.21859640e-01 3.22431386e-01 -1.29440629e+00 -1.06900275e+00 5.11681199e-01 -8.92766535e-01 2.74943322e-01 4.54241633e-01 -1.18115932e-01 1.04756045e+00 -1.24898267e+00 3.36974025e-01 8.65523696e-01 7.32682884e-01 4.90994573e-01 -9.70289528e-01 -7.85304844e-01 9.46557671e-02 3.65921378e-01 -1.44042170e+00 -2.71408737e-01 1.06839752e+00 -6.86472535e-01 9.11459684e-01 2.80639917e-01 3.36090714e-01 7.09598303e-01 -3.70146148e-02 1.00365043e+00 6.81357861e-01 -6.83791876e-01 2.48679817e-02 1.66633546e-01 3.80276263e-01 1.01797616e+00 1.58121392e-01 -6.09591529e-02 -7.75170550e-02 -7.58700371e-01 5.55143952e-01 3.22362810e-01 -3.05002868e-01 -6.82721436e-01 -1.06999969e+00 9.42218244e-01 5.82961857e-01 3.90251964e-01 -1.42193794e-01 2.76873976e-01 8.04867744e-01 2.47522399e-01 3.36590588e-01 2.71220624e-01 -5.37477255e-01 -2.45979667e-01 -1.00352502e+00 2.29477286e-01 5.06437421e-01 1.38884819e+00 1.04866660e+00 8.48207846e-02 -2.06342757e-01 1.14822066e+00 5.40588498e-01 3.09202790e-01 9.76549923e-01 -3.15216899e-01 8.05866659e-01 1.13739657e+00 -2.33624205e-01 -1.02457821e+00 -5.86329848e-02 -4.57129717e-01 -8.04749727e-01 8.21563751e-02 -1.66504875e-01 1.75392672e-01 -7.15611756e-01 1.52019846e+00 5.42898066e-02 1.18272014e-01 -8.45943019e-03 4.53152508e-01 1.00333381e+00 6.21951640e-01 2.83750352e-02 -6.65032938e-02 1.25993633e+00 -1.52315629e+00 -4.38552320e-01 -2.78632373e-01 1.02100599e+00 -8.88358116e-01 1.59715962e+00 1.25681281e-01 -9.30030227e-01 -7.88423419e-01 -9.09662306e-01 -2.91765004e-01 -4.43394750e-01 4.90049630e-01 8.24128270e-01 6.23771131e-01 -1.00603032e+00 2.41539106e-01 -7.54030526e-01 -2.90867269e-01 3.10843349e-01 2.56959736e-01 -1.40698075e-01 -1.19324945e-01 -9.10766661e-01 4.33821261e-01 4.67836082e-01 -3.11489403e-01 -7.03694701e-01 -6.92058802e-01 -1.04635239e+00 3.49141359e-01 4.30675566e-01 -2.78368682e-01 1.27004504e+00 -8.93699765e-01 -9.07296002e-01 7.43680656e-01 -1.92494944e-01 -1.89588040e-01 1.40125722e-01 -2.07101896e-01 -3.01880300e-01 -3.06826860e-01 1.32387325e-01 2.65925735e-01 6.02760315e-01 -1.16405797e+00 -3.30855101e-01 -1.70622781e-01 2.56426096e-01 -1.22172900e-01 -7.86246002e-01 2.87195891e-01 -8.21507633e-01 -1.06778300e+00 -4.67047766e-02 -8.74828219e-01 -2.83214748e-01 9.60538685e-02 -5.82571700e-02 -4.75416392e-01 7.18722880e-01 -7.56772816e-01 1.74372029e+00 -2.40018749e+00 7.90587738e-02 7.08185360e-02 8.12268704e-02 3.78674120e-01 -3.21278840e-01 5.12989938e-01 -2.80674785e-01 2.76201814e-01 -6.65125191e-01 -2.56535143e-01 -3.14426385e-02 -1.96386114e-01 -3.33498836e-01 2.05957651e-01 9.60142612e-02 9.55616415e-01 -8.35863352e-01 -6.55640721e-01 -1.79086208e-01 2.39840567e-01 -9.52731013e-01 3.20938289e-01 -3.16615939e-01 1.09150305e-01 -6.86808467e-01 8.54291320e-01 6.30212665e-01 -4.79400516e-01 -3.52076352e-01 1.36518747e-01 -2.78714716e-01 2.54953206e-01 -9.11230505e-01 2.19475102e+00 -8.18407953e-01 4.10066336e-01 -2.70580292e-01 -1.12589300e+00 1.18899953e+00 3.59876677e-02 3.85218769e-01 -5.32418489e-01 -1.09738871e-01 5.40828228e-01 -1.33525312e-01 -6.52391315e-01 5.32777727e-01 6.38239026e-01 -2.50469416e-01 5.93310475e-01 -1.97437674e-01 8.08485597e-02 2.75356621e-01 7.14754611e-02 1.16041780e+00 2.16883138e-01 3.75518352e-01 -4.47641909e-01 1.19094479e+00 1.85009822e-01 4.41816598e-01 5.98511636e-01 1.01261206e-01 2.63944328e-01 3.26396912e-01 -4.18016553e-01 -9.80183184e-01 -3.36279869e-01 8.23634565e-02 1.43466794e+00 2.18495354e-01 -7.10796714e-01 -6.93545341e-01 -8.21934760e-01 -9.70542282e-02 4.61645037e-01 -5.86004734e-01 -3.96887779e-01 -6.82669938e-01 -5.95444500e-01 6.90659881e-01 6.82973266e-01 7.96987057e-01 -1.11824214e+00 -4.52676505e-01 -5.21359295e-02 -1.50649235e-01 -5.75670898e-01 -8.24850142e-01 6.08216152e-02 -8.87385488e-01 -1.22795630e+00 -7.43350267e-01 -1.28415847e+00 8.70925725e-01 4.49518740e-01 1.10314000e+00 9.59128499e-01 -2.22756073e-01 3.06346744e-01 -6.54084682e-01 -4.28257063e-02 -6.69905007e-01 3.85578811e-01 -3.74095082e-01 -3.97754669e-01 3.56878638e-01 -4.06977922e-01 -3.98928195e-01 2.63699144e-02 -1.12006700e+00 5.15439734e-02 9.22606885e-01 9.49174225e-01 5.27265191e-01 -2.19278336e-01 4.37866688e-01 -9.24587429e-01 8.66958797e-01 -8.49831820e-01 -8.79654586e-01 5.48789918e-01 -9.13224995e-01 3.19931000e-01 7.29774773e-01 -4.37464118e-01 -1.00008810e+00 1.76882669e-01 -3.87163311e-01 -5.32520771e-01 7.69008845e-02 1.06790698e+00 1.14828080e-01 -4.49820220e-01 8.49042237e-01 7.72216856e-01 -8.25238526e-02 -8.49687159e-01 2.28940591e-01 8.41208756e-01 4.34459031e-01 -9.58037138e-01 9.56263125e-01 3.56256254e-02 -3.83183628e-01 -4.47735518e-01 -1.86895445e-01 -6.16972387e-01 -5.21668911e-01 3.10758799e-01 3.80841464e-01 -8.37136328e-01 -3.43082815e-01 2.48663455e-01 -1.25020647e+00 -7.08991885e-02 3.43674757e-02 3.48662972e-01 -5.40592670e-01 6.19436860e-01 -3.78248900e-01 -5.42575777e-01 -5.94008565e-01 -1.55812359e+00 9.74556804e-01 7.63113499e-02 1.73040837e-01 -7.49639690e-01 3.18346649e-01 9.06573012e-02 5.59426963e-01 7.82189518e-02 1.42055809e+00 -8.14579487e-01 -6.78636253e-01 -1.32531792e-01 -3.15222561e-01 2.18455344e-01 2.14056551e-01 3.32196765e-02 -6.80207491e-01 -4.32843029e-01 3.51035781e-02 -4.08690482e-01 8.23005617e-01 -1.60576507e-01 1.73424125e+00 -4.95207250e-01 -3.75744939e-01 6.62117958e-01 1.74326587e+00 4.45749879e-01 5.47194123e-01 5.35641015e-01 7.60569692e-01 2.75376201e-01 6.92294538e-01 4.52591836e-01 3.48942846e-01 8.27531159e-01 5.20591021e-01 6.08922797e-04 1.93943959e-02 -2.18533501e-01 2.39666000e-01 1.19198108e+00 5.23391180e-02 1.68444753e-01 -1.31331551e+00 7.07811952e-01 -1.96601307e+00 -6.31421685e-01 -5.80072105e-02 2.25556087e+00 1.09221554e+00 -1.42852545e-01 8.87528062e-02 1.05895132e-01 7.28787005e-01 2.26469757e-03 -6.13984585e-01 -4.00688678e-01 3.62042129e-01 2.48457596e-01 2.43265882e-01 3.10620636e-01 -1.14396477e+00 8.15215886e-01 5.16587639e+00 1.00001085e+00 -9.29129362e-01 2.51206815e-01 3.12650442e-01 5.39209008e-01 -4.97805387e-01 9.24112201e-02 -5.16164482e-01 5.46943545e-01 5.87012112e-01 -6.59572423e-01 5.95169544e-01 1.57764328e+00 -3.23903114e-01 2.72950828e-01 -1.20582879e+00 1.12795687e+00 1.96873948e-01 -1.43632936e+00 -2.85931975e-02 -2.34593689e-01 7.88065255e-01 8.22552145e-02 3.33583765e-02 8.99243176e-01 3.25773098e-02 -6.94224477e-01 5.31668782e-01 1.98407844e-01 6.84239805e-01 -6.28381491e-01 6.97580636e-01 4.67262030e-01 -1.57831442e+00 -3.08990300e-01 -2.66809344e-01 1.72088355e-01 -5.53729177e-01 2.93686122e-01 -3.23225379e-01 6.05457485e-01 6.38484299e-01 7.34171450e-01 -1.07486320e+00 1.35355639e+00 1.44464271e-02 3.30045521e-01 7.99147338e-02 -2.20519677e-01 9.38831046e-02 1.12282999e-01 1.93662405e-01 1.36710310e+00 4.64484990e-01 -3.50078970e-01 3.13400805e-01 1.34755158e+00 -3.36481690e-01 5.10090411e-01 -6.18563235e-01 -6.07417971e-02 5.73198557e-01 9.64678824e-01 -4.53302115e-01 -3.66490543e-01 -9.20001209e-01 7.28623986e-01 3.60057354e-01 2.21069261e-01 -9.63872373e-01 -8.29754114e-01 4.33635622e-01 -1.81952551e-01 1.34186789e-01 -1.08913362e-01 -1.03010215e-01 -1.20447707e+00 6.22514069e-01 -1.17940724e+00 4.05877560e-01 -3.78058195e-01 -8.91471028e-01 6.92012429e-01 9.15806964e-02 -1.46771526e+00 -2.61507064e-01 -2.46331394e-01 -6.69200718e-01 8.38136971e-01 -1.74503541e+00 -1.07420754e+00 -5.96258104e-01 5.74986815e-01 9.00010765e-01 -4.23938304e-01 7.01795697e-01 7.41088033e-01 -4.06830221e-01 8.97805870e-01 1.66175395e-01 2.84518898e-01 3.47660393e-01 -9.86628473e-01 6.72341228e-01 8.95369411e-01 1.44645825e-01 1.06719613e+00 1.46344423e-01 -6.17400229e-01 -1.61894679e+00 -1.04883838e+00 6.81989849e-01 -9.97107178e-02 7.73503065e-01 -2.66163856e-01 -1.30193090e+00 5.52461922e-01 1.41688481e-01 2.38710314e-01 5.45330346e-01 -5.97011894e-02 -5.08071721e-01 4.41500880e-02 -9.59745228e-01 3.24199915e-01 8.78940046e-01 -7.10927367e-01 -5.59331596e-01 3.54197741e-01 8.97204578e-01 -4.16784346e-01 -8.41662049e-01 6.09038413e-01 2.49314457e-01 -7.25449324e-01 9.56154585e-01 -5.68046272e-01 6.63799047e-01 -4.28208709e-01 -8.47054049e-02 -9.37648714e-01 -2.81449974e-01 -2.88232803e-01 -2.44733796e-01 1.32516980e+00 2.44933978e-01 -4.50644642e-01 6.86294436e-01 3.43396336e-01 -3.27703536e-01 -1.00665545e+00 -6.66096568e-01 -1.01567233e+00 1.75016984e-01 -2.49782875e-01 7.97990799e-01 1.21533906e+00 1.95570648e-01 -2.39106975e-02 -7.72073492e-02 1.83882806e-02 2.81381130e-01 4.52792168e-01 7.68664956e-01 -1.22573996e+00 -4.04765695e-01 -7.24650323e-01 -2.55044550e-01 -1.20476806e+00 3.66786093e-01 -1.29498208e+00 8.27127416e-03 -1.37066543e+00 4.46399122e-01 -7.35468209e-01 -3.80219668e-01 7.32024789e-01 -4.35174376e-01 -3.20183307e-01 2.68399362e-02 6.86309814e-01 -3.83403182e-01 7.53860414e-01 7.15341866e-01 -5.48402846e-01 -2.63986737e-01 6.84475303e-02 -7.21055150e-01 6.26227915e-01 8.08158875e-01 -4.96612728e-01 -6.35189652e-01 -6.96016431e-01 3.48400235e-01 4.89920974e-02 2.27774560e-01 -1.06838870e+00 4.21529770e-01 -3.09923086e-02 -2.30248734e-01 -3.95027250e-01 -2.01567799e-01 -9.39112246e-01 -3.73946913e-02 7.99579501e-01 -6.03560686e-01 4.75460440e-01 5.11108816e-01 4.62913334e-01 -4.24177080e-01 -7.67617226e-01 7.15070665e-01 -1.70218721e-01 -8.37406099e-01 3.75254780e-01 7.45193809e-02 1.82099819e-01 8.27533126e-01 1.22282311e-01 -4.32830691e-01 1.29409671e-01 -1.62498921e-01 1.32873446e-01 5.88288546e-01 8.29384923e-01 6.75086617e-01 -1.63768339e+00 -5.93197405e-01 2.37785697e-01 7.06496596e-01 -1.17052056e-01 -9.99439731e-02 6.39587045e-01 -6.53258920e-01 5.01780391e-01 -6.77942634e-02 -3.80459785e-01 -1.21107125e+00 1.03560495e+00 -2.50427797e-02 -2.99811572e-01 -2.83597499e-01 7.92991757e-01 1.55273393e-01 -7.04136670e-01 3.51174414e-01 -4.95091289e-01 -3.35662335e-01 -4.37810898e-01 4.72135365e-01 2.45853662e-01 1.38038903e-01 -4.42181677e-01 -4.77380753e-01 7.13344693e-01 -2.59898007e-01 5.46246827e-01 1.20940697e+00 3.58927161e-01 -4.97826874e-01 1.77569881e-01 1.40534675e+00 1.88525155e-01 -5.51810980e-01 -5.43034196e-01 4.28968698e-01 -6.89687610e-01 -6.03999682e-02 -3.54604572e-01 -1.21431708e+00 8.28650951e-01 7.07944572e-01 1.99852765e-01 1.27200675e+00 -1.05114253e-02 6.27245009e-01 6.48316979e-01 4.90937173e-01 -7.38501072e-01 1.07802391e-01 3.58003050e-01 9.96764898e-01 -1.38555646e+00 3.24727967e-02 -4.51299995e-01 -2.02431545e-01 1.16707551e+00 1.05956733e+00 1.42281381e-02 6.58647656e-01 1.68953344e-01 -3.82492840e-01 -1.41749769e-01 -6.25272751e-01 -7.68586174e-02 3.01554769e-01 2.69812703e-01 7.20132887e-01 -4.35712606e-01 -6.05338752e-01 5.77386260e-01 2.75585383e-01 3.08018979e-02 3.67935002e-01 1.31564546e+00 -3.44173729e-01 -1.45984566e+00 -2.77267188e-01 5.55829227e-01 -2.28419274e-01 -5.90006828e-01 -2.20747218e-01 7.29431570e-01 4.02267314e-02 6.56084299e-01 -2.55196840e-01 -5.03428757e-01 3.52542073e-01 4.52864431e-02 9.85738076e-03 -8.86558831e-01 -6.43875718e-01 -3.33465844e-01 -5.46161115e-01 -4.73255694e-01 -3.62319052e-01 -5.43085396e-01 -1.23274291e+00 2.10131332e-02 -5.58421135e-01 4.61163521e-01 6.69301808e-01 4.64835942e-01 6.24904931e-01 4.93807286e-01 6.73090279e-01 -4.87214357e-01 -5.40113628e-01 -9.15450871e-01 1.19342655e-01 2.39606366e-01 4.37472194e-01 -7.33075917e-01 -3.20481509e-01 2.09709436e-01]

[7.449542999267578, 8.038731575012207]

68997de3-8590-4679-acda-75e1f770aaf0

sgem-test-time-adaptation-for-automatic

2306.01981

https://arxiv.org/abs/2306.01981v4

https://arxiv.org/pdf/2306.01981v4.pdf

SGEM: Test-Time Adaptation for Automatic Speech Recognition via Sequential-Level Generalized Entropy Minimization

Automatic speech recognition (ASR) models are frequently exposed to data distribution shifts in many real-world scenarios, leading to erroneous predictions. To tackle this issue, an existing test-time adaptation (TTA) method has recently been proposed to adapt the pre-trained ASR model on unlabeled test instances without source data. Despite decent performance gain, this work relies solely on naive greedy decoding and performs adaptation across timesteps at a frame level, which may not be optimal given the sequential nature of the model output. Motivated by this, we propose a novel TTA framework, dubbed SGEM, for general ASR models. To treat the sequential output, SGEM first exploits beam search to explore candidate output logits and selects the most plausible one. Then, it utilizes generalized entropy minimization and negative sampling as unsupervised objectives to adapt the model. SGEM achieves state-of-the-art performance for three mainstream ASR models under various domain shifts.

['Eunho Yang', 'Hajin Shim', 'Joonhyung Park', 'Changhun Kim']

2023-06-03

['automatic-speech-recognition']

['speech']

[ 5.61364710e-01 -8.94003361e-02 -1.80103078e-01 -4.64760661e-01 -1.10414183e+00 -3.97108406e-01 5.24050236e-01 -2.49681219e-01 -3.90301883e-01 8.91672611e-01 1.24292485e-01 -5.10105968e-01 1.60601035e-01 -1.45102143e-01 -6.16361916e-01 -6.35799229e-01 3.96432519e-01 5.47309697e-01 3.04681689e-01 -1.11576907e-01 2.44957373e-01 3.78585935e-01 -1.33387673e+00 1.80193454e-01 9.16597664e-01 8.63851845e-01 4.68077004e-01 8.31107974e-01 -1.11323543e-01 6.50099277e-01 -9.18264925e-01 -5.37968934e-01 1.11988164e-01 -7.34480917e-01 -5.71862996e-01 2.49098524e-01 -6.46873862e-02 -6.63888780e-03 -5.74486434e-01 1.02999878e+00 7.95687199e-01 6.23919249e-01 5.80199540e-01 -6.90245926e-01 -6.39200747e-01 7.25816965e-01 -3.65482539e-01 6.26193166e-01 2.27276251e-01 2.73292899e-01 6.99444056e-01 -1.25380874e+00 2.70152628e-01 1.27057981e+00 3.38728309e-01 9.48134959e-01 -1.19872117e+00 -7.07052231e-01 5.35572708e-01 4.36615944e-01 -1.43015289e+00 -9.72342551e-01 9.49860990e-01 -6.02961658e-03 1.05777526e+00 3.03879708e-01 4.43991244e-01 1.46643209e+00 -1.07658170e-01 1.08534563e+00 1.15126634e+00 -4.94371474e-01 7.26836681e-01 1.76918413e-02 -2.98922639e-02 1.63788557e-01 -2.71315396e-01 2.48043984e-01 -9.39574540e-01 -3.22668217e-02 2.85811245e-01 -2.97005475e-01 -3.79692048e-01 3.94885428e-02 -1.05399561e+00 5.85547745e-01 -1.80364937e-01 3.22741643e-02 -4.70597029e-01 -2.31395617e-01 3.22047621e-01 4.34686840e-01 7.24117160e-01 2.51738250e-01 -8.44078183e-01 -5.67871749e-01 -1.13747489e+00 9.47818086e-02 5.61302364e-01 8.35321546e-01 4.71116841e-01 4.82626617e-01 -3.82661372e-01 1.30560839e+00 5.04157007e-01 4.26190794e-01 9.32691991e-01 -4.89466757e-01 6.55919969e-01 1.54164761e-01 -5.85428253e-02 -3.53388011e-01 6.53546900e-02 -7.38864243e-01 -7.14873314e-01 -1.65933445e-01 3.85500565e-02 2.86154393e-02 -1.43270016e+00 1.59471059e+00 4.70991701e-01 3.80352527e-01 3.32862258e-01 8.57551634e-01 4.58404154e-01 6.77066922e-01 1.42561868e-01 -6.83055997e-01 7.68128872e-01 -1.05384350e+00 -8.52010250e-01 -5.32116413e-01 4.79716867e-01 -7.73152947e-01 1.17350030e+00 3.96725595e-01 -1.07516515e+00 -4.68158454e-01 -9.45523739e-01 4.16720390e-01 -1.27960980e-01 1.43870845e-01 -1.09931313e-01 6.11929059e-01 -9.77863550e-01 2.61464059e-01 -8.53201509e-01 -3.45013589e-01 3.44326109e-01 2.45343149e-01 9.17190164e-02 -2.97259539e-01 -1.24382222e+00 8.23648632e-01 4.91046578e-01 2.03999877e-01 -1.07521868e+00 -4.75032508e-01 -6.20629191e-01 -5.52594922e-02 6.47452652e-01 -3.05377752e-01 1.79024971e+00 -1.12934172e+00 -2.27870226e+00 3.90426040e-01 -5.69670975e-01 -6.95378363e-01 6.20311201e-01 -1.98446080e-01 -7.56005943e-01 -1.52041107e-01 -4.93585706e-01 3.07579905e-01 1.21175992e+00 -1.07379842e+00 -4.24851924e-01 -4.03007656e-01 -4.79170173e-01 4.70040292e-01 -3.59022349e-01 9.74930674e-02 -3.47572058e-01 -1.09871876e+00 3.45953435e-01 -8.68874371e-01 -3.64357561e-01 -7.94477165e-01 -3.63324106e-01 -3.26404899e-01 8.00747871e-01 -7.45345175e-01 1.83611739e+00 -2.15079594e+00 1.35153905e-01 1.07852481e-01 -2.76637375e-01 6.83906674e-01 -1.91602200e-01 1.79700747e-01 3.80826853e-02 2.80553959e-02 -4.32291806e-01 -4.47827250e-01 -7.07077608e-02 1.99289829e-01 -6.29459620e-01 2.43439153e-01 1.87080339e-01 6.12187445e-01 -7.74170101e-01 -2.55153447e-01 1.92777053e-01 2.93132365e-01 -5.49531758e-01 5.46651244e-01 -4.25721914e-01 6.69414699e-01 -5.15943408e-01 7.41543293e-01 5.68749547e-01 -8.93690512e-02 6.75845966e-02 1.14468761e-01 9.31483880e-02 6.31622255e-01 -8.62161279e-01 1.50621212e+00 -3.89224470e-01 4.06431198e-01 -3.03612709e-01 -1.18162239e+00 1.22105253e+00 5.17703891e-01 2.45554969e-01 -6.87167645e-01 -1.63593605e-01 4.17269260e-01 1.26856908e-01 -2.80361384e-01 4.78577912e-01 -1.00178033e-01 1.57959059e-01 1.51300326e-01 4.93723117e-02 -8.72484408e-03 -2.91687757e-01 -1.07982121e-01 1.21005023e+00 2.19791114e-01 5.31179070e-01 7.39075840e-02 6.44294202e-01 -1.83075428e-01 7.09667146e-01 8.17877591e-01 -4.42318916e-01 7.00039804e-01 -2.14906439e-01 -2.68394589e-01 -9.17151749e-01 -1.14433968e+00 -5.74044548e-02 1.09361064e+00 -3.73273194e-02 -1.07903853e-01 -6.36839688e-01 -9.33367550e-01 -2.94508934e-01 1.11624646e+00 -1.86212942e-01 -4.35896039e-01 -6.73486173e-01 -8.21538329e-01 3.88312787e-01 2.27514789e-01 4.16603833e-01 -1.18290353e+00 -2.53319740e-01 5.45149922e-01 -1.13804080e-01 -1.08669329e+00 -6.38375938e-01 4.40134555e-01 -9.03608084e-01 -4.57506090e-01 -9.45367754e-01 -4.59921777e-01 5.87014735e-01 1.45785034e-01 1.04178238e+00 -2.04491153e-01 2.61748821e-01 1.91102952e-01 -8.19714487e-01 -2.25402221e-01 -8.45910251e-01 3.97919267e-01 2.77680904e-01 3.08081150e-01 5.74098527e-01 -3.63144517e-01 -4.94192004e-01 5.21919727e-01 -8.11324835e-01 -1.02370828e-01 6.29539907e-01 1.03859496e+00 8.57133925e-01 -2.02229023e-01 1.18267751e+00 -8.22318912e-01 6.13872588e-01 -5.65495491e-01 -3.65774602e-01 6.27535403e-01 -8.23386848e-01 5.43750599e-02 8.56998682e-01 -8.84086668e-01 -1.30294204e+00 2.84677688e-02 -3.24687988e-01 -7.31085420e-01 -2.10954040e-01 5.22790015e-01 -2.21817330e-01 2.25114375e-01 6.29150450e-01 7.69787908e-01 -4.30649370e-01 -6.10278726e-01 1.28671274e-01 1.09007192e+00 3.78399521e-01 -5.19245982e-01 6.64784193e-01 -1.59948945e-01 -5.86822033e-01 -9.81386006e-01 -9.37328935e-01 -3.96818399e-01 -4.36844856e-01 -2.59858787e-01 4.12175626e-01 -8.19183588e-01 4.11000326e-02 4.33691382e-01 -1.03644919e+00 -5.99754155e-01 -2.10958093e-01 6.21700525e-01 -7.13471234e-01 2.77533025e-01 -2.07886145e-01 -1.17879391e+00 -3.28115970e-01 -1.24669504e+00 8.33202541e-01 1.63577735e-01 -2.65859246e-01 -7.07756877e-01 2.13175137e-02 3.46854657e-01 5.24383724e-01 -5.80195189e-01 6.97614133e-01 -1.22340310e+00 -5.45906484e-01 -7.14306235e-02 2.98238099e-01 4.33064431e-01 1.02556050e-01 -1.45154417e-01 -1.08142734e+00 -5.82233667e-01 1.24489821e-01 -2.99350709e-01 7.52760649e-01 4.18832481e-01 1.45435166e+00 -4.64108765e-01 -2.67690420e-01 5.06504655e-01 8.70680630e-01 7.53726065e-01 6.07904375e-01 3.14245641e-01 3.77908498e-01 2.70362824e-01 8.23609352e-01 5.97570598e-01 1.03401937e-01 9.16873038e-01 -1.46409636e-02 2.96860009e-01 -8.25506672e-02 -4.55682874e-01 7.94848979e-01 1.35734868e+00 3.81471872e-01 -7.78831542e-01 -9.24171150e-01 5.64706147e-01 -1.73984158e+00 -8.24095011e-01 5.78370810e-01 2.38539171e+00 1.00559545e+00 2.43961990e-01 3.22420672e-02 8.15923512e-02 8.64297807e-01 2.87931979e-01 -1.07539868e+00 -3.36827189e-01 -1.95156604e-01 1.82032496e-01 2.41225764e-01 3.38370144e-01 -7.12160885e-01 1.20341766e+00 6.45787239e+00 1.25678265e+00 -1.17180634e+00 3.24980855e-01 8.44691277e-01 -2.14587376e-01 -2.90458113e-01 -1.10542670e-01 -9.54029918e-01 6.85418248e-01 1.35058367e+00 -2.39625454e-01 7.24007487e-01 7.06969738e-01 4.71735001e-01 2.53104776e-01 -9.64214683e-01 1.01029956e+00 8.77576619e-02 -8.45029235e-01 1.86642349e-01 -1.80568695e-01 6.81611061e-01 1.26025304e-01 2.80994564e-01 5.70163786e-01 4.11537766e-01 -8.54380250e-01 7.40021884e-01 2.21904695e-01 8.65893602e-01 -7.64249027e-01 4.24011201e-01 3.96409988e-01 -1.03379011e+00 -2.09229261e-01 -2.55350530e-01 3.89593214e-01 2.80237734e-01 4.42875832e-01 -1.18560016e+00 2.12755948e-01 5.20241797e-01 5.42769492e-01 -4.19205159e-01 1.01637459e+00 -1.50983185e-01 1.34457600e+00 -2.17150107e-01 3.78847048e-02 2.92963460e-02 6.82960972e-02 9.68690574e-01 1.28521872e+00 6.19996250e-01 9.57852080e-02 2.16473043e-01 4.67258841e-01 -1.13872111e-01 1.97402034e-02 -4.77466732e-01 -1.13915205e-01 9.98408318e-01 5.80126941e-01 -4.17656928e-01 -3.00619453e-01 -2.86958605e-01 1.17005193e+00 3.21895689e-01 7.82881796e-01 -7.63682604e-01 1.40750119e-02 6.36898041e-01 -1.03809277e-03 4.03880537e-01 -1.96226045e-01 -1.98606864e-01 -1.21194100e+00 1.01768315e-01 -1.23810005e+00 3.96789759e-01 -5.52559912e-01 -1.23214853e+00 8.13297629e-01 -1.60544693e-01 -1.35718846e+00 -4.36166525e-01 -7.55831301e-02 -3.43819112e-01 7.11648583e-01 -1.68704140e+00 -7.06336141e-01 2.93574184e-01 5.35601854e-01 1.53707111e+00 -6.76999450e-01 5.38199604e-01 1.59478515e-01 -7.58855581e-01 1.12506700e+00 2.93759942e-01 -3.22027922e-01 6.53344393e-01 -9.34656858e-01 6.97016239e-01 1.11484289e+00 1.94268182e-01 4.21266556e-01 7.22866237e-01 -6.61600292e-01 -1.17020106e+00 -1.28987777e+00 8.65572572e-01 -2.84065604e-01 5.22035718e-01 -1.74675778e-01 -1.24747753e+00 5.60642362e-01 -7.81622231e-02 -4.67014350e-02 5.62497258e-01 -2.63353109e-01 -1.88371211e-01 -1.01263449e-01 -9.31375384e-01 7.35690832e-01 1.20252156e+00 -6.27389193e-01 -6.72102034e-01 1.09371737e-01 9.88752067e-01 -5.48346996e-01 -6.60215259e-01 5.86721301e-01 2.22796604e-01 -5.43449342e-01 7.90811539e-01 -5.92143655e-01 1.50526091e-02 -3.36673647e-01 -4.81618732e-01 -1.59814835e+00 -1.44507006e-01 -9.70671177e-01 -7.66009688e-01 1.27565527e+00 5.99645257e-01 -5.83343208e-01 5.97094715e-01 5.22613466e-01 -4.88936871e-01 -8.71056676e-01 -1.36313677e+00 -8.87743771e-01 -2.78015643e-01 -6.79252267e-01 6.48824692e-01 6.71107411e-01 -2.96931148e-01 2.20259160e-01 -3.69285941e-01 3.89588237e-01 4.30452734e-01 -2.95319676e-01 5.25190115e-01 -8.10511112e-01 -4.09933805e-01 -3.43974859e-01 -1.66949153e-01 -1.55140281e+00 3.63960177e-01 -7.27020562e-01 4.36189055e-01 -1.01078176e+00 -1.91001043e-01 -4.03442740e-01 -6.73410892e-01 2.10040987e-01 -3.68115038e-01 -1.26106635e-01 1.16069861e-01 3.92154157e-01 -7.27645636e-01 9.56247151e-01 9.34899628e-01 -9.28572044e-02 -5.18930495e-01 4.68877643e-01 -4.49027061e-01 4.92783785e-01 8.28184783e-01 -6.28499985e-01 -5.64566791e-01 -4.14689124e-01 -1.32140741e-01 2.65318394e-01 -2.41175443e-02 -9.30494368e-01 2.41714224e-01 -4.69005793e-01 1.12571873e-01 -7.07720339e-01 2.57638931e-01 -7.16831505e-01 1.38547212e-01 1.69403329e-01 -7.80234933e-01 1.27841324e-01 2.74560321e-02 8.81539643e-01 -3.61834973e-01 -2.62225121e-01 1.09784138e+00 2.04763010e-01 -7.82860816e-01 3.61095458e-01 -4.50072795e-01 1.23837389e-01 7.75935173e-01 -3.18085104e-01 -6.27920553e-02 -2.73161471e-01 -6.69432104e-01 5.26973456e-02 2.31721595e-01 6.05193377e-01 8.60200167e-01 -1.23150671e+00 -7.47397244e-01 3.04437965e-01 1.70014530e-01 1.39597252e-01 3.79531294e-01 4.57904369e-01 1.46495655e-01 2.15979919e-01 6.04002833e-01 -5.42117894e-01 -1.19832516e+00 4.65201885e-01 3.42331558e-01 -3.81848574e-01 -4.88174319e-01 8.80599558e-01 8.65702629e-02 -3.91548991e-01 3.91494364e-01 -7.40024116e-05 -1.22928567e-01 -2.37040222e-01 4.28958595e-01 2.86341578e-01 3.70068967e-01 -5.77777147e-01 -2.54952669e-01 2.33963415e-01 -3.63652229e-01 -2.92218626e-01 1.11525655e+00 -5.10250747e-01 6.55517399e-01 6.59110725e-01 9.42805707e-01 -2.47630700e-01 -1.62965560e+00 -5.98519146e-01 1.86400279e-01 -5.28449357e-01 7.63805583e-02 -1.08361030e+00 -7.57297158e-01 8.06576729e-01 6.17339671e-01 6.16461113e-02 1.24242306e+00 -1.04148611e-01 9.47387874e-01 4.65690523e-01 3.17954183e-01 -1.43366838e+00 1.07029654e-01 7.93071866e-01 7.34826148e-01 -1.16616857e+00 -3.35320115e-01 8.97221565e-02 -7.45584607e-01 9.03203785e-01 7.71815538e-01 4.38662738e-01 4.77288514e-01 6.26667216e-02 8.26341808e-02 4.69066650e-01 -1.08191502e+00 -1.16470300e-01 2.06077293e-01 6.11681283e-01 1.95403427e-01 8.19745809e-02 -2.39670426e-01 4.81878638e-01 -1.38141006e-01 9.06994753e-03 2.08311915e-01 7.20947266e-01 -4.12146330e-01 -9.82570112e-01 -2.76503563e-01 4.17280525e-01 -4.94703710e-01 -1.62952110e-01 -1.96329013e-01 1.88626602e-01 -4.07624274e-01 1.02546299e+00 -1.10612698e-01 -5.61095774e-01 3.25274587e-01 3.33264589e-01 1.43080398e-01 -5.90742290e-01 -3.72523874e-01 2.24703938e-01 -1.16526768e-01 -1.16984174e-01 -2.78650641e-01 -8.09106171e-01 -9.19235468e-01 2.15126336e-01 -5.01855791e-01 1.41398802e-01 6.31325543e-01 1.15185583e+00 5.70089459e-01 5.65430403e-01 1.03593361e+00 -6.35944903e-01 -1.08731675e+00 -1.18006194e+00 -2.41989687e-01 2.58275032e-01 4.11308080e-01 -5.83918273e-01 -5.28907478e-01 1.32660285e-01]

[14.447153091430664, 6.644789695739746]

10107c44-8c5b-4200-99b1-e8260a0644a9

quantum-inspired-representation-for-long-tail

https://openreview.net/forum?id=r6z-A8wyD-W

https://openreview.net/pdf?id=r6z-A8wyD-W

Quantum-inspired Representation for Long-tail Senses of Word Sense Disambiguation

Data imbalance, also known as the long-tailed distribution of data, is an important challenge for data-driven models. Due to the long tail phenomenon of word sense distribution in linguistics, it is difficult to learn accurate representations for Long-Tail Senses (LTSs) in Word Sense Disambiguation (WSD) tasks. Without data augmentation, exploring representation methods that do not rely on large sample sizes is an important means to combat the long tail. In this paper, inspired by the superposition state in quantum mechanics, a representation method in Hilbert space is proposed to reduce the dependence on large sample sizes. We theoretically prove the correctness of the method, verify its effectiveness on the WSD task, and achieve state-of-the-art performance.

['Anonymous']

2021-11-16

acl-arr-november-2021-11

['word-sense-disambiguation']

['natural-language-processing']

[ 1.21306315e-01 -8.15825909e-02 -3.85255754e-01 -3.76407862e-01 -6.38467014e-01 -1.98695138e-01 3.86312306e-01 4.95309085e-01 -5.70912123e-01 7.44256556e-01 3.27636242e-01 -4.69286978e-01 -2.01191783e-01 -7.66566217e-01 -3.45235437e-01 -7.54267097e-01 5.81231900e-02 3.21551025e-01 -1.44176528e-01 -6.12197161e-01 3.02671343e-01 1.14750467e-01 -1.64608669e+00 -1.08836219e-01 1.30364501e+00 7.95406044e-01 7.21596554e-02 6.26426190e-02 -5.56353211e-01 3.73807371e-01 -7.59150445e-01 -9.05754268e-02 1.42124519e-01 -4.39993232e-01 -8.06509674e-01 -3.54691267e-01 2.35489324e-01 2.08447158e-01 -4.24126625e-01 1.42534196e+00 8.91150475e-01 2.89156765e-01 7.06666768e-01 -1.42622280e+00 -7.98841417e-01 7.67438889e-01 -8.25898111e-01 5.31904697e-01 3.60690318e-02 -3.20828974e-01 1.32060802e+00 -5.75237274e-01 3.72716069e-01 1.38160920e+00 5.41031718e-01 6.83927596e-01 -1.25761962e+00 -1.01767302e+00 2.05005124e-01 5.06839037e-01 -1.52970338e+00 -7.52032548e-02 7.27114201e-01 -2.39176318e-01 8.76922011e-01 1.02670319e-01 3.74290526e-01 1.10065448e+00 1.94765747e-01 1.03864634e+00 1.11185598e+00 -4.29815024e-01 5.34380853e-01 -2.44417563e-01 5.82462847e-01 2.02935711e-01 8.95306706e-01 -4.29159366e-02 -7.85953403e-01 -3.55423987e-01 2.84332275e-01 5.45796473e-03 -2.81967163e-01 -3.01619411e-01 -8.49119365e-01 9.92512882e-01 3.84513259e-01 3.76793146e-01 -2.15242803e-01 6.02257475e-02 5.26238501e-01 3.92967820e-01 7.91701734e-01 5.42133212e-01 -6.18468106e-01 -2.70155102e-01 -8.74547482e-01 4.10268515e-01 6.59454167e-01 6.17200136e-01 7.44522095e-01 -5.91198690e-02 -1.97029486e-01 8.33505094e-01 1.17047504e-01 7.53673732e-01 1.01427484e+00 -1.59295067e-01 3.89345795e-01 7.56608963e-01 -2.78146379e-02 -9.07727361e-01 -6.57178223e-01 -7.25165009e-01 -1.35988951e+00 -2.32277021e-01 1.82005510e-01 1.85516641e-01 -7.95777500e-01 2.11770129e+00 4.00156796e-01 2.36024067e-01 -3.52399908e-02 9.40282881e-01 4.97176021e-01 2.86440879e-01 2.94287860e-01 -3.82163405e-01 1.32028663e+00 -2.11546063e-01 -9.44052160e-01 -4.13217247e-01 1.01195240e+00 -4.25001234e-01 1.36476719e+00 3.82440776e-01 -4.73634422e-01 -2.82005548e-01 -1.15685189e+00 -3.49957854e-01 -3.02515924e-01 -3.70194852e-01 9.19323444e-01 6.83310986e-01 -6.08251572e-01 5.25533080e-01 -8.32033098e-01 -2.16453776e-01 6.05083227e-01 1.83792323e-01 -2.09626749e-01 -1.79916069e-01 -1.62831545e+00 8.25203359e-01 5.44014156e-01 -4.63871807e-02 -1.53385252e-01 -8.93349171e-01 -8.31481099e-01 1.75621867e-01 2.85315901e-01 -6.44117236e-01 9.75002468e-01 -4.48404878e-01 -9.68173623e-01 9.47884560e-01 -2.11601883e-01 -7.27435887e-01 1.44974634e-01 -4.81390089e-01 -4.31053877e-01 -4.52711999e-01 1.90050080e-01 3.59643251e-02 7.26606548e-01 -7.80476928e-01 -2.13846743e-01 -6.55685961e-01 -1.36968404e-01 5.26097603e-02 -6.07249975e-01 -5.21679938e-01 2.23910600e-01 -6.80049837e-01 3.35819483e-01 -8.57644260e-01 -4.14728671e-01 -4.83095080e-01 -5.35321534e-01 -7.24806368e-01 3.09228539e-01 -2.33162731e-01 1.50061786e+00 -2.36879444e+00 -1.45506173e-01 1.89898133e-01 3.81761819e-01 4.33092743e-01 -3.73369396e-01 4.32427198e-01 -3.44011813e-01 1.50176644e-01 -2.65176624e-01 -2.14775741e-01 2.84997761e-01 2.94340849e-01 -6.31751835e-01 6.06758595e-01 1.09750129e-01 6.68945312e-01 -1.07576096e+00 -1.97809756e-01 -8.58009160e-02 1.21216767e-01 -4.87288505e-01 1.61268443e-01 -1.81539491e-01 1.91594049e-01 -5.05911648e-01 1.76982403e-01 1.02244794e+00 -3.59497309e-01 1.66354284e-01 -2.26326540e-01 2.07105532e-01 7.63334930e-01 -1.32312047e+00 1.89839303e+00 -5.20020783e-01 3.37669283e-01 -3.20522696e-01 -1.27713311e+00 9.90742385e-01 -1.74643788e-02 4.48832035e-01 -9.96846676e-01 5.71135767e-02 3.72101188e-01 1.28044546e-01 -3.28435153e-01 7.01016724e-01 -6.45412445e-01 -2.59982586e-01 6.43362999e-01 -8.53037229e-04 -1.96781561e-01 4.16398913e-01 1.90666467e-01 1.09689105e+00 -3.32463861e-01 4.93511230e-01 -5.15081763e-01 6.69702888e-02 -3.22119832e-01 9.13074493e-01 6.40002429e-01 -7.68203288e-02 4.78573829e-01 6.47114813e-01 -5.22937238e-01 -6.87183797e-01 -8.18098366e-01 -2.87560225e-01 1.14289200e+00 3.63311857e-01 -8.24212193e-01 -5.64624310e-01 -5.37472129e-01 1.36121184e-01 9.63627100e-01 -5.84036767e-01 -8.25228631e-01 -1.78307950e-01 -1.42211294e+00 4.62563634e-01 4.13077593e-01 2.38579482e-01 -6.81127429e-01 -5.16977489e-01 7.69730508e-02 -4.48057562e-01 -9.14934993e-01 -2.63324171e-01 4.32156593e-01 -8.51505697e-01 -1.10160518e+00 -4.82138634e-01 -1.81310251e-01 4.63287145e-01 5.07398605e-01 1.38785863e+00 1.63097754e-02 -3.37088138e-01 -1.53969854e-01 -5.63331842e-01 -8.53028238e-01 -2.02728316e-01 3.61782134e-01 5.22542596e-01 -2.61306733e-01 9.17867482e-01 -6.30631804e-01 -5.57406723e-01 -9.22802463e-02 -1.37009573e+00 -3.36271197e-01 5.80558956e-01 1.02683246e+00 5.67222118e-01 1.76607773e-01 6.27927959e-01 -1.09114051e+00 7.51520097e-01 -6.15538836e-01 -4.52836484e-01 5.26827574e-02 -8.86622965e-01 5.72841465e-01 3.41985852e-01 -3.82792950e-01 -5.50715864e-01 -4.17920738e-01 -8.22989196e-02 -1.77122042e-01 9.51765329e-02 7.45174348e-01 -1.88577026e-01 5.01192689e-01 8.96989405e-01 4.75584343e-02 -2.80681960e-02 -6.05486870e-01 4.56543475e-01 7.56780267e-01 1.46056339e-01 -4.70788717e-01 7.25175142e-01 2.87796021e-01 2.61923149e-02 -9.25859451e-01 -1.63979721e+00 -7.13234127e-01 -5.00286222e-01 5.33422112e-01 5.61809301e-01 -9.80537891e-01 -6.08106375e-01 3.72292131e-01 -9.12157476e-01 3.64638269e-02 -4.70955461e-01 5.93028188e-01 -4.01922047e-01 3.26230645e-01 -1.34299442e-01 -1.12313163e+00 -5.40615439e-01 -5.39425433e-01 1.00849342e+00 2.04714999e-01 -3.59930545e-01 -8.09620082e-01 3.13132793e-01 2.10663453e-01 5.15152812e-01 -6.12358563e-02 1.20945835e+00 -1.04090607e+00 -3.17891911e-02 -8.24377313e-02 -2.83838570e-01 3.05182248e-01 2.03855351e-01 -5.21040738e-01 -1.02932370e+00 -4.96474713e-01 2.62080520e-01 -4.12036330e-01 1.21568930e+00 3.07505995e-01 1.39849889e+00 -9.52065513e-02 -4.19176035e-02 3.75056297e-01 1.29044890e+00 -5.82264364e-01 5.68078041e-01 2.04563335e-01 6.30652189e-01 3.29348117e-01 7.83948481e-01 7.48191357e-01 5.77719629e-01 3.49605381e-01 4.87203926e-01 9.47466046e-02 2.71868035e-02 -3.33802521e-01 1.65695235e-01 1.06208348e+00 1.75152481e-01 -2.00358719e-01 -9.69643712e-01 5.90629935e-01 -1.75556445e+00 -6.75782084e-01 -2.65063107e-01 2.48666835e+00 1.09099221e+00 6.59706518e-02 -1.98458191e-02 4.98623967e-01 6.35263026e-01 3.83534849e-01 -6.86838746e-01 -2.04568803e-01 -3.43756080e-01 4.25851077e-01 5.78239918e-01 1.64140955e-01 -9.61479425e-01 1.00401211e+00 5.65626764e+00 1.11812568e+00 -9.74730313e-01 9.16145146e-02 5.09173989e-01 1.47048950e-01 -7.47724354e-01 -1.54151186e-01 -6.28533304e-01 5.02850354e-01 8.15119028e-01 -4.27856863e-01 2.17646465e-01 6.45654082e-01 -7.98853934e-02 -9.04461220e-02 -1.03295696e+00 1.33235860e+00 -2.88900416e-02 -9.28266108e-01 4.53208014e-02 1.33051857e-01 7.51922131e-01 1.08265087e-01 -7.95915723e-02 5.54019153e-01 2.60497242e-01 -1.20199680e+00 3.01981598e-01 2.19674140e-01 8.89839649e-01 -9.01689768e-01 9.47618008e-01 4.85908061e-01 -8.21639836e-01 -8.50373358e-02 -8.95005584e-01 -2.11102903e-01 -9.85018685e-02 1.55882895e+00 -5.89213252e-01 6.29233718e-01 5.05587876e-01 6.34398997e-01 -4.29289371e-01 8.84777367e-01 -2.10276529e-01 8.56684625e-01 -3.53874624e-01 -8.37173462e-02 -2.75422424e-01 -2.02466592e-01 4.54271317e-01 8.21791887e-01 3.41846496e-01 6.37584552e-02 5.12777902e-02 7.40451515e-01 -2.12530226e-01 9.56232920e-02 -6.51925743e-01 -4.10599738e-01 6.64129198e-01 8.79739702e-01 -3.52373630e-01 -8.80565122e-02 -1.76243305e-01 8.02268803e-01 4.34994578e-01 3.25878143e-01 -2.76872456e-01 -2.65606105e-01 8.79662037e-01 6.51775440e-03 8.86537433e-02 -2.82308310e-01 -4.83208150e-01 -1.41296422e+00 1.64305016e-01 -9.04261291e-01 6.22946382e-01 -4.38835770e-02 -1.87081325e+00 2.95953572e-01 -2.91054964e-01 -1.30651796e+00 -1.82952121e-01 -4.81277019e-01 -3.45000148e-01 7.59652555e-01 -1.89154291e+00 -7.33007789e-01 -3.04215532e-02 4.86229956e-01 1.31327584e-01 1.56484142e-01 1.13754618e+00 3.58014584e-01 -4.56673175e-01 8.49557459e-01 2.22472683e-01 -6.48756772e-02 7.44153738e-01 -1.45945585e+00 4.15208399e-01 4.19692606e-01 3.78186315e-01 7.83676624e-01 9.29441333e-01 -6.02408648e-01 -1.39922452e+00 -8.15525591e-01 9.83654797e-01 -1.68116033e-01 7.79817402e-01 -3.66648614e-01 -1.18753302e+00 1.83123589e-01 -3.16194236e-01 2.35124394e-01 1.09150636e+00 7.43538439e-01 -7.60475338e-01 -2.29992211e-01 -9.25449610e-01 3.09215993e-01 1.13560987e+00 -6.46741271e-01 -9.45802212e-01 3.09725314e-01 6.21255457e-01 -3.14261675e-01 -4.58044738e-01 3.45026940e-01 1.56515747e-01 -8.48273873e-01 8.37460637e-01 -9.64092731e-01 9.02275443e-02 -8.51041302e-02 -3.15470308e-01 -1.74708021e+00 -1.08935781e-01 -5.54445565e-01 -2.20524773e-01 1.05008554e+00 1.93735227e-01 -6.48755133e-01 6.48638010e-01 5.02694428e-01 2.47721821e-01 -5.82476079e-01 -1.30282998e+00 -8.82881999e-01 4.12325978e-01 -5.33893824e-01 8.36220741e-01 1.10836387e+00 1.36772692e-01 6.52929544e-01 -3.36680114e-01 1.34424657e-01 5.83335936e-01 4.57157373e-01 4.83856440e-01 -1.68503654e+00 -2.79881936e-02 -2.90739805e-01 -7.54946291e-01 -1.08262897e+00 3.88990015e-01 -1.02124524e+00 6.82570487e-02 -1.31606162e+00 3.80995631e-01 -5.78187227e-01 -9.64460790e-01 3.70211869e-01 -7.69047916e-01 -6.39098808e-02 1.02543257e-01 3.83778811e-02 -7.71800995e-01 1.14502728e+00 1.02543437e+00 -9.02185962e-02 7.00091273e-02 3.69606279e-02 -1.03487694e+00 6.97489679e-01 6.68692291e-01 -7.42303967e-01 -5.51996350e-01 -2.86752969e-01 8.19658697e-01 -2.63472110e-01 8.33754539e-02 -7.94094265e-01 -8.15374311e-03 -1.00010939e-01 -1.01017848e-01 -5.21429956e-01 9.88794267e-02 -6.07581675e-01 -6.24352038e-01 5.29537082e-01 -4.50162977e-01 5.75126112e-02 -1.05229899e-01 7.66977549e-01 -3.23216408e-01 -6.72726780e-02 4.69283134e-01 7.16834888e-02 -3.95555675e-01 3.83735150e-01 3.14119719e-02 7.12541103e-01 5.38349271e-01 4.72508103e-01 -3.13061893e-01 -2.21258953e-01 -1.94105491e-01 3.57044995e-01 2.37182245e-01 6.50752544e-01 4.10137922e-01 -1.42591524e+00 -7.98658788e-01 2.60759473e-01 3.87725592e-01 3.12785000e-01 4.74496096e-01 7.29614735e-01 1.30109817e-01 1.27709880e-01 1.22641467e-01 -3.88589561e-01 -7.31316745e-01 5.54224610e-01 -2.25278805e-03 -4.52416480e-01 -3.73924822e-01 8.91106188e-01 2.94992924e-01 -6.57072484e-01 -4.64536482e-03 -2.55683333e-01 -1.77016526e-01 4.02623475e-01 9.37080443e-01 3.42897177e-01 4.28864628e-01 -2.72282839e-01 -4.74283874e-01 1.76801026e-01 -7.63710588e-02 2.39990696e-01 1.46630800e+00 -1.76539198e-01 -2.14889780e-01 9.64583218e-01 1.10725677e+00 -6.20647892e-02 -7.11850822e-01 -5.63163221e-01 3.43370199e-01 -5.21475971e-01 1.23723634e-01 -4.90318090e-01 -8.78306925e-01 1.14830554e+00 8.35160732e-01 4.66882169e-01 9.74588335e-01 -6.43506050e-02 1.18916249e+00 3.97555888e-01 5.71152270e-01 -1.33711207e+00 -4.78106104e-02 7.64401138e-01 6.81552589e-01 -1.26932824e+00 7.20731448e-04 -2.56625891e-01 -5.77933490e-01 7.63505340e-01 3.72251660e-01 -9.83140096e-02 8.10424089e-01 9.60734859e-02 1.07121602e-01 -2.27949709e-01 -7.19910502e-01 -4.67743427e-01 2.68778473e-01 6.09974265e-01 4.92819995e-01 2.88231671e-01 -6.32948697e-01 7.54286766e-01 -4.56758320e-01 -9.90748256e-02 4.31841105e-01 6.41122997e-01 -2.88101166e-01 -1.29987240e+00 -4.88931499e-02 7.22713590e-01 -5.86672664e-01 -3.73601496e-01 -1.52235776e-01 4.11294401e-01 -3.46368179e-02 1.08273351e+00 -7.79803377e-03 -3.34360778e-01 2.57065892e-01 1.78783625e-01 2.38842472e-01 -5.78300834e-01 4.39741611e-02 -4.41554278e-01 -1.60080999e-01 -7.37344325e-01 -4.76136357e-01 -4.42939192e-01 -1.49120128e+00 -3.20470989e-01 -4.95665044e-01 3.18457156e-01 6.97583139e-01 1.34854305e+00 6.70783818e-01 4.12063658e-01 6.46034598e-01 -3.15174848e-01 -1.32438087e+00 -1.25267994e+00 -1.08321154e+00 9.11916375e-01 5.27142286e-01 -8.46872926e-01 -4.66590822e-01 -5.80715954e-01]

[10.268644332885742, 8.900481224060059]

5de06204-dfab-4ff7-b1b1-b4dae5859585

regression-and-classification-for-direction

1904.08452

https://arxiv.org/abs/1904.08452v3

https://arxiv.org/pdf/1904.08452v3.pdf

Regression and Classification for Direction-of-Arrival Estimation with Convolutional Recurrent Neural Networks

We present a novel learning-based approach to estimate the direction-of-arrival (DOA) of a sound source using a convolutional recurrent neural network (CRNN) trained via regression on synthetic data and Cartesian labels. We also describe an improved method to generate synthetic data to train the neural network using state-of-the-art sound propagation algorithms that model specular as well as diffuse reflections of sound. We compare our model against three other CRNNs trained using different formulations of the same problem: classification on categorical labels, and regression on spherical coordinate labels. In practice, our model achieves up to 43% decrease in angular error over prior methods. The use of diffuse reflection results in 34% and 41% reduction in angular prediction errors on LOCATA and SOFA datasets, respectively, over prior methods based on image-source methods. Our method results in an additional 3% error reduction over prior schemes that use classification based networks, and we use 36% fewer network parameters.

['Kevin Hogan', 'Dinesh Manocha', 'John D. Kanu', 'Zhenyu Tang']

2019-04-17

['direction-of-arrival-estimation']

['audio']

[ 3.35493237e-01 1.17691346e-01 8.72305691e-01 -5.66003799e-01 -1.35367203e+00 -4.30723786e-01 5.87461770e-01 -3.51515919e-01 -1.45833910e-01 5.01431644e-01 3.75254542e-01 -4.79780883e-01 1.79907352e-01 -9.89142656e-01 -9.23974574e-01 -8.91138792e-01 -8.28510746e-02 2.70718426e-01 1.07021883e-01 -5.20236455e-02 4.35878187e-02 3.87656927e-01 -1.27792251e+00 3.04056376e-01 2.17467189e-01 1.20410502e+00 -3.22514802e-01 1.11365044e+00 1.84728116e-01 1.13570476e+00 -8.71397555e-01 1.17722347e-01 1.84424087e-01 -4.32898641e-01 -5.93793869e-01 -7.11821258e-01 7.22579777e-01 -2.64001280e-01 -3.10351223e-01 3.66637528e-01 8.14852178e-01 2.48981431e-01 8.10757577e-01 -8.26628029e-01 -4.44661617e-01 6.80754662e-01 -1.49445489e-01 4.95058633e-02 2.42659867e-01 -2.91428208e-01 7.08566248e-01 -9.76283967e-01 2.23389640e-01 9.36540186e-01 1.36831450e+00 4.12736207e-01 -1.20615900e+00 -7.62199044e-01 -4.72659111e-01 -2.54055709e-01 -1.36813843e+00 -6.78054810e-01 7.97915995e-01 -2.69998312e-01 9.14868057e-01 3.57247829e-01 3.65540594e-01 1.15486157e+00 -3.18975300e-02 2.41842061e-01 1.23756933e+00 -7.73858607e-01 3.98229748e-01 -3.08454446e-02 -1.61107451e-01 7.59685636e-01 -8.32380652e-02 1.83697715e-01 -5.58201790e-01 -2.46073678e-01 9.07038212e-01 -4.57157642e-01 -2.93291122e-01 2.25596711e-01 -1.14928150e+00 7.89925575e-01 4.49832261e-01 -7.75873140e-02 -3.36161911e-01 8.44664037e-01 -7.86648765e-02 2.82510836e-02 9.97018397e-01 6.23806179e-01 -4.77669835e-01 -9.03973803e-02 -1.29771090e+00 2.61618495e-01 1.01396775e+00 7.86886871e-01 6.64612770e-01 7.51398742e-01 8.53982717e-02 9.69757020e-01 7.96360850e-01 1.10265374e+00 8.91082585e-02 -1.30388665e+00 1.88918561e-01 -4.54668283e-01 4.30026978e-01 -1.08983314e+00 -8.41768026e-01 -8.74956369e-01 -7.01793790e-01 3.51260990e-01 4.56774771e-01 -7.39613831e-01 -1.17058444e+00 1.58550656e+00 1.35570560e-02 6.90092921e-01 2.20079228e-01 7.65612304e-01 1.05683517e+00 1.05527008e+00 -4.91186112e-01 1.23001687e-01 8.29932809e-01 -1.23159885e+00 -3.28202337e-01 4.52007540e-02 5.11059642e-01 -1.10517550e+00 5.20294666e-01 4.03755158e-01 -1.02873433e+00 -4.93537337e-01 -1.06275713e+00 2.49575198e-01 -1.31485373e-01 3.15445364e-01 5.61707616e-01 9.61867988e-01 -1.53295624e+00 4.92111593e-01 -9.89145756e-01 1.72158599e-01 -2.58620866e-02 1.44931510e-01 3.04015547e-01 1.93143636e-01 -1.00814247e+00 4.38562989e-01 -6.52254224e-01 2.72027075e-01 -1.17738414e+00 -1.12609065e+00 -7.80010164e-01 -4.91750166e-02 -4.67854023e-01 -5.03551304e-01 1.62125957e+00 -8.31139088e-01 -1.94834042e+00 2.69511968e-01 -2.03676820e-01 -5.16721964e-01 2.46589407e-01 -3.57723325e-01 -5.04181087e-01 3.00720066e-01 1.40619185e-02 5.43809116e-01 8.09407651e-01 -1.50156689e+00 -3.38543743e-01 3.71234268e-01 1.97725017e-02 -6.34615943e-02 2.73663998e-01 1.63801074e-01 5.12817316e-02 -7.67054677e-01 3.72079492e-01 -9.80921805e-01 -2.88239151e-01 1.13082957e-02 -4.16304231e-01 1.58462003e-01 3.85259390e-01 -6.70894265e-01 5.51879704e-01 -1.80720901e+00 -5.55516660e-01 5.01202345e-01 6.94147721e-02 -2.29421295e-02 -2.60000288e-01 3.00913423e-01 -1.55258626e-01 9.30625275e-02 -3.17366868e-01 -6.80628777e-01 -2.04104066e-01 -2.89317757e-01 -7.92177141e-01 3.87109041e-01 -6.08756319e-02 2.59996861e-01 -8.43089402e-01 1.10399783e-01 -8.77379104e-02 1.06632376e+00 -7.17551172e-01 2.83150047e-01 -1.15304127e-01 6.36063457e-01 -1.24230288e-01 4.25156265e-01 8.56660187e-01 -3.13088536e-01 -2.50096202e-01 -6.50917143e-02 -2.24600583e-01 7.12120295e-01 -1.15995741e+00 1.42750776e+00 -1.32125413e+00 1.15308070e+00 9.87637267e-02 -6.27479196e-01 1.19483817e+00 6.86369836e-01 3.75209004e-01 -4.17697519e-01 -1.25572130e-01 4.47744817e-01 -2.27281824e-01 -2.00078785e-01 2.92653054e-01 -1.35417521e-01 3.70327175e-01 8.88624430e-01 4.21821438e-02 -3.83042604e-01 -5.78932941e-01 -1.13976605e-01 1.07720768e+00 4.94746685e-01 -6.24781430e-01 -1.94591790e-01 9.91916955e-02 -2.02988446e-01 1.65376872e-01 1.08313835e+00 5.28403401e-01 1.25203478e+00 4.34741788e-02 -7.38322556e-01 -8.02031100e-01 -8.55475783e-01 -4.31170076e-01 1.17140841e+00 -1.27985463e-01 -2.11279660e-01 -6.99460745e-01 -2.35224545e-01 -4.11921680e-01 1.10535157e+00 -5.94816029e-01 2.12848619e-01 -8.36293519e-01 -1.15504503e+00 1.23969889e+00 4.95555878e-01 4.76373851e-01 -9.95845616e-01 -5.27494788e-01 2.44687289e-01 -2.74323910e-01 -1.05886173e+00 1.07082196e-01 1.94497719e-01 -5.25671005e-01 -5.22436798e-01 -1.02614367e+00 -5.76300442e-01 6.15632117e-01 1.55193478e-01 1.29009545e+00 -1.47506371e-01 8.70018546e-03 5.60371459e-01 -3.52888614e-01 -7.06407964e-01 -6.40953183e-01 -1.58775449e-01 8.93673524e-02 1.33118153e-01 -2.03603491e-01 -8.99475873e-01 -8.66990089e-01 3.28451991e-01 -4.22789067e-01 1.33364588e-01 9.69560817e-02 5.56039751e-01 3.05914819e-01 -3.20128232e-01 5.39388895e-01 -7.24103987e-01 3.68103504e-01 -6.21450543e-01 -5.75035691e-01 -1.75055847e-01 -5.91436446e-01 1.84980974e-01 5.60589433e-01 -4.67894338e-02 -1.40689766e+00 -5.40780053e-02 -6.42014563e-01 -1.93765730e-01 -3.53460461e-01 2.95861989e-01 8.70447099e-01 -3.93781424e-01 1.11325681e+00 1.76296845e-01 -4.00387287e-01 -4.68351424e-01 1.52731821e-01 6.78636014e-01 3.14092398e-01 -5.92797160e-01 6.03794038e-01 6.07607126e-01 3.66209596e-01 -9.19696093e-01 -1.05971384e+00 -2.75496602e-01 -3.11500132e-01 -1.44299224e-01 7.04776943e-01 -1.01325107e+00 -3.67661536e-01 7.34498858e-01 -1.50731885e+00 -8.10181916e-01 -8.51418823e-02 8.28630209e-01 -5.51828325e-01 -1.51372671e-01 -5.60353696e-01 -9.47389543e-01 -5.71082354e-01 -9.26169634e-01 1.20714998e+00 7.18215182e-02 1.20087616e-01 -1.12241852e+00 5.70245624e-01 9.03802440e-02 1.00111699e+00 2.96560913e-01 5.42998910e-01 -4.72454339e-01 -6.18514180e-01 -4.29834686e-02 -2.93104231e-01 4.83434498e-01 -2.72561349e-02 9.87642109e-02 -1.58271801e+00 -5.92304952e-03 -3.53838466e-02 -3.09423596e-01 9.23982561e-01 7.81537771e-01 9.71395791e-01 -3.86369258e-01 -7.90361613e-02 1.08423913e+00 1.32805347e+00 1.27157390e-01 5.34991443e-01 1.98693331e-02 7.81100035e-01 2.83099502e-01 3.63889001e-02 4.30698633e-01 2.70797342e-01 5.46167552e-01 4.06410635e-01 -3.79685849e-01 -5.29223561e-01 -1.23280570e-01 2.09043160e-01 8.36790442e-01 -5.44080198e-01 -3.16105574e-01 -1.12550473e+00 4.41018730e-01 -1.22826433e+00 -7.40675330e-01 -5.14503419e-01 2.00825310e+00 8.76572132e-01 -2.18263373e-01 -2.81585455e-01 4.06052656e-02 2.77155340e-01 2.46284470e-01 1.15989417e-01 -3.90954524e-01 2.12411538e-01 7.03687549e-01 5.44678330e-01 8.73318195e-01 -1.03207517e+00 7.76757419e-01 7.85680389e+00 4.03524607e-01 -1.73131025e+00 9.10030901e-02 6.76999629e-01 6.45508990e-02 -4.67690945e-01 -2.97441959e-01 -8.01552415e-01 9.68912765e-02 1.52559268e+00 7.02550113e-01 5.72873294e-01 7.20751703e-01 1.69569463e-01 8.34530443e-02 -9.15234327e-01 7.88572431e-01 3.39441329e-01 -1.52793574e+00 -2.81708986e-01 -4.63256478e-01 1.06199074e+00 6.27141953e-01 1.62995622e-01 4.44004172e-03 5.88659823e-01 -1.06987989e+00 9.30105507e-01 8.92520308e-01 1.07886839e+00 -5.88280201e-01 6.74690843e-01 5.55362999e-02 -7.97676086e-01 5.19955873e-01 -3.87201786e-01 -1.26591891e-01 2.08035782e-01 6.80701494e-01 -1.26589262e+00 3.11177522e-01 8.82112801e-01 2.67623007e-01 -2.02018425e-01 1.02411485e+00 -4.03807312e-01 1.60728264e+00 -7.54782557e-01 -1.09990031e-01 2.33283252e-01 1.31006390e-01 5.58398485e-01 1.55039179e+00 7.20977962e-01 -1.74298137e-01 -3.02740604e-01 8.28752995e-01 -7.34362230e-02 -5.83601929e-02 -4.64095771e-01 5.67277014e-01 3.68085027e-01 1.25011766e+00 -6.33651197e-01 -8.89488980e-02 -1.67091966e-01 5.63614428e-01 -2.62332708e-01 8.27683270e-01 -9.43605959e-01 -8.02085340e-01 2.82899499e-01 1.23922534e-01 2.97411382e-01 -3.10558975e-01 -4.47021186e-01 -6.49889112e-01 -3.05645257e-01 -4.73872364e-01 -3.51806164e-01 -1.27839804e+00 -8.47508430e-01 9.84638870e-01 -9.30276811e-02 -1.11397767e+00 -5.49405038e-01 -6.34792507e-01 -8.23118448e-01 1.10249233e+00 -1.68078971e+00 -1.25313759e+00 -5.38507700e-01 1.89449489e-01 2.30465651e-01 -1.97021827e-01 1.34612179e+00 1.62981123e-01 3.91367860e-02 4.97142613e-01 4.95221466e-01 2.68855929e-01 7.75052488e-01 -1.22713697e+00 6.86126530e-01 5.72353959e-01 4.80206490e-01 5.29368877e-01 8.24237525e-01 -9.50267389e-02 -8.22395682e-01 -1.31005454e+00 7.92860091e-01 -5.99652052e-01 4.16429490e-01 -3.74927700e-01 -4.28868443e-01 7.39372790e-01 2.21196279e-01 3.46788377e-01 7.13002682e-01 2.34030023e-01 -6.72575951e-01 -3.19930464e-01 -9.22163963e-01 2.72869647e-01 5.35348892e-01 -4.81137872e-01 -1.33535758e-01 5.76919198e-01 8.02183509e-01 -7.23979294e-01 -5.73293507e-01 4.71952111e-01 6.66740596e-01 -1.05681491e+00 1.02476943e+00 -9.03001577e-02 6.73574865e-01 -2.44672194e-01 -2.86371142e-01 -1.60120595e+00 -2.03007475e-01 -7.57086217e-01 1.86272398e-01 9.18407917e-01 8.43255341e-01 -6.07310355e-01 4.99441206e-01 7.50204325e-02 -5.07395029e-01 -7.13167846e-01 -8.85363162e-01 -3.94095302e-01 2.15483695e-01 -8.71944606e-01 4.43932205e-01 6.25734091e-01 -6.27857625e-01 1.69578180e-01 -3.63759428e-01 6.86573505e-01 6.33359253e-01 6.98015317e-02 6.08320892e-01 -1.10961020e+00 -3.25320452e-01 -8.09983313e-02 1.44363090e-01 -1.33917117e+00 -4.78789099e-02 -7.06649840e-01 6.26078308e-01 -1.59370279e+00 -5.44765413e-01 -1.14834487e+00 -3.19871753e-01 5.26228547e-01 5.28663695e-01 1.04116237e+00 -3.77801389e-01 7.15255365e-03 -3.39936614e-01 2.94222742e-01 8.94242704e-01 5.16395345e-02 -2.01669037e-01 4.06472921e-01 -3.63210738e-01 1.06065226e+00 7.80285776e-01 -6.89469874e-01 -2.13535830e-01 -9.28144336e-01 8.08385193e-01 1.58051312e-01 5.83400607e-01 -1.41248918e+00 2.21214533e-01 1.52462810e-01 4.84025747e-01 -3.43359709e-01 5.29365301e-01 -3.56706858e-01 -6.53987890e-03 6.00642115e-02 -5.55566013e-01 -3.23506564e-01 1.62399054e-01 3.49299580e-01 4.41002846e-02 -1.91563353e-01 8.14689279e-01 -1.01302564e-02 3.13722007e-02 -1.99231073e-01 -6.44447148e-01 -3.95480217e-03 1.84693843e-01 2.66724527e-01 -3.55011225e-01 -1.06181669e+00 -5.61789989e-01 -6.04359925e-01 -2.82682270e-01 8.20099115e-02 4.94272918e-01 -1.08386326e+00 -1.07272470e+00 3.21132511e-01 -3.76576900e-01 1.89310864e-01 9.83785763e-02 7.29940236e-01 -1.20018232e+00 4.97539639e-01 2.35374168e-01 -5.91041207e-01 -9.85212684e-01 -3.31234097e-01 7.44370878e-01 7.27125332e-02 -4.80544895e-01 1.46059501e+00 1.41068548e-01 -8.15328717e-01 1.46341130e-01 -5.40134549e-01 -5.97308874e-02 -3.99306089e-01 3.39879811e-01 7.45736480e-01 2.15467378e-01 -5.09743333e-01 -1.88705906e-01 6.48045838e-01 4.91403520e-01 -5.69816470e-01 1.49757552e+00 1.74422711e-01 -2.81280968e-02 4.97158974e-01 1.15200961e+00 7.34932780e-01 -1.29388463e+00 -1.87334716e-01 -7.72152662e-01 1.04012132e-01 4.87930030e-01 -1.32390094e+00 -1.05068803e+00 1.06369054e+00 6.05311215e-01 3.35214704e-01 8.37675750e-01 -1.02960477e-02 5.58938563e-01 6.39111400e-01 2.79864073e-01 -6.20801389e-01 6.90036789e-02 6.75751805e-01 1.02622128e+00 -8.31352293e-01 -2.34847024e-01 -2.91533798e-01 -1.46686554e-01 1.30039465e+00 2.41382867e-01 -4.78873700e-01 1.17792916e+00 6.43272042e-01 7.22277939e-01 -6.82715774e-02 -4.45908070e-01 3.75869215e-01 2.79094189e-01 6.28760457e-01 6.60352468e-01 -2.53504738e-02 4.77194965e-01 4.81941909e-01 -6.91053748e-01 -2.81497747e-01 6.57959342e-01 4.85708654e-01 -1.93957403e-01 -5.90703011e-01 -5.93568921e-01 1.67944357e-01 -7.49307334e-01 -5.25191903e-01 2.11045384e-01 2.28449807e-01 4.19581197e-02 1.14366853e+00 3.63518983e-01 -3.47364128e-01 -1.54015943e-01 5.63980453e-02 2.10299700e-01 -4.95868027e-01 -2.95235783e-01 2.48203352e-01 4.07810211e-01 -2.90570527e-01 -6.34926736e-01 -5.14860272e-01 -1.43408990e+00 1.41141057e-01 -3.25036168e-01 2.17524126e-01 1.05991566e+00 8.08034122e-01 3.33372265e-01 8.37592542e-01 1.06509554e+00 -1.11259353e+00 -3.47239256e-01 -1.24380994e+00 -3.96580160e-01 -2.91823655e-01 8.17791104e-01 -3.22152138e-01 -8.57580185e-01 2.07349762e-01]

[15.186945915222168, 5.771124839782715]

674734a0-8a2a-40be-af7e-3ce4b1f6643a

algorithmic-probability-guided-supervised

1910.02758

https://arxiv.org/abs/1910.02758v2

https://arxiv.org/pdf/1910.02758v2.pdf

Algorithmic Probability-guided Supervised Machine Learning on Non-differentiable Spaces

We show how complexity theory can be introduced in machine learning to help bring together apparently disparate areas of current research. We show that this new approach requires less training data and is more generalizable as it shows greater resilience to random attacks. We investigate the shape of the discrete algorithmic space when performing regression or classification using a loss function parametrized by algorithmic complexity, demonstrating that the property of differentiation is not necessary to achieve results similar to those obtained using differentiable programming approaches such as deep learning. In doing so we use examples which enable the two approaches to be compared (small, given the computational power required for estimations of algorithmic complexity). We find and report that (i) machine learning can successfully be performed on a non-smooth surface using algorithmic complexity; (ii) that parameter solutions can be found using an algorithmic-probability classifier, establishing a bridge between a fundamentally discrete theory of computability and a fundamentally continuous mathematical theory of optimization methods; (iii) a formulation of an algorithmically directed search technique in non-smooth manifolds can be defined and conducted; (iv) exploitation techniques and numerical methods for algorithmic search to navigate these discrete non-differentiable spaces can be performed; in application of the (a) identification of generative rules from data observations; (b) solutions to image classification problems more resilient against pixel attacks compared to neural networks; (c) identification of equation parameters from a small data-set in the presence of noise in continuous ODE system problem, (d) classification of Boolean NK networks by (1) network topology, (2) underlying Boolean function, and (3) number of incoming edges.

['Jesper Tegnér', 'Jürgen Riedel', 'Santiago Hernández-Orozco', 'Narsis A. Kiani', 'Adam Uccello', 'Hector Zenil']

2019-10-07

['small-data']

['computer-vision']

[ 5.79138160e-01 1.78039789e-01 4.61455528e-03 5.53651899e-02 -5.43177903e-01 -7.18494952e-01 8.10738504e-01 2.73899376e-01 -4.94383782e-01 7.73061931e-01 -4.34029907e-01 -6.71761572e-01 -7.37474740e-01 -8.22122514e-01 -6.88147008e-01 -1.06886792e+00 -7.00658202e-01 2.88396508e-01 1.95292756e-01 -3.91695470e-01 5.46748877e-01 1.08670235e+00 -1.50903130e+00 -3.27697366e-01 5.49044728e-01 1.16770601e+00 -5.40390193e-01 9.46201026e-01 2.41370678e-01 5.05713582e-01 -3.61931652e-01 -1.32660821e-01 6.09167457e-01 -4.74084139e-01 -9.07467306e-01 -5.08423485e-02 2.93096751e-01 2.95655638e-01 -1.06578611e-01 1.39771032e+00 3.22674155e-01 1.41328022e-01 1.20465815e+00 -1.24927843e+00 -4.93818283e-01 1.40478566e-01 -1.33600652e-01 2.51427561e-01 1.37597054e-01 3.18111122e-01 6.77221835e-01 -3.48998159e-01 5.14194012e-01 1.11380732e+00 8.82550538e-01 4.15444404e-01 -1.60981226e+00 -2.85180479e-01 -4.11247402e-01 -1.09559126e-01 -1.38583255e+00 -1.57487080e-01 7.47455180e-01 -5.15189052e-01 7.02694297e-01 5.98037839e-01 6.31449640e-01 7.87441254e-01 3.83471787e-01 2.28437290e-01 1.30436754e+00 -5.74631751e-01 5.47866523e-01 2.01004595e-01 1.80795863e-01 9.83898401e-01 1.81275651e-01 5.33643544e-01 1.11351274e-01 -4.60220903e-01 8.76680255e-01 -2.87095159e-01 -3.53498399e-01 -3.98609787e-01 -8.70240390e-01 1.13062322e+00 3.36308718e-01 4.76163208e-01 -1.42611966e-01 1.51072010e-01 5.28885841e-01 8.23456585e-01 1.27786398e-01 9.76707220e-01 -4.65937644e-01 8.44783187e-02 -6.00372672e-01 1.58922046e-01 1.22892809e+00 3.64837080e-01 7.35792816e-01 2.87480950e-01 5.00338793e-01 2.68278033e-01 7.39284977e-02 2.15300992e-01 4.31516320e-01 -9.01896596e-01 5.32822832e-02 3.40535432e-01 -1.24384895e-01 -1.16293883e+00 -4.86321479e-01 -1.99528784e-01 -9.24363196e-01 9.47553158e-01 7.39584506e-01 -6.34750724e-02 -5.13580322e-01 1.77323341e+00 2.75489956e-01 -2.46675336e-03 2.98627406e-01 3.90233427e-01 1.61506400e-01 5.96587777e-01 -3.15339088e-01 -4.53512758e-01 1.04645312e+00 -1.89996481e-01 -2.40990520e-01 4.87226009e-01 8.48059356e-01 -3.65300238e-01 9.09652412e-01 5.63646376e-01 -1.02461970e+00 -1.67641923e-01 -1.40017605e+00 3.80884171e-01 -8.04501951e-01 -3.20340782e-01 5.42511344e-01 1.09419334e+00 -1.18260384e+00 1.13699698e+00 -8.44799757e-01 -2.18341798e-01 4.87833768e-01 8.73977780e-01 -3.27203512e-01 4.58104342e-01 -1.04111195e+00 9.12998438e-01 2.70208120e-01 6.97775707e-02 -6.10298872e-01 -6.24615788e-01 -9.13793981e-01 -1.60737839e-02 1.23916864e-01 -6.12977624e-01 5.14230788e-01 -1.34497523e+00 -1.61661160e+00 8.46788466e-01 3.52062643e-01 -6.15582764e-01 6.56508565e-01 2.65334696e-01 -2.05964506e-01 3.37608159e-01 -3.31918895e-01 3.13790977e-01 1.06415081e+00 -1.09455478e+00 -1.81424245e-01 -4.49047297e-01 6.05524890e-02 -1.60974190e-01 -3.30311328e-01 -1.38360053e-01 5.86521566e-01 -6.64223194e-01 2.81011686e-03 -8.63937199e-01 -2.52488226e-01 3.34689498e-01 -4.50743407e-01 -1.95371345e-01 9.97906148e-01 -3.42036098e-01 8.82128417e-01 -1.95621502e+00 3.69543940e-01 6.74984932e-01 1.40785590e-01 3.21798950e-01 2.52251178e-02 5.84137440e-01 -3.53314072e-01 5.90440929e-01 -4.65683550e-01 2.30116308e-01 1.94725320e-02 1.55466035e-01 -4.01669651e-01 1.07230723e+00 4.01739419e-01 5.27542889e-01 -5.96093237e-01 -2.70566911e-01 7.28536546e-02 4.26142395e-01 -4.08866167e-01 -1.46724090e-01 3.01301498e-02 1.59516752e-01 -5.83707511e-01 5.00864148e-01 3.80592316e-01 -3.63143906e-02 -1.59512863e-01 -2.77845580e-02 -4.15740535e-02 2.55518258e-02 -1.42829692e+00 8.83290648e-01 -3.06695491e-01 9.17532206e-01 3.17701340e-01 -1.69712782e+00 1.03588283e+00 2.25488916e-01 2.08227172e-01 -1.81691051e-01 5.44751585e-01 3.46094489e-01 7.12639466e-02 -4.14703041e-01 -1.79739118e-01 -2.65565485e-01 -2.06848439e-02 3.91534328e-01 -8.67037848e-03 -2.57699281e-01 -8.68398622e-02 -2.28159413e-01 1.40558112e+00 -3.98605525e-01 1.40183911e-01 -8.46075892e-01 8.17999542e-01 4.75952215e-02 5.94339073e-02 7.26765692e-01 -1.82082415e-01 1.53576687e-01 7.47068405e-01 -6.21958554e-01 -9.57018912e-01 -1.04934442e+00 -7.14394450e-01 4.83189881e-01 8.35340098e-02 1.34770840e-01 -8.24257791e-01 -2.97970653e-01 7.64401853e-02 2.79172391e-01 -9.45177555e-01 -2.79158741e-01 -6.32571757e-01 -8.77110064e-01 8.89426649e-01 2.29528159e-01 4.01251942e-01 -8.81953239e-01 -5.35585582e-01 -8.66244584e-02 7.82571733e-01 -7.81916201e-01 -1.54647544e-01 6.55703902e-01 -1.12984276e+00 -1.35274982e+00 -3.41995806e-01 -9.52123821e-01 7.78289199e-01 -4.80200499e-01 7.17655897e-01 3.35200459e-01 -4.58241731e-01 7.96442986e-01 1.79222912e-01 -1.75650835e-01 -8.80752861e-01 -1.21204235e-01 3.01510036e-01 3.56953032e-03 -2.25332472e-02 -9.03368473e-01 -3.19388747e-01 3.69630575e-01 -1.19112742e+00 -6.67481065e-01 2.45460674e-01 9.37617064e-01 2.94712573e-01 5.38071930e-01 6.66266382e-01 -5.80598056e-01 7.32224286e-01 -3.34285408e-01 -9.18453217e-01 5.31383306e-02 -6.81791484e-01 2.97138005e-01 8.69896770e-01 -7.09305167e-01 -2.44791925e-01 7.91961849e-02 7.30266795e-02 -3.37823093e-01 -1.42013729e-01 3.30833137e-01 3.05019114e-02 -8.57804298e-01 9.46976066e-01 2.89970040e-01 6.01415098e-01 -1.11408308e-01 2.49561787e-01 3.34367663e-01 5.46475410e-01 -7.23459542e-01 1.04581118e+00 5.62868059e-01 7.63383925e-01 -1.20523012e+00 3.33693027e-02 2.76451204e-02 -6.06212854e-01 -2.18993798e-02 7.05559492e-01 -1.57055870e-01 -1.08230114e+00 3.89158934e-01 -8.19861114e-01 -3.06462795e-01 -5.77898681e-01 3.53072077e-01 -8.20935011e-01 3.51637810e-01 -6.77866697e-01 -1.20805943e+00 -1.57218147e-02 -1.07900143e+00 6.83676422e-01 1.26944333e-01 -2.04395890e-01 -1.64965141e+00 1.38828121e-02 -1.36771351e-01 2.40489066e-01 8.79245520e-01 1.29465163e+00 -1.09625506e+00 -4.96227086e-01 -5.76334238e-01 1.39984176e-01 6.29328191e-01 -7.86923915e-02 3.94143969e-01 -7.48695672e-01 -4.20388252e-01 6.11624241e-01 -3.38038743e-01 4.98657316e-01 3.60921592e-01 9.96896863e-01 -6.51930273e-01 -2.37413824e-01 7.71511078e-01 1.63583815e+00 2.18181059e-01 7.34669447e-01 4.56254154e-01 3.39212775e-01 9.17577684e-01 -8.16845223e-02 -5.53891435e-02 -4.14945424e-01 4.17950213e-01 4.75655466e-01 -1.02985121e-01 3.48112911e-01 1.63839519e-01 4.58191782e-01 4.83921736e-01 5.42801879e-02 2.32116178e-01 -7.54279196e-01 3.32329124e-01 -1.59869683e+00 -1.02815187e+00 -1.18693300e-01 2.39285707e+00 9.09872115e-01 5.24629653e-01 3.48000407e-01 5.10688365e-01 7.69094646e-01 -1.14519864e-01 -5.16644061e-01 -8.97066474e-01 -1.49688810e-01 5.77797830e-01 7.05351651e-01 6.77161276e-01 -1.05188155e+00 3.94575745e-01 7.00922632e+00 1.14260840e+00 -1.22178280e+00 -2.68449247e-01 7.73262143e-01 3.60844076e-01 -1.89249814e-01 1.93783212e-02 -6.06917739e-01 4.10107434e-01 7.63302743e-01 3.05302534e-03 4.52717990e-01 6.85694933e-01 -1.91387385e-01 4.67639714e-02 -1.20501542e+00 8.11624467e-01 -7.73705468e-02 -1.53645444e+00 -1.69988170e-01 3.96860301e-01 7.14683473e-01 -3.74874651e-01 1.10498950e-01 3.08005698e-02 2.59538651e-01 -1.52111113e+00 3.24292123e-01 4.44308728e-01 4.64206487e-01 -7.04136372e-01 4.28478539e-01 4.27446544e-01 -8.62328649e-01 -2.23147944e-01 -1.04665369e-01 -6.21295050e-02 -3.68019581e-01 3.17379981e-01 -3.85129154e-01 3.67108583e-01 3.58247966e-01 4.03936923e-01 -4.13727105e-01 9.48980272e-01 1.86730146e-01 5.30745506e-01 -7.40575135e-01 -4.37712759e-01 3.26693565e-01 -4.54893261e-01 1.00667763e+00 1.03273606e+00 1.98709080e-03 -3.85779189e-03 -7.48018101e-02 1.20404124e+00 3.95182133e-01 6.16645999e-02 -8.74405742e-01 -7.02087581e-02 1.48376942e-01 1.10644615e+00 -1.09017038e+00 1.51993439e-01 -2.72930507e-02 4.92559880e-01 2.02742126e-02 4.42354083e-01 -3.43032032e-01 -6.70000672e-01 5.52290559e-01 1.75646245e-02 4.32504594e-01 -2.74408132e-01 -5.17577708e-01 -8.43438923e-01 -5.91953583e-02 -7.92371690e-01 3.25256795e-01 -3.31995368e-01 -1.39072847e+00 5.12783289e-01 5.46263531e-02 -8.87721360e-01 -3.87936532e-01 -1.15173686e+00 -8.95624101e-01 8.32443357e-01 -1.04973686e+00 -4.81168866e-01 3.48271221e-01 6.32053673e-01 -8.74607712e-02 -4.25180763e-01 9.41161931e-01 -1.22674406e-01 -4.19717968e-01 5.95162034e-01 2.67773181e-01 1.89789787e-01 -8.79165828e-02 -1.42063880e+00 3.65303159e-02 6.52273595e-01 3.89847048e-02 5.36755979e-01 7.79467165e-01 -3.72006834e-01 -1.62556982e+00 -6.22627854e-01 3.32317859e-01 -5.40410876e-01 9.71904874e-01 -3.20099980e-01 -1.01390326e+00 1.94179505e-01 -2.36346751e-01 1.43979058e-01 3.72859627e-01 -1.17327034e-01 -2.39434451e-01 4.55558561e-02 -1.53690159e+00 6.07329845e-01 5.77176988e-01 -5.56295276e-01 -4.56004083e-01 4.40803111e-01 3.23426574e-01 -1.00182079e-01 -8.15596104e-01 3.88257056e-01 4.47808117e-01 -7.72482812e-01 1.02143848e+00 -8.61762404e-01 1.73891604e-01 -1.21573091e-01 -1.22447267e-01 -7.95395255e-01 1.58650920e-01 -1.40107393e+00 4.06167470e-02 9.31680024e-01 3.06662560e-01 -1.16449976e+00 6.59406364e-01 5.75389326e-01 2.04418272e-01 -1.32154083e+00 -1.28047168e+00 -1.11888742e+00 4.36657012e-01 -1.52377665e-01 -1.30449021e-02 1.09331632e+00 1.75166521e-02 -5.09914830e-02 2.81406909e-01 1.50970459e-01 4.17182356e-01 -3.53014976e-01 6.27576947e-01 -1.35048723e+00 -4.45529431e-01 -9.12412465e-01 -1.21333027e+00 -6.43564582e-01 2.49335647e-01 -1.06340933e+00 -2.97173589e-01 -7.53232002e-01 -5.13397336e-01 -6.07467294e-01 1.35455579e-02 1.88157469e-01 3.95384759e-01 -8.85137357e-03 -2.11517662e-01 2.24336922e-01 1.12984367e-01 3.15106899e-01 8.41988444e-01 5.87009154e-02 -1.93520516e-01 2.02622265e-01 -4.69945222e-01 8.86834025e-01 6.39576614e-01 -5.42786598e-01 -3.16495299e-01 4.82350618e-01 4.52157617e-01 9.17819142e-02 7.19990075e-01 -1.03322434e+00 2.38734886e-01 6.97878376e-02 2.41937220e-01 2.20540047e-01 2.25406438e-01 -9.77240324e-01 -7.25165159e-02 7.96616375e-01 -4.44419950e-01 1.47150913e-02 3.03209662e-01 6.64313078e-01 3.41288932e-02 -7.20310330e-01 1.01761115e+00 2.06018999e-01 -1.86917365e-01 6.79744110e-02 -5.05389452e-01 1.17591612e-01 1.00424957e+00 -5.34367263e-01 -2.79375792e-01 -4.20529485e-01 -8.86489034e-01 -8.84376764e-02 5.03574729e-01 -2.08891258e-01 2.60709167e-01 -1.22600925e+00 -4.18998659e-01 2.98811406e-01 -5.55445433e-01 -2.61232823e-01 -2.93508500e-01 8.72903168e-01 -8.04465353e-01 1.69243850e-02 -1.61407530e-01 -6.75531447e-01 -1.13137424e+00 6.16742790e-01 8.15209329e-01 -1.06152922e-01 -4.96783048e-01 7.26726770e-01 -6.28715754e-02 -3.35556209e-01 4.01974171e-01 -2.05638245e-01 5.27993329e-02 -5.07176332e-02 3.14361662e-01 4.71807361e-01 -1.16873132e-02 -4.66403097e-01 -1.48230508e-01 7.73315072e-01 1.32905468e-01 -9.61744413e-02 1.36661983e+00 2.96643049e-01 -1.93793416e-01 3.43348682e-01 1.64342332e+00 -1.93237975e-01 -1.12391424e+00 8.29501674e-02 -2.71233171e-02 -6.24679253e-02 7.09396377e-02 -4.46831882e-01 -7.59163797e-01 8.65543723e-01 7.82533348e-01 1.01225972e+00 1.13088727e+00 -2.07184970e-01 4.14997965e-01 5.58175743e-01 6.96032867e-02 -9.96499002e-01 1.66158341e-02 3.16063732e-01 8.96306038e-01 -9.67833340e-01 4.75071222e-02 -4.94826943e-01 1.35303974e-01 1.69014728e+00 -3.59761119e-02 -7.59201050e-01 1.03809571e+00 4.02496994e-01 -2.50862628e-01 -4.31390613e-01 -4.99448121e-01 -4.07267846e-02 3.65059197e-01 7.42928088e-01 -2.50371933e-01 -4.20154303e-01 -2.23488986e-01 2.00226203e-01 -2.86861241e-01 -3.80754083e-01 4.70982701e-01 8.64577770e-01 -3.32520723e-01 -9.53908384e-01 -4.32671458e-01 3.08293134e-01 -5.83660126e-01 1.09098099e-01 -4.85428333e-01 1.25493252e+00 -1.66861281e-01 5.55444300e-01 -8.92421380e-02 -1.85986578e-01 2.30773333e-02 9.01640430e-02 6.01268053e-01 -2.47818425e-01 -6.64951026e-01 -3.80283296e-01 -6.38253763e-02 -2.68080175e-01 -3.09120417e-01 -6.79208875e-01 -1.02707350e+00 -2.89247066e-01 -4.88856405e-01 1.70025349e-01 8.32830191e-01 1.12619400e+00 1.19299017e-01 9.37175304e-02 8.13386619e-01 -1.04985762e+00 -9.82948065e-01 -4.74820912e-01 -8.45948279e-01 1.72158673e-01 4.79630262e-01 -5.59289694e-01 -1.31034195e+00 -6.47261962e-02]

[7.7939982414245605, 3.790597915649414]

e0650a89-dd29-42a1-b04a-7a2125cecb4d

scod-active-object-detection-for-embodied

2107.02069

https://arxiv.org/abs/2107.02069v1

https://arxiv.org/pdf/2107.02069v1.pdf

SCOD: Active Object Detection for Embodied Agents using Sensory Commutativity of Action Sequences

We introduce SCOD (Sensory Commutativity Object Detection), an active method for movable and immovable object detection. SCOD exploits the commutative properties of action sequences, in the scenario of an embodied agent equipped with first-person sensors and a continuous motor space with multiple degrees of freedom. SCOD is based on playing an action sequence in two different orders from the same starting point and comparing the two final observations obtained after each sequence. Our experiments on 3D realistic robotic setups (iGibson) demonstrate the accuracy of SCOD and its generalization to unseen environments and objects. We also successfully apply SCOD on a real robot to further illustrate its generalization properties. With SCOD, we aim at providing a novel way of approaching the problem of object discovery in the context of a naive embodied agent. We provide code and a supplementary video.

['David Filliat', 'Michael Garcia-Ortiz', 'Hugo Caselles-Dupré']

2021-07-05

['active-object-detection']

['computer-vision']

[ 3.30157608e-01 1.79979682e-01 2.18980432e-01 2.01028273e-01 2.36948747e-02 -8.30392420e-01 8.57812762e-01 -1.81099981e-01 -8.24097931e-01 5.77497780e-01 -1.54216483e-01 2.09280685e-01 -4.33454782e-01 -2.71424174e-01 -8.15676570e-01 -6.22829199e-01 -7.46721923e-01 3.63899529e-01 5.85846543e-01 -3.47575396e-01 3.29146892e-01 6.91605926e-01 -1.73144126e+00 -6.04696609e-02 1.54023454e-01 7.60191441e-01 8.86090100e-01 8.25400591e-01 6.02502942e-01 1.01518965e+00 -3.14931124e-01 6.41759038e-02 7.12839067e-01 -2.78999299e-01 -8.90804708e-01 2.14794427e-01 7.41580501e-02 -4.29512590e-01 -4.34437245e-01 1.01244366e+00 2.99019545e-01 5.29522002e-01 6.51326180e-01 -1.73911202e+00 -4.03984696e-01 6.55451953e-01 2.80451048e-02 1.38350397e-01 8.95557642e-01 5.17819107e-01 8.01657736e-01 -7.46939719e-01 9.70651567e-01 1.48227668e+00 2.98114866e-01 8.09698761e-01 -9.73408282e-01 1.07029425e-02 5.26392944e-02 3.58220994e-01 -1.15490007e+00 -5.10746419e-01 4.93731022e-01 -6.46545649e-01 1.01514101e+00 2.63336599e-01 7.29502916e-01 1.38539636e+00 9.92664546e-02 1.08492649e+00 7.55821764e-01 -4.63578075e-01 5.51929474e-01 -1.05737261e-01 -3.63147296e-02 5.68779349e-01 3.14909697e-01 4.80223179e-01 -7.26458907e-01 -1.11633567e-02 7.15862691e-01 9.89280194e-02 -1.85603485e-01 -1.02640760e+00 -1.89165616e+00 3.76235992e-01 4.70685154e-01 4.36671764e-01 -6.12002075e-01 2.65831470e-01 -5.08843511e-02 4.62936372e-01 -4.57793981e-01 7.32494354e-01 -4.41745706e-02 -2.51271814e-01 1.34276086e-02 6.22932971e-01 7.66892195e-01 1.32483017e+00 2.65812576e-01 -2.31845394e-01 1.28779396e-01 -5.98047785e-02 2.14467511e-01 5.99644244e-01 2.97452062e-01 -1.39755821e+00 1.71258584e-01 6.09311581e-01 6.38687134e-01 -7.73499787e-01 -7.88785160e-01 1.06182911e-01 -2.14850098e-01 9.43734646e-01 4.35319990e-01 4.23227400e-02 -4.80576724e-01 1.72855294e+00 4.24848169e-01 -4.15869296e-01 3.78223777e-01 1.22104037e+00 3.73483121e-01 2.47244030e-01 -1.51699975e-01 -3.85022648e-02 1.14590681e+00 -7.67117858e-01 -3.96817148e-01 -3.67392421e-01 7.03528225e-01 -1.50975212e-02 7.83079088e-01 6.64432764e-01 -9.09848213e-01 -4.00572389e-01 -1.19744575e+00 5.99903949e-02 -3.65810335e-01 7.76175931e-02 5.73581040e-01 -6.53993860e-02 -8.54520202e-01 7.15937674e-01 -1.24545467e+00 -7.83932328e-01 1.88942671e-01 5.36223829e-01 -7.47383475e-01 3.21536124e-01 -6.18547380e-01 1.13525760e+00 8.15633535e-01 2.09284455e-01 -1.38333881e+00 1.35312065e-01 -8.41781616e-01 -3.81169289e-01 8.18036914e-01 -4.94696856e-01 1.30273438e+00 -8.66064310e-01 -1.64339459e+00 8.60845864e-01 2.88899124e-01 -4.62587744e-01 8.17671478e-01 -5.14670908e-01 -2.40898401e-01 3.38245183e-01 1.62512735e-01 6.90059841e-01 8.17793906e-01 -9.54964042e-01 -5.92164040e-01 -6.84631407e-01 3.24837983e-01 3.79528046e-01 7.74163604e-02 -2.57931650e-01 1.70697838e-01 -2.39767671e-01 2.14323997e-01 -1.30758131e+00 -2.45508194e-01 2.00956300e-01 -4.46540356e-01 -3.51755947e-01 6.16190076e-01 4.58366722e-02 2.60308653e-01 -2.42279911e+00 7.24680722e-01 -7.70159438e-02 1.86601594e-01 -3.48675959e-02 -2.81209558e-01 6.80359423e-01 2.80354291e-01 -4.09864426e-01 -1.17420673e-01 -1.42215833e-01 3.90247703e-01 1.69983819e-01 -1.43694207e-01 7.82481194e-01 -7.05844536e-02 9.34635758e-01 -1.28492212e+00 -2.49493271e-01 2.95528889e-01 2.91201565e-02 -1.61784679e-01 2.46708676e-01 -1.70779854e-01 6.15655124e-01 -5.45793951e-01 4.32515174e-01 3.25991750e-01 3.26806791e-02 2.94900298e-01 3.81062359e-01 -3.76237839e-01 1.11111134e-01 -1.54351163e+00 1.78890550e+00 6.18186779e-02 6.35533094e-01 2.02389434e-01 -7.59277701e-01 6.73069358e-01 1.44414842e-01 1.82497606e-01 -5.89887381e-01 3.37437421e-01 2.74648160e-01 1.85430855e-01 -1.08214343e+00 7.26948142e-01 8.05355310e-02 -2.41381973e-01 4.74024892e-01 2.02541769e-01 4.00938280e-03 2.73111403e-01 2.97985911e-01 1.38429046e+00 4.16523337e-01 5.45430779e-01 -3.11459363e-01 1.50769725e-01 2.02733576e-01 8.09804201e-02 1.15632415e+00 -3.63197803e-01 1.94636807e-01 1.38029367e-01 -2.33059257e-01 -8.02641034e-01 -1.39998686e+00 1.29465476e-01 1.04848063e+00 9.15216684e-01 -8.83360133e-02 -5.39224684e-01 -4.95082557e-01 1.46143466e-01 6.57453060e-01 -8.35776925e-01 -1.66558251e-02 -5.80859601e-01 -7.27922050e-03 2.42078781e-01 5.95339596e-01 4.22596961e-01 -1.50669646e+00 -1.72243381e+00 3.20902094e-02 3.63974981e-02 -1.17361224e+00 5.58100194e-02 3.67319047e-01 -5.42984366e-01 -1.10402906e+00 -3.76778811e-01 -8.76520753e-01 7.87370026e-01 3.57952923e-01 6.48677886e-01 -1.57678068e-01 -2.02886313e-01 8.30136240e-01 -4.99231130e-01 -4.76216108e-01 -6.16834819e-01 -4.15822178e-01 7.72530675e-01 -1.60054818e-01 1.18129328e-01 -3.79579663e-01 -4.14932787e-01 3.85614008e-01 -7.43054926e-01 2.39929818e-02 5.31346500e-01 4.89123940e-01 -1.93691421e-02 -1.50893345e-01 -2.96657160e-02 1.87355056e-01 3.18767101e-01 -2.13686317e-01 -7.44674265e-01 4.80793975e-02 7.86501095e-02 9.12293047e-02 2.60066152e-01 -8.74333382e-01 -6.93763494e-01 5.21243095e-01 5.21845222e-01 -1.81721449e-01 -3.15466493e-01 1.09297350e-01 -2.05638379e-01 1.02143995e-01 8.48498702e-01 9.85763595e-02 2.65362501e-01 -4.06124115e-01 3.36121291e-01 5.55783927e-01 9.22094405e-01 -2.94609159e-01 5.68689883e-01 8.00922573e-01 2.44744286e-01 -7.41349161e-01 -8.34842492e-03 -3.21530819e-01 -1.06656373e+00 -4.75636601e-01 7.36018956e-01 -6.78348780e-01 -1.59063900e+00 6.28906786e-01 -1.26373255e+00 -6.83769822e-01 -4.94191289e-01 9.16446865e-01 -1.01943243e+00 2.67848432e-01 -2.83718556e-01 -1.13303840e+00 4.05024052e-01 -9.14251924e-01 9.77486730e-01 -1.71583463e-02 -4.50774223e-01 -4.99390721e-01 1.57493129e-01 -2.64620155e-01 -2.16853783e-01 2.71926075e-01 2.08481774e-01 -8.53370667e-01 -7.37811267e-01 -1.14880450e-01 4.77589995e-01 -9.02694538e-02 1.36891291e-01 -4.44292367e-01 -5.63062549e-01 -3.70038688e-01 7.09167197e-02 -3.47142637e-01 5.68539977e-01 -8.51170495e-02 2.64870107e-01 -2.46619165e-01 -7.31658936e-01 -8.28803107e-02 1.11168253e+00 4.09588039e-01 4.12569344e-01 7.40772724e-01 3.58932376e-01 7.75400579e-01 8.39808822e-01 4.32718068e-01 -7.61547312e-02 8.86894226e-01 1.00849044e+00 5.59665918e-01 2.36094430e-01 -2.35529453e-01 6.82618558e-01 1.28188923e-01 -3.78470004e-01 -4.71812308e-01 -6.01760685e-01 6.55931532e-01 -2.01506925e+00 -1.27134025e+00 -1.10596769e-01 2.19348073e+00 8.19804072e-02 1.50269523e-01 4.77177769e-01 2.10963190e-01 6.55040443e-01 -3.97952855e-01 -7.31703281e-01 -1.05157875e-01 -1.41962767e-01 -5.09953141e-01 4.21889365e-01 3.87652516e-01 -9.88841951e-01 5.50543606e-01 6.58955431e+00 1.63417995e-01 -7.05804169e-01 -6.56633750e-02 -6.46160185e-01 -2.34998330e-01 3.49500775e-01 3.90610017e-04 -4.28413481e-01 3.60454857e-01 1.69973522e-01 -5.10965101e-03 4.91111577e-01 9.76732850e-01 -9.67804864e-02 -6.41552746e-01 -1.67277956e+00 8.96294713e-01 -6.12732247e-02 -8.97997499e-01 -2.88074315e-01 3.19715291e-02 1.36263222e-01 6.58325031e-02 -5.07034138e-02 1.32895663e-01 4.49326575e-01 -6.84840381e-01 1.43103325e+00 4.69359279e-01 7.20524788e-02 6.36965036e-03 2.13908255e-01 8.77367496e-01 -8.24889004e-01 -5.59080184e-01 -2.11159244e-01 -6.54731452e-01 2.82371163e-01 -3.11766833e-01 -1.03603649e+00 3.30062866e-01 6.96341753e-01 6.37253582e-01 -1.29431278e-01 9.20391321e-01 -2.85468847e-01 -1.47193715e-01 -5.61374664e-01 -5.55090487e-01 1.25908583e-01 -9.33489054e-02 1.24960840e+00 7.19371736e-01 4.12456989e-01 4.52352703e-01 -1.28616467e-01 8.81291568e-01 4.63219225e-01 -5.27543068e-01 -8.60476732e-01 6.52352348e-02 2.03137487e-01 9.38089609e-01 -8.16247642e-01 -2.19892994e-01 3.51772159e-02 1.26997924e+00 2.18670562e-01 2.23329559e-01 -5.64872146e-01 -3.58447194e-01 5.72121024e-01 -2.03135028e-01 6.47087216e-01 -6.29234493e-01 2.77646005e-01 -9.41718340e-01 2.94707090e-01 -5.92635751e-01 9.46288779e-02 -1.05060267e+00 -8.52510691e-01 4.71132755e-01 1.56806409e-01 -1.68801522e+00 -2.63533175e-01 -9.93089974e-01 -2.82712042e-01 1.27897542e-02 -5.81618428e-01 -7.23651171e-01 -4.31045651e-01 6.83269322e-01 4.27055627e-01 -9.78515074e-02 7.10994780e-01 -4.60078806e-01 -3.55920494e-02 -1.21235348e-01 9.20257065e-03 2.68920027e-02 1.08363107e-01 -1.06549680e+00 3.40477705e-01 8.67383659e-01 1.60346314e-01 6.05593681e-01 9.96981621e-01 -6.11724794e-01 -1.86944401e+00 -3.75934422e-01 2.80147821e-01 -9.25843418e-01 6.77656591e-01 -7.73049891e-01 -4.32745457e-01 1.10476434e+00 -6.17547929e-02 -1.58367097e-01 -6.90598041e-02 -3.43089879e-01 6.10647202e-02 2.72923887e-01 -1.16819537e+00 8.47274542e-01 1.92084908e+00 -1.59013793e-01 -1.19951117e+00 2.74800569e-01 5.85919023e-01 -4.13753331e-01 -3.31862330e-01 4.38470036e-01 8.66253674e-01 -9.65829909e-01 8.09491098e-01 -5.83753943e-01 3.24769989e-02 -5.68972707e-01 -2.91788548e-01 -1.16620636e+00 -4.56021070e-01 -9.11318481e-01 -1.21032313e-01 5.02368689e-01 4.47398424e-02 -6.24211550e-01 3.28590602e-01 1.83012769e-01 6.70658052e-02 -4.61504087e-02 -1.04681313e+00 -1.49898016e+00 -5.97235262e-01 -5.36591172e-01 3.83304894e-01 4.17470455e-01 6.89030826e-01 2.28938699e-01 -3.44848961e-01 4.32019562e-01 6.57212198e-01 6.89008608e-02 1.04583418e+00 -1.24746060e+00 -3.97947431e-01 -1.40096754e-01 -1.04434741e+00 -1.31141126e+00 1.50581211e-01 -6.82851017e-01 5.21594584e-01 -1.27729297e+00 2.08403081e-01 -6.77936152e-02 2.87631508e-02 4.19351190e-01 2.90840417e-01 1.10070072e-01 4.97394383e-01 3.97453576e-01 -1.04639447e+00 4.77961838e-01 1.18462121e+00 -2.38108575e-01 -3.21357518e-01 -1.01471066e-01 -7.83130527e-02 7.93038368e-01 5.27498186e-01 -5.08089304e-01 -2.66610354e-01 -4.62084621e-01 3.29356670e-01 -7.59942457e-02 1.11135709e+00 -1.24627209e+00 4.01806533e-01 -2.47887075e-01 1.73503369e-01 -4.89111930e-01 7.00717151e-01 -1.15010917e+00 4.07369494e-01 9.84788060e-01 -4.85108882e-01 -5.53483749e-03 8.62329528e-02 7.01233864e-01 4.15061235e-01 -4.04230028e-01 3.32808703e-01 -3.04550290e-01 -1.45205998e+00 -2.06608579e-01 -8.02787721e-01 -3.47165495e-01 1.41932106e+00 -3.64864737e-01 -3.49725515e-01 -1.71814635e-01 -1.21345997e+00 2.54667073e-01 6.86772168e-01 5.54059029e-01 6.51868641e-01 -1.20851886e+00 -3.73927623e-01 2.52024531e-01 3.61583799e-01 -3.35494369e-01 -3.10504418e-02 9.69940543e-01 -2.86178261e-01 2.24556655e-01 -5.77028930e-01 -7.27846503e-01 -1.27300227e+00 1.05256462e+00 2.77284592e-01 4.94164884e-01 -8.62830043e-01 5.58567882e-01 2.90874451e-01 -5.19626617e-01 3.04373533e-01 -3.43915373e-01 -4.89803925e-02 -4.03036088e-01 6.65158331e-01 7.65480638e-01 -2.76844144e-01 -6.57706976e-01 -8.23219538e-01 4.04416233e-01 4.39194292e-01 -5.95119238e-01 1.23268175e+00 -2.71116048e-01 4.18297648e-02 6.76489830e-01 8.60788941e-01 -2.93514095e-02 -1.43500185e+00 -4.10543084e-02 4.64329012e-02 -3.06407392e-01 -6.07260466e-01 -5.53526342e-01 -1.73976585e-01 4.05003488e-01 6.12677932e-01 4.33940351e-01 7.51022696e-01 5.64501226e-01 1.50662482e-01 1.17268980e+00 9.92730319e-01 -1.01073229e+00 4.46754634e-01 4.25571412e-01 1.17983961e+00 -1.10899210e+00 -4.25773151e-02 -2.02988297e-01 -6.25477076e-01 1.09102702e+00 3.68299603e-01 -1.66147590e-01 1.05344452e-01 2.96566308e-01 -7.77406394e-02 -3.21306944e-01 -5.44804215e-01 -5.43581009e-01 -1.52543470e-01 9.28980708e-01 -6.30508661e-01 1.40059486e-01 -5.19205742e-02 1.97442919e-01 -2.04296023e-01 2.01698497e-01 7.77327597e-01 1.45303357e+00 -6.05905771e-01 -4.93816823e-01 -4.05133456e-01 -2.18182549e-01 2.08659634e-01 5.35391688e-01 -5.66926181e-01 1.06389630e+00 8.63263160e-02 1.08852530e+00 2.82356739e-01 -4.77284014e-01 4.75822806e-01 -2.61814952e-01 1.01541877e+00 -4.16637868e-01 -9.81678218e-02 -4.39529419e-01 -1.16358534e-01 -8.41950417e-01 -8.28442931e-01 -1.14846635e+00 -1.54156387e+00 1.01526350e-01 -2.08118618e-01 8.34342185e-03 7.04162538e-01 9.96867537e-01 3.35591584e-01 1.16110578e-01 4.97500569e-01 -1.26972115e+00 -7.03565478e-01 -9.24372196e-01 -8.67080271e-01 4.66947168e-01 7.79715240e-01 -1.10843241e+00 -5.84031284e-01 6.77932054e-02]

[4.576280117034912, 0.8487626910209656]

680e29bb-31e6-49df-9996-5cf25e096dd3

among-us-adversarially-robust-collaborative

2303.09495

https://arxiv.org/abs/2303.09495v2

https://arxiv.org/pdf/2303.09495v2.pdf

Among Us: Adversarially Robust Collaborative Perception by Consensus

Multiple robots could perceive a scene (e.g., detect objects) collaboratively better than individuals, although easily suffer from adversarial attacks when using deep learning. This could be addressed by the adversarial defense, but its training requires the often-unknown attacking mechanism. Differently, we propose ROBOSAC, a novel sampling-based defense strategy generalizable to unseen attackers. Our key idea is that collaborative perception should lead to consensus rather than dissensus in results compared to individual perception. This leads to our hypothesize-and-verify framework: perception results with and without collaboration from a random subset of teammates are compared until reaching a consensus. In such a framework, more teammates in the sampled subset often entail better perception performance but require longer sampling time to reject potential attackers. Thus, we derive how many sampling trials are needed to ensure the desired size of an attacker-free subset, or equivalently, the maximum size of such a subset that we can successfully sample within a given number of trials. We validate our method on the task of collaborative 3D object detection in autonomous driving scenarios.

['Chen Feng', 'Felix Juefei-Xu', 'Siheng Chen', 'Jiamu Bai', 'Qi Fang', 'Yiming Li']

2023-03-16

['adversarial-defense']

['adversarial']

[ 1.21789023e-01 4.60185528e-01 6.88901842e-01 -1.64836556e-01 -7.37247288e-01 -1.14322269e+00 5.84837794e-01 1.74107343e-01 -7.60447919e-01 6.18782759e-01 -3.85887742e-01 1.72545135e-01 -7.23963529e-02 -9.34522450e-01 -9.06920552e-01 -8.00063252e-01 -3.29908103e-01 6.23586297e-01 2.91370660e-01 -2.09358901e-01 4.20981869e-02 3.92246693e-01 -1.43310964e+00 -5.91743598e-03 8.25875640e-01 6.18509293e-01 -7.48909339e-02 8.15962791e-01 8.23256373e-01 6.68235838e-01 -1.40085721e+00 -3.86375248e-01 6.76737607e-01 -1.88705176e-01 -4.56562012e-01 1.62041664e-01 5.97228169e-01 -4.66273308e-01 6.58942526e-03 1.34391868e+00 6.20073378e-01 2.19465569e-01 5.21701634e-01 -1.50391138e+00 -1.74836352e-01 9.64271903e-01 -2.18786314e-01 -2.67748743e-01 4.49731499e-01 5.89484990e-01 7.02496350e-01 -3.03271085e-01 3.30766171e-01 1.59626436e+00 4.02053297e-01 7.21536040e-01 -1.57418692e+00 -8.58249784e-01 2.09781021e-01 -1.22269168e-02 -1.35515904e+00 -3.00964147e-01 5.56145668e-01 -3.73742044e-01 2.36249968e-01 3.83939207e-01 5.49540162e-01 1.64809120e+00 2.97250986e-01 3.71411473e-01 1.29342854e+00 1.69981062e-01 7.37097859e-01 1.59226060e-01 -1.98689207e-01 4.27044690e-01 8.14226747e-01 3.08628947e-01 -6.11925185e-01 -4.97029990e-01 2.51704961e-01 -2.89085984e-01 -2.08036631e-01 -5.56599975e-01 -1.17444670e+00 8.65445912e-01 4.74200159e-01 -3.12302321e-01 -4.55336034e-01 3.23270649e-01 3.06338489e-01 6.99163795e-01 -4.79052737e-02 8.68468940e-01 -1.08199894e-01 1.27816722e-01 -2.05278322e-01 5.83891749e-01 9.78429914e-01 5.31803429e-01 5.84780276e-01 1.44255519e-01 2.49094889e-01 1.92071289e-01 2.72713955e-02 9.54221845e-01 -2.68163443e-01 -1.45438039e+00 3.50915521e-01 1.67345583e-01 5.46465635e-01 -1.16553414e+00 -1.48652852e-01 -4.03272957e-01 -6.22240543e-01 1.21564925e+00 5.76909304e-01 -6.98125124e-01 -3.10027748e-01 2.10233307e+00 6.21863663e-01 -1.11429179e-02 4.24783468e-01 1.03746188e+00 2.86618829e-01 1.36197150e-01 -2.57780045e-01 -4.83978242e-02 1.11381972e+00 -4.20210630e-01 -1.56708375e-01 -4.68540788e-01 2.03817070e-01 -3.06241959e-01 6.04189157e-01 8.84948552e-01 -7.72513092e-01 -5.02032936e-01 -1.37154460e+00 7.67921686e-01 -1.56938029e-03 -4.53899741e-01 2.71984190e-01 1.05667698e+00 -7.73177922e-01 5.27228415e-01 -8.68002295e-01 -2.40809456e-01 3.86660933e-01 4.80292231e-01 -4.63415265e-01 3.70877758e-02 -8.67080510e-01 9.59262311e-01 9.23829153e-02 -6.67078719e-02 -2.11948800e+00 -2.55848408e-01 -6.06619895e-01 -1.21505298e-01 8.96319449e-01 -6.95428014e-01 1.01137853e+00 -8.92410815e-01 -1.32811773e+00 4.94522780e-01 4.46215808e-01 -8.82967532e-01 8.39083612e-01 -3.53641152e-01 3.73998970e-01 2.24255264e-01 1.58442453e-01 5.11823535e-01 1.21871686e+00 -1.76509857e+00 -4.01023090e-01 -4.67999279e-01 6.39186442e-01 3.20791483e-01 -2.08399460e-01 -1.44169018e-01 4.68289226e-01 -2.04257816e-01 -1.11938946e-01 -1.22461963e+00 -5.11472464e-01 4.82803434e-02 -5.13602018e-01 -9.29633752e-02 5.53487480e-01 7.21701756e-02 -5.90457879e-02 -2.22947335e+00 3.73302579e-01 3.31822038e-01 5.65858364e-01 8.93096700e-02 -3.22195709e-01 6.33857310e-01 3.27207714e-01 1.94965690e-01 1.40153006e-01 -6.03235006e-01 2.43443564e-01 2.85227388e-01 -4.16320145e-01 8.62087131e-01 -1.20858224e-02 3.12140286e-01 -1.05557930e+00 -5.19154258e-02 -1.73916724e-02 -6.96719810e-02 -6.95507228e-01 3.08915883e-01 -2.45221555e-02 5.25328279e-01 -6.47046983e-01 5.29651821e-01 7.02516258e-01 4.12172675e-01 3.67344707e-01 2.80971915e-01 1.44916624e-01 -7.82246813e-02 -1.37264645e+00 1.10434628e+00 -2.19755590e-01 4.16680962e-01 9.00471449e-01 -1.03031218e+00 9.71617699e-01 1.73569456e-01 7.17880055e-02 1.06397927e-01 4.44387436e-01 1.17590606e-01 4.97324884e-01 -1.82682965e-02 2.60527074e-01 -4.32307944e-02 -5.67796767e-01 6.22720659e-01 -2.59015620e-01 -4.65771437e-01 -3.23610902e-01 3.04844975e-01 1.59102976e+00 -3.88771474e-01 -1.58388540e-01 -1.32977247e-01 1.91823184e-01 7.08504468e-02 7.65449643e-01 1.74596584e+00 -6.20277524e-01 1.87161088e-01 5.26482821e-01 -2.69034892e-01 -7.69945741e-01 -1.37703836e+00 3.14722776e-01 5.93497694e-01 7.09683359e-01 -2.65492767e-01 -7.10658252e-01 -1.03031778e+00 2.68672287e-01 7.42797673e-01 -7.88380682e-01 -5.78292191e-01 -2.57635385e-01 -1.26352414e-01 1.00970268e+00 -1.02859633e-02 3.54110032e-01 -1.08710575e+00 -1.12366605e+00 -1.08644530e-01 2.16037109e-02 -9.89618123e-01 7.01059848e-02 2.72759616e-01 -1.23678908e-01 -1.37196505e+00 -5.35770059e-02 -3.14316928e-01 7.56110549e-01 5.63497245e-01 6.33303046e-01 1.37181692e-02 8.18197578e-02 8.58897686e-01 -5.52141070e-01 -7.10363269e-01 -9.18517649e-01 -4.88679260e-01 7.86647499e-01 -1.22462787e-01 6.76022517e-03 -7.33285189e-01 -2.80052632e-01 4.66885298e-01 -5.76499820e-01 -5.78419507e-01 4.36138511e-01 6.97621644e-01 -1.09542102e-01 5.14500856e-01 4.37235355e-01 -2.10050300e-01 8.36534917e-01 -2.93771565e-01 -8.36355388e-01 4.21271920e-02 1.48960188e-01 -3.04778516e-01 7.94792295e-01 -1.04666233e+00 -3.45483124e-01 -9.90833119e-02 4.07583833e-01 -6.80849731e-01 -2.94716954e-01 8.85608271e-02 -2.74752229e-01 -3.27104717e-01 1.07778251e+00 1.40283659e-01 4.10976231e-01 -9.10400879e-03 2.68636167e-01 3.95053297e-01 3.43642026e-01 -8.66119385e-01 1.39695907e+00 6.56549335e-01 7.91334361e-02 -6.09057128e-01 -5.18154263e-01 3.85597199e-01 -2.09848672e-01 -6.54517174e-01 7.81673670e-01 -9.94515896e-01 -1.54957056e+00 5.45898080e-01 -1.34339929e+00 -3.54323804e-01 -3.64831060e-01 4.68394101e-01 -4.04723912e-01 4.59263474e-01 -9.45703462e-02 -1.24449742e+00 7.46256784e-02 -1.33765352e+00 7.92418897e-01 -2.29380396e-03 -1.88588411e-01 -2.70619512e-01 -1.77902997e-01 4.35985744e-01 1.09997816e-01 5.80889463e-01 3.58948737e-01 -1.12071955e+00 -7.12393939e-01 -4.52222288e-01 5.20265102e-01 4.67561841e-01 -1.34514822e-02 -1.98030770e-01 -6.57972276e-01 -8.62606704e-01 3.64244580e-01 -9.41978991e-01 6.25281870e-01 -1.28750905e-01 6.94720149e-01 -4.05630618e-01 -3.46022129e-01 9.76766832e-03 7.72629797e-01 2.54880279e-01 1.90591842e-01 2.58860886e-01 4.01542246e-01 8.59288216e-01 7.54927695e-01 5.25280178e-01 1.44283652e-01 6.22931719e-01 1.01751995e+00 4.60856199e-01 3.16140622e-01 -2.02145010e-01 8.84339273e-01 -4.45263125e-02 2.58918077e-01 -3.86059940e-01 -6.81137919e-01 3.35256130e-01 -1.68831730e+00 -9.18833673e-01 8.79504383e-02 2.54645085e+00 4.22910362e-01 5.78981280e-01 1.40527666e-01 1.07255302e-01 8.62122059e-01 -6.81789368e-02 -6.71760142e-01 -2.62947470e-01 -4.15433943e-02 -2.56922305e-01 5.25045753e-01 6.53882921e-01 -1.03503489e+00 6.84893012e-01 5.87769508e+00 7.03342557e-01 -6.69593573e-01 9.58833843e-02 2.63681382e-01 -3.81144911e-01 -4.07475047e-03 3.45998228e-01 -5.49176812e-01 2.18414143e-01 5.01581609e-01 -2.65829355e-01 6.38575196e-01 1.07918584e+00 4.80451509e-02 -3.29200715e-01 -1.30326498e+00 6.22743666e-01 1.70070857e-01 -7.78775513e-01 -3.29064667e-01 4.46930438e-01 4.86264616e-01 8.54419265e-03 2.20757108e-02 2.47080654e-01 1.19425392e+00 -8.80442441e-01 1.16086698e+00 1.02563746e-01 7.73835778e-02 -7.89266765e-01 7.44343698e-01 9.60424960e-01 -5.66867411e-01 -4.74463999e-01 -5.60083687e-01 -2.62376785e-01 -3.75381112e-02 3.62396300e-01 -7.69000113e-01 5.37977099e-01 7.42239177e-01 2.86672041e-02 -3.82072538e-01 7.09724665e-01 -4.83448714e-01 3.80087554e-01 -6.36277616e-01 -4.53500003e-01 9.42474082e-02 -6.59270883e-02 1.34521282e+00 2.63826996e-01 1.60443529e-01 -4.48262691e-02 8.37347567e-01 9.19990599e-01 2.77522326e-01 -5.43980181e-01 -8.23450029e-01 1.60533413e-01 1.12861598e+00 1.08408165e+00 -3.96403730e-01 -1.73593476e-01 1.94518253e-01 8.55911076e-01 3.87406766e-01 3.06062326e-02 -1.04299593e+00 -3.02684903e-01 9.16459441e-01 -3.15346420e-01 2.56291002e-01 -4.20761555e-01 -7.25012571e-02 -9.11924362e-01 -6.35819659e-02 -1.48115504e+00 7.37348152e-03 -4.43635315e-01 -1.53552270e+00 4.66849774e-01 -4.19579595e-02 -1.28301895e+00 9.18132812e-02 -3.38512599e-01 -7.98922658e-01 4.11085844e-01 -5.88624418e-01 -1.04672408e+00 -4.06419277e-01 4.88131583e-01 2.46705160e-01 -3.73086810e-01 6.61951184e-01 -3.51526082e-01 -5.23202717e-01 6.58184946e-01 -3.99463445e-01 6.36643078e-03 7.61283576e-01 -1.13917696e+00 8.45495090e-02 1.05022013e+00 1.45367095e-02 6.66256309e-01 1.16786015e+00 -8.81572545e-01 -1.46806717e+00 -7.86561549e-01 1.55958086e-02 -8.12656522e-01 7.43932545e-01 -8.32791686e-01 -3.42057228e-01 4.39839333e-01 -2.58470010e-02 -3.13301653e-01 2.79791772e-01 1.61085188e-01 -4.68499690e-01 -1.36737302e-01 -1.42136037e+00 8.93064380e-01 8.65023077e-01 -3.00712407e-01 -6.53214931e-01 2.71017522e-01 7.87892103e-01 -1.69608936e-01 -6.37683094e-01 3.64091009e-01 3.67144018e-01 -1.15435660e+00 8.52418065e-01 -3.46823990e-01 -4.90506478e-02 -6.28420532e-01 -3.78748178e-01 -1.37755764e+00 -9.35877487e-03 -9.40693140e-01 3.30394208e-01 9.92086947e-01 1.08740844e-01 -9.08596218e-01 7.23251164e-01 3.01187485e-01 -4.01468202e-02 -1.58376664e-01 -1.27338362e+00 -1.15683603e+00 2.28496805e-01 -2.84537822e-01 3.29037368e-01 6.74216390e-01 -4.73777764e-02 1.04174972e-01 -6.13935411e-01 9.87911046e-01 1.19736600e+00 -2.63605297e-01 1.55828166e+00 -1.26643634e+00 -5.76980889e-01 -6.00457527e-02 -5.65055132e-01 -8.97771537e-01 2.99074620e-01 -4.61974949e-01 4.17227775e-01 -1.00292587e+00 8.26305971e-02 -6.01012349e-01 3.68371636e-01 5.16021430e-01 -8.65264758e-02 7.42014870e-02 4.93751079e-01 1.47500679e-01 -7.71417320e-01 5.48949182e-01 1.03034449e+00 -2.95363009e-01 -4.96890070e-03 2.39881620e-01 -7.49056697e-01 7.70085096e-01 7.61572242e-01 -7.22026527e-01 -3.79279077e-01 -1.90000013e-01 4.92618904e-02 3.70986760e-02 9.63762820e-01 -1.22997046e+00 2.99862862e-01 -2.74491489e-01 2.81027913e-01 -2.59984761e-01 7.17028379e-01 -8.26920867e-01 1.33519411e-01 9.48208749e-01 -3.70073050e-01 -2.25127757e-01 -1.02902696e-01 8.85208309e-01 2.99693882e-01 -2.47235745e-01 6.61982954e-01 -2.54674345e-01 -4.93609086e-02 -1.28869772e-01 -1.00994158e+00 -2.26895437e-01 1.40307403e+00 -2.27167904e-01 -6.59975648e-01 -7.60030150e-01 -7.89415717e-01 4.59756315e-01 6.99439406e-01 3.41094971e-01 5.63302755e-01 -8.15295935e-01 -9.73926306e-01 5.27983755e-02 -1.04904650e-02 9.21426043e-02 7.52503797e-02 6.26328826e-01 -2.30106805e-02 -5.08203149e-01 -7.16611370e-02 -5.04512846e-01 -1.58747315e+00 5.61207473e-01 5.13762355e-01 2.14754641e-01 -3.35284352e-01 9.98543620e-01 3.02433878e-01 -4.45709646e-01 3.43134493e-01 2.70145565e-01 1.74245000e-01 -2.80609429e-01 4.26086277e-01 3.29328299e-01 -3.76354516e-01 -3.08738887e-01 -5.83191574e-01 9.75199416e-02 -1.00667998e-01 -3.83549124e-01 9.47563589e-01 -2.96843387e-02 -9.38362256e-03 1.60160270e-02 4.39573765e-01 5.15342832e-01 -1.57310438e+00 1.59535110e-01 -4.93335009e-01 -6.36922598e-01 -2.91743845e-01 -6.02730989e-01 -5.51380098e-01 4.42682564e-01 5.34981608e-01 7.55066872e-01 6.16250932e-01 1.16154984e-01 1.63562447e-01 8.56925309e-01 1.06581640e+00 -7.69252002e-01 5.59718788e-01 3.18437070e-01 9.27279115e-01 -1.09155786e+00 1.30020589e-01 -3.57977837e-01 -7.15178370e-01 6.79124773e-01 9.61968839e-01 -5.96548259e-01 1.74100138e-02 3.10241908e-01 1.98556989e-01 -9.81603861e-02 -7.90876985e-01 -1.78345352e-01 -3.96810442e-01 1.24920619e+00 -8.39788735e-01 1.76039025e-01 1.41300008e-01 4.00287449e-01 -5.17776072e-01 -9.06172395e-01 1.10358059e+00 8.85952830e-01 -7.99528599e-01 -9.48200881e-01 -8.61294866e-01 -5.87662980e-02 -1.51319444e-01 4.14742410e-01 -9.04738903e-01 5.92437088e-01 3.43123913e-01 1.72930038e+00 -7.23827258e-02 -7.36253798e-01 2.88060844e-01 -5.62323570e-01 5.05147398e-01 -7.37422943e-01 -6.75292015e-01 -2.37070158e-01 2.57721543e-01 -6.13664806e-01 -1.17255986e-01 -8.04434419e-01 -6.71870887e-01 -4.34539795e-01 -4.45963144e-01 3.43688786e-01 4.47806180e-01 8.27393115e-01 7.81292021e-02 2.50008076e-01 1.04871523e+00 -9.82297480e-01 -1.12228715e+00 -8.60197425e-01 -3.86438578e-01 1.43330559e-01 4.11126018e-01 -8.58431518e-01 -9.00744915e-01 -5.78313351e-01]

[3.9270174503326416, 2.4597625732421875]

432e5eb4-dd5e-47af-ae00-35d35877a661

festa-flow-estimation-via-spatial-temporal

2104.00798

https://arxiv.org/abs/2104.00798v2

https://arxiv.org/pdf/2104.00798v2.pdf

FESTA: Flow Estimation via Spatial-Temporal Attention for Scene Point Clouds

Scene flow depicts the dynamics of a 3D scene, which is critical for various applications such as autonomous driving, robot navigation, AR/VR, etc. Conventionally, scene flow is estimated from dense/regular RGB video frames. With the development of depth-sensing technologies, precise 3D measurements are available via point clouds which have sparked new research in 3D scene flow. Nevertheless, it remains challenging to extract scene flow from point clouds due to the sparsity and irregularity in typical point cloud sampling patterns. One major issue related to irregular sampling is identified as the randomness during point set abstraction/feature extraction -- an elementary process in many flow estimation scenarios. A novel Spatial Abstraction with Attention (SA^2) layer is accordingly proposed to alleviate the unstable abstraction problem. Moreover, a Temporal Abstraction with Attention (TA^2) layer is proposed to rectify attention in temporal domain, leading to benefits with motions scaled in a larger range. Extensive analysis and experiments verified the motivation and significant performance gains of our method, dubbed as Flow Estimation via Spatial-Temporal Attention (FESTA), when compared to several state-of-the-art benchmarks of scene flow estimation.

['Dong Tian', 'YingLi Tian', 'Muhammad A. Lodhi', 'Jiahao Pang', 'HaiYan Wang']

2021-04-01

http://openaccess.thecvf.com//content/CVPR2021/html/Wang_FESTA_Flow_Estimation_via_Spatial-Temporal_Attention_for_Scene_Point_Clouds_CVPR_2021_paper.html

http://openaccess.thecvf.com//content/CVPR2021/papers/Wang_FESTA_Flow_Estimation_via_Spatial-Temporal_Attention_for_Scene_Point_Clouds_CVPR_2021_paper.pdf

cvpr-2021-1

['scene-flow-estimation']

['computer-vision']

[ 7.22251311e-02 -4.46997941e-01 -1.97732113e-02 -7.46075958e-02 -9.72704291e-02 -2.50454664e-01 4.74104226e-01 -2.06667557e-02 -2.25451544e-01 5.79800129e-01 2.59996086e-01 -1.65941536e-01 -2.05359071e-01 -8.47827494e-01 -5.88607907e-01 -6.91768050e-01 9.40340385e-02 -1.21892132e-02 3.71137321e-01 -3.06089878e-01 5.79173803e-01 8.88933480e-01 -1.58861578e+00 -1.77248746e-01 9.43662047e-01 1.08971703e+00 2.92430103e-01 5.80923438e-01 -6.18068695e-01 8.51314604e-01 -4.10064340e-01 -2.29466930e-01 4.49037552e-01 -2.12049484e-01 -6.08873188e-01 2.79908746e-01 5.11145592e-01 -6.35038853e-01 -8.38011146e-01 9.87528026e-01 1.83281720e-01 4.03846204e-01 3.03096712e-01 -1.30942631e+00 -3.87529880e-01 -1.92124337e-01 -9.15436327e-01 6.07243538e-01 4.09578919e-01 4.93572950e-01 6.79113328e-01 -7.85939336e-01 6.28136098e-01 1.36900723e+00 2.78218806e-01 3.27028304e-01 -8.33418012e-01 -6.39446855e-01 4.18210506e-01 3.98943692e-01 -1.24144340e+00 -2.32675001e-01 1.12390685e+00 -4.82176661e-01 8.84571373e-01 9.18151736e-02 9.31822419e-01 8.15380573e-01 1.00959241e-01 9.09237981e-01 6.95021331e-01 1.80373654e-01 9.51739550e-02 -1.87852174e-01 1.32640335e-03 4.77388412e-01 2.75023431e-01 -2.41519837e-03 -5.53347707e-01 3.87414992e-01 1.18235540e+00 3.18665981e-01 -5.04548550e-01 -4.60618764e-01 -1.39750624e+00 5.35347581e-01 7.93041825e-01 1.07624494e-02 -5.42660475e-01 1.98121116e-01 5.25096416e-01 -1.47369998e-02 4.67334718e-01 1.92894861e-01 -1.73137069e-01 -3.95829469e-01 -7.20792353e-01 3.38272423e-01 3.22205901e-01 1.23458612e+00 8.95020902e-01 2.99803406e-01 -1.30384564e-02 3.35303962e-01 1.00994319e-01 5.51222444e-01 2.65639365e-01 -1.11568844e+00 8.79787505e-01 8.92002940e-01 3.25849831e-01 -1.51292682e+00 -3.92151773e-01 -3.03604960e-01 -1.29014957e+00 1.33477271e-01 4.44641143e-01 3.03344131e-01 -6.15608215e-01 1.40077770e+00 6.14543438e-01 6.61532104e-01 -1.69949070e-01 1.35387969e+00 7.26776779e-01 8.51148307e-01 -5.38456589e-02 -2.66883314e-01 1.11102450e+00 -7.55317688e-01 -7.21499324e-01 -1.74322128e-01 4.05186743e-01 -4.16828007e-01 1.01393378e+00 2.70217717e-01 -9.28888857e-01 -8.05290639e-01 -9.94981170e-01 -4.30511862e-01 -1.61905095e-01 -2.82669276e-01 8.37601364e-01 3.33188981e-01 -6.97975099e-01 3.63094777e-01 -9.09649491e-01 -2.42319405e-01 8.08457196e-01 1.13964558e-01 -5.55114627e-01 -3.22126746e-01 -9.23940122e-01 4.29348826e-01 9.06384811e-02 4.51514155e-01 -5.33291876e-01 -9.84162152e-01 -9.30454731e-01 1.83547493e-02 2.27730557e-01 -7.91739404e-01 8.14985991e-01 -3.76290739e-01 -1.45610535e+00 5.70028424e-01 -5.32885909e-01 -3.69698793e-01 6.43793762e-01 -4.95037138e-01 -1.55385807e-01 2.38795713e-01 2.34830409e-01 5.15047789e-01 8.71584892e-01 -1.08793664e+00 -7.31634080e-01 -4.77073222e-01 1.52622581e-01 3.93514931e-01 2.56849285e-02 -4.78256434e-01 -4.78759140e-01 -3.99594545e-01 3.92033339e-01 -6.97036505e-01 -4.01076764e-01 1.26219064e-01 -3.81809205e-01 -1.73339069e-01 1.20264399e+00 -3.16504776e-01 1.13746238e+00 -2.04967690e+00 3.18737298e-01 -2.01279387e-01 4.91977304e-01 1.43473446e-01 1.72920778e-01 5.96378930e-02 5.90469595e-03 -2.20313799e-02 -3.16843450e-01 -3.88849705e-01 -3.88047367e-01 1.08811624e-01 -7.07814991e-01 7.77733743e-01 5.35608053e-01 7.31680453e-01 -1.26511812e+00 -4.64380682e-01 1.05812013e+00 5.52673101e-01 -6.85923934e-01 1.95105061e-01 -1.35639295e-01 1.00848508e+00 -8.98413420e-01 5.61298311e-01 1.03583074e+00 -3.18486154e-01 -5.78026414e-01 -3.07711303e-01 -5.15657544e-01 1.53486222e-01 -1.22289550e+00 2.11604404e+00 -5.98982036e-01 8.79412711e-01 -1.23003669e-01 -9.29019809e-01 9.79226708e-01 1.02029564e-02 8.17254305e-01 -7.28537738e-01 1.90440297e-01 5.66815399e-02 -2.59269267e-01 -6.90987110e-01 9.06091094e-01 1.09780520e-01 -5.42415865e-03 5.77506889e-03 -3.37594241e-01 -4.24759269e-01 1.92509014e-02 1.06123708e-01 1.05086386e+00 2.02920556e-01 2.03159228e-01 -1.18551105e-01 9.67139661e-01 2.81233370e-01 8.18576336e-01 3.55607212e-01 -6.35489225e-01 7.32602239e-01 4.04857516e-01 -7.41859972e-01 -9.76752698e-01 -8.04196060e-01 -6.43442646e-02 2.00944930e-01 8.11550975e-01 -2.31876373e-01 -2.68904835e-01 -4.24100101e-01 3.38523835e-02 3.72817844e-01 -4.71324116e-01 -1.04206234e-01 -9.23994660e-01 -4.17306215e-01 9.16086882e-02 4.77974147e-01 9.13926840e-01 -1.05860865e+00 -1.18220365e+00 3.47223401e-01 -2.60016501e-01 -1.52571952e+00 -3.88305038e-01 -3.26804906e-01 -1.17994893e+00 -1.00437284e+00 -7.69778967e-01 -2.83549875e-01 5.10302305e-01 8.25644970e-01 9.67062354e-01 -4.78441119e-02 -3.57443482e-01 1.93203852e-01 -2.99489737e-01 -1.84852764e-01 1.56626627e-01 -2.73352917e-02 3.26870009e-03 3.04765344e-01 4.41653907e-01 -6.58449471e-01 -1.05308282e+00 2.49132365e-01 -8.42868626e-01 1.25658289e-01 2.77855903e-01 6.25604510e-01 4.65047866e-01 1.21177725e-01 2.63610929e-01 -5.26839733e-01 1.57767370e-01 -4.88742828e-01 -7.38699257e-01 -2.99162924e-01 1.71525761e-01 -7.41330013e-02 6.97044551e-01 -7.70494565e-02 -1.06523991e+00 1.18019544e-01 -3.44748497e-02 -9.82321680e-01 -2.40456954e-01 3.62330722e-03 -3.26504916e-01 6.68207183e-02 3.09205383e-01 3.24570000e-01 -1.86212376e-01 -9.07443613e-02 2.67852366e-01 3.62599671e-01 6.06095970e-01 -3.39415401e-01 8.45204115e-01 9.61499870e-01 4.02031779e-01 -1.14548302e+00 -7.20940173e-01 -7.72124767e-01 -7.42293417e-01 -3.40066493e-01 1.03875291e+00 -9.50422764e-01 -1.07913256e+00 5.68884730e-01 -1.52153921e+00 -1.10752337e-01 -2.37794235e-01 6.44343376e-01 -6.45085752e-01 5.04554093e-01 -3.55878532e-01 -9.16123807e-01 -1.45240173e-01 -1.51167703e+00 1.18821120e+00 3.63909394e-01 -2.21922230e-02 -8.21235120e-01 -1.10528946e-01 3.36904943e-01 2.90109307e-01 4.63996083e-01 5.34823656e-01 3.38262081e-01 -1.29039192e+00 1.76933929e-01 -5.93952358e-01 3.60174961e-02 2.93264806e-01 2.97267810e-02 -9.56278563e-01 8.80208910e-02 1.95099145e-01 3.84203829e-02 6.69045687e-01 6.41899884e-01 1.31636095e+00 5.21980785e-02 -1.48437470e-01 1.06670392e+00 1.35728717e+00 1.72608748e-01 7.47674346e-01 3.11660558e-01 1.15454268e+00 5.81395566e-01 8.01082850e-01 5.48600554e-01 3.54439467e-01 7.26791620e-01 7.69546926e-01 8.99441615e-02 -1.48771659e-01 -3.32096100e-01 -1.55166209e-01 7.89738178e-01 -1.54338956e-01 -9.43461061e-02 -8.18259895e-01 6.40184522e-01 -1.78238952e+00 -9.28665817e-01 -3.57096076e-01 2.15889931e+00 2.53638744e-01 2.53031939e-01 -1.36662230e-01 4.86405969e-01 5.39283097e-01 4.94054794e-01 -8.92083406e-01 -1.05129994e-01 -1.35086447e-01 -2.98552364e-02 5.56792617e-01 5.12397707e-01 -1.00755656e+00 1.03456461e+00 4.63047361e+00 6.34154737e-01 -1.21778011e+00 -1.58711940e-01 5.45284450e-01 -1.21918125e-02 -2.74528652e-01 -1.12682655e-01 -7.78926909e-01 6.24870420e-01 3.21097642e-01 -1.74262375e-01 7.00717047e-02 6.75916910e-01 5.40347099e-01 -1.12088144e-01 -8.56024742e-01 1.36065388e+00 -2.44925097e-01 -1.44979358e+00 2.16192082e-01 1.20543770e-01 7.00319469e-01 -2.04045195e-02 -1.01999752e-01 5.98022342e-02 -3.88882519e-03 -6.80035233e-01 5.67647636e-01 3.92765492e-01 7.75438845e-01 -7.10482121e-01 5.49079776e-01 2.83917934e-01 -1.54018652e+00 -4.54227068e-02 -4.52783585e-01 -1.41134292e-01 5.99341869e-01 6.90693736e-01 -3.26519340e-01 8.20167780e-01 8.19385588e-01 1.44131303e+00 -1.98807254e-01 1.07298708e+00 -1.66023541e-02 1.29235893e-01 -3.40010732e-01 1.59641072e-01 4.25230950e-01 -5.38206339e-01 9.11422431e-01 8.69978905e-01 4.19437885e-01 4.19621021e-01 -5.82748689e-02 9.65630352e-01 2.44337860e-02 -1.62257731e-01 -8.50440323e-01 3.13834071e-01 4.31952834e-01 1.07724190e+00 -8.08705091e-01 -2.11570367e-01 -4.57409680e-01 9.10075188e-01 1.05606385e-01 4.26424384e-01 -7.49272883e-01 -3.41015518e-01 1.13463271e+00 2.38414541e-01 6.34537488e-02 -7.36647248e-01 -4.09215450e-01 -1.27535117e+00 4.75460067e-02 -2.57524431e-01 7.67854527e-02 -8.32116485e-01 -9.75979805e-01 5.95673501e-01 -6.87105395e-03 -1.81397510e+00 1.67868212e-01 -4.14834291e-01 -5.92144608e-01 9.66870010e-01 -1.93951714e+00 -5.31711578e-01 -1.06974661e+00 7.21573114e-01 9.17343915e-01 2.38809332e-01 2.29758490e-02 4.55264688e-01 -6.93769336e-01 4.23180535e-02 -4.12096798e-01 5.56721762e-02 2.46858791e-01 -9.63569880e-01 6.43111765e-01 9.96120632e-01 -1.20516583e-01 3.67749423e-01 4.75577325e-01 -5.12059689e-01 -1.76224267e+00 -1.18643057e+00 5.85655153e-01 -4.24331903e-01 5.31418800e-01 -1.17872015e-01 -1.06272340e+00 4.02236611e-01 -2.06127957e-01 5.17898738e-01 1.74282584e-02 -5.70680797e-01 1.69926416e-02 -2.79555291e-01 -9.07528996e-01 5.89055836e-01 1.37648094e+00 -3.56114030e-01 -3.03355813e-01 6.65764809e-02 1.08344281e+00 -6.90028071e-01 -7.07606554e-01 4.71453011e-01 2.57991403e-01 -1.18861747e+00 1.22730124e+00 -2.61641771e-01 6.97425127e-01 -6.05785787e-01 -3.07450686e-02 -9.71139312e-01 -1.68816462e-01 -6.94063067e-01 -4.34107244e-01 8.11848640e-01 -3.80622745e-01 -5.27307808e-01 1.12725210e+00 4.76547509e-01 -3.21508765e-01 -4.06258076e-01 -1.07800663e+00 -4.93947566e-01 -2.46116981e-01 -5.83380878e-01 7.05577374e-01 9.09309626e-01 -4.15219456e-01 2.42610753e-01 -2.35071242e-01 3.30750227e-01 8.26406598e-01 2.45398507e-01 1.15207481e+00 -1.14659214e+00 1.87020734e-01 -7.02801168e-01 -9.31971431e-01 -1.79067636e+00 1.64668933e-02 -5.25870502e-01 -1.13558754e-01 -1.50810528e+00 -4.91799474e-01 -6.35997295e-01 -3.08557563e-02 -2.92630017e-01 -3.35908145e-01 1.14718534e-01 3.10232431e-01 3.59907895e-01 -4.20561194e-01 1.06673944e+00 1.66471243e+00 -4.55586091e-02 -5.49825072e-01 2.46098526e-02 -2.28691563e-01 5.36508739e-01 5.33121228e-01 -1.35680854e-01 -6.21766686e-01 -6.83464050e-01 1.32542878e-01 4.44389105e-01 4.55009460e-01 -1.24603188e+00 3.62895519e-01 -2.90571302e-01 1.81591153e-01 -1.04035103e+00 5.77035666e-01 -9.99844611e-01 -7.81815127e-02 3.67670476e-01 6.45412281e-02 2.29715928e-01 2.91277289e-01 8.06260645e-01 -4.77385044e-01 3.50477695e-01 5.87372005e-01 -1.75469995e-01 -1.05505729e+00 8.82189393e-01 3.54291610e-02 1.10127620e-01 1.06437266e+00 -4.90780681e-01 -2.89546520e-01 -2.11349353e-01 -2.09199548e-01 3.50442052e-01 3.46175551e-01 5.99699616e-01 8.75926733e-01 -1.28497112e+00 -6.55174911e-01 5.37504375e-01 2.02800691e-01 9.02959645e-01 7.27910340e-01 9.61910963e-01 -1.01436210e+00 5.77419698e-01 -3.08564335e-01 -1.12079632e+00 -6.07460797e-01 4.51608688e-01 1.13333263e-01 -6.05580173e-02 -1.13291156e+00 6.24311686e-01 5.36755860e-01 -9.93400719e-03 7.05720559e-02 -7.18376577e-01 -1.29259512e-01 -1.60572320e-01 5.24971545e-01 6.11622036e-01 -4.31820098e-03 -7.62540579e-01 -3.41583848e-01 1.05619061e+00 2.09703282e-01 2.63018966e-01 9.96019125e-01 -4.17523354e-01 1.52705714e-01 4.00215358e-01 1.20575964e+00 -2.77870446e-01 -1.75948834e+00 -2.31786281e-01 -2.78729677e-01 -1.05326796e+00 2.37211093e-01 2.65235871e-01 -1.30773771e+00 1.33154190e+00 1.99603841e-01 2.07608536e-01 1.09932923e+00 -3.96217942e-01 1.07684886e+00 1.92460760e-01 5.29095888e-01 -5.17035365e-01 -1.63358390e-01 4.37251389e-01 6.94808960e-01 -1.34201097e+00 1.34767428e-01 -5.84206402e-01 -4.69758064e-01 1.02254295e+00 6.67647421e-01 -3.11730206e-01 6.29715800e-01 -2.89339703e-02 -2.65498519e-01 -8.32558423e-02 -4.68068719e-01 -1.15810409e-01 6.12606406e-02 4.26896960e-01 -2.80787200e-02 -3.29354972e-01 2.14892060e-01 -1.00176007e-01 -9.78831351e-02 1.63393959e-01 5.76274455e-01 7.72979736e-01 -2.25155264e-01 -4.43296969e-01 -2.27331012e-01 1.89869434e-01 -8.48771781e-02 2.29530916e-01 2.77488023e-01 9.59985793e-01 -6.77612945e-02 8.62265289e-01 4.07624334e-01 -3.20947021e-01 5.31114519e-01 -5.17081022e-01 3.70380521e-01 -2.62817234e-01 -2.38811240e-01 -2.37880677e-01 -4.70104992e-01 -1.02151632e+00 -7.01189101e-01 -6.19873583e-01 -1.21927369e+00 -4.38783258e-01 1.58290323e-02 -4.90848757e-02 5.88698924e-01 7.26253390e-01 3.72355253e-01 6.30567908e-01 8.37611437e-01 -1.24362075e+00 9.10253674e-02 -6.29998267e-01 -4.28103924e-01 6.11322939e-01 7.96782553e-01 -9.33611512e-01 -5.51819384e-01 2.20543966e-02]

[8.540590286254883, -2.057718515396118]

344f7121-f06b-44da-9878-d5c53d126421

multi-dimensional-and-multi-scale-modeling

2303.03737

https://arxiv.org/abs/2303.03737v1

https://arxiv.org/pdf/2303.03737v1.pdf

Multi-Dimensional and Multi-Scale Modeling for Speech Separation Optimized by Discriminative Learning

Transformer has shown advanced performance in speech separation, benefiting from its ability to capture global features. However, capturing local features and channel information of audio sequences in speech separation is equally important. In this paper, we present a novel approach named Intra-SE-Conformer and Inter-Transformer (ISCIT) for speech separation. Specifically, we design a new network SE-Conformer that can model audio sequences in multiple dimensions and scales, and apply it to the dual-path speech separation framework. Furthermore, we propose Multi-Block Feature Aggregation to improve the separation effect by selectively utilizing information from the intermediate blocks of the separation network. Meanwhile, we propose a speaker similarity discriminative loss to optimize the speech separation model to address the problem of poor performance when speakers have similar voices. Experimental results on the benchmark datasets WSJ0-2mix and WHAM! show that ISCIT can achieve state-of-the-art results.

