paperID
stringlengths
36
36
pwc_id
stringlengths
8
47
arxiv_id
stringlengths
6
16
nips_id
float64
url_abs
stringlengths
18
329
url_pdf
stringlengths
18
742
title
stringlengths
8
325
abstract
stringlengths
1
7.27k
authors
stringlengths
2
7.06k
published
stringlengths
10
10
conference
stringlengths
12
47
conference_url_abs
stringlengths
16
198
conference_url_pdf
stringlengths
27
199
proceeding
stringlengths
6
47
taskID
stringlengths
7
1.44k
areaID
stringclasses
688 values
embedding
stringlengths
9.26k
12.5k
umap_embedding
stringlengths
29
44
ab2aaa99-7126-41fe-9b38-80290269be7f
anda-a-novel-data-augmentation-technique
1910.01256
null
https://arxiv.org/abs/1910.01256v1
https://arxiv.org/pdf/1910.01256v1.pdf
ANDA: A Novel Data Augmentation Technique Applied to Salient Object Detection
In this paper, we propose a novel data augmentation technique (ANDA) applied to the Salient Object Detection (SOD) context. Standard data augmentation techniques proposed in the literature, such as image cropping, rotation, flipping, and resizing, only generate variations of the existing examples, providing a limited generalization. Our method has the novelty of creating new images, by combining an object with a new background while retaining part of its salience in this new context; To do so, the ANDA technique relies on the linear combination between labeled salient objects and new backgrounds, generated by removing the original salient object in a process known as image inpainting. Our proposed technique allows for more precise control of the object's position and size while preserving background information. Aiming to evaluate our proposed method, we trained multiple deep neural networks and compared the effect that our technique has in each one. We also compared our method with other data augmentation techniques. Our findings show that depending on the network improvement can be up to 14.1% in the F-measure and decay of up to 2.6% in the Mean Absolute Error.
['Bruno A. Krinski', 'Eduardo Todt', 'Daniel V. Ruiz']
2019-10-03
null
null
null
null
['image-cropping']
['computer-vision']
[ 8.19587648e-01 3.52030218e-01 5.10958880e-02 1.47861764e-02 -1.85755238e-01 -3.42856556e-01 6.91986382e-01 4.30024683e-01 -5.81210852e-01 7.64021337e-01 1.88275352e-01 2.34177932e-02 2.79598206e-01 -7.45299995e-01 -9.97639894e-01 -9.10137475e-01 1.19496271e-01 1.32088372e-02 5.59400320e-01 -1.79297552e-01 3.38167340e-01 7.65082181e-01 -1.77420938e+00 1.01099551e-01 7.81090796e-01 9.20310616e-01 4.04774457e-01 4.14509058e-01 -2.04215363e-01 9.29158032e-01 -8.56491983e-01 -1.15122810e-01 5.71144164e-01 -4.53661829e-01 -7.27911592e-01 3.64058465e-01 7.10983276e-01 -3.48314613e-01 -1.38531372e-01 8.77140820e-01 4.26611662e-01 2.08335444e-01 3.37843299e-01 -1.14041841e+00 -6.86959386e-01 4.12166953e-01 -1.04242802e+00 3.29374731e-01 1.23668075e-01 -9.46571305e-02 4.96221989e-01 -8.63909960e-01 7.81294048e-01 9.16463554e-01 5.52716017e-01 7.30932415e-01 -1.35731137e+00 -4.77238923e-01 2.38844976e-01 7.36012608e-02 -1.08266544e+00 -3.76416773e-01 1.09413338e+00 -2.61069089e-01 5.05262792e-01 4.28367555e-01 7.32176661e-01 8.39515567e-01 -7.30116665e-02 7.59292483e-01 1.24016035e+00 -9.40278292e-01 2.67938554e-01 2.70654470e-01 -2.39822678e-02 3.67535144e-01 5.02783716e-01 -1.27650753e-01 -3.28548461e-01 -1.29889976e-02 8.57091963e-01 5.81608154e-02 -1.56124219e-01 -6.61498189e-01 -1.14526427e+00 5.97631991e-01 5.82842588e-01 3.74958813e-01 -4.45237696e-01 1.24089807e-01 2.90106505e-01 -1.63062707e-01 5.25036275e-01 5.18629670e-01 -2.44159326e-01 3.31815600e-01 -1.01796150e+00 4.00235802e-01 2.87344605e-01 7.61615694e-01 7.57107377e-01 2.40243390e-01 -3.55188131e-01 6.31603956e-01 -1.37612090e-01 2.67162800e-01 7.23528683e-01 -7.57627785e-01 1.76144630e-01 6.88001037e-01 3.19257081e-01 -1.17874372e+00 -4.25850958e-01 -4.40607399e-01 -6.73364103e-01 5.95929980e-01 5.22004187e-01 -2.12301910e-01 -1.25541079e+00 1.97346926e+00 5.65318227e-01 2.89201766e-01 -2.98773032e-02 7.15177894e-01 5.51504612e-01 5.02730489e-01 2.21848190e-01 -2.90978491e-01 1.24386537e+00 -1.02656853e+00 -8.66634429e-01 -4.32567835e-01 2.97354460e-01 -1.02761137e+00 1.06537974e+00 2.85239160e-01 -1.31290781e+00 -6.72604978e-01 -1.17730296e+00 -8.86665881e-02 -5.00534058e-01 9.48743597e-02 4.30446357e-01 5.73233485e-01 -1.15130317e+00 5.91452003e-01 -6.45446360e-01 -4.70121205e-01 4.35212404e-01 3.44936609e-01 -3.81301343e-01 2.69103289e-01 -8.22884560e-01 1.12003505e+00 3.83094788e-01 -2.26471573e-01 -6.15105987e-01 -6.88220918e-01 -7.29242265e-01 1.45369366e-01 4.50656056e-01 -5.57870567e-01 1.03631103e+00 -1.28009975e+00 -1.20700884e+00 8.66484463e-01 -1.81536242e-01 -7.93963969e-01 5.67757607e-01 -5.14181077e-01 -1.61087468e-01 1.36389390e-01 4.38064374e-02 9.55889583e-01 1.19355786e+00 -1.51383746e+00 -5.51460922e-01 -2.72656411e-01 1.00868933e-01 8.19837004e-02 -3.16244423e-01 9.32074338e-02 -1.67614743e-01 -1.07155848e+00 7.74068981e-02 -8.86504591e-01 -3.18352342e-01 9.07194242e-02 -1.53354511e-01 3.71994555e-01 1.36653256e+00 -7.90039241e-01 1.14377272e+00 -2.29180479e+00 -2.62202206e-03 -7.52491206e-02 3.24308962e-01 6.30652249e-01 -1.62446320e-01 3.18562835e-01 -4.12017077e-01 2.08188873e-02 -4.80085671e-01 -4.62360084e-01 -4.15369481e-01 1.40758023e-01 -3.72293949e-01 2.98559844e-01 6.32868111e-01 6.20944262e-01 -8.39200377e-01 -4.66993690e-01 4.05956686e-01 6.26812696e-01 -4.98282641e-01 -3.32227759e-02 -9.26470682e-02 3.12513918e-01 1.01850867e-01 4.37047571e-01 9.48776186e-01 7.96264634e-02 -6.02613240e-02 -1.65694803e-01 -2.41814807e-01 6.30115718e-02 -1.27661598e+00 1.40277243e+00 -2.66068012e-01 7.35936284e-01 -4.72555980e-02 -9.05379951e-01 1.07124758e+00 -3.31106223e-02 3.84734243e-01 -6.58478439e-01 1.42888099e-01 -3.87349129e-02 1.05121560e-01 -3.09604615e-01 9.21293437e-01 -2.13888008e-02 3.44206959e-01 2.95379579e-01 -2.22228065e-01 2.68293411e-01 3.38330418e-01 1.39185935e-01 9.03131127e-01 2.07376495e-01 4.61207896e-01 -1.72323972e-01 3.99379820e-01 -1.85958609e-01 4.62613732e-01 7.51105070e-01 -2.68368006e-01 8.79156530e-01 3.93795460e-01 -4.43427771e-01 -1.30973446e+00 -7.79863179e-01 3.24278809e-02 9.11395550e-01 2.09225222e-01 -1.99666396e-02 -9.02997315e-01 -6.88779712e-01 -1.06732227e-01 6.73024833e-01 -9.43934798e-01 -3.24933082e-01 -7.66032279e-01 -8.67134690e-01 1.72722667e-01 5.66176474e-01 6.84272468e-01 -1.26763165e+00 -1.09362268e+00 1.08452283e-01 -1.42394193e-02 -1.05842221e+00 -3.46838176e-01 2.15527937e-01 -9.92747605e-01 -8.35075319e-01 -9.07735765e-01 -8.38882327e-01 9.89280522e-01 5.34558654e-01 9.85601366e-01 1.66373789e-01 -3.35142881e-01 1.13774324e-02 -4.96533006e-01 -6.38038754e-01 -4.95082885e-01 -1.02569766e-01 -7.21764788e-02 2.25586280e-01 2.06761993e-02 -5.98870635e-01 -6.16737425e-01 6.76915273e-02 -1.30929375e+00 2.92035729e-01 7.48610497e-01 7.18386352e-01 3.85586470e-01 -8.41180608e-02 5.60350060e-01 -9.28354502e-01 4.11018372e-01 -1.84119299e-01 -5.42387366e-01 1.90453697e-02 -4.83172864e-01 1.07122421e-01 4.26881582e-01 -6.78990543e-01 -1.10459626e+00 3.89109462e-01 1.32985607e-01 -4.67459142e-01 -2.07553595e-01 2.57860664e-02 6.42900961e-03 -3.48848522e-01 7.62673557e-01 2.04001322e-01 -8.44000280e-03 -4.51470584e-01 5.15863180e-01 2.34751672e-01 6.34584844e-01 -2.09045112e-01 8.83952439e-01 7.55598903e-01 1.61146149e-01 -6.26522481e-01 -7.69337296e-01 -2.63635516e-01 -8.48943412e-01 -1.18249379e-01 5.99328101e-01 -5.99181533e-01 -1.71028346e-01 4.96887892e-01 -1.16310561e+00 -1.10877819e-01 -7.41137743e-01 2.00706691e-01 -2.98115075e-01 4.60444540e-01 -2.04173937e-01 -8.90564680e-01 -2.98991889e-01 -8.55752885e-01 8.47128332e-01 3.94661963e-01 -2.23356292e-01 -6.38404250e-01 -1.60186499e-01 1.40299471e-02 6.94434166e-01 7.25906253e-01 7.72066653e-01 -6.19582236e-01 -4.12194461e-01 -2.60656595e-01 -3.10720205e-01 4.89702255e-01 3.88562351e-01 -6.10610545e-02 -1.01567614e+00 -2.47453481e-01 1.48181170e-01 7.99766406e-02 8.44909370e-01 4.40854907e-01 1.12567389e+00 -3.48553747e-01 -2.84952879e-01 1.45936877e-01 1.38876402e+00 3.97160351e-01 8.79019737e-01 6.70066237e-01 5.82970619e-01 5.35292685e-01 6.27542555e-01 3.92346203e-01 -2.96881884e-01 6.86012447e-01 6.72326148e-01 -5.83186328e-01 -5.55642486e-01 4.47731512e-03 5.03485985e-02 2.12248400e-01 -1.43175870e-01 7.84915015e-02 -5.58568299e-01 7.99128830e-01 -1.71990991e+00 -9.34111893e-01 2.37835571e-02 2.44680977e+00 8.60943675e-01 3.17167759e-01 2.60219634e-01 3.80150914e-01 8.95102143e-01 1.62206754e-01 -3.43651026e-01 -4.08560365e-01 -1.77103505e-01 3.67143005e-01 6.81753993e-01 4.17831153e-01 -1.13119030e+00 7.45155632e-01 6.44363928e+00 5.47228575e-01 -1.31037521e+00 -8.20197016e-02 5.93736827e-01 -2.17104211e-01 -1.25027085e-02 3.23859565e-02 -6.60743773e-01 4.98647034e-01 5.02587557e-01 -8.22715834e-02 2.67873973e-01 8.81397307e-01 2.38074675e-01 -4.56466168e-01 -7.13296413e-01 7.16601074e-01 4.29162890e-01 -1.17424750e+00 2.63247788e-01 -2.86058784e-02 7.82952547e-01 -4.69356984e-01 1.18309990e-01 8.21919590e-02 -1.59360051e-01 -5.60928822e-01 8.16417217e-01 3.46927136e-01 3.34746569e-01 -8.45724165e-01 7.55230546e-01 1.58053904e-03 -6.59895003e-01 -1.37080774e-01 -1.20088339e-01 -3.36958081e-01 8.78664106e-03 5.98221838e-01 -1.07322359e+00 3.09961289e-01 5.83733618e-01 1.93715587e-01 -8.13486278e-01 1.10320437e+00 -7.92130679e-02 4.22029942e-01 -2.52822757e-01 2.16010332e-01 3.48010547e-02 -3.28186452e-02 6.38013184e-01 1.12573946e+00 2.19711438e-01 -2.38049537e-01 -1.55735254e-01 9.24515426e-01 -1.60439417e-01 1.55264020e-01 -5.18244028e-01 2.44948044e-01 3.43638659e-01 1.28987384e+00 -1.00057709e+00 -5.74537575e-01 -3.31128418e-01 1.01515830e+00 1.28713235e-01 1.36992857e-01 -8.22087646e-01 -5.38262486e-01 2.95768321e-01 5.01011193e-01 4.87097144e-01 4.61172611e-02 -3.91755968e-01 -8.02815318e-01 3.23810130e-01 -6.69752896e-01 6.74040094e-02 -9.68652725e-01 -8.57733011e-01 6.05002880e-01 2.01051861e-01 -1.17051494e+00 -2.24498324e-02 -4.06226873e-01 -4.97463226e-01 8.07415783e-01 -1.47929358e+00 -1.19173598e+00 -5.33929348e-01 2.70852357e-01 5.65726995e-01 7.18722306e-03 5.56088269e-01 3.96403760e-01 -5.31828761e-01 3.39566797e-01 -5.89629263e-02 -3.61556262e-02 6.73585474e-01 -1.20777965e+00 5.36798060e-01 1.24341357e+00 -4.03306866e-03 5.22868514e-01 1.10322857e+00 -5.28750062e-01 -7.52637327e-01 -1.08736610e+00 7.67600119e-01 -1.95195898e-01 3.62354577e-01 -3.56811851e-01 -1.14099801e+00 6.42851710e-01 3.64032954e-01 1.09833844e-01 1.66285947e-01 -2.46394768e-01 -2.02188268e-01 -1.62537903e-01 -1.46431684e+00 7.75425375e-01 6.87032104e-01 3.56702097e-02 -5.49947917e-01 8.05629231e-03 7.67235160e-01 -4.50003266e-01 -2.21343011e-01 5.23405254e-01 3.10645103e-01 -1.08981872e+00 1.01467097e+00 -5.05499125e-01 5.80871880e-01 -5.61922967e-01 9.06102657e-02 -1.11813998e+00 -3.44728798e-01 -4.64387268e-01 -2.23369494e-01 1.29385281e+00 6.51745424e-02 -4.58962560e-01 9.29744482e-01 4.62773681e-01 -8.92800540e-02 -5.69167554e-01 -7.57473826e-01 -5.46515822e-01 -2.70584404e-01 -1.20013794e-02 3.93972427e-01 9.69126225e-01 -3.61090690e-01 -1.09070409e-02 -6.00969136e-01 1.96364343e-01 4.72043395e-01 -4.21700440e-02 8.88331532e-01 -1.09720659e+00 -1.44891888e-01 -3.22774500e-01 -4.76168066e-01 -4.86591876e-01 -2.97149122e-01 -4.61036772e-01 -1.76556572e-01 -1.25215471e+00 3.22089344e-01 -1.97747454e-01 -4.17355627e-01 6.60541236e-01 -3.27960759e-01 4.83496517e-01 5.20411253e-01 -5.29393293e-02 -1.93638310e-01 3.32271010e-01 1.20937586e+00 -3.61610844e-04 -4.14814740e-01 -1.92973316e-01 -8.25460792e-01 8.40665698e-01 9.29928541e-01 -4.81007665e-01 -2.36540049e-01 -3.25916946e-01 -8.09856355e-02 -5.28117836e-01 5.63594580e-01 -1.28331590e+00 -2.07269847e-01 4.09456007e-02 6.89647794e-01 -4.95650023e-01 3.56432259e-01 -8.73212278e-01 -1.11226641e-01 5.92366040e-01 -2.42996886e-01 3.54348123e-02 8.24838936e-01 3.84032249e-01 -9.52080041e-02 -3.84037971e-01 1.07650661e+00 -7.18064085e-02 -7.65324831e-01 -3.67965966e-01 -2.81624436e-01 -3.19704980e-01 1.22129035e+00 -3.60665053e-01 -2.34903619e-01 -3.04518789e-01 -7.94166923e-01 -3.36110622e-01 5.50926626e-01 4.24632311e-01 4.58951920e-01 -1.20649993e+00 -4.90211964e-01 3.32540065e-01 -8.64318982e-02 -1.17815332e-02 1.38958588e-01 7.82121181e-01 -5.47369480e-01 6.13649599e-02 -6.74521387e-01 -3.52533311e-01 -1.55741847e+00 9.81145144e-01 1.37592703e-01 -1.75821587e-01 -5.11887074e-01 5.06332695e-01 3.98309678e-01 7.90371522e-02 1.93207398e-01 -5.25564551e-01 -2.57638723e-01 -1.78038143e-02 6.72214806e-01 3.45988244e-01 1.20391294e-01 -4.89180654e-01 -1.86664417e-01 3.91466528e-01 -3.99405271e-01 5.64542338e-02 1.14100635e+00 -4.35780026e-02 -7.60850459e-02 2.89051980e-01 7.82672405e-01 4.17428613e-01 -1.26258790e+00 -1.95957080e-01 -1.40826970e-01 -6.70638502e-01 1.09316565e-01 -8.40424001e-01 -1.04934990e+00 6.57745361e-01 9.23753142e-01 3.84648114e-01 1.57909989e+00 -2.20201254e-01 5.69408834e-01 -3.89068164e-02 1.12491632e-02 -8.92427146e-01 6.94333240e-02 1.13800891e-01 1.03277564e+00 -1.20201695e+00 1.99508697e-01 -5.36490738e-01 -5.14585018e-01 7.84017026e-01 7.02903032e-01 -2.63580352e-01 1.92885607e-01 3.31514597e-01 4.88395952e-02 1.27287969e-01 -4.31312084e-01 -2.85845131e-01 4.75031137e-02 6.80854380e-01 2.75052458e-01 -3.04535747e-01 -5.11165857e-01 1.12017758e-01 4.08370756e-02 9.97373485e-04 7.78647900e-01 1.25596547e+00 -5.18746138e-01 -1.06299567e+00 -7.04462349e-01 1.80890054e-01 -5.74947417e-01 -1.36783317e-01 -4.92139369e-01 1.12088966e+00 3.38148177e-01 5.40308237e-01 1.80691406e-01 -1.21357165e-01 4.43521380e-01 8.32505990e-03 4.20217156e-01 -4.75134283e-01 -4.31031138e-01 1.03385732e-01 -2.71649897e-01 -1.92700431e-01 -7.48243809e-01 -7.30255485e-01 -1.09235299e+00 6.69739544e-02 -3.23253602e-01 -2.07144618e-01 7.71629930e-01 7.60581553e-01 3.96081060e-01 7.56461084e-01 4.13508803e-01 -1.12548029e+00 -1.58188626e-01 -1.11380124e+00 -3.70138407e-01 7.59241760e-01 5.70510030e-01 -7.84141123e-01 -3.05524528e-01 4.33535963e-01]
[10.833586692810059, -0.9801509380340576]
efeb7e50-0e82-434b-bf8c-5919d332bca0
function-words-enhanced-attention-networks
2204.12111
null
https://arxiv.org/abs/2204.12111v1
https://arxiv.org/pdf/2204.12111v1.pdf
Function-words Enhanced Attention Networks for Few-Shot Inverse Relation Classification
The relation classification is to identify semantic relations between two entities in a given text. While existing models perform well for classifying inverse relations with large datasets, their performance is significantly reduced for few-shot learning. In this paper, we propose a function words adaptively enhanced attention framework (FAEA) for few-shot inverse relation classification, in which a hybrid attention model is designed to attend class-related function words based on meta-learning. As the involvement of function words brings in significant intra-class redundancy, an adaptive message passing mechanism is introduced to capture and transfer inter-class differences.We mathematically analyze the negative impact of function words from dot-product measurement, which explains why message passing mechanism effectively reduces the impact. Our experimental results show that FAEA outperforms strong baselines, especially the inverse relation accuracy is improved by 14.33% under 1-shot setting in FewRel1.0.
['Kewen Wang', 'Zhiyong Feng', 'Xiaowang Zhang', 'Shaojuan Wu', 'Chunliu Dou']
2022-04-26
null
null
null
null
['relation-classification']
['natural-language-processing']
[ 1.36088133e-01 4.58449841e-01 -6.10464752e-01 -3.85286599e-01 -4.34599042e-01 1.07805748e-02 7.92092562e-01 5.11645734e-01 -3.43414277e-01 6.76786780e-01 3.25644612e-01 -5.04474863e-02 -4.79475290e-01 -1.17841649e+00 -5.34288943e-01 -3.41142118e-01 -1.45698518e-01 6.12756252e-01 2.94975251e-01 -8.64767432e-01 -8.54835007e-03 8.67413729e-02 -1.51895452e+00 3.47292125e-01 1.02911615e+00 9.78501558e-01 6.57289326e-02 6.86923027e-01 -4.72031921e-01 1.22734952e+00 -5.98282218e-01 -8.51848841e-01 -1.32562652e-01 -5.24380684e-01 -1.27451289e+00 -3.51590693e-01 -2.56465256e-01 -1.75036103e-01 -5.72878242e-01 9.55709517e-01 4.91309851e-01 6.96001112e-01 7.83953011e-01 -1.46933293e+00 -1.30442131e+00 8.82980287e-01 -4.23561424e-01 5.39050817e-01 3.99296999e-01 -2.38730699e-01 1.70700908e+00 -9.51872766e-01 5.26443839e-01 1.49330211e+00 4.77818042e-01 1.91349968e-01 -1.02773643e+00 -6.44336343e-01 4.43929099e-02 8.15729678e-01 -1.54230773e+00 -3.48707557e-01 4.68784600e-01 -7.78607428e-02 1.79744530e+00 3.16173077e-01 4.41645563e-01 7.40706205e-01 2.73931026e-01 6.38211310e-01 7.79686794e-02 -4.86738592e-01 3.22635961e-03 -7.39092454e-02 7.02633917e-01 4.78288621e-01 1.54497236e-01 -1.96032733e-01 -3.96670759e-01 -6.19715871e-03 3.30493599e-01 2.27383316e-01 -2.49679863e-01 -1.58082647e-03 -7.94905841e-01 9.51167464e-01 8.46168876e-01 5.03259718e-01 -2.48221457e-01 -1.60421561e-02 5.43127537e-01 4.43356246e-01 7.05505013e-01 4.98121381e-01 -6.59567356e-01 -9.51210111e-02 5.52408546e-02 -6.77225292e-02 7.80740023e-01 1.28263986e+00 8.51108611e-01 -5.54664791e-01 -6.05091453e-01 1.22305655e+00 1.97546408e-01 -1.19177543e-01 5.78174949e-01 -5.44516265e-01 6.19370580e-01 1.07014680e+00 -2.37162545e-01 -1.18182540e+00 -3.72380406e-01 -3.84660602e-01 -7.26219356e-01 -4.08309221e-01 -2.56340623e-01 2.07616046e-01 -7.28962183e-01 1.56100011e+00 3.94288987e-01 2.81271845e-01 3.31331223e-01 6.49613619e-01 1.34503162e+00 6.04335845e-01 5.96450925e-01 -3.17774117e-01 1.63451672e+00 -1.27342474e+00 -1.33300149e+00 -2.72477835e-01 1.17444646e+00 -4.20348614e-01 1.10967946e+00 -5.47278345e-01 -6.94505274e-01 -6.15594149e-01 -1.22438931e+00 -6.06988370e-01 -9.42902625e-01 -4.46910441e-01 9.41715479e-01 2.90778935e-01 -2.33672246e-01 7.85806000e-01 -3.69613528e-01 -5.88135302e-01 6.76782548e-01 3.17236960e-01 -3.27226043e-01 9.19640958e-02 -2.10045671e+00 1.11867166e+00 7.00240135e-01 -3.31829876e-01 -2.18321636e-01 -1.07170117e+00 -1.19860208e+00 6.96269572e-01 5.78318954e-01 -5.96407294e-01 1.14124954e+00 -4.18484956e-01 -1.14888573e+00 7.40451217e-01 -1.25851825e-01 -4.89174902e-01 7.59679591e-03 -3.03080410e-01 -7.14941919e-01 -9.32863653e-02 2.72646457e-01 4.00458604e-01 2.37337157e-01 -9.05703723e-01 -6.25234962e-01 -2.05522582e-01 3.42741847e-01 3.32367688e-01 -4.78284299e-01 -3.72780226e-02 -4.46746945e-01 -5.24302840e-01 -2.28177130e-01 -5.33067942e-01 9.69096944e-02 -3.06297898e-01 -2.35921100e-01 -7.60387361e-01 8.64572048e-01 -2.07975805e-01 1.62781048e+00 -1.95557892e+00 -7.49205276e-02 -3.86061192e-01 3.40734214e-01 4.45845783e-01 -1.75510749e-01 4.09088045e-01 -5.40746868e-01 1.47187933e-01 -1.65625140e-01 -8.58630687e-02 -1.34490281e-01 3.52164984e-01 -1.48920134e-01 1.54555812e-01 4.21524018e-01 1.52594745e+00 -1.21755731e+00 -5.54286182e-01 9.58248898e-02 4.13086861e-01 -1.73697308e-01 2.99937576e-01 5.98925576e-02 -3.87840092e-01 -3.65036428e-01 6.74459338e-01 5.18247247e-01 -5.28354049e-01 1.78774178e-01 -4.42648917e-01 4.70242172e-01 4.16304857e-01 -6.51997685e-01 1.63688695e+00 -4.64625388e-01 4.56651062e-01 -5.50597548e-01 -1.19293487e+00 9.24415052e-01 5.35482347e-01 4.38273013e-01 -8.01422954e-01 5.03525734e-01 -2.99632698e-01 4.44485247e-01 -6.18549466e-01 4.41335052e-01 -2.76359260e-01 3.26202735e-02 2.64562637e-01 5.21257341e-01 2.23469988e-01 2.13291734e-01 4.47592676e-01 1.40774727e+00 -9.82221365e-02 9.16984916e-01 -1.73193142e-01 6.11060619e-01 -1.94323674e-01 5.14416575e-01 6.55603826e-01 -4.12533909e-01 2.60659754e-01 5.22248507e-01 -4.20844525e-01 -6.08541131e-01 -8.27187121e-01 -9.60144922e-02 1.52974713e+00 5.51462471e-01 -7.28850126e-01 -1.98935091e-01 -8.75397384e-01 1.58529524e-02 8.84674251e-01 -1.00001717e+00 -8.80698144e-01 -1.53899714e-01 -9.76950407e-01 3.09230477e-01 6.28582418e-01 6.28954172e-01 -1.10569274e+00 -2.60249496e-01 2.74834007e-01 -2.09231675e-01 -1.03945673e+00 -4.47176486e-01 3.39898705e-01 -5.19303024e-01 -1.15049279e+00 -3.63305569e-01 -7.81746268e-01 1.59631327e-01 4.43884611e-01 1.38989151e+00 1.52312741e-01 -3.80787492e-01 1.13642924e-01 -7.31164694e-01 -3.54199290e-01 -3.25366110e-02 3.84657443e-01 -1.74749061e-01 2.52170004e-02 9.12064254e-01 -6.12519860e-01 -3.67780715e-01 1.20832838e-01 -5.13642430e-01 -1.33065119e-01 4.97460425e-01 1.21872878e+00 2.28126422e-01 1.48926312e-02 7.91689575e-01 -1.31953144e+00 8.57768655e-01 -9.19534743e-01 1.45060986e-01 6.54612422e-01 -9.11157846e-01 1.32715464e-01 5.00751555e-01 -5.25293708e-01 -1.28631604e+00 -3.97705406e-01 6.74193725e-02 -3.40658784e-01 1.28247842e-01 4.74736631e-01 -4.07171279e-01 1.87259540e-01 8.00910413e-01 -2.20876530e-01 -4.06069189e-01 -2.23764762e-01 6.27415240e-01 6.48729980e-01 2.49988422e-01 -2.79664814e-01 3.36492956e-01 2.22156510e-01 -1.01675689e-01 -5.13966978e-01 -1.38825357e+00 -6.70987785e-01 -5.40506005e-01 1.75799876e-01 9.91680920e-01 -7.84998059e-01 -9.50126231e-01 -1.22350737e-01 -1.34941733e+00 -6.39313385e-02 -3.40447307e-01 3.77696693e-01 -3.10212970e-01 1.20491898e-02 -7.77307391e-01 -7.56215215e-01 -6.18485510e-01 -6.97650194e-01 8.70757818e-01 2.49747768e-01 -4.57780302e-01 -1.14280617e+00 4.82054874e-02 3.16364914e-01 2.18276083e-01 -1.87210679e-01 1.09804523e+00 -9.61682796e-01 -1.27679691e-01 -1.83428928e-01 -5.45384407e-01 -2.16821074e-01 4.73134369e-01 -4.36511874e-01 -9.86611485e-01 3.06322724e-01 -2.63510197e-01 -3.11897635e-01 1.00252140e+00 1.13387980e-01 9.04752493e-01 -6.19115233e-02 -7.16110229e-01 3.86711717e-01 1.03385317e+00 3.27479035e-01 8.53500009e-01 1.70906410e-01 9.25893903e-01 7.09855378e-01 9.71018195e-01 3.42110157e-01 5.93405724e-01 6.81192338e-01 9.67601985e-02 5.53059988e-02 -3.11246812e-01 -1.58762187e-01 -2.67253488e-01 6.68127418e-01 -7.18431398e-02 -4.07523423e-01 -8.53008449e-01 5.65622091e-01 -2.18844843e+00 -1.05554628e+00 -2.64831454e-01 1.78178370e+00 9.48191583e-01 2.41524249e-01 -3.46799523e-01 6.22502640e-02 8.47327411e-01 1.35322109e-01 -4.24878180e-01 -3.17198902e-01 -1.84814371e-02 2.87883431e-01 2.33711883e-01 6.74871922e-01 -1.16909361e+00 1.20191348e+00 5.67528820e+00 1.09512973e+00 -3.86940807e-01 4.46434408e-01 5.83551109e-01 1.10394284e-01 -2.34518982e-02 -3.34626213e-02 -6.48115218e-01 2.28225246e-01 1.08718777e+00 -5.95835447e-01 -1.92108676e-02 8.45524013e-01 -3.77894431e-01 7.53536150e-02 -1.11177397e+00 9.48846102e-01 -1.05804903e-02 -1.22465336e+00 4.92685288e-03 -1.96175188e-01 5.09936929e-01 -2.04450950e-01 -3.55827332e-01 1.00177634e+00 2.79522836e-01 -1.02301872e+00 6.32410794e-02 3.95347267e-01 6.38173223e-01 -1.00530529e+00 9.00766730e-01 3.47288176e-02 -1.64227128e+00 -5.49030080e-02 -5.32348275e-01 -4.85924363e-01 1.48748085e-01 4.10412937e-01 -8.40046525e-01 6.67072296e-01 6.29220486e-01 9.20266807e-01 -4.01845813e-01 4.16664630e-01 -2.95939207e-01 2.12387726e-01 1.64574668e-01 -2.63283730e-01 -1.67827189e-01 8.95979479e-02 2.61220813e-01 1.12523592e+00 -8.62068310e-02 6.21197164e-01 2.28768699e-02 8.56115043e-01 -3.32931250e-01 1.34116411e-01 -5.67147493e-01 -1.66664943e-01 6.73592150e-01 1.48599255e+00 -6.88190818e-01 -6.75097167e-01 -6.95499063e-01 1.15867543e+00 8.71889174e-01 2.93923467e-01 -1.10441625e+00 -9.05302107e-01 9.93941605e-01 -3.15733224e-01 2.54380763e-01 2.99391806e-01 -2.76505589e-01 -1.02705538e+00 -3.79165977e-01 -1.16570681e-01 7.95929790e-01 -6.11005127e-01 -1.46779799e+00 4.62747782e-01 -4.63840812e-02 -9.81371045e-01 -2.13093627e-02 -2.95743972e-01 -6.56025469e-01 6.67394578e-01 -1.32418716e+00 -1.17058110e+00 -3.90054077e-01 3.04498225e-01 7.48276591e-01 -3.53963189e-02 1.21201336e+00 5.58326781e-01 -7.09363818e-01 7.82984734e-01 -4.57848191e-01 2.32116044e-01 6.35874271e-01 -1.20269227e+00 5.42377830e-01 3.61441463e-01 1.67351514e-01 7.35623837e-01 6.57131493e-01 -7.43765771e-01 -1.04916549e+00 -1.06662858e+00 1.23209465e+00 -4.43024367e-01 9.11011577e-01 -3.11727196e-01 -1.26456571e+00 7.70453155e-01 1.95272148e-01 5.39311051e-01 9.98110473e-01 5.60808301e-01 -6.15092754e-01 -4.12152037e-02 -1.01742923e+00 5.43117464e-01 1.40031123e+00 -7.09939539e-01 -9.22645032e-01 3.35359603e-01 1.33332014e+00 -3.12093869e-02 -1.27916038e+00 5.65425634e-01 4.51401681e-01 -4.95292664e-01 1.12661529e+00 -1.02273524e+00 6.90415621e-01 -5.66771701e-02 -1.04485117e-01 -1.37012160e+00 -6.82359338e-01 -1.99121565e-01 -8.98540556e-01 1.46954513e+00 4.93233711e-01 -5.76704383e-01 4.11329269e-01 7.67692029e-01 -9.50422511e-03 -8.53845894e-01 -6.07583463e-01 -8.34023595e-01 -1.48669630e-01 -2.63965458e-01 6.10290408e-01 1.34492588e+00 5.27928770e-01 1.45022511e+00 -2.90027082e-01 -1.01958044e-01 2.28145033e-01 5.42469025e-02 4.13802177e-01 -1.17987549e+00 -3.43607128e-01 -3.34319711e-01 -6.88022256e-01 -7.33622134e-01 5.40304124e-01 -9.95364487e-01 -3.53350490e-02 -1.43202138e+00 4.81583744e-01 -1.40635818e-01 -6.84066236e-01 4.90095764e-01 -9.25314903e-01 5.06992936e-02 -1.10783920e-01 -7.76666030e-02 -9.49030817e-01 1.09407961e+00 1.08062589e+00 -3.78778636e-01 -1.37132302e-01 -1.96294397e-01 -6.98983908e-01 3.46458554e-01 5.90505540e-01 -4.62741137e-01 -7.57319808e-01 -5.31519577e-02 5.29130995e-02 -1.08591534e-01 -5.34369610e-02 -6.59358621e-01 3.13501656e-01 7.72074461e-02 3.09000999e-01 -3.03607255e-01 4.64683801e-01 -7.19156563e-01 -1.66223317e-01 3.63339156e-01 -5.83325624e-01 -3.74968946e-01 -6.48439527e-02 8.22525084e-01 -2.18574822e-01 -2.65914232e-01 4.82855469e-01 -4.38420512e-02 -9.94533956e-01 3.91574353e-01 1.21915471e-02 2.21709162e-01 1.13450515e+00 1.74497560e-01 -4.49077487e-01 -3.27293783e-01 -5.76570988e-01 2.73080707e-01 -2.52626538e-01 8.70966077e-01 5.28924167e-01 -1.69676137e+00 -5.01219869e-01 -1.39096603e-01 7.72088110e-01 -3.40081602e-01 4.01210517e-01 7.37037122e-01 6.05296120e-02 3.61772656e-01 1.97542027e-01 -3.12896306e-03 -1.26470697e+00 9.04081583e-01 2.58429945e-01 -3.77899349e-01 -5.87235808e-01 1.27415705e+00 3.09086174e-01 -3.49571407e-01 8.38333927e-03 -1.07877754e-01 -4.53405768e-01 2.68911362e-01 8.30234170e-01 4.15372819e-01 1.36055410e-01 -5.99239707e-01 -4.13942307e-01 2.50211924e-01 -3.56064260e-01 2.24578053e-01 1.30063307e+00 -1.01596475e-01 -1.54780284e-01 6.16678953e-01 1.51187134e+00 -5.48867941e-01 -7.52723455e-01 -3.50652575e-01 3.12171578e-01 -4.85377342e-01 4.69514802e-02 -5.73854268e-01 -8.05705965e-01 6.32500231e-01 2.86116332e-01 4.08017993e-01 8.84221315e-01 4.63282615e-01 8.72609258e-01 4.32506472e-01 2.09791921e-02 -9.14937675e-01 8.75187442e-02 8.25766683e-01 7.40839839e-01 -1.55350482e+00 1.55327082e-01 -8.56100142e-01 -6.14398539e-01 7.82726109e-01 9.28843498e-01 2.49873642e-02 8.21496129e-01 1.69414610e-01 -4.64944214e-01 -5.60120046e-01 -1.10253537e+00 -7.13896692e-01 5.23572087e-01 5.56819260e-01 8.63480270e-01 -4.01937449e-03 -6.30155325e-01 1.02187991e+00 1.93982739e-02 -2.77414829e-01 -8.70945677e-02 8.53964210e-01 -3.39372307e-01 -9.30831075e-01 2.07323924e-01 6.37299001e-01 -2.72757202e-01 -3.85571033e-01 -4.26044136e-01 6.57999158e-01 1.74238637e-01 1.14206064e+00 2.43899956e-01 -6.34391248e-01 4.13037449e-01 2.20360860e-01 3.08025330e-01 -8.78060222e-01 -4.79260564e-01 -4.44694340e-01 3.51315439e-01 -5.57511926e-01 -2.25439712e-01 -1.56405926e-01 -1.61256540e+00 -2.49107808e-01 -6.71531856e-01 7.46251196e-02 -6.86150938e-02 1.19820893e+00 5.25438666e-01 1.14252841e+00 2.64055073e-01 -2.62331486e-01 -6.47464469e-02 -1.32425272e+00 -6.07856333e-01 7.47329891e-01 7.88617432e-02 -1.28000748e+00 -1.75502643e-01 -2.74989963e-01]
[9.254343032836914, 8.549166679382324]
5f033c6c-b7af-47df-96b1-ac89a26b7038
propagating-over-phrase-relations-for-one
null
null
https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/3304_ECCV_2020_paper.php
https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123640579.pdf
Propagating Over Phrase Relations for One-Stage Visual Grounding
Phrase level visual grounding aims to locate in an image the corresponding visual regions referred to by multiple noun phrases in a given sentence. Its challenge comes not only from large variations in visual contents and unrestricted phrase descriptions but also from unambiguous referrals derived from phrase relational reasoning. In this paper, we propose a linguistic structure guided propagation network for one-stage phrase grounding. It explicitly explores the linguistic structure of the sentence and performs relational propagation among noun phrases under the guidance of the linguistic relations between them. Specifically, we first construct a linguistic graph parsed from the sentence and then capture multimodal feature maps for all the phrasal nodes independently. The node features are then propagated over the edges with a tailor-designed relational propagation module and ultimately integrated for final prediction. Experiments on Flicker30K Entities dataset show that our model outperforms state-of-the-art methods and demonstrate the effectiveness of propagating among phrases with linguistic relations.
['Yizhou Yu', 'Sibei Yang', 'Guanbin Li']
null
null
null
null
eccv-2020-8
['phrase-grounding']
['natural-language-processing']
[ 1.73658449e-02 5.12117863e-01 -4.40529227e-01 -4.28369999e-01 -7.35930085e-01 -6.40429974e-01 4.61314380e-01 4.92785633e-01 -1.54834956e-01 4.15055096e-01 4.88126606e-01 -1.43867552e-01 8.20079166e-03 -7.68547535e-01 -7.16580033e-01 -4.84273165e-01 -4.28194217e-02 4.30332065e-01 3.21308881e-01 -1.41761467e-01 5.36758192e-02 3.74243796e-01 -1.12141693e+00 8.03054631e-01 4.68688667e-01 8.34069669e-01 1.74091846e-01 5.26857734e-01 -4.23225909e-01 1.01894045e+00 -2.57572263e-01 -9.21376407e-01 -2.82820374e-01 -2.80056804e-01 -8.48985076e-01 4.66675758e-01 4.48380083e-01 -7.17910454e-02 -3.23254526e-01 9.83471930e-01 2.25947589e-01 -6.11492014e-03 5.11875987e-01 -1.16523170e+00 -9.92426395e-01 1.05290663e+00 -7.16089725e-01 2.87086098e-03 6.92869544e-01 -1.29182667e-01 1.82302904e+00 -1.22124159e+00 8.22516143e-01 1.37184370e+00 4.64519471e-01 3.73259068e-01 -1.35790181e+00 -3.27987850e-01 6.20737970e-01 1.27221420e-01 -1.55448925e+00 -9.54864696e-02 8.16593468e-01 -4.59865779e-01 1.16604841e+00 8.69371966e-02 6.13115668e-01 1.03889418e+00 6.22667968e-02 7.59160936e-01 7.17936873e-01 -4.15200382e-01 -2.99854040e-01 4.08285946e-01 3.38521034e-01 1.44426346e+00 -1.68324023e-01 -3.50624263e-01 -9.77626026e-01 2.18589567e-02 3.67557317e-01 -3.47662657e-01 -2.62713909e-01 -1.69071972e-01 -1.25978899e+00 7.40534544e-01 9.59918618e-01 8.50810017e-03 -4.77222979e-01 1.89881846e-01 1.49525464e-01 -1.79363295e-01 3.33988458e-01 2.12397203e-01 -1.82505846e-02 5.19832790e-01 -8.15121651e-01 1.34943396e-01 6.20818615e-01 1.09874892e+00 1.07574213e+00 -7.07461953e-01 -6.73282683e-01 7.29110658e-01 6.81038380e-01 3.96352947e-01 -1.43196091e-01 -9.14656937e-01 7.82700896e-01 9.96653914e-01 -9.59831700e-02 -1.61587727e+00 -4.99277502e-01 -3.71889412e-01 -4.68513489e-01 -3.22866648e-01 8.80127698e-02 -5.38259894e-02 -7.43557692e-01 1.63425565e+00 3.39139223e-01 -1.45785168e-01 -8.00592452e-02 8.57500970e-01 1.49915326e+00 6.64852262e-01 3.78856838e-01 2.68747173e-02 1.65648293e+00 -1.04273415e+00 -5.19763768e-01 -3.29122305e-01 3.46529484e-01 -4.27914858e-01 1.01156521e+00 -1.52603522e-01 -8.58444035e-01 -2.93365568e-01 -6.27624989e-01 -4.19644833e-01 -3.69976014e-01 1.50426552e-01 3.86225462e-01 2.85674352e-02 -1.31996703e+00 -1.36146238e-02 -5.24543047e-01 -5.18642783e-01 4.57633674e-01 4.03196692e-01 -4.01119351e-01 7.37814307e-02 -1.15476739e+00 8.02623808e-01 5.88784099e-01 3.51124763e-01 -4.49753642e-01 -3.85988593e-01 -1.07238281e+00 1.51842505e-01 4.18621272e-01 -1.06105292e+00 8.13709199e-01 -5.72273493e-01 -9.56503749e-01 1.33429730e+00 -4.82643366e-01 -3.02380174e-01 -5.10436743e-02 2.12018698e-01 -2.85825193e-01 6.46243215e-01 4.11492258e-01 1.30039513e+00 7.32758880e-01 -1.51610684e+00 -9.12203491e-01 -1.28269896e-01 6.09313130e-01 4.99649793e-01 -2.36242026e-01 1.93065360e-01 -1.12236559e+00 -2.31628373e-01 2.12530077e-01 -7.95755386e-01 3.00341174e-02 -3.12296506e-02 -1.06887543e+00 -4.33438867e-01 3.85740221e-01 -6.43445432e-01 1.35253692e+00 -2.22015572e+00 5.08560777e-01 5.00872850e-01 6.22512460e-01 -6.01755142e-01 -6.46597520e-02 4.10797805e-01 1.57648399e-01 2.22188190e-01 1.88489184e-02 -5.29227853e-01 2.01461941e-01 3.81790131e-01 -4.51199025e-01 1.38600513e-01 4.57179725e-01 1.21782839e+00 -9.27425742e-01 -1.09708488e+00 -8.91873986e-03 5.30220449e-01 -3.80272865e-01 -1.48818996e-02 -3.34004968e-01 1.86839357e-01 -5.23793399e-01 7.84058273e-01 1.22174367e-01 -6.51526392e-01 -3.28855254e-02 -7.42428541e-01 7.30464607e-02 1.46799803e-01 -5.74795187e-01 1.66595554e+00 -1.47134393e-01 5.41885555e-01 -1.30468860e-01 -4.96569812e-01 8.55729461e-01 1.46427810e-01 1.98141739e-01 -5.76831281e-01 -1.36404447e-02 -1.78606048e-01 -2.27524713e-01 -7.12805331e-01 4.82044131e-01 -6.23456985e-02 -3.83776188e-01 3.09505202e-02 6.98620453e-02 1.22771077e-01 3.38123351e-01 6.95822001e-01 8.10735941e-01 1.73724517e-01 1.17948942e-01 1.80769444e-01 6.70928001e-01 2.42562711e-01 3.54954869e-01 6.97024465e-01 -6.95030168e-02 5.38738310e-01 8.03227901e-01 -1.59079239e-01 -6.83747888e-01 -1.20482767e+00 -4.38904390e-02 1.29483640e+00 3.82446587e-01 -8.64983320e-01 -5.29660046e-01 -8.49816740e-01 -7.27573335e-02 6.80610299e-01 -8.47294390e-01 1.97365910e-01 -3.07180405e-01 -3.66895318e-01 4.56265181e-01 6.11459851e-01 2.63455719e-01 -1.24634302e+00 -2.89048076e-01 -1.01165332e-01 -3.87069792e-01 -1.56821597e+00 -2.92950958e-01 -8.90343860e-02 -2.79798925e-01 -8.10138643e-01 -4.12615299e-01 -1.04433107e+00 1.01226747e+00 -1.29761800e-01 1.24703944e+00 1.16609573e-01 1.05439601e-02 5.02108037e-01 -2.92444855e-01 -5.66142388e-02 -2.16232151e-01 8.08561370e-02 -3.16908449e-01 2.76096135e-01 4.36856836e-01 -1.42990708e-01 -4.64079410e-01 3.37086767e-02 -3.38223010e-01 4.03472275e-01 5.39838374e-01 6.39361441e-01 1.04753900e+00 1.59535259e-01 -6.46089688e-02 -8.66052032e-01 6.08741224e-01 -3.84590924e-01 -4.55135107e-01 7.74072409e-01 -1.77136436e-01 5.58157377e-02 1.33081347e-01 -2.13322192e-02 -1.00478983e+00 1.76352456e-01 1.47363558e-01 -3.45864177e-01 -2.26961970e-01 7.09070683e-01 5.11090867e-02 1.61809430e-01 4.09862965e-01 1.14621662e-01 -3.97203743e-01 4.56974208e-02 8.25795472e-01 4.61216599e-01 7.25105464e-01 -5.85287333e-01 8.27451587e-01 5.95380902e-01 1.08310692e-01 -7.36561656e-01 -1.24470723e+00 -3.20897728e-01 -7.07674742e-01 -4.73644257e-01 1.34436440e+00 -8.46909404e-01 -8.11947823e-01 -2.20217451e-01 -1.37093186e+00 6.16605617e-02 -6.99079782e-02 2.24281311e-01 -3.23786736e-01 7.68402293e-02 -7.60112464e-01 -5.74261725e-01 -2.83442855e-01 -1.03632116e+00 1.20929098e+00 2.77696580e-01 -2.37408876e-01 -1.05041933e+00 -1.71264172e-01 4.76596713e-01 -3.00959378e-01 2.17314854e-01 1.20299387e+00 -3.94242227e-01 -7.54463553e-01 -7.71371722e-02 -6.93822742e-01 -1.30612209e-01 -3.29399630e-02 3.06137979e-01 -8.31757009e-01 1.67144299e-01 -5.38488925e-01 -2.44453311e-01 8.52019906e-01 3.08346331e-01 6.53314471e-01 -3.34372312e-01 -6.04990780e-01 5.21194398e-01 1.39443326e+00 -2.12985381e-01 1.12831980e-01 3.00503790e-01 1.16151619e+00 1.09687102e+00 4.01754886e-01 1.44234067e-02 7.05005944e-01 4.92264420e-01 5.61438203e-01 -2.45034501e-01 -1.10201076e-01 -5.13481379e-01 1.44225568e-01 5.37439108e-01 1.74725771e-01 -3.72270405e-01 -1.13401783e+00 6.60627902e-01 -1.91811728e+00 -8.07568610e-01 -2.60271370e-01 1.42623830e+00 7.80498683e-01 1.30707666e-01 7.08027631e-02 -4.13235664e-01 8.95395577e-01 8.38184580e-02 -1.83743894e-01 -3.70669663e-01 -2.80582637e-01 -2.29223594e-01 1.50247663e-01 6.79065406e-01 -1.03926229e+00 1.33803582e+00 5.56544924e+00 5.46572387e-01 -7.56865799e-01 -2.57713571e-02 5.70763230e-01 1.30961034e-02 -6.06008768e-01 -1.78800561e-02 -1.09884250e+00 8.06434825e-02 3.70002717e-01 9.65504721e-03 2.11828768e-01 3.47280443e-01 8.28202888e-02 -3.33705619e-02 -1.25174236e+00 7.50049889e-01 2.75956482e-01 -1.43426335e+00 3.54437143e-01 -9.86522362e-02 4.50423181e-01 4.44101505e-02 1.13360301e-01 6.78498074e-02 1.02208704e-01 -1.06255805e+00 8.84221554e-01 6.16140246e-01 6.03906751e-01 -7.10284710e-01 6.05525136e-01 1.79325804e-01 -1.37158155e+00 2.34633628e-02 -1.92915246e-01 3.28463197e-01 2.85086095e-01 1.55598119e-01 -9.39498723e-01 4.90184665e-01 6.95053041e-01 7.56681025e-01 -7.78404057e-01 7.21834242e-01 -8.97999406e-01 3.96463633e-01 -1.97684109e-01 -1.83318391e-01 5.44914782e-01 9.09280479e-02 6.35742724e-01 1.54511893e+00 -6.10664375e-02 1.48891941e-01 2.57951826e-01 1.23330307e+00 -2.19667494e-01 3.78883719e-01 -6.80940390e-01 -7.58369416e-02 3.31137270e-01 1.50388098e+00 -8.72738659e-01 3.43982838e-02 -8.78595889e-01 9.50610518e-01 8.16052675e-01 7.24348903e-01 -7.70460963e-01 -2.85118133e-01 1.42097056e-01 4.43615131e-02 4.15976942e-01 -5.70782349e-02 -2.17627808e-01 -9.20296311e-01 1.42663747e-01 -2.12769911e-01 5.29111624e-01 -1.31234646e+00 -1.44543195e+00 8.17959368e-01 2.42315065e-02 -8.44479322e-01 -2.32058093e-02 -5.95361948e-01 -3.83086622e-01 1.04938209e+00 -1.54934347e+00 -1.63047719e+00 -2.19041646e-01 6.54026449e-01 2.24845991e-01 1.62219517e-02 8.93120646e-01 -1.36196300e-01 -7.32417762e-01 4.53232259e-01 -7.38709509e-01 5.72924376e-01 3.70284140e-01 -1.27742124e+00 6.75632432e-02 9.08508480e-01 6.90326750e-01 7.94905245e-01 5.66733241e-01 -6.82903290e-01 -8.96115363e-01 -1.16459060e+00 1.52179480e+00 -4.75616038e-01 1.06118596e+00 -5.13712823e-01 -8.57078612e-01 8.28961134e-01 5.93691409e-01 -8.23233500e-02 7.55307615e-01 3.24649274e-01 -5.71062028e-01 -3.79547663e-02 -7.11056054e-01 7.26014435e-01 8.82566631e-01 -8.18366647e-01 -8.63767564e-01 4.94603634e-01 1.01331961e+00 -5.25926769e-01 -7.25050211e-01 4.56711501e-02 3.18694770e-01 -6.14814341e-01 1.05251801e+00 -6.65539145e-01 7.68118262e-01 -3.74266177e-01 -2.87443668e-01 -9.30276334e-01 -5.95733643e-01 -2.77544439e-01 -7.93529153e-02 1.54323208e+00 1.13857722e+00 -6.02867417e-02 7.07416058e-01 7.56218195e-01 8.22023973e-02 -8.44224572e-01 -5.93143821e-01 -1.64508045e-01 -4.03221428e-01 -4.99293417e-01 1.76124662e-01 6.37981951e-01 3.52133214e-01 8.77602041e-01 -9.35018212e-02 7.01496720e-01 6.63731217e-01 3.59984994e-01 1.89310119e-01 -8.21085632e-01 -3.79209936e-01 -3.36652637e-01 -3.94569933e-01 -9.62602675e-01 6.01606488e-01 -1.11728895e+00 1.57577336e-01 -2.03259254e+00 5.13105512e-01 -2.56187081e-01 -3.10914487e-01 8.77592981e-01 -2.67695516e-01 4.49242204e-01 2.29458794e-01 2.51896858e-01 -9.48060989e-01 1.80963904e-01 1.20023322e+00 -3.98157865e-01 -8.08638781e-02 -3.08327824e-01 -8.90153885e-01 8.42407644e-01 4.56951648e-01 -4.26941335e-01 -4.58010942e-01 -7.43690133e-01 8.71657014e-01 7.80986026e-02 6.70531809e-01 -4.42170531e-01 5.41936994e-01 -1.05999187e-01 2.15579405e-01 -8.87929797e-01 2.48701483e-01 -9.21627402e-01 -2.82290637e-01 -1.23150744e-01 -6.07740641e-01 8.82397592e-02 -1.16168903e-02 7.45360672e-01 -2.87055165e-01 -1.15776286e-01 2.98995942e-01 -1.22664973e-01 -8.55161786e-01 1.37677804e-01 -3.53709608e-02 4.81902901e-03 9.29826081e-01 -8.40137228e-02 -4.92572099e-01 -3.07134718e-01 -1.30494571e+00 5.97674370e-01 2.26166844e-01 2.99252242e-01 8.75623226e-01 -1.28476465e+00 -7.40002871e-01 -1.46883607e-01 4.97089267e-01 3.68850082e-02 8.44799578e-02 9.94738281e-01 -3.99562657e-01 4.78779018e-01 1.41583234e-01 -8.68025601e-01 -1.40070796e+00 5.70805073e-01 3.39882255e-01 -5.00061316e-03 -6.68939829e-01 1.33893204e+00 3.63274276e-01 -1.23229772e-01 3.86516660e-01 -3.84990871e-01 -5.99092066e-01 1.99161097e-01 2.26456344e-01 -2.09544316e-01 -3.10505897e-01 -1.23530376e+00 -6.54980063e-01 8.71940672e-01 -1.02820672e-01 -2.39357278e-01 1.02761090e+00 -4.29483891e-01 -4.85781431e-01 5.08694828e-01 1.17571652e+00 1.26245424e-01 -1.03525519e+00 -5.00156105e-01 -5.77713037e-03 -4.53484766e-02 1.07271262e-02 -8.43326092e-01 -1.09016621e+00 7.58027256e-01 -1.49137620e-03 1.89852506e-01 1.09254205e+00 7.98690319e-01 4.73320693e-01 3.84156942e-01 1.34538800e-01 -7.08246052e-01 2.73385607e-02 3.37656856e-01 8.31078053e-01 -1.07913506e+00 -7.46114179e-02 -8.30278873e-01 -9.20880556e-01 9.27184582e-01 5.86934447e-01 1.07891761e-01 4.00736928e-01 1.82413086e-01 6.13826737e-02 -6.32430017e-01 -7.31319606e-01 -4.16885287e-01 7.75395572e-01 5.24692416e-01 3.81776720e-01 2.89673544e-02 -9.96918306e-02 5.05635560e-01 -1.06491826e-01 -4.37230915e-01 1.05532393e-01 4.47426498e-01 -4.15129453e-01 -8.92881513e-01 -8.29378068e-02 1.60663053e-01 -3.08287382e-01 -5.00060320e-01 -7.84088790e-01 6.49340868e-01 4.10115302e-01 1.02368855e+00 1.99480444e-01 -4.13560033e-01 2.69390970e-01 9.08438209e-03 3.54660273e-01 -9.96191204e-01 -7.40468979e-01 1.96083739e-01 3.17192495e-01 -5.45198083e-01 -6.33611560e-01 -6.14004016e-01 -1.78436947e+00 3.99954990e-02 -7.99177364e-02 1.84493624e-02 3.90830964e-01 1.02917111e+00 3.27920318e-01 5.97407281e-01 2.78713256e-01 -4.08555686e-01 2.23327056e-01 -4.27668095e-01 -3.87603194e-01 5.05025744e-01 3.31103474e-01 -4.96779412e-01 -1.73035324e-01 1.91769391e-01]
[10.493812561035156, 1.4765198230743408]
fb5ea972-9b3a-4fd1-883b-e5cd3d00816f
chat-crowd-a-dialog-based-platform-for-visual
1812.04081
null
http://arxiv.org/abs/1812.04081v3
http://arxiv.org/pdf/1812.04081v3.pdf
Chat-crowd: A Dialog-based Platform for Visual Layout Composition
In this paper we introduce Chat-crowd, an interactive environment for visual layout composition via conversational interactions. Chat-crowd supports multiple agents with two conversational roles: agents who play the role of a designer are in charge of placing objects in an editable canvas according to instructions or commands issued by agents with a director role. The system can be integrated with crowdsourcing platforms for both synchronous and asynchronous data collection and is equipped with comprehensive quality controls on the performance of both types of agents. We expect that this system will be useful to build multimodal goal-oriented dialog tasks that require spatial and geometric reasoning.
['Paola Cascante-Bonilla', 'Vicente Ordonez', 'Song Feng', 'Xuwang Yin']
2018-12-10
chat-crowd-a-dialog-based-platform-for-visual-1
https://aclanthology.org/N19-4024
https://aclanthology.org/N19-4024.pdf
naacl-2019-6
['goal-oriented-dialog']
['natural-language-processing']
[-2.88181961e-01 1.55091837e-01 5.95615923e-01 -4.00333285e-01 -2.42808491e-01 -1.17812085e+00 9.82616246e-01 3.51331204e-01 -3.81357461e-01 5.50347626e-01 2.60228723e-01 -2.88083255e-01 1.38437571e-02 -7.88110673e-01 5.76858781e-03 -3.66359025e-01 1.03436261e-01 9.78979826e-01 6.77725852e-01 -8.41515303e-01 3.46929669e-01 7.49653459e-01 -1.71599209e+00 5.98344445e-01 3.73551100e-01 2.58531243e-01 6.33042216e-01 1.15586483e+00 -5.02699673e-01 1.35538208e+00 -9.41729724e-01 -2.49591306e-01 1.76837116e-01 -2.32705981e-01 -8.69923472e-01 4.55345660e-02 1.35374703e-02 -4.36667532e-01 -1.09505616e-02 5.73850036e-01 5.47097266e-01 4.48498040e-01 5.05554557e-01 -1.60547924e+00 -2.62775302e-01 4.34299588e-01 6.18220344e-02 -3.17258567e-01 1.27119100e+00 5.51126897e-01 8.25076938e-01 -6.88643038e-01 9.59946156e-01 1.59731638e+00 2.30989933e-01 6.01118445e-01 -7.38261461e-01 -6.02613501e-02 2.29562655e-01 4.56438102e-02 -1.22499132e+00 -5.13293743e-01 4.62828130e-01 -7.09168851e-01 8.58747184e-01 6.93563640e-01 4.47056890e-01 6.91578567e-01 -2.49042377e-01 5.47676742e-01 7.42278695e-01 -6.15525126e-01 4.53518331e-01 3.76216739e-01 -1.44783417e-02 8.12746346e-01 -4.68611360e-01 -3.78275722e-01 -7.90656805e-01 -4.41274524e-01 8.80079329e-01 -2.54262924e-01 -8.03647786e-02 -3.36430907e-01 -1.42273152e+00 6.31948948e-01 7.91916400e-02 4.69116092e-01 -2.63779521e-01 1.05646789e-01 3.77641112e-01 5.71391821e-01 1.58662289e-01 8.43771040e-01 -4.13457211e-03 -4.35909152e-01 2.94091608e-02 7.85777628e-01 1.23846173e+00 1.36346233e+00 4.98386353e-01 -3.70281219e-01 -3.86183470e-01 7.88150668e-01 4.13948029e-01 3.59694391e-01 -1.97278887e-01 -1.49074697e+00 4.23600823e-01 1.04654253e+00 8.58215213e-01 -8.36856127e-01 -5.98586202e-01 6.73843920e-01 -1.67408735e-01 1.10511804e+00 7.90341437e-01 -2.97037750e-01 -1.44186392e-01 9.13851857e-01 6.92856491e-01 -7.87306964e-01 -4.37661670e-02 9.99878228e-01 1.25955641e+00 3.95571321e-01 1.22675911e-01 2.52266079e-01 1.58795714e+00 -9.36356306e-01 -1.00939775e+00 3.19029152e-01 7.67659068e-01 -9.76840556e-01 1.29597449e+00 3.55815470e-01 -1.38365197e+00 -3.77090991e-01 -5.88704407e-01 -3.09781969e-01 -6.18200243e-01 9.43337977e-02 4.72984701e-01 5.25209725e-01 -1.24469316e+00 1.99355423e-01 -5.06004632e-01 -4.89915729e-01 -2.04931572e-01 2.14054853e-01 -2.87964880e-01 3.34810019e-01 -6.83013558e-01 1.13663614e+00 -2.23233223e-01 4.89726849e-02 -6.71718836e-01 -2.42791653e-01 -8.48990977e-01 -1.97884142e-01 4.07885402e-01 -7.06412435e-01 1.70134115e+00 -6.84636295e-01 -1.83962381e+00 7.74790883e-01 2.48995014e-02 1.34335831e-01 1.06785512e+00 -1.74069360e-01 5.09000905e-02 1.43090934e-01 1.78578779e-01 5.79004586e-01 4.34213690e-02 -1.51027429e+00 -7.40625918e-01 -3.01920980e-01 6.88610792e-01 6.06971800e-01 3.64495218e-01 6.80362821e-01 -6.10466003e-01 -2.03670621e-01 -4.26775038e-01 -9.13611829e-01 -1.31734505e-01 4.89508837e-01 -4.22910750e-01 -4.49289590e-01 1.05333555e+00 -4.09666359e-01 7.51657903e-01 -1.85033667e+00 1.31332994e-01 1.16237499e-01 2.35539794e-01 1.87072575e-01 5.55345695e-03 1.13636827e+00 6.35758936e-01 -1.31368756e-01 4.28624868e-01 -6.85040236e-01 6.91729009e-01 -1.17114522e-01 3.25885750e-02 1.22793041e-01 -3.85011733e-01 6.56681657e-01 -9.05451596e-01 -4.78011757e-01 6.48237586e-01 2.28596553e-01 -6.64215088e-02 7.24437237e-01 -6.71153545e-01 6.81786716e-01 -4.29961532e-01 3.44731480e-01 4.42320049e-01 -7.21890554e-02 5.54053009e-01 4.65860069e-01 -7.61696577e-01 2.83734709e-01 -1.32391524e+00 1.68880558e+00 -5.28855741e-01 8.49716783e-01 8.89462829e-01 -8.46284553e-02 9.67723489e-01 4.13924307e-01 7.41330087e-02 -2.85950571e-01 6.74191788e-02 -1.62200123e-01 -2.30358288e-01 -8.34801495e-01 8.71662676e-01 4.67005223e-01 -4.18268032e-02 1.04176819e+00 -5.45234859e-01 -4.94736165e-01 3.53217095e-01 3.98771554e-01 1.14838457e+00 2.42501557e-01 6.98489323e-02 -5.87342083e-02 3.69115442e-01 3.29235405e-01 -1.64962828e-01 8.58665049e-01 1.15052350e-01 2.43525639e-01 4.29044455e-01 -6.56409383e-01 -1.09784949e+00 -9.70746994e-01 6.67209029e-01 1.80173659e+00 2.59067774e-01 -3.85946602e-01 -5.57908595e-01 -2.66191512e-01 -2.97833830e-01 6.50585771e-01 -3.89846623e-01 7.67296255e-01 -5.46508849e-01 3.12180847e-01 4.81945783e-01 3.02349240e-01 2.65688986e-01 -1.39813709e+00 -1.05563390e+00 5.64140007e-02 -6.66920319e-02 -9.40852106e-01 -4.11171407e-01 -4.79497850e-01 -5.34463301e-02 -1.22434032e+00 -6.46822035e-01 -9.31462586e-01 7.41394877e-01 4.20006186e-01 1.04552233e+00 7.58104742e-01 -1.44056782e-01 9.24350023e-01 -8.35507452e-01 -6.07108951e-01 -7.13277042e-01 7.47195035e-02 -3.57904375e-01 -2.08115771e-01 2.33637206e-02 -4.15132433e-01 -4.93819207e-01 8.05611312e-01 -7.30117202e-01 5.81859231e-01 -3.81767869e-01 -2.98200157e-02 -4.25044686e-01 -4.48829859e-01 8.50801766e-02 -5.08343816e-01 1.14463139e+00 -6.53455853e-02 -7.63428748e-01 3.78402740e-01 3.22526693e-01 -2.37228051e-01 3.63857567e-01 -1.20368659e-01 -1.26820982e+00 2.69236773e-01 1.59118086e-01 3.90285909e-01 -5.83350122e-01 8.30456521e-03 -4.42570359e-01 -6.54471442e-02 5.06763101e-01 -3.24516356e-01 3.58453020e-02 -3.33459109e-01 4.46984023e-01 8.41482878e-01 3.49903703e-01 -7.21716464e-01 5.83157301e-01 5.03674448e-01 -3.02516073e-01 -9.28304255e-01 2.87671864e-01 -4.10976768e-01 -7.55822062e-01 -8.30616593e-01 1.05871534e+00 -6.46229088e-01 -1.62623084e+00 6.40319049e-01 -1.61755693e+00 -9.03665960e-01 1.59285422e-02 4.00182903e-02 -4.39679593e-01 1.86461568e-01 -5.52124381e-01 -1.13062263e+00 2.40278989e-02 -1.14020133e+00 1.12507224e+00 3.41920495e-01 -7.04111993e-01 -1.08175015e+00 1.03273652e-01 3.44929665e-01 4.94166523e-01 2.89886057e-01 3.73821497e-01 -6.22646689e-01 -7.26681888e-01 -1.38085648e-01 -1.56845301e-02 -5.17479777e-01 2.48136129e-02 3.89493197e-01 -7.33835697e-01 2.08133474e-01 -7.85596013e-01 -3.90157461e-01 -1.72764882e-01 -1.79065615e-01 4.82330054e-01 -2.50884324e-01 -3.76584619e-01 -2.98610598e-01 8.29129875e-01 7.04940617e-01 5.90957522e-01 3.66931677e-01 5.05480528e-01 1.09505022e+00 6.26072288e-01 6.31536245e-01 7.94599891e-01 1.30755532e+00 3.46957326e-01 -2.12533742e-01 3.21536899e-01 -5.13871349e-02 4.91286218e-02 6.68094084e-02 -3.20678651e-01 -5.71135402e-01 -1.10152411e+00 2.92366326e-01 -2.25771189e+00 -1.08674872e+00 -6.85274839e-01 1.76689994e+00 4.57971305e-01 -4.26674217e-01 4.65625167e-01 -5.63845746e-02 8.07721734e-01 8.58992785e-02 1.37266442e-01 -6.64696574e-01 2.80657351e-01 -3.67099792e-01 -5.16930129e-03 9.82562304e-01 -6.90067351e-01 8.48819852e-01 6.21665144e+00 2.32198507e-01 -4.79317576e-01 1.72769040e-01 8.06033388e-02 -4.11373936e-03 -2.85075158e-01 -9.07228291e-02 -3.98164034e-01 3.39114100e-01 2.59175390e-01 1.24503843e-01 7.50851274e-01 5.52086592e-01 7.96076119e-01 -7.34797418e-01 -1.23992598e+00 6.68732762e-01 -1.83401451e-01 -1.60836887e+00 -4.29416686e-01 1.47253852e-02 2.75385827e-01 -4.15155709e-01 -4.12137419e-01 1.41569581e-02 8.66696894e-01 -9.38274384e-01 8.65530074e-01 7.71145284e-01 4.89467472e-01 -4.96152997e-01 3.63365054e-01 6.47826016e-01 -1.26386797e+00 3.21728051e-01 1.60304889e-01 -6.75041080e-01 6.55911505e-01 -2.81955838e-01 -1.14984751e+00 2.35543638e-01 5.04231751e-01 -3.11902255e-01 -2.88355559e-01 1.11912489e+00 -1.76196799e-01 -2.73636222e-01 -1.06618479e-01 -8.13301980e-01 -1.42672792e-01 -3.45594585e-01 5.87142408e-01 1.22555292e+00 -6.26094220e-03 3.54728460e-01 4.86994594e-01 5.36407530e-01 4.40177977e-01 2.06615478e-01 -6.96961284e-01 3.26847434e-01 7.26351976e-01 1.17878032e+00 -8.91367018e-01 -3.65003705e-01 -5.33660240e-02 1.07603323e+00 2.06979677e-01 3.47088486e-01 -5.67861319e-01 -6.60340607e-01 6.38597190e-01 2.58116156e-01 -9.95276645e-02 -6.83857739e-01 -1.66877538e-01 -3.57839525e-01 -3.68936993e-02 -8.32183719e-01 -4.39472273e-02 -1.45177531e+00 -8.69838834e-01 6.83787704e-01 6.54650433e-03 -9.63753462e-01 -3.55419815e-01 -6.71583056e-01 -1.02759266e+00 9.02513385e-01 -4.99858856e-01 -1.26751435e+00 -8.03701341e-01 5.15847862e-01 6.29517019e-01 -3.47084910e-01 1.06865382e+00 -3.63174789e-02 -7.70282000e-02 -1.37619272e-01 -5.04932761e-01 2.77208894e-01 5.35432518e-01 -1.52206087e+00 4.27135438e-01 3.53858657e-02 -9.59609896e-02 5.26258051e-01 8.99833441e-01 -5.40479362e-01 -1.38903487e+00 -3.95923972e-01 8.78820479e-01 -7.22066402e-01 4.53915268e-01 -6.42646074e-01 -4.74891782e-01 5.42488873e-01 8.61809373e-01 -6.00709260e-01 7.22667813e-01 -9.87485498e-02 1.03678569e-01 2.48065054e-01 -1.10094178e+00 8.85945439e-01 1.03504825e+00 -5.81901789e-01 -5.64788640e-01 6.99726343e-01 4.64800537e-01 -8.89908075e-01 -3.04115891e-01 -4.51222748e-01 5.10770679e-01 -1.12855732e+00 5.81620276e-01 -5.03196657e-01 4.18927819e-02 -7.58195698e-01 2.40585551e-01 -1.32093608e+00 1.47082284e-02 -1.18186855e+00 8.12842131e-01 1.27641308e+00 4.18903321e-01 -5.78205347e-01 4.30959016e-01 1.24124706e+00 -4.22506072e-02 2.89693087e-01 -6.04838908e-01 -2.70821720e-01 -4.96097714e-01 -4.64668900e-01 7.72563159e-01 6.50470078e-01 6.72408581e-01 2.87522614e-01 -1.86769560e-01 2.83903986e-01 -1.69460177e-01 -3.21607202e-01 1.60362077e+00 -1.20086443e+00 -1.22362144e-01 -5.62665164e-01 4.75145765e-02 -9.49070096e-01 -9.46533158e-02 -2.95404643e-01 2.08918154e-01 -2.01803112e+00 -4.69761699e-01 -4.29107755e-01 1.04150069e+00 2.41491333e-01 3.05305868e-01 -2.68776506e-01 6.91446364e-01 1.66746661e-01 -1.03866208e+00 2.79263079e-01 1.19697821e+00 9.81524214e-02 -6.36658072e-01 1.53279230e-01 -1.60666287e-01 8.02210927e-01 7.86328077e-01 -7.59366602e-02 -3.86430085e-01 -4.21711475e-01 5.65174401e-01 2.70994425e-01 5.77255785e-01 -8.86043727e-01 7.51717210e-01 -3.97409081e-01 -8.23105052e-02 -3.37570101e-01 5.58065951e-01 -1.07192707e+00 4.44095612e-01 1.02604248e-01 -6.57629192e-01 6.34000719e-01 -1.82837490e-02 2.04266295e-01 1.72993958e-01 -3.96381199e-01 4.75557856e-02 -4.46422637e-01 -4.78994966e-01 -4.91084784e-01 -1.19850147e+00 -3.26525956e-01 1.38685882e+00 -1.95084825e-01 -8.80204260e-01 -1.00396562e+00 -8.40318918e-01 5.22318244e-01 8.27895463e-01 4.39010084e-01 4.78559703e-01 -1.17929125e+00 -5.47001004e-01 -2.21024632e-01 1.02930531e-01 -9.11087394e-02 -2.43911315e-02 3.24156225e-01 -1.32241261e+00 5.83818331e-02 -1.63746357e-01 -4.46877152e-01 -1.75679410e+00 3.24843496e-01 2.27260187e-01 3.47482935e-02 -3.91451985e-01 6.13401830e-01 -7.01917037e-02 -8.15588892e-01 6.07011616e-01 -2.95147691e-02 -3.19852322e-01 9.21577960e-02 1.10952163e+00 7.66787648e-01 -3.72042879e-02 -5.55163503e-01 -3.98861349e-01 4.25762944e-02 5.20711899e-01 -7.71239996e-01 1.10674238e+00 -3.04064602e-01 -2.35107943e-01 6.30459666e-01 4.40210938e-01 4.02912259e-01 -1.18899071e+00 3.12357545e-01 -9.61713567e-02 -4.83154178e-01 -6.25673115e-01 -9.28703547e-01 -3.06410581e-01 4.81239498e-01 3.64348739e-01 7.48625696e-01 4.04103547e-01 2.98104901e-02 -1.21596837e-02 7.15109408e-01 7.44497418e-01 -1.35775459e+00 3.64849687e-01 5.08777320e-01 1.42869031e+00 -9.09794986e-01 -2.27170840e-01 -5.02162099e-01 -1.18474686e+00 1.21209741e+00 7.36484468e-01 1.66468069e-01 2.34629259e-01 6.50785863e-01 6.65573597e-01 -6.50462091e-01 -9.47015166e-01 -3.89452666e-01 1.01133503e-01 1.13757443e+00 4.93582577e-01 2.44086325e-01 -1.59445122e-01 2.56181192e-02 -2.91112095e-01 -1.63986176e-01 7.32287526e-01 1.24853194e+00 -6.82135284e-01 -1.13715613e+00 -9.44380701e-01 -3.92181396e-01 3.03186953e-01 2.91064650e-01 -1.03316402e+00 8.53006065e-01 1.04540102e-01 1.52452791e+00 2.45072082e-01 -7.43031129e-02 7.18850553e-01 -1.98164657e-02 1.64229020e-01 -7.50436246e-01 -1.14534855e+00 -2.07086191e-01 8.77573788e-01 -4.57222909e-01 -3.49112540e-01 -6.00529850e-01 -1.28661668e+00 -4.52896178e-01 5.18939421e-02 3.64658862e-01 6.66058123e-01 7.75074542e-01 2.09418610e-01 8.93023983e-02 2.74848998e-01 -1.23460937e+00 9.41917822e-02 -1.07263160e+00 -3.01771462e-01 2.06739232e-01 8.70797634e-02 -3.69470686e-01 -5.04635759e-02 1.21016920e-01]
[5.31419038772583, 0.2904457747936249]
f9453a5c-8045-4375-b46d-8e5a52e559f2
what-s-cracking-a-review-and-analysis-of-deep
2202.03714
null
https://arxiv.org/abs/2202.03714v1
https://arxiv.org/pdf/2202.03714v1.pdf
What's Cracking? A Review and Analysis of Deep Learning Methods for Structural Crack Segmentation, Detection and Quantification
Surface cracks are a very common indicator of potential structural faults. Their early detection and monitoring is an important factor in structural health monitoring. Left untreated, they can grow in size over time and require expensive repairs or maintenance. With recent advances in computer vision and deep learning algorithms, the automatic detection and segmentation of cracks for this monitoring process have become a major topic of interest. This review aims to give researchers an overview of the published work within the field of crack analysis algorithms that make use of deep learning. It outlines the various tasks that are solved through applying computer vision algorithms to surface cracks in a structural health monitoring setting and also provides in-depth reviews of recent fully, semi and unsupervised approaches that perform crack classification, detection, segmentation and quantification. Additionally, this review also highlights popular datasets used for cracks and the metrics that are used to evaluate the performance of those algorithms. Finally, potential research gaps are outlined and further research directions are provided.
['Gordon Morison', 'Peter Barrie', 'Mike Mannion', 'Mark Jenkins', 'Jacob König']
2022-02-08
null
null
null
null
['crack-segmentation']
['computer-vision']
[ 2.98663616e-01 -4.28706482e-02 -1.61450818e-01 -1.36903867e-01 -7.96469331e-01 -8.76291245e-02 -3.46021861e-01 6.44812047e-01 -2.35442385e-01 1.95863426e-01 -4.26780656e-02 -1.00538626e-01 -7.92188477e-03 -9.91589367e-01 -3.91428113e-01 -8.92524540e-01 -6.07932545e-02 4.03542489e-01 2.84032315e-01 -5.66903576e-02 7.25166261e-01 5.67068875e-01 -1.69396842e+00 4.14766550e-01 5.87270677e-01 1.03069448e+00 -1.00059681e-01 9.06472981e-01 4.05845672e-01 6.91841245e-01 -6.28461182e-01 2.77624995e-01 -3.93411130e-01 -1.50408342e-01 -1.00625277e+00 4.71297562e-01 4.90606755e-01 -6.20354414e-01 -2.31363222e-01 5.80261171e-01 9.22597587e-01 -2.37991050e-01 8.30248237e-01 -6.05596662e-01 -4.81203198e-01 4.97571826e-01 -6.68487370e-01 6.61693037e-01 3.64452064e-01 1.06950561e-02 1.03813887e+00 -1.36758566e+00 1.98474199e-01 7.66404390e-01 1.18209076e+00 2.82305688e-01 -7.76193142e-01 -1.47871569e-01 -3.35129827e-01 3.35024267e-01 -8.95749032e-01 -2.11672217e-01 1.09621358e+00 -1.02080083e+00 1.14206970e+00 4.71276976e-02 7.95372367e-01 6.05033278e-01 3.03126544e-01 8.84137213e-01 6.42778814e-01 -7.20298350e-01 3.37300003e-01 -6.52227402e-01 7.23576784e-01 9.80963945e-01 4.92335439e-01 1.33030443e-03 -2.64796436e-01 -5.85265495e-02 8.21694136e-01 -1.78483110e-02 -1.07532904e-01 3.36019732e-02 -7.87872255e-01 1.23931861e+00 3.43883544e-01 5.98955572e-01 -5.01738071e-01 3.61895800e-01 6.12968564e-01 2.61069983e-01 4.00729954e-01 3.50582391e-01 -1.69414699e-01 1.74846470e-01 -1.23225951e+00 3.32448244e-01 2.41091773e-01 7.18375444e-02 5.82684457e-01 3.14378291e-01 -9.67220441e-02 1.22182262e+00 5.63284039e-01 5.39263666e-01 1.32710189e-01 -1.12039232e+00 9.92737636e-02 7.05575466e-01 -2.85204649e-01 -1.13856971e+00 -7.77456939e-01 -3.81566137e-01 -6.60004854e-01 6.08162761e-01 2.44319022e-01 -4.60616022e-01 -9.89331126e-01 3.85482371e-01 2.60962725e-01 -2.74800509e-01 -5.27207136e-01 5.61564684e-01 1.05479491e+00 2.44841561e-01 -3.37325841e-01 -4.27375408e-03 1.26038003e+00 -1.06141210e+00 -8.24643552e-01 -4.60702449e-01 4.93905693e-01 -5.34176946e-01 8.39844048e-01 4.81695235e-01 -9.70862985e-01 -4.04366076e-01 -1.28075361e+00 3.70141938e-02 -1.73575744e-01 3.60758185e-01 3.58429521e-01 9.17274594e-01 -6.49398088e-01 6.51822746e-01 -1.17993522e+00 -2.71713644e-01 9.31777775e-01 1.75807729e-01 3.64997163e-02 -1.05977111e-01 -9.00742948e-01 6.48872912e-01 -1.00060530e-01 4.61712360e-01 -1.25319386e+00 -4.11789387e-01 -8.86668801e-01 -4.21344668e-01 3.49681303e-02 -2.69440979e-01 1.25278080e+00 -3.39048635e-03 -9.54042375e-01 1.06808925e+00 1.32309496e-02 -1.78989530e-01 8.18176344e-02 -6.36848330e-01 1.29673630e-01 4.76855069e-01 1.39433503e-01 7.35009611e-02 1.05431247e+00 -1.30458796e+00 -7.96960592e-01 -4.13082510e-01 -5.32542691e-02 -2.74614275e-01 -1.77441344e-01 1.98671937e-01 2.12725669e-01 -7.29057908e-01 4.81393009e-01 -5.59273183e-01 -2.98300683e-01 3.95290628e-02 -7.30751634e-01 -4.19455469e-01 1.05033088e+00 -9.73206758e-01 1.43073022e+00 -1.63147020e+00 5.46343103e-02 1.79490373e-01 4.25943047e-01 5.33383787e-02 6.11775517e-01 8.72124255e-01 -8.68462026e-02 1.00739505e-02 -1.18161809e+00 -3.33353519e-01 -5.10745883e-01 1.78472757e-01 3.22311223e-01 7.84785628e-01 1.20638106e-02 6.17635846e-01 -7.59573638e-01 -4.41427231e-01 2.04310015e-01 2.64126539e-01 -3.13672781e-01 -7.25166202e-02 2.46289685e-01 5.01767278e-01 -2.25575328e-01 1.37735021e+00 3.84099662e-01 -8.59264731e-02 -6.41642034e-01 -1.65260643e-01 -2.14184761e-01 -3.36545646e-01 -1.12278044e+00 1.24332774e+00 -1.04319789e-01 7.61610866e-01 2.11247444e-01 -1.52963543e+00 7.35374868e-01 7.10516870e-01 8.08256209e-01 -1.33514926e-01 3.85263443e-01 5.47665894e-01 -4.83646542e-01 -1.18464053e+00 -3.22706215e-02 -2.13219151e-01 1.40341192e-01 7.28847921e-01 -3.34118843e-01 -2.39459753e-01 2.03793883e-01 -2.49010727e-01 1.16826618e+00 -5.90862751e-01 -2.18813151e-01 -1.13717854e-01 6.09411061e-01 1.80784196e-01 2.96950072e-01 5.46277821e-01 -3.25580865e-01 8.99324358e-01 1.44885346e-01 -7.59480059e-01 -7.95524657e-01 -6.94171548e-01 -3.32496285e-01 7.06872582e-01 -2.57107109e-01 4.00540709e-01 -1.04966748e+00 -6.11833632e-01 8.06639716e-02 -4.62377779e-02 -9.08192158e-01 -1.17957756e-01 -7.67900586e-01 -1.14031899e+00 8.34968388e-01 9.25792098e-01 2.81995863e-01 -1.27022552e+00 -1.25337875e+00 3.19079846e-01 -3.26915920e-01 -6.75659418e-01 1.28726095e-01 1.30649814e-02 -1.42296171e+00 -1.73916614e+00 -1.08923149e+00 -1.30280674e+00 7.62944400e-01 1.37454912e-01 8.86984348e-01 9.70613420e-01 -1.04218245e+00 7.70917952e-01 -6.44256353e-01 -4.62239832e-01 -3.45997334e-01 -3.02208737e-02 -3.06505919e-01 -1.24161772e-01 5.74692786e-02 -2.27571309e-01 -8.95872653e-01 3.40991579e-02 -8.44663262e-01 -6.29958570e-01 3.69514495e-01 9.98664975e-01 6.44953787e-01 5.23883581e-01 7.32436895e-01 -8.83894026e-01 8.91045451e-01 -4.14006352e-01 4.55057919e-02 -2.10758019e-02 -8.91950667e-01 -5.29670417e-01 -4.25755501e-01 1.76655918e-01 -7.92950988e-01 3.71226445e-02 -6.50448084e-01 1.60234928e-01 -4.00548905e-01 9.71222579e-01 4.38187629e-01 -1.34837851e-01 1.01020575e+00 -3.24884981e-01 1.72345921e-01 -6.77521288e-01 3.09953541e-02 8.11817706e-01 3.84899318e-01 -4.68469560e-01 6.60422266e-01 8.91899407e-01 -1.25459462e-01 -1.41495717e+00 -1.04095972e+00 -8.01077068e-01 -1.03832293e+00 -8.30595613e-01 1.06330311e+00 -4.65578526e-01 -3.62014659e-02 1.52003789e+00 -1.05346048e+00 -3.08211297e-01 -3.05807680e-01 1.43985227e-01 -2.32723534e-01 7.92089701e-01 -1.23157656e+00 -9.15872872e-01 -8.15582037e-01 -1.26003671e+00 1.12762225e+00 2.10058078e-01 -2.86304634e-02 -1.23177874e+00 4.17686671e-01 1.00661922e+00 3.61605525e-01 9.23235357e-01 9.53012288e-01 1.53052315e-01 1.00795910e-01 -7.03714907e-01 6.25384971e-02 6.69927120e-01 3.51922601e-01 4.51228619e-01 -1.09739649e+00 -4.07893986e-01 2.25542739e-01 -3.67049187e-01 1.11667407e+00 9.02847826e-01 7.38225877e-01 1.98225170e-01 -5.58863640e-01 -1.16970398e-01 1.53279257e+00 1.77231878e-01 4.86839652e-01 4.06656504e-01 8.58521700e-01 9.03119445e-01 5.97635508e-01 4.29252714e-01 1.07984573e-01 3.14716324e-02 9.12248492e-01 -3.50650847e-01 -2.19166636e-01 5.20736039e-01 -7.37477988e-02 9.32220459e-01 -5.27822852e-01 2.33429849e-01 -1.42382860e+00 9.93645728e-01 -1.27712405e+00 -9.42990839e-01 -1.01166677e+00 1.59407842e+00 5.49988329e-01 1.84775203e-01 1.15460329e-01 1.40259576e+00 8.56204569e-01 -1.56676605e-01 -5.15342891e-01 -1.48978099e-01 5.77541022e-03 4.59784329e-01 2.70783544e-01 3.90503943e-01 -1.46188366e+00 4.11757618e-01 7.49584723e+00 2.38778770e-01 -9.41828549e-01 2.41933271e-01 4.50956076e-01 5.14460325e-01 1.58815295e-01 -2.12800145e-01 -2.36606672e-01 -8.38672444e-02 3.97288501e-01 9.64307606e-01 -1.62886739e-01 5.35617828e-01 3.24533284e-01 -4.79595453e-01 -7.79606640e-01 6.57437146e-01 -6.30197451e-02 -1.30957472e+00 -6.10875905e-01 -6.64357617e-02 8.43353927e-01 1.44352525e-01 5.80144301e-02 -3.02818567e-01 -4.16180402e-01 -8.55864048e-01 7.38043904e-01 4.05407101e-01 5.96949935e-01 -6.34745717e-01 1.08572280e+00 3.79738472e-02 -1.19284081e+00 -6.10061646e-01 -2.78390020e-01 -2.95364112e-01 5.67222238e-01 1.18106794e+00 -2.10332230e-01 2.80367762e-01 1.23935032e+00 8.72182369e-01 -5.27413130e-01 1.40305901e+00 -4.45560098e-01 1.22388804e+00 -1.56370968e-01 5.59332848e-01 1.30228996e-01 9.42473412e-02 3.71337086e-01 1.13311255e+00 1.07389592e-01 -1.88919142e-01 1.94287643e-01 6.50187433e-01 3.18322003e-01 -3.94085534e-02 -3.39781553e-01 2.18634933e-01 3.80429715e-01 8.42753410e-01 -1.10630083e+00 -8.22184794e-03 -5.41056216e-01 5.33772528e-01 -1.73804313e-01 1.42716661e-01 -6.69077694e-01 -6.18855476e-01 1.73802331e-01 4.46420550e-01 6.23753965e-02 -2.66759187e-01 -8.27580273e-01 -4.25470501e-01 -1.49514675e-01 -6.63942516e-01 6.19815648e-01 -2.66325384e-01 -1.27786994e+00 6.54403642e-02 -1.27020674e-02 -6.39393091e-01 4.89802122e-01 -6.80027306e-01 -1.00709820e+00 1.10102192e-01 -1.23073375e+00 -1.09947813e+00 -4.42823738e-01 3.05128753e-01 1.11162841e+00 6.31561950e-02 4.34427023e-01 5.65794587e-01 -1.07744634e+00 2.92421728e-01 4.97674085e-02 6.46166801e-01 2.30526298e-01 -1.23394811e+00 3.37830186e-01 8.25416267e-01 -1.05104670e-01 1.55445099e-01 6.17592275e-01 -9.17468548e-01 -9.25889850e-01 -8.88837516e-01 5.92685282e-01 -3.64637673e-01 3.12797755e-01 3.22376043e-01 -1.12346232e+00 3.61360461e-01 1.99695140e-01 -2.20103025e-01 5.88302612e-01 -1.79155305e-01 1.53480530e-01 2.85758823e-01 -1.22103608e+00 -1.18848398e-01 4.66276348e-01 -5.75631976e-01 -5.87478101e-01 4.91110176e-01 1.37384117e-01 -3.93282413e-01 -1.21742964e+00 5.27482629e-01 4.92954195e-01 -1.20259511e+00 1.09031343e+00 2.60585081e-02 5.47516346e-01 -1.12354085e-02 2.59526372e-01 -7.51552284e-01 -3.73379230e-01 6.20474741e-02 -2.71910846e-01 7.64979601e-01 3.33928406e-01 -3.01222771e-01 1.12441778e+00 2.45251372e-01 -8.09890091e-01 -1.16148841e+00 -9.07611191e-01 -1.95115447e-01 3.87570530e-01 -5.79779267e-01 5.38819470e-03 7.42245615e-01 -1.32446453e-01 1.07558668e-01 1.12982228e-01 2.78015107e-01 9.00146961e-01 -1.70650825e-01 7.48777613e-02 -1.82272434e+00 3.66751820e-01 -5.03843427e-01 -2.46169209e-01 -4.99295682e-01 -4.14945155e-01 -5.06779075e-01 3.46188694e-01 -2.23711181e+00 -1.09574638e-01 -2.62935996e-01 -4.08911407e-02 5.87222159e-01 -7.70782754e-02 4.83341038e-01 -6.86643243e-01 3.99316013e-01 2.31554404e-01 1.86522707e-01 1.27503181e+00 -6.10854685e-01 1.24850295e-01 4.29705352e-01 -3.47120166e-01 8.77213717e-01 1.13267875e+00 -6.24794543e-01 -8.61235708e-02 -6.89376771e-01 4.41943616e-01 -2.01460779e-01 3.72242242e-01 -1.41718531e+00 1.93555623e-01 1.52386293e-01 -5.36923222e-02 -9.67613101e-01 3.13777849e-02 -6.58583701e-01 -4.10142690e-01 1.09240103e+00 -6.40913919e-02 5.71140125e-02 -8.93397182e-02 8.12427342e-01 -3.22065920e-01 -9.21788454e-01 1.20982707e+00 -2.29060292e-01 -4.74972934e-01 8.02459195e-02 -1.15009093e+00 -1.00166216e-01 1.09949899e+00 -7.31180668e-01 5.00865020e-02 -9.07598436e-02 -1.01347268e+00 1.34059086e-01 1.92363843e-01 9.54744890e-02 1.21200526e+00 -9.47537363e-01 -8.07721972e-01 -6.87295198e-02 6.01496063e-02 4.60291594e-01 4.73789841e-01 1.10420036e+00 -1.23667574e+00 1.52154705e-02 -1.30906450e-02 -8.87478650e-01 -1.20178282e+00 3.53611350e-01 5.97665966e-01 -7.77161196e-02 -8.26182783e-01 1.25836849e+00 -5.43511450e-01 -5.66010959e-02 2.75079221e-01 -6.71967626e-01 -7.74499178e-01 2.47538313e-01 3.26264560e-01 1.28732848e+00 5.88502645e-01 -7.10591972e-01 -1.61239579e-01 1.18253076e+00 3.31007510e-01 2.85227567e-01 1.47007895e+00 -2.81565547e-01 -5.01829445e-01 6.04117215e-01 7.50313640e-01 -3.31889063e-01 -9.74369764e-01 5.26855700e-02 2.86021322e-01 1.98198140e-01 4.00740117e-01 -3.30427617e-01 -1.82636952e+00 1.41170120e+00 1.12936592e+00 4.18199599e-01 9.64268923e-01 2.88232446e-01 1.29137230e+00 2.20128760e-01 1.83204487e-01 -1.57289064e+00 6.70610428e-01 4.98872429e-01 9.97160733e-01 -1.40812945e+00 1.09623618e-01 -4.49724019e-01 -1.79069400e-01 1.46061492e+00 3.05320948e-01 -3.94922167e-01 1.17128825e+00 2.14670941e-01 3.17671716e-01 -1.28520548e+00 1.90994263e-01 -2.73568220e-02 4.35108319e-02 8.79600108e-01 5.74993730e-01 -2.01676503e-01 -1.69657409e-01 1.20648645e-01 4.40782309e-01 -5.06142914e-01 5.98554730e-01 1.65563226e+00 -1.30054855e+00 -9.14736688e-01 -8.77965212e-01 7.07794666e-01 -8.14052582e-01 4.22021866e-01 -3.76546025e-01 3.41841489e-01 2.54501313e-01 1.59019887e+00 -3.93388242e-01 -2.91941792e-01 3.88162047e-01 -1.55930072e-01 1.75796524e-01 -9.15547431e-01 -6.30190909e-01 -8.28429237e-02 -4.15844098e-02 -1.70774817e-01 -6.56834602e-01 -7.81835020e-01 -1.64172804e+00 3.33712250e-01 -9.04136300e-01 -1.14015408e-01 6.00446165e-01 1.07701552e+00 -3.81232202e-01 8.43635619e-01 4.41956103e-01 -9.46808338e-01 -1.77631378e-01 -1.01343906e+00 -6.84108019e-01 -9.95848142e-03 5.74636042e-01 -1.03274703e+00 -3.79676133e-01 5.19344568e-01]
[7.492422580718994, 1.5522392988204956]
e135eca3-f474-4fb3-90ad-238e9f4a62ee
invariant-deep-compressible-covariance
2011.05702
null
https://arxiv.org/abs/2011.05702v1
https://arxiv.org/pdf/2011.05702v1.pdf
Invariant Deep Compressible Covariance Pooling for Aerial Scene Categorization
Learning discriminative and invariant feature representation is the key to visual image categorization. In this article, we propose a novel invariant deep compressible covariance pooling (IDCCP) to solve nuisance variations in aerial scene categorization. We consider transforming the input image according to a finite transformation group that consists of multiple confounding orthogonal matrices, such as the D4 group. Then, we adopt a Siamese-style network to transfer the group structure to the representation space, where we can derive a trivial representation that is invariant under the group action. The linear classifier trained with trivial representation will also be possessed with invariance. To further improve the discriminative power of representation, we extend the representation to the tensor space while imposing orthogonal constraints on the transformation matrix to effectively reduce feature dimensions. We conduct extensive experiments on the publicly released aerial scene image data sets and demonstrate the superiority of this method compared with state-of-the-art methods. In particular, with using ResNet architecture, our IDCCP model can reduce the dimension of the tensor representation by about 98% without sacrificing accuracy (i.e., <0.5%).
['Ling Shao', 'Yu Guan', 'Gerard Parr', 'Yi Ren', 'Shidong Wang']
2020-11-11
null
null
null
null
['image-categorization']
['computer-vision']
[ 1.89652860e-01 -4.09102023e-01 9.25201774e-02 -2.15403110e-01 -8.38251337e-02 -8.57752621e-01 3.67332846e-01 -5.39053380e-01 -2.34778181e-01 -7.84991831e-02 2.87560165e-01 -2.39696756e-01 -3.79135996e-01 -4.70427334e-01 -4.87881511e-01 -7.13996112e-01 -1.54460073e-01 -2.80157447e-01 -1.08371764e-01 -3.42473418e-01 1.83148190e-01 5.80149174e-01 -1.31050861e+00 2.51868546e-01 8.35780621e-01 1.31950903e+00 -1.92350633e-02 3.80031466e-01 3.24191868e-01 8.49222243e-01 -4.17028189e-01 -1.26870707e-01 8.25361729e-01 -1.79535627e-01 -1.00582600e+00 4.96973485e-01 6.99366391e-01 -4.73320931e-01 -8.08272004e-01 1.27568161e+00 1.30412653e-01 3.43366951e-01 7.08476484e-01 -1.21971881e+00 -1.24893725e+00 4.62598503e-01 -5.85869253e-01 9.73773748e-02 -7.64986053e-02 6.59499259e-04 1.20156252e+00 -1.02946341e+00 4.43907499e-01 1.34924889e+00 2.46533915e-01 3.08408558e-01 -1.28037810e+00 -5.93211174e-01 5.44613421e-01 2.06168756e-01 -1.69106328e+00 -2.11118788e-01 8.06347907e-01 -6.74037576e-01 6.33133054e-01 4.56281155e-01 4.25776452e-01 6.65710390e-01 2.93029517e-01 6.26941264e-01 7.59871960e-01 7.01360628e-02 6.99113542e-03 -4.20045733e-01 2.56371409e-01 9.61824775e-01 1.78092465e-01 -2.56869704e-01 -8.94482061e-02 1.89619628e-03 7.92361856e-01 5.35364032e-01 -3.23205709e-01 -6.52243376e-01 -1.64156663e+00 9.20364380e-01 1.12331963e+00 2.05640823e-01 -2.34926358e-01 1.33367285e-01 5.87605417e-01 4.12015945e-01 3.37698609e-01 6.68155193e-01 -1.86237171e-01 5.95910311e-01 -4.81540889e-01 2.01225765e-02 3.52567911e-01 1.01509607e+00 5.85052073e-01 2.31978565e-01 -4.54158396e-01 9.17006373e-01 2.53179461e-01 6.49092972e-01 7.00663328e-01 -9.75928068e-01 5.44992626e-01 9.92135286e-01 -2.05499694e-01 -1.51128316e+00 -3.95360678e-01 -4.77511555e-01 -1.44843924e+00 -2.38325685e-01 4.77651469e-02 2.45319121e-02 -1.07461238e+00 1.73219514e+00 9.47059914e-02 -5.91763519e-02 7.77333826e-02 1.19179821e+00 6.54451251e-01 6.19696081e-01 -2.72066295e-01 2.94213265e-01 1.34069479e+00 -1.00974619e+00 -4.36691761e-01 -3.28836329e-02 7.62945473e-01 -7.39005387e-01 1.14927888e+00 1.41511187e-01 -4.14479762e-01 -5.75789213e-01 -1.39679253e+00 -3.19012880e-01 -2.20509380e-01 7.26888537e-01 8.16397429e-01 1.70437455e-01 -7.80899584e-01 6.46193683e-01 -8.60601842e-01 -2.91027993e-01 4.37795043e-01 4.23925221e-01 -6.87513292e-01 -4.20263052e-01 -8.76383781e-01 4.46923912e-01 2.91248083e-01 3.17946136e-01 -9.11528885e-01 -5.97074807e-01 -9.51039314e-01 7.02155381e-02 2.48779193e-01 -6.96346164e-01 7.34154820e-01 -6.67354047e-01 -1.40738773e+00 5.50758719e-01 -2.03829557e-02 -2.14831308e-01 2.93041348e-01 -1.28486708e-01 -1.13809235e-01 2.59839833e-01 2.74654120e-01 5.65032303e-01 1.24666846e+00 -7.49180436e-01 -2.14941561e-01 -3.17801625e-01 5.69378853e-01 1.43487468e-01 -7.01581359e-01 1.17302805e-01 -2.44318396e-01 -8.96983147e-01 7.77763903e-01 -1.42329776e+00 -3.11617136e-01 2.63329953e-01 -6.51626348e-01 -2.50757903e-01 8.21852505e-01 -6.59387410e-01 1.02888405e+00 -2.59247971e+00 7.62163758e-01 2.26933092e-01 5.23368657e-01 -6.57819882e-02 -4.95399147e-01 1.01434231e-01 -4.21901733e-01 3.35370570e-01 -4.24130201e-01 -7.81526566e-02 5.52331246e-02 1.49641693e-01 -7.47609735e-01 6.47657633e-01 3.77756655e-01 8.07999372e-01 -6.36642516e-01 -4.59342822e-02 3.98583710e-02 2.61718541e-01 -8.68255496e-01 1.53381407e-01 3.07057023e-01 1.83426872e-01 -5.90521038e-01 5.81064045e-01 9.97120619e-01 -3.26761037e-01 5.73396273e-02 -6.66107237e-01 -7.15209246e-02 9.36324969e-02 -1.07730222e+00 1.81378925e+00 -2.67389297e-01 5.30888915e-01 -1.57777980e-01 -1.13599539e+00 8.47887456e-01 -2.62713969e-01 4.72104520e-01 -1.02924228e-01 3.01179141e-01 1.09770549e-02 4.10860419e-01 -1.35875612e-01 5.25501609e-01 1.97539613e-01 -2.15663850e-01 3.73700321e-01 2.57750183e-01 -8.07504505e-02 2.16805220e-01 2.92026490e-01 8.88692379e-01 -1.85970008e-01 1.09229356e-01 -5.66682160e-01 6.75589800e-01 -4.04466510e-01 6.56297565e-01 3.92267853e-01 -2.31478378e-01 5.71580112e-01 6.76719427e-01 -7.74813890e-01 -8.87274444e-01 -8.17082942e-01 -1.58119872e-01 9.85604465e-01 3.13808560e-01 -6.50374711e-01 -5.94620883e-01 -7.93849349e-01 5.89226596e-02 8.22415277e-02 -8.43499541e-01 -5.92781723e-01 -4.29218233e-01 -8.72428179e-01 3.60436112e-01 5.55008471e-01 1.16321933e+00 -3.19131523e-01 -1.26045555e-01 -2.58827478e-01 -1.64328292e-01 -1.20026517e+00 -7.75789380e-01 -7.84907490e-02 -7.09632874e-01 -9.79595363e-01 -7.26645470e-01 -6.57745898e-01 1.05298710e+00 8.13476980e-01 3.64117742e-01 1.20230965e-01 -1.81182683e-01 2.91401237e-01 -4.53688711e-01 2.86137089e-02 2.27785304e-01 1.48825422e-01 4.15792972e-01 4.09237832e-01 1.59902707e-01 -2.23648876e-01 -6.50578260e-01 5.92497051e-01 -1.10946226e+00 3.58484918e-03 4.30976868e-01 1.04064894e+00 4.73083466e-01 8.74836668e-02 2.11413484e-03 -2.62480110e-01 5.37453353e-01 -3.13989103e-01 -4.91469413e-01 2.29264006e-01 -1.42450303e-01 2.83099920e-01 8.94305408e-01 -6.49438441e-01 -4.19974715e-01 1.82653040e-01 3.08242261e-01 -8.08345556e-01 2.67147183e-01 5.87015927e-01 -3.22000496e-02 -5.91143608e-01 6.29140615e-01 2.51967251e-01 7.53996242e-03 -4.21923250e-01 4.16621149e-01 5.14903069e-01 1.86075017e-01 -4.67868567e-01 1.33250928e+00 6.10603452e-01 2.73353696e-01 -8.26930821e-01 -1.09372461e+00 -2.65991390e-01 -1.01673639e+00 2.14689389e-01 8.20900977e-01 -9.84794378e-01 -6.73921227e-01 4.61082429e-01 -1.01529169e+00 1.75307374e-02 -1.48760840e-01 6.47576451e-01 -2.51904696e-01 6.43068671e-01 -5.57243884e-01 -1.18837342e-01 -2.16828659e-01 -1.26266181e+00 8.70363295e-01 -1.32405847e-01 1.82916775e-01 -5.34649551e-01 -2.97189146e-01 1.07501730e-01 4.27516669e-01 1.73292279e-01 8.45491230e-01 -3.90129477e-01 -6.93615973e-01 -2.85224080e-01 -5.21516740e-01 7.37908185e-01 2.33398244e-01 -3.56918313e-02 -6.51702642e-01 -6.64252579e-01 1.54551435e-02 -3.00219893e-01 1.24538541e+00 -4.49803611e-03 1.63688886e+00 -3.91873747e-01 -3.22575383e-02 1.20449460e+00 1.09208345e+00 -1.73699372e-02 4.08257097e-01 2.59760201e-01 1.15633261e+00 4.11540657e-01 3.91011626e-01 2.88274467e-01 3.62481475e-01 5.23276627e-01 3.20083290e-01 -2.55551711e-02 -6.19253796e-03 -2.43448377e-01 3.70969117e-01 1.19448316e+00 -3.07199687e-01 1.78740323e-01 -6.54967248e-01 1.93697274e-01 -1.77243090e+00 -5.88452637e-01 2.70043999e-01 1.91155279e+00 3.35042685e-01 -1.73557654e-01 -3.44313562e-01 1.28388911e-01 5.59533060e-01 4.77286637e-01 -7.15074480e-01 -1.86526656e-01 -9.43233445e-02 -2.97295660e-01 7.23219514e-01 1.34786189e-01 -1.58046329e+00 1.04097164e+00 5.87936831e+00 6.35312259e-01 -1.43326283e+00 -2.59389639e-01 2.38360688e-01 2.53392905e-01 -2.05064621e-02 -2.27818033e-03 -4.16134417e-01 2.43120000e-01 3.54683578e-01 -4.01536942e-01 8.54285419e-01 1.06775033e+00 -3.11571300e-01 5.96134603e-01 -1.13289881e+00 1.21069121e+00 4.73855615e-01 -1.05809677e+00 6.49384975e-01 1.66815445e-01 7.20340908e-01 -4.67042625e-02 3.91730458e-01 4.68001753e-01 1.57487273e-01 -8.96699727e-01 6.33700192e-01 3.90845150e-01 9.71092224e-01 -5.21001160e-01 6.36721849e-01 2.62774676e-02 -1.46414959e+00 -3.82905215e-01 -1.13615119e+00 -1.11694552e-01 -3.92115027e-01 2.51256913e-01 -2.80108303e-01 6.92353606e-01 6.81965232e-01 1.32585275e+00 -8.66292775e-01 7.68184841e-01 -1.07327044e-01 3.82593513e-01 -2.66533673e-01 2.06046745e-01 4.53701764e-01 -4.47106808e-01 5.59890926e-01 7.85035670e-01 4.54542130e-01 2.90455431e-01 3.81062537e-01 8.25037658e-01 -4.40358400e-01 1.72030896e-01 -9.29837942e-01 -2.34577850e-01 1.01717763e-01 1.41267180e+00 -6.07434869e-01 -2.95290530e-01 -3.28953117e-01 1.21183538e+00 3.71045679e-01 4.73021686e-01 -5.05690932e-01 -7.88763821e-01 9.06441629e-01 -4.58122551e-01 5.03513455e-01 -5.93349576e-01 8.93226787e-02 -1.92382729e+00 1.15234584e-01 -9.10337508e-01 2.63410211e-01 -5.22543490e-01 -1.41533303e+00 8.22030246e-01 -4.97708842e-03 -1.72632253e+00 1.58101574e-01 -1.13283873e+00 -4.56367522e-01 7.67159998e-01 -1.23032260e+00 -1.43266249e+00 -5.56148946e-01 7.48386443e-01 3.75147432e-01 -4.89833474e-01 8.32896471e-01 1.79046169e-01 -8.64659369e-01 7.65241504e-01 1.00763306e-01 5.59987426e-01 5.15851974e-01 -9.40593302e-01 3.90532434e-01 1.04920757e+00 1.38303687e-04 1.13928199e+00 1.25594735e-01 -1.38093054e-01 -1.61756361e+00 -1.52908671e+00 2.63139755e-01 -3.11907262e-01 8.96911621e-01 -6.36407793e-01 -8.76754642e-01 9.21924770e-01 -1.15464360e-01 4.56842244e-01 7.62646735e-01 1.70657784e-02 -9.45057690e-01 -3.46610278e-01 -8.47429395e-01 8.89117539e-01 1.26564264e+00 -8.35650504e-01 -5.53851068e-01 5.15860260e-01 1.05584836e+00 -2.92344689e-01 -1.04442334e+00 4.89359945e-01 4.68552709e-01 -4.64312077e-01 1.00373471e+00 -9.85908926e-01 4.85379070e-01 -6.63122892e-01 -6.31860018e-01 -1.58705676e+00 -1.08342338e+00 -4.50292230e-01 2.43290991e-01 8.00147653e-01 9.77598131e-02 -8.61313283e-01 6.78942427e-02 4.67927963e-01 -2.24361584e-01 -6.84231341e-01 -7.79161274e-01 -1.02402747e+00 1.55401126e-01 -1.87567160e-01 8.53519380e-01 1.07053161e+00 -1.22506775e-01 3.81351799e-01 -5.10376453e-01 2.88975328e-01 5.42544782e-01 3.59913647e-01 7.11601138e-01 -1.08003759e+00 7.07908422e-02 -1.90540254e-01 -9.43304181e-01 -1.37596262e+00 4.91128147e-01 -1.38004982e+00 -2.80136406e-01 -1.11809170e+00 3.30695719e-01 -1.59561172e-01 -5.57492971e-01 6.72741950e-01 -2.82869842e-02 3.49223554e-01 6.95642829e-01 5.50311863e-01 -5.28777540e-01 9.34221745e-01 1.50709748e+00 -6.32285416e-01 9.28214118e-02 -4.36816245e-01 -9.85469341e-01 7.47794986e-01 7.30540097e-01 -2.37565264e-01 -3.89749080e-01 -7.38025069e-01 3.30289491e-02 -5.20708501e-01 3.13855231e-01 -9.61427271e-01 1.53263822e-01 -2.32578248e-01 5.07941723e-01 -2.41966650e-01 2.17782885e-01 -6.35555089e-01 -3.23849916e-01 5.39002180e-01 -5.46488404e-01 2.05930267e-02 9.19712484e-02 5.07240057e-01 -3.01233113e-01 -2.13681925e-02 7.58266628e-01 6.07803836e-02 -5.83474815e-01 7.48441041e-01 -8.18053782e-02 -2.11682707e-01 9.09271061e-01 1.62767351e-01 -4.93335754e-01 -6.66225925e-02 -4.22524095e-01 2.26115406e-01 3.98760408e-01 6.00620031e-01 7.68925309e-01 -1.77238417e+00 -6.10218048e-01 5.28952897e-01 3.75099897e-01 -1.96425959e-01 2.71843523e-01 7.58585811e-01 -5.64807296e-01 5.91573060e-01 -5.00785291e-01 -6.31440759e-01 -1.03172350e+00 8.33554804e-01 4.07467216e-01 4.11094315e-02 -7.12734103e-01 6.73583090e-01 7.28094220e-01 -6.62423313e-01 -1.38620570e-01 -7.60114253e-01 -3.55293185e-01 -7.72923529e-02 5.90437651e-01 1.61037207e-01 -7.83420876e-02 -8.92999053e-01 -5.35280287e-01 1.06099689e+00 -2.13747621e-01 3.43765557e-01 1.13886523e+00 -5.75344190e-02 -4.82244372e-01 1.85665667e-01 1.73198199e+00 -1.98342308e-01 -9.70655620e-01 -2.87991911e-01 -3.70190412e-01 -6.68229878e-01 2.09061936e-01 -1.54531389e-01 -1.19840586e+00 9.34025943e-01 5.29705226e-01 1.98034137e-01 1.12597549e+00 -2.50989109e-01 4.38503355e-01 1.03961313e+00 2.36970440e-01 -6.91301525e-01 1.48978651e-01 8.64064693e-01 1.52692366e+00 -1.14009356e+00 2.25792363e-01 -4.74742413e-01 -7.57251680e-01 1.09449673e+00 6.05827987e-01 -6.25803471e-01 8.04211974e-01 -4.94619250e-01 -1.03609316e-01 -2.16042340e-01 -5.67437649e-01 -8.81838575e-02 8.29641521e-01 3.42073262e-01 2.10755765e-01 2.81938195e-01 -8.11475590e-02 4.40170884e-01 -3.72697949e-01 -4.33875442e-01 5.05767703e-01 5.66780210e-01 -7.97142759e-02 -6.03304386e-01 -2.59295970e-01 5.42095244e-01 -2.46206179e-01 -5.89187853e-02 -4.66993690e-01 4.53832805e-01 -2.54922286e-02 7.09981024e-01 6.64505810e-02 -9.10728991e-01 4.59412545e-01 -3.20149362e-01 1.93452150e-01 -5.93122900e-01 -1.13425907e-02 -2.66910285e-01 -4.01602924e-01 -7.89412379e-01 -2.48953104e-01 -4.65609580e-01 -9.09113407e-01 -2.68079132e-01 -1.07493967e-01 -7.89601952e-02 4.59433615e-01 6.32493734e-01 4.48118180e-01 5.43962359e-01 1.12935162e+00 -9.86681223e-01 -1.03758049e+00 -9.82954502e-01 -7.40180552e-01 5.08408248e-01 4.27513838e-01 -9.23946738e-01 -6.94424868e-01 1.04069076e-01]
[9.019206047058105, 2.2378528118133545]
3c9c8d4f-a972-4ef8-bb3a-ae72e2c118d3
automatic-annotation-of-semantic-term-types
null
null
https://aclanthology.org/L18-1586
https://aclanthology.org/L18-1586.pdf
Automatic Annotation of Semantic Term Types in the Complete ACL Anthology Reference Corpus
null
["H{\\'e}ctor Mart{\\'\\i}nez Alonso", 'Anne-Kathrin Schumann']
2018-05-01
automatic-annotation-of-semantic-term-types-1
https://aclanthology.org/L18-1586
https://aclanthology.org/L18-1586.pdf
lrec-2018-5
['lexical-analysis']
['natural-language-processing']
[-8.63703638e-02 1.71006292e-01 -6.22772932e-01 -4.08054382e-01 -8.41685571e-03 -9.08429027e-01 6.55310392e-01 -6.53472245e-01 -2.85945535e-01 1.06888819e+00 -4.63127941e-02 -1.01159286e+00 -3.91567826e-01 -9.63214397e-01 -4.95059669e-01 -6.31337762e-01 -9.79754329e-01 7.25764990e-01 3.30370307e-01 -6.93831444e-01 7.03166842e-01 7.88774848e-01 -1.68942046e+00 7.18545914e-01 7.04417467e-01 8.52217197e-01 2.49141872e-01 1.14950800e+00 -1.95044339e-01 1.55633950e+00 -7.48382092e-01 -5.46825826e-01 3.13719302e-01 -1.23176083e-01 -7.22945035e-01 -1.01074085e-01 9.28529128e-02 -8.59008506e-02 -2.09758401e-01 9.22211111e-01 5.37373662e-01 4.49454933e-02 1.08379531e+00 -1.42548037e+00 -5.91619551e-01 6.10313773e-01 -4.01565880e-02 1.21627934e-01 1.03678203e+00 -5.39447069e-01 1.19919395e+00 -1.13026452e+00 7.20913768e-01 1.26888943e+00 8.66221786e-01 5.44149756e-01 -1.22286928e+00 -1.94712028e-01 -3.26822817e-01 -9.51717794e-02 -1.46558487e+00 -3.25250506e-01 4.25783843e-02 -2.08119690e-01 1.66093647e+00 1.26596653e+00 1.20609856e+00 1.01401424e+00 1.26658809e+00 8.34431887e-01 1.04267764e+00 -5.13792276e-01 3.35295945e-01 3.66983831e-01 1.54683650e-01 6.33519173e-01 8.40953708e-01 5.26628852e-01 -7.06372619e-01 -9.13127720e-01 9.33553874e-01 -2.94925272e-01 1.71355158e-01 -5.05680561e-01 -9.05919552e-01 6.91228509e-01 1.78732842e-01 3.83959889e-01 -1.39880210e-01 9.89067405e-02 1.26390755e-01 5.30987144e-01 -2.58292928e-02 6.47037446e-01 -9.11868811e-01 -1.33165747e-01 -8.71728659e-01 5.10332465e-01 1.25398111e+00 1.52653182e+00 1.24482810e-01 2.94908643e-01 -9.34252143e-02 3.17179203e-01 8.92314315e-01 1.01808000e+00 4.28362608e-01 -1.36146402e+00 -6.87414408e-02 1.72361732e-01 5.01781464e-01 -8.52631688e-01 -6.33224547e-01 -9.64177120e-03 -8.93263519e-01 4.49267089e-01 3.49161088e-01 4.57367361e-01 -8.02827001e-01 5.07305264e-01 4.33481112e-02 -2.34125629e-01 4.53833073e-01 5.55570945e-02 4.99930978e-01 3.76208365e-01 -1.34477139e-01 -5.73289394e-01 1.06082785e+00 -1.36716676e+00 -1.35299087e+00 2.33215362e-01 9.05734658e-01 -1.07320261e+00 4.35900748e-01 5.33875942e-01 -1.55548143e+00 -1.37560293e-01 -1.08699942e+00 1.78573877e-01 -7.27255583e-01 -3.14239264e-01 8.57801437e-01 1.43120694e+00 -1.60129595e+00 9.73287821e-01 -4.91727620e-01 6.59165755e-02 1.20568443e-02 8.24621081e-01 -2.64718989e-03 4.62812334e-01 -1.33193445e+00 1.08501506e+00 2.22979754e-01 -1.21242590e-01 -1.65216476e-01 -2.13068098e-01 -8.23704481e-01 -5.63443303e-01 -4.78693932e-01 -5.29636025e-01 1.44139910e+00 -2.59346128e-01 -1.65295815e+00 9.71794367e-01 -1.42069459e-01 -1.97814897e-01 6.14786744e-01 -1.28011424e-02 -8.31891418e-01 2.42498964e-01 -1.89849049e-01 5.76383233e-01 9.28263724e-01 -1.35132408e+00 -7.59897232e-01 -1.67359829e-01 -1.23336017e-01 2.66287565e-01 -1.25510961e-01 1.89734384e-01 2.11616129e-01 -1.12999000e-01 3.27147305e-01 -7.20919967e-01 -2.53068686e-01 -5.32041907e-01 -1.46512717e-01 -7.10518599e-01 7.70373225e-01 -4.51523662e-01 1.83705616e+00 -1.67618537e+00 -2.06720144e-01 3.98590982e-01 3.57815564e-01 -1.24705513e-03 2.40583986e-01 1.08380008e+00 -2.76906848e-01 7.32199550e-01 4.11965609e-01 -1.10722095e-01 2.24991128e-01 4.85861301e-01 -4.16602850e-01 3.05609167e-01 -1.29282743e-01 1.15307164e+00 -1.16605783e+00 -5.23096442e-01 4.11106765e-01 9.42391157e-02 -4.58732933e-01 4.79237735e-01 2.99364805e-01 1.47170946e-01 -3.56553018e-01 1.39399457e+00 1.15709066e+00 -1.31984919e-01 1.45911396e-01 5.30878425e-01 -3.79135728e-01 3.55090618e-01 -6.96863770e-01 1.05554795e+00 7.08333924e-02 5.00173986e-01 1.02364108e-01 -7.94621468e-01 3.33247900e-01 8.61540735e-01 4.37155962e-01 -9.67555881e-01 -2.26398129e-02 6.92409754e-01 1.12803578e-01 -6.07703328e-01 7.58228302e-01 3.94563079e-02 -3.64872098e-01 6.40070081e-01 -2.37588286e-01 -6.59476995e-01 9.64643434e-02 3.08100313e-01 6.22585893e-01 -5.18246442e-02 5.62923312e-01 -1.05613089e+00 7.86340594e-01 -1.82965681e-01 -1.81299388e-01 1.03415680e+00 -3.09923887e-01 3.19085121e-01 2.99841821e-01 -6.65102363e-01 -6.45341039e-01 -1.12307119e+00 -4.89381433e-01 1.30636716e+00 3.24267983e-01 -4.39044595e-01 -9.54439282e-01 -2.49762803e-01 1.77620783e-01 6.89606130e-01 -5.90509653e-01 3.84124845e-01 -5.03739953e-01 -8.32535863e-01 7.39044368e-01 3.45434904e-01 -5.07752821e-02 -1.33414865e+00 -6.58416986e-01 1.25490099e-01 -2.20292807e-01 -6.63697243e-01 -6.23428151e-02 4.48765576e-01 -1.35989368e+00 -5.18594682e-01 -6.66252747e-02 -8.20914626e-01 5.87345481e-01 2.46782884e-01 1.27047324e+00 5.39230824e-01 -2.31483161e-01 4.26904231e-01 -1.21292919e-01 -4.95818377e-01 -4.59671497e-01 -8.00336525e-02 5.28869390e-01 -5.87835789e-01 5.19427478e-01 -2.50617653e-01 -7.29350567e-01 5.37953973e-01 -6.88540697e-01 1.62748516e-01 1.79803044e-01 1.04410267e+00 1.35816500e-01 -9.34035778e-02 1.22507080e-01 -6.38007045e-01 8.72274399e-01 -1.69219792e-01 -3.78732830e-01 5.77745810e-02 -6.77108407e-01 -3.74140263e-01 3.21430594e-01 -3.25342178e-01 -1.01981449e+00 -4.87835288e-01 -9.82677937e-02 2.45538145e-01 1.11353043e-02 -1.46784872e-01 6.47139177e-02 -5.24923325e-01 8.02199244e-01 9.25758183e-02 1.99174434e-02 -6.80815242e-03 3.01039815e-01 7.09525108e-01 -6.82967342e-03 -6.68678164e-01 8.44880998e-01 4.91470337e-01 7.98524171e-02 -9.57177758e-01 -1.52186140e-01 -2.60129690e-01 -9.51962709e-01 -6.54426932e-01 6.56643391e-01 -6.78531289e-01 -9.10833478e-01 3.91110867e-01 -9.38691139e-01 -3.38627815e-01 -3.91645581e-01 4.25431967e-01 -1.01278400e+00 2.75717527e-02 -3.90154392e-01 -1.27895141e+00 -5.10977268e-01 -1.02017939e+00 9.43384409e-01 5.30070923e-02 -5.10597289e-01 -1.26927447e+00 5.87685481e-02 2.71537274e-01 1.81734428e-01 -1.73075795e-01 6.90226793e-01 -2.38256708e-01 -4.24233019e-01 -1.53791070e-01 2.34436691e-02 -1.39755070e-01 1.70832314e-02 4.95917559e-01 -9.81751978e-01 -5.31145096e-01 6.65065646e-02 -1.92070693e-01 -1.08835101e-01 6.52520418e-01 5.91872573e-01 -2.29931593e-01 -8.56000841e-01 5.40386558e-01 1.38545322e+00 3.85070026e-01 5.32770038e-01 7.28214979e-01 1.41836226e-01 5.53460240e-01 9.17806149e-01 4.63203549e-01 1.30579369e-02 3.28798652e-01 2.40537539e-01 1.49327129e-01 1.11720070e-01 -1.54819340e-01 3.77893507e-01 1.16112018e+00 -8.18235934e-01 -2.69281328e-01 -5.07867396e-01 4.42987174e-01 -1.72482407e+00 -1.40330648e+00 -4.32368398e-01 6.90478683e-01 6.25676990e-01 1.56016424e-01 -1.48347050e-01 3.35214496e-01 4.99015123e-01 -2.03574806e-01 -1.19133167e-01 -1.06291151e+00 -1.43546045e-01 3.15233678e-01 7.37729073e-01 1.00061214e+00 -7.20721722e-01 1.03317809e+00 1.29781246e+01 1.02230716e+00 2.21112028e-01 1.03134915e-01 5.16071796e-01 3.48020852e-01 -4.36954498e-01 -4.56139445e-02 -1.04416132e+00 2.72933897e-02 1.38140702e+00 -4.30666685e-01 6.85999811e-01 5.44219851e-01 3.44648361e-01 -4.23268199e-01 -1.26188684e+00 5.26221812e-01 9.73738134e-02 -1.40886843e+00 -2.83300440e-04 6.85225725e-01 7.73699820e-01 -5.08050561e-01 6.22419357e-01 3.24184299e-01 6.09259963e-01 -1.14389277e+00 8.60300779e-01 2.53660440e-01 1.03040910e+00 -6.05088234e-01 5.67372203e-01 1.68872893e-01 -1.14389896e+00 -2.20873043e-01 -8.77727985e-01 -1.00755692e+00 3.93533185e-02 -1.81779593e-01 -4.29956943e-01 3.48861217e-01 9.58353162e-01 2.99398601e-01 -3.93658698e-01 9.95779395e-01 -4.78694476e-02 1.04875881e-02 -2.71853864e-01 -4.48467314e-01 4.83122796e-01 -3.54241252e-01 4.66730654e-01 1.00164843e+00 2.48499006e-01 3.51035744e-01 -9.84472036e-02 4.01770771e-01 5.45058846e-01 3.29446048e-02 -1.19659424e+00 -1.78908288e-01 2.83276141e-01 9.16795909e-01 -4.83487815e-01 -4.22520459e-01 -2.00212970e-01 8.62069130e-01 -3.55488248e-02 5.01107454e-01 -6.11489356e-01 -4.35615242e-01 9.72222984e-01 -1.27327025e-01 -1.14700586e-01 -3.48497719e-01 -6.23769283e-01 -7.30352640e-01 -5.89872956e-01 -4.54965204e-01 5.93606755e-02 -5.53365827e-01 -1.39813089e+00 5.79277515e-01 -2.27688253e-02 -1.40553558e+00 -6.99901402e-01 -1.27676582e+00 -4.76714373e-01 4.92853165e-01 -1.11898029e+00 -1.10984349e+00 2.50124663e-01 4.52870727e-01 1.64141744e-01 -5.34416080e-01 1.39563632e+00 3.57715860e-02 1.00637585e-01 9.24474537e-01 6.69434488e-01 -7.37814724e-01 5.56605101e-01 -1.27867436e+00 5.68737745e-01 -1.37897313e-01 -4.31265175e-01 9.05828118e-01 6.28349900e-01 -5.39804697e-01 -1.41196322e+00 -3.66917729e-01 1.08350635e+00 -9.83769417e-01 6.55218959e-01 -3.86345625e-01 4.23767231e-02 7.88592756e-01 7.15902448e-01 -6.18741751e-01 8.21781039e-01 -1.83753878e-01 1.80774391e-01 5.75296998e-01 -1.39248300e+00 6.12354755e-01 1.66275799e+00 -4.63594139e-01 -6.25784039e-01 7.60327101e-01 8.13696027e-01 -6.94087505e-01 -1.30082703e+00 3.34633321e-01 8.65424156e-01 -8.75409484e-01 1.61978090e+00 -1.32660246e+00 -4.43697497e-02 2.81152606e-01 -2.61993498e-01 -9.32519078e-01 -5.96193194e-01 -1.23518765e+00 -5.33532679e-01 -5.83747849e-02 5.96577883e-01 -1.13057327e+00 3.42365682e-01 8.74560475e-01 -2.82833427e-01 -6.42737269e-01 -1.06996536e+00 -1.32016802e+00 -3.35779637e-02 -1.45572275e-01 4.90409225e-01 7.63798356e-01 6.83744550e-01 1.09839931e-01 -6.36873543e-02 -1.00294888e-01 5.33176839e-01 9.55312885e-03 4.41501856e-01 -1.34294486e+00 3.86843324e-01 -5.75816095e-01 -3.07655483e-01 -9.45992947e-01 -8.85957032e-02 -8.12076271e-01 -6.53862000e-01 -1.28511906e+00 -8.32044985e-03 -1.91056758e-01 -1.11109078e-01 -1.65725678e-01 3.67937148e-01 2.13746816e-01 1.20859891e-02 1.03788137e-01 -3.71160030e-01 6.18435517e-02 1.29639816e+00 6.91750320e-05 -1.62315920e-01 4.85058486e-01 -4.81304944e-01 7.84440815e-01 8.58408585e-02 -2.99253196e-01 -6.78878546e-01 6.11881316e-02 6.69384480e-01 4.61409837e-02 3.24159935e-02 -7.42885649e-01 5.37211418e-01 -3.75702560e-01 4.78586555e-01 -1.32223868e+00 1.30741090e-01 -9.61415648e-01 6.95283338e-02 9.40189242e-01 2.74610907e-01 1.20424610e-02 8.66204947e-02 5.39500564e-02 -1.44871444e-01 -5.70943117e-01 9.21121240e-01 -4.07591403e-01 -4.92852688e-01 -5.20386267e-03 -1.03226590e+00 8.97834301e-02 9.92593169e-01 -7.84614205e-01 -3.59281451e-01 -4.20183957e-01 -8.29068601e-01 -1.95836127e-02 6.50830388e-01 3.03609259e-02 7.30431557e-01 -1.51530886e+00 -2.30721906e-01 7.18729138e-01 -3.22939813e-01 -3.74400020e-01 -1.70157343e-01 6.58265352e-01 -1.32361674e+00 1.02442718e+00 -5.41665435e-01 -4.55340147e-01 -1.14228773e+00 4.78126436e-01 4.28307921e-01 -2.41845414e-01 -2.15481281e-01 1.11954463e+00 2.71224789e-02 -8.23025763e-01 1.85185194e-01 -9.21545625e-02 -7.54407048e-01 3.55081353e-03 6.88606799e-01 1.05194807e+00 -2.91290224e-01 -6.04341030e-01 -4.56784427e-01 6.65885091e-01 2.32151806e-01 -2.87484169e-01 9.17833567e-01 -2.19243199e-01 -9.89108324e-01 4.28274393e-01 8.38715494e-01 -1.36269778e-01 -3.69319022e-02 4.16855574e-01 1.25943512e-01 -8.02164078e-01 -4.32406247e-01 -3.55811834e-01 -1.85641110e-01 5.54822803e-01 5.34874737e-01 8.99602413e-01 8.57008278e-01 -3.02566767e-01 8.18335712e-01 9.66778398e-01 5.72402716e-01 -1.68019545e+00 -2.13140488e-01 6.89524531e-01 9.12339568e-01 -9.28350806e-01 5.44190466e-01 -7.27165341e-01 -4.14997995e-01 1.32979155e+00 4.68304873e-01 -1.55325383e-01 1.27306652e+00 5.74917436e-01 1.14069022e-02 -3.70670199e-01 -9.44949508e-01 1.12705544e-01 3.75366658e-01 1.11147714e+00 5.06513238e-01 5.07374525e-01 -9.66500878e-01 3.21953118e-01 -7.28706717e-01 -2.34555230e-01 4.90474731e-01 1.41972518e+00 -6.43810987e-01 -1.20391917e+00 -7.23931909e-01 4.61561680e-01 -5.49773455e-01 -1.16372630e-01 -5.28106689e-01 8.46754074e-01 -5.76629937e-02 1.49448860e+00 -1.97535474e-03 -5.03491640e-01 4.11356747e-01 1.41089618e-01 7.21762300e-01 -1.23501487e-01 -9.04846430e-01 4.15413082e-01 3.84890139e-01 -1.23056793e+00 -8.58632207e-01 -1.05834293e+00 -1.40667629e+00 -1.19437599e+00 -5.12782812e-01 1.89310342e-01 3.83317530e-01 3.90289724e-01 -2.06836104e-01 2.85260603e-02 9.80917513e-01 -1.07949340e+00 -3.90341938e-01 -9.39418614e-01 -1.00026262e+00 -7.84516707e-02 2.89751232e-01 -8.00943017e-01 -7.83523321e-01 2.83909619e-01]
[-7.269936561584473, 3.7788901329040527]
41ed5ddf-204b-4175-a61b-e84e75a731fd
a-framework-for-bidirectional-decoding-case
2305.1258
null
https://arxiv.org/abs/2305.12580v1
https://arxiv.org/pdf/2305.12580v1.pdf
A Framework for Bidirectional Decoding: Case Study in Morphological Inflection
Transformer-based encoder-decoder models that generate outputs in a left-to-right fashion have become standard for sequence-to-sequence tasks. In this paper, we propose a framework for decoding that produces sequences from the "outside-in": at each step, the model chooses to generate a token on the left, on the right, or join the left and right sequences. We argue that this is more principled than prior bidirectional decoders. Our proposal supports a variety of model architectures and includes several training methods, such as a dynamic programming algorithm that marginalizes out the latent ordering variable. Our model improves considerably over a simple baseline based on unidirectional transformers on the SIGMORPHON 2023 inflection task and sets SOTA on the 2022 shared task. The model performs particularly well on long sequences, can learn the split point of words composed of stem and affix (without supervision), and performs better relative to the baseline on datasets that have fewer unique lemmas (but more examples per lemma).
['Julia Hockenmaier', 'Marc E. Canby']
2023-05-21
null
null
null
null
['morphological-inflection']
['natural-language-processing']
[ 6.25183165e-01 3.48714411e-01 -2.32431769e-01 -4.50678855e-01 -1.21519005e+00 -9.68081653e-01 9.14451122e-01 -1.58734366e-01 -3.25268090e-01 9.17918324e-01 7.73672462e-01 -8.27036858e-01 4.99183327e-01 -7.35079169e-01 -9.53731358e-01 -6.82731688e-01 3.24374586e-01 7.83150792e-01 -9.37035158e-02 -4.14977580e-01 1.07517391e-01 -7.39248618e-02 -1.15499914e+00 7.58336484e-01 5.90313435e-01 3.77850860e-01 4.68142480e-01 6.78839266e-01 -2.34926939e-01 1.00277042e+00 -4.64188516e-01 -9.36653435e-01 1.86903268e-01 -8.30191016e-01 -8.46116483e-01 -3.20498735e-01 3.86848897e-01 -3.50784063e-01 -1.84054583e-01 7.33406544e-01 3.34380418e-01 -2.91236967e-01 9.88788962e-01 -9.05124962e-01 -8.70002508e-01 1.56321013e+00 -4.11852270e-01 3.39301415e-02 3.41030657e-01 7.14747235e-02 1.64678204e+00 -9.32066023e-01 8.85364532e-01 1.28627610e+00 6.34745002e-01 6.21339679e-01 -1.56049192e+00 -5.84257782e-01 1.47333488e-01 4.71640863e-02 -1.09873199e+00 -7.66928434e-01 2.75151163e-01 -5.95313549e-01 1.45163059e+00 1.60894588e-01 5.25535226e-01 1.34981585e+00 2.09067792e-01 9.25876439e-01 8.62699747e-01 -6.56765282e-01 -1.63713112e-01 -1.84607044e-01 -3.80321220e-02 6.55116796e-01 1.28085732e-01 1.30644232e-01 -7.92983949e-01 -3.14450860e-02 5.49899518e-01 -4.76730019e-01 -1.55952379e-01 -9.91903692e-02 -1.46189845e+00 6.87054157e-01 -1.12881258e-01 2.11780399e-01 -2.15348288e-01 4.42750603e-01 4.91392165e-01 4.73727137e-01 4.14619356e-01 3.61365527e-01 -6.93762779e-01 -4.59772199e-01 -1.09983730e+00 3.10211599e-01 8.99642646e-01 1.26460028e+00 6.00840449e-01 -3.78617160e-02 -4.18258280e-01 7.19146967e-01 1.87294871e-01 3.97132456e-01 4.22782868e-01 -7.64011443e-01 9.76212442e-01 -4.58711460e-02 4.61171456e-02 -2.77706772e-01 9.25743431e-02 -2.43037939e-01 -4.95772332e-01 -1.69684649e-01 5.47530591e-01 -2.04318687e-01 -1.04581761e+00 1.99728167e+00 -2.70371199e-01 -2.03700483e-01 -1.13466484e-02 4.32778001e-01 2.74553329e-01 9.52968955e-01 -2.99509503e-02 -1.17235355e-01 1.22395980e+00 -1.00769103e+00 -7.18446791e-01 -3.41221839e-01 1.00841630e+00 -7.93360174e-01 1.15468943e+00 4.06348288e-01 -1.53320324e+00 -2.97562063e-01 -7.99605668e-01 -6.22649670e-01 -1.96693420e-01 1.86633274e-01 4.79853451e-01 4.74491030e-01 -1.38133180e+00 6.36485696e-01 -8.90354335e-01 -1.64035484e-01 -6.93868995e-02 -1.62836730e-01 -3.23491573e-01 1.19338073e-01 -1.29321015e+00 1.14093065e+00 5.25643229e-01 -4.18601707e-02 -1.04595006e+00 -6.08448684e-01 -9.78662014e-01 1.07343867e-01 -1.36711344e-01 -8.30163538e-01 1.54161489e+00 -9.33144152e-01 -1.70338619e+00 1.01202714e+00 -5.78353107e-01 -6.83652520e-01 5.40545940e-01 -3.12095970e-01 -6.51718769e-03 -2.17783839e-01 2.91321933e-01 7.76396632e-01 5.85483849e-01 -9.06585336e-01 -6.23171091e-01 -1.14581227e-01 -2.26257771e-01 1.81626678e-01 4.98743542e-02 2.02627480e-01 -3.13907653e-01 -7.90237188e-01 -1.90367892e-01 -9.87909794e-01 1.74479991e-01 -3.72441322e-01 -6.49276257e-01 -3.09336424e-01 1.09661460e-01 -9.15593266e-01 1.30608582e+00 -1.85032320e+00 5.18093944e-01 -4.83895056e-02 -2.42778033e-01 -4.95990478e-02 -1.52159631e-01 8.81016731e-01 -4.01849121e-01 2.10966349e-01 -4.07128572e-01 -5.08963287e-01 1.83101594e-01 2.11225495e-01 -6.22087657e-01 1.91197738e-01 1.86736494e-01 1.11555469e+00 -9.45835650e-01 -3.59375447e-01 -1.92858636e-01 2.25338057e-01 -6.81861401e-01 7.00002536e-02 -4.15514141e-01 5.41681312e-02 1.97944596e-01 2.34042078e-01 1.83957562e-01 -2.68262684e-01 5.37294924e-01 1.22406594e-01 -2.69357115e-01 1.33799517e+00 -7.44495928e-01 1.89204800e+00 -6.45162523e-01 8.11544955e-01 -6.52637482e-02 -7.24583447e-01 7.33528316e-01 7.02506959e-01 -1.46613389e-01 -3.93846691e-01 -1.10883452e-01 3.92216712e-01 1.79551169e-01 -3.33405286e-01 6.70290351e-01 -4.55117613e-01 -2.30009302e-01 6.45668924e-01 2.34398931e-01 -2.62003720e-01 4.87798959e-01 4.35206175e-01 1.15001333e+00 7.14083970e-01 4.18292999e-01 -2.79314578e-01 1.71658576e-01 -1.14830539e-01 3.89622480e-01 7.56186485e-01 3.32363009e-01 6.42544746e-01 8.24850917e-01 -1.63993105e-01 -1.53754890e+00 -1.28032434e+00 9.66202244e-02 1.36638331e+00 -3.91356587e-01 -6.31520748e-01 -6.32378101e-01 -5.94041348e-01 -7.01634064e-02 1.36316252e+00 -4.87472296e-01 -4.00097594e-02 -8.39392781e-01 -3.35329592e-01 1.07573104e+00 6.33833885e-01 -1.52570441e-01 -1.10253274e+00 -3.46143097e-02 3.29058707e-01 -6.47320032e-01 -7.76990414e-01 -6.64660871e-01 5.70867956e-01 -6.59059346e-01 -5.53011954e-01 -5.45965970e-01 -9.91100550e-01 4.94577974e-01 -2.47017398e-01 1.39718211e+00 -2.77788132e-01 3.51899028e-01 -3.08068067e-01 -2.19957471e-01 -2.56298631e-01 -8.30457032e-01 4.73765790e-01 -3.23682308e-01 -2.85046041e-01 3.02321166e-01 -5.99648356e-01 5.92121854e-03 -1.40390009e-01 -6.26455307e-01 4.75480050e-01 6.51913226e-01 9.24198210e-01 2.26603970e-01 -5.67511916e-01 5.16721964e-01 -1.18455327e+00 5.67831516e-01 -5.45891106e-01 -3.96156311e-01 3.32473606e-01 -4.34327185e-01 5.92734218e-01 9.40875947e-01 -7.93505013e-02 -1.07342827e+00 2.12534331e-02 -3.64591867e-01 6.81772828e-02 2.83612832e-02 4.31702644e-01 -3.24968666e-01 8.94183636e-01 5.75681150e-01 6.06069207e-01 2.70173827e-04 -5.84894657e-01 8.82620990e-01 6.48477018e-01 5.51768482e-01 -7.22244263e-01 8.09243143e-01 2.15668604e-01 -3.13675821e-01 -4.93640810e-01 -6.51580513e-01 -1.36294857e-01 -6.32390738e-01 2.62869805e-01 7.29629278e-01 -1.11846673e+00 -3.55598986e-01 3.01969677e-01 -1.61648715e+00 -6.45604312e-01 -3.30640018e-01 2.51983047e-01 -7.72010207e-01 8.97913352e-02 -1.17412961e+00 -5.04910827e-01 -4.40461874e-01 -9.91159618e-01 1.10037339e+00 -3.93132448e-01 -7.58073092e-01 -9.25373435e-01 1.39969841e-01 4.35351208e-02 2.44634062e-01 -2.78376967e-01 1.32536769e+00 -6.55814648e-01 -6.78584516e-01 1.29962638e-01 7.04871565e-02 4.30333108e-01 1.33015171e-01 1.99273050e-01 -5.89872718e-01 -1.00597598e-01 -3.49919111e-01 -4.32169378e-01 1.18137598e+00 1.43521857e-02 7.52232671e-01 -5.36949813e-01 -3.96245629e-01 6.03846431e-01 1.11325884e+00 3.68721895e-02 7.66600490e-01 1.79264888e-01 6.68713868e-01 5.44203639e-01 -4.66158278e-02 1.25381753e-01 6.78352594e-01 3.61755282e-01 6.08302318e-02 3.43903482e-01 -4.54126626e-01 -9.03096199e-01 9.13466632e-01 1.22649789e+00 2.48578876e-01 -5.92520535e-01 -7.88166106e-01 1.00049138e+00 -1.49212158e+00 -1.18393743e+00 -2.27749303e-01 1.95884717e+00 1.42687690e+00 2.23517835e-01 -6.82004467e-02 4.72410098e-02 7.41994679e-01 3.63786608e-01 -1.06416002e-01 -6.95295751e-01 -2.08671883e-01 5.32488644e-01 6.41068161e-01 8.49865913e-01 -6.16698742e-01 1.30623388e+00 7.50574160e+00 9.12184000e-01 -9.15793300e-01 1.39899865e-01 4.89988208e-01 -2.43316159e-01 -1.04859352e+00 3.32922101e-01 -1.16861629e+00 7.37250209e-01 1.20753765e+00 -4.47818898e-02 7.00958788e-01 4.98540908e-01 1.33359775e-01 1.91553831e-01 -1.54673362e+00 5.25329769e-01 1.63245186e-01 -1.39317167e+00 2.76719362e-01 -6.90115690e-02 6.88948631e-01 8.38475451e-02 -3.75539884e-02 2.34215692e-01 9.51143682e-01 -1.06913984e+00 1.24948370e+00 3.06971818e-01 1.02412474e+00 -4.95579660e-01 1.75012171e-01 5.44950783e-01 -1.03005123e+00 2.39727184e-01 -2.13584956e-02 -3.10121417e-01 5.97536325e-01 5.17196774e-01 -1.09674728e+00 2.72109002e-01 1.30502060e-01 8.13950181e-01 -2.06949770e-01 3.99089664e-01 -1.05745184e+00 1.08269656e+00 -1.44109473e-01 -2.37095520e-01 1.63834065e-01 -3.16900223e-01 4.71038193e-01 1.69784546e+00 3.94327074e-01 -2.63330847e-01 -2.35363245e-01 7.65895188e-01 -2.56549329e-01 4.18570414e-02 -8.95022988e-01 -1.17800429e-01 6.38432145e-01 9.02220547e-01 -3.47292513e-01 -6.32889688e-01 -4.11469251e-01 1.07712018e+00 7.28408813e-01 3.99568737e-01 -7.79127777e-01 -4.91878957e-01 5.49261630e-01 1.92837507e-01 6.88485086e-01 -4.47822034e-01 -4.49625731e-01 -1.22890937e+00 -1.72021762e-02 -1.05730617e+00 2.64354259e-01 -8.35840881e-01 -1.11008418e+00 4.25844669e-01 -1.23896692e-02 -8.09018672e-01 -7.83527911e-01 -5.15810966e-01 -4.67411190e-01 1.11623836e+00 -1.31270134e+00 -1.07971883e+00 6.29616857e-01 2.41304085e-01 5.59003890e-01 6.79682270e-02 6.64423347e-01 2.96212554e-01 -2.78553486e-01 6.31134331e-01 2.42207110e-01 4.88126218e-01 7.60553718e-01 -1.43424690e+00 1.06057143e+00 1.20776904e+00 4.60079908e-01 9.71772313e-01 7.25020766e-01 -6.48769736e-01 -1.12818134e+00 -9.01219308e-01 1.79108441e+00 -6.30322337e-01 8.99132013e-01 -8.14781010e-01 -5.36189854e-01 1.48993313e+00 6.99885666e-01 -4.67919677e-01 5.95745265e-01 2.23561898e-01 -7.00042903e-01 1.49362296e-01 -4.33009833e-01 8.97037983e-01 1.33002985e+00 -7.93685257e-01 -8.83733571e-01 2.28534415e-01 8.66690814e-01 -3.79366755e-01 -4.80100006e-01 -4.61874157e-02 6.94969475e-01 -8.37344527e-01 5.39442897e-01 -7.42054820e-01 9.88044918e-01 -2.60752112e-01 -2.35722154e-01 -1.70070243e+00 -4.98177290e-01 -1.04569244e+00 -5.43612950e-02 1.28329015e+00 9.59365249e-01 -4.40353006e-01 5.71611762e-01 1.46688193e-01 -3.88074458e-01 -4.53536004e-01 -7.14693427e-01 -8.09733450e-01 4.67131793e-01 -4.43789661e-01 5.80950737e-01 5.68684399e-01 1.47338212e-01 8.70030403e-01 -5.05197883e-01 -2.99805820e-01 2.53421754e-01 6.83915541e-02 5.44144809e-01 -6.81746244e-01 -6.00831330e-01 -5.38072288e-01 9.68934447e-02 -1.75414109e+00 2.93552935e-01 -1.55571675e+00 3.85091186e-01 -1.61154413e+00 1.55717716e-01 -2.63184726e-01 8.63974262e-03 7.06023574e-01 -2.66895711e-01 -1.20391892e-02 2.35949904e-01 6.05856851e-02 -2.72840440e-01 5.25781453e-01 8.90334785e-01 -2.35604439e-02 2.24418938e-01 -3.24410528e-01 -8.78438890e-01 4.67185885e-01 6.28880024e-01 -7.02846467e-01 -2.88591802e-01 -9.25921261e-01 6.62235320e-01 2.66369045e-01 -8.94589499e-02 -3.74171913e-01 1.29268155e-01 -2.23799143e-02 1.19287491e-01 -7.00437486e-01 2.72240311e-01 -3.11817378e-01 1.66802108e-01 3.90569270e-01 -9.21032012e-01 4.30305451e-01 -2.04646200e-01 2.41315052e-01 -4.12041992e-02 -2.62323648e-01 5.37041783e-01 -4.20404613e-01 -1.13957651e-01 -4.06462587e-02 -6.72340035e-01 4.89430219e-01 6.53210580e-01 5.90255931e-02 -3.60710740e-01 -4.54614818e-01 -4.68409270e-01 -2.42498443e-02 4.73367453e-01 3.92911494e-01 2.09981859e-01 -1.29280150e+00 -1.01913786e+00 1.88041151e-01 9.16247163e-03 -2.33141288e-01 -5.14435053e-01 5.77929974e-01 -5.77244639e-01 6.14007831e-01 4.41397019e-02 -2.52634466e-01 -9.64264452e-01 4.60013002e-01 2.26038277e-01 -5.37765026e-01 -4.52630669e-01 1.16056895e+00 2.00512215e-01 -5.83014548e-01 -3.70352343e-03 -3.88427168e-01 2.36439288e-01 2.26165652e-01 3.93881530e-01 2.60831416e-01 -1.47074200e-02 -6.23170674e-01 -2.51462162e-01 1.26965106e-01 -2.86962569e-01 -7.49506772e-01 1.21259153e+00 -1.20251320e-01 -4.08726156e-01 8.43855321e-01 1.27204788e+00 3.29191953e-01 -1.18959677e+00 -1.62964404e-01 1.39193177e-01 -1.30400330e-01 -4.84668285e-01 -8.20996106e-01 -6.01160347e-01 1.05399930e+00 -4.49686527e-01 1.54121667e-01 5.40903866e-01 -7.82202836e-03 9.83994067e-01 1.93933249e-01 2.61374086e-01 -1.02618694e+00 -1.69934049e-01 9.69739437e-01 6.76403701e-01 -6.03363812e-01 -3.56020153e-01 -2.09625617e-01 -7.76723444e-01 1.01360011e+00 2.57371426e-01 -5.16552813e-02 2.32226580e-01 6.69285059e-01 -2.06066705e-02 2.04126492e-01 -1.35190594e+00 -1.00366779e-01 -1.26389548e-01 3.86204004e-01 1.01832306e+00 2.87733316e-01 -4.72475886e-01 1.82844460e-01 -8.03404093e-01 -9.70056877e-02 5.88690519e-01 7.18131244e-01 -3.15572917e-01 -1.46277010e+00 -9.20541212e-02 4.07768548e-01 -5.84121764e-01 -9.15823698e-01 -2.46044680e-01 5.95974445e-01 1.54716268e-01 9.39877510e-01 2.77362108e-01 -1.08234033e-01 -6.86083287e-02 7.03375041e-01 6.94759965e-01 -9.11264002e-01 -6.06555998e-01 5.72797917e-02 5.86028278e-01 -2.55712777e-01 8.80173966e-02 -1.03174484e+00 -1.15987551e+00 -2.84932524e-01 -1.17940173e-01 1.25713050e-01 4.46850210e-01 1.02193105e+00 3.83502126e-01 4.11890119e-01 3.72081518e-01 -5.31736910e-01 -7.63234615e-01 -1.06378770e+00 -4.34925526e-01 3.29202056e-01 3.67287576e-01 -7.74902776e-02 -2.94703811e-01 4.10819411e-01]
[11.34865665435791, 9.185531616210938]
87e58a1a-370e-46eb-8f92-455ec2c9c9e5
learning-to-predict-3d-lane-shape-and-camera
2112.15351
null
https://arxiv.org/abs/2112.15351v1
https://arxiv.org/pdf/2112.15351v1.pdf
Learning to Predict 3D Lane Shape and Camera Pose from a Single Image via Geometry Constraints
Detecting 3D lanes from the camera is a rising problem for autonomous vehicles. In this task, the correct camera pose is the key to generating accurate lanes, which can transform an image from perspective-view to the top-view. With this transformation, we can get rid of the perspective effects so that 3D lanes would look similar and can accurately be fitted by low-order polynomials. However, mainstream 3D lane detectors rely on perfect camera poses provided by other sensors, which is expensive and encounters multi-sensor calibration issues. To overcome this problem, we propose to predict 3D lanes by estimating camera pose from a single image with a two-stage framework. The first stage aims at the camera pose task from perspective-view images. To improve pose estimation, we introduce an auxiliary 3D lane task and geometry constraints to benefit from multi-task learning, which enhances consistencies between 3D and 2D, as well as compatibility in the above two tasks. The second stage targets the 3D lane task. It uses previously estimated pose to generate top-view images containing distance-invariant lane appearances for predicting accurate 3D lanes. Experiments demonstrate that, without ground truth camera pose, our method outperforms the state-of-the-art perfect-camera-pose-based methods and has the fewest parameters and computations. Codes are available at https://github.com/liuruijin17/CLGo.
['Zejian yuan', 'Zhiliang Xiong', 'Tie Liu', 'Dapeng Chen', 'Ruijin Liu']
2021-12-31
null
null
null
null
['3d-lane-detection']
['computer-vision']
[-2.28972808e-01 -1.09327145e-01 -3.59378219e-01 -4.95506495e-01 -8.08850229e-01 -6.14342093e-01 4.84264612e-01 -7.44693756e-01 -1.86295241e-01 1.69185400e-01 -1.28028482e-01 -5.22181988e-01 5.34565628e-01 -5.02711296e-01 -9.69896197e-01 -7.28066146e-01 5.36007643e-01 5.36568701e-01 5.90661883e-01 -2.00647444e-01 3.36426795e-01 7.37371206e-01 -1.39813542e+00 -1.58503488e-01 8.74109387e-01 7.10308969e-01 4.64805037e-01 6.79257751e-01 -5.66612966e-02 3.92279625e-01 2.84519717e-02 -4.02101964e-01 5.80206096e-01 1.86314464e-01 5.42187840e-02 6.29046500e-01 9.03228641e-01 -8.16076934e-01 -3.87407333e-01 1.11695218e+00 5.31386444e-03 -1.93644807e-01 4.65497881e-01 -1.52421832e+00 -1.48559501e-02 -3.85495424e-01 -8.47277224e-01 -3.68409514e-01 4.70724285e-01 4.43315804e-01 7.30993450e-01 -1.14731932e+00 5.09940386e-01 1.32930887e+00 8.45773757e-01 2.61105925e-01 -9.87878621e-01 -7.07301497e-01 2.58567452e-01 6.83120936e-02 -1.27025020e+00 -5.59194863e-01 1.17711961e+00 -5.98510981e-01 5.48082411e-01 -9.57401693e-02 4.80371147e-01 1.02834141e+00 2.70633459e-01 7.26731539e-01 1.03710711e+00 -6.59462363e-02 -3.17707211e-01 2.50748903e-01 -1.42856434e-01 8.88948560e-01 4.11840230e-01 2.35603273e-01 -1.14900947e-01 3.51440996e-01 8.68864059e-01 2.50084817e-01 -1.66786581e-01 -9.69442308e-01 -1.21102238e+00 7.16188252e-01 3.63847405e-01 -3.87142092e-01 9.56717040e-03 -1.45796522e-01 6.02809526e-02 -1.57840669e-01 3.77397686e-01 1.79762185e-01 -3.51670176e-01 1.68993294e-01 -4.59430993e-01 3.42101365e-01 4.01669890e-01 1.39144599e+00 1.18427444e+00 1.31235234e-02 4.83533978e-01 4.95996237e-01 5.76501787e-01 1.17316711e+00 -1.39323562e-01 -1.24126256e+00 9.19415057e-01 6.69060409e-01 2.26945311e-01 -1.10794938e+00 -5.87543786e-01 -2.09634468e-01 -6.89310133e-01 3.14961553e-01 6.45111382e-01 -1.30306676e-01 -8.53455484e-01 1.27991283e+00 5.42106450e-01 1.42511979e-01 1.02141919e-02 1.17754781e+00 5.31179786e-01 7.96615481e-01 -5.01660347e-01 3.12914513e-02 1.14280546e+00 -1.17625904e+00 -4.72090155e-01 -1.10241592e+00 6.80423498e-01 -9.29692745e-01 7.12720215e-01 2.02518404e-01 -6.35680914e-01 -7.23251343e-01 -1.02864933e+00 -3.60986233e-01 -2.51654357e-01 5.48536658e-01 4.19768691e-01 4.76710290e-01 -8.52907360e-01 -3.75447780e-01 -5.65530837e-01 -2.18466491e-01 -4.93431576e-02 -2.31456682e-02 -6.60926580e-01 -4.24043894e-01 -9.51174796e-01 1.22119009e+00 2.13224590e-01 3.27846199e-01 -5.80417514e-01 -5.18558502e-01 -1.30505228e+00 -4.61499870e-01 8.57538462e-01 -5.90287626e-01 1.12622213e+00 -6.75896585e-01 -1.38507533e+00 1.02249026e+00 -4.67240125e-01 -9.56216976e-02 7.32854605e-01 -2.10791379e-01 -1.67088076e-01 7.09798858e-02 2.23584726e-01 7.49896646e-01 8.57217848e-01 -1.71653509e+00 -8.86227548e-01 -5.12831628e-01 -2.00334117e-01 4.06612098e-01 4.18949425e-01 -5.33454716e-01 -1.02333355e+00 1.30405854e-02 5.43114364e-01 -1.28833163e+00 -3.80192131e-01 1.71980634e-01 -4.25368577e-01 1.22346364e-01 1.07655108e+00 -7.42535114e-01 4.92881000e-01 -2.12545896e+00 -8.58982056e-02 1.64994016e-01 2.62526035e-01 1.84539273e-01 -5.92859462e-02 1.53988525e-01 1.49333760e-01 -3.07246834e-01 2.37090379e-01 -5.75417459e-01 -1.54786780e-01 2.07795501e-01 -3.18671346e-01 6.20844781e-01 2.11671621e-01 7.73909450e-01 -6.97607815e-01 -3.84268492e-01 7.00549603e-01 3.50179732e-01 -4.07172948e-01 3.34908366e-01 -4.38393233e-03 5.73802888e-01 -4.80767727e-01 5.63707530e-01 1.18092215e+00 7.65587017e-02 -1.91887971e-02 -5.27924299e-01 -4.45424676e-01 -7.61687942e-03 -1.11323607e+00 1.30737245e+00 -5.73297679e-01 7.16423810e-01 1.38675883e-01 -7.33109415e-01 1.13364673e+00 -1.74045533e-01 3.33273828e-01 -6.98998988e-01 2.50748456e-01 3.91448855e-01 -3.04639041e-01 -3.97473961e-01 4.76155400e-01 1.82564750e-01 -4.68538046e-01 -1.64103329e-01 -3.41574728e-01 -6.75387144e-01 3.25991213e-02 4.11097407e-02 5.12623906e-01 2.35065743e-01 3.16891193e-01 2.75957972e-01 5.63402236e-01 1.54796258e-01 8.21464360e-01 3.64631534e-01 -1.52958378e-01 6.51067674e-01 5.25776982e-01 -5.56650698e-01 -1.49608529e+00 -1.00080562e+00 2.01569274e-01 3.22110206e-01 6.52563572e-01 4.77368459e-02 -6.49591088e-01 -5.84951937e-01 3.14115018e-01 6.52606845e-01 -2.55637437e-01 -1.06669046e-01 -7.00595438e-01 -1.31415144e-01 -1.46332886e-02 4.63325620e-01 5.12523055e-01 -1.39038369e-01 -5.46442747e-01 -1.37814224e-01 -2.37922087e-01 -1.75152183e+00 -8.66189480e-01 -8.44761655e-02 -7.57109880e-01 -1.21345806e+00 -5.33181787e-01 -6.17187381e-01 7.99175322e-01 1.13456964e+00 7.93280959e-01 -1.94947109e-01 1.72398090e-01 4.16679122e-02 -1.46521255e-01 -3.35036665e-01 -4.56018269e-01 -2.38343894e-01 1.29129574e-01 1.81918263e-01 2.45360434e-01 -1.92632690e-01 -4.17515934e-01 7.86598802e-01 -2.29463935e-01 7.47809172e-01 1.14949620e+00 5.68421125e-01 5.95537305e-01 -1.72658652e-01 -8.89585912e-02 -7.27437437e-01 -3.40617508e-01 -1.05318017e-01 -1.24208534e+00 -2.12303489e-01 -4.20141459e-01 -8.86212811e-02 7.79545307e-01 -1.90900311e-01 -1.17187214e+00 8.38899612e-01 -4.93134521e-02 -9.51084435e-01 -2.61687160e-01 1.35154441e-01 -5.99516630e-01 -1.30292013e-01 2.11911649e-01 2.40155742e-01 4.69705731e-01 -1.74387679e-01 3.49323720e-01 5.97402453e-01 6.11721516e-01 -2.51598079e-02 1.24888682e+00 6.45792902e-01 2.97789484e-01 -9.06360090e-01 -1.03451180e+00 -8.46738696e-01 -1.00762141e+00 -6.27990782e-01 8.86684000e-01 -1.26464677e+00 -4.84655857e-01 7.55271316e-01 -1.25203228e+00 -1.66143522e-01 4.40637290e-01 4.55890357e-01 -6.30203247e-01 4.62347627e-01 -1.53583542e-01 -6.29513979e-01 1.66854724e-01 -1.48196304e+00 1.54673600e+00 1.61747113e-01 4.16746140e-01 -8.96830380e-01 -2.45066255e-01 8.41701090e-01 -1.02076881e-01 1.29786104e-01 4.71753359e-01 -7.70991519e-02 -1.15449297e+00 -4.62798327e-01 -2.99270183e-01 2.62336165e-01 -1.73560932e-01 6.80317208e-02 -8.52895617e-01 -1.27788603e-01 -1.75986364e-01 5.37599018e-03 7.55572498e-01 3.78048569e-01 5.69403708e-01 -2.41229907e-02 -5.16139150e-01 8.32908988e-01 1.29071569e+00 1.09907754e-01 5.49423516e-01 3.83094221e-01 1.21275151e+00 6.10342801e-01 1.05604756e+00 2.48617213e-02 9.48141694e-01 1.05534244e+00 5.51753640e-01 -2.29177356e-01 -9.47378501e-02 -6.95300341e-01 4.87841636e-01 6.24619365e-01 1.75186783e-01 -9.79093239e-02 -1.01232743e+00 2.46076539e-01 -1.91570139e+00 -8.30526769e-01 -6.24802172e-01 2.18860149e+00 2.66487122e-01 3.90508503e-01 1.39182970e-01 -2.55865425e-01 7.03191459e-01 2.08670527e-01 -6.54444396e-01 -9.51005071e-02 1.47554101e-02 -1.10740566e+00 1.08605123e+00 7.93090284e-01 -1.09178269e+00 1.16636539e+00 4.79934406e+00 4.87709820e-01 -1.20829642e+00 -1.51635632e-01 5.61938763e-01 4.33614045e-01 -2.33696043e-01 2.00357065e-01 -1.45237577e+00 4.13462132e-01 5.18199682e-01 1.87059343e-01 1.74137056e-01 1.05962980e+00 5.31197786e-01 -3.40203315e-01 -1.01088035e+00 1.12911677e+00 4.02749479e-01 -1.03856516e+00 -8.36796686e-02 3.36653471e-01 7.73209453e-01 1.07509032e-01 -3.51116620e-02 1.28870636e-01 2.04264205e-02 -4.55676526e-01 9.68094409e-01 5.37230372e-01 7.11728036e-01 -6.57545626e-01 8.83830547e-01 8.77712369e-01 -1.34566557e+00 -8.79975706e-02 -3.28285366e-01 2.15949222e-01 5.28516769e-01 6.21682346e-01 -1.05600274e+00 4.71474230e-01 3.00292313e-01 9.50340092e-01 -6.88490272e-01 9.84813213e-01 -3.43210220e-01 2.74113063e-02 -3.07322174e-01 2.06931546e-01 3.09301704e-01 -5.62658250e-01 5.38748801e-01 8.77786219e-01 5.10269403e-01 -3.64559237e-03 6.02744818e-01 6.36097968e-01 2.43938446e-01 -4.31506157e-01 -1.09535027e+00 6.78099155e-01 4.95454669e-01 1.28500450e+00 -4.58915502e-01 7.42558949e-03 -7.18561292e-01 6.83086932e-01 5.58488257e-02 2.77625561e-01 -1.03500986e+00 7.47981993e-03 6.36982441e-01 3.31140190e-01 2.05390647e-01 -6.60632551e-01 -2.16545448e-01 -1.14915216e+00 2.43519545e-01 -6.53798461e-01 -2.54957110e-01 -1.13717675e+00 -8.59986186e-01 1.68432310e-01 1.08972326e-01 -1.79719150e+00 -3.13083112e-01 -9.72209632e-01 -3.70119512e-01 7.96321869e-01 -1.72493291e+00 -1.58387363e+00 -4.99739021e-01 2.34892175e-01 7.80337632e-01 9.76674631e-02 6.84113428e-02 1.71487704e-01 -4.48225826e-01 4.73118842e-01 3.38106113e-03 1.09048411e-01 7.81667352e-01 -9.12076116e-01 5.07684529e-01 1.08749032e+00 -2.44151145e-01 9.24963951e-02 6.88948333e-01 -5.92245042e-01 -1.92079532e+00 -1.30493224e+00 8.77768815e-01 -8.02046597e-01 4.35770065e-01 -5.74710250e-01 -6.43016100e-01 6.87035620e-01 -4.30827320e-01 9.11542326e-02 -1.51746556e-01 -3.64680529e-01 -3.30191672e-01 -4.40601259e-01 -6.57098889e-01 6.63852572e-01 1.00787997e+00 -4.18268919e-01 -2.18689993e-01 1.73313096e-01 4.22995389e-01 -8.29639137e-01 -3.38700712e-01 2.91841179e-01 6.21228635e-01 -1.05586326e+00 1.03393912e+00 2.43105665e-01 8.66537318e-02 -7.14584708e-01 -1.74528323e-02 -1.25863969e+00 -4.86116260e-02 -3.44913453e-01 2.47255132e-01 9.09647226e-01 4.46487755e-01 -6.72650516e-01 8.58366668e-01 4.96427804e-01 -5.63285232e-01 -4.32460457e-01 -7.04302847e-01 -7.55746365e-01 -2.56105721e-01 -5.16002774e-01 2.36664236e-01 5.26763916e-01 -5.49850941e-01 6.46574020e-01 -5.83503246e-01 7.43069708e-01 8.45510423e-01 2.54908174e-01 1.52545929e+00 -1.18951750e+00 2.01764330e-01 -2.64504641e-01 -5.64605355e-01 -1.71645534e+00 4.48881537e-01 -5.02066255e-01 4.35194403e-01 -1.36354184e+00 1.22212663e-01 -2.71084964e-01 5.92733264e-01 1.14906877e-01 -1.89905554e-01 1.51494499e-02 2.54296422e-01 2.84683734e-01 -5.49303412e-01 4.04847562e-01 1.46926296e+00 2.62317825e-02 -1.60402656e-02 2.57450104e-01 -3.69162291e-01 9.75137949e-01 6.78175509e-01 -1.71533797e-03 -4.37351257e-01 -5.74580252e-01 8.29900503e-02 3.84406209e-01 4.97631997e-01 -8.39191854e-01 4.05906737e-01 -3.29547018e-01 3.71366948e-01 -1.18025112e+00 7.06920326e-01 -1.09956563e+00 1.67947978e-01 2.63408750e-01 2.76810437e-01 1.76105857e-01 -3.70206982e-02 5.54112017e-01 -5.24808094e-02 4.57902923e-02 8.32918108e-01 -5.40351048e-02 -1.21278763e+00 4.15383577e-01 -2.06956267e-01 -2.94578344e-01 1.26457942e+00 -7.42096722e-01 -3.76036674e-01 -5.21341443e-01 -8.46015960e-02 6.43269598e-01 9.60332513e-01 6.47378862e-01 8.46867740e-01 -1.21327078e+00 -5.27954638e-01 5.84887564e-01 4.01427537e-01 4.53873158e-01 2.91761220e-01 1.04009891e+00 -7.43210971e-01 7.30655015e-01 -4.52573448e-02 -1.13496518e+00 -1.19543672e+00 6.18913114e-01 3.95032257e-01 1.48196191e-01 -5.45277655e-01 2.38878295e-01 4.39043313e-01 -7.29251623e-01 -9.43692401e-02 -3.51337373e-01 -4.28392813e-02 -1.70368806e-01 3.38858455e-01 1.91027775e-01 4.81591001e-02 -1.25348508e+00 -2.29534253e-01 1.20948505e+00 -7.03918710e-02 2.04390123e-01 9.65544999e-01 -5.11357486e-01 3.43252659e-01 3.34341556e-01 1.32753456e+00 1.18185185e-01 -1.95585752e+00 7.88950697e-02 -2.48768777e-01 -8.04758012e-01 1.76423833e-01 -1.77717462e-01 -1.03042579e+00 9.91891265e-01 4.07961160e-01 -3.05708945e-01 7.84436047e-01 -6.61463737e-02 7.58572698e-01 4.45231736e-01 5.69362760e-01 -9.28675473e-01 5.89116998e-02 7.91788876e-01 6.80874646e-01 -1.66919827e+00 -5.92278466e-02 -8.75793219e-01 -7.93623388e-01 1.25119638e+00 1.00173259e+00 -3.43724824e-02 3.28598708e-01 1.97606206e-01 3.70077074e-01 1.18212603e-01 -4.31193441e-01 -1.28490299e-01 3.56269777e-01 6.23665035e-01 -1.68850660e-01 3.84819508e-02 3.77325028e-01 2.04446297e-02 -6.09974712e-02 -4.75800127e-01 6.84583962e-01 5.27415097e-01 -5.61129928e-01 -8.25922132e-01 -6.54621899e-01 1.01211779e-01 4.41634506e-01 3.90810132e-01 -3.21405411e-01 1.00643885e+00 9.58206430e-02 7.21385777e-01 7.21067116e-02 -5.07612050e-01 6.08092964e-01 -1.54031664e-01 2.16490239e-01 -4.48625922e-01 5.52249432e-01 2.11455345e-01 2.31480718e-01 -7.27434695e-01 -1.15317523e-01 -8.42382431e-01 -9.36631382e-01 -3.97280216e-01 -3.91311318e-01 -2.32282802e-01 7.53973842e-01 8.28929484e-01 3.47450048e-01 -1.02994137e-01 1.14525878e+00 -1.50621891e+00 -4.74699348e-01 -4.76125151e-01 -3.58091801e-01 2.34843969e-01 6.18267417e-01 -7.92530358e-01 -4.92321670e-01 4.96459007e-02]
[8.005998611450195, -1.7175087928771973]
d9f9bf7e-25b4-4430-a9df-ee447cd89ca4
handling-noisy-labels-for-robustly-learning
1903.12008
null
http://arxiv.org/abs/1903.12008v1
http://arxiv.org/pdf/1903.12008v1.pdf
Handling Noisy Labels for Robustly Learning from Self-Training Data for Low-Resource Sequence Labeling
In this paper, we address the problem of effectively self-training neural networks in a low-resource setting. Self-training is frequently used to automatically increase the amount of training data. However, in a low-resource scenario, it is less effective due to unreliable annotations created using self-labeling of unlabeled data. We propose to combine self-training with noise handling on the self-labeled data. Directly estimating noise on the combined clean training set and self-labeled data can lead to corruption of the clean data and hence, performs worse. Thus, we propose the Clean and Noisy Label Neural Network which trains on clean and noisy self-labeled data simultaneously by explicitly modelling clean and noisy labels separately. In our experiments on Chunking and NER, this approach performs more robustly than the baselines. Complementary to this explicit approach, noise can also be handled implicitly with the help of an auxiliary learning task. To such a complementary approach, our method is more beneficial than other baseline methods and together provides the best performance overall.
['Michael A. Hedderich', 'Mittul Singh', 'Debjit Paul', 'Dietrich Klakow']
2019-03-28
handling-noisy-labels-for-robustly-learning-1
https://aclanthology.org/N19-3005
https://aclanthology.org/N19-3005.pdf
naacl-2019-6
['auxiliary-learning']
['methodology']
[ 4.45143729e-02 4.41373438e-01 6.42286465e-02 -4.95069802e-01 -1.24056149e+00 -5.81416070e-01 5.18900871e-01 1.02371559e-01 -8.69340241e-01 1.00126398e+00 4.43833590e-01 -2.14847967e-01 5.89524329e-01 -5.68352997e-01 -8.19130480e-01 -5.89752138e-01 4.68059033e-01 2.82961518e-01 -2.12753154e-02 7.73967728e-02 -3.33253175e-01 6.42135646e-03 -1.58866072e+00 2.07824498e-01 1.17182446e+00 6.54907942e-01 2.08095685e-01 4.41025734e-01 -3.48897696e-01 1.07298434e+00 -7.51951814e-01 -3.20841372e-01 3.15453917e-01 -6.25226498e-01 -9.28761423e-01 -4.32719477e-02 7.01809943e-01 -2.25508422e-01 6.77123591e-02 1.19280112e+00 6.66984737e-01 3.32454622e-01 3.24225634e-01 -6.58761322e-01 -3.48939836e-01 9.67558682e-01 -2.24554852e-01 5.54947257e-02 -1.49915710e-01 1.01834219e-02 7.22294211e-01 -8.97940338e-01 6.25835717e-01 1.01774681e+00 9.37258244e-01 7.82261968e-01 -1.46090078e+00 -6.68522954e-01 8.94262716e-02 -7.80241042e-02 -1.11465621e+00 -1.01957834e+00 5.74193299e-01 -1.06874414e-01 8.36812258e-01 6.13282919e-02 2.52898447e-02 1.27682865e+00 -5.52779436e-01 8.72320831e-01 1.30405807e+00 -7.31983781e-01 3.52069616e-01 2.93949097e-01 4.50074285e-01 3.25024813e-01 2.74738044e-01 1.04408562e-01 -3.78155977e-01 4.20603417e-02 3.01765472e-01 -2.20746860e-01 -1.31829813e-01 -7.66894370e-02 -9.70003366e-01 5.67488372e-01 3.08637589e-01 5.69927096e-01 -4.05244619e-01 6.61049709e-02 5.57687998e-01 3.22349697e-01 1.04083383e+00 5.81030250e-01 -7.08002687e-01 -2.75463372e-01 -1.40985429e+00 -8.34607482e-02 1.00917065e+00 8.71585965e-01 8.45920146e-01 3.22349995e-01 -2.49239087e-01 1.03646195e+00 -1.31347805e-01 4.33878124e-01 4.38383758e-01 -1.03871047e+00 5.84905624e-01 1.53979972e-01 2.50972271e-01 -3.59874696e-01 -5.37834108e-01 -6.69624567e-01 -1.04442024e+00 9.64532271e-02 5.06408632e-01 -7.49634326e-01 -1.38785219e+00 2.01382113e+00 2.92652786e-01 4.45318818e-01 2.55766839e-01 6.65652812e-01 1.04269707e+00 3.59370261e-01 3.96426916e-01 -3.42811227e-01 9.57573593e-01 -1.28887045e+00 -1.09241211e+00 -5.56598961e-01 1.01447618e+00 -6.54032648e-01 9.00661051e-01 2.94912100e-01 -1.01381981e+00 -5.55181503e-01 -7.85539031e-01 -3.34329568e-02 -3.99874955e-01 2.49548152e-01 2.08825424e-01 6.58389568e-01 -1.11909759e+00 8.98581386e-01 -9.22541261e-01 -2.50704110e-01 5.66767156e-01 1.59784421e-01 -5.98209262e-01 -2.59406358e-01 -1.14339089e+00 1.21403587e+00 5.68245351e-01 6.00608774e-02 -4.61270034e-01 -5.85691750e-01 -1.18775094e+00 1.75249949e-01 8.23880374e-01 -4.16072190e-01 1.55357289e+00 -1.03667462e+00 -1.32848263e+00 6.31015658e-01 -2.87560999e-01 -4.88725454e-01 4.05490607e-01 -5.96886992e-01 -1.85149297e-01 -1.93270057e-01 1.90138951e-01 5.81519783e-01 7.23070383e-01 -1.58004117e+00 -4.33811963e-01 -2.38861606e-01 -2.96333522e-01 2.14731053e-01 -3.70604068e-01 -1.25244454e-01 -3.71792018e-01 -7.68851519e-01 -3.96249108e-02 -8.38657260e-01 -2.97016799e-01 -8.87195587e-01 -5.65756619e-01 -9.17488262e-02 6.46855772e-01 -8.08175504e-01 1.19860125e+00 -2.18728590e+00 -5.74275181e-02 -2.38038711e-02 1.58804044e-01 9.37784612e-01 -3.49751949e-01 1.88293055e-01 -2.32635751e-01 4.90882814e-01 -3.35536987e-01 -9.37818289e-01 -2.02083230e-01 5.20484209e-01 6.12446666e-02 2.50548899e-01 4.42602545e-01 9.36391413e-01 -1.00652456e+00 -4.93098438e-01 5.41723042e-04 5.10629058e-01 -2.65769213e-01 3.53891134e-01 -6.16116598e-02 5.91775596e-01 6.98181316e-02 3.49161804e-01 7.57255614e-01 1.16345380e-02 1.78582802e-01 -1.31517611e-02 -1.90189406e-02 5.63132763e-01 -1.38819027e+00 1.64447677e+00 -8.38921070e-01 4.92572308e-01 4.73808318e-01 -1.00944924e+00 5.60743749e-01 5.92475951e-01 6.03358187e-02 -6.50103927e-01 9.55316275e-02 1.23061910e-01 -2.25593597e-01 -6.12949371e-01 4.87519354e-01 -3.93623322e-01 2.28348188e-02 6.66611671e-01 4.83738720e-01 -6.01879247e-02 2.61339009e-01 2.38097981e-01 1.28522372e+00 2.60373652e-01 2.85953879e-01 -1.23633280e-01 1.35962605e-01 -1.25888303e-01 7.69672811e-01 1.08696747e+00 -1.87871724e-01 5.56046844e-01 9.95291024e-02 -7.77877197e-02 -1.02283239e+00 -4.48657513e-01 -8.07966664e-02 1.15378845e+00 -2.01543450e-01 -4.72245634e-01 -1.12256539e+00 -1.16021407e+00 -2.30566964e-01 8.43316495e-01 -5.06368101e-01 -5.92181049e-02 -7.27364719e-01 -5.32562912e-01 7.34414756e-01 7.01539218e-01 5.33004105e-01 -1.14579761e+00 -2.27443472e-01 3.17689657e-01 -4.19048339e-01 -1.16608882e+00 -2.75876313e-01 1.07271612e+00 -8.71724367e-01 -7.82283068e-01 -4.81057435e-01 -7.49793053e-01 6.75752461e-01 3.01614106e-01 1.34051442e+00 3.75388473e-01 9.84440669e-02 -1.28821850e-01 -6.66324973e-01 -4.64418262e-01 -7.18282044e-01 3.78986448e-01 1.08562872e-01 -3.00665766e-01 4.20912683e-01 -4.81095612e-01 -4.62882482e-02 -3.19076069e-02 -1.04021895e+00 -1.46422878e-01 6.52012050e-01 1.25251198e+00 2.90232301e-01 2.32336238e-01 7.41786778e-01 -1.50461113e+00 4.35141504e-01 -5.82225800e-01 -2.52026856e-01 8.27723444e-02 -5.09105504e-01 4.28714842e-01 9.20089960e-01 -3.30787390e-01 -1.46834528e+00 2.07537979e-01 -4.24123704e-01 -1.62487954e-01 -8.39219868e-01 5.19293845e-01 -3.76382530e-01 1.54320449e-01 9.56636608e-01 -3.18615615e-01 -2.36988083e-01 -1.00584984e+00 6.53416157e-01 9.58588481e-01 6.78357065e-01 -2.69153118e-01 7.84757137e-01 1.83792040e-01 -4.19903964e-01 -6.25290751e-01 -1.43314087e+00 -6.72345281e-01 -8.72580528e-01 2.59526879e-01 5.52087843e-01 -1.10704482e+00 9.50569212e-02 5.32087743e-01 -1.06872606e+00 -6.26224399e-01 -6.00009859e-01 4.01453376e-01 -1.60712451e-01 6.01120114e-01 -7.57928491e-01 -1.07252491e+00 -3.73195887e-01 -8.57887328e-01 1.08140182e+00 9.94769409e-02 -1.68782696e-01 -9.99278963e-01 1.32220592e-02 2.93251544e-01 5.31633258e-01 -1.75244361e-02 4.14441615e-01 -1.12047577e+00 -6.35233968e-02 -2.24321827e-01 -2.42829874e-01 9.32827294e-01 2.28760600e-01 -4.83071923e-01 -1.51228952e+00 -1.66904658e-01 3.04773837e-01 -7.72027910e-01 1.32864451e+00 1.33015975e-01 8.78966272e-01 -3.33557278e-01 -7.29582831e-02 4.01937246e-01 1.34328043e+00 -2.07594186e-01 4.72140342e-01 2.05208570e-01 8.68461549e-01 7.10032225e-01 5.94669700e-01 1.01881884e-01 3.43914360e-01 4.43002313e-01 2.29312763e-01 -4.32548195e-01 -4.26503688e-01 -3.98831815e-02 1.68097347e-01 9.52243507e-01 1.47752449e-01 -2.68314153e-01 -8.72597158e-01 5.44854701e-01 -1.86325443e+00 -1.04426670e+00 -2.40106910e-01 2.21191645e+00 1.34494543e+00 -2.64827777e-02 -2.70041581e-02 2.22863361e-01 7.44285345e-01 -6.85004145e-02 -3.11547577e-01 -1.37269109e-01 -2.28708774e-01 3.25483680e-01 7.26224482e-01 5.11995912e-01 -1.57593358e+00 1.07982635e+00 6.68387175e+00 9.06000197e-01 -8.24119449e-01 5.68727374e-01 6.19171739e-01 -1.64436132e-01 -1.10242367e-01 -2.01565072e-01 -8.28202903e-01 5.95318079e-01 1.26337266e+00 5.07477462e-01 4.51540291e-01 7.95393467e-01 9.97423679e-02 -4.74235088e-01 -1.09081829e+00 7.84450054e-01 1.04217879e-01 -9.44554567e-01 -6.41279578e-01 -1.43555164e-01 9.87040102e-01 2.52778828e-01 -3.46398860e-01 6.29632592e-01 8.24281216e-01 -9.24430013e-01 7.35621691e-01 2.34837398e-01 7.80743778e-01 -6.18290246e-01 1.39995122e+00 8.56135666e-01 -6.69996619e-01 -3.03099514e-03 -2.32421175e-01 -1.25705034e-01 1.45962432e-01 1.05381453e+00 -5.33826590e-01 5.57661355e-01 6.40137613e-01 4.16069150e-01 -5.97254097e-01 1.00164008e+00 -6.19573236e-01 1.07313645e+00 -5.63791275e-01 4.47442055e-01 1.42488137e-01 1.01725325e-01 2.69571811e-01 1.57440543e+00 -6.40706047e-02 -7.97243789e-03 2.38777235e-01 6.03401542e-01 -3.93217802e-01 7.51860514e-02 -5.36355138e-01 1.77502483e-01 6.13937259e-01 1.34849918e+00 -5.79136491e-01 -7.35439718e-01 -4.29185152e-01 9.60240304e-01 9.31729198e-01 4.21765089e-01 -3.83802325e-01 -2.98950464e-01 2.61778295e-01 -2.39281401e-01 4.42845196e-01 -3.59837562e-02 -5.69648921e-01 -1.44218314e+00 1.14706442e-01 -8.41657817e-01 1.92583576e-01 -3.86906534e-01 -1.39548612e+00 7.16780066e-01 -2.44811878e-01 -8.45102429e-01 -4.85407501e-01 -1.19543336e-01 -5.66479445e-01 1.02130520e+00 -1.83695841e+00 -9.11525846e-01 -2.85813272e-01 1.24840431e-01 5.36530256e-01 2.34327585e-01 1.03661168e+00 4.35305923e-01 -6.46997988e-01 6.77626729e-01 4.43633378e-01 2.49788329e-01 1.18122470e+00 -1.61557472e+00 5.66614568e-01 1.22450054e+00 3.88836980e-01 5.49602389e-01 7.19029546e-01 -9.18503821e-01 -6.11360073e-01 -1.33039427e+00 9.86280918e-01 -5.21613419e-01 3.52619886e-01 -4.84903038e-01 -1.26825511e+00 5.42953551e-01 3.36847961e-01 1.02739103e-01 8.50767553e-01 4.35241759e-01 -5.46702564e-01 1.92848310e-01 -1.08625829e+00 2.78422385e-01 9.52734649e-01 -5.07053614e-01 -6.90362692e-01 3.68840218e-01 8.75082254e-01 -4.59080309e-01 -6.91031098e-01 2.76082158e-01 -1.78330213e-01 -8.37922573e-01 5.40991485e-01 -5.31475246e-01 1.64615005e-01 -1.49650037e-01 1.50463879e-01 -1.86897922e+00 -2.85672307e-01 -4.45767105e-01 -1.16567805e-01 1.63097668e+00 5.43974280e-01 -2.28039905e-01 8.63411486e-01 7.88776696e-01 -2.33466357e-01 -2.51781702e-01 -9.21396971e-01 -9.66724098e-01 1.82901874e-01 -4.60832477e-01 3.69463921e-01 1.11315072e+00 3.01653054e-02 4.64113742e-01 -6.59338474e-01 -1.25712126e-01 6.55105352e-01 -3.63366634e-01 6.38921201e-01 -1.24705577e+00 -1.16185270e-01 2.04655714e-02 1.66086167e-01 -6.75414324e-01 5.54699004e-01 -8.23742211e-01 8.16861331e-01 -1.45330870e+00 2.58215927e-02 -7.25724518e-01 -1.98937654e-01 9.39305246e-01 -7.85681903e-01 5.32436848e-01 2.21741647e-01 2.10293263e-01 -8.32971811e-01 2.81258732e-01 6.63340509e-01 1.30278200e-01 -3.21855456e-01 -1.77588716e-01 -8.97179067e-01 6.82021141e-01 9.11156118e-01 -7.27471769e-01 -5.22042252e-03 -6.33699059e-01 -5.99989817e-02 -3.18906605e-01 9.24869180e-02 -9.51020002e-01 2.70908684e-01 2.86724359e-01 2.07806021e-01 -1.82700351e-01 1.93503216e-01 -9.00765777e-01 -9.87121239e-02 -1.50760990e-02 -3.91036779e-01 -5.17535388e-01 1.91605240e-01 4.80951786e-01 -3.14293712e-01 -6.91620886e-01 8.81381273e-01 -4.43493217e-01 -4.68003422e-01 -2.04782590e-01 -3.96167874e-01 3.59515309e-01 5.57098985e-01 7.77789801e-02 -3.25812101e-01 -5.31801760e-01 -9.75449741e-01 1.52326569e-01 5.45154691e-01 1.84296995e-01 -3.80747877e-02 -1.16962492e+00 -5.59176207e-01 1.71584129e-01 -8.10710415e-02 4.09633130e-01 2.37320721e-01 5.96290827e-01 4.91290689e-02 2.13373944e-01 2.32655033e-01 -2.35777363e-01 -1.20418715e+00 6.68694615e-01 7.77900368e-02 -5.60273767e-01 -4.54168171e-01 8.24870348e-01 -1.85962811e-01 -7.20068336e-01 6.77251995e-01 -1.86115876e-01 -3.91644210e-01 2.01518029e-01 6.44552529e-01 3.44343901e-01 5.59879661e-01 -8.36317360e-01 -1.24690486e-02 -6.42690901e-03 -1.18731581e-01 -1.01412125e-01 1.50157559e+00 -2.96719819e-01 -3.94558460e-02 3.80763292e-01 1.07980609e+00 1.46947548e-01 -1.11478984e+00 -6.49903834e-01 2.59312809e-01 -1.61429435e-01 3.24766129e-01 -1.00288260e+00 -1.05047834e+00 7.92865217e-01 2.23522499e-01 2.34699488e-01 1.00126851e+00 -9.10666138e-02 7.18618095e-01 5.21650076e-01 5.57462983e-02 -1.42910957e+00 -1.72455668e-01 6.19834721e-01 2.71617085e-01 -1.84703994e+00 -2.25999177e-01 -5.44364452e-01 -7.29415596e-01 7.35921144e-01 7.39397824e-01 1.32055312e-01 4.74831074e-01 6.40437722e-01 5.85484028e-01 1.33549944e-01 -5.81836998e-01 -6.27876461e-01 1.31524578e-01 7.78061926e-01 4.73757625e-01 -1.09042346e-01 -4.52872850e-02 8.29682350e-01 8.49324465e-03 5.58600463e-02 5.55375457e-01 1.08594179e+00 -4.55505311e-01 -1.37087095e+00 -6.11056685e-01 6.15330637e-01 -7.59855390e-01 -5.71413577e-01 -2.50983238e-01 4.40003812e-01 3.75236988e-01 1.36434150e+00 -1.06178634e-01 -2.49821588e-01 3.85975569e-01 5.55484712e-01 -5.12056202e-02 -1.11854160e+00 -9.42534208e-01 2.49845862e-01 4.35988635e-01 -4.51668292e-01 -7.42991805e-01 -4.71511841e-01 -9.79843199e-01 9.80302393e-02 -7.83845484e-01 4.01462764e-01 7.70480275e-01 1.19424927e+00 4.13320929e-01 4.40156102e-01 3.57484579e-01 -1.02305913e+00 -7.08177984e-01 -1.33480656e+00 -5.42651415e-01 7.07520366e-01 4.55136925e-01 -4.60722834e-01 -5.44159532e-01 1.21448867e-01]
[9.431920051574707, 4.05098819732666]
520aa276-957b-4b78-96a4-80f87f86b6f6
seeking-common-but-distinguishing-difference-1
2111.09634
null
https://arxiv.org/abs/2111.09634v1
https://arxiv.org/pdf/2111.09634v1.pdf
Seeking Common but Distinguishing Difference, A Joint Aspect-based Sentiment Analysis Model
Aspect-based sentiment analysis (ABSA) task consists of three typical subtasks: aspect term extraction, opinion term extraction, and sentiment polarity classification. These three subtasks are usually performed jointly to save resources and reduce the error propagation in the pipeline. However, most of the existing joint models only focus on the benefits of encoder sharing between subtasks but ignore the difference. Therefore, we propose a joint ABSA model, which not only enjoys the benefits of encoder sharing but also focuses on the difference to improve the effectiveness of the model. In detail, we introduce a dual-encoder design, in which a pair encoder especially focuses on candidate aspect-opinion pair classification, and the original encoder keeps attention on sequence labeling. Empirical results show that our proposed model shows robustness and significantly outperforms the previous state-of-the-art on four benchmark datasets.
['Shu Jiang', 'Hai Zhao', 'Zuchao Li', 'Hongjiang Jing']
2021-11-18
seeking-common-but-distinguishing-difference
https://aclanthology.org/2021.emnlp-main.318
https://aclanthology.org/2021.emnlp-main.318.pdf
emnlp-2021-11
['term-extraction', 'aspect-based-sentiment-analysis']
['natural-language-processing', 'natural-language-processing']
[ 1.62842304e-01 -1.27733245e-01 -2.05391780e-01 -6.60909414e-01 -9.25276816e-01 -5.19186497e-01 4.82868284e-01 -3.06206550e-02 -2.22469732e-01 3.90250415e-01 4.29517090e-01 -3.56447071e-01 3.45699787e-01 -8.38541150e-01 -6.53900921e-01 -4.89692688e-01 2.64326572e-01 2.49606129e-02 1.71026677e-01 -1.69097036e-01 7.97141418e-02 -2.73460627e-01 -1.25822961e+00 5.35916865e-01 8.34892690e-01 1.26669109e+00 -2.07631394e-01 3.38756531e-01 -4.42544281e-01 1.00185549e+00 -6.14759445e-01 -1.07856476e+00 -3.04692145e-02 -2.94744492e-01 -7.02666163e-01 -1.36256898e-02 -2.34322563e-01 -4.06197488e-01 -1.36506811e-01 1.10011947e+00 5.59405565e-01 -2.71609753e-01 3.19705963e-01 -1.29997849e+00 -8.97469819e-01 8.19072306e-01 -8.37571084e-01 -1.80573016e-01 2.66739458e-01 -5.25920885e-04 1.66200006e+00 -7.46371806e-01 2.40947410e-01 9.56996977e-01 6.93151832e-01 1.74505085e-01 -5.96672475e-01 -6.36666298e-01 7.20001817e-01 1.32645190e-01 -1.01014960e+00 -3.44618827e-01 7.83043087e-01 -2.84517765e-01 1.36870968e+00 8.94469768e-02 9.93372917e-01 7.46759534e-01 2.41825685e-01 1.47899091e+00 9.04248416e-01 -9.01296083e-03 4.31472920e-02 6.21698610e-02 4.98483688e-01 7.89902449e-01 2.92231798e-01 -2.98153698e-01 -7.16544509e-01 7.92483520e-03 -7.44541781e-03 -5.59989251e-02 -1.07613884e-01 -1.58853903e-01 -1.09950590e+00 7.24795163e-01 1.22208282e-01 6.26933277e-02 -4.21515137e-01 1.27644313e-03 7.15123177e-01 2.48205826e-01 8.29026699e-01 2.29101226e-01 -8.79650533e-01 -3.14712912e-01 -6.87228203e-01 1.38405561e-01 9.15101945e-01 1.27509820e+00 7.42889583e-01 6.54099807e-02 -5.11993945e-01 6.96508825e-01 5.54620445e-01 4.79426801e-01 5.19471347e-01 -3.54763061e-01 5.28586268e-01 1.21011019e+00 -2.94036478e-01 -7.13165760e-01 -2.19722569e-01 -5.79536438e-01 -5.65199137e-01 -1.55942112e-01 -2.83564329e-01 -4.58922982e-01 -9.47104514e-01 1.49908125e+00 3.96048158e-01 -4.78508510e-02 3.18880565e-02 6.27260804e-01 1.03818083e+00 6.24375045e-01 -1.19535990e-01 -2.92600721e-01 1.85882318e+00 -1.72770834e+00 -1.01840329e+00 -4.94891942e-01 6.62961364e-01 -1.03677630e+00 8.10511112e-01 8.56605619e-02 -1.11764216e+00 -2.84736603e-01 -1.42505503e+00 -3.38959128e-01 -4.03314203e-01 3.60163033e-01 1.10683191e+00 6.62116289e-01 -7.71099150e-01 1.42199546e-01 -8.64650786e-01 1.95692867e-01 5.26724339e-01 4.02241647e-01 -1.44142732e-01 1.64360315e-01 -1.28823137e+00 4.16233093e-01 -2.22315565e-01 2.70089716e-01 -4.84600782e-01 -6.29499018e-01 -1.13604951e+00 1.99815571e-01 2.43434578e-01 -1.04580355e+00 1.59055173e+00 -1.04964387e+00 -1.82147789e+00 6.17347360e-01 -6.93836033e-01 -2.17027426e-01 8.46343115e-02 -5.55702209e-01 -4.30742621e-01 -3.34000617e-01 2.33439445e-01 4.19803977e-01 7.85386920e-01 -7.93411434e-01 -8.08146477e-01 -3.71840358e-01 6.65983438e-01 2.42691532e-01 -5.65561473e-01 2.32568607e-01 -8.98243368e-01 -7.45192230e-01 -3.27638626e-01 -8.25331330e-01 -2.46082887e-01 -3.85529369e-01 -5.40174246e-01 -3.22950184e-01 5.79179108e-01 -4.99909788e-01 1.52227235e+00 -2.15723252e+00 -8.38181823e-02 -1.57469109e-01 1.70615777e-01 1.71622291e-01 -1.94037378e-01 3.39661896e-01 -2.03612313e-01 1.14268258e-01 -3.41192514e-01 -6.47131205e-01 9.81356800e-02 -2.31735721e-01 -3.33712935e-01 1.38828382e-01 3.34320128e-01 1.22809029e+00 -8.53681147e-01 -4.50063825e-01 -2.96815991e-01 5.21851182e-01 -5.57296157e-01 3.22936028e-01 -2.06994832e-01 -3.82565632e-02 -7.12251902e-01 7.65199006e-01 7.94000745e-01 -4.48508114e-01 6.32129684e-02 -4.57709908e-01 -1.67325690e-01 9.43843603e-01 -9.70035017e-01 1.64192045e+00 -6.30517304e-01 4.83521432e-01 3.92478518e-02 -7.38170087e-01 6.12903595e-01 4.16965842e-01 5.17727852e-01 -6.79309309e-01 3.24404627e-01 2.21823782e-01 5.34848273e-02 -3.65851462e-01 6.67318344e-01 -7.55986273e-02 -9.00268033e-02 7.10381448e-01 8.21843296e-02 6.42222688e-02 3.88184845e-01 1.30260319e-01 9.20116842e-01 2.26932466e-01 4.59178120e-01 -5.25140204e-02 8.18227530e-01 -2.05399320e-01 9.32001412e-01 1.56727299e-01 -1.48549259e-01 4.91085708e-01 8.58040929e-01 -3.75165045e-01 -7.51272500e-01 -4.61004287e-01 1.87050626e-01 9.44393754e-01 3.42943192e-01 -8.99889469e-01 -5.67030132e-01 -1.19277966e+00 -3.50578964e-01 4.62161660e-01 -7.43421495e-01 -1.21316046e-01 -4.39853281e-01 -1.06534445e+00 2.30296552e-01 8.01422119e-01 6.62444472e-01 -9.49303389e-01 -1.61797032e-01 -2.99119391e-02 -2.26168916e-01 -1.08221400e+00 -7.50872791e-01 6.47243485e-02 -7.01725662e-01 -1.00701082e+00 -5.50660312e-01 -9.00484920e-01 5.30520082e-01 5.89845240e-01 1.28969514e+00 -7.38756508e-02 2.36754403e-01 5.37396967e-02 -5.45634508e-01 -7.28008568e-01 1.66334823e-01 4.26302940e-01 -5.04724264e-01 2.83076674e-01 7.58486390e-01 -5.30255437e-01 -8.14437687e-01 1.32072531e-02 -8.36318791e-01 2.05679297e-01 1.05958235e+00 7.69184709e-01 7.55138159e-01 4.45496291e-02 5.57754457e-01 -1.46182024e+00 7.12055624e-01 -3.14427614e-01 -4.81169403e-01 3.38257551e-01 -9.08808351e-01 6.08329102e-02 5.40014863e-01 5.05982004e-02 -1.27487862e+00 -2.06114445e-02 -4.42238241e-01 1.17906965e-02 3.29104185e-01 8.73684466e-01 -5.09517491e-01 3.10413808e-01 -2.87090927e-01 3.54350209e-01 -2.59169698e-01 -2.62089342e-01 2.69859761e-01 8.49799216e-01 -2.73073852e-01 -5.17601147e-02 4.42770392e-01 4.97976869e-01 -3.11003119e-01 -3.88792843e-01 -1.27219582e+00 -5.04748881e-01 -1.73983589e-01 9.16334242e-02 7.45034516e-01 -1.37117076e+00 -7.30264008e-01 7.87405133e-01 -1.31377256e+00 2.80684769e-01 -2.59361088e-01 4.32643354e-01 -2.92360038e-01 3.34146887e-01 -8.45943272e-01 -7.83201337e-01 -9.35469329e-01 -1.60229778e+00 1.36518502e+00 2.85776615e-01 -1.24231845e-01 -6.71730757e-01 1.97784558e-01 6.31525099e-01 4.05910462e-01 -4.19828385e-01 8.68058562e-01 -5.98160863e-01 -6.74687326e-01 -3.20601046e-01 -2.82984108e-01 4.25762802e-01 3.37175250e-01 -3.57786454e-02 -1.27930665e+00 -8.22171718e-02 3.09433937e-02 -2.02348337e-01 1.03878868e+00 1.97924197e-01 1.00602818e+00 -1.64913371e-01 -1.89578280e-01 5.03314555e-01 1.02113628e+00 3.29488546e-01 6.05468571e-01 4.37504739e-01 7.16230392e-01 4.79398459e-01 6.61825418e-01 1.29727125e-01 9.64084089e-01 4.33385044e-01 3.74652982e-01 -1.80217072e-01 -8.78303126e-02 -1.86143428e-01 7.01022863e-01 1.67757094e+00 1.76522195e-01 -2.42701948e-01 -2.67839313e-01 7.98353493e-01 -1.90048206e+00 -4.77417737e-01 -8.44220817e-03 1.80398738e+00 9.83983397e-01 3.14882755e-01 -2.83468775e-02 6.66636676e-02 3.27771068e-01 7.07176805e-01 -6.07899249e-01 -5.75449288e-01 -4.76106778e-02 1.38865605e-01 1.20896719e-01 5.27154326e-01 -1.34989524e+00 6.63599968e-01 6.33503771e+00 8.02385390e-01 -8.17620456e-01 1.40999869e-01 7.43440568e-01 -9.18827578e-02 -8.26689363e-01 2.09276661e-01 -6.78710461e-01 4.37559664e-01 6.66081429e-01 -1.77103937e-01 9.19933096e-02 1.14981687e+00 -3.59697849e-01 2.00941190e-01 -8.66748750e-01 6.67391956e-01 1.79043189e-01 -1.08276296e+00 2.32177973e-01 -2.63672262e-01 7.23650455e-01 -4.79165055e-02 -2.22132206e-02 6.30721271e-01 1.57560945e-01 -5.89771688e-01 5.65703094e-01 2.96894848e-01 4.30185020e-01 -1.01060593e+00 1.19429934e+00 -2.02979028e-01 -1.67445993e+00 2.33807191e-01 -9.78071988e-02 -3.30707543e-02 5.33321619e-01 1.15269291e+00 -1.56048477e-01 9.68920231e-01 5.71699560e-01 1.16271913e+00 -4.18524504e-01 8.33926201e-01 -5.90732038e-01 6.04534924e-01 1.10309534e-01 -2.85392404e-01 2.66463250e-01 -3.77403080e-01 3.95428091e-01 1.29083943e+00 1.33929402e-01 -3.60055804e-01 -8.65672827e-02 4.28045392e-01 -3.00458252e-01 3.67583066e-01 -4.42020953e-01 -3.65937263e-01 1.85555965e-01 1.51349092e+00 -4.14691031e-01 -5.07942021e-01 -1.22175288e+00 9.53520298e-01 3.79209220e-01 1.78466216e-01 -9.13082719e-01 -8.65382969e-01 1.06696475e+00 -3.51202458e-01 7.79315889e-01 1.62225336e-01 -5.75787544e-01 -1.55302715e+00 4.63821948e-01 -1.13821912e+00 2.40450963e-01 -5.14958739e-01 -1.13465083e+00 8.70912790e-01 -5.44726908e-01 -1.34652424e+00 7.88982492e-03 -5.78977168e-01 -7.81673193e-01 8.46370637e-01 -1.87178695e+00 -1.59624481e+00 -8.55144784e-02 2.04487309e-01 7.09588885e-01 8.47694054e-02 6.28777981e-01 4.88237858e-01 -8.29905689e-01 7.79244363e-01 -3.39971036e-01 1.87190965e-01 6.92744255e-01 -1.32487524e+00 7.98370302e-01 8.94034564e-01 -1.18262134e-01 9.71131265e-01 1.53080851e-01 -4.79769140e-01 -1.41721809e+00 -9.94514465e-01 1.37401617e+00 -3.35654497e-01 8.13829660e-01 -4.41551477e-01 -6.86781645e-01 7.42115974e-01 6.86190128e-01 -3.51810217e-01 1.08309436e+00 5.25915921e-01 -6.30066931e-01 -4.10271943e-01 -3.47416162e-01 5.94198227e-01 7.97389269e-01 -7.60628343e-01 -5.12255490e-01 1.44333720e-01 1.17198718e+00 -2.38039777e-01 -6.96498275e-01 7.18151450e-01 7.63018787e-01 -9.81870711e-01 7.01605439e-01 -4.97819602e-01 9.03052330e-01 -5.48366725e-01 8.68552327e-02 -1.21200299e+00 -2.20717967e-01 -4.74076658e-01 -4.62953180e-01 1.64019740e+00 8.68600547e-01 -5.64574838e-01 5.52639782e-01 3.87641191e-01 -1.54072374e-01 -1.44183242e+00 -4.80147392e-01 -2.73807496e-01 -2.57930219e-01 -4.98514384e-01 1.03680301e+00 6.88718557e-01 2.05268621e-01 1.27170825e+00 -4.05364990e-01 6.48506358e-02 1.11859821e-01 8.95316124e-01 5.74074924e-01 -6.37255311e-01 -6.21658146e-01 -5.92412174e-01 -4.92375828e-02 -1.41708827e+00 9.59048569e-02 -6.41128302e-01 -6.83058985e-03 -1.67071819e+00 6.52348638e-01 -4.29250263e-02 -4.40593630e-01 4.15663689e-01 -8.84991229e-01 4.51028198e-02 -9.87712517e-02 1.31584313e-02 -1.05786753e+00 1.02141964e+00 1.32655251e+00 -3.73568505e-01 -3.71914841e-02 1.55175343e-01 -1.27645791e+00 7.67538488e-01 5.16750097e-01 -3.77244115e-01 -6.02800965e-01 -8.59070778e-01 9.50555325e-01 -3.60639006e-01 -4.92980331e-01 -3.91321659e-01 2.39896402e-01 2.81523943e-01 -1.75123885e-02 -6.64661169e-01 2.14396700e-01 -6.17812157e-01 -4.47930396e-01 1.89622074e-01 -2.20039189e-01 3.28091770e-01 1.20502405e-01 6.33051038e-01 -7.91211605e-01 -9.89388674e-02 2.49935910e-01 9.79280993e-02 -4.20042068e-01 6.03305876e-01 -4.06540215e-01 -2.51203356e-03 8.21539581e-01 3.94519091e-01 -2.69059926e-01 -4.25477505e-01 -1.27367288e-01 4.91657704e-01 2.05737665e-01 5.47628462e-01 3.09596211e-01 -1.29419923e+00 -4.31384683e-01 2.39961118e-01 3.56339812e-01 2.51568526e-01 3.46635252e-01 1.01675642e+00 -2.06833154e-01 5.53164303e-01 3.35256577e-01 -5.09700775e-02 -1.18644595e+00 5.98419607e-01 2.38397226e-01 -7.82545924e-01 -1.79290906e-01 1.04166627e+00 6.69945717e-01 -6.29502118e-01 2.97217388e-02 -2.01376542e-01 -5.56046247e-01 2.65309781e-01 7.35183418e-01 8.69291872e-02 2.91857511e-01 -3.99553001e-01 -4.76833552e-01 6.10998213e-01 -5.90114176e-01 1.59762681e-01 1.17303240e+00 -2.28565261e-01 -4.87441391e-01 4.30252969e-01 1.21881652e+00 3.36061448e-01 -8.34380567e-01 -2.42236555e-01 -2.69894093e-01 -2.27889344e-01 1.26376510e-01 -7.63358474e-01 -1.50817001e+00 1.05453241e+00 6.96227001e-03 1.38557076e-01 1.50298226e+00 -2.01653570e-01 1.20335591e+00 3.55134994e-01 3.55599448e-03 -9.00510311e-01 -7.50553012e-02 7.52728462e-01 4.08314466e-01 -1.18575978e+00 1.59545183e-01 -6.87448740e-01 -9.55851972e-01 6.68621480e-01 6.70073390e-01 -2.12539621e-02 8.81436110e-01 5.39154351e-01 2.85939008e-01 -2.50621170e-01 -1.12897229e+00 -4.17799830e-01 2.49122441e-01 1.11241117e-01 9.82171893e-01 -2.25133616e-02 -7.96796203e-01 1.34051645e+00 -1.74182072e-01 -7.04414770e-02 9.02967080e-02 1.14203346e+00 6.25102445e-02 -1.32328510e+00 4.83534843e-01 4.88000989e-01 -8.79311502e-01 -6.90223277e-01 -4.86525118e-01 3.78531873e-01 2.45933123e-02 8.67585480e-01 -1.67470470e-01 -6.46003604e-01 5.39134920e-01 5.71031943e-02 -4.55978103e-02 -3.68908197e-01 -9.04570401e-01 1.52471811e-01 5.00446498e-01 -7.01399148e-01 -7.21243680e-01 -5.39240539e-01 -8.71249735e-01 -9.54992175e-02 -6.59206629e-01 2.95275867e-01 6.27993762e-01 1.17381465e+00 8.73886645e-01 1.08449185e+00 8.34286869e-01 -1.63111359e-01 -1.29071683e-01 -1.11808932e+00 -4.25013602e-01 1.59864113e-01 4.09914047e-01 -4.26571816e-01 -7.65476525e-02 -9.47770849e-02]
[11.529488563537598, 6.580546855926514]
b38b310d-71f5-414a-824e-3a09c18d5844
fundamental-limits-and-tradeoffs-in-invariant-1
2012.10713
null
https://arxiv.org/abs/2012.10713v4
https://arxiv.org/pdf/2012.10713v4.pdf
Fundamental Limits and Tradeoffs in Invariant Representation Learning
A wide range of machine learning applications such as privacy-preserving learning, algorithmic fairness, and domain adaptation/generalization among others, involve learning invariant representations of the data that aim to achieve two competing goals: (a) maximize information or accuracy with respect to a target response, and (b) maximize invariance or independence with respect to a set of protected features (e.g., for fairness, privacy, etc). Despite their wide applicability, theoretical understanding of the optimal tradeoffs -- with respect to accuracy, and invariance -- achievable by invariant representations is still severely lacking. In this paper, we provide an information theoretic analysis of such tradeoffs under both classification and regression settings. More precisely, we provide a geometric characterization of the accuracy and invariance achievable by any representation of the data; we term this feasible region the information plane. We provide an inner bound for this feasible region for the classification case, and an exact characterization for the regression case, which allows us to either bound or exactly characterize the Pareto optimal frontier between accuracy and invariance. Although our contributions are mainly theoretical, a key practical application of our results is in certifying the potential sub-optimality of any given representation learning algorithm for either classification or regression tasks. Our results shed new light on the fundamental interplay between accuracy and invariance, and may be useful in guiding the design of future representation learning algorithms.
['Pradeep Ravikumar', 'Geoffrey J. Gordon', 'Tommi S. Jaakkola', 'Bryon Aragam', 'Chen Dan', 'Han Zhao']
2020-12-19
fundamental-limits-and-tradeoffs-in-invariant
https://openreview.net/forum?id=9CG8RW_p3Y
https://openreview.net/pdf?id=9CG8RW_p3Y
null
['information-plane']
['methodology']
[ 6.42384648e-01 1.31718397e-01 -7.46377528e-01 -5.15423298e-01 -7.97185838e-01 -7.38553762e-01 3.28539103e-01 4.58113849e-01 -4.08685416e-01 7.30726779e-01 2.66937554e-01 -4.26694214e-01 -6.60533905e-01 -7.36424208e-01 -4.71251220e-01 -8.27198803e-01 -1.08422726e-01 1.12472959e-01 -5.46582282e-01 2.15798125e-01 2.40780383e-01 7.24990368e-01 -1.49734378e+00 -4.15647812e-02 9.57155526e-01 1.27498305e+00 -6.16599739e-01 3.99853706e-01 2.46852726e-01 4.85416681e-01 -2.53518492e-01 -4.81090993e-01 6.28855884e-01 -4.27684098e-01 -7.65466452e-01 -1.24772206e-01 4.13718790e-01 -1.40729204e-01 -2.06245974e-01 1.28883731e+00 4.15994376e-01 2.16283977e-01 9.28398073e-01 -1.56080031e+00 -7.69851327e-01 3.31656277e-01 -4.89163995e-01 4.15748768e-02 5.45854568e-02 -1.58666059e-01 1.38246429e+00 -1.42409921e-01 4.02024865e-01 9.78821039e-01 4.27562684e-01 7.93030858e-01 -1.49096155e+00 -7.73123980e-01 5.29442132e-02 -1.54072538e-01 -1.33000910e+00 -7.84137368e-01 5.10446727e-01 -5.41920364e-01 1.47652254e-01 6.93186462e-01 2.71873385e-01 7.19579160e-01 2.34279707e-01 8.16501915e-01 1.20621049e+00 -3.93571645e-01 4.14814711e-01 3.57580811e-01 5.09208500e-01 5.99511027e-01 6.53396130e-01 3.29932481e-01 -4.25349981e-01 -5.23504972e-01 6.83258176e-01 1.75178915e-01 -4.77477044e-01 -9.29665685e-01 -5.73169947e-01 1.23333406e+00 3.30485761e-01 -5.30371554e-02 -3.25734735e-01 7.31836334e-02 3.53460431e-01 6.75562918e-01 4.76137668e-01 6.72962844e-01 -3.44371200e-01 2.28931233e-01 -7.46215820e-01 3.20423394e-01 9.17754591e-01 8.18117797e-01 6.76076293e-01 -1.21757656e-01 -3.06547076e-01 5.34818828e-01 -2.45600399e-02 3.33295584e-01 2.62281924e-01 -1.01434577e+00 4.71133351e-01 4.19424385e-01 2.34628171e-01 -9.50893998e-01 -3.46122980e-01 -6.84840441e-01 -8.79181325e-01 5.15619814e-02 6.31889164e-01 -1.61365598e-01 -2.72541404e-01 2.32997680e+00 3.49722542e-02 -3.57372403e-01 1.10942699e-01 8.11896503e-01 4.42446470e-02 3.74457568e-01 2.45008886e-01 -6.13579214e-01 1.23142481e+00 -1.21182382e-01 -4.70235944e-01 -3.64244655e-02 7.35598683e-01 -2.87239164e-01 9.93777514e-01 -1.92814823e-02 -9.57025409e-01 8.32061768e-02 -9.00378942e-01 -2.87884735e-02 -1.17102064e-01 -2.53682166e-01 8.08329821e-01 1.18117332e+00 -7.91805029e-01 4.73912418e-01 -5.74714422e-01 -4.07829434e-01 7.59253442e-01 5.91170132e-01 -4.12662685e-01 -1.33231342e-01 -1.09426379e+00 5.99634051e-01 8.13748762e-02 -3.81528914e-01 -4.34087396e-01 -1.03149605e+00 -7.29629040e-01 3.82515371e-01 3.88737410e-01 -7.91565478e-01 1.05005562e+00 -1.26339436e+00 -9.95118737e-01 1.03053570e+00 -8.16429965e-03 -6.73697412e-01 6.56500697e-01 2.40703914e-02 -2.04777107e-01 -1.34439915e-01 -3.78435329e-02 2.19358355e-01 6.35480285e-01 -8.68013263e-01 -6.63878739e-01 -8.48847210e-01 2.81792969e-01 3.16218525e-01 -6.87042773e-01 -6.95369300e-03 2.51008391e-01 -6.71003520e-01 -1.39139235e-01 -8.53283048e-01 -3.45919579e-01 4.66784984e-01 -2.59591877e-01 -5.99731430e-02 5.97951591e-01 -5.07950187e-01 1.17453802e+00 -2.22867250e+00 -4.39205067e-03 5.07961810e-01 1.63868457e-01 -1.20901354e-02 -2.81762421e-01 2.75202274e-01 -5.44515178e-02 3.67844403e-01 -3.92783374e-01 2.10853647e-02 9.69860479e-02 -4.16905200e-03 -5.97277522e-01 1.10460126e+00 7.05865920e-02 8.19034278e-01 -5.99031985e-01 -2.12162271e-01 6.70294650e-03 2.36366019e-01 -7.42969692e-01 -1.54333720e-02 4.95686084e-02 3.46571326e-01 -7.73713231e-01 5.41243553e-01 5.91946900e-01 -1.09390356e-02 1.68781817e-01 2.30583861e-01 -2.63270587e-02 1.61447853e-01 -9.61639285e-01 1.18270385e+00 -5.09260595e-01 5.89875340e-01 2.71213412e-01 -1.23109853e+00 8.08860064e-01 1.80226229e-02 7.22238481e-01 -5.98789692e-01 4.89572473e-02 6.59281835e-02 -1.44398198e-01 -8.92925709e-02 3.63934487e-01 -4.70628202e-01 -2.72548079e-01 7.37061203e-01 -3.53656203e-01 4.04545486e-01 -4.20922846e-01 -1.03611447e-01 8.34894478e-01 -3.42153639e-01 7.98927724e-01 -7.06104279e-01 3.95544589e-01 -3.62592727e-01 7.06435204e-01 9.43027556e-01 -3.98918241e-01 3.58600020e-01 9.29546893e-01 -2.99951881e-01 -9.22067046e-01 -1.00958765e+00 -5.52411199e-01 1.08240986e+00 4.09749635e-02 -2.33359588e-03 -4.30414706e-01 -6.83778644e-01 5.54459095e-01 7.19564974e-01 -8.55943263e-01 -5.05212247e-01 -1.64813176e-01 -8.77516270e-01 3.60531330e-01 3.04968327e-01 4.23233002e-01 -2.95139760e-01 -7.27933705e-01 -2.80151159e-01 6.23544492e-02 -7.47215152e-01 -6.55085742e-01 1.19221143e-01 -9.83259618e-01 -1.17385197e+00 -4.15485471e-01 -2.07853660e-01 7.08579838e-01 4.04917479e-01 7.33118355e-01 -1.13693215e-01 -4.07293905e-03 6.82945907e-01 5.81940543e-03 -5.32331705e-01 -2.10440412e-01 8.40118453e-02 1.98234960e-01 3.77283424e-01 2.08600447e-01 -6.04225338e-01 -6.59290493e-01 3.24063718e-01 -1.04051173e+00 -3.30733716e-01 4.54340786e-01 6.71955407e-01 5.53061604e-01 1.72780249e-02 5.90419173e-01 -1.10957789e+00 7.63794363e-01 -7.13661611e-01 -8.52170467e-01 5.41342616e-01 -8.28637540e-01 4.10486758e-01 7.38566399e-01 -2.49723420e-01 -7.64619887e-01 1.13392830e-01 2.99946427e-01 -3.12249511e-01 2.41821095e-01 4.12717521e-01 -4.78078455e-01 -2.49393225e-01 6.71705484e-01 1.82690457e-01 1.63708359e-01 -3.40100318e-01 4.32702541e-01 7.57678211e-01 4.39090192e-01 -9.68000174e-01 5.58764100e-01 4.53816414e-01 4.74669904e-01 -7.89274037e-01 -8.87954056e-01 -3.89840692e-01 -3.24761242e-01 2.25963280e-01 4.87423241e-01 -6.52473688e-01 -9.26984549e-01 -1.98630229e-01 -5.13966143e-01 4.64325659e-02 -7.14673102e-01 3.22608858e-01 -9.53701198e-01 3.63478303e-01 8.47506374e-02 -1.32332397e+00 -2.77234316e-01 -8.03676128e-01 5.81523955e-01 6.07261918e-02 -2.12701172e-01 -9.84225392e-01 -6.53815866e-02 3.52971643e-01 4.20285165e-01 3.97312880e-01 1.25819457e+00 -9.51817155e-01 -4.47984993e-01 -5.20564258e-01 -3.08721989e-01 1.77027300e-01 2.20563978e-01 -3.34719121e-01 -1.00793016e+00 -5.49629271e-01 3.00803129e-02 -3.42351496e-01 9.67634320e-01 5.38536012e-01 1.62358940e+00 -9.97793794e-01 -1.69900090e-01 7.98374176e-01 1.35394502e+00 -2.66125780e-02 3.67111295e-01 4.13315296e-02 1.54562429e-01 9.65921760e-01 4.71657991e-01 7.84101605e-01 7.29713887e-02 8.06219578e-01 2.94386625e-01 2.00664699e-01 3.46864760e-01 -4.33667868e-01 1.55232832e-01 -1.63088471e-01 1.13639697e-01 -1.58279151e-01 -4.94319111e-01 3.84011120e-01 -1.90773153e+00 -1.09960759e+00 4.96683568e-01 2.99515295e+00 6.87317133e-01 -3.32325846e-01 5.57302117e-01 1.31113008e-01 7.85978854e-01 5.27000204e-02 -9.89286542e-01 -6.41542375e-01 -1.20892331e-01 1.97555125e-01 8.91809881e-01 3.83749664e-01 -1.10944676e+00 4.02428120e-01 6.66648960e+00 7.29663312e-01 -1.00103498e+00 -1.26208588e-01 1.02365828e+00 -1.95309952e-01 -7.04256833e-01 8.25423896e-02 -4.49797660e-01 1.91799670e-01 8.86246204e-01 -8.72544050e-01 6.36882544e-01 9.74794686e-01 1.28684670e-01 3.08903545e-01 -1.35730183e+00 8.41934562e-01 -2.46539801e-01 -1.23182082e+00 1.85247064e-01 5.90896010e-01 7.19480455e-01 -5.16324043e-01 4.65304106e-01 8.73527750e-02 4.55945075e-01 -1.20875108e+00 4.99818116e-01 2.58919895e-01 1.14756203e+00 -1.10728085e+00 3.84147286e-01 3.46638888e-01 -7.94097126e-01 -4.09081042e-01 -4.21623588e-01 -1.26109108e-01 -5.72227299e-01 5.01797736e-01 -3.09922189e-01 6.25338972e-01 9.66868103e-02 4.53332216e-01 -1.32175654e-01 9.07701194e-01 1.08948320e-01 3.47776771e-01 -2.52304584e-01 7.86592364e-02 -6.63487837e-02 -1.81862697e-01 5.48897266e-01 1.04488444e+00 7.77030364e-02 1.81636244e-01 -1.30139226e-02 8.90128553e-01 -4.97149736e-01 2.98708856e-01 -8.00112307e-01 -2.57181823e-01 7.08618462e-01 9.35069859e-01 -2.63852239e-01 1.85082555e-01 -2.91795433e-01 6.34417534e-01 4.31266963e-01 3.18071574e-01 -4.07105297e-01 -3.35394502e-01 1.43568599e+00 1.53009459e-01 -1.23476230e-01 4.74269390e-02 -7.82577753e-01 -1.24971271e+00 1.03069752e-01 -7.52178729e-01 1.05402839e+00 1.22860670e-01 -1.38445914e+00 -8.16727709e-03 -6.60439432e-02 -1.06510484e+00 -2.58539230e-01 -5.14253080e-01 -4.22975183e-01 9.20267224e-01 -1.40618694e+00 -7.96478629e-01 2.30025440e-01 6.79010212e-01 1.89951267e-02 -1.84909239e-01 9.23352361e-01 -2.98422761e-02 -5.46890199e-01 1.18978429e+00 4.89243597e-01 -9.90393981e-02 3.20506215e-01 -1.05015337e+00 -1.14056520e-01 7.41219282e-01 1.59220412e-01 7.59490430e-01 5.70997715e-01 -2.19250366e-01 -1.68609726e+00 -1.25246108e+00 6.34655595e-01 -3.66119713e-01 4.60566074e-01 -1.86337605e-01 -6.17975175e-01 7.14659870e-01 -6.49734378e-01 1.80896997e-01 1.04768622e+00 4.22085911e-01 -7.26928592e-01 -3.63764197e-01 -1.66709769e+00 6.10429645e-01 1.00930190e+00 -5.37547171e-01 -3.46524804e-03 1.98966861e-01 5.73945880e-01 -7.08722547e-02 -9.44016993e-01 3.11110765e-01 9.31077302e-01 -8.65001857e-01 9.34385896e-01 -1.31586778e+00 2.56080449e-01 1.55308068e-01 -6.87515557e-01 -1.01009262e+00 -4.00470495e-01 -8.78721833e-01 -8.16466212e-02 9.67658877e-01 2.12870106e-01 -1.03928053e+00 8.82459104e-01 1.25323248e+00 4.36018974e-01 -9.13766444e-01 -1.24881744e+00 -1.00126195e+00 4.90588456e-01 -3.92405599e-01 6.96258962e-01 1.17292130e+00 -3.39366123e-02 3.59418727e-02 -5.58430135e-01 1.51276380e-01 8.26175153e-01 3.34846467e-01 6.71808243e-01 -1.31879103e+00 -1.19071163e-01 -7.22557068e-01 -6.69149816e-01 -9.03777122e-01 3.98290008e-01 -1.02738607e+00 -5.31627893e-01 -8.86522114e-01 5.22206962e-01 -7.48636484e-01 -6.63896143e-01 4.71568197e-01 4.17621667e-03 -1.74609855e-01 3.20980966e-01 3.97018373e-01 -3.30040991e-01 5.45353174e-01 7.78255522e-01 5.04185222e-02 -2.63763011e-01 6.21836662e-01 -1.54984808e+00 3.14166307e-01 8.31408203e-01 -5.06275594e-01 -5.17846048e-01 -2.57383231e-02 1.04426377e-01 3.22336286e-01 3.40039551e-01 -4.96521443e-01 -1.79798491e-02 -6.57755077e-01 2.48897210e-01 1.47713751e-01 8.49489346e-02 -9.43493128e-01 -5.63118942e-02 6.90128386e-01 -1.10464132e+00 -2.35215336e-01 -7.07972273e-02 8.69365036e-01 3.05830181e-01 -1.57570630e-01 1.11232865e+00 2.99271911e-01 -2.02492997e-01 6.17256939e-01 -1.44479917e-02 4.90686744e-01 1.06595993e+00 -1.95665836e-01 -4.25114334e-01 -7.52203405e-01 -2.30294019e-01 2.93983996e-01 6.95800364e-01 2.73817033e-01 3.88619930e-01 -1.25983715e+00 -8.64396751e-01 3.10731471e-01 2.88411409e-01 -5.46185791e-01 1.38707999e-02 4.98278171e-01 1.11672923e-01 6.84270978e-01 -2.75540024e-01 -5.21385074e-02 -1.12591982e+00 5.46815395e-01 4.06043231e-01 -3.01922560e-01 -3.26839238e-01 3.40250999e-01 6.85894549e-01 -1.43080518e-01 3.17802459e-01 1.17610425e-01 2.62335837e-01 -1.84694782e-01 5.14860094e-01 5.11788189e-01 -1.86525553e-01 -4.22840685e-01 -3.80495876e-01 2.89386332e-01 -1.37424067e-01 -1.27732053e-01 1.10558879e+00 -9.02786665e-03 4.96711507e-02 4.07855175e-02 1.36758506e+00 -2.80071981e-02 -1.27372169e+00 -4.27210510e-01 5.86066879e-02 -8.81453574e-01 3.13443914e-02 -6.29670620e-01 -1.07831955e+00 6.88165963e-01 5.05267680e-01 3.55269402e-01 1.33967280e+00 -9.26827416e-02 2.05687717e-01 3.55085105e-01 2.83802360e-01 -6.68213844e-01 -4.19272542e-01 -3.77221406e-03 7.83883154e-01 -1.01633263e+00 1.65545061e-01 -2.38836735e-01 -4.71865535e-01 9.35987115e-01 2.03202367e-01 6.87070787e-02 5.00513434e-01 -5.06809391e-02 -4.66293842e-01 2.47390583e-01 -7.48464346e-01 -1.95116946e-03 4.23721343e-01 8.42767954e-01 4.31360364e-01 3.73190254e-01 -4.89742249e-01 7.24097490e-01 -2.92016208e-01 -2.27425978e-01 3.15131366e-01 6.91824555e-01 -4.57438886e-01 -1.02702820e+00 -2.38557503e-01 7.41584837e-01 -6.67967379e-01 1.08590312e-01 -6.43694997e-01 7.12706983e-01 -3.49655032e-01 9.42763448e-01 1.31628111e-01 -5.96867427e-02 2.75214136e-01 -8.65543410e-02 5.78518093e-01 -3.51048440e-01 -1.95976034e-01 -4.42240208e-01 -7.27652162e-02 -4.86993462e-01 -1.06983587e-01 -8.96742344e-01 -7.27765262e-01 -7.36385345e-01 -2.20150664e-01 2.06175193e-01 5.21880984e-01 7.20656872e-01 4.17603046e-01 -1.47703141e-01 1.10432434e+00 -8.90335348e-03 -1.22536600e+00 -2.72006363e-01 -8.20647478e-01 2.95333177e-01 6.03593171e-01 -3.45501751e-01 -3.30995351e-01 -5.53159773e-01]
[6.188374996185303, 6.732506275177002]
6a08b505-0ec6-4994-b649-150822f443d3
dip-dual-incongruity-perceiving-network-for
null
null
http://openaccess.thecvf.com//content/CVPR2023/html/Wen_DIP_Dual_Incongruity_Perceiving_Network_for_Sarcasm_Detection_CVPR_2023_paper.html
http://openaccess.thecvf.com//content/CVPR2023/papers/Wen_DIP_Dual_Incongruity_Perceiving_Network_for_Sarcasm_Detection_CVPR_2023_paper.pdf
DIP: Dual Incongruity Perceiving Network for Sarcasm Detection
Sarcasm indicates the literal meaning is contrary to the real attitude. Considering the popularity and complementarity of image-text data, we investigate the task of multi-modal sarcasm detection. Different from other multi-modal tasks, for the sarcastic data, there exists intrinsic incongruity between a pair of image and text as demonstrated in psychological theories. To tackle this issue, we propose a Dual Incongruity Perceiving (DIP) network consisting of two branches to mine the sarcastic information from factual and affective levels. For the factual aspect, we introduce a channel-wise reweighting strategy to obtain semantically discriminative embeddings, and leverage gaussian distribution to model the uncertain correlation caused by the incongruity. The distribution is generated from the latest data stored in the memory bank, which can adaptively model the difference of semantic similarity between sarcastic and non-sarcastic data. For the affective aspect, we utilize siamese layers with shared parameters to learn cross-modal sentiment information. Furthermore, we use the polarity value to construct a relation graph for the mini-batch, which forms the continuous contrastive loss to acquire affective embeddings. Extensive experiments demonstrate that our proposed method performs favorably against state-of-the-art approaches. Our code is released on https://github.com/downdric/MSD.
['Jufeng Yang', 'Guoli Jia', 'Changsong Wen']
2023-01-01
null
null
null
cvpr-2023-1
['sarcasm-detection', 'semantic-textual-similarity', 'semantic-similarity']
['natural-language-processing', 'natural-language-processing', 'natural-language-processing']
[-3.58598202e-01 -6.93275258e-02 8.89812782e-03 -5.46277046e-01 -5.21930695e-01 -2.75599211e-01 5.22445679e-01 9.90239233e-02 -5.93896329e-01 6.70520216e-02 6.30912840e-01 4.27531153e-01 3.15836191e-01 -4.61808294e-01 -4.29672331e-01 -6.67816937e-01 5.91013908e-01 3.10249060e-01 -2.02321306e-01 -3.98977965e-01 2.91888356e-01 -2.32436046e-01 -1.22137558e+00 6.45410061e-01 5.01906753e-01 1.20965171e+00 4.29857261e-02 1.41696990e-01 -3.87956440e-01 9.05287683e-01 -3.41429621e-01 -9.33250606e-01 -4.04594392e-02 -5.76893330e-01 -4.26959485e-01 2.28812888e-01 3.39157492e-01 -6.02604225e-02 -3.14554542e-01 1.50748920e+00 6.96898580e-01 -1.49089722e-02 5.22246778e-01 -1.32405782e+00 -1.10765684e+00 5.68438232e-01 -1.16145790e+00 3.35198194e-01 1.77640736e-01 1.24707282e-01 1.04000902e+00 -1.41584730e+00 3.53609115e-01 1.48564970e+00 6.12444818e-01 4.96046931e-01 -9.77271616e-01 -8.14716756e-01 8.18326548e-02 5.07123470e-01 -1.39649761e+00 -4.14344251e-01 1.35940313e+00 -4.30934072e-01 3.39635283e-01 -4.34353538e-02 1.00156701e+00 1.27473962e+00 2.26986229e-01 1.08257973e+00 1.12482297e+00 -1.24787502e-01 -5.91052175e-02 4.79611158e-01 8.36079717e-02 6.35691464e-01 -2.19098121e-01 -3.96393746e-01 -6.94187224e-01 2.38195527e-02 3.45436633e-01 2.08817959e-01 -2.15145439e-01 -3.75866234e-01 -1.17690229e+00 1.13885748e+00 4.83842939e-01 3.27251077e-01 -3.45730215e-01 -3.79702412e-02 7.34970033e-01 2.46846199e-01 5.66708088e-01 1.75177440e-01 -1.01580791e-01 -1.01059988e-01 -7.42508292e-01 -9.41437632e-02 4.20584470e-01 4.15961564e-01 6.07317626e-01 -5.95966242e-02 4.28062379e-02 1.20671737e+00 4.24069285e-01 4.54897374e-01 9.72588480e-01 -4.98065859e-01 5.21351099e-01 7.25998282e-01 -2.91100949e-01 -1.67134213e+00 -4.07351613e-01 -3.45778793e-01 -1.08149707e+00 -9.72983539e-02 5.54285161e-02 -1.81973781e-02 -2.57806212e-01 1.86835325e+00 3.82423997e-01 1.43550143e-01 2.67249197e-01 1.40372860e+00 1.06395996e+00 6.38771474e-01 1.73757263e-02 -2.37531617e-01 1.66105056e+00 -1.19157708e+00 -8.27423215e-01 -4.86151248e-01 4.08076882e-01 -8.92088950e-01 1.50140953e+00 2.34407425e-01 -1.01852643e+00 -5.85408390e-01 -1.11751878e+00 -2.63379276e-01 -2.92228252e-01 3.66662860e-01 3.40576559e-01 1.69550702e-01 -6.01300895e-01 1.08607054e-01 -5.64638615e-01 -2.55405623e-02 3.39352787e-01 -1.16998851e-01 -3.68382186e-01 -1.46460943e-02 -1.48720980e+00 8.62793803e-01 3.92773300e-01 3.50060463e-01 -4.15801555e-01 -4.77625400e-01 -1.00175834e+00 -7.64335021e-02 2.40296125e-01 -5.80594838e-01 7.18109012e-01 -1.61851263e+00 -1.24301958e+00 1.33931637e+00 1.07315749e-01 -1.33368641e-01 2.78507620e-01 -2.21907884e-01 -7.11608291e-01 2.55240500e-01 1.96363911e-01 6.03938997e-01 1.06586945e+00 -1.12226307e+00 -3.29036191e-02 -6.35481954e-01 -7.13519901e-02 8.40678990e-01 -8.42539728e-01 5.38323037e-02 -5.85261762e-01 -7.50064492e-01 2.45951161e-01 -8.59373093e-01 8.54275376e-02 -6.28463086e-03 -4.02810961e-01 -2.09131658e-01 7.26625621e-01 -4.69464749e-01 1.17031693e+00 -2.60559940e+00 3.37456018e-01 -1.71766102e-01 4.00012225e-01 -1.94604084e-01 -2.09859580e-01 2.78167903e-01 -2.63458699e-01 -2.95215279e-01 -1.53170019e-01 -5.55285633e-01 2.65486185e-02 6.84922189e-02 -3.03123116e-01 6.51796818e-01 3.97269987e-02 9.99436259e-01 -6.62317634e-01 -8.59259784e-01 1.65229172e-01 4.97854173e-01 -3.18737805e-01 2.94033438e-01 1.59880251e-01 2.59412050e-01 -4.83892560e-01 3.97712559e-01 8.22076023e-01 -4.90435094e-01 1.13700911e-01 -8.39207590e-01 1.94383889e-01 -9.26685333e-02 -9.68643844e-01 2.08893156e+00 -4.56894219e-01 2.76772588e-01 -2.86777895e-02 -1.28582978e+00 1.09380066e+00 1.18044548e-01 4.08673525e-01 -8.89736295e-01 4.79948014e-01 -2.59357095e-02 -8.33961442e-02 -7.76498854e-01 4.63388264e-01 -6.66888833e-01 -3.96156400e-01 4.36491191e-01 8.05798844e-02 -3.57809812e-01 -2.23714799e-01 3.65266353e-01 4.66299802e-01 -2.25882620e-01 1.05185360e-01 -3.01144958e-01 8.27771366e-01 -1.77382648e-01 6.08697414e-01 -4.64225002e-02 -3.68483007e-01 6.41061604e-01 6.86816633e-01 -3.68231267e-01 -8.27164531e-01 -1.29596913e+00 1.00765802e-01 1.03298461e+00 7.31508851e-01 -3.40104967e-01 -5.03135443e-01 -5.83102584e-01 -1.23823017e-01 5.63015163e-01 -8.46106708e-01 -5.03793657e-01 -3.43710303e-01 -1.06161356e+00 1.86463118e-01 3.32478732e-01 7.21866488e-01 -1.00031674e+00 -3.58015656e-01 -8.72553140e-02 -4.06698674e-01 -1.01939535e+00 -7.24622548e-01 -1.32532865e-01 -3.94355685e-01 -9.16422963e-01 -4.51886177e-01 -9.49546278e-01 5.17480850e-01 1.51186913e-01 1.19652998e+00 -1.20922700e-01 -1.29868165e-01 4.75998938e-01 -1.88745216e-01 1.48231806e-02 -8.66909176e-02 -4.20251846e-01 -2.13126782e-02 3.26141000e-01 8.21966171e-01 -5.42019665e-01 -9.94334817e-01 1.36522010e-01 -1.06544495e+00 3.57866168e-01 4.82996464e-01 1.08591449e+00 7.36305475e-01 -3.04699615e-02 6.59078062e-01 -6.57496750e-01 9.76416528e-01 -8.87978196e-01 1.23380229e-01 1.15825593e-01 -6.40823305e-01 -1.83831602e-01 7.32733011e-01 -7.04796672e-01 -1.06973922e+00 1.78256736e-03 -1.08991817e-01 -8.05824161e-01 1.79551736e-01 6.15840614e-01 -1.64088726e-01 5.62623918e-01 2.17137918e-01 3.67289752e-01 3.93813223e-01 -1.10094815e-01 7.90222347e-01 6.86187863e-01 7.38896549e-01 -4.11580980e-01 4.11949545e-01 8.02675188e-01 -4.65562522e-01 -5.24062634e-01 -1.33402300e+00 -5.66828012e-01 -3.66208047e-01 -3.95219892e-01 1.02342880e+00 -1.24252856e+00 -4.99208957e-01 5.94927430e-01 -1.07850063e+00 2.37056717e-01 -3.46853763e-01 5.47352195e-01 -6.18406594e-01 5.68415403e-01 -8.67021084e-01 -4.44958806e-01 -3.92537862e-01 -1.04496467e+00 9.86885905e-01 3.58405739e-01 -2.22276181e-01 -1.15335369e+00 3.10145944e-01 7.96332538e-01 5.39072268e-02 2.59629972e-02 6.48581922e-01 -6.49031520e-01 2.66519785e-01 -1.47353500e-01 -3.71623456e-01 5.83386004e-01 -6.03209212e-02 -4.14842933e-01 -8.18025589e-01 -2.85245240e-01 6.87547684e-01 -1.07140100e+00 8.33807766e-01 1.32583588e-01 1.15566528e+00 -3.02664280e-01 2.14781120e-01 3.81853998e-01 1.39787662e+00 -3.16802472e-01 5.77047884e-01 1.79237068e-01 8.18765700e-01 7.11621583e-01 7.38946497e-01 6.83425426e-01 8.21449518e-01 4.56488967e-01 4.99167740e-01 -2.18787178e-01 -8.04628283e-02 -3.44593495e-01 5.19130349e-01 1.39177012e+00 5.98602653e-01 -5.22096157e-02 -6.60324514e-01 5.41375101e-01 -1.90260983e+00 -1.13371944e+00 -7.34313652e-02 1.76245296e+00 1.14818859e+00 2.98346747e-02 -1.01290777e-01 -1.98885143e-01 7.94346631e-01 5.45575976e-01 -6.77289844e-01 -3.24080348e-01 -5.42854428e-01 -2.43703008e-01 -3.56059633e-02 3.25512201e-01 -1.02998364e+00 9.59723175e-01 4.52237320e+00 1.18534732e+00 -1.22093832e+00 4.44885463e-01 8.86361778e-01 -3.65309685e-01 -5.78041017e-01 -2.61112332e-01 -3.35315645e-01 7.80952930e-01 6.20211244e-01 -1.59344658e-01 1.45675346e-01 7.72734821e-01 2.69975513e-01 -1.14794120e-01 -6.92271352e-01 1.32775486e+00 8.53620529e-01 -1.02862275e+00 -1.20624632e-01 -5.34084558e-01 4.94347215e-01 1.64334234e-02 4.92388666e-01 3.93007338e-01 4.62631695e-02 -8.72105718e-01 7.51732051e-01 6.53400958e-01 6.01724923e-01 -7.35079646e-01 7.67339885e-01 2.07768559e-01 -1.15387058e+00 1.14864018e-02 -3.42019349e-01 3.54731381e-02 2.88365096e-01 8.25093567e-01 -5.42073734e-02 2.01482713e-01 7.66718566e-01 1.09505415e+00 -6.61153913e-01 2.43040964e-01 -6.90663680e-02 3.03175151e-01 -2.15978518e-01 1.61548294e-02 2.63930023e-01 -5.02224803e-01 2.75176376e-01 1.03798568e+00 5.65394200e-02 -9.56624076e-02 1.81629166e-01 8.63244355e-01 -7.53146932e-02 4.73183036e-01 -3.97798389e-01 -1.51935786e-01 2.04485878e-01 1.65303361e+00 -5.13424933e-01 -4.45388436e-01 -6.37183249e-01 1.55406082e+00 6.30336821e-01 1.79272052e-02 -1.11183977e+00 -6.87062368e-02 3.09459180e-01 -3.14950407e-01 1.15149148e-01 -5.90540059e-02 -4.01645839e-01 -1.44174337e+00 1.69801325e-01 -8.09190035e-01 7.49889255e-01 -1.05107915e+00 -1.88376486e+00 6.17749393e-01 -3.51961285e-01 -1.17038953e+00 1.61843181e-01 -3.15674067e-01 -7.44433939e-01 6.68087780e-01 -1.27030134e+00 -1.33146727e+00 -3.62132221e-01 7.45588362e-01 5.20862818e-01 -7.12417886e-02 5.37534595e-01 3.77096415e-01 -5.62983274e-01 4.88728106e-01 -3.96510363e-02 7.41826892e-02 9.21987236e-01 -1.07960057e+00 -3.48376185e-01 4.66541022e-01 1.02640383e-01 4.04325038e-01 8.01767945e-01 -3.84614468e-01 -1.21816957e+00 -7.11782038e-01 5.70519328e-01 -2.64351666e-01 1.21201706e+00 -2.46015936e-01 -9.10308063e-01 2.86826730e-01 5.29799700e-01 5.78331240e-02 1.01087117e+00 -1.89360045e-02 -6.84560955e-01 -8.79456028e-02 -8.41197729e-01 6.86931551e-01 6.34173810e-01 -7.01194763e-01 -9.12679374e-01 2.88642228e-01 6.33058548e-01 -1.21180959e-01 -8.61442029e-01 2.43252754e-01 4.64804828e-01 -1.18574524e+00 8.63802135e-01 -5.13690948e-01 1.09760439e+00 -2.11219281e-01 -3.10795397e-01 -1.40832114e+00 -1.68723315e-01 5.49345240e-02 1.19974218e-01 1.26501417e+00 1.42073125e-01 -2.74557173e-01 7.47167468e-01 2.10407436e-01 -8.46049041e-02 -1.04676270e+00 -9.73797858e-01 -3.82205069e-01 2.97657877e-01 -1.62126034e-01 2.29346976e-01 1.30222416e+00 3.17806363e-01 8.62595320e-01 -6.00332975e-01 -1.32688850e-01 3.51842314e-01 5.91185391e-01 5.28846085e-01 -6.63137853e-01 -3.55904788e-01 -5.21070063e-01 -3.37219894e-01 -1.00467169e+00 5.79051316e-01 -9.20389473e-01 -1.78172011e-02 -1.13713360e+00 6.53288245e-01 -1.64178714e-01 -6.28980458e-01 1.67226672e-01 -3.19414973e-01 5.38477778e-01 4.71358150e-02 3.32319796e-01 -9.47954834e-01 1.20239007e+00 1.34978998e+00 -3.25332701e-01 1.78879991e-01 -5.14603555e-01 -8.02307189e-01 1.06270146e+00 9.72430170e-01 -2.62591869e-01 -6.16515994e-01 -3.96394879e-01 5.84010005e-01 8.88049006e-02 4.98530239e-01 -7.60374725e-01 2.08501071e-01 -8.67227390e-02 1.63546994e-01 -6.32569849e-01 8.31672192e-01 -8.76298904e-01 -2.03533068e-01 2.93895811e-01 -5.10642529e-01 4.54849690e-01 -7.18721673e-02 7.02183783e-01 -6.15529895e-01 -1.10094093e-01 9.04680252e-01 -7.12541267e-02 -6.34878933e-01 3.20905864e-01 -7.61549501e-03 5.72302639e-01 9.82383788e-01 8.07734281e-02 -4.63890910e-01 -4.43443537e-01 -6.47172213e-01 3.61304760e-01 4.99463141e-01 6.60712957e-01 9.14995909e-01 -1.66365361e+00 -7.10661709e-01 4.27175313e-02 4.60403562e-01 -4.14709300e-01 8.58418524e-01 1.32101250e+00 -9.58480388e-02 -3.55611831e-01 -2.42738649e-01 -5.01138866e-01 -1.24490321e+00 6.72002077e-01 3.51287186e-01 -1.13624312e-01 -5.07906437e-01 8.71957958e-01 4.75906670e-01 -3.56556535e-01 -1.64219588e-01 4.87503916e-01 -4.08012986e-01 4.87720966e-01 4.32705194e-01 -1.35360226e-01 -3.58612984e-01 -1.04126096e+00 -4.34638619e-01 4.96015579e-01 -1.29295319e-01 -1.83504090e-01 1.07469475e+00 -5.48037469e-01 -3.69623154e-01 9.61518586e-01 1.49963605e+00 1.06217034e-01 -9.61681604e-01 -3.31809282e-01 -4.32747543e-01 -3.62838626e-01 9.45526138e-02 -6.13957167e-01 -1.40385818e+00 1.03365326e+00 6.59153283e-01 5.03126346e-02 1.16629934e+00 2.88636297e-01 9.34282959e-01 1.51615804e-02 -3.65635574e-01 -1.50714183e+00 6.64696574e-01 2.33870253e-01 1.01750672e+00 -1.57292879e+00 -5.59348911e-02 -1.10207453e-01 -1.40283990e+00 9.80646133e-01 9.46376622e-01 -3.04722369e-01 8.48317087e-01 2.02622280e-01 3.84595901e-01 -6.83976293e-01 -6.77924573e-01 1.85851559e-01 1.28531888e-01 1.89214408e-01 4.13150460e-01 1.50538739e-02 -7.19686747e-01 9.07156110e-01 -2.67872512e-01 -3.59306246e-01 4.41538811e-01 2.12694034e-01 -3.00843179e-01 -6.26772523e-01 -1.08735159e-01 3.25526029e-01 -3.22740257e-01 -3.59877855e-01 -2.70853609e-01 3.32696646e-01 1.33462206e-01 8.29700589e-01 3.92799467e-01 -5.22559524e-01 8.22933465e-02 -6.54694065e-02 1.46175161e-01 -2.38042250e-01 -4.88621742e-01 -2.36463011e-03 4.89424430e-02 -5.34975410e-01 -6.04281902e-01 -5.23140609e-01 -1.27880955e+00 -2.47904629e-01 -1.65691927e-01 2.83895254e-01 5.10745287e-01 7.76295602e-01 3.21072876e-01 2.64109790e-01 8.54382336e-01 -3.39909047e-01 -6.24436140e-01 -8.81004870e-01 -7.23024249e-01 1.11173451e+00 -8.70700628e-02 -6.73456550e-01 -6.38070524e-01 -1.14694700e-01]
[13.076189994812012, 5.0307159423828125]
c2ebc69d-a694-48f4-ba49-80bb625594f2
codetrek-flexible-modeling-of-code-using-an
null
null
https://openreview.net/forum?id=WQc075jmBmf
https://openreview.net/pdf?id=WQc075jmBmf
CodeTrek: Flexible Modeling of Code using an Extensible Relational Representation
Designing a suitable representation for code-reasoning tasks is challenging in aspects such as the kinds of program information to model, how to combine them, and how much context to consider. We propose CodeTrek, a deep learning approach that addresses these challenges by representing codebases as databases that conform to rich relational schemas. The relational representation not only allows CodeTrek to uniformly represent diverse kinds of program information, but also to leverage program-analysis queries to derive new semantic relations, which can be readily incorporated without further architectural engineering. CodeTrek embeds this relational representation using a set of walks that can traverse different relations in an unconstrained fashion, and incorporates all relevant attributes along the way. We evaluate CodeTrek on four diverse and challenging Python tasks: variable misuse, exception prediction, unused definition, and variable shadowing. CodeTrek achieves an accuracy of 91%, 63%, 98%, and 94% on these tasks respectively, and outperforms state-of-the-art neural models by 2-19% points.
['Mayur Naik', 'Petros Maniatis', 'Hanjun Dai', 'Yuepeng Wang', 'Aaditya Naik', 'Pardis Pashakhanloo']
2021-09-29
null
null
null
iclr-2022-4
['variable-misuse', 'exception-type']
['computer-code', 'computer-code']
[-2.81601608e-01 7.79276341e-02 -6.96380556e-01 -6.71655655e-01 -5.25257587e-01 -5.32330334e-01 3.20570230e-01 6.14482701e-01 -1.56806987e-02 2.43551999e-01 6.72694743e-02 -7.04890072e-01 -8.16400629e-03 -1.19610119e+00 -9.01534617e-01 1.72423616e-01 -2.84458827e-02 1.84035093e-01 5.41334093e-01 -3.38733107e-01 1.19884528e-01 2.07890585e-01 -1.59753573e+00 7.28309035e-01 8.52345169e-01 8.61002505e-01 -9.87066776e-02 5.44264555e-01 -5.00494242e-01 1.36802697e+00 -4.64075327e-01 -7.25347579e-01 -6.12447560e-02 2.01354355e-01 -8.74170482e-01 -6.80363774e-01 2.59671837e-01 -2.33005673e-01 -2.62649477e-01 8.55231464e-01 1.46245331e-01 -1.21167943e-01 2.37129137e-01 -1.22145772e+00 -8.39869022e-01 1.17245257e+00 -4.63343859e-01 4.52103093e-02 3.35456640e-01 2.18490630e-01 1.19754577e+00 -6.69312119e-01 8.87617171e-01 1.10202622e+00 1.02594519e+00 4.25105840e-01 -1.43836379e+00 -5.54773211e-01 6.06525913e-02 7.71575794e-02 -1.26885021e+00 -5.47096372e-01 4.40490991e-01 -7.66071856e-01 1.67707086e+00 4.25854266e-01 4.03003722e-01 8.97579014e-01 3.74805927e-02 5.54486215e-01 6.08743429e-01 -1.86052501e-01 1.15957163e-01 1.74534261e-01 8.10838461e-01 1.07455146e+00 3.92083347e-01 -2.85892934e-01 -3.95095825e-01 -4.32728440e-01 2.80952841e-01 2.06087649e-01 -2.24152654e-02 -5.84567130e-01 -1.01221085e+00 6.50695920e-01 6.84003413e-01 -3.16468161e-03 -8.91528726e-02 5.26508510e-01 9.45752501e-01 3.79328460e-01 -1.72113068e-02 7.31774509e-01 -8.53145957e-01 -3.89438033e-01 -5.98700285e-01 3.61072153e-01 1.03442717e+00 1.49684930e+00 1.05605781e+00 -2.18451070e-03 -1.16273068e-01 9.07111645e-01 4.89606500e-01 2.45262727e-01 3.51274192e-01 -8.66872132e-01 9.16738570e-01 1.25597227e+00 -3.13632220e-01 -9.01347756e-01 -2.73028165e-01 -2.86221176e-01 -3.61949414e-01 1.22959666e-01 9.81886089e-02 1.39301002e-01 -6.01136506e-01 1.56984949e+00 6.77858442e-02 -3.56642425e-01 3.14523019e-02 3.99477571e-01 1.06768310e+00 2.27876708e-01 1.17755659e-01 5.38299799e-01 1.18928874e+00 -9.90994275e-01 -1.92990825e-01 -4.71904635e-01 1.07145739e+00 -3.31799805e-01 1.30161762e+00 1.41365379e-01 -9.07953203e-01 -2.56528020e-01 -1.18856251e+00 -4.85315531e-01 -5.73364019e-01 -1.43733639e-02 1.01143169e+00 5.46046078e-01 -9.60003018e-01 6.11559689e-01 -9.34591770e-01 -2.68632919e-01 4.95642006e-01 2.11049616e-01 -3.70519012e-01 -3.64043601e-02 -7.77754664e-01 7.26445675e-01 4.51435208e-01 -2.75017858e-01 -7.63242662e-01 -1.08146894e+00 -1.04386735e+00 3.60547960e-01 5.19524753e-01 -6.12541914e-01 1.37267268e+00 -4.74470705e-01 -1.00530934e+00 8.11834753e-01 -1.04019545e-01 -4.01549786e-01 2.56947607e-01 -1.94568098e-01 -3.85273546e-01 -4.85847175e-01 1.75546795e-01 -1.05862580e-02 1.53009504e-01 -1.04549789e+00 -3.80180508e-01 -5.07201612e-01 5.19207358e-01 -4.26159352e-01 -1.44842461e-01 2.25747392e-01 -9.10024822e-01 -3.73785675e-01 -1.93545952e-01 -8.06858599e-01 -1.04150280e-01 2.42448121e-01 -5.52610636e-01 -2.63371676e-01 5.36022902e-01 -7.27791727e-01 1.50260115e+00 -2.22430682e+00 1.32668018e-01 1.25100657e-01 5.78963399e-01 1.43490717e-01 3.26955728e-02 3.73696864e-01 7.50286430e-02 4.92973596e-01 -4.16071475e-01 -1.91344023e-02 4.55347955e-01 3.42680752e-01 -3.76158148e-01 -7.62086138e-02 2.92641640e-01 8.86410654e-01 -8.28235745e-01 -4.90131944e-01 -9.84884575e-02 3.37532818e-01 -9.81942058e-01 1.75654143e-01 -9.09837842e-01 -4.59486425e-01 -5.46151578e-01 8.19660842e-01 4.96624738e-01 -3.65109742e-01 4.86247241e-01 -9.73482355e-02 7.13358447e-03 6.79509819e-01 -9.69665945e-01 1.64430118e+00 -8.75064969e-01 6.57502532e-01 -2.79164553e-01 -5.68723083e-01 1.15612841e+00 -1.33640468e-01 -1.06322626e-02 -7.79619157e-01 -4.48053449e-01 1.74598277e-01 -2.36970708e-01 -8.77904892e-01 7.09381104e-01 5.23782849e-01 -5.64586282e-01 5.48090518e-01 -1.12870768e-01 1.30798250e-01 1.83683768e-01 2.80242473e-01 1.58796275e+00 3.95524383e-01 4.44635630e-01 -1.63473397e-01 2.84626603e-01 -1.56932175e-02 8.13785732e-01 8.02156210e-01 2.11969152e-01 1.57755449e-01 1.20045829e+00 -7.86792934e-01 -8.41795206e-01 -1.00604904e+00 -9.49171409e-02 1.42564249e+00 3.04738875e-03 -1.07899225e+00 -4.62806165e-01 -9.10552859e-01 3.75538290e-01 8.58228862e-01 -6.61525428e-01 -1.70689359e-01 -7.04656780e-01 -5.73904395e-01 9.07574534e-01 8.94600034e-01 3.51366609e-01 -8.65649223e-01 -7.08578587e-01 1.50267884e-01 -1.46513144e-02 -8.68292391e-01 -6.17117137e-02 5.16851008e-01 -6.29606247e-01 -1.50209188e+00 3.26765478e-01 -4.05228764e-01 3.65833163e-01 -6.86298609e-02 1.88150632e+00 5.08288085e-01 -1.86129987e-01 -2.96550803e-02 -1.36505678e-01 -7.90692940e-02 -7.16065347e-01 4.00648624e-01 -6.40286326e-01 -7.13374555e-01 6.93384469e-01 -6.83813632e-01 -1.87079459e-01 2.16670766e-01 -8.35660040e-01 1.10867275e-02 2.94487178e-01 6.05672598e-01 4.63897407e-01 -1.03140652e-01 3.11865360e-01 -1.46830702e+00 4.19477463e-01 -8.68133843e-01 -8.15088034e-01 5.72954476e-01 -7.31792271e-01 5.72817624e-01 8.66311669e-01 -7.59932175e-02 -1.00606287e+00 -1.77026123e-01 -1.50379434e-01 -2.50339627e-01 9.14436355e-02 8.15593898e-01 -2.64051825e-01 2.73040477e-02 1.12718880e+00 -5.15666939e-02 -4.08369824e-02 -6.69391692e-01 6.11516714e-01 6.56247795e-01 5.19832671e-01 -1.19258237e+00 5.11391878e-01 7.23473877e-02 -2.89632767e-01 -3.58808011e-01 -3.08441341e-01 -1.16644390e-01 -5.88900208e-01 3.69002998e-01 4.66387302e-01 -7.78884828e-01 -7.26557672e-01 2.28945404e-01 -1.10287881e+00 -6.39114738e-01 -2.28621855e-01 -1.78512543e-01 -3.57507646e-01 1.54124796e-01 -6.69341803e-01 -2.96094745e-01 -4.55210239e-01 -1.67503333e+00 8.07070196e-01 -1.51350275e-01 -3.36065859e-01 -7.87833154e-01 -5.61790317e-02 2.51638412e-01 7.12356329e-01 4.10669953e-01 1.73194420e+00 -7.55822897e-01 -1.13547468e+00 -8.32651109e-02 -4.78399009e-01 5.71389087e-02 -1.51707456e-01 3.05892318e-01 -7.02749133e-01 9.22808796e-02 -6.70197427e-01 -4.02365863e-01 6.39499724e-01 -4.91656065e-01 1.49939215e+00 -4.03991073e-01 -5.29819787e-01 1.05056131e+00 1.61618745e+00 1.46924183e-01 6.17967904e-01 5.88450849e-01 8.83479536e-01 2.74845511e-01 -8.13820213e-02 4.26063567e-01 9.62985516e-01 7.99636781e-01 6.97504580e-01 4.06305581e-01 -1.10084154e-01 -2.50289649e-01 1.01115450e-01 8.20859551e-01 3.73548895e-01 3.07211488e-01 -1.44565010e+00 6.29182935e-01 -1.98314726e+00 -8.04421604e-01 -2.42843986e-01 1.99968338e+00 1.18752265e+00 8.03487226e-02 -5.01189604e-02 -1.68751225e-01 3.74071777e-01 1.50279835e-01 -8.23185146e-01 -7.18797624e-01 1.80108070e-01 1.56492665e-01 4.25563544e-01 1.85727254e-01 -9.35265839e-01 8.81608307e-01 6.07033014e+00 3.10829937e-01 -1.13285661e+00 -8.92053265e-03 2.65657336e-01 1.28048256e-01 -6.70894444e-01 2.79797763e-01 -7.85035253e-01 4.40992653e-01 9.51117992e-01 -3.88162106e-01 8.47969592e-01 1.41745472e+00 -5.13604045e-01 1.85761258e-01 -1.52378094e+00 6.72795475e-01 -1.31699383e-01 -1.50975966e+00 -1.04555927e-01 -2.86328048e-01 4.50601429e-01 4.92328405e-01 -2.14856088e-01 9.04000461e-01 8.91622961e-01 -1.11437583e+00 8.08502734e-01 7.47749805e-01 6.50642633e-01 -4.74316835e-01 4.39464778e-01 -3.14132422e-02 -1.27114975e+00 -2.44478360e-01 -2.72135407e-01 1.31292477e-01 -6.81774259e-01 4.57080394e-01 -8.52068663e-01 4.74822134e-01 9.37672615e-01 8.56774151e-01 -1.10098445e+00 9.47289705e-01 -2.72126496e-01 3.28193337e-01 -2.35107824e-01 7.56549044e-03 -1.88515738e-01 3.84566933e-01 6.57264963e-02 1.54296434e+00 1.83419228e-01 -4.07811850e-01 9.85775441e-02 1.49044955e+00 -5.35194218e-01 -1.07841276e-01 -6.48457825e-01 1.10981502e-02 9.39825773e-01 9.92132008e-01 -1.93792403e-01 -3.44229549e-01 -7.63044715e-01 3.03505391e-01 8.29770029e-01 4.19936776e-01 -7.55744100e-01 -7.93612957e-01 1.04757321e+00 4.04981486e-02 3.91232312e-01 -2.73460224e-02 -5.23734093e-01 -1.32179117e+00 5.17272353e-01 -1.23229134e+00 4.58544344e-01 -7.50328183e-01 -9.99391973e-01 7.21357703e-01 2.00063866e-02 -5.79315424e-01 -3.25935572e-01 -4.46716040e-01 -4.32853222e-01 9.85879660e-01 -1.34912252e+00 -1.16400862e+00 -4.55937803e-01 3.22726965e-01 7.29671568e-02 -3.74615967e-01 8.05239737e-01 4.43065435e-01 -7.32480884e-01 9.54693973e-01 1.17269076e-01 4.94768053e-01 4.09267664e-01 -1.34389198e+00 1.03663766e+00 9.16973770e-01 -2.45102257e-01 1.25291598e+00 3.66464645e-01 -4.05473381e-01 -1.85947704e+00 -1.48822725e+00 6.90823734e-01 -8.10830593e-01 8.40393841e-01 -4.53268558e-01 -1.34351885e+00 1.11744535e+00 -2.26383373e-01 4.72226828e-01 7.04929769e-01 5.41592479e-01 -1.43030131e+00 -3.26107770e-01 -1.02673066e+00 5.93599081e-01 1.06010103e+00 -1.00029612e+00 -5.80062985e-01 3.06890495e-02 1.11818445e+00 -6.49789453e-01 -1.42160118e+00 1.67186543e-01 7.02184916e-01 -1.07205188e+00 1.00337148e+00 -8.74726534e-01 9.34422851e-01 -3.19183648e-01 -5.54683685e-01 -9.83762920e-01 -1.64908037e-01 -2.69691914e-01 -5.59863865e-01 1.30788291e+00 5.68810523e-01 -5.45535088e-01 5.94536602e-01 9.84157145e-01 -2.68457890e-01 -1.00025976e+00 -3.84898275e-01 -5.26429236e-01 9.98423845e-02 -7.71633148e-01 1.48645365e+00 1.21423209e+00 2.11848572e-01 1.11388341e-01 1.14827268e-01 1.65146172e-01 3.88125420e-01 4.80003834e-01 1.01071405e+00 -1.24277604e+00 -6.08604372e-01 -7.24817872e-01 -4.30975348e-01 -7.14869797e-01 3.97398949e-01 -1.34337616e+00 -3.11279655e-01 -1.63915753e+00 1.89492017e-01 -9.36036170e-01 -2.39597738e-01 1.03920460e+00 -1.08214788e-01 -4.35059786e-01 1.54299483e-01 2.53720105e-01 -6.83179617e-01 1.31877303e-01 4.34569329e-01 -4.75450516e-01 -8.83156285e-02 -1.98807016e-01 -1.09570789e+00 6.08112752e-01 4.40890819e-01 -4.39989418e-01 -3.66301984e-01 -1.11715925e+00 6.97956562e-01 3.91714543e-01 3.45931858e-01 -8.67292643e-01 2.39427447e-01 -2.64116675e-01 -1.02161039e-02 -2.69035995e-01 1.02338098e-01 -6.33578300e-01 3.60210389e-01 5.99446520e-02 -6.92478299e-01 3.00574541e-01 3.45020980e-01 4.27280992e-01 3.85086983e-02 -2.71354169e-01 5.75783670e-01 -2.92225003e-01 -1.00497448e+00 2.05374971e-01 -5.55809252e-02 3.76856536e-01 8.11365962e-01 1.84034094e-01 -9.75087225e-01 2.86173999e-01 -3.32453668e-01 2.55962789e-01 8.20646942e-01 7.65726626e-01 3.12625110e-01 -1.17365789e+00 -1.84737399e-01 3.78195286e-01 7.27177441e-01 1.43061027e-01 -9.07607824e-02 3.06154966e-01 -8.21111917e-01 2.86127239e-01 -8.18472058e-02 -3.63630146e-01 -1.03708720e+00 7.59888887e-01 5.28598905e-01 -2.80226260e-01 -7.59226084e-01 5.87607205e-01 -2.17105240e-01 -1.10918915e+00 2.24510744e-01 -7.48363495e-01 9.68811214e-02 -3.34726006e-01 4.50003237e-01 1.74489662e-01 4.27963704e-01 -2.24907756e-01 -5.48670769e-01 3.34721178e-01 -2.89083302e-01 6.21376395e-01 1.42903304e+00 4.20425594e-01 -5.96657217e-01 3.63979846e-01 1.32021821e+00 -6.90300986e-02 -9.31951940e-01 -6.35382354e-01 5.28719246e-01 -5.94452858e-01 -1.27427116e-01 -1.16517103e+00 -1.14411759e+00 8.96479130e-01 -1.25508085e-02 2.89102137e-01 7.40121782e-01 -1.41613316e-02 6.27166033e-01 8.34859729e-01 6.26904070e-01 -5.70542455e-01 -2.35685602e-01 7.03343093e-01 6.75691307e-01 -9.95080173e-01 -4.69235592e-02 -2.65069067e-01 -2.56013393e-01 1.23271215e+00 8.98194730e-01 1.61105439e-01 4.69009876e-01 7.18565822e-01 -1.90083697e-01 -3.56540740e-01 -1.21617889e+00 1.00645438e-01 -1.06203735e-01 6.68581426e-01 8.22284341e-01 1.52641043e-01 2.34604746e-01 9.08649921e-01 -1.14613503e-01 1.83744252e-01 5.57045877e-01 1.05775499e+00 -7.15269819e-02 -1.19259727e+00 -4.03725430e-02 8.49400878e-01 -3.18446100e-01 -2.34397009e-01 -1.72839925e-01 8.75161350e-01 -8.01028386e-02 4.75838780e-01 -7.33911321e-02 -5.98280013e-01 5.49211979e-01 8.59567076e-02 8.91717076e-02 -7.39008248e-01 -9.48767602e-01 -7.86769569e-01 3.71635586e-01 -8.15176189e-01 3.03232968e-01 -4.44021285e-01 -1.26867151e+00 -5.50761819e-01 6.40632063e-02 4.65356885e-03 6.24791443e-01 6.20046020e-01 8.02584708e-01 9.54097450e-01 1.25245571e-01 -1.32597670e-01 -6.51798964e-01 -4.44377542e-01 4.69129570e-02 5.02982736e-01 3.66149396e-01 -4.58958685e-01 4.13373634e-02 -8.20905045e-02]
[7.621532440185547, 7.829018592834473]
33531134-5b9b-4ddb-a625-86948ef7222c
improving-non-autoregressive-generation-with
2110.11115
null
https://arxiv.org/abs/2110.11115v1
https://arxiv.org/pdf/2110.11115v1.pdf
Improving Non-autoregressive Generation with Mixup Training
While pre-trained language models have achieved great success on various natural language understanding tasks, how to effectively leverage them into non-autoregressive generation tasks remains a challenge. To solve this problem, we present a non-autoregressive generation model based on pre-trained transformer models. To bridge the gap between autoregressive and non-autoregressive models, we propose a simple and effective iterative training method called MIx Source and pseudo Target (MIST). Unlike other iterative decoding methods, which sacrifice the inference speed to achieve better performance based on multiple decoding iterations, MIST works in the training stage and has no effect on inference time. Our experiments on three generation benchmarks including question generation, summarization and paraphrase generation, show that the proposed framework achieves the new state-of-the-art results for fully non-autoregressive models. We also demonstrate that our method can be used to a variety of pre-trained models. For instance, MIST based on the small pre-trained model also obtains comparable performance with seq2seq models.
['Qi Zhang', 'Liangjie Zhang', 'Haizhen Huang', 'Furu Wei', 'Fuzhen Zhuang', 'Deqing Wang', 'Zihan Zhang', 'Shaohan Huang', 'Ting Jiang']
2021-10-21
null
null
null
null
['paraphrase-generation', 'paraphrase-generation']
['computer-code', 'natural-language-processing']
[ 4.81725395e-01 4.24630016e-01 -5.95755987e-02 -3.53064865e-01 -1.29219544e+00 -4.47356373e-01 9.69088912e-01 -2.29055032e-01 -2.14902498e-02 8.03444028e-01 7.22795069e-01 -6.28437936e-01 3.82972091e-01 -9.13870752e-01 -7.68089712e-01 -3.61711204e-01 5.75483322e-01 7.43446350e-01 3.39522921e-02 -5.18024325e-01 1.69704050e-01 -1.74306393e-01 -1.11417794e+00 5.70580006e-01 1.18064451e+00 6.19106174e-01 1.87234417e-01 9.34949994e-01 -4.80421901e-01 1.15440083e+00 -7.90930688e-01 -6.83208883e-01 -8.34598392e-02 -9.26556945e-01 -8.37386012e-01 -3.32925349e-01 2.37243071e-01 -4.95748103e-01 -1.81024492e-01 5.87698579e-01 5.98219812e-01 1.02653094e-01 8.63347590e-01 -8.21173608e-01 -8.82945418e-01 1.20516407e+00 -3.14596236e-01 1.27098143e-01 4.97741193e-01 -8.86653289e-02 1.01957309e+00 -8.61817718e-01 2.62443990e-01 1.44989407e+00 6.60758555e-01 6.84170187e-01 -1.16194510e+00 -7.46823370e-01 1.77171603e-01 -3.87266651e-02 -8.30523849e-01 -6.41308427e-01 6.24184668e-01 -2.14080155e-01 1.20025539e+00 1.57434314e-01 3.03471714e-01 1.47926521e+00 2.54401058e-01 9.45249975e-01 9.98704314e-01 -5.34789026e-01 3.67253758e-02 -7.01414496e-02 2.19426006e-01 4.70539600e-01 -1.86926704e-02 -9.17702094e-02 -6.40987754e-01 -1.28903478e-01 6.20472074e-01 -3.27136993e-01 -1.97010487e-01 3.22787583e-01 -1.10598898e+00 1.06585932e+00 7.37593621e-02 2.20925570e-01 -4.48332936e-01 3.22711051e-01 3.85403216e-01 2.67910033e-01 7.75847673e-01 5.66677809e-01 -3.32815707e-01 -6.79518938e-01 -1.28753078e+00 2.17637360e-01 8.95476878e-01 1.07070684e+00 4.50427860e-01 5.64473629e-01 -6.94827318e-01 1.00234187e+00 1.92352384e-01 5.07458985e-01 8.72905374e-01 -6.45257056e-01 6.77658975e-01 3.76125127e-01 -1.70147568e-01 -4.81250942e-01 1.32238632e-03 -5.57807684e-01 -1.09874141e+00 -5.12613833e-01 1.79843605e-01 -2.99519032e-01 -1.24984598e+00 1.63756585e+00 -1.78211004e-01 2.29948372e-01 4.46705818e-01 4.84173268e-01 1.00191104e+00 1.11102295e+00 2.15744972e-02 -1.90125570e-01 1.17997003e+00 -1.41936624e+00 -8.65515411e-01 -3.92510563e-01 5.84906161e-01 -8.60868692e-01 1.02155447e+00 1.34555548e-01 -1.43856573e+00 -5.26841402e-01 -7.45929301e-01 -3.18021625e-01 -1.43779933e-01 3.04477572e-01 7.03294218e-01 6.70620024e-01 -1.06581759e+00 3.74016553e-01 -8.66472900e-01 -1.48186520e-01 2.71200445e-02 6.12220392e-02 -5.20969704e-02 1.75538920e-02 -1.33977151e+00 8.95149350e-01 3.52596730e-01 2.08530083e-01 -8.71184289e-01 -7.88557172e-01 -9.74147499e-01 2.23230332e-01 1.14073053e-01 -1.19351149e+00 1.71593845e+00 -7.94984519e-01 -2.25607753e+00 3.26889992e-01 -6.99403346e-01 -8.81801188e-01 2.83788502e-01 -6.84949934e-01 -6.32051155e-02 -1.27735317e-01 1.75763443e-02 5.90729356e-01 7.63594985e-01 -8.33720803e-01 -1.89856350e-01 1.48654007e-03 8.36251602e-02 1.77768961e-01 -2.88285106e-01 1.15893967e-01 -4.11274076e-01 -8.67867112e-01 -2.04038963e-01 -8.65209937e-01 -2.89237469e-01 -9.29327309e-01 -5.37199259e-01 -3.79804164e-01 4.49198127e-01 -8.18937004e-01 1.32740331e+00 -1.32252932e+00 2.48969331e-01 -3.04806709e-01 -1.76208898e-01 4.28393811e-01 -3.19682181e-01 7.64221728e-01 -2.23548971e-02 1.72455207e-01 -3.86907220e-01 -6.23430371e-01 -4.73746322e-02 2.50770867e-01 -9.33702767e-01 -3.64799768e-01 4.75672513e-01 1.27896690e+00 -1.05234063e+00 -3.25598300e-01 1.30151096e-03 3.85410994e-01 -6.58376276e-01 5.53324461e-01 -4.14611191e-01 3.07931006e-01 -4.18135583e-01 3.45346510e-01 4.03538585e-01 -2.96566695e-01 1.70910582e-01 1.13099605e-01 1.28093421e-01 8.87194097e-01 -5.05607009e-01 1.69816971e+00 -8.00019324e-01 5.40756166e-01 -4.36492592e-01 -1.09789705e+00 9.75669980e-01 4.95003670e-01 -1.18920438e-01 -5.81434846e-01 -1.04596123e-01 2.05340937e-01 -2.15466525e-02 -1.75697401e-01 1.00796175e+00 -2.39916012e-01 -6.65532425e-02 6.77492380e-01 3.12858284e-01 -3.89378607e-01 2.94895977e-01 4.51926917e-01 1.04434526e+00 4.04153466e-01 3.06417674e-01 1.09706372e-01 4.51671302e-01 -1.01163879e-01 4.47716653e-01 1.10474658e+00 5.14521122e-01 8.53830218e-01 4.86638814e-01 -2.66471468e-02 -9.85521019e-01 -1.11969066e+00 3.60040486e-01 1.25477040e+00 -3.59805912e-01 -7.74037004e-01 -8.72071266e-01 -5.80702126e-01 -4.94114876e-01 1.29272270e+00 -3.33253205e-01 -2.99567074e-01 -8.03989768e-01 -8.68373573e-01 8.66439402e-01 7.59653687e-01 5.30859888e-01 -1.07195735e+00 -1.32371396e-01 4.53355998e-01 -5.78625083e-01 -1.14569783e+00 -4.87076908e-01 -1.54212251e-01 -9.61554587e-01 -4.16684896e-01 -8.27895999e-01 -5.94848514e-01 4.52635735e-01 1.65346771e-01 1.48599565e+00 -2.11302966e-01 3.64149362e-01 2.98901200e-01 -5.94046593e-01 -5.97817481e-01 -9.76812065e-01 5.84278643e-01 -3.98289680e-01 -2.28825361e-01 1.27469748e-01 -6.79745793e-01 -2.80896664e-01 -3.39795500e-02 -8.80309820e-01 5.34491777e-01 7.68587470e-01 1.09358168e+00 3.75005454e-01 -5.14011145e-01 9.88668203e-01 -1.13129163e+00 1.06456757e+00 -4.90199983e-01 -3.07801366e-01 3.97973567e-01 -5.66961706e-01 4.85321701e-01 9.28534806e-01 -4.02614295e-01 -1.37965751e+00 -2.16230154e-01 -4.66840953e-01 -1.44956052e-01 2.32765704e-01 7.92118788e-01 1.57736853e-01 5.84070265e-01 4.89293158e-01 6.83292329e-01 -2.87346318e-02 -5.67208529e-01 7.15576053e-01 6.87334061e-01 5.24981856e-01 -4.80317324e-01 8.61796439e-01 -6.85577542e-02 -3.03044230e-01 -5.27497649e-01 -9.66744661e-01 -2.47705445e-01 -2.09560096e-01 7.98295438e-02 6.16954088e-01 -1.03100073e+00 -7.59673938e-02 4.95921969e-01 -1.58019114e+00 -3.67904216e-01 -2.74944901e-01 3.33574980e-01 -6.95641100e-01 4.03537303e-01 -7.35410571e-01 -9.29753542e-01 -1.04398358e+00 -9.23216164e-01 1.30975842e+00 1.10058255e-01 -3.59953195e-01 -1.06911004e+00 3.39569092e-01 4.66946810e-01 8.21003854e-01 -2.96761006e-01 1.09737146e+00 -8.32229912e-01 -6.00480855e-01 -2.01815873e-01 7.65333399e-02 3.97374511e-01 9.92401540e-02 -1.46608442e-01 -8.37413728e-01 1.09518781e-01 -5.87722566e-03 -4.88489300e-01 1.16318393e+00 2.61060119e-01 1.07585669e+00 -5.46256900e-01 -5.01561165e-02 4.47261661e-01 8.87866735e-01 -1.87917277e-02 1.11914372e+00 2.32201815e-02 5.81243157e-01 2.96627969e-01 4.06414658e-01 1.84912458e-01 6.63126528e-01 7.13834763e-01 7.88408443e-02 -9.85241123e-03 -1.79068670e-01 -8.13034713e-01 8.14061046e-01 1.38877559e+00 -1.54495269e-01 -5.40135264e-01 -7.37084508e-01 5.54657817e-01 -1.90839326e+00 -1.16291118e+00 -1.92600265e-01 2.18873191e+00 1.16655934e+00 6.44624140e-03 -7.60514438e-02 -2.52182722e-01 5.02718806e-01 2.78750181e-01 -2.32905552e-01 -8.44041526e-01 -9.96102467e-02 6.03875637e-01 1.32732093e-01 4.68046635e-01 -8.07560265e-01 1.31264985e+00 7.04104900e+00 1.05670714e+00 -1.04804349e+00 1.79296404e-01 6.42605364e-01 4.95013744e-02 -6.41832650e-01 1.31085992e-01 -1.05919957e+00 4.77874964e-01 1.50384641e+00 -5.69939554e-01 3.22909862e-01 8.15219998e-01 3.59720021e-01 1.86712801e-01 -1.09143627e+00 6.99391723e-01 2.79997289e-01 -1.40460527e+00 6.20806634e-01 -2.23351091e-01 8.62078249e-01 -6.91568032e-02 -2.95147859e-02 1.00874352e+00 5.94477713e-01 -1.21752119e+00 4.95660186e-01 4.98117387e-01 5.92981279e-01 -4.73397464e-01 6.68308556e-01 6.12276673e-01 -9.37799752e-01 7.04339221e-02 -5.43911040e-01 -1.04975060e-01 6.71989024e-01 6.28501832e-01 -1.13422859e+00 8.84451985e-01 2.45803166e-02 6.08127356e-01 -3.56344014e-01 7.47634888e-01 -6.10959351e-01 1.19768906e+00 -9.19995308e-02 -4.30312082e-02 2.24565178e-01 -2.92114735e-01 5.89328170e-01 1.25948632e+00 5.61493039e-01 2.13716067e-02 -1.75605804e-01 8.47709298e-01 -3.09187949e-01 1.56724229e-01 -6.09768331e-01 -2.83221990e-01 3.04999650e-01 1.08749187e+00 -1.04270920e-01 -7.25381136e-01 -3.06260407e-01 1.05530047e+00 4.27144796e-01 3.77045751e-01 -1.03208184e+00 -3.00880760e-01 2.99387962e-01 9.30556804e-02 2.74194032e-01 -2.71420181e-01 -1.32851660e-01 -1.35975528e+00 -2.92567518e-02 -1.23056173e+00 2.53279746e-01 -9.80588794e-01 -1.16416025e+00 8.70463431e-01 1.46572217e-01 -1.01305878e+00 -1.30273557e+00 -1.59050241e-01 -8.81122887e-01 9.62425947e-01 -1.77188039e+00 -1.31954896e+00 -8.77030790e-02 1.89071238e-01 1.11772275e+00 -2.31318340e-01 1.03613257e+00 -7.55038708e-02 -4.46552962e-01 7.38325655e-01 1.70745134e-01 -1.13357995e-02 7.22316206e-01 -1.25485301e+00 8.48286986e-01 9.42377687e-01 2.49975100e-01 7.96718001e-01 5.81013441e-01 -6.11903667e-01 -1.32911801e+00 -1.31823909e+00 1.22682750e+00 -3.96663994e-01 5.26619017e-01 -4.16009754e-01 -8.47746015e-01 9.22389984e-01 6.48035944e-01 -7.68036902e-01 4.78911370e-01 2.18161255e-01 -3.40134889e-01 1.01123899e-01 -4.73308593e-01 7.52918005e-01 8.58182967e-01 -4.50748593e-01 -8.55048835e-01 4.99088764e-01 1.05427504e+00 -5.18495679e-01 -5.65367103e-01 4.42363530e-01 4.77272183e-01 -7.68028855e-01 7.74787366e-01 -8.80142927e-01 1.14091170e+00 1.87355369e-01 1.67801306e-01 -1.59065986e+00 -1.66932404e-01 -1.08372521e+00 -3.56507093e-01 1.42831612e+00 7.57323146e-01 -6.77798569e-01 6.00166738e-01 4.41624045e-01 -4.45219636e-01 -7.98721075e-01 -6.95365310e-01 -8.18449676e-01 3.72141063e-01 -3.75512183e-01 5.92588246e-01 4.23643112e-01 -9.88698751e-02 9.44894910e-01 -7.22976804e-01 -2.86300510e-01 2.33899489e-01 2.38644019e-01 1.09768784e+00 -8.07284713e-01 -5.60303926e-01 -2.51665622e-01 1.23642206e-01 -1.77495933e+00 4.31636661e-01 -9.20612454e-01 2.38150701e-01 -1.80405188e+00 3.14888000e-01 -6.23023137e-02 1.69396728e-01 4.28369641e-01 -5.78770876e-01 2.11876072e-02 9.29395705e-02 9.91593674e-03 -3.58373344e-01 9.43690956e-01 1.03169703e+00 -3.99092212e-02 -1.87061384e-01 1.53764084e-01 -9.31570053e-01 4.52272773e-01 8.08377504e-01 -3.65016192e-01 -6.69439197e-01 -6.86505079e-01 1.85660183e-01 2.38599122e-01 -3.06504387e-02 -6.94470286e-01 1.98029935e-01 1.70279324e-01 -3.20117213e-02 -6.59585238e-01 5.30819654e-01 -5.72489435e-03 4.67438474e-02 1.93048656e-01 -6.94125533e-01 3.13641846e-01 1.59744978e-01 5.71500242e-01 -3.62461537e-01 -4.95756269e-01 3.73660922e-01 -1.57974720e-01 -1.37058944e-01 3.68289575e-02 -6.74236894e-01 2.82552451e-01 4.15059686e-01 -3.67333740e-02 -5.46941102e-01 -1.03249657e+00 -2.97937244e-01 1.00504145e-01 -1.74613558e-02 5.62442005e-01 6.24726951e-01 -1.09943318e+00 -1.13933361e+00 5.63128255e-02 -2.12653875e-01 -9.16873515e-02 1.01247162e-01 7.32552111e-01 -2.37152666e-01 9.06424165e-01 3.45743686e-01 -4.37679619e-01 -1.12701225e+00 1.59148723e-01 2.00427607e-01 -1.09762633e+00 -3.01231831e-01 6.62788272e-01 3.64667684e-01 -4.98814732e-01 -1.58662230e-01 -2.92194545e-01 -3.23804319e-01 -1.75144091e-01 5.51708877e-01 1.45317301e-01 1.34274974e-01 -4.02782589e-01 1.35387197e-01 2.07943544e-01 -4.74150926e-01 -2.59296298e-01 1.30892920e+00 4.66720499e-02 -1.16518535e-01 4.52767819e-01 9.66167569e-01 2.01582573e-02 -6.43024564e-01 -2.42105410e-01 -2.51552254e-01 -9.60919857e-02 1.06309541e-02 -7.93161631e-01 -7.32363164e-01 1.06822324e+00 -1.39082432e-01 1.69028863e-01 1.06699657e+00 -2.12697104e-01 1.10668755e+00 4.74205166e-01 1.38167873e-01 -7.82903075e-01 2.37204015e-01 1.08310783e+00 1.11033845e+00 -8.45234752e-01 -2.44507238e-01 -5.54335713e-01 -8.98995459e-01 1.04279053e+00 5.79054832e-01 -4.95621786e-02 1.26517788e-01 1.85901344e-01 -1.43583804e-01 2.82907188e-01 -1.32469070e+00 -9.62883607e-02 2.22172588e-01 5.25460720e-01 8.92729044e-01 -1.78702101e-01 -4.38566059e-01 8.03693533e-01 -6.89414561e-01 1.59331307e-01 7.26845086e-01 5.31765103e-01 -1.84202597e-01 -1.43562376e+00 -1.23395868e-01 6.24187052e-01 -6.56208754e-01 -7.62314439e-01 -4.32347953e-01 4.91093338e-01 -7.98936844e-01 1.11923814e+00 9.96769145e-02 -2.24788025e-01 1.73213229e-01 3.12811196e-01 4.35050160e-01 -7.98772752e-01 -7.93380857e-01 2.93601248e-02 5.07755280e-01 -2.59138554e-01 -3.10088009e-01 -2.81419545e-01 -8.17126215e-01 -1.60779104e-01 -5.52011311e-01 4.67239976e-01 5.58303237e-01 9.19206619e-01 6.90038443e-01 6.81865275e-01 5.20596266e-01 -6.61515772e-01 -1.02046621e+00 -1.42451596e+00 2.81107333e-02 1.31072685e-01 1.80887487e-02 -1.86118498e-01 -1.60881668e-01 2.46409267e-01]
[11.840886116027832, 9.084722518920898]
d13a4d47-5e7b-41eb-8a7f-eb11f2f856e4
emergent-resource-exchange-and-tolerated
2307.01862
null
https://arxiv.org/abs/2307.01862v1
https://arxiv.org/pdf/2307.01862v1.pdf
Emergent Resource Exchange and Tolerated Theft Behavior using Multi-Agent Reinforcement Learning
For decades, the evolution of cooperation has piqued the interest of numerous academic disciplines such as game theory, economics, biology, and computer science. In this work, we demonstrate the emergence of a novel and effective resource exchange protocol formed by dropping and picking up resources in a foraging environment. This form of cooperation is made possible by the introduction of a campfire, which adds an extended period of congregation and downtime for agents to explore otherwise unlikely interactions. We find that the agents learn to avoid getting cheated by their exchange partners, but not always from a third party. We also observe the emergence of behavior analogous to tolerated theft, despite the lack of any punishment, combat, or larceny mechanism in the environment.
['Jordan Pollack', 'Jack Garbus']
2023-07-04
null
null
null
null
['multi-agent-reinforcement-learning']
['methodology']
[-7.80565292e-02 1.31403059e-01 4.39136356e-01 2.29885310e-01 4.54100549e-01 -9.34272647e-01 6.38829947e-01 1.82629600e-01 -9.73212481e-01 1.35316348e+00 -3.27497631e-01 -1.23027638e-01 -3.28496814e-01 -7.42489219e-01 -9.96243134e-02 -8.60699177e-01 -6.68383479e-01 3.19963992e-01 2.16026410e-01 -7.12496817e-01 1.55040637e-01 2.94303387e-01 -1.39352310e+00 -7.40170896e-01 6.26764476e-01 3.55164289e-01 1.25602737e-01 4.91465598e-01 5.16152978e-01 5.86436033e-01 -8.45601916e-01 -5.56127787e-01 6.27604783e-01 -7.76207268e-01 -3.56694728e-01 -1.00560620e-01 -8.42198074e-01 -3.40860128e-01 -4.60344590e-02 7.71291077e-01 3.72393221e-01 3.58755201e-01 2.34153017e-01 -1.34508812e+00 -3.86475772e-01 8.25128317e-01 -6.43843293e-01 1.44138768e-01 3.83518785e-01 1.39444947e-01 8.13722014e-01 -1.94435522e-01 8.91480505e-01 8.04290056e-01 4.10220385e-01 7.57666647e-01 -8.56713057e-01 -6.52675867e-01 -3.11029196e-01 -2.62519419e-01 -1.49871325e+00 -1.85593963e-01 5.36505938e-01 3.14868912e-02 8.61212671e-01 1.87710553e-01 1.16413844e+00 9.01877046e-01 3.30956638e-01 2.47405142e-01 1.09875655e+00 -4.35759902e-01 5.86278856e-01 -1.14645042e-01 -5.13823748e-01 3.78762007e-01 8.22177231e-01 5.28136313e-01 -6.19348705e-01 -6.09159648e-01 9.00634944e-01 -5.87592125e-02 -3.51196736e-01 -2.62590408e-01 -8.19639206e-01 6.77228689e-01 3.38090211e-01 3.85515094e-01 -6.09866202e-01 -5.48347607e-02 -8.11407715e-02 7.49671400e-01 1.54073507e-01 9.03677702e-01 -3.63834202e-01 -5.28429270e-01 -2.49940440e-01 8.93466249e-02 1.11442077e+00 4.63716418e-01 6.16764665e-01 -7.78602064e-02 6.75574064e-01 3.06560010e-01 6.95777759e-02 1.62795559e-01 1.78990826e-01 -1.22541511e+00 -2.42660418e-01 5.33940554e-01 6.69586897e-01 -9.17734563e-01 -4.77708906e-01 -4.42674667e-01 -6.25899851e-01 5.54231524e-01 3.58137816e-01 -7.58425653e-01 5.60641102e-02 1.98580968e+00 4.34720963e-01 -9.26238522e-02 1.31647721e-01 6.95945203e-01 -2.16731817e-01 2.45363235e-01 -3.41731995e-01 -7.08678365e-01 6.62483811e-01 -6.41701043e-01 -3.99647743e-01 -1.41882244e-02 4.48353261e-01 -2.46301934e-01 4.32184339e-01 4.10982102e-01 -1.07539880e+00 3.97922009e-01 -9.53852355e-01 7.50941694e-01 -2.27128342e-01 -1.07768214e+00 7.85683870e-01 7.22583532e-01 -1.23197210e+00 7.98101544e-01 -7.86737978e-01 -6.26891673e-01 1.18481994e-01 4.78416085e-01 -5.74826181e-01 3.48549247e-01 -8.82441401e-01 8.93991470e-01 -7.41412640e-02 2.31573526e-02 -6.03051424e-01 2.49678129e-03 -4.22189653e-01 1.07999824e-01 6.67238533e-01 -6.48100615e-01 8.85524690e-01 -1.25120950e+00 -1.60395038e+00 8.64728689e-01 3.56374443e-01 -4.37926441e-01 5.69891751e-01 1.15532637e-01 1.41989216e-01 3.54488455e-02 2.08326593e-01 4.16736543e-01 3.76978308e-01 -1.00177503e+00 -4.44442451e-01 -3.55885178e-01 3.21273297e-01 5.54043472e-01 -1.83514476e-01 2.02220872e-01 4.70353633e-01 -4.75476593e-01 -3.44747782e-01 -9.16153491e-01 -4.95618552e-01 -2.39175498e-01 6.94238544e-02 -1.38902634e-01 4.09740865e-01 1.22602515e-01 7.38080442e-01 -2.13348317e+00 4.14703310e-01 9.89025012e-02 4.96584773e-01 1.84967201e-02 -1.79003775e-01 1.12427878e+00 5.08377552e-01 5.00251234e-01 -1.94038317e-01 -1.29290357e-01 2.13625245e-02 4.58191484e-01 3.35383981e-01 5.14219105e-01 -6.57249168e-02 5.47400236e-01 -1.17063987e+00 8.70492309e-03 -3.56253713e-01 1.40486211e-01 -6.51873708e-01 2.92301774e-01 2.50784189e-01 5.66531241e-01 -4.53229576e-01 6.00534856e-01 2.94959813e-01 -2.74613231e-01 6.58674002e-01 1.25514162e+00 -6.36327744e-01 1.93948686e-01 -7.94227898e-01 1.14280593e+00 -3.59337479e-02 4.22829896e-01 7.60242105e-01 -7.38062024e-01 5.37700951e-01 2.87668914e-01 5.60274899e-01 -4.03562278e-01 7.23248065e-01 4.30885226e-01 7.03309119e-01 6.94110766e-02 4.38448936e-01 -3.17265391e-01 -2.39784539e-01 1.18968034e+00 -7.72576928e-02 -1.45612329e-01 2.91799635e-01 1.64740384e-01 1.37771153e+00 -1.02096051e-01 6.25376284e-01 -3.00242722e-01 3.16559477e-03 7.59809166e-02 7.62450218e-01 1.09347653e+00 -5.88907242e-01 -1.86960116e-01 2.49864861e-01 -2.88277596e-01 -8.32463205e-01 -9.71966565e-01 1.64876491e-01 9.76982832e-01 7.07319796e-01 -4.58075047e-01 -6.50308192e-01 -2.02710852e-01 -2.53982712e-02 2.64689118e-01 -7.08544731e-01 -4.41810548e-01 -4.80141699e-01 -5.91655850e-01 4.79315192e-01 -3.55128050e-02 6.86599612e-01 -1.31354165e+00 -1.40471840e+00 2.65048832e-01 8.95363837e-02 -4.88792151e-01 -3.78148794e-01 5.44596076e-01 -5.38567126e-01 -9.57268298e-01 -4.60006058e-01 -4.35368031e-01 6.33516073e-01 4.59102064e-01 8.22240114e-01 1.04904461e+00 -3.76026362e-01 4.72722381e-01 -7.09882796e-01 -2.96392381e-01 -1.97358191e-01 -9.23360586e-02 7.15183496e-01 -3.93425763e-01 1.61728740e-01 -1.08591247e+00 -6.10855281e-01 4.59267527e-01 -6.29216135e-01 -3.29004735e-01 2.14418620e-01 8.59803557e-01 -2.59137839e-01 3.43891561e-01 7.28303134e-01 1.16931926e-02 1.07183349e+00 -7.03452468e-01 -6.62418544e-01 -8.48899558e-02 -2.13683322e-01 -2.02069730e-01 5.55941880e-01 -5.81257403e-01 -6.50823653e-01 -2.61797577e-01 5.31120062e-01 2.83725679e-01 5.86521663e-02 3.42296869e-01 1.15259208e-01 -5.69745719e-01 5.76282322e-01 8.78105462e-02 1.26830429e-01 -2.30116934e-01 8.60029161e-02 5.96044302e-01 6.28297329e-02 -4.50681239e-01 8.29829097e-01 4.14808244e-01 -2.21739113e-02 -1.05305123e+00 9.07305628e-02 7.40811452e-02 -3.70794445e-01 -4.07246232e-01 5.37687659e-01 -2.89210975e-01 -1.25335491e+00 7.78481722e-01 -1.01621294e+00 -5.35619259e-01 -4.84911412e-01 4.47446227e-01 -2.68538654e-01 3.02824765e-01 -4.13915992e-01 -1.17814326e+00 1.77143857e-01 -6.00342929e-01 1.03548951e-01 7.14647651e-01 -1.83242068e-01 -6.03891730e-01 3.78660649e-01 5.27526997e-02 7.40535915e-01 1.41828150e-01 2.01324344e-01 -6.78548276e-01 -7.99616396e-01 5.20976679e-03 3.54114890e-01 -2.20296785e-01 5.32897651e-01 -1.97679773e-02 -6.38584793e-02 -4.13754433e-01 1.76312283e-01 -3.85956466e-01 2.38328204e-01 8.89894366e-02 -2.13029489e-01 -3.08248818e-01 -4.44069415e-01 3.29234332e-01 9.15545046e-01 8.52804065e-01 1.60266370e-01 4.76101488e-01 -3.48568082e-01 9.23576653e-01 1.49129838e-01 7.81372190e-01 2.64161557e-01 3.62493783e-01 5.93749881e-01 2.08561853e-01 6.07930005e-01 1.98584963e-02 1.49793759e-01 5.83605826e-01 -6.96493268e-01 -5.18898487e-01 -8.01805735e-01 5.23685992e-01 -1.76667070e+00 -1.01234925e+00 3.74201655e-01 2.24298215e+00 7.48867393e-01 2.73891129e-02 3.14771533e-01 -4.19390090e-02 5.55237949e-01 -3.48977864e-01 -4.38253582e-01 -3.86988163e-01 -2.91601062e-01 1.52801082e-01 6.02169633e-01 7.00021446e-01 -4.55676168e-01 1.04126561e+00 7.46935797e+00 1.16359048e-01 -6.92725956e-01 -1.34616438e-02 3.17612827e-01 -2.74599314e-01 3.68079031e-03 2.15773895e-01 -9.11371335e-02 3.26574683e-01 4.11055237e-01 -7.96133935e-01 1.05026555e+00 3.06811154e-01 1.37665182e-01 -7.84112811e-01 -6.32132292e-01 3.13838601e-01 -1.34511665e-01 -8.17556024e-01 -6.52206123e-01 3.75210106e-01 4.38633114e-01 -1.98713496e-01 -3.00681025e-01 -4.88852896e-02 1.15115416e+00 -9.85536098e-01 4.82391357e-01 1.39778823e-01 6.11803569e-02 -7.12135494e-01 5.37638366e-01 8.81623626e-01 -1.00189495e+00 -4.07239616e-01 -6.56146109e-02 -1.19490230e+00 2.82542586e-01 -1.65270329e-01 -5.05325139e-01 3.23155969e-01 7.60246098e-01 7.15988502e-02 -7.87879452e-02 1.02883303e+00 -3.44352454e-01 1.53637052e-01 -6.20809197e-01 -6.51173174e-01 1.52628109e-01 -6.98773682e-01 8.38711321e-01 5.00226542e-02 2.02727810e-01 6.25651300e-01 -1.16062991e-01 8.49303246e-01 -1.01296306e-01 -2.51985162e-01 -7.87466407e-01 -2.59957910e-01 7.47161448e-01 1.08111489e+00 -9.59462881e-01 1.43691123e-01 1.04242796e-02 1.15082502e+00 3.44610691e-01 1.90995097e-01 -5.57448208e-01 -3.76185060e-01 8.27357769e-01 -8.93810168e-02 1.53452829e-01 -5.55940747e-01 4.59760660e-03 -1.09727848e+00 -9.56321135e-02 -5.61941564e-01 4.80510555e-02 -4.91399378e-01 -1.21580505e+00 5.13281643e-01 -1.60477191e-01 -6.29496753e-01 -3.35861415e-01 6.05992526e-02 -7.94589341e-01 4.57487941e-01 -8.77318621e-01 -3.58993471e-01 1.04355142e-01 4.40836132e-01 -9.39163119e-02 -2.76623309e-01 5.47074676e-01 -1.21116370e-01 -4.52336878e-01 2.44844362e-01 1.14745967e-01 -5.76152280e-02 4.05324936e-01 -8.04189980e-01 1.03651933e-01 7.66743660e-01 6.62280247e-02 8.63142371e-01 8.33546758e-01 -8.60101402e-01 -1.22148514e+00 -2.34085381e-01 7.11093128e-01 -2.66141444e-01 8.91294241e-01 -6.84144378e-01 -2.72436321e-01 4.81666714e-01 5.50432861e-01 -3.79555315e-01 6.88449383e-01 -5.76099195e-02 5.13532646e-02 2.48502403e-01 -1.24218273e+00 1.08021522e+00 1.33137918e+00 -1.59921944e-01 -5.42262971e-01 -3.02877836e-02 6.72631860e-01 2.86935180e-01 -1.07091293e-01 -4.96980175e-02 7.50263393e-01 -1.17740273e+00 5.94862700e-01 -5.58414638e-01 -1.09379543e-02 -3.62231061e-02 5.68069853e-02 -1.37980974e+00 -2.37054542e-01 -1.38419998e+00 7.62613297e-01 8.51697206e-01 3.74982715e-01 -1.04885721e+00 7.72623301e-01 4.96122122e-01 3.17048341e-01 -1.63102597e-01 -1.30417335e+00 -8.98736477e-01 2.49294505e-01 5.21002829e-01 3.10907096e-01 9.84604299e-01 7.37253368e-01 1.63134202e-01 -5.94593346e-01 -2.06450582e-01 7.89758682e-01 -1.00601412e-01 6.51570737e-01 -1.35423660e+00 -5.13232708e-01 -5.38458645e-01 -2.50251323e-01 -6.20453775e-01 -1.40962377e-01 -3.71893615e-01 2.20547631e-01 -8.96985888e-01 6.85851555e-03 -4.24242914e-01 -1.31418154e-01 3.64138961e-01 1.35351315e-01 1.09147087e-01 3.34654748e-01 1.71841294e-01 -8.49812567e-01 5.29316843e-01 9.87116873e-01 3.89631540e-01 -6.03041053e-01 -1.59212500e-01 -8.46692801e-01 5.75843394e-01 9.30391014e-01 -7.46525586e-01 -3.19345713e-01 -5.00768004e-03 5.71641266e-01 4.74758923e-01 1.68597236e-01 -8.03904593e-01 4.66437727e-01 -4.96895969e-01 -4.12288725e-01 3.74863535e-01 3.54267836e-01 -9.39861536e-01 4.68268752e-01 9.30512667e-01 8.15660432e-02 1.27061978e-01 -2.09144115e-01 5.56213677e-01 2.25780487e-01 -9.48585421e-02 5.55789351e-01 -3.60052526e-01 -8.29002336e-02 -1.53397322e-01 -1.13187969e+00 -9.42425355e-02 1.46102440e+00 -3.31192285e-01 -5.34777045e-01 -6.35779440e-01 -3.51401895e-01 4.83671367e-01 1.08066630e+00 4.80641611e-03 1.67134762e-01 -7.40250528e-01 -6.92292988e-01 1.72722384e-01 -5.65030277e-01 -4.54609364e-01 3.65793370e-02 7.24811852e-01 -6.32654428e-01 -2.41976753e-01 -8.40637803e-01 1.37180731e-01 -1.10294139e+00 4.70385373e-01 6.10385001e-01 -1.80377051e-01 -2.98091739e-01 8.79053354e-01 7.80657306e-02 -1.72284424e-01 -1.83681287e-02 3.79276574e-01 1.89834446e-01 -1.38123900e-01 4.12418962e-01 2.01200441e-01 -4.94944662e-01 -4.19959456e-01 -6.40280843e-01 1.07255526e-01 2.44753286e-01 -4.53241944e-01 1.30234766e+00 -2.50263274e-01 -4.35459197e-01 9.60258469e-02 1.60794824e-01 2.80677795e-01 -1.13129115e+00 2.40849078e-01 -1.47862406e-02 -5.37225306e-01 -4.51198936e-01 -6.88461900e-01 -6.63123846e-01 1.94743261e-01 -2.23613739e-01 9.49582517e-01 1.06565809e+00 -9.85274613e-02 3.69557559e-01 5.73034286e-01 1.16655493e+00 -9.73018408e-01 2.29501322e-01 5.90957165e-01 4.63767797e-01 -9.04509187e-01 -3.64500254e-01 -2.24942826e-02 -4.30585742e-01 4.85831797e-01 4.67242420e-01 -3.84374708e-01 6.13565683e-01 6.92418635e-01 -1.24709144e-01 -3.88015173e-02 -1.07828701e+00 -3.57023805e-01 -8.23733091e-01 1.12899435e+00 -2.42379513e-02 4.91803028e-02 -8.62688303e-01 7.89334774e-01 -2.93898582e-01 -2.59300292e-01 1.15577030e+00 1.33998764e+00 -7.44902372e-01 -1.45028925e+00 -1.72198519e-01 -8.50356445e-02 -4.40990716e-01 -1.17104046e-01 -1.20064902e+00 9.16557610e-01 1.47450209e-01 1.30495214e+00 1.25719041e-01 -2.80311853e-01 -1.96313381e-01 -1.78413361e-01 2.99663067e-01 -4.01195168e-01 -1.01801324e+00 -8.19722340e-02 5.64635992e-02 -2.25714132e-01 -6.73158228e-01 -6.43262506e-01 -1.04930675e+00 -7.20713675e-01 -4.39685762e-01 7.62874842e-01 1.72440767e-01 7.36138523e-01 2.24651724e-01 -5.08463122e-02 9.61228251e-01 -6.80209696e-01 -2.17708841e-01 -6.23216689e-01 -1.18097794e+00 -6.12396114e-02 2.80790687e-01 -9.02321577e-01 -9.16998923e-01 -4.04719174e-01]
[3.8922555446624756, 2.1788299083709717]
6c7a0fb6-fb28-4dfa-bc39-3ca470e66613
from-synthetic-to-real-image-dehazing
2108.02934
null
https://arxiv.org/abs/2108.02934v1
https://arxiv.org/pdf/2108.02934v1.pdf
From Synthetic to Real: Image Dehazing Collaborating with Unlabeled Real Data
Single image dehazing is a challenging task, for which the domain shift between synthetic training data and real-world testing images usually leads to degradation of existing methods. To address this issue, we propose a novel image dehazing framework collaborating with unlabeled real data. First, we develop a disentangled image dehazing network (DID-Net), which disentangles the feature representations into three component maps, i.e. the latent haze-free image, the transmission map, and the global atmospheric light estimate, respecting the physical model of a haze process. Our DID-Net predicts the three component maps by progressively integrating features across scales, and refines each map by passing an independent refinement network. Then a disentangled-consistency mean-teacher network (DMT-Net) is employed to collaborate unlabeled real data for boosting single image dehazing. Specifically, we encourage the coarse predictions and refinements of each disentangled component to be consistent between the student and teacher networks by using a consistency loss on unlabeled real data. We make comparison with 13 state-of-the-art dehazing methods on a new collected dataset (Haze4K) and two widely-used dehazing datasets (i.e., SOTS and HazeRD), as well as on real-world hazy images. Experimental results demonstrate that our method has obvious quantitative and qualitative improvements over the existing methods.
['Wei Feng', 'Liang Wan', 'Qing Zhang', 'Jing Qin', 'Huazhu Fu', 'Shunda Pei', 'Lei Zhu', 'Ye Liu']
2021-08-06
null
null
null
null
['image-dehazing']
['computer-vision']
[ 3.17004770e-01 1.57739922e-01 2.61647463e-01 -3.78001451e-01 -6.39202237e-01 -2.32077822e-01 7.93956101e-01 -4.49366897e-01 -1.17438756e-01 7.49029517e-01 -1.45883977e-01 -7.84777701e-02 -1.05074920e-01 -8.77933025e-01 -9.12093878e-01 -1.38551617e+00 2.27073461e-01 3.30467880e-01 4.93406236e-01 -3.10875803e-01 -1.98391601e-01 -3.33973691e-02 -1.73427761e+00 1.09250717e-01 1.35539448e+00 8.26382399e-01 4.29208353e-02 6.26437724e-01 2.63236821e-01 1.04834235e+00 -7.16126025e-01 -3.75061125e-01 3.15371126e-01 -4.80878055e-01 -3.98383617e-01 2.94719785e-01 1.02131414e+00 -5.77441096e-01 -5.26426256e-01 1.33488357e+00 2.37541541e-01 1.81428954e-01 6.42190993e-01 -1.60032260e+00 -1.35605681e+00 2.04650119e-01 -9.00007427e-01 5.43485098e-02 -4.92319614e-01 2.58945525e-01 5.58603704e-01 -1.01607847e+00 2.29787737e-01 1.19965136e+00 3.84958327e-01 6.29353642e-01 -1.35370803e+00 -1.35427797e+00 2.93677598e-01 2.72391826e-01 -1.40007722e+00 -2.97768116e-01 9.48117197e-01 -5.80044448e-01 2.68639803e-01 1.14347376e-01 6.15087211e-01 9.66838658e-01 2.84542501e-01 7.23413467e-01 1.52106380e+00 -3.04412216e-01 1.06558479e-01 4.31077838e-01 4.33688797e-02 6.84232533e-01 4.48595673e-01 2.70429313e-01 -6.97737753e-01 1.78833842e-01 6.59197569e-01 -2.09795218e-02 -4.99206126e-01 -4.33812141e-01 -1.23787844e+00 7.54286647e-01 5.58914363e-01 -2.24903330e-01 -2.40154207e-01 -8.22574198e-02 -4.09140229e-01 4.78110880e-01 1.15902257e+00 1.83145404e-01 -1.40232801e-01 7.29915917e-01 -1.05358350e+00 2.95547575e-01 3.82304907e-01 8.79124224e-01 1.29293525e+00 2.36492559e-01 1.04341477e-01 6.49659812e-01 5.48739433e-01 8.94383430e-01 1.20215364e-01 -7.82741249e-01 3.05213332e-01 3.61183316e-01 1.75618067e-01 -6.97038054e-01 2.99126029e-01 -3.55794638e-01 -1.18628287e+00 7.88942575e-01 2.47706890e-01 -9.83938575e-03 -1.34235883e+00 1.65431547e+00 6.73136592e-01 8.42129230e-01 1.96256742e-01 1.12443054e+00 9.00017977e-01 9.88868535e-01 -2.21935228e-01 1.29190400e-01 1.21190739e+00 -1.51907158e+00 -7.97082007e-01 -3.51120383e-01 -7.94860721e-02 -6.91975296e-01 7.48052120e-01 7.61101723e-01 -1.04093301e+00 -6.04454100e-01 -1.53269589e+00 -1.37153268e-01 -4.26305801e-01 -3.03074151e-01 1.62685588e-01 5.18590629e-01 -1.13095891e+00 2.13227376e-01 -7.59235740e-01 2.14707270e-01 3.64014566e-01 9.06201154e-02 -3.42115074e-01 -4.65671659e-01 -1.46876907e+00 8.71596098e-01 2.70276397e-01 2.39528105e-01 -1.42497075e+00 -1.14283252e+00 -8.31986547e-01 -1.63371995e-01 2.30444133e-01 -8.90720069e-01 1.05256343e+00 -1.03913474e+00 -1.48099720e+00 7.90782571e-01 2.48195007e-01 -1.81649506e-01 3.91746610e-01 -3.13050389e-01 -6.73959792e-01 2.78525054e-01 -5.36526293e-02 8.58922541e-01 1.64370203e+00 -1.81487846e+00 -5.80685318e-01 -3.45678896e-01 -2.24897619e-02 4.13675636e-01 -2.09398031e-01 -4.23701644e-01 -1.48382306e-01 -7.74773419e-01 7.23643824e-02 -9.11114395e-01 1.90945622e-02 3.09609324e-01 -3.40829283e-01 1.30799785e-01 1.05876160e+00 -7.64035702e-01 8.24603558e-01 -2.29449344e+00 3.30532730e-01 -9.16886330e-02 9.56310809e-01 2.07697555e-01 -2.64665663e-01 -4.21657041e-03 -3.05353612e-01 -1.58642575e-01 -2.75601655e-01 -5.61859608e-01 -1.97224259e-01 4.18248504e-01 -4.87261146e-01 8.06031227e-01 5.02903700e-01 5.95228612e-01 -1.05631804e+00 -4.84609425e-01 3.37027639e-01 7.93947518e-01 -3.26057047e-01 6.13301396e-01 -3.31742167e-01 7.84393489e-01 -1.19262047e-01 2.49274567e-01 1.20983827e+00 -3.15975398e-01 -3.01597506e-01 -1.19378105e-01 -1.66446865e-01 1.48610130e-01 -8.55929077e-01 1.45313227e+00 -4.07243580e-01 6.76168621e-01 2.23561257e-01 -7.67223895e-01 8.32998931e-01 3.50722194e-01 1.85625717e-01 -7.29770243e-01 -2.06250206e-01 -8.28088745e-02 -1.85018897e-01 -4.41685408e-01 2.95100272e-01 -3.75583500e-01 5.10771155e-01 4.94638741e-01 3.85408670e-01 -7.44220793e-01 -3.43913257e-01 3.58157009e-01 5.46409369e-01 1.33260516e-02 -1.02568291e-01 -5.31637013e-01 4.16052103e-01 -5.72560914e-02 4.52593982e-01 6.79151237e-01 -2.75755227e-01 9.03824329e-01 5.44355698e-02 -5.23280084e-01 -1.01080120e+00 -1.49103498e+00 -1.00866131e-01 8.53998482e-01 5.73947430e-01 3.74975093e-02 -8.59160542e-01 -5.91195226e-01 -1.96415991e-01 7.68697441e-01 -9.46080506e-01 -3.32255483e-01 -3.08955163e-01 -8.38803768e-01 2.59786934e-01 -1.40763283e-01 9.93517041e-01 -5.01914978e-01 -8.80146101e-02 -2.15947807e-01 -2.28972942e-01 -1.00914490e+00 -5.14541566e-01 7.88401216e-02 -4.33487415e-01 -9.60853100e-01 -6.73389673e-01 -7.53222883e-01 6.25220120e-01 9.93755817e-01 1.28568828e+00 1.38700470e-01 7.65218437e-02 9.71557945e-02 -1.89092115e-01 -6.42384589e-01 -5.18988788e-01 -2.16204837e-01 -4.83935811e-02 3.13203961e-01 2.19976425e-01 -8.37358892e-01 -9.83759880e-01 4.79058772e-01 -1.39252496e+00 6.37051225e-01 4.83845234e-01 9.03689444e-01 5.11885226e-01 3.41133922e-01 2.14564368e-01 -8.65166306e-01 6.33296371e-02 -7.41656244e-01 -7.38699496e-01 4.25323397e-01 -1.03322303e+00 1.57862693e-01 3.28549355e-01 -6.53191030e-01 -1.46856236e+00 -3.43662471e-01 2.42761225e-01 -8.32727671e-01 -1.90911591e-01 1.43599138e-01 -2.13679835e-01 -3.12469691e-01 7.82634616e-01 3.70151669e-01 4.80452254e-02 -1.65134579e-01 6.11140490e-01 5.15249610e-01 8.10692549e-01 -6.26928568e-01 1.69382346e+00 7.05019653e-01 -3.38850886e-01 -7.93306649e-01 -1.29158533e+00 -4.12769586e-01 -5.00315428e-01 -3.05833101e-01 1.12149298e+00 -1.50143826e+00 -2.26611167e-01 9.37948883e-01 -1.10875404e+00 -4.20250088e-01 -2.42792830e-01 5.21934569e-01 -1.53305009e-01 1.26197681e-01 -5.83312869e-01 -6.07895970e-01 -1.70789137e-01 -1.17943120e+00 1.20795810e+00 3.16686213e-01 6.04872942e-01 -1.00938106e+00 3.30441475e-01 7.12532222e-01 3.78407121e-01 2.34769061e-01 9.70108211e-01 -1.52689040e-01 -9.93164897e-01 3.70534748e-01 -4.27113533e-01 7.56588995e-01 2.26888314e-01 7.05615133e-02 -1.33300579e+00 -2.62895942e-01 3.17397326e-01 -5.24792373e-01 1.15832996e+00 1.30205169e-01 9.12797928e-01 -3.31187576e-01 1.29025534e-01 9.06524837e-01 1.13929081e+00 -9.82679352e-02 7.08502591e-01 3.29248101e-01 9.32006001e-01 8.14790308e-01 3.82296354e-01 -4.82459441e-02 5.65761387e-01 3.12384039e-01 8.31005454e-01 -4.66315627e-01 -4.24999177e-01 -2.74897546e-01 3.58363003e-01 1.23648369e+00 7.26631358e-02 -4.43536937e-01 -6.54540718e-01 4.19899821e-01 -1.72247446e+00 -6.18865967e-01 -1.79131702e-01 1.94838202e+00 8.34535360e-01 -8.38820636e-02 -3.54764163e-01 -1.11574307e-01 6.36970818e-01 6.41317308e-01 -6.76332176e-01 1.40125230e-01 -1.74975425e-01 2.38758251e-01 3.77946228e-01 6.63130820e-01 -1.15157294e+00 7.88560808e-01 5.75061417e+00 7.25400329e-01 -1.03088880e+00 4.10828769e-01 5.62112808e-01 -9.97663513e-02 -6.10347569e-01 -1.51675448e-01 -4.33165431e-01 4.41536397e-01 7.80689001e-01 -9.13574845e-02 6.12495840e-01 5.63021898e-01 7.64880516e-03 -3.29753431e-03 -9.33271170e-01 7.76031494e-01 2.45606273e-01 -1.27387786e+00 2.24790573e-01 -1.75619107e-02 1.23865950e+00 1.81415156e-01 6.16087317e-01 2.06255883e-01 8.17541182e-01 -1.03639078e+00 9.63627100e-01 2.89482296e-01 7.53275394e-01 -3.68389726e-01 4.88719404e-01 4.55850154e-01 -9.34533060e-01 5.18023193e-01 -1.62521511e-01 1.79051280e-01 -3.82222533e-01 7.62152791e-01 -5.33648670e-01 7.33615637e-01 9.44855273e-01 7.94041812e-01 -5.69773853e-01 5.29067695e-01 -5.73389649e-01 9.14733469e-01 -4.04435508e-02 5.34642220e-01 2.47467458e-01 -6.05150759e-01 3.05492848e-01 7.01438069e-01 9.07656103e-02 2.22789645e-01 -2.48333924e-02 1.13179755e+00 1.85726458e-04 -6.09835744e-01 -4.16099280e-01 3.44461292e-01 2.37079993e-01 1.17802882e+00 -3.47043812e-01 -4.36109751e-01 -6.20259523e-01 1.03523207e+00 2.78994054e-01 8.16636622e-01 -1.05229962e+00 -1.02286689e-01 9.67202425e-01 -1.29880697e-01 1.17177010e-01 -1.70076802e-01 1.30157053e-01 -1.56863284e+00 -2.39411175e-01 -8.78518105e-01 1.32096052e-01 -1.25214362e+00 -1.70465028e+00 6.39967322e-01 3.05165023e-01 -1.27319860e+00 1.61068007e-01 -4.70902890e-01 -7.27693319e-01 1.16801143e+00 -2.15724397e+00 -1.38236952e+00 -8.74716640e-01 7.68113315e-01 5.21857440e-01 2.12731317e-01 4.33689028e-01 2.58603126e-01 -6.84262276e-01 3.21527034e-01 3.08378369e-01 -7.48976246e-02 9.90841329e-01 -1.35815787e+00 4.12906647e-01 1.02993417e+00 2.41953865e-01 3.63354743e-01 8.68039906e-01 -3.89560878e-01 -1.02954078e+00 -1.39297044e+00 2.33025864e-01 -5.91275156e-01 8.24150085e-01 -6.75266683e-01 -1.36119401e+00 7.75303364e-01 6.35304093e-01 4.41336840e-01 5.48024774e-01 -3.66083741e-01 -8.59415114e-01 -2.32068047e-01 -9.93161380e-01 5.28382361e-01 7.47762024e-01 -5.59309185e-01 -6.56472266e-01 3.37253571e-01 1.30710387e+00 -4.72765714e-01 -6.95294082e-01 3.00678134e-01 4.93826747e-01 -1.22109473e+00 1.03126132e+00 -3.60963196e-01 6.36012733e-01 -5.05178690e-01 2.43892591e-03 -1.78145003e+00 -3.79621923e-01 -6.01192951e-01 -1.82104200e-01 1.01252913e+00 1.90013453e-01 -7.39846289e-01 8.12584639e-01 3.08500975e-01 -2.24917963e-01 -4.74688619e-01 -7.50426412e-01 -5.75753033e-01 3.51350993e-01 -2.08198875e-01 9.16681111e-01 1.34723961e+00 -7.84900963e-01 3.08845431e-01 -7.40034342e-01 9.65333700e-01 1.24951732e+00 1.09357826e-01 6.96042538e-01 -1.17585242e+00 -2.36921877e-01 7.96651840e-02 -2.13529497e-01 -1.00202525e+00 2.58149981e-01 -5.34467280e-01 3.80712748e-01 -1.15856671e+00 1.85606837e-01 -3.44183803e-01 -4.44680542e-01 2.75283962e-01 -4.90302145e-01 4.13091928e-01 -6.21201731e-02 3.20165485e-01 -3.97976905e-01 1.09119463e+00 1.74610722e+00 -6.64733291e-01 2.00324491e-01 -2.04412431e-01 -5.29506624e-01 5.81268311e-01 6.76759660e-01 -8.44205081e-01 -8.28706563e-01 -7.20400631e-01 5.95743544e-02 -9.63865668e-02 6.73187852e-01 -8.44850898e-01 3.37979078e-01 -4.35750574e-01 3.34337354e-01 -4.24981594e-01 3.85860056e-01 -7.77308941e-01 2.35209763e-01 2.04433784e-01 -1.27109751e-01 -2.63484269e-01 -1.19853780e-01 7.24814773e-01 -4.43488061e-01 1.08449228e-01 9.44504380e-01 6.69708923e-02 -5.35081685e-01 5.88524222e-01 -1.34487838e-01 3.18327323e-02 8.96970987e-01 -1.28537729e-01 -9.90885913e-01 -3.63425404e-01 -4.36689019e-01 4.16602612e-01 3.78587127e-01 4.95987922e-01 8.40651155e-01 -1.18088627e+00 -9.00023699e-01 5.52767456e-01 3.68381202e-01 7.16138899e-01 4.63965893e-01 5.98467290e-01 -5.69088161e-01 -1.98151678e-01 -1.08185992e-01 -6.65933192e-01 -1.13767421e+00 5.43174922e-01 5.22823453e-01 -3.09680980e-02 -6.48392320e-01 9.52562451e-01 1.02159262e+00 -5.65347552e-01 -2.72257496e-02 -2.52384037e-01 6.27776682e-02 -3.24406624e-01 6.81856096e-01 1.13090977e-01 1.10426163e-02 -5.35430908e-01 -5.42082824e-04 4.29992378e-01 -9.58958417e-02 -8.67971107e-02 1.31210113e+00 -3.55523825e-01 -3.76845330e-01 4.23406899e-01 1.00792861e+00 -1.25207156e-01 -1.89883351e+00 -5.03742635e-01 -5.61287940e-01 -6.00885332e-01 4.06280339e-01 -5.98842740e-01 -1.33947778e+00 1.05406034e+00 7.69776106e-01 2.16341808e-01 1.18432724e+00 2.68836934e-02 6.70912743e-01 1.61215186e-01 1.94476262e-01 -3.53545934e-01 4.80904430e-01 1.11568674e-01 6.63185179e-01 -1.47530782e+00 1.05621561e-01 -5.34914494e-01 -5.07273614e-01 6.69751167e-01 1.11555719e+00 -1.57953426e-01 9.41769361e-01 1.04672059e-01 3.62216949e-01 -4.39847261e-01 -9.43889022e-01 5.12394160e-02 5.65288603e-01 6.70692027e-01 -1.23001821e-01 3.42800319e-02 6.94024324e-01 2.44389474e-01 -7.70915896e-02 -3.46886009e-01 5.07240951e-01 5.33554256e-01 -4.45594400e-01 -5.81570923e-01 -5.38977265e-01 4.87909280e-02 2.42598169e-02 -1.15577050e-01 -2.81852245e-01 9.72623467e-01 5.20927012e-01 1.08151126e+00 1.07115343e-01 -5.04634082e-01 5.67612201e-02 -2.05801621e-01 6.22089267e-01 -6.10355258e-01 -3.74823302e-01 -1.30976588e-01 -4.79529023e-01 -2.72827238e-01 -6.61254704e-01 -1.18710756e-01 -7.25518584e-01 -4.48953658e-01 -5.93173385e-01 3.69913399e-01 4.81449455e-01 9.15113986e-01 7.88735077e-02 7.19831169e-01 8.46738756e-01 -1.02560878e+00 -4.35510576e-01 -8.89026999e-01 -7.86283314e-01 3.83338422e-01 9.80090320e-01 -7.65320122e-01 -7.91788936e-01 2.92988986e-01]
[10.943320274353027, -3.142702341079712]
9e712718-6155-4925-8252-5a3579d54ee2
roam-random-layer-mixup-for-semi-supervised
2003.09439
null
https://arxiv.org/abs/2003.09439v4
https://arxiv.org/pdf/2003.09439v4.pdf
ROAM: Random Layer Mixup for Semi-Supervised Learning in Medical Imaging
Medical image segmentation is one of the major challenges addressed by machine learning methods. Yet, deep learning methods profoundly depend on a large amount of annotated data, which is time-consuming and costly. Though, semi-supervised learning methods approach this problem by leveraging an abundant amount of unlabeled data along with a small amount of labeled data in the training process. Recently, MixUp regularizer has been successfully introduced to semi-supervised learning methods showing superior performance. MixUp augments the model with new data points through linear interpolation of the data at the input space. We argue that this option is limited. Instead, we propose ROAM, a RandOm lAyer Mixup, which encourages the network to be less confident for interpolated data points at randomly selected space. ROAM generates more data points that have never seen before, and hence it avoids over-fitting and enhances the generalization ability. We conduct extensive experiments to validate our method on three publicly available datasets on whole-brain image segmentation. ROAM achieves state-of-the-art (SOTA) results in fully supervised (89.5%) and semi-supervised (87.0%) settings with a relative improvement of up to 2.40% and 16.50%, respectively for the whole-brain segmentation.