['Wenjing Zhu', 'Xinyu Yang', 'Zhaoxi Mu']

2023-03-07

['speech-separation']

['speech']

[-0.03181819 -0.5565229 -0.10706517 -0.35037386 -0.94813675 -0.3813049 0.18629506 -0.2387086 -0.07354158 0.299 0.38322937 -0.19241309 -0.4214418 0.07603276 -0.09501217 -0.9178478 -0.139646 0.08608121 0.18544869 -0.10879592 -0.13610446 0.49723244 -1.303584 0.3561807 1.1158435 1.0901062 0.4076201 0.76617104 -0.15434991 0.5217425 -0.6767467 -0.22037894 0.37019536 -0.66732 -0.38110873 0.10944238 0.39184448 -0.05622912 -0.4323602 0.9508987 1.019653 0.39184082 0.32916498 -1.2922484 -0.07247107 0.98738855 -0.7178526 0.5997529 0.05436086 -0.18272606 1.2089776 -0.9957646 -0.22382912 1.539538 0.6098005 0.3285601 -1.2247827 -1.2238678 0.4726575 0.39214346 -1.3964207 -1.1631547 0.9483483 -0.12000186 0.8975226 0.583444 0.43718722 0.76025695 -0.40651673 1.2409744 0.8381131 -0.23080216 -0.02555294 -0.11780389 0.33567324 0.33301637 -0.27191037 -0.00870467 -1.069836 -0.05462337 0.33624798 -0.21353449 -0.55776566 -0.01612095 -1.0186719 0.42705083 0.17642494 0.44208807 -0.1626675 -0.1952644 0.48484573 0.5333633 0.57180876 0.23819721 -0.43219492 -0.35440072 -1.2145046 -0.02652265 0.738908 0.88698196 0.3325902 0.5224937 -0.31955332 1.2931018 0.19353424 0.54950136 0.8004361 -0.77502227 0.8445588 -0.05668438 -0.37274733 -0.8001665 -0.26792765 -1.0143812 -0.9533127 -0.32006636 -0.04111052 -0.18412538 -0.8006828 1.6076562 0.21205245 0.75478554 0.29953474 0.9101951 0.8446335 0.7807955 -0.33442855 -0.49219522 1.037539 -1.392675 -0.8462931 -0.23081277 0.09592875 -0.9221371 0.65939224 0.56819534 -1.362313 -0.83221275 -1.0304573 0.39181846 0.02766315 0.39806148 0.19771487 0.7185111 -0.8310872 0.36516026 -0.8403467 0.03760276 0.18786241 0.46443617 -0.1352835 0.17801993 -1.0434368 0.24917863 0.242378 0.19054922 -0.74395424 -0.76383376 -0.6781939 0.6313774 0.36568382 -0.32212064 1.1444991 -0.8115923 -1.8249158 0.02103672 -0.55589604 -0.5548418 0.15563759 -0.29922184 -0.970252 0.34073594 -0.04396974 0.31735152 1.3588744 -1.2007164 -0.93984085 -0.35492986 -0.4931275 0.34994245 -0.8267409 0.2487903 -0.6319939 -1.0702971 0.27347285 -0.87766844 -0.02169165 -0.6118602 -0.6332719 -0.06254674 0.913748 -1.039491 1.6355847 -2.8756943 0.53755647 0.33001986 0.20476753 0.7736857 -0.40028006 0.28438267 -0.15432864 -0.18402189 -0.02089283 -0.90455323 0.15361278 -0.05519482 -0.47186324 0.2510347 -0.04495139 0.32491353 -0.5704699 -0.4203372 0.10408846 0.47407264 -0.62304527 0.29461792 0.5390009 0.40172487 -0.16049026 0.41159785 0.94064 0.2111183 0.1582503 -0.19431005 -0.08342731 0.6981529 -1.2194239 1.6509869 -0.5589079 0.6499634 0.85286695 -1.0167241 0.8403906 0.66525424 0.41366875 -0.5231461 -0.00663544 0.35765067 0.32433355 -0.15787573 0.2222106 -0.07754886 0.17117558 -0.00608555 0.17504822 -0.07636692 0.14479426 0.08314487 0.86546385 -0.68535024 0.12681203 0.01445223 0.8616057 -0.85084486 0.9142674 0.5059109 -0.35644254 0.5354366 0.25219414 0.4250267 -0.3554986 -1.2676331 0.02411474 1.1640399 0.1723746 -0.74855226 -0.6918113 -0.5576125 -0.05442797 0.48780423 0.02571508 -0.2908593 -0.5971449 -0.48127326 0.8203034 0.61364067 0.5027335 -0.4425793 0.11975485 0.2871168 -0.34697193 -1.2152777 -1.007982 0.3947262 -0.6043625 -0.44215325 -0.7710425 -0.8198023 0.21111098 0.900666 0.5745442 -0.32623765 0.12811239 0.00957715 -0.46385568 -0.05659792 -0.1799103 0.3647288 0.3317777 0.46971938 0.15511674 -0.9836897 -0.39884403 0.48696172 -0.6091541 -0.21863015 0.5625303 0.874451 0.18561766 0.53377163 0.96307194 -0.12849681 0.82495916 -0.3522968 -0.26062983 0.19635591 -0.42337963 -0.14472827 0.924959 -0.5502811 -1.1861975 -0.29108685 -0.20454668 -0.7816671 0.08680556 0.4454478 -0.67018056 0.06314413 -0.10718948 0.45859513 -0.07506036 -0.88976586 0.07891415 1.0770957 0.39291567 -0.3629578 0.895567 0.22574905 -0.39977697 -1.0063325 -0.6433719 -1.0185149 -0.28564695 0.18022095 0.46863514 -1.1168073 -0.66941756 0.56583196 -0.9121658 0.02863917 -0.08750627 0.9000845 -0.13764022 0.34390092 -0.62102675 -0.8139942 -0.42621085 -1.4675652 0.9270362 0.14787686 -0.08868511 -0.61770636 -0.08457746 0.4736923 0.48518202 -0.76419723 0.4389322 -1.0264672 -0.32596558 0.1656325 -0.05122744 0.8716409 0.52069557 -0.2867768 -1.1691678 -0.5828942 0.12346286 0.18524353 1.0627872 0.44826144 0.9751268 -0.27138853 -0.397536 0.8278014 0.638397 0.53597534 0.24543679 -0.14046553 0.8684098 0.6280957 0.5134781 0.41250098 0.1417754 1.0150652 -0.11889628 -0.18874657 -0.49414438 0.11532946 0.7623753 1.6730837 0.3878053 -0.23261641 -0.62225014 0.58445096 -1.797155 -1.0563757 0.08721737 2.0857573 0.76029843 0.12929747 0.30789813 0.47620672 0.75384855 0.36822933 -0.37230653 -0.22244059 -0.17519285 0.25141203 0.12767665 0.5527006 -1.0173954 0.89304054 5.647451 1.6100729 -1.2681111 0.21897215 0.24483821 -0.5629731 0.01816239 -0.24450798 -0.9741229 0.68456215 0.92945546 -0.19221081 0.6246481 0.2820679 0.34319717 0.4096074 -0.99039155 1.1968898 0.26715776 -0.73792547 -0.11411836 -0.13191201 0.47575957 -0.07973789 0.37134865 0.23234007 -0.05461632 -0.6812865 0.71277857 0.0575682 0.47762474 -1.0853591 0.3977323 0.21291444 -1.7308955 -0.37555912 -0.02450642 0.40040702 0.16907486 0.7334645 -0.8902026 0.848749 0.6882039 0.86411405 -0.3505551 1.1482208 0.02670349 0.7906136 -0.19625512 0.43280286 0.15853466 -0.11086514 1.0631123 1.4736208 0.5016388 -0.20384139 0.24574439 0.27214006 0.0283692 0.02111286 -0.10489112 0.07122616 0.7687185 1.0958552 -0.44474947 -0.2841554 -0.43264273 0.9454429 0.09878997 0.5445129 -0.793187 -0.6338735 1.2282915 -0.2704701 0.8364591 -0.19179253 0.08227223 -1.1934111 0.08739524 -1.3599069 0.04012774 -0.23519175 -1.1325463 0.8621073 -0.25744045 -1.5018405 -0.05305377 -0.28509873 -0.60270774 0.840761 -1.6596327 -1.0012996 0.10623771 0.7879154 0.911358 -0.57242227 0.4058268 0.81379724 -0.98998743 1.024899 0.43847337 0.01768375 0.89637506 -1.0504855 0.4509475 0.97010446 0.30641618 0.64418155 0.48071322 -0.24028784 -1.0395916 -0.904371 0.61560154 0.23546635 0.6124141 -0.51512164 -0.9230165 0.40182284 0.24045053 -0.15976384 1.0636551 0.24088144 -0.46228755 -0.729012 -0.6778043 0.5106051 1.0320672 -0.6998379 -0.5001153 -0.17114475 0.99524266 -0.29094163 -0.5838179 0.22739506 0.49830252 -1.0275034 1.3811376 -0.24638614 -0.10186541 -0.38606855 -0.25825575 -1.7841939 -0.43190172 -1.113762 -0.16669485 1.793368 0.3361926 -0.73685473 0.3611128 -0.04408048 -0.5552603 -0.35479826 -1.226309 -1.2711806 -0.29296845 -0.27426133 0.8211796 0.83713317 0.22413872 0.47396818 -0.5216102 0.44951013 0.52863985 0.2003859 0.668403 -0.99644774 -0.70432997 -0.7046338 -0.2076396 -1.5365918 0.5437838 -0.79815346 0.04323087 -1.0513722 -0.14265957 -0.32495418 -0.67902076 0.14153486 -0.43208233 -0.07867623 0.62807345 0.1990022 -0.6432396 0.8697952 1.1058487 -0.30926898 -0.42824304 0.35273862 -0.67179346 0.58408695 0.7391707 -0.3928863 -0.58029646 -0.45820794 -0.5314725 0.16963695 0.02177724 -1.2631769 0.50343996 0.16450953 -0.11141064 -0.73301524 0.90698195 -0.9871864 -0.08268634 0.10117318 -0.4058412 -0.39827564 0.31022394 0.5431338 -0.80247504 0.03501926 0.6178327 0.5457744 -0.26275283 0.20818467 -0.16459309 -0.04494032 0.6108609 0.06023124 -0.19844232 -0.3775634 -0.6726707 0.36945605 -0.148324 0.47585094 0.6487646 -1.3558733 -0.800432 0.5263943 -0.18004076 -0.26658022 0.4008953 1.126563 0.20683162 0.49266705 0.23085532 -0.5690813 -1.716333 0.36086303 0.23195705 -0.21347675 -0.4352383 1.2421331 0.5098391 -0.24372306 0.5413669 -0.34659547 -0.20370346 0.22165923 0.6002455 0.7106608 0.2418961 -0.82876366 -0.52654105 0.5260736 -0.31545138 -0.24974497 1.2070324 -0.54447985 -0.03861957 0.33441365 1.3676469 0.6730937 -1.1869504 -0.53989327 -0.24889137 -0.70106924 0.3265102 -0.4963848 -1.4417304 0.99357826 0.47828892 0.50476533 1.5187454 -0.10798344 1.2273813 0.0626815 -0.15422541 -0.8883347 0.17138042 0.5524158 0.7278093 -0.9040038 -0.39570892 -0.5142634 -0.5587565 0.7891354 0.44483995 0.30678824 0.71066 0.438914 0.05815044 0.15981394 -0.5888743 -0.2341976 0.48318547 0.42628348 0.41129443 0.23264515 0.05891338 0.82828593 -0.2927996 -0.57139796 -0.08174199 0.5970849 -0.38021827 -1.3490491 -0.67695314 0.2893897 -0.4289711 -0.46312702 -0.38463628 0.19183 -0.10538866 1.4461138 -0.0365755 -0.6386857 0.3272287 -0.04460104 0.1182258 -0.48815197 -0.6708682 0.7770266 0.22040543 -0.28499478 -0.41530633 -0.73972774 -1.139167 -0.21716055 -0.5557068 0.6408701 0.469933 0.85617596 0.5642098 1.0632722 1.0228262 -0.8796922 -0.6034571 -0.98201257 -0.89402646 0.02019721 0.9794245 -0.4841724 -0.6993895 -0.10722163]

[14.884148597717285, 5.888002872467041]

ec6c9442-9c06-4956-b2e2-6349509f0e7a

skeleton-cloud-colorization-for-unsupervised

2108.01959

https://arxiv.org/abs/2108.01959v3

https://arxiv.org/pdf/2108.01959v3.pdf

Skeleton Cloud Colorization for Unsupervised 3D Action Representation Learning

Skeleton-based human action recognition has attracted increasing attention in recent years. However, most of the existing works focus on supervised learning which requiring a large number of annotated action sequences that are often expensive to collect. We investigate unsupervised representation learning for skeleton action recognition, and design a novel skeleton cloud colorization technique that is capable of learning skeleton representations from unlabeled skeleton sequence data. Specifically, we represent a skeleton action sequence as a 3D skeleton cloud and colorize each point in the cloud according to its temporal and spatial orders in the original (unannotated) skeleton sequence. Leveraging the colorized skeleton point cloud, we design an auto-encoder framework that can learn spatial-temporal features from the artificial color labels of skeleton joints effectively. We evaluate our skeleton cloud colorization approach with action classifiers trained under different configurations, including unsupervised, semi-supervised and fully-supervised settings. Extensive experiments on NTU RGB+D and NW-UCLA datasets show that the proposed method outperforms existing unsupervised and semi-supervised 3D action recognition methods by large margins, and it achieves competitive performance in supervised 3D action recognition as well.

['Alex C. Kot', 'Meng Hwa Er', 'Shijian Lu', 'Jun Liu', 'Siyuan Yang']

2021-08-04

http://openaccess.thecvf.com//content/ICCV2021/html/Yang_Skeleton_Cloud_Colorization_for_Unsupervised_3D_Action_Representation_Learning_ICCV_2021_paper.html

http://openaccess.thecvf.com//content/ICCV2021/papers/Yang_Skeleton_Cloud_Colorization_for_Unsupervised_3D_Action_Representation_Learning_ICCV_2021_paper.pdf

iccv-2021-1

['3d-human-action-recognition']

['computer-vision']

[ 6.28166258e-01 -2.60520846e-01 -6.35627031e-01 -3.39372814e-01 -5.88907182e-01 -3.52662683e-01 3.99499416e-01 -5.41675687e-01 -5.21363080e-01 4.01412904e-01 4.16385800e-01 2.02495605e-02 3.23193878e-01 -2.64448583e-01 -7.12531865e-01 -6.93054259e-01 -1.20688854e-02 4.83428299e-01 4.28688347e-01 2.69513786e-01 2.17958733e-01 6.48598075e-01 -1.44330835e+00 2.60309905e-01 4.78533179e-01 1.16948080e+00 -2.63931036e-01 8.05906892e-01 -2.22983569e-01 1.21199989e+00 -5.39941549e-01 -6.47386611e-02 5.18552959e-01 -6.99567556e-01 -8.98303151e-01 8.83290052e-01 4.75934386e-01 -4.41941917e-01 -4.84912574e-01 8.66449177e-01 2.53672749e-01 1.35254368e-01 8.28815937e-01 -1.31116009e+00 -5.90023875e-01 1.51401788e-01 -1.10691893e+00 1.83565289e-01 4.83558238e-01 2.73195624e-01 7.60115087e-01 -7.70467997e-01 5.02742469e-01 1.17734170e+00 4.30085838e-01 8.22800994e-01 -1.00089395e+00 -5.50164104e-01 3.39356631e-01 5.02016425e-01 -9.72131491e-01 -2.92989254e-01 9.21678782e-01 -4.21700984e-01 9.86784697e-01 -2.23232195e-01 1.04178941e+00 1.25114059e+00 -1.35436729e-01 1.47369158e+00 9.47241604e-01 -5.54656565e-01 2.97143728e-01 -9.81204331e-01 -9.15525034e-02 1.02745819e+00 -1.14136383e-01 -7.50489533e-02 -8.54971707e-01 1.68645605e-01 1.24729681e+00 4.77322340e-01 -4.27928939e-03 -9.54095304e-01 -1.34638536e+00 3.77585888e-01 2.43948743e-01 1.52696511e-02 -3.51736039e-01 5.42012990e-01 5.33750832e-01 1.30283916e-02 3.95989448e-01 -2.14677289e-01 -5.08480310e-01 -6.44472599e-01 -6.40540957e-01 -2.74558693e-01 2.58567631e-01 8.80816638e-01 6.28783345e-01 2.12735608e-01 1.40066162e-01 7.81515718e-01 3.00027162e-01 7.47137785e-01 8.75419259e-01 -1.16839504e+00 5.53731620e-01 1.02734232e+00 -6.60887957e-02 -5.87590337e-01 -2.22843811e-01 2.84730494e-01 -6.56926095e-01 5.24612725e-01 6.63198948e-01 1.37256816e-01 -1.23298550e+00 1.32228327e+00 4.64740306e-01 5.54463983e-01 7.77132809e-02 1.12581241e+00 2.75640845e-01 2.92026848e-01 2.21357241e-01 4.65764813e-02 9.07790720e-01 -1.34006786e+00 -4.35545206e-01 -1.13257349e-01 7.49117613e-01 -3.68423402e-01 1.14149439e+00 4.08966869e-01 -8.70324314e-01 -6.71454787e-01 -1.15027881e+00 -5.61329313e-02 -1.50590539e-01 5.53246915e-01 7.93474674e-01 6.31620705e-01 -4.38183397e-01 5.11668980e-01 -1.64879990e+00 -4.75149333e-01 7.23724842e-01 2.09717542e-01 -6.47064388e-01 -2.17099309e-01 -5.64327359e-01 5.15224338e-01 3.57820332e-01 1.02914669e-01 -1.12881279e+00 -3.12961750e-02 -9.64030623e-01 -4.58064944e-01 4.88723755e-01 -3.29155773e-01 1.17017245e+00 -1.23505473e+00 -1.85758793e+00 1.02756214e+00 9.15483460e-02 -3.64044040e-01 4.94972765e-01 -5.11437535e-01 -3.08102071e-01 6.08106673e-01 3.43116224e-02 6.35676563e-01 1.11844778e+00 -8.22421193e-01 -7.00005710e-01 -9.02905464e-01 -7.70423636e-02 4.54939723e-01 -4.25821960e-01 -5.78978099e-02 -6.77297950e-01 -8.03610086e-01 4.19606715e-01 -1.00917017e+00 -2.84626096e-01 4.84379113e-01 -3.01335573e-01 -1.46880463e-01 1.01075280e+00 -3.46831411e-01 5.99566042e-01 -2.10891962e+00 3.37044716e-01 2.13547293e-02 -1.52174383e-01 3.40762854e-01 -1.58164471e-01 7.55730793e-02 -2.39514336e-01 -4.09944564e-01 -4.27078545e-01 -2.43510738e-01 -6.24152459e-02 6.98943675e-01 9.52234194e-02 7.06067562e-01 2.66797543e-01 9.77686107e-01 -1.01294160e+00 -8.03664148e-01 3.81670982e-01 3.14952374e-01 -3.65410537e-01 1.74362719e-01 -2.51176476e-01 5.78403294e-01 -8.14818442e-01 1.14792252e+00 1.66440666e-01 -4.41593081e-01 1.75514683e-01 1.01148739e-01 3.06725353e-01 -1.23400226e-01 -1.08256221e+00 2.35727358e+00 -7.97487944e-02 2.83441454e-01 -4.32199270e-01 -1.34505320e+00 8.48105669e-01 4.05250676e-02 1.08634067e+00 -6.08082592e-01 -7.04481229e-02 6.43415153e-02 -3.83057445e-01 -6.10449433e-01 9.10043940e-02 1.07406778e-02 -3.25325765e-02 9.30105031e-01 7.71605074e-02 1.86207697e-01 1.91309750e-01 7.18951002e-02 1.41608191e+00 1.02386308e+00 1.64808854e-01 4.38988209e-01 7.48973489e-01 3.05242520e-02 6.10149622e-01 1.79676980e-01 -6.35315299e-01 7.25283325e-01 5.20798326e-01 -4.85965967e-01 -8.15304399e-01 -1.14931870e+00 4.48270142e-01 1.16869545e+00 1.71990305e-01 -2.80637801e-01 -6.59949601e-01 -1.28222430e+00 4.27747071e-02 6.69835359e-02 -7.18046784e-01 -2.56653905e-01 -8.07762563e-01 -2.42731139e-01 6.57196581e-01 1.17787313e+00 6.38890624e-01 -1.10857177e+00 -1.10367131e+00 -1.10786952e-01 -5.79674952e-02 -1.20013893e+00 -4.97918904e-01 1.48597881e-01 -1.24441040e+00 -1.45708179e+00 -1.06717825e+00 -7.01307178e-01 9.47141945e-01 2.99205035e-01 5.73478341e-01 -1.50948554e-01 -5.30941129e-01 1.06094825e+00 -7.42875040e-01 4.26259786e-02 -2.19875537e-02 -3.50209743e-01 2.50982761e-01 4.74458665e-01 6.58853650e-01 -5.56457639e-01 -7.32883930e-01 4.41917360e-01 -9.85451400e-01 -2.99840476e-02 8.87367308e-01 7.71028638e-01 8.70769739e-01 -2.79936761e-01 1.92102883e-02 -6.64841354e-01 -5.31141907e-02 6.30777255e-02 -1.65424317e-01 3.39718103e-01 -3.16087604e-01 1.48835644e-01 3.72349054e-01 -4.65653807e-01 -1.14510596e+00 8.77900660e-01 2.21233979e-01 -1.00732219e+00 -4.41339433e-01 2.34566614e-01 -2.14715526e-01 6.50118710e-03 5.71682692e-01 4.71196324e-01 3.22137803e-01 -7.58638084e-01 5.83554208e-01 5.63301086e-01 6.82839990e-01 -6.62404180e-01 7.35977054e-01 8.93195748e-01 -3.14854085e-02 -7.25837350e-01 -8.26817453e-01 -6.96329296e-01 -1.44669223e+00 -6.56353176e-01 1.10710216e+00 -8.34268272e-01 -3.33208948e-01 7.95211256e-01 -8.44467580e-01 -5.26261508e-01 -5.31990349e-01 7.44942605e-01 -1.09836090e+00 9.53679860e-01 -4.21503484e-01 -7.67005861e-01 1.35064414e-02 -8.62505138e-01 1.44228959e+00 2.34225653e-02 -5.80239855e-02 -6.00918472e-01 3.79661709e-01 6.72409356e-01 -3.50492835e-01 5.27167201e-01 7.14094162e-01 -3.29264730e-01 -4.12862331e-01 -4.32875842e-01 -1.44109353e-01 6.17981255e-01 3.99576783e-01 -2.16348395e-02 -4.99556333e-01 7.87412748e-02 -3.39623213e-01 -9.16955054e-01 9.51842308e-01 1.31946266e-01 1.33688378e+00 8.04699436e-02 -1.93734974e-01 4.64536220e-01 9.65743542e-01 5.65098412e-02 6.12724423e-01 1.92596704e-01 9.51829374e-01 3.59333545e-01 7.73642182e-01 7.25059032e-01 -9.33412369e-03 5.27856469e-01 4.32042956e-01 7.91465212e-03 -2.33697534e-01 -5.29352009e-01 7.19954431e-01 6.95295513e-01 -7.50612676e-01 5.06506801e-01 -7.85015345e-01 1.63478956e-01 -2.11292768e+00 -9.11885023e-01 1.86235309e-01 2.00857496e+00 8.45380664e-01 1.87026262e-01 5.55662155e-01 4.58115786e-01 5.23237944e-01 3.12382728e-01 -9.61959541e-01 1.62394255e-01 7.30033815e-02 4.18487132e-01 4.76597667e-01 -1.21855050e-01 -1.38963866e+00 9.46509123e-01 5.76339960e+00 5.60653090e-01 -7.92086065e-01 -1.46353707e-01 3.36907178e-01 -1.06204748e-01 4.95455652e-01 -1.59027606e-01 -2.00385064e-01 1.92205846e-01 4.42514092e-01 1.46822810e-01 9.53541175e-02 1.22617054e+00 1.15636531e-02 -7.35556707e-02 -1.24959934e+00 1.23366880e+00 4.22168761e-01 -8.19759846e-01 5.79843186e-02 -8.05982575e-02 7.86668599e-01 -1.46914169e-01 -1.43199921e-01 1.61590368e-01 3.26484948e-01 -8.50026131e-01 5.11692643e-01 7.55031168e-01 8.73991191e-01 -5.94746530e-01 2.07764447e-01 2.11459443e-01 -1.28896713e+00 -1.66198909e-01 -2.36477360e-01 -1.32700324e-01 1.59337670e-01 -1.73656512e-02 -2.55719274e-01 2.92977333e-01 7.12883115e-01 1.53517473e+00 -6.77747905e-01 8.80804479e-01 -5.55474877e-01 5.34336090e-01 -1.81280181e-01 7.99549222e-02 3.08093667e-01 -2.57816195e-01 2.12463439e-02 8.14937651e-01 -5.18251620e-02 3.44789594e-01 4.43429619e-01 3.84179324e-01 7.51007199e-02 5.16678626e-03 -4.11689311e-01 -5.15344262e-01 -1.57526061e-01 8.23534489e-01 -9.47329521e-01 -3.27196568e-01 -7.08657265e-01 1.52264678e+00 2.59453058e-01 3.75156879e-01 -8.06636155e-01 -1.13811925e-01 4.90252465e-01 -2.06145301e-01 5.30842662e-01 -6.79445744e-01 -5.86990081e-02 -1.30643415e+00 -5.28997649e-03 -7.90332556e-01 6.29244804e-01 -8.28421593e-01 -1.09101725e+00 -1.18738919e-01 -5.46677597e-02 -1.73965180e+00 -1.90525427e-01 -9.91607785e-01 -3.23278874e-01 2.57687792e-02 -1.19296956e+00 -1.21726358e+00 -3.09359938e-01 1.01268470e+00 7.23912895e-01 -3.65716308e-01 7.78142989e-01 1.83678210e-01 -6.76841080e-01 3.79303843e-01 2.31585409e-02 6.13711357e-01 4.69354212e-01 -1.26275063e+00 2.68371940e-01 6.97239399e-01 7.21797228e-01 1.83643013e-01 -1.07678473e-01 -6.85408235e-01 -1.70635092e+00 -9.44156349e-01 2.53368050e-01 -5.34646809e-01 5.89464605e-01 -6.91936910e-02 -4.69065279e-01 7.72840261e-01 -2.97730267e-01 4.14625645e-01 8.41725707e-01 -2.23197386e-01 -6.19900584e-01 -8.68793577e-02 -7.08387554e-01 5.71625412e-01 1.56054068e+00 -4.59542811e-01 -6.83375239e-01 5.19024432e-01 1.39741912e-01 -3.50160390e-01 -6.81406319e-01 3.25521618e-01 8.13959241e-01 -9.12234128e-01 1.05398858e+00 -1.19161987e+00 6.54483080e-01 -4.32920814e-01 -9.90305543e-02 -7.47109413e-01 4.91065606e-02 -2.45441675e-01 -4.57348973e-01 5.67880630e-01 5.41415485e-03 -5.86570427e-02 1.51577222e+00 4.35330272e-01 -1.22899033e-01 -7.39684582e-01 -1.01425707e+00 -9.59644020e-01 -2.82864720e-01 -5.42572320e-01 1.46788940e-01 6.34909511e-01 1.31527662e-01 6.09822460e-02 -4.32244360e-01 -2.76646346e-01 7.94221580e-01 3.16461861e-01 1.14260972e+00 -8.91143262e-01 -3.39418113e-01 -3.51153731e-01 -1.06244397e+00 -1.58717084e+00 3.45367104e-01 -7.33216703e-01 9.56267715e-02 -1.33302224e+00 1.97233930e-01 -2.66644340e-02 -4.65494543e-01 8.70478928e-01 6.88683987e-02 4.62280542e-01 2.19944119e-02 4.14721906e-01 -1.07645750e+00 8.72076273e-01 1.31141639e+00 -4.36245173e-01 1.07794881e-01 5.81060052e-02 1.04532801e-01 1.04329455e+00 4.56238508e-01 -1.95220158e-01 -4.84341919e-01 -4.82350856e-01 -4.30339485e-01 6.22342341e-02 3.63737404e-01 -1.29040205e+00 8.04055557e-02 -4.04376924e-01 7.78185666e-01 -6.88678563e-01 4.90711570e-01 -9.60647404e-01 -2.53197432e-01 4.36478734e-01 -5.12735486e-01 -2.74713278e-01 -3.05213422e-01 1.11598945e+00 -1.01184450e-01 -1.42631009e-02 6.99705958e-01 -3.89371485e-01 -1.12593257e+00 6.19363844e-01 -4.14615870e-01 1.09484173e-01 1.22372043e+00 -6.61186039e-01 3.07248741e-01 -1.25746176e-01 -8.21501732e-01 -3.76473693e-03 5.21218479e-01 5.11023641e-01 9.53349710e-01 -1.66284716e+00 -2.64485508e-01 3.08984518e-01 4.89488274e-01 -1.40758500e-01 1.28397122e-01 8.36936593e-01 -6.69907928e-01 2.21711263e-01 -6.65363669e-01 -8.06772172e-01 -1.24940681e+00 4.72188354e-01 1.69609100e-01 -6.98137954e-02 -9.70463455e-01 6.98992729e-01 -1.79712325e-01 -2.24913478e-01 7.26214230e-01 -2.60087639e-01 -1.57511756e-01 -3.09388071e-01 4.09275055e-01 5.29756010e-01 -3.94156247e-01 -8.90530348e-01 -3.91915023e-01 1.00918138e+00 1.02095552e-01 -1.59581631e-01 1.09850204e+00 1.81223676e-01 2.79751778e-01 5.02551794e-01 1.27916074e+00 -6.34457469e-01 -1.69124210e+00 -3.93313795e-01 2.94532239e-01 -9.01975036e-01 -3.71356606e-01 -4.68001992e-01 -1.47032154e+00 9.90180910e-01 6.92425370e-01 -3.08647841e-01 1.10886931e+00 -5.90119474e-02 9.34005320e-01 6.05923533e-01 4.98017907e-01 -1.45267987e+00 9.53608692e-01 4.06359702e-01 5.95279813e-01 -1.19502604e+00 2.39126891e-01 -6.60350546e-02 -9.23904419e-01 1.31205738e+00 7.42519498e-01 -2.96773762e-01 5.90052664e-01 -1.75550178e-01 2.10323617e-01 -2.24676788e-01 -2.62925178e-01 -5.61439991e-01 1.31183267e-01 7.63557315e-01 2.03773201e-01 -7.48472661e-02 -1.37603045e-01 3.44421387e-01 4.13646638e-01 1.30031362e-01 1.28114700e-01 1.45298803e+00 -3.57400477e-01 -1.26260507e+00 -2.25584283e-01 3.31764549e-01 -1.73100740e-01 5.43546617e-01 -7.24101722e-01 6.74517214e-01 -7.16221258e-02 4.60379809e-01 -5.52594252e-02 -4.57628280e-01 4.06033814e-01 3.92516822e-01 8.74471903e-01 -6.51839614e-01 1.59142658e-01 2.63835192e-01 -6.58637807e-02 -9.55729842e-01 -1.06521809e+00 -1.00431001e+00 -1.69515276e+00 4.34663028e-01 -6.30093962e-02 -2.15526879e-01 5.13840437e-01 1.03980470e+00 1.37156844e-01 1.87219277e-01 6.36747777e-01 -1.05462885e+00 -5.65816164e-01 -8.19564283e-01 -9.31933939e-01 7.69541264e-01 4.13719527e-02 -1.12929213e+00 -1.81826293e-01 6.84087813e-01]

[7.860272407531738, 0.38868314027786255]

0be47ba3-1917-4ce3-a695-8cbd7419ca98

generalized-label-propagation-methods-for

1901.09993

https://arxiv.org/abs/1901.09993v3

https://arxiv.org/pdf/1901.09993v3.pdf

Label Efficient Semi-Supervised Learning via Graph Filtering

Graph-based methods have been demonstrated as one of the most effective approaches for semi-supervised learning, as they can exploit the connectivity patterns between labeled and unlabeled data samples to improve learning performance. However, existing graph-based methods either are limited in their ability to jointly model graph structures and data features, such as the classical label propagation methods, or require a considerable amount of labeled data for training and validation due to high model complexity, such as the recent neural-network-based methods. In this paper, we address label efficient semi-supervised learning from a graph filtering perspective. Specifically, we propose a graph filtering framework that injects graph similarity into data features by taking them as signals on the graph and applying a low-pass graph filter to extract useful data representations for classification, where label efficiency can be achieved by conveniently adjusting the strength of the graph filter. Interestingly, this framework unifies two seemingly very different methods -- label propagation and graph convolutional networks. Revisiting them under the graph filtering framework leads to new insights that improve their modeling capabilities and reduce model complexity. Experiments on various semi-supervised classification tasks on four citation networks and one knowledge graph and one semi-supervised regression task for zero-shot image recognition validate our findings and proposals.

['Xiao-Ming Wu', 'Zhichao Guan', 'Xiaotong Zhang', 'Qimai Li', 'Han Liu']

2019-01-28

label-efficient-semi-supervised-learning-via

http://openaccess.thecvf.com/content_CVPR_2019/html/Li_Label_Efficient_Semi-Supervised_Learning_via_Graph_Filtering_CVPR_2019_paper.html

http://openaccess.thecvf.com/content_CVPR_2019/papers/Li_Label_Efficient_Semi-Supervised_Learning_via_Graph_Filtering_CVPR_2019_paper.pdf

cvpr-2019-6

['graph-similarity']

['graphs']

[ 4.54460591e-01 4.45585757e-01 -4.71502244e-01 -4.51452315e-01 -1.50784358e-01 -6.82993114e-01 7.84139931e-01 7.30034053e-01 -2.28300378e-01 4.46854293e-01 -8.62615854e-02 -3.41058075e-01 -3.52432191e-01 -1.05888844e+00 -5.09360909e-01 -6.14648700e-01 -1.30673930e-01 4.40493315e-01 3.62558335e-01 -1.12938425e-02 2.25748166e-01 5.33713639e-01 -1.59529293e+00 5.62634394e-02 9.17867243e-01 9.18610334e-01 -6.12011813e-02 4.77192998e-01 -5.38937032e-01 1.00809216e+00 -6.45481497e-02 -4.93158609e-01 1.52043045e-01 -5.38144529e-01 -9.65951025e-01 3.30612838e-01 5.02732158e-01 3.49013478e-01 -5.50072551e-01 1.29108417e+00 1.91703647e-01 1.67034134e-01 7.58897066e-01 -1.09686327e+00 -6.34172559e-01 7.02027977e-01 -3.43744904e-01 7.75081366e-02 1.70472950e-01 -1.43350393e-01 1.17315412e+00 -5.69209099e-01 8.34930837e-01 1.09345388e+00 7.37564266e-01 4.46348518e-01 -1.49427831e+00 -4.10893708e-01 2.72717804e-01 2.22957194e-01 -1.10507941e+00 -2.87045658e-01 9.79083717e-01 -6.38490856e-01 6.91370606e-01 7.56017789e-02 7.70439625e-01 7.47778654e-01 -8.99479762e-02 7.41168797e-01 1.29982352e+00 -6.42399669e-01 3.16762239e-01 1.09988086e-01 7.18741834e-01 1.32910788e+00 3.32509398e-01 1.58396572e-01 -4.93598789e-01 -4.08828557e-02 5.87090671e-01 2.04604745e-01 -3.44480842e-01 -6.56054914e-01 -9.80845213e-01 9.38691676e-01 7.80678213e-01 4.47536141e-01 -1.14019915e-01 4.62895678e-03 2.80544162e-01 5.68792403e-01 8.98287475e-01 5.10509014e-01 -6.34573027e-02 5.00510097e-01 -1.01636767e+00 -3.04175287e-01 9.12653685e-01 8.35939467e-01 1.18306267e+00 -6.68564215e-02 -2.28025585e-01 7.16206491e-01 5.49608648e-01 -2.95259729e-02 2.72881508e-01 -4.39233571e-01 8.37026164e-02 1.19250393e+00 -5.29637694e-01 -1.02881420e+00 -6.75069690e-01 -7.66367793e-01 -8.02906692e-01 2.41112694e-01 5.63155711e-01 2.25173011e-01 -1.17613900e+00 1.51529312e+00 2.06166193e-01 3.36813509e-01 -2.33484164e-01 6.94586158e-01 1.09838617e+00 2.26623729e-01 1.08831935e-01 -3.48860353e-01 1.19648623e+00 -1.10022342e+00 -6.04319930e-01 -2.43636265e-01 1.09386098e+00 -2.94637412e-01 9.22134876e-01 2.15206087e-01 -7.58322179e-01 -3.41258764e-01 -1.04878891e+00 2.20132262e-01 -8.54741454e-01 -1.02188051e-01 9.52984095e-01 6.73555315e-01 -1.01776004e+00 1.05043924e+00 -7.82112300e-01 -6.85801208e-01 7.86890984e-01 4.45679635e-01 -3.25458169e-01 -3.15084010e-01 -9.66769040e-01 6.84771717e-01 5.07823527e-01 -9.23478901e-02 -5.90068579e-01 -3.59530956e-01 -8.63541901e-01 3.50094378e-01 5.83817303e-01 -5.98710775e-01 7.95512259e-01 -1.09447742e+00 -1.39606428e+00 1.19031036e+00 2.24078327e-01 -6.46785319e-01 2.86032975e-01 3.40619206e-01 -2.59945869e-01 3.12357605e-01 -1.46077633e-01 4.56954360e-01 9.76342559e-01 -1.04452634e+00 -3.89357805e-01 -4.97901022e-01 1.19676791e-01 -1.51805710e-02 -6.43283069e-01 -2.04594240e-01 -3.76533836e-01 -6.71763659e-01 3.97303402e-01 -9.35100555e-01 -2.88351029e-01 -1.83652371e-01 -5.89173734e-01 -2.61756569e-01 6.42345607e-01 -1.10889874e-01 1.26624668e+00 -1.89360213e+00 2.32031479e-01 4.37484264e-01 8.31728160e-01 4.18242902e-01 -1.63364723e-01 4.53745812e-01 -7.37727582e-02 1.69962481e-01 -3.42398584e-01 -2.96977192e-01 -1.52410984e-01 1.26185387e-01 4.90223579e-02 5.77495575e-01 2.90665269e-01 1.18265045e+00 -1.13056374e+00 -5.31306982e-01 2.34671161e-01 3.33685279e-01 -2.27129206e-01 1.70587808e-01 -3.58083278e-01 4.91681576e-01 -3.96601200e-01 4.89645749e-01 4.36550379e-01 -7.29727983e-01 3.77746433e-01 -2.22101375e-01 1.99767515e-01 1.83746025e-01 -1.00447190e+00 1.54563522e+00 -1.48037434e-01 5.57158172e-01 -1.59721226e-01 -1.60233581e+00 9.35694993e-01 1.87135246e-02 5.12270451e-01 -5.61895609e-01 2.61210948e-01 1.18066825e-01 -2.36603677e-01 -2.34456643e-01 8.73048510e-03 -1.75653338e-01 2.88264483e-01 6.38768971e-01 5.56892574e-01 5.48007414e-02 4.88597214e-01 5.25232792e-01 1.29450488e+00 -1.40490845e-01 2.78740287e-01 -3.35071236e-01 5.95628262e-01 -3.05773057e-02 1.58424065e-01 9.30304468e-01 6.17935956e-02 4.69774246e-01 4.90292281e-01 -3.98825973e-01 -4.36918825e-01 -8.54528069e-01 1.01146944e-01 1.17680085e+00 4.41157296e-02 -6.51019573e-01 -6.62097633e-01 -1.29085267e+00 -4.43237349e-02 2.97563165e-01 -6.65002465e-01 -4.78744805e-01 -1.94618136e-01 -6.85980737e-01 3.21705252e-01 3.29725295e-01 2.28973851e-01 -9.08343256e-01 3.52996737e-02 1.19993873e-01 3.53897601e-01 -9.55132663e-01 -2.54341990e-01 5.35997987e-01 -1.06324613e+00 -1.51085067e+00 -4.41264898e-01 -1.04514730e+00 1.05603826e+00 4.72943842e-01 1.17491221e+00 5.23900867e-01 -3.32517117e-01 6.39731586e-01 -4.65219021e-01 -1.64065748e-01 -5.08926153e-01 2.59521931e-01 -2.12965682e-01 3.39257240e-01 3.77943546e-01 -6.03433788e-01 -2.92065293e-01 7.79116005e-02 -8.23179126e-01 2.97993366e-02 5.61739802e-01 9.67359781e-01 6.61780000e-01 2.18206421e-01 6.49762273e-01 -1.61169600e+00 6.48559868e-01 -3.69777650e-01 -6.93472564e-01 4.10265595e-01 -1.17163372e+00 3.44723701e-01 7.40041912e-01 -3.78953665e-01 -7.55223215e-01 2.41567269e-01 2.13416204e-01 -4.26887721e-01 -1.03334114e-01 8.66056025e-01 6.75332174e-02 -6.27597272e-01 7.46093631e-01 1.45472154e-01 1.46389455e-01 -4.61208075e-01 7.00910091e-01 4.04592395e-01 2.75927663e-01 -1.21296152e-01 9.01986718e-01 5.79196870e-01 3.77385736e-01 -7.68229067e-01 -1.18536222e+00 -6.49160922e-01 -9.61263537e-01 -3.62024486e-01 8.53134334e-01 -5.54857552e-01 -5.94204485e-01 3.94541711e-01 -6.93768024e-01 -3.38308960e-01 -3.63760591e-01 5.44331491e-01 -4.39957380e-01 5.29256761e-01 -6.51605129e-01 -6.34298027e-01 -2.21874684e-01 -8.42938662e-01 7.51581788e-01 2.67012805e-01 2.20983192e-01 -1.47371840e+00 5.26291355e-02 1.82311952e-01 3.90914619e-01 1.38989747e-01 1.11357367e+00 -1.00468993e+00 -5.19688904e-01 -4.23796743e-01 -4.50857311e-01 2.95377731e-01 1.72215328e-01 -1.32818460e-01 -9.27724838e-01 -4.48966116e-01 -3.88264358e-01 -4.47257996e-01 1.33309281e+00 2.80162185e-01 9.37040508e-01 -1.62786633e-01 -6.19040966e-01 6.49039209e-01 1.29753184e+00 -3.59153211e-01 1.34855017e-01 -2.41679419e-02 1.04825842e+00 6.80474281e-01 2.25083187e-01 2.43222024e-02 1.35891974e-01 4.64910388e-01 4.85479116e-01 -1.53713673e-01 -4.36766773e-01 -2.19976082e-01 -3.69164087e-02 9.81577098e-01 -8.52889270e-02 -2.26166502e-01 -8.58964801e-01 1.78207204e-01 -2.01200032e+00 -7.51121104e-01 -5.21319330e-01 2.23821068e+00 5.74110448e-01 4.39797997e-01 1.20164797e-01 2.11470008e-01 7.67439008e-01 2.96915472e-01 -4.49755251e-01 2.83123031e-02 -9.74123627e-02 3.66084963e-01 4.90235955e-01 3.70740801e-01 -1.25058460e+00 1.03146935e+00 5.95499754e+00 6.44971311e-01 -1.11478293e+00 1.17409369e-03 3.81250024e-01 2.20316038e-01 -2.17559144e-01 2.19311103e-01 -5.21287680e-01 1.10105246e-01 9.61273074e-01 -1.45295277e-01 5.80504596e-01 8.13769698e-01 -1.23301439e-01 1.54097572e-01 -1.23444128e+00 8.72028053e-01 7.99441785e-02 -1.62715876e+00 1.43123537e-01 -2.13745097e-03 5.33820212e-01 1.54133633e-01 -2.36629859e-01 2.76292473e-01 4.27740693e-01 -1.02186739e+00 3.53222042e-01 6.00805521e-01 5.12079179e-01 -5.11396110e-01 5.01470268e-01 3.98718417e-01 -1.34060490e+00 -1.01452284e-01 -3.06828976e-01 -1.22499689e-01 -9.14227441e-02 7.92154968e-01 -7.06392348e-01 6.17510200e-01 2.96409100e-01 1.09360600e+00 -1.05803490e+00 1.01808989e+00 -3.61384422e-01 9.10789669e-01 -1.11524828e-01 -1.02576219e-01 2.42223978e-01 -3.74391824e-01 4.18310851e-01 1.27802849e+00 -9.98014286e-02 -1.38968304e-01 4.60894704e-01 9.02223527e-01 -3.71541560e-01 2.96098769e-01 -8.01186740e-01 -5.36584973e-01 1.99166194e-01 1.52393413e+00 -1.42559695e+00 -3.45347315e-01 -6.59356356e-01 7.16830075e-01 8.35786581e-01 3.15272808e-01 -4.08383787e-01 -5.35256445e-01 -6.96692988e-02 2.16280177e-01 6.97431713e-02 -2.88920045e-01 -1.14706919e-01 -1.32285643e+00 -4.63544309e-01 -5.44287860e-01 5.94485760e-01 -3.23489368e-01 -1.47769034e+00 4.32266563e-01 -2.75383264e-01 -9.66513276e-01 5.38312383e-02 -6.86837375e-01 -6.11135662e-01 4.73316818e-01 -1.66063964e+00 -1.26637685e+00 -4.78005081e-01 6.07415140e-01 1.84853643e-01 -1.64225474e-01 7.77976930e-01 2.53901064e-01 -5.25552928e-01 3.60185146e-01 1.28761396e-01 1.68059200e-01 5.22987127e-01 -1.48162985e+00 2.67990112e-01 7.82194436e-01 7.77250469e-01 5.65918803e-01 2.40344599e-01 -5.92285395e-01 -1.53312480e+00 -1.20979238e+00 7.66170800e-01 -1.58149853e-01 8.00231099e-01 -5.97035885e-01 -1.20746350e+00 5.81677318e-01 -1.65385023e-01 5.71711242e-01 7.48394668e-01 1.96154118e-01 -4.87520009e-01 1.58389837e-01 -9.16328013e-01 4.43897218e-01 1.31445730e+00 -7.49614537e-01 -3.13097596e-01 6.63195789e-01 6.00572705e-01 8.77250433e-02 -8.46446991e-01 3.29574585e-01 2.41421372e-01 -9.31928515e-01 8.01310539e-01 -9.30700541e-01 5.15804850e-02 -1.70265242e-01 2.48033836e-01 -1.34854758e+00 -5.84219754e-01 -6.18975520e-01 -3.95564377e-01 1.21054900e+00 3.80843937e-01 -7.49400854e-01 1.01635063e+00 1.31927192e-01 1.32384924e-02 -7.14033842e-01 -4.85984296e-01 -7.85942256e-01 -2.56895661e-01 -3.31708997e-01 1.96656540e-01 1.29542840e+00 2.59722412e-01 7.79923975e-01 -4.49557267e-02 -1.52856726e-02 8.45988691e-01 2.37282127e-01 5.26818752e-01 -1.90034974e+00 -2.55071133e-01 -6.46891356e-01 -7.17861354e-01 -7.76879907e-01 5.62563837e-01 -1.66935802e+00 -3.03752393e-01 -1.76905394e+00 2.73424666e-02 -4.34835315e-01 -4.79079098e-01 6.29812777e-01 -1.88514978e-01 3.96797031e-01 4.20266837e-02 2.60130256e-01 -8.32432687e-01 3.09114128e-01 9.73250091e-01 -3.14724624e-01 -2.20912158e-01 1.26714110e-01 -5.01174748e-01 7.94546843e-01 5.34958601e-01 -5.87139130e-01 -6.52475595e-01 -8.90586227e-02 3.33660424e-01 -1.36477530e-01 2.60499775e-01 -9.80756521e-01 4.74237204e-01 1.02520511e-01 1.20013244e-02 -1.42204925e-01 -1.41491950e-01 -6.83325052e-01 -1.20494835e-01 5.80718398e-01 -5.83535552e-01 -5.81794381e-01 -2.72858679e-01 1.12259746e+00 -2.37277538e-01 -2.94130355e-01 9.51327622e-01 -2.65381455e-01 -6.33475363e-01 5.25191307e-01 -2.35185847e-01 -5.08033251e-03 9.14897025e-01 -1.91328034e-01 -3.31509918e-01 -2.56750792e-01 -1.06365454e+00 1.15622349e-01 3.68356973e-01 2.80854195e-01 4.77360815e-01 -1.19786823e+00 -4.18061346e-01 2.50676811e-01 3.15897822e-01 -2.89651334e-01 -7.98074976e-02 1.05190516e+00 -2.04438478e-01 3.02088350e-01 -1.57321021e-02 -5.52076101e-01 -1.28465390e+00 8.93896937e-01 2.57440209e-01 -5.15290380e-01 -6.47331834e-01 8.04685652e-01 5.83900437e-02 -5.19092143e-01 3.08157086e-01 -9.83187482e-02 -4.66617405e-01 3.48727018e-01 1.91437781e-01 3.26592743e-01 2.22988635e-01 -5.53347409e-01 -3.02466720e-01 6.01123869e-01 -9.86521989e-02 4.94863570e-01 1.40590394e+00 -1.03035316e-01 -2.59758055e-01 4.82008308e-01 1.25886357e+00 -2.35687837e-01 -9.57806110e-01 -6.84591830e-01 6.15111232e-01 -1.77930564e-01 3.01786661e-01 -4.68063295e-01 -1.24810863e+00 9.56740201e-01 3.92236590e-01 7.62290061e-01 9.86845493e-01 2.76382774e-01 3.39336634e-01 6.19395137e-01 3.41356754e-01 -8.66909802e-01 2.36360610e-01 3.44965190e-01 3.31354111e-01 -1.43057287e+00 1.68366536e-01 -7.19313443e-01 -2.13252708e-01 1.34325874e+00 2.48423070e-01 -2.45894462e-01 9.88224089e-01 -6.11392856e-02 -2.21695662e-01 -7.87420988e-01 -5.82321107e-01 -7.07178831e-01 6.74472034e-01 5.69808304e-01 5.49281001e-01 -1.19309323e-02 -3.76991272e-01 4.54759210e-01 2.39130124e-01 4.69130836e-02 3.49573195e-01 8.92553627e-01 -6.11979604e-01 -1.05203581e+00 2.41096094e-01 9.19606984e-01 -2.62583941e-01 -6.75464645e-02 -7.40527630e-01 4.93285149e-01 -2.52790242e-01 9.85853195e-01 -2.50484318e-01 -5.67450047e-01 1.52439967e-01 3.05438370e-01 5.08286655e-01 -1.13293231e+00 -6.15851045e-01 -2.08536431e-01 9.87100825e-02 -5.41278720e-01 -7.40015686e-01 -1.57056510e-01 -1.23497188e+00 -5.78761008e-03 -8.66728246e-01 1.84681460e-01 4.64077830e-01 1.03606343e+00 3.53252172e-01 4.89152104e-01 5.47722161e-01 -7.35266447e-01 -3.19242626e-01 -7.97204018e-01 -8.31178188e-01 5.71937442e-01 -2.12709680e-02 -7.32904911e-01 -4.96216297e-01 -3.42735425e-02]

[7.2489800453186035, 6.23140287399292]

b3b2dab6-e2b6-4f6f-aba2-afc706f1ae58

adaptive-selection-of-informative-path

2108.06618

https://arxiv.org/abs/2108.06618v1

https://arxiv.org/pdf/2108.06618v1.pdf

Adaptive Selection of Informative Path Planning Strategies via Reinforcement Learning

In our previous work, we designed a systematic policy to prioritize sampling locations to lead significant accuracy improvement in spatial interpolation by using the prediction uncertainty of Gaussian Process Regression (GPR) as "attraction force" to deployed robots in path planning. Although the integration with Traveling Salesman Problem (TSP) solvers was also shown to produce relatively short travel distance, we here hypothesise several factors that could decrease the overall prediction precision as well because sub-optimal locations may eventually be included in their paths. To address this issue, in this paper, we first explore "local planning" approaches adopting various spatial ranges within which next sampling locations are prioritized to investigate their effects on the prediction performance as well as incurred travel distance. Also, Reinforcement Learning (RL)-based high-level controllers are trained to adaptively produce blended plans from a particular set of local planners to inherit unique strengths from that selection depending on latest prediction states. Our experiments on use cases of temperature monitoring robots demonstrate that the dynamic mixtures of planners can not only generate sophisticated, informative plans that a single planner could not create alone but also ensure significantly reduced travel distances at no cost of prediction reliability without any assist of additional modules for shortest path calculation.

['Grzegorz Cielniak', 'Taeyeong Choi']

2021-08-14

['gpr', 'gpr']

['computer-vision', 'miscellaneous']

[ 2.56716073e-01 5.46392918e-01 -5.90614751e-02 -2.35046551e-01 -9.60366607e-01 -4.55519617e-01 5.55707872e-01 2.07410142e-01 -3.48419040e-01 1.11472964e+00 -7.80919120e-02 -4.41433579e-01 -7.59933829e-01 -1.01755667e+00 -7.28965878e-01 -7.09745526e-01 -4.90725249e-01 9.37141359e-01 3.90241206e-01 -3.19064170e-01 6.40828013e-01 6.52858913e-01 -1.54502845e+00 -1.27178296e-01 1.31172466e+00 4.92448092e-01 7.06042349e-01 4.71482098e-01 1.62181735e-01 5.42833209e-01 -5.95584571e-01 4.87496734e-01 3.97063226e-01 -1.92366652e-02 -7.08380222e-01 -2.61685014e-01 -3.81293833e-01 -3.58916938e-01 2.00548753e-01 5.30232966e-01 3.59477073e-01 4.95785326e-01 7.60137141e-01 -1.33219385e+00 -1.13163427e-01 9.13128912e-01 -4.47460830e-01 -1.99699283e-01 4.21708971e-01 7.14672923e-01 5.05496204e-01 -2.78027773e-01 3.95845324e-01 1.28141642e+00 6.73401177e-01 1.57135040e-01 -1.53098202e+00 -4.90921021e-01 3.01873118e-01 -1.89019263e-01 -1.63033175e+00 -2.33971804e-01 4.64626521e-01 -1.71770766e-01 1.22936058e+00 8.85146260e-02 4.39852923e-01 7.01817811e-01 7.64982402e-01 2.48198003e-01 9.17138338e-01 -2.16678068e-01 5.80327570e-01 2.77110729e-02 -5.55981100e-01 4.55475599e-01 2.18992770e-01 3.51114243e-01 -1.85381994e-01 -2.16004461e-01 8.94929886e-01 -4.81144249e-01 -2.07835898e-01 -3.71648669e-01 -9.97308135e-01 8.61500680e-01 4.73680764e-01 2.98133586e-02 -7.87941575e-01 5.05203843e-01 -1.03444993e-01 -6.42127320e-02 5.28016239e-02 9.84956086e-01 -5.05528033e-01 -3.72505277e-01 -6.77597940e-01 7.61708260e-01 8.89811099e-01 1.18514192e+00 9.67421234e-01 2.05833260e-02 -8.77288952e-02 4.33754414e-01 4.54465896e-01 4.71070558e-01 5.37177920e-02 -1.42119467e+00 7.38125503e-01 4.05453950e-01 7.99873352e-01 -9.48435366e-01 -7.83426940e-01 -9.88853872e-02 -1.63139150e-01 5.75646579e-01 4.88533199e-01 -6.66706562e-01 -9.45973158e-01 1.52991581e+00 1.98925838e-01 -2.48752430e-01 6.65129162e-03 8.21330667e-01 -4.88423258e-01 9.56787705e-01 2.27707639e-01 -1.17639441e-03 7.17086434e-01 -7.52561092e-01 -3.71484309e-01 -3.16189229e-01 6.13012493e-01 -4.44020301e-01 8.58865023e-01 5.46049893e-01 -1.06785178e+00 -4.21688706e-01 -9.50828135e-01 4.76017535e-01 -2.67181784e-01 -2.31515244e-01 5.12624979e-01 5.60640454e-01 -1.23078108e+00 1.14112437e+00 -1.07959461e+00 -4.67838973e-01 6.91816509e-02 6.03568316e-01 1.99276805e-01 1.84255779e-01 -9.98685598e-01 1.49456728e+00 2.80476749e-01 1.55567139e-01 -1.17178071e+00 -6.85731113e-01 -6.11972034e-01 8.83462280e-02 3.83454412e-01 -4.37097728e-01 1.33289504e+00 -5.11227846e-01 -1.76863587e+00 -4.32241946e-01 -1.87259674e-01 -5.42278349e-01 4.59167391e-01 -3.32323201e-02 -1.89223811e-02 -7.47294724e-02 1.38092950e-01 1.15135491e+00 4.30601746e-01 -1.46872473e+00 -1.08742929e+00 -7.69981295e-02 1.55360356e-01 5.84070981e-01 2.19830170e-01 -3.63228083e-01 1.81129292e-01 1.00163579e-01 2.12848872e-01 -1.10281503e+00 -1.19531631e+00 -3.36659342e-01 -2.85168946e-01 -2.84403145e-01 4.46677834e-01 -2.94773936e-01 8.73739600e-01 -1.51241541e+00 8.65435451e-02 5.62875986e-01 -5.86527646e-01 -3.99895787e-01 -1.48832053e-01 8.53096485e-01 5.00405312e-01 1.54705897e-01 -2.36022577e-01 -1.78766474e-01 2.97992341e-02 4.91331786e-01 -3.11717689e-01 3.86217982e-01 4.75867331e-01 3.58092576e-01 -1.15776408e+00 -2.49987975e-01 3.70164543e-01 2.58713752e-01 -4.48197871e-01 -1.17268197e-01 -5.36059141e-01 7.11892247e-01 -6.47576928e-01 4.90275919e-01 5.41905701e-01 1.73131153e-01 1.54125765e-01 4.00353670e-01 -6.76402330e-01 5.06087065e-01 -1.29761362e+00 1.53850698e+00 -6.54827178e-01 1.18716344e-01 1.96584985e-01 -5.73694408e-01 1.21254003e+00 1.05087280e-01 5.57979465e-01 -5.42922020e-01 -1.05691195e-01 3.85884225e-01 -1.99724901e-02 -3.40759575e-01 1.07584584e+00 3.77203450e-02 -1.36566415e-01 2.23054618e-01 -6.36801541e-01 -6.58420920e-01 2.00557280e-02 -4.39761907e-01 1.16347516e+00 8.27535450e-01 -2.44401544e-02 -5.95720589e-01 2.78494388e-01 8.95166814e-01 6.49460077e-01 8.05901170e-01 -2.17786670e-01 3.99758399e-01 3.94008666e-01 -2.35208005e-01 -9.43829000e-01 -1.01510859e+00 -3.20182256e-02 8.52956712e-01 6.11340582e-01 -1.07171513e-01 -4.14751083e-01 -2.03314096e-01 1.26297520e-02 1.33617270e+00 -2.14468524e-01 3.61592360e-02 -7.57454991e-01 -7.09143341e-01 5.71404219e-01 4.46277261e-01 1.37938544e-01 -9.61764693e-01 -1.19267881e+00 6.20044947e-01 2.44572423e-02 -4.80137885e-01 -1.19103648e-01 7.13472605e-01 -8.70740294e-01 -7.96153784e-01 -4.06727910e-01 -5.24211943e-01 7.63549328e-01 2.27214158e-01 6.27246559e-01 -2.50678897e-01 2.69846737e-01 2.34628916e-01 -3.26728374e-01 -5.52560449e-01 -4.71400291e-01 2.87682354e-01 -3.72737944e-02 -7.16484785e-01 -7.53851011e-02 -6.30462050e-01 -5.02686620e-01 6.15033388e-01 -3.99843335e-01 5.80759086e-02 6.41571820e-01 4.36952978e-01 5.76004148e-01 6.32266581e-01 7.21964598e-01 -2.54808128e-01 9.35624361e-01 -7.26113200e-01 -9.65183139e-01 6.01053238e-02 -8.97221804e-01 4.07715589e-01 8.19007754e-01 -3.70960265e-01 -1.14345098e+00 4.49470967e-01 1.42560735e-01 -6.25629649e-02 -3.62639278e-01 6.11398876e-01 4.78521436e-02 1.08521851e-02 6.85405433e-01 9.90464464e-02 1.65200204e-01 -1.00280128e-01 3.38134706e-01 2.82158881e-01 5.13045639e-02 -9.39639747e-01 5.77702940e-01 2.59677768e-01 2.33509630e-01 -5.16221642e-01 2.95602351e-01 -1.49919331e-01 -4.07210916e-01 -1.53645098e-01 6.59264445e-01 -6.23274863e-01 -8.40749562e-01 -7.50747919e-02 -1.21484458e+00 -8.35425973e-01 -1.03345081e-01 2.27332339e-01 -1.01007342e+00 -1.29896522e-01 -3.40501040e-01 -1.27315986e+00 5.97370863e-02 -1.61417568e+00 8.42697084e-01 5.45126379e-01 -5.38745463e-01 -7.62463033e-01 5.05775921e-02 -2.70189404e-01 7.74390221e-01 3.73710543e-01 7.04298735e-01 -2.39499167e-01 -1.04896259e+00 1.61146596e-01 1.72263041e-01 -3.68333936e-01 3.71676050e-02 5.22082560e-02 -6.50569737e-01 -1.03578143e-01 -3.90005976e-01 1.55508935e-01 1.79614395e-01 5.67257404e-01 5.76806128e-01 -3.39432210e-01 -7.67904103e-01 7.58297667e-02 1.59245276e+00 8.13941121e-01 7.32528746e-01 7.36195087e-01 2.28574336e-01 1.09991550e+00 1.15163529e+00 4.64993030e-01 4.83861595e-01 7.94824839e-01 7.66355872e-01 3.74861002e-01 5.48050821e-01 -4.93518531e-01 4.77233678e-01 -2.24041775e-01 -2.76065618e-01 -7.12841451e-02 -1.11578786e+00 6.82881534e-01 -2.01946926e+00 -8.13338220e-01 -1.89386010e-01 2.35992932e+00 5.74167848e-01 2.47542202e-01 1.19200207e-01 -1.07744619e-01 5.95909595e-01 -2.42714241e-01 -3.68697643e-01 -5.96163094e-01 4.15839821e-01 -2.05172319e-02 1.37289524e+00 8.33546698e-01 -8.12008798e-01 8.04013729e-01 6.16510582e+00 6.44593358e-01 -9.17392671e-01 -3.77890438e-01 4.98267114e-01 2.32075080e-02 -4.04063582e-01 3.59073728e-01 -7.66789436e-01 4.68737006e-01 1.41990960e+00 -1.37482077e-01 7.36425877e-01 1.00278485e+00 9.98005271e-01 -6.44650519e-01 -8.80553603e-01 9.64087062e-03 -7.28512704e-01 -1.15067971e+00 -2.55670726e-01 2.49172255e-01 5.99658191e-01 -1.11020925e-02 7.12790526e-03 3.84764373e-01 8.01701844e-01 -1.15205026e+00 1.04386079e+00 5.15952826e-01 1.07081063e-01 -1.16189289e+00 5.48921764e-01 7.65609384e-01 -1.17615867e+00 -4.37509507e-01 -5.04348218e-01 -3.53732079e-01 8.08996558e-01 6.59258664e-02 -1.62738335e+00 7.97082186e-01 5.07055759e-01 9.93312104e-04 -1.23448491e-01 1.31498969e+00 -2.74363190e-01 2.20371217e-01 -6.58808410e-01 -4.23913687e-01 5.97518086e-01 -2.33892605e-01 6.45793438e-01 8.17931652e-01 6.52827263e-01 -2.33214065e-01 2.09162965e-01 1.23832691e+00 1.03930199e+00 -2.86694795e-01 -7.06068575e-01 3.92482787e-01 8.95652831e-01 8.11355293e-01 -7.84074366e-01 1.82041273e-01 1.95963487e-01 4.73899633e-01 2.95268167e-02 4.16803867e-01 -1.11961794e+00 -3.22328925e-01 6.74143136e-01 3.42407703e-01 2.31779873e-01 -6.15798712e-01 -6.79458559e-01 -6.11493140e-02 -4.61018026e-01 -4.19047087e-01 -2.19861969e-01 -8.00471425e-01 -7.60894001e-01 3.80591899e-01 4.32115287e-01 -1.13165379e+00 -5.64541519e-01 -2.32154429e-01 -7.17371047e-01 1.30847478e+00 -1.46938360e+00 -8.38859379e-01 -3.00982036e-02 1.02922328e-01 6.28888071e-01 2.22776771e-01 7.48471498e-01 -1.62184581e-01 -3.03924978e-01 5.39581105e-03 -6.05364926e-02 -6.31636024e-01 4.26188976e-01 -1.24402988e+00 2.85712868e-01 7.00499117e-01 -7.06730425e-01 7.72859097e-01 1.26082051e+00 -1.06049252e+00 -1.44686282e+00 -1.03680015e+00 5.80761135e-01 -3.59301180e-01 5.60577989e-01 2.02490315e-01 -6.49943173e-01 5.87883949e-01 1.06034398e-01 -8.97103846e-01 -6.79347143e-02 7.84147382e-02 5.42532384e-01 2.25599739e-03 -1.43147969e+00 9.77865279e-01 7.21567690e-01 2.11356893e-01 -2.28436619e-01 -4.56142193e-03 8.51147592e-01 -5.45528829e-01 -8.06852221e-01 4.86508131e-01 2.91475534e-01 -7.41263568e-01 8.21978509e-01 3.28662731e-02 3.16606313e-01 -6.22828782e-01 -5.69187067e-02 -1.50582397e+00 -4.15395170e-01 -7.11275578e-01 3.84369224e-01 1.20782113e+00 9.98911917e-01 -6.76191270e-01 9.69976664e-01 1.26416659e+00 -5.93195379e-01 -9.13070261e-01 -1.01714873e+00 -8.36436391e-01 1.68879986e-01 -4.51831937e-01 9.28944111e-01 2.26478130e-01 3.37106436e-01 -1.26434684e-01 -3.09372514e-01 7.50969172e-01 4.74666357e-01 -1.18238434e-01 7.78731525e-01 -8.16152155e-01 -1.86711088e-01 -5.24316669e-01 5.34707308e-02 -9.35730815e-01 -3.45616311e-01 -4.30428207e-01 9.32632089e-01 -1.83682871e+00 -6.37484550e-01 -1.23846054e+00 1.21944003e-01 3.31064910e-01 1.44220263e-01 -7.72638261e-01 8.13812837e-02 1.92909718e-01 -2.67696470e-01 4.73674774e-01 1.16338754e+00 4.17672954e-02 -9.35042083e-01 2.64090717e-01 -5.64787686e-01 3.94914001e-01 9.99252200e-01 -4.26659226e-01 -8.24484408e-01 -4.25545484e-01 1.55796364e-01 4.76644546e-01 3.77276577e-02 -1.21258795e+00 4.99677181e-01 -8.98699701e-01 4.69687134e-02 -6.88023627e-01 4.06695813e-01 -1.10667717e+00 5.40012479e-01 6.67752981e-01 -3.69925559e-01 2.71362394e-01 4.22775477e-01 8.35024357e-01 3.37557882e-01 -5.02413571e-01 4.44613636e-01 -7.15269744e-02 -7.82761991e-01 -1.88068271e-01 -1.05830455e+00 -7.72324383e-01 1.36586463e+00 -4.25070882e-01 -2.41886571e-01 -3.19222093e-01 -2.92839497e-01 7.41237342e-01 6.70069396e-01 2.22345501e-01 3.71317565e-01 -7.34512687e-01 -3.79319489e-01 -2.39704132e-01 -2.91108787e-01 4.86376464e-01 -5.78619447e-03 8.53685498e-01 -7.49468207e-01 6.63034856e-01 -2.71076500e-01 -4.23448920e-01 -3.64219129e-01 5.31058550e-01 4.47538525e-01 -3.68265510e-01 -5.79095244e-01 6.54018939e-01 -3.73202890e-01 -5.65232098e-01 2.85973191e-01 -7.19751179e-01 -6.87292293e-02 -2.65633941e-01 2.43256241e-01 6.60793841e-01 -2.72483230e-01 -8.02310482e-02 -3.65371764e-01 2.24416807e-01 4.05029714e-01 -4.62335140e-01 1.24987757e+00 -4.48926151e-01 2.27700129e-01 2.38338098e-01 3.03123772e-01 -1.57758221e-01 -1.86922789e+00 5.86860716e-01 4.77879673e-01 -3.48685175e-01 -2.59860665e-01 -8.12211275e-01 -5.00954866e-01 3.01766753e-01 2.78760552e-01 2.50315964e-01 8.04497421e-01 -2.90979058e-01 4.53709006e-01 2.92634547e-01 1.13442588e+00 -1.35149133e+00 -3.52847606e-01 4.12318081e-01 7.93107152e-01 -8.95467281e-01 -1.15826152e-01 -3.97252411e-01 -9.21884120e-01 1.07685852e+00 7.25607634e-01 -3.74452084e-01 2.59849519e-01 4.05069858e-01 -2.85135061e-01 3.89301926e-02 -7.75572062e-01 -1.14209071e-01 -4.90582436e-01 1.03899086e+00 3.69780436e-02 2.36465842e-01 -3.81855935e-01 3.74548167e-01 -1.39402360e-01 -6.37603505e-03 7.64128268e-01 1.16584802e+00 -9.55387414e-01 -1.11884725e+00 -7.82912552e-01 2.42486119e-01 1.37913287e-01 2.76497215e-01 2.73743421e-01 9.29652750e-01 2.97270596e-01 1.27108884e+00 2.20435247e-01 -4.13179189e-01 3.47691298e-01 -1.96965650e-01 2.60727972e-01 -5.18824399e-01 -5.72808743e-01 -1.38180871e-02 5.91491044e-01 -7.17519820e-01 -9.32201892e-02 -8.96078706e-01 -1.69336319e+00 -2.68047810e-01 -1.41242415e-01 2.38114059e-01 8.22672904e-01 8.76815677e-01 3.60357434e-01 5.02547979e-01 4.23612088e-01 -1.42299366e+00 -6.75196469e-01 -7.58632541e-01 -3.82603109e-01 -5.39620578e-01 -1.18046232e-01 -8.59674633e-01 -2.90628701e-01 -3.44343215e-01]

[4.830384254455566, 1.745438575744629]

e93a3294-1777-4cee-a53d-27e55ee42b78

first-place-solution-to-the-cvpr-2023-aqtc

2306.1338

https://arxiv.org/abs/2306.13380v1

https://arxiv.org/pdf/2306.13380v1.pdf

First Place Solution to the CVPR'2023 AQTC Challenge: A Function-Interaction Centric Approach with Spatiotemporal Visual-Language Alignment

Affordance-Centric Question-driven Task Completion (AQTC) has been proposed to acquire knowledge from videos to furnish users with comprehensive and systematic instructions. However, existing methods have hitherto neglected the necessity of aligning spatiotemporal visual and linguistic signals, as well as the crucial interactional information between humans and objects. To tackle these limitations, we propose to combine large-scale pre-trained vision-language and video-language models, which serve to contribute stable and reliable multimodal data and facilitate effective spatiotemporal visual-textual alignment. Additionally, a novel hand-object-interaction (HOI) aggregation module is proposed which aids in capturing human-object interaction information, thereby further augmenting the capacity to understand the presented scenario. Our method achieved first place in the CVPR'2023 AQTC Challenge, with a Recall@1 score of 78.7\%. The code is available at https://github.com/tomchen-ctj/CVPR23-LOVEU-AQTC.

['Chen Chen', 'Wei Sun', 'Wei Miao', 'Jingwen Wang', 'Zechuan Li', 'Ming Li', 'Zhengeng Yang', 'Hongshan Yu', 'Tom Tongjia Chen']

2023-06-23

['human-object-interaction-detection']

['computer-vision']

[ 1.13364249e-01 -1.92173988e-01 -1.68360211e-02 -4.16515946e-01 -9.63869929e-01 -5.63187540e-01 6.39554858e-01 6.57633021e-02 -5.20565510e-01 4.52831239e-01 4.74950999e-01 -1.28889561e-01 4.40540873e-02 6.33335188e-02 -6.30322516e-01 -2.59785146e-01 8.11376497e-02 1.11941926e-01 1.59594476e-01 -2.23788261e-01 2.73873717e-01 1.04662605e-01 -1.72889185e+00 6.65025473e-01 9.73843515e-01 8.86419654e-01 5.87409854e-01 9.07944083e-01 1.53980479e-01 7.59949267e-01 -1.22940652e-01 -4.46281433e-01 -6.03428446e-02 -2.88111985e-01 -8.66586268e-01 6.02535158e-02 8.32153618e-01 -6.00118637e-01 -4.32612777e-01 7.82778203e-01 5.70371091e-01 5.27970850e-01 5.40969133e-01 -1.34278345e+00 -6.57604098e-01 1.10090204e-01 -5.93087196e-01 3.25597793e-01 8.03015769e-01 6.15997910e-01 1.05440819e+00 -1.10071325e+00 6.91808701e-01 1.06748343e+00 5.50619438e-02 7.63989806e-01 -9.90231454e-01 -3.86452973e-01 2.40478903e-01 5.62522352e-01 -1.21333206e+00 -7.21817732e-01 6.98507905e-01 -5.71031988e-01 9.62537229e-01 4.47739452e-01 7.30997801e-01 1.51800811e+00 -2.28472024e-01 1.33234763e+00 7.58780122e-01 -4.14958060e-01 -1.27953112e-01 2.80009490e-02 5.80054298e-02 6.60679221e-01 1.49200425e-01 2.37531587e-02 -8.31862807e-01 3.62213910e-01 7.07902312e-01 -2.21883692e-02 -4.70433503e-01 -4.52909052e-01 -1.55473650e+00 3.10910195e-01 4.92018193e-01 5.59785724e-01 -5.48120260e-01 2.74854809e-01 4.34106320e-01 -6.68317303e-02 2.18573675e-01 3.02197278e-01 -1.60015434e-01 -3.84477764e-01 -7.53202438e-01 2.59342849e-01 2.99734533e-01 1.26804996e+00 3.43955189e-01 -1.21978149e-01 -6.11764491e-01 6.08139455e-01 4.15507674e-01 6.78711057e-01 5.61872981e-02 -1.24103498e+00 9.33752477e-01 4.91725087e-01 5.07664263e-01 -1.09927714e+00 -3.44809085e-01 -5.84733933e-02 -4.85998511e-01 -4.55342345e-02 6.56560659e-01 1.92161649e-01 -8.06632876e-01 1.69597816e+00 3.42595726e-01 -9.47533026e-02 -2.17295140e-01 1.45854974e+00 8.44126463e-01 3.86987835e-01 5.01458406e-01 1.13115735e-01 1.66454458e+00 -9.99885023e-01 -7.80710220e-01 -2.90472537e-01 5.75112879e-01 -7.21946001e-01 1.55681872e+00 4.15317476e-01 -1.15382504e+00 -9.26307023e-01 -7.76207387e-01 -4.28952217e-01 -1.95656881e-01 4.04311687e-01 5.73334575e-01 2.94105560e-01 -1.05476880e+00 -1.74519084e-02 -8.32034945e-01 -4.57160205e-01 5.55358171e-01 1.66695356e-01 -7.31347620e-01 -3.54150116e-01 -1.02572000e+00 9.38328505e-01 3.37866366e-01 3.97391051e-01 -1.01466167e+00 -4.65875775e-01 -9.77640986e-01 -3.50105226e-01 6.36354983e-01 -7.45499969e-01 1.29648256e+00 -8.07524264e-01 -1.13823009e+00 7.47090578e-01 -5.22115827e-01 -2.23532453e-01 5.33456206e-01 -6.93498909e-01 -3.12991589e-01 5.53292692e-01 3.13441567e-02 1.11756909e+00 6.84157908e-01 -1.41812134e+00 -6.26643836e-01 -3.33618850e-01 1.06723770e-01 5.71792364e-01 -3.05890262e-01 -8.93733129e-02 -8.31702471e-01 -4.49554563e-01 -1.07410006e-01 -8.01738977e-01 4.46014572e-03 7.30457827e-02 -3.07521701e-01 -3.51649135e-01 5.77895880e-01 -1.07394230e+00 1.01305270e+00 -2.19268799e+00 4.61638391e-01 -1.61863580e-01 3.57713223e-01 4.11780655e-01 -4.18869823e-01 4.13013667e-01 7.07116872e-02 -1.28074750e-01 -4.69481945e-02 -7.81517088e-01 9.23300683e-02 -9.15988833e-02 -1.49597824e-01 4.15749907e-01 3.07787299e-01 1.42615747e+00 -1.13122439e+00 -5.77815175e-01 6.90713465e-01 6.52483582e-01 -5.10227203e-01 5.41369081e-01 -4.50250924e-01 1.11549628e+00 -4.42053854e-01 7.28850901e-01 3.21311623e-01 -2.35962361e-01 -8.79644975e-02 -4.13099647e-01 -1.10769235e-01 1.31593989e-02 -9.18974161e-01 2.25459909e+00 -2.54388541e-01 7.44575620e-01 -4.17814068e-02 -7.53189385e-01 4.49651062e-01 4.13185239e-01 5.14254749e-01 -1.21440411e+00 2.67024666e-01 -2.75001936e-02 -1.69095382e-01 -1.13718557e+00 5.22175133e-01 4.78502274e-01 -4.30818051e-02 1.59105331e-01 1.62588224e-01 -3.68834808e-02 3.35005224e-01 4.16318089e-01 7.45596826e-01 7.04154432e-01 1.19718462e-01 1.55091941e-01 6.36710525e-01 -5.49242226e-03 -1.83354691e-02 4.82190937e-01 -6.38235390e-01 8.39738548e-01 9.86584425e-02 -2.55060971e-01 -9.69717085e-01 -9.28477526e-01 4.49679881e-01 1.12098932e+00 1.77710831e-01 -3.28234166e-01 -5.75549483e-01 -4.90242958e-01 -3.33560795e-01 9.08850729e-01 -5.45683861e-01 5.87055683e-02 -5.66199303e-01 8.53644907e-02 2.96549171e-01 4.71533954e-01 5.15360892e-01 -1.29227650e+00 -8.10006142e-01 -9.49277356e-02 -1.04008269e+00 -1.47632337e+00 -8.52223635e-01 -4.62904155e-01 -4.88213688e-01 -1.19043851e+00 -8.62268746e-01 -5.91914296e-01 6.50620401e-01 5.49895048e-01 9.36266840e-01 -1.40191475e-02 -4.71037388e-01 8.34154248e-01 -4.37801301e-01 -1.63735524e-01 -7.85048865e-03 -9.79460403e-02 1.59465231e-03 1.70469865e-01 4.57676679e-01 -2.84229904e-01 -9.46910441e-01 1.69829190e-01 -6.77291989e-01 4.13092673e-01 6.67022109e-01 4.97895122e-01 3.72120917e-01 -7.39521325e-01 2.76757896e-01 -1.78121641e-01 4.44690257e-01 -1.95161179e-01 -3.67482245e-01 4.34332103e-01 -8.69119465e-02 -1.95980474e-01 1.01908818e-01 -5.63580811e-01 -1.20018756e+00 2.67183125e-01 -8.76302868e-02 -6.32701457e-01 -5.38920343e-01 2.57472485e-01 -2.21558586e-01 1.40739918e-01 5.21430910e-01 2.43334547e-01 -8.35515335e-02 -2.88074464e-01 8.56370568e-01 6.50360227e-01 7.86705256e-01 -5.86110711e-01 4.54032719e-01 4.67700958e-01 -3.04515719e-01 -7.19754755e-01 -7.34291196e-01 -9.09762323e-01 -1.03363574e+00 -6.20127916e-01 1.16251040e+00 -1.00876927e+00 -1.05027771e+00 2.64992714e-01 -1.37126422e+00 -3.83770883e-01 -2.55689155e-02 7.04760134e-01 -7.27535903e-01 5.11201918e-01 -2.45825365e-01 -1.06299615e+00 -1.91250399e-01 -1.00747144e+00 1.24904585e+00 2.29295835e-01 -4.12910640e-01 -5.67969084e-01 -8.79486501e-02 8.94609392e-01 2.25210890e-01 1.89075917e-01 3.32739890e-01 -4.52222198e-01 -7.08695412e-01 -1.34729639e-01 -4.53376263e-01 1.99375361e-01 -6.82565616e-03 -3.08038890e-01 -8.58676314e-01 -2.63530523e-01 -1.21277101e-01 -5.71655393e-01 6.34266973e-01 2.58412540e-01 8.42782974e-01 -7.17233494e-02 -1.32788152e-01 1.52431712e-01 1.08715653e+00 1.57869428e-01 6.86608016e-01 -2.96495426e-02 1.00688100e+00 8.97635698e-01 8.79880667e-01 1.85918793e-01 8.20576727e-01 1.02950835e+00 4.77081031e-01 -6.80447817e-02 -3.71900916e-01 -3.02146971e-01 3.80274922e-01 6.77589536e-01 -2.41386384e-01 -2.93633491e-01 -9.76865530e-01 7.13910460e-01 -2.00337267e+00 -8.37673903e-01 -4.70927685e-01 2.00128913e+00 6.59621954e-01 -1.42196685e-01 3.03650618e-01 -1.32898495e-01 4.81554687e-01 4.50775810e-02 -3.99986565e-01 5.90052791e-02 8.81830454e-02 -3.87500644e-01 1.14279523e-01 6.11420810e-01 -1.11829865e+00 1.05153906e+00 5.03218079e+00 5.32347500e-01 -8.50610316e-01 2.82342464e-01 2.35793889e-01 -2.62639314e-01 -3.29610556e-02 -2.48106256e-01 -5.37777066e-01 3.24112833e-01 8.72629762e-01 3.59043479e-01 4.96995866e-01 3.74620020e-01 5.56777835e-01 -4.19673502e-01 -1.13875782e+00 1.18189299e+00 4.01457131e-01 -7.48769462e-01 -5.20933494e-02 -1.33269265e-01 3.72624755e-01 -1.26709361e-02 1.23949125e-01 2.16939718e-01 -1.96110830e-01 -8.75369191e-01 8.52387071e-01 1.00620306e+00 8.40332925e-01 -3.52155179e-01 4.79904354e-01 2.98128486e-01 -1.33155763e+00 -4.00696583e-02 2.11477190e-01 3.64471860e-02 4.13684875e-01 -4.34242748e-02 -7.06288695e-01 7.48184323e-01 8.01080406e-01 6.11915708e-01 -7.08417296e-01 1.04760158e+00 -2.28045151e-01 3.04962456e-01 -1.17225207e-01 9.19632707e-03 1.35293677e-01 2.15235930e-02 6.54142797e-01 1.12374544e+00 6.91296235e-02 2.91108251e-01 1.07930258e-01 7.11202919e-01 2.28011429e-01 1.52530104e-01 -6.68141305e-01 -2.90001899e-01 1.89968735e-01 1.11883545e+00 -6.16685987e-01 -1.68214828e-01 -6.53226852e-01 1.21357000e+00 4.30207193e-01 6.36711836e-01 -9.01485145e-01 -1.71131149e-01 4.70333308e-01 -1.09951496e-02 2.46093526e-01 -8.40870798e-01 -1.02107629e-01 -1.32364452e+00 4.16427553e-01 -8.16736519e-01 2.73866475e-01 -1.19191658e+00 -9.10990655e-01 5.15592456e-01 1.78227812e-01 -1.42209828e+00 -2.04818398e-01 -7.05061853e-01 -5.84139377e-02 7.37772107e-01 -1.26471984e+00 -1.64141059e+00 -6.36933982e-01 7.54795671e-01 7.78228521e-01 1.15083113e-01 5.53081274e-01 5.01595259e-01 -3.67298633e-01 3.91591817e-01 -5.32068491e-01 1.52403805e-02 8.18076074e-01 -1.12848449e+00 1.46763533e-01 9.27222133e-01 3.13554972e-01 7.32891321e-01 8.18255126e-01 -6.71379805e-01 -1.65779853e+00 -8.46742213e-01 9.44499373e-01 -1.17913353e+00 4.43854183e-01 -5.83415389e-01 -8.63256156e-01 5.59111297e-01 3.99171025e-01 -1.23121887e-02 4.19306606e-01 -2.02159658e-01 -4.44769919e-01 1.38212191e-02 -7.05636740e-01 9.95831907e-01 1.25598586e+00 -8.55645418e-01 -6.55848503e-01 2.97281623e-01 7.98736513e-01 -5.14788151e-01 -5.96860647e-01 2.63010859e-01 8.27446401e-01 -7.70423770e-01 1.08216882e+00 -5.32674253e-01 3.46806735e-01 -4.90087867e-01 -3.17059219e-01 -7.87017167e-01 -9.64095965e-02 -3.78599465e-01 -3.63850415e-01 1.09111381e+00 2.07141325e-01 -8.21637213e-02 3.51117760e-01 8.60220253e-01 -7.47557729e-02 -4.24413890e-01 -8.61815512e-01 -5.22107124e-01 -5.37319481e-01 -8.33404005e-01 9.58336592e-02 5.68958342e-01 1.75469652e-01 3.42845112e-01 -7.00986743e-01 1.70273706e-01 4.70072299e-01 -2.85071284e-01 8.55689168e-01 -8.24298978e-01 -1.02081988e-02 -5.08052051e-01 -1.55058235e-01 -1.11631036e+00 -8.58339071e-02 -6.58569753e-01 1.73109189e-01 -1.72539222e+00 2.68321186e-01 1.64089099e-01 -2.94992805e-01 5.04597068e-01 -2.17456609e-01 4.15527195e-01 3.32121700e-01 3.13217610e-01 -1.28355098e+00 6.98140860e-01 1.57558751e+00 -4.29025441e-02 -2.43492395e-01 -2.62515277e-01 -5.33694625e-01 2.59527892e-01 6.38308167e-01 3.44653279e-02 -5.81923485e-01 -6.26161814e-01 -6.07317127e-02 8.90992209e-02 7.93829501e-01 -8.23307931e-01 3.80745083e-01 -1.37793288e-01 3.33148509e-01 -7.56719828e-01 5.85240006e-01 -8.58891606e-01 6.58282712e-02 3.64920914e-01 -3.92065942e-01 1.90009903e-02 3.99153918e-01 6.01621687e-01 -2.05224767e-01 1.68668061e-01 2.29716465e-01 -7.98731223e-02 -1.02798271e+00 1.79858863e-01 -1.67164415e-01 -5.24588190e-02 1.06665444e+00 -9.45366174e-02 -4.21269625e-01 -5.44145644e-01 -7.49229372e-01 5.95368743e-01 2.31903076e-01 8.51572931e-01 8.46923530e-01 -1.28124607e+00 -6.71658278e-01 1.65911227e-01 4.93948340e-01 -1.46300286e-01 7.93918669e-01 1.19645202e+00 -2.97598690e-01 9.22871351e-01 -3.23105395e-01 -7.14382410e-01 -1.28405261e+00 6.05124414e-01 9.19721499e-02 1.94763824e-01 -7.09062159e-01 7.56832123e-01 7.54567189e-03 -2.13997811e-01 5.13200641e-01 -2.95499206e-01 -4.42910880e-01 1.51512489e-01 5.82302451e-01 1.40788928e-01 -1.18016973e-01 -9.98970270e-01 -4.76006657e-01 5.61860383e-01 7.52781332e-02 -4.79191989e-01 8.78431201e-01 -6.95981622e-01 2.35722765e-01 5.59562802e-01 1.02215040e+00 -2.45943695e-01 -1.49180198e+00 -3.19275260e-01 8.19822475e-02 -5.32074094e-01 -2.53430784e-01 -1.00835145e+00 -6.76458776e-01 1.19905889e+00 6.76145136e-01 -1.49247512e-01 9.83602881e-01 7.63044506e-02 7.01409638e-01 4.64150220e-01 3.99862617e-01 -7.75746763e-01 4.74887043e-01 2.54482001e-01 1.38329971e+00 -1.65462363e+00 -2.05491841e-01 -2.52747208e-01 -1.10758603e+00 8.22121263e-01 9.10983920e-01 3.53207648e-01 7.43464604e-02 -4.43766236e-01 8.31411220e-03 -2.03692332e-01 -6.99876487e-01 -6.08089983e-01 9.39206660e-01 6.79956019e-01 4.10939276e-01 -3.84755805e-02 -6.53550625e-02 6.15387201e-01 2.15584666e-01 1.76258057e-01 -6.34463727e-02 8.59145820e-01 -1.52277097e-01 -7.62048841e-01 -3.88877392e-02 8.33983496e-02 -1.46883398e-01 -1.13108687e-01 -1.52201116e-01 7.95836985e-01 -3.30821276e-02 1.05407381e+00 -6.84750229e-02 -2.78611630e-01 5.71081936e-01 1.47976547e-01 6.45455122e-01 -4.39913213e-01 -2.09913954e-01 1.73663870e-01 1.59344539e-01 -1.02759981e+00 -6.55714095e-01 -7.37368882e-01 -9.80827749e-01 1.49864063e-01 8.94867480e-02 -1.30259171e-01 6.15232527e-01 1.05563676e+00 4.49335843e-01 4.66220558e-01 8.85703042e-02 -1.18113029e+00 4.88458900e-03 -9.79217529e-01 -4.08705324e-02 6.30593538e-01 5.20746768e-01 -7.68881142e-01 -5.52772842e-02 3.16995174e-01]

[9.84041690826416, 0.8389043807983398]

a1f59634-2c8b-4532-9451-e993a240c6be

improving-cnn-based-person-re-identification

2307.00397

https://arxiv.org/abs/2307.00397v2

https://arxiv.org/pdf/2307.00397v2.pdf

Improving CNN-based Person Re-identification using score Normalization

Person re-identification (PRe-ID) is a crucial task in security, surveillance, and retail analysis, which involves identifying an individual across multiple cameras and views. However, it is a challenging task due to changes in illumination, background, and viewpoint. Efficient feature extraction and metric learning algorithms are essential for a successful PRe-ID system. This paper proposes a novel approach for PRe-ID, which combines a Convolutional Neural Network (CNN) based feature extraction method with Cross-view Quadratic Discriminant Analysis (XQDA) for metric learning. Additionally, a matching algorithm that employs Mahalanobis distance and a score normalization process to address inconsistencies between camera scores is implemented. The proposed approach is tested on four challenging datasets, including VIPeR, GRID, CUHK01, and PRID450S, and promising results are obtained. For example, without normalization, the rank-20 rate accuracies of the GRID, CUHK01, VIPeR and PRID450S datasets were 61.92%, 83.90%, 92.03%, 96.22%; however, after score normalization, they have increased to 64.64%, 89.30%, 92.78%, and 98.76%, respectively. Accordingly, the promising results on four challenging datasets indicate the effectiveness of the proposed approach.

['Chahrazed Boudellal', 'Afaf Benzaibak', 'Shadi Atalla', 'Wathiq Mansoor', 'Yassine Himeur', 'Abdelmalik Ouamane', 'Ammar Chouchane']

2023-07-01

['person-re-identification', 'metric-learning', 'metric-learning']

['computer-vision', 'computer-vision', 'methodology']

[-1.99817806e-01 -7.35138834e-01 2.98833221e-01 -3.88399482e-01 -6.23653769e-01 -4.93022323e-01 4.85923976e-01 1.12909622e-01 -6.64158881e-01 7.10490942e-01 -7.20242113e-02 3.49260956e-01 -3.01546484e-01 -6.15277946e-01 -9.75815579e-02 -7.15876579e-01 1.13660865e-01 1.53795496e-01 -1.35964036e-01 -9.52772945e-02 2.79972196e-01 4.95023876e-01 -1.51244438e+00 -1.11129530e-01 7.29605019e-01 1.06983030e+00 -3.96152049e-01 5.14977157e-01 3.61283690e-01 3.15277368e-01 -7.18313038e-01 -9.71106946e-01 4.35929328e-01 -1.06048055e-01 -2.58593321e-01 1.95914894e-01 5.14011919e-01 -4.53565657e-01 -3.80058944e-01 1.16597748e+00 6.65100515e-01 1.04098544e-01 7.55678892e-01 -1.36248207e+00 -7.07612395e-01 -8.02439600e-02 -9.05529857e-01 5.99707402e-02 4.12788302e-01 -8.07989668e-03 7.12799430e-01 -8.68553817e-01 2.73583114e-01 9.80124831e-01 9.25981462e-01 3.62430930e-01 -9.66981471e-01 -1.01470673e+00 -3.40176135e-01 4.88886058e-01 -1.96753931e+00 -2.13470027e-01 6.73120081e-01 -6.40876651e-01 5.14524639e-01 2.96165258e-01 4.94906694e-01 7.81199038e-01 -1.05693322e-02 2.39275008e-01 1.18460095e+00 -3.06596905e-01 -5.35510890e-02 3.90984625e-01 2.25154877e-01 5.14047921e-01 6.44410968e-01 7.31689930e-02 -4.54962224e-01 -4.48639616e-02 4.96330678e-01 4.20811146e-01 -6.48485497e-02 -2.51056671e-01 -1.07564342e+00 5.61985075e-01 3.10115874e-01 2.00614467e-01 -2.36687765e-01 -6.15637541e-01 4.74986702e-01 6.54262155e-02 2.95336992e-01 1.60356238e-01 -2.78068751e-01 -1.58132866e-01 -8.33825409e-01 2.98087537e-01 4.34739411e-01 8.87152374e-01 6.66666567e-01 -2.54466653e-01 -1.56562239e-01 1.01133907e+00 1.07708789e-01 7.62843013e-01 4.30941820e-01 -8.14701438e-01 5.57831049e-01 7.38857210e-01 3.87187570e-01 -1.61729681e+00 -5.65506399e-01 -4.11363006e-01 -1.21997821e+00 -2.22684741e-02 3.62664670e-01 -9.26582143e-02 -4.44092870e-01 1.54141080e+00 2.88090020e-01 1.44577492e-02 2.45825782e-01 1.00860453e+00 9.39840734e-01 4.16811168e-01 4.28087562e-02 -1.41146630e-01 1.28724587e+00 -5.86251795e-01 -6.38963878e-01 2.61240989e-01 4.77014035e-01 -8.30029845e-01 7.12226093e-01 5.29218793e-01 -7.27078438e-01 -9.14004087e-01 -1.17371309e+00 3.67572159e-01 -3.19579124e-01 6.28752232e-01 2.25410312e-01 9.25140917e-01 -7.76827812e-01 3.59110683e-01 -2.06877545e-01 -6.38006628e-01 3.20026934e-01 5.28664351e-01 -7.07064033e-01 -3.44418019e-01 -9.29790616e-01 6.28217757e-01 3.35437059e-01 9.36992541e-02 -4.35141176e-01 -3.41661513e-01 -5.90791225e-01 4.72863056e-02 -3.74863222e-02 -3.01827937e-01 7.58864880e-01 -5.83501041e-01 -1.20623779e+00 8.76208901e-01 6.33096881e-03 -1.67653739e-01 4.01675731e-01 -7.74810985e-02 -8.05082977e-01 7.87473768e-02 1.92331046e-01 2.68030584e-01 3.37002724e-01 -9.62508738e-01 -8.60018015e-01 -9.59006190e-01 -5.19777164e-02 3.29745531e-01 -5.91627419e-01 3.58366370e-01 -4.33603168e-01 -3.83177102e-01 5.98357208e-02 -8.72159123e-01 1.69228181e-01 -2.52153397e-01 -3.17586094e-01 -2.97052771e-01 5.37965119e-01 -1.11428177e+00 1.14036000e+00 -2.28768086e+00 -1.14310585e-01 3.69289190e-01 1.31668597e-01 4.83515590e-01 1.37146741e-01 1.39816970e-01 -1.80870499e-02 -1.74111694e-01 4.32844907e-02 -1.74966544e-01 4.14650189e-03 -3.60351503e-01 4.64765579e-01 5.56617796e-01 -9.98039730e-03 4.36340511e-01 -4.29474473e-01 -4.52761352e-01 4.95879769e-01 5.08055627e-01 -3.18738610e-01 2.76277244e-01 8.16067338e-01 3.26425284e-01 -2.83618029e-02 8.99218500e-01 9.67872322e-01 1.24226034e-01 -4.68684873e-03 -5.40853798e-01 -3.14155310e-01 -6.39497578e-01 -1.63782883e+00 1.26534867e+00 -1.33167312e-01 4.55344677e-01 -1.49728224e-01 -1.02709544e+00 1.29501510e+00 2.55135924e-01 6.54138982e-01 -1.03188050e+00 2.24441141e-01 9.15273055e-02 -9.86913219e-02 -4.93970335e-01 4.95364070e-01 1.25466004e-01 -1.20476924e-01 1.51778102e-01 -3.11265644e-02 6.42147362e-01 3.01947504e-01 6.65461179e-03 6.41466022e-01 -3.15824360e-01 3.60619992e-01 -1.17074244e-01 1.09147787e+00 -2.68747061e-01 7.05750525e-01 4.11970407e-01 -5.49669743e-01 7.17189252e-01 3.79131623e-02 -8.07423890e-01 -1.14124048e+00 -8.98391545e-01 -1.30826995e-01 5.73956430e-01 4.41325456e-01 -4.06464666e-01 -7.79177427e-01 -4.17860478e-01 -9.14335158e-03 1.86353803e-01 -3.47470641e-01 -2.21089542e-01 -3.74331027e-01 -9.21199620e-01 4.88127738e-01 4.53612238e-01 1.23387063e+00 -5.33747971e-01 -1.95665300e-01 1.46572394e-02 -2.26190895e-01 -1.18951142e+00 -4.45453376e-01 -6.49022937e-01 -4.46640372e-01 -1.43838251e+00 -8.94321144e-01 -7.34301627e-01 6.10207021e-01 6.03840709e-01 7.22257555e-01 6.45813067e-03 -4.32076216e-01 3.81184220e-01 -1.68717265e-01 -6.55597746e-02 1.96510866e-01 -9.45285410e-02 4.86523628e-01 3.63129586e-01 6.58962965e-01 -2.19102502e-01 -7.05621183e-01 7.47819483e-01 -4.10196811e-01 -2.79346138e-01 5.53898156e-01 7.13714242e-01 4.95244801e-01 3.19160432e-01 4.16954577e-01 -2.20936298e-01 4.96431947e-01 -2.20384344e-01 -7.20205367e-01 3.27599615e-01 -7.08729744e-01 -6.47130370e-01 6.06281161e-01 -3.15431446e-01 -9.09087956e-01 5.02932966e-02 1.77319739e-02 -1.96232051e-01 -3.97489727e-01 2.47269757e-02 -4.59046721e-01 -1.80976614e-01 5.85821152e-01 1.51577786e-01 -1.22575708e-01 -4.27823633e-01 -7.54324794e-02 1.10893679e+00 6.85584843e-01 -1.86695084e-01 1.03298843e+00 3.37772787e-01 -7.11620227e-02 -8.64631712e-01 -5.55626929e-01 -6.18805528e-01 -8.20123494e-01 -3.05556536e-01 1.02113068e+00 -1.11543941e+00 -1.11228251e+00 9.34603631e-01 -9.75249231e-01 4.75437909e-01 1.84037313e-01 8.05326283e-01 3.36173475e-02 5.61741650e-01 -2.47178689e-01 -8.82666349e-01 -5.76530457e-01 -9.77929533e-01 6.95254683e-01 7.39159048e-01 -6.90018237e-02 -5.81684768e-01 -1.93345249e-01 7.17447102e-01 4.51057613e-01 4.50812131e-01 3.22709173e-01 -7.12067366e-01 -2.45049521e-01 -5.86207032e-01 -7.13624716e-01 3.76649261e-01 3.54810268e-01 1.67870254e-03 -8.17127407e-01 -4.54317927e-01 -3.34696680e-01 -6.21389672e-02 2.50210702e-01 1.78830415e-01 1.02689195e+00 -3.64039540e-01 -5.38151190e-02 7.36402571e-01 1.29542792e+00 5.98522544e-01 5.13531625e-01 6.14269614e-01 6.58298135e-01 5.44643700e-01 6.05444074e-01 7.23226607e-01 7.17090189e-01 9.96050179e-01 2.98058689e-01 -8.29381570e-02 1.12940460e-01 1.16002515e-01 1.40242606e-01 7.33303547e-01 -1.94269612e-01 1.61065981e-01 -7.65655398e-01 4.15074199e-01 -1.63178074e+00 -1.11054611e+00 -2.44421542e-01 2.41368914e+00 2.25812435e-01 -1.17914148e-01 3.98582727e-01 5.40836513e-01 1.08487821e+00 -1.75074711e-01 -4.58548337e-01 -1.83282569e-01 -1.91497579e-01 9.81420185e-03 6.05486333e-01 1.34968907e-01 -1.46631932e+00 4.23638284e-01 4.72738934e+00 5.69880009e-01 -7.73816884e-01 4.66414057e-02 6.91490412e-01 1.74296618e-01 5.08127213e-01 -5.06165087e-01 -1.00025105e+00 6.64392352e-01 6.44770801e-01 -2.07783446e-01 5.67681134e-01 7.07407951e-01 1.04899004e-01 -1.20127080e-02 -7.05047965e-01 1.89542413e+00 5.17015755e-01 -8.27597678e-01 -3.67063046e-01 8.47947225e-02 7.45086908e-01 -5.35666764e-01 2.04475701e-01 2.14092061e-01 3.03041656e-02 -9.64873254e-01 2.38421187e-01 5.69719553e-01 7.29120910e-01 -1.28129971e+00 1.29946387e+00 1.95985436e-01 -1.28006566e+00 -2.14444861e-01 -4.60676223e-01 5.76124303e-02 -1.45405874e-01 4.53961223e-01 -4.41676497e-01 7.39048779e-01 1.18796754e+00 7.38743305e-01 -8.49576116e-01 1.15183032e+00 7.40829408e-02 2.06712991e-01 -1.05164945e-01 1.93880070e-02 -2.33675539e-01 -3.66370887e-01 1.23859823e-01 1.03885221e+00 5.42850852e-01 2.03873888e-01 1.14812553e-01 4.35514480e-01 -3.41876559e-02 3.81654799e-01 -4.74108815e-01 5.11263967e-01 3.20594668e-01 1.43391752e+00 -2.06518456e-01 -2.27328643e-01 -5.08851290e-01 9.09065485e-01 8.35505277e-02 1.53742209e-01 -1.18368375e+00 -6.03222847e-01 8.30541015e-01 -1.30191550e-01 1.56196967e-01 -3.12531859e-01 -1.19137496e-01 -1.29853547e+00 1.43689692e-01 -8.53947103e-01 6.15769267e-01 -5.83142400e-01 -1.36867094e+00 4.07093763e-01 4.86648129e-03 -1.40977037e+00 7.54501112e-03 -6.39865577e-01 -4.63011801e-01 9.00919616e-01 -1.31654930e+00 -1.07109678e+00 -9.34359848e-01 8.62455904e-01 2.03391850e-01 -7.19968617e-01 5.98612010e-01 6.96646929e-01 -9.62869704e-01 1.08202922e+00 4.62503701e-01 4.22344208e-01 7.64835715e-01 -9.01421130e-01 -9.25379544e-02 8.92751634e-01 -1.55384824e-01 4.79295552e-01 2.12585449e-01 -3.35486829e-01 -1.22259593e+00 -1.16377723e+00 8.25488091e-01 -1.37526035e-01 1.14415713e-01 -1.60845011e-01 -6.61117375e-01 2.78921306e-01 5.90622202e-02 -3.21659148e-02 9.53842998e-01 -1.19502563e-02 -3.09704155e-01 -7.84062624e-01 -1.57719243e+00 4.18853641e-01 8.41611445e-01 -3.03658813e-01 -1.54675558e-01 2.29506269e-01 4.58242409e-02 -1.57313570e-01 -1.14209890e+00 2.95650303e-01 8.55885923e-01 -1.13308930e+00 1.10323131e+00 -1.93630040e-01 -1.29403502e-01 -5.71069479e-01 -4.26265568e-01 -1.00513589e+00 -5.06897449e-01 -8.82517844e-02 1.92955509e-01 1.69935858e+00 -1.09513095e-02 -7.45681703e-01 7.35145569e-01 8.54968369e-01 2.38598168e-01 -2.15483561e-01 -9.60167348e-01 -8.06982160e-01 -3.49813193e-01 -4.48747650e-02 7.45716095e-01 9.25892532e-01 -3.27279031e-01 2.61427388e-02 -6.96723223e-01 4.42850351e-01 1.02086508e+00 -2.34203160e-01 1.11835003e+00 -1.56952357e+00 1.32089123e-01 -1.28261968e-01 -8.32622528e-01 -3.84737164e-01 -2.44830787e-01 -5.51236570e-01 -4.82749581e-01 -1.23256946e+00 4.80081171e-01 -3.87973905e-01 -5.22957921e-01 1.76779762e-01 -8.74615386e-02 5.89705229e-01 2.31644437e-01 4.13905263e-01 -5.54288030e-01 4.38863069e-01 8.21655095e-01 -3.03908944e-01 -6.79029152e-02 1.41249865e-01 -7.12734818e-01 5.63441455e-01 1.17729795e+00 -2.07098395e-01 -1.29452758e-02 -3.73502046e-01 -1.12713352e-01 -2.98770756e-01 4.38722283e-01 -1.36430991e+00 3.48349869e-01 2.81230975e-02 9.89628732e-01 -6.55072570e-01 4.05301213e-01 -8.49089801e-01 4.09672260e-01 5.75508714e-01 -3.20801437e-02 4.95974362e-01 -6.03953302e-02 2.64623016e-01 -2.38141164e-01 1.20371394e-02 8.18240762e-01 -3.20611037e-02 -8.02421868e-01 4.00999218e-01 1.43022418e-01 -2.58004278e-01 1.31635714e+00 -5.54028988e-01 -3.21752667e-01 -1.88301384e-01 -4.67711329e-01 1.98502243e-02 2.97599137e-01 3.92528772e-01 7.49544322e-01 -1.75275278e+00 -8.24916899e-01 3.29614550e-01 4.14455920e-01 -2.98626125e-01 4.84857738e-01 7.18536496e-01 -4.39199507e-01 4.70725656e-01 -5.36922634e-01 -5.56605995e-01 -1.64082861e+00 4.81835634e-01 2.41416752e-01 -1.02083571e-01 -1.36926830e-01 3.92810374e-01 -7.49475658e-02 -4.94717777e-01 2.35939801e-01 1.91145629e-01 -6.63387239e-01 9.06303525e-02 6.46788061e-01 8.65357161e-01 1.49798334e-01 -1.22049487e+00 -5.40577710e-01 1.10514748e+00 -8.02829936e-02 1.53547361e-01 1.12253916e+00 -2.59839416e-01 -1.23485993e-03 -1.58184186e-01 1.24743605e+00 -1.00154206e-01 -8.39973032e-01 -2.36428395e-01 -6.40657619e-02 -7.62345374e-01 -4.03446764e-01 -6.21947706e-01 -1.00376368e+00 8.10264647e-01 1.28653097e+00 1.15678422e-01 1.12580740e+00 -3.76779437e-01 7.26405799e-01 2.26932585e-01 3.05622786e-01 -1.19677043e+00 1.05170466e-01 2.40830988e-01 6.75288737e-01 -1.39150703e+00 5.63090742e-02 -1.31097138e-01 -5.20339549e-01 1.14877462e+00 8.47771585e-01 3.78379561e-02 4.62104380e-01 -3.71154845e-01 1.25386836e-02 9.43963975e-02 -1.11023849e-02 -1.17464989e-01 2.60785311e-01 7.59733021e-01 1.40207723e-01 1.24006495e-01 -3.08699429e-01 6.94660664e-01 -2.28030905e-01 -1.02299064e-01 2.71103591e-01 5.71527362e-01 -1.77749619e-01 -7.56487489e-01 -7.05498755e-01 2.91774690e-01 -4.64820743e-01 3.00347269e-01 -2.87995428e-01 8.71725738e-01 5.25539994e-01 1.22930193e+00 1.55511387e-02 -8.82831573e-01 5.86813569e-01 -2.04116285e-01 1.63207129e-01 -4.22231779e-02 -4.39958900e-01 -2.35675365e-01 -1.51936188e-01 -2.94106752e-01 -6.15416527e-01 -7.25103915e-01 -6.83564484e-01 -6.56220734e-01 -2.91245818e-01 1.51000455e-01 6.97879434e-01 7.53039539e-01 3.25236410e-01 1.44890115e-01 1.02938783e+00 -3.95025074e-01 -3.93026173e-01 -8.21915209e-01 -5.96621394e-01 6.39900327e-01 3.92135940e-02 -7.69397736e-01 -2.50896156e-01 -1.55214937e-02]

[14.659543991088867, 0.9857585430145264]

caf0d8ab-e68a-429a-9312-2dcaabfca4d9

v3ctron-data-retrieval-access-system-for

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4430463

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4430463

V3CTRON | Data Retrieval & Access System For Flexible Semantic Search & Retrieval Of Proprietary Document Collections Using Natural Language Queries.

V3CTRON is an open source vector database that allows users to upload text based documents & document collections, which are automatically embedded for super-accurate semantic search & retrieval using natural language queries. V3CTRON supports multiple vector database providers, including Milvus, LlamaIndex and Qdrant, giving developers flexibility & customization - and end users the ability to leverage the benefits of LLMs that can directly access their use-case specific data. A notable feature of V3CTRON is its memory capability, which allows ChatGPT to remember and retrieve information from previous conversations by saving conversation snippets to the vector database. V3CTRON is a practical solution for enhancing document accessibility and enabling context-aware and memory-relevant chat experiences with effective neural search capabilities.

['Devin Schumacher']

2023-04-26

social-science-research-network-ssrn-2023-4

['conversational-search']

['natural-language-processing']

[-3.89071673e-01 -3.85644317e-01 -4.79589224e-01 -2.01393411e-01 -7.54245996e-01 -8.06330681e-01 5.71607292e-01 1.27629980e-01 -5.19009233e-01 5.71291268e-01 7.44546294e-01 -2.53870577e-01 -3.98126870e-01 -7.22794533e-01 1.55144662e-01 -1.24654226e-01 2.63802558e-01 7.20092952e-01 3.89567703e-01 -7.59911597e-01 7.20561981e-01 3.30104411e-01 -1.65603161e+00 6.08126163e-01 5.07923305e-01 8.25273573e-01 1.06181443e+00 7.95588315e-01 -1.17356110e+00 7.29837060e-01 -7.81799614e-01 -3.88779581e-01 -1.60647571e-01 2.53112793e-01 -1.34470046e+00 -7.52050757e-01 2.43262485e-01 -5.20660400e-01 -3.53726953e-01 4.94304359e-01 7.04309940e-01 4.04006779e-01 2.82425702e-01 -1.30376649e+00 -7.04117179e-01 6.84718549e-01 2.23895907e-01 5.50979853e-01 7.78284431e-01 4.11855221e-01 1.06803679e+00 -8.39432418e-01 1.02580988e+00 1.25237584e+00 4.93046433e-01 5.44159591e-01 -7.84490943e-01 -6.44199848e-01 -2.15006948e-01 6.01579845e-01 -1.23436832e+00 -8.29782724e-01 2.87612587e-01 -1.55659154e-01 2.30721998e+00 8.41156602e-01 7.35976219e-01 1.34971690e+00 1.40042361e-02 1.03288054e+00 2.64154613e-01 -2.16628551e-01 1.94595337e-01 4.96198565e-01 6.03249311e-01 6.50611937e-01 -6.55220151e-01 -4.12320048e-01 -1.10720956e+00 -5.14578879e-01 5.86999476e-01 6.29038960e-02 -2.14837119e-01 -3.66157703e-02 -1.04626393e+00 5.03450572e-01 2.69462585e-01 6.32892251e-01 -4.07537997e-01 1.96415499e-01 9.39380050e-01 4.81109291e-01 2.40529805e-01 6.94957316e-01 -6.45618081e-01 -1.02826071e+00 -5.51161289e-01 2.93262899e-01 8.52642536e-01 1.40362728e+00 5.17211556e-01 -1.15819804e-01 -6.47607625e-01 1.33760643e+00 1.81650043e-01 6.05584443e-01 1.07299459e+00 -1.14420545e+00 6.05523705e-01 9.01927471e-01 -3.11898202e-01 -8.18002999e-01 -1.68676838e-01 -1.19482286e-01 -1.73560247e-01 -3.20849955e-01 -2.67124295e-01 3.79723877e-01 -2.85834759e-01 1.36631775e+00 3.66605110e-02 -3.76791358e-01 3.53473395e-01 5.83404541e-01 1.16575551e+00 7.49121606e-01 6.76368624e-02 -5.88606372e-02 1.39505458e+00 -8.59900415e-01 -1.01193082e+00 -3.59025657e-01 1.02921021e+00 -8.04745317e-01 1.69752908e+00 8.42305943e-02 -1.04397786e+00 -2.05106996e-02 -8.16690564e-01 -4.81905222e-01 -1.04700565e+00 -3.99091631e-01 9.68087912e-01 4.48137075e-01 -1.42369306e+00 4.58084524e-01 -7.54571021e-01 -6.95236802e-01 4.63586271e-01 1.44028947e-01 -3.16619664e-01 2.32725609e-02 -1.25906765e+00 1.34101760e+00 3.12764615e-01 -5.11808157e-01 -4.02249962e-01 -1.20719969e+00 -6.59187198e-01 4.07201916e-01 2.23220050e-01 -5.11019170e-01 1.46568155e+00 -2.44726315e-01 -1.35003424e+00 6.14963889e-01 -3.58729690e-01 -1.28772870e-01 -3.34657691e-02 9.51236486e-02 -3.46603543e-01 2.65979707e-01 2.41064787e-01 5.25055766e-01 2.43126675e-01 -1.19087316e-01 -8.70022103e-02 -5.38438678e-01 -2.29060233e-01 4.65856910e-01 -9.73740280e-01 2.48176858e-01 -9.34271932e-01 -2.93473840e-01 -3.81087929e-01 -4.08100009e-01 1.87745899e-01 1.73340321e-01 -3.64938259e-01 -6.22416735e-01 1.15704143e+00 -8.74602914e-01 1.37920654e+00 -1.94551814e+00 8.07998031e-02 1.94546103e-01 1.61839128e-01 6.06655121e-01 -2.25479335e-01 1.15261042e+00 5.28531969e-01 7.64681473e-02 5.11876285e-01 -1.77836075e-01 2.85445124e-01 8.74901414e-02 -3.89532775e-01 -1.62781626e-01 -4.78889793e-01 1.36006308e+00 -6.74313724e-01 -8.55481923e-01 5.90633333e-01 7.10434735e-01 -5.60408711e-01 4.98585850e-02 -3.30347896e-01 -3.36188763e-01 -6.07214391e-01 4.34372425e-01 2.20375210e-01 -4.10200119e-01 2.20453933e-01 8.04822594e-02 -7.43530840e-02 6.16771936e-01 -6.92800820e-01 2.09993815e+00 -1.08110023e+00 1.13402820e+00 4.75030616e-02 -4.10979271e-01 6.40402675e-01 4.15100247e-01 1.93530872e-01 -1.05264831e+00 -2.42365617e-02 -1.90653801e-01 -1.01198447e+00 -7.03950882e-01 9.35861051e-01 6.71351790e-01 6.24525920e-03 8.47822368e-01 4.25914079e-01 7.57295266e-02 -1.60020927e-03 8.42055678e-01 1.28004360e+00 -3.53998452e-01 1.49495363e-01 -2.43423179e-01 1.42911240e-01 7.08149001e-02 -1.86461836e-01 6.18398368e-01 2.93133467e-01 -9.04785916e-02 1.20745122e-01 4.43835370e-02 -7.27022648e-01 -1.11175191e+00 -2.38047391e-01 1.65904772e+00 -1.64335474e-01 -9.33961093e-01 -7.40722954e-01 -1.49153531e-01 1.20122686e-01 1.21204901e+00 -1.17220387e-01 -3.63187820e-01 -1.37336493e-01 -2.04009842e-02 8.37579370e-01 7.80825198e-01 5.83471060e-01 -1.42321646e+00 -6.35045707e-01 1.10473312e-01 -5.50481558e-01 -8.11478972e-01 -9.35334504e-01 -2.57883836e-02 -7.44484961e-01 -5.35535395e-01 -5.35272360e-01 -6.80756330e-01 1.18129633e-01 4.15038109e-01 1.08024490e+00 3.23415816e-01 -7.59365797e-01 6.93627954e-01 -3.01735699e-01 -1.06473178e-01 -3.29448670e-01 2.64544189e-01 1.34704085e-02 -8.43921185e-01 5.38660645e-01 -7.07521319e-01 -4.73045498e-01 1.46348238e-01 -6.01640284e-01 -4.43603136e-02 -5.48373051e-02 5.88047028e-01 6.96424171e-02 -5.90775371e-01 6.88709795e-01 -4.78724837e-01 1.20281243e+00 -4.63894576e-01 -7.66936362e-01 4.89270687e-01 -1.01254892e+00 2.13813975e-01 2.79646724e-01 -2.56181121e-01 -1.14295983e+00 -3.96942019e-01 -4.10119236e-01 -3.72659177e-01 2.90193051e-01 5.87416887e-01 -1.30919516e-01 7.09162503e-02 6.26381993e-01 5.92660010e-01 2.12340191e-01 -6.22414291e-01 6.46614313e-01 1.23540354e+00 6.05414152e-01 -1.93121329e-01 -3.53803486e-02 -1.04050204e-01 -9.81457174e-01 -1.11055088e+00 -4.13843580e-02 -8.31236660e-01 -3.32209706e-01 -2.90708125e-01 8.49329948e-01 -5.79395652e-01 -1.29770815e+00 2.04731062e-01 -1.24551272e+00 -3.93623829e-01 -3.89253832e-02 6.29233718e-02 -3.37434322e-01 1.45768166e-01 -7.98868954e-01 -6.74067616e-01 -9.12286997e-01 -1.10125196e+00 8.86233330e-01 1.24803789e-01 -8.84983778e-01 -1.20133865e+00 -6.44409359e-02 6.43952012e-01 9.56193328e-01 -7.86858261e-01 1.16122150e+00 -1.06061327e+00 -5.59673250e-01 -3.43452215e-01 -2.41230890e-01 4.80225235e-02 -1.16552889e-01 -3.82079095e-01 -1.04851997e+00 -1.23088330e-01 -5.90009153e-01 -4.63085443e-01 4.62272704e-01 3.36026922e-02 1.16036248e+00 -6.42416775e-01 -8.26405287e-01 5.46652615e-01 1.18505001e+00 3.01885962e-01 5.55754960e-01 3.07986528e-01 2.92014092e-01 2.43598387e-01 2.42831111e-01 4.07055169e-01 4.45819825e-01 1.17146635e+00 1.56972766e-01 4.95328158e-01 -2.05604762e-01 -4.27278787e-01 8.25007930e-02 9.54932153e-01 4.42877442e-01 -3.58655423e-01 -7.97645807e-01 6.74156308e-01 -1.76972687e+00 -1.06780946e+00 3.12341154e-01 1.76098704e+00 1.01951098e+00 -2.40164667e-01 -2.05341384e-01 -2.00282797e-01 1.62343517e-01 1.69461612e-02 -7.91510880e-01 -4.28256869e-01 -5.20699937e-03 2.84945369e-01 3.97974998e-01 6.56857014e-01 -1.69767559e-01 1.30384803e+00 7.19559145e+00 1.19547260e+00 -1.06492090e+00 3.68451953e-01 -9.83527899e-02 -7.30405450e-01 -5.20409644e-01 -4.20695037e-01 -8.63376200e-01 4.42530274e-01 9.54088509e-01 -7.46049583e-01 9.42700565e-01 1.11725152e+00 -1.57698900e-01 9.21002701e-02 -9.39397991e-01 1.39288080e+00 3.51090766e-02 -2.33725619e+00 1.18211545e-01 -1.00568689e-01 -1.17698446e-01 3.14275384e-01 5.19333519e-02 4.65085238e-01 3.87560517e-01 -1.07894099e+00 3.66507918e-01 4.45394009e-01 8.60772729e-01 -8.05105150e-01 3.52189213e-01 3.15409958e-01 -8.46955597e-01 2.74972115e-02 -2.25349471e-01 1.64167449e-01 1.22993343e-01 -1.46391824e-01 -1.38495147e+00 -2.17478909e-02 9.58382547e-01 4.12126422e-01 -6.17000341e-01 4.87028122e-01 5.94025254e-01 1.45020783e-01 -1.03721000e-01 -7.95929253e-01 -6.17858581e-02 1.63719982e-01 5.55804908e-01 1.40801942e+00 2.22270608e-01 1.36250947e-02 -2.34271929e-01 9.39019799e-01 2.03713626e-01 2.92700768e-01 -7.97654808e-01 -2.91101336e-01 1.31887257e+00 1.21487486e+00 -5.08201003e-01 -3.66585374e-01 -7.30263218e-02 1.36842835e+00 4.73023921e-01 3.34460616e-01 -2.68951386e-01 -9.97850418e-01 1.23453426e+00 -1.10160470e-01 2.39132807e-01 -2.14782909e-01 -1.39500782e-01 -7.62669086e-01 1.25921503e-01 -9.21597302e-01 3.27868342e-01 -1.40743566e+00 -9.12799895e-01 8.32760751e-01 1.02480613e-01 -6.00586295e-01 -7.16577947e-01 -5.02855003e-01 -4.86867815e-01 9.46351826e-01 -5.36220670e-01 -8.77147198e-01 -5.17713428e-02 1.02832508e+00 8.44540775e-01 -7.71669090e-01 1.36512077e+00 2.26926580e-01 -3.21670830e-01 8.48218083e-01 2.06160650e-01 4.40410450e-02 3.46305668e-01 -1.03387761e+00 5.24762273e-01 -1.63697160e-03 8.58223513e-02 1.31014729e+00 5.14032662e-01 -6.60677910e-01 -1.96380913e+00 -7.97239780e-01 9.78547633e-01 -9.25019264e-01 5.65225363e-01 -7.39880741e-01 -9.18287754e-01 8.56071532e-01 4.90859121e-01 -6.22296572e-01 8.14327478e-01 2.79105276e-01 -4.30375844e-01 -5.24609499e-02 -1.18643796e+00 8.41065943e-01 1.29011619e+00 -1.22948599e+00 -5.80850780e-01 6.88399673e-01 1.15058708e+00 -3.42676729e-01 -8.83198440e-01 -5.40718615e-01 6.15101218e-01 -6.00471914e-01 1.51172197e+00 -4.57761288e-01 5.70890307e-02 4.56799984e-01 -3.13676596e-01 -1.34090590e+00 5.03688715e-02 -7.59815454e-01 -1.03766978e-01 1.38078094e+00 6.59570158e-01 -7.47841239e-01 5.56039631e-01 9.07346547e-01 -1.62566900e-01 -3.07297379e-01 -1.12661171e+00 -7.11377919e-01 -2.58625776e-01 -9.14528966e-01 1.07253122e+00 1.08631837e+00 1.04370093e+00 5.72721183e-01 -4.65061702e-02 -4.13277864e-01 1.02660671e-01 -1.49900734e-01 4.04913485e-01 -1.02861977e+00 -1.74731076e-01 -6.06160820e-01 -2.20936820e-01 -1.08323181e+00 3.64173561e-01 -1.54512310e+00 -4.03254390e-01 -1.66271293e+00 4.13942672e-02 -3.79233658e-01 1.54627100e-01 7.40478158e-01 3.47202152e-01 -8.28356445e-02 7.42646828e-02 1.40841171e-01 -8.04880440e-01 5.90618730e-01 1.12027431e+00 -2.53890276e-01 -4.94531184e-01 -3.36461216e-01 -8.36583138e-01 1.71744332e-01 5.06791472e-01 -9.81833562e-02 -9.48411524e-01 -7.25191891e-01 2.51473904e-01 3.56357157e-01 4.13818836e-01 -6.24303043e-01 5.78813970e-01 -1.82467297e-01 3.08893204e-01 -7.41173089e-01 8.64213943e-01 -5.05587161e-01 -4.61341552e-02 1.95919573e-02 -8.63380790e-01 3.42156112e-01 5.07119477e-01 1.24691188e-01 2.16252729e-01 -1.36827022e-01 8.02347809e-02 -2.48377711e-01 -1.03509724e+00 4.92570587e-02 -1.09762359e+00 1.78609386e-01 6.28739417e-01 -1.81593701e-01 -7.92077780e-01 -7.84530640e-01 -2.79431880e-01 4.07175571e-01 2.31088057e-01 9.87242818e-01 9.87885296e-01 -1.41194046e+00 -6.39953166e-02 2.41330311e-01 5.63395500e-01 -6.39521718e-01 4.54056889e-01 3.31955582e-01 -4.53319818e-01 1.04971814e+00 -3.49962562e-01 -3.13866287e-01 -1.58422422e+00 5.15276968e-01 4.62592021e-02 1.59142703e-01 -9.82860088e-01 9.51197445e-01 -5.57331264e-01 -9.06408250e-01 5.94744027e-01 -1.69985712e-01 -2.16034979e-01 -1.53867500e-02 1.22863233e+00 4.79065090e-01 5.09068310e-01 -1.31152570e-01 -6.55715704e-01 -7.38700256e-02 -2.09178001e-01 -5.66908836e-01 1.30777872e+00 -1.82087898e-01 -2.23347351e-01 3.85292470e-01 1.73579931e+00 -3.64273816e-01 -3.83077979e-01 -3.30381274e-01 6.66932261e-04 -4.24037874e-01 5.91719270e-01 -1.07136428e+00 -9.51099277e-01 7.11530149e-01 6.33224249e-01 -1.10475317e-01 6.71238422e-01 3.02784890e-01 1.06788015e+00 1.14615536e+00 3.93446058e-01 -1.28821814e+00 2.42132530e-01 5.36789119e-01 1.10718191e+00 -5.86128294e-01 -3.28675479e-01 1.64928138e-02 -8.07506084e-01 8.24736059e-01 4.96860564e-01 6.56366765e-01 5.48850834e-01 4.04178917e-01 4.24539030e-01 -6.59013629e-01 -1.37594092e+00 1.65183291e-01 5.43196350e-02 9.21929300e-01 3.79912734e-01 -2.70127892e-01 2.51511335e-01 5.98740876e-01 -6.06860995e-01 1.00284042e-02 5.53577133e-02 1.00151598e+00 -2.57059187e-01 -9.15147960e-01 -3.16995941e-03 7.44252443e-01 2.70120986e-03 -6.06618345e-01 -4.97674406e-01 4.10804510e-01 -9.54315543e-01 1.06959391e+00 4.11615312e-01 -4.55999762e-01 2.31171846e-01 5.20118952e-01 6.75169006e-02 -4.00163531e-01 -9.44744527e-01 -4.47346658e-01 5.56733191e-01 -1.12810147e+00 5.57573736e-01 -4.49447870e-01 -1.34146094e+00 -5.50632000e-01 -3.62918615e-01 4.15449589e-01 9.79680240e-01 7.22640038e-01 9.49527740e-01 4.55509633e-01 -9.80134457e-02 -3.88709217e-01 -4.10569042e-01 -1.02382684e+00 -3.85842770e-01 4.12544124e-02 1.57335490e-01 -4.21041936e-01 -5.93162924e-02 -3.23621005e-01]

[11.48068904876709, 7.904420375823975]

f30cde1e-cf99-41a7-b04a-92ca3ea443fe

graph-neural-networks-for-3d-multi-object

2008.09506

https://arxiv.org/abs/2008.09506v1

https://arxiv.org/pdf/2008.09506v1.pdf

Graph Neural Networks for 3D Multi-Object Tracking

3D Multi-object tracking (MOT) is crucial to autonomous systems. Recent work often uses a tracking-by-detection pipeline, where the feature of each object is extracted independently to compute an affinity matrix. Then, the affinity matrix is passed to the Hungarian algorithm for data association. A key process of this pipeline is to learn discriminative features for different objects in order to reduce confusion during data association. To that end, we propose two innovative techniques: (1) instead of obtaining the features for each object independently, we propose a novel feature interaction mechanism by introducing Graph Neural Networks; (2) instead of obtaining the features from either 2D or 3D space as in prior work, we propose a novel joint feature extractor to learn appearance and motion features from 2D and 3D space. Through experiments on the KITTI dataset, our proposed method achieves state-of-the-art 3D MOT performance. Our project website is at http://www.xinshuoweng.com/projects/GNN3DMOT.

['Kris Kitani', 'Xinshuo Weng', 'Yunze Man', 'Yongxin Wang']

2020-08-20

['3d-multi-object-tracking']

['computer-vision']

[-4.39284518e-02 -5.01949310e-01 -1.09947257e-01 -1.42742053e-01 -4.30220097e-01 -4.64664310e-01 5.90819716e-01 -1.50820076e-01 -4.37499762e-01 3.80331814e-01 -1.99167818e-01 -8.10536742e-02 -1.24237493e-01 -7.14920759e-01 -6.30143642e-01 -7.50573099e-01 -4.03594822e-02 6.92209065e-01 7.84753561e-01 5.28595224e-02 1.48925126e-01 9.53474224e-01 -1.71696424e+00 -2.77203172e-01 3.85473251e-01 1.03070486e+00 2.56884933e-01 8.45084310e-01 -4.66508009e-02 5.08047402e-01 -2.87980646e-01 1.61688104e-01 6.73135817e-01 -4.74319279e-01 -4.18395847e-01 2.57180661e-01 7.26978719e-01 -3.88874978e-01 -4.99473393e-01 1.15652621e+00 4.52128172e-01 1.15347989e-01 6.37473822e-01 -1.41541791e+00 -3.38396668e-01 1.59437194e-01 -6.81022406e-01 2.74079323e-01 1.77997053e-01 2.05806702e-01 8.88266742e-01 -8.03344071e-01 6.56572282e-01 1.18508995e+00 4.99255687e-01 5.15144229e-01 -1.11572850e+00 -7.53369093e-01 1.28167018e-01 3.25638562e-01 -1.51007676e+00 -3.30180645e-01 9.50366139e-01 -4.10906434e-01 6.81036592e-01 1.05786741e-01 1.03799200e+00 6.72766387e-01 2.43711948e-01 8.54796946e-01 7.91189849e-01 -3.73375237e-01 3.75368372e-02 -4.53654714e-02 1.70785323e-01 1.09866571e+00 4.72257525e-01 2.69289702e-01 -5.00237346e-01 -1.74067125e-01 7.85223722e-01 2.95851022e-01 1.56403542e-01 -9.07418251e-01 -1.23241413e+00 6.50316000e-01 5.92694700e-01 2.15304255e-01 -2.48961017e-01 2.70618200e-01 2.72477269e-02 3.63794178e-01 2.44650751e-01 -9.76279154e-02 -1.21108092e-01 1.95111588e-01 -6.05006397e-01 4.38835382e-01 6.44452512e-01 1.13710308e+00 1.12195647e+00 -2.93124676e-01 -1.07484281e-01 4.13223863e-01 7.52704263e-01 7.11906195e-01 2.10710511e-01 -6.79512918e-01 2.43841544e-01 8.35955203e-01 1.20147437e-01 -9.52840745e-01 -5.93060076e-01 -2.24626392e-01 -5.65497398e-01 5.94638348e-01 5.02752244e-01 -2.56448667e-02 -9.02496755e-01 1.32982743e+00 9.15666759e-01 2.01050729e-01 -1.19714677e-01 1.12539017e+00 7.52331853e-01 1.93686694e-01 -2.74852723e-01 1.73157286e-02 1.38924301e+00 -1.03085709e+00 -4.42615569e-01 -2.40521520e-01 6.47996664e-01 -6.33312583e-01 4.79232579e-01 9.84801799e-02 -8.14266324e-01 -7.58784473e-01 -1.12290096e+00 -3.99432033e-02 -3.94073308e-01 2.18012735e-01 5.75406134e-01 4.82603341e-01 -9.15974557e-01 3.74032080e-01 -1.14938581e+00 -6.33498073e-01 3.56122732e-01 6.22762263e-01 -3.35318834e-01 7.14204982e-02 -6.75090075e-01 9.15152013e-01 5.32444954e-01 5.46641760e-02 -8.81637514e-01 -1.98064998e-01 -8.90839040e-01 -3.59080225e-01 5.58516026e-01 -9.36702073e-01 9.32920814e-01 -4.81277525e-01 -1.50121725e+00 7.76432693e-01 -1.77835077e-01 -2.76799709e-01 3.99653554e-01 -5.83263040e-01 -1.71964481e-01 -2.48428974e-02 -3.55130360e-02 5.33501863e-01 1.02768528e+00 -1.14540935e+00 -9.51891840e-01 -6.09433949e-01 -2.54973143e-01 3.02480072e-01 -2.41156742e-01 2.53986325e-02 -8.31030786e-01 -2.44116485e-01 3.39980245e-01 -1.29739141e+00 -2.64244676e-01 3.38426411e-01 -3.33748937e-01 -4.70200419e-01 1.32717538e+00 -3.17274153e-01 7.50309050e-01 -2.20936871e+00 3.24673653e-01 2.53290772e-01 6.40050769e-01 1.53130904e-01 1.22472206e-02 1.80788487e-02 5.07617772e-01 -5.28680205e-01 7.78295845e-02 -4.08875525e-01 7.04589039e-02 1.25131279e-01 2.64148563e-01 8.38221133e-01 4.43000764e-01 8.76728356e-01 -9.15381849e-01 -7.63198078e-01 5.00238419e-01 3.80710095e-01 -4.22530830e-01 1.94038138e-01 -2.21118465e-01 5.42157173e-01 -8.26989472e-01 8.58867824e-01 7.55676389e-01 -3.71029735e-01 7.99734965e-02 -3.13646346e-01 -3.54302347e-01 7.71519095e-02 -1.38268924e+00 1.73962164e+00 1.03791229e-01 4.60703760e-01 -5.00492752e-02 -8.25052619e-01 1.11779523e+00 -4.27074954e-02 7.24497855e-01 -4.48381096e-01 3.66249472e-01 7.89118856e-02 4.22304198e-02 -2.46431664e-01 5.39189398e-01 2.43564084e-01 -1.27992585e-01 3.33654493e-01 1.75667226e-01 -4.57128510e-02 9.09248441e-02 4.81104255e-02 1.26843059e+00 4.86016750e-01 3.40646446e-01 -9.18502659e-02 5.62538326e-01 2.11021528e-01 7.49199092e-01 7.51897335e-01 -4.84070152e-01 2.63565719e-01 -9.31442082e-02 -5.48965454e-01 -9.63983834e-01 -1.10664499e+00 1.35429204e-01 8.84937644e-01 6.38204515e-01 -2.84846783e-01 -1.51972637e-01 -1.03332722e+00 4.41251367e-01 7.75113255e-02 -5.71363449e-01 -1.05990402e-01 -6.59262717e-01 -3.19638878e-01 2.19451085e-01 5.49865663e-01 4.44726110e-01 -7.02988982e-01 -9.53131318e-01 1.40134051e-01 2.07761213e-01 -1.11131620e+00 -5.86939335e-01 2.91421980e-01 -9.49620068e-01 -9.32318807e-01 -3.88801694e-01 -6.22548282e-01 6.66997969e-01 7.52582252e-01 7.58238733e-01 1.99639365e-01 -3.83686036e-01 3.84174675e-01 -4.44094241e-01 -3.77071828e-01 -4.81016114e-02 1.52320433e-02 3.84932578e-01 1.21679660e-02 7.53312826e-01 -5.05240977e-01 -4.98230994e-01 5.31538665e-01 -4.13080066e-01 1.55846691e-02 9.79625046e-01 5.55809617e-01 8.52766395e-01 -2.00791553e-01 -4.91497219e-02 -3.70454729e-01 -4.73625921e-02 -2.71811068e-01 -1.07057035e+00 6.42630160e-02 -2.47004673e-01 8.35895985e-02 3.00714642e-01 -6.72426045e-01 -6.42997801e-01 7.71349251e-01 2.14236587e-01 -9.12752748e-01 -4.40132707e-01 -2.76366412e-03 -1.54332042e-01 -4.15333629e-01 3.76348227e-01 2.30097428e-01 9.30942297e-02 -5.85488021e-01 3.32216024e-01 5.92920125e-01 3.85103762e-01 -2.93656766e-01 1.28610075e+00 7.37429738e-01 3.64399940e-01 -8.67686391e-01 -7.31514931e-01 -7.53195882e-01 -1.24005616e+00 -5.34539580e-01 1.14833522e+00 -9.60652232e-01 -8.06169987e-01 2.91458905e-01 -8.66234362e-01 -1.55863509e-01 -7.68254176e-02 9.63613927e-01 -4.53338236e-01 3.74130249e-01 -2.15261787e-01 -9.07668054e-01 -2.35696033e-01 -9.28910673e-01 1.22075903e+00 5.17315984e-01 4.00066823e-02 -7.55632937e-01 2.77173132e-01 2.01117396e-01 1.89463794e-01 1.70020133e-01 2.50737965e-01 -7.08048940e-01 -8.16929519e-01 -2.73661464e-01 -3.55168849e-01 -1.37568355e-01 3.48153263e-01 -8.16723565e-04 -6.35312200e-01 -5.70411444e-01 -1.51019797e-01 -1.61896810e-01 7.50440776e-01 2.88921595e-01 6.64978147e-01 1.47882760e-01 -7.67668247e-01 6.29886508e-01 1.25671160e+00 1.24468654e-01 1.64439023e-01 4.51642573e-01 9.27413285e-01 2.79494733e-01 8.41920495e-01 4.07039940e-01 5.16685247e-01 9.60394263e-01 5.00174105e-01 1.23654127e-01 -3.06422412e-01 -9.02896151e-02 5.17424405e-01 7.92857587e-01 -1.33147687e-01 1.16847061e-01 -8.87044966e-01 5.02207756e-01 -2.19392729e+00 -7.86345005e-01 -1.35548458e-01 2.09672689e+00 1.89763889e-01 3.33198607e-01 5.27352870e-01 -3.78111154e-01 7.62021005e-01 -4.63539883e-02 -6.95986450e-01 3.37768197e-01 7.22055882e-02 -1.92742184e-01 6.63759053e-01 2.04450995e-01 -1.32492995e+00 9.87929344e-01 4.93800783e+00 4.72307593e-01 -9.28191543e-01 -5.73691502e-02 -2.97768861e-01 -2.26838455e-01 3.58340472e-01 2.54574955e-01 -1.45923793e+00 2.08493859e-01 5.66321373e-01 2.39058062e-02 1.06695972e-01 8.57917726e-01 -1.74846485e-01 -8.82772133e-02 -9.74855781e-01 9.23549891e-01 1.00849546e-01 -1.08119988e+00 -1.27659559e-01 2.68932045e-01 2.64856666e-01 3.22333723e-01 -3.43349427e-01 1.98120043e-01 5.27882814e-01 -3.39780360e-01 7.26841331e-01 6.28896654e-01 2.83669025e-01 -5.76035023e-01 4.83180374e-01 4.19145435e-01 -1.77605927e+00 4.99521345e-02 -4.90295053e-01 -8.20643380e-02 7.34459087e-02 4.67852443e-01 -8.37189496e-01 9.16989625e-01 9.26069260e-01 9.85840082e-01 -6.45588338e-01 1.32467997e+00 -4.66029868e-02 2.03802183e-01 -5.83354294e-01 -2.75685459e-01 9.42944065e-02 -1.81507125e-01 9.10659552e-01 1.01187754e+00 3.34899843e-01 -1.15164608e-01 6.86084449e-01 7.32268989e-01 8.74451324e-02 -1.63510576e-01 -8.32444787e-01 7.65935779e-02 4.46481258e-01 1.53521681e+00 -9.21581864e-01 -2.36984894e-01 -6.75115824e-01 9.25901115e-01 4.08347756e-01 -5.89656681e-02 -8.20902944e-01 -3.69816333e-01 6.63015604e-01 -2.59496361e-01 8.92323971e-01 -6.76528275e-01 1.19998433e-01 -9.53253686e-01 -4.41568457e-02 -3.39849293e-01 5.39464056e-01 -3.88669580e-01 -1.21359622e+00 3.25580865e-01 -8.90773237e-02 -1.60576701e+00 9.24491882e-03 -5.76466501e-01 -4.34282392e-01 7.44710267e-01 -1.37900007e+00 -1.34093261e+00 -6.04806662e-01 7.03393340e-01 3.37578952e-01 -1.27319664e-01 3.53920579e-01 3.46267283e-01 -5.35042048e-01 3.08022857e-01 7.02582393e-03 3.22587699e-01 6.98276341e-01 -1.12020421e+00 5.01938879e-01 8.35920453e-01 3.43070388e-01 3.14763337e-01 5.35699725e-01 -9.38519955e-01 -2.01570344e+00 -1.19510567e+00 5.45630932e-01 -5.56811690e-01 6.17399395e-01 -5.67016304e-01 -6.85166240e-01 7.25512207e-01 -5.15636876e-02 3.74070406e-01 5.07246971e-01 -3.28474678e-02 -8.90329201e-03 -8.93878117e-02 -9.58721876e-01 4.58704621e-01 1.42448187e+00 -1.85921937e-01 -5.74073136e-01 2.29563072e-01 5.75907707e-01 -5.42066455e-01 -8.22516203e-01 3.85017425e-01 8.21348667e-01 -6.80704176e-01 9.64646459e-01 -3.92786950e-01 -3.20970625e-01 -1.09178424e+00 -2.82131225e-01 -9.15819347e-01 -6.80272281e-01 -4.63772416e-01 -6.25466049e-01 9.06118631e-01 9.61436778e-02 -4.81322199e-01 1.00545037e+00 7.44762719e-02 -1.27504006e-01 -3.43237519e-01 -9.84396338e-01 -1.08843172e+00 -5.23051798e-01 -2.08307400e-01 3.67174000e-01 6.36803865e-01 -4.79860336e-01 4.98460501e-01 -4.50524479e-01 5.57839453e-01 1.13731885e+00 5.56798100e-01 1.42205501e+00 -1.47903883e+00 -3.39173853e-01 -2.97627658e-01 -7.93571949e-01 -1.35860419e+00 -3.34803835e-02 -8.46167743e-01 2.65846819e-01 -1.33196902e+00 2.60916084e-01 -4.95960504e-01 -3.21131438e-01 4.88003373e-01 -8.52221027e-02 4.38494891e-01 3.76210839e-01 3.23861897e-01 -1.00216460e+00 6.82458520e-01 1.25698352e+00 -1.18946537e-01 -3.85717511e-01 3.04674469e-02 -2.88615704e-01 4.92870986e-01 7.34604955e-01 -6.86284900e-01 2.02538788e-01 -4.00569856e-01 -3.84993851e-01 -2.16982350e-01 6.77830756e-01 -1.42134488e+00 6.35350168e-01 -1.69475138e-01 6.19357109e-01 -1.28651762e+00 5.05510926e-01 -1.10168409e+00 2.52307922e-01 7.30757594e-01 3.24586779e-01 1.00378022e-01 2.49279618e-01 6.48201823e-01 1.62731826e-01 2.15832423e-02 6.84477746e-01 7.26598054e-02 -1.01341641e+00 4.30027038e-01 -2.14681238e-01 -4.25272167e-01 1.26818371e+00 -3.79547685e-01 -1.86994210e-01 5.76240458e-02 -4.94907439e-01 5.08965850e-01 6.29971623e-01 5.40572762e-01 6.28852665e-01 -1.63531983e+00 -6.56922340e-01 3.31608951e-01 1.83537170e-01 2.32871864e-02 7.12181851e-02 1.16959012e+00 -1.44901872e-01 2.59398848e-01 -4.04530942e-01 -1.16892552e+00 -1.72458792e+00 5.60085952e-01 2.05979615e-01 -3.59787159e-02 -9.69667792e-01 6.87555611e-01 -1.74438935e-02 -3.06871295e-01 1.65803626e-01 -8.40904713e-02 -1.34370103e-01 -6.21385612e-02 4.44732457e-01 3.19648623e-01 -8.09783638e-02 -7.65149117e-01 -6.66161776e-01 8.99527252e-01 -3.57373506e-01 5.87059110e-02 1.42087626e+00 -1.75241709e-01 1.32042751e-01 4.54022676e-01 9.58609879e-01 -1.13700382e-01 -1.43773842e+00 -4.71483797e-01 1.32488683e-01 -7.02370584e-01 6.61758929e-02 -1.67186618e-01 -8.91661584e-01 5.09274065e-01 9.92250681e-01 1.78911552e-01 9.45426106e-01 3.72628212e-01 6.10765278e-01 5.66527903e-01 3.59046280e-01 -8.52424204e-01 1.15096688e-01 7.05951095e-01 3.47806692e-01 -1.30722773e+00 6.54325262e-02 -7.67131522e-02 -2.56448656e-01 1.00175142e+00 7.21181214e-01 -2.28832766e-01 5.80899000e-01 2.51559228e-01 1.79638505e-01 -4.63975370e-01 -5.04802704e-01 -7.66676664e-01 4.72578883e-01 5.44792771e-01 8.35247561e-02 -2.32836097e-01 -2.02469304e-02 -3.30850831e-03 1.72342867e-01 -1.48076192e-01 5.44722266e-02 1.38383436e+00 -7.12939322e-01 -1.23381627e+00 -5.45371115e-01 4.58018214e-01 3.58981341e-02 4.25754249e-01 -6.39820635e-01 8.47514629e-01 7.25276917e-02 7.12478042e-01 2.66577899e-02 -7.79391706e-01 4.55694675e-01 -7.67979175e-02 6.96616709e-01 -5.24300754e-01 -3.79272848e-01 2.48991236e-01 -1.07563712e-01 -6.88105702e-01 -7.24741518e-01 -9.44710314e-01 -1.24438429e+00 -9.98894572e-02 -5.91534317e-01 1.04837030e-01 7.07064390e-01 9.14125979e-01 4.71205533e-01 5.49612880e-01 5.06829262e-01 -1.33853006e+00 -1.73281491e-01 -7.72925138e-01 -3.94694716e-01 2.76032269e-01 5.17123222e-01 -1.23949134e+00 -1.59048051e-01 -1.50081336e-01]

[6.515285015106201, -2.2098731994628906]

0f7fa78d-f966-413e-8a1b-d22fa8ce7377

pebble-feedback-efficient-interactive

2106.05091

https://arxiv.org/abs/2106.05091v1

https://arxiv.org/pdf/2106.05091v1.pdf

PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training

Conveying complex objectives to reinforcement learning (RL) agents can often be difficult, involving meticulous design of reward functions that are sufficiently informative yet easy enough to provide. Human-in-the-loop RL methods allow practitioners to instead interactively teach agents through tailored feedback; however, such approaches have been challenging to scale since human feedback is very expensive. In this work, we aim to make this process more sample- and feedback-efficient. We present an off-policy, interactive RL algorithm that capitalizes on the strengths of both feedback and off-policy learning. Specifically, we learn a reward model by actively querying a teacher's preferences between two clips of behavior and use it to train an agent. To enable off-policy learning, we relabel all the agent's past experience when its reward model changes. We additionally show that pre-training our agents with unsupervised exploration substantially increases the mileage of its queries. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods, including a variety of locomotion and robotic manipulation skills. We also show that our method is able to utilize real-time human feedback to effectively prevent reward exploitation and learn new behaviors that are difficult to specify with standard reward functions.