['Shadi Albarqouni', 'Benedikt Wiestler', 'Tariq Bdair', 'Nassir Navab']
2020-03-20
null
null
null
null
['brain-image-segmentation']
['medical']
[ 2.31434152e-01 3.64757210e-01 -4.04671103e-01 -7.07048118e-01 -9.31963086e-01 -2.42056012e-01 2.10868120e-01 4.68841977e-02 -9.72586811e-01 7.43031561e-01 -1.75370425e-01 -1.24281771e-01 2.68267214e-01 -6.13989770e-01 -7.71287501e-01 -9.27152395e-01 1.21952057e-01 6.33750916e-01 1.77124947e-01 1.24946877e-01 -1.67446151e-01 1.55309513e-01 -1.01288676e+00 1.23894125e-01 1.37619436e+00 9.49740946e-01 3.44678760e-01 9.96705443e-02 -1.61974758e-01 4.64033216e-01 -3.15862834e-01 -1.37748331e-01 3.15527081e-01 -2.30598733e-01 -9.13355947e-01 2.45232537e-01 2.82793492e-01 -1.90569222e-01 -1.86335057e-01 1.15690088e+00 4.31812763e-01 -4.92792763e-03 6.44436300e-01 -1.06217694e+00 -5.99243402e-01 8.32838595e-01 -8.82601023e-01 1.10923097e-01 -3.53525251e-01 1.19145542e-01 6.85832322e-01 -9.11076903e-01 5.79982460e-01 6.81221426e-01 6.54983580e-01 7.97302961e-01 -1.39178300e+00 -8.14222991e-01 1.05864458e-01 -1.30969211e-01 -1.32550406e+00 -2.11347863e-01 8.61058295e-01 -4.65211451e-01 4.32062894e-01 1.80549379e-02 6.15095437e-01 8.72099042e-01 -1.36513010e-01 1.23189592e+00 1.39377475e+00 -2.38354251e-01 3.31759542e-01 2.13498875e-01 3.78846467e-01 7.12783635e-01 3.99719551e-02 -7.56171718e-02 -1.69793040e-01 6.65866658e-02 9.19456065e-01 1.57344386e-01 -2.34198153e-01 -3.93465042e-01 -1.17778873e+00 8.84506047e-01 8.07191133e-01 2.60712236e-01 -4.12146688e-01 -3.22708488e-01 3.00099045e-01 -1.34235783e-03 6.04113162e-01 4.44529474e-01 -5.93744040e-01 7.97219500e-02 -1.21962714e+00 -1.69889927e-01 3.22047800e-01 8.41006398e-01 7.53436804e-01 -8.95316005e-02 -5.84671535e-02 1.17758548e+00 2.45367765e-01 2.92444944e-01 8.29783618e-01 -5.62182248e-01 5.31481445e-01 7.84707904e-01 -1.94819197e-01 -5.02173781e-01 -5.77863634e-01 -8.03495169e-01 -1.12913191e+00 9.85103548e-02 6.57498956e-01 -4.23234344e-01 -1.41547883e+00 1.70775485e+00 2.91182905e-01 2.37486541e-01 -5.65467961e-02 9.43829417e-01 8.91851604e-01 5.42542219e-01 4.75798883e-02 -2.77983457e-01 9.67773676e-01 -1.41598046e+00 -5.77977777e-01 -4.93325770e-01 7.43734419e-01 -4.56149638e-01 1.29121721e+00 4.69292223e-01 -9.39578176e-01 -5.29195011e-01 -8.71246636e-01 8.51959176e-03 -1.06371589e-01 3.14870209e-01 7.82949746e-01 5.09423792e-01 -8.51702273e-01 5.50271690e-01 -1.15252113e+00 1.78859785e-01 1.11470568e+00 4.63049442e-01 -4.20171589e-01 -1.91601604e-01 -8.91903400e-01 4.77271050e-01 4.29733038e-01 2.38571003e-01 -6.54040635e-01 -7.81545103e-01 -8.49615216e-01 -3.12690228e-01 4.93109703e-01 -2.70635009e-01 1.03217506e+00 -1.03938067e+00 -1.20843184e+00 1.00487244e+00 -1.04576312e-01 -6.63035572e-01 9.25471425e-01 -3.04692715e-01 -2.42662117e-01 2.41369739e-01 1.92773372e-01 1.20541096e+00 7.18232930e-01 -1.22103775e+00 -4.43088740e-01 -4.79932129e-01 -3.52493018e-01 2.93689761e-02 -2.72481769e-01 -3.52497250e-01 -4.89926130e-01 -8.21565807e-01 4.15501177e-01 -1.03899288e+00 -5.77038050e-01 2.43246425e-02 -5.56874871e-01 -1.96806818e-01 6.52017891e-01 -4.85332847e-01 9.33506370e-01 -2.04629135e+00 -3.90105583e-02 1.02260798e-01 3.59379947e-01 5.72885096e-01 -4.55782041e-02 -2.82958984e-01 -9.04691517e-02 1.54434681e-01 -7.65026629e-01 -5.56954145e-01 -3.14250916e-01 1.84205443e-01 -1.57457665e-01 4.03116077e-01 1.28369316e-01 1.01674831e+00 -9.72501278e-01 -6.84719086e-01 1.80884868e-01 3.72466624e-01 -5.27464926e-01 3.59184332e-02 -9.87225845e-02 8.21260691e-01 -3.97439152e-01 5.55074751e-01 8.25555086e-01 -5.68931162e-01 -1.85237780e-01 -1.61806848e-02 1.80392563e-01 -4.52606045e-02 -9.78126824e-01 1.90183842e+00 -3.49981636e-01 5.19084036e-01 -3.93775970e-01 -1.14392757e+00 9.66257393e-01 1.56917617e-01 6.07268870e-01 -5.68768561e-01 1.97086781e-01 3.11408728e-01 2.90482845e-02 -4.99811798e-01 7.35861734e-02 -4.92101870e-02 1.94008887e-01 4.11746144e-01 1.64321512e-01 5.17238863e-02 1.07740544e-01 1.38681337e-01 7.60170043e-01 1.28629252e-01 8.74655843e-02 -2.42570788e-01 3.61945033e-01 9.63430181e-02 9.03892457e-01 7.24731684e-01 -3.99441659e-01 8.37151647e-01 4.09270316e-01 -4.08764720e-01 -8.82954359e-01 -8.56020331e-01 -5.44516027e-01 8.17295432e-01 1.58314005e-01 -5.26891463e-02 -1.12812674e+00 -1.07769704e+00 -2.05038667e-01 5.27481139e-01 -8.01683843e-01 -7.05380961e-02 -5.89658678e-01 -1.12877297e+00 3.75119686e-01 8.28838050e-01 8.82465780e-01 -1.13580906e+00 -4.91707116e-01 2.15785488e-01 -1.61667734e-01 -1.25481081e+00 -5.08421957e-01 3.25797290e-01 -1.18803370e+00 -9.04416263e-01 -1.08325922e+00 -9.65915203e-01 1.29307246e+00 8.48542377e-02 8.70260537e-01 1.38916343e-01 -2.22083509e-01 -3.48040462e-01 -2.76917547e-01 -1.64097890e-01 -2.47424856e-01 2.10222244e-01 -1.05237469e-01 4.71521877e-02 3.89567494e-01 -2.72797823e-01 -6.53432429e-01 5.47583997e-01 -7.91073263e-01 4.70811546e-01 6.51344180e-01 1.24611902e+00 1.00529647e+00 -1.07076295e-01 8.64327610e-01 -1.32668161e+00 1.03699833e-01 -3.71589452e-01 -5.24586558e-01 6.07810430e-02 -6.41765893e-01 2.11753324e-02 7.19383061e-01 -7.24976778e-01 -9.00776505e-01 4.33001876e-01 -3.46511185e-01 -3.28923851e-01 -3.33137482e-01 5.18205464e-01 -6.11974336e-02 -1.04165472e-01 6.46979928e-01 1.70342147e-01 8.47645327e-02 -5.14776409e-01 2.86155015e-01 7.61378169e-01 5.66782176e-01 -3.49820763e-01 5.56154251e-01 6.16568565e-01 -2.63493657e-01 -6.13362372e-01 -1.12574828e+00 -5.18599331e-01 -8.77892017e-01 -7.24848360e-02 8.61389279e-01 -8.49544883e-01 -1.16770916e-01 7.09061921e-01 -5.97939909e-01 -8.07077587e-01 -3.46675545e-01 6.32890224e-01 -3.49100143e-01 3.54344159e-01 -6.27098083e-01 -4.08053905e-01 -3.78265738e-01 -1.54847360e+00 8.29964221e-01 3.94957542e-01 -3.36216204e-03 -9.84260678e-01 -3.39113474e-01 6.18148625e-01 1.29790634e-01 3.06763887e-01 6.11617386e-01 -9.05992568e-01 -2.68614680e-01 -2.50896156e-01 -3.16492945e-01 5.46690345e-01 2.69453794e-01 -2.88281828e-01 -1.04774547e+00 -3.28448355e-01 -8.95301849e-02 -6.09372973e-01 1.09594774e+00 6.12998307e-01 1.54191732e+00 3.68513800e-02 -3.91987562e-01 7.55341351e-01 1.16748333e+00 2.57017650e-02 4.42809165e-01 1.96454391e-01 8.17534149e-01 4.82501775e-01 6.57066643e-01 1.63146257e-01 3.38374525e-01 4.13990170e-01 3.10835719e-01 -5.62827945e-01 -4.26967330e-02 -2.35275492e-01 -1.84378743e-01 6.90305948e-01 1.57332987e-01 1.11642726e-01 -1.11526048e+00 5.38345814e-01 -1.89091289e+00 -4.63137090e-01 -1.38287395e-01 2.16588640e+00 1.30085742e+00 4.27780241e-01 2.28369385e-01 2.09599689e-01 7.55864203e-01 -6.84851035e-02 -9.65462029e-01 3.28614175e-01 -1.99483167e-02 2.31675893e-01 6.03928506e-01 3.50297570e-01 -1.42811859e+00 1.06394660e+00 6.14109802e+00 9.34494615e-01 -1.33956826e+00 1.20593823e-01 1.17011893e+00 3.57248634e-02 1.04242541e-01 -4.11041945e-01 -7.14159667e-01 7.60838985e-01 5.44317126e-01 2.70578593e-01 2.57754624e-01 9.47362900e-01 9.37037468e-02 -1.42727539e-01 -1.01993763e+00 1.02470839e+00 -5.34004197e-02 -1.41088295e+00 -1.38936982e-01 -5.42375520e-02 1.05382228e+00 2.55991071e-01 1.14656329e-01 2.90573120e-01 1.55914426e-01 -1.04221308e+00 5.29414058e-01 1.19494587e-01 7.93055296e-01 -6.54317260e-01 9.39857304e-01 6.89488709e-01 -7.61065185e-01 2.66245920e-02 -3.57835114e-01 2.27401733e-01 1.43194884e-01 8.32574010e-01 -7.63050735e-01 1.64767876e-01 6.58794343e-01 7.28692830e-01 -6.28892362e-01 1.24085534e+00 -5.08008122e-01 9.73135233e-01 -4.46428150e-01 2.96177477e-01 4.77481484e-01 -2.44717136e-01 1.29193008e-01 9.99924362e-01 -1.93492338e-01 4.33053300e-02 3.86016726e-01 8.64103496e-01 -2.44643316e-01 1.14530213e-01 -1.21196523e-01 1.69828668e-01 3.29268277e-01 1.29930520e+00 -1.13111019e+00 -3.82619768e-01 -3.30025315e-01 7.33111441e-01 4.34388250e-01 3.92742723e-01 -7.79889286e-01 -2.08170384e-01 2.69119199e-02 -7.80827329e-02 7.39765838e-02 -1.18308768e-01 -7.06951559e-01 -1.09302330e+00 -5.27081592e-03 -5.83357811e-01 4.51932371e-01 -5.12853861e-01 -1.26846564e+00 8.11768055e-01 -1.20895788e-01 -1.20464361e+00 -1.09661609e-01 -4.97602940e-01 -4.35459346e-01 7.32423604e-01 -1.63850653e+00 -1.06670368e+00 -4.08528179e-01 4.16986197e-01 6.21868074e-01 -1.87344745e-01 5.45573354e-01 3.82435113e-01 -7.75660992e-01 7.35544205e-01 3.39917913e-02 5.14186323e-01 7.14558005e-01 -1.42195594e+00 3.72742474e-01 7.48998821e-01 2.31484219e-01 5.64570844e-01 3.50048423e-01 -5.42040288e-01 -8.23135018e-01 -1.18191540e+00 5.03539681e-01 -9.07119140e-02 5.65108895e-01 -3.25626642e-01 -1.28079784e+00 6.36290073e-01 -1.60106614e-01 6.38812840e-01 7.90761590e-01 -5.35051674e-02 -1.11091435e-01 -9.26016923e-03 -1.45870435e+00 5.15267372e-01 8.38366807e-01 -2.60853529e-01 -6.36201143e-01 4.56214756e-01 6.53352737e-01 -6.39932156e-01 -8.09805274e-01 4.91117597e-01 2.12894067e-01 -6.76823735e-01 8.06405246e-01 -5.00728369e-01 4.41527545e-01 -2.59223521e-01 2.73881972e-01 -1.38829923e+00 3.85329011e-03 -5.43597639e-01 5.57838678e-02 1.00252008e+00 6.34138465e-01 -7.67971992e-01 1.14450610e+00 7.67027378e-01 -2.28263050e-01 -1.31454647e+00 -7.01309860e-01 -7.60440588e-01 4.27322686e-01 -4.19222146e-01 5.22624254e-01 1.10247397e+00 -8.13080892e-02 7.95358494e-02 -1.62571192e-01 3.77371199e-02 8.21506500e-01 9.70521644e-02 5.52394509e-01 -1.15288424e+00 -8.87263864e-02 -3.22295606e-01 -1.72038540e-01 -1.28603160e+00 1.56240001e-01 -1.05358601e+00 1.35796130e-01 -1.51406455e+00 2.59702533e-01 -8.73639405e-01 -3.64124626e-01 8.20727766e-01 -4.68231022e-01 5.23477495e-01 -8.81071836e-02 4.14281160e-01 -4.29743081e-01 4.44087654e-01 1.56868851e+00 -1.37720153e-01 -5.05223155e-01 1.45976290e-01 -6.14004791e-01 1.03054738e+00 1.05903590e+00 -4.38321590e-01 -4.47968125e-01 -5.97679079e-01 -3.00767809e-01 -9.70320031e-02 3.05600405e-01 -8.70964825e-01 2.25182861e-01 -3.68491150e-02 5.13068318e-01 -6.37067676e-01 1.05201498e-01 -7.11358786e-01 -2.83027381e-01 3.59273672e-01 -6.06232822e-01 -3.78751695e-01 2.75937039e-02 4.57596481e-01 -2.06068873e-01 -3.74631256e-01 1.13590157e+00 4.83329175e-03 -6.80238128e-01 6.73947573e-01 7.04132542e-02 2.86014378e-01 1.00993931e+00 -1.75721407e-01 -1.23094298e-01 5.13641462e-02 -1.08816612e+00 4.96398717e-01 2.87040442e-01 1.97176948e-01 4.89367306e-01 -1.07462442e+00 -4.89565909e-01 3.03622484e-01 2.31752284e-02 6.09117985e-01 2.32330710e-01 9.16372418e-01 -3.29714715e-01 1.42119512e-01 -1.24555953e-01 -8.99765968e-01 -9.38338101e-01 3.94777566e-01 2.61911124e-01 -1.75614536e-01 -9.76717234e-01 9.74462569e-01 7.66136050e-02 -4.86686379e-01 3.69842172e-01 -5.86265087e-01 -2.36719266e-01 -5.90266809e-02 5.07907152e-01 1.13468170e-01 -4.24795374e-02 -5.42243123e-01 -2.74086982e-01 4.79330510e-01 -4.45184797e-01 1.32115737e-01 1.36120975e+00 1.37126982e-01 1.65606782e-01 3.40717196e-01 1.17874229e+00 -2.26582617e-01 -1.68040895e+00 -6.00545108e-01 -1.76322572e-02 -3.35812956e-01 3.36021870e-01 -9.87932622e-01 -1.50908005e+00 9.91290867e-01 5.78323781e-01 -9.32043940e-02 1.03198457e+00 -2.50550807e-02 1.03460586e+00 3.13711911e-01 4.98162895e-01 -1.14249897e+00 1.77934952e-02 1.51622549e-01 5.60705006e-01 -1.76832414e+00 -3.91580611e-02 -5.53013384e-01 -9.56131995e-01 8.75763118e-01 6.74440145e-01 -1.95603609e-01 7.54095554e-01 3.17744970e-01 2.35092103e-01 6.84023052e-02 -2.82378018e-01 -6.69298545e-02 1.93265021e-01 4.93684173e-01 3.10287774e-01 1.79161370e-01 -1.86916232e-01 8.37519407e-01 -1.06166929e-01 1.82946622e-01 3.01491052e-01 8.86246383e-01 -3.05664361e-01 -9.86803055e-01 -8.63008797e-02 6.97457194e-01 -4.24428165e-01 -7.19698519e-02 3.44923064e-02 7.41683304e-01 1.79634213e-01 7.97837436e-01 3.78189385e-02 -8.15823004e-02 -4.74621914e-03 -7.42061213e-02 2.61464208e-01 -6.21001601e-01 -3.27378839e-01 2.49554619e-01 -3.48481297e-01 -3.29352349e-01 -4.16362613e-01 -6.57952309e-01 -1.82381785e+00 2.45993435e-01 -3.85688454e-01 1.58629149e-01 5.33679903e-01 1.08001971e+00 2.13675141e-01 5.26068568e-01 6.38017952e-01 -6.05778933e-01 -5.31461358e-01 -8.87420833e-01 -4.92204964e-01 4.86941844e-01 2.97221929e-01 -7.14423060e-01 -1.93913862e-01 1.67645827e-01]
[14.621512413024902, -2.1378211975097656]
6bb01454-2d03-4731-8a45-a3b61dad94d1
tcgm-an-information-theoretic-framework-for
2007.06793
null
https://arxiv.org/abs/2007.06793v1
https://arxiv.org/pdf/2007.06793v1.pdf
TCGM: An Information-Theoretic Framework for Semi-Supervised Multi-Modality Learning
Fusing data from multiple modalities provides more information to train machine learning systems. However, it is prohibitively expensive and time-consuming to label each modality with a large amount of data, which leads to a crucial problem of semi-supervised multi-modal learning. Existing methods suffer from either ineffective fusion across modalities or lack of theoretical guarantees under proper assumptions. In this paper, we propose a novel information-theoretic approach, namely \textbf{T}otal \textbf{C}orrelation \textbf{G}ain \textbf{M}aximization (TCGM), for semi-supervised multi-modal learning, which is endowed with promising properties: (i) it can utilize effectively the information across different modalities of unlabeled data points to facilitate training classifiers of each modality (ii) it has theoretical guarantee to identify Bayesian classifiers, i.e., the ground truth posteriors of all modalities. Specifically, by maximizing TC-induced loss (namely TC gain) over classifiers of all modalities, these classifiers can cooperatively discover the equivalent class of ground-truth classifiers; and identify the unique ones by leveraging limited percentage of labeled data. We apply our method to various tasks and achieve state-of-the-art results, including news classification, emotion recognition and disease prediction.
['Shanghang Zhang', 'Lingjing Hu', 'Xinwei Sun', 'Yilun Xu', 'Yuqing Kong', 'Peng Cao', 'Yizhou Wang']
2020-07-14
null
https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/6209_ECCV_2020_paper.php
https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123480171.pdf
eccv-2020-8
['news-classification']
['natural-language-processing']
[ 3.70716095e-01 2.20260337e-01 -4.88602787e-01 -3.66956592e-01 -1.35173607e+00 -4.87591267e-01 4.19566810e-01 3.43900442e-01 -1.77042708e-01 8.83603692e-01 5.45912012e-02 1.50857434e-01 -2.89639890e-01 -5.07272720e-01 -8.02546382e-01 -1.21236646e+00 2.42110133e-01 3.71512085e-01 -6.46989467e-03 1.99596718e-01 -2.53189832e-01 7.32452720e-02 -1.64439380e+00 5.26918590e-01 7.99936354e-01 1.60465169e+00 -6.88437670e-02 2.10986510e-01 1.01898745e-01 8.92353296e-01 -1.96451530e-01 -4.62218881e-01 -1.65654138e-01 -4.60010171e-01 -7.67353475e-01 2.48722360e-01 3.09909731e-02 9.58251283e-02 -2.12427750e-01 1.36085367e+00 3.05103898e-01 5.68568856e-02 9.26820099e-01 -1.55257905e+00 -3.63186806e-01 6.92166030e-01 -9.09375489e-01 -3.24265324e-02 2.90741205e-01 -4.50483918e-01 1.01931357e+00 -7.55348265e-01 3.97214770e-01 1.10722232e+00 4.63769138e-01 5.19375801e-01 -1.04123914e+00 -5.52370608e-01 2.72656351e-01 1.45805702e-01 -1.58562005e+00 -5.55122375e-01 7.79359877e-01 -2.24734455e-01 8.83911848e-02 3.95985365e-01 2.26594970e-01 9.91031408e-01 -1.52962692e-02 1.24391019e+00 1.38924670e+00 -5.31617045e-01 2.49097392e-01 4.04804289e-01 1.28309965e-01 7.21870720e-01 -1.85483530e-01 -8.11163615e-03 -8.82635117e-01 -3.29243690e-01 1.51414990e-01 2.24404648e-01 -3.95595610e-01 -2.30009571e-01 -1.36683023e+00 6.12339616e-01 2.02265799e-01 2.73590088e-01 -4.88373905e-01 -2.03583956e-01 3.07183594e-01 2.25747690e-01 4.80679005e-01 -3.82799745e-01 -5.34332037e-01 3.25289696e-01 -6.91598356e-01 -2.32337117e-01 5.04299283e-01 9.03781593e-01 8.91201079e-01 -3.63183975e-01 4.35292944e-02 7.83660293e-01 6.32291019e-01 7.19504297e-01 4.04724926e-01 -8.19691539e-01 4.67861772e-01 6.16389096e-01 -4.88521606e-02 -7.21313238e-01 -4.42983627e-01 -1.23924591e-01 -1.20495689e+00 -2.60020405e-01 2.93275744e-01 -3.93002570e-01 -5.62563479e-01 1.93642664e+00 6.74903274e-01 1.38458595e-01 4.33116078e-01 8.65183294e-01 9.30491447e-01 5.78806520e-01 2.24650666e-01 -6.98990941e-01 1.41266334e+00 -5.50295055e-01 -6.96786821e-01 -1.36918709e-01 5.91458857e-01 -5.31064034e-01 5.55041671e-01 3.99782896e-01 -7.16850400e-01 -2.26306394e-01 -9.24447060e-01 5.72692275e-01 -2.19707340e-01 3.81020874e-01 6.51317596e-01 6.99250519e-01 -5.35486639e-01 1.38111979e-01 -6.88343048e-01 -1.35369882e-01 6.25205040e-01 3.49464715e-01 -6.38574779e-01 -3.48056018e-01 -1.17572582e+00 6.51212037e-01 7.61953354e-01 1.96389437e-01 -8.96425068e-01 -3.11966866e-01 -8.75788927e-01 -2.42039829e-01 6.62280083e-01 -5.68035960e-01 9.87999201e-01 -1.20501280e+00 -1.15034389e+00 7.76570618e-01 -2.83183694e-01 -3.18088718e-02 3.77649337e-01 3.54826376e-02 -7.09651411e-01 2.71335483e-01 5.48823774e-02 5.39511383e-01 9.63525712e-01 -1.56548905e+00 -1.07490563e+00 -6.72247708e-01 -1.10617548e-01 4.04646903e-01 -5.86053073e-01 1.41582545e-02 -2.53757000e-01 -4.33197021e-01 5.91661751e-01 -8.11853588e-01 1.64192975e-01 -2.14628145e-01 -7.53637612e-01 -4.06789482e-01 8.66655231e-01 -3.57803583e-01 9.72529411e-01 -2.20725608e+00 3.25829536e-01 2.16900587e-01 3.17114234e-01 -7.22989067e-02 2.24402830e-01 3.14599313e-02 1.50215402e-01 -1.16612606e-01 -1.24683686e-01 -2.89881378e-01 -6.10023066e-02 2.96167254e-01 -3.49550992e-01 6.23473048e-01 -1.00135421e-02 5.95319927e-01 -7.70281017e-01 -9.26960468e-01 2.82480240e-01 2.73515642e-01 -1.35748954e-02 8.71539395e-03 -2.20525935e-01 5.65888286e-01 -7.08579540e-01 1.11247301e+00 5.92328012e-01 -4.73135501e-01 1.59629673e-01 -5.64115703e-01 3.34054947e-01 -3.75791728e-01 -1.35038114e+00 1.64011717e+00 -1.60211891e-01 8.88353288e-02 2.08488449e-01 -1.39822865e+00 5.85787833e-01 7.55630255e-01 7.10996568e-01 -3.09789836e-01 3.45140308e-01 1.77574158e-01 -5.43246567e-01 -5.29980183e-01 -1.22177154e-01 -4.96758431e-01 -3.78388911e-01 6.03549123e-01 3.52250993e-01 3.08724791e-01 -1.34500548e-01 2.14373559e-01 6.29885018e-01 -1.07476048e-01 2.21855193e-01 7.87431549e-04 5.55274010e-01 -1.41228229e-01 6.86641395e-01 7.13326514e-01 -2.21731916e-01 2.09553540e-01 2.78832346e-01 5.90226464e-02 -6.00911558e-01 -1.06096816e+00 -2.79346704e-01 1.18520331e+00 4.11908060e-01 -6.75370842e-02 -4.54377651e-01 -9.66826260e-01 -2.03537598e-01 4.71581042e-01 -6.37913585e-01 -3.73925030e-01 1.24236248e-01 -1.07662082e+00 6.10122561e-01 4.39966381e-01 5.53693891e-01 -5.33217788e-01 -2.80054331e-01 -1.40694097e-01 -5.71333826e-01 -1.02409637e+00 -1.99309573e-01 2.73756564e-01 -7.94429779e-01 -1.16054344e+00 -5.45675457e-01 -5.25802195e-01 7.94164121e-01 1.81262434e-01 7.15101600e-01 -2.41551742e-01 1.54283419e-01 3.94659102e-01 -4.93926436e-01 -5.06350040e-01 -3.12030613e-01 -2.42393911e-01 2.38008425e-01 5.83686471e-01 4.11544770e-01 -2.46150136e-01 -2.95766920e-01 5.92646599e-01 -9.28093553e-01 1.09795414e-01 7.05911696e-01 8.90748739e-01 9.15668130e-01 4.58905935e-01 8.85096908e-01 -8.20230067e-01 2.35552087e-01 -6.02427423e-01 -2.60429680e-01 8.52084994e-01 -4.09741163e-01 -8.43130201e-02 4.29326385e-01 -5.36118686e-01 -1.28535879e+00 2.21091449e-01 2.58027405e-01 -5.23485243e-01 -3.20509791e-01 8.93324614e-01 -4.91008341e-01 1.36508480e-01 4.59714949e-01 2.69819647e-01 -1.37004763e-01 -3.18255931e-01 4.40044314e-01 9.42859471e-01 5.44350684e-01 -6.93111658e-01 5.40393114e-01 7.02153504e-01 1.41477704e-01 -5.88854432e-01 -1.40517724e+00 -3.68704766e-01 -5.42976379e-01 -5.22620678e-01 6.79709494e-01 -1.16258633e+00 -8.98131311e-01 4.71127272e-01 -6.81284010e-01 4.13708657e-01 5.54528087e-02 8.61844957e-01 -2.98248649e-01 4.57701772e-01 -3.36745471e-01 -1.08880699e+00 -1.61080495e-01 -9.31525350e-01 1.16124606e+00 5.24475336e-01 2.19664887e-01 -9.40913975e-01 -3.12565774e-01 7.36254394e-01 -1.51562810e-01 3.28771442e-01 8.51296484e-01 -7.85447061e-01 -2.61756480e-01 -6.26719773e-01 -2.97911316e-01 4.22929704e-01 1.18727967e-01 -2.59401351e-01 -1.30413628e+00 -1.87697098e-01 5.74163208e-03 -9.57581460e-01 9.43627298e-01 3.91897351e-01 1.13561249e+00 -3.07773471e-01 -5.93610525e-01 1.05845943e-01 1.09073079e+00 2.63424758e-02 1.06633361e-02 -2.81466097e-01 5.87648511e-01 6.27699018e-01 8.07793915e-01 5.93182206e-01 5.61178863e-01 4.17716444e-01 3.51006657e-01 1.28551096e-01 3.68900150e-01 -1.08734593e-01 3.26310545e-01 8.07027102e-01 2.40972312e-03 -3.53455156e-01 -6.40751600e-01 3.01455289e-01 -2.19461560e+00 -7.64122486e-01 2.79563665e-01 2.25723076e+00 8.75527143e-01 -8.72176439e-02 7.99490064e-02 3.20971787e-01 1.02899396e+00 -1.10003375e-01 -7.12023318e-01 4.59389657e-01 -4.67045695e-01 -3.75646889e-01 1.60518557e-01 8.62882286e-02 -1.51815236e+00 3.66502732e-01 5.35001659e+00 1.21628499e+00 -8.91985893e-01 4.42848533e-01 7.23102570e-01 -1.38284862e-01 -2.36672133e-01 -1.73239745e-02 -6.99716032e-01 4.44662511e-01 7.04291165e-01 -2.47301087e-02 3.25351864e-01 6.32721663e-01 -2.31839493e-01 -4.32460487e-01 -1.15874851e+00 9.63608205e-01 3.17460060e-01 -1.00612879e+00 -1.24753058e-01 -1.13909587e-01 8.33703220e-01 -1.23258464e-01 -7.03568161e-02 3.12803030e-01 3.04896593e-01 -7.82906711e-01 8.09251130e-01 6.70240819e-01 7.81866252e-01 -6.94061756e-01 8.23302507e-01 7.71745980e-01 -1.15966606e+00 -2.23553255e-01 -1.53415054e-01 3.95087957e-01 8.31488669e-02 8.09211791e-01 -1.58658892e-01 1.06190765e+00 6.06573522e-01 6.81667805e-01 -3.23434323e-01 6.99659824e-01 -4.50985357e-02 5.58579803e-01 -6.04237795e-01 6.20234013e-02 -2.40505859e-02 8.42387080e-02 3.24936301e-01 7.11613834e-01 2.43563801e-01 2.90429503e-01 5.72057366e-01 2.83879310e-01 -3.28758329e-01 2.63736565e-02 -1.76602513e-01 -1.67870834e-01 7.44586706e-01 1.26560223e+00 -6.78399265e-01 -3.31705272e-01 -4.94018853e-01 5.75237989e-01 2.09519535e-01 2.45473057e-01 -8.39293301e-01 -1.08702518e-01 -9.48052555e-02 -6.11589849e-01 -8.42156559e-02 5.04931688e-01 -2.88093209e-01 -1.38807058e+00 1.17378738e-02 -6.22640908e-01 1.12771499e+00 -7.66494989e-01 -1.67134500e+00 3.59778047e-01 1.40288040e-01 -1.40955079e+00 3.94051261e-02 -3.85241896e-01 -2.37838164e-01 6.31897688e-01 -1.26133716e+00 -1.47344828e+00 -2.64690399e-01 1.02291727e+00 1.70257706e-02 -3.10072303e-01 8.75618935e-01 4.32297617e-01 -6.63123965e-01 6.18053854e-01 2.52402842e-01 3.95908244e-02 7.02498853e-01 -1.10233867e+00 -9.85868990e-01 5.92888832e-01 1.45732373e-01 1.64849296e-01 5.03724337e-01 -4.38299000e-01 -1.51729774e+00 -9.80071306e-01 6.53522968e-01 -2.09777325e-01 5.13120711e-01 1.39116004e-01 -7.33515441e-01 5.95997632e-01 -9.67587680e-02 2.19815508e-01 1.01319385e+00 3.36952180e-01 -4.07040328e-01 -2.65706331e-01 -1.20696652e+00 2.52658278e-01 5.06558239e-01 -6.94529235e-01 -4.82094139e-01 6.62706077e-01 4.38373357e-01 -2.73469180e-01 -1.17293537e+00 6.36264324e-01 4.95019048e-01 -6.74572110e-01 7.53478289e-01 -6.46290660e-01 3.78260046e-01 -4.11678374e-01 -5.67871809e-01 -1.07472301e+00 2.65831470e-01 -2.41678193e-01 -4.81575996e-01 1.37427199e+00 4.80049372e-01 -6.18385792e-01 5.75414717e-01 6.90071046e-01 6.03101887e-02 -7.20145762e-01 -1.43318188e+00 -4.51949447e-01 -1.95789173e-01 -5.11377990e-01 2.93742716e-01 1.13253582e+00 3.65848422e-01 2.61394024e-01 -5.18077493e-01 4.38625962e-01 8.49846900e-01 4.11796093e-01 2.10657254e-01 -1.30625451e+00 -2.09691346e-01 -2.18891844e-01 -2.07447380e-01 -7.77709603e-01 2.84781486e-01 -8.36386621e-01 1.42359138e-01 -1.21158338e+00 6.95325553e-01 -6.47673547e-01 -7.51395583e-01 8.43630493e-01 -9.26435962e-02 3.35996389e-01 -5.54566979e-02 5.08391976e-01 -1.12905383e+00 5.62320232e-01 1.09317183e+00 -1.55403957e-01 8.36192071e-02 2.36243650e-01 -7.26906955e-01 8.55506778e-01 3.75997037e-01 -5.45139015e-01 -3.83415729e-01 -1.46232113e-01 3.03247660e-01 5.75271070e-01 5.40507674e-01 -6.90627217e-01 3.62566113e-01 -2.83266574e-01 4.54551369e-01 -7.34393239e-01 6.08770251e-01 -8.87789428e-01 2.15242356e-01 8.95613357e-02 -4.30340886e-01 -6.19062364e-01 -9.26107317e-02 9.77281272e-01 -3.53575319e-01 -2.81009436e-01 7.64463484e-01 7.09394217e-02 -5.08754671e-01 2.57523775e-01 -2.12140262e-01 -3.81915532e-02 1.21995127e+00 1.91576570e-01 -4.87961859e-01 -3.05075884e-01 -8.81154120e-01 3.51210117e-01 8.30188543e-02 2.08740845e-01 4.11663532e-01 -1.50518000e+00 -5.72151184e-01 -6.61491156e-02 4.06078100e-01 1.82099771e-02 7.07248569e-01 1.15596437e+00 2.98924983e-01 2.75674671e-01 1.19149916e-01 -8.77303123e-01 -1.20479953e+00 5.56952596e-01 1.79439634e-01 2.11510677e-02 -8.73413160e-02 9.21649992e-01 4.08114903e-02 -4.72623229e-01 2.71140367e-01 2.98498511e-01 -1.66394085e-01 4.53509837e-01 4.09236968e-01 2.32088134e-01 4.02322598e-02 -8.60523403e-01 -5.13970435e-01 3.44917685e-01 -1.03861883e-01 -1.34506553e-01 9.52886403e-01 -4.11664218e-01 -1.99873462e-01 6.45112336e-01 9.95266974e-01 -3.90390635e-01 -1.08876610e+00 -8.78829122e-01 -1.76688194e-01 -2.52277911e-01 4.14656729e-01 -1.01880741e+00 -1.09293580e+00 7.50647783e-01 6.50594115e-01 1.82363480e-01 1.45340967e+00 5.25383711e-01 4.97916490e-01 4.66082036e-01 4.90707397e-01 -1.22897589e+00 -5.21170795e-02 -6.44504139e-03 3.74179542e-01 -1.50978851e+00 1.83260869e-02 -3.98710281e-01 -9.05802190e-01 9.38364387e-01 3.15505505e-01 6.26777649e-01 1.00190163e+00 -4.50636353e-03 -6.74279481e-02 -2.44507670e-01 -7.50184476e-01 -2.02108875e-01 5.22754788e-01 3.23106140e-01 1.31811351e-01 3.25862557e-01 9.20430273e-02 9.65066850e-01 4.39758420e-01 5.79165155e-03 -4.15679179e-02 1.05399120e+00 -4.97897685e-01 -8.11972976e-01 -5.76933682e-01 7.75071263e-01 -4.13486123e-01 1.97287321e-01 -1.86300352e-01 3.70019436e-01 3.09471607e-01 1.32843578e+00 -4.54837352e-01 -5.52046776e-01 -1.12208478e-01 1.73280373e-01 4.92210329e-01 -2.44214892e-01 5.95711805e-02 4.22151804e-01 1.20509090e-02 -4.78109419e-02 -1.00588036e+00 -1.11121428e+00 -1.14829445e+00 -8.17185119e-02 -8.06543112e-01 1.85639799e-01 4.29145932e-01 1.52798796e+00 1.40426487e-01 2.74239123e-01 7.55911767e-01 -5.91381192e-01 -6.36363268e-01 -9.62085485e-01 -8.68509650e-01 3.87246937e-01 1.57364935e-01 -9.24568892e-01 -3.69342327e-01 2.75418758e-01]
[12.878022193908691, 4.872745037078857]
c99fc710-1bde-4934-b23c-067fe8df19c9
improving-eeg-decoding-via-clustering-based
2012.06813
null
https://arxiv.org/abs/2012.06813v1
https://arxiv.org/pdf/2012.06813v1.pdf
Improving EEG Decoding via Clustering-based Multi-task Feature Learning
Accurate electroencephalogram (EEG) pattern decoding for specific mental tasks is one of the key steps for the development of brain-computer interface (BCI), which is quite challenging due to the considerably low signal-to-noise ratio of EEG collected at the brain scalp. Machine learning provides a promising technique to optimize EEG patterns toward better decoding accuracy. However, existing algorithms do not effectively explore the underlying data structure capturing the true EEG sample distribution, and hence can only yield a suboptimal decoding accuracy. To uncover the intrinsic distribution structure of EEG data, we propose a clustering-based multi-task feature learning algorithm for improved EEG pattern decoding. Specifically, we perform affinity propagation-based clustering to explore the subclasses (i.e., clusters) in each of the original classes, and then assign each subclass a unique label based on a one-versus-all encoding strategy. With the encoded label matrix, we devise a novel multi-task learning algorithm by exploiting the subclass relationship to jointly optimize the EEG pattern features from the uncovered subclasses. We then train a linear support vector machine with the optimized features for EEG pattern decoding. Extensive experimental studies are conducted on three EEG datasets to validate the effectiveness of our algorithm in comparison with other state-of-the-art approaches. The improved experimental results demonstrate the outstanding superiority of our algorithm, suggesting its prominent performance for EEG pattern decoding in BCI applications.
['Andrzej Cichocki', 'Guoxu Zhou', 'Hongru Zhu', 'Hua Xie', 'Wei Wu', 'Tao Zhou', 'Yu Zhang']
2020-12-12
null
null
null
null
['eeg-decoding', 'eeg-decoding']
['medical', 'time-series']
[ 5.38940191e-01 -5.79809070e-01 1.56493887e-01 -4.38232094e-01 -6.02127552e-01 -1.40179649e-01 -1.31973699e-01 5.18972389e-02 -7.50895739e-02 8.09655905e-01 -8.78043100e-02 -3.41821797e-02 -8.88415158e-01 -3.32358539e-01 -7.17180669e-01 -1.13909459e+00 -9.56225246e-02 4.28800464e-01 -2.27161348e-01 1.64075315e-01 5.08985221e-01 1.07351713e-01 -1.53867209e+00 5.98369300e-01 1.19069338e+00 1.20110893e+00 4.85571325e-01 2.05127940e-01 1.01143189e-01 2.89586931e-01 -6.60130024e-01 -2.02791411e-02 -1.17492743e-01 -5.36635041e-01 -4.44480300e-01 1.35579690e-01 -3.51577759e-01 2.36829117e-01 -3.38241249e-01 1.26412106e+00 5.02425015e-01 -1.24491900e-01 8.46337020e-01 -1.29864132e+00 -4.84028816e-01 3.73362571e-01 -7.97630727e-01 4.46301430e-01 2.36628011e-01 -2.59669662e-01 8.07940364e-01 -8.15066576e-01 -2.58707136e-01 4.95608836e-01 1.26949534e-01 2.76671767e-01 -1.37010241e+00 -1.27563214e+00 9.94255468e-02 7.88588822e-01 -1.95903349e+00 -3.31475407e-01 9.33662295e-01 -4.72662657e-01 6.47536159e-01 2.34486938e-01 9.36398566e-01 1.07696962e+00 6.28065526e-01 8.61797750e-01 1.44529450e+00 -9.40907300e-02 2.95625865e-01 1.65186837e-01 3.48949522e-01 5.20526290e-01 9.54860225e-02 -2.53620416e-01 -9.93171990e-01 -2.10515007e-01 4.53432620e-01 1.28580138e-01 -7.27561474e-01 -2.16624781e-01 -1.35422122e+00 4.22346294e-01 1.01858713e-01 3.13673496e-01 -5.81842661e-01 -2.21074745e-01 1.44132122e-01 2.86737204e-01 3.23280811e-01 3.12808931e-01 -4.67975169e-01 -2.27516547e-01 -9.29432154e-01 1.36952683e-01 5.43454289e-01 8.53171051e-01 9.67180371e-01 -2.78990924e-01 -3.04569811e-01 1.12020469e+00 1.76918745e-01 4.08249974e-01 6.55260563e-01 -3.71237785e-01 4.40030992e-01 6.09736562e-01 -1.66501656e-01 -1.06505680e+00 -5.66241801e-01 -6.87267601e-01 -1.08203721e+00 -3.01159441e-01 8.44904408e-02 -2.37001389e-01 -4.84599262e-01 1.48819625e+00 -1.25597462e-01 6.17333889e-01 -1.23083010e-01 7.96761394e-01 2.74094462e-01 6.50530815e-01 -2.30255783e-01 -3.63505453e-01 1.26679265e+00 -5.02075613e-01 -6.09828949e-01 -2.21222356e-01 4.30379927e-01 -2.92147994e-01 8.79426718e-01 8.78013015e-01 -5.63762486e-01 -4.05147880e-01 -1.11043453e+00 8.32431972e-01 1.56639069e-01 2.24913448e-01 6.44456506e-01 6.58900559e-01 -6.83669209e-01 3.17718714e-01 -7.50341535e-01 2.56549269e-01 7.56164789e-01 7.94282854e-01 -3.03196967e-01 -7.67867193e-02 -9.28338230e-01 5.84805012e-01 4.94136393e-01 2.03938559e-01 -6.59050226e-01 -6.74832642e-01 -4.86272454e-01 2.27569282e-01 2.16894343e-01 -3.38002741e-01 6.11047626e-01 -8.94257545e-01 -1.44192135e+00 3.54171604e-01 -5.41307867e-01 -4.28734310e-02 -2.99822360e-01 1.99402452e-01 -4.97965544e-01 6.11927062e-02 -6.92149326e-02 4.68750596e-01 1.05973554e+00 -1.26332390e+00 -9.40204680e-01 -6.53310895e-01 -4.92837697e-01 2.03757271e-01 -9.24017072e-01 -9.18296427e-02 -2.47012705e-01 -4.95689094e-01 2.45580435e-01 -7.36584246e-01 1.97148383e-01 -6.03838921e-01 -3.61102402e-01 -3.55739623e-01 5.10123014e-01 -4.67850447e-01 1.45301330e+00 -2.30552030e+00 6.07470334e-01 7.33888388e-01 3.76364410e-01 -4.41443063e-02 -1.01783104e-01 -1.19238779e-01 -1.71249017e-01 -3.73658538e-01 -4.01477665e-01 -5.57297207e-02 -2.56371319e-01 1.90997440e-02 -6.51060790e-02 5.97526312e-01 1.07861208e-02 7.48991251e-01 -7.64170706e-01 -1.81077272e-01 1.19339943e-01 3.12418342e-01 -4.43417758e-01 3.75280529e-01 2.62398452e-01 8.89379501e-01 -6.04129970e-01 5.62136114e-01 4.59737867e-01 -4.26539421e-01 -1.28327729e-02 -2.94272304e-01 9.09797549e-02 -1.90545097e-02 -9.67533648e-01 1.74886000e+00 -2.52496153e-01 5.44859111e-01 -3.07412684e-01 -1.50367141e+00 8.82097125e-01 3.30903620e-01 8.48366201e-01 -7.46122897e-01 2.64270574e-01 3.64058673e-01 6.62941396e-01 -6.67090774e-01 -2.70913273e-01 -1.80096045e-01 2.64021568e-02 5.47579646e-01 1.51812375e-01 2.78960913e-01 -4.29706722e-02 -2.71884471e-01 1.00924850e+00 -2.89664656e-01 1.18964866e-01 -5.54361165e-01 5.97068071e-01 -4.32293922e-01 6.23938024e-01 5.72926819e-01 -3.49587612e-02 4.46726412e-01 4.32649702e-01 -2.73838073e-01 -5.58547497e-01 -7.73949444e-01 -5.82727730e-01 7.87802696e-01 2.58905202e-01 -3.13491344e-01 -8.78676593e-01 -2.94747889e-01 -1.18923977e-01 5.99618196e-01 -5.46834946e-01 -5.36757588e-01 -2.41455004e-01 -1.44428384e+00 4.29717809e-01 2.88781703e-01 3.22059393e-01 -9.86773372e-01 -5.55404186e-01 3.76910806e-01 -3.82700354e-01 -8.00164580e-01 -1.56566292e-01 4.40730482e-01 -5.88802516e-01 -9.42902505e-01 -7.36158967e-01 -8.93866777e-01 6.80968583e-01 2.31083661e-01 3.56389552e-01 2.65696421e-02 -1.26145199e-01 1.50805056e-01 -4.42752123e-01 -4.82346952e-01 1.39921412e-01 1.35252878e-01 2.89166212e-01 6.49403214e-01 7.21650124e-01 -9.71855402e-01 -7.15520918e-01 4.59201485e-01 -6.22636914e-01 2.24443763e-01 8.36854100e-01 1.02649546e+00 6.10783458e-01 5.56214154e-01 9.61659253e-01 -5.77011585e-01 7.58870840e-01 -6.64604962e-01 -3.18833590e-01 5.32687128e-01 -4.30635005e-01 9.73626878e-03 6.27262235e-01 -4.96755809e-01 -6.72518373e-01 6.92811608e-02 -8.71134996e-02 -3.76502246e-01 -2.73296744e-01 6.89828575e-01 -4.32770103e-01 -4.46538031e-01 2.45060682e-01 8.83721590e-01 -1.84413835e-01 -2.19858453e-01 -2.86495119e-01 1.04841828e+00 3.83569926e-01 -7.45423198e-01 2.31994241e-01 6.53133690e-02 -2.19770789e-01 -8.66071403e-01 -6.31987333e-01 -3.86793852e-01 -5.99714339e-01 -2.47172773e-01 7.23024547e-01 -7.28649914e-01 -9.06661212e-01 6.98858023e-01 -9.93528724e-01 -1.16544075e-01 4.92761075e-01 6.78824008e-01 -4.89914745e-01 2.67612755e-01 -2.83692002e-01 -7.88650215e-01 -3.04022521e-01 -1.58461487e+00 9.00608242e-01 4.53818478e-02 -2.64721036e-01 -6.47200346e-01 -1.78000778e-01 2.34068900e-01 6.41052946e-02 -2.10846841e-01 1.44793487e+00 -7.77880549e-01 -3.86919767e-01 -6.13112673e-02 -4.52027798e-01 7.79533237e-02 3.41363072e-01 -5.55504501e-01 -9.28777635e-01 -3.85343790e-01 2.31187329e-01 -1.90022275e-01 5.65001190e-01 4.51073319e-01 1.89003766e+00 -1.81048457e-02 -5.49103200e-01 8.53482604e-01 1.20259583e+00 6.92370892e-01 5.54697812e-01 2.69048095e-01 7.42230415e-01 5.38319409e-01 1.40702561e-01 7.91528940e-01 4.67295259e-01 5.38789034e-01 1.56873703e-01 3.14152390e-01 4.04496014e-01 1.82974353e-01 6.09913468e-02 1.17422819e+00 -3.76004986e-02 -2.50523984e-01 -9.22747493e-01 3.69516015e-01 -1.74894559e+00 -4.78300363e-01 7.42237419e-02 2.15504265e+00 7.89909661e-01 -3.45857069e-02 -5.12399897e-02 5.25484800e-01 6.73656464e-01 -3.01826477e-01 -6.20205760e-01 1.52669013e-01 -4.62133326e-02 5.48204958e-01 1.37854636e-01 -7.41353557e-02 -9.40993130e-01 5.58116019e-01 5.52690220e+00 9.99962747e-01 -1.17456853e+00 1.93730786e-01 5.66599429e-01 -2.91942328e-01 -1.16569720e-01 -4.28680986e-01 -7.30438352e-01 9.96624470e-01 9.10924673e-01 -2.85399616e-01 8.97210419e-01 2.18576834e-01 1.77469701e-01 7.44132251e-02 -1.17664754e+00 1.61876607e+00 3.33261728e-01 -1.13759327e+00 1.15924716e-01 1.26774669e-01 4.64384019e-01 -1.28631890e-01 9.58135650e-02 6.80647343e-02 -4.79658365e-01 -1.04751503e+00 7.69226372e-01 5.48579276e-01 8.50756288e-01 -9.40719008e-01 5.83738387e-01 4.58010256e-01 -1.05469716e+00 -4.91669238e-01 -4.72298831e-01 3.25398780e-02 -5.08540347e-02 5.62066317e-01 -1.90045401e-01 4.65408802e-01 7.85043299e-01 1.03684819e+00 -5.34326494e-01 1.46967769e+00 -3.94987352e-02 6.70497596e-01 -7.79519379e-02 -3.44324529e-01 -1.26643050e-02 -4.56129849e-01 4.85141098e-01 1.05868566e+00 6.63624048e-01 4.79043335e-01 1.82178300e-02 7.30754077e-01 2.05532741e-02 2.33802572e-01 -3.58965099e-01 9.06218365e-02 5.18751025e-01 1.09831333e+00 -7.75116503e-01 -7.47935474e-02 -4.97568697e-01 1.10716426e+00 4.96953517e-01 3.72164100e-01 -7.01502860e-01 -6.57309532e-01 6.02368355e-01 -3.26642334e-01 9.75195691e-02 -6.93816990e-02 -7.51292944e-01 -1.21023285e+00 1.75723314e-01 -1.02883577e+00 2.27635190e-01 -4.94593382e-01 -1.24460459e+00 6.96429491e-01 5.31919338e-02 -1.14157367e+00 1.54599816e-01 -6.30156279e-01 -6.61831439e-01 6.87872529e-01 -1.35224545e+00 -6.92900300e-01 -2.62484908e-01 9.17695045e-01 5.59338808e-01 -4.88145590e-01 8.64934266e-01 5.48055649e-01 -9.43059087e-01 7.41741896e-01 4.03321415e-01 -1.56988263e-01 4.65449661e-01 -8.82146299e-01 -3.86734277e-01 5.34479618e-01 3.08172792e-01 5.03440142e-01 2.73543805e-01 -3.62495154e-01 -1.69150639e+00 -8.93521011e-01 3.66119593e-01 -5.10160252e-02 7.72228539e-01 -5.43506324e-01 -1.15964580e+00 3.85372490e-01 9.21962336e-02 -5.41316628e-01 1.25705624e+00 1.58980697e-01 1.73705593e-01 -4.46596891e-01 -6.39696360e-01 3.88920724e-01 9.35468912e-01 -3.44237119e-01 -5.96991301e-01 4.40972805e-01 1.97623923e-01 -2.35669818e-02 -8.52649510e-01 3.91761929e-01 7.26630390e-01 -6.57094717e-01 7.19860554e-01 -2.64425844e-01 3.35477293e-01 -2.91144729e-01 -1.21326946e-01 -1.83225286e+00 -6.69186771e-01 -3.68013859e-01 -1.81000680e-02 8.95973504e-01 5.80849290e-01 -8.68420839e-01 5.75396240e-01 4.06345338e-01 -3.31846774e-01 -1.16158438e+00 -9.34625506e-01 -4.57620263e-01 -1.67771205e-01 -6.01012111e-01 8.82384181e-01 6.07763231e-01 4.82100517e-01 2.42280513e-01 -4.50685233e-01 3.50832373e-01 7.37235427e-01 3.23997080e-01 3.07031751e-01 -1.18456638e+00 -3.62768531e-01 -6.27755582e-01 -5.64294100e-01 -1.02897894e+00 5.35908937e-01 -1.14403689e+00 2.71038890e-01 -1.23228240e+00 5.97725749e-01 -6.47131383e-01 -9.85795081e-01 5.16362190e-01 -4.32970732e-01 2.87270129e-01 -2.28726640e-01 4.04094428e-01 -4.86258835e-01 6.96850538e-01 9.25345004e-01 -2.34244674e-01 -2.69562632e-01 1.93820044e-01 -8.43124211e-01 5.15669882e-01 7.75602341e-01 -7.04162896e-01 -5.89602292e-01 -6.40325367e-01 1.00424469e-01 8.82308185e-02 1.11404069e-01 -1.28350282e+00 4.46651042e-01 3.62264365e-02 6.73883975e-01 -3.84446442e-01 2.74865210e-01 -7.47791231e-01 1.72106624e-01 2.25908250e-01 -1.94715649e-01 -1.26127452e-01 1.22742511e-01 6.85700357e-01 -2.21297190e-01 -1.77014366e-01 6.00780904e-01 3.67641151e-01 -5.81333935e-01 4.99819934e-01 -7.09645212e-01 -2.31894821e-01 1.03188789e+00 -2.97439605e-01 6.51830211e-02 2.20624823e-02 -5.11878610e-01 2.46764541e-01 -9.43356007e-02 3.35634112e-01 1.05921912e+00 -1.23138738e+00 -7.65428424e-01 8.60809743e-01 3.59641016e-01 -2.97915310e-01 2.78963536e-01 1.11873269e+00 -6.28477335e-03 3.76043409e-01 -5.44336975e-01 -8.94556344e-01 -1.06221473e+00 2.63760328e-01 1.68037698e-01 1.61975741e-01 -7.28294611e-01 8.72479498e-01 5.05695522e-01 1.13901496e-01 2.53879637e-01 2.63194507e-03 -4.34883356e-01 -7.03733414e-02 7.31218219e-01 1.25485316e-01 3.10112834e-01 -4.57142889e-01 -4.93538648e-01 5.45864344e-01 -9.47306752e-02 7.40873814e-02 1.55358922e+00 2.09225621e-02 -4.23198938e-01 3.57567996e-01 1.24679542e+00 -5.69476366e-01 -1.03949034e+00 -6.83908984e-02 1.08338460e-01 -6.84507072e-01 1.63743421e-01 -5.15060961e-01 -1.10721445e+00 1.00706160e+00 5.29850662e-01 -1.26907542e-01 1.46432388e+00 -1.05516769e-01 8.38962674e-01 4.24533427e-01 8.92580032e-01 -8.24294567e-01 -3.73536721e-02 1.19983494e-01 6.44650519e-01 -9.55172420e-01 -2.08756313e-01 -1.15946867e-01 -6.35132551e-01 1.16664827e+00 5.02308905e-01 -7.23375082e-02 8.02831113e-01 3.00812095e-01 -4.25134510e-01 -3.15902829e-01 -5.53490818e-01 2.69193977e-01 5.63712120e-01 5.52967012e-01 8.74863490e-02 2.37740055e-01 -1.83000833e-01 1.15500915e+00 -8.52668211e-02 2.85463985e-02 7.69110471e-02 5.77116787e-01 -5.27153909e-01 -9.71809149e-01 -2.45064080e-01 1.15035832e+00 -3.68520707e-01 -1.63793504e-01 -9.38795507e-03 1.48108900e-01 7.47028068e-02 1.12375474e+00 -3.14909779e-02 -9.11698222e-01 2.33794644e-01 2.10901603e-01 8.22508931e-01 -5.16466320e-01 -2.86067486e-01 1.98964939e-01 -5.46117306e-01 -2.91039199e-01 -1.50786072e-01 -7.63355076e-01 -1.13259208e+00 1.42581776e-01 -2.85559446e-01 3.16554755e-01 5.65087259e-01 1.32006967e+00 4.80918705e-01 6.81562126e-01 8.40137720e-01 -9.36711967e-01 -1.59317970e-01 -7.96837866e-01 -8.91505897e-01 1.62170216e-01 1.79777026e-01 -1.04150152e+00 -1.86843410e-01 -1.01910777e-01]
[13.133993148803711, 3.468426465988159]
2a090fd6-d756-41c7-93ee-a4c952901bb7
action-localization-through-continual
2003.12185
null
https://arxiv.org/abs/2003.12185v1
https://arxiv.org/pdf/2003.12185v1.pdf
Action Localization through Continual Predictive Learning
The problem of action recognition involves locating the action in the video, both over time and spatially in the image. The dominant current approaches use supervised learning to solve this problem, and require large amounts of annotated training data, in the form of frame-level bounding box annotations around the region of interest. In this paper, we present a new approach based on continual learning that uses feature-level predictions for self-supervision. It does not require any training annotations in terms of frame-level bounding boxes. The approach is inspired by cognitive models of visual event perception that propose a prediction-based approach to event understanding. We use a stack of LSTMs coupled with CNN encoder, along with novel attention mechanisms, to model the events in the video and use this model to predict high-level features for the future frames. The prediction errors are used to continuously learn the parameters of the models. This self-supervised framework is not complicated as other approaches but is very effective in learning robust visual representations for both labeling and localization. It should be noted that the approach outputs in a streaming fashion, requiring only a single pass through the video, making it amenable for real-time processing. We demonstrate this on three datasets - UCF Sports, JHMDB, and THUMOS'13 and show that the proposed approach outperforms weakly-supervised and unsupervised baselines and obtains competitive performance compared to fully supervised baselines. Finally, we show that the proposed framework can generalize to egocentric videos and obtain state-of-the-art results in unsupervised gaze prediction.
['Sudeep Sarkar', 'Sathyanarayanan N. Aakur']
2020-03-26
null
https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/2129_ECCV_2020_paper.php
https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123590290.pdf
eccv-2020-8
['eye-tracking']
['computer-vision']
[ 2.96405762e-01 1.61176056e-01 -4.45181102e-01 -5.82202852e-01 -6.03598714e-01 -1.51947737e-01 5.40035248e-01 2.42572606e-01 -6.70555711e-01 5.22856116e-01 3.89707178e-01 1.23821408e-01 1.69090420e-01 -5.46864033e-01 -1.10001564e+00 -5.75758457e-01 -2.53914922e-01 1.66137695e-01 6.89316928e-01 5.51034175e-02 3.00532579e-01 2.22326592e-01 -2.05258512e+00 7.30736136e-01 4.09426421e-01 1.08563900e+00 3.03015590e-01 7.31836081e-01 1.74920887e-01 1.56110501e+00 -2.81541884e-01 -1.52014703e-01 2.11379435e-02 -4.41645384e-01 -9.55651999e-01 4.72304553e-01 6.02714241e-01 -5.41715384e-01 -4.62776214e-01 6.54508948e-01 1.32588997e-01 3.92536610e-01 4.46004003e-01 -1.21384048e+00 -2.09803790e-01 1.59631535e-01 -4.85394627e-01 6.45208776e-01 4.43099618e-01 1.21725472e-02 9.19585705e-01 -8.00339162e-01 8.23661983e-01 1.08299172e+00 5.27361095e-01 5.53978443e-01 -1.04678428e+00 -3.38296115e-01 5.04158318e-01 7.77659357e-01 -1.14683366e+00 -6.27353072e-01 6.19845450e-01 -5.31799138e-01 1.27974951e+00 -1.92366585e-01 7.97858715e-01 1.17267156e+00 1.61553457e-01 1.06425416e+00 8.75581682e-01 -7.18657196e-01 2.15904921e-01 -1.51706532e-01 1.97087154e-02 1.00943577e+00 -3.71601611e-01 2.11370409e-01 -1.08822727e+00 2.88058847e-01 8.88440967e-01 2.89486319e-01 -2.25880519e-01 -5.15425026e-01 -1.28605294e+00 6.93872094e-01 4.27087069e-01 1.96883246e-01 -3.44505757e-01 4.42977756e-01 5.23882926e-01 -3.30568501e-03 7.31263995e-01 -2.99538523e-02 -3.77693355e-01 -3.26873004e-01 -1.15131915e+00 7.88612962e-02 4.37831759e-01 7.75983989e-01 8.01027596e-01 -1.99853465e-01 -2.93358177e-01 4.62418646e-01 3.46473873e-01 1.03384331e-02 5.97728968e-01 -1.22347915e+00 3.30721825e-01 5.42607367e-01 2.05317333e-01 -9.14991975e-01 -5.46342731e-01 9.71215069e-02 -2.60405183e-01 4.48090285e-01 5.58766842e-01 -1.55298654e-02 -9.67994809e-01 1.77444410e+00 2.76843190e-01 7.04109728e-01 -6.64914995e-02 8.32297146e-01 6.09830678e-01 5.63241243e-01 2.89984584e-01 -3.29612464e-01 1.16657078e+00 -1.28638542e+00 -7.80739009e-01 -1.63091674e-01 7.44259059e-01 -4.07238662e-01 9.71357226e-01 3.84539872e-01 -1.18447614e+00 -7.45571911e-01 -8.79500806e-01 -2.40307137e-01 -3.61362994e-01 3.34810317e-01 5.45968473e-01 2.01706246e-01 -1.17707527e+00 6.41560435e-01 -1.21356702e+00 -7.54040420e-01 5.90702355e-01 3.66498709e-01 -4.10981447e-01 1.16363332e-01 -8.11818957e-01 8.12169790e-01 6.17530704e-01 -7.31718689e-02 -1.15589571e+00 -3.02217185e-01 -1.09589601e+00 1.35909423e-01 4.94720846e-01 -3.18913788e-01 1.38528299e+00 -1.48383093e+00 -1.42859924e+00 9.72227633e-01 -5.87253511e-01 -8.06522131e-01 2.42575735e-01 -5.59795022e-01 -2.14929268e-01 6.57796502e-01 1.97132230e-02 1.03901374e+00 9.18957114e-01 -9.86805797e-01 -9.98441160e-01 -1.87451616e-01 4.24081624e-01 2.94431180e-01 -4.13071811e-01 2.56059140e-01 -5.88037312e-01 -5.00811934e-01 -1.45369887e-01 -8.45723331e-01 -6.80966005e-02 6.74589723e-02 -1.65527184e-02 -5.38059711e-01 7.49654472e-01 -4.33715731e-01 1.07456887e+00 -2.26063395e+00 1.89732946e-02 -2.00451061e-01 2.19926480e-02 5.57690971e-02 1.59774534e-02 2.44405955e-01 -2.88962483e-01 -2.58602917e-01 1.05307169e-01 -6.46377444e-01 -2.24961266e-01 3.73453677e-01 -4.03729260e-01 5.14194250e-01 4.01215136e-01 8.62682462e-01 -1.06467104e+00 -8.39843750e-01 4.68435109e-01 4.29762483e-01 -6.26844823e-01 2.85346776e-01 -5.10442138e-01 5.00407100e-01 -9.55284685e-02 5.33366740e-01 -5.59478849e-02 -4.73077595e-01 6.76286519e-02 -2.50132471e-01 -2.50886053e-01 2.97793001e-01 -9.16936636e-01 1.93401706e+00 -1.37456566e-01 9.11384046e-01 -5.51545143e-01 -1.30624270e+00 4.81872767e-01 4.46865410e-01 6.47789538e-01 -5.02679884e-01 1.28614217e-01 -2.13057876e-01 -3.05854857e-01 -8.24840605e-01 3.40081006e-01 3.44129279e-02 1.10942133e-01 4.47724581e-01 6.11768246e-01 5.52038431e-01 6.19140148e-01 1.17568970e-01 1.05158746e+00 8.45124185e-01 2.76529878e-01 1.39842063e-01 5.41926265e-01 1.23540245e-01 6.16978943e-01 6.20431781e-01 -3.65268975e-01 5.72556794e-01 5.07728696e-01 -5.70803225e-01 -7.76740253e-01 -6.79999352e-01 1.86085418e-01 1.53982651e+00 1.52115852e-01 -6.25641882e-01 -8.32852781e-01 -8.49475443e-01 -4.88560140e-01 5.37399292e-01 -8.93078387e-01 -8.76063332e-02 -6.35613561e-01 -2.32949212e-01 2.32659638e-01 9.75529015e-01 4.84258831e-01 -1.41382670e+00 -1.00184286e+00 1.12649038e-01 -4.04626667e-01 -1.27963305e+00 -3.53775114e-01 2.32741311e-01 -9.09105897e-01 -1.31713176e+00 -4.13535744e-01 -8.07891667e-01 7.65933633e-01 1.61244273e-01 1.09665966e+00 4.54774983e-02 -1.99163258e-01 7.46807396e-01 -5.40757477e-01 -4.28344637e-01 5.23340814e-02 -2.96737820e-01 -1.66685749e-02 3.66379231e-01 6.77285373e-01 -3.42908204e-01 -6.34245038e-01 3.02134961e-01 -7.17801571e-01 1.31393358e-01 3.77090067e-01 7.98824131e-01 7.79786646e-01 -8.76109526e-02 3.41104627e-01 -6.53007090e-01 -1.53028280e-01 -2.90570050e-01 -4.51899499e-01 2.54193187e-01 -2.39112362e-01 -1.82887465e-01 3.25111181e-01 -4.00913745e-01 -1.26364315e+00 5.04568100e-01 2.32035547e-01 -7.63619184e-01 -4.75715995e-01 3.43366653e-01 1.10353082e-01 -5.43696061e-02 5.70993781e-01 1.77876398e-01 -1.11511767e-01 -2.75037140e-01 3.20180535e-01 3.51352185e-01 6.69297874e-01 -1.36915281e-01 2.66085476e-01 8.41644466e-01 -1.52152359e-01 -5.89285910e-01 -1.35323524e+00 -6.49216115e-01 -1.10146248e+00 -6.98287904e-01 1.19437420e+00 -1.11272025e+00 -8.17394912e-01 3.66365731e-01 -1.05000234e+00 -6.89523876e-01 -5.04513383e-01 7.30798185e-01 -1.12713981e+00 2.97455996e-01 -5.41487098e-01 -7.65255153e-01 1.82874337e-01 -8.15772355e-01 1.17216825e+00 2.64776200e-01 -2.14039698e-01 -1.06151175e+00 9.90055576e-02 4.60864455e-01 -1.62917510e-01 2.51214772e-01 5.06856024e-01 -5.82904220e-01 -7.49602199e-01 -8.08574185e-02 -1.02537982e-01 3.14522117e-01 -1.10027067e-01 -1.42782256e-01 -1.03386056e+00 -1.15197122e-01 -9.40067768e-02 -7.38246620e-01 1.03339660e+00 5.85830271e-01 1.43058002e+00 -9.17131919e-03 -4.90401298e-01 5.84666669e-01 1.18098569e+00 8.71857703e-02 8.18864048e-01 3.45322579e-01 5.34235120e-01 6.78153098e-01 9.85254109e-01 4.61033225e-01 6.37555361e-01 6.59217298e-01 5.78724682e-01 -1.42591760e-01 -1.94891825e-01 -3.38619709e-01 6.38743222e-01 2.20633924e-01 -5.10005474e-01 -2.59141594e-01 -6.87304616e-01 6.42794132e-01 -2.33678341e+00 -1.39740300e+00 3.36040482e-02 2.18795586e+00 6.53912902e-01 9.64066684e-02 3.62378240e-01 1.08608477e-01 6.07509673e-01 2.05689043e-01 -4.64598417e-01 -1.81218654e-01 1.66464344e-01 9.21303034e-02 3.42392415e-01 2.79093623e-01 -1.56588900e+00 1.10599136e+00 6.72557974e+00 3.35823715e-01 -1.10067952e+00 3.21184754e-01 5.09102166e-01 -4.66197610e-01 5.42242289e-01 4.20583272e-03 -7.69495010e-01 3.64979625e-01 1.12797213e+00 2.92741179e-01 1.24203451e-01 6.80816948e-01 4.83056068e-01 -4.60324943e-01 -1.38005781e+00 9.53357756e-01 5.72657764e-01 -1.41850340e+00 -3.76506805e-01 -1.31320879e-01 5.00524282e-01 1.50494009e-01 -2.47099116e-01 2.36468196e-01 9.47797075e-02 -8.72643173e-01 8.38207304e-01 7.93759823e-01 5.16409636e-01 -6.29603863e-01 5.11961877e-01 3.46059144e-01 -1.08683932e+00 -3.73022974e-01 -2.93054372e-01 -3.09748799e-01 2.97810793e-01 -4.87454459e-02 -6.60336971e-01 1.04467586e-01 1.09476030e+00 1.26572704e+00 -6.15680337e-01 1.07615626e+00 -3.34271222e-01 6.90799892e-01 -1.43574968e-01 2.84302235e-01 2.22399250e-01 2.51983255e-01 2.51800448e-01 1.09703493e+00 1.73167303e-01 1.66440755e-01 3.92930686e-01 4.41595942e-01 5.67546599e-02 7.44225979e-02 -4.64915186e-01 3.97345684e-02 -1.03008121e-01 1.00437784e+00 -8.38536322e-01 -4.91503686e-01 -7.22057402e-01 1.00324571e+00 6.14256978e-01 4.48142380e-01 -9.63072836e-01 -3.98422666e-02 2.86084890e-01 3.24783862e-01 7.59595931e-01 -2.24022716e-01 2.93522686e-01 -1.25719607e+00 -8.82581249e-02 -5.73643923e-01 6.22471690e-01 -1.09347713e+00 -9.30800974e-01 4.31072265e-01 1.26636207e-01 -1.45659769e+00 -5.05190909e-01 -6.37899876e-01 -4.80719060e-01 2.49777004e-01 -1.64838374e+00 -1.29004419e+00 -4.10664380e-01 9.79245424e-01 8.30613911e-01 -6.45309985e-02 8.05584311e-01 2.06390873e-01 -5.29267311e-01 2.60582447e-01 -2.31145725e-01 2.92467028e-01 9.59711134e-01 -1.24053371e+00 -8.38151872e-02 9.32016253e-01 5.68969309e-01 2.09328309e-01 5.23669600e-01 -5.06879151e-01 -8.96862924e-01 -1.13484359e+00 1.07503045e+00 -6.08291507e-01 6.08781517e-01 -3.43985200e-01 -6.74529016e-01 1.10979199e+00 3.96633923e-01 4.73824263e-01 8.26900780e-01 1.11635260e-01 -2.06480488e-01 -5.08794934e-02 -7.23847866e-01 2.83636808e-01 1.23133349e+00 -5.31751990e-01 -7.97436476e-01 6.23256385e-01 4.15742964e-01 -5.24180353e-01 -6.09121621e-01 2.95086205e-01 4.13278013e-01 -1.26970160e+00 8.06516886e-01 -8.65720868e-01 5.68977833e-01 -3.69179666e-01 6.97791427e-02 -7.48485923e-01 -1.75221726e-01 -4.52725530e-01 -5.30857921e-01 1.02960181e+00 1.24674633e-01 -6.79902583e-02 9.05049980e-01 3.69710326e-01 -1.51307538e-01 -6.85558617e-01 -7.69082248e-01 -4.42874640e-01 -4.60120380e-01 -5.25955319e-01 -1.24814689e-01 6.03753090e-01 2.50583947e-01 1.93872675e-01 -5.71020901e-01 1.54261842e-01 4.73319978e-01 1.07614011e-01 6.61378324e-01 -1.03811979e+00 -1.96372345e-01 2.45995224e-02 -7.96667516e-01 -1.17977154e+00 5.95351458e-01 -5.83648205e-01 2.97944456e-01 -1.32523441e+00 2.35833272e-01 1.31369665e-01 -4.79526341e-01 8.36174190e-01 6.82555065e-02 5.42355359e-01 1.58441797e-01 3.42455119e-01 -1.31469882e+00 3.62950385e-01 7.35247314e-01 1.23951927e-01 -1.57187521e-01 -5.46612889e-02 -1.46558121e-01 1.17208672e+00 5.17677426e-01 -3.91625255e-01 -5.12665570e-01 -3.98998052e-01 -3.11375745e-02 3.77893113e-02 7.50051856e-01 -1.25929284e+00 5.14511049e-01 -6.65241107e-02 5.42860866e-01 -7.01365769e-01 4.59576488e-01 -7.75936425e-01 -4.99986410e-01 1.49525583e-01 -5.91856539e-01 -1.01133294e-01 1.02882050e-02 8.95374477e-01 -4.40619856e-01 -1.92659512e-01 6.31576955e-01 -2.28738964e-01 -1.27818370e+00 2.14707568e-01 -5.01240373e-01 -8.29599351e-02 1.34177899e+00 -4.19939846e-01 -1.33005366e-01 -5.43348491e-01 -1.21114457e+00 2.59929359e-01 5.03483117e-01 4.59229439e-01 6.21640444e-01 -1.12817419e+00 -3.54366541e-01 1.48569658e-01 2.63800740e-01 -1.52808994e-01 3.44836265e-01 1.06624115e+00 -3.19596529e-01 3.46276820e-01 -2.56495029e-01 -9.41514969e-01 -1.34861970e+00 7.22568214e-01 3.05781871e-01 -1.32904977e-01 -6.92803562e-01 7.99159110e-01 3.11547637e-01 2.05064878e-01 5.59357941e-01 -3.15000325e-01 -5.38403392e-01 1.68399051e-01 7.92991638e-01 9.02203470e-02 -1.37702793e-01 -9.47739065e-01 -2.64516652e-01 5.27485073e-01 4.98905964e-03 -5.45746163e-02 1.36055410e+00 -3.90208215e-01 4.92721982e-02 6.89409018e-01 1.00327361e+00 -4.32864219e-01 -1.83294368e+00 -3.33505362e-01 1.62082016e-01 -4.89915520e-01 1.68840826e-01 -6.17452621e-01 -9.22180176e-01 1.00768805e+00 7.58677483e-01 1.66472811e-02 1.27702391e+00 2.07318172e-01 5.59441984e-01 3.53326321e-01 2.69343227e-01 -1.23929584e+00 5.87753177e-01 4.79679316e-01 5.74496329e-01 -1.48129761e+00 -9.52308401e-02 -2.71997750e-01 -6.91661835e-01 1.15902781e+00 8.39616179e-01 -3.07295412e-01 5.28223813e-01 3.92781273e-02 9.46556125e-03 -2.70133764e-01 -1.04576075e+00 -6.44574940e-01 4.40051317e-01 5.95818639e-01 4.99261200e-01 -5.45895576e-01 8.35629553e-03 2.97210753e-01 3.26259255e-01 4.73105520e-01 4.46889013e-01 1.09930003e+00 -4.97754842e-01 -8.94424021e-01 -1.66102424e-01 3.10217351e-01 -5.29809713e-01 7.35686719e-02 -6.05475754e-02 7.19653785e-01 4.41107363e-01 9.31677938e-01 5.05276084e-01 -1.75917953e-01 1.34727240e-01 3.85391921e-01 7.12655067e-01 -8.61849546e-01 -3.16144288e-01 2.18518496e-01 7.83561319e-02 -1.08303916e+00 -1.29612160e+00 -9.11169112e-01 -1.39490044e+00 3.20141762e-01 -3.28959733e-01 -1.83889434e-01 1.71063274e-01 1.23552787e+00 3.23758096e-01 4.84281480e-01 3.73570412e-01 -1.18494046e+00 8.98868675e-05 -7.90606320e-01 -3.97385269e-01 5.12106836e-01 3.65750402e-01 -1.03507781e+00 -1.28954515e-01 8.18690062e-01]
[8.479859352111816, 0.5899574756622314]
0d7c255b-80b6-4b1f-9b14-88aa2ba720f3
sood-towards-semi-supervised-oriented-object
2304.04515
null
https://arxiv.org/abs/2304.04515v1
https://arxiv.org/pdf/2304.04515v1.pdf
SOOD: Towards Semi-Supervised Oriented Object Detection
Semi-Supervised Object Detection (SSOD), aiming to explore unlabeled data for boosting object detectors, has become an active task in recent years. However, existing SSOD approaches mainly focus on horizontal objects, leaving multi-oriented objects that are common in aerial images unexplored. This paper proposes a novel Semi-supervised Oriented Object Detection model, termed SOOD, built upon the mainstream pseudo-labeling framework. Towards oriented objects in aerial scenes, we design two loss functions to provide better supervision. Focusing on the orientations of objects, the first loss regularizes the consistency between each pseudo-label-prediction pair (includes a prediction and its corresponding pseudo label) with adaptive weights based on their orientation gap. Focusing on the layout of an image, the second loss regularizes the similarity and explicitly builds the many-to-many relation between the sets of pseudo-labels and predictions. Such a global consistency constraint can further boost semi-supervised learning. Our experiments show that when trained with the two proposed losses, SOOD surpasses the state-of-the-art SSOD methods under various settings on the DOTA-v1.5 benchmark. The code will be available at https://github.com/HamPerdredes/SOOD.
['Xiang Bai', 'Xiaoqing Ye', 'Zhikang Zou', 'Xiaolong Liu', 'Jingyu Li', 'Dingkang Liang', 'Wei Hua']
2023-04-10
null
http://openaccess.thecvf.com//content/CVPR2023/html/Hua_SOOD_Towards_Semi-Supervised_Oriented_Object_Detection_CVPR_2023_paper.html
http://openaccess.thecvf.com//content/CVPR2023/papers/Hua_SOOD_Towards_Semi-Supervised_Oriented_Object_Detection_CVPR_2023_paper.pdf
cvpr-2023-1
['semi-supervised-object-detection', 'pseudo-label']
['computer-vision', 'miscellaneous']
[ 9.80517343e-02 6.60539139e-03 -4.44777429e-01 -5.64847648e-01 -5.61410725e-01 -4.19895887e-01 4.92998004e-01 2.26462781e-01 -2.22568259e-01 3.47693413e-01 -5.34703434e-02 -1.02754153e-01 -1.30089328e-01 -5.83717406e-01 -7.65349984e-01 -8.91169667e-01 -4.89419959e-02 2.04499811e-01 6.39847100e-01 -3.30672637e-02 -3.40337791e-02 1.92825466e-01 -1.50876224e+00 3.22530776e-01 7.01062262e-01 1.38003314e+00 3.58450204e-01 2.45296225e-01 9.10416842e-02 7.53685415e-01 -1.29740581e-01 -3.69482279e-01 5.36475062e-01 -1.72915801e-01 -6.55886114e-01 4.39413995e-01 8.41550708e-01 -1.57215878e-01 -9.19154063e-02 1.29974365e+00 4.23866183e-01 -2.56802201e-01 5.49732029e-01 -1.38581860e+00 -6.63161218e-01 6.26429558e-01 -8.07101548e-01 3.58514577e-01 -2.88321733e-01 9.89085138e-02 1.31551707e+00 -1.00550318e+00 3.94831926e-01 1.13227296e+00 7.44861543e-01 3.78650934e-01 -1.28157055e+00 -5.53061306e-01 5.97623229e-01 2.87332118e-01 -1.43478739e+00 6.36018217e-02 8.99227738e-01 -6.10895038e-01 2.83989877e-01 2.11825088e-01 4.87623453e-01 8.92827809e-01 -1.14571594e-01 1.07931745e+00 1.16612065e+00 -4.51137334e-01 2.61672616e-01 3.14306319e-01 6.36649311e-01 8.41485620e-01 3.38736475e-01 1.74126253e-01 -3.30054194e-01 1.00891121e-01 2.71601647e-01 2.11741194e-01 -2.80233443e-01 -9.98759091e-01 -9.34367895e-01 6.62813008e-01 1.02548862e+00 -6.46671206e-02 -2.19277978e-01 -9.78476554e-02 2.87163287e-01 -2.18587786e-01 6.65015042e-01 9.53470320e-02 -4.30148542e-01 5.15784740e-01 -5.92157304e-01 2.89346844e-01 2.29534939e-01 9.98450756e-01 7.60494351e-01 -2.58736044e-01 -4.30391848e-01 8.61258328e-01 5.51479697e-01 4.28371459e-01 2.63103843e-01 -4.22197819e-01 6.04789674e-01 9.38580751e-01 1.82137668e-01 -8.89089525e-01 -5.27486205e-01 -8.64273190e-01 -7.27235079e-01 3.02166760e-01 4.81007665e-01 1.00146560e-03 -9.13740158e-01 1.59116042e+00 8.89864564e-01 -5.17372899e-02 -3.04239362e-01 1.25735998e+00 8.11046422e-01 5.23972809e-01 2.35110983e-01 8.49406570e-02 1.52225494e+00 -1.46228552e+00 -3.23398530e-01 -6.82004452e-01 6.13771558e-01 -7.03330576e-01 1.10753417e+00 1.74190924e-01 -7.21545279e-01 -7.53611863e-01 -1.20409775e+00 9.77699235e-02 -2.19964802e-01 6.38470829e-01 4.13564831e-01 5.56026518e-01 -5.70756733e-01 4.21934605e-01 -8.62674713e-01 -2.71834791e-01 8.20096016e-01 1.64547905e-01 3.01441755e-02 -2.33598784e-01 -8.23891342e-01 5.81537843e-01 4.06569779e-01 2.05266938e-01 -9.40914035e-01 -5.48248529e-01 -6.70957446e-01 -7.81715363e-02 5.99889696e-01 -4.73272651e-01 1.13833106e+00 -1.00835824e+00 -9.67772722e-01 1.16119456e+00 1.00289740e-01 -5.28881192e-01 6.13417685e-01 -2.88849831e-01 -1.90479651e-01 -5.60785346e-02 3.31975162e-01 8.58418345e-01 9.89873409e-01 -1.32825398e+00 -1.00450432e+00 -7.59471953e-01 2.66069412e-01 3.80854219e-01 -4.52401489e-01 -3.24702039e-02 -3.62354100e-01 -7.55318820e-01 3.18305105e-01 -1.03185666e+00 -1.80092424e-01 6.01453602e-01 -6.48902833e-01 -4.28914696e-01 8.90711188e-01 -3.22765052e-01 1.07017851e+00 -2.25693464e+00 1.16247728e-01 -2.63352752e-01 1.16759069e-01 3.38061541e-01 -1.94005407e-02 -2.11789086e-02 -9.94797051e-02 -3.70944500e-01 -3.46376866e-01 -4.29251999e-01 -3.24944779e-02 -2.26384923e-01 -2.63446838e-01 6.06599510e-01 3.00511867e-01 7.00445175e-01 -8.96575868e-01 -4.97597963e-01 1.78101301e-01 2.96722949e-01 -3.14289957e-01 2.43323281e-01 -2.34481514e-01 2.14827120e-01 -4.96923774e-01 7.90134728e-01 1.02394569e+00 -5.41031420e-01 -8.36359151e-03 -2.56547809e-01 -4.47972231e-02 1.69095933e-01 -1.18642175e+00 1.36718309e+00 -2.65005916e-01 2.73863316e-01 -8.24928656e-03 -1.07875645e+00 9.85887527e-01 -1.50274754e-01 1.45343974e-01 -4.07361925e-01 6.87167048e-02 9.47278813e-02 -1.53310046e-01 -2.59741634e-01 1.81545258e-01 -7.03920797e-02 2.82789499e-01 6.59247637e-02 4.12103422e-02 8.31045285e-02 3.38224769e-01 1.78753898e-01 5.96553206e-01 4.03931826e-01 2.96070963e-01 -4.48476344e-01 4.81570095e-01 1.66216031e-01 6.66175604e-01 5.96892059e-01 -3.55082363e-01 5.99037528e-01 3.84732634e-01 -5.66766620e-01 -7.69743145e-01 -9.38928485e-01 -5.88501036e-01 1.21771979e+00 5.98370552e-01 -2.19731241e-01 -7.10052669e-01 -1.24908853e+00 1.31931379e-01 4.69885051e-01 -7.77148664e-01 -9.73951351e-03 -2.38312691e-01 -1.02543485e+00 -5.23527414e-02 6.63467407e-01 5.75471997e-01 -8.89051139e-01 -5.47587931e-01 -1.41323870e-03 -4.57248539e-02 -1.07406700e+00 -5.56068599e-01 5.91178417e-01 -8.83260787e-01 -1.09818661e+00 -8.16185832e-01 -9.89748240e-01 8.32558751e-01 7.64023304e-01 9.94319201e-01 -8.33114758e-02 -3.81303787e-01 -4.99051325e-02 -5.25491655e-01 -5.67299962e-01 4.25937288e-02 7.93373585e-02 -1.61838531e-02 3.11000675e-01 3.95804197e-01 -1.47189274e-01 -7.36187100e-01 7.15929389e-01 -6.48672640e-01 2.49040142e-01 5.23556828e-01 8.40082407e-01 7.53066540e-01 -1.12363495e-01 2.64445692e-01 -8.31352830e-01 -2.83931226e-01 -4.28258330e-01 -9.83315885e-01 3.35576922e-01 -7.05291867e-01 2.17504203e-02 5.94546735e-01 -4.61573511e-01 -9.00138497e-01 3.27994198e-01 3.34577560e-02 -3.48257005e-01 -1.92085192e-01 1.28683701e-01 -2.94759244e-01 -2.51946338e-02 6.95853651e-01 -1.91542413e-02 -4.33935136e-01 -5.55833340e-01 1.37560889e-01 6.65392578e-01 2.14057639e-01 -2.78630883e-01 7.73022830e-01 7.52578020e-01 2.02039573e-02 -5.62406659e-01 -1.60106814e+00 -8.40155482e-01 -6.32274091e-01 -1.34661511e-01 7.33215392e-01 -1.10229909e+00 -2.20910162e-01 5.58380902e-01 -8.38727415e-01 -4.02978122e-01 -3.81494284e-01 4.66880679e-01 -1.95559859e-01 3.47780347e-01 -3.68643641e-01 -7.92198598e-01 -2.68083572e-01 -1.12346315e+00 1.24621427e+00 2.70435065e-01 2.88290650e-01 -6.25044286e-01 -1.06395938e-01 4.71618474e-01 1.81523766e-02 -1.04054742e-01 6.64684296e-01 -5.41278481e-01 -6.71328962e-01 -2.77828753e-01 -6.06542706e-01 6.12516940e-01 -4.10864223e-03 -2.90690750e-01 -9.99253452e-01 -3.24873269e-01 -1.04167908e-01 -6.42356932e-01 1.09700739e+00 4.56504852e-01 1.20984733e+00 -2.31031463e-01 -4.27264512e-01 5.99453568e-01 1.24504113e+00 -1.61075860e-01 1.03028238e-01 4.51530069e-01 7.85678983e-01 7.59636045e-01 1.00020003e+00 4.62250710e-01 4.26405936e-01 9.79233623e-01 8.41660678e-01 -2.39470199e-01 -2.92792916e-01 -3.73273611e-01 1.19330429e-01 2.97887295e-01 1.47503227e-01 -1.94278419e-01 -8.00807238e-01 4.94134724e-01 -1.91240633e+00 -5.54854810e-01 -3.99147868e-01 2.20185637e+00 6.37472987e-01 4.40056294e-01 2.20079169e-01 1.29972562e-01 9.49571431e-01 3.28684181e-01 -6.38593853e-01 2.78515399e-01 -1.23365737e-01 -2.67024666e-01 5.78249097e-01 1.98357671e-01 -1.72769892e+00 7.12677658e-01 4.58050299e+00 9.34167504e-01 -9.82275069e-01 2.06326112e-01 7.82029092e-01 1.66309252e-01 2.90502936e-01 -1.32386256e-02 -1.41570497e+00 6.02012157e-01 1.56377628e-01 2.58831650e-01 1.17348328e-01 1.50237632e+00 7.44995335e-03 -3.33224870e-02 -9.79805708e-01 7.00070143e-01 5.56294844e-02 -9.55744028e-01 -2.89992765e-02 -1.75869659e-01 8.33414853e-01 1.09725311e-01 1.35015056e-01 2.06194267e-01 6.04938418e-02 -4.66207474e-01 1.16242576e+00 -2.45240163e-02 4.15030450e-01 -3.76020849e-01 7.71628559e-01 4.65730131e-01 -1.47396207e+00 -3.63320827e-01 -4.63030547e-01 -3.63833942e-02 4.05996405e-02 7.00985491e-01 -4.99771297e-01 4.18878824e-01 9.84148920e-01 8.98925304e-01 -9.30049419e-01 1.18690157e+00 -5.12144923e-01 8.05877626e-01 -2.78388411e-01 -1.12517394e-01 3.80082250e-01 -1.37362435e-01 4.97356385e-01 1.06182957e+00 -5.45441732e-02 -1.12493262e-01 4.69350189e-01 6.89203501e-01 -4.09995168e-02 1.27494007e-01 -2.58530140e-01 3.38931918e-01 2.15719521e-01 1.53593504e+00 -8.84648442e-01 -3.19157332e-01 -3.57336432e-01 7.27296710e-01 4.62116957e-01 6.33557364e-02 -9.28529620e-01 -1.77753821e-01 2.93335408e-01 3.66813332e-01 5.51862538e-01 2.31879145e-01 -3.47768158e-01 -1.00411105e+00 3.05603743e-01 -6.39890432e-01 5.76697886e-01 -7.11152792e-01 -1.38415718e+00 6.25492096e-01 5.92830889e-02 -1.50784385e+00 4.93971050e-01 -8.00961554e-01 -5.05064189e-01 4.30503160e-01 -1.59876108e+00 -1.45182729e+00 -6.11400127e-01 1.07065387e-01 7.71997213e-01 1.41815230e-01 3.67464840e-01 3.21648389e-01 -7.84746230e-01 3.89391541e-01 -4.56187688e-03 7.29273632e-02 5.64257503e-01 -1.33141232e+00 2.53440291e-01 8.97355318e-01 3.35702151e-01 2.56697595e-01 5.81385732e-01 -3.33130211e-01 -8.55760872e-01 -1.41916573e+00 7.58283019e-01 -2.87015200e-01 5.20221710e-01 -4.88443583e-01 -9.32270944e-01 5.51850200e-01 -8.13331380e-02 6.22327149e-01 2.44530931e-01 4.70845476e-02 -4.12194937e-01 -4.15096968e-01 -7.83416867e-01 3.36202890e-01 1.16465592e+00 -9.27891731e-02 -2.95866489e-01 8.15065145e-01 6.68505013e-01 -3.30015182e-01 -2.25758284e-01 7.82014370e-01 3.63257796e-01 -1.01835752e+00 1.04702032e+00 -5.29721975e-01 5.25926530e-01 -6.46508694e-01 -9.30954292e-02 -1.01764643e+00 -5.29667497e-01 -1.12111829e-01 -8.77452567e-02 1.18403578e+00 3.45918447e-01 -3.63903522e-01 8.72344732e-01 1.37888968e-01 -1.43471092e-01 -1.10844445e+00 -7.03028679e-01 -1.01484632e+00 -2.77340204e-01 -1.83491722e-01 2.65145093e-01 7.09564686e-01 -2.68224388e-01 5.62631667e-01 -3.53660554e-01 5.18127322e-01 9.62273896e-01 4.32501823e-01 6.15335703e-01 -1.13037467e+00 -4.26960021e-01 -3.28974992e-01 -4.36361879e-01 -1.41374874e+00 -1.60879865e-01 -9.57063377e-01 2.20782310e-01 -1.33645058e+00 4.50843304e-01 -7.61402190e-01 -3.50382745e-01 6.27447724e-01 -5.61405420e-01 5.97091615e-01 1.79186255e-01 1.82263151e-01 -9.20781255e-01 7.78580129e-01 9.39755678e-01 -3.42496097e-01 -5.33815697e-02 2.97324985e-01 -5.09362459e-01 9.44617391e-01 8.22202921e-01 -7.31145382e-01 -1.86900288e-01 -3.48583460e-01 -1.45402849e-01 -4.08653229e-01 6.54181838e-01 -1.06501627e+00 6.14264794e-02 1.39743113e-03 2.37956598e-01 -7.46718824e-01 1.43836468e-01 -7.83158004e-01 -3.29271674e-01 6.15214765e-01 -4.55199987e-01 -2.78029084e-01 2.20994391e-02 8.32867920e-01 -1.45038500e-01 -3.59019488e-01 1.17399931e+00 1.70302093e-01 -6.89761579e-01 3.67268533e-01 2.24367782e-01 1.98646307e-01 1.20323873e+00 4.45408709e-02 -2.54339755e-01 2.08320752e-01 -4.96125579e-01 4.15377349e-01 2.81417102e-01 4.61045444e-01 3.94032031e-01 -1.23087192e+00 -6.54806852e-01 2.18238868e-02 5.84381461e-01 2.55377024e-01 2.51907706e-01 8.82347584e-01 -2.48874664e-01 4.53588486e-01 -4.15642001e-02 -8.96315753e-01 -1.41531241e+00 8.51190805e-01 2.86721319e-01 -4.34302956e-01 -5.88781118e-01 1.20672071e+00 7.90188909e-01 -4.87168789e-01 5.63580573e-01 -2.48988405e-01 -1.68413922e-01 2.68877037e-02 4.27030653e-01 3.48370314e-01 1.03131868e-01 -5.28478265e-01 -5.28219044e-01 5.89538097e-01 -2.16545746e-01 3.37499827e-01 1.17007899e+00 -5.45154214e-02 2.23746002e-01 4.01063621e-01 1.01663411e+00 -3.60338032e-01 -1.68536520e+00 -5.59818804e-01 -1.89402569e-02 -5.50536990e-01 2.41671830e-01 -7.17500210e-01 -1.04073751e+00 9.17541564e-01 9.61064875e-01 6.42505288e-02 9.73510087e-01 1.97411105e-01 4.48428690e-01 1.81396097e-01 4.03952330e-01 -7.93363690e-01 3.48407626e-01 1.55160815e-01 7.03098953e-01 -1.64945054e+00 1.40582696e-01 -7.44196653e-01 -8.55989099e-01 7.59447873e-01 8.51711929e-01 -4.52515110e-02 6.42991245e-01 -5.27524464e-02 6.72661290e-02 -9.46888328e-02 -4.20320868e-01 -4.18387502e-01 6.27644360e-01 3.35602522e-01 2.62714922e-01 1.38879657e-01 -2.74510801e-01 5.76937854e-01 3.53745818e-01 -5.07253781e-02 -5.70440479e-02 9.20713603e-01 -4.68567431e-01 -9.02776718e-01 -4.31956798e-01 4.24927413e-01 -2.91341156e-01 -9.02546570e-02 -2.34361321e-01 4.87123013e-01 3.88424158e-01 7.03450501e-01 -7.85949454e-02 -1.97808787e-01 3.94274890e-01 -2.91136861e-01 2.64897555e-01 -7.14767396e-01 -2.96380430e-01 -5.08704083e-03 -2.62205184e-01 -3.39945167e-01 -5.37065208e-01 -6.53252900e-01 -9.10996675e-01 2.91209608e-01 -8.63436460e-01 6.24696054e-02 5.62199354e-01 5.82341075e-01 2.20451668e-01 3.81648123e-01 7.80455351e-01 -8.39623094e-01 -9.62098897e-01 -9.57003951e-01 -7.23383784e-01 3.94861549e-01 2.85154760e-01 -9.79724050e-01 -4.20358330e-01 7.12023005e-02]
[9.21463680267334, 1.13712477684021]
72dff537-3221-43a0-8a34-d377656de9be
learning-rich-representation-of-keyphrases-1
2112.08547
null
https://arxiv.org/abs/2112.08547v2
https://arxiv.org/pdf/2112.08547v2.pdf
Learning Rich Representation of Keyphrases from Text
In this work, we explore how to train task-specific language models aimed towards learning rich representation of keyphrases from text documents. We experiment with different masking strategies for pre-training transformer language models (LMs) in discriminative as well as generative settings. In the discriminative setting, we introduce a new pre-training objective - Keyphrase Boundary Infilling with Replacement (KBIR), showing large gains in performance (upto 8.16 points in F1) over SOTA, when the LM pre-trained using KBIR is fine-tuned for the task of keyphrase extraction. In the generative setting, we introduce a new pre-training setup for BART - KeyBART, that reproduces the keyphrases related to the input text in the CatSeq format, instead of the denoised original input. This also led to gains in performance (upto 4.33 points in F1@M) over SOTA for keyphrase generation. Additionally, we also fine-tune the pre-trained language models on named entity recognition (NER), question answering (QA), relation extraction (RE), abstractive summarization and achieve comparable performance with that of the SOTA, showing that learning rich representation of keyphrases is indeed beneficial for many other fundamental NLP tasks.
['Rajarshi Bhowmik', 'Ravneet Arora', 'Debanjan Mahata', 'Mayank Kulkarni']
2021-12-16
null
https://aclanthology.org/2022.findings-naacl.67
https://aclanthology.org/2022.findings-naacl.67.pdf
findings-naacl-2022-7
['keyphrase-generation', 'keyphrase-extraction']
['natural-language-processing', 'natural-language-processing']
[ 4.04317468e-01 5.16906202e-01 1.16250187e-01 1.77619427e-01 -1.62944973e+00 -7.52443910e-01 1.05863047e+00 5.34850836e-01 -7.50116646e-01 1.00506461e+00 7.52139568e-01 -3.53871852e-01 -1.89744443e-01 -8.06270719e-01 -9.46832538e-01 -5.28652847e-01 1.74774043e-02 5.24652123e-01 1.66694000e-01 -5.72932184e-01 2.07741573e-01 3.51087987e-01 -1.13971674e+00 6.62200511e-01 8.65032852e-01 6.83137596e-01 1.73654124e-01 1.17396486e+00 -5.44089019e-01 1.01286602e+00 -1.09373260e+00 -5.57365417e-01 1.51390452e-02 -4.30201501e-01 -1.17942095e+00 -1.53997809e-01 6.85263455e-01 -2.47614197e-02 -4.07203257e-01 6.62095785e-01 6.29077435e-01 1.37692124e-01 7.68087268e-01 -7.78074503e-01 -5.28364480e-01 1.19275784e+00 -3.38967860e-01 4.91276264e-01 5.15710592e-01 -2.59394675e-01 1.32646132e+00 -9.28314447e-01 6.74101412e-01 1.24311411e+00 4.79511619e-01 3.32045078e-01 -1.25554860e+00 -4.17620867e-01 -1.17591098e-01 2.00577065e-01 -1.24763000e+00 -4.76702362e-01 3.38435143e-01 -4.25002389e-02 1.37498927e+00 5.07383466e-01 2.30533704e-01 1.21340346e+00 2.47435063e-01 1.06541216e+00 9.29105163e-01 -7.75324702e-01 -7.53413290e-02 2.45871231e-01 1.58034906e-01 3.14007312e-01 1.65216953e-01 -3.46235335e-01 -5.56838393e-01 -2.72332042e-01 4.05212671e-01 -5.43560207e-01 -2.66421169e-01 2.80817866e-01 -1.43353975e+00 8.09578240e-01 4.56380397e-02 6.63470268e-01 -6.06113195e-01 1.41677484e-01 6.22740626e-01 6.04939282e-01 3.83231968e-01 9.19799030e-01 -8.59903157e-01 -7.93687478e-02 -1.26909542e+00 5.07914901e-01 9.74017143e-01 1.01452732e+00 6.44989014e-01 4.56756987e-02 -1.07614446e+00 8.54524493e-01 -2.70900428e-01 5.32682776e-01 7.35704303e-01 -6.08999550e-01 9.21052158e-01 5.40059209e-01 -2.24569663e-02 -5.10227561e-01 -2.76949823e-01 -6.74327672e-01 -1.07421446e+00 -6.46795511e-01 1.30822048e-01 -3.67471069e-01 -1.23424733e+00 1.46319175e+00 6.52004853e-02 -1.37671515e-01 5.12274504e-01 6.37226850e-02 1.08038926e+00 1.18267143e+00 -5.88371307e-02 -2.84748346e-01 1.73227048e+00 -9.77392912e-01 -8.63022685e-01 -2.54019409e-01 7.89729834e-01 -1.05745602e+00 1.09899187e+00 4.35551405e-01 -1.09740305e+00 -5.70181847e-01 -7.89734423e-01 -3.40119958e-01 -6.01370633e-01 5.60712397e-01 1.46106228e-01 2.66454786e-01 -1.11732483e+00 4.85356122e-01 -3.30171198e-01 -4.97492492e-01 1.39781818e-01 9.17035863e-02 -4.25561070e-01 1.37346551e-01 -1.37736952e+00 1.01003981e+00 8.23406696e-01 -2.54290998e-01 -7.86293447e-01 -9.92762923e-01 -8.57916474e-01 1.26292109e-01 7.34527588e-01 -8.82527173e-01 1.34430742e+00 -1.92103878e-01 -1.44336569e+00 8.73132885e-01 -1.16977334e-01 -9.29840028e-01 2.74605423e-01 -7.10397363e-01 -3.44751239e-01 1.26539335e-01 1.72995374e-01 6.72559738e-01 9.69424903e-01 -9.76444960e-01 -6.45573139e-01 8.62287655e-02 -2.85160225e-02 1.33397609e-01 -2.96365112e-01 2.48675972e-01 -2.66054600e-01 -1.16263497e+00 -3.76270294e-01 -5.98769307e-01 -1.28654346e-01 -1.00483990e+00 -8.72010946e-01 -4.38055992e-01 5.42590678e-01 -1.12030637e+00 1.68587863e+00 -1.63855076e+00 1.95372492e-01 8.37690160e-02 -1.13887705e-01 4.86203074e-01 -1.48447782e-01 9.81131136e-01 -1.56684950e-01 2.96628714e-01 -3.92175883e-01 -2.52299219e-01 4.61013578e-02 2.74901718e-01 -1.00540543e+00 -1.94736153e-01 5.47542691e-01 1.32144749e+00 -7.48529553e-01 -5.72055757e-01 -6.83343485e-02 1.88327402e-01 -2.92809576e-01 4.53411847e-01 -3.35581303e-01 1.50383234e-01 -3.56257051e-01 1.02045506e-01 3.29264522e-01 -6.36076629e-02 -7.29753673e-02 -3.28723013e-01 5.47804944e-02 8.00257266e-01 -1.19388616e+00 1.58475852e+00 -8.33946228e-01 5.72088122e-01 -1.91127315e-01 -8.54661524e-01 7.03759313e-01 5.60265422e-01 1.01981349e-01 -5.72994351e-01 7.15399673e-03 1.56368628e-01 -2.14161605e-01 -3.15273970e-01 9.57975805e-01 2.46374719e-02 -3.32412452e-01 4.92665410e-01 6.40306413e-01 -5.55821598e-01 5.22446275e-01 6.93092644e-01 1.26997650e+00 -1.52191356e-01 6.25555158e-01 -1.56091183e-01 8.52619827e-01 -6.64240122e-02 -1.52699575e-01 9.52280879e-01 6.73829257e-01 6.91527843e-01 5.99119127e-01 9.02891532e-02 -1.08466506e+00 -7.72162974e-01 2.67340392e-02 1.20834398e+00 -5.44009328e-01 -1.04210079e+00 -8.90417993e-01 -8.41257036e-01 -1.33479431e-01 1.17977536e+00 -5.49376249e-01 -3.74458015e-01 -1.06689489e+00 -6.90757871e-01 1.10436749e+00 3.96848410e-01 5.03662109e-01 -1.23625934e+00 -2.42165208e-01 4.00617599e-01 -3.33495796e-01 -1.38354659e+00 -5.58868051e-01 5.23895502e-01 -7.30105579e-01 -6.72611713e-01 -1.01972520e+00 -6.32974029e-01 3.41512650e-01 -2.15006709e-01 1.35986257e+00 -2.21462682e-01 -3.86228412e-02 4.84898418e-01 -5.08673966e-01 -2.77970046e-01 -8.23461235e-01 5.52628994e-01 -2.89411724e-01 -8.66696388e-02 -9.14453939e-02 -4.16888922e-01 -8.68727937e-02 -3.23170125e-01 -1.29573560e+00 1.92052554e-02 9.28397715e-01 8.15745413e-01 4.40636575e-01 7.55724013e-02 4.35226679e-01 -1.02247918e+00 1.05167878e+00 -3.20149034e-01 -1.97336957e-01 5.67215264e-01 -4.32545096e-01 7.83420563e-01 8.09997916e-01 -5.10033250e-01 -9.58462775e-01 -5.71205020e-01 -4.07817274e-01 -8.05864669e-03 -1.60064697e-01 4.24632847e-01 -1.84745386e-01 4.82363641e-01 8.29453349e-01 4.85532194e-01 -6.69477582e-01 -8.50414455e-01 8.66696358e-01 6.49243832e-01 7.69294202e-01 -6.81605339e-01 1.06129682e+00 6.03509955e-02 -1.88354760e-01 -9.01199520e-01 -1.05072141e+00 -5.61447680e-01 -4.64691907e-01 5.26908875e-01 7.37700403e-01 -8.81855428e-01 -2.13466167e-01 2.45163381e-01 -1.35708702e+00 -1.67502314e-01 -6.77486658e-01 1.76179558e-01 -3.43282610e-01 4.89606410e-01 -6.36859417e-01 -6.08339667e-01 -1.15552175e+00 -6.21287704e-01 1.30699098e+00 9.74285081e-02 -4.65798199e-01 -9.16491985e-01 1.62853166e-01 2.89669186e-01 3.62419993e-01 6.06465228e-02 1.26784265e+00 -1.17768955e+00 -2.46126965e-01 -1.92481965e-01 -1.41151935e-01 6.27750993e-01 1.35524645e-01 -2.60520905e-01 -9.54216897e-01 -2.05959082e-01 -1.03346698e-01 -3.58990997e-01 1.26345503e+00 8.46954137e-02 7.97141790e-01 -7.67636359e-01 -8.06133375e-02 2.84418076e-01 1.01053298e+00 -2.64365315e-01 8.01612318e-01 3.30699205e-01 6.42449498e-01 5.78066349e-01 4.69460130e-01 1.68587029e-01 4.18760657e-01 6.41627729e-01 -1.64042428e-01 -8.07930455e-02 -3.48522514e-01 -5.89088917e-01 6.18328273e-01 9.68383908e-01 1.70872718e-01 -4.12054449e-01 -7.28586137e-01 7.23988235e-01 -1.58948946e+00 -7.88540125e-01 -1.37354866e-01 1.93642032e+00 1.46212196e+00 2.52042592e-01 7.87357520e-03 2.55117178e-01 4.15935218e-01 2.64469445e-01 5.95678762e-02 -4.81272906e-01 -3.66554439e-01 9.17576313e-01 5.34532726e-01 5.33974111e-01 -1.12450266e+00 1.30628145e+00 5.57446623e+00 1.39212322e+00 -8.57906818e-01 1.49166375e-01 3.80826950e-01 2.94726819e-01 -2.84465939e-01 1.07219994e-01 -1.19739306e+00 1.10598445e-01 1.34956467e+00 -3.59050691e-01 1.84502468e-01 4.86565411e-01 -9.12642702e-02 -2.21895218e-01 -9.99013066e-01 9.10762548e-01 1.23773664e-01 -1.45792580e+00 4.34193075e-01 -1.81671083e-01 6.23714983e-01 -1.36541158e-01 -3.25852275e-01 7.07303762e-01 4.46327418e-01 -1.00557148e+00 7.21645176e-01 3.71093422e-01 7.20513105e-01 -6.51629567e-01 7.74676919e-01 5.21154821e-01 -1.02585351e+00 1.94081977e-01 -3.09168100e-01 3.24915528e-01 1.66663706e-01 7.10012317e-01 -1.30888367e+00 1.05083740e+00 4.82362807e-01 2.44687855e-01 -8.65081549e-01 6.69736624e-01 -4.96497691e-01 8.74853015e-01 -4.73163396e-01 1.39873356e-01 3.12022179e-01 1.73922807e-01 6.75554693e-01 1.79438841e+00 2.47307166e-01 -1.06491424e-01 -1.56852722e-01 6.16719961e-01 -4.53661263e-01 4.30996478e-01 -3.55560184e-01 -4.01950747e-01 2.98822910e-01 1.55303288e+00 -6.44209743e-01 -7.83570588e-01 1.48929387e-01 1.17787576e+00 1.89917311e-01 3.11473995e-01 -5.61094046e-01 -7.11475730e-01 2.10571155e-01 2.51034889e-02 6.79559231e-01 -1.02342144e-01 1.43575147e-01 -1.26461220e+00 1.27021790e-01 -1.21790802e+00 6.04703546e-01 -6.86443865e-01 -1.06463075e+00 8.23451102e-01 2.79333502e-01 -7.00040340e-01 -6.98999226e-01 -2.96957910e-01 -6.26342118e-01 8.99310410e-01 -1.51536036e+00 -1.28663027e+00 1.84397861e-01 3.23707432e-01 6.90177619e-01 -7.83408955e-02 9.35018957e-01 1.78653494e-01 -3.33945602e-01 7.01156557e-01 7.24971443e-02 2.26704136e-01 7.93845117e-01 -1.63625598e+00 8.33771348e-01 9.42309201e-01 7.55328238e-01 8.01330566e-01 7.37739027e-01 -5.99955440e-01 -1.34033406e+00 -1.24135518e+00 1.24532163e+00 -7.10314929e-01 8.05364847e-01 -5.83464265e-01 -1.08262610e+00 6.18330002e-01 5.99478483e-01 -3.86193037e-01 2.84790844e-01 -1.88202009e-01 -2.96397507e-01 -5.68784960e-02 -7.87474036e-01 5.03072917e-01 5.83008289e-01 -7.37854123e-01 -1.09332609e+00 3.78601253e-01 1.08660710e+00 -4.34826195e-01 -8.58369172e-01 3.87671053e-01 -3.37018184e-02 -4.14842159e-01 1.11227536e+00 -8.05570245e-01 2.40133032e-01 -1.86020777e-01 -1.13002904e-01 -1.36266232e+00 8.10863357e-03 -1.20479977e+00 -4.53673333e-01 1.77539480e+00 5.92746615e-01 -3.95660073e-01 2.80044377e-01 -1.64957613e-01 -5.59944399e-02 -7.36259222e-01 -7.45744586e-01 -6.14654958e-01 2.73065716e-01 -4.21044230e-01 4.41594154e-01 4.96889621e-01 -3.83810401e-01 1.14795899e+00 -3.35853219e-01 -3.85921858e-02 1.18483923e-01 -9.75729302e-02 9.32874739e-01 -8.18060517e-01 -3.87007385e-01 -2.11076975e-01 1.09909117e-01 -1.18756878e+00 -5.35591925e-03 -9.47946250e-01 -2.27741793e-01 -1.76882589e+00 -3.29299690e-03 -2.13550068e-02 -7.99180716e-02 5.83404303e-01 -4.80852991e-01 5.88499662e-03 2.65865952e-01 1.05349325e-01 -4.36258733e-01 5.32985389e-01 1.00481665e+00 -6.40103370e-02 -2.31901005e-01 1.10005163e-01 -8.90275657e-01 4.17847693e-01 5.67071438e-01 -6.45685315e-01 -1.84704140e-01 -9.68253538e-02 4.09558177e-01 -7.36160725e-02 3.04166555e-01 -7.47570097e-01 1.32514745e-01 1.59809858e-01 2.04405084e-01 -8.96445990e-01 1.50343254e-01 -3.05546552e-01 -1.38405606e-01 1.86210215e-01 -5.04489243e-01 1.60608500e-01 5.27783692e-01 2.36881271e-01 -2.69412488e-01 -4.27765727e-01 3.62203658e-01 -1.73614547e-01 -2.45320335e-01 -1.34019211e-01 -3.96384090e-01 6.90535784e-01 2.99184710e-01 1.38806015e-01 -2.31337905e-01 -4.67337400e-01 -4.16283488e-01 7.25171119e-02 -1.05458133e-01 3.31919432e-01 3.67163181e-01 -9.95634317e-01 -1.03210747e+00 2.53746621e-02 -4.15910855e-02 1.53877646e-01 5.33823948e-03 9.12517488e-01 -3.47325057e-01 9.81913328e-01 3.83158267e-01 -1.11296907e-01 -1.12922406e+00 3.47786665e-01 5.06901629e-02 -1.19014716e+00 -6.94880545e-01 8.86273026e-01 2.66826786e-02 -5.57607651e-01 2.20243275e-01 -9.88459408e-01 -3.90296519e-01 4.60988641e-01 5.25046468e-01 2.83554196e-01 7.02562749e-01 -2.95821190e-01 -1.45445123e-01 3.29048961e-01 -4.79965210e-01 -3.35205078e-01 1.27595687e+00 9.05341431e-02 -2.20136479e-01 3.60878766e-01 1.27158558e+00 5.70112348e-01 -4.75267708e-01 -5.34588933e-01 5.91212988e-01 3.77191693e-01 1.19371630e-01 -8.83799791e-01 -4.39330667e-01 6.99922144e-01 4.96635586e-02 2.26946369e-01 1.04107666e+00 2.54063994e-01 1.22251892e+00 7.91946888e-01 -1.46494890e-02 -1.04990947e+00 1.05646454e-01 8.51361513e-01 1.17977345e+00 -8.14993322e-01 1.14589788e-01 -1.25044391e-01 -7.77262807e-01 1.13708699e+00 5.25865890e-02 -1.09993950e-01 2.93829858e-01 2.69158840e-01 -3.47055010e-02 -2.20393881e-01 -8.82751584e-01 -4.04127777e-01 7.03347206e-01 2.17093125e-01 3.70545834e-01 -1.63905606e-01 -1.91195041e-01 7.16282487e-01 -8.04371953e-01 -3.93153459e-01 4.58919019e-01 8.34105074e-01 -2.91483551e-01 -1.25508237e+00 -4.56056118e-01 5.58399916e-01 -7.86388993e-01 -6.93242133e-01 -6.47623181e-01 8.28461349e-01 -1.35137811e-01 8.19321215e-01 -1.92699328e-01 -2.64163017e-01 4.80783939e-01 4.81747895e-01 4.62506682e-01 -9.41843867e-01 -1.06613517e+00 1.32777527e-01 4.31406707e-01 -1.34295225e-01 -5.96421510e-02 -2.35904425e-01 -1.18657303e+00 6.79378957e-02 -5.12099028e-01 5.66770434e-01 3.97583842e-01 1.12900531e+00 3.48032862e-01 6.99439824e-01 3.50570679e-01 -4.25043970e-01 -6.79642975e-01 -1.51420546e+00 -1.19335651e-01 2.40556926e-01 3.88363272e-01 -8.60311091e-02 -3.14959079e-01 3.07938695e-01]
[12.299161911010742, 9.030123710632324]
458802f8-31bf-407e-a4e8-dcfeca0ec2a6
learning-in-imperfect-environment-multi-label
2304.10539
null
https://arxiv.org/abs/2304.10539v1
https://arxiv.org/pdf/2304.10539v1.pdf
Learning in Imperfect Environment: Multi-Label Classification with Long-Tailed Distribution and Partial Labels
Conventional multi-label classification (MLC) methods assume that all samples are fully labeled and identically distributed. Unfortunately, this assumption is unrealistic in large-scale MLC data that has long-tailed (LT) distribution and partial labels (PL). To address the problem, we introduce a novel task, Partial labeling and Long-Tailed Multi-Label Classification (PLT-MLC), to jointly consider the above two imperfect learning environments. Not surprisingly, we find that most LT-MLC and PL-MLC approaches fail to solve the PLT-MLC, resulting in significant performance degradation on the two proposed PLT-MLC benchmarks. Therefore, we propose an end-to-end learning framework: \textbf{CO}rrection $\rightarrow$ \textbf{M}odificat\textbf{I}on $\rightarrow$ balan\textbf{C}e, abbreviated as \textbf{\method{}}. Our bootstrapping philosophy is to simultaneously correct the missing labels (Correction) with convinced prediction confidence over a class-aware threshold and to learn from these recall labels during training. We next propose a novel multi-focal modifier loss that simultaneously addresses head-tail imbalance and positive-negative imbalance to adaptively modify the attention to different samples (Modification) under the LT class distribution. In addition, we develop a balanced training strategy by distilling the model's learning effect from head and tail samples, and thus design a balanced classifier (Balance) conditioned on the head and tail learning effect to maintain stable performance for all samples. Our experimental study shows that the proposed \method{} significantly outperforms general MLC, LT-MLC and PL-MLC methods in terms of effectiveness and robustness on our newly created PLT-MLC datasets.
['Yueting Zhuang', 'Siliang Tang', 'Beng Chin Ooi', 'Lingze Zeng', 'Changshuo Liu', 'Wenqiao Zhang']
2023-04-20
null
null
null
null
['philosophy']
['miscellaneous']
[ 3.41859370e-01 -2.27576613e-01 -4.85226482e-01 -8.32802474e-01 -1.27283084e+00 -5.43143690e-01 1.75440386e-01 3.08817536e-01 -4.65663821e-01 1.01566470e+00 -3.59055251e-01 -4.32393312e-01 -2.43099853e-01 -4.15392488e-01 -7.05980301e-01 -8.36859226e-01 3.68928343e-01 7.20966280e-01 1.69483960e-01 1.45331711e-01 1.70027971e-01 2.12035790e-01 -1.60885680e+00 5.92527092e-01 9.06287491e-01 1.12445772e+00 2.01106574e-02 5.99000394e-01 -5.25243431e-02 8.69150817e-01 -6.74355984e-01 -6.63207769e-01 1.95097819e-01 -3.82272214e-01 -6.01512253e-01 -1.10086463e-01 6.02115571e-01 -3.99399661e-02 4.78838533e-01 1.00492275e+00 7.38961935e-01 -1.86553240e-01 1.11458778e+00 -1.72036004e+00 -1.51527077e-01 6.48500919e-01 -1.36937535e+00 -2.35326700e-02 -1.38565168e-01 -1.90301668e-02 1.15562034e+00 -9.69353914e-01 2.05403894e-01 1.26488292e+00 1.01454318e+00 5.46585798e-01 -1.12656581e+00 -1.28096235e+00 3.72482508e-01 -6.65369332e-02 -1.44253111e+00 -1.89165711e-01 5.78886628e-01 -3.42309535e-01 5.11692464e-01 4.12681341e-01 -1.43931121e-01 9.06660140e-01 2.36099929e-01 1.03981268e+00 1.53683376e+00 -6.26652360e-01 1.49935305e-01 2.61491865e-01 4.98934746e-01 6.25797272e-01 2.37293288e-01 -8.95149931e-02 -3.20981681e-01 -3.63403976e-01 4.97439466e-02 5.44251241e-02 6.62843511e-02 -9.40733030e-02 -8.20958316e-01 8.72653186e-01 -1.00998431e-01 -6.58640778e-03 -1.69184860e-02 3.62970471e-01 5.32644272e-01 2.45316207e-01 6.19701624e-01 -1.69243127e-01 -9.44602549e-01 2.74497241e-01 -9.66472149e-01 3.77939016e-01 4.97453213e-01 1.13957238e+00 7.62925446e-01 -2.76543349e-01 -5.20561278e-01 1.23124516e+00 2.68940926e-01 7.67419279e-01 5.75068831e-01 -8.13639462e-01 6.45215929e-01 4.53439087e-01 1.77530516e-02 -7.08934247e-01 -6.65531576e-01 -8.41798484e-01 -9.75719988e-01 2.30624110e-01 3.55642587e-01 -2.32601434e-01 -7.88835585e-01 2.04918528e+00 4.54629391e-01 1.90783411e-01 -1.60686061e-01 4.45060223e-01 5.09801865e-01 3.18068475e-01 4.93274093e-01 -3.98600668e-01 1.33522964e+00 -9.79177833e-01 -6.18196070e-01 -3.52592200e-01 1.02039278e+00 -7.21492529e-01 1.28465748e+00 4.41679865e-01 -7.07642257e-01 -4.70135182e-01 -9.95032072e-01 2.14619398e-01 -2.74930090e-01 3.59990895e-01 3.35162610e-01 9.49173152e-01 -6.06916666e-01 3.63149077e-01 -3.46768558e-01 1.66138977e-01 5.27665257e-01 3.65140706e-01 -1.54905885e-01 -3.58769625e-01 -1.04163134e+00 5.60944498e-01 5.21854579e-01 -1.05714671e-01 -7.56777167e-01 -6.88344181e-01 -6.04601562e-01 -1.59593523e-01 5.30271351e-01 -3.39710921e-01 1.28915012e+00 -9.32979465e-01 -1.06427491e+00 1.05993772e+00 -1.07876122e-01 -1.48839727e-01 7.19861150e-01 3.55820335e-03 -3.83078843e-01 -2.98151255e-01 4.52841789e-01 7.08084464e-01 7.66672194e-01 -1.66776919e+00 -1.19227242e+00 -4.56376821e-01 -5.34348011e-01 9.96680781e-02 -2.00586259e-01 3.75702754e-02 -2.36749202e-01 -8.96583438e-01 1.64572015e-01 -8.93408179e-01 1.17784157e-01 -3.31281036e-01 -4.72839117e-01 -4.84618098e-01 8.28602850e-01 -2.89691269e-01 1.48041964e+00 -2.10558820e+00 -4.41820949e-01 2.42169172e-01 8.77251551e-02 2.40798131e-01 -2.33065993e-01 -2.85239611e-03 -2.92740166e-01 1.79735467e-01 -2.48558149e-01 -7.72162735e-01 -3.44050229e-02 3.70659024e-01 -7.76827559e-02 5.39849222e-01 -1.10478088e-01 6.25517309e-01 -6.88400209e-01 -7.29593277e-01 -1.28875464e-01 -1.65293038e-01 -6.05895221e-01 2.02133097e-02 -3.69336754e-01 2.23937213e-01 -3.37072521e-01 8.86153340e-01 9.44582105e-01 -3.84332776e-01 1.69415727e-01 2.36532744e-02 2.95063645e-01 -3.85396391e-01 -1.34413683e+00 9.64559436e-01 -3.75868082e-01 -2.38269940e-01 1.08489275e-01 -1.12493932e+00 9.22003746e-01 2.44457394e-01 4.44646925e-01 -5.58032155e-01 2.62033641e-01 5.80606878e-01 -3.25028449e-01 -1.14335611e-01 1.39560774e-01 -6.29094064e-01 -3.80579352e-01 6.77300632e-01 -2.31576711e-02 5.38589172e-02 4.26416062e-02 9.93755367e-03 8.28799903e-01 -2.96657719e-02 3.16162080e-01 -2.59931147e-01 6.01802647e-01 -3.33659858e-01 1.00293112e+00 8.91520858e-01 -6.18257225e-01 5.95268786e-01 5.06642461e-01 -7.85084814e-02 -6.89421415e-01 -7.09444165e-01 -1.76067173e-01 1.71536410e+00 7.38525158e-03 -8.77803862e-02 -6.16033554e-01 -1.25316036e+00 2.54865140e-01 1.00931644e+00 -5.55655658e-01 -1.52579591e-01 -4.56050336e-01 -1.37473106e+00 8.11838210e-01 3.42330128e-01 4.59417641e-01 -9.59150851e-01 -2.36676410e-01 5.24714850e-02 -3.73368561e-01 -8.54341149e-01 -5.95908463e-01 7.45460391e-01 -5.74833632e-01 -1.10935235e+00 -5.70945501e-01 -7.26330757e-01 5.52304626e-01 3.71740647e-02 1.15501511e+00 2.27700099e-02 -2.77684126e-02 -1.33961001e-02 -4.21373785e-01 -5.45946062e-01 -4.29962456e-01 1.47837400e-01 -5.40774837e-02 1.48279250e-01 4.00409579e-01 -3.83026451e-01 -4.03783202e-01 7.01229334e-01 -9.03446436e-01 -7.15631526e-03 6.09596550e-01 9.32530105e-01 8.71980965e-01 2.77357101e-01 1.26664662e+00 -1.39492285e+00 3.69260043e-01 -4.78333652e-01 -2.79971510e-01 6.83447421e-01 -1.15302706e+00 -1.32322401e-01 5.94056070e-01 -5.28939307e-01 -9.37380672e-01 -5.72963320e-02 -3.25129926e-01 -5.24124026e-01 -1.44645080e-01 2.90313870e-01 -3.20566684e-01 2.88371623e-01 4.74071890e-01 -2.57761888e-02 -1.73544884e-01 -5.10069847e-01 2.51888007e-01 1.03813708e+00 3.46823484e-01 -7.96870112e-01 3.90632331e-01 2.11148486e-01 -1.24749849e-02 1.55689036e-02 -1.44816661e+00 -5.71715891e-01 -5.21192491e-01 -1.92071974e-01 6.05793118e-01 -9.93090510e-01 -8.33987236e-01 8.26228738e-01 -6.94665134e-01 -5.30646145e-01 -1.12131514e-01 2.60672659e-01 -4.61872160e-01 2.77900040e-01 -6.19885325e-01 -9.48225915e-01 -4.44601715e-01 -1.12327909e+00 1.33884633e+00 1.13293372e-01 -2.81111784e-02 -8.66222799e-01 -1.67349368e-01 6.58509672e-01 8.98578689e-02 2.34458432e-01 1.31549239e+00 -1.03653061e+00 3.74451187e-03 -4.13770169e-01 -4.28125530e-01 6.31398261e-01 -5.13193868e-02 -3.34610999e-01 -1.02988696e+00 -6.45533919e-01 -7.97599256e-02 -7.83509254e-01 8.86307061e-01 2.50016958e-01 1.41047788e+00 -9.43082273e-02 -4.39972937e-01 3.26197982e-01 1.59254813e+00 8.22200328e-02 2.59144694e-01 1.27148643e-01 5.35307884e-01 5.42524576e-01 8.44086885e-01 6.02124333e-01 5.60293376e-01 4.82339472e-01 3.83540392e-01 3.22390832e-02 -1.47424012e-01 -2.22043961e-01 1.85909420e-01 6.18189812e-01 4.69535083e-01 -5.21666527e-01 -8.88110280e-01 2.41624057e-01 -1.70273733e+00 -5.90594947e-01 -2.44778693e-01 2.23803473e+00 1.29728985e+00 2.91474938e-01 -7.53769577e-02 3.18041354e-01 9.29706395e-01 -9.59125757e-02 -6.24173284e-01 -1.26318648e-01 -3.52234125e-01 2.13137925e-01 6.91174328e-01 4.04490590e-01 -1.30295038e+00 6.17201209e-01 5.36284781e+00 1.57652307e+00 -8.93082023e-01 4.41034675e-01 1.23128879e+00 -1.14616621e-02 -2.88253695e-01 -1.50947690e-01 -1.40466702e+00 5.98003447e-01 8.73605967e-01 2.83845603e-01 -4.16688435e-02 8.01630855e-01 -7.70760477e-02 -3.55452031e-01 -1.02533066e+00 9.11180198e-01 2.43918136e-01 -6.94843829e-01 -8.20011944e-02 -6.58703446e-02 8.64372313e-01 -2.38918990e-01 1.50335804e-01 1.01796651e+00 5.76821029e-01 -8.26032579e-01 9.83771145e-01 1.24036908e-01 1.26422310e+00 -7.21444666e-01 8.89391899e-01 8.14381838e-01 -1.05384707e+00 -3.28936756e-01 -1.67575851e-01 2.08939895e-01 -5.35144545e-02 9.37468588e-01 -4.18492109e-01 7.30848610e-01 4.54162121e-01 3.77858102e-01 -8.52287054e-01 6.38908029e-01 -9.02897418e-02 8.11854005e-01 -1.65315166e-01 1.10930920e-01 6.84743747e-02 2.43763953e-01 2.32893024e-02 1.12524760e+00 1.68000281e-01 -2.38186847e-02 6.02231801e-01 3.14250320e-01 -2.66836494e-01 2.81842858e-01 6.07556291e-02 5.60571909e-01 5.45280933e-01 1.15640450e+00 -9.17110980e-01 -3.97369474e-01 -4.32672858e-01 7.37814844e-01 5.84566593e-01 1.77092522e-01 -1.07557237e+00 -2.79525310e-01 -1.08037062e-01 -8.54331255e-02 1.72748744e-01 5.35328150e-01 -5.72568715e-01 -9.45277274e-01 -4.91349883e-02 -9.99755144e-01 8.81973386e-01 -4.25898314e-01 -1.74916303e+00 2.32540593e-01 9.68865380e-02 -1.04701281e+00 2.37885360e-02 -5.47812581e-01 -1.85697913e-01 9.49354172e-01 -1.69120634e+00 -1.39091766e+00 -1.26832798e-01 6.23175859e-01 4.52288121e-01 -1.08496286e-01 6.77711248e-01 6.88907623e-01 -8.15056324e-01 1.22345507e+00 3.56383175e-01 -1.55337498e-01 1.21254456e+00 -1.27952528e+00 -4.03958589e-01 4.74529415e-01 -3.07748020e-01 1.32592544e-01 4.36056018e-01 -7.35979557e-01 -5.50794423e-01 -1.51308787e+00 9.04838383e-01 -3.76660317e-01 1.69074163e-01 -3.58229816e-01 -7.97841966e-01 7.75403738e-01 -3.30571383e-01 3.37213308e-01 9.63305473e-01 1.39097497e-01 -6.07456803e-01 -3.57549638e-01 -1.49469209e+00 2.11263433e-01 6.48199201e-01 -2.57798016e-01 -1.94091916e-01 4.76020336e-01 6.46611750e-01 -1.85104206e-01 -6.96024239e-01 8.81557643e-01 5.57133079e-01 -9.21080410e-01 6.98419929e-01 -3.44371825e-01 2.52299070e-01 -1.19281292e-01 -4.56595868e-01 -9.80873764e-01 -7.94776082e-02 -1.03406228e-01 1.78899780e-01 1.54838896e+00 6.16992295e-01 -6.73826635e-01 8.50157857e-01 4.03923184e-01 -1.11348227e-01 -9.47775304e-01 -9.51355815e-01 -6.72572851e-01 5.93464613e-01 -5.40242493e-01 4.68777359e-01 1.10363090e+00 -4.85253751e-01 2.64852375e-01 -5.38856447e-01 5.41742034e-02 7.87891448e-01 2.12846011e-01 5.70052147e-01 -1.36873245e+00 -3.61460030e-01 -3.11022818e-01 3.42958212e-01 -7.97329187e-01 3.52977097e-01 -1.08441782e+00 2.47556373e-01 -9.62354183e-01 6.70431852e-01 -1.13106596e+00 -6.33274078e-01 7.27121770e-01 -6.25903428e-01 3.55470777e-01 6.76035434e-02 3.59548330e-01 -8.70935142e-01 3.99750978e-01 1.05056059e+00 3.80695052e-02 1.42417386e-01 3.80588293e-01 -7.74123669e-01 6.27614498e-01 6.91084802e-01 -8.57693493e-01 -3.28167349e-01 2.32217144e-02 1.68883160e-01 1.66729897e-01 1.41119733e-01 -8.28103244e-01 1.16941862e-01 -2.23726586e-01 3.43757808e-01 -9.60747004e-01 -1.14634685e-01 -7.22671568e-01 -4.65391129e-02 4.08491731e-01 -6.15880311e-01 -9.29579958e-02 3.38497804e-03 7.05054939e-01 -2.58354358e-02 -4.40448821e-01 1.20391810e+00 -6.24781996e-02 -2.19898403e-01 1.79818496e-01 -4.87057306e-03 3.50006908e-01 1.23483980e+00 1.35673389e-01 -4.04117435e-01 -2.90614925e-02 -5.34420192e-01 6.43086076e-01 6.47801086e-02 7.93739110e-02 1.85003966e-01 -1.38976884e+00 -8.00058365e-01 2.91643262e-01 3.42397720e-01 1.01075940e-01 4.11141187e-01 8.62599790e-01 -1.69710368e-01 3.45375001e-01 2.46099412e-01 -5.35262108e-01 -1.34896255e+00 6.17114663e-01 3.98782969e-01 -8.33848119e-01 6.62998948e-03 1.04737043e+00 2.36817226e-01 -1.08296847e+00 4.99321222e-01 -1.11676678e-01 -3.29661393e-03 2.22056746e-01 2.62868822e-01 4.88306314e-01 2.05582231e-01 -4.89625394e-01 -2.69891471e-01 4.46877360e-01 -2.50945151e-01 8.71248990e-02 8.92131507e-01 -3.30260694e-01 -1.50305182e-01 5.56251168e-01 1.32475674e+00 -9.62582827e-02 -1.05193472e+00 -3.44484687e-01 2.18874961e-01 -2.80566454e-01 -1.83347940e-01 -1.31742644e+00 -9.62244153e-01 7.02853858e-01 7.71889150e-01 -2.55430311e-01 1.15797877e+00 -1.71740562e-01 7.13618517e-01 7.14385286e-02 5.17841816e-01 -1.16092491e+00 3.14022392e-01 3.68476629e-01 4.62993324e-01 -1.38582873e+00 2.60247607e-02 -4.11601007e-01 -6.80799901e-01 5.69517314e-01 8.62445772e-01 3.48288000e-01 8.70474041e-01 3.23697031e-01 9.08155516e-02 -3.93000580e-02 -8.19549620e-01 1.37341410e-01 -9.98827219e-02 2.56818652e-01 3.66710842e-01 2.04433590e-01 -4.46289539e-01 8.54658306e-01 1.92660525e-01 -5.48072997e-03 1.63023725e-01 1.01184177e+00 -4.81159329e-01 -1.33397293e+00 -5.07480025e-01 7.25915432e-01 -8.03874195e-01 3.31376642e-02 1.16809040e-01 6.56013906e-01 6.85843885e-01 1.01901901e+00 -2.09282592e-01 -2.70216078e-01 2.68102527e-01 4.25502449e-01 1.35790318e-01 -5.67572236e-01 -6.83175445e-01 3.70968997e-01 -1.36834964e-01 -8.50913823e-02 -4.59364057e-01 -7.79040873e-01 -1.26553667e+00 -1.29169643e-01 -6.92832530e-01 1.34759545e-01 3.71590823e-01 8.18431318e-01 2.12041456e-02 4.90399361e-01 8.90998363e-01 -3.19571704e-01 -1.06242907e+00 -1.20626056e+00 -1.02124107e+00 5.09880602e-01 1.69626012e-01 -8.62622261e-01 -6.20865703e-01 -1.80454552e-01]
[9.44477367401123, 4.152299404144287]
0e2ed912-87a2-4ed8-95e0-0a913f5eb732
a-simple-and-optimal-policy-design-for-online
2206.02969
null
https://arxiv.org/abs/2206.02969v5
https://arxiv.org/pdf/2206.02969v5.pdf
A Simple and Optimal Policy Design with Safety against Heavy-tailed Risk for Stochastic Bandits
We study the stochastic multi-armed bandit problem and design new policies that enjoy both worst-case optimality for expected regret and light-tailed risk for regret distribution. Starting from the two-armed bandit setting with time horizon $T$, we propose a simple policy and prove that the policy (i) enjoys the worst-case optimality for the expected regret at order $O(\sqrt{T\ln T})$ and (ii) has the worst-case tail probability of incurring a linear regret decay at an exponential rate $\exp(-\Omega(\sqrt{T}))$, a rate that we prove to be best achievable for all worst-case optimal policies. Briefly, our proposed policy achieves a delicate balance between doing more exploration at the beginning of the time horizon and doing more exploitation when approaching the end, compared to the standard Successive Elimination policy and Upper Confidence Bound policy. We then improve the policy design and analysis to work for the general $K$-armed bandit setting. Specifically, the worst-case probability of incurring a regret larger than any $x>0$ is upper bounded by $\exp(-\Omega(x/\sqrt{KT}))$. We then enhance the policy design to accommodate the "any-time" setting where $T$ is not known a priori, and prove equivalently desired policy performances as compared to the "fixed-time" setting with known $T$. A brief account of numerical experiments is conducted to illustrate the theoretical findings. We conclude by extending our proposed policy design to the general stochastic linear bandit setting and proving that the policy leads to both worst-case optimality in terms of expected regret order and light-tailed risk on the regret distribution.
['Feng Zhu', 'Zeyu Zheng', 'David Simchi-Levi']
2022-06-07
null
null
null
null
['thompson-sampling']
['methodology']
[-1.04583383e-01 2.80028999e-01 -5.99101603e-01 -2.95395792e-01 -1.34659386e+00 -1.05576313e+00 -2.17415065e-01 1.02866665e-01 -8.39337170e-01 1.04903650e+00 -1.16850957e-01 -1.17489302e+00 -1.07140934e+00 -8.63621116e-01 -9.94648695e-01 -9.03739154e-01 -3.79747510e-01 4.37761962e-01 -3.25267673e-01 1.32758766e-01 2.42798567e-01 3.22366863e-01 -9.78880942e-01 -3.88649583e-01 8.62579525e-01 1.78851318e+00 -8.12299252e-02 5.47181845e-01 4.83932346e-03 5.70442677e-01 -3.81750911e-01 -6.55203342e-01 8.35795283e-01 -3.66273731e-01 -7.37241805e-01 -8.27908888e-02 -8.40709955e-02 -4.80762035e-01 -1.52249679e-01 1.22471070e+00 3.00829500e-01 3.86071205e-01 2.74980426e-01 -8.53848159e-01 -1.91247210e-01 9.13485527e-01 -9.86926079e-01 3.93138498e-01 3.06429178e-03 -2.20853239e-01 1.09351814e+00 1.46843567e-01 4.89957184e-02 9.62575078e-01 4.23107564e-01 3.82231146e-01 -1.11157978e+00 -7.53453195e-01 6.18637741e-01 -3.09666038e-01 -1.07876885e+00 -1.38748512e-01 4.20334190e-01 -1.51474983e-01 5.87835968e-01 7.15734959e-01 5.48682570e-01 3.21715921e-01 -2.77239960e-02 9.87684667e-01 1.24151015e+00 -5.27698457e-01 6.52422190e-01 6.45186240e-03 2.51635671e-01 5.12610078e-01 3.34808737e-01 4.63765830e-01 9.59318504e-03 -2.79285312e-01 5.67168415e-01 1.30146503e-01 -3.18701893e-01 -1.45706967e-01 -6.54861033e-01 9.04312611e-01 2.44801328e-01 -6.70770109e-02 -6.66961491e-01 5.36334097e-01 3.04220051e-01 5.25468349e-01 6.15522504e-01 1.30709931e-01 -7.04739094e-01 -5.09211600e-01 -9.01111364e-01 2.82648116e-01 8.23397398e-01 1.19990742e+00 3.39221507e-01 -7.94317946e-02 -5.73767304e-01 5.79122603e-01 -1.29517257e-01 7.20140874e-01 -7.57526457e-02 -1.09797943e+00 1.05081213e+00 -2.10684333e-02 1.06871748e+00 -3.45154971e-01 -2.04623967e-01 -9.78655040e-01 -5.44377863e-01 1.52153313e-01 6.65471613e-01 -6.64843857e-01 -6.37737513e-01 2.04980612e+00 1.28276303e-01 -4.60362017e-01 -3.43413465e-02 7.26426363e-01 -3.86040896e-01 6.70336545e-01 -3.70523840e-01 -1.07736409e+00 1.18222296e+00 -6.12743318e-01 -5.25424540e-01 -2.62028664e-01 4.41405565e-01 -4.84051287e-01 9.08711195e-01 4.41257030e-01 -1.51961422e+00 2.54072756e-01 -6.97103918e-01 7.31975019e-01 2.41205424e-01 -3.70826095e-01 6.63722038e-01 1.32571089e+00 -5.01821399e-01 4.82232451e-01 -7.29580402e-01 2.04922184e-01 2.92907774e-01 3.92787546e-01 2.76716888e-01 -1.01655520e-01 -6.39803946e-01 4.94416386e-01 2.95943469e-01 2.21670777e-01 -7.23793387e-01 -6.89715862e-01 -3.39257687e-01 3.13356817e-01 9.94970977e-01 -5.66654444e-01 1.70218682e+00 -7.85390139e-01 -1.55590320e+00 2.62213856e-01 -5.78041077e-02 -5.88898480e-01 1.02354276e+00 -2.52811402e-01 2.48404667e-01 -4.59164195e-02 1.93214940e-03 -3.61099958e-01 4.25634652e-01 -9.14668977e-01 -1.01852489e+00 -5.66103339e-01 6.83354914e-01 2.51454145e-01 -3.03893089e-01 -6.74520656e-02 -1.89633802e-01 -6.35031641e-01 1.58378825e-01 -1.06177688e+00 -5.07168949e-01 -4.04287994e-01 -1.79018691e-01 1.50629878e-01 1.93478204e-02 -4.04692143e-01 1.36840951e+00 -1.95494938e+00 -2.85580277e-01 4.27579612e-01 -4.47850823e-01 4.43359464e-02 2.39729911e-01 3.81475270e-01 2.94996295e-02 3.78619522e-01 -4.93718050e-02 -1.81317776e-01 1.45443946e-01 1.32785887e-01 -3.69397163e-01 5.94392478e-01 -7.18147755e-01 4.81644481e-01 -8.03404987e-01 1.80756167e-01 -1.75750256e-03 -5.31367779e-01 -6.33656621e-01 1.42532915e-01 -3.93993467e-01 5.34800440e-02 -8.55738521e-01 4.46485907e-01 6.64245129e-01 -1.75997004e-01 1.35105744e-01 4.36240703e-01 -3.18445802e-01 -4.06571403e-02 -1.26201737e+00 1.15160811e+00 -7.29859591e-01 -8.88268575e-02 4.85119760e-01 -1.24882865e+00 4.47411925e-01 1.34843171e-01 5.39655149e-01 -6.55405760e-01 3.47075671e-01 2.04890579e-01 -2.52816528e-01 -2.41895452e-01 2.26426065e-01 -8.26004207e-01 -2.90440202e-01 8.70030522e-01 -5.42745233e-01 2.82903194e-01 -2.75085941e-02 -2.73675192e-02 1.20555246e+00 -3.32623273e-01 1.25804782e-01 -4.66123819e-01 9.02256891e-02 -3.43155235e-01 7.00101614e-01 1.30741477e+00 -2.22267255e-01 -1.33410782e-01 8.55533719e-01 -1.42405108e-01 -8.67920995e-01 -9.38131332e-01 4.16554287e-02 1.31172144e+00 2.99649000e-01 2.29823232e-01 -4.41303700e-01 -6.51363313e-01 4.14098561e-01 1.24107194e+00 -9.00404036e-01 6.93976209e-02 -1.22019991e-01 -8.61520648e-01 1.19120799e-01 3.29953760e-01 6.24760926e-01 -4.78389859e-01 -7.23089635e-01 2.98229426e-01 -1.44146338e-01 -8.15915942e-01 -7.01488137e-01 4.98978585e-01 -8.28060627e-01 -7.19804525e-01 -8.14705133e-01 3.95626277e-02 6.30113304e-01 2.54415542e-01 5.16073227e-01 -6.87984347e-01 1.84594274e-01 5.95443606e-01 -3.57908905e-01 -7.19623327e-01 1.02095507e-01 -2.11039454e-01 -1.22290723e-01 -6.21945374e-02 -2.25092947e-01 -3.10546398e-01 -1.05848682e+00 3.20600957e-01 -8.99692476e-01 -2.96066403e-01 4.66647476e-01 7.36947000e-01 5.39665461e-01 3.40473562e-01 6.79446101e-01 -5.22077799e-01 7.03727663e-01 -3.64663482e-01 -1.25898325e+00 4.23990637e-01 -7.65622973e-01 3.09551179e-01 7.02471137e-01 -2.98302591e-01 -1.19857073e+00 -4.23322916e-01 7.02403020e-03 -3.93410265e-01 4.78730083e-01 5.68286538e-01 2.07805470e-01 1.01304382e-01 3.37381512e-01 3.10076863e-01 -4.27010618e-02 -5.12001038e-01 3.26053351e-01 6.23798728e-01 2.10505053e-01 -1.12229884e+00 3.92085075e-01 4.65030849e-01 1.27353579e-01 -2.66976088e-01 -1.33868754e+00 -3.27663809e-01 2.99824774e-01 -4.35079075e-02 7.99712166e-02 -5.52733362e-01 -1.61304247e+00 -1.17481612e-01 -5.37118256e-01 -2.50775397e-01 -5.34171581e-01 7.23649740e-01 -1.25783944e+00 2.59136111e-01 -2.90781409e-01 -1.96880496e+00 -5.67564309e-01 -7.53641129e-01 3.90854567e-01 1.73827440e-01 4.75280046e-01 -6.88704789e-01 -2.16219217e-01 4.71708179e-01 3.59312326e-01 2.34113470e-01 8.68642271e-01 -3.15157264e-01 -4.17755961e-01 -3.58687103e-01 -3.12898815e-01 4.09071445e-01 -2.25192606e-02 -7.29257107e-01 -2.90873379e-01 -7.63372064e-01 2.45599136e-01 -1.35657758e-01 7.28879333e-01 8.26481164e-01 1.40008903e+00 -1.01653779e+00 -2.92450041e-01 5.08897901e-01 1.73484898e+00 7.15237796e-01 2.34727696e-01 6.34171307e-01 -4.03483391e-01 2.28338048e-01 9.94420707e-01 1.13867748e+00 6.24575326e-03 4.14530814e-01 8.28762352e-01 4.59113568e-01 9.54582632e-01 7.25650266e-02 1.86806142e-01 -5.20514399e-02 -2.26062775e-01 -3.91128540e-01 -6.02156699e-01 8.14223886e-01 -2.17349100e+00 -1.05463159e+00 4.80081797e-01 3.11973786e+00 7.60774732e-01 4.42842096e-01 5.46176910e-01 8.46006498e-02 6.52281046e-01 -1.76617965e-01 -8.89856756e-01 -7.87660599e-01 2.50424027e-01 2.89406776e-01 1.35481274e+00 5.24463117e-01 -7.42939532e-01 5.48499346e-01 5.86385012e+00 1.08048332e+00 -9.09412026e-01 1.72537550e-01 8.60848486e-01 -9.51532066e-01 -2.11456254e-01 1.20650701e-01 -6.79198205e-01 7.29380012e-01 9.27566707e-01 -6.68877423e-01 7.90065050e-01 1.07456148e+00 3.78448665e-01 -3.43971223e-01 -8.13370943e-01 7.91410506e-01 -6.96760476e-01 -1.25531244e+00 -5.25826752e-01 3.39377433e-01 5.98606050e-01 -1.62438944e-01 1.94885164e-01 2.41877690e-01 8.83763075e-01 -6.21791065e-01 9.41184759e-01 1.90248549e-01 8.72907102e-01 -1.35651839e+00 6.72271252e-01 7.99222171e-01 -8.09963167e-01 -8.87320995e-01 -2.07212970e-01 -2.95155525e-01 3.37568432e-01 7.25109577e-01 -5.29392064e-01 9.38143730e-01 7.10016370e-01 -3.14559758e-01 6.66537702e-01 1.30937696e+00 1.55305728e-01 4.72483903e-01 -8.02003384e-01 -3.76143783e-01 7.24265814e-01 -3.24769586e-01 3.90389919e-01 8.64007711e-01 6.30208850e-01 6.03565156e-01 2.59542406e-01 3.23245704e-01 1.42906249e-01 1.89284921e-01 -1.51500970e-01 5.57697415e-02 5.10955155e-01 5.70295632e-01 -5.47302425e-01 -2.24092066e-01 -2.21179411e-01 6.17974520e-01 3.09462816e-01 4.11629140e-01 -9.46954250e-01 -5.47696948e-01 8.18016112e-01 -8.75320211e-02 7.83014655e-01 5.28915897e-02 -3.81333619e-01 -7.45212018e-01 4.56762910e-01 -2.24450082e-01 8.03371370e-01 -2.15294510e-01 -1.20029211e+00 1.50441289e-01 1.44656941e-01 -8.98925543e-01 -2.08083346e-01 -3.63956690e-01 -1.85147047e-01 8.83716285e-01 -1.40542603e+00 -5.72562575e-01 5.50122201e-01 5.96808910e-01 2.09560171e-01 1.73786774e-01 6.12502992e-01 9.40988213e-02 -5.97652137e-01 9.58595634e-01 9.78658795e-01 -1.99225664e-01 -1.39328122e-01 -1.00985432e+00 -4.52618599e-01 6.35283172e-01 -5.84191799e-01 5.85252643e-01 9.15169120e-01 -3.00729215e-01 -1.52838457e+00 -9.39796805e-01 1.78210348e-01 2.76010692e-01 8.19360435e-01 2.12348420e-02 -1.51925664e-02 9.49457288e-01 -2.37896368e-01 -7.13255554e-02 6.89531028e-01 4.58187491e-01 -2.53596753e-01 -5.67903519e-01 -1.68192387e+00 4.54754889e-01 1.04520881e+00 -9.92394984e-02 -2.43865088e-01 3.70960057e-01 6.98071718e-01 -3.65521103e-01 -9.20928776e-01 5.15673280e-01 9.16622281e-01 -9.68592048e-01 5.78142107e-01 -7.91889727e-01 -1.26067564e-01 3.05098742e-01 -4.59605783e-01 -1.17610288e+00 -2.69431949e-01 -1.30087817e+00 -2.16575321e-02 7.52384245e-01 5.10387480e-01 -8.41739714e-01 1.07978809e+00 9.57016468e-01 6.87910095e-02 -9.04218733e-01 -1.53617227e+00 -1.35331893e+00 4.71266121e-01 -8.64020050e-01 5.21988690e-01 4.50977534e-01 2.17371389e-01 -3.06260109e-01 -5.16554236e-01 2.54957706e-01 7.48540878e-01 6.38201475e-01 4.36911076e-01 -4.74752307e-01 -7.71424234e-01 -5.41297138e-01 3.11714888e-01 -1.48603976e+00 -2.78371513e-01 -3.44715208e-01 1.15294866e-02 -1.24573088e+00 4.63736981e-01 -8.56279492e-01 -8.65009010e-01 4.20253932e-01 1.57774352e-02 -4.83406484e-01 4.41542476e-01 -2.57633746e-01 -6.90457642e-01 4.27777141e-01 1.20999014e+00 1.27060097e-02 -2.43315160e-01 7.26921678e-01 -1.13413537e+00 4.10066545e-01 6.50152028e-01 -4.89880294e-01 -4.82736111e-01 -2.51958251e-01 4.64781374e-01 1.23013067e+00 -1.56604499e-01 -5.14199376e-01 -1.70091733e-01 -7.71596372e-01 -2.29520276e-01 -6.84811711e-01 2.74631590e-01 -9.61593151e-01 1.60023257e-01 6.64910018e-01 -3.67295235e-01 -2.99403906e-01 1.67637035e-01 8.97660673e-01 3.87009770e-01 -4.39019144e-01 6.66660070e-01 -3.06836814e-01 3.15289855e-01 4.93762225e-01 -2.44928911e-01 -4.39006723e-02 1.37236893e+00 8.14958755e-03 -3.53162825e-01 -8.69117737e-01 -9.52325284e-01 6.37408495e-01 1.54427318e-02 -1.30861893e-01 4.43693687e-04 -9.11376178e-01 -4.39513326e-01 -2.79866934e-01 -1.99864998e-01 -1.19446009e-01 4.06212062e-01 7.11885870e-01 4.47883317e-03 6.96980774e-01 2.86162168e-01 -2.06888348e-01 -9.48921978e-01 9.80894625e-01 3.08954537e-01 -6.97894633e-01 -1.57203063e-01 9.55726624e-01 -2.48366743e-02 1.57760680e-01 5.17994046e-01 -3.48234951e-01 7.00309575e-01 -1.88563541e-01 4.22318310e-01 5.04444599e-01 1.43814562e-02 2.46259034e-01 -8.03659111e-02 2.45485246e-01 -2.37915292e-01 -5.28849900e-01 1.40219700e+00 -3.57906371e-01 1.34727940e-01 2.68635545e-02 1.00929642e+00 1.51000144e-02 -1.33158875e+00 -3.93400848e-01 -9.50764045e-02 -8.41092527e-01 7.26517588e-02 -1.09830105e+00 -1.04171669e+00 3.82985890e-01 5.53630054e-01 7.93923795e-01 1.29924035e+00 -1.08465865e-01 6.14019752e-01 4.40178931e-01 8.87787700e-01 -1.12933755e+00 -3.07745874e-01 4.24136788e-01 6.28941119e-01 -8.40426207e-01 -4.91115823e-02 -9.59274396e-02 -4.25805181e-01 6.75919533e-01 5.32812141e-02 -1.12575330e-01 7.48903036e-01 2.28224978e-01 -4.08583164e-01 2.54798591e-01 -6.46338344e-01 -3.27314198e-01 -2.04907998e-01 -1.08535387e-01 1.26256034e-01 4.85686630e-01 -8.16716969e-01 9.01326835e-01 -3.28815788e-01 1.14778772e-01 2.89628953e-01 1.15896714e+00 -6.31653190e-01 -9.44371819e-01 -5.90947509e-01 7.48309135e-01 -1.13445723e+00 1.19599581e-01 2.42052644e-01 5.69556296e-01 -4.00175840e-01 1.11496723e+00 8.62356946e-02 9.55317393e-02 4.04481918e-01 -5.38532343e-03 6.90092981e-01 -5.09354360e-02 -3.26350927e-01 3.84956032e-01 3.21059763e-01 -4.54773486e-01 -1.30506769e-01 -5.72725832e-01 -9.16664541e-01 -7.32531190e-01 -5.13320267e-01 6.42385483e-01 6.68622136e-01 1.03126609e+00 1.82234004e-01 2.65015364e-01 1.07988882e+00 -2.27422908e-01 -1.25322545e+00 -7.49016166e-01 -9.08708155e-01 -1.14092473e-02 4.92142111e-01 -4.83883411e-01 -4.54440683e-01 -7.33846962e-01]
[4.552914619445801, 3.3278088569641113]
ab8e2377-228a-4076-8dff-d7051d9af571
scheduling-techniques-for-liver-segmentation
2202.06373
null
https://arxiv.org/abs/2202.06373v1
https://arxiv.org/pdf/2202.06373v1.pdf
Scheduling Techniques for Liver Segmentation: ReduceLRonPlateau Vs OneCycleLR
Machine learning and computer vision techniques have influenced many fields including the biomedical one. The aim of this paper is to investigate the important concept of schedulers in manipulating the learning rate (LR), for the liver segmentation task, throughout the training process, focusing on the newly devised OneCycleLR against the ReduceLRonPlateau. A dataset, published in 2018 and produced by the Medical Segmentation Decathlon Challenge organizers, called Task 8 Hepatic Vessel (MSDC-T8) has been used for testing and validation. The reported results that have the same number of maximum epochs (75), and are the average of 5-fold cross-validation, indicate that ReduceLRonPlateau converges faster while maintaining a similar or even better loss score on the validation set when compared to OneCycleLR. The epoch at which the peak LR occurs perhaps should be made early for the OneCycleLR such that the super-convergence feature can be observed. Moreover, the overall results outperform the state-of-the-art results from the researchers who published the liver masks for this dataset. To conclude, both schedulers are suitable for medical segmentation challenges, especially the MSDC-T8 dataset, and can be used confidently in rapidly converging the validation loss with a minimal number of epochs.
['Sarada Prasad Dakua', 'Faycal Bensaali', 'Ayman Al-Kababji']
2022-02-13
null
null
null
null
['liver-segmentation']
['medical']
[ 1.22999735e-02 2.52609730e-01 -2.12684095e-01 -2.00464830e-01 -6.62470639e-01 -3.23527515e-01 5.04909337e-01 3.94841135e-01 -6.90518975e-01 6.28289521e-01 -2.29568958e-01 -4.27821875e-01 -2.75612742e-01 -1.09622471e-01 -4.05755132e-01 -1.08909798e+00 -5.45633316e-01 3.65501195e-01 2.86596894e-01 2.30705068e-01 5.88277951e-02 4.34390694e-01 -9.75575686e-01 3.27244967e-01 9.12559032e-01 1.05638945e+00 7.31050447e-02 5.91097832e-01 2.51476139e-01 5.25594234e-01 -5.33779860e-01 -4.04270023e-01 4.63332772e-01 -5.20855546e-01 -8.70361805e-01 3.80563363e-02 2.87487954e-01 1.49544403e-01 -1.27095124e-02 8.55710268e-01 6.98932171e-01 -1.34010985e-01 4.81791496e-01 -7.83213258e-01 3.67387235e-02 7.41596878e-01 -6.03243589e-01 7.13231146e-01 -1.20619416e-01 2.68219709e-01 3.35970789e-01 -5.79363346e-01 4.95275199e-01 6.58927023e-01 9.11604583e-01 6.16880536e-01 -1.18922472e+00 -5.42278230e-01 -1.83615103e-01 4.80473489e-02 -1.43997538e+00 -8.03725496e-02 1.80532545e-01 -5.12755394e-01 5.42765915e-01 5.20674407e-01 5.99581122e-01 8.79976213e-01 6.92536473e-01 5.72430491e-01 1.46410966e+00 -5.74359417e-01 2.59224862e-01 5.02265871e-01 9.52150971e-02 6.93237782e-01 1.36030033e-01 2.97367364e-01 -2.04047337e-01 4.05124724e-02 4.06963289e-01 -5.10704041e-01 -4.58873868e-01 -2.03327790e-01 -1.17626095e+00 8.41805339e-01 4.28237945e-01 6.18799984e-01 -3.33276719e-01 -4.87780422e-02 8.27128589e-01 4.42074150e-01 5.90186298e-01 5.83431423e-01 -5.03939688e-01 4.49648723e-02 -1.18488181e+00 -1.62646741e-01 9.85075533e-01 7.56400049e-01 1.64013684e-01 -2.17984676e-01 -5.09381831e-01 5.27951062e-01 1.33245081e-01 4.52438854e-02 7.57272542e-01 -4.06457216e-01 3.75118926e-02 5.13270020e-01 -3.04826558e-01 -2.93604672e-01 -8.81011724e-01 -9.99951482e-01 -9.51992393e-01 3.15583527e-01 7.68174767e-01 -3.88489574e-01 -1.04719281e+00 1.31487834e+00 3.62430573e-01 3.39794129e-01 1.10703386e-01 1.15433502e+00 9.53281105e-01 2.78917253e-01 3.59351754e-01 -4.21987742e-01 1.73329711e+00 -9.24808919e-01 -3.55500519e-01 6.97744489e-02 9.73112047e-01 -9.09821868e-01 9.02172863e-01 6.96506500e-01 -7.90603697e-01 -4.73232031e-01 -1.15840626e+00 5.86234927e-01 -1.22113891e-01 5.44695318e-01 7.76162863e-01 1.05885506e+00 -1.19987082e+00 7.25199223e-01 -9.99934018e-01 -5.10262728e-01 3.90956700e-01 3.30176383e-01 -8.98452178e-02 1.11880146e-01 -1.03708255e+00 1.14464307e+00 4.60709572e-01 2.40847796e-01 -9.79639709e-01 -1.23743069e+00 -4.57962930e-01 -2.91756004e-01 1.86462209e-01 -3.81051809e-01 1.07713032e+00 -9.99561310e-01 -1.42340040e+00 1.33654308e+00 5.08284926e-01 -1.06085026e+00 1.13885665e+00 9.92167648e-03 -1.08710460e-01 2.19800830e-01 -2.15743676e-01 6.73503757e-01 5.82717597e-01 -9.21309710e-01 -6.11722708e-01 -1.86123222e-01 -1.57502249e-01 3.05642374e-02 8.18108916e-02 1.57429054e-01 -3.96458745e-01 -6.15153551e-01 -1.88467667e-01 -1.08702409e+00 -3.08685064e-01 -1.99067056e-01 -4.00314033e-01 -1.81000739e-01 5.21236241e-01 -6.52682185e-01 1.02755296e+00 -1.98892081e+00 -7.95550197e-02 1.41170993e-01 -2.37243976e-02 4.01917666e-01 2.09917113e-01 -2.31037691e-01 -4.14664447e-01 8.04522038e-02 -3.03263187e-01 -2.91173279e-01 -5.17357171e-01 2.43777875e-02 1.42100647e-01 9.51660931e-01 6.85463175e-02 6.73072636e-01 -8.10162604e-01 -5.48435390e-01 2.04498336e-01 3.62880677e-01 -1.86180994e-01 2.93621253e-02 2.23665208e-01 8.39878738e-01 -2.96280146e-01 3.58734459e-01 6.17400289e-01 -1.37380153e-01 6.46430925e-02 -3.39552820e-01 -2.82932073e-01 -2.70013452e-01 -7.20962107e-01 1.80338335e+00 -3.17428291e-01 7.53022790e-01 -6.36479333e-02 -9.51926351e-01 9.04417634e-01 5.35673618e-01 7.93161035e-01 -8.25644553e-01 2.04530463e-01 1.92274377e-01 3.63425940e-01 -5.38242161e-01 -1.71009749e-01 -2.12507010e-01 2.44931310e-01 1.15568139e-01 6.34431168e-02 -8.85027945e-02 5.20632863e-01 -7.87764415e-02 7.57939041e-01 1.57424230e-02 2.38125890e-01 -9.86753106e-01 9.02777851e-01 1.32721905e-02 3.87739033e-01 6.52913570e-01 -5.26511192e-01 6.23812616e-01 8.57398212e-01 -5.19820273e-01 -9.26518142e-01 -6.66271150e-01 -7.86159098e-01 7.38246500e-01 1.12579137e-01 -7.02666119e-02 -1.10255253e+00 -1.14094341e+00 -2.19660208e-01 7.32356727e-01 -8.60307395e-01 -7.12658912e-02 -6.17250562e-01 -1.34423137e+00 9.11804736e-01 1.91692606e-01 3.23489457e-01 -9.60961878e-01 -9.94810104e-01 1.75515592e-01 2.28291810e-01 -1.15734506e+00 -3.51590484e-01 2.55042493e-01 -9.96346951e-01 -1.34851706e+00 -1.06055701e+00 -7.78077006e-01 7.58512437e-01 -4.10814524e-01 1.11253774e+00 2.09115133e-01 -7.35518277e-01 3.08210284e-01 -4.55925763e-01 -5.94108224e-01 -6.01438761e-01 2.90603817e-01 -2.17493057e-01 -1.27543760e-02 1.46655843e-01 -1.19025216e-01 -9.01682854e-01 4.84050304e-01 -5.91112316e-01 1.14511631e-01 7.55160570e-01 8.36489379e-01 6.80710733e-01 -2.61630505e-01 6.43714786e-01 -1.01764858e+00 2.95752943e-01 -2.17122957e-01 -7.18673408e-01 3.64945918e-01 -1.13566196e+00 -1.72885627e-01 6.48472250e-01 -2.99572319e-01 -6.25803709e-01 -1.36704668e-01 2.00879993e-03 -5.08193016e-01 -2.16504604e-01 2.45183051e-01 6.02171302e-01 -3.38969529e-01 9.39476371e-01 1.68786213e-01 2.59978920e-01 -3.66338998e-01 -6.33492544e-02 2.62281418e-01 9.88728181e-02 -3.34077060e-01 3.47338855e-01 3.42615902e-01 1.61826551e-01 -6.64873838e-01 -6.19047463e-01 -6.02361023e-01 -5.71193039e-01 -3.85541290e-01 9.49163139e-01 -7.12584317e-01 -3.83752197e-01 6.21169865e-01 -7.71696568e-01 -6.11287236e-01 -4.44633126e-01 7.26032257e-01 -4.89280075e-01 2.35264227e-01 -6.19439900e-01 -5.47985911e-01 -8.15454066e-01 -1.48530066e+00 7.10830271e-01 5.79646766e-01 -1.23323724e-01 -1.29674625e+00 -2.03887060e-01 1.37323499e-01 5.61979592e-01 5.54858804e-01 7.89871037e-01 -1.01524770e+00 -2.31584370e-01 -6.24323227e-02 -2.42362216e-01 4.37004536e-01 -3.91515195e-02 -2.30636634e-02 -8.61575544e-01 -8.03622425e-01 2.31628105e-01 -1.84195504e-01 8.02714348e-01 7.23197460e-01 1.20139730e+00 2.92406380e-02 -2.54468709e-01 6.24586582e-01 1.43262279e+00 1.34970531e-01 5.87927103e-01 4.49738562e-01 6.64187297e-02 3.32314074e-01 6.78007364e-01 3.94691378e-01 -3.23840380e-02 5.89056432e-01 3.93630236e-01 -6.33302271e-01 -4.24291611e-01 4.29372966e-01 2.96188910e-02 5.80864608e-01 -6.67011589e-02 2.09950700e-01 -1.12846458e+00 3.66226286e-01 -1.36984992e+00 -2.43192375e-01 -3.10731620e-01 2.45374227e+00 8.45038056e-01 2.86799431e-01 1.61829278e-01 2.30550412e-02 4.85119134e-01 -2.78827511e-02 -3.87665451e-01 -5.23230016e-01 1.10717595e-01 2.21633911e-01 8.20289850e-01 3.66542101e-01 -1.32767105e+00 5.71661592e-01 6.40271902e+00 9.37897563e-01 -1.56543040e+00 3.31293166e-01 1.29728830e+00 -1.58829372e-02 3.86100262e-01 -1.26504928e-01 -5.82336187e-01 5.66265583e-01 1.10702527e+00 -9.51606855e-02 2.37182170e-01 6.74391448e-01 3.84964883e-01 -5.41392803e-01 -1.14140999e+00 6.34760857e-01 1.07956447e-01 -1.14227200e+00 -2.53869325e-01 -2.21027568e-01 6.49482548e-01 2.14654952e-01 8.43998324e-03 2.92174399e-01 -2.79841006e-01 -1.17854559e+00 4.64944839e-01 4.76185501e-01 1.07376349e+00 -6.99267149e-01 1.06065106e+00 2.96194524e-01 -8.56484294e-01 1.65650591e-01 -2.78033018e-01 5.41817367e-01 -1.78268611e-01 8.69748592e-01 -1.26734948e+00 9.18332458e-01 7.95141995e-01 5.38781106e-01 -8.29469442e-01 1.61374187e+00 -1.13748394e-01 9.10798013e-01 -1.92124009e-01 5.11634722e-02 3.87922734e-01 -1.79686472e-01 7.63140440e-01 1.62233233e+00 5.60623445e-02 -2.22425535e-01 1.79933354e-01 6.39566898e-01 2.19404057e-01 3.31098229e-01 4.76823598e-02 3.19645405e-01 -5.53740710e-02 1.56734276e+00 -1.23481309e+00 -1.56442970e-01 -1.59351334e-01 5.67717671e-01 -2.27148458e-01 1.53250396e-01 -1.20722938e+00 -1.40362859e-01 1.86945051e-02 1.49445236e-01 1.12367213e-01 2.54769236e-01 -5.45967937e-01 -6.04748726e-01 -2.07473353e-01 -7.78975368e-01 7.83295214e-01 -2.77545124e-01 -1.06131589e+00 8.38535249e-01 1.34371385e-01 -1.31475472e+00 1.06270835e-01 -6.72356069e-01 -6.29878521e-01 8.55284512e-01 -1.70731497e+00 -8.67621303e-01 -2.94377089e-01 2.76286960e-01 6.63014889e-01 -8.11078921e-02 6.46343887e-01 3.72190863e-01 -6.61484480e-01 9.87236023e-01 -6.52532130e-02 7.85124302e-02 6.81621552e-01 -1.38226414e+00 -1.40714183e-01 7.77866781e-01 -1.50542155e-01 2.79295534e-01 8.41626883e-01 -3.42435509e-01 -9.58022833e-01 -1.04019558e+00 3.71076554e-01 -2.45410249e-01 3.17829102e-01 9.25809816e-02 -7.87709892e-01 3.32329422e-01 3.95395458e-01 2.05128998e-01 6.14595890e-01 -1.70132935e-01 3.30108315e-01 -2.08679974e-01 -1.45928323e+00 2.78561503e-01 5.07991314e-01 1.06820323e-01 -3.53883952e-01 5.04466832e-01 3.50319594e-01 -1.01288974e+00 -1.20150506e+00 6.88004673e-01 3.67512584e-01 -1.02251136e+00 7.63239443e-01 -4.50544864e-01 1.26528531e-01 -1.61899880e-01 5.95878184e-01 -1.30673826e+00 1.33004319e-02 -7.03034818e-01 1.50323465e-01 8.16925287e-01 7.86758780e-01 -6.15778685e-01 8.98019075e-01 3.11274499e-01 -2.62244016e-01 -1.43512261e+00 -1.53428578e+00 -6.61003411e-01 2.65704781e-01 -1.50217265e-01 -5.31587154e-02 7.69361913e-01 -3.42429578e-01 -1.78680450e-01 -9.29293633e-02 -1.10711336e-01 6.76131427e-01 -1.97540954e-01 3.97476137e-01 -1.06238616e+00 -6.74348921e-02 -4.80931014e-01 -5.02891421e-01 -4.98463154e-01 -3.30864131e-01 -1.11864340e+00 -3.87434997e-02 -1.12675214e+00 1.97852299e-01 -7.99303114e-01 -5.61669469e-01 5.85084438e-01 -2.84350485e-01 2.75535971e-01 2.57632464e-01 1.92688316e-01 -4.59569305e-01 2.72971969e-02 1.52662218e+00 -1.92893967e-02 -1.91779688e-01 3.18115622e-01 -4.46491957e-01 4.79003847e-01 6.29461229e-01 -5.57608843e-01 -2.59212643e-01 3.93745117e-02 -3.49159449e-01 -8.84660631e-02 2.05020055e-01 -1.28938842e+00 3.18793952e-01 2.95167744e-01 4.37043041e-01 -3.61295491e-01 -2.92567968e-01 -7.32386589e-01 1.58972427e-01 1.17196476e+00 -3.04518372e-01 -4.57048863e-02 4.73109245e-01 1.93922013e-01 3.00001279e-02 -4.12432104e-01 1.29857719e+00 -1.27122298e-01 -5.62709510e-01 2.58740693e-01 -2.34146059e-01 2.52126753e-01 1.58925438e+00 -3.28035980e-01 -1.91347003e-02 2.59544164e-01 -9.19564009e-01 4.42666024e-01 4.33105789e-02 3.22251827e-01 2.38471553e-01 -8.38253915e-01 -1.19744730e+00 2.14963406e-01 -1.07281290e-01 -7.56861717e-02 3.91970217e-01 1.78462219e+00 -8.26209724e-01 4.58931804e-01 -1.83650700e-03 -1.17629027e+00 -1.21978760e+00 3.79797071e-01 9.62975800e-01 -7.70340860e-01 -1.07478690e+00 9.62448955e-01 -7.10103065e-02 -8.15530121e-02 3.75516504e-01 -5.42596221e-01 -2.78034925e-01 1.99995801e-01 4.16495711e-01 4.27654624e-01 5.80975950e-01 -1.27403766e-01 -4.98666286e-01 3.50177765e-01 -3.23613703e-01 3.83005112e-01 1.13208687e+00 1.54815376e-01 3.93544734e-02 1.35803595e-01 9.72524345e-01 -3.41967911e-01 -1.37864089e+00 7.54212886e-02 1.81383401e-01 -2.14203998e-01 1.37433082e-01 -1.20217299e+00 -1.42451465e+00 5.82983017e-01 1.35819006e+00 1.90278038e-01 1.13370323e+00 -1.04228958e-01 3.49904895e-01 -2.24033311e-01 3.85423154e-01 -9.20395553e-01 -2.61084676e-01 2.90227622e-01 9.75471973e-01 -1.08207154e+00 2.33713448e-01 -4.21145588e-01 -8.25903594e-01 1.28902900e+00 5.15092731e-01 -1.45800680e-01 5.61478257e-01 5.21735370e-01 3.13051730e-01 -4.81713265e-02 -4.92224842e-01 5.19092679e-02 2.85969049e-01 4.65841919e-01 5.20504653e-01 5.93701974e-02 -8.51280093e-01 4.59067345e-01 3.91205493e-03 8.28407258e-02 4.04521406e-01 4.75826025e-01 -1.36062175e-01 -9.20681417e-01 -1.20860018e-01 5.09436905e-01 -8.52968216e-01 1.76840741e-02 6.75270334e-02 1.14412212e+00 1.94905549e-01 5.06421745e-01 -1.13207899e-01 3.91408466e-02 5.10600507e-01 -1.06030911e-01 4.54609871e-01 -3.00579011e-01 -1.18956769e+00 -3.98417674e-02 -1.46357361e-02 -6.38160586e-01 -3.07348311e-01 -5.57096958e-01 -1.25216520e+00 -8.87910575e-02 -5.17047822e-01 4.35730934e-01 6.65074229e-01 8.81102204e-01 -7.36224512e-03 8.24296653e-01 6.13514900e-01 -6.29746556e-01 -7.89198697e-01 -1.00083554e+00 -4.47475404e-01 2.20896870e-01 6.66846931e-02 -4.87674385e-01 -5.34550965e-01 -8.77456591e-02]
[14.56166934967041, -2.547041893005371]
a6904c00-d323-4d16-a897-c49428ed54b3
temporal-dynamic-convolutional-neural-network
2110.03213
null
https://arxiv.org/abs/2110.03213v2
https://arxiv.org/pdf/2110.03213v2.pdf
Temporal Dynamic Convolutional Neural Network for Text-Independent Speaker Verification and Phonemetic Analysis
In the field of text-independent speaker recognition, dynamic models that adapt along the time axis have been proposed to consider the phoneme-varying characteristics of speech. However, a detailed analysis of how dynamic models work depending on phonemes is insufficient. In this paper, we propose temporal dynamic CNN (TDY-CNN) that considers temporal variation of phonemes by applying kernels optimally adapting to each time bin. These kernels adapt to time bins by applying weighted sum of trained basis kernels. Then, an analysis of how adaptive kernels work on different phonemes in various layers is carried out. TDY-ResNet-38(x0.5) using six basis kernels improved an equal error rate (EER), the speaker verification performance, by 17.3% compared to the baseline model ResNet-38(x0.5). In addition, we showed that adaptive kernels depend on phoneme groups and are more phoneme-specific at early layers. The temporal dynamic model adapts itself to phonemes without explicitly given phoneme information during training, and results show the necessity to consider phoneme variation within utterances for more accurate and robust text-independent speaker verification.
['Yong-Hwa Park', 'Hyeonuk Nam', 'Seong-Hu Kim']
2021-10-07
null
null
null
null
['text-independent-speaker-recognition', 'text-independent-speaker-verification']
['speech', 'speech']
[-1.28799677e-01 -4.99798596e-01 1.59565628e-01 -6.33092821e-01 -4.61547792e-01 -6.44209087e-01 4.34317559e-01 -1.90001711e-01 -7.44200170e-01 3.05454731e-01 1.49417192e-01 -3.54394853e-01 -1.77411884e-01 -3.60051870e-01 -5.38235784e-01 -9.20455337e-01 -4.97389108e-01 2.52413660e-01 3.14213008e-01 -2.53158152e-01 -3.34308147e-02 6.80477083e-01 -1.41048169e+00 3.95557582e-01 6.84249401e-01 7.01311409e-01 3.44749633e-03 1.13057780e+00 -1.54540583e-01 2.34099537e-01 -8.37002754e-01 -7.02123344e-02 -2.12305156e-03 -3.92505884e-01 -4.25595552e-01 -1.98289245e-01 4.61083293e-01 -5.96180111e-02 -2.25670248e-01 8.34427178e-01 8.59690368e-01 4.81282353e-01 6.25643253e-01 -1.00488091e+00 -6.83312416e-01 1.03098702e+00 3.55421789e-02 6.55261755e-01 -2.39927903e-01 -3.40382010e-02 3.64013672e-01 -1.05202532e+00 1.21624231e-01 1.21723306e+00 9.53783989e-01 9.34080958e-01 -1.05266118e+00 -7.76606321e-01 5.07405877e-01 8.73463750e-01 -1.50342643e+00 -7.92724788e-01 6.72158003e-01 -2.98349440e-01 1.37738287e+00 4.86948133e-01 3.80447745e-01 1.07243836e+00 1.61689088e-01 4.68243420e-01 9.53931212e-01 -6.58643723e-01 2.51558423e-01 2.22984672e-01 7.79656887e-01 1.96624041e-01 -5.34727037e-01 2.77191192e-01 -6.89066172e-01 2.80945331e-01 3.83012474e-01 -2.50116795e-01 -4.40962404e-01 9.42295790e-02 -1.01761615e+00 5.21847010e-01 1.94979936e-01 9.03204024e-01 -3.07082862e-01 8.23897198e-02 6.98523045e-01 5.31637132e-01 6.68663442e-01 -6.79482892e-02 -8.80974770e-01 -4.38805252e-01 -1.02929771e+00 -3.66846085e-01 6.51369035e-01 5.20956695e-01 4.83841002e-01 6.46830022e-01 -2.82692760e-01 1.23801339e+00 2.00285129e-02 4.92208481e-01 9.67173636e-01 -2.62620449e-01 4.41578567e-01 2.05118313e-01 -2.43248358e-01 -4.67371076e-01 -4.98466909e-01 -5.05110979e-01 -9.21557486e-01 3.13182592e-01 6.01136565e-01 -1.50167108e-01 -1.27544391e+00 1.76114261e+00 1.60749361e-01 4.61244255e-01 2.39616111e-01 5.42155504e-01 6.11105084e-01 9.15367305e-01 1.72802284e-01 -1.90442458e-01 1.34000957e+00 -8.92796814e-01 -1.16137767e+00 2.42851973e-01 5.33562660e-01 -7.71102667e-01 1.08209097e+00 3.75662118e-01 -8.35903823e-01 -8.44388425e-01 -9.93660331e-01 3.90787572e-01 -9.49385345e-01 2.13185892e-01 8.68844166e-02 1.26550078e+00 -1.60231066e+00 5.48146605e-01 -8.97472024e-01 -3.84497792e-01 -1.37734905e-01 6.69708848e-01 -2.28910580e-01 3.21792185e-01 -1.42126644e+00 1.10283875e+00 4.36226308e-01 4.03999746e-01 -8.68056417e-01 -7.93462574e-01 -7.10369527e-01 2.69195259e-01 -2.11076826e-01 -1.80680260e-01 1.33013213e+00 -1.06924188e+00 -2.19897270e+00 3.50732654e-01 -5.17150164e-01 -6.72161520e-01 3.86876225e-01 -4.62115072e-02 -1.10334826e+00 -4.96403351e-02 -6.45068645e-01 3.58533502e-01 1.14652920e+00 -8.71742189e-01 -2.65226543e-01 -2.67418861e-01 -3.03876400e-01 1.23128124e-01 -7.36487985e-01 2.94199824e-01 -4.84681517e-01 -6.89366639e-01 -2.27369070e-01 -9.61189568e-01 1.31229788e-01 -4.96277332e-01 -1.75191015e-01 -3.99803638e-01 1.10854983e+00 -1.04841316e+00 1.47083902e+00 -2.41693139e+00 5.81977284e-03 -9.84223839e-03 -2.71294713e-01 5.49385548e-01 -1.36926919e-01 2.27995828e-01 -3.90098840e-01 8.30954537e-02 -2.35607222e-01 -5.80282450e-01 1.20063178e-01 1.12782396e-01 -2.86711991e-01 4.43482667e-01 -3.27954255e-02 8.39327037e-01 -3.75194401e-01 -1.13901116e-01 4.87098157e-01 1.00290501e+00 -1.74532190e-01 -1.94233298e-01 2.63137341e-01 2.46413946e-01 2.44209096e-01 3.01850319e-01 9.07637835e-01 3.49727452e-01 -3.00911903e-01 -1.57835558e-01 -3.12282115e-01 1.64948508e-01 -9.38463867e-01 1.37078238e+00 -8.61416221e-01 1.05000627e+00 1.00276142e-01 -7.98183918e-01 7.53384829e-01 7.16017962e-01 1.59785956e-01 -4.60760325e-01 7.79327378e-02 1.44766748e-01 2.66074091e-01 -1.87750071e-01 4.67331707e-01 -1.53181195e-01 2.02993065e-01 1.71823874e-01 2.91502953e-01 2.79837042e-01 -1.95209041e-01 -1.96112618e-01 7.44134426e-01 -4.67462718e-01 -1.57791257e-01 -2.82539219e-01 7.94854164e-01 -6.87907279e-01 2.73488760e-01 6.75196886e-01 -4.14871722e-01 6.27172828e-01 -1.60617173e-01 -4.12324280e-01 -7.28793859e-01 -1.08555377e+00 -4.73664880e-01 1.38954794e+00 -3.17272633e-01 -6.72490299e-02 -1.09101987e+00 -5.50777674e-01 -1.80336952e-01 1.01895249e+00 -8.70991051e-01 -2.60108948e-01 -9.44138944e-01 -9.48331833e-01 8.75246882e-01 5.16132534e-01 6.23583913e-01 -1.06623483e+00 -1.91050544e-01 5.23739398e-01 1.26700372e-01 -1.07293284e+00 -9.13532197e-01 5.07931232e-01 -8.14045191e-01 -5.40262878e-01 -1.02963507e+00 -6.36579871e-01 4.46492165e-01 -4.95409742e-02 6.58242524e-01 -2.18791738e-01 1.01886466e-01 6.79516256e-01 -4.81934339e-01 -3.06678236e-01 -5.77399373e-01 2.93482184e-01 4.93049324e-01 4.77409780e-01 3.95736039e-01 -5.47103167e-01 -2.26792216e-01 6.03842199e-01 -7.70369768e-01 -4.55544978e-01 1.59058392e-01 7.69011557e-01 6.37250021e-02 1.50267228e-01 5.47377050e-01 -4.02293205e-01 5.06095588e-01 -1.34757325e-01 -4.91152048e-01 5.02980351e-01 -6.48391128e-01 -7.49410018e-02 7.59769738e-01 -1.07266748e+00 -1.17342973e+00 -7.30825961e-02 -4.79691654e-01 -4.46449190e-01 -2.56343186e-01 1.89371228e-01 -2.27124378e-01 -1.09504253e-01 7.04469562e-01 6.44715488e-01 -2.05958307e-01 -5.73143303e-01 4.09676224e-01 7.20908582e-01 4.58797872e-01 -1.39527529e-01 7.07568169e-01 2.09237084e-01 -8.26300383e-01 -1.18054199e+00 -7.40515143e-02 -4.84211475e-01 -7.84182370e-01 -3.48238975e-01 8.85510206e-01 -6.89466000e-01 -7.26580083e-01 1.12897515e+00 -1.23085678e+00 -7.67000437e-01 -2.69330591e-01 7.41947889e-01 -2.62321293e-01 2.18949810e-01 -6.38772309e-01 -8.85363638e-01 -3.96645844e-01 -1.00774479e+00 5.95607698e-01 1.44701570e-01 8.92511979e-02 -1.36931109e+00 7.70445019e-02 -1.26454771e-01 1.03822422e+00 -4.05825257e-01 5.70290089e-01 -8.47872615e-01 -2.82118320e-01 -2.72407115e-01 1.42889872e-01 6.47389293e-01 3.09833378e-01 1.80411518e-01 -1.51298606e+00 -3.37248504e-01 2.24862427e-01 4.47690487e-01 1.03354061e+00 7.99940646e-01 1.09927356e+00 -4.80506361e-01 -2.07556114e-01 7.02072859e-01 9.72899258e-01 6.04562342e-01 6.52521372e-01 2.16433957e-01 5.97788751e-01 5.15351832e-01 -8.01379830e-02 8.58266950e-02 2.80510724e-01 1.13134480e+00 -6.55077547e-02 6.49927109e-02 -4.53292400e-01 2.02145308e-01 9.14130509e-01 1.34902501e+00 1.70963220e-04 -5.31516671e-01 -9.09415483e-01 6.76874220e-01 -1.44230866e+00 -1.06517196e+00 -1.11631295e-02 2.12390971e+00 7.50543833e-01 1.58686727e-01 1.07749380e-01 4.26873147e-01 1.09864461e+00 3.18461880e-02 -5.76841712e-01 -7.52967358e-01 -3.48726153e-01 9.98740196e-02 5.39510071e-01 9.19356167e-01 -8.44347537e-01 1.09012449e+00 6.40079260e+00 1.17481804e+00 -1.67553043e+00 5.07719874e-01 4.06194508e-01 3.67529481e-03 -3.72421965e-02 -3.31341654e-01 -1.11453593e+00 5.53094923e-01 1.58881295e+00 -1.14401884e-01 5.30348480e-01 5.37456810e-01 3.05138350e-01 1.84201211e-01 -9.44619715e-01 1.02396226e+00 2.10727096e-01 -1.04448187e+00 -1.29536569e-01 -1.50781497e-01 4.83344048e-01 1.56428069e-01 4.68309999e-01 5.39892018e-01 2.12633967e-01 -9.66385603e-01 7.65048027e-01 5.76241910e-01 7.26794124e-01 -7.19471812e-01 6.44982457e-01 1.48616701e-01 -1.35446942e+00 -1.12869099e-01 -1.64257258e-01 2.98318833e-01 2.80017436e-01 6.50042295e-01 -1.26371574e+00 2.22376376e-01 9.01319206e-01 5.31602204e-01 -5.68098485e-01 9.35466826e-01 1.03071488e-01 1.11100316e+00 -3.13158959e-01 -6.19540997e-02 -8.92317668e-03 2.97806174e-01 5.33975780e-01 1.80909538e+00 5.05475044e-01 -1.55975342e-01 -5.58476985e-01 3.64632964e-01 3.28396320e-01 1.35231957e-01 -2.13082746e-01 1.34151608e-01 4.87558782e-01 8.49252880e-01 -6.64795816e-01 -4.42986280e-01 6.27712905e-02 1.31968415e+00 6.87624589e-02 7.92237639e-01 -8.22022438e-01 -4.78672206e-01 8.46007645e-01 -2.69642800e-01 5.14152706e-01 -5.45937896e-01 -3.59560788e-01 -1.03161180e+00 1.06784506e-02 -5.42638481e-01 2.28687748e-01 -3.83588523e-01 -1.17661464e+00 1.00073791e+00 4.74155135e-02 -1.11471879e+00 -4.10849810e-01 -7.75882602e-01 -8.05701792e-01 1.11816430e+00 -1.57274473e+00 -9.70619678e-01 5.34614958e-02 7.62985051e-01 8.53823960e-01 -3.36250693e-01 1.00176239e+00 3.53508085e-01 -6.88370407e-01 1.29981267e+00 2.70999879e-01 1.60508171e-01 7.17329025e-01 -1.14326441e+00 8.42181623e-01 8.33950102e-01 2.60958165e-01 7.74061859e-01 5.46454251e-01 -1.77199766e-01 -9.32224751e-01 -1.01745188e+00 1.17890179e+00 -5.73671639e-01 4.70243573e-01 -6.26599193e-01 -1.40806198e+00 4.99044627e-01 3.94650638e-01 1.49713904e-02 6.78750217e-01 1.57925338e-01 -5.89809656e-01 -5.07290781e-01 -1.02351928e+00 4.27436471e-01 7.14909136e-01 -9.21830177e-01 -3.87720048e-01 5.25227413e-02 8.83218527e-01 -3.37846786e-01 -7.48897612e-01 3.46686602e-01 5.01400292e-01 -1.04660094e+00 1.05530655e+00 -1.52935684e-01 -6.59224331e-01 -3.72295648e-01 -7.93685243e-02 -1.54932892e+00 -3.36127490e-01 -5.14341354e-01 -1.74912632e-01 1.34776163e+00 6.28753543e-01 -9.98877823e-01 4.03296292e-01 4.24046904e-01 -5.72887480e-01 -2.00277984e-01 -1.60213709e+00 -1.28597641e+00 1.84436843e-01 -9.44715261e-01 6.84642673e-01 9.83430088e-01 -2.19827548e-01 -2.40861550e-01 -2.85830408e-01 4.67005134e-01 9.95912254e-02 -6.27087295e-01 4.45067972e-01 -7.38218725e-01 -2.64357775e-01 -7.69783139e-01 -5.28937697e-01 -9.31294858e-01 2.69645393e-01 -7.27728546e-01 1.96764156e-01 -9.48068857e-01 -3.49638790e-01 -3.95002186e-01 -6.99371696e-01 5.87577522e-01 -2.80440420e-01 1.13099262e-01 1.96150005e-01 -9.20406580e-02 -2.01675788e-01 6.64534152e-01 7.13608742e-01 -3.54206681e-01 -5.79790831e-01 2.58294314e-01 8.66958797e-02 4.17727858e-01 1.11787295e+00 -2.88387269e-01 -2.69252628e-01 -3.85833859e-01 -4.04796928e-01 -2.56470859e-01 3.07086170e-01 -1.38407362e+00 5.23899913e-01 2.81693697e-01 3.03281248e-01 -7.19546974e-01 6.21282160e-01 -8.32961202e-01 2.17330262e-01 4.81147110e-01 -2.72754818e-01 1.66295782e-01 8.30208242e-01 3.85605127e-01 -3.42848569e-01 -1.42667040e-01 8.15447092e-01 2.01832891e-01 -6.99507535e-01 2.01246008e-01 -7.50120401e-01 -5.30245781e-01 7.45990872e-01 -5.58478355e-01 5.08644572e-03 -2.84290105e-01 -9.57801521e-01 -2.04638302e-01 1.49330914e-01 6.23687983e-01 5.84680736e-01 -1.25963557e+00 -7.66680479e-01 2.92107552e-01 -4.18096632e-02 -6.29137099e-01 9.04752731e-01 7.81114399e-01 -7.95966834e-02 4.86382365e-01 -2.32428424e-02 -7.51580477e-01 -1.48268378e+00 5.14438331e-01 7.78157651e-01 4.44412790e-02 -3.44846278e-01 1.10157382e+00 1.33160815e-01 -6.62088156e-01 6.26394749e-01 -6.86116397e-01 -4.90880400e-01 3.43838006e-01 6.64477646e-01 3.43998075e-01 4.14886236e-01 -8.00341964e-01 -5.40339470e-01 9.43690181e-01 -2.33325407e-01 -4.42222923e-01 1.16475725e+00 -3.02586615e-01 1.78828344e-01 9.29181457e-01 1.31271267e+00 1.00889407e-01 -1.30803907e+00 -1.73491642e-01 -5.95776457e-03 7.73948431e-02 -2.58553866e-02 -1.06098652e+00 -1.01630127e+00 1.11330163e+00 1.19209898e+00 2.58889586e-01 1.28604805e+00 -4.10273641e-01 6.56532764e-01 1.26975015e-01 -2.37989426e-02 -1.06162095e+00 -1.30172372e-01 1.04578161e+00 9.10119057e-01 -8.80724132e-01 -6.99862599e-01 -7.82696977e-02 -4.19303089e-01 1.49399817e+00 4.38401937e-01 4.02528137e-01 1.04852509e+00 1.63410038e-01 5.52326143e-01 3.55621517e-01 -5.57242930e-01 4.21550684e-02 5.34814715e-01 7.59777367e-01 3.84460002e-01 2.03377441e-01 4.57416661e-02 3.40333164e-01 -3.77816886e-01 -3.79723221e-01 2.74946570e-01 4.02442902e-01 -2.98832417e-01 -1.19058740e+00 -6.74281120e-01 -1.92254037e-01 -1.99519649e-01 -3.76956075e-01 -1.59748226e-01 5.55966020e-01 1.95464611e-01 1.02967048e+00 -1.03160376e-02 -6.33438706e-01 3.23386908e-01 6.97063744e-01 2.79679686e-01 -3.35846990e-01 -8.92116308e-01 -1.32939056e-01 -2.32590303e-01 -3.28921266e-02 -3.02657902e-01 -7.26008177e-01 -1.24257874e+00 -1.94237202e-01 -4.58153337e-01 1.39907971e-01 1.20256293e+00 7.43723571e-01 3.48907590e-01 8.45118284e-01 7.44168162e-01 -8.32946301e-01 -4.49279070e-01 -1.31276405e+00 -5.33503711e-01 -3.16161774e-02 8.22849691e-01 -3.63830864e-01 -7.36890554e-01 2.36489877e-01]
[14.394279479980469, 6.107877254486084]
e22b8ce6-e186-4d34-b539-7e8c0ef11554
retrieval-as-attention-end-to-end-learning-of
2212.02027
null
https://arxiv.org/abs/2212.02027v1
https://arxiv.org/pdf/2212.02027v1.pdf
Retrieval as Attention: End-to-end Learning of Retrieval and Reading within a Single Transformer
Systems for knowledge-intensive tasks such as open-domain question answering (QA) usually consist of two stages: efficient retrieval of relevant documents from a large corpus and detailed reading of the selected documents to generate answers. Retrievers and readers are usually modeled separately, which necessitates a cumbersome implementation and is hard to train and adapt in an end-to-end fashion. In this paper, we revisit this design and eschew the separate architecture and training in favor of a single Transformer that performs Retrieval as Attention (ReAtt), and end-to-end training solely based on supervision from the end QA task. We demonstrate for the first time that a single model trained end-to-end can achieve both competitive retrieval and QA performance, matching or slightly outperforming state-of-the-art separately trained retrievers and readers. Moreover, end-to-end adaptation significantly boosts its performance on out-of-domain datasets in both supervised and unsupervised settings, making our model a simple and adaptable solution for knowledge-intensive tasks. Code and models are available at https://github.com/jzbjyb/ReAtt.
['Graham Neubig', 'Jamie Callan', 'Zhiruo Wang', 'Haibo Ding', 'Jun Araki', 'Luyu Gao', 'Zhengbao Jiang']
2022-12-05
null
null
null
null
['passage-retrieval', 'open-domain-question-answering']
['natural-language-processing', 'natural-language-processing']
[ 4.06329334e-02 3.37163150e-01 9.76833049e-03 -4.70639944e-01 -1.76587951e+00 -8.50269616e-01 6.90679610e-01 2.43399873e-01 -5.55454195e-01 5.01617432e-01 2.19693318e-01 -2.95717269e-01 -3.04816872e-01 -6.08303487e-01 -8.65653872e-01 -2.35026613e-01 4.77477044e-01 1.34850562e+00 4.98008788e-01 -5.02441168e-01 1.45464540e-01 -4.21960540e-02 -1.51500297e+00 4.61892962e-01 1.09954095e+00 9.12797332e-01 3.59355628e-01 8.40262771e-01 -2.24271774e-01 8.31441402e-01 -4.56362426e-01 -7.80713916e-01 8.29053223e-02 -2.98792511e-01 -1.38969541e+00 -3.06631356e-01 5.93431771e-01 -4.26728457e-01 -3.70683879e-01 5.14887094e-01 8.41217935e-01 2.72776514e-01 6.15283072e-01 -5.27869105e-01 -1.25142074e+00 5.18602729e-01 -9.74531621e-02 1.86779499e-01 5.60614467e-01 6.11650087e-02 1.27138484e+00 -9.78423536e-01 4.86171693e-01 9.73003507e-01 2.36467436e-01 6.96601808e-01 -1.03286397e+00 -2.16596305e-01 9.01519582e-02 2.91250736e-01 -1.14015019e+00 -8.12585413e-01 5.99964559e-01 -1.47353886e-02 1.11010623e+00 1.41864553e-01 -1.87127534e-02 9.31675375e-01 -3.50626320e-01 1.29334652e+00 6.51137412e-01 -7.44946420e-01 1.94723625e-02 -1.25883976e-02 4.11725670e-01 4.41089779e-01 -7.83220455e-02 -3.43982935e-01 -3.64269018e-01 -1.38387024e-01 2.16045320e-01 -2.74571143e-02 -4.16619092e-01 -2.86887676e-01 -1.13384068e+00 7.85307407e-01 6.12600505e-01 1.30745947e-01 -4.19914812e-01 -6.21246472e-02 2.56627917e-01 6.93730056e-01 4.41431910e-01 6.57915533e-01 -5.99917471e-01 1.24610085e-02 -8.82711411e-01 4.48012143e-01 9.39666629e-01 1.00157797e+00 7.37558782e-01 -6.61495745e-01 -5.20920157e-01 1.21019590e+00 1.10818200e-01 7.58053422e-01 6.23696446e-01 -8.18647623e-01 6.37232661e-01 4.87021416e-01 3.38746518e-01 -3.20790172e-01 -2.37256125e-01 -6.03959262e-01 -4.28865612e-01 -3.68917376e-01 5.91953516e-01 -2.22684257e-02 -1.07072437e+00 1.56296468e+00 3.06455255e-01 -4.08096403e-01 1.68300152e-01 9.93919969e-01 1.01524007e+00 9.07785594e-01 4.36408892e-02 6.58508763e-02 1.54275954e+00 -1.62663770e+00 -6.39079571e-01 -5.31748414e-01 5.75187683e-01 -9.40580368e-01 1.47614598e+00 2.13235945e-01 -1.55303240e+00 -4.54163045e-01 -6.57240689e-01 -8.36905599e-01 -4.10292447e-01 1.91473160e-02 1.53291181e-01 -6.58672750e-02 -1.11142635e+00 1.81864604e-01 -6.22260749e-01 -2.99663961e-01 1.86138734e-01 2.40093365e-01 -1.54142395e-01 -4.93804872e-01 -1.30617917e+00 1.01586592e+00 1.91912636e-01 -1.97442807e-02 -8.14274848e-01 -6.32468879e-01 -5.74494839e-01 3.12055051e-01 6.20956302e-01 -1.19394815e+00 2.07413244e+00 -7.85834432e-01 -1.63982415e+00 1.00396442e+00 -3.70896101e-01 -4.01575446e-01 2.77739048e-01 -7.62267470e-01 -1.77108824e-01 4.74457830e-01 1.81507990e-01 4.39311534e-01 7.52321243e-01 -9.80475187e-01 -4.06574398e-01 -3.92902404e-01 2.69534886e-01 4.69052553e-01 -1.96528524e-01 2.71821797e-01 -9.12441075e-01 -3.39984804e-01 -1.27809480e-01 -7.13650227e-01 -3.30600590e-02 -3.33399802e-01 -1.60382420e-01 -6.39324367e-01 2.40065679e-01 -9.79008913e-01 1.23128235e+00 -1.78849113e+00 2.18324646e-01 -1.42818391e-01 1.91186443e-01 5.21272242e-01 -4.66931939e-01 7.93529630e-01 2.39586279e-01 -3.51321131e-01 -1.74758866e-01 -4.20584321e-01 2.70430356e-01 -2.10198984e-01 -5.28949499e-01 -1.39738889e-02 2.97207624e-01 1.37293041e+00 -1.10200405e+00 -3.21560591e-01 -2.69058079e-01 2.46411666e-01 -4.22820359e-01 6.15327954e-01 -8.18851769e-01 2.41043150e-01 -7.35334277e-01 5.55820882e-01 2.40577370e-01 -7.53425300e-01 -1.64582908e-01 2.02950791e-01 4.04956013e-01 9.04185057e-01 -5.90062976e-01 1.92725945e+00 -7.95893431e-01 4.34684604e-01 2.57714856e-02 -8.71558309e-01 8.19839239e-01 4.15592164e-01 -1.36571705e-01 -1.15531421e+00 2.29698829e-02 4.80251461e-01 -2.27503821e-01 -5.74829459e-01 5.78914523e-01 1.88563243e-01 -1.20542645e-01 8.08160841e-01 3.89867187e-01 -3.82271525e-03 3.15459847e-01 4.78095412e-01 1.24218285e+00 1.26096718e-02 1.96452793e-02 8.79161656e-02 4.48312640e-01 1.64229065e-01 -1.02921665e-01 1.01310110e+00 1.74613819e-01 7.98706174e-01 1.64625086e-02 -9.55078378e-02 -9.32266414e-01 -1.07038784e+00 5.61194383e-02 1.68827784e+00 -5.00229113e-02 -2.37356946e-01 -6.32370889e-01 -7.96893537e-01 -1.46835446e-01 7.16143906e-01 -2.85141766e-01 -2.46192738e-01 -6.16633713e-01 -2.60786146e-01 4.50713396e-01 4.19654191e-01 3.84189159e-01 -1.17206430e+00 -3.43628526e-01 3.26674104e-01 -5.88334441e-01 -8.91602993e-01 -4.39274698e-01 1.17872022e-02 -7.82910764e-01 -8.84313881e-01 -1.21959603e+00 -9.99986291e-01 5.79218388e-01 3.19908082e-01 1.72283828e+00 2.67219216e-01 1.85390115e-01 6.03145421e-01 -6.73517764e-01 -3.44445646e-01 -2.03484356e-01 6.24076188e-01 -4.48233277e-01 -1.00818411e-01 5.18415570e-01 -3.27469975e-01 -9.34424460e-01 3.11211079e-01 -8.38954866e-01 -6.66278601e-02 8.97436202e-01 9.22056019e-01 5.18836915e-01 -5.64903975e-01 8.64695370e-01 -9.38161671e-01 8.57900798e-01 -5.07349849e-01 -4.98603433e-01 7.61336148e-01 -5.91077805e-01 2.45500475e-01 6.20091140e-01 -2.44905695e-01 -1.14188325e+00 -2.66559958e-01 -3.12572569e-01 -2.55504817e-01 -4.11829092e-02 7.08049357e-01 1.14327751e-01 3.36278886e-01 9.59051073e-01 3.08954388e-01 -3.46303657e-02 -8.78173530e-01 6.50510788e-01 1.08265710e+00 6.87008977e-01 -5.03367722e-01 8.19787681e-01 6.78829700e-02 -7.52889335e-01 -2.41671845e-01 -1.39425254e+00 -7.56759405e-01 -4.35648888e-01 7.13266656e-02 5.02762318e-01 -1.13446355e+00 -3.21316063e-01 3.17429096e-01 -1.18739247e+00 -5.64997077e-01 -2.32235312e-01 8.21796060e-02 -4.54927742e-01 1.39591232e-01 -6.23962581e-01 -5.50120354e-01 -1.06966269e+00 -7.79506385e-01 1.24850869e+00 3.49218220e-01 -6.08806387e-02 -9.19131339e-01 2.29942262e-01 1.13540912e+00 7.14379728e-01 -4.95717466e-01 7.92621911e-01 -1.10686851e+00 -7.70413041e-01 -6.60493970e-01 -1.92774415e-01 2.91145295e-01 -2.82433718e-01 -5.11829674e-01 -1.11280847e+00 -2.92220742e-01 -2.22279206e-01 -9.30408537e-01 1.09227622e+00 -3.04747280e-03 9.11405027e-01 -2.65661985e-01 -1.93644688e-01 1.54847309e-01 1.10411489e+00 -1.88621879e-01 5.89137971e-01 4.62115377e-01 4.48425174e-01 6.87373459e-01 6.29178226e-01 -1.40027314e-01 7.72464335e-01 7.13083744e-01 4.37043458e-02 5.49267139e-03 -3.59560728e-01 -3.42746258e-01 2.90466368e-01 7.67575443e-01 2.73819149e-01 -4.69386458e-01 -1.09726512e+00 1.04053330e+00 -2.03459859e+00 -8.18131030e-01 1.03721254e-01 2.27732444e+00 1.32249260e+00 -6.22694977e-02 9.24742501e-03 -2.97992826e-01 4.10630375e-01 -2.20564995e-02 -5.71442544e-01 -2.82160878e-01 1.60231411e-01 5.40286124e-01 5.20631624e-03 7.26629078e-01 -8.14315438e-01 1.03889430e+00 5.90783405e+00 8.60087812e-01 -8.08625281e-01 2.91645199e-01 5.17607570e-01 -2.20634252e-01 -4.22686607e-01 1.48875356e-01 -8.01242888e-01 2.56511331e-01 1.22087932e+00 -1.02291971e-01 4.58295226e-01 7.21417904e-01 -2.47466624e-01 -3.43038477e-02 -1.20338440e+00 6.24532878e-01 1.54030010e-01 -1.06338573e+00 2.49732077e-01 -4.82585818e-01 5.95683992e-01 4.47478890e-01 -5.73804276e-03 7.22629905e-01 3.49826366e-01 -8.55120242e-01 4.71140772e-01 5.69410563e-01 4.94859248e-01 -4.11330014e-01 8.03069472e-01 6.14406407e-01 -5.77595472e-01 -6.88872039e-02 -4.44565833e-01 7.50190252e-03 3.01963031e-01 4.66875106e-01 -8.15450728e-01 6.92056477e-01 6.96482897e-01 3.00299793e-01 -7.36685753e-01 1.07298124e+00 -4.94313240e-01 7.10867763e-01 -3.16067606e-01 -1.46010727e-01 2.64953852e-01 1.64135709e-01 2.83321708e-01 1.01808786e+00 8.46131295e-02 1.09482482e-01 -1.23959132e-01 6.31578147e-01 -4.68418807e-01 1.91519290e-01 -2.85119087e-01 -8.05960819e-02 5.83471239e-01 1.11958420e+00 -5.58458231e-02 -5.21834850e-01 -5.40786564e-01 1.00744045e+00 8.58210027e-01 5.38406134e-01 -5.20619929e-01 -6.90547884e-01 2.35724747e-02 1.48533449e-01 6.12863600e-01 1.91096902e-01 -1.55245569e-02 -1.35619426e+00 4.40313399e-01 -1.03922176e+00 8.16805482e-01 -9.63593900e-01 -1.42671573e+00 7.99328506e-01 -1.24894612e-01 -9.88677859e-01 -6.28870487e-01 -3.38103414e-01 -4.76649076e-01 1.03561831e+00 -2.04149914e+00 -1.11580527e+00 -2.51598209e-01 6.96926415e-01 6.97232664e-01 1.88495908e-02 9.06387985e-01 3.62689942e-01 -3.49084884e-01 7.91629672e-01 5.35713911e-01 2.31674030e-01 1.06259298e+00 -1.25810194e+00 5.01152039e-01 7.62500286e-01 1.61483929e-01 8.14387202e-01 4.98112410e-01 -2.31647491e-01 -1.43854940e+00 -8.52401018e-01 1.54111016e+00 -9.56998050e-01 6.55442715e-01 -3.61651689e-01 -1.08310652e+00 7.64433265e-01 6.64820194e-01 -2.91244477e-01 5.88624179e-01 5.03458381e-01 -4.49741602e-01 -1.94247842e-01 -7.28646815e-01 5.84195793e-01 7.71280229e-01 -7.94794858e-01 -1.13109636e+00 7.49400735e-01 7.90701449e-01 -5.84686935e-01 -7.48361945e-01 2.15969279e-01 3.82515758e-01 -6.29773259e-01 1.09130275e+00 -6.31271362e-01 5.21333516e-01 -1.43641829e-01 1.41171739e-01 -1.16153800e+00 -2.23526552e-01 -7.40911901e-01 -4.88627344e-01 1.19743586e+00 8.09175909e-01 -5.52246034e-01 2.69356340e-01 7.81665206e-01 -2.27325857e-01 -8.44171882e-01 -6.58058107e-01 -5.60946643e-01 2.24608615e-01 -7.12026879e-02 4.92284477e-01 6.09129965e-01 4.17268984e-02 9.94342387e-01 -5.47456406e-02 1.00525767e-01 3.47552001e-01 5.08278608e-01 8.78733277e-01 -1.06677687e+00 -4.84785825e-01 -3.55433196e-01 3.53037804e-01 -1.69214308e+00 3.48402113e-02 -9.80701745e-01 1.93106994e-01 -1.91363037e+00 1.93402156e-01 -2.91721940e-01 -3.11739773e-01 5.29856086e-01 -5.46235204e-01 1.00067094e-01 7.22177103e-02 4.74286914e-01 -1.27264380e+00 7.21611261e-01 1.23356926e+00 -2.43162885e-01 -1.80159122e-01 9.24625695e-02 -1.17275190e+00 1.73502743e-01 6.50836468e-01 -4.86724406e-01 -5.88646233e-01 -1.18530619e+00 3.53703409e-01 2.34903544e-01 3.10745507e-01 -6.59119666e-01 4.02283013e-01 3.36159974e-01 1.38697073e-01 -4.71070647e-01 3.12652260e-01 -5.26702523e-01 -5.14704466e-01 -2.46060714e-02 -6.67324960e-01 1.55660100e-02 1.09430835e-01 4.62080985e-01 -4.03675675e-01 -4.79605347e-01 3.67615521e-01 -1.66136384e-01 -4.46760327e-01 2.10267961e-01 -4.71352506e-03 5.40030956e-01 4.56983626e-01 1.86937153e-01 -7.60379136e-01 -7.44781494e-01 -4.48453009e-01 7.54921794e-01 2.28981718e-01 5.13028502e-01 3.83350790e-01 -1.03933275e+00 -9.95090902e-01 -1.47926390e-01 3.77775162e-01 4.63415474e-01 3.90589327e-01 8.01208556e-01 -4.16676164e-01 8.19083929e-01 3.27805310e-01 -3.58168304e-01 -9.00162697e-01 4.97864246e-01 3.15446347e-01 -7.60729134e-01 -4.71358389e-01 1.03855264e+00 1.14690252e-01 -7.72828043e-01 2.47414783e-01 -1.33063616e-02 -1.90569520e-01 -1.52975526e-02 7.14258015e-01 3.00296526e-02 4.63267595e-01 -1.78642303e-01 -1.51790440e-01 3.96800965e-01 -6.76074505e-01 -1.42304122e-01 1.20671451e+00 -2.26577118e-01 -4.82459255e-02 3.87967117e-02 1.09029567e+00 -2.03934386e-01 -9.55642343e-01 -8.97837043e-01 -1.16477096e-02 -1.91305086e-01 1.33752048e-01 -1.39705312e+00 -6.02808654e-01 8.15508783e-01 1.61977082e-01 8.43251571e-02 1.27781475e+00 4.35971260e-01 1.23165011e+00 1.09757698e+00 1.41866848e-01 -1.11255300e+00 1.96451798e-01 7.14448929e-01 1.11553383e+00 -1.34838843e+00 -2.47222275e-01 9.25468653e-02 -6.25185013e-01 7.39382327e-01 4.74804103e-01 -4.86614890e-02 3.61266822e-01 -4.52295393e-01 4.33088094e-01 -3.85235220e-01 -1.10500026e+00 -4.49247807e-01 6.74081743e-01 3.56643587e-01 4.81714278e-01 -3.51217449e-01 -1.76774994e-01 6.94834709e-01 -1.90293327e-01 4.74523008e-02 5.03062382e-02 1.20236349e+00 -4.89728689e-01 -1.11092424e+00 -2.03090072e-01 4.82351780e-01 -4.52031076e-01 -3.47450644e-01 -4.40563679e-01 5.26922584e-01 -6.66804612e-01 1.22623301e+00 -2.46549249e-02 5.52807190e-02 5.58307171e-01 5.02348483e-01 4.28220689e-01 -7.60475159e-01 -7.22485840e-01 -1.66824132e-01 3.24331015e-01 -4.16182697e-01 -1.97479457e-01 -4.18642312e-01 -1.08813405e+00 -5.44075444e-02 -5.39345503e-01 5.17308235e-01 2.72581190e-01 1.02181375e+00 8.85669887e-01 2.67759562e-01 4.89909708e-01 -3.88251692e-01 -1.01749456e+00 -1.23395395e+00 -2.19659768e-02 4.34570044e-01 5.28223038e-01 -2.65370637e-01 -1.57259867e-01 -1.30944292e-03]
[11.345090866088867, 7.8913798332214355]
661590e5-9f1d-42fb-aca1-b0cbe1d7483d
sampling-matters-an-empirical-study-of
null
null
https://aclanthology.org/D19-1128
https://aclanthology.org/D19-1128.pdf
Sampling Matters! An Empirical Study of Negative Sampling Strategies for Learning of Matching Models in Retrieval-based Dialogue Systems
We study how to sample negative examples to automatically construct a training set for effective model learning in retrieval-based dialogue systems. Following an idea of dynamically adapting negative examples to matching models in learning, we consider four strategies including minimum sampling, maximum sampling, semi-hard sampling, and decay-hard sampling. Empirical studies on two benchmarks with three matching models indicate that compared with the widely used random sampling strategy, although the first two strategies lead to performance drop, the latter two ones can bring consistent improvement to the performance of all the models on both benchmarks.
['Chongyang Tao', 'Wei Wu', 'Rui Yan', 'Dongyan Zhao', 'Yansong Feng', 'Jia Li']
2019-11-01
null
null
null
ijcnlp-2019-11
['conversational-response-selection']
['natural-language-processing']
[ 2.60706931e-01 3.56150657e-01 -5.54704249e-01 -3.93541068e-01 -1.10657585e+00 -5.42112112e-01 1.11000276e+00 1.74458757e-01 -8.56601417e-01 1.00859797e+00 6.38416186e-02 -3.88718903e-01 -1.39617827e-02 -6.12399280e-01 -6.91121519e-02 -5.14824986e-01 5.63041344e-02 1.17554712e+00 6.56498671e-01 -8.06209266e-01 6.12001181e-01 2.03864262e-01 -1.25436485e+00 5.26975811e-01 8.17326427e-01 6.76276267e-01 1.03693649e-01 9.07993078e-01 -6.75154746e-01 7.68160522e-01 -1.06097031e+00 -4.40913916e-01 2.71692663e-01 -5.46301365e-01 -1.25387025e+00 -1.37113467e-01 5.49572408e-02 -1.88389525e-01 1.35040116e-02 8.34679246e-01 7.62438953e-01 3.99115562e-01 8.94896924e-01 -1.02480805e+00 -3.21804702e-01 5.12391090e-01 -4.40720677e-01 2.32729658e-01 7.71556556e-01 5.58754876e-02 7.95378923e-01 -8.93537104e-01 6.42994523e-01 1.71831715e+00 5.17697930e-01 1.15729725e+00 -1.25601387e+00 -3.86097312e-01 -1.46025300e-01 3.55467051e-02 -9.47649181e-01 -7.15643644e-01 5.63156962e-01 -1.34636134e-01 1.05854130e+00 7.11083770e-01 8.31579626e-01 8.29139888e-01 -2.03704029e-01 1.00980330e+00 1.31243157e+00 -9.83265698e-01 2.77700722e-01 8.10871303e-01 5.16345322e-01 3.93753260e-01 -7.74414986e-02 7.65921324e-02 -5.90135098e-01 -8.34299743e-01 2.53433019e-01 -4.88993198e-01 -2.24757567e-01 -1.86722860e-01 -6.41935170e-01 9.96154010e-01 -5.58814555e-02 2.81041026e-01 -2.74174035e-01 -3.68550450e-01 5.93443930e-01 7.18171179e-01 6.81739926e-01 9.08874452e-01 -7.27639854e-01 -3.95911425e-01 -4.77787226e-01 6.18941307e-01 1.28166628e+00 8.25136006e-01 6.08834028e-01 -6.14397600e-02 -6.01956487e-01 1.33325028e+00 1.97924912e-01 3.13442320e-01 8.17644000e-01 -9.12718773e-01 5.72780609e-01 5.93493819e-01 4.98933047e-01 -3.12388629e-01 -3.22787613e-01 3.51693830e-03 -3.19717199e-01 -1.24394149e-01 6.12254679e-01 -2.95410067e-01 -8.52627873e-01 1.47569597e+00 3.54242504e-01 -3.18348855e-01 2.67148525e-01 4.38636035e-01 9.08891380e-01 6.43165767e-01 2.47437179e-01 -8.27526033e-01 7.50596285e-01 -1.14382660e+00 -9.03242767e-01 -2.15628535e-01 1.08382154e+00 -8.85793626e-01 1.31284106e+00 6.03504479e-01 -1.19497871e+00 -2.95841515e-01 -8.18391263e-01 3.39007616e-01 -4.03457999e-01 -2.88187325e-01 6.72453582e-01 7.38712072e-01 -1.20735598e+00 5.30574620e-01 -1.36924475e-01 -2.67225832e-01 -1.78093776e-01 4.34553206e-01 -3.46047953e-02 -1.67452861e-02 -1.52668369e+00 1.20130742e+00 4.13251519e-01 -1.94774270e-01 -7.12017834e-01 -2.56275117e-01 -5.97892284e-01 -2.36905724e-01 1.95326984e-01 -3.22369367e-01 1.73667455e+00 -1.03729475e+00 -1.76963592e+00 8.59643340e-01 -3.04730266e-01 -5.71521521e-01 7.90117145e-01 -2.84627259e-01 -4.59029526e-02 5.40888980e-02 -3.27469289e-01 9.54072416e-01 6.32424772e-01 -1.50543201e+00 -4.50538903e-01 -2.99194027e-02 2.64212519e-01 5.69469213e-01 -6.07299685e-01 3.45420092e-01 -5.15398502e-01 -1.26261219e-01 -1.41723692e-01 -7.53386915e-01 -4.43345845e-01 -4.42227840e-01 -1.84217140e-01 -9.46340203e-01 7.44014204e-01 -3.60901922e-01 1.41559494e+00 -1.60367095e+00 7.63436407e-02 2.03606397e-01 -1.04956858e-01 6.70860469e-01 -1.65320292e-01 4.83758718e-01 8.63095224e-02 2.84481198e-01 1.94541305e-01 -2.04894722e-01 -2.06784606e-01 2.20781207e-01 -2.59893350e-02 -9.56969559e-02 2.31643081e-01 5.69445252e-01 -1.30515265e+00 -6.06943607e-01 1.65046453e-01 1.89186916e-01 -5.34604132e-01 5.76534152e-01 -4.94361877e-01 -8.84780735e-02 -4.58701909e-01 4.08177137e-01 2.45183930e-01 -1.20452732e-01 4.21552241e-01 3.67751420e-01 5.74291825e-01 6.32026494e-01 -1.04889381e+00 7.45882630e-01 -3.21230650e-01 3.72216851e-01 1.16026187e-02 -9.02079105e-01 1.20050609e+00 5.75355411e-01 -6.14021197e-02 -7.95443296e-01 -1.80914938e-01 3.50044698e-01 1.44948989e-01 -5.69383621e-01 8.61721635e-01 -3.31909388e-01 1.09642625e-01 4.87784654e-01 9.71038043e-02 -4.97337967e-01 4.13735688e-01 3.18303525e-01 7.22556770e-01 -2.12739304e-01 2.91102201e-01 -1.59374207e-01 7.02647865e-01 -7.67265037e-02 3.43519598e-01 1.14004803e+00 -5.00265777e-01 2.10495442e-01 6.96939707e-01 -3.83798063e-01 -6.84629560e-01 -6.54554844e-01 -3.18257883e-03 1.56896198e+00 -3.31542119e-02 -6.55369222e-01 -8.16431403e-01 -1.13787150e+00 -3.22513491e-01 8.79561782e-01 -5.66218972e-01 -4.61082041e-01 -6.85345650e-01 -9.19116616e-01 3.39502692e-01 2.22672641e-01 2.82551527e-01 -1.30175567e+00 -7.23521635e-02 7.23363683e-02 -1.84317276e-01 -2.58225352e-01 -2.10098088e-01 5.23079932e-01 -8.61409903e-01 -9.20641243e-01 -7.86262512e-01 -7.37972796e-01 5.40478826e-01 2.75762171e-01 1.50020289e+00 8.02354813e-01 4.67794091e-02 5.25695801e-01 -4.50414747e-01 -6.85310304e-01 -9.50380623e-01 2.03943655e-01 -7.74345174e-02 -5.77846229e-01 6.74175501e-01 2.93942809e-01 -2.23040298e-01 2.10018739e-01 -6.85858667e-01 -1.75237566e-01 2.64477909e-01 1.46483052e+00 -2.34561786e-03 -4.64421719e-01 1.08832181e+00 -1.25333202e+00 1.31175256e+00 -3.40918005e-01 -2.02278510e-01 5.22209287e-01 -1.07605755e+00 -5.09406533e-03 5.09176493e-01 -7.44810939e-01 -1.14153278e+00 -5.00398986e-02 -2.01711223e-01 1.51998669e-01 -2.99196094e-02 2.19515145e-01 6.53162673e-02 -8.30480903e-02 1.05282402e+00 1.90549150e-01 7.19862431e-02 -2.85087526e-01 2.18906000e-01 1.00211012e+00 -1.49797872e-01 -6.76099479e-01 3.47175539e-01 -1.37434199e-01 -5.93411267e-01 -9.39922988e-01 -8.53931546e-01 -5.33539653e-01 -4.79987115e-01 -3.85912567e-01 4.04064089e-01 -3.96555930e-01 -2.47853488e-01 2.57852942e-01 -1.00436413e+00 -5.25098562e-01 -3.67067486e-01 1.98355123e-01 -6.68987215e-01 4.78598654e-01 -7.61384785e-01 -1.29498482e+00 -4.89429027e-01 -1.07489610e+00 9.13800776e-01 3.70319754e-01 -6.80677831e-01 -1.17095053e+00 1.99368045e-01 3.75638485e-01 4.81607467e-01 -4.14981872e-01 1.08597708e+00 -1.38095939e+00 9.53608826e-02 -4.16306704e-01 3.27643037e-01 4.25684273e-01 5.96841378e-03 -1.60257313e-02 -1.04282212e+00 -3.75035226e-01 1.16378009e-01 -1.01304460e+00 6.28031492e-01 6.03004508e-02 7.07453668e-01 -4.32078868e-01 -3.91178578e-01 -2.17555240e-01 8.77624869e-01 5.53104162e-01 6.47493601e-01 5.53293645e-01 4.24753455e-03 8.12505484e-01 1.27454543e+00 3.83034378e-01 9.11918655e-02 7.09690392e-01 1.40601799e-01 -2.08461910e-01 3.98532152e-01 9.74670500e-02 3.64573985e-01 6.66787624e-01 7.36588687e-02 -1.58660889e-01 -9.09379721e-01 4.98378694e-01 -1.73493350e+00 -1.06980276e+00 5.19070737e-02 2.41603947e+00 1.58852303e+00 5.38697958e-01 4.25525129e-01 7.17799887e-02 6.18598521e-01 2.05872446e-01 -2.08859697e-01 -7.80648947e-01 1.29846886e-01 1.11699834e-01 -7.19637647e-02 8.32894802e-01 -8.15883398e-01 9.83632565e-01 8.00882912e+00 9.84214723e-01 -9.94974792e-01 -2.15168111e-02 8.09416056e-01 -1.38114482e-01 -4.55271274e-01 -6.81079924e-03 -1.03459823e+00 1.97968423e-01 1.29251516e+00 -3.25545788e-01 2.22311318e-01 8.84598494e-01 7.26592094e-02 -1.32221326e-01 -9.98428464e-01 4.66545045e-01 -1.15960243e-03 -1.08873594e+00 2.94794917e-01 -3.31448406e-01 7.67443120e-01 -2.05887362e-01 -3.75422955e-01 9.29537654e-01 4.00882065e-01 -8.36845577e-01 3.72600287e-01 4.00969028e-01 2.37277001e-01 -8.81418824e-01 9.46282864e-01 7.62308896e-01 -4.16926861e-01 1.74555220e-02 -5.16272664e-01 5.71411252e-02 -6.22225851e-02 5.09830862e-02 -1.28362834e+00 1.58381075e-01 4.67791915e-01 1.38335332e-01 -8.46080065e-01 9.71152365e-01 6.74128113e-03 8.81840110e-01 -1.45513296e-01 -5.77637911e-01 2.66572654e-01 -1.18823625e-01 3.73423576e-01 1.38050330e+00 -4.42202806e-01 -4.07209098e-02 2.29140788e-01 3.36630046e-01 2.22139955e-01 3.20620209e-01 -7.33374536e-01 2.33018279e-01 7.78127849e-01 1.11644411e+00 -4.61246938e-01 -8.92639101e-01 -1.07391596e-01 5.29467762e-01 5.43113887e-01 2.18703777e-01 -3.24679255e-01 -4.18909788e-01 -7.35863298e-02 8.20691586e-02 -2.21890047e-01 3.39794338e-01 -1.13449171e-01 -5.13540685e-01 -2.02733666e-01 -1.40742970e+00 4.51274693e-01 -5.78275025e-01 -1.16252840e+00 5.87430894e-01 4.65858936e-01 -1.01021302e+00 -9.44922507e-01 -3.53023052e-01 -6.83087707e-01 8.97475243e-01 -1.21944737e+00 -4.93448138e-01 7.83499926e-02 2.44907051e-01 9.29119647e-01 -3.27714592e-01 1.23698223e+00 3.24724987e-02 -2.46619627e-01 6.26480401e-01 1.11383438e-01 -7.69734606e-02 8.89908314e-01 -1.62715793e+00 1.50315002e-01 -2.57320739e-02 -1.23152524e-01 9.94665980e-01 8.19738984e-01 -5.29094398e-01 -1.03694165e+00 -5.51700234e-01 1.00853503e+00 -3.53543818e-01 3.40626568e-01 -1.80617183e-01 -1.22338295e+00 4.90552366e-01 5.92359364e-01 -4.14565146e-01 7.13846087e-01 2.24451438e-01 -7.60694966e-02 1.14701144e-01 -1.34763801e+00 6.40672684e-01 3.77617925e-01 -1.98332056e-01 -8.55383337e-01 8.97205412e-01 5.84637225e-01 -4.83097374e-01 -7.44003713e-01 4.15580750e-01 3.70964706e-01 -1.01069343e+00 8.94155145e-01 -9.51583862e-01 7.92902559e-02 3.14717919e-01 2.38425553e-01 -1.46585798e+00 9.87522006e-02 -9.19180512e-01 -2.37670779e-01 1.18294549e+00 7.98051178e-01 -7.08309770e-01 8.58117044e-01 8.50158155e-01 1.55313700e-01 -1.10437787e+00 -8.71309698e-01 -6.34532332e-01 4.55521584e-01 -1.81065593e-02 2.53519803e-01 8.50057840e-01 2.97744036e-01 7.44484782e-01 -4.05940652e-01 -7.80568719e-01 1.85490057e-01 -2.76099503e-01 9.80583012e-01 -1.23306096e+00 -4.95575428e-01 -4.52469707e-01 1.08086802e-01 -1.20680439e+00 2.40715250e-01 -4.10268039e-01 2.04380497e-01 -1.38249087e+00 3.08973372e-01 -6.08618736e-01 -5.93305193e-02 3.47923785e-01 -7.02938318e-01 1.16480682e-02 -7.56979594e-03 1.31235927e-01 -8.21038723e-01 3.76716405e-01 1.10157776e+00 -1.78510338e-01 -4.87415493e-01 2.83106655e-01 -5.25312424e-01 7.96358705e-01 7.98640788e-01 -5.07409096e-01 -6.50070071e-01 1.04507856e-01 1.67364925e-01 3.23111653e-01 -2.63493896e-01 -4.11605477e-01 -5.85280359e-02 -4.72870916e-01 3.01694930e-01 -6.42689168e-01 5.08774102e-01 -3.97674382e-01 -5.34634948e-01 7.49739766e-01 -9.99572217e-01 2.01406345e-01 1.55428231e-01 4.88179892e-01 -1.60929635e-01 -9.97755945e-01 9.56653714e-01 -4.50842649e-01 -5.27548254e-01 -4.08606619e-01 -5.09858847e-01 3.97372305e-01 8.93255711e-01 -4.51915003e-02 -4.45788950e-01 -7.22931862e-01 -7.10820794e-01 4.59899694e-01 2.30297670e-01 5.56206465e-01 5.90725064e-01 -1.12037051e+00 -5.82285583e-01 2.94265971e-02 1.46524057e-01 -1.99214056e-01 -3.57298493e-01 7.53843963e-01 -3.07749093e-01 4.73840863e-01 3.25393528e-01 -6.42367780e-01 -1.70867455e+00 3.03964674e-01 5.64297497e-01 -7.85765350e-01 5.94657958e-02 8.80735993e-01 -2.74536818e-01 -1.00164735e+00 7.84150422e-01 2.18279094e-01 -6.90833271e-01 -3.99224274e-02 6.33042753e-01 3.88006598e-01 1.24527872e-01 -3.23259950e-01 -1.09496098e-02 3.04337610e-02 -5.12407541e-01 -3.31231922e-01 1.00340843e+00 1.15589373e-01 -9.57860947e-02 7.85873950e-01 8.98899376e-01 -2.34926969e-01 -6.71979487e-01 -3.01532656e-01 4.86987412e-01 -3.94528985e-01 -2.80178219e-01 -7.67290294e-01 -4.64807808e-01 7.90672004e-01 5.87191403e-01 8.60535920e-01 7.62511611e-01 -2.14798763e-01 2.71868944e-01 1.00596619e+00 4.33108419e-01 -1.58726799e+00 2.96824008e-01 8.06093633e-01 9.14755762e-01 -1.34855926e+00 1.39512852e-01 -1.73020750e-01 -8.91436458e-01 1.07867932e+00 1.18722892e+00 7.37159774e-02 4.84848440e-01 -4.75658253e-02 3.01525623e-01 -1.37395099e-01 -1.46218324e+00 8.05751681e-02 2.29837537e-01 6.78898990e-01 8.49204004e-01 -2.21912593e-01 -8.45854759e-01 1.36699975e-01 -5.62130436e-02 -2.16921315e-01 5.13093054e-01 1.04754031e+00 -7.17177153e-01 -1.35055292e+00 -2.71930844e-01 8.56790543e-01 -4.20187056e-01 -2.83811409e-02 -1.15407526e+00 1.20095861e+00 -5.50824046e-01 1.23202765e+00 -1.97029203e-01 -2.70839304e-01 4.74080324e-01 5.71006536e-01 3.75805795e-01 -1.04464936e+00 -1.25571883e+00 2.32624024e-01 7.28388071e-01 -8.74545127e-02 -4.67880756e-01 -4.01226699e-01 -1.14385188e+00 4.61778715e-02 -1.17929220e+00 7.51814544e-01 2.25070164e-01 9.01177049e-01 -1.06869943e-01 -4.29794304e-02 8.24607253e-01 -7.24789500e-01 -1.54142499e+00 -1.72636843e+00 -3.01584154e-01 3.71700794e-01 1.92906708e-01 -5.03570735e-01 -7.31776059e-01 -3.27407002e-01]
[12.724601745605469, 8.103229522705078]
621648d3-afc4-4f3a-bba2-686fbc9640bd
saliency-augmented-memory-completion-for
2212.13242
null
https://arxiv.org/abs/2212.13242v1
https://arxiv.org/pdf/2212.13242v1.pdf
Saliency-Augmented Memory Completion for Continual Learning
Continual Learning is considered a key step toward next-generation Artificial Intelligence. Among various methods, replay-based approaches that maintain and replay a small episodic memory of previous samples are one of the most successful strategies against catastrophic forgetting. However, since forgetting is inevitable given bounded memory and unbounded tasks, how to forget is a problem continual learning must address. Therefore, beyond simply avoiding catastrophic forgetting, an under-explored issue is how to reasonably forget while ensuring the merits of human memory, including 1. storage efficiency, 2. generalizability, and 3. some interpretability. To achieve these simultaneously, our paper proposes a new saliency-augmented memory completion framework for continual learning, inspired by recent discoveries in memory completion separation in cognitive neuroscience. Specifically, we innovatively propose to store the part of the image most important to the tasks in episodic memory by saliency map extraction and memory encoding. When learning new tasks, previous data from memory are inpainted by an adaptive data generation module, which is inspired by how humans complete episodic memory. The module's parameters are shared across all tasks and it can be jointly trained with a continual learning classifier as bilevel optimization. Extensive experiments on several continual learning and image classification benchmarks demonstrate the proposed method's effectiveness and efficiency.
['Liang Zhao', 'Yuyang Gao', 'Chen Ling', 'Guangji Bai']
2022-12-26
null
null
null
null
['bilevel-optimization']
['methodology']
[ 3.00376445e-01 2.10734636e-01 -1.31119087e-01 -1.58655107e-01 -4.64344054e-01 2.60966029e-02 5.12463629e-01 2.31155664e-01 -5.34181178e-01 1.14634788e+00 1.13527328e-01 1.54040396e-01 -3.45224261e-01 -8.25812697e-01 -1.08415663e+00 -7.36402571e-01 1.46521956e-01 2.02221036e-01 2.75754213e-01 -7.45813921e-02 6.59292877e-01 2.45218217e-01 -1.93916237e+00 2.73413211e-01 1.18534017e+00 9.96676743e-01 7.51226485e-01 2.83869177e-01 -3.12620074e-01 9.09722626e-01 -5.72447896e-01 -1.79571480e-01 -1.07299395e-01 -6.31887197e-01 -8.41866076e-01 2.73503780e-01 2.13836059e-01 -2.05263019e-01 -3.21849734e-01 9.05252993e-01 3.42756480e-01 3.24803203e-01 3.63625854e-01 -1.10639310e+00 -1.14418399e+00 5.96487701e-01 -4.91034031e-01 6.11613095e-01 1.92744106e-01 1.81974590e-01 4.44363296e-01 -1.29939902e+00 5.10569811e-01 1.02465546e+00 5.14438033e-01 6.72545254e-01 -9.58913267e-01 -4.73333985e-01 2.82611579e-01 5.62588394e-01 -1.39948285e+00 -6.05268240e-01 7.91859448e-01 -4.89605665e-02 8.35255742e-01 2.23139435e-01 9.62955773e-01 1.03324783e+00 6.07369363e-01 1.00041580e+00 1.03965664e+00 -4.89415884e-01 5.74868917e-01 2.64299065e-01 1.74952075e-01 7.59885669e-01 3.99661392e-01 3.20701301e-03 -1.06580210e+00 1.55740052e-01 5.73735118e-01 4.69158709e-01 -4.89087284e-01 -2.55028903e-01 -1.03283656e+00 5.79418182e-01 4.04130042e-01 2.31038466e-01 -5.44410706e-01 -2.64507551e-02 1.06030084e-01 4.48311150e-01 5.19899964e-01 3.93309921e-01 -2.00313441e-02 1.40851051e-01 -1.25340581e+00 1.79674223e-01 2.15439409e-01 1.00575900e+00 9.56460953e-01 2.37762690e-01 -2.15185910e-01 6.83153272e-01 -4.75775935e-02 2.70452738e-01 1.13797140e+00 -6.16786838e-01 8.33892226e-02 6.81016266e-01 6.17475547e-02 -9.96928453e-01 -3.33661377e-01 -7.24978268e-01 -9.60679650e-01 7.46593177e-02 -4.72730137e-02 2.35862732e-01 -8.86119664e-01 1.74248147e+00 1.54336333e-01 3.73469323e-01 -8.16298574e-02 9.45997655e-01 5.69458067e-01 5.80855072e-01 2.25403666e-01 -6.94959819e-01 1.06930041e+00 -1.13224590e+00 -8.33737433e-01 -5.75940967e-01 1.60275713e-01 -4.40838993e-01 1.28828466e+00 2.12283239e-01 -1.28211617e+00 -6.84388220e-01 -1.35183513e+00 -2.08545506e-01 -3.90600502e-01 -1.41163066e-01 5.20536482e-01 4.01045948e-01 -1.01085842e+00 6.72953784e-01 -7.79727519e-01 -2.31302053e-01 6.20180190e-01 1.18902333e-01 -6.42753989e-02 -4.01620660e-03 -1.21461618e+00 1.10077703e+00 7.51433909e-01 -7.33986497e-02 -1.06362200e+00 -7.04482377e-01 -5.51142335e-01 2.20479071e-01 4.05483663e-01 -8.12419713e-01 9.60775018e-01 -9.71059978e-01 -1.09626436e+00 6.82248890e-01 -3.54311824e-01 -8.88350666e-01 3.17761511e-01 -4.48466599e-01 -3.97592068e-01 1.16053179e-01 5.57348244e-02 7.06494689e-01 1.32287574e+00 -1.07629859e+00 -6.26168013e-01 -3.32618028e-01 -2.50580013e-01 3.50317538e-01 -7.85338223e-01 -6.56561255e-01 -1.01483300e-01 -8.40637147e-01 3.66410166e-01 -7.66515017e-01 1.14617422e-01 -4.37992588e-02 3.03608421e-02 -1.02141470e-01 9.43499207e-01 -6.61832631e-01 1.29149616e+00 -2.27083445e+00 2.54087389e-01 -3.43024075e-01 1.98933095e-01 2.72210956e-01 -1.56932458e-01 1.22348122e-01 1.05075397e-01 -1.50382563e-01 -3.83745164e-01 -4.50422734e-01 -2.22416267e-01 4.77968045e-02 -8.59392643e-01 2.26837918e-01 2.65252322e-01 1.19218135e+00 -1.00289130e+00 -4.74281579e-01 -1.02770753e-01 3.40020031e-01 -2.17100695e-01 1.70730799e-01 -3.28957617e-01 3.00058782e-01 -1.58350557e-01 6.42991722e-01 7.08674610e-01 -4.56587493e-01 -1.55190051e-01 4.64404374e-02 -1.53898954e-01 -2.97950697e-03 -9.90344286e-01 1.98989296e+00 -2.66291469e-01 4.44686174e-01 -4.63893563e-01 -9.76162851e-01 1.00195026e+00 1.01096466e-01 1.36602111e-02 -1.11237669e+00 -1.94730416e-01 3.00585121e-01 -4.04863268e-01 -4.09850031e-01 7.73243010e-01 -2.67344862e-01 1.67221829e-01 7.55002260e-01 2.23888353e-01 1.05000496e-01 -6.28861890e-04 2.35948920e-01 1.00117755e+00 2.11580172e-02 3.78558159e-01 -2.14610353e-01 4.12715822e-01 1.01072378e-01 7.01848328e-01 9.09105659e-01 -3.01132441e-01 7.61409521e-01 -6.70521846e-03 -8.43849540e-01 -9.46849763e-01 -1.18614089e+00 2.05172047e-01 1.06230986e+00 4.86802042e-01 -5.99252097e-02 -6.14057422e-01 -4.18232769e-01 -1.45651400e-01 9.42439973e-01 -7.35925555e-01 -8.50596726e-01 -6.35002017e-01 -8.61982048e-01 1.14625119e-01 3.44570518e-01 1.03226423e+00 -1.49073625e+00 -1.03979468e+00 3.34816664e-01 -3.32046986e-01 -4.82160896e-01 -5.43012261e-01 1.70344681e-01 -1.19336259e+00 -8.74837935e-01 -8.55485976e-01 -9.45245624e-01 6.29789293e-01 8.91991735e-01 1.04449105e+00 3.78806204e-01 -3.47301722e-01 3.66709232e-01 -2.31369257e-01 -4.06813174e-01 -5.57530001e-02 3.45803499e-01 2.17616305e-01 8.25588852e-02 2.27355212e-01 -7.99444437e-01 -7.87414551e-01 -8.98499861e-02 -1.16499317e+00 1.90954924e-01 9.97749209e-01 1.06735528e+00 9.05251086e-01 1.81629524e-01 1.16521645e+00 -8.24351609e-01 6.74718261e-01 -6.33141935e-01 -1.60046801e-01 5.20420790e-01 -8.65329623e-01 1.94530800e-01 5.89724600e-01 -6.36125803e-01 -1.29825151e+00 -6.82915896e-02 3.24041367e-01 -4.75968748e-01 1.69162065e-01 4.15491968e-01 -1.02183446e-01 -3.32349725e-02 6.71971560e-01 9.94980276e-01 1.26857966e-01 -2.11773664e-01 3.93797249e-01 1.98525995e-01 5.90999186e-01 -3.11026901e-01 4.87227410e-01 4.97075737e-01 -2.45777965e-01 -7.61407018e-01 -1.15630937e+00 -2.49058623e-02 -5.42175293e-01 -2.68997222e-01 4.58216906e-01 -8.31392527e-01 -3.90090793e-01 5.31750381e-01 -1.07455838e+00 1.06574148e-02 -6.13347590e-01 2.41642594e-01 -7.27858603e-01 2.05457583e-01 -4.02726084e-01 -7.41033852e-01 -5.34255326e-01 -6.73678756e-01 6.46019220e-01 5.81458688e-01 -1.65185943e-01 -7.49759734e-01 -3.80724780e-02 5.89876361e-02 7.33909309e-01 1.27834305e-02 9.55904901e-01 -2.25509644e-01 -9.24129546e-01 2.40231037e-01 -8.70818496e-02 1.40741602e-01 2.15898946e-01 -8.28295648e-01 -8.58278692e-01 -5.68062365e-01 5.24337471e-01 -4.75712299e-01 1.39249325e+00 5.32204285e-02 1.12662363e+00 -5.66350877e-01 -3.87718469e-01 3.91090661e-01 1.35280943e+00 2.02109277e-01 8.40755880e-01 5.86368740e-01 2.74522990e-01 4.57484126e-01 7.18102634e-01 5.45320988e-01 2.88619161e-01 9.69691351e-02 3.74644816e-01 3.46751869e-01 -4.05646831e-01 -4.77445751e-01 1.97699532e-01 1.01747739e+00 2.35565662e-01 2.15605110e-01 -6.44026041e-01 6.70545161e-01 -1.92078662e+00 -1.28525865e+00 4.27316606e-01 2.18123722e+00 1.08005595e+00 2.79973596e-01 -2.58651346e-01 2.29286149e-01 7.50579774e-01 2.03387946e-01 -1.07882750e+00 -5.90610169e-02 -5.05608678e-01 1.73006833e-01 3.32999676e-02 2.32788607e-01 -8.67763460e-01 9.55586135e-01 5.98492861e+00 8.86501908e-01 -1.14396966e+00 4.22100842e-01 6.90230787e-01 -4.17840183e-01 -3.34479094e-01 7.08118379e-02 -8.20855975e-01 6.25027239e-01 8.10545444e-01 -6.50748551e-01 6.37655556e-01 7.94186175e-01 -7.60142580e-02 -3.62857819e-01 -8.20924222e-01 9.58642960e-01 3.91785532e-01 -1.53668940e+00 4.38991368e-01 -3.89417589e-01 8.22790384e-01 -5.16301394e-01 6.16634905e-01 5.50558388e-01 -4.97606874e-01 -9.03260648e-01 9.08825278e-01 1.08967519e+00 3.99697244e-01 -8.64427626e-01 3.56692791e-01 6.73847735e-01 -8.56883407e-01 -4.09516901e-01 -7.84318864e-01 -8.19399804e-02 1.87191591e-01 9.39479053e-01 -5.95678806e-01 2.10407093e-01 7.03973055e-01 5.40995061e-01 -8.86422992e-01 1.19899261e+00 -1.36968732e-01 4.24741775e-01 1.54815450e-01 -7.60635640e-03 -9.96421427e-02 9.86323059e-02 5.04746497e-01 8.76577735e-01 5.89313507e-01 1.85621262e-01 -2.14486897e-01 9.92983162e-01 -1.31140918e-01 -1.43416282e-02 -5.96049190e-01 4.82148081e-02 7.94737875e-01 9.48095620e-01 -7.94194758e-01 -3.59640062e-01 -7.97814727e-02 1.28357053e+00 7.57977307e-01 2.34827176e-01 -8.73576224e-01 -4.24385250e-01 5.82413897e-02 2.39226714e-01 2.30464920e-01 -2.73080587e-01 -5.19275069e-01 -1.03868842e+00 2.56962091e-01 -5.85240960e-01 2.95450509e-01 -7.87528574e-01 -9.51853156e-01 5.73498547e-01 -3.74544352e-01 -1.10616553e+00 7.69101232e-02 1.56684797e-02 -6.18497968e-01 4.97087866e-01 -1.70449352e+00 -9.57880497e-01 -5.42956054e-01 6.09608710e-01 9.44084466e-01 -3.70763391e-01 7.50089228e-01 2.66159326e-01 -4.29480344e-01 5.50412595e-01 -1.22646563e-01 -6.03122115e-01 6.13102615e-01 -8.96452665e-01 1.21781081e-01 7.38297164e-01 1.72592565e-01 8.35771799e-01 6.64683163e-01 -9.00168955e-01 -1.61790621e+00 -1.38377213e+00 9.68175709e-01 -1.17705464e-01 2.80710697e-01 -1.90036207e-01 -1.50657427e+00 6.29165947e-01 1.35561809e-01 -1.69073403e-01 4.13323879e-01 -2.16855183e-01 -1.14617601e-01 -1.44575730e-01 -9.72619295e-01 6.45122766e-01 1.26448488e+00 -3.71569514e-01 -1.05885220e+00 3.15275639e-01 9.86209810e-01 -2.06302539e-01 -2.90669054e-01 2.26563111e-01 3.81296217e-01 -1.07300901e+00 7.93291032e-01 -2.76587516e-01 2.83949763e-01 -2.83989251e-01 1.11321710e-01 -1.30745888e+00 -4.47054982e-01 -5.48604488e-01 -6.20624959e-01 1.20766056e+00 9.80204716e-02 -5.28067768e-01 7.95904398e-01 3.97747576e-01 -2.26787031e-01 -6.05098009e-01 -1.07915413e+00 -9.87493396e-01 -3.01749706e-02 5.93966059e-02 6.71033621e-01 7.28409708e-01 -7.64162913e-02 3.58802259e-01 -7.10841775e-01 -4.35173959e-02 7.55680025e-01 3.54006886e-01 4.01500672e-01 -1.05174267e+00 -8.69579837e-02 -2.69698620e-01 -5.90724275e-02 -9.76554453e-01 1.79862857e-01 -8.58856201e-01 2.61537842e-02 -1.34127331e+00 4.58590448e-01 -2.17425689e-01 -5.55635273e-01 5.16735196e-01 -3.41333836e-01 6.86572939e-02 2.07461327e-01 6.60608172e-01 -9.51802969e-01 1.11095989e+00 1.33531046e+00 -3.38500261e-01 -3.46046835e-01 -3.30972016e-01 -7.86601663e-01 3.92316341e-01 8.12003076e-01 -4.98337656e-01 -6.59304917e-01 -5.36360502e-01 1.40411496e-01 3.13728750e-02 4.69030678e-01 -1.32253301e+00 8.51978660e-01 -1.87920049e-01 5.09780347e-01 -5.74884176e-01 3.39563310e-01 -3.76865476e-01 2.48166084e-01 6.50849760e-01 -3.49369854e-01 1.45191684e-01 1.38468161e-01 8.40560913e-01 -3.08441132e-01 -2.84659237e-01 8.50202024e-01 -4.67744708e-01 -1.09926271e+00 2.90330887e-01 -2.87469864e-01 1.96760371e-02 1.07844210e+00 -2.56979585e-01 -5.07480323e-01 -1.73203334e-01 -9.03967977e-01 7.78353736e-02 2.85480291e-01 5.99295676e-01 1.26513302e+00 -1.43671310e+00 -5.15255809e-01 3.03798795e-01 -7.39416480e-02 -1.85851812e-01 5.00736058e-01 7.62161434e-01 -4.43920232e-02 3.72915208e-01 -5.29865503e-01 -3.26403618e-01 -6.72483325e-01 9.91065621e-01 1.18749477e-02 3.13793421e-02 -7.24286139e-01 6.72820270e-01 7.69035667e-02 2.00871438e-01 2.30316207e-01 2.36103609e-01 -1.19397312e-01 1.86142266e-01 9.01826382e-01 5.32874584e-01 2.34379053e-01 -1.86217993e-01 -1.24155201e-01 1.29262164e-01 -3.85262311e-01 -6.04058430e-03 1.40565205e+00 -5.67299426e-01 -2.52324849e-01 7.86297917e-01 7.08998740e-01 -4.56247032e-01 -1.33182228e+00 -3.44795018e-01 2.63297200e-01 -4.24232036e-01 -1.13612764e-01 -6.17100656e-01 -9.39573407e-01 9.17372227e-01 7.16085613e-01 4.28022183e-02 1.32528532e+00 -2.95234561e-01 1.09972179e+00 4.81966704e-01 7.58777618e-01 -1.43920124e+00 6.77093208e-01 5.42738318e-01 1.23362923e+00 -1.20888877e+00 3.61938588e-03 5.36351316e-02 -4.77878511e-01 9.81052101e-01 8.01257432e-01 -1.29359886e-01 6.16156995e-01 -1.00899622e-01 -3.46150637e-01 -2.35857777e-02 -9.37437713e-01 -1.43597061e-02 2.48905532e-02 3.19856912e-01 1.27960620e-02 -2.19243914e-01 -6.09836161e-01 8.32612514e-01 -1.98423833e-01 2.44718209e-01 4.73784924e-01 1.24299431e+00 -1.08006978e+00 -6.07044756e-01 -3.16525638e-01 4.56287891e-01 -2.11585946e-02 -2.82879956e-02 -7.64191151e-02 4.31639522e-01 1.03997827e-01 7.38827884e-01 -1.13488756e-01 -2.69507229e-01 7.23885149e-02 3.24388862e-01 4.58853334e-01 -6.27982914e-01 -1.94965005e-01 -4.03891206e-01 -6.57616019e-01 -3.61792892e-01 -3.02859426e-01 -5.97518027e-01 -1.35470641e+00 -1.80023178e-01 -2.94016212e-01 1.59223616e-01 3.97836447e-01 1.00019479e+00 7.57174194e-01 4.55729395e-01 5.74481845e-01 -6.52210355e-01 -6.39752984e-01 -7.71283984e-01 -6.87927186e-01 3.67094010e-01 3.04904878e-01 -9.15238082e-01 -4.26354438e-01 2.17333063e-01]
[9.845317840576172, 3.40586256980896]
3a82e8b3-c8ce-4525-9281-ba4c12f4181e
gender-stereotyping-impact-in-facial
2210.05332
null
https://arxiv.org/abs/2210.05332v1
https://arxiv.org/pdf/2210.05332v1.pdf
Gender Stereotyping Impact in Facial Expression Recognition
Facial Expression Recognition (FER) uses images of faces to identify the emotional state of users, allowing for a closer interaction between humans and autonomous systems. Unfortunately, as the images naturally integrate some demographic information, such as apparent age, gender, and race of the subject, these systems are prone to demographic bias issues. In recent years, machine learning-based models have become the most popular approach to FER. These models require training on large datasets of facial expression images, and their generalization capabilities are strongly related to the characteristics of the dataset. In publicly available FER datasets, apparent gender representation is usually mostly balanced, but their representation in the individual label is not, embedding social stereotypes into the datasets and generating a potential for harm. Although this type of bias has been overlooked so far, it is important to understand the impact it may have in the context of FER. To do so, we use a popular FER dataset, FER+, to generate derivative datasets with different amounts of stereotypical bias by altering the gender proportions of certain labels. We then proceed to measure the discrepancy between the performance of the models trained on these datasets for the apparent gender groups. We observe a discrepancy in the recognition of certain emotions between genders of up to $29 \%$ under the worst bias conditions. Our results also suggest a safety range for stereotypical bias in a dataset that does not appear to produce stereotypical bias in the resulting model. Our findings support the need for a thorough bias analysis of public datasets in problems like FER, where a global balance of demographic representation can still hide other types of bias that harm certain demographic groups.
['Mikel Galar', 'Daniel Paternain', 'Iris Dominguez-Catena']
2022-10-11
null
null
null
null
['facial-expression-recognition']
['computer-vision']
[ 1.56352166e-02 2.73715585e-01 -7.22115785e-02 -9.82892156e-01 6.12070225e-02 -4.98445094e-01 6.69253290e-01 4.40299846e-02 -5.51301241e-01 5.64977944e-01 3.75043154e-01 -5.29951714e-02 1.71154156e-01 -7.59359837e-01 -4.99221325e-01 -5.96514106e-01 6.68105707e-02 3.01139086e-01 -3.57111841e-01 -4.54045027e-01 1.96635410e-01 4.66240823e-01 -1.87574029e+00 3.20646405e-01 6.17444277e-01 1.14380467e+00 -5.15489697e-01 1.43525973e-01 7.90346563e-02 6.86155975e-01 -7.96327829e-01 -7.72941470e-01 4.25584793e-01 -4.23407853e-01 -5.20031393e-01 -3.41534428e-02 8.69948149e-01 -2.96940565e-01 1.26245711e-03 1.03663123e+00 5.90682030e-01 -1.57346874e-01 9.07347977e-01 -1.54402494e+00 -3.62702072e-01 4.19597328e-01 -8.62457097e-01 -5.48230931e-02 4.76712376e-01 2.45955482e-01 8.80819499e-01 -6.12208903e-01 8.37263405e-01 1.71446466e+00 7.29284167e-01 7.62412488e-01 -1.55695641e+00 -1.39978826e+00 4.39234450e-02 -1.60522759e-01 -1.51719189e+00 -7.07691789e-01 6.58941627e-01 -7.36801147e-01 2.28214234e-01 3.08811218e-01 6.99045181e-01 1.36670935e+00 -2.88997144e-02 1.59680992e-01 1.71435523e+00 -2.06291318e-01 2.56520599e-01 6.96242213e-01 2.10012309e-02 5.23631096e-01 2.85888255e-01 9.11082551e-02 -5.92638791e-01 -6.87229753e-01 3.77241671e-01 -2.31714055e-01 -1.28641594e-02 -4.12210584e-01 -6.59325182e-01 1.04153776e+00 3.70585054e-01 1.66843221e-01 -2.59139687e-01 -1.24849685e-01 5.48551798e-01 2.73730636e-01 6.05849564e-01 7.73939729e-01 -2.97761977e-01 -2.11455114e-02 -8.10671031e-01 5.79087615e-01 7.60362327e-01 4.04532313e-01 9.46193039e-01 -2.68758461e-02 -1.46683753e-01 1.01245964e+00 2.33594719e-02 3.64000678e-01 4.42081928e-01 -1.10438955e+00 -7.28865787e-02 6.89779580e-01 1.03124984e-01 -1.51520181e+00 -3.42373759e-01 1.56964421e-01 -4.14186269e-01 5.29913306e-01 7.67184794e-01 -3.68062288e-01 -6.00714922e-01 2.36201811e+00 2.51376957e-01 -4.01704371e-01 -1.47661179e-01 8.63335609e-01 5.83874762e-01 1.31841794e-01 5.20608902e-01 6.75826566e-03 1.40381193e+00 -1.93447769e-02 -5.03226697e-01 -3.86851013e-01 7.01690018e-01 -6.07336521e-01 1.08772981e+00 1.10683076e-01 -7.21954644e-01 -1.93091735e-01 -8.41438890e-01 3.41872126e-01 -4.82497096e-01 -1.08902514e-01 9.41153049e-01 1.17917192e+00 -9.41368639e-01 4.29652125e-01 -1.36519894e-01 -6.91249847e-01 6.27767205e-01 4.69453365e-01 -5.62088192e-01 8.24524388e-02 -1.17725217e+00 9.88835633e-01 -3.75953391e-02 -1.86138928e-01 -3.89851987e-01 -7.42489100e-01 -8.63453209e-01 -1.81328088e-01 2.22512297e-02 -2.49352574e-01 9.09087658e-01 -1.98025155e+00 -1.01046491e+00 1.40967488e+00 -7.12069124e-02 -1.01057321e-01 6.07202172e-01 2.69833654e-01 -5.15799999e-01 -2.66673595e-01 2.09490255e-01 1.05354548e+00 8.96027267e-01 -1.37598419e+00 -3.23644727e-01 -7.83027709e-01 1.68808952e-01 5.49599715e-02 -4.18500274e-01 4.94331419e-01 3.12456280e-01 -4.06681925e-01 -4.33783889e-01 -1.17261076e+00 -4.12856154e-02 1.85906366e-01 1.12383794e-02 -2.42068425e-01 7.11065292e-01 -4.39040750e-01 1.04885674e+00 -2.34304523e+00 -2.59980798e-01 4.16123748e-01 1.74230710e-01 -5.23182936e-02 2.01039091e-02 1.44900441e-01 -4.26218569e-01 4.27979171e-01 8.08754861e-02 -1.76019087e-01 -8.70936215e-02 1.34409592e-01 -1.78113192e-01 6.64831638e-01 3.38353723e-01 3.80833060e-01 -5.91608882e-01 -5.24883807e-01 -2.82577753e-01 3.94838154e-01 -8.66644263e-01 9.69405696e-02 1.21663183e-01 3.32041174e-01 -6.72359765e-02 6.46622777e-01 6.71988249e-01 4.34929609e-01 3.22472841e-01 -1.69562832e-01 1.15650399e-02 -9.05178860e-03 -9.02405560e-01 9.01852071e-01 -1.62425086e-01 8.20503652e-01 3.17909658e-01 -6.92236364e-01 1.09882402e+00 1.10613368e-02 4.39693421e-01 -6.71401978e-01 4.53035325e-01 2.91195542e-01 5.04085720e-01 -3.68211597e-01 6.40807271e-01 -6.95952833e-01 -2.08006456e-01 4.11561280e-01 -2.03886911e-01 -9.99779999e-02 3.45411189e-02 -5.04681915e-02 7.32420087e-01 -2.90685833e-01 -2.20207796e-02 -5.92120469e-01 9.59121287e-02 -8.14964771e-02 7.58622050e-01 4.23410952e-01 -5.01749396e-01 5.05407810e-01 8.87943923e-01 -3.83431196e-01 -8.60556304e-01 -7.83172429e-01 -4.41059798e-01 1.23891652e+00 -1.19216733e-01 -1.96639225e-01 -6.83060706e-01 -6.20144367e-01 3.31213653e-01 7.33536541e-01 -1.05504882e+00 -6.28178596e-01 -1.51295274e-01 -9.52108204e-01 9.59371448e-01 2.58522689e-01 1.93588808e-01 -8.51384759e-01 -7.31855452e-01 -4.34821844e-01 6.00793883e-02 -9.68728304e-01 2.78834179e-02 -1.69701815e-01 -5.17792463e-01 -1.08489203e+00 -6.29242778e-01 -4.53111321e-01 8.88575196e-01 -7.64491782e-02 1.32269859e+00 2.38190100e-01 -2.55795777e-01 5.09799898e-01 -1.08042091e-01 -8.55707347e-01 -5.29295504e-01 -1.05384476e-01 3.11445683e-01 4.03390348e-01 7.26222813e-01 -4.09233600e-01 -5.92268705e-01 5.97515702e-01 -8.93543661e-01 -1.94088712e-01 1.91697881e-01 5.23002088e-01 -2.63198793e-01 -2.35115290e-01 6.32825732e-01 -1.15869439e+00 6.45444274e-01 -7.18917131e-01 -1.47112325e-01 -1.43068448e-01 -6.57336414e-01 -2.01352552e-01 3.03466052e-01 -6.50852501e-01 -1.13356066e+00 -2.26765364e-01 2.09370270e-01 -2.85780400e-01 -2.71532595e-01 2.70463228e-01 -7.47816786e-02 -6.62851483e-02 1.01875019e+00 -4.91873890e-01 4.59502041e-01 -1.71894968e-01 -1.03953019e-01 9.54592466e-01 6.01922236e-02 -8.27304006e-01 4.82422471e-01 6.24625027e-01 -1.77225098e-01 -7.64331639e-01 -6.35668933e-01 1.25169262e-01 -2.29664862e-01 -4.91076618e-01 6.73954785e-01 -1.01293421e+00 -6.76082492e-01 3.67149562e-01 -6.79790080e-01 -3.73598099e-01 -1.63377315e-01 2.88700968e-01 -3.09139103e-01 4.58889641e-03 -2.61032760e-01 -8.72967184e-01 8.43093619e-02 -1.16219664e+00 8.51804554e-01 1.03955649e-01 -1.02694464e+00 -7.39408731e-01 -7.90608376e-02 3.49584132e-01 4.22065884e-01 5.38395166e-01 1.02925849e+00 -6.85823619e-01 2.72084564e-01 -2.64436543e-01 -2.82796323e-01 2.64863431e-01 2.91060656e-01 3.69513094e-01 -1.25538528e+00 -1.72249570e-01 -1.97689459e-01 -7.87466526e-01 5.02561390e-01 3.20429616e-02 1.08329630e+00 -2.72972494e-01 -1.20665699e-01 2.37042755e-01 1.19654191e+00 2.01304406e-01 5.90034604e-01 1.63508937e-01 4.72169548e-01 1.27979994e+00 6.35299146e-01 4.96291280e-01 3.58114958e-01 6.04603946e-01 2.93318301e-01 -2.98645467e-01 3.27239364e-01 -1.90149501e-01 5.80468416e-01 -2.00287551e-01 -9.95767042e-02 2.67954201e-01 -9.13535833e-01 3.40735793e-01 -1.29799938e+00 -1.01603913e+00 1.67563975e-01 2.30505943e+00 8.28742445e-01 -1.20959371e-01 4.50092584e-01 2.69404929e-02 8.50935698e-01 3.43088776e-01 -4.40781295e-01 -9.74672854e-01 -2.53650606e-01 8.88204295e-03 5.03670454e-01 1.62413001e-01 -6.56755805e-01 6.26074255e-01 7.08724976e+00 5.15860319e-01 -1.60287678e+00 -3.28507036e-01 1.25714672e+00 -4.33291525e-01 -3.25987518e-01 -9.88744050e-02 -5.05903840e-01 5.83320916e-01 8.84119034e-01 -2.71812886e-01 3.83737952e-01 8.67844582e-01 2.91514099e-01 -3.28686357e-01 -1.25540817e+00 9.79681492e-01 2.33797416e-01 -6.19579375e-01 -1.14691503e-01 2.91892648e-01 4.78767872e-01 -3.36507082e-01 2.71540433e-01 4.20763344e-01 2.89964497e-01 -1.42913461e+00 8.59547436e-01 1.22998059e-01 1.06541991e+00 -7.32308626e-01 5.90554237e-01 -1.90754067e-02 -3.23917776e-01 -2.72522628e-01 -3.49517345e-01 -5.60784161e-01 -3.38730246e-01 3.55874270e-01 -6.90546215e-01 -2.56986499e-01 9.86284196e-01 3.11647952e-01 -8.51374447e-01 2.69922405e-01 3.61626089e-01 4.63236004e-01 -3.21983069e-01 1.06512234e-01 -1.52219296e-01 -2.35624388e-01 8.90340954e-02 1.01548493e+00 2.02588156e-01 -3.97819169e-02 -2.76444733e-01 7.36400843e-01 -2.27560565e-01 2.88926244e-01 -1.01659894e+00 -1.78802356e-01 3.95165622e-01 1.38441586e+00 -4.72634584e-01 -2.13585421e-01 -5.05912602e-01 5.43334007e-01 3.96060020e-01 2.50483096e-01 -6.65912151e-01 1.65568545e-01 1.13715613e+00 4.61691171e-01 -3.30184013e-01 2.85908133e-01 -3.76190037e-01 -9.32394505e-01 -2.07789138e-01 -1.39952850e+00 4.58809644e-01 -7.44215310e-01 -1.37628472e+00 2.66826153e-01 1.36754721e-01 -6.83126330e-01 -3.13377202e-01 -5.78916967e-01 -4.03448105e-01 7.69045711e-01 -9.57304239e-01 -9.84856188e-01 -3.17631185e-01 4.78996545e-01 -1.31311705e-02 -1.10162064e-01 7.85563111e-01 3.17429990e-01 -5.64111650e-01 8.41894627e-01 -4.51941729e-01 2.50237942e-01 1.35635555e+00 -9.76077497e-01 -3.84638876e-01 1.53457507e-01 -3.08814645e-01 7.46890128e-01 9.89859283e-01 -4.09144670e-01 -8.73888433e-01 -6.82250440e-01 7.39936650e-01 -5.22945106e-01 4.49106365e-01 -4.76698607e-01 -5.74331999e-01 7.16204941e-01 7.52406344e-02 -1.77334696e-01 1.15360522e+00 5.13052702e-01 -8.13643396e-01 -2.29852915e-01 -1.57046795e+00 9.38926101e-01 1.06912196e+00 -5.72222114e-01 -1.55768976e-01 -8.96503851e-02 3.86990756e-02 -5.29285744e-02 -8.92121315e-01 4.26097572e-01 9.78623211e-01 -1.42814684e+00 5.99301755e-01 -7.32132256e-01 6.77377880e-01 1.27631739e-01 -2.49107450e-01 -1.38615859e+00 -1.56427249e-01 -1.79136470e-01 6.34750247e-01 1.64327550e+00 4.48330462e-01 -7.75238514e-01 8.37713301e-01 1.47334695e+00 4.90407377e-01 -7.90901780e-01 -7.19481051e-01 -3.90414894e-01 4.32324082e-01 -2.64492184e-01 7.67099202e-01 1.26957524e+00 -8.11519995e-02 1.91189796e-01 -2.55110800e-01 -2.16104746e-01 2.07437664e-01 -1.23688683e-01 1.10168850e+00 -1.45837724e+00 1.40399411e-01 -5.70303440e-01 -6.76388085e-01 1.52462889e-02 6.63175344e-01 -6.90008402e-01 -2.52852917e-01 -5.69359601e-01 3.63984525e-01 -6.53949201e-01 2.32690573e-02 4.72804576e-01 -6.10068403e-02 4.51285988e-01 2.60685742e-01 -1.65361576e-02 -2.32128035e-02 4.04552400e-01 9.31526780e-01 -2.92487610e-02 -3.97595577e-02 -3.65564287e-01 -1.33055067e+00 1.02046943e+00 7.45159626e-01 -3.70723456e-01 -3.32105964e-01 -1.50989190e-01 3.35669726e-01 -4.19518232e-01 3.83639663e-01 -7.30736196e-01 -4.01692688e-01 -3.74568731e-01 6.33107364e-01 4.19572413e-01 5.11282265e-01 -9.69667971e-01 2.40833491e-01 3.42493951e-01 -4.49840993e-01 1.53683767e-01 2.09179446e-01 1.49537906e-01 -8.20756778e-02 -4.43124166e-03 9.39688206e-01 -2.11360037e-01 -4.17074561e-01 1.25295624e-01 -4.75389719e-01 2.77165264e-01 1.01386559e+00 -3.86268556e-01 -1.56769320e-01 -6.04167700e-01 -3.02393168e-01 -3.92510891e-02 1.14807343e+00 6.43901408e-01 -7.77713880e-02 -1.20458984e+00 -6.22741163e-01 2.44571850e-01 4.36120540e-01 -6.02427602e-01 2.11368054e-02 7.25812316e-01 -9.72266793e-02 -2.97994971e-01 -5.89810908e-01 -3.55442673e-01 -1.54600275e+00 2.68880069e-01 5.56599975e-01 2.96697319e-01 1.75068945e-01 6.38622820e-01 5.31246841e-01 -4.39171582e-01 -1.09510891e-01 2.69426554e-01 -3.03908497e-01 7.38215327e-01 2.73124546e-01 3.57586384e-01 -2.93514311e-01 -1.18705118e+00 -3.16487730e-01 2.89638340e-01 -3.02594956e-02 -2.06142619e-01 1.07956886e+00 -4.35985683e-04 -2.82005042e-01 4.72452074e-01 1.10690343e+00 3.59471083e-01 -9.68438625e-01 1.98531285e-01 -7.43387416e-02 -8.51468146e-01 -3.19981307e-01 -5.93117058e-01 -1.18991363e+00 5.86025238e-01 7.53820479e-01 1.16755910e-01 9.57071006e-01 -1.26270175e-01 2.01076806e-01 -7.66939968e-02 4.46671337e-01 -1.17822397e+00 -1.85651645e-01 1.47179201e-01 8.53388309e-01 -1.40226901e+00 4.83739115e-02 -4.24232662e-01 -8.01065624e-01 5.97775459e-01 9.70125794e-01 -9.80427396e-03 5.16863525e-01 6.00266941e-02 5.90431094e-01 -2.87280500e-01 -6.04362547e-01 2.66456362e-02 -1.43668935e-01 7.15497434e-01 7.51730442e-01 1.82511792e-01 -6.61181331e-01 5.79573452e-01 -7.14156985e-01 -7.72545263e-02 5.64022124e-01 7.51133442e-01 1.62659645e-01 -1.05967486e+00 -5.82199395e-01 7.05356002e-01 -8.10753465e-01 2.47012839e-01 -9.96394217e-01 8.37233841e-01 4.40552682e-01 8.02626729e-01 4.00066733e-01 -4.96531725e-01 3.27215612e-01 3.60191286e-01 3.70806962e-01 -5.30978262e-01 -8.37019563e-01 -6.59474015e-01 4.99484211e-01 -5.07173717e-01 -4.23786432e-01 -9.16649580e-01 -8.59985888e-01 -5.86518466e-01 1.25101414e-02 -7.94083551e-02 4.77948427e-01 6.08259499e-01 4.46308106e-01 -3.04266602e-01 6.65961981e-01 -9.51074421e-01 -4.21214342e-01 -8.40640008e-01 -8.35451365e-01 1.15157890e+00 1.16408179e-02 -9.78256047e-01 -5.92579067e-01 -2.13764727e-01]
[13.043142318725586, 1.3560817241668701]
8bf7aca5-dcd0-4b7f-83f8-eb0e1f29c232
uncertainty-inspired-open-set-learning-for
2304.03981
null
https://arxiv.org/abs/2304.03981v1
https://arxiv.org/pdf/2304.03981v1.pdf
Uncertainty-inspired Open Set Learning for Retinal Anomaly Identification
Failure to recognize samples from the classes unseen during training is a major limit of artificial intelligence (AI) in real-world implementation of retinal anomaly classification. To resolve this obstacle, we propose an uncertainty-inspired open-set (UIOS) model which was trained with fundus images of 9 common retinal conditions. Besides the probability of each category, UIOS also calculates an uncertainty score to express its confidence. Our UIOS model with thresholding strategy achieved an F1 score of 99.55%, 97.01% and 91.91% for the internal testing set, external testing set and non-typical testing set, respectively, compared to the F1 score of 92.20%, 80.69% and 64.74% by the standard AI model. Furthermore, UIOS correctly predicted high uncertainty scores, which prompted the need for a manual check, in the datasets of rare retinal diseases, low-quality fundus images, and non-fundus images. This work provides a robust method for real-world screening of retinal anomalies.
['Huazhu Fu', 'Haoyu Chen', 'Chi Pui Pang', 'Yong liu', 'Rick Siow Mong Goh', 'Daoqiang Zhang', 'Xinjian Chen', 'Changqing Zhang', 'Weifang Zhu', 'Mingzhi Zhang', 'Jianhong Lin', 'Junhong Chen', 'Zhiqun Wu', 'Guoyao Deng', 'Yiming Qian', 'Qingquan Meng', 'Yuanyuan Peng', 'Yi Zhou', 'Xinxing Xu', 'Ke Zou', 'Aidi Lin', 'Lianyu Wang', 'Tian Lin', 'Meng Wang']
2023-04-08
null
null
null
null
['anomaly-classification', 'open-set-learning']
['computer-vision', 'miscellaneous']
[ 4.75892350e-02 5.07820189e-01 1.09598130e-01 -4.63781029e-01 -4.38099623e-01 -2.11009055e-01 1.92122281e-01 4.07491699e-02 -2.45236084e-01 1.09126377e+00 -3.73281419e-01 -2.23830879e-01 -4.95749801e-01 -6.57990813e-01 -6.24164402e-01 -6.38297796e-01 2.86942143e-02 4.55630898e-01 3.36012721e-01 3.78329068e-01 4.86347705e-01 2.10007265e-01 -2.34630632e+00 5.20834982e-01 1.59030151e+00 1.29583979e+00 -3.03742468e-01 6.46303535e-01 -1.37649020e-02 6.29798293e-01 -7.20967829e-01 -4.95209038e-01 3.52011174e-01 -4.33871686e-01 -7.10042417e-01 1.17477730e-01 7.53893852e-01 -2.53263921e-01 2.52533346e-01 1.37074113e+00 3.21532696e-01 -1.06042050e-01 1.10055232e+00 -9.46374893e-01 -6.62512243e-01 1.22128516e-01 -5.88132560e-01 1.07600354e-01 1.01569511e-01 1.86753288e-01 5.30523002e-01 -2.87560850e-01 5.29000521e-01 6.87716424e-01 3.55490685e-01 4.35644418e-01 -7.61926651e-01 -3.08255821e-01 -1.51099876e-01 1.55254602e-01 -1.22069299e+00 -1.45991355e-01 -1.84772626e-01 -8.58185351e-01 7.56129801e-01 4.15663093e-01 8.97049010e-01 3.55836719e-01 4.71927345e-01 3.33136111e-01 1.41161442e+00 -7.39629865e-01 1.77097082e-01 3.56382459e-01 1.88464522e-01 8.63903284e-01 7.21879900e-01 4.31577176e-01 -5.27304187e-02 -3.35274875e-01 7.44188666e-01 -4.28923070e-01 -1.38121367e-01 1.47185400e-01 -5.77205300e-01 2.96006173e-01 3.03919405e-01 -5.29117957e-02 -3.22259933e-01 -5.89907944e-01 4.07734402e-02 1.54648751e-01 3.81275147e-01 5.46103954e-01 -3.14922750e-01 6.13194592e-02 -5.56097448e-01 -1.39917389e-01 6.84075832e-01 7.65114069e-01 4.55869317e-01 -1.52604416e-01 -3.52299511e-01 7.63877749e-01 5.13998985e-01 4.21983063e-01 7.29962647e-01 -9.61662889e-01 -2.88969576e-01 1.09217739e+00 1.94600046e-01 -6.32982969e-01 -2.85079002e-01 -5.57111621e-01 -6.91707611e-01 5.71317911e-01 6.15662575e-01 -4.64477055e-02 -1.58254015e+00 1.19296598e+00 1.50530636e-01 4.26430613e-01 2.32419908e-01 7.85167038e-01 7.87609339e-01 1.76253438e-01 -1.91382125e-01 -4.38218683e-01 1.12345529e+00 -4.82430756e-01 -4.98468608e-01 4.73782755e-02 5.84004462e-01 -7.14323580e-01 7.68291414e-01 9.16734397e-01 -8.78495991e-01 -3.40223283e-01 -8.71846557e-01 3.64740521e-01 9.41520464e-03 5.99515021e-01 5.25610268e-01 7.31818259e-01 -9.54407632e-01 3.35249603e-01 -5.55489779e-01 -2.33935446e-01 7.46582270e-01 4.11018223e-01 -3.89669985e-01 -5.70575222e-02 -5.80152154e-01 9.64378536e-01 3.03486407e-01 -1.65570676e-01 -4.07142848e-01 -5.02517045e-01 -4.91866261e-01 -2.14851290e-01 1.62621498e-01 -6.55268908e-01 1.04249525e+00 -9.18514252e-01 -1.31493998e+00 1.14931750e+00 -2.00785950e-01 -7.70559072e-01 4.78021502e-01 -2.33185440e-01 -6.95641041e-01 1.91767275e-01 6.66916370e-02 4.26410884e-01 7.55248129e-01 -7.82567799e-01 -8.82006288e-01 -5.64494312e-01 -3.74252915e-01 -3.27629238e-01 2.09397748e-01 9.88554731e-02 -6.30368069e-02 -1.66178778e-01 2.40305558e-01 -7.67806292e-01 -2.13498294e-01 -3.50678451e-02 -5.90667486e-01 -4.74490672e-01 -1.35126607e-02 -5.06519318e-01 1.29444134e+00 -2.09702945e+00 -4.85282779e-01 5.51568687e-01 4.21439141e-01 8.33250344e-01 2.76186734e-01 -3.92177492e-01 -2.01453656e-01 3.04270238e-01 -9.48684886e-02 3.94072682e-01 -4.83385950e-01 1.06291682e-01 6.59992248e-02 2.14552909e-01 6.75559163e-01 2.12789312e-01 -6.93954408e-01 -5.75951159e-01 2.43628979e-01 1.02789346e-02 -6.97079062e-01 1.04688488e-01 -2.57444352e-01 3.21070343e-01 -3.88287812e-01 9.91593957e-01 5.93453646e-01 -2.85823226e-01 -2.13769466e-01 -1.36775166e-01 -2.15287179e-01 -2.77990848e-01 -1.21731758e+00 8.98305297e-01 2.22472593e-01 5.23639143e-01 -9.77500737e-01 -6.84050500e-01 1.12286627e+00 2.24025831e-01 1.25576824e-01 -4.93051261e-01 3.04869682e-01 5.64933717e-01 6.79044485e-01 -1.05999446e+00 3.09361387e-02 7.23967105e-02 5.83160341e-01 -4.71490733e-02 2.42910370e-01 2.60598540e-01 1.82525009e-01 -1.81509808e-01 7.70632207e-01 1.59863919e-01 6.99352205e-01 -3.13488096e-02 5.68383873e-01 -7.11958334e-02 8.35671902e-01 7.34822750e-01 -6.09302282e-01 1.04570258e+00 7.38212168e-01 -6.36511564e-01 -7.25127459e-01 -8.95831048e-01 -9.96401131e-01 -3.02194394e-02 -3.57507206e-02 -1.87394410e-01 -7.02777624e-01 -5.64006686e-01 1.28949434e-01 3.77743661e-01 -5.43042362e-01 -1.58744246e-01 1.55834436e-01 -9.55115795e-01 2.61570096e-01 8.29780623e-02 5.01894236e-01 -4.90708470e-01 -3.56033325e-01 -8.52152780e-02 1.15852594e-01 -6.93411887e-01 3.38764310e-01 -5.74090123e-01 -7.53421009e-01 -1.74131846e+00 -4.51548606e-01 -4.02272999e-01 8.48988950e-01 -4.81804043e-01 1.01353562e+00 1.09007820e-01 -8.87223244e-01 -7.77984932e-02 -2.07517669e-01 -9.22905028e-01 -3.97793680e-01 -8.68459880e-01 3.11462075e-01 4.76812601e-01 8.01250160e-01 -2.06200883e-01 -7.10778952e-01 4.21196789e-01 -3.41514409e-01 -4.43340749e-01 8.26943994e-01 9.95869756e-01 1.08317149e+00 -4.59485166e-02 6.22065365e-01 -7.26644874e-01 1.99399561e-01 -4.82096583e-01 -9.06617105e-01 4.96828884e-01 -9.00251567e-01 -1.46561652e-01 1.23205267e-01 -3.00704658e-01 -8.92289519e-01 -6.39997572e-02 2.52103209e-01 -4.00942981e-01 -4.38000113e-01 3.26555341e-01 3.34456712e-01 -3.14563066e-01 1.23159838e+00 -1.99247718e-01 4.02850062e-01 -3.52563292e-01 -2.19454572e-01 1.20194721e+00 6.46628320e-01 -1.87029257e-01 1.39382437e-01 2.07167029e-01 1.16800591e-01 -7.26166785e-01 -9.26760018e-01 -3.47732544e-01 -3.00690502e-01 -2.55671740e-01 7.09489644e-01 -7.24217892e-01 -5.78007698e-01 6.36068523e-01 -8.15379918e-01 3.81061137e-01 -4.14463669e-01 1.05818343e+00 -3.22969854e-01 3.84002894e-01 1.13409802e-01 -1.19883513e+00 -1.74686283e-01 -1.11817813e+00 5.52559793e-01 7.63693213e-01 -6.68899715e-02 -4.26129639e-01 -1.39353275e-01 4.22731936e-01 -1.67689081e-02 4.66627359e-01 8.38067591e-01 -7.12292731e-01 -6.60793185e-01 -5.19950807e-01 -4.03624505e-01 9.25797820e-01 1.29920244e-01 8.95168960e-01 -1.15071642e+00 1.54298916e-01 -1.46389022e-01 -3.98346394e-01 8.79069030e-01 8.49149704e-01 1.25366604e+00 -2.10295040e-02 -2.76524991e-01 6.37802482e-01 1.46966982e+00 6.50746405e-01 1.16719103e+00 3.38672251e-01 -8.45835805e-02 4.47722733e-01 9.25158143e-01 5.34534216e-01 4.31372449e-02 1.45896837e-01 5.36514819e-01 1.75218940e-01 -8.42221081e-02 1.18835412e-01 -1.41609892e-01 6.79809377e-02 -5.31372786e-01 -2.78422683e-01 -9.71496701e-01 5.18476248e-01 -1.53594196e+00 -7.63515651e-01 -4.51717168e-01 2.57290053e+00 6.01014555e-01 2.26994097e-01 -4.74382974e-02 6.49063513e-02 7.52094328e-01 -8.31656039e-01 -6.51765823e-01 -4.75188613e-01 -2.43261978e-01 2.46056885e-01 1.90067708e-01 2.58541793e-01 -1.12771535e+00 5.32656074e-01 6.29577971e+00 4.95212823e-01 -8.25018466e-01 -4.07833219e-01 8.77597988e-01 3.15030105e-02 1.28739208e-01 1.96293090e-02 -7.89448202e-01 6.94426119e-01 1.03092420e+00 -2.00051486e-01 -9.04785469e-02 4.99328554e-01 -8.04316327e-02 -6.11661434e-01 -7.78661966e-01 8.76417816e-01 -1.18752485e-02 -1.13989913e+00 2.48951226e-01 1.20155802e-02 8.36715639e-01 7.68762231e-02 1.43926263e-01 -1.05881780e-01 -7.75502026e-02 -1.29714429e+00 -1.13216758e-01 1.29604411e+00 1.09508431e+00 -5.17618120e-01 1.18715799e+00 1.94586009e-01 -1.17083848e-01 -1.93220258e-01 -6.56751990e-01 -1.55683666e-01 -4.21967387e-01 7.60847628e-01 -1.03248084e+00 5.84473193e-01 9.38590288e-01 5.14950275e-01 -7.23882854e-01 1.95165837e+00 9.28334668e-02 6.92884982e-01 -5.70639551e-01 6.43430604e-03 1.96344629e-02 -4.61285591e-01 5.95974267e-01 4.86262441e-01 4.89430219e-01 2.10713163e-01 -4.40073162e-01 7.20547080e-01 4.06953067e-01 2.29021594e-01 -4.99306649e-01 -1.21853203e-01 3.29855621e-01 6.90648556e-01 -7.67834708e-02 -1.14386216e-01 -3.17995429e-01 2.40313768e-01 4.78123389e-02 1.72583371e-01 -5.03137529e-01 -6.33536756e-01 4.29172218e-01 2.60381605e-02 -3.16595793e-01 6.48178160e-01 -3.13176423e-01 -1.11260390e+00 2.45687738e-01 -6.36104882e-01 5.41963279e-01 -1.01462638e+00 -1.24524188e+00 8.35686445e-01 -2.32779592e-01 -1.81807518e+00 -2.55938411e-01 -1.01116419e+00 -5.70395350e-01 1.19445968e+00 -1.39851427e+00 -7.13115990e-01 -4.31716472e-01 1.98176190e-01 -1.29924551e-01 -9.88789916e-01 8.39487493e-01 -2.54838318e-02 -6.60186172e-01 6.56751394e-01 2.08149031e-01 -1.27210081e-01 1.01115930e+00 -1.25725079e+00 -3.86300415e-01 9.39056754e-01 -3.01687449e-01 4.56562400e-01 3.85283619e-01 -6.37813866e-01 -4.60573435e-01 -1.03009427e+00 7.73453295e-01 -4.87774670e-01 2.94178039e-01 8.37542057e-01 -9.76627111e-01 4.95569080e-01 -2.13471889e-01 5.18695176e-01 1.06099093e+00 -2.75589693e-02 -1.94842573e-02 -1.14113532e-01 -1.61513889e+00 3.20230693e-01 8.05555105e-01 -1.82052441e-02 -7.34013081e-01 5.19495308e-01 2.93992519e-01 -5.75454354e-01 -1.23518872e+00 8.55598629e-01 6.98542714e-01 -1.42458653e+00 6.92759752e-01 -1.00047195e+00 4.48309183e-01 -3.73924762e-01 -2.39880569e-02 -9.38683450e-01 -1.26156777e-01 -3.55534226e-01 3.11495531e-02 9.00674045e-01 6.45226896e-01 -1.17069161e+00 5.16280711e-01 6.84740424e-01 -2.66094655e-01 -9.95846748e-01 -9.06004548e-01 -4.59891737e-01 -2.52190083e-01 -2.22336575e-01 3.76992077e-01 5.44863105e-01 -3.39843452e-01 -4.76295739e-01 -1.01629935e-01 6.39392495e-01 7.43982196e-01 -5.00618592e-02 4.01942641e-01 -1.84231055e+00 -8.27101395e-02 -1.71844617e-01 -1.30917609e+00 7.65249040e-03 -4.05297756e-01 -5.67695081e-01 -5.42513490e-01 -1.37432742e+00 4.97352853e-02 -5.83048940e-01 -4.74851191e-01 4.05696303e-01 -1.94324419e-01 3.32818151e-01 -5.11213958e-01 3.50675166e-01 -9.62923765e-02 4.15989347e-02 1.42096531e+00 1.57162309e-01 -2.22003371e-01 5.67247510e-01 -8.17542970e-01 1.02640331e+00 9.05299604e-01 -1.87707543e-01 -5.59321880e-01 -7.43730962e-02 1.91055760e-01 -1.05303705e-01 4.34890151e-01 -1.11236358e+00 3.80293019e-02 -3.23203504e-01 5.29683113e-01 -4.60635453e-01 -2.88722161e-02 -2.84132808e-01 -6.45289570e-02 3.70600373e-01 3.52448635e-02 -7.37140000e-01 1.17005236e-01 6.54215753e-01 -4.44622487e-01 -4.00541037e-01 9.67822552e-01 -1.10313848e-01 -6.86751366e-01 1.46203861e-01 -5.22771999e-02 6.92307577e-02 1.32461739e+00 -6.45284235e-01 -6.96312070e-01 1.62247509e-01 -1.08686173e+00 1.52123079e-01 3.27206284e-01 1.06642529e-01 8.26482892e-01 -7.35711277e-01 -9.08080637e-01 6.58136964e-01 4.96107161e-01 2.76869625e-01 3.75374168e-01 1.01971173e+00 -7.12020099e-01 3.87937665e-01 -4.04061019e-01 -8.91105294e-01 -1.35421050e+00 5.42888464e-03 8.18482041e-01 4.04088259e-01 -3.21998715e-01 9.89787281e-01 -1.22566335e-01 8.24218057e-03 1.00034602e-01 -4.85490948e-01 -7.02693820e-01 -2.54602581e-01 6.71052456e-01 4.66367245e-01 3.15142214e-01 -2.63209432e-01 -1.54684499e-01 8.32095206e-01 -2.94601560e-01 4.54538852e-01 8.16225410e-01 2.42547885e-01 -4.60443169e-01 2.54547089e-01 4.85855699e-01 -1.55474216e-01 -6.78098381e-01 -7.18045533e-02 -9.18768868e-02 -7.67791092e-01 6.33948594e-02 -1.49214554e+00 -4.50056165e-01 5.62808573e-01 1.28152847e+00 1.94830060e-01 1.17116559e+00 -1.75159097e-01 -6.41990900e-02 4.55343574e-01 4.00959581e-01 -9.70799506e-01 -4.70203668e-01 5.56101836e-02 8.68032634e-01 -1.55015528e+00 1.37985293e-02 -5.06507099e-01 -6.49529278e-01 1.04889059e+00 1.03138459e+00 -8.85584131e-02 6.08481646e-01 -1.16833597e-01 2.43932620e-01 -3.38403769e-02 -8.81969571e-01 -3.10801744e-01 7.79676974e-01 8.27575088e-01 2.98380941e-01 1.79593533e-01 -6.84547305e-01 7.30023563e-01 -2.43360400e-02 5.53626299e-01 8.34808350e-01 6.07009172e-01 -6.97462440e-01 -8.32224131e-01 -1.56613395e-01 1.20635676e+00 -5.50998986e-01 1.53202444e-01 -4.04571086e-01 7.66022503e-01 5.21645844e-01 8.58760655e-01 4.15519446e-01 -4.55337077e-01 8.54004025e-02 1.91841796e-01 4.90770966e-01 -6.33959293e-01 -5.66533394e-02 -5.70204966e-02 3.30353260e-01 -5.36291480e-01 -4.90333438e-01 -4.56559896e-01 -1.06360590e+00 3.32255960e-01 -5.97440183e-01 2.12848127e-01 3.94489199e-01 7.36242175e-01 6.36503518e-01 2.66793996e-01 5.54722488e-01 1.42606094e-01 -6.62962556e-01 -1.04995334e+00 -7.03932464e-01 6.48844019e-02 4.72159803e-01 -8.69504213e-01 -6.04646683e-01 1.38118514e-04]
[15.785842895507812, -3.9077773094177246]
55ad182d-6bde-4b89-b614-46c552705c8f
jointformer-single-frame-lifting-transformer
2208.03704
null
https://arxiv.org/abs/2208.03704v1
https://arxiv.org/pdf/2208.03704v1.pdf
Jointformer: Single-Frame Lifting Transformer with Error Prediction and Refinement for 3D Human Pose Estimation
Monocular 3D human pose estimation technologies have the potential to greatly increase the availability of human movement data. The best-performing models for single-image 2D-3D lifting use graph convolutional networks (GCNs) that typically require some manual input to define the relationships between different body joints. We propose a novel transformer-based approach that uses the more generalised self-attention mechanism to learn these relationships within a sequence of tokens representing joints. We find that the use of intermediate supervision, as well as residual connections between the stacked encoders benefits performance. We also suggest that using error prediction as part of a multi-task learning framework improves performance by allowing the network to compensate for its confidence level. We perform extensive ablation studies to show that each of our contributions increases performance. Furthermore, we show that our approach outperforms the recent state of the art for single-frame 3D human pose estimation by a large margin. Our code and trained models are made publicly available on Github.
['Aljosa Smolic', 'Ciaran Simms', 'Matthew Moynihan', 'Koustav Ghosal', 'Richard Blythman', 'Sebastian Lutz']
2022-08-07
null
null
null
null
['monocular-3d-human-pose-estimation']
['computer-vision']
[-1.19610289e-02 2.49817282e-01 -1.81644157e-01 -3.54605049e-01 -6.16046846e-01 -3.00277680e-01 5.05784750e-01 -1.95785388e-01 -6.17919564e-01 5.42167306e-01 5.36869466e-01 -5.32102771e-02 1.79321989e-01 -3.27239960e-01 -1.08909297e+00 -1.40316889e-01 -3.26535016e-01 6.02352321e-01 4.72552717e-01 -4.07760382e-01 -7.03590885e-02 3.82546127e-01 -1.30519307e+00 1.27341092e-01 3.22638512e-01 8.97672892e-01 -3.56747545e-02 7.87346900e-01 4.66559023e-01 7.80691147e-01 -5.37132323e-01 -5.81520379e-01 4.35428321e-01 -1.94524959e-01 -9.20100987e-01 5.11072204e-02 6.78660691e-01 -6.78668857e-01 -3.57402176e-01 4.23529953e-01 7.79028058e-01 1.07448749e-01 2.72203475e-01 -1.26773083e+00 -2.66242087e-01 3.04902047e-01 -4.82791871e-01 4.70156074e-02 6.35555446e-01 2.65438437e-01 9.52173471e-01 -8.04822028e-01 7.90812969e-01 1.30584967e+00 1.00077772e+00 4.93972927e-01 -1.08360279e+00 -5.14349818e-01 2.48574793e-01 1.10785112e-01 -1.18842316e+00 -3.55172783e-01 7.01443911e-01 -2.79228181e-01 1.45595908e+00 -9.80964229e-02 1.03997850e+00 1.21569729e+00 3.08658630e-01 9.30099666e-01 6.26939774e-01 -4.84326810e-01 -2.70681947e-01 -5.25514364e-01 -2.53799409e-01 1.18995392e+00 2.41241828e-01 9.74942222e-02 -7.81050146e-01 -6.30941913e-02 9.72554922e-01 -1.40902877e-01 -6.24790229e-02 -9.74619389e-01 -1.29226720e+00 6.80769026e-01 8.32523406e-01 -2.65384410e-02 -1.32671148e-01 8.52891386e-01 6.19607151e-01 1.38306841e-01 5.20327926e-01 3.12421769e-01 -6.62171781e-01 -2.40583166e-01 -8.42842162e-01 6.17498279e-01 5.87289870e-01 8.77024770e-01 4.96886522e-01 -1.94995731e-01 -4.01928872e-02 6.21587455e-01 3.28191280e-01 3.93237732e-02 2.44309470e-01 -1.24991596e+00 5.20593643e-01 4.07282054e-01 2.03568563e-01 -8.10758114e-01 -5.56525648e-01 -2.94079781e-01 -1.79233477e-01 4.04605687e-01 5.77623606e-01 -2.79663384e-01 -1.22896183e+00 1.82959580e+00 4.32051390e-01 2.82609817e-02 -4.66503233e-01 1.03975677e+00 5.01176238e-01 -3.30716521e-02 1.67730391e-01 5.80194294e-01 1.29497302e+00 -1.11484087e+00 -4.74985272e-01 -5.41045427e-01 7.82246232e-01 -6.72057748e-01 9.17609096e-01 2.85854816e-01 -1.18230057e+00 -5.95805824e-01 -1.16442490e+00 -6.03787184e-01 -1.01103269e-01 3.59248549e-01 9.11556900e-01 4.62514013e-01 -1.05768943e+00 8.68020117e-01 -1.35910618e+00 -4.27108079e-01 3.61981720e-01 6.29967809e-01 -6.87387586e-01 5.82484975e-02 -1.30984056e+00 1.28144014e+00 5.46511896e-02 1.84462339e-01 -7.52111673e-01 -4.00300026e-01 -1.26700652e+00 -2.55336791e-01 5.20340919e-01 -1.16026378e+00 1.25813627e+00 -6.97072744e-01 -1.53007817e+00 9.72243309e-01 8.34888406e-03 -6.53336048e-01 9.10023630e-01 -9.80603695e-01 3.63538444e-01 2.90475011e-01 5.75962253e-02 1.16198933e+00 8.37686658e-01 -8.47426414e-01 -3.73878479e-01 -2.76977539e-01 3.21452856e-01 4.01143014e-01 6.77758679e-02 -8.51631388e-02 -7.38826573e-01 -5.55838704e-01 -7.71318972e-02 -1.39411080e+00 -2.94814408e-01 3.91112745e-01 -4.04975206e-01 -2.32159927e-01 4.74985540e-01 -8.43492150e-01 8.44474733e-01 -1.69039118e+00 4.39311594e-01 6.18346930e-02 1.43179819e-01 2.28265468e-02 1.06743768e-01 2.63979524e-01 -1.50865428e-02 -7.27001950e-02 2.99900807e-02 -6.50485575e-01 5.95537610e-02 2.48191565e-01 3.09756666e-01 5.30768394e-01 4.62603033e-01 1.12342501e+00 -7.14346826e-01 -4.33231741e-01 4.60399121e-01 7.94252813e-01 -8.77117217e-01 6.03438951e-02 -5.45845889e-02 3.54413271e-01 -1.32747233e-01 3.40607345e-01 1.40313625e-01 -3.62319559e-01 2.70988882e-01 -4.14243758e-01 2.48010680e-01 6.44184589e-01 -1.09141576e+00 2.45989847e+00 -4.29796308e-01 5.42884827e-01 -1.41896605e-01 -7.46984363e-01 5.12274861e-01 3.69113117e-01 5.95373690e-01 -1.65573761e-01 2.02140033e-01 7.09683895e-02 1.88607618e-01 -1.26607031e-01 4.80981320e-01 3.10990456e-02 -1.72686338e-01 2.30200797e-01 3.41287524e-01 -1.02911159e-01 9.06260312e-02 1.93886653e-01 1.26340151e+00 1.03139639e+00 2.35425159e-01 -4.50463630e-02 1.24616064e-01 1.57781690e-02 3.22947115e-01 5.27737021e-01 -2.90074855e-01 7.14014053e-01 3.74441504e-01 -5.54716289e-01 -1.30016136e+00 -1.07210374e+00 2.76471227e-01 1.05865097e+00 -1.24941528e-01 -5.77818990e-01 -6.38202965e-01 -6.20231092e-01 1.95128202e-01 1.51830003e-01 -7.61354446e-01 -1.70843497e-01 -9.62402403e-01 -3.03053617e-01 7.09531605e-01 9.55036819e-01 3.26570749e-01 -8.47360730e-01 -1.10175395e+00 2.29957640e-01 -2.18096688e-01 -1.23501194e+00 -4.50995386e-01 4.55777884e-01 -7.92495191e-01 -1.10357976e+00 -1.02160764e+00 -6.36631966e-01 4.94905263e-01 -7.93455765e-02 1.22347701e+00 1.30958185e-01 -4.24402803e-01 3.81586641e-01 -3.23998511e-01 -1.46318913e-01 3.77528891e-02 3.73066217e-01 1.16078062e-02 -6.56547606e-01 2.41283551e-01 -4.38666463e-01 -6.43279016e-01 1.99381948e-01 -2.93532312e-01 2.70121813e-01 4.67321366e-01 8.63061011e-01 2.33789697e-01 -6.60135031e-01 9.16608348e-02 -6.12337768e-01 4.40242261e-01 2.17716172e-02 -2.00961232e-01 -8.36506113e-02 -2.86780894e-01 4.12659317e-01 1.52693361e-01 -3.28289211e-01 -8.42000186e-01 4.89283562e-01 -2.73558289e-01 -6.46330655e-01 -7.88264573e-02 3.50401282e-01 2.24201083e-01 -1.38779894e-01 6.19045436e-01 -2.95616150e-01 2.03506082e-01 -5.11347890e-01 4.92657989e-01 1.05044603e-01 6.55944228e-01 -5.93287706e-01 6.41800880e-01 4.46926534e-01 1.03556871e-01 -2.69244075e-01 -8.31989348e-01 -2.99470484e-01 -1.01796591e+00 -2.21455738e-01 1.05972600e+00 -1.29997218e+00 -8.29446912e-01 2.52241790e-01 -1.15272558e+00 -6.43360615e-01 -6.21472746e-02 5.46020448e-01 -8.49749923e-01 4.31891203e-01 -9.28647459e-01 -6.13900244e-01 -2.83816814e-01 -1.15892804e+00 1.54152381e+00 -2.85172462e-01 -7.10805714e-01 -6.92130506e-01 3.39399017e-02 3.79514515e-01 1.48095876e-01 4.74324703e-01 5.32241404e-01 -3.15462261e-01 -3.33427072e-01 -3.75990987e-01 5.91325611e-02 2.48747349e-01 7.57889152e-02 -2.08167836e-01 -6.59203410e-01 -4.31787521e-01 -5.77432096e-01 -7.25581527e-01 9.50191557e-01 4.79450703e-01 9.07094717e-01 4.36906628e-02 -3.61638188e-01 5.34105837e-01 9.03968453e-01 -4.48187947e-01 6.18967831e-01 4.60041940e-01 1.02943540e+00 5.24203479e-01 4.89367008e-01 2.91223288e-01 5.56311548e-01 9.70288694e-01 4.43071067e-01 -1.88003585e-01 -3.41865689e-01 -3.89948308e-01 3.21720481e-01 4.84797865e-01 -4.65354949e-01 -4.08814549e-02 -8.77693892e-01 5.09667814e-01 -1.88154864e+00 -8.93828273e-01 2.62899965e-01 2.14112902e+00 7.62997389e-01 6.27100408e-01 4.48132992e-01 2.47008041e-01 4.74880487e-01 1.77295804e-01 -3.83777589e-01 -2.91270703e-01 4.16982025e-01 5.83489239e-01 7.64126122e-01 6.72968149e-01 -1.22512436e+00 1.04406691e+00 6.76476860e+00 2.98772424e-01 -9.81130898e-01 -1.31925456e-02 2.18283340e-01 -5.05668402e-01 1.60424277e-01 -3.73072848e-02 -7.18058288e-01 1.23415351e-01 7.65508175e-01 3.42495531e-01 1.74480647e-01 7.44435191e-01 8.67235810e-02 -1.01623997e-01 -1.31500673e+00 7.89098263e-01 1.83991924e-01 -9.81159329e-01 -1.97614610e-01 9.95346978e-02 4.32143599e-01 2.40579486e-01 -2.27982223e-01 1.14953015e-02 5.47974348e-01 -9.82061148e-01 9.59499002e-01 3.75975400e-01 6.98416829e-01 -7.39797175e-01 5.47654331e-01 1.21261060e-01 -1.31761992e+00 1.58147171e-01 -1.51437044e-01 -4.30529445e-01 3.44628274e-01 2.17050537e-01 -8.34601641e-01 6.01867139e-01 6.53542936e-01 8.66744041e-01 -6.64820433e-01 9.46892619e-01 -5.53263009e-01 2.83774644e-01 -6.25757158e-01 7.25199208e-02 1.70139581e-01 4.56477940e-01 3.29212099e-01 9.32830334e-01 7.76555240e-02 -2.96065629e-01 3.70154142e-01 4.60402817e-01 -1.55530190e-02 -3.03749114e-01 -4.68342334e-01 2.53071785e-01 4.60685566e-02 8.60762954e-01 -5.40149093e-01 -2.95935035e-01 -3.57232690e-01 1.30982828e+00 6.01296306e-01 -1.66721735e-02 -8.84354413e-01 -3.08108330e-01 7.99287915e-01 3.13369870e-01 5.04442215e-01 -6.85078323e-01 -1.66645631e-01 -1.12482750e+00 2.76768237e-01 -6.67856157e-01 3.08450699e-01 -9.12411332e-01 -9.62107360e-01 3.67422611e-01 1.96049735e-01 -1.10653448e+00 -7.42546141e-01 -7.47459173e-01 -6.07039966e-02 7.19608247e-01 -1.21668553e+00 -1.33800673e+00 -2.46523365e-01 4.63631153e-01 3.74629229e-01 4.37181979e-01 8.16670716e-01 2.95722574e-01 -1.90469265e-01 6.14239931e-01 -7.57970154e-01 4.57180649e-01 8.03787231e-01 -1.22812581e+00 9.63732839e-01 6.97896302e-01 2.77664363e-01 6.65470064e-01 7.59546936e-01 -7.28153884e-01 -1.01810694e+00 -6.33167624e-01 1.01604009e+00 -8.70003700e-01 3.27132285e-01 -6.54103160e-01 -4.63738263e-01 1.16918576e+00 3.39515135e-02 2.60948181e-01 4.06488448e-01 4.25100625e-01 -3.34183127e-01 2.14181229e-01 -7.43070602e-01 5.26865304e-01 1.53800750e+00 -4.62682515e-01 -8.47465575e-01 2.23292977e-01 6.50863469e-01 -8.84406269e-01 -9.38626230e-01 4.99768883e-01 9.56096768e-01 -8.67845953e-01 1.18742037e+00 -6.69108510e-01 6.01121366e-01 -2.07240939e-01 1.43581972e-01 -1.18959749e+00 -2.69559771e-01 -4.43217278e-01 -1.96778595e-01 4.04912204e-01 3.29399467e-01 -2.42207617e-01 1.16594136e+00 5.62738836e-01 -1.20716661e-01 -8.27204406e-01 -1.04270256e+00 -5.54983377e-01 3.18752229e-02 -4.73264128e-01 2.06792906e-01 4.61758018e-01 1.16044939e-01 4.43861693e-01 -8.85353804e-01 -1.81367788e-02 5.03743827e-01 -2.78239489e-01 1.16596866e+00 -1.08570504e+00 -4.77552891e-01 -6.24299571e-02 -8.10494065e-01 -1.36110282e+00 8.19277167e-02 -6.95413291e-01 1.00239530e-01 -1.60028553e+00 -1.10677153e-01 -1.47676736e-01 -1.29604682e-01 8.48339498e-01 -1.39497101e-01 5.46068549e-01 3.39691132e-01 9.46450140e-03 -6.63231373e-01 4.91785645e-01 1.22573698e+00 1.82795957e-01 2.84431074e-02 -1.89002812e-01 -2.91474462e-01 7.19617605e-01 5.86662173e-01 -4.59996760e-01 -3.01213592e-01 -7.35277593e-01 1.75911918e-01 -1.42627239e-01 8.44689131e-01 -1.26063025e+00 9.31935161e-02 3.61756444e-01 8.19086373e-01 -4.94803309e-01 7.32793152e-01 -5.77517867e-01 3.58595140e-02 7.79312491e-01 -4.39955711e-01 3.48000735e-01 1.11432835e-01 3.91360939e-01 6.89527839e-02 2.90951073e-01 5.34228384e-01 -5.04158974e-01 -5.92480958e-01 2.55354255e-01 -1.64923817e-01 6.06370438e-03 6.38047874e-01 -3.16496521e-01 2.00693876e-01 -5.36438346e-01 -9.75646973e-01 1.32396787e-01 4.86694723e-01 6.11370146e-01 3.49482447e-01 -1.33278561e+00 -4.75155503e-01 -5.00667328e-03 -4.86637503e-02 -3.32740806e-02 -1.35545805e-01 7.25282848e-01 -6.25305295e-01 5.22819221e-01 -4.88202572e-01 -6.72022164e-01 -1.28573549e+00 1.18784949e-01 4.81512219e-01 -3.86793345e-01 -8.97112191e-01 9.91469085e-01 -3.52742314e-01 -5.34262776e-01 3.06261688e-01 -4.63278532e-01 2.50566155e-01 -3.46923351e-01 1.05727650e-01 2.01796487e-01 1.47967011e-01 -6.83927119e-01 -6.51808143e-01 4.79571640e-01 -6.77172560e-04 -2.59733915e-01 1.29992139e+00 -1.30175054e-02 4.25861239e-01 2.78262794e-01 1.25862288e+00 -1.74339443e-01 -1.75910532e+00 6.23054756e-03 -4.15245555e-02 -4.94511306e-01 -1.36105061e-01 -8.58872950e-01 -8.65620494e-01 8.25028718e-01 4.96290028e-01 -4.36186641e-01 7.59950757e-01 -1.80393849e-02 8.32027078e-01 3.38956892e-01 5.88471949e-01 -1.10606265e+00 3.22825253e-01 4.19393986e-01 7.50462234e-01 -1.19816566e+00 4.08888459e-01 -5.01764834e-01 -4.77404892e-01 9.54522014e-01 6.82882547e-01 -6.26206279e-01 5.04878521e-01 3.41009438e-01 2.44582742e-02 -2.24465489e-01 -7.88524568e-01 -3.84080738e-01 4.47563648e-01 5.54324269e-01 9.02773023e-01 -1.79423630e-01 -2.68889815e-01 1.67037115e-01 -5.08862913e-01 2.23488882e-01 9.42375213e-02 1.21517313e+00 -2.43043512e-01 -1.29822290e+00 -1.53144166e-01 3.32637817e-01 -5.82099319e-01 -1.51636926e-02 -3.76342744e-01 1.09099150e+00 1.41985372e-01 5.61279297e-01 4.27973680e-02 -4.07302320e-01 5.09682178e-01 2.53203779e-01 1.06795490e+00 -6.29192948e-01 -6.19831145e-01 1.19047821e-01 4.37804610e-01 -1.03468418e+00 -6.11903548e-01 -6.37923837e-01 -1.17687631e+00 -1.94730997e-01 -2.98896790e-01 -3.26791465e-01 5.25824487e-01 1.06669259e+00 4.17181313e-01 6.02099657e-01 -1.41199917e-01 -1.42580545e+00 -4.92433041e-01 -9.63019311e-01 -1.29011944e-01 5.58387995e-01 2.12451324e-01 -1.22947860e+00 -7.81454984e-03 -2.54392102e-02]
[7.000821113586426, -0.9118217825889587]
38f4d85a-eb56-4532-b2d7-89727cf73b6f
closing-the-loop-testing-chatgpt-to-generate
2306.05115
null
https://arxiv.org/abs/2306.05115v1
https://arxiv.org/pdf/2306.05115v1.pdf
Closing the Loop: Testing ChatGPT to Generate Model Explanations to Improve Human Labelling of Sponsored Content on Social Media
Regulatory bodies worldwide are intensifying their efforts to ensure transparency in influencer marketing on social media through instruments like the Unfair Commercial Practices Directive (UCPD) in the European Union, or Section 5 of the Federal Trade Commission Act. Yet enforcing these obligations has proven to be highly problematic due to the sheer scale of the influencer market. The task of automatically detecting sponsored content aims to enable the monitoring and enforcement of such regulations at scale. Current research in this field primarily frames this problem as a machine learning task, focusing on developing models that achieve high classification performance in detecting ads. These machine learning tasks rely on human data annotation to provide ground truth information. However, agreement between annotators is often low, leading to inconsistent labels that hinder the reliability of models. To improve annotation accuracy and, thus, the detection of sponsored content, we propose using chatGPT to augment the annotation process with phrases identified as relevant features and brief explanations. Our experiments show that this approach consistently improves inter-annotator agreement and annotation accuracy. Additionally, our survey of user experience in the annotation task indicates that the explanations improve the annotators' confidence and streamline the process. Our proposed methods can ultimately lead to more transparency and alignment with regulatory requirements in sponsored content detection.
['Adriana Iamnitchi', 'Gerasimos Spanakis', 'Catalina Goanta', 'Stefan Huber', 'Thales Bertaglia']
2023-06-08
null
null
null
null
['marketing']
['miscellaneous']
[ 3.47881019e-01 5.77261925e-01 -4.96413499e-01 -5.89366198e-01 -9.82474029e-01 -8.50042045e-01 5.61216354e-01 4.50004429e-01 -4.00177598e-01 5.35314977e-01 4.91452843e-01 -4.44204569e-01 9.40856040e-02 -5.08097708e-01 -2.25442126e-01 -3.46070677e-01 4.13753927e-01 4.43527550e-01 2.07254916e-01 -1.40892729e-01 4.71490979e-01 1.36680260e-01 -1.16908276e+00 6.40768230e-01 1.01505792e+00 8.42014968e-01 -1.20640695e-01 2.58305758e-01 -4.01227236e-01 6.91422284e-01 -6.03267014e-01 -9.40366626e-01 3.10801685e-01 -3.72346461e-01 -8.65765452e-01 1.16434231e-01 3.31686825e-01 -1.76637799e-01 2.74538040e-01 1.27077711e+00 5.34038097e-02 -1.86821371e-01 5.01961350e-01 -1.02954769e+00 -8.02567601e-01 1.01930130e+00 -7.26155758e-01 4.32759821e-02 5.06615460e-01 -1.71895221e-01 1.57939327e+00 -5.46331465e-01 6.17417991e-01 1.05980527e+00 4.12553430e-01 2.79879957e-01 -1.26855266e+00 -8.06245208e-01 4.07454610e-01 -1.15225554e-01 -1.18105519e+00 -2.07089394e-01 5.25023878e-01 -7.61164188e-01 4.88522023e-01 4.32274848e-01 4.46545333e-01 8.62223387e-01 -1.11536123e-01 6.54302299e-01 1.09580946e+00 -5.00702560e-01 1.86270222e-01 7.40045905e-01 2.24623933e-01 3.54947358e-01 4.40608144e-01 -2.85894871e-01 -4.33958769e-01 -3.74468416e-01 3.70830864e-01 -3.07361573e-01 -1.73884109e-01 -6.70635328e-02 -6.58910692e-01 1.21299100e+00 2.68281013e-01 7.32988536e-01 -3.84013027e-01 -2.73364097e-01 5.16401589e-01 2.77005315e-01 8.26856673e-01 7.57756531e-01 -2.26439819e-01 -9.53684151e-02 -6.82023406e-01 4.20184106e-01 9.19128776e-01 5.77049136e-01 4.71305311e-01 -5.27352095e-01 -1.97954744e-01 9.76343215e-01 5.50862789e-01 3.31755996e-01 1.84526369e-01 -8.61705661e-01 5.70070207e-01 9.56879377e-01 6.17632389e-01 -1.28857565e+00 -1.87769920e-01 -5.95457792e-01 -6.91738203e-02 -6.58900067e-02 5.53947628e-01 -1.97517917e-01 -4.77590293e-01 1.43792844e+00 4.33002383e-01 -5.40912986e-01 -3.09307605e-01 1.09278893e+00 3.82570982e-01 4.89983261e-01 5.29926002e-01 -1.54184699e-01 1.45394933e+00 -4.98877019e-01 -1.06453049e+00 -3.79687287e-02 1.04653120e+00 -1.00420785e+00 1.09265673e+00 4.44352835e-01 -6.01809859e-01 -2.47659773e-01 -7.64380753e-01 2.55333751e-01 -2.07683951e-01 9.09642279e-02 9.04506683e-01 9.54063237e-01 -3.02140146e-01 2.84184039e-01 -2.49673933e-01 -2.52526581e-01 4.19331968e-01 2.83728093e-01 -2.47611403e-01 9.06612426e-02 -1.28816640e+00 8.05635929e-01 1.99189007e-01 2.04664026e-03 -1.83375970e-01 -3.95413309e-01 -5.76269329e-01 3.27604078e-02 5.21001995e-01 1.09914839e-01 1.49027967e+00 -1.40050209e+00 -1.05707562e+00 9.12113130e-01 3.08239013e-01 -4.80211794e-01 4.42258686e-01 -3.18477839e-01 -6.91703320e-01 6.61442569e-03 4.88337398e-01 5.43949604e-01 4.01412427e-01 -1.20096207e+00 -1.08831954e+00 -4.07319218e-01 2.77179986e-01 1.33984461e-01 -6.35321021e-01 3.72654468e-01 -1.65645584e-01 -4.12513584e-01 3.39428820e-02 -1.23790205e+00 -2.05001712e-01 -4.61791337e-01 -2.81845331e-01 -5.78858256e-01 7.54003763e-01 -7.62493491e-01 1.39081299e+00 -2.15573382e+00 -4.64958876e-01 5.05304337e-01 2.77201474e-01 2.61261880e-01 1.63218737e-01 3.53352904e-01 2.42868155e-01 4.96537030e-01 2.66744792e-01 4.95434441e-02 1.20101534e-01 1.54415965e-01 -3.56398970e-01 4.14626420e-01 -9.88829210e-02 5.69069684e-01 -8.57908070e-01 -4.11431462e-01 -1.92509070e-01 1.96283191e-01 -4.87435579e-01 -2.50118785e-02 -3.67252290e-01 6.05549455e-01 -7.41689086e-01 5.66018820e-01 2.62321144e-01 -3.83879900e-01 5.68773448e-01 -3.24443653e-02 -2.41999432e-01 5.81050396e-01 -9.58719850e-01 1.12097049e+00 -3.13311279e-01 4.88898337e-01 2.69582570e-01 -7.83182740e-01 9.84863877e-01 4.80847478e-01 6.51097894e-01 -7.25077152e-01 3.81130904e-01 3.85016412e-01 3.28332573e-01 -7.44725049e-01 5.78678787e-01 -9.80172306e-02 -2.86194623e-01 5.07532537e-01 -4.59235907e-01 7.81041905e-02 3.04712325e-01 3.61138046e-01 7.93817580e-01 -4.00764532e-02 3.12322736e-01 -6.56049401e-02 5.51480293e-01 3.06740969e-01 6.35239005e-01 6.19446576e-01 -2.32492194e-01 3.54078342e-03 4.90387291e-01 -2.62780458e-01 -1.01272321e+00 -1.85709372e-01 -1.37286246e-01 1.21000218e+00 -1.37623876e-01 -4.95099545e-01 -8.20328951e-01 -1.20244992e+00 3.21109556e-02 8.23935032e-01 -5.13631105e-01 3.02511513e-01 -2.63597220e-01 -5.40709794e-01 2.28093475e-01 2.28132606e-01 4.40570801e-01 -7.14522898e-01 -4.27368462e-01 3.49602193e-01 -3.26252520e-01 -1.24936080e+00 -4.70373929e-01 8.44067056e-03 -4.60822910e-01 -1.28238511e+00 -3.80486369e-01 -4.96222824e-01 7.05537498e-01 1.25313461e-01 7.16508687e-01 1.30555958e-01 9.03814584e-02 8.32325816e-02 -6.03391469e-01 -5.86274505e-01 -7.53226757e-01 1.81616172e-01 -2.36381054e-01 9.51622352e-02 1.00009203e+00 -2.30607823e-01 -4.67527300e-01 8.30730438e-01 -9.07887101e-01 -1.89557672e-01 5.84024489e-01 4.51977462e-01 2.25414589e-01 -1.31408721e-01 7.78045654e-01 -1.53576136e+00 9.57939923e-01 -4.13815469e-01 -8.09723139e-01 9.29417089e-02 -9.21302557e-01 -2.08670840e-01 1.84067070e-01 -2.67398238e-01 -1.13632119e+00 1.02608539e-01 -8.46490636e-02 3.91957611e-01 -2.61944622e-01 5.58688700e-01 8.85601938e-02 -2.04189435e-01 9.49529469e-01 -4.91511762e-01 -1.53920010e-01 -5.11484146e-01 2.26866618e-01 1.07283735e+00 6.25497997e-02 -2.84924954e-01 8.32324743e-01 1.61358163e-01 -5.83008289e-01 -4.89730835e-01 -1.44671345e+00 -9.13842380e-01 -3.69014263e-01 -5.27290583e-01 8.92761350e-01 -8.98736358e-01 -6.04184091e-01 -3.71947408e-01 -1.11036098e+00 2.08121911e-01 1.53959498e-01 6.18034184e-01 3.71489078e-02 2.77549654e-01 -4.89537120e-01 -9.77552712e-01 -2.79208213e-01 -1.00737309e+00 8.27588499e-01 -9.58162919e-02 -8.76182437e-01 -7.83206642e-01 1.32374525e-01 1.19183755e+00 3.67017627e-01 1.49436414e-01 7.56823421e-01 -1.13070738e+00 -3.10991764e-01 -7.21066415e-01 -1.33456483e-01 3.80533397e-01 2.81236451e-02 -8.92715752e-02 -8.22482049e-01 2.06963763e-01 -9.96802151e-02 -2.61264145e-01 3.79591525e-01 1.41023695e-01 7.85729766e-01 -6.40324950e-01 -4.25911486e-01 -5.17867804e-01 1.07295442e+00 4.10350680e-01 3.47206384e-01 3.63026798e-01 3.51040572e-01 1.17111290e+00 1.10182524e+00 3.13158751e-01 1.12210706e-01 9.99048412e-01 3.36125731e-01 2.46998062e-03 2.50286847e-01 -2.61148453e-01 2.82203585e-01 5.47712266e-01 -1.80992573e-01 -5.89155331e-02 -7.71714866e-01 3.64622265e-01 -1.95580757e+00 -8.79029751e-01 -5.23596585e-01 2.03421283e+00 7.80056179e-01 3.71859044e-01 2.50662327e-01 2.52942681e-01 8.95001590e-01 -2.22972199e-01 2.64818594e-03 -6.68481290e-01 2.94444799e-01 -3.25860560e-01 6.24811292e-01 5.13834596e-01 -8.70203674e-01 8.79694223e-01 5.85001230e+00 5.51658809e-01 -7.85696268e-01 3.23045284e-01 7.24727929e-01 1.24946341e-01 -4.93071616e-01 -1.05241805e-01 -9.86863017e-01 3.95416558e-01 7.17087090e-01 -1.37973621e-01 1.38288170e-01 1.01844954e+00 4.96145189e-01 -3.39833014e-02 -7.95797348e-01 5.13470888e-01 -2.06690785e-02 -1.19945514e+00 -2.63671458e-01 5.01616120e-01 8.21783781e-01 -3.12485397e-01 9.37621444e-02 3.49439770e-01 4.31119025e-01 -6.42513812e-01 7.52354443e-01 -2.04184830e-01 2.66800940e-01 -6.49769723e-01 1.02706420e+00 2.01551348e-01 -7.90107310e-01 -2.40006089e-01 -7.23443478e-02 -1.39142901e-01 3.33355397e-01 8.11717987e-01 -1.33021605e+00 3.29254046e-02 4.19900954e-01 3.17578882e-01 -2.81339735e-01 8.49480212e-01 -5.43155253e-01 7.37716615e-01 5.61084971e-02 -4.05862451e-01 5.11551738e-01 -1.51538491e-01 4.43770409e-01 1.06715751e+00 4.32886137e-03 1.07728481e-01 3.67378980e-01 7.14568138e-01 -2.50167161e-01 7.12981105e-01 -5.40372014e-01 -4.82669890e-01 2.30268463e-01 1.35827959e+00 -9.10459101e-01 -1.47075266e-01 -6.69602156e-01 4.91202205e-01 7.59405270e-02 9.52061862e-02 -9.27014768e-01 4.72954428e-03 4.56703186e-01 5.83070874e-01 1.05841517e-01 6.32864609e-02 -3.91809434e-01 -6.71701789e-01 -1.08656339e-01 -1.03836548e+00 4.30098951e-01 -2.49342516e-01 -1.14263725e+00 5.50891817e-01 -1.29856184e-01 -1.03116858e+00 3.08908778e-03 -4.82783407e-01 -1.95167020e-01 5.80263495e-01 -1.28876483e+00 -9.71873045e-01 -5.87868094e-02 1.02928586e-01 4.72659141e-01 1.49854153e-01 7.20091820e-01 5.71711659e-01 -2.52208620e-01 3.36039186e-01 -3.91782969e-01 1.41981885e-01 8.78165364e-01 -1.07167888e+00 -1.80031750e-02 4.72791553e-01 2.80159146e-01 6.58315539e-01 1.02203798e+00 -9.61267173e-01 -8.23892355e-01 -8.14468205e-01 1.26177824e+00 -3.20578992e-01 1.09304428e+00 -4.20963705e-01 -7.70683825e-01 5.20835459e-01 6.47222325e-02 -5.86627722e-01 1.21043372e+00 6.41059637e-01 -4.29852694e-01 1.69052109e-01 -1.05607557e+00 3.95284355e-01 7.33199358e-01 -4.56435084e-01 -4.95003194e-01 7.73447633e-01 4.94801491e-01 -8.59470889e-02 -8.44676495e-01 6.20879196e-02 6.25100374e-01 -5.59632957e-01 2.07226768e-01 -5.97102106e-01 3.77116889e-01 -2.16301262e-01 -7.66775981e-02 -9.54929352e-01 -2.32545897e-01 -3.49016368e-01 4.65709984e-01 1.42321718e+00 1.07584441e+00 -3.50725263e-01 8.12044680e-01 1.27099323e+00 8.54989290e-02 -4.20360893e-01 -5.14236271e-01 -5.00444412e-01 -4.76954401e-01 -5.92077672e-01 2.03265175e-01 1.17231405e+00 3.60410184e-01 5.65246880e-01 -3.97637337e-01 2.49575347e-01 2.09289312e-01 9.75553021e-02 8.44874084e-01 -1.43035579e+00 -3.59521061e-01 -1.78922296e-01 -2.67372012e-01 -7.26599872e-01 5.89690693e-02 -7.68461585e-01 -1.30716741e-01 -1.41386223e+00 1.84497476e-01 -6.46987379e-01 -4.54345010e-02 5.58655620e-01 -1.05526103e-02 3.88005048e-01 3.17316093e-02 4.40277696e-01 -7.84626961e-01 -4.44707423e-02 1.12826514e+00 -1.68948263e-01 -3.40639174e-01 2.12310165e-01 -1.17672992e+00 7.56742477e-01 6.89195573e-01 -7.61361837e-01 -1.65378541e-01 -1.33083627e-01 7.05212831e-01 -3.80765319e-01 -8.08172990e-06 -5.96194804e-01 3.97019833e-02 -2.39552364e-01 -1.37246391e-02 -1.99819088e-01 5.89472475e-03 -1.16686141e+00 1.95410952e-01 3.94065589e-01 -8.49963427e-01 -2.80126929e-01 -6.46061674e-02 6.60921276e-01 -4.37730014e-01 -4.66261744e-01 3.94148022e-01 -1.37663350e-01 -2.43283808e-01 -1.38324216e-01 -3.91879231e-01 -2.26775244e-01 1.07491016e+00 -3.64065617e-02 -2.81901389e-01 -5.64749599e-01 -8.33089590e-01 1.53867662e-01 2.67629176e-01 5.18730938e-01 -7.87265152e-02 -1.07400477e+00 -7.10434496e-01 -3.37972224e-01 2.56738901e-01 -5.87703288e-01 -6.70628771e-02 9.76585448e-01 -1.12191707e-01 5.50365448e-01 1.95671573e-01 -4.43230450e-01 -1.64409196e+00 3.29757750e-01 3.48538533e-02 -4.61736321e-01 -3.03583831e-01 5.31239748e-01 7.58193657e-02 -3.99896242e-02 4.13535208e-01 1.98358409e-02 -7.19248533e-01 2.61266977e-01 6.51821434e-01 1.72646597e-01 -3.67295481e-02 -6.49806023e-01 -2.56607104e-02 2.30065305e-02 -5.14925182e-01 -8.00196975e-02 1.28991652e+00 -1.44205809e-01 -2.01712474e-02 1.04543425e-01 8.66913259e-01 4.54968333e-01 -7.43195295e-01 -1.16297729e-01 4.82425809e-01 -7.31047153e-01 1.19476080e-01 -1.10725701e+00 -8.47361267e-01 3.30682814e-01 4.01467979e-01 8.89232099e-01 5.61590850e-01 1.30347898e-02 6.06552958e-01 2.26965964e-01 2.92355150e-01 -1.37382174e+00 -1.03679359e-01 -2.32811347e-02 7.48812735e-01 -1.34503067e+00 5.51215257e-04 -9.18056071e-01 -8.13208044e-01 8.60667109e-01 4.27789152e-01 4.14232373e-01 2.78210431e-01 6.91124052e-02 2.91822940e-01 -4.58669841e-01 -4.37651575e-01 -4.29128148e-02 4.56015706e-01 2.50843644e-01 9.59432840e-01 1.48690715e-01 -1.19062054e+00 6.15844548e-01 -5.34937456e-02 -9.96086001e-02 2.70341694e-01 6.95403397e-01 -7.03362226e-01 -1.47258103e+00 -4.08819884e-01 5.68067133e-01 -9.89975274e-01 3.17344852e-02 -9.09318507e-01 6.91309631e-01 2.30255142e-01 1.22797358e+00 -3.33248705e-01 -2.14897990e-01 3.28315765e-01 5.69640798e-03 -9.78640392e-02 -9.13628101e-01 -8.78149331e-01 3.84235054e-01 8.79027605e-01 -3.87969822e-01 -7.10087478e-01 -5.92182398e-01 -9.74473178e-01 -1.41667277e-01 -8.47123861e-01 8.43050480e-01 9.78225887e-01 1.05736482e+00 2.19954506e-01 2.08068579e-01 5.74394524e-01 -9.84814167e-02 -7.14389622e-01 -1.02302217e+00 -5.89659214e-01 8.16134334e-01 -3.42144310e-01 -5.64773619e-01 -3.52144450e-01 1.32446736e-01]
[10.025749206542969, 6.406435966491699]
bf368426-7c13-455e-ad2b-27182dd961d0
distributionally-robust-learning-for-2
null
null
https://openreview.net/forum?id=qRdED5QjM9e
https://openreview.net/pdf?id=qRdED5QjM9e
Distributionally Robust Learning for Unsupervised Domain Adaptation
We propose a distributionally robust learning (DRL) method for unsupervised domain adaptation (UDA) that scales to modern computer-vision benchmarks. DRL can be naturally formulated as a competitive two-player game between a predictor and an adversary that is allowed to corrupt the labels, subject to certain constraints, and reduces to incorporating a density ratio between the source and target domains (under the standard log loss). This formulation motivates the use of two neural networks that are jointly trained --- a discriminative network between the source and target domains for density-ratio estimation, in addition to the standard classification network. The use of a density ratio in DRL prevents the model from being overconfident on target inputs far away from the source domain. Thus, DRL provides conservative confidence estimation in the target domain, even when the target labels are not available. This conservatism motivates the use of DRL in self-training for sample selection, and we term the approach distributionally robust self-training (DRST). In our experiments, DRST generates more calibrated probabilities and achieves state-of-the-art self-training accuracy on benchmark datasets. We demonstrate that DRST captures shape features more effectively, and reduces the extent of distributional shift during self-training.