['Pieter Abbeel', 'Laura Smith', 'Kimin Lee']

2021-06-09

['unsupervised-pre-training']

['methodology']

[ 2.17880487e-01 3.64109755e-01 -2.68903702e-01 -2.62728125e-01 -8.95826697e-01 -8.21043909e-01 6.67407572e-01 2.34413415e-01 -7.32925773e-01 1.12844551e+00 -2.65494317e-01 -3.36489052e-01 -1.26724884e-01 -7.00109661e-01 -9.04538453e-01 -5.97419083e-01 -4.49825764e-01 7.03759313e-01 4.28089261e-01 -5.10045230e-01 3.25163037e-01 5.76540112e-01 -1.89307153e+00 -1.06236048e-01 8.51082921e-01 6.42133176e-01 4.29447174e-01 9.08991039e-01 2.04351678e-01 1.05314076e+00 -5.73885858e-01 1.74112409e-01 1.76551685e-01 -5.00796795e-01 -7.63062775e-01 3.65073755e-02 -3.62919718e-02 -6.21125102e-01 5.95121123e-02 6.31331086e-01 4.09882367e-01 4.22182977e-01 5.76061904e-01 -1.26112914e+00 -3.16916555e-01 5.71713746e-01 -1.29608721e-01 -4.88119200e-02 5.84363997e-01 7.73467183e-01 9.39406812e-01 -2.94662863e-01 7.46266663e-01 1.23727834e+00 1.51304290e-01 8.76790226e-01 -1.40509260e+00 -5.15113294e-01 4.22741085e-01 6.03932142e-03 -5.84170043e-01 -2.74251461e-01 6.01197779e-01 -3.45678955e-01 8.28828573e-01 1.51073083e-01 1.09312689e+00 8.78397465e-01 -5.63285723e-02 9.79813993e-01 1.11255360e+00 -4.16603953e-01 5.04879296e-01 1.37630120e-01 -7.08929896e-01 8.49082291e-01 -2.22711951e-01 6.20977342e-01 -4.44257826e-01 -7.78195709e-02 1.09622955e+00 -1.69419289e-01 -1.05966106e-01 -1.00335014e+00 -1.14728045e+00 9.18451190e-01 4.78058219e-01 4.48577143e-02 -4.91816938e-01 5.40201604e-01 2.28284985e-01 7.99549699e-01 -1.46837026e-01 1.07571638e+00 -5.53180516e-01 -5.07029891e-01 -4.00715202e-01 5.68800628e-01 8.76417935e-01 8.05331528e-01 8.86208057e-01 1.69913039e-01 3.32238502e-03 6.78268254e-01 6.06986061e-02 3.79781216e-01 4.34561223e-01 -1.67353964e+00 2.34890580e-02 4.13760453e-01 6.38995051e-01 -4.97191191e-01 -2.71711498e-01 -1.72853589e-01 -8.08711872e-02 1.01979148e+00 4.85400170e-01 -4.06160891e-01 -7.44755983e-01 1.80452538e+00 5.60963869e-01 -2.26714134e-01 2.11536303e-01 9.03750956e-01 1.38235748e-01 5.07492185e-01 1.94069028e-01 -3.60138834e-01 9.03335392e-01 -9.52003121e-01 -4.21948493e-01 -2.59365171e-01 8.71448278e-01 -2.94340223e-01 1.50119436e+00 6.54813766e-01 -1.22805047e+00 -2.32580259e-01 -8.63103688e-01 2.71355689e-01 -7.79385045e-02 -2.15515807e-01 8.20029020e-01 2.56549716e-01 -8.92880261e-01 9.70985115e-01 -1.06546283e+00 -3.34507853e-01 2.36119896e-01 6.45191193e-01 -1.10592216e-01 2.73336262e-01 -9.01339352e-01 1.18171442e+00 4.51240987e-01 -4.09950405e-01 -1.40394616e+00 -4.54431593e-01 -7.62968421e-01 -2.32255924e-02 1.00100493e+00 -4.96075839e-01 2.00684023e+00 -1.15877926e+00 -2.17009759e+00 3.92250270e-01 4.69162405e-01 -3.93216461e-01 6.33723080e-01 -1.69118255e-01 1.97019160e-01 2.96663851e-01 -5.10687977e-02 1.08862281e+00 8.69690180e-01 -1.41580236e+00 -6.83830142e-01 5.34854867e-02 4.87901598e-01 4.26971972e-01 -1.82604134e-01 -2.78126508e-01 -6.56911805e-02 -3.84077668e-01 -4.13925648e-01 -1.09060729e+00 -6.35333598e-01 2.50860602e-01 1.07453600e-01 -2.44036973e-01 7.06777453e-01 1.69306368e-01 7.48948514e-01 -1.88063765e+00 1.06197990e-01 1.47373363e-01 -8.44374821e-02 1.71160281e-01 -3.10905874e-01 7.14543045e-01 3.28648806e-01 -1.69406548e-01 -1.32565349e-01 -1.04257902e-02 1.97773680e-01 6.91099048e-01 -1.37285650e-01 2.61402965e-01 2.67692477e-01 8.04260492e-01 -1.43448639e+00 -5.40648401e-01 1.41564861e-01 1.88450500e-01 -1.03292370e+00 8.03777993e-01 -8.45205486e-01 8.68496776e-01 -7.10222900e-01 4.24955577e-01 -2.82018483e-01 -1.40329227e-01 3.78337592e-01 5.48610628e-01 -2.44951233e-01 4.19143915e-01 -1.18961430e+00 1.51666605e+00 -5.97801208e-01 3.09323221e-01 1.98128283e-01 -8.73486102e-01 7.69396722e-01 2.25522533e-01 5.45497715e-01 -5.09704351e-01 -2.78114267e-02 3.04813027e-01 2.53907591e-01 -6.53069973e-01 4.22703445e-01 -1.27110034e-01 -6.48930818e-02 7.41046906e-01 7.95914456e-02 -8.46861005e-01 3.80804658e-01 2.71791071e-02 1.23843753e+00 6.90668106e-01 3.47002894e-01 9.29868221e-03 1.82191655e-01 7.94672519e-02 4.20380235e-01 1.08927953e+00 -1.10559009e-01 -1.54589359e-02 7.00353503e-01 -3.10904175e-01 -1.00648093e+00 -9.83195543e-01 3.72760475e-01 1.64797699e+00 -9.70780700e-02 -8.87724385e-02 -5.36433041e-01 -9.00975764e-01 9.08679515e-02 8.37892354e-01 -6.67424262e-01 -6.05547838e-02 -7.41919219e-01 -5.22392727e-02 2.09243327e-01 4.20022935e-01 -7.67475367e-02 -1.80469239e+00 -1.53830862e+00 4.66956854e-01 3.66234362e-01 -4.97764200e-01 -4.77803051e-01 6.74475849e-01 -8.95626068e-01 -9.33239937e-01 -5.63568354e-01 -5.85035741e-01 8.78586233e-01 -1.23301066e-01 1.12229013e+00 2.72868037e-01 -3.19435030e-01 7.46519506e-01 -4.00662154e-01 -2.83192664e-01 -6.64604187e-01 2.35849842e-02 -2.75570471e-02 -6.45171702e-01 -2.10893288e-01 -8.50244343e-01 -6.01511836e-01 2.62795120e-01 -8.84505332e-01 -3.24547440e-02 7.32241988e-01 1.04858112e+00 3.54943097e-01 -2.04752862e-01 7.44835675e-01 -9.41096127e-01 9.05005455e-01 -3.19301069e-01 -1.05012548e+00 2.84899734e-02 -7.81668365e-01 5.76293826e-01 7.51174629e-01 -9.03813720e-01 -7.46272266e-01 3.49505424e-01 9.13119391e-02 -3.07065278e-01 -2.05801472e-01 2.43705541e-01 2.57762134e-01 -2.74210542e-01 7.77228534e-01 1.39878884e-01 1.87962100e-01 -1.21985130e-01 6.58061802e-01 2.29081228e-01 4.60185975e-01 -1.04251254e+00 7.91483700e-01 -5.20862006e-02 -1.36068508e-01 -4.16956425e-01 -5.77211618e-01 -7.63512999e-02 -2.83369750e-01 -4.69240099e-01 5.10122061e-01 -3.99139047e-01 -1.47544992e+00 -2.33920887e-02 -7.34387696e-01 -1.11253643e+00 -7.04796493e-01 5.10422111e-01 -1.30498660e+00 -8.55335519e-02 -4.53657418e-01 -1.11733925e+00 1.35768697e-01 -1.38742530e+00 7.84923255e-01 3.50816637e-01 -2.78470993e-01 -6.93799078e-01 6.16552345e-02 -1.86286658e-01 4.91987258e-01 6.98060170e-02 7.79071212e-01 -6.04289234e-01 -7.60294497e-01 3.20826292e-01 3.13070744e-01 -5.44519015e-02 3.84132750e-02 -9.28229094e-02 -6.09437943e-01 -4.41931665e-01 -3.95951033e-01 -1.25348258e+00 4.62805927e-01 -7.74932057e-02 1.13809788e+00 -6.10060573e-01 -9.06976089e-02 2.16509208e-01 1.07578325e+00 3.99072707e-01 2.13511810e-01 6.78364396e-01 1.79250285e-01 8.09645176e-01 1.02268982e+00 6.31330192e-01 3.55909854e-01 6.77416682e-01 6.54498458e-01 6.82084709e-02 4.32571173e-01 -4.76646990e-01 3.87584150e-01 1.57696322e-01 -1.64887071e-01 1.01312008e-02 -5.75816572e-01 4.76921380e-01 -2.05836105e+00 -9.54007328e-01 6.69418931e-01 2.13761210e+00 1.41811419e+00 3.80298883e-01 5.18685699e-01 -1.11888826e-01 1.75516028e-02 -1.37834311e-01 -1.02967155e+00 -6.51966214e-01 5.78632951e-01 1.89475492e-01 4.20897454e-01 7.11059511e-01 -8.02396119e-01 1.20261455e+00 6.28117752e+00 2.86190659e-01 -9.88214910e-01 -2.92278916e-01 2.12462738e-01 -1.39995009e-01 -4.77137655e-01 3.44058499e-02 -3.79320592e-01 7.44275227e-02 7.83468127e-01 9.58430246e-02 9.75402951e-01 1.05715132e+00 2.94565350e-01 -4.37565386e-01 -1.52754378e+00 4.47312772e-01 -4.24029201e-01 -1.06533515e+00 -3.68436664e-01 1.15084192e-02 4.88492697e-01 -2.92769253e-01 6.70743063e-02 6.89280212e-01 1.12360871e+00 -1.10382211e+00 6.21211112e-01 4.11573201e-01 5.08846760e-01 -9.20396388e-01 1.33762687e-01 8.25322092e-01 -6.80363834e-01 -3.75465721e-01 9.66838747e-03 -2.39645883e-01 9.92447240e-05 -2.34782040e-01 -1.19441450e+00 -1.05258748e-01 4.15558040e-01 2.83134371e-01 -1.59613267e-01 8.96304309e-01 -5.67056477e-01 3.37851673e-01 -3.87210965e-01 -6.83885276e-01 4.67769057e-01 -1.61632001e-01 4.22884881e-01 8.73956144e-01 1.82816282e-01 3.46200854e-01 7.12470412e-01 7.91162789e-01 1.81502923e-01 9.58350208e-03 -8.10327172e-01 -2.28708118e-01 5.38987339e-01 9.72569644e-01 -4.93736774e-01 -2.25689396e-01 -2.74393577e-02 6.67074323e-01 7.88614035e-01 2.46852860e-01 -5.76988816e-01 -2.63507783e-01 5.62265873e-01 2.24030018e-02 5.86073875e-01 -3.69374335e-01 3.75850022e-01 -8.76602829e-01 -3.21547180e-01 -1.15796971e+00 1.96693361e-01 -6.49905384e-01 -9.45386887e-01 3.50373507e-01 7.67475888e-02 -1.03628433e+00 -1.02655196e+00 -2.88066328e-01 -2.70024478e-01 4.30285245e-01 -1.46208322e+00 -7.27119088e-01 4.39825095e-02 4.77476507e-01 5.46036005e-01 -5.58635518e-02 9.78767872e-01 -1.28109381e-01 -7.61822760e-02 3.53203148e-01 -2.44397864e-01 -2.91200012e-01 6.27179027e-01 -1.53181994e+00 -4.29354347e-02 1.28627345e-01 -3.85067016e-02 4.77668583e-01 1.05323243e+00 -4.78648633e-01 -1.53206277e+00 -5.39700747e-01 8.94428641e-02 -9.92307737e-02 8.06750476e-01 -2.03103170e-01 -7.49994934e-01 7.27030635e-01 1.78835198e-01 -5.29573485e-02 3.13911766e-01 -2.02095625e-03 9.87537950e-03 3.33023220e-02 -1.08757913e+00 9.11709845e-01 6.99964464e-01 -2.80680388e-01 -6.03279114e-01 2.27001995e-01 6.59838200e-01 -5.45024633e-01 -6.76917970e-01 3.21592480e-01 5.63543379e-01 -8.65695536e-01 7.88598835e-01 -7.17800319e-01 2.71101892e-01 -2.60714918e-01 2.03169465e-01 -1.49876308e+00 -4.00830284e-02 -1.01257777e+00 -1.36840224e-01 7.37883329e-01 4.02236998e-01 -6.63877606e-01 9.67332542e-01 4.62362826e-01 5.15049882e-02 -1.02843928e+00 -6.18551135e-01 -6.75601780e-01 8.58699903e-02 -2.04837099e-01 3.88393998e-01 6.51563406e-01 2.96462297e-01 2.14060053e-01 -4.44436103e-01 -3.00351083e-02 3.89065057e-01 1.37630329e-01 1.00974834e+00 -1.03921127e+00 -7.87203312e-01 -4.28465784e-01 1.91501245e-01 -1.34748960e+00 1.26495272e-01 -4.34114486e-01 6.03769183e-01 -1.17867935e+00 -1.76143974e-01 -7.88823247e-01 -7.50863180e-02 7.95706391e-01 4.75085229e-02 -2.55363703e-01 4.22727913e-01 1.32995367e-01 -7.84693897e-01 7.53910005e-01 1.78191721e+00 1.23553917e-01 -5.77515066e-01 9.77766141e-02 -3.79118890e-01 7.36503661e-01 9.25123155e-01 -6.08517468e-01 -7.52679765e-01 -6.21822737e-02 1.85925663e-01 6.03533566e-01 2.66365349e-01 -7.92908609e-01 2.80559272e-01 -7.25666046e-01 1.54101640e-01 -1.61232829e-01 4.90392148e-01 -7.77772188e-01 -2.86896855e-01 6.88985705e-01 -8.93240988e-01 6.39728978e-02 2.37469114e-02 6.03270710e-01 1.80894181e-01 -4.07869071e-01 7.23921120e-01 -5.10837018e-01 -5.53556085e-01 1.40484512e-01 -8.24546099e-01 8.02442580e-02 1.13625026e+00 1.61131863e-02 -3.45283113e-02 -7.79005051e-01 -7.65731394e-01 7.13602424e-01 5.53139865e-01 9.70284864e-02 6.12426817e-01 -9.27799881e-01 -2.18407646e-01 6.08544871e-02 4.19199727e-02 3.14550810e-02 -3.21864128e-01 5.48016548e-01 -2.88217038e-01 1.48070768e-01 -4.30075556e-01 -4.46796149e-01 -1.10538423e+00 6.29744768e-01 4.21179116e-01 -4.26680118e-01 -5.94740808e-01 6.05184793e-01 -3.66205499e-02 -8.11363637e-01 6.52537107e-01 -2.93513596e-01 -3.30080241e-01 -1.56313613e-01 1.99356377e-01 7.35579431e-02 -4.24708754e-01 1.74161017e-01 -2.61757467e-02 1.95770964e-01 -9.71641243e-02 -6.94884062e-01 1.53877223e+00 1.48988217e-01 2.77715147e-01 5.42354465e-01 6.09950602e-01 -1.63476840e-01 -2.02569342e+00 5.26806228e-02 1.24309227e-01 -4.83274192e-01 -2.80335933e-01 -1.03822267e+00 -5.23756564e-01 6.56378925e-01 4.30136949e-01 4.22626168e-01 1.00100267e+00 -6.70527816e-02 5.14170766e-01 1.00931156e+00 6.44024432e-01 -1.30012846e+00 6.86899483e-01 3.94012958e-01 8.62651408e-01 -1.24323845e+00 -3.83735523e-02 3.05460803e-02 -8.25287640e-01 1.24021316e+00 8.88944864e-01 -1.82787120e-01 1.70060322e-01 3.35943252e-01 1.79279655e-01 -1.09623030e-01 -1.26814425e+00 -3.40618700e-01 -1.50812268e-01 6.89154804e-01 1.62925154e-01 -1.21385949e-02 -5.41016795e-02 -2.04583302e-01 -2.89228469e-01 1.23874977e-01 7.14719594e-01 1.44756389e+00 -9.10890520e-01 -1.47609520e+00 -2.00822055e-01 2.51101762e-01 -2.18718871e-01 2.97636330e-01 -2.66740143e-01 9.40142989e-01 -2.38921002e-01 8.19766164e-01 -2.64559120e-01 1.10533740e-02 2.60598332e-01 -1.78889573e-01 7.37410605e-01 -8.62109065e-01 -7.19007373e-01 9.08708647e-02 2.71455348e-01 -8.36470962e-01 -2.80068427e-01 -6.98463678e-01 -1.55296457e+00 1.17265724e-01 -2.91874185e-02 3.81801665e-01 5.68696916e-01 8.29886615e-01 3.79990116e-02 5.72128773e-01 7.41648734e-01 -1.15956056e+00 -1.02042997e+00 -6.39038801e-01 -2.78505296e-01 1.10162906e-01 6.73445702e-01 -9.13464844e-01 -2.44108737e-01 -7.60908201e-02]

[4.12812614440918, 1.6280169486999512]

78f20d98-92a5-4bc1-b7f2-523c4344ba37

hub-dravidianlangtech-eacl2021-meme

https://aclanthology.org/2021.dravidianlangtech-1.28

https://aclanthology.org/2021.dravidianlangtech-1.28.pdf

HUB@DravidianLangTech-EACL2021: Meme Classification for Tamil Text-Image Fusion

This article describes our system for task DravidianLangTech - EACL2021: Meme classification for Tamil. In recent years, we have witnessed the rapid development of the Internet and social media. Compared with traditional TV and radio media platforms, there are not so many restrictions on the use of online social media for individuals and many functions of online social media platforms are free. Based on this feature of social media, it is difficult for people’s posts/comments on social media to be strictly and effectively controlled like TV and radio content. Therefore, the detection of negative information in social media has attracted attention from academic and industrial fields in recent years. The task of classifying memes is also driven by offensive posts/comments prevalent on social media. The data of the meme classification task is the fusion data of text and image information. To identify the content expressed by the meme, we develop a system that combines BiGRU and CNN. It can fuse visual features and text features to achieve the purpose of using multi-modal information from memetic data. In this article, we discuss our methods, models, experiments, and results.

['Yang Bai', 'Bo Huang']

eacl-dravidianlangtech-2021-4

['meme-classification']

['natural-language-processing']

[-3.49633187e-01 -4.89955992e-01 1.03964970e-01 8.04583877e-02 -1.96329311e-01 -4.69141573e-01 7.79335260e-01 4.87096906e-01 -6.60198927e-01 3.93164843e-01 1.72746763e-01 2.32015513e-02 4.64785546e-01 -1.08843076e+00 -1.00455955e-01 -4.43536192e-01 3.36174130e-01 1.89790595e-02 3.66436630e-01 -7.83615232e-01 6.29138708e-01 1.15738757e-01 -1.57190835e+00 6.94955826e-01 5.11482775e-01 1.08413577e+00 2.23661780e-01 4.30351704e-01 -6.94504023e-01 1.09260380e+00 -6.73181534e-01 -3.99878442e-01 -1.42729759e-01 -3.94944727e-01 -5.32952428e-01 2.14618295e-01 1.99620411e-01 -1.47639871e-01 -5.91405213e-01 1.20465767e+00 5.90298116e-01 -5.12249731e-02 5.10096133e-01 -1.11290443e+00 -9.87611949e-01 5.06210089e-01 -8.88259709e-01 5.40862024e-01 5.23532629e-01 -3.35269332e-01 5.64578593e-01 -8.65653455e-01 7.05935001e-01 1.30683804e+00 3.86227399e-01 3.10707837e-01 -4.82291013e-01 -8.27019513e-01 -7.10655525e-02 1.08489811e-01 -1.42381525e+00 -1.85883999e-01 8.18840384e-01 -7.84597754e-01 3.86523277e-01 3.82744223e-01 9.05903697e-01 1.27005363e+00 5.74030638e-01 9.61786747e-01 1.15952778e+00 -3.24257493e-01 -1.76968917e-01 7.32183635e-01 1.56236619e-01 7.38172472e-01 -3.60480279e-01 -4.62090909e-01 -7.30024695e-01 -1.06740162e-01 1.61916733e-01 4.90234047e-01 2.15297863e-02 4.01343048e-01 -1.09187520e+00 1.01135087e+00 4.65714216e-01 6.91748023e-01 4.74911630e-02 -2.62726545e-01 6.84241831e-01 5.66736102e-01 1.33039999e+00 -2.49035042e-02 3.68936419e-01 -1.78170446e-02 -8.82290542e-01 1.49188727e-01 6.16292953e-01 5.64331114e-01 5.87246835e-01 -3.47313970e-01 3.19504403e-02 1.35790586e+00 1.69321448e-01 6.83983386e-01 6.03242218e-01 -9.04972479e-02 6.13322318e-01 1.11311519e+00 -8.25922564e-02 -2.08235621e+00 -5.54491282e-01 -2.43319288e-01 -1.06796312e+00 -3.90060693e-01 9.40154195e-02 -1.63817599e-01 -5.54709494e-01 1.01685548e+00 2.86391556e-01 -2.69618481e-01 -5.57265699e-01 8.72086167e-01 1.30599546e+00 1.11591315e+00 -1.01079373e-02 -1.36611477e-01 1.29564679e+00 -7.00518847e-01 -1.10382867e+00 -1.12279959e-01 4.87785965e-01 -1.05826461e+00 8.88148785e-01 1.24615572e-01 -8.70327234e-01 -4.31784660e-01 -8.89392972e-01 -5.53520508e-02 -1.03477740e+00 -1.16651230e-01 2.90525228e-01 4.46295440e-01 -7.07935989e-01 3.02603602e-01 -3.46844584e-01 -7.50267506e-01 5.46227455e-01 5.02675772e-02 -4.02151674e-01 2.72010535e-01 -1.37988758e+00 6.41391218e-01 2.30032697e-01 1.54882312e-01 -3.06003779e-01 -1.10960297e-01 -5.49505889e-01 -3.25234711e-01 1.26802966e-01 1.01844603e-02 6.21806920e-01 -1.43298602e+00 -9.67678845e-01 1.51426697e+00 3.00429285e-01 8.21339488e-02 7.36761153e-01 -9.21664089e-02 -9.10595298e-01 2.24311456e-01 2.91165113e-01 4.70434755e-01 7.79214382e-01 -7.89926469e-01 -8.11561823e-01 -3.99814695e-01 2.12981954e-01 -4.09326516e-02 -1.32982111e+00 6.26007617e-01 -6.46395326e-01 -6.19672179e-01 1.40755132e-01 -9.99400079e-01 3.32441151e-01 -2.42190138e-01 -7.43489861e-01 -2.95540720e-01 1.37459981e+00 -6.99963033e-01 1.42522943e+00 -2.41604114e+00 -1.27715856e-01 2.03304023e-01 6.86382532e-01 -8.46234113e-02 1.74510196e-01 6.23073697e-01 3.72158706e-01 4.01218325e-01 1.18945450e-01 -5.90369366e-02 -2.34675452e-01 -5.54724336e-01 -3.11421156e-01 7.03422427e-01 -2.43277386e-01 6.77376330e-01 -8.18164229e-01 -9.51997876e-01 3.97732519e-02 3.75694156e-01 -3.54453921e-01 1.00943714e-01 7.97801688e-02 5.26353300e-01 -7.93269753e-01 6.40468478e-01 6.07615888e-01 -5.58194518e-01 -3.00595406e-02 1.86590534e-02 -5.48592865e-01 -9.94184464e-02 -7.75471091e-01 1.03022528e+00 -3.30179244e-01 1.15708280e+00 1.70317844e-01 -8.01455736e-01 9.17661309e-01 1.77761674e-01 7.02274203e-01 -8.34524035e-01 8.58381748e-01 6.77573830e-02 -5.84757566e-01 -8.59596252e-01 7.04654217e-01 5.88029251e-02 -3.40904355e-01 6.51143968e-01 -3.84251326e-01 1.58027187e-01 3.35064828e-01 5.27904570e-01 6.36594832e-01 -2.98011124e-01 8.30611736e-02 -2.14960292e-01 4.74301964e-01 5.22477962e-02 1.13746218e-01 4.39706415e-01 -2.51852274e-01 5.77464402e-01 4.10898954e-01 -6.47463441e-01 -8.88216376e-01 -4.77598310e-01 -1.93482593e-01 1.56147981e+00 5.08380175e-01 -5.66829562e-01 -6.91117883e-01 -5.18557906e-01 -8.82585496e-02 -1.56334206e-01 -8.44673991e-01 1.68090686e-01 -3.66216332e-01 -8.96549881e-01 2.04578623e-01 -1.28799260e-01 1.12487066e+00 -1.26361537e+00 -7.13822246e-02 1.01022571e-01 -4.38546270e-01 -1.06082582e+00 -5.85153043e-01 -4.30872291e-01 -3.69618177e-01 -1.02317357e+00 -8.83151829e-01 -1.07719696e+00 6.80105865e-01 6.32847965e-01 9.42696273e-01 4.53732729e-01 -5.97405434e-02 3.79069060e-01 -5.74509144e-01 -4.64806646e-01 -3.42344016e-01 2.50108778e-01 -1.12134673e-01 5.00273764e-01 4.51172292e-01 -2.51622140e-01 -5.67070246e-01 5.03932595e-01 -1.03779650e+00 5.10672331e-01 -6.00289889e-02 4.60461050e-01 1.77320670e-02 1.38633937e-01 4.66084450e-01 -1.06764460e+00 7.60192633e-01 -9.55390692e-01 -2.43981540e-01 6.71344548e-02 6.90558925e-02 -8.95425439e-01 3.94922644e-01 -5.51048696e-01 -7.93891907e-01 -4.20444816e-01 -6.54815212e-02 2.24538241e-02 1.63115487e-01 6.51217222e-01 1.73183531e-01 -3.82237211e-02 5.45774519e-01 9.57773626e-02 -4.57091369e-02 -3.04533660e-01 -3.24585348e-01 1.61415470e+00 -1.82789192e-01 2.02622831e-01 6.40896916e-01 7.07470059e-01 -4.52917039e-01 -1.26827729e+00 -1.20393598e+00 -6.66319013e-01 -7.60546699e-02 -8.96297336e-01 1.08874023e+00 -8.35924625e-01 -1.06520092e+00 9.48797107e-01 -1.04407334e+00 2.43210971e-01 6.19355738e-01 1.08625472e-01 5.70357665e-02 3.39891702e-01 -9.87591147e-01 -7.81575143e-01 -3.80750000e-01 -7.31761873e-01 6.25178397e-01 2.84593374e-01 -1.29550710e-01 -1.05903697e+00 8.74877870e-02 6.33941233e-01 5.80077708e-01 3.40394020e-01 3.22034955e-01 -6.09162748e-01 -3.56428534e-01 -4.90722448e-01 -4.37443435e-01 2.37973571e-01 -9.06029195e-02 9.51148197e-02 -1.01325786e+00 -2.53948331e-01 8.44077207e-03 -6.08488262e-01 9.79283750e-01 1.49895579e-01 1.23248887e+00 -3.61596733e-01 -4.97655988e-01 1.08807795e-01 1.03435171e+00 3.50940600e-02 6.51408076e-01 7.35713422e-01 9.23478484e-01 8.47548783e-01 5.66018879e-01 5.23993969e-01 3.78304839e-01 4.14747775e-01 5.27392626e-01 -1.72027811e-01 4.73362356e-01 -1.48060650e-01 4.24127936e-01 1.09712839e+00 -3.31776850e-02 -4.94970858e-01 -9.98265505e-01 4.09201384e-01 -1.74771547e+00 -1.14243317e+00 -2.85833567e-01 1.61762965e+00 5.00316083e-01 8.53452161e-02 1.99602693e-01 2.42563710e-02 1.37920511e+00 3.63492429e-01 -1.04899459e-01 -4.56812233e-02 -3.49361688e-01 -6.09282255e-01 1.43595517e-01 7.82129318e-02 -1.40562594e+00 6.37948930e-01 5.39563799e+00 9.88485456e-01 -1.41105378e+00 4.76340204e-01 8.58259559e-01 -2.02556640e-01 -6.53406158e-02 -5.82939267e-01 -8.30285847e-01 9.70069289e-01 6.55156374e-01 1.74155757e-01 2.97664821e-01 6.59633636e-01 2.35451013e-01 -1.09325089e-01 -4.47499514e-01 1.20128632e+00 4.13953543e-01 -1.53585935e+00 -1.82562143e-01 2.10772783e-01 8.84242833e-01 1.39178053e-01 3.55039060e-01 3.01187605e-01 -4.16685998e-01 -6.46813989e-01 9.40025270e-01 3.36591750e-01 4.06164467e-01 -7.24509776e-01 7.70567000e-01 3.80195320e-01 -8.46085906e-01 -2.38271937e-01 -3.11485738e-01 -1.31857827e-01 -2.59463699e-03 7.53339887e-01 -2.45242506e-01 -9.34063345e-02 1.01401675e+00 1.10676181e+00 -7.26585329e-01 8.66913915e-01 1.78276494e-01 5.32328010e-01 -1.52016938e-01 -6.21203065e-01 2.51810908e-01 -3.32988948e-01 5.16882002e-01 1.40725660e+00 2.67292082e-01 -2.14093342e-01 3.65386933e-01 5.71969628e-01 -2.86363155e-01 6.55721605e-01 -5.90203524e-01 -7.11394250e-01 1.24251105e-01 1.50247216e+00 -1.15198123e+00 -2.52991170e-01 -5.29538274e-01 8.43679965e-01 3.19289237e-01 5.90214171e-02 -9.05516803e-01 -3.59921426e-01 -1.04473732e-01 5.48441112e-01 -4.66487378e-01 -1.63375333e-01 1.03168719e-01 -1.27418363e+00 -7.73906857e-02 -6.40246451e-01 4.87224668e-01 -6.21385217e-01 -1.61447597e+00 6.01486862e-01 -3.96759361e-01 -1.40714049e+00 3.67620885e-01 -5.50301075e-01 -5.52079141e-01 4.44563240e-01 -1.16380548e+00 -1.24735284e+00 -4.39414740e-01 7.28876472e-01 5.24647415e-01 -4.88845110e-01 5.69371760e-01 6.41336918e-01 -6.42745078e-01 1.77082896e-01 2.59206235e-01 5.11375487e-01 7.43817568e-01 -7.02349603e-01 -1.00656249e-01 3.53471249e-01 -2.29128540e-01 5.24957180e-01 6.70095265e-01 -8.33371580e-01 -1.07223797e+00 -9.49657798e-01 9.93484378e-01 -4.10053581e-01 1.01526845e+00 -5.47897339e-01 -5.30494869e-01 4.06769723e-01 4.47091430e-01 -3.06859523e-01 8.85538578e-01 1.07653514e-01 -1.41147734e-03 1.01497717e-01 -9.44031000e-01 7.00789034e-01 6.82473600e-01 -6.33024931e-01 -2.71111459e-01 9.20740962e-01 1.79813012e-01 -1.74338117e-01 -4.82827395e-01 -2.50962019e-01 5.10616124e-01 -9.47175503e-01 5.31940043e-01 -3.32005322e-01 8.54352891e-01 -1.29551783e-01 1.52196866e-02 -1.10103655e+00 -7.75603727e-02 -2.99354911e-01 2.85299599e-01 1.36641431e+00 1.43516645e-01 -5.79940081e-01 6.24231815e-01 4.70565856e-02 1.51766613e-01 -2.38725886e-01 -5.62389612e-01 -7.73021281e-02 -1.34230092e-01 -2.38600209e-01 3.56873195e-03 1.47593343e+00 2.80126184e-01 4.42391962e-01 -7.06122696e-01 -2.09452540e-01 2.05586821e-01 9.83794257e-02 6.45178318e-01 -1.28423560e+00 1.77092910e-01 -4.23642576e-01 -5.24267614e-01 -6.71084166e-01 2.97453776e-02 -8.19077194e-01 -4.67519403e-01 -1.59706116e+00 7.70769835e-01 -2.21659437e-01 -2.29733199e-01 2.18895644e-01 2.01547354e-01 1.11191404e+00 1.57532170e-01 5.94258130e-01 -9.88882542e-01 2.78632075e-01 1.58520043e+00 -6.77903771e-01 -2.89316624e-01 -2.79435188e-01 -5.78311980e-01 8.37081254e-01 8.29639018e-01 -4.42800701e-01 1.32971048e-01 -1.07469216e-01 1.36866784e+00 -2.19682068e-01 2.50628114e-01 -7.98565030e-01 2.50406832e-01 -2.13931084e-01 4.15452033e-01 -9.25139606e-01 2.76443273e-01 -6.70968533e-01 -1.48957685e-01 3.06860089e-01 -3.85578513e-01 -4.76023816e-02 -1.83049757e-02 3.96745801e-01 -4.51351553e-01 2.33360723e-01 9.14705396e-01 -2.13394031e-01 -4.87932801e-01 1.84863240e-01 -9.00226951e-01 -2.52485536e-02 1.05047023e+00 -1.92653060e-01 -6.47147954e-01 -5.51899493e-01 -8.42295527e-01 2.19511852e-01 3.24081481e-01 7.21287906e-01 5.71521640e-01 -1.37040150e+00 -7.99098432e-01 -2.30909176e-02 3.87058020e-01 -7.03312039e-01 6.83394194e-01 1.11958992e+00 -5.65366149e-01 1.23365484e-02 -5.03167689e-01 -4.14269418e-01 -1.33123195e+00 3.90535295e-01 2.26727709e-01 1.27791569e-01 -5.49358845e-01 5.25462627e-01 2.70871133e-01 -3.69809985e-01 -4.60740961e-02 5.31709433e-01 -9.88036692e-01 7.96208501e-01 1.04487872e+00 4.48914826e-01 -7.17664361e-02 -9.31159437e-01 -1.43646941e-01 3.02487671e-01 -2.55255282e-01 1.82012618e-01 1.41647828e+00 -5.55916250e-01 -5.93072653e-01 7.71757960e-01 1.44326758e+00 3.24668825e-01 -3.31213355e-01 8.96554589e-02 -3.78316283e-01 -5.16018331e-01 2.52216905e-01 -3.73585731e-01 -1.26939023e+00 9.07713950e-01 6.81333959e-01 1.03872526e+00 5.57001770e-01 2.81434227e-03 1.01225078e+00 2.99385250e-01 1.93587288e-01 -1.89250410e+00 5.01633048e-01 6.63573861e-01 1.02350116e+00 -1.47281349e+00 -5.49201556e-02 -3.38682801e-01 -5.14748514e-01 9.76827681e-01 5.10915637e-01 -1.89990420e-02 1.00838864e+00 3.81082855e-03 3.24265510e-01 -4.73554999e-01 -2.05247894e-01 7.62246549e-02 2.81425804e-01 -8.79451260e-03 5.84585011e-01 -2.04265818e-01 -5.25146842e-01 3.08473200e-01 -1.97724223e-01 -3.14459771e-01 7.22687542e-01 1.08155966e+00 -8.36328506e-01 -6.27324581e-01 -6.98234439e-01 6.75572991e-01 -1.08074212e+00 1.27069861e-01 -5.32248199e-01 5.99881351e-01 1.41402230e-01 1.28630495e+00 3.90448570e-01 -6.96887076e-01 -1.64959401e-01 -2.09016934e-01 -1.09355941e-01 -5.62539220e-01 -7.08301783e-01 5.12243211e-02 1.83209285e-01 -1.98736086e-01 -6.26321316e-01 -3.08490545e-01 -9.77421045e-01 -9.21053588e-01 -5.59522986e-01 1.78857416e-01 9.59379911e-01 8.61464024e-01 6.47013187e-02 2.98668176e-01 1.08356476e+00 -8.11921537e-01 2.81385362e-01 -1.01504815e+00 -8.88212800e-01 6.05627179e-01 5.21349870e-02 -6.74798727e-01 -5.44159055e-01 -1.90844797e-02]

[8.516146659851074, 10.72548770904541]

e6920023-7a6f-490f-a00d-29749fdf4fcb

large-language-models-as-counterfactual

2305.14791

https://arxiv.org/abs/2305.14791v1

https://arxiv.org/pdf/2305.14791v1.pdf

Large Language Models as Counterfactual Generator: Strengths and Weaknesses

Large language models (LLMs) have demonstrated remarkable performance in a range of natural language understanding and generation tasks. Yet, their ability to generate counterfactuals, which can be used for areas like data augmentation, remains under-explored. This study aims to investigate the counterfactual generation capabilities of LLMs and analysis factors that influence this ability. First, we evaluate how effective are LLMs in counterfactual generation through data augmentation experiments for small language models (SLMs) across four tasks: sentiment analysis, natural language inference, named entity recognition, and relation extraction. While LLMs show promising enhancements in various settings, they struggle in complex tasks due to their self-limitations and the lack of logical guidance to produce counterfactuals that align with commonsense. Second, our analysis reveals the pivotal role of providing accurate task definitions and detailed step-by-step instructions to LLMs in generating counterfactuals. Interestingly, we also find that LLMs can generate reasonable counterfactuals even with unreasonable demonstrations, which illustrates that demonstrations are primarily to regulate the output format.This study provides the first comprehensive insight into counterfactual generation abilities of LLMs, and offers a novel perspective on utilizing LLMs for data augmentation to enhance SLMs.

['Tieyun Qian', 'Shen Zhou', 'Xin Miao', 'Mayi Xu', 'Yongqi Li']

2023-05-24

['sentiment-analysis', 'relation-extraction']

['natural-language-processing', 'natural-language-processing']

[ 4.56129909e-01 7.00211346e-01 -4.05072957e-01 -3.75062495e-01 -6.03341937e-01 -4.92867380e-01 1.18046439e+00 4.00921553e-02 -4.05010074e-01 1.29504979e+00 5.89653730e-01 -7.85105228e-01 1.28959715e-01 -7.74993598e-01 -8.28116179e-01 -5.58134839e-02 -9.22980011e-02 2.04516321e-01 -6.91351593e-01 -4.74886566e-01 2.96843320e-01 2.55898386e-01 -1.45463955e+00 4.80343103e-01 1.30173969e+00 3.89687449e-01 -4.07280736e-02 4.43147331e-01 -1.47965044e-01 1.05506659e+00 -1.03273726e+00 -6.80352390e-01 1.98998377e-01 -4.24494594e-01 -6.85757041e-01 -1.65387332e-01 2.98710704e-01 -4.24547702e-01 3.33642587e-02 7.54368246e-01 5.51711619e-01 8.31701979e-02 6.66475773e-01 -1.68993533e+00 -8.92492592e-01 1.23858356e+00 -1.83425203e-01 1.08476885e-01 5.52583039e-01 7.43748307e-01 1.10596740e+00 -4.11944002e-01 7.19456017e-01 1.60689318e+00 5.34527659e-01 7.87528336e-01 -1.53697562e+00 -1.04298353e+00 2.44883999e-01 -3.16089213e-01 -7.96664059e-01 -6.49165213e-01 8.23525786e-01 -2.86704153e-01 1.20343161e+00 3.96450847e-01 6.31368101e-01 1.53393364e+00 1.63823962e-01 8.05032015e-01 1.58275545e+00 -5.39935291e-01 2.61730850e-01 4.22703385e-01 -8.49372968e-02 2.61901677e-01 8.78242970e-01 3.67320538e-01 -6.30396426e-01 -2.86279947e-01 6.99041128e-01 -5.14747858e-01 -3.05532575e-01 -8.30367133e-02 -1.37170327e+00 1.05355895e+00 3.83986652e-01 1.58910826e-01 -4.05220687e-01 2.80752242e-01 3.84957224e-01 4.21411902e-01 4.74436760e-01 1.65320611e+00 -7.54313469e-01 -4.72066589e-02 -6.43522143e-01 7.11760342e-01 8.64789784e-01 6.57003641e-01 2.52808094e-01 3.97703975e-01 -4.81905431e-01 6.93786144e-01 -6.83546439e-02 5.07008135e-01 1.06019998e+00 -7.22726107e-01 8.12501609e-01 6.97038710e-01 2.74553746e-01 -6.46017432e-01 -4.45170343e-01 -3.97880882e-01 -5.88674188e-01 2.66032636e-01 5.57205021e-01 -4.58136320e-01 -6.32395506e-01 2.25372291e+00 1.29524395e-02 -2.80509204e-01 3.58299077e-01 6.32844448e-01 6.29761934e-01 3.85398299e-01 4.69785511e-01 -3.85481417e-01 1.25345290e+00 -4.85822350e-01 -7.44218946e-01 -6.83380246e-01 1.14262962e+00 -5.17801046e-01 1.77987325e+00 3.44293378e-02 -9.89959359e-01 -4.17367578e-01 -1.10799921e+00 1.02254078e-01 -4.60221410e-01 -3.54668014e-02 1.27354944e+00 8.80510211e-01 -6.53856158e-01 6.47119761e-01 -3.09767812e-01 8.05382133e-02 5.62270463e-01 1.23908110e-01 -2.64580280e-01 1.80655882e-01 -1.68902469e+00 1.11051965e+00 6.43986285e-01 -2.14976221e-01 -6.16446435e-01 -9.61215436e-01 -1.30352521e+00 -7.13548511e-02 5.11040986e-01 -8.61710727e-01 1.39589405e+00 -9.51586068e-01 -1.32043624e+00 5.63928068e-01 6.67788163e-02 -1.00238371e+00 6.93131506e-01 -1.50715709e-01 -3.92025620e-01 -5.43229520e-01 3.78806621e-01 9.97223377e-01 5.42368293e-01 -1.14974248e+00 -2.32709110e-01 -1.67247266e-01 3.89862806e-01 4.34824944e-01 -2.46422932e-01 -2.71172941e-01 5.23559809e-01 -9.52236593e-01 -3.36433411e-01 -7.49768734e-01 -4.42496866e-01 -5.95070243e-01 -7.89511085e-01 -2.61709511e-01 3.06411654e-01 -2.82713830e-01 1.28924334e+00 -1.69920623e+00 -6.38266981e-01 -1.07042238e-01 -3.99162769e-02 1.91837043e-01 -2.14480549e-01 1.78687364e-01 -4.93763715e-01 7.96701968e-01 -3.48493713e-03 -1.53909937e-01 3.52849871e-01 -1.99981600e-01 -7.21013725e-01 -1.09314382e-01 4.97304857e-01 1.45505619e+00 -8.50841880e-01 -3.23208004e-01 2.08102241e-01 -1.06853925e-01 -6.70691192e-01 5.56584038e-02 -6.13393307e-01 5.71649559e-02 -1.40800118e-01 3.05476427e-01 3.58095318e-01 -1.36811778e-01 2.98541635e-01 -4.21926863e-02 -9.45442021e-02 1.05408263e+00 -1.13127995e+00 1.15912855e+00 -6.78112984e-01 6.44674301e-01 -5.51308095e-01 -6.25318468e-01 5.74018180e-01 1.02683447e-01 -2.28180870e-01 -7.62891054e-01 6.34611538e-03 2.44809300e-01 4.25989717e-01 -4.10331368e-01 7.48548388e-01 -7.13216960e-01 -3.44720125e-01 8.58989418e-01 -3.28208536e-01 -7.93042243e-01 4.76414979e-01 2.40295365e-01 5.35239875e-01 1.31122023e-01 8.20456862e-01 -3.52663323e-02 -6.05894998e-03 2.65787512e-01 4.26478803e-01 1.00902152e+00 1.15710489e-01 1.55174911e-01 7.06135273e-01 -1.70467392e-01 -9.88972843e-01 -7.63988018e-01 6.97709918e-02 9.49412882e-01 -4.07307267e-01 -1.89473778e-01 -5.68503201e-01 -8.42356861e-01 1.45459712e-01 1.79021454e+00 -6.38354063e-01 -4.31778044e-01 -3.15607667e-01 -1.16115797e+00 7.95928359e-01 5.80054581e-01 6.40383542e-01 -1.38369274e+00 -6.49760008e-01 3.84619087e-02 -3.56514305e-01 -9.43483472e-01 -3.85594517e-01 -1.28503874e-01 -1.11279583e+00 -1.04722917e+00 -2.06465065e-01 -1.75121203e-01 6.67716742e-01 1.07640713e-01 1.22702932e+00 -2.21214861e-01 8.00680071e-02 -3.87277678e-02 1.11466140e-01 -1.02246571e+00 -8.63746405e-01 -1.64897546e-01 2.73161083e-01 -7.07736611e-01 2.13652536e-01 -5.41556478e-01 -2.78035522e-01 2.99652526e-03 -9.38236415e-01 5.16630292e-01 8.10631812e-01 9.19652998e-01 6.38218299e-02 -8.22012797e-02 1.08541799e+00 -1.33761621e+00 1.55518746e+00 -2.46111959e-01 -3.45947295e-01 1.67590573e-01 -1.00023031e+00 3.83446366e-01 8.50497305e-01 -6.37406409e-01 -1.40057826e+00 -3.92968088e-01 7.20121786e-02 2.70201445e-01 -2.29702249e-01 5.44477701e-01 -2.40692183e-01 5.15274405e-01 1.01286495e+00 6.27356721e-03 1.18739299e-01 -5.71270734e-02 8.06971729e-01 4.09351379e-01 2.91206062e-01 -6.26634419e-01 7.32257843e-01 2.68589139e-01 -3.07925612e-01 -4.92728412e-01 -9.48831141e-01 3.78566384e-01 -1.86639488e-01 2.17381030e-01 3.52884054e-01 -9.35396791e-01 -7.12374628e-01 1.96054906e-01 -9.40032542e-01 -7.04002798e-01 -6.96187854e-01 5.51221311e-01 -4.81804729e-01 3.22991051e-02 -3.20755631e-01 -1.02831161e+00 -4.25855070e-01 -8.89843941e-01 6.51714563e-01 2.23620355e-01 -8.97775829e-01 -9.68630075e-01 -2.37871289e-01 5.69072008e-01 2.86910921e-01 3.55390131e-01 1.19962466e+00 -9.71297383e-01 -3.16936940e-01 -1.42184600e-01 -1.80966798e-02 3.52891952e-01 2.72487581e-01 -1.86404809e-01 -1.00098658e+00 3.77319683e-03 -1.65569857e-02 -6.60287321e-01 6.88796461e-01 4.05810148e-01 9.27091718e-01 -9.64658499e-01 -2.41063565e-01 2.23493367e-01 8.97897065e-01 1.87035307e-01 5.14491320e-01 4.88022864e-01 2.92952716e-01 6.56824589e-01 6.73935115e-01 2.79426932e-01 5.96905708e-01 3.22790504e-01 2.23965034e-01 -1.86116382e-01 -3.35541666e-02 -7.48083830e-01 5.56797087e-01 -1.35344252e-01 1.72832355e-01 -1.37806252e-01 -7.07292199e-01 5.32257259e-01 -1.40581167e+00 -1.07865345e+00 -4.01372015e-02 2.14753079e+00 1.25706804e+00 6.44485056e-01 -4.47371863e-02 2.51140714e-01 3.62337857e-01 3.29504490e-01 -5.85758150e-01 -6.56446278e-01 -4.07104462e-01 5.95109686e-02 2.82113850e-01 3.18524569e-01 -8.51054609e-01 9.49262619e-01 6.64013100e+00 6.93746090e-01 -1.01913679e+00 -2.53307372e-01 9.84025061e-01 -2.01868832e-01 -8.39874387e-01 -6.79010106e-03 -4.44516718e-01 4.27145362e-01 7.86070764e-01 -6.21233046e-01 2.36076295e-01 8.84428144e-01 6.13973856e-01 -2.47596785e-01 -1.41537154e+00 5.62720001e-01 -8.23023245e-02 -1.72145069e+00 6.53565884e-01 -8.93067941e-02 1.00233269e+00 -4.28645074e-01 2.26334170e-01 7.09267855e-01 7.53573358e-01 -1.37789237e+00 8.13041747e-01 4.96106073e-02 9.20826137e-01 -6.78976893e-01 7.89169967e-01 6.32265151e-01 -4.11046684e-01 -2.39994362e-01 -4.49440181e-02 -8.18335712e-01 1.64829716e-01 4.61830080e-01 -1.55355728e+00 1.48105279e-01 1.89432129e-02 -1.94881726e-02 -6.48776829e-01 1.79344296e-01 -6.87738121e-01 6.08715296e-01 -3.52206640e-02 -2.70841897e-01 1.74281076e-02 2.28978723e-01 3.30173045e-01 1.16221321e+00 3.62290710e-04 7.61734322e-02 -1.58859551e-01 1.25283992e+00 -3.90437543e-01 1.26608752e-03 -1.00789416e+00 -5.59593797e-01 6.16231441e-01 8.62852395e-01 -3.88048202e-01 -5.76139748e-01 -7.81802461e-02 4.90450978e-01 3.68212342e-01 3.52647543e-01 -6.89573646e-01 -2.09405959e-01 8.34035873e-01 1.43798888e-01 -5.99724948e-01 -2.38577928e-02 -1.03340685e+00 -1.19722843e+00 -4.16454114e-02 -1.26485753e+00 3.50281447e-01 -1.04930770e+00 -1.10161579e+00 2.12382987e-01 2.05170214e-01 -8.39112282e-01 -7.48129725e-01 -4.65949416e-01 -8.20757389e-01 7.92787313e-01 -1.38676798e+00 -1.07695520e+00 1.48064777e-01 1.82803497e-02 6.27910614e-01 -4.88243774e-02 8.04029286e-01 -3.75830293e-01 -6.19784892e-01 6.79034591e-01 -4.53528911e-01 1.12813704e-01 7.30456948e-01 -1.23307908e+00 8.54432464e-01 8.92741799e-01 2.06426963e-01 1.22084284e+00 9.63727951e-01 -9.14250433e-01 -1.05574179e+00 -9.69009578e-01 1.10386741e+00 -5.64740717e-01 5.59116006e-01 -4.20239478e-01 -5.35929561e-01 7.53712416e-01 3.96839306e-02 -5.84554195e-01 7.98414052e-01 2.82976478e-01 -3.99540484e-01 2.66538322e-01 -1.23017871e+00 1.27646124e+00 1.11471200e+00 -3.29952866e-01 -1.08699632e+00 2.57428616e-01 9.94614124e-01 -3.70051265e-01 -3.10997874e-01 4.43223923e-01 4.44329441e-01 -7.34989941e-01 9.43656564e-01 -1.24955881e+00 1.04033017e+00 -3.89756896e-02 1.96918428e-01 -1.82633710e+00 6.69453442e-02 -7.81739950e-01 8.31870064e-02 1.30427933e+00 9.39134240e-01 -8.93626809e-01 5.21355808e-01 1.22401774e+00 1.23450585e-01 -6.52905703e-01 -4.73945409e-01 -6.81812048e-01 8.64173323e-02 -1.01061797e+00 9.75382030e-01 1.28629637e+00 3.21535319e-01 5.52636564e-01 -2.21488833e-01 -1.78028286e-01 3.67637485e-01 3.13480586e-01 1.01371956e+00 -7.95915604e-01 -1.57981455e-01 -5.12431502e-01 2.22118258e-01 -7.31057644e-01 4.21442389e-01 -1.05126989e+00 -3.22410703e-01 -1.49528730e+00 2.47993752e-01 -4.62648600e-01 1.34459853e-01 4.86293256e-01 -6.16606593e-01 7.22237751e-02 5.32963097e-01 -1.62233829e-01 -1.24883510e-01 5.31823277e-01 1.34036875e+00 3.11875418e-02 -5.39243221e-01 8.86043087e-02 -1.51564598e+00 9.65906680e-01 9.47892606e-01 -2.01866418e-01 -6.87160373e-01 -1.37768745e-01 2.91562408e-01 -1.70518443e-01 5.09885788e-01 -3.42374593e-01 -3.06107968e-01 -6.62805319e-01 5.09945095e-01 -1.16206817e-01 1.02667145e-01 -1.67065650e-01 -8.13764781e-02 4.58175957e-01 -7.47208297e-01 1.12598263e-01 5.34452796e-01 3.01256448e-01 -3.01937964e-02 -1.01268686e-01 3.49132240e-01 -4.01766419e-01 -6.35278881e-01 -3.35446805e-01 -4.18655396e-01 3.99869800e-01 8.00248682e-01 -1.95741788e-01 -5.53400040e-01 -5.92335045e-01 -2.86504209e-01 3.24832827e-01 3.99802357e-01 5.44175208e-01 5.24110556e-01 -1.23602104e+00 -7.36048162e-01 2.40511551e-01 7.65126944e-02 -8.03989917e-02 -6.34555612e-03 5.41882396e-01 -7.33160786e-03 6.45081043e-01 -1.21710472e-01 7.13565499e-02 -9.53533947e-01 4.58761632e-01 1.43057361e-01 -6.49427712e-01 -1.49777427e-01 8.34766388e-01 4.57363546e-01 -5.75705230e-01 -1.58351675e-01 -7.16412663e-01 -2.00845242e-01 1.84033483e-01 4.95530844e-01 2.82827854e-01 -1.51025340e-01 2.41510645e-02 -4.33605611e-02 -4.24273759e-01 -5.21122962e-02 -3.31159949e-01 1.14028406e+00 1.34142876e-01 1.15229063e-01 4.75023180e-01 4.74456608e-01 1.48879111e-01 -1.09565032e+00 1.42804980e-01 1.25127420e-01 -4.08909917e-01 -3.89631152e-01 -1.22580397e+00 -2.87210047e-01 5.31593978e-01 -1.73903152e-01 1.61555916e-01 7.89076984e-01 -1.67269647e-01 3.57785404e-01 3.96672428e-01 3.18152219e-01 -1.16746211e+00 2.92587411e-02 3.37636560e-01 1.33533025e+00 -1.57667708e+00 1.81049258e-01 -2.95344830e-01 -1.12886012e+00 6.59792185e-01 9.79141057e-01 3.69992793e-01 -5.09553775e-02 3.34405035e-01 -4.67977934e-02 8.91060755e-02 -1.12917960e+00 -4.90109362e-02 1.02162391e-01 6.32184148e-01 9.79215682e-01 4.64610487e-01 -5.17952025e-01 8.29371154e-01 -1.13423824e+00 1.36544267e-02 7.53619075e-01 4.66773003e-01 7.54427835e-02 -7.74371803e-01 -4.88474607e-01 9.18414712e-01 -4.07007903e-01 -5.10990083e-01 -6.12149477e-01 1.19405723e+00 -3.84514853e-02 1.05083251e+00 -1.10184968e-01 -1.40029460e-01 3.86709511e-01 4.77649242e-01 2.52690762e-01 -9.29294407e-01 -6.79572880e-01 -5.52545607e-01 7.73090839e-01 -3.37342083e-01 -1.23066299e-01 -6.36598229e-01 -1.28469980e+00 -2.57469684e-01 -4.23304856e-01 1.26498416e-01 5.40368199e-01 1.00588143e+00 2.73958594e-01 5.79067886e-01 2.89677322e-01 -3.76825124e-01 -1.04340041e+00 -1.31532025e+00 -2.33563095e-01 7.35013843e-01 1.53273180e-01 -4.14100438e-01 -3.61101806e-01 -1.60991624e-01]

[10.220510482788086, 8.157829284667969]

69088377-090e-49ae-b3b1-2e0d44584b8a

space-air-ground-integrated-multi-domain

2202.02459

https://arxiv.org/abs/2202.02459v1

https://arxiv.org/pdf/2202.02459v1.pdf

Space-Air-Ground Integrated Multi-domain Network Resource Orchestration based on Virtual Network Architecture: a DRL Method

Traditional ground wireless communication networks cannot provide high-quality services for artificial intelligence (AI) applications such as intelligent transportation systems (ITS) due to deployment, coverage and capacity issues. The space-air-ground integrated network (SAGIN) has become a research focus in the industry. Compared with traditional wireless communication networks, SAGIN is more flexible and reliable, and it has wider coverage and higher quality of seamless connection. However, due to its inherent heterogeneity, time-varying and self-organizing characteristics, the deployment and use of SAGIN still faces huge challenges, among which the orchestration of heterogeneous resources is a key issue. Based on virtual network architecture and deep reinforcement learning (DRL), we model SAGIN's heterogeneous resource orchestration as a multi-domain virtual network embedding (VNE) problem, and propose a SAGIN cross-domain VNE algorithm. We model the different network segments of SAGIN, and set the network attributes according to the actual situation of SAGIN and user needs. In DRL, the agent is acted by a five-layer policy network. We build a feature matrix based on network attributes extracted from SAGIN and use it as the agent training environment. Through training, the probability of each underlying node being embedded can be derived. In test phase, we complete the embedding process of virtual nodes and links in turn based on this probability. Finally, we verify the effectiveness of the algorithm from both training and testing.

['Lei Liu', 'Neeraj Kumar', 'Chao Wang', 'Peiying Zhang']

2022-02-03

['network-embedding']

['methodology']

[-4.18871582e-01 5.46816625e-02 -7.02421963e-01 8.98139849e-02 2.56463587e-01 -4.17478591e-01 2.38824159e-01 -3.15977246e-01 -1.45909443e-01 1.10818708e+00 -2.14551777e-01 -4.46125537e-01 -7.14797854e-01 -1.49983454e+00 -3.23930353e-01 -8.18291664e-01 -3.92013699e-01 7.77246475e-01 3.23274434e-01 -2.87372947e-01 -3.05529863e-01 7.55318820e-01 -1.03584862e+00 -5.60807288e-01 8.71404052e-01 1.20952189e+00 1.08242154e-01 3.07999134e-01 -2.70138204e-01 6.58943772e-01 -8.76235783e-01 7.52318054e-02 1.80776387e-01 -3.30940560e-02 -6.20907903e-01 9.42623913e-02 -5.44160187e-01 -7.72918463e-02 -1.00474215e+00 7.05286622e-01 4.20619041e-01 3.09222609e-01 4.92747843e-01 -2.14449239e+00 -1.98912144e-01 6.40206635e-01 -3.37994576e-01 1.96018070e-01 -1.29773632e-01 2.24662676e-01 1.02400994e+00 -1.60225421e-01 5.48816860e-01 1.04605722e+00 4.68018353e-01 4.15269762e-01 -9.59284544e-01 -1.13843560e+00 5.89369059e-01 5.43572426e-01 -1.30107391e+00 -1.58407971e-01 9.03930843e-01 -2.25192860e-01 7.53018320e-01 5.91416359e-02 1.17187703e+00 1.01036239e+00 1.27602413e-01 7.77108252e-01 3.48198324e-01 1.21050693e-01 3.68025839e-01 2.08663061e-01 -2.59303331e-01 6.07456565e-01 3.52575570e-01 2.97610819e-01 7.92663097e-02 -7.69186541e-02 7.93376148e-01 -6.49482384e-02 -9.79070961e-02 -4.21165437e-01 -9.75924671e-01 7.88476944e-01 9.60283637e-01 2.02010900e-01 -6.15735054e-01 2.99000829e-01 3.08042347e-01 3.69056612e-01 1.38112709e-01 2.67035842e-01 -3.09784710e-01 6.66008741e-02 -6.13150835e-01 -6.97245970e-02 7.48206675e-01 1.09112430e+00 7.14402795e-01 5.87702215e-01 3.64437103e-01 7.61594474e-01 2.94421554e-01 8.26179981e-01 9.17019472e-02 -8.47326875e-01 3.14593643e-01 5.30619860e-01 -1.14162222e-01 -1.19090116e+00 -8.87737274e-01 -9.14203227e-01 -1.27477443e+00 1.35396853e-01 -2.32687995e-01 -7.75320888e-01 -3.96856725e-01 1.86003363e+00 5.72551429e-01 6.68276191e-01 3.30158591e-01 7.65195489e-01 4.43319172e-01 1.02300155e+00 -4.03664447e-02 -3.26028556e-01 8.31187248e-01 -1.11360514e+00 -9.19494629e-01 -1.82055786e-01 5.00974357e-01 -2.02657014e-01 2.77537733e-01 1.02901101e-01 -7.42915571e-01 -3.38204563e-01 -1.34246349e+00 8.04719687e-01 -4.48180735e-01 -1.55216441e-01 6.88222408e-01 5.06456256e-01 -8.24330032e-01 1.63138099e-02 -5.15318155e-01 -3.54747355e-01 4.99034375e-01 6.17419660e-01 -3.27503383e-01 -1.19671874e-01 -1.67422700e+00 5.17421007e-01 4.92807448e-01 5.88615090e-02 -7.31378019e-01 -4.65722144e-01 -9.14270997e-01 -1.13082323e-02 6.25865579e-01 -7.53080070e-01 7.68647432e-01 -7.44929910e-01 -1.44981742e+00 -1.52245194e-01 3.31258476e-01 -3.65115911e-01 2.91693062e-01 4.24531966e-01 -1.26350904e+00 2.56877512e-01 4.81099822e-02 2.41300821e-01 5.81444144e-01 -1.19974279e+00 -1.02747142e+00 2.67621260e-02 4.24158961e-01 -8.20896700e-02 -3.42369109e-01 -2.61781454e-01 -3.69886845e-01 -3.50230396e-01 5.37370294e-02 -9.07391846e-01 -2.85280406e-01 5.30600548e-02 -5.85703671e-01 -3.71337324e-01 1.18593371e+00 -2.63177395e-01 1.33434796e+00 -2.12306595e+00 1.60830945e-01 9.41908181e-01 5.79944313e-01 1.56620100e-01 -3.42521369e-01 5.97100556e-01 2.34860137e-01 1.44753326e-02 2.00805694e-01 2.22991690e-01 7.15395063e-02 4.38115031e-01 1.43490508e-01 2.42695004e-01 -1.77698433e-01 5.07225811e-01 -1.04338348e+00 -4.82584864e-01 2.65847325e-01 1.90337166e-01 -5.00604212e-01 1.76772833e-01 -3.08365285e-01 4.68741834e-01 -8.46493185e-01 7.06129789e-01 7.11597860e-01 -2.85474867e-01 3.66860032e-01 -3.83195192e-01 2.09707484e-01 -4.18326169e-01 -1.22716665e+00 1.22455251e+00 -7.16060698e-01 7.36840189e-01 2.76832908e-01 -1.23356879e+00 9.63885546e-01 4.33252394e-01 8.45828831e-01 -8.60727549e-01 1.99385419e-01 3.17100018e-01 1.74384266e-01 -8.03881645e-01 1.13609001e-01 8.65391642e-02 -3.78610902e-02 2.45690927e-01 1.48056105e-01 3.15007538e-01 -1.78350825e-02 1.94091246e-01 1.25225484e+00 -4.84863013e-01 -9.29814205e-02 1.23310387e-01 3.93436968e-01 -3.51521790e-01 7.10183501e-01 4.62544203e-01 -2.30993122e-01 -3.45515132e-01 5.12520254e-01 -3.47262323e-01 -8.28064561e-01 -1.08717179e+00 -3.21357325e-02 4.09275204e-01 5.11470437e-01 1.54603124e-02 -4.68763232e-01 -7.64807999e-01 2.61601120e-01 7.34323263e-01 -2.75475949e-01 -5.72955310e-01 -1.72612131e-01 -6.64067030e-01 5.37437797e-01 8.57215822e-02 8.56532574e-01 -8.65392268e-01 8.11714155e-04 7.34090745e-01 -1.75100327e-01 -1.48477793e+00 -1.86489746e-01 -1.95025057e-01 -3.26418877e-01 -1.20655072e+00 -1.06699131e-01 -8.26972544e-01 3.31013143e-01 2.64336705e-01 7.10294068e-01 2.45336950e-01 5.52232899e-02 4.04525638e-01 -1.27755016e-01 -1.65929019e-01 -2.52243251e-01 4.44059312e-01 4.59053218e-01 4.16108936e-01 -5.41885057e-03 -9.82886434e-01 -5.82521796e-01 6.85177326e-01 -6.80372298e-01 -1.35105878e-01 6.56851292e-01 7.42987335e-01 2.82447994e-01 9.06347096e-01 1.11383677e+00 -6.89438164e-01 6.04175210e-01 -1.11159098e+00 -7.22918332e-01 8.75341818e-02 -5.17751932e-01 -3.28107893e-01 7.56160617e-01 -3.43921900e-01 -4.19908375e-01 -7.00917721e-01 -9.78004113e-02 -5.46627760e-01 2.04698384e-01 8.43483925e-01 -7.52682745e-01 -2.38028511e-01 1.74116716e-01 5.69872791e-03 2.87706405e-01 7.20072910e-02 9.08761099e-02 9.10922825e-01 -2.41967849e-03 -3.79368961e-01 1.19067466e+00 3.08744192e-01 2.62669057e-01 -9.41305816e-01 -5.40590286e-01 1.50143147e-01 5.44326454e-02 -6.21430695e-01 4.99063790e-01 -6.81030333e-01 -9.74295974e-01 4.01976079e-01 -9.01616752e-01 -5.21348774e-01 -2.25853562e-01 7.78639734e-01 -2.37549037e-01 -1.71519265e-01 -2.77585030e-01 -8.25667560e-01 1.13819733e-01 -1.00790441e+00 1.68657377e-01 3.21907967e-01 2.14466885e-01 -1.34377706e+00 -2.57005602e-01 2.35633045e-01 7.23623276e-01 3.44748318e-01 1.13050520e+00 -4.47389632e-01 -9.35663104e-01 -3.48879367e-01 -3.07805419e-01 4.06194389e-01 1.44892976e-01 1.41918957e-01 -4.37585890e-01 -5.22923291e-01 -3.98868024e-01 -3.75211872e-02 2.63694584e-01 4.18572903e-01 1.27052236e+00 -5.97412884e-01 -6.89804614e-01 9.03885663e-01 1.59185481e+00 5.36664844e-01 5.49716890e-01 4.80331779e-01 6.76171720e-01 7.27970839e-01 6.29061639e-01 5.01014709e-01 5.88973224e-01 5.68589211e-01 1.03772807e+00 -2.13622794e-01 1.08238295e-01 -3.47796351e-01 2.11420968e-01 7.45691240e-01 9.50806364e-02 -8.25828552e-01 -4.86687779e-01 3.47087115e-01 -1.81970191e+00 -1.08470678e+00 2.92140394e-01 1.84324217e+00 -3.45914997e-02 4.53161448e-01 3.02183658e-01 2.05481440e-01 7.14422345e-01 3.13088804e-01 -8.82544219e-01 4.17228341e-02 -3.84776533e-01 -3.69622916e-01 6.97518647e-01 4.88427848e-01 -7.35755026e-01 8.12610388e-01 5.54899454e+00 1.14214528e+00 -1.26917076e+00 9.94487777e-02 2.77068287e-01 6.93080723e-02 -3.41311574e-01 -1.99396059e-01 -4.28847551e-01 9.44411457e-01 8.72192264e-01 -2.42704004e-01 8.39548886e-01 5.42774081e-01 3.81699353e-01 3.64423364e-01 -5.84970891e-01 9.95781064e-01 -2.56662011e-01 -1.63907158e+00 4.19127867e-02 2.61577040e-01 5.08253813e-01 3.54915977e-01 -1.17619775e-01 4.09181774e-01 3.65299791e-01 -8.48235488e-01 4.49561119e-01 4.19887275e-01 8.29313517e-01 -9.95096803e-01 9.16991591e-01 2.99193352e-01 -1.44709718e+00 -6.34516716e-01 -1.58819437e-01 1.48541838e-01 2.23355129e-01 5.23097754e-01 -2.59701312e-01 8.96182299e-01 3.61232728e-01 8.76129031e-01 1.70877457e-01 1.20016646e+00 -6.09716326e-02 6.40268564e-01 -4.70098764e-01 -2.02837005e-01 2.78487742e-01 -4.57054168e-01 8.26759934e-01 4.90933150e-01 3.52965534e-01 -3.27688307e-02 5.57154834e-01 5.77731490e-01 -2.51717538e-01 -1.22544557e-01 -7.66066909e-01 9.84798521e-02 1.21157706e+00 1.24439812e+00 -1.83963269e-01 -4.83650118e-02 -5.28658986e-01 4.48661834e-01 2.18028247e-01 9.48335528e-01 -9.94463205e-01 -7.15470672e-01 1.04520082e+00 1.97723120e-01 5.44951297e-02 -1.62479922e-01 2.98073351e-01 -7.62345314e-01 -7.56128579e-02 -7.48577833e-01 2.83782899e-01 -7.16911733e-01 -1.45059991e+00 9.99987185e-01 -2.83922225e-01 -1.64877498e+00 1.88683942e-01 -3.64310652e-01 -6.80255473e-01 4.72128481e-01 -1.80157113e+00 -1.15031087e+00 -4.94093329e-01 6.97677732e-01 7.35266358e-02 -8.83978724e-01 7.81251431e-01 8.30393076e-01 -1.08919084e+00 8.30915093e-01 2.69119948e-01 2.19247207e-01 -5.17806150e-02 -6.48958623e-01 -3.30514222e-01 6.10441387e-01 -2.77958393e-01 -1.96409784e-03 4.74132866e-01 -4.02702719e-01 -1.35008478e+00 -1.38523519e+00 2.85319865e-01 1.96727768e-01 1.00652754e+00 -1.34373635e-01 -4.21125084e-01 7.14096248e-01 1.28195599e-01 3.40977043e-01 6.82732999e-01 -6.56667724e-02 9.60116088e-02 -5.85047305e-01 -1.32703960e+00 5.39072752e-01 1.11186898e+00 -1.93315789e-01 1.86325490e-01 6.09588087e-01 1.01040936e+00 -1.36495098e-01 -8.25308084e-01 3.24247301e-01 2.79034793e-01 -4.93088245e-01 8.75771880e-01 -6.93874955e-01 -1.80484816e-01 -4.12106365e-01 -2.21679583e-01 -1.68376911e+00 -6.26108944e-01 -5.61796248e-01 -1.97939038e-01 1.13630891e+00 2.58039415e-01 -1.40517271e+00 1.00735128e+00 -1.23823643e-01 -1.42666548e-01 -1.01809371e+00 -1.24811661e+00 -1.03679717e+00 -5.93747377e-01 -4.32517618e-01 1.34471714e+00 1.05000842e+00 -1.93724096e-01 6.04322314e-01 -3.99332553e-01 6.10131681e-01 5.91417849e-01 -1.39881700e-01 7.82708704e-01 -1.43187404e+00 -1.31371424e-01 -4.51341808e-01 -7.89741516e-01 -1.07385433e+00 5.82073450e-01 -8.95573437e-01 -5.14658928e-01 -1.94358230e+00 -5.71693003e-01 -1.44409144e+00 -7.15874732e-01 2.85141885e-01 5.18513680e-01 -2.46753156e-01 -8.82134959e-02 1.11844555e-01 -5.57370901e-01 8.87703359e-01 1.31236792e+00 -6.28924251e-01 -4.09988090e-02 4.72448826e-01 -3.65066171e-01 4.68759269e-01 1.03237784e+00 -1.84513554e-01 -1.03736806e+00 -4.50097412e-01 4.93151508e-02 5.39522946e-01 1.98101133e-01 -1.30505157e+00 2.37763882e-01 -5.88591635e-01 4.65615839e-02 -2.74118215e-01 5.26078939e-01 -1.66474986e+00 3.30608755e-01 4.01555657e-01 2.08350718e-01 1.85442176e-02 -3.61923911e-02 7.93245256e-01 -2.01032013e-01 1.76791057e-01 4.88975555e-01 5.18236041e-01 -7.85584092e-01 1.08980417e+00 -6.19295061e-01 -9.78691652e-02 1.30162632e+00 -4.15923625e-01 -4.89956230e-01 -5.60363412e-01 -6.67689323e-01 9.61946964e-01 2.06193164e-01 2.87774652e-01 7.99224913e-01 -1.50011981e+00 -4.77987975e-01 5.06538987e-01 8.03057030e-02 -1.59175351e-01 6.03488445e-01 7.84103632e-01 -5.01512885e-01 1.36999153e-02 -3.10613364e-01 -3.70769262e-01 -6.72392607e-01 4.63708341e-01 7.26983130e-01 -1.87330037e-01 -3.98483664e-01 5.77854276e-01 -2.41345093e-01 -5.20515561e-01 4.04861540e-01 3.52293134e-01 -5.07829726e-01 -1.36412447e-03 1.66463748e-01 4.10805523e-01 -3.05943549e-01 -5.25167286e-01 -2.46754616e-01 3.75381112e-01 2.04451680e-01 4.35612462e-02 1.24455059e+00 -3.02503526e-01 -1.75300211e-01 2.24739209e-01 1.01847732e+00 -4.22028601e-02 -8.36098731e-01 -4.52813476e-01 -4.69454288e-01 -4.21093315e-01 3.26148033e-01 -5.20638704e-01 -1.88580990e+00 3.66398901e-01 5.17553568e-01 5.83738685e-01 1.10603046e+00 -2.67821431e-01 1.01601827e+00 3.61047626e-01 7.50895381e-01 -1.22423708e+00 7.26234019e-02 3.98997843e-01 4.05294150e-01 -8.53706121e-01 -3.50435257e-01 -4.94593918e-01 -1.84622064e-01 9.67086613e-01 1.00286663e+00 -3.78737971e-02 1.34913015e+00 7.83963501e-02 -3.10493242e-02 -2.49937817e-01 -6.99563742e-01 7.14585092e-03 -4.25105184e-01 8.87466073e-01 -5.77408612e-01 3.37596595e-01 2.18538433e-01 4.14776206e-01 -1.59955129e-01 -2.69886672e-01 4.93666410e-01 5.93011141e-01 -4.58450615e-01 -1.13344562e+00 9.65849608e-02 8.04915905e-01 2.78273433e-01 3.99945021e-01 4.54329878e-01 8.22444201e-01 2.35410169e-01 1.12628543e+00 2.39336759e-01 -1.07468271e+00 4.63367879e-01 -6.30360603e-01 2.99379765e-03 -1.85207948e-01 1.11125819e-01 -3.87426227e-01 3.05937797e-01 -4.70062882e-01 -1.67298019e-01 -2.27257371e-01 -1.43525445e+00 -8.31341445e-01 -3.38166207e-01 6.72173202e-01 6.10291421e-01 8.34676147e-01 4.55138743e-01 1.25644839e+00 1.43388510e+00 -4.59933430e-01 -1.19049370e-01 -5.09985149e-01 -1.14948511e+00 -7.47473724e-03 4.89049494e-01 -1.03800023e+00 -3.82549822e-01 -1.00237560e+00]

[5.875757694244385, 1.7134358882904053]

5207146f-3b67-48f5-93ad-fffa3733e8f3

this-email-could-save-your-life-introducing

1906.03497

https://arxiv.org/abs/1906.03497v1

https://arxiv.org/pdf/1906.03497v1.pdf

This Email Could Save Your Life: Introducing the Task of Email Subject Line Generation

Given the overwhelming number of emails, an effective subject line becomes essential to better inform the recipient of the email's content. In this paper, we propose and study the task of email subject line generation: automatically generating an email subject line from the email body. We create the first dataset for this task and find that email subject line generation favor extremely abstractive summary which differentiates it from news headline generation or news single document summarization. We then develop a novel deep learning method and compare it to several baselines as well as recent state-of-the-art text summarization systems. We also investigate the efficacy of several automatic metrics based on correlations with human judgments and propose a new automatic evaluation metric. Our system outperforms competitive baselines given both automatic and human evaluations. To our knowledge, this is the first work to tackle the problem of effective email subject line generation.

['Joel Tetreault', 'Rui Zhang']

2019-06-08

this-email-could-save-your-life-introducing-1

https://aclanthology.org/P19-1043

https://aclanthology.org/P19-1043.pdf

acl-2019-7

['headline-generation']

['natural-language-processing']

[ 5.72278023e-01 4.07058716e-01 -3.44445743e-02 -2.53170580e-01 -1.25601053e+00 -5.74427068e-01 1.20892131e+00 5.35364449e-01 -4.37631369e-01 1.19608176e+00 9.42649066e-01 -2.06863225e-01 1.43589705e-01 -5.81677794e-01 -5.93184173e-01 -1.83606502e-02 5.48791587e-01 8.40477824e-01 7.67211840e-02 -3.34594518e-01 9.02922153e-01 1.02374859e-01 -1.10411847e+00 5.94871283e-01 1.13431108e+00 7.03958809e-01 1.09861270e-01 1.17225492e+00 -2.64319390e-01 8.01820397e-01 -1.40785551e+00 -5.38446128e-01 -1.94613740e-01 -9.86553550e-01 -1.45418215e+00 8.63443464e-02 8.73004675e-01 -5.05388439e-01 -3.00317913e-01 6.53290808e-01 1.08737147e+00 1.70787528e-01 8.61454546e-01 -1.07672107e+00 -7.51567006e-01 1.01044142e+00 -3.78255874e-01 3.41320843e-01 7.90988863e-01 -6.69021159e-02 1.24825883e+00 -4.87159371e-01 9.29320335e-01 1.32188022e+00 6.18940532e-01 5.50795138e-01 -1.12686038e+00 -1.76960617e-01 2.54005045e-02 -2.33602021e-02 -5.59811354e-01 -7.48358309e-01 6.43868208e-01 -2.85005450e-01 9.29456532e-01 3.73077393e-01 2.06972823e-01 1.46703446e+00 3.36429179e-01 1.11099315e+00 7.88931668e-01 -5.78150094e-01 2.34264627e-01 6.90989792e-02 4.79858667e-01 4.72526759e-01 2.65448630e-01 -6.29800975e-01 -6.90099835e-01 -4.32071716e-01 2.35666677e-01 -6.30171716e-01 -4.66559649e-01 2.81587362e-01 -1.35721648e+00 7.99761593e-01 1.56286806e-01 3.08865786e-01 -7.17056215e-01 2.63906181e-01 5.83611548e-01 1.96071282e-01 7.01732874e-01 9.28879142e-01 -3.37312698e-01 -3.57125700e-01 -1.31441653e+00 6.43380642e-01 1.31826854e+00 8.62281978e-01 2.79261827e-01 -1.02906257e-01 -8.15878034e-01 9.25008535e-01 -1.00724123e-01 1.93379954e-01 6.44407570e-01 -1.05019045e+00 6.31977975e-01 5.25279522e-01 3.02616894e-01 -7.53571570e-01 -4.31388557e-01 -6.45295739e-01 -7.75878370e-01 -3.73588353e-01 2.02917278e-01 -5.04704237e-01 -4.14219022e-01 1.30594647e+00 -1.48425266e-01 -3.31620306e-01 5.25474511e-02 3.21063578e-01 1.14926922e+00 9.38806951e-01 -3.29325855e-01 -6.31824553e-01 1.24299359e+00 -1.04410088e+00 -8.69462609e-01 -1.09018549e-01 5.60691655e-01 -1.10093570e+00 8.99999201e-01 2.32222825e-01 -1.37805533e+00 -5.52558482e-01 -9.20316815e-01 -1.24906585e-01 -1.06527740e-02 4.33108896e-01 3.20777357e-01 3.03683549e-01 -1.27714372e+00 6.96497738e-01 -3.55343282e-01 -7.46460676e-01 2.91952074e-01 5.81982583e-02 -7.65877515e-02 2.54565120e-01 -1.13415909e+00 8.61171603e-01 2.16488853e-01 -3.84995401e-01 -3.64556432e-01 -7.47877836e-01 -4.72090214e-01 -2.76967380e-02 1.75866649e-01 -1.31722367e+00 2.01219916e+00 -6.92013264e-01 -1.42207170e+00 7.37236023e-01 -3.74452859e-01 -7.78262198e-01 9.10226703e-01 -5.66333532e-01 5.26402192e-03 3.67847085e-01 3.26530248e-01 6.72337294e-01 7.12778211e-01 -1.41603374e+00 -6.86012506e-01 -1.02325268e-01 3.25382724e-02 1.42363966e-01 -3.66884410e-01 1.00489840e-01 -3.67445052e-02 -7.46958852e-01 -5.02651513e-01 -8.16908002e-01 -2.73844786e-02 -7.44514823e-01 -1.00380671e+00 -6.60028696e-01 5.64764738e-01 -9.36925828e-01 1.38052690e+00 -1.40472960e+00 2.05618776e-02 -3.79455179e-01 2.31237903e-01 2.21655026e-01 -2.79580116e-01 9.04730320e-01 1.49607524e-01 4.53612208e-01 -1.81707427e-01 -5.03572881e-01 1.91018477e-01 -5.24759114e-01 -6.40855610e-01 -9.97275021e-03 5.11221439e-02 1.16697586e+00 -8.94082546e-01 -6.70774639e-01 -3.37906867e-01 2.42804319e-01 -3.90432119e-01 3.13869178e-01 -4.17431831e-01 4.60855253e-02 -4.04557765e-01 5.51894400e-03 2.49877289e-01 -3.00851017e-01 -7.06944317e-02 -1.91445619e-01 -2.98929513e-02 7.64903009e-01 -5.75180292e-01 1.34083474e+00 -5.76566219e-01 9.88793969e-01 -2.72405863e-01 -6.30435705e-01 7.65301883e-01 4.20746803e-01 2.48680353e-01 -4.71785456e-01 1.21798493e-01 2.13174671e-01 -2.05363780e-01 -2.35430583e-01 1.27985036e+00 2.31012017e-01 -1.52324617e-01 1.14586389e+00 8.26785266e-02 -4.07849550e-01 5.13582230e-01 8.43553364e-01 1.22697222e+00 -6.82804212e-02 5.02148032e-01 -2.48503745e-01 4.19728965e-01 -2.24626735e-01 -6.22288138e-02 1.36801648e+00 1.27957225e-01 7.85106540e-01 7.78071105e-01 -1.17797248e-01 -9.73286986e-01 -7.05256879e-01 2.41467759e-01 1.17117262e+00 -1.93483561e-01 -5.38315415e-01 -1.24882698e+00 -1.04853797e+00 -1.86861366e-01 1.22568560e+00 -5.05393267e-01 -8.32242072e-02 -9.16254818e-01 -6.82855546e-01 4.84543443e-01 5.29075861e-01 4.02961373e-01 -1.24208319e+00 -3.82106781e-01 3.74061167e-01 -7.22142935e-01 -1.09102535e+00 -7.41072237e-01 -2.52748460e-01 -7.72941589e-01 -7.43625700e-01 -8.65257800e-01 -7.23602831e-01 4.08849299e-01 2.56743371e-01 1.44357228e+00 1.72020286e-01 7.23425373e-02 4.49733526e-01 -5.01344919e-01 -6.89754128e-01 -1.04151428e+00 8.12007248e-01 -5.43516129e-02 -2.25832909e-01 -3.35727893e-02 -3.18628401e-01 -6.17802262e-01 -2.07972065e-01 -9.67730820e-01 6.17526211e-02 7.18171656e-01 7.05393195e-01 -9.68049541e-02 -2.66727954e-01 1.20061648e+00 -1.04896700e+00 1.56312180e+00 -1.95448846e-01 8.06813594e-03 2.20503435e-01 -4.85914558e-01 2.15590492e-01 6.36924505e-01 1.35841612e-02 -1.23017120e+00 -4.51235384e-01 -1.10099420e-01 6.13823175e-01 -4.34753112e-02 3.30830693e-01 9.99337658e-02 5.25654852e-01 8.43498826e-01 3.51241559e-01 -3.07877541e-01 -5.55410862e-01 4.26173508e-01 1.03461576e+00 5.35558045e-01 -4.12758231e-01 5.47519803e-01 2.64099330e-01 -3.77467722e-01 -9.88832951e-01 -1.10725832e+00 -4.45948184e-01 -5.69330215e-01 -2.21096516e-01 5.89624822e-01 -4.29696113e-01 -5.54704964e-01 4.88896519e-01 -1.91417980e+00 -1.72497913e-01 -2.78449357e-01 2.03276295e-02 -6.89850986e-01 7.32579410e-01 -8.52884769e-01 -6.44138098e-01 -9.51545775e-01 -7.17647254e-01 1.41625941e+00 6.85774460e-02 -1.02507949e+00 -9.54459190e-01 2.09936738e-01 6.40753388e-01 3.85257453e-01 3.70269977e-02 7.50881076e-01 -1.20529318e+00 -2.59617120e-01 -4.99154419e-01 -3.10861766e-01 4.12120163e-01 1.53900176e-01 1.20820999e-01 -7.09932804e-01 -1.52618155e-01 -9.63305235e-02 -3.77405494e-01 1.32469928e+00 4.54889536e-01 7.60155141e-01 -9.27836895e-01 -3.37321937e-01 -1.17350034e-01 7.97362745e-01 -5.85651435e-02 6.65977299e-01 4.31169480e-01 4.65258837e-01 7.29837120e-01 3.61220539e-01 5.83444774e-01 5.20824075e-01 4.51262861e-01 -1.14653744e-02 8.25355649e-02 -2.70178080e-01 -2.63112456e-01 2.79282749e-01 1.04818058e+00 1.86598599e-01 -1.04774308e+00 -6.11488938e-01 4.94533271e-01 -1.80685699e+00 -1.17479014e+00 -3.17604095e-01 1.78797138e+00 1.21659017e+00 2.66530156e-01 1.78296044e-01 1.24794662e-01 8.40725839e-01 4.74381834e-01 -5.09522595e-02 -5.76950788e-01 -1.11985914e-02 2.25295663e-01 4.21544127e-02 7.87168443e-01 -1.09776771e+00 8.75787437e-01 6.70522356e+00 6.54778302e-01 -6.93160295e-01 -1.99482486e-01 6.91884220e-01 5.66436537e-02 -3.17179829e-01 -2.23620743e-01 -1.04344988e+00 4.38583285e-01 9.05553937e-01 -5.42394400e-01 -7.84919858e-02 6.01579547e-01 3.92831355e-01 -4.84275445e-02 -1.38633251e+00 5.03906310e-01 6.32264376e-01 -1.53532088e+00 5.08376598e-01 -1.34369023e-02 1.08017743e+00 -3.71069998e-01 -2.63613760e-01 1.77649528e-01 4.11476105e-01 -7.81575084e-01 5.70692658e-01 7.44511664e-01 3.57515484e-01 -5.55658817e-01 1.07988322e+00 2.64730960e-01 -3.86854708e-01 2.77514309e-01 -4.43536565e-02 1.76528934e-02 5.08532047e-01 7.11668968e-01 -1.34934735e+00 5.36965668e-01 -7.24596903e-02 5.06799877e-01 -8.18401814e-01 9.91138756e-01 -1.81198418e-01 6.96445227e-01 1.69512272e-01 -4.05719787e-01 2.06001386e-01 2.69269377e-01 9.07001972e-01 1.59678566e+00 2.44656578e-01 -1.92209437e-01 1.19326942e-01 5.78202426e-01 -7.41450429e-01 3.11034441e-01 -3.23328912e-01 -2.66702086e-01 4.10471797e-01 1.20686412e+00 -6.64627254e-01 -7.23402023e-01 1.05699949e-01 1.26973784e+00 1.48283303e-01 3.57625395e-01 -4.38629359e-01 -8.27736735e-01 1.03242658e-01 1.32874355e-01 9.01607350e-02 -8.62066969e-02 -4.33745056e-01 -9.68067884e-01 1.01483464e-01 -1.02238262e+00 8.73337090e-02 -8.09172034e-01 -1.28014505e+00 8.07712495e-01 -2.01940000e-01 -8.40201139e-01 -7.46865153e-01 -1.02364890e-01 -9.07002807e-01 5.81334591e-01 -1.37540662e+00 -8.94758224e-01 -3.36528361e-01 -1.13921925e-01 1.00611246e+00 -2.97039717e-01 6.09780729e-01 -2.46561676e-01 -3.38355601e-01 5.98757148e-01 2.17868164e-01 7.33522251e-02 1.01079273e+00 -1.43498516e+00 9.85696673e-01 6.85266078e-01 -1.17011189e-01 6.24957979e-01 1.12834513e+00 -7.67117441e-01 -8.81698191e-01 -9.98385310e-01 1.52428603e+00 -6.99539959e-01 5.43767691e-01 -1.37144446e-01 -6.98039949e-01 5.39033890e-01 9.75100458e-01 -1.04878855e+00 6.88310266e-01 -6.68846965e-02 -6.01538531e-02 8.77604187e-02 -7.12059200e-01 7.65015185e-01 9.70201135e-01 -1.72211096e-01 -1.00953472e+00 8.37072551e-01 1.07082188e+00 -9.50375795e-02 -5.42607963e-01 1.95027918e-01 3.49589109e-01 -9.16765392e-01 6.74560010e-01 -7.69518733e-01 1.04269731e+00 1.85858056e-01 3.04975659e-01 -1.65156329e+00 -2.61657923e-01 -1.02853847e+00 -1.70006141e-01 1.72847128e+00 4.63915706e-01 -4.76748168e-01 5.87350965e-01 1.88001990e-01 -4.24165308e-01 -5.71159124e-01 -4.67493206e-01 -4.90241140e-01 4.18802276e-02 -1.58807114e-02 4.30802584e-01 6.79038405e-01 7.75693208e-02 1.00955856e+00 -2.72868037e-01 -6.45618379e-01 4.93573993e-01 2.18319759e-01 1.05605912e+00 -1.38221145e+00 -1.52405128e-01 -8.62109423e-01 -1.96475331e-02 -1.20029712e+00 4.04177338e-01 -7.60404170e-01 1.23746037e-01 -2.35924053e+00 3.86979252e-01 2.93539464e-01 2.49615923e-01 1.60900593e-01 -4.75572139e-01 5.51946349e-02 1.71388328e-01 7.32126981e-02 -9.24348295e-01 6.70421779e-01 1.06452167e+00 -2.46460825e-01 -2.38301754e-01 2.74635792e-01 -1.25825191e+00 5.01360059e-01 9.75832045e-01 -2.14453235e-01 -7.97135532e-02 -2.60561049e-01 2.16186851e-01 4.46451977e-02 1.51082754e-01 -7.51344681e-01 2.19648376e-01 5.02134711e-02 2.59681731e-01 -8.08545887e-01 -4.09725420e-02 1.15064837e-01 -7.49729276e-01 3.42941612e-01 -1.17653668e+00 1.52702972e-01 -6.10494725e-02 4.31155175e-01 -7.34679624e-02 -4.63664532e-01 6.43830955e-01 -1.80602133e-01 -8.96711573e-02 -1.47401601e-01 -7.00570107e-01 5.70168316e-01 4.77225572e-01 7.74895549e-02 -7.92510629e-01 -1.04735351e+00 1.86058450e-02 1.38301298e-01 3.77342999e-01 5.42211473e-01 5.82433760e-01 -1.06216323e+00 -1.40362906e+00 -4.96368319e-01 -1.61418319e-01 -3.10400873e-01 -5.64778633e-02 5.93601286e-01 -6.35812044e-01 8.32554519e-01 5.27722016e-02 -1.78117931e-01 -1.26658857e+00 1.89517453e-01 -1.03539973e-01 -5.20757258e-01 -4.08572465e-01 7.65319884e-01 1.45093545e-01 -2.94952154e-01 3.11337203e-01 -1.58682153e-01 -3.25515330e-01 4.84207809e-01 8.56829166e-01 7.74789631e-01 2.91709542e-01 -4.78997320e-01 -5.81433550e-02 1.66770574e-02 -4.86983210e-01 -3.34714711e-01 1.13710177e+00 -1.57066211e-01 -3.06629598e-01 3.97268832e-01 1.39259684e+00 1.72395855e-01 -5.42287588e-01 -1.64961740e-01 1.62740454e-01 -1.79447919e-01 -1.93171293e-01 -9.32032228e-01 -2.83096880e-01 6.96557581e-01 -1.48732707e-01 6.08682156e-01 7.75899649e-01 1.38611734e-01 1.31370902e+00 6.84006155e-01 -2.43434131e-01 -1.37996995e+00 6.72352672e-01 9.10740793e-01 1.30307710e+00 -1.15528333e+00 3.14097643e-01 -1.05935335e-01 -6.48990989e-01 1.04707551e+00 5.08747458e-01 9.84286144e-03 -1.26021072e-01 3.53365205e-03 -5.42245917e-02 -3.36085483e-02 -1.04899669e+00 2.66830307e-02 4.72079337e-01 4.14721191e-01 1.01500773e+00 -2.48218790e-01 -7.22052515e-01 4.64991629e-01 -9.28078651e-01 -1.97842699e-02 1.00350511e+00 7.89487541e-01 -6.86890781e-01 -9.51110661e-01 -1.56712949e-01 7.61410773e-01 -7.43428171e-01 -1.64848983e-01 -1.26587331e+00 4.16186810e-01 -7.28286803e-01 1.26886845e+00 -7.70526603e-02 -1.80136517e-01 3.58217269e-01 2.96956062e-01 3.68341774e-01 -7.74516761e-01 -9.34057236e-01 -8.03981870e-02 5.72489679e-01 4.37161624e-02 -2.54111350e-01 -7.54641771e-01 -1.00088751e+00 -4.28020358e-01 -3.68059278e-01 4.38353211e-01 6.42181039e-01 8.63970637e-01 5.18298030e-01 7.01865315e-01 6.77489281e-01 -1.00053823e+00 -8.32633376e-01 -1.37664342e+00 1.38136903e-02 4.13986295e-01 3.77202958e-01 1.31749824e-01 -2.77974963e-01 2.32872054e-01]

[12.397648811340332, 9.393540382385254]

3ba2f93b-087a-499b-969f-922935a96a1c

a-subjective-study-of-the-perceptual

2212.01686

https://arxiv.org/abs/2212.01686v1

https://arxiv.org/pdf/2212.01686v1.pdf

A subjective study of the perceptual acceptability of audio-video desynchronization in sports videos

This paper presents the results of a study conducted on the perceptual acceptability of audio-video desynchronization for sports videos. The study was conducted with 45 videos generated by applying 8 audio-video offsets on 5 source contents. 20 subjects participated in the study. The results show that humans are more sensitive to audio-video offset errors for speech stimuli, and the complex events that occur in sports broadcasts have higher thresholds of acceptability. This suggests the tuning of audio-video synchronization requirements in broadcasting to the content of the broadcast.

['Joshua Peter Ebenezer']

2022-12-03

['video-synchronization']

['computer-vision']

[ 2.58021027e-01 -2.49438882e-01 -1.32222697e-01 -3.57882708e-01 -9.36983347e-01 -4.30218846e-01 2.31525321e-02 2.19533727e-01 -3.92047942e-01 5.24336100e-01 6.27578318e-01 1.69538528e-01 1.34296298e-01 -4.81969193e-02 -8.29271376e-01 -4.23643142e-01 -4.99884039e-01 -4.06434268e-01 6.86158955e-01 -1.89197600e-01 4.70894575e-01 2.84168571e-01 -1.93305373e+00 7.26029694e-01 2.79054213e-02 7.16596663e-01 1.21914282e-01 1.13315678e+00 6.96068823e-01 6.68536365e-01 -1.24372005e+00 -2.52252549e-01 4.08204235e-02 -5.78331232e-01 -2.49614209e-01 9.42269191e-02 7.25040495e-01 -6.80977225e-01 -5.13402820e-01 1.04867768e+00 1.05483770e+00 1.72733128e-01 1.17880575e-01 -1.35162640e+00 -1.44375756e-01 7.61735260e-01 -1.69825345e-01 8.73005688e-01 1.40726686e+00 1.14822343e-01 3.06969255e-01 -4.58089292e-01 4.74886507e-01 1.35240519e+00 6.40335202e-01 2.71946371e-01 -1.12944400e+00 -8.29997182e-01 -3.47255558e-01 5.17403245e-01 -1.63618934e+00 -9.77169096e-01 5.77822685e-01 -3.87802780e-01 9.85620320e-01 5.15322685e-01 9.95654404e-01 1.14386714e+00 5.78846395e-01 1.63179368e-01 8.28331769e-01 -5.76299489e-01 2.20969290e-01 2.80504197e-01 -1.83589175e-01 -4.62878168e-01 2.56093621e-01 4.14667875e-01 -1.04015231e+00 -1.17235437e-01 4.75356758e-01 -9.74212646e-01 -4.12240952e-01 4.00059253e-01 -7.50726640e-01 4.85130250e-01 -3.70464653e-01 3.84473264e-01 -1.92176491e-01 3.15222502e-01 8.55978668e-01 5.29399633e-01 3.67439896e-01 9.51462537e-02 2.12903455e-01 -6.82228923e-01 -7.51690924e-01 4.66580510e-01 5.10775387e-01 8.98767710e-01 -3.01475435e-01 4.18549478e-01 -1.41890168e-01 6.19692981e-01 7.45686963e-02 6.08641088e-01 3.32376122e-01 -1.11658072e+00 4.33167726e-01 -5.07748425e-01 1.65041864e-01 -1.32656121e+00 -2.63839483e-01 8.81559327e-02 4.73874927e-01 -8.44289642e-03 3.50681841e-01 -2.72995204e-01 -2.64939845e-01 1.45943201e+00 2.51030028e-01 1.99228644e-01 -6.91296905e-02 1.02405918e+00 7.24703312e-01 8.59315217e-01 2.73469865e-01 -6.63029313e-01 1.33642232e+00 -1.64614655e-02 -1.37217581e+00 1.51093230e-01 2.65711308e-01 -1.19575965e+00 9.95050251e-01 7.14752972e-01 -1.32662868e+00 -9.17302668e-01 -1.14262831e+00 4.26673949e-01 4.36420441e-01 -2.42170066e-01 -1.88038386e-02 1.31606340e+00 -9.98475373e-01 2.36597136e-01 -3.37709099e-01 -2.65613228e-01 -6.94789052e-01 2.55136013e-01 -2.94085234e-01 3.57750833e-01 -1.55664253e+00 7.44764268e-01 3.85078073e-01 1.66087165e-01 -9.60643411e-01 -5.05651772e-01 -7.83926249e-01 -1.76720411e-01 1.01507582e-01 1.28650680e-01 1.42240274e+00 -1.39357615e+00 -1.42433655e+00 8.55023682e-01 1.00958228e-01 -4.59300786e-01 3.84204835e-01 -5.55021048e-01 -1.04957712e+00 8.87832165e-01 -5.80599084e-02 4.97070670e-01 9.88804638e-01 -9.77501333e-01 -5.74702621e-01 2.05348302e-02 3.90389678e-03 4.22161967e-01 2.28287857e-02 8.29392493e-01 -4.92815599e-02 -8.27559114e-01 -1.11087695e-01 -8.96712303e-01 4.63121295e-01 -7.20020652e-01 -8.36870819e-02 9.45118815e-02 5.41341007e-01 -8.89434755e-01 1.52877474e+00 -2.77608275e+00 -2.25822479e-01 3.17627311e-01 -4.41651911e-01 -3.58186960e-02 1.01863302e-01 6.87785804e-01 -4.06968176e-01 -5.79169393e-02 8.25275481e-01 4.41222221e-01 -1.15547284e-01 -2.16963366e-01 -2.04161555e-01 8.03111076e-01 -5.27789116e-01 -1.78158239e-01 -6.33453071e-01 -6.85173750e-01 4.42277156e-02 4.53894496e-01 -5.90719402e-01 6.94334880e-02 3.35362285e-01 1.38360143e-01 -4.02827933e-02 4.85119760e-01 5.36860585e-01 9.33234155e-01 1.07532069e-02 -4.13331658e-01 -1.79102808e-01 3.10179949e-01 -1.29848528e+00 1.34251893e+00 2.22923458e-01 1.12926996e+00 3.22010219e-02 -3.29864860e-01 5.23773491e-01 9.59313869e-01 3.83158058e-01 -7.36056805e-01 3.57537597e-01 5.09244911e-02 2.85661548e-01 -1.32711279e+00 7.35465944e-01 -2.32744679e-01 -1.79996058e-01 8.10442343e-02 -1.69408157e-01 -4.19011861e-01 4.76901144e-01 3.34534019e-01 4.06094760e-01 -2.04971671e-01 -2.19182283e-01 -1.77579537e-01 3.14471304e-01 -3.20771545e-01 3.52058768e-01 5.82626700e-01 -4.53945816e-01 4.78979439e-01 5.98147631e-01 6.00472279e-02 -7.42681563e-01 -1.12047589e+00 -5.97584099e-02 1.12553728e+00 2.32842818e-01 -6.41287148e-01 -1.25499213e+00 2.79520690e-01 -3.22308004e-01 8.93037975e-01 -1.23886377e-01 -3.31569374e-01 -4.84156549e-01 -5.46694025e-02 8.32174718e-01 4.35277253e-01 -1.76272556e-01 -8.67220223e-01 -6.84230804e-01 2.25510657e-01 -7.07387269e-01 -1.46489239e+00 -9.08085287e-01 -4.29462254e-01 -7.41825938e-01 -8.00456285e-01 -4.96457934e-01 -7.47640908e-01 2.15119526e-01 3.41485053e-01 5.87040901e-01 -7.91071206e-02 -3.36361736e-01 9.97853577e-01 -6.68190122e-01 -5.04308760e-01 -7.46444404e-01 -6.26069248e-01 4.49764758e-01 -1.44274637e-01 2.83956349e-01 -3.77809703e-01 -3.91318411e-01 7.02928841e-01 -7.72549152e-01 -5.90687096e-01 -2.74642594e-02 1.89668939e-01 2.39206821e-01 4.07553881e-01 5.35777569e-01 -6.93147406e-02 9.78705049e-01 -1.91544086e-01 -3.52947235e-01 -2.88760632e-01 1.24696560e-01 -8.89029860e-01 -3.62470113e-02 -8.61508012e-01 -1.01756263e+00 -2.75867343e-01 -1.40981779e-01 -1.16908453e-01 -2.89277285e-01 1.01629436e-01 -2.58393824e-01 -2.17739642e-01 9.61513162e-01 -1.79576039e-01 2.73155272e-02 1.04534566e-01 -2.86598921e-01 8.04276764e-01 6.61326289e-01 -5.17041743e-01 2.10270271e-01 1.60460733e-02 -4.39003050e-01 -1.52632880e+00 -3.99994589e-02 -2.62783378e-01 1.43991902e-01 -1.23649299e+00 9.15782988e-01 -1.17736173e+00 -5.26206732e-01 7.08249390e-01 -8.68565083e-01 -4.41703871e-02 3.24398763e-02 1.33257377e+00 -6.96803689e-01 6.03825569e-01 -8.22421193e-01 -9.28731382e-01 2.35156521e-01 -1.19430566e+00 7.09194839e-01 1.89553380e-01 -6.91505015e-01 -2.07642063e-01 7.32898042e-02 5.72974563e-01 8.81883055e-02 1.17477521e-01 2.26828232e-01 -2.76533365e-01 -1.34674326e-01 -3.97665322e-01 5.90095401e-01 3.77867252e-01 3.14318314e-02 3.23122561e-01 -1.10647452e+00 -3.89903784e-01 2.53375322e-01 -4.69227374e-01 -1.16954548e-02 9.89191115e-01 6.80263579e-01 -9.27871764e-02 7.94556290e-02 7.87711740e-02 1.09089017e+00 7.20386982e-01 1.06040537e+00 4.72727343e-02 -4.13473770e-02 9.34149265e-01 9.48159277e-01 4.33082163e-01 -2.94585973e-01 8.27238679e-01 5.30786067e-02 1.75202683e-01 -3.24760340e-02 -2.70290911e-01 8.54663372e-01 7.26783097e-01 -1.89537987e-01 -4.69389707e-01 -3.39397609e-01 4.01589304e-01 -7.82256663e-01 -1.33502829e+00 -3.86743307e-01 2.39037704e+00 5.68382323e-01 5.23400962e-01 6.07095540e-01 6.83631182e-01 1.40600502e+00 1.00651473e-01 1.09553456e-01 -8.82157445e-01 -5.21463305e-02 -5.01602739e-02 3.44071805e-01 6.02669120e-01 -7.76682436e-01 4.08362240e-01 8.35792065e+00 8.02820027e-01 -1.09622943e+00 -1.62397642e-02 2.68892139e-01 -7.18226373e-01 -1.79847032e-01 -1.44956440e-01 -5.29710531e-01 6.92147732e-01 1.46617639e+00 -4.41875517e-01 -4.68247719e-02 4.13580090e-01 1.04582632e+00 -8.48403215e-01 -9.94464099e-01 7.95592546e-01 4.32743788e-01 -6.17018640e-01 -1.05337001e-01 -1.46103874e-01 1.23364389e-01 -5.81196129e-01 1.88253582e-01 -1.79777384e-01 -9.64263797e-01 -6.71148181e-01 1.41397524e+00 1.01243764e-01 9.11941409e-01 -8.66990447e-01 4.72290605e-01 -1.98655009e-01 -1.00005662e+00 5.25534116e-02 -1.54290512e-01 -1.86598405e-01 6.83322430e-01 1.17745265e-01 -5.29039800e-01 -7.66061395e-02 9.52194333e-01 -1.11289978e-01 -2.75434822e-01 1.41570878e+00 -4.84995060e-02 9.62624848e-01 -3.76735628e-01 3.89855995e-04 -3.78754824e-01 2.35118940e-01 9.14520919e-01 1.46537411e+00 4.19388682e-01 2.71461606e-01 -2.82396406e-01 -8.10238644e-02 6.48230195e-01 3.85232091e-01 -7.00678408e-01 -7.70637393e-02 5.72285473e-01 2.84293950e-01 -8.10791016e-01 -1.54453740e-01 -2.71299124e-01 4.19639379e-01 -9.20381606e-01 3.24409097e-01 -1.24635851e+00 -6.27555728e-01 4.48261559e-01 5.11359155e-01 -8.72140676e-02 -2.38189586e-02 2.98564285e-01 -4.59684998e-01 1.93352357e-01 -1.30545378e+00 4.36987221e-01 -1.21444130e+00 -5.76814651e-01 4.98085678e-01 8.28939915e-01 -1.46598792e+00 -2.94062674e-01 -1.65333375e-01 -4.36065257e-01 3.63766640e-01 -4.51701403e-01 -2.52232015e-01 2.52771564e-02 6.96997106e-01 7.20209956e-01 1.18839636e-01 3.66860151e-01 6.70788288e-01 -6.59732567e-03 7.24820256e-01 -5.10682642e-01 -4.21341330e-01 1.00749421e+00 -4.34710294e-01 -2.33814836e-01 8.06872308e-01 -2.77970493e-01 3.66239965e-01 1.52158225e+00 -8.55166435e-01 -1.09349191e+00 -3.13776210e-02 9.51437175e-01 2.14666110e-02 5.80849051e-01 -1.41833931e-01 -4.65772569e-01 4.78896588e-01 5.00182688e-01 -8.38772774e-01 1.04337537e+00 -4.42266703e-01 1.94568813e-01 -3.70402634e-01 -1.07524621e+00 3.91676337e-01 7.26282775e-01 -8.67999554e-01 -8.24539304e-01 2.02782720e-01 7.20020175e-01 -7.15597212e-01 -8.56814265e-01 6.37215152e-02 1.04330635e+00 -1.01628327e+00 8.93551648e-01 -2.26429537e-01 1.46606371e-01 -1.42394379e-01 -1.63027748e-01 -1.16477454e+00 -2.64873598e-02 -1.01589108e+00 8.78791511e-01 1.21833396e+00 2.26981193e-01 -3.23082894e-01 2.12305352e-01 4.19526875e-01 -1.93908110e-01 5.19438744e-01 -1.13953793e+00 -6.45601809e-01 -6.59096062e-01 -6.04391575e-01 -1.39936090e-01 4.85420167e-01 7.00451136e-01 -1.01084396e-01 -6.76585734e-01 2.68635988e-01 3.84311885e-01 -6.94644988e-01 4.01321113e-01 -5.02288222e-01 -2.90910244e-01 2.32860316e-02 -9.52289104e-01 -3.97850960e-01 -2.43173584e-01 6.14672303e-02 1.03901722e-01 -8.01850796e-01 -3.25912684e-01 5.72562933e-01 1.37315048e-02 -3.87486309e-01 1.02145754e-01 2.96259135e-01 3.38668674e-01 -3.19895983e-01 -5.60170934e-02 1.25405446e-01 8.63091528e-01 -8.92392173e-03 -2.84905463e-01 2.06781238e-01 -1.73549831e-01 6.39438272e-01 5.68129003e-01 -7.66451001e-01 -7.38513589e-01 -1.81443140e-01 4.21307325e-01 6.81718528e-01 1.84889421e-01 -1.32019114e+00 8.76749586e-03 2.24418435e-02 1.82999492e-01 -5.24768949e-01 5.63212991e-01 -7.40404665e-01 8.69288146e-01 7.29705095e-01 -5.14244318e-01 4.43680882e-01 6.52792752e-01 6.06806040e-01 -4.17605251e-01 -4.16077912e-01 5.56553543e-01 1.72077030e-01 -4.82744247e-01 -6.18936121e-01 -1.45655310e+00 8.05931985e-02 1.18239784e+00 -6.90612674e-01 -2.05992341e-01 -8.87207448e-01 -1.19214678e+00 -2.37000778e-01 1.40159354e-01 3.76476854e-01 7.14466810e-01 -1.24424756e+00 -5.34315169e-01 2.04311401e-01 -5.02872281e-02 -1.15112638e+00 7.95705140e-01 9.86147165e-01 -9.53678489e-01 1.60996422e-01 -7.71703482e-01 -7.08537459e-01 -1.86304951e+00 2.82251209e-01 2.13517025e-01 8.95812273e-01 -5.10338604e-01 7.81031847e-01 -1.53283566e-01 1.14385831e+00 7.48456120e-01 -3.79201233e-01 -3.33009362e-01 3.36855233e-01 7.65192449e-01 8.05106044e-01 3.64479236e-02 -7.95969784e-01 -4.27443802e-01 4.13635194e-01 1.42676800e-01 -6.43562436e-01 5.04416943e-01 -3.74170840e-01 2.67238200e-01 6.93816841e-01 8.55017960e-01 6.24724329e-01 -7.40644753e-01 5.48160672e-01 -4.25551713e-01 -9.58659649e-01 -4.06784974e-02 -2.43507102e-01 -7.62101352e-01 5.28499305e-01 9.37977791e-01 3.21728408e-01 1.35467160e+00 -1.93281621e-01 5.41563451e-01 -8.56595784e-02 4.48614627e-01 -1.78970540e+00 1.64889857e-01 -2.58872330e-01 1.01899922e+00 -1.22373782e-01 2.35293210e-02 -6.06314659e-01 -8.00939620e-01 1.07628095e+00 4.16559458e-01 -1.39611229e-01 4.90234941e-01 4.09726024e-01 1.44346997e-01 9.45641659e-03 -6.10420465e-01 3.50541264e-01 -1.43802315e-01 9.49214518e-01 5.18302441e-01 -7.58143365e-02 -1.02434969e+00 1.69115573e-01 -3.54637146e-01 1.83271214e-01 1.11743307e+00 8.88229251e-01 -5.06349802e-01 -5.88046432e-01 -1.01748693e+00 -1.29913092e-01 -1.00060284e+00 1.24862187e-01 -3.73963833e-01 6.96719110e-01 1.03518870e-02 1.52369261e+00 1.08380027e-01 -5.68568766e-01 7.07108259e-01 -3.60745676e-02 6.93722010e-01 -1.23344980e-01 -7.21680820e-01 7.80348122e-01 7.03657031e-01 -7.95309484e-01 -7.59978771e-01 -9.24861789e-01 -8.70138049e-01 -2.76365042e-01 -4.01958406e-01 5.31456053e-01 6.31764829e-01 4.86092657e-01 -6.34441003e-02 5.85079968e-01 5.16818225e-01 -8.55211079e-01 -3.54336768e-01 -9.80432928e-01 -9.98902977e-01 4.10407454e-01 1.88705161e-01 -6.05844021e-01 -7.77388752e-01 4.23028260e-01]

[15.052556991577148, 5.661042213439941]

5497b00a-1a6e-44e3-902e-63ce3f76dfd5

coordinet-uncertainty-aware-pose-regressor

2103.10796

https://arxiv.org/abs/2103.10796v2

https://arxiv.org/pdf/2103.10796v2.pdf

CoordiNet: uncertainty-aware pose regressor for reliable vehicle localization

In this paper, we investigate visual-based camera re-localization with neural networks for robotics and autonomous vehicles applications. Our solution is a CNN-based algorithm which predicts camera pose (3D translation and 3D rotation) directly from a single image. It also provides an uncertainty estimate of the pose. Pose and uncertainty are learned together with a single loss function and are fused at test time with an EKF. Furthermore, we propose a new fully convolutional architecture, named CoordiNet, designed to embed some of the scene geometry. Our framework outperforms comparable methods on the largest available benchmark, the Oxford RobotCar dataset, with an average error of 8 meters where previous best was 19 meters. We have also investigated the performance of our method on large scenes for real time (18 fps) vehicle localization. In this setup, structure-based methods require a large database, and we show that our proposal is a reliable alternative, achieving 29cm median error in a 1.9km loop in a busy urban area

['Arnaud de La Fortelle', 'Bogdan Stanciulescu', 'Dzmitry Tsishkou', 'Nathan Piasco', 'Arthur Moreau']

2021-03-19

['camera-localization']

['computer-vision']

[-2.80806482e-01 2.88606673e-01 8.78822953e-02 -4.82804418e-01 -8.23911071e-01 -7.69906402e-01 7.21282184e-01 -1.18331723e-02 -1.08551073e+00 6.06173158e-01 -3.49453211e-01 -2.37609029e-01 2.99411416e-01 -5.65644622e-01 -1.53214216e+00 -4.53069955e-01 -1.30694807e-01 9.86204743e-01 4.80961502e-01 -1.68788627e-01 1.70839638e-01 8.85124862e-01 -1.40578997e+00 -5.05978525e-01 4.20519024e-01 9.72029626e-01 4.01729643e-01 9.48899508e-01 4.62958366e-01 6.32548869e-01 -3.39852631e-01 -2.55070865e-01 5.74065268e-01 3.75035822e-01 -6.85405016e-01 2.27287248e-01 8.90936494e-01 -6.12397850e-01 -4.21962619e-01 9.33445692e-01 3.56097519e-01 -5.43517545e-02 5.73973775e-01 -1.10568309e+00 1.40452728e-01 1.40638098e-01 -1.97801232e-01 -1.43892482e-01 3.65417987e-01 3.00399065e-01 4.54563797e-01 -1.02107775e+00 8.10082734e-01 1.14024651e+00 1.16617000e+00 1.40758738e-01 -1.12673235e+00 -4.00778025e-01 -1.92902699e-01 3.33929569e-01 -1.75872803e+00 -5.46139598e-01 5.33868790e-01 -4.88982558e-01 1.23716331e+00 -3.39293003e-01 5.32115817e-01 9.40518677e-01 2.60429263e-01 2.81986326e-01 6.16394937e-01 -1.97699308e-01 2.70758450e-01 3.12853724e-01 -4.08893198e-01 9.82180297e-01 4.85358328e-01 3.11777651e-01 1.04475945e-01 2.76701391e-01 6.81092978e-01 -4.97862585e-02 4.09214757e-02 -1.10630918e+00 -1.42514241e+00 8.86105239e-01 8.67389381e-01 -1.03530742e-01 -1.00383610e-01 7.63643801e-01 2.33045354e-01 1.67880580e-01 1.51700988e-01 3.74131203e-01 -5.80607474e-01 3.10582779e-02 -6.90385818e-01 2.83692896e-01 7.99118340e-01 1.46931422e+00 1.34980726e+00 -3.24487425e-02 5.30638635e-01 1.73902899e-01 4.83567983e-01 1.00852311e+00 3.03571075e-01 -1.28646684e+00 5.66341162e-01 2.17963710e-01 5.54637253e-01 -1.18583477e+00 -8.84682298e-01 -3.79206836e-01 -7.23204792e-01 5.37714005e-01 4.24165219e-01 -3.63744259e-01 -7.97509968e-01 1.44755697e+00 2.30387613e-01 3.75911772e-01 2.50804573e-01 9.62368965e-01 5.40218294e-01 3.16716045e-01 -5.25150537e-01 3.31374049e-01 1.01459789e+00 -1.11410809e+00 -5.77455126e-02 -6.55479550e-01 8.01087201e-01 -5.84301949e-01 2.67147690e-01 2.68064201e-01 -6.82452321e-01 -6.56285226e-01 -1.32531619e+00 -4.23699105e-03 -4.47274119e-01 5.55676162e-01 1.83590978e-01 3.43333811e-01 -1.63654566e+00 6.42415464e-01 -9.06619728e-01 -8.97071123e-01 -2.43699141e-02 5.19561827e-01 -8.63103449e-01 3.67987603e-02 -7.85969615e-01 1.41721785e+00 5.60478687e-01 2.26781905e-01 -1.12901711e+00 -2.07997739e-01 -1.31426454e+00 -1.69443533e-01 2.50888884e-01 -6.02677882e-01 1.27890301e+00 -5.44236362e-01 -1.61067355e+00 7.52602935e-01 -1.00236557e-01 -1.01231873e+00 8.44538569e-01 -2.85510421e-01 7.77162984e-02 1.25511531e-02 1.38769597e-01 1.29952621e+00 8.07651401e-01 -1.44533849e+00 -5.74879467e-01 -3.31793368e-01 2.22703815e-02 7.05517363e-04 4.13727552e-01 -6.06393099e-01 -6.41611874e-01 8.06847587e-02 2.91143566e-01 -1.39173937e+00 -6.36045754e-01 1.26518428e-01 -3.99920568e-02 1.50985181e-01 6.87963963e-01 -3.63298029e-01 1.50817961e-01 -1.83326685e+00 -7.61299580e-03 1.01800337e-01 7.53040910e-02 8.86253044e-02 -2.64547020e-01 2.94407278e-01 2.15932250e-01 -1.80359781e-01 -1.04946785e-01 -9.04155850e-01 3.60312499e-02 3.21295887e-01 2.23194715e-02 1.08308339e+00 2.87743449e-01 1.14457500e+00 -7.64866948e-01 -2.53599435e-01 7.80233741e-01 6.22138262e-01 -5.29568970e-01 -4.00759615e-02 -1.50681600e-01 5.55472255e-01 -3.74174751e-02 4.55122262e-01 1.02412856e+00 -5.63164912e-02 -6.69492707e-02 -1.37198746e-01 -4.06009912e-01 -1.74464911e-01 -1.16991985e+00 2.08090448e+00 -6.19737923e-01 1.10404670e+00 -1.34381885e-02 -8.26494515e-01 1.12181568e+00 -2.94183996e-02 2.90648758e-01 -5.98251283e-01 4.58562136e-01 2.23387823e-01 -4.14834768e-01 -3.51565361e-01 8.07295620e-01 4.70306218e-01 -1.51944920e-01 -3.35302770e-01 2.37605140e-01 -4.75072235e-01 9.25938785e-02 8.74582399e-03 1.23298967e+00 2.18409449e-01 3.13023418e-01 -3.15658659e-01 5.82999527e-01 2.58005500e-01 2.33204946e-01 7.59579480e-01 -2.08043873e-01 8.17271948e-01 3.04137826e-01 -7.39295304e-01 -1.44541001e+00 -7.74151087e-01 5.31216990e-03 3.54999751e-01 5.46457112e-01 -1.50219709e-01 -5.42512238e-01 -5.59432983e-01 1.88174799e-01 2.91855156e-01 -5.18464148e-01 2.02773333e-01 -7.45929778e-01 -1.64885044e-01 5.00700414e-01 6.54880524e-01 5.63516259e-01 -5.76972127e-01 -9.03824866e-01 2.52509385e-01 4.37648073e-02 -1.69143891e+00 -1.01850592e-01 3.75292271e-01 -6.43544495e-01 -1.19202971e+00 -5.90442181e-01 -8.49409401e-01 6.50429368e-01 3.39493036e-01 9.96432304e-01 -2.04364151e-01 6.65066913e-02 4.84683454e-01 -2.81350881e-01 -2.44680345e-01 -3.14389765e-01 1.86550274e-01 2.16024488e-01 -2.49559656e-01 9.08230096e-02 -4.29732770e-01 -4.71408755e-01 4.09476131e-01 -2.94855863e-01 -3.19345415e-01 8.03236723e-01 5.68063557e-01 4.74447638e-01 -3.52251500e-01 -5.50804660e-02 -2.70430028e-01 -2.17316464e-01 -3.44799668e-01 -1.53917134e+00 -3.26697975e-01 -6.32919192e-01 2.71486998e-01 5.18831015e-01 -1.85017139e-01 -5.40268600e-01 1.01259303e+00 -2.57003009e-01 -6.03771329e-01 -4.34947848e-01 2.14453667e-01 7.25319982e-02 -6.35320425e-01 6.88892245e-01 3.39180231e-02 1.52242631e-01 -3.91203165e-01 5.58915913e-01 3.50109816e-01 9.33187664e-01 -5.24269082e-02 1.03516817e+00 7.40865052e-01 3.48840177e-01 -6.83574140e-01 -2.24228278e-01 -7.96863616e-01 -1.06578040e+00 -1.77604422e-01 8.49170148e-01 -1.53400671e+00 -1.08496547e+00 4.08942997e-01 -1.68018854e+00 -3.96143138e-01 6.53675422e-02 7.86592543e-01 -7.91725338e-01 4.01994348e-01 -2.09422588e-01 -5.44390917e-01 1.89102720e-02 -1.36620975e+00 1.60621989e+00 -2.15418190e-02 2.42373914e-01 -8.83397698e-01 2.00301290e-01 -6.25661761e-02 3.65682662e-01 3.13569635e-01 -8.79001170e-02 -3.97317678e-01 -1.06162381e+00 -4.76730168e-01 -3.65077764e-01 2.50699580e-01 -5.09579957e-01 -3.07820112e-01 -9.35721934e-01 -5.77226341e-01 -3.88100356e-01 -3.84609461e-01 1.08368134e+00 2.30494604e-01 4.95103896e-01 -3.23453955e-02 -6.33417249e-01 8.32431614e-01 1.85789776e+00 -4.61519696e-02 5.72853863e-01 6.85708344e-01 7.88445175e-01 2.15279713e-01 5.24272382e-01 4.33752209e-01 7.08374023e-01 7.84414411e-01 1.16800511e+00 -1.70061958e-03 9.13649201e-02 -1.89275995e-01 4.27914977e-01 3.87901843e-01 -2.05589738e-02 -2.88014740e-01 -1.06256819e+00 6.41092658e-01 -1.98601115e+00 -5.16284704e-01 -2.11259112e-01 2.17304873e+00 5.11618033e-02 2.17835367e-01 -1.97701052e-01 -3.11586916e-01 3.66509855e-01 2.63186940e-03 -4.14773732e-01 -2.08135284e-02 -3.58456895e-02 -3.16654295e-01 1.53388059e+00 9.84302938e-01 -1.37851214e+00 1.10953426e+00 6.29147482e+00 1.30263343e-01 -1.29949212e+00 -1.24528315e-02 7.00254962e-02 4.47956473e-01 2.35062748e-01 1.28035367e-01 -1.10355687e+00 1.57437339e-01 1.07074988e+00 2.77467132e-01 2.90132970e-01 1.31877875e+00 2.56698336e-02 -4.61686611e-01 -1.13012564e+00 1.17043555e+00 3.78127307e-01 -1.33244872e+00 -4.80538279e-01 2.44678482e-01 6.39743328e-01 9.92989480e-01 -2.63802975e-01 2.76805550e-01 3.67511511e-01 -8.69387031e-01 1.00682688e+00 5.24622858e-01 8.35689247e-01 -8.16741645e-01 1.25013435e+00 4.73939270e-01 -1.28118229e+00 7.27529675e-02 -6.07555687e-01 -1.10915065e-01 6.28708228e-02 2.12758422e-01 -1.44368827e+00 5.12112260e-01 6.80644333e-01 9.26333666e-01 -8.75109375e-01 1.18826115e+00 -1.70524448e-01 -8.79154131e-02 -6.62140965e-01 -9.01872758e-03 4.76549000e-01 1.21389970e-01 6.14141107e-01 1.26434720e+00 5.78390121e-01 -5.10059357e-01 3.42863411e-01 5.25757074e-01 2.27368642e-02 -3.53626966e-01 -1.00497949e+00 6.90951049e-01 3.70164812e-01 1.29624128e+00 -6.07040167e-01 -1.72307789e-01 -2.42594779e-01 9.65420604e-01 4.53272760e-01 1.44532576e-01 -9.74328160e-01 -4.16636050e-01 5.42872608e-01 -1.57987088e-01 8.07605267e-01 -6.35854721e-01 1.51042983e-01 -1.13167202e+00 4.72774878e-02 -1.34849384e-01 -4.84167993e-01 -9.42563057e-01 -6.31768465e-01 6.87069893e-01 -6.83578178e-02 -1.48739088e+00 -7.62730479e-01 -1.02502978e+00 -1.00253642e-01 5.49013495e-01 -1.74278569e+00 -1.09355986e+00 -6.22312427e-01 3.51065934e-01 5.23083270e-01 -1.77023649e-01 5.99047124e-01 1.65270254e-01 -9.60449800e-02 3.95384312e-01 4.25399542e-01 2.87740648e-01 6.20883584e-01 -1.23256099e+00 6.96698308e-01 7.61014402e-01 5.28969392e-02 9.79457274e-02 9.51946378e-01 -3.42353493e-01 -1.58444154e+00 -1.53912306e+00 9.90959823e-01 -7.84330249e-01 6.11375630e-01 -5.24192512e-01 -4.61353362e-01 1.09384811e+00 2.23080859e-01 3.66068453e-01 -5.31469166e-01 -3.90398711e-01 -1.76041752e-01 -3.03971320e-01 -1.19067109e+00 1.42332435e-01 9.08309996e-01 -2.64159948e-01 -2.38748223e-01 2.49511719e-01 9.60842431e-01 -7.94580698e-01 -5.46362221e-01 6.08923018e-01 5.19321263e-01 -1.04210079e+00 8.54949951e-01 1.78187087e-01 -1.60993904e-01 -6.31334901e-01 -2.64104545e-01 -1.30032361e+00 -2.77279317e-01 -3.55554193e-01 2.68807650e-01 6.43301487e-01 4.66065794e-01 -6.55636013e-01 9.41756189e-01 6.85051782e-03 -3.11037213e-01 -2.07133561e-01 -1.11398900e+00 -9.63237166e-01 -1.16217606e-01 -7.54912555e-01 4.74553436e-01 3.12469035e-01 -4.95490640e-01 3.38153809e-01 -5.28637111e-01 6.81772470e-01 6.18649423e-01 -3.62854838e-01 1.28585076e+00 -1.19745493e+00 1.61989927e-01 -1.28604412e-01 -1.22007644e+00 -1.35384452e+00 5.34632444e-01 -5.90742111e-01 5.59401572e-01 -1.35934615e+00 -1.52866900e-01 -1.54457152e-01 2.52719581e-01 1.36256829e-01 6.38136685e-01 2.60350585e-01 1.14339896e-01 -8.01069438e-02 -1.00189447e+00 4.55258578e-01 5.79641938e-01 -2.94203073e-01 2.30575830e-01 -8.65188986e-02 1.09913729e-01 9.07556236e-01 7.17159748e-01 -3.95043582e-01 -8.83692279e-02 -7.75411248e-01 1.19189411e-01 3.75587605e-02 8.86337280e-01 -1.68030620e+00 5.88567495e-01 2.97944874e-01 5.07427156e-01 -9.66182649e-01 5.86786270e-01 -1.28359210e+00 1.20080985e-01 5.68723321e-01 1.65581390e-01 3.67359370e-01 3.54475260e-01 6.98642850e-01 -1.52805597e-01 -1.11404456e-01 7.02814519e-01 -9.80297700e-02 -1.30759561e+00 2.30039746e-01 -3.80696535e-01 -4.41899359e-01 9.70428944e-01 -1.88393474e-01 -9.85702276e-02 -4.97290403e-01 -5.36759853e-01 2.62397259e-01 8.38550091e-01 4.11277652e-01 5.48339903e-01 -1.31483996e+00 -5.82902491e-01 2.58967429e-01 4.14935470e-01 4.18496341e-01 -1.85391173e-01 7.31097043e-01 -1.19530416e+00 9.29222584e-01 -9.86258034e-03 -1.19966543e+00 -9.76543844e-01 6.75819993e-01 5.79023778e-01 4.92289141e-02 -3.87887686e-01 6.23716116e-01 -2.33904451e-01 -7.97670722e-01 4.33723897e-01 -7.04420269e-01 -2.15382259e-02 -3.59862119e-01 2.77466923e-01 1.74908996e-01 2.55510122e-01 -1.12453401e+00 -6.54825151e-01 8.75997901e-01 3.90412897e-01 -2.16983855e-01 1.07945395e+00 -4.93214816e-01 1.70904323e-01 2.14744225e-01 1.50874949e+00 -1.40214145e-01 -1.70239031e+00 -1.45671979e-01 2.21260682e-01 -2.67212361e-01 -2.55437158e-02 -3.00567448e-01 -8.30546498e-01 6.54330909e-01 9.75136220e-01 -2.97961205e-01 5.88996828e-01 2.23935425e-01 3.63052368e-01 1.15554869e+00 7.81801224e-01 -7.64356375e-01 -1.82123333e-01 1.14321697e+00 8.47904861e-01 -1.93019342e+00 -7.59795904e-02 2.80040503e-02 -3.60444814e-01 1.20599818e+00 6.64376318e-01 -4.65958178e-01 5.01145363e-01 3.73333871e-01 2.10588858e-01 1.43204585e-01 -4.85860139e-01 -5.06842911e-01 2.42521688e-02 7.09429562e-01 -8.99421647e-02 -8.59043002e-02 3.72794777e-01 -2.11837545e-01 -2.26956874e-01 -2.00807601e-01 5.67486882e-01 7.23098516e-01 -7.04336584e-01 -6.47464991e-01 -4.23655957e-01 -3.19121987e-01 2.02405080e-01 1.28770322e-01 -2.43414462e-01 1.21494234e+00 3.60528201e-01 8.31636965e-01 3.11878502e-01 -6.42239273e-01 4.47380632e-01 -3.16025496e-01 4.42101479e-01 -3.03100616e-01 4.50387634e-02 -5.98205924e-02 4.50551361e-02 -9.95691001e-01 -4.84548360e-01 -6.96366191e-01 -1.10752213e+00 -2.24256769e-01 -2.05815941e-01 -9.73599553e-02 1.26034164e+00 9.50220108e-01 3.97368461e-01 6.85592219e-02 6.59371316e-01 -1.53609967e+00 -4.38661486e-01 -1.03434706e+00 -3.27259272e-01 -8.78140852e-02 8.74506831e-01 -5.86641252e-01 -2.91457176e-01 4.99217287e-02]

[7.618219375610352, -2.125690460205078]

b620f5cc-fa42-44a8-8b4d-3faa836d2664

utility-theory-of-synthetic-data-generation

2305.10015

https://arxiv.org/abs/2305.10015v1

https://arxiv.org/pdf/2305.10015v1.pdf

Utility Theory of Synthetic Data Generation

Evaluating the utility of synthetic data is critical for measuring the effectiveness and efficiency of synthetic algorithms. Existing results focus on empirical evaluations of the utility of synthetic data, whereas the theoretical understanding of how utility is affected by synthetic data algorithms remains largely unexplored. This paper establishes utility theory from a statistical perspective, aiming to quantitatively assess the utility of synthetic algorithms based on a general metric. The metric is defined as the absolute difference in generalization between models trained on synthetic and original datasets. We establish analytical bounds for this utility metric to investigate critical conditions for the metric to converge. An intriguing result is that the synthetic feature distribution is not necessarily identical to the original one for the convergence of the utility metric as long as the model specification in downstream learning tasks is correct. Another important utility metric is model comparison based on synthetic data. Specifically, we establish sufficient conditions for synthetic data algorithms so that the ranking of generalization performances of models trained on the synthetic data is consistent with that from the original data. Finally, we conduct extensive experiments using non-parametric models and deep neural networks to validate our theoretical findings.

['Guang Cheng', 'Will Wei Sun', 'SHIRONG XU']

2023-05-17

['synthetic-data-generation', 'synthetic-data-generation']

['medical', 'miscellaneous']

[ 2.98160970e-01 9.42414925e-02 -1.05926074e-01 -4.61329669e-01 -7.13000536e-01 -6.64840043e-01 6.46810830e-01 6.25410154e-02 -5.45784891e-01 8.64174724e-01 -4.81765941e-02 -1.99097902e-01 -4.33817595e-01 -7.62899876e-01 -8.25506568e-01 -8.04890037e-01 5.11592533e-03 2.55573392e-01 -2.20567614e-01 1.39865875e-01 2.92519838e-01 5.24766028e-01 -1.51529443e+00 -7.98652470e-02 1.22246850e+00 1.07058144e+00 -2.42278855e-02 5.76311827e-01 3.19739461e-01 2.04887524e-01 -7.92943060e-01 -3.99473608e-01 5.37301004e-01 -4.87603694e-01 -6.06677175e-01 -9.35780033e-02 1.63288131e-01 -1.03022397e-01 -9.61536244e-02 1.26124036e+00 4.04020816e-01 2.65693869e-02 1.08788431e+00 -1.71800148e+00 -7.59975016e-01 6.70462549e-01 2.08172053e-01 2.13570237e-01 -1.00168034e-01 1.80905238e-01 1.06366372e+00 -7.55932868e-01 4.85634923e-01 1.09018600e+00 4.79033351e-01 4.30296570e-01 -1.37166429e+00 -5.18436670e-01 -1.30681276e-01 -1.57125071e-01 -1.22764874e+00 -4.10071373e-01 3.66895497e-01 -5.63930094e-01 2.95178354e-01 1.23609431e-01 4.99834418e-01 1.36587691e+00 2.26615593e-01 5.10468423e-01 9.20250833e-01 -3.90017867e-01 4.94136602e-01 4.48744893e-01 4.15589452e-01 4.18142140e-01 7.51953363e-01 4.95601386e-01 -5.93738914e-01 -4.71536163e-03 6.77667141e-01 -3.18786889e-01 -3.31525981e-01 -5.96049547e-01 -1.02993596e+00 8.34257483e-01 1.03016652e-01 1.95890039e-01 -2.76242942e-01 9.35820118e-02 3.69088203e-01 6.43952608e-01 4.18087631e-01 1.03315234e+00 -4.91837233e-01 -1.95509598e-01 -7.48577476e-01 4.57313687e-01 5.97261429e-01 1.04061651e+00 5.43224454e-01 2.84859836e-01 -4.07800823e-01 6.73743665e-01 -3.37455720e-02 3.43173087e-01 6.63101077e-01 -1.21789753e+00 3.03176731e-01 7.20850646e-01 2.67548352e-01 -5.68301737e-01 -2.12477297e-01 -7.91381717e-01 -7.35538006e-01 -4.74410970e-03 8.15813720e-01 -3.34975660e-01 -2.97371417e-01 2.14082241e+00 -8.70622024e-02 -1.81336686e-01 3.76426667e-01 5.47890842e-01 5.57414472e-01 4.50572550e-01 -3.06105185e-02 -2.68890530e-01 5.64120352e-01 -5.70426941e-01 -9.97097492e-02 1.01079322e-01 9.49824452e-01 -3.14761698e-01 1.58682382e+00 7.22897425e-02 -1.02607524e+00 -4.93802339e-01 -1.21802473e+00 3.26362997e-01 -2.60095865e-01 1.15345128e-01 1.58679992e-01 8.15754354e-01 -1.01891983e+00 8.05868089e-01 -5.92161834e-01 -3.00512463e-01 2.56569505e-01 3.16320777e-01 -2.21583501e-01 1.46387458e-01 -1.23933923e+00 7.79802561e-01 6.54828668e-01 -1.98667616e-01 -9.93909597e-01 -1.03840077e+00 -6.92299545e-01 4.10717279e-01 5.41786803e-03 -7.86723256e-01 1.49699938e+00 -1.18904901e+00 -1.11718702e+00 6.76793993e-01 6.48482665e-02 -7.69403398e-01 7.75411367e-01 7.45961592e-02 -1.79874882e-01 -1.31068349e-01 9.93771702e-02 4.21359599e-01 5.46209633e-01 -1.21585023e+00 -5.90058386e-01 -3.63946408e-01 1.43685520e-01 1.03184931e-01 -6.05368018e-01 -2.27868780e-01 -5.43687085e-04 -7.81805933e-01 -9.76975039e-02 -9.36394274e-01 -3.38493623e-02 4.09548320e-02 -4.19155538e-01 -8.26908201e-02 4.54145581e-01 -7.31534436e-02 1.25776148e+00 -2.21963453e+00 -1.30192444e-01 2.13030174e-01 6.04112782e-02 5.06128147e-02 -1.78774223e-01 2.88837671e-01 -2.81003892e-01 4.41666245e-01 -3.83979291e-01 9.42344666e-02 1.47098467e-01 -1.18102483e-01 -3.26953143e-01 4.90302622e-01 2.75633097e-01 8.22664142e-01 -8.10006022e-01 -1.92761838e-01 -2.07742020e-01 4.58472669e-02 -5.94554663e-01 2.44427294e-01 -1.13416668e-02 2.66819060e-01 -5.58962524e-01 2.36887753e-01 4.51750994e-01 -3.00646991e-01 -5.97173069e-03 1.15852021e-01 -4.95047979e-02 1.38146013e-01 -8.29794168e-01 9.89417732e-01 -4.57566649e-01 6.98951602e-01 -3.41266632e-01 -1.22099352e+00 9.79714930e-01 1.41438752e-01 4.34442103e-01 -6.47043169e-01 1.02649912e-01 1.81234643e-01 2.05415934e-01 -1.39914587e-01 2.87038088e-01 -2.81587690e-01 -2.20361173e-01 6.98637009e-01 4.86554839e-02 -7.23157823e-02 3.44273537e-01 4.17959169e-02 8.76259565e-01 -8.14518556e-02 2.37827048e-01 -7.51622319e-01 1.64293602e-01 -4.76137549e-02 5.58481574e-01 8.56119275e-01 -9.11362767e-02 3.66875619e-01 9.84556258e-01 -1.32522821e-01 -1.28481150e+00 -1.52473533e+00 -2.96655893e-01 7.43182123e-01 1.98208213e-01 -1.45473838e-01 -9.88412976e-01 -5.99517763e-01 6.32502735e-02 1.10244024e+00 -9.09964681e-01 -8.32565784e-01 -1.67693540e-01 -1.09699190e+00 7.20260501e-01 6.40406787e-01 4.83189851e-01 -7.19832122e-01 -8.56531441e-01 -1.40609607e-01 1.02876365e-01 -8.65704298e-01 -2.60547847e-01 6.33547604e-02 -9.89512384e-01 -1.26250911e+00 -5.77048719e-01 -5.34318507e-01 7.24437237e-01 3.87547426e-02 1.03554177e+00 -1.64728582e-01 1.54030398e-01 1.82857722e-01 -2.90245891e-01 -4.60813642e-01 -8.13427806e-01 1.41503751e-01 3.81000459e-01 -4.55647148e-02 6.44631088e-02 -5.28670728e-01 -5.27408183e-01 5.64348459e-01 -9.91034329e-01 -7.43417591e-02 5.33194363e-01 9.52728808e-01 3.56596500e-01 7.09677413e-02 1.05444014e+00 -6.64482594e-01 1.09303236e+00 -4.49043721e-01 -5.48013687e-01 4.02355999e-01 -9.54105496e-01 4.86642271e-01 1.01253867e+00 -4.83240634e-01 -1.09092546e+00 -1.86926097e-01 2.79485643e-01 -3.64007413e-01 2.07402989e-01 3.53088886e-01 -2.24471539e-01 2.81069368e-01 9.94976938e-01 2.62201667e-01 1.07573336e-02 -3.06226164e-01 3.31256539e-01 4.78180140e-01 3.49578589e-01 -8.96091759e-01 6.73176527e-01 1.65867537e-01 3.06905899e-02 -8.07616889e-01 -7.96615064e-01 9.18072239e-02 -7.77458906e-01 -9.24195573e-02 2.89204717e-01 -5.13917983e-01 -6.29231215e-01 2.61895239e-01 -8.62089992e-01 -4.60811019e-01 -4.50615764e-01 6.15515769e-01 -8.93898726e-01 6.04154021e-02 -3.40884238e-01 -7.72109866e-01 -2.17429757e-01 -1.30581832e+00 7.05928385e-01 6.15625419e-02 -3.51268500e-01 -9.77373779e-01 -1.65476233e-01 -2.43605837e-01 1.99694261e-01 3.98826241e-01 1.43318987e+00 -8.53824198e-01 -3.42547566e-01 -1.89264372e-01 -3.52173418e-01 5.52171052e-01 4.44360189e-02 1.83139309e-01 -8.57186854e-01 -1.63985908e-01 9.17664841e-02 -2.57428616e-01 8.73606920e-01 5.86236298e-01 1.35924351e+00 -4.83274668e-01 -1.86975196e-01 5.08766770e-01 1.22182071e+00 3.47501755e-01 1.95926070e-01 2.54883826e-01 1.90855131e-01 7.34310448e-01 7.56217659e-01 4.81056362e-01 -2.24835232e-01 4.36138928e-01 2.91190147e-01 2.82037944e-01 4.06050652e-01 -2.84204543e-01 2.85426438e-01 3.54788393e-01 2.66440660e-01 -2.75305539e-01 -9.00287151e-01 4.78922874e-01 -1.59495652e+00 -8.35883796e-01 1.65187061e-01 2.70565748e+00 6.62064016e-01 5.55299699e-01 2.64064193e-01 2.97937334e-01 6.76037669e-01 -2.71921664e-01 -8.19585145e-01 -4.80563670e-01 -1.79216251e-01 -9.54253674e-02 3.79829288e-01 2.95267522e-01 -7.81326294e-01 4.81112808e-01 7.43823481e+00 8.58166039e-01 -1.03830314e+00 -1.86837092e-01 9.72013772e-01 -2.93311328e-01 -4.43836212e-01 -1.65271029e-01 -5.69555044e-01 5.51492810e-01 1.06170428e+00 -1.01604283e+00 6.64396733e-02 9.50960815e-01 3.85992378e-01 1.00633159e-01 -1.55797088e+00 6.95292711e-01 -2.64267236e-01 -1.28069353e+00 3.12086731e-01 1.57891065e-01 6.84653938e-01 -2.85409749e-01 4.92326289e-01 3.17850888e-01 3.17901820e-01 -1.12765670e+00 8.09996724e-01 4.84555632e-01 9.75609899e-01 -9.33527887e-01 5.62394023e-01 5.94623685e-01 -6.94137275e-01 -3.01000953e-01 -2.67513305e-01 7.52396584e-02 -3.11023355e-01 5.40253341e-01 -1.09606040e+00 2.06463426e-01 3.03686827e-01 4.44927007e-01 -7.74020314e-01 8.95315230e-01 1.86010346e-01 6.38392150e-01 -1.40970811e-01 -1.81482494e-01 1.85426533e-01 -2.21998245e-01 3.94865304e-01 9.24586892e-01 5.51334620e-01 -3.78021181e-01 -4.03324485e-01 1.22717619e+00 -1.32016376e-01 6.73764944e-02 -9.46039140e-01 -3.11340570e-01 6.72462821e-01 6.50439918e-01 -5.52578092e-01 -8.90233219e-02 -1.90833554e-01 5.05929768e-01 3.55733812e-01 5.04896164e-01 -8.05902779e-01 -2.84842223e-01 8.12585771e-01 1.31502315e-01 -2.98746943e-01 -1.74439818e-01 -7.02832282e-01 -8.87131453e-01 2.10125461e-01 -6.07284248e-01 2.73872823e-01 -4.60294336e-01 -1.17796850e+00 5.33528447e-01 2.75429070e-01 -1.40189457e+00 -4.41586733e-01 -6.55315101e-01 -5.97144186e-01 7.56375194e-01 -7.96722949e-01 -4.84610677e-01 -1.63850188e-01 1.34296253e-01 4.56020802e-01 -3.99031490e-01 7.84172058e-01 -1.48696387e-02 -7.58503199e-01 1.11000800e+00 6.22947454e-01 2.02902388e-02 4.92672026e-01 -1.11178768e+00 5.58725238e-01 8.61736953e-01 -8.85531530e-02 8.19571555e-01 9.00378168e-01 -3.53030264e-01 -1.06522846e+00 -1.18751836e+00 4.33797717e-01 -6.00470483e-01 4.34601158e-01 -1.65456161e-01 -6.41044915e-01 6.05560720e-01 -5.01494825e-01 -2.86861539e-01 7.01549590e-01 -7.34929889e-02 -3.20285022e-01 -1.39118269e-01 -1.14085901e+00 9.60574448e-01 1.06697381e+00 -1.56545669e-01 -3.21309209e-01 8.50678515e-03 9.87813890e-01 1.66014776e-01 -9.38173056e-01 7.42743790e-01 6.92052603e-01 -1.14279282e+00 8.64588559e-01 -9.52594221e-01 7.23377943e-01 2.12254807e-01 -5.48603237e-01 -1.67856705e+00 -2.23843381e-02 -2.75557280e-01 7.33742639e-02 9.08559561e-01 7.82145321e-01 -8.16082835e-01 8.00803244e-01 7.69774377e-01 -8.05886183e-03 -1.15796661e+00 -8.60745192e-01 -1.20721567e+00 5.57962418e-01 -3.47633362e-01 6.50888383e-01 7.29537427e-01 -1.07553497e-01 3.02203536e-01 6.72889128e-02 -2.31785178e-01 5.40186703e-01 1.09376021e-01 6.52872860e-01 -1.36620355e+00 -1.33152828e-01 -7.98936188e-01 -3.62542599e-01 -7.51972079e-01 4.71538991e-01 -8.91305208e-01 -3.07738125e-01 -1.23332751e+00 1.76156685e-01 -4.58742976e-01 -3.41493458e-01 -1.63218230e-01 -9.18233022e-02 -1.00557148e-01 1.52024686e-01 8.49473923e-02 -1.68346897e-01 8.99096608e-01 1.16485083e+00 1.05818875e-01 -2.22038671e-01 4.37889874e-01 -1.00573909e+00 6.45442545e-01 1.03984809e+00 -4.08996612e-01 -8.74918878e-01 -9.98836234e-02 1.03564940e-01 -1.70197859e-02 2.54619569e-01 -8.20201159e-01 -1.28909975e-01 -2.84202129e-01 3.19609612e-01 -1.79317862e-01 2.69228429e-01 -4.99130517e-01 -8.69132653e-02 4.43636358e-01 -7.50209033e-01 3.21556121e-01 1.02399506e-01 5.07170260e-01 -2.21850593e-02 -4.55136836e-01 1.01167536e+00 2.84329373e-02 -4.37237412e-01 2.99188435e-01 -2.39041299e-01 4.84143227e-01 1.15554059e+00 -3.48991603e-01 -2.53887802e-01 -4.56449300e-01 -5.54485023e-01 1.89381808e-01 6.66658342e-01 3.03122729e-01 5.12485981e-01 -1.33688176e+00 -7.59209335e-01 2.69070596e-01 3.52757990e-01 -2.66051739e-01 -1.45177022e-01 5.47335923e-01 -4.15024787e-01 4.51741040e-01 -3.26606691e-01 -4.28793877e-01 -9.59448695e-01 6.43504798e-01 7.36549497e-01 -6.86950162e-02 -1.17386810e-01 4.83808309e-01 5.79604805e-01 -3.91514748e-01 2.31744036e-01 -3.87230515e-01 2.30360731e-01 -6.11724518e-02 2.03294083e-01 6.40431166e-01 -2.96988934e-02 -3.16620082e-01 -5.66116720e-02 2.44791776e-01 1.85081691e-01 -3.96455795e-01 9.43449855e-01 1.46845266e-01 3.42121154e-01 6.18865132e-01 1.39130390e+00 -5.43131948e-01 -1.25117099e+00 -2.37531602e-01 1.39405370e-01 -3.94154876e-01 -3.12627465e-01 -5.54559946e-01 -6.96776807e-01 8.63648593e-01 4.27205056e-01 3.27492505e-01 1.11234713e+00 -2.84119517e-01 2.66993381e-02 5.43120086e-01 2.54899710e-01 -1.14564848e+00 1.93762034e-01 3.47411275e-01 1.15340352e+00 -1.12574255e+00 -1.82922497e-01 -2.18592331e-01 -6.75863624e-01 7.67000437e-01 7.40058422e-01 -8.48396420e-02 7.35952616e-01 6.65633827e-02 -2.14573756e-01 1.42479718e-01 -8.31291318e-01 1.29294008e-01 1.20867021e-01 5.77076852e-01 4.25931364e-01 1.44101053e-01 -4.28591758e-01 7.83976734e-01 -6.60802901e-01 -1.65341407e-01 5.92725098e-01 4.12011504e-01 -3.61059725e-01 -8.24566722e-01 -2.01546419e-02 8.06159496e-01 -2.76952028e-01 -5.86785637e-02 -5.36597490e-01 1.00477529e+00 -3.16794664e-01 8.43637466e-01 8.09480995e-02 -4.22823101e-01 4.13767844e-01 2.71473885e-01 3.82067800e-01 -6.40935063e-01 -1.38194665e-01 -5.03725469e-01 1.81820728e-02 -2.96106458e-01 -2.88925208e-02 -5.66397548e-01 -9.62907314e-01 -3.28280777e-01 -1.19860642e-01 3.03429395e-01 4.75323349e-01 8.22972476e-01 2.36291885e-01 3.99531364e-01 7.63121367e-01 -4.39532846e-01 -1.24186993e+00 -9.76982176e-01 -7.44279087e-01 4.16246682e-01 2.33800784e-01 -8.33079994e-01 -4.49432939e-01 -1.24886245e-01]

[8.593117713928223, 4.244696140289307]

fffbfedb-9074-4837-8324-2f966c3ac459

flexible-and-structured-knowledge-grounded

2209.08284

https://arxiv.org/abs/2209.08284v3

https://arxiv.org/pdf/2209.08284v3.pdf

Structured Knowledge Grounding for Question Answering

Can language models (LM) ground question-answering (QA) tasks in the knowledge base via inherent relational reasoning ability? While previous models that use only LMs have seen some success on many QA tasks, more recent methods include knowledge graphs (KG) to complement LMs with their more logic-driven implicit knowledge. However, effectively extracting information from structured data, like KGs, empowers LMs to remain an open question, and current models rely on graph techniques to extract knowledge. In this paper, we propose to solely leverage the LMs to combine the language and knowledge for knowledge based question-answering with flexibility, breadth of coverage and structured reasoning. Specifically, we devise a knowledge construction method that retrieves the relevant context with a dynamic hop, which expresses more comprehensivenes than traditional GNN-based techniques. And we devise a deep fusion mechanism to further bridge the information exchanging bottleneck between the language and the knowledge. Extensive experiments show that our model consistently demonstrates its state-of-the-art performance over CommensenseQA benchmark, showcasing the possibility to leverage LMs solely to robustly ground QA into the knowledge base.

['Kairui Zhou', 'Siqi Ouyang', 'Yujie Lu']

2022-09-17

['relational-reasoning']

['natural-language-processing']

[-1.27450541e-01 8.63772631e-01 -3.53941709e-01 -2.46425405e-01 -1.05402088e+00 -8.52279544e-01 6.55278444e-01 2.91309118e-01 -1.53152063e-01 7.10655034e-01 2.78089702e-01 -8.55279505e-01 -5.03240407e-01 -1.44452286e+00 -9.50299740e-01 1.14097677e-01 1.45450160e-01 8.14445376e-01 8.67142379e-01 -8.06791067e-01 -3.04133520e-02 1.16613522e-01 -1.38282287e+00 6.34192824e-01 1.12033653e+00 8.62368822e-01 -2.78439492e-01 3.73656243e-01 -7.44269729e-01 1.93020129e+00 -3.22649956e-01 -8.76117110e-01 8.42796862e-02 -2.67028153e-01 -1.59631658e+00 -5.41240871e-01 6.07936025e-01 -7.21714720e-02 -3.95019472e-01 7.08060741e-01 1.27958581e-01 -8.44152048e-02 1.31355077e-01 -1.14264917e+00 -5.90792358e-01 1.14671826e+00 -2.29577553e-02 6.71129627e-03 6.90474391e-01 1.94191784e-02 1.45282304e+00 -5.00931859e-01 1.01361287e+00 1.24011815e+00 8.72105539e-01 4.72622991e-01 -8.99443686e-01 -1.26611143e-01 1.48103237e-01 7.59109914e-01 -1.33192110e+00 -4.98485535e-01 6.76934481e-01 -8.02170336e-02 1.35634136e+00 4.22362119e-01 6.72513306e-01 5.20609975e-01 -7.81446099e-02 8.06477010e-01 1.12096763e+00 -6.89225435e-01 6.22232109e-02 1.61340952e-01 4.38075453e-01 1.19040537e+00 5.23562729e-01 -4.78532702e-01 -8.89108300e-01 -2.44689673e-01 3.79907817e-01 -5.14748633e-01 -3.48342746e-01 -6.13302827e-01 -9.25353467e-01 7.10243285e-01 4.63056803e-01 2.28951931e-01 -2.82170951e-01 3.08721364e-01 2.49428973e-01 6.24480844e-01 -1.43119125e-02 8.26080441e-01 -7.99334764e-01 -1.15570910e-01 -7.42690086e-01 3.64288509e-01 1.45722592e+00 9.44447875e-01 9.69781399e-01 -5.29219508e-01 -1.54812634e-01 2.95282513e-01 3.73214394e-01 4.77347910e-01 -9.49757025e-02 -1.24893689e+00 6.45002246e-01 1.51220632e+00 -1.71895139e-02 -8.50881279e-01 -3.03275526e-01 -3.87549132e-01 -6.55641183e-02 -2.80309141e-01 4.92153823e-01 7.53813535e-02 -6.65443242e-01 1.58940291e+00 6.14987493e-01 1.27600497e-02 4.35390621e-01 5.85879982e-01 9.84756291e-01 1.33372977e-01 -9.01266525e-04 2.18797281e-01 1.48191512e+00 -8.51734638e-01 -6.16519570e-01 -3.10760438e-01 1.05980730e+00 -8.78033787e-02 1.13985407e+00 2.53164858e-01 -1.24583900e+00 -7.43114622e-03 -9.99213696e-01 -5.35641789e-01 -5.96440554e-01 -5.36453843e-01 9.86229658e-01 6.93598032e-01 -1.30244029e+00 2.98290312e-01 -7.74937868e-01 -3.40121627e-01 4.25633937e-01 3.08537483e-01 -3.79690856e-01 -4.94245827e-01 -1.72593272e+00 1.23749757e+00 6.89873040e-01 6.23999834e-02 -5.14632106e-01 -8.37662578e-01 -1.05171871e+00 4.54049511e-03 1.41588652e+00 -1.47035766e+00 1.18477082e+00 -4.48177904e-01 -1.42099643e+00 6.03739560e-01 -1.93560332e-01 -7.22449362e-01 2.71692514e-01 -2.54838407e-01 -4.39129174e-01 4.52624410e-01 -9.81037766e-02 3.87645423e-01 3.07483017e-01 -1.25225294e+00 -5.50701201e-01 -3.48350376e-01 1.12764251e+00 1.18187964e-01 -1.52163818e-01 -2.07939833e-01 -6.84586108e-01 1.99522749e-01 3.36317152e-01 -5.11793733e-01 -5.55277728e-02 -3.59222323e-01 -3.56423229e-01 -4.91907358e-01 4.97248083e-01 -6.76129878e-01 1.52120543e+00 -1.44526410e+00 2.11840272e-01 4.70817715e-01 6.20875657e-01 2.95252830e-01 1.36143893e-01 7.88851738e-01 5.73639870e-01 1.04814745e-01 -1.15439646e-01 1.07586585e-01 3.72406304e-01 7.30175078e-01 -5.32357097e-01 -2.90991575e-01 4.43577379e-01 1.76356137e+00 -1.04716563e+00 -8.91934395e-01 -2.24561051e-01 1.89811271e-02 -7.84727395e-01 -8.78231898e-02 -9.27188218e-01 -7.80654550e-02 -6.19845390e-01 8.70518982e-01 2.66294837e-01 -6.77573860e-01 7.13287055e-01 -3.08471322e-01 4.06480908e-01 7.73963034e-01 -1.11914027e+00 1.81782055e+00 -5.06084800e-01 1.27756014e-01 1.34396166e-01 -5.83540976e-01 5.19030094e-01 7.50606135e-02 1.11776427e-03 -7.67859697e-01 -3.36516500e-01 3.34478617e-01 -4.28817682e-02 -6.74953341e-01 5.54850757e-01 -1.60404414e-01 2.09610015e-02 2.58615434e-01 1.35103807e-01 -4.18160170e-01 2.74903744e-01 7.73587704e-01 1.60193908e+00 4.19251710e-01 1.60976678e-01 -9.31670964e-02 6.89499319e-01 6.04652584e-01 2.88480073e-01 9.93470848e-01 3.31249565e-01 -2.39766717e-01 7.62399852e-01 -1.93260550e-01 -4.12671834e-01 -9.21238899e-01 2.32249096e-01 1.01717854e+00 1.23779312e-01 -9.90380943e-01 -7.01560497e-01 -1.17178428e+00 6.39984980e-02 9.53929722e-01 -2.83049256e-01 -2.00487629e-01 -8.36185157e-01 -1.62589371e-01 1.12422621e+00 5.42386532e-01 5.58854699e-01 -8.81358504e-01 -6.32394731e-01 2.30873242e-01 -4.80713964e-01 -1.21900821e+00 4.69319224e-01 6.30076826e-02 -8.10597897e-01 -1.47834504e+00 2.09793806e-01 -3.62710863e-01 2.95689881e-01 -1.38935372e-01 1.75631297e+00 3.50563139e-01 2.16882452e-01 9.21171606e-01 -5.24652660e-01 -3.90900940e-01 -4.75750685e-01 3.83838385e-01 -3.87603819e-01 -5.03390372e-01 5.46581149e-01 -5.02855837e-01 -4.40283567e-01 1.22282349e-01 -1.02754450e+00 2.84951180e-02 5.88855028e-01 4.92365211e-01 5.18150032e-01 1.19122542e-01 8.20737064e-01 -1.22118521e+00 5.60491562e-01 -5.74279666e-01 -4.79923576e-01 8.88852298e-01 -8.39165449e-01 4.77974415e-01 4.32947397e-01 2.06508726e-01 -1.10037553e+00 -4.45256978e-01 -6.88033625e-02 -3.49870361e-02 2.68371254e-01 1.21858537e+00 -2.95630217e-01 -2.43136451e-01 7.55331874e-01 4.79773208e-02 -1.15811437e-01 -2.90194482e-01 1.13744342e+00 2.27270722e-01 5.73782444e-01 -1.20349252e+00 9.40192282e-01 5.32413363e-01 2.59765089e-01 -2.10023999e-01 -1.23990440e+00 -4.45856452e-01 -4.41280812e-01 1.72864512e-01 5.24922490e-01 -8.02283645e-01 -8.94232810e-01 -1.90694630e-01 -9.18211162e-01 -4.60212171e-01 -6.79617822e-01 -2.60331780e-02 -4.62146670e-01 5.79296947e-01 -5.92647970e-01 -7.61084378e-01 -4.87445474e-01 -7.12880611e-01 9.22478735e-01 -5.69700226e-02 -1.85155451e-01 -1.15215027e+00 1.39087766e-01 9.30005550e-01 4.38525826e-01 2.06708740e-02 1.39980984e+00 -8.65463793e-01 -1.21101284e+00 -1.00617811e-01 -3.41756582e-01 1.74272045e-01 -1.08931787e-01 -5.97392797e-01 -8.98596287e-01 1.59539849e-01 -1.41044199e-01 -6.47983968e-01 6.49615288e-01 -3.94809455e-01 8.09775174e-01 -4.55798745e-01 -2.10624859e-01 3.14187706e-01 1.39440656e+00 -3.55019778e-01 8.51633370e-01 6.34266138e-01 6.72094941e-01 6.98524952e-01 3.26609790e-01 -1.77053064e-01 1.22584224e+00 4.52253282e-01 3.76731008e-01 2.75978059e-01 -3.72186899e-01 -5.85388303e-01 1.66695997e-01 9.48909819e-01 -2.37048879e-01 5.63454591e-02 -1.37488377e+00 5.77795923e-01 -1.95505512e+00 -8.46164107e-01 -2.51097292e-01 1.75778472e+00 1.31573296e+00 2.66870439e-01 -2.98544407e-01 5.28163686e-02 -1.84053004e-01 -2.09383871e-02 -4.82424170e-01 -1.14901714e-01 -3.26598763e-01 6.39283717e-01 4.54895228e-01 6.79627419e-01 -5.15995383e-01 1.20473814e+00 6.17518568e+00 8.23527098e-01 -5.47614515e-01 3.56447399e-02 -1.55942261e-01 9.68544185e-02 -9.30117905e-01 5.57896197e-01 -7.60302007e-01 -2.54241735e-01 1.03448546e+00 -1.83810309e-01 5.75067937e-01 6.99710548e-01 -5.38375497e-01 -1.61088943e-01 -1.24125230e+00 4.56235647e-01 -6.55798391e-02 -1.80446529e+00 4.64700669e-01 -1.22379504e-01 4.94067967e-01 -1.90012273e-03 -3.54127049e-01 8.14302087e-01 5.90777218e-01 -1.05236924e+00 6.17556393e-01 9.82262254e-01 5.99061131e-01 -4.90801722e-01 8.64978373e-01 4.76916730e-01 -1.17804468e+00 -1.98231131e-01 -7.66931623e-02 -9.22459960e-02 1.26291931e-01 4.73095357e-01 -1.07519722e+00 1.57159841e+00 3.80965918e-01 2.33751535e-01 -9.55717027e-01 5.41312158e-01 -5.74778855e-01 6.95656180e-01 -3.80986273e-01 1.47145092e-01 3.05640429e-01 6.90500811e-02 3.14701259e-01 1.02159083e+00 -1.13802716e-01 2.17818230e-01 1.22835040e-01 9.66491163e-01 -2.69455969e-01 8.70629866e-03 -4.57861841e-01 -1.94058344e-01 4.39937562e-01 1.00679004e+00 -2.04849347e-01 -6.12446964e-01 -6.65162325e-01 4.78625178e-01 7.01496780e-01 2.72838682e-01 -4.84172374e-01 -3.52817774e-01 1.50236860e-01 2.42415026e-01 3.03448051e-01 -3.00581425e-01 -9.07503664e-02 -1.21106505e+00 4.09228176e-01 -1.12973607e+00 8.74229610e-01 -8.57452095e-01 -1.23683643e+00 3.72506797e-01 2.14416072e-01 -3.72178942e-01 -4.78131920e-01 -5.62718093e-01 -1.73292756e-01 7.77125359e-01 -2.04367948e+00 -1.48460054e+00 -2.00346455e-01 8.89713645e-01 -1.48497418e-01 2.17794523e-01 8.17047775e-01 1.07237130e-01 -2.11767182e-02 5.15514851e-01 -5.27619123e-01 1.26189277e-01 4.79915619e-01 -1.46258676e+00 2.25913852e-01 6.64559960e-01 4.37884986e-01 1.23203194e+00 3.61877233e-01 -7.59674072e-01 -2.29319906e+00 -7.61252165e-01 1.03031313e+00 -1.27640092e+00 1.14297867e+00 -1.47916660e-01 -1.20933843e+00 9.80596364e-01 5.94507493e-02 -1.04892924e-01 5.21981001e-01 4.96406287e-01 -8.67334068e-01 -1.00343756e-01 -8.39668930e-01 5.72381735e-01 1.39449644e+00 -9.38779175e-01 -1.25658488e+00 1.23838134e-01 1.22353017e+00 -4.32800144e-01 -1.12894714e+00 5.85402727e-01 3.85702342e-01 -8.36505532e-01 9.58636284e-01 -9.28225815e-01 3.24679077e-01 -6.71832860e-01 -3.04303706e-01 -7.48210967e-01 6.39597178e-02 -6.42661452e-01 -9.68578100e-01 1.09291506e+00 6.62810326e-01 -7.39453077e-01 8.55797350e-01 8.28751802e-01 -1.40929803e-01 -9.59519506e-01 -8.96086514e-01 -7.60534048e-01 -4.57792170e-02 -8.28423858e-01 8.95283520e-01 9.85665917e-01 4.56951708e-01 5.71179688e-01 2.70549923e-01 3.09894472e-01 3.29712123e-01 4.12910253e-01 7.54235744e-01 -1.27102387e+00 -4.51427221e-01 -1.60129070e-01 -2.97065407e-01 -1.01842678e+00 2.74033457e-01 -1.33728349e+00 -3.90714228e-01 -2.39226103e+00 6.48160428e-02 -6.22705221e-01 -3.02671373e-01 9.42841470e-01 -1.57332748e-01 -2.49720454e-01 4.08084132e-02 1.74690951e-02 -1.24643624e+00 2.56230980e-01 1.17056191e+00 -7.84606412e-02 -4.75140177e-02 -5.20798028e-01 -1.05772495e+00 6.10167921e-01 3.80569518e-01 -1.15447849e-01 -8.86395633e-01 -5.40466249e-01 1.19123638e+00 9.70195681e-02 5.36805809e-01 -8.06091428e-01 7.99310446e-01 -1.22165255e-01 -3.18809450e-01 -1.96810499e-01 2.47553602e-01 -8.72941554e-01 5.25502898e-02 1.46759450e-01 -1.37880757e-01 -1.80227056e-01 2.98236549e-01 6.17685616e-01 -3.63196045e-01 -2.07221687e-01 -1.02932937e-01 -3.93897235e-01 -1.01828969e+00 2.64649726e-02 1.99653596e-01 5.42410970e-01 4.31516260e-01 -8.39694403e-03 -8.94183934e-01 -3.00601959e-01 -4.78822380e-01 5.15577793e-01 3.81646276e-01 1.36377504e-02 4.34827238e-01 -9.31721449e-01 -5.09307444e-01 -1.88195080e-01 3.51044118e-01 2.97045946e-01 1.03583507e-01 9.86008286e-01 -4.60515171e-01 6.59989417e-01 2.37908453e-01 -2.23416522e-01 -7.60142624e-01 6.00720286e-01 4.63309556e-01 -7.71426260e-01 -5.45005202e-01 6.85865700e-01 -3.69526953e-01 -6.71231985e-01 -8.55514929e-02 -4.44563478e-01 -1.06368244e-01 8.78210925e-03 3.15051705e-01 2.66136855e-01 4.59074616e-01 6.10619318e-03 -5.16824722e-01 2.63364613e-01 1.00527614e-01 -3.26675065e-02 1.08613873e+00 -1.01036951e-01 -7.12440848e-01 3.10898751e-01 3.96943599e-01 4.54905987e-01 -5.75188577e-01 -6.65620625e-01 6.19808018e-01 -9.95868444e-02 -1.74135223e-01 -1.37472916e+00 -5.22815406e-01 5.40479124e-01 -2.21310839e-01 4.10363197e-01 9.15559530e-01 4.17633325e-01 8.78258467e-01 1.19417119e+00 9.06683922e-01 -7.83735275e-01 -1.27315179e-01 8.35730791e-01 6.57650292e-01 -9.66384888e-01 1.69224869e-02 -6.18654013e-01 -4.68755752e-01 9.37791824e-01 6.16251647e-01 4.16759670e-01 2.12477118e-01 2.64937758e-01 9.98248383e-02 -7.68272221e-01 -9.97648001e-01 -6.12123668e-01 4.07446027e-01 4.67224628e-01 1.40032843e-01 -1.31590053e-01 7.76887313e-02 8.44249427e-01 -4.00812268e-01 3.60969812e-01 1.75391808e-01 1.32804775e+00 -5.31295776e-01 -1.40633130e+00 -1.06303073e-01 4.28751945e-01 -3.45044732e-01 -4.28122103e-01 -6.30596638e-01 1.05550671e+00 1.62555113e-01 1.06494379e+00 -5.74802458e-01 -2.82782763e-01 6.23168945e-01 6.93573654e-01 8.99400890e-01 -7.37147212e-01 -7.09817886e-01 -8.21826994e-01 7.89014637e-01 -7.73320138e-01 -4.06319410e-01 -3.93172204e-02 -1.47882426e+00 -3.49883944e-01 -3.31746131e-01 3.54384840e-01 3.40532005e-01 1.20282590e+00 5.89348316e-01 4.75795805e-01 -2.50167042e-01 3.22146177e-01 -5.58593690e-01 -5.03241301e-01 -1.59186631e-01 1.68018550e-01 -6.01608306e-02 -4.55298096e-01 -3.02647594e-02 -1.00184187e-01]

[10.200711250305176, 7.835104942321777]

c067f6bb-8b54-42a3-b728-e16674b8bef9

spatialsense-an-adversarially-crowdsourced

1908.0266

https://arxiv.org/abs/1908.02660v2

https://arxiv.org/pdf/1908.02660v2.pdf

SpatialSense: An Adversarially Crowdsourced Benchmark for Spatial Relation Recognition

Understanding the spatial relations between objects in images is a surprisingly challenging task. A chair may be "behind" a person even if it appears to the left of the person in the image (depending on which way the person is facing). Two students that appear close to each other in the image may not in fact be "next to" each other if there is a third student between them. We introduce SpatialSense, a dataset specializing in spatial relation recognition which captures a broad spectrum of such challenges, allowing for proper benchmarking of computer vision techniques. SpatialSense is constructed through adversarial crowdsourcing, in which human annotators are tasked with finding spatial relations that are difficult to predict using simple cues such as 2D spatial configuration or language priors. Adversarial crowdsourcing significantly reduces dataset bias and samples more interesting relations in the long tail compared to existing datasets. On SpatialSense, state-of-the-art recognition models perform comparably to simple baselines, suggesting that they rely on straightforward cues instead of fully reasoning about this complex task. The SpatialSense benchmark provides a path forward to advancing the spatial reasoning capabilities of computer vision systems. The dataset and code are available at https://github.com/princeton-vl/SpatialSense.