['Anima Anandkumar', 'Yisong Yue', 'Zhiding Yu', 'Anqi Liu', 'Haoxuan Wang']
2020-09-28
null
null
null
null
['density-ratio-estimation']
['methodology']
[ 3.83498847e-01 2.79086471e-01 -3.95853281e-01 -4.31265146e-01 -1.10464907e+00 -8.22952747e-01 6.69112921e-01 -3.06242588e-03 -7.62072921e-01 9.28990483e-01 -1.61881790e-01 -3.26743364e-01 6.92796484e-02 -7.61520505e-01 -1.02794135e+00 -9.50082481e-01 1.98383987e-01 8.22015882e-01 2.89936364e-01 2.41117164e-01 -2.16308057e-01 4.76683468e-01 -1.22372234e+00 -1.52265772e-01 9.38524246e-01 1.01608968e+00 -3.45395841e-02 5.72941482e-01 1.31576687e-01 5.97377062e-01 -7.24909961e-01 -6.31767213e-01 4.63190705e-01 -4.15901244e-01 -6.96808815e-01 -1.82575751e-02 5.50424039e-01 -4.04438525e-01 -1.40454963e-01 1.38065553e+00 5.76399922e-01 1.52962819e-01 1.33420980e+00 -1.34126937e+00 -8.23663294e-01 5.94386280e-01 -7.07499862e-01 7.24145398e-02 -7.70148039e-02 8.32551271e-02 8.81093085e-01 -6.21566772e-01 4.96768802e-01 1.23745513e+00 6.96849406e-01 8.96321714e-01 -1.57682991e+00 -8.43811154e-01 1.68958046e-02 -1.77472696e-01 -1.38343477e+00 -4.66399759e-01 8.38464379e-01 -5.46309590e-01 4.09478664e-01 -8.82940367e-03 -6.72618151e-02 1.55532503e+00 -4.08015633e-03 9.38351810e-01 1.23075378e+00 -5.15530825e-01 7.62302995e-01 5.56241572e-01 4.42089848e-02 2.10828409e-01 3.55577677e-01 4.95177686e-01 -1.64077550e-01 -5.94220579e-01 8.31926286e-01 -3.88732165e-01 6.41412884e-02 -9.39498663e-01 -8.19929004e-01 1.07526958e+00 1.06054343e-01 -1.30339777e-02 -1.47669718e-01 -1.52511686e-01 6.11683965e-01 3.51308078e-01 7.06011415e-01 4.13497120e-01 -3.85988593e-01 1.24411888e-01 -8.21832240e-01 1.30709618e-01 7.37316310e-01 9.83910978e-01 6.37486160e-01 1.79896444e-01 -1.83180690e-01 1.02168047e+00 8.78500119e-02 7.97720134e-01 4.71756399e-01 -1.05933452e+00 2.35114962e-01 1.00244381e-01 1.01426519e-01 -5.58786213e-01 -1.22276368e-02 -4.37434673e-01 -1.00219131e+00 4.83944982e-01 7.91723132e-01 -3.89797628e-01 -8.61510396e-01 2.24965978e+00 4.50248301e-01 3.02393407e-01 2.82234818e-01 7.57213175e-01 1.89422905e-01 4.16255981e-01 1.91093057e-01 -2.43573382e-01 8.92817616e-01 -4.86693501e-01 -2.96507835e-01 -3.58492762e-01 5.94474137e-01 -2.88426697e-01 1.21336460e+00 3.79434526e-01 -8.20435524e-01 -5.54324985e-01 -1.09248340e+00 3.09844613e-01 -2.88067788e-01 -7.35369474e-02 1.44025594e-01 8.23253810e-01 -8.72045219e-01 6.05627120e-01 -6.10792994e-01 -2.74977505e-01 8.09823275e-01 3.77534807e-01 -5.24833322e-01 -1.23961054e-01 -1.25859940e+00 8.48492742e-01 4.08995867e-01 -5.78045726e-01 -1.05314577e+00 -7.37425566e-01 -1.09692800e+00 -2.81176239e-01 2.62390196e-01 -5.58970273e-01 1.35703957e+00 -1.34976053e+00 -1.61900485e+00 1.26029038e+00 -2.81627662e-02 -9.09542024e-01 8.51043284e-01 -1.11669108e-01 -2.44297087e-01 -3.42482254e-02 4.00303751e-01 5.78855813e-01 1.28045225e+00 -1.38243258e+00 -6.53093219e-01 -3.98004442e-01 -1.98145896e-01 1.23328485e-01 -3.26693863e-01 -9.76840630e-02 -1.91108435e-01 -7.72342324e-01 -3.86176109e-01 -9.24502015e-01 -1.38347208e-01 8.69671702e-02 -5.17335236e-01 -3.95167977e-01 8.99592638e-01 -2.23134354e-01 8.28218281e-01 -2.37298441e+00 -8.96608829e-02 3.90178144e-01 1.51381880e-01 4.48421389e-01 -2.71250337e-01 -2.29789123e-01 -2.79684722e-01 -1.55912163e-02 -5.74162185e-01 -6.16980493e-01 8.27224702e-02 4.51262087e-01 -6.91397667e-01 8.43814015e-01 2.17335135e-01 6.19843543e-01 -1.01846170e+00 -5.09981573e-01 2.48004422e-02 2.00487643e-01 -4.57180411e-01 4.02614504e-01 -3.10135275e-01 4.57555890e-01 -2.80001909e-01 3.05846304e-01 9.06203389e-01 -4.33122739e-02 -3.81313488e-02 3.14400375e-01 4.00202572e-01 1.21841401e-01 -1.20953417e+00 1.22524309e+00 -3.36658508e-01 3.65153998e-01 2.50275973e-02 -1.11070907e+00 9.88733053e-01 2.90490631e-02 2.90954441e-01 -5.06219566e-01 2.72779733e-01 2.92657405e-01 -2.19698057e-01 -4.98551801e-02 -1.30213471e-02 -3.72610301e-01 -2.14902237e-01 4.69431311e-01 3.09423983e-01 -1.20006666e-01 -9.83353555e-02 1.33483052e-01 9.03218389e-01 8.56041983e-02 5.02798855e-01 -3.66750717e-01 3.13999176e-01 -1.35251313e-01 8.32417667e-01 1.09754944e+00 -5.55991709e-01 5.38397610e-01 4.67011929e-01 -6.09803312e-02 -1.24261248e+00 -1.52658463e+00 -3.04464072e-01 1.33683729e+00 5.71640059e-02 2.28193954e-01 -7.84815490e-01 -1.30647302e+00 3.76380861e-01 1.11962402e+00 -7.88819671e-01 -5.48841715e-01 -3.53257239e-01 -4.11839098e-01 8.38101149e-01 5.91741741e-01 4.39061046e-01 -8.89260173e-01 -2.24567235e-01 -1.30751595e-01 2.28135616e-01 -9.67612267e-01 -6.80642068e-01 6.74196601e-01 -6.18482947e-01 -9.01560068e-01 -8.87523770e-01 -9.35678065e-01 6.79346144e-01 -1.50780424e-01 1.07876706e+00 -7.70884216e-01 1.78845346e-01 3.96471173e-01 -7.02453628e-02 -5.43830991e-01 -8.05236042e-01 -4.68167104e-02 5.74870765e-01 1.07236713e-01 4.97835428e-01 -6.72285378e-01 -9.13405418e-02 3.97451669e-01 -7.94982314e-01 -4.83931601e-01 3.98572415e-01 1.02599037e+00 8.06747496e-01 1.69188529e-01 6.70785546e-01 -1.45166671e+00 5.04178524e-01 -6.64589405e-01 -7.44133890e-01 1.31113023e-01 -4.86687422e-01 2.53782868e-01 9.57751751e-01 -1.12158895e+00 -9.71951723e-01 1.58545181e-01 -5.57781197e-04 -9.03606415e-01 -4.58997399e-01 -2.84953369e-03 -5.27196586e-01 -4.07717191e-02 1.08115900e+00 3.02751064e-01 1.31996050e-01 -2.87824869e-01 4.43171352e-01 8.73327315e-01 8.80003572e-01 -9.29806530e-01 1.08431816e+00 4.15756345e-01 -1.93086252e-01 -6.60002172e-01 -1.05901742e+00 -3.97175580e-01 -6.62885666e-01 1.62584558e-01 4.92409825e-01 -9.04543161e-01 -2.99655795e-01 6.33952737e-01 -8.93464148e-01 -6.34561837e-01 -9.75713611e-01 3.71879458e-01 -8.15592945e-01 3.00222903e-01 -2.57538557e-01 -1.04026127e+00 -8.01226497e-02 -8.89209926e-01 8.24529827e-01 1.29424661e-01 -2.78471619e-01 -1.23832476e+00 2.42828339e-01 -7.34557770e-03 2.46360362e-01 2.73908585e-01 8.01930130e-01 -1.30601084e+00 1.49697214e-01 -2.28432789e-01 -1.15652800e-01 1.04900110e+00 1.18174076e-01 -3.44343603e-01 -1.26913261e+00 -5.47400653e-01 2.59811163e-01 -7.57242143e-01 8.84466767e-01 5.35752296e-01 1.25437164e+00 -4.78945166e-01 -2.79372931e-01 6.87852025e-01 1.08652306e+00 1.03431270e-01 3.44043761e-01 3.30316812e-01 6.17793858e-01 4.89169627e-01 6.03239179e-01 4.61513221e-01 1.77931696e-01 3.57271284e-01 2.67662346e-01 -1.28349140e-01 -6.21288195e-02 -4.96817052e-01 4.64172989e-01 2.60662824e-01 6.31292641e-01 -2.32623369e-01 -7.25905597e-01 7.84440517e-01 -1.65786958e+00 -8.16168487e-01 4.10093009e-01 2.58645368e+00 1.25360608e+00 3.50003511e-01 4.82812196e-01 -8.57050717e-02 9.16479528e-01 -2.59279087e-03 -1.16149819e+00 -3.23883384e-01 -3.10354978e-01 2.58796811e-01 8.83816838e-01 5.77769518e-01 -1.52725339e+00 8.64473820e-01 6.36819077e+00 1.40039694e+00 -9.93733943e-01 1.58901572e-01 8.97755682e-01 2.00592041e-01 -1.46106601e-01 -2.83726722e-01 -7.50670075e-01 5.70212305e-01 8.65629077e-01 -2.52208382e-01 3.04472685e-01 1.40103662e+00 -2.93603122e-01 6.89959328e-04 -1.25160778e+00 9.72140670e-01 1.36756711e-02 -9.41855609e-01 -7.65801966e-02 3.45918462e-02 7.94137776e-01 5.64489067e-02 5.29862165e-01 4.37889695e-01 1.00357771e+00 -9.55439866e-01 7.33256876e-01 1.43682435e-01 1.30044305e+00 -9.51715052e-01 7.59741247e-01 6.50987625e-01 -6.26703799e-01 -3.18059400e-02 -5.71765482e-01 3.32053185e-01 -3.76277775e-01 7.20375478e-01 -9.11301732e-01 2.32630596e-01 5.64267695e-01 5.00367403e-01 -2.88358659e-01 7.98372388e-01 -1.52102530e-01 8.89768362e-01 -4.97064918e-01 2.42183566e-01 -3.26157175e-02 -8.83393548e-03 7.51074493e-01 1.20983267e+00 -1.01278611e-01 -2.99933165e-01 2.80484736e-01 9.39993799e-01 -3.43208492e-01 -1.38756737e-01 -7.78149366e-01 1.34345740e-01 7.13751495e-01 6.71228945e-01 -2.12955683e-01 -3.34884346e-01 -1.41547816e-02 1.13330269e+00 6.15791440e-01 3.94284457e-01 -7.96032488e-01 -3.27076703e-01 8.38490665e-01 1.21104799e-01 2.30431631e-01 1.57200947e-01 -3.79202396e-01 -9.88807797e-01 -1.60113141e-01 -9.10952687e-01 6.92472398e-01 -3.08212101e-01 -2.14265299e+00 4.39974874e-01 9.77335796e-02 -1.22863340e+00 -3.88395578e-01 -6.24943495e-01 -5.74679315e-01 1.00055146e+00 -1.43162537e+00 -9.99242485e-01 3.16731989e-01 8.11641812e-01 1.94262743e-01 -3.90935570e-01 9.14141178e-01 -2.98043545e-02 -4.44950610e-01 1.15767539e+00 5.79736292e-01 4.11756754e-01 1.13881433e+00 -1.60945141e+00 3.51605684e-01 8.33934724e-01 -1.97417606e-02 4.86944795e-01 6.86495781e-01 -5.63421786e-01 -6.42446518e-01 -1.45742154e+00 6.03272557e-01 -5.31154633e-01 6.20471716e-01 -5.19078553e-01 -1.04554427e+00 6.24941468e-01 -2.44058087e-01 3.66402686e-01 7.51119375e-01 -5.24836406e-02 -9.10312295e-01 -1.52411595e-01 -1.67809498e+00 4.62168902e-01 6.98300302e-01 -7.87173152e-01 -6.44925117e-01 2.05154940e-01 6.08833432e-01 -2.93526053e-01 -6.15395248e-01 3.10862899e-01 3.02081436e-01 -7.43225574e-01 9.85655427e-01 -7.05934763e-01 7.42590949e-02 -2.06534103e-01 -2.14374691e-01 -1.46524823e+00 -3.78103226e-01 -5.82698464e-01 -1.69653833e-01 1.41394615e+00 2.59839088e-01 -7.82010138e-01 8.80659342e-01 5.23207188e-01 2.73839951e-01 -4.29517210e-01 -1.24108350e+00 -1.26649022e+00 7.24169016e-01 -5.78572929e-01 3.24470282e-01 1.16068363e+00 -2.94728905e-01 2.63413906e-01 -3.71133089e-01 3.03936511e-01 1.08036554e+00 -2.25680739e-01 6.90537453e-01 -1.31512761e+00 -4.02373314e-01 -3.66825014e-01 -2.60570079e-01 -1.17306066e+00 7.20167339e-01 -8.21903408e-01 2.60536581e-01 -7.93488622e-01 1.29362196e-01 -6.56451344e-01 -3.94334406e-01 4.89448845e-01 -1.18647136e-01 2.21897215e-01 -6.74861819e-02 2.32941315e-01 -5.89820683e-01 6.46847844e-01 9.33211803e-01 -2.92445540e-01 -1.38219550e-01 3.14026654e-01 -9.32565808e-01 7.93074250e-01 7.45489657e-01 -7.62131333e-01 -4.85334575e-01 5.97101031e-03 -4.04350460e-01 -3.26152831e-01 3.08762133e-01 -8.74484539e-01 1.56542554e-01 -2.50646919e-01 5.31737089e-01 -2.79076785e-01 4.54911441e-02 -8.30279112e-01 -2.74420470e-01 2.28194892e-01 -6.14073277e-01 -6.02312207e-01 2.52959400e-01 7.17787743e-01 4.19775285e-02 -1.91307560e-01 1.48487806e+00 2.30137900e-01 -4.24504131e-01 3.00407916e-01 -1.31527901e-01 5.65571547e-01 1.08355844e+00 -2.68156528e-01 -1.91850856e-01 -3.38842273e-01 -5.47037184e-01 1.46512851e-01 6.56330109e-01 1.15840130e-01 3.93294632e-01 -1.42963850e+00 -7.39708483e-01 3.92981768e-01 3.01596820e-01 3.07372570e-01 5.58077246e-02 3.31819803e-01 1.60513725e-02 1.66492257e-02 -5.35274111e-02 -6.33097947e-01 -9.69197989e-01 7.42021024e-01 4.24539149e-01 -3.30829173e-01 -2.96264172e-01 1.22601151e+00 6.46465659e-01 -7.78367162e-01 3.84234697e-01 3.73817198e-02 7.41486549e-02 -2.26977065e-01 5.25389433e-01 1.92402184e-01 -3.08465958e-01 -6.66531920e-01 -2.83072859e-01 1.96921751e-01 -2.16763765e-01 -1.04734272e-01 9.76723313e-01 -4.17101160e-02 2.26176694e-01 5.18745482e-01 1.26276886e+00 1.47669911e-01 -1.69841886e+00 -6.80647194e-01 -8.89587998e-02 -3.59598845e-01 -5.18494248e-02 -8.74097288e-01 -7.20518112e-01 7.21369684e-01 6.11001194e-01 1.49130821e-01 9.84029710e-01 5.36014661e-02 5.26622117e-01 2.03771695e-01 3.33887517e-01 -1.14756668e+00 5.01412759e-03 6.65182829e-01 5.10856807e-01 -1.32161748e+00 -2.80666351e-01 -1.04997136e-01 -1.00276780e+00 7.53334343e-01 7.67907798e-01 -4.94183570e-01 6.09924257e-01 4.65227127e-01 6.13418296e-02 5.48627853e-01 -3.97505820e-01 -8.46045613e-02 2.97160745e-01 1.28209925e+00 -2.08210379e-01 2.21403182e-01 4.34773624e-01 8.07705402e-01 -1.69076636e-01 -2.94690758e-01 1.76841319e-01 6.20290339e-01 -2.45755479e-01 -1.08958828e+00 -4.58412081e-01 4.53787297e-01 -3.22483093e-01 5.21410219e-02 -4.40657258e-01 6.84986293e-01 2.98585624e-01 7.25642264e-01 2.46615767e-01 -3.46235931e-01 2.10491210e-01 1.15986057e-01 2.90607065e-01 -6.28423512e-01 -1.67235985e-01 -1.92059800e-02 -1.66213766e-01 -2.94928312e-01 -1.99465111e-01 -6.39492691e-01 -9.59965706e-01 -2.80662268e-01 -3.76061738e-01 1.75709546e-01 1.91165388e-01 9.41971302e-01 2.49482378e-01 6.09989390e-02 8.45543087e-01 -5.62414825e-01 -1.18313491e+00 -9.67726111e-01 -9.85996902e-01 4.78144079e-01 5.56668520e-01 -7.57830977e-01 -6.62347257e-01 5.77601306e-02]
[10.295648574829102, 3.2252726554870605]
48c7a5fd-6940-4f89-8714-cdab8bb1c6ac
mtfnet-mutual-transformer-fusion-network-for
2112.01177
null
https://arxiv.org/abs/2112.01177v3
https://arxiv.org/pdf/2112.01177v3.pdf
MutualFormer: Multi-Modality Representation Learning via Cross-Diffusion Attention
Aggregating multi-modality data to obtain reliable data representation attracts more and more attention. Recent studies demonstrate that Transformer models usually work well for multi-modality tasks. Existing Transformers generally either adopt the Cross-Attention (CA) mechanism or simple concatenation to achieve the information interaction among different modalities which generally ignore the issue of modality gap. In this work, we re-think Transformer and extend it to MutualFormer for multi-modality data representation. Rather than CA in Transformer, MutualFormer employs our new design of Cross-Diffusion Attention (CDA) to conduct the information communication among different modalities. Comparing with CA, the main advantages of the proposed CDA are three aspects. First, the crossaffinities in CDA are defined based on the individual modality affinities in the metric space which thus can naturally avoid the issue of modality/domain gap in feature based CA definition. Second, CDA provides a general scheme which can either be used for multimodality representation or serve as the post-optimization for existing CA models. Third, CDA is implemented efficiently. We successfully apply the MutualFormer on different multi-modality learning tasks (i.e., RGB-Depth SOD, RGB-NIR object ReID). Extensive experiments demonstrate the effectiveness of the proposed MutualFormer.
['Bin Luo', 'Jin Tang', 'Bo Jiang', 'Xiao Wang', 'Xixi Wang']
2021-12-02
null
null
null
null
['rgb-d-salient-object-detection']
['computer-vision']
[-5.85849397e-02 -3.00345898e-01 -1.48144916e-01 -3.65311921e-01 -9.92158532e-01 -3.99906695e-01 5.89566469e-01 -2.02978905e-02 -3.94341528e-01 3.67432505e-01 5.53214431e-01 -5.44672161e-02 -3.19328487e-01 -6.87058568e-01 -4.99952674e-01 -9.93559837e-01 5.89734554e-01 2.77353246e-02 8.08460638e-02 -2.65934736e-01 1.09465890e-01 2.95433700e-01 -1.36638415e+00 3.16781342e-01 9.71745193e-01 1.17815602e+00 3.68734270e-01 1.22333206e-01 -3.96544486e-01 5.64909339e-01 -2.66079724e-01 -4.93584692e-01 6.14270195e-02 -4.29988235e-01 -7.77197778e-01 8.82846639e-02 1.29052565e-01 -1.49250954e-01 -4.27034587e-01 1.06531537e+00 6.99701965e-01 2.39040852e-01 6.30875349e-01 -1.38672042e+00 -9.97634053e-01 6.59209609e-01 -9.28440750e-01 1.94870815e-01 3.31455112e-01 -1.25529006e-01 1.15732837e+00 -1.15910709e+00 2.06352592e-01 1.47109616e+00 5.05139589e-01 4.58449334e-01 -8.22823584e-01 -6.15909278e-01 4.66259092e-01 3.41531277e-01 -1.49875021e+00 -2.96063095e-01 9.57901716e-01 -2.18916282e-01 6.92495704e-01 4.06155735e-01 5.27731955e-01 9.06570196e-01 -2.70585380e-02 1.27654433e+00 1.13456988e+00 -3.51432979e-01 -1.87647179e-01 3.14007364e-02 1.69800326e-01 4.80788112e-01 -9.84912664e-02 -2.87702411e-01 -5.83936334e-01 -2.25962717e-02 8.68601918e-01 3.04148436e-01 -4.35765415e-01 -2.71578103e-01 -1.69457173e+00 7.16243207e-01 7.20356643e-01 5.16664982e-01 -3.44716966e-01 -2.25867145e-02 4.62950051e-01 2.88132310e-01 3.40185344e-01 5.78377172e-02 -1.94080502e-01 1.10813500e-02 -3.36952150e-01 -1.88841939e-01 2.73175873e-02 9.40802276e-01 7.59230971e-01 -6.95764869e-02 -1.76248223e-01 1.14618003e+00 6.71821833e-01 4.64234263e-01 8.77990603e-01 -5.09110689e-01 6.71247244e-01 7.29619086e-01 -2.43127510e-01 -1.05068898e+00 -3.82886708e-01 -2.59247720e-01 -1.05909407e+00 -1.48728967e-01 1.17093399e-01 -6.32768869e-02 -7.81667829e-01 1.80573034e+00 3.47326219e-01 -1.99523754e-02 2.56390363e-01 1.12262464e+00 1.22725511e+00 5.24797857e-01 1.90114632e-01 -1.72380909e-01 1.38297224e+00 -1.11432433e+00 -9.90418851e-01 8.49079266e-02 2.95019299e-01 -7.27029204e-01 1.02047873e+00 1.46894678e-01 -9.80748117e-01 -6.25288486e-01 -9.28960502e-01 -3.09355706e-01 -5.71853817e-01 -3.77570428e-02 7.98246324e-01 3.82051021e-01 -9.92490530e-01 -1.49254967e-02 -6.82822227e-01 -4.10794854e-01 3.24687839e-01 3.72438997e-01 -6.75198913e-01 -1.09443121e-01 -1.25739467e+00 8.48767519e-01 5.07978737e-01 4.57714587e-01 -5.17597258e-01 -4.54679966e-01 -9.19915497e-01 -7.44770542e-02 2.88795769e-01 -7.60420918e-01 9.07770216e-01 -1.20655954e+00 -1.41608512e+00 6.85597599e-01 -2.88260162e-01 2.00685516e-01 2.31732130e-01 -2.41196603e-01 -5.55918634e-01 2.24138480e-02 -8.48618057e-03 6.74278438e-01 7.03921854e-01 -1.38410723e+00 -7.70597100e-01 -4.39672083e-01 2.63938695e-01 7.25346029e-01 -7.16843843e-01 -1.74261883e-01 -7.90332317e-01 -7.78015316e-01 5.64269245e-01 -7.40331590e-01 1.06691509e-01 -1.04826326e-02 -4.15661097e-01 -6.28081083e-01 8.31298709e-01 -2.84379780e-01 1.29964519e+00 -2.27541423e+00 7.05002069e-01 1.90155789e-01 3.84330004e-01 -3.56189758e-02 -2.59567022e-01 4.94802564e-01 -1.80563390e-01 1.73854828e-03 -1.96859643e-01 -5.77505291e-01 5.87353893e-02 2.21946031e-01 -3.75398286e-02 4.63069588e-01 2.18138054e-01 1.00180113e+00 -7.60257483e-01 -5.96237957e-01 2.58703530e-01 7.32897162e-01 -2.61353225e-01 3.14711660e-01 2.95142531e-01 4.50274974e-01 -6.36714280e-01 9.53429937e-01 7.37833917e-01 -2.80414641e-01 -1.18254565e-01 -8.33367348e-01 -8.68214890e-02 -9.63957906e-02 -1.15693796e+00 2.16677928e+00 -6.08988822e-01 2.97456801e-01 1.70965604e-02 -8.69990528e-01 6.94841087e-01 2.64935315e-01 6.07707262e-01 -8.34923804e-01 3.22525501e-01 1.51773557e-01 -9.48749632e-02 -5.78373373e-01 4.70904648e-01 -3.78221452e-01 -2.18448765e-03 4.12586242e-01 1.20745242e-01 2.17898563e-01 -1.01657659e-01 1.08437240e-01 4.97737348e-01 8.08940977e-02 1.49323568e-01 -2.12694436e-01 7.96662569e-01 -3.38449627e-01 6.22540891e-01 3.89942139e-01 -2.53311872e-01 7.85018623e-01 1.13199279e-01 3.85207636e-03 -4.13704395e-01 -9.85902965e-01 -6.07975274e-02 1.22882771e+00 6.96145892e-01 -2.82085896e-01 -2.38535658e-01 -7.81971395e-01 -6.73774332e-02 1.16565391e-01 -6.96484804e-01 -2.01180249e-01 -2.47981787e-01 -7.84214556e-01 3.68647486e-01 7.34998584e-01 8.43829453e-01 -8.64288330e-01 -2.03693032e-01 4.58890572e-02 -5.08592486e-01 -8.21871459e-01 -6.43902898e-01 -2.55491491e-02 -7.07301676e-01 -8.18783939e-01 -1.01425529e+00 -8.94936860e-01 5.13464391e-01 7.20529735e-01 7.23217726e-01 -7.00363666e-02 2.11671650e-01 7.23834395e-01 -5.96982360e-01 -2.01080114e-01 1.67765796e-01 2.33056620e-02 -2.61316989e-02 4.90945578e-01 4.07927066e-01 -5.23280203e-01 -7.75293648e-01 2.36753866e-01 -1.26177526e+00 1.69412464e-01 7.28790402e-01 1.03825164e+00 7.22246528e-01 8.42695236e-02 7.49441981e-01 -5.51443815e-01 7.33059406e-01 -7.18245983e-01 -5.14177196e-02 6.96620703e-01 -3.81199211e-01 4.06678254e-03 3.11023176e-01 -5.75641453e-01 -1.27662659e+00 -1.37004375e-01 -2.31222615e-01 -6.63305044e-01 6.29593804e-02 8.80191505e-01 -5.33536255e-01 -1.62636653e-01 4.16029468e-02 3.24576706e-01 1.14845159e-02 -6.17460966e-01 4.70370680e-01 7.35494435e-01 2.83934832e-01 -6.60697281e-01 4.81949896e-01 5.03311336e-01 -2.29182094e-01 -4.35880035e-01 -6.23948812e-01 -4.88449097e-01 -5.21336555e-01 -2.26011455e-01 9.15760696e-01 -1.16768324e+00 -8.25268567e-01 7.85561621e-01 -9.48147893e-01 1.90007523e-01 6.20772280e-02 6.50406241e-01 -1.60018921e-01 5.68006396e-01 -5.57623327e-01 -7.66825616e-01 -3.61565530e-01 -1.29551339e+00 1.18644321e+00 5.68454802e-01 3.91269416e-01 -1.24714196e+00 -2.13287592e-01 4.43325311e-01 5.52469969e-01 6.34570001e-03 8.47158670e-01 -3.62504065e-01 -3.72585684e-01 9.16934982e-02 -5.76393485e-01 2.26892561e-01 5.44852495e-01 -2.97675759e-01 -1.03039622e+00 -2.75693715e-01 -6.42503947e-02 -4.48713034e-01 8.65375578e-01 2.36707926e-01 1.14218271e+00 7.80193508e-02 -2.99264997e-01 4.49355721e-01 1.55392277e+00 7.63261765e-02 6.05414331e-01 4.02310789e-01 1.17114913e+00 5.66720307e-01 6.66230440e-01 3.40968817e-01 9.40663397e-01 6.90364301e-01 3.72722238e-01 -4.39986795e-01 -1.33482307e-01 -1.29626602e-01 3.03215861e-01 1.18000495e+00 -3.53313722e-02 -2.66460240e-01 -7.09006727e-01 6.11752570e-01 -2.05467463e+00 -7.32011676e-01 -7.55197182e-02 2.11097646e+00 7.76287735e-01 -2.90082157e-01 2.45301262e-01 -5.43529093e-02 7.42061198e-01 -5.58044994e-04 -5.58816969e-01 1.20783979e-02 -5.31154335e-01 -2.93645710e-01 1.36956528e-01 3.16778034e-01 -1.12743759e+00 5.50683916e-01 5.43556881e+00 1.01441431e+00 -1.25107002e+00 3.90866697e-01 3.80520910e-01 2.28054309e-03 -7.29198635e-01 -2.33663201e-01 -4.99298364e-01 3.85656089e-01 2.12225094e-01 -8.99139419e-02 2.90196061e-01 3.66975486e-01 -1.15260586e-01 -1.56076103e-01 -9.01243210e-01 1.50698519e+00 2.00552434e-01 -8.85847807e-01 2.02264875e-01 -6.78743348e-02 5.10293663e-01 -1.01336753e-02 2.04269692e-01 3.09044659e-01 3.84690426e-02 -7.78275013e-01 7.99803376e-01 5.63256800e-01 5.98888993e-01 -7.24010706e-01 7.94924259e-01 1.38184562e-01 -1.44763184e+00 -1.54756621e-01 -2.48096943e-01 3.64253551e-01 2.29637533e-01 3.83570164e-01 1.62308738e-01 1.17149627e+00 7.15546429e-01 1.05620372e+00 -7.10209012e-01 9.44910109e-01 1.40108451e-01 9.25077870e-02 -4.19651777e-01 2.14382201e-01 2.96308696e-01 -3.09637487e-01 3.87150228e-01 1.10619807e+00 2.76565731e-01 7.69460723e-02 1.94765195e-01 5.19364655e-01 1.06125645e-01 2.45670602e-01 -3.42774570e-01 6.70759305e-02 3.92011553e-01 1.21838474e+00 -3.70267600e-01 -2.46083990e-01 -8.41955483e-01 1.00258374e+00 3.90242100e-01 4.94274527e-01 -9.12226737e-01 -3.53573799e-01 5.32171905e-01 -4.33661610e-01 3.42243314e-01 -4.23211195e-02 -2.14429989e-01 -1.43812799e+00 1.37657747e-01 -8.68538082e-01 7.40956008e-01 -8.96892369e-01 -1.81300318e+00 8.66266608e-01 1.84155270e-01 -1.48438776e+00 4.06077653e-01 -3.37736696e-01 -3.56673241e-01 1.03145766e+00 -1.83631527e+00 -1.66016710e+00 -3.66521537e-01 1.06728935e+00 4.90672201e-01 -1.08739892e-02 5.69499075e-01 6.17287695e-01 -7.65444040e-01 7.20707357e-01 -1.08317966e-02 -1.42215090e-02 7.38330603e-01 -1.16067922e+00 -4.34690267e-01 7.79679894e-01 -6.03871420e-02 8.47925842e-01 2.83966660e-01 -3.42118531e-01 -1.66639161e+00 -7.60672271e-01 5.78530729e-01 -2.88372129e-01 7.00328648e-01 3.50548476e-02 -9.80324626e-01 4.92999226e-01 6.17086589e-01 -6.70003965e-02 9.55415964e-01 2.36502618e-01 -5.42680502e-01 -3.85126382e-01 -9.36962068e-01 4.35539365e-01 9.73729551e-01 -6.75279915e-01 -6.45440578e-01 8.41982216e-02 5.74041307e-01 -3.60068023e-01 -1.20110941e+00 5.66503108e-01 6.00159883e-01 -8.43448281e-01 7.66498148e-01 -3.69699925e-01 3.20466816e-01 -5.81036568e-01 -3.96689415e-01 -1.24749863e+00 -3.95695448e-01 -2.07342729e-01 -8.58201925e-03 1.63055956e+00 2.98033297e-01 -8.07452798e-01 1.64745525e-02 6.23039663e-01 -7.52573907e-02 -7.30728507e-01 -1.03764319e+00 -4.09424394e-01 9.44792405e-02 -3.43001455e-01 7.49189019e-01 1.12205470e+00 1.71702310e-01 4.13801372e-01 -3.81190509e-01 1.10005476e-01 4.92375582e-01 3.51191431e-01 2.81798661e-01 -9.05777693e-01 -2.72034347e-01 -6.66191161e-01 -2.79985607e-01 -1.40784287e+00 9.23284963e-02 -8.47818315e-01 5.01641724e-03 -1.65799296e+00 4.59632039e-01 -7.23545969e-01 -9.08768058e-01 6.70614064e-01 -4.39230561e-01 1.93342403e-01 2.87742972e-01 4.38546747e-01 -6.31448209e-01 9.95841503e-01 1.43080080e+00 -1.79755747e-01 -1.22004375e-01 -3.13712060e-01 -1.03591394e+00 3.72322798e-01 5.85061431e-01 2.79300194e-02 -5.71144164e-01 -8.68667841e-01 1.39964044e-01 -8.11125636e-02 2.98066407e-01 -6.47040546e-01 4.10856634e-01 -1.35265157e-01 1.76659688e-01 -5.38485646e-01 3.99596393e-01 -1.14362001e+00 7.58441612e-02 -1.31949514e-01 -2.31515303e-01 2.65507072e-01 1.42910510e-01 7.37910330e-01 -6.43229306e-01 1.78528488e-01 5.09670973e-01 4.70550582e-02 -9.52572107e-01 4.05954987e-01 -7.57743493e-02 -3.09993565e-01 8.32856894e-01 -2.84353793e-01 -4.04323399e-01 -3.49901080e-01 -6.02636993e-01 4.64491427e-01 1.79458678e-01 8.09345007e-01 7.41649389e-01 -1.82734787e+00 -5.16965985e-01 1.52706042e-01 4.10589576e-01 1.45439841e-02 6.07372224e-01 1.35902762e+00 -8.16699769e-03 2.43837163e-01 -1.74622938e-01 -7.43240118e-01 -1.15220797e+00 5.43594837e-01 4.11472857e-01 1.29404347e-02 -4.00175929e-01 9.68358755e-01 5.51855087e-01 -3.03258151e-01 1.65300786e-01 -1.08751133e-01 -4.57827657e-01 2.76663482e-01 5.00986457e-01 2.58525293e-02 -1.24413349e-01 -1.07419348e+00 -5.71798027e-01 8.14216495e-01 -1.31355494e-01 -1.38203323e-01 1.19946384e+00 -5.60588658e-01 -2.55819678e-01 6.79925501e-01 1.28971910e+00 -8.70553851e-02 -1.06880891e+00 -5.66115201e-01 -3.89217705e-01 -5.18182039e-01 3.37605059e-01 -6.74603939e-01 -1.55795097e+00 9.01451707e-01 7.62650371e-01 2.60323465e-01 1.63806367e+00 1.26642019e-01 7.51625359e-01 -4.44202572e-02 2.71861643e-01 -8.71314585e-01 -6.58249308e-04 1.75809294e-01 1.02176893e+00 -1.40805030e+00 -8.02615136e-02 -4.03451413e-01 -1.03714228e+00 1.03683007e+00 8.51651073e-01 3.31122965e-01 7.75762677e-01 -1.44788951e-01 2.64219195e-01 -2.92194158e-01 -5.37694991e-01 -6.41598165e-01 5.43991208e-01 4.71331745e-01 5.22564411e-01 -8.11061263e-02 -3.88636976e-01 7.29201198e-01 4.43079084e-01 -2.02033296e-01 1.14280619e-01 9.37010527e-01 -8.35505202e-02 -1.17656827e+00 -3.68271202e-01 2.14944199e-01 -1.66993186e-01 -5.44287935e-02 -1.32274985e-01 5.77679396e-01 1.90089494e-01 1.15476263e+00 -1.17492452e-01 -4.98602211e-01 4.14121717e-01 -1.76556617e-01 5.11306286e-01 -2.89076269e-01 -6.44190371e-01 4.54535604e-01 -3.00345480e-01 -3.43318820e-01 -1.13115907e+00 -6.65313840e-01 -9.94675815e-01 -1.91367015e-01 -7.37235665e-01 9.65938787e-04 5.36651254e-01 1.12454295e+00 5.22146165e-01 5.13889015e-01 6.98517144e-01 -6.04634464e-01 -9.75666493e-02 -9.89789248e-01 -5.56433499e-01 5.26697993e-01 4.05361176e-01 -9.26179051e-01 -1.95593014e-01 -1.37471169e-01]
[13.0941801071167, 4.965790748596191]
59f4460d-088c-4497-bd63-9563657c283f
vast-the-valence-assessing-semantics-test-for
2203.07504
null
https://arxiv.org/abs/2203.07504v1
https://arxiv.org/pdf/2203.07504v1.pdf
VAST: The Valence-Assessing Semantics Test for Contextualizing Language Models
VAST, the Valence-Assessing Semantics Test, is a novel intrinsic evaluation task for contextualized word embeddings (CWEs). VAST uses valence, the association of a word with pleasantness, to measure the correspondence of word-level LM semantics with widely used human judgments, and examines the effects of contextualization, tokenization, and LM-specific geometry. Because prior research has found that CWEs from GPT-2 perform poorly on other intrinsic evaluations, we select GPT-2 as our primary subject, and include results showing that VAST is useful for 7 other LMs, and can be used in 7 languages. GPT-2 results show that the semantics of a word incorporate the semantics of context in layers closer to model output, such that VAST scores diverge between our contextual settings, ranging from Pearson's rho of .55 to .77 in layer 11. We also show that multiply tokenized words are not semantically encoded until layer 8, where they achieve Pearson's rho of .46, indicating the presence of an encoding process for multiply tokenized words which differs from that of singly tokenized words, for which rho is highest in layer 0. We find that a few neurons with values having greater magnitude than the rest mask word-level semantics in GPT-2's top layer, but that word-level semantics can be recovered by nullifying non-semantic principal components: Pearson's rho in the top layer improves from .32 to .76. After isolating semantics, we show the utility of VAST for understanding LM semantics via improvements over related work on four word similarity tasks, with a score of .50 on SimLex-999, better than the previous best of .45 for GPT-2. Finally, we show that 8 of 10 WEAT bias tests, which compare differences in word embedding associations between groups of words, exhibit more stereotype-congruent biases after isolating semantics, indicating that non-semantic structures in LMs also mask biases.
['Aylin Caliskan', 'Robert Wolfe']
2022-03-14
null
null
null
null
['word-similarity']
['natural-language-processing']
[ 9.42451321e-03 -6.93787187e-02 -2.00872645e-01 -2.34732226e-01 -5.33987284e-01 -8.48981261e-01 6.26114130e-01 4.72674459e-01 -8.92166436e-01 2.02412114e-01 7.22755492e-01 -2.81638712e-01 6.59806058e-02 -6.32554114e-01 -4.79081511e-01 -5.21892786e-01 -2.38272667e-01 1.06067315e-01 -7.73461536e-02 -4.83364940e-01 3.37837815e-01 1.58000931e-01 -1.54298615e+00 5.17148733e-01 4.59561378e-01 9.40423906e-01 2.00277105e-01 3.10932994e-01 -2.51145273e-01 1.41782641e-01 -5.58846176e-01 -3.35889518e-01 5.92515953e-02 -3.57054889e-01 -6.11884892e-01 -6.32711470e-01 1.00864041e+00 3.48022014e-01 4.44870368e-02 8.39663386e-01 5.95191956e-01 4.58105087e-01 9.43999946e-01 -9.92540717e-01 -1.19240701e+00 8.15863013e-01 -1.51521146e-01 2.36116678e-01 2.93110192e-01 3.86766344e-01 1.77477145e+00 -1.30751848e+00 6.02749109e-01 1.60090363e+00 9.71831679e-01 4.76787835e-01 -1.50407505e+00 -7.14199603e-01 3.44816506e-01 -7.93661475e-02 -1.44379759e+00 -2.57474124e-01 5.29150069e-01 -3.64932775e-01 1.46998656e+00 2.20218942e-01 7.82003403e-01 1.28239012e+00 2.77079016e-01 2.74420023e-01 1.41790617e+00 -3.03800255e-01 3.10187310e-01 2.86663324e-01 1.39995679e-01 1.78848699e-01 3.75843555e-01 2.80631091e-02 -7.58114398e-01 9.69383691e-04 4.55341309e-01 -4.88575071e-01 -1.47505805e-01 -8.54731277e-02 -1.34915388e+00 8.72693002e-01 5.28260946e-01 4.57508087e-01 -1.07729070e-01 2.35048145e-01 6.21720135e-01 3.55151147e-01 5.22122383e-01 1.09730840e+00 -4.58768427e-01 -7.91802406e-02 -7.77924955e-01 3.87732238e-01 5.51494837e-01 7.27232993e-01 6.69157863e-01 3.69873494e-01 -2.85352528e-01 1.09604931e+00 1.29183531e-01 6.34349823e-01 8.82612050e-01 -6.81324601e-01 2.60230780e-01 3.47591519e-01 -1.49554878e-01 -1.17523050e+00 -5.43278635e-01 -4.73733246e-01 -1.44252151e-01 -9.36223716e-02 4.17244583e-02 5.06932288e-02 -5.95550179e-01 2.48148727e+00 -2.75294602e-01 -1.61396980e-01 7.96506777e-02 8.39220583e-01 9.69868064e-01 4.69326943e-01 7.47615039e-01 4.53948416e-02 1.58179688e+00 -4.55932021e-01 -5.10274291e-01 -8.85561764e-01 9.86136317e-01 -6.97185338e-01 2.05174565e+00 1.32222846e-01 -1.02305257e+00 -8.51648092e-01 -1.35178280e+00 -2.22310126e-01 -9.29986238e-01 -4.12943691e-01 6.29505336e-01 7.64392853e-01 -1.33173525e+00 6.27869964e-01 -2.29698405e-01 -5.63974380e-01 3.42322737e-01 1.88497722e-01 -4.39779729e-01 2.16972809e-02 -1.61996794e+00 1.58918643e+00 2.93671161e-01 -3.14850330e-01 -4.62676704e-01 -9.16921020e-01 -1.18228090e+00 1.14213973e-01 -2.83080637e-01 -3.96987110e-01 8.37677479e-01 -1.05027866e+00 -8.59972000e-01 1.09857702e+00 -3.32038313e-01 -4.86439764e-02 -3.79670292e-01 -9.87633914e-02 -7.81432807e-01 -1.46728158e-01 2.03503653e-01 9.56774771e-01 5.66635966e-01 -1.19058239e+00 -3.43856364e-01 -1.50967345e-01 3.42146046e-02 4.74137485e-01 -1.03810644e+00 -2.18806788e-01 1.38027281e-01 -8.05343449e-01 -8.56377631e-02 -8.26626182e-01 1.28754020e-01 -2.04095349e-01 8.07125643e-02 -4.32745427e-01 3.30959231e-01 -5.42805910e-01 1.33342350e+00 -2.32992959e+00 -1.52715966e-01 3.14492583e-01 5.73300160e-02 -1.03579275e-01 -7.02160656e-01 4.80555534e-01 -5.34889817e-01 6.46849036e-01 -2.05986634e-01 -4.42926675e-01 4.84345704e-01 1.33470044e-01 -2.01837540e-01 3.90553355e-01 3.90902370e-01 1.11431146e+00 -8.76001716e-01 -2.39010632e-01 1.61299452e-01 4.95288163e-01 -6.68307781e-01 -3.24613482e-01 -5.40313348e-02 -4.12305117e-01 4.27047700e-01 6.42435327e-02 4.65458751e-01 1.03928745e-01 1.72183916e-01 -2.71090358e-01 -1.41019374e-01 7.57101297e-01 -8.89795482e-01 1.72705841e+00 -8.25232029e-01 1.07697618e+00 -2.47392088e-01 -4.78445292e-01 8.80605459e-01 1.15888231e-02 -9.68244001e-02 -1.07102311e+00 6.46643341e-02 8.64433870e-02 4.56405342e-01 -2.04516068e-01 8.44616771e-01 -7.94867098e-01 -4.11250830e-01 4.69875693e-01 -2.09983606e-02 -4.80431437e-01 -8.27539265e-02 1.30694523e-01 8.99666369e-01 -1.59908324e-01 4.10248153e-02 -8.59637916e-01 1.19983116e-02 -6.19488955e-02 4.24259841e-01 4.32605743e-01 -9.37780216e-02 5.32611668e-01 5.60726225e-01 1.71280563e-01 -1.03324711e+00 -1.56938624e+00 -5.07453740e-01 1.34962809e+00 2.80105546e-02 -6.10717416e-01 -5.64851701e-01 -1.78857788e-01 2.05029368e-01 1.49749517e+00 -9.19133008e-01 -5.67355573e-01 -2.54910380e-01 -6.51684105e-01 7.09682763e-01 8.24680150e-01 -1.06416963e-01 -1.14939463e+00 -5.04793763e-01 -5.92077076e-02 -2.94125248e-02 -9.56062615e-01 -3.91063511e-01 4.25452083e-01 -7.41778135e-01 -5.39375603e-01 -2.60856241e-01 -9.52214658e-01 4.13130760e-01 2.28052996e-02 1.44492924e+00 8.46636370e-02 -2.44371563e-01 5.27353585e-01 -2.63287425e-01 -3.33490610e-01 -1.19891718e-01 7.03862682e-02 3.11546654e-01 -7.36961961e-01 6.94682777e-01 -4.38134611e-01 -4.86493617e-01 1.07679367e-01 -9.39429104e-01 -3.50890160e-01 3.77586544e-01 7.72270143e-01 4.14051712e-01 -9.07028019e-02 7.74101079e-01 -5.39494693e-01 9.06044304e-01 -5.38424551e-01 8.90887156e-02 -1.41975552e-01 -7.50775278e-01 4.92002852e-02 4.77720499e-01 -6.51910484e-01 -7.24606156e-01 -8.09222937e-01 -1.28515456e-02 -2.24229127e-01 4.76067606e-03 3.65024209e-01 -2.78932191e-02 2.66056180e-01 8.93030643e-01 -2.68547624e-01 -1.04252070e-01 -1.63422093e-01 5.57101905e-01 3.23743999e-01 2.27596909e-01 -8.43466520e-01 7.31432676e-01 2.64839053e-01 -2.40703791e-01 -8.91094029e-01 -8.22916567e-01 -1.63340315e-01 -3.20740283e-01 3.40677947e-01 1.03537691e+00 -1.05312312e+00 -7.46562064e-01 -1.20364703e-01 -8.39682102e-01 -5.01801670e-01 -5.47937870e-01 6.54901147e-01 -3.17385405e-01 5.35744876e-02 -6.52782798e-01 -3.96269381e-01 -1.10709555e-01 -9.31037009e-01 7.78804481e-01 -3.31649661e-01 -1.30490327e+00 -1.58009648e+00 1.86718285e-01 -1.53830528e-01 5.65034330e-01 -1.28880711e-02 1.51588261e+00 -9.71871614e-01 4.39886928e-01 1.45192131e-01 -2.35974327e-01 7.01112747e-01 1.54613435e-01 -1.94174856e-01 -1.30217862e+00 -2.53082603e-01 -8.32086354e-02 -3.77777368e-01 1.05272770e+00 2.13097230e-01 8.88870835e-01 -4.79648970e-02 -1.24135586e-02 5.15123129e-01 1.48818052e+00 -5.38557619e-02 5.60669899e-01 3.74665827e-01 6.78838909e-01 8.41037571e-01 4.29579973e-01 7.03344168e-03 3.49333107e-01 3.80072474e-01 1.89934999e-01 -3.51441205e-02 -2.53970146e-01 -3.93121481e-01 8.60275805e-01 9.70955193e-01 5.01556754e-01 -1.28394872e-01 -9.14846659e-01 8.04266155e-01 -1.16007507e+00 -8.40756357e-01 -1.61941260e-01 2.37864828e+00 8.73828351e-01 3.67235333e-01 -3.05325210e-01 -8.52689072e-02 4.34953988e-01 4.91832078e-01 -4.43628073e-01 -1.14054310e+00 -5.77990174e-01 8.36654186e-01 4.54831392e-01 6.55630410e-01 -6.37161613e-01 1.44261479e+00 7.07258892e+00 9.55592811e-01 -9.86174107e-01 8.10422823e-02 3.43119740e-01 -5.44289231e-01 -1.06370759e+00 3.87836457e-03 -4.76679921e-01 2.71531790e-01 1.24660540e+00 -1.02516659e-01 3.78431171e-01 7.19894886e-01 1.85183659e-02 -1.15502112e-01 -1.58943701e+00 9.01881099e-01 3.92689198e-01 -9.06918049e-01 2.67987490e-01 -1.65092107e-02 8.21756303e-01 -6.36242032e-02 7.01186121e-01 5.68811417e-01 1.02853946e-01 -1.66076326e+00 1.00156832e+00 4.86302562e-02 9.60472703e-01 -9.83955383e-01 5.24377882e-01 -2.94316292e-01 -1.15687275e+00 9.00152847e-02 -6.84036076e-01 -5.91050625e-01 -1.51897490e-01 4.95289773e-01 -7.03844249e-01 -2.05242038e-01 5.76169133e-01 6.56603098e-01 -7.39631951e-01 2.89211810e-01 -3.01884830e-01 6.42569661e-01 -1.26049802e-01 -1.67962641e-01 3.28743219e-01 2.29194120e-01 2.59954959e-01 1.61648679e+00 3.13691258e-01 -2.91557223e-01 -2.69313008e-01 1.09255981e+00 -1.07878551e-01 4.82615292e-01 -8.12576115e-01 -2.86980093e-01 7.22853541e-01 1.18658030e+00 -6.30916536e-01 -2.29647160e-01 -4.47794557e-01 1.05456984e+00 3.50593567e-01 4.05139536e-01 -7.88641810e-01 -4.19701546e-01 1.24111736e+00 1.59284398e-01 2.30005562e-01 -3.12080294e-01 -7.72597969e-01 -7.54089534e-01 -1.14238881e-01 -7.20462561e-01 6.12605922e-02 -1.01105189e+00 -1.58543956e+00 4.55318570e-01 -5.28269671e-02 -7.35164940e-01 8.09326172e-02 -9.59087968e-01 -6.34749115e-01 1.25252962e+00 -1.20554876e+00 -8.00756514e-01 -3.47419418e-02 4.30719882e-01 2.72859603e-01 1.19726852e-01 1.10262024e+00 -8.67243037e-02 -1.48311600e-01 8.61572862e-01 -2.86726713e-01 3.34063470e-02 1.15553570e+00 -1.29708374e+00 6.94446087e-01 4.08881158e-01 2.32412830e-01 1.19313443e+00 6.27735317e-01 -6.33312285e-01 -1.04265034e+00 -7.82185316e-01 1.11800623e+00 -8.55055690e-01 7.86528230e-01 -5.89481831e-01 -8.42794657e-01 6.24659002e-01 4.63309586e-01 -4.31793064e-01 1.10811031e+00 5.84922075e-01 -9.64210808e-01 1.51665002e-01 -8.52381766e-01 1.09452045e+00 1.30186629e+00 -1.05139875e+00 -1.04394794e+00 5.49230799e-02 1.09010577e+00 1.65200159e-01 -9.66026366e-01 1.33443266e-01 5.85744798e-01 -7.09591091e-01 1.08138883e+00 -6.24496818e-01 6.07688546e-01 1.88513175e-02 -6.95511580e-01 -1.89684081e+00 -5.69696128e-01 6.01697937e-02 6.58192873e-01 1.12192953e+00 6.90474570e-01 -7.36889660e-01 2.69678831e-01 5.08996129e-01 -2.61363626e-01 -6.63752615e-01 -8.44021440e-01 -9.84619677e-01 7.87262440e-01 -9.66153741e-01 3.87688905e-01 1.32900202e+00 2.17189372e-01 6.54546976e-01 4.65299994e-01 -2.46896878e-01 3.16646278e-01 -4.40505207e-01 -4.80688252e-02 -1.10520315e+00 -8.08177739e-02 -8.01472604e-01 -3.40058357e-01 -4.75009680e-01 6.70417786e-01 -1.43180525e+00 -3.69050652e-02 -1.63840449e+00 1.31826997e-01 -3.85469049e-01 -5.53716481e-01 5.39137244e-01 -7.70261586e-02 4.30872649e-01 3.11785400e-01 -1.59707069e-01 -3.69506925e-02 4.66648638e-01 1.00288272e+00 -5.38844950e-02 -2.45704744e-02 -1.06737173e+00 -1.25940216e+00 7.72969067e-01 8.22769344e-01 -2.05539778e-01 -6.66146874e-01 -5.16299486e-01 7.08809495e-01 -8.00757647e-01 4.06104356e-01 -1.03693259e+00 -3.89234096e-01 2.70394236e-02 7.68837333e-01 -5.66657782e-02 6.20135427e-01 -6.30001903e-01 -2.25108325e-01 3.18642646e-01 -6.44226432e-01 4.23591465e-01 7.28243470e-01 7.13586509e-02 9.53191519e-03 -1.79145738e-01 6.55253470e-01 -1.15003966e-01 -8.69568348e-01 -1.90001756e-01 -4.91331488e-01 6.89189732e-01 6.01276934e-01 -4.72441494e-01 -3.87374490e-01 -1.32091820e-01 -5.39145291e-01 4.24549878e-02 7.61291564e-01 7.51304626e-01 6.44914806e-01 -1.62650156e+00 -5.75643539e-01 1.34583116e-01 4.97773826e-01 -5.98258495e-01 1.00840837e-01 6.40560448e-01 -5.29099554e-02 2.53519267e-01 -1.51342139e-01 -3.58662218e-01 -9.48603511e-01 5.09878874e-01 5.97071797e-02 2.89482117e-01 -1.82220459e-01 1.06624913e+00 4.99212116e-01 -3.53550673e-01 -1.36535436e-01 -6.46602035e-01 -9.09475386e-02 5.91744423e-01 3.00198823e-01 2.47251928e-01 -3.61756831e-02 -8.05579066e-01 -6.04251325e-01 8.14494073e-01 1.45821497e-01 -6.04772925e-01 1.06164408e+00 4.40517887e-02 -2.20720962e-01 1.12118351e+00 1.65777552e+00 2.85645306e-01 -6.55509174e-01 1.98726699e-01 -5.17193414e-02 -1.66073650e-01 2.98480298e-02 -8.56664598e-01 -6.14358902e-01 1.07586396e+00 6.48034751e-01 -4.04081792e-01 6.42831743e-01 -6.77124336e-02 7.03468919e-01 3.11384916e-01 -2.62808390e-02 -1.50742435e+00 1.72374502e-01 8.42852414e-01 8.71420622e-01 -7.71047950e-01 -9.10270512e-02 3.78080308e-02 -8.96443963e-01 6.55622184e-01 8.66781473e-01 -2.19818935e-01 4.48905259e-01 1.83687538e-01 -3.05405371e-02 -1.56258658e-01 -8.65339339e-01 -1.64870918e-01 3.48289728e-01 5.88067532e-01 9.33757305e-01 3.24679643e-01 -3.70608747e-01 8.19057763e-01 -8.34370732e-01 -9.33328986e-01 1.56150728e-01 6.11248076e-01 -5.58451891e-01 -8.72505784e-01 -2.64938116e-01 4.29049611e-01 -2.08731428e-01 -1.03677964e+00 -5.13275206e-01 1.06614912e+00 2.71372408e-01 7.30628252e-01 7.59592652e-01 -6.29680753e-01 2.88705945e-01 5.27255356e-01 4.51949120e-01 -9.63572085e-01 -8.59693289e-01 -2.61467427e-01 3.14935446e-01 -6.70003116e-01 -1.15453936e-02 -5.49103916e-01 -1.68087828e+00 -3.59597564e-01 4.05647419e-02 8.40154737e-02 7.05477238e-01 6.57419741e-01 1.70001447e-01 4.39268082e-01 1.43040031e-01 -6.26039028e-01 -2.51946032e-01 -9.85336244e-01 -6.84340835e-01 9.02108133e-01 1.47672594e-01 -7.17673659e-01 -8.12000573e-01 -4.11742389e-01]
[10.404618263244629, 8.959829330444336]
6e908737-1765-4610-abfa-4829c47dda97
understanding-dataset-design-choices-for
1904.12106
null
http://arxiv.org/abs/1904.12106v1
http://arxiv.org/pdf/1904.12106v1.pdf
Understanding Dataset Design Choices for Multi-hop Reasoning
Learning multi-hop reasoning has been a key challenge for reading comprehension models, leading to the design of datasets that explicitly focus on it. Ideally, a model should not be able to perform well on a multi-hop question answering task without doing multi-hop reasoning. In this paper, we investigate two recently proposed datasets, WikiHop and HotpotQA. First, we explore sentence-factored models for these tasks; by design, these models cannot do multi-hop reasoning, but they are still able to solve a large number of examples in both datasets. Furthermore, we find spurious correlations in the unmasked version of WikiHop, which make it easy to achieve high performance considering only the questions and answers. Finally, we investigate one key difference between these datasets, namely span-based vs. multiple-choice formulations of the QA task. Multiple-choice versions of both datasets can be easily gamed, and two models we examine only marginally exceed a baseline in this setting. Overall, while these datasets are useful testbeds, high-performing models may not be learning as much multi-hop reasoning as previously thought.
['Jifan Chen', 'Greg Durrett']
2019-04-27
understanding-dataset-design-choices-for-1
https://aclanthology.org/N19-1405
https://aclanthology.org/N19-1405.pdf
naacl-2019-6
['multi-hop-question-answering']
['knowledge-base']
[-1.70398932e-02 5.05921721e-01 1.63270459e-01 -2.81094253e-01 -1.26878464e+00 -9.09771442e-01 4.87615883e-01 2.76880056e-01 -5.88754177e-01 8.16995740e-01 4.70243424e-01 -8.16380441e-01 -6.05790436e-01 -9.40082490e-01 -7.99365938e-01 -1.65689453e-01 3.17549944e-01 8.93490255e-01 5.08561909e-01 -6.95769489e-01 1.67042106e-01 -8.10850486e-02 -1.54821694e+00 6.46943092e-01 1.19601727e+00 4.47009295e-01 1.59769624e-01 1.10570800e+00 6.28123879e-02 1.22963750e+00 -6.22410953e-01 -6.02091253e-01 1.15740307e-01 -5.46956360e-01 -1.53074217e+00 -5.69665849e-01 9.43601787e-01 -4.09972548e-01 -2.46024325e-01 6.05483890e-01 4.21726674e-01 3.53422731e-01 5.90949059e-01 -1.11459303e+00 -6.73105001e-01 7.53671348e-01 -1.92044377e-01 4.34519172e-01 8.31068277e-01 3.48111480e-01 1.72010887e+00 -4.11138326e-01 5.68573952e-01 1.34573483e+00 5.33972144e-01 4.98449355e-01 -1.25734484e+00 -1.52105749e-01 1.75828952e-02 5.41546226e-01 -7.94707835e-01 -3.26701492e-01 5.50228596e-01 -1.99932277e-01 1.07921410e+00 5.19200385e-01 3.90665382e-01 1.01324654e+00 -1.38103083e-01 1.00337756e+00 1.39343286e+00 -4.33931142e-01 3.45928334e-02 -2.61420816e-01 5.99774718e-01 7.60283411e-01 8.08305219e-02 -8.83057714e-02 -7.18646824e-01 -1.12925097e-02 3.13886017e-01 -4.59752560e-01 -4.14245844e-01 -3.06247652e-01 -1.38288987e+00 9.01477277e-01 3.18978012e-01 2.31245041e-01 -1.20756805e-01 -2.72191074e-02 1.56702906e-01 8.29368651e-01 4.86252308e-02 1.14401615e+00 -7.43142128e-01 -5.05835414e-01 -8.06060135e-01 9.93741572e-01 1.19725263e+00 6.83019519e-01 6.45758927e-01 -6.03888512e-01 -5.62089741e-01 6.48173809e-01 -1.85667232e-01 1.94767386e-01 3.09675127e-01 -1.29251766e+00 9.46480036e-01 5.33564091e-01 2.98372179e-01 -8.56689215e-01 -8.03169012e-01 -1.85616210e-01 -2.22711012e-01 -1.49558574e-01 1.21290243e+00 -1.71832725e-01 -5.13441205e-01 1.97609806e+00 4.12236415e-02 -1.89292535e-01 1.54588372e-01 9.24282432e-01 9.30247307e-01 4.19528186e-01 -1.11256197e-01 1.53620556e-01 1.63796163e+00 -1.40299153e+00 -4.89295840e-01 -4.84258682e-01 9.84812677e-01 -4.37720776e-01 1.71622181e+00 5.51114380e-01 -1.54611051e+00 -3.93761754e-01 -8.84066105e-01 -4.95601058e-01 -2.91064620e-01 -4.85985309e-01 7.88175762e-01 4.15982753e-01 -1.18747449e+00 3.62359226e-01 -3.82495552e-01 -2.24305108e-01 1.45681649e-01 -1.20323583e-01 -1.35716796e-01 -5.83051026e-01 -1.47355783e+00 1.29018974e+00 2.96113580e-01 -2.68188238e-01 -6.71205103e-01 -7.74977028e-01 -6.24039471e-01 3.84264261e-01 6.66118801e-01 -7.62918413e-01 1.72939920e+00 -4.33894753e-01 -1.15941882e+00 8.12214792e-01 -2.57217199e-01 -2.53974795e-01 7.09199488e-01 -3.21695954e-01 -1.10028058e-01 2.89145559e-01 2.25444928e-01 5.86038530e-01 1.99088618e-01 -9.37895536e-01 -4.11131769e-01 -3.34935963e-01 8.23937058e-01 4.38018709e-01 4.13053483e-02 -3.31261218e-01 -1.23307914e-01 -3.28849256e-01 2.11175624e-02 -6.30783200e-01 9.52024013e-02 -2.66019285e-01 -3.10179949e-01 -4.77079451e-01 3.85557950e-01 -7.52097547e-01 1.08633685e+00 -1.80680895e+00 2.51391053e-01 -2.15660393e-01 4.13939923e-01 1.47148773e-01 -6.47326827e-01 6.05121732e-01 1.11319341e-01 3.48675400e-01 -6.46812394e-02 -1.43243298e-01 3.84214491e-01 3.27883571e-01 -1.87831506e-01 -9.84173790e-02 2.75087863e-01 1.45731688e+00 -9.82461214e-01 -4.35708880e-01 -2.98997909e-01 -3.55316550e-01 -9.37679648e-01 2.67026871e-01 -9.33436990e-01 1.47281274e-01 -2.32292846e-01 3.07062387e-01 5.77924848e-01 -5.07646799e-01 4.49597798e-02 1.29691899e-01 2.66673684e-01 8.55902433e-01 -8.18274379e-01 1.76742744e+00 -5.23191035e-01 6.35704279e-01 -1.56113133e-01 -9.83457327e-01 2.33236134e-01 9.97087061e-02 1.24479737e-02 -1.18499064e+00 -2.59722084e-01 -2.23036557e-02 4.82264668e-01 -8.29990923e-01 6.73166633e-01 -1.76287830e-01 -3.95003967e-02 8.25290442e-01 -1.28206667e-02 -3.36102843e-01 5.52363992e-01 5.45798182e-01 1.61673737e+00 -1.50004476e-01 -1.00471042e-01 -2.07825273e-01 1.80374280e-01 2.67233789e-01 3.24820876e-01 1.26015317e+00 -1.96032077e-01 6.42566085e-01 1.06903791e+00 -1.58439666e-01 -7.78942823e-01 -1.15548265e+00 5.14125861e-02 1.27576518e+00 1.37834743e-01 -5.51049232e-01 -6.86914504e-01 -7.76005864e-01 -1.61395520e-01 8.61073434e-01 -5.19596159e-01 -7.62957707e-02 -5.57757914e-01 -5.28073788e-01 8.68693590e-01 4.62085903e-01 6.18378699e-01 -9.42161560e-01 -7.28316486e-01 1.79137215e-01 -6.91167116e-01 -1.09631920e+00 -1.03884578e-01 1.90309912e-01 -5.18196404e-01 -1.41030669e+00 -4.81295079e-01 -6.96161628e-01 7.35938177e-02 4.18295354e-01 1.92911124e+00 5.65231979e-01 8.84556919e-02 5.90005994e-01 -6.42245531e-01 -3.09800595e-01 -2.80577838e-01 4.73211199e-01 -5.07971227e-01 -6.27403021e-01 5.63996673e-01 -4.61857885e-01 -5.51998854e-01 2.68329114e-01 -7.34373689e-01 1.98029190e-01 6.63795829e-01 9.93672907e-01 -2.52515972e-02 -3.56352255e-02 7.66970038e-01 -1.08374119e+00 1.08209264e+00 -6.01463556e-01 -3.25438797e-01 5.89794636e-01 -3.68789315e-01 3.02214533e-01 6.58323348e-01 -1.08395278e-01 -9.70378697e-01 -7.53374219e-01 -1.97488114e-01 2.54623890e-01 -1.73576087e-01 7.54644692e-01 -3.36831063e-02 2.35953584e-01 8.83093059e-01 -4.80639227e-02 -8.12882781e-02 -3.91198665e-01 5.93958139e-01 3.16082239e-01 3.79163206e-01 -1.08292973e+00 8.35515618e-01 -3.87385450e-02 -2.72372067e-01 -6.86653614e-01 -1.25945055e+00 -2.63720214e-01 -2.99559146e-01 1.54440820e-01 1.00378323e+00 -8.28365743e-01 -1.03950143e+00 3.27627271e-01 -1.21304941e+00 -8.35913479e-01 -2.31497645e-01 1.13939978e-01 -6.69050455e-01 2.89840907e-01 -7.99516499e-01 -4.73112732e-01 1.56370223e-01 -1.06737053e+00 7.40750492e-01 2.41823792e-01 -6.11222804e-01 -1.21440637e+00 3.95467505e-02 9.85399663e-01 4.89314765e-01 -1.14671558e-01 1.60344684e+00 -9.67785776e-01 -8.30807745e-01 2.48905301e-01 -3.83044749e-01 6.15049489e-02 -3.50335419e-01 -3.13996106e-01 -9.62202370e-01 -9.17246044e-02 3.65942828e-02 -1.14806473e+00 9.56673324e-01 -1.37738669e-02 1.06985116e+00 -2.59553343e-01 1.32952750e-01 1.45328850e-01 9.93810475e-01 -2.36066580e-01 7.14384496e-01 3.18621516e-01 4.47858244e-01 7.40210295e-01 5.01062989e-01 -6.66174367e-02 1.22514391e+00 5.86363435e-01 3.12972039e-01 2.40697160e-01 -5.56818657e-02 -3.64914268e-01 6.44531175e-02 7.55638778e-01 2.64948130e-01 -4.62785691e-01 -1.23335993e+00 6.78449571e-01 -1.87570739e+00 -1.08881402e+00 -3.14681888e-01 1.84592152e+00 1.01054788e+00 2.16760576e-01 2.42264420e-01 1.55024573e-01 5.30708581e-02 4.39779103e-01 -4.13322240e-01 -4.96541858e-01 -3.06626916e-01 6.01535797e-01 -5.77869415e-02 7.13141203e-01 -6.45910859e-01 7.89447248e-01 7.01058626e+00 5.86820245e-01 -5.85735261e-01 4.77995314e-02 4.84008789e-01 -9.11975875e-02 -8.86439800e-01 1.51550323e-01 -3.57960582e-01 2.36797467e-01 1.03470743e+00 6.47421032e-02 6.77532136e-01 2.72216439e-01 -9.21094790e-02 -5.52883863e-01 -1.45978498e+00 6.21983051e-01 3.05169541e-02 -1.17371976e+00 7.49806836e-02 -3.10805738e-01 5.07636905e-01 -1.62268654e-01 3.95982303e-02 7.97992229e-01 5.76916814e-01 -1.45367825e+00 6.21508956e-01 2.84291655e-01 4.37827259e-01 -4.96846139e-01 5.21744013e-01 8.20455849e-01 -4.68594968e-01 -2.61419177e-01 -2.57907838e-01 -6.26132607e-01 8.11087489e-02 3.56286943e-01 -4.45308685e-01 5.12320817e-01 5.29890239e-01 2.37710953e-01 -1.05413103e+00 8.75111699e-01 -5.24477303e-01 7.52867460e-01 -2.20848188e-01 -3.51505488e-01 3.04545313e-01 5.68892919e-02 1.61116317e-01 6.98872030e-01 -5.48277535e-02 3.61958027e-01 8.82268772e-02 7.94029653e-01 -2.25656286e-01 -2.10377946e-01 -3.34146798e-01 -2.21074149e-01 4.98479813e-01 9.01999831e-01 -2.17029139e-01 -9.21307504e-02 -7.28501022e-01 6.44265294e-01 9.83876467e-01 3.93019080e-01 -7.22989202e-01 -2.85711378e-01 5.35937548e-01 6.02597445e-02 -1.71823781e-02 -1.61385685e-01 -4.18231934e-01 -1.50866950e+00 3.55723888e-01 -1.45020366e+00 5.93833268e-01 -1.06598294e+00 -1.44416142e+00 1.62438974e-01 -1.02632754e-01 -3.86902511e-01 -3.44062269e-01 -6.30034983e-01 -6.37585402e-01 9.21441197e-01 -1.71260118e+00 -9.63072598e-01 -5.10620952e-01 5.30121088e-01 4.93316114e-01 2.14482620e-01 8.79739642e-01 9.03370455e-02 -3.25159460e-01 6.81420386e-01 -1.55300424e-01 1.59283862e-01 8.50104570e-01 -1.60705829e+00 4.04819876e-01 7.84290671e-01 4.63714927e-01 6.16426766e-01 6.97312951e-01 -1.62269726e-01 -1.37482548e+00 -5.21740735e-01 9.63455737e-01 -1.00234234e+00 6.34039879e-01 -3.24069709e-01 -1.24547052e+00 9.16982591e-01 4.04396147e-01 -3.58189374e-01 4.75893587e-01 7.43929386e-01 -7.08156109e-01 3.10458541e-01 -8.75875652e-01 8.75267804e-01 1.16916203e+00 -6.95892870e-01 -1.17335081e+00 5.92498362e-01 9.69245255e-01 -6.57674551e-01 -6.67131662e-01 3.37805673e-02 3.94227624e-01 -1.32474935e+00 8.48705769e-01 -1.04838359e+00 9.96862710e-01 3.75612937e-02 4.17015441e-02 -1.58412361e+00 -2.63014853e-01 -5.08529484e-01 -1.57975778e-01 1.03084970e+00 6.92757964e-01 -5.62178075e-01 8.88990283e-01 8.89923871e-01 4.78604957e-02 -8.43565702e-01 -8.07009399e-01 -6.53584421e-01 8.21926296e-01 -3.78831953e-01 6.34747684e-01 8.84315252e-01 3.35795254e-01 7.10424483e-01 4.72395867e-02 -2.18856707e-02 3.69135141e-01 1.44110963e-01 8.10390532e-01 -1.08613908e+00 -7.43114114e-01 -5.39433300e-01 1.70857951e-01 -1.35113513e+00 2.56785810e-01 -9.25136685e-01 -8.18725526e-02 -1.74226630e+00 1.49803832e-01 -6.41153812e-01 2.98907682e-02 4.17099029e-01 -5.97129047e-01 -2.05139861e-01 2.80487448e-01 -1.67679921e-01 -8.82823825e-01 2.48878077e-01 1.52581120e+00 -6.07828721e-02 1.58887193e-01 -1.92640111e-01 -1.14540255e+00 4.41812605e-01 7.98083544e-01 -1.43719390e-01 -7.32032776e-01 -1.02229571e+00 7.64747620e-01 4.41916555e-01 6.24318838e-01 -9.11718071e-01 3.41521025e-01 -2.43768290e-01 7.11298827e-03 -3.22644204e-01 2.75173813e-01 -4.31169033e-01 -4.52971429e-01 5.76034971e-02 -8.00916791e-01 4.36208427e-01 7.42739439e-02 3.98548663e-01 -2.77628541e-01 -4.21414763e-01 5.41901588e-01 -4.20247793e-01 -5.48171520e-01 -1.22575626e-01 -3.35404575e-01 1.05684876e+00 6.90900862e-01 6.94700405e-02 -8.78665328e-01 -6.81200862e-01 -4.13617194e-01 7.19031274e-01 3.52886707e-01 3.88786137e-01 2.94375360e-01 -8.15614641e-01 -8.69466305e-01 -1.98729560e-01 2.96700388e-01 4.11963582e-01 4.44106787e-01 7.67783701e-01 -6.31005406e-01 5.39451838e-01 -1.00761279e-01 -2.59517193e-01 -1.02744341e+00 6.09471142e-01 3.22919279e-01 -7.59914756e-01 -3.60728204e-01 9.28157151e-01 8.91467370e-03 -8.06515276e-01 3.51658128e-02 -3.43570352e-01 -1.89221531e-01 1.52754635e-01 5.44257402e-01 2.57550120e-01 1.67255729e-01 -2.81168967e-02 -3.25179920e-02 2.99694926e-01 -2.82868683e-01 -1.96234345e-01 1.18905163e+00 -1.47528648e-02 -4.62499224e-02 3.44901770e-01 9.99448836e-01 -3.82746868e-02 -9.03140366e-01 -2.57520229e-01 1.81564584e-01 -3.02123219e-01 -4.35822397e-01 -1.28841507e+00 -5.13183832e-01 1.07505381e+00 -2.97689974e-01 3.45050812e-01 9.24295843e-01 -2.89196689e-02 8.57245624e-01 8.42788696e-01 3.19411367e-01 -8.66804063e-01 3.84013206e-01 8.79223883e-01 7.08083987e-01 -1.35465121e+00 -2.68559188e-01 -1.55145213e-01 -5.23806691e-01 8.02422881e-01 9.85707462e-01 2.08616525e-01 5.14357314e-02 5.79662547e-02 1.11009464e-01 -4.07548398e-01 -1.35676992e+00 -3.33616227e-01 4.19087522e-02 5.51027775e-01 4.19268280e-01 -1.06021591e-01 -1.99348345e-01 6.05914056e-01 -7.60045648e-01 -2.00993717e-01 7.64254093e-01 9.35732007e-01 -3.62431377e-01 -9.62148368e-01 -1.90645158e-01 6.23569548e-01 -2.41704702e-01 -2.60505110e-01 -5.12259245e-01 9.18293357e-01 -4.12699640e-01 1.26842868e+00 -1.00146927e-01 -3.05190235e-01 5.25035441e-01 3.36356252e-01 6.95590079e-01 -5.87801754e-01 -5.58364153e-01 -9.34672713e-01 3.95493090e-01 -5.27670503e-01 -1.32090807e-01 -4.27800655e-01 -8.77749324e-01 -6.52376115e-01 -1.04688346e-01 3.81695241e-01 3.91336232e-02 1.27263641e+00 2.22363800e-01 5.72259963e-01 3.36525007e-03 -2.22817466e-01 -1.00084937e+00 -8.94345582e-01 -4.16704655e-01 6.68969810e-01 4.60370034e-01 -4.43344593e-01 -4.93373752e-01 -4.00713950e-01]
[11.058050155639648, 7.993850231170654]
db72d3bf-96b2-4453-8623-5c71215998ce
a-latent-feature-analysis-based-approach-for
2208.07739
null
https://arxiv.org/abs/2208.07739v1
https://arxiv.org/pdf/2208.07739v1.pdf
A Latent Feature Analysis-based Approach for Spatio-Temporal Traffic Data Recovery
Missing data is an inevitable and common problem in data-driven intelligent transportation systems (ITS). In the past decade, scholars have done many research on the recovery of missing traffic data, however how to make full use of spatio-temporal traffic patterns to improve the recovery performance is still an open problem. Aiming at the spatio-temporal characteristics of traffic speed data, this paper regards the recovery of missing data as a matrix completion problem, and proposes a spatio-temporal traffic data completion method based on hidden feature analysis, which discovers spatio-temporal patterns and underlying structures from incomplete data to complete the recovery task. Therefore, we introduce spatial and temporal correlation to capture the main underlying features of each dimension. Finally, these latent features are applied to recovery traffic data through latent feature analysis. The experimental and evaluation results show that the evaluation criterion value of the model is small, which indicates that the model has better performance. The results show that the model can accurately estimate the continuous missing data.
['Di wu', 'Yuting Ding']
2022-08-16
null
null
null
null
['matrix-completion']
['methodology']
[-3.62904556e-02 -6.71485543e-01 -4.59153265e-01 -4.04913694e-01 -5.77874064e-01 2.84686297e-01 2.97707260e-01 -5.02438247e-01 -2.37624310e-02 8.18818688e-01 7.10243881e-01 -3.17999333e-01 -6.75338805e-01 -8.17519069e-01 -3.30096960e-01 -8.30859721e-01 -1.93279739e-02 2.34572351e-01 1.59180671e-01 -2.84466594e-01 1.51635811e-01 3.05476606e-01 -1.74494338e+00 2.87069887e-01 1.07819355e+00 6.79973960e-01 3.91583979e-01 7.38146752e-02 -3.11275274e-01 8.47589970e-01 -2.26767406e-01 -1.35985967e-02 1.45605743e-01 -1.57203794e-01 -3.41637313e-01 4.35225695e-01 -3.57599556e-01 -5.46867430e-01 -7.82910943e-01 7.27640986e-01 9.66461748e-02 2.37645015e-01 5.52040219e-01 -1.80534899e+00 -4.71027195e-01 2.09223986e-01 -5.93794465e-01 2.59615272e-01 5.03109694e-02 1.39970019e-01 5.75754166e-01 -1.11812484e+00 2.05445960e-01 1.35975015e+00 5.65302610e-01 1.33986965e-01 -1.01351273e+00 -8.29888046e-01 8.12716112e-02 5.81157804e-01 -1.75478339e+00 -6.94831789e-01 8.41095269e-01 -6.79808259e-01 5.15407801e-01 2.10876226e-01 3.17789584e-01 8.24006200e-01 6.08793944e-02 9.75389659e-01 8.60639691e-01 1.45658627e-02 -7.69046620e-02 -7.52153993e-02 1.78266734e-01 2.75104910e-01 4.39235508e-01 2.40108356e-01 -3.75768453e-01 -2.50342965e-01 4.99900728e-01 7.30322182e-01 1.08721264e-01 -1.36551529e-01 -1.14629459e+00 7.07410157e-01 -6.80232281e-03 1.60244614e-01 -6.48960769e-01 -2.30476886e-01 1.94547728e-01 1.97456211e-01 4.80290592e-01 -7.47542322e-01 -1.41535267e-01 -3.26094657e-01 -9.77964282e-01 2.70540059e-01 2.23106265e-01 1.18739176e+00 1.02648556e+00 2.87480503e-01 -1.61620036e-01 7.24140286e-01 2.43809924e-01 1.00233841e+00 3.75914350e-02 -9.89153802e-01 1.03473413e+00 8.88236880e-01 2.60074645e-01 -1.38588631e+00 -3.35744321e-01 -8.73292387e-02 -1.29308975e+00 -2.11015463e-01 3.25219244e-01 -1.61247492e-01 -5.08920908e-01 1.39904988e+00 3.12738836e-01 5.32317519e-01 3.40438634e-02 9.26908791e-01 3.51018876e-01 8.57035816e-01 -5.33303013e-03 -5.76212645e-01 9.71011877e-01 -4.45581347e-01 -1.21966434e+00 2.04233393e-01 4.60178673e-01 -7.25057065e-01 6.23417199e-01 1.91185728e-01 -6.75066650e-01 -7.40056992e-01 -5.41640520e-01 5.85110746e-02 -6.59487844e-02 4.42952216e-01 5.07243454e-01 2.42516145e-01 -2.55160213e-01 1.03682987e-01 -8.39653611e-01 -5.51116429e-02 2.90864050e-01 1.59697592e-01 -4.66299117e-01 -6.45245433e-01 -1.30554605e+00 5.16035557e-01 3.00950527e-01 6.05794132e-01 -6.36066735e-01 -6.11265481e-01 -5.34197271e-01 -7.96290413e-02 5.74889183e-01 -4.34963018e-01 5.45699477e-01 -1.84452459e-01 -6.66202426e-01 2.69505605e-02 -8.11367750e-01 1.12754935e-02 3.91123414e-01 1.97831094e-01 -1.14802551e+00 -3.23685884e-01 5.95048130e-01 -1.58362389e-01 8.21914613e-01 -1.06547487e+00 -1.01046634e+00 -4.30813164e-01 -5.04581690e-01 -1.22595020e-01 -1.98905185e-01 -3.25874686e-01 -3.54524881e-01 -5.83717644e-01 4.51397568e-01 -8.36261809e-01 -2.18890622e-01 -3.16013098e-01 -3.10015202e-01 -2.64270872e-01 1.04601681e+00 -9.07046318e-01 1.64771962e+00 -2.28252649e+00 -4.64491732e-02 3.80327523e-01 2.05113754e-01 4.33649644e-02 -3.11419349e-02 8.80673170e-01 7.68714249e-02 -2.19016656e-01 -4.72115427e-01 -2.23988593e-01 -1.02537259e-01 6.11472607e-01 -5.32462299e-01 5.03208458e-01 8.51417482e-02 7.03254580e-01 -6.21022224e-01 -4.23779488e-01 5.08392632e-01 4.44698244e-01 -1.78436086e-01 -6.50988007e-03 3.18052948e-01 6.26341820e-01 -8.21483016e-01 4.46764588e-01 1.14881384e+00 3.59446518e-02 -1.65064752e-01 -1.42483979e-01 -5.65984726e-01 -1.71621203e-01 -1.49285173e+00 1.43204570e+00 -1.53894380e-01 3.44145030e-01 -2.25754417e-02 -9.71565068e-01 1.19565070e+00 2.31607124e-01 8.72293413e-01 -1.15524828e+00 -1.93918511e-01 2.03079030e-01 -3.26756090e-01 -1.16633475e+00 5.94885945e-01 -4.60203290e-02 3.47592793e-02 3.61067593e-01 -7.70350754e-01 7.77383864e-01 1.70353919e-01 2.17126176e-01 9.89346027e-01 -1.57404423e-01 -3.62638146e-01 1.47743896e-01 7.56029606e-01 3.62560451e-01 1.12481618e+00 2.07805231e-01 -1.36341020e-01 3.59944969e-01 2.37940654e-01 -5.77803195e-01 -1.10896909e+00 -6.94243431e-01 -8.48136395e-02 6.03313565e-01 1.35215819e-01 -1.57285869e-01 -1.31192267e-01 -2.29968384e-01 2.94447392e-01 4.80324298e-01 -4.81417358e-01 -2.60567516e-01 -6.32378995e-01 -8.17747474e-01 2.77325183e-01 3.49102259e-01 6.50188029e-01 -6.51216745e-01 1.62319630e-01 3.70428443e-01 -8.81052732e-01 -1.02913463e+00 -3.09528530e-01 -6.82956755e-01 -9.88705873e-01 -1.14903474e+00 -5.75299025e-01 -5.51070452e-01 7.33271897e-01 1.19356692e+00 3.89659286e-01 1.60175875e-01 -9.72122177e-02 -4.15988341e-02 -3.84074420e-01 2.20129713e-01 -2.48742457e-02 2.32750140e-02 1.46804243e-01 6.38971329e-01 7.46945739e-01 -7.31601834e-01 -3.97231132e-01 7.16918945e-01 -1.00647533e+00 2.35464856e-01 9.10185695e-01 6.55894935e-01 4.45453674e-01 6.92837477e-01 6.02564752e-01 -4.07065541e-01 7.24133313e-01 -1.01372719e+00 -4.55134898e-01 7.48982728e-02 -5.56722343e-01 6.70194551e-02 3.28174561e-01 -1.69378698e-01 -1.15584826e+00 5.29559776e-02 1.38063550e-01 -5.76567650e-01 -1.75459459e-01 7.54401684e-01 -6.31970465e-01 5.24092495e-01 6.32092431e-02 8.04906011e-01 3.09592217e-01 -8.49501789e-01 2.42474838e-03 9.33061063e-01 3.25044572e-01 -3.88956964e-01 9.56063807e-01 8.01081479e-01 2.75425881e-01 -9.64753926e-01 -4.93508279e-01 -7.80232847e-01 -7.42375255e-01 -1.29692197e-01 4.94164407e-01 -1.08121693e+00 -8.03412199e-01 2.24932730e-01 -9.57644165e-01 2.08758473e-01 -7.96092302e-02 9.32074428e-01 -4.06431824e-01 4.89658803e-01 -2.93798327e-01 -9.70597565e-01 2.70884454e-01 -9.25826013e-01 8.20549190e-01 -2.90553629e-01 2.76926428e-01 -7.36357868e-01 2.63835728e-01 5.67622066e-01 5.14970422e-01 1.83642954e-01 9.09418225e-01 8.99385586e-02 -9.56309855e-01 -2.22242519e-01 -4.99243885e-01 6.03890195e-02 3.56551111e-01 1.02682877e-02 -5.38561881e-01 -7.83758983e-02 1.23473750e-02 4.81955439e-01 9.55693722e-01 4.19507802e-01 8.22701216e-01 -5.25449395e-01 -3.61281931e-01 3.23145479e-01 1.13370025e+00 3.58175300e-02 9.54413712e-01 1.75240740e-01 7.24130213e-01 9.80886757e-01 9.62877631e-01 6.89001739e-01 9.13728893e-01 4.99603271e-01 3.81261140e-01 -1.54099974e-03 -6.42919913e-03 -5.29146552e-01 1.69533402e-01 1.22450292e+00 -7.42716268e-02 1.61014840e-01 -8.49445224e-01 7.98839271e-01 -2.43059850e+00 -1.61647677e+00 -1.01983964e+00 2.09409165e+00 1.41075537e-01 -1.43360987e-01 3.54935616e-01 5.93692005e-01 9.66392338e-01 8.27208906e-03 -5.06605208e-01 3.15264910e-01 -2.65649170e-01 -7.05745578e-01 4.78615314e-01 4.20789719e-01 -7.00247884e-01 5.19958973e-01 6.15124130e+00 9.47169125e-01 -6.08390212e-01 -7.03363074e-03 1.66587695e-01 1.92593187e-01 -5.41574895e-01 2.99627781e-01 -7.88515329e-01 9.22471404e-01 8.41704965e-01 -1.87816754e-01 4.56351191e-01 4.80918229e-01 1.08926165e+00 1.04589416e-02 -4.28607225e-01 1.03761625e+00 -2.90941447e-01 -1.11266196e+00 2.92262971e-01 6.36896312e-01 7.24897146e-01 -1.84626564e-01 3.65248695e-02 3.00773084e-01 1.34755090e-01 -8.25640619e-01 2.15780526e-01 1.23038399e+00 5.73494792e-01 -1.00363982e+00 6.32824898e-01 8.97990108e-01 -1.46705091e+00 -3.42301071e-01 -6.91591144e-01 -2.89391845e-01 5.89362085e-01 9.71536875e-01 -3.21634829e-01 8.22524488e-01 3.42061043e-01 1.13371706e+00 -4.33156282e-01 1.24082184e+00 2.16927931e-01 6.15788937e-01 -5.60677461e-02 5.91244400e-01 1.17518632e-02 -6.81411147e-01 4.11028564e-01 7.17782974e-01 5.52239597e-01 2.36823261e-01 4.21863228e-01 7.73858726e-01 3.87200147e-01 -2.82198424e-03 -7.80480504e-01 2.57567674e-01 5.98936856e-01 8.90556395e-01 5.26576117e-02 -1.63119808e-01 -5.68711460e-01 3.87771368e-01 -1.34547919e-01 5.92652321e-01 -8.57837021e-01 -1.93978205e-01 6.78658128e-01 4.00468260e-01 1.96639746e-01 -7.38112926e-01 -2.25356668e-01 -1.14378440e+00 4.36081946e-01 -4.70854193e-01 2.93333292e-01 -7.05893099e-01 -1.36436343e+00 1.61039338e-01 1.72350332e-01 -1.82846212e+00 -4.14417200e-02 1.10219754e-01 -5.21116555e-01 9.81199801e-01 -1.60460925e+00 -1.21011007e+00 -4.53105390e-01 1.19361699e+00 3.37130308e-01 -2.79823005e-01 3.64107013e-01 8.36555898e-01 -9.34457242e-01 2.12719408e-03 5.50758660e-01 1.61642894e-01 2.05374941e-01 -2.98622102e-01 2.72026688e-01 1.00593519e+00 -3.78215820e-01 6.51101410e-01 4.85757947e-01 -9.71003354e-01 -1.63707507e+00 -1.28789985e+00 1.30685043e+00 -1.86962143e-01 5.10133684e-01 -9.13324505e-02 -1.01758909e+00 5.92021644e-01 -4.66516405e-01 5.92935905e-02 6.77026570e-01 1.17868006e-01 -1.59884587e-01 -5.70003688e-01 -1.01852763e+00 3.22125405e-01 1.05358922e+00 -5.23603201e-01 -4.42457557e-01 1.57493025e-01 6.92751348e-01 3.53996783e-01 -6.51514292e-01 4.09205645e-01 4.24394161e-01 -7.21841574e-01 8.18196177e-01 -6.59264803e-01 -2.32758597e-02 -7.88142800e-01 -4.46143359e-01 -8.25092912e-01 -6.54756665e-01 -3.22496831e-01 -6.84123933e-02 1.41277552e+00 1.66949570e-01 -4.95500624e-01 7.61517227e-01 8.90774190e-01 -1.44399554e-02 -3.50678340e-02 -1.29197526e+00 -8.37086082e-01 -2.96352714e-01 -7.56108403e-01 1.11632001e+00 9.00819123e-01 -2.43096292e-01 2.68570274e-01 -1.10743225e+00 2.74253100e-01 8.02453220e-01 1.90917626e-01 1.05782557e+00 -1.51272786e+00 3.75449657e-01 8.06407332e-02 -3.00585508e-01 -1.09491026e+00 1.18773349e-01 -6.29596829e-01 -2.58388847e-01 -1.53334320e+00 2.62395024e-01 -7.41542757e-01 -2.33695164e-01 3.59729826e-01 -1.23031614e-02 -3.08234990e-01 -1.89354986e-01 8.09085846e-01 -5.78395307e-01 1.07002854e+00 1.21225297e+00 -3.16931546e-01 -2.59033740e-01 3.82866770e-01 -5.78846395e-01 1.25162631e-01 6.72842681e-01 -6.13121867e-01 -5.36845267e-01 -6.20375991e-01 6.03234768e-02 5.54955244e-01 4.16740835e-01 -9.56808925e-01 4.85883981e-01 -7.90709615e-01 1.26878664e-01 -1.18838751e+00 3.43866229e-01 -1.39001155e+00 5.70986032e-01 2.81279981e-01 3.56892087e-02 2.79013157e-01 -8.52070600e-02 9.10608768e-01 -2.95382619e-01 3.33313763e-01 -6.62735105e-02 1.60119817e-01 -7.99543560e-01 7.34045029e-01 -5.97868741e-01 -1.24758393e-01 9.36966181e-01 -3.40651453e-01 -2.18752980e-01 -3.36171687e-01 -7.44302213e-01 5.81024945e-01 7.19598979e-02 5.97337902e-01 9.67601895e-01 -1.92969525e+00 -1.02571845e+00 5.35517097e-01 2.35413104e-01 -2.72681057e-01 9.09265578e-01 1.12123513e+00 5.28665865e-03 6.18246257e-01 -1.65173247e-01 -7.21888125e-01 -1.03428137e+00 8.34866464e-01 -2.03448132e-01 9.81770009e-02 -6.04623258e-01 -2.62042165e-01 -2.60169744e-01 -3.69340628e-01 -8.62492025e-02 -8.17496628e-02 -2.50495583e-01 3.10672168e-02 6.91518843e-01 1.11161768e+00 -5.01788072e-02 -1.25220263e+00 -2.69964129e-01 5.77915549e-01 3.31638783e-01 -6.98846504e-02 1.36895382e+00 -8.36580694e-01 -1.62215590e-01 4.91223097e-01 1.22478461e+00 -1.01828665e-01 -1.30700684e+00 -6.13637984e-01 5.34962490e-02 -9.89112675e-01 -3.93835828e-02 -1.78961232e-01 -1.17167687e+00 1.02496588e+00 4.64609444e-01 6.28707260e-02 8.76403570e-01 -5.91687143e-01 1.04324543e+00 2.03164771e-01 5.02646029e-01 -9.80285585e-01 -3.48166376e-01 4.38132674e-01 6.25649095e-01 -1.16469657e+00 4.63486016e-02 -5.47275007e-01 -6.87828541e-01 8.22479248e-01 2.90203035e-01 8.75941738e-02 8.73196781e-01 -1.77052960e-01 -3.18053186e-01 -2.74525732e-02 -7.81619787e-01 -6.90511763e-01 1.31109253e-01 5.75397074e-01 -2.45890722e-01 2.31441513e-01 -2.73700118e-01 4.63098258e-01 6.33375049e-02 2.85741538e-01 3.72543454e-01 9.41607654e-01 -6.95865214e-01 -1.20359719e+00 -4.93744373e-01 3.43988746e-01 1.61672756e-01 3.54253769e-01 2.22130209e-01 5.82462549e-01 2.85457790e-01 1.52163422e+00 -1.05321087e-01 -6.50145948e-01 6.01817191e-01 -8.19030404e-02 -3.20154727e-01 -1.07570268e-01 4.42299694e-01 -2.83682533e-02 -1.63071722e-01 -4.68555033e-01 -5.03776610e-01 -7.72763968e-01 -1.13460159e+00 -9.68005776e-01 -2.37424776e-01 4.58641469e-01 4.97556120e-01 1.02077222e+00 6.06842816e-01 3.09594691e-01 1.16830170e+00 -2.90678352e-01 -2.85471618e-01 -7.23198235e-01 -8.39322627e-01 6.50531411e-01 5.08437514e-01 -9.01845813e-01 -2.48947337e-01 -1.13027304e-01]
[6.55159854888916, 2.0862841606140137]
7b356c16-c684-419c-9bba-b67fd24e213c
attentive-memory-networks-efficient-machine
1712.07229
null
http://arxiv.org/abs/1712.07229v1
http://arxiv.org/pdf/1712.07229v1.pdf
Attentive Memory Networks: Efficient Machine Reading for Conversational Search
Recent advances in conversational systems have changed the search paradigm. Traditionally, a user poses a query to a search engine that returns an answer based on its index, possibly leveraging external knowledge bases and conditioning the response on earlier interactions in the search session. In a natural conversation, there is an additional source of information to take into account: utterances produced earlier in a conversation can also be referred to and a conversational IR system has to keep track of information conveyed by the user during the conversation, even if it is implicit. We argue that the process of building a representation of the conversation can be framed as a machine reading task, where an automated system is presented with a number of statements about which it should answer questions. The questions should be answered solely by referring to the statements provided, without consulting external knowledge. The time is right for the information retrieval community to embrace this task, both as a stand-alone task and integrated in a broader conversational search setting. In this paper, we focus on machine reading as a stand-alone task and present the Attentive Memory Network (AMN), an end-to-end trainable machine reading algorithm. Its key contribution is in efficiency, achieved by having an hierarchical input encoder, iterating over the input only once. Speed is an important requirement in the setting of conversational search, as gaps between conversational turns have a detrimental effect on naturalness. On 20 datasets commonly used for evaluating machine reading algorithms we show that the AMN achieves performance comparable to the state-of-the-art models, while using considerably fewer computations.
['Maarten de Rijke', 'Tom Kenter']
2017-12-19
null
null
null
null
['conversational-search']
['natural-language-processing']
[ 6.87916756e-01 6.82487130e-01 -1.33751750e-01 -4.88411993e-01 -9.37212586e-01 -8.55190575e-01 1.00787044e+00 3.35337162e-01 -5.84667563e-01 6.09429598e-01 5.22969365e-01 -6.77736521e-01 -1.81035623e-01 -8.66824269e-01 -4.83558178e-01 -2.86121517e-01 2.20523119e-01 1.13097703e+00 2.11945325e-01 -6.15077615e-01 2.24497586e-01 6.22307137e-02 -1.37347698e+00 4.44582075e-01 4.75233942e-01 8.32397461e-01 3.25560093e-01 1.00519001e+00 -3.43360603e-01 1.16252494e+00 -5.06050944e-01 -5.71059585e-01 -1.69101894e-01 -7.02080369e-01 -1.95001602e+00 -4.26575057e-02 -1.34826396e-02 -2.92954177e-01 -4.78704758e-02 6.64088249e-01 2.77455300e-01 4.69904602e-01 4.74516392e-01 -8.39165032e-01 -3.80917013e-01 7.30458856e-01 1.62316158e-01 1.96638301e-01 8.51031005e-01 -6.73796842e-03 1.46962476e+00 -6.30395055e-01 5.83562493e-01 1.32345009e+00 1.08497620e-01 7.10708439e-01 -1.25288332e+00 -1.07481323e-01 3.03594291e-01 1.72257096e-01 -7.28395343e-01 -5.01544595e-01 5.53422809e-01 -3.52649182e-01 1.19042647e+00 6.95494652e-01 3.13234746e-01 9.26769435e-01 -1.54981658e-01 7.61898100e-01 7.54392922e-01 -7.90818870e-01 2.52532452e-01 4.61119354e-01 5.87065101e-01 4.69802201e-01 -5.47282934e-01 -7.89512545e-02 -5.94848037e-01 -3.07659805e-01 2.70492226e-01 -1.88928127e-01 -4.84655052e-01 -1.21060595e-01 -8.51655304e-01 1.07086527e+00 4.79115874e-01 5.33978581e-01 -5.03094316e-01 -2.24285051e-01 2.53130347e-01 7.94907629e-01 4.49684530e-01 6.86420619e-01 -4.80288088e-01 -4.40422773e-01 -4.91101742e-01 2.43830249e-01 1.48947930e+00 5.89460731e-01 6.91285610e-01 -9.86840069e-01 -1.92700669e-01 9.19897854e-01 9.09233093e-02 2.70482242e-01 4.96070296e-01 -9.14224982e-01 4.14364576e-01 6.34390235e-01 2.48655081e-01 -9.28831756e-01 -2.30897710e-01 -2.15826228e-01 -4.72299039e-01 -2.91171432e-01 3.99674714e-01 -1.56297132e-01 -4.66870368e-01 2.03395271e+00 3.24702829e-01 -2.81373262e-01 3.63839567e-01 7.64990926e-01 7.19015896e-01 7.69230902e-01 -6.34500161e-02 -4.84962761e-01 1.39856207e+00 -1.01943803e+00 -5.16992092e-01 -4.34621245e-01 6.99489415e-01 -8.26809764e-01 9.56879258e-01 2.67132938e-01 -1.27936482e+00 -2.99560368e-01 -5.97622931e-01 -3.83494049e-01 -3.98506880e-01 -4.56928074e-01 3.12478006e-01 1.06766060e-01 -1.25732732e+00 3.58933061e-01 -4.17908281e-01 -6.20160758e-01 -3.22443664e-01 3.52683008e-01 -1.26455024e-01 5.28567433e-02 -1.47222793e+00 1.18938029e+00 1.41409948e-01 2.40152955e-01 -5.06674349e-01 -1.43382624e-01 -6.36052012e-01 3.47998828e-01 6.45286858e-01 -7.40002573e-01 2.10611606e+00 -1.25662935e+00 -1.71356833e+00 1.05905497e+00 -6.34638131e-01 -5.46018183e-01 4.83981222e-01 -2.36968398e-01 -1.42912641e-01 1.74079016e-01 4.52785604e-02 5.24673760e-01 4.53921884e-01 -1.06486773e+00 -5.89655638e-01 -3.35141569e-01 8.52212250e-01 4.80260670e-01 5.24970703e-02 1.81077078e-01 -5.27302265e-01 1.03175901e-01 -7.08331168e-02 -1.06866074e+00 8.76087695e-02 -4.38773811e-01 -2.73165286e-01 -7.69811273e-01 6.14252567e-01 -6.01551056e-01 1.29670262e+00 -1.70841670e+00 4.36924458e-01 2.02201009e-01 1.87964663e-01 2.88768411e-01 -1.30636707e-01 9.21065032e-01 5.70511073e-03 6.40237406e-02 -4.72392738e-02 -2.91413248e-01 1.34163156e-01 9.43259746e-02 -6.52109325e-01 -1.51601732e-01 8.84975772e-03 9.34200168e-01 -1.00088668e+00 -3.26776326e-01 -3.25653069e-02 2.08473757e-01 -4.60414559e-01 6.26224160e-01 -6.23098016e-01 2.74433076e-01 -5.58498979e-01 -1.26796097e-01 -2.31359396e-02 -6.74757719e-01 1.70910627e-01 2.82550216e-01 6.44815415e-02 9.21811104e-01 -7.99338639e-01 1.43388391e+00 -9.01523411e-01 6.66535258e-01 3.07060659e-01 -9.68859315e-01 5.77349007e-01 6.10210299e-01 -3.97679359e-02 -8.02315712e-01 1.32361472e-01 3.33778863e-03 -7.33738393e-02 -5.29050529e-01 3.83300960e-01 -2.20275596e-01 -2.62091421e-02 1.15129471e+00 -9.07921642e-02 -2.77825929e-02 8.33971873e-02 5.13713896e-01 1.15615427e+00 -4.34587002e-01 1.83977813e-01 -8.21056217e-03 8.16743851e-01 4.81382124e-02 -1.08162940e-01 1.05654597e+00 2.20484927e-01 1.19443916e-01 5.23246586e-01 -3.54711771e-01 -5.95030785e-01 -6.43218696e-01 7.69410282e-02 1.57183337e+00 2.11371303e-01 -3.47787350e-01 -7.94619799e-01 -6.18961930e-01 -4.22331810e-01 8.24755788e-01 -6.40561104e-01 -1.56954959e-01 -6.35251343e-01 -1.54814154e-01 2.07483768e-01 1.76109031e-01 4.56952661e-01 -1.42885947e+00 -7.94112265e-01 3.72201920e-01 -6.64610445e-01 -9.22954679e-01 -4.39723432e-01 2.96543300e-01 -5.18051207e-01 -1.02086473e+00 -4.38399404e-01 -8.80131364e-01 3.51936787e-01 2.25048214e-01 1.54102921e+00 4.82173920e-01 5.75280190e-02 7.81972408e-01 -3.30513746e-01 -3.63490164e-01 -7.52104998e-01 3.26642424e-01 -5.29158652e-01 8.21716040e-02 4.69166934e-01 -4.04965103e-01 -5.26422203e-01 3.31601322e-01 -9.06924605e-01 3.85484308e-01 4.60232019e-01 9.42246616e-01 -1.38714150e-01 -2.91052908e-01 5.85379779e-01 -1.12635589e+00 1.14878011e+00 -4.96144295e-01 -3.82641673e-01 4.72647429e-01 -4.19320673e-01 3.78877342e-01 3.11893284e-01 -2.42537171e-01 -1.32085967e+00 -2.24330112e-01 -1.63319066e-01 3.27753961e-01 -1.13757893e-01 7.88620293e-01 1.27954334e-01 3.09653938e-01 6.95002437e-01 3.83775473e-01 2.23437160e-01 -4.05481517e-01 4.31171566e-01 9.48356450e-01 4.02943313e-01 -6.57169878e-01 4.36132520e-01 1.84515771e-02 -4.57316816e-01 -9.40273762e-01 -1.13555467e+00 -9.10676956e-01 -4.28445756e-01 -2.26184562e-01 7.50182867e-01 -4.63848025e-01 -1.08895588e+00 9.18207318e-02 -1.48596549e+00 -4.81914520e-01 4.89805527e-02 5.94955608e-02 -5.85278630e-01 8.57545957e-02 -5.61629891e-01 -1.01712346e+00 -4.32367265e-01 -1.02186275e+00 9.97973800e-01 6.92554563e-02 -7.55328774e-01 -1.14777017e+00 2.53194958e-01 7.28483558e-01 6.38146460e-01 -3.94468307e-01 1.18518615e+00 -1.21007860e+00 -5.04775882e-01 -1.93216816e-01 -1.19200185e-01 1.70876428e-01 1.06614269e-01 -5.85967541e-01 -1.26526701e+00 -1.34163827e-01 1.92526743e-01 -5.52327335e-01 5.65444350e-01 -1.85646713e-01 7.56538689e-01 -7.00646281e-01 -2.28860423e-01 -3.22481573e-01 9.91496921e-01 2.61194885e-01 3.66327107e-01 1.79430753e-01 -6.33013397e-02 1.06137300e+00 1.42187357e-01 -8.31740499e-02 5.01213372e-01 7.82544076e-01 1.46832556e-01 2.01530941e-02 1.90529779e-01 -1.80703536e-01 9.90383700e-02 6.13224387e-01 1.85124934e-01 -5.41797876e-01 -9.28052902e-01 5.28517902e-01 -2.02984691e+00 -1.17544484e+00 2.80573756e-01 2.30811238e+00 1.28813195e+00 1.82527512e-01 5.74251935e-02 -7.20558837e-02 5.31040251e-01 8.76374096e-02 -5.41195035e-01 -6.64970577e-01 3.96139950e-01 2.99540192e-01 -3.24271739e-01 1.29259932e+00 -8.07616115e-01 7.78140306e-01 5.88657093e+00 1.83676332e-01 -1.18786144e+00 -1.02893457e-01 6.73860729e-01 9.65184793e-02 -2.94095457e-01 1.77520722e-01 -5.88394225e-01 2.22878411e-01 1.20051098e+00 -3.78237426e-01 6.50834441e-01 6.27740741e-01 1.28652498e-01 -3.51923168e-01 -1.70326805e+00 6.79283142e-01 1.22525789e-01 -1.21814466e+00 3.48447785e-02 -1.19300105e-01 1.43923849e-01 1.48514728e-03 -2.46576831e-01 5.76816022e-01 3.85006160e-01 -1.12037206e+00 1.64293095e-01 5.46773732e-01 9.19631273e-02 -4.47606266e-01 8.33178699e-01 8.47692490e-01 -7.33147502e-01 -6.59128372e-03 1.10426925e-01 -3.50178391e-01 2.45496467e-01 4.64020818e-02 -1.27932477e+00 2.31407806e-01 2.78160810e-01 -2.04817578e-03 -2.93670028e-01 5.50408185e-01 -1.08929969e-01 5.86980999e-01 -3.27534616e-01 -4.42250162e-01 4.59258169e-01 -6.57864809e-02 4.06785578e-01 1.25024152e+00 -2.21531585e-01 4.79284674e-01 2.29171589e-01 6.87533855e-01 -2.47437954e-01 1.86036676e-01 -5.66430449e-01 -4.39818539e-02 3.12148631e-01 1.01150346e+00 -2.83761054e-01 -5.33085227e-01 -3.67314845e-01 9.79792893e-01 5.05514264e-01 4.16120708e-01 -2.44049594e-01 -3.01260173e-01 3.49615455e-01 2.79101189e-02 9.43590105e-02 1.57931820e-01 2.04914421e-01 -9.53198731e-01 2.19319433e-01 -1.27048230e+00 5.01570880e-01 -8.62522900e-01 -1.04174876e+00 8.08232427e-01 -7.12499395e-02 -5.94540119e-01 -1.01407647e+00 -2.88175017e-01 -6.38415158e-01 1.07656038e+00 -1.40957808e+00 -7.48328805e-01 -2.12714911e-01 4.34443086e-01 8.55251908e-01 3.24988663e-01 1.16519165e+00 -1.80850513e-02 -1.29344627e-01 3.04991305e-01 -3.59951973e-01 2.74039805e-01 6.21593952e-01 -1.11782861e+00 1.19475592e-02 4.58356321e-01 4.04062688e-01 1.04015529e+00 8.16503704e-01 -1.97950438e-01 -1.38815773e+00 -5.41719317e-01 1.71948624e+00 -7.41630971e-01 7.68181205e-01 -3.05499077e-01 -1.17526376e+00 8.49404216e-01 6.28539920e-01 -6.77561462e-01 6.66281164e-01 5.90842903e-01 -1.73810720e-01 1.00684240e-01 -7.19685674e-01 5.13684392e-01 6.80833161e-01 -1.08299518e+00 -1.20914626e+00 5.18223882e-01 9.47390914e-01 -3.32374454e-01 -4.07651603e-01 2.62452252e-02 3.83128494e-01 -8.12913895e-01 7.62461960e-01 -8.75084281e-01 3.58077198e-01 1.40626714e-01 9.65191573e-02 -1.19275880e+00 8.47520977e-02 -9.55959022e-01 2.02468093e-02 1.06213319e+00 8.46669495e-01 -7.27350652e-01 4.92079884e-01 1.30906260e+00 2.03604534e-01 -6.38658285e-01 -8.33892703e-01 -2.42522672e-01 -3.72080645e-03 -3.35087121e-01 3.82031351e-01 6.09499335e-01 6.63781822e-01 1.30328929e+00 -1.72673538e-01 -1.28809293e-03 3.65807414e-02 4.16503251e-01 8.01258385e-01 -1.41503859e+00 -5.81140697e-01 -4.86836344e-01 2.50331819e-01 -1.65968740e+00 2.18403265e-01 -7.62523293e-01 3.60770285e-01 -1.38006413e+00 2.86432117e-01 -1.52011186e-01 -7.01611042e-02 3.83804888e-01 -1.38721168e-01 -3.27773064e-01 1.40780255e-01 2.75190443e-01 -7.39874244e-01 2.91833490e-01 1.06826019e+00 -2.31296375e-01 -2.91946232e-01 4.18883324e-01 -7.24133611e-01 7.03550100e-01 5.62199533e-01 -1.63511187e-01 -7.38478541e-01 -4.90389019e-01 5.12864769e-01 6.34104848e-01 3.26096237e-01 -4.97073174e-01 8.92955780e-01 -5.09155635e-03 -3.08396399e-01 -3.82238716e-01 5.78197241e-01 -7.33170748e-01 -7.50568956e-02 1.91606432e-01 -1.21225739e+00 1.26730070e-01 -6.38997778e-02 5.19002616e-01 -4.82072204e-01 -4.14444774e-01 4.26767528e-01 -2.83065259e-01 -3.52620125e-01 -7.84449279e-02 -5.32952666e-01 1.80762023e-01 5.33249259e-01 9.59338620e-02 -2.02846572e-01 -1.04126596e+00 -9.52555060e-01 4.49073941e-01 8.16563517e-02 4.42005873e-01 2.72519231e-01 -7.17704296e-01 -5.39863467e-01 2.05549803e-02 1.45146877e-01 -3.89691209e-03 -9.92601439e-02 5.40317237e-01 -4.42238152e-02 1.03051460e+00 4.88073528e-01 -4.92582887e-01 -1.47756028e+00 3.50442946e-01 4.15098995e-01 -5.90840697e-01 -3.30340445e-01 9.24767137e-01 3.37166876e-01 -5.61796010e-01 6.02320015e-01 -2.48430297e-01 -3.76471460e-01 2.64300644e-01 7.67503023e-01 -1.13842808e-01 2.82420963e-01 -3.69157046e-01 -1.22628592e-01 3.17544639e-01 -2.90229321e-01 -5.06970525e-01 1.19096434e+00 -3.31005514e-01 -2.69155413e-01 5.86946070e-01 1.32102513e+00 -2.04799265e-01 -4.89866346e-01 -8.48789811e-01 1.84480712e-01 -1.20629393e-01 2.81570922e-03 -1.24884295e+00 -3.49320799e-01 7.63367891e-01 7.54064992e-02 8.66129220e-01 8.50184083e-01 4.55889046e-01 8.27020943e-01 1.12273943e+00 3.17265332e-01 -1.01974940e+00 1.99193388e-01 8.48297477e-01 1.18985534e+00 -1.40698981e+00 -4.05215651e-01 -1.89403862e-01 -5.24278343e-01 1.14855468e+00 3.13305050e-01 3.57717037e-01 4.29189295e-01 -1.99999049e-01 2.80802906e-01 -4.97873485e-01 -1.32365060e+00 -1.50395945e-01 3.60330939e-01 1.15214303e-01 6.16114259e-01 -1.96165651e-01 -1.72641963e-01 2.36936495e-01 -3.74715388e-01 -8.09014067e-02 2.06852227e-01 9.38834369e-01 -5.70667863e-01 -1.29143262e+00 1.46783320e-02 2.02186346e-01 -4.24897105e-01 -3.10360491e-01 -9.27774131e-01 5.29838979e-01 -5.74957788e-01 1.43246794e+00 8.56095999e-02 9.87183489e-03 2.46761799e-01 5.41850746e-01 8.03444535e-02 -8.22297871e-01 -8.39428902e-01 -3.01151574e-01 6.13072574e-01 -5.29721737e-01 -4.42010492e-01 -3.62752229e-01 -1.03805637e+00 -1.35593325e-01 -4.66811717e-01 8.65700722e-01 5.41779220e-01 1.33996797e+00 4.22254950e-01 1.66162252e-01 7.01938629e-01 -3.78758818e-01 -7.46902347e-01 -9.42640662e-01 2.70298541e-01 4.78587061e-01 6.80456161e-01 -2.13799670e-01 -5.34917772e-01 7.90709481e-02]
[12.174091339111328, 7.840701580047607]
f6f13b06-8753-4522-8707-b36ec5f9fbb7
protecting-the-intellectual-properties-of
2104.09203
null
https://arxiv.org/abs/2104.09203v1
https://arxiv.org/pdf/2104.09203v1.pdf
Protecting the Intellectual Properties of Deep Neural Networks with an Additional Class and Steganographic Images
Recently, the research on protecting the intellectual properties (IP) of deep neural networks (DNN) has attracted serious concerns. A number of DNN copyright protection methods have been proposed. However, most of the existing watermarking methods focus on verifying the copyright of the model, which do not support the authentication and management of users' fingerprints, thus can not satisfy the requirements of commercial copyright protection. In addition, the query modification attack which was proposed recently can invalidate most of the existing backdoor-based watermarking methods. To address these challenges, in this paper, we propose a method to protect the intellectual properties of DNN models by using an additional class and steganographic images. Specifically, we use a set of watermark key samples to embed an additional class into the DNN, so that the watermarked DNN will classify the watermark key sample as the predefined additional class in the copyright verification stage. We adopt the least significant bit (LSB) image steganography to embed users' fingerprints into watermark key images. Each user will be assigned with a unique fingerprint image so that the user's identity can be authenticated later. Experimental results demonstrate that, the proposed method can protect the copyright of DNN models effectively. On Fashion-MNIST and CIFAR-10 datasets, the proposed method can obtain 100% watermark accuracy and 100% fingerprint authentication success rate. In addition, the proposed method is demonstrated to be robust to the model fine-tuning attack, model pruning attack, and the query modification attack. Compared with three existing watermarking methods (the logo-based, noise-based, and adversarial frontier stitching watermarking methods), the proposed method has better performance on watermark accuracy and robustness against the query modification attack.
['Weiqiang Liu', 'Jian Wang', 'Mingfu Xue', 'Shichang Sun']
2021-04-19
null
null
null
null
['image-steganography']
['computer-vision']
[ 6.51650190e-01 -3.26225579e-01 -5.12374461e-01 1.44843921e-01 8.09487998e-02 -5.99984109e-01 2.91213304e-01 -3.00012112e-01 -6.65303290e-01 6.51073635e-01 -3.22204411e-01 -5.02523422e-01 1.18316654e-02 -8.62082243e-01 -6.17883980e-01 -6.98555410e-01 1.37620538e-01 -3.96836251e-01 6.82957351e-01 -1.90557949e-02 6.19508922e-01 5.51697493e-01 -1.25162697e+00 -2.40025539e-02 4.93152678e-01 1.17047107e+00 1.52524188e-01 2.71687448e-01 -1.91814810e-01 4.22412574e-01 -9.86629367e-01 -3.90822500e-01 6.65652931e-01 -2.99739063e-01 -3.43743414e-01 -2.50412613e-01 3.28798741e-01 -8.32735240e-01 -8.79080117e-01 1.66825461e+00 5.62434316e-01 -3.49428147e-01 1.13770641e-01 -1.70962763e+00 -7.75236309e-01 5.94487071e-01 -7.46106148e-01 2.37108335e-01 -2.33315825e-01 -1.36919037e-01 5.53934395e-01 -3.50158781e-01 7.17721701e-01 1.16942370e+00 6.74377322e-01 7.93113649e-01 -9.00649130e-01 -1.69570327e+00 -2.02807337e-02 3.90838265e-01 -1.48397195e+00 -4.50628102e-01 9.55190599e-01 -5.35413213e-02 2.48159990e-01 2.21306190e-01 5.51539421e-01 8.46301615e-01 2.42162302e-01 4.77011949e-01 9.20530200e-01 -4.24083561e-01 2.27022246e-02 2.00227574e-01 -2.43180897e-02 6.26775444e-01 9.89952385e-01 4.47029203e-01 -2.87701666e-01 -3.07973832e-01 1.14620912e+00 2.31761038e-01 -7.37329423e-01 -2.44041175e-01 -1.16753125e+00 7.15278387e-01 3.00254434e-01 3.28053623e-01 1.10979497e-01 5.06513119e-01 4.26540613e-01 5.14839768e-01 -2.46364683e-01 -1.91374093e-01 -4.49909508e-01 3.74571145e-01 -8.76581490e-01 9.61194485e-02 6.51026547e-01 1.05458319e+00 5.06894648e-01 2.22398296e-01 3.73446643e-01 3.18778545e-01 9.97435331e-01 6.05077088e-01 8.69164407e-01 -6.96500361e-01 4.74317849e-01 2.61382192e-01 -2.35297114e-01 -1.38811243e+00 -1.37220994e-02 -3.91807377e-01 -8.13978255e-01 3.20367575e-01 2.10381970e-01 -2.95622528e-01 -9.73127723e-01 1.71487474e+00 1.87386438e-01 5.67418754e-01 3.84342611e-01 4.47194487e-01 8.42025340e-01 7.27360785e-01 -9.89654213e-02 -6.98136836e-02 1.51176310e+00 -7.24573672e-01 -1.02470517e+00 -1.46750897e-01 2.66553730e-01 -1.02730191e+00 1.88795194e-01 1.98680326e-01 -6.00922227e-01 -6.80689096e-01 -1.70721161e+00 3.01739007e-01 -4.08277065e-01 -6.05600514e-02 3.33194017e-01 1.55437207e+00 -7.27775335e-01 3.38421702e-01 -5.38286805e-01 -9.88517404e-02 6.69785261e-01 7.61282980e-01 -4.47658956e-01 -7.32139573e-02 -1.68580317e+00 4.38990057e-01 8.47660959e-01 2.76749671e-01 -7.30215549e-01 -5.24300635e-01 -7.13294566e-01 -5.57402298e-02 -1.54787481e-01 -5.25345020e-02 8.01988900e-01 -1.00603950e+00 -1.16326380e+00 4.38894749e-01 4.34878618e-01 -5.95770419e-01 4.89460975e-01 5.82279325e-01 -1.04164219e+00 4.22372341e-01 -3.34443212e-01 7.90377915e-01 9.54755843e-01 -1.16704071e+00 -6.74104989e-01 -1.93631098e-01 -8.28887001e-02 -4.74811047e-01 -7.36314714e-01 5.99598736e-02 -2.26603791e-01 -1.07726026e+00 4.07102525e-01 -8.09884727e-01 6.80788830e-02 6.03629410e-01 -2.37241104e-01 6.25686169e-01 1.59108996e+00 -6.87255859e-01 1.30824292e+00 -2.46437120e+00 -6.17460251e-01 7.95151353e-01 2.41509844e-02 8.09011221e-01 -2.07730860e-01 2.42505729e-01 5.54048270e-02 4.77878243e-01 -1.85421005e-01 2.93282419e-01 3.46294767e-03 2.67453432e-01 -4.32016850e-01 6.46553099e-01 -1.96528673e-01 6.92771852e-01 -5.40496409e-01 -5.89759231e-01 -1.59040943e-01 7.37348914e-01 -3.74106258e-01 -5.00184357e-01 1.14485078e-01 -1.40867615e-02 -5.91222882e-01 6.13476098e-01 1.34740162e+00 2.04297140e-01 3.42005253e-01 -2.87098199e-01 1.91847146e-01 -3.52880687e-01 -1.62526286e+00 1.08595335e+00 2.47222811e-01 7.46844649e-01 -3.11235040e-02 -6.98957086e-01 1.15843713e+00 6.32072628e-01 2.64598399e-01 -5.12208223e-01 2.92678654e-01 5.38593471e-01 1.63442999e-01 -5.03886878e-01 2.62617230e-01 -1.35918379e-01 3.27333510e-01 4.58411962e-01 -1.46633536e-01 5.97953379e-01 -2.09368244e-01 -2.87437826e-01 8.22553456e-01 -1.35159537e-01 -1.97969645e-01 -1.64854392e-01 9.76544380e-01 -4.77664888e-01 8.95945191e-01 4.59691733e-01 -4.93962824e-01 1.07071489e-01 4.19941097e-01 -1.63104028e-01 -8.44986856e-01 -7.22013891e-01 -3.09523761e-01 2.77492225e-01 4.16498333e-01 1.12782367e-01 -9.05807316e-01 -5.98748446e-01 1.89227015e-01 1.19656876e-01 -2.57855177e-01 -5.04554629e-01 -5.31746089e-01 -4.04067457e-01 1.50990117e+00 2.57173926e-01 1.38896644e+00 -9.69097197e-01 -4.74599957e-01 3.10561866e-01 4.22219634e-02 -1.15940225e+00 -7.22954690e-01 -3.91776919e-01 -1.13753211e+00 -1.09139895e+00 -9.59460795e-01 -1.55505323e+00 9.36174989e-01 3.12114924e-01 -1.95070043e-01 5.13455629e-01 3.11675202e-02 2.43231878e-02 -1.91931263e-01 -3.61530513e-01 -5.61462939e-01 1.09880362e-02 5.19851670e-02 2.27992356e-01 2.20440626e-01 -4.57844138e-01 -5.38768053e-01 6.04638994e-01 -1.47526538e+00 -4.13623273e-01 6.62554860e-01 6.40998423e-01 3.64890784e-01 7.52259493e-01 7.31688321e-01 -7.03505218e-01 5.05221009e-01 -3.40750515e-02 -8.72865677e-01 3.66030008e-01 -1.00696719e+00 2.86610611e-02 4.50594932e-01 -9.07613337e-01 -6.08790696e-01 -1.50931194e-01 -5.20578548e-02 -4.46501851e-01 3.16784412e-01 4.73168164e-01 -6.78271115e-01 -1.04093838e+00 1.96182624e-01 4.31229651e-01 2.97818154e-01 -5.85052133e-01 -1.03399962e-01 9.16647673e-01 5.35215735e-01 -2.92683035e-01 1.35568297e+00 3.96499187e-01 1.72084987e-01 -7.38734484e-01 2.86682248e-01 1.18991807e-01 -1.33212611e-01 -2.24839915e-02 5.43894410e-01 -7.18905628e-01 -7.64701784e-01 1.18048072e+00 -1.30656874e+00 3.26869369e-01 3.01915169e-01 6.67568505e-01 3.80078293e-02 9.12563026e-01 -5.71451724e-01 -6.23448133e-01 -4.28247720e-01 -1.35037982e+00 1.68015540e-01 5.28620481e-01 4.68221515e-01 -8.50763798e-01 -4.20262814e-01 1.55828819e-01 4.75399137e-01 4.74511981e-01 1.06247687e+00 -7.33776867e-01 -8.23664010e-01 -5.68033516e-01 -3.80284607e-01 5.21870375e-01 2.11659372e-01 4.45453562e-02 -8.43741894e-01 -5.81849515e-01 1.00376219e-01 2.47294769e-01 7.11064517e-01 8.11248347e-02 8.62161696e-01 -5.01739860e-01 -3.69830668e-01 6.61917508e-01 1.51971436e+00 7.88751125e-01 1.04818964e+00 7.44453132e-01 5.52348435e-01 2.76886165e-01 2.45718211e-01 2.87525505e-01 4.32097614e-02 2.21423477e-01 5.12368560e-01 9.59124416e-02 2.11155117e-01 -3.53287607e-01 4.39329654e-01 6.81473374e-01 2.17290759e-01 -1.82959199e-01 -4.16231394e-01 1.49390653e-01 -1.46998847e+00 -1.05878401e+00 -9.83485505e-02 2.10133743e+00 8.16034794e-01 5.85978866e-01 -4.20100719e-01 7.08261728e-01 1.19914448e+00 3.93965811e-01 -5.01235723e-01 -1.17282562e-01 -1.36433482e-01 2.19655067e-01 1.17815578e+00 2.06274763e-01 -9.58018601e-01 6.30766928e-01 5.07638121e+00 8.83883953e-01 -1.19761920e+00 9.43806916e-02 9.79146436e-02 4.94819939e-01 -4.10416424e-01 2.01910749e-01 -9.91334915e-01 7.08782852e-01 6.23214483e-01 -1.38836175e-01 3.99650216e-01 3.42798322e-01 -3.71895619e-02 3.81615251e-01 -5.76901436e-01 8.36855114e-01 -1.03657812e-01 -1.35513198e+00 2.93551147e-01 3.76816958e-01 5.72788179e-01 -4.57101971e-01 3.19482148e-01 4.64378856e-03 -5.16402841e-01 -5.65045536e-01 8.18131030e-01 4.48242158e-01 8.14831972e-01 -9.87591743e-01 8.48639965e-01 1.50380373e-01 -1.29911005e+00 -1.88921511e-01 -6.71085417e-01 3.67741734e-01 -1.53487489e-01 1.37634575e-01 -2.45164007e-01 5.44900060e-01 4.39747840e-01 3.96291852e-01 -3.35641146e-01 1.23071074e+00 -1.33791193e-01 4.11286980e-01 -2.51639009e-01 4.22768220e-02 2.28817314e-01 1.11340493e-01 3.42540890e-01 7.45132387e-01 5.07948458e-01 -5.25158308e-02 -1.72370940e-01 6.77313268e-01 -5.19402921e-01 7.57335573e-02 -3.76133949e-01 -3.92562419e-01 7.91077018e-01 6.71946824e-01 -7.55975366e-01 -1.32993668e-01 -3.54634821e-01 6.99172378e-01 -7.92741537e-01 4.09578592e-01 -9.13032234e-01 -1.20001495e+00 5.10369241e-01 -9.05072689e-02 7.45597780e-01 -2.92473614e-01 1.59157105e-02 -7.86826670e-01 -1.99922966e-03 -1.14741158e+00 1.47776887e-01 -4.59987521e-01 -7.63482392e-01 2.68925816e-01 -3.16638529e-01 -1.43205476e+00 5.78946233e-01 -5.81184804e-01 -4.64840084e-01 8.39212835e-01 -1.90705287e+00 -1.01418316e+00 2.08748318e-02 6.62774980e-01 -1.88958541e-01 -4.22803968e-01 7.00785637e-01 8.53869557e-01 -5.34120500e-01 1.06333220e+00 3.27243090e-01 7.07614303e-01 6.55095756e-01 -1.62831530e-01 5.21214724e-01 1.03796482e+00 -1.45427644e-01 9.48585987e-01 4.55504596e-01 -9.21510518e-01 -1.25546181e+00 -1.00485218e+00 8.55454624e-01 5.48186541e-01 4.69225734e-01 -4.32983898e-02 -8.65063012e-01 7.91955471e-01 -4.11410118e-03 2.03097075e-01 7.19703436e-01 -1.17315602e+00 -5.09108722e-01 -4.81303275e-01 -1.78170812e+00 4.50801700e-01 5.77130198e-01 -4.95989472e-01 -3.79944056e-01 -2.23091543e-01 8.38561952e-01 -3.68726730e-01 -8.05304289e-01 3.05299848e-01 1.06423807e+00 -3.99095029e-01 1.07256973e+00 -3.32053840e-01 -3.66932638e-02 -7.22030282e-01 -2.30669335e-01 -3.22648168e-01 9.62206051e-02 -6.81206822e-01 -6.34654909e-02 1.50187433e+00 1.95996135e-01 -1.20611823e+00 8.54462802e-01 4.24659461e-01 5.08829415e-01 -2.71718323e-01 -1.13967061e+00 -1.02079690e+00 -2.19529286e-01 -1.41092151e-01 1.23206747e+00 1.11387670e+00 -4.30976331e-01 -5.46611071e-01 -3.94356251e-01 5.59611976e-01 8.55293334e-01 -5.69580019e-01 5.03268361e-01 -1.39497435e+00 1.17565461e-01 -4.39989746e-01 -9.07054007e-01 -8.83901715e-01 4.97486070e-02 -8.51987243e-01 -5.46769738e-01 -1.14330935e+00 -3.71830702e-01 -4.77963388e-01 -8.98001909e-01 4.78588253e-01 3.90308321e-01 4.06973690e-01 2.89837807e-01 3.70808899e-01 2.48543695e-01 1.43266350e-01 1.35050881e+00 -4.43717778e-01 9.88939311e-03 9.46495160e-02 -5.57998478e-01 4.09492046e-01 8.46929252e-01 -7.99662054e-01 -6.28596485e-01 -3.59368920e-01 -4.37044837e-02 -1.60456479e-01 4.41470087e-01 -1.18261135e+00 5.06935000e-01 -9.40778106e-02 4.21346962e-01 -4.97940034e-01 -1.00902013e-01 -1.41503203e+00 5.02108574e-01 1.24722600e+00 -1.08826205e-01 1.60836391e-02 3.73873115e-01 6.80248082e-01 -5.54814488e-02 -7.69654036e-01 7.50440419e-01 1.32759035e-01 -8.35559547e-01 4.55285281e-01 -3.36531013e-01 -2.58858651e-01 1.00790632e+00 -7.06823230e-01 -6.20066285e-01 7.23164454e-02 -4.77356017e-02 1.76547199e-01 4.00420010e-01 3.43035102e-01 9.30149019e-01 -1.52998078e+00 -2.96239048e-01 7.59218812e-01 -6.17945082e-02 -4.81667101e-01 5.68017848e-02 4.04661596e-01 -9.97751474e-01 2.28917405e-01 -6.07984602e-01 8.26617796e-03 -1.31600094e+00 5.88208318e-01 4.00703043e-01 8.68576989e-02 -3.14171314e-01 5.44229984e-01 -3.80962908e-01 -1.83362752e-01 5.09906590e-01 -3.00751776e-01 -3.56923491e-01 2.23285165e-02 5.82006633e-01 4.35728192e-01 -2.77399391e-01 -6.74088061e-01 -3.74364018e-01 8.70572865e-01 -2.84802407e-01 -3.04708034e-01 8.68586361e-01 -7.16293752e-02 -2.06424817e-01 -3.35768729e-01 1.49886787e+00 8.80206600e-02 -9.31039751e-01 -3.60955536e-01 -3.34070483e-03 -5.46198249e-01 2.83956915e-01 -4.89146382e-01 -1.65701663e+00 7.25530028e-01 1.06691062e+00 1.83381781e-01 9.77644205e-01 -9.86562133e-01 1.51397777e+00 3.08209568e-01 4.66131151e-01 -8.78802657e-01 -3.94469947e-01 6.24168385e-03 1.63930148e-01 -7.41414785e-01 1.58235297e-01 -5.75397536e-02 -5.91723761e-03 1.43898964e+00 3.81959587e-01 4.70425896e-02 1.01300585e+00 3.43198739e-02 2.22926423e-01 1.92041978e-01 -6.50758222e-02 5.27318537e-01 -2.77707893e-02 6.85062170e-01 -3.39687377e-01 -3.66983503e-01 -5.46055198e-01 6.72631025e-01 8.67617782e-03 3.52040440e-01 6.94678426e-01 1.27035081e+00 -7.54210889e-01 -1.58595526e+00 -6.42229676e-01 8.69399309e-02 -7.42065012e-01 -6.98202057e-03 -6.16590353e-03 7.89164603e-01 3.98771793e-01 8.00648272e-01 -2.79461414e-01 -6.52814567e-01 2.75649935e-01 -4.30201106e-02 2.90900856e-01 4.01698686e-02 -3.63939553e-01 3.20765097e-03 -3.64306152e-01 5.04946262e-02 -4.39034730e-01 -3.31186444e-01 -1.49271727e+00 -6.29242778e-01 -7.08399773e-01 1.11800656e-02 8.59259307e-01 6.50906980e-01 -2.08399668e-02 4.14087623e-01 7.90675223e-01 -1.94217965e-01 -6.52915657e-01 -3.49403203e-01 -8.10052156e-01 -1.98938310e-01 5.49892128e-01 -4.39721733e-01 -4.64771897e-01 1.16343714e-01]
[5.339414596557617, 7.86707878112793]
cca4b9d9-6fe2-4685-8955-a37e70bbffb8
denoising-bottleneck-with-mutual-information
2305.14652
null
https://arxiv.org/abs/2305.14652v3
https://arxiv.org/pdf/2305.14652v3.pdf
Denoising Bottleneck with Mutual Information Maximization for Video Multimodal Fusion
Video multimodal fusion aims to integrate multimodal signals in videos, such as visual, audio and text, to make a complementary prediction with multiple modalities contents. However, unlike other image-text multimodal tasks, video has longer multimodal sequences with more redundancy and noise in both visual and audio modalities. Prior denoising methods like forget gate are coarse in the granularity of noise filtering. They often suppress the redundant and noisy information at the risk of losing critical information. Therefore, we propose a denoising bottleneck fusion (DBF) model for fine-grained video multimodal fusion. On the one hand, we employ a bottleneck mechanism to filter out noise and redundancy with a restrained receptive field. On the other hand, we use a mutual information maximization module to regulate the filter-out module to preserve key information within different modalities. Our DBF model achieves significant improvement over current state-of-the-art baselines on multiple benchmarks covering multimodal sentiment analysis and multimodal summarization tasks. It proves that our model can effectively capture salient features from noisy and redundant video, audio, and text inputs. The code for this paper is publicly available at https://github.com/WSXRHFG/DBF.
['Shaoxiang Wu', 'Zhifang Sui', 'Yunbo Cao', 'Binghuai Lin', 'Tianyu Liu', 'Ziwei Qin', 'Damai Dai']
2023-05-24
null
null
null
null
['multimodal-sentiment-analysis', 'sentiment-analysis', 'multimodal-sentiment-analysis']
['computer-vision', 'natural-language-processing', 'natural-language-processing']
[ 2.33661950e-01 -3.28028381e-01 -1.61811598e-02 -1.78867102e-01 -1.22426748e+00 -4.37993377e-01 4.66783792e-01 1.51859179e-01 -4.00054544e-01 5.48997462e-01 9.16253924e-01 2.34804705e-01 2.38101214e-01 -1.73490882e-01 -8.26704621e-01 -8.43864202e-01 3.60767037e-01 -4.18975353e-01 4.09485102e-02 -3.12376976e-01 8.66505504e-02 -1.75862700e-01 -1.72147298e+00 9.94960606e-01 7.76246607e-01 1.08286524e+00 3.97650123e-01 8.22252095e-01 2.80087572e-02 1.16493917e+00 -3.78058344e-01 -5.53493142e-01 -1.69178694e-02 -3.61463994e-01 -5.30417383e-01 2.15221241e-01 5.86143374e-01 -6.06631815e-01 -8.67187440e-01 1.31035399e+00 8.25022995e-01 3.83480459e-01 3.98611456e-01 -1.34874833e+00 -3.53486180e-01 7.73019314e-01 -9.58474576e-01 1.06670298e-01 7.25109279e-01 2.48181313e-01 1.02425754e+00 -1.20052230e+00 6.66757941e-01 1.47771883e+00 4.68445271e-01 4.83057350e-01 -1.03956234e+00 -4.74201381e-01 5.04425824e-01 3.43250006e-01 -1.29721558e+00 -8.56719017e-01 6.59181714e-01 -1.71770588e-01 7.38088012e-01 4.22457755e-01 4.47021365e-01 1.38497519e+00 1.82739720e-01 1.34424305e+00 6.28326416e-01 -1.54590696e-01 -1.14717208e-01 -2.39135265e-01 1.47606228e-02 5.61651409e-01 -2.73308188e-01 -4.18417960e-01 -1.07893825e+00 -1.67387426e-01 2.67580509e-01 4.46493387e-01 -5.86137235e-01 2.21584216e-01 -1.40931773e+00 5.48234344e-01 2.42757890e-02 2.20056057e-01 -4.76857811e-01 1.94322228e-01 5.93138635e-01 4.84225929e-01 3.53786170e-01 -1.59971118e-01 -2.73343116e-01 -2.54835725e-01 -9.53103006e-01 1.39852852e-01 5.25806785e-01 9.47165191e-01 6.96405232e-01 -3.96757089e-02 -5.45219064e-01 9.43496346e-01 3.44932586e-01 5.61458290e-01 3.36770087e-01 -1.35614514e+00 7.83512533e-01 3.59306753e-01 -2.19681486e-01 -1.02163208e+00 -2.44093210e-01 4.20386121e-02 -1.07304740e+00 -2.39864364e-01 -4.13086824e-02 -2.35694274e-01 -8.99454176e-01 1.68290806e+00 1.99197039e-01 2.29016513e-01 -8.54698271e-02 1.11318994e+00 1.36506498e+00 8.86801958e-01 -1.36399781e-02 -4.52243358e-01 1.42510498e+00 -1.10405433e+00 -1.08609605e+00 4.56245616e-02 2.00580046e-01 -1.02837777e+00 9.23708975e-01 4.78299260e-01 -1.39983141e+00 -4.63752538e-01 -8.25038433e-01 -4.78320867e-01 -9.42432508e-02 -2.28949562e-02 2.58106858e-01 1.45493984e-01 -1.04586387e+00 3.75656366e-01 -8.36651921e-01 -2.62637377e-01 3.51049930e-01 1.37477204e-01 -6.26585007e-01 -4.43308353e-01 -1.09420645e+00 4.58033651e-01 1.63835108e-01 9.82036814e-02 -1.02212584e+00 -5.23451507e-01 -1.04658365e+00 6.16815984e-02 6.42542899e-01 -8.99358213e-01 1.26781857e+00 -1.10241437e+00 -1.23990214e+00 3.41521382e-01 -6.49525344e-01 -2.18444854e-01 3.56101334e-01 -3.88140291e-01 -3.29215974e-01 5.94741106e-01 -3.28086577e-02 1.00497103e+00 1.33811927e+00 -1.28249419e+00 -7.20846772e-01 -2.01073736e-01 6.54428527e-02 6.00201130e-01 -4.83004838e-01 1.08789653e-01 -9.90117848e-01 -9.59812760e-01 1.06639653e-01 -5.34669995e-01 -3.09062060e-02 -2.72768468e-01 -3.96942705e-01 7.73702338e-02 9.57581937e-01 -8.94491255e-01 1.47314072e+00 -2.35579848e+00 6.58690929e-01 -1.15131892e-01 5.02574801e-01 -1.73239619e-01 -5.60103238e-01 6.00260377e-01 1.48021653e-02 5.98964132e-02 -3.20207775e-02 -9.09001470e-01 2.55255103e-02 8.46513659e-02 -2.59843379e-01 3.78018826e-01 1.87523872e-01 8.28789651e-01 -7.07019269e-01 -6.26035392e-01 2.87904739e-01 7.40898550e-01 -4.92996335e-01 4.97475341e-02 -2.83663999e-03 5.51119566e-01 -2.36941263e-01 9.89317715e-01 7.78658152e-01 -1.20750569e-01 -7.04187676e-02 -7.49308407e-01 7.41338283e-02 -1.22625800e-02 -1.22723997e+00 2.20428038e+00 -4.75948900e-02 6.70929730e-01 6.20654464e-01 -7.75968075e-01 1.71714768e-01 5.61810732e-01 6.44568980e-01 -6.91068292e-01 4.44525063e-01 -4.70281504e-02 -4.55702841e-01 -5.90340257e-01 8.88988674e-01 1.67980209e-01 -2.09212467e-01 8.14068764e-02 5.23657560e-01 8.93478394e-02 4.41522866e-01 7.76781261e-01 1.13227296e+00 -8.55086818e-02 -2.14680117e-02 3.16109359e-01 6.22730553e-01 -3.81627947e-01 6.27677560e-01 8.40049803e-01 -2.65890628e-01 1.02293515e+00 5.88471353e-01 1.05951890e-01 -7.12766051e-01 -8.63153398e-01 3.42348397e-01 1.50706208e+00 3.62788469e-01 -9.03492033e-01 -6.83802247e-01 -5.30858040e-01 -1.83608308e-01 3.07477921e-01 -4.54243988e-01 -2.33511180e-01 -2.64978856e-01 -5.29181302e-01 4.59078789e-01 2.84029126e-01 2.87058741e-01 -6.70778632e-01 1.80533361e-02 3.99404466e-02 -1.01481485e+00 -1.23716414e+00 -9.61552978e-01 4.02585827e-02 -6.47348464e-01 -8.98782492e-01 -7.67144501e-01 -5.69638789e-01 4.28413600e-01 7.68953145e-01 9.71874714e-01 -6.74198940e-02 -7.74166640e-03 8.49805832e-01 -6.49312496e-01 -1.60071224e-01 -1.47315562e-01 -1.35618329e-01 -5.65919355e-02 3.06958079e-01 2.04645991e-01 -3.75193208e-01 -6.87601924e-01 2.10423842e-01 -1.30207491e+00 9.33714136e-02 4.27326620e-01 9.72778678e-01 7.57319808e-01 6.09047599e-02 3.43249172e-01 -2.33183801e-01 6.63231194e-01 -5.91479480e-01 -6.54211864e-02 1.96428418e-01 1.30615622e-01 -1.92997336e-01 3.46381336e-01 -5.20061612e-01 -1.18322384e+00 1.28633440e-01 -6.34851158e-02 -7.73436129e-01 -7.78812543e-02 5.96800625e-01 -3.36384565e-01 1.86050832e-01 2.40120918e-01 2.15444326e-01 4.09170352e-02 -4.63997692e-01 3.97410661e-01 5.81246078e-01 6.78578496e-01 -3.88457298e-01 3.96273315e-01 6.55604899e-01 -2.06061035e-01 -8.93094659e-01 -6.16951585e-01 -7.01153517e-01 -3.02058190e-01 -5.26102066e-01 7.03713894e-01 -1.55262649e+00 -7.76472807e-01 5.20330250e-01 -1.08180261e+00 1.26069695e-01 -1.03431627e-01 5.54412723e-01 -3.76445979e-01 7.24952042e-01 -9.21855867e-01 -7.28479028e-01 -3.49495023e-01 -1.31901419e+00 1.50694489e+00 3.07399541e-01 -6.89506829e-02 -4.73358333e-01 -2.72200435e-01 6.66477919e-01 1.58915982e-01 -5.95590584e-02 2.61558592e-01 -2.30473533e-01 -5.36255956e-01 -9.84112248e-02 -7.24841952e-02 4.78834808e-01 -1.26046285e-01 1.60066336e-01 -1.04724944e+00 -3.44193399e-01 7.84417391e-02 -5.35205126e-01 1.69348300e+00 6.41088188e-01 9.64644969e-01 -1.95433944e-01 -4.50183339e-02 4.21465337e-01 1.04682302e+00 -1.47975326e-01 8.01108718e-01 -4.66572605e-02 8.04362237e-01 5.37244976e-01 6.86570823e-01 7.11013198e-01 5.56762338e-01 2.35513836e-01 4.83650506e-01 7.34722754e-03 -1.52929619e-01 -1.84982240e-01 8.11058879e-01 1.08643317e+00 -3.29946615e-02 -5.35949469e-01 -5.03800213e-01 4.01052028e-01 -2.22666502e+00 -1.07864666e+00 2.03029495e-02 1.81845033e+00 7.63752401e-01 -2.96207666e-01 1.39366567e-01 6.57245070e-02 7.61726618e-01 3.62892240e-01 -3.18629682e-01 1.16974376e-01 -7.27949739e-01 -3.91966820e-01 2.53502697e-01 4.09597307e-01 -1.31891513e+00 5.95970809e-01 4.98900032e+00 1.14908421e+00 -8.33452582e-01 3.05623889e-01 5.45517027e-01 -7.48690665e-01 -4.14655149e-01 -2.42091820e-01 -5.72839081e-01 5.43116391e-01 7.02055216e-01 2.50609726e-01 5.71459949e-01 2.63389926e-02 4.41879809e-01 -4.62656796e-01 -8.81478250e-01 1.35603380e+00 2.91621804e-01 -1.21715713e+00 2.28215396e-01 -3.67863953e-01 7.27606356e-01 8.01050588e-02 2.16347307e-01 1.70810372e-01 -8.16982090e-02 -8.50735486e-01 9.77168322e-01 9.66555893e-01 4.03801382e-01 -8.45380962e-01 8.88264775e-01 1.78911492e-01 -1.32328284e+00 -1.51611775e-01 -9.50980633e-02 2.57618368e-01 3.57685804e-01 6.12304568e-01 5.66272251e-02 8.57108653e-01 1.02863765e+00 9.75132883e-01 -4.81741518e-01 7.73977339e-01 8.48781690e-02 3.70791018e-01 -3.16105992e-01 4.21911210e-01 1.36390224e-01 1.11210893e-03 7.92881489e-01 1.28212881e+00 3.81359428e-01 7.11165965e-02 1.96179315e-01 1.63067579e-01 -3.58358115e-01 6.72114938e-02 -4.22149092e-01 -1.71810672e-01 3.87636930e-01 1.26773512e+00 -3.48800808e-01 -3.40325832e-01 -8.13797951e-01 9.92296576e-01 -1.13333911e-01 6.01392031e-01 -9.80355263e-01 -1.57596394e-01 6.19147420e-01 -3.35041910e-01 4.50478137e-01 1.71844158e-02 -1.41181856e-01 -1.72845888e+00 1.81039348e-01 -1.30599964e+00 6.09876215e-01 -8.64702702e-01 -1.22267473e+00 3.04673195e-01 -2.28741392e-01 -1.31896031e+00 9.34642479e-02 -2.27468818e-01 -2.44398132e-01 5.33973277e-01 -1.29002082e+00 -1.09690309e+00 -3.25569302e-01 1.12113166e+00 7.25021183e-01 -1.18254557e-01 2.78083146e-01 6.53510690e-01 -5.96636355e-01 5.12589633e-01 1.75967902e-01 -5.84530309e-02 1.16385913e+00 -8.10082734e-01 -3.26042831e-01 9.61438715e-01 -8.97830501e-02 3.97990376e-01 6.99173093e-01 -6.64616048e-01 -1.88095355e+00 -9.41051006e-01 4.61662769e-01 -9.10169855e-02 6.38862133e-01 -2.25690678e-01 -8.41237307e-01 4.65445608e-01 7.86620855e-01 -1.96983412e-01 6.71724916e-01 -2.59238124e-01 -4.54167694e-01 -1.95259973e-01 -8.17710936e-01 6.57973349e-01 7.46772110e-01 -7.15921998e-01 -3.35353613e-01 8.31643716e-02 9.15956259e-01 -2.83636868e-01 -8.11021507e-01 4.91267383e-01 5.98510444e-01 -9.30285990e-01 8.93729568e-01 -2.60965109e-01 6.39893949e-01 -4.91287857e-01 -6.09184921e-01 -1.12500274e+00 -1.22034021e-01 -9.11985397e-01 -4.99268621e-01 1.42976391e+00 1.53468758e-01 1.18014999e-02 3.02605987e-01 3.33901763e-01 -2.01605037e-01 -4.01081026e-01 -9.20843959e-01 -1.92021236e-01 -5.03599405e-01 -5.77236235e-01 1.93762794e-01 7.26689041e-01 1.50238082e-01 5.19258440e-01 -9.57171619e-01 5.24081551e-02 6.27983630e-01 -1.50210649e-01 6.21418476e-01 -6.88876987e-01 -3.63917723e-02 -3.96643758e-01 -1.55438066e-01 -1.07564771e+00 7.71387368e-02 -4.64453250e-01 1.07613824e-01 -1.43493629e+00 6.24177396e-01 7.28662372e-01 -4.16228116e-01 4.07801837e-01 -3.96571070e-01 4.08709913e-01 5.26217401e-01 1.13430798e-01 -1.28157592e+00 8.64226162e-01 1.27168536e+00 -4.64602292e-01 -8.43404531e-02 -3.73413444e-01 -8.97267699e-01 8.22593451e-01 5.66299796e-01 -3.41943145e-01 -2.13558510e-01 -6.58477664e-01 3.22004557e-01 2.29259193e-01 4.01520163e-01 -6.59172595e-01 5.93300641e-01 1.54768284e-02 4.20816272e-01 -9.50099707e-01 7.48585045e-01 -7.98321486e-01 4.67827767e-02 2.62906067e-02 -3.37957114e-01 9.67959687e-03 2.44363487e-01 8.06495011e-01 -7.97269762e-01 3.49564701e-02 5.55802166e-01 -5.48609793e-02 -7.85432875e-01 2.39812747e-01 -6.94073558e-01 -8.76329988e-02 6.26145363e-01 7.88200945e-02 -5.96691012e-01 -9.13031399e-01 -8.30276251e-01 6.46942139e-01 3.91441554e-01 4.64970678e-01 9.80049729e-01 -1.29439962e+00 -8.62279952e-01 -6.69112280e-02 2.07046643e-02 -1.79580629e-01 9.49666500e-01 1.33443177e+00 -8.89492556e-02 4.50692475e-02 1.83830075e-02 -6.12520814e-01 -1.63582897e+00 5.31962991e-01 2.30621528e-02 -3.47971208e-02 -2.59083331e-01 7.76086092e-01 3.11773270e-01 -1.19252736e-02 6.40581310e-01 -2.24042267e-01 -2.63793677e-01 5.80337465e-01 8.95670891e-01 5.33074081e-01 -4.26843390e-02 -9.12191927e-01 -2.91263878e-01 2.70907193e-01 -2.31761903e-01 -1.67552546e-01 1.22445464e+00 -8.91553581e-01 -2.93105781e-01 3.99534613e-01 1.29284716e+00 6.90296665e-02 -1.26990247e+00 -3.87547940e-01 -5.49874127e-01 -3.70579869e-01 1.95379883e-01 -5.43170154e-01 -1.19507563e+00 7.47015119e-01 3.81218374e-01 1.32551298e-01 1.70888436e+00 4.64190431e-02 7.90358722e-01 3.75716418e-01 -1.65995270e-01 -1.32164395e+00 2.84641683e-01 6.60488784e-01 1.03702962e+00 -1.40073812e+00 7.08666071e-02 -2.74659961e-01 -1.08959734e+00 1.04737353e+00 3.59183669e-01 2.48262629e-01 5.52837074e-01 5.17691135e-01 -8.10949728e-02 4.77019250e-02 -1.20907879e+00 -3.06717724e-01 4.50101107e-01 2.72194892e-01 2.31983766e-01 -1.87277228e-01 -5.92296161e-02 8.77577901e-01 3.79394293e-01 -8.67544860e-02 4.50328171e-01 1.01442814e+00 -4.67309356e-01 -7.34914362e-01 -6.60142183e-01 3.13648313e-01 -7.95575678e-01 -3.81356388e-01 -3.04620087e-01 3.02505672e-01 -1.51592242e-02 1.46050680e+00 -1.19679287e-01 -6.24395072e-01 2.66304731e-01 1.01539819e-02 3.84650320e-01 -1.53292909e-01 -6.85264230e-01 9.58797336e-01 2.15966459e-02 -9.48248208e-01 -8.39354157e-01 -7.45030761e-01 -9.02743101e-01 -5.06803930e-01 -1.82103738e-01 -1.28791645e-01 3.56344819e-01 7.39489555e-01 5.05887508e-01 6.69991791e-01 4.47862417e-01 -1.15776885e+00 -1.60863280e-01 -1.06910443e+00 -6.02984309e-01 5.28708994e-01 6.89091861e-01 -4.11473751e-01 -3.20979953e-01 3.55041683e-01]
[13.546647071838379, 4.764918804168701]
ad7068d5-f7cc-4d14-b9e8-fc601480c5b5
employing-weak-annotations-for-medical-image
1708.06297
null
http://arxiv.org/abs/1708.06297v1
http://arxiv.org/pdf/1708.06297v1.pdf
Employing Weak Annotations for Medical Image Analysis Problems
To efficiently establish training databases for machine learning methods, collaborative and crowdsourcing platforms have been investigated to collectively tackle the annotation effort. However, when this concept is ported to the medical imaging domain, reading expertise will have a direct impact on the annotation accuracy. In this study, we examine the impact of expertise and the amount of available annotations on the accuracy outcome of a liver segmentation problem in an abdominal computed tomography (CT) image database. In controlled experiments, we study this impact for different types of weak annotations. To address the decrease in accuracy associated with lower expertise, we propose a method for outlier correction making use of a weakly labelled atlas. Using this approach, we demonstrate that weak annotations subject to high error rates can achieve a similarly high accuracy as state-of-the-art multi-atlas segmentation approaches relying on a large amount of expert manual segmentations. Annotations of this nature can realistically be obtained from a non-expert crowd and can potentially enable crowdsourcing of weak annotation tasks for medical image analysis.