['Olga Russakovsky', 'Kaiyu Yang', 'Jia Deng']

2019-08-07

spatialsense-an-adversarially-crowdsourced-1

http://openaccess.thecvf.com/content_ICCV_2019/html/Yang_SpatialSense_An_Adversarially_Crowdsourced_Benchmark_for_Spatial_Relation_Recognition_ICCV_2019_paper.html

http://openaccess.thecvf.com/content_ICCV_2019/papers/Yang_SpatialSense_An_Adversarially_Crowdsourced_Benchmark_for_Spatial_Relation_Recognition_ICCV_2019_paper.pdf

iccv-2019-10

['spatial-relation-recognition']

['computer-vision']

[-1.19059309e-01 -1.13255540e-02 6.98656291e-02 -4.20384020e-01 -6.56130910e-01 -1.03120303e+00 9.30790603e-01 1.52249545e-01 -5.83550453e-01 4.10060614e-01 1.67699650e-01 -4.91291076e-01 -1.16945006e-01 -6.09470129e-01 -1.08515561e+00 -7.18232632e-01 1.93160757e-01 7.21484721e-01 4.60713059e-01 -3.66405666e-01 2.60878325e-01 6.23775244e-01 -1.43519771e+00 3.42674702e-01 4.56577092e-01 6.74853146e-01 1.24027964e-03 8.14288676e-01 2.84051389e-01 9.83886123e-01 -6.10554516e-01 -6.27080142e-01 3.76488119e-01 -2.48370990e-02 -1.08433402e+00 -1.41116068e-01 1.28827477e+00 -1.40605316e-01 -6.69250786e-01 9.20672476e-01 2.39545718e-01 5.09041190e-01 7.00716615e-01 -1.44788504e+00 -1.06152916e+00 1.64706334e-01 -6.32208586e-01 5.15398085e-01 7.96807766e-01 2.72805125e-01 1.07466590e+00 -8.91955197e-01 6.47306144e-01 1.15245318e+00 6.32321358e-01 7.68096745e-02 -1.22226250e+00 -4.49886113e-01 4.28749353e-01 5.54807544e-01 -1.58562291e+00 -3.82694185e-01 5.90071499e-01 -7.85991311e-01 9.32048321e-01 5.56323349e-01 4.38807636e-01 1.14518082e+00 6.01151697e-02 6.38012350e-01 1.23136842e+00 -5.03886223e-01 1.40597299e-01 1.06821824e-02 1.39106989e-01 6.09384120e-01 -1.16496138e-01 1.11292105e-03 -7.12425828e-01 1.15603600e-02 7.80713856e-01 1.10687036e-02 -2.43487164e-01 -7.55864501e-01 -1.37044406e+00 5.27165055e-01 9.46058929e-01 1.26537085e-01 4.84217517e-03 1.19465806e-01 -1.42650455e-01 1.02610841e-01 1.86328456e-01 7.04671681e-01 -6.31501824e-02 -4.90013510e-02 -6.96917236e-01 5.22812665e-01 6.42865360e-01 1.21533692e+00 6.64954782e-01 -5.21801472e-01 -2.18780756e-01 5.90108335e-01 1.17182381e-01 4.18225944e-01 1.61531106e-01 -1.27841663e+00 7.26356149e-01 7.05511451e-01 3.70988041e-01 -1.31789052e+00 -3.84755969e-01 -1.36016548e-01 -5.15867651e-01 4.56371725e-01 1.19180644e+00 3.51065487e-01 -6.93160534e-01 1.45800138e+00 4.87363935e-01 3.73188853e-01 -2.16964200e-01 1.10663033e+00 1.06446731e+00 3.71084958e-01 3.04867066e-02 7.12099671e-01 1.61274755e+00 -1.20150042e+00 -1.06590316e-01 -4.59747970e-01 5.30269742e-01 -8.87744129e-01 1.18728507e+00 2.60337919e-01 -9.95177627e-01 -5.38391948e-01 -6.98814571e-01 -5.14359355e-01 -9.59293842e-01 -1.59668624e-01 5.68391562e-01 3.64936024e-01 -1.05293381e+00 4.91390049e-01 -5.35843790e-01 -3.31915170e-01 5.36354482e-01 8.17116722e-02 -8.41056883e-01 -4.05399919e-01 -9.26146090e-01 1.17733395e+00 -3.02403063e-01 7.01807961e-02 -5.52524865e-01 -7.68312573e-01 -1.13172460e+00 -2.90733695e-01 4.54822153e-01 -5.70040703e-01 1.40311515e+00 -7.22033679e-01 -6.94397688e-01 1.27182961e+00 -2.80819952e-01 -1.73373207e-01 8.04169774e-01 -2.04901278e-01 -2.62066066e-01 3.89623493e-02 6.25175357e-01 7.92268872e-01 4.42450553e-01 -1.25987792e+00 -4.49721754e-01 -6.55218720e-01 4.68502015e-01 3.04071814e-01 3.64531308e-01 7.24042729e-02 -7.19598591e-01 -6.34074390e-01 1.89835459e-01 -1.38597298e+00 -3.33728306e-02 4.89591777e-01 -7.08125114e-01 -3.65445077e-01 7.37159908e-01 -3.65631968e-01 5.49360931e-01 -2.02196264e+00 -9.63183120e-02 5.06653637e-02 1.09509200e-01 2.23220661e-01 -8.59896168e-02 3.59736413e-01 -3.55293781e-01 3.85542326e-02 7.99193606e-02 -2.56286979e-01 2.00124189e-01 4.05006558e-01 -4.80487883e-01 8.27210069e-01 -1.28887116e-03 9.34944034e-01 -1.10183787e+00 -3.71224105e-01 2.89597988e-01 4.12583292e-01 -1.28243655e-01 8.10277835e-02 5.80672137e-02 7.50824094e-01 -1.25025600e-01 5.59872031e-01 8.47642958e-01 -3.83277625e-01 -2.37352759e-01 1.65057048e-01 -4.24765870e-02 3.27378362e-01 -1.29890287e+00 1.57490110e+00 -3.06456685e-01 9.38540578e-01 -1.03476644e-01 -7.37066269e-01 4.98731762e-01 1.92606682e-03 4.74551320e-02 -9.13192928e-01 -2.39306703e-01 -2.25038201e-01 -1.27757475e-01 -6.60559058e-01 4.42554057e-01 2.44863987e-01 -7.45425373e-02 2.30219960e-01 -3.30386043e-01 -2.68492758e-01 1.38765320e-01 2.92350858e-01 1.18098950e+00 2.89114296e-01 1.83690146e-01 -2.02424914e-01 3.17022830e-01 1.67558104e-01 3.18472862e-01 9.33715999e-01 -4.94045168e-01 8.56252193e-01 5.48086405e-01 -7.72680044e-01 -9.48499858e-01 -1.39671874e+00 -1.84985965e-01 1.51261389e+00 5.79955459e-01 -8.79251212e-02 -4.10879284e-01 -7.85608292e-01 2.77367383e-01 5.60945928e-01 -1.00057924e+00 3.35344672e-01 -3.94828737e-01 5.50244674e-02 6.80633903e-01 8.86661768e-01 4.38526958e-01 -8.40320349e-01 -3.29864860e-01 -4.23382342e-01 -1.68775126e-01 -1.43076432e+00 -4.85301018e-01 1.57714877e-02 -1.17510103e-01 -1.45806932e+00 -7.51322567e-01 -7.83443272e-01 8.46252322e-01 6.47458315e-01 1.55949306e+00 2.14924946e-01 -2.65163839e-01 5.58517694e-01 -7.77840093e-02 -4.31427240e-01 1.14625365e-01 -8.01328495e-02 -8.57279897e-02 -3.01546931e-01 5.38982332e-01 -4.77058560e-01 -7.54732490e-01 7.03352571e-01 -4.89900887e-01 -1.63573891e-01 2.76629120e-01 5.38287342e-01 4.67346966e-01 -6.22889139e-02 -1.04215503e-01 -8.37389708e-01 1.93896264e-01 -6.77881122e-01 -5.20458281e-01 2.81868547e-01 -8.28397796e-02 -1.72346339e-01 5.56547582e-01 -3.94135296e-01 -7.00061202e-01 2.17874154e-01 2.68486619e-01 -3.68737668e-01 -7.24939108e-01 -2.88204968e-01 -2.35409494e-02 -1.62436366e-01 9.44453835e-01 8.68085027e-02 -2.38366961e-01 -1.54639512e-01 4.25065517e-01 1.33482471e-01 8.56165409e-01 -7.29394317e-01 9.73508239e-01 6.93766236e-01 1.96525335e-01 -6.81684792e-01 -9.81830895e-01 -6.93182945e-01 -1.14966321e+00 -1.69875339e-01 1.04857588e+00 -1.13515854e+00 -9.78464007e-01 2.29289278e-01 -1.20670199e+00 -5.62711120e-01 -7.94196352e-02 2.41099820e-01 -3.18737000e-01 7.72524402e-02 -4.44818348e-01 -4.02672678e-01 5.62129319e-01 -1.21813941e+00 1.31473672e+00 2.47707754e-01 -5.40090859e-01 -1.12991405e+00 -6.90272078e-02 9.53214288e-01 3.50683331e-02 2.83351958e-01 5.68376124e-01 -5.87911725e-01 -8.66032481e-01 -2.29817018e-01 -2.23180279e-01 -1.29977018e-01 -1.76185042e-01 3.57859731e-02 -1.02158749e+00 1.85266629e-01 -6.54785931e-01 -3.39440078e-01 7.39705145e-01 4.62496392e-02 1.13060629e+00 -2.35216245e-01 -5.17511249e-01 4.70329791e-01 1.07412016e+00 -6.24252521e-02 6.61370635e-01 6.06380284e-01 9.89528775e-01 8.46439838e-01 4.52780247e-01 -4.98396950e-03 8.76372159e-01 9.94192183e-01 5.00568449e-01 -1.86921597e-01 -9.74728614e-02 -2.46863395e-01 -1.42833993e-01 -3.20360959e-02 -8.57012570e-02 -2.24240944e-01 -1.52839327e+00 7.93583453e-01 -1.83517134e+00 -1.11940563e+00 -6.60966396e-01 1.93672693e+00 7.33117223e-01 -1.04948826e-01 7.88837746e-02 -6.36656508e-02 5.45534730e-01 2.29947895e-01 -4.03115243e-01 -2.87325919e-01 -8.48521739e-02 -1.52486965e-01 5.05602062e-01 7.82169819e-01 -1.39055252e+00 8.94536376e-01 5.88413000e+00 5.55931926e-01 -8.30140591e-01 -1.95009649e-01 7.05320299e-01 -2.72110403e-01 -2.24185377e-01 2.80406773e-02 -7.84622788e-01 4.29931760e-01 2.26487607e-01 4.70290661e-01 4.56231862e-01 8.16836655e-01 -1.39090503e-02 -5.74102521e-01 -1.45010722e+00 9.73265648e-01 2.09952563e-01 -1.24144506e+00 -3.68034095e-01 4.87175509e-02 8.60488713e-01 -2.77015641e-02 4.31649715e-01 1.84054583e-01 7.36291766e-01 -1.67593396e+00 8.56213748e-01 6.70790374e-01 3.69495839e-01 -4.70862657e-01 4.73354280e-01 5.68336606e-01 -1.10351324e+00 1.37591019e-01 -1.36424616e-01 -4.32337075e-01 -2.80984849e-01 2.65925497e-01 -8.91049981e-01 1.16858020e-01 1.15540493e+00 3.49146396e-01 -1.09982359e+00 9.07153964e-01 -6.28792584e-01 1.18706688e-01 -2.16006115e-01 9.84036773e-02 3.44235361e-01 -1.08650491e-01 4.25342768e-01 1.00882447e+00 7.01598674e-02 1.19701967e-01 8.56461078e-02 1.00767505e+00 -1.61019061e-02 -1.34532586e-01 -1.23914218e+00 6.30052626e-01 6.27202034e-01 1.04771388e+00 -5.65382421e-01 -4.96029891e-02 -5.34491479e-01 9.43921626e-01 6.02411032e-01 5.81016660e-01 -8.30950558e-01 -1.45644128e-01 7.99311519e-01 4.40171063e-01 2.77071416e-01 -3.38100910e-01 -5.28752089e-01 -7.90661573e-01 3.97806853e-01 -7.82177687e-01 2.56739885e-01 -1.41458023e+00 -1.34704971e+00 4.11170572e-01 -1.38720527e-01 -1.28556180e+00 -7.12193698e-02 -8.63098502e-01 -6.98585808e-01 1.03524590e+00 -1.32268214e+00 -1.29152060e+00 -7.27573931e-01 8.88643205e-01 3.15852731e-01 -9.31164473e-02 8.60091269e-01 -7.42102414e-02 -4.20541465e-01 6.14041209e-01 -1.12648636e-01 5.32665193e-01 1.01725042e+00 -1.65667617e+00 5.14961123e-01 8.76461029e-01 6.34348452e-01 8.37017655e-01 6.56514108e-01 -3.59464109e-01 -1.03896916e+00 -1.02154279e+00 1.03714383e+00 -1.28083634e+00 7.47279644e-01 -4.99472201e-01 -8.98797631e-01 1.02810049e+00 2.84588307e-01 7.12980866e-01 7.64877260e-01 3.89945328e-01 -1.00293469e+00 -1.24556990e-02 -9.28747773e-01 1.06312597e+00 1.13473105e+00 -9.01441872e-01 -5.73284447e-01 6.24459982e-01 1.74452811e-01 -8.12822580e-01 -7.41873920e-01 8.15835968e-02 3.46150428e-01 -1.29730022e+00 1.37223339e+00 -6.84230924e-01 6.42820477e-01 -5.38419306e-01 -2.53195256e-01 -1.38327646e+00 -3.82674456e-01 -3.79003793e-01 1.75228462e-01 1.05253351e+00 4.57182735e-01 -4.96949017e-01 9.26082313e-01 1.03147650e+00 2.31099203e-01 -7.47884452e-01 -9.17889655e-01 -9.23902214e-01 4.25851792e-01 -5.36502063e-01 6.19398773e-01 1.21342754e+00 -7.30107352e-02 3.03384304e-01 7.60798901e-02 5.95818043e-01 4.11013037e-01 1.07916951e-01 1.13500047e+00 -1.05727386e+00 -2.24035725e-01 -7.05660999e-01 -6.86262071e-01 -1.27503109e+00 2.98079789e-01 -7.62816370e-01 2.82972381e-02 -1.53583598e+00 -5.51005118e-02 -6.17145240e-01 4.70132120e-02 6.32260799e-01 -2.39453152e-01 5.04322588e-01 1.74965322e-01 1.86291844e-01 -9.10445750e-01 1.28003016e-01 1.47915733e+00 -3.44272763e-01 3.35753709e-01 1.33657217e-01 -7.82038808e-01 1.04329777e+00 5.08408129e-01 -3.05855781e-01 -4.87234473e-01 -7.31977046e-01 2.60295004e-01 1.61464080e-01 1.13545978e+00 -1.03472435e+00 8.25000167e-01 -1.95068851e-01 8.03244054e-01 -3.36675018e-01 6.46857977e-01 -1.04172945e+00 -8.34755152e-02 -1.00904755e-01 -4.95846361e-01 4.35817510e-01 1.99990988e-01 6.11710429e-01 -5.03214225e-02 -8.61404762e-02 6.05107188e-01 -2.83252656e-01 -9.33546364e-01 1.70024276e-01 -2.43725106e-01 4.12160903e-01 1.37975001e+00 -3.10334116e-01 -8.46870840e-01 -5.98876476e-01 -8.04180622e-01 1.75155640e-01 6.74576283e-01 4.96544957e-01 3.83723766e-01 -1.47844243e+00 -5.05385578e-01 3.71454731e-02 4.36852753e-01 4.50907856e-01 2.98643082e-01 8.15195918e-01 -7.29856253e-01 2.93228447e-01 -9.51777212e-03 -7.75747299e-01 -1.41292155e+00 6.53490663e-01 4.57569569e-01 4.78011966e-02 -6.11651719e-01 1.30946541e+00 2.67839372e-01 -6.98361993e-01 2.90585041e-01 -2.45756194e-01 -1.34673953e-01 -9.35608447e-02 5.09163916e-01 2.75005817e-01 -3.10387220e-02 -1.08507252e+00 -7.23238051e-01 6.69907570e-01 8.19751248e-02 -3.49298194e-02 1.07849371e+00 9.35940966e-02 2.44128387e-02 4.60353047e-01 1.06583822e+00 3.98539335e-01 -1.41562808e+00 -4.93255645e-01 -1.26731871e-02 -8.93632889e-01 -3.62747699e-01 -8.32475960e-01 -7.63024271e-01 8.94028246e-01 2.63582647e-01 2.86499798e-01 5.56417048e-01 3.30219656e-01 2.72297919e-01 4.88977760e-01 3.43521833e-01 -9.55325723e-01 3.03137779e-01 6.07752681e-01 1.02556682e+00 -1.66329706e+00 1.50840104e-01 -5.76123059e-01 -9.21722889e-01 7.97337890e-01 9.14294302e-01 -2.02617601e-01 6.55266404e-01 3.44039649e-01 3.71364295e-01 -2.79806167e-01 -4.86841321e-01 -1.68740869e-01 6.70299113e-01 9.48780239e-01 4.23298687e-01 1.83442950e-01 7.83448696e-01 1.28392294e-01 -6.94780529e-01 -6.38270438e-01 2.91103005e-01 8.26170921e-01 -2.56299302e-02 -7.33381450e-01 -7.57555604e-01 -1.45052940e-01 -1.25038505e-01 -2.22466271e-02 -6.64502025e-01 9.19811010e-01 4.66870546e-01 1.13936830e+00 3.68648857e-01 -1.30949050e-01 6.20838165e-01 -1.54395103e-01 5.14234126e-01 -5.14080167e-01 -3.56111646e-01 -5.42960346e-01 1.82100188e-03 -7.35055387e-01 -9.73645300e-02 -8.33034635e-01 -1.11011720e+00 -5.98579288e-01 2.51422763e-01 -2.68591553e-01 2.99581468e-01 9.53476727e-01 2.11918458e-01 3.16685915e-01 1.75006956e-01 -8.89421284e-01 -2.05866978e-01 -7.14042902e-01 -4.10289139e-01 8.02317321e-01 3.98105472e-01 -7.38049924e-01 -6.76847398e-02 8.38214085e-02]

[10.463980674743652, 1.7307357788085938]

c618b637-fc6e-4fea-8fd7-7117388506c5

a-survey-on-evolutionary-computation-for

2209.06399

https://arxiv.org/abs/2209.06399v1

https://arxiv.org/pdf/2209.06399v1.pdf

A Survey on Evolutionary Computation for Computer Vision and Image Analysis: Past, Present, and Future Trends

Computer vision (CV) is a big and important field in artificial intelligence covering a wide range of applications. Image analysis is a major task in CV aiming to extract, analyse and understand the visual content of images. However, image-related tasks are very challenging due to many factors, e.g., high variations across images, high dimensionality, domain expertise requirement, and image distortions. Evolutionary computation (EC) approaches have been widely used for image analysis with significant achievement. However, there is no comprehensive survey of existing EC approaches to image analysis. To fill this gap, this paper provides a comprehensive survey covering all essential EC approaches to important image analysis tasks including edge detection, image segmentation, image feature analysis, image classification, object detection, and others. This survey aims to provide a better understanding of evolutionary computer vision (ECV) by discussing the contributions of different approaches and exploring how and why EC is used for CV and image analysis. The applications, challenges, issues, and trends associated to this research field are also discussed and summarised to provide further guidelines and opportunities for future research.

['Mengjie Zhang', 'Stefano Cagnoni', 'Pablo Mesejo', 'Bing Xue', 'Ying Bi']

2022-09-14

['edge-detection']

['computer-vision']

[ 7.19164848e-01 -6.03091896e-01 7.35874996e-02 -1.42751291e-01 -9.34883878e-02 -5.08263409e-01 1.71610162e-01 1.54095158e-01 -3.35109890e-01 2.85479844e-01 -6.14792824e-01 2.71038488e-02 -2.63983518e-01 -4.97491062e-01 -2.62188286e-01 -9.91380453e-01 2.00845838e-01 -6.10223822e-02 1.67804897e-01 -7.15226457e-02 7.57529318e-01 6.93381190e-01 -2.22014952e+00 5.96937053e-02 1.03001177e+00 1.10460758e+00 6.14744842e-01 7.96431124e-01 -3.23712468e-01 3.93845767e-01 -7.30349422e-01 -3.52761596e-01 -3.79378125e-02 -8.86707067e-01 -5.61913848e-01 5.94798028e-01 1.21010374e-02 4.66011673e-01 2.02865675e-01 1.37646139e+00 6.80895627e-01 2.05250010e-01 7.95262933e-01 -1.42849433e+00 -7.83600688e-01 1.18165771e-02 -6.82055712e-01 3.91265035e-01 -1.13754809e-01 7.57401660e-02 5.59651315e-01 -8.95429850e-01 4.70601618e-01 9.37580347e-01 5.49035013e-01 6.52079403e-01 -8.01689148e-01 -1.56561762e-01 1.03868790e-01 7.96374798e-01 -1.20776308e+00 -1.09148279e-01 9.36828494e-01 -3.64164233e-01 9.09280777e-01 6.29315674e-01 9.84229207e-01 4.19060111e-01 2.69778371e-01 1.02667367e+00 1.04075134e+00 -8.70568514e-01 4.69434172e-01 3.66269141e-01 1.96371138e-01 5.32750487e-01 5.38670607e-02 -6.36790469e-02 -3.51140618e-01 9.71013308e-02 3.56328845e-01 -1.48213416e-01 -2.66770244e-01 -4.94089663e-01 -7.51103878e-01 1.04546583e+00 4.04075950e-01 3.57514977e-01 -4.04208779e-01 -2.03276858e-01 5.35361409e-01 1.11125432e-01 1.13307111e-01 6.32141829e-01 -3.99990261e-01 -1.25469267e-02 -5.76359153e-01 1.12818182e-01 3.12669933e-01 6.49379075e-01 4.40164119e-01 2.88679957e-01 6.13735691e-02 1.25986421e+00 6.10777251e-02 3.09135556e-01 6.51512802e-01 -9.88380492e-01 -1.92164183e-01 6.50887489e-01 -5.03145456e-01 -1.33967364e+00 -2.42338404e-01 -2.52394855e-01 -8.98514211e-01 6.91954732e-01 -8.08251202e-02 1.61672354e-01 -8.56288671e-01 1.19754648e+00 3.04699868e-01 -6.30332157e-02 -7.09348097e-02 7.32657194e-01 1.22056365e+00 5.99596500e-01 -6.74660178e-03 -5.07269859e-01 1.65341723e+00 -1.15806878e+00 -6.41766489e-01 -6.07621431e-01 2.77564954e-02 -9.19820249e-01 7.10393906e-01 4.38698441e-01 -9.81097162e-01 -6.04675293e-01 -1.00989914e+00 1.46866590e-02 -5.52181065e-01 3.39612722e-01 6.02192581e-01 9.98998284e-01 -8.73135626e-01 1.27544671e-01 -6.16000533e-01 -6.26491606e-01 6.27459586e-01 2.48622328e-01 1.52853038e-02 -2.83929948e-02 -5.88160396e-01 8.22440565e-01 5.98566175e-01 1.80900678e-01 -1.44341201e-01 -4.49614018e-01 -8.23710322e-01 -2.33126715e-01 3.74950230e-01 -5.57857275e-01 8.58047366e-01 -1.39753318e+00 -1.35047030e+00 1.20957172e+00 -2.64870644e-01 -4.11069185e-01 2.73427069e-01 2.06850484e-01 -3.12529981e-01 1.28630117e-01 -2.51176178e-01 6.99753344e-01 9.72662807e-01 -1.21270859e+00 -8.92043829e-01 -5.87574124e-01 -4.61614639e-01 2.59516060e-01 -3.38368744e-01 3.81625116e-01 -8.02066445e-01 -7.64824986e-01 -1.45976283e-02 -9.78862107e-01 -2.91601717e-01 1.76052675e-01 8.71418801e-04 -2.37894848e-01 1.11701739e+00 -4.23322439e-01 1.25577879e+00 -2.25802755e+00 2.43928015e-01 2.38041952e-02 4.74202298e-02 7.82155454e-01 -2.13210471e-02 2.02809319e-01 3.12334616e-02 7.97579214e-02 -7.06035256e-01 2.92231794e-02 -4.38233644e-01 1.65042251e-01 7.99315721e-02 2.54239321e-01 2.76855856e-01 1.10281301e+00 -6.75005734e-01 -6.01904750e-01 7.26025581e-01 4.97230709e-01 -1.55838415e-01 -1.68607324e-01 -1.43082798e-01 4.59149331e-01 -3.17405373e-01 8.90220284e-01 6.99641109e-01 -1.75645407e-02 1.13973944e-02 -2.09778428e-01 -3.31188738e-01 -7.06133246e-01 -1.07458425e+00 1.22620857e+00 -2.04911917e-01 1.07544804e+00 7.36027658e-02 -1.49448121e+00 9.60444808e-01 -5.44309616e-02 5.09026289e-01 -8.37544441e-01 3.04475546e-01 1.71247602e-01 2.52734989e-01 -8.93159688e-01 4.07974720e-01 3.58391941e-01 3.01956654e-01 3.03698897e-01 -1.50211096e-01 -4.78381544e-01 4.35181201e-01 -2.60344476e-01 4.04763490e-01 -9.20161530e-02 6.64382696e-01 -2.52145343e-02 1.03618610e+00 3.61208111e-01 7.09186375e-01 5.90594232e-01 -7.09983349e-01 7.12709427e-01 -3.80949676e-02 -6.05999947e-01 -9.25991178e-01 -6.14958167e-01 -1.66230083e-01 7.06947267e-01 6.21409297e-01 -1.14563458e-01 -9.95692253e-01 -2.49221772e-01 -2.28233546e-01 5.22015691e-01 -7.28299499e-01 -2.51107991e-01 -5.52382827e-01 -1.32213950e+00 1.59348935e-01 3.29682797e-01 8.41472447e-01 -1.40614748e+00 -1.14183664e+00 8.50983337e-02 -1.23172224e-01 -1.02447987e+00 -1.23865865e-01 -2.93717355e-01 -9.90561664e-01 -1.15741158e+00 -7.01176167e-01 -1.17950606e+00 7.47324109e-01 7.44546175e-01 1.00209475e+00 3.65172297e-01 -1.20141530e+00 4.58252758e-01 -3.88625771e-01 -8.46298516e-01 -3.67650717e-01 -3.45024347e-01 -3.00540805e-01 -3.58570404e-02 5.54799557e-01 -5.87692000e-02 -6.69488311e-01 4.22149837e-01 -9.09284353e-01 -1.74490839e-01 6.08629704e-01 9.62970197e-01 9.21258509e-01 6.02859497e-01 2.46617526e-01 -5.40452421e-01 7.87401795e-01 -4.45786230e-02 -5.88573337e-01 7.57996976e-01 -7.54819751e-01 -3.24925423e-01 3.74111593e-01 -2.50758946e-01 -1.06655085e+00 1.50532335e-01 -2.56425086e-02 -2.30488092e-01 -2.30168268e-01 2.43109360e-01 -1.51698366e-01 -6.14832699e-01 6.03558242e-01 5.26242554e-01 3.85843247e-01 -1.90686986e-01 2.31322512e-01 7.11258590e-01 4.72228795e-01 -8.73946771e-02 2.73697764e-01 4.04923141e-01 1.07506484e-01 -1.32310665e+00 -4.43435013e-01 -5.14778137e-01 -5.44293821e-01 -5.56777239e-01 1.01899552e+00 -1.75599024e-01 -5.72188020e-01 8.69603574e-01 -8.69395673e-01 6.88227415e-02 -6.46194518e-02 3.50415915e-01 -4.74632889e-01 6.18647814e-01 -2.06666574e-01 -1.04208302e+00 -6.92266762e-01 -1.53843713e+00 7.09854543e-01 8.52999747e-01 -5.22334501e-02 -1.06972277e+00 -2.02552617e-01 5.64888954e-01 3.91090602e-01 3.47327769e-01 1.02636015e+00 -1.15839280e-02 -2.08468050e-01 -6.06057532e-02 -8.59709084e-02 4.45735365e-01 2.14047208e-01 6.38179839e-01 -8.83274019e-01 -3.01046297e-03 2.24454165e-01 -1.34614736e-01 7.99494743e-01 8.84625614e-01 1.39174795e+00 1.65429458e-01 -5.53994656e-01 5.72115719e-01 1.58282793e+00 8.18024576e-01 6.64949000e-01 5.10812104e-01 3.60937566e-01 8.93064022e-01 8.76708329e-01 2.49762863e-01 -5.05210049e-02 6.89684093e-01 4.35339838e-01 -7.66627789e-02 -3.29698026e-01 5.40090799e-01 -3.75824645e-02 6.58156514e-01 -2.72978753e-01 -3.42234999e-01 -8.64640594e-01 4.03572649e-01 -1.72641087e+00 -9.90754247e-01 -4.24647599e-01 2.06427050e+00 2.20588297e-01 -3.78870636e-01 6.05955571e-02 4.19737607e-01 1.19122231e+00 -2.89875895e-01 -9.05485332e-01 -6.94187224e-01 -4.48916823e-01 3.05236783e-02 7.67183006e-02 4.90542687e-02 -1.21706939e+00 7.77240455e-01 6.71075249e+00 9.40429151e-01 -1.34344733e+00 -2.75844604e-01 8.09619844e-01 1.84473202e-01 1.84583187e-01 -3.37251037e-01 -3.39125663e-01 5.99825442e-01 1.17306091e-01 -1.59529299e-01 4.69330817e-01 8.48587573e-01 -3.53174247e-02 -4.28303182e-01 -4.81155187e-01 1.51563776e+00 4.97725099e-01 -1.20086050e+00 -6.86687976e-02 -1.43732339e-01 9.74857986e-01 -6.98489323e-02 2.97197878e-01 -2.30964944e-01 -2.60007143e-01 -8.94136190e-01 5.63474059e-01 1.53612047e-01 3.92232835e-01 -9.31538880e-01 6.30526602e-01 2.34641626e-01 -1.09988153e+00 -3.44126165e-01 -4.98975486e-01 2.03149378e-01 1.25503510e-01 5.36178648e-01 -1.82529420e-01 5.43347836e-01 1.19600534e+00 6.35292709e-01 -6.96260631e-01 1.40412843e+00 -9.99509022e-02 1.91733271e-01 -4.51247506e-02 -2.85640597e-01 9.91041213e-02 -6.34902418e-01 6.57477617e-01 9.15693820e-01 2.75388002e-01 1.26866281e-01 -7.60872066e-02 7.47091174e-01 3.02188128e-01 2.22561091e-01 -4.13634658e-01 -1.85539216e-01 4.56614673e-01 1.38770425e+00 -1.30837142e+00 -1.45508707e-01 -4.40864772e-01 1.18872964e+00 -3.27883437e-02 2.03505069e-01 -6.57259047e-01 -6.66584134e-01 6.47342205e-01 -7.19106019e-01 3.50070089e-01 -5.16562499e-02 -6.02500975e-01 -8.76461625e-01 -8.50487195e-05 -8.90612602e-01 5.73749900e-01 -6.35397971e-01 -9.68217194e-01 5.66251338e-01 -1.21003583e-01 -1.19033897e+00 -1.79207042e-01 -7.86151707e-01 -6.96045697e-01 5.42793036e-01 -1.31882751e+00 -8.89364421e-01 -6.33501470e-01 3.46427977e-01 1.08193922e+00 -3.91453654e-01 5.90371549e-01 -4.01700065e-02 -8.00461650e-01 2.87027121e-01 3.15185726e-01 -1.91908851e-01 2.41212681e-01 -8.85452151e-01 3.50442439e-01 9.56218779e-01 2.15548083e-01 2.15122819e-01 6.68780982e-01 -3.28758031e-01 -1.56947136e+00 -7.38908887e-01 4.55942571e-01 -1.09855585e-01 -5.06298169e-02 1.11874729e-01 -8.44793558e-01 1.23640187e-02 2.82065004e-01 -4.76633385e-02 7.31663823e-01 -6.35471463e-01 2.71814704e-01 -3.75574008e-02 -1.40484166e+00 7.17336595e-01 8.97388935e-01 5.93234859e-02 -2.50687927e-01 -5.35199493e-02 -4.12960653e-04 -2.04319879e-01 -4.94060338e-01 3.88531059e-01 5.84820092e-01 -1.15357959e+00 1.20607162e+00 -1.13523632e-01 1.58732682e-01 -4.18133795e-01 2.81616956e-01 -1.26595175e+00 -3.92624378e-01 -2.65804648e-01 1.09117240e-01 1.15168333e+00 -7.18039349e-02 -8.09746325e-01 6.55741811e-01 5.76076746e-01 -1.78862195e-02 -8.35358083e-01 -6.29227877e-01 -6.29755020e-01 -1.76867843e-01 -4.34500605e-01 4.02552217e-01 6.93479776e-01 -3.01833689e-01 1.20073110e-01 -8.06835443e-02 -1.79687828e-01 8.53130579e-01 4.75899160e-01 4.66150761e-01 -1.25297129e+00 6.55908659e-02 -1.08864486e+00 -7.91215837e-01 -3.74746680e-01 -1.48013741e-01 -6.45796597e-01 -5.68834133e-02 -1.63622224e+00 2.73964196e-01 -2.80345678e-01 -3.51549722e-02 -1.14008069e-01 -4.75728035e-01 6.26159668e-01 3.60585243e-01 2.50897050e-01 -2.22943887e-01 3.98120075e-01 1.25575197e+00 -3.17636341e-01 -1.67832300e-01 1.55624896e-01 -7.18129277e-01 7.62336493e-01 1.16120839e+00 -2.19489530e-01 -6.06841624e-01 -5.24666727e-01 5.75968660e-02 -5.91322899e-01 2.01126426e-01 -7.55192041e-01 3.76504719e-01 -3.47564429e-01 5.66244662e-01 -4.99825776e-01 2.25883618e-01 -7.25561023e-01 2.34030247e-01 7.83300698e-01 4.55713794e-02 2.96747267e-01 3.10918123e-01 4.29674119e-01 -5.94226956e-01 -7.60579228e-01 1.11122322e+00 -4.29692924e-01 -1.43309927e+00 1.07466429e-01 -6.65144742e-01 -1.90136731e-01 1.68436742e+00 -8.94763470e-01 8.23825076e-02 -9.12612677e-02 -3.69355023e-01 2.57421672e-01 5.44280529e-01 5.23540258e-01 9.31650996e-01 -1.01800752e+00 -5.92510521e-01 3.61939609e-01 2.69124418e-01 -1.75639406e-01 6.18607283e-01 7.40988195e-01 -7.28558421e-01 2.41874903e-01 -4.76808637e-01 -8.56127441e-01 -1.98264503e+00 7.94981062e-01 3.05367440e-01 3.62397075e-01 -4.71220881e-01 9.21406746e-01 1.55290425e-01 -8.91938508e-02 8.05745423e-02 2.41865024e-01 -6.43026650e-01 1.53800976e-02 6.45512640e-01 7.53435373e-01 2.75159419e-01 -7.54143178e-01 -5.29426217e-01 1.30360889e+00 2.82643735e-02 2.10534379e-01 1.14634717e+00 -4.21342134e-01 -5.02108872e-01 9.63757858e-02 9.40485656e-01 -5.19491076e-01 -8.69357884e-01 3.01333927e-02 -1.05746418e-01 -5.40433466e-01 1.14488766e-01 -6.54184759e-01 -1.45271432e+00 1.07313061e+00 9.35067892e-01 1.22899212e-01 1.78055394e+00 -6.47558048e-02 4.75760370e-01 1.06676385e-01 2.31024221e-01 -1.48191667e+00 1.93674773e-01 2.50448942e-01 9.54121470e-01 -1.35783899e+00 1.93263516e-01 -6.16345942e-01 -9.00442958e-01 1.37511814e+00 5.29545844e-01 1.84816867e-01 5.82466364e-01 5.66897616e-02 2.71181732e-01 -1.13365635e-01 -3.53209436e-01 -3.27919126e-01 3.86015594e-01 9.68624830e-01 2.37029955e-01 -2.13111743e-01 -5.77235103e-01 -9.29435417e-02 7.43982475e-03 -1.56348273e-01 1.70010805e-01 1.11012757e+00 -4.28984046e-01 -1.20635927e+00 -6.09089434e-01 3.80571485e-01 -5.18851757e-01 2.24333301e-01 -6.30895376e-01 5.22720516e-01 1.54678270e-01 1.11022091e+00 7.58502334e-02 -1.95691064e-01 -4.27882709e-02 -3.19427907e-01 7.29134023e-01 -1.16051979e-01 -5.86103618e-01 -1.56748340e-01 -4.71880287e-01 -1.20600693e-01 -7.38391578e-01 -6.86751783e-01 -1.01600063e+00 -7.28042498e-02 -5.41110694e-01 4.90541495e-02 1.19528043e+00 6.92400455e-01 5.06686687e-01 5.37186503e-01 5.72757423e-01 -7.57491112e-01 9.49492007e-02 -4.51813072e-01 -4.01296139e-01 3.20346683e-01 -2.86120847e-02 -6.17580593e-01 -1.04197972e-01 3.55936557e-01]

[9.770865440368652, -0.44271019101142883]

5c63e00b-e5a9-4200-b4a7-1c8a8b0a80c5

spatial-temporal-attention-network-for-open

2211.1394

https://arxiv.org/abs/2211.13940v1

https://arxiv.org/pdf/2211.13940v1.pdf

Spatial-Temporal Attention Network for Open-Set Fine-Grained Image Recognition

Triggered by the success of transformers in various visual tasks, the spatial self-attention mechanism has recently attracted more and more attention in the computer vision community. However, we empirically found that a typical vision transformer with the spatial self-attention mechanism could not learn accurate attention maps for distinguishing different categories of fine-grained images. To address this problem, motivated by the temporal attention mechanism in brains, we propose a spatial-temporal attention network for learning fine-grained feature representations, called STAN, where the features learnt by implementing a sequence of spatial self-attention operations corresponding to multiple moments are aggregated progressively. The proposed STAN consists of four modules: a self-attention backbone module for learning a sequence of features with self-attention operations, a spatial feature self-organizing module for facilitating the model training, a spatial-temporal feature learning module for aggregating the re-organized features via a Long Short-Term Memory network, and a context-aware module that is implemented as the forget block of the spatial-temporal feature learning module for preserving/forgetting the long-term memory by utilizing contextual information. Then, we propose a STAN-based method for open-set fine-grained recognition by integrating the proposed STAN network with a linear classifier, called STAN-OSFGR. Extensive experimental results on 3 fine-grained datasets and 2 coarse-grained datasets demonstrate that the proposed STAN-OSFGR outperforms 9 state-of-the-art open-set recognition methods significantly in most cases.

['Qiulei Dong', 'Hong Wang', 'Jiayin Sun']

2022-11-25

['fine-grained-image-recognition', 'open-set-learning']

['computer-vision', 'miscellaneous']

[ 8.02144930e-02 -3.12974244e-01 1.75855577e-01 -4.31957275e-01 -5.11850297e-01 -2.42699504e-01 6.35870814e-01 -7.73419961e-02 -2.01284558e-01 5.07092476e-01 1.61289126e-01 7.44284317e-02 -6.13760889e-01 -8.44889998e-01 -8.49838436e-01 -8.73749077e-01 -1.28754765e-01 2.05921665e-01 6.65383399e-01 1.19298451e-01 5.77770591e-01 4.91518319e-01 -2.01392174e+00 4.82676655e-01 1.08305359e+00 1.54156256e+00 5.57082355e-01 3.13755840e-01 -4.26763743e-01 7.96261311e-01 -2.14783162e-01 7.35263228e-02 1.45426676e-01 -1.51406527e-01 -7.48291671e-01 1.97849378e-01 7.48656332e-01 -9.16154161e-02 -4.25884575e-01 1.04675317e+00 1.86111465e-01 3.41381073e-01 7.37294614e-01 -1.04887986e+00 -1.51207101e+00 3.18531752e-01 -4.87476915e-01 6.41438425e-01 2.74867695e-02 1.34138212e-01 8.65698099e-01 -1.29099000e+00 1.32361963e-01 1.38670862e+00 7.05329418e-01 2.95052797e-01 -1.00402093e+00 -7.51672626e-01 4.05071437e-01 7.07374275e-01 -1.57715845e+00 -1.98949054e-01 6.96379960e-01 -5.21649539e-01 1.20170927e+00 4.46661450e-02 8.96762848e-01 7.32875586e-01 6.54293537e-01 5.58141112e-01 1.34773839e+00 -2.76216120e-01 1.44240260e-01 -1.32446855e-01 6.93589568e-01 9.68240559e-01 1.45405859e-01 3.40476364e-01 -6.81876600e-01 4.10414524e-02 1.01880884e+00 6.52935445e-01 -1.61992550e-01 -3.75916272e-01 -1.19063997e+00 5.49205780e-01 9.64076579e-01 6.32293344e-01 -4.70368326e-01 1.38758827e-04 1.59889996e-01 3.79775375e-01 5.36235213e-01 2.58807659e-01 -3.98153186e-01 4.05112207e-01 -9.14834082e-01 -9.75015238e-02 2.73769021e-01 7.79749691e-01 1.07619619e+00 -4.50906195e-02 -5.98505318e-01 8.83617222e-01 3.53498548e-01 3.59958231e-01 1.13163304e+00 -4.06604111e-01 1.88610200e-02 9.89824235e-01 -1.89047709e-01 -9.96395290e-01 -3.23911697e-01 -5.20115316e-01 -1.05762851e+00 1.76881716e-01 1.09732300e-01 4.84060049e-01 -1.23283255e+00 1.57442689e+00 7.95274228e-02 5.30294299e-01 -1.17628567e-01 7.95449972e-01 9.36648428e-01 6.09224021e-01 1.02871075e-01 -1.76529929e-01 1.41207290e+00 -1.27835011e+00 -4.92611408e-01 -1.37848139e-01 7.68287927e-02 -3.62633020e-01 1.15035856e+00 5.24129085e-02 -7.42197990e-01 -9.98461425e-01 -1.14564657e+00 -1.77839115e-01 -8.22491884e-01 -1.00021571e-01 5.61237931e-01 1.10940740e-01 -1.14988577e+00 6.30958855e-01 -5.90526223e-01 -5.38978875e-01 8.30615044e-01 3.63291293e-01 -4.98716742e-01 2.63121948e-02 -9.35067832e-01 7.25949049e-01 4.49117541e-01 2.77891941e-02 -1.00719547e+00 -7.54778743e-01 -8.10666144e-01 4.88570452e-01 5.68041019e-02 -8.19965839e-01 8.69479239e-01 -1.04211986e+00 -1.31226492e+00 8.81253064e-01 -2.52281606e-01 -3.88800651e-01 -8.08171183e-02 -1.08466193e-01 -4.28688526e-01 -9.73358285e-03 3.57515126e-01 6.02534533e-01 1.41036212e+00 -1.02410424e+00 -8.52917969e-01 -7.88059473e-01 -1.76516712e-01 9.85828564e-02 -5.63095629e-01 -3.42485696e-01 -1.90419659e-01 -9.59004939e-01 2.33200282e-01 -5.73071778e-01 4.10397202e-02 -1.51544092e-02 -1.19637005e-01 -5.79899848e-01 1.07446361e+00 -3.45977932e-01 1.23374379e+00 -2.14440870e+00 4.66303863e-02 -3.22639346e-02 3.70368212e-01 2.57039726e-01 -3.10971677e-01 -3.13066281e-02 -3.71513128e-01 -1.02287963e-01 -1.12703130e-01 -1.86773762e-01 -1.94966316e-01 7.34955817e-02 -6.83126450e-01 3.13858420e-01 2.75526494e-01 1.20323312e+00 -9.71717775e-01 -4.73942220e-01 4.74132478e-01 4.53330964e-01 -2.52689332e-01 4.33887750e-01 -5.31687401e-02 2.97306061e-01 -4.32044566e-01 6.93575740e-01 6.15208566e-01 -5.45835555e-01 -4.70773816e-01 -3.82461339e-01 -4.56118345e-01 -1.52320713e-01 -8.48785102e-01 1.60445106e+00 -3.12738597e-01 1.95167542e-01 -4.04287517e-01 -8.14006567e-01 1.09999764e+00 -5.88897467e-02 9.96467024e-02 -1.03106689e+00 1.14307642e-01 9.28331241e-02 -1.78594306e-01 -3.77585441e-01 3.75888973e-01 -1.36539876e-01 -8.87985621e-03 5.06900787e-01 6.33709490e-01 2.02656314e-01 -6.46336749e-02 -7.08378553e-02 1.13545048e+00 -1.41815096e-01 2.75659889e-01 -5.19114792e-01 6.26397133e-01 -1.82113931e-01 5.70913255e-01 7.75159001e-01 -3.84557217e-01 5.48930764e-01 -2.21607149e-01 -9.87973630e-01 -8.48859608e-01 -1.10490859e+00 -3.28683734e-01 1.47348428e+00 3.11916649e-01 -1.98060676e-01 -6.65735722e-01 -6.71424568e-01 1.95689797e-01 2.63930380e-01 -1.15987301e+00 -5.43444932e-01 -1.38643548e-01 -4.12758768e-01 1.52958810e-01 7.70238042e-01 9.35264051e-01 -1.50809634e+00 -7.29398489e-01 1.29026979e-01 2.13097632e-01 -5.49498081e-01 -8.37444067e-01 5.39161742e-01 -8.35149169e-01 -1.03783381e+00 -7.52323091e-01 -1.08605194e+00 6.92745686e-01 7.43641615e-01 8.69236410e-01 -2.14243270e-02 -3.71398419e-01 3.68666798e-01 -2.98862427e-01 -2.92583048e-01 4.71569091e-01 -1.07212067e-02 2.35268891e-01 4.80784297e-01 4.75143671e-01 -8.19500983e-01 -5.98307967e-01 3.53806794e-01 -8.93204391e-01 1.44596153e-04 8.39065254e-01 1.26782584e+00 8.93306613e-01 -9.52235088e-02 7.19976842e-01 -6.70024753e-01 5.09878635e-01 -3.30543846e-01 -5.46090364e-01 4.73214328e-01 -5.03683448e-01 4.30207253e-01 7.70471334e-01 -5.24150789e-01 -1.08465695e+00 -1.63944125e-01 1.04360275e-01 -7.51295149e-01 -2.34425962e-01 1.55884698e-01 -1.57296360e-01 -5.48920751e-01 4.26588446e-01 8.89110327e-01 -2.85581887e-01 -4.94448781e-01 4.11635131e-01 6.14031255e-01 6.63879752e-01 -3.66925091e-01 7.64608324e-01 5.53480923e-01 -3.49372447e-01 -7.41273463e-01 -1.35157609e+00 -4.19885159e-01 -9.22094941e-01 7.20652984e-03 9.92386341e-01 -7.56609023e-01 -6.26248479e-01 8.11247289e-01 -9.02667046e-01 -1.60022721e-01 -6.34690702e-01 2.98534501e-02 -7.29485512e-01 2.19930008e-01 -3.64526510e-01 -4.46195841e-01 -5.46877444e-01 -9.62256432e-01 1.18494785e+00 6.41651213e-01 5.79694621e-02 -8.90646160e-01 2.22315624e-01 5.10491133e-02 5.56806087e-01 -1.99771792e-01 9.98320401e-01 -4.42755461e-01 -7.71779835e-01 2.60179639e-01 -6.22633100e-01 1.62462756e-01 3.43345761e-01 -4.73430157e-01 -1.24493647e+00 -3.92688215e-01 4.46292087e-02 -4.77905512e-01 1.40682387e+00 5.00135839e-01 1.60135865e+00 -3.36585641e-01 -4.86996502e-01 1.01966274e+00 1.33687544e+00 2.79317528e-01 5.48740029e-01 4.01001185e-01 8.68091941e-01 2.17561886e-01 4.04721469e-01 4.10015196e-01 4.12339211e-01 2.80872762e-01 3.99251908e-01 9.19234455e-02 -2.13595673e-01 -2.79877305e-01 3.26306336e-02 8.21187437e-01 -5.47328927e-02 3.78558517e-01 -5.79633653e-01 8.02677870e-01 -1.95005167e+00 -1.17371726e+00 5.23784995e-01 2.13035941e+00 6.31922483e-01 -2.77729984e-02 -2.06573516e-01 4.31198487e-03 7.90328383e-01 4.19094384e-01 -8.94599557e-01 -2.46262878e-01 -2.09705487e-01 1.47283792e-01 9.18680206e-02 1.76877901e-01 -1.26069963e+00 1.09828365e+00 5.75633860e+00 9.04734373e-01 -1.18697548e+00 1.70388624e-01 5.54074287e-01 2.52558708e-01 -2.01370329e-01 -2.50347674e-01 -7.96532273e-01 4.76790488e-01 5.92404902e-01 -2.15734094e-01 4.20336634e-01 9.02111292e-01 -2.57338077e-01 1.54241428e-01 -1.07486379e+00 1.23889649e+00 3.59218776e-01 -1.46386516e+00 5.22504270e-01 -1.85158700e-01 7.11554706e-01 2.77641695e-02 3.42409074e-01 4.08380538e-01 9.58250985e-02 -1.12762678e+00 6.78910971e-01 1.11106086e+00 9.26975965e-01 -5.27660429e-01 5.01428306e-01 2.75346071e-01 -1.59181035e+00 -5.10561705e-01 -7.20940411e-01 -7.06452727e-02 -4.52451259e-01 4.57404345e-01 -2.20898896e-01 2.85739660e-01 1.17096174e+00 1.02140808e+00 -8.71708035e-01 1.24982834e+00 3.41670550e-02 1.70049906e-01 1.07149571e-01 1.15191996e-01 3.89915854e-01 2.96974834e-02 3.24814975e-01 1.00918365e+00 3.55043977e-01 2.84989178e-01 1.04261763e-01 8.83057058e-01 -3.36966515e-02 -2.03225404e-01 -6.62440419e-01 1.67386413e-01 4.87228245e-01 1.35820520e+00 -6.15432382e-01 -3.00748706e-01 -4.42475110e-01 1.08946502e+00 8.96306694e-01 2.33274058e-01 -5.40777504e-01 -6.49922788e-01 6.45380199e-01 2.75805909e-02 8.52157652e-01 -1.19586168e-02 -3.07925850e-01 -1.12120473e+00 -1.06781311e-01 -4.91205692e-01 5.19254565e-01 -9.51970756e-01 -1.64635193e+00 1.04139388e+00 -4.14358407e-01 -1.08170950e+00 1.51046798e-01 -6.17794037e-01 -8.29597771e-01 8.43431652e-01 -1.73444915e+00 -1.47523522e+00 -6.72112823e-01 1.14174116e+00 7.92133093e-01 -3.77656788e-01 8.89635861e-01 -6.26596063e-02 -3.72714549e-01 6.47404313e-01 1.30757213e-01 -1.14001237e-01 7.26467669e-01 -1.21174586e+00 1.72285929e-01 6.04236186e-01 2.23888442e-01 7.53977537e-01 1.19657256e-01 -5.36970019e-01 -1.33719790e+00 -1.61968958e+00 8.70207012e-01 -2.83482641e-01 5.92657030e-01 -2.67976910e-01 -1.34145641e+00 8.04819465e-01 1.58213809e-01 6.54123962e-01 4.31110710e-01 8.55650976e-02 -6.81713164e-01 -4.76061732e-01 -1.06000090e+00 1.74534991e-01 1.20183074e+00 -7.03077078e-01 -1.29105747e+00 1.03717804e-01 8.54718983e-01 -7.69670457e-02 -6.97537839e-01 2.49121010e-01 6.07469916e-01 -1.06307495e+00 9.39536095e-01 -4.46425498e-01 4.17665504e-02 -5.07124782e-01 -1.46856830e-01 -1.52098644e+00 -1.40784454e+00 -1.05052322e-01 -2.13024408e-01 1.05356383e+00 -1.31734684e-01 -7.73686707e-01 4.60912913e-01 5.65995462e-02 -4.35472786e-01 -8.30734730e-01 -1.12633932e+00 -7.47231603e-01 -7.19849095e-02 -3.59616950e-02 6.71627641e-01 7.05411732e-01 -2.11579427e-01 5.34022868e-01 -1.37908414e-01 1.69924811e-01 8.47417772e-01 7.41799593e-01 1.67143241e-01 -1.56674969e+00 8.57526213e-02 -4.61591065e-01 -6.17897987e-01 -1.03205347e+00 3.04047853e-01 -7.93554485e-01 1.20464876e-01 -1.38909030e+00 7.05097497e-01 -3.63452315e-01 -8.38478625e-01 6.99422061e-01 -2.12934494e-01 3.55784535e-01 8.30006078e-02 3.57281268e-01 -1.14169812e+00 9.99512911e-01 1.28722703e+00 -4.76067215e-01 -3.21908891e-02 -3.37914646e-01 -8.62223327e-01 8.03963780e-01 3.04753691e-01 -2.40356743e-01 -3.03147316e-01 -3.74944210e-01 -4.31419462e-01 -4.03414637e-01 6.13801956e-01 -1.44742179e+00 5.87234378e-01 -9.45255309e-02 8.76160502e-01 -7.74390876e-01 1.20853439e-01 -9.25481498e-01 -1.42227650e-01 3.23236436e-01 -1.31233037e-01 2.20793928e-03 2.88403392e-01 6.05660915e-01 -3.26389164e-01 1.76947936e-01 9.23005581e-01 -2.51635253e-01 -1.31571448e+00 8.80164027e-01 -1.66412652e-01 -3.49302068e-02 1.19575953e+00 -5.05700111e-01 -5.51691055e-01 2.84055740e-01 -7.82713652e-01 2.61810422e-01 2.53415704e-01 6.04208231e-01 1.04295743e+00 -1.67466927e+00 -3.12999219e-01 7.53622413e-01 4.65632260e-01 -1.00997977e-01 7.65168250e-01 4.37387705e-01 1.79968312e-01 7.58755744e-01 -8.01489294e-01 -7.10717440e-01 -9.61966455e-01 9.29779947e-01 3.74975294e-01 -3.49423617e-01 -6.10056102e-01 1.04664063e+00 8.68708551e-01 -3.30848932e-01 1.69051424e-01 -6.46745265e-01 -4.40993696e-01 3.86806168e-02 9.22587276e-01 2.25123361e-01 8.01867470e-02 -7.98687279e-01 -4.59914923e-01 1.09960377e+00 -3.85115862e-01 5.11808097e-01 1.66047752e+00 -3.32875520e-01 -2.68185794e-01 7.61285722e-01 1.06575763e+00 -6.68489397e-01 -1.58192265e+00 -6.97554708e-01 -1.58861190e-01 -4.19577718e-01 1.57803431e-01 -6.92407966e-01 -8.86714935e-01 9.06789422e-01 8.84520590e-01 2.17627898e-01 1.43991220e+00 1.48625284e-01 7.37544060e-01 4.36521173e-01 6.50956094e-01 -7.69847929e-01 3.47020954e-01 1.15004694e+00 1.25908399e+00 -1.27236462e+00 -4.03296381e-01 -3.40819806e-02 -2.71929681e-01 9.79216039e-01 9.52187240e-01 -5.46741664e-01 1.04494226e+00 3.64037827e-02 -1.76921472e-01 -2.48668909e-01 -9.37183619e-01 -2.18255579e-01 6.35128736e-01 7.46894777e-01 7.74495006e-02 1.97420660e-02 2.47387066e-01 1.12071383e+00 -8.26102048e-02 1.20101981e-01 -1.91639915e-01 6.29246354e-01 -9.44989383e-01 -5.71234405e-01 -3.74894708e-01 7.81116545e-01 6.73970878e-02 -2.43353918e-01 -1.71736464e-01 3.09772879e-01 4.32482213e-01 4.56544638e-01 5.54165840e-01 -6.92378581e-01 2.87387997e-01 9.29930806e-02 5.34435093e-01 -7.55771935e-01 -6.65590346e-01 -3.27482671e-01 -7.66353071e-01 -8.17215800e-01 -2.92759597e-01 -4.59464908e-01 -1.06071103e+00 -1.24279462e-01 -1.99697539e-01 5.86682297e-02 -1.12452276e-01 1.14481544e+00 6.81741059e-01 8.61754239e-01 5.98083258e-01 -1.12205625e+00 -4.78728145e-01 -1.10678589e+00 -8.47657084e-01 4.23187852e-01 6.14059567e-01 -1.00912035e+00 -2.20652118e-01 7.52109066e-02]

[9.677848815917969, 2.0548439025878906]

1a80665f-26bd-40dc-86c3-16ab88c9613c

a-novel-method-for-comparative-analysis-of

1403.1523

http://arxiv.org/abs/1403.1523v2

http://arxiv.org/pdf/1403.1523v2.pdf

A Novel Method for Comparative Analysis of DNA Sequences by Ramanujan-Fourier Transform

Alignment-free sequence analysis approaches provide important alternatives over multiple sequence alignment (MSA) in biological sequence analysis because alignment-free approaches have low computation complexity and are not dependent on high level of sequence identity, however, most of the existing alignment-free methods do not employ true full information content of sequences and thus can not accurately reveal similarities and differences among DNA sequences. We present a novel alignment-free computational method for sequence analysis based on Ramanujan-Fourier transform (RFT), in which complete information of DNA sequences is retained. We represent DNA sequences as four binary indicator sequences and apply RFT on the indicator sequences to convert them into frequency domain. The Euclidean distance of the complete RFT coefficients of DNA sequences are used as similarity measure. To address the different lengths in Euclidean space of RFT coefficients, we pad zeros to short DNA binary sequences so that the binary sequences equal the longest length in the comparison sequence data. Thus, the DNA sequences are compared in the same dimensional frequency space without information loss. We demonstrate the usefulness of the proposed method by presenting experimental results on hierarchical clustering of genes and genomes. The proposed method opens a new channel to biological sequence analysis, classification, and structural module identification.

['Changchuan Yin', 'Xuemeng E. Yin', 'Jiasong Wang']

2014-03-06

['multiple-sequence-alignment']

['medical']

[ 8.99765074e-01 -7.53436208e-01 -1.27958804e-01 -2.58437663e-01 -4.01279628e-01 -8.20956230e-01 1.28071070e-01 4.70274925e-01 -6.31032884e-01 8.59875321e-01 -1.75309107e-01 -3.26878786e-01 -3.43202889e-01 -7.94385135e-01 -3.35145414e-01 -1.33750987e+00 -1.58283412e-01 2.91500181e-01 3.33896995e-01 -2.05608636e-01 6.79529071e-01 4.31194544e-01 -1.50455308e+00 3.07046801e-01 9.66106653e-01 4.00302798e-01 7.12722242e-01 8.54468465e-01 -5.37161410e-01 8.26619864e-02 -7.67718434e-01 1.16199836e-01 2.07570359e-01 -6.13126040e-01 -8.04634213e-01 -7.96234906e-02 -3.05927515e-01 1.29882663e-01 5.23412563e-02 1.17859519e+00 3.22985619e-01 2.17519850e-02 8.45684588e-01 -1.14893353e+00 -5.33155799e-01 2.67293751e-01 -7.41489291e-01 1.36959419e-01 5.03718555e-01 -8.15086886e-02 7.19455242e-01 -8.37907851e-01 6.76594198e-01 1.27881205e+00 8.12795281e-01 9.68678668e-02 -1.17023563e+00 -4.61439729e-01 -4.83367831e-01 6.81652427e-01 -1.60551107e+00 9.99324918e-02 4.38553989e-01 -5.47818601e-01 9.15434778e-01 5.69735944e-01 5.41173756e-01 5.62121212e-01 3.70782346e-01 2.98742980e-01 1.00795794e+00 -5.67071617e-01 1.01408381e-02 -4.02806729e-01 3.89742732e-01 4.81049538e-01 1.55538678e-01 -4.97243941e-01 -5.87604232e-02 -5.40809512e-01 3.37468743e-01 2.59849101e-01 -4.12111372e-01 -1.77628174e-01 -1.50951171e+00 8.93517256e-01 -9.19510126e-02 6.89173222e-01 -2.83576995e-01 -2.94274926e-01 5.04993677e-01 2.32841223e-01 -1.92538440e-01 -6.83653541e-03 -1.10068142e-01 -1.63380310e-01 -6.16965353e-01 5.79418316e-02 4.81311500e-01 6.98443711e-01 9.62921977e-01 -2.84209996e-01 2.72661060e-01 8.96062195e-01 7.75645440e-03 4.50203627e-01 9.85478401e-01 -6.95557475e-01 1.00237004e-01 5.73020518e-01 3.90112540e-03 -1.29395497e+00 -3.99300575e-01 5.10131977e-02 -9.72357988e-01 -2.11060733e-01 3.54625225e-01 3.08600515e-01 -6.13603234e-01 1.37224936e+00 4.18065012e-01 3.60134006e-01 2.51745462e-01 7.51788020e-01 3.55026871e-01 1.01264763e+00 -2.47895092e-01 -6.03370130e-01 1.50568306e+00 -5.18139899e-01 -6.62849605e-01 4.04079735e-01 5.77588558e-01 -1.12599862e+00 9.38117743e-01 3.55404526e-01 -6.09094977e-01 -4.81951356e-01 -1.22745788e+00 2.18118414e-01 -2.48669565e-01 -2.23667383e-01 2.21569106e-01 7.11196542e-01 -7.71433532e-01 6.89586282e-01 -7.00561464e-01 -4.12971079e-01 -4.36405212e-01 3.37738901e-01 -5.13451159e-01 2.96391658e-02 -1.20985663e+00 6.59316480e-01 8.87180030e-01 -2.35532783e-02 -3.09605151e-01 -2.00030699e-01 -6.50856853e-01 1.70024429e-02 6.14123382e-02 -2.07020387e-01 6.23508394e-01 -7.41067111e-01 -1.13904440e+00 6.99070036e-01 -4.25964236e-01 -2.67802149e-01 9.79641005e-02 2.86730498e-01 -3.69601101e-01 3.22520971e-01 -1.43948682e-02 6.26435429e-02 4.91768032e-01 -8.87028456e-01 -4.72726166e-01 -1.35743544e-01 -7.27332950e-01 -1.60497203e-02 -2.50407904e-01 2.91574169e-02 2.19045375e-02 -7.24814296e-01 3.83165538e-01 -9.33173656e-01 -2.55380660e-01 -4.18257922e-01 -1.03347637e-01 -3.03951278e-02 8.68303955e-01 -8.39038551e-01 1.27602983e+00 -2.30645585e+00 2.64125794e-01 5.50647378e-01 -1.17122322e-01 3.57915848e-01 -3.65913033e-01 8.43688965e-01 -3.99653733e-01 1.87701546e-02 -6.74139857e-01 7.31399953e-01 -3.67796868e-01 4.76852208e-01 -5.25509305e-02 7.58537233e-01 7.70716444e-02 2.41297156e-01 -1.00286651e+00 -5.66051006e-01 1.16813555e-01 4.33986902e-01 -2.68575668e-01 2.61433739e-02 1.06340006e-01 4.80393201e-01 -4.75678504e-01 2.94156373e-01 1.02533853e+00 -1.36272207e-01 6.96624517e-01 -2.49361902e-01 -3.80881786e-01 1.35987580e-01 -1.20756876e+00 1.42884016e+00 1.24740317e-01 3.18977445e-01 -2.74268240e-01 -1.41206670e+00 1.27973783e+00 2.14799270e-01 7.71496117e-01 -1.91631019e-01 -2.48477891e-01 3.61177951e-01 4.72254872e-01 -5.61487198e-01 4.24628198e-01 3.15404981e-02 1.23043410e-01 2.80422837e-01 -3.44904065e-01 1.37223795e-01 4.10485148e-01 -2.71348692e-02 1.01440775e+00 -1.83069659e-03 8.29712391e-01 -5.62441766e-01 1.09369028e+00 -7.89027195e-03 8.45816553e-01 3.55785608e-01 -1.67710766e-01 4.15567726e-01 3.43824983e-01 -4.36762422e-01 -1.56274664e+00 -9.54115510e-01 -3.15445662e-01 8.64674032e-01 1.96170509e-01 -3.78977269e-01 -8.12447727e-01 -1.41454861e-01 7.12314025e-02 2.62109101e-01 -3.12711418e-01 -4.71225642e-02 -8.71304512e-01 -1.17211843e+00 8.24428499e-01 -1.14761353e-01 3.74646932e-01 -6.77466512e-01 -4.95837122e-01 3.43523324e-01 -5.65132678e-01 -6.41865849e-01 -5.53071320e-01 1.03716858e-01 -7.19014406e-01 -1.13404214e+00 -8.83569419e-01 -1.01809084e+00 4.19580281e-01 6.81724906e-01 4.09376115e-01 2.26471916e-01 -6.58378124e-01 -2.68811673e-01 -4.30506617e-01 8.72306153e-02 -7.71812379e-01 -3.31895389e-02 9.54233631e-02 -4.87168320e-02 6.84063256e-01 -7.61731446e-01 -7.56200194e-01 6.19271100e-01 -9.83892858e-01 -2.82969922e-01 5.06028891e-01 1.07219934e+00 5.30564249e-01 1.52583793e-01 5.11811137e-01 -4.08352137e-01 5.53140402e-01 -6.24615192e-01 -4.80730772e-01 5.64686000e-01 -3.55613619e-01 2.91570067e-01 7.61590004e-01 -2.84566671e-01 -8.00522745e-01 1.08833171e-01 -3.51264745e-01 5.84905371e-02 -5.49677908e-02 5.32260537e-01 -2.70462744e-02 -1.24103606e-01 6.59146905e-01 9.82003689e-01 3.84318262e-01 -4.28389847e-01 -9.52950343e-02 1.01041842e+00 7.04972982e-01 -4.44385350e-01 3.47229242e-01 3.07955623e-01 3.69240731e-01 -1.01516521e+00 6.38802871e-02 -9.41284716e-01 -9.03170586e-01 1.69032976e-01 8.14957201e-01 -4.42330390e-01 -1.14583743e+00 4.87961680e-01 -9.95857894e-01 5.49975395e-01 5.37639380e-01 5.41622043e-01 -6.51064098e-01 1.55222642e+00 -6.15295410e-01 -7.79811859e-01 -2.86816567e-01 -1.27007961e+00 8.57146204e-01 -1.52545139e-01 -3.39574128e-01 -5.88359416e-01 3.37141991e-01 -5.49918301e-02 -2.84992531e-02 4.34325725e-01 1.52216673e+00 -6.74596488e-01 -2.17004299e-01 2.49325726e-02 6.84554130e-02 8.86675492e-02 6.14751756e-01 3.26890945e-01 -3.74299467e-01 -4.63176012e-01 2.26988837e-01 1.68340951e-01 6.56862855e-01 1.66295931e-01 9.43504274e-01 -2.50017405e-01 -5.62500000e-01 5.04468501e-01 1.48191047e+00 8.08589816e-01 7.64015079e-01 6.15154803e-01 3.36660057e-01 8.27315450e-01 9.56780076e-01 6.30762577e-01 -2.04439566e-01 7.23448277e-01 5.78419715e-02 2.02220052e-01 3.99842054e-01 1.01449519e-01 2.66649038e-01 1.28246772e+00 -1.50549114e-01 -1.83337420e-01 -1.08927095e+00 3.61429274e-01 -1.81381929e+00 -1.25618792e+00 -5.00159085e-01 2.31846857e+00 1.08015037e+00 -2.58250147e-01 2.89141804e-01 6.04507506e-01 1.20728028e+00 -1.79733440e-01 -4.74503636e-01 -5.27896345e-01 -4.66536820e-01 -3.19037996e-02 5.90312898e-01 4.21340942e-01 -8.51232708e-01 2.76085109e-01 6.82694960e+00 1.15591562e+00 -9.10799623e-01 -3.37361395e-01 1.74730524e-01 3.66478294e-01 -2.70671695e-01 -6.14099810e-03 -4.76416945e-01 8.14939678e-01 9.90876615e-01 -3.71818751e-01 1.64213777e-01 6.44528866e-01 8.74109790e-02 -1.71287909e-01 -6.71501756e-01 9.72756028e-01 -3.53406489e-01 -1.17495334e+00 3.21793288e-01 8.59082714e-02 3.08160096e-01 -3.74868214e-01 -6.84793591e-02 -4.99377936e-01 1.04364835e-01 -8.84469628e-01 1.78739458e-01 2.85438478e-01 5.84191382e-01 -1.09542358e+00 8.25149179e-01 3.79559040e-01 -1.37454605e+00 3.82757336e-02 -8.49439442e-01 7.84482583e-02 1.84492305e-01 6.43491924e-01 -1.13680494e+00 8.90916288e-01 4.10843730e-01 6.12416267e-01 -9.71780717e-02 9.48025048e-01 4.70520020e-01 4.67562199e-01 -2.42879957e-01 -2.67548084e-01 3.14173758e-01 -9.51410949e-01 5.46957433e-01 1.51817095e+00 7.11571813e-01 2.46234462e-01 2.67029732e-01 1.18858971e-01 6.47372127e-01 5.02378404e-01 -4.71787632e-01 -1.59970894e-01 6.87220097e-01 6.97409809e-01 -9.86351371e-01 -4.05755788e-01 -4.90602791e-01 9.49192762e-01 -7.34811351e-02 -3.68531756e-02 -6.32369041e-01 -8.57801139e-01 8.31729352e-01 -2.30922356e-01 2.95402408e-01 -5.74435651e-01 2.72861123e-02 -7.92939365e-01 -2.31281102e-01 -1.10022020e+00 6.16710722e-01 -3.95180792e-01 -1.12912548e+00 4.36675668e-01 -7.95726851e-02 -1.29342461e+00 -5.52325428e-01 -5.20577490e-01 -1.37117635e-02 1.04412282e+00 -1.04541755e+00 -4.19463634e-01 -1.08716130e-01 6.16441905e-01 4.66906309e-01 -1.31917760e-01 1.04335737e+00 3.42347115e-01 -3.07196587e-01 4.91427451e-01 1.05789566e+00 -5.01109939e-03 6.47643983e-01 -8.49975944e-01 6.33943617e-01 5.88165343e-01 -3.60645533e-01 1.06804144e+00 8.96475911e-01 -9.25185442e-01 -1.40083992e+00 -5.98059773e-01 6.59571886e-01 7.44483173e-02 5.43662727e-01 -2.41011962e-01 -1.21403861e+00 3.67339283e-01 3.71586829e-02 -7.28942513e-01 1.20272958e+00 -6.22420192e-01 -3.63499999e-01 3.35969657e-01 -1.21021748e+00 5.56237340e-01 8.11677694e-01 -3.31481487e-01 -5.82565963e-01 1.98161662e-01 6.28662288e-01 1.68619707e-01 -1.12977982e+00 5.20596504e-01 8.04573357e-01 -9.50859547e-01 1.27035046e+00 -3.46853554e-01 3.77024012e-03 -9.13534760e-01 -3.51395071e-01 -8.71584475e-01 -7.13080823e-01 -4.47264314e-01 7.85099030e-01 9.63239491e-01 -6.64561838e-02 -7.22913563e-01 6.04322374e-01 -4.44902927e-01 -4.21069041e-02 -2.61840850e-01 -9.89296913e-01 -1.06030798e+00 -1.04252003e-01 3.46564651e-01 8.47768903e-01 1.23786533e+00 3.51427019e-01 4.46833692e-05 -4.29655790e-01 1.73777491e-02 7.40331471e-01 3.27682734e-01 4.48025078e-01 -1.16337192e+00 -2.76896030e-01 -4.27554965e-01 -8.19195807e-01 -7.07137823e-01 -4.98983525e-02 -9.25840974e-01 8.25524479e-02 -1.04682410e+00 5.27252257e-01 -3.08699429e-01 -2.43431941e-01 1.07251123e-01 -3.09985012e-01 1.36384323e-01 -2.03456059e-01 5.36535919e-01 9.69840959e-02 2.86971480e-01 8.86909008e-01 5.06581552e-02 9.87841263e-02 -3.02288532e-01 1.55819923e-01 4.56316978e-01 7.36763179e-01 -4.43364143e-01 -4.25178379e-01 6.25749826e-02 -9.41669643e-02 1.69496775e-01 -2.18499362e-01 -7.30545759e-01 5.01696542e-02 -7.05315590e-01 2.33546644e-01 -7.99704015e-01 1.67791769e-01 -6.80249035e-01 7.97071040e-01 1.13702738e+00 -1.12068921e-01 5.26050210e-01 1.97478081e-03 5.06231606e-01 -2.14869156e-01 -7.22939670e-01 7.92030931e-01 -1.83985710e-01 -7.89429128e-01 -2.02519625e-01 -8.93934190e-01 -7.64740884e-01 1.12963426e+00 -7.04435170e-01 -1.63208112e-01 7.43799806e-02 -4.00845945e-01 -2.26348877e-01 6.72644615e-01 9.41103995e-02 7.06267536e-01 -1.24977767e+00 -7.50006199e-01 3.23448509e-01 1.28678441e-01 -5.08010149e-01 3.87604564e-01 5.65779448e-01 -1.13540661e+00 6.21987462e-01 -7.89937019e-01 -9.93804276e-01 -1.93811011e+00 1.02699411e+00 -7.47732818e-02 1.11919999e-01 -4.55385298e-01 5.33565938e-01 2.18053624e-01 -4.63513374e-01 -2.16780931e-01 1.14191636e-01 -3.36551607e-01 -4.87151220e-02 6.07449114e-01 4.53837067e-01 8.09465274e-02 -8.18248510e-01 -5.38191736e-01 1.07354558e+00 -5.49752489e-02 -8.72933492e-02 1.08148992e+00 -2.92339355e-01 -7.26648629e-01 4.56140578e-01 1.42856324e+00 7.79539198e-02 -5.25253654e-01 -5.68030402e-02 5.50146043e-01 -8.52543950e-01 -8.72997284e-01 -1.04637459e-01 -2.58626670e-01 6.81345880e-01 5.87559462e-01 7.46548027e-02 1.21099412e+00 -5.18746376e-01 7.52416432e-01 4.92999554e-01 3.34460765e-01 -7.48956919e-01 -1.17681593e-01 4.07073408e-01 3.72369975e-01 -5.87798715e-01 -1.47316948e-01 -8.24669480e-01 -2.04230353e-01 1.41860151e+00 1.28251165e-01 1.30722642e-01 1.06952101e-01 3.01370144e-01 -5.80354743e-02 1.63174674e-01 -5.27866602e-01 1.06785849e-01 -1.69725493e-01 8.02773356e-01 6.83298111e-01 -1.59598440e-01 -1.04318333e+00 -7.85982236e-02 -4.01373565e-01 -2.73639441e-01 6.01952314e-01 1.16825640e+00 -8.92234504e-01 -1.52325988e+00 -8.25694919e-01 8.27571079e-02 -4.62845534e-01 -1.64923266e-01 -1.83567122e-01 4.87572342e-01 -1.71212465e-01 7.48334408e-01 1.04642712e-01 -4.61283743e-01 -4.69713621e-02 3.53062749e-01 4.69853967e-01 -1.46756798e-01 -2.83533484e-01 2.76900202e-01 -1.83502734e-01 1.51473731e-02 -2.81908661e-01 -8.74036312e-01 -1.53649151e+00 -5.94849825e-01 -1.14474811e-01 7.49937952e-01 8.17190051e-01 5.99431157e-01 2.47511566e-01 1.38214335e-01 9.49486613e-01 -3.27323645e-01 -5.59673905e-01 -7.22610891e-01 -5.40052056e-01 5.53686023e-01 3.04145545e-01 -4.05012399e-01 -6.64896965e-02 1.47987828e-01]

[4.874583721160889, 5.296675682067871]

99d9432d-84c5-4841-9385-ad407805d017

docut5-seq2seq-sql-generation-with-table

2211.06193

https://arxiv.org/abs/2211.06193v1

https://arxiv.org/pdf/2211.06193v1.pdf

DocuT5: Seq2seq SQL Generation with Table Documentation

Current SQL generators based on pre-trained language models struggle to answer complex questions requiring domain context or understanding fine-grained table structure. Humans would deal with these unknowns by reasoning over the documentation of the tables. Based on this hypothesis, we propose DocuT5, which uses off-the-shelf language model architecture and injects knowledge from external `documentation' to improve domain generalization. We perform experiments on the Spider family of datasets that contain complex questions that are cross-domain and multi-table. Specifically, we develop a new text-to-SQL failure taxonomy and find that 19.6% of errors are due to foreign key mistakes, and 49.2% are due to a lack of domain knowledge. We proposed DocuT5, a method that captures knowledge from (1) table structure context of foreign keys and (2) domain knowledge through contextualizing tables and columns. Both types of knowledge improve over state-of-the-art T5 with constrained decoding on Spider, and domain knowledge produces state-of-the-art comparable effectiveness on Spider-DK and Spider-SYN datasets.

['Jeffrey Dalton', 'Iain Mackie', 'Elena Soare']

2022-11-11

['text-to-sql']

['computer-code']

[-2.53400922e-01 3.36161733e-01 -2.88379937e-01 -5.36624134e-01 -1.37160897e+00 -1.01653397e+00 3.28981340e-01 3.63373429e-01 -4.43502590e-02 8.80035520e-01 4.08468992e-01 -6.60181046e-01 7.46558793e-03 -1.15161407e+00 -1.32464850e+00 4.48179007e-01 1.71463639e-01 9.76658821e-01 5.66483974e-01 -6.80049002e-01 1.81092232e-01 -1.85176358e-01 -1.23295474e+00 1.19372964e+00 1.23110604e+00 1.03554630e+00 1.62483603e-02 3.76377076e-01 -7.21738517e-01 1.45119333e+00 -8.16407919e-01 -1.00896466e+00 1.45831406e-01 6.24036565e-02 -1.02527058e+00 -4.34698999e-01 7.12088645e-01 -4.18996841e-01 -2.28122830e-01 6.36102796e-01 2.98202813e-01 -4.31924731e-01 3.94096851e-01 -1.11156356e+00 -8.79431129e-01 1.05780387e+00 -3.06374719e-03 1.76469181e-02 7.05958188e-01 1.77835375e-01 1.22867215e+00 -1.02436328e+00 9.01174247e-01 1.29722023e+00 9.28164780e-01 4.60008800e-01 -1.37316763e+00 -3.90130192e-01 1.24986179e-01 3.13175529e-01 -1.23171055e+00 -2.84787297e-01 2.62331456e-01 -3.39447111e-01 1.70676982e+00 3.06776941e-01 -1.42430902e-01 1.42880082e+00 3.76053065e-01 7.35326529e-01 8.70878577e-01 -4.45859134e-01 1.41940415e-01 6.57076001e-01 2.17819467e-01 6.84866726e-01 5.97501218e-01 -2.29727894e-01 -7.98935413e-01 -2.54234910e-01 2.40785226e-01 -3.34000111e-01 1.44355178e-01 -3.11136425e-01 -9.37490821e-01 7.37198055e-01 1.53913349e-01 -1.14105433e-01 -2.17205599e-01 -1.36331946e-01 5.40681899e-01 4.44439292e-01 -6.02358989e-02 8.46134722e-01 -1.19832420e+00 -3.65456700e-01 -6.76666796e-01 5.69565773e-01 1.58415735e+00 1.60999465e+00 8.07903707e-01 -2.07874537e-01 -7.97309354e-02 7.58008659e-01 1.19621992e-01 7.47177184e-01 4.35478210e-01 -8.31302643e-01 1.28557360e+00 1.25746548e+00 1.60303116e-02 -7.14895248e-01 -2.45440945e-01 -3.22937399e-01 -2.89886057e-01 -2.89773494e-01 5.44461310e-01 1.83740747e-03 -8.64630044e-01 1.64067400e+00 -1.63238987e-01 -5.57277679e-01 3.88048798e-01 2.89986521e-01 1.00392365e+00 4.89760607e-01 -9.60513577e-02 3.29732507e-01 1.47893727e+00 -6.49998546e-01 -6.04462206e-01 -7.07382977e-01 8.31453264e-01 -7.25815117e-01 1.53394544e+00 5.70998728e-01 -1.16683316e+00 -5.21344185e-01 -1.15872908e+00 -4.16248053e-01 -9.03746068e-01 2.02190816e-01 5.45083106e-01 7.19834685e-01 -7.64253557e-01 2.71068215e-01 -7.15748966e-01 -5.14245272e-01 4.70479205e-02 2.98383050e-02 -2.74605453e-01 -2.75999546e-01 -1.26915669e+00 1.14254415e+00 5.35108864e-01 -5.20596266e-01 -8.21868360e-01 -1.02503717e+00 -8.44596446e-01 1.29615799e-01 1.12474847e+00 -7.41618395e-01 1.50430226e+00 1.34887591e-01 -8.44375134e-01 7.36672699e-01 -2.51402706e-01 -4.86770302e-01 3.07574004e-01 -7.50218272e-01 -5.11111081e-01 -1.69176087e-01 5.25460839e-01 2.16807961e-01 8.83910581e-02 -1.19833660e+00 -4.79261756e-01 -4.22570527e-01 1.75239369e-01 -2.79227555e-01 -1.36801735e-01 -2.12584972e-01 -5.91698229e-01 -6.43186927e-01 -5.27258310e-03 -6.62102640e-01 1.86765626e-01 -3.59445423e-01 -6.87666416e-01 -1.12525366e-01 3.83841872e-01 -1.01563442e+00 1.87847519e+00 -1.65458763e+00 -1.81317374e-01 2.86032349e-01 1.39215842e-01 8.21645334e-02 -7.86020160e-02 9.13307786e-01 5.84217757e-02 4.75259990e-01 -1.28355056e-01 1.71418712e-01 4.05092567e-01 2.91635185e-01 -9.19361830e-01 -6.80956721e-01 5.45582712e-01 9.30825472e-01 -5.12046874e-01 -5.24773002e-01 -2.85018265e-01 -3.51510108e-01 -1.02446389e+00 4.10316467e-01 -1.06597722e+00 -4.60215658e-01 -2.58160055e-01 1.08332324e+00 6.74624383e-01 -4.02016878e-01 6.07659996e-01 -1.07568279e-01 3.90219033e-01 1.13147950e+00 -1.36492395e+00 1.68897843e+00 -4.69448149e-01 -2.13352423e-02 -3.06946963e-01 -3.88339728e-01 1.07563114e+00 4.53691855e-02 -2.94307489e-02 -9.59135056e-01 -4.90754128e-01 5.98959565e-01 -2.53715187e-01 -8.64884853e-01 4.36190218e-01 1.64425597e-01 -4.84662592e-01 4.03043121e-01 4.33054745e-01 -1.96034953e-01 5.01436353e-01 6.49583459e-01 1.62444115e+00 2.53236294e-01 3.54710191e-01 -8.56909454e-02 4.68628824e-01 3.61171544e-01 4.52368915e-01 1.06840372e+00 3.84745836e-01 3.86920601e-01 8.13161969e-01 -4.73280966e-01 -8.85418236e-01 -1.55560708e+00 5.76819479e-02 1.23083591e+00 -3.19297850e-01 -1.13516307e+00 -6.35697007e-01 -1.09412527e+00 4.79025871e-01 1.28291476e+00 -4.40027833e-01 -2.11960331e-01 -7.69630194e-01 -4.16644633e-01 9.01990116e-01 9.69483674e-01 4.75548238e-01 -1.01208293e+00 -2.69651443e-01 5.74559867e-01 -5.26626706e-01 -1.34004247e+00 -1.66588083e-01 6.62856877e-01 -7.60058641e-01 -1.33685315e+00 2.42385238e-01 -5.67840457e-01 1.15277074e-01 -3.10503840e-01 2.03086591e+00 -2.45588664e-02 -1.60220191e-01 2.83882558e-01 -3.29846174e-01 -4.98341203e-01 -5.94797015e-01 5.05090952e-01 -1.64706275e-01 -8.14751029e-01 9.69755828e-01 -3.54486465e-01 1.33782163e-01 5.65002024e-01 -9.90075648e-01 -2.84789413e-01 7.29549646e-01 8.39361191e-01 3.73588860e-01 6.11556321e-02 5.50143421e-01 -1.31629705e+00 7.05974638e-01 -6.91526413e-01 -5.19035161e-01 9.61403906e-01 -7.81283319e-01 7.22273767e-01 7.24287510e-01 -1.04554385e-01 -1.14632142e+00 -3.67673427e-01 -1.21464968e-01 7.43081868e-02 -6.19301833e-02 8.04956675e-01 -5.42611182e-01 4.15497035e-01 1.22845948e+00 2.48551637e-01 -4.01379466e-01 -6.94568574e-01 4.13011730e-01 4.83160228e-01 7.46471584e-01 -1.47488213e+00 7.02712595e-01 6.89751795e-03 -4.53178793e-01 -1.27442658e-01 -8.55496764e-01 -1.32450894e-01 -4.22790736e-01 4.34455186e-01 4.48309511e-01 -9.09399629e-01 -6.13902569e-01 1.71718806e-01 -1.16213310e+00 -1.94086850e-01 -1.68582559e-01 -5.79160405e-03 -5.82664609e-01 1.75746500e-01 -7.59061396e-01 -4.82566297e-01 -2.29094267e-01 -9.96143699e-01 9.85635817e-01 -2.90663779e-01 -5.51126361e-01 -8.19564044e-01 3.23657431e-02 4.16345984e-01 4.06959683e-01 -1.21224783e-01 1.66372395e+00 -9.82917011e-01 -1.03684783e+00 3.39957699e-02 -2.60082960e-01 4.07053798e-01 -5.05131260e-02 -3.01342547e-01 -6.30736768e-01 1.01106755e-01 7.81005397e-02 -7.94318080e-01 5.50045788e-01 -3.42896521e-01 1.01137805e+00 -6.25638664e-01 -3.32227439e-01 5.11500776e-01 1.54269028e+00 2.15568990e-01 7.15236783e-01 6.09879434e-01 4.54594642e-01 6.58753812e-01 6.51479065e-01 5.38299143e-01 9.93095994e-01 6.11454487e-01 1.64556116e-01 5.07545710e-01 -5.94832227e-02 -9.61404681e-01 3.68869662e-01 6.78270578e-01 6.79666400e-01 -3.09083462e-01 -1.46353519e+00 6.94293797e-01 -1.65968657e+00 -6.15092933e-01 -9.35591683e-02 1.89246321e+00 1.51271343e+00 6.84891224e-01 -8.97424221e-02 -1.08289562e-01 1.40864491e-01 -3.95354331e-01 -6.45771682e-01 -4.55168068e-01 -3.50856185e-01 4.16664183e-01 4.16627645e-01 3.71842861e-01 -7.46969938e-01 1.14129519e+00 6.27095985e+00 7.47995973e-01 -7.30017126e-01 -2.64933288e-01 1.93410829e-01 -4.26825928e-03 -6.60256267e-01 1.21920116e-01 -1.40471697e+00 4.07319129e-01 1.12179565e+00 -1.46557748e-01 5.20255327e-01 1.13496935e+00 -6.27283871e-01 -1.06549457e-01 -1.54340899e+00 5.95070183e-01 1.17770396e-01 -1.32899594e+00 3.62255424e-01 -2.96218961e-01 4.36847538e-01 -2.14828581e-01 -9.87103358e-02 1.00110877e+00 6.68734133e-01 -1.01361179e+00 7.87550151e-01 5.69046617e-01 7.86027491e-01 -4.50196147e-01 6.70315325e-01 4.83504802e-01 -8.93555641e-01 -2.31437296e-01 -3.60909224e-01 1.83057249e-01 -1.86467350e-01 4.67286497e-01 -1.36108184e+00 5.00488341e-01 1.02320158e+00 3.38309169e-01 -1.17039526e+00 3.53864998e-01 -4.00636584e-01 5.85186064e-01 -4.57590640e-01 -8.33652020e-02 -3.27575319e-02 4.43346858e-01 1.03369035e-01 1.33761609e+00 2.74312139e-01 -4.55798991e-02 8.58421251e-02 1.17943597e+00 -2.48398483e-01 -1.88649490e-01 -7.15531468e-01 -1.59888715e-01 8.09388995e-01 5.66245973e-01 -4.20544073e-02 -6.59708619e-01 -6.58616543e-01 4.49587345e-01 6.13919616e-01 3.05352241e-01 -6.85767770e-01 -5.53639531e-01 5.97182751e-01 4.83385921e-01 5.59527636e-01 -3.32011245e-02 -9.54626918e-01 -1.46193099e+00 8.46837223e-01 -1.67393601e+00 7.08969951e-01 -9.36310172e-01 -1.42182565e+00 6.08442187e-01 1.24449678e-01 -8.53190243e-01 -6.63019419e-01 -8.69370699e-01 -3.96267287e-02 8.51124108e-01 -1.28857756e+00 -1.00947928e+00 -1.65430099e-01 5.77022791e-01 3.48955095e-01 -4.28179115e-01 1.03759432e+00 3.87149036e-01 -6.53604120e-02 1.05046487e+00 -1.58269808e-01 2.87837863e-01 1.07174575e+00 -1.59670758e+00 9.58835125e-01 8.29142392e-01 -1.69709340e-01 1.36074746e+00 4.95938480e-01 -1.17872167e+00 -1.69597864e+00 -1.01721370e+00 1.42978740e+00 -1.43477285e+00 7.69034266e-01 -7.69731164e-01 -1.23116267e+00 1.07831907e+00 1.19238675e-01 -3.68897021e-01 6.80111408e-01 4.07549709e-01 -1.08679843e+00 -3.68129700e-01 -1.26962709e+00 4.26194221e-01 1.11149800e+00 -9.05198514e-01 -1.21527123e+00 2.26663694e-01 9.26273227e-01 -5.93513370e-01 -1.09645748e+00 6.53213561e-01 4.75916713e-01 -8.82950306e-01 1.00874078e+00 -1.05644011e+00 8.31037402e-01 -2.45386422e-01 -7.89509773e-01 -1.06044972e+00 -1.78686325e-02 -4.09835488e-01 -4.95719403e-01 1.32852519e+00 9.36464429e-01 -3.90489012e-01 9.23818707e-01 7.95857549e-01 -2.18733326e-01 -6.03716671e-01 -5.14193296e-01 -8.98883224e-01 1.54047385e-01 -6.02410257e-01 8.57531905e-01 8.25706840e-01 2.86354363e-01 5.07666647e-01 -9.62571129e-02 8.40526000e-02 3.45469117e-01 -2.81231422e-02 9.90584493e-01 -9.92838740e-01 -4.75444913e-01 2.42049638e-02 2.57265326e-02 -1.20929396e+00 -1.39500096e-01 -8.61111224e-01 -8.45433474e-02 -1.49137127e+00 3.04141734e-02 -3.73739541e-01 2.91857962e-02 7.67611742e-01 -2.42470875e-01 -4.92561579e-01 2.09423751e-01 -1.35788128e-01 -7.42460787e-01 2.14460671e-01 6.75864875e-01 -2.11125553e-01 9.61437151e-02 -1.91192925e-01 -1.18097496e+00 3.39026153e-01 4.52628583e-01 -5.40219307e-01 -5.37290394e-01 -8.39385927e-01 8.06744039e-01 4.03196245e-01 4.57995161e-02 -1.18903112e+00 3.32313448e-01 -2.97891200e-02 3.91547561e-01 -8.69432449e-01 1.13185734e-01 -8.27517807e-01 -2.06649169e-01 3.28160852e-01 -5.60446203e-01 5.27287304e-01 6.16960764e-01 2.47408018e-01 -3.69919538e-01 -1.16712868e-01 2.46187016e-01 -4.94193852e-01 -7.19626963e-01 -3.22713733e-01 -2.41747737e-01 9.49720621e-01 5.26381671e-01 1.89977050e-01 -9.93148685e-01 -1.92046270e-01 -4.95531172e-01 3.92039239e-01 3.06113333e-01 7.15616822e-01 5.79192340e-01 -1.15807354e+00 -6.79173291e-01 4.93627995e-01 5.99399328e-01 -9.77589563e-02 -9.08827409e-02 1.15857266e-01 -6.09212637e-01 9.47224557e-01 -1.71944290e-01 -4.94524807e-01 -7.26606548e-01 7.01486349e-01 1.84912041e-01 -7.16850638e-01 -4.62908670e-02 9.55134809e-01 -5.93510270e-02 -1.11933315e+00 4.82422322e-01 -9.10138190e-01 3.18619609e-01 -1.84956819e-01 3.72745395e-01 1.93057686e-01 6.97747409e-01 3.66264492e-01 -6.07056320e-01 2.08261162e-01 -5.68187177e-01 -4.20125090e-02 1.09014332e+00 2.26599529e-01 -3.13739479e-02 2.80693203e-01 8.78064334e-01 2.18857631e-01 -6.04824483e-01 -7.05362082e-01 6.45130992e-01 -3.15766066e-01 -6.53775215e-01 -1.97227037e+00 -4.55804825e-01 9.13759351e-01 4.98175099e-02 7.32130324e-03 9.60281789e-01 1.60203758e-03 8.96467626e-01 1.06970489e+00 7.08859861e-01 -1.08580315e+00 2.91730434e-01 9.45630789e-01 9.63182032e-01 -1.16335988e+00 -8.13978463e-02 -4.99544263e-01 -8.41858745e-01 1.05522871e+00 1.41477454e+00 1.82846650e-01 3.98554444e-01 7.37640917e-01 2.61607319e-01 -1.78167403e-01 -1.51424158e+00 1.19503863e-01 1.30437925e-01 5.90692043e-01 5.96668303e-01 -1.59343570e-01 8.04582015e-02 1.29804611e+00 -5.89518845e-01 1.89783469e-01 4.61903840e-01 1.37389612e+00 -2.67375797e-01 -1.37790990e+00 -3.47010434e-01 6.66387200e-01 -4.51822996e-01 -5.12843609e-01 -6.30353153e-01 8.88455391e-01 1.33295193e-01 9.23314929e-01 -3.09047341e-01 -6.80199564e-01 8.67025793e-01 4.85509336e-01 4.23478305e-01 -9.41101611e-01 -8.66938889e-01 -4.24199492e-01 5.32359719e-01 -7.82797754e-01 7.01685607e-01 -5.18117726e-01 -1.09625411e+00 -5.17890990e-01 1.50330260e-01 1.43220276e-01 4.79914308e-01 6.00146353e-01 7.11115241e-01 5.31524062e-01 1.67358853e-02 4.45317775e-01 -1.26236570e+00 -1.01405156e+00 -1.94181919e-01 4.66472536e-01 3.37631479e-02 -5.17034054e-01 4.43329662e-02 2.32563224e-02]

[9.912863731384277, 7.84554386138916]

46b20ca8-e3f2-4510-b583-2004db1dc816

patchwork-learning-a-paradigm-towards

2305.06217

https://arxiv.org/abs/2305.06217v2

https://arxiv.org/pdf/2305.06217v2.pdf

Patchwork Learning: A Paradigm Towards Integrative Analysis across Diverse Biomedical Data Sources

Machine learning (ML) in healthcare presents numerous opportunities for enhancing patient care, population health, and healthcare providers' workflows. However, the real-world clinical and cost benefits remain limited due to challenges in data privacy, heterogeneous data sources, and the inability to fully leverage multiple data modalities. In this perspective paper, we introduce "patchwork learning" (PL), a novel paradigm that addresses these limitations by integrating information from disparate datasets composed of different data modalities (e.g., clinical free-text, medical images, omics) and distributed across separate and secure sites. PL allows the simultaneous utilization of complementary data sources while preserving data privacy, enabling the development of more holistic and generalizable ML models. We present the concept of patchwork learning and its current implementations in healthcare, exploring the potential opportunities and applicable data sources for addressing various healthcare challenges. PL leverages bridging modalities or overlapping feature spaces across sites to facilitate information sharing and impute missing data, thereby addressing related prediction tasks. We discuss the challenges associated with PL, many of which are shared by federated and multimodal learning, and provide recommendations for future research in this field. By offering a more comprehensive approach to healthcare data integration, patchwork learning has the potential to revolutionize the clinical applicability of ML models. This paradigm promises to strike a balance between personalization and generalizability, ultimately enhancing patient experiences, improving population health, and optimizing healthcare providers' workflows.

['Yong Chen', 'Fei Wang', 'Jiayu Zhou', 'Mert R. Sabuncu', 'Weishen Pan', 'Suraj Rajendran']

2023-05-10

['data-integration']

['knowledge-base']

[ 1.61751315e-01 1.12666413e-01 -7.38310635e-01 -4.59602058e-01 -1.15497017e+00 -5.53428233e-01 1.19005233e-01 9.02383029e-01 -2.82114357e-01 6.01579666e-01 5.81801832e-01 -4.07795787e-01 -4.58594292e-01 -4.99407858e-01 -3.10849965e-01 -6.25646770e-01 -3.81821468e-02 2.50446141e-01 -7.69364715e-01 2.49672219e-01 -3.19398820e-01 5.45361698e-01 -9.65623081e-01 8.73407483e-01 6.00161910e-01 9.53871131e-01 -3.38284850e-01 2.48938695e-01 -1.99761525e-01 6.63924932e-01 -2.09562838e-01 -4.83989984e-01 4.05527741e-01 -2.25908518e-01 -5.36962926e-01 -1.65486068e-01 2.98339784e-01 -3.89444619e-01 -4.74888384e-02 6.60938919e-01 8.11851919e-01 -4.08825397e-01 3.36667672e-02 -1.38611495e+00 -6.87060952e-01 2.34029964e-01 -2.32426763e-01 -3.17504406e-01 3.88514072e-01 2.93319583e-01 6.43159568e-01 -5.93864262e-01 9.30572808e-01 8.00232351e-01 1.05411601e+00 8.37940633e-01 -1.32392955e+00 -5.58770597e-01 -1.82048649e-01 -1.17573105e-01 -1.18222582e+00 -7.17615902e-01 1.78655237e-01 -5.15630305e-01 7.77379274e-01 7.08775997e-01 6.36119068e-01 1.08928347e+00 3.96995902e-01 7.90486038e-01 9.12722588e-01 -1.83023900e-01 1.92063838e-01 4.68945235e-01 1.09332213e-02 6.34273350e-01 4.53179806e-01 4.23473157e-02 -8.16489339e-01 -9.55893517e-01 2.34689206e-01 9.05988336e-01 -4.30853397e-01 -6.57364070e-01 -1.46316266e+00 4.82467920e-01 2.06203803e-01 5.91165088e-02 -4.97304112e-01 -4.43614602e-01 6.05649292e-01 1.87015995e-01 1.38945773e-01 3.74036431e-01 -8.06774318e-01 6.95670247e-02 -7.03400135e-01 8.40058476e-02 9.49863732e-01 8.70842695e-01 5.66579878e-01 -5.10307372e-01 -1.22833334e-01 4.30802852e-01 2.88713247e-01 5.40776432e-01 3.31798047e-01 -9.76569891e-01 4.09461558e-01 8.40040267e-01 7.89220333e-02 -7.74463654e-01 -6.69208050e-01 -3.46949339e-01 -1.01718104e+00 -7.32025504e-02 2.35778779e-01 -3.69801491e-01 -4.90383595e-01 1.78551149e+00 6.43936694e-01 -1.18679270e-01 4.89601374e-01 6.36352241e-01 9.06609654e-01 8.58112201e-02 5.21005690e-01 -2.63916910e-01 1.46439385e+00 -4.52225953e-01 -1.00129735e+00 1.40247568e-01 8.84802818e-01 -4.87599462e-01 6.89256430e-01 3.12577277e-01 -9.49725807e-01 3.01652011e-02 -6.33976638e-01 -1.48196176e-01 -5.64661741e-01 -3.28754395e-01 5.39454877e-01 7.87504196e-01 -7.44153798e-01 4.31008607e-01 -1.19835567e+00 -5.25305510e-01 9.72239375e-01 4.29228038e-01 -9.36223984e-01 -4.51100975e-01 -8.61443222e-01 6.40423834e-01 -2.75748428e-02 -2.31175438e-01 -4.59274352e-01 -1.36050129e+00 -9.19384897e-01 -3.61762824e-03 1.86443925e-01 -1.44189823e+00 7.49121368e-01 -3.76990765e-01 -7.17313349e-01 7.89579809e-01 -4.14979570e-02 -3.15347731e-01 6.08197093e-01 -2.29006231e-01 -5.70109427e-01 -3.84650677e-02 1.01205409e-01 3.13719898e-01 1.79639935e-01 -1.02156711e+00 -6.75992250e-01 -8.53959322e-01 -7.54229605e-01 9.59028602e-02 -5.74036002e-01 -1.66120857e-01 -1.23358279e-01 -3.39913130e-01 -1.82016060e-01 -7.20682442e-01 -4.99452174e-01 3.72561216e-01 -4.28868264e-01 4.10608679e-01 8.29524875e-01 -7.61282027e-01 1.06652772e+00 -2.37224722e+00 -7.59929717e-02 2.37410784e-01 7.08103776e-01 3.48617047e-01 4.03890759e-03 7.63575852e-01 2.17259675e-01 3.72546762e-01 -1.31284639e-01 -4.53096867e-01 -2.84802496e-01 1.34185493e-01 3.60678136e-02 4.91729409e-01 -3.27285789e-02 1.15874302e+00 -7.51844168e-01 -3.45816493e-01 2.32596949e-01 8.33936453e-01 -5.45551717e-01 2.28385821e-01 1.56449452e-02 9.10505593e-01 -4.21336532e-01 1.15884054e+00 5.77681899e-01 -5.56616783e-01 6.88075185e-01 -2.96084076e-01 1.69957448e-02 -2.05905810e-01 -9.01291609e-01 1.76752341e+00 -1.89094231e-01 -4.51807603e-02 4.55638647e-01 -4.00727630e-01 6.84744179e-01 5.20228744e-01 1.17149341e+00 -4.86326337e-01 -5.01134694e-02 1.91454530e-01 -3.10166061e-01 -9.39478576e-01 -2.13584732e-02 -3.03904265e-01 -3.04279625e-02 4.92192477e-01 -2.87489831e-01 5.54827690e-01 -6.58171356e-01 1.22330794e-02 1.22971582e+00 -1.91296145e-01 6.09775245e-01 1.22261763e-01 1.00391172e-01 5.77816740e-02 7.69043684e-01 6.85127079e-01 -5.35465300e-01 4.20216501e-01 1.95969075e-01 -6.20943606e-01 -6.84839487e-01 -1.25910950e+00 -4.54291850e-01 8.29533041e-01 -1.78038359e-01 -6.62250340e-01 -1.45181090e-01 -7.50550330e-01 7.00207889e-01 2.10905790e-01 -5.32477319e-01 -1.45963043e-01 -2.67827630e-01 -9.11926508e-01 6.15981460e-01 4.12092805e-01 1.95642784e-02 -6.80201650e-01 -7.41754532e-01 1.97001636e-01 -4.00438398e-01 -8.70801449e-01 -4.45399940e-01 -8.27875733e-02 -8.58021855e-01 -1.40922129e+00 -3.73377651e-01 -3.17786455e-01 5.30759156e-01 1.30328506e-01 8.35865736e-01 -1.21535830e-01 -8.48393857e-01 7.36774087e-01 -4.74281572e-02 -7.23207355e-01 -4.43239778e-01 4.44560172e-03 6.95449859e-02 6.96065575e-02 4.51375425e-01 -1.76110983e-01 -9.59442019e-01 -1.41103342e-01 -1.08437884e+00 -2.31836528e-01 6.08546376e-01 9.37802017e-01 8.23337197e-01 -5.61985135e-01 9.38004017e-01 -1.45791757e+00 5.36641777e-01 -1.04462981e+00 -3.77459116e-02 6.34004831e-01 -1.24818349e+00 -2.41658881e-01 4.21810538e-01 -2.36426461e-02 -9.23652828e-01 5.80031201e-02 -8.83982331e-03 -3.23957026e-01 -2.19835848e-01 7.23173082e-01 -2.66751260e-01 -7.56284222e-02 5.46504378e-01 -9.29515064e-02 8.36441278e-01 -7.26233959e-01 3.52150887e-01 9.02325153e-01 3.55821729e-01 -3.35247159e-01 1.11316614e-01 6.59339130e-01 6.00960664e-02 -2.38899171e-01 -5.52267730e-01 -6.00175500e-01 -2.98302323e-01 3.00867587e-01 7.84187675e-01 -1.06534517e+00 -9.80060935e-01 1.17403753e-01 -5.01297176e-01 7.42716789e-02 -5.56461215e-01 4.95337367e-01 -1.86398655e-01 3.53099138e-01 -6.14888251e-01 -6.04182899e-01 -8.18756640e-01 -9.76608753e-01 8.63386452e-01 -2.80978531e-03 -4.59235102e-01 -9.04728651e-01 1.80313185e-01 8.19081247e-01 6.62471473e-01 7.19176412e-01 1.09126103e+00 -1.01969123e+00 -5.60722888e-01 -4.98522311e-01 -4.90558259e-02 1.40713483e-01 7.36375809e-01 -3.56376827e-01 -8.66427779e-01 -6.78774774e-01 6.20232150e-02 -3.08313042e-01 2.97972351e-01 1.99419722e-01 1.13333452e+00 -5.23933351e-01 -6.59356773e-01 8.77483070e-01 1.42644191e+00 8.37055072e-02 1.69823065e-01 6.04320988e-02 6.41522050e-01 6.94873393e-01 5.19187510e-01 9.02297199e-01 7.36998796e-01 1.86581194e-01 3.95490944e-01 -5.93283653e-01 7.98693448e-02 -2.30241194e-01 -2.10166335e-01 5.28990805e-01 4.48348314e-01 2.16178030e-01 -1.14136386e+00 5.37002325e-01 -1.93646646e+00 -7.72006631e-01 1.71241447e-01 2.44660759e+00 9.52163637e-01 -6.62166059e-01 1.17769446e-02 -4.77101833e-01 3.04842085e-01 -2.91138887e-01 -1.03393292e+00 -1.67038471e-01 -1.60187498e-01 1.17175937e-01 4.81515646e-01 1.24710135e-01 -9.20211434e-01 1.80074587e-01 6.61369848e+00 -4.27743010e-02 -1.17399764e+00 3.54860872e-01 8.09783161e-01 -5.33252656e-01 -5.44719517e-01 -2.55536735e-01 -5.23922920e-01 2.02513620e-01 8.78830552e-01 -2.88003296e-01 2.31955007e-01 7.66227603e-01 1.72343269e-01 3.31637412e-01 -1.33970225e+00 9.37071621e-01 -1.97466686e-02 -1.81429684e+00 -1.18323855e-01 4.23865765e-01 6.50493741e-01 2.99462140e-01 3.71958852e-01 -1.06016085e-01 4.67375778e-02 -1.05348730e+00 2.05394160e-02 7.81450868e-01 1.07065821e+00 -6.11660302e-01 8.23219419e-01 1.88491076e-01 -6.37566984e-01 -3.42521906e-01 2.72304922e-01 4.19006407e-01 1.39660150e-01 5.77063382e-01 -1.06794584e+00 9.61952507e-01 7.77500093e-01 6.61075056e-01 -2.50455946e-01 9.25505340e-01 8.31830382e-01 -2.54173167e-02 -1.42115533e-01 5.93973339e-01 -3.21069866e-01 7.64955708e-04 1.31151378e-01 1.29626143e+00 2.33551443e-01 4.32083234e-02 4.90955293e-01 4.56017584e-01 -1.80142358e-01 5.42315423e-01 -8.68921220e-01 -1.32847369e-01 7.42600560e-01 1.21302199e+00 1.49865642e-01 -2.00384587e-01 -7.12693870e-01 4.35284913e-01 1.24519430e-01 1.61870196e-01 -2.96659589e-01 -8.84087663e-03 1.20620394e+00 2.67905027e-01 -3.09711009e-01 2.30504781e-01 -5.81399262e-01 -1.26846957e+00 -1.68468073e-01 -1.35700178e+00 1.01975346e+00 -7.87765905e-02 -1.69157267e+00 2.56716877e-01 -3.36500585e-01 -1.16194105e+00 -1.24380119e-01 -2.43411183e-01 1.70392543e-01 9.13846433e-01 -1.52461743e+00 -1.54345047e+00 -1.73241407e-01 9.19381201e-01 -3.33403438e-01 -3.35871041e-01 1.39899004e+00 5.37493169e-01 -6.26864016e-01 9.35138643e-01 4.09159213e-01 -2.07535699e-01 1.15538096e+00 -7.64519215e-01 -2.27199271e-01 2.93046534e-01 -2.66122937e-01 1.12322974e+00 -4.71590422e-02 -7.19645560e-01 -1.90688658e+00 -1.37428284e+00 8.73611867e-01 -7.95241177e-01 8.73104930e-02 -2.18793899e-01 -9.02111530e-01 9.18991566e-01 -5.60906418e-02 2.26335704e-01 2.02834105e+00 3.81795973e-01 -5.89893937e-01 -4.69317943e-01 -1.92422938e+00 4.43260491e-01 6.51636958e-01 -7.44876087e-01 -7.57509544e-02 3.76399070e-01 5.93725085e-01 -3.21507901e-01 -1.64254773e+00 5.62832236e-01 8.16650271e-01 -8.19084048e-01 1.03763986e+00 -1.16078269e+00 1.65237114e-02 -6.48941919e-02 -3.16589892e-01 -7.02241361e-01 -1.78348929e-01 -7.16348112e-01 -2.17125624e-01 1.11795962e+00 3.48143995e-01 -1.04091048e+00 8.21632266e-01 1.45284390e+00 1.39627397e-01 -9.42208886e-01 -8.66108835e-01 -1.72623768e-01 -9.15537104e-02 -2.68219888e-01 1.03728378e+00 1.43103540e+00 4.85007316e-01 -3.65651667e-01 -4.80546266e-01 2.87318081e-01 7.56846130e-01 1.74159199e-01 7.72954166e-01 -1.05723047e+00 -3.39387476e-01 2.39619315e-02 -1.56845346e-01 -7.66036734e-02 -3.11840802e-01 -1.20729899e+00 -5.32498121e-01 -1.46296608e+00 5.19803345e-01 -9.34621394e-01 -7.30449319e-01 1.01412272e+00 -1.39320269e-01 9.00633410e-02 2.89924294e-01 5.43401778e-01 -4.88398999e-01 -1.93520244e-02 1.12955403e+00 -3.91045325e-02 -2.92033136e-01 -4.77745496e-02 -1.21175718e+00 1.40973300e-01 7.45201468e-01 -3.57231528e-01 -3.53874773e-01 -4.32050079e-01 -1.32028654e-01 3.85639608e-01 3.28726172e-01 -5.75898647e-01 4.86322850e-01 -4.24818844e-01 4.99844670e-01 -1.90308541e-01 2.02371731e-01 -1.15089953e+00 8.47619236e-01 8.48365366e-01 -3.65095288e-01 -2.43346989e-02 2.23605722e-01 7.30037034e-01 -5.37485965e-02 5.02816617e-01 5.47888279e-01 -2.78947890e-01 -9.97524187e-02 5.14129639e-01 7.28824362e-02 -1.34448066e-01 1.25221467e+00 1.78288773e-01 -5.16630948e-01 -9.57366079e-04 -1.13085520e+00 3.85651678e-01 5.96620560e-01 3.82273644e-01 5.95639646e-01 -1.11290598e+00 -6.61529243e-01 6.91365778e-01 5.44169903e-01 1.01495255e-02 7.07994819e-01 1.03858960e+00 -2.23486930e-01 3.53976727e-01 -2.14840084e-01 -5.55291176e-01 -1.35669923e+00 8.06100190e-01 2.32125938e-01 -1.00141510e-01 -6.89587533e-01 2.86201507e-01 -1.38929281e-02 -7.85823643e-01 3.76867533e-01 1.39938101e-01 4.73659635e-01 -1.05371460e-01 7.64848411e-01 3.48694414e-01 3.36349219e-01 -1.96439818e-01 -7.00782418e-01 -4.97571826e-02 -2.98318356e-01 3.45833868e-01 1.41845775e+00 -3.29165280e-01 -3.81048828e-01 2.52917647e-01 1.18801653e+00 2.16913193e-01 -9.81495500e-01 -3.12939703e-01 -1.38423881e-02 -5.48469543e-01 -3.39027822e-01 -1.40337837e+00 -1.05513835e+00 4.93647695e-01 8.92974973e-01 -1.43305898e-01 1.24566317e+00 1.61228869e-02 1.00330877e+00 1.12912403e-02 3.97283047e-01 -5.58880210e-01 -4.25782204e-01 -3.86343718e-01 3.67800504e-01 -1.43521810e+00 7.36597404e-02 -1.52435854e-01 -8.15227270e-01 9.23102438e-01 2.59393871e-01 8.35255682e-01 8.74648511e-01 5.30668557e-01 6.26807213e-01 -2.11106658e-01 -9.09114540e-01 3.58633906e-01 -4.57186289e-02 8.31057727e-01 4.88258600e-01 3.77668589e-01 -1.47049502e-01 7.68564939e-01 3.84529501e-01 4.36515361e-01 9.33253169e-02 1.23062336e+00 2.52289534e-01 -1.71831822e+00 -4.01425213e-01 8.21647584e-01 -7.96490252e-01 -6.37429580e-02 -1.49907380e-01 4.50355917e-01 4.80388820e-01 8.34557652e-01 -3.32274437e-01 -2.18115970e-01 4.08305645e-01 3.44343036e-01 8.68821815e-02 -5.44401467e-01 -9.42988098e-01 3.61598469e-02 -4.40927707e-02 -7.90731549e-01 -8.80349725e-02 -1.03559780e+00 -1.04149127e+00 -3.69986057e-01 2.35160381e-01 -1.41288698e-01 9.19831276e-01 6.07392311e-01 1.45966256e+00 2.43910924e-01 5.05824804e-01 4.62519489e-02 -7.84133911e-01 -1.34070903e-01 -3.62194240e-01 6.17343485e-01 7.89973795e-01 9.56737399e-02 1.20711856e-01 5.39558157e-02]

[6.192500591278076, 6.502936840057373]

b751f020-2d4a-4d3d-ba5d-d6ad6766ffb8

a-robust-iris-authentication-system-on-gpu

1912.00756

https://arxiv.org/abs/1912.00756v2

https://arxiv.org/pdf/1912.00756v2.pdf

Learning scale-variant features for robust iris authentication with deep learning based ensemble framework

In recent years, mobile Internet has accelerated the proliferation of smart mobile development. The mobile payment, mobile security and privacy protection have become the focus of widespread attention. Iris recognition becomes a high-security authentication technology in these fields, it is widely used in distinct science fields in biometric authentication fields. The Convolutional Neural Network (CNN) is one of the mainstream deep learning approaches for image recognition, whereas its anti-noise ability is weak and needs a certain amount of memory to train in image classification tasks. Under these conditions we put forward a fine-tuning neural network model based on the Mask R-CNN and Inception V4 neural network model, which integrates every component in an overall system that combines the iris detection, extraction, and recognition function as an iris recognition system. The proposed framework has the characteristics of scalability and high availability; it not only can learn part-whole relationships of the iris image but also enhancing the robustness of the whole framework. Importantly, the proposed model can be trained using the different spectrum of samples, such as Visible Wavelength (VW) and Near Infrared (NIR) iris biometric databases. The recognition average accuracy of 99.10% is achieved while executing in the mobile edge calculation device of the Jetson Nano.

['Nurul Amelina Nasharuddin', 'Rahmita Wirza O. K. Rahmat', 'Fatimah Khalid', 'Siming Zheng']

2019-12-02

['mobile-security']

['miscellaneous']

[ 2.24393964e-01 -5.22518277e-01 -2.23364934e-01 -1.61132038e-01 1.60153195e-01 -1.08770445e-01 1.79271758e-01 -2.90441632e-01 -5.56604147e-01 2.16176212e-01 -1.41197518e-01 -6.25444353e-01 -3.84623557e-01 -7.35938728e-01 -2.97538221e-01 -9.61423337e-01 3.20224315e-01 -1.49172783e-01 -2.34675273e-01 -2.10811213e-01 3.03475618e-01 7.33941376e-01 -1.74562609e+00 1.37614995e-01 9.59843934e-01 1.33062160e+00 -4.99188453e-01 5.93752146e-01 1.11907721e-01 6.83967352e-01 -2.86948413e-01 -4.48875844e-01 2.12112710e-01 -1.81699932e-01 -5.15724003e-01 -2.92160660e-01 1.45433545e-01 -2.87829310e-01 -3.08408439e-01 1.08023930e+00 9.29517090e-01 -1.99978948e-01 3.46684247e-01 -6.62964225e-01 -9.16082561e-01 1.46975651e-01 -1.00204074e+00 4.34408009e-01 1.61167547e-01 3.25019568e-01 1.64828241e-01 -5.66815257e-01 -1.10008858e-01 6.68703616e-01 8.73413682e-01 6.84797406e-01 -6.18871033e-01 -7.44865894e-01 -5.04366815e-01 2.82995403e-01 -1.33418429e+00 -3.79019082e-01 6.47140265e-01 -3.42437327e-01 1.04512167e+00 3.11049521e-01 8.11660111e-01 7.64010251e-01 4.27337617e-01 3.78760010e-01 1.15687299e+00 -4.12349671e-01 -3.53252351e-01 1.93203330e-01 1.98981017e-01 7.35945225e-01 4.26934272e-01 4.43066835e-01 -2.08169073e-01 1.40200034e-01 9.16976094e-01 4.41380411e-01 -2.38492921e-01 3.47809255e-01 -8.17992449e-01 2.28138000e-01 5.69518983e-01 4.00669783e-01 -3.20360392e-01 -2.43313700e-01 2.82179922e-01 2.06875980e-01 2.31710970e-01 9.64263976e-02 -4.67138350e-01 -1.83264375e-01 -5.24746537e-01 -2.95324147e-01 4.60535109e-01 3.05955231e-01 3.79488140e-01 2.14827046e-01 -1.02503514e-02 6.40663981e-01 3.75207335e-01 5.69505155e-01 6.86759949e-01 1.59727428e-02 1.17625698e-01 1.10306776e+00 -4.23976853e-02 -1.01389658e+00 -7.49990165e-01 -8.45052183e-01 -1.34509814e+00 2.48941362e-01 3.31160009e-01 -3.85764331e-01 -1.01915109e+00 1.09567797e+00 5.30519962e-01 5.15017688e-01 2.31666416e-02 1.04102314e+00 1.24406552e+00 3.71443719e-01 -1.02798663e-01 -3.40807736e-02 1.45527875e+00 -6.41749918e-01 -5.50815046e-01 1.84322610e-01 3.01605582e-01 -1.01233578e+00 7.31807828e-01 4.04116541e-01 -8.79793167e-01 -9.47622240e-01 -1.16701758e+00 -2.28276163e-01 -5.52257657e-01 5.35274625e-01 8.68452847e-01 1.29455781e+00 -7.81275332e-01 3.51483524e-01 -5.52722454e-01 -2.81034738e-01 5.53231001e-01 1.06865788e+00 -3.67358476e-01 1.61717907e-01 -9.18968201e-01 5.91557264e-01 1.79140463e-01 6.68272674e-01 7.42524266e-02 -2.67425597e-01 -5.13996243e-01 1.96640909e-01 -6.53806701e-02 -5.83771765e-01 7.45903194e-01 -1.26102221e+00 -1.86776495e+00 1.07535303e+00 1.20965168e-01 -1.75342768e-01 4.46088389e-02 6.02976279e-03 -9.51666296e-01 -2.51974225e-01 -6.30983889e-01 5.79485903e-03 7.14187562e-01 -3.42605799e-01 -5.75695455e-01 -7.56918609e-01 -1.80443272e-01 1.09866917e-01 -3.91905993e-01 4.58668381e-01 -2.26451665e-01 -4.24380034e-01 1.47255450e-01 -6.75901353e-01 1.43413261e-01 -3.57944608e-01 -2.42326945e-01 -1.54180467e-01 7.74127722e-01 -7.13788450e-01 1.22432780e+00 -2.21656752e+00 -3.43282640e-01 4.58673090e-01 5.76318987e-02 1.01074052e+00 8.50424170e-02 -9.23001841e-02 -3.68788511e-01 -7.42614195e-02 1.50390506e-01 1.75498098e-01 -5.20475805e-01 -3.23155582e-01 1.84808180e-01 7.45931983e-01 6.19737571e-03 1.08959460e+00 -3.96673381e-01 -4.99356873e-02 4.36731219e-01 8.17251861e-01 -1.51039198e-01 3.98583636e-02 4.45200652e-01 6.57111466e-01 -2.55890876e-01 1.01859617e+00 1.00788999e+00 -3.80384594e-01 -2.91798174e-01 -3.44215512e-01 -1.58755451e-01 -2.83504426e-01 -1.30152488e+00 1.29858565e+00 -1.06630079e-01 3.32101077e-01 4.72486652e-02 -9.77333426e-01 1.04979289e+00 4.38139409e-01 4.53212380e-01 -9.00150418e-01 6.58796847e-01 2.74435431e-01 2.80836165e-01 -1.00683570e+00 -3.30738770e-03 -1.38079980e-02 5.80151439e-01 3.53756160e-01 -1.10702462e-01 7.83144653e-01 -5.52805364e-01 -5.75945675e-01 4.52764332e-01 -1.40916303e-01 1.82460502e-01 7.05379760e-03 1.02587688e+00 -4.86102521e-01 5.14838338e-01 3.99756551e-01 -2.22476214e-01 2.96108574e-01 -4.88786511e-02 -1.00376976e+00 -6.96994960e-01 -3.39610547e-01 -5.33447385e-01 3.92078459e-01 3.43036413e-01 2.65748411e-01 -8.31409276e-01 -5.19006193e-01 -9.21087563e-02 -3.66879702e-01 -5.04900277e-01 -1.33729562e-01 -3.65751952e-01 -1.12085378e+00 7.96777904e-01 3.55414599e-01 1.13316441e+00 -1.11835015e+00 -3.90450925e-01 -9.81976986e-02 3.20062697e-01 -8.33978117e-01 -1.72428265e-01 -3.61040711e-01 -6.68336570e-01 -1.53604484e+00 -7.29648530e-01 -9.52756047e-01 6.71477854e-01 1.02936901e-01 5.03308892e-01 6.62632465e-01 -6.14857078e-01 4.10569198e-02 -2.01114099e-02 -5.85129261e-01 1.68417528e-01 1.51605085e-01 2.67996013e-01 7.83732474e-01 1.10392320e+00 -2.53540426e-01 -8.97460282e-01 1.73477426e-01 -6.61334276e-01 -6.54399246e-02 6.98414564e-01 1.10985708e+00 2.89533257e-01 1.83923423e-01 4.88272756e-01 -8.28842461e-01 4.77121353e-01 -1.60809547e-01 -7.03795731e-01 2.37322867e-01 -8.59608829e-01 -4.60614055e-01 6.02706134e-01 -3.38432550e-01 -1.01132405e+00 -3.04826573e-02 -4.65089977e-01 5.17960750e-02 -3.45531642e-01 5.31728625e-01 -1.48176625e-01 -7.76870966e-01 6.70505285e-01 2.31675327e-01 2.66732931e-01 -4.79305327e-01 -1.46082595e-01 1.24933600e+00 5.63175440e-01 -6.72602206e-02 5.26839793e-01 1.90273717e-01 2.08329737e-01 -8.36479485e-01 -2.26333827e-01 -6.06531441e-01 -2.48873144e-01 -2.13979110e-01 8.39735925e-01 -7.89925635e-01 -1.71533000e+00 1.33439267e+00 -1.01816499e+00 2.95862943e-01 2.93146521e-01 7.76496649e-01 2.01122776e-01 3.80860060e-01 -7.65226007e-01 -1.01928771e+00 -8.16882014e-01 -1.26740432e+00 6.39766693e-01 1.16390383e+00 3.70421708e-01 -7.25175202e-01 -3.19999099e-01 5.97232938e-01 6.21056437e-01 2.14645118e-01 7.13971198e-01 -3.95186961e-01 -6.25384569e-01 -5.72059691e-01 -5.85164428e-01 4.30947661e-01 4.44533885e-01 1.34480253e-01 -1.23699164e+00 -3.57029259e-01 1.77799374e-01 -1.18048908e-02 7.69516110e-01 5.79267442e-01 1.36876118e+00 -2.47182295e-01 -1.96884990e-01 1.16322255e+00 1.62487447e+00 6.12741530e-01 9.68023896e-01 2.57739007e-01 7.24077642e-01 4.65599179e-01 5.07356152e-02 1.28266260e-01 1.31189570e-01 4.97899324e-01 3.07473123e-01 -7.36907244e-01 1.35480344e-01 7.61201605e-02 -1.20646574e-01 5.15075147e-01 -6.96912110e-01 1.36395365e-01 -9.81510758e-01 1.51033551e-01 -1.68309319e+00 -1.05005956e+00 -9.79623795e-02 2.20760226e+00 5.56781471e-01 -2.12833509e-01 -3.40233482e-02 3.50191534e-01 7.12797344e-01 -2.43436724e-01 -6.96278751e-01 -3.81992042e-01 -2.44256660e-01 8.19923639e-01 5.75393200e-01 3.02532375e-01 -1.19717717e+00 7.42177188e-01 5.49794245e+00 5.68318307e-01 -1.67150211e+00 -1.21266976e-01 9.54653144e-01 2.23145381e-01 2.17612937e-01 -3.79727006e-01 -6.17459178e-01 5.28706789e-01 7.16362119e-01 3.97023737e-01 5.75823724e-01 5.12625396e-01 1.43679649e-01 -6.77503133e-03 -4.86903816e-01 1.56182086e+00 2.70607863e-02 -1.35403764e+00 -1.39273137e-01 1.20609671e-01 5.89860380e-01 -1.91367641e-01 6.38958752e-01 -7.57677183e-02 -6.16473138e-01 -1.58021903e+00 -3.39052141e-01 8.76776814e-01 1.21716535e+00 -1.04757154e+00 1.34681928e+00 2.88845807e-01 -1.01751947e+00 -3.29527318e-01 -2.82368928e-01 -3.39633554e-01 -4.50975746e-01 4.43396479e-01 -3.99295390e-01 7.08195329e-01 8.46974671e-01 6.79659069e-01 -5.20880520e-01 1.16947794e+00 1.72191754e-01 4.70513880e-01 -4.15329300e-02 -1.63533464e-01 -7.53601640e-02 -5.23419738e-01 2.74524596e-02 8.98860693e-01 3.18126649e-01 3.53162855e-01 -2.91783273e-01 5.13867557e-01 -5.10327891e-02 3.93053681e-01 -4.89789873e-01 -2.36910377e-02 1.34492442e-02 1.26716709e+00 -5.28788865e-01 -3.35609876e-02 -5.90925455e-01 5.52751839e-01 -1.75761431e-01 3.38621199e-01 -6.52387559e-01 -7.86346555e-01 6.57246232e-01 -6.48942366e-02 -5.77277206e-02 1.21788725e-01 -7.70485818e-01 -1.23613834e+00 9.99906734e-02 -1.19549191e+00 1.37678161e-01 -4.29391742e-01 -1.03989422e+00 5.28425515e-01 -8.85072052e-01 -1.18251204e+00 2.47993991e-01 -1.11336637e+00 -6.93285584e-01 1.45449948e+00 -1.58468127e+00 -1.33457899e+00 -6.62777543e-01 8.35699260e-01 -1.71196256e-02 -7.83118129e-01 9.55817759e-01 6.22209609e-01 -1.05843806e+00 1.00052524e+00 1.91934332e-02 5.66673160e-01 4.62594599e-01 -7.13706315e-01 3.21067870e-01 1.05088508e+00 -1.03305921e-01 1.01101267e+00 -9.75501761e-02 -4.97052193e-01 -1.50059438e+00 -5.69870412e-01 7.50234663e-01 -1.68578357e-01 -1.44245233e-02 1.50995180e-01 -6.85303867e-01 2.57239610e-01 1.60019934e-01 1.46055356e-01 9.87226188e-01 2.40333706e-01 -1.30028114e-01 -6.16931021e-01 -1.38598299e+00 3.21726710e-01 6.48001909e-01 -6.27991378e-01 3.17934118e-02 2.94694543e-01 9.87994745e-02 -8.06103587e-01 -7.37040758e-01 6.79396987e-01 9.69017148e-01 -1.26622689e+00 9.49157238e-01 -6.90640807e-01 7.03227594e-02 -5.03334224e-01 4.98277366e-01 -4.88231301e-01 -3.30958277e-01 -6.61473155e-01 -1.64361611e-01 1.01761830e+00 2.39594668e-01 -1.05227017e+00 1.02676725e+00 6.25824392e-01 3.40365648e-01 -1.22587204e+00 -9.17929888e-01 -2.34620094e-01 -4.75445628e-01 -5.85444495e-02 1.07096457e+00 8.43223810e-01 -1.05187602e-01 1.77140608e-01 -4.28868890e-01 4.65946436e-01 4.45294112e-01 -1.55570149e-01 6.77010417e-01 -1.35418391e+00 -1.97606921e-01 -4.29133385e-01 -8.05018723e-01 -9.58463669e-01 -3.23747456e-01 -5.31819105e-01 -7.65624523e-01 -1.10878134e+00 -8.37126821e-02 -5.13804138e-01 -7.49172926e-01 3.98728490e-01 -1.55214980e-01 3.93034160e-01 -4.07432795e-01 1.15202397e-01 8.21553171e-02 9.55544561e-02 1.13671076e+00 -3.41580153e-01 -3.19286913e-01 5.51573992e-01 -7.62034535e-01 4.31808829e-01 7.28238881e-01 2.43844762e-01 -2.83006817e-01 -3.72222275e-01 1.98605657e-01 -5.74164651e-02 2.19140828e-01 -1.16333258e+00 7.26743996e-01 1.18748374e-01 9.04162884e-01 -9.28661227e-02 7.75196105e-02 -9.60463345e-01 2.14127019e-01 5.02584398e-01 1.99415818e-01 3.18515860e-02 2.94905365e-01 1.49935052e-01 -3.44865441e-01 1.31720841e-01 8.40455353e-01 1.27101526e-01 -6.00390017e-01 7.00609684e-01 2.10843787e-01 -6.45104766e-01 9.34908569e-01 -7.68793702e-01 -5.09282172e-01 1.14947088e-01 -2.88118869e-01 -1.12736098e-01 2.43019179e-01 3.76986533e-01 8.37066174e-01 -1.08530378e+00 -4.60668623e-01 1.02250743e+00 -2.40553636e-02 -1.50046065e-01 6.26067042e-01 1.04828703e+00 -8.12939525e-01 3.70467186e-01 -3.52489650e-01 -5.71676075e-01 -1.54246640e+00 5.37747085e-01 9.20019448e-01 1.40078291e-01 -4.00391936e-01 9.07062888e-01 -3.85916352e-01 -2.48737082e-01 4.33388442e-01 -1.33355811e-01 -9.37612593e-01 -4.65230763e-01 8.94167721e-01 3.44569802e-01 2.79005796e-01 -7.31736839e-01 -2.68804729e-01 1.06105864e+00 -2.14274719e-01 4.63497251e-01 1.07035804e+00 1.80807441e-01 -5.06144822e-01 -1.77494198e-01 8.65900159e-01 -3.17931399e-02 -6.21692836e-01 -9.31346491e-02 -3.62104118e-01 -3.84361088e-01 1.42732561e-01 -1.03889382e+00 -1.21815073e+00 8.51406217e-01 1.29202724e+00 8.85979235e-02 1.49623525e+00 -7.82542408e-01 8.61786485e-01 1.89578727e-01 1.90652445e-01 -9.74667668e-01 -4.73364085e-01 3.41704637e-01 1.80579975e-01 -1.42298889e+00 -1.72020361e-01 -1.75826967e-01 -2.04037696e-01 1.39768565e+00 7.89600134e-01 1.11358784e-01 8.88655007e-01 8.27742293e-02 4.90445346e-01 -2.09242508e-01 1.10302389e-01 -1.26717255e-01 6.74471676e-01 7.08895504e-01 6.94671869e-01 1.04697123e-02 -3.14777792e-01 7.75339246e-01 5.31568266e-02 4.07120138e-01 9.72479731e-02 4.95503008e-01 -1.21062689e-01 -1.05872440e+00 -2.45849684e-01 6.71147227e-01 -8.17209661e-01 -7.86204636e-02 -1.13243409e-01 3.69094104e-01 7.25102186e-01 1.09354675e+00 -1.02582395e-01 -8.52672815e-01 2.10264355e-01 -6.88472241e-02 3.19700330e-01 -1.57185897e-01 -9.86503601e-01 -2.06077784e-01 -4.06946599e-01 -3.89090985e-01 -6.89726293e-01 -9.60363373e-02 -8.10468912e-01 -6.46167338e-01 -5.33894837e-01 1.19201355e-02 7.39639342e-01 1.13874543e+00 6.12884998e-01 3.01521301e-01 7.26531863e-01 -3.22002232e-01 -1.79167822e-01 -9.45805609e-01 -6.38884783e-01 8.71959329e-02 6.43404841e-01 -2.98883200e-01 5.88829741e-02 -2.79488832e-01]

[3.7703969478607178, -3.6165342330932617]

ba0bc57d-5f5d-49ee-baa0-3b6dba7d74f7

survival-analysis-meets-counterfactual

2006.07756

https://arxiv.org/abs/2006.07756v2

https://arxiv.org/pdf/2006.07756v2.pdf

Enabling Counterfactual Survival Analysis with Balanced Representations

Balanced representation learning methods have been applied successfully to counterfactual inference from observational data. However, approaches that account for survival outcomes are relatively limited. Survival data are frequently encountered across diverse medical applications, i.e., drug development, risk profiling, and clinical trials, and such data are also relevant in fields like manufacturing (e.g., for equipment monitoring). When the outcome of interest is a time-to-event, special precautions for handling censored events need to be taken, as ignoring censored outcomes may lead to biased estimates. We propose a theoretically grounded unified framework for counterfactual inference applicable to survival outcomes. Further, we formulate a nonparametric hazard ratio metric for evaluating average and individualized treatment effects. Experimental results on real-world and semi-synthetic datasets, the latter of which we introduce, demonstrate that the proposed approach significantly outperforms competitive alternatives in both survival-outcome prediction and treatment-effect estimation.

['Michael J. Pencina', 'Paidamoyo Chapfuwa', 'Shuxi Zeng', 'Lawrence Carin', 'Serge Assaad', 'Ricardo Henao']

2020-06-14

enabling-counterfactual-survival-analysis

https://openreview.net/forum?id=3ZeGLibhFo0

https://openreview.net/pdf?id=3ZeGLibhFo0

['counterfactual-inference']

['miscellaneous']

[ 5.47351837e-01 -2.15065077e-01 -8.66846025e-01 -5.53189754e-01 -1.00460112e+00 -1.61161825e-01 2.76584208e-01 5.95781624e-01 -2.18506634e-01 1.39306772e+00 2.50153124e-01 -7.95764506e-01 -5.37312388e-01 -6.97140694e-01 -6.02498651e-01 -9.58192885e-01 -4.00282204e-01 3.11966985e-01 -7.65830576e-01 3.20901781e-01 2.41041183e-01 4.95583475e-01 -1.23044527e+00 -2.55328566e-01 9.79876757e-01 9.21524644e-01 -3.96094859e-01 1.46230295e-01 2.05296442e-01 6.55935943e-01 -5.84531367e-01 -3.55328262e-01 -2.36522570e-01 -3.02908361e-01 -3.11218441e-01 -2.26206511e-01 -9.42177549e-02 -2.64191210e-01 -2.41928771e-01 6.92357659e-01 6.38413787e-01 1.43856287e-01 1.04016376e+00 -1.38511562e+00 -4.87612188e-01 6.26446426e-01 -7.53131628e-01 -2.59019192e-02 2.05204576e-01 1.06876704e-03 6.85119569e-01 -5.11855185e-01 9.83653963e-02 1.10166609e+00 4.58702564e-01 4.69090670e-01 -1.33512104e+00 -7.29973078e-01 6.40432015e-02 8.88248980e-02 -1.13987589e+00 -3.30832928e-01 8.15188766e-01 -4.55794930e-01 2.26924226e-01 4.37470049e-01 2.79556304e-01 1.47138309e+00 9.04179573e-01 7.08418071e-01 1.26751041e+00 -2.97280669e-01 6.89473689e-01 -1.22053988e-01 1.02763183e-01 -5.60671724e-02 7.34458745e-01 6.32675111e-01 -5.49355559e-02 -5.55981159e-01 7.26034462e-01 4.68908191e-01 -4.14760828e-01 -4.71244663e-01 -1.21714926e+00 1.02636385e+00 1.53806701e-01 -3.45823355e-02 -7.26461709e-01 9.94757190e-02 6.96275115e-01 1.99229896e-01 4.85073000e-01 1.07956119e-01 -5.86870611e-01 8.67657140e-02 -9.02235985e-01 3.06012452e-01 7.39033222e-01 7.64376700e-01 -1.83741711e-02 -2.79990281e-03 -6.01921141e-01 7.98985243e-01 -6.96650986e-03 3.90877336e-01 3.92531693e-01 -6.43365741e-01 2.54791439e-01 1.79725274e-01 5.86786807e-01 -5.35359263e-01 -4.83970314e-01 -3.98850352e-01 -1.39439332e+00 9.64485407e-02 5.67487538e-01 -1.98483169e-01 -7.38723695e-01 1.80295873e+00 3.23969394e-01 4.80294079e-01 2.39210710e-01 7.13026166e-01 5.16326129e-01 3.33770722e-01 6.95985377e-01 -1.08277524e+00 1.27566183e+00 -3.28459702e-02 -9.18577731e-01 1.62928864e-01 7.93489993e-01 -4.58956212e-01 8.08940887e-01 3.16291153e-01 -9.39092755e-01 4.58183289e-02 -8.24856222e-01 3.52709979e-01 -8.75868052e-02 -5.17303348e-02 9.92092609e-01 7.86860764e-01 -6.83903471e-02 7.97993898e-01 -4.54262018e-01 -6.11662045e-02 7.26389766e-01 2.24900514e-01 -6.76145330e-02 -1.96652710e-01 -1.36290526e+00 7.02021956e-01 2.25891918e-01 7.43575469e-02 -1.08331168e+00 -9.36782598e-01 -9.35861051e-01 2.50337809e-01 5.83634019e-01 -1.00804079e+00 1.20788383e+00 -4.31703717e-01 -1.30925763e+00 4.45352584e-01 -6.08399995e-02 -5.47347546e-01 7.73500144e-01 -7.19008073e-02 -6.14046514e-01 -5.04844069e-01 3.93102644e-03 -1.66266143e-01 6.31739676e-01 -8.79826188e-01 -4.29021806e-01 -5.06723106e-01 4.67419736e-02 -2.79407226e-03 -1.41159892e-01 3.29040326e-02 4.55644011e-01 -9.90288436e-01 -2.15221614e-01 -6.48889661e-01 -6.49193406e-01 -9.43355560e-02 -5.31999707e-01 -2.65623331e-01 4.45472926e-01 -5.57883322e-01 1.29145634e+00 -1.91799378e+00 -1.81859478e-01 -4.29624245e-02 -7.68758506e-02 -3.23376358e-01 2.71902949e-01 2.88882375e-01 -5.05988777e-01 1.56748772e-01 -4.68878657e-01 -8.27184692e-02 -7.86527768e-02 -1.78222552e-01 -3.63797218e-01 7.70432234e-01 2.09222566e-02 6.75880253e-01 -7.94064164e-01 -6.14072919e-01 4.03503090e-01 3.09097588e-01 -1.27450779e-01 2.31389344e-01 -1.29952893e-01 5.13085246e-01 -7.58840501e-01 8.08859289e-01 6.22335851e-01 -3.32323819e-01 2.81636775e-01 2.01159015e-01 3.24223675e-02 1.95049122e-01 -9.24399495e-01 1.39043212e+00 -9.37097728e-01 5.87696321e-02 -5.26488185e-01 -1.19857192e+00 5.47278464e-01 4.99672830e-01 4.88827735e-01 -2.93062568e-01 4.63295221e-01 1.53726429e-01 -1.94560289e-02 -3.36227834e-01 1.99562907e-02 -7.51094878e-01 -4.29671258e-01 2.58574009e-01 -4.52360898e-01 9.54377577e-02 -2.11402893e-01 -2.73733288e-01 7.20762432e-01 -8.49520904e-04 1.01390088e+00 -2.78943211e-01 5.33466756e-01 -2.31848329e-01 9.48886037e-01 6.45114005e-01 -1.23409495e-01 4.08716559e-01 7.29525805e-01 -4.17977929e-01 -5.94695330e-01 -1.15297914e+00 -5.81988454e-01 5.62988877e-01 -1.19900182e-02 4.63583916e-01 -2.53908068e-01 -8.08908522e-01 2.54224092e-01 1.32800460e+00 -6.99145854e-01 -4.44010109e-01 -2.49486238e-01 -1.27696991e+00 1.67716682e-01 7.22808242e-01 -1.94341883e-01 -8.57643783e-01 -4.61874217e-01 5.67074418e-01 -1.26129892e-02 -5.06889999e-01 -4.98277813e-01 7.58527741e-02 -1.21313488e+00 -1.18109894e+00 -1.13299239e+00 -3.14991295e-01 2.94056803e-01 -1.15214311e-01 1.02732408e+00 -3.06125641e-01 -3.03539425e-01 -3.51555683e-02 8.96242633e-02 -6.47980034e-01 -3.96715194e-01 -4.34604883e-01 2.03471124e-01 5.67293912e-03 3.18000883e-01 -5.35216689e-01 -8.87927413e-01 4.09079850e-01 -7.25064218e-01 -2.21102446e-01 6.66214108e-01 1.22172129e+00 7.50818431e-01 -1.18382894e-01 1.41373396e+00 -1.07242668e+00 5.41341603e-01 -8.34485292e-01 -6.82632387e-01 4.47151601e-01 -8.21335316e-01 -9.10234079e-02 7.19006717e-01 -7.12259829e-01 -1.35375452e+00 -3.48808438e-01 1.08332209e-01 -3.20455313e-01 -2.44760752e-01 7.99936593e-01 -5.74056029e-01 4.93032306e-01 3.86962652e-01 2.88857240e-02 -1.70839429e-01 -2.81440884e-01 4.58158739e-02 9.18819666e-01 2.48843744e-01 -5.98515451e-01 1.56331211e-01 3.99015099e-01 4.90851551e-01 -4.58007246e-01 -6.25112236e-01 -2.48764440e-01 5.03927805e-02 1.31603196e-01 3.51970255e-01 -8.75888407e-01 -1.22617233e+00 2.80835807e-01 -8.91394615e-01 5.66430204e-02 -4.45346087e-01 1.05962682e+00 -1.09775889e+00 2.09646180e-01 -2.94924766e-01 -1.17146635e+00 -3.07597786e-01 -1.00626242e+00 9.01624799e-01 2.93412715e-01 -3.57666254e-01 -1.20072496e+00 1.42926667e-02 9.97397527e-02 1.32563993e-01 7.78981149e-01 1.42522311e+00 -6.62782073e-01 -1.45425692e-01 -6.03193700e-01 -7.28291124e-02 -2.89042424e-02 4.37564939e-01 -2.70034939e-01 -8.49950254e-01 -3.46023649e-01 9.70430598e-02 -3.07665229e-01 5.72906494e-01 1.08420432e+00 1.79288018e+00 -3.35864365e-01 -6.07072413e-01 2.65840590e-01 1.23343205e+00 4.88638937e-01 6.54329419e-01 -2.18504027e-01 2.81712741e-01 8.17992687e-01 9.92384493e-01 8.69934082e-01 1.48286983e-01 6.39714420e-01 4.13491219e-01 6.59674555e-02 4.25350338e-01 -1.11616842e-01 -3.97170112e-02 1.32089794e-01 4.28105332e-02 -5.64104021e-01 -5.67641318e-01 7.60486841e-01 -1.80613410e+00 -8.72529685e-01 -1.78296149e-01 2.96762538e+00 9.11443532e-01 3.19362618e-02 1.44251473e-02 1.24474585e-01 9.12577927e-01 -8.52971971e-02 -1.13836300e+00 -3.76912206e-01 3.07177324e-02 1.68582350e-01 5.94405472e-01 -4.17774841e-02 -1.05906868e+00 7.33147264e-02 5.87987041e+00 9.22902465e-01 -9.18622673e-01 9.34553221e-02 1.20491946e+00 -9.34555009e-02 -3.78181607e-01 -6.35421509e-03 -1.77721977e-01 6.57264888e-01 1.08513653e+00 -7.87558854e-01 -2.43901819e-01 7.08092034e-01 6.79263115e-01 -5.03391912e-03 -1.42197728e+00 8.97138238e-01 -4.43102092e-01 -1.11839521e+00 -7.88821951e-02 1.32734731e-01 6.71638310e-01 -6.92206502e-01 1.11482725e-01 3.41645837e-01 3.25141698e-01 -1.15695775e+00 3.06704521e-01 6.55463040e-01 1.24609339e+00 -1.07043517e+00 1.00121224e+00 2.92887390e-01 -5.58290303e-01 -1.34637937e-01 -1.81699514e-01 2.70113791e-03 3.86879951e-01 1.11664200e+00 -5.52140236e-01 8.62740755e-01 3.25621337e-01 5.13565660e-01 1.64484382e-01 1.35152066e+00 -1.33758232e-01 7.00700343e-01 6.32769382e-03 -6.91623837e-02 -4.07184690e-01 -8.62363726e-02 2.83027381e-01 5.90844393e-01 5.94755948e-01 -6.59679994e-02 -3.14063936e-01 6.82566404e-01 -1.48077518e-01 1.60249323e-01 -7.06643343e-01 -3.67841683e-02 5.07476389e-01 7.97031581e-01 -4.45575863e-01 -1.83926135e-01 -5.28607905e-01 6.39431655e-01 -1.71467841e-01 4.48590010e-01 -1.04955590e+00 -3.03456873e-01 8.48640621e-01 1.72867268e-01 -1.27101630e-01 3.73719126e-01 -6.57571435e-01 -1.02687204e+00 -2.41567940e-01 -7.03139961e-01 7.82037616e-01 -2.39858299e-01 -1.53998125e+00 -1.13033606e-02 3.46109867e-01 -1.52904046e+00 -4.79863137e-01 -4.36367899e-01 -7.87325740e-01 9.56878841e-01 -1.37701607e+00 -8.56019855e-01 1.18475668e-01 2.36505285e-01 4.72021222e-01 1.51847869e-01 8.08070004e-01 2.94947416e-01 -9.10335600e-01 7.20416188e-01 4.32024598e-01 -2.98736781e-01 7.90886104e-01 -1.09253836e+00 8.20568502e-02 2.49220669e-01 -2.50756115e-01 6.37898326e-01 8.24877977e-01 -6.92802429e-01 -1.16358638e+00 -1.26449955e+00 7.60449946e-01 -1.80289596e-01 5.20611882e-01 3.31812650e-02 -8.25352609e-01 4.26887512e-01 -2.53573328e-01 2.51704571e-03 9.70468462e-01 3.88975233e-01 4.21503298e-02 -1.07299641e-01 -1.39656568e+00 7.57469773e-01 8.18281293e-01 -2.02185228e-01 -5.06664753e-01 5.12185872e-01 5.17956436e-01 -3.29611987e-01 -9.74598527e-01 6.95293665e-01 7.75781751e-01 -6.06554806e-01 9.62669075e-01 -1.09808862e+00 5.06345689e-01 -9.89364684e-02 -1.54096195e-02 -1.41008365e+00 -9.69955549e-02 -6.12516165e-01 -2.94298381e-01 1.13512254e+00 2.96961695e-01 -8.72662783e-01 6.71321809e-01 5.71991503e-01 2.72375584e-01 -9.75369990e-01 -1.31857264e+00 -9.88793552e-01 2.85631031e-01 -4.26797271e-01 9.58240271e-01 9.36670661e-01 -3.36898305e-02 8.21063593e-02 -6.78476512e-01 1.78795919e-01 9.79271591e-01 5.21763086e-01 3.43756050e-01 -1.34147918e+00 -1.28499582e-01 -2.56533325e-01 -2.93511719e-01 -5.13717592e-01 3.69378835e-01 -3.38514984e-01 -1.31191775e-01 -1.31059754e+00 4.94433105e-01 -5.29061615e-01 -7.84550786e-01 7.73938894e-02 -5.38737833e-01 -2.82382518e-01 -4.33715820e-01 -1.01359919e-01 -6.20566942e-02 9.19071019e-01 8.75355840e-01 -1.95589289e-01 -5.18384278e-02 6.23765230e-01 -8.66213024e-01 6.54835939e-01 8.73934984e-01 -6.54401898e-01 -4.85908598e-01 3.76479059e-01 -1.84051409e-01 8.51115227e-01 4.77532804e-01 -5.21224916e-01 -2.29056701e-01 -7.70535886e-01 3.46009851e-01 -3.83081138e-01 -7.36610144e-02 -9.11680043e-01 4.42276478e-01 5.33512771e-01 -3.53933364e-01 -2.44227260e-01 6.09343983e-02 1.11849117e+00 -6.59588724e-02 -6.46124408e-02 6.92038596e-01 3.87824357e-01 -1.49954900e-01 4.22922343e-01 -1.33324772e-01 -1.59622151e-02 1.38889933e+00 -1.55922949e-01 -2.92702138e-01 -5.99919736e-01 -6.76913619e-01 5.21868348e-01 2.38819733e-01 4.65626508e-01 6.67385817e-01 -1.49369037e+00 -7.81142890e-01 -1.76908046e-01 3.83197248e-01 -2.25598603e-01 7.40123093e-01 9.55852687e-01 1.23097159e-01 4.13879037e-01 1.43387765e-01 -2.42381439e-01 -8.65737617e-01 9.28737700e-01 5.65075874e-02 -5.12404084e-01 -2.94949323e-01 2.31099784e-01 6.47319376e-01 -6.46524727e-02 1.31307960e-01 -2.59996235e-01 -1.93460852e-01 4.04404141e-02 5.30618370e-01 8.68904293e-01 2.01705080e-02 -1.39813632e-01 -4.65734839e-01 8.64361301e-02 -2.37811148e-01 2.71951318e-01 1.20945549e+00 -1.14392757e-01 3.09614509e-01 8.03447425e-01 1.00510430e+00 -3.90298486e-01 -1.21912336e+00 -5.98396501e-03 1.29183695e-01 -4.35962558e-01 7.01403394e-02 -8.08423817e-01 -6.39698148e-01 7.62551248e-01 6.16502821e-01 1.14609390e-01 1.22641504e+00 -1.74775526e-01 4.98672009e-01 2.97487956e-02 7.31446862e-01 -8.12518656e-01 -6.07244194e-01 -2.01098770e-01 9.46853220e-01 -1.40059626e+00 1.88592196e-01 -4.13866907e-01 -3.99739861e-01 5.77777445e-01 1.67434990e-01 2.12145463e-01 6.87190413e-01 -2.04106662e-02 -1.11140087e-01 2.77138174e-01 -7.18471766e-01 3.06523621e-01 7.86951631e-02 5.28405488e-01 8.00012648e-01 5.94840884e-01 -8.87994826e-01 9.19869781e-01 2.49379218e-01 5.42200029e-01 5.94488978e-01 8.47032189e-01 4.14233595e-01 -9.86256599e-01 -5.01129866e-01 9.78229582e-01 -7.76128173e-01 -1.05852569e-02 2.14083999e-01 8.84318709e-01 -3.39941293e-01 9.15200412e-01 -7.02256113e-02 2.73783147e-01 5.15368104e-01 4.28008996e-02 2.92504907e-01 -3.91377389e-01 1.20883482e-02 -1.01643987e-01 4.22758311e-02 -4.06435490e-01 -3.07755679e-01 -7.86440730e-01 -9.81067479e-01 -2.76222378e-01 -4.69036102e-01 1.92688152e-01 4.89394993e-01 7.40988731e-01 1.43434033e-01 8.24067116e-01 9.79955554e-01 -4.76263046e-01 -1.14405560e+00 -8.09647560e-01 -9.70221341e-01 1.90244883e-01 4.41117555e-01 -9.96654272e-01 -4.24219996e-01 -2.19411775e-01]

[8.007707595825195, 5.415359020233154]

e697e367-25c6-44e9-9781-466d312facb0

urban-scene-semantic-segmentation-with-low

2212.07911

https://arxiv.org/abs/2212.07911v1

https://arxiv.org/pdf/2212.07911v1.pdf

Urban Scene Semantic Segmentation with Low-Cost Coarse Annotation

For best performance, today's semantic segmentation methods use large and carefully labeled datasets, requiring expensive annotation budgets. In this work, we show that coarse annotation is a low-cost but highly effective alternative for training semantic segmentation models. Considering the urban scene segmentation scenario, we leverage cheap coarse annotations for real-world captured data, as well as synthetic data to train our model and show competitive performance compared with finely annotated real-world data. Specifically, we propose a coarse-to-fine self-training framework that generates pseudo labels for unlabeled regions of the coarsely annotated data, using synthetic data to improve predictions around the boundaries between semantic classes, and using cross-domain data augmentation to increase diversity. Our extensive experimental results on Cityscapes and BDD100k datasets demonstrate that our method achieves a significantly better performance vs annotation cost tradeoff, yielding a comparable performance to fully annotated data with only a small fraction of the annotation budget. Also, when used as pretraining, our framework performs better compared to the standard fully supervised setting.

['Bernt Schiele', 'Zeynep Akata', 'Yang He', 'Yongqin Xian', 'Anurag Das']

2022-12-15

['scene-segmentation']

['computer-vision']

[ 2.41284370e-01 4.95873898e-01 -4.70804483e-01 -6.02245510e-01 -1.14782214e+00 -6.36945307e-01 4.70650047e-01 2.66997628e-02 -5.88371933e-01 8.36854577e-01 -1.14137689e-02 -1.04598194e-01 4.75955039e-01 -7.77033269e-01 -8.77175570e-01 -3.38207394e-01 3.24505448e-01 1.05788124e+00 7.62407601e-01 -1.68109108e-02 -2.47014731e-01 2.62180362e-02 -1.50990427e+00 1.89590484e-01 1.26431596e+00 1.00057340e+00 2.37318262e-01 3.75659436e-01 -1.89008310e-01 5.88064075e-01 -4.82501537e-01 -3.98553550e-01 5.28668046e-01 -9.19379815e-02 -1.25602913e+00 6.01971984e-01 4.53760207e-01 -3.04734826e-01 -2.56662369e-02 9.38838661e-01 1.87526479e-01 5.10334000e-02 5.30660987e-01 -1.03342986e+00 -3.50704610e-01 5.49535990e-01 -6.36025071e-01 -1.68377072e-01 -1.33034242e-02 1.57977358e-01 1.04268789e+00 -6.60897493e-01 6.44142985e-01 1.07357144e+00 8.20801973e-01 4.74025130e-01 -1.35118127e+00 -5.26184797e-01 4.30150658e-01 -2.11253971e-01 -1.58905423e+00 -3.67931813e-01 5.11064112e-01 -4.13561523e-01 5.95256090e-01 -1.68376751e-02 5.86155653e-01 1.06180692e+00 -7.48901248e-01 1.09586406e+00 1.18103814e+00 -3.41720790e-01 4.05984998e-01 1.94464535e-01 2.47117028e-01 6.79384649e-01 2.65199602e-01 -1.15929402e-01 -4.46970984e-02 -7.48477355e-02 6.33040071e-01 -1.48720160e-01 -2.95854993e-02 -4.61417198e-01 -1.07695496e+00 7.95265138e-01 4.59049493e-01 -5.90222364e-04 -2.99903583e-02 1.28784895e-01 2.75243431e-01 -9.29923579e-02 7.36836672e-01 2.80349851e-01 -7.72836626e-01 -6.73244745e-02 -1.08658910e+00 5.48710562e-02 6.80951059e-01 1.22496843e+00 1.11156809e+00 -6.74485266e-02 -7.18647838e-02 1.07740092e+00 1.19078286e-01 5.42603910e-01 4.66902912e-01 -1.21037233e+00 7.59982109e-01 6.31090760e-01 2.96527624e-01 -4.05615509e-01 -3.97780925e-01 -4.22980279e-01 -5.59997678e-01 -9.77895260e-02 7.62616396e-01 -1.96896702e-01 -1.39022148e+00 1.69274259e+00 4.18122798e-01 3.98827612e-01 4.54942659e-02 7.84710944e-01 5.43903887e-01 4.20970887e-01 3.92939895e-01 2.85023421e-01 1.13181007e+00 -1.41051650e+00 -2.77849078e-01 -6.00122511e-01 9.13147390e-01 -3.99658799e-01 1.57678545e+00 2.16698460e-02 -7.72700906e-01 -5.51560223e-01 -7.72404432e-01 2.53610741e-02 -4.24754590e-01 4.00127828e-01 7.51597345e-01 6.78221285e-01 -8.94174576e-01 3.99662077e-01 -1.00360799e+00 -4.59841669e-01 8.26317251e-01 1.32287920e-01 -1.81610003e-01 -2.49635994e-01 -9.27386940e-01 3.54186565e-01 7.80990303e-01 -2.61967301e-01 -9.90662456e-01 -7.08982408e-01 -1.04832983e+00 -5.51130921e-02 7.11900592e-01 -3.82954538e-01 1.38970065e+00 -9.47573721e-01 -1.33743691e+00 1.24614131e+00 1.12138636e-01 -6.03585124e-01 7.65355825e-01 -2.61233389e-01 -1.37660593e-01 1.51754946e-01 5.23546219e-01 1.24185443e+00 4.54390794e-01 -1.45515501e+00 -8.06887507e-01 -2.13711292e-01 1.98074117e-01 1.19848520e-01 -8.41515213e-02 -5.14242709e-01 -9.36957598e-01 -6.70948148e-01 1.66011661e-01 -1.24147522e+00 -6.41014040e-01 -2.17069179e-01 -7.04397202e-01 -1.48192998e-02 6.98699892e-01 -3.72368217e-01 8.48963022e-01 -2.14551187e+00 -4.19611484e-01 1.66889697e-01 4.96726222e-02 3.86890858e-01 -2.34091312e-01 -9.47012380e-02 2.76485503e-01 2.63218939e-01 -7.37702310e-01 -5.37930727e-01 1.05083115e-01 5.22566080e-01 -1.34892553e-01 1.18635692e-01 3.57473642e-01 1.12859869e+00 -9.83380020e-01 -6.24223948e-01 2.23553583e-01 6.15008697e-02 -6.88678026e-01 1.22331128e-01 -5.46154261e-01 6.17690682e-01 -6.98054075e-01 7.77065396e-01 5.89509904e-01 -6.69090390e-01 1.58570707e-01 8.19181502e-02 3.10665905e-01 2.86139529e-02 -1.08074582e+00 1.83926129e+00 -5.62039435e-01 4.90668118e-01 -6.56225383e-02 -1.04583907e+00 7.95126379e-01 -4.40405682e-02 3.50808203e-01 -7.62816191e-01 1.16068728e-01 2.95238912e-01 -5.50303221e-01 -2.41090864e-01 4.76318568e-01 1.74628809e-01 -3.77362996e-01 3.64809662e-01 6.72689825e-02 -4.16926324e-01 3.69250655e-01 1.09702334e-01 9.16615844e-01 2.79467851e-01 1.56987816e-01 -1.97191551e-01 1.97926119e-01 5.80167949e-01 6.38172805e-01 7.37620592e-01 -2.71600991e-01 8.45772207e-01 3.25516820e-01 -4.01585639e-01 -1.16007388e+00 -8.31752896e-01 -1.53154120e-01 1.07868493e+00 4.24308270e-01 -2.10928202e-01 -1.27943242e+00 -1.06340992e+00 -1.26905933e-01 5.88321269e-01 -5.12392282e-01 1.70063123e-01 -4.73534077e-01 -9.53744352e-01 7.18955576e-01 7.57802904e-01 1.03376234e+00 -7.68048704e-01 -5.15960634e-01 1.40723839e-01 -3.22435528e-01 -1.73939455e+00 -2.33899012e-01 2.09082305e-01 -8.43422234e-01 -1.11033309e+00 -7.99482107e-01 -8.87296379e-01 7.18271613e-01 2.63120443e-01 1.43854988e+00 -6.46632677e-03 -1.81123763e-01 5.85074238e-02 -4.47130352e-01 -1.40843421e-01 -3.27661574e-01 4.57572430e-01 -4.25288349e-01 -1.51177257e-01 2.42472947e-01 -3.05012226e-01 -4.79719460e-01 5.97596049e-01 -7.81794786e-01 2.73783237e-01 5.13225436e-01 9.02553976e-01 8.94167244e-01 -1.77613664e-02 5.32370925e-01 -1.48021722e+00 8.65643192e-03 -4.09248859e-01 -7.82253981e-01 2.00100094e-01 -5.73006630e-01 7.22554550e-02 5.17929375e-01 -2.92551875e-01 -1.21759784e+00 4.65623170e-01 -3.18340659e-01 -3.07278275e-01 -6.82862639e-01 1.90256298e-01 -2.77185231e-01 1.34813245e-02 7.59070039e-01 -1.60477504e-01 -5.19042373e-01 -6.78368628e-01 6.08952105e-01 5.91116726e-01 6.80640876e-01 -8.03336263e-01 7.39356697e-01 7.87231088e-01 -4.67570335e-01 -5.92777848e-01 -1.36360168e+00 -5.20462275e-01 -9.52686906e-01 2.30543360e-01 9.58485246e-01 -1.23854041e+00 9.02600214e-03 3.82894665e-01 -6.36462867e-01 -1.13777566e+00 -6.90911829e-01 2.29832426e-01 -6.68600678e-01 1.75499037e-01 -5.00243545e-01 -4.52425957e-01 5.34566045e-02 -1.15747130e+00 1.59240162e+00 1.20942518e-01 2.38219136e-03 -1.01729012e+00 -5.85963912e-02 7.66055942e-01 5.31221367e-02 4.41725612e-01 5.31388223e-01 -7.64575660e-01 -7.42513120e-01 -1.61874846e-01 -5.17775238e-01 2.95007020e-01 1.87010039e-02 -2.70639300e-01 -1.16546869e+00 2.00469028e-02 -5.83011866e-01 -8.39632154e-01 9.86699820e-01 1.91446692e-01 1.43253696e+00 -7.93166608e-02 -5.54451406e-01 7.14376748e-01 1.39670300e+00 -1.33221641e-01 3.63680482e-01 3.51098508e-01 8.78612936e-01 5.09641528e-01 9.65513587e-01 3.68964374e-01 7.14689851e-01 6.07619047e-01 2.64675200e-01 -5.03299594e-01 -2.70324171e-01 -4.98041600e-01 -2.37514526e-01 3.70409161e-01 2.12704480e-01 -3.86492938e-01 -1.18275046e+00 8.53442311e-01 -1.90971458e+00 -5.77946782e-01 -1.10594146e-01 2.06721616e+00 9.44021881e-01 4.78651732e-01 1.85569584e-01 -9.35326971e-04 6.44966841e-01 9.94373113e-02 -6.09587371e-01 1.35028303e-01 -2.12606832e-01 3.23078007e-01 9.41741526e-01 4.30151284e-01 -1.49113965e+00 1.56706977e+00 6.95335388e+00 1.12994480e+00 -9.42685485e-01 3.74515563e-01 1.05928349e+00 1.59633905e-01 -2.24156305e-01 -4.56059501e-02 -7.93701589e-01 5.28898299e-01 8.78688157e-01 3.95117611e-01 1.36450678e-01 1.15345132e+00 1.81969419e-01 -1.83769450e-01 -8.53860855e-01 8.96068871e-01 -2.85223693e-01 -1.42709088e+00 -1.59128666e-01 -1.64987333e-02 1.28545904e+00 4.23745811e-01 -2.19577953e-01 4.01836693e-01 9.31195676e-01 -8.07655275e-01 8.06319237e-01 -1.20592840e-01 9.00065660e-01 -4.76977021e-01 7.17234373e-01 5.39070606e-01 -1.25653255e+00 9.59101990e-02 -3.22300762e-01 9.38887745e-02 2.93247014e-01 6.45980120e-01 -9.30828452e-01 3.54755551e-01 7.59042799e-01 7.31656432e-01 -7.84143746e-01 9.31151509e-01 -4.69167769e-01 9.37894523e-01 -5.22421122e-01 4.72790837e-01 5.72356641e-01 -1.13456376e-01 -9.12467539e-02 1.22551906e+00 -2.17768364e-02 1.15905270e-01 8.20413828e-01 8.29047322e-01 -2.75290698e-01 -4.42201225e-03 -4.52915341e-01 1.26113951e-01 5.51475465e-01 1.10205376e+00 -1.15986073e+00 -6.75120413e-01 -3.45447838e-01 1.04127705e+00 4.84833539e-01 4.59389389e-01 -9.35008049e-01 -1.11092918e-01 4.35752809e-01 7.03741312e-02 2.86279231e-01 -1.07568614e-01 -5.37709117e-01 -1.12274039e+00 -1.28746986e-01 -5.45133829e-01 4.11809325e-01 -6.70994520e-01 -1.04525542e+00 5.63985527e-01 5.97713254e-02 -1.12178981e+00 -1.63683861e-01 -4.29862380e-01 -3.59960496e-01 5.02426803e-01 -1.80989015e+00 -1.47399986e+00 -6.46401942e-01 4.29949731e-01 8.12937021e-01 2.40983114e-01 6.29560888e-01 3.94300938e-01 -6.41771972e-01 5.78704894e-01 1.36468098e-01 4.12372261e-01 4.89731342e-01 -1.32589829e+00 7.60551333e-01 7.74133921e-01 2.48744816e-01 -5.47652207e-02 3.05159748e-01 -5.62204063e-01 -4.19209957e-01 -1.60413861e+00 3.86046529e-01 -3.94545197e-01 5.63846111e-01 -4.83180493e-01 -8.46254289e-01 8.58340204e-01 -2.12540731e-01 3.19514066e-01 5.39055526e-01 4.86641601e-02 -3.01471889e-01 7.73449466e-02 -1.25923109e+00 4.32232559e-01 1.29853225e+00 -3.06200862e-01 -4.53536928e-01 5.88002443e-01 1.05487084e+00 -5.78479052e-01 -6.80189848e-01 5.74575305e-01 1.85000256e-01 -8.33135962e-01 9.03678179e-01 -3.76221597e-01 2.44883984e-01 -2.77788043e-01 -2.54331142e-01 -1.21461046e+00 5.89836165e-02 -3.26052666e-01 2.47058317e-01 1.30744994e+00 7.51781106e-01 -4.70268190e-01 1.36669290e+00 6.34260714e-01 -3.26667279e-01 -6.44330621e-01 -5.69433749e-01 -9.43851888e-01 1.94164719e-02 -5.49799800e-01 6.73529983e-01 1.04551148e+00 -5.02718508e-01 2.15526596e-01 -1.15801148e-01 2.57308692e-01 6.68881655e-01 3.02964151e-01 9.48513031e-01 -1.27822959e+00 -2.39673004e-01 -1.62328392e-01 -2.38487154e-01 -1.39421976e+00 5.33615053e-01 -7.44391084e-01 3.29353601e-01 -1.53550291e+00 1.87697411e-01 -1.17172825e+00 6.33430481e-02 7.60438859e-01 -2.52545416e-01 8.57590914e-01 -1.09981455e-01 2.98690975e-01 -9.65722144e-01 4.86681581e-01 1.25765920e+00 -3.10940385e-01 -2.09430188e-01 -4.24059527e-03 -5.67695260e-01 9.88132477e-01 7.92587280e-01 -3.46603185e-01 -6.28229856e-01 -7.22576082e-01 -2.80405462e-01 -3.12723219e-01 2.98106670e-01 -1.12408864e+00 -1.76523685e-01 -2.44280890e-01 1.25305474e-01 -5.19959211e-01 4.15980309e-01 -8.16158414e-01 -1.19982280e-01 1.86214924e-01 -3.17436874e-01 -4.77722108e-01 1.83946550e-01 6.57284975e-01 -3.93845618e-01 -7.93288499e-02 1.00164998e+00 -2.90059716e-01 -1.10821283e+00 4.48689431e-01 -4.28571515e-02 7.93931246e-01 1.18326664e+00 -3.99220675e-01 -2.49087080e-01 -1.23104863e-01 -9.45282638e-01 5.53519905e-01 8.15178812e-01 2.42022067e-01 7.30025619e-02 -1.11405826e+00 -4.03203219e-01 1.06327310e-01 4.09045547e-01 6.79679215e-01 1.65733889e-01 4.10874188e-01 -6.26200736e-01 4.00288433e-01 4.14576083e-02 -9.33413744e-01 -8.46698105e-01 3.54707539e-01 2.05625042e-01 -3.89349312e-01 -6.00485384e-01 9.18105423e-01 3.67550820e-01 -7.80810893e-01 1.42347038e-01 -5.84655643e-01 8.11243355e-02 -1.02436796e-01 1.16246238e-01 1.94464773e-01 8.85178894e-02 -6.45801246e-01 -2.58078933e-01 5.72420299e-01 2.11839527e-01 -7.43126497e-02 1.19377053e+00 -2.95799077e-01 5.33847570e-01 2.15513214e-01 9.69223082e-01 -1.37296304e-01 -1.70716798e+00 -4.22676027e-01 2.98867732e-01 -5.76674581e-01 -1.43802479e-01 -8.41031730e-01 -1.17473364e+00 6.84906602e-01 3.77269804e-01 1.56006724e-01 9.72597182e-01 3.77362192e-01 9.37618852e-01 3.49319458e-01 7.02378571e-01 -1.33927405e+00 7.83507153e-02 4.55936939e-01 1.48044929e-01 -1.67324710e+00 -4.00365025e-01 -9.53485608e-01 -9.89150107e-01 4.80079770e-01 7.63716042e-01 -1.05733760e-01 5.38904190e-01 1.76104292e-01 2.76913226e-01 -3.86917405e-02 -3.50131631e-01 -6.80717111e-01 4.85435911e-02 7.44313836e-01 8.40396583e-02 2.32281938e-01 1.35750785e-01 4.34647053e-01 -2.70549417e-01 -2.34467406e-02 2.80938178e-01 8.08747649e-01 -6.08877778e-01 -1.18333220e+00 -2.13532999e-01 3.92982185e-01 -3.32099617e-01 -1.45138845e-01 -3.06615621e-01 9.64766860e-01 2.98712283e-01 9.74476397e-01 1.93430796e-01 -6.72915727e-02 3.88656408e-01 -1.95159949e-02 1.22308485e-01 -9.23591077e-01 -1.54003471e-01 3.97408754e-02 2.84158021e-01 -7.16345549e-01 -6.07327104e-01 -5.08288682e-01 -1.41440380e+00 -2.68791318e-02 -2.35761791e-01 1.94586352e-01 4.56409603e-01 1.18584168e+00 3.66882116e-01 4.31836575e-01 4.96064991e-01 -9.55353141e-01 -2.66658247e-01 -7.55720139e-01 -4.49070901e-01 6.59045875e-01 2.11714894e-01 -7.38099217e-01 -9.02836472e-02 3.01981002e-01]

[9.508354187011719, 0.6279402375221252]

f9adf6a4-4f30-4efb-a8f7-1ae8eb99b2d9

restricted-forensic-levenshtein-distance

2203.06138

https://arxiv.org/abs/2203.06138v3

https://arxiv.org/pdf/2203.06138v3.pdf

A New String Edit Distance and Applications

String edit distances have been used for decades in applications ranging from spelling correction and web search suggestions to DNA analysis. Most string edit distances are variations of the Levenshtein distance and consider only single-character edits. In forensic applications polymorphic genetic markers such as short tandem repeats (STRs) are used. At these repetitive motifs the DNA copying errors consist of more than just single base differences. More often the phenomenon of ``stutter'' is observed, where the number of repeated units differs (by whole units) from the template. To adapt the Levenshtein distance to be suitable for forensic applications where DNA sequence similarity is of interest, a generalized string edit distance is defined that accommodates the addition or deletion of whole motifs in addition to single-nucleotide edits. A dynamic programming implementation is developed for computing this distance between sequences. The novelty of this algorithm is in handling the complex interactions that arise between multiple- and single-character edits. Forensic examples illustrate the purpose and use of the Restricted Forensic Levenshtein (RFL) distance measure, but applications extend to sequence alignment and string similarity in other biological areas, as well as dynamic programming algorithms more broadly.

['Hari Iyer', 'Tunde I Huszar', 'Jan Hannig', 'Taylor Petty']

2022-03-11

['dna-analysis', 'spelling-correction']

['medical', 'natural-language-processing']

[ 8.60683322e-01 -3.48343283e-01 1.25466689e-01 -4.53258395e-01 -2.12277532e-01 -9.55735266e-01 4.60287601e-01 8.69076312e-01 -7.29718626e-01 8.17213178e-01 -2.79141694e-01 -3.81366879e-01 -2.03007817e-01 -8.32736313e-01 -4.15421307e-01 -9.01866317e-01 -3.26077431e-01 5.68544984e-01 5.99434972e-01 -1.98059946e-01 8.02581370e-01 9.73556757e-01 -1.57228589e+00 1.67405039e-01 8.79029572e-01 7.26839751e-02 2.22853616e-01 9.74523067e-01 -7.10370123e-01 -1.84102491e-01 -8.53898346e-01 -7.42389262e-01 2.20266819e-01 -5.33286691e-01 -4.70666200e-01 -4.23084050e-01 2.55438924e-01 -6.82025915e-04 2.30318069e-01 1.20012641e+00 5.43190956e-01 2.22903132e-01 6.94888413e-01 -8.27237844e-01 -5.02397895e-01 5.72200179e-01 -6.10962033e-01 4.19753045e-01 7.63830781e-01 7.75618199e-03 8.32037807e-01 -7.78800845e-01 8.60908806e-01 1.23213971e+00 8.74270022e-01 1.23368777e-01 -1.21969748e+00 -3.96603048e-02 -5.21684408e-01 4.00163710e-01 -1.35037172e+00 1.98202237e-01 3.19780022e-01 -6.44959152e-01 1.19052374e+00 7.44726241e-01 5.19649625e-01 5.26417792e-01 6.67451084e-01 2.33155534e-01 7.14128613e-01 -7.78276443e-01 2.19737068e-01 -3.10450166e-01 2.19348058e-01 3.16902310e-01 6.80069566e-01 -1.87131703e-01 -2.56288290e-01 -5.25153697e-01 3.02497834e-01 1.78939611e-01 2.19356745e-01 -3.09270117e-02 -1.08783913e+00 9.35273528e-01 -4.31252092e-01 6.59842312e-01 -7.86902830e-02 -8.18711892e-02 8.18995833e-01 2.45910078e-01 1.27257869e-01 3.23558182e-01 9.72980484e-02 -6.13351882e-01 -1.03461039e+00 3.30902398e-01 8.75192583e-01 7.47565389e-01 7.92028487e-01 -2.72986084e-01 4.70668487e-02 1.01982427e+00 -3.52565348e-02 2.60877311e-01 5.95477998e-01 -7.54777908e-01 5.72908670e-02 3.99172425e-01 -1.07873276e-01 -1.13167024e+00 -9.34386775e-02 1.60692543e-01 -6.31127357e-01 3.21062326e-01 5.27829528e-01 4.17499900e-01 -5.35067201e-01 1.36202800e+00 4.92747754e-01 1.28708370e-02 -4.12649095e-01 4.38013047e-01 2.91381180e-01 5.79404950e-01 -1.96909398e-01 -4.83270884e-01 1.36765516e+00 -2.43569613e-01 -5.58208644e-01 1.56985730e-01 6.78402722e-01 -9.65377927e-01 7.75020778e-01 2.67217338e-01 -1.18964338e+00 -1.91746920e-01 -9.05237913e-01 -1.61453456e-01 -8.08684111e-01 -6.91131234e-01 7.22765997e-02 1.09417295e+00 -7.24260092e-01 1.06164968e+00 -6.22612953e-01 -5.48649073e-01 -2.43864119e-01 2.42687568e-01 -4.44229513e-01 1.70029506e-01 -1.01516819e+00 1.02956069e+00 5.77904165e-01 1.15110777e-01 8.08009580e-02 -6.00132883e-01 -6.19418204e-01 -6.46095872e-02 6.37070164e-02 -2.13685632e-01 6.41906619e-01 -7.97647536e-01 -1.28807878e+00 1.37928820e+00 -1.09782368e-01 -4.69933182e-01 7.53548741e-01 2.45013252e-01 -3.93775314e-01 1.54636368e-01 -1.69241728e-04 -3.37027647e-02 4.08967286e-01 -4.18256491e-01 -2.19058141e-01 -2.07598329e-01 -3.19115251e-01 -2.57714242e-01 1.37137637e-01 4.51565892e-01 -4.11560200e-02 -1.03171003e+00 -1.88903973e-01 -8.89021337e-01 -3.99354398e-02 2.81715065e-01 3.70028391e-02 -1.07907824e-01 5.12329102e-01 -9.13330317e-01 1.31653965e+00 -2.17780089e+00 1.98274359e-01 6.50698721e-01 -1.66904509e-01 6.83194578e-01 -8.40324238e-02 1.14505720e+00 -3.40048820e-01 8.67092535e-02 -9.75144088e-01 2.88390368e-01 -2.54246164e-02 5.02162695e-01 -3.35582606e-02 7.55924940e-01 -7.03689009e-02 2.89161891e-01 -1.15610087e+00 -5.17344475e-01 6.18224218e-03 2.05620661e-01 -3.59841019e-01 -1.44440040e-01 -1.28126633e-03 -2.20135469e-02 -1.22391293e-03 3.63278657e-01 8.94044042e-01 4.13909972e-01 4.25426185e-01 2.05422431e-01 -5.00171840e-01 1.42555311e-01 -1.21982062e+00 1.45793808e+00 -4.09742296e-02 5.98098874e-01 -3.58168602e-01 -9.57919300e-01 1.37716603e+00 -1.91948295e-01 3.38153660e-01 -2.97763079e-01 -2.64889807e-01 4.11460668e-01 3.50902081e-01 -5.84207773e-01 7.32809663e-01 -4.27510202e-01 1.92283213e-01 5.81913948e-01 -4.62131441e-01 -1.78140149e-01 7.35814393e-01 7.60671403e-03 1.23782420e+00 -5.37840091e-02 8.11122537e-01 -3.49010378e-01 9.24506307e-01 -2.23062158e-01 6.90738261e-01 7.41282940e-01 -1.27223685e-01 5.51574647e-01 6.26614749e-01 -2.06410468e-01 -1.50141370e+00 -1.18299985e+00 -3.30523252e-01 9.41646457e-01 -1.74538299e-01 -3.56348991e-01 -7.13711202e-01 -1.63698241e-01 3.91642988e-01 7.94489086e-01 -5.43915451e-01 -2.47481033e-01 -8.93304944e-01 -6.83932364e-01 1.13126194e+00 1.75520971e-01 1.28677458e-01 -1.15406764e+00 -8.66037846e-01 4.74249601e-01 1.42054647e-01 -4.04791534e-01 -7.34581351e-01 -5.47117880e-03 -1.01293039e+00 -9.73915458e-01 -6.89069748e-01 -4.90679771e-01 3.65895033e-01 -3.00098676e-02 5.15217841e-01 3.95256728e-01 -1.14653134e+00 2.88094819e-01 -5.71865916e-01 -2.49032766e-01 -8.65778446e-01 -3.14911842e-01 8.02636296e-02 -1.56202003e-01 5.48473656e-01 -7.14729130e-01 -2.53914446e-01 3.89555484e-01 -1.39304864e+00 -6.72361374e-01 3.02725494e-01 7.93053567e-01 4.78028268e-01 -2.82883435e-01 2.20409364e-01 -8.87247562e-01 7.06939697e-01 -3.22667032e-01 -4.46640998e-01 4.69628811e-01 -3.06149989e-01 1.28424287e-01 6.09170556e-01 -4.46603566e-01 -7.39990115e-01 -3.20255637e-01 -2.98786938e-01 1.13182403e-01 -1.74128953e-02 5.90409756e-01 -7.26127326e-02 -1.88867837e-01 5.69318295e-01 6.39701605e-01 3.82360965e-01 -5.61854482e-01 7.11730868e-02 8.00668836e-01 6.68361068e-01 -6.02587044e-01 4.21240628e-01 4.25936699e-01 3.82365942e-01 -1.18364894e+00 3.43258440e-01 -5.35889924e-01 -9.49628592e-01 -1.76353306e-01 6.78030372e-01 5.89299574e-02 -7.86026120e-01 5.66770434e-01 -1.17905509e+00 1.07801117e-01 -4.08161074e-01 2.18929112e-01 -6.14811897e-01 1.57430124e+00 -5.77025890e-01 -7.48130441e-01 -1.38215095e-01 -1.01891136e+00 6.56939566e-01 5.67757487e-02 -4.45169657e-01 -1.03402305e+00 6.64105952e-01 -2.12724224e-01 9.01211202e-02 6.41413689e-01 1.23910189e+00 -8.89705598e-01 9.48397815e-02 -5.07835448e-01 2.81896859e-01 1.54283091e-01 2.42698401e-01 6.50880873e-01 -2.34103918e-01 -2.06202611e-01 -1.24983087e-01 5.18310964e-01 5.54917514e-01 -1.31206870e-01 1.01916337e+00 -1.19481981e-01 -1.96706742e-01 3.98435682e-01 1.23046577e+00 6.93295956e-01 1.14282525e+00 7.04127133e-01 2.83503532e-01 8.09043825e-01 7.39229023e-01 6.36697829e-01 -4.49053168e-01 9.22764421e-01 1.71400964e-01 6.11436248e-01 2.42279023e-01 1.52183756e-01 4.65479255e-01 6.74355388e-01 -2.53383458e-01 -6.02453807e-03 -9.17958915e-01 4.24744517e-01 -1.59720266e+00 -1.52865303e+00 -5.85073829e-01 2.80381536e+00 1.04076111e+00 -5.43405078e-02 2.58013755e-01 4.13570434e-01 1.36882687e+00 1.64282650e-01 -6.14669621e-01 -1.33469260e+00 -3.28241140e-01 3.46888989e-01 7.11492181e-01 6.35303080e-01 -4.34916914e-01 4.87402767e-01 6.44937420e+00 1.02343976e+00 -8.26394975e-01 -1.91459909e-01 -1.37223616e-01 1.79719031e-02 -3.98652822e-01 1.28135487e-01 -6.06964171e-01 1.08312261e+00 9.34600055e-01 -3.16598713e-01 2.75706947e-01 5.99638939e-01 1.19254000e-01 -3.53897721e-01 -9.40499187e-01 1.07178855e+00 1.37602746e-01 -1.12431431e+00 6.59299120e-02 3.31706226e-01 1.48670390e-01 -3.97929817e-01 -2.68898785e-01 -5.31318367e-01 4.21558954e-02 -6.49803519e-01 5.95405877e-01 6.68903768e-01 8.52073252e-01 -8.43715310e-01 5.54998338e-01 1.99926691e-03 -1.03816032e+00 2.38737404e-01 -7.10338473e-01 1.67448789e-01 4.93200004e-01 1.13866091e+00 -8.04180324e-01 4.29076552e-01 3.01062852e-01 5.96837282e-01 -3.53533983e-01 1.56304276e+00 -8.16587508e-02 1.17197774e-01 -4.75932658e-01 -2.94814914e-01 1.76155910e-01 -9.46889818e-01 1.02624869e+00 1.95520401e+00 6.13382399e-01 3.03428769e-02 -4.80173469e-01 5.34037292e-01 4.14204210e-01 2.34676257e-01 -5.66891849e-01 -2.73202807e-01 6.10492945e-01 8.05975676e-01 -8.89757395e-01 -3.08771610e-01 -1.51400238e-01 1.34718370e+00 1.14397320e-04 -2.15395212e-01 -7.53100932e-01 -1.23372769e+00 1.08001983e+00 1.30623877e-01 4.74367976e-01 -5.25523543e-01 -6.95601106e-02 -6.49349034e-01 1.92591324e-01 -8.48063827e-01 5.21669507e-01 -4.20672268e-01 -1.14240718e+00 7.47496216e-03 1.75702587e-01 -1.06831872e+00 -4.15423185e-01 -6.65934503e-01 -8.12917233e-01 8.61717939e-01 -8.63267362e-01 -4.64650095e-01 2.27016449e-01 1.54233709e-01 3.56840402e-01 -9.09112860e-03 6.48284674e-01 2.85682738e-01 -3.15509051e-01 6.88183069e-01 8.97056520e-01 -1.92476511e-01 8.92651379e-01 -1.22461057e+00 6.29663765e-01 8.23896825e-01 -2.60211915e-01 1.06196427e+00 1.19143021e+00 -1.07310963e+00 -1.29671335e+00 -6.18375659e-01 1.28080809e+00 -5.17919995e-02 6.98098540e-01 -3.57710212e-01 -1.19707441e+00 5.36380351e-01 -1.93594575e-01 -5.17957926e-01 1.10874248e+00 -4.26073730e-01 -6.40571415e-01 1.80356979e-01 -1.54790688e+00 6.14600658e-01 1.08610809e+00 -5.35914242e-01 -7.51125574e-01 3.04820925e-01 6.97143525e-02 1.07314216e-03 -9.78257954e-01 -1.47132844e-01 9.09955561e-01 -1.22672009e+00 1.10663223e+00 -5.47508538e-01 7.47152343e-02 -5.53100824e-01 -3.77944931e-02 -8.67780745e-01 -3.31200957e-01 -8.15938711e-01 4.06691015e-01 1.13893008e+00 -2.14949325e-02 -8.55274796e-01 4.84099209e-01 2.17989936e-01 -1.65301442e-01 -1.21670872e-01 -1.27026117e+00 -1.28091085e+00 6.69853464e-02 1.05458066e-01 6.59692287e-01 9.74497914e-01 4.34958220e-01 -5.51344573e-01 -3.56445879e-01 -3.03068757e-01 7.05063760e-01 -1.76548839e-01 5.30929923e-01 -1.28174329e+00 -4.84310180e-01 -6.92917049e-01 -1.07382667e+00 -5.75796545e-01 -9.55938101e-02 -1.04544353e+00 2.51268595e-02 -8.06327760e-01 5.31971604e-02 -1.78443521e-01 2.00776309e-01 7.73014352e-02 -1.00424141e-01 1.17975049e-01 1.82896480e-01 1.12810642e-01 1.84363335e-01 -5.35420328e-02 3.75157326e-01 1.21066468e-02 -7.74473399e-02 -1.39739424e-01 -5.62094301e-02 5.32221258e-01 5.99562645e-01 -7.17308223e-01 1.22979701e-01 -1.29878551e-01 5.97158134e-01 -2.65114933e-01 -2.49700006e-02 -5.47066927e-01 2.13451102e-01 -3.59480113e-01 -2.07937524e-01 -4.92949933e-01 8.59290212e-02 -5.89114249e-01 7.66422629e-01 9.45078313e-01 -2.36650765e-01 5.38898051e-01 -8.50974843e-02 6.85811698e-01 9.37022343e-02 -1.22108114e+00 9.66026485e-01 -2.17511669e-01 -4.85067189e-01 -1.84572041e-01 -9.11829948e-01 -4.89278167e-01 1.26759911e+00 -1.13756967e+00 -2.07347631e-01 6.29619360e-02 -7.49207795e-01 -3.82484943e-01 9.13103163e-01 9.79834646e-02 6.53112292e-01 -1.00736666e+00 -7.15687394e-01 4.73741330e-02 1.63469225e-01 -5.66172719e-01 2.61846989e-01 8.39302123e-01 -1.46030831e+00 1.10742718e-01 -6.78334534e-01 -4.36356157e-01 -1.91854203e+00 6.49890423e-01 8.61051977e-02 -2.14854740e-02 -7.25010276e-01 6.42961323e-01 -3.78049493e-01 -3.22375864e-01 -9.45916101e-02 -3.96582298e-02 1.38567552e-01 2.85292953e-01 9.70341861e-01 1.01901412e+00 2.32474372e-01 -6.45192087e-01 -5.31728923e-01 9.12784278e-01 -2.74278134e-01 -8.63616094e-02 1.26856506e+00 -3.70712765e-02 -8.42568159e-01 6.12514257e-01 1.21447122e+00 2.18602389e-01 -5.34676552e-01 1.97745413e-01 3.81122887e-01 -9.47452903e-01 -9.41736639e-01 -3.50692689e-01 -2.93102562e-01 8.54414523e-01 4.35717940e-01 2.27098227e-01 6.58229411e-01 -3.46368074e-01 5.91875196e-01 4.06840175e-01 2.63370782e-01 -1.18404543e+00 -4.17971283e-01 4.94654775e-01 7.84195185e-01 -3.00978035e-01 4.24073748e-02 -3.75631273e-01 -2.28780359e-01 1.54438186e+00 -1.35124773e-01 -2.11179882e-01 1.64024904e-01 3.52923632e-01 -4.46443975e-01 1.91709399e-01 -6.54763132e-02 1.56265616e-01 -3.20609093e-01 7.72016168e-01 5.85874557e-01 -1.02320001e-01 -1.47699499e+00 -3.32015395e-01 -1.54535353e-01 -9.06963795e-02 7.89832354e-01 1.32520723e+00 -7.65059948e-01 -1.73932970e+00 -6.59327984e-01 4.36446965e-01 -2.92904675e-01 -1.47174358e-01 -6.38945282e-01 4.85738635e-01 1.18624948e-01 4.54454988e-01 3.34753871e-01 1.81327923e-03 1.64379776e-01 3.57042223e-01 6.81929350e-01 -4.66839015e-01 -9.71340358e-01 -2.23934785e-01 -4.23988979e-03 -2.71435708e-01 -2.89676636e-01 -1.25231409e+00 -1.15142465e+00 -7.84938455e-01 -1.47267014e-01 3.47862571e-01 9.38792884e-01 8.01059425e-01 1.20996222e-01 -9.20018777e-02 3.04320753e-01 -5.05033851e-01 -6.19995892e-01 -6.74689531e-01 -1.04844189e+00 4.26746070e-01 -3.25256214e-02 -4.30709243e-01 -3.39519829e-01 1.59998372e-01]

[4.909133434295654, 5.192310810089111]

ae6d9ebd-c864-4099-8f5d-3f2db8a36c8c

anea-distant-supervision-for-low-resource

2102.13129

https://arxiv.org/abs/2102.13129v2

https://arxiv.org/pdf/2102.13129v2.pdf

ANEA: Distant Supervision for Low-Resource Named Entity Recognition

Distant supervision allows obtaining labeled training corpora for low-resource settings where only limited hand-annotated data exists. However, to be used effectively, the distant supervision must be easy to gather. In this work, we present ANEA, a tool to automatically annotate named entities in texts based on entity lists. It spans the whole pipeline from obtaining the lists to analyzing the errors of the distant supervision. A tuning step allows the user to improve the automatic annotation with their linguistic insights without labelling or checking all tokens manually. In six low-resource scenarios, we show that the F1-score can be increased by on average 18 points through distantly supervised data obtained by ANEA.

['Dietrich Klakow', 'Lukas Lange', 'Michael A. Hedderich']

2021-02-25

['low-resource-named-entity-recognition']

['natural-language-processing']

[-8.35129693e-02 5.06036341e-01 -2.66107589e-01 -7.21744776e-01 -1.21709561e+00 -1.03004670e+00 4.08705175e-01 5.56041121e-01 -7.42248297e-01 1.07009017e+00 1.74906403e-01 -3.43332946e-01 -2.41269413e-02 -2.92939395e-01 -6.59954906e-01 -2.27308005e-01 2.29404554e-01 8.53014648e-01 2.95008123e-01 -1.27245143e-01 -2.58555375e-02 3.01600307e-01 -1.03304923e+00 3.79923165e-01 9.27880943e-01 4.54587668e-01 2.22220287e-01 3.65706682e-01 -3.22114557e-01 7.59834886e-01 -5.87054372e-01 -7.92613924e-01 2.13959962e-01 -1.16641544e-01 -1.33925593e+00 -2.40978464e-01 3.68263751e-01 -1.19114950e-01 5.13388515e-01 1.01119792e+00 4.63288814e-01 1.39753729e-01 3.78655523e-01 -8.37951660e-01 -4.31189150e-01 1.36758113e+00 -1.59674257e-01 1.19575135e-01 4.37195718e-01 -1.31302148e-01 1.36601603e+00 -1.09714544e+00 9.42025065e-01 7.90099561e-01 8.28451991e-01 4.12841409e-01 -1.13735092e+00 -5.21306932e-01 2.64613740e-02 1.43256495e-02 -1.41232705e+00 -5.54234326e-01 2.30861321e-01 -3.86410296e-01 1.15387917e+00 8.65515918e-02 -9.95448977e-02 7.62803137e-01 -7.51980424e-01 8.75939488e-01 8.33358526e-01 -9.56341624e-01 1.07410019e-02 5.46131790e-01 4.11943704e-01 5.67250490e-01 2.59394884e-01 -4.56235081e-01 -6.01785839e-01 -1.26523569e-01 1.13284856e-01 -5.28446734e-01 -2.00460479e-01 -1.51494965e-01 -1.15458417e+00 5.03984928e-01 3.76434565e-01 8.33516300e-01 -2.45292664e-01 -3.63490164e-01 6.45556509e-01 1.54546693e-01 6.49765909e-01 8.43967617e-01 -1.19220138e+00 -1.69338658e-01 -9.94723260e-01 -3.01428646e-01 8.52639139e-01 1.25856411e+00 7.90245771e-01 -3.63646835e-01 -5.44969924e-02 8.26879561e-01 1.62349284e-01 3.76469016e-01 4.05524045e-01 -7.51026988e-01 7.71703839e-01 6.53691292e-01 4.51471329e-01 -4.18219656e-01 -4.64920640e-01 -2.69774675e-01 -3.72714132e-01 -3.39031667e-01 8.82538259e-01 -4.29819584e-01 -4.38522011e-01 1.54625356e+00 5.61787724e-01 -1.41170278e-01 1.17510132e-01 5.38529336e-01 8.34996700e-01 3.10864538e-01 3.34939390e-01 -1.25465617e-01 1.39442873e+00 -1.18873358e+00 -9.84695017e-01 -2.01320291e-01 1.40445006e+00 -9.03779209e-01 1.22258139e+00 7.67596886e-02 -1.01980960e+00 -5.18862188e-01 -6.15903616e-01 -3.92417252e-01 -5.49757838e-01 5.81786752e-01 3.45730811e-01 3.96631390e-01 -8.24355662e-01 7.41202056e-01 -7.93144047e-01 -2.88094431e-01 1.50653198e-01 4.05581295e-01 -8.61986697e-01 4.00221124e-02 -1.51503813e+00 1.12585735e+00 8.26749802e-01 2.89370436e-02 -3.08607459e-01 -8.51088703e-01 -1.03081322e+00 1.87437132e-01 5.55988669e-01 -3.63332987e-01 1.49830306e+00 -7.64140546e-01 -1.13679957e+00 1.17573225e+00 -2.67911583e-01 -4.48179990e-01 5.55675507e-01 -3.86260539e-01 -2.05300122e-01 -1.28416136e-01 5.83212137e-01 4.32831675e-01 2.93835625e-02 -8.84211659e-01 -6.82347417e-01 -2.51292378e-01 3.96441594e-02 1.68137610e-01 -2.98456073e-01 4.69654560e-01 -3.84328127e-01 -4.26680863e-01 -3.15270096e-01 -8.71610403e-01 -2.08439872e-01 -3.99683654e-01 -4.90099818e-01 -5.81344604e-01 4.87827927e-01 -9.05906498e-01 1.24727190e+00 -2.19946003e+00 -1.63996786e-01 1.70783605e-02 6.87528476e-02 5.34333766e-01 2.39461794e-01 2.56834000e-01 -1.89903930e-01 4.59553450e-01 -1.58686310e-01 -5.99567592e-01 1.66932121e-01 1.81261212e-01 -1.72492731e-02 2.23625347e-01 3.53837669e-01 7.96941042e-01 -1.22813106e+00 -7.47609556e-01 -4.70256545e-02 2.55414128e-01 -3.26526016e-01 3.79542619e-01 -7.01045841e-02 4.54682559e-01 -2.82618940e-01 3.30091774e-01 3.98010254e-01 -3.83228630e-01 3.99457723e-01 -2.57103853e-02 -1.50605306e-01 6.80553794e-01 -1.13935947e+00 1.68571293e+00 -6.79257691e-01 6.08333468e-01 1.38662264e-01 -6.82296216e-01 7.95356631e-01 5.57424545e-01 4.37349565e-02 -2.69341767e-01 1.07288651e-01 6.71563625e-01 -2.51051515e-01 -3.91937703e-01 5.27029693e-01 -2.54525006e-01 -3.68289083e-01 5.15063405e-01 4.62498069e-01 3.69442135e-01 4.63005722e-01 2.61058450e-01 1.08277380e+00 2.09104851e-01 8.13409746e-01 -1.85574859e-01 6.91195726e-01 3.43934834e-01 6.01749837e-01 4.82218057e-01 -2.86706686e-01 1.62887186e-01 2.30737403e-01 -2.06497222e-01 -1.04627240e+00 -5.88531792e-01 -5.31098127e-01 1.56125569e+00 -4.02989030e-01 -5.24236739e-01 -8.46861899e-01 -1.28661251e+00 -2.79192567e-01 9.45884407e-01 -5.65178514e-01 3.15596879e-01 -6.40633881e-01 -2.22668633e-01 8.74341309e-01 6.05707884e-01 4.06024873e-01 -1.11705315e+00 -3.00671440e-02 1.56383738e-01 -5.70335209e-01 -1.38765335e+00 -3.57176036e-01 6.19199097e-01 -5.10117590e-01 -7.55396485e-01 -4.10141617e-01 -8.39008510e-01 7.38364697e-01 -2.12305605e-01 1.33631802e+00 3.00324783e-02 2.16121450e-01 -2.24974886e-01 -4.86755878e-01 -2.88988054e-01 -5.05021989e-01 6.18063688e-01 -2.30514854e-02 -3.58871371e-01 8.41160357e-01 -1.08864531e-01 -8.17483664e-02 3.86920452e-01 -5.45901418e-01 -1.90268174e-01 4.14058238e-01 8.72220933e-01 6.25214934e-01 -1.52509764e-01 6.00850224e-01 -1.52977943e+00 6.59481138e-02 -5.65897524e-01 -4.37675327e-01 4.61308628e-01 -3.50177735e-01 2.38420382e-01 8.44143569e-01 -6.68047462e-03 -1.43708694e+00 3.48127365e-01 -4.31259334e-01 -1.25410361e-02 -6.22951508e-01 5.11763036e-01 -3.81390959e-01 4.51331407e-01 8.42107952e-01 -4.66125548e-01 -8.37624669e-01 -8.59647512e-01 5.53410232e-01 1.02469254e+00 6.58564448e-01 -5.94231904e-01 7.27766633e-01 1.48436993e-01 -3.44807565e-01 -4.74382102e-01 -1.77180779e+00 -8.60856354e-01 -1.48456204e+00 1.77641958e-01 8.08191657e-01 -9.75672603e-01 -2.76893497e-01 -1.49407908e-01 -1.05599773e+00 -3.87392849e-01 -4.75249529e-01 4.88720566e-01 -8.95978138e-02 1.72515810e-01 -5.52291155e-01 -5.27225137e-01 -2.33981669e-01 -6.76461518e-01 1.13173413e+00 1.04836300e-02 -6.78264081e-01 -1.36272311e+00 1.91614311e-03 6.75294399e-01 3.12498957e-02 1.73764918e-02 5.14103770e-01 -1.69982755e+00 -1.55848294e-01 -2.63143480e-01 2.44350117e-02 2.50188380e-01 2.85277873e-01 5.52646769e-03 -1.14324081e+00 -1.25212856e-02 -4.49744105e-01 -5.38021207e-01 2.94422567e-01 -1.07252419e-01 7.78760076e-01 -3.01116496e-01 -4.03843999e-01 9.43850651e-02 1.16482019e+00 -2.79303998e-01 5.85682578e-02 6.25120878e-01 7.15316713e-01 9.78604853e-01 1.02364159e+00 1.46712571e-01 2.48858824e-01 6.32811189e-01 -2.26279214e-01 -2.02271551e-01 -2.59569716e-02 -5.32006919e-01 1.24077320e-01 9.13744211e-01 2.64724523e-01 -6.56061620e-02 -1.26919913e+00 9.10467505e-01 -1.68331218e+00 -8.52405488e-01 -4.88877237e-01 2.04087400e+00 1.39228368e+00 1.64776519e-01 -7.09857121e-02 6.97841421e-02 1.09559798e+00 -3.87285709e-01 -1.66418597e-01 -2.90634453e-01 -1.64953247e-02 3.78056467e-02 5.02407312e-01 7.64975727e-01 -1.15680325e+00 1.34228039e+00 6.47694445e+00 8.91274989e-01 -5.29142082e-01 4.23050046e-01 4.22737241e-01 -1.66368894e-02 -2.14479491e-01 1.84033364e-01 -1.35600805e+00 3.50485146e-01 1.42173398e+00 -2.27182090e-01 -1.39046669e-01 1.09682918e+00 1.78226084e-01 4.18080389e-03 -1.31570506e+00 4.60307240e-01 -2.02606007e-01 -1.04298329e+00 -3.54277164e-01 -2.19647914e-01 8.60591769e-01 4.10748035e-01 -4.51701671e-01 5.30786872e-01 8.57687950e-01 -7.09634185e-01 6.49589717e-01 1.09172903e-01 9.21165824e-01 -7.51320601e-01 1.25246274e+00 7.75383055e-01 -1.01890266e+00 3.10421288e-01 -2.50678688e-01 9.08282772e-02 4.30365592e-01 8.73638570e-01 -1.35700440e+00 6.37176335e-01 6.83136404e-01 5.36988080e-01 -6.73058808e-01 8.04475605e-01 -7.43094981e-01 8.19748223e-01 -4.81769323e-01 7.43293464e-02 3.27458471e-01 9.31264758e-02 2.75545329e-01 1.59691405e+00 9.30905715e-02 -9.97318178e-02 2.70952672e-01 5.97706378e-01 -5.32645464e-01 5.49483597e-01 -6.54943049e-01 1.83807641e-01 6.58641219e-01 1.40802860e+00 -7.07034171e-01 -6.01517856e-01 -3.24764460e-01 8.50328982e-01 8.96897495e-01 -1.19362630e-01 -5.90906441e-01 -6.09626055e-01 6.98389038e-02 -2.16621514e-02 4.84260172e-01 1.11379558e-02 -2.63654232e-01 -1.18752849e+00 1.84558034e-01 -7.45098829e-01 6.70753181e-01 -8.33055794e-01 -1.32359242e+00 7.58839607e-01 -1.09832376e-01 -9.25326407e-01 -4.01670069e-01 -5.54422855e-01 -9.92017612e-02 9.32042658e-01 -1.56201935e+00 -1.06234562e+00 -2.87118312e-02 3.21423024e-01 4.03946698e-01 1.77091390e-01 8.78896773e-01 6.03227556e-01 -6.50263190e-01 7.17225671e-01 1.61016837e-01 7.02432096e-01 1.34038436e+00 -1.64205444e+00 3.98953766e-01 8.31293464e-01 4.35327679e-01 9.58677173e-01 6.91841722e-01 -7.08370209e-01 -3.57995927e-01 -1.13100946e+00 1.76891065e+00 -8.52954984e-01 1.11234057e+00 -3.52045506e-01 -1.16883039e+00 1.15835607e+00 2.68800378e-01 2.29908854e-01 1.20713222e+00 6.33487046e-01 -4.32752788e-01 2.42105246e-01 -1.21526384e+00 9.12410542e-02 9.66149330e-01 -6.83188200e-01 -9.74356234e-01 5.33963919e-01 6.49524331e-01 -5.65780342e-01 -1.21586692e+00 2.01152489e-01 7.13292286e-02 -4.40382838e-01 4.17080224e-01 -7.81404495e-01 1.54736117e-01 -5.64390063e-01 -7.94315897e-03 -1.36643660e+00 -7.64360651e-02 -4.10669446e-01 3.56517315e-01 1.92556179e+00 1.05794728e+00 -3.60658914e-01 3.65489751e-01 7.79165268e-01 -1.34930238e-01 -2.57559102e-02 -7.59262502e-01 -7.89093852e-01 1.52164087e-01 -4.09224808e-01 6.49358928e-01 1.48487461e+00 4.56633121e-01 7.38718867e-01 -7.07945600e-02 4.24124718e-01 2.44274810e-01 -7.95686767e-02 6.08358443e-01 -1.31111848e+00 -2.44245544e-01 7.64459670e-02 -7.78043689e-03 -9.73277092e-01 5.97447753e-01 -1.18766296e+00 2.70624310e-01 -1.33669865e+00 2.58957535e-01 -7.71779120e-01 -3.72616984e-02 9.98765051e-01 -4.99046713e-01 1.34160116e-01 -1.14599511e-01 2.21897036e-01 -1.12411070e+00 1.32351533e-01 7.20085859e-01 2.45280892e-01 -1.47292679e-02 -6.48558959e-02 -7.14198232e-01 1.02826858e+00 7.91560173e-01 -9.05828238e-01 2.17820574e-02 -6.56525970e-01 5.02587974e-01 -3.50112706e-01 -2.25035772e-01 -6.30755246e-01 3.62247884e-01 1.37535840e-01 2.02041700e-01 -5.20876944e-01 -3.70257437e-01 -1.00398982e+00 -1.36878490e-01 -7.66264275e-03 -6.80053174e-01 -1.01358667e-01 2.03978464e-01 3.33228409e-01 -5.02986431e-01 -8.09139848e-01 6.18058860e-01 -2.19223619e-01 -5.76122224e-01 -1.92549825e-01 -2.36633882e-01 3.68559510e-01 1.04403222e+00 3.53665024e-01 -3.50539953e-01 5.35174422e-02 -1.09380245e+00 2.86302328e-01 4.11307931e-01 1.60170570e-02 -3.51727068e-01 -1.25169945e+00 -8.34522307e-01 -1.75630882e-01 4.16418523e-01 2.74539232e-01 -2.41467077e-02 6.39108121e-01 -3.99033368e-01 7.28081584e-01 8.60222876e-02 -5.75845659e-01 -1.62869513e+00 5.33756137e-01 1.91692874e-01 -8.20108175e-01 -2.28487059e-01 8.62906575e-01 -2.74164468e-01 -1.14275682e+00 7.13420361e-02 -6.20333813e-02 -5.49821258e-01 2.48279825e-01 6.25467360e-01 2.82880574e-01 5.72949588e-01 -8.43116820e-01 -4.28467870e-01 1.77708045e-01 -1.25296816e-01 -4.95473295e-02 1.36599028e+00 -3.24361712e-01 -1.21835805e-01 6.61235332e-01 1.05684268e+00 5.20949244e-01 -7.55766690e-01 -5.70732415e-01 7.17962503e-01 -3.86849791e-01 -7.28568584e-02 -8.53210032e-01 -6.58455849e-01 7.17975616e-01 -8.22093859e-02 3.35523784e-01 6.84730589e-01 3.16133082e-01 5.88214278e-01 8.08134556e-01 4.35030490e-01 -1.15326011e+00 -3.32556337e-01 7.37765551e-01 4.26489323e-01 -1.46145320e+00 -1.65408283e-01 -6.21755540e-01 -8.39924514e-01 7.72396266e-01 5.53623199e-01 4.43523407e-01 4.66806054e-01 3.34661782e-01 2.75822699e-01 -1.76683113e-01 -6.32129669e-01 -4.19432580e-01 3.62436175e-01 4.91939574e-01 8.69495869e-01 -8.85971636e-02 -2.04719484e-01 7.46014655e-01 -4.49060768e-01 -3.77238512e-01 3.43150258e-01 7.05621660e-01 -3.66412342e-01 -1.37240958e+00 -1.59711182e-01 6.25908598e-02 -9.78908360e-01 -3.24017256e-01 -6.11089349e-01 7.29200661e-01 2.34797120e-01 1.05578506e+00 -1.08646467e-01 2.77590066e-01 4.01691347e-01 4.68116641e-01 -9.43639502e-02 -1.07558584e+00 -8.98551643e-01 -1.58137321e-01 8.23256433e-01 -1.62747473e-01 -7.52564013e-01 -9.48111892e-01 -1.51942766e+00 -1.96467951e-01 -9.08236325e-01 7.31885016e-01 6.22464478e-01 1.25819862e+00 2.46992260e-01 3.67965907e-01 4.66799557e-01 -4.10597354e-01 -3.36385190e-01 -1.19313812e+00 -3.85951310e-01 5.80594778e-01 9.40930173e-02 -4.67853695e-01 -6.42417014e-01 3.30669045e-01]

[9.697881698608398, 9.245559692382812]

78567fbf-8799-4d17-b524-fdb7ad5dc593

mgtr-end-to-end-mutual-gaze-detection-with

2209.1093

https://arxiv.org/abs/2209.10930v2

https://arxiv.org/pdf/2209.10930v2.pdf

MGTR: End-to-End Mutual Gaze Detection with Transformer

People's looking at each other or mutual gaze is ubiquitous in our daily interactions, and detecting mutual gaze is of great significance for understanding human social scenes. Current mutual gaze detection methods focus on two-stage methods, whose inference speed is limited by the two-stage pipeline and the performance in the second stage is affected by the first one. In this paper, we propose a novel one-stage mutual gaze detection framework called Mutual Gaze TRansformer or MGTR to perform mutual gaze detection in an end-to-end manner. By designing mutual gaze instance triples, MGTR can detect each human head bounding box and simultaneously infer mutual gaze relationship based on global image information, which streamlines the whole process with simplicity. Experimental results on two mutual gaze datasets show that our method is able to accelerate mutual gaze detection process without losing performance. Ablation study shows that different components of MGTR can capture different levels of semantic information in images. Code is available at https://github.com/Gmbition/MGTR

['Jingtai Liu', 'Zhengxi Hu', 'Hang Guo']

2022-09-22

['mutual-gaze']

['computer-vision']

[-9.64009762e-02 9.46813822e-02 -2.27429047e-02 -4.62606549e-01 -9.60506871e-02 -2.13165253e-01 3.65044355e-01 -1.83527201e-01 -3.19620341e-01 1.10070765e-01 8.37119371e-02 -1.20230347e-01 -6.02617511e-05 -3.75953317e-01 -4.03211117e-01 -4.59475100e-01 3.15232962e-01 -1.01125045e-02 4.42545921e-01 -1.38528794e-01 4.54322249e-01 -1.59446627e-01 -2.04447699e+00 1.08796768e-01 7.45965302e-01 7.82821894e-01 3.58125895e-01 9.24227238e-01 -1.56602003e-02 1.04987061e+00 -2.97802955e-01 -5.68016410e-01 -8.69113281e-02 -5.67478776e-01 -7.79195070e-01 -2.49981776e-01 4.29520011e-01 -5.23693085e-01 4.19131108e-02 1.02884436e+00 4.56764936e-01 -2.05360502e-02 2.79768229e-01 -1.74759912e+00 -5.19374788e-01 5.17150164e-01 -1.28287327e+00 4.17299747e-01 8.20310414e-01 3.74064475e-01 1.16756678e+00 -7.62133777e-01 1.18559137e-01 1.44186294e+00 4.49434161e-01 6.91940069e-01 -7.58811831e-01 -1.24268878e+00 2.22346798e-01 4.09657091e-01 -1.46795523e+00 -6.91486776e-01 7.56357789e-01 -4.08130080e-01 5.41995287e-01 5.85749626e-01 4.22524512e-01 7.27526546e-01 -1.53603286e-01 1.19693971e+00 1.10446191e+00 -4.28093463e-01 -4.34537649e-01 1.46900624e-01 3.43584299e-01 8.73283207e-01 1.37064396e-03 -3.65094058e-02 -9.64930415e-01 2.20337078e-01 3.83390248e-01 1.37715220e-01 -4.41711128e-01 -1.26173586e-01 -1.17670703e+00 5.62429965e-01 7.78945625e-01 2.75464863e-01 1.18136987e-01 1.15815625e-01 1.62908789e-02 1.17241435e-01 5.93810320e-01 -5.55611588e-03 -2.00789981e-02 -2.99413592e-01 -8.00344944e-01 1.78058609e-01 6.44459426e-01 9.58626509e-01 1.00988626e+00 -1.04716122e+00 -2.35736147e-01 5.42510033e-01 8.77341688e-01 6.77315593e-01 3.38670671e-01 -4.84710753e-01 2.91160524e-01 9.10099030e-01 3.27816233e-03 -1.18273807e+00 -6.85968161e-01 1.37419522e-01 -2.98205167e-01 4.72593075e-03 5.33897460e-01 -6.74384683e-02 -5.29265404e-01 1.79413319e+00 7.61776388e-01 2.59823918e-01 -5.11189222e-01 1.22907090e+00 1.18127680e+00 1.87557653e-01 1.79538220e-01 -1.11341022e-01 1.86544204e+00 -1.06080580e+00 -8.55715871e-01 -3.92524123e-01 8.96468282e-01 -9.71248627e-01 1.30034268e+00 -3.35419588e-02 -1.06498170e+00 -4.35306162e-01 -7.13720858e-01 -5.68861365e-01 -5.31249940e-02 6.08254224e-02 4.45809573e-01 6.79157734e-01 -1.11528301e+00 1.27382115e-01 -7.44858921e-01 -5.63793004e-01 5.07902861e-01 4.90441144e-01 -1.17729709e-01 1.30980819e-01 -9.41561222e-01 6.54098928e-01 -2.91458108e-02 2.26698756e-01 -2.19665363e-01 -5.39201438e-01 -8.83253813e-01 1.02925844e-01 4.42001134e-01 -7.90798187e-01 1.53118420e+00 -9.59247470e-01 -1.35368180e+00 1.15119720e+00 -9.21533227e-01 4.94795069e-02 6.69929028e-01 -5.43411553e-01 -2.56133497e-01 7.27598220e-02 1.89320564e-01 8.12725663e-01 9.42101300e-01 -8.78843606e-01 -9.49856222e-01 -6.94218636e-01 1.31092459e-01 4.25358474e-01 -2.80677468e-01 6.40961111e-01 -7.59258986e-01 3.27981524e-02 -9.48666222e-03 -1.00331008e+00 3.31667781e-01 9.18831974e-02 -5.84404528e-01 -7.36314058e-01 1.03655446e+00 -3.72928649e-01 1.51793778e+00 -2.22135043e+00 -4.26998213e-02 -6.85411617e-02 8.79978657e-01 1.03725558e-02 3.13804209e-01 1.05872311e-01 -9.21839848e-02 4.23007682e-02 -7.23227635e-02 -9.90336895e-01 4.32308763e-02 -3.08818698e-01 8.25863928e-02 5.38650870e-01 4.64647189e-02 1.10030401e+00 -1.06839418e+00 -7.92758346e-01 1.66256458e-01 5.37666261e-01 -4.30310428e-01 4.56418633e-01 1.88742101e-01 5.47110558e-01 -2.86376834e-01 4.28635895e-01 8.84301603e-01 -6.63712204e-01 -7.45136142e-02 -1.37810975e-01 -2.73677170e-01 1.83513999e-01 -7.08279371e-01 1.52176547e+00 -2.32970268e-01 9.61164355e-01 -8.27567205e-02 -8.71145651e-02 5.92961133e-01 -4.98993322e-02 5.10510467e-02 -9.13247168e-01 6.16385937e-01 -3.06248724e-01 8.94177109e-02 -8.66193891e-01 4.96127307e-01 1.26336396e-01 9.79968831e-02 1.00221479e+00 -3.55274618e-01 6.34827614e-01 1.21439703e-01 3.19207162e-01 6.85443938e-01 6.73952177e-02 3.09134215e-01 -1.69497564e-01 6.56438708e-01 -4.49124813e-01 3.64705950e-01 2.79941171e-01 -6.59343958e-01 6.54379487e-01 7.11110175e-01 -2.63830507e-03 -4.00289983e-01 -6.61182106e-01 1.09890297e-01 1.66033113e+00 5.99026144e-01 -6.23507142e-01 -1.01280344e+00 -6.56450152e-01 -2.16969147e-01 5.62137604e-01 -1.11688876e+00 -1.09538607e-01 -4.25996929e-01 -4.27133471e-01 3.23011816e-01 1.88641623e-01 5.38181484e-01 -1.11108744e+00 -1.00469112e+00 -6.83478653e-01 -3.98996055e-01 -9.79297996e-01 -7.64517426e-01 -6.50395453e-01 -2.12929472e-01 -1.48828423e+00 -7.03232527e-01 -4.63889956e-01 6.93573058e-01 9.33850408e-01 9.65488613e-01 4.20286596e-01 -1.74624741e-01 2.67821491e-01 -3.32871199e-01 -5.67776442e-01 8.88938904e-02 2.35903189e-01 -1.55744836e-01 3.20298016e-01 1.04049671e+00 -2.13751331e-01 -1.08146381e+00 4.85004246e-01 -4.74176288e-01 4.27294999e-01 3.10022026e-01 3.41360241e-01 -1.03237763e-01 -4.77252334e-01 6.60903379e-02 -9.68998313e-01 5.62762201e-01 -6.11355960e-01 -4.98679847e-01 1.49200574e-01 -4.09834713e-01 -1.85273886e-01 -2.06926972e-01 -1.13493793e-01 -1.21280706e+00 -2.34608427e-01 2.17496455e-02 -5.60785234e-01 -2.41593435e-01 6.89542517e-02 -1.30213335e-01 1.76610976e-01 4.47243482e-01 -5.87375276e-02 2.92609662e-01 -4.79735970e-01 1.93864852e-01 8.11758637e-01 3.54055703e-01 1.14311554e-01 5.28856575e-01 5.34378111e-01 -3.42956066e-01 -5.88717341e-01 -1.19789147e+00 -8.51589620e-01 -6.46284223e-01 -5.68217874e-01 1.00613678e+00 -9.46807206e-01 -1.59514225e+00 7.25682199e-01 -9.58219886e-01 -1.94246382e-01 3.83006692e-01 2.42257729e-01 -1.05495468e-01 2.82708377e-01 -3.56891096e-01 -8.86286616e-01 -5.53635061e-01 -1.23691142e+00 1.41429627e+00 6.80124879e-01 -3.51715863e-01 -8.85189295e-01 1.19442843e-01 6.55710638e-01 1.26652122e-01 -1.95563495e-01 5.62810563e-02 -2.85517693e-01 -5.69363654e-01 8.29455331e-02 -7.13206232e-01 -2.33549833e-01 2.02961400e-01 2.50549525e-01 -1.31079280e+00 -1.87911659e-01 -1.03255816e-01 -5.29970489e-02 6.71253920e-01 2.15268925e-01 8.56283069e-01 -1.74251691e-01 -6.81533873e-01 5.72184622e-01 9.84694004e-01 -3.38007748e-01 5.17594814e-01 2.36826822e-01 1.27949202e+00 9.11740184e-01 7.86056459e-01 3.14704269e-01 1.07919383e+00 5.48118949e-01 5.82467079e-01 -2.28185743e-01 -2.62177698e-02 -1.75387651e-01 3.27051014e-01 5.15205920e-01 1.00733355e-01 -2.02018648e-01 -1.27685356e+00 4.72076118e-01 -1.96511173e+00 -9.68564510e-01 -6.83467746e-01 1.97941041e+00 6.65883660e-01 -7.07653090e-02 5.64849377e-01 8.96668062e-02 1.02930939e+00 8.66514072e-02 -4.76367384e-01 1.80489023e-03 4.63981211e-01 -3.16040248e-01 7.52676725e-02 3.73962522e-01 -9.69669700e-01 8.63079071e-01 5.54394054e+00 2.99494058e-01 -1.17316043e+00 3.69174153e-01 4.46011275e-01 -4.62624222e-01 1.55350175e-02 -1.64153129e-02 -9.46507275e-01 8.45125496e-01 7.20558405e-01 -8.24247152e-02 7.04473257e-02 5.95857859e-01 1.83398098e-01 -4.46043730e-01 -1.03786683e+00 1.45519185e+00 2.58700192e-01 -6.97873235e-01 -5.23798108e-01 6.04544505e-02 2.39965841e-01 1.16747096e-01 2.86042124e-01 -3.37392278e-02 6.20033443e-02 -8.12847376e-01 6.20596230e-01 6.01364076e-01 7.22384274e-01 -9.02552485e-01 5.16398132e-01 4.88909602e-01 -1.10549319e+00 -1.55927405e-01 9.53275710e-02 -3.48077983e-01 1.69480786e-01 3.47747684e-01 -9.23929751e-01 1.13937490e-01 1.16854870e+00 9.71474528e-01 -8.87373149e-01 1.01779759e+00 -4.11749393e-01 3.19770753e-01 -3.35578948e-01 -2.01040804e-01 -4.32626940e-02 -3.89282405e-03 5.59822857e-01 1.02962410e+00 1.12728365e-02 7.77166411e-02 -2.88946748e-01 9.27250922e-01 -2.26597600e-02 -1.05334550e-01 -2.86404878e-01 4.41919327e-01 4.03074980e-01 1.42892110e+00 -6.75479054e-01 -6.89818636e-02 -4.17836696e-01 9.93820131e-01 5.16156316e-01 2.37562601e-02 -1.15738142e+00 -4.03543383e-01 7.63845384e-01 2.91803807e-01 1.18730053e-01 2.10399777e-01 -2.47626588e-01 -1.12108767e+00 1.27057031e-01 -4.64651197e-01 4.23471093e-01 -1.05978918e+00 -9.02476251e-01 3.54189605e-01 -1.38666481e-01 -1.11412454e+00 -8.88514742e-02 -2.07401931e-01 -6.78750813e-01 9.66273308e-01 -1.59170759e+00 -1.33743203e+00 -1.03714943e+00 8.72310758e-01 3.18843186e-01 4.64349091e-01 4.17728394e-01 2.12709859e-01 -1.15476596e+00 9.56609488e-01 -6.00941956e-01 1.39954269e-01 1.02299571e+00 -1.13957882e+00 4.30211663e-01 9.04921770e-01 -1.69279188e-01 7.53456593e-01 7.28566766e-01 -3.42427254e-01 -1.02659655e+00 -7.25719154e-01 1.02632308e+00 -8.36238205e-01 6.31247044e-01 -4.50369745e-01 -8.48809719e-01 7.94530630e-01 4.87977564e-01 -1.99870080e-01 8.97580206e-01 5.70849121e-01 -3.89889538e-01 6.63803741e-02 -8.44136894e-01 7.30535865e-01 1.25289869e+00 -6.13897681e-01 -5.21298826e-01 1.39139310e-01 6.09735131e-01 -5.94374239e-01 -4.26769435e-01 2.13812441e-01 6.90063894e-01 -1.44674087e+00 5.57128966e-01 1.10060694e-02 4.57130104e-01 -5.09984076e-01 5.99148810e-01 -7.92142630e-01 -2.15227067e-01 -8.19709897e-01 -3.75100940e-01 1.35330808e+00 1.00700788e-01 -7.28570402e-01 5.11469483e-01 8.48275483e-01 4.06362385e-01 -5.43303311e-01 -4.13787216e-01 -7.25527853e-02 -5.85158706e-01 -2.70453781e-01 9.00783181e-01 9.36230123e-01 3.68732274e-01 8.52286458e-01 -2.59117156e-01 1.21253841e-01 6.39685512e-01 1.67101815e-01 1.19120944e+00 -1.23554206e+00 -2.90268417e-02 -6.07112527e-01 -4.56994325e-01 -1.26041198e+00 -5.88456392e-02 -3.71554017e-01 6.33164719e-02 -1.25979805e+00 5.98225594e-01 -4.50732797e-01 -3.40134323e-01 5.06108105e-01 -7.46481717e-01 4.47268158e-01 3.21105599e-01 5.77437460e-01 -1.04888403e+00 3.52888346e-01 1.36689806e+00 2.68417090e-01 -2.22891286e-01 -1.00606019e-02 -1.04668832e+00 7.89883375e-01 5.85229576e-01 -4.00007665e-01 -5.64468563e-01 -5.70008278e-01 5.50779641e-01 -2.68565416e-01 5.51832616e-01 -8.91661108e-01 7.25973725e-01 1.88423559e-01 9.87591222e-02 -8.42356026e-01 1.56408206e-01 -5.56293964e-01 -3.18627626e-01 1.10534281e-01 -8.37057456e-02 1.41231149e-01 7.47878551e-02 5.23513377e-01 -6.15259819e-02 4.31589559e-02 7.39046693e-01 1.83173910e-01 -6.60283804e-01 2.18056768e-01 1.97458625e-01 -1.28971547e-01 1.21344006e+00 -3.02823722e-01 -4.49636757e-01 -2.52102703e-01 -5.55830598e-01 5.46642125e-01 7.66805410e-01 7.32183397e-01 2.99834251e-01 -8.91777515e-01 -5.38998127e-01 1.76143661e-01 3.53797853e-01 1.77853405e-01 4.04719144e-01 1.37372744e+00 -3.01107228e-01 2.59396583e-01 -6.96641803e-02 -1.12275910e+00 -1.84995067e+00 5.55599034e-01 2.64644355e-01 -2.01397762e-02 -3.85319561e-01 1.41302001e+00 8.57169330e-01 -1.15402237e-01 -3.07709370e-02 -1.02668166e-01 -6.13142014e-01 2.37278193e-01 1.10774386e+00 2.61958867e-01 -1.79075524e-01 -1.19127023e+00 -5.14076173e-01 5.74586511e-01 -4.56702888e-01 1.91414565e-01 9.13265944e-01 -7.93735027e-01 -3.22393954e-01 5.19176543e-01 1.30491197e+00 -6.47462904e-02 -1.14465404e+00 -2.26049066e-01 -1.10238463e-01 -7.17781901e-01 1.70659542e-01 -3.87521446e-01 -1.19910991e+00 9.07550931e-01 6.73648894e-01 2.69755512e-01 1.37441194e+00 2.05499381e-01 7.63382971e-01 -1.65087998e-01 5.24883084e-02 -5.58712840e-01 -1.02801643e-01 2.58588761e-01 5.84130764e-01 -1.63772106e+00 -4.26243097e-02 -6.38137519e-01 -7.27052391e-01 6.09952688e-01 9.29668009e-01 1.29782021e-01 8.69591236e-01 -6.81350902e-02 8.06251541e-02 -5.84497392e-01 -7.05581009e-01 -5.56864500e-01 4.76900280e-01 1.17538191e-01 6.56478465e-01 6.21026866e-02 -9.09453407e-02 2.65917659e-01 -4.63810325e-01 -5.77746406e-02 3.77251729e-02 8.08227599e-01 -3.03637296e-01 -7.48802960e-01 -3.89668375e-01 2.47542590e-01 -4.28707510e-01 -8.73253867e-02 -3.94066125e-01 6.89380467e-01 5.02677299e-02 1.16630685e+00 5.03001332e-01 -4.78551865e-01 1.84351400e-01 -1.91284493e-01 4.55876797e-01 -5.02969742e-01 -5.50711393e-01 -1.33157402e-01 -1.44429341e-01 -9.84460235e-01 -7.13169873e-01 -8.67071211e-01 -1.10560024e+00 -6.83270395e-01 -5.72786093e-01 -9.31833237e-02 4.97596592e-01 9.96678889e-01 6.29239857e-01 3.74073833e-01 5.70244849e-01 -8.83783698e-01 1.86121479e-01 -1.14111221e+00 -1.27143338e-01 4.75222409e-01 7.28990078e-01 -9.09486473e-01 -3.27595890e-01 1.20242722e-01]

[14.093866348266602, 0.04494749382138252]

7cb390fa-413d-4be6-a847-c3b4500a42ab

towards-end-to-end-generative-modeling-of

2303.11251

https://arxiv.org/abs/2303.11251v3

https://arxiv.org/pdf/2303.11251v3.pdf

Towards End-to-End Generative Modeling of Long Videos with Memory-Efficient Bidirectional Transformers

Autoregressive transformers have shown remarkable success in video generation. However, the transformers are prohibited from directly learning the long-term dependency in videos due to the quadratic complexity of self-attention, and inherently suffering from slow inference time and error propagation due to the autoregressive process. In this paper, we propose Memory-efficient Bidirectional Transformer (MeBT) for end-to-end learning of long-term dependency in videos and fast inference. Based on recent advances in bidirectional transformers, our method learns to decode the entire spatio-temporal volume of a video in parallel from partially observed patches. The proposed transformer achieves a linear time complexity in both encoding and decoding, by projecting observable context tokens into a fixed number of latent tokens and conditioning them to decode the masked tokens through the cross-attention. Empowered by linear complexity and bidirectional modeling, our method demonstrates significant improvement over the autoregressive Transformers for generating moderately long videos in both quality and speed. Videos and code are available at https://sites.google.com/view/mebt-cvpr2023 .

['Seunghoon Hong', 'Chiheon Kim', 'Doyup Lee', 'Semin Kim', 'Jaehoon Yoo']

2023-03-20

http://openaccess.thecvf.com//content/CVPR2023/html/Yoo_Towards_End-to-End_Generative_Modeling_of_Long_Videos_With_Memory-Efficient_Bidirectional_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Yoo_Towards_End-to-End_Generative_Modeling_of_Long_Videos_With_Memory-Efficient_Bidirectional_CVPR_2023_paper.pdf

cvpr-2023-1

['video-generation']

['computer-vision']

[ 1.68524772e-01 4.21770439e-02 7.83363059e-02 -2.20856071e-01 -1.10811317e+00 -4.39798594e-01 5.78196585e-01 -4.68679518e-01 3.14589753e-03 5.88577628e-01 4.71347481e-01 -1.81799129e-01 2.70722806e-02 -5.71999192e-01 -1.35597646e+00 -7.86122143e-01 -3.08295429e-01 2.21621007e-01 -5.19635826e-02 3.31292301e-01 -1.69051178e-02 1.19893309e-02 -1.35468841e+00 5.48215330e-01 8.59531641e-01 1.10987723e+00 5.79459906e-01 1.06483507e+00 1.74245715e-01 1.41574037e+00 -1.13129556e-01 -4.03064251e-01 6.73164101e-03 -4.65414196e-01 -5.93187749e-01 5.32831857e-03 7.44846404e-01 -6.48514271e-01 -7.75941312e-01 6.37720108e-01 4.91024703e-01 -8.86281654e-02 4.11913037e-01 -1.05265808e+00 -8.29048097e-01 6.29261494e-01 -4.36334193e-01 3.10699970e-01 2.39840642e-01 1.19724788e-01 1.04859614e+00 -1.29643118e+00 4.06728953e-01 1.11955523e+00 6.32519245e-01 4.90206569e-01 -1.23196435e+00 -8.26626658e-01 3.52477163e-01 5.10623336e-01 -1.30897796e+00 -7.71059275e-01 5.28995097e-01 -5.74489474e-01 1.23503995e+00 2.11458683e-01 6.57077134e-01 1.34012949e+00 3.11835021e-01 8.77431870e-01 8.91613662e-01 -1.56436548e-01 -1.36377532e-02 -2.37242013e-01 -4.02347028e-01 9.38871384e-01 -3.78812313e-01 1.59853622e-01 -8.98757517e-01 6.23904318e-02 1.06797183e+00 7.27201775e-02 -1.84226349e-01 -1.14889547e-01 -1.18152046e+00 6.01256013e-01 3.53962272e-01 -1.64187744e-01 -4.40262616e-01 7.85512745e-01 2.49978572e-01 1.58487841e-01 6.47611678e-01 4.71988805e-02 -3.69420201e-01 -4.74443227e-01 -1.03381872e+00 4.11925539e-02 4.22297180e-01 1.20689464e+00 4.53922838e-01 2.47486830e-01 -4.33174849e-01 6.62277222e-01 3.25587541e-01 9.48921859e-01 2.49813154e-01 -9.79403615e-01 6.59916699e-01 -6.08606972e-02 -1.41924620e-02 -8.87331903e-01 1.01349004e-01 -5.13780534e-01 -9.63618279e-01 -8.25259909e-02 6.06856570e-02 -2.45521978e-01 -1.08300316e+00 1.89135361e+00 1.59161329e-01 6.96469903e-01 -1.76898986e-01 8.41567159e-01 5.19952595e-01 1.14771354e+00 1.01719480e-02 -6.40406758e-02 1.23703492e+00 -1.31195951e+00 -6.64776683e-01 -2.14112684e-01 3.46049279e-01 -7.26236999e-01 9.62094307e-01 4.05744970e-01 -1.50048149e+00 -6.52545094e-01 -7.62401879e-01 -5.54629624e-01 1.63081944e-01 2.71621108e-01 4.76422399e-01 1.00343242e-01 -1.13744295e+00 3.63542676e-01 -1.17461824e+00 2.17003614e-01 6.08089685e-01 3.13851237e-01 -2.61767238e-01 -2.00271398e-01 -1.04622245e+00 3.68627250e-01 -2.80270457e-01 4.22199786e-01 -1.66623199e+00 -8.95079136e-01 -9.07257676e-01 4.02933925e-01 4.32532489e-01 -8.82153511e-01 1.19196284e+00 -1.01493466e+00 -1.56143999e+00 3.62354994e-01 -6.58073306e-01 -6.37363911e-01 5.91292322e-01 -7.47582257e-01 1.14612758e-01 1.92671552e-01 8.10645744e-02 6.88998401e-01 1.19953918e+00 -7.76198626e-01 -5.17636657e-01 -1.26683265e-01 2.28675261e-01 2.00618535e-01 -2.67381936e-01 -1.07279517e-01 -7.68002748e-01 -8.92553926e-01 -9.12933722e-02 -1.06434000e+00 -4.55788597e-02 -5.32217547e-02 -2.26106733e-01 1.06451355e-01 6.90695643e-01 -9.99357343e-01 1.07224703e+00 -2.12111163e+00 4.09805954e-01 -2.68448710e-01 1.99446976e-01 -9.82612148e-02 -3.18182796e-01 4.39498037e-01 -3.11064385e-02 5.71851730e-02 -6.39623925e-02 -6.88182235e-01 1.65423024e-02 1.76852971e-01 -7.49499142e-01 2.73266077e-01 3.43713850e-01 1.09424841e+00 -9.41558003e-01 -4.64441031e-01 1.75701320e-01 9.91569996e-01 -9.09273148e-01 5.00481427e-01 -3.69323969e-01 6.26955569e-01 -2.21777871e-01 3.84322822e-01 5.07514715e-01 -4.20233965e-01 9.19797719e-02 -1.82432890e-01 4.56222966e-02 4.24106002e-01 -5.71554601e-01 1.78336489e+00 -8.81374896e-01 9.89439487e-01 -1.62047110e-02 -8.56387675e-01 4.00499940e-01 6.60340071e-01 2.96224207e-01 -7.82879531e-01 -1.68496355e-01 -1.06148690e-01 -4.66032356e-01 -7.47664988e-01 2.88804591e-01 -8.27214420e-02 1.22924641e-01 2.83337712e-01 2.66186386e-01 1.69077218e-01 1.54739767e-01 3.30471277e-01 9.85294938e-01 5.15799344e-01 -3.55070949e-01 7.98121542e-02 2.95883268e-01 -4.82539058e-01 6.42009854e-01 6.41229808e-01 4.12781477e-01 6.56145036e-01 6.46130145e-01 -3.21346253e-01 -1.30577898e+00 -1.28286362e+00 1.88462377e-01 1.17397654e+00 -2.98656344e-01 -5.96419275e-01 -7.58904994e-01 -4.25329894e-01 -2.81803191e-01 4.41886574e-01 -6.55294120e-01 -4.29130010e-02 -7.45580256e-01 -4.42051679e-01 4.57330763e-01 7.12941945e-01 3.72885227e-01 -1.06433129e+00 -5.46121597e-01 1.35526553e-01 -5.00020742e-01 -1.26214933e+00 -7.71186054e-01 -1.38294980e-01 -9.59321260e-01 -6.48444295e-01 -8.71136248e-01 -7.82631159e-01 8.35366428e-01 2.37634748e-01 1.12557340e+00 -1.61570117e-01 -2.49311358e-01 3.91735196e-01 -1.71364084e-01 5.80077879e-02 -9.46990028e-02 -1.30144984e-01 -1.95608556e-01 1.21921800e-01 -1.84141964e-01 -7.54972816e-01 -7.52197921e-01 1.23948976e-01 -7.90250838e-01 4.94188249e-01 7.39008248e-01 1.15447307e+00 7.55409598e-01 -1.76573068e-01 1.60439625e-01 -7.43274212e-01 8.86412710e-02 -5.69464147e-01 -7.86379218e-01 1.76885411e-01 -3.63915741e-01 8.47186297e-02 5.73265553e-01 -3.87940884e-01 -1.22902048e+00 -1.67633481e-02 -1.78830624e-01 -7.71694124e-01 2.73112893e-01 5.75121701e-01 -3.80585268e-02 3.28785330e-01 1.33169621e-01 4.54593629e-01 -1.91612452e-01 -3.89217347e-01 4.21536326e-01 2.72633970e-01 4.44700360e-01 -5.61564386e-01 5.19656181e-01 5.19787610e-01 -1.23723447e-01 -6.37947202e-01 -9.51261163e-01 -5.37031963e-02 -3.47962260e-01 -1.99464142e-01 8.93121004e-01 -1.43689024e+00 -6.07674360e-01 5.76938212e-01 -1.13224506e+00 -7.51855731e-01 -2.25742146e-01 4.62044090e-01 -8.15369129e-01 1.91337794e-01 -9.10414338e-01 -6.88193917e-01 -4.02214259e-01 -1.11996734e+00 1.22613978e+00 -1.00184679e-01 1.05375737e-01 -9.01513100e-01 -9.70327295e-03 4.44565743e-01 4.34210956e-01 -9.14288834e-02 6.82560623e-01 2.56720245e-01 -1.21642697e+00 -6.03814796e-02 -2.11171001e-01 4.60449994e-01 -1.77010223e-01 -7.41079375e-02 -1.07117701e+00 -4.74114805e-01 2.46001724e-02 -3.09177637e-01 9.95415151e-01 6.84153616e-01 1.39727354e+00 -6.70418143e-01 -9.39570069e-02 1.00089467e+00 1.40303826e+00 -6.90834373e-02 7.86907017e-01 -2.29260758e-01 9.97156739e-01 3.55846941e-01 4.52912509e-01 5.43840945e-01 3.60680044e-01 6.46873832e-01 3.45659316e-01 -6.53943270e-02 -2.87065029e-01 -5.88075221e-01 8.39883566e-01 1.14812064e+00 -1.77667439e-01 -3.54619592e-01 -7.10986197e-01 7.74125695e-01 -1.85655797e+00 -1.19876027e+00 2.30357181e-02 2.16047478e+00 7.53965139e-01 5.65257156e-03 -1.70548692e-01 -2.85997957e-01 3.67267907e-01 2.68637419e-01 -4.76725370e-01 -1.92060918e-01 1.70449391e-02 2.20671162e-01 4.86152679e-01 7.90710390e-01 -9.69033659e-01 8.01035225e-01 6.07312918e+00 8.23166192e-01 -1.17340577e+00 4.73218173e-01 8.38232994e-01 -6.86067820e-01 -3.25059414e-01 7.74973929e-02 -6.25071287e-01 6.19971037e-01 1.24599254e+00 2.11135242e-02 5.93446195e-01 6.29839540e-01 4.04623240e-01 1.21970452e-01 -1.01686454e+00 9.98384535e-01 1.26944274e-01 -1.45485342e+00 9.90054756e-02 1.20208845e-01 8.63888323e-01 2.23187253e-01 4.12038714e-01 2.74233192e-01 1.29442334e-01 -1.10418618e+00 1.02909756e+00 5.47042191e-01 9.61975813e-01 -5.73846459e-01 3.81713927e-01 1.57564893e-01 -1.31918406e+00 -2.82600731e-01 -3.75055730e-01 -5.29508665e-02 4.16265339e-01 5.95389485e-01 -5.60394585e-01 1.37707531e-01 8.68062675e-01 8.88763845e-01 -3.57767135e-01 6.90301478e-01 -3.79729092e-01 9.87662196e-01 -2.28245884e-01 3.37203413e-01 1.87838346e-01 -2.44075954e-01 4.03068155e-01 1.35034668e+00 6.67147040e-01 6.89589418e-03 -8.43739361e-02 8.74510586e-01 -1.51377335e-01 -2.93736637e-01 -3.84756833e-01 -1.23384461e-01 2.73525745e-01 9.75961983e-01 -2.37103358e-01 -6.01882219e-01 -4.06192452e-01 1.26833546e+00 4.24594522e-01 5.64844668e-01 -1.42495501e+00 -7.04901293e-02 5.11793673e-01 2.60447145e-01 8.31715286e-01 -3.71904314e-01 -2.18748853e-01 -1.28428662e+00 4.94486064e-01 -7.73570538e-01 1.81993216e-01 -1.04172647e+00 -9.92111146e-01 6.44147575e-01 -1.30391002e-01 -1.15158105e+00 -1.94677874e-01 -3.65362018e-01 -3.24727416e-01 7.26275384e-01 -1.39666522e+00 -1.41141379e+00 -3.60201627e-01 5.90801358e-01 8.14880073e-01 1.44185767e-01 7.39762485e-01 5.70946038e-01 -4.24673706e-01 5.68161309e-01 1.71047628e-01 2.30413917e-02 5.83674908e-01 -1.18780255e+00 6.75215364e-01 9.11823034e-01 3.43707800e-01 4.56532270e-01 5.87946594e-01 -5.38765967e-01 -1.51176608e+00 -1.13118577e+00 9.97377217e-01 -3.35283011e-01 7.34246254e-01 -8.08247387e-01 -7.22920716e-01 1.02051449e+00 3.33793253e-01 1.96060762e-01 4.48662221e-01 5.76361679e-02 -5.92562616e-01 -2.36719936e-01 -5.13986707e-01 5.36880732e-01 1.20095527e+00 -9.34663832e-01 -1.04702234e-01 4.46784914e-01 7.98313558e-01 -5.59510767e-01 -8.47652197e-01 1.52728543e-01 8.42454851e-01 -9.01306570e-01 9.54049110e-01 -4.26424056e-01 9.11030054e-01 -1.22569792e-01 -1.66915637e-02 -9.38462675e-01 -4.96969193e-01 -8.55923951e-01 -4.84759927e-01 1.14266908e+00 5.38324118e-01 -4.59785163e-01 8.24004889e-01 4.04367477e-01 -1.69158906e-01 -9.51033115e-01 -9.87207055e-01 -6.28565967e-01 -6.85860813e-02 -4.83960867e-01 2.57577628e-01 4.11300719e-01 -1.60117492e-01 4.39574093e-01 -1.04993069e+00 4.17854667e-01 6.64847374e-01 1.78755820e-01 6.96603715e-01 -4.07773167e-01 -8.62230837e-01 6.22869059e-02 -2.28168234e-01 -1.79825628e+00 -8.28381162e-03 -6.44772112e-01 3.77859861e-01 -1.35024142e+00 5.86200118e-01 -3.76496613e-01 -2.52052516e-01 2.72703648e-01 -1.88083678e-01 4.10058558e-01 3.18360448e-01 1.37641087e-01 -6.83097064e-01 7.94685781e-01 1.15897107e+00 -1.46949127e-01 1.90727159e-01 -3.08654070e-01 -2.53619552e-01 4.94561136e-01 5.93718946e-01 -5.98047256e-01 -6.44424438e-01 -1.13261676e+00 4.57807034e-01 4.43646252e-01 5.94281077e-01 -1.01309764e+00 1.85614601e-01 5.83373047e-02 3.65412712e-01 -4.97822762e-01 7.83702075e-01 -6.25489473e-01 3.50084156e-01 3.45960736e-01 -4.59144950e-01 2.93825030e-01 1.22615546e-01 7.43481934e-01 -3.00760716e-01 1.65170431e-01 4.57795769e-01 -3.85004692e-02 -3.84506077e-01 6.21100426e-01 -5.19199491e-01 5.67057468e-02 7.86160767e-01 3.68730500e-02 -1.94275886e-01 -6.21762156e-01 -1.02485561e+00 1.70049548e-01 2.73286551e-01 3.72869551e-01 7.87174523e-01 -1.23147845e+00 -8.74114931e-01 3.59232336e-01 -3.20497572e-01 2.04641506e-01 8.26921999e-01 1.12258637e+00 -4.22887564e-01 5.68677485e-01 -5.64412102e-02 -7.33837247e-01 -1.36991358e+00 5.76413512e-01 2.51775056e-01 -4.25454497e-01 -6.86302841e-01 1.19277024e+00 6.64348543e-01 1.51529744e-01 1.98670506e-01 -3.40450048e-01 3.14283133e-01 -2.59571046e-01 5.76141417e-01 1.63776219e-01 -3.14626813e-01 -5.40842056e-01 -6.73116520e-02 5.49196720e-01 -8.18863064e-02 -2.70604968e-01 1.36846435e+00 -2.58774728e-01 -3.33925746e-02 4.36753452e-01 1.27798748e+00 1.95034638e-01 -1.94941843e+00 5.01701783e-04 -3.59985918e-01 -5.83641112e-01 8.12408254e-02 -5.75720251e-01 -1.27193499e+00 1.18563426e+00 4.84455109e-01 -1.14441127e-01 1.13237667e+00 -5.65053225e-02 1.09362388e+00 2.78855432e-02 2.00195968e-01 -7.20651805e-01 1.95835784e-01 5.65900981e-01 1.08177793e+00 -8.92705917e-01 -1.43914506e-01 -3.55185241e-01 -5.40677488e-01 7.60957241e-01 3.69687915e-01 -3.91733915e-01 5.98912895e-01 4.76753950e-01 -1.78430364e-01 -1.68127324e-02 -1.45121467e+00 2.50738710e-01 2.39472792e-01 4.22071964e-01 5.58531761e-01 -5.44435233e-02 3.10649693e-01 1.64757133e-01 -2.63447780e-02 2.48313367e-01 3.11794817e-01 6.63216829e-01 2.16173809e-02 -8.05968404e-01 -1.16649777e-01 2.95222640e-01 -6.98964715e-01 -4.83398497e-01 1.81246325e-01 1.47523373e-01 3.16192098e-02 8.33371341e-01 2.81580538e-01 -8.36999491e-02 -1.17661553e-02 -1.12596817e-01 7.44307995e-01 -3.66658509e-01 -1.63063392e-01 2.88424313e-01 6.89407960e-02 -8.89824867e-01 -2.48643517e-01 -6.70842707e-01 -9.59573030e-01 -2.70556778e-01 -1.50826484e-01 1.23836935e-01 5.10963023e-01 6.89925075e-01 8.05983305e-01 5.97878575e-01 5.86215734e-01 -1.03509784e+00 -4.95612025e-01 -9.70954597e-01 -1.07213743e-01 3.13513935e-01 6.20839417e-01 -3.98866981e-01 -2.19245762e-01 6.71264231e-01]

[10.668998718261719, -0.668192446231842]

9224b1d1-dfe9-43de-9670-697be4bbcbb2

cut-and-approximate-3d-shape-reconstruction

2210.12509

https://arxiv.org/abs/2210.12509v1

https://arxiv.org/pdf/2210.12509v1.pdf

Cut-and-Approximate: 3D Shape Reconstruction from Planar Cross-sections with Deep Reinforcement Learning

Current methods for 3D object reconstruction from a set of planar cross-sections still struggle to capture detailed topology or require a considerable number of cross-sections. In this paper, we present, to the best of our knowledge the first 3D shape reconstruction network to solve this task which additionally uses orthographic projections of the shape. Our method is based on applying a Reinforcement Learning algorithm to learn how to effectively parse the shape using a trial-and-error scheme relying on scalar rewards. This method cuts a part of a 3D shape in each step which is then approximated as a polygon mesh. The agent aims to maximize the reward that depends on the accuracy of surface reconstruction for the approximated parts. We also consider pre-training of the network for faster learning using demonstrations generated by a heuristic approach. Experiments show that our training algorithm which benefits from both imitation learning and also self exploration, learns efficient policies faster, which results the agent to produce visually compelling results.

['Azimkhon Ostonov']

2022-10-22

['3d-object-reconstruction', '3d-shape-reconstruction', 'object-reconstruction']

['computer-vision', 'computer-vision', 'computer-vision']

[ 8.41015428e-02 5.62907100e-01 2.10004076e-01 -1.00357033e-01 -5.48521876e-01 -5.12443841e-01 6.39979482e-01 1.77512094e-01 -3.28759491e-01 8.86545062e-01 -3.41645390e-01 -1.89739168e-01 -3.04090418e-02 -8.74947786e-01 -1.14043415e+00 -3.21145117e-01 -2.89654285e-01 1.17413330e+00 2.44886130e-01 -1.89444840e-01 4.31006521e-01 9.96318698e-01 -1.38501906e+00 -1.30680688e-02 7.65960634e-01 8.51786494e-01 5.10952294e-01 5.16938686e-01 -1.43879607e-01 3.69686991e-01 -3.95026505e-01 -4.58726883e-02 5.31356037e-01 -2.84249663e-01 -8.51763248e-01 3.52643639e-01 -7.47029036e-02 -4.53000873e-01 2.88732588e-01 7.23820925e-01 2.43563980e-01 -7.47542176e-03 7.79874504e-01 -8.38849425e-01 -1.97398752e-01 3.21694493e-01 -4.25528020e-01 -4.34036523e-01 5.52338183e-01 3.17723542e-01 6.70840144e-01 -7.60237217e-01 1.08988512e+00 1.34256470e+00 7.36635685e-01 6.62535191e-01 -1.34696424e+00 -3.41845453e-01 -8.66909847e-02 -4.03102994e-01 -8.01946998e-01 -4.63930406e-02 7.56908238e-01 -2.82948285e-01 1.21840286e+00 -1.37884736e-01 1.27228761e+00 7.03541160e-01 2.01016024e-01 8.75400841e-01 1.37178779e+00 -5.86710095e-01 6.23235345e-01 4.80411910e-02 -6.51790559e-01 9.58790421e-01 7.54959369e-03 6.19415343e-01 -4.01212880e-03 -1.95635241e-02 1.49105108e+00 -2.04967856e-01 -1.60509311e-02 -1.28209209e+00 -8.03198338e-01 7.31994152e-01 6.29257262e-01 3.21078412e-02 -6.31059885e-01 4.82226372e-01 5.29580079e-02 2.45665759e-01 5.98436333e-02 8.97172451e-01 -4.42329884e-01 -4.61669452e-02 -6.29265904e-01 5.13558865e-01 9.81095850e-01 7.73487866e-01 8.25748861e-01 4.14857686e-01 5.56542218e-01 5.60297608e-01 4.09302473e-01 3.67489964e-01 -2.70119892e-03 -1.46587682e+00 1.62977487e-01 7.38543153e-01 2.96505362e-01 -3.64086032e-01 -3.06170553e-01 -1.22551188e-01 -2.81984121e-01 1.31750345e+00 4.36949104e-01 -3.28343451e-01 -8.95075440e-01 1.46133947e+00 6.36031210e-01 -4.26477343e-02 1.62428632e-01 7.47028947e-01 1.46615326e-01 5.66258907e-01 -2.43508175e-01 -4.88125272e-02 6.14443362e-01 -7.82102942e-01 -1.97430089e-01 1.71512663e-01 4.15045381e-01 -4.24913079e-01 7.90551424e-01 6.53542101e-01 -1.72304988e+00 -3.47890437e-01 -1.05168951e+00 3.58699143e-01 -1.76043540e-01 -2.14583962e-03 4.47063059e-01 3.06654274e-01 -1.08520293e+00 1.38224804e+00 -1.01923740e+00 -2.50066191e-01 7.81932712e-01 5.56487262e-01 -3.38908315e-01 1.54666513e-01 -4.85266984e-01 1.22000563e+00 2.58928895e-01 -2.20092401e-01 -1.16533601e+00 -6.55106843e-01 -7.46311307e-01 -1.05499052e-01 4.05553907e-01 -7.94582546e-01 1.51479793e+00 -1.27515280e+00 -2.16578436e+00 8.12570214e-01 3.81005347e-01 -4.73991662e-01 6.82614982e-01 8.84581730e-03 4.49830860e-01 3.89921576e-01 -2.09946290e-01 1.11862314e+00 9.30939198e-01 -1.88728368e+00 -3.86207581e-01 -3.51910084e-01 1.99664459e-01 3.04566562e-01 4.04525876e-01 -4.84063089e-01 1.58762988e-02 -4.52434033e-01 -2.63001323e-02 -8.88664007e-01 -5.60020506e-01 4.31834519e-01 -9.80393887e-02 -3.15307796e-01 7.47184038e-01 -4.06262338e-01 1.95699558e-01 -1.66210902e+00 4.33661550e-01 4.26393062e-01 -9.18677002e-02 1.93596005e-01 -1.64247528e-01 6.93054497e-01 1.53528228e-01 1.28337085e-01 -4.35595959e-01 -5.24995625e-01 -7.37042651e-02 4.42336559e-01 -2.18928695e-01 3.52434248e-01 3.51386815e-01 8.13003004e-01 -9.60324287e-01 -2.05903441e-01 1.57337353e-01 4.44917828e-01 -5.58615088e-01 3.35549355e-01 -6.36878908e-01 4.67169136e-01 -7.04987228e-01 4.09767419e-01 4.69943851e-01 1.06963165e-01 2.02264622e-01 1.83922678e-01 -3.23171943e-01 1.09733060e-01 -1.10714173e+00 1.98720026e+00 -5.36434352e-01 2.07120016e-01 5.27705312e-01 -1.18212783e+00 1.25129843e+00 1.77435726e-01 5.03741920e-01 -5.01371384e-01 1.91950902e-01 4.76055980e-01 -1.52752414e-01 -3.93479645e-01 1.38756782e-01 -2.13293806e-01 2.38619640e-01 6.22120023e-01 -9.90560502e-02 -9.75468099e-01 -1.02986030e-01 -1.25953421e-01 9.66743946e-01 1.02588153e+00 1.76957279e-01 -1.76907033e-01 2.77439356e-01 2.83266723e-01 2.43146360e-01 5.07113516e-01 2.22932979e-01 4.48234767e-01 4.76864100e-01 -6.46582484e-01 -1.60598099e+00 -1.06814051e+00 1.13261044e-02 1.44400299e-01 1.72571838e-02 1.01680391e-01 -9.73692596e-01 -6.53467238e-01 1.79501623e-01 9.41925824e-01 -6.19925797e-01 1.38079271e-01 -9.32436347e-01 1.91359848e-01 1.39285289e-02 5.50927699e-01 2.22935468e-01 -1.73075962e+00 -1.38623452e+00 4.67161089e-01 7.64873445e-01 -6.45979345e-01 -2.52084639e-02 3.32763910e-01 -1.42283428e+00 -1.13264549e+00 -7.35561907e-01 -9.47790742e-01 9.53375280e-01 -7.75963217e-02 9.28146303e-01 2.74417460e-01 -4.29364145e-01 6.50832236e-01 -2.01906189e-01 -3.99908721e-01 -7.51278579e-01 -2.53489554e-01 -1.94205493e-01 -5.44824362e-01 -3.76231760e-01 -1.03127646e+00 -3.33339572e-01 1.09805256e-01 -6.33108616e-01 2.57198215e-01 8.22507799e-01 7.54654229e-01 8.36029708e-01 -1.39423728e-01 4.38302279e-01 -4.36078191e-01 6.45908713e-01 -1.23720147e-01 -9.82433975e-01 1.69307198e-02 -4.92931396e-01 4.53841567e-01 7.57575929e-01 -5.31875134e-01 -9.14761782e-01 6.27760291e-01 2.76480615e-02 -7.76949048e-01 -3.28567326e-01 -1.62494183e-02 2.89490551e-01 -3.23470622e-01 4.93715763e-01 1.81820229e-01 5.96937597e-01 -4.64288890e-01 3.57525527e-01 9.89796147e-02 1.70231298e-01 -8.10246885e-01 6.36034667e-01 4.90738869e-01 3.56041938e-01 -5.74644804e-01 -3.00158262e-01 5.76124787e-02 -7.37008929e-01 -2.55745292e-01 6.25451684e-01 -3.42337489e-01 -1.11367452e+00 1.27542689e-01 -1.18893015e+00 -8.79943788e-01 -7.89420068e-01 3.70696545e-01 -1.35329103e+00 2.13693753e-01 -4.02214199e-01 -9.89495873e-01 -2.70434976e-01 -1.08195615e+00 1.01570594e+00 1.38939500e-01 -8.45906511e-03 -6.99259162e-01 3.11636984e-01 -1.44683599e-01 3.78372103e-01 5.58965206e-01 8.83386672e-01 -2.23640889e-01 -8.94262731e-01 7.18490481e-02 7.34040067e-02 7.34939277e-02 -2.66838521e-01 -8.85352318e-04 -5.48111916e-01 -2.46556729e-01 -2.13376075e-01 -8.44752669e-01 5.70926011e-01 3.86499226e-01 1.04715312e+00 -3.49153966e-01 -3.88270885e-01 3.79481912e-01 1.42242908e+00 2.73748875e-01 5.44440925e-01 3.99866551e-01 3.17241430e-01 8.32256913e-01 5.10862112e-01 4.36386377e-01 2.45388031e-01 5.30868471e-01 9.84601378e-01 7.45402202e-02 -1.76667869e-01 -4.35641289e-01 1.88859805e-01 4.90316711e-02 -2.38567472e-01 1.57456309e-01 -7.53291905e-01 4.39853340e-01 -1.64812410e+00 -8.03981423e-01 2.64753878e-01 2.18836999e+00 8.24776351e-01 1.95484549e-01 2.82388777e-01 -1.50623517e-02 3.51222873e-01 -4.15777922e-01 -7.02349186e-01 -7.71599770e-01 3.91980946e-01 4.14356261e-01 3.18566740e-01 7.62157798e-01 -5.34558475e-01 9.79937971e-01 6.09802866e+00 4.50005323e-01 -9.56098795e-01 -4.85165387e-01 2.72599429e-01 8.62888247e-02 -4.03604686e-01 1.86317682e-01 -3.49413216e-01 1.94991622e-02 4.86113071e-01 3.02847981e-01 1.01762950e+00 8.00225496e-01 2.25561649e-01 -3.03187042e-01 -1.13522899e+00 4.95758593e-01 -2.29069248e-01 -1.45711935e+00 -2.23748609e-02 1.37949437e-01 7.89950669e-01 -1.89194769e-01 -1.44222990e-01 1.85685858e-01 7.74145603e-01 -1.06693327e+00 9.93445039e-01 6.08097374e-01 5.55738151e-01 -9.34867144e-01 1.40993103e-01 8.32610786e-01 -8.20199728e-01 -8.11414607e-03 -3.13054264e-01 -3.22766118e-02 2.78342873e-01 4.16230559e-02 -1.08170867e+00 2.77718693e-01 4.31350917e-01 4.46012110e-01 1.59075186e-01 1.17752385e+00 -3.47191453e-01 1.08238921e-01 -4.64357674e-01 -4.57853407e-01 4.74089473e-01 -3.46778512e-01 7.64994979e-01 7.79597819e-01 3.41001362e-01 2.33243331e-01 3.93740714e-01 1.23995185e+00 7.42551386e-02 6.03325404e-02 -7.97552288e-01 1.25398904e-01 3.35102350e-01 1.04144549e+00 -8.03619385e-01 -1.44086823e-01 1.37143552e-01 6.86160684e-01 7.94215262e-01 1.22241996e-01 -3.16730231e-01 -1.10438792e-02 1.18776403e-01 2.37072378e-01 7.83453941e-01 -4.75362092e-01 -3.05984527e-01 -5.92702150e-01 -1.34050176e-01 -6.40248954e-01 -1.77128106e-01 -1.01686358e+00 -9.64207292e-01 5.32897651e-01 -1.59921739e-02 -1.04371309e+00 -5.14799356e-01 -4.60279256e-01 -5.84401846e-01 6.65618122e-01 -1.58397448e+00 -9.67761338e-01 1.27679873e-02 2.74415672e-01 6.10225737e-01 -1.65052548e-01 9.80386138e-01 -5.14659047e-01 4.23180312e-02 -3.40498388e-02 -1.46977333e-02 -3.41824889e-01 6.17736839e-02 -1.45510757e+00 3.26867968e-01 5.55765368e-02 -4.53135483e-02 1.84242517e-01 6.44046843e-01 -7.94222414e-01 -1.48691452e+00 -6.05903924e-01 2.99217939e-01 -1.37678877e-01 2.39568979e-01 -9.78189856e-02 -8.77439678e-01 4.91625220e-01 3.22931260e-01 -3.56368460e-02 -1.01431727e-01 -3.39893311e-01 -5.16048744e-02 9.58027989e-02 -1.46610820e+00 4.89020675e-01 8.93525720e-01 1.01097651e-01 -6.61732137e-01 9.48178768e-02 2.32296214e-01 -6.67293906e-01 -8.10038745e-01 2.83496171e-01 5.88211596e-01 -9.53327537e-01 9.82366681e-01 -5.73092759e-01 5.01754522e-01 -2.24310800e-01 2.11326927e-01 -1.67244947e+00 -2.20334157e-02 -8.10157239e-01 -1.33759364e-01 5.37225783e-01 3.59113723e-01 -3.41119289e-01 1.08698428e+00 1.76554099e-01 -2.59118319e-01 -1.23064888e+00 -8.50114465e-01 -8.01715076e-01 3.02547842e-01 -3.38503532e-02 3.67777348e-01 5.52714705e-01 -8.51670429e-02 -4.92604449e-02 7.50927348e-03 -1.17320521e-02 8.99419308e-01 4.66719329e-01 7.07462907e-01 -1.23962045e+00 -3.85142744e-01 -5.29183030e-01 9.10094157e-02 -1.09113121e+00 2.03817040e-01 -8.05250645e-01 2.08291203e-01 -1.63865483e+00 -2.43640736e-01 -8.38530540e-01 4.72121060e-01 5.25353670e-01 5.09122729e-01 -2.21688777e-01 3.14421088e-01 2.38085672e-01 -3.36737365e-01 6.85737371e-01 1.82239997e+00 2.64379352e-01 -5.27055323e-01 7.16052875e-02 -1.85567990e-01 8.37565064e-01 9.63546038e-01 -5.05100429e-01 -2.73005843e-01 -4.31265771e-01 1.21558271e-01 5.72150111e-01 3.97552103e-01 -7.70265400e-01 -3.64284366e-02 -1.76443473e-01 6.38472855e-01 -5.82302392e-01 7.38480926e-01 -1.05068576e+00 5.13203368e-02 8.47469687e-01 -3.81524622e-01 1.65560454e-01 3.36546272e-01 6.14215732e-01 2.49239936e-01 -5.73542178e-01 9.72296000e-01 -7.49321461e-01 -1.56388640e-01 5.88433519e-02 -4.14151102e-01 -1.45625919e-01 1.21583116e+00 -5.10311544e-01 4.28016990e-01 -2.84644961e-01 -8.70029569e-01 2.70945549e-01 8.54106963e-01 -4.17906679e-02 8.32794368e-01 -1.24768901e+00 -6.05516732e-01 1.42294869e-01 -6.26014054e-01 3.35122406e-01 -2.31045127e-01 2.54541397e-01 -8.11158478e-01 9.02518071e-03 -5.92109680e-01 -6.16120517e-01 -9.41731095e-01 5.25010884e-01 6.49234891e-01 -3.85386080e-01 -9.23191190e-01 4.28354859e-01 -2.78509051e-01 -7.03089833e-01 2.98965931e-01 -3.19998562e-01 -1.32493660e-01 -3.83524597e-01 4.70228791e-02 4.02528822e-01 -3.20587635e-01 -1.98975787e-01 -4.30934597e-03 9.08771634e-01 1.16138831e-01 -4.74387169e-01 1.74693060e+00 2.84286231e-01 2.37639874e-01 3.73377874e-02 8.89212251e-01 -1.07695796e-01 -2.05212808e+00 1.47045702e-01 -2.43466139e-01 -4.32291210e-01 -6.50097281e-02 -9.98897433e-01 -9.10392940e-01 8.01296771e-01 2.46830553e-01 1.96700320e-01 7.21236229e-01 -5.91568984e-02 6.24762177e-01 5.65975368e-01 4.50495213e-01 -1.15422857e+00 6.24083936e-01 3.50507468e-01 1.37750697e+00 -8.57597589e-01 1.57706246e-01 -1.92820027e-01 -5.03675699e-01 1.67580545e+00 5.13862073e-01 -8.61080825e-01 3.18232596e-01 4.66236442e-01 -1.70054480e-01 -1.56481400e-01 -5.37882566e-01 -8.21662173e-02 -3.62268426e-02 8.37792158e-01 -2.62599856e-01 -1.23631969e-01 -1.24180235e-01 -1.61333039e-01 -1.92177236e-01 -9.14293528e-02 5.29974341e-01 1.16703212e+00 -8.50846767e-01 -1.39131999e+00 -3.49873483e-01 3.35170962e-02 -1.44552156e-01 5.33615589e-01 -4.66634542e-01 1.08413219e+00 -1.56734809e-02 2.47918904e-01 -6.22622855e-03 2.11776316e-01 4.24317837e-01 -2.61494935e-01 1.18447447e+00 -5.22118151e-01 -6.39316380e-01 1.04642846e-01 7.80525729e-02 -6.73747718e-01 -2.74192601e-01 -8.53608966e-01 -1.87534297e+00 -4.86005703e-03 -3.00445575e-02 -1.22352410e-02 1.05272293e+00 6.81222677e-01 3.02204013e-01 2.38369286e-01 7.29430199e-01 -1.53031611e+00 -9.71765161e-01 -4.33569223e-01 -3.21178198e-01 2.23707393e-01 2.84576148e-01 -8.10413480e-01 -1.85121186e-02 -2.32113481e-01]

[4.758941650390625, 0.4571479558944702]

aa4f2f9e-f628-4e83-9468-67935ebdfa5a

learning-emotion-aware-contextual-1

https://openreview.net/forum?id=8yJ0RpqNppY

https://openreview.net/pdf?id=8yJ0RpqNppY

Learning Emotion-Aware Contextual Representations for Emotion Cause Analysis

Emotion Cause Analysis has been a key topic in natural language processing. Previous works focus on Emotion Cause Extraction (ECE), a clause-level classification task aimed at extracting causes of certain given emotion in text. The task has been expanded to Emotion Cause Pair Extraction (ECPE) that focus on extracting both emotions and corresponding causes in the context. Most existing methods for the ECPE task implement a joint model that performs extracting and matching of emotion and cause clauses simultaneously. However, we argue that different input features are needed for the two subtasks, thus sharing contextual representations may be suboptimal. In this work, we propose a pipelined approach that builds on two independent pre-trained encoders, in which the emotion extraction model only provide input features for the cause extraction model. Based on a series of careful experiments, we validate that our model can create distinct contextual representations according to specific emotional texts, and thus achieve state-of-the-art performance in both ECE and ECPE tasks, with the absolute F1 improvements of 1.5% and 4.72% over best previous works respectively. Besides, we apply a set of simple clause selection rules to extract multiple pairs in the document, strengthening the applicability of our approach in real world scenarios.

['Anonymous']

2021-09-17

acl-arr-september-2021-9

['emotion-cause-pair-extraction', 'emotion-cause-extraction']

['natural-language-processing', 'natural-language-processing']

[ 4.36650485e-01 2.67890126e-01 -3.04235388e-02 -6.02031648e-01 -9.10937786e-01 -5.59005916e-01 6.68739796e-01 4.66645062e-01 -3.32598805e-01 6.71685159e-01 4.06670123e-01 -1.05336038e-02 -1.12536503e-03 -6.52097285e-01 -4.30003852e-01 -4.89628136e-01 -8.90539438e-02 -4.97241393e-02 -1.09566934e-01 -2.93669194e-01 1.84771493e-01 5.63180335e-02 -1.72084689e+00 7.31616437e-01 7.69427180e-01 9.99684036e-01 -1.57833219e-01 6.70471430e-01 -3.70407283e-01 1.20625818e+00 -7.42131174e-01 -6.24586940e-01 -2.63914794e-01 -4.98037845e-01 -1.06497300e+00 -2.87920684e-01 -2.95388252e-01 4.42899838e-02 1.77131832e-01 8.51022243e-01 4.40695941e-01 2.96256039e-03 7.99590290e-01 -1.50042140e+00 -4.60891634e-01 8.56330514e-01 -5.67356050e-01 -6.93873921e-03 7.17827320e-01 -4.63427514e-01 1.42086458e+00 -6.93388939e-01 5.07754982e-01 1.11669481e+00 4.89289165e-01 6.62171960e-01 -6.56920552e-01 -7.09897637e-01 3.57064754e-01 4.20451343e-01 -1.13994002e+00 -1.93031892e-01 1.10514486e+00 -7.83347934e-02 1.47492719e+00 3.90112430e-01 4.43555385e-01 1.09503961e+00 4.21409935e-01 1.05982113e+00 1.14984787e+00 -5.55147886e-01 3.01801413e-01 1.72195315e-01 3.02054524e-01 2.76816815e-01 -1.25521272e-01 -2.86082447e-01 -5.78869522e-01 -2.18790084e-01 1.49009421e-01 -3.67574096e-01 -2.88322866e-01 2.31149852e-01 -9.51679289e-01 9.93717134e-01 1.96005598e-01 5.97138464e-01 -5.88192284e-01 8.88162702e-02 6.41111314e-01 2.22761661e-01 4.24931228e-01 4.58179891e-01 -6.69233441e-01 -2.53643274e-01 -7.62181938e-01 4.69705552e-01 9.20549572e-01 9.60782707e-01 5.07756710e-01 -4.70075697e-01 -3.06251824e-01 6.95507586e-01 1.74404964e-01 1.99291587e-01 4.57891077e-01 -3.12176257e-01 3.85994822e-01 9.16566133e-01 -3.53579596e-02 -1.23852777e+00 -4.71243352e-01 -4.76610988e-01 -6.86905622e-01 -3.42917740e-01 -8.92398804e-02 -4.80962783e-01 -6.82010770e-01 1.92436767e+00 4.48540092e-01 2.05810353e-01 4.57274705e-01 8.94857228e-01 1.10987651e+00 7.44892776e-01 4.15964454e-01 -4.50858474e-01 1.64479315e+00 -8.76430988e-01 -1.25474763e+00 -4.01451588e-01 7.54681766e-01 -9.43962097e-01 8.72886658e-01 5.75598538e-01 -8.59366417e-01 -1.27877578e-01 -1.12031627e+00 -1.99664444e-01 -5.95478714e-01 3.91443223e-01 9.56231356e-01 2.84610033e-01 -5.89784026e-01 2.53474891e-01 -2.84727186e-01 -1.47674814e-01 1.11977026e-01 4.45059061e-01 -2.84662396e-01 1.31903961e-01 -1.71319211e+00 7.45015442e-01 3.24178696e-01 2.14744076e-01 -2.34958559e-01 -4.72199351e-01 -8.23723316e-01 1.19074643e-01 4.73800778e-01 -3.14871579e-01 1.22267556e+00 -1.19101667e+00 -1.24284554e+00 8.54945362e-01 -4.23051804e-01 -2.82662481e-01 -1.04370579e-01 -7.27288842e-01 -7.74438322e-01 4.91181761e-02 1.38035029e-01 5.59527576e-01 6.66991889e-01 -1.30543399e+00 -6.95921004e-01 -1.17845990e-01 1.18874975e-01 2.72332311e-01 -2.08683133e-01 6.38288856e-01 -5.43801785e-01 -6.08788490e-01 -2.47299105e-01 -7.12665081e-01 -6.95336163e-02 -6.87951744e-01 -4.68293875e-01 -6.84005737e-01 8.87174428e-01 -6.07359231e-01 1.58788919e+00 -2.14970088e+00 9.50384438e-02 1.20440006e-01 1.34023711e-01 -5.67050800e-02 -2.04784218e-02 4.95423824e-01 -4.31055635e-01 3.15156370e-01 -1.66377008e-01 -2.60993421e-01 2.45060652e-01 1.73693642e-01 -5.37362933e-01 -6.77158535e-02 8.30353379e-01 8.22479486e-01 -8.44944179e-01 -6.73786223e-01 -1.37558192e-01 5.70933521e-01 -5.39502144e-01 5.21037877e-01 -1.33000106e-01 -7.96009824e-02 -5.02618253e-01 4.47082341e-01 6.83023155e-01 2.66261417e-02 4.04928267e-01 -2.12168157e-01 1.16069473e-01 5.78429937e-01 -1.29310608e+00 1.38619757e+00 -5.66020668e-01 4.30895299e-01 -1.13162901e-02 -8.90903771e-01 1.04637563e+00 6.81933820e-01 4.04734790e-01 -5.96139193e-01 4.76584613e-01 1.65230826e-01 4.88584600e-02 -6.78225040e-01 4.94089037e-01 -3.84560198e-01 -6.88140810e-01 3.93585175e-01 1.26839250e-01 2.13521644e-02 2.29108810e-01 2.28183001e-01 1.11345387e+00 1.00057080e-01 6.12707078e-01 -8.59022439e-02 7.50928998e-01 -6.39962405e-02 8.53856564e-01 2.75683433e-01 -3.63581479e-01 3.80265355e-01 9.95856345e-01 -3.57143939e-01 -5.33534169e-01 -4.74612683e-01 1.28871590e-01 9.48286235e-01 1.51471898e-01 -7.12228179e-01 -7.03257740e-01 -9.26737249e-01 -4.28453535e-01 8.86324584e-01 -8.26842666e-01 -1.30820349e-01 -7.02772975e-01 -8.72157514e-01 6.07125163e-01 4.74987507e-01 5.05108178e-01 -1.29151392e+00 -9.05083418e-01 2.83803374e-01 -4.63361979e-01 -1.26124156e+00 7.17219859e-02 6.56454325e-01 -2.73952901e-01 -1.19469535e+00 -1.23152092e-01 -7.25241423e-01 3.49683553e-01 -2.07618073e-01 1.31674385e+00 1.13644429e-01 -1.85264260e-01 6.31831735e-02 -8.73481512e-01 -6.98985875e-01 -1.53681561e-01 1.87119499e-01 -2.69517541e-01 1.22167118e-01 9.76803482e-01 -4.12304461e-01 -3.16765994e-01 -1.10777296e-01 -9.44426179e-01 2.69565612e-01 7.29750514e-01 6.37968302e-01 4.55643237e-01 2.47714967e-01 8.03016782e-01 -1.04607141e+00 9.55207348e-01 -6.91486895e-01 8.83934349e-02 3.27596724e-01 -5.48045993e-01 1.45556137e-01 7.28595793e-01 -2.09790468e-01 -1.34944999e+00 1.61109954e-01 -2.77739614e-01 -1.30653441e-01 -5.09071529e-01 6.73166513e-01 -4.00908828e-01 6.54546559e-01 1.86956048e-01 -2.03337558e-02 -5.73505580e-01 -1.05848059e-01 3.86763543e-01 8.25875044e-01 3.65695804e-01 -5.35591602e-01 4.38889563e-01 4.17636074e-02 -1.91257775e-01 -2.36438051e-01 -1.06103230e+00 -5.15919685e-01 -4.48541552e-01 -2.78514206e-01 1.08732092e+00 -9.70471084e-01 -5.96039653e-01 6.30341545e-02 -1.38247740e+00 -4.81192954e-03 1.77265272e-01 3.87965441e-01 -2.66817182e-01 1.14701398e-01 -5.70950806e-01 -9.45962548e-01 -5.51580906e-01 -8.64066780e-01 1.46164155e+00 2.24512979e-01 -6.87382221e-01 -7.48967111e-01 3.94921601e-02 1.46684438e-01 9.83093083e-02 5.89312673e-01 1.11902487e+00 -9.06386077e-01 9.51314718e-03 -1.44251779e-01 -1.88014120e-01 -1.00816667e-01 1.89297065e-01 2.09378645e-01 -1.23512578e+00 2.48319834e-01 1.44150421e-01 -5.50480068e-01 6.46654010e-01 -8.55457187e-02 9.50742543e-01 -4.37206149e-01 -2.72101790e-01 1.19116440e-01 1.56345558e+00 3.78797919e-01 6.49506509e-01 2.85512775e-01 4.55860347e-01 9.23084080e-01 8.76616359e-01 4.92411822e-01 3.77500564e-01 5.31795621e-01 4.23250705e-01 -3.24433655e-01 1.31726459e-01 -2.96594072e-02 3.08714449e-01 9.06308353e-01 1.00524612e-01 -3.02668422e-01 -7.37863600e-01 6.65017724e-01 -1.98531902e+00 -9.12208676e-01 -3.90127689e-01 1.46518302e+00 1.13838875e+00 -3.16712148e-02 -1.25077605e-01 5.61331987e-01 5.99663138e-01 1.80850253e-01 -1.21928819e-01 -9.69069898e-01 -4.76266816e-02 4.25646633e-01 -8.60248134e-02 3.89121115e-01 -1.35435605e+00 9.43551123e-01 5.68539953e+00 8.79820943e-01 -1.06720936e+00 -1.56253390e-02 6.31034911e-01 -9.32491869e-02 -4.59064543e-01 -2.24222317e-02 -5.58084190e-01 1.62209705e-01 8.27160656e-01 -1.50287719e-02 1.15529902e-01 9.10879314e-01 8.68444145e-02 -1.46748237e-02 -1.11099076e+00 8.82555783e-01 1.29312739e-01 -7.90717006e-01 -1.33297578e-01 -2.45967597e-01 5.90563178e-01 -4.84888285e-01 -3.29443842e-01 6.38915002e-01 -1.47824854e-01 -9.92135763e-01 6.55384600e-01 2.83299804e-01 5.11343598e-01 -1.29530931e+00 1.30979562e+00 1.19659096e-01 -1.30209255e+00 5.58675732e-03 -8.25623274e-02 -3.73982012e-01 1.16190203e-01 8.48228097e-01 -6.31083012e-01 7.87945449e-01 7.51087606e-01 5.53351760e-01 -4.88820940e-01 3.92442018e-01 -7.40981936e-01 6.65862918e-01 -1.82050705e-01 -4.18919891e-01 2.03101277e-01 3.07859719e-01 2.91071951e-01 1.65870976e+00 1.72709096e-02 3.46658975e-01 -1.16500393e-01 9.05075014e-01 -1.31561801e-01 5.24911940e-01 -3.28172117e-01 9.14292261e-02 3.37539911e-01 1.61582947e+00 -7.49728680e-01 -4.32553113e-01 -3.40140760e-01 9.69990432e-01 4.86708373e-01 1.70334086e-01 -1.21591520e+00 -8.57344270e-01 4.67481285e-01 -6.04079306e-01 3.59131336e-01 2.00404391e-01 -3.30832213e-01 -1.21899438e+00 1.88492775e-01 -9.53320324e-01 3.81843984e-01 -7.49980152e-01 -1.22262859e+00 8.67065907e-01 5.00926077e-02 -1.04378831e+00 -5.71537971e-01 -5.83365321e-01 -9.63659167e-01 6.21288478e-01 -1.69972348e+00 -1.12270665e+00 -2.20630288e-01 5.45685410e-01 4.44907486e-01 3.47063094e-01 1.06665552e+00 2.38723308e-01 -7.48878002e-01 5.65992653e-01 -6.06748760e-01 1.77841559e-01 7.45514274e-01 -1.33019006e+00 -1.82590529e-01 1.02580798e+00 8.08660313e-02 6.34467900e-01 8.41299653e-01 -6.42858326e-01 -1.36640716e+00 -9.80535984e-01 1.73085392e+00 -2.19194427e-01 4.15160775e-01 -6.16796732e-01 -8.04567397e-01 3.89890671e-01 7.94990301e-01 -2.55711019e-01 1.02826512e+00 4.09033746e-01 -3.74592304e-01 1.43280521e-01 -1.10431910e+00 5.55365324e-01 6.94223046e-01 -3.61093253e-01 -8.90407622e-01 -4.31202166e-02 8.77039194e-01 -1.87985584e-01 -7.95048118e-01 5.47875226e-01 3.82891029e-01 -1.00258756e+00 3.74174565e-01 -5.70191622e-01 1.00689256e+00 -2.85944730e-01 -2.30750799e-01 -9.58684266e-01 -2.05263585e-01 -5.41522980e-01 -1.75822765e-01 1.82654822e+00 5.62949181e-01 -1.44336969e-01 3.18225086e-01 8.74604702e-01 7.44766295e-02 -1.20543635e+00 -6.87878191e-01 -2.40548089e-01 -1.17895961e-01 -8.67861092e-01 9.07836080e-01 1.03303719e+00 4.63475555e-01 8.43171597e-01 -2.73695290e-01 1.17722698e-01 8.13829973e-02 5.13529658e-01 4.07443076e-01 -8.60037625e-01 -2.66264886e-01 -4.18545008e-01 3.12434565e-02 -5.40432632e-01 4.02110934e-01 -6.89932585e-01 2.91578680e-01 -1.48197675e+00 3.65318656e-01 -1.67249918e-01 -4.21827376e-01 8.10688734e-01 -6.87195539e-01 -5.21163903e-02 7.87987635e-02 -1.36437222e-01 -7.12017357e-01 6.49897456e-01 7.67759919e-01 2.16361750e-02 -2.00007111e-01 -4.38450634e-01 -9.60006833e-01 7.42688775e-01 8.71810436e-01 -5.36861777e-01 -4.88875926e-01 -1.81685999e-01 6.17300689e-01 -2.90040821e-01 1.14442207e-01 -6.48700058e-01 2.23578382e-02 -2.63836145e-01 2.70868629e-01 -4.11672652e-01 1.62943024e-02 -8.24494660e-01 -2.55632252e-01 6.63232058e-02 -4.46265191e-01 8.39444920e-02 3.85659039e-01 2.85157025e-01 -6.25853240e-01 -2.85259306e-01 2.75396615e-01 9.56122428e-02 -7.61424363e-01 -2.16100693e-01 -4.75128323e-01 -4.99001937e-03 9.89249408e-01 2.17194855e-01 -3.75667438e-02 -3.81516725e-01 -3.30944151e-01 1.39145717e-01 -1.51276350e-01 4.68854010e-01 6.65934503e-01 -1.43831778e+00 -6.22245669e-01 -3.66699174e-02 1.70345530e-01 -6.86593130e-02 7.36989081e-02 7.30044723e-01 8.93074200e-02 4.38399941e-01 9.12366137e-02 -4.66649886e-03 -1.47393775e+00 6.98090315e-01 4.88508306e-02 -7.44184732e-01 -2.46346831e-01 8.39356184e-01 7.37289712e-02 -2.83627808e-01 1.84021026e-01 -4.69629586e-01 -6.66459858e-01 1.80746764e-01 4.67496604e-01 -1.36194721e-01 1.33124501e-01 -6.45832062e-01 -7.22207725e-01 4.75989580e-01 -5.09081483e-02 -1.01090856e-01 1.43754864e+00 -2.24493793e-04 -3.57788861e-01 3.88199598e-01 1.30579746e+00 1.79811418e-01 -5.69700360e-01 8.51703286e-02 9.10623670e-02 3.61479856e-02 1.03189617e-01 -1.02340496e+00 -8.99663746e-01 7.94201195e-01 1.70007437e-01 1.85955614e-01 1.72882020e+00 2.96450611e-02 9.17115271e-01 1.12159483e-01 -2.39792233e-03 -1.18601215e+00 -1.85459957e-01 5.13093889e-01 1.10309196e+00 -1.14169610e+00 -1.02784082e-01 -8.09020877e-01 -7.31999934e-01 1.12089539e+00 6.98726118e-01 -1.82753280e-01 4.75010276e-01 5.86032450e-01 1.24776304e-01 -3.75580966e-01 -1.21162963e+00 -3.16741914e-01 4.57633525e-01 1.08503960e-01 9.90846217e-01 8.16588327e-02 -7.36748338e-01 1.57438052e+00 -2.90097684e-01 -8.25290903e-02 1.91865921e-01 1.10114503e+00 -1.01000078e-01 -1.36674309e+00 -1.76096171e-01 2.46340200e-01 -8.71496260e-01 -2.21928492e-01 -1.07569551e+00 8.84145439e-01 4.92740273e-01 1.30328763e+00 -1.81266874e-01 -6.04906499e-01 2.26954237e-01 3.60961378e-01 2.20526844e-01 -4.47220385e-01 -9.67674971e-01 2.74106562e-01 5.79256952e-01 -6.40908003e-01 -8.13662887e-01 -5.32721221e-01 -1.70285928e+00 9.80518609e-02 -4.22896951e-01 2.26217180e-01 4.14441973e-01 1.27923131e+00 4.10868555e-01 8.15786481e-01 7.17522144e-01 -3.00731897e-01 -6.43285215e-02 -9.80579019e-01 -3.37913215e-01 7.47074783e-01 6.79998845e-02 -5.41902959e-01 -2.99147218e-01 1.56738237e-01]

[12.630476951599121, 6.216465950012207]

66b9a793-2659-4b04-b5f8-6829fc98b16a

counterfactual-reasoning-for-fair-clinical

1907.0626

https://arxiv.org/abs/1907.06260v1

https://arxiv.org/pdf/1907.06260v1.pdf

Counterfactual Reasoning for Fair Clinical Risk Prediction

The use of machine learning systems to support decision making in healthcare raises questions as to what extent these systems may introduce or exacerbate disparities in care for historically underrepresented and mistreated groups, due to biases implicitly embedded in observational data in electronic health records. To address this problem in the context of clinical risk prediction models, we develop an augmented counterfactual fairness criteria to extend the group fairness criteria of equalized odds to an individual level. We do so by requiring that the same prediction be made for a patient, and a counterfactual patient resulting from changing a sensitive attribute, if the factual and counterfactual outcomes do not differ. We investigate the extent to which the augmented counterfactual fairness criteria may be applied to develop fair models for prolonged inpatient length of stay and mortality with observational electronic health records data. As the fairness criteria is ill-defined without knowledge of the data generating process, we use a variational autoencoder to perform counterfactual inference in the context of an assumed causal graph. While our technique provides a means to trade off maintenance of fairness with reduction in predictive performance in the context of a learned generative model, further work is needed to assess the generality of this approach.

['Nigam H. Shah', 'Tony Duan', 'Stephen Pfohl', 'Daisy Yi Ding']

2019-07-14

['counterfactual-inference']

['miscellaneous']

[ 5.87730348e-01 9.67630088e-01 -4.02243406e-01 -5.61115682e-01 -1.62088960e-01 -2.97216445e-01 6.16052806e-01 5.81427753e-01 -6.65897429e-01 8.85825336e-01 7.60785758e-01 -9.35753405e-01 -5.55145204e-01 -1.07872605e+00 -7.64903724e-01 -4.86799330e-01 -2.26701081e-01 6.35141969e-01 -6.84706211e-01 2.12361097e-01 -3.11059654e-02 3.24545979e-01 -1.09648347e+00 3.14192712e-01 9.26847100e-01 4.55514669e-01 -4.10935014e-01 3.80320579e-01 2.09846333e-01 9.66920435e-01 -3.62147212e-01 -7.66514421e-01 2.90589392e-01 -6.74889147e-01 -7.03307390e-01 -2.08046481e-01 2.60729909e-01 -5.77537715e-01 -2.11119503e-01 1.00494325e+00 6.55319154e-01 -7.28062391e-02 1.06845450e+00 -1.30629754e+00 -6.96166337e-01 1.00013411e+00 5.37300482e-02 3.00492789e-03 1.06594451e-01 1.98718727e-01 9.75945175e-01 4.58140485e-02 5.41068852e-01 1.16058743e+00 6.84181809e-01 5.64936519e-01 -1.54585814e+00 -7.22382426e-01 4.01784927e-02 -2.51562893e-01 -9.37908649e-01 -6.28739774e-01 5.13645232e-01 -8.17554832e-01 6.20372593e-01 3.44562829e-01 6.90423429e-01 1.19184005e+00 7.88500309e-01 1.95228495e-02 1.14316905e+00 -2.99195260e-01 4.98284936e-01 7.41130039e-02 2.88087707e-02 3.89929801e-01 8.54928493e-01 7.87545204e-01 -1.40843345e-02 -9.20179427e-01 5.53506017e-01 2.21683905e-01 -3.43417525e-01 -4.77818966e-01 -1.06555963e+00 1.28978837e+00 3.89386356e-01 1.95807535e-02 -7.96601355e-01 -4.75357547e-02 5.65454066e-01 3.48679811e-01 4.28279817e-01 8.13865721e-01 -5.02870679e-01 3.20867270e-01 -1.01442373e+00 5.24373174e-01 9.71430480e-01 2.55720615e-01 9.05744359e-02 -6.87869638e-02 -3.03922504e-01 2.37875372e-01 2.77167171e-01 2.60543168e-01 4.65171903e-01 -1.39910793e+00 1.47719666e-01 3.26071858e-01 2.24008992e-01 -8.08367193e-01 -2.68021017e-01 -2.56763309e-01 -8.04857373e-01 4.40174788e-01 6.32334352e-01 -4.81534958e-01 -5.56112885e-01 2.03994989e+00 3.04752111e-01 -8.25972185e-02 4.13513571e-01 7.70740330e-01 2.18878433e-01 7.82254785e-02 5.90689063e-01 -6.45168066e-01 1.19370210e+00 9.11368355e-02 -8.20063412e-01 1.15293585e-01 9.56016839e-01 -2.32568890e-01 5.83727002e-01 -5.82029112e-02 -1.01711261e+00 1.55326188e-01 -7.08106220e-01 1.83618069e-01 6.84594586e-02 -7.59472609e-01 4.67012376e-01 9.46750104e-01 -6.93800569e-01 1.05782342e+00 -6.74978554e-01 -1.52708992e-01 7.62052715e-01 2.26743162e-01 -1.82175979e-01 1.79145053e-01 -1.39105606e+00 1.04542720e+00 2.56330192e-01 -3.60413581e-01 -5.53399563e-01 -1.17221129e+00 -9.43348885e-01 4.08628434e-01 3.96696091e-01 -1.36726546e+00 1.06493938e+00 -1.35626519e+00 -8.62810135e-01 7.85689950e-01 2.57050276e-01 -7.89755642e-01 1.02345932e+00 2.60068804e-01 -2.39271015e-01 -2.61016816e-01 2.58484721e-01 1.11225553e-01 4.32992697e-01 -1.12087870e+00 -3.13972503e-01 -7.95243382e-01 8.71838816e-03 2.42798328e-01 1.67829484e-01 -1.83228835e-01 8.26119184e-01 -7.60535359e-01 -4.13493633e-01 -8.66886854e-01 -3.61141950e-01 1.19814932e-01 -2.31808454e-01 -1.47004202e-02 1.83485180e-01 -6.39836013e-01 1.00513268e+00 -1.87806082e+00 -3.00677329e-01 1.79806113e-01 2.67499685e-01 -2.67370015e-01 2.83696532e-01 1.86202466e-01 -2.70269781e-01 4.50874448e-01 -6.39062881e-01 9.26950723e-02 1.69138499e-02 2.22955465e-01 -2.21570939e-01 6.61096334e-01 -1.30226780e-02 5.78342855e-01 -8.10672998e-01 -5.34924924e-01 1.56545252e-01 3.68933797e-01 -1.01428711e+00 1.03077874e-01 4.19613533e-02 1.06022388e-01 -2.36265495e-01 -5.83777837e-02 4.72540498e-01 -1.04562290e-01 7.95742989e-01 5.12232967e-02 -2.41470337e-03 4.84415799e-01 -1.01015460e+00 9.92455602e-01 -1.04550980e-01 2.91153193e-01 -7.47480541e-02 -1.00048578e+00 3.09528470e-01 5.31519353e-01 4.99407172e-01 -4.80731308e-01 2.67443061e-01 -2.49753036e-02 5.86348534e-01 -5.96003234e-01 1.35031626e-01 -1.23942673e+00 6.46453025e-03 7.59584904e-01 -2.03266591e-01 1.75247461e-01 -5.45705199e-01 1.20566152e-01 8.83317053e-01 -8.09740350e-02 6.69867575e-01 -6.12259030e-01 -2.87809074e-01 7.63494745e-02 8.53267670e-01 9.30026889e-01 -4.33652312e-01 6.10666811e-01 7.48280525e-01 -4.93202150e-01 -1.10362172e+00 -1.03985476e+00 -6.53817534e-01 5.73534727e-01 -3.38158101e-01 3.23189318e-01 -5.44355452e-01 -7.89519489e-01 2.86113918e-01 1.49235308e+00 -9.49457645e-01 -5.57733834e-01 -1.66692868e-01 -1.25612712e+00 5.90657830e-01 4.27443832e-01 -1.23248197e-01 -7.85644531e-01 -1.33168077e+00 3.18885773e-01 -8.01327005e-02 -3.64664644e-01 -3.98215085e-01 -7.40608722e-02 -9.69135940e-01 -1.54605663e+00 -6.84221327e-01 -4.53653149e-02 3.73104364e-01 -6.12607896e-01 1.21451259e+00 -1.14265129e-01 -8.81149992e-02 2.68709332e-01 2.03374505e-01 -9.41632807e-01 -9.03042495e-01 -5.76548636e-01 2.04301953e-01 -2.40909532e-01 3.93566728e-01 -4.31405216e-01 -1.03284800e+00 -1.94172695e-01 -1.07954645e+00 -1.30404592e-01 1.83350965e-01 1.02321947e+00 1.23986050e-01 -3.21673155e-01 7.36647069e-01 -1.43223524e+00 8.18557501e-01 -8.48657846e-01 -3.60566288e-01 2.41314873e-01 -1.37780285e+00 2.79552162e-01 3.84474665e-01 -3.80820900e-01 -1.21963656e+00 -4.65431213e-01 1.64496705e-01 -2.01518402e-01 -1.36985868e-01 7.31571317e-01 -1.76960424e-01 5.99410594e-01 8.34818840e-01 -4.55766678e-01 4.22051072e-01 -1.23097748e-01 2.54368067e-01 5.82387388e-01 -1.61996260e-02 -4.53213334e-01 -9.20523480e-02 6.77581668e-01 -4.17959839e-02 -1.64502695e-01 -5.49597383e-01 2.10739896e-01 -3.63693386e-02 2.40584031e-01 1.03368878e+00 -8.37878883e-01 -1.08747709e+00 -3.25007588e-01 -8.68791580e-01 -3.85779947e-01 -7.71647990e-01 7.93282688e-01 -6.69582903e-01 8.84582624e-02 -3.19389343e-01 -9.65300500e-01 -3.03715497e-01 -1.00649214e+00 5.18270969e-01 -3.02574337e-01 -7.84972310e-01 -1.16435218e+00 3.65537703e-01 1.71616927e-01 2.28765622e-01 6.69188797e-01 1.45182979e+00 -9.41061437e-01 -7.78812245e-02 -2.57194549e-01 3.82147320e-02 6.06777370e-02 3.89957845e-01 -3.32372189e-02 -8.13272178e-01 -2.18109101e-01 4.49777812e-01 5.45217772e-04 6.47178173e-01 8.85038018e-01 9.30651546e-01 -1.00856614e+00 -3.61411631e-01 3.31655532e-01 1.41014445e+00 4.74733531e-01 4.32401329e-01 1.12102196e-01 4.37492788e-01 1.09801149e+00 -1.14253916e-01 5.68325698e-01 6.67379200e-01 2.96928793e-01 3.93421799e-01 5.34287766e-02 4.30088371e-01 -2.37231508e-01 -9.51973125e-02 1.85396764e-02 -1.40233800e-01 -1.25828609e-01 -9.46997881e-01 6.65749371e-01 -1.64830387e+00 -1.29039872e+00 1.64516166e-01 2.49422026e+00 8.53860199e-01 3.46879810e-02 2.31119365e-01 -5.67113012e-02 6.84229374e-01 3.83993387e-02 -5.87701321e-01 -8.46677899e-01 1.09155029e-02 -7.81713054e-02 6.94843650e-01 6.70535564e-01 -6.11572385e-01 1.27028138e-03 6.18386030e+00 -7.42615713e-03 -8.58726561e-01 4.54393439e-02 1.16105354e+00 -2.34324798e-01 -8.39489579e-01 2.06838146e-01 2.04424068e-01 5.32393694e-01 1.15900648e+00 -4.99728113e-01 1.18195184e-01 4.98913080e-01 5.40977240e-01 1.55096591e-01 -1.49583554e+00 2.69670069e-01 -3.50016683e-01 -1.32848120e+00 1.33268774e-01 3.79476577e-01 6.86841905e-01 -2.06759974e-01 4.55219187e-02 4.41112593e-02 8.52287531e-01 -1.26407206e+00 7.53891468e-01 7.75699019e-01 7.20116079e-01 -6.90290093e-01 9.16861653e-01 2.95653343e-01 -5.62321320e-02 -9.60456505e-02 -4.47454937e-02 -4.07459050e-01 4.52225050e-03 6.31992459e-01 -9.71484184e-01 4.35686052e-01 3.84574741e-01 1.61715746e-01 2.46831775e-01 6.45562112e-01 2.65183955e-01 6.86790168e-01 4.17307429e-02 2.68161714e-01 -9.66563746e-02 1.67139694e-02 5.09423196e-01 7.75690556e-01 2.16448873e-01 2.55221575e-01 -3.55899692e-01 1.20306420e+00 -1.40142262e-01 7.10778758e-02 -1.01684809e+00 9.14157927e-02 3.04887742e-01 4.44524229e-01 -1.55924469e-01 -4.56989020e-01 -2.50017166e-01 5.78569770e-01 3.18373693e-03 4.57778960e-01 -3.78706694e-01 2.61072904e-01 7.47536540e-01 5.15595853e-01 -3.00798357e-01 6.74851775e-01 -6.66817367e-01 -1.16086698e+00 -3.05392414e-01 -1.09006226e+00 9.03818965e-01 -4.77031231e-01 -1.45775378e+00 2.97224242e-02 1.53155312e-01 -6.38054967e-01 -5.81413031e-01 -2.77738094e-01 -4.92673695e-01 1.15260684e+00 -1.20715058e+00 -5.11345029e-01 5.26048481e-01 2.26023719e-01 -3.96039300e-02 1.15667611e-01 1.11903441e+00 -2.17271134e-01 -3.07770669e-01 4.58625883e-01 1.90046757e-01 7.33643547e-02 6.51882887e-01 -1.32780266e+00 3.10096052e-02 4.68000352e-01 -2.59337932e-01 7.24565029e-01 9.44553733e-01 -9.81373489e-01 -6.56457186e-01 -1.04586172e+00 1.01416683e+00 -6.24778450e-01 2.26716325e-01 1.28135696e-01 -9.63133037e-01 8.99518967e-01 1.14406377e-01 -3.65226954e-01 9.97717381e-01 2.87125528e-01 -2.73080289e-01 2.89551437e-01 -1.58854926e+00 7.36464977e-01 8.38541090e-01 -5.60908556e-01 -1.07349288e+00 4.74900529e-02 8.19585264e-01 1.14557862e-01 -1.06387997e+00 5.78998446e-01 6.96123540e-01 -8.84236097e-01 7.16420949e-01 -1.25890231e+00 8.30110669e-01 2.11286679e-01 -2.12642342e-01 -1.34476686e+00 -3.68045598e-01 -2.85761923e-01 2.03605503e-01 8.09145689e-01 5.19655704e-01 -1.00859320e+00 5.05075872e-01 1.42759681e+00 1.44796163e-01 -6.52316391e-01 -1.17609274e+00 -2.50407934e-01 5.96423090e-01 -3.09659600e-01 9.55182731e-01 1.63466001e+00 9.07090828e-02 -8.97059496e-03 -2.28299186e-01 2.15233088e-01 8.30724537e-01 2.21209750e-01 1.82711005e-01 -1.28456175e+00 -3.47328037e-01 -3.23539883e-01 -1.51626408e-01 2.63300836e-01 2.28412643e-01 -8.34777176e-01 -3.39607388e-01 -1.34441769e+00 5.52950144e-01 -3.27883631e-01 -5.31025708e-01 2.08952546e-01 -4.90346849e-01 -4.95543361e-01 2.12793455e-01 8.53029788e-02 2.19799533e-01 5.84731638e-01 8.79236400e-01 2.20252350e-02 -1.58893138e-01 7.10356683e-02 -1.17080164e+00 7.62755334e-01 5.82791030e-01 -8.51799130e-01 -3.93535346e-01 -2.45875567e-01 1.87656596e-01 4.20695633e-01 7.76941597e-01 -1.62323758e-01 -7.18544275e-02 -4.51446414e-01 4.62390125e-01 4.24186051e-01 -3.14147353e-01 -1.02081680e+00 5.73937118e-01 9.81042862e-01 -9.33929801e-01 4.83648069e-02 1.47415981e-01 7.39997506e-01 1.54507801e-01 2.34906077e-02 6.13141775e-01 -2.75941581e-01 1.34934247e-01 4.76674251e-02 -4.74560201e-01 2.03166351e-01 8.03264439e-01 -1.20507345e-01 -2.76892394e-01 -3.74512672e-01 -9.45660830e-01 6.97229952e-02 7.00428963e-01 1.57112032e-01 3.67405325e-01 -1.06201327e+00 -1.00761926e+00 -4.06721048e-02 1.72128249e-02 -3.51216435e-01 4.16335672e-01 5.72853804e-01 -2.78238416e-01 1.57487229e-01 -1.50909647e-01 -5.38146980e-02 -8.63355100e-01 9.10251319e-01 8.02953184e-01 -1.03933893e-01 -5.20796955e-01 2.22745091e-01 6.60211742e-01 -3.63712251e-01 -2.24148914e-01 -9.75484103e-02 8.84819105e-02 6.73313886e-02 2.70021081e-01 4.53837037e-01 -1.50704443e-01 -3.20757151e-01 -2.90915400e-01 -3.26109231e-01 9.79274884e-02 -2.52750486e-01 1.34406829e+00 -1.49999246e-01 1.01254761e-01 4.92058307e-01 9.57704663e-01 -1.38937742e-01 -1.28375900e+00 2.36395493e-01 -2.35391229e-01 -2.77857691e-01 1.55416086e-01 -9.90823030e-01 -5.17612457e-01 5.67706585e-01 8.55088055e-01 2.10616842e-01 9.19068098e-01 -2.38412559e-01 6.82828343e-03 -1.53915122e-01 -1.41854286e-01 -8.09452772e-01 -7.91061819e-01 -3.93780649e-01 8.15101981e-01 -1.34911931e+00 2.32210048e-02 9.33565348e-02 -6.91042244e-01 5.05625188e-01 2.43990812e-02 -1.48366481e-01 7.48604298e-01 1.17140152e-01 1.09007210e-01 -2.90732294e-01 -7.95856297e-01 3.42751503e-01 1.94448858e-01 4.16703701e-01 6.69391274e-01 7.48030722e-01 -7.27489352e-01 6.68014109e-01 -3.47477376e-01 2.97951609e-01 6.52566969e-01 7.90894806e-01 3.05659562e-01 -5.83064020e-01 -3.49064112e-01 1.03980100e+00 -9.46645319e-01 -3.81214827e-01 -1.91452086e-01 8.27765524e-01 2.25401565e-01 7.15429246e-01 4.23337638e-01 1.30202621e-01 4.90779817e-01 3.72317463e-01 2.01500952e-01 -5.79927802e-01 -5.64560711e-01 -1.64427444e-01 3.17364663e-01 -3.39359075e-01 -5.64795375e-01 -9.77296948e-01 -8.87574911e-01 -4.09520179e-01 -1.03889875e-01 1.48958355e-01 2.24775016e-01 1.03565121e+00 5.65211236e-01 6.62714183e-01 1.68889776e-01 -7.10034445e-02 -9.25426543e-01 -5.36621094e-01 -4.51279104e-01 7.53830791e-01 6.38473928e-01 -2.79933125e-01 -4.91144538e-01 -1.62474409e-01]

[8.2918701171875, 5.519822120666504]

1fce28c9-d175-4df1-a61d-df71506aec38

word-substitution-in-short-answer-extraction

https://aclanthology.org/2016.gwc-1.11

https://aclanthology.org/2016.gwc-1.11.pdf

Word Substitution in Short Answer Extraction: A WordNet-based Approach

We describe the implementation of a short answer extraction system. It consists of a simple sentence selection front-end and a two phase approach to answer extraction from a sentence. In the first phase sentence classification is performed with a classifier trained with the passive aggressive algorithm utilizing the UIUC dataset and taxonomy and a feature set including word vectors. This phase outperforms the current best published results on that dataset. In the second phase, a sieve algorithm consisting of a series of increasingly general extraction rules is applied, using WordNet to find word types aligned with the UIUC classifications determined in the first phase. Some very preliminary performance metrics are presented.