['Kensaku MORI', 'Kazunari Misawa', 'Jonathan Passerat-Palmbach', 'Christian Ledig', 'Daniel Rueckert', 'Martin Rajchl', 'Lisa M. Koch']
2017-08-21
null
null
null
null
['liver-segmentation']
['medical']
[-1.74146250e-03 6.50736809e-01 2.07481131e-01 -3.08154881e-01 -1.14400399e+00 -6.33485436e-01 2.90551960e-01 8.26542377e-01 -9.31289971e-01 8.11734974e-01 -2.06010304e-02 -2.07862779e-01 5.66164441e-02 -3.94755512e-01 -6.60936892e-01 -5.52533984e-01 8.07826743e-02 8.09355795e-01 5.68007112e-01 8.50096643e-02 -4.21045087e-02 2.21780419e-01 -1.08799696e+00 3.23172897e-01 8.95996988e-01 7.29895294e-01 7.79548660e-02 5.85137546e-01 9.83298048e-02 7.80484259e-01 -8.12951565e-01 -5.34199119e-01 5.90318561e-01 -2.61028945e-01 -8.58106613e-01 1.72178298e-01 2.62495577e-01 -3.12266082e-01 3.17732185e-01 9.02311027e-01 9.02662277e-01 -1.85182959e-01 3.78528953e-01 -1.11932576e+00 -1.36567131e-01 6.65081084e-01 -2.76863456e-01 4.08185333e-01 6.04431868e-01 2.00703353e-01 5.12855411e-01 -6.91161692e-01 7.09432662e-01 3.94593298e-01 1.10716164e+00 2.91289091e-01 -1.34797263e+00 -2.77655929e-01 -1.72138929e-01 -2.72965133e-01 -1.53334415e+00 -2.76215643e-01 2.12653860e-01 -7.85242379e-01 4.75363284e-01 2.58477002e-01 6.44367278e-01 8.08237076e-01 -2.14066962e-03 4.44703192e-01 1.34573925e+00 -7.22814620e-01 5.35487890e-01 4.87130314e-01 -3.02958429e-01 7.37535954e-01 4.90569115e-01 -4.03230458e-01 -3.95946920e-01 -4.84989345e-01 5.91520905e-01 -2.90614009e-01 -3.59809399e-01 -4.36743647e-01 -1.30152023e+00 7.50771701e-01 3.43498439e-01 6.58686936e-01 -4.21658188e-01 -4.17789705e-02 6.54345810e-01 1.23419374e-01 8.12938929e-01 8.09949398e-01 -4.09819901e-01 -1.71049708e-03 -1.30420160e+00 2.67512146e-02 1.17765296e+00 7.23225117e-01 3.47542018e-01 -5.03535092e-01 -5.80454648e-01 4.66626465e-01 1.63476437e-01 1.17878921e-01 6.07945919e-01 -7.19612837e-01 2.88424492e-01 7.93544710e-01 3.85240793e-01 -7.48876154e-01 -4.64587450e-01 -1.04677618e-01 -3.14920068e-01 3.51558357e-01 1.04329169e+00 -2.93265909e-01 -8.96416068e-01 1.18648946e+00 7.51943648e-01 9.30068567e-02 -3.43563795e-01 1.12705207e+00 8.06551278e-01 -4.37332094e-01 4.36392337e-01 -2.18786836e-01 1.41992939e+00 -8.49037588e-01 -6.04007185e-01 2.12366581e-01 9.80152845e-01 -9.28941846e-01 1.20758986e+00 2.76846796e-01 -1.10972834e+00 -2.21849948e-01 -7.36399651e-01 8.12812373e-02 -2.97326952e-01 3.57314736e-01 1.55214682e-01 1.03045881e+00 -1.08807659e+00 6.04750931e-01 -1.08974135e+00 -5.93788564e-01 6.37144387e-01 5.03150582e-01 -4.02718544e-01 -1.65325403e-03 -8.16193640e-01 1.26952326e+00 2.33015418e-01 -6.30395189e-02 -8.50746095e-01 -8.26939285e-01 -4.98024732e-01 -3.42486262e-01 6.06103539e-01 -6.80798590e-01 1.24904370e+00 -1.39638257e+00 -1.29386437e+00 1.28713989e+00 1.71313941e-01 -4.17512655e-01 1.36436319e+00 -1.41931906e-01 6.14012666e-02 3.06873471e-01 3.68315130e-01 5.54389358e-01 6.26198947e-01 -1.27816784e+00 -3.76828611e-01 -1.87827513e-01 1.32453457e-01 1.94239907e-03 -1.59339771e-01 2.16857120e-01 -1.52538210e-01 -5.78683019e-01 -1.52928784e-01 -1.31040967e+00 -5.71799159e-01 3.42579782e-02 -1.97380438e-01 -5.19819520e-02 2.16912031e-01 -7.65782058e-01 1.00973248e+00 -1.90092039e+00 -4.20703962e-02 2.93318301e-01 3.85392398e-01 4.36062336e-01 3.84641081e-01 1.24273099e-01 3.78313243e-01 3.52787346e-01 -2.59788662e-01 -4.92633522e-01 -1.92954317e-01 3.82788360e-01 4.43885654e-01 7.68510520e-01 1.86841693e-02 9.38036740e-01 -1.03187180e+00 -1.01175523e+00 -9.88153964e-02 1.66146860e-01 -5.05570531e-01 1.93085119e-01 -1.61610376e-02 1.13419294e+00 -4.21307117e-01 5.82466125e-01 3.03876936e-01 -4.53470260e-01 4.22227651e-01 2.16374025e-02 -3.39449309e-02 -3.08472037e-01 -1.05599773e+00 1.80787122e+00 -3.44081700e-01 2.21255571e-01 2.68373549e-01 -5.88552356e-01 4.56459075e-01 7.05220222e-01 7.27989435e-01 -4.75444496e-01 2.78507650e-01 5.94469726e-01 1.31883487e-01 -7.06696212e-01 5.04380286e-01 -2.09803715e-01 1.52099296e-01 5.04224002e-01 9.53852087e-02 -1.87971547e-01 1.55743241e-01 9.05399472e-02 1.35194409e+00 1.15236044e-01 5.32444715e-01 -6.50559187e-01 5.07214427e-01 4.74422872e-01 3.95850092e-01 1.02823842e+00 -7.51351058e-01 6.42264664e-01 2.90758759e-01 -6.32751405e-01 -1.23637748e+00 -5.73174417e-01 -1.56612575e-01 1.18453574e+00 4.46096361e-02 -2.10207433e-01 -1.04422903e+00 -1.00645804e+00 -7.33737051e-02 1.19768701e-01 -6.62948012e-01 2.58389503e-01 -5.57161987e-01 -8.73336196e-01 9.34755743e-01 3.92796963e-01 1.63957566e-01 -7.94362843e-01 -1.11520469e+00 2.50867784e-01 -2.01130539e-01 -1.44206965e+00 -2.77492434e-01 -8.31838474e-02 -7.42904782e-01 -1.21323657e+00 -1.05218995e+00 -3.85092497e-01 1.05649602e+00 -2.75961667e-01 1.57278025e+00 6.31204903e-01 -4.02232051e-01 7.67069638e-01 -7.26026535e-01 -8.08514655e-01 -5.80924928e-01 4.13017720e-01 -4.30599265e-02 -1.15683213e-01 2.36165941e-01 -1.92012355e-01 -7.85448670e-01 4.99401420e-01 -1.10738325e+00 -2.65451044e-01 3.88567179e-01 5.62339425e-01 5.41945994e-01 -6.08989298e-01 5.05836368e-01 -1.26555037e+00 5.91075838e-01 -4.02650625e-01 -5.41510701e-01 3.15232843e-01 -6.53479099e-01 -2.76993841e-01 3.10185492e-01 -6.56423032e-01 -7.58626282e-01 3.09018940e-01 1.97814971e-01 -3.36923242e-01 -1.99095160e-01 5.15926242e-01 5.37748933e-01 -6.09063506e-01 1.08631790e+00 -4.30317938e-01 -5.93357999e-03 -2.22433001e-01 1.53582156e-01 3.13985288e-01 8.58278722e-02 -6.28849208e-01 4.26948279e-01 6.02579534e-01 -2.78270599e-02 -3.12453300e-01 -9.19948399e-01 -6.51949883e-01 -1.00064659e+00 -2.75456369e-01 1.14762676e+00 -9.57851231e-01 -2.73621023e-01 2.81932466e-02 -9.72598493e-01 -5.72961569e-01 -4.40086812e-01 5.94585299e-01 -3.62925470e-01 3.91671926e-01 -4.68319863e-01 -6.10588133e-01 -2.07012653e-01 -1.43473148e+00 1.24057376e+00 6.89103082e-02 -5.73945701e-01 -1.10620654e+00 1.28047884e-01 5.08898735e-01 5.85002482e-01 6.21398211e-01 3.06242406e-01 -1.29939187e+00 -4.66106415e-01 -3.79089504e-01 9.40527488e-03 1.01593815e-01 -2.21741572e-01 -4.05862153e-01 -9.84515309e-01 -3.45641136e-01 6.00576028e-02 -4.27223414e-01 2.28525952e-01 1.56097770e-01 8.49700809e-01 -1.43380478e-01 -2.81275660e-01 4.18108441e-02 1.40214050e+00 -4.29020047e-01 3.61461848e-01 6.19300246e-01 6.16700411e-01 5.60263693e-01 7.12011218e-01 5.72107732e-01 2.93971807e-01 5.93506217e-01 1.53894931e-01 -4.25128281e-01 7.71816969e-02 3.14275235e-01 -2.51404464e-01 4.97503936e-01 -4.47495401e-01 -6.14284389e-02 -1.55220866e+00 7.07510173e-01 -1.85471976e+00 -3.38868052e-01 -3.05228919e-01 2.24568105e+00 1.00865793e+00 -5.31815365e-02 2.35866860e-01 4.63718399e-02 6.70067549e-01 -3.78716528e-01 5.80740422e-02 -5.37076406e-02 2.11556956e-01 5.47720790e-02 8.63735616e-01 1.94219664e-01 -9.38656092e-01 4.73789483e-01 6.80848169e+00 5.55839956e-01 -9.11508262e-01 8.38697195e-01 6.66859746e-01 -1.92122191e-01 3.49103846e-02 -1.31037280e-01 -3.14076960e-01 6.50318325e-01 9.32370365e-01 9.72547531e-02 1.72229577e-02 7.73298681e-01 2.14526668e-01 -6.97975516e-01 -1.27926779e+00 6.35476649e-01 1.35895804e-01 -1.11252296e+00 -3.83107752e-01 1.17113635e-01 9.12345350e-01 1.64280012e-01 -4.50238943e-01 5.80940433e-02 2.40030989e-01 -1.05518317e+00 6.37516975e-01 6.51995540e-01 6.20812058e-01 -3.04654658e-01 1.31963372e+00 5.13576210e-01 -6.44718349e-01 9.14114267e-02 1.80493649e-02 9.49267820e-02 2.68649399e-01 8.48151326e-01 -1.32508123e+00 6.56626403e-01 6.87871099e-01 -2.35390097e-01 -8.40899289e-01 1.20802987e+00 -1.08713850e-01 5.48981547e-01 -3.65830272e-01 1.18639447e-01 1.11310646e-01 2.19865292e-01 4.64837044e-01 1.21522021e+00 1.24508522e-01 2.27858946e-01 6.64497316e-01 5.42115211e-01 -7.63975531e-02 4.08405334e-01 -4.09809381e-01 2.90915728e-01 3.43497574e-01 1.44408607e+00 -1.30405700e+00 -5.52126467e-01 -3.57232720e-01 9.28424478e-01 3.14223021e-01 -7.57100955e-02 -9.24090922e-01 2.80900747e-01 -2.80250341e-01 5.08132637e-01 8.24100450e-02 6.49667308e-02 -3.39385182e-01 -8.48713756e-01 2.31209651e-01 -9.60597813e-01 3.71831000e-01 -5.73954999e-01 -1.28816390e+00 6.62178636e-01 -3.17458473e-02 -1.18483114e+00 2.21187975e-02 -3.18466932e-01 -4.16001469e-01 7.28943825e-01 -1.34674299e+00 -1.33208811e+00 -7.01692462e-01 3.37862194e-01 2.03244627e-01 9.34103876e-02 7.60718942e-01 4.76056218e-01 -2.09768161e-01 5.85045218e-01 -2.70084113e-01 1.97913349e-01 9.65280175e-01 -1.32169008e+00 -2.37443015e-01 7.00726926e-01 -5.49624003e-02 3.35069835e-01 7.69464910e-01 -7.94412732e-01 -8.23447049e-01 -1.06944883e+00 6.32409394e-01 -1.16418111e+00 5.01141429e-01 -4.25663777e-02 -8.59731853e-01 6.29458964e-01 8.82067084e-02 6.68485045e-01 9.98338938e-01 -1.28352866e-01 1.75523818e-01 4.44174796e-01 -1.64620423e+00 2.26551950e-01 8.66422772e-01 -4.46124166e-01 -4.65427727e-01 6.35587692e-01 4.10899460e-01 -8.94495904e-01 -1.33903039e+00 3.96984279e-01 3.10173184e-01 -9.52068210e-01 5.95146239e-01 -3.92810762e-01 2.29108453e-01 -3.21519196e-01 2.14893401e-01 -1.12963700e+00 1.74088344e-01 -4.46862817e-01 3.52783918e-01 1.13459241e+00 5.44673979e-01 -4.79941607e-01 8.53708684e-01 1.06243384e+00 -5.99432550e-02 -6.92068458e-01 -1.29896629e+00 -5.84488869e-01 9.22780484e-02 4.15283367e-02 3.14219713e-01 1.16009653e+00 -3.53862420e-02 -2.72353232e-01 3.95585783e-02 2.89213866e-01 2.77097434e-01 -3.97170752e-01 9.49492097e-01 -1.25291514e+00 -2.31670633e-01 3.70483622e-02 -6.03404224e-01 7.38651305e-03 -6.79658651e-02 -8.75232816e-01 1.87949136e-01 -1.32993615e+00 1.34505510e-01 -7.12830186e-01 9.18052997e-03 3.30955029e-01 -4.49808955e-01 7.31411517e-01 2.84246713e-01 6.52078450e-01 -1.14750290e+00 -1.55336976e-01 1.07937312e+00 2.92031199e-01 -1.23262867e-01 -2.61177778e-01 -2.72710770e-01 8.72414947e-01 5.86603582e-01 -9.28458810e-01 -2.50619669e-02 -5.08009374e-01 4.31587577e-01 -1.67623580e-01 4.97130156e-01 -1.07577658e+00 6.14085317e-01 2.69578844e-01 3.95681292e-01 1.60633937e-01 -1.38979211e-01 -9.82109427e-01 2.03262627e-01 4.63438600e-01 -2.84847051e-01 2.33922750e-01 1.02194928e-01 6.55336559e-01 -2.00629607e-01 -5.84908664e-01 6.19531453e-01 -7.27714658e-01 -5.73587604e-02 -1.33758157e-01 -3.45437586e-01 2.06018463e-01 1.45214164e+00 -3.12301964e-01 3.07592358e-02 -1.75209977e-02 -1.07904065e+00 1.97764650e-01 7.39636838e-01 -1.43165171e-01 -1.02373190e-01 -9.77256894e-01 -8.37152183e-01 -2.81527907e-01 2.06935942e-01 2.62726754e-01 -5.39243780e-03 1.61694014e+00 -7.75697291e-01 -1.09286886e-02 -3.43539193e-02 -8.75147462e-01 -1.20393932e+00 3.41656625e-01 5.65744281e-01 -7.00828612e-01 -4.34700966e-01 6.90567553e-01 -3.60423326e-01 -4.44923609e-01 4.19731960e-02 -3.63938808e-01 -5.21429479e-02 2.83656031e-01 2.81078190e-01 4.72395748e-01 5.84480405e-01 -5.22677124e-01 -4.89169568e-01 1.32440567e-01 8.13852400e-02 -1.49209946e-01 9.98852015e-01 -2.00054914e-01 -7.40656704e-02 4.25666064e-01 4.57963079e-01 4.33422804e-01 -9.27730441e-01 -1.01741469e-02 4.27154571e-01 -5.22526026e-01 -1.03379704e-01 -9.85983074e-01 -7.76338577e-01 2.35940099e-01 7.93828845e-01 4.61973906e-01 7.65193999e-01 1.40252590e-01 4.04831052e-01 7.03125298e-02 5.85333169e-01 -1.32022607e+00 2.52090823e-02 -5.02005741e-02 5.36621809e-01 -1.82626665e+00 2.70281494e-01 -5.18458247e-01 -8.59392107e-01 7.93835759e-01 4.26222414e-01 -3.42258699e-02 4.82738554e-01 3.90074432e-01 5.15147448e-01 -4.16749686e-01 -1.01780705e-01 -2.72940397e-01 2.03589991e-01 4.30566669e-01 6.41679108e-01 6.40346855e-02 -6.30425990e-01 4.80023950e-01 1.77168578e-01 4.19546396e-01 7.33035028e-01 1.27266479e+00 -1.06539957e-01 -9.25540209e-01 -6.67884886e-01 3.92287880e-01 -1.20594418e+00 -1.88775603e-02 -3.62860143e-01 8.44815135e-01 5.14320314e-01 9.59084809e-01 -2.14753434e-01 3.32057655e-01 4.16527539e-01 1.39418244e-01 4.66861725e-01 -8.97043645e-01 -1.33828413e+00 -1.64471567e-01 2.52777576e-01 -5.62624156e-01 -9.30645525e-01 -5.18416405e-01 -1.02079034e+00 1.52430370e-01 -6.10525727e-01 1.79324657e-01 6.52056932e-01 1.13632786e+00 2.98746318e-01 5.84245980e-01 2.56800279e-02 -9.57365930e-01 -5.34004509e-01 -1.08057177e+00 -2.35092267e-01 7.93681741e-01 3.45628113e-02 -5.79468131e-01 -3.77426207e-01 2.47787371e-01]
[14.863517761230469, -2.502560615539551]
d4d22155-a2a1-4051-9628-65499dae3cef
generalized-lstm-based-end-to-end-text
2011.04896
null
https://arxiv.org/abs/2011.04896v4
https://arxiv.org/pdf/2011.04896v4.pdf
An Empirical Study on Text-Independent Speaker Verification based on the GE2E Method
While many researchers in the speaker recognition area have started to replace the former classical state-of-the-art methods with deep learning techniques, some of the traditional i-vector-based methods are still state-of-the-art in the context of text-independent speaker verification. Google's Generalized End-to-End Loss for Speaker Verification (GE2E), a deep learning-based technique using long short-term memory units, has recently gained a lot of attention due to its speed in convergence and generalization. In this study, we aim at further studying the GE2E method and comparing different scenarios in order to investigate all of its aspects. Various experiments including the effects of a random sampling of test and enrollment utterances, test utterance duration, and the number of enrollment utterances are discussed in this article. Furthermore, we compare the GE2E method with the baseline state-of-the-art i-vector-based methods for text-independent speaker verification and show that it outperforms them by resulting in lower error rates while being end-to-end and requiring less training time for convergence.
['Soroosh Tayebi Arasteh']
2020-11-10
null
null
null
null
['text-independent-speaker-verification']
['speech']
[-1.01522394e-01 -4.94593143e-01 1.32406633e-02 -8.68742824e-01 -1.05763829e+00 -3.43327522e-01 4.66686040e-01 -7.34274387e-02 -4.81875360e-01 3.78220558e-01 2.60934770e-01 -6.26302421e-01 9.63614658e-02 -6.19437173e-02 -4.37286139e-01 -7.28974640e-01 -1.40436172e-01 3.12048405e-01 -5.42401485e-02 -1.06138304e-01 2.44289771e-01 3.66400719e-01 -1.59795558e+00 -1.33759588e-01 7.15872467e-01 1.00280595e+00 -2.04204693e-01 6.45186543e-01 6.99267304e-03 3.70762914e-01 -8.54733407e-01 -5.02844036e-01 -5.49450293e-02 -4.15186137e-01 -6.16768360e-01 1.34548634e-01 5.60116410e-01 -2.10851297e-01 -3.71143490e-01 9.16230202e-01 1.21634042e+00 4.26207155e-01 3.76382530e-01 -1.14925373e+00 -6.97202027e-01 6.24848664e-01 -3.27198654e-01 2.93615013e-01 3.24593782e-01 -2.07496807e-01 7.27183998e-01 -1.02671802e+00 6.74405620e-02 1.30852544e+00 7.88476408e-01 7.56422997e-01 -8.72856557e-01 -9.69477773e-01 3.39850903e-01 7.21362233e-01 -1.58139122e+00 -9.69744623e-01 8.09594572e-01 -1.82110235e-01 9.18044925e-01 2.24986792e-01 -9.89338160e-02 1.25014818e+00 -6.10768013e-02 1.10073817e+00 8.04252267e-01 -7.70403981e-01 2.65177995e-01 3.22502673e-01 4.88907546e-01 4.02866870e-01 -3.27465802e-01 4.13649976e-01 -8.06440175e-01 -8.93452317e-02 1.99845508e-01 -2.91670114e-01 -4.05447751e-01 3.07341721e-02 -9.42593694e-01 9.94833529e-01 1.44732162e-01 6.38288498e-01 -3.23583663e-01 -2.27772370e-01 7.26140916e-01 3.95136446e-01 5.94091952e-01 -2.38658115e-01 -3.86211485e-01 -3.50204706e-01 -1.32494187e+00 1.39238015e-01 8.07603657e-01 5.06589592e-01 1.15452647e-01 5.69996953e-01 -3.74011129e-01 1.02524447e+00 5.78992009e-01 3.12267274e-01 1.09768331e+00 -1.93228483e-01 5.04970610e-01 2.91590691e-01 -1.44716740e-01 -3.44129771e-01 -1.04357943e-01 -7.36215532e-01 -7.44620442e-01 2.24962998e-02 2.01743737e-01 -2.83464611e-01 -1.07206905e+00 1.80838585e+00 1.24703668e-01 3.33292782e-01 2.57289708e-01 7.58174837e-01 8.78650010e-01 6.78255141e-01 -1.23599343e-01 -5.83949424e-02 1.15166771e+00 -1.03313994e+00 -9.16820943e-01 -1.39960036e-01 4.52526420e-01 -8.51600945e-01 9.09863174e-01 2.58809656e-01 -7.90525079e-01 -7.34232247e-01 -1.09620929e+00 2.23447368e-01 -4.59541172e-01 3.47651504e-02 8.43499675e-02 1.41983485e+00 -1.25548911e+00 2.76568949e-01 -6.44788802e-01 -2.48262882e-01 2.94718176e-01 4.02927935e-01 -3.29632282e-01 -8.85175988e-02 -1.37784851e+00 9.71622348e-01 -2.20501926e-02 2.37591863e-01 -9.21818912e-01 -5.84974885e-01 -9.47776496e-01 2.67071068e-01 5.30681126e-02 -2.33576760e-01 1.55708146e+00 -7.78431296e-01 -1.95787430e+00 5.77910066e-01 -6.71442032e-01 -4.55901712e-01 3.77261400e-01 -1.29338965e-01 -7.74661899e-01 -3.78758967e-01 -3.19610655e-01 3.35803241e-01 9.23100293e-01 -8.12932849e-01 -3.73573154e-01 -6.77099884e-01 -4.07048374e-01 7.53809288e-02 -4.46964413e-01 4.76614535e-01 -3.11146051e-01 -4.80194360e-01 -1.95023678e-02 -8.15676153e-01 7.97431394e-02 -4.44191962e-01 -5.42812288e-01 -5.63107014e-01 1.05488420e+00 -7.67977059e-01 1.25311184e+00 -2.34840393e+00 3.73433493e-02 -4.12085913e-02 -3.91006321e-01 5.76641440e-01 -1.42468095e-01 3.91037196e-01 -2.81460196e-01 1.72916148e-02 -2.03912809e-01 -1.14367020e+00 2.27506846e-01 -1.65837795e-01 -1.24767661e-01 7.30421603e-01 -3.13610077e-01 6.34833336e-01 -3.82195622e-01 -1.70461372e-01 3.85639697e-01 1.00310647e+00 -6.62799506e-03 9.68497768e-02 3.34299475e-01 3.99863154e-01 6.51529804e-02 7.04752386e-01 7.30468214e-01 1.98437825e-01 -9.57674980e-02 3.04890603e-01 -1.40963688e-01 4.32325691e-01 -1.11755419e+00 1.55450547e+00 -5.31967819e-01 1.07633817e+00 1.94397792e-01 -9.66693103e-01 8.96773338e-01 8.03454399e-01 -3.79934423e-02 -4.97632951e-01 3.54142278e-01 3.57049942e-01 8.08609501e-02 -1.72856569e-01 2.82592952e-01 -3.57462466e-01 1.08289361e-01 3.96810532e-01 2.26437803e-02 2.55742550e-01 -2.16739506e-01 -1.39108196e-01 5.48574805e-01 -3.39896530e-01 2.46180240e-02 -7.66756311e-02 9.19942558e-01 -8.75030100e-01 3.86758298e-01 4.12731975e-01 -5.70230126e-01 5.38039267e-01 -1.58690244e-01 -1.26629218e-01 -6.32230639e-01 -6.69419169e-01 -2.86278188e-01 1.09430540e+00 -2.55793244e-01 2.71179248e-02 -9.30477798e-01 -4.65890586e-01 7.41479322e-02 1.03066015e+00 -4.86970901e-01 4.09024097e-02 -3.71367544e-01 -6.11537814e-01 9.46349025e-01 5.39764464e-01 6.92484498e-01 -7.98850179e-01 -2.33381182e-01 2.81595439e-01 -1.98600739e-01 -9.16594505e-01 -7.30653465e-01 7.73534626e-02 -6.99750960e-01 -4.95510161e-01 -1.22153497e+00 -7.76959836e-01 2.82746851e-01 1.69740334e-01 7.15704322e-01 -2.84631103e-01 9.76008996e-02 2.30733961e-01 -4.07886177e-01 -6.52719557e-01 -5.53891480e-01 1.56792521e-01 3.27354431e-01 4.44528252e-01 7.39535809e-01 -2.81777561e-01 -2.84216732e-01 5.32608569e-01 -7.39661634e-01 -7.22826660e-01 2.82548815e-01 8.97743881e-01 1.28268287e-01 -4.54041734e-02 8.56935203e-01 -3.15777302e-01 8.38114619e-01 -1.52075872e-01 -5.10059476e-01 3.64998609e-01 -6.45043314e-01 4.55052927e-02 1.98707417e-01 -5.22141397e-01 -7.85755098e-01 -2.38483682e-01 -5.72569549e-01 -6.26768053e-01 -2.64571130e-01 7.26441801e-01 -3.77983004e-01 -1.88041449e-01 3.65669519e-01 6.55869007e-01 1.24617651e-01 -6.57514513e-01 1.24603257e-01 1.34223557e+00 2.92713553e-01 6.51786476e-02 5.13606489e-01 -3.23069133e-02 -6.37955487e-01 -1.13599157e+00 -4.71241415e-01 -7.86599994e-01 -3.63762796e-01 -1.85956553e-01 5.72019458e-01 -9.12263930e-01 -8.10503840e-01 1.02582109e+00 -1.03270888e+00 -1.34913042e-01 3.48595418e-02 6.25283778e-01 -1.20362952e-01 6.42384529e-01 -5.06848454e-01 -1.23170853e+00 -6.70156002e-01 -1.53418934e+00 9.83102381e-01 2.16353014e-01 -4.16743420e-02 -1.01949584e+00 1.57920673e-01 4.87175196e-01 9.41707194e-01 -3.72656137e-01 5.50538182e-01 -1.05873609e+00 -1.05067335e-01 -7.48684227e-01 5.84063865e-02 7.38103569e-01 1.17998138e-01 -2.32863396e-01 -1.42300498e+00 -6.27631307e-01 3.65653068e-01 -1.83853563e-02 7.81866372e-01 6.87481284e-01 9.34057593e-01 -2.85178095e-01 -1.72127277e-01 3.60895634e-01 1.23936450e+00 2.86834300e-01 5.50037205e-01 2.95699269e-01 3.15430969e-01 5.10088980e-01 1.16874710e-01 2.40383863e-01 3.86811286e-01 9.79959488e-01 1.79626331e-01 5.08495513e-03 -4.86266799e-02 7.84756616e-02 7.03072488e-01 7.85212576e-01 2.78079152e-01 -6.20002210e-01 -8.65333557e-01 7.37226188e-01 -1.52985251e+00 -1.06804240e+00 2.89413750e-01 2.50355482e+00 4.64365751e-01 -7.39675239e-02 4.26747650e-01 7.64918983e-01 1.07988560e+00 2.56316483e-01 -5.51030993e-01 -5.53153992e-01 -5.46239875e-02 4.89130383e-03 3.11719000e-01 6.96173847e-01 -1.00992084e+00 5.97771347e-01 6.30890656e+00 7.56524444e-01 -1.68898940e+00 3.21510643e-01 7.07490146e-01 1.22571234e-02 1.10708676e-01 -4.81759697e-01 -1.25011754e+00 4.41534579e-01 1.56501067e+00 -1.23059452e-02 9.93060991e-02 9.99118626e-01 2.07038298e-01 2.33804330e-01 -1.11189532e+00 1.21444917e+00 5.41209877e-01 -9.00045514e-01 -2.44425341e-01 2.16302159e-03 4.64141518e-01 2.88896412e-01 2.77616888e-01 6.58862293e-01 -3.03736717e-01 -9.92341101e-01 9.17797506e-01 -8.03548470e-02 7.45490134e-01 -7.75616527e-01 1.24344254e+00 2.17786938e-01 -9.62097466e-01 -2.52289087e-01 -7.05091730e-02 1.34357452e-01 4.41784233e-01 5.17848969e-01 -7.21257627e-01 5.25059998e-01 4.56206143e-01 2.77266651e-01 -2.59301186e-01 1.13823593e+00 -1.37308789e-02 1.02882230e+00 -3.05350780e-01 -2.62857825e-01 3.87571037e-01 2.62675911e-01 5.93373716e-01 1.31071448e+00 3.54218215e-01 -2.76254654e-01 -6.02235682e-02 4.90028650e-01 -1.71434134e-01 1.20339684e-01 -2.48366371e-01 2.71842983e-02 4.80096757e-01 6.22847676e-01 -6.20614290e-02 -4.25643414e-01 -4.37165529e-01 9.15558636e-01 -7.89724365e-02 5.04268765e-01 -7.72449195e-01 -6.02843165e-01 7.17031896e-01 -1.72689576e-02 6.50011361e-01 -2.58762896e-01 -1.92666769e-01 -8.98208499e-01 1.81195691e-01 -9.98222470e-01 1.19156256e-01 -1.48699418e-01 -1.28582215e+00 9.18362319e-01 -1.63116366e-01 -1.12002981e+00 -4.92519587e-01 -3.02382141e-01 -7.57402956e-01 1.21337378e+00 -1.75215280e+00 -9.23327565e-01 -7.13672787e-02 6.74996972e-01 8.35153162e-01 -5.08592725e-01 9.97279763e-01 6.28200114e-01 -7.93117821e-01 1.40082932e+00 5.37834048e-01 3.79655182e-01 7.37069547e-01 -7.68493593e-01 7.36004531e-01 1.01697087e+00 1.53916970e-01 6.35233343e-01 7.56662309e-01 -1.21181332e-01 -1.29316688e+00 -7.30098665e-01 1.31900346e+00 -1.28724009e-01 1.97035447e-01 -4.39574540e-01 -9.18367088e-01 5.32435417e-01 2.54578054e-01 -1.59941941e-01 9.40293491e-01 3.88959646e-01 -5.79803050e-01 -2.31027991e-01 -1.23925149e+00 3.17262113e-01 4.46787089e-01 -7.60813296e-01 -5.04270911e-01 7.15607777e-02 3.97566259e-01 -2.86863357e-01 -4.87308949e-01 4.09489363e-01 6.43692017e-01 -1.03951705e+00 8.25565457e-01 -2.01239616e-01 -3.32685024e-01 -1.50571004e-01 -4.25428540e-01 -1.36374211e+00 -6.76292703e-02 -5.94275773e-01 3.41325291e-02 1.29342043e+00 6.59505546e-01 -7.59348869e-01 7.82882690e-01 5.56478381e-01 -2.69928753e-01 -6.37849331e-01 -1.60746551e+00 -1.21938682e+00 6.89012855e-02 -5.75097919e-01 6.24053657e-01 7.27727711e-01 -3.93211003e-03 5.67622244e-01 -4.15414482e-01 1.47007063e-01 5.41584551e-01 -1.72691047e-02 5.78828037e-01 -1.00675642e+00 -7.97865614e-02 -5.01642168e-01 -7.92924881e-01 -1.14214587e+00 3.37209672e-01 -5.91298103e-01 3.05596322e-01 -1.32257247e+00 -3.61973383e-02 -4.57814261e-02 -3.90518337e-01 3.63282651e-01 -3.27685833e-01 -6.16173558e-02 1.24402866e-01 -1.23435974e-01 -2.54967988e-01 8.55784893e-01 5.09984314e-01 -3.56853724e-01 -4.56905842e-01 4.46849674e-01 -4.09736603e-01 7.33826086e-02 8.02615821e-01 -2.10212946e-01 -2.87216902e-01 -4.33577389e-01 -8.48153532e-01 8.65172893e-02 -5.02764247e-02 -1.14420438e+00 3.35933208e-01 5.73153853e-01 1.06160246e-01 -7.21990585e-01 5.92238963e-01 -4.92878497e-01 -1.93815187e-01 4.03421879e-01 -3.46783429e-01 -2.65041869e-02 4.75049406e-01 4.81940448e-01 -6.61799312e-01 -2.28575930e-01 9.53040183e-01 2.61079669e-01 -4.81405318e-01 3.09458017e-01 -6.93213999e-01 -3.72834951e-01 7.29978025e-01 -4.13839668e-01 1.11575685e-01 -7.33372331e-01 -5.36466956e-01 -8.41414556e-02 2.27399766e-02 6.27371728e-01 6.59926295e-01 -9.94635284e-01 -1.21507394e+00 4.22955096e-01 5.07342815e-02 -6.05541348e-01 5.63687801e-01 6.98676050e-01 -5.04696108e-02 8.89777184e-01 1.52633160e-01 -8.26065660e-01 -1.87416756e+00 4.16669548e-01 4.48095173e-01 -8.79120603e-02 -2.32808039e-01 1.26893342e+00 -1.16210245e-01 -6.30954087e-01 9.93038118e-01 -1.24644496e-01 5.41958325e-02 -5.97726041e-03 7.82640874e-01 3.21153700e-01 4.79491889e-01 -8.05173993e-01 -7.33306944e-01 4.01524752e-01 -2.89390743e-01 -5.18145740e-01 1.00364625e+00 -1.62256688e-01 4.25747961e-01 4.34610903e-01 1.44360185e+00 8.15057009e-03 -6.16557598e-01 -3.85108501e-01 -7.08938763e-02 -3.03392649e-01 5.29843986e-01 -8.91409218e-01 -1.09610033e+00 1.18179345e+00 1.17792809e+00 -9.20472816e-02 9.80833888e-01 -3.30085695e-01 8.93351912e-01 2.50928760e-01 2.10940555e-01 -7.19141543e-01 -3.72320026e-01 5.75080991e-01 8.27384055e-01 -1.52114022e+00 -2.40600377e-01 2.64133252e-02 -4.72649515e-01 8.74852002e-01 7.52956346e-02 4.35861349e-01 9.51729774e-01 -3.28028873e-02 5.91321111e-01 2.70103842e-01 -4.04170960e-01 4.21396457e-02 3.39036763e-01 4.76998538e-01 8.03049445e-01 1.26325816e-01 6.28787978e-03 3.58404309e-01 -1.90512165e-01 -3.60077545e-02 2.90642250e-02 7.75452197e-01 -2.62187660e-01 -1.35190701e+00 -5.33039749e-01 6.98986873e-02 -6.02358460e-01 -2.21109986e-01 -3.39649051e-01 4.82464373e-01 -3.40121835e-01 1.47429645e+00 -2.80081421e-01 -5.36208153e-01 2.96189606e-01 5.06416500e-01 2.59938389e-01 -3.45657259e-01 -7.97134519e-01 -4.76789810e-02 2.36855745e-02 -1.25073954e-01 -4.32451516e-01 -7.80307353e-01 -7.80605733e-01 -5.71563721e-01 -9.92335618e-01 3.65108073e-01 1.46370101e+00 9.55580175e-01 4.91175324e-01 4.25056934e-01 7.52195477e-01 -9.97688830e-01 -9.91758049e-01 -1.54038072e+00 -5.51332533e-01 1.66572526e-01 6.61172807e-01 -5.43141901e-01 -6.29734993e-01 -2.06535369e-01]
[14.321465492248535, 6.084474563598633]
9a8fb870-fbfd-4887-a302-8eb5999e10a7
semi-supervised-learning-with-normalizing-1
1912.13025
null
https://arxiv.org/abs/1912.13025v1
https://arxiv.org/pdf/1912.13025v1.pdf
Semi-Supervised Learning with Normalizing Flows
Normalizing flows transform a latent distribution through an invertible neural network for a flexible and pleasingly simple approach to generative modelling, while preserving an exact likelihood. We propose FlowGMM, an end-to-end approach to generative semi supervised learning with normalizing flows, using a latent Gaussian mixture model. FlowGMM is distinct in its simplicity, unified treatment of labelled and unlabelled data with an exact likelihood, interpretability, and broad applicability beyond image data. We show promising results on a wide range of applications, including AG-News and Yahoo Answers text data, tabular data, and semi-supervised image classification. We also show that FlowGMM can discover interpretable structure, provide real-time optimization-free feature visualizations, and specify well calibrated predictive distributions.
['Pavel Izmailov', 'Andrew Gordon Wilson', 'Marc Finzi', 'Polina Kirichenko']
2019-12-30
null
https://proceedings.icml.cc/static/paper_files/icml/2020/3378-Paper.pdf
https://proceedings.icml.cc/static/paper_files/icml/2020/3378-Paper.pdf
icml-2020-1
['semi-supervised-text-classification-1']
['natural-language-processing']
[ 9.69952568e-02 2.33048841e-01 -3.68661553e-01 -8.14655662e-01 -7.14941204e-01 -7.04360306e-01 9.33395028e-01 -4.84301507e-01 1.50213838e-01 6.27024531e-01 6.54282212e-01 -3.76236349e-01 -4.31799352e-01 -6.60653234e-01 -4.22629297e-01 -7.85024822e-01 1.34604096e-01 1.15486681e+00 -4.72955376e-01 1.61468431e-01 -2.48119980e-02 5.08356571e-01 -1.19456887e+00 3.20167363e-01 7.74374008e-01 7.71952510e-01 -1.50239974e-01 7.76162803e-01 -2.68948466e-01 1.05445230e+00 -5.45798600e-01 -6.92481637e-01 1.78619355e-01 -4.66831356e-01 -8.44300747e-01 6.22194588e-01 5.98428667e-01 -2.03796670e-01 -1.60628334e-01 9.06751871e-01 3.65108103e-01 9.64002460e-02 1.30751014e+00 -1.73046780e+00 -1.13869488e+00 7.53254294e-01 -4.09526289e-01 -8.65405500e-02 5.79206422e-02 2.91762531e-01 9.80176210e-01 -7.31829226e-01 7.07670927e-01 1.81984973e+00 3.58675241e-01 3.77526432e-01 -1.70891023e+00 -3.41197103e-01 1.63480937e-02 -3.23417902e-01 -1.07795465e+00 -3.97141725e-01 6.03103340e-01 -8.58301103e-01 7.69836545e-01 3.81462663e-01 3.66188645e-01 1.23425543e+00 2.09647402e-01 9.73270476e-01 9.56727922e-01 -2.66852081e-01 2.07907036e-01 2.90261090e-01 4.32003103e-02 7.66378045e-01 -2.49194041e-01 1.42268064e-02 -6.23344779e-01 -3.05179358e-01 8.73230338e-01 2.27902189e-01 -1.25031933e-01 -7.74101377e-01 -1.18730760e+00 1.15224349e+00 2.76882499e-01 -1.84506804e-01 -9.04946849e-02 3.30092549e-01 1.57726243e-01 1.49590999e-01 7.67338574e-01 2.21459925e-01 -5.16019426e-02 -2.26895973e-01 -1.16814411e+00 3.08956336e-02 8.51074398e-01 1.16290772e+00 7.14740038e-01 5.47385037e-01 -1.56478420e-01 7.45885849e-01 7.26739645e-01 6.28003657e-01 6.10091209e-01 -1.25673378e+00 1.12063505e-01 5.14406443e-01 -2.07190603e-01 -7.28516281e-01 -3.13808680e-01 -7.85474002e-01 -1.20958078e+00 5.13985991e-01 2.33870521e-01 -5.78119606e-02 -1.10806298e+00 1.72710991e+00 5.95413521e-02 -4.96688157e-01 6.51051998e-02 5.65583706e-01 8.23412538e-01 6.74987137e-01 1.59552731e-02 5.56932837e-02 9.14223075e-01 -1.02690983e+00 -6.16688192e-01 -3.65489691e-01 3.41788769e-01 -6.48967624e-01 1.18729854e+00 3.56299341e-01 -9.73841250e-01 -4.47545201e-01 -6.25769198e-01 -3.60246927e-01 -4.58805859e-01 2.68899411e-01 8.79428923e-01 6.35429502e-01 -1.11830199e+00 6.14572704e-01 -1.06906009e+00 -2.67438948e-01 7.56961167e-01 9.81079787e-02 -3.57819945e-01 1.15781352e-01 -5.26421189e-01 4.81405288e-01 6.67363703e-01 -1.41919240e-01 -9.14956868e-01 -8.69256139e-01 -9.86452878e-01 2.89175183e-01 1.11191615e-01 -1.01668739e+00 1.25613356e+00 -8.43394995e-01 -1.50548851e+00 7.31956005e-01 -1.52846903e-01 -3.53166938e-01 8.65126014e-01 -6.71762750e-02 -1.57958955e-01 -2.12459013e-01 1.85931727e-01 1.05634320e+00 1.04178488e+00 -9.90118563e-01 -1.78814426e-01 -1.54602259e-01 -4.88442034e-01 1.17945880e-01 -3.68806273e-01 -2.94078946e-01 -1.97645217e-01 -8.81823599e-01 2.56308727e-02 -6.83754325e-01 -2.14698061e-01 1.12265430e-01 -7.47893929e-01 4.25493456e-02 9.92042065e-01 -5.55123806e-01 1.02616823e+00 -2.05594826e+00 2.38805220e-01 2.77470231e-01 4.89094257e-01 -9.69609693e-02 -9.93749127e-02 3.44191641e-01 -3.64503205e-01 2.09451020e-01 -4.82251167e-01 -7.28768349e-01 4.46330875e-01 4.30543095e-01 -5.89686096e-01 2.46206135e-01 1.92476049e-01 1.46242356e+00 -7.81092167e-01 -3.77143711e-01 4.48971629e-01 6.83265984e-01 -4.82893735e-01 2.20395193e-01 -3.64790708e-01 3.88114542e-01 9.13709924e-02 6.61176622e-01 3.84670913e-01 -7.27995574e-01 -6.08285218e-02 -9.10094082e-02 2.50748396e-01 2.91411970e-02 -1.29118335e+00 1.52164125e+00 -2.84594864e-01 8.93737197e-01 -2.41163760e-01 -4.61765528e-01 1.08736086e+00 -9.45134163e-02 3.34874451e-01 -3.32163036e-01 -3.96574615e-03 -7.86359757e-02 -4.51427996e-01 -1.27043441e-01 5.18413126e-01 -2.45778501e-01 -2.82771196e-02 8.85823131e-01 7.06365883e-01 -8.02931860e-02 5.15117764e-01 6.42502308e-01 4.55408901e-01 3.62902254e-01 4.11816806e-01 -2.87205726e-01 -4.86463606e-02 -2.28848174e-01 1.50965713e-02 7.10635126e-01 4.49199051e-01 8.21536124e-01 6.74500108e-01 -4.76613164e-01 -1.02924430e+00 -1.46239924e+00 -4.49533761e-02 1.27384686e+00 -4.83276993e-01 -5.23765862e-01 -4.94124979e-01 -5.09314239e-01 1.41718900e-02 1.03113520e+00 -8.20755482e-01 -1.95433393e-01 -1.47521511e-01 -1.09711611e+00 3.19660425e-01 8.39876771e-01 2.84422666e-01 -1.08244562e+00 -1.80297524e-01 1.14118092e-01 -2.62754057e-02 -7.43535876e-01 -7.04983473e-01 4.49334651e-01 -9.11384106e-01 -7.28094518e-01 -6.94781899e-01 -4.31255490e-01 8.84197772e-01 -9.37637240e-02 1.34897792e+00 -6.99812412e-01 -3.41187477e-01 4.40885514e-01 2.89182752e-01 -2.32661471e-01 -1.06168103e+00 2.39260733e-01 -3.11819643e-01 7.62639269e-02 2.54469756e-02 -6.58524990e-01 -8.27693120e-02 3.82071972e-01 -1.11757147e+00 3.89628351e-01 3.35649788e-01 9.13726330e-01 6.73622191e-01 -3.68563771e-01 3.30255300e-01 -1.12718260e+00 9.10087466e-01 -4.53869700e-01 -6.44344807e-01 4.66583431e-01 -1.03615367e+00 4.82427150e-01 4.53235537e-01 -4.86084998e-01 -1.26295531e+00 1.53554246e-01 4.96531092e-02 -6.79112971e-01 -3.09639424e-01 4.12248969e-01 -2.99290866e-01 3.47132772e-01 1.02365506e+00 2.11564228e-01 3.48919839e-01 -5.82794070e-01 1.15950549e+00 5.31855881e-01 1.21809185e+00 -3.11511666e-01 7.90753901e-01 6.16389573e-01 6.73015937e-02 -6.48833334e-01 -6.89626276e-01 -3.98839623e-01 -8.41968954e-01 1.82791967e-02 5.77182174e-01 -7.55148768e-01 -5.33981442e-01 4.06731427e-01 -8.53125393e-01 -4.03196484e-01 -7.25703955e-01 1.34008408e-01 -8.58577311e-01 3.19760948e-01 -5.07997155e-01 -5.65390944e-01 -5.24593115e-01 -9.06136036e-01 1.09215653e+00 1.76777273e-01 -6.18274808e-01 -1.60073256e+00 1.69871047e-01 1.48558110e-01 5.95734537e-01 2.79698998e-01 9.52388644e-01 -6.36260629e-01 -6.86420977e-01 -1.04233630e-01 -1.75582096e-01 4.42391515e-01 3.67282033e-01 3.41345042e-01 -1.08411860e+00 -1.36667073e-01 -4.09267336e-01 -3.49291146e-01 1.12429178e+00 5.11115074e-01 1.08287680e+00 -6.17568791e-01 -1.43854842e-01 1.15858245e+00 9.71607327e-01 -9.95659158e-02 5.79933524e-01 -7.69492760e-02 9.10068631e-01 4.56805736e-01 -3.17186527e-02 4.02829766e-01 1.49891242e-01 1.05168968e-01 4.37599182e-01 -3.98686200e-01 -1.79216504e-01 -5.45141101e-01 2.18320057e-01 5.16808748e-01 3.09368908e-01 -7.33078480e-01 -7.03605831e-01 3.09803694e-01 -1.91822374e+00 -1.06229258e+00 -2.31635477e-02 1.87134349e+00 6.59097910e-01 1.75339997e-01 2.49077007e-01 -1.95938900e-01 5.48953891e-01 1.67085513e-01 -6.25525832e-01 -2.30745509e-01 -4.78648752e-01 -1.02251425e-01 3.65833133e-01 6.33484662e-01 -1.23152137e+00 7.97695398e-01 7.65276861e+00 1.09754133e+00 -1.05403471e+00 -7.90772513e-02 1.06594634e+00 -3.59794408e-01 -7.19722211e-01 -8.90522674e-02 -7.62984276e-01 3.34166050e-01 7.52199411e-01 -3.26082319e-01 4.81245100e-01 1.02005148e+00 4.87925150e-02 1.09484009e-01 -1.20211458e+00 1.15198588e+00 1.79401323e-01 -1.78817368e+00 3.52547228e-01 9.18836072e-02 7.52221584e-01 -2.75624190e-02 3.78715813e-01 -3.67816426e-02 8.70939791e-01 -1.30595899e+00 9.66536403e-01 7.55206883e-01 1.00533056e+00 -6.81155562e-01 2.00362325e-01 4.32970226e-01 -6.67932749e-01 1.79782823e-01 -2.66740620e-01 2.24698320e-01 5.25567472e-01 5.94559610e-01 -1.02202249e+00 2.60396898e-01 4.53441858e-01 1.01901722e+00 -8.98654521e-01 9.39315200e-01 -3.22627574e-01 8.41415763e-01 -3.87564212e-01 2.05255866e-01 1.42449021e-01 -3.17871153e-01 4.43195105e-01 1.54810500e+00 1.36418283e-01 -5.91081679e-01 1.19173028e-01 1.41536176e+00 -4.27726917e-02 -6.17226772e-02 -5.15890956e-01 -2.59379268e-01 2.34505851e-02 1.37231159e+00 -1.10676301e+00 -4.99936104e-01 1.70657501e-01 8.07819426e-01 1.08203717e-01 5.69374263e-01 -6.68579400e-01 1.79495215e-02 2.52644897e-01 6.69946745e-02 7.52578005e-02 -8.38362202e-02 -4.72273678e-01 -1.23341191e+00 -3.26881677e-01 -6.14823282e-01 5.99600196e-01 -1.09099150e+00 -1.55245388e+00 7.24320114e-01 4.12801176e-01 -1.11202586e+00 -9.91571188e-01 -6.49736464e-01 -5.26082397e-01 9.71535027e-01 -1.05841804e+00 -1.34568977e+00 -3.90908897e-01 3.61964613e-01 3.92141700e-01 -4.28263843e-01 8.46452534e-01 2.70019658e-02 -4.40809518e-01 4.21497792e-01 4.08767521e-01 1.08352296e-01 7.32363880e-01 -1.62448740e+00 6.48364663e-01 7.75601566e-01 6.03997946e-01 6.58127844e-01 5.82045257e-01 -6.22624278e-01 -8.78629267e-01 -1.25601757e+00 4.97307599e-01 -5.27995646e-01 4.50732082e-01 -5.63440979e-01 -8.56160998e-01 1.08879840e+00 4.08740669e-01 -1.99977443e-01 8.18345487e-01 -2.04204023e-02 -4.94762361e-01 1.39057636e-01 -9.00894523e-01 3.74476224e-01 8.75671029e-01 -4.67906654e-01 -2.93418109e-01 5.06247103e-01 5.36885381e-01 -4.73532766e-01 -6.25681460e-01 -7.63794556e-02 5.32309771e-01 -9.63835895e-01 9.34121072e-01 -8.22158813e-01 6.04941428e-01 -1.21993199e-01 -4.20864299e-02 -1.44578063e+00 -6.17550254e-01 -1.10237885e+00 -3.10350865e-01 1.42960298e+00 4.62908149e-01 -6.13426507e-01 8.86658013e-01 6.76307797e-01 6.61812648e-02 -5.65851450e-01 -5.43590069e-01 -7.03255177e-01 5.25858663e-02 -4.21999425e-01 6.73555434e-01 8.65835965e-01 -3.89271021e-01 4.74256009e-01 -6.34833694e-01 -2.97598541e-01 7.65668631e-01 3.99416029e-01 9.10245359e-01 -1.35487890e+00 -5.75511813e-01 -6.78257048e-01 -3.46521705e-01 -1.12215531e+00 9.93276238e-02 -1.28265786e+00 -2.16925055e-01 -1.63303649e+00 2.97409534e-01 -5.93929291e-02 2.73139894e-01 7.45157838e-01 1.59970269e-01 3.06335896e-01 3.21394764e-02 4.07696217e-01 -5.08448005e-01 5.86610138e-01 1.04607844e+00 -2.19760433e-01 -1.33527666e-01 3.63431685e-02 -9.29213524e-01 7.44054019e-01 6.32066011e-01 -4.96816963e-01 -7.20005453e-01 -3.15378666e-01 1.78969041e-01 -1.50367871e-01 6.76683187e-01 -3.80080104e-01 2.41468474e-02 -2.76888013e-01 6.69372678e-01 -7.77164221e-01 1.54838398e-01 -3.77165258e-01 6.13609493e-01 9.65387821e-02 -6.05968416e-01 -2.05028895e-02 5.61285727e-02 3.67648304e-01 -1.15359634e-01 -1.46550313e-01 8.24005723e-01 -8.21261853e-02 -5.69554865e-01 3.61486226e-01 -2.77375102e-01 2.20471784e-01 6.00042641e-01 -2.62097001e-01 -5.97595334e-01 -8.60653758e-01 -1.00682437e+00 1.52194470e-01 3.88687432e-01 4.73346859e-01 4.66563851e-01 -1.35264933e+00 -6.94917858e-01 5.43634176e-01 -2.12610816e-03 1.23682372e-01 4.09204923e-02 3.74173462e-01 -3.13793272e-01 2.76605338e-01 -3.12414914e-01 -9.24285889e-01 -9.82133389e-01 3.76956046e-01 3.65704596e-01 -2.64210463e-01 -3.66720378e-01 7.25597620e-01 3.82670939e-01 -6.87665820e-01 1.99700549e-01 -2.84619063e-01 1.74468771e-01 2.12506548e-01 3.79499853e-01 5.20233810e-01 -1.57435745e-01 -6.08940780e-01 5.23562618e-02 1.84570730e-01 1.45528316e-01 -2.60802209e-01 1.48260188e+00 -1.28311157e-01 -5.32382093e-02 5.18087149e-01 1.29904699e+00 -2.54801631e-01 -1.68381977e+00 -2.42808431e-01 -3.17476183e-01 -3.08401972e-01 -1.22186504e-01 -9.74906266e-01 -1.12952077e+00 1.03838599e+00 2.97335923e-01 1.74189284e-01 8.97272050e-01 2.93065757e-01 3.40602174e-02 3.52589816e-01 -2.83747941e-01 -5.66385567e-01 2.73373574e-01 2.86732376e-01 1.06025004e+00 -1.12677538e+00 1.70404658e-01 -3.70364606e-01 -7.73411632e-01 1.33121908e+00 3.51466686e-01 2.76408702e-01 6.09897912e-01 5.57791114e-01 1.93797797e-01 -2.01720640e-01 -8.34584892e-01 1.16054580e-01 6.14383757e-01 8.25014532e-01 2.86040306e-01 -1.42386109e-02 6.46046996e-01 3.44463289e-01 -6.06857359e-01 -1.43251702e-01 3.54482114e-01 6.33954704e-01 -1.60976380e-01 -8.63834620e-01 -3.52848440e-01 6.50305033e-01 -6.93039745e-02 -1.75227493e-01 -3.26542348e-01 7.35652745e-01 -2.95673370e-01 4.25999939e-01 3.00790787e-01 2.43467297e-02 -2.10245222e-01 4.37045991e-01 2.24899322e-01 -4.43276912e-01 -1.57577366e-01 6.90461099e-01 -2.02785403e-01 -3.18367511e-01 -4.11921203e-01 -6.93765044e-01 -9.92105603e-01 -3.07985812e-01 -1.53288633e-01 -3.94500345e-02 5.41374028e-01 8.04383337e-01 5.16460776e-01 4.86598819e-01 3.36410135e-01 -6.55756831e-01 -6.28809512e-01 -1.09580243e+00 -4.80314076e-01 4.65723723e-01 2.53442168e-01 -4.15928334e-01 -3.76528680e-01 5.83447695e-01]
[11.369890213012695, -0.0861077532172203]
b4a0df00-264a-434c-9d68-d4f8755bd3d8
visual-depth-mapping-from-monocular-images
1812.04082
null
http://arxiv.org/abs/1812.04082v1
http://arxiv.org/pdf/1812.04082v1.pdf
Visual Depth Mapping from Monocular Images using Recurrent Convolutional Neural Networks
A reliable sense-and-avoid system is critical to enabling safe autonomous operation of unmanned aircraft. Existing sense-and-avoid methods often require specialized sensors that are too large or power intensive for use on small unmanned vehicles. This paper presents a method to estimate object distances based on visual image sequences, allowing for the use of low-cost, on-board monocular cameras as simple collision avoidance sensors. We present a deep recurrent convolutional neural network and training method to generate depth maps from video sequences. Our network is trained using simulated camera and depth data generated with Microsoft's AirSim simulator. Empirically, we show that our model achieves superior performance compared to models generated using prior methods.We further demonstrate that the method can be used for sense-and-avoid of obstacles in simulation.
['Rachael E. Tompa', 'John Mern', 'Mykel J. Kochenderfer', 'Kyle Julian']
2018-12-10
null
null
null
null
['depth-and-camera-motion']
['computer-vision']
[ 5.88234551e-02 -5.53602651e-02 2.15267837e-01 -3.67073536e-01 -5.09537637e-01 -7.39539623e-01 5.19348145e-01 -3.65193337e-01 -7.27060020e-01 6.24316931e-01 -3.76468480e-01 -6.37518525e-01 1.90139577e-01 -7.61094630e-01 -8.34878922e-01 -3.16350937e-01 -3.38242650e-01 3.32250893e-01 5.13388515e-01 -4.75654334e-01 4.47998121e-02 9.40847516e-01 -1.58363855e+00 -1.24171972e-01 8.45548809e-02 8.83086026e-01 5.13349771e-01 1.26206458e+00 6.40084088e-01 7.80248940e-01 -7.81094372e-01 3.20544422e-01 7.31595635e-01 -2.65541613e-01 -2.20773295e-01 6.23900928e-02 3.83171678e-01 -7.81112373e-01 -7.67584503e-01 8.38414490e-01 3.93085301e-01 3.90156567e-01 3.92612636e-01 -1.28100610e+00 1.31540716e-01 -2.03739479e-01 -1.84672445e-01 2.88385957e-01 3.68676126e-01 4.02161270e-01 4.09823656e-01 -5.14026523e-01 6.91349089e-01 9.81745422e-01 6.90176368e-01 7.69915164e-01 -7.90275097e-01 -4.70162630e-01 1.03483990e-01 6.03174642e-02 -1.54372251e+00 -3.65922540e-01 5.13822377e-01 -2.67485797e-01 1.38302815e+00 -3.60483788e-02 7.55398870e-01 9.98791933e-01 7.02203035e-01 4.02293086e-01 4.70285058e-01 -3.58414948e-02 3.90299231e-01 -3.84233028e-01 -5.36756158e-01 1.07079685e+00 5.22996306e-01 7.29660690e-01 -6.13074228e-02 1.65055871e-01 1.11565280e+00 -5.36096729e-02 -2.79264987e-01 -6.93413794e-01 -1.23130226e+00 6.86217248e-01 7.73962557e-01 -2.96928138e-01 -2.37974569e-01 6.41918361e-01 1.16785593e-01 2.10770056e-01 1.00369193e-01 5.53768456e-01 -1.51593477e-01 -2.26011470e-01 -7.13796258e-01 4.79179859e-01 5.27584493e-01 1.33373535e+00 6.04859352e-01 5.97404778e-01 4.41929370e-01 1.49850264e-01 2.29956269e-01 6.69532478e-01 2.44556069e-01 -1.30720294e+00 4.20466751e-01 2.59955049e-01 5.70936799e-01 -8.25382590e-01 -6.19111538e-01 -3.10216129e-01 -3.86576504e-01 1.17692125e+00 -1.59841344e-01 -6.91759348e-01 -1.05324304e+00 1.29284346e+00 1.60878330e-01 2.16213271e-01 4.30330396e-01 1.29563367e+00 4.57656026e-01 6.79149985e-01 -4.00750905e-01 1.55403823e-01 7.75545299e-01 -9.85521972e-01 -4.17628348e-01 -8.08245838e-01 7.62068629e-01 -3.96170616e-01 3.56074721e-01 4.58185256e-01 -9.74692523e-01 -6.09723806e-01 -1.58285522e+00 4.42444794e-02 -2.94243187e-01 1.85370758e-01 5.15089869e-01 2.64388412e-01 -1.16695404e+00 6.02152586e-01 -1.18790412e+00 -2.31085137e-01 1.57384336e-01 4.43754137e-01 -5.26256859e-01 1.30270958e-01 -8.96487296e-01 1.23515570e+00 2.67032057e-01 2.82103509e-01 -1.70206380e+00 -6.10847995e-02 -1.32421899e+00 -3.46566051e-01 4.17287111e-01 -7.30419815e-01 1.58234715e+00 -5.89087784e-01 -1.52414727e+00 5.91556311e-01 -4.92293425e-02 -9.65297818e-01 3.40862989e-01 -5.46281815e-01 -1.53909519e-01 3.21803898e-01 -8.95380080e-02 9.74027395e-01 6.22126698e-01 -1.23543143e+00 -8.05233777e-01 -6.55528009e-02 2.64468640e-01 5.89422643e-01 3.97429466e-01 -3.00619066e-01 -2.32650712e-01 -3.20029169e-01 2.01643538e-02 -1.28345239e+00 -8.12470913e-01 3.44318658e-01 -1.88700616e-01 5.20930886e-01 1.03350151e+00 -2.78039962e-01 5.58965564e-01 -1.71352470e+00 1.91435844e-01 3.02537549e-02 5.99046536e-02 3.92565370e-01 -8.46205205e-02 2.80338913e-01 2.53340185e-01 -3.31888497e-01 -1.28906488e-01 -2.10491136e-01 -3.16108197e-01 2.87211508e-01 -3.74343038e-01 4.47883993e-01 -3.00733373e-02 5.66994607e-01 -9.65688825e-01 -2.39172019e-02 6.28829658e-01 5.17001569e-01 -5.22184193e-01 5.34412444e-01 -1.93175778e-01 4.47406352e-01 -1.61931887e-01 5.92424572e-01 4.00298357e-01 3.09189111e-01 4.96480316e-02 -2.18869317e-02 -3.14562589e-01 2.51425892e-01 -8.72047782e-01 1.70052493e+00 -5.98508835e-01 1.13366687e+00 3.76810789e-01 -4.95637715e-01 9.53827083e-01 1.33210078e-01 9.26269144e-02 -4.55229491e-01 5.56589782e-01 1.44862235e-01 -1.56189904e-01 -2.02986360e-01 8.02153111e-01 -2.00456619e-01 -1.97078213e-01 3.94018777e-02 -1.64502218e-01 -8.25757623e-01 7.17592016e-02 7.55767003e-02 1.32522559e+00 2.44767338e-01 1.62928313e-01 4.99012694e-02 8.19244832e-02 3.72885585e-01 5.20501673e-01 5.09364486e-01 -1.61428511e-01 6.18959665e-01 -6.29215455e-03 -7.27920055e-01 -1.02560520e+00 -1.18674946e+00 2.71755457e-01 3.55836064e-01 7.49780834e-01 -1.93676040e-01 -6.94192946e-01 -5.36235094e-01 -9.87941474e-02 7.37236381e-01 -2.72007883e-01 -2.86661237e-01 -5.95227718e-01 -2.01829553e-01 5.40775418e-01 9.29639816e-01 5.20657599e-01 -6.04865193e-01 -1.60411096e+00 3.59216154e-01 2.16606170e-01 -1.44833112e+00 -1.04660198e-01 2.41676301e-01 -7.44893014e-01 -1.14122498e+00 -3.52078348e-01 -5.96514225e-01 8.14014971e-01 7.58813202e-01 9.11177814e-01 8.35528225e-03 -4.96304870e-01 5.13176799e-01 1.93239376e-02 -6.24302030e-01 -2.48774067e-01 -2.17406288e-01 5.01511574e-01 -6.14581823e-01 2.09492952e-01 -2.89144069e-01 -7.20086575e-01 5.39360106e-01 -5.54631710e-01 9.60688740e-02 5.36323428e-01 5.23495913e-01 5.13702512e-01 3.80300842e-02 -5.31927273e-02 -8.18741396e-02 4.27275091e-01 -1.98799744e-01 -1.31139886e+00 -3.56896579e-01 -2.59299725e-01 -1.06205508e-01 6.96546078e-01 -1.66615292e-01 -7.18504608e-01 5.28916180e-01 -1.68798491e-01 -9.19806957e-01 -1.89090371e-01 6.03341050e-02 -3.47484909e-02 -5.40042341e-01 7.00448573e-01 -1.03294633e-01 1.68572456e-01 5.21907322e-02 1.28221869e-01 5.00010371e-01 8.02927375e-01 7.68362591e-03 9.91542697e-01 8.54448020e-01 2.95276791e-01 -9.17266011e-01 -6.00887179e-01 1.27698360e-02 -6.90903604e-01 -4.36028838e-01 8.36345553e-01 -1.25189602e+00 -8.19736362e-01 2.13001654e-01 -1.11577737e+00 -6.73630655e-01 -2.14469880e-02 8.27387512e-01 -6.88438535e-01 1.97323099e-01 -5.77015698e-01 -7.58091331e-01 1.81839205e-02 -1.28521848e+00 1.25745523e+00 2.85065114e-01 -2.06336036e-01 -1.02706492e+00 1.27725035e-01 2.43172562e-03 3.84890825e-01 3.79492193e-01 -5.47812879e-02 -3.82900536e-02 -1.00821972e+00 -4.75128651e-01 1.45251453e-01 2.34188080e-01 1.30179310e-02 -1.56648517e-01 -7.17317104e-01 -6.10283196e-01 -1.81447491e-01 -3.63793135e-01 7.56228626e-01 4.42117989e-01 9.86720920e-01 -2.31159300e-01 -6.81692898e-01 7.76610255e-01 1.37442863e+00 5.74593961e-01 4.35850263e-01 3.38930309e-01 7.85375357e-01 2.83543944e-01 8.97396684e-01 3.67217183e-01 2.62856930e-01 7.33708858e-01 9.12051380e-01 -2.24425092e-01 2.78356522e-01 -3.22375268e-01 4.01896864e-01 3.88743460e-01 -4.73469980e-02 -4.00627136e-01 -9.90377665e-01 6.26181901e-01 -1.64280641e+00 -7.50929117e-01 2.68293262e-01 2.26225090e+00 1.99959159e-01 4.68205959e-01 -2.28012398e-01 -1.41701922e-01 3.80599260e-01 1.56326979e-01 -4.73177791e-01 -6.33440793e-01 2.42174178e-01 -1.52050108e-01 1.08158910e+00 1.00102854e+00 -1.14837098e+00 1.16583276e+00 7.18435860e+00 9.42736492e-02 -1.10722017e+00 -3.14439178e-01 -5.43677947e-03 -5.80460131e-01 -5.78726791e-02 -1.63272135e-02 -9.72313404e-01 4.97558666e-03 1.13554132e+00 -6.43541887e-02 3.65145683e-01 1.09884655e+00 1.56154990e-01 -3.34857345e-01 -1.14810920e+00 9.74445224e-01 1.64029211e-01 -1.32704198e+00 -8.57133642e-02 -5.89275965e-03 5.65234601e-01 3.32069069e-01 -3.83059122e-02 4.29558270e-02 8.52380633e-01 -1.04707432e+00 6.27979636e-01 2.03860879e-01 7.06638277e-01 -9.91132021e-01 8.43149722e-01 3.97538215e-01 -1.30982244e+00 -9.67485458e-02 -6.89874589e-01 -4.88812715e-01 5.49492896e-01 2.13851407e-01 -1.19124222e+00 2.74872363e-01 4.91884947e-01 6.44148827e-01 -1.44690320e-01 9.42653060e-01 -2.72669613e-01 2.07048599e-02 -3.79053175e-01 -1.07981473e-01 6.56171560e-01 7.05497786e-02 7.77315319e-01 8.82810414e-01 5.45189440e-01 1.17496490e-01 3.28874499e-01 5.48542798e-01 1.71970531e-01 -8.59183967e-01 -1.58775687e+00 2.10412934e-01 3.87601554e-01 1.09576023e+00 -5.86326897e-01 -2.94696540e-01 -2.22827300e-01 9.92328584e-01 1.28478929e-01 3.62675577e-01 -9.73606646e-01 -7.91403294e-01 1.07775772e+00 1.57079756e-01 3.85253608e-01 -9.92657125e-01 8.08683708e-02 -8.99829447e-01 -1.24549881e-01 -5.37492812e-01 -1.32065237e-01 -1.20014524e+00 -5.88585019e-01 1.02388918e+00 2.59773824e-02 -1.65856671e+00 -7.71853507e-01 -9.68355060e-01 -4.56298977e-01 4.85103875e-01 -1.40516436e+00 -8.26515317e-01 -7.44021833e-01 3.91882569e-01 6.14372551e-01 -3.65370363e-01 6.86802149e-01 -1.49123505e-01 -2.39777416e-01 -7.06805512e-02 -3.37400138e-01 8.25743154e-02 2.60039330e-01 -9.15231049e-01 8.98906589e-01 1.06377840e+00 -4.07704934e-02 3.01535577e-01 9.10269260e-01 -7.39584446e-01 -1.53515112e+00 -1.23035276e+00 2.40579516e-01 -6.88439250e-01 3.27130377e-01 -3.46243590e-01 -3.54965270e-01 1.12978125e+00 2.96437532e-01 -2.10758448e-02 1.42883375e-01 -5.01890421e-01 2.06719339e-01 -1.59258202e-01 -9.61466074e-01 9.42143679e-01 1.08092225e+00 -4.04706776e-01 -5.69377303e-01 1.24083839e-01 5.76003253e-01 -9.29049075e-01 -2.45175675e-01 4.80263501e-01 4.69018221e-01 -9.60976303e-01 9.98380005e-01 -3.33101541e-01 1.11143865e-01 -6.99179471e-01 -5.50345182e-02 -1.55555034e+00 -1.55357003e-01 -7.15151966e-01 1.11370325e-01 1.30749688e-01 3.37551892e-01 -2.94149846e-01 8.69193375e-01 4.55824107e-01 -4.71254885e-01 -5.10975957e-01 -1.06406045e+00 -8.73845398e-01 -3.49621892e-01 -4.88734692e-01 2.88211972e-01 2.71768779e-01 -1.06697224e-01 2.18472898e-01 -3.98137271e-01 6.35096729e-01 6.21858299e-01 -2.14699730e-01 9.38575625e-01 -8.71510804e-01 6.67936280e-02 -6.35345757e-04 -8.80526125e-01 -1.36274397e+00 3.44811618e-01 -3.35879773e-01 5.39457321e-01 -1.74836445e+00 -6.36162102e-01 -8.56047049e-02 2.07620740e-01 2.54897267e-01 4.14031804e-01 2.72497475e-01 7.79127404e-02 -5.39560094e-02 -6.91732585e-01 5.86689830e-01 1.05058634e+00 -4.86563295e-02 -1.70607604e-02 1.82973163e-03 -1.87738821e-01 1.00113344e+00 9.15869176e-01 -2.32416227e-01 -6.80887759e-01 -7.63932765e-01 1.72767907e-01 2.81546921e-01 6.95314407e-01 -1.79202724e+00 5.68675101e-01 -1.22312084e-02 5.45027196e-01 -7.78876007e-01 9.47221756e-01 -1.08549321e+00 -9.90791991e-02 8.13992739e-01 -9.12556574e-02 6.31753325e-01 6.17023826e-01 6.49984956e-01 -2.33559549e-01 -1.11102775e-01 7.40435004e-01 -3.83582264e-01 -1.08813858e+00 1.47660434e-01 -1.09501481e+00 -1.99525103e-01 1.47940874e+00 -3.85031253e-01 -1.35059565e-01 -7.57592320e-01 -4.81800348e-01 2.78793484e-01 6.96843624e-01 5.50546527e-01 1.28117740e+00 -1.27302456e+00 -3.43616396e-01 5.37882507e-01 -9.65942368e-02 2.78924644e-01 1.03581361e-01 4.17387765e-03 -1.27703917e+00 7.54166663e-01 -4.87238944e-01 -7.10641325e-01 -1.17359269e+00 5.87269247e-01 7.16439545e-01 1.60090998e-01 -6.60617113e-01 7.69306540e-01 2.21667543e-01 -4.23580229e-01 1.89211786e-01 -4.74099278e-01 1.86573565e-01 -6.90335333e-01 6.39598727e-01 2.29923368e-01 -7.08642527e-02 -4.91671622e-01 -4.76930887e-01 5.99974632e-01 1.32196456e-01 -5.75065076e-01 8.58483195e-01 -1.79197311e-01 5.46567023e-01 3.08804929e-01 9.17580426e-01 -1.94790110e-01 -1.92911375e+00 4.27459717e-01 -3.88033509e-01 -4.70746458e-01 2.51493067e-01 -3.11403096e-01 -9.03708875e-01 1.05292797e+00 6.16926968e-01 -3.97881955e-01 7.75101066e-01 -5.29024482e-01 8.85947168e-01 9.99873757e-01 8.86785924e-01 -1.03643882e+00 2.36107800e-02 7.04675317e-01 8.05210054e-01 -1.15254354e+00 -1.33594096e-01 -1.89365923e-01 -8.00085723e-01 1.06686080e+00 9.86842334e-01 -4.22137409e-01 3.94584447e-01 8.87116432e-01 3.97088230e-01 -2.24646300e-01 -1.08745623e+00 -1.65159971e-01 2.14829203e-02 8.56552482e-01 -2.36840118e-02 -9.10136849e-02 5.37725091e-01 -1.51791140e-01 -2.64061809e-01 -2.91274209e-02 9.47277844e-01 1.37781775e+00 -7.78623879e-01 -5.38327456e-01 -2.82321632e-01 -3.26179266e-02 -6.90814033e-02 6.03742301e-02 -2.82938778e-01 8.94288123e-01 -1.78761452e-01 8.85125220e-01 4.71995682e-01 -7.52928436e-01 4.33595717e-01 -5.21765590e-01 5.24561226e-01 -5.38596749e-01 -2.91863263e-01 -2.79728115e-01 4.07116830e-01 -9.53023434e-01 -1.01413853e-01 -1.47986993e-01 -1.45093513e+00 -1.35787413e-01 8.89808312e-02 1.05466004e-02 8.22080255e-01 6.27393782e-01 6.00874305e-01 7.24603415e-01 5.41706383e-01 -1.43200910e+00 -2.72972286e-01 -4.58881766e-01 -3.44722569e-01 -1.83196694e-01 6.87137008e-01 -8.22980702e-01 -2.80860305e-01 -1.23802423e-01]
[4.893213272094727, 0.7395444512367249]
60a8c495-b466-484f-8e68-b7dd22110462
receptive-field-regularized-cnns-for-music
2007.13503
null
https://arxiv.org/abs/2007.13503v1
https://arxiv.org/pdf/2007.13503v1.pdf
Receptive-Field Regularized CNNs for Music Classification and Tagging
Convolutional Neural Networks (CNNs) have been successfully used in various Music Information Retrieval (MIR) tasks, both as end-to-end models and as feature extractors for more complex systems. However, the MIR field is still dominated by the classical VGG-based CNN architecture variants, often in combination with more complex modules such as attention, and/or techniques such as pre-training on large datasets. Deeper models such as ResNet -- which surpassed VGG by a large margin in other domains -- are rarely used in MIR. One of the main reasons for this, as we will show, is the lack of generalization of deeper CNNs in the music domain. In this paper, we present a principled way to make deep architectures like ResNet competitive for music-related tasks, based on well-designed regularization strategies. In particular, we analyze the recently introduced Receptive-Field Regularization and Shake-Shake, and show that they significantly improve the generalization of deep CNNs on music-related tasks, and that the resulting deep CNNs can outperform current more complex models such as CNNs augmented with pre-training and attention. We demonstrate this on two different MIR tasks and two corresponding datasets, thus offering our deep regularized CNNs as a new baseline for these datasets, which can also be used as a feature-extracting module in future, more complex approaches.
['Gerhard Widmer', 'Hamid Eghbal-zadeh', 'Paul Primus', 'Khaled Koutini', 'Shreyan Chowdhury', 'Verena Haunschmid']
2020-07-27
null
null
null
null
['music-classification']
['music']
[ 1.16954610e-01 -3.09971366e-02 4.88202460e-02 -2.84692086e-02 -5.71447432e-01 -3.98738474e-01 6.63093746e-01 -5.54105081e-02 -6.61979675e-01 2.99074113e-01 3.96234363e-01 6.79169893e-02 -4.25035566e-01 -6.21098518e-01 -7.47698545e-01 -4.84906077e-01 -3.82028073e-02 3.60176474e-01 1.25262722e-01 -5.90080380e-01 1.78737685e-01 4.73428428e-01 -1.81866372e+00 3.33509862e-01 3.53510678e-01 1.02426577e+00 1.86720744e-01 3.16219151e-01 2.05964800e-02 7.66229928e-01 -4.69218582e-01 -3.60048115e-01 3.74699652e-01 -2.74974018e-01 -9.85043883e-01 -3.45224917e-01 5.33113599e-01 1.96737483e-01 -5.90020120e-01 7.11177528e-01 8.69250357e-01 4.43309486e-01 5.89512110e-01 -7.95610726e-01 -6.06413722e-01 1.02753556e+00 -2.84562230e-01 -3.39019345e-04 -3.28793153e-02 1.25056967e-01 1.37240136e+00 -8.33557725e-01 6.87639236e-01 1.27161324e+00 9.61585641e-01 7.91100562e-01 -1.16355026e+00 -6.05435491e-01 5.05999252e-02 3.80899519e-01 -1.41489720e+00 -4.29294318e-01 9.29879308e-01 -2.04020306e-01 1.03184998e+00 3.03470880e-01 6.32762194e-01 1.34001493e+00 -6.47950917e-02 1.22948265e+00 6.82527781e-01 -3.32463652e-01 -1.06116369e-01 -2.30531707e-01 -2.52475232e-01 2.56286830e-01 -1.22026943e-01 -2.19627377e-03 -5.24471164e-01 1.78398594e-01 7.37025142e-01 -1.39909804e-01 -4.49894905e-01 -1.58311814e-01 -1.42080808e+00 7.99399137e-01 1.02059901e+00 7.75490880e-01 -3.24388564e-01 5.90114415e-01 6.33233845e-01 5.38163900e-01 5.69259822e-01 1.00073850e+00 -5.82594275e-01 1.35063892e-02 -1.36706209e+00 4.44800228e-01 5.67969441e-01 5.98518550e-01 5.53863585e-01 2.57460088e-01 -4.61301714e-01 1.11957705e+00 5.96421957e-02 -6.89581037e-02 6.63200378e-01 -5.89390337e-01 3.11435312e-01 5.52303553e-01 -4.02148515e-01 -8.69126201e-01 -7.27411151e-01 -1.31326640e+00 -1.10365903e+00 1.04172595e-01 3.71238679e-01 2.26797774e-01 -6.56508088e-01 2.01273847e+00 -1.95572674e-01 3.14512253e-01 -1.36094764e-01 1.04533410e+00 8.89455557e-01 2.20864519e-01 -1.34781718e-01 2.58652180e-01 1.22319341e+00 -1.15767395e+00 -4.41636980e-01 -1.33188859e-01 6.06892467e-01 -9.08195853e-01 1.27396202e+00 6.16160929e-01 -1.09293842e+00 -9.11884129e-01 -1.11121047e+00 -2.97241181e-01 -5.20714045e-01 2.66160250e-01 7.66757309e-01 1.97473899e-01 -1.31604636e+00 1.44166076e+00 -5.65557003e-01 -4.20887798e-01 6.80832207e-01 5.33272505e-01 -1.50935978e-01 2.34922826e-01 -1.18928647e+00 7.16641307e-01 2.94607043e-01 3.36858809e-01 -8.44102204e-01 -6.33801043e-01 -5.57441771e-01 3.56534302e-01 1.92000911e-01 -9.99291658e-01 1.11400557e+00 -1.08493268e+00 -1.58936334e+00 1.09634888e+00 4.30083603e-01 -7.25210249e-01 4.57540154e-01 -5.13105929e-01 -1.51486680e-01 -1.63878709e-01 -2.41841733e-01 7.89377451e-01 1.00669098e+00 -8.41385782e-01 1.01261092e-02 -3.35462630e-01 1.58853829e-01 -2.83033997e-02 -4.36460108e-01 -3.51729356e-02 -3.94688785e-01 -1.00938332e+00 5.95615283e-02 -1.17971599e+00 -4.22083110e-01 -4.73259658e-01 -6.81605935e-01 -3.48987877e-01 4.65387434e-01 -2.75025457e-01 1.18471777e+00 -2.22775984e+00 5.66472054e-01 9.51302983e-03 1.13937728e-01 5.70012510e-01 -6.60867691e-01 5.20640016e-01 -3.17399919e-01 9.61259454e-02 -3.29764187e-02 -4.53633666e-01 1.40700415e-01 -1.79699630e-01 -2.28878066e-01 4.27385539e-01 3.99777025e-01 1.30144322e+00 -6.57356977e-01 1.76802680e-01 2.16701493e-01 6.92454815e-01 -6.48693442e-01 3.97407673e-02 -4.50018734e-01 7.32791245e-01 -1.35371521e-01 2.24690855e-01 4.14944947e-01 -2.02386826e-01 -1.23831593e-01 -2.87333101e-01 6.61567273e-03 6.65869951e-01 -1.08862913e+00 2.44643855e+00 -6.60216331e-01 9.11479235e-01 -1.76035780e-02 -1.27212405e+00 9.65813458e-01 2.95582414e-01 5.64503312e-01 -8.07724535e-01 3.97413582e-01 3.94547462e-01 1.51326701e-01 -2.25964800e-01 4.88559425e-01 -5.47345057e-02 4.45588119e-02 3.62408608e-01 4.30895329e-01 -8.03235471e-02 1.79121479e-01 -8.13831296e-03 1.09263003e+00 2.14597315e-01 -7.79626817e-02 -4.21730280e-01 6.83370650e-01 -3.33492279e-01 2.59142458e-01 9.39847052e-01 3.09235007e-01 1.00323129e+00 2.51785934e-01 -4.91374671e-01 -9.06184614e-01 -5.02606690e-01 -3.33482593e-01 1.30787575e+00 -2.60462761e-01 -7.89339900e-01 -6.73717558e-01 -4.77878839e-01 -1.07995771e-01 2.11659104e-01 -5.06332040e-01 -3.48602623e-01 -6.53148949e-01 -6.75586283e-01 8.03730965e-01 4.97215629e-01 5.15558779e-01 -1.59865344e+00 -3.12896699e-01 4.96310294e-01 7.37402439e-02 -9.79734600e-01 -2.72912439e-02 4.53933626e-01 -9.24205005e-01 -1.07509935e+00 -1.03329456e+00 -7.46743977e-01 -8.72761607e-02 8.30258429e-02 1.43416917e+00 3.60483021e-01 -3.05358678e-01 2.33803853e-01 -4.87933010e-01 -4.71392870e-01 -8.85316357e-02 7.96465576e-01 -7.92081058e-02 3.90049294e-02 1.87290549e-01 -9.61840451e-01 -8.73045981e-01 1.34765297e-01 -1.11298275e+00 5.19684479e-02 5.61617613e-01 9.14836824e-01 3.11226040e-01 -3.05804878e-01 7.80185997e-01 -9.11220908e-01 6.60276473e-01 -1.35832191e-01 -2.45586619e-01 -1.86284408e-01 -3.46479207e-01 2.03723535e-01 6.37460113e-01 -4.45180625e-01 -3.11774462e-01 -1.61992431e-01 -6.76627100e-01 -5.07492006e-01 -2.08704203e-01 7.17808604e-01 -1.12751434e-02 -1.67221904e-01 9.50984776e-01 2.21518930e-02 -1.78285465e-01 -1.04816711e+00 2.92641550e-01 5.99210501e-01 4.74937886e-01 -4.18775290e-01 8.67830157e-01 2.54276365e-01 2.75582343e-01 -6.79482162e-01 -1.13740385e+00 -5.36620438e-01 -5.01013517e-01 5.11869155e-02 7.86691248e-01 -1.03940415e+00 -7.47165561e-01 4.34589088e-01 -9.79022563e-01 -5.07122517e-01 -5.80471337e-01 5.47765791e-01 -7.20726252e-01 1.48041770e-01 -8.17791104e-01 -4.82531279e-01 -4.92854267e-01 -1.14550996e+00 1.20230114e+00 6.13469966e-02 -1.83257699e-01 -9.15490508e-01 1.43609464e-01 2.13288054e-01 9.29923177e-01 4.38080691e-02 9.35773730e-01 -8.76897812e-01 -4.69027817e-01 -8.83589610e-02 -1.74899310e-01 5.68050683e-01 -3.95199805e-01 -3.19806099e-01 -1.51252532e+00 -3.88213575e-01 -3.48002255e-01 -4.75195348e-01 1.61936951e+00 3.80094647e-01 1.58315313e+00 1.07090265e-01 3.69430706e-02 1.02460122e+00 1.33263159e+00 -5.01292706e-01 8.49780321e-01 6.41772151e-01 7.24982798e-01 3.83411139e-01 1.23634867e-01 2.19982207e-01 1.81993283e-02 1.08533835e+00 7.60643601e-01 -3.56642008e-01 -5.13095915e-01 -1.37920231e-01 1.08720735e-01 1.00345314e+00 -4.97674584e-01 -1.61141559e-01 -5.40030777e-01 4.17180896e-01 -2.11893296e+00 -7.96183288e-01 -1.56119213e-01 2.22174883e+00 5.99163055e-01 -7.68103004e-02 2.55800724e-01 5.11582971e-01 3.31990570e-01 3.88877124e-01 -3.78269732e-01 -2.26001218e-01 -3.68594170e-01 9.88616526e-01 1.15224697e-01 -2.08107755e-02 -1.43462980e+00 1.14828026e+00 5.62827873e+00 1.09877121e+00 -1.39332056e+00 1.13413505e-01 1.22620307e-01 -1.40617967e-01 -1.54173508e-01 -1.38913453e-01 -4.71809149e-01 1.57809928e-02 8.39754283e-01 4.78016645e-01 5.45524538e-01 7.53347456e-01 1.40390778e-02 5.39714038e-01 -1.24793053e+00 1.22053719e+00 -1.24822162e-01 -1.52269471e+00 1.42295912e-01 -5.00184763e-03 6.27566755e-01 4.91521478e-01 9.59490091e-02 8.06331754e-01 -2.87364602e-01 -1.26890612e+00 6.97644651e-01 3.40370119e-01 5.78205287e-01 -6.23845994e-01 8.37541819e-01 1.89046711e-01 -1.14951265e+00 -8.55757296e-02 -6.91212296e-01 -1.87087685e-01 -1.37475714e-01 5.55901349e-01 -3.03143471e-01 7.88214445e-01 6.02985203e-01 1.24651039e+00 -8.26387048e-01 1.29909372e+00 -1.29928514e-01 5.33933401e-01 -9.92677361e-02 1.16746485e-01 4.34263498e-01 6.68943394e-03 6.55100822e-01 1.37076557e+00 2.10093647e-01 -7.28149891e-01 -1.25204578e-01 9.90300179e-01 -3.57702672e-01 3.82777572e-01 -4.86421198e-01 -1.26466468e-01 -2.50821799e-01 1.53464365e+00 -5.24860084e-01 -1.36891929e-02 -3.01656634e-01 8.40500832e-01 4.82344329e-01 5.20904243e-01 -5.88914096e-01 -4.55420285e-01 6.47864580e-01 3.19386065e-01 4.37212288e-01 -1.24218844e-01 -8.58525708e-02 -1.43531394e+00 -1.67336658e-01 -8.91529560e-01 3.00610691e-01 -7.56117463e-01 -1.29484248e+00 8.83946180e-01 -5.00687182e-01 -1.42000544e+00 -2.49305919e-01 -7.90616930e-01 -5.75572371e-01 6.96843565e-01 -1.88628995e+00 -1.23486722e+00 -2.17386141e-01 9.59372222e-01 4.82487798e-01 -4.18075025e-01 9.96443689e-01 6.61878943e-01 -4.12613988e-01 6.99394464e-01 -4.71924394e-02 2.14722291e-01 7.39610612e-01 -1.17787457e+00 3.93799305e-01 3.73384774e-01 8.46816897e-01 5.64425409e-01 3.32738578e-01 9.85103473e-02 -1.27873588e+00 -1.08225930e+00 7.03527927e-01 -1.70008764e-01 5.36824942e-01 -6.21787429e-01 -9.11321580e-01 4.83941823e-01 2.47618049e-01 -5.54734766e-02 4.79881316e-01 7.25887835e-01 -4.89850461e-01 -1.13706812e-01 -4.94732082e-01 4.95978743e-01 1.39208877e+00 -6.72621310e-01 -4.03034568e-01 3.68955404e-01 5.18273354e-01 -1.93305179e-01 -7.05808043e-01 8.56563687e-01 5.73666394e-01 -1.13103294e+00 1.21721005e+00 -8.74050140e-01 5.17251790e-01 -3.64893638e-02 -3.40446644e-02 -1.36180532e+00 -5.00143468e-01 -6.74715281e-01 1.52942449e-01 1.09785068e+00 4.27288115e-01 -2.94339746e-01 7.02559233e-01 -2.46543288e-01 -4.15918320e-01 -8.14604044e-01 -9.33669150e-01 -8.09246898e-01 2.49323070e-01 -7.98079550e-01 3.89333010e-01 9.56424832e-01 -2.72987783e-01 6.29721224e-01 -4.64537501e-01 -4.76146549e-01 1.78802088e-01 1.05373248e-01 9.34446096e-01 -1.57370257e+00 -4.76287186e-01 -9.17868376e-01 -7.60242760e-01 -9.07206237e-01 3.54931265e-01 -1.45685923e+00 -1.97541595e-01 -1.40699267e+00 1.37408271e-01 -3.36647540e-01 -7.66243577e-01 5.20942807e-01 4.58229221e-02 7.02658951e-01 4.51887369e-01 2.82689124e-01 -7.28623688e-01 6.71634436e-01 1.33826768e+00 -2.61212647e-01 -2.45942518e-01 1.68714270e-01 -6.30648017e-01 7.50979364e-01 8.36340547e-01 -3.74362051e-01 -1.60605356e-01 -3.64181757e-01 7.49327123e-01 -3.11241299e-01 6.43389821e-01 -1.39633572e+00 -1.71109233e-02 5.36733747e-01 3.43755960e-01 -3.88428926e-01 3.73683721e-01 -6.56033933e-01 -4.76182718e-03 1.94603160e-01 -5.93077064e-01 -3.50571394e-01 3.66068423e-01 2.71655113e-01 -5.40049791e-01 -2.02807948e-01 5.46343386e-01 -2.31240138e-01 -5.07348061e-01 4.42419201e-01 -1.69797286e-01 9.09011960e-02 1.41109392e-01 1.03200048e-01 -1.04081385e-01 -4.67390716e-01 -9.94463801e-01 -2.05217376e-01 1.49665430e-01 7.47697651e-01 3.23533624e-01 -1.44860148e+00 -9.40370917e-01 1.32189602e-01 1.30390391e-01 -7.51697943e-02 2.98084348e-01 1.09162879e+00 -2.96132892e-01 5.09975672e-01 -2.50547260e-01 -7.15067029e-01 -9.52178657e-01 5.05972385e-01 3.84273440e-01 -5.59530318e-01 -8.53318930e-01 9.23631132e-01 2.56187946e-01 -5.48386455e-01 5.18440783e-01 -3.70821804e-01 -3.64096701e-01 1.36798933e-01 4.22914565e-01 -2.51231380e-02 4.67729896e-01 -4.74054724e-01 -2.98262686e-01 7.93004155e-01 1.31881654e-01 1.32310405e-01 1.80810320e+00 2.26962820e-01 -2.37541813e-02 3.06196928e-01 1.14255643e+00 -3.68674286e-02 -8.83561254e-01 -3.64514738e-01 1.68671191e-01 -1.75901711e-01 1.44417718e-01 -6.56923831e-01 -1.53512442e+00 1.22739768e+00 5.09900093e-01 2.19086707e-01 1.20109046e+00 1.79853588e-01 6.98585451e-01 4.06147301e-01 2.70668715e-01 -9.78716969e-01 1.47055954e-01 6.73427045e-01 1.26455843e+00 -9.90299106e-01 -2.45046213e-01 3.84790935e-02 -1.62189856e-01 1.26900172e+00 2.17455819e-01 -6.37937486e-01 6.44899428e-01 4.32175249e-02 -4.20358963e-02 -3.47778231e-01 -6.09109938e-01 -7.67385602e-01 8.02388787e-01 2.81782776e-01 9.15770590e-01 -1.95937291e-01 -2.52764195e-01 8.64038467e-01 -4.65844214e-01 1.35027498e-01 1.19119041e-01 4.66537207e-01 -1.39065146e-01 -1.25772774e+00 4.18223515e-02 2.87101656e-01 -7.79073417e-01 -3.05575341e-01 -4.99063730e-01 9.18230891e-01 2.29025245e-01 7.94964612e-01 -1.28149360e-01 -6.47382975e-01 5.21605849e-01 1.36385843e-01 6.88873589e-01 -6.63558364e-01 -1.34005141e+00 3.21864307e-01 -2.31524743e-02 -7.44939446e-01 -6.34153664e-01 -2.85056174e-01 -7.92957723e-01 -1.65516809e-01 -3.94198447e-01 -1.43627256e-01 7.10006714e-01 8.97971809e-01 4.53416288e-01 9.77685869e-01 2.87818760e-01 -1.34324825e+00 -5.22351921e-01 -1.35516512e+00 -7.17908323e-01 7.89439857e-01 1.65174529e-01 -6.11236453e-01 -3.66199076e-01 -4.81056184e-01]
[15.683359146118164, 5.2280049324035645]
620c2020-f8dd-4b96-8b81-ee05ba679662
turning-to-a-teacher-for-timestamp-supervised
2207.00712
null
https://arxiv.org/abs/2207.00712v1
https://arxiv.org/pdf/2207.00712v1.pdf
Turning to a Teacher for Timestamp Supervised Temporal Action Segmentation
Temporal action segmentation in videos has drawn much attention recently. Timestamp supervision is a cost-effective way for this task. To obtain more information to optimize the model, the existing method generated pseudo frame-wise labels iteratively based on the output of a segmentation model and the timestamp annotations. However, this practice may introduce noise and oscillation during the training, and lead to performance degeneration. To address this problem, we propose a new framework for timestamp supervised temporal action segmentation by introducing a teacher model parallel to the segmentation model to help stabilize the process of model optimization. The teacher model can be seen as an ensemble of the segmentation model, which helps to suppress the noise and to improve the stability of pseudo labels. We further introduce a segmentally smoothing loss, which is more focused and cohesive, to enforce the smooth transition of the predicted probabilities within action instances. The experiments on three datasets show that our method outperforms the state-of-the-art method and performs comparably against the fully-supervised methods at a much lower annotation cost.
['Yan Song', 'Yang Zhao']
2022-07-02
null
null
null
null
['action-segmentation']
['computer-vision']
[ 3.85687768e-01 8.57105702e-02 -4.64992255e-01 -6.20158494e-01 -6.53540552e-01 -2.17604294e-01 4.57950562e-01 5.02187200e-03 -4.45945084e-01 5.15536547e-01 1.78103477e-01 1.75340936e-01 2.38647938e-01 -3.48168343e-01 -6.37240827e-01 -9.13976550e-01 2.00289309e-01 1.97729632e-01 8.52551818e-01 3.27895522e-01 2.85303384e-01 -1.03310019e-01 -1.44566035e+00 3.50486249e-01 1.21462238e+00 9.96114552e-01 1.58562854e-01 1.39152631e-01 -1.47705778e-01 9.74015772e-01 -4.80859101e-01 -2.49359921e-01 2.28811234e-01 -7.30214953e-01 -9.04194832e-01 3.61880094e-01 1.02944344e-01 -1.30207196e-01 -7.58770481e-02 9.93537486e-01 1.66549966e-01 3.14710975e-01 5.16626835e-01 -1.23445356e+00 -1.43309534e-01 7.47812867e-01 -7.38998055e-01 -7.41062090e-02 1.55372515e-01 7.57906064e-02 8.51981282e-01 -3.35800201e-01 5.69149375e-01 1.13163579e+00 5.58216214e-01 5.62701821e-01 -1.11465931e+00 -5.94326556e-01 5.72663784e-01 3.77515674e-01 -1.17296171e+00 -2.00850114e-01 7.65759528e-01 -3.51217449e-01 4.62435961e-01 8.96860063e-02 7.95593858e-01 1.03002870e+00 4.25292253e-02 1.16847920e+00 9.39937711e-01 -4.56888527e-01 3.01654994e-01 -3.32307704e-02 9.61932763e-02 6.65190220e-01 -3.38195056e-01 -2.86989272e-01 -5.54883480e-01 -6.44860789e-03 7.08973587e-01 -1.16942927e-01 -1.99042231e-01 -3.57111722e-01 -1.13643658e+00 5.88523030e-01 2.09030539e-01 4.05892074e-01 -2.67481238e-01 7.04498589e-02 4.80399668e-01 -1.66125864e-01 6.55692697e-01 1.33476079e-01 -4.73391324e-01 -6.34468198e-01 -1.20072079e+00 3.36450920e-03 5.70173323e-01 7.33654559e-01 5.79953611e-01 -1.09093696e-01 -5.50556421e-01 7.71920204e-01 4.72373039e-01 -7.13479966e-02 5.44959009e-01 -1.36724746e+00 4.46632385e-01 8.17621768e-01 1.09709695e-01 -5.82022429e-01 -1.77417591e-01 -6.19965196e-02 -5.23588181e-01 2.29900517e-02 7.08060086e-01 1.23655736e-01 -1.10760903e+00 1.71581721e+00 6.28213227e-01 6.70578122e-01 -3.14505786e-01 9.27796125e-01 2.79684126e-01 6.38493001e-01 3.78520191e-01 -4.97859687e-01 1.07881141e+00 -1.36491764e+00 -1.02569711e+00 -1.33092538e-01 8.20320129e-01 -6.05868876e-01 1.10694253e+00 2.95606136e-01 -8.53241980e-01 -7.18975604e-01 -8.68780613e-01 6.40158355e-02 3.79482168e-03 3.40789288e-01 4.12860870e-01 4.13655579e-01 -6.32387817e-01 8.33383858e-01 -1.38429630e+00 -2.81598687e-01 5.91076314e-01 2.51446068e-01 -5.85812517e-02 1.75929710e-01 -1.11867821e+00 6.87236309e-01 4.57847416e-01 1.33564979e-01 -6.60500824e-01 -3.92731220e-01 -8.77268970e-01 -3.78956273e-02 7.43544638e-01 -2.80045718e-01 1.49457800e+00 -1.33758783e+00 -1.88608825e+00 6.66224897e-01 -2.74606854e-01 -5.66118538e-01 7.34325349e-01 -3.87223512e-01 1.04795618e-03 2.32204542e-01 2.60355055e-01 8.99580419e-01 9.94434237e-01 -1.01470435e+00 -8.53611648e-01 -1.16256453e-01 2.92698778e-02 2.80485094e-01 -3.84618223e-01 5.37040196e-02 -9.52514529e-01 -7.49603033e-01 1.59759358e-01 -1.14008927e+00 -4.45608646e-01 -2.98349798e-01 -3.12463343e-01 -5.16298354e-01 8.56194854e-01 -7.16824412e-01 1.53244579e+00 -2.35120368e+00 1.73012480e-01 -1.17350996e-01 -1.73895270e-01 3.31229568e-01 1.62911534e-01 1.80009007e-01 2.94352788e-02 -1.07101006e-02 -5.95175862e-01 -7.65266538e-01 -2.02061921e-01 4.27607447e-01 -4.62302640e-02 4.46913332e-01 7.73326606e-02 5.79598129e-01 -9.83990729e-01 -9.50509846e-01 2.84033477e-01 3.42897505e-01 -4.28739965e-01 4.05045301e-01 -5.55990994e-01 9.42372859e-01 -6.85519457e-01 3.99551243e-01 3.13161910e-01 -3.38438839e-01 1.92096576e-01 2.09288560e-02 -1.35624751e-01 3.36617559e-01 -1.09217894e+00 1.94357300e+00 9.58665162e-02 2.55011320e-01 -2.09472477e-01 -1.23073471e+00 6.69828713e-01 3.58077198e-01 8.84871006e-01 -4.83922571e-01 2.25629747e-01 -1.08045109e-01 -1.75957635e-01 -6.87180161e-01 3.36742193e-01 6.57375604e-02 1.19651191e-01 3.87226850e-01 -1.81736395e-01 1.04586527e-01 4.12809193e-01 7.33170733e-02 6.86841488e-01 8.85136783e-01 5.54063171e-02 5.42780124e-02 5.74502289e-01 2.60486007e-02 1.09500027e+00 4.20579970e-01 -3.73693883e-01 6.34504974e-01 6.39712274e-01 -2.45089889e-01 -7.13524103e-01 -5.34533918e-01 3.87696028e-02 9.72036421e-01 3.19173068e-01 -6.24331951e-01 -1.26825356e+00 -1.08082438e+00 -3.03030640e-01 5.89690149e-01 -4.45761740e-01 -2.08683416e-01 -7.94077873e-01 -6.82959199e-01 2.95585573e-01 7.79526353e-01 7.30618358e-01 -1.01528931e+00 -5.72773695e-01 3.14878851e-01 -5.53514004e-01 -1.33085382e+00 -6.34750068e-01 3.45087685e-02 -1.14179111e+00 -1.04736900e+00 -6.86430752e-01 -6.77040398e-01 7.46892810e-01 -1.70962065e-02 6.93211675e-01 -3.83005515e-02 1.43007740e-01 1.92912430e-01 -5.95941067e-01 -1.58179864e-01 -4.49839294e-01 2.16218337e-01 -9.52827930e-02 4.11319256e-01 2.73293346e-01 -3.70107472e-01 -6.02529109e-01 5.43092132e-01 -9.47503567e-01 2.88572520e-01 2.03191370e-01 6.52426720e-01 7.28533685e-01 8.08508694e-02 3.33712965e-01 -8.86190295e-01 2.14202181e-01 -1.03838764e-01 -6.80645823e-01 2.35148057e-01 -6.73394442e-01 9.94678810e-02 6.61277354e-01 -6.26113296e-01 -1.30747557e+00 5.21103084e-01 8.19117650e-02 -5.17327487e-01 -1.80617139e-01 3.59327585e-01 -1.82908565e-01 1.35431454e-01 2.79134840e-01 9.41094235e-02 2.32656579e-02 -5.84697783e-01 3.24679822e-01 5.21071553e-01 3.36588800e-01 -4.13918734e-01 4.06383783e-01 5.54859817e-01 -1.90892249e-01 -5.07736802e-01 -1.08275223e+00 -5.53207636e-01 -9.55463290e-01 -4.70209360e-01 1.03056490e+00 -7.68879116e-01 -3.35650951e-01 7.31463075e-01 -1.01618516e+00 -6.55200243e-01 -3.45113426e-01 5.51571548e-01 -6.27766609e-01 6.51252329e-01 -7.48767674e-01 -8.06811333e-01 9.01268721e-02 -1.19701529e+00 1.11288881e+00 4.26450938e-01 -2.50028819e-01 -8.52920651e-01 4.62709926e-02 5.00274956e-01 -3.11650839e-02 1.06006481e-01 5.14162242e-01 -4.84234184e-01 -6.17517948e-01 -2.10157543e-01 2.94116028e-02 6.56903028e-01 2.07665488e-01 1.73526481e-01 -8.45840991e-01 2.37077363e-02 2.20098495e-01 -2.49901518e-01 1.01219273e+00 5.37330389e-01 1.32881594e+00 -9.19033289e-02 -4.94252950e-01 3.30976993e-01 8.97363484e-01 3.09898257e-01 7.11461365e-01 4.53623563e-01 8.82684827e-01 7.25348175e-01 1.28065300e+00 4.39094245e-01 4.31818545e-01 8.33493531e-01 2.29008332e-01 -4.94618751e-02 1.05381787e-01 -3.13128859e-01 4.75578487e-01 7.88254797e-01 -2.51489401e-01 -1.24707781e-01 -6.44589186e-01 3.28688771e-01 -2.39142060e+00 -8.48497272e-01 -1.57895029e-01 2.15482259e+00 1.03792894e+00 3.28467250e-01 3.40880185e-01 3.25762480e-01 7.69267261e-01 4.68572378e-01 -4.07442182e-01 4.51123379e-02 2.98635930e-01 -2.40129083e-01 3.08144897e-01 3.15065056e-01 -1.34706581e+00 1.23106647e+00 6.21801043e+00 9.43141401e-01 -1.13417602e+00 1.28433943e-01 7.33322263e-01 1.31208792e-01 2.17369527e-01 2.83203989e-01 -8.05126309e-01 7.72992790e-01 8.16492379e-01 7.54082054e-02 8.57570693e-02 8.84317696e-01 7.00095534e-01 -5.16630054e-01 -1.13367367e+00 7.93866634e-01 -1.35737052e-02 -8.52879405e-01 -2.49902338e-01 -1.23491451e-01 8.35571706e-01 -2.64916897e-01 -2.48411566e-01 1.33241102e-01 2.97739580e-02 -5.86035728e-01 8.60823393e-01 6.21797502e-01 2.98948497e-01 -5.17317295e-01 6.42894089e-01 6.04276001e-01 -1.26121891e+00 -4.81224693e-02 -1.21657252e-01 -1.18420921e-01 4.79685038e-01 5.68175137e-01 -4.73370671e-01 5.61186552e-01 6.46154881e-01 1.09841228e+00 -4.74843889e-01 1.10814917e+00 -5.95841527e-01 8.90777171e-01 -2.87131488e-01 2.69351453e-01 3.86220425e-01 -4.70809162e-01 3.70552868e-01 1.04190564e+00 -3.17029618e-02 1.21305346e-01 5.91615260e-01 5.95751703e-01 2.46451557e-01 1.63357109e-01 -1.84018999e-01 -1.83299854e-01 2.04142198e-01 1.12970567e+00 -1.05274653e+00 -5.99880159e-01 -3.69439960e-01 1.06798828e+00 1.56965002e-01 3.50564331e-01 -1.30433023e+00 6.98316246e-02 2.18881547e-01 1.46210164e-01 2.97210485e-01 -2.71271497e-01 -2.79875487e-01 -1.16744304e+00 2.47448072e-01 -7.16064870e-01 5.50936222e-01 -4.45070297e-01 -8.58604670e-01 3.74466896e-01 1.75074786e-01 -1.50683391e+00 -2.02935860e-01 -1.01179034e-01 -5.25330245e-01 3.49353969e-01 -1.40497601e+00 -9.23311770e-01 -2.10540950e-01 4.24802035e-01 7.75541127e-01 3.09795082e-01 3.98725420e-01 3.35339189e-01 -8.52698445e-01 3.92508417e-01 -2.35294178e-01 2.99481004e-02 9.09808099e-01 -1.09898961e+00 3.48775685e-02 1.00521314e+00 1.39692903e-01 2.90466905e-01 5.35196185e-01 -8.27128708e-01 -6.64669156e-01 -1.13944232e+00 8.71852636e-01 -3.03848863e-01 4.89398748e-01 -1.07603796e-01 -1.09992671e+00 6.59351528e-01 1.42425060e-01 -1.48834765e-01 5.76282263e-01 -1.64803818e-01 2.93672215e-02 -2.06037313e-01 -8.23537529e-01 6.16539657e-01 1.02002621e+00 -2.08581135e-01 -6.14521980e-01 3.22441906e-01 8.07204068e-01 -5.49163222e-01 -7.25645959e-01 5.86618066e-01 4.78768498e-01 -1.04912853e+00 5.33395350e-01 -1.99568048e-01 3.25972795e-01 -4.53763276e-01 3.87900412e-01 -9.99718666e-01 -2.10244283e-01 -7.42169440e-01 -1.98099196e-01 1.55624306e+00 2.22886875e-01 -3.62617642e-01 9.53446984e-01 7.07520068e-01 -1.56533495e-01 -8.29308748e-01 -1.00763440e+00 -8.66652846e-01 -3.96109253e-01 -4.35479164e-01 2.50070721e-01 7.74171233e-01 1.07072368e-01 1.80136159e-01 -4.44825023e-01 2.53549386e-02 5.08395314e-01 2.78544612e-02 6.15517795e-01 -1.02707863e+00 -6.81305304e-02 -3.86642516e-01 -2.00595111e-01 -1.41053939e+00 3.13604027e-01 -4.74255830e-01 4.13021147e-01 -1.43171191e+00 1.51372701e-01 -3.82573187e-01 -3.41691643e-01 6.57227159e-01 -4.01493341e-01 5.03071360e-02 1.17701188e-01 3.50859940e-01 -1.02184749e+00 8.29861999e-01 1.38792288e+00 8.78867283e-02 -5.50435126e-01 2.29501471e-01 -2.48452932e-01 9.54830289e-01 7.30764687e-01 -7.41943598e-01 -4.69961077e-01 -2.37494528e-01 -3.48757505e-01 -4.31692153e-02 8.49866122e-02 -1.09898627e+00 3.41555119e-01 -3.34754765e-01 -1.63265057e-02 -5.82313418e-01 2.71121234e-01 -7.89661050e-01 2.54537482e-02 3.66362184e-01 -3.94038707e-01 -3.34291518e-01 -1.41957924e-01 6.72123730e-01 -4.80271310e-01 -2.38061234e-01 9.66551185e-01 -1.14693537e-01 -6.59659505e-01 3.80085051e-01 -2.92438477e-01 -6.55407310e-02 1.18115520e+00 -4.01166379e-01 1.30770102e-01 -2.18011960e-01 -6.45209134e-01 5.71333945e-01 4.66253072e-01 4.79243577e-01 2.56540656e-01 -1.20993936e+00 -2.53373206e-01 1.58640593e-02 -1.12393849e-01 2.21635684e-01 1.23297811e-01 1.21048486e+00 -1.99912280e-01 2.09586054e-01 5.36814630e-02 -8.41317356e-01 -1.30234146e+00 4.49076027e-01 2.31272295e-01 -5.02062440e-01 -7.43587315e-01 7.21110821e-01 1.79743558e-01 -1.50747597e-02 6.55102491e-01 -5.48580050e-01 -4.52452272e-01 1.95474342e-01 3.04004163e-01 4.18465108e-01 -3.35992008e-01 -5.67110598e-01 -3.00307602e-01 7.11370826e-01 -5.02018221e-02 -2.34464318e-01 1.19696021e+00 -2.68990546e-01 -9.70595106e-02 7.91814864e-01 8.40230703e-01 -3.30310792e-01 -1.80183697e+00 -1.54068112e-01 3.05234075e-01 -4.75937843e-01 -2.51646578e-01 -4.30294216e-01 -1.01624715e+00 6.41043961e-01 3.65736425e-01 3.36796075e-01 1.25068855e+00 -1.43194735e-01 1.00108457e+00 -1.51368380e-01 2.92209297e-01 -1.53469670e+00 3.07506144e-01 4.45683897e-01 3.66376609e-01 -1.25550473e+00 -6.22484349e-02 -7.07981765e-01 -9.25651431e-01 9.68093276e-01 7.47794151e-01 3.08325309e-02 4.40174729e-01 2.76645888e-02 1.64106354e-01 1.12160392e-01 -4.98097479e-01 -1.98312968e-01 3.06403220e-01 2.39746496e-01 3.08907598e-01 -2.46136263e-01 -7.01163948e-01 5.62853277e-01 2.79029906e-01 3.17565471e-01 5.59157208e-02 9.73763943e-01 -4.12218571e-01 -1.45811236e+00 -1.44061059e-01 1.83991328e-01 -6.19464457e-01 1.72117591e-01 -2.67132550e-01 5.09079337e-01 3.04967552e-01 9.65797186e-01 -2.22998381e-01 -2.61344165e-01 2.21068412e-01 3.66152704e-01 3.00358683e-01 -6.35311663e-01 -4.78969604e-01 5.25489092e-01 4.04611044e-02 -8.52127671e-01 -1.09537613e+00 -8.07380736e-01 -1.53659725e+00 3.18096280e-01 -5.73371768e-01 2.76837826e-01 3.29403698e-01 1.23702598e+00 2.33702525e-01 6.05749130e-01 7.34462261e-01 -7.78105140e-01 -4.25741464e-01 -1.02654958e+00 -4.72353816e-01 5.49558997e-01 -7.19551649e-03 -6.67178154e-01 -3.70830715e-01 5.70960462e-01]
[8.480764389038086, 0.652700662612915]
53d339f4-1f8f-400a-aa8f-c374ae5ff2a4
the-theory-of-artificial-immutability
2205.01166
null
https://arxiv.org/abs/2205.01166v1
https://arxiv.org/pdf/2205.01166v1.pdf
The Theory of Artificial Immutability: Protecting Algorithmic Groups Under Anti-Discrimination Law
Artificial Intelligence (AI) is increasingly used to make important decisions about people. While issues of AI bias and proxy discrimination are well explored, less focus has been paid to the harms created by profiling based on groups that do not map to or correlate with legally protected groups such as sex or ethnicity. This raises a question: are existing equality laws able to protect against emergent AI-driven inequality? This article examines the legal status of algorithmic groups in North American and European non-discrimination doctrine, law, and jurisprudence and will show that algorithmic groups are not comparable to traditional protected groups. Nonetheless, these new groups are worthy of protection. I propose a new theory of harm - "the theory of artificial immutability" - that aims to bring AI groups within the scope of the law. My theory describes how algorithmic groups act as de facto immutable characteristics in practice that limit people's autonomy and prevent them from achieving important goals.
['Sandra Wachter']
2022-05-02
null
null
null
null
['jurisprudence']
['miscellaneous']
[ 5.96429765e-01 8.63997638e-01 -6.76320314e-01 -5.18876731e-01 -4.69802320e-02 -4.84798461e-01 7.15243220e-01 2.27696270e-01 -7.50129163e-01 8.55925620e-01 9.98547912e-01 -7.66072631e-01 -4.99843150e-01 -7.72127151e-01 -1.34739146e-01 -4.72784013e-01 4.29027826e-01 5.33143103e-01 -6.06796980e-01 -4.32626903e-01 6.37946546e-01 4.04974282e-01 -1.40726924e+00 -1.18612193e-01 1.78126550e+00 1.81957811e-01 -1.21145976e+00 1.45105511e-01 1.82142667e-02 1.15310073e+00 -7.89085567e-01 -8.51423800e-01 6.99180484e-01 -8.43277872e-01 -5.02573669e-01 -4.55327511e-01 8.51135790e-01 -5.11112809e-01 -2.04468727e-01 1.34626102e+00 5.46344757e-01 -2.70869285e-01 9.40828323e-01 -1.43845963e+00 -1.33442485e+00 1.05861068e+00 -3.96970034e-01 1.55927557e-02 2.35419422e-01 3.84178251e-01 7.79341996e-01 1.14149123e-01 6.30649805e-01 1.36766756e+00 8.20250034e-01 1.14621520e+00 -1.30029976e+00 -9.76426125e-01 -8.64262786e-03 -2.02770293e-01 -1.32791686e+00 -6.93483710e-01 5.14786780e-01 -9.79328632e-01 2.46963024e-01 9.15070832e-01 1.05579019e+00 8.70015979e-01 5.16181588e-01 1.57043800e-01 9.80271101e-01 -2.69791842e-01 3.83339494e-01 9.23050568e-02 2.98229069e-01 4.69054937e-01 1.53737128e+00 1.90921545e-01 -3.67117912e-01 -7.38672733e-01 3.24467152e-01 -2.87822962e-01 -5.90276979e-02 -1.93732113e-01 -1.03665292e+00 9.81280386e-01 2.23772690e-01 5.22652447e-01 -3.49240363e-01 7.96181858e-02 2.78611481e-01 6.15959764e-02 4.22047526e-01 1.12127161e+00 -5.30440956e-02 -1.98594466e-01 -4.95589852e-01 6.74949110e-01 6.53251946e-01 2.26616323e-01 1.95445642e-01 1.68309852e-01 -1.50591776e-01 2.26867661e-01 3.77012789e-01 6.51943922e-01 2.20792189e-01 -1.38560510e+00 2.16190383e-01 9.37432408e-01 1.73861429e-01 -1.31284535e+00 -3.42322677e-01 -2.90283591e-01 -6.84303403e-01 8.45253050e-01 9.20498252e-01 -5.74433446e-01 -7.11835623e-01 1.90867352e+00 1.83764696e-01 -8.12560081e-01 7.58977979e-02 9.26463246e-01 3.53260964e-01 4.98141833e-02 6.68946087e-01 2.39526723e-02 1.04102075e+00 -9.35579836e-02 -7.95673370e-01 -3.28461707e-01 8.62318575e-01 -5.03686070e-02 7.99536824e-01 5.17066009e-02 -1.12348342e+00 4.17125784e-02 -9.84990120e-01 -3.39558184e-01 -5.17268777e-01 -7.67445147e-01 1.18970001e+00 1.88468528e+00 -5.87082922e-01 4.48963344e-01 -1.19322680e-01 -1.66788146e-01 1.05719197e+00 4.52347755e-01 7.67213032e-02 5.41228235e-01 -1.11780131e+00 1.07169724e+00 1.56769589e-01 8.36783126e-02 -6.79294318e-02 -9.71745312e-01 -7.44273841e-01 -1.57385945e-01 1.72014192e-01 -7.74634659e-01 4.64874774e-01 -1.81729424e+00 -1.03706932e+00 1.30973876e+00 3.89427453e-01 -7.20419049e-01 8.40567648e-01 -4.04060811e-01 -7.94129610e-01 -1.67766005e-01 4.20485497e-01 7.31898308e-01 1.85497090e-01 -1.22463274e+00 -5.14754951e-01 -8.15589130e-01 2.50070155e-01 3.23686510e-01 -3.72634739e-01 3.94883335e-01 1.24715006e+00 -5.39996684e-01 -2.45393291e-01 -8.16462934e-01 -3.02575618e-01 1.34530485e-01 -4.07602668e-01 -8.63039643e-02 3.49841624e-01 -4.48828399e-01 1.21908092e+00 -1.97778857e+00 -3.77609402e-01 3.24075460e-01 6.69768393e-01 4.36178088e-01 3.57253432e-01 6.29048198e-02 2.83105075e-01 1.01481211e+00 -2.17225194e-01 7.78947413e-01 5.68590105e-01 5.67412712e-02 -6.55027181e-02 7.47361958e-01 -3.73698980e-01 8.61130297e-01 -7.44218528e-01 -4.40603316e-01 -1.05448686e-01 2.07506821e-01 -8.27242017e-01 -5.43330669e-01 1.64569467e-01 4.45526034e-01 -4.09728795e-01 6.11786842e-01 5.93213439e-01 3.32219660e-01 2.96661526e-01 5.47790110e-01 -4.51986283e-01 2.47723192e-01 -5.02029479e-01 8.45506132e-01 6.82644248e-01 6.39615297e-01 7.88395032e-02 -5.23312271e-01 7.50949502e-01 3.28470208e-02 3.46813619e-01 -6.24672830e-01 5.37441254e-01 3.79232466e-01 1.25455117e+00 -4.89483148e-01 5.80866337e-01 -3.81065607e-01 -3.39182734e-01 5.50367534e-01 -8.29903424e-01 -2.18674690e-01 -1.85191020e-01 -1.11847691e-01 7.30946422e-01 3.03894542e-02 4.43033636e-01 -1.06581914e+00 3.09048444e-01 3.17725748e-01 1.18756056e+00 8.84255230e-01 -9.18843806e-01 1.68609425e-01 4.36862081e-01 -6.76230788e-01 -1.24479759e+00 -9.55768645e-01 -2.45402366e-01 1.04909694e+00 2.84581661e-01 1.72619760e-01 -1.09204590e+00 -7.95746386e-01 5.63732088e-01 1.37631214e+00 -9.30914760e-01 -5.45734882e-01 -2.87936449e-01 -7.11409509e-01 9.63174284e-01 -2.79318020e-02 5.78200877e-01 -7.65434325e-01 -1.21594667e+00 -3.27005893e-01 3.66739452e-01 -3.73457998e-01 -2.22882956e-01 -5.48762202e-01 -3.45883310e-01 -8.07415247e-01 -3.45868766e-01 -8.58501494e-02 6.41820490e-01 -3.36416841e-01 7.85735130e-01 3.77282977e-01 -1.63021490e-01 2.21645758e-01 3.65336537e-02 -1.28697073e+00 -8.44655097e-01 7.62616172e-02 8.62380713e-02 4.06151861e-02 1.08529687e+00 -6.01556897e-01 -6.28383338e-01 -3.12789865e-02 -6.74540699e-01 6.59323558e-02 2.83505231e-01 2.56649754e-03 -3.79400373e-01 -2.81379431e-01 1.04078901e+00 -1.37269163e+00 8.02809656e-01 -5.18425882e-01 -1.64152414e-01 3.41458358e-02 -8.54133248e-01 -2.76719093e-01 3.39244872e-01 -4.03595835e-01 -1.13094544e+00 -7.03909278e-01 1.05888493e-01 6.13698125e-01 -3.90988797e-01 7.55816549e-02 -5.35566270e-01 -1.80031851e-01 1.11088991e+00 -5.27209282e-01 2.52980411e-01 -9.30386186e-02 3.30668390e-01 9.13384259e-01 4.05078322e-01 -7.29629219e-01 7.25417674e-01 6.58236742e-01 -1.42683968e-01 -7.74103880e-01 -8.84675026e-01 4.82375741e-01 -2.98785776e-01 -3.75968575e-01 1.09435022e+00 -5.22249341e-01 -9.42465901e-01 -7.02013075e-02 -5.03489375e-01 -3.09857816e-01 -8.74463260e-01 6.46590710e-01 -3.07536513e-01 1.67973727e-01 -2.57649899e-01 -1.27224123e+00 -4.65685844e-01 -5.41337550e-01 2.86969304e-01 4.51301217e-01 -9.43790972e-01 -7.11261511e-01 -2.10956708e-01 9.25663829e-01 5.16084194e-01 8.73503029e-01 9.43919301e-01 -1.09934771e+00 -1.08326729e-02 -1.90526918e-01 1.31873965e-01 -2.70526279e-02 1.56441957e-01 -5.58935851e-02 -7.26536036e-01 2.90225178e-01 1.93470076e-01 -4.64692637e-02 2.54250199e-01 1.95850834e-01 4.77042824e-01 -8.88161480e-01 -2.11283177e-01 3.13690871e-01 1.21671510e+00 8.66752505e-01 9.34463799e-01 3.85067016e-01 7.38301992e-01 1.10951340e+00 1.63229555e-01 2.89980710e-01 2.93051809e-01 1.61356434e-01 3.60679291e-02 2.86564813e-03 1.74166903e-01 -2.08830252e-01 2.23480120e-01 1.13581643e-01 -5.46058536e-01 -1.05074070e-01 -1.33271086e+00 4.14702982e-01 -1.72850871e+00 -1.29823709e+00 -4.55976337e-01 2.16282129e+00 6.65837765e-01 3.95453662e-01 4.99260545e-01 8.39145854e-02 8.42150509e-01 -1.83257051e-02 -7.81376183e-01 -1.06721139e+00 -2.36469567e-01 -2.04306126e-01 7.66374528e-01 6.01464331e-01 -7.41734982e-01 8.04793775e-01 7.02085209e+00 5.36767542e-01 -5.69948375e-01 5.77639788e-02 9.81947899e-01 -2.10536104e-02 -9.90743458e-01 8.12376067e-02 -3.57290000e-01 5.68971932e-01 7.05699563e-01 -1.03391051e+00 3.48499939e-02 3.95849556e-01 2.14510188e-01 -4.41370457e-02 -8.59670520e-01 2.83183783e-01 1.86036915e-01 -8.83431077e-01 2.13498682e-01 8.54246616e-01 8.77746284e-01 -6.13749683e-01 3.60041648e-01 -5.89957461e-02 7.94446409e-01 -1.48166132e+00 1.02199721e+00 4.06316280e-01 5.41848123e-01 -8.63233805e-01 5.36555052e-01 2.51071185e-01 7.15966299e-02 -3.42622578e-01 -4.89915848e-01 -9.23818648e-01 -5.12980334e-02 4.57247257e-01 -1.18069842e-01 3.14643583e-03 9.68377963e-02 -3.03785969e-02 -2.89505750e-01 6.09100580e-01 -5.57712540e-02 3.57926905e-01 -1.44402869e-02 -1.65057003e-01 2.48769611e-01 -4.65562165e-01 7.49233127e-01 6.66278839e-01 1.48995826e-02 4.19891328e-01 -2.62109220e-01 1.10356843e+00 2.13609710e-01 6.00310639e-02 -9.23144698e-01 -4.21677887e-01 5.61121523e-01 6.99352920e-01 -5.54799557e-01 -2.73581445e-01 -3.19319576e-01 2.57999569e-01 -4.31104958e-01 1.48656130e-01 -8.54048789e-01 -3.05439293e-01 1.24550128e+00 5.38251817e-01 -7.79522777e-01 3.29806693e-02 -9.84385014e-01 -9.94869709e-01 -7.93074846e-01 -1.33600354e+00 3.78711760e-01 -1.26378298e-01 -1.17191470e+00 -2.46805444e-01 -1.91168115e-01 -4.36045200e-01 -6.09266646e-02 -4.23008353e-01 -3.90589759e-02 6.72272742e-01 -4.70286459e-01 -1.13365686e+00 4.51548919e-02 -4.98972945e-02 -2.52104223e-01 -2.44833991e-01 7.61742055e-01 -1.86422169e-02 -1.44163430e-01 8.00348878e-01 -5.19034825e-02 2.82520592e-01 3.54240924e-01 -9.69798446e-01 4.14040536e-01 7.95834303e-01 -4.28732395e-01 7.42312253e-01 8.59799564e-01 -1.02937603e+00 -1.13417554e+00 -4.18239921e-01 1.05098689e+00 -8.03246379e-01 4.19014990e-01 -3.37510735e-01 -5.58742940e-01 9.46480334e-01 3.42950404e-01 -7.38804221e-01 1.20339417e+00 3.22738171e-01 -4.66025710e-01 7.37708062e-02 -1.83090734e+00 1.16074491e+00 1.76665401e+00 -3.04356486e-01 -8.00674975e-01 8.79694670e-02 6.20560706e-01 5.13688922e-02 -7.91610658e-01 3.48950446e-01 1.03905058e+00 -1.08579278e+00 7.90414035e-01 -8.36405158e-01 3.56790304e-01 -8.90797004e-02 7.05409199e-02 -7.23434389e-01 -6.50765121e-01 -7.62352884e-01 6.51048899e-01 1.38394177e+00 5.28960526e-01 -1.51644945e+00 8.81625473e-01 1.75087249e+00 4.23312411e-02 -2.92429447e-01 -8.67190421e-01 -6.25346065e-01 8.00779223e-01 8.93500149e-02 1.01355135e+00 1.85659313e+00 4.37131643e-01 2.72933781e-01 -2.27749377e-01 -4.02311720e-02 9.11762476e-01 -2.28012592e-01 7.60075927e-01 -1.69172096e+00 3.28719258e-01 -9.09742117e-01 -8.83229315e-01 1.24372229e-01 4.66501899e-02 -7.53931403e-01 -3.56268108e-01 -1.28296924e+00 1.81110725e-01 -3.87505949e-01 5.03607653e-02 1.93374917e-01 -6.53248504e-02 1.27669601e-02 5.35556376e-01 8.38548243e-02 -1.00885190e-01 1.30990937e-01 1.03035784e+00 -4.57982868e-02 -1.68432325e-01 -5.35665035e-01 -2.01504993e+00 1.12278152e+00 1.16172171e+00 -3.38444263e-01 -3.42075795e-01 -2.43868649e-01 6.29230142e-01 -5.70883930e-01 2.73066074e-01 -1.25418580e+00 8.19749907e-02 -9.41904724e-01 4.47055310e-01 3.48122388e-01 -5.29912747e-02 -9.82260108e-01 5.68655431e-01 8.75874639e-01 -7.72361636e-01 -3.97378564e-01 -6.31681532e-02 -1.98018596e-01 6.18866682e-01 -2.06451699e-01 6.69832170e-01 6.18742704e-02 4.25360650e-02 -1.18065976e-01 -6.55620694e-01 4.59668756e-01 1.22980070e+00 -6.43433332e-01 -1.09167206e+00 -3.52724701e-01 -3.29022378e-01 1.20572373e-01 1.10305643e+00 1.54241174e-01 -5.08312657e-02 -1.06688356e+00 -1.06131530e+00 -2.58605331e-01 -1.04081936e-01 -5.96798241e-01 -1.48206977e-02 4.12221551e-01 -6.56963587e-01 2.36583635e-01 -5.47797382e-01 2.93867409e-01 -1.09787679e+00 7.15429306e-01 5.63842535e-01 5.90091407e-01 -6.99443936e-01 4.74119008e-01 4.36313331e-01 -2.17588767e-01 -7.94793814e-02 3.43946576e-01 -2.10483074e-01 -1.16474731e-02 6.08425736e-01 6.83365583e-01 -1.02454042e+00 -1.01972008e+00 -4.02849495e-01 3.51424932e-01 1.33332163e-01 -3.53664219e-01 9.36469495e-01 -5.62335551e-02 -3.03115100e-01 4.03055191e-01 4.42149192e-01 6.63463533e-01 -7.09111214e-01 7.74344146e-01 9.59471092e-02 -1.06700075e+00 -6.08275592e-01 -1.22983372e+00 -4.50900972e-01 5.05590975e-01 4.69585270e-01 7.13083088e-01 6.38280094e-01 -3.57671946e-01 3.44743013e-01 -4.35532182e-02 1.92937925e-01 -1.47398996e+00 -7.90700018e-01 -1.56812772e-01 8.50721538e-01 -4.88269061e-01 2.38231584e-01 -4.82325882e-01 -6.88895583e-01 3.16246897e-01 8.20624948e-01 -1.10445090e-01 1.45216689e-01 1.86836123e-01 3.79553080e-01 -1.00817546e-01 -2.42168471e-01 -2.72419959e-01 6.56638667e-02 1.05571985e+00 5.36405206e-01 7.10189760e-01 -1.55206835e+00 3.68410140e-01 -7.68345416e-01 1.96374089e-01 6.03251100e-01 7.68317342e-01 -8.63368690e-01 -8.02905321e-01 -7.14543462e-01 6.02938473e-01 -9.97753978e-01 5.55056110e-02 -1.33951545e+00 1.05310547e+00 1.04627693e+00 1.01508021e+00 3.93312037e-01 -2.64764488e-01 4.56566364e-03 9.75975841e-02 2.18412414e-01 -2.95922279e-01 -9.90217149e-01 -4.64507341e-01 7.33887672e-01 -1.77346781e-01 -4.80985016e-01 -7.75734007e-01 -1.15862429e+00 -9.40719187e-01 6.19588420e-02 4.00346756e-01 2.50364035e-01 7.11694598e-01 5.99463761e-01 6.08323365e-02 -1.09166287e-01 -6.79228231e-02 -1.94868132e-01 -3.19543302e-01 -6.06110454e-01 5.44374406e-01 1.84242316e-02 -1.86233521e-01 -4.20123249e-01 -3.81134897e-01]
[8.968252182006836, 5.7589569091796875]
ab1e4101-f99a-4993-a465-9791fc3d9fc4
tackling-provably-hard-representative
2205.10403
null
https://arxiv.org/abs/2205.10403v1
https://arxiv.org/pdf/2205.10403v1.pdf
Tackling Provably Hard Representative Selection via Graph Neural Networks
Representative selection (RS) is the problem of finding a small subset of exemplars from an unlabeled dataset, and has numerous applications in summarization, active learning, data compression and many other domains. In this paper, we focus on finding representatives that optimize the accuracy of a model trained on the selected representatives. We study RS for data represented as attributed graphs. We develop RS-GNN, a representation learning-based RS model based on Graph Neural Networks. Empirically, we demonstrate the effectiveness of RS-GNN on problems with predefined graph structures as well as problems with graphs induced from node feature similarities, by showing that RS-GNN achieves significant improvements over established baselines that optimize surrogate functions. Theoretically, we establish a new hardness result for RS by proving that RS is hard to approximate in polynomial time within any reasonable factor, which implies a significant gap between the optimum solution of widely-used surrogate functions and the actual accuracy of the model, and provides justification for the superiority of representation learning-based approaches such as RS-GNN over surrogate functions.
['Vahab Mirrokni', 'Bryan Perozzi', 'Deepak Ramachandran', 'Mohammadhossein Bateni', 'Hossein Esfandiari', 'Anton Tsitsulin', 'Seyed Mehran Kazemi']
2022-05-20
null
null
null
null
['data-compression']
['time-series']
[ 7.50046372e-01 8.09648216e-01 -7.74145782e-01 -2.95739233e-01 -1.10033178e+00 -4.26910698e-01 2.42014125e-01 8.58636260e-01 2.42277049e-02 7.22966909e-01 2.98527092e-01 -1.60403088e-01 -7.56885886e-01 -9.89823520e-01 -9.87970114e-01 -5.37796199e-01 -7.10654736e-01 9.73467708e-01 -1.45678684e-01 -1.65929615e-01 4.07403916e-01 6.11801565e-01 -1.25567889e+00 -6.51220232e-02 9.87683117e-01 7.66470134e-01 -2.08252594e-01 7.04650760e-01 -3.95679846e-02 9.97434676e-01 -6.75735354e-01 -3.16589713e-01 2.72770166e-01 -5.35516262e-01 -1.03125644e+00 1.88794836e-01 5.58821857e-01 2.17270136e-01 -7.61383057e-01 9.82337594e-01 4.10036564e-01 5.74355304e-01 8.27384174e-01 -1.49041724e+00 -7.41624296e-01 1.17122018e+00 -5.80630720e-01 -3.66704762e-02 3.82064611e-01 -3.88887972e-01 1.39465427e+00 -4.75432277e-01 7.29297221e-01 1.16345787e+00 1.00307596e+00 5.77998161e-01 -1.45315707e+00 -1.59871444e-01 1.08197615e-01 2.28916138e-01 -1.49066329e+00 -4.63608503e-01 5.39359570e-01 1.21654868e-01 9.27413344e-01 6.56259656e-01 5.50854385e-01 4.28101063e-01 -4.67348844e-01 9.74564254e-01 4.32417065e-01 -6.99224293e-01 3.09146643e-01 -1.09096110e-01 5.21340609e-01 9.88765061e-01 1.03434956e+00 -3.99548799e-01 -5.08028984e-01 -5.89280963e-01 4.56840545e-01 -1.12195790e-01 -7.40394592e-01 -7.26607800e-01 -7.73909271e-01 1.16276395e+00 6.71633065e-01 4.29370813e-02 -2.22595185e-01 4.88697410e-01 4.43967164e-01 3.26254815e-01 5.01288235e-01 8.57234299e-01 -2.52937406e-01 4.17610466e-01 -8.46540511e-01 -8.12969543e-03 1.18827176e+00 1.22494352e+00 7.84516335e-01 1.28503874e-01 -1.89733565e-01 8.16062629e-01 -5.85730281e-03 1.55289933e-01 3.30151528e-01 -9.00912881e-01 4.87979949e-01 8.86590838e-01 -4.07932580e-01 -1.15105033e+00 -3.64970326e-01 -5.96650302e-01 -8.95313919e-01 -6.38731778e-01 1.03773750e-01 1.68700457e-01 -7.63813496e-01 1.89332724e+00 3.22382450e-01 1.82265118e-01 4.36232006e-03 3.70906442e-01 9.02337134e-01 6.43339932e-01 -2.83520848e-01 -7.68354356e-01 5.19283712e-01 -1.04062438e+00 -4.25915837e-01 -1.54102132e-01 1.15951741e+00 7.67090544e-03 8.14717710e-01 1.95867792e-01 -1.09635031e+00 -7.75774270e-02 -1.18301797e+00 6.23462349e-02 -4.09516133e-02 -2.04048619e-01 8.90463948e-01 5.63543499e-01 -1.37301481e+00 1.13406110e+00 -4.37436312e-01 -5.58033168e-01 6.02346778e-01 6.64475918e-01 -4.69895840e-01 -2.50454485e-01 -8.75846744e-01 8.41296196e-01 6.09824479e-01 -8.93954486e-02 -6.35231614e-01 -5.33081770e-01 -1.10003531e+00 2.26211131e-01 5.98871708e-01 -7.84806013e-01 1.08700168e+00 -9.46098924e-01 -8.50820839e-01 6.24347389e-01 -2.41209537e-01 -9.07671928e-01 1.50287598e-01 1.04126483e-01 -2.48103172e-01 3.23281676e-01 -6.65594488e-02 9.38079972e-03 5.38861752e-01 -1.22696066e+00 -2.18733013e-01 -5.48051596e-01 1.20921642e-01 3.32985997e-01 -5.95726132e-01 -4.16833550e-01 -3.25906783e-01 -2.40505770e-01 1.47659749e-01 -8.30507517e-01 -7.86701977e-01 -2.65388995e-01 -7.01295257e-01 -6.97920620e-01 4.07497108e-01 -2.58675933e-01 1.47119391e+00 -1.70898640e+00 1.38357833e-01 8.75258267e-01 7.38885760e-01 2.96898603e-01 -3.86783242e-01 6.50167525e-01 -1.26146898e-01 4.38794255e-01 -4.65339810e-01 -7.85686150e-02 -1.61545295e-02 3.73063892e-01 -1.67632222e-01 9.07857060e-01 8.36112723e-02 1.00198305e+00 -9.97389138e-01 -5.23627281e-01 -3.01875681e-01 -3.13520022e-02 -1.88175410e-01 1.61542714e-01 -2.40929842e-01 -3.71688485e-01 -3.83717239e-01 4.49509472e-01 4.86755252e-01 -7.29837120e-01 5.87007523e-01 -4.36445400e-02 6.78940833e-01 2.83851743e-01 -1.04576552e+00 1.41246510e+00 -3.04748684e-01 5.29476821e-01 -1.51448071e-01 -1.63910508e+00 1.02092111e+00 -6.20590858e-02 5.35646558e-01 -2.66447097e-01 -1.58006474e-02 3.53849649e-01 -2.22002000e-01 -3.14141214e-01 6.48783684e-01 3.24043036e-01 -6.88969195e-02 7.49478102e-01 6.06115088e-02 -3.29337418e-02 6.72427356e-01 8.77005100e-01 1.62195027e+00 -2.80642241e-01 7.56027818e-01 -8.30687284e-02 3.41542184e-01 2.14840189e-01 3.42740893e-01 1.23753929e+00 1.50269672e-01 6.06119812e-01 5.71736157e-01 -3.31871390e-01 -8.97082806e-01 -7.07372963e-01 2.37585142e-01 9.23968613e-01 1.33564651e-01 -9.44532573e-01 -6.14191473e-01 -9.67232764e-01 1.83446016e-02 7.87694395e-01 -8.16201210e-01 -5.53799629e-01 -6.89993203e-01 -8.23152065e-01 3.62887681e-01 3.68374795e-01 -3.75863016e-02 -7.58710623e-01 -7.75113404e-02 1.72193363e-01 -3.29626761e-02 -7.64753401e-01 -5.19758642e-01 4.03407812e-01 -1.30164254e+00 -1.40612304e+00 -4.44904387e-01 -8.17083776e-01 1.12334156e+00 5.96010327e-01 1.48718059e+00 5.16706765e-01 -4.43612076e-02 4.53034610e-01 -3.99914145e-01 -3.43864381e-01 -6.18548393e-01 5.26533246e-01 -1.21775351e-01 -3.72737408e-01 1.93797365e-01 -6.23014510e-01 -2.35764399e-01 -1.13353640e-01 -9.26869273e-01 -2.10633442e-01 4.62898076e-01 7.51308858e-01 8.76230776e-01 -3.18791829e-02 7.28076994e-01 -1.64317465e+00 8.00717473e-01 -7.11995244e-01 -3.79224151e-01 8.76337290e-01 -9.67718422e-01 4.71325129e-01 7.26753950e-01 -2.89333701e-01 -1.23064555e-01 1.53777406e-01 2.62041092e-01 -2.60253489e-01 5.12393355e-01 9.08618510e-01 -1.50199965e-01 -2.54216164e-01 9.94865119e-01 1.53205708e-01 4.69714217e-02 -7.26131573e-02 6.55440032e-01 4.97853607e-01 5.29848576e-01 -5.26414275e-01 7.18109548e-01 2.62850791e-01 4.04343754e-01 -8.36429596e-01 -1.13596535e+00 -5.93496859e-01 -3.12219054e-01 4.33491953e-02 -2.54632890e-01 -5.22856355e-01 -6.31175637e-01 -4.25424695e-01 -9.76227403e-01 -1.88101143e-01 -7.79813826e-01 1.63432568e-01 -7.97942996e-01 6.46470428e-01 -3.90312910e-01 -9.13320601e-01 -8.96022320e-01 -4.93023485e-01 6.40282452e-01 1.13404870e-01 -3.83576244e-01 -9.94569838e-01 2.08389446e-01 1.06824882e-01 1.80248842e-01 4.79505062e-01 1.23866403e+00 -1.41898477e+00 -4.46466982e-01 -5.79648137e-01 -7.15503693e-02 9.01470557e-02 -1.07268142e-02 1.46319140e-02 -5.62064052e-01 -5.40739000e-01 -2.71238565e-01 -4.10256356e-01 1.15258920e+00 4.71420854e-01 1.46689165e+00 -7.39828169e-01 -5.85235953e-01 7.40526676e-01 1.65867913e+00 -1.57739460e-01 5.01713276e-01 1.70452759e-01 8.71440649e-01 5.36964536e-01 3.68462175e-01 2.82501608e-01 2.08552778e-01 2.64703423e-01 4.45755303e-01 -1.87576320e-02 1.65482927e-02 -4.17663068e-01 1.13650486e-01 1.05109215e+00 -5.45345172e-02 -5.95872700e-01 -5.49859464e-01 6.16237998e-01 -2.13656235e+00 -7.77851760e-01 -1.90460667e-01 2.65243816e+00 6.54524326e-01 -1.32677509e-02 3.09562117e-01 3.54944557e-01 1.21676981e+00 7.13481084e-02 -7.60838985e-01 -6.10843599e-01 -3.40613246e-01 5.39613128e-01 8.36288571e-01 1.68277755e-01 -9.77513611e-01 5.83953857e-01 6.65818834e+00 1.05865026e+00 -5.81341624e-01 -2.51968056e-01 6.53291643e-01 1.02259889e-01 -6.58968508e-01 1.10047236e-01 -4.41535830e-01 -2.20705226e-01 1.35053921e+00 -9.41250801e-01 5.72894454e-01 1.08091116e+00 -1.98671862e-01 3.98086399e-01 -1.44653726e+00 8.00362945e-01 4.76592809e-01 -1.64637876e+00 2.67506123e-01 3.75388302e-02 1.10406566e+00 -2.15437561e-02 -2.09099069e-01 1.66507646e-01 6.15503967e-01 -1.19981551e+00 2.99740434e-01 1.13715053e-01 8.33079219e-01 -1.00270975e+00 4.86062139e-01 2.52586782e-01 -1.19837213e+00 -1.04890503e-01 -7.82018423e-01 3.60944569e-01 -6.84670582e-02 5.40453911e-01 -1.26705837e+00 9.80006516e-01 -3.29545699e-02 8.84529710e-01 -5.38595080e-01 1.35523140e+00 -1.08791359e-01 7.43999183e-01 -4.31456238e-01 -3.66288096e-01 -8.96623358e-02 -6.39479905e-02 5.37491918e-01 1.10650361e+00 2.84116477e-01 -5.07770590e-02 2.80495107e-01 4.47164774e-01 -7.05161333e-01 3.79253536e-01 -1.05782068e+00 -3.34515482e-01 8.04510713e-01 1.23424661e+00 -8.10259938e-01 -4.32266891e-01 1.48814684e-02 5.80729187e-01 9.32115078e-01 2.28967205e-01 -6.28095686e-01 -6.38701081e-01 -8.92194882e-02 1.91370770e-01 2.95827538e-01 3.94399352e-02 -1.50175020e-01 -9.00674880e-01 -9.89900827e-02 -9.07144785e-01 8.63036036e-01 -3.36968482e-01 -1.37647510e+00 7.06431210e-01 1.18585564e-02 -1.17218184e+00 -3.83022875e-01 -3.40386719e-01 -5.42852640e-01 3.92424464e-01 -1.07644534e+00 -7.77141333e-01 -1.74954683e-01 3.90234977e-01 2.26997659e-01 -1.45049077e-02 1.12804163e+00 -1.81523129e-01 -5.23348689e-01 7.82894075e-01 3.43881220e-01 -2.61941366e-02 2.57574052e-01 -1.48065019e+00 5.34433842e-01 9.47586477e-01 5.38240731e-01 7.42399514e-01 6.60382092e-01 -5.10344088e-01 -1.73928690e+00 -1.32613254e+00 8.50002348e-01 -2.96557192e-02 4.68196154e-01 1.57113403e-01 -8.38515222e-01 7.47209609e-01 -1.99061200e-01 3.37799370e-01 7.19950020e-01 1.86382219e-01 -4.29965764e-01 -1.50096819e-01 -1.13336825e+00 4.48150486e-01 1.51470399e+00 -3.39815795e-01 -2.65632749e-01 7.25041509e-01 8.75546336e-01 -2.26639539e-01 -8.45603943e-01 4.68789160e-01 1.76117271e-01 -4.76936698e-01 9.53009129e-01 -1.07911718e+00 3.38108420e-01 -1.89239755e-02 -2.21790776e-01 -1.49429893e+00 -2.94702679e-01 -8.83191824e-01 -9.19926643e-01 9.94229972e-01 6.42781913e-01 -6.23341978e-01 1.09949148e+00 4.44626421e-01 -9.60196257e-02 -1.05327106e+00 -7.41617858e-01 -8.88332844e-01 -2.45256171e-01 -1.87108312e-02 6.96700573e-01 9.28000987e-01 1.47004113e-01 7.01720178e-01 -1.96187422e-01 -1.31742686e-01 8.35864007e-01 3.52483511e-01 8.73296201e-01 -1.46565306e+00 -3.12132001e-01 -3.35940242e-01 -4.81022716e-01 -9.22506452e-01 3.81073028e-01 -1.41728306e+00 -3.66819724e-02 -2.23247695e+00 2.98198521e-01 -7.17934966e-01 -3.30020219e-01 3.51376861e-01 -2.82327443e-01 -7.04707354e-02 -5.57646900e-02 3.00714225e-01 -1.10015774e+00 4.62348878e-01 1.03742731e+00 -3.69409770e-01 -3.52627099e-01 1.89304635e-01 -1.20980465e+00 4.76000488e-01 8.19409132e-01 -6.19879246e-01 -6.79605365e-01 -1.19884543e-01 4.29041117e-01 1.30846277e-01 -2.06728846e-01 -6.54596150e-01 3.68155718e-01 1.60856396e-02 6.70883507e-02 -4.57577258e-01 5.78490570e-02 -5.98228872e-01 1.85405657e-01 5.81234574e-01 -9.58743393e-01 1.44530654e-01 -3.17041755e-01 1.00196266e+00 3.15256640e-02 -6.20965183e-01 4.58273143e-01 -1.29634067e-01 -5.03470540e-01 6.81077421e-01 2.40072012e-01 4.65711981e-01 7.97052681e-01 -4.43400979e-01 -7.40733147e-01 -6.07488215e-01 -3.40860724e-01 2.03562200e-01 3.87117058e-01 -2.59350598e-01 8.26212049e-01 -1.19044292e+00 -8.72967541e-01 -1.84883758e-01 3.62549514e-01 3.33158731e-01 -9.12952498e-02 7.54223883e-01 -6.50744319e-01 8.64554346e-02 4.50301409e-01 -3.10133278e-01 -1.31162524e+00 7.51160622e-01 1.44519165e-01 -5.71500838e-01 -6.33744538e-01 1.01475310e+00 -2.60040402e-01 -3.38235527e-01 3.63311380e-01 1.78363442e-01 -2.25363031e-01 -2.55902827e-01 4.04125899e-01 7.80101418e-01 2.26638094e-01 -4.81772363e-01 -3.34249616e-01 1.47659391e-01 -1.59024760e-01 3.32206666e-01 1.62748086e+00 9.60982889e-02 -4.20031637e-01 2.37586528e-01 1.39514565e+00 -1.30903507e-02 -5.38490415e-01 -5.70207894e-01 2.94697106e-01 -3.20644528e-01 -2.47911096e-01 -1.70620561e-01 -1.06855190e+00 2.90333122e-01 -3.34210806e-02 7.14628518e-01 1.14469516e+00 1.96491212e-01 6.56642199e-01 7.96742976e-01 4.04252023e-01 -8.62117827e-01 -9.24846008e-02 1.16003342e-01 8.07021797e-01 -7.93635190e-01 6.25515640e-01 -6.69323444e-01 -4.61296022e-01 1.25270486e+00 3.13666940e-01 -5.14508605e-01 3.09775144e-01 -1.37907386e-01 -6.14466488e-01 -3.25515330e-01 -1.06682622e+00 -2.68688530e-01 4.28541929e-01 8.09665382e-01 2.66272813e-01 2.63130635e-01 -3.35550934e-01 3.02967876e-01 -2.50882119e-01 -1.89368740e-01 6.97825670e-01 7.70654440e-01 -5.79670012e-01 -9.79055464e-01 1.38010699e-02 1.18855703e+00 -1.96817026e-01 -1.47953823e-01 -6.82327151e-01 7.27911294e-01 -3.97137493e-01 9.57619369e-01 -1.43605784e-01 -4.93672431e-01 2.07040429e-01 -4.03123170e-01 7.86815882e-01 -8.21949661e-01 -4.81807351e-01 -3.90391886e-01 5.60643911e-01 -3.18699211e-01 -5.40367007e-01 -2.95444161e-01 -1.27995384e+00 -6.15403831e-01 -8.51099133e-01 6.48996890e-01 4.14127827e-01 7.43870437e-01 3.64385724e-01 2.18631491e-01 9.52226579e-01 -4.27452415e-01 -1.08927464e+00 -9.20850337e-01 -7.33293474e-01 2.57516831e-01 1.67482123e-01 -1.96941555e-01 -6.23474956e-01 -4.43373829e-01]
[7.117372989654541, 6.101532459259033]
06272b4a-426c-4207-bc8f-03ac65180eb6
towards-addressing-training-data-scarcity
2304.1248
null
https://arxiv.org/abs/2304.12480v1
https://arxiv.org/pdf/2304.12480v1.pdf
Towards Addressing Training Data Scarcity Challenge in Emerging Radio Access Networks: A Survey and Framework
The future of cellular networks is contingent on artificial intelligence (AI) based automation, particularly for radio access network (RAN) operation, optimization, and troubleshooting. To achieve such zero-touch automation, a myriad of AI-based solutions are being proposed in literature for modeling and optimizing network behavior to achieve the zero-touch automation goal. However, to work reliably, AI based automation, requires a deluge of training data. Consequently, the success of AI solutions is limited by a fundamental challenge faced by cellular network research community: scarcity of training data. We present an extensive review of classic and emerging techniques to address this challenge. We first identify the common data types in RAN and their known use-cases. We then present a taxonomized survey of techniques to address training data scarcity for various data types. This is followed by a framework to address the training data scarcity. The framework builds on available information and combination of techniques including interpolation, domain-knowledge based, generative adversarial neural networks, transfer learning, autoencoders, few-shot learning, simulators, and testbeds. Potential new techniques to enrich scarce data in cellular networks are also proposed, such as by matrix completion theory, and domain knowledge-based techniques leveraging different network parameters and geometries. An overview of state-of-the art simulators and testbeds is also presented to make readers aware of current and emerging platforms for real data access. The extensive survey of training data scarcity addressing techniques combined with proposed framework to select a suitable technique for given type of data, can assist researchers and network operators in choosing appropriate methods to overcome the data scarcity challenge in leveraging AI to radio access network automation.
['Ali Imran', 'Ali Rizwan', 'Per Karlsson', 'Shruti Bothe', 'Maxime Bouton', 'Julien Forgeat', 'Hasan Farooq', 'Syed Muhammad Asad Zaidi', 'Marvin Manalastas', 'Usama Masood', 'Haneya Naeem Qureshi']
2023-04-24
null
null
null
null
['matrix-completion']
['methodology']
[ 6.33227453e-02 -8.08718354e-02 -9.70933288e-02 7.60346800e-02 -1.05400562e-01 -4.57012296e-01 1.34488627e-01 -4.23812509e-01 -9.21389181e-03 1.42120719e+00 -3.03145677e-01 -6.62657678e-01 -6.64983213e-01 -1.06854761e+00 -2.70413697e-01 -7.93892860e-01 -7.35275924e-01 6.72527194e-01 -3.73340845e-01 -6.96146011e-01 -1.32705733e-01 9.97211516e-01 -8.53196084e-01 -2.85292149e-01 6.24641001e-01 1.28927064e+00 8.28258321e-02 9.31875288e-01 1.29093200e-01 8.16128671e-01 -9.52041209e-01 -1.64576977e-01 6.02030814e-01 -4.30406660e-01 -5.77827752e-01 7.97596294e-03 -5.38120449e-01 -2.72171110e-01 -9.57791388e-01 4.27774727e-01 1.19744492e+00 3.58081944e-02 6.44883811e-01 -1.58433461e+00 -5.88427007e-01 4.71895307e-01 7.04305843e-02 7.01477528e-01 1.10509664e-01 3.84430975e-01 5.00872195e-01 -5.51340759e-01 4.18616593e-01 6.45448864e-01 9.65135634e-01 4.24286485e-01 -1.13037932e+00 -7.90621340e-01 -2.98749655e-01 9.58471298e-02 -1.63348734e+00 -6.63692117e-01 7.75976479e-01 -1.93473026e-01 7.53210545e-01 1.91925272e-01 8.24803114e-01 9.56623971e-01 -1.27332196e-01 1.39405325e-01 8.30376089e-01 -5.34324765e-01 5.69009602e-01 4.14089829e-01 -3.83803993e-01 7.19426870e-01 6.54508248e-02 4.39453125e-01 -6.03535511e-02 -3.79167378e-01 1.26093125e+00 -1.81130037e-01 -1.92506135e-01 -2.37312362e-01 -1.01935959e+00 4.80662525e-01 2.85298079e-01 3.99425089e-01 -7.77712166e-01 2.41721258e-01 1.58058286e-01 8.37370276e-01 -1.81596324e-01 5.81565678e-01 -6.17534339e-01 -1.54642314e-01 -1.05795336e+00 1.31788656e-01 1.25286043e+00 1.42539048e+00 7.92134225e-01 9.66990292e-01 -2.72672862e-01 5.97079039e-01 2.72047758e-01 6.29228890e-01 7.90246725e-02 -1.18375719e+00 2.43584812e-01 1.46738198e-02 -4.56571989e-02 -8.19186926e-01 -7.12416530e-01 -1.14399219e+00 -1.41038144e+00 6.30239025e-02 2.02047735e-01 -1.10740590e+00 -7.46982157e-01 1.60987818e+00 -1.01796553e-01 8.33191633e-01 2.96880662e-01 5.77324450e-01 6.72295272e-01 6.26428425e-01 -2.36928254e-01 -4.18884546e-01 8.88886034e-01 -8.23769808e-01 -5.94784617e-01 1.92970142e-01 3.10631216e-01 -5.21228671e-01 2.92774945e-01 8.68604630e-02 -1.07999074e+00 -3.51715237e-01 -1.24978483e+00 7.67719150e-01 -5.14492512e-01 -3.16576660e-01 6.53107345e-01 1.55486643e+00 -1.30127859e+00 5.71853042e-01 -3.46129119e-01 -9.37770903e-01 7.93179572e-01 1.00911939e+00 8.55500624e-02 7.94876143e-02 -1.47682869e+00 7.05761254e-01 2.10619748e-01 2.91643739e-02 -1.21143699e+00 -1.13082170e+00 -4.82867926e-01 2.56701082e-01 1.96336985e-01 -1.07063460e+00 8.77033293e-01 -6.52889490e-01 -1.92237639e+00 2.39267081e-01 3.93103212e-01 -1.01862872e+00 8.21676180e-02 4.82259095e-01 -1.16420937e+00 1.63751841e-01 -3.37814987e-01 1.96043834e-01 6.70432270e-01 -1.24166405e+00 -5.15621424e-01 2.30012849e-01 1.30793557e-01 -1.83802620e-01 -2.32377186e-01 -1.02178827e-01 -5.30517064e-02 -7.86272168e-01 -3.09000760e-01 -8.29836011e-01 -7.92498887e-01 -1.32467508e-01 -2.75008619e-01 5.24362862e-01 1.14739382e+00 -2.76401162e-01 1.19952607e+00 -1.65708280e+00 -2.10558534e-01 8.64334702e-01 3.45542282e-01 4.02707726e-01 -8.80089253e-02 7.81204820e-01 9.01153833e-02 3.47684681e-01 1.40709043e-01 1.65357247e-01 -2.18428046e-01 4.04410243e-01 -2.22239837e-01 2.52140433e-01 5.72410077e-02 8.71782541e-01 -7.49835014e-01 -2.22394079e-01 3.83648396e-01 3.41047794e-01 -6.41201675e-01 1.21514469e-01 7.17115998e-02 8.63504410e-01 -6.31310642e-01 1.10013974e+00 5.16635001e-01 -2.46386588e-01 5.57392277e-02 -3.63970667e-01 9.65750366e-02 -4.69653815e-01 -1.15717876e+00 1.25050604e+00 -8.55180860e-01 6.24254465e-01 2.87452579e-01 -1.36505604e+00 8.04421663e-01 9.99491751e-01 1.01629543e+00 -2.48868987e-01 5.05200744e-01 3.30077320e-01 2.12092549e-01 -3.60298008e-01 -2.06949055e-01 -2.71433949e-01 1.37486473e-01 4.00466919e-01 5.16414106e-01 -1.65376998e-02 -4.02637869e-02 7.26514012e-02 1.68864441e+00 -5.30259132e-01 4.20527399e-01 -1.64366856e-01 6.90157413e-01 -2.42463157e-01 6.89038634e-01 1.12403893e+00 -9.03240502e-01 1.62111625e-01 2.65864551e-01 -5.67449033e-01 -1.34554493e+00 -1.11017895e+00 8.35570320e-03 6.73239529e-01 1.14657298e-01 1.64790094e-01 -5.00641942e-01 -1.80683687e-01 -2.69585997e-01 3.52387279e-01 -2.46940017e-01 -5.72512485e-02 -1.22604780e-01 -1.05754638e+00 1.16667521e+00 7.90718421e-02 8.75652730e-01 -7.75660992e-01 1.27661243e-01 5.68673849e-01 7.54237995e-02 -1.52277970e+00 1.65254831e-01 2.56043196e-01 -7.44093597e-01 -4.47837681e-01 -8.22539628e-01 -8.48357975e-01 3.87595475e-01 9.67569351e-02 1.16258454e+00 1.42832026e-01 -2.67375439e-01 6.25877440e-01 -2.44622707e-01 -2.83595085e-01 -4.83611912e-01 1.57232225e-01 5.84839344e-01 -1.48622282e-02 -3.82688679e-02 -1.56231546e+00 -6.26483798e-01 4.89549041e-01 -4.64325130e-01 -5.68083525e-01 9.44891155e-01 9.37407792e-01 2.15714395e-01 5.70386231e-01 1.16504121e+00 -8.32184911e-01 5.66273630e-01 -1.07180917e+00 -3.48688096e-01 1.64213225e-01 -5.65554857e-01 -1.74875408e-01 9.91541982e-01 -1.93688571e-01 -7.88569212e-01 -3.04744422e-01 -2.49980297e-02 -4.00349051e-01 -1.43739551e-01 4.39656913e-01 -1.78908214e-01 -8.03638518e-01 1.06461096e+00 6.55547157e-02 2.74522841e-01 1.45904332e-01 2.83765435e-01 8.49893034e-01 1.91971511e-01 -5.46657383e-01 1.37121761e+00 6.47845209e-01 5.04880726e-01 -1.08345854e+00 -3.86058211e-01 -1.84685498e-01 -2.91851431e-01 -4.67866004e-01 2.78456122e-01 -7.37603962e-01 -5.51884711e-01 1.80959582e-01 -7.03715146e-01 -3.20255190e-01 -5.00313401e-01 4.81239706e-01 -8.99428427e-01 -1.57959275e-02 -5.76813042e-01 -6.57690167e-01 -5.05825102e-01 -9.24108088e-01 6.02473579e-02 4.28650409e-01 1.55254692e-01 -1.10006762e+00 -8.69256184e-02 4.09294605e-01 1.24150336e+00 2.57687122e-01 1.08271694e+00 -6.93716109e-01 -8.72112215e-01 -4.49817657e-01 -1.27075821e-01 7.41829276e-02 -2.99752727e-02 -3.97221968e-02 -8.86616290e-01 -4.02836859e-01 -1.95587710e-01 -1.05359823e-01 8.55372660e-03 7.51783848e-01 1.04133058e+00 -2.95247644e-01 -5.25661111e-01 8.71874928e-01 1.63160825e+00 3.73989284e-01 9.86199021e-01 -3.39419618e-02 1.27196023e-02 1.17519051e-02 1.92116335e-01 1.00978076e+00 6.02536909e-02 4.55621898e-01 5.13226449e-01 -3.40561211e-01 -1.15653954e-01 2.25378171e-01 -1.72036260e-01 6.49333179e-01 -4.62873310e-01 -6.58147454e-01 -6.34633183e-01 1.07589200e-01 -1.53438890e+00 -1.39663410e+00 3.56527299e-01 1.94421840e+00 1.96078315e-01 2.21517473e-01 2.73759723e-01 4.68563259e-01 7.89609194e-01 -3.09681594e-01 -4.11116064e-01 -1.73650548e-01 -1.05581917e-01 4.52458292e-01 9.22217548e-01 2.20430419e-01 -9.42462623e-01 7.47531235e-01 6.78980875e+00 9.92071152e-01 -9.87529218e-01 1.36828065e-01 5.74249387e-01 3.17709029e-01 1.57789275e-01 1.49291679e-01 -3.13059568e-01 3.05197150e-01 1.33528912e+00 -2.67446727e-01 1.02428699e+00 6.14068687e-01 6.13746881e-01 2.46757925e-01 -7.76069641e-01 1.17937541e+00 -3.24818879e-01 -1.76453626e+00 -1.26271039e-01 2.57315248e-01 9.09267247e-01 1.10761248e-01 1.29893363e-01 8.50197196e-01 2.38134772e-01 -1.03523982e+00 -4.31661904e-02 6.06972635e-01 9.27374542e-01 -8.37568343e-01 9.15717185e-01 5.47875874e-02 -1.16833425e+00 -5.83354056e-01 -2.22201973e-01 -2.20965713e-01 3.59811217e-01 6.76949322e-01 -9.79499280e-01 6.77981555e-01 3.32820594e-01 4.02961284e-01 -8.82752240e-02 1.55903125e+00 5.46879172e-01 7.38523901e-01 -4.98433799e-01 -2.80858800e-02 5.15827574e-02 -1.09472990e-01 8.67077470e-01 8.03986073e-01 5.07534683e-01 2.88852394e-01 1.90663189e-01 7.68068790e-01 -1.88428327e-01 -9.10004377e-02 -7.33922780e-01 -3.93659584e-02 1.26638114e+00 1.36986160e+00 -4.23832864e-01 -1.49792805e-01 -5.32354891e-01 6.75814092e-01 -2.41215631e-01 8.33056748e-01 -9.16026831e-01 -6.90014064e-01 7.16563463e-01 1.80823028e-01 -1.95380636e-02 -5.03387928e-01 -4.43645060e-01 -5.36530137e-01 -6.31656468e-01 -8.61209631e-01 2.30668366e-01 -5.78640699e-01 -1.17946780e+00 6.83459222e-01 -2.57519692e-01 -1.60414624e+00 -2.44589552e-01 -4.25572455e-01 -6.54034078e-01 8.22436333e-01 -1.40728819e+00 -1.08345485e+00 -6.03686094e-01 6.44095063e-01 3.04807842e-01 -1.28698552e+00 1.26517737e+00 9.42492604e-01 -7.44156539e-01 6.94386661e-01 2.78399557e-01 1.26863763e-01 2.56983161e-01 -8.50607812e-01 -5.99071458e-02 6.12621248e-01 -5.21949045e-02 2.78992027e-01 7.70429730e-01 -1.77132741e-01 -1.44180071e+00 -1.01315165e+00 -1.17641725e-02 1.27622887e-01 6.32957876e-01 4.52810377e-02 -1.54472351e-01 4.34218764e-01 8.79850388e-02 4.26446795e-01 9.80809450e-01 -1.76083576e-02 4.42879289e-01 -5.98154068e-01 -1.62980449e+00 9.79693949e-01 9.20355320e-01 -3.37170064e-01 3.82886738e-01 4.46159601e-01 4.04594451e-01 -1.12920187e-01 -1.07816255e+00 3.61555099e-01 3.23359191e-01 -7.17142522e-01 1.15168357e+00 -6.71862602e-01 -3.99756163e-01 -3.63595158e-01 -3.09699208e-01 -1.32551670e+00 -4.18553948e-01 -1.47281218e+00 -5.53874731e-01 1.00513208e+00 5.64498007e-01 -5.84198415e-01 1.23011076e+00 3.29428613e-01 -5.11470921e-02 -7.86644936e-01 -8.98953915e-01 -9.24524724e-01 -1.22174621e-01 -3.19339961e-01 7.85955906e-01 8.97332191e-01 -2.43011359e-02 3.92458141e-01 -4.75267678e-01 5.50909758e-01 7.45434403e-01 -6.19198561e-01 1.01415384e+00 -1.17092407e+00 -3.83600205e-01 -3.05976808e-01 -1.06760442e+00 -8.90592217e-01 -2.12826580e-01 -6.97370648e-01 -5.05258977e-01 -1.26291990e+00 -4.30670679e-01 -1.19774890e+00 -3.76069933e-01 -9.67368707e-02 6.65041029e-01 2.03162894e-01 -9.56619829e-02 1.75526410e-01 -2.99010217e-01 3.27512383e-01 1.05621552e+00 -1.96824297e-02 -3.90561789e-01 6.46624267e-01 -6.82753921e-01 3.48374903e-01 1.41896152e+00 -1.07641377e-01 -6.74504101e-01 5.72339855e-02 7.39766210e-02 5.88746190e-01 3.81601572e-01 -1.91227210e+00 5.13170779e-01 -1.52444020e-01 3.55731994e-01 -2.53376573e-01 5.14609694e-01 -1.35889041e+00 3.15283298e-01 4.92584139e-01 2.02981085e-01 -2.05751196e-01 -7.64695415e-03 9.07288134e-01 1.53021172e-01 -1.02041833e-01 9.52386379e-01 7.35447705e-02 -7.47869253e-01 8.27637196e-01 -8.77507925e-01 2.16119751e-01 1.28789759e+00 -5.36098659e-01 4.08443399e-02 -1.11747527e+00 -1.18947864e+00 3.47820312e-01 -1.52004942e-01 -2.40355074e-01 4.59373802e-01 -1.31626546e+00 -6.62566423e-01 1.40571862e-01 -3.73567820e-01 -5.47728658e-01 3.52573872e-01 9.32759941e-01 -8.31632078e-01 5.35401404e-01 -5.24221599e-01 -2.50926286e-01 -5.98960757e-01 6.29805744e-01 9.57363784e-01 -2.98204005e-01 -1.14451628e-02 5.75415790e-01 -7.57386804e-01 -2.74187446e-01 2.87900925e-01 7.38761723e-02 -3.13198045e-02 -6.41145051e-01 4.16392982e-02 6.65857434e-01 1.49015784e-01 -1.40498519e-01 -1.23946965e-01 1.45014748e-01 2.28844598e-01 -7.14466944e-02 1.06944215e+00 -4.92737085e-01 2.01439425e-01 -6.17498159e-02 7.89570808e-01 -1.92627415e-01 -7.26296604e-01 -5.29966116e-01 -2.42035463e-01 -1.17149122e-01 2.81267941e-01 -8.48868310e-01 -1.34049559e+00 5.28360009e-01 8.02824855e-01 4.50852275e-01 1.37495768e+00 -4.10691887e-01 7.96065688e-01 7.22868323e-01 8.60958457e-01 -9.86349285e-01 3.78263816e-02 5.65382063e-01 3.72142494e-01 -1.09614384e+00 -1.25190333e-01 -5.25763392e-01 -2.10202590e-01 1.10496807e+00 4.41675633e-01 -5.02105094e-02 1.45213675e+00 5.42455494e-01 2.09945608e-02 -5.95345683e-02 -6.26794338e-01 -4.42359209e-01 -3.62597674e-01 1.42985129e+00 1.75041169e-01 -7.71032423e-02 2.59439915e-01 6.77291214e-01 -6.05030917e-03 8.02118629e-02 6.74722970e-01 9.02751803e-01 -4.97903019e-01 -1.20324528e+00 -3.50661516e-01 9.36023831e-01 -3.84439051e-01 -1.82359070e-01 1.15708277e-01 6.65815592e-01 1.23192497e-01 1.19587243e+00 -2.52416253e-01 -5.55846274e-01 1.38876244e-01 -4.07093018e-01 2.38319561e-01 -3.61770719e-01 -3.43266308e-01 -5.94931364e-01 1.57429650e-01 -1.37967123e-02 -2.99249142e-01 -6.12769835e-02 -1.04103732e+00 -9.86426890e-01 -4.16232526e-01 7.15080202e-02 4.18443829e-01 9.85457242e-01 5.83372355e-01 7.80530512e-01 1.08613753e+00 -8.23616624e-01 -3.62599075e-01 -7.12398350e-01 -7.83505917e-01 -3.10553104e-01 2.61481136e-01 -6.90002322e-01 -1.25808492e-01 -1.17005423e-01]
[6.031755447387695, 1.6166057586669922]
3e48ad7e-5e02-4ff2-88b3-914e6abd128f
graphing-the-future-activity-and-next-active
2209.05194
null
https://arxiv.org/abs/2209.05194v1
https://arxiv.org/pdf/2209.05194v1.pdf
Graphing the Future: Activity and Next Active Object Prediction using Graph-based Activity Representations
We present a novel approach for the visual prediction of human-object interactions in videos. Rather than forecasting the human and object motion or the future hand-object contact points, we aim at predicting (a)the class of the on-going human-object interaction and (b) the class(es) of the next active object(s) (NAOs), i.e., the object(s) that will be involved in the interaction in the near future as well as the time the interaction will occur. Graph matching relies on the efficient Graph Edit distance (GED) method. The experimental evaluation of the proposed approach was conducted using two well-established video datasets that contain human-object interactions, namely the MSR Daily Activities and the CAD120. High prediction accuracy was obtained for both action prediction and NAO forecasting.
['Antonis Argyros', 'Konstantinos Papoutsakis', 'Victoria Manousaki']
2022-09-12
null
null
null
null
['human-object-interaction-detection', 'graph-matching']
['computer-vision', 'graphs']
[ 3.98141116e-01 -1.72236003e-02 -1.08968116e-01 -1.87165588e-01 2.20512435e-01 -1.92504182e-01 7.19184101e-01 2.83078700e-01 -1.08766973e-01 4.18996453e-01 -4.40872572e-02 1.15375882e-02 -2.89367169e-01 -5.45103192e-01 -4.71377581e-01 -4.67265666e-01 -4.89165366e-01 7.14018404e-01 7.38698006e-01 -3.02536525e-02 3.79322737e-01 1.08610249e+00 -1.79703116e+00 2.84134716e-01 3.83517385e-01 1.25583017e+00 3.88292581e-01 8.56892228e-01 2.41585404e-01 1.14719164e+00 -4.12847042e-01 -2.29119807e-01 2.69396722e-01 -4.73273367e-01 -7.42793858e-01 4.30197179e-01 4.71113384e-01 -2.09138244e-01 -5.65397799e-01 6.99816287e-01 7.62657896e-02 5.80848634e-01 6.46899223e-01 -1.69873142e+00 1.40697092e-01 1.17433108e-01 -5.10305643e-01 4.39893544e-01 9.26228881e-01 1.14858642e-01 5.84700406e-01 -9.13969517e-01 1.42486811e+00 1.10163760e+00 3.02339077e-01 3.21327835e-01 -6.89058006e-01 -3.81978422e-01 -4.43286337e-02 9.29761529e-01 -1.47847724e+00 -3.94908071e-01 9.98600543e-01 -9.02459800e-01 8.14578354e-01 1.88290045e-01 1.00621986e+00 7.73746669e-01 3.65738004e-01 7.28008151e-01 4.09340620e-01 -6.27194583e-01 2.69479811e-01 -4.94973175e-02 4.97375540e-02 6.29405797e-01 -1.76344767e-01 2.01656036e-02 -6.64467990e-01 -1.20943896e-01 5.28100193e-01 1.48583323e-01 -1.07385859e-01 -7.13359833e-01 -1.64382339e+00 3.24468791e-01 1.84510037e-01 3.82653147e-01 -7.40107238e-01 -1.75582282e-02 3.08982342e-01 -3.12695727e-02 1.71141565e-01 -2.71760851e-01 -2.16379464e-01 -2.05109358e-01 -7.41991997e-01 4.12537158e-01 7.86647081e-01 1.03176355e+00 5.57911158e-01 -2.77506143e-01 -2.53948390e-01 1.27295673e-01 1.04375668e-01 5.44543304e-02 5.53901047e-02 -8.34084749e-01 7.28477478e-01 9.49278414e-01 5.53334892e-01 -1.44787955e+00 -7.12532774e-02 4.24557239e-01 -5.06635845e-01 4.41306293e-01 5.43933511e-01 9.95261744e-02 -7.45591104e-01 1.03801346e+00 6.66264296e-01 4.05945361e-01 -2.50101805e-01 6.22012973e-01 6.69487476e-01 9.13868725e-01 2.29714707e-01 -4.64935660e-01 1.14666724e+00 -9.74944234e-01 -8.76379609e-01 -4.04902510e-02 6.80938184e-01 -6.68195605e-01 2.47986376e-01 1.60180598e-01 -1.08310664e+00 -1.02350092e+00 -6.77212715e-01 1.82692736e-01 -2.92698801e-01 2.30552703e-01 3.94121796e-01 -3.02329715e-02 -5.76879859e-01 6.94326162e-01 -9.46755648e-01 -6.33973360e-01 9.89796370e-02 3.53105813e-01 -5.75991631e-01 8.84518847e-02 -7.56938159e-01 7.66758025e-01 6.75455034e-01 3.44640911e-01 -9.17509615e-01 -3.73878986e-01 -5.96451998e-01 4.77165990e-02 6.70336545e-01 -2.70668596e-01 6.71497405e-01 -1.35321999e+00 -8.70315075e-01 8.41342628e-01 -3.21469009e-01 -4.91020709e-01 6.88013494e-01 -1.42182395e-01 -5.59583426e-01 2.72768885e-01 -2.33619064e-01 7.59649873e-01 7.35656202e-01 -9.14557278e-01 -1.14798880e+00 -3.68068397e-01 -1.87883265e-02 5.03959775e-01 -9.18865949e-02 1.55004740e-01 -5.75696886e-01 -6.62599206e-01 1.30020395e-01 -1.12113154e+00 1.07818849e-01 6.35784149e-01 -1.68290645e-01 -7.43537068e-01 1.33173525e+00 -9.24944520e-01 1.27484167e+00 -2.25868678e+00 4.40510303e-01 2.50824153e-01 6.77409619e-02 4.45266992e-01 1.97964817e-01 5.79806447e-01 -2.81002432e-01 -4.99071211e-01 3.09863478e-01 -4.46705110e-02 -4.07631814e-01 9.33866426e-02 -9.67567042e-02 4.19559717e-01 -3.48185420e-01 7.14384794e-01 -9.10506904e-01 -7.79724598e-01 5.64574182e-01 2.94510901e-01 -4.54985648e-02 5.62745869e-01 -1.86236545e-01 6.70523942e-01 -3.63702506e-01 4.77270901e-01 2.53863603e-01 -7.67935216e-02 1.41730905e-01 -5.14056206e-01 -1.52614459e-01 -2.07817987e-01 -1.48172832e+00 1.21749389e+00 1.19837023e-01 7.42582619e-01 -2.35712603e-01 -7.09753156e-01 9.03755844e-01 4.35841709e-01 9.67446029e-01 -4.13443089e-01 -1.48389071e-01 -8.57034922e-02 -8.38546380e-02 -6.50815964e-01 4.01266426e-01 5.44429421e-01 4.76392567e-01 1.55553579e-01 -2.64258385e-01 4.77563888e-01 6.34432137e-01 2.29645863e-01 9.80867445e-01 4.63906467e-01 6.38780296e-01 -4.38267551e-02 7.83964276e-01 1.03961267e-01 5.21843910e-01 3.55129302e-01 -3.16511631e-01 6.32552207e-02 4.22035158e-01 -7.38872468e-01 -8.70359957e-01 -8.05913448e-01 2.72262722e-01 8.19442034e-01 4.35045600e-01 -2.09610477e-01 -4.57936704e-01 -7.62253702e-01 -1.89190388e-01 8.92030597e-01 -6.85844660e-01 1.10199034e-01 -7.67771065e-01 2.63008893e-01 -3.76134485e-01 6.04064822e-01 2.77440310e-01 -1.60390043e+00 -8.42077792e-01 2.12381527e-01 -2.83749044e-01 -1.06484115e+00 -8.47007155e-01 -2.95338243e-01 -9.30518627e-01 -1.28503954e+00 -5.89218020e-01 -8.34220767e-01 7.41407692e-01 6.24415139e-03 8.00592005e-01 1.70227304e-01 -5.58543921e-01 7.13940322e-01 -3.37116003e-01 -1.60745308e-01 -5.60343623e-01 -6.02367043e-01 2.03415770e-02 6.20495617e-01 2.19546929e-01 -1.58919781e-01 -7.59201586e-01 7.30638027e-01 -4.54341829e-01 2.96469301e-01 -3.48718534e-03 2.87674993e-01 6.98463321e-01 2.87842780e-01 -9.94212776e-02 -4.04233575e-01 -3.63845192e-02 -3.44358802e-01 -6.84048653e-01 7.59751737e-01 -4.79842603e-01 -4.47228163e-01 9.93704721e-02 -6.83365762e-01 -1.18899560e+00 6.96525514e-01 3.88708591e-01 -5.24256825e-01 -4.25832629e-01 2.15239689e-01 -8.87766331e-02 9.75182429e-02 2.66673476e-01 1.59377992e-01 -4.19423133e-01 -2.81905472e-01 9.16913711e-03 5.30766726e-01 5.39316058e-01 3.25280130e-02 4.71586138e-01 5.24065554e-01 2.64043510e-01 -8.70598435e-01 -2.92030096e-01 -7.32546091e-01 -1.20747828e+00 -9.74360883e-01 1.29466307e+00 -6.54083252e-01 -8.74089181e-01 6.14040017e-01 -1.39270008e+00 -8.80107880e-02 -5.06957471e-02 6.98668778e-01 -6.52336657e-01 4.39889669e-01 -1.57767862e-01 -9.68581319e-01 -1.50152788e-01 -8.90386283e-01 8.93633306e-01 2.48200428e-02 -5.41549683e-01 -8.88760269e-01 -1.15029700e-01 5.41506588e-01 -3.52189183e-01 5.32643080e-01 8.45475137e-01 -5.78172147e-01 -8.88212502e-01 -5.36260188e-01 1.96116008e-02 -1.01263128e-01 9.07182172e-02 3.46023738e-01 -1.41780376e-01 -2.40661219e-01 -3.83355856e-01 3.30223143e-01 6.04257919e-03 3.18627298e-01 9.48903978e-01 -1.04411311e-01 -8.77943218e-01 2.33212970e-02 1.01800776e+00 9.32208538e-01 1.07685578e+00 -2.34937333e-02 9.09067810e-01 6.99415326e-01 1.42117643e+00 6.18091464e-01 1.87589481e-01 1.48863709e+00 4.26185459e-01 2.78355688e-01 -2.35816792e-01 -2.86726475e-01 4.36196417e-01 4.45977688e-01 -5.10863543e-01 -5.32595515e-01 -1.05691516e+00 4.89021391e-01 -2.08771467e+00 -1.22304308e+00 -5.33988237e-01 2.31058002e+00 1.24308832e-01 9.79723483e-02 1.95791125e-01 1.76286548e-01 9.64791298e-01 -4.11553755e-02 -4.89578158e-01 -8.05461928e-02 4.40895587e-01 -3.38178277e-01 7.30812475e-02 3.35673124e-01 -1.00351763e+00 6.25510514e-01 5.61014652e+00 8.03195298e-01 -6.71335697e-01 -1.15851432e-01 4.30992723e-01 1.13358407e-03 6.28880560e-01 -5.88325188e-02 -8.76006007e-01 4.67967242e-01 5.43029428e-01 -1.81239873e-01 3.85833234e-01 8.71382475e-01 2.46411458e-01 -6.36786222e-01 -1.42845011e+00 1.13512421e+00 2.37437233e-01 -1.19227099e+00 4.87623215e-02 -1.27612904e-01 4.39837754e-01 -6.06778204e-01 -5.57347000e-01 -2.76529253e-01 -3.68690968e-01 -4.87836361e-01 9.37791288e-01 1.11619616e+00 5.38214862e-01 -5.58196187e-01 5.89101851e-01 5.17758191e-01 -1.72409689e+00 -2.79610246e-01 2.49140680e-01 -1.61215663e-02 5.02976894e-01 4.02325392e-02 -8.60591888e-01 3.38198543e-01 8.52443695e-01 9.78720963e-01 -6.16501153e-01 9.84652340e-01 -9.67121311e-03 2.41020128e-01 -7.75034949e-02 -8.44016150e-02 -2.27357849e-01 -3.75048518e-01 9.53845143e-01 6.92551434e-01 5.15693009e-01 3.65503132e-01 1.92944020e-01 4.09663677e-01 4.08670276e-01 -5.41940145e-02 -6.95782959e-01 -2.29329079e-01 2.15277091e-01 9.99468386e-01 -9.08546209e-01 -5.57420909e-01 -2.94735610e-01 1.17632389e+00 1.04645211e-02 2.76488632e-01 -9.19456005e-01 -3.19280744e-01 3.29753071e-01 6.83399975e-01 6.84765160e-01 -4.94264036e-01 3.54885548e-01 -6.09851480e-01 3.93542916e-01 -3.75680506e-01 4.92888331e-01 -1.41931152e+00 -7.82951415e-01 5.14005244e-01 2.69924611e-01 -1.63521349e+00 -5.41673124e-01 -4.10118341e-01 -5.77741921e-01 4.57663268e-01 -4.70939279e-01 -1.07825553e+00 -4.68532860e-01 7.45602012e-01 5.46978533e-01 -1.66225418e-01 4.81821984e-01 3.66924196e-01 -1.55665159e-01 2.07911506e-02 -2.44884193e-01 8.16153884e-02 6.69413358e-02 -6.65122151e-01 1.61274254e-01 5.90036869e-01 3.59950125e-01 -1.76938269e-02 7.81007111e-01 -1.16254747e+00 -1.26335585e+00 -9.35195208e-01 1.37204742e+00 -5.86106837e-01 6.40692055e-01 -2.76231796e-01 -6.49969220e-01 8.31381619e-01 -2.28175864e-01 1.08182319e-01 -1.41307972e-02 -6.98467255e-01 5.00503242e-01 -1.77430227e-01 -1.02085602e+00 6.99512541e-01 1.42988861e+00 -3.79463047e-01 -5.23927927e-01 5.57250082e-01 1.21441387e-01 -5.72368741e-01 -7.11068809e-01 4.83683646e-01 6.22008324e-01 -1.05328965e+00 1.04111123e+00 -6.74782932e-01 1.26091123e-01 -3.66302341e-01 3.99828292e-02 -6.28021479e-01 -2.85608917e-01 -4.81096447e-01 -6.92000866e-01 1.02284980e+00 -1.81288093e-01 1.60708860e-01 9.00157750e-01 7.81097412e-01 1.56066537e-01 -7.64676094e-01 -1.17629421e+00 -7.45156050e-01 -1.19270754e+00 -1.52188480e-01 2.49024317e-01 6.16897583e-01 -2.47390628e-01 -1.24240488e-01 -6.66213691e-01 1.73137784e-01 4.61787075e-01 1.30414858e-01 9.35704887e-01 -1.37818646e+00 3.32336165e-02 8.71732682e-02 -1.16687286e+00 -9.39325988e-01 9.59911197e-02 -5.40299773e-01 2.21057751e-04 -1.40901494e+00 1.66717052e-01 -2.53610834e-02 4.65864278e-02 2.48968109e-01 2.36519575e-01 -8.39541852e-02 2.88804322e-01 2.88879126e-01 -7.40432560e-01 2.28518024e-01 1.14078701e+00 -9.32465401e-03 -3.53635609e-01 3.58469278e-01 7.87748277e-01 8.10138226e-01 2.79317379e-01 -5.52104712e-01 -3.17013383e-01 1.93747640e-01 -3.81605164e-03 8.25555921e-01 6.67637467e-01 -1.26799476e+00 3.25375855e-01 -4.32209074e-01 3.72107059e-01 -9.18764174e-01 4.87738252e-01 -1.46575356e+00 9.78907287e-01 7.48249054e-01 -3.82209331e-01 1.42483532e-01 -4.28421833e-02 1.00335634e+00 -6.11405522e-02 -3.76704663e-01 3.44158411e-01 1.95117697e-01 -1.17493939e+00 5.91517687e-01 -5.09878993e-01 -5.03837824e-01 1.74386096e+00 -7.07964003e-01 2.15251148e-01 -4.84829962e-01 -1.35229766e+00 2.86212638e-02 1.98919311e-01 7.87676215e-01 7.29471147e-01 -1.38627672e+00 -3.27375889e-01 7.22656026e-02 2.55189538e-01 -6.19011581e-01 2.16443613e-01 8.10153961e-01 -3.85924846e-01 1.56947374e-01 -4.24403459e-01 -4.39688742e-01 -2.11496162e+00 6.66740358e-01 1.77918896e-01 -1.28640786e-01 -7.28860676e-01 2.65967518e-01 7.89441913e-02 3.71817440e-01 5.75581491e-01 1.16899908e-01 -5.75133801e-01 9.48226526e-02 4.95744288e-01 1.15616810e+00 -1.45392969e-01 -1.22031629e+00 -4.22509313e-01 5.27464032e-01 1.86151311e-01 2.59308666e-01 1.22844720e+00 -2.19273761e-01 -2.40808740e-01 6.64836645e-01 8.47093940e-01 -4.09214705e-01 -1.43377960e+00 -2.79063918e-02 1.81348890e-01 -6.82544231e-01 -1.42706633e-01 -6.22378111e-01 -1.06630111e+00 7.17742264e-01 1.23819208e+00 7.44940862e-02 8.34230304e-01 1.28190875e-01 7.45379746e-01 5.51596105e-01 7.08174050e-01 -1.11215866e+00 2.18563512e-01 -1.08169578e-02 1.17224145e+00 -8.54148746e-01 1.82356596e-01 -7.11679637e-01 -8.10228705e-01 1.30156779e+00 5.91255784e-01 3.04056287e-01 9.60364103e-01 -3.05899978e-01 -2.69868106e-01 -5.05649745e-01 -4.00053769e-01 1.11459628e-01 9.01206195e-01 4.75816071e-01 1.75851002e-01 -2.01836620e-02 -2.29504943e-01 -1.73287153e-01 4.84616876e-01 4.91950482e-01 7.76771456e-02 1.04754555e+00 -2.42842540e-01 -5.89048743e-01 -2.36043304e-01 5.22808552e-01 1.12976685e-01 4.53877747e-01 -3.25076371e-01 6.57376468e-01 4.06020254e-01 7.80068099e-01 4.13303435e-01 -2.26557314e-01 4.79754746e-01 9.52612702e-03 5.52946746e-01 -4.42780256e-01 -2.42738619e-01 -3.26289803e-01 4.60385568e-02 -8.64240289e-01 -6.76771641e-01 -9.20291185e-01 -1.36119235e+00 -1.87704831e-01 -2.70358086e-01 -1.57836825e-01 3.72944564e-01 9.13746834e-01 3.96637499e-01 1.75943658e-01 4.31413293e-01 -1.27016544e+00 7.42516518e-02 -6.37326479e-01 -7.90787995e-01 9.79023218e-01 -1.45014316e-01 -9.18077290e-01 -2.88234860e-01 6.70260906e-01]
[8.428596496582031, 0.40439894795417786]
be5f6cc8-d7f2-47ba-b1f8-0424f3de3499
generalization-bounds-with-data-dependent
2302.02766
null
https://arxiv.org/abs/2302.02766v2
https://arxiv.org/pdf/2302.02766v2.pdf
Generalization Bounds with Data-dependent Fractal Dimensions
Providing generalization guarantees for modern neural networks has been a crucial task in statistical learning. Recently, several studies have attempted to analyze the generalization error in such settings by using tools from fractal geometry. While these works have successfully introduced new mathematical tools to apprehend generalization, they heavily rely on a Lipschitz continuity assumption, which in general does not hold for neural networks and might make the bounds vacuous. In this work, we address this issue and prove fractal geometry-based generalization bounds without requiring any Lipschitz assumption. To achieve this goal, we build up on a classical covering argument in learning theory and introduce a data-dependent fractal dimension. Despite introducing a significant amount of technical complications, this new notion lets us control the generalization error (over either fixed or random hypothesis spaces) along with certain mutual information (MI) terms. To provide a clearer interpretation to the newly introduced MI terms, as a next step, we introduce a notion of "geometric stability" and link our bounds to the prior art. Finally, we make a rigorous connection between the proposed data-dependent dimension and topological data analysis tools, which then enables us to compute the dimension in a numerically efficient way. We support our theory with experiments conducted on various settings.
['Umut Şimşekli', 'George Deligiannidis', 'Benjamin Dupuis']
2023-02-06
null
null
null
null
['topological-data-analysis']
['graphs']
[ 4.49976772e-02 1.00413516e-01 9.98220295e-02 -2.51884729e-01 -9.40774381e-02 -5.40633321e-01 3.85289222e-01 5.20587921e-01 -5.21700144e-01 7.88843751e-01 -3.03377777e-01 -3.85841250e-01 -6.93025649e-01 -9.38443482e-01 -7.06578612e-01 -1.05190170e+00 -3.17117542e-01 -8.84297416e-02 2.87188292e-01 -3.81754339e-01 3.90523076e-01 6.92483008e-01 -1.49150467e+00 -4.27320212e-01 9.47896242e-01 1.04387975e+00 -6.46604225e-03 4.29167509e-01 -1.15000978e-02 2.35383794e-01 -3.06719810e-01 -5.47623992e-01 4.04851228e-01 -4.15916741e-01 -9.29628193e-01 -4.26414311e-02 3.26946765e-01 2.39504099e-01 -3.03722303e-02 1.44713795e+00 3.54027808e-01 2.55777746e-01 8.36443901e-01 -1.01788425e+00 -6.05469942e-01 6.44091427e-01 -3.59641284e-01 3.82453710e-01 -1.29998714e-01 -3.81319076e-01 9.47542071e-01 -6.37522817e-01 3.60969007e-01 9.08725917e-01 1.00220656e+00 5.04574955e-01 -1.21093762e+00 -3.07554394e-01 -2.39452552e-02 1.09584652e-01 -1.35878229e+00 -9.07026976e-02 9.54782009e-01 -5.49249351e-01 6.38755038e-02 2.55992830e-01 5.41882575e-01 7.89259076e-01 2.84816802e-01 4.37082201e-01 1.08912766e+00 -6.62084639e-01 4.05541033e-01 2.72871286e-01 4.05876189e-01 8.19681942e-01 6.10565126e-01 -8.42128918e-02 1.64663061e-01 2.91294754e-01 9.76320326e-01 -1.04665220e-01 -7.37220466e-01 -7.20598042e-01 -1.04988992e+00 1.09917378e+00 6.23856902e-01 8.04791570e-01 -5.61256986e-03 -1.32123232e-01 3.56114537e-01 4.60583180e-01 5.75869679e-01 5.13783634e-01 -2.87968844e-01 1.04567625e-01 -4.97812539e-01 6.03187305e-04 7.93952763e-01 7.65861750e-01 6.11008763e-01 -2.31300741e-01 3.68447632e-01 7.31080353e-01 2.46016290e-02 6.43951073e-02 3.88323098e-01 -5.65206349e-01 2.71737307e-01 2.62676865e-01 -9.80689526e-02 -1.42283058e+00 -5.95388472e-01 -8.61313283e-01 -1.21556294e+00 1.86942771e-01 8.75344396e-01 1.30196452e-01 -1.73051730e-01 2.04710650e+00 2.19976932e-01 8.06602836e-02 1.20615922e-01 8.20485771e-01 -1.18217044e-01 4.89404127e-02 -2.88433582e-01 -3.54585975e-01 9.61329222e-01 -5.51718473e-01 -4.75250989e-01 6.34577811e-01 9.70576584e-01 -2.79148400e-01 1.39388061e+00 5.16186893e-01 -9.61009800e-01 -1.97084114e-01 -1.34776962e+00 1.88765228e-01 -5.50229549e-01 -1.20237194e-01 6.73229039e-01 8.83765042e-01 -1.01842248e+00 1.08614922e+00 -8.18358481e-01 -6.60644889e-01 3.73059392e-01 1.06408201e-01 -3.22321862e-01 3.33334416e-01 -1.09285092e+00 8.45430374e-01 4.42378074e-01 3.07448376e-02 2.70771515e-02 -5.04811704e-01 -6.39579654e-01 2.52997410e-02 2.83235669e-01 -7.08805501e-01 8.85041475e-01 -7.11551905e-01 -1.33248734e+00 5.71593344e-01 2.72119582e-01 -6.90077245e-01 8.40577722e-01 4.05284809e-03 -9.95131508e-02 2.31707141e-01 -2.66514122e-01 -3.27351093e-02 4.66439039e-01 -9.91985142e-01 -2.28660345e-01 -6.53709948e-01 4.60009366e-01 -2.38549471e-01 -6.66059077e-01 -5.69332182e-01 2.75696307e-01 -7.88926125e-01 3.18526596e-01 -6.89669967e-01 -1.19590610e-01 5.04864492e-02 -1.83002546e-01 -2.84789264e-01 3.40843529e-01 7.71630332e-02 9.51172113e-01 -2.07966089e+00 2.81991184e-01 2.41774485e-01 4.16459382e-01 1.41135335e-01 2.15337977e-01 3.41689080e-01 -1.77111417e-01 3.40220183e-01 -5.63143253e-01 -4.09357548e-01 6.48110136e-02 8.16624090e-02 -4.28835750e-01 9.37790334e-01 2.06680015e-01 5.10774672e-01 -8.66969407e-01 -4.00229692e-01 2.03457117e-01 6.47289157e-01 -8.08745205e-01 -5.61267018e-01 9.02165100e-03 5.36511719e-01 -4.33959842e-01 1.08356543e-01 8.34150553e-01 -1.70535907e-01 -1.38958097e-01 -1.09300673e-01 -1.84245691e-01 9.11472142e-02 -1.28858399e+00 1.54253817e+00 -6.02504194e-01 4.79498923e-01 5.13393022e-02 -1.84638488e+00 7.99756348e-01 -8.30309466e-02 3.94292891e-01 -2.82076925e-01 4.25807506e-01 3.12598646e-01 -1.44668028e-01 -4.18658257e-01 2.58213818e-01 -6.86259985e-01 6.48161322e-02 2.70047843e-01 -9.71859694e-02 2.80106336e-01 1.55412987e-01 -1.69064730e-01 9.43856120e-01 -1.61553890e-01 1.40935883e-01 -9.26082313e-01 8.03759456e-01 -2.44578615e-01 2.17375085e-01 5.68022549e-01 -1.22551709e-01 2.84268618e-01 6.56417310e-01 -4.12801743e-01 -1.01504588e+00 -1.04395449e+00 -7.42899120e-01 6.97948277e-01 2.38501012e-01 -2.32374936e-01 -8.55864048e-01 -6.57489359e-01 -1.46931484e-01 4.20691997e-01 -1.00233090e+00 -2.19892204e-01 -2.12681815e-01 -8.03926051e-01 6.96275592e-01 4.69564646e-01 6.06591880e-01 -4.24121231e-01 -6.79956853e-01 -1.15565725e-01 1.20301053e-01 -8.97246957e-01 -1.26856148e-01 3.45665991e-01 -1.04408097e+00 -1.11285090e+00 -9.18000996e-01 -6.31289482e-01 5.09192467e-01 2.23306954e-01 7.35889792e-01 1.92335933e-01 -4.72717807e-02 2.88668215e-01 -4.22540784e-01 -3.39863777e-01 -3.43125582e-01 3.22625041e-01 4.57218647e-01 1.48919066e-02 2.30210155e-01 -9.38391805e-01 -4.59311545e-01 3.27986598e-01 -1.35302734e+00 7.96617642e-02 5.38006902e-01 7.13504791e-01 3.56437594e-01 5.03118992e-01 6.73356652e-01 -5.58270156e-01 5.51342130e-01 -2.93270826e-01 -8.64048362e-01 1.50528356e-01 -6.78959310e-01 5.86545944e-01 1.10166526e+00 -5.06699085e-01 -4.29066956e-01 -1.40713379e-01 -2.14577302e-01 -3.88591588e-01 3.91907282e-02 7.13468134e-01 -2.07794785e-01 -3.13475221e-01 6.88544214e-01 3.06246012e-01 7.91733861e-02 -5.99913359e-01 3.87101620e-01 4.31226850e-01 5.50625622e-01 -6.00304723e-01 8.01496029e-01 6.24154866e-01 4.22856629e-01 -9.98634100e-01 -9.14412260e-01 -2.05481142e-01 -9.35203314e-01 6.75113946e-02 5.71248591e-01 -1.80932850e-01 -9.34747934e-01 1.04256161e-01 -9.48906720e-01 -2.21999899e-01 -3.38937879e-01 5.20693064e-01 -7.97605038e-01 6.73647761e-01 -3.27815950e-01 -1.05870807e+00 2.43293524e-01 -9.01189387e-01 5.88646948e-01 -5.81308687e-03 2.09996641e-01 -1.53466046e+00 1.45287320e-01 -2.44027168e-01 4.83034402e-01 4.37919825e-01 9.36506510e-01 -5.98723352e-01 -1.25519007e-01 -1.19047709e-01 -3.40101272e-01 4.05525237e-01 1.95209861e-01 -2.33795106e-01 -6.27737641e-01 -1.47684023e-01 5.59670091e-01 2.25600809e-01 1.14512265e+00 2.28718847e-01 1.22112668e+00 -3.21167856e-01 -1.63914964e-01 7.06653059e-01 1.69840360e+00 -3.50106686e-01 3.81746143e-01 4.01833206e-01 5.81525683e-01 7.87371099e-01 2.27181017e-01 3.95692974e-01 1.52852371e-01 6.87183082e-01 3.57424736e-01 2.59880811e-01 1.90754294e-01 -4.42723837e-03 -2.35562958e-02 8.97171080e-01 -2.32668787e-01 8.69762972e-02 -7.49479055e-01 2.82629222e-01 -1.68218470e+00 -9.14444506e-01 -3.09634328e-01 2.61307096e+00 8.22409570e-01 1.90933332e-01 4.45794225e-01 7.72316217e-01 8.48379076e-01 -2.48257697e-01 -2.67225623e-01 -3.13499123e-01 -4.16417509e-01 1.10567436e-01 6.14435554e-01 7.51054823e-01 -1.09387600e+00 5.33244431e-01 5.42648458e+00 7.91393936e-01 -1.36828601e+00 3.86266634e-02 2.63543725e-01 1.72614500e-01 -2.77683586e-01 -1.42354250e-01 -5.81141770e-01 4.10927653e-01 6.85937107e-01 -2.52275348e-01 2.66967475e-01 8.87687564e-01 -3.81743954e-03 1.09757647e-01 -1.09255481e+00 9.37000930e-01 -4.23533544e-02 -1.11334276e+00 -4.71700095e-02 4.46014494e-01 4.26582992e-01 -3.48617703e-01 1.95216537e-01 -2.46886499e-02 -2.19333813e-01 -1.07419062e+00 5.22517920e-01 5.32363594e-01 5.42163610e-01 -9.76014435e-01 7.15316296e-01 4.43046302e-01 -1.11925125e+00 -1.40094176e-01 -6.65288925e-01 -3.43929946e-01 -1.70198426e-01 9.88954186e-01 -3.69552851e-01 8.44371438e-01 1.50152057e-01 6.06983364e-01 -5.49143314e-01 1.18602562e+00 2.03806669e-01 2.84772336e-01 -4.31777954e-01 -2.89041579e-01 1.56119511e-01 -3.92264217e-01 5.17058790e-01 1.08073080e+00 3.37686121e-01 -5.02064675e-02 -3.97322178e-01 9.04851317e-01 -3.31077464e-02 4.90856618e-01 -7.50550807e-01 9.94932577e-02 1.11513720e-04 1.11501479e+00 -1.21502936e+00 -5.80897406e-02 -2.58395851e-01 8.33755910e-01 3.68896961e-01 -1.91979129e-02 -7.82743633e-01 -6.79200351e-01 4.95716214e-01 2.59533584e-01 2.39760354e-01 -6.29304111e-01 -4.60171461e-01 -1.40945363e+00 4.96624261e-01 -2.63221741e-01 1.46775633e-01 6.93719387e-02 -1.37016332e+00 4.49586570e-01 8.27571973e-02 -1.23441589e+00 8.12638476e-02 -8.49336743e-01 -4.42631900e-01 6.16190374e-01 -1.41359329e+00 -5.96344590e-01 9.24087071e-04 5.00482857e-01 1.47396654e-01 1.61793768e-01 6.72872603e-01 4.23158318e-01 -6.33528233e-01 7.16453731e-01 3.75580490e-01 1.64727628e-01 4.09628659e-01 -1.48803592e+00 3.81533392e-02 6.96296573e-01 1.37081504e-01 8.21363151e-01 1.03828549e+00 -2.49578342e-01 -1.26543641e+00 -8.48161459e-01 7.97128081e-01 -5.28327167e-01 1.00470877e+00 -3.20925683e-01 -1.03354883e+00 2.83752620e-01 -5.05859613e-01 1.33271724e-01 5.60563505e-01 2.76724488e-01 -3.48911643e-01 -2.60059591e-02 -1.09035671e+00 5.83115458e-01 1.31208706e+00 -3.26659411e-01 -5.69091797e-01 1.76066965e-01 6.03644013e-01 1.62471145e-01 -1.05852723e+00 6.07432246e-01 6.23421967e-01 -1.21597159e+00 7.25074708e-01 -6.62932754e-01 2.30022594e-01 -1.52242035e-01 -3.97849560e-01 -1.06019390e+00 1.49732800e-02 -5.23932159e-01 3.99152078e-02 1.10309339e+00 2.01483041e-01 -9.71185088e-01 6.37029409e-01 2.75067359e-01 -1.15653500e-01 -1.20188332e+00 -1.10346210e+00 -1.34880424e+00 8.05311918e-01 -7.50966370e-01 3.68715137e-01 1.10409987e+00 4.90350008e-01 1.65862814e-01 -7.18047544e-02 1.15895763e-01 6.19449675e-01 -2.00094983e-01 4.09489959e-01 -1.69658303e+00 -3.86819690e-01 -1.14905357e+00 -9.41148281e-01 -1.12402415e+00 1.57631055e-01 -9.13969338e-01 -1.59491614e-01 -8.99748802e-01 -1.44534737e-01 -7.68366158e-01 -3.97585064e-01 -4.61670347e-02 1.32227495e-01 3.14864606e-01 -1.14620440e-01 3.24069262e-01 -4.24825370e-01 7.18835354e-01 1.12274504e+00 2.26667419e-01 -2.86893249e-01 3.06169957e-01 -7.94847190e-01 8.94786060e-01 9.71038222e-01 -1.38515353e-01 -5.29662192e-01 -1.79189205e-01 5.02629519e-01 -3.50713044e-01 5.71703970e-01 -1.29936993e+00 1.93389609e-01 1.61748707e-01 1.39750302e-01 -9.27960053e-02 1.17956914e-01 -7.32349694e-01 -5.04897416e-01 4.77838695e-01 -5.84273636e-01 -1.37430251e-01 -3.30428109e-02 6.51558280e-01 2.07667723e-01 -4.30412322e-01 1.04788566e+00 3.00541848e-01 5.64874485e-02 1.91008076e-01 -6.17657863e-02 -2.63627786e-02 8.52075100e-01 -2.12788448e-01 -3.00865751e-02 -2.51364261e-01 -9.10624802e-01 -2.44553179e-01 7.50786006e-01 1.87512323e-01 3.22365373e-01 -1.32261419e+00 -3.82772148e-01 1.49151266e-01 4.23876531e-02 -2.55650014e-01 1.61690086e-01 1.49516702e+00 -4.25964743e-01 6.48410738e-01 -5.60481176e-02 -6.07682824e-01 -7.70951569e-01 1.03900290e+00 3.35612178e-01 4.04874273e-02 -8.71399164e-01 5.33196628e-01 1.93793893e-01 3.05996137e-03 4.05683577e-01 -7.32461572e-01 -8.64722729e-02 -9.09653492e-04 5.43084085e-01 3.05549473e-01 2.11973250e-01 -5.98859370e-01 -2.74852782e-01 8.46409321e-01 1.77865475e-01 -2.71621764e-01 1.17229497e+00 -4.42839831e-01 2.90220380e-01 9.24789548e-01 1.49817789e+00 2.60146141e-01 -1.13023412e+00 -2.49790281e-01 3.25679481e-01 -1.64809957e-01 -1.57814413e-01 -1.14562102e-01 -9.15930450e-01 1.14954317e+00 4.96238261e-01 9.91288722e-01 1.20232284e+00 1.45412952e-01 6.16663337e-01 5.31367898e-01 4.93415385e-01 -8.18087459e-01 -1.91160053e-01 4.42692041e-01 7.74753273e-01 -1.16051972e+00 -3.86192858e-01 -4.93917286e-01 2.51929294e-02 1.44383395e+00 2.88427342e-02 -3.29340726e-01 8.90175521e-01 -1.91549938e-02 -2.62533098e-01 7.11900815e-02 -2.40750343e-01 -4.15798694e-01 1.58041224e-01 4.55725193e-01 6.15373492e-01 -1.61025226e-01 -6.83309495e-01 6.19840384e-01 -6.63258433e-01 -4.09590490e-02 5.06995678e-01 5.47784865e-01 -6.28188014e-01 -1.06345665e+00 -8.65074322e-02 3.10326573e-02 -6.59605563e-01 2.86975168e-02 -1.89106673e-01 9.95419323e-01 4.59997952e-02 6.93341076e-01 -1.88167274e-01 -5.26964962e-01 1.00261204e-01 1.58386994e-02 8.33808720e-01 -2.59371251e-01 9.67514738e-02 -2.82966524e-01 -4.83205497e-01 -1.69793084e-01 -5.93126953e-01 -5.26962519e-01 -1.02807021e+00 -4.93203729e-01 -4.97314632e-01 5.45370340e-01 8.36962163e-01 1.23102725e+00 3.62695009e-03 3.00483674e-01 6.65579557e-01 -6.23476326e-01 -7.80297637e-01 -8.82407784e-01 -7.80710280e-01 1.42693341e-01 5.22513032e-01 -9.16585863e-01 -6.91312432e-01 -3.17594558e-01]
[7.605167388916016, 3.940443277359009]
8762b5d9-1f00-4edd-afff-8c1be4d850d1
for-the-underrepresented-in-gender-bias
2302.00419
null
https://arxiv.org/abs/2302.00419v1
https://arxiv.org/pdf/2302.00419v1.pdf
For the Underrepresented in Gender Bias Research: Chinese Name Gender Prediction with Heterogeneous Graph Attention Network
Achieving gender equality is an important pillar for humankind's sustainable future. Pioneering data-driven gender bias research is based on large-scale public records such as scientific papers, patents, and company registrations, covering female researchers, inventors and entrepreneurs, and so on. Since gender information is often missing in relevant datasets, studies rely on tools to infer genders from names. However, available open-sourced Chinese gender-guessing tools are not yet suitable for scientific purposes, which may be partially responsible for female Chinese being underrepresented in mainstream gender bias research and affect their universality. Specifically, these tools focus on character-level information while overlooking the fact that the combinations of Chinese characters in multi-character names, as well as the components and pronunciations of characters, convey important messages. As a first effort, we design a Chinese Heterogeneous Graph Attention (CHGAT) model to capture the heterogeneity in component relationships and incorporate the pronunciations of characters. Our model largely surpasses current tools and also outperforms the state-of-the-art algorithm. Last but not least, the most popular Chinese name-gender dataset is single-character based with far less female coverage from an unreliable source, naturally hindering relevant studies. We open-source a more balanced multi-character dataset from an official source together with our code, hoping to help future research promoting gender equality.
['Haipeng Zhang', 'Shuai Ling', 'Kai Peng', 'Zihao Pan']
2023-02-01
null
null
null
null
['gender-prediction']
['computer-vision']
[-2.56283224e-01 4.22134064e-02 -8.20372224e-01 -3.92286956e-01 -1.70992717e-01 -5.89091063e-01 7.06899762e-01 3.74322474e-01 -4.75481689e-01 6.69458389e-01 6.25893533e-01 -6.40560448e-01 1.19905032e-01 -8.60678434e-01 -3.58644336e-01 -4.34439212e-01 6.15448534e-01 5.53152680e-01 -2.95229286e-01 -3.48574370e-01 6.91331625e-01 3.45727019e-02 -1.26010430e+00 -2.60454327e-01 1.18920267e+00 3.61473173e-01 -1.59340113e-01 2.00358257e-01 -5.02889335e-01 4.16758984e-01 -6.97957516e-01 -1.22274017e+00 -1.45905212e-01 -5.77887714e-01 -5.66450000e-01 -4.48335022e-01 6.09039187e-01 -1.70156822e-01 -3.43405634e-01 1.32470250e+00 6.92005754e-01 -4.60917920e-01 7.27402091e-01 -1.12758064e+00 -1.00905490e+00 9.88899052e-01 -1.11678243e+00 5.16685471e-02 3.05192061e-02 7.42862001e-02 1.39618540e+00 -6.02332294e-01 6.17806733e-01 1.49665225e+00 4.55421060e-01 6.10066891e-01 -9.55833912e-01 -1.08682418e+00 3.59454066e-01 2.20039576e-01 -1.38896620e+00 -2.49979973e-01 9.07566965e-01 -5.06899178e-01 3.35067123e-01 5.35343051e-01 7.28659451e-01 1.58240640e+00 2.16659680e-01 4.85439390e-01 1.13155615e+00 -2.71670967e-01 -1.30593345e-01 1.20515093e-01 1.16972953e-01 4.79580104e-01 8.88234019e-01 -2.61170059e-01 -5.75692236e-01 -4.15754803e-02 5.44735909e-01 9.88379866e-02 2.30668280e-02 8.95029381e-02 -1.35563409e+00 8.19167316e-01 1.99003100e-01 5.67955196e-01 3.92611511e-02 1.28757164e-01 4.21355844e-01 4.87221330e-02 6.93705976e-01 5.93023777e-01 -2.59181231e-01 -4.12062109e-01 -1.01522005e+00 4.43188220e-01 8.60037625e-01 9.71788645e-01 7.12920368e-01 1.28578842e-02 -3.65323484e-01 7.46814728e-01 2.52554297e-01 6.77159548e-01 3.41396987e-01 -6.95421875e-01 4.93830234e-01 8.36998224e-01 -2.29475841e-01 -1.63442349e+00 -2.75926232e-01 -7.95051396e-01 -9.59683716e-01 -5.54489911e-01 6.95795119e-01 -1.20920926e-01 -6.21607006e-01 1.80585909e+00 1.17891990e-01 -2.21763447e-01 -5.79826593e-01 9.58089352e-01 1.03984511e+00 1.04976490e-01 3.36217970e-01 4.07031849e-02 1.60372770e+00 -5.39833844e-01 -8.03722024e-01 -4.00897980e-01 5.26636481e-01 -6.78403318e-01 1.00014675e+00 6.12862594e-02 -4.42936480e-01 -3.13561082e-01 -7.55426586e-01 -1.16369449e-01 -5.70805788e-01 2.13081494e-01 1.06261182e+00 1.17316210e+00 -6.29536450e-01 5.08142054e-01 -4.41775858e-01 -3.03079426e-01 6.57021463e-01 1.27333105e-01 -9.45192799e-02 -1.31943151e-01 -1.11745536e+00 6.47328973e-01 -1.60665497e-01 1.09937012e-01 -4.97205019e-01 -1.07117379e+00 -8.33980083e-01 -1.70844391e-01 6.46444201e-01 -6.47066534e-01 7.15782940e-01 -6.87955976e-01 -9.43881154e-01 1.01398289e+00 -3.05134386e-01 2.31705114e-01 6.00732327e-01 -4.22277302e-02 -6.38401568e-01 -2.88170159e-01 4.87427175e-01 5.27508616e-01 5.97598314e-01 -1.12068486e+00 -5.21668613e-01 -8.50748658e-01 -1.48315623e-01 -3.67707349e-02 -7.16423631e-01 8.47671852e-02 -6.53672934e-01 -7.30108202e-01 -3.95477889e-03 -9.39671874e-01 -2.05368415e-01 -3.69779348e-01 -8.11391115e-01 -3.41351748e-01 4.76408303e-01 -8.92628193e-01 1.56997657e+00 -2.09346247e+00 -7.54176825e-02 7.16346353e-02 7.48962402e-01 -1.22514874e-01 9.11718700e-03 4.33196157e-01 1.98676199e-01 6.04848683e-01 5.66042513e-02 -2.23451898e-01 1.40784234e-01 -1.79189771e-01 -1.65456489e-01 6.55143499e-01 1.06060423e-01 1.00107670e+00 -1.07973814e+00 -5.46250641e-01 -1.79116651e-01 2.84794033e-01 -6.25178099e-01 -3.71166617e-01 -7.14576151e-03 5.90238273e-01 -4.22652930e-01 1.09024501e+00 8.96138430e-01 -1.39286304e-02 4.11648214e-01 -1.09885767e-01 -3.75674546e-01 3.35444957e-01 -8.49431813e-01 1.23798680e+00 -1.39574766e-01 8.14154565e-01 -1.48590833e-01 -7.90866852e-01 1.08873856e+00 -2.61428684e-01 2.89345056e-01 -7.61133373e-01 2.99528807e-01 5.89106143e-01 4.16436732e-01 -2.98287481e-01 8.79104078e-01 7.75151653e-03 -4.48773205e-01 1.33423224e-01 -2.65469998e-01 -1.17535354e-03 4.82849538e-01 3.87228519e-01 6.75941050e-01 4.03366499e-02 -1.36171430e-01 -5.04036009e-01 2.63551205e-01 -1.42624021e-01 9.28143799e-01 6.92450583e-01 -2.40953833e-01 7.88502812e-01 1.31453574e+00 -1.65806204e-01 -1.04803252e+00 -4.78479713e-01 -1.54715210e-01 1.14576674e+00 5.02875075e-02 -6.31511867e-01 -6.11867726e-01 -6.47226691e-01 2.42117748e-01 5.93206227e-01 -8.91786158e-01 -1.23358175e-01 -4.37942713e-01 -9.20704901e-01 6.96813703e-01 3.85701507e-01 4.33910459e-01 -6.20819747e-01 -8.17692578e-02 -1.59390405e-01 -2.67964929e-01 -1.00304103e+00 -6.29296660e-01 -2.33663753e-01 -4.24001068e-01 -1.14041185e+00 -1.13357151e+00 -4.78686243e-01 6.43232405e-01 -4.29378673e-02 1.27614498e+00 1.76944375e-01 -1.61788434e-01 -1.73498675e-01 -1.27818689e-01 -7.56359816e-01 -1.03856318e-01 5.60650170e-01 -1.49359182e-01 3.94134037e-02 6.96033180e-01 -4.42840189e-01 -6.37000799e-01 1.60558373e-01 -3.84628445e-01 -7.09811077e-02 6.24351501e-01 5.00747442e-01 1.30992323e-01 -1.59958094e-01 5.49137771e-01 -1.42599547e+00 4.53081906e-01 -6.93332136e-01 -3.14862996e-01 -6.08911328e-02 -8.95298481e-01 -1.49442345e-01 4.18630868e-01 -3.85776669e-01 -8.43111992e-01 -8.15870523e-01 -5.61553910e-02 -3.60053144e-02 7.14378357e-02 6.06397808e-01 -3.66823256e-01 3.07804286e-01 3.03589642e-01 -8.32861662e-02 -5.97102642e-02 -5.41315973e-01 8.52080211e-02 8.56032073e-01 3.92949671e-01 -7.51834750e-01 6.78323865e-01 2.19111145e-01 3.04961111e-02 -7.12705731e-01 -7.19167411e-01 -2.74381787e-01 -2.69229800e-01 -1.27956681e-02 6.39846206e-01 -1.05815351e+00 -8.57739747e-01 5.96108913e-01 -9.09434199e-01 1.11033171e-01 2.04212710e-01 4.84945655e-01 3.09168965e-01 1.99105546e-01 -5.84482849e-01 -7.11501658e-01 -2.58709639e-02 -1.00631261e+00 9.59561229e-01 4.38143164e-01 -4.16281283e-01 -8.43712330e-01 -1.34283110e-01 5.74611664e-01 4.82359439e-01 1.25361726e-01 1.26597393e+00 -7.41463482e-01 -2.57676780e-01 -1.51330903e-01 -6.31472886e-01 -2.32719630e-01 1.29884854e-02 4.80895609e-01 -7.36873925e-01 -6.08266424e-03 -5.93171000e-01 9.89564210e-02 1.02228975e+00 3.21275890e-01 1.27448678e+00 -2.30648234e-01 -5.06868064e-01 3.51604849e-01 1.21517003e+00 -1.54490724e-01 4.75418955e-01 1.11802742e-01 1.41010547e+00 9.38413203e-01 2.04913348e-01 5.37420213e-01 1.13624167e+00 3.42649847e-01 3.39148700e-01 -9.53437462e-02 -2.07837280e-02 -4.83410716e-01 -5.59544340e-02 8.36763978e-01 -3.41403067e-01 -9.80889648e-02 -1.15092731e+00 8.10129881e-01 -1.46403873e+00 -8.11919451e-01 -6.38310015e-01 2.06379485e+00 9.86561298e-01 1.73021466e-01 2.02078000e-01 -5.65260649e-02 8.91630530e-01 3.67849708e-01 -4.41568017e-01 -3.26933116e-01 -2.77418703e-01 3.39923091e-02 7.21444726e-01 4.65755612e-02 -8.52150798e-01 9.68241632e-01 5.76186466e+00 9.23038065e-01 -1.15121472e+00 -2.19964787e-01 1.00869167e+00 5.76955229e-02 -1.04994655e+00 2.51636267e-01 -8.63800347e-01 7.71187901e-01 6.61848843e-01 -2.75750101e-01 2.21855432e-01 6.81906581e-01 9.70336124e-02 -1.31455123e-01 -8.19143951e-01 1.06929493e+00 3.71622071e-02 -1.15422618e+00 -1.98388342e-02 5.26418626e-01 6.75504148e-01 -3.57378542e-01 2.68590957e-01 4.43637460e-01 1.38443053e-01 -1.07145774e+00 8.93072188e-01 1.65931597e-01 1.04408562e+00 -9.28030372e-01 8.74061942e-01 2.07472183e-02 -7.52192140e-01 -6.04812130e-02 -3.45371068e-01 -2.25422010e-01 -3.40010434e-01 1.02127099e+00 -3.47898364e-01 6.51825905e-01 6.91402197e-01 9.36638772e-01 -1.04537666e+00 6.71352148e-01 -1.96034521e-01 8.34377825e-01 1.77983075e-01 -3.53504717e-01 3.64894122e-02 -3.18999857e-01 3.03796619e-01 1.17147100e+00 3.45145494e-01 -2.27246478e-01 -1.85664281e-01 1.13215649e+00 -3.15724224e-01 3.69084805e-01 -5.35265267e-01 -8.45030069e-01 7.55291283e-01 1.45022476e+00 -1.07320869e+00 2.23267339e-02 -6.64565802e-01 4.74378854e-01 4.61941808e-02 4.50772166e-01 -5.79427958e-01 -3.90199840e-01 8.75437677e-01 2.73738444e-01 4.25294740e-03 -1.76969722e-01 -8.27736795e-01 -1.14220858e+00 -2.39999622e-01 -1.18199492e+00 3.81585449e-01 -4.26403396e-02 -1.17898357e+00 -1.62089791e-03 -2.39313543e-01 -7.70485699e-01 1.02523588e-01 -4.98357803e-01 -4.60115880e-01 9.10698175e-01 -1.22551203e+00 -1.37271369e+00 -2.04417706e-01 7.93596134e-02 6.24914318e-02 -3.74625951e-01 4.97437268e-01 6.01583481e-01 -1.07113123e+00 9.55746293e-01 -3.40818405e-01 4.72884655e-01 1.04347467e+00 -1.18671823e+00 5.06454468e-01 6.12379193e-01 1.89218763e-02 9.90153790e-01 7.73821115e-01 -1.00706697e+00 -1.54653418e+00 -6.30396426e-01 1.43132174e+00 -6.27382159e-01 6.12198710e-01 -5.95776856e-01 -5.70085764e-01 4.22493368e-01 4.50796224e-02 -2.32867315e-01 7.46555328e-01 6.77165747e-01 -5.10631204e-01 -6.03326708e-02 -7.20170081e-01 9.87284839e-01 1.26169503e+00 -4.77663636e-01 3.69716622e-02 -2.11571902e-01 6.38266087e-01 -2.87723720e-01 -5.39420545e-01 3.02378923e-01 5.88068187e-01 -9.08342481e-01 6.64529026e-01 -4.51823950e-01 1.07745266e+00 5.93413785e-02 -1.73515715e-02 -1.06719136e+00 -3.41200411e-01 -6.10757172e-01 9.74421501e-02 1.81362259e+00 5.52385032e-01 -5.91631830e-01 8.07053685e-01 7.25183785e-01 5.75455688e-02 -6.97972894e-01 -6.63362861e-01 -3.23727906e-01 4.28114623e-01 -2.82544762e-01 8.96844804e-01 1.27093482e+00 -2.62626350e-01 4.27407801e-01 -3.15850198e-01 -3.85849923e-01 5.51115990e-01 2.65688479e-01 7.64209449e-01 -1.37668502e+00 1.32501319e-01 -9.48331058e-01 -2.81180918e-01 -4.66050684e-01 5.26673555e-01 -8.78983498e-01 -4.66131479e-01 -1.26334560e+00 7.25895166e-01 -5.91915131e-01 -3.36789759e-03 3.63733679e-01 -6.02367580e-01 4.61889029e-01 -9.38734785e-02 -4.07973602e-02 -2.88101315e-01 4.29662108e-01 1.60313547e+00 -3.53391439e-01 7.41398428e-03 -1.59545690e-01 -1.66671193e+00 5.47254622e-01 5.11997163e-01 -5.43608963e-01 -6.82016686e-02 -3.49674851e-01 6.56062424e-01 -3.46308559e-01 2.97806650e-01 -4.35952514e-01 6.28428459e-02 -4.64655638e-01 4.44181532e-01 -4.58731323e-01 -2.18595341e-01 -3.60724002e-01 -4.20590751e-02 3.34907591e-01 -1.72382995e-01 1.14200942e-01 -3.20046693e-02 3.68450135e-01 -1.02563947e-01 -4.11556149e-03 2.13624209e-01 -2.35623345e-01 -3.98160219e-01 4.30336118e-01 -1.42781302e-01 4.70999062e-01 5.38081110e-01 -1.81456104e-01 -6.24107599e-01 -3.68497699e-01 5.16009480e-02 1.44177422e-01 6.70549512e-01 7.23347247e-01 4.14046161e-02 -1.20836473e+00 -1.03904951e+00 1.62427768e-01 2.28160933e-01 -3.07899565e-01 2.47492254e-01 8.78110588e-01 -2.91007370e-01 3.10013294e-01 -1.07907914e-01 -1.39105424e-01 -9.34103608e-01 3.12033474e-01 -1.90336615e-01 1.89581029e-02 -2.63202429e-01 6.95186257e-01 1.02666095e-01 -6.34500086e-01 5.08774929e-02 6.99874386e-03 -4.24548954e-01 5.76344073e-01 2.75658935e-01 6.17301464e-01 -1.37231320e-01 -6.71994567e-01 -6.88972235e-01 4.84064162e-01 -6.55081943e-02 -9.30144191e-02 1.22661996e+00 -1.20483264e-01 -5.28572083e-01 4.67706531e-01 1.03784025e+00 7.65632808e-01 -7.05493748e-01 5.08011989e-02 8.03377703e-02 -8.00540328e-01 -2.09994361e-01 -5.63872755e-01 -1.22566783e+00 7.34064162e-01 -8.44420046e-02 -1.04919665e-01 5.60432494e-01 -4.17417251e-02 5.62773347e-01 -2.12321028e-01 2.87453026e-01 -1.28988600e+00 -5.38950637e-02 4.03504729e-01 4.33436990e-01 -1.37081504e+00 2.23866329e-01 -5.96029758e-01 -6.18855536e-01 8.63581240e-01 7.82147706e-01 5.12793303e-01 4.47082490e-01 6.50868788e-02 1.60990685e-01 -4.70159985e-02 -4.43141520e-01 -1.43759832e-01 3.65483969e-01 5.37459970e-01 9.72783446e-01 2.20579669e-01 -9.15530741e-01 1.15213406e+00 -5.48147559e-01 -4.61098939e-01 4.68163282e-01 3.63229066e-01 8.58439580e-02 -1.34492612e+00 -2.12678209e-01 7.40347564e-01 -9.21511650e-01 -3.23748022e-01 -6.71682775e-01 6.25744402e-01 3.20371360e-01 9.30835664e-01 1.87039942e-01 -4.34704870e-01 8.13980401e-02 -2.51132369e-01 4.02201146e-01 -4.15595591e-01 -5.71347654e-01 -5.98846376e-02 2.76148528e-01 -1.49163887e-01 -6.74078241e-02 -7.59431958e-01 -8.92112076e-01 -9.66905296e-01 -1.10788494e-01 -8.85679200e-02 7.33615696e-01 7.81315863e-01 4.22898889e-01 4.91068393e-01 5.47388196e-01 -4.42218482e-01 -2.47925758e-01 -9.20614481e-01 -7.57422090e-01 3.33972961e-01 -2.26850845e-02 -5.54874480e-01 -2.53143430e-01 -3.98328185e-01]
[9.388835906982422, 10.261147499084473]
5195a86e-6ee5-4c06-b907-bf5653317e79
2305-14984
2305.14984
null
https://arxiv.org/abs/2305.14984v1
https://arxiv.org/pdf/2305.14984v1.pdf
Adversarial robustness of amortized Bayesian inference
Bayesian inference usually requires running potentially costly inference procedures separately for every new observation. In contrast, the idea of amortized Bayesian inference is to initially invest computational cost in training an inference network on simulated data, which can subsequently be used to rapidly perform inference (i.e., to return estimates of posterior distributions) for new observations. This approach has been applied to many real-world models in the sciences and engineering, but it is unclear how robust the approach is to adversarial perturbations in the observed data. Here, we study the adversarial robustness of amortized Bayesian inference, focusing on simulation-based estimation of multi-dimensional posterior distributions. We show that almost unrecognizable, targeted perturbations of the observations can lead to drastic changes in the predicted posterior and highly unrealistic posterior predictive samples, across several benchmark tasks and a real-world example from neuroscience. We propose a computationally efficient regularization scheme based on penalizing the Fisher information of the conditional density estimator, and show how it improves the adversarial robustness of amortized Bayesian inference.
['Jakob H. Macke', 'Michael Deistler', 'Manuel Glöckler']
2023-05-24
null
null
null
null
['bayesian-inference']
['methodology']
[ 4.72296327e-01 1.31991580e-01 4.94204760e-01 -4.06157643e-01 -9.42539096e-01 -6.12942517e-01 6.33759558e-01 -5.12730144e-02 -6.61313474e-01 1.11783290e+00 -2.15365335e-01 -3.13717753e-01 -1.49517432e-01 -6.94988072e-01 -1.31078875e+00 -9.00137246e-01 -1.52010083e-01 6.21205270e-01 9.01679881e-03 3.52921695e-01 3.07955116e-01 6.47464752e-01 -1.17511284e+00 -4.32016671e-01 6.89754844e-01 7.20987260e-01 -1.45198599e-01 8.00033927e-01 5.41815460e-01 1.67928472e-01 -6.15397692e-01 -7.36856699e-01 1.28155574e-01 -2.66115427e-01 -4.79389846e-01 -2.24305436e-01 4.24656868e-01 -4.08017665e-01 -2.42855385e-01 1.44426262e+00 5.17051280e-01 4.22407478e-01 1.08675480e+00 -9.66645181e-01 -1.52576268e-01 5.94213903e-01 -5.92103958e-01 2.73429126e-01 2.15692788e-01 3.81392479e-01 7.05026567e-01 -6.21620059e-01 5.31710029e-01 1.50144446e+00 7.56916046e-01 4.80014414e-01 -1.85118461e+00 -6.92222774e-01 1.45078480e-01 -2.55475223e-01 -1.33354533e+00 -5.24360597e-01 5.28299272e-01 -5.73135495e-01 3.61017674e-01 3.56500968e-02 4.11305696e-01 1.61503339e+00 5.45714378e-01 3.09522867e-01 1.05478370e+00 -4.97090295e-02 7.82118261e-01 7.63027444e-02 -2.40809117e-02 3.78983259e-01 6.01196170e-01 4.22284037e-01 -3.23375940e-01 -4.48020071e-01 4.38721061e-01 -8.44586343e-02 -2.67849594e-01 -1.11718491e-01 -9.34375644e-01 7.79228032e-01 9.07403678e-02 -2.65611082e-01 -3.81360561e-01 4.96926546e-01 3.47352445e-01 2.04050183e-01 6.08125389e-01 3.59367609e-01 -4.23907518e-01 -1.06386356e-01 -7.98127234e-01 7.23873556e-01 9.13927853e-01 5.32397747e-01 6.05074704e-01 1.46793500e-01 -7.91482329e-02 5.51461518e-01 3.95764112e-01 9.65742648e-01 -1.49631826e-02 -1.27999115e+00 2.95775056e-01 -3.04360181e-01 4.49264735e-01 -1.05557048e+00 -2.57269979e-01 -3.11101913e-01 -9.50887263e-01 3.12814713e-01 8.54665220e-01 -5.80410063e-01 -8.99371326e-01 2.15177083e+00 3.27790111e-01 4.20985609e-01 -7.40804747e-02 6.65200591e-01 1.69494718e-01 5.93480468e-01 5.39504960e-02 -2.18726233e-01 9.21496511e-01 4.84142639e-03 -5.80691636e-01 -3.24115127e-01 -8.69368538e-02 -5.46536326e-01 8.15045118e-01 5.00208080e-01 -1.27969956e+00 -1.21392019e-01 -9.40229833e-01 3.73964876e-01 -3.04580837e-01 -3.97203773e-01 3.36508721e-01 7.47588992e-01 -4.90183383e-01 9.53356862e-01 -1.13960826e+00 -1.43546471e-02 6.90573215e-01 2.65161127e-01 -3.01742464e-01 -1.76246196e-01 -1.14501858e+00 9.25855815e-01 4.02997524e-01 2.06966087e-01 -1.32639778e+00 -1.09206724e+00 -7.46873081e-01 1.89536631e-01 4.25083607e-01 -7.28278577e-01 9.81754422e-01 -6.07009768e-01 -1.69199955e+00 4.15924251e-01 -2.31354386e-02 -6.68343544e-01 7.40827203e-01 -2.22281292e-01 -9.29928478e-03 -1.59719978e-02 -2.29114473e-01 4.41740274e-01 1.19897485e+00 -1.07746661e+00 -4.53058928e-02 -3.82875681e-01 -1.11626290e-01 -2.22129673e-01 6.33136034e-02 -1.98769048e-01 -7.06605539e-02 -7.04956770e-01 1.22085012e-01 -1.10523045e+00 -3.73268425e-01 8.51735920e-02 -5.05998135e-01 2.50895828e-01 3.94541949e-01 -6.40966177e-01 6.35627866e-01 -2.08318138e+00 3.30283731e-01 5.21248460e-01 6.55899718e-02 -1.35299742e-01 2.17189103e-01 2.46640295e-02 -1.63782039e-03 6.82484061e-02 -7.09214628e-01 -3.46154213e-01 2.05447286e-01 2.13572130e-01 -6.31278515e-01 8.41929376e-01 1.67060018e-01 6.75602198e-01 -8.86711359e-01 -1.04500905e-01 -8.63023289e-03 4.35234308e-01 -8.24284613e-01 2.35368654e-01 -3.21730196e-01 5.81474841e-01 -2.56699204e-01 4.02359590e-02 1.00267351e+00 -2.11687386e-01 1.44921124e-01 -4.08251658e-02 4.05017257e-01 -6.33659661e-02 -1.21303046e+00 1.24191236e+00 -3.85624796e-01 5.88424802e-01 5.12215793e-02 -1.26417196e+00 4.05826122e-01 1.79037496e-01 -4.91004856e-03 1.92970708e-02 2.43893266e-01 5.94346933e-02 -7.96913058e-02 -1.00688033e-01 2.22939365e-02 -5.74582398e-01 -3.51064742e-01 5.79736114e-01 1.20208539e-01 -5.03786683e-01 -1.81326885e-02 3.35514873e-01 1.02072394e+00 9.58062336e-02 1.06251322e-01 -3.28797460e-01 1.22350007e-01 -4.17834073e-01 5.94120502e-01 1.19370091e+00 4.32934612e-02 4.92473841e-01 6.37563407e-01 -2.13867977e-01 -1.13339722e+00 -1.65898824e+00 -3.39073718e-01 6.64170384e-01 -1.91001520e-01 1.27934366e-01 -8.40938032e-01 -5.57936013e-01 3.49628001e-01 9.42771196e-01 -6.75417960e-01 -5.52879870e-01 -2.71789938e-01 -1.18241453e+00 6.25464261e-01 3.49585980e-01 1.54241428e-01 -6.77798748e-01 -5.38238525e-01 2.24871412e-01 4.90069427e-02 -1.01766860e+00 -2.82623589e-01 1.62639305e-01 -7.64484525e-01 -7.89126456e-01 -7.08250999e-01 2.50744581e-01 8.24419498e-01 -3.75022560e-01 1.05093932e+00 -5.31652510e-01 -3.28897774e-01 4.09373194e-01 1.93737298e-01 -6.79670513e-01 -6.35538697e-01 -3.51172924e-01 5.08529484e-01 6.22633994e-02 3.40790600e-02 -8.41407359e-01 -3.73959810e-01 8.61707404e-02 -8.87322307e-01 -4.56843853e-01 3.42337340e-01 8.91797781e-01 4.60084260e-01 -1.40203638e-02 7.27625310e-01 -1.08684230e+00 6.06780589e-01 -6.92592740e-01 -1.16955829e+00 2.97515150e-02 -3.89322042e-01 4.09646988e-01 8.68384242e-01 -6.11416519e-01 -1.26343703e+00 -1.25457153e-01 -5.72579205e-02 -5.75726926e-01 -6.78528249e-02 4.75393534e-01 8.05556104e-02 -6.15992062e-02 5.75875938e-01 3.99205051e-02 4.50299717e-02 -3.82077545e-01 3.02758306e-01 2.38284439e-01 6.22116029e-01 -8.97016466e-01 9.79564488e-01 6.48294449e-01 5.09166062e-01 -6.54784739e-01 -1.05861366e+00 3.03766608e-01 -2.81889021e-01 -1.38734788e-01 8.12935650e-01 -7.04938471e-01 -1.03987408e+00 5.61333179e-01 -1.09663785e+00 -2.85672158e-01 -3.15619022e-01 7.26912022e-01 -8.28884423e-01 3.61172646e-01 -2.82258898e-01 -9.76305127e-01 3.26016024e-02 -1.10600376e+00 7.54936755e-01 1.64826095e-01 -2.07621098e-01 -1.11586905e+00 1.35445133e-01 -3.20663378e-02 3.27880055e-01 2.77682155e-01 8.51845205e-01 -5.60719132e-01 -4.32311863e-01 -3.81929636e-01 -9.76217017e-02 5.87558448e-01 -2.52196699e-01 2.26098463e-01 -1.03751028e+00 -3.83388281e-01 5.18825576e-02 -4.11632240e-01 1.03073406e+00 6.77748680e-01 1.35200620e+00 -3.29518527e-01 -2.75776803e-01 6.25387490e-01 1.21363115e+00 -1.21228531e-01 5.30361891e-01 -2.79500306e-01 3.00165445e-01 4.08134282e-01 4.78391171e-01 6.35160685e-01 -2.08362162e-01 4.17715073e-01 2.83224851e-01 5.17123640e-01 4.63367254e-01 -1.90695107e-01 3.52723539e-01 3.32615197e-01 -7.20365122e-02 -2.52137929e-01 -6.93909287e-01 2.69572645e-01 -1.64592695e+00 -1.11179650e+00 4.13066953e-01 2.51090193e+00 1.06175435e+00 3.40113521e-01 -2.09951863e-01 -2.33212084e-01 7.57896662e-01 -1.75738007e-01 -9.38864768e-01 -2.80960947e-01 5.49075380e-02 3.52003187e-01 6.05653346e-01 6.60379112e-01 -1.04695046e+00 5.78441262e-01 7.69159126e+00 8.72869134e-01 -7.87226677e-01 6.44213334e-02 7.24673152e-01 -4.06815588e-01 -1.56913072e-01 -1.23525530e-01 -6.24979913e-01 8.02787721e-01 1.26048100e+00 -3.60019475e-01 5.64025283e-01 6.35178685e-01 2.10353360e-01 -3.23293567e-01 -1.31987512e+00 7.35446274e-01 -2.34288216e-01 -1.29022801e+00 -4.95495461e-02 3.04776542e-02 7.27087080e-01 -2.29694620e-02 2.37299874e-01 1.57679915e-01 9.22527432e-01 -1.08108699e+00 4.63202953e-01 1.03657210e+00 4.84006584e-01 -8.88369322e-01 6.86424971e-01 4.25683469e-01 -3.84313911e-01 1.14086673e-01 -6.06328309e-01 9.65557173e-02 2.06316069e-01 9.98179793e-01 -6.30663157e-01 1.75819904e-01 7.57190704e-01 4.57347989e-01 -1.46812260e-01 7.45600820e-01 -1.39101163e-01 9.23062444e-01 -7.91144073e-01 5.56627437e-02 -8.56988356e-02 -3.60040605e-01 8.68654966e-01 1.14457452e+00 3.75551641e-01 -1.55516475e-01 -2.55119920e-01 1.28871787e+00 -1.74834922e-01 -4.77959812e-01 -6.53196335e-01 3.23004201e-02 5.67555606e-01 7.79820025e-01 -5.84386945e-01 -3.48830640e-01 -1.17668882e-01 6.84132636e-01 3.00074965e-01 6.74320757e-01 -9.83481109e-01 -3.31538409e-01 7.19434679e-01 -2.21141547e-01 5.86977780e-01 -4.47995141e-02 -1.81329533e-01 -1.13736248e+00 -2.24210601e-02 -7.26100981e-01 3.65724742e-01 -6.90539181e-01 -1.66662467e+00 -1.08843163e-01 3.22984606e-01 -8.25956523e-01 -3.15813750e-01 -4.48679775e-01 -6.62905276e-01 1.06201816e+00 -1.08905292e+00 -4.03633177e-01 1.25092417e-01 3.70061100e-01 4.47727926e-02 -1.10895149e-01 6.72093987e-01 9.54003185e-02 -8.13503146e-01 7.01626718e-01 5.80291331e-01 -4.53258716e-02 7.49725342e-01 -1.20027030e+00 5.10003328e-01 9.34628367e-01 -2.27667660e-01 7.78589487e-01 1.31537390e+00 -5.79021454e-01 -1.34664106e+00 -1.20941305e+00 1.59812525e-01 -4.77772802e-01 8.80842686e-01 -3.85490924e-01 -9.25747275e-01 9.30462182e-01 -2.34861985e-01 2.55984306e-01 4.17003393e-01 -2.16715299e-02 -4.78917420e-01 -1.32705748e-01 -1.53315485e+00 6.84367299e-01 6.49330556e-01 -3.45163196e-01 -5.79916596e-01 3.72574925e-01 5.50220013e-01 -1.89963505e-01 -9.06606972e-01 1.79383740e-01 7.60475874e-01 -6.99967504e-01 1.20341766e+00 -8.18780243e-01 4.26028758e-01 -1.41612142e-01 -2.79257476e-01 -1.48147643e+00 1.42196506e-01 -8.39421690e-01 -2.66384989e-01 9.15464699e-01 4.44206893e-01 -9.39779580e-01 5.42802334e-01 7.13697851e-01 2.71911055e-01 -4.29546654e-01 -1.28716958e+00 -8.78908932e-01 4.34073538e-01 -6.52543128e-01 3.00354630e-01 5.19095659e-01 -2.45298564e-01 -8.34381673e-03 -3.49107832e-01 4.83473748e-01 1.30202508e+00 -2.09326506e-01 7.19311595e-01 -1.20651579e+00 -6.81038678e-01 -2.28999004e-01 -5.04486561e-01 -9.37938392e-01 4.65030313e-01 -6.21785462e-01 4.06634867e-01 -7.28350043e-01 3.52772236e-01 -8.45027938e-02 -1.27770916e-01 2.74599442e-05 -4.88768816e-01 2.59778917e-01 2.19667312e-02 -1.05621047e-01 -3.85150820e-01 5.59993148e-01 9.64418590e-01 -1.26118839e-01 2.76489466e-01 2.69700944e-01 -4.81814206e-01 9.96526659e-01 5.69812775e-01 -9.80655372e-01 -2.68268198e-01 -3.09144333e-02 6.14387631e-01 3.36027220e-02 7.92381704e-01 -8.56758654e-01 1.31835759e-01 -2.58791834e-01 4.62907135e-01 -4.39409196e-01 5.09961486e-01 -6.40047908e-01 1.30073369e-01 4.53212351e-01 -3.90010476e-01 -5.14682651e-01 2.85231650e-01 1.13734055e+00 1.98107064e-01 -5.15422940e-01 1.31181300e+00 -9.92157310e-02 1.27187297e-01 2.32713297e-01 -7.18241155e-01 3.98730606e-01 9.56800938e-01 2.52281070e-01 -1.86149657e-01 -4.91938144e-01 -1.08549154e+00 1.63917858e-02 2.67575711e-01 -3.93598050e-01 5.87247491e-01 -9.21326816e-01 -8.93548548e-01 9.78691503e-02 -2.60593057e-01 1.34884054e-02 3.01119119e-01 7.14739084e-01 -2.78120875e-01 -9.58004817e-02 -1.08038466e-02 -6.21772289e-01 -8.91687214e-01 4.62078065e-01 5.01093447e-01 -6.44415841e-02 -4.87007976e-01 9.04686093e-01 9.41556245e-02 -4.00340885e-01 2.16812640e-01 -2.95117289e-01 3.42387229e-01 -1.93315998e-01 4.96311575e-01 5.26411414e-01 -2.09584385e-01 -3.23618948e-02 -2.54032373e-01 4.26137239e-01 -2.41639376e-01 -4.21819240e-01 1.23557341e+00 -3.81800197e-02 -4.24780883e-02 7.46756792e-01 1.10802197e+00 -1.26450494e-01 -1.71953046e+00 -8.83312300e-02 -2.21445948e-01 -4.66994017e-01 8.65712315e-02 -6.97439909e-01 -9.37979102e-01 9.47947860e-01 3.84600163e-01 1.34121686e-01 7.45177329e-01 -3.18804160e-02 4.68796521e-01 8.65040720e-01 2.60170102e-01 -1.03982556e+00 -1.16255648e-01 3.65762204e-01 7.85938978e-01 -1.11744618e+00 3.34556609e-01 -1.23507038e-01 -3.16035122e-01 8.59127879e-01 5.07760420e-02 -3.99260521e-01 1.06099796e+00 3.83364230e-01 -4.53904897e-01 2.01624800e-02 -7.54390180e-01 3.98522913e-01 3.90166938e-02 6.29470587e-01 -7.12739676e-02 -5.75400516e-02 1.45773411e-01 5.48946738e-01 -9.34710503e-02 -1.02897055e-01 7.97726750e-01 6.51693106e-01 -2.50993282e-01 -5.40039062e-01 -4.08882767e-01 6.72947466e-01 -7.19757676e-01 -1.11032113e-01 2.94599850e-02 4.91493434e-01 -2.86566198e-01 6.94485009e-01 1.39664114e-01 6.97801337e-02 1.04776258e-02 2.49500796e-01 6.43847227e-01 -3.42454135e-01 3.76104526e-02 -1.82255715e-01 -1.05705947e-01 -5.41767657e-01 -2.68782198e-01 -1.04094100e+00 -8.07503998e-01 -5.38668811e-01 -1.92153022e-01 3.17924581e-02 6.82780504e-01 1.13790405e+00 2.13832438e-01 4.91676152e-01 4.79164630e-01 -1.11712193e+00 -1.10669243e+00 -1.00137448e+00 -6.68128550e-01 4.77932036e-01 3.88198704e-01 -7.91567445e-01 -1.11580658e+00 8.43794942e-02]
[6.963768005371094, 3.893561601638794]
421fd8f4-8013-49d7-8656-9ba6d6f83291
v2c-visual-voice-cloning
2111.1289
null
https://arxiv.org/abs/2111.12890v1
https://arxiv.org/pdf/2111.12890v1.pdf
V2C: Visual Voice Cloning
Existing Voice Cloning (VC) tasks aim to convert a paragraph text to a speech with desired voice specified by a reference audio. This has significantly boosted the development of artificial speech applications. However, there also exist many scenarios that cannot be well reflected by these VC tasks, such as movie dubbing, which requires the speech to be with emotions consistent with the movie plots. To fill this gap, in this work we propose a new task named Visual Voice Cloning (V2C), which seeks to convert a paragraph of text to a speech with both desired voice specified by a reference audio and desired emotion specified by a reference video. To facilitate research in this field, we construct a dataset, V2C-Animation, and propose a strong baseline based on existing state-of-the-art (SoTA) VC techniques. Our dataset contains 10,217 animated movie clips covering a large variety of genres (e.g., Comedy, Fantasy) and emotions (e.g., happy, sad). We further design a set of evaluation metrics, named MCD-DTW-SL, which help evaluate the similarity between ground-truth speeches and the synthesised ones. Extensive experimental results show that even SoTA VC methods cannot generate satisfying speeches for our V2C task. We hope the proposed new task together with the constructed dataset and evaluation metric will facilitate the research in the field of voice cloning and the broader vision-and-language community.
['Qi Wu', 'Mingkui Tan', 'Jiaqiu Zhou', 'Yuankai Qi', 'Yuanqing Li', 'Qi Chen']
2021-11-25
null
http://openaccess.thecvf.com//content/CVPR2022/html/Chen_V2C_Visual_Voice_Cloning_CVPR_2022_paper.html
http://openaccess.thecvf.com//content/CVPR2022/papers/Chen_V2C_Visual_Voice_Cloning_CVPR_2022_paper.pdf
cvpr-2022-1
['voice-cloning']
['speech']
[ 3.53262983e-02 -1.08285420e-01 1.03790581e-01 -4.45732832e-01 -6.65055037e-01 -6.40401304e-01 6.81969464e-01 -6.37981653e-01 8.72727185e-02 4.57811087e-01 6.15558565e-01 -1.60931647e-01 6.21521413e-01 -3.86417210e-01 -5.50918043e-01 -5.60819983e-01 3.67356002e-01 -3.22434977e-02 5.01880720e-02 -3.44567955e-01 6.03393055e-02 2.94512540e-01 -1.69359255e+00 6.24696434e-01 5.75270593e-01 1.11440456e+00 3.77486169e-01 7.37109959e-01 -3.55909079e-01 6.95130289e-01 -8.29711199e-01 -6.08546138e-01 -1.35483623e-01 -9.65419114e-01 -7.82602310e-01 3.65530252e-01 2.71092743e-01 -7.19605237e-02 -3.69056091e-02 1.09268844e+00 8.31147552e-01 4.15182739e-01 6.87453091e-01 -1.55490339e+00 -9.22749817e-01 5.91639876e-01 -1.93958461e-01 -2.27102965e-01 6.53521299e-01 2.73635209e-01 1.14768088e+00 -1.23401761e+00 9.99744058e-01 1.63356936e+00 3.73085380e-01 1.07113397e+00 -9.04103220e-01 -7.56543934e-01 2.10864127e-01 8.52577612e-02 -1.20808446e+00 -7.62932956e-01 1.19859469e+00 -2.33239442e-01 5.73963702e-01 7.37713397e-01 6.76045537e-01 1.86442387e+00 -2.93523699e-01 8.73648763e-01 9.24104214e-01 -1.34035990e-01 3.03232789e-01 3.53018552e-01 -4.99842793e-01 1.91621989e-01 -5.42404830e-01 1.67132095e-02 -5.39938033e-01 2.50831693e-01 4.54088390e-01 -5.23330569e-01 -6.82513833e-01 -5.80407195e-02 -1.45171809e+00 8.34020078e-01 1.25359461e-01 4.00662631e-01 -1.79391980e-01 -1.31269991e-01 6.03963733e-01 4.57419932e-01 4.06970888e-01 3.85892093e-01 -5.32184392e-02 -3.01665038e-01 -9.38443482e-01 2.89579928e-01 8.02319765e-01 1.21170878e+00 -5.63040972e-02 5.60425699e-01 -3.82876217e-01 1.12356615e+00 2.99779952e-01 5.47344983e-01 5.37438631e-01 -1.00767660e+00 4.91655886e-01 -4.64725234e-02 2.12337792e-01 -1.02389610e+00 -1.66501239e-01 -3.40675861e-01 -1.00367999e+00 1.41849473e-01 1.65577922e-02 -2.12609872e-01 -4.85850513e-01 1.76300514e+00 3.62347603e-01 2.09741578e-01 2.71139115e-01 1.21012795e+00 1.44236517e+00 1.20876813e+00 -3.54341596e-01 -6.78533554e-01 1.19402277e+00 -1.32338870e+00 -1.19096613e+00 6.40385002e-02 4.04538549e-02 -1.11024857e+00 1.77236247e+00 5.06465554e-01 -1.00934005e+00 -8.97719502e-01 -9.57894683e-01 1.52401784e-02 5.23388907e-02 2.14615330e-01 1.55907243e-01 5.07448852e-01 -9.58650589e-01 2.61791229e-01 -3.90794784e-01 -3.13134283e-01 6.38758093e-02 -4.21157569e-01 -3.53202522e-01 3.40192050e-01 -1.21100128e+00 5.39362967e-01 -8.57608542e-02 1.03368917e-02 -1.07510960e+00 -4.25435752e-01 -8.26194346e-01 -1.51172683e-01 3.18464160e-01 -4.51311558e-01 1.46096373e+00 -1.39773381e+00 -1.80499613e+00 8.59660625e-01 -8.14821273e-02 -1.54236272e-01 6.76156580e-01 -3.96928042e-02 -8.63230586e-01 9.81182382e-02 8.29693899e-02 9.60145175e-01 1.26329339e+00 -1.74929810e+00 -5.29976189e-01 1.38586506e-01 -1.06727801e-01 1.83009803e-01 -3.06509227e-01 4.23017979e-01 -6.13757432e-01 -1.11243510e+00 -2.21650779e-01 -8.73358548e-01 2.65717119e-01 9.51290167e-06 -7.46395171e-01 -2.60307759e-01 9.10802305e-01 -6.92097068e-01 1.27915096e+00 -2.31765914e+00 4.50852871e-01 -3.26662540e-01 1.00460425e-02 1.29718140e-01 -4.04953808e-01 4.93496895e-01 -1.75075278e-01 2.29695693e-01 -1.39132366e-01 -7.55212069e-01 7.97468275e-02 1.28707299e-02 -6.88834727e-01 2.18621522e-01 2.04410151e-01 5.78371346e-01 -8.92181218e-01 -5.77124536e-01 -8.48115534e-02 5.17618179e-01 -5.44887066e-01 5.77123046e-01 -2.83469737e-01 6.85494065e-01 -7.43127316e-02 5.90182781e-01 5.32043695e-01 2.03959510e-01 -1.52149469e-01 -2.60494530e-01 -2.55749255e-01 1.98052347e-01 -1.07691717e+00 1.63849437e+00 -4.84532773e-01 8.83004367e-01 2.55418509e-01 -6.24790907e-01 1.24894619e+00 7.52842486e-01 2.18665391e-01 -3.29805851e-01 2.44539693e-01 3.20486128e-02 -1.46099240e-01 -7.52714872e-01 5.40079355e-01 -3.94688785e-01 -3.42883021e-02 1.75850257e-01 -1.39242858e-02 -7.86674380e-01 -2.37350427e-02 3.09948549e-02 5.70366025e-01 6.02675900e-02 3.69185396e-02 9.94027182e-02 6.34252548e-01 -1.16040699e-01 6.10300064e-01 3.30478609e-01 -5.08373857e-01 9.67233658e-01 5.67700624e-01 6.74032122e-02 -1.09084451e+00 -9.40890312e-01 9.52034369e-02 9.26384568e-01 2.41935581e-01 -4.20657486e-01 -9.30340767e-01 -4.96640384e-01 -4.01880890e-01 9.18633640e-01 -4.50269371e-01 -1.16039254e-01 -3.16173524e-01 1.45218596e-02 7.89122105e-01 2.67397314e-01 4.18373376e-01 -1.57287800e+00 -2.44005486e-01 1.35967299e-01 -7.40102530e-01 -1.49789178e+00 -9.25746560e-01 -4.49928373e-01 -2.74932951e-01 -6.02795184e-01 -9.90759075e-01 -1.22264802e+00 2.92193145e-01 4.32663232e-01 1.08841062e+00 -2.04536214e-01 1.89770997e-01 3.46281707e-01 -7.67607868e-01 -2.91447848e-01 -9.43899870e-01 -3.88801754e-01 1.92199454e-01 4.16311353e-01 -2.36957550e-01 -4.99622434e-01 -2.06688643e-01 4.65169281e-01 -8.99347961e-01 5.52323818e-01 2.43586581e-02 8.44960809e-01 5.36440134e-01 -1.12655781e-01 7.18673348e-01 -3.67449135e-01 1.08543217e+00 -4.23273265e-01 -8.40886757e-02 1.39076963e-01 -1.18316472e-01 -2.81377226e-01 1.00049949e+00 -1.00835252e+00 -1.16370690e+00 -1.68587819e-01 -4.26702470e-01 -8.05797100e-01 -1.77674443e-01 3.47998321e-01 -5.40300786e-01 4.12600249e-01 4.84087735e-01 3.70074868e-01 -1.50280511e-02 -4.72238779e-01 5.86797714e-01 1.13238609e+00 1.11415482e+00 -4.90709215e-01 7.30877697e-01 2.09003076e-01 -3.65266830e-01 -1.00164974e+00 -5.51206946e-01 -2.10893020e-01 -1.29378140e-01 -6.48335040e-01 9.58661675e-01 -8.37782204e-01 -5.59697986e-01 5.42324603e-01 -1.65201974e+00 -1.32370502e-01 -5.59805185e-02 3.92715007e-01 -6.93920910e-01 3.66283506e-01 -4.58487242e-01 -8.82423103e-01 -4.04529989e-01 -1.34207475e+00 1.12844038e+00 7.15183690e-02 -3.86403441e-01 -5.76490581e-01 1.66220397e-01 4.16458130e-01 3.72414023e-01 3.09998691e-01 8.11414361e-01 -4.98991638e-01 9.20985863e-02 1.85721770e-01 6.64057210e-02 6.26591980e-01 3.02160472e-01 4.39468473e-01 -1.04201710e+00 -6.39212653e-02 6.24419004e-02 -4.91542250e-01 4.56926584e-01 3.42778265e-02 9.85047579e-01 -4.36917722e-01 2.32346699e-01 6.11625910e-01 7.13636935e-01 5.23630142e-01 6.90290213e-01 -1.67580366e-01 5.47947884e-01 8.00505280e-01 8.88361990e-01 5.73379695e-01 3.38373810e-01 9.54961419e-01 5.18005550e-01 -1.14782639e-01 -5.84630251e-01 -5.19497275e-01 6.88402057e-01 1.42288613e+00 -2.19397582e-02 -6.66232884e-01 -4.95016575e-01 5.78899145e-01 -1.50744605e+00 -1.10970283e+00 3.60956509e-03 1.82102501e+00 1.11216068e+00 7.69298598e-02 1.52467072e-01 3.19821835e-01 1.03898370e+00 4.63026315e-01 -4.96184707e-01 -6.29320383e-01 -2.09854901e-01 1.68140121e-02 -6.31800711e-01 5.20277798e-01 -1.03102100e+00 1.17314410e+00 5.12827969e+00 1.12794566e+00 -1.62856400e+00 -1.81281881e-03 4.24290508e-01 -1.22095436e-01 -5.88986933e-01 -3.14802408e-01 -4.39831227e-01 4.98402566e-01 7.12763011e-01 -3.44182760e-01 7.22779155e-01 8.92261147e-01 7.95442820e-01 6.15910888e-01 -1.07527471e+00 1.46800482e+00 3.29573423e-01 -1.19742429e+00 2.56902158e-01 -4.76025939e-01 6.83223784e-01 -4.51783389e-01 9.24656540e-02 4.68196064e-01 -1.03576876e-01 -9.15275037e-01 1.22654581e+00 2.06618682e-01 1.27245963e+00 -5.62183857e-01 3.32212716e-01 2.71339715e-01 -1.38445437e+00 3.15734655e-01 -2.31017768e-01 2.22691149e-01 3.14024895e-01 4.38348874e-02 -7.38932371e-01 4.30661052e-01 6.40589237e-01 7.28508055e-01 -9.73337367e-02 7.05934823e-01 -2.57987797e-01 7.85088658e-01 1.29713774e-01 -3.48006278e-01 2.39473820e-01 -1.08793400e-01 9.21027422e-01 1.26376331e+00 4.26838428e-01 -2.55589541e-02 -7.88164288e-02 9.74207044e-01 -3.24045718e-01 6.34281158e-01 -6.67890847e-01 -4.58041489e-01 6.96736276e-01 1.28374505e+00 -5.56470633e-01 -2.46298566e-01 -3.37474227e-01 1.24460840e+00 -9.12764966e-02 4.45106447e-01 -1.20308959e+00 -5.17476320e-01 8.32107961e-01 -2.13403314e-01 2.98863381e-01 -5.47441915e-02 3.80873494e-02 -1.21419060e+00 8.29028860e-02 -1.31020808e+00 -2.11788565e-01 -1.26998723e+00 -1.18363023e+00 1.14055002e+00 -2.97133982e-01 -1.68562901e+00 -3.32592398e-01 -1.64519072e-01 -1.05997169e+00 4.19660360e-01 -1.11069798e+00 -1.09436977e+00 -5.60556412e-01 7.26942420e-01 1.19747210e+00 -2.74346381e-01 6.51248515e-01 2.22389430e-01 -4.86176699e-01 5.79971015e-01 -2.39458770e-01 3.17793526e-02 9.37584758e-01 -8.42986941e-01 6.18696332e-01 7.46373415e-01 3.59425813e-01 7.30166882e-02 1.10937095e+00 -4.54595536e-01 -1.30238652e+00 -1.24148035e+00 9.05063570e-01 -8.96954536e-03 5.20526886e-01 -4.77213681e-01 -8.51156592e-01 2.95215100e-01 3.59032720e-01 -8.22215900e-02 4.37329292e-01 -4.91972655e-01 -2.71052897e-01 1.93103254e-02 -1.02403092e+00 7.84506500e-01 1.17871916e+00 -6.80542290e-01 -5.46764255e-01 8.21607560e-02 1.42686665e+00 -3.48860174e-01 -6.15714729e-01 2.72979558e-01 5.08234203e-01 -9.83328223e-01 7.74246395e-01 -1.00071430e+00 6.94138467e-01 -1.61235854e-01 -4.11819279e-01 -1.53656578e+00 -6.76241368e-02 -9.24275339e-01 1.65789664e-01 1.79650795e+00 3.16349357e-01 -1.44606173e-01 2.18713224e-01 2.10099339e-01 -4.68830556e-01 -6.39602959e-01 -8.95105004e-01 -8.28269303e-01 2.08889190e-02 -8.49191666e-01 7.78369963e-01 1.11495161e+00 -3.23874392e-02 5.33024907e-01 -8.30452859e-01 -4.45707934e-03 1.36031285e-01 2.25515336e-01 9.23613906e-01 -9.72278833e-01 -1.95910424e-01 -5.69297671e-01 -2.28956297e-01 -1.13303185e+00 3.67361963e-01 -7.78917968e-01 2.64454603e-01 -1.29370892e+00 -1.04504891e-01 4.82395813e-02 2.34445065e-01 1.52936518e-01 1.79334878e-04 6.20366111e-02 4.32250619e-01 8.19420740e-02 -3.47868413e-01 1.14460933e+00 1.78916156e+00 -2.35312149e-01 -2.82100409e-01 6.02170452e-02 -5.95502496e-01 6.24422073e-01 6.38962805e-01 -3.71014953e-01 -5.44447482e-01 -1.72284618e-01 2.46371683e-02 6.04944468e-01 2.00646356e-01 -7.61001766e-01 2.78727692e-02 -3.99733424e-01 -1.75857037e-01 -5.04336238e-01 6.51347399e-01 -5.41653454e-01 2.37206474e-01 7.12078512e-02 -5.75475335e-01 4.39748131e-02 -8.27295035e-02 3.26801300e-01 -6.08477771e-01 -4.74957600e-02 8.58331800e-01 1.11334637e-01 -5.98525047e-01 2.24040985e-01 -2.92559117e-01 3.57949167e-01 8.86534452e-01 -2.00005174e-02 -3.52539659e-01 -1.11541224e+00 -7.29500830e-01 6.96979314e-02 3.31330568e-01 8.97206545e-01 1.05541134e+00 -1.83213222e+00 -9.40863669e-01 2.60866322e-02 1.38864219e-01 -2.80444264e-01 5.20128496e-02 6.65393591e-01 -2.35748500e-01 1.49755731e-01 -2.63995171e-01 -3.79300892e-01 -1.52887940e+00 6.17091656e-01 2.43545681e-01 4.15004164e-01 -5.84426284e-01 8.40561807e-01 2.44771317e-01 -3.94998282e-01 6.15949810e-01 -4.45669770e-01 -4.31829840e-01 1.19846649e-01 4.90778923e-01 1.49582803e-01 -3.33313853e-01 -9.84176815e-01 -2.25407302e-01 3.73494864e-01 5.04941165e-01 -4.33219254e-01 1.11751854e+00 -2.20898718e-01 1.52530298e-01 6.92338228e-01 1.25547385e+00 3.96676958e-01 -1.08948898e+00 -4.48093675e-02 -3.96370232e-01 -3.67178649e-01 -1.19456023e-01 -6.17800474e-01 -1.25090146e+00 1.04785502e+00 3.00366789e-01 3.53512257e-01 1.10656750e+00 3.64016816e-02 1.15193260e+00 1.70239180e-01 1.41700014e-01 -9.64836836e-01 3.61928254e-01 3.24711651e-01 1.70523942e+00 -1.10693407e+00 -6.80650890e-01 -4.56939101e-01 -1.30941594e+00 1.10357010e+00 6.76014721e-01 2.92819709e-01 3.60777497e-01 7.59904906e-02 4.32609618e-01 1.65882945e-01 -1.06280887e+00 -2.97801197e-01 4.14733738e-01 7.30612457e-01 3.94512922e-01 1.48577467e-01 -1.88891217e-01 1.21075058e+00 -6.41972661e-01 -2.41898060e-01 6.65332973e-01 2.45253295e-01 -4.50937152e-01 -8.48523140e-01 -4.51791167e-01 -1.29045010e-01 -2.32462972e-01 -3.47272679e-02 -7.81076252e-01 5.82771719e-01 -3.01132482e-02 1.47943485e+00 -2.17475072e-02 -8.39131117e-01 3.88993025e-01 -2.87606120e-02 1.99667588e-01 -4.03886557e-01 -4.60088491e-01 3.88309538e-01 4.31919456e-01 -3.40402275e-01 -3.50420952e-01 -4.41544592e-01 -1.34442544e+00 -2.57020473e-01 -1.23649582e-01 1.58608675e-01 7.10763156e-01 7.25995362e-01 1.28924817e-01 5.54709136e-01 1.12469459e+00 -9.30409968e-01 -6.30581677e-01 -1.06385875e+00 -5.29507458e-01 7.01185703e-01 3.55759531e-01 -5.90042293e-01 -6.24319971e-01 2.50686288e-01]
[14.478862762451172, 5.943192005157471]
fa27c2b7-bab0-49e6-80a1-19fb192af3c0
image-harmonization-with-region-wise
2205.14058
null
https://arxiv.org/abs/2205.14058v2
https://arxiv.org/pdf/2205.14058v2.pdf
Image Harmonization with Region-wise Contrastive Learning
Image harmonization task aims at harmonizing different composite foreground regions according to specific background image. Previous methods would rather focus on improving the reconstruction ability of the generator by some internal enhancements such as attention, adaptive normalization and light adjustment, $etc.$. However, they pay less attention to discriminating the foreground and background appearance features within a restricted generator, which becomes a new challenge in image harmonization task. In this paper, we propose a novel image harmonization framework with external style fusion and region-wise contrastive learning scheme. For the external style fusion, we leverage the external background appearance from the encoder as the style reference to generate harmonized foreground in the decoder. This approach enhances the harmonization ability of the decoder by external background guidance. Moreover, for the contrastive learning scheme, we design a region-wise contrastive loss function for image harmonization task. Specifically, we first introduce a straight-forward samples generation method that selects negative samples from the output harmonized foreground region and selects positive samples from the ground-truth background region. Our method attempts to bring together corresponding positive and negative samples by maximizing the mutual information between the foreground and background styles, which desirably makes our harmonization network more robust to discriminate the foreground and background style features when harmonizing composite images. Extensive experiments on the benchmark datasets show that our method can achieve a clear improvement in harmonization quality and demonstrate the good generalization capability in real-scenario applications.
['Chi-Man Pun', 'Jingtang Liang']
2022-05-27
null
null
null
null
['image-harmonization']
['computer-vision']
[ 4.77724969e-01 -3.00263315e-01 -1.57035850e-02 -2.24989235e-01 -7.26707935e-01 -2.89758056e-01 4.50323731e-01 -4.30591434e-01 -3.12955797e-01 6.17125750e-01 9.51592475e-02 2.42734909e-01 1.96545869e-01 -9.17364955e-01 -7.60988176e-01 -1.06933188e+00 8.36001992e-01 -1.85399526e-03 1.98963046e-01 -3.74660820e-01 1.24416895e-01 2.29390174e-01 -1.36090660e+00 2.24941373e-01 1.23331642e+00 1.08186162e+00 5.88336051e-01 4.22740310e-01 -1.42679393e-01 6.13727510e-01 -7.68248498e-01 -6.09358430e-01 4.27906722e-01 -9.88353372e-01 -2.78369099e-01 3.59999329e-01 5.00537395e-01 4.39729579e-02 -2.51537412e-01 1.34876740e+00 9.67234910e-01 4.42684442e-02 3.63608629e-01 -1.29718328e+00 -7.32737899e-01 4.75118458e-01 -9.42793608e-01 1.57409951e-01 4.19226997e-02 5.57851970e-01 8.58469605e-01 -7.08075285e-01 5.23387671e-01 1.23849010e+00 4.72762376e-01 5.07442713e-01 -1.27865314e+00 -1.03606582e+00 4.88194637e-02 3.28411996e-01 -1.48933828e+00 -4.38488275e-01 1.28623712e+00 -1.77825883e-01 1.99556634e-01 5.37238002e-01 7.25495875e-01 8.39189887e-01 2.43728295e-01 1.01585019e+00 1.15789485e+00 -3.25095803e-01 -5.31412847e-02 2.59310573e-01 -5.09224236e-01 5.92435181e-01 4.65106182e-02 -8.07236657e-02 -4.46217149e-01 2.25229859e-01 7.86826432e-01 -3.46322596e-01 -6.49834454e-01 -4.03820246e-01 -1.31872463e+00 6.00399673e-01 5.79553008e-01 2.44609743e-01 -2.02847794e-01 -8.53243172e-02 2.25525320e-01 1.20286524e-01 2.66422331e-01 4.60475713e-01 -5.75115383e-02 3.77609074e-01 -9.74505663e-01 6.59211800e-02 2.03129485e-01 9.25911784e-01 8.60691488e-01 3.54099751e-01 -6.21640086e-01 1.17199683e+00 8.95653293e-02 7.96687782e-01 5.16979396e-01 -9.20830667e-01 5.57160556e-01 5.22398353e-01 -1.60389580e-02 -1.33031332e+00 1.99690219e-02 -8.83044362e-01 -1.25716138e+00 2.54395723e-01 9.25005078e-02 -2.26203948e-01 -6.14499927e-01 2.04092574e+00 2.82785237e-01 2.99352467e-01 1.08423144e-01 9.44759786e-01 8.80756974e-01 7.84279168e-01 -1.14517421e-01 -5.57101250e-01 1.14830303e+00 -1.07914782e+00 -7.95966983e-01 -1.91979975e-01 -1.90968718e-02 -1.16381848e+00 1.11072040e+00 2.18648046e-01 -1.45603526e+00 -1.05093682e+00 -1.12476873e+00 1.32323667e-01 7.42478520e-02 3.64014179e-01 1.78557426e-01 5.51994145e-01 -8.83022606e-01 1.93699837e-01 -3.27363908e-01 -2.83335987e-02 4.14610118e-01 2.99000472e-01 -8.80920812e-02 2.02537432e-01 -1.11649120e+00 7.25346029e-01 5.69538176e-01 2.43515834e-01 -5.88710368e-01 -4.76816922e-01 -8.32576871e-01 -1.35823637e-01 2.47735411e-01 -9.20979738e-01 7.57119596e-01 -1.47946668e+00 -1.66885114e+00 1.00096786e+00 -1.02906600e-01 -1.00687020e-01 6.83111489e-01 2.17910483e-01 -5.91250300e-01 -1.57691717e-01 2.05402672e-01 9.04644310e-01 9.92696702e-01 -1.70202518e+00 -9.30919945e-01 -9.22766179e-02 -2.54535109e-01 5.84540248e-01 -3.53043973e-01 -2.03157634e-01 -9.71240401e-01 -1.10260797e+00 4.86678109e-02 -7.79048145e-01 -1.10558771e-01 -1.61435932e-01 -5.09472251e-01 2.51892209e-01 8.55149090e-01 -7.62788475e-01 1.29057217e+00 -2.27923608e+00 4.17508870e-01 2.39283413e-01 5.00888377e-02 3.43859702e-01 -3.06746274e-01 -3.20994645e-01 -9.28894058e-02 -2.18745142e-01 -4.04739857e-01 -5.89565821e-02 -1.11026675e-01 5.77085800e-02 -1.87332958e-01 2.70546407e-01 3.77422035e-01 8.78261149e-01 -8.27544034e-01 -8.23934257e-01 2.39097774e-01 3.92772943e-01 -5.94868839e-01 5.59779108e-01 -1.23701565e-01 6.57432377e-01 -3.12166333e-01 5.05727828e-01 8.95322502e-01 -4.93892170e-02 1.43160475e-02 -7.16267765e-01 1.05682351e-01 -5.04709780e-01 -1.34825802e+00 1.53454292e+00 -5.19306302e-01 4.85936731e-01 2.44786665e-01 -9.25094426e-01 1.18837345e+00 -2.68278196e-02 3.00305218e-01 -1.05626953e+00 1.97630510e-01 1.50573552e-01 -6.97460817e-03 -2.75800020e-01 3.89930159e-01 -2.02857211e-01 -1.62388325e-01 1.21834435e-01 -1.73962161e-01 -4.27508920e-01 3.45036350e-02 -2.07581669e-01 4.01593000e-01 1.49111241e-01 1.93583146e-01 -8.09921771e-02 1.03259826e+00 -2.76937723e-01 9.63851571e-01 4.70119387e-01 -1.60250261e-01 8.89916420e-01 2.65145361e-01 -2.13317171e-01 -1.10646331e+00 -1.19651079e+00 -2.40093286e-04 8.98521960e-01 6.40472889e-01 -6.14491217e-02 -8.91723394e-01 -5.27831256e-01 -4.26195234e-01 5.88582218e-01 -4.40782070e-01 -4.96343017e-01 -8.74561548e-01 -1.06794643e+00 3.96394312e-01 3.60869348e-01 1.27115977e+00 -1.11231995e+00 -2.98646510e-01 8.16527754e-02 -6.47238970e-01 -8.66871655e-01 -8.40658128e-01 -6.64891601e-02 -3.42194289e-01 -6.43250763e-01 -1.02888393e+00 -1.03797805e+00 5.74972689e-01 2.28970513e-01 1.06383717e+00 -5.91187067e-02 -1.78700551e-01 -3.54187861e-02 -5.50510958e-02 -1.42213821e-01 -4.89352167e-01 -1.54867157e-01 -2.78602451e-01 4.08751696e-01 -1.41141430e-01 -3.93115759e-01 -8.23293924e-01 5.71624517e-01 -9.63507295e-01 4.29218054e-01 7.21998811e-01 1.03440857e+00 8.12272668e-01 4.06843036e-01 3.93805087e-01 -5.63851237e-01 6.87291920e-01 -1.53028339e-01 -5.13318181e-01 4.78995472e-01 -4.02238578e-01 -3.16557921e-02 7.43533254e-01 -3.98964792e-01 -1.38947499e+00 2.72252932e-02 -1.01740837e-01 -5.59171677e-01 2.01417252e-01 -3.82133760e-02 -8.53699565e-01 -1.75583899e-01 4.85633552e-01 6.28346562e-01 -6.10121079e-02 -4.24457490e-02 5.75448811e-01 5.06577075e-01 8.88898253e-01 -5.44885337e-01 9.42362845e-01 3.18449646e-01 -8.98633450e-02 -5.27273297e-01 -5.34207404e-01 -3.37854736e-02 -3.46457392e-01 -3.66061181e-01 1.06389856e+00 -9.63308454e-01 -5.69437444e-01 6.21149540e-01 -9.54114020e-01 -3.18496674e-01 -3.52355957e-01 1.76877707e-01 -7.16929793e-01 5.18133163e-01 -3.35758537e-01 -4.56657290e-01 -4.61130589e-01 -1.43546939e+00 1.04529226e+00 4.94323105e-01 3.03169936e-02 -6.63952827e-01 -1.03384696e-01 4.44944292e-01 4.86844718e-01 2.94196934e-01 9.19950187e-01 -6.46322444e-02 -5.47853887e-01 1.68148980e-01 -4.80338752e-01 5.84444344e-01 3.34443420e-01 2.29765996e-02 -7.92853713e-01 -3.16398889e-01 7.67554715e-02 3.14099863e-02 8.70902598e-01 3.22525591e-01 1.24195421e+00 -3.12637836e-01 -1.34291112e-01 9.37336981e-01 1.46028805e+00 5.07697165e-01 6.58341646e-01 2.76532352e-01 9.57726717e-01 5.96426845e-01 6.30759656e-01 4.29040253e-01 1.35197327e-01 1.03392673e+00 2.65396029e-01 -6.82653606e-01 -4.93754357e-01 -9.93131846e-02 3.73591751e-01 7.68117487e-01 4.42481972e-02 -4.01298344e-01 -5.67419052e-01 3.14489424e-01 -1.86672461e+00 -1.19993603e+00 1.91937402e-01 2.20399332e+00 1.14453721e+00 4.05885167e-02 1.14000319e-02 -6.75419495e-02 1.25306165e+00 2.58585483e-01 -6.57989383e-01 3.07586044e-03 -7.68762290e-01 1.95906475e-01 1.96209043e-01 3.26649755e-01 -1.15596282e+00 9.74278331e-01 5.38355589e+00 1.35020149e+00 -1.41500497e+00 3.60330530e-02 1.07838285e+00 -1.16974667e-01 -3.92267257e-01 -3.22715610e-01 -6.36498868e-01 8.73917639e-01 1.49783701e-01 -2.04769075e-01 5.11902273e-01 5.11472583e-01 2.91585803e-01 8.70854706e-02 -6.19065464e-01 1.29633534e+00 2.42768601e-01 -1.32511210e+00 3.99321020e-01 -3.52889746e-01 1.01688135e+00 -5.19121587e-01 3.75048369e-01 1.44119874e-01 1.11209303e-01 -8.23067963e-01 8.85338664e-01 5.98495960e-01 7.99752951e-01 -8.99681628e-01 7.24301457e-01 5.44256009e-02 -1.28704643e+00 -6.80901930e-02 -3.27257782e-01 4.06727523e-01 1.74439922e-01 3.99909019e-01 -4.31149751e-01 7.19872415e-01 5.92290401e-01 6.18523061e-01 -7.76560664e-01 9.16748047e-01 -1.03268236e-01 2.56836057e-01 1.66172665e-02 3.72038901e-01 -1.45607572e-02 -5.00599980e-01 5.70371628e-01 1.24230707e+00 3.24516743e-01 -3.13586672e-03 2.28912771e-01 9.71743286e-01 3.49478573e-02 4.10320133e-01 -3.17714214e-01 3.83457601e-01 2.79718310e-01 1.28029966e+00 -6.97614431e-01 -3.81992161e-01 -2.34090894e-01 1.17743802e+00 3.32208984e-02 3.45505327e-01 -1.13176644e+00 -2.40537226e-01 3.75074863e-01 -1.69933602e-01 3.17594230e-01 3.68226022e-01 -4.93033648e-01 -1.16710746e+00 9.96626168e-03 -1.18135488e+00 2.38164663e-01 -9.77434218e-01 -1.32129586e+00 7.52026558e-01 -1.96132764e-01 -1.50700235e+00 8.75498205e-02 -1.94572017e-01 -8.67608190e-01 9.60000873e-01 -1.29697704e+00 -1.30165517e+00 -6.39615297e-01 8.54565680e-01 5.71694970e-01 -3.26959342e-01 1.95804521e-01 3.68408054e-01 -8.72198999e-01 8.83168459e-01 8.61804038e-02 1.37663424e-01 9.49272811e-01 -1.08270240e+00 -3.35516743e-02 1.06072700e+00 -1.86013177e-01 3.65943819e-01 7.93090761e-01 -6.14129007e-01 -1.04916334e+00 -1.32366562e+00 3.37870508e-01 8.56470168e-02 2.22672984e-01 -1.30849212e-01 -7.91686416e-01 2.11506829e-01 6.27522111e-01 -8.19787532e-02 4.67431396e-01 -5.17650962e-01 2.09817430e-04 -5.97446024e-01 -1.02423203e+00 9.49636936e-01 9.36223447e-01 -2.09843799e-01 -3.64913493e-01 1.38382286e-01 6.08173132e-01 -4.19251472e-01 -6.18927121e-01 7.91894197e-01 4.34146613e-01 -1.00995147e+00 1.16776431e+00 -1.60080463e-01 6.49571657e-01 -8.21418285e-01 -2.31287792e-01 -1.22011220e+00 -5.13527513e-01 -6.01829529e-01 4.09717232e-01 1.60261071e+00 2.17797890e-01 -2.91468859e-01 5.96835136e-01 1.85788661e-01 -1.13099493e-01 -7.15697169e-01 -5.69067776e-01 -5.72432578e-01 -8.06395784e-02 -1.67775655e-03 6.37518883e-01 8.91097426e-01 -5.50821841e-01 4.22232717e-01 -5.96274734e-01 5.10483794e-02 6.50629818e-01 4.95899141e-01 1.04863656e+00 -6.76764548e-01 -5.81877053e-01 -8.01748097e-01 -1.91658735e-01 -8.57983530e-01 2.54397541e-01 -7.85374105e-01 2.20577389e-01 -1.26171935e+00 4.44206685e-01 -4.87766087e-01 -4.30156887e-01 2.27557167e-01 -6.34972930e-01 6.75023913e-01 4.28910613e-01 1.83322832e-01 -6.15680277e-01 8.69776964e-01 1.75236487e+00 -5.12575448e-01 -2.58477509e-01 -1.20690521e-02 -9.13782895e-01 6.55012906e-01 7.74284482e-01 -3.22304554e-02 -5.41550875e-01 -3.87101829e-01 -8.92409235e-02 2.43648484e-01 2.90018797e-01 -1.10146511e+00 4.38030288e-02 -3.58356088e-01 7.62438476e-01 -3.75666857e-01 2.63253570e-01 -6.14087999e-01 5.00858903e-01 4.54023629e-01 -2.71786779e-01 1.72775723e-02 1.36646688e-01 5.33638000e-01 -3.94432098e-01 2.12899502e-02 1.26882887e+00 -1.06779434e-01 -7.30662882e-01 2.84883022e-01 7.69097582e-02 1.85338199e-01 1.04628921e+00 -4.11987185e-01 -1.89134374e-01 -2.78306454e-01 -5.42124927e-01 1.03644535e-01 7.59429932e-01 3.89569253e-01 4.34673250e-01 -1.55498207e+00 -8.40118349e-01 3.96783471e-01 7.53971711e-02 -1.44507468e-01 4.63305295e-01 6.84716046e-01 -4.36823636e-01 -2.01996282e-01 -5.67966878e-01 -6.36045694e-01 -1.35936940e+00 6.76867783e-01 5.00981867e-01 -1.34842888e-01 -3.09382915e-01 8.76446068e-01 9.08603668e-01 -2.17554141e-02 4.75652367e-02 7.77487596e-03 -2.97634870e-01 4.87253182e-02 2.91288853e-01 2.19174206e-01 -1.73870713e-01 -9.84934866e-01 -1.63081095e-01 9.03606594e-01 5.92911541e-02 -3.83624695e-02 9.19317842e-01 -3.56707782e-01 -2.33560711e-01 1.61528274e-01 1.12598228e+00 3.14088911e-01 -1.34326577e+00 -3.50186497e-01 -4.86945719e-01 -5.81707001e-01 -2.01353561e-02 -7.74425566e-01 -1.55743456e+00 6.58492565e-01 8.60049844e-01 -5.21538705e-02 1.81563830e+00 -2.90122032e-01 8.77792120e-01 -1.87525183e-01 1.76744685e-02 -1.03286612e+00 3.89989287e-01 1.50138279e-03 9.52230513e-01 -1.21316159e+00 -1.95294291e-01 -4.55378324e-01 -8.33954692e-01 8.29848230e-01 9.22150373e-01 -6.94389939e-02 2.02758208e-01 1.83385506e-01 2.82796800e-01 1.01374887e-01 -2.64225513e-01 -2.00600296e-01 4.12658960e-01 5.21766961e-01 2.76450008e-01 2.17801072e-02 -4.13007945e-01 5.31560659e-01 -3.47774267e-01 -4.24381256e-01 1.89577997e-01 4.57909286e-01 -3.59295607e-01 -1.12401164e+00 -5.32971680e-01 1.40401751e-01 -3.02426040e-01 -1.11100048e-01 -2.57734835e-01 4.92744744e-01 7.11681783e-01 9.16865051e-01 2.07168516e-02 -4.31555897e-01 4.14093554e-01 -3.75283569e-01 5.10792971e-01 -2.25227728e-01 -6.80879891e-01 4.66979921e-01 -3.45707446e-01 -4.21325207e-01 -3.85854900e-01 -4.45516527e-01 -9.61478353e-01 -2.31737286e-01 -2.86359727e-01 -1.97476409e-02 1.26618221e-01 6.96355283e-01 1.83729231e-01 6.89561486e-01 9.48904932e-01 -8.51916969e-01 -1.06354669e-01 -5.08469105e-01 -4.13429886e-01 7.61635423e-01 1.32621363e-01 -5.24756014e-01 -1.96285710e-01 2.82124728e-01]
[11.21369457244873, -1.1563202142715454]
9ccd1d31-2b36-4919-a706-77874597edff
sequentialpointnet-a-strong-parallelized
2111.08492
null
https://arxiv.org/abs/2111.08492v2
https://arxiv.org/pdf/2111.08492v2.pdf
SequentialPointNet: A strong frame-level parallel point cloud sequence network for 3D action recognition
The point cloud sequence of 3D human actions consists of a set of ordered point cloud frames. Compared to static point clouds, point cloud sequences have huge data sizes proportional to the time dimension. Therefore, developing an efficient and lightweight point cloud sequence model is pivotal for 3D action recognition. In this paper, we propose a strong frame-level parallel point cloud sequence network referred to as SequentialPointNet for 3D action recognition. The key to our approach is to divide the main modeling operations into frame-level units executed in parallel, which greatly improves the efficiency of modeling point cloud sequences.Moreover, we propose to flatten the point cloud sequence into a new point data type named hyperpoint sequence that preserves the complete spatial structure of each frame. Then, a novel Hyperpoint-Mixer module is introduced to mix intra-frame spatial features and inter-frame temporal features of the hyperpoint sequence. By doing so, SequentialPointNet maximizes the appearance encoding ability and extracts sufficient motion information for effective human action recognition. Extensive experiments show that SequentialPointNet achieves up to 10X faster than existing point cloud sequence models. Additionally, our SequentialPointNet surpasses state-of-the-art approaches for human action recognition on both large-scale datasets (i.e., NTU RGB+D 60 and NTU RGB+D 120) and small-scale datasets (i.e., MSR Action3D and UTD-MHAD).
['Tianjin Yang', 'Zhenjie Hou', 'Zhijian Wang', 'Qian Huang', 'Xing Li']
2021-11-16
null
null
null
null
['3d-human-action-recognition']
['computer-vision']
[ 1.67946257e-02 -6.86368227e-01 -3.93032581e-01 3.58767994e-02 -3.07123244e-01 -2.81592399e-01 4.64083463e-01 -1.83157131e-01 -4.18597668e-01 1.41145766e-01 -1.49235517e-01 -1.99004769e-01 2.39295185e-01 -7.82504022e-01 -6.27374470e-01 -4.93599832e-01 -4.14442122e-02 3.68560284e-01 8.54739845e-01 -1.84543088e-01 2.63300478e-01 1.17940831e+00 -1.60878682e+00 2.46102735e-01 4.45426166e-01 1.22884178e+00 2.84833729e-01 8.47279847e-01 -4.58132684e-01 9.84432995e-01 -6.09397590e-01 -6.57073408e-02 5.70189416e-01 -2.57409334e-01 -5.74082136e-01 4.58197564e-01 4.44079816e-01 -8.21503282e-01 -7.63729632e-01 7.39988148e-01 4.59582567e-01 2.64700532e-01 5.66743091e-02 -1.54675388e+00 4.94105145e-02 -1.04594961e-01 -7.57243872e-01 3.53540242e-01 6.27985120e-01 5.32754004e-01 4.57296729e-01 -8.80838811e-01 7.02206552e-01 1.31826651e+00 6.55592024e-01 4.41037327e-01 -6.79369926e-01 -7.59399533e-01 2.29667142e-01 5.98984122e-01 -1.22996294e+00 -1.91289008e-01 9.80294287e-01 -2.40484461e-01 1.40582573e+00 3.37108403e-01 1.43861830e+00 1.00147009e+00 1.70299917e-01 1.21188807e+00 5.76550841e-01 -1.78307787e-01 2.48213068e-01 -9.85076308e-01 -1.83992341e-01 5.16721964e-01 -2.32254386e-01 1.26328692e-01 -7.90149033e-01 -6.65844679e-02 1.41978049e+00 6.89628124e-01 -2.01771334e-01 -6.00272357e-01 -1.62868392e+00 2.54728913e-01 3.95353317e-01 2.20293835e-01 -5.32661498e-01 5.47186494e-01 5.59327900e-01 9.78131741e-02 4.33296204e-01 -2.27593586e-01 -4.27228689e-01 -7.83164799e-01 -8.58388603e-01 2.92467773e-01 2.71115124e-01 1.29427254e+00 6.32614851e-01 8.16552341e-02 1.51986524e-01 5.06854653e-01 2.36941025e-01 9.09508646e-01 3.27608645e-01 -1.28630984e+00 8.00798774e-01 1.01006949e+00 1.57053366e-01 -1.28395069e+00 -3.82806033e-01 3.72550748e-02 -9.04258907e-01 3.97071332e-01 2.15675205e-01 3.05663526e-01 -8.38963926e-01 9.30401027e-01 6.45783424e-01 7.76576996e-01 -1.85690477e-01 1.04196179e+00 6.15746617e-01 7.24000096e-01 6.87724426e-02 8.17478746e-02 1.21704292e+00 -8.81748140e-01 -4.21329707e-01 -2.50333659e-02 7.64614403e-01 -5.52532256e-01 9.45087254e-01 1.47687614e-01 -1.23652720e+00 -8.42913270e-01 -9.89428461e-01 -2.74305552e-01 2.84269881e-02 -5.59879816e-04 5.79147696e-01 1.79534212e-01 -7.72096097e-01 5.37375569e-01 -1.34768558e+00 -1.33239746e-01 4.72083479e-01 3.16093504e-01 -5.43306708e-01 -2.54458249e-01 -8.11813831e-01 5.33380210e-01 4.06719476e-01 4.44221422e-02 -5.15577734e-01 -6.90559804e-01 -8.47200990e-01 -1.02621503e-01 3.73872161e-01 -7.79837668e-01 1.29423964e+00 -6.97891176e-01 -1.47399020e+00 7.09919214e-01 -1.67424798e-01 -4.52122331e-01 6.98471189e-01 -2.57263958e-01 -3.39786977e-01 6.69178963e-01 2.49981210e-02 7.83946276e-01 8.04976940e-01 -1.03928220e+00 -1.12731850e+00 -5.78275740e-01 2.01474294e-01 4.14911568e-01 -4.60589938e-02 6.53700754e-02 -9.38958406e-01 -5.94034493e-01 4.69392121e-01 -1.02099836e+00 -2.73323536e-01 4.24355775e-01 -3.02375127e-02 -1.89987808e-01 1.27121639e+00 -4.74560291e-01 9.10791099e-01 -2.35978818e+00 1.61062792e-01 9.25027207e-02 4.06166971e-01 4.86500651e-01 -3.04982774e-02 2.85468280e-01 -1.58554718e-01 -4.20144260e-01 6.29904121e-02 -3.75293911e-01 -7.40570650e-02 5.17284691e-01 -1.12558663e-01 5.89331567e-01 -6.44800216e-02 1.17816281e+00 -1.01494741e+00 -6.62459731e-01 9.47431922e-01 5.19813776e-01 -3.96044582e-01 -5.50553016e-02 -2.03882962e-01 3.10731411e-01 -6.92848802e-01 9.57148552e-01 7.56659329e-01 -3.90452415e-01 -2.41969928e-01 -1.58016518e-01 -2.53335804e-01 -6.04062229e-02 -1.22692490e+00 2.27907395e+00 -1.48969293e-01 3.92633557e-01 -2.34323800e-01 -7.82201350e-01 9.02174771e-01 3.20396423e-01 1.26986921e+00 -7.13273108e-01 2.32249752e-01 1.99365497e-01 -4.08969879e-01 -1.72353134e-01 5.86755037e-01 2.33316228e-01 8.20284486e-02 2.00506315e-01 -1.88583478e-01 -1.45642638e-01 -3.95423211e-02 1.29241524e-02 1.35738516e+00 3.84293586e-01 3.18940401e-01 4.28978860e-01 5.87238729e-01 3.32125455e-01 5.84486127e-01 3.62529248e-01 -5.45679927e-01 6.66112483e-01 2.33341560e-01 -7.38967776e-01 -1.23913205e+00 -8.95601630e-01 3.94221157e-01 5.23718953e-01 6.07273340e-01 -6.61530972e-01 -4.27985787e-01 -5.43674469e-01 7.78579563e-02 2.25764945e-01 -2.94261307e-01 -9.98383686e-02 -9.77773130e-01 1.95157677e-02 4.37197626e-01 9.22426581e-01 1.01238286e+00 -9.30917978e-01 -1.17986524e+00 2.60161728e-01 -2.53428251e-01 -1.38516772e+00 -6.00744724e-01 -2.29205459e-01 -1.27144623e+00 -1.11148727e+00 -8.58731389e-01 -4.84814793e-01 3.15741777e-01 8.29471171e-01 1.08688200e+00 1.88585594e-01 -7.29598105e-02 5.37810564e-01 -7.98533022e-01 -2.58560300e-01 2.66632717e-02 -2.42640957e-01 8.54470804e-02 -1.49799660e-01 8.21597278e-01 -7.27358580e-01 -7.88377404e-01 4.86568689e-01 -9.88461435e-01 3.58884156e-01 5.10880947e-01 4.08556640e-01 9.72000003e-01 -1.37549676e-02 -3.67778808e-01 1.54739663e-01 -1.08625211e-01 -8.08689557e-03 -5.04382610e-01 2.15831846e-02 2.36996710e-01 -5.20801663e-01 3.46455365e-01 -6.05624914e-01 -4.87827331e-01 5.55007577e-01 -1.41306326e-01 -1.38357913e+00 -2.34452486e-01 2.23930515e-02 -1.61463976e-01 -1.89966455e-01 2.97373205e-01 4.17419851e-01 1.96945414e-01 -3.70671242e-01 3.20612371e-01 4.48390514e-01 6.68067515e-01 -2.52584249e-01 7.86126494e-01 8.89858127e-01 1.42680824e-01 -9.35352266e-01 -3.19050133e-01 -8.32577050e-01 -1.14089167e+00 -6.51527882e-01 1.17605853e+00 -9.43102181e-01 -1.02092803e+00 9.41349864e-01 -1.41281462e+00 -3.49820256e-01 -5.58263898e-01 5.91724694e-01 -9.65409160e-01 5.83704710e-01 -6.12021744e-01 -5.38082004e-01 -1.57430202e-01 -1.05867505e+00 1.66448867e+00 -1.30539849e-01 -1.81115754e-02 -6.70015097e-01 1.14548893e-03 2.17783555e-01 -3.23187321e-01 4.42490965e-01 2.93614030e-01 -2.79653892e-02 -9.23693717e-01 -1.52988046e-01 -2.37672433e-01 1.65226236e-01 3.63592058e-02 -9.18217674e-02 -3.17030191e-01 -4.89672041e-03 1.76043436e-01 8.05713329e-03 4.35514033e-01 3.69187653e-01 1.23896897e+00 1.91189170e-01 -4.14320946e-01 8.54163229e-01 1.16340709e+00 4.65597451e-01 8.67822051e-01 5.52727401e-01 1.08008015e+00 3.77513207e-02 9.03721392e-01 7.16737568e-01 2.71137625e-01 1.04888260e+00 4.17062700e-01 -1.62976339e-01 -1.58947840e-01 -3.70206773e-01 3.34089696e-01 8.69678319e-01 -4.47197855e-01 6.40750453e-02 -1.13062024e+00 9.15740505e-02 -2.03940368e+00 -1.29997933e+00 -2.31807321e-01 1.92419517e+00 2.96018481e-01 3.77830341e-02 3.08342844e-01 5.07877529e-01 5.44872344e-01 3.47087771e-01 -7.07011759e-01 2.67612755e-01 -1.29899025e-01 5.75244687e-02 7.22956717e-01 1.38988808e-01 -1.04227495e+00 9.07798946e-01 5.49150085e+00 9.51002896e-01 -1.03076053e+00 6.35797605e-02 8.16276148e-02 -3.32499027e-01 2.43525133e-01 -1.95944130e-01 -6.74543619e-01 5.68604469e-01 5.15753806e-01 1.98807176e-02 -7.68058226e-02 1.05105555e+00 3.30242246e-01 5.26547059e-02 -9.58957374e-01 1.67873228e+00 -3.75998183e-03 -1.48124528e+00 -7.04747587e-02 2.52515405e-01 4.45992768e-01 1.64811060e-01 -3.76067668e-01 2.40796972e-02 1.54559702e-01 -5.92272460e-01 9.25035417e-01 5.06017625e-01 8.38588715e-01 -8.78137469e-01 3.90490264e-01 5.18646061e-01 -1.58050656e+00 -4.52572219e-02 -4.53676283e-01 -5.64821847e-02 5.69434524e-01 3.36870402e-01 -2.88296342e-01 6.84224546e-01 8.51012945e-01 1.43707478e+00 -4.04398829e-01 1.00251758e+00 1.74366646e-02 6.81699738e-02 -5.11713266e-01 1.01483278e-01 4.92088258e-01 -2.05597654e-01 5.56288719e-01 7.13830113e-01 5.23681402e-01 7.38089323e-01 4.59342867e-01 3.28643233e-01 3.58074248e-01 -2.22767964e-01 -5.89850903e-01 1.98225453e-01 2.83044040e-01 7.53029108e-01 -7.94650733e-01 -5.59677303e-01 -5.93926489e-01 1.31588781e+00 -6.38548061e-02 1.16321772e-01 -9.20367301e-01 -5.06140403e-02 9.99905527e-01 1.36362985e-01 5.68471909e-01 -9.91202652e-01 -1.55696779e-01 -1.34438407e+00 1.14295155e-01 -8.91315579e-01 1.88669160e-01 -1.05320811e+00 -8.84127021e-01 3.25234056e-01 1.80703074e-01 -2.13801122e+00 -2.84282595e-01 -6.07701302e-01 -1.78805947e-01 5.96649468e-01 -1.28897798e+00 -1.09871697e+00 -7.04338372e-01 1.23428893e+00 8.43046665e-01 1.07751088e-02 3.97781461e-01 4.03629959e-01 -9.52944532e-02 -3.32915387e-03 -5.22534028e-02 2.74573088e-01 1.00802451e-01 -8.87900233e-01 9.89547908e-01 7.35078692e-01 2.58778900e-01 9.28057507e-02 1.43413231e-01 -8.86973500e-01 -1.74786711e+00 -9.50225413e-01 6.83232605e-01 -6.96156740e-01 4.70736235e-01 -2.94560492e-01 -7.61972964e-01 5.49792707e-01 -4.94737267e-01 2.92345434e-01 3.35724443e-01 -5.51207304e-01 7.59089412e-03 -4.87721451e-02 -8.31820965e-01 6.87095702e-01 1.66784334e+00 -3.93895060e-01 -4.89776701e-01 4.49865013e-01 6.99443400e-01 -9.71031964e-01 -8.68279636e-01 2.88096696e-01 5.12624323e-01 -1.09314597e+00 1.43253875e+00 -3.92913789e-01 4.44603026e-01 -5.07795155e-01 -3.03111732e-01 -9.15140152e-01 -3.89174789e-01 -4.03073281e-01 -6.23811662e-01 5.37033677e-01 -5.29759109e-01 -2.73907661e-01 1.26479661e+00 3.75832170e-01 -2.65771061e-01 -7.31787860e-01 -1.19338965e+00 -9.29257452e-01 -4.31714416e-01 -8.92371118e-01 8.50025296e-01 6.79817796e-01 -3.17432016e-01 -2.47924760e-01 -2.96387821e-01 4.55632769e-02 6.81252182e-01 1.69675142e-01 1.26822042e+00 -9.67353463e-01 2.61048898e-02 -3.26248288e-01 -1.29378963e+00 -1.79529893e+00 1.03915393e-01 -4.18700278e-01 -1.87886208e-01 -1.45892835e+00 -3.22703511e-01 -4.73364532e-01 -4.45229886e-03 3.96260828e-01 -1.05883084e-01 3.09237689e-01 6.25605762e-01 6.33623421e-01 -6.07910335e-01 6.99170291e-01 1.49205863e+00 -1.73683792e-01 -1.84186324e-01 -2.41157208e-02 2.85988152e-01 7.50169575e-01 5.46896577e-01 -1.37496769e-01 -5.14773905e-01 -6.10291779e-01 -3.54490280e-01 3.75494033e-01 8.02191496e-01 -1.47541928e+00 3.33383262e-01 -3.81127506e-01 6.12372160e-01 -1.28389704e+00 8.83051574e-01 -1.23617995e+00 3.85341555e-01 5.22543371e-01 1.95139512e-01 4.07982230e-01 5.29168509e-02 4.95234787e-01 -3.11157852e-01 2.71224409e-01 4.66791153e-01 -3.51479352e-01 -1.24911690e+00 9.58157897e-01 8.69201869e-02 -3.16700488e-01 1.42919695e+00 -9.28780079e-01 8.87934342e-02 -1.06723547e-01 -4.13698584e-01 1.10097900e-01 8.40595365e-01 7.45542228e-01 1.06660235e+00 -1.76303458e+00 -4.05793786e-01 2.83509284e-01 8.05896446e-02 3.34495634e-01 5.85714340e-01 9.55746591e-01 -9.55182254e-01 5.31456769e-01 -4.66981232e-01 -1.20375800e+00 -1.61544883e+00 5.04921973e-01 2.72164375e-01 -2.97690835e-02 -1.34304190e+00 6.13911092e-01 -7.59769529e-02 -1.46986097e-01 1.46706790e-01 -6.84224784e-01 1.50988936e-01 -2.73276865e-01 7.61610150e-01 6.08681798e-01 -3.72263677e-02 -1.07571995e+00 -4.33161110e-01 8.01527500e-01 3.05611074e-01 1.12403901e-02 1.21743488e+00 2.50355061e-02 5.02077788e-02 4.76477265e-01 1.29190564e+00 -5.69103360e-01 -1.66792166e+00 -1.43983781e-01 -4.10023659e-01 -1.16256893e+00 -1.42211422e-01 3.98134887e-02 -1.06091571e+00 8.48196328e-01 5.64620316e-01 -6.49887994e-02 1.20285010e+00 -1.09149657e-01 1.28000426e+00 2.51158357e-01 8.44187021e-01 -9.30587351e-01 1.50739908e-01 5.79778790e-01 6.41948879e-01 -9.42401290e-01 1.30939066e-01 -4.58729804e-01 -4.62123156e-01 1.09406686e+00 5.80363750e-01 -1.56480923e-01 4.61842179e-01 1.53327376e-01 4.08629365e-02 -3.26868355e-01 -4.57759947e-01 -4.74407487e-02 1.40949458e-01 7.76063859e-01 -1.86084092e-01 -4.18107398e-02 -1.25555592e-02 -6.02971613e-02 -1.12169497e-01 3.70188653e-01 -4.24825493e-03 1.39852226e+00 -3.69029194e-01 -1.05862594e+00 -5.81153214e-01 1.69473067e-01 3.23474146e-02 2.92237431e-01 -2.31989667e-01 1.03114116e+00 1.18988559e-01 6.28350317e-01 2.89129972e-01 -7.29389787e-01 6.73513949e-01 -1.97188884e-01 5.16377985e-01 -3.12234521e-01 -4.22312230e-01 -3.58528607e-02 -4.00689214e-01 -1.27074671e+00 -8.78718138e-01 -8.50659549e-01 -1.41559339e+00 -5.83457053e-01 -2.94323806e-02 -2.82514483e-01 6.35262549e-01 8.38586688e-01 4.76691842e-01 3.55100423e-01 4.63034034e-01 -1.46329582e+00 -2.78604805e-01 -5.06376863e-01 -6.61161184e-01 5.43834329e-01 2.03102812e-01 -8.19066584e-01 5.85331023e-02 1.82015166e-01]
[8.189865112304688, 0.10553991794586182]
d7c07c0a-ab46-4637-bd7f-46b236a2fee4
automatic-pulmonary-nodule-detection-in-ct
1904.05956
null
https://arxiv.org/abs/1904.05956v2
https://arxiv.org/pdf/1904.05956v2.pdf
Automatic Pulmonary Nodule Detection in CT Scans Using Convolutional Neural Networks Based on Maximum Intensity Projection
Accurate pulmonary nodule detection is a crucial step in lung cancer screening. Computer-aided detection (CAD) systems are not routinely used by radiologists for pulmonary nodule detection in clinical practice despite their potential benefits. Maximum intensity projection (MIP) images improve the detection of pulmonary nodules in radiological evaluation with computed tomography (CT) scans. Inspired by the clinical methodology of radiologists, we aim to explore the feasibility of applying MIP images to improve the effectiveness of automatic lung nodule detection using convolutional neural networks (CNNs). We propose a CNN-based approach that takes MIP images of different slab thicknesses (5 mm, 10 mm, 15 mm) and 1 mm axial section slices as input. Such an approach augments the two-dimensional (2-D) CT slice images with more representative spatial information that helps discriminate nodules from vessels through their morphologies. Our proposed method achieves sensitivity of 92.67% with 1 false positive per scan and sensitivity of 94.19% with 2 false positives per scan for lung nodule detection on 888 scans in the LIDC-IDRI dataset. The use of thick MIP images helps the detection of small pulmonary nodules (3 mm-10 mm) and results in fewer false positives. Experimental results show that utilizing MIP images can increase the sensitivity and lower the number of false positives, which demonstrates the effectiveness and significance of the proposed MIP-based CNNs framework for automatic pulmonary nodule detection in CT scans. The proposed method also shows the potential that CNNs could gain benefits for nodule detection by combining the clinical procedure.
['Sunyi Zheng', 'Raymond N. J. Veldhuis', 'Peter M. A. van Ooijen', 'Matthijs Oudkerk', 'Jiapan Guo', 'Xiaonan Cui']
2019-04-11
null
null
null
null
['lung-nodule-detection']
['medical']
[ 2.63848990e-01 4.26534474e-01 -3.70407909e-01 1.05371997e-01 -6.33451641e-01 -3.63177001e-01 2.82278836e-01 -1.32623464e-01 -5.63637733e-01 3.44474167e-01 1.52113438e-01 -9.82053280e-01 -2.22714677e-01 -1.05464709e+00 -3.20101678e-01 -5.76527297e-01 -1.85148582e-01 6.31293833e-01 7.94827700e-01 3.74171376e-01 -2.41488546e-01 8.41537237e-01 -1.02358544e+00 5.59860706e-01 4.83106583e-01 8.72440338e-01 4.70004112e-01 1.05125475e+00 -6.17555007e-02 7.55946636e-01 -1.01549342e-01 6.94883522e-03 6.05144024e-01 -3.50911558e-01 -7.62180209e-01 2.29466066e-01 1.53126597e-01 -7.19161332e-01 -3.17848980e-01 6.62848294e-01 4.90103751e-01 -3.90819967e-01 7.96260536e-01 -4.78647858e-01 -3.59412283e-01 4.33489501e-01 -6.63229346e-01 6.25308871e-01 -4.05805111e-01 3.72698486e-01 5.56955755e-01 -8.47551703e-01 2.02216372e-01 6.19658947e-01 8.33011866e-01 5.21469653e-01 -5.08512259e-01 -6.15562379e-01 -8.01762640e-01 -7.04794154e-02 -1.17803800e+00 3.29469919e-01 1.09629385e-01 -5.28693736e-01 6.71843469e-01 5.68536699e-01 9.84361470e-01 3.25118631e-01 2.56751746e-01 4.81139332e-01 9.05895591e-01 -3.28604937e-01 -1.91882536e-01 1.53726622e-01 -4.06982362e-01 1.21799874e+00 8.47138643e-01 2.98014671e-01 4.67764646e-01 -1.69294789e-01 1.43311703e+00 3.86898369e-01 -2.96200573e-01 -2.79905111e-01 -1.46352935e+00 8.49460483e-01 9.49755669e-01 8.24080348e-01 -6.29433393e-01 1.99616686e-01 3.11863810e-01 -4.06985134e-01 -1.63434699e-01 4.67263639e-01 -3.27686965e-02 2.45477542e-01 -9.67389226e-01 -2.82558441e-01 4.37492281e-01 3.90965015e-01 1.27902329e-01 6.15889765e-02 -5.38226247e-01 6.91696942e-01 1.87596962e-01 4.64091808e-01 8.41825485e-01 -7.06302762e-01 1.71097666e-01 8.64041507e-01 -5.54706082e-02 -5.56441605e-01 -6.06456339e-01 -7.63687670e-01 -9.86763954e-01 3.72844897e-02 3.91230792e-01 2.77305543e-02 -1.33476377e+00 8.26624036e-01 1.83162645e-01 3.38861495e-02 -1.81265354e-01 9.53247964e-01 9.69925702e-01 1.53647229e-01 1.47411734e-01 5.85250370e-02 1.75131536e+00 -9.90695119e-01 -1.70429096e-01 9.34489667e-02 8.61404777e-01 -6.92560315e-01 9.64793801e-01 -1.30599946e-01 -9.48021472e-01 -3.93399805e-01 -1.01328170e+00 4.63893890e-01 5.44961914e-02 7.34361470e-01 5.28614998e-01 9.07074392e-01 -8.38415027e-01 2.30141163e-01 -1.18699384e+00 -4.73225683e-01 8.42290640e-01 9.00529921e-01 -2.36726612e-01 -1.02199633e-02 -7.62386560e-01 7.59358644e-01 5.32721221e-01 -7.38920420e-02 -8.08925271e-01 -8.69555652e-01 -2.37822756e-01 1.93560719e-01 5.55514276e-01 -9.67887521e-01 1.46075368e+00 -5.28572261e-01 -1.05219531e+00 8.76072407e-01 1.25572547e-01 -5.25882602e-01 6.53312802e-01 2.70615906e-01 3.98298465e-02 6.15629017e-01 6.40017018e-02 7.86839128e-01 3.94020259e-01 -7.59618521e-01 -8.35331798e-01 -2.21495777e-01 -1.17677756e-01 1.45075306e-01 -9.53830406e-02 -2.02484652e-01 -4.70237017e-01 -2.71968901e-01 2.71700561e-01 -1.06388557e+00 -5.54269433e-01 2.67748266e-01 -4.43170309e-01 1.49972394e-01 9.89696681e-01 -5.57071090e-01 8.48045945e-01 -1.60945129e+00 -6.88274145e-01 3.24816793e-01 5.81967592e-01 7.11637676e-01 3.30963045e-01 -3.47735405e-01 -7.59606138e-02 5.60941398e-01 -1.45267606e-01 5.33029258e-01 -5.87153196e-01 1.28135249e-01 7.21922398e-01 1.80502892e-01 3.13590884e-01 1.49873686e+00 -5.49432635e-01 -8.10956597e-01 5.59109747e-01 5.75784326e-01 -3.73983860e-01 1.88259352e-02 2.63994217e-01 5.67744792e-01 -6.75165176e-01 6.77279294e-01 5.85681498e-01 -8.53964984e-01 1.58516154e-01 -2.84838408e-01 -1.07850999e-01 1.24285214e-01 -8.04361761e-01 7.78569698e-01 -5.60772419e-01 5.70073724e-01 -8.26920420e-02 -4.44224417e-01 5.43020666e-01 8.43810856e-01 5.83788633e-01 -5.05273283e-01 3.32933724e-01 3.61637354e-01 8.65692258e-01 -7.36256957e-01 -2.31849819e-01 -4.33507800e-01 6.52478516e-01 5.41363895e-01 -4.09822673e-01 -1.28678739e-01 1.54354304e-01 -1.99833035e-01 1.30276918e+00 -6.13056421e-01 8.58520806e-01 -7.53019899e-02 7.03136683e-01 2.53288656e-01 5.00615574e-02 8.49509299e-01 -4.21203524e-01 7.31859744e-01 3.19334656e-01 -5.29191673e-01 -1.24206138e+00 -1.05489373e+00 -3.46987724e-01 3.50134850e-01 -4.64527965e-01 4.43136245e-01 -4.10191149e-01 -9.11742985e-01 -2.40350693e-01 -5.98896332e-02 -8.34320247e-01 3.16004276e-01 -9.18435216e-01 -9.12846029e-01 6.71622694e-01 8.05821419e-01 9.06641603e-01 -1.13816082e+00 -1.23859262e+00 6.13359176e-02 -1.73852786e-01 -8.89180541e-01 -1.12622045e-01 3.63730460e-01 -1.34843862e+00 -1.38968349e+00 -1.16054964e+00 -8.56808305e-01 8.53156209e-01 4.64805782e-01 9.70045149e-01 3.62023175e-01 -8.28077078e-01 1.06089160e-01 -1.19844630e-01 -3.57865572e-01 -5.82515299e-01 1.81504220e-01 -5.41006148e-01 -5.66297114e-01 1.55828252e-01 -2.02562571e-01 -9.22982812e-01 3.42886060e-01 -9.72624421e-01 1.40586674e-01 1.61055613e+00 7.90409029e-01 7.45154083e-01 -3.83658037e-02 3.03320348e-01 -1.28234088e+00 2.92215496e-01 -4.83614475e-01 -3.17316562e-01 2.51626819e-02 -2.21595317e-01 -1.69743866e-01 4.93355364e-01 -4.01802033e-01 -9.48910654e-01 2.01310724e-01 4.23289426e-02 -3.78691733e-01 -3.15649211e-01 4.32888031e-01 6.24128222e-01 -3.60429347e-01 9.17493105e-01 1.98961318e-01 7.27562085e-02 4.72794548e-02 -1.45021781e-01 5.06589174e-01 4.63543862e-01 2.15572417e-01 8.31201017e-01 7.06724644e-01 4.59415466e-01 -8.32738578e-01 -6.90402865e-01 -9.09006178e-01 -9.38344657e-01 -1.44354776e-01 1.05614579e+00 -7.53983080e-01 -4.18868005e-01 -3.66529107e-01 -6.37467146e-01 -2.44337581e-02 -3.49735498e-01 8.95242035e-01 -2.12370172e-01 3.54744524e-01 -6.72492683e-01 -3.95691067e-01 -7.14296520e-01 -1.19058657e+00 6.51672482e-01 3.82095337e-01 -1.21487387e-01 -8.68191481e-01 -2.54135251e-01 3.80520284e-01 9.22996640e-01 1.10340722e-01 1.12251842e+00 -7.93435037e-01 -8.77213955e-01 -5.44428587e-01 -7.20670819e-01 5.62379919e-02 4.50778455e-01 -8.70015770e-02 -7.38671362e-01 1.16936870e-01 -7.46460457e-04 1.52182475e-01 7.80465424e-01 9.36600268e-01 1.15619314e+00 -1.41763419e-01 -6.17660522e-01 5.82706809e-01 1.56001091e+00 4.71448064e-01 5.28686702e-01 3.72557878e-01 5.77296138e-01 2.01762676e-01 2.21279755e-01 1.73986197e-01 -3.98864359e-01 4.65915613e-02 8.14488530e-01 -6.09019041e-01 -5.79088092e-01 6.87922016e-02 -5.15067995e-01 4.89391983e-01 -5.07231891e-01 1.98977068e-02 -1.21789479e+00 7.37714410e-01 -1.11179316e+00 -7.67266452e-01 -5.73164344e-01 2.04862833e+00 3.19409221e-01 1.03411369e-01 6.81645945e-02 1.17539778e-01 7.42654681e-01 -3.81871611e-01 -4.35332716e-01 2.25432552e-02 5.18478572e-01 6.29114985e-01 8.81830513e-01 9.66252908e-02 -1.24584138e+00 2.80367672e-01 5.98924446e+00 4.56334859e-01 -1.45052302e+00 2.02855572e-01 7.23663509e-01 1.09630249e-01 1.75551236e-01 -5.04223585e-01 -4.44075286e-01 -3.83736715e-02 6.32516146e-01 8.54687579e-03 -3.41977268e-01 8.84859860e-01 3.45932364e-01 -3.38993728e-01 -8.74758363e-01 5.17020285e-01 -3.67453396e-01 -1.55220687e+00 1.24841280e-01 3.08403105e-01 7.21929431e-01 2.62579739e-01 3.37328583e-01 1.14111826e-01 -7.32779428e-02 -1.28303838e+00 -1.31826237e-01 8.78435373e-03 9.79446948e-01 -5.59709072e-01 1.46744633e+00 4.10950154e-01 -1.28959072e+00 5.45490012e-02 -4.54883218e-01 3.32293540e-01 -2.49166965e-01 3.08375001e-01 -2.29036331e+00 3.03520381e-01 5.98928630e-01 1.36512890e-01 -8.70102704e-01 1.47891068e+00 7.58974180e-02 8.81082892e-01 -5.02030790e-01 -3.04314435e-01 4.72228616e-01 2.19912469e-01 3.11983228e-01 1.30006802e+00 6.25844777e-01 2.75437742e-01 -9.14276112e-03 1.02320218e+00 -2.13286523e-02 2.84388721e-01 -6.33935928e-01 4.70215306e-02 3.24414968e-01 1.46602368e+00 -1.40232468e+00 -4.00826663e-01 -5.91740549e-01 3.71567130e-01 -2.78466642e-01 -1.27193615e-01 -8.08635056e-01 5.90883903e-02 -4.43530560e-01 7.29851544e-01 6.09901428e-01 2.26384804e-01 -3.59033495e-01 -4.16449577e-01 -2.43079364e-01 -4.76981252e-01 3.84277970e-01 -5.32247484e-01 -8.37768137e-01 7.61437178e-01 -1.30666539e-01 -1.43235552e+00 -2.07779005e-01 -7.89425731e-01 -8.88179362e-01 7.06218660e-01 -1.35074973e+00 -1.31244183e+00 -7.22323656e-01 3.32961231e-01 4.74144310e-01 4.63897064e-02 6.34714603e-01 6.01264499e-02 -1.94435164e-01 2.44926751e-01 -1.35972112e-01 4.43372875e-01 2.80666709e-01 -1.34589720e+00 -8.73097312e-03 5.40834129e-01 -3.13327640e-01 3.72499615e-01 4.88409437e-02 -6.18524492e-01 -8.23898017e-01 -1.46223497e+00 4.79038686e-01 -1.25149965e-01 3.14569354e-01 5.78025639e-01 -6.53341949e-01 5.91715872e-01 -1.49888203e-01 3.96359265e-01 8.00029516e-01 -6.22929394e-01 1.40665904e-01 4.60938990e-01 -1.33285809e+00 5.35672247e-01 3.24170679e-01 -1.25001520e-01 -2.60757089e-01 3.97602737e-01 2.56116152e-01 -4.28972989e-01 -7.54966736e-01 7.99701095e-01 6.14972353e-01 -1.18146253e+00 1.15477979e+00 -1.37060791e-01 5.03149569e-01 -1.59532636e-01 2.83733845e-01 -5.64805210e-01 -6.36433065e-01 4.19718146e-01 3.59566152e-01 5.52414395e-02 7.20852017e-01 -3.39096457e-01 1.48838127e+00 2.34029949e-01 -1.79948777e-01 -1.19244361e+00 -7.32102811e-01 -4.75997388e-01 1.55685753e-01 -2.90316671e-01 2.44786665e-02 4.59196508e-01 -5.39133966e-01 -1.81930587e-01 2.23551959e-01 1.21657848e-01 2.69805610e-01 -1.88431844e-01 3.35639924e-01 -1.20839560e+00 -4.82836217e-01 -5.98428488e-01 -3.92715216e-01 -6.55192971e-01 -8.63710105e-01 -1.12058604e+00 -2.64420062e-01 -1.91516709e+00 8.41549516e-01 -2.20072061e-01 -1.95135444e-01 3.86961192e-01 -1.30071178e-01 5.64645886e-01 7.02948570e-02 4.96714115e-01 -2.03611404e-01 -1.95964471e-01 2.00261426e+00 3.23527753e-02 -2.80781351e-02 5.81009984e-01 -3.56699258e-01 9.44338918e-01 9.11863327e-01 -5.99309206e-01 -1.90061420e-01 4.19171304e-02 -2.96101123e-01 4.06154007e-01 5.13522506e-01 -1.31637824e+00 1.84370503e-01 3.06331068e-02 8.63248229e-01 -8.98480237e-01 1.71385273e-01 -8.45129907e-01 1.06015004e-01 1.45790219e+00 -1.27805755e-01 -2.63820350e-01 1.88750222e-01 5.50801635e-01 -1.11907586e-01 -4.72929895e-01 1.10008204e+00 -9.07164097e-01 -2.43221566e-01 3.74573231e-01 -7.59117424e-01 -4.41382051e-01 1.21852064e+00 -6.65229380e-01 1.30641768e-02 -1.50681764e-01 -9.25373793e-01 -2.16945410e-01 1.29936170e-03 -2.58920550e-01 6.13775492e-01 -1.11002111e+00 -7.46293485e-01 9.48092714e-02 -1.11942507e-01 1.38545379e-01 1.12122655e-01 1.43987715e+00 -1.21499848e+00 1.16176784e+00 -3.20661128e-01 -9.39567804e-01 -1.56562936e+00 1.57156557e-01 9.04437363e-01 -8.53177011e-01 -6.14975572e-01 1.03503716e+00 5.61841667e-01 -3.21090847e-01 1.30102346e-02 -7.55826831e-01 -3.03787231e-01 -5.82915545e-01 2.14085609e-01 2.72022665e-01 3.01806986e-01 -2.23697305e-01 -2.06882730e-01 3.82958949e-01 -2.65240103e-01 4.84423600e-02 9.02325332e-01 3.03635597e-01 2.43670255e-01 5.33999316e-02 9.75774646e-01 -5.74352816e-02 -7.76396632e-01 -1.64416984e-01 -1.05636390e-02 -3.98409784e-01 1.40017033e-01 -1.08545280e+00 -1.15073752e+00 9.86166239e-01 9.81644034e-01 1.03163853e-01 8.27306032e-01 2.35862479e-01 6.60461426e-01 3.65538627e-01 -1.23067334e-01 -2.40210533e-01 4.49267328e-01 2.04319149e-01 4.79479283e-01 -1.49049711e+00 7.44844675e-02 -8.24088752e-01 -5.17534852e-01 1.58079112e+00 8.09870183e-01 -1.16308078e-01 5.44006050e-01 3.09462816e-01 8.99371728e-02 -5.32762170e-01 -5.67259669e-01 -4.48432684e-01 6.35411978e-01 5.95785081e-01 9.26176190e-01 4.54462826e-01 -2.28156775e-01 3.88543546e-01 -6.32164115e-03 6.64971918e-02 7.02647686e-01 1.06375360e+00 -9.59104717e-01 -6.89729571e-01 -5.03328323e-01 1.10565090e+00 -7.93024004e-01 -1.54184652e-02 -4.79586929e-01 1.44547868e+00 2.78470397e-01 4.10087436e-01 4.73343469e-02 8.15280527e-02 1.55903384e-01 -5.11505492e-02 3.55470449e-01 -1.01591933e+00 -8.46801937e-01 3.45409900e-01 -1.33043945e-01 -8.99796337e-02 -3.16094011e-01 -3.00663292e-01 -1.42415094e+00 1.27308127e-02 -5.12650907e-01 -5.10834251e-03 4.97893572e-01 7.62979150e-01 -1.23635530e-02 9.72359061e-01 2.22209439e-01 -3.36672276e-01 -4.33113784e-01 -1.04160416e+00 -3.16170096e-01 -5.09868860e-02 1.33938998e-01 -3.73833746e-01 -2.47630835e-01 8.22747722e-02]
[15.359626770019531, -2.1555259227752686]
dd14a0a2-8aee-4d8d-bc04-6ab9b26e5dba
modeling-hierarchical-syntax-structure-with-1
null
null
https://aclanthology.org/2022.acl-long.37
https://aclanthology.org/2022.acl-long.37.pdf
Modeling Hierarchical Syntax Structure with Triplet Position for Source Code Summarization
Automatic code summarization, which aims to describe the source code in natural language, has become an essential task in software maintenance. Our fellow researchers have attempted to achieve such a purpose through various machine learning-based approaches. One key challenge keeping these approaches from being practical lies in the lacking of retaining the semantic structure of source code, which has unfortunately been overlooked by the state-of-the-art. Existing approaches resort to representing the syntax structure of code by modeling the Abstract Syntax Trees (ASTs). However, the hierarchical structures of ASTs have not been well explored. In this paper, we propose CODESCRIBE to model the hierarchical syntax structure of code by introducing a novel triplet position for code summarization. Specifically, CODESCRIBE leverages the graph neural network and Transformer to preserve the structural and sequential information of code, respectively. In addition, we propose a pointer-generator network that pays attention to both the structure and sequential tokens of code for a better summary generation. Experiments on two real-world datasets in Java and Python demonstrate the effectiveness of our proposed approach when compared with several state-of-the-art baselines.