['Adam Pease', 'Gerald Kurlandski', 'Maochen Guan', 'James Gung', 'Qingqing Cai']

gwc-2016-1

['sentence-classification']

['natural-language-processing']

[ 4.39009845e-01 4.72111069e-02 -5.24467342e-02 -5.89198232e-01 -9.64458883e-01 -8.34370792e-01 4.28008169e-01 8.01952660e-01 -1.05489445e+00 9.78423417e-01 4.86371219e-01 -5.07284343e-01 -4.48337160e-02 -7.79101968e-01 2.60862559e-01 -1.85660452e-01 1.72587961e-01 8.91201854e-01 7.28186011e-01 -6.94471478e-01 9.14701581e-01 9.00996104e-02 -1.52641654e+00 7.36032665e-01 7.79240310e-01 9.47451234e-01 -1.84631422e-02 1.45600903e+00 -7.35766947e-01 9.49478090e-01 -8.61117601e-01 -3.75449896e-01 -7.35061988e-02 -4.35420036e-01 -1.55674422e+00 -2.39138201e-01 2.56459266e-01 -1.07177041e-01 -4.62312363e-02 7.06948519e-01 7.61318982e-01 4.84053850e-01 6.66772425e-01 -4.28531319e-01 7.41725191e-02 5.02016068e-01 1.18243463e-01 5.25353253e-01 1.40226758e+00 -1.04568116e-02 1.38740885e+00 -9.86603618e-01 9.35662031e-01 1.07798374e+00 5.81818402e-01 5.83151877e-01 -1.04576588e+00 -1.60186931e-01 -2.21912414e-01 1.43661559e-01 -9.53633368e-01 -2.88953692e-01 5.65390646e-01 -4.18045223e-01 1.69900393e+00 7.59318054e-01 5.57621717e-01 5.39263964e-01 2.50877529e-01 6.76634490e-01 9.89790320e-01 -1.00256681e+00 5.87586761e-01 5.31237483e-01 1.26258266e+00 5.00204980e-01 -6.92337304e-02 -3.49116087e-01 -5.41011035e-01 -6.23525083e-01 -3.76379728e-01 -7.10747778e-01 -1.63041994e-01 6.40451089e-02 -5.82886994e-01 9.55706239e-01 1.43313678e-02 4.61609304e-01 -2.97234178e-01 -4.09329087e-01 9.44111764e-01 7.63626337e-01 5.59008837e-01 9.40618217e-01 -7.79506385e-01 -3.11741590e-01 -1.05390584e+00 6.67423129e-01 1.44317389e+00 6.93965673e-01 2.96185464e-01 -6.08370185e-01 -5.38057864e-01 8.50770950e-01 4.08054471e-01 3.13975736e-02 6.67074621e-01 -3.72145534e-01 6.68809354e-01 8.40819478e-01 6.58460483e-02 -8.86470497e-01 -6.40130699e-01 -3.71549100e-01 1.35542043e-02 -1.47031218e-01 1.89234331e-01 -5.12820899e-01 -7.84347653e-01 1.06458318e+00 3.73487234e-01 -6.33401215e-01 3.38056296e-01 4.69964832e-01 1.57006097e+00 4.47101057e-01 4.36205715e-02 -2.89343894e-01 1.62869012e+00 -7.01258898e-01 -6.62209928e-01 -2.16349453e-01 9.97130334e-01 -9.86751914e-01 7.76244402e-01 2.58888394e-01 -9.35085833e-01 -5.40261030e-01 -1.38583422e+00 -1.56941608e-01 -5.95576704e-01 2.03128569e-02 5.31068206e-01 9.81665194e-01 -1.22521126e+00 3.48892033e-01 -7.58416951e-02 -6.94352388e-01 -1.89558081e-02 6.80236518e-01 -2.17439026e-01 5.96857853e-02 -1.52433050e+00 1.09941649e+00 6.35359406e-01 -2.69453079e-01 1.41614407e-01 -2.02999383e-01 -8.79450738e-01 -1.25099376e-01 1.76167071e-01 -7.06975400e-01 1.46419597e+00 -6.41242325e-01 -1.50508296e+00 9.07517314e-01 -4.06405449e-01 -5.73948562e-01 -1.08837420e-02 -1.11897714e-01 -3.12795609e-01 6.20794535e-01 1.58455908e-01 3.88279319e-01 7.07458973e-01 -6.97670937e-01 -8.41746867e-01 -2.82895356e-01 2.18796149e-01 3.13954920e-01 -3.27424109e-01 4.80078697e-01 -2.17098385e-01 -1.99297592e-01 1.13391969e-02 -4.71191347e-01 -1.17491744e-01 -8.07010412e-01 -2.03224838e-01 -7.69031286e-01 5.97773373e-01 -7.90300071e-01 1.97023368e+00 -1.43888223e+00 9.27428082e-02 4.15828049e-01 3.68818641e-01 3.07881713e-01 -9.89450738e-02 6.71834826e-01 -3.36270034e-02 2.02910528e-02 -2.99890004e-02 -2.00001568e-01 -1.13061108e-01 -9.88230556e-02 -3.20572406e-01 -7.76013061e-02 1.77838877e-01 5.72763979e-01 -1.24855363e+00 -8.38421941e-01 -6.27950057e-02 -4.47370201e-01 -5.79424739e-01 3.84049594e-01 -3.90045881e-01 -2.74568975e-01 -5.19131660e-01 4.52723235e-01 3.29691380e-01 2.43216857e-01 2.71585822e-01 1.56676337e-01 -8.78288597e-02 1.01817286e+00 -1.05923617e+00 1.35119867e+00 -2.94762254e-01 5.87869346e-01 -1.50119811e-01 -7.23621011e-01 9.99383390e-01 5.15297472e-01 1.54424712e-01 -2.11625144e-01 4.20800626e-01 6.69659734e-01 6.63807988e-02 -8.47468913e-01 9.58161652e-01 -1.13582285e-02 -3.86556387e-01 4.44705278e-01 7.06372917e-01 -6.12737834e-01 9.00338829e-01 7.26547122e-01 1.55975556e+00 2.38297395e-02 6.74612403e-01 -4.23626572e-01 1.10056436e+00 4.92544830e-01 1.97356939e-01 9.39854443e-01 -1.45916879e-01 5.40783405e-01 5.01269221e-01 -4.45098698e-01 -5.13161242e-01 -6.59299672e-01 -2.41456792e-01 7.69642293e-01 -5.08454442e-01 -9.53235030e-01 -8.99889827e-01 -1.21352136e+00 -1.71133265e-01 8.75814736e-01 -7.11377621e-01 -6.72965273e-02 -5.06571412e-01 -8.23954165e-01 3.56737971e-01 1.59587547e-01 2.57588804e-01 -1.07973051e+00 -7.97231257e-01 2.56570339e-01 -2.12944210e-01 -7.42567658e-01 -2.34034076e-01 7.05810547e-01 -7.75298119e-01 -9.73703206e-01 -2.76288480e-01 -9.79533315e-01 5.12281716e-01 -1.98652208e-01 1.45341551e+00 4.34837967e-01 -1.73266809e-02 3.08790445e-01 -8.47536802e-01 -4.77501333e-01 -4.32545424e-01 7.49187171e-01 -1.11469992e-01 -2.45571703e-01 1.00691462e+00 1.03553303e-01 -3.67530823e-01 -3.68617892e-01 -6.24912620e-01 -4.71602201e-01 2.78231114e-01 9.44295466e-01 3.98092233e-02 -3.07967931e-01 8.77137780e-01 -1.15091980e+00 1.18002737e+00 -5.83427131e-01 -2.19418816e-02 1.85241967e-01 -5.97733676e-01 -5.13781980e-02 4.40755367e-01 -1.52434289e-01 -9.29133475e-01 2.34164298e-01 -8.71044636e-01 6.29419506e-01 -2.15162769e-01 7.22332954e-01 3.81881446e-02 -1.73675731e-01 9.11421418e-01 1.10030681e-01 -1.56814724e-01 -3.55292499e-01 1.19242325e-01 1.37159419e+00 7.93089792e-02 -1.70977011e-01 5.65945029e-01 -2.58839399e-01 -3.86614174e-01 -1.15658033e+00 -1.09389544e+00 -9.93930280e-01 -1.02542329e+00 -5.59043586e-01 9.72622991e-01 -2.01383740e-01 -1.39675990e-01 2.44056806e-01 -1.30051017e+00 1.30301595e-01 -5.00195801e-01 2.56255597e-01 -1.76551610e-01 4.44914460e-01 -5.40365994e-01 -8.91790330e-01 -7.89535165e-01 -5.99205673e-01 7.77671874e-01 3.79140586e-01 -1.09449041e+00 -9.16105509e-01 6.01909101e-01 4.85901356e-01 1.87600851e-01 -3.51830244e-01 1.06978643e+00 -1.47573650e+00 1.59781381e-01 -6.96206629e-01 2.65008330e-01 2.85537094e-01 -8.95043463e-02 -3.20463151e-01 -8.87229383e-01 6.27863966e-03 4.30374354e-01 -5.27887225e-01 9.40541506e-01 -2.07306385e-01 5.79274237e-01 -5.82999503e-03 -1.94831073e-01 1.58811919e-02 1.26192367e+00 3.46447259e-01 3.56725961e-01 5.10921538e-01 9.55207720e-02 7.85934448e-01 6.91509843e-01 -1.86005048e-02 1.96672127e-01 3.69841486e-01 -5.36255836e-02 4.39506084e-01 -2.54003704e-02 1.49425119e-01 2.17390940e-01 1.12458456e+00 5.07739782e-01 -2.52887934e-01 -1.05413961e+00 4.76971567e-01 -1.56225777e+00 -8.12835336e-01 -4.58256930e-01 1.97128212e+00 1.02089489e+00 7.70000756e-01 2.49591410e-01 6.26000106e-01 8.73410106e-02 1.97509021e-01 5.63278720e-02 -9.79983091e-01 -7.01930001e-02 7.45332778e-01 1.22260200e-02 9.59851027e-01 -1.06221604e+00 9.41310525e-01 7.16924906e+00 7.74926245e-01 -5.21620035e-01 4.96018752e-02 3.73279601e-01 -6.88236728e-02 -3.28556448e-01 1.15455449e-01 -1.01468098e+00 1.94602311e-01 1.07259536e+00 -1.37978655e-04 -2.71575451e-01 4.59393114e-01 -3.15925598e-01 -7.34787464e-01 -7.88196743e-01 3.79898131e-01 5.73050559e-01 -1.35628676e+00 3.70890610e-02 -5.81577599e-01 2.27552876e-01 -8.01836252e-02 -7.18337893e-01 6.34865940e-01 6.23881258e-02 -5.61284721e-01 4.79292750e-01 5.39334536e-01 4.86391634e-01 -7.59061694e-01 1.27641177e+00 4.04878587e-01 -1.14721310e+00 -6.08291775e-02 -2.53060669e-01 -4.00492519e-01 -1.06895477e-01 3.79035145e-01 -7.44134128e-01 5.84792733e-01 3.58724147e-01 2.82093197e-01 -8.31126511e-01 1.03124857e+00 -3.35212409e-01 7.71675467e-01 -2.44839683e-01 -1.00741434e+00 2.37253293e-01 -6.10040948e-02 9.53916311e-01 1.64921772e+00 -3.76540959e-01 3.29315990e-01 1.53805822e-01 3.98499109e-02 1.63121745e-01 6.54577255e-01 -4.67085123e-01 4.18388486e-01 3.49171609e-01 1.40871692e+00 -9.07597542e-01 -5.78137398e-01 -2.60743856e-01 8.86179566e-01 3.49676698e-01 1.09546453e-01 1.57850757e-02 -1.16856682e+00 6.75861314e-02 -2.89815277e-01 4.47564162e-02 -8.30767155e-02 -7.13120580e-01 -1.23765516e+00 -6.73626736e-02 -9.79098856e-01 7.35328138e-01 -3.85115802e-01 -8.86569679e-01 9.99000967e-01 -3.85229141e-02 -9.16751385e-01 -6.00145459e-01 -6.68691933e-01 -8.18191290e-01 1.07242894e+00 -9.24568594e-01 -4.47194278e-01 2.43673429e-01 4.53788787e-02 1.00014341e+00 -3.63815248e-01 1.36245072e+00 6.13065846e-02 -6.23506546e-01 4.96905506e-01 -2.23076642e-01 2.76336193e-01 3.02170873e-01 -1.70101190e+00 2.13729799e-01 8.91474247e-01 1.75940380e-01 7.54163623e-01 9.22645926e-01 -8.70481730e-01 -1.14513493e+00 -6.19399786e-01 1.86746860e+00 -7.23407507e-01 8.16553175e-01 -5.70337355e-01 -8.04696977e-01 1.44250706e-01 5.75384557e-01 -6.95907950e-01 1.07285357e+00 3.06998461e-01 1.75787598e-01 1.25081450e-01 -1.10732031e+00 4.95287478e-01 4.17450666e-01 -5.26055813e-01 -1.32071888e+00 5.05714893e-01 6.18201911e-01 -4.64919329e-01 -6.98663533e-01 3.22112232e-01 3.67964417e-01 -3.16205293e-01 5.79871893e-01 -7.91521490e-01 2.90817022e-01 -6.59461543e-02 -1.22335427e-01 -1.19254684e+00 -1.23197153e-01 -7.36917436e-01 -2.29465365e-01 1.26432347e+00 9.97525275e-01 -5.45128047e-01 6.06162012e-01 3.49662900e-01 1.35714799e-01 -1.21495605e+00 -9.92296219e-01 -3.04534972e-01 7.17236148e-03 -4.34686542e-01 1.13811873e-01 2.97944158e-01 7.02025115e-01 1.32977474e+00 2.58725137e-01 -5.63066840e-01 2.74837703e-01 6.29506260e-02 3.18187892e-01 -1.28981662e+00 -1.37802184e-01 -4.13477689e-01 -2.46251017e-01 -1.02020049e+00 1.04877606e-01 -1.02657044e+00 1.91114321e-01 -1.64584422e+00 4.09286767e-02 3.18480730e-02 -1.00699089e-01 -8.61721695e-04 -5.83306909e-01 1.08535714e-01 -6.09201519e-03 -3.59475985e-02 -6.81209683e-01 2.49410048e-01 4.10723746e-01 -4.40796874e-02 -2.41035774e-01 2.29414061e-01 -5.62257886e-01 7.88001776e-01 9.85969067e-01 -6.58171713e-01 -4.25888658e-01 1.44007709e-02 5.22671700e-01 -2.61409312e-01 -3.92367095e-01 -9.56094384e-01 2.90001243e-01 3.02241474e-01 2.21345991e-01 -9.78620887e-01 8.24762732e-02 -2.53623158e-01 -8.45175445e-01 7.06659257e-01 -5.75091183e-01 3.27919155e-01 6.86469376e-02 2.62471110e-01 -3.09989214e-01 -1.18002594e+00 5.95309019e-01 -6.89021796e-02 -6.02238595e-01 -2.63530284e-01 -1.08524585e+00 2.27068469e-01 7.75995791e-01 -1.05774797e-01 -4.46613207e-02 -2.77043432e-01 -5.98125041e-01 5.40844858e-01 -7.45347664e-02 3.42668474e-01 9.18690383e-01 -8.47845972e-01 -8.28102112e-01 1.08994199e-02 1.31075636e-01 -5.49568713e-01 -2.40066588e-01 7.25227654e-01 -7.73555875e-01 7.38009751e-01 1.78290844e-01 -2.13990927e-01 -1.74906683e+00 3.64958167e-01 3.50914478e-01 -8.08411479e-01 -3.77868235e-01 1.07982004e+00 -8.37221026e-01 -7.79856622e-01 7.91440830e-02 -8.50728601e-02 -1.25900817e+00 6.49005473e-01 6.22081995e-01 9.14231911e-02 6.53199911e-01 -4.71249044e-01 -4.90922689e-01 1.46224707e-01 -2.35510364e-01 -5.44420362e-01 9.81546938e-01 -1.51895687e-01 -3.44285905e-01 6.20029151e-01 1.25803089e+00 1.73647553e-01 5.93530852e-03 -2.72317231e-01 7.77751863e-01 -1.53768092e-01 -5.62436171e-02 -8.45432520e-01 -2.60651588e-01 4.43388134e-01 3.20578575e-01 5.05171657e-01 1.19630396e+00 -1.77297443e-01 8.98847759e-01 6.00922525e-01 2.45038912e-01 -1.52923226e+00 -1.30257621e-01 1.23109186e+00 8.65681231e-01 -8.70998085e-01 -4.87009920e-02 -5.24891257e-01 -2.67599195e-01 1.33111703e+00 5.07119536e-01 -2.76296854e-01 8.16625834e-01 3.47883910e-01 2.17395827e-01 -4.08083767e-01 -1.14623380e+00 -4.81738389e-01 4.90588665e-01 2.27544397e-01 7.94919491e-01 -2.35419944e-01 -1.36057508e+00 6.77028716e-01 -4.69659746e-01 -2.85284489e-01 4.87555653e-01 1.14983916e+00 -7.94394493e-01 -1.40738630e+00 -1.30191103e-01 1.08391058e+00 -6.86432958e-01 -3.58652323e-01 -1.08112347e+00 5.13303876e-01 -2.91972328e-02 1.52711713e+00 -1.48413867e-01 -8.29738021e-01 5.43218136e-01 4.55097854e-01 3.28086376e-01 -1.27673244e+00 -1.55695498e+00 -4.45478529e-01 1.00594735e+00 -2.56206632e-01 -9.38269645e-02 -1.08905768e+00 -1.03915811e+00 3.41170877e-01 -5.25786757e-01 7.56222725e-01 4.26088661e-01 1.23147452e+00 -5.57546094e-02 5.06973803e-01 6.16274774e-01 -4.83654976e-01 -6.68643892e-01 -1.21877682e+00 -5.39344773e-02 7.93397799e-02 1.56587318e-01 -1.74932823e-01 -3.51867825e-01 -4.54252481e-01]

[11.987848281860352, 9.427606582641602]

dc35a310-19f6-41b5-aaa5-a578f8294a9b

attention-based-modeling-for-emotion

1906.0702

https://arxiv.org/abs/1906.07020v1

https://arxiv.org/pdf/1906.07020v1.pdf

Attention-based Modeling for Emotion Detection and Classification in Textual Conversations

This paper addresses the problem of modeling textual conversations and detecting emotions. Our proposed model makes use of 1) deep transfer learning rather than the classical shallow methods of word embedding; 2) self-attention mechanisms to focus on the most important parts of the texts and 3) turn-based conversational modeling for classifying the emotions. The approach does not rely on any hand-crafted features or lexicons. Our model was evaluated on the data provided by the SemEval-2019 shared task on contextual emotion detection in text. The model shows very competitive results.

['Jérôme Azé', 'Waleed Ragheb', 'Sandra Bringay', 'Maximilien Servajean']

2019-06-14

['emotion-recognition-in-conversation']

['natural-language-processing']

[-1.31878287e-01 2.86256760e-01 -1.01011053e-01 -6.38339520e-01 -5.60278714e-01 -1.50343224e-01 9.44692612e-01 4.53310281e-01 -7.21353531e-01 6.68028355e-01 7.41741598e-01 -3.08920264e-01 3.90653163e-01 -5.84523857e-01 -1.84659749e-01 -4.58422363e-01 -7.44946003e-02 5.70949674e-01 -2.33974785e-01 -6.86231136e-01 4.09919798e-01 1.39876723e-01 -1.26477444e+00 7.62155831e-01 4.84443426e-01 1.12103498e+00 -6.95205554e-02 1.11894095e+00 -6.48255408e-01 1.24678993e+00 -6.37971699e-01 -4.27749872e-01 -3.26228201e-01 -6.03008747e-01 -1.01365125e+00 -8.97380188e-02 -1.08579502e-01 7.40794837e-02 -1.33132607e-01 7.08566308e-01 5.44663310e-01 3.79731745e-01 9.08262432e-01 -1.16374111e+00 -5.78392506e-01 8.45782578e-01 -1.52071759e-01 4.29412663e-01 4.21298057e-01 -1.51247233e-01 1.34277272e+00 -1.08681357e+00 4.67072010e-01 1.38729191e+00 5.90222955e-01 5.88541687e-01 -7.66590595e-01 -4.24704939e-01 1.87932923e-01 4.16920334e-01 -9.06642497e-01 -2.76949257e-01 1.18386531e+00 -4.02915925e-01 1.61786962e+00 2.33986512e-01 5.31782746e-01 1.54347920e+00 3.08663696e-01 7.70198345e-01 1.18238628e+00 -7.63526320e-01 2.56215364e-01 6.56468749e-01 6.53375387e-01 4.98794734e-01 -6.02495134e-01 -3.02950174e-01 -6.21367455e-01 -4.59762633e-01 -1.45110833e-02 -2.79920667e-01 -1.39549419e-01 1.24029800e-01 -8.70693922e-01 1.32596695e+00 1.58919394e-01 8.30141425e-01 -7.50292897e-01 3.75196710e-02 9.43920791e-01 5.28742194e-01 1.16749191e+00 5.01832843e-01 -7.90914178e-01 -5.37534356e-01 -6.45330429e-01 -1.13797128e-01 1.00541222e+00 5.39960861e-01 6.34786427e-01 -1.53117687e-01 -4.68246303e-02 9.50830162e-01 3.00123781e-01 6.22803229e-04 6.27875209e-01 -2.93299139e-01 4.43478763e-01 5.70351362e-01 1.13076650e-01 -1.14078724e+00 -5.04329503e-01 6.81452639e-03 -4.08743441e-01 -1.83826119e-01 -1.07962780e-01 -7.40670323e-01 -4.89168644e-01 1.58613515e+00 3.56213190e-02 -5.97856119e-02 4.27476019e-01 4.08053130e-01 1.09032917e+00 1.05289936e+00 2.56264448e-01 -1.34488270e-01 1.36288917e+00 -1.20986438e+00 -1.25784039e+00 -2.50648588e-01 6.91741705e-01 -7.79646873e-01 1.04652524e+00 2.30101809e-01 -9.92325604e-01 -3.74046683e-01 -9.42294776e-01 -2.16416195e-01 -9.71560776e-01 -6.43766522e-02 6.61532581e-01 6.28308117e-01 -1.12177694e+00 1.03644811e-01 -1.71295464e-01 -5.97895861e-01 -2.51956731e-02 2.67471999e-01 -1.61956698e-01 4.90190595e-01 -1.56946433e+00 1.16958380e+00 1.08197518e-01 9.03512090e-02 -4.08010274e-01 -2.65173763e-01 -1.08699083e+00 1.49903268e-01 9.33218747e-02 -2.32797369e-01 1.22501004e+00 -1.41463387e+00 -2.00062656e+00 9.56612468e-01 -3.71517003e-01 -3.04301798e-01 7.40069598e-02 -4.37676251e-01 -5.11067092e-01 1.44520655e-01 -4.12780225e-01 4.50478435e-01 6.06530964e-01 -9.76243854e-01 -3.90753895e-01 -1.24009751e-01 8.23664963e-02 2.14651823e-01 -7.98665106e-01 5.34535170e-01 -2.11955607e-02 -4.11364108e-01 -6.81625485e-01 -8.21153343e-01 -2.87089258e-01 -6.59537673e-01 -1.63384810e-01 -7.99580872e-01 8.38029742e-01 -6.62137032e-01 9.82220054e-01 -1.93979919e+00 1.66657344e-01 3.57206687e-02 9.06026065e-02 1.71187758e-01 -2.98960537e-01 1.03183413e+00 -2.37095967e-01 8.91354531e-02 2.76783377e-01 -5.89890599e-01 2.86235809e-01 5.04544191e-02 -3.52381885e-01 2.13858560e-01 2.50456095e-01 9.30193901e-01 -8.71161580e-01 -5.23102164e-01 4.04950887e-01 5.86187065e-01 -4.53938991e-01 4.36784506e-01 -3.05213213e-01 1.62973553e-01 -4.96447682e-01 5.95338084e-02 2.12689012e-01 3.11312750e-02 2.45606333e-01 -2.19357863e-01 -2.14514643e-01 7.86832929e-01 -5.41263938e-01 1.38609588e+00 -9.49958384e-01 9.17342067e-01 -7.52742887e-02 -1.15293765e+00 1.15232778e+00 7.76881754e-01 4.75920707e-01 -4.60397661e-01 6.53299868e-01 -3.23285103e-01 -1.06192216e-01 -7.26685524e-01 5.61111689e-01 -3.89681548e-01 -5.02110004e-01 6.31164789e-01 4.02795821e-01 6.60758838e-02 -7.05192089e-02 2.90254265e-01 8.90116811e-01 -3.59984815e-01 4.53542352e-01 -5.33043683e-01 7.06185699e-01 -1.48077860e-01 2.79431194e-01 3.16363335e-01 -3.01755458e-01 5.64080365e-02 7.24960566e-01 -6.40036941e-01 -7.42777526e-01 -1.72760397e-01 9.62491333e-02 1.61255813e+00 -2.29101360e-01 -5.99804044e-01 -6.77222073e-01 -8.46570611e-01 -3.10322583e-01 9.57773864e-01 -1.21697247e+00 -1.58455685e-01 -8.22084323e-02 -6.36412501e-01 5.01206160e-01 5.06044447e-01 3.24539505e-02 -1.71053505e+00 -4.35446233e-01 2.59179324e-01 -5.55151224e-01 -1.11353481e+00 -3.77477020e-01 5.82503200e-01 -4.87299949e-01 -8.14028680e-01 -3.49014610e-01 -8.96518171e-01 2.55128771e-01 -1.06442630e-01 1.24024498e+00 -2.16424227e-01 -1.42589107e-01 6.99414670e-01 -7.43943572e-01 -8.43387246e-01 -4.07806337e-01 1.45400643e-01 -3.39258432e-01 3.42961431e-01 1.07243931e+00 -3.29514146e-01 -2.12631688e-01 -1.73962384e-01 -4.61662859e-01 -2.23443151e-01 4.41847950e-01 9.54222560e-01 -2.35606283e-01 -2.92885154e-01 9.87702489e-01 -9.38105702e-01 1.29468703e+00 -6.27758741e-01 2.40197241e-01 1.60962343e-01 -4.35301930e-01 3.26826260e-03 4.56419706e-01 -5.60405552e-01 -1.19772708e+00 -2.97430426e-01 -4.83996004e-01 3.90003361e-02 -3.23055089e-01 5.92965186e-01 -1.53963380e-02 3.53950441e-01 3.89529884e-01 6.32951483e-02 -9.34938192e-02 -3.04086506e-01 4.96229857e-01 1.20240641e+00 -6.03187978e-02 -4.76216018e-01 -1.80103723e-02 1.76378697e-01 -6.37791634e-01 -1.22566295e+00 -7.85901785e-01 -5.86455464e-01 -5.00437975e-01 -3.71265590e-01 1.29089189e+00 -8.41664851e-01 -7.52137959e-01 1.75933599e-01 -1.37273955e+00 -3.91383469e-01 -6.82066530e-02 4.72120941e-01 -5.93187094e-01 2.33999804e-01 -1.18870986e+00 -1.12619972e+00 -7.87498534e-01 -7.70471871e-01 9.11622703e-01 1.01257926e-02 -6.61975443e-01 -1.28084636e+00 5.25574744e-01 3.32554162e-01 5.94600379e-01 1.76102549e-01 9.99868095e-01 -1.20125568e+00 5.12583971e-01 -3.80649328e-01 9.38801393e-02 4.53733832e-01 -2.77038198e-02 5.18289059e-02 -1.32111371e+00 7.44603053e-02 2.33292468e-02 -9.43941593e-01 6.66383684e-01 1.56878904e-01 7.64593661e-01 -2.69370764e-01 -1.57087892e-01 -1.02884687e-01 1.04655099e+00 3.22348207e-01 5.10818839e-01 1.51271820e-01 2.63367206e-01 1.02077305e+00 4.30733860e-01 7.23115861e-01 6.00412309e-01 4.69873369e-01 3.43690097e-01 -2.34692052e-01 3.87516648e-01 1.70920454e-02 5.13287127e-01 1.24912453e+00 -4.33236994e-02 -5.09252906e-01 -8.46105635e-01 7.63480544e-01 -1.94717455e+00 -9.32659030e-01 -1.91914156e-01 1.29533720e+00 8.05199862e-01 1.45456925e-01 -1.05149290e-02 -2.04976052e-02 5.52836657e-01 4.79175866e-01 -3.41789126e-02 -1.48153174e+00 8.33577439e-02 4.88815904e-01 -1.93249941e-01 8.33289325e-01 -1.15521872e+00 1.04680109e+00 6.65763426e+00 4.38189566e-01 -1.03680837e+00 3.41024637e-01 6.15698814e-01 -4.35644314e-02 -1.67926043e-01 -4.32716846e-01 -4.88043606e-01 3.60050529e-01 1.37221098e+00 -1.93568602e-01 5.20985387e-02 8.77630949e-01 4.67942022e-02 3.74690443e-02 -9.68247652e-01 8.05503905e-01 4.61554080e-01 -9.60243046e-01 -1.80569604e-01 -1.04958311e-01 4.69369650e-01 1.70436967e-02 -2.03965992e-01 9.66536283e-01 2.23086566e-01 -9.90748107e-01 4.15729463e-01 4.50130641e-01 1.93761587e-01 -1.17780364e+00 1.20749748e+00 2.78312922e-01 -7.55591035e-01 7.24651590e-02 -4.40151572e-01 -4.40433741e-01 1.71227697e-02 4.27201718e-01 -8.68009865e-01 1.70993477e-01 5.77840149e-01 6.67539537e-01 -1.49590656e-01 9.52053145e-02 -2.68021405e-01 8.75564575e-01 3.96326408e-02 -7.85075068e-01 6.59735084e-01 -1.34778589e-01 2.87477225e-01 1.94516706e+00 -1.18202128e-01 1.69587508e-01 -6.04042225e-02 4.59741235e-01 -1.08426385e-01 7.01402724e-01 -7.19466805e-01 -3.51918906e-01 1.48800970e-03 1.52734745e+00 -4.43570405e-01 -5.90995193e-01 -6.86417818e-01 9.83863413e-01 4.68671530e-01 2.77797669e-01 -6.95061922e-01 -5.81658959e-01 8.28301668e-01 -4.09394234e-01 5.45127094e-01 5.15280627e-02 -3.17530637e-03 -1.03535485e+00 -3.75374168e-01 -9.02519822e-01 2.47452125e-01 -6.75978780e-01 -1.56448972e+00 1.25783157e+00 -3.06455463e-01 -4.97170746e-01 -5.26813149e-01 -7.04516888e-01 -1.01328719e+00 8.87443364e-01 -1.33089709e+00 -9.92747784e-01 3.51425447e-03 7.13509202e-01 7.65753508e-01 -4.99924943e-02 1.37353790e+00 2.79868782e-01 -4.61110353e-01 4.77229893e-01 -9.12050158e-02 1.87417105e-01 9.76667345e-01 -1.44117546e+00 1.75619155e-01 1.79532260e-01 -2.90633421e-02 3.52159351e-01 9.72965419e-01 -2.29588911e-01 -9.83349860e-01 -6.82958245e-01 1.65372741e+00 -6.16962492e-01 9.54186320e-01 -8.19153845e-01 -7.88821340e-01 8.48094702e-01 1.21047258e+00 -3.73999864e-01 1.32163107e+00 7.39927888e-01 -2.90251046e-01 1.86328128e-01 -1.01199472e+00 3.70367408e-01 2.89729655e-01 -6.45793200e-01 -1.08582890e+00 3.58148992e-01 7.30151355e-01 1.44101635e-01 -6.91179514e-01 1.57656059e-01 4.01150376e-01 -7.24828422e-01 6.11318231e-01 -1.14380276e+00 7.02959418e-01 5.90107203e-01 -1.43889025e-01 -1.56565619e+00 -1.31035700e-01 -5.98728001e-01 -3.83953191e-02 1.27616286e+00 5.13991535e-01 -4.42645371e-01 4.18053240e-01 5.61499417e-01 -1.45361498e-01 -9.37870741e-01 -5.84970236e-01 -9.22308676e-03 7.69675523e-02 -5.14029205e-01 1.74110249e-01 1.30804598e+00 6.85813427e-01 1.00188124e+00 -5.93309939e-01 -2.42750749e-01 4.99392226e-02 1.36343881e-01 7.27562070e-01 -1.09477460e+00 -8.34812820e-02 -5.37006736e-01 -3.42696667e-01 -6.09571159e-01 6.16809011e-01 -6.37827218e-01 1.47723243e-01 -1.39278674e+00 2.18253419e-01 1.29654646e-01 -6.28157377e-01 4.75966543e-01 -1.92380130e-01 3.58104445e-02 -4.00890261e-02 -5.45724034e-01 -8.80201936e-01 1.00868690e+00 6.74764872e-01 -9.14603323e-02 3.43681430e-03 -2.79181331e-01 -7.85334826e-01 7.03331113e-01 1.15616465e+00 -4.35175925e-01 -3.24639410e-01 6.04941510e-02 3.61172050e-01 -8.85170419e-03 -5.92377298e-02 -5.56685388e-01 5.83851673e-02 -1.69690384e-03 1.02258436e-02 -6.62217498e-01 5.80438018e-01 -6.99370086e-01 -5.26281297e-01 2.37976789e-01 -8.03940237e-01 1.54855892e-01 4.47359294e-01 4.55207348e-01 -3.83402914e-01 -2.78574646e-01 5.61306894e-01 6.48951381e-02 -5.91690421e-01 -2.05220893e-01 -1.17897391e+00 -1.28738703e-02 9.67138410e-01 2.52881885e-01 -9.32461917e-02 -8.75059426e-01 -9.01779175e-01 1.14631623e-01 -8.28679726e-02 7.56129205e-01 3.43082160e-01 -1.23005295e+00 -6.42734230e-01 -8.22556615e-02 2.80150265e-01 -8.15596700e-01 1.73795059e-01 5.11516333e-01 -2.33079139e-02 6.52931035e-01 -2.28584632e-01 -8.31952915e-02 -1.36970890e+00 7.61093020e-01 2.13376999e-01 -6.39953613e-01 -3.36660713e-01 9.06190693e-01 1.24424115e-01 -6.48571730e-01 2.76082814e-01 -2.22409025e-01 -5.06862402e-01 5.04658401e-01 6.74427092e-01 9.96431112e-02 7.27595836e-02 -7.39286482e-01 -5.00783443e-01 2.00600281e-01 -7.12627172e-02 -3.84355485e-01 1.60411584e+00 -1.60396710e-01 -2.44686335e-01 1.02708781e+00 1.52864385e+00 4.88486812e-02 -5.56118786e-01 -5.49786150e-01 1.26487106e-01 1.95940733e-01 3.93872738e-01 -1.02944481e+00 -7.00770497e-01 1.28551209e+00 3.02298576e-01 4.01930898e-01 7.87275851e-01 -7.94772953e-02 9.10076261e-01 5.16980767e-01 6.01200908e-02 -1.36648083e+00 3.80323976e-01 8.84621203e-01 9.85944390e-01 -1.27048159e+00 -4.76646066e-01 1.35296835e-02 -1.04366684e+00 1.30023777e+00 5.64901590e-01 -2.07441911e-01 8.64707589e-01 3.92113924e-01 5.37708521e-01 -4.27176625e-01 -1.45218456e+00 -2.00385526e-01 2.43546799e-01 2.94961065e-01 9.16060090e-01 9.68820453e-02 -6.28912210e-01 8.65142167e-01 6.50583729e-02 -3.11858416e-01 3.08096766e-01 9.53632414e-01 -4.43432689e-01 -1.03933358e+00 -6.45797551e-02 1.73921064e-01 -6.98579013e-01 -2.84932613e-01 -1.03211761e+00 6.11076891e-01 -5.53396046e-02 1.25377059e+00 1.27932981e-01 -5.53868651e-01 2.40732506e-01 7.26796687e-01 1.69450313e-01 -6.87606215e-01 -9.70753133e-01 9.21954438e-02 5.01654625e-01 -3.60451907e-01 -5.80188394e-01 -5.89808643e-01 -9.91969347e-01 -9.84639078e-02 -2.22548410e-01 6.84642375e-01 8.94514859e-01 9.90498960e-01 5.25579214e-01 7.19544172e-01 7.66993523e-01 -9.33760941e-01 -2.77154565e-01 -1.63215673e+00 -4.45087612e-01 3.96315575e-01 2.26310626e-01 -3.15577239e-01 -4.88784313e-01 -1.32333711e-01]

[12.979376792907715, 6.19352388381958]

54ee8427-921a-4a8f-9c0d-2bf94998743f

aim-adapting-image-models-for-efficient-video

2302.03024

https://arxiv.org/abs/2302.03024v1

https://arxiv.org/pdf/2302.03024v1.pdf

AIM: Adapting Image Models for Efficient Video Action Recognition

Recent vision transformer based video models mostly follow the ``image pre-training then finetuning" paradigm and have achieved great success on multiple video benchmarks. However, full finetuning such a video model could be computationally expensive and unnecessary, given the pre-trained image transformer models have demonstrated exceptional transferability. In this work, we propose a novel method to Adapt pre-trained Image Models (AIM) for efficient video understanding. By freezing the pre-trained image model and adding a few lightweight Adapters, we introduce spatial adaptation, temporal adaptation and joint adaptation to gradually equip an image model with spatiotemporal reasoning capability. We show that our proposed AIM can achieve competitive or even better performance than prior arts with substantially fewer tunable parameters on four video action recognition benchmarks. Thanks to its simplicity, our method is also generally applicable to different image pre-trained models, which has the potential to leverage more powerful image foundation models in the future. The project webpage is \url{https://adapt-image-models.github.io/}.

['Mu Li', 'Chen Chen', 'Aston Zhang', 'Yusheng Xie', 'Yi Zhu', 'Taojiannan Yang']

2023-02-06

['action-classification', 'video-understanding']

['computer-vision', 'computer-vision']

[ 3.27272087e-01 6.79525435e-02 -3.74495745e-01 -2.42623851e-01 -5.54511309e-01 -6.06925368e-01 6.62869573e-01 -6.65308475e-01 -3.01486820e-01 4.45071161e-01 3.44774187e-01 -2.36400589e-01 1.28641814e-01 -5.20964444e-01 -1.23680639e+00 -3.71951759e-01 2.08106995e-01 1.84062093e-01 5.36490262e-01 -1.25491202e-01 1.67215452e-01 2.06500888e-01 -1.48701525e+00 6.39560342e-01 5.65092325e-01 9.61050034e-01 3.43551785e-01 9.11872566e-01 1.94797173e-01 1.37581635e+00 -7.60118738e-02 -4.57646757e-01 4.02659565e-01 -3.90639603e-01 -1.03044069e+00 2.45945171e-01 8.87486279e-01 -8.05421114e-01 -8.54205906e-01 8.12176108e-01 1.16184458e-01 2.22896874e-01 1.72995523e-01 -1.23914087e+00 -8.94871891e-01 6.02375388e-01 -4.35705334e-01 4.69726354e-01 3.03457916e-01 5.11519432e-01 8.53696883e-01 -7.39173770e-01 7.14596748e-01 1.02802598e+00 7.53444910e-01 8.42179716e-01 -1.14047539e+00 -6.85235739e-01 4.49608207e-01 7.06618905e-01 -1.05325627e+00 -7.53208220e-01 5.41089118e-01 -3.53096426e-01 1.19556451e+00 1.13020509e-01 8.35247040e-01 1.39068198e+00 -7.63293132e-02 9.59362626e-01 9.44648087e-01 -2.46071726e-01 4.69936691e-02 -1.75434455e-01 -2.18724087e-01 9.42092657e-01 -1.59604073e-01 1.12976789e-01 -7.93710470e-01 3.49937409e-01 1.10929775e+00 2.99370646e-01 -4.01864499e-01 -4.93428856e-01 -1.30051649e+00 4.61592734e-01 5.89531898e-01 2.33910248e-01 -2.01107621e-01 7.21623659e-01 4.35764283e-01 2.10807264e-01 1.90897331e-01 2.82961398e-01 -4.86892998e-01 -4.70644891e-01 -1.00833392e+00 1.83948398e-01 4.66615140e-01 9.31501627e-01 6.98680162e-01 1.55781791e-01 -1.07474051e-01 6.36250675e-01 1.55176716e-02 3.57661724e-01 5.02765357e-01 -1.67897117e+00 3.71200770e-01 3.68446350e-01 9.11067352e-02 -7.32673168e-01 -1.16333611e-01 -2.41296038e-01 -6.91316545e-01 2.40464032e-01 4.71275300e-01 1.86420068e-01 -1.13011694e+00 1.76466048e+00 1.27178654e-01 7.54866421e-01 -1.25122685e-02 8.61275315e-01 7.21566975e-01 5.83397746e-01 2.01082930e-01 1.00192986e-01 1.09178996e+00 -1.48176551e+00 -3.42438757e-01 -3.41847956e-01 3.93738300e-01 -5.95316648e-01 1.40016592e+00 3.83980334e-01 -1.41731727e+00 -6.90921128e-01 -8.95416141e-01 -1.80218801e-01 -8.87614340e-02 -1.20489029e-02 8.67280900e-01 2.50506699e-01 -1.43027794e+00 7.61381626e-01 -1.09632039e+00 -6.89829767e-01 7.61693597e-01 3.97377133e-01 -5.64130604e-01 -3.06778431e-01 -8.21469665e-01 6.65939212e-01 2.51743823e-01 -3.97231057e-02 -1.38209665e+00 -9.65021431e-01 -7.59152055e-01 -7.16938078e-02 4.68657583e-01 -1.18156767e+00 1.45314837e+00 -1.38366103e+00 -1.72724175e+00 8.48011017e-01 -1.51790023e-01 -7.52740443e-01 5.39622068e-01 -4.86967713e-01 -8.86555687e-02 4.36137229e-01 -3.20537575e-02 1.09410608e+00 1.13992512e+00 -9.95842040e-01 -6.41284227e-01 -1.04305446e-01 5.49189627e-01 2.27431893e-01 -5.16804039e-01 -8.49604905e-02 -9.42454398e-01 -7.49424875e-01 -2.60953486e-01 -1.01597059e+00 -2.09483653e-01 1.24059491e-01 9.96438712e-02 4.88597527e-02 8.00097108e-01 -5.78378856e-01 1.03202116e+00 -2.06679463e+00 4.06248719e-01 -3.08308840e-01 1.49384469e-01 3.51047456e-01 -3.91954839e-01 2.01728746e-01 -2.54938193e-02 -5.24137430e-02 -1.18493050e-01 -3.22063923e-01 -9.45729688e-02 3.45090449e-01 -4.54790205e-01 1.48102701e-01 1.02798998e-01 1.17332423e+00 -8.83117616e-01 -4.32052910e-01 4.52991813e-01 5.71933448e-01 -1.01209569e+00 1.81222662e-01 -4.79553998e-01 5.78264236e-01 -3.01617473e-01 7.03259945e-01 3.60369265e-01 -5.85329354e-01 1.84532285e-01 -5.41974843e-01 1.26758426e-01 -2.34939810e-02 -7.62876153e-01 2.03807783e+00 -4.07941878e-01 7.47536540e-01 -8.03021640e-02 -1.07092476e+00 2.48054296e-01 1.55852392e-01 5.57623446e-01 -8.38183701e-01 -2.48987272e-01 -8.02106708e-02 -1.77819118e-01 -6.83219314e-01 3.76511842e-01 1.67081673e-02 1.88485250e-01 3.36262017e-01 3.06101769e-01 -6.26405329e-02 1.95785463e-01 3.38725775e-01 1.41750026e+00 9.16947067e-01 5.09791076e-02 5.28591760e-02 4.88209456e-01 1.09213352e-01 5.07942498e-01 7.20955312e-01 -4.76324588e-01 6.54354095e-01 9.83453915e-02 -6.85862422e-01 -1.11308682e+00 -1.19809401e+00 2.00848937e-01 1.41402817e+00 1.92603588e-01 -6.05829000e-01 -8.79538655e-01 -7.72347152e-01 -1.53708786e-01 4.38839555e-01 -7.36526549e-01 -1.94656312e-01 -7.50797331e-01 -2.72082418e-01 7.68457890e-01 1.01131439e+00 8.32189441e-01 -1.05197072e+00 -3.56438577e-01 5.50456531e-02 -4.09308493e-01 -1.28012705e+00 -6.46204710e-01 -2.06476182e-01 -9.33006823e-01 -9.22795415e-01 -7.76801288e-01 -7.14911222e-01 5.51739633e-01 5.48386931e-01 1.24557364e+00 2.26493314e-01 -1.37956128e-01 8.54996681e-01 -4.29757774e-01 8.38490129e-02 -1.91724464e-01 -6.34837300e-02 5.45442477e-02 -9.32096988e-02 2.43873149e-01 -8.84250462e-01 -1.05450106e+00 4.29616332e-01 -1.05514324e+00 3.89356941e-01 5.65989137e-01 8.10800970e-01 5.68754375e-01 -2.82179177e-01 2.48253196e-01 -7.06133425e-01 -2.51942500e-02 -3.24486732e-01 -3.82507265e-01 4.07833606e-01 -5.13728440e-01 1.33922964e-01 6.46793842e-01 -4.92694259e-01 -1.30319369e+00 1.49250299e-01 -1.30129978e-02 -9.52738941e-01 -2.25297019e-01 1.80807441e-01 2.74411682e-02 -3.14451009e-01 5.93255162e-01 3.92472357e-01 -8.14735666e-02 -3.45036119e-01 6.14273727e-01 7.83600882e-02 7.98263609e-01 -6.76315546e-01 7.75037169e-01 7.84939706e-01 -2.34401599e-01 -3.96470606e-01 -9.86883759e-01 -3.04379255e-01 -5.66532135e-01 -3.30748588e-01 9.46884513e-01 -1.20534313e+00 -7.05757022e-01 3.59405190e-01 -8.61877680e-01 -1.05531585e+00 -2.99363852e-01 4.34708059e-01 -1.00605154e+00 4.84575361e-01 -8.22244406e-01 -7.44125396e-02 -2.32838944e-01 -1.22087622e+00 1.01063955e+00 1.96952209e-01 -1.31396472e-01 -1.03817368e+00 1.14794567e-01 8.62078190e-01 5.38800538e-01 -1.08116470e-01 4.07338470e-01 -5.59637211e-02 -1.02516580e+00 1.69841468e-01 -3.72290820e-01 3.91376704e-01 8.59995559e-03 6.08347878e-02 -9.12118554e-01 -3.37052584e-01 -2.76071191e-01 -5.99727750e-01 1.20101142e+00 3.77608657e-01 1.32488656e+00 -3.03817332e-01 -2.75848508e-01 1.06994045e+00 1.41135144e+00 -1.93847381e-02 9.21280146e-01 5.76627612e-01 9.03926671e-01 5.26325144e-02 4.83091176e-01 3.22693408e-01 6.22665048e-01 8.47481430e-01 5.27392805e-01 1.36435434e-01 -6.16205692e-01 -2.92521745e-01 8.85327816e-01 5.27863741e-01 -7.35869229e-01 -1.12292312e-01 -7.97768235e-01 4.64243859e-01 -2.11730075e+00 -1.46115267e+00 3.66994172e-01 1.93400633e+00 7.25331426e-01 -4.05675098e-02 2.12949902e-01 -3.72898072e-01 3.48384529e-01 2.17560947e-01 -6.86604977e-01 -1.13271236e-01 2.47310121e-02 3.61885041e-01 5.09695590e-01 4.02052104e-01 -1.13136530e+00 1.28704751e+00 6.24875832e+00 6.97138011e-01 -1.09892368e+00 3.18686396e-01 4.69995648e-01 -3.50988805e-01 -3.14280123e-01 2.81376451e-01 -6.70277655e-01 2.88373172e-01 8.84999931e-01 2.65287887e-02 7.73201525e-01 7.52269983e-01 1.56383276e-01 1.35610208e-01 -1.19305146e+00 1.11692619e+00 2.01282576e-01 -1.68651772e+00 3.80140066e-01 -6.28453344e-02 7.48603702e-01 3.05476338e-01 2.60061324e-01 4.62733299e-01 3.41717929e-01 -9.17048633e-01 8.58729661e-01 6.39445364e-01 7.87375629e-01 -2.67055988e-01 2.07737520e-01 -9.15260520e-03 -1.22666740e+00 -2.69476980e-01 -2.99853534e-01 -1.63639888e-01 1.44564956e-01 -5.23190387e-02 -4.49623078e-01 3.03065151e-01 1.12467134e+00 1.15176773e+00 -7.66560316e-01 8.48124921e-01 -1.33818105e-01 7.17594326e-01 -1.96249798e-01 5.21785617e-01 2.92764693e-01 1.16399117e-02 3.04941058e-01 1.03502715e+00 3.27634871e-01 1.90081164e-01 7.79303312e-02 5.26317298e-01 -1.73794240e-01 -1.68766677e-01 -5.88807344e-01 3.26057337e-02 1.07694909e-01 1.05476034e+00 -6.25233471e-01 -5.46913087e-01 -7.76845038e-01 1.45755506e+00 5.81793010e-01 5.72678626e-01 -1.28464508e+00 1.97389737e-01 7.02501416e-01 2.03267172e-01 6.26145363e-01 -1.67369932e-01 1.32039189e-02 -1.56604433e+00 2.04799473e-02 -1.10937929e+00 6.04308128e-01 -1.10310519e+00 -9.18145001e-01 4.90048468e-01 -9.31752250e-02 -1.28152537e+00 -3.16571772e-01 -7.56342471e-01 -4.51684117e-01 2.06192993e-02 -1.36960888e+00 -1.56803989e+00 -5.72473228e-01 1.07692134e+00 8.92906368e-01 -5.83432689e-02 6.69048965e-01 3.46562266e-01 -6.25612974e-01 6.98209822e-01 -1.20021462e-01 1.42712727e-01 8.60250533e-01 -1.08453357e+00 3.21890354e-01 9.47865725e-01 4.14799422e-01 4.86319125e-01 6.38568044e-01 -3.80325526e-01 -1.56325638e+00 -1.14150465e+00 2.59406775e-01 -8.02658439e-01 7.98805475e-01 -1.50778726e-01 -6.71008348e-01 1.32875574e+00 6.83559060e-01 2.39426538e-01 4.66701120e-01 -2.35841591e-02 -6.94637418e-01 -3.02916706e-01 -7.05563128e-01 8.65416348e-01 1.53833091e+00 -5.90916872e-01 -4.02470201e-01 4.11296427e-01 6.92213416e-01 -3.71529520e-01 -9.83976007e-01 4.86301720e-01 7.67144978e-01 -1.17163420e+00 1.27578712e+00 -7.51755834e-01 5.90695262e-01 -3.44777375e-01 -2.64794320e-01 -9.35494304e-01 -5.92685103e-01 -7.50359476e-01 -5.10923088e-01 1.07893002e+00 2.55604744e-01 -3.32983375e-01 9.81018901e-01 6.78072274e-01 -1.90644607e-01 -6.25175595e-01 -6.54300332e-01 -7.86632061e-01 -1.24090360e-02 -5.69940150e-01 4.03906822e-01 6.92677915e-01 7.91156385e-03 1.57951549e-01 -6.22956753e-01 9.38546732e-02 6.37877345e-01 1.29975155e-01 9.82334197e-01 -4.95745957e-01 -7.94369996e-01 -5.73058069e-01 -4.95573372e-01 -1.34957802e+00 1.90924168e-01 -6.96812868e-01 -1.59129292e-01 -1.32068515e+00 5.07645488e-01 -2.12541431e-01 -5.06316304e-01 7.45966852e-01 -9.50125381e-02 7.60072470e-01 4.18095261e-01 3.83551002e-01 -1.10551286e+00 4.48126346e-01 1.33834720e+00 -2.62295872e-01 -3.18701454e-02 -1.49125129e-01 -6.17015541e-01 8.48105729e-01 8.57274950e-01 -9.99228135e-02 -6.29671991e-01 -8.14006805e-01 2.72334237e-02 -1.74807817e-01 7.43697166e-01 -1.11729419e+00 2.54857063e-01 -2.50321269e-01 4.00083899e-01 -3.72477099e-02 5.08605361e-01 -7.12923169e-01 3.66449356e-01 3.72499138e-01 -2.78809994e-01 6.17094077e-02 4.15233135e-01 6.60851836e-01 -1.63288504e-01 1.56140208e-01 6.73543751e-01 -4.39776927e-01 -1.36464274e+00 5.07712364e-01 -1.64874226e-01 -4.24373113e-02 1.16148794e+00 -4.22331750e-01 -4.05133516e-01 -4.77594584e-01 -8.81641209e-01 8.70884303e-03 8.27998400e-01 5.34010172e-01 7.35608339e-01 -1.30523622e+00 -4.91056383e-01 2.09576562e-02 2.23822251e-01 -4.11834866e-01 5.90761423e-01 1.00211442e+00 -6.57530904e-01 3.48353714e-01 -3.42030376e-01 -7.87111878e-01 -1.24839902e+00 8.93430769e-01 4.19578403e-01 -1.74382523e-01 -8.94671321e-01 9.16231692e-01 5.49782991e-01 -2.99763456e-02 2.21633092e-01 -3.74799341e-01 1.88911378e-01 -4.79488283e-01 5.76130748e-01 1.48604929e-01 -2.73475289e-01 -5.79804063e-01 -3.19036543e-01 8.00805628e-01 -2.81325161e-01 -6.46640733e-02 1.41167307e+00 -3.22621614e-01 1.32143006e-01 6.72413036e-02 9.27885115e-01 -3.50083053e-01 -1.86801171e+00 -2.37220511e-01 -3.27883452e-01 -4.97096896e-01 -4.94145080e-02 -6.33355498e-01 -1.27769554e+00 7.21355140e-01 4.51519638e-01 -2.33324111e-01 1.44025552e+00 1.08829908e-01 8.03930163e-01 3.74670267e-01 4.73436356e-01 -1.04296947e+00 5.07379711e-01 3.51404428e-01 8.46842468e-01 -1.23848701e+00 -1.17104702e-01 -2.12675601e-01 -7.33733892e-01 1.06794322e+00 8.55214417e-01 -2.42606997e-01 5.34666657e-01 1.81870207e-01 9.90726277e-02 -7.10101351e-02 -9.91898060e-01 -2.80994684e-01 1.59262538e-01 6.20025516e-01 3.88799399e-01 -2.70078719e-01 2.27186963e-01 2.44819105e-01 1.50335431e-01 6.23053610e-01 2.63284206e-01 6.82305992e-01 -1.52619049e-01 -1.03965032e+00 -7.57349655e-02 1.69515982e-01 -4.64095533e-01 -2.71341443e-01 -6.45446777e-02 7.61439621e-01 1.40458360e-01 6.64891779e-01 4.99093579e-03 -4.37563181e-01 2.03478172e-01 -7.01915398e-02 1.02102971e+00 -3.57992023e-01 -3.22126210e-01 -1.53378487e-01 -2.50026807e-02 -1.23637903e+00 -7.44066477e-01 -5.81434250e-01 -1.08446610e+00 -4.28409308e-01 1.07660353e-01 -2.91113794e-01 1.21759467e-01 9.56773221e-01 6.86331332e-01 4.09457237e-01 1.21894926e-01 -1.18012333e+00 -2.12845176e-01 -6.78507030e-01 -2.45462991e-02 6.12000108e-01 2.17071593e-01 -6.46972418e-01 3.56090628e-02 7.76992440e-01]

[9.11372184753418, 0.7227661609649658]

79cd2143-6069-4100-8a5f-691e2940e99a

a-densely-connected-criss-cross-attention

2203.13953

https://arxiv.org/abs/2203.13953v1

https://arxiv.org/pdf/2203.13953v1.pdf

A Densely Connected Criss-Cross Attention Network for Document-level Relation Extraction

Document-level relation extraction (RE) aims to identify relations between two entities in a given document. Compared with its sentence-level counterpart, document-level RE requires complex reasoning. Previous research normally completed reasoning through information propagation on the mention-level or entity-level document-graph, but rarely considered reasoning at the entity-pair-level.In this paper, we propose a novel model, called Densely Connected Criss-Cross Attention Network (Dense-CCNet), for document-level RE, which can complete logical reasoning at the entity-pair-level. Specifically, the Dense-CCNet performs entity-pair-level logical reasoning through the Criss-Cross Attention (CCA), which can collect contextual information in horizontal and vertical directions on the entity-pair matrix to enhance the corresponding entity-pair representation. In addition, we densely connect multiple layers of the CCA to simultaneously capture the features of single-hop and multi-hop logical reasoning.We evaluate our Dense-CCNet model on three public document-level RE datasets, DocRED, CDR, and GDA. Experimental results demonstrate that our model achieves state-of-the-art performance on these three datasets.

['Yidong Cheng', 'Liang Zhang']

2022-03-26

['document-level-relation-extraction']

['natural-language-processing']

[-3.37757796e-01 4.97505516e-01 -4.33945835e-01 -3.10798645e-01 -6.94100559e-01 -4.30777699e-01 7.25584686e-01 5.16952813e-01 -1.10598959e-01 6.30961239e-01 6.51416838e-01 -6.15469217e-01 -5.15354276e-01 -1.35691857e+00 -9.35627520e-01 1.28784357e-02 -7.16237025e-03 6.65995777e-01 1.55255094e-01 -4.77617592e-01 -6.19128942e-02 3.05124700e-01 -7.87202477e-01 8.26876044e-01 9.54833448e-01 9.63372350e-01 -4.29725975e-01 5.37314475e-01 -5.69444597e-01 1.66151571e+00 -4.37853605e-01 -1.10068011e+00 -1.16879590e-01 -1.21044129e-01 -1.36254728e+00 -2.97659785e-01 3.28934550e-01 -2.69036621e-01 -6.83952212e-01 1.10378015e+00 -1.63729951e-01 -1.47220924e-01 5.98119378e-01 -1.24562204e+00 -1.13044429e+00 1.31707048e+00 -6.89502835e-01 2.51610190e-01 6.48168564e-01 -3.49556327e-01 1.91266000e+00 -1.09407628e+00 1.01472938e+00 1.44642282e+00 5.76184690e-01 1.02695031e-02 -7.71412432e-01 -5.76411664e-01 6.31027341e-01 6.44354761e-01 -1.49247360e+00 1.27758965e-01 7.65312552e-01 -2.52783209e-01 1.59610033e+00 2.55856574e-01 4.31718707e-01 5.96735537e-01 2.64579922e-01 1.07032502e+00 4.31419045e-01 -1.77759394e-01 -2.12439537e-01 -2.86971033e-01 7.14154482e-01 7.55783796e-01 5.66061795e-01 -6.96137607e-01 -3.68195266e-01 5.80921806e-02 3.82228315e-01 -2.17486974e-02 -2.48323992e-01 -1.76130217e-02 -1.24457896e+00 5.88192523e-01 1.18912160e+00 4.05052304e-01 -5.90199292e-01 8.14805329e-02 3.35825026e-01 2.07318127e-01 4.85128880e-01 5.71356118e-01 -7.91603208e-01 2.31026635e-01 -4.99427021e-01 2.14493558e-01 1.09496963e+00 1.48362505e+00 3.51524055e-01 -7.54933238e-01 -6.42418563e-01 4.92844433e-01 5.76683283e-01 2.92769372e-01 -4.88785282e-02 -3.73005301e-01 1.30081058e+00 1.29703462e+00 -3.41237754e-01 -1.47316909e+00 -4.76818413e-01 -7.74773538e-01 -1.18407595e+00 -6.85880899e-01 -1.62256151e-01 -2.46706799e-01 -5.77747226e-01 1.32957721e+00 3.27369958e-01 1.39626026e-01 4.32783097e-01 6.98695600e-01 1.53854048e+00 4.98985201e-01 3.16682085e-02 6.92053959e-02 1.57818580e+00 -1.34491754e+00 -9.47347879e-01 -2.08952427e-01 8.25631201e-01 -3.39767903e-01 5.71323931e-01 -2.09054649e-02 -9.70690906e-01 -3.25513393e-01 -1.11822689e+00 -7.77628005e-01 -7.70065546e-01 2.07471982e-01 8.01333368e-01 -2.63723768e-02 -7.23410070e-01 2.16181010e-01 -4.80263889e-01 -1.15565427e-01 4.86778200e-01 -2.40353346e-02 -4.05890763e-01 -4.18162346e-01 -1.87592185e+00 8.50219965e-01 4.87960368e-01 6.15224898e-01 -2.22047597e-01 -8.12924206e-01 -1.11361837e+00 7.64564812e-01 8.97797227e-01 -9.46657598e-01 9.04652655e-01 4.44558084e-01 -5.53283930e-01 6.17090821e-01 -3.19776356e-01 -4.75846112e-01 3.00593317e-01 -3.58673722e-01 -6.67845666e-01 -1.12685241e-01 2.96121180e-01 3.28345746e-01 -2.84147896e-02 -1.31580508e+00 -5.71752787e-01 -2.34968275e-01 5.42668283e-01 1.94725350e-01 6.46589836e-03 -1.07543416e-01 -1.01319563e+00 -5.06746352e-01 4.08425957e-01 -4.22247618e-01 1.29864827e-01 -5.62834203e-01 -1.25003743e+00 -7.07982004e-01 6.16173387e-01 -8.15658212e-01 1.78692782e+00 -1.84026766e+00 3.60564530e-01 3.12185794e-01 7.77654469e-01 4.95231077e-02 3.12843435e-02 7.52588511e-01 -1.50421590e-01 4.62802500e-01 -5.37306182e-02 -1.54689685e-01 2.33161792e-01 8.70900825e-02 -5.04270196e-01 -1.06261760e-01 6.26151741e-01 1.45830083e+00 -1.06899917e+00 -6.73068285e-01 -3.39916438e-01 2.98958331e-01 -6.26659811e-01 -5.22198454e-02 -4.42672223e-01 -1.85284361e-01 -6.56162679e-01 8.13704550e-01 7.43587792e-01 -7.28697360e-01 6.40764356e-01 -6.59166992e-01 4.35457408e-01 8.50849330e-01 -8.51025403e-01 1.29102457e+00 -5.40593743e-01 6.53306067e-01 -3.77234817e-01 -6.55270696e-01 7.76342213e-01 -4.80193906e-02 2.96579629e-01 -8.25548172e-01 -2.19308306e-02 -5.62133119e-02 1.14113269e-02 -4.52096313e-01 6.92954004e-01 3.42592806e-01 -3.51428360e-01 1.70057938e-01 2.54103467e-02 1.19908281e-01 6.31253183e-01 1.07251430e+00 1.42000651e+00 -1.37664452e-01 3.69031727e-01 1.23184785e-01 9.48505044e-01 -4.46848199e-02 5.55076420e-01 6.53639376e-01 5.25205910e-01 1.37195110e-01 1.17473984e+00 -1.96610063e-01 -4.61852610e-01 -7.85552442e-01 -8.76583830e-02 6.34225309e-01 2.80681610e-01 -1.09520340e+00 -3.95793080e-01 -1.09907854e+00 3.86239588e-01 8.19950223e-01 -6.48257434e-01 -1.60506666e-01 -6.77811563e-01 -6.01661384e-01 6.10308230e-01 8.68434131e-01 1.01302886e+00 -9.62700546e-01 3.82965982e-01 2.45760500e-01 -5.37502229e-01 -1.56724715e+00 -1.05804898e-01 3.41480337e-02 -3.12479198e-01 -1.51169062e+00 -1.02794364e-01 -6.56201065e-01 8.08920085e-01 -1.19254086e-02 1.54650700e+00 4.05299008e-01 1.13021724e-01 7.89372772e-02 -4.94638771e-01 1.05257030e-03 1.80903673e-01 3.91601294e-01 -4.71120149e-01 -1.70504972e-01 7.59348035e-01 -2.65110284e-01 -1.48637041e-01 2.35767872e-03 -6.70740247e-01 1.32456928e-01 8.41641247e-01 9.41014290e-01 4.96160835e-01 6.29093230e-01 2.58363456e-01 -1.37430978e+00 1.00876474e+00 -8.80898774e-01 -1.80327490e-01 8.46370161e-01 -6.31530523e-01 2.17090547e-01 6.17357254e-01 2.05633968e-01 -1.15808654e+00 -6.78391397e-01 -1.40042171e-01 -3.18764639e-03 2.91341752e-01 1.47419977e+00 -4.53232437e-01 5.45208216e-01 1.39156729e-01 -1.40687199e-02 -1.02594006e+00 -3.53402346e-01 6.90200865e-01 5.99953711e-01 8.26089323e-01 -7.34701395e-01 1.01319325e+00 1.34547040e-01 7.65352473e-02 -1.37093022e-01 -1.48067093e+00 -3.65779489e-01 -8.15559924e-01 1.14351578e-01 8.37829530e-01 -1.03160787e+00 -1.14839101e+00 2.23348558e-01 -1.47285640e+00 -8.91631767e-02 -9.13466662e-02 1.54309779e-01 2.66307980e-01 4.12401557e-02 -9.58563089e-01 -4.62725520e-01 -4.47053909e-01 -8.40076864e-01 1.05283308e+00 1.61034152e-01 -9.92694423e-02 -1.14215517e+00 -1.84809238e-01 6.63633108e-01 -1.35369435e-01 4.61540706e-02 1.39338624e+00 -6.88928843e-01 -1.07086098e+00 -4.58616853e-01 -1.01184893e+00 -5.45085594e-02 7.57132843e-02 1.77128636e-03 -4.54726338e-01 3.23899955e-01 -6.47056639e-01 -1.29957750e-01 1.12153506e+00 -1.18029214e-01 1.23953927e+00 -4.45879281e-01 -5.40907919e-01 5.02816677e-01 1.22086906e+00 -9.57941860e-02 6.17268503e-01 3.99332166e-01 1.30493975e+00 3.30750465e-01 6.80723190e-01 8.67602974e-02 1.19047511e+00 4.59868371e-01 2.27620885e-01 -9.14429352e-02 -3.04136097e-01 -6.49326086e-01 -2.35520393e-01 9.46351647e-01 3.40527967e-02 -6.45142615e-01 -1.10844171e+00 4.73549843e-01 -2.14480996e+00 -1.04138625e+00 -5.14993787e-01 1.24089706e+00 9.16634619e-01 5.09081304e-01 -5.27486980e-01 1.70690626e-01 4.55182672e-01 1.80651948e-01 -3.80275667e-01 -1.59205928e-01 -2.64123380e-01 -1.69326022e-01 3.64230663e-01 6.27247036e-01 -1.05532348e+00 1.02891123e+00 4.93469858e+00 6.16129220e-01 -4.44091529e-01 -1.51631013e-01 4.13722605e-01 1.90768957e-01 -7.33380914e-01 4.98102754e-02 -1.18403423e+00 2.74847597e-01 4.47030604e-01 -1.92782432e-01 1.84200540e-01 7.59854138e-01 -5.52125931e-01 -2.19103638e-02 -1.28810501e+00 5.37258327e-01 7.11728707e-02 -1.68647206e+00 1.25072166e-01 8.18090700e-03 6.71684325e-01 -1.99467629e-01 -2.62846380e-01 8.72463226e-01 7.09994614e-01 -9.05517697e-01 4.52342719e-01 6.07665122e-01 7.73065448e-01 -9.00838733e-01 1.31343412e+00 9.43562314e-02 -1.76153517e+00 -4.85726595e-02 -2.52068155e-02 2.29462311e-01 1.46303356e-01 8.47749114e-01 -6.72093093e-01 1.24940217e+00 6.04076147e-01 1.22234702e+00 -7.77694464e-01 4.85408038e-01 -7.28954196e-01 2.90396750e-01 -8.15000534e-02 -2.31758490e-01 5.32110453e-01 -1.70758925e-02 1.65066585e-01 1.51698136e+00 -5.30634597e-02 2.66467422e-01 -1.35117844e-01 9.06232178e-01 -8.35808694e-01 -2.12047994e-01 -8.21294859e-02 -1.67399943e-01 6.53189480e-01 1.23496699e+00 -4.32360858e-01 -5.58062971e-01 -5.82724988e-01 7.25723922e-01 8.66002381e-01 4.39448833e-01 -7.87521899e-01 -6.69023871e-01 5.52471876e-01 -3.87089670e-01 4.86476123e-01 -8.85310173e-02 -4.48603928e-01 -1.41048110e+00 4.63466257e-01 -5.95644236e-01 7.72072673e-01 -1.00330555e+00 -1.56108046e+00 6.66078329e-01 6.96403012e-02 -8.24769735e-01 5.40851839e-02 -5.01263082e-01 -5.17060101e-01 8.69713485e-01 -1.67665350e+00 -1.51684821e+00 -4.11156178e-01 4.67579067e-01 1.99588373e-01 -2.84294859e-02 5.02896547e-01 5.70892870e-01 -8.23821843e-01 7.71659732e-01 -2.95208037e-01 9.76549029e-01 2.23975748e-01 -1.49635684e+00 7.92496026e-01 8.92838597e-01 3.43288422e-01 1.17304337e+00 1.21907644e-01 -1.12904406e+00 -1.46537149e+00 -1.16895318e+00 1.57321286e+00 -4.85055864e-01 1.00406921e+00 -3.41540784e-01 -8.90667498e-01 1.24350786e+00 7.47521371e-02 7.77688473e-02 4.47105527e-01 9.56891358e-01 -8.31403911e-01 -1.47962123e-01 -9.35247183e-01 7.91649103e-01 1.29167354e+00 -9.05707061e-01 -8.53117347e-01 4.74234462e-01 1.21471429e+00 -7.65639007e-01 -1.27744484e+00 5.02667725e-01 3.58210683e-01 -4.33306187e-01 1.18043876e+00 -9.50727582e-01 1.05422330e+00 -3.37740928e-01 -3.79417427e-02 -1.23863709e+00 -5.98120511e-01 -1.22142397e-01 -9.58259940e-01 1.58216166e+00 1.06813216e+00 -4.59061176e-01 4.06043619e-01 6.22346938e-01 -1.68002188e-01 -1.19831073e+00 -4.61140335e-01 -4.99296367e-01 5.03535233e-02 -4.62071806e-01 1.09883010e+00 9.89073694e-01 4.33383465e-01 8.79368722e-01 -5.59301041e-02 6.20267928e-01 3.47135663e-01 3.91725004e-01 4.00351256e-01 -1.18541503e+00 -1.23622179e-01 -4.90796268e-01 -2.38100111e-01 -1.33051622e+00 5.06008446e-01 -1.23363078e+00 -2.39277840e-01 -2.34529138e+00 3.19186002e-01 -3.97894084e-01 -3.54933083e-01 5.09744406e-01 -4.81144577e-01 -3.56550276e-01 1.14233688e-01 4.85036559e-02 -8.57322335e-01 5.01341939e-01 1.43316233e+00 -6.40966237e-01 1.38622195e-01 -4.50180113e-01 -1.09227109e+00 4.08133894e-01 3.00172657e-01 -4.16317493e-01 -4.25166696e-01 -6.47490323e-01 1.07968092e+00 -1.20326273e-01 3.20843786e-01 -6.24548614e-01 8.16526592e-01 -7.37501457e-02 4.01782393e-01 -1.16121984e+00 3.30132470e-02 -7.67342865e-01 -3.09127241e-01 1.43095888e-02 -6.83820069e-01 -1.31317317e-01 -1.28465161e-01 6.26325250e-01 -6.11518621e-01 1.67613193e-01 -9.98511463e-02 6.30650902e-03 -6.66669071e-01 3.88112813e-01 2.54666686e-01 3.62217456e-01 6.23780191e-01 5.62861025e-01 -1.05643582e+00 -1.93568036e-01 -5.03141761e-01 8.57101083e-01 -2.39568070e-01 4.76083636e-01 5.40335715e-01 -1.53171134e+00 -7.51865745e-01 -8.98639951e-03 4.05124605e-01 6.97762907e-01 3.92504692e-01 9.38302100e-01 -4.38422173e-01 9.73959684e-01 3.88210028e-01 -8.43071938e-02 -1.09248459e+00 6.88653409e-01 1.29517049e-01 -1.07968926e+00 -7.58035004e-01 1.18137717e+00 -1.82926744e-01 -6.62392080e-01 7.15532005e-02 -7.75518298e-01 -4.62508023e-01 2.34148763e-02 4.48744416e-01 8.67334232e-02 2.25960925e-01 -4.30966258e-01 -7.12999344e-01 3.64893466e-01 -4.33437079e-01 2.14082509e-01 1.25520599e+00 -5.49236825e-03 -7.02711463e-01 6.82426617e-02 1.09184980e+00 2.02714190e-01 -6.12426758e-01 -5.90515792e-01 2.56753623e-01 -1.39954641e-01 2.56461650e-01 -9.83013749e-01 -1.20266509e+00 6.90742373e-01 -7.75962532e-01 4.06008095e-01 9.75302339e-01 3.53651851e-01 9.38192248e-01 8.41535568e-01 1.26195997e-01 -8.00952196e-01 -2.32966140e-01 9.75491583e-01 1.09855807e+00 -1.09303832e+00 3.85839880e-01 -1.04868340e+00 -7.77564108e-01 9.10799265e-01 8.97506893e-01 2.11009890e-01 7.81747758e-01 4.72342819e-01 -4.22797024e-01 -7.09382653e-01 -1.02136421e+00 -3.62523407e-01 7.80578613e-01 2.18545303e-01 5.87341189e-01 1.70691043e-01 -1.78020433e-01 1.00096190e+00 -2.44723856e-01 -3.29414196e-02 1.97909057e-01 9.03763950e-01 1.17481738e-01 -7.91089356e-01 1.49972975e-01 5.92699945e-01 -7.11098388e-02 -6.34536147e-01 -8.26607764e-01 8.19265783e-01 8.49565938e-02 1.03671420e+00 2.28706703e-01 -7.14932740e-01 3.84653151e-01 -4.49612923e-03 1.78352386e-01 -5.68362415e-01 -6.10339522e-01 -4.63841259e-01 7.42259562e-01 -3.76249820e-01 -1.53909385e-01 -3.58270526e-01 -1.64799547e+00 -5.93539059e-01 -1.98738798e-01 1.03684112e-01 1.99555874e-01 1.26514578e+00 5.70232928e-01 1.35811985e+00 2.77825743e-01 2.10032895e-01 -4.89945188e-02 -1.14214051e+00 -4.47584271e-01 2.43871018e-01 2.51539618e-01 -5.14802516e-01 -5.82501441e-02 -3.26953918e-01]

[9.178617477416992, 8.530251502990723]

3fcd544e-c51a-4d80-97fd-74e64693b2aa

trajectory-guided-control-prediction-for-end

2206.08129

https://arxiv.org/abs/2206.08129v2

https://arxiv.org/pdf/2206.08129v2.pdf

Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline

Current end-to-end autonomous driving methods either run a controller based on a planned trajectory or perform control prediction directly, which have spanned two separately studied lines of research. Seeing their potential mutual benefits to each other, this paper takes the initiative to explore the combination of these two well-developed worlds. Specifically, our integrated approach has two branches for trajectory planning and direct control, respectively. The trajectory branch predicts the future trajectory, while the control branch involves a novel multi-step prediction scheme such that the relationship between current actions and future states can be reasoned. The two branches are connected so that the control branch receives corresponding guidance from the trajectory branch at each time step. The outputs from two branches are then fused to achieve complementary advantages. Our results are evaluated in the closed-loop urban driving setting with challenging scenarios using the CARLA simulator. Even with a monocular camera input, the proposed approach ranks first on the official CARLA Leaderboard, outperforming other complex candidates with multiple sensors or fusion mechanisms by a large margin. The source code is publicly available at https://github.com/OpenPerceptionX/TCP

['Yu Qiao', 'Hongyang Li', 'Junchi Yan', 'Li Chen', 'Xiaosong Jia', 'Penghao Wu']

2022-06-16

['trajectory-planning']

['robots']

[-4.30968404e-02 4.64368105e-01 -4.60195661e-01 -4.41745281e-01 -6.61520243e-01 -6.84361398e-01 1.05904448e+00 1.92406196e-02 -2.91193843e-01 6.61194503e-01 6.04305193e-02 -5.88397563e-01 -1.86854154e-01 -8.96216929e-01 -7.28225589e-01 -6.85272753e-01 -4.01800647e-02 4.58416015e-01 5.16737044e-01 -5.32090366e-01 1.99957609e-01 4.11868423e-01 -1.91201806e+00 -2.25540191e-01 8.17774594e-01 1.02086043e+00 2.80218810e-01 7.02685535e-01 3.22212249e-01 7.14637339e-01 3.47013503e-01 -3.83947380e-02 5.97994626e-01 -8.61631334e-02 -3.44850719e-01 3.28969955e-02 -2.63993144e-02 -2.52578259e-01 -4.72569674e-01 8.38051021e-01 2.40282044e-01 5.98705336e-02 4.68335859e-02 -1.70880663e+00 3.12230378e-01 3.41296971e-01 2.13118140e-02 -3.06543559e-01 3.18616301e-01 7.62602746e-01 7.40427852e-01 -5.88550746e-01 6.66027486e-01 1.10595870e+00 3.41851264e-01 3.58455479e-01 -1.01432657e+00 -6.20724201e-01 4.08608675e-01 3.82082492e-01 -1.26158822e+00 -7.85528719e-01 4.93556499e-01 -6.59984231e-01 7.87960112e-01 2.93114968e-02 6.77830637e-01 9.60275114e-01 4.91812944e-01 6.66025698e-01 1.05345440e+00 1.93850607e-01 3.65697771e-01 2.16528118e-01 6.61974624e-02 5.72759449e-01 7.33877495e-02 1.03091276e+00 -3.10872942e-01 7.07684234e-02 -1.12753306e-02 -2.29208395e-02 -4.03767198e-01 -5.97922564e-01 -1.60781574e+00 6.16006672e-01 3.55912387e-01 -2.31641799e-01 -6.15492821e-01 1.56407863e-01 1.57305285e-01 3.47678214e-01 -4.77072597e-02 1.89169869e-01 -2.66941160e-01 -3.43053550e-01 -6.83316171e-01 7.85520852e-01 7.41897464e-01 1.20321310e+00 9.04488027e-01 -5.79912849e-02 -2.45857105e-01 7.93565810e-02 6.15062058e-01 7.23914981e-01 8.84644538e-02 -1.30204022e+00 6.86055481e-01 5.49209416e-01 5.90196908e-01 -8.84317696e-01 -6.38213336e-01 -3.42931479e-01 -4.31594223e-01 8.13377559e-01 5.83200336e-01 -4.60537255e-01 -6.33443356e-01 1.74968338e+00 5.60407817e-01 3.49693358e-01 4.66022879e-01 1.03017569e+00 1.43199176e-01 7.60644376e-01 -2.07423285e-01 -1.91416945e-02 1.09605265e+00 -1.09548283e+00 -7.83940911e-01 -4.66601700e-01 6.42622113e-01 -6.83399141e-01 5.16044557e-01 4.85993356e-01 -8.43885720e-01 -7.15422511e-01 -1.41150224e+00 1.18944153e-01 -4.66231167e-01 3.04801702e-01 2.11500764e-01 1.89838812e-01 -1.14957941e+00 6.84058547e-01 -1.12398028e+00 -5.47660232e-01 -1.27605973e-02 2.44215295e-01 -1.57277793e-01 1.81073695e-02 -1.21576071e+00 1.18944478e+00 4.02943760e-01 9.90996435e-02 -1.00937080e+00 -5.00290573e-01 -7.84963727e-01 -2.67387658e-01 6.86982393e-01 -5.02820015e-01 1.40403008e+00 -3.02166373e-01 -1.94851732e+00 3.64152342e-01 -2.10688457e-01 -9.39472914e-01 9.51023638e-01 -3.70177716e-01 -2.62714714e-01 -2.54609883e-01 7.80018121e-02 9.01520312e-01 5.25116444e-01 -1.15098894e+00 -1.29137623e+00 -3.69403332e-01 2.12652162e-01 3.04753393e-01 2.37282887e-01 -7.09544122e-01 -4.16225851e-01 1.98379233e-02 -7.68293068e-02 -1.51291728e+00 -5.74069023e-01 7.87295625e-02 -2.45013416e-01 -1.65415436e-01 9.59952176e-01 -1.59288555e-01 1.22051466e+00 -1.89985645e+00 2.49570847e-01 2.77354866e-02 1.01098500e-01 2.42331013e-01 -1.75210200e-02 9.14733469e-01 2.21847638e-01 -4.56191808e-01 -3.83749679e-02 -3.08157146e-01 1.34626254e-01 1.79012880e-01 -6.48750305e-01 5.90579093e-01 2.66354810e-03 7.36785352e-01 -1.04252303e+00 -7.46536553e-02 6.75777972e-01 1.94089279e-01 -2.98060715e-01 2.58799642e-01 -3.63202780e-01 7.77006388e-01 -6.99299634e-01 2.63899475e-01 5.68140447e-01 2.33109191e-01 9.10967961e-02 9.63582024e-02 -9.09909785e-01 3.14447433e-01 -1.32542801e+00 1.61958408e+00 -3.94098788e-01 6.53826594e-01 3.60554934e-01 -7.00919390e-01 1.08143353e+00 3.21690828e-01 4.95495200e-01 -5.64914644e-01 2.01752335e-01 4.15490955e-01 -6.77605905e-03 -4.35514510e-01 6.86885595e-01 9.39791501e-02 -2.14775324e-01 -4.39520143e-02 -3.35604697e-01 -3.54561567e-01 2.21940354e-01 3.99391241e-02 1.11368108e+00 7.13622272e-01 4.36174750e-01 -1.55851543e-01 7.85281360e-01 6.18701696e-01 7.86724031e-01 5.99320114e-01 -7.30625093e-01 3.05749357e-01 3.81118268e-01 -3.52352440e-01 -7.92764544e-01 -9.42541361e-01 2.75579765e-02 5.05136192e-01 7.75038362e-01 -4.34132606e-01 -4.88782346e-01 -4.12318021e-01 1.02629766e-01 1.09935737e+00 -3.72584611e-01 -2.06427693e-01 -4.89171654e-01 -7.44927973e-02 3.15161467e-01 2.68657237e-01 4.71479326e-01 -6.63012683e-01 -1.10926580e+00 3.39885533e-01 -1.64292976e-01 -1.25963438e+00 -3.18947695e-02 1.82806030e-01 -5.58063269e-01 -1.17030466e+00 -1.94265798e-01 -2.19726220e-01 6.64342791e-02 5.43080091e-01 5.93582094e-01 -2.16950342e-01 4.13103551e-01 4.08934891e-01 -2.80391574e-01 -5.89855015e-01 -5.50871432e-01 1.43653313e-02 9.67417359e-02 2.10591152e-01 1.49408251e-01 -4.80052054e-01 -6.60603642e-01 5.65662801e-01 -2.72780567e-01 5.22187233e-01 4.89028037e-01 3.62856865e-01 5.71656466e-01 -2.27655202e-01 5.29309630e-01 -3.22037071e-01 2.13169992e-01 -7.96006322e-01 -1.18463278e+00 -1.94025680e-01 -8.63213181e-01 8.08420703e-02 7.72413611e-01 5.91338240e-02 -1.19515944e+00 5.95426738e-01 -3.98547314e-02 -4.39010739e-01 -5.35404682e-01 4.36326772e-01 -3.35403904e-02 3.30465585e-01 4.10477489e-01 2.15830103e-01 3.39296192e-01 -4.63993214e-02 5.07632077e-01 5.88673174e-01 6.84041917e-01 -2.99692631e-01 9.53409553e-01 7.27729857e-01 1.49540350e-01 -4.47024763e-01 -3.92971605e-01 -5.41963339e-01 -6.73753381e-01 -6.93280697e-01 9.35351312e-01 -1.18222976e+00 -8.31360698e-01 3.70355159e-01 -1.17329764e+00 -5.29678822e-01 -2.13243335e-01 7.68790483e-01 -8.88422430e-01 -6.50264025e-02 -1.23567380e-01 -8.38389874e-01 9.86162275e-02 -1.55578291e+00 1.15256691e+00 2.48749867e-01 -9.62195098e-02 -7.10780621e-01 3.60835731e-01 1.74315616e-01 3.77677530e-01 4.06477481e-01 2.70170629e-01 -4.38011289e-01 -9.66978550e-01 -4.39772487e-01 1.33875251e-01 -9.52664986e-02 -1.81290984e-01 2.53067184e-02 -8.47149670e-01 -3.19094330e-01 -2.42547005e-01 -2.41194498e-02 7.11410701e-01 1.69447303e-01 3.48830193e-01 -2.18626242e-02 -7.02946007e-01 4.03466612e-01 1.43843150e+00 5.02238452e-01 5.54461539e-01 4.68440324e-01 3.53864580e-01 9.78611708e-01 1.25017917e+00 4.12555277e-01 9.25427794e-01 9.38417494e-01 9.96743798e-01 3.40427428e-01 2.05520447e-02 -3.84359092e-01 7.52316535e-01 5.40456712e-01 4.22347784e-02 -3.65835965e-01 -1.11734986e+00 6.24068081e-01 -2.35710907e+00 -1.01717412e+00 -4.55545843e-01 2.39838529e+00 1.10124141e-01 3.40214103e-01 8.23163912e-02 -4.05837968e-02 4.87309396e-01 2.34128267e-01 -7.37712562e-01 -2.34780684e-01 3.36539000e-01 -4.06243235e-01 8.77653837e-01 7.13053823e-01 -1.12408590e+00 8.12291563e-01 5.18859863e+00 5.52946329e-01 -1.23131239e+00 -8.86910334e-02 1.31276265e-01 -3.34264003e-02 2.41096914e-02 4.59630698e-01 -1.02915931e+00 4.12226260e-01 1.31067801e+00 -3.76548946e-01 3.00251782e-01 8.14657688e-01 9.05761957e-01 -3.72804970e-01 -1.13080335e+00 4.40158725e-01 -6.35319710e-01 -1.37798202e+00 -4.48266804e-01 3.16530854e-01 5.52265227e-01 5.34986079e-01 -5.04015461e-02 4.73728597e-01 4.87150371e-01 -5.86467206e-01 1.17967117e+00 8.06758821e-01 2.04317540e-01 -6.75228477e-01 5.49254835e-01 8.89583230e-01 -1.44721675e+00 -3.71680558e-01 1.33643031e-01 -4.25132453e-01 5.90376973e-01 2.63088584e-01 -7.62901545e-01 9.66769934e-01 2.95758128e-01 1.05898392e+00 -3.53256345e-01 1.03760111e+00 -1.91088364e-01 5.77437699e-01 -2.15115905e-01 -1.32518486e-04 6.26197755e-01 -5.08294106e-01 1.12520456e+00 8.12776029e-01 4.71970379e-01 3.15582156e-02 5.63714981e-01 8.01545918e-01 6.34000182e-01 -3.97656858e-01 -9.75457609e-01 3.03452939e-01 3.78967047e-01 1.29719305e+00 -3.11368465e-01 -4.15456086e-01 -4.99908358e-01 3.66313696e-01 1.09235816e-01 2.38159716e-01 -1.13818121e+00 -2.96985984e-01 1.20200872e+00 1.52828231e-01 3.10120493e-01 -4.55969572e-01 -2.45594323e-01 -8.63652289e-01 8.47020652e-03 -4.75715637e-01 -1.66097954e-01 -7.26272106e-01 -4.75494325e-01 6.32214069e-01 1.94888160e-01 -1.91213882e+00 -4.25320715e-01 -4.27782148e-01 -6.34694815e-01 6.15305603e-01 -1.91549790e+00 -8.65032971e-01 -4.22017038e-01 3.43788058e-01 6.34609163e-01 -2.33991798e-02 5.34855843e-01 1.85221359e-01 -5.87522447e-01 -8.10772777e-02 9.61849019e-02 -4.74454075e-01 5.01875639e-01 -1.05914462e+00 3.84385496e-01 9.78338182e-01 -4.58145440e-01 1.18333161e-01 9.42883551e-01 -6.22760952e-01 -1.68008316e+00 -1.40911698e+00 8.97538424e-01 -4.30621147e-01 6.43586934e-01 -1.59587979e-01 -5.33688724e-01 7.24269092e-01 3.65962058e-01 -7.63604566e-02 -1.54658437e-01 -3.80324453e-01 1.19413085e-01 -2.57744074e-01 -7.64955163e-01 8.31236601e-01 8.96580219e-01 9.31100175e-02 -3.32331985e-01 1.06601641e-01 6.87349796e-01 -5.66939890e-01 -6.43272340e-01 5.65024614e-01 5.01961708e-01 -1.27540708e+00 5.18717110e-01 -1.21034138e-01 2.91540593e-01 -8.64117026e-01 -1.67090729e-01 -1.36550081e+00 -2.84509491e-02 -9.01518762e-01 -2.22642608e-02 8.99057627e-01 4.97657835e-01 -9.51706886e-01 6.14414513e-01 3.56727302e-01 -5.92322707e-01 -9.40381765e-01 -1.01035523e+00 -8.03206921e-01 -8.17984641e-02 -8.95091832e-01 5.59217513e-01 2.61096597e-01 1.19994268e-01 2.78745323e-01 -4.12876695e-01 5.21368742e-01 5.20204008e-01 2.99208492e-01 1.32189572e+00 -9.35868680e-01 4.54768725e-02 -4.48371053e-01 -4.45545763e-01 -1.42775130e+00 1.44628957e-01 -8.31919372e-01 5.10073006e-01 -1.36512363e+00 -4.77574944e-01 -3.82908374e-01 1.39423117e-01 3.65062088e-01 1.54847518e-01 -4.39905375e-01 4.38432276e-01 1.68036446e-01 -5.63464642e-01 8.14638913e-01 1.08747852e+00 -5.52584827e-02 -3.99680436e-01 4.23105389e-01 -2.80880958e-01 8.15225542e-01 9.76571143e-01 -1.43497929e-01 -7.40786493e-01 -2.64367521e-01 -1.26883283e-01 7.03880489e-01 5.65922081e-01 -1.64310563e+00 6.15819037e-01 -3.29286426e-01 -5.08249104e-01 -8.98893297e-01 5.27879119e-01 -1.05598915e+00 2.81211644e-01 6.66772187e-01 -3.82300109e-01 1.41645670e-01 1.57464698e-01 9.38179374e-01 -3.45037431e-01 3.07486892e-01 7.41481602e-01 2.01423645e-01 -9.76889431e-01 3.74871701e-01 -8.01665604e-01 -3.49574536e-01 1.63485050e+00 -3.09391916e-01 -2.07071617e-01 -5.40331662e-01 -5.57755053e-01 9.62405145e-01 2.99664438e-01 8.73576701e-01 4.53520358e-01 -1.32848871e+00 -6.48803473e-01 1.76672146e-01 2.83453822e-01 -1.13594905e-01 8.57052505e-02 1.30132234e+00 -9.89949331e-02 8.56808782e-01 -1.57339722e-01 -8.56380224e-01 -7.83566833e-01 5.41610897e-01 3.95162225e-01 -3.27572107e-01 -6.88241065e-01 6.22113831e-02 1.31035537e-01 -8.07224691e-01 1.44857585e-01 -3.58706325e-01 -2.76088953e-01 -7.88520128e-02 4.34683502e-01 6.60753489e-01 2.13930290e-02 -1.03934312e+00 -2.11052373e-01 4.97045964e-01 3.81176263e-01 -3.35363895e-01 1.00417054e+00 -5.76573670e-01 4.13385212e-01 5.77520967e-01 1.01994419e+00 -1.82644814e-01 -1.78819394e+00 -3.63406241e-02 9.31551903e-02 -2.27600813e-01 8.76775756e-02 -7.24786639e-01 -9.12144244e-01 6.37865543e-01 5.97468793e-01 1.08966559e-01 8.71019959e-01 -3.28567743e-01 8.85079443e-01 3.18427265e-01 8.49519670e-01 -9.26014721e-01 -5.63894808e-01 7.15170562e-01 9.65787411e-01 -1.28843522e+00 -1.88427165e-01 -3.96862656e-01 -7.65312016e-01 1.09099007e+00 7.39999235e-01 -2.32207075e-01 6.97805643e-01 2.46427745e-01 1.92509398e-01 -6.11955635e-02 -1.34574318e+00 -5.23801863e-01 2.06622705e-02 4.44676816e-01 -1.34313703e-01 1.93953335e-01 -4.44958597e-01 2.99354643e-01 -1.83592051e-01 1.07538030e-01 6.15683377e-01 8.38802040e-01 -5.91664433e-01 -9.32441533e-01 -4.37049747e-01 7.43253008e-02 2.75335938e-01 5.21177769e-01 1.42409921e-01 1.00176430e+00 2.18839437e-01 1.34526420e+00 -2.79435460e-02 -7.17155635e-01 6.89088166e-01 -5.48707508e-02 -2.26609141e-01 -2.83465385e-01 -4.23066586e-01 -1.60960123e-01 3.18916053e-01 -1.33081162e+00 -3.11009169e-01 -1.06627595e+00 -1.43514168e+00 -2.43332446e-01 -5.84165119e-02 -5.87809421e-02 6.32551968e-01 9.16243434e-01 8.11201334e-01 3.94986600e-01 7.36732960e-01 -1.30582619e+00 -6.12131059e-01 -5.73430479e-01 6.00265851e-03 -1.34738490e-01 5.92108369e-01 -8.03436816e-01 -1.12421751e-01 -9.35078263e-02]

[5.692533016204834, 0.9589014053344727]

abf420c1-9b3e-4468-aaf1-67ac4062063e

detecting-label-errors-in-token

2210.0392

https://arxiv.org/abs/2210.03920v1

https://arxiv.org/pdf/2210.03920v1.pdf

Detecting Label Errors in Token Classification Data

Mislabeled examples are a common issue in real-world data, particularly for tasks like token classification where many labels must be chosen on a fine-grained basis. Here we consider the task of finding sentences that contain label errors in token classification datasets. We study 11 different straightforward methods that score tokens/sentences based on the predicted class probabilities output by a (any) token classification model (trained via any procedure). In precision-recall evaluations based on real-world label errors in entity recognition data from CoNLL-2003, we identify a simple and effective method that consistently detects those sentences containing label errors when applied with different token classification models.

['Jonas Mueller', 'Wei-Chen Wang']

2022-10-08

['classification']

['methodology']

[ 1.49526700e-01 -1.49788022e-01 -4.06869739e-01 -7.70281732e-01 -1.03254378e+00 -6.55834973e-01 5.75826228e-01 8.45733583e-01 -8.96654606e-01 1.08927071e+00 9.62937996e-03 -1.20657295e-01 1.90682545e-01 -7.13290751e-01 -4.85484660e-01 -3.48819077e-01 1.76117733e-01 6.72627509e-01 7.47904703e-02 2.27862328e-01 7.73592234e-01 2.50094891e-01 -1.47648954e+00 6.44516885e-01 8.53523493e-01 8.46492648e-01 -1.33126915e-01 5.81098676e-01 -6.94194019e-01 9.70298350e-01 -1.19480896e+00 -5.73852658e-01 -9.68632624e-02 1.81950051e-02 -1.37308562e+00 -7.01571861e-03 7.27609038e-01 4.04649645e-01 4.21295404e-01 1.31002855e+00 4.00465846e-01 -8.59483704e-02 1.10151601e+00 -1.08597314e+00 -3.14393103e-01 9.79183257e-01 -7.79625634e-03 2.78372228e-01 4.98685032e-01 -1.30269185e-01 1.13024902e+00 -1.12427223e+00 7.41693914e-01 1.20165455e+00 9.75905597e-01 4.38900650e-01 -1.36959779e+00 -6.03215277e-01 -6.68954477e-03 2.78072864e-01 -1.63073802e+00 -3.43525350e-01 1.13973521e-01 -5.08797407e-01 1.45910060e+00 3.42412621e-01 -1.07093096e-01 7.77287424e-01 2.52240449e-01 5.55848956e-01 1.04636753e+00 -1.01306880e+00 3.26355428e-01 3.62285852e-01 5.57677209e-01 5.13191760e-01 5.06634951e-01 -1.15746818e-01 -5.98747253e-01 -4.44751799e-01 -1.69686601e-01 -1.75016463e-01 2.91394383e-01 3.43876809e-01 -1.27134049e+00 6.50209427e-01 7.22112954e-02 8.39167178e-01 -2.59089112e-01 4.00471717e-01 1.14637470e+00 5.80124855e-01 8.28333497e-01 9.19611156e-01 -8.88529122e-01 -2.94990361e-01 -8.78594279e-01 3.21527541e-01 7.16477334e-01 1.09776509e+00 9.84483361e-01 -9.49851945e-02 -6.36455476e-01 1.04716802e+00 1.66318446e-01 3.19864899e-01 7.96163023e-01 -4.21335250e-01 4.92456228e-01 5.32429039e-01 3.65675300e-01 -5.16296387e-01 -5.07004857e-01 -1.01560377e-01 -2.71772414e-01 -1.57024801e-01 5.03680825e-01 -4.64438871e-02 -1.00648618e+00 1.39960337e+00 2.79176921e-01 1.06494583e-01 1.12375878e-01 2.38299832e-01 8.62354994e-01 2.44728848e-01 9.79699671e-01 3.28434296e-02 1.52593064e+00 -4.57512319e-01 -6.87858224e-01 -3.94977778e-02 1.59976768e+00 -9.67512488e-01 8.28968883e-01 2.03176141e-01 -4.50552344e-01 -4.58254665e-01 -5.98765016e-01 7.97529742e-02 -9.10734177e-01 3.59945029e-01 3.62738103e-01 9.13728237e-01 -6.90756083e-01 6.78336620e-01 -2.47533828e-01 -5.22947431e-01 -2.65019163e-02 2.66434729e-01 -4.12709802e-01 -4.57765199e-02 -1.43490720e+00 1.29584706e+00 7.95572281e-01 2.41271164e-02 -5.44757724e-01 -4.82896686e-01 -9.11556184e-01 2.11909831e-01 1.13540113e-01 -1.86184168e-01 1.67864680e+00 -6.80467606e-01 -6.50773108e-01 1.47312856e+00 -3.75006080e-01 -4.11857247e-01 5.05733788e-01 7.92271718e-02 -5.18934965e-01 -2.28248596e-01 6.78558767e-01 3.22696298e-01 4.59565490e-01 -1.13646877e+00 -1.00376451e+00 -5.54774217e-02 -2.71366656e-01 -4.44957726e-02 -1.27097845e-01 4.12289709e-01 6.15029573e-01 -6.98765099e-01 9.27140489e-02 -6.46062434e-01 -1.59906656e-01 -7.06795633e-01 -8.32552314e-01 -1.14491272e+00 4.02213871e-01 -2.07602888e-01 1.07578862e+00 -2.03592110e+00 -9.11802053e-01 3.02186221e-01 1.38787746e-01 3.15674782e-01 -2.32848134e-02 5.60272515e-01 -2.70805717e-01 7.50911236e-01 8.37227330e-02 -5.29154420e-01 4.16128546e-01 1.18281081e-01 -4.03594017e-01 2.67280310e-01 1.56165317e-01 4.49337184e-01 -1.22971916e+00 -9.59535658e-01 1.48550585e-01 -2.84516841e-01 9.18193012e-02 1.35648638e-01 -1.82667628e-01 -2.49027312e-01 -5.01468897e-01 8.39883566e-01 5.10465145e-01 5.63088767e-02 1.71604276e-01 4.75524180e-02 -1.53017670e-01 6.31489277e-01 -1.25507104e+00 9.80546951e-01 -5.77177703e-01 5.75565815e-01 -7.67717898e-01 -1.03860629e+00 9.79712605e-01 7.51508296e-01 9.50800553e-02 -3.17110866e-01 2.96293367e-02 7.20194519e-01 -4.31380451e-01 -8.04015279e-01 1.09243906e+00 -6.23552740e-01 -6.27518058e-01 3.84122282e-01 2.71746457e-01 3.16748098e-02 5.67401767e-01 9.31163803e-02 1.27871168e+00 -5.53373694e-01 4.60208267e-01 -2.51105309e-01 3.68689030e-01 3.50684226e-01 7.81690598e-01 1.36086094e+00 -4.33889180e-01 4.89115626e-01 4.62476879e-01 -4.87891316e-01 -1.06189430e+00 -7.91426301e-01 -6.78722203e-01 1.25232208e+00 -1.94153547e-01 -4.79356289e-01 -4.26595598e-01 -1.06857908e+00 2.11645603e-01 7.79830694e-01 -5.09015024e-01 3.11309639e-02 -2.56712347e-01 -6.57208025e-01 1.19575346e+00 4.44561541e-01 2.84239471e-01 -1.65424752e+00 -1.02885395e-01 4.22260463e-01 -2.23432735e-01 -1.13892293e+00 -2.06930071e-01 8.79420936e-01 -4.24733073e-01 -1.08943212e+00 -2.70715028e-01 -1.10534310e+00 9.42568362e-01 -5.94521202e-02 1.36264157e+00 4.87079978e-01 -3.04773003e-01 6.45020008e-02 -7.37427652e-01 -5.10259032e-01 -7.00650752e-01 2.13709608e-01 3.11018735e-01 -3.09198290e-01 9.81277227e-01 3.18750560e-01 1.91893891e-01 1.60785168e-01 -8.11882138e-01 -6.69056356e-01 3.01866084e-01 1.11206472e+00 4.09682363e-01 3.79721969e-01 9.02687311e-01 -1.70451212e+00 7.25043476e-01 -5.40981114e-01 -2.14840412e-01 7.13834286e-01 -6.77051723e-01 1.23806968e-01 6.68947697e-01 -1.50902271e-01 -9.82555389e-01 5.79680651e-02 -2.38240451e-01 1.91825569e-01 -6.36854827e-01 5.86620629e-01 1.75738081e-01 3.21117118e-02 9.10055161e-01 7.13724867e-02 -9.24782634e-01 -4.39920753e-01 -1.20905619e-02 1.01947403e+00 2.79450864e-01 -9.59381044e-01 4.00651366e-01 -2.05016464e-01 -2.90525198e-01 -5.68061829e-01 -1.08345270e+00 -8.62397194e-01 -6.99415684e-01 -1.61480889e-01 5.66326499e-01 -7.45272815e-01 -5.72492659e-01 6.14789188e-01 -1.39400506e+00 8.21321923e-03 -4.90437150e-01 6.48748636e-01 -1.73077390e-01 2.49280140e-01 -5.74304938e-01 -8.43382835e-01 -2.28040546e-01 -6.89970851e-01 1.06365526e+00 2.58506507e-01 -5.92930496e-01 -1.09829855e+00 1.89404562e-01 -2.06079800e-02 1.64889470e-01 -7.50184283e-02 1.00655663e+00 -1.34887135e+00 -4.45757732e-02 -6.10905886e-01 -1.81162998e-01 3.64311695e-01 8.69260505e-02 2.14342952e-01 -1.21381497e+00 2.38917936e-02 -5.38150728e-01 -6.81300461e-01 9.41956878e-01 1.71494618e-01 1.13550258e+00 -2.57057518e-01 -5.03845692e-01 -2.20391005e-01 1.55290484e+00 4.19023000e-02 3.10414135e-01 4.32338148e-01 6.11307919e-01 4.63354051e-01 8.66736233e-01 4.29686576e-01 1.60955667e-01 3.29180092e-01 -2.10443169e-01 1.69264212e-01 2.42268398e-01 -2.79457897e-01 1.04948588e-01 4.68054771e-01 3.19518775e-01 -3.93762499e-01 -1.11591017e+00 8.71111333e-01 -1.70947874e+00 -1.21649706e+00 -2.54008263e-01 1.99870467e+00 1.26777434e+00 5.00854969e-01 -2.34723553e-01 4.33307707e-01 1.25705278e+00 -7.88804442e-02 -1.48916870e-01 -9.31519032e-01 4.14635688e-02 3.22033256e-01 6.67657495e-01 2.97865927e-01 -1.32816398e+00 1.17455304e+00 7.41166782e+00 1.16464996e+00 -9.46005404e-01 -2.20550094e-02 7.01213658e-01 4.97480482e-01 -9.61330533e-02 -8.92781466e-02 -1.28041923e+00 5.96796334e-01 9.50145304e-01 -2.68156469e-01 -4.00736064e-01 1.04902864e+00 1.07546814e-01 -5.00805080e-01 -1.26492095e+00 6.65444493e-01 -1.84659988e-01 -1.21480584e+00 -2.31122836e-01 -4.33512300e-01 8.03164959e-01 -6.97730184e-02 -2.61588663e-01 7.79720485e-01 6.03781939e-01 -9.58910286e-01 9.54151988e-01 2.19511092e-01 9.13016438e-01 -6.10519469e-01 1.18521404e+00 3.67214531e-01 -1.06344938e+00 2.15634644e-01 -7.27603018e-01 -1.12246573e-01 -2.58172035e-01 1.18252039e+00 -1.68559849e+00 2.42750257e-01 5.15867472e-01 5.98391473e-01 -6.69890106e-01 1.40163195e+00 -4.79154259e-01 8.32808375e-01 -1.98450610e-01 -4.57141161e-01 1.16201311e-01 6.61681533e-01 5.21634109e-02 1.97122049e+00 7.74048418e-02 -2.17140749e-01 3.20225716e-01 5.95117509e-01 -5.02863646e-01 2.46248245e-01 -7.69763827e-01 2.67587095e-01 1.11735821e+00 1.20732641e+00 -1.02852762e+00 -8.00661743e-01 -1.71433881e-01 4.84580159e-01 4.64525670e-01 8.94941613e-02 -3.54779065e-01 -7.97570467e-01 4.24138814e-01 -4.02003646e-01 3.04786731e-02 2.21487343e-01 -6.05040669e-01 -1.05419159e+00 1.37303829e-01 -3.59138370e-01 4.31956828e-01 -4.15564090e-01 -1.87544274e+00 5.43717384e-01 -1.45775571e-01 -1.35619462e+00 -3.44844460e-01 -6.84913397e-01 -7.25423455e-01 8.06969762e-01 -1.47763407e+00 -6.52803302e-01 6.49118721e-02 1.50049135e-01 3.63711059e-01 -9.49662179e-03 1.07736933e+00 4.29295927e-01 -4.49749798e-01 8.25600386e-01 2.16744468e-01 7.15392411e-01 1.11578929e+00 -1.66026843e+00 4.66550052e-01 6.35031939e-01 3.07116389e-01 4.59685504e-01 8.15072417e-01 -6.76781178e-01 -2.12196827e-01 -1.27323043e+00 2.26242042e+00 -5.34243464e-01 4.45439130e-01 -1.95334211e-01 -7.05214262e-01 7.82595456e-01 -1.44483358e-01 2.72594333e-01 8.43664885e-01 5.91389358e-01 -5.40114284e-01 1.57487690e-01 -1.54238796e+00 3.45436670e-02 5.78506351e-01 -8.50150824e-01 -9.09264326e-01 7.24897683e-01 2.40227222e-01 -4.04598117e-01 -8.20160389e-01 1.32920846e-01 2.50335336e-01 -4.35337424e-01 5.78773022e-01 -9.59448338e-01 1.92447275e-01 -3.08856487e-01 -4.20017317e-02 -1.57451439e+00 -7.59228468e-02 1.45275638e-01 8.28723013e-01 1.69731629e+00 7.31615603e-01 -6.28604114e-01 6.29804075e-01 9.45669770e-01 -7.41875619e-02 -2.36259028e-01 -1.26338243e+00 -1.12003529e+00 1.13379329e-01 -3.92808050e-01 5.09475708e-01 1.38728869e+00 5.87232888e-01 3.08215432e-02 -3.03195920e-02 -5.30450977e-02 4.47647810e-01 -2.24681035e-01 3.41159701e-01 -1.20376027e+00 3.66016507e-01 -1.55375376e-01 -7.61586964e-01 -4.68407482e-01 7.02935815e-01 -1.15961790e+00 7.06668615e-01 -1.17217267e+00 2.57728040e-01 -1.10353661e+00 -4.22249615e-01 9.97967541e-01 -4.33232307e-01 2.11108401e-01 2.24537417e-01 1.84323967e-01 -9.32593584e-01 1.27276897e-01 4.37000334e-01 -4.19308752e-01 4.10089314e-01 4.59608957e-02 -3.89292508e-01 7.71285355e-01 6.96159005e-01 -1.48947978e+00 3.52166891e-01 -3.11161757e-01 2.97608972e-01 -1.18073404e-01 1.43794358e-01 -9.74796772e-01 1.65967360e-01 -3.08011204e-01 4.00748312e-01 -4.06453878e-01 -3.70157421e-01 -5.05603731e-01 -3.01257074e-01 5.28569400e-01 -1.05812454e+00 2.19997764e-01 -1.56024873e-01 2.71848530e-01 -5.81296146e-01 -1.17564774e+00 6.57591164e-01 -4.00949568e-01 -1.04507911e+00 -1.08863182e-01 -8.03722858e-01 2.89610028e-01 9.66540396e-01 1.74897797e-02 -6.31492376e-01 3.84148985e-01 -8.05215240e-01 1.32317081e-01 1.70735255e-01 9.83106568e-02 3.46178979e-01 -1.31496680e+00 -8.84539187e-01 -1.93598792e-01 6.57848060e-01 -1.84966177e-01 -8.38131532e-02 2.22012892e-01 -5.20921886e-01 5.69274008e-01 1.02336265e-01 -4.92294520e-01 -1.43467176e+00 2.66632438e-01 4.89124686e-01 -7.50792623e-01 -1.05822042e-01 1.02925718e+00 -4.36267644e-01 -7.78229594e-01 1.88943431e-01 -6.01625144e-01 -2.14798316e-01 1.48733944e-01 4.13807601e-01 3.06383342e-01 5.70897758e-01 -5.07299662e-01 -6.08914673e-01 2.05403402e-01 -3.00081223e-01 1.54189333e-01 8.01105380e-01 1.99130684e-01 -1.61230385e-01 6.73663259e-01 1.12502599e+00 -4.39901724e-02 -2.25184217e-01 -5.18010080e-01 7.89993763e-01 -5.97764969e-01 -3.37475419e-01 -8.37853193e-01 -5.46481550e-01 5.32349348e-01 4.78615195e-01 7.10173607e-01 4.37545240e-01 -8.04450512e-02 4.15559441e-01 7.62967885e-01 5.07634521e-01 -1.62187612e+00 -2.31823668e-01 8.70067537e-01 1.82081312e-01 -1.48967373e+00 -1.62835285e-01 -5.32987595e-01 -5.77805758e-01 1.03631258e+00 6.40206754e-01 -1.25707150e-01 7.22353995e-01 4.81380299e-02 3.39043826e-01 -1.06313653e-01 -8.64125788e-01 -2.31671244e-01 -1.83990747e-01 5.29030204e-01 9.28548992e-01 4.59604204e-01 -8.33519220e-01 2.96306729e-01 -6.80750012e-02 -6.79628998e-02 5.60622156e-01 1.07461405e+00 -7.61443198e-01 -1.47252178e+00 -5.64925849e-01 8.82090211e-01 -7.01624036e-01 -3.55775654e-01 -3.29120517e-01 3.47264826e-01 4.51251686e-01 1.12629974e+00 -9.09262337e-03 -2.43649229e-01 2.93324441e-01 4.88007933e-01 1.54057935e-01 -1.30005753e+00 -1.22243404e+00 -7.52128422e-01 6.05003953e-01 4.16096300e-02 -5.85331321e-01 -9.57556188e-01 -1.62835848e+00 -3.32032025e-01 -6.95805728e-01 6.47267818e-01 6.68752909e-01 1.50582254e+00 1.18577899e-02 2.19850436e-01 8.27410161e-01 -4.26351398e-01 -8.63980591e-01 -1.28204083e+00 -1.11962545e+00 8.14427316e-01 1.84154212e-01 -6.55849814e-01 -5.06398439e-01 2.51955420e-01]

[9.78543758392334, 9.60146427154541]

b47e7e2b-bb76-4a31-a7e2-4ede716ba62c

reviewrobot-explainable-paper-review

2010.06119

https://arxiv.org/abs/2010.06119v3

https://arxiv.org/pdf/2010.06119v3.pdf

ReviewRobot: Explainable Paper Review Generation based on Knowledge Synthesis

To assist human review process, we build a novel ReviewRobot to automatically assign a review score and write comments for multiple categories such as novelty and meaningful comparison. A good review needs to be knowledgeable, namely that the comments should be constructive and informative to help improve the paper; and explainable by providing detailed evidence. ReviewRobot achieves these goals via three steps: (1) We perform domain-specific Information Extraction to construct a knowledge graph (KG) from the target paper under review, a related work KG from the papers cited by the target paper, and a background KG from a large collection of previous papers in the domain. (2) By comparing these three KGs, we predict a review score and detailed structured knowledge as evidence for each review category. (3) We carefully select and generalize human review sentences into templates, and apply these templates to transform the review scores and evidence into natural language comments. Experimental results show that our review score predictor reaches 71.4%-100% accuracy. Human assessment by domain experts shows that 41.7%-70.5% of the comments generated by ReviewRobot are valid and constructive, and better than human-written ones for 20% of the time. Thus, ReviewRobot can serve as an assistant for paper reviewers, program chairs and authors.

['Nazneen Fatema Rajani', 'Heng Ji', 'Kevin Knight', 'Lifu Huang', 'Qi Zeng', 'Qingyun Wang']

2020-10-13

https://aclanthology.org/2020.inlg-1.44

https://aclanthology.org/2020.inlg-1.44.pdf

inlg-acl-2020-12

['review-generation']

['natural-language-processing']

[-6.87164292e-02 6.24305665e-01 -7.32410252e-01 -4.95180726e-01 -9.36578870e-01 -6.45734489e-01 4.74780440e-01 5.08191288e-01 -8.22811052e-02 9.23003435e-01 9.64534134e-02 -3.44992906e-01 -1.51250586e-01 -7.16896057e-01 -5.60988247e-01 4.13679667e-02 4.94902760e-01 2.27257088e-01 3.75559449e-01 1.03405029e-01 9.00423110e-01 1.50708199e-01 -1.24037087e+00 2.94495374e-01 1.24242091e+00 8.03773224e-01 4.72103447e-01 8.40629816e-01 -4.79747295e-01 7.92101920e-01 -1.02370536e+00 -7.89033949e-01 -1.29281670e-01 -5.04201770e-01 -9.42787051e-01 2.55489200e-01 2.47228086e-01 -1.20518528e-01 2.52807200e-01 1.01499045e+00 4.16758507e-02 -1.73379064e-01 7.67049193e-01 -1.40208888e+00 -1.13665235e+00 9.61952150e-01 -8.02586734e-01 -5.06520383e-02 6.34842396e-01 -1.69660062e-01 1.18853295e+00 -1.11576200e+00 8.98972332e-01 1.04663634e+00 2.49467775e-01 3.11609983e-01 -2.94247866e-01 -7.40792572e-01 4.95650887e-01 2.92921096e-01 -9.15556133e-01 2.18927003e-02 6.97897553e-01 -5.81211686e-01 8.47100556e-01 1.26730919e-01 6.05494857e-01 5.19092560e-01 4.14026678e-01 6.18476093e-01 6.25726402e-01 -6.71177506e-01 2.99627304e-01 8.37632954e-01 6.60973132e-01 6.96502447e-01 6.85670495e-01 -8.57115149e-01 -6.28309667e-01 -2.95025092e-02 1.96256325e-01 -1.55734615e-02 -2.94144660e-01 5.03274389e-02 -1.05937016e+00 4.18653816e-01 2.14389652e-01 1.77447155e-01 -5.50776005e-01 -1.19003475e-01 2.84905553e-01 1.95754394e-01 3.03453445e-01 6.68540657e-01 -4.54170436e-01 -2.54208565e-01 -6.87545836e-01 2.26076722e-01 1.20851076e+00 1.39099073e+00 1.05135524e+00 -4.51732576e-01 -4.86556701e-02 8.96212995e-01 5.61024725e-01 9.37716544e-01 1.73662171e-01 -9.51291621e-01 6.71780288e-01 1.09630418e+00 2.34799534e-01 -1.51037979e+00 6.78009987e-02 -4.23710883e-01 -4.87706393e-01 -1.61275119e-01 -3.68857682e-01 -1.06708527e-01 -3.66616249e-01 1.06975102e+00 -5.80063015e-02 -5.72089851e-01 3.13798226e-02 7.96931624e-01 1.39152706e+00 8.39128315e-01 -1.53034534e-02 -4.97284293e-01 1.37882841e+00 -1.09783816e+00 -1.07720351e+00 -2.41559908e-01 5.80107927e-01 -9.83553469e-01 1.19532621e+00 8.39791536e-01 -1.13871360e+00 -7.31411576e-01 -1.27345157e+00 6.47936836e-02 -2.87558734e-01 7.72636831e-01 2.80330092e-01 2.41418809e-01 -9.21267867e-01 4.00106966e-01 -5.78639060e-02 -5.25657594e-01 1.37585938e-01 -4.46516722e-02 -2.16415539e-01 -2.34706819e-01 -1.12084877e+00 9.26746666e-01 -1.27511352e-01 -1.28822684e-01 -4.19215322e-01 -6.26489401e-01 -6.82159126e-01 -1.06781824e-02 4.46841389e-01 -6.76584840e-01 1.42198026e+00 -6.78902745e-01 -1.08447099e+00 1.09178388e+00 -4.49178576e-01 1.28099751e-02 1.96009548e-03 -5.22582769e-01 -5.78887165e-01 5.25626838e-01 3.73168379e-01 3.53441238e-01 5.81454039e-01 -1.36699581e+00 -9.44704354e-01 -8.50469023e-02 3.29201490e-01 1.18560560e-01 -5.07866204e-01 3.10468256e-01 -1.04176688e+00 -4.25204635e-01 -1.82466224e-01 -7.76514530e-01 -9.45929140e-02 -6.22586422e-02 -4.49590653e-01 -6.09357595e-01 7.51427591e-01 -9.11253452e-01 1.69980085e+00 -1.65339673e+00 -1.35853902e-01 3.48888785e-01 6.01492584e-01 3.18150192e-01 -4.51803431e-02 6.14934802e-01 1.89115778e-01 6.37755096e-01 -1.96785759e-02 -3.95589173e-02 2.94211134e-02 -1.10565580e-01 -4.23646539e-01 -1.29981086e-01 3.20514232e-01 9.85140502e-01 -1.20213056e+00 -6.23782992e-01 -2.56651610e-01 -1.54643908e-01 -7.49706551e-02 2.29520544e-01 -1.59237862e-01 -3.40992928e-01 -6.89390361e-01 7.74823964e-01 5.62223136e-01 -6.56046569e-01 4.31237929e-02 -8.06870759e-02 -3.91553016e-03 4.84996587e-01 -9.88061547e-01 1.10069263e+00 -4.38541502e-01 8.12077463e-01 -9.40155312e-02 -7.31369138e-01 1.66923594e+00 8.75779912e-02 9.46033597e-02 -3.90810549e-01 -5.46635799e-02 3.17459136e-01 -2.98309118e-01 -7.57820129e-01 7.62505054e-01 2.32934639e-01 -9.55078825e-02 9.77905512e-01 -3.55017692e-01 -6.26410067e-01 6.07677460e-01 1.01111138e+00 1.32949650e+00 -2.76535004e-01 5.67816377e-01 4.99652065e-02 8.10232818e-01 -5.65129006e-03 5.55356562e-01 9.04406786e-01 -8.53861049e-02 4.35065240e-01 9.07363057e-01 -1.21310718e-01 -7.59210229e-01 -7.32137918e-01 3.91788214e-01 6.48525536e-01 2.32921869e-01 -1.02292991e+00 -4.86615062e-01 -1.04354370e+00 1.88226588e-02 6.55503690e-01 -4.24717009e-01 -1.67914882e-01 -2.28937492e-01 8.75926018e-02 3.31671648e-02 6.68837547e-01 3.18077594e-01 -1.16010642e+00 -2.29658931e-01 1.37569159e-01 -2.56989479e-01 -1.03589594e+00 -4.80811387e-01 -1.34494171e-01 -4.16836739e-01 -1.47985446e+00 -6.95331037e-01 -7.94188797e-01 8.12006652e-01 5.96524537e-01 1.31554472e+00 6.52485490e-01 9.14967731e-02 7.32525468e-01 -8.84501338e-01 -8.43683660e-01 -5.18466949e-01 6.96252137e-02 -4.88500819e-02 -5.98512530e-01 5.30865610e-01 -1.76882789e-01 -2.55352676e-01 4.55132306e-01 -5.16218662e-01 -1.31672230e-02 7.39965558e-01 3.30298722e-01 4.78615046e-01 -4.33209017e-02 1.08387709e+00 -1.02041006e+00 1.15239882e+00 -4.04572248e-01 -3.00288051e-01 7.76287079e-01 -1.06800711e+00 -2.00895235e-01 5.08216858e-01 -1.98021457e-01 -9.79275048e-01 -4.88445759e-01 4.14271057e-01 -3.07921786e-02 2.99905568e-01 1.02521968e+00 -1.62076026e-01 2.79565006e-01 7.17878282e-01 -2.31915072e-01 -1.44187510e-01 -1.30392266e-02 1.62532806e-01 9.79179323e-01 3.70393813e-01 -5.88709414e-01 8.85753632e-01 -1.18090875e-01 -2.14894503e-01 -6.40414715e-01 -1.10311663e+00 -7.73911178e-01 -6.45617306e-01 -6.26340210e-01 5.56970716e-01 -6.67173386e-01 -5.10370851e-01 -1.40665740e-01 -1.49206805e+00 9.17940661e-02 -7.70655647e-02 4.44702357e-01 -6.55227108e-03 5.44457555e-01 -3.69033396e-01 -8.66029561e-01 -5.86311340e-01 -8.75927269e-01 8.64013553e-01 6.73128426e-01 -6.54872775e-01 -7.75476038e-01 -2.11263761e-01 4.63189274e-01 4.80737500e-02 -1.42034560e-01 8.33531022e-01 -5.63974679e-01 -6.51306152e-01 -5.29848754e-01 -4.74404484e-01 6.17598414e-01 1.53757825e-01 8.20648611e-01 -4.91232604e-01 2.27631614e-01 -3.64160568e-01 -2.15749651e-01 6.58499658e-01 1.82784960e-01 1.02974641e+00 -4.03358161e-01 -5.54325223e-01 -2.15615928e-01 7.89418936e-01 5.27816653e-01 5.67296863e-01 2.28384465e-01 5.39030373e-01 7.30770111e-01 1.16098464e+00 5.23004651e-01 6.51374400e-01 3.88698608e-01 7.26626208e-03 2.33133629e-01 4.90461364e-02 -3.29836786e-01 3.83115053e-01 1.64643693e+00 2.47164249e-01 -4.12221104e-01 -1.00581229e+00 7.16722906e-01 -2.04319572e+00 -7.09453285e-01 -5.82196891e-01 1.79178953e+00 8.73487175e-01 5.69482982e-01 -8.69425014e-02 1.87890440e-01 7.90239275e-01 -1.58892885e-01 -3.31340164e-01 -7.91646481e-01 9.43989679e-02 3.82533558e-02 -2.82366928e-02 3.86979073e-01 -4.84253019e-01 6.97153449e-01 6.02306652e+00 4.23813671e-01 -8.36236060e-01 -3.56336147e-01 3.46808046e-01 2.54521251e-01 -7.15297997e-01 3.21151376e-01 -1.12192023e+00 2.56421328e-01 6.45980597e-01 -1.06737518e+00 -1.10399939e-01 1.08795583e+00 3.55198056e-01 -3.46946180e-01 -1.03637683e+00 8.67813349e-01 3.75978500e-01 -1.39592767e+00 2.60225892e-01 -2.53527820e-01 8.74555886e-01 -6.66694820e-01 -2.83396214e-01 5.46425164e-01 2.65406370e-01 -6.78941071e-01 4.64313686e-01 8.70086074e-01 5.36554813e-01 -6.24423802e-01 1.00020778e+00 4.63276118e-01 -1.11096883e+00 2.43947566e-01 -5.54829359e-01 -2.00249597e-01 1.66945666e-01 1.13866079e+00 -1.12526381e+00 7.68445909e-01 5.91158450e-01 1.36470282e+00 -6.85423613e-01 9.14942622e-01 -1.08757353e+00 7.08791971e-01 1.94915920e-01 -7.86436498e-01 -4.43361662e-02 1.79732442e-02 2.94174939e-01 1.33752441e+00 2.56010443e-01 3.93134713e-01 2.52031982e-01 8.70029032e-01 -5.42811155e-01 2.89222211e-01 -3.44723076e-01 -4.46854055e-01 7.51345873e-01 1.57831311e+00 -5.69988072e-01 -5.57136595e-01 -2.94543833e-01 4.70866323e-01 1.03500396e-01 3.66784990e-01 -4.49913532e-01 -1.24123132e+00 2.47899126e-02 6.16769046e-02 2.22427368e-01 1.84709355e-01 -6.68773532e-01 -9.49615240e-01 7.23623216e-01 -7.22757697e-01 1.65891841e-01 -1.50843310e+00 -1.22401559e+00 4.81707662e-01 -3.24765086e-01 -1.41026878e+00 -8.38020071e-02 -5.41085720e-01 -8.59862864e-01 1.07446909e+00 -1.51522636e+00 -8.08527350e-01 -6.68575466e-01 -2.20605075e-01 6.27664387e-01 -2.72951156e-01 4.38897818e-01 -9.71264578e-03 -4.36204076e-01 3.61069053e-01 -6.36922002e-01 -1.44798189e-01 9.73369002e-01 -1.31935298e+00 2.47678339e-01 8.46357048e-01 -9.64533985e-02 1.18822372e+00 4.98630911e-01 -1.16233361e+00 -1.25691617e+00 -7.89229870e-01 1.56576025e+00 -5.82378387e-01 7.76065469e-01 1.87580958e-01 -1.21972144e+00 1.80873722e-01 2.20503643e-01 -5.30268788e-01 7.29408026e-01 3.87366235e-01 -2.85350025e-01 -3.06225508e-01 -6.35202885e-01 4.13310856e-01 7.15917110e-01 -5.38900971e-01 -1.08771157e+00 2.96610177e-01 8.98064673e-01 -1.84907675e-01 -7.76577532e-01 2.56167591e-01 4.97698009e-01 -4.07875717e-01 3.45073730e-01 -3.98705184e-01 1.21053684e+00 -5.76274216e-01 3.34425777e-01 -1.28140974e+00 -3.18599194e-01 -5.21845877e-01 -1.75389409e-01 1.52680910e+00 7.27664411e-01 -5.10297902e-02 4.56934869e-01 8.28959227e-01 -3.65478665e-01 -1.07249820e+00 2.82650045e-03 -5.83660185e-01 -3.51294577e-01 -4.30214286e-01 3.93687487e-01 9.15816009e-01 7.44059265e-01 5.88528991e-01 -1.20235972e-01 -6.96049780e-02 1.43247232e-01 1.12486511e-01 1.20595813e+00 -1.44173169e+00 9.22764912e-02 -3.70504260e-01 -7.75499716e-02 -1.11192894e+00 2.08783478e-01 -8.19101095e-01 1.42390534e-01 -2.56200624e+00 5.43973446e-01 -2.93547243e-01 -2.20860660e-01 5.81286252e-01 -4.70102012e-01 -4.36932266e-01 9.29920375e-02 4.56103891e-01 -1.01277053e+00 2.27985501e-01 1.44318235e+00 -1.69539332e-01 -2.08056450e-01 9.23869759e-02 -1.50209320e+00 6.60576761e-01 4.38872337e-01 -3.02775651e-01 -4.22361195e-01 -4.97646350e-03 9.25054550e-01 5.79667389e-02 8.27062353e-02 -6.73377335e-01 5.24295807e-01 -4.82740849e-01 2.04793483e-01 -9.80935276e-01 -2.62740552e-01 -5.38579345e-01 -3.06109071e-01 9.85561907e-02 -6.07247651e-01 2.31975153e-01 5.59253013e-03 4.83250886e-01 -2.38288015e-01 -6.27962530e-01 2.07624570e-01 -9.07015502e-02 -6.25334442e-01 -6.34363368e-02 -3.23617399e-01 -6.62586140e-03 8.36457253e-01 -1.72714651e-01 -8.18931401e-01 -7.52290368e-01 -2.70606816e-01 5.68497062e-01 4.10853535e-01 7.41904914e-01 8.70021462e-01 -9.90066767e-01 -6.72223985e-01 -3.96042287e-01 5.23293376e-01 -1.61716461e-01 -1.74444184e-01 5.37074149e-01 -1.85678810e-01 5.25385022e-01 2.00203240e-01 -4.04021591e-01 -1.34300303e+00 2.31940895e-01 -4.25217479e-01 -3.88657242e-01 -4.06454086e-01 6.66518211e-01 -6.75372034e-02 -2.03903377e-01 3.54535252e-01 -5.54870605e-01 -5.45696735e-01 2.00706452e-01 1.02253747e+00 2.86668986e-01 1.82778955e-01 -7.32750222e-02 -5.19139051e-01 6.98854268e-01 -4.19664502e-01 -1.00650996e-01 1.38209248e+00 -2.17873231e-01 -1.95581630e-01 6.05987906e-01 8.48660529e-01 4.81091380e-01 -6.23037100e-01 -2.43936822e-01 1.85181677e-01 -2.93630928e-01 -2.46418461e-01 -1.30520046e+00 -7.61837304e-01 7.47561038e-01 -5.27983367e-01 3.13288510e-01 8.91172528e-01 2.46550813e-01 5.44208467e-01 7.79058218e-01 3.12716633e-01 -1.33468390e+00 4.50875789e-01 6.17018819e-01 1.36770809e+00 -1.14164722e+00 5.86575210e-01 -7.02027619e-01 -9.94361460e-01 1.30374503e+00 8.14969897e-01 1.42740741e-01 5.81083357e-01 -2.57498045e-02 1.76033616e-01 -5.30999243e-01 -1.14323342e+00 2.75802255e-01 7.69753993e-01 5.85307956e-01 4.41037118e-01 -9.89267081e-02 -7.94694483e-01 1.21549618e+00 -1.99686170e-01 3.09526682e-01 1.00294745e+00 1.03164136e+00 -7.97623038e-01 -9.92958367e-01 -1.04517154e-01 7.26200461e-01 -1.47522926e-01 2.01464906e-01 -1.07732928e+00 5.62078714e-01 -2.92149425e-01 1.30663788e+00 -4.38782156e-01 -5.55981338e-01 6.00994945e-01 -1.86008543e-01 -1.03568204e-01 -1.09623027e+00 -7.05817878e-01 -2.26997763e-01 3.20684463e-01 -8.58934410e-03 -4.97179270e-01 -3.70396137e-01 -1.50592315e+00 -1.42782167e-01 -4.92910266e-01 6.82129383e-01 8.68119657e-01 1.01808023e+00 4.50718492e-01 6.00425184e-01 7.01297104e-01 -1.21980466e-01 -2.51986384e-01 -8.91494870e-01 -4.75211978e-01 2.96006471e-01 -1.24121673e-01 -5.07008433e-01 -4.96822089e-01 4.33458447e-01]

[12.294317245483398, 9.500164031982422]

c141237d-be41-4fc4-8fad-730795ba32e1

uncertainty-sensitive-learning-and-planning

https://openreview.net/forum?id=SkglVlSFPS

https://openreview.net/pdf?id=SkglVlSFPS

Uncertainty - sensitive learning and planning with ensembles

We propose a reinforcement learning framework for discrete environments in which an agent optimizes its behavior on two timescales. For the short one, it uses tree search methods to perform tactical decisions. The long strategic level is handled with an ensemble of value functions learned using $TD$-like backups. Combining these two techniques brings synergies. The planning module performs \textit{what-if} analysis allowing to avoid short-term pitfalls and boost backups of the value function. Notably, our method performs well in environments with sparse rewards where standard $TD(1)$ backups fail. On the other hand, the value functions compensate for inherent short-sightedness of planning. Importantly, we use ensembles to measure the epistemic uncertainty of value functions. This serves two purposes: a) it stabilizes planning, b) it guides exploration. We evaluate our methods on discrete environments with sparse rewards: the Deep sea chain environment, toy Montezuma's Revenge, and Sokoban. In all the cases, we obtain speed-up of learning and boost to the final performance.

['Maciej Klimek', 'Piotr Kozakowski', 'Konrad Czechowski', 'Łukasz Kuciński', 'Piotr Miłoś']

2019-09-25

['montezumas-revenge']

['playing-games']

[-2.31865510e-01 4.02754515e-01 1.87844597e-02 8.61710906e-02 -7.72848189e-01 -6.93625569e-01 6.27984703e-01 4.02707636e-01 -8.73859704e-01 1.29589593e+00 1.38117343e-01 -2.66280502e-01 -5.72407067e-01 -9.93882000e-01 -7.20034063e-01 -1.03543973e+00 -6.84059680e-01 5.05131900e-01 1.69846609e-01 -6.38923466e-01 3.87163937e-01 1.74819499e-01 -1.29525483e+00 -3.79045576e-01 1.00016475e+00 1.05408251e+00 2.76568025e-01 4.82075214e-01 1.99780926e-01 1.02839947e+00 -4.89539385e-01 -1.34300277e-01 6.89805806e-01 -6.32845610e-02 -6.39528573e-01 -3.06091577e-01 -5.16969681e-01 -4.78476763e-01 -9.18670818e-02 1.05569863e+00 3.74940813e-01 4.12966996e-01 5.23650885e-01 -1.02051282e+00 -4.14781906e-02 1.01265979e+00 -5.24666011e-01 5.58763556e-02 1.33867949e-01 6.26353920e-01 1.17406344e+00 -1.12577565e-01 3.61011177e-01 1.21747875e+00 3.30497801e-01 2.24582568e-01 -1.07441509e+00 -3.72348487e-01 5.60066760e-01 1.03903547e-01 -9.55468476e-01 -2.85955131e-01 3.75437438e-01 -4.12969738e-01 8.27876627e-01 -4.75180820e-02 7.90200889e-01 7.93026567e-01 4.46392208e-01 6.49933457e-01 1.24206841e+00 -1.84191987e-01 9.24233139e-01 -2.09663004e-01 -4.35087591e-01 5.16223133e-01 2.14626387e-01 8.28116119e-01 -6.07635617e-01 -9.94593352e-02 5.48430860e-01 -2.29761124e-01 -1.29972830e-01 -3.18142354e-01 -9.43415523e-01 7.28364110e-01 4.34082568e-01 -3.22680809e-02 -6.79169953e-01 3.80159378e-01 1.72913551e-01 7.99650848e-01 -6.87261019e-03 8.17444503e-01 -4.41533715e-01 -5.26700497e-01 -6.08950615e-01 4.49208945e-01 8.15814316e-01 5.59670150e-01 8.78312945e-01 1.26731977e-01 -6.00016192e-02 4.23305243e-01 2.09982187e-01 3.06164116e-01 2.44230613e-01 -1.34980571e+00 6.24477446e-01 1.41055539e-01 8.37555707e-01 -6.47000313e-01 -5.46325088e-01 -6.40827298e-01 -3.62213284e-01 7.87209392e-01 5.95338285e-01 -4.75944042e-01 -6.42810643e-01 1.98451042e+00 3.13005626e-01 -1.99230894e-01 3.08862686e-01 7.50203490e-01 -1.23887353e-01 3.53298247e-01 -1.58943728e-01 -3.58125120e-01 9.00448918e-01 -7.97580600e-01 -4.96711671e-01 -4.63209838e-01 4.80610907e-01 -9.85778943e-02 1.02934229e+00 4.75508362e-01 -1.12806523e+00 1.96167566e-02 -9.77383494e-01 6.78495049e-01 -6.18992709e-02 -4.65113014e-01 6.36202037e-01 4.83592361e-01 -1.08796048e+00 9.98594105e-01 -9.89188612e-01 8.80524293e-02 2.27515906e-01 3.48587215e-01 3.57735828e-02 3.92377794e-01 -1.31143367e+00 1.13645732e+00 3.47217500e-01 5.41784093e-02 -1.42870796e+00 -3.61527234e-01 -7.02908337e-01 3.33356649e-01 9.07997072e-01 -3.30403596e-01 1.46652925e+00 -6.38258100e-01 -1.95421433e+00 1.28976405e-01 4.58079487e-01 -6.84000552e-01 8.56463730e-01 -2.79606551e-01 2.68849403e-01 -7.57701918e-02 3.70652467e-01 4.20326769e-01 5.25897026e-01 -1.04959559e+00 -6.25043452e-01 -3.72969329e-01 4.46375459e-01 6.44209862e-01 2.24954356e-02 -5.85630953e-01 2.74585783e-01 -3.52586269e-01 -7.55388215e-02 -9.63582993e-01 -6.59750998e-01 -3.53585720e-01 -6.85492679e-02 1.01246819e-01 -1.77135035e-01 -2.21371144e-01 1.18588996e+00 -1.98346841e+00 3.74247819e-01 3.15482467e-01 -1.47599086e-01 -3.87682348e-01 -1.19055912e-01 7.77519345e-01 5.46431363e-01 3.33337858e-02 -3.58641952e-01 -1.63742989e-01 2.74376124e-01 4.39948052e-01 -3.43487322e-01 3.25704247e-01 -1.00790158e-01 6.16027832e-01 -1.10047257e+00 -1.37813404e-01 -2.26189028e-02 -3.37872356e-01 -6.82968974e-01 1.39379323e-01 -4.25501287e-01 5.19511342e-01 -6.49919510e-01 5.30177057e-01 2.24492922e-01 1.07568309e-01 5.46311915e-01 6.30940139e-01 -4.99280274e-01 2.82326996e-01 -1.33213973e+00 1.52071416e+00 -3.19680035e-01 3.10159791e-02 4.55743134e-01 -8.37584138e-01 7.40363777e-01 -1.28316507e-01 4.31187153e-01 -9.32035744e-01 2.72649806e-02 4.58802849e-01 1.38015449e-01 -3.53198230e-01 5.41303098e-01 -2.89051950e-01 -3.95311356e-01 6.18147016e-01 -2.05451682e-01 -4.07237262e-01 1.39661804e-01 7.75911147e-03 1.35921800e+00 4.82556403e-01 5.26984751e-01 -5.95747828e-01 1.30893618e-01 -4.91444767e-02 8.15038502e-01 1.07802927e+00 -4.09692049e-01 -1.09948143e-01 1.02108634e+00 -3.99805069e-01 -5.85280299e-01 -9.93234456e-01 1.79854900e-01 1.18244970e+00 5.29223919e-01 -1.62055343e-01 -4.35762018e-01 -5.82947195e-01 1.72638208e-01 9.75592077e-01 -9.64880645e-01 -1.02272511e-01 -5.37202001e-01 -6.30026817e-01 2.72644341e-01 3.67781967e-01 5.35337746e-01 -1.03660572e+00 -1.32556570e+00 3.12998980e-01 2.17839070e-02 -4.21686471e-01 -3.33916873e-01 7.81420052e-01 -7.27938831e-01 -9.99803066e-01 -4.95922089e-01 -1.52269885e-01 2.31050581e-01 -3.01965810e-02 9.42899406e-01 -2.04581097e-01 4.11373824e-01 3.08892310e-01 -3.49741638e-01 -3.43115598e-01 -6.62266016e-02 -1.56018868e-01 3.29391539e-01 -3.37179929e-01 -2.12095663e-01 -9.03305709e-01 -5.94417155e-01 2.13269904e-01 -5.69345295e-01 -2.39696294e-01 5.25245130e-01 1.08168769e+00 3.55570406e-01 1.84510589e-01 7.17104077e-01 -3.85098398e-01 7.20809460e-01 -6.19513929e-01 -1.08161652e+00 2.46009588e-01 -6.34206891e-01 5.85812986e-01 6.66647613e-01 -2.41596311e-01 -1.03592408e+00 -2.64942318e-01 1.57591894e-01 1.99737884e-02 2.52008826e-01 5.84292769e-01 8.87882933e-02 -3.57832015e-03 7.31230617e-01 1.41653076e-01 1.67688102e-01 -2.28556260e-01 9.20374542e-02 2.35054612e-01 1.89591601e-01 -1.13570607e+00 6.78258240e-01 2.64509946e-01 9.61524397e-02 -4.85450119e-01 -5.27430236e-01 2.55770862e-01 -1.36770338e-01 -3.52714121e-01 3.89790535e-01 -7.28839576e-01 -1.12810516e+00 2.43822441e-01 -6.16641462e-01 -9.26832616e-01 -7.00976074e-01 3.81903350e-01 -1.07547200e+00 2.04469338e-01 -3.73514175e-01 -1.23115253e+00 2.72927247e-02 -1.33450449e+00 6.45276606e-01 4.23917860e-01 1.90834686e-01 -7.24532425e-01 4.12893623e-01 -1.45525828e-01 5.24024129e-01 4.51831430e-01 6.85924232e-01 -4.22011048e-01 -6.81376040e-01 4.77668285e-01 3.31621319e-01 -2.70640496e-02 -9.62686166e-02 -3.48741442e-01 -7.24680305e-01 -4.25156564e-01 1.13235041e-01 -4.71982211e-01 6.95911765e-01 4.10211891e-01 6.75303817e-01 -6.75052583e-01 6.91361427e-02 4.35824424e-01 1.18592513e+00 6.58977807e-01 4.02013034e-01 1.01726735e+00 -2.64298469e-01 7.57962108e-01 9.27131593e-01 1.06948841e+00 6.11815035e-01 6.43999219e-01 1.07302737e+00 6.67651474e-01 4.97205168e-01 -2.81491429e-01 6.37468755e-01 2.65173942e-01 -1.63763762e-01 7.80935660e-02 -9.30496633e-01 4.56063211e-01 -2.22948694e+00 -9.78460491e-01 6.08943403e-01 2.41445422e+00 9.10275877e-01 3.72271717e-01 3.03322464e-01 -2.29653701e-01 1.81474403e-01 2.25553825e-01 -8.84443939e-01 -3.22516203e-01 1.15489298e-02 -1.44942120e-01 5.90848505e-01 7.89689362e-01 -7.29567528e-01 7.82677352e-01 6.07128906e+00 6.92752004e-01 -8.63599360e-01 -3.45974565e-02 5.49328744e-01 -4.73618805e-01 -4.68575388e-01 1.81593135e-01 -4.30227339e-01 5.40766060e-01 7.18159318e-01 -1.82523698e-01 9.49340940e-01 9.10698891e-01 2.87057817e-01 -7.27529466e-01 -9.53085840e-01 4.48185295e-01 -6.38902962e-01 -1.26228142e+00 -6.34679377e-01 2.62573719e-01 5.55258393e-01 1.91518679e-01 -2.80026509e-03 6.09238207e-01 1.01203048e+00 -9.24502015e-01 1.32084107e+00 5.47626376e-01 4.19504791e-01 -1.05956423e+00 6.86083794e-01 7.83992767e-01 -9.65676188e-01 -5.82215190e-01 -4.98050712e-02 -5.15237987e-01 1.10637836e-01 3.67639691e-01 -6.82396591e-01 5.23542702e-01 5.92334569e-01 1.59887746e-01 -1.04826294e-01 8.55034709e-01 -3.78116667e-01 2.29422539e-01 -5.68715930e-01 -3.11971903e-01 7.24580228e-01 -5.69686174e-01 6.73657835e-01 5.64711332e-01 2.92600662e-01 2.53322810e-01 3.69910777e-01 8.53200793e-01 2.70478517e-01 -2.88345009e-01 -5.85766137e-01 1.01633435e-02 7.47556686e-01 8.38240266e-01 -5.39209425e-01 1.57809764e-01 2.66126782e-01 3.49011689e-01 6.30562484e-01 3.31911743e-01 -7.20986843e-01 -2.78936356e-01 6.69623673e-01 -2.91693717e-01 1.24070011e-01 -3.70362818e-01 -3.83032143e-01 -1.08357930e+00 -7.03897029e-02 -8.81591797e-01 2.70476758e-01 -1.95116207e-01 -8.97268176e-01 3.71432632e-01 1.24565467e-01 -1.06275845e+00 -5.37095249e-01 -3.16863388e-01 -5.80725014e-01 5.66143692e-01 -1.68047416e+00 -5.86571753e-01 9.15778205e-02 2.53404200e-01 4.19392645e-01 -1.60990939e-01 6.64084971e-01 -2.34207273e-01 -6.47061169e-01 2.11510047e-01 3.13053131e-01 -3.70089471e-01 2.85036832e-01 -1.44913375e+00 -3.91792096e-02 7.11295545e-01 -3.87741119e-01 3.63617688e-01 9.36493933e-01 -6.58936918e-01 -1.28500497e+00 -5.17363489e-01 1.73217505e-01 -1.34250239e-01 8.68017077e-01 3.98158357e-02 -3.88353080e-01 4.60669994e-01 1.66991249e-01 -3.90278190e-01 3.63914907e-01 3.52080733e-01 -1.18102707e-01 -1.92835778e-01 -1.15120101e+00 7.44959354e-01 7.99086392e-01 -3.88336480e-02 -4.93385464e-01 1.07854672e-01 5.57991803e-01 -4.29531157e-01 -7.04297185e-01 3.16223621e-01 6.85320020e-01 -1.23359406e+00 5.28989017e-01 -5.74314535e-01 3.59214067e-01 -3.50278556e-01 -4.86847341e-01 -1.75041103e+00 -2.12561518e-01 -1.20014572e+00 -9.87277180e-02 6.77969277e-01 4.08354670e-01 -7.68603683e-01 5.78840256e-01 5.45676649e-01 -1.33974463e-01 -9.10739064e-01 -1.29187238e+00 -8.91296446e-01 3.55281115e-01 -1.42630711e-01 7.71396518e-01 4.91434872e-01 1.78887099e-01 -1.12165615e-01 -4.39543933e-01 4.27453443e-02 9.02487218e-01 1.55583262e-01 5.60534894e-01 -8.46639335e-01 -7.73205698e-01 -4.62489277e-01 2.33414739e-01 -9.18930590e-01 -2.26388016e-04 -3.56021255e-01 2.82269537e-01 -1.08344138e+00 -1.49567097e-01 -8.82815897e-01 -3.33101124e-01 5.51545143e-01 9.62118879e-02 -5.90133786e-01 5.07875502e-01 1.72256932e-01 -7.57389367e-01 9.18793261e-01 1.16016865e+00 -2.57860925e-02 -4.81803417e-01 1.24073282e-01 -7.96074450e-01 7.45679438e-01 8.44216347e-01 -3.23488086e-01 -3.68183941e-01 -4.68955427e-01 5.80934107e-01 7.12112546e-01 9.08597708e-02 -8.46249700e-01 2.58891702e-01 -9.46062624e-01 -2.06457451e-01 -1.18346557e-01 2.39912197e-01 -6.08262479e-01 1.88493967e-01 7.36213028e-01 -5.02556980e-01 8.42517987e-02 4.20010313e-02 8.15560520e-01 -3.66900899e-02 -4.24625009e-01 8.98885369e-01 -5.14037788e-01 -5.08902431e-01 -2.81678326e-02 -5.65070450e-01 2.01208800e-01 1.10986102e+00 -7.23829269e-02 -3.93558443e-01 -6.43808365e-01 -7.04870522e-01 9.27130342e-01 5.27175069e-01 -1.27125174e-01 2.88426667e-01 -9.19763148e-01 -4.60241169e-01 1.93492117e-04 -2.17572525e-01 1.26503348e-01 2.46209487e-01 9.46797729e-01 -3.90822083e-01 1.13156335e-02 -4.83847469e-01 -1.88113928e-01 -4.22339350e-01 1.64632529e-01 6.75727546e-01 -6.91641927e-01 -3.39822829e-01 8.57106090e-01 5.24587929e-02 -3.30980033e-01 3.81267607e-01 -1.47188351e-01 -1.76128462e-01 3.91906410e-01 3.92351627e-01 4.71493810e-01 -9.96300802e-02 6.90675974e-02 -3.94000322e-01 1.35552332e-01 1.76617250e-01 -6.45274818e-01 1.71372437e+00 -2.76586264e-01 1.29182428e-01 3.26556414e-01 2.61966646e-01 -3.85917351e-02 -1.94732320e+00 -1.62351243e-02 1.75411686e-01 -3.85407627e-01 9.83741693e-03 -1.13395548e+00 -5.71596801e-01 6.28308594e-01 1.44731253e-01 4.02516365e-01 1.04903901e+00 -4.25918043e-01 2.10506946e-01 7.79963017e-01 1.02747500e+00 -1.47918475e+00 1.50237188e-01 8.88792574e-01 8.56462121e-01 -1.00870097e+00 -1.65586874e-01 4.80111748e-01 -9.92287338e-01 1.05869496e+00 4.81525481e-01 -1.36459574e-01 2.23735780e-01 4.14440662e-01 -1.97620317e-01 4.36030291e-02 -1.11478961e+00 -4.37486142e-01 -4.72518444e-01 5.07814109e-01 -3.07290614e-01 3.42420250e-01 -1.85917512e-01 5.99039197e-01 -2.96395302e-01 -3.34307015e-01 7.34180331e-01 1.01532638e+00 -8.93723488e-01 -9.61023748e-01 -3.38849396e-01 3.27172995e-01 -2.11537018e-01 -3.59040424e-02 -6.68053105e-02 7.52909720e-01 -1.38569444e-01 8.84922206e-01 -2.08509102e-01 -3.11226845e-01 1.91664487e-01 -1.67339846e-01 3.25466692e-01 -3.19618285e-01 -8.01611781e-01 9.64352563e-02 3.64011109e-01 -7.97376573e-01 -1.15637355e-01 -8.93103480e-01 -1.23847210e+00 -1.94084808e-01 -2.33450141e-02 5.86837590e-01 4.11145627e-01 9.40444887e-01 2.46925786e-01 4.18260962e-01 9.22549069e-01 -9.04060304e-01 -1.51575458e+00 -7.15718389e-01 -8.45044553e-01 -2.05099985e-01 5.63758433e-01 -1.06107688e+00 -4.95083600e-01 -7.26986766e-01]

[3.9930639266967773, 2.0512382984161377]

610155cc-4988-4f1b-8c96-6c41a8c616a4

bayesian-optimization-with-formal-safety

2306.17815

https://arxiv.org/abs/2306.17815v1

https://arxiv.org/pdf/2306.17815v1.pdf

Bayesian Optimization with Formal Safety Guarantees via Online Conformal Prediction

Black-box zero-th order optimization is a central primitive for applications in fields as diverse as finance, physics, and engineering. In a common formulation of this problem, a designer sequentially attempts candidate solutions, receiving noisy feedback on the value of each attempt from the system. In this paper, we study scenarios in which feedback is also provided on the safety of the attempted solution, and the optimizer is constrained to limit the number of unsafe solutions that are tried throughout the optimization process. Focusing on methods based on Bayesian optimization (BO), prior art has introduced an optimization scheme -- referred to as SAFEOPT -- that is guaranteed not to select any unsafe solution with a controllable probability over feedback noise as long as strict assumptions on the safety constraint function are met. In this paper, a novel BO-based approach is introduced that satisfies safety requirements irrespective of properties of the constraint function. This strong theoretical guarantee is obtained at the cost of allowing for an arbitrary, controllable but non-zero, rate of violation of the safety constraint. The proposed method, referred to as SAFE-BOCP, builds on online conformal prediction (CP) and is specialized to the cases in which feedback on the safety constraint is either noiseless or noisy. Experimental results on synthetic and real-world data validate the advantages and flexibility of the proposed SAFE-BOCP.

['Osvaldo Simeone', 'Sangwoo Park', 'Yunchuan Zhang']

2023-06-30

['conformal-prediction', 'bayesian-optimization', 'conformal-prediction']

['computer-vision', 'methodology', 'reasoning']

[ 2.54210621e-01 3.76465172e-01 -4.38198969e-02 -1.33918494e-03 -4.72568661e-01 -3.76600206e-01 3.04345667e-01 4.97238159e-01 -3.90322745e-01 7.92433858e-01 -3.34984094e-01 -2.77477652e-01 -6.87249780e-01 -8.86071801e-01 -7.92132437e-01 -9.87787068e-01 6.85921311e-02 -6.24512555e-03 4.21639793e-02 -1.05696060e-01 3.27267647e-01 4.07476127e-01 -1.18440950e+00 -4.98080641e-01 8.30031276e-01 1.28559780e+00 4.73195314e-02 4.22580302e-01 5.61954141e-01 2.13096872e-01 -3.36168557e-01 -2.73188263e-01 5.64281285e-01 -2.19461679e-01 -2.25977361e-01 2.10541621e-01 -1.38018310e-01 -6.96791187e-02 -1.56714767e-01 1.29333627e+00 4.09117639e-01 2.56252289e-01 4.60391551e-01 -9.84668553e-01 -1.12150021e-01 2.72271723e-01 -8.62402469e-02 1.35782212e-02 2.27434769e-01 1.01868235e-01 9.35352027e-01 -6.12323999e-01 3.33830118e-01 5.74480414e-01 3.22038919e-01 5.95903337e-01 -1.34913158e+00 -4.93564367e-01 2.15653524e-01 -1.11014113e-01 -1.70073307e+00 -2.93103606e-01 7.72872388e-01 -5.97817004e-01 2.48148873e-01 5.53869665e-01 4.18981522e-01 7.40726709e-01 7.03219891e-01 3.34834009e-01 9.32495236e-01 -4.82715428e-01 7.92761862e-01 4.92987990e-01 6.63028583e-02 4.34186608e-01 3.33758026e-01 6.29290223e-01 -5.75897694e-01 -2.90605932e-01 4.54811305e-01 -4.34335232e-01 -4.93593663e-01 -4.70997781e-01 -7.10759640e-01 8.33698273e-01 2.08688706e-01 1.22407153e-01 -3.69822025e-01 -9.03986543e-02 -6.37627915e-02 2.46840701e-01 4.60298657e-01 4.60654289e-01 -7.95799494e-02 8.61614794e-02 -7.76878715e-01 4.73172992e-01 7.65842080e-01 9.97934878e-01 2.77507484e-01 2.09742785e-01 -4.52189296e-01 3.32946211e-01 3.73934358e-01 3.23247850e-01 -2.79716272e-02 -6.45749509e-01 4.24600482e-01 2.92408705e-01 6.40636444e-01 -1.15930784e+00 -1.46562800e-01 -9.61779892e-01 -7.05856144e-01 3.45916986e-01 2.51442820e-01 -2.20960364e-01 -5.98152816e-01 1.72955608e+00 4.80650455e-01 8.64095688e-02 -1.64095536e-01 1.08525801e+00 4.97119203e-02 7.03895330e-01 -3.00535440e-01 -8.56297195e-01 6.99640274e-01 -1.15764625e-01 -8.26166332e-01 1.64294556e-01 9.24428999e-02 -4.59305257e-01 8.71824980e-01 8.04078937e-01 -9.60557103e-01 -1.50870606e-01 -1.19481421e+00 8.26538146e-01 5.99828884e-02 -1.55683421e-02 3.76057550e-02 1.17766368e+00 -6.22082114e-01 6.67683482e-01 -6.38438165e-01 2.70787273e-02 9.43138450e-02 5.49799681e-01 3.63222621e-02 2.61709601e-01 -8.79124820e-01 7.00962782e-01 4.28214818e-01 3.93726379e-01 -1.02753532e+00 -7.41270959e-01 -6.26852155e-01 3.02449524e-01 9.14546847e-01 -3.58567894e-01 1.12900841e+00 -6.14762783e-01 -1.86190236e+00 9.24459025e-02 5.97136766e-02 -5.72350800e-01 9.67706680e-01 -3.80827129e-01 -1.74181446e-01 -8.90089124e-02 -2.20428750e-01 -1.86376125e-01 1.08651364e+00 -1.17771494e+00 -3.48188758e-01 -2.69795626e-01 3.24224122e-02 2.55310815e-02 -3.12964737e-01 -2.50058293e-01 -2.17576548e-01 -5.31852961e-01 7.18622506e-02 -1.04789209e+00 -5.17892003e-01 -5.42438328e-02 -6.30399168e-01 3.34920041e-04 5.84895790e-01 -3.08838576e-01 1.42110038e+00 -2.06628370e+00 2.40199581e-01 8.50715816e-01 -2.78483123e-01 9.50754806e-02 2.88594753e-01 5.25067866e-01 1.41723469e-01 9.63785201e-02 -5.55477142e-01 -1.96246088e-01 -6.67779893e-02 1.41670048e-01 -4.51384038e-01 9.85861063e-01 2.21359462e-01 1.26143083e-01 -7.14974105e-01 4.94786575e-02 4.25470382e-01 2.98778772e-01 -6.43945813e-01 2.81055331e-01 -3.53795201e-01 8.18265021e-01 -7.82164514e-01 3.72214407e-01 4.40212876e-01 5.40553816e-02 -3.60366516e-02 2.83651769e-01 -3.73974919e-01 -2.17128634e-01 -1.60432661e+00 1.01201749e+00 -5.81663132e-01 1.01927608e-01 2.52229691e-01 -1.06478179e+00 8.77890468e-01 4.33264107e-01 4.44837570e-01 -3.98045242e-01 4.69083339e-01 1.02528155e-01 4.43664715e-02 -4.74980799e-03 2.69460112e-01 -1.35815904e-01 -1.60445809e-01 -9.56163406e-02 -1.53536394e-01 -1.79832295e-01 1.22583829e-01 1.82536058e-02 8.41090381e-01 -1.11832693e-01 3.84420663e-01 -5.67843556e-01 8.17534983e-01 -4.62415427e-01 8.32533002e-01 9.31149662e-01 8.11416283e-02 3.71913433e-01 5.02206087e-01 1.25110477e-01 -8.73887181e-01 -8.92962933e-01 -4.26725894e-01 3.33966434e-01 4.58283305e-01 -4.19366688e-01 -5.71767747e-01 -4.77090567e-01 -3.78388390e-02 9.84826148e-01 -5.04076183e-01 -3.36885333e-01 -2.53553659e-01 -6.69414461e-01 -1.91258252e-01 -1.30888626e-01 1.94039896e-01 -4.94859517e-01 -8.83298993e-01 2.88049042e-01 3.28596026e-01 -9.58032191e-01 -3.83746415e-01 2.32153550e-01 -6.11620367e-01 -8.21147561e-01 -4.87045020e-01 -9.88432392e-02 7.61819601e-01 -3.46422762e-01 5.27914286e-01 -4.95566204e-02 -2.27046564e-01 6.41415656e-01 -2.39618212e-01 -3.60439062e-01 -4.86778408e-01 -3.01796943e-01 2.29044452e-01 7.25808382e-01 -5.08429646e-01 -2.88217425e-01 -3.20768386e-01 4.73721057e-01 -7.54113495e-01 -3.66928220e-01 2.56140113e-01 8.58073890e-01 6.48802102e-01 5.57951629e-01 5.46103358e-01 -6.80739284e-01 6.45365894e-01 -4.80346918e-01 -1.51096416e+00 2.51402229e-01 -9.68730748e-01 2.26395860e-01 8.85909438e-01 -4.39011961e-01 -9.32090759e-01 1.82260320e-01 1.13328882e-01 -4.84686047e-01 1.12780221e-01 4.35972959e-01 -3.60041350e-01 -4.61021096e-01 4.44149196e-01 2.47147560e-01 -3.38048428e-01 -2.69024611e-01 1.04993880e-01 3.09008867e-01 3.49933267e-01 -7.12379873e-01 9.98832166e-01 2.17717677e-01 5.76246619e-01 -9.95534480e-01 -6.78536773e-01 -3.80968869e-01 -9.71384645e-02 -7.00354278e-01 5.17333508e-01 -2.78772742e-01 -1.16495752e+00 5.06579988e-02 -8.48723114e-01 7.42192566e-02 -2.37112865e-01 4.59279418e-01 -6.86928272e-01 2.92742133e-01 -4.95399069e-03 -1.78364253e+00 -5.30765355e-02 -1.09478605e+00 5.86191595e-01 3.72263789e-01 -1.94332600e-02 -7.20104933e-01 -1.74326852e-01 -4.92670797e-02 4.40436274e-01 4.50100094e-01 5.27716398e-01 -3.66687030e-01 -8.40671360e-01 -5.58678687e-01 3.53711307e-01 4.08147991e-01 -1.81060910e-01 -1.43533215e-01 -5.79505563e-01 -2.98182279e-01 5.50098658e-01 1.11181661e-03 4.93655533e-01 5.06772995e-01 1.23495722e+00 -7.70093083e-01 -1.27745643e-01 3.66200477e-01 1.63191044e+00 3.20791423e-01 2.52289325e-01 1.69019461e-01 7.59398937e-02 6.10450149e-01 8.40273738e-01 9.84794199e-01 -2.87325472e-01 8.98570538e-01 7.29852021e-01 4.67608839e-01 5.91418445e-01 -1.22640282e-01 2.58565217e-01 3.16842943e-01 1.84413612e-01 -5.29241502e-01 -4.20536757e-01 4.22025204e-01 -1.86633849e+00 -6.73544765e-01 8.56919363e-02 2.97567487e+00 7.61839271e-01 4.75320935e-01 -1.41130999e-01 3.70546997e-01 7.51040816e-01 -1.39499620e-01 -4.73267436e-01 -5.20526171e-01 1.68655798e-01 9.45205539e-02 8.57473433e-01 8.01806748e-01 -9.66820240e-01 2.36232609e-01 5.42878866e+00 9.29002166e-01 -1.09297156e+00 -6.59256056e-02 6.94250882e-01 -1.66476965e-01 -2.07895055e-01 2.07407296e-01 -8.85800540e-01 7.81476557e-01 8.72525990e-01 -3.98533881e-01 4.39059466e-01 8.84020686e-01 4.32964295e-01 -4.68914598e-01 -9.29168105e-01 6.20410264e-01 -2.40910351e-01 -1.13411129e+00 -5.02557993e-01 1.93533972e-01 5.88442028e-01 -7.48131931e-01 3.28771621e-01 -1.27243906e-01 -2.47732885e-02 -8.15347970e-01 9.48417902e-01 7.40869403e-01 3.94857228e-01 -1.08685124e+00 6.16702020e-01 6.99567199e-01 -8.97657573e-01 -4.82274652e-01 -2.98305862e-02 9.83904749e-02 3.79353851e-01 8.98089826e-01 -6.74229324e-01 8.96739900e-01 3.52711558e-01 2.30948746e-01 3.83059345e-02 1.28998387e+00 -4.65417862e-01 8.06600392e-01 -6.13152683e-01 -1.96979702e-01 4.92799431e-02 -2.63153404e-01 1.11698854e+00 8.18288624e-01 4.12844360e-01 2.13071123e-01 3.78551960e-01 1.05149126e+00 4.16985810e-01 1.39949530e-01 -4.78696764e-01 1.66315690e-01 3.96091670e-01 9.72953260e-01 -5.09541929e-01 1.39160573e-01 -2.98892297e-02 4.96873468e-01 -8.77891257e-02 2.53218830e-01 -7.98566878e-01 -1.42350078e-01 8.85470062e-02 1.46508992e-01 3.04152280e-01 -3.57017219e-01 -6.31044745e-01 -6.37149811e-01 2.57305264e-01 -4.25083131e-01 4.26490277e-01 -1.45418242e-01 -1.12557232e+00 2.07789972e-01 7.89566562e-02 -1.25883484e+00 -1.46022886e-01 -3.61544609e-01 -6.37539923e-01 8.45484734e-01 -1.24843633e+00 -5.27489066e-01 9.65202376e-02 3.88579756e-01 3.62854689e-01 -1.03947900e-01 4.61559653e-01 1.43027753e-01 -8.58774066e-01 6.28023148e-01 8.59795585e-02 -5.31016350e-01 1.27478972e-01 -1.07182717e+00 -5.98461986e-01 1.11149561e+00 -8.07461962e-02 5.09476066e-01 1.32277310e+00 -7.87042141e-01 -1.35035574e+00 -1.05477619e+00 6.19717777e-01 -4.46594767e-02 5.60874939e-01 -3.84515613e-01 -7.01682210e-01 3.35093625e-02 -2.35642090e-01 7.92787224e-02 2.46749684e-01 -2.41216242e-01 2.95821697e-01 -2.25606650e-01 -1.21199346e+00 5.84491313e-01 4.38652605e-01 -5.36192618e-02 -1.55519277e-01 3.41189802e-01 5.97153425e-01 -3.55717957e-01 -7.83600032e-01 7.21384764e-01 2.47725829e-01 -7.50923991e-01 7.77638733e-01 -2.95501143e-01 -3.11519150e-02 -4.09107149e-01 -2.77026892e-01 -1.08012950e+00 -7.06100240e-02 -1.17089844e+00 -3.39387655e-01 9.17787910e-01 3.48141104e-01 -5.47259927e-01 6.51274085e-01 9.59779203e-01 -8.20587128e-02 -9.47416067e-01 -1.47684491e+00 -1.13734472e+00 -2.34408647e-01 -4.55018878e-01 6.40154704e-02 3.77297580e-01 3.17572430e-02 -1.23011470e-01 -6.95171893e-01 8.76140058e-01 7.35781133e-01 -1.79334611e-01 4.46704894e-01 -9.49976325e-01 -6.55224860e-01 -2.28244409e-01 -3.20810854e-01 -7.29236841e-01 -2.27783173e-02 -4.66254145e-01 3.41884226e-01 -8.01699400e-01 -2.48220056e-01 -4.59971100e-01 -3.72669876e-01 -1.33060277e-01 2.45647114e-02 -2.93128490e-01 1.51789173e-01 -1.54681146e-01 -2.62562692e-01 8.57892931e-01 8.66249859e-01 8.24367553e-02 -4.37748551e-01 7.01961577e-01 -2.32841238e-01 5.29448450e-01 5.58862150e-01 -5.68801939e-01 -4.80679870e-01 3.01527172e-01 3.01447988e-01 5.54912508e-01 3.89789432e-01 -9.77581739e-01 3.20369124e-01 -4.67800021e-01 -1.72532827e-01 -4.45756525e-01 4.66158271e-01 -1.21614277e+00 1.71422869e-01 6.81437194e-01 -5.16577959e-01 -4.28134680e-01 -5.48673561e-03 1.02454603e+00 1.44596249e-01 -6.02765739e-01 9.86043274e-01 2.62729466e-01 -1.42363891e-01 1.44382805e-01 -3.93124610e-01 -3.32073808e-01 1.34938931e+00 9.17702392e-02 1.58670783e-01 -5.57413876e-01 -9.35580432e-01 3.14127535e-01 3.85439284e-02 1.98168114e-01 4.92801547e-01 -9.27204907e-01 -5.19138634e-01 1.42704561e-01 -1.07839135e-02 -2.64589965e-01 2.67492831e-01 9.08810556e-01 5.36200069e-02 5.80116451e-01 4.44021553e-01 -6.51182532e-01 -9.82918322e-01 6.50282323e-01 3.91066372e-01 -1.60095826e-01 -5.29473960e-01 6.52617931e-01 5.82559891e-02 5.30344360e-02 4.81155455e-01 -8.65385607e-02 5.64225651e-02 -4.42318231e-01 2.42627099e-01 3.01230341e-01 2.67034501e-01 -3.31200331e-01 -2.07491085e-01 4.48332965e-01 2.39472836e-01 -3.96725893e-01 1.05767012e+00 -1.73614532e-01 2.22416043e-01 5.05451500e-01 5.70783556e-01 2.27420703e-01 -1.30606282e+00 -1.38579533e-01 1.53029814e-01 -7.77591109e-01 3.09723169e-01 -7.25266337e-01 -8.47532690e-01 4.57550496e-01 7.33550549e-01 3.18176001e-01 1.04584730e+00 -4.18565184e-01 2.87571490e-01 2.83485383e-01 6.32406592e-01 -1.25519609e+00 -1.49837345e-01 2.89812356e-01 1.04718304e+00 -7.85024703e-01 -5.51388264e-02 -5.01871884e-01 -4.74584907e-01 8.85971427e-01 5.67355752e-01 -2.76691049e-01 8.68952155e-01 1.00774124e-01 -6.27850235e-01 2.94216394e-01 -7.01265991e-01 7.76954815e-02 5.70506632e-01 -9.28368513e-03 -4.97374088e-02 8.96458477e-02 -7.28116035e-01 8.18860352e-01 -8.03761464e-03 -1.13526344e-01 5.68906784e-01 8.99700224e-01 -5.42918980e-01 -9.46854353e-01 -6.50050581e-01 2.54812658e-01 -4.93838817e-01 2.06769302e-01 -1.56627297e-01 6.57254159e-01 5.78264445e-02 1.19421446e+00 -3.03570509e-01 -1.24596469e-01 4.00740862e-01 -2.80805916e-01 3.55426222e-01 -5.33849716e-01 -4.88601863e-01 3.79314035e-01 2.01636389e-01 -6.45648479e-01 -5.23133874e-02 -4.80369866e-01 -8.81452918e-01 1.08829796e-01 -7.43791282e-01 3.43138814e-01 7.26497710e-01 1.08635843e+00 7.55062848e-02 5.20874798e-01 1.06823897e+00 -4.56180960e-01 -1.12893593e+00 -5.05202174e-01 -6.92763090e-01 -1.35030910e-01 3.68982285e-01 -8.23738396e-01 -4.69654560e-01 -3.38894159e-01]

[5.338906288146973, 3.343168258666992]

a9c8a7f0-a6b8-413d-aa60-608721b2e731

robust-channel-wise-illumination-estimation

2111.05681

https://arxiv.org/abs/2111.05681v1

https://arxiv.org/pdf/2111.05681v1.pdf

Robust channel-wise illumination estimation

Recently, Convolutional Neural Networks (CNNs) have been widely used to solve the illuminant estimation problem and have often led to state-of-the-art results. Standard approaches operate directly on the input image. In this paper, we argue that this problem can be decomposed into three channel-wise independent and symmetric sub-problems and propose a novel CNN-based illumination estimation approach based on this decomposition. The proposed method substantially reduces the number of parameters needed to solve the task while achieving competitive experimental results compared to state-of-the-art methods. Furthermore, the practical application of illumination estimation techniques typically requires identifying the extreme error cases. This can be achieved using an uncertainty estimation technique. In this work, we propose a novel color constancy uncertainty estimation approach that augments the trained model with an auxiliary branch which learns to predict the error based on the feature representation. Intuitively, the model learns which feature combinations are robust and are thus likely to yield low errors and which combinations result in erroneous estimates. We test this approach on the proposed method and show that it can indeed be used to avoid several extreme error cases and, thus, improves the practicality of the proposed technique.

['Moncef Gabbouj', 'Alexandros Iosifidis', 'Jarno Nikkanen', 'Jenni Raitoharju', 'Firas Laakom']

2021-11-10

['color-constancy']

['computer-vision']

[ 2.96998739e-01 -1.94814846e-01 2.41982684e-01 -3.90238047e-01 -8.55495751e-01 -3.48826319e-01 3.70386899e-01 -1.23149976e-01 -4.86414909e-01 8.86823058e-01 -4.96640176e-01 -1.92403734e-01 -1.13442034e-01 -5.54116964e-01 -8.63228142e-01 -1.05587375e+00 3.14238727e-01 1.00953966e-01 -6.33841455e-02 2.56279856e-01 3.81091744e-01 3.86225224e-01 -1.76963472e+00 -1.23372138e-01 1.08678114e+00 1.58714747e+00 -1.12369858e-01 6.74079001e-01 8.86033475e-02 5.38929760e-01 -4.17518765e-01 -2.42379859e-01 4.44762617e-01 -5.38010836e-01 -4.90142703e-01 1.63296908e-01 6.01717591e-01 -3.26056749e-01 2.30346620e-01 1.14000511e+00 4.96582538e-01 1.93814605e-01 7.15220690e-01 -1.13885808e+00 -3.47259462e-01 4.18082736e-02 -7.08467662e-01 -1.08512662e-01 -2.39091858e-01 -7.15943798e-02 9.45554376e-01 -8.54934394e-01 2.16680989e-01 8.22526157e-01 7.28228033e-01 3.55381250e-01 -1.32677770e+00 -3.82142574e-01 -4.76815179e-02 3.32298458e-01 -1.34323883e+00 -5.13895214e-01 8.99601519e-01 -2.10301369e-01 6.72414839e-01 4.52649742e-02 2.59173393e-01 9.06893015e-01 -1.94580294e-02 7.46582568e-01 1.41443801e+00 -6.78425312e-01 4.73917931e-01 2.24261031e-01 -2.37989444e-02 7.50796199e-01 4.10079777e-01 1.72174796e-01 -2.82624781e-01 8.68237913e-02 7.32361674e-01 -3.84624660e-01 -4.42746371e-01 -4.35634345e-01 -7.90448487e-01 7.61397541e-01 6.33674562e-01 1.69680700e-01 -2.86621213e-01 5.09915948e-01 1.64787441e-01 -1.81243510e-03 6.85260713e-01 3.68053794e-01 -4.59612936e-01 -1.33999428e-02 -9.91374135e-01 8.05512816e-02 6.74936056e-01 6.14547849e-01 8.80795181e-01 1.37972282e-02 5.97112142e-02 8.32334638e-01 2.09861994e-01 3.18313301e-01 -6.68472946e-02 -8.95898104e-01 1.40929624e-01 2.92249203e-01 4.57374722e-01 -8.97901297e-01 -4.40562636e-01 -6.75288737e-01 -7.45684028e-01 3.84784997e-01 7.37272859e-01 -3.14473093e-01 -8.65778863e-01 1.76381946e+00 2.25733310e-01 2.94598192e-01 1.00010060e-01 9.25019443e-01 2.31264725e-01 4.42791253e-01 -3.39487910e-01 -1.61969915e-01 1.00320733e+00 -8.75134647e-01 -5.94759583e-01 -9.03081894e-03 2.63119936e-01 -8.85755360e-01 7.50187159e-01 7.76270509e-01 -1.00975847e+00 -5.48391223e-01 -1.20710731e+00 7.18060285e-02 -2.75472641e-01 6.19696438e-01 4.77645248e-01 8.94844770e-01 -9.49900687e-01 8.53491664e-01 -6.05182946e-01 -2.33578101e-01 4.20740753e-01 3.70039105e-01 -1.50625557e-02 -1.27261788e-01 -8.28377247e-01 1.00011694e+00 4.23414201e-01 6.45077348e-01 -7.02290237e-01 -3.15652311e-01 -7.14285433e-01 1.36985049e-01 6.78704560e-01 -6.22642398e-01 1.16128528e+00 -1.33546722e+00 -1.68063688e+00 3.38024348e-01 -1.56017512e-01 -2.89706737e-01 6.20589554e-01 -3.40944558e-01 -3.95980850e-02 1.36273548e-01 -2.99608350e-01 4.65229213e-01 1.30267918e+00 -1.62066483e+00 -6.73953056e-01 -2.59923995e-01 2.92983232e-03 -7.81026483e-02 -3.07501882e-01 -2.08536938e-01 -4.36001420e-01 -2.85466701e-01 1.25791535e-01 -9.98041809e-01 -9.50448215e-02 2.39552230e-01 -4.54086423e-01 3.24754156e-02 3.99375468e-01 -3.43807876e-01 7.58141875e-01 -2.03338122e+00 1.68745339e-01 3.88108462e-01 5.95373427e-03 1.35590389e-01 -8.07506889e-02 3.85154895e-02 -1.63843744e-02 -1.86377034e-01 -4.24381226e-01 -6.68467462e-01 7.95152932e-02 2.47036472e-01 -9.26747397e-02 8.48997772e-01 4.79993314e-01 6.70004666e-01 -7.42162645e-01 -1.76411018e-01 4.29523528e-01 7.25497723e-01 -3.53606939e-01 2.71503031e-01 -2.34439805e-01 5.43274701e-01 3.97200063e-02 4.96505260e-01 9.77968097e-01 -1.45458445e-01 9.92539153e-02 -5.73956192e-01 -3.24415356e-01 -1.60901710e-01 -1.31478941e+00 1.41714859e+00 -8.68271410e-01 8.40701461e-01 4.46781144e-02 -1.14458227e+00 8.97283375e-01 6.76538572e-02 2.61982709e-01 -4.03832346e-01 5.25707364e-01 4.61770236e-01 -5.45933545e-02 -4.39584017e-01 3.27323109e-01 -1.86993077e-01 3.60274076e-01 2.07739383e-01 6.97444305e-02 -2.50326782e-01 -6.42853603e-02 -3.12637001e-01 6.85443580e-01 4.17888582e-01 3.71148050e-01 -2.70111024e-01 7.37541676e-01 -5.36501467e-01 4.88982975e-01 7.83469200e-01 -2.30475560e-01 6.02628827e-01 5.12948871e-01 -4.31265086e-01 -8.85571778e-01 -8.57354760e-01 -3.03368300e-01 4.96022284e-01 1.72723919e-01 5.96724637e-02 -9.15220499e-01 -6.17694318e-01 -2.29079068e-01 8.01724553e-01 -7.87239790e-01 6.06731623e-02 -1.97686732e-01 -7.74879515e-01 2.93773204e-01 5.60695350e-01 7.09430575e-01 -6.39030635e-01 -8.64543438e-01 4.32692915e-02 -3.01096171e-01 -1.16671050e+00 -1.39425776e-03 4.67076749e-01 -6.75279677e-01 -1.07799125e+00 -7.51435757e-01 -2.15189353e-01 9.16659713e-01 2.43029729e-01 9.37581778e-01 6.73425421e-02 -2.95004338e-01 5.31374514e-01 -3.56245458e-01 -5.74037135e-01 -9.80409905e-02 -6.98903501e-02 -3.14698070e-02 5.13536870e-01 2.19920948e-01 -5.00462890e-01 -6.39117301e-01 2.21052498e-01 -1.02780187e+00 -1.31232068e-01 7.82816589e-01 1.11377323e+00 5.25636792e-01 2.19817981e-01 4.96454418e-01 -7.22390175e-01 2.84224570e-01 -1.95290968e-01 -1.03561771e+00 2.97685504e-01 -8.65322411e-01 4.49600607e-01 7.73089170e-01 -2.05859274e-01 -1.26242721e+00 3.11534196e-01 -3.17470953e-02 -4.84600574e-01 -2.65067965e-01 4.35004324e-01 -8.27049464e-02 -6.19913459e-01 4.88758355e-01 1.77374855e-01 -2.48531029e-01 -3.89404863e-01 3.45942020e-01 5.31438708e-01 5.72157562e-01 -7.13726103e-01 7.88841665e-01 5.20430028e-01 5.15946507e-01 -8.53112698e-01 -1.00617754e+00 -4.38397735e-01 -5.57553947e-01 -5.71932614e-01 6.36307597e-01 -6.39049649e-01 -7.75906444e-01 5.60262561e-01 -1.27764249e+00 -1.13160498e-01 2.78508198e-02 3.66680413e-01 -6.86189950e-01 5.39230883e-01 -5.56462444e-02 -1.28412724e+00 -5.57676665e-02 -1.26346493e+00 1.01016259e+00 3.38760048e-01 3.36682796e-01 -9.82974529e-01 -6.11775555e-02 -2.23833993e-02 5.64938843e-01 2.16439486e-01 7.63948321e-01 -2.25406155e-01 -7.38670409e-01 -1.20222054e-01 -6.75311267e-01 7.39702284e-01 8.52060132e-03 2.08571091e-01 -1.49772155e+00 -1.50490999e-01 1.78628340e-01 -3.38988394e-01 1.22881269e+00 7.15083122e-01 1.49081326e+00 1.01854213e-01 1.08726382e-01 6.57125831e-01 1.90649319e+00 -1.82677165e-01 8.08147907e-01 2.26429924e-01 4.53776687e-01 6.19471014e-01 4.21360523e-01 4.58814442e-01 9.49063227e-02 7.51060009e-01 9.44885433e-01 -8.14307034e-02 5.82149439e-02 1.71202525e-01 1.04749642e-01 4.15199041e-01 -2.38382161e-01 -4.24027711e-01 -4.83417839e-01 5.44416606e-01 -1.99614811e+00 -6.06360734e-01 -6.45810589e-02 2.47610354e+00 6.00241244e-01 2.32209545e-02 -2.07350865e-01 3.37880552e-01 4.63948965e-01 -6.14399388e-02 -4.72541004e-01 -4.54751909e-01 7.40589797e-02 3.11609149e-01 5.78214407e-01 3.81175458e-01 -1.16881800e+00 5.20521641e-01 5.96621466e+00 7.02841222e-01 -1.19907355e+00 -1.07845262e-01 6.84236407e-01 2.27515712e-01 -1.33386776e-01 -9.74549055e-02 -5.63244939e-01 2.59641618e-01 6.58112884e-01 2.52514958e-01 5.13711751e-01 7.40202308e-01 2.94631068e-02 -6.82786167e-01 -1.13705456e+00 1.17431462e+00 3.14236850e-01 -9.33281600e-01 -3.33822578e-01 -3.77784856e-02 9.27351236e-01 -3.63848269e-01 3.41097265e-01 -9.42509621e-02 -1.92186773e-01 -8.94452333e-01 5.70576847e-01 6.77171767e-01 5.92005193e-01 -9.51446176e-01 9.59384143e-01 1.91736788e-01 -9.33117568e-01 -2.75559127e-01 -6.02709174e-01 -1.03231035e-01 -2.23882683e-02 1.01880574e+00 -5.46371460e-01 8.58692944e-01 4.91117060e-01 4.52970743e-01 -4.81650174e-01 1.28242409e+00 -4.37508345e-01 4.41115558e-01 -4.14768845e-01 -1.45967096e-01 2.78705031e-01 -2.61969179e-01 4.37655747e-01 1.04988980e+00 5.52550912e-01 -2.48699233e-01 -2.27584705e-01 1.08838415e+00 -1.19097099e-01 4.47486416e-02 -4.26861316e-01 2.44321316e-01 -4.97984923e-02 1.44511724e+00 -9.45238531e-01 -1.21899739e-01 -5.29606104e-01 1.07421327e+00 5.09066701e-01 4.58614439e-01 -8.95821691e-01 -4.37229335e-01 4.76802528e-01 -3.35938722e-01 6.50686383e-01 -1.84987649e-01 -3.27100188e-01 -1.25984526e+00 1.35795251e-01 -5.20447731e-01 -2.46075410e-02 -8.18908989e-01 -1.29273462e+00 5.81056893e-01 -1.20023102e-01 -1.24905860e+00 -9.35604647e-02 -9.80883241e-01 -5.99933326e-01 8.39136839e-01 -2.15859771e+00 -1.03565872e+00 -5.86792886e-01 3.85561377e-01 4.00838405e-01 2.73984432e-01 8.66724074e-01 2.71682560e-01 -6.78125024e-01 5.88149250e-01 4.70129669e-01 -1.78203389e-01 8.94544780e-01 -1.38122451e+00 -7.71723315e-02 1.18710160e+00 1.21624134e-01 3.84810686e-01 8.14871609e-01 -1.44035295e-01 -1.13609886e+00 -9.93146539e-01 6.11433208e-01 -6.51259422e-02 4.26981479e-01 -1.60019249e-01 -5.55836022e-01 2.63583422e-01 1.69194728e-01 2.40237266e-01 4.86774355e-01 1.03422135e-01 -4.83051628e-01 -3.30397099e-01 -1.15497470e+00 3.60108286e-01 4.98306274e-01 -2.92845905e-01 -1.33867890e-01 1.01106331e-01 1.85816318e-01 -2.62567222e-01 -4.67800021e-01 3.54481250e-01 6.46195292e-01 -1.39717889e+00 6.99174702e-01 -1.86123461e-01 6.66763365e-01 -3.07140946e-01 -2.33466968e-01 -1.54789507e+00 7.81848207e-02 -3.59812051e-01 -2.81842530e-01 9.64530826e-01 3.28382224e-01 -7.36740470e-01 4.75149214e-01 6.16396427e-01 -9.67490897e-02 -7.58753181e-01 -1.01287127e+00 -8.42269957e-01 -1.24942318e-01 -6.09526694e-01 2.24869549e-01 4.34826016e-01 -4.03972834e-01 1.61329377e-02 -6.21775329e-01 3.34046036e-01 9.70259666e-01 1.58692360e-01 6.57428861e-01 -1.27640533e+00 -3.69797111e-01 -4.27319020e-01 -3.37238818e-01 -1.09217179e+00 2.71838486e-01 -2.61555642e-01 6.12275243e-01 -1.17322552e+00 8.68777558e-02 -4.13323969e-01 -3.23399156e-01 2.77414680e-01 -2.98560172e-01 3.81787121e-01 7.94282481e-02 -1.86196506e-01 -4.28092629e-01 4.47293639e-01 8.85171771e-01 -6.85631633e-02 1.20907612e-02 1.49952054e-01 -5.03415048e-01 6.78689003e-01 7.88730919e-01 -2.05277979e-01 -2.89135307e-01 -3.46962839e-01 6.03447258e-01 -3.55342686e-01 6.15440667e-01 -1.07998288e+00 1.59852058e-01 1.83856189e-02 5.76993287e-01 -3.77298981e-01 3.81913513e-01 -1.15586555e+00 -1.78329408e-01 1.66595981e-01 -2.25806668e-01 -4.92125809e-01 3.32250744e-01 5.77100575e-01 -6.24858849e-02 -5.74450910e-01 1.12948513e+00 -1.43002663e-02 -4.78190541e-01 -2.55778665e-03 -9.35375169e-02 -4.34219658e-01 8.03989351e-01 -4.04862761e-02 -7.92633072e-02 -5.23313582e-01 -5.01332641e-01 -2.67685860e-01 3.65174949e-01 -1.15782470e-01 4.92797792e-01 -1.10142434e+00 -4.93561894e-01 1.34676918e-01 2.21059740e-01 -1.09837770e-01 1.64643586e-01 1.04194391e+00 -3.05137247e-01 2.95826614e-01 -1.47166580e-01 -7.10520804e-01 -1.01637125e+00 5.25621176e-01 5.02128303e-01 -1.07018560e-01 -1.69744745e-01 8.92598450e-01 1.49414847e-02 -1.11879464e-02 2.50654757e-01 -4.97700244e-01 -3.56035046e-02 -1.41004607e-01 4.48283136e-01 4.54154849e-01 2.91183025e-01 -4.31707352e-01 -9.29297358e-02 7.36274779e-01 2.34058499e-01 -6.74497262e-02 1.26089728e+00 -3.46228629e-01 -1.56071261e-01 4.26949650e-01 1.31092668e+00 -2.55074557e-02 -1.46219492e+00 -2.12728083e-01 -3.10569406e-01 -7.09248364e-01 4.17067856e-01 -8.68171930e-01 -1.26692319e+00 1.17768955e+00 8.39502394e-01 1.86161250e-01 1.55675244e+00 -4.39236850e-01 5.49365044e-01 4.57676888e-01 2.43789762e-01 -1.17618322e+00 1.25588523e-02 3.13730836e-01 5.85272908e-01 -1.55347216e+00 1.16688639e-01 -4.09029037e-01 -1.89573437e-01 1.55491292e+00 4.54375774e-01 -4.74188775e-02 4.63751376e-01 8.27594772e-02 1.58609189e-02 -7.17289932e-03 -3.85249257e-01 -5.04165709e-01 4.70957607e-01 5.01352251e-01 3.64246726e-01 -6.04961812e-02 -2.69701183e-01 3.11055481e-01 4.20121372e-01 8.64101872e-02 4.96687680e-01 5.78382969e-01 -3.61094892e-01 -9.71833289e-01 -4.46213543e-01 2.91330904e-01 -2.60640591e-01 -1.57786354e-01 -2.41851091e-01 5.70617557e-01 2.76308417e-01 1.02091420e+00 -1.46197736e-01 -9.20367390e-02 6.02920773e-03 9.05104876e-02 8.70336711e-01 -1.10242337e-01 -1.76068202e-01 2.68098861e-01 -4.63368222e-02 -6.02934003e-01 -7.59776413e-01 -5.60716152e-01 -7.60072887e-01 -4.66507860e-02 -8.26084018e-01 -1.52417764e-01 1.02069366e+00 1.06428540e+00 8.30961764e-02 5.04231095e-01 8.94970059e-01 -1.10928214e+00 -7.19202876e-01 -8.48458350e-01 -5.27126908e-01 6.01336770e-02 4.53137070e-01 -9.45159256e-01 -5.43635786e-01 -6.71846569e-02]

[10.310770034790039, -2.5118136405944824]

e9eeec4c-67f4-45c2-90c7-28493bec7826

back-to-the-future-knowledge-distillation-for

1904.04868

https://arxiv.org/abs/1904.04868v2

https://arxiv.org/pdf/1904.04868v2.pdf

Knowledge Distillation for Human Action Anticipation

We consider the task of training a neural network to anticipate human actions in video. This task is challenging given the complexity of video data, the stochastic nature of the future, and the limited amount of annotated training data. In this paper, we propose a novel knowledge distillation framework that uses an action recognition network to supervise the training of an action anticipation network, guiding the latter to attend to the relevant information needed for correctly anticipating the future actions. This framework is possible thanks to a novel loss function to account for positional shifts of semantic concepts in a dynamic video. The knowledge distillation framework is a form of self-supervised learning, and it takes advantage of unlabeled data. Experimental results on JHMDB and EPIC-KITCHENS dataset show the effectiveness of our approach.

['Vinh Tran', 'Minh Hoai', 'Yang Wang']

2019-04-09

['action-anticipation']

['computer-vision']

[ 5.02159655e-01 3.14186662e-01 -2.68870592e-01 -5.07596970e-01 -1.67843997e-01 -3.01409483e-01 5.85953295e-01 -1.50745228e-01 -6.90713286e-01 7.12980390e-01 3.27352524e-01 -5.04554398e-02 -1.75925121e-01 -5.78931212e-01 -9.38530684e-01 -5.13746500e-01 -3.45209718e-01 3.67106318e-01 6.24008358e-01 -8.73402283e-02 1.11634083e-01 3.35378051e-01 -1.54910696e+00 5.35590708e-01 4.36691582e-01 1.09833252e+00 4.79367614e-01 8.40613782e-01 2.86748022e-01 1.38503909e+00 -5.28551675e-02 -2.21402813e-02 4.38874006e-01 -5.73362708e-01 -1.03026438e+00 4.02410030e-01 1.00408107e-01 -5.01583636e-01 -6.26992285e-01 7.87374794e-01 1.46553636e-01 6.54066384e-01 3.91993254e-01 -1.15776527e+00 -6.83444068e-02 5.29841840e-01 4.14956771e-02 5.33349931e-01 4.82927710e-01 3.16197664e-01 7.91179836e-01 -4.57835257e-01 8.57952774e-01 1.07325947e+00 4.77816492e-01 7.92696059e-01 -6.49276435e-01 -1.64423078e-01 5.42590201e-01 8.50950599e-01 -1.10754561e+00 -4.67206359e-01 8.91003668e-01 -5.76797605e-01 9.28210735e-01 -1.84013680e-01 9.39299166e-01 1.34774411e+00 -5.51577210e-02 1.23515093e+00 6.77446544e-01 -5.36434710e-01 6.09927893e-01 -1.53144389e-01 -2.46216089e-01 7.19303548e-01 -2.22151458e-01 3.59642208e-01 -6.73304558e-01 2.94566423e-01 6.46742165e-01 1.27087161e-01 -2.31629953e-01 -7.10641682e-01 -1.04823375e+00 5.38257718e-01 1.96421012e-01 2.46839032e-01 -6.88031495e-01 2.80447125e-01 5.64717829e-01 2.66212761e-01 3.97605181e-01 3.03943932e-01 -7.04728246e-01 -4.85826790e-01 -9.14386749e-01 1.90207481e-01 7.80265272e-01 7.15215743e-01 5.33971727e-01 -1.40447527e-01 -2.04949409e-01 3.37120503e-01 2.16041565e-01 1.28646016e-01 7.31275320e-01 -1.32814693e+00 4.98921990e-01 5.54731309e-01 1.62871793e-01 -9.21836138e-01 -2.62441874e-01 1.60844937e-01 -3.17401350e-01 2.42557883e-01 3.89100671e-01 -2.98156440e-01 -8.98008704e-01 1.64244783e+00 4.21804845e-01 6.97768271e-01 2.71829218e-01 8.17073822e-01 1.58688232e-01 5.13900876e-01 2.29767367e-01 -2.86627740e-01 7.93527186e-01 -9.89517510e-01 -5.83711624e-01 -3.44623268e-01 6.75021887e-01 -5.27349636e-02 5.52486360e-01 4.10470843e-01 -9.86931562e-01 -8.23799551e-01 -7.28864074e-01 2.42235959e-01 -3.26707602e-01 -9.84040555e-03 5.39088964e-01 1.03412136e-01 -9.86153841e-01 9.52136934e-01 -1.22200072e+00 -5.80470562e-01 6.34066224e-01 3.26492459e-01 -3.97344649e-01 -1.17059506e-01 -1.24297345e+00 6.36609077e-01 1.17177498e+00 3.00707489e-01 -1.12698662e+00 -3.48830849e-01 -8.90572011e-01 3.97523008e-02 8.89432013e-01 -4.77846652e-01 1.38087535e+00 -1.47833490e+00 -1.66920078e+00 5.95556200e-01 1.95600718e-01 -9.94538903e-01 7.42995620e-01 -4.76141930e-01 -2.39714041e-01 6.06330037e-01 -9.62462500e-02 7.55249143e-01 8.42986703e-01 -7.57754922e-01 -9.66025293e-01 -2.88750619e-01 2.85534471e-01 4.51578289e-01 -2.42741853e-01 -2.24573746e-01 -5.27888417e-01 -4.39408839e-01 -2.48049319e-01 -1.14304388e+00 -4.63798851e-01 -1.70879036e-01 -1.70219690e-01 -2.43408591e-01 9.81747985e-01 -5.77242255e-01 1.06827986e+00 -2.20160413e+00 3.51238042e-01 -4.60595861e-02 -2.56524175e-01 4.66365606e-01 -1.20647259e-01 3.89782310e-01 -2.40915015e-01 -4.46472555e-01 -1.07573718e-01 -9.57790911e-02 -2.65374899e-01 5.67466021e-01 -3.47622216e-01 2.58178025e-01 3.40782583e-01 8.50024343e-01 -1.18572176e+00 -4.83492345e-01 2.90507317e-01 3.36616665e-01 -5.58156550e-01 5.00630856e-01 -7.06111670e-01 7.26043820e-01 -7.27174640e-01 4.53281909e-01 2.82338429e-02 -1.02713205e-01 4.06074911e-01 -2.12526321e-02 -1.12320045e-02 -2.14762110e-02 -1.10091102e+00 1.93831265e+00 -2.78269589e-01 6.47018909e-01 -4.29794401e-01 -1.42023122e+00 5.74165285e-01 4.28924948e-01 7.04632282e-01 -4.26088333e-01 9.72666964e-02 -4.81912924e-04 -1.52633503e-01 -1.17209280e+00 2.76182145e-01 -9.90098342e-02 7.45401606e-02 2.73497999e-01 1.82245657e-01 1.81289822e-01 5.74523330e-01 2.40680382e-01 1.24971497e+00 7.89880633e-01 2.85781235e-01 2.10440204e-01 5.90521276e-01 2.90286131e-02 8.17580700e-01 7.42721677e-01 -5.08970141e-01 3.33199322e-01 4.95895088e-01 -9.18097556e-01 -7.49123931e-01 -5.48263431e-01 5.36186159e-01 1.02675450e+00 3.81211808e-04 -1.81984574e-01 -7.34633923e-01 -1.13881898e+00 -3.03790063e-01 7.93568552e-01 -7.09188998e-01 -5.02722204e-01 -7.73442864e-01 -3.31637293e-01 8.61868411e-02 8.36408198e-01 6.21118248e-01 -1.56421626e+00 -1.19931149e+00 3.52379024e-01 -2.81611085e-01 -1.42581356e+00 -2.01264098e-01 8.42033476e-02 -9.05620635e-01 -1.40021789e+00 -3.91743392e-01 -6.94455445e-01 5.55906117e-01 -1.58274382e-01 9.25850809e-01 -1.19383641e-01 -2.68737614e-01 9.27753925e-01 -5.30555189e-01 -2.51970321e-01 -5.08262634e-01 -1.94316566e-01 1.69368505e-01 4.34847504e-01 4.28962946e-01 -4.67516154e-01 -7.36081898e-01 5.51380105e-02 -8.59903991e-01 8.95843357e-02 5.17196298e-01 8.68078351e-01 4.66335654e-01 4.05337930e-01 2.97843635e-01 -7.85333335e-01 2.39645541e-01 -4.54728603e-01 -4.86776471e-01 3.80454481e-01 -3.89787406e-01 3.27475220e-02 5.64346313e-01 -5.35045326e-01 -1.46069360e+00 6.86321259e-01 2.72053927e-01 -6.18966699e-01 -3.85676473e-01 4.79581326e-01 -1.39286712e-01 2.55816996e-01 3.23285818e-01 2.98586488e-01 -1.13153353e-01 -3.09811294e-01 3.30842942e-01 3.54501635e-01 7.00205266e-01 -4.39293832e-01 3.13649625e-01 6.37238681e-01 2.97584776e-02 -5.59019983e-01 -1.04471099e+00 -4.59546179e-01 -1.08552909e+00 -6.11778259e-01 9.26050365e-01 -9.07798409e-01 -7.47355580e-01 4.34747577e-01 -1.07875085e+00 -7.63798952e-01 -7.65575945e-01 7.40087330e-01 -1.10824251e+00 4.68267292e-01 -3.24020207e-01 -8.79056513e-01 1.44123256e-01 -9.19752777e-01 7.91394055e-01 2.01931119e-01 -1.35815531e-01 -1.16235435e+00 2.84545332e-01 4.85156268e-01 -6.88336268e-02 3.50742757e-01 4.38454479e-01 -8.96682084e-01 -7.34622717e-01 -4.14381593e-01 1.42604351e-01 6.35761559e-01 -4.13360521e-02 -1.69431835e-01 -6.55467391e-01 -8.32765996e-02 6.11161776e-02 -6.70208037e-01 1.05214417e+00 4.57763582e-01 1.31530821e+00 -3.88205051e-01 -3.66994649e-01 2.80638605e-01 1.20535684e+00 3.79114836e-01 6.24272466e-01 3.16118598e-01 5.59953630e-01 7.74765909e-01 9.95725989e-01 6.83760524e-01 4.32046264e-01 5.67599595e-01 6.22825503e-01 3.21226984e-01 -4.78468277e-02 -3.57262075e-01 4.89610791e-01 4.16589797e-01 -3.33056509e-01 -3.06855977e-01 -8.56595814e-01 7.08446383e-01 -2.42649961e+00 -1.35721803e+00 3.36810082e-01 2.00643706e+00 6.69492126e-01 1.50435135e-01 1.17768869e-01 4.55865338e-02 5.37260115e-01 1.86379135e-01 -6.91624761e-01 -2.01503783e-01 1.87170133e-01 -1.83171272e-01 3.23577493e-01 2.72437692e-01 -1.49841022e+00 1.11276221e+00 6.39764118e+00 4.68613446e-01 -8.86364460e-01 -1.20246120e-01 5.69833457e-01 -1.45335361e-01 5.67230403e-01 2.62276251e-02 -5.99707663e-01 5.08060694e-01 9.88592386e-01 9.86801460e-02 3.68120730e-01 8.57572377e-01 4.43674713e-01 -3.81596684e-01 -1.35110915e+00 6.78300381e-01 2.49001071e-01 -1.17199337e+00 -1.43916056e-01 -3.75023693e-01 6.43524289e-01 1.12009324e-01 -1.37383193e-01 3.65864158e-01 3.64689708e-01 -6.00537658e-01 5.72226942e-01 8.71062934e-01 3.13260794e-01 -5.84636390e-01 5.01064062e-01 5.54121792e-01 -8.75810146e-01 -5.97271919e-01 -1.18618615e-01 -3.97666305e-01 3.77722681e-01 2.38111362e-01 -1.11474979e+00 3.39831710e-01 6.66089177e-01 1.09965634e+00 -3.40261698e-01 8.74950469e-01 -5.50347984e-01 5.30790150e-01 -8.67813155e-02 2.04886571e-01 3.33481252e-01 -4.00816463e-02 4.92362112e-01 9.87265348e-01 1.86704025e-01 2.64931470e-01 5.34063101e-01 2.98601925e-01 1.61046833e-01 -2.30212852e-01 -7.16259956e-01 -1.68329805e-01 -6.50459826e-02 9.37378526e-01 -7.20346212e-01 -6.09023869e-01 -3.19609255e-01 1.28299260e+00 3.63904208e-01 4.23503608e-01 -6.31677806e-01 -1.76577885e-02 2.22933829e-01 -1.40347511e-01 7.16483176e-01 -5.97825311e-02 4.33273941e-01 -1.18475735e+00 1.24637291e-01 -7.05203414e-01 6.32086217e-01 -7.90934741e-01 -7.54395187e-01 4.42062855e-01 1.04106233e-01 -1.25175309e+00 -7.88309872e-01 -6.31205618e-01 -5.62203109e-01 1.79681689e-01 -1.60129344e+00 -1.01365495e+00 -9.60869119e-02 6.98714077e-01 8.06532979e-01 -2.60763496e-01 5.69206893e-01 4.26568501e-02 -5.04996121e-01 -6.06660917e-02 -8.55228826e-02 1.74169570e-01 4.27918404e-01 -1.16516376e+00 6.94149435e-02 9.13823664e-01 1.91074625e-01 1.86069235e-02 7.56568551e-01 -8.18247259e-01 -1.11601901e+00 -1.17418063e+00 8.81715417e-01 -5.28533638e-01 7.44285941e-01 -4.70188111e-02 -8.46925676e-01 9.51184571e-01 2.66920477e-02 3.70131522e-01 6.79869473e-01 -2.14724943e-01 -1.01410344e-01 -1.23424172e-01 -8.93427610e-01 4.62220162e-01 1.17607832e+00 -4.02418554e-01 -8.91729951e-01 6.60802901e-01 6.26323581e-01 -2.89119363e-01 -5.33453107e-01 4.40152824e-01 5.34497678e-01 -1.03868651e+00 9.22939301e-01 -1.05673862e+00 4.72317964e-01 -7.68960044e-02 2.50883922e-02 -1.23477590e+00 -1.26998931e-01 -6.59224689e-01 -5.31400740e-01 7.49726832e-01 6.07888065e-02 -1.23615554e-02 1.00532901e+00 7.88441777e-01 2.72866432e-02 -6.22171342e-01 -9.50138032e-01 -9.15839136e-01 -5.20748317e-01 -5.20570457e-01 1.54573530e-01 7.10099280e-01 2.61595964e-01 3.55950519e-02 -7.52493680e-01 -3.41568477e-02 4.38592583e-01 -5.70674874e-02 5.11731923e-01 -1.00101650e+00 -6.06139779e-01 1.02687761e-01 -7.58321285e-01 -1.08177137e+00 4.86468345e-01 -5.20323932e-01 2.99413234e-01 -1.30485940e+00 1.01882093e-01 5.33218309e-02 -3.86202216e-01 5.79833627e-01 -3.96804065e-02 -8.39392245e-02 2.91014045e-01 2.84613539e-02 -1.32045460e+00 6.96915030e-01 9.90407407e-01 -3.28958482e-02 -3.79423022e-01 2.54460126e-01 -4.70038764e-02 1.04177880e+00 7.06500888e-01 -3.10617894e-01 -7.47357130e-01 -4.47648048e-01 2.83674002e-02 1.57268062e-01 5.65867305e-01 -1.28924882e+00 4.88560855e-01 -2.48054653e-01 3.22271466e-01 -4.43645120e-01 4.75857437e-01 -1.10281348e+00 -9.85961258e-02 7.13521540e-01 -6.58698380e-01 -2.42107227e-01 1.53080681e-02 1.06633604e+00 -3.05291146e-01 -2.20270991e-01 4.38429922e-01 -3.90326291e-01 -1.40484750e+00 4.15819615e-01 -4.37943786e-01 1.35926068e-01 1.37145364e+00 -3.43301296e-01 2.95321316e-01 -2.94672877e-01 -1.05164981e+00 3.27876925e-01 5.66972159e-02 5.95420241e-01 6.49701416e-01 -1.14155900e+00 -3.87153059e-01 1.44645050e-01 1.09978445e-01 -5.94208203e-02 4.16823417e-01 6.45969212e-01 -2.27085158e-01 3.00522625e-01 -2.94452816e-01 -4.06905740e-01 -1.05919147e+00 8.61926973e-01 2.76644588e-01 -3.53810221e-01 -7.35143483e-01 7.63605237e-01 1.42014846e-01 -1.12223022e-01 4.03543353e-01 -1.92357957e-01 -5.17954171e-01 -4.10531601e-03 6.65936947e-01 4.79837388e-01 -2.39270121e-01 -5.69575787e-01 -2.72801578e-01 1.47810146e-01 -8.30322802e-02 -4.59551997e-02 1.50357246e+00 -2.86761492e-01 2.20152855e-01 4.86918658e-01 1.02174938e+00 -7.55778193e-01 -1.99980962e+00 -4.31674212e-01 3.84005755e-01 -3.99591297e-01 -7.81072304e-02 -8.35137308e-01 -1.02240610e+00 7.11276054e-01 4.52314556e-01 -4.86588925e-02 1.27185404e+00 -1.09596230e-01 6.07412517e-01 6.81679904e-01 2.66096443e-01 -1.66594613e+00 5.09770095e-01 6.06485546e-01 6.19837463e-01 -1.36060262e+00 -1.11424714e-01 -9.28703249e-02 -9.44169343e-01 1.30547237e+00 8.06943297e-01 4.93431315e-02 7.14698195e-01 -2.00305969e-01 -8.92224833e-02 -1.97649807e-01 -1.01306176e+00 -2.97875881e-01 1.38246939e-01 5.75072885e-01 -1.40132502e-01 -2.55790889e-01 -9.57728326e-02 3.16812754e-01 4.59745973e-01 6.99739933e-01 5.04725516e-01 1.18823862e+00 -4.49865192e-01 -8.96568537e-01 2.77036987e-03 1.59390196e-01 -2.73578048e-01 2.75230348e-01 -2.02126965e-01 5.38398147e-01 2.71801889e-01 7.52056479e-01 8.90811831e-02 -2.85177082e-01 3.52356762e-01 3.37517589e-01 4.41360652e-01 -6.16521776e-01 -9.57221538e-02 -9.33565944e-02 1.11328632e-01 -9.71864045e-01 -9.36022401e-01 -9.58053291e-01 -1.20623469e+00 3.68151814e-01 6.70003518e-02 -7.17398301e-02 3.55404019e-01 1.30877244e+00 4.83575881e-01 6.34640396e-01 7.55952656e-01 -7.16314614e-01 -6.10642552e-01 -5.92434525e-01 -4.90849674e-01 6.28400028e-01 2.04937801e-01 -6.86525762e-01 -1.16179973e-01 6.05830133e-01]

[8.395657539367676, 0.5376204252243042]

4dcc1a41-9e82-4a7e-89d0-c5adfb07b273

learning-symmetry-consistent-deep-cnns-for

1812.07741

http://arxiv.org/abs/1812.07741v1

http://arxiv.org/pdf/1812.07741v1.pdf

Learning Symmetry Consistent Deep CNNs for Face Completion

Deep convolutional networks (CNNs) have achieved great success in face completion to generate plausible facial structures. These methods, however, are limited in maintaining global consistency among face components and recovering fine facial details. On the other hand, reflectional symmetry is a prominent property of face image and benefits face recognition and consistency modeling, yet remaining uninvestigated in deep face completion. In this work, we leverage two kinds of symmetry-enforcing subnets to form a symmetry-consistent CNN model (i.e., SymmFCNet) for effective face completion. For missing pixels on only one of the half-faces, an illumination-reweighted warping subnet is developed to guide the warping and illumination reweighting of the other half-face. As for missing pixels on both of half-faces, we present a generative reconstruction subnet together with a perceptual symmetry loss to enforce symmetry consistency of recovered structures. The SymmFCNet is constructed by stacking generative reconstruction subnet upon illumination-reweighted warping subnet, and can be end-to-end learned from training set of unaligned face images. Experiments show that SymmFCNet can generate high quality results on images with synthetic and real occlusion, and performs favorably against state-of-the-arts.

['WangMeng Zuo', 'Guosheng Hu', 'Meng Wang', 'Lei Zhang', 'Jieru Zhu', 'Xiaoming Li', 'Ming Liu']

2018-12-19

['facial-inpainting']

['computer-vision']

[ 1.66456893e-01 3.80959570e-01 9.64195952e-02 -5.66599667e-01 -3.31197023e-01 -2.47489899e-01 5.50212502e-01 -1.03802443e+00 3.36816519e-01 5.04832208e-01 5.27121842e-01 1.83736056e-01 -3.74624468e-02 -8.03127348e-01 -9.53685999e-01 -7.32765615e-01 3.51784497e-01 4.90790419e-02 -4.88079607e-01 -1.86526865e-01 -1.96828976e-01 8.10093582e-01 -1.53140616e+00 4.27283287e-01 4.96672720e-01 9.61116016e-01 -1.92004681e-01 -1.17624849e-02 3.60744521e-02 4.52871740e-01 -1.52291805e-01 -6.06429994e-01 5.52971959e-01 -5.29959381e-01 -4.86469150e-01 4.71766442e-01 9.18457747e-01 -6.81302249e-01 -6.98594630e-01 8.80913734e-01 4.70875174e-01 2.14941930e-02 5.99651933e-01 -1.37322927e+00 -9.58992302e-01 1.85423091e-01 -8.45872462e-01 -4.25638288e-01 2.65278071e-01 1.56398356e-01 6.68892205e-01 -1.44925344e+00 8.51274252e-01 1.69601715e+00 6.75441206e-01 8.35830271e-01 -1.37723696e+00 -1.11786473e+00 8.99113268e-02 -1.71146646e-01 -1.42486477e+00 -9.80369031e-01 1.12035441e+00 -1.76472887e-01 4.88422424e-01 4.83710272e-03 4.85333741e-01 1.08997607e+00 2.07923323e-01 3.47529382e-01 8.37558150e-01 -9.11533907e-02 -8.56951997e-02 -4.20609802e-01 -5.18189549e-01 8.81147742e-01 2.33167931e-01 1.98849604e-01 -6.27725184e-01 -7.95197636e-02 1.26133168e+00 4.26661044e-01 -4.63052362e-01 -3.69869053e-01 -7.70768881e-01 6.83734417e-01 6.79425240e-01 -1.20149828e-01 -5.64570844e-01 8.83753225e-02 -2.20838469e-02 1.75064206e-01 5.80778003e-01 -4.60545458e-02 -6.63182139e-02 5.74877858e-01 -1.06409979e+00 3.84520501e-01 3.76782566e-01 9.38737452e-01 1.03943515e+00 4.28189129e-01 -3.25814605e-01 1.09822345e+00 4.46816266e-01 4.05042738e-01 3.04385480e-02 -1.13531637e+00 3.17164689e-01 6.34551883e-01 -2.31987759e-01 -1.06587756e+00 -1.10233769e-01 -3.80937010e-01 -1.28507042e+00 4.34027851e-01 1.40357256e-01 1.02763707e-02 -1.17120004e+00 2.06409550e+00 4.63640004e-01 5.21725476e-01 -9.43522081e-02 8.29014242e-01 1.04045200e+00 5.74785531e-01 -1.63469017e-01 -2.25361466e-01 1.32667935e+00 -8.86853933e-01 -6.61609948e-01 -1.21833853e-01 -1.18294246e-01 -9.72645879e-01 7.26932466e-01 1.80964515e-01 -1.42911386e+00 -7.68114448e-01 -8.92664731e-01 -2.89992750e-01 2.94222653e-01 6.31872788e-02 4.54377025e-01 2.08339289e-01 -1.23498797e+00 5.94136775e-01 -7.26448417e-01 -1.10692522e-02 9.11877334e-01 3.28940600e-01 -8.58698547e-01 -6.27848268e-01 -6.91101432e-01 5.20647466e-01 -1.56880572e-01 4.17995572e-01 -1.27161002e+00 -1.04844749e+00 -1.16962039e+00 1.90597311e-01 -1.70709603e-02 -8.14694285e-01 8.41982067e-01 -1.00920427e+00 -1.40419292e+00 8.16146970e-01 -3.70465666e-01 2.08428130e-01 4.47344601e-01 1.81225777e-01 -4.34875250e-01 -1.33919027e-02 6.21299446e-02 1.04958045e+00 1.33701777e+00 -1.55948937e+00 -7.19697997e-02 -5.19935131e-01 -2.84771740e-01 -4.88807596e-02 -3.66581053e-01 3.63399610e-02 -6.18262112e-01 -9.25379753e-01 5.27558029e-01 -8.64741504e-01 1.30356982e-01 5.24855554e-01 -4.71104831e-01 2.81443819e-02 1.24401462e+00 -9.43653345e-01 7.99162269e-01 -2.11831522e+00 1.65603667e-01 2.77806222e-01 3.05041343e-01 3.08670342e-01 -7.99247742e-01 2.59826273e-01 -6.27967834e-01 -7.48303905e-02 -2.91060060e-01 -8.95987630e-01 -1.28963575e-01 3.02463800e-01 -5.49467087e-01 5.72521627e-01 7.37698853e-01 9.92752969e-01 -5.45813680e-01 -1.16259627e-01 1.94321528e-01 1.26850021e+00 -9.47527111e-01 3.08818430e-01 -3.74966376e-02 6.25364780e-01 -1.66809633e-01 8.56114984e-01 1.29843247e+00 1.26377977e-02 2.36567020e-01 -5.39320350e-01 1.71913877e-01 8.36705640e-02 -1.08413470e+00 1.77395034e+00 -5.15030086e-01 3.13910365e-01 3.25707257e-01 -6.44836962e-01 9.92713928e-01 2.83885092e-01 3.31402123e-01 -7.01402307e-01 1.33429319e-01 8.84334370e-02 -9.80439633e-02 -2.52554238e-01 9.99920443e-02 -2.19325155e-01 6.52032137e-01 4.59883392e-01 3.22004050e-01 -7.50838965e-02 -8.94194841e-02 -7.41461366e-02 6.66413307e-01 3.85295361e-01 -1.99429572e-01 -2.47931406e-01 5.18810034e-01 -8.27126324e-01 7.81615734e-01 -9.35702547e-02 1.36903033e-01 1.21188760e+00 3.44390690e-01 -7.67571688e-01 -1.14784145e+00 -1.10939515e+00 -1.66568190e-01 7.45714009e-01 -1.76238552e-01 -3.69069815e-01 -9.25931394e-01 -3.95028532e-01 -1.94446817e-02 2.03294501e-01 -7.85146117e-01 -2.03080580e-01 -7.54003584e-01 -4.90981877e-01 4.07654583e-01 5.72414577e-01 7.92971969e-01 -1.19137216e+00 1.31490692e-01 8.65914226e-02 -2.61644810e-01 -1.05593240e+00 -7.38435864e-01 -5.66123366e-01 -7.06908464e-01 -1.05204237e+00 -7.88039029e-01 -9.18046951e-01 1.27118278e+00 5.40706575e-01 1.03334689e+00 4.19731230e-01 -4.40106511e-01 7.33645400e-03 1.49213508e-01 -6.93124831e-02 -2.30489388e-01 -5.33178687e-01 9.93092135e-02 4.96873528e-01 -2.09894031e-01 -1.07331288e+00 -8.39537382e-01 5.64569712e-01 -1.06777596e+00 1.64270565e-01 3.94017160e-01 9.49088156e-01 6.67923152e-01 4.00458165e-02 3.09888631e-01 -6.13832176e-01 2.02744946e-01 -2.60119438e-01 -3.62068206e-01 1.13667868e-01 -2.14611307e-01 2.08854163e-03 5.40048182e-01 -4.20208424e-01 -1.24664629e+00 7.93573782e-02 -3.54535908e-01 -9.57040608e-01 1.19255900e-01 -3.40913758e-02 -7.36995101e-01 -3.44358534e-01 3.31368834e-01 1.47121251e-01 4.66472864e-01 -4.22446430e-01 1.98234990e-01 7.91032314e-02 6.80466831e-01 -6.04184806e-01 1.13504744e+00 7.56219149e-01 2.31059477e-01 -7.17366457e-01 -7.21303046e-01 2.17378810e-01 -4.22063023e-01 -1.36163905e-01 6.08324409e-01 -1.13163054e+00 -6.38481140e-01 6.51831508e-01 -1.25816977e+00 -1.17935970e-01 -1.89725742e-01 6.74017370e-02 -3.82746547e-01 1.70621872e-01 -4.53466475e-01 -4.24914598e-01 -4.58474547e-01 -1.16302812e+00 1.27474368e+00 3.56858313e-01 3.87817770e-02 -6.75725222e-01 -2.06051588e-01 4.66846794e-01 5.45003474e-01 4.68147844e-01 7.79052496e-01 1.57620072e-01 -5.18846095e-01 -6.17121421e-02 -3.87871623e-01 5.73430598e-01 3.33961517e-01 2.01054692e-01 -1.32300901e+00 -6.04048669e-01 1.30629251e-02 -3.44796777e-01 9.89367604e-01 3.60578865e-01 1.39870560e+00 -5.76616764e-01 -1.22296222e-01 1.08483958e+00 1.04432154e+00 -1.87575907e-01 1.18418348e+00 -3.42058867e-01 8.83671820e-01 8.12017798e-01 1.12284109e-01 3.91859084e-01 2.68354654e-01 6.22124910e-01 6.40619278e-01 -3.86444718e-01 -7.28512347e-01 -5.73598206e-01 3.64877999e-01 2.99622566e-01 -3.46446663e-01 -5.98098105e-03 -4.13394451e-01 1.72177851e-01 -1.48541081e+00 -1.05327094e+00 3.04787427e-01 2.09450507e+00 7.44879663e-01 -3.29199195e-01 -1.64666861e-01 -2.98798140e-02 8.23381007e-01 3.26940507e-01 -5.83952785e-01 1.40216634e-01 -1.85928404e-01 5.52369714e-01 -1.91691801e-01 6.13370121e-01 -6.86845779e-01 9.24654901e-01 5.36524296e+00 8.41858923e-01 -1.18811476e+00 8.76205713e-02 9.09965932e-01 -3.20799537e-02 -5.56418955e-01 8.53079278e-03 -8.29715073e-01 2.53454208e-01 1.85270265e-01 2.75762141e-01 6.32298350e-01 6.06741011e-01 2.75830030e-01 4.89569485e-01 -1.06016660e+00 1.17641521e+00 3.60202610e-01 -1.44395983e+00 5.18028259e-01 2.51022875e-01 1.07893968e+00 -4.81839657e-01 2.87027746e-01 7.78875500e-02 1.76345333e-01 -1.52453911e+00 6.07904553e-01 4.89208281e-01 1.25914383e+00 -1.06195068e+00 3.56250465e-01 -6.01430386e-02 -1.26114631e+00 4.56099436e-02 -5.41981339e-01 1.35679036e-01 9.04327407e-02 5.30473948e-01 -4.01352406e-01 4.83855873e-01 5.90332747e-01 6.55147433e-01 -2.56212324e-01 5.08690298e-01 -4.61478114e-01 2.81093776e-01 -1.29339710e-01 1.00174558e+00 5.60897216e-02 -3.43630224e-01 2.70752400e-01 6.87990069e-01 4.52115953e-01 2.69052297e-01 -5.24113066e-02 1.33658159e+00 -6.71421230e-01 -3.18204343e-01 -5.94186246e-01 1.87865824e-01 4.58937794e-01 1.70355201e+00 -4.04969305e-01 -9.01051983e-03 -3.17861885e-01 9.06040728e-01 3.48021477e-01 5.04175663e-01 -6.86826766e-01 9.30457562e-02 1.16527998e+00 5.40973485e-01 2.47537225e-01 -6.89717829e-02 -7.85925367e-04 -1.07364368e+00 1.77154809e-01 -8.28116894e-01 4.00519110e-02 -7.54725993e-01 -1.40753853e+00 7.40492821e-01 -2.48750299e-01 -1.08152235e+00 -1.34763464e-01 -5.11367679e-01 -1.01170754e+00 1.10402560e+00 -1.71020019e+00 -1.74006951e+00 -7.08718181e-01 9.92562115e-01 3.93784612e-01 -2.40650117e-01 6.99731112e-01 3.09239715e-01 -6.54433191e-01 8.67623925e-01 -4.08035338e-01 3.05288851e-01 7.43696153e-01 -4.05381233e-01 4.69200879e-01 9.53772306e-01 -6.49673864e-02 9.77431059e-01 3.55834544e-01 -6.09338403e-01 -1.61258101e+00 -1.68008673e+00 7.17465341e-01 -3.91117930e-02 -5.48603870e-02 -4.76045936e-01 -9.73934472e-01 6.88785076e-01 1.59414753e-01 6.03439867e-01 4.86360759e-01 -2.42866516e-01 -9.02418673e-01 -2.94009089e-01 -1.35068119e+00 7.27749825e-01 1.40036952e+00 -4.68583196e-01 -1.97534427e-01 2.30132446e-01 4.56810564e-01 -3.91455263e-01 -6.69244766e-01 6.56459510e-01 7.33038008e-01 -1.08038640e+00 1.18319404e+00 -4.80229050e-01 8.54735374e-01 -3.79124075e-01 -6.80587888e-02 -1.20873106e+00 -4.40529317e-01 -8.83364379e-01 1.04921408e-01 1.45059550e+00 2.34433841e-02 -5.61053157e-01 8.67628276e-01 5.39646506e-01 -3.46244901e-01 -7.55203426e-01 -8.35431099e-01 -6.59257531e-01 1.01379834e-01 -1.07014284e-01 9.45836902e-01 9.95096862e-01 -5.62599063e-01 9.75685120e-02 -5.29809117e-01 1.01868279e-01 7.81222284e-01 4.48666066e-02 9.48186159e-01 -1.01195157e+00 8.18676432e-04 -3.26635569e-01 -1.74006909e-01 -8.36427331e-01 5.46947122e-01 -8.61309469e-01 -1.40044749e-01 -1.27252078e+00 3.15030545e-01 -2.21950799e-01 1.57683089e-01 8.34449828e-01 8.35791305e-02 9.18009996e-01 8.80643502e-02 1.12327851e-01 1.59083873e-01 9.37077403e-01 1.75587833e+00 3.70348841e-02 -4.12826352e-02 -2.78193116e-01 -9.11194623e-01 8.34704041e-01 4.15381879e-01 -2.24835351e-01 -4.80677575e-01 -4.93046552e-01 -8.52659345e-03 2.29189228e-02 7.03401387e-01 -8.15145791e-01 6.27764389e-02 -1.95333049e-01 9.81697619e-01 -3.75721037e-01 5.63173234e-01 -7.52323806e-01 6.32999837e-01 1.62028879e-01 -4.72807698e-02 -5.16228229e-02 1.21293284e-01 2.44040981e-01 -2.70755231e-01 3.11986417e-01 1.14953327e+00 5.56830224e-03 -2.56875992e-01 1.02462363e+00 3.88127148e-01 -7.66003355e-02 7.79912055e-01 -2.85020262e-01 -2.14280710e-01 -3.28434616e-01 -3.99882138e-01 -5.19238487e-02 5.62630236e-01 6.35532796e-01 9.89724576e-01 -1.74905682e+00 -1.08705699e+00 9.69832182e-01 -1.29322886e-01 4.07368153e-01 5.11510432e-01 5.22179902e-01 -4.45394486e-01 -2.42263023e-02 -6.65574789e-01 -4.21051383e-01 -1.26102889e+00 3.01649183e-01 3.70864362e-01 8.29166919e-02 -6.59583926e-01 9.18966830e-01 7.98514545e-01 -4.37519759e-01 1.14989594e-01 -3.65621573e-03 3.12176123e-02 -6.54325038e-02 8.36277544e-01 6.34686276e-02 2.25719750e-01 -8.74737859e-01 -2.07641244e-01 8.82854044e-01 -1.86613113e-01 -4.81280461e-02 1.65313816e+00 1.54910356e-01 -4.70734686e-01 -6.02735460e-01 1.27464080e+00 -2.21806616e-02 -1.77337313e+00 -3.14207047e-01 -8.87540460e-01 -7.43928313e-01 -2.75381282e-02 -4.12610799e-01 -1.81210637e+00 8.88559163e-01 2.37042293e-01 -6.11557364e-01 1.38070643e+00 -2.46195823e-01 7.68613219e-01 -2.65478268e-02 3.00332904e-01 -5.08673012e-01 4.21642184e-01 3.89257967e-01 1.56784034e+00 -1.05993974e+00 -4.84921671e-02 -7.27828026e-01 -3.64672154e-01 1.17548203e+00 7.84737051e-01 -3.55842471e-01 8.19154441e-01 3.03750038e-01 -2.71444559e-01 -2.23182693e-01 -5.90479732e-01 2.53288805e-01 4.86290336e-01 6.71407819e-01 4.38818544e-01 -1.49646685e-01 1.98082030e-01 5.01045406e-01 -2.94547498e-01 -8.02273527e-02 1.04586184e-01 6.80970967e-01 -2.60461532e-02 -9.86668587e-01 -5.33382535e-01 2.25802243e-01 -2.64173716e-01 -1.83415115e-01 -1.49450719e-01 6.71291709e-01 3.68669361e-01 8.29616666e-01 2.52230227e-01 -3.28714937e-01 3.39318007e-01 -4.09850068e-02 6.61696315e-01 -6.14544630e-01 -3.30779612e-01 1.91336438e-01 -2.24466681e-01 -7.93573916e-01 -1.16533637e-01 -4.92617607e-01 -1.11456180e+00 -7.09685206e-01 3.52189578e-02 -4.28784788e-01 5.01144171e-01 6.96619391e-01 6.02264762e-01 3.69451463e-01 9.28780556e-01 -1.34479928e+00 -3.95625859e-01 -9.42302108e-01 -5.57940423e-01 5.39171040e-01 3.46625537e-01 -8.12694430e-01 -3.33952904e-01 1.94150016e-01]

[12.776800155639648, -0.05653838440775871]

4c4f0f12-0de7-48f9-96e1-aff5f2098406

uncertainty-aware-label-distribution-learning

2209.10448

https://arxiv.org/abs/2209.10448v1

https://arxiv.org/pdf/2209.10448v1.pdf

Uncertainty-aware Label Distribution Learning for Facial Expression Recognition

Despite significant progress over the past few years, ambiguity is still a key challenge in Facial Expression Recognition (FER). It can lead to noisy and inconsistent annotation, which hinders the performance of deep learning models in real-world scenarios. In this paper, we propose a new uncertainty-aware label distribution learning method to improve the robustness of deep models against uncertainty and ambiguity. We leverage neighborhood information in the valence-arousal space to adaptively construct emotion distributions for training samples. We also consider the uncertainty of provided labels when incorporating them into the label distributions. Our method can be easily integrated into a deep network to obtain more training supervision and improve recognition accuracy. Intensive experiments on several datasets under various noisy and ambiguous settings show that our method achieves competitive results and outperforms recent state-of-the-art approaches. Our code and models are available at https://github.com/minhnhatvt/label-distribution-learning-fer-tf.

['Anh Nguyen', 'Bac Le', 'Erman Tjiputra', 'Quang Tran', 'Khanh Nguyen', 'Nhat Le']

2022-09-21

['facial-expression-recognition']

['computer-vision']

[-2.77533904e-02 -1.30774662e-01 -2.95581967e-01 -1.24577272e+00 -9.06380594e-01 -4.65272665e-01 1.62898704e-01 -3.44246656e-01 -3.67085278e-01 8.06286931e-01 1.93658508e-02 1.03495635e-01 1.59000233e-01 -2.79262275e-01 -5.12139201e-01 -6.37156844e-01 2.24047273e-01 3.52315634e-01 -3.90763313e-01 3.85941081e-02 -2.32932970e-01 3.66312176e-01 -1.64668977e+00 3.92436773e-01 7.19315350e-01 1.66776252e+00 -3.50729793e-01 -6.42932765e-03 -2.20954478e-01 6.46980703e-01 -5.76444209e-01 -6.29707992e-01 8.12641978e-02 -2.30873361e-01 -6.85318053e-01 1.17603146e-01 3.91244590e-01 -3.62796694e-01 -1.31907806e-01 1.22628295e+00 7.83793211e-01 1.71008721e-01 6.77809179e-01 -1.55433583e+00 -5.85445166e-01 4.79215682e-01 -6.94683194e-01 -9.02834907e-03 -8.99932384e-02 -2.26148680e-01 8.16764772e-01 -1.12993622e+00 3.21681947e-01 1.35944426e+00 6.17475808e-01 8.87937188e-01 -1.10944879e+00 -1.29545343e+00 4.27606165e-01 3.69777292e-01 -1.68065155e+00 -8.03341746e-01 9.02387381e-01 -1.50001571e-01 3.84239405e-01 1.81050617e-02 2.95351148e-01 1.52746177e+00 -2.43905649e-01 1.07590461e+00 1.17891681e+00 -2.63616443e-01 3.11918586e-01 5.39729111e-02 1.08580671e-01 7.04434395e-01 -1.04930937e-01 -2.30118468e-01 -6.63112819e-01 -2.33286306e-01 4.65528011e-01 1.58574358e-02 -2.66500950e-01 8.30442924e-03 -3.31927091e-01 7.48051405e-01 3.25164944e-01 1.61701456e-01 -1.53758496e-01 2.94690907e-01 5.51308572e-01 5.11622988e-02 9.12729502e-01 1.28813460e-01 -6.07507825e-01 -2.51923084e-01 -6.74627900e-01 1.59831539e-01 4.44923759e-01 7.30708838e-01 7.57507145e-01 3.42133082e-02 -3.38163793e-01 1.30173957e+00 4.16633576e-01 3.18075240e-01 1.70973137e-01 -1.18908715e+00 1.09163515e-01 2.60000497e-01 1.86660826e-01 -9.80890930e-01 -3.90772402e-01 -4.00193095e-01 -7.25066483e-01 1.01108747e-02 3.59372467e-01 -4.85626459e-01 -1.09349823e+00 2.13251710e+00 3.38353813e-01 5.33790529e-01 -3.59431207e-01 1.17074740e+00 7.70420849e-01 3.83653283e-01 3.52979839e-01 -2.02292204e-01 1.30198765e+00 -9.00015712e-01 -1.10684025e+00 -1.96339026e-01 4.66059119e-01 -5.23486376e-01 1.00883567e+00 6.48370922e-01 -6.95837915e-01 -3.53594124e-01 -8.21216941e-01 -1.03240564e-01 -7.53134936e-02 4.11009938e-01 8.80393624e-01 6.51541054e-01 -8.41024756e-01 3.86164576e-01 -7.62939274e-01 1.66669518e-01 1.05661118e+00 4.10319984e-01 -4.09368366e-01 -2.84478456e-01 -1.44654059e+00 5.13262630e-01 1.87018216e-01 5.34228325e-01 -8.38925242e-01 -4.13459808e-01 -8.75725329e-01 -1.04127310e-01 5.62231839e-01 -2.07304686e-01 1.58925891e+00 -1.37630582e+00 -1.61959112e+00 8.51396620e-01 -4.28139418e-01 -3.91648524e-02 4.63790417e-01 -3.48523825e-01 -3.78653169e-01 -1.56089544e-01 -1.53833076e-01 9.15350735e-01 8.83716941e-01 -1.29943943e+00 -4.56067801e-01 -5.62754393e-01 -1.23256467e-01 1.34095550e-01 -4.26991522e-01 1.50019944e-01 -5.87267697e-01 -5.61637998e-01 -1.90242827e-02 -9.42335010e-01 -1.08758889e-01 2.65318692e-01 -1.21360265e-01 -6.85488701e-01 5.50395429e-01 -3.94571334e-01 1.18094587e+00 -2.27309227e+00 -3.25111821e-02 1.00581340e-01 -1.73627138e-02 3.03115398e-01 -2.10512757e-01 -2.21108794e-01 -5.18179685e-02 3.19753081e-01 -1.55103967e-01 -9.43823457e-01 2.42558137e-01 4.77872014e-01 -2.66654164e-01 4.55282331e-01 3.84236306e-01 6.98260188e-01 -7.78171659e-01 -5.04417002e-01 -2.14882910e-01 7.80205786e-01 -3.64427567e-01 2.88388073e-01 -3.33273977e-01 5.32205284e-01 -4.52470690e-01 1.01371968e+00 9.75527108e-01 -1.34209991e-01 1.44482583e-01 -3.08531672e-01 5.52993834e-01 9.11154374e-02 -1.11150646e+00 1.81918061e+00 -4.56975669e-01 5.29214501e-01 1.07667066e-01 -9.48177755e-01 1.09713805e+00 2.34846756e-01 4.09028471e-01 -5.13840199e-01 6.29351914e-01 1.83983162e-01 -2.13347554e-01 -2.95650423e-01 3.12438130e-01 -1.97592035e-01 -8.56653303e-02 3.31629395e-01 3.29125494e-01 4.28856909e-01 -6.80143237e-02 -2.39226699e-01 6.04196429e-01 1.67103067e-01 -8.93008783e-02 1.70628857e-02 3.14317226e-01 -6.74555302e-01 1.15265298e+00 4.51416761e-01 -6.78631425e-01 6.54440105e-01 5.52356958e-01 -4.71555531e-01 -3.92832845e-01 -7.91586995e-01 -4.89038736e-01 1.43133414e+00 -9.95386913e-02 -3.75809520e-01 -7.92340040e-01 -9.66481507e-01 1.91036314e-02 5.49606144e-01 -7.99613833e-01 -2.22756162e-01 -4.86133471e-02 -8.47481847e-01 7.23043084e-01 7.12611198e-01 5.02416134e-01 -8.37785959e-01 1.42863300e-02 2.20610239e-02 -5.14389098e-01 -1.10129130e+00 -3.08746189e-01 1.60265535e-01 -4.45104420e-01 -6.97817624e-01 -4.16547000e-01 -3.87198061e-01 6.33145452e-01 -2.48284906e-01 1.06226814e+00 1.57533269e-02 -1.24920048e-01 1.05788879e-01 -4.03528571e-01 -8.24925482e-01 4.05433252e-02 -6.29597232e-02 1.94130018e-01 3.18843424e-01 8.07287991e-01 -4.50069040e-01 -6.26639664e-01 4.54598218e-01 -9.15160716e-01 -3.46372724e-01 1.12662323e-01 1.07656610e+00 8.05508792e-01 -8.55932385e-02 8.68498981e-01 -8.07408810e-01 6.92009151e-01 -6.09965265e-01 -4.94947433e-01 2.07947195e-01 -5.10981023e-01 -1.47838099e-02 2.66537070e-01 -5.34281135e-01 -1.15099823e+00 1.61753297e-01 -3.99431705e-01 -7.99039066e-01 -3.48657966e-01 5.17389834e-01 -4.66063350e-01 -3.04860473e-02 5.20164430e-01 -3.81270021e-01 1.36723258e-02 -5.11666000e-01 2.43698373e-01 1.04833794e+00 3.03412944e-01 -8.93654704e-01 1.52837941e-02 3.78780782e-01 -1.87289506e-01 -2.25624323e-01 -1.39446259e+00 -1.88162297e-01 -3.27648491e-01 -4.03696597e-01 3.87068480e-01 -1.17565346e+00 -5.65870523e-01 5.99903345e-01 -1.10773861e+00 -2.05139160e-01 1.65173307e-01 1.59811080e-01 -3.34503680e-01 6.88361228e-02 -6.34594917e-01 -9.72445488e-01 -3.57065737e-01 -1.28120041e+00 1.28322411e+00 5.22492468e-01 -1.17470205e-01 -8.33114326e-01 -1.31116390e-01 4.81271207e-01 3.63258094e-01 2.04366028e-01 3.73604655e-01 -6.98656142e-01 -2.38633379e-02 -3.42528373e-02 -3.08180571e-01 7.19158411e-01 2.42540404e-01 1.59357693e-02 -1.49278951e+00 -7.62450472e-02 -9.64813828e-02 -1.02723372e+00 1.09660935e+00 2.28202984e-01 1.86049163e+00 -8.93882364e-02 -1.55438364e-01 7.34207571e-01 9.31861520e-01 -1.06049225e-01 5.33897758e-01 -2.61237342e-02 5.94273806e-01 5.54683506e-01 9.80353951e-01 7.74073124e-01 4.66327339e-01 6.27957106e-01 4.48112845e-01 4.50399518e-02 1.58410251e-01 -8.95755067e-02 2.50367343e-01 3.21785241e-01 1.08388893e-01 -3.69397253e-01 -8.33051145e-01 3.79623771e-01 -2.02864790e+00 -7.23240674e-01 4.91504103e-01 1.74564421e+00 1.24277222e+00 -2.45021999e-01 -6.15120679e-02 -8.67680460e-02 6.85644150e-01 2.48860061e-01 -8.21814239e-01 -4.81119573e-01 3.42226699e-02 1.95174396e-01 9.29026604e-02 4.15094972e-01 -1.34828997e+00 1.17437565e+00 5.83294582e+00 1.12147295e+00 -1.34591305e+00 2.57772803e-01 1.33969367e+00 -2.55564421e-01 -2.83053845e-01 -5.16367137e-01 -9.86952484e-01 5.98341167e-01 8.05270553e-01 1.81224808e-01 3.38910401e-01 1.07156479e+00 1.69121251e-02 5.67364357e-02 -1.00626659e+00 1.34226692e+00 1.48828551e-01 -8.98487687e-01 -1.72000393e-01 -1.59344494e-01 6.36679471e-01 9.22485590e-02 3.54852498e-01 4.36260611e-01 4.18947160e-01 -1.26969051e+00 5.76637626e-01 5.35559535e-01 9.74888086e-01 -1.04949093e+00 9.74542975e-01 2.19450276e-02 -7.23839402e-01 -2.15759110e-02 -6.04825616e-01 1.06543638e-01 2.43427102e-02 8.40226233e-01 -6.76678777e-01 3.68867993e-01 8.52613151e-01 7.27839172e-01 -4.09567863e-01 6.34488881e-01 -4.90285933e-01 8.19685042e-01 -5.10072947e-01 1.49056047e-01 1.04254171e-01 -8.40628222e-02 6.13029711e-02 1.11170042e+00 3.40290278e-01 1.26929134e-01 1.42918840e-01 8.67057502e-01 -5.89057267e-01 4.05776054e-02 -2.78055876e-01 -9.09952223e-02 7.78966904e-01 1.46772659e+00 -3.30188572e-01 -4.83283810e-02 -1.42733052e-01 8.98371398e-01 8.04970324e-01 3.99974406e-01 -1.03040171e+00 -1.93263334e-03 1.04939139e+00 -3.72239351e-01 1.52304262e-01 -1.85231841e-03 -1.44811168e-01 -1.20838189e+00 4.78633046e-02 -9.59379971e-01 5.91599882e-01 -6.99476302e-01 -1.71739662e+00 7.70086706e-01 -1.20595992e-01 -8.62614512e-01 -2.49869436e-01 -7.41743982e-01 -1.77738667e-01 6.91765845e-01 -1.61557400e+00 -1.00989795e+00 -2.44609296e-01 5.55097342e-01 3.41883987e-01 -7.35104978e-02 9.72416401e-01 4.98596907e-01 -7.98743367e-01 1.11599696e+00 -1.15812654e-02 3.23377937e-01 1.20134544e+00 -9.81015086e-01 -1.03735141e-01 4.98657912e-01 1.68883622e-01 3.28303277e-01 5.24542868e-01 -3.10707331e-01 -9.19004858e-01 -1.27047050e+00 4.34577346e-01 -3.13280255e-01 5.29542446e-01 -5.43475032e-01 -9.91862059e-01 6.16708219e-01 1.71922237e-01 5.90920985e-01 1.20773911e+00 4.43926126e-01 -7.69622803e-01 -3.70762795e-01 -1.15611243e+00 4.13089335e-01 1.07955778e+00 -5.32365859e-01 -9.61394534e-02 1.67487055e-01 4.72953767e-01 -6.02043629e-01 -7.77072787e-01 7.56420493e-01 6.75685406e-01 -7.11875618e-01 4.32318807e-01 -7.08866417e-01 1.37124553e-01 -1.65325508e-01 -2.58004367e-01 -1.44696856e+00 -1.68710798e-01 -5.12588978e-01 -4.87083830e-02 1.37023103e+00 4.06509489e-01 -4.77081299e-01 6.98212445e-01 1.01422811e+00 6.52710795e-02 -1.06413841e+00 -1.10383356e+00 -4.54742193e-01 1.95996817e-02 -7.00960875e-01 6.81963682e-01 1.04268253e+00 -1.38999298e-01 1.61705792e-01 -5.60482621e-01 7.29988366e-02 4.72291857e-01 -6.99368268e-02 4.12075371e-01 -1.14130306e+00 -3.04849502e-02 -2.59690970e-01 -3.42909306e-01 -7.37194657e-01 8.64754617e-01 -6.88900471e-01 1.76001713e-01 -1.10230625e+00 1.16480775e-01 -6.30760074e-01 -8.03501427e-01 1.13372171e+00 -2.76265562e-01 4.82303232e-01 -1.28153265e-02 -5.77993244e-02 -9.94328558e-01 9.79132533e-01 8.45749855e-01 -2.06922218e-02 1.30613312e-01 -1.47226617e-01 -9.35783684e-01 8.31851602e-01 1.10311806e+00 -6.06719255e-01 -4.42965806e-01 -6.27734780e-01 2.54406065e-01 -3.73365164e-01 -2.90352502e-03 -6.64741457e-01 -5.45556955e-02 -1.56424135e-01 4.91448909e-01 -1.50852859e-01 5.34608066e-01 -8.46217811e-01 -7.39912614e-02 -3.67297858e-01 -4.90843505e-01 -2.76597887e-01 4.58225548e-01 5.05920410e-01 -3.84198755e-01 -1.49546534e-01 7.62977004e-01 1.37225360e-01 -7.17009783e-01 7.18080878e-01 -4.17861678e-02 7.80825242e-02 7.26906419e-01 3.61397654e-01 -3.51165295e-01 -6.23007834e-01 -8.71703625e-01 4.47809279e-01 3.34675908e-01 6.37008071e-01 5.29583871e-01 -1.60472941e+00 -5.69280267e-01 7.17820153e-02 3.11361194e-01 1.21783368e-01 3.23291242e-01 6.66979432e-01 7.18327165e-02 -3.30441669e-02 -1.44851312e-01 -6.58008456e-01 -1.41992915e+00 2.47078519e-02 6.73229814e-01 2.19521555e-03 1.74085021e-01 1.09739947e+00 5.96043002e-03 -4.89464641e-01 6.25519693e-01 5.07610962e-02 -1.05954722e-01 7.69621208e-02 7.47545838e-01 -2.22913064e-02 1.69411674e-02 -6.98013544e-01 -4.62120265e-01 2.43868008e-01 -3.65625799e-01 -5.44654131e-02 1.24597037e+00 -2.03279451e-01 -1.13532253e-01 5.73229909e-01 1.31930089e+00 -4.06653672e-01 -1.43717396e+00 -4.63714689e-01 -1.34938613e-01 -5.93408227e-01 4.23809469e-01 -1.06466508e+00 -1.39272106e+00 9.77151215e-01 8.09350669e-01 -4.23663288e-01 1.33051682e+00 -3.17871198e-02 5.09635627e-01 4.28338021e-01 3.97215426e-01 -1.20538533e+00 -7.26354821e-03 4.24254119e-01 8.58174384e-01 -1.52619708e+00 -2.39713207e-01 -4.28236365e-01 -8.93830001e-01 9.27845538e-01 8.58747005e-01 2.47630268e-01 8.89789641e-01 4.96971726e-01 5.68153322e-01 -2.32028845e-03 -9.12614286e-01 -1.62231952e-01 2.32869789e-01 4.66814041e-01 7.08496094e-01 6.98794946e-02 -2.19394818e-01 1.15468967e+00 2.14285702e-01 3.31194341e-01 1.21249303e-01 7.66560316e-01 -2.02675417e-01 -1.36059070e+00 -8.79183933e-02 2.54513830e-01 -8.48921955e-01 4.89232019e-02 -4.26172376e-01 3.53466541e-01 3.57906550e-01 9.86947417e-01 9.58328620e-02 -4.73921925e-01 2.96092555e-02 4.22169268e-01 5.21546304e-01 -4.41723764e-01 -1.04286976e-01 1.85340479e-01 3.85472834e-01 -7.69756794e-01 -5.79435885e-01 -6.39795244e-01 -1.29767370e+00 -1.47776995e-02 -4.35404956e-01 2.21171439e-01 6.86265469e-01 8.62987041e-01 8.31609845e-01 1.77574471e-01 7.15560257e-01 -6.91747785e-01 -6.44764245e-01 -1.10537839e+00 -6.75854206e-01 4.16475356e-01 1.22391135e-01 -1.11907852e+00 -4.12617534e-01 -1.11916460e-01]

[13.67630672454834, 1.6823756694793701]

f76747c9-c520-43af-81f1-188ed60f1e0d

few-shot-fine-grained-image-classification

2207.08547

https://arxiv.org/abs/2207.08547v2

https://arxiv.org/pdf/2207.08547v2.pdf

Few-shot Fine-grained Image Classification via Multi-Frequency Neighborhood and Double-cross Modulation

Traditional fine-grained image classification typically relies on large-scale training samples with annotated ground-truth. However, some sub-categories have few available samples in real-world applications, and current few-shot models still have difficulty in distinguishing subtle differences among fine-grained categories. To solve this challenge, we propose a novel few-shot fine-grained image classification network (FicNet) using multi-frequency neighborhood (MFN) and double-cross modulation (DCM). MFN focuses on both spatial domain and frequency domain to capture multi-frequency structural representations, which reduces the influence of appearance and background changes to the intra-class distance. DCM consists of bi-crisscross component and double 3D cross-attention component. It modulates the representations by considering global context information and inter-class relationship respectively, which enables the support and query samples respond to the same parts and accurately identify the subtle inter-class differences. The comprehensive experiments on three fine-grained benchmark datasets for two few-shot tasks verify that FicNet has excellent performance compared to the state-of-the-art methods. Especially, the experiments on two datasets, "Caltech-UCSD Birds" and "Stanford Cars", can obtain classification accuracy 93.17\% and 95.36\%, respectively. They are even higher than that the general fine-grained image classification methods can achieve.

['Chengqing Li', 'Yange Zhou', 'Jiayi Wang', 'Zhan Gao', 'Hegui Zhu']

2022-07-18

['fine-grained-image-classification']

['computer-vision']

[ 3.24859768e-02 -6.59000456e-01 -3.81908834e-01 -4.27165866e-01 -6.69795930e-01 -2.42183074e-01 5.57850599e-01 6.14399500e-02 -2.66714752e-01 6.94769382e-01 1.52612731e-01 4.31285173e-01 -2.44497493e-01 -8.90289009e-01 -4.57590342e-01 -7.24571884e-01 2.28164509e-01 -6.24731109e-02 8.38040113e-01 -2.58615136e-01 3.94505709e-01 1.38928205e-01 -2.15168524e+00 6.34184659e-01 8.31786692e-01 1.50360942e+00 3.80805850e-01 2.95628697e-01 -3.43631148e-01 6.75300062e-01 -7.57115066e-01 1.66186109e-01 -3.05665582e-02 -2.47685552e-01 -5.87631524e-01 1.42200679e-01 5.62766194e-01 -1.80952266e-01 -3.33771348e-01 1.40223157e+00 5.13413548e-01 4.42564577e-01 8.57997835e-01 -1.23813438e+00 -1.21952808e+00 2.03085542e-01 -7.50543058e-01 8.14566731e-01 -3.34895588e-02 2.00975388e-01 7.77968407e-01 -9.13071096e-01 3.17497432e-01 1.48871493e+00 5.91031671e-01 4.14408416e-01 -9.27396119e-01 -8.17929029e-01 5.16237617e-01 6.19348705e-01 -1.58479905e+00 -2.68346876e-01 5.46889186e-01 -6.15566313e-01 8.67655933e-01 1.34563774e-01 3.05487365e-01 9.00598705e-01 2.52818882e-01 4.27173972e-01 1.14370298e+00 -2.28983000e-01 1.24117315e-01 -2.26222388e-02 6.70456171e-01 5.91896951e-01 6.72831908e-02 2.99303710e-01 -3.06339115e-01 7.83482715e-02 6.83700025e-01 5.15087783e-01 -4.21272129e-01 -1.91825643e-01 -1.16499376e+00 9.03589606e-01 7.69438744e-01 6.57140493e-01 -9.42054465e-02 -1.38734415e-01 6.07017517e-01 1.81200355e-01 5.36522746e-01 1.81008410e-02 -2.72220105e-01 1.69600397e-01 -7.07901597e-01 -2.74383537e-02 2.50673354e-01 9.51257408e-01 9.85641062e-01 1.36034250e-01 -6.75629914e-01 1.28096545e+00 1.38337627e-01 2.71188676e-01 1.04698348e+00 -4.37484890e-01 1.93652272e-01 6.10348344e-01 -1.82975158e-01 -1.24688482e+00 -1.27618372e-01 -7.60099590e-01 -1.17385519e+00 8.75918791e-02 8.26884061e-02 3.08499724e-01 -1.17202032e+00 1.47072613e+00 3.13887417e-01 5.82252920e-01 -2.95127422e-01 1.12949765e+00 1.47968650e+00 7.75347650e-01 3.36162716e-01 -2.21321851e-01 1.61867547e+00 -1.11923707e+00 -6.04005754e-01 -1.66920751e-01 2.41303027e-01 -6.44788384e-01 1.45991600e+00 3.62796187e-02 -5.56220055e-01 -1.34362388e+00 -1.40685856e+00 2.79623382e-02 -7.95387208e-01 -8.43785480e-02 4.83213246e-01 3.61631364e-01 -5.18097043e-01 5.62694430e-01 -1.57663599e-01 -2.88995564e-01 6.66125119e-01 -1.03813715e-01 -3.58142644e-01 -4.73685294e-01 -1.49365807e+00 4.76060838e-01 4.58068311e-01 -1.46284923e-01 -9.02826667e-01 -9.54668164e-01 -8.07237864e-01 2.30288282e-01 2.78700054e-01 -4.65693414e-01 9.48592067e-01 -8.69807899e-01 -9.31673825e-01 9.00202155e-01 1.03423655e-01 -5.88305444e-02 5.69527708e-02 1.96303844e-01 -8.19481730e-01 1.55401304e-01 5.70926249e-01 6.07842028e-01 9.52620149e-01 -8.63462687e-01 -8.57805192e-01 -3.72320503e-01 7.14583099e-02 4.80875373e-02 -2.50162482e-01 -1.07302807e-01 -4.17477965e-01 -1.00219095e+00 -1.11857511e-01 -4.14415687e-01 1.29258499e-01 7.94585720e-02 -9.83440056e-02 -4.52729076e-01 1.10695255e+00 -9.62765887e-02 1.30762899e+00 -2.47910428e+00 -3.67518008e-01 -3.76990169e-01 1.76684976e-01 5.32400787e-01 -4.47125226e-01 1.19891368e-01 -1.19431801e-01 3.39979976e-02 1.26548067e-01 2.74735361e-01 3.65432277e-02 9.81233492e-02 -3.26938212e-01 3.34581792e-01 2.24985212e-01 8.52426887e-01 -8.63254786e-01 -5.92668712e-01 4.43024039e-01 2.99863786e-01 -1.81464016e-01 1.28236383e-01 -8.85042250e-02 7.38698095e-02 -5.54217398e-01 7.48910367e-01 7.12595463e-01 -4.62978959e-01 -4.89804804e-01 -6.38847530e-01 1.35788685e-02 -4.30346519e-01 -1.15192401e+00 1.50263548e+00 -2.52087593e-01 2.55219340e-01 -2.38321975e-01 -1.19424498e+00 9.61641848e-01 -4.35167402e-02 3.64835449e-02 -1.17168748e+00 2.74364293e-01 -5.07714674e-02 -1.70160502e-01 -5.28565884e-01 2.73326993e-01 -3.45219642e-01 -3.35924238e-01 1.09169073e-01 2.91428298e-01 2.54227638e-01 1.82110116e-01 9.81523916e-02 7.88248777e-01 -2.33058736e-01 5.07144570e-01 -5.84052205e-01 5.35847008e-01 -7.95331001e-02 6.70656741e-01 7.41922200e-01 -7.32310951e-01 6.55591369e-01 -3.94146405e-02 -5.36412835e-01 -5.72640717e-01 -9.30260360e-01 -2.34034151e-01 1.50781250e+00 8.83813858e-01 -1.95856422e-01 -6.32932186e-01 -6.92922652e-01 7.34080225e-02 4.80421901e-01 -1.01553655e+00 -5.78267157e-01 -4.52813134e-02 -7.79362321e-01 2.47502118e-01 5.93836427e-01 9.23261821e-01 -1.03182340e+00 -3.54853928e-01 1.54500827e-01 -2.15493649e-01 -9.22290504e-01 -8.49209547e-01 5.39502092e-02 -4.19514447e-01 -1.22589552e+00 -7.71161139e-01 -9.59691763e-01 2.43852586e-01 9.79968011e-01 1.09200752e+00 1.44008666e-01 -7.69013822e-01 -4.66715656e-02 -6.34040475e-01 -1.69518650e-01 2.61311978e-01 -1.71183184e-01 -1.47593379e-01 2.18555346e-01 6.25997543e-01 -4.06331897e-01 -7.96909273e-01 7.56709397e-01 -7.82407105e-01 -2.66212612e-01 6.01917148e-01 1.22313511e+00 9.31609690e-01 3.69130015e-01 8.17747951e-01 -8.43107283e-01 4.68475223e-01 -6.50976181e-01 -1.93862841e-01 2.33382806e-01 -4.36760753e-01 -3.07782024e-01 7.57414520e-01 -6.79609299e-01 -9.70019042e-01 -5.38428128e-01 6.47685155e-02 -8.34982634e-01 -5.83382964e-01 2.74779290e-01 -2.07454890e-01 -1.37826815e-01 7.91888714e-01 4.25447613e-01 -3.67270470e-01 -4.83401656e-01 1.97350383e-01 7.38929391e-01 4.74122375e-01 -3.82967442e-01 4.69130605e-01 3.08896691e-01 -3.98360610e-01 -8.06107998e-01 -1.19704103e+00 -7.04822838e-01 -4.56831604e-01 -6.00417852e-02 8.64142418e-01 -1.04135156e+00 -3.51145416e-01 6.15843952e-01 -7.43027568e-01 -7.70002007e-02 -3.93633038e-01 3.06135327e-01 -3.24780136e-01 1.89546451e-01 -5.14215887e-01 -2.75197566e-01 -3.19729269e-01 -1.00364769e+00 1.17523658e+00 5.49935997e-01 2.15695426e-01 -6.76717043e-01 -2.14832783e-01 1.73505008e-01 5.38141310e-01 1.29493788e-01 9.62920368e-01 -4.35439825e-01 -4.26822037e-01 -1.06635898e-01 -7.27790952e-01 1.75510153e-01 2.64448673e-01 -2.15192646e-01 -1.06016517e+00 -3.37153226e-01 -5.58795072e-02 -4.73526657e-01 1.15212405e+00 4.11676049e-01 1.63003778e+00 -1.44267753e-01 -4.78947490e-01 5.75550139e-01 1.56659448e+00 1.97616652e-01 5.99859774e-01 9.83695313e-02 6.73384547e-01 3.38799119e-01 1.20190263e+00 4.26486582e-01 1.70752734e-01 7.43159413e-01 4.70049083e-01 1.93430752e-01 -4.97298807e-01 -1.31126836e-01 -2.25973800e-01 6.72576845e-01 1.15991630e-01 -4.52853143e-02 -5.11858463e-01 6.85034573e-01 -1.80460072e+00 -1.30593288e+00 1.45673871e-01 1.72309732e+00 6.23902738e-01 2.11758316e-01 3.14476783e-03 6.72634318e-02 1.16033399e+00 5.04320681e-01 -7.75760472e-01 -2.27251723e-02 -1.39408097e-01 9.91224125e-02 2.70257026e-01 5.19403927e-02 -1.41097641e+00 8.14873159e-01 5.23347330e+00 1.68132997e+00 -1.28816652e+00 3.32638353e-01 8.95239472e-01 1.65453218e-02 5.79867214e-02 -5.25153279e-01 -9.75619256e-01 8.39075208e-01 5.88742137e-01 -9.91178751e-02 1.85783565e-01 1.03603721e+00 -8.32131132e-02 7.67034516e-02 -7.68410027e-01 1.38399339e+00 1.84583262e-01 -1.50594532e+00 2.24831700e-01 -3.55346799e-01 7.39944816e-01 1.04397982e-01 1.44007560e-02 7.64657438e-01 9.08378139e-02 -1.09784210e+00 7.59173453e-01 6.64948404e-01 1.33652592e+00 -6.81800604e-01 7.25702167e-01 5.50050497e-01 -1.85442042e+00 -1.99070007e-01 -8.92405748e-01 1.51277008e-02 -2.31910110e-01 6.01056695e-01 -3.06746215e-02 4.94454384e-01 1.05483365e+00 9.29929912e-01 -6.01396322e-01 9.18048501e-01 2.84543425e-01 2.97366500e-01 2.56247908e-01 -1.47516012e-01 4.43448871e-01 1.67468250e-01 1.75151661e-01 1.17834044e+00 1.03510171e-01 4.57786143e-01 5.68739831e-01 6.30611837e-01 4.05586287e-02 -1.32281646e-01 -1.63272768e-01 2.95785572e-02 6.44010723e-01 1.28187358e+00 -7.54723251e-01 -5.73330522e-01 -5.29204667e-01 8.56024027e-01 3.16960037e-01 2.50991881e-01 -8.80443633e-01 -9.76824343e-01 8.22229683e-01 -1.37322850e-03 6.99555635e-01 3.22178513e-01 8.96268040e-02 -1.27103245e+00 -2.11219445e-01 -8.95188510e-01 7.47061014e-01 -8.71998668e-01 -1.86768162e+00 7.67456234e-01 -1.38609365e-01 -1.40922892e+00 1.46524832e-01 -4.43099052e-01 -7.21661866e-01 8.85784268e-01 -1.74926376e+00 -1.27870452e+00 -8.42668116e-01 1.00181925e+00 9.91536736e-01 -1.43124580e-01 8.07780504e-01 5.20506859e-01 -4.77026820e-01 8.95212531e-01 -2.91619878e-02 4.02162448e-02 6.97904766e-01 -8.49690318e-01 -6.66691642e-03 5.85059166e-01 -1.99610561e-01 3.81489486e-01 3.49590868e-01 -4.41790372e-01 -1.02846944e+00 -1.60486090e+00 2.85479635e-01 6.84400350e-02 5.18827617e-01 -8.48068818e-02 -1.09882045e+00 1.59884796e-01 -8.65727141e-02 7.94364452e-01 5.13778031e-01 -9.66056138e-02 -8.41254115e-01 -4.46300179e-01 -1.28412688e+00 1.29721031e-01 1.28169298e+00 -6.98673248e-01 -7.02995241e-01 9.23549607e-02 1.01526582e+00 -1.29391719e-02 -7.88878620e-01 6.21192098e-01 4.72618520e-01 -1.06933272e+00 1.11544967e+00 -6.50211394e-01 2.72223890e-01 -5.11956751e-01 -6.45969689e-01 -1.39578378e+00 -1.01569295e+00 1.40076563e-01 -3.58595699e-03 1.23978770e+00 -3.89966547e-01 -6.26238048e-01 4.53543156e-01 -1.89233541e-01 -3.64256740e-01 -8.16312253e-01 -6.86838627e-01 -8.75126123e-01 -4.77423370e-02 -1.71385288e-01 7.74293959e-01 1.03018129e+00 -3.61824304e-01 5.98471820e-01 -2.74543434e-01 1.03865489e-01 6.58532023e-01 7.05514908e-01 4.06443357e-01 -1.39585054e+00 -2.44269088e-01 -5.76520801e-01 -6.99497223e-01 -9.69681859e-01 3.69898006e-02 -5.60344815e-01 1.38870493e-01 -1.35098267e+00 5.18074989e-01 -5.19880235e-01 -7.08756804e-01 3.18411797e-01 -3.75975072e-01 7.61766374e-01 6.86917678e-02 3.97397488e-01 -8.57183576e-01 7.11680651e-01 1.40815997e+00 -6.18808687e-01 2.62561381e-01 -2.44868666e-01 -8.61754239e-01 6.46300137e-01 5.17446280e-01 -1.44694403e-01 -5.94387412e-01 -1.28312051e-01 -5.30027866e-01 -1.81028500e-01 4.32528377e-01 -1.14280641e+00 2.03850240e-01 -4.08603847e-01 6.59753382e-01 -7.45371938e-01 2.89037794e-01 -4.75553095e-01 9.74099264e-02 4.12451476e-01 -1.28828064e-01 -5.18250346e-01 2.81952560e-01 8.69284630e-01 -6.58179224e-01 -7.47469813e-02 1.39504516e+00 -3.92487317e-01 -1.57499945e+00 8.91073763e-01 1.09830894e-01 4.90468413e-01 1.24346066e+00 -3.67050648e-01 -7.86486983e-01 4.61993478e-02 -5.79571009e-01 1.55020226e-02 2.40762249e-01 6.61669195e-01 7.06583619e-01 -1.72688019e+00 -6.12109840e-01 4.78520840e-01 6.89371049e-01 -3.53500962e-01 1.09624863e+00 5.49607933e-01 2.25175427e-05 4.62512821e-01 -5.43648243e-01 -7.14121342e-01 -1.19310427e+00 9.73552644e-01 4.29549277e-01 -1.02111744e-02 -3.93051773e-01 1.08968663e+00 9.03181911e-01 -1.26663044e-01 1.03281468e-01 -1.41343072e-01 -4.72966433e-01 3.24442625e-01 1.04251230e+00 3.28025877e-01 1.73690431e-02 -8.71026814e-01 -4.63633150e-01 9.64158595e-01 -2.70431399e-01 7.43194044e-01 1.17989039e+00 -3.41786027e-01 1.32294476e-01 6.44100487e-01 1.41952717e+00 -4.52429533e-01 -1.23953617e+00 -5.72673500e-01 -5.68603516e-01 -7.96565831e-01 6.86130896e-02 -8.60061944e-01 -1.02962351e+00 1.17543554e+00 9.46982145e-01 4.63239253e-01 1.23839557e+00 1.75250053e-01 7.59723842e-01 -6.29987419e-02 5.09777427e-01 -9.87333715e-01 2.78687954e-01 6.34382844e-01 8.47798824e-01 -1.38283563e+00 -1.83735535e-01 -4.31568772e-01 -5.46983182e-01 8.71439815e-01 8.94708991e-01 -4.56765294e-01 1.08294439e+00 -1.93208069e-01 -9.47854072e-02 -1.62623718e-01 -7.79880047e-01 -3.11229318e-01 5.81192434e-01 7.09821224e-01 2.16779821e-02 1.24739148e-01 -1.11506961e-01 1.01863623e+00 2.05323160e-01 1.04310261e-02 -4.30264212e-02 6.64961159e-01 -1.01632237e+00 -4.55294341e-01 -2.12535918e-01 7.61085987e-01 -3.29934478e-01 -1.22669592e-01 7.42738172e-02 6.59613490e-01 6.25374138e-01 9.63739753e-01 3.40739936e-01 -6.23036206e-01 4.10265416e-01 -3.21992755e-01 3.24345618e-01 -7.88225174e-01 -2.75917858e-01 5.28895259e-02 -2.06814021e-01 -7.86413193e-01 -4.27449763e-01 -2.26875067e-01 -9.56656456e-01 -3.75041962e-01 -4.19629484e-01 1.45769194e-01 2.82295458e-02 9.37247455e-01 5.48621297e-01 8.93646777e-01 6.96033895e-01 -9.21620309e-01 -6.12516105e-01 -1.14106429e+00 -9.68194246e-01 6.24484301e-01 3.74388725e-01 -1.21733630e+00 -4.46272463e-01 -1.10384725e-01]

[9.686807632446289, 2.0359251499176025]

688872ca-8253-43bc-81b8-694c806abbff

spts-v2-single-point-scene-text-spotting

2301.01635

https://arxiv.org/abs/2301.01635v2

https://arxiv.org/pdf/2301.01635v2.pdf

SPTS v2: Single-Point Scene Text Spotting

End-to-end scene text spotting has made significant progress due to its intrinsic synergy between text detection and recognition. Previous methods commonly regard manual annotations such as horizontal rectangles, rotated rectangles, quadrangles, and polygons as a prerequisite, which are much more expensive than using single-point. For the first time, we demonstrate that training scene text spotting models can be achieved with an extremely low-cost single-point annotation by the proposed framework, termed SPTS v2. SPTS v2 reserves the advantage of the auto-regressive Transformer with an Instance Assignment Decoder (IAD) through sequentially predicting the center points of all text instances inside the same predicting sequence, while with a Parallel Recognition Decoder (PRD) for text recognition in parallel. These two decoders share the same parameters and are interactively connected with a simple but effective information transmission process to pass the gradient and information. Comprehensive experiments on various existing benchmark datasets demonstrate the SPTS v2 can outperform previous state-of-the-art single-point text spotters with fewer parameters while achieving 19$\times$ faster inference speed. Most importantly, within the scope of our SPTS v2, extensive experiments further reveal an important phenomenon that single-point serves as the optimal setting for the scene text spotting compared to non-point, rectangular bounding box, and polygonal bounding box. Such an attempt provides a significant opportunity for scene text spotting applications beyond the realms of existing paradigms. Code will be available at https://github.com/bytedance/SPTSv2.

['Lianwen Jin', 'Xiang Bai', 'Chunhua Shen', 'Dahua Lin', 'Can Huang', 'Jingqun Tang', 'Xinyu Wang', 'Mingxin Huang', 'Dezhi Peng', 'Jiaxin Zhang', 'Yuliang Liu']

2023-01-04

['text-spotting']

['computer-vision']

[ 4.51312065e-01 -2.09586307e-01 -7.40643218e-02 -2.31986508e-01 -9.03996766e-01 -4.42544252e-01 5.79771161e-01 -2.08142456e-02 -3.72300625e-01 1.64645329e-01 -1.67792231e-01 -5.39450765e-01 2.53419697e-01 -6.51974857e-01 -8.41920257e-01 -6.48344815e-01 4.35888380e-01 6.88341856e-01 3.40342015e-01 -8.55804980e-02 5.65780282e-01 3.00625443e-01 -1.28543305e+00 3.70183200e-01 9.10968900e-01 9.09413934e-01 5.28963566e-01 6.23558283e-01 -3.81713301e-01 4.05100614e-01 -5.28464675e-01 -4.93986219e-01 2.82431215e-01 -1.14722457e-02 -3.84871155e-01 1.50627449e-01 5.45104802e-01 -5.50662577e-01 -4.83902037e-01 7.71694839e-01 5.39759994e-01 1.18016012e-01 3.73027325e-01 -1.04538560e+00 -3.09641153e-01 6.15798771e-01 -1.01358902e+00 -1.22661427e-01 4.56095189e-01 2.81972766e-01 1.17923915e+00 -1.32459414e+00 2.87665427e-01 9.45581555e-01 8.20968747e-01 1.73213229e-01 -1.12786484e+00 -7.32213676e-01 2.77737975e-01 1.27940783e-02 -1.63497436e+00 -3.99244219e-01 6.64228618e-01 -2.64614463e-01 1.17705941e+00 6.05701566e-01 3.91224623e-01 1.13032627e+00 -1.38360456e-01 1.20456171e+00 6.26002133e-01 -4.49530482e-01 -1.01903968e-01 3.81328054e-02 1.08844839e-01 8.84239137e-01 1.96376629e-02 -3.82667482e-01 -6.84626460e-01 1.26770496e-01 7.62200475e-01 2.38002598e-01 -3.13648969e-01 -1.76134169e-01 -1.46917379e+00 4.68341470e-01 1.91208929e-01 1.54613212e-01 4.83890288e-02 1.46242544e-01 5.05218625e-01 5.20639643e-02 5.32913089e-01 9.90480408e-02 -3.73189181e-01 -3.73415619e-01 -1.28895617e+00 -1.48393828e-02 6.34260535e-01 1.22143817e+00 6.15411580e-01 -9.90283117e-03 -3.34432453e-01 9.78983283e-01 2.03064635e-01 7.80008733e-01 4.24718082e-01 8.44553262e-02 1.21233881e+00 7.80915916e-01 -2.26467289e-02 -9.84634757e-01 -3.66456121e-01 -2.52474844e-01 -8.25815856e-01 -6.02815934e-02 5.31140506e-01 -9.62546170e-02 -8.22117209e-01 1.01752377e+00 3.70369822e-01 1.69506073e-01 -2.87475288e-01 1.04967594e+00 6.39942586e-01 8.57928395e-01 -1.75712079e-01 3.19487065e-01 1.58887482e+00 -1.11270773e+00 -3.70087713e-01 -3.40274423e-01 9.25468564e-01 -9.83108461e-01 1.51343036e+00 3.64439726e-01 -8.11570048e-01 -3.52880985e-01 -9.23423290e-01 -4.89676207e-01 -4.23328161e-01 8.70172322e-01 3.25765401e-01 6.11139774e-01 -8.53670180e-01 2.52966374e-01 -9.08051848e-01 -3.47086817e-01 4.73368764e-01 3.73685569e-01 -1.42745271e-01 2.30704978e-01 -7.46689379e-01 4.90250349e-01 2.60911793e-01 2.84403324e-01 -4.73196447e-01 -7.70194769e-01 -7.73041487e-01 1.88268960e-01 6.19342446e-01 -4.04958844e-01 1.08027685e+00 -6.40145838e-01 -1.60204065e+00 7.96701014e-01 -4.27957743e-01 -3.27176332e-01 9.09552336e-01 -4.85673428e-01 -1.04787871e-01 1.97754558e-02 1.56643674e-01 5.12649000e-01 1.06811047e+00 -8.39530408e-01 -6.59637332e-01 -3.16503286e-01 -3.65598679e-01 5.47300339e-01 -4.66319680e-01 4.97604720e-02 -9.65398371e-01 -7.76577592e-01 2.39588946e-01 -8.17211568e-01 1.60781145e-01 2.56202579e-01 -1.13144803e+00 -3.11096430e-01 1.13584030e+00 -7.07390904e-01 1.37250888e+00 -2.33068681e+00 2.69857887e-02 1.50293440e-01 1.79747045e-01 1.07388288e-01 2.11296201e-01 5.14243901e-01 -4.88595553e-02 1.20234005e-01 -2.26271182e-01 -7.22653389e-01 2.77905524e-01 -3.03432554e-01 -5.92122197e-01 6.37759030e-01 2.99040060e-02 7.93088317e-01 -4.65708286e-01 -6.16148531e-01 5.99129677e-01 3.86179835e-01 -2.88258374e-01 -5.76203242e-02 -3.69585425e-01 3.11017148e-02 -6.03800476e-01 7.16298699e-01 6.11304760e-01 -6.38627410e-01 1.25590950e-01 -1.31334469e-01 -4.14789885e-01 4.13753092e-01 -1.25540805e+00 1.67225111e+00 -1.85908243e-01 1.01630640e+00 -1.56227782e-01 -9.23168540e-01 9.03927445e-01 1.59486651e-01 3.25724036e-01 -8.05935204e-01 1.03326403e-01 9.80382226e-03 -5.75977743e-01 -3.80881637e-01 8.09238732e-01 4.12565440e-01 -3.50925364e-02 4.23198849e-01 -4.11681265e-01 -5.35208127e-03 7.11489189e-03 1.68002829e-01 1.02972007e+00 2.88820624e-01 1.29766345e-01 6.32596612e-02 3.67949903e-01 2.23789494e-02 1.96473658e-01 8.44289362e-01 1.69571459e-01 8.05109620e-01 4.66930449e-01 -2.90342450e-01 -1.33361137e+00 -7.09511757e-01 -3.45451862e-01 1.20348811e+00 3.95928532e-01 -6.64771676e-01 -6.11525893e-01 -6.22345924e-01 -4.38383371e-02 8.54817688e-01 -4.56234455e-01 5.17863810e-01 -6.49647295e-01 -5.96979558e-01 8.56286466e-01 5.47640502e-01 6.81837440e-01 -6.36164606e-01 -7.77963400e-01 -1.89732268e-01 -1.77891955e-01 -1.25623190e+00 -8.24395478e-01 3.04699421e-01 -7.37363815e-01 -8.74923348e-01 -8.88714015e-01 -6.53954029e-01 8.03471267e-01 5.77815175e-01 7.29827583e-01 1.63253695e-01 -2.51561970e-01 2.21307561e-01 -3.97956073e-01 -1.27045572e-01 5.55221401e-02 2.51606792e-01 -3.68949026e-01 1.14661865e-01 2.51649320e-01 -2.28677437e-01 -7.30965137e-01 6.11427844e-01 -7.99925447e-01 8.57980728e-01 6.09800458e-01 7.55930901e-01 6.47751689e-01 -5.12760095e-02 -8.08194503e-02 -8.85148406e-01 1.41832754e-01 -3.38067681e-01 -8.10668588e-01 3.05757940e-01 -4.01766896e-01 -7.73085207e-02 8.71487081e-01 -2.21687764e-01 -9.87703621e-01 2.26726711e-01 -2.59834770e-02 -5.08152783e-01 -2.09440261e-01 2.94112653e-01 -8.01617652e-02 7.99588785e-02 4.91576195e-01 6.08541071e-01 -2.92937428e-01 -5.04436851e-01 2.13088185e-01 7.30278134e-01 1.93103686e-01 -4.52639759e-01 9.67146933e-01 6.39866829e-01 -2.34088674e-01 -1.12616611e+00 -4.67498362e-01 -6.41290843e-01 -7.54028976e-01 -9.76143554e-02 7.02941656e-01 -8.63220513e-01 -9.78621721e-01 5.42822361e-01 -1.02992630e+00 -5.80641210e-01 1.28808215e-01 1.07105412e-01 -3.00315410e-01 7.53237605e-01 -4.32550341e-01 -8.32050860e-01 -5.48101425e-01 -1.18949449e+00 1.75236559e+00 2.22729240e-02 3.41942981e-02 -7.65493512e-01 -3.75859708e-01 6.06335342e-01 2.68859249e-02 -2.15452075e-01 7.28836000e-01 -8.05298030e-01 -8.59731495e-01 -3.23559284e-01 -5.21380126e-01 -3.18035245e-01 -6.91861287e-02 1.13406442e-01 -9.00073826e-01 -2.40305647e-01 -3.85112166e-01 -7.52355009e-02 7.53200948e-01 2.01200679e-01 1.49417126e+00 -2.34849513e-01 -5.56081474e-01 8.90353501e-01 1.39524209e+00 1.89162046e-01 6.60360754e-01 4.07336593e-01 9.89510536e-01 1.47690549e-01 5.27329862e-01 7.26726711e-01 3.65892053e-01 1.02317154e+00 2.85977840e-01 -3.54685277e-01 -4.76978458e-02 -4.60674882e-01 3.33152890e-01 5.88753998e-01 3.35130453e-01 -5.68042338e-01 -1.12102759e+00 1.00634947e-01 -1.84015751e+00 -8.01586032e-01 -3.27521861e-01 2.30358815e+00 6.61578476e-01 1.26834065e-01 -4.70572338e-02 2.56472707e-01 8.91559899e-01 2.76798189e-01 -6.00127816e-01 3.27964090e-02 -2.20275357e-01 -1.34640723e-01 6.59510136e-01 3.21692824e-01 -1.25903773e+00 1.20001113e+00 5.21716642e+00 1.26988363e+00 -1.28217220e+00 -2.38854259e-01 7.21307755e-01 -1.73398346e-01 -7.25900978e-02 8.37189481e-02 -1.16625106e+00 6.64641082e-01 4.74220932e-01 1.92545608e-01 5.07157862e-01 8.80258441e-01 3.07139367e-01 -6.88182861e-02 -1.06730580e+00 1.32104015e+00 1.57632574e-01 -1.33552480e+00 9.91284102e-02 -1.11803360e-01 2.94330269e-01 1.02397896e-01 1.70566410e-01 1.26982018e-01 -1.11014895e-01 -8.95808220e-01 9.22066927e-01 2.47760326e-01 1.28379524e+00 -2.72701144e-01 1.87050775e-01 3.29422683e-01 -1.43334091e+00 1.00532226e-01 -3.35632056e-01 1.75255731e-01 6.09037690e-02 5.48866868e-01 -1.30547178e+00 4.99678761e-01 5.76894403e-01 8.21271956e-01 -6.13000154e-01 8.88864517e-01 -3.74109633e-02 7.04980016e-01 -7.24066138e-01 -3.59148264e-01 2.16405287e-01 -2.18510658e-01 6.72123969e-01 1.45582128e+00 3.90282422e-01 -1.23631902e-01 1.56957895e-01 8.93442512e-01 -8.99888724e-02 3.30031157e-01 -5.62900007e-01 1.21819876e-01 5.16469002e-01 1.12499666e+00 -1.18403733e+00 -4.52266455e-01 -5.79671919e-01 1.21337831e+00 2.37321973e-01 3.10459912e-01 -1.17859042e+00 -5.13782561e-01 1.36549145e-01 1.62795871e-01 5.98556280e-01 -3.43379229e-01 -8.32899332e-01 -1.39927912e+00 2.71569341e-01 -8.49337399e-01 2.16590196e-01 -8.74695957e-01 -8.50964487e-01 3.13433409e-01 -2.44946301e-01 -1.42928576e+00 3.72951239e-01 -7.93432772e-01 -5.20072401e-01 6.95576906e-01 -1.22340894e+00 -1.21658206e+00 -5.58486342e-01 6.68806434e-01 1.11035883e+00 1.53121923e-03 5.01261294e-01 2.32075840e-01 -1.11242044e+00 9.66420472e-01 2.97119766e-01 4.26285207e-01 6.35565221e-01 -1.19610834e+00 7.90310085e-01 8.50769401e-01 4.27222282e-01 4.02807504e-01 5.34585357e-01 -7.25635767e-01 -1.82251942e+00 -9.17763948e-01 5.87869346e-01 -3.99249196e-01 6.12231314e-01 -7.40714133e-01 -8.32912385e-01 7.33751774e-01 -3.06132250e-02 -3.79517883e-01 3.39812070e-01 2.21287444e-01 -3.98691922e-01 -9.90695283e-02 -6.05849326e-01 1.04301214e+00 9.44389343e-01 -4.92840171e-01 -2.51450002e-01 6.73499346e-01 5.72116077e-01 -8.00196588e-01 -3.83802831e-01 7.08900616e-02 5.39791763e-01 -8.34558785e-01 9.33691680e-01 1.11094184e-01 4.58320439e-01 -3.99509281e-01 2.52774055e-03 -6.57033026e-01 1.47841364e-01 -7.87717283e-01 1.93247899e-01 9.49683547e-01 6.36950076e-01 -7.06048489e-01 8.93021166e-01 5.88037252e-01 -4.08677608e-01 -7.64279306e-01 -1.00102699e+00 -3.90790403e-01 -2.69148827e-01 -6.14662409e-01 3.87250692e-01 8.03100765e-01 4.69150916e-02 3.51599723e-01 -4.96284366e-01 2.68710613e-01 4.13990825e-01 2.52309740e-01 1.01065958e+00 -8.63018930e-01 -2.98430353e-01 -6.78368926e-01 -2.31335908e-01 -1.89966559e+00 -7.52767250e-02 -8.90954733e-01 2.04623476e-01 -1.22076249e+00 1.49563372e-01 -6.35740280e-01 2.60883987e-01 6.80903018e-01 -1.69847488e-01 1.10364288e-01 2.82455921e-01 4.33440983e-01 -8.19435596e-01 5.48781991e-01 1.09409571e+00 -1.96068600e-01 -1.89495385e-01 1.51280006e-02 -3.30555230e-01 6.03942871e-01 7.44814575e-01 -2.43934602e-01 -2.07035929e-01 -7.81245708e-01 3.71821046e-01 1.54729620e-01 3.76780480e-01 -7.84368992e-01 7.19942451e-01 5.73868789e-02 5.21933079e-01 -9.55493867e-01 4.95131165e-01 -8.81150842e-01 -1.26709476e-01 1.52603546e-02 -2.40546942e-01 -3.11850607e-02 3.57746035e-01 5.96298575e-01 1.73601389e-01 -2.50035584e-01 4.49423045e-01 2.54325122e-01 -6.09975517e-01 1.92849502e-01 -2.03572899e-01 -1.88447773e-01 9.90675390e-01 -6.92213178e-01 -4.33164656e-01 -1.11596696e-01 -3.04654956e-01 1.23255283e-01 5.32062113e-01 2.44987324e-01 7.21174181e-01 -6.88353121e-01 -5.72000504e-01 3.09721917e-01 -6.67612106e-02 3.11345190e-01 3.26229930e-01 1.02119148e+00 -7.06246614e-01 6.82405353e-01 3.06055605e-01 -9.84069943e-01 -1.43915141e+00 4.17634189e-01 2.41460383e-01 -2.25590974e-01 -1.30704331e+00 6.34209096e-01 4.78060424e-01 -1.91518456e-01 3.85927916e-01 -3.18618774e-01 2.61919439e-01 -1.62957847e-01 2.67771572e-01 3.55859905e-01 9.16241407e-02 -3.87272537e-01 -1.97105139e-01 7.29844034e-01 -3.29648286e-01 1.93905719e-02 9.52817917e-01 -1.58077151e-01 1.74536586e-01 3.96280140e-01 8.25132728e-01 1.35947511e-01 -1.46184075e+00 -1.73956558e-01 -2.86684241e-02 -5.43979049e-01 -4.13610078e-02 -7.78527558e-01 -7.67515600e-01 8.81232619e-01 2.24558845e-01 1.65765256e-01 1.03887618e+00 -2.55530387e-01 8.11786890e-01 5.42058945e-01 1.79247960e-01 -1.02486074e+00 2.93745063e-02 4.96534437e-01 7.56815076e-01 -1.21158946e+00 3.03701181e-02 -5.54625750e-01 -7.20209658e-01 1.22171354e+00 6.10491335e-01 6.22461215e-02 3.32547069e-01 5.98629653e-01 -2.30502218e-01 -1.36368737e-01 -7.39336848e-01 7.01461732e-02 2.03013152e-01 -4.75945137e-03 4.65381086e-01 1.89607292e-02 1.25266820e-01 2.49903888e-01 -1.36963606e-01 -3.80124092e-01 1.31559893e-01 6.89668655e-01 -2.92939901e-01 -7.80450344e-01 -4.54798341e-01 5.48816919e-01 -3.53918314e-01 -4.95002389e-01 -2.62452066e-01 8.89529288e-01 -4.88851488e-01 5.90470374e-01 1.67642131e-01 -3.42589110e-01 2.92421192e-01 9.73096341e-02 -5.25191100e-03 -4.21716601e-01 -6.24332964e-01 3.06790888e-01 -1.73469812e-01 -4.53742951e-01 2.90893376e-01 -5.30878782e-01 -1.37322557e+00 -4.58746612e-01 -5.53116739e-01 -1.66421011e-01 7.46791363e-01 8.74103844e-01 5.51453412e-01 3.11694175e-01 5.65903544e-01 -8.85765195e-01 -4.17805851e-01 -9.03807998e-01 -2.68297434e-01 1.54820666e-01 2.11744070e-01 -4.21544760e-01 -3.00470561e-01 2.56631255e-01]

[12.027835845947266, 2.271174907684326]

0c161e8b-782c-4007-a712-fa02046e4f31

parametrically-retargetable-decision-makers

2206.13477

https://arxiv.org/abs/2206.13477v2

https://arxiv.org/pdf/2206.13477v2.pdf

Parametrically Retargetable Decision-Makers Tend To Seek Power

If capable AI agents are generally incentivized to seek power in service of the objectives we specify for them, then these systems will pose enormous risks, in addition to enormous benefits. In fully observable environments, most reward functions have an optimal policy which seeks power by keeping options open and staying alive. However, the real world is neither fully observable, nor must trained agents be even approximately reward-optimal. We consider a range of models of AI decision-making, from optimal, to random, to choices informed by learning and interacting with an environment. We discover that many decision-making functions are retargetable, and that retargetability is sufficient to cause power-seeking tendencies. Our functional criterion is simple and broad. We show that a range of qualitatively dissimilar decision-making procedures incentivize agents to seek power. We demonstrate the flexibility of our results by reasoning about learned policy incentives in Montezuma's Revenge. These results suggest a safety risk: Eventually, retargetable training procedures may train real-world agents which seek power over humans.

['Prasad Tadepalli', 'Alexander Matt Turner']

2022-06-27

['montezumas-revenge']

['playing-games']

[ 7.88304433e-02 7.60183215e-01 -5.65945864e-01 -1.73198104e-01 -2.33545408e-01 -1.00006497e+00 8.06689382e-01 -3.46992075e-01 -8.93031836e-01 1.17091048e+00 2.51787066e-01 -5.78968167e-01 -4.54738975e-01 -7.03528941e-01 -5.34518898e-01 -6.37839019e-01 -5.85220814e-01 6.62821054e-01 -2.04212740e-01 -2.57540166e-01 3.12300116e-01 5.11022687e-01 -1.22480023e+00 -3.89321357e-01 9.29937840e-01 4.02551860e-01 3.81628075e-03 9.69313085e-01 5.72831810e-01 1.16376162e+00 -5.05058348e-01 -3.25600386e-01 5.98797977e-01 -2.18290344e-01 -8.83060873e-01 -1.26871821e-02 -1.78408831e-01 -8.95837009e-01 -3.55037302e-01 1.04694664e+00 3.27124834e-01 1.41790584e-01 6.32049441e-01 -1.56612778e+00 -9.67622101e-01 1.29798055e+00 -8.21641982e-02 8.40873867e-02 2.68817335e-01 8.31822693e-01 1.14623725e+00 1.35543570e-01 5.20742238e-01 1.42626703e+00 5.48920929e-02 1.26603293e+00 -1.46631193e+00 -3.39245677e-01 1.80630878e-01 -2.50821084e-01 -5.86888194e-01 -7.54623294e-01 4.44570303e-01 -6.58312201e-01 7.76935697e-01 5.13515234e-01 9.09374893e-01 1.38024282e+00 2.45619833e-01 7.72310495e-01 1.39183187e+00 -1.37106493e-01 6.04676843e-01 8.95424373e-03 -2.59406507e-01 4.93536830e-01 8.16286922e-01 1.01597524e+00 -3.21004510e-01 -6.01261139e-01 1.03883135e+00 4.65634651e-02 -1.55402005e-01 -6.14910960e-01 -1.49079001e+00 7.91396022e-01 2.84254521e-01 2.47203618e-01 -6.26472592e-01 6.40843868e-01 -3.08684334e-02 7.79795468e-01 -2.94098139e-01 1.50864732e+00 -4.22747940e-01 -2.45000198e-01 -1.58080742e-01 6.61862969e-01 9.54252183e-01 7.92958438e-01 6.21611118e-01 3.62516761e-01 -1.56930670e-01 3.46060693e-02 3.30864877e-01 6.38122976e-01 4.11686629e-01 -1.97015095e+00 2.64559418e-01 2.63099074e-01 8.21282864e-01 -6.52122080e-01 -4.76326823e-01 -4.82370406e-01 -3.92348856e-01 8.54699433e-01 4.44756448e-01 -8.05616260e-01 -4.24970359e-01 2.25754452e+00 -1.65471569e-01 -6.19128168e-01 4.38969225e-01 9.79418159e-01 -2.48884708e-01 4.92143780e-01 2.74140865e-01 -4.11178082e-01 9.46730793e-01 -4.51579034e-01 -3.97746980e-01 -3.77238244e-01 6.95323110e-01 6.81722015e-02 1.12156010e+00 1.97291270e-01 -1.48094273e+00 2.89081812e-01 -8.01957846e-01 3.59090179e-01 3.31628561e-01 -5.64942956e-01 1.15867591e+00 5.69477379e-01 -1.17926598e+00 6.90637648e-01 -7.52377033e-01 -3.67846727e-01 4.16241109e-01 3.47662896e-01 -1.23010285e-01 6.31447971e-01 -1.07687593e+00 8.43875349e-01 2.55197883e-01 -2.00045198e-01 -1.72580099e+00 -2.05709204e-01 -4.67473149e-01 4.34560239e-01 9.46247756e-01 -1.02451491e+00 1.71022463e+00 -1.33587646e+00 -1.65140164e+00 7.49657393e-01 5.90368390e-01 -5.96210778e-01 7.85605669e-01 2.38697953e-03 -4.06663716e-02 -7.57748112e-02 4.02597785e-01 7.53777623e-01 7.66197741e-01 -1.20242679e+00 -6.89540744e-01 -2.66562283e-01 7.63387740e-01 6.79638267e-01 -2.88405955e-01 5.19087762e-02 8.50911200e-01 -3.98896605e-01 -5.17425239e-01 -1.08875084e+00 -7.09141910e-01 -2.28538200e-01 -4.88346487e-01 -2.57452786e-01 -9.81017053e-02 2.07331285e-01 8.50545406e-01 -1.84831297e+00 -7.28027448e-02 6.27234578e-02 6.34511888e-01 -4.34683263e-01 -1.86325490e-01 2.75943935e-01 2.55062103e-01 6.17629945e-01 2.12017912e-03 2.91597128e-01 7.86766469e-01 2.80109137e-01 -3.05330127e-01 5.96289039e-01 1.32260993e-02 1.07656693e+00 -8.95804346e-01 -2.10983008e-01 -2.75685936e-01 -5.31134069e-01 -8.28800917e-01 2.91683167e-01 -4.32799160e-01 3.58096868e-01 -1.16697764e+00 5.83992541e-01 -1.67326331e-01 -2.05714688e-01 3.76473814e-01 8.72943342e-01 -3.34180832e-01 2.23596752e-01 -7.67473102e-01 1.05955887e+00 1.30519524e-01 4.42570090e-01 3.56088281e-01 -7.46881425e-01 1.94595233e-01 3.71542960e-01 3.77933085e-01 -4.30600792e-01 4.49750006e-01 3.57403725e-01 7.34245718e-01 -5.04452229e-01 5.01060188e-01 -3.38276029e-01 -4.18136209e-01 8.85378242e-01 -3.64871413e-01 -1.84538946e-01 -4.99182865e-02 4.40046728e-01 1.37286377e+00 -7.21967965e-02 4.49429899e-01 -7.73372233e-01 -7.00372458e-02 4.68277603e-01 9.77484226e-01 1.26185894e+00 -6.08819902e-01 -3.57610732e-01 9.40914690e-01 -3.99533480e-01 -1.30443013e+00 -9.49492574e-01 1.74137652e-01 1.26764309e+00 1.81262583e-01 2.61660129e-01 -7.66491234e-01 -4.42255020e-01 3.01242441e-01 1.03522766e+00 -7.13874698e-01 -1.01846501e-01 -4.39108163e-01 -2.97861546e-01 5.31502545e-01 2.69523889e-01 5.01223028e-01 -1.25458074e+00 -1.45445502e+00 3.57336141e-02 3.70731622e-01 -4.17196304e-01 -6.91785336e-01 1.40513778e-01 -7.62280881e-01 -7.74621904e-01 -8.22406650e-01 -6.51787817e-02 7.03061759e-01 8.23843777e-02 9.42430735e-01 1.97720498e-01 4.96945046e-02 8.62254083e-01 1.75396711e-01 -4.34206069e-01 -6.67538404e-01 -1.35862172e-01 6.71598315e-01 -4.31324720e-01 1.57304704e-01 -5.50418198e-01 -8.09278131e-01 3.03342819e-01 -6.02515161e-01 -1.44548491e-01 7.34926462e-01 6.44596100e-01 -1.63175225e-01 -8.77648592e-03 7.04169393e-01 -7.25790501e-01 1.12824106e+00 -6.96045935e-01 -9.80146825e-01 2.29585826e-01 -7.60918677e-01 3.01015586e-01 5.75505495e-01 -6.83634043e-01 -1.12971091e+00 -2.89983362e-01 7.29331613e-01 -8.04143026e-02 -1.25893906e-01 -9.95337963e-04 -2.32316419e-01 7.30544254e-02 1.20079172e+00 -1.42496049e-01 1.43099025e-01 -1.49604484e-01 5.19782305e-01 4.10756648e-01 3.19830507e-01 -1.27154374e+00 8.50321472e-01 1.15612783e-01 9.35695097e-02 -4.48826551e-01 -1.20140873e-01 4.97603089e-01 3.28678399e-01 -1.53126583e-01 7.76935101e-01 -6.86214983e-01 -1.58633733e+00 3.30191493e-01 -7.63901770e-01 -7.47667849e-01 -7.55783319e-01 4.91209924e-01 -1.06787467e+00 9.48977843e-02 -4.85887796e-01 -1.25125992e+00 1.72477379e-01 -1.19352615e+00 3.30337465e-01 5.41008294e-01 -2.01595128e-01 -6.88422322e-01 1.79223239e-01 7.76569918e-02 8.12373281e-01 7.58668222e-03 8.15741479e-01 -8.99580777e-01 -1.09418273e+00 3.34219337e-01 3.15466434e-01 -2.78183997e-01 -1.03086367e-01 -4.29742411e-02 -6.79324269e-01 -3.74654204e-01 1.05931565e-01 -5.80718100e-01 2.58028746e-01 5.57934940e-01 6.60250366e-01 -1.37130845e+00 -3.91099513e-01 2.39030838e-01 9.35248971e-01 7.25854874e-01 1.12530708e-01 5.60563028e-01 1.15995258e-01 8.27086568e-01 3.87910008e-01 6.46502614e-01 3.52839291e-01 2.68926978e-01 5.75766683e-01 3.08376521e-01 8.05044055e-01 -3.82474661e-01 5.08373916e-01 -1.38286620e-01 -4.13873255e-01 -2.55691677e-01 -8.21519136e-01 5.14278412e-01 -2.02327824e+00 -1.45396733e+00 6.17748797e-01 2.29471755e+00 1.05331779e+00 3.42828363e-01 6.55696571e-01 -4.84720767e-01 5.99544525e-01 -1.08990066e-01 -1.27869105e+00 -5.37224710e-01 -1.03343554e-01 -6.18480802e-01 8.25732529e-01 6.78840995e-01 -6.07639849e-01 7.35609293e-01 7.82433987e+00 3.61243695e-01 -6.89754367e-01 5.69164939e-02 9.98324513e-01 -4.71421272e-01 -9.09180522e-01 2.61923015e-01 -2.68619925e-01 2.42662966e-01 8.89169395e-01 -8.34445417e-01 9.17047560e-01 1.20044649e+00 3.04104000e-01 -1.28463760e-01 -1.53516078e+00 5.31504691e-01 -6.91154838e-01 -1.14372301e+00 -4.44617271e-01 4.78268236e-01 7.12015927e-01 -2.31867850e-01 1.03285648e-01 3.25818777e-01 1.56873119e+00 -1.34649515e+00 1.03344905e+00 4.32639241e-01 6.76480591e-01 -5.69219053e-01 1.33083791e-01 8.08848798e-01 -1.92910269e-01 -7.16265500e-01 -3.45865309e-01 -6.04106784e-01 -1.75881177e-01 5.56820072e-02 -7.05995142e-01 -3.56901318e-01 2.97623068e-01 1.28444448e-01 3.41545083e-02 5.26684821e-01 -2.62908876e-01 2.78209865e-01 -2.84626693e-01 -5.05985677e-01 3.83569598e-01 -2.61974812e-01 9.32338238e-01 4.32947010e-01 1.68534979e-01 6.17726386e-01 1.27148747e-01 1.36772752e+00 8.12826026e-03 -4.02638257e-01 -9.85899746e-01 -7.16720581e-01 6.97159648e-01 9.72001374e-01 -3.46153617e-01 -2.60522872e-01 1.29694939e-01 5.30609369e-01 2.06307217e-01 5.92403352e-01 -3.96882892e-01 -9.89332870e-02 1.06751335e+00 -1.51800334e-01 -5.31974673e-01 -1.40097082e-01 -2.63198167e-01 -1.33920658e+00 -4.08288121e-01 -1.09246671e+00 1.70041934e-01 -7.09096134e-01 -1.28528392e+00 2.01273844e-01 1.89520761e-01 -5.83898067e-01 -5.58387995e-01 -4.19322789e-01 -4.70535070e-01 5.87436616e-01 -1.05816126e+00 -5.45041263e-01 4.74428594e-01 3.59067291e-01 3.95604521e-01 -4.45623726e-01 4.39384907e-01 -4.45105314e-01 -5.37230372e-01 3.34711641e-01 -1.24248937e-01 -1.85734689e-01 5.17344028e-02 -1.46301353e+00 2.80397445e-01 7.97115326e-01 -3.04361284e-01 8.51176918e-01 9.70345080e-01 -4.86646652e-01 -1.77206481e+00 -4.40939367e-01 2.57005125e-01 -6.56417549e-01 8.80110562e-01 -1.29683301e-01 -4.02443469e-01 9.75529373e-01 3.02536517e-01 -6.51384473e-01 1.20478868e-01 1.35775775e-01 -3.30515690e-02 1.96401775e-01 -1.23754370e+00 1.27594280e+00 1.46381044e+00 -2.60404944e-01 -8.33558917e-01 2.68459767e-01 1.05112970e+00 9.89838690e-02 -5.07523656e-01 9.99576449e-02 5.87050498e-01 -9.34761763e-01 6.35663927e-01 -1.11480510e+00 3.80729049e-01 -3.57971922e-03 -1.39667541e-01 -1.36273587e+00 -6.89451098e-01 -1.54384863e+00 2.24755988e-01 6.96958244e-01 5.13101459e-01 -1.16424465e+00 7.41542518e-01 1.35250759e+00 9.67946947e-02 -1.79808050e-01 -8.51818800e-01 -8.85649860e-01 4.86092508e-01 1.18195869e-01 8.54423404e-01 1.07824051e+00 5.54798603e-01 2.52042979e-01 -4.43883538e-01 -1.00864423e-02 1.06528330e+00 -9.08606872e-02 6.03428602e-01 -1.07632852e+00 -7.47310698e-01 -9.21313226e-01 1.56219490e-02 -1.03702819e+00 8.91905501e-02 -4.85474795e-01 1.23925038e-01 -1.10940373e+00 3.55403900e-01 -9.92007017e-01 -5.30154631e-02 7.05101073e-01 9.43820029e-02 -7.99757779e-01 4.08269465e-01 7.40010440e-01 -5.43195367e-01 3.92277390e-01 1.36256576e+00 -6.41940013e-02 -2.08185583e-01 1.14101849e-01 -1.35688019e+00 8.47849131e-01 9.81756270e-01 -3.61432403e-01 -8.37372541e-01 -3.29824597e-01 6.30820572e-01 5.60925126e-01 4.55575138e-01 -4.73692000e-01 1.16807528e-01 -1.33555579e+00 2.69736089e-02 3.48476410e-01 -6.34885132e-02 -8.85852158e-01 3.93804342e-01 9.09022868e-01 -1.01187205e+00 7.95577988e-02 -5.11522770e-01 4.44982409e-01 7.15256035e-01 -3.16225380e-01 6.16658092e-01 -3.98272634e-01 -4.86269206e-01 2.18294874e-01 -9.30775046e-01 1.35223612e-01 1.31371665e+00 -4.53757793e-02 -8.07120740e-01 -7.18831539e-01 -4.79846239e-01 6.67581260e-01 9.88373518e-01 -9.75314528e-02 2.38341361e-01 -9.32739973e-01 -7.01644778e-01 -1.52545437e-01 -7.57877901e-02 -2.31026188e-01 -1.46438673e-04 3.26101422e-01 -2.86543787e-01 4.11063612e-01 -5.53310871e-01 -7.60285407e-02 -6.28733158e-01 8.62512708e-01 7.84731150e-01 1.04263127e-01 -3.12751204e-01 6.43432736e-01 6.91832244e-01 -2.82396883e-01 2.24733859e-01 -3.37430239e-01 -6.75943419e-02 -4.65677083e-01 3.16052258e-01 3.36283088e-01 -9.94521260e-01 8.50944519e-02 -2.06017360e-01 8.40317607e-02 3.13163660e-02 -7.61578619e-01 9.87196624e-01 -2.86228627e-01 1.14072032e-01 6.20139800e-02 2.01034173e-01 2.78336965e-02 -1.68120492e+00 -3.11361551e-02 -9.91510674e-02 -6.19636774e-01 -3.03367347e-01 -9.59097326e-01 -5.55385649e-01 4.07368422e-01 6.92525357e-02 8.67437303e-01 8.45357001e-01 -3.89177054e-02 -1.70380098e-03 9.86234903e-01 8.01883280e-01 -1.11861694e+00 4.25610282e-02 5.00229038e-02 7.81836927e-01 -9.96582747e-01 -2.13969707e-01 3.24882895e-01 -1.00361145e+00 5.38194656e-01 5.96682429e-01 -2.21345305e-01 1.24228582e-01 1.92304149e-01 -1.75256759e-01 -2.78389573e-01 -1.31636441e+00 -1.64937764e-01 -4.43489432e-01 7.75505364e-01 -2.13082850e-01 6.52687073e-01 -3.27090591e-01 5.11837840e-01 -3.45883340e-01 -1.23239122e-02 9.80661333e-01 9.29108679e-01 -1.04688430e+00 -9.10201550e-01 -4.60278779e-01 3.93374115e-01 -5.39940715e-01 1.67328194e-01 -6.65019453e-01 6.68382883e-01 -3.30944508e-01 1.02520859e+00 7.97587335e-02 1.03664715e-02 3.55215147e-02 -2.84029067e-01 2.76581973e-01 -4.24821407e-01 -3.44304025e-01 1.06428368e-02 2.36607730e-01 -7.05625296e-01 -3.20693552e-01 -7.18155563e-01 -1.10875559e+00 -8.38159323e-01 1.02569118e-01 2.72338897e-01 2.75284439e-01 6.27646983e-01 2.85453528e-01 2.15526417e-01 9.19155419e-01 -3.87483418e-01 -1.43921387e+00 -5.77069342e-01 -8.28393579e-01 1.51225671e-01 6.90467954e-01 -5.42219996e-01 -6.32664979e-01 -3.56417239e-01]

[4.163037300109863, 2.4064903259277344]

eb01338c-382a-4dfb-aa93-dbd3e16e4a05

machine-love

2302.09248

https://arxiv.org/abs/2302.09248v2

https://arxiv.org/pdf/2302.09248v2.pdf

Machine Love

While ML generates much economic value, many of us have problematic relationships with social media and other ML-powered applications. One reason is that ML often optimizes for what we want in the moment, which is easy to quantify but at odds with what is known scientifically about human flourishing. Thus, through its impoverished models of us, ML currently falls far short of its exciting potential, which is for it to help us to reach ours. While there is no consensus on defining human flourishing, from diverse perspectives across psychology, philosophy, and spiritual traditions, love is understood to be one of its primary catalysts. Motivated by this view, this paper explores whether there is a useful conception of love fitting for machines to embody, as historically it has been generative to explore whether a nebulous concept, such as life or intelligence, can be thoughtfully abstracted and reimagined, as in the fields of machine intelligence or artificial life. This paper forwards a candidate conception of machine love, inspired in particular by work in positive psychology and psychotherapy: to provide unconditional support enabling humans to autonomously pursue their own growth and development. Through proof of concept experiments, this paper aims to highlight the need for richer models of human flourishing in ML, provide an example framework through which positive psychology can be combined with ML to realize a rough conception of machine love, and demonstrate that current language models begin to enable embodying qualitative humanistic principles. The conclusion is that though at present ML may often serve to addict, distract, or divide us, an alternative path may be opening up: We may align ML to support our growth, through it helping us to align ourselves towards our highest aspirations.

['Joel Lehman']

2023-02-18

['artificial-life', 'philosophy']

['miscellaneous', 'miscellaneous']

[ 3.03396005e-02 6.46958530e-01 -5.34595490e-01 -2.39871949e-01 2.21400574e-01 -2.22565636e-01 7.65923500e-01 2.88943022e-01 -3.77989262e-01 4.09288675e-01 6.71146214e-01 -4.87196267e-01 -2.32248828e-01 -8.23079348e-01 -1.54820368e-01 -3.42882276e-01 1.67865589e-01 4.98943418e-01 -7.22998142e-01 -7.71009266e-01 3.17951173e-01 3.87954533e-01 -1.34963059e+00 -3.80804300e-01 9.97782826e-01 2.03110099e-01 7.99023286e-02 2.27573588e-01 -2.08038151e-01 1.18248391e+00 -7.77756870e-02 -6.92191005e-01 -2.87859201e-01 -7.95313776e-01 -8.77762675e-01 -7.98800290e-02 -2.77354658e-01 -3.75835180e-01 1.06494278e-01 7.04745352e-01 3.06415081e-01 -2.18599543e-01 4.85041708e-01 -1.22622561e+00 -1.22852135e+00 7.27084875e-01 -2.44203463e-01 -2.66664654e-01 3.20390701e-01 6.09568894e-01 8.74870837e-01 -3.06589961e-01 6.29849076e-01 1.36916733e+00 6.70697689e-01 5.97787559e-01 -1.26919150e+00 -3.67880732e-01 -3.62296492e-01 -2.79628426e-01 -9.19179678e-01 -7.15603054e-01 7.13780105e-01 -5.30615389e-01 7.66231060e-01 3.60845536e-01 1.59305418e+00 9.28749442e-01 3.98734272e-01 5.58264136e-01 9.71860826e-01 -5.53243935e-01 -3.70251760e-02 4.24304515e-01 1.09026276e-01 6.54949844e-01 5.65765202e-01 1.62691157e-02 -5.13265789e-01 2.15987097e-02 8.29801977e-01 -2.51824945e-01 -1.58835560e-01 -4.54730354e-02 -1.17809272e+00 8.66268516e-01 2.48093903e-01 1.04782891e+00 -4.76681888e-01 1.57542124e-01 2.22615942e-01 3.16034764e-01 3.51806343e-01 8.98772657e-01 2.10052803e-01 -8.05584550e-01 -5.00737786e-01 4.02247310e-01 9.48825300e-01 2.92693879e-02 3.82043779e-01 -6.31849542e-02 2.89070785e-01 8.02064776e-01 5.05458713e-01 4.35071230e-01 2.86260128e-01 -1.17445934e+00 -4.73755330e-01 8.57960701e-01 2.46642139e-02 -1.26396894e+00 -5.24581671e-01 -5.00351787e-01 -3.91959041e-01 4.16101426e-01 3.84414464e-01 2.56401533e-03 7.22871646e-02 1.87060583e+00 2.62815319e-02 -3.82033288e-01 8.18933360e-03 9.91580367e-01 5.77884495e-01 3.30245614e-01 4.91697699e-01 -2.50854999e-01 9.59988534e-01 -4.14400011e-01 -6.85648620e-01 -7.14633524e-01 9.48931396e-01 -6.71543956e-01 1.44970369e+00 4.29023772e-01 -1.42153251e+00 8.96324739e-02 -1.14361453e+00 -3.75138789e-01 -2.99993102e-02 -5.72668076e-01 1.42751932e+00 1.01950514e+00 -1.16947651e+00 8.89901280e-01 -6.62089288e-01 -8.74818861e-01 4.81524199e-01 1.17735527e-01 -1.97194442e-01 1.20010532e-01 -8.93872201e-01 1.31202507e+00 7.56125972e-02 9.39597413e-02 -1.04943343e-01 -3.59222144e-01 -7.05388844e-01 -1.70405015e-01 1.63352117e-01 -1.24816597e+00 8.87346566e-01 -1.64222062e+00 -1.15411162e+00 1.39859021e+00 1.73692871e-02 -4.23662901e-01 2.62306690e-01 -2.17580050e-01 -4.28551137e-01 6.25203177e-02 1.69673607e-01 6.11054361e-01 2.13693693e-01 -1.34673178e+00 1.83839872e-02 -6.21946216e-01 1.84067115e-01 2.29005456e-01 -7.02502728e-01 3.65718037e-01 3.69114101e-01 -2.77460784e-01 2.65894890e-01 -5.88820517e-01 -5.36205210e-02 1.09407634e-01 -2.22692993e-02 -2.38041997e-01 4.48964566e-01 -3.30524087e-01 1.24600768e+00 -2.03072405e+00 -8.36042035e-03 -1.37759387e-01 8.55330825e-01 -1.55511141e-01 2.09486395e-01 1.00097573e+00 3.14179659e-01 5.44506311e-01 5.55557646e-02 -2.37160549e-01 2.61767119e-01 4.27499324e-01 -7.98845738e-02 6.65073156e-01 2.69036591e-01 1.08091676e+00 -1.21342909e+00 -5.17140031e-01 5.51028587e-02 8.04662526e-01 -6.48180902e-01 -2.00319737e-01 2.38619968e-01 2.73372531e-01 -3.61780524e-01 6.43512845e-01 1.94180191e-01 -4.68062162e-01 4.44558382e-01 4.84518856e-01 -1.90411851e-01 1.51440352e-01 -4.07870710e-01 1.26361716e+00 -1.94535241e-01 5.95676899e-01 3.21468949e-01 -9.48005974e-01 9.72296894e-01 1.56940579e-01 5.49859345e-01 -9.20853853e-01 5.55949748e-01 2.58019835e-01 5.31538248e-01 -6.14543974e-01 6.09898925e-01 -9.95490968e-01 2.48348251e-01 7.84314394e-01 -3.91841263e-01 -5.89757204e-01 -3.92510653e-01 2.74112254e-01 8.26315999e-01 3.74847054e-01 1.96715236e-01 -5.07619023e-01 1.02524117e-01 6.42074421e-02 4.22690690e-01 1.24451771e-01 -4.25281674e-01 -7.42161050e-02 4.25823152e-01 -2.36001864e-01 -1.02102327e+00 -1.12106001e+00 -3.24634999e-01 9.77364659e-01 1.96188629e-01 -3.03364933e-01 -6.83983445e-01 1.69669688e-01 -2.85101961e-02 1.29324412e+00 -3.42747986e-01 -2.69663811e-01 -1.96796790e-01 -5.93048871e-01 4.45797354e-01 -1.76226661e-01 2.84254640e-01 -1.20921099e+00 -1.18440771e+00 2.01832369e-01 7.28899315e-02 -6.08817756e-01 4.51459259e-01 -5.20607643e-02 -7.89796174e-01 -6.55486941e-01 -3.02733511e-01 -5.25190353e-01 3.38543534e-01 2.62066633e-01 1.21044743e+00 6.88998520e-01 1.65779877e-03 6.05123818e-01 -2.49301016e-01 -6.41063809e-01 -8.78879547e-01 -3.27172607e-01 2.03832060e-01 -4.32769388e-01 6.34595692e-01 -1.07774031e+00 -3.86923909e-01 -3.47660303e-01 -9.38447177e-01 5.21362126e-01 3.36954534e-01 5.61052263e-01 -2.31155500e-01 -3.13177735e-01 1.24961007e+00 -8.14818382e-01 7.21690655e-01 -8.51236939e-01 5.43939948e-01 1.72899980e-02 -9.33782816e-01 -3.32316458e-01 1.45542234e-01 -3.12128276e-01 -8.77782464e-01 -7.76651442e-01 -2.00175911e-01 1.16902232e-01 1.54236585e-01 5.89719594e-01 4.72486056e-02 1.09704033e-01 7.78284192e-01 -1.30654767e-01 7.37197459e-01 -2.84513772e-01 5.73738158e-01 6.50564909e-01 5.58169067e-01 -9.28193033e-01 5.17588556e-01 4.50608045e-01 -1.66260436e-01 -1.12807190e+00 -6.40462995e-01 -5.55707514e-02 -3.27478319e-01 -5.58471620e-01 9.74527955e-01 -6.73224211e-01 -1.16076767e+00 9.50926989e-02 -7.13641524e-01 -4.46223944e-01 -5.42313159e-01 4.43376184e-01 -8.43247294e-01 1.95917338e-01 -6.73471630e-01 -1.30128121e+00 -2.61015475e-01 -5.75761259e-01 5.36461532e-01 3.76448691e-01 -8.64606738e-01 -1.31065667e+00 -2.32660547e-01 7.27718651e-01 6.05170429e-01 6.36501372e-01 1.22021854e+00 -5.74356675e-01 -1.00352950e-01 -4.81203943e-02 1.76794410e-01 2.47989357e-01 1.75324157e-01 -1.85439419e-02 -8.82979870e-01 4.31813337e-02 6.67749763e-01 -7.54442930e-01 2.85277963e-02 -1.90178588e-01 2.28267208e-01 -5.56422532e-01 -2.35638604e-01 1.68208688e-01 1.47064757e+00 2.34945491e-01 9.13813472e-01 4.83205676e-01 4.42829907e-01 1.11229861e+00 6.18089080e-01 4.49402481e-01 7.70549476e-01 1.62210688e-01 1.90811247e-01 -2.25179374e-01 2.27147892e-01 -4.13377494e-01 2.72878140e-01 1.00653505e+00 -5.16568013e-02 2.14572072e-01 -1.29774535e+00 3.44120741e-01 -1.68884885e+00 -1.26880860e+00 -3.26127261e-01 2.13445187e+00 9.86038983e-01 5.03484488e-01 2.75068194e-01 8.97751972e-02 1.45932734e-01 1.15530655e-01 -2.72153169e-01 -9.36347783e-01 -8.09372738e-02 8.90082046e-02 -1.84190333e-01 4.93362308e-01 -1.80111915e-01 8.47364604e-01 6.63181448e+00 3.44519258e-01 -1.13896179e+00 3.95469703e-02 1.03179109e+00 -2.23307833e-01 -9.32664871e-01 2.45854795e-01 -1.51159510e-01 4.04467508e-02 9.33176279e-01 -5.59341550e-01 5.35884202e-01 4.79222149e-01 6.06492281e-01 -3.41910213e-01 -1.21908092e+00 7.38748074e-01 1.70950159e-01 -9.86962438e-01 -4.60222423e-01 3.78455073e-01 1.56793445e-01 -1.83222488e-01 -2.96101421e-02 1.03918701e-01 2.42200404e-01 -1.56150961e+00 9.32597160e-01 5.45642078e-01 3.65213424e-01 -6.73805535e-01 2.96331644e-01 8.15881848e-01 -1.27486840e-01 -4.96575385e-02 -1.09850802e-01 -1.05173409e+00 1.57005489e-01 4.84676540e-01 -6.54701769e-01 -1.32775516e-03 5.14726490e-02 5.36725163e-01 -3.00814360e-01 4.43821937e-01 1.66998789e-01 4.67159122e-01 -1.86733857e-01 -2.93521672e-01 1.76294982e-01 -5.94996929e-01 6.11641586e-01 7.55635500e-01 4.57916372e-02 4.10699338e-01 -1.96094126e-01 1.25662220e+00 4.69194382e-01 5.15641570e-01 -1.03724372e+00 -9.12050664e-01 4.32185173e-01 1.20974779e+00 -7.39517629e-01 1.36514641e-02 -3.96089107e-01 6.73234463e-01 1.23801030e-01 -5.19505376e-03 -6.51696801e-01 2.13261172e-01 6.20690763e-01 5.88138998e-01 -8.04672897e-01 -3.16450745e-01 -9.73281741e-01 -1.00433779e+00 -4.74880785e-01 -9.00122106e-01 -4.02248025e-01 -6.80959165e-01 -1.17168999e+00 -1.41496196e-01 -4.49798495e-01 -4.36660945e-01 -2.63150722e-01 -9.24114063e-02 -5.69704533e-01 7.36627340e-01 -8.44984412e-01 -1.62696624e+00 1.47031128e-01 -2.74833068e-02 -1.21866325e-02 4.70550001e-01 7.97027528e-01 -9.05279443e-02 -2.94861585e-01 3.78572106e-01 -2.07412139e-01 -3.35247874e-01 4.03496027e-01 -7.06345260e-01 1.35280132e-01 4.42432910e-01 6.63297698e-02 9.70325708e-01 9.22197461e-01 -6.75058603e-01 -1.93980587e+00 -5.94346561e-02 1.24348438e+00 -6.06679320e-01 1.00425720e+00 -1.25967950e-01 -8.05421889e-01 6.77051425e-01 2.17978805e-01 -8.40115309e-01 7.83596575e-01 5.43630898e-01 -1.06269881e-01 9.96779874e-02 -1.29961443e+00 1.01894140e+00 1.20589471e+00 -4.27896202e-01 -4.82686609e-01 3.14580798e-01 8.83652210e-01 2.06414312e-01 -1.06991661e+00 2.00606529e-02 7.18397021e-01 -1.37479925e+00 8.70117188e-01 -2.60729939e-01 8.15843880e-01 2.86546379e-01 -6.39225841e-02 -7.26351976e-01 -2.76968330e-01 -6.99935496e-01 5.04351199e-01 1.78623998e+00 1.03465535e-01 -1.03050375e+00 6.26846015e-01 1.09979165e+00 -2.84703344e-01 -1.14337599e+00 -7.43703485e-01 -3.90686333e-01 6.59808874e-01 -7.36406326e-01 5.74058473e-01 1.48890507e+00 8.71327043e-01 7.51427293e-01 -6.88216239e-02 -5.72038591e-01 5.27976990e-01 -3.01050276e-01 8.56624246e-01 -1.49891329e+00 -5.15270352e-01 -8.97385299e-01 -2.23627135e-01 -2.70374119e-01 -1.05072692e-01 -1.03020310e+00 -2.49824107e-01 -1.77060890e+00 3.63793433e-01 -7.00412095e-01 2.86374778e-01 3.88058066e-01 3.71002167e-01 1.30693063e-01 6.16199434e-01 3.21537346e-01 3.37216258e-02 2.98586607e-01 1.48924017e+00 3.71156782e-01 -5.08401811e-01 -7.43440032e-01 -1.79995453e+00 1.00812256e+00 7.88399637e-01 2.10660566e-02 -5.19023716e-01 1.02231242e-01 8.64373028e-01 3.85165989e-01 5.72718859e-01 -8.89155686e-01 -1.57102585e-01 -6.66356206e-01 1.99113652e-01 2.02855751e-01 4.60180402e-01 -6.68827653e-01 5.01501501e-01 6.51241541e-01 -1.39508480e-02 -4.46384162e-01 -3.73056382e-02 -2.22810224e-01 3.86185050e-01 -2.39961714e-01 6.82230890e-01 4.85645258e-04 -2.77626604e-01 -3.67828161e-01 -2.62811542e-01 2.61714924e-02 9.47342157e-01 -6.01021647e-01 -5.81002295e-01 -5.83561063e-01 -5.40203571e-01 2.15740755e-01 1.26707184e+00 4.01065648e-01 4.37287271e-01 -1.15387404e+00 -5.67873955e-01 -2.03603625e-01 3.25219333e-02 -4.90238458e-01 1.33252308e-01 9.13014293e-01 -4.72556412e-01 1.68206647e-01 -2.60698676e-01 -2.16887519e-01 -7.92014241e-01 5.76384187e-01 1.76009238e-01 3.65021914e-01 -7.96246946e-01 6.50388956e-01 3.52256507e-01 1.26855865e-01 -2.61926323e-01 1.86651826e-01 -1.81179389e-01 -4.72176112e-02 4.60218430e-01 1.64418966e-01 -7.61183143e-01 -8.24363589e-01 -2.08479241e-01 2.58193433e-01 5.03180385e-01 -1.67060912e-01 1.53657913e+00 -3.08905542e-01 -7.38548279e-01 8.21215570e-01 7.59363294e-01 2.19583675e-01 -6.50762677e-01 4.77542043e-01 -7.11052343e-02 -5.88650346e-01 -2.48598233e-01 -1.03003776e+00 -3.01430643e-01 8.44044328e-01 1.80763558e-01 3.77358437e-01 1.03909421e+00 2.98460454e-01 7.42023945e-01 -4.90402468e-02 4.95239794e-01 -9.41468954e-01 -1.31075475e-02 -1.39966518e-01 9.38776195e-01 -9.80402708e-01 4.03149799e-02 -1.91395834e-01 -6.90594316e-01 9.49050426e-01 3.87523383e-01 -3.30304354e-02 4.14330363e-01 2.76514888e-01 -1.73815429e-01 -4.08165902e-01 -7.90444076e-01 -1.68176010e-01 -4.17719007e-01 7.21503317e-01 1.00762391e+00 4.11003977e-01 -1.12512076e+00 6.22524500e-01 -6.98158741e-01 4.11525726e-01 6.66839957e-01 5.96426606e-01 -1.08893836e+00 -1.36619496e+00 -3.80000621e-01 4.02412176e-01 -6.68859482e-01 4.58155796e-02 -7.02601314e-01 9.76549506e-01 4.79329377e-01 1.14163220e+00 2.79881619e-02 -3.20032567e-01 -2.88047463e-01 1.10776342e-01 6.84600592e-01 -5.10250807e-01 -9.37284708e-01 7.75954798e-02 5.07259786e-01 -2.43617415e-01 -4.08662081e-01 -6.99353933e-01 -1.58101058e+00 -1.16613591e+00 1.13576524e-01 -1.38450980e-01 5.75968683e-01 8.54782403e-01 4.89501506e-02 -2.43067387e-02 1.59923404e-01 -5.27008772e-01 -2.34468415e-01 -4.82234329e-01 -7.33773708e-01 1.70535445e-01 -6.46052824e-04 -3.73892546e-01 -2.36764848e-01 -2.85112262e-01]

[9.046924591064453, 6.318998336791992]

b946326e-3ad2-40b5-8bb9-58d340536b3b

a-new-dataset-and-transformer-for

2204.10039

https://arxiv.org/abs/2204.10039v1

https://arxiv.org/pdf/2204.10039v1.pdf

A New Dataset and Transformer for Stereoscopic Video Super-Resolution

Stereo video super-resolution (SVSR) aims to enhance the spatial resolution of the low-resolution video by reconstructing the high-resolution video. The key challenges in SVSR are preserving the stereo-consistency and temporal-consistency, without which viewers may experience 3D fatigue. There are several notable works on stereoscopic image super-resolution, but there is little research on stereo video super-resolution. In this paper, we propose a novel Transformer-based model for SVSR, namely Trans-SVSR. Trans-SVSR comprises two key novel components: a spatio-temporal convolutional self-attention layer and an optical flow-based feed-forward layer that discovers the correlation across different video frames and aligns the features. The parallax attention mechanism (PAM) that uses the cross-view information to consider the significant disparities is used to fuse the stereo views. Due to the lack of a benchmark dataset suitable for the SVSR task, we collected a new stereoscopic video dataset, SVSR-Set, containing 71 full high-definition (HD) stereo videos captured using a professional stereo camera. Extensive experiments on the collected dataset, along with two other datasets, demonstrate that the Trans-SVSR can achieve competitive performance compared to the state-of-the-art methods. Project code and additional results are available at https://github.com/H-deep/Trans-SVSR/

['Lai-Kuan Wong', 'Md Baharul Islam', 'Hassan Imani']

2022-04-21

['video-super-resolution']

['computer-vision']

[ 2.77495444e-01 -4.69007879e-01 -1.22696437e-01 -1.96288243e-01 -8.08317840e-01 -1.36425197e-01 2.49510854e-01 -8.98424089e-01 2.95357071e-02 8.04545760e-01 7.46336639e-01 1.35543302e-01 3.59065272e-02 -6.64615035e-01 -8.01747978e-01 -6.58630311e-01 1.38297752e-01 -4.11446124e-01 7.23771274e-01 -3.80424798e-01 4.07280207e-01 3.49624962e-01 -1.94114661e+00 7.75010884e-01 9.34108019e-01 9.97716546e-01 7.06958175e-01 6.86055481e-01 2.18222111e-01 1.09446228e+00 -5.71446307e-02 1.04098044e-01 5.01605511e-01 -5.10185301e-01 -8.33119988e-01 -6.53439909e-02 8.79898250e-01 -9.62881267e-01 -8.42935383e-01 1.15086091e+00 6.35554492e-01 2.45772928e-01 -1.42195567e-01 -5.73848963e-01 -8.86042953e-01 3.02774429e-01 -8.14621150e-01 1.03437257e+00 7.01403141e-01 2.63126969e-01 6.29196644e-01 -9.70687211e-01 8.72197151e-01 1.33965635e+00 1.92869946e-01 6.60338640e-01 -1.05703735e+00 -9.28221285e-01 -3.54120508e-02 5.68429708e-01 -1.29820049e+00 -7.04211593e-01 5.78183830e-01 -2.88595259e-01 6.81736290e-01 1.63380444e-01 5.74810982e-01 1.12036955e+00 1.58739701e-01 4.71537024e-01 1.30202973e+00 6.55432343e-02 -7.99926594e-02 -2.83333004e-01 -8.61233249e-02 3.55474174e-01 5.67364041e-03 7.36875951e-01 -9.19412851e-01 3.35638195e-01 1.61034882e+00 7.63820931e-02 -8.67908657e-01 -2.70689189e-01 -1.22151947e+00 2.94253230e-01 4.67567146e-01 4.23751354e-01 -3.04633796e-01 -2.49535576e-01 2.37992719e-01 1.88692182e-01 5.66728652e-01 1.55847237e-01 -2.81927049e-01 -4.20293957e-03 -9.02624786e-01 1.69490412e-01 3.12711969e-02 9.27629113e-01 6.04434967e-01 2.59578526e-01 -2.56124109e-01 8.06196332e-01 -8.62938762e-02 2.87710130e-01 5.53393781e-01 -1.56062353e+00 5.45985281e-01 1.54832482e-01 2.55837351e-01 -9.35704470e-01 -2.44280379e-02 -3.48775297e-01 -1.07347393e+00 4.56924498e-01 -6.99190199e-02 1.36338621e-01 -5.27331471e-01 1.60201788e+00 3.36021245e-01 6.38131559e-01 3.94674838e-02 1.61139357e+00 1.23580229e+00 7.06288040e-01 -3.28795224e-01 -5.41263521e-01 1.06289744e+00 -1.00705922e+00 -8.59033883e-01 5.92411086e-02 -3.56334865e-01 -6.95197344e-01 9.51605201e-01 2.85824358e-01 -1.58483803e+00 -1.03186095e+00 -1.05449200e+00 -6.55379236e-01 5.27837761e-02 -3.68970633e-01 3.03267241e-01 -6.02031536e-02 -1.31494963e+00 6.07019722e-01 -7.00671434e-01 -8.00992735e-03 4.38245654e-01 1.24335341e-01 -4.16367948e-01 -3.51602077e-01 -1.47922349e+00 5.73461711e-01 2.39469498e-01 -7.75317699e-02 -9.19088960e-01 -1.00551558e+00 -9.36453819e-01 3.86205837e-02 3.35167408e-01 -9.66446280e-01 9.94895577e-01 -1.05296886e+00 -1.50571227e+00 8.16212237e-01 -6.36357963e-01 -1.43871680e-01 4.70056057e-01 -2.54067451e-01 -6.78716362e-01 4.20710623e-01 2.32823610e-01 4.07391518e-01 9.80309010e-01 -1.36688972e+00 -1.04898190e+00 -4.40324962e-01 1.27885967e-01 5.90879381e-01 1.48788253e-02 1.80320144e-01 -6.33250356e-01 -6.17480278e-01 5.38150489e-04 -3.65606219e-01 -6.13015480e-02 -2.09944278e-01 -1.43502623e-01 2.55988210e-01 9.49685097e-01 -8.73238087e-01 1.22895336e+00 -2.07176161e+00 4.56844896e-01 -5.04320621e-01 5.41565180e-01 4.84660655e-01 -7.10559040e-02 -1.58631295e-01 -3.59813780e-01 -2.20046207e-01 2.81587224e-02 -1.70959324e-01 -7.58504748e-01 -1.34686738e-01 -5.30616641e-01 4.21999782e-01 -1.35704249e-01 6.68053150e-01 -1.00014079e+00 -5.09729803e-01 6.20198667e-01 1.07713580e+00 -4.41014349e-01 4.94200706e-01 2.50455469e-01 1.06876957e+00 -3.49541754e-01 5.62496006e-01 1.10083556e+00 -3.81921709e-01 -1.32185638e-01 -5.43303728e-01 -6.30717993e-01 2.54815191e-01 -1.17830253e+00 1.97720385e+00 -3.23536783e-01 6.65941775e-01 1.38588503e-01 -2.94854522e-01 5.76181829e-01 9.85556543e-02 6.22136474e-01 -1.37739921e+00 -4.64289375e-02 1.44565627e-01 -5.14880538e-01 -6.60424531e-01 7.90768445e-01 1.05123192e-01 5.10659635e-01 5.84663711e-02 -9.71276239e-02 3.47476065e-01 1.66394457e-01 1.64217263e-01 7.33642519e-01 3.62166822e-01 2.22212389e-01 -8.95654559e-02 9.91202414e-01 -4.05793667e-01 8.63273323e-01 2.29393482e-01 -1.70105159e-01 1.20914698e+00 5.39108831e-03 -6.29388630e-01 -1.23306918e+00 -1.31678319e+00 -2.62198329e-01 7.41399169e-01 7.47827530e-01 -4.01818395e-01 -5.60656607e-01 -9.11021903e-02 -4.02272016e-01 1.95000201e-01 -5.91686845e-01 1.27487347e-01 -8.06518316e-01 -1.77739590e-01 -1.76137120e-01 4.96504754e-01 1.04087472e+00 -9.19866264e-01 -7.97394097e-01 -6.95855021e-02 -7.20427632e-01 -1.47428870e+00 -8.22667778e-01 -5.97280204e-01 -9.00065958e-01 -1.11800003e+00 -9.07087088e-01 -6.83546364e-01 3.48014385e-01 9.62273121e-01 1.04668820e+00 -1.49415538e-01 -1.82409286e-01 -9.13527161e-02 -2.16311768e-01 3.81284863e-01 -9.28305089e-02 -2.22391352e-01 4.81461734e-03 1.02635592e-01 8.25369433e-02 -8.29523027e-01 -1.06338322e+00 4.77852851e-01 -8.62586677e-01 6.24045610e-01 4.59380776e-01 7.59063065e-01 9.75334466e-01 8.04537013e-02 5.23064397e-02 -6.01117134e-01 1.42055914e-01 -3.74146491e-01 -6.99332535e-01 -2.43247017e-01 -3.68239939e-01 -1.76729709e-01 7.65263796e-01 -2.04100847e-01 -1.49069774e+00 -2.25327119e-01 -5.65466657e-02 -1.00438952e+00 -9.22158808e-02 -1.29983380e-01 -3.00756961e-01 -1.43600062e-01 3.59672636e-01 6.28195405e-01 -6.76858872e-02 -6.17393672e-01 4.29082960e-02 5.90883970e-01 9.92618859e-01 -1.68829951e-02 7.40505159e-01 8.88417244e-01 -6.50139600e-02 -6.99566722e-01 -1.09318721e+00 -5.03253400e-01 -5.88640273e-01 -1.63958356e-01 1.02239931e+00 -1.40097702e+00 -6.48058236e-01 4.02259499e-01 -9.36619103e-01 -2.40489259e-01 -1.96283340e-01 4.98604804e-01 -8.74071777e-01 4.82654214e-01 -8.42411280e-01 -3.59194458e-01 -4.30133432e-01 -1.20798707e+00 1.17358887e+00 5.60711682e-01 2.89137691e-01 -6.08747840e-01 1.78850412e-01 7.25983381e-01 7.50778198e-01 1.68335229e-01 3.41636539e-01 3.40865463e-01 -1.11039817e+00 6.58296585e-01 -5.57442605e-01 3.83265376e-01 2.55412728e-01 -2.83434987e-01 -9.67167497e-01 -4.74656343e-01 1.94717437e-01 -5.70285320e-02 9.29176867e-01 7.03743756e-01 1.41652679e+00 -1.69586867e-01 5.69800138e-02 1.35287070e+00 1.63932920e+00 1.41515300e-01 1.18926561e+00 5.16998470e-01 9.89227414e-01 2.80369878e-01 7.89599359e-01 3.37724924e-01 5.90574503e-01 9.89073217e-01 4.28268075e-01 -2.64095515e-01 -6.12777352e-01 -3.12203109e-01 4.32541013e-01 6.05760276e-01 -6.74948037e-01 2.18589455e-01 -3.16813976e-01 3.97000819e-01 -1.69352973e+00 -1.53023291e+00 -1.26162812e-01 2.28114843e+00 7.76510119e-01 -3.02367330e-01 -6.14700764e-02 -2.16552000e-02 9.60300744e-01 6.50664926e-01 -7.95610845e-01 3.06953583e-02 -5.06663501e-01 3.17659937e-02 3.38849485e-01 6.89060509e-01 -1.08296633e+00 8.96738350e-01 5.64426374e+00 6.95459306e-01 -1.26239145e+00 1.95771962e-01 6.57571852e-01 -3.95251513e-01 -1.46154925e-01 -2.41650879e-01 -7.27249503e-01 7.22274423e-01 6.08453572e-01 -2.66255707e-01 7.98981905e-01 4.45160538e-01 6.51386023e-01 2.42305566e-02 -7.92906940e-01 1.38274801e+00 2.34086424e-01 -1.62562943e+00 1.02632634e-01 -2.39029005e-02 1.03327870e+00 1.53328493e-01 3.60620886e-01 -1.73113540e-01 2.24941894e-02 -9.95522916e-01 5.49252689e-01 6.26576006e-01 1.39254630e+00 -8.22045922e-01 4.94436890e-01 -7.89329968e-03 -1.49928367e+00 -2.36086443e-01 -4.28470671e-01 1.00216560e-01 3.36829424e-01 2.72329748e-01 2.70020127e-01 8.23656440e-01 1.47381568e+00 1.37059355e+00 -3.99038583e-01 8.00289273e-01 2.49695033e-02 -1.20269224e-01 3.33368570e-01 9.64687407e-01 -2.05854520e-01 -2.32368812e-01 8.16196740e-01 8.59792650e-01 3.42199802e-01 4.89393145e-01 -2.62149543e-01 9.93499041e-01 7.61810094e-02 -3.20759058e-01 -4.15483624e-01 5.56374013e-01 3.50007772e-01 1.18565428e+00 5.14549427e-02 -4.21866179e-01 -6.23102963e-01 1.16335297e+00 9.78082865e-02 5.28677464e-01 -7.75048435e-01 -5.00229970e-02 1.04591024e+00 5.11493862e-01 5.35183191e-01 5.19120134e-02 -2.10752487e-01 -1.66022718e+00 1.15710437e-01 -1.14808512e+00 4.86800253e-01 -1.34151089e+00 -1.04207170e+00 8.87986839e-01 -1.54773593e-01 -1.68030989e+00 -2.46982694e-01 -6.08131438e-02 -2.54620850e-01 1.11774647e+00 -2.09429789e+00 -8.19060683e-01 -9.35675025e-01 1.18468094e+00 9.60561633e-01 2.20731646e-02 3.00842017e-01 4.66346860e-01 -4.33663607e-01 1.80371702e-01 5.61786219e-02 -3.91504690e-02 9.01265681e-01 -8.17629457e-01 3.24177682e-01 1.15699756e+00 -5.22795916e-01 4.91444021e-01 6.94564521e-01 -5.65383852e-01 -1.43714309e+00 -1.10986376e+00 5.83476245e-01 -2.71541536e-01 2.38027930e-01 1.89645104e-02 -1.15405047e+00 5.27720928e-01 2.99018085e-01 4.82052088e-01 1.23596072e-01 -5.69074512e-01 -4.07603830e-01 -2.66338944e-01 -1.09778321e+00 4.96628284e-01 1.54742455e+00 -6.01900518e-01 -4.11034077e-01 -1.86157897e-01 9.92281914e-01 -8.41158152e-01 -1.02523339e+00 4.66847360e-01 5.79509079e-01 -1.87015247e+00 1.32930338e+00 -1.65472656e-01 1.07931519e+00 -6.70544326e-01 -1.72202066e-01 -1.16873777e+00 -6.88991606e-01 -6.12323046e-01 -3.83906066e-01 9.55062926e-01 -3.39133233e-01 -5.35163403e-01 4.62674379e-01 3.72333288e-01 -1.55352443e-01 -5.26460528e-01 -9.24800515e-01 -5.41552007e-01 -2.56313294e-01 1.30504653e-01 6.33698761e-01 9.82966781e-01 -2.80688256e-01 2.38056749e-01 -6.61620378e-01 3.64128500e-01 1.06580007e+00 4.58736926e-01 5.80049574e-01 -8.38450611e-01 -2.75822639e-01 -2.81677246e-01 -2.82272547e-01 -1.27798450e+00 2.70161834e-02 -3.94003898e-01 -3.52897316e-01 -1.39803648e+00 5.53458929e-01 1.25332206e-01 -2.65410334e-01 -1.75592229e-01 -2.86220998e-01 3.41842920e-01 1.10350721e-01 4.15870219e-01 -6.77154541e-01 6.18815422e-01 1.82596004e+00 2.71830559e-01 -3.07814956e-01 -3.10694695e-01 -8.15057516e-01 5.60078561e-01 5.42939246e-01 1.47204295e-01 -4.22304928e-01 -6.69621110e-01 -1.74646810e-01 6.77017689e-01 4.65501606e-01 -9.51871157e-01 3.07375371e-01 -2.89208978e-01 6.73228145e-01 -7.02411354e-01 3.60011041e-01 -6.88958883e-01 3.50843549e-01 1.84706390e-01 -4.21547711e-01 8.84679556e-02 -1.14911370e-01 4.52332795e-01 -4.76578534e-01 6.37451828e-01 1.26491785e+00 -2.51415998e-01 -1.15496159e+00 7.69162476e-01 1.09525390e-01 1.62738264e-01 7.98895895e-01 -3.73625606e-01 -7.37038016e-01 -3.87694329e-01 -3.41836721e-01 1.73988953e-01 8.36080432e-01 7.35816061e-01 1.14910865e+00 -1.30731058e+00 -9.08893049e-01 4.59302187e-01 -8.06217417e-02 1.79847986e-01 1.07179677e+00 7.66378403e-01 -4.34473842e-01 4.27689433e-01 -7.91941166e-01 -6.20958209e-01 -1.40071821e+00 8.24878812e-01 5.38390994e-01 3.23646069e-02 -1.18667531e+00 4.90267605e-01 7.06514180e-01 3.16225678e-01 -1.11541431e-03 -6.20819032e-02 -6.30799532e-01 -4.29231942e-01 1.32575810e+00 6.07203126e-01 -2.95665741e-01 -1.04512787e+00 -1.47143066e-01 1.06291771e+00 -5.30244708e-02 2.21583713e-02 1.43378842e+00 -9.21108365e-01 -4.10331264e-02 2.54270673e-01 1.19510257e+00 -9.13391560e-02 -1.78753531e+00 -4.92281139e-01 -7.53744781e-01 -1.34388721e+00 3.01919937e-01 -4.21123743e-01 -1.58099413e+00 7.26913571e-01 8.40416133e-01 -1.86103165e-01 1.61538374e+00 -1.27521083e-01 1.12510931e+00 -5.27307868e-01 6.13577425e-01 -7.42322028e-01 -1.34043805e-02 5.05892515e-01 1.03437626e+00 -1.42598569e+00 -1.48325684e-02 -6.58372521e-01 -7.32941091e-01 1.01670265e+00 9.07263815e-01 -1.87545568e-01 3.36845636e-01 3.12775671e-01 -1.17695078e-01 1.16350368e-01 -1.00056362e+00 -2.59670913e-01 2.36832350e-01 7.24649131e-01 3.57227653e-01 -3.34905297e-01 -1.19217493e-01 4.47234690e-01 -1.26931578e-01 4.67337489e-01 7.07352996e-01 4.24980074e-01 -3.33749086e-01 -5.49796462e-01 -2.37052575e-01 -6.24671578e-03 -5.23957968e-01 -2.03234091e-01 1.65023565e-01 3.47755373e-01 1.27061978e-01 9.11054552e-01 2.25599945e-01 -5.61262071e-01 3.61518443e-01 -7.39138901e-01 5.16087770e-01 -2.20964983e-01 -2.22287998e-01 1.52148187e-01 -8.13650414e-02 -1.51324511e+00 -7.88300931e-01 -4.34897184e-01 -1.12421131e+00 -6.12107515e-01 4.01533484e-01 -2.27429554e-01 1.82855740e-01 4.57401663e-01 6.16621315e-01 6.91068172e-01 8.81272793e-01 -1.15549135e+00 1.05289957e-02 -6.01969600e-01 -6.24730349e-01 5.09621382e-01 8.29149425e-01 -6.30435109e-01 -4.97968704e-01 2.13317275e-01]

[10.92693042755127, -2.0580759048461914]

20aa37ed-ff55-438d-9ddb-65086a1aa539

simple-online-and-realtime-tracking-with-a

1703.07402

http://arxiv.org/abs/1703.07402v1

http://arxiv.org/pdf/1703.07402v1.pdf

Simple Online and Realtime Tracking with a Deep Association Metric

Simple Online and Realtime Tracking (SORT) is a pragmatic approach to multiple object tracking with a focus on simple, effective algorithms. In this paper, we integrate appearance information to improve the performance of SORT. Due to this extension we are able to track objects through longer periods of occlusions, effectively reducing the number of identity switches. In spirit of the original framework we place much of the computational complexity into an offline pre-training stage where we learn a deep association metric on a large-scale person re-identification dataset. During online application, we establish measurement-to-track associations using nearest neighbor queries in visual appearance space. Experimental evaluation shows that our extensions reduce the number of identity switches by 45%, achieving overall competitive performance at high frame rates.

['Alex Bewley', 'Nicolai Wojke', 'Dietrich Paulus']

2017-03-21

['large-scale-person-re-identification', 'video-instance-segmentation']

['computer-vision', 'computer-vision']

[-1.38235509e-01 -4.27639991e-01 2.41257716e-02 -4.92835075e-01 -7.53494263e-01 -8.01812649e-01 4.82279956e-01 8.80458504e-02 -7.86285579e-01 4.40311253e-01 5.10156602e-02 -1.58577319e-02 9.67979133e-02 -4.65863347e-01 -8.00377071e-01 -9.07905176e-02 -1.24464452e-01 6.48955524e-01 4.98052418e-01 1.53115228e-01 -6.33883178e-02 6.25635624e-01 -1.67086136e+00 1.21305048e-01 2.39225343e-01 8.92024159e-01 -2.18107641e-01 9.55945969e-01 4.80263866e-02 4.73508477e-01 -6.31018519e-01 -8.56916845e-01 7.12474942e-01 -1.89447161e-02 -5.49950182e-01 1.34460345e-01 1.39350891e+00 -6.08021498e-01 -6.01014912e-01 8.15214515e-01 7.18872607e-01 2.81643540e-01 1.79948032e-01 -1.36906826e+00 -6.10133290e-01 5.73240928e-02 -5.79617143e-01 4.10798848e-01 6.24195635e-01 2.34813184e-01 7.31795073e-01 -8.49389672e-01 6.37264729e-01 1.31782186e+00 1.25314796e+00 6.56263649e-01 -1.51370597e+00 -6.72275543e-01 3.79872501e-01 2.37986788e-01 -1.65010524e+00 -7.70571232e-01 4.91470098e-02 -4.58623171e-01 7.65104294e-01 4.58690524e-01 7.22902894e-01 7.73442090e-01 -3.44562113e-01 6.36611879e-01 8.19053590e-01 -3.00461650e-01 -1.25755340e-01 2.18264192e-01 3.78036171e-01 6.30237401e-01 4.00504142e-01 3.23819786e-01 -5.19731939e-01 -3.08478713e-01 8.96190107e-01 1.12428360e-01 4.26314659e-02 -6.10198498e-01 -1.00854659e+00 5.05271077e-01 5.29236972e-01 -4.97860685e-02 2.26031058e-02 4.09349769e-01 3.47300768e-01 4.18720186e-01 1.85925916e-01 6.63447231e-02 -2.65869200e-01 -2.45625824e-01 -8.20898473e-01 4.77533966e-01 6.07829988e-01 1.20631850e+00 6.60220206e-01 -3.06899935e-01 -4.06511366e-01 5.80071270e-01 1.80386290e-01 6.18845463e-01 6.11989535e-02 -1.21005726e+00 2.20954016e-01 4.40586418e-01 4.22436893e-01 -8.03376436e-01 -4.05716240e-01 -5.33565998e-01 -2.04121262e-01 4.64826077e-01 8.16660047e-01 -1.04632042e-01 -8.16143036e-01 1.83988333e+00 6.08152986e-01 4.25870568e-01 -5.07213116e-01 9.24440980e-01 2.92751908e-01 -5.16713560e-02 2.12068498e-01 1.33317217e-01 1.55311513e+00 -8.59167874e-01 -4.22659665e-01 -1.60881326e-01 4.50611144e-01 -8.32448125e-01 8.29175234e-01 6.46295696e-02 -1.12120843e+00 -8.31555307e-01 -8.41881692e-01 -1.94469199e-01 -4.56197083e-01 9.86165702e-02 8.71754646e-01 1.10105264e+00 -1.37915063e+00 6.27546072e-01 -7.69356370e-01 -6.22991443e-01 6.08608246e-01 9.11906540e-01 -4.75922436e-01 -4.34803665e-02 -6.61444724e-01 7.49717534e-01 8.08893070e-02 -1.70885827e-02 -4.36155558e-01 -9.59478915e-01 -5.94269514e-01 -1.29940882e-01 1.68300509e-01 -1.05469871e+00 1.26496458e+00 -8.54345083e-01 -1.24448657e+00 1.04169464e+00 -5.25869906e-01 -5.76885581e-01 8.75259876e-01 -6.10024393e-01 -4.62866068e-01 -6.73356503e-02 1.00910135e-01 7.60218263e-01 6.38594687e-01 -1.02071702e+00 -9.13302481e-01 -4.07914668e-01 2.01238602e-01 1.83139175e-01 -5.66281915e-01 4.24096823e-01 -9.90579247e-01 -5.68347216e-01 -2.21405789e-01 -1.22039866e+00 -1.00309588e-01 6.98853791e-01 1.46766007e-01 -1.71599746e-01 8.78526509e-01 -6.14500582e-01 8.18444073e-01 -2.23960042e+00 -1.79695785e-01 1.86044782e-01 4.39134151e-01 3.11262578e-01 -6.76179081e-02 -1.17733531e-01 1.88385546e-01 -3.29450011e-01 3.84579748e-01 -8.50193441e-01 2.00144306e-01 7.03615993e-02 -1.01407506e-02 7.71468997e-01 -1.71568200e-01 1.07286894e+00 -8.17340553e-01 -4.77480054e-01 3.31777275e-01 6.01571620e-01 -6.18121743e-01 7.87825584e-02 1.59266621e-01 2.78696567e-01 -3.88948880e-02 6.18102908e-01 7.01965749e-01 -2.91736633e-01 1.33143319e-03 -2.30851844e-01 -1.40578508e-01 1.09793313e-01 -1.54158354e+00 1.86024976e+00 -1.63849846e-01 7.75702059e-01 -4.79584187e-03 -4.07009810e-01 4.60159600e-01 9.47022960e-02 6.88302934e-01 -6.35609448e-01 -5.98974973e-02 -6.73568323e-02 -1.97547048e-01 9.26678106e-02 6.24940872e-01 1.86980739e-01 9.13753435e-02 3.37824494e-01 -2.08328828e-01 8.20329964e-01 1.68768972e-01 2.28825092e-01 1.03174472e+00 3.64474058e-01 -1.07418373e-01 -1.11831687e-01 3.69008929e-01 8.11686069e-02 4.95647907e-01 1.12385130e+00 -4.98577863e-01 5.34133494e-01 -4.08212274e-01 -8.20212662e-01 -1.12280965e+00 -1.50107503e+00 -1.23154253e-01 1.29430127e+00 3.30156237e-01 -5.04035294e-01 -5.62346578e-01 -4.59556937e-01 4.11022723e-01 6.50466979e-02 -5.68948627e-01 2.08299875e-01 -9.84699786e-01 -6.19413316e-01 5.81903875e-01 8.52663755e-01 3.74567300e-01 -5.98700523e-01 -5.98711908e-01 3.18049818e-01 7.44279027e-02 -1.38487697e+00 -8.70767474e-01 -4.42750543e-01 -7.08000004e-01 -1.06665540e+00 -7.32999980e-01 -5.37609041e-01 6.40741289e-01 6.07159734e-01 1.06591690e+00 1.38522029e-01 -7.82339156e-01 7.64855087e-01 -6.24666102e-02 -3.25788170e-01 -4.03253594e-03 -1.55859008e-01 3.92607123e-01 1.48461293e-02 6.51784718e-01 -4.29017335e-01 -7.53043950e-01 4.52567101e-01 -1.96045399e-01 -2.66447484e-01 4.18585211e-01 3.29338282e-01 4.22559232e-01 -2.70429045e-01 5.91126643e-02 -4.50867414e-01 5.05364388e-02 1.32751644e-01 -9.28645432e-01 2.83906460e-01 -4.93884474e-01 -1.29220024e-01 1.47855490e-01 -7.86625326e-01 -9.03495133e-01 3.06683153e-01 1.63048148e-01 -5.87296009e-01 -3.60277109e-02 -4.61589932e-01 8.27627853e-02 -6.06787682e-01 5.29303014e-01 1.07754521e-01 -4.17789109e-02 -6.74868405e-01 5.96091151e-01 3.23570043e-01 9.00025308e-01 -5.75327396e-01 1.07317615e+00 9.00677741e-01 -1.71424355e-02 -3.23145270e-01 -7.59768248e-01 -8.48216057e-01 -8.49792004e-01 -2.86627442e-01 7.54998207e-01 -1.25696433e+00 -1.29245198e+00 3.14656079e-01 -9.35671389e-01 -3.13378990e-01 -2.96018928e-01 4.90672499e-01 -2.78197318e-01 4.45132822e-01 -6.42929792e-01 -8.16890717e-01 -3.69033217e-01 -6.64963245e-01 1.30096006e+00 3.39888662e-01 -2.92822510e-01 -7.42608249e-01 5.08016832e-02 3.33112299e-01 3.69411081e-01 6.41766191e-02 -5.58617897e-02 -4.08288151e-01 -9.78865445e-01 -4.02270645e-01 -5.31980276e-01 -2.95865804e-01 9.50805694e-02 -4.49948847e-01 -9.97764766e-01 -7.34480262e-01 -5.35453439e-01 1.20847993e-01 6.63111567e-01 2.45688438e-01 9.36935544e-01 -3.63973267e-02 -7.39549339e-01 8.81307662e-01 1.30627966e+00 -8.95391926e-02 4.34823036e-01 5.89692652e-01 9.42819536e-01 2.31300980e-01 5.61670542e-01 3.52670163e-01 5.75744033e-01 1.37404382e+00 -1.46503255e-01 -1.49612278e-01 -6.09172165e-01 -1.15415357e-01 4.22840655e-01 2.12737560e-01 -2.14422181e-01 1.93099186e-01 -5.14326096e-01 5.55733323e-01 -1.99597228e+00 -1.27323484e+00 -2.15901911e-01 2.54630899e+00 6.30767941e-01 1.08303629e-01 7.42104471e-01 -2.84627289e-01 9.10654664e-01 -3.71784240e-01 -5.23046434e-01 1.24426328e-01 -7.71135762e-02 -1.26077635e-02 8.16657007e-01 7.63267040e-01 -1.39272058e+00 8.67143214e-01 6.98888016e+00 4.46378380e-01 -7.07792938e-01 2.19002679e-01 1.71569064e-01 -6.22382820e-01 3.54475439e-01 -1.05587713e-01 -1.34695077e+00 4.72569048e-01 9.33035135e-01 4.47955243e-02 5.30193508e-01 7.78217971e-01 -1.37146041e-01 1.78790301e-01 -1.31980014e+00 1.30699813e+00 3.85943316e-02 -1.34146023e+00 -2.74229050e-01 3.02090675e-01 4.36462700e-01 7.00294897e-02 9.31749295e-04 2.06450671e-01 5.47829151e-01 -6.71158254e-01 8.93642008e-01 5.25793314e-01 7.78964639e-01 -4.49498951e-01 3.47296596e-01 -2.26759553e-01 -1.68650401e+00 -1.07501455e-01 -4.53391165e-01 -1.45082429e-01 2.80311018e-01 2.72648096e-01 -5.52149177e-01 3.86064947e-01 1.03633749e+00 4.23772693e-01 -9.30136800e-01 1.50345457e+00 2.89054513e-01 -3.18787880e-02 -8.18013489e-01 3.75543445e-01 -2.73325801e-01 1.13735728e-01 4.65592235e-01 1.37135088e+00 4.55830805e-02 1.27368886e-02 5.20691335e-01 6.05335355e-01 2.09473148e-02 -2.94405937e-01 -3.14574093e-01 4.80314851e-01 6.56378448e-01 1.19649434e+00 -5.74349880e-01 -6.10556424e-01 -5.50644755e-01 1.25703990e+00 4.71874714e-01 6.15792125e-02 -1.09919119e+00 -2.07551364e-02 1.02696037e+00 2.38533482e-01 5.70301473e-01 -2.71942884e-01 -6.25902936e-02 -1.08994186e+00 3.05099338e-01 -6.63063049e-01 7.36669481e-01 -3.31247985e-01 -1.27880144e+00 1.78144380e-01 -9.38444808e-02 -1.13817286e+00 -1.86976027e-02 -4.99414712e-01 -2.93388456e-01 8.10679138e-01 -1.28164721e+00 -1.44584441e+00 -5.15857518e-01 7.39190876e-01 3.24162632e-01 5.47977090e-02 7.06967533e-01 9.27769065e-01 -5.02726495e-01 1.22341633e+00 1.43698841e-01 3.85527492e-01 9.54869628e-01 -1.29631579e+00 8.90485466e-01 9.89416897e-01 4.46101397e-01 9.59353328e-01 7.76789606e-01 -6.42074406e-01 -1.46162057e+00 -1.00614738e+00 8.15779626e-01 -1.10654914e+00 5.78776240e-01 -6.89482152e-01 -6.28646433e-01 1.00107670e+00 -1.97539881e-01 3.85656446e-01 7.38290191e-01 5.29666305e-01 -6.44091666e-01 -2.27251902e-01 -1.00319779e+00 5.40457845e-01 1.63789892e+00 -5.34264266e-01 -3.06572974e-01 4.19604897e-01 4.97496337e-01 -5.98676860e-01 -1.04617512e+00 4.30380665e-02 1.03299034e+00 -7.92502642e-01 1.60118008e+00 -4.68481600e-01 -7.61645079e-01 -7.10142136e-01 1.04104159e-02 -5.37661076e-01 -6.57429576e-01 -8.43271196e-01 -3.99643242e-01 1.36265898e+00 -9.23360363e-02 -5.74346960e-01 1.09516573e+00 9.85515177e-01 1.95957392e-01 -2.24327192e-01 -9.39098358e-01 -1.09192991e+00 -3.83457720e-01 -1.42346799e-01 6.29209399e-01 6.81499779e-01 -3.38830292e-01 4.49775718e-03 -3.25366497e-01 4.62886423e-01 1.20183563e+00 2.40110919e-01 1.18967569e+00 -1.31441629e+00 -6.33916676e-01 -4.36519235e-01 -6.93326592e-01 -1.22403216e+00 -1.49708256e-01 -5.99739254e-01 -2.30485171e-01 -1.11786544e+00 4.45816636e-01 -7.29173005e-01 -4.32465523e-01 4.88310456e-01 -3.16379160e-01 7.12540686e-01 5.90279758e-01 4.05508876e-01 -1.09610736e+00 1.35277227e-01 5.17797291e-01 -1.41124681e-01 -4.57985736e-02 -1.73426513e-02 -5.72355151e-01 5.33261418e-01 4.71941829e-01 -5.25989652e-01 -3.17777991e-02 -7.48509288e-01 -2.17772976e-01 -4.54228282e-01 8.48667681e-01 -1.40271807e+00 4.96755123e-01 2.41190523e-01 8.65305841e-01 -5.44097662e-01 6.93303704e-01 -8.58049214e-01 4.84436035e-01 5.22151351e-01 -9.46025327e-02 4.02996123e-01 4.84468967e-01 5.19353747e-01 4.17259365e-01 2.45829806e-01 7.34941602e-01 4.84387875e-02 -9.43192959e-01 5.19967854e-01 1.69758096e-01 -1.28310829e-01 9.92597580e-01 -6.95982397e-01 -2.91516155e-01 -1.17545605e-01 -9.34056342e-01 1.98248193e-01 8.91721010e-01 5.02977848e-01 1.95036069e-01 -1.53386450e+00 -5.80707908e-01 4.19459492e-02 3.85503843e-02 -5.20956993e-01 1.07789956e-01 8.05806100e-01 -2.93080777e-01 4.85232979e-01 -2.36729100e-01 -7.34465003e-01 -1.76298428e+00 7.19452739e-01 5.14922678e-01 -7.42207002e-03 -1.08043170e+00 9.33954298e-01 2.40405098e-01 -1.84095517e-01 4.58894134e-01 3.41024399e-01 1.50815532e-01 -2.52967834e-01 9.59803283e-01 4.88763481e-01 -1.51990369e-01 -7.18473256e-01 -6.76809311e-01 7.56094873e-01 -3.30870390e-01 -1.93323761e-01 9.38359737e-01 -4.08738554e-01 2.52031624e-01 -1.19029889e-02 1.12073100e+00 1.07270844e-01 -1.61580265e+00 -2.94716924e-01 1.36278570e-01 -9.03021514e-01 -1.89086184e-01 -6.92725897e-01 -8.58207464e-01 3.58442694e-01 1.31053007e+00 -2.48428494e-01 7.91005194e-01 -2.44085826e-02 9.44952130e-01 4.73698050e-01 5.82613289e-01 -9.17249560e-01 -2.34928817e-01 8.76910836e-02 2.10492939e-01 -1.26614630e+00 3.35016161e-01 -4.36749429e-01 -8.41757655e-02 7.87564516e-01 6.63688004e-01 -1.18549220e-01 3.32760841e-01 4.48420048e-01 2.02660024e-01 -1.73187211e-01 -3.12521636e-01 -3.65932226e-01 3.27438444e-01 9.37377334e-01 1.34143293e-01 -1.11452192e-01 1.35720894e-01 1.02585992e-02 -9.03943703e-02 9.50308144e-03 -1.85066890e-02 8.26831460e-01 -3.53118658e-01 -1.40935326e+00 -6.17620230e-01 1.76054463e-01 -3.68560940e-01 1.40729293e-01 -1.29479244e-01 8.15333247e-01 2.06358626e-01 8.30831110e-01 3.82900298e-01 -4.49655689e-02 4.92464274e-01 -1.05727054e-01 9.93785501e-01 -3.41333717e-01 -7.77350545e-01 -1.47029966e-01 1.69336691e-01 -8.80625129e-01 -4.18937355e-01 -1.02893507e+00 -9.30472672e-01 -6.87357366e-01 -2.39043862e-01 -1.49453193e-01 4.15214747e-01 7.92884648e-01 7.36229718e-01 3.14913779e-01 1.66644171e-01 -8.79900217e-01 -3.83621901e-01 -4.86168861e-01 -9.03673843e-02 8.35201263e-01 2.92958349e-01 -8.99481893e-01 2.43317932e-01 1.62265748e-01]

[6.504465103149414, -1.8163576126098633]

9c2ef47d-37ec-43e6-b6c9-5ca36d70119b

conversations-with-search-engines

2004.14162

https://arxiv.org/abs/2004.14162v2

https://arxiv.org/pdf/2004.14162v2.pdf

Conversations with Search Engines: SERP-based Conversational Response Generation

In this paper, we address the problem of answering complex information needs by conversing conversations with search engines, in the sense that users can express their queries in natural language, and directly receivethe information they need from a short system response in a conversational manner. Recently, there have been some attempts towards a similar goal, e.g., studies on Conversational Agents (CAs) and Conversational Search (CS). However, they either do not address complex information needs, or they are limited to the development of conceptual frameworks and/or laboratory-based user studies. We pursue two goals in this paper: (1) the creation of a suitable dataset, the Search as a Conversation (SaaC) dataset, for the development of pipelines for conversations with search engines, and (2) the development of astate-of-the-art pipeline for conversations with search engines, the Conversations with Search Engines (CaSE), using this dataset. SaaC is built based on a multi-turn conversational search dataset, where we further employ workers from a crowdsourcing platform to summarize each relevant passage into a short, conversational response. CaSE enhances the state-of-the-art by introducing a supporting token identification module and aprior-aware pointer generator, which enables us to generate more accurate responses. We carry out experiments to show that CaSE is able to outperform strong baselines. We also conduct extensive analyses on the SaaC dataset to show where there is room for further improvement beyond CaSE. Finally, we release the SaaC dataset and the code for CaSE and all models used for comparison to facilitate future research on this topic.

['Evangelos Kanoulas', 'Zhaochun Ren', 'Christof Monz', 'Zhumin Chen', 'Pengjie Ren', 'Maarten de Rijke']

2020-04-29

['conversational-search', 'conversational-response-generation']

['natural-language-processing', 'natural-language-processing']

[ 7.23192282e-03 8.37130770e-02 -4.42465022e-02 -2.75042832e-01 -1.28893661e+00 -8.85819614e-01 1.32238066e+00 7.99260139e-02 -5.55066168e-01 6.09960258e-01 7.14717925e-01 -4.73693699e-01 2.42814124e-01 -5.24295390e-01 -3.17612410e-01 -1.79764077e-01 4.66098875e-01 8.85911644e-01 4.88141388e-01 -7.31589913e-01 2.77150661e-01 1.69354171e-01 -1.61366594e+00 8.65138829e-01 6.98319316e-01 6.75653040e-01 3.42479408e-01 9.18679237e-01 -3.78973037e-01 9.18682098e-01 -8.42070878e-01 -6.38860583e-01 -2.51826584e-01 -5.74848771e-01 -1.56520808e+00 -3.38401824e-01 6.38883561e-02 -1.92705020e-01 -7.52178803e-02 3.93851846e-01 7.35779583e-01 3.45940709e-01 3.40922952e-01 -1.46724057e+00 -4.94093508e-01 6.39563501e-01 2.67622918e-01 1.17491141e-01 1.06071234e+00 4.26419258e-01 9.46705699e-01 -8.28308463e-01 7.34347343e-01 1.33517194e+00 5.84979117e-01 8.01745892e-01 -8.81467164e-01 -4.28387791e-01 -8.19637254e-02 1.84228852e-01 -9.42591071e-01 -8.57497275e-01 4.15496677e-01 -3.89443099e-01 1.56701291e+00 7.36670077e-01 4.48174328e-01 1.59837592e+00 -5.56210995e-01 9.28847134e-01 1.01379216e+00 -5.31476736e-01 8.69382694e-02 3.50152284e-01 2.15618104e-01 5.44350624e-01 -2.72385389e-01 -5.57835326e-02 -7.01399505e-01 -4.96320337e-01 3.18167269e-01 -2.97338158e-01 -2.82492518e-01 3.32122028e-01 -1.25927365e+00 7.94707954e-01 1.54317394e-01 5.07900298e-01 -3.79662842e-01 -7.44887441e-02 6.47771835e-01 4.76568043e-01 4.85044628e-01 7.86876917e-01 -3.21003765e-01 -7.00681090e-01 -5.48136890e-01 8.72771561e-01 1.66351414e+00 1.13143718e+00 5.37992299e-01 -5.37125409e-01 -7.06367254e-01 1.07661128e+00 1.42126486e-01 3.94624650e-01 6.21042967e-01 -1.04862368e+00 7.55096734e-01 6.58603966e-01 5.55587411e-01 -8.32052648e-01 -4.98869389e-01 3.83730382e-01 -1.44834787e-01 -4.37260896e-01 5.97652555e-01 -4.14448708e-01 -2.04278439e-01 1.63510954e+00 1.60883874e-01 -2.21655071e-01 2.63957709e-01 7.93415725e-01 1.39150107e+00 5.37594736e-01 2.62523834e-02 -1.66787151e-02 1.68790460e+00 -1.14874244e+00 -6.53763950e-01 -2.86934972e-01 8.71812999e-01 -9.93670762e-01 1.46011972e+00 -3.61605771e-02 -1.26171160e+00 -2.97067612e-01 -3.84587377e-01 -4.19129997e-01 -4.84601080e-01 -5.12808040e-02 6.29883885e-01 4.22030985e-01 -1.30290818e+00 -3.91615108e-02 -6.23717725e-01 -8.58624518e-01 -2.88443238e-01 -9.01117399e-02 -2.74848882e-02 2.02137828e-01 -1.54852724e+00 1.10976279e+00 -6.86123744e-02 -1.77608818e-01 -6.46456003e-01 -4.52658623e-01 -8.42065334e-01 -5.33984192e-02 3.84659648e-01 -7.00309634e-01 2.07694769e+00 -5.37280202e-01 -1.67055070e+00 1.00433457e+00 -5.51068842e-01 -3.37722123e-01 5.52840292e-01 -8.22276771e-02 -2.07432359e-01 9.77705270e-02 2.99399674e-01 6.64086282e-01 4.67719417e-03 -1.03081024e+00 -7.50936031e-01 -5.57965115e-02 6.37058198e-01 3.12445939e-01 -1.20433182e-01 5.62833786e-01 -7.29981482e-01 -3.75763178e-01 -5.23488283e-01 -1.18914163e+00 4.80946265e-02 -5.57401001e-01 -6.09869063e-01 -8.55542421e-01 4.17873055e-01 -6.46624446e-01 1.34925699e+00 -1.73524165e+00 -6.88705817e-02 -1.94599688e-01 -2.74453331e-02 4.44454968e-01 -3.22605669e-01 1.14554060e+00 3.39902014e-01 3.95220339e-01 2.56762262e-02 -7.25756168e-01 2.31073156e-01 9.04264743e-04 -3.60482275e-01 -1.18281081e-01 1.59698099e-01 1.18637490e+00 -1.17384756e+00 -4.42138106e-01 -1.83933839e-01 2.58708864e-01 -5.23042262e-01 5.23090363e-01 -4.91425008e-01 3.82432669e-01 -5.62885106e-01 3.45793188e-01 -1.43364146e-02 -2.52988130e-01 -7.11526200e-02 1.46499500e-01 -3.26644212e-01 1.03275669e+00 -7.01930523e-01 1.55079329e+00 -9.24412549e-01 6.83011532e-01 3.92559767e-01 -5.38024783e-01 5.64562678e-01 7.72517204e-01 2.74708003e-01 -6.10869825e-01 -1.22284126e-02 2.38731772e-01 -3.08761358e-01 -9.70660090e-01 7.55347371e-01 1.09168686e-01 -3.81287485e-01 1.02477658e+00 -2.50455141e-01 -4.52576339e-01 3.88256282e-01 3.88138413e-01 1.20178056e+00 -1.02031603e-01 -5.73260374e-02 -2.50935904e-03 6.75849497e-01 4.29462165e-01 -7.39074964e-03 1.10495543e+00 -1.72967151e-01 4.94796097e-01 3.57835948e-01 -2.05797657e-01 -7.20303059e-01 -4.04720604e-01 2.34918639e-01 1.48383772e+00 -4.10538688e-02 -5.44427216e-01 -9.07384574e-01 -5.06367087e-01 -1.56596035e-01 7.90488422e-01 -2.01161221e-01 2.11564407e-01 -5.17650247e-01 -3.74949247e-01 1.03646624e+00 3.30976456e-01 6.84887469e-01 -1.46125019e+00 -6.89125240e-01 1.74535185e-01 -1.04839623e+00 -1.30987203e+00 -7.11041451e-01 -3.66629034e-01 -8.90200958e-02 -1.10410607e+00 -6.03203416e-01 -7.81664789e-01 -1.05352603e-01 5.36258817e-01 1.29946971e+00 3.14869463e-01 -3.92879685e-03 7.97022343e-01 -7.44003892e-01 -5.23163736e-01 -7.09779739e-01 4.70217198e-01 -1.45742491e-01 -4.02062744e-01 7.28710175e-01 -1.93034485e-01 -5.51894367e-01 5.69359004e-01 -6.41837895e-01 6.70085028e-02 8.64746869e-02 6.75372183e-01 -2.48242974e-01 -5.06044328e-01 8.53359044e-01 -7.69917786e-01 1.62991309e+00 -6.41060472e-01 -2.44017035e-01 3.02681386e-01 -4.13502872e-01 -8.78582746e-02 4.00743008e-01 -3.11251909e-01 -1.17998350e+00 -2.56990403e-01 -5.67355454e-01 3.23673844e-01 -2.34957412e-01 5.56145728e-01 1.66994795e-01 1.79284021e-01 9.60574806e-01 2.59871095e-01 8.93227980e-02 -4.22440231e-01 5.88402212e-01 1.31105852e+00 4.41421837e-01 -8.07970941e-01 2.88905293e-01 9.28277224e-02 -7.15402007e-01 -8.25459540e-01 -5.72216451e-01 -9.02966022e-01 -6.64688200e-02 -5.19767031e-02 8.56712699e-01 -8.35605979e-01 -1.23094285e+00 3.65707844e-01 -1.58182800e+00 -7.04917490e-01 1.48890510e-01 1.58749819e-01 -6.78960502e-01 3.02449614e-01 -1.00145268e+00 -1.05277693e+00 -5.05101144e-01 -1.25444412e+00 1.05854571e+00 1.00115910e-01 -9.52845275e-01 -9.91848409e-01 2.37977833e-01 7.83683956e-01 6.84168994e-01 -2.57046133e-01 5.79551280e-01 -1.28994238e+00 -3.21756631e-01 -2.89227039e-01 -2.24722661e-02 -1.18205182e-01 -1.05752625e-01 -1.74877822e-01 -1.05656886e+00 7.88310915e-02 -1.26936838e-01 -8.02926540e-01 5.79475939e-01 -2.23688915e-01 7.13812351e-01 -6.31724000e-01 -4.07478541e-01 -5.35485409e-02 6.42453194e-01 1.16960458e-01 5.08447945e-01 3.91745150e-01 1.85255006e-01 1.02278173e+00 3.12568367e-01 3.36614281e-01 1.14530909e+00 1.05654192e+00 -2.64166947e-02 8.65968317e-02 -1.06328785e-01 -2.37759873e-01 3.59073132e-01 7.32162416e-01 4.52575274e-02 -5.61523616e-01 -9.72513258e-01 7.07409143e-01 -2.00583887e+00 -1.15239978e+00 -2.59500176e-01 1.89439976e+00 1.24576867e+00 -2.50046939e-01 5.14350772e-01 -2.90537447e-01 4.10061747e-01 1.68271922e-02 -1.93781137e-01 -6.25380754e-01 2.42345899e-01 -2.38161311e-02 -1.61290482e-01 8.94831896e-01 -7.60929942e-01 1.11284316e+00 6.32759237e+00 4.07700151e-01 -8.97138774e-01 1.89697996e-01 4.32648540e-01 1.60201102e-01 -3.89596611e-01 -1.36287302e-01 -9.83832121e-01 4.93592739e-01 1.02902019e+00 -2.74941564e-01 8.52762759e-01 7.31251419e-01 3.46977770e-01 -1.42966792e-01 -1.24461555e+00 8.50600660e-01 1.07195035e-01 -1.35319650e+00 -1.59704715e-01 -3.82278502e-01 2.51538038e-01 2.36612886e-01 -3.68849784e-01 8.24240863e-01 7.83934832e-01 -7.82143056e-01 4.44917887e-01 3.73057246e-01 4.42545116e-01 -2.03073785e-01 7.80238926e-01 7.50507534e-01 -8.79226565e-01 3.07921525e-02 1.12995781e-01 -2.15972260e-01 3.71005893e-01 -2.00407282e-02 -1.23499227e+00 3.04399461e-01 6.94785058e-01 1.17166102e-01 -4.06048179e-01 9.05773520e-01 -2.35028803e-01 5.78232050e-01 -3.15296859e-01 -6.23762906e-01 2.89100766e-01 8.05135444e-02 5.26861012e-01 1.70035017e+00 2.91949902e-02 3.63986403e-01 4.86917138e-01 8.79967093e-01 -1.27214506e-01 1.35329753e-01 -5.97145021e-01 -2.14969456e-01 9.51496780e-01 1.32807469e+00 -3.72201532e-01 -4.77567732e-01 -5.56535125e-01 8.96221995e-01 3.08999389e-01 4.73567069e-01 -3.76172125e-01 -4.91840750e-01 6.02450252e-01 -2.10754741e-02 -2.62177169e-01 -5.52464696e-03 6.26541907e-03 -1.06533694e+00 8.70692432e-02 -1.39423048e+00 3.53024662e-01 -8.29542100e-01 -1.40444553e+00 8.60018909e-01 2.55725473e-01 -6.74360335e-01 -1.07536149e+00 -1.85935363e-01 -8.38679552e-01 1.23189676e+00 -1.38327181e+00 -9.91987407e-01 -4.83158916e-01 5.54606020e-01 9.54167962e-01 -5.60485050e-02 1.10071683e+00 1.70175195e-01 -3.84315044e-01 4.91752237e-01 -4.76200402e-01 1.59524158e-01 8.96618605e-01 -9.53684270e-01 7.59426653e-01 3.61226737e-01 -1.65670455e-01 9.62858915e-01 8.02953780e-01 -6.18593991e-01 -1.38207591e+00 -5.51693380e-01 1.56429946e+00 -1.08699489e+00 7.75236547e-01 -5.76785803e-01 -8.61508489e-01 6.31053388e-01 5.48566997e-01 -5.28602004e-01 6.88247025e-01 2.54616618e-01 -1.11450955e-01 4.05939341e-01 -1.06191576e+00 7.79578626e-01 9.74945724e-01 -9.79874551e-01 -7.30136931e-01 6.31696939e-01 1.01662171e+00 -4.20938551e-01 -5.92130363e-01 -1.18887238e-01 4.96359646e-01 -9.37778234e-01 8.46444011e-01 -6.60062313e-01 3.22870642e-01 6.13325946e-02 5.45958951e-02 -1.46483970e+00 9.56937075e-02 -1.02949727e+00 1.12127826e-01 1.54776490e+00 7.06218719e-01 -7.70348012e-01 4.52199250e-01 1.15178609e+00 -2.80438185e-01 -5.71651876e-01 -6.76850915e-01 -4.76697296e-01 7.63739422e-02 -4.12367761e-01 8.03202808e-01 8.57873738e-01 7.26816058e-01 7.31456399e-01 1.45029500e-02 -2.74361730e-01 -2.65694261e-01 -7.15339184e-02 1.14756548e+00 -1.00228858e+00 -2.23910093e-01 -6.31643176e-01 3.39008927e-01 -1.32676840e+00 3.87232304e-01 -8.71340632e-01 4.36477244e-01 -1.83146954e+00 1.41206995e-01 -4.33860481e-01 4.73011255e-01 4.75365132e-01 -3.16254109e-01 -1.44279152e-01 2.69972861e-01 3.25170666e-01 -7.27527142e-01 4.08908248e-01 9.76064801e-01 4.16911654e-02 -4.34853554e-01 3.09346855e-01 -9.18435156e-01 4.37440813e-01 7.07339823e-01 -1.84909180e-01 -5.08596778e-01 -4.46801215e-01 3.37347865e-01 3.13352138e-01 3.24501514e-01 -5.60746849e-01 5.79750538e-01 -8.82291645e-02 -3.19496930e-01 -3.84070754e-01 5.22450030e-01 -3.40828717e-01 -2.78492779e-01 1.64316997e-01 -8.42297077e-01 4.12345111e-01 1.16681471e-01 1.62517652e-01 -1.60806105e-01 -4.28614259e-01 3.80668677e-02 -4.51358795e-01 -4.28444892e-01 -1.49757773e-01 -8.83151948e-01 4.66483265e-01 5.16602397e-01 1.56379014e-01 -7.31321454e-01 -1.10329270e+00 -3.12224984e-01 5.91582358e-01 3.83479416e-01 6.68344915e-01 2.16337413e-01 -1.01665008e+00 -8.58186722e-01 -2.01387614e-01 4.30631608e-01 -1.90730929e-01 -2.91424930e-01 6.28583014e-01 -3.12621325e-01 8.97124052e-01 3.66686732e-01 -2.38011658e-01 -1.25281024e+00 4.10996914e-01 2.26340607e-01 -3.56112659e-01 -2.12560043e-01 7.79049218e-01 -1.81433707e-01 -9.65304375e-01 3.58837932e-01 -3.32772195e-01 -5.11247635e-01 2.40600109e-01 9.73286867e-01 4.48212743e-01 7.77881518e-02 -6.23091161e-01 -3.97775978e-01 -8.85343999e-02 1.09137565e-01 -6.36480629e-01 9.92633581e-01 -2.95385182e-01 -1.81180075e-01 2.94599712e-01 1.05389369e+00 1.70970529e-01 -7.32280195e-01 -3.25771123e-01 3.29465747e-01 -1.61253884e-01 -4.19331431e-01 -1.04999363e+00 -3.12616795e-01 5.46994150e-01 3.01973559e-02 6.77582681e-01 5.99015832e-01 2.70736605e-01 9.97086704e-01 8.79832566e-01 3.28508288e-01 -9.96181905e-01 2.44015515e-01 9.79147732e-01 1.17647910e+00 -1.33301818e+00 -4.52723563e-01 -3.82243454e-01 -9.65793550e-01 1.03181052e+00 5.80185950e-01 4.85977352e-01 2.39573017e-01 1.23409212e-01 3.28507155e-01 -3.61336708e-01 -1.14116275e+00 -4.97177035e-01 1.32477418e-01 6.61725104e-01 8.01857591e-01 -1.42322063e-01 -3.94156545e-01 7.10449219e-01 -6.30000234e-01 1.10205799e-01 4.04038727e-01 8.88302922e-01 -2.68947363e-01 -1.11421907e+00 -1.80396587e-01 1.29259348e-01 -3.67057651e-01 -2.88812131e-01 -9.97825682e-01 6.70195222e-01 -4.83546317e-01 1.85690260e+00 -1.01032801e-01 -4.03368443e-01 6.53491080e-01 4.53473508e-01 1.00163603e-03 -7.82000780e-01 -1.12360489e+00 -3.97704899e-01 9.50244427e-01 -5.35665095e-01 -4.04425949e-01 -7.33570874e-01 -1.09342337e+00 -4.17001873e-01 -2.55207628e-01 5.41608393e-01 6.43799186e-01 1.09532988e+00 5.45146704e-01 6.16360120e-02 3.53164911e-01 -8.14377248e-01 -5.30588508e-01 -1.24634898e+00 2.74764061e-01 6.24410033e-01 2.51972198e-01 -2.50746608e-01 -2.50429749e-01 -4.89447340e-02]

[12.368642807006836, 7.945367813110352]

47f6f5a0-95b7-4ae8-bf24-7bee9a12c0b9

do-not-train-it-a-linear-neural-architecture

2305.14065

https://arxiv.org/abs/2305.14065v3

https://arxiv.org/pdf/2305.14065v3.pdf

Do Not Train It: A Linear Neural Architecture Search of Graph Neural Networks

Neural architecture search (NAS) for Graph neural networks (GNNs), called NAS-GNNs, has achieved significant performance over manually designed GNN architectures. However, these methods inherit issues from the conventional NAS methods, such as high computational cost and optimization difficulty. More importantly, previous NAS methods have ignored the uniqueness of GNNs, where GNNs possess expressive power without training. With the randomly-initialized weights, we can then seek the optimal architecture parameters via the sparse coding objective and derive a novel NAS-GNNs method, namely neural architecture coding (NAC). Consequently, our NAC holds a no-update scheme on GNNs and can efficiently compute in linear time. Empirical evaluations on multiple GNN benchmark datasets demonstrate that our approach leads to state-of-the-art performance, which is up to $200\times$ faster and $18.8\%$ more accurate than the strong baselines.

['Haiqin Yang', 'Bei Yu', 'Yue Zhao', 'Jiaqi Sun', 'Xuanzhou Liu', 'Lin Zhang', 'Peng Xu']

2023-05-23

['architecture-search']

['methodology']

[ 1.08638786e-01 3.31779987e-01 -3.72534603e-01 -2.34243661e-01 -4.73550469e-01 -2.82156318e-01 1.66508108e-01 -3.87688935e-01 -3.77668262e-01 4.64047343e-01 1.06564343e-01 -5.86533189e-01 -2.25846320e-01 -6.87250495e-01 -8.70369256e-01 -5.89240432e-01 -1.10636607e-01 4.59962904e-01 -4.32842597e-02 -2.64473081e-01 5.55115975e-02 2.61094838e-01 -1.23515165e+00 -1.12780727e-01 6.18240535e-01 1.31976044e+00 1.37422532e-01 4.51617569e-01 -3.19196910e-01 9.21767354e-01 -2.29601175e-01 -5.30618370e-01 5.39041460e-01 -4.55790222e-01 -7.04857409e-01 -2.72303909e-01 7.84405529e-01 -3.51950377e-01 -8.69994044e-01 1.40643632e+00 3.94703537e-01 2.91478336e-01 3.46242309e-01 -1.16441131e+00 -1.13675022e+00 1.14279211e+00 -6.60799980e-01 2.40992293e-01 -3.87914866e-01 3.74194533e-02 1.54136062e+00 -9.30856407e-01 4.96328413e-01 1.18661892e+00 9.29280579e-01 7.19753802e-01 -1.27977836e+00 -8.50223780e-01 5.70906281e-01 1.31402448e-01 -1.59218359e+00 -6.23794496e-01 7.40172505e-01 1.02386139e-01 1.25472033e+00 -7.89083615e-02 5.47217190e-01 1.08615673e+00 -3.66258562e-01 7.96582222e-01 4.17107493e-01 -2.75260240e-01 2.03408882e-01 -5.15437722e-01 2.63899088e-01 1.18162560e+00 7.85325825e-01 8.36538002e-02 -4.27492917e-01 -1.08057819e-01 1.06201243e+00 1.71458572e-02 -1.57408938e-01 -4.54029590e-01 -8.56984377e-01 1.00415039e+00 9.83551860e-01 1.46552920e-01 -4.52434629e-01 8.48833680e-01 3.84773254e-01 2.94946283e-01 9.10545290e-02 4.45764869e-01 -4.79683042e-01 -4.60510105e-02 -8.17537367e-01 3.41929868e-02 8.22659314e-01 1.11917877e+00 7.13896334e-01 9.14852381e-01 2.45829746e-01 9.06450808e-01 3.15072358e-01 4.11592454e-01 5.04130661e-01 -1.05345619e+00 5.92623293e-01 7.99948037e-01 -8.15096915e-01 -1.23247600e+00 -4.05354470e-01 -9.39712703e-01 -1.46958363e+00 -1.84854314e-01 2.60530319e-03 -2.81088740e-01 -1.12267041e+00 2.05336070e+00 3.41331474e-02 5.37130654e-01 7.54200295e-02 7.49920189e-01 1.15556967e+00 5.74384630e-01 -1.41632617e-01 1.99772269e-01 9.00746703e-01 -1.46770310e+00 -3.77910554e-01 -6.79862440e-01 6.90848053e-01 -4.08408679e-02 1.06306875e+00 2.23661765e-01 -1.23833930e+00 -1.85289204e-01 -1.09767962e+00 -1.18151745e-02 -3.52198556e-02 6.60470128e-02 1.09751654e+00 6.87519073e-01 -1.63768721e+00 6.01865947e-01 -9.41353023e-01 -2.21204773e-01 6.94409192e-01 7.86815286e-01 -1.76858664e-01 -1.83684647e-01 -8.82091522e-01 3.77587825e-01 5.91111720e-01 2.73615927e-01 -1.06641746e+00 -8.04990947e-01 -1.03876603e+00 4.22403097e-01 5.99796891e-01 -8.00243795e-01 1.29743719e+00 -9.31566536e-01 -1.38156128e+00 3.95713270e-01 1.93491150e-02 -7.98219562e-01 -1.98317528e-01 -2.83732880e-02 -5.17727852e-01 1.02353439e-01 -2.41848931e-01 6.75018311e-01 5.68601906e-01 -1.06798625e+00 -2.49436736e-01 -8.84395849e-04 1.90435976e-01 1.15868896e-01 -8.71410370e-01 -3.65062177e-01 -1.00025535e+00 -7.53976583e-01 4.97340143e-01 -1.05364490e+00 -6.05337083e-01 -4.20463905e-02 -6.31803215e-01 -6.84360564e-02 4.86780912e-01 -4.19571072e-01 1.62004077e+00 -2.01994276e+00 2.33353749e-01 5.88722408e-01 8.70127797e-01 4.58024293e-01 -6.85028374e-01 2.54890677e-02 -8.55864398e-03 1.15318514e-01 -2.46973455e-01 -4.36123967e-01 1.61554903e-01 3.88312042e-01 -1.01506747e-01 1.07550777e-01 -1.13714471e-01 1.28229213e+00 -6.55400574e-01 -3.47511441e-01 -3.09687644e-01 4.68862206e-01 -9.18060780e-01 -2.65672892e-01 -3.38260293e-01 -4.03871983e-01 -3.70780855e-01 8.53894591e-01 3.71801317e-01 -1.08553565e+00 4.85199392e-01 -2.67688215e-01 5.09325862e-01 2.42231175e-01 -8.92305017e-01 1.88191521e+00 -1.70399815e-01 5.89673638e-01 1.95041478e-01 -1.06976950e+00 8.36193085e-01 -9.37927291e-02 2.71988928e-01 -8.23488772e-01 2.21828923e-01 2.53285199e-01 1.40626445e-01 1.78543329e-01 5.25469720e-01 3.61939877e-01 2.59261161e-01 5.04710555e-01 3.63786995e-01 4.81906742e-01 1.50483444e-01 5.03136396e-01 1.46066654e+00 -3.84315133e-01 1.07687436e-01 -3.96084458e-01 1.65472209e-01 -2.14413062e-01 5.80629230e-01 8.64916861e-01 -9.12147835e-02 5.01815856e-01 7.11209893e-01 -5.85190773e-01 -1.01931059e+00 -9.16344881e-01 5.32162726e-01 1.17897463e+00 -1.63726255e-01 -6.24249697e-01 -8.02362025e-01 -8.23917091e-01 -1.78973392e-01 3.40878576e-01 -6.66888356e-01 -2.62232542e-01 -8.69846225e-01 -7.68965304e-01 8.17580521e-01 8.05801690e-01 5.37914395e-01 -1.01251304e+00 -5.72449267e-02 2.50537008e-01 3.02915517e-02 -1.14629519e+00 -6.72038376e-01 1.54176503e-01 -1.19962609e+00 -8.56680155e-01 -6.24641597e-01 -1.14412749e+00 9.30404603e-01 3.47973555e-01 1.39258528e+00 6.97086215e-01 3.03106289e-02 -5.92154488e-02 -1.68393645e-02 -2.98907384e-02 -3.12387161e-02 6.56513214e-01 -3.40318158e-02 -2.42445514e-01 1.77340135e-01 -9.65472400e-01 -7.47452497e-01 -3.10095977e-02 -7.91821957e-01 -6.28215000e-02 1.01154685e+00 9.15728390e-01 8.16482544e-01 -5.09969220e-02 6.41262472e-01 -1.19726789e+00 7.07796991e-01 -4.31511372e-01 -7.35692143e-01 2.49692053e-01 -1.26990104e+00 4.03607547e-01 7.55508006e-01 -4.12623107e-01 -4.35741007e-01 -5.68041280e-02 -1.47179738e-01 -9.21156585e-01 4.09501970e-01 1.04985380e+00 2.18129531e-03 -5.75533271e-01 8.03912699e-01 2.67828465e-01 1.28299117e-01 -4.42016125e-01 3.49170625e-01 -2.31013656e-01 6.36642158e-01 -5.29704928e-01 1.02274692e+00 1.61299437e-01 1.38628870e-01 -5.13422906e-01 -8.45988333e-01 -1.84149295e-01 -5.53532317e-02 2.08537728e-01 2.46327490e-01 -1.08037543e+00 -6.38271391e-01 3.22404206e-01 -8.94420445e-01 -6.01079345e-01 -9.71416011e-02 2.57524669e-01 -8.37022737e-02 3.36248726e-01 -8.60920727e-01 -3.44049573e-01 -7.74565279e-01 -9.85630155e-01 4.35485154e-01 1.78597763e-01 1.21259943e-01 -1.09890509e+00 -1.60538092e-01 1.12650439e-01 9.18592632e-01 4.82071005e-02 1.24675643e+00 -6.04936481e-01 -8.35205257e-01 -3.39400247e-02 -6.97154880e-01 3.03995073e-01 -2.22634628e-01 -3.09984714e-01 -6.03382707e-01 -4.96467382e-01 -3.28815430e-01 -4.05193269e-01 1.18255556e+00 5.50675571e-01 1.49614227e+00 -7.70872176e-01 -3.42701435e-01 1.23181760e+00 1.84005952e+00 3.40177938e-02 6.77637875e-01 1.40910819e-01 1.17429435e+00 -7.20541850e-02 -2.77124375e-01 2.23801270e-01 4.84883755e-01 2.47426808e-01 8.98241401e-01 -1.37423158e-01 -2.85819560e-01 -3.83129984e-01 2.62702197e-01 1.30433238e+00 -2.80502558e-01 -4.42261100e-01 -9.77905750e-01 5.52267730e-01 -1.90092897e+00 -5.96552312e-01 3.31105202e-01 1.73249888e+00 6.67906880e-01 1.60066962e-01 9.82539915e-03 -2.00462341e-01 6.25215113e-01 5.35706878e-01 -1.04002297e+00 -2.85558313e-01 -1.69115052e-01 4.08256829e-01 7.96086192e-01 2.24393278e-01 -8.26850951e-01 1.05969274e+00 6.93960381e+00 8.20164323e-01 -8.97497714e-01 3.31493169e-02 6.72295868e-01 -3.89856100e-01 -5.53286850e-01 -1.90820038e-01 -9.78848875e-01 1.90976962e-01 1.19414163e+00 -1.89927876e-01 1.01883137e+00 1.20324087e+00 -5.40558279e-01 8.41351271e-01 -8.38994384e-01 1.17684817e+00 3.26905489e-01 -1.92078876e+00 4.35649067e-01 1.16321340e-01 9.73094404e-01 5.97144902e-01 2.43041381e-01 5.82404196e-01 7.98228979e-01 -1.35104370e+00 4.62181240e-01 8.57043639e-02 8.80361974e-01 -7.99607813e-01 5.03362477e-01 -1.12744235e-01 -1.59602523e+00 -2.59671539e-01 -6.31720662e-01 1.43350840e-01 -8.78208280e-02 2.75472432e-01 -5.87927759e-01 3.22920293e-01 5.75488150e-01 7.08268583e-01 -7.16682136e-01 1.04576504e+00 -7.72383735e-02 8.39948714e-01 -4.07738537e-01 -3.37680012e-01 6.74771190e-01 -9.82443988e-02 3.02009106e-01 7.92610228e-01 4.68537927e-01 2.88568996e-02 1.77509896e-02 1.07409680e+00 -7.02706754e-01 -1.05701730e-01 -5.01626670e-01 -4.11194652e-01 7.64645338e-01 1.04391956e+00 -7.27039993e-01 -4.62650582e-02 -4.97653097e-01 6.80543244e-01 8.06149721e-01 6.80753648e-01 -7.85941243e-01 -4.99542028e-01 6.69828475e-01 -1.21720694e-01 6.46628857e-01 -2.00506255e-01 -3.09367776e-01 -9.66779649e-01 7.10434318e-02 -1.14994955e+00 5.77348650e-01 -5.37708402e-01 -1.26032960e+00 1.04171610e+00 -3.54352891e-01 -7.79516518e-01 -1.94616467e-01 -6.62890017e-01 -4.02475089e-01 3.92960578e-01 -1.53959024e+00 -1.15590763e+00 -2.63734818e-01 6.65365219e-01 3.12386572e-01 -5.51991701e-01 7.45999098e-01 5.05396485e-01 -8.89845550e-01 1.28460336e+00 1.21824600e-01 5.02440512e-01 6.05490021e-02 -1.11356449e+00 9.07035947e-01 1.04940712e+00 4.32076156e-01 7.89764881e-01 1.77614585e-01 -4.52168196e-01 -1.73016000e+00 -1.36050713e+00 7.13350654e-01 5.22359945e-02 8.55397105e-01 -2.39026457e-01 -8.77487004e-01 1.02901173e+00 6.04084656e-02 2.71079212e-01 4.80726957e-01 4.49728668e-01 -8.02733243e-01 -2.21312135e-01 -6.60348833e-01 8.58374476e-01 1.71034455e+00 -5.33172548e-01 -1.58347845e-01 2.83396721e-01 1.36678147e+00 -6.05864525e-01 -5.67042291e-01 3.54496688e-01 3.67095917e-01 -6.74815357e-01 1.33216786e+00 -8.36938560e-01 4.31715131e-01 -7.62950704e-02 -3.29399526e-01 -1.09108138e+00 -7.01597095e-01 -7.43463337e-01 -6.79738104e-01 8.12511563e-01 8.71243298e-01 -9.05470729e-01 1.35974300e+00 6.53451264e-01 -4.82239187e-01 -1.06796658e+00 -7.86633074e-01 -9.49662149e-01 -3.56469899e-02 -3.58862400e-01 9.81061220e-01 9.61074769e-01 -5.81978142e-01 4.04965848e-01 -4.06054735e-01 2.67405450e-01 8.00870538e-01 -2.37063095e-02 4.34088230e-01 -1.32451212e+00 -4.99514669e-01 -8.41658890e-01 -4.25742567e-01 -1.29828513e+00 3.90978515e-01 -1.04378271e+00 -1.96435407e-01 -1.63568377e+00 1.33671358e-01 -6.33466661e-01 -7.86616504e-01 9.02301788e-01 2.74215210e-02 2.15105399e-01 2.21841838e-02 1.67325005e-01 -9.04667139e-01 5.70843279e-01 9.93857980e-01 -2.27931768e-01 -7.08980188e-02 -4.11333412e-01 -1.35139775e+00 7.06626296e-01 7.40726650e-01 -4.42941785e-01 -8.59385371e-01 -1.05045903e+00 6.24259889e-01 -2.50381559e-01 3.24949294e-01 -1.09586668e+00 6.76433086e-01 4.18372378e-02 -1.35579873e-02 -4.33461189e-01 2.20610961e-01 -6.39303088e-01 1.65846497e-01 4.61909533e-01 -2.85285205e-01 6.21009111e-01 1.42527640e-01 8.50405276e-01 -1.88694224e-01 -1.91982985e-01 4.54297304e-01 -1.32069543e-01 -9.68212128e-01 9.85026956e-01 2.04342484e-01 2.89439648e-01 4.03443784e-01 -2.18298852e-01 -7.07600892e-01 -3.78875256e-01 -3.60885292e-01 3.30142766e-01 2.51147181e-01 1.29466340e-01 8.85984838e-01 -1.50585711e+00 -4.32992727e-01 1.81608900e-01 -2.64182221e-02 3.30872655e-01 3.56840521e-01 7.26680815e-01 -7.14486659e-01 5.42690754e-01 4.06895168e-02 -2.25714415e-01 -8.45453620e-01 5.63813448e-01 3.27768803e-01 -4.38313782e-01 -7.76929796e-01 1.37390018e+00 2.10009605e-01 -4.86363053e-01 6.93172455e-01 -1.85973838e-01 -1.01929531e-01 -4.92358088e-01 3.91539395e-01 2.46647269e-01 1.65473238e-01 -2.81302065e-01 -2.39357173e-01 3.04224104e-01 -3.25713098e-01 2.93178350e-01 1.56155741e+00 6.93942904e-02 -1.62691236e-01 -1.85274422e-01 1.34275675e+00 -5.36102295e-01 -1.16085100e+00 -6.50676131e-01 -6.30335808e-02 -3.69113386e-02 3.24351788e-01 -5.98614037e-01 -1.96692121e+00 6.07158542e-01 3.54077250e-01 7.21237659e-02 1.25530219e+00 -1.19726092e-01 1.16733384e+00 9.69262421e-01 2.74815947e-01 -8.08791459e-01 1.94005892e-01 8.51870298e-01 7.06132531e-01 -8.66991520e-01 -1.40943453e-02 -2.02309385e-01 -3.95716786e-01 7.79166102e-01 9.61710572e-01 -2.66340286e-01 7.69234538e-01 1.81772187e-01 -1.51762918e-01 -6.01332247e-01 -1.04057145e+00 5.21177873e-02 3.48601043e-01 5.25394559e-01 6.69039274e-03 -9.73959193e-02 2.08356008e-01 7.56297648e-01 -2.97704607e-01 -2.72539586e-01 2.45535791e-01 6.01848900e-01 -2.07212016e-01 -8.99606824e-01 3.70369405e-01 8.61306071e-01 -3.60954911e-01 -7.17336059e-01 -2.56554395e-01 7.88246989e-01 -4.77829099e-01 5.11535466e-01 -9.52790380e-02 -7.42669761e-01 1.93284065e-01 -9.37897414e-02 1.84250370e-01 -4.94544685e-01 -6.31996334e-01 -2.15705305e-01 1.82859436e-01 -9.75828350e-01 -9.85584557e-02 -2.02997774e-01 -1.27454686e+00 -7.50423968e-01 -3.43936682e-01 5.24694882e-02 5.41020632e-01 6.33522332e-01 8.60438228e-01 6.50799811e-01 2.80467838e-01 -6.25265181e-01 -7.12945282e-01 -4.85062778e-01 -4.40784812e-01 -9.20592900e-03 2.45792791e-01 -3.71200413e-01 -4.84028220e-01 -3.68915081e-01]

[8.575203895568848, 3.3967554569244385]

55fec37a-7fc9-416a-b1b5-09098a2c0498

reinforcement-learning-approach-for-real-time

1602.04936

http://arxiv.org/abs/1602.04936v1

http://arxiv.org/pdf/1602.04936v1.pdf

Reinforcement Learning approach for Real Time Strategy Games Battle city and S3

In this paper we proposed reinforcement learning algorithms with the generalized reward function. In our proposed method we use Q-learning and SARSA algorithms with generalised reward function to train the reinforcement learning agent. We evaluated the performance of our proposed algorithms on two real-time strategy games called BattleCity and S3. There are two main advantages of having such an approach as compared to other works in RTS. (1) We can ignore the concept of a simulator which is often game specific and is usually hard coded in any type of RTS games (2) our system can learn from interaction with any opponents and quickly change the strategy according to the opponents and do not need any human traces as used in previous works. Keywords : Reinforcement learning, Machine learning, Real time strategy, Artificial intelligence.

['Harshit Sethy', 'Amit Patel']

2016-02-16

['real-time-strategy-games']

['playing-games']

[-2.16738030e-01 2.89982617e-01 3.35204214e-01 -1.13653038e-02 -1.19201951e-01 -7.13745296e-01 5.20896673e-01 8.17470402e-02 -1.07645595e+00 1.39786398e+00 -3.73070985e-01 -5.73920071e-01 -3.95239800e-01 -1.08127725e+00 -4.33243483e-01 -4.24096704e-01 -2.56241322e-01 9.99047577e-01 9.70423996e-01 -1.20312369e+00 7.62933254e-01 3.40318859e-01 -1.60918963e+00 -1.43564060e-01 8.06233823e-01 5.27553976e-01 5.28834760e-01 1.32852280e+00 -4.35262695e-02 1.30753624e+00 -9.81289268e-01 1.40821695e-01 5.56733847e-01 -7.33876705e-01 -1.04950035e+00 -2.50124544e-01 -9.05179381e-01 -3.55451554e-01 -2.24062189e-01 6.53169274e-01 7.50057459e-01 4.34998512e-01 6.28746927e-01 -1.34824038e+00 2.97494084e-01 7.26895392e-01 -5.53078473e-01 6.05692327e-01 6.31499350e-01 3.37889612e-01 3.83050740e-01 4.09534514e-01 5.25401711e-01 1.17504191e+00 2.72947848e-01 6.09380484e-01 -7.15218544e-01 -6.83430910e-01 -2.04954311e-01 5.21777987e-01 -1.03966594e+00 2.64898747e-01 5.61602056e-01 -5.66289164e-02 8.48400950e-01 1.11316346e-01 9.74876881e-01 7.61275649e-01 3.76210004e-01 6.64950192e-01 1.87104166e+00 -6.18170798e-01 7.42556334e-01 1.39091894e-01 -2.98102409e-01 5.08243740e-01 -1.87072366e-01 7.39295185e-01 1.49179280e-01 -2.11912654e-02 8.91300440e-01 -2.93105215e-01 4.08104122e-01 -2.30627015e-01 -6.88977420e-01 1.10599411e+00 4.32093553e-02 5.37776113e-01 -6.77810073e-01 3.52925926e-01 6.84361160e-01 8.88162434e-01 -4.10674334e-01 5.93869328e-01 -6.30801499e-01 -8.71623755e-01 -5.41299939e-01 6.51553869e-01 9.82577145e-01 5.56384504e-01 4.86334413e-01 5.42639732e-01 1.82400957e-01 5.36256909e-01 1.01136975e-01 3.48698258e-01 8.94351840e-01 -8.64789665e-01 4.04576249e-02 3.21453482e-01 6.52226448e-01 -4.79604304e-01 -7.46931612e-01 -1.83556408e-01 -6.04038723e-02 9.48555768e-01 3.64409447e-01 -8.02830100e-01 -6.18907750e-01 1.28541791e+00 3.42821270e-01 2.85366327e-01 5.10661185e-01 5.52955806e-01 7.15024352e-01 6.17607474e-01 -5.56818321e-02 -2.88112640e-01 1.04162443e+00 -6.39128923e-01 -6.14661992e-01 2.28247821e-01 4.00101513e-01 -7.15619504e-01 8.27766716e-01 8.12108696e-01 -1.26909268e+00 -4.68993992e-01 -1.16193593e+00 9.84595478e-01 -5.29321134e-01 -5.52474618e-01 7.54921913e-01 1.12590253e+00 -1.15487075e+00 8.05244088e-01 -6.24440074e-01 -4.95693684e-01 -2.15172082e-01 8.15906644e-01 1.75506771e-01 5.46469212e-01 -1.41956341e+00 1.21247363e+00 9.93547678e-01 -2.61603624e-01 -1.23830593e+00 1.64380357e-01 -3.61415535e-01 -8.54143500e-02 9.16810334e-01 -6.99665546e-02 1.66138780e+00 -1.27481413e+00 -2.41849208e+00 3.68501604e-01 6.83972776e-01 -6.04337871e-01 7.35476553e-01 2.55245775e-01 -3.07515085e-01 1.66022815e-02 -3.60944092e-01 4.15763766e-01 5.14225900e-01 -1.03679669e+00 -1.07777750e+00 7.05340281e-02 8.08050811e-01 5.07744551e-01 3.71655196e-01 -7.05047622e-02 4.03918952e-01 -3.14237118e-01 -4.69732076e-01 -9.20688987e-01 -5.38690805e-01 -1.12241375e+00 3.64238530e-01 -2.91763484e-01 6.38113797e-01 -3.79021801e-02 8.02316129e-01 -1.65559912e+00 8.63686427e-02 2.64588654e-01 -1.99491024e-01 4.37671751e-01 -1.81923896e-01 9.62118387e-01 8.41584057e-02 -1.94729820e-01 2.82938808e-01 6.79568946e-01 9.95080099e-02 5.69516599e-01 1.60591930e-01 6.56340122e-02 -4.63201016e-01 5.68632782e-01 -1.11058247e+00 -4.56170857e-01 2.08612561e-01 -1.33755684e-01 -4.27358896e-01 4.50172335e-01 -2.24118948e-01 4.91140395e-01 -7.11562634e-01 8.09397325e-02 4.02337044e-01 5.19603848e-01 3.48581970e-01 4.66520727e-01 -4.17166263e-01 1.47299662e-01 -1.64183760e+00 1.21069014e+00 -4.99344349e-01 6.10167831e-02 -1.39900789e-01 -1.47125304e+00 1.06094408e+00 5.97970009e-01 6.17246687e-01 -9.68571901e-01 6.51370823e-01 7.63405636e-02 5.47720194e-01 -5.32506704e-01 4.03768748e-01 -3.35834265e-01 -1.71870202e-01 8.37311268e-01 2.55668163e-01 -4.27863002e-01 4.76797760e-01 -1.88964739e-01 1.17699158e+00 4.35907781e-01 6.90336466e-01 -2.51556545e-01 7.71659017e-01 3.04513156e-01 5.94855785e-01 1.30397403e+00 -3.68633300e-01 -3.51239592e-01 7.19344795e-01 -4.09172833e-01 -8.45848441e-01 -8.08183968e-01 4.59423542e-01 1.20814574e+00 1.25311241e-01 -6.78740740e-02 -5.58608592e-01 -7.46155739e-01 -3.96258563e-01 6.73056304e-01 -5.07824779e-01 -1.78000145e-02 -4.82551396e-01 -5.23908257e-01 5.76409817e-01 5.48679121e-02 6.50710702e-01 -1.59071970e+00 -1.21006000e+00 6.50835812e-01 3.61522764e-01 -6.41368330e-01 1.26404449e-01 3.92589837e-01 -9.63903189e-01 -1.19183958e+00 -2.36619368e-01 -6.98503792e-01 8.57117921e-02 -5.54144047e-02 9.79430854e-01 1.68743670e-01 -1.39225543e-01 6.59344673e-01 -9.17279601e-01 -9.90227222e-01 -6.02160037e-01 -2.95013227e-02 -1.45480245e-01 -6.45712256e-01 2.20692307e-01 -6.80932283e-01 -5.46754599e-01 3.18979770e-01 -9.77591217e-01 -2.33968541e-01 6.80954933e-01 9.85765219e-01 -1.73351079e-01 7.64637709e-01 8.07958305e-01 -1.16878986e+00 1.24968183e+00 -3.23175222e-01 -1.10840631e+00 6.31701052e-02 -5.00132561e-01 2.69950807e-01 8.73165488e-01 -3.99078190e-01 -8.39378834e-01 -9.89500359e-02 -1.07569054e-01 4.07383919e-01 -1.55505210e-01 2.84937233e-01 3.05209279e-01 -2.87431061e-01 6.19039237e-01 3.93534124e-01 1.90458834e-01 -4.60855551e-02 4.17242721e-02 7.01181948e-01 -1.30671218e-01 -6.70174003e-01 7.94516623e-01 -2.27009490e-01 1.67797491e-01 -5.26291251e-01 2.70511955e-01 -1.43342361e-01 -3.20803523e-01 -4.55635160e-01 5.85520804e-01 -4.40233558e-01 -1.63354719e+00 5.81473410e-01 -6.23532772e-01 -6.13060772e-01 -2.62443066e-01 5.84808588e-01 -1.04664600e+00 2.67811924e-01 -5.62904060e-01 -1.29560673e+00 -1.24968708e-01 -1.07782924e+00 7.36041367e-02 8.00251782e-01 9.63281319e-02 -9.53420877e-01 4.35028464e-01 -6.61308765e-02 5.89837670e-01 2.13897750e-01 6.13176107e-01 -6.71917677e-01 -1.80971503e-01 1.58843324e-02 4.89623249e-01 4.54258807e-02 1.12939909e-01 -1.76471040e-01 -4.27231312e-01 -3.01226616e-01 1.62770689e-01 -9.16594565e-01 -6.35944530e-02 2.25828528e-01 5.45014322e-01 -1.78811610e-01 1.71056718e-01 -9.45053548e-02 1.62049937e+00 1.22458267e+00 9.56034601e-01 9.03089583e-01 -2.62697875e-01 3.99024785e-01 1.33021522e+00 8.77414763e-01 2.40895182e-01 6.75232410e-01 4.98426616e-01 6.24107430e-04 6.61466837e-01 9.88904387e-04 5.65936625e-01 4.51418519e-01 -6.79189742e-01 4.83374037e-02 -7.59399533e-01 1.62213534e-01 -2.10408807e+00 -1.31197596e+00 1.82548344e-01 2.23260236e+00 7.48521566e-01 5.77256680e-01 8.82587671e-01 3.66724461e-01 4.46322471e-01 -2.06644401e-01 -3.39037985e-01 -1.13397396e+00 2.47971043e-01 8.80077720e-01 8.32644999e-01 7.69595742e-01 -7.40814030e-01 1.25457251e+00 6.29590845e+00 7.93741167e-01 -1.07705629e+00 2.26750284e-01 6.10062554e-02 -1.77683383e-01 1.37758642e-01 1.75407648e-01 -2.82597601e-01 3.10913324e-01 1.00126755e+00 -4.97420043e-01 8.74733448e-01 7.03682840e-01 6.55902505e-01 -7.05868423e-01 -4.05341804e-01 6.31099999e-01 -4.55561101e-01 -8.90846610e-01 -4.75676030e-01 -2.26034209e-01 3.81678700e-01 -2.31269553e-01 -2.44970873e-01 9.02485013e-01 1.05961657e+00 -1.12266886e+00 3.92928928e-01 1.41312823e-01 1.26796111e-01 -1.37285757e+00 1.20798016e+00 5.82564533e-01 -9.17502284e-01 -3.19459975e-01 -5.16634703e-01 -5.71793973e-01 -2.11487874e-01 -5.83845675e-01 -1.18763638e+00 7.34102726e-01 4.29362029e-01 -1.34515613e-01 -2.69173741e-01 1.19536674e+00 -1.61496758e-01 7.52363741e-01 -3.06771815e-01 -8.40125382e-01 8.37380826e-01 -4.24195558e-01 2.75011122e-01 7.11150050e-01 1.10822074e-01 4.09044713e-01 2.30542406e-01 2.65689015e-01 8.29134583e-01 3.04756910e-01 -5.38983941e-01 1.41775221e-01 3.28488857e-01 9.44574296e-01 -1.10126293e+00 -2.47825548e-01 -4.92884479e-02 7.96069562e-01 -1.40386865e-01 1.59180626e-01 -9.04389560e-01 -6.23512387e-01 -2.27146149e-02 9.85040590e-02 4.19241458e-01 -1.43425956e-01 4.16456282e-01 -4.62184072e-01 -6.38740182e-01 -1.11384463e+00 4.57634062e-01 -5.41490793e-01 -6.98674142e-01 7.61646986e-01 2.18348980e-01 -1.24732351e+00 -7.01408863e-01 -5.52385807e-01 -6.35037601e-01 5.84717691e-01 -1.20737970e+00 -6.24083459e-01 -5.23446947e-02 8.40450943e-01 4.98047203e-01 -8.46609950e-01 8.07489932e-01 -7.12222094e-03 -1.32653728e-01 2.08567783e-01 2.20384911e-01 -1.61841050e-01 2.98841804e-01 -1.51292443e+00 -2.85500288e-01 1.90487206e-01 -2.79764384e-01 9.80876852e-03 1.11465204e+00 -5.05870402e-01 -1.23125184e+00 -1.26697540e-01 -5.22422008e-02 2.25725785e-01 7.63768554e-01 -2.09290668e-01 -2.40868121e-01 3.47561449e-01 6.64789200e-01 -4.43828136e-01 4.53262895e-01 -1.15943862e-04 6.22852981e-01 -2.80172259e-01 -1.31027186e+00 5.70317864e-01 5.04495263e-01 6.80346489e-02 -7.75740683e-01 2.08347589e-01 1.33407235e-01 -4.48184192e-01 -5.20628273e-01 -5.86602278e-02 3.48410159e-01 -1.28424919e+00 5.94935775e-01 -8.33418071e-01 -1.37230337e-01 -3.49584401e-01 2.68137962e-01 -1.64111209e+00 -1.94664672e-01 -9.75474656e-01 4.66606826e-01 7.28050351e-01 9.33223888e-02 -7.96559691e-01 1.11808777e+00 7.61922523e-02 4.12049741e-01 -3.90353769e-01 -9.52990174e-01 -8.77173126e-01 1.82641163e-01 -3.26207504e-02 6.02155268e-01 5.90789258e-01 2.95679599e-01 2.70396799e-01 -6.34089291e-01 -2.21505344e-01 2.23373070e-01 -2.57595956e-01 8.49364638e-01 -9.93090868e-01 -1.08300531e+00 -2.38771871e-01 -7.43838549e-01 -4.53689337e-01 -2.03119278e-01 -2.69809246e-01 -4.92684171e-02 -1.26296520e+00 -2.23522767e-01 -6.62646592e-01 -5.25004208e-01 3.15209895e-01 -3.37939635e-02 -3.42431158e-01 2.45641544e-01 -3.49367261e-01 -7.38950193e-01 5.02614021e-01 1.30111468e+00 1.65786579e-01 -4.66697693e-01 4.66570556e-01 -3.11080545e-01 5.10754287e-01 1.06455410e+00 -6.61594748e-01 -9.23663974e-01 2.90126890e-01 2.03052908e-01 8.22591245e-01 -1.08807072e-01 -1.20432460e+00 7.97039419e-02 -7.37113416e-01 -8.23642388e-02 -2.33027056e-01 3.11322697e-02 -1.00855577e+00 1.26980260e-01 1.14481544e+00 1.09422468e-01 5.89349866e-01 2.64273345e-01 4.75133300e-01 -1.21869847e-01 -8.73182178e-01 7.21743405e-01 -6.46250486e-01 -8.64669025e-01 -3.60073566e-01 -1.17843211e+00 -3.16905901e-02 1.71910584e+00 -3.90669286e-01 -4.03353944e-02 -9.88625050e-01 -5.84018707e-01 3.07021797e-01 1.70939103e-01 1.48284376e-01 3.02488208e-01 -8.08505177e-01 -5.56650698e-01 -4.17380601e-01 -4.20019895e-01 -6.50566459e-01 1.82256237e-01 4.82794076e-01 -1.13642144e+00 1.79996967e-01 -1.26338387e+00 -3.34338434e-02 -1.29596484e+00 5.69486082e-01 6.62572384e-01 -6.63283288e-01 -3.22321713e-01 1.34208456e-01 -4.19066131e-01 -5.40531099e-01 1.37807831e-01 1.75704345e-01 -9.30738211e-01 -2.29062870e-01 2.15927899e-01 3.52686644e-01 -2.46985257e-01 -6.75596371e-02 -1.44904166e-01 2.17856497e-01 1.30164092e-02 -7.91128218e-01 1.53409469e+00 2.65792549e-01 2.07816303e-01 5.14991939e-01 2.30757505e-01 -1.07555762e-01 -9.50107813e-01 2.70200908e-01 2.32173681e-01 -4.36635703e-01 -7.82685578e-02 -9.60642874e-01 -7.75259256e-01 3.96377385e-01 1.11517060e+00 4.28856581e-01 1.09643328e+00 -7.81241715e-01 4.18052971e-01 6.11470044e-01 8.68641317e-01 -1.75195849e+00 2.80949950e-01 7.28340209e-01 5.68675637e-01 -1.04736960e+00 -1.93867926e-02 1.26052842e-01 -1.10082734e+00 1.20876288e+00 8.26911271e-01 -6.05387092e-01 4.54771399e-01 4.88750517e-01 2.21253574e-01 -1.46271497e-01 -9.44990337e-01 -5.45686185e-01 -6.81136727e-01 1.04915023e+00 1.85661376e-01 5.72490804e-02 -1.10133088e+00 2.95527339e-01 -3.21545362e-01 5.59805453e-01 1.33232796e+00 1.51846683e+00 -7.22135007e-01 -1.83997107e+00 -5.02236485e-01 2.38503948e-01 -4.65344012e-01 3.40368062e-01 -2.34325796e-01 1.13215041e+00 2.12154299e-01 1.12358165e+00 -3.10651213e-01 -3.85040969e-01 5.80437720e-01 -2.54777610e-01 8.76474202e-01 -5.02819955e-01 -1.15774310e+00 -9.13736522e-02 1.75081342e-01 -4.74134773e-01 -3.23669463e-01 -5.59263706e-01 -1.44769323e+00 -4.24694985e-01 -2.22902689e-02 7.25994706e-01 5.98131180e-01 9.34751511e-01 -2.02944696e-01 6.15515828e-01 7.94725537e-01 -4.33390886e-01 -7.23675668e-01 -1.03256249e+00 -9.30403292e-01 1.76690638e-01 -2.46800005e-01 -9.81071115e-01 4.04914618e-02 -6.73706114e-01]

[3.5862066745758057, 1.5481213331222534]

79691bcf-65e4-4d3b-97dd-f5f0fb4951a9

identifying-and-extracting-rare-disease

2306.12656

https://arxiv.org/abs/2306.12656v1

https://arxiv.org/pdf/2306.12656v1.pdf

Identifying and Extracting Rare Disease Phenotypes with Large Language Models

Rare diseases (RDs) are collectively common and affect 300 million people worldwide. Accurate phenotyping is critical for informing diagnosis and treatment, but RD phenotypes are often embedded in unstructured text and time-consuming to extract manually. While natural language processing (NLP) models can perform named entity recognition (NER) to automate extraction, a major bottleneck is the development of a large, annotated corpus for model training. Recently, prompt learning emerged as an NLP paradigm that can lead to more generalizable results without any (zero-shot) or few labeled samples (few-shot). Despite growing interest in ChatGPT, a revolutionary large language model capable of following complex human prompts and generating high-quality responses, none have studied its NER performance for RDs in the zero- and few-shot settings. To this end, we engineered novel prompts aimed at extracting RD phenotypes and, to the best of our knowledge, are the first the establish a benchmark for evaluating ChatGPT's performance in these settings. We compared its performance to the traditional fine-tuning approach and conducted an in-depth error analysis. Overall, fine-tuning BioClinicalBERT resulted in higher performance (F1 of 0.689) than ChatGPT (F1 of 0.472 and 0.591 in the zero- and few-shot settings, respectively). Despite this, ChatGPT achieved similar or higher accuracy for certain entities (i.e., rare diseases and signs) in the one-shot setting (F1 of 0.776 and 0.725). This suggests that with appropriate prompt engineering, ChatGPT has the potential to match or outperform fine-tuned language models for certain entity types with just one labeled sample. While the proliferation of large language models may provide opportunities for supporting RD diagnosis and treatment, researchers and clinicians should critically evaluate model outputs and be well-informed of their limitations.

['Hua Xu', 'Paul A. Harris', 'Yan Hu', 'Cathy Shyr']

2023-06-22

['prompt-engineering', 'named-entity-recognition-ner']

['natural-language-processing', 'natural-language-processing']

[ 1.20498240e-01 1.82923213e-01 -1.06938697e-01 -3.62073094e-01 -1.07139754e+00 -5.79517126e-01 2.79844373e-01 5.57772100e-01 -4.60637122e-01 8.68929267e-01 1.19947232e-01 -4.26213950e-01 -2.98088372e-01 -6.95157111e-01 -5.21541238e-01 -2.37465620e-01 8.62190407e-03 7.85289168e-01 -1.43343598e-01 2.10990787e-01 -1.77143306e-01 2.94620186e-01 -1.20777154e+00 3.53668988e-01 1.24148548e+00 3.90832335e-01 2.14140922e-01 8.24196160e-01 -1.62249789e-01 7.10657895e-01 -8.58364105e-01 -5.51936328e-01 -1.12964377e-01 -3.52551371e-01 -8.13611209e-01 -3.33436459e-01 1.16430648e-01 -2.03708261e-01 1.28226072e-01 5.70705891e-01 9.51986313e-01 -9.31809098e-02 4.69708174e-01 -9.31211233e-01 -7.12588251e-01 5.57784796e-01 -1.41237825e-01 6.19900487e-02 6.22814178e-01 6.61013246e-01 9.60726380e-01 -5.84394097e-01 8.97243142e-01 1.02128792e+00 9.63747263e-01 8.10496390e-01 -1.61743891e+00 -5.16691387e-01 -1.81149859e-02 -1.49483100e-01 -1.22007871e+00 -4.61635232e-01 -3.21975276e-02 -5.63024580e-01 1.45912826e+00 2.74035424e-01 5.37015378e-01 1.37069273e+00 2.23992944e-01 4.58403468e-01 1.06097186e+00 -2.02272758e-01 2.27528647e-01 1.47213429e-01 2.05282167e-01 5.49451649e-01 2.83859581e-01 4.03034836e-02 -4.06665206e-01 -6.75068319e-01 4.52672005e-01 -5.22264875e-02 -1.84688315e-01 4.54979718e-01 -1.21967089e+00 6.20191574e-01 -2.84461558e-01 2.28368253e-01 -6.26412213e-01 -1.54827237e-01 3.22577238e-01 2.62317687e-01 5.34568489e-01 1.02437174e+00 -9.86826181e-01 -5.09472489e-01 -9.79201555e-01 2.09190175e-01 1.13207340e+00 7.85664499e-01 2.55662471e-01 -4.60420996e-02 -5.81887186e-01 1.03050113e+00 -2.28613257e-01 5.78471661e-01 3.72557193e-01 -6.83138549e-01 3.11409086e-01 7.91192353e-01 1.44221395e-01 -8.25076580e-01 -8.41555536e-01 -4.04399514e-01 -7.51303732e-01 -4.30241942e-01 5.67787707e-01 -6.90261006e-01 -8.76422167e-01 1.93309712e+00 1.95955679e-01 3.45405459e-01 1.00911148e-01 5.33302367e-01 8.57471287e-01 5.01676619e-01 6.43568695e-01 -1.26370236e-01 1.50858307e+00 -3.13511312e-01 -4.15177643e-01 -3.04059535e-01 9.76063669e-01 -7.58213580e-01 1.22131753e+00 4.57556784e-01 -8.05119395e-01 -1.61936626e-01 -5.30123234e-01 9.41766985e-03 -2.42410228e-01 2.08920240e-01 7.84664035e-01 5.89842021e-01 -8.74375105e-01 6.56356275e-01 -9.72828507e-01 -8.14511061e-01 5.39384484e-01 4.45157111e-01 -4.47815120e-01 -2.93630570e-01 -1.14834547e+00 8.54062140e-01 7.16558769e-02 -2.93416262e-01 -6.06300056e-01 -1.31054878e+00 -6.06349409e-01 6.99879378e-02 3.64640534e-01 -1.06502891e+00 1.19029248e+00 -3.98898244e-01 -1.12557662e+00 9.35460567e-01 -3.08505356e-01 -4.25215125e-01 1.96046367e-01 -1.32885620e-01 -7.10980415e-01 1.74515814e-01 2.17622861e-01 6.29162610e-01 2.39322037e-01 -4.19802397e-01 -6.02528811e-01 -2.60321170e-01 -2.64547408e-01 -1.82168797e-01 -1.98681742e-01 5.60928345e-01 -5.79164028e-02 -5.90331316e-01 -4.08617049e-01 -6.98578954e-01 -3.28961909e-01 -5.82754314e-02 -4.50388163e-01 -3.74685854e-01 1.76496625e-01 -7.63202310e-01 1.23989260e+00 -1.80799699e+00 -3.10445487e-01 -1.98888302e-01 4.90032524e-01 6.14834428e-01 -3.46018106e-01 5.37184358e-01 -1.45681560e-01 6.21079087e-01 -1.39678791e-01 -4.27371338e-02 -1.34148747e-01 3.01394053e-02 2.28262669e-03 1.33795321e-01 8.42335939e-01 1.06851828e+00 -1.08731616e+00 -3.20547223e-01 -1.00132801e-01 6.45875037e-01 -8.74885082e-01 3.15070540e-01 -3.78004521e-01 4.74378467e-01 -5.10526419e-01 6.20644987e-01 1.86370268e-01 -5.89571536e-01 3.04410905e-01 6.98651224e-02 -1.05832733e-01 4.14161980e-01 -8.72185946e-01 1.18796444e+00 -3.24189693e-01 3.33749324e-01 -1.33758947e-01 -6.50383353e-01 8.27451766e-01 5.29758453e-01 6.40323818e-01 -5.07055283e-01 -1.20941393e-01 3.06553930e-01 2.86265850e-01 -8.86881232e-01 -1.24293882e-02 -3.27838868e-01 -5.20775206e-02 4.22571868e-01 1.11735752e-02 1.81797922e-01 2.63174027e-01 8.67170021e-02 1.73223042e+00 -1.66412160e-01 4.23959732e-01 -2.51855962e-02 6.01977631e-02 2.61498719e-01 9.47154105e-01 8.69701087e-01 -1.77643616e-02 5.97904742e-01 6.57994986e-01 -3.77054602e-01 -8.55150223e-01 -7.54327595e-01 -3.00827444e-01 8.54613662e-01 -7.10568488e-01 -7.22258925e-01 -5.80188394e-01 -5.87388873e-01 1.85765475e-01 8.84283364e-01 -2.75396317e-01 -1.54961288e-01 -2.90419489e-01 -1.17388570e+00 1.18344653e+00 4.17496204e-01 5.02999164e-02 -1.19749904e+00 -4.63618279e-01 5.40982604e-01 -2.46537164e-01 -1.18156147e+00 -3.58781219e-01 2.12939307e-01 -6.63298070e-01 -1.26310647e+00 -7.14388490e-01 -5.51070333e-01 5.39244294e-01 -4.56549138e-01 1.29477429e+00 -3.33181359e-02 -6.29247308e-01 3.55284840e-01 -3.70474845e-01 -4.96996760e-01 -4.78103399e-01 1.75582156e-01 1.18826345e-01 -3.57318401e-01 7.76941538e-01 -5.37075698e-01 -4.78542119e-01 1.03765748e-01 -7.09472060e-01 -9.91943525e-04 8.37681174e-01 1.05143464e+00 5.41142881e-01 -1.77974433e-01 9.23208654e-01 -1.31001341e+00 8.34626794e-01 -7.31528044e-01 -3.62158179e-01 4.02295858e-01 -8.84192228e-01 8.50327387e-02 8.99987757e-01 -6.57642186e-01 -9.24169838e-01 -3.36734615e-02 -4.11151022e-01 9.92654040e-02 -7.00697064e-01 6.34017587e-01 7.33750593e-03 2.98206568e-01 8.33566725e-01 7.20063448e-02 -2.24435136e-01 -8.03365648e-01 1.09428570e-01 8.15487146e-01 2.96393484e-01 -7.50853360e-01 3.18266660e-01 -8.31152275e-02 -2.27522016e-01 -8.29042673e-01 -8.16605628e-01 -3.81206989e-01 -1.23198651e-01 3.82091314e-01 8.41924846e-01 -8.15571129e-01 -9.07238662e-01 4.93086725e-01 -8.97712231e-01 -6.16845548e-01 -2.68676966e-01 4.23002571e-01 -2.99137920e-01 2.08042905e-01 -7.42623985e-01 -5.65511882e-01 -6.25026286e-01 -9.17572498e-01 1.09995353e+00 2.17294738e-01 -8.58766496e-01 -9.41051960e-01 2.32909888e-01 2.90932089e-01 2.68948108e-01 2.39002958e-01 1.41026628e+00 -1.17946601e+00 -1.34270102e-01 -1.71130210e-01 -2.51054704e-01 2.99258213e-02 2.57840812e-01 1.09393476e-02 -9.43463147e-01 -1.29385162e-02 -1.99558973e-01 -3.94096613e-01 4.24253792e-01 3.12627673e-01 8.76388788e-01 -4.18607414e-01 -4.49959993e-01 6.08166218e-01 1.23684287e+00 2.67600983e-01 3.97290647e-01 -2.89518293e-02 4.76256013e-01 6.15450442e-01 4.95038688e-01 4.71226573e-01 4.49220181e-01 4.98621374e-01 -4.07632649e-01 -9.29566100e-02 -1.03489444e-01 -4.01319593e-01 2.69180745e-01 3.44680637e-01 2.95043707e-01 -4.18922454e-01 -1.37064207e+00 6.05680048e-01 -1.47001433e+00 -7.88485050e-01 -1.78898036e-01 2.12213588e+00 1.30295062e+00 -5.53820701e-03 1.23712778e-01 -2.59390563e-01 5.94242811e-01 -3.15797001e-01 -7.63357580e-01 -4.78796571e-01 -3.64092916e-01 5.50417542e-01 7.67105594e-02 1.15903653e-01 -6.08825684e-01 9.28248584e-01 6.61363077e+00 6.00606143e-01 -1.04126978e+00 -3.99322398e-02 7.51322150e-01 -2.84293383e-01 -1.42414868e-01 -2.51442604e-02 -1.03052342e+00 6.47016644e-01 1.44340992e+00 -1.91277698e-01 4.84767824e-01 5.43692172e-01 5.00023961e-01 2.39917636e-02 -1.37074363e+00 8.91349494e-01 -2.79302031e-01 -1.26015544e+00 -1.55988514e-01 2.82624718e-02 7.26676464e-01 1.75504223e-01 -1.82886615e-01 4.63903219e-01 6.22694433e-01 -1.22409213e+00 1.18282735e-01 6.53030634e-01 1.12041295e+00 -5.36378920e-01 6.95486903e-01 4.71703440e-01 -7.64468670e-01 -1.27303109e-01 -3.41622472e-01 2.07041539e-02 2.20034137e-01 9.51678276e-01 -1.35056281e+00 3.48311305e-01 5.61452925e-01 5.64004421e-01 -4.41023350e-01 1.09531999e+00 -1.97707027e-01 1.08080757e+00 -3.83942157e-01 -4.18061353e-02 -1.77140296e-01 2.32403055e-01 4.74693924e-01 1.37113357e+00 3.30169559e-01 4.68520105e-01 2.21695215e-01 9.22856390e-01 -9.89975333e-02 2.85406321e-01 -3.23473036e-01 -7.60450184e-01 6.76787198e-01 1.19019794e+00 -3.56508583e-01 -3.11359733e-01 -4.09574389e-01 6.46457672e-01 4.44576979e-01 3.72548908e-01 -5.64624250e-01 -4.62122262e-01 7.01654255e-01 2.14739203e-01 -5.12611158e-02 3.21867824e-01 -3.03711295e-01 -1.03421140e+00 -1.95072696e-01 -1.31535661e+00 6.84352577e-01 -7.34830916e-01 -1.62611163e+00 5.68160474e-01 -3.42457116e-01 -7.39464760e-01 -3.46741974e-01 -6.14776433e-01 -3.91365051e-01 1.06733406e+00 -1.04881310e+00 -9.63540196e-01 -5.51075041e-02 2.87339449e-01 2.67833799e-01 -1.42211076e-02 1.23878467e+00 3.45993906e-01 -9.67026234e-01 7.75822282e-01 -8.49344060e-02 1.24786355e-01 1.19655812e+00 -1.20013976e+00 4.41431880e-01 5.58172464e-01 -1.74829409e-01 1.03625345e+00 6.11515582e-01 -1.12395740e+00 -1.09076798e+00 -1.29101574e+00 1.37994730e+00 -6.70469165e-01 5.73314905e-01 -2.99778849e-01 -9.24409211e-01 6.51992202e-01 -3.60189945e-01 -3.06157678e-01 1.06648314e+00 4.67287779e-01 -2.92861283e-01 4.24686261e-02 -1.39826214e+00 6.31037593e-01 1.08217180e+00 -5.59159279e-01 -4.84756231e-01 5.20424485e-01 8.15274477e-01 -1.41839936e-01 -1.38291168e+00 3.64031941e-01 4.87405717e-01 -5.39799869e-01 8.36877644e-01 -1.07137644e+00 3.83385599e-01 -3.78713235e-02 1.27742946e-01 -1.29944360e+00 -3.80229652e-01 -8.25576544e-01 1.99704506e-02 1.41618693e+00 8.01652730e-01 -8.66147578e-01 6.08710289e-01 1.11394572e+00 -7.95746669e-02 -9.63081002e-01 -3.54472548e-01 -6.27581894e-01 1.81243084e-02 -4.56626773e-01 7.41021156e-01 9.98791397e-01 7.46302158e-02 4.15325582e-01 -2.83131957e-01 1.96985647e-01 2.75251985e-01 -2.02934608e-01 5.25759518e-01 -1.49178112e+00 -5.66162586e-01 -2.34721571e-01 -1.44327134e-01 -4.88041371e-01 -1.25923371e-02 -9.54241037e-01 -1.39595181e-01 -1.66121721e+00 4.68016058e-01 -6.97012663e-01 -1.53011724e-01 1.02127838e+00 -5.87831497e-01 -1.35366797e-01 -2.11478844e-01 -3.62419561e-02 -4.01929080e-01 8.13129246e-02 8.63924444e-01 8.91660824e-02 -5.15676856e-01 -5.58991618e-02 -1.20635760e+00 5.11361837e-01 1.03734767e+00 -7.85599113e-01 -3.65731984e-01 -3.39287698e-01 3.42470884e-01 5.44473380e-02 2.90338039e-01 -6.13903999e-01 2.82379240e-01 -2.55508125e-01 3.81471127e-01 -2.65489250e-01 9.32431743e-02 -1.73383996e-01 3.66585076e-01 4.76077706e-01 -3.93865705e-01 8.59476626e-03 4.05458182e-01 2.90325433e-01 2.40795404e-01 -3.70563865e-02 5.05516112e-01 -2.14559257e-01 -3.30603808e-01 2.34502330e-01 -5.85974157e-01 5.42602718e-01 7.05938399e-01 -7.84280673e-02 -5.70980966e-01 -1.55670047e-01 -8.15763533e-01 2.90285766e-01 3.70734692e-01 2.26918161e-01 3.49818826e-01 -6.52943730e-01 -9.46630180e-01 7.96756968e-02 1.93619609e-01 -5.02441712e-02 3.79943192e-01 1.07367575e+00 -3.02228421e-01 7.19305992e-01 5.94887771e-02 -4.15581822e-01 -1.19785738e+00 4.63226318e-01 1.27065212e-01 -5.84402502e-01 -6.10800564e-01 8.20112824e-01 9.65842456e-02 -6.74010158e-01 7.90264979e-02 -3.64316672e-01 -5.94842480e-03 4.07076487e-03 6.26777470e-01 4.42530841e-01 4.77109030e-02 -7.35318363e-02 -4.22320366e-01 2.11696550e-01 -2.14173198e-01 1.69535220e-01 1.62145591e+00 3.74339312e-01 -1.65998876e-01 2.99915671e-01 7.58578360e-01 2.03239784e-01 -9.02042925e-01 1.77017555e-01 3.48904669e-01 -5.30224629e-02 -3.25152814e-01 -1.40722382e+00 -6.67647958e-01 5.52816927e-01 2.80658156e-01 -1.03497453e-01 1.11315620e+00 2.89856866e-02 7.60218024e-01 5.07179737e-01 4.02665257e-01 -7.93424189e-01 -3.70421112e-01 2.84149766e-01 4.69772220e-01 -9.14786875e-01 -2.27664083e-01 -3.20938021e-01 -6.18028641e-01 6.40309572e-01 6.60771847e-01 3.20912480e-01 1.99619532e-01 5.46169937e-01 4.28990051e-02 -2.36357674e-01 -1.29001379e+00 -6.21369630e-02 2.01868907e-01 9.63193178e-01 7.49229610e-01 1.58885866e-01 -4.39454079e-01 1.24056482e+00 -2.28061542e-01 2.72090316e-01 2.92681515e-01 5.73341966e-01 -1.12136036e-01 -1.50208592e+00 -1.81834862e-01 1.06267428e+00 -7.91182160e-01 -3.98086429e-01 -7.14759529e-01 6.31804764e-01 2.39540920e-01 1.21767163e+00 -2.73722738e-01 -2.96754628e-01 4.40126210e-01 4.81101602e-01 1.99789450e-01 -1.13632345e+00 -9.96616125e-01 -9.58735868e-02 5.66329777e-01 -6.08315766e-01 6.62842989e-02 -6.47984624e-01 -1.30521750e+00 -2.02431038e-01 -1.76953971e-01 8.11122283e-02 1.75449312e-01 8.04963648e-01 9.43850935e-01 4.35827106e-01 6.96721897e-02 1.47396192e-01 -5.59528708e-01 -8.63335133e-01 -4.15568769e-01 3.39116842e-01 -8.86933587e-04 -2.88800061e-01 -1.23604350e-01 -5.30362092e-02]

[8.5006685256958, 8.577235221862793]

4ed565ff-4d68-402a-bbbf-f29567c0a8ea

locally-smoothed-gaussian-process-regression

2210.09998

https://arxiv.org/abs/2210.09998v1

https://arxiv.org/pdf/2210.09998v1.pdf

Locally Smoothed Gaussian Process Regression

We develop a novel framework to accelerate Gaussian process regression (GPR). In particular, we consider localization kernels at each data point to down-weigh the contributions from other data points that are far away, and we derive the GPR model stemming from the application of such localization operation. Through a set of experiments, we demonstrate the competitive performance of the proposed approach compared to full GPR, other localized models, and deep Gaussian processes. Crucially, these performances are obtained with considerable speedups compared to standard global GPR due to the sparsification effect of the Gram matrix induced by the localization operation.

['Maurizio Filippone', 'Bogdan Kozyrskiy', 'Davit Gogolashvili']

2022-10-18

['gpr', 'gpr']

['computer-vision', 'miscellaneous']

[-1.50579259e-01 -1.07917935e-01 4.41546410e-01 -9.44512710e-02 -1.19090319e+00 -3.64631154e-02 8.19781423e-01 4.66659129e-01 -3.91510338e-01 2.92516083e-01 2.82996148e-01 -1.27505928e-01 -1.15788572e-01 -6.70906544e-01 -7.00727284e-01 -1.08225453e+00 -2.76087046e-01 6.97969615e-01 3.55121158e-02 2.60538220e-01 3.71718317e-01 4.30997938e-01 -9.12403822e-01 -1.69721022e-01 6.70738101e-01 9.05061483e-01 2.19042271e-01 5.85241854e-01 2.19034195e-01 6.77487671e-01 -2.60051668e-01 -3.24481636e-01 9.25986618e-02 2.16522053e-01 -1.72649741e-01 -1.80368781e-01 1.46261659e-02 -1.31190136e-01 -4.08265650e-01 1.05847740e+00 4.75734651e-01 6.39496505e-01 7.90269971e-01 -9.47608948e-01 -6.89862311e-01 4.37895447e-01 -1.19241810e+00 5.62286973e-02 1.41842574e-01 -2.11450577e-01 1.02085698e+00 -1.61481726e+00 8.78195390e-02 1.45774865e+00 9.86408710e-01 1.75999682e-02 -1.39023721e+00 -4.36233819e-01 3.95187110e-01 -3.70267242e-01 -1.61938369e+00 -2.14666069e-01 3.65546405e-01 -3.18857044e-01 6.85956717e-01 -1.73941538e-01 2.22248673e-01 1.33856750e+00 3.07927936e-01 8.11185002e-01 9.35653806e-01 -1.58138022e-01 4.31054533e-01 -2.91989952e-01 5.19201636e-01 4.76000458e-01 3.41035753e-01 5.96299432e-02 -6.54744565e-01 -6.75108969e-01 8.44558060e-01 2.40071565e-01 -3.15382659e-01 -4.05495435e-01 -1.05925345e+00 9.19662476e-01 2.87139922e-01 -2.56948620e-01 -8.31760347e-01 4.77309346e-01 3.46118324e-02 -1.11269794e-01 7.90026546e-01 8.15222855e-04 -4.50673521e-01 -3.30384374e-01 -1.06754923e+00 1.91278040e-01 1.07550073e+00 1.12560129e+00 8.77139747e-01 -1.21956483e-01 -2.80971199e-01 7.56793916e-01 5.06428421e-01 7.24760890e-01 2.84652710e-01 -5.12573779e-01 5.25402606e-01 8.65479484e-02 4.68255170e-02 -1.07898140e+00 -3.16259772e-01 -4.87329900e-01 -9.99275088e-01 -2.50480741e-01 1.97143808e-01 -5.16257405e-01 -7.11720884e-01 1.52418911e+00 3.33278984e-01 6.88266575e-01 1.15552515e-01 3.11333537e-01 2.44232565e-01 7.98569798e-01 5.19761086e-01 1.29777774e-01 1.19681811e+00 -1.13866639e+00 -4.60802853e-01 -2.00322464e-01 3.68985385e-01 -6.97757900e-01 7.30009437e-01 4.15403247e-01 -9.78071153e-01 -6.59658611e-01 -7.04104483e-01 -1.46410570e-01 -2.11548358e-02 2.76967973e-01 7.46680796e-01 4.82045650e-01 -1.14581871e+00 8.09614122e-01 -1.28325486e+00 -1.39485374e-01 3.16843390e-01 2.32611522e-01 -2.48879388e-01 -1.04941465e-02 -7.72585213e-01 6.29125476e-01 1.35622054e-01 4.17985708e-01 -9.18221951e-01 -1.00599182e+00 -7.01076210e-01 2.81308770e-01 1.35995686e-01 -7.49885440e-01 1.14133799e+00 -1.96480229e-01 -1.26097453e+00 3.29372644e-01 -6.26758814e-01 -5.59879601e-01 5.35648525e-01 -8.32445323e-01 -1.41390294e-01 6.19125776e-02 -4.83785607e-02 9.91677493e-02 1.19101167e+00 -1.03757644e+00 -7.94177294e-01 -4.98365045e-01 -5.54190695e-01 2.45138898e-01 9.22184845e-05 3.33240747e-01 -9.39951479e-01 -5.78435302e-01 2.36048192e-01 -1.08994687e+00 -8.40970933e-01 -5.15902460e-01 -3.40083033e-01 -6.20349050e-02 6.23291016e-01 -9.59312201e-01 9.90458012e-01 -2.61177707e+00 9.17146504e-02 5.71668983e-01 4.99552399e-01 -2.19934016e-01 2.75123511e-02 5.74036062e-01 -5.07451966e-02 -4.75260839e-02 -8.90577883e-02 -1.03965366e+00 4.24020179e-02 1.16121426e-01 -6.11400902e-01 6.13210380e-01 2.01057851e-01 8.21778119e-01 -9.94785905e-01 -4.66052443e-02 -7.86895456e-04 8.71521413e-01 -3.34683388e-01 4.82697673e-02 2.94722021e-01 4.58710968e-01 -5.44102669e-01 5.36897779e-01 8.51340711e-01 -3.40860128e-01 -1.20850019e-01 -2.94615254e-02 1.11867145e-01 1.24026351e-01 -1.28623009e+00 1.33502793e+00 -4.51137215e-01 3.00373971e-01 2.23514333e-01 -5.76701522e-01 8.72219920e-01 2.99392253e-01 4.66423303e-01 -2.32191738e-02 -9.40044075e-02 9.43783820e-02 -5.25783539e-01 3.06815326e-01 7.13426411e-01 -8.04993287e-02 1.36719663e-02 2.15174362e-01 -1.80892050e-02 7.85096884e-02 -1.27588421e-01 1.01117551e-01 1.19029987e+00 2.54330695e-01 2.93680131e-01 -3.57069045e-01 4.13424462e-01 -3.45801055e-01 2.76354253e-01 1.05889893e+00 -7.95481950e-02 6.61490142e-01 4.84510660e-01 -1.45263061e-01 -7.48409331e-01 -1.30305791e+00 5.94333373e-02 1.29202425e+00 -1.18925162e-01 -5.15129268e-01 -5.12313962e-01 -5.49400330e-01 1.82653666e-02 5.06132483e-01 -6.16319835e-01 -7.24899918e-02 -6.27362490e-01 -1.23601389e+00 5.08277833e-01 9.54387069e-01 2.08142444e-01 -5.79520285e-01 1.03413738e-01 2.77279347e-01 1.83509991e-01 -1.28755915e+00 -3.64045382e-01 4.09608305e-01 -1.26476490e+00 -7.21652329e-01 -7.51461327e-01 -7.05173552e-01 7.38056362e-01 4.65509772e-01 7.24296033e-01 -5.78789532e-01 3.62881362e-01 2.58815050e-01 -1.40055194e-01 -2.66510606e-01 2.01513115e-02 4.28476408e-02 8.49341676e-02 2.91387793e-02 4.84051883e-01 -6.93564653e-01 -4.56157953e-01 -5.88533469e-02 -5.55471718e-01 -2.67594308e-01 7.85393596e-01 6.83592081e-01 7.52808690e-01 3.87778193e-01 5.39843477e-02 -8.77133012e-01 1.03685737e+00 -7.30571032e-01 -5.22390544e-01 2.30500894e-03 -3.31762254e-01 1.23674542e-01 3.44796002e-01 -5.22617102e-01 -1.15758944e+00 -5.91229647e-02 -2.12312862e-01 -5.62985957e-01 2.76039913e-02 5.80359638e-01 -9.77054238e-02 -1.25601217e-01 3.69672865e-01 1.73822924e-01 -1.34051114e-01 -7.11777389e-01 6.25417829e-01 2.61323780e-01 5.00582218e-01 -9.72777605e-01 8.91691446e-01 7.51753747e-01 2.08321169e-01 -1.05283141e+00 -7.38452435e-01 -8.55242491e-01 -6.19279385e-01 3.96520883e-01 6.57009304e-01 -1.29384255e+00 -6.02739751e-01 5.20115495e-01 -1.12609923e+00 -2.85989642e-01 -2.11730376e-01 6.18446469e-01 -5.26131153e-01 5.86806417e-01 -1.02063656e+00 -9.06410158e-01 -4.88599330e-01 -9.75869596e-01 1.34131372e+00 3.69892530e-02 -1.18843719e-01 -1.30689359e+00 2.29698583e-01 -1.23962760e-01 2.44638115e-01 2.45933607e-02 6.98636353e-01 -1.05348229e+00 -2.86809474e-01 -5.05735338e-01 -3.44576627e-01 5.23678422e-01 -1.38517559e-01 -9.72740352e-03 -9.69301879e-01 -2.61179060e-01 4.79341924e-01 3.00740600e-01 8.25099468e-01 5.44468999e-01 9.24192369e-01 5.54926023e-02 -4.65174973e-01 7.92961299e-01 1.44807434e+00 -2.63261080e-01 6.02525115e-01 1.22166686e-01 1.13866127e+00 3.86887580e-01 4.70314175e-01 4.40356761e-01 2.61077195e-01 2.55779386e-01 -6.62297308e-02 -2.56753027e-01 2.44650871e-01 -5.32159150e-01 5.56682885e-01 1.00823092e+00 -4.32746470e-01 2.98521519e-01 -9.01696086e-01 2.66266406e-01 -2.07106495e+00 -5.81166267e-01 -6.95500314e-01 2.48722720e+00 3.99935901e-01 -7.16334656e-02 -2.61394709e-01 -3.87959778e-01 6.00884438e-01 1.28957957e-01 -2.59086601e-02 -1.52155504e-01 -4.69988212e-02 6.39191210e-01 9.38951135e-01 5.82816660e-01 -1.42029095e+00 9.90964949e-01 7.66193581e+00 1.06132245e+00 -7.06556499e-01 3.71610016e-01 4.91964251e-01 9.72719714e-02 2.71550447e-01 1.13958828e-01 -1.23444390e+00 2.79246569e-01 1.23524117e+00 -1.60384640e-01 2.30353728e-01 1.30799413e+00 2.86125392e-01 1.62412282e-02 -1.14807642e+00 8.41967165e-01 -6.32149279e-02 -4.73893821e-01 2.35579666e-02 3.58492166e-01 8.22350442e-01 3.10443044e-01 3.04871053e-01 5.81013799e-01 8.79769683e-01 -9.27520633e-01 4.59988058e-01 4.92927551e-01 6.96706697e-02 -8.75828087e-01 5.90969324e-01 3.17962855e-01 -1.29956949e+00 8.91135260e-02 -8.28351676e-01 3.57690826e-02 2.74387568e-01 9.38112795e-01 -8.11858594e-01 7.77035832e-01 4.47946578e-01 4.99092996e-01 -4.14037228e-01 1.03319025e+00 -7.68370271e-01 6.96148813e-01 -3.64836276e-01 4.84291136e-01 4.44764227e-01 -5.81637025e-01 5.30291140e-01 1.51009321e+00 7.42755592e-01 -4.77062345e-01 2.02478394e-02 6.86823308e-01 -1.15186563e-02 1.67910725e-01 -3.47642303e-01 1.89135522e-01 3.34688276e-01 1.36754823e+00 -3.48082662e-01 -3.30872595e-01 -7.11547852e-01 1.15294302e+00 4.66736883e-01 9.06147301e-01 -9.72114265e-01 9.32047442e-02 8.63896012e-01 -1.24983221e-01 6.31355524e-01 -6.41956389e-01 -1.68268993e-01 -1.03860545e+00 -1.54617906e-01 -6.13041520e-01 2.61211842e-01 -6.62049055e-01 -1.63248658e+00 3.15771818e-01 -1.95878237e-01 -6.49607360e-01 -8.63260180e-02 -5.29758334e-01 -5.49956381e-01 1.41059196e+00 -1.46598792e+00 -1.23437941e+00 -4.36437726e-02 5.11164367e-01 2.57865667e-01 3.54622722e-01 7.80777216e-01 1.59819841e-01 -6.03218615e-01 2.50430256e-01 4.82660353e-01 -6.36740327e-02 7.84087181e-01 -1.47314548e+00 9.61177409e-01 1.03891575e+00 1.34774312e-01 1.29336965e+00 6.42172813e-01 -8.21667552e-01 -1.35166490e+00 -1.19072974e+00 7.21070409e-01 -6.37947500e-01 1.14893544e+00 -3.07348013e-01 -1.00146687e+00 1.08537161e+00 -3.43815744e-01 -1.88112676e-01 7.97195017e-01 6.90360546e-01 -3.43506932e-01 1.86473206e-01 -7.65313745e-01 6.71866596e-01 5.62744737e-01 -5.55006683e-01 -6.85679138e-01 9.95986834e-02 6.17639720e-01 -2.79811859e-01 -1.02200675e+00 3.18522960e-01 2.58366704e-01 -3.99571240e-01 1.39305902e+00 -4.87413049e-01 3.45192283e-01 -2.52973288e-01 -2.56832331e-01 -1.26640952e+00 -6.71917737e-01 -8.44917834e-01 -5.50061166e-01 1.05814290e+00 4.03748274e-01 -6.41258776e-01 6.79598510e-01 6.89738929e-01 -1.12446718e-01 -5.61696947e-01 -5.74295580e-01 -6.55500531e-01 7.55794346e-02 -6.80019200e-01 2.72289664e-01 5.93203902e-01 -2.41574168e-01 3.66962820e-01 -7.05464244e-01 8.08627009e-01 9.50878680e-01 -2.76829243e-01 8.85949910e-01 -1.28757727e+00 -5.85855007e-01 -2.23670214e-01 -2.63265163e-01 -1.59069633e+00 -1.19796395e-01 -5.10728061e-01 2.62307584e-01 -1.48012686e+00 3.17619979e-01 -4.86568987e-01 -4.17306274e-01 4.91701290e-02 -8.36317956e-01 8.19028914e-02 9.37674567e-02 6.83712125e-01 -6.33844912e-01 6.57133698e-01 9.81648505e-01 4.08991158e-01 -3.18942189e-01 2.60939360e-01 -9.64396834e-01 1.11365676e+00 5.89131653e-01 -4.33612555e-01 -1.61522463e-01 -2.97814697e-01 2.42546186e-01 -1.83601543e-01 2.46360958e-01 -1.03599608e+00 5.63933134e-01 2.40852594e-01 4.93272871e-01 -7.35838413e-01 5.69982052e-01 -6.58253789e-01 9.35009643e-02 8.86916593e-02 -3.67124565e-02 1.91516913e-02 5.20592779e-02 1.17279756e+00 -2.50300139e-01 -1.42546341e-01 5.56381762e-01 1.27347872e-01 -6.76940680e-01 4.21576828e-01 -3.67067724e-01 -1.97652310e-01 8.79389763e-01 1.12410888e-01 3.80590819e-02 -2.86556661e-01 -9.36266005e-01 7.13023096e-02 3.26990604e-01 4.43470031e-02 4.55921173e-01 -1.08766198e+00 -5.71119070e-01 2.19711199e-01 -1.64646775e-01 1.35304704e-01 2.95885712e-01 1.40645957e+00 -3.11377048e-01 2.53813654e-01 4.71603781e-01 -4.44006622e-01 -1.00694561e+00 7.14531660e-01 -2.33519554e-01 -7.77984083e-01 -8.72735381e-01 8.98701191e-01 6.67249560e-01 -2.34461606e-01 2.34278083e-01 -6.16430223e-01 1.23192787e-01 -1.58291802e-01 7.50757277e-01 6.41557574e-01 1.19326010e-01 -3.75274330e-01 -5.33298701e-02 4.64950621e-01 -4.89967167e-02 -2.50818521e-01 1.25480139e+00 -2.15286076e-01 -1.11390211e-01 5.14776289e-01 9.20397758e-01 4.67009217e-01 -1.51982629e+00 -2.74401039e-01 1.69323802e-01 -2.47467235e-01 7.20984414e-02 -1.49793372e-01 -6.79017782e-01 9.29473639e-01 1.07436962e-01 -8.52424055e-02 7.88683116e-01 -4.52809408e-02 4.68301296e-01 3.25407505e-01 2.61004508e-01 -1.02829790e+00 -2.07029894e-01 7.99441695e-01 4.76843327e-01 -7.44108140e-01 -1.08277611e-01 -5.65825760e-01 -6.47504628e-01 8.35887849e-01 -9.84656438e-02 -5.30297339e-01 9.58033144e-01 3.33469272e-01 -3.03377777e-01 -2.47098938e-01 -5.19342899e-01 -1.56662688e-01 8.15157145e-02 6.64417326e-01 2.41543055e-01 1.13153763e-01 -6.30945293e-03 7.43267715e-01 2.10550260e-02 -2.58171797e-01 1.97744854e-02 8.42618644e-01 -5.18150568e-01 -9.86025155e-01 -5.99802673e-01 2.00563267e-01 -7.10520089e-01 -6.82455003e-01 1.72128305e-01 8.26592565e-01 -4.96314466e-01 9.50292408e-01 5.93836745e-03 -5.61594069e-02 1.71501935e-01 1.50441960e-01 1.92873389e-01 -6.79071367e-01 -4.33583319e-01 5.34414053e-01 -5.31932004e-02 -6.34625554e-01 1.46804461e-02 -7.27407753e-01 -1.11343324e+00 -4.28098261e-01 -1.32925734e-01 4.55878899e-02 9.02448416e-01 8.01709473e-01 3.73719484e-01 4.79553878e-01 3.54798734e-01 -9.69806612e-01 -1.00464916e+00 -1.09512460e+00 -9.09605086e-01 2.56496847e-01 -1.78899802e-02 -5.32054186e-01 -6.25602722e-01 2.77353544e-03]

[6.980855464935303, 3.763324737548828]