['Pingyi Zhou', 'Li Li', 'Yao Wan', 'Jin Liu', 'Juncai Guo']
null
null
null
null
acl-2022-5
['code-summarization']
['computer-code']
[ 2.60583252e-01 1.98542252e-01 -3.77891272e-01 -2.02547684e-01 -5.96110582e-01 -4.47963953e-01 4.09542859e-01 5.09163380e-01 3.54843408e-01 2.37197146e-01 5.94320774e-01 -5.23697078e-01 2.36988097e-01 -5.54151475e-01 -6.82252705e-01 -1.83021545e-01 -6.08050898e-02 -4.29057717e-01 4.56859469e-01 -2.09582657e-01 8.02139163e-01 -2.15284660e-01 -1.64038992e+00 4.28269088e-01 1.35666239e+00 5.82871616e-01 3.20477635e-01 2.46977940e-01 -8.18114281e-01 1.16304696e+00 -5.84580183e-01 -6.03991032e-01 -1.36891633e-01 -5.49779654e-01 -8.89422536e-01 -2.21291110e-02 4.31536257e-01 -3.14078256e-02 -2.23736748e-01 1.37264764e+00 1.53083101e-01 -3.85200500e-01 3.28545094e-01 -1.27797687e+00 -6.77740216e-01 1.04672408e+00 -9.66694832e-01 -7.65504539e-02 4.10770983e-01 7.50200078e-02 1.29089093e+00 -5.95680296e-01 4.31350678e-01 1.03661764e+00 7.05491662e-01 4.65234697e-01 -1.04314566e+00 -3.75081688e-01 1.26943052e-01 8.24391097e-02 -1.14233506e+00 -3.84453803e-01 1.05043495e+00 -5.89934528e-01 1.09103990e+00 1.13313869e-01 3.91912580e-01 6.54160738e-01 5.50566673e-01 7.31156707e-01 6.33333266e-01 -4.27719772e-01 5.69832921e-02 6.34727404e-02 4.60131705e-01 1.11135411e+00 3.76012444e-01 -4.79310960e-01 -2.90205449e-01 -3.81996095e-01 1.65718317e-01 7.20214918e-02 -2.05972821e-01 -6.47289097e-01 -1.07521200e+00 7.04655647e-01 4.26543802e-01 2.82976180e-01 -1.37046188e-01 2.90595055e-01 9.17263806e-01 6.66353703e-02 1.99669883e-01 3.75107884e-01 -2.50976801e-01 -2.97484785e-01 -1.04488420e+00 2.13634655e-01 7.71744788e-01 1.31227541e+00 8.19418609e-01 1.28490746e-01 -1.36271715e-01 6.84523761e-01 4.34807241e-01 4.42324728e-02 5.65084696e-01 -7.85856247e-01 8.09517443e-01 1.48509490e+00 -2.72138268e-01 -1.19921327e+00 -9.26491395e-02 -5.49564064e-01 -8.20412159e-01 -1.41436905e-01 -1.59768417e-01 3.08327526e-01 -4.42907125e-01 1.47149444e+00 4.31944616e-02 -8.66268110e-03 1.87572073e-02 2.38502994e-01 8.47314715e-01 6.49285555e-01 -2.52626449e-01 -1.20330364e-01 1.27746665e+00 -1.33395076e+00 -5.57743609e-01 -2.81084418e-01 9.01071846e-01 -6.83368206e-01 1.16642439e+00 9.87600684e-02 -1.04220796e+00 -5.30107677e-01 -1.10252333e+00 -1.69459835e-01 -2.79659890e-02 2.43744522e-01 6.57713413e-01 6.62044227e-01 -1.05725992e+00 4.85663563e-01 -9.99665320e-01 -3.49854916e-01 4.29238230e-01 5.23960255e-02 -2.06945941e-01 1.94927007e-01 -5.77029526e-01 4.10784513e-01 5.91067374e-01 -3.45971547e-02 -5.42890549e-01 -6.74064279e-01 -1.05613899e+00 4.72078979e-01 5.77592731e-01 -7.55623460e-01 1.47742128e+00 -7.62324929e-01 -1.25015795e+00 4.56527978e-01 -3.71424079e-01 -3.18609565e-01 -8.32292661e-02 -4.29506414e-02 -1.32824108e-01 -1.48915946e-01 1.87076762e-01 1.20698735e-01 5.80407441e-01 -1.43578994e+00 -6.68168306e-01 -2.73723394e-01 2.51353472e-01 -5.91161922e-02 -6.34833574e-01 2.42546842e-01 -5.18462896e-01 -7.30067074e-01 -2.80861650e-02 -7.51936197e-01 -9.32409540e-02 -2.26781204e-01 -5.98363042e-01 -3.98398399e-01 7.58372545e-01 -8.78726184e-01 2.12115622e+00 -2.28479648e+00 3.16102743e-01 -1.26965746e-01 4.51441169e-01 2.13062420e-01 -6.61959127e-02 8.98328006e-01 1.10318348e-01 3.33569914e-01 -7.09687889e-01 -3.94862086e-01 2.07414478e-01 2.90942769e-02 -5.73857665e-01 1.72099248e-01 1.65557861e-01 8.44241858e-01 -9.93697464e-01 -8.00420523e-01 -1.96058854e-01 4.47130576e-02 -1.13766348e+00 3.25186700e-01 -5.01852393e-01 8.69708210e-02 -6.00935698e-01 6.65459991e-01 6.33041322e-01 -3.66027236e-01 4.00753260e-01 -9.82296839e-02 -3.43529969e-01 5.91023862e-01 -8.51499557e-01 2.11766267e+00 -4.57976401e-01 1.73342749e-01 -1.34935811e-01 -1.01203370e+00 8.75179410e-01 6.67974874e-02 3.07436228e-01 -5.83617806e-01 -2.14683130e-01 2.70325273e-01 9.12048505e-04 -7.24869192e-01 7.72335172e-01 2.79056132e-01 -4.47811037e-01 5.86096287e-01 -4.04629588e-01 -1.39459759e-01 3.75629961e-01 7.11710155e-01 1.35586941e+00 5.71252286e-01 5.13778210e-01 -2.63575822e-01 7.22656369e-01 9.42163914e-02 6.52510643e-01 6.29417241e-01 1.93418950e-01 4.30722505e-01 9.58666921e-01 -1.90752417e-01 -9.61323321e-01 -6.90518677e-01 2.47303680e-01 9.17431891e-01 6.81705251e-02 -1.15391922e+00 -1.03369617e+00 -9.99908209e-01 -2.25263730e-01 7.20868587e-01 -4.90627229e-01 -3.55912209e-01 -8.92005563e-01 -5.10923564e-01 6.74377978e-01 4.61109042e-01 4.82328206e-01 -8.68344843e-01 -8.00268292e-01 2.39291564e-01 -3.95355135e-01 -7.76689351e-01 -8.28613162e-01 -1.39757782e-01 -9.14963603e-01 -1.05408621e+00 -2.77709246e-01 -8.62814486e-01 7.39090204e-01 4.10451174e-01 1.08024383e+00 7.09175289e-01 -5.92790954e-02 -1.38311600e-02 -6.87647820e-01 -5.78680038e-02 -8.85885775e-01 5.04693449e-01 -5.90895534e-01 -2.37699598e-01 1.08270288e-01 -7.65268087e-01 -4.13517088e-01 -1.28616542e-02 -1.22758281e+00 3.63013268e-01 7.18844414e-01 6.26532555e-01 9.95650068e-02 1.82450384e-01 3.23668361e-01 -1.18468785e+00 5.46991289e-01 -6.70557678e-01 -5.97640574e-01 4.54525888e-01 -5.81673443e-01 5.38240850e-01 9.33073342e-01 6.08605333e-02 -1.16794276e+00 -6.77270070e-02 -1.32581070e-01 6.68816641e-02 2.30694607e-01 1.00966728e+00 -2.36335799e-01 1.39453784e-01 3.51184011e-01 5.87501109e-01 3.66231659e-04 -5.77712834e-01 2.55997390e-01 8.71412575e-01 4.55942601e-01 -9.17239904e-01 8.38046372e-01 2.50014484e-01 -1.58829659e-01 -5.22434950e-01 -6.14417851e-01 -3.52414519e-01 -4.72020179e-01 3.00837427e-01 5.10719359e-01 -6.52481318e-01 -3.28798145e-01 4.70684826e-01 -1.48700166e+00 8.40886906e-02 5.92014566e-02 -2.75818020e-01 -4.31389362e-01 1.06768882e+00 -4.42848086e-01 -4.65566099e-01 -4.92126793e-01 -1.52449346e+00 1.23373747e+00 2.36565806e-02 -2.06647113e-01 -7.01517403e-01 2.30565786e-01 2.22479731e-01 6.78169966e-01 1.78594500e-01 1.53664362e+00 -3.62739444e-01 -7.89623320e-01 4.42417301e-02 -3.66230279e-01 1.99333593e-01 3.84388059e-01 3.66839230e-01 -4.59263384e-01 -2.53645986e-01 -4.72950153e-02 -6.90103248e-02 7.07990706e-01 -1.19262561e-01 1.29412174e+00 -6.96578741e-01 -4.67531055e-01 6.16955757e-01 1.45688951e+00 1.97833985e-01 6.70494020e-01 3.83719206e-01 8.74691665e-01 6.75264180e-01 1.99131250e-01 6.29305243e-01 9.05564725e-01 6.75520241e-01 7.58949876e-01 2.85847902e-01 -2.42058396e-01 -5.26033938e-01 4.68371570e-01 1.46080554e+00 4.24447030e-01 -3.55880223e-02 -1.05386472e+00 7.73206890e-01 -1.90591705e+00 -8.84930313e-01 -4.65593904e-01 1.95073760e+00 9.61429656e-01 8.69000554e-02 -2.17430703e-02 7.14049935e-02 6.68618381e-01 2.21443862e-01 -3.28652233e-01 -4.06520814e-01 2.37599179e-01 -1.52743071e-01 1.70067534e-01 4.13001850e-02 -7.98200607e-01 7.08305717e-01 5.12003899e+00 5.91844857e-01 -8.90236080e-01 -3.26372571e-02 -9.48896185e-02 4.59307909e-01 -5.81914544e-01 6.90343559e-01 -6.85555220e-01 6.10353172e-01 8.42644811e-01 -6.78029180e-01 4.01018232e-01 9.42149162e-01 9.81279239e-02 1.18650300e-02 -1.14201605e+00 5.73620319e-01 2.04631835e-01 -1.32316911e+00 3.06869447e-01 -1.68981537e-01 7.89112449e-01 -2.24204049e-01 -2.27050513e-01 5.60543954e-01 2.16985956e-01 -5.66783369e-01 9.94046271e-01 3.65212351e-01 5.04826665e-01 -4.77069557e-01 6.47655189e-01 4.46985155e-01 -1.51030707e+00 -1.50895491e-01 -2.50140041e-01 3.27831320e-02 -8.81171152e-02 5.20003140e-01 -6.06468797e-01 9.94647622e-01 5.80535114e-01 9.89657462e-01 -1.09744811e+00 1.19294989e+00 -1.06983095e-01 5.47111750e-01 1.85121715e-01 -7.91990682e-02 2.22468331e-01 1.69838276e-02 3.74284416e-01 1.44201398e+00 4.85306770e-01 -3.43773007e-01 2.96984464e-01 1.15839696e+00 -2.69188285e-01 2.21736044e-01 -6.32824659e-01 -3.82385522e-01 4.98628706e-01 1.08181429e+00 -6.00940049e-01 -3.13145250e-01 -7.97474861e-01 6.84756577e-01 3.42554837e-01 7.23765939e-02 -8.34007621e-01 -8.59451175e-01 2.29639068e-01 2.02027723e-01 3.80731702e-01 -2.29922488e-01 -3.23915422e-01 -1.45062363e+00 4.85545188e-01 -1.16974568e+00 2.93940455e-01 -6.09969437e-01 -8.79318357e-01 6.87449932e-01 1.49755120e-01 -1.24991941e+00 -1.97364762e-02 -6.22445494e-02 -8.32434297e-01 6.10507548e-01 -1.52971721e+00 -1.19806898e+00 -3.34359020e-01 1.24886483e-01 6.71561778e-01 -1.10147260e-01 5.72917998e-01 3.23832244e-01 -7.51552463e-01 5.55157065e-01 3.89764607e-02 1.53252512e-01 4.44939137e-01 -1.28544617e+00 9.11016166e-01 1.25495982e+00 -1.80268943e-01 1.45506167e+00 6.49633646e-01 -6.99470878e-01 -1.77878606e+00 -1.15839052e+00 1.01786685e+00 -1.87863812e-01 9.11291778e-01 -3.23598176e-01 -1.23767567e+00 7.59352207e-01 5.56446791e-01 -2.52636880e-01 4.49718833e-01 -2.41347596e-01 -6.28461957e-01 -1.18713178e-01 -5.70081890e-01 6.04818523e-01 1.03046644e+00 -5.19563675e-01 -9.16861236e-01 3.25705968e-02 9.11872685e-01 -3.59894991e-01 -5.66139638e-01 2.07948819e-01 3.83257985e-01 -1.14149475e+00 5.48729777e-01 -3.62081259e-01 1.06468630e+00 -5.93191624e-01 -2.99772806e-02 -1.23807573e+00 -3.92869301e-02 -7.09522605e-01 -2.20788598e-01 1.68879497e+00 1.40275657e-01 -4.82938349e-01 6.20608628e-01 1.24802351e-01 -6.36212051e-01 -6.36877716e-01 -5.37601709e-01 -6.75201893e-01 1.28953354e-04 -1.81589916e-01 7.97741711e-01 8.32777977e-01 4.54278409e-01 4.73513037e-01 -2.68487036e-01 -1.21992780e-03 6.08796597e-01 5.27469099e-01 1.00075948e+00 -1.19770849e+00 -3.44158202e-01 -6.62074566e-01 -2.04388991e-01 -1.10369599e+00 5.16760349e-01 -1.23775041e+00 4.00704518e-02 -1.91879976e+00 6.65065527e-01 -2.64484793e-01 3.74058336e-02 5.88113248e-01 -3.00149590e-01 -4.86417353e-01 4.41046841e-02 3.99493128e-01 -6.75411105e-01 7.19583690e-01 7.69252062e-01 -3.78066093e-01 -3.26584056e-02 -9.87430289e-02 -1.05438101e+00 6.13117516e-01 7.55392194e-01 -7.88807929e-01 -5.18731236e-01 -8.15161526e-01 4.34257030e-01 3.12114745e-01 1.49029061e-01 -9.17174518e-01 4.98510897e-01 -1.45978212e-01 -6.98543489e-01 -6.01613343e-01 -2.95198381e-01 -7.51619279e-01 1.92493841e-01 5.57792187e-01 -3.27719659e-01 5.07790267e-01 1.75304696e-01 4.58762407e-01 -4.23398018e-01 -6.43584967e-01 5.83938956e-01 -2.96002656e-01 -7.03104675e-01 8.29843953e-02 -1.32435635e-01 2.83636123e-01 7.89128482e-01 -1.79589480e-01 -6.95492804e-01 1.55582920e-01 2.11437061e-01 1.92347929e-01 9.02781904e-01 6.28329873e-01 6.20338082e-01 -1.11114347e+00 -5.73255360e-01 2.24566117e-01 5.82195759e-01 -4.70449999e-02 7.48147890e-02 7.62327850e-01 -6.98068798e-01 4.30648863e-01 -1.22416154e-01 -3.76291424e-01 -1.26610875e+00 7.04012275e-01 5.29599153e-02 -4.88833547e-01 -6.87825620e-01 2.39020377e-01 2.29654625e-01 -4.86502707e-01 1.70314744e-01 -5.46904504e-01 -1.53258443e-01 -3.35731983e-01 4.17047650e-01 2.40824863e-01 1.31842211e-01 -4.55906540e-01 -4.18745458e-01 5.68681121e-01 -3.47210467e-01 4.48354244e-01 1.37466753e+00 -1.43611148e-01 -8.19611132e-01 2.21005917e-01 1.32740617e+00 3.26742828e-01 -8.36318254e-01 -3.11311334e-01 6.33537173e-01 -4.13551807e-01 -1.52250037e-01 -4.38956439e-01 -9.44631040e-01 9.65743065e-01 -5.11817262e-02 3.27803433e-01 9.45976794e-01 -1.59293041e-01 9.71847653e-01 3.99108499e-01 5.47043920e-01 -5.79995036e-01 1.84941098e-01 5.17476082e-01 6.58828557e-01 -9.18849885e-01 -2.48781387e-02 -5.09225070e-01 -4.19312865e-01 1.19519150e+00 6.00939810e-01 1.81382880e-01 1.74128368e-01 3.38064224e-01 -3.60770971e-01 -2.58498877e-01 -7.91373849e-01 1.31489038e-01 1.37484938e-01 2.56945044e-01 6.60585880e-01 -3.76127481e-01 -5.15768707e-01 6.29075110e-01 2.82406937e-02 1.12199031e-01 1.05268776e+00 1.49642563e+00 -5.01484871e-01 -1.44894898e+00 -9.21890065e-02 4.88904208e-01 -6.55568540e-01 -4.85903800e-01 -2.87462980e-01 5.76837003e-01 -1.33800358e-01 7.99436271e-01 -3.47194791e-01 -3.06345910e-01 4.92097169e-01 -2.51797363e-02 1.93044752e-01 -1.14724779e+00 -8.44007850e-01 -2.24871948e-01 -2.15150103e-01 -4.33538288e-01 -4.42316085e-01 -5.77647090e-01 -1.35863459e+00 -2.13493943e-01 -2.45524749e-01 4.32610750e-01 5.41260421e-01 5.14605582e-01 6.76103473e-01 7.47476339e-01 5.90690970e-01 -3.74714285e-01 -7.17172503e-01 -6.90773129e-01 -2.78560668e-01 4.03298289e-01 3.53135616e-01 -3.90762717e-01 -3.21458250e-01 4.04119283e-01]
[7.614256381988525, 7.937649726867676]
2917cbf6-b8a5-4a87-ad06-842d8d108535
edict-exact-diffusion-inversion-via-coupled
2211.12446
null
https://arxiv.org/abs/2211.12446v2
https://arxiv.org/pdf/2211.12446v2.pdf
EDICT: Exact Diffusion Inversion via Coupled Transformations
Finding an initial noise vector that produces an input image when fed into the diffusion process (known as inversion) is an important problem in denoising diffusion models (DDMs), with applications for real image editing. The state-of-the-art approach for real image editing with inversion uses denoising diffusion implicit models (DDIMs) to deterministically noise the image to the intermediate state along the path that the denoising would follow given the original conditioning. However, DDIM inversion for real images is unstable as it relies on local linearization assumptions, which result in the propagation of errors, leading to incorrect image reconstruction and loss of content. To alleviate these problems, we propose Exact Diffusion Inversion via Coupled Transformations (EDICT), an inversion method that draws inspiration from affine coupling layers. EDICT enables mathematically exact inversion of real and model-generated images by maintaining two coupled noise vectors which are used to invert each other in an alternating fashion. Using Stable Diffusion, a state-of-the-art latent diffusion model, we demonstrate that EDICT successfully reconstructs real images with high fidelity. On complex image datasets like MS-COCO, EDICT reconstruction significantly outperforms DDIM, improving the mean square error of reconstruction by a factor of two. Using noise vectors inverted from real images, EDICT enables a wide range of image edits--from local and global semantic edits to image stylization--while maintaining fidelity to the original image structure. EDICT requires no model training/finetuning, prompt tuning, or extra data and can be combined with any pretrained DDM. Code is available at https://github.com/salesforce/EDICT.
['Nikhil Naik', 'Akash Gokul', 'Bram Wallace']
2022-11-22
null
http://openaccess.thecvf.com//content/CVPR2023/html/Wallace_EDICT_Exact_Diffusion_Inversion_via_Coupled_Transformations_CVPR_2023_paper.html
http://openaccess.thecvf.com//content/CVPR2023/papers/Wallace_EDICT_Exact_Diffusion_Inversion_via_Coupled_Transformations_CVPR_2023_paper.pdf
cvpr-2023-1
['text-based-image-editing', 'text-guided-image-editing', 'image-stylization']
['computer-vision', 'computer-vision', 'computer-vision']
[ 5.84183931e-01 1.37852058e-01 2.24226952e-01 -1.04317747e-01 -5.39303958e-01 -6.51624084e-01 8.74860764e-01 -4.02599007e-01 -3.73375207e-01 3.55434716e-01 1.51052266e-01 -2.59033442e-01 7.13332891e-02 -7.00485885e-01 -9.35525358e-01 -6.10669971e-01 4.82942581e-01 3.15771103e-01 -7.80859217e-03 -2.90747106e-01 2.00690642e-01 4.29341108e-01 -9.56916928e-01 2.49546215e-01 1.01316619e+00 5.83464503e-01 4.24135029e-01 8.66834998e-01 -2.53271043e-01 9.28462088e-01 -5.32305002e-01 -6.17896318e-01 4.08965826e-01 -7.24041939e-01 -6.39477909e-01 1.17253289e-01 5.41378736e-01 -4.96623307e-01 -5.05255520e-01 1.29313636e+00 3.82608533e-01 -7.49554336e-02 7.52153635e-01 -1.10513663e+00 -1.32507765e+00 4.96253818e-01 -7.17035651e-01 -1.04619563e-01 8.40622708e-02 4.83226448e-01 4.78579402e-01 -8.93198073e-01 1.13883221e+00 1.26643610e+00 8.19899321e-01 7.19700456e-01 -1.82779062e+00 -6.52286112e-01 -1.60153806e-01 -1.73687980e-01 -1.29644644e+00 -6.09835804e-01 7.19594955e-01 -6.17862105e-01 7.14135349e-01 1.99912593e-01 5.67091525e-01 1.23834324e+00 2.29264081e-01 6.18297338e-01 1.32859421e+00 -3.30386788e-01 2.87262001e-03 9.06075910e-02 -3.38161349e-01 6.82862341e-01 -1.41796516e-02 2.74768233e-01 -3.21771890e-01 1.09761529e-01 1.18822181e+00 -2.37388667e-02 -3.05690527e-01 -2.33681351e-01 -1.28215694e+00 7.00546384e-01 5.17565310e-01 2.14239359e-01 -5.45495689e-01 5.01753688e-01 1.34177446e-01 6.67525589e-01 4.94505763e-01 4.30569619e-01 1.14502329e-02 -9.03210640e-02 -1.21582830e+00 2.22687900e-01 5.89941084e-01 8.15650702e-01 8.75556529e-01 1.73482373e-01 -1.28964081e-01 7.86086202e-01 2.73127705e-02 7.14761555e-01 2.23576546e-01 -1.51892328e+00 2.85395771e-01 2.94018298e-01 -6.91363439e-02 -1.16024756e+00 2.03722239e-01 -4.08894271e-01 -1.17377019e+00 4.83768612e-01 2.69131452e-01 -3.57294977e-02 -1.22060943e+00 1.85015976e+00 1.21143162e-01 2.58751780e-01 -9.48870257e-02 7.92846680e-01 3.61960709e-01 8.32514822e-01 -1.42822623e-01 -1.38031900e-01 8.58380735e-01 -9.47174251e-01 -8.78351688e-01 -2.70688802e-01 3.83690655e-01 -9.98357117e-01 1.24500787e+00 5.95231354e-01 -1.48404801e+00 -3.69161695e-01 -9.12178874e-01 -4.39870089e-01 -8.24125707e-02 -1.63296789e-01 3.57085943e-01 3.65098387e-01 -1.30612326e+00 7.25531161e-01 -9.68704760e-01 -1.66926101e-01 5.56035459e-01 2.56430060e-01 -4.68082309e-01 -2.54778117e-01 -8.96817803e-01 8.68708134e-01 -2.02691093e-01 1.71531364e-01 -1.14287627e+00 -1.02102137e+00 -7.96111166e-01 -2.47557595e-01 3.25587064e-01 -9.20989871e-01 1.03798568e+00 -1.43366647e+00 -1.71043158e+00 8.88782859e-01 -2.59078711e-01 -4.06993270e-01 1.07237005e+00 -1.45562083e-01 -2.16539592e-01 1.28262505e-01 1.65208146e-01 9.08529162e-01 1.45748317e+00 -1.57697511e+00 5.05257323e-02 -3.37896054e-03 -1.05100900e-01 -3.79350781e-02 -1.35607451e-01 -1.22623757e-01 -7.56036699e-01 -1.03118694e+00 4.89634164e-02 -1.11248922e+00 -2.64242738e-01 4.94664043e-01 -4.11631852e-01 4.93069977e-01 1.01871300e+00 -9.63195622e-01 1.09943020e+00 -2.12402487e+00 6.99502647e-01 3.31235975e-01 4.26088423e-01 3.30531955e-01 -4.90608066e-01 3.22010726e-01 -1.02481641e-01 2.58529812e-01 -7.16839552e-01 -7.86525548e-01 -1.13613062e-01 5.23590982e-01 -4.68956947e-01 3.34387094e-01 1.03183873e-01 1.21166039e+00 -8.75741541e-01 -2.12466851e-01 2.48524949e-01 9.15039182e-01 -6.99922144e-01 2.07756862e-01 -1.75326750e-01 7.72586644e-01 1.45582542e-01 1.04594916e-01 8.79552245e-01 -1.53743982e-01 2.39456333e-02 -4.62214530e-01 -3.95212807e-02 -9.75497663e-02 -1.17313218e+00 1.99664056e+00 -6.59968615e-01 7.18649685e-01 4.26224083e-01 -7.46071756e-01 5.04368722e-01 8.26193094e-02 2.30887547e-01 -7.95315266e-01 -6.19385466e-02 2.52128273e-01 -2.58884370e-01 -2.44592369e-01 5.18328547e-01 -2.19678074e-01 3.20865452e-01 7.59133041e-01 1.33786067e-01 -5.58516145e-01 2.72052735e-01 5.88102460e-01 9.49682117e-01 1.78199679e-01 -3.35926354e-01 -1.37435794e-01 2.42235720e-01 -1.79722324e-01 4.05439168e-01 7.97700584e-01 2.77341843e-01 1.02721107e+00 4.94487971e-01 -1.68162331e-01 -1.31530881e+00 -1.19105339e+00 1.88612312e-01 4.16106045e-01 4.98992316e-02 -3.36163044e-01 -1.27558672e+00 -3.97237211e-01 -1.94876045e-01 8.57842982e-01 -7.22383738e-01 -1.77105084e-01 -6.24005318e-01 -4.54827845e-01 7.40020275e-01 1.75948143e-01 6.25075221e-01 -8.43943417e-01 -1.27939194e-01 1.76763520e-01 -2.68235147e-01 -9.23608720e-01 -9.82001662e-01 -8.75295177e-02 -8.39571774e-01 -7.04219520e-01 -1.05381250e+00 -5.28569400e-01 1.00497472e+00 2.49060541e-01 9.92345572e-01 3.16318154e-01 -2.72426307e-01 3.44011515e-01 1.87663689e-01 1.58192907e-02 -1.00211334e+00 -1.61278442e-01 -9.97345820e-02 1.61658242e-01 -3.59447718e-01 -8.11467290e-01 -7.28826106e-01 3.61555338e-01 -1.48414898e+00 3.76015633e-01 4.30150419e-01 9.34547186e-01 6.95971310e-01 -1.71207581e-02 6.54944554e-02 -1.16517365e+00 7.87735879e-01 -1.96468934e-01 -6.22365236e-01 1.12679258e-01 -8.05180311e-01 3.74091208e-01 4.93695855e-01 -7.41281390e-01 -1.16653621e+00 -1.17979087e-01 -1.76836863e-01 -7.88554430e-01 2.75802344e-01 4.67891186e-01 3.64673920e-02 -2.52416164e-01 8.11008096e-01 3.54952574e-01 4.71695453e-01 -4.94751662e-01 8.38358879e-01 2.97595501e-01 7.39517927e-01 -4.80691731e-01 8.39770973e-01 6.90217495e-01 -1.64500237e-01 -7.21287787e-01 -5.47873616e-01 2.32609361e-01 -5.36453366e-01 1.80371869e-02 7.15259194e-01 -8.33634615e-01 -3.18910688e-01 8.90542388e-01 -1.19576883e+00 -8.22654605e-01 -4.79418963e-01 -7.50274435e-02 -5.05656958e-01 4.06985492e-01 -8.89581382e-01 -2.27569759e-01 -2.39112541e-01 -1.46596503e+00 8.40681076e-01 -1.53604403e-01 -3.92536819e-01 -1.07734549e+00 -1.69043303e-01 2.52383500e-01 7.72293925e-01 1.33602992e-01 9.10308659e-01 3.50608140e-01 -7.92964399e-01 1.26201004e-01 -1.78087831e-01 7.41959810e-01 1.51132733e-01 1.40034765e-01 -6.43483877e-01 -2.95345187e-01 1.63656771e-01 -1.07990853e-01 9.65927660e-01 4.68492627e-01 9.64092374e-01 -4.48428959e-01 -4.26832438e-02 9.66811776e-01 1.34257376e+00 -1.55298769e-01 1.01374030e+00 1.63193092e-01 9.27013397e-01 3.34264964e-01 -2.42908355e-02 1.85877793e-02 2.95969486e-01 4.77506548e-01 2.30661884e-01 -4.31958973e-01 -7.27846920e-01 -4.91994411e-01 4.92138743e-01 9.38481987e-01 -5.56667335e-03 -1.33301184e-01 -7.03092217e-01 6.05032682e-01 -1.62735379e+00 -1.01331544e+00 -1.26541346e-01 2.11410356e+00 1.12003720e+00 3.76395658e-02 -3.50626647e-01 -9.70452875e-02 5.19115686e-01 2.16680497e-01 -7.15857983e-01 -3.39934319e-01 -3.23745012e-01 2.03640908e-01 5.51097274e-01 1.05343342e+00 -6.03251874e-01 9.91290927e-01 6.03276777e+00 9.16437447e-01 -1.40874541e+00 4.63631481e-01 6.16645336e-01 -1.13509744e-01 -7.52169371e-01 2.51856685e-01 -1.60362035e-01 4.37981606e-01 6.89501345e-01 -1.65422284e-03 1.14438248e+00 1.78318560e-01 4.87286180e-01 -2.57523078e-02 -8.23470533e-01 1.06793869e+00 -2.69377418e-02 -1.65912294e+00 4.02738452e-01 1.80177823e-01 1.07875669e+00 -3.06130070e-02 4.40629780e-01 -1.84623197e-01 4.74922419e-01 -9.83324289e-01 1.14014137e+00 8.09202313e-01 1.10652637e+00 -5.84580183e-01 1.67005941e-01 2.93807924e-01 -6.32129312e-01 1.99161276e-01 -2.74532080e-01 2.30222583e-01 4.93366569e-01 8.28157604e-01 -1.97282940e-01 2.05477878e-01 5.05817235e-01 7.89379597e-01 -3.42880309e-01 3.05496067e-01 -5.30425131e-01 6.51496172e-01 -3.55641067e-01 8.42110634e-01 1.04609504e-01 -6.56789362e-01 7.11807013e-01 1.14474487e+00 4.24497753e-01 6.25597686e-02 -3.67817163e-01 1.39274573e+00 -3.25929999e-01 -4.24659044e-01 -6.19572222e-01 -7.58100003e-02 1.33001745e-01 1.03235960e+00 -5.29230297e-01 -4.08879697e-01 -7.38809928e-02 1.74488425e+00 1.79127648e-01 6.84242904e-01 -8.73141527e-01 -2.20578238e-01 7.21689701e-01 3.88598144e-01 2.97060758e-01 -4.67599094e-01 -4.04642671e-01 -1.29926014e+00 -5.03282137e-02 -1.10955250e+00 -2.47808293e-01 -1.02262306e+00 -1.18990993e+00 6.65961564e-01 -1.80472136e-01 -9.44364846e-01 -7.31793940e-02 -1.49495229e-01 -5.74648261e-01 1.08694160e+00 -1.39677429e+00 -1.21687484e+00 -3.85211825e-01 5.55869281e-01 3.98412853e-01 2.15867668e-01 4.92447525e-01 5.03806412e-01 -3.63114476e-01 5.64728856e-01 1.87092006e-01 -4.98648323e-02 8.47361207e-01 -1.03830719e+00 7.05010772e-01 1.04001284e+00 1.27842247e-01 8.96752059e-01 8.21369231e-01 -8.06479037e-01 -1.39607537e+00 -1.14355612e+00 7.36170769e-01 -4.40624446e-01 6.09584987e-01 -3.25450510e-01 -1.01286304e+00 8.33585560e-01 5.61751366e-01 -3.18689682e-02 1.37193963e-01 -5.63350320e-01 -3.59445661e-01 4.81452830e-02 -1.12395775e+00 9.66888249e-01 1.35030282e+00 -8.57244730e-01 -1.59452975e-01 1.56690031e-01 7.30143011e-01 -5.71610689e-01 -8.68853331e-01 -3.19570974e-02 4.07947689e-01 -9.26432610e-01 1.05378330e+00 -1.73774227e-01 7.27734327e-01 -4.58001316e-01 7.81645477e-02 -1.54018390e+00 -3.00940603e-01 -1.15662742e+00 -8.76692608e-02 1.24226773e+00 4.84607488e-01 -6.39361858e-01 3.55532438e-01 7.41540670e-01 8.28600749e-02 -4.56638277e-01 -5.66183627e-01 -7.09310591e-01 1.79849654e-01 -5.09683669e-01 4.04391944e-01 1.05130148e+00 -7.27774799e-01 1.31318733e-01 -5.52488148e-01 -2.31464189e-02 7.63856232e-01 -2.76715308e-01 8.27449262e-01 -6.62305653e-01 -3.36880118e-01 -4.22706604e-01 -8.42441432e-03 -1.40524602e+00 7.93366060e-02 -9.78708863e-01 -1.44675598e-01 -1.54633975e+00 -3.10607105e-02 -4.93083745e-01 1.13432780e-01 4.32897270e-01 1.98865049e-02 5.27684450e-01 4.50740844e-01 5.74292362e-01 -2.68084500e-02 6.77526534e-01 1.56186128e+00 -3.66801888e-01 -1.57909870e-01 -4.60324973e-01 -7.58755922e-01 6.63155079e-01 4.58071470e-01 -7.76804566e-01 -5.68742514e-01 -1.02086020e+00 3.41759980e-01 -6.77084252e-02 5.51816165e-01 -6.63809299e-01 2.65633881e-01 -4.06506322e-02 2.06224188e-01 -6.21807063e-03 3.17770362e-01 -6.98886395e-01 8.66721690e-01 4.47037280e-01 -4.08813119e-01 2.32466429e-01 1.29761785e-01 5.10932267e-01 -2.68959552e-01 -2.70641625e-01 9.50103104e-01 -2.28143990e-01 -4.35756177e-01 2.55664229e-01 -2.85454690e-01 -7.40427233e-04 6.98775768e-01 -2.74772584e-01 -1.48249805e-01 -7.08907068e-01 -7.88667917e-01 -1.76426962e-01 9.91207778e-01 3.25905621e-01 5.88660479e-01 -1.20681369e+00 -7.26794362e-01 4.18902695e-01 -3.49008054e-01 3.75154652e-02 3.14735681e-01 1.03868330e+00 -8.04759204e-01 -2.47491285e-01 4.05277610e-02 -6.06992543e-01 -1.00797951e+00 4.88642216e-01 4.44543868e-01 -2.86715627e-01 -7.24517465e-01 7.71274745e-01 3.76063019e-01 -5.81927419e-01 -1.69255301e-01 -3.46297532e-01 5.26496589e-01 -1.67087436e-01 4.75899011e-01 2.15604752e-01 8.12415220e-03 -5.73322833e-01 1.12290896e-01 7.78648615e-01 -8.77334476e-02 -5.86511016e-01 1.33561647e+00 -4.33138072e-01 -4.27930266e-01 4.86225598e-02 1.31929934e+00 8.23443234e-02 -1.76161981e+00 -2.78001547e-01 -4.70274627e-01 -5.97398996e-01 3.61833096e-01 -8.39215279e-01 -1.39901149e+00 9.46445107e-01 5.64679682e-01 -1.53657839e-01 1.16611648e+00 -3.56287926e-01 9.86559093e-01 5.27518578e-02 1.32514924e-01 -8.80207956e-01 2.06442028e-01 3.95700663e-01 1.30659783e+00 -1.00347483e+00 -6.70716688e-02 -2.87612468e-01 -8.51368189e-01 7.83056557e-01 1.16001062e-01 -1.93689376e-01 6.89655542e-01 5.34295082e-01 3.13166291e-01 -1.42612040e-01 -4.67874914e-01 3.43476444e-01 7.07501248e-02 5.04901707e-01 1.08469814e-01 -1.81091964e-01 -5.71962968e-02 -3.93608632e-03 -9.91143845e-03 3.28164726e-01 5.23413539e-01 8.33248496e-01 1.96463451e-01 -1.30378604e+00 -3.16343814e-01 2.56338298e-01 -2.63398916e-01 -3.73234868e-01 -2.87378132e-01 4.00606066e-01 1.11148901e-01 7.04816043e-01 3.76400910e-02 -1.53175801e-01 2.93441951e-01 -2.70438492e-01 6.32760882e-01 -2.22343892e-01 -6.02073610e-01 1.22281089e-01 -2.11656153e-01 -6.79462492e-01 -3.94111216e-01 -5.08074582e-01 -1.21554768e+00 -7.83217132e-01 -2.07110658e-01 -2.46309772e-01 7.56975889e-01 7.21332371e-01 7.48852491e-01 4.72720861e-01 2.52202541e-01 -1.07258499e+00 -3.86883467e-01 -7.31266439e-01 -3.42734933e-01 8.11599433e-01 4.28099424e-01 -4.33827043e-01 -3.47259611e-01 5.98199844e-01]
[11.454667091369629, -0.4751112461090088]
0f15642d-30c1-4606-be6c-52cff4348691
unsupervised-hdr-image-and-video-tone-mapping
2303.07327
null
https://arxiv.org/abs/2303.07327v2
https://arxiv.org/pdf/2303.07327v2.pdf
Unsupervised HDR Image and Video Tone Mapping via Contrastive Learning
Capturing high dynamic range (HDR) images (videos) is attractive because it can reveal the details in both dark and bright regions. Since the mainstream screens only support low dynamic range (LDR) content, tone mapping algorithm is required to compress the dynamic range of HDR images (videos). Although image tone mapping has been widely explored, video tone mapping is lagging behind, especially for the deep-learning-based methods, due to the lack of HDR-LDR video pairs. In this work, we propose a unified framework (IVTMNet) for unsupervised image and video tone mapping. To improve unsupervised training, we propose domain and instance based contrastive learning loss. Instead of using a universal feature extractor, such as VGG to extract the features for similarity measurement, we propose a novel latent code, which is an aggregation of the brightness and contrast of extracted features, to measure the similarity of different pairs. We totally construct two negative pairs and three positive pairs to constrain the latent codes of tone mapped results. For the network structure, we propose a spatial-feature-enhanced (SFE) module to enable information exchange and transformation of nonlocal regions. For video tone mapping, we propose a temporal-feature-replaced (TFR) module to efficiently utilize the temporal correlation and improve the temporal consistency of video tone-mapped results. We construct a large-scale unpaired HDR-LDR video dataset to facilitate the unsupervised training process for video tone mapping. Experimental results demonstrate that our method outperforms state-of-the-art image and video tone mapping methods. Our code and dataset are available at https://github.com/cao-cong/UnCLTMO.
['Jingyu Yang', 'Xin Liu', 'Huanjing Yue', 'Cong Cao']
2023-03-13
null
null
null
null
['tone-mapping']
['computer-vision']
[ 3.79118204e-01 -6.33463979e-01 -3.33909571e-01 -3.09974819e-01 -6.10449553e-01 -2.10853070e-01 3.59054387e-01 -7.65062690e-01 -2.10106716e-01 5.82587540e-01 1.53107971e-01 -6.21908270e-02 -1.23422192e-02 -9.11630452e-01 -7.50061393e-01 -8.25017154e-01 1.81425780e-01 -3.12455386e-01 4.19433206e-01 -2.59689987e-01 1.21474557e-01 9.99249443e-02 -1.54204488e+00 5.80222130e-01 8.97070944e-01 1.22293150e+00 6.09281659e-01 3.38666469e-01 -2.78806183e-02 1.03566301e+00 -1.86704680e-01 9.73953400e-03 4.84399468e-01 -7.00842559e-01 -5.00557303e-01 6.19078279e-02 3.56671989e-01 -7.63409257e-01 -1.07137847e+00 1.20728970e+00 4.62719023e-01 2.48886883e-01 2.37911060e-01 -1.19802761e+00 -1.01368427e+00 3.23342085e-01 -7.79306233e-01 3.81003082e-01 3.72229367e-01 2.82226771e-01 6.27766550e-01 -9.00124550e-01 7.44022429e-01 8.98940384e-01 2.81119138e-01 4.99294102e-01 -1.00051010e+00 -1.07845867e+00 -2.62332037e-02 4.82123613e-01 -1.53691816e+00 -4.51740026e-01 1.02506912e+00 -1.41944498e-01 7.05763042e-01 3.53025585e-01 7.84583926e-01 9.46257412e-01 9.91794933e-03 5.89728415e-01 1.40464127e+00 -1.85059264e-01 -2.39237890e-01 -6.75603151e-02 -6.85256958e-01 5.64113915e-01 -3.61065686e-01 4.54620004e-01 -7.72377014e-01 3.88176858e-01 1.26104295e+00 2.24241257e-01 -6.40847981e-01 -2.05446765e-01 -1.41656888e+00 4.64993566e-01 4.45968747e-01 3.42619032e-01 2.22937968e-02 2.09222719e-01 3.15116495e-01 6.24956489e-01 4.07731265e-01 -1.03457719e-02 -1.87788770e-01 -3.46049219e-01 -9.35512125e-01 -2.35828876e-01 9.80801582e-02 9.84673738e-01 9.55986023e-01 2.08998308e-01 -2.40605295e-01 1.11193800e+00 4.49996330e-02 5.58010101e-01 5.94482780e-01 -1.04841852e+00 5.69352090e-01 3.52722794e-01 -1.18246816e-01 -1.24600804e+00 9.01771337e-02 5.41038848e-02 -1.02696097e+00 1.15034245e-02 -5.82124889e-02 2.65344799e-01 -9.31697428e-01 1.54108989e+00 1.52306110e-01 5.33304572e-01 -2.22901627e-01 1.32072973e+00 8.09747756e-01 1.11735225e+00 -1.27934813e-01 -3.64037722e-01 9.88186955e-01 -7.31272399e-01 -8.53169382e-01 1.60036087e-01 3.24801058e-01 -8.28768075e-01 1.42323554e+00 2.95856416e-01 -1.17538810e+00 -8.58871698e-01 -1.12598515e+00 -4.44128156e-01 -3.81549954e-01 -2.88199503e-02 4.09740955e-01 3.38956565e-01 -9.47688818e-01 5.20953238e-01 -5.36537945e-01 -5.96281849e-02 2.97769129e-01 2.61392593e-01 -4.06081676e-01 -3.93355250e-01 -1.67877853e+00 5.50864816e-01 3.71344358e-01 3.09019923e-01 -8.33625436e-01 -7.34453559e-01 -7.58975685e-01 -1.31292254e-01 2.76058286e-01 -3.32339615e-01 5.71676970e-01 -1.18308759e+00 -1.71942806e+00 8.98740351e-01 2.66929150e-01 -7.22617209e-02 6.23710990e-01 1.73429102e-01 -6.88108325e-01 4.83616620e-01 -1.16129503e-01 9.15024936e-01 9.36711490e-01 -1.11333489e+00 -6.65425241e-01 1.91836193e-01 -2.32244469e-02 4.34529006e-01 -6.87490880e-01 1.20253392e-01 -8.66590858e-01 -9.77730751e-01 1.24140598e-01 -8.46670151e-01 2.32047930e-01 2.54588217e-01 -2.15308964e-01 3.65034401e-01 1.14934766e+00 -9.10525799e-01 1.39971602e+00 -2.39406514e+00 -2.75831688e-02 1.98825851e-01 3.02812070e-01 1.47898301e-01 -2.02812105e-01 -5.35991564e-02 -2.46385574e-01 -5.75415008e-02 -2.81070381e-01 1.38994768e-01 -9.72850174e-02 9.60510783e-03 -5.10954857e-01 4.94039178e-01 7.50126317e-02 1.08508706e+00 -7.66313374e-01 -5.77053547e-01 6.79858923e-01 6.37109816e-01 -3.29498470e-01 1.76110238e-01 1.02701895e-02 5.56105375e-01 -3.64420414e-01 6.21456504e-01 8.43960881e-01 -1.81962103e-01 3.53189968e-02 -7.23979473e-01 -3.76993239e-01 1.64745580e-02 -9.02925909e-01 1.96897817e+00 -5.76841474e-01 8.84679317e-01 -3.78685296e-01 -7.51367986e-01 1.15447903e+00 1.63986206e-01 7.40702093e-01 -1.40222466e+00 8.83380622e-02 2.47535706e-01 -1.15502402e-01 -5.71259499e-01 5.65013587e-01 1.23074010e-01 -6.37598932e-02 3.59233916e-01 -8.11197683e-02 3.06334645e-02 -1.86220314e-02 7.69424215e-02 8.84672225e-01 2.77232468e-01 -1.36966258e-01 -2.50200965e-02 4.47880983e-01 -3.70762795e-01 5.67240298e-01 3.52115214e-01 -2.75560409e-01 1.00929630e+00 2.07561687e-01 -5.00176609e-01 -1.34940958e+00 -1.17051697e+00 -2.23458767e-01 9.24546421e-01 7.54358947e-01 -1.82534188e-01 -4.62013304e-01 -2.00087771e-01 -3.48976731e-01 1.49014831e-01 -6.42193854e-01 -4.97567266e-01 -7.07395315e-01 -4.72428560e-01 3.35133553e-01 2.73082167e-01 1.19413340e+00 -1.07783556e+00 -2.73702592e-01 1.02115318e-01 -5.61723888e-01 -1.17744207e+00 -9.58100975e-01 -2.83608288e-02 -4.94436949e-01 -6.44380629e-01 -8.87938201e-01 -9.14067924e-01 3.72896761e-01 4.75962579e-01 7.34896898e-01 9.33495760e-02 -3.14625293e-01 8.48178566e-02 -5.64220309e-01 4.38030034e-01 -1.49538100e-01 -1.61956742e-01 -5.40368669e-02 1.63451001e-01 2.61850297e-01 -7.51267850e-01 -9.72822368e-01 7.11548090e-01 -1.28044069e+00 4.87870604e-01 6.03794277e-01 7.88775980e-01 7.97389030e-01 1.21072009e-01 3.82831693e-01 -5.46005845e-01 3.14026326e-01 -2.04697520e-01 -4.38486904e-01 3.79448593e-01 -3.09066683e-01 -3.86381298e-01 7.51831412e-01 -8.13718557e-01 -1.07082772e+00 -1.46410272e-01 1.73825890e-01 -1.00417650e+00 1.42809287e-01 2.38187864e-01 -3.33943695e-01 -4.54678297e-01 2.53505975e-01 7.34031856e-01 -1.95855647e-01 -4.45518456e-02 2.69060612e-01 7.78051138e-01 7.22689569e-01 -3.62094700e-01 1.00578988e+00 5.28330386e-01 -3.15465212e-01 -4.61823672e-01 -4.18427765e-01 -1.06726028e-01 -2.69167989e-01 -5.91110587e-01 9.87749696e-01 -1.21644592e+00 -4.96005237e-01 5.14070332e-01 -6.92652762e-01 -7.10511386e-01 -1.05804071e-01 5.49181700e-01 -6.59958184e-01 3.47509116e-01 -8.48201871e-01 -1.52720615e-01 -2.15592355e-01 -1.14691281e+00 8.93505514e-01 3.01918745e-01 3.66789699e-01 -6.84844732e-01 -1.10603534e-01 2.01037019e-01 6.41016424e-01 1.28773183e-01 5.89528501e-01 2.60600179e-01 -9.97851908e-01 2.40367681e-01 -5.89788556e-01 3.40996921e-01 3.88496369e-01 -6.13783784e-02 -8.13817978e-01 -3.22648436e-01 -3.62771563e-02 -3.39694470e-01 9.56817985e-01 4.13233846e-01 1.61945724e+00 -1.02218725e-01 -7.11743385e-02 1.11217690e+00 1.37822616e+00 3.25544178e-01 1.27787757e+00 4.85251397e-01 1.02363861e+00 2.93790102e-01 8.91668618e-01 2.80835748e-01 2.03847244e-01 9.39867198e-01 -6.21803850e-02 -4.98723090e-01 -4.02226180e-01 -4.44385707e-01 5.77650845e-01 9.17588115e-01 -7.66719654e-02 -8.33437741e-02 -4.39487398e-01 1.58419341e-01 -1.66334617e+00 -1.15394700e+00 3.96389812e-01 2.08417892e+00 1.05257845e+00 2.17619129e-02 -1.50026843e-01 -5.56067191e-02 1.08147562e+00 3.66226315e-01 -6.69293225e-01 -7.20040277e-02 -3.69134337e-01 3.72845158e-02 5.12653649e-01 2.44933248e-01 -1.02860999e+00 9.66378808e-01 5.10393572e+00 1.39928603e+00 -1.56652331e+00 1.88014016e-01 9.58829582e-01 -2.78409928e-01 -5.24247169e-01 -1.56888947e-01 -2.45103016e-01 9.18685675e-01 4.31443095e-01 -1.28350565e-02 9.29819822e-01 4.66445446e-01 5.02038479e-01 1.29977375e-01 -7.66020775e-01 1.42346478e+00 2.62884051e-03 -1.33691692e+00 9.62356031e-02 4.76722047e-02 1.05670476e+00 -1.38607815e-01 6.92176282e-01 7.49289915e-02 -2.31376335e-01 -7.92824924e-01 6.98955774e-01 5.23555934e-01 1.64620662e+00 -7.94912040e-01 2.78718024e-01 -3.32787395e-01 -1.55659831e+00 -7.02724084e-02 -5.74976563e-01 4.25133318e-01 -8.46366026e-03 4.38091874e-01 5.82839064e-02 4.44734961e-01 1.11135697e+00 1.10130012e+00 -4.13337588e-01 7.02857792e-01 1.24041259e-01 2.06302449e-01 -1.37641236e-01 4.41349953e-01 2.83465795e-02 -4.09332633e-01 2.52449155e-01 9.49792564e-01 4.90805715e-01 3.25465977e-01 8.38617608e-02 9.83698070e-01 -2.18656197e-01 -7.63149634e-02 -5.84294438e-01 9.20657068e-02 6.52737916e-01 1.41391122e+00 -7.61233270e-01 -2.37329587e-01 -6.13107800e-01 1.36863875e+00 -2.79852375e-02 5.15833259e-01 -1.14568138e+00 -6.74006283e-01 3.60354662e-01 1.55865207e-01 4.07298654e-01 -1.06455110e-01 1.21324249e-01 -1.28590691e+00 1.95186317e-01 -9.03187275e-01 1.20333210e-01 -1.20413017e+00 -1.17249298e+00 5.73381186e-01 -1.39963686e-01 -1.99983191e+00 9.36076045e-02 -3.01123738e-01 -6.00335062e-01 6.36064529e-01 -1.78243315e+00 -1.14207721e+00 -7.20369101e-01 1.00051022e+00 3.96927267e-01 -1.68599695e-01 1.97781816e-01 8.62681329e-01 -3.79520923e-01 7.70292938e-01 1.68446034e-01 2.43233472e-01 9.82880414e-01 -7.06436753e-01 2.24841237e-01 8.86677027e-01 -1.22496851e-01 5.43628931e-01 3.74923438e-01 -5.86166441e-01 -1.45809448e+00 -1.33981156e+00 1.56419456e-01 1.00707278e-01 5.65402806e-01 -5.16268253e-01 -9.61169243e-01 3.92153144e-01 1.94971971e-02 3.05195451e-01 4.71892834e-01 -5.91477871e-01 -3.73296678e-01 -5.36709487e-01 -9.68741357e-01 6.42771840e-01 1.22829485e+00 -9.69509482e-01 7.77927488e-02 -2.67070774e-02 9.89830852e-01 -3.90596867e-01 -1.07076013e+00 4.94681537e-01 6.62300110e-01 -1.11980259e+00 9.70047593e-01 3.36241096e-01 7.01271594e-01 -6.83061600e-01 -2.73163140e-01 -9.70614135e-01 -2.41639689e-01 -5.71921825e-01 -9.47890058e-03 1.32957304e+00 8.91880095e-02 -5.65597117e-01 6.10419214e-01 4.16529685e-01 -2.17260659e-01 -6.03215873e-01 -8.91545475e-01 -5.84327519e-01 -2.59261549e-01 -1.77794963e-01 5.45222461e-01 1.28351271e+00 -1.33563206e-01 -2.50104189e-01 -8.65324974e-01 -8.41814429e-02 4.55175161e-01 3.88982445e-01 4.09074873e-01 -4.16719109e-01 -2.37644434e-01 -2.59310722e-01 -4.65587586e-01 -1.11172318e+00 6.92739189e-02 -7.27601469e-01 1.93335898e-02 -1.07591558e+00 4.23892528e-01 -4.89402711e-01 -5.51833391e-01 3.30558896e-01 7.59757906e-02 8.30658913e-01 1.97221160e-01 3.49000394e-01 -7.39500284e-01 8.11976492e-01 1.68165338e+00 -8.26575160e-02 -2.43444473e-01 -7.34296739e-01 -2.80020803e-01 2.31501311e-01 7.72930801e-01 -2.24312916e-01 -4.74389106e-01 -5.23468614e-01 1.67211667e-01 1.58395275e-01 4.80235636e-01 -1.05923033e+00 2.87440479e-01 -2.30872065e-01 8.09627831e-01 -4.53871101e-01 2.94440478e-01 -7.29902625e-01 4.76218849e-01 2.15362698e-01 -5.26637077e-01 3.01094241e-02 -2.24306323e-02 3.67013961e-01 -5.32626867e-01 3.66082817e-01 9.52392042e-01 -4.29810248e-02 -1.01006353e+00 8.97279084e-01 1.22526139e-02 -1.57988459e-01 8.49008203e-01 -4.94387329e-01 -5.34260988e-01 -4.46186453e-01 -3.25196326e-01 3.10383569e-02 6.89382792e-01 6.76738501e-01 1.07994044e+00 -1.74494803e+00 -4.83029932e-01 3.76229018e-01 1.87029660e-01 -2.86466002e-01 8.97357106e-01 8.14477623e-01 -6.20353281e-01 -1.58841815e-02 -7.49268711e-01 -4.97454315e-01 -9.93322194e-01 6.12373590e-01 2.88594186e-01 1.04100458e-01 -7.03969181e-01 5.12552917e-01 5.94587564e-01 -8.44021887e-02 -2.55728606e-02 -1.44907057e-01 7.79657206e-03 -3.41997951e-01 6.67153597e-01 1.07680634e-01 -2.89352566e-01 -6.63396776e-01 2.30014301e-03 8.75970125e-01 -2.46042177e-01 -6.46022186e-02 1.13596320e+00 -6.11521780e-01 -1.30276516e-01 3.07267606e-01 1.67305958e+00 -2.41672710e-01 -1.50140393e+00 -2.44226798e-01 -6.38541043e-01 -9.46180284e-01 3.49421859e-01 -4.47045773e-01 -1.46151412e+00 7.60058522e-01 1.06697428e+00 -2.65982635e-02 1.71314418e+00 -2.52520651e-01 1.14055514e+00 -4.19035694e-03 2.79805422e-01 -1.23911464e+00 3.71501565e-01 1.86326668e-01 6.05605423e-01 -1.15474784e+00 -1.58905268e-01 -3.52633566e-01 -6.16136849e-01 1.11283851e+00 8.51838470e-01 -5.92067651e-02 4.61471379e-01 2.03654408e-01 5.69590628e-02 6.37905151e-02 -6.69171154e-01 -1.67928264e-01 2.22505823e-01 6.50250137e-01 1.79241329e-01 -3.04429859e-01 -8.97558965e-03 1.08251646e-01 3.37909572e-02 1.90818369e-01 3.89751345e-01 5.95350742e-01 -3.23510617e-01 -8.50490332e-01 -4.47827484e-03 4.28974092e-01 -2.37662524e-01 -3.32289726e-01 7.39123896e-02 6.21120930e-01 1.75341681e-01 6.74660802e-01 2.20805466e-01 -9.09740686e-01 4.74979356e-02 -5.96711338e-01 5.81784546e-01 -8.77096355e-02 -1.35454223e-01 2.85028219e-01 -3.67001086e-01 -7.62851894e-01 -6.15753949e-01 -2.01986700e-01 -1.14577746e+00 -5.66271245e-01 -1.87817425e-01 -2.33729750e-01 2.65612751e-01 5.36936402e-01 1.73175409e-01 4.79206026e-01 1.18291581e+00 -8.21802437e-01 1.54930398e-01 -6.00600600e-01 -9.23624158e-01 5.64873934e-01 2.44154871e-01 -6.34928405e-01 -2.49345690e-01 3.33118528e-01]
[10.941662788391113, -2.1558117866516113]
1d26f207-4eeb-40e8-a6f3-e024ca00daae
inverse-consistency-by-construction-for
2305.00087
null
https://arxiv.org/abs/2305.00087v1
https://arxiv.org/pdf/2305.00087v1.pdf
Inverse Consistency by Construction for Multistep Deep Registration
Inverse consistency is a desirable property for image registration. We propose a simple technique to make a neural registration network inverse consistent by construction, as a consequence of its structure, as long as it parameterizes its output transform by a Lie group. We extend this technique to multi-step neural registration by composing many such networks in a way that preserves inverse consistency. This multi-step approach also allows for inverse-consistent coarse to fine registration. We evaluate our technique on synthetic 2-D data and four 3-D medical image registration tasks and obtain excellent registration accuracy while assuring inverse consistency.
['Marc Niethammer', 'Richard Rushmore', 'Raul San Jose Estepar', 'Sylvain Bouix', 'Roland Kwitt', 'Francois-Xavier Vialard', 'Lin Tian', 'Hastings Greer']
2023-04-28
null
null
null
null
['image-registration', 'medical-image-registration']
['computer-vision', 'medical']
[ 2.65422940e-01 1.88688844e-01 -8.93636644e-02 -6.77898228e-01 -7.13187456e-01 -4.80615944e-01 7.67472923e-01 -2.40841046e-01 -5.93088269e-01 4.60578412e-01 3.50784272e-01 8.15976709e-02 -3.37048769e-01 -8.07420135e-01 -7.16633499e-01 -5.85093737e-01 -1.48070008e-01 5.89176118e-01 1.16452150e-01 -5.68792820e-01 7.87819922e-02 7.78613031e-01 -9.23759818e-01 -8.29385221e-02 7.32352197e-01 5.29696763e-01 -2.37283021e-01 5.12304187e-01 4.23216224e-01 3.15500766e-01 -8.82321075e-02 -1.51001140e-01 7.12976396e-01 -7.37195313e-01 -1.13787019e+00 -1.44562826e-01 8.86825323e-01 -1.94868520e-01 -1.51468456e-01 9.25208330e-01 4.98831451e-01 3.23357671e-01 7.89452374e-01 -1.11776006e+00 -9.76545691e-01 3.46103698e-01 -5.35481870e-01 1.22558884e-01 1.74686104e-01 -3.62285048e-01 8.05446446e-01 -6.98516965e-01 7.23813951e-01 1.06644940e+00 1.33581984e+00 6.41684711e-01 -1.59579647e+00 -5.37720680e-01 -4.38519210e-01 -1.78316742e-01 -1.43669307e+00 -6.92139685e-01 7.12295413e-01 -2.76753515e-01 6.55657113e-01 2.54524350e-01 5.33691406e-01 6.38385475e-01 7.42337286e-01 -3.86382230e-02 1.35543215e+00 -3.02342981e-01 -6.91499263e-02 -5.66743970e-01 1.18357442e-01 8.19124937e-01 7.60440677e-02 3.65718752e-01 -4.18136150e-01 -2.43501961e-01 1.27520680e+00 -2.31249616e-01 -2.71979272e-01 -4.60756779e-01 -1.50615978e+00 6.40502751e-01 7.44599402e-01 7.12286294e-01 -3.34779173e-01 3.80846471e-01 1.02737822e-01 5.83423913e-01 4.84575212e-01 4.18126523e-01 2.30630785e-02 3.91297966e-01 -9.58603263e-01 2.63858810e-02 8.49946618e-01 6.17557943e-01 7.82113791e-01 8.52950513e-02 1.78912640e-01 8.11579764e-01 4.44292784e-01 4.00301754e-01 6.09820604e-01 -1.45653689e+00 7.34930299e-03 -3.71505320e-02 -5.19823372e-01 -1.02024043e+00 -6.17954791e-01 -3.75263065e-01 -1.49140203e+00 6.13943875e-01 4.04407531e-01 3.13184947e-01 -8.42814207e-01 2.24709153e+00 3.07732880e-01 2.38177240e-01 1.21015981e-01 7.35668957e-01 7.70050943e-01 1.36742637e-01 -1.74068063e-02 -8.37909728e-02 1.28067625e+00 -5.12817562e-01 -8.84723186e-01 1.60137013e-01 4.90601033e-01 -7.49024749e-01 7.40941942e-01 -2.64985040e-02 -1.36854219e+00 -5.77952087e-01 -1.06029129e+00 -1.95254162e-01 -1.71945989e-02 -3.93065900e-01 4.35115665e-01 4.56976086e-01 -1.69034016e+00 1.05667102e+00 -9.85704839e-01 -3.65883976e-01 2.51133502e-01 8.40353906e-01 -1.04952419e+00 4.19208616e-01 -1.14745295e+00 1.23350811e+00 2.37157136e-01 2.09371746e-01 -1.52496502e-01 -6.80857956e-01 -1.06863034e+00 -4.66219097e-01 -3.93291503e-01 -9.97246742e-01 9.63276327e-01 -8.22209358e-01 -1.34849441e+00 1.33980024e+00 -2.76677966e-01 -2.43144974e-01 5.05225539e-01 2.73598373e-01 -3.49382102e-01 -1.45734437e-02 1.45705730e-01 7.69766867e-01 6.57812059e-01 -1.35705519e+00 1.85151249e-02 -5.97682297e-01 -2.32699677e-01 2.50757664e-01 -1.31135538e-01 -9.07176360e-02 -2.26818472e-01 -1.07608843e+00 8.69577706e-01 -1.26102710e+00 -4.60210323e-01 4.28414226e-01 -3.20994318e-01 8.63641724e-02 6.72997117e-01 -6.91820085e-01 6.20909512e-01 -1.79511821e+00 1.19582899e-01 7.05964029e-01 5.11800468e-01 -2.14785308e-01 -3.65420043e-01 -2.03227535e-01 -4.83785719e-01 6.17383681e-02 -6.26782060e-01 -5.04619122e-01 -2.10758567e-01 3.29581261e-01 -1.39835015e-01 1.11852181e+00 -1.53572872e-01 1.13635433e+00 -5.79498827e-01 -6.56946719e-01 7.40481913e-02 7.84936130e-01 -6.06694460e-01 1.71570718e-01 5.68610609e-01 9.57989097e-01 -4.07979101e-01 2.49184266e-01 7.30016351e-01 -1.02360256e-01 -1.23186044e-01 -6.90454602e-01 -1.60627570e-02 -5.45392483e-02 -1.00163209e+00 1.76489222e+00 -5.78218877e-01 5.24489880e-01 2.92073995e-01 -1.01916230e+00 9.76943731e-01 6.13636196e-01 9.42139506e-01 -6.80847347e-01 2.55868107e-01 3.11620742e-01 5.82293086e-02 -3.95438299e-02 4.24589843e-01 -5.56434393e-01 -6.64040400e-03 9.01827514e-01 -1.29337937e-01 -3.67198020e-01 -2.56599039e-01 -1.14212744e-01 7.94750094e-01 1.75701171e-01 5.71236849e-01 -7.46539533e-01 6.32820606e-01 -1.82541221e-01 5.36331415e-01 5.83983839e-01 -1.88523009e-01 1.00902140e+00 -1.52387008e-01 -6.08073711e-01 -1.09836662e+00 -1.30878437e+00 -3.20857167e-01 5.38009882e-01 2.10389197e-01 2.80058771e-01 -9.72008824e-01 -2.90267497e-01 -1.87745705e-01 -9.25362669e-03 -8.61577511e-01 -3.23579431e-01 -1.10807741e+00 -6.89430952e-01 8.53655338e-01 4.59375024e-01 8.42200935e-01 -1.03647077e+00 -2.40829457e-02 1.33343965e-01 -2.45252550e-01 -9.34911668e-01 -1.06675756e+00 8.82891268e-02 -1.22529995e+00 -1.07246411e+00 -9.07436728e-01 -1.16974163e+00 1.04970825e+00 5.51943434e-03 1.29646909e+00 4.26596224e-01 1.08960360e-01 2.93868363e-01 3.85917187e-01 3.60386133e-01 -9.55333173e-01 -3.49184498e-02 4.31492150e-01 -1.25075966e-01 -1.22318164e-01 -1.03553367e+00 -2.79003799e-01 6.16286099e-01 -9.47076321e-01 -1.60054535e-01 3.35382581e-01 8.66913617e-01 8.88924122e-01 -2.69455194e-01 3.87037396e-01 -8.17580521e-01 7.75592446e-01 4.35893098e-03 -4.48195249e-01 3.39299440e-01 -7.09100902e-01 3.07078868e-01 4.38498676e-01 -4.32107985e-01 -8.72644544e-01 2.38896564e-01 -6.25012815e-01 -2.38327727e-01 -1.34740949e-01 1.01188026e-01 1.02355555e-01 -9.13330615e-01 1.02847576e+00 1.67492583e-01 6.88300133e-01 -3.55069011e-01 3.82314980e-01 2.64375567e-01 1.03794324e+00 -6.73237383e-01 1.34762347e+00 8.07833016e-01 4.84838188e-01 -6.05434120e-01 -5.13284445e-01 -1.52961284e-01 -1.19962549e+00 -1.04428411e-01 1.10289097e+00 -7.68295109e-01 -9.75602865e-01 5.70781410e-01 -1.34865308e+00 -3.53134722e-01 -4.14183915e-01 6.52817309e-01 -1.04027498e+00 4.22553569e-01 -7.53618240e-01 1.77707165e-01 -4.78624225e-01 -1.26452696e+00 8.05725336e-01 -1.74707144e-01 -5.91992557e-01 -1.53651333e+00 5.11628211e-01 -1.33799285e-01 6.59311771e-01 5.53976059e-01 5.12101233e-01 -5.03291786e-01 -2.92875886e-01 1.22913182e-01 -1.50615662e-01 2.64252543e-01 4.68915880e-01 -4.37951446e-01 -7.28685796e-01 -3.15692842e-01 4.36952204e-01 4.61514369e-02 5.54148555e-01 6.47760034e-01 8.62233162e-01 -2.61529058e-01 -1.65271580e-01 1.12727964e+00 1.33677053e+00 -3.63798067e-02 4.94935811e-01 3.63918602e-01 9.89778399e-01 5.07770896e-01 3.14675346e-02 -4.46369320e-01 5.09987712e-01 1.07055795e+00 1.53297856e-01 -6.91286623e-01 -4.73460555e-01 9.18313861e-02 -5.33160195e-02 1.25563729e+00 -5.41474819e-01 4.31964785e-01 -9.02132690e-01 3.31681460e-01 -1.51448333e+00 -9.82672274e-01 -2.35860109e-01 2.09786868e+00 1.15373504e+00 -3.90717387e-01 -1.18059032e-01 -1.56460591e-02 7.60107219e-01 1.36539340e-01 -1.62303448e-01 -3.26527923e-01 -1.79017842e-01 3.67888123e-01 6.62332356e-01 7.88025737e-01 -1.04377687e+00 6.55505121e-01 7.94449472e+00 5.36586583e-01 -9.45191085e-01 3.41056436e-01 5.32703280e-01 6.70239627e-01 -5.27799129e-01 -2.15734363e-01 -3.57473344e-01 -9.97721627e-02 5.03749669e-01 -3.22464079e-01 5.50486326e-01 2.87248641e-01 1.29239440e-01 2.50389308e-01 -1.27012110e+00 8.61917078e-01 1.45331398e-01 -1.60765910e+00 -4.83259223e-02 1.75526857e-01 8.52975667e-01 3.36539410e-02 -2.34251767e-02 -4.17653680e-01 4.20846462e-01 -1.08663583e+00 5.76581836e-01 4.71717954e-01 9.52333093e-01 -4.28449750e-01 3.11719775e-01 1.28018677e-01 -1.31141496e+00 7.38577425e-01 -7.40228668e-02 3.02782267e-01 4.21459466e-01 4.05966550e-01 -2.51971841e-01 3.66305321e-01 6.58859909e-01 8.50773633e-01 -3.18840295e-01 8.24840128e-01 -1.32286549e-01 -2.15010196e-02 -2.99953610e-01 6.90864623e-01 -1.24501906e-01 -6.07116461e-01 5.86762607e-01 1.00156426e+00 4.43964332e-01 3.47255349e-01 -9.35651287e-02 1.04641938e+00 -1.64001137e-01 -1.14768997e-01 -1.00846863e+00 7.72950411e-01 2.91486055e-01 1.07370663e+00 -8.27782512e-01 -1.34161457e-01 -4.65600863e-02 1.00882816e+00 5.47827557e-02 2.00670168e-01 -5.23361027e-01 -1.83229104e-01 7.87814736e-01 -2.70687670e-01 -3.37379277e-01 -5.03139555e-01 -5.62220633e-01 -9.80023623e-01 -1.23935509e-02 -7.40718305e-01 2.36235842e-01 -5.12006760e-01 -1.38854039e+00 1.06776953e+00 -2.05225102e-03 -1.41739893e+00 -4.91087347e-01 -2.60911256e-01 -5.30992091e-01 1.15008581e+00 -1.20552111e+00 -1.28495467e+00 -3.60898763e-01 9.96663988e-01 -2.15361163e-01 2.94647459e-02 1.01383579e+00 3.62728983e-01 1.37314931e-01 6.93188727e-01 -2.90229470e-01 2.51662314e-01 8.43373835e-01 -1.19800019e+00 8.35930824e-01 7.97921240e-01 2.52577931e-01 9.61986959e-01 4.41070795e-01 -5.17363667e-01 -1.17807221e+00 -9.29884791e-01 1.08697474e+00 -4.82434094e-01 4.86350060e-01 -1.68890711e-02 -1.14410079e+00 9.40648317e-01 2.05246001e-01 3.82397652e-01 4.52834100e-01 -4.87259701e-02 -3.58524859e-01 -1.33901313e-01 -1.42879295e+00 6.33961678e-01 1.30071342e+00 -7.21315503e-01 -8.85847986e-01 2.88193315e-01 4.64464039e-01 -7.68512607e-01 -1.61461532e+00 5.88541448e-01 6.08069658e-01 -8.86987269e-01 1.49083626e+00 -1.21548802e-01 1.45197988e-01 -4.00667846e-01 3.57200317e-02 -1.29906428e+00 -4.75705564e-01 -7.29740024e-01 4.13711548e-01 8.37588012e-01 2.03481376e-01 -1.05860448e+00 6.01500928e-01 6.04851067e-01 -1.48192719e-01 -3.70666683e-01 -1.22579801e+00 -1.15201318e+00 2.43543595e-01 -1.72888845e-01 5.97041011e-01 1.29604197e+00 -3.73291820e-01 -1.51881546e-01 -3.28761309e-01 3.99604887e-02 1.05205989e+00 -1.18431233e-01 5.51505566e-01 -1.23929524e+00 -5.55403344e-02 -5.28220654e-01 -4.16281492e-01 -9.73815382e-01 5.74596941e-01 -1.25166440e+00 4.05786157e-01 -1.26398349e+00 7.97823910e-03 -1.03009355e+00 -5.65490723e-02 6.53894007e-01 3.02620441e-01 1.27400267e+00 -1.40606776e-01 6.63292110e-01 1.51464166e-02 2.10154757e-01 1.53753400e+00 1.53054059e-01 -3.01454365e-01 9.25095305e-02 -6.32993042e-01 7.45514572e-01 6.83836043e-01 -5.31997621e-01 -1.45144194e-01 -5.09985805e-01 -2.29246721e-01 7.30655640e-02 5.39355278e-01 -8.97702873e-01 4.57365930e-01 -7.05109909e-04 2.73284525e-01 -8.76103714e-02 3.55636477e-01 -1.00561869e+00 7.88478315e-01 6.87905967e-01 -4.83243763e-01 4.61283714e-01 1.43858576e-02 2.27582920e-02 -3.74631315e-01 1.30292520e-01 1.20827246e+00 2.63354313e-02 -5.80920935e-01 6.38839543e-01 1.32231906e-01 1.44905644e-02 6.92164660e-01 -5.45724571e-01 -3.89469005e-02 -3.66325259e-01 -8.79577756e-01 -2.86482394e-01 7.97224224e-01 3.88696760e-01 5.42157114e-01 -1.84767556e+00 -8.30496788e-01 5.58152616e-01 -1.96412131e-01 -1.72196016e-01 -1.03573985e-01 1.25879788e+00 -7.22743452e-01 5.97211048e-02 -6.25475228e-01 -8.05300951e-01 -1.50018263e+00 -4.78118006e-03 8.02271843e-01 -7.79583231e-02 -9.16328132e-01 5.66896379e-01 1.80798769e-01 -8.20723057e-01 -1.24469422e-01 -1.12922385e-01 -1.75827846e-01 -4.01386857e-01 2.28832394e-01 1.29670471e-01 2.38425195e-01 -1.24156940e+00 -4.96173561e-01 1.41597462e+00 5.45195997e-01 -4.28354800e-01 1.37379241e+00 -9.93249789e-02 -7.85499036e-01 -3.85597609e-02 1.65017879e+00 -2.51872707e-02 -9.74507332e-01 -4.50146317e-01 -2.05736920e-01 -2.25607097e-01 5.38828149e-02 -1.86555699e-01 -1.43086076e+00 3.24365824e-01 5.75276077e-01 1.30535709e-02 1.23831689e+00 5.58166318e-02 6.63004458e-01 4.08165932e-01 3.68296981e-01 -6.63400769e-01 -2.81663448e-01 6.19384646e-01 1.07634270e+00 -1.03768647e+00 2.99324602e-01 -4.52636361e-01 -1.40009999e-01 1.06130672e+00 1.79337800e-01 -5.91260612e-01 1.02527678e+00 4.41007853e-01 3.39447260e-01 -2.32539862e-01 -2.62591317e-02 1.40678167e-01 8.78093660e-01 7.96421230e-01 6.60057724e-01 -1.24767065e-01 -4.93769556e-01 1.57592855e-02 -4.22349632e-01 -2.49302089e-02 3.41231018e-01 6.68058097e-01 -5.48266321e-02 -1.44257748e+00 -5.65584362e-01 5.91349155e-02 -4.06448364e-01 5.21968491e-02 -1.23410791e-01 1.04081726e+00 -1.03673130e-01 3.84833395e-01 3.06780040e-01 -2.62334496e-01 2.70804018e-01 -4.99904126e-01 6.67669356e-01 -3.38582337e-01 -8.00845504e-01 -6.52390644e-02 -1.53106511e-01 -9.53860581e-01 -9.06355977e-01 -5.71606934e-01 -1.39244342e+00 -4.68927860e-01 -1.41574800e-01 5.04328609e-02 7.95177817e-01 9.90982175e-01 1.00210682e-01 2.82093346e-01 6.59972548e-01 -1.10654354e+00 -4.21910316e-01 -5.36652088e-01 -5.24075806e-01 6.38083220e-01 5.87832272e-01 -4.33891684e-01 -2.54570127e-01 4.02182162e-01]
[13.938958168029785, -2.5451152324676514]
80cf2c73-de9d-42a1-8b2f-40e2dae90bcd
deep-hdr-imaging-via-a-non-local-network
null
null
https://ieeexplore.ieee.org/abstract/document/8989959
https://ieeexplore.ieee.org/abstract/document/8989959
Deep HDR Imaging via A Non-Local Network
One of the most challenging problems in reconstructing a high dynamic range (HDR) image from multiple low dynamic range (LDR) inputs is the ghosting artifacts caused by the object motion across different inputs. When the object motion is slight, most existing methods can well suppress the ghosting artifacts through aligning LDR inputs based on optical flow or detecting anomalies among them. However, they often fail to produce satisfactory results in practice, since the real object motion can be very large. In this study, we present a novel deep framework, termed NHDRRnet, which adopts an alternative direction and attempts to remove ghosting artifacts by exploiting the non-local correlation in inputs. In NHDRRnet, we first adopt an Unet architecture to fuse all inputs and map the fusion results into a low-dimensional deep feature space. Then, we feed the resultant features into a novel global non-local module which reconstructs each pixel by weighted averaging all the other pixels using the weights determined by their correspondences. By doing this, the proposed NHDRRnet is able to adaptively select the useful information (e.g., which are not corrupted by large motions or adverse lighting conditions) in the whole deep feature space to accurately reconstruct each pixel. In addition, we also incorporate a triple-pass residual module to capture more powerful local features, which proves to be effective in further boosting the performance. Extensive experiments on three benchmark datasets demonstrate the superiority of the proposed NDHRnet in terms of suppressing the ghosting artifacts in HDR reconstruction, especially when the objects have large motions.
['Q. Yan and L. Zhang and Y. Liu and Y. Zhu and J. Sun and Q. Shi and Y. Zhang']
2020-02-10
null
null
null
null
['hdr-reconstruction']
['computer-vision']
[ 9.19405296e-02 -5.85230768e-01 1.15717329e-01 -6.59567714e-02 -4.00810540e-01 -2.07258448e-01 4.59625185e-01 -4.88853216e-01 -2.43927956e-01 6.71414196e-01 3.16202521e-01 3.82616878e-01 -2.14595869e-01 -7.07671165e-01 -5.51130474e-01 -1.06857312e+00 2.82459706e-01 -1.33952081e-01 3.29449415e-01 -1.71665460e-01 1.75089553e-01 5.77224791e-01 -1.72256947e+00 7.38212243e-02 1.01243675e+00 9.77766335e-01 3.67224157e-01 3.21051270e-01 1.51318774e-01 1.07481408e+00 -4.89526898e-01 1.33317813e-01 5.34957647e-01 -4.32065606e-01 -4.50574040e-01 1.68476805e-01 5.47094762e-01 -6.58698201e-01 -6.89961731e-01 1.11760724e+00 5.56308508e-01 4.76278931e-01 1.16297066e-01 -8.59221160e-01 -5.28275669e-01 1.79730952e-01 -9.71329570e-01 4.16493744e-01 3.12601537e-01 5.16450107e-01 6.66103363e-01 -1.01564074e+00 7.05665827e-01 1.43488598e+00 3.97653222e-01 4.66537565e-01 -1.29233825e+00 -6.91636503e-01 1.06017590e-01 2.98178256e-01 -1.40251434e+00 -4.89837319e-01 1.03206778e+00 -2.48898193e-01 5.66736042e-01 3.05705339e-01 5.99361122e-01 9.96536136e-01 2.43961215e-01 6.57673657e-01 1.10731149e+00 3.40058878e-02 -2.92269029e-02 -3.25003564e-01 2.91527417e-02 4.76263553e-01 2.16018841e-01 2.04680726e-01 -6.40372097e-01 1.20193973e-01 1.02242184e+00 2.78186768e-01 -8.83804440e-01 -2.70551533e-01 -1.62043655e+00 3.60950321e-01 7.89426506e-01 4.53784645e-01 -4.95480418e-01 7.54225478e-02 1.63555294e-01 1.66799769e-01 2.37117916e-01 2.29365170e-01 -2.45018512e-01 2.69324511e-01 -8.50137770e-01 1.09610669e-01 2.17676893e-01 4.56828505e-01 9.23252583e-01 1.39684036e-01 -2.96921104e-01 8.89223218e-01 3.45370591e-01 4.32232738e-01 6.11319661e-01 -1.05514169e+00 5.52350044e-01 4.21409994e-01 3.57596725e-01 -1.40504980e+00 -4.01961505e-01 -5.27198434e-01 -1.31309438e+00 3.44434500e-01 3.16008955e-01 2.98538692e-02 -9.19048190e-01 1.73872769e+00 4.53391403e-01 5.42676032e-01 2.33084932e-01 1.50879037e+00 8.36123586e-01 9.16324615e-01 -1.58813834e-01 -4.14821655e-01 9.78946328e-01 -8.76872122e-01 -9.32906449e-01 -2.02845737e-01 1.48127183e-01 -7.74036169e-01 7.55035043e-01 3.43138844e-01 -9.01113272e-01 -1.02494335e+00 -1.15273964e+00 -3.22935611e-01 2.91681588e-02 2.55270842e-02 3.06628138e-01 -8.52999315e-02 -7.87815154e-01 7.22945571e-01 -7.37281680e-01 7.87608176e-02 1.77247390e-01 1.97242618e-01 -4.23152983e-01 -4.62396353e-01 -1.24467206e+00 7.65050113e-01 3.83998066e-01 7.27819145e-01 -7.20732510e-01 -6.63680375e-01 -8.23251665e-01 -1.08918965e-01 3.61004859e-01 -6.90747321e-01 5.11693120e-01 -1.00319004e+00 -1.29672587e+00 3.15623224e-01 -2.19013453e-01 -5.09862229e-02 7.70492315e-01 -2.74566531e-01 -4.90765184e-01 1.28635868e-01 8.29530582e-02 5.65616071e-01 1.03437340e+00 -1.35172331e+00 -5.78995585e-01 -3.54197174e-01 -2.20783442e-01 4.20958430e-01 -2.21928716e-01 -3.15408051e-01 -6.17601573e-01 -8.39486957e-01 3.92373383e-01 -8.53364766e-01 -2.92927116e-01 8.40562806e-02 -2.59520531e-01 9.35234204e-02 1.03604865e+00 -7.11429894e-01 1.19472170e+00 -2.33498120e+00 2.90666491e-01 1.15994662e-01 3.81789327e-01 4.63366807e-01 -2.65127838e-01 -4.33874428e-02 -5.81173450e-02 -3.22592139e-01 -2.56595671e-01 -1.41314283e-01 -4.54678178e-01 2.28805885e-01 -2.90580601e-01 8.04848969e-01 8.53098258e-02 8.22401106e-01 -1.02796853e+00 -2.75647104e-01 7.95868993e-01 8.41507435e-01 -8.71021673e-02 2.23615125e-01 1.36550501e-01 9.10617828e-01 -5.33119261e-01 3.39175880e-01 1.07878542e+00 -1.95238948e-01 -1.12090729e-01 -5.33073545e-01 -3.04923534e-01 -9.50203985e-02 -1.57790637e+00 1.78493440e+00 -3.59147549e-01 6.26655817e-01 -1.00207143e-02 -6.70283139e-01 1.03854847e+00 -2.14442667e-02 8.90584707e-01 -9.41721976e-01 4.28264253e-02 2.78756917e-01 2.10576952e-02 -4.86003846e-01 4.03324306e-01 1.28518909e-01 2.32463285e-01 1.86849311e-01 -1.87813610e-01 3.60217363e-01 -4.04076315e-02 -6.39451966e-02 1.02609360e+00 6.61227405e-02 3.00492257e-01 -4.71538939e-02 9.18191373e-01 -4.19585764e-01 9.34594691e-01 5.63817561e-01 -3.02314460e-01 1.00083947e+00 -3.17368860e-04 -7.08526134e-01 -9.15307283e-01 -9.69679534e-01 -1.13147564e-01 5.08406818e-01 7.87573814e-01 -7.25187361e-02 -2.47577861e-01 -5.35626292e-01 -7.15023056e-02 2.77977049e-01 -5.30486226e-01 -3.04642081e-01 -8.92482936e-01 -8.78346562e-01 1.13545738e-01 3.61329079e-01 8.85812640e-01 -1.04160202e+00 -7.58907318e-01 3.58199149e-01 -5.56216121e-01 -1.22351635e+00 -5.63764393e-01 -5.35795689e-02 -7.80697763e-01 -8.85244608e-01 -9.32180941e-01 -5.68498254e-01 6.09334171e-01 8.11548114e-01 7.18533814e-01 2.38410994e-01 -2.78236687e-01 -1.67625904e-01 -2.85249442e-01 3.20250183e-01 -6.78191632e-02 -3.44869673e-01 -2.92501152e-02 5.05458474e-01 2.02101339e-02 -4.17254001e-01 -9.41039860e-01 5.33297777e-01 -1.17137158e+00 1.51195183e-01 5.68503916e-01 9.31916833e-01 6.33457005e-01 3.80780995e-01 3.91684622e-01 -4.99933362e-01 1.61396176e-01 -2.89298475e-01 -5.09269178e-01 8.07343498e-02 -2.92300642e-01 1.12788312e-01 8.80412936e-01 -5.23858249e-01 -1.23713815e+00 1.97277457e-01 -8.08097273e-02 -8.93690288e-01 -5.88462986e-02 -3.29774879e-02 -2.40367740e-01 -1.93582907e-01 3.25262338e-01 4.00643915e-01 -8.53181109e-02 -4.04433250e-01 2.65644908e-01 3.45809639e-01 7.06491411e-01 -1.11977190e-01 1.15985990e+00 6.94714010e-01 1.43240497e-01 -6.92493975e-01 -6.94232464e-01 -5.77278018e-01 -4.73191082e-01 -3.66865098e-01 8.23969126e-01 -1.10516930e+00 -6.36195719e-01 8.49737406e-01 -9.18514013e-01 -5.66822514e-02 -3.04044187e-01 6.00771129e-01 -3.68387818e-01 5.33368409e-01 -5.15760422e-01 -5.51453829e-01 -3.54470700e-01 -1.23724139e+00 9.29483116e-01 5.65677822e-01 2.00144097e-01 -6.58394277e-01 1.82663146e-02 1.57197371e-01 4.62413490e-01 3.99719536e-01 4.53058392e-01 -8.05360079e-02 -8.72705996e-01 1.66845545e-01 -3.76848280e-01 4.11049515e-01 2.92049676e-01 -1.81618556e-02 -9.25077319e-01 -4.44119245e-01 1.77083269e-01 -7.53415748e-02 1.07029486e+00 4.94751900e-01 1.07752407e+00 -1.09133795e-01 -1.16327949e-01 9.64819491e-01 1.78306377e+00 4.38590571e-02 8.79507840e-01 4.08027083e-01 1.08752418e+00 5.11505961e-01 9.19224203e-01 5.04241467e-01 7.10870400e-02 8.31291318e-01 5.77075243e-01 -3.56584281e-01 -4.68310237e-01 5.06228302e-03 4.74252760e-01 6.54315472e-01 -2.45394528e-01 -1.30158916e-01 -4.42789823e-01 3.85188639e-01 -2.04125905e+00 -9.77741539e-01 -3.21986914e-01 2.20077562e+00 6.35160267e-01 -7.00078532e-02 -2.96880126e-01 7.36095682e-02 8.20125163e-01 6.72217369e-01 -7.53106773e-01 2.34594569e-01 -4.39727962e-01 -6.06378764e-02 4.24834967e-01 3.42273593e-01 -1.02655435e+00 6.76474512e-01 4.90408182e+00 5.76613307e-01 -1.28680527e+00 -4.11546417e-02 5.02046943e-01 -1.38852075e-01 -1.72617316e-01 -1.20822668e-01 -6.82067037e-01 6.06200576e-01 3.54541421e-01 7.52177089e-02 5.07233024e-01 4.80564564e-01 4.45746750e-01 -1.88172564e-01 -7.73341596e-01 1.17716730e+00 7.37072155e-02 -1.13600039e+00 -2.45827679e-02 -8.81558657e-02 9.88912702e-01 1.54525379e-03 1.27949223e-01 -3.96424271e-02 1.01759017e-01 -7.82059133e-01 6.04994833e-01 8.29037428e-01 6.25717998e-01 -8.51602256e-01 8.45317602e-01 2.58894861e-01 -1.25476575e+00 -2.41647869e-01 -5.25096416e-01 2.10550025e-01 1.69203252e-01 8.32383335e-01 -1.99224725e-01 7.88246036e-01 9.61956978e-01 1.18715954e+00 -5.96138179e-01 1.02427578e+00 -8.85716528e-02 -7.01453909e-02 -2.30348647e-01 5.45382679e-01 1.20009378e-01 -3.06728542e-01 7.30515599e-01 8.50911736e-01 3.92836183e-01 1.77573249e-01 8.75950828e-02 7.95220852e-01 -2.70706490e-02 -1.58812374e-01 -5.96394479e-01 6.02670550e-01 2.04805464e-01 1.42633557e+00 -5.24262309e-01 -2.33999714e-01 -4.83352274e-01 1.21684408e+00 1.03830040e-01 6.15880013e-01 -8.86755407e-01 -4.00870621e-01 8.41789603e-01 -7.62394816e-02 4.13596064e-01 -4.98072952e-02 -3.37961353e-02 -1.55858195e+00 2.38630623e-01 -8.28512132e-01 2.55409539e-01 -8.42139661e-01 -1.15831530e+00 5.42694211e-01 -3.82796615e-01 -1.65135837e+00 -9.68680084e-02 -1.87270463e-01 -4.06798840e-01 9.02051747e-01 -1.82106245e+00 -7.58660197e-01 -7.98092902e-01 8.96632075e-01 6.03330135e-01 1.91481262e-01 1.77512228e-01 5.52415907e-01 -7.07974613e-01 2.50043392e-01 1.58324242e-01 6.67710826e-02 8.68311465e-01 -9.39043045e-01 3.90684009e-02 1.25900745e+00 -7.14345425e-02 5.33540189e-01 5.91678560e-01 -5.56935072e-01 -1.47189307e+00 -1.38419604e+00 3.82862031e-01 -4.64028902e-02 3.13020349e-01 -1.03501089e-01 -1.28910983e+00 4.49009150e-01 -1.11408740e-01 7.55658627e-01 -5.00126858e-04 -5.95039010e-01 -2.16768831e-01 -5.27988970e-01 -1.03475296e+00 4.32901919e-01 1.08338058e+00 -2.67051786e-01 -3.59551847e-01 -3.22812311e-02 7.14622140e-01 -4.63680774e-01 -9.41293776e-01 6.23429954e-01 4.42083359e-01 -1.22955859e+00 1.16013086e+00 3.84521857e-02 4.24028397e-01 -9.11490023e-01 -5.77716455e-02 -1.15553558e+00 -4.89530563e-01 -5.38616180e-01 -3.22251320e-01 1.28845584e+00 -3.46079886e-01 -6.38815165e-01 3.82298172e-01 3.74754131e-01 -5.25398366e-02 -5.09422660e-01 -8.80877495e-01 -4.94704276e-01 -4.87821668e-01 -6.18346110e-02 5.07869244e-01 9.32474971e-01 -7.39471912e-01 1.31533682e-01 -8.29820275e-01 3.16041321e-01 8.86571527e-01 3.26952815e-01 7.73712814e-01 -1.02128875e+00 -1.76643401e-01 -1.66358843e-01 -5.47187209e-01 -1.14602828e+00 1.22084588e-01 -5.59764385e-01 4.11778808e-01 -1.30430853e+00 1.70091555e-01 -3.97147834e-01 -5.96833229e-01 2.99501717e-01 -5.06748974e-01 4.61472034e-01 2.95769125e-01 5.01213908e-01 -5.71656048e-01 7.36804307e-01 1.65654063e+00 -2.82461774e-02 -3.75699580e-01 -1.84921086e-01 -4.65962619e-01 6.66019380e-01 5.14114082e-01 -3.15053165e-01 -2.83639669e-01 -5.37782490e-01 -2.62706071e-01 1.46196544e-01 5.21582544e-01 -1.24321032e+00 1.45177439e-01 -2.04066843e-01 1.01263845e+00 -6.41759038e-01 1.42853618e-01 -1.03261924e+00 5.04512429e-01 3.83496463e-01 -1.46532521e-01 -1.19325191e-01 -6.21951707e-02 7.21039295e-01 -4.51668024e-01 2.14242518e-01 1.23720610e+00 -2.14887131e-02 -1.02973318e+00 4.39398676e-01 -1.33961644e-02 -2.20852360e-01 1.10023892e+00 -2.15909228e-01 -4.38442171e-01 -1.28162101e-01 -3.98344904e-01 2.03695580e-01 6.21370018e-01 5.71436048e-01 9.21445251e-01 -1.38028049e+00 -5.89395761e-01 5.21559298e-01 -8.35030898e-03 2.26219088e-01 7.08177745e-01 9.82130110e-01 -3.97555351e-01 1.52864665e-01 -5.10346830e-01 -7.45174170e-01 -1.00046623e+00 6.17310345e-01 3.82961303e-01 -2.55737007e-01 -1.10578287e+00 4.73193765e-01 3.71012956e-01 8.89560506e-02 1.21092513e-01 -1.94736868e-01 -3.62272739e-01 7.06108063e-02 8.40847015e-01 4.61077422e-01 8.34044768e-04 -9.63368356e-01 -4.70134139e-01 1.00773227e+00 -5.62127307e-02 1.46829262e-01 1.35477400e+00 -5.92815161e-01 -7.44402185e-02 2.66898870e-01 1.41009402e+00 -4.76908460e-02 -1.67141747e+00 -3.66240233e-01 -3.82445484e-01 -9.39082086e-01 2.65520394e-01 -3.63115162e-01 -1.75323701e+00 6.23307109e-01 8.02085102e-01 -1.68076575e-01 1.69995439e+00 -3.17949623e-01 1.05121207e+00 7.59720281e-02 2.04030022e-01 -8.05437684e-01 1.95719495e-01 2.73031056e-01 7.57559538e-01 -1.22970510e+00 1.26360938e-01 -1.70243934e-01 -6.07381344e-01 1.30986774e+00 7.62290120e-01 -4.03855532e-01 3.11115086e-01 2.56541395e-03 1.67857468e-01 9.40744057e-02 -5.61062992e-01 -3.28581810e-01 2.40402997e-01 5.02021134e-01 -7.56235374e-03 -3.28806877e-01 -2.06213281e-01 -8.16885978e-02 3.27597141e-01 6.93883076e-02 5.92996895e-01 6.37948096e-01 -3.15911233e-01 -6.91557646e-01 -5.07320642e-01 2.69963622e-01 -3.80033225e-01 2.44674012e-01 1.05074145e-01 6.55875206e-01 2.27111429e-01 8.38527679e-01 4.70801108e-02 -5.35602331e-01 4.15282786e-01 -6.00084007e-01 2.26682365e-01 -1.76812977e-01 -4.50187862e-01 3.09274912e-01 -3.85406315e-01 -1.10996795e+00 -6.87009811e-01 -5.54879129e-01 -1.31308866e+00 -3.25643182e-01 -2.44136691e-01 -2.99610466e-01 2.79256314e-01 7.08319306e-01 2.31559962e-01 6.71185553e-01 9.60830092e-01 -9.54237938e-01 -3.74063730e-01 -6.64715767e-01 -6.63579285e-01 7.35405982e-01 9.11857963e-01 -7.24340200e-01 -6.05938911e-01 -5.02288155e-02]
[10.95154094696045, -1.91062593460083]
77a5d721-782b-43fd-9a6b-35716e054a6c
block-bilinear-superdiagonal-fusion-for
1902.00038
null
http://arxiv.org/abs/1902.00038v2
http://arxiv.org/pdf/1902.00038v2.pdf
BLOCK: Bilinear Superdiagonal Fusion for Visual Question Answering and Visual Relationship Detection
Multimodal representation learning is gaining more and more interest within the deep learning community. While bilinear models provide an interesting framework to find subtle combination of modalities, their number of parameters grows quadratically with the input dimensions, making their practical implementation within classical deep learning pipelines challenging. In this paper, we introduce BLOCK, a new multimodal fusion based on the block-superdiagonal tensor decomposition. It leverages the notion of block-term ranks, which generalizes both concepts of rank and mode ranks for tensors, already used for multimodal fusion. It allows to define new ways for optimizing the tradeoff between the expressiveness and complexity of the fusion model, and is able to represent very fine interactions between modalities while maintaining powerful mono-modal representations. We demonstrate the practical interest of our fusion model by using BLOCK for two challenging tasks: Visual Question Answering (VQA) and Visual Relationship Detection (VRD), where we design end-to-end learnable architectures for representing relevant interactions between modalities. Through extensive experiments, we show that BLOCK compares favorably with respect to state-of-the-art multimodal fusion models for both VQA and VRD tasks. Our code is available at https://github.com/Cadene/block.bootstrap.pytorch.
['Rémi Cadene', 'Hedi Ben-Younes', 'Matthieu Cord', 'Nicolas Thome']
2019-01-31
null
null
null
null
['visual-relationship-detection']
['computer-vision']
[-2.38181978e-01 -3.28163326e-01 -3.13208662e-02 -4.69689578e-01 -1.17333841e+00 -8.57502937e-01 8.07749033e-01 2.90814489e-01 -2.32992351e-01 2.30190679e-01 6.06479585e-01 -3.69289458e-01 -1.95212334e-01 -4.47552502e-01 -6.48166597e-01 -6.49733067e-01 -2.78660059e-01 3.46629143e-01 -1.26471937e-01 -4.03230667e-01 -1.74801067e-01 2.91119218e-01 -1.59033859e+00 8.98495913e-01 5.39766550e-01 1.20083022e+00 -1.97730333e-01 8.09049487e-01 1.06914043e-01 1.07870007e+00 -6.51760250e-02 -7.10285664e-01 6.88676462e-02 -4.24290970e-02 -9.26102698e-01 -1.13777593e-01 9.60458755e-01 -4.13064361e-01 -6.95670068e-01 6.45817637e-01 4.82352495e-01 2.25670487e-01 6.84206247e-01 -1.50232720e+00 -6.91287518e-01 7.35853910e-01 -6.79996490e-01 8.15058798e-02 5.47825396e-01 6.13995269e-02 1.59592438e+00 -1.05395722e+00 3.86585057e-01 1.64592326e+00 6.82275176e-01 2.26335645e-01 -1.43505156e+00 -3.11950803e-01 5.85940816e-02 2.68930316e-01 -1.32637000e+00 -5.00934124e-01 5.43769717e-01 -7.31623292e-01 8.36515903e-01 5.52972913e-01 2.65899211e-01 1.03063202e+00 -6.30640835e-02 1.17516649e+00 9.96751964e-01 -1.24266088e-01 -7.57599175e-02 -2.11304605e-01 4.25500631e-01 9.71557140e-01 -2.43203074e-01 -1.05166540e-01 -8.55629385e-01 -5.42978525e-01 5.51111698e-01 1.49100274e-01 -2.85870075e-01 -7.36368537e-01 -1.58808708e+00 8.24288726e-01 6.87435448e-01 2.46660933e-01 -1.54216304e-01 6.69210672e-01 4.75422800e-01 3.04300308e-01 2.38292933e-01 1.17416099e-01 -3.63953441e-01 -1.65392399e-01 -5.48424721e-01 4.03682739e-01 7.40039468e-01 5.67196369e-01 7.68489659e-01 -3.90757561e-01 -4.17563617e-01 1.09219217e+00 4.92066264e-01 3.87085021e-01 -1.22132808e-01 -1.18476021e+00 5.52743673e-01 7.79878497e-01 -1.54756457e-01 -9.99235988e-01 -5.05632937e-01 -2.14089900e-01 -9.92393076e-01 -1.03662070e-03 5.45387626e-01 6.43975884e-02 -8.93855095e-01 1.91420293e+00 3.22120219e-01 -1.68182552e-01 -8.10447559e-02 1.08799863e+00 1.20159042e+00 6.24329507e-01 1.81084916e-01 3.32365751e-01 1.80331278e+00 -9.16444063e-01 -5.39000034e-01 3.23092490e-02 7.08645105e-01 -7.65748560e-01 9.70531583e-01 2.31114775e-01 -1.29963815e+00 -2.98805505e-01 -7.25286424e-01 -8.08240414e-01 -4.94256616e-01 7.79857859e-02 1.12879252e+00 4.33583260e-01 -1.30066299e+00 2.95652092e-01 -8.88990402e-01 -3.96667957e-01 3.62394333e-01 4.51301873e-01 -7.52402246e-01 -3.22017670e-01 -1.09404802e+00 7.88833737e-01 1.05425030e-01 2.90352941e-01 -9.28989351e-01 -8.09959650e-01 -9.73915815e-01 1.91872448e-01 2.25621462e-01 -1.20752335e+00 1.30544221e+00 -5.00807643e-01 -1.06133366e+00 9.15189624e-01 -2.09233731e-01 -8.31478238e-02 4.05958831e-01 -1.62991479e-01 -4.94043753e-02 1.93202719e-01 -1.97323471e-01 9.19894695e-01 7.55293369e-01 -1.36192274e+00 -5.22818267e-01 -4.20578003e-01 6.11669958e-01 2.60714591e-01 -2.67588317e-01 -1.24893747e-02 -6.56130433e-01 -4.31580633e-01 5.72178438e-02 -9.08019781e-01 -4.10775505e-02 2.33612746e-01 -2.93384314e-01 -4.44708586e-01 6.61196530e-01 -7.46760368e-01 9.48369145e-01 -2.15882659e+00 8.13091218e-01 2.11647391e-01 7.38874197e-01 -2.03853235e-01 -4.33186948e-01 7.44039297e-01 -2.99655385e-02 1.68267414e-02 -3.62045705e-01 -7.75281668e-01 4.92398560e-01 3.54982823e-01 -3.40095401e-01 3.54095101e-01 2.35851184e-01 1.08738542e+00 -7.71491528e-01 -3.21424365e-01 1.19424559e-01 8.42679739e-01 -6.39984131e-01 2.80982375e-01 -8.63566995e-02 2.35808641e-01 -1.39720470e-01 9.76958454e-01 7.09165454e-01 -5.55302143e-01 2.95012414e-01 -7.89676070e-01 1.80610120e-01 2.27592066e-01 -9.84944642e-01 1.89833331e+00 -3.75835627e-01 6.17494166e-01 4.02455389e-01 -7.07230985e-01 4.04776454e-01 2.10545972e-01 3.72139454e-01 -4.79108572e-01 2.38539979e-01 4.81227450e-02 -2.53298789e-01 -3.51628751e-01 6.94244742e-01 6.98968694e-02 -2.71686673e-01 4.97668475e-01 4.55948591e-01 2.17465609e-02 3.97836834e-01 6.41327500e-01 9.93397892e-01 -5.37530482e-02 -2.75871269e-02 -6.15750551e-02 3.63959014e-01 -2.40057558e-01 -1.18149579e-01 6.12124801e-01 1.89988792e-01 5.52559793e-01 9.18142557e-01 -3.16121906e-01 -7.63939559e-01 -1.27080870e+00 -1.97986364e-02 1.69870591e+00 1.74608547e-02 -6.74211562e-01 -3.27125639e-01 -5.76297104e-01 3.28257382e-01 1.00402780e-01 -9.02466297e-01 5.88362478e-02 -3.27532589e-01 -7.17000306e-01 6.87299907e-01 5.43763757e-01 1.61275119e-01 -3.30682039e-01 -1.62902027e-01 -3.77946854e-01 -5.40810227e-01 -1.09272027e+00 -4.27296549e-01 -4.88700904e-03 -7.40903795e-01 -1.06818080e+00 -7.51631618e-01 -5.25930285e-01 4.37070251e-01 5.14435828e-01 1.40206861e+00 9.90656912e-02 -1.15042843e-01 1.00795496e+00 -3.10622603e-01 1.29829437e-01 -1.41664326e-01 1.61306292e-01 -2.21673787e-01 2.13174403e-01 7.58423358e-02 -5.49640536e-01 -6.40298665e-01 1.70396671e-01 -1.27471459e+00 2.46049210e-01 5.29766560e-01 8.80295753e-01 3.73262465e-01 -5.34428775e-01 -4.49115895e-02 -3.90938103e-01 7.52422988e-01 -6.45848393e-01 -3.67272586e-01 4.11043108e-01 7.03416318e-02 2.60072470e-01 -1.76231675e-02 -3.69660765e-01 -6.08031034e-01 -1.34800121e-01 -2.38638464e-03 -6.20602369e-01 1.47754937e-01 9.53644574e-01 -4.94111814e-02 -1.74554944e-01 5.01910806e-01 -2.63827324e-01 2.44249552e-02 -6.54185176e-01 9.64326024e-01 3.08827549e-01 6.04870200e-01 -8.75241220e-01 7.38113761e-01 5.91758549e-01 1.03628471e-01 -6.05441213e-01 -7.32672930e-01 -5.72170019e-01 -4.83252406e-01 -1.58137709e-01 7.32271314e-01 -1.18301427e+00 -1.23782289e+00 2.75448620e-01 -1.17837512e+00 -1.69482291e-01 -1.00907190e-02 2.51886964e-01 -2.71382809e-01 5.97921968e-01 -7.45989561e-01 -6.46162927e-01 -1.91873282e-01 -1.21600604e+00 1.44934952e+00 -7.81182125e-02 -8.57948810e-02 -1.04282272e+00 1.75208956e-01 8.68493199e-01 3.83276582e-01 3.31766069e-01 1.05579138e+00 -2.91989923e-01 -7.44470358e-01 -6.77470267e-02 -6.03696108e-01 3.23367834e-01 -2.67508745e-01 -3.28482352e-02 -1.08736241e+00 -3.68507445e-01 -4.57722127e-01 -6.44103408e-01 1.39693975e+00 6.95445314e-02 1.00481248e+00 -1.73404187e-01 -1.26087904e-01 5.77301383e-01 1.17607772e+00 -5.61589241e-01 4.56999898e-01 -1.06139496e-01 1.20369232e+00 6.05471492e-01 1.09331533e-01 4.31833655e-01 1.09654701e+00 7.11807966e-01 6.30960822e-01 -2.81335056e-01 -7.73727149e-02 -4.99758050e-02 3.64608854e-01 8.81307840e-01 -2.00164765e-01 -1.40774012e-01 -1.22234559e+00 5.76357961e-01 -2.27250528e+00 -7.29971349e-01 -1.21999353e-01 1.92895460e+00 7.01450706e-01 -5.37072778e-01 3.22616249e-01 -1.62738815e-01 3.11762035e-01 2.22444028e-01 -1.84594885e-01 -3.25799525e-01 -2.77909577e-01 -7.31741451e-03 2.34540984e-01 6.68306947e-01 -1.27377951e+00 6.47520661e-01 6.14246988e+00 6.81707501e-01 -8.96304309e-01 2.47631222e-01 4.95363355e-01 -2.13104531e-01 -6.63874447e-01 5.01747150e-03 -3.72902751e-01 -3.48918140e-03 6.97364271e-01 3.07929873e-01 6.82673335e-01 3.00618708e-01 -3.17033827e-01 -3.45723666e-02 -1.51447701e+00 1.39150810e+00 7.69392252e-02 -1.38487148e+00 2.66916007e-01 -5.75614497e-02 5.45792460e-01 2.54898101e-01 3.91266882e-01 4.58468974e-01 3.98582578e-01 -1.16807461e+00 7.40856469e-01 6.72346294e-01 4.99336839e-01 -6.08509183e-01 5.94370604e-01 -1.21278331e-01 -1.23138678e+00 -1.31794304e-01 -1.62245914e-01 1.22411653e-01 1.16714202e-01 5.45719862e-01 -4.61374283e-01 6.81795478e-01 7.78569221e-01 6.29801631e-01 -8.14289808e-01 8.14378977e-01 1.30299330e-01 2.50386715e-01 -4.79849756e-01 3.30370277e-01 3.42831045e-01 -9.94971246e-02 4.88729686e-01 1.38314748e+00 2.76375175e-01 -2.99037211e-02 4.07916307e-02 7.96818018e-01 -4.43096995e-01 -4.24216315e-02 -4.90261972e-01 -3.07070911e-01 1.35867208e-01 1.49364507e+00 -1.76183730e-01 -1.86323315e-01 -4.62642133e-01 6.64881408e-01 6.33827448e-01 5.24566233e-01 -8.81680489e-01 1.17222235e-01 8.92167151e-01 -1.69828147e-01 3.54559422e-01 -5.43699145e-01 -2.75235355e-01 -1.55310297e+00 4.61170822e-02 -1.13736343e+00 8.30125928e-01 -8.36462855e-01 -1.69799399e+00 4.33048725e-01 2.28805780e-01 -1.03993690e+00 -1.22402653e-01 -8.20801735e-01 -1.24507941e-01 7.71485090e-01 -1.34369588e+00 -1.80366373e+00 -4.09721017e-01 9.42774236e-01 -1.45133346e-01 1.28514946e-01 7.74159074e-01 4.29677337e-01 -4.20854330e-01 7.40875125e-01 3.97619382e-02 3.72239873e-02 7.43282080e-01 -1.28563368e+00 4.96774763e-02 5.00760317e-01 2.04948738e-01 7.87244141e-01 6.16728544e-01 -1.25474036e-02 -2.06611013e+00 -5.65862298e-01 7.72183895e-01 -7.55875945e-01 1.06606364e+00 -5.79869092e-01 -8.22050810e-01 6.85434103e-01 5.14809489e-01 -2.22150199e-02 1.05109119e+00 6.30316615e-01 -1.14306366e+00 -1.77719831e-01 -7.39002585e-01 7.00284421e-01 8.62584949e-01 -1.02805960e+00 -2.96635151e-01 3.69095862e-01 6.63214564e-01 -4.14768279e-01 -1.29557085e+00 5.54552436e-01 7.25918770e-01 -1.02641022e+00 1.07929742e+00 -6.02375090e-01 5.05029738e-01 -4.89035428e-01 -6.80171490e-01 -1.07327509e+00 -3.71074855e-01 -5.83993196e-01 -6.03620827e-01 1.12420344e+00 4.22528327e-01 -3.33917141e-01 3.14995438e-01 7.59902000e-01 1.16633527e-01 -8.82224798e-01 -9.42559242e-01 -2.54901797e-01 2.04056092e-02 -4.48839605e-01 4.86806065e-01 1.21815658e+00 1.29818320e-01 6.33595705e-01 -4.22821283e-01 1.23116791e-01 5.68655908e-01 2.36672536e-01 8.67668092e-01 -1.03176010e+00 -5.24743080e-01 -7.80079424e-01 -4.50923622e-01 -1.07227707e+00 -5.06247059e-02 -1.08792567e+00 -3.55990112e-01 -1.68726206e+00 4.77268636e-01 -1.63637549e-01 -4.98165071e-01 8.78453612e-01 -1.33374020e-01 4.97608751e-01 5.05853534e-01 2.09865972e-01 -8.96113992e-01 6.04705691e-01 1.10166931e+00 -4.71490443e-01 5.30791655e-02 -4.91142571e-01 -9.02025402e-01 4.65833068e-01 5.23415089e-01 6.84745461e-02 -3.05506527e-01 -9.60247040e-01 6.71805263e-01 7.92365298e-02 7.62962341e-01 -5.36774039e-01 2.18565077e-01 1.34065181e-01 2.09167033e-01 -6.36083066e-01 8.11185837e-01 -7.44070768e-01 8.69911909e-02 -7.09296986e-02 -5.38702250e-01 2.06011400e-01 3.96256685e-01 3.88795972e-01 -4.01954591e-01 3.77924174e-01 4.91142094e-01 2.10408986e-01 -6.27977252e-01 3.54600489e-01 -1.16792567e-01 -1.72058225e-01 5.43005109e-01 1.63601965e-01 -7.24195004e-01 -6.09253645e-01 -7.20858574e-01 5.08228660e-01 2.78239667e-01 5.91442108e-01 6.44449532e-01 -1.64638734e+00 -7.89278150e-01 -2.50547111e-01 3.63218367e-01 -2.80143917e-01 6.42625391e-01 1.28413022e+00 -3.61349791e-01 2.55995125e-01 -9.49679166e-02 -8.33841205e-01 -1.42427182e+00 5.04398346e-01 2.68570721e-01 -4.17846143e-01 -1.09054185e-01 9.31141853e-01 4.52856243e-01 -7.10713804e-01 4.40116346e-01 -4.00909990e-01 -6.34742081e-02 3.81445110e-01 5.00615597e-01 3.90013963e-01 -3.11727747e-02 -7.32360482e-01 -5.33177555e-01 3.73358041e-01 -1.38847023e-01 -2.01479822e-01 1.20204651e+00 -7.13544786e-02 -7.04912186e-01 5.61424851e-01 1.48104656e+00 -2.60616064e-01 -9.98024285e-01 -3.92667145e-01 -1.65583640e-01 -3.98588806e-01 1.01622701e-01 -8.95885646e-01 -1.02408314e+00 1.07080281e+00 7.03907669e-01 2.77150869e-01 1.06499803e+00 3.34497184e-01 6.65342152e-01 5.18164933e-01 9.47131217e-02 -4.26794976e-01 1.01443194e-01 6.63284600e-01 1.13278854e+00 -1.24038005e+00 5.41798770e-02 -2.42634699e-01 -8.16172123e-01 9.88357663e-01 2.45473325e-01 1.66045412e-01 6.49257898e-01 9.19788927e-02 8.15866441e-02 -4.19545412e-01 -9.86122131e-01 -3.43591154e-01 8.58068824e-01 3.47116917e-01 5.83527327e-01 2.53496379e-01 1.03690147e-01 2.87550539e-01 1.53534487e-02 -3.06473255e-01 1.50930956e-01 8.47731054e-01 -1.27430737e-01 -1.17464113e+00 -4.54857945e-01 2.81876087e-01 -3.36654603e-01 -1.67137474e-01 -5.15361726e-01 5.89756310e-01 5.82891144e-02 8.85050476e-01 -6.07971586e-02 -5.84479332e-01 1.95779756e-01 -3.45919877e-02 7.39772320e-01 -1.76917225e-01 -5.96731186e-01 3.57759884e-03 3.22987080e-01 -7.04449177e-01 -5.78954279e-01 -5.39137721e-01 -7.68518567e-01 -7.16320753e-01 6.00517988e-02 -1.83341615e-02 5.78775942e-01 8.28036964e-01 5.04284382e-01 2.26295933e-01 3.40971559e-01 -1.07208693e+00 -4.09500957e-01 -8.42519045e-01 -3.54906976e-01 5.06315351e-01 6.14570260e-01 -7.56123722e-01 -1.59246385e-01 -2.87189074e-02]
[10.755874633789062, 1.54763925075531]
96f0aeef-6f16-473b-8df6-bc928a8dc4b9
stochastic-pitch-prediction-improves-the
2305.17724
null
https://arxiv.org/abs/2305.17724v1
https://arxiv.org/pdf/2305.17724v1.pdf
Stochastic Pitch Prediction Improves the Diversity and Naturalness of Speech in Glow-TTS
Flow-based generative models are widely used in text-to-speech (TTS) systems to learn the distribution of audio features (e.g., Mel-spectrograms) given the input tokens and to sample from this distribution to generate diverse utterances. However, in the zero-shot multi-speaker TTS scenario, the generated utterances lack diversity and naturalness. In this paper, we propose to improve the diversity of utterances by explicitly learning the distribution of fundamental frequency sequences (pitch contours) of each speaker during training using a stochastic flow-based pitch predictor, then conditioning the model on generated pitch contours during inference. The experimental results demonstrate that the proposed method yields a significant improvement in the naturalness and diversity of speech generated by a Glow-TTS model that uses explicit stochastic pitch prediction, over a Glow-TTS baseline and an improved Glow-TTS model that uses a stochastic duration predictor.
['Emmanuel Vincent', 'Vincent Colotte', 'Sewade Ogun']
2023-05-28
null
null
null
null
['zero-shot-multi-speaker-tts']
['audio']
[ 2.22466290e-01 1.11581467e-01 5.36511913e-02 -3.57497543e-01 -1.01940119e+00 -4.36698139e-01 6.44228458e-01 -6.22736476e-02 -6.16157707e-03 6.47718787e-01 6.37579918e-01 -1.89808980e-01 1.94210902e-01 -6.10539079e-01 -4.81049567e-01 -9.66383159e-01 1.88736152e-02 3.90644044e-01 6.34191707e-02 -2.66558677e-01 1.78789988e-01 2.29382485e-01 -1.96193218e+00 1.19260594e-01 7.59319425e-01 7.16856539e-01 4.85660404e-01 1.45277190e+00 -2.51947045e-01 8.05277884e-01 -1.19608426e+00 3.11858989e-02 9.39810649e-02 -9.84368145e-01 -3.30465525e-01 7.00912401e-02 1.61089197e-01 -3.29487920e-01 -1.95351347e-01 7.90170908e-01 6.59383059e-01 6.20580375e-01 9.47490692e-01 -9.19885874e-01 -2.87443876e-01 8.56478631e-01 -9.04362351e-02 2.62499750e-01 3.81680697e-01 2.37344280e-01 9.95062172e-01 -7.30332077e-01 3.50021213e-01 1.41377842e+00 3.36276144e-01 6.89519227e-01 -1.23003697e+00 -7.87808836e-01 -1.82774514e-01 -1.14357218e-01 -1.35287666e+00 -8.43844831e-01 8.56620908e-01 -4.67012137e-01 8.37085187e-01 2.83713877e-01 5.92784047e-01 1.23412836e+00 1.87050149e-01 7.37198770e-01 5.60345829e-01 -8.11452210e-01 6.30215704e-01 1.16746323e-02 -3.37162346e-01 3.13459903e-01 -5.05683005e-01 4.29554015e-01 -9.09807265e-01 -2.06241161e-01 6.57233775e-01 -4.59601671e-01 -1.14651643e-01 1.95422679e-01 -9.09309864e-01 1.02598810e+00 -2.42648751e-01 2.66476214e-01 -4.65826362e-01 3.05765271e-01 3.58775496e-01 1.68864712e-01 4.93766159e-01 2.81905353e-01 6.77503273e-02 -6.00518823e-01 -1.49068058e+00 6.62283182e-01 1.07366765e+00 1.01390803e+00 5.63030541e-01 8.25070858e-01 -5.39511859e-01 7.71425366e-01 3.87281358e-01 9.32224154e-01 7.16685414e-01 -9.84333575e-01 4.45558488e-01 -3.49973381e-01 6.77677318e-02 -5.15287220e-01 1.41207948e-01 -2.32533216e-01 -5.14684796e-01 1.08552620e-01 2.86409557e-01 -5.16586244e-01 -1.08477283e+00 1.77420843e+00 3.30337673e-01 3.85070056e-01 2.97066033e-01 5.71311653e-01 4.79307860e-01 1.27698302e+00 -3.15740556e-02 -4.51643199e-01 9.27414656e-01 -8.95603299e-01 -1.06846678e+00 3.10707837e-02 2.72108406e-01 -1.02492034e+00 1.28639936e+00 3.37865919e-01 -1.14701724e+00 -8.85875762e-01 -9.86560166e-01 3.40939432e-01 5.01896665e-02 -1.82289451e-01 -1.46419689e-01 1.09369135e+00 -7.96905339e-01 5.59444487e-01 -7.55734384e-01 1.89558342e-01 -1.48740172e-01 -1.35525176e-02 3.76668304e-01 3.98182631e-01 -1.31611788e+00 4.25374687e-01 5.32161713e-01 -4.65131044e-01 -1.19772208e+00 -9.50402260e-01 -8.95518899e-01 3.53793383e-01 9.63397920e-02 -3.47826451e-01 1.74493086e+00 -5.83589375e-01 -2.21785045e+00 -1.10433713e-01 -4.97978836e-01 -5.76642275e-01 2.51434237e-01 -2.12475657e-01 -4.83249068e-01 3.13740045e-01 -2.36762792e-01 7.25573838e-01 1.25587845e+00 -1.10709488e+00 -7.77184963e-01 3.64023536e-01 -5.12213826e-01 3.25591624e-01 -2.17669994e-01 -2.38533646e-01 1.22504883e-01 -8.72830808e-01 -2.29792163e-01 -8.95912230e-01 -1.49927139e-01 -5.77533126e-01 -4.93775308e-01 -2.33821750e-01 8.62729490e-01 -6.61860704e-01 1.50743091e+00 -2.29055262e+00 2.73473151e-02 6.20454475e-02 -3.60343486e-01 3.46800447e-01 -1.11358970e-01 7.06048667e-01 2.50937611e-01 -1.07499920e-01 -1.52643412e-01 -6.23672247e-01 8.77605900e-02 2.92307317e-01 -7.89313018e-01 -4.37332056e-02 2.11707339e-01 4.72743630e-01 -1.00033224e+00 -4.54185188e-01 4.29931939e-01 7.82547057e-01 -6.95088506e-01 7.05384135e-01 -5.42333841e-01 6.15980685e-01 -5.41682467e-02 2.03562900e-02 1.80269897e-01 4.38719898e-01 -1.36150569e-01 1.72453016e-01 -2.15096280e-01 5.56773245e-01 -1.17778301e+00 1.57101488e+00 -7.93032467e-01 6.68748379e-01 -2.96788841e-01 -4.34881270e-01 1.36610329e+00 8.39807212e-01 2.10172534e-01 -2.17984423e-01 9.54908058e-02 1.31310225e-01 2.95552313e-01 -3.59555215e-01 7.82669902e-01 -4.90723401e-01 -2.08971277e-02 5.05989015e-01 4.14104104e-01 -7.60731757e-01 3.10767710e-01 5.15797064e-02 5.82148433e-01 -4.70082797e-02 7.70394132e-02 -1.89017877e-02 5.14399469e-01 -5.02332747e-01 2.51065165e-01 6.87820911e-01 1.66525260e-01 8.66937399e-01 2.50600189e-01 4.09079120e-02 -1.37206864e+00 -1.30643845e+00 9.56024006e-02 1.20808923e+00 -4.14606363e-01 -5.05065978e-01 -1.02466786e+00 -8.72206688e-02 -2.86743194e-01 1.44847846e+00 -1.93882167e-01 -2.76978999e-01 -6.97917402e-01 -2.15920553e-01 8.42790425e-01 3.02724600e-01 -6.93741217e-02 -1.28976011e+00 -6.07911527e-01 6.74551964e-01 -3.03057373e-01 -9.85763967e-01 -9.30355608e-01 1.87066883e-01 -6.46267354e-01 -1.36981294e-01 -9.79425132e-01 -5.54199159e-01 1.40960872e-01 -2.75839537e-01 7.41434693e-01 -6.19675636e-01 -1.64307222e-01 7.99579844e-02 -5.89886904e-01 -6.45916283e-01 -1.28659332e+00 1.37992740e-01 1.56362116e-01 1.37579560e-01 -5.53255752e-02 -6.33877516e-01 -3.51980448e-01 2.92395577e-02 -1.03948486e+00 -6.11362569e-02 1.69105291e-01 8.86133909e-01 4.13447738e-01 2.03382120e-01 9.98964548e-01 -6.05699062e-01 8.26485515e-01 -3.51059943e-01 -4.07895446e-01 5.35555296e-02 -4.15853322e-01 2.02481404e-01 9.57284510e-01 -7.87985921e-01 -1.37284255e+00 -5.03871180e-02 -5.47323823e-01 -6.35084093e-01 -3.11220437e-01 3.16217184e-01 -5.11848880e-03 5.60070813e-01 7.36097395e-01 5.50391436e-01 4.58046198e-02 -3.51745456e-01 6.37595773e-01 1.00519109e+00 7.69355714e-01 -6.30553961e-01 7.00986743e-01 -1.11221626e-01 -3.01011831e-01 -1.27258623e+00 -6.16393805e-01 -3.66667271e-01 -3.33704203e-01 -1.74280792e-01 6.22763455e-01 -7.80872703e-01 -4.12744045e-01 4.15815622e-01 -1.20185542e+00 -2.37847686e-01 -7.01896489e-01 7.49144554e-01 -1.10576451e+00 2.52164245e-01 -5.60494602e-01 -1.58409858e+00 -5.16145825e-01 -9.54675019e-01 1.10195088e+00 4.76234138e-01 -5.18682361e-01 -9.57967639e-01 4.25204933e-01 -2.85014566e-02 5.11742353e-01 1.29898012e-01 8.68018031e-01 -5.57989657e-01 -2.14342877e-01 -1.90450419e-02 6.67538762e-01 4.85010445e-01 5.28056324e-01 3.14682275e-01 -1.28036320e+00 -1.35583624e-01 6.07387014e-02 -1.32621199e-01 4.46129173e-01 6.60584092e-01 7.04364657e-01 -6.51773691e-01 1.49377704e-01 4.22167540e-01 1.02841640e+00 8.06473076e-01 5.90141475e-01 -5.49285948e-01 4.00386631e-01 6.51225626e-01 5.35830915e-01 8.15181196e-01 -2.49297917e-02 4.65107799e-01 -2.79638786e-02 2.07000285e-01 -3.54576349e-01 -6.11809313e-01 5.46639025e-01 1.33820939e+00 9.91002694e-02 -5.74357808e-01 -6.03768110e-01 7.34698057e-01 -1.38332331e+00 -1.26766849e+00 3.55924636e-01 2.37530136e+00 1.21154118e+00 2.46114388e-01 4.31784689e-01 5.23608029e-01 8.71621549e-01 3.16064566e-01 -4.59742874e-01 -8.45408976e-01 7.31872097e-02 5.87039053e-01 6.25778213e-02 8.29638004e-01 -6.57528698e-01 9.15470719e-01 6.58501911e+00 1.14540124e+00 -1.23303044e+00 -2.16362000e-01 3.68088216e-01 -2.06099078e-01 -4.35952485e-01 -2.99423099e-01 -1.07061064e+00 4.83266175e-01 1.74594295e+00 -7.63164341e-01 5.37700057e-01 5.85693777e-01 5.83806753e-01 9.09596607e-02 -1.10765135e+00 7.77062953e-01 -2.64731925e-02 -1.25964975e+00 1.22317471e-01 -1.31495014e-01 7.56851494e-01 -4.59587783e-01 4.28632408e-01 3.47061306e-01 3.03016514e-01 -1.02275419e+00 1.11440170e+00 3.85005444e-01 8.51584673e-01 -1.02310717e+00 2.77682632e-01 5.11165500e-01 -1.22759271e+00 2.05427200e-01 -1.98536068e-01 1.22936167e-01 6.38674021e-01 5.56538939e-01 -1.59114540e+00 3.82697523e-01 1.28474295e-01 7.26017589e-03 1.80106610e-01 9.08557236e-01 -1.42955348e-01 1.27257109e+00 -2.35062867e-01 -2.84402668e-01 4.39127922e-01 1.86198488e-01 6.94554627e-01 1.59766865e+00 6.22631729e-01 -2.98004765e-02 1.58498988e-01 9.74453211e-01 2.46001586e-01 1.16881169e-01 -4.40399051e-01 -4.01338369e-01 8.99316967e-01 7.32148707e-01 -3.63069028e-01 -4.54009563e-01 1.95310768e-02 7.09503472e-01 -4.53257769e-01 5.41880071e-01 -8.55823278e-01 -8.32495570e-01 5.72392106e-01 -2.24167164e-02 6.23291552e-01 -3.21727604e-01 -1.73935611e-02 -5.81145644e-01 -4.76139337e-01 -7.40813971e-01 -1.19700938e-01 -5.97891331e-01 -1.04400182e+00 7.59241462e-01 1.82982340e-01 -1.23070991e+00 -1.38608181e+00 6.53523952e-02 -8.12232018e-01 1.41109252e+00 -1.23943424e+00 -6.72585368e-01 3.34302306e-01 3.63658160e-01 1.13167715e+00 -3.25253218e-01 9.52049851e-01 -1.70268685e-01 -2.65212413e-02 6.12071991e-01 2.34922439e-01 -6.60313070e-02 5.11316419e-01 -1.26663613e+00 7.67631650e-01 5.59190631e-01 3.89913946e-01 3.98408622e-01 1.23520553e+00 -4.93534029e-01 -9.62841153e-01 -1.12851405e+00 9.75190520e-01 8.87051374e-02 3.07539880e-01 -4.32570428e-01 -9.53387439e-01 2.82506436e-01 5.54552555e-01 -3.65931064e-01 8.65352094e-01 -2.42709324e-01 -6.51143864e-02 1.01906836e-01 -1.08179545e+00 5.01871049e-01 3.77559304e-01 -5.48991382e-01 -8.55219066e-01 -2.76810192e-02 1.10616946e+00 -4.91624951e-01 -7.26663113e-01 -8.70797485e-02 5.47606170e-01 -9.43831801e-01 7.64592230e-01 -2.58824468e-01 2.13589609e-01 -2.63810188e-01 -3.03273261e-01 -1.85431159e+00 -3.92779224e-02 -1.12601674e+00 -3.11576784e-01 1.50508678e+00 4.65256214e-01 -3.11988682e-01 6.76957428e-01 1.94612846e-01 -2.74718702e-01 -3.41737360e-01 -9.53671515e-01 -1.09189320e+00 1.13657296e-01 -3.98837566e-01 7.23595023e-01 2.38999695e-01 5.42622176e-04 3.79341483e-01 -5.27734637e-01 1.91006497e-01 5.32142401e-01 -1.46557838e-01 7.22940803e-01 -8.27174366e-01 -7.07386792e-01 -2.62846231e-01 -2.21087560e-02 -1.21789002e+00 -2.16550641e-02 -4.42867428e-01 7.59205520e-01 -1.03477418e+00 -5.38096130e-01 -1.70555100e-01 -1.31534636e-02 -1.62757173e-01 -3.58929634e-01 -3.36077839e-01 5.68727911e-01 -1.41023770e-01 7.06447735e-02 1.00965357e+00 1.11819804e+00 -2.14147754e-02 -6.30664349e-01 3.23325366e-01 -1.42650977e-01 4.70949233e-01 7.44755983e-01 -6.53826237e-01 -9.05339837e-01 2.26810724e-02 -5.10489643e-01 6.21422589e-01 -1.85124382e-01 -1.15520775e+00 2.93273836e-01 -3.07613730e-01 8.29784796e-02 -9.18408573e-01 6.40468895e-01 -2.07533196e-01 2.04625785e-01 5.35653770e-01 -6.78241611e-01 -2.35828176e-01 1.31706089e-01 4.58486497e-01 -4.22481030e-01 -5.13296723e-01 8.35844994e-01 5.73259499e-03 -1.37767166e-01 6.19692765e-02 -8.34539592e-01 2.46681169e-01 6.45968437e-01 -1.97458863e-01 9.93938148e-02 -8.28497112e-01 -5.84282815e-01 -3.28002125e-01 -1.43673465e-01 4.02501106e-01 6.72117412e-01 -1.29419470e+00 -9.13113058e-01 4.45364565e-01 -1.78661972e-01 5.52791208e-02 1.78734556e-01 5.32198772e-02 -1.95260644e-01 5.33599555e-01 8.25807676e-02 -6.79925382e-01 -1.02986443e+00 2.54091054e-01 9.57840309e-02 -3.18793431e-02 -4.74223912e-01 8.20770264e-01 3.43394697e-01 -1.35408208e-01 4.58001643e-01 -4.71016377e-01 -8.42230767e-02 -7.86533356e-02 7.19099343e-01 4.33267921e-01 -1.38621643e-01 -6.16356373e-01 5.84482513e-02 3.64514351e-01 1.77600145e-01 -9.82438207e-01 9.00923431e-01 -2.31477499e-01 7.08806098e-01 1.00138843e+00 1.06466889e+00 3.58815074e-01 -1.46940243e+00 -5.06923683e-02 -1.16158113e-01 -4.64603782e-01 -7.44840130e-02 -5.97182512e-01 -6.30153418e-01 1.13073564e+00 2.54578292e-01 4.13480222e-01 9.17361200e-01 -2.51288176e-01 1.08939981e+00 6.96662515e-02 1.76836982e-01 -9.28705931e-01 3.12666655e-01 7.92666972e-01 8.15633178e-01 -5.65726876e-01 -6.60540760e-01 -2.13607788e-01 -8.74685764e-01 1.31548548e+00 2.40218669e-01 -4.97682579e-02 6.16781950e-01 5.07308722e-01 2.93756783e-01 5.05914748e-01 -9.53817248e-01 -1.22435123e-01 2.04819605e-01 8.07735443e-01 5.76152384e-01 1.54392615e-01 8.43898728e-02 1.94255814e-01 -8.91187370e-01 -2.32431337e-01 6.23622835e-01 6.06936336e-01 -8.89297724e-01 -1.20210469e+00 -6.76126659e-01 1.12863071e-01 -2.60799408e-01 -3.24963003e-01 8.15581381e-02 2.52658635e-01 -5.98542280e-02 1.24284649e+00 3.44000429e-01 -3.74264151e-01 2.67561615e-01 5.72797656e-01 4.01082933e-01 -8.24678123e-01 -5.53330958e-01 7.07983077e-01 3.53314471e-03 1.90915421e-01 -3.98176117e-03 -8.32910061e-01 -1.43001068e+00 -2.41605893e-01 -4.50933248e-01 5.22172689e-01 8.26335490e-01 7.13530600e-01 3.95884626e-02 8.52049649e-01 1.14646244e+00 -8.17726970e-01 -8.75388622e-01 -1.22093475e+00 -7.41976500e-01 2.14291334e-01 6.65512025e-01 -2.18363509e-01 -6.05402112e-01 4.99898702e-01]
[15.205803871154785, 6.348029136657715]
61a7f2fa-ab0e-4267-989d-f902c353dd01
meshwalker-deep-mesh-understanding-by-random
2006.05353
null
https://arxiv.org/abs/2006.05353v3
https://arxiv.org/pdf/2006.05353v3.pdf
MeshWalker: Deep Mesh Understanding by Random Walks
Most attempts to represent 3D shapes for deep learning have focused on volumetric grids, multi-view images and point clouds. In this paper we look at the most popular representation of 3D shapes in computer graphics - a triangular mesh - and ask how it can be utilized within deep learning. The few attempts to answer this question propose to adapt convolutions & pooling to suit Convolutional Neural Networks (CNNs). This paper proposes a very different approach, termed MeshWalker, to learn the shape directly from a given mesh. The key idea is to represent the mesh by random walks along the surface, which "explore" the mesh's geometry and topology. Each walk is organized as a list of vertices, which in some manner imposes regularity on the mesh. The walk is fed into a Recurrent Neural Network (RNN) that "remembers" the history of the walk. We show that our approach achieves state-of-the-art results for two fundamental shape analysis tasks: shape classification and semantic segmentation. Furthermore, even a very small number of examples suffices for learning. This is highly important, since large datasets of meshes are difficult to acquire.
['Alon Lahav', 'Ayellet Tal']
2020-06-09
null
null
null
null
['3d-object-recognition', '3d-classification']
['computer-vision', 'computer-vision']
[-7.61202946e-02 2.70718962e-01 2.30785072e-01 -2.68844843e-01 -3.66953462e-01 -6.29638553e-01 5.58633566e-01 3.38860840e-01 -1.67725384e-01 2.10713610e-01 -2.56619602e-01 -2.83818990e-01 1.06209978e-01 -1.33033776e+00 -1.03925288e+00 -6.43380105e-01 -1.64137945e-01 1.00427830e+00 2.45802104e-01 -1.96477637e-01 3.68999869e-01 1.25798619e+00 -1.56548464e+00 4.65823680e-01 2.39763916e-01 1.01767170e+00 2.49128506e-01 5.81219792e-01 -6.53094053e-01 1.43129811e-01 -1.96212187e-01 -3.81327212e-01 1.26228154e-01 8.80910549e-03 -1.12460411e+00 9.25929099e-02 4.44898337e-01 1.55984731e-02 -9.71541107e-02 9.48094070e-01 4.22895461e-01 -6.84408296e-04 7.95656323e-01 -7.22411215e-01 -5.64457655e-01 3.90765876e-01 -4.85993743e-01 9.02060345e-02 1.42744486e-03 -1.01546273e-01 9.14820433e-01 -1.09344304e+00 9.48012650e-01 1.37504232e+00 1.02976823e+00 5.70533216e-01 -1.30551648e+00 -3.02298009e-01 8.42127353e-02 -2.65535057e-01 -1.23772693e+00 -8.52448717e-02 1.07074153e+00 -5.61166823e-01 1.06534326e+00 2.32437462e-01 9.71356750e-01 7.85167575e-01 2.37850294e-01 8.65159929e-01 7.96511173e-01 -1.70971543e-01 3.55112672e-01 -1.57499015e-01 -4.00764644e-02 1.13432968e+00 1.58891454e-02 -3.08040559e-01 -4.69606854e-02 -2.06964627e-01 1.25257063e+00 1.47598628e-02 -2.82531716e-02 -6.74365878e-01 -9.58370090e-01 8.51864755e-01 8.18331361e-01 4.14799660e-01 -4.14217263e-01 4.51737702e-01 3.86307895e-01 1.73204482e-01 8.09743822e-01 4.31145966e-01 -5.28864145e-01 2.15316713e-01 -9.50474858e-01 4.10981208e-01 9.04358149e-01 6.36543751e-01 9.70057905e-01 3.73697765e-02 2.39187390e-01 6.99486315e-01 3.09140831e-01 2.05188975e-01 -6.00827411e-02 -8.39425921e-01 1.75399050e-01 8.58190715e-01 -7.01034069e-02 -1.11884892e+00 -6.61662936e-01 -3.76916230e-01 -1.02651286e+00 6.75096035e-01 4.33335125e-01 1.06978096e-01 -1.17947841e+00 1.22956240e+00 3.00091237e-01 3.02058756e-01 -3.90425503e-01 7.49080837e-01 1.02866578e+00 6.44210875e-01 -7.77234584e-02 4.13335919e-01 1.30451024e+00 -5.94478071e-01 -2.95252930e-02 8.39116871e-02 4.31588382e-01 -5.74160397e-01 9.03510392e-01 2.84914434e-01 -1.47774220e+00 -5.46976388e-01 -8.69103134e-01 -1.75827533e-01 -6.45731032e-01 5.91595173e-02 5.74506402e-01 3.07765394e-01 -1.31777012e+00 1.25501204e+00 -1.04175115e+00 -3.56151044e-01 9.94661152e-01 4.58040237e-01 -1.11504897e-01 1.71787724e-01 -6.75378084e-01 6.64003730e-01 -8.13912824e-02 1.14553511e-01 -7.47004986e-01 -6.69181406e-01 -9.14185345e-01 2.33016655e-01 4.25641611e-02 -9.84608114e-01 1.08007061e+00 -8.39647293e-01 -1.35145867e+00 1.47884703e+00 -2.77139872e-01 -4.05777872e-01 5.34422994e-01 1.34445131e-01 2.58743554e-01 2.41257660e-02 -1.09804504e-01 7.79218256e-01 9.73305643e-01 -1.73318589e+00 -8.98722261e-02 -5.87263167e-01 8.08127820e-02 -1.28684103e-01 3.33727866e-01 -2.42740825e-01 -5.03177881e-01 -6.09676361e-01 5.08697748e-01 -8.38606596e-01 -6.17533207e-01 2.40411490e-01 -4.91868228e-01 -5.86480498e-01 7.33902633e-01 -2.78490275e-01 6.78261340e-01 -1.92764151e+00 3.82452071e-01 4.39736456e-01 3.55607867e-01 2.17165321e-01 -2.98770983e-02 3.22239608e-01 -1.08078495e-01 5.65633059e-01 -4.94411796e-01 -6.97179198e-01 -1.21416256e-01 3.69755000e-01 -3.23736250e-01 4.90503967e-01 4.67139632e-01 1.30403471e+00 -7.14547336e-01 -1.58535451e-01 2.46514425e-01 7.41005599e-01 -5.61665416e-01 6.68925270e-02 -6.48432434e-01 6.10265851e-01 -5.06953359e-01 4.87378508e-01 8.77302766e-01 -5.87480605e-01 -8.11028630e-02 -2.08874911e-01 -2.08344907e-01 1.47539571e-01 -1.04377997e+00 1.93178666e+00 -4.07656550e-01 4.18255240e-01 1.20008074e-01 -1.09410465e+00 1.10731578e+00 2.17404053e-01 5.63515961e-01 -3.88209611e-01 2.66534895e-01 2.66098708e-01 -4.24699903e-01 -3.92970055e-01 1.68933198e-01 -1.29695445e-01 9.00346786e-02 4.78639871e-01 -8.10957700e-02 -4.12352443e-01 -1.91060394e-01 -2.48405844e-01 7.72575915e-01 2.46015280e-01 -2.21948661e-02 -4.79612231e-01 4.91232932e-01 -5.50483353e-02 1.39580294e-01 5.28596878e-01 4.28636611e-01 9.56608474e-01 6.18672848e-01 -1.10926592e+00 -1.24467599e+00 -1.23520732e+00 -1.67897746e-01 8.55077386e-01 6.23643808e-02 -6.02421686e-02 -8.97275746e-01 -4.53778952e-01 1.47584766e-01 3.61448199e-01 -8.75287235e-01 3.73003632e-01 -1.00676644e+00 -3.17349017e-01 2.42882788e-01 5.76456547e-01 3.41044158e-01 -1.47169733e+00 -9.10175562e-01 3.33431125e-01 3.99043173e-01 -8.23013425e-01 -5.19535691e-02 1.64312989e-01 -1.26457584e+00 -1.11854005e+00 -8.30447316e-01 -1.01896513e+00 8.15849781e-01 -2.33983118e-02 1.53502262e+00 4.14477646e-01 -2.14908734e-01 3.75749081e-01 6.68001473e-02 -3.98087859e-01 -2.24535137e-01 3.59948248e-01 -4.26946968e-01 1.55258566e-01 8.78040045e-02 -9.46411014e-01 -5.33148825e-01 -1.74019322e-01 -7.29969263e-01 -1.18737081e-02 3.75304937e-01 4.25509036e-01 1.00457633e+00 -1.53875366e-01 2.03110993e-01 -1.18323088e+00 3.79340857e-01 -4.94234204e-01 -5.36815047e-01 -1.13676563e-02 -6.13699257e-02 1.02490567e-01 6.21083379e-01 -1.89219952e-01 -5.28839827e-01 2.43171245e-01 -6.93703711e-01 -7.08321631e-01 -5.00804484e-01 4.36717570e-01 8.65217671e-02 -2.26612926e-01 3.98783714e-01 1.16733417e-01 3.89826670e-02 -8.62573385e-01 3.82973373e-01 1.27071254e-02 3.89744729e-01 -6.64785564e-01 5.15463471e-01 9.18923199e-01 3.19441378e-01 -1.02015197e+00 -7.37690210e-01 -1.95810884e-01 -9.84298825e-01 -1.51806548e-01 9.64071333e-01 -4.45907980e-01 -9.73889232e-01 4.35882300e-01 -1.58066976e+00 -5.34950793e-01 -4.25729364e-01 -1.93494171e-01 -8.43649566e-01 9.79327783e-03 -7.23424435e-01 -7.26677835e-01 -3.78239810e-01 -1.15232325e+00 1.36708975e+00 1.43731788e-01 -9.53568667e-02 -1.17403460e+00 -4.95693162e-02 -2.93507069e-01 4.10894603e-01 6.60851002e-01 1.34378564e+00 -3.24436516e-01 -6.82354152e-01 -2.64367610e-02 -3.54255408e-01 3.38057205e-02 -5.99342920e-02 4.12157215e-02 -1.08630550e+00 -2.36867651e-01 3.99154648e-02 -1.79070607e-01 1.02331889e+00 6.08214915e-01 1.94573629e+00 -4.19447273e-02 -4.30203259e-01 8.22620511e-01 1.61190081e+00 -6.27592579e-02 5.86628973e-01 -1.60336599e-03 9.35306311e-01 5.68505883e-01 -3.36755961e-01 2.80970991e-01 3.53159606e-01 4.64074522e-01 7.05604732e-01 -2.81518996e-01 -1.23107120e-01 -1.21165574e-01 -2.89345831e-01 8.18853378e-01 -2.75941968e-01 -1.19715579e-01 -1.17525578e+00 5.20718455e-01 -1.62226784e+00 -5.84441245e-01 -3.32438886e-01 1.87922883e+00 4.68152881e-01 3.29477876e-01 -2.30560265e-02 -1.19577870e-02 4.77471650e-01 2.45110780e-01 -6.34636462e-01 -7.73244500e-01 -1.01945922e-02 7.19185591e-01 2.51738966e-01 4.82779235e-01 -1.09769356e+00 9.78836834e-01 6.11249495e+00 5.39070845e-01 -1.42123413e+00 -6.49121255e-02 8.62820566e-01 4.09543335e-01 -4.96717721e-01 -4.27911639e-01 -4.72801894e-01 1.87934831e-01 6.39768541e-01 3.10817122e-01 3.58178645e-01 7.33913839e-01 -1.52445049e-03 1.38186976e-01 -1.21282721e+00 1.04581928e+00 -1.34977520e-01 -1.89601874e+00 3.07379901e-01 8.31950456e-02 7.26600885e-01 4.17201042e-01 1.12490542e-01 -1.27206519e-01 2.12609455e-01 -1.47570825e+00 6.46195054e-01 8.28021228e-01 8.54850352e-01 -7.98279405e-01 4.91361827e-01 3.72539729e-01 -1.45817113e+00 3.51655960e-01 -5.48108697e-01 9.55448449e-02 6.39394000e-02 5.92560887e-01 -4.91073728e-01 4.80088860e-01 5.93032956e-01 7.01423109e-01 -3.96030545e-01 1.07618141e+00 2.45262403e-02 3.44970822e-01 -4.81064290e-01 3.39242741e-02 6.25516236e-01 -2.40249544e-01 4.99022901e-01 1.29475927e+00 3.23983371e-01 1.37278110e-01 2.48724312e-01 1.37503958e+00 -3.39702219e-01 5.25931418e-02 -9.98677433e-01 1.42754376e-01 1.21566646e-01 1.15139318e+00 -1.36504912e+00 -3.26894879e-01 -2.25033119e-01 8.72100711e-01 7.02126861e-01 3.24784160e-01 -3.19958657e-01 -7.29910061e-02 5.90944886e-01 3.94174933e-01 6.31823838e-01 -5.57693601e-01 -7.52731502e-01 -7.08599210e-01 -2.06706658e-01 -3.24234337e-01 8.08546767e-02 -7.70616353e-01 -1.42146695e+00 6.97932184e-01 -3.45854372e-01 -8.81222665e-01 1.64243817e-01 -8.12269568e-01 -8.99961829e-01 9.04951930e-01 -1.44023240e+00 -1.16488278e+00 -1.69382825e-01 3.93602699e-01 6.25215471e-01 1.27128139e-01 9.72277761e-01 9.48091075e-02 -9.53218490e-02 9.52602550e-02 -1.91884890e-01 3.14896405e-01 -2.49023557e-01 -1.43257666e+00 1.06331003e+00 5.94823770e-02 4.24079448e-01 3.63498926e-01 4.64155197e-01 -6.22049510e-01 -1.52838337e+00 -1.13515425e+00 9.21704352e-01 -6.09015167e-01 3.03394854e-01 -5.42425632e-01 -1.36919153e+00 5.54597676e-01 5.24235256e-02 2.84070134e-01 2.51683086e-01 1.66044347e-02 -2.76128650e-01 3.62435788e-01 -1.16716981e+00 4.99659121e-01 1.17984116e+00 -4.62552816e-01 -4.70739782e-01 2.42142946e-01 5.22390306e-01 -5.92821598e-01 -7.72078037e-01 4.65713292e-01 3.47574592e-01 -1.00926888e+00 1.14760566e+00 -9.70587432e-01 5.89927554e-01 -5.86162508e-02 4.51435819e-02 -1.33808792e+00 -3.44724655e-01 -4.84201819e-01 -1.00590445e-01 6.46979094e-01 2.81266958e-01 -3.75326544e-01 1.26942706e+00 2.16860667e-01 -3.03932965e-01 -1.47610235e+00 -1.02450931e+00 -3.98800284e-01 5.55054963e-01 -5.33374846e-01 7.29460776e-01 9.64306533e-01 -6.53582752e-01 1.37504950e-01 1.85490742e-01 1.83267176e-01 5.22618055e-01 5.53980052e-01 4.68185425e-01 -1.67305207e+00 2.35731155e-01 -8.85813355e-01 -3.22172105e-01 -1.15522790e+00 2.18381748e-01 -1.30958152e+00 -1.91953227e-01 -1.84706903e+00 -2.35327363e-01 -7.99889743e-01 -2.70739365e-02 3.13343704e-01 2.86070734e-01 3.99921954e-01 4.60228696e-02 -2.75951810e-02 -3.62907946e-01 4.25838411e-01 1.52307022e+00 -1.65414497e-01 -2.22255483e-01 2.39698678e-01 -2.01240554e-01 1.16200542e+00 8.21734667e-01 -3.55460584e-01 -4.14827839e-02 -8.40251684e-01 4.56105053e-01 1.42729543e-02 6.35727465e-01 -7.46546865e-01 3.21800947e-01 2.53315479e-01 5.35348475e-01 -9.84013200e-01 4.75034028e-01 -7.68168390e-01 1.70754403e-01 4.02397335e-01 -8.40229690e-02 3.18464726e-01 3.07272673e-01 4.97101098e-01 4.82152328e-02 -2.79350579e-01 8.56186628e-01 -8.16557646e-01 -3.27533782e-01 7.17397690e-01 -1.57557353e-01 3.39537337e-02 6.04352474e-01 -2.98155278e-01 2.55376220e-01 1.06003657e-02 -1.25299561e+00 9.46346074e-02 5.16058803e-01 2.23372325e-01 8.55164886e-01 -1.44049275e+00 -6.14495993e-01 4.12728727e-01 -3.87464225e-01 6.55763507e-01 1.02409266e-01 3.88845056e-01 -7.26325572e-01 2.12969258e-01 -1.91467628e-01 -8.85685205e-01 -8.10725093e-01 3.83354187e-01 6.94108903e-01 -1.71752930e-01 -1.08849204e+00 1.18151140e+00 1.35614097e-01 -6.84262335e-01 2.09990665e-01 -6.42574906e-01 -2.32458442e-01 1.98252574e-01 2.16398016e-01 1.42902315e-01 1.95886999e-01 -5.99572420e-01 -2.11648658e-01 1.11331701e+00 1.98416978e-01 2.56562501e-01 1.70585752e+00 2.26244763e-01 -3.99158150e-01 8.00291657e-01 1.33153903e+00 -2.84451723e-01 -1.22216928e+00 -2.23972917e-01 1.92494884e-01 -2.59872705e-01 -2.76619405e-01 -2.61448652e-01 -1.37803233e+00 1.41365123e+00 3.00173998e-01 6.71966791e-01 5.82936525e-01 3.61855805e-01 9.11305010e-01 3.61304194e-01 4.83099937e-01 -7.31200099e-01 6.03815056e-02 9.02831495e-01 1.11141968e+00 -1.00052667e+00 -2.46590018e-01 -4.30692881e-01 -1.13902204e-01 1.53618646e+00 2.73754835e-01 -7.89924204e-01 1.21053815e+00 2.65555352e-01 -2.44944260e-01 -9.03917253e-01 -4.13192242e-01 -2.53163755e-01 4.20987993e-01 3.28722924e-01 3.00774276e-01 3.80060747e-02 2.76328802e-01 3.19631547e-01 -2.97254473e-01 -2.78730422e-01 2.65452117e-01 6.45033717e-01 -5.93432665e-01 -8.67511153e-01 -7.46102482e-02 6.06300890e-01 -4.18054312e-01 1.24341682e-01 -4.14462119e-01 5.90543330e-01 3.33261907e-01 1.74200416e-01 4.94497389e-01 -9.74192619e-02 4.22727883e-01 2.42956892e-01 6.66321635e-01 -8.87400508e-01 -6.46983862e-01 3.09312344e-02 -2.65694708e-01 -6.18066728e-01 -4.31799501e-01 -5.57722092e-01 -1.56798720e+00 -4.00423557e-01 2.22578973e-01 -2.00186163e-01 7.43064046e-01 9.60377395e-01 3.03069532e-01 5.24562716e-01 5.51264048e-01 -1.57270885e+00 -1.98701859e-01 -4.13105965e-01 -4.81183797e-01 4.62603271e-01 3.87308568e-01 -6.01355016e-01 -1.69686526e-02 -1.55989155e-01]
[8.267056465148926, -3.6855616569519043]
0e38564a-5998-475a-bb5c-e14fc88265ae
an-end-to-end-network-for-co-saliency
1910.11819
null
https://arxiv.org/abs/1910.11819v2
https://arxiv.org/pdf/1910.11819v2.pdf
An End-to-End Network for Co-Saliency Detection in One Single Image
Co-saliency detection within a single image is a common vision problem that has received little attention and has not yet been well addressed. Existing methods often used a bottom-up strategy to infer co-saliency in an image in which salient regions are firstly detected using visual primitives such as color and shape and then grouped and merged into a co-saliency map. However, co-saliency is intrinsically perceived complexly with bottom-up and top-down strategies combined in human vision. To address this problem, this study proposes a novel end-to-end trainable network comprising a backbone net and two branch nets. The backbone net uses ground-truth masks as top-down guidance for saliency prediction, whereas the two branch nets construct triplet proposals for regional feature mapping and clustering, which drives the network to be bottom-up sensitive to co-salient regions. We construct a new dataset of 2,019 natural images with co-saliency in each image to evaluate the proposed method. Experimental results show that the proposed method achieves state-of-the-art accuracy with a running speed of 28 fps.
['Song Wang', 'Zhongyuan Wang', 'Qian Wang', 'Yuanhao Yue', 'Qin Zou', 'Hongkai Yu']
2019-10-25
null
null
null
null
['co-saliency-detection']
['computer-vision']
[ 5.25849283e-01 1.32031664e-01 2.83838660e-02 -5.24704635e-01 -4.70264703e-01 -1.00859590e-01 4.22295600e-01 1.71195790e-01 -3.63715529e-01 4.10140544e-01 7.10508367e-03 1.13299772e-01 2.14498237e-01 -5.08197665e-01 -8.84172916e-01 -4.02369887e-01 -3.92546281e-02 4.98869382e-02 1.16605604e+00 -2.12549359e-01 4.74429309e-01 1.84413984e-01 -1.82354593e+00 3.85231823e-01 1.16532874e+00 1.07021034e+00 8.52462351e-01 3.79819214e-01 1.49710238e-01 5.93491077e-01 -1.76925614e-01 -1.85604930e-01 2.66294420e-01 -4.10820037e-01 -8.11539233e-01 2.35976100e-01 4.96481508e-01 -1.16473392e-01 4.71484959e-02 1.36559713e+00 3.98916185e-01 -9.75172445e-02 2.16312200e-01 -1.20296752e+00 -6.34613931e-01 4.62119877e-01 -9.11117375e-01 4.50593352e-01 -4.28887904e-02 1.25542939e-01 1.13933218e+00 -1.16309464e+00 3.67489308e-01 1.21683681e+00 3.65907103e-01 1.64943546e-01 -1.05004132e+00 -4.61195856e-01 3.92267376e-01 3.39815497e-01 -1.24147999e+00 -1.71821564e-01 1.01021123e+00 -1.81411952e-01 5.99035084e-01 1.01836771e-01 7.02657640e-01 3.62264723e-01 1.63719177e-01 1.17265069e+00 1.18957281e+00 -2.29973078e-01 7.67137036e-02 1.82127729e-01 -1.06468238e-01 8.83506417e-01 2.87024558e-01 9.21738986e-03 -6.64268553e-01 1.02775678e-01 1.01341939e+00 2.16371119e-01 -4.85915579e-02 -7.58730173e-01 -1.46369708e+00 6.86580241e-01 1.15605175e+00 1.02582492e-01 -5.09106040e-01 -3.56481411e-02 2.02658892e-01 -2.06063494e-01 4.52290982e-01 3.13782096e-01 -3.98143351e-01 3.97215933e-01 -1.21417046e+00 2.14822486e-01 1.50103495e-01 8.28865230e-01 1.04904914e+00 -6.97826520e-02 -1.73062757e-01 8.91192615e-01 3.04072648e-01 3.17707717e-01 4.24456507e-01 -6.26155078e-01 3.09777290e-01 8.95104289e-01 1.83149993e-01 -1.29828584e+00 -4.86656398e-01 -7.76135385e-01 -6.20336473e-01 3.10009480e-01 1.64333418e-01 5.76636828e-02 -9.95496154e-01 1.59846270e+00 4.24063444e-01 3.49744707e-01 -1.93703637e-01 1.52048898e+00 7.45117903e-01 2.94881970e-01 4.82579581e-02 1.21588141e-01 1.41376245e+00 -1.63249409e+00 -2.71823406e-01 -5.95042408e-01 1.35571897e-01 -7.87813246e-01 9.69414473e-01 2.95244134e-03 -1.27185059e+00 -8.31472933e-01 -9.83375847e-01 -3.13346356e-01 -2.66520977e-01 2.75396138e-01 5.27548432e-01 1.33200988e-01 -1.32129300e+00 3.52722317e-01 -7.29330361e-01 -3.27847302e-01 7.54177570e-01 2.37423599e-01 -5.45234196e-02 5.63507080e-02 -1.07516229e+00 8.63511086e-01 6.50216937e-01 1.77807957e-01 -1.12235296e+00 -4.88864034e-01 -8.46702993e-01 1.95603326e-01 5.26354134e-01 -7.61666656e-01 1.11024356e+00 -1.34119296e+00 -1.17690015e+00 9.13569927e-01 -3.89302999e-01 -4.73381490e-01 2.88499683e-01 -3.13910306e-01 -2.50480503e-01 3.41460109e-01 5.35961092e-01 1.09600413e+00 1.13883018e+00 -1.48023784e+00 -1.29627144e+00 -2.89573848e-01 -3.98538150e-02 5.43696225e-01 6.16822839e-02 2.95015901e-01 -5.06065488e-01 -6.00204527e-01 4.43448722e-01 -7.85583794e-01 -2.36654803e-01 2.74037272e-02 -6.08783901e-01 -1.30289063e-01 9.06383932e-01 -4.56773818e-01 8.98297548e-01 -1.90252531e+00 9.09779221e-02 -1.18565366e-01 4.21184689e-01 2.54763156e-01 -1.13919780e-01 -6.64302660e-03 -1.41763508e-01 -1.58667922e-01 -4.67910975e-01 -2.99600184e-01 -3.13592076e-01 -2.76436299e-01 -2.99922705e-01 3.86131078e-01 5.74559271e-01 1.00809681e+00 -1.28096938e+00 -7.02276826e-01 4.30782616e-01 2.49418810e-01 -4.85446274e-01 1.94608197e-01 -1.60062477e-01 2.48270601e-01 -4.93030906e-01 9.05750096e-01 6.99761331e-01 -5.30579627e-01 -2.38980949e-01 -1.92840025e-01 -4.26386744e-01 9.82210711e-02 -8.85997593e-01 1.74015248e+00 -8.08426645e-03 5.76263249e-01 2.19160896e-02 -9.33316171e-01 1.01085854e+00 -2.64135689e-01 2.63202339e-01 -6.09469891e-01 5.66969626e-02 3.66145283e-01 4.88229059e-02 -2.29146749e-01 6.81310594e-01 8.93800333e-03 1.67123839e-01 1.35846168e-01 3.57344933e-02 -9.06934123e-03 1.68433730e-02 9.46598575e-02 5.51306009e-01 2.03269094e-01 1.85159683e-01 -4.39892650e-01 5.65406919e-01 1.49958581e-01 6.85612619e-01 5.29306531e-01 -4.27362740e-01 9.69018579e-01 2.33279243e-01 -3.80352378e-01 -8.79356921e-01 -1.01912427e+00 1.40165865e-01 1.37913823e+00 8.73831332e-01 -1.53871462e-01 -7.97515035e-01 -6.25941634e-01 -1.78635374e-01 3.76839936e-01 -6.78244412e-01 -1.22575380e-01 -3.52298260e-01 -4.02143389e-01 -6.85684010e-02 4.06607598e-01 9.50419605e-01 -1.49208629e+00 -1.25843811e+00 1.74810752e-01 -2.58635789e-01 -1.07709730e+00 -6.91563249e-01 3.42348479e-02 -8.59378219e-01 -1.00549603e+00 -9.55098391e-01 -1.20914829e+00 8.95712435e-01 9.89907026e-01 1.09593892e+00 2.61560857e-01 -2.65549034e-01 -1.94771871e-01 -2.41754636e-01 -5.67853808e-01 2.11054355e-01 9.26315412e-02 -1.91127822e-01 4.45551544e-01 2.71503419e-01 -3.79600465e-01 -9.15334225e-01 4.91468251e-01 -8.07199717e-01 6.07503712e-01 9.25059080e-01 8.05222809e-01 7.53421187e-01 -2.13103190e-01 5.54488361e-01 -6.11763716e-01 2.81179279e-01 -3.02038342e-01 -6.11886740e-01 2.51930803e-01 -2.55979389e-01 -2.28460804e-01 4.14742410e-01 2.94725001e-02 -9.50688183e-01 5.00358760e-01 3.16096604e-01 -4.98298794e-01 -1.31817594e-01 5.11527419e-01 -6.44662231e-02 -2.34103233e-01 4.17344034e-01 5.89204311e-01 2.48157848e-02 -2.51124054e-01 3.84637326e-01 4.45546806e-01 8.77018034e-01 -1.22564144e-01 6.53713822e-01 6.29055858e-01 -1.86335444e-01 -7.33083665e-01 -1.32478988e+00 -6.44831836e-01 -8.37345719e-01 -2.94342667e-01 9.58657920e-01 -1.17560303e+00 -2.24931091e-01 6.43000662e-01 -1.05595458e+00 -8.79613534e-02 -3.84424515e-02 1.81363806e-01 -6.47728741e-01 2.42171243e-01 -2.01874614e-01 -5.59958756e-01 -4.89920110e-01 -1.17468143e+00 1.33677948e+00 7.09847748e-01 1.73619702e-01 -6.57465994e-01 -3.03192794e-01 2.13994473e-01 4.16802496e-01 2.69120246e-01 4.11239862e-01 -3.46721619e-01 -8.47715974e-01 1.87597841e-01 -8.52467358e-01 9.98712033e-02 2.69705534e-01 -1.86103925e-01 -1.02162051e+00 -3.45936388e-01 -1.11588597e-01 -2.68017381e-01 1.06222594e+00 4.83065456e-01 1.26102471e+00 1.00190267e-01 -4.19191748e-01 5.94437480e-01 1.41913807e+00 -9.98922437e-02 4.43200022e-01 4.97260034e-01 8.44669878e-01 7.02774644e-01 1.00486529e+00 2.44381502e-01 6.37437344e-01 6.00970030e-01 8.26256931e-01 -6.77371562e-01 -1.17487945e-01 -3.80147010e-01 1.18589200e-01 4.26223069e-01 8.91507119e-02 3.08005452e-01 -7.76753604e-01 1.04557788e+00 -2.02596784e+00 -7.72970676e-01 -2.22631805e-02 2.02298331e+00 8.82345319e-01 3.38321000e-01 3.71706814e-01 -2.90351421e-01 1.07314885e+00 2.11786613e-01 -8.56694579e-01 6.18419722e-02 -2.07723796e-01 -2.48355687e-01 4.31017727e-01 2.24784508e-01 -1.31354332e+00 1.32212770e+00 5.79722357e+00 7.02920258e-01 -1.48486805e+00 3.04819737e-02 8.41511548e-01 1.84239537e-01 -1.16072290e-01 1.34432033e-01 -7.14484990e-01 6.86845005e-01 2.29043871e-01 -1.34555519e-01 1.22674629e-01 9.80173171e-01 1.87580466e-01 -4.37004566e-01 -8.32836628e-01 9.26526964e-01 1.66126147e-01 -1.30037713e+00 -3.61643396e-02 -1.82166815e-01 9.81108785e-01 3.45470637e-01 2.55708545e-01 -1.23201935e-02 2.38210768e-01 -9.45986807e-01 1.03662074e+00 4.65389520e-01 5.99053144e-01 -7.16187835e-01 6.36828363e-01 3.71576697e-01 -1.39260900e+00 -1.86764047e-01 -5.86997449e-01 -5.88742681e-02 1.81330264e-01 6.21384919e-01 -8.14886332e-01 4.90042657e-01 8.26706648e-01 9.98809814e-01 -9.28897679e-01 1.39679027e+00 -2.60531276e-01 4.27420884e-01 -2.63963521e-01 -1.54440347e-02 5.08991420e-01 -9.65545550e-02 4.97042358e-01 1.08014250e+00 6.22910336e-02 -1.52485862e-01 4.14145172e-01 1.16336286e+00 2.22054675e-01 -1.99808151e-01 -3.03776175e-01 2.79869109e-01 4.11771119e-01 1.42612779e+00 -1.19459808e+00 -4.44680929e-01 -3.19461197e-01 1.04447675e+00 4.92775857e-01 2.18632892e-01 -8.38693917e-01 -4.41768736e-01 4.36416656e-01 2.09123194e-01 5.23102760e-01 -5.23382006e-03 -5.69653809e-01 -1.10710430e+00 -2.62054261e-02 -5.59169590e-01 1.46039292e-01 -1.04798269e+00 -1.09264708e+00 5.82864225e-01 -1.61971539e-01 -1.29409599e+00 1.54097944e-01 -3.16922635e-01 -7.26150930e-01 1.04791296e+00 -1.94785368e+00 -1.30000842e+00 -5.79190969e-01 5.27232111e-01 8.19814622e-01 -3.39150764e-02 2.86201745e-01 8.53517745e-03 -4.85738188e-01 2.54918605e-01 -3.33000898e-01 4.46227640e-02 5.44100463e-01 -1.24838257e+00 5.20962954e-01 1.21386552e+00 5.62513769e-02 6.14821255e-01 6.53952599e-01 -7.08255172e-01 -7.76874006e-01 -1.31298459e+00 8.60418677e-01 -2.56417155e-01 5.44593275e-01 -2.37953380e-01 -9.84833062e-01 2.39080086e-01 1.89657912e-01 4.20725524e-01 1.71241537e-02 -1.84452057e-01 -5.78232072e-02 -2.12060392e-01 -1.01774180e+00 6.49735212e-01 9.74029660e-01 -5.46577096e-01 -6.66046977e-01 1.45463869e-01 8.32755208e-01 -3.89057219e-01 -1.54954284e-01 4.33613420e-01 2.89671093e-01 -1.31798673e+00 9.94200110e-01 -2.03367203e-01 6.60915971e-01 -7.48780549e-01 1.82418093e-01 -1.32929802e+00 -6.09811842e-01 -3.30947608e-01 1.88467935e-01 8.46634030e-01 3.33977014e-01 -4.26325947e-01 6.66697443e-01 1.76165625e-01 -3.89313221e-01 -1.09983385e+00 -5.28844297e-01 -4.25581485e-01 -4.33557004e-01 3.32739055e-02 4.65979427e-01 7.76881635e-01 -1.86290845e-01 3.79129022e-01 -2.81175822e-01 3.77073854e-01 7.15106845e-01 5.96822619e-01 5.36944985e-01 -1.26511836e+00 1.09218404e-01 -6.27744019e-01 -4.78005350e-01 -1.34328699e+00 -2.04217359e-01 -7.90444434e-01 4.08645421e-01 -1.65503788e+00 5.20437062e-01 -3.24760854e-01 -6.05370700e-01 5.04976869e-01 -6.07926786e-01 3.80627722e-01 3.28953981e-01 2.45995343e-01 -1.02541590e+00 6.54568970e-01 1.45213747e+00 -8.97982866e-02 -1.71253458e-01 -2.05623671e-01 -9.55432117e-01 8.37375760e-01 7.94967651e-01 -2.62993991e-01 -5.08921027e-01 -2.95648664e-01 -1.30509660e-01 -9.59975943e-02 6.44519329e-01 -1.13165259e+00 3.90202969e-01 -1.75683722e-01 3.95810783e-01 -9.43940163e-01 1.28866583e-01 -6.84768319e-01 -4.77278769e-01 4.57040757e-01 -3.05631459e-01 1.20755315e-01 1.18731938e-01 6.57317817e-01 -4.32347357e-01 1.02549031e-01 1.05319047e+00 -1.82884619e-01 -1.11134398e+00 3.17916304e-01 2.03859851e-01 5.98721765e-02 1.13209355e+00 -4.87531722e-01 -2.35922962e-01 -1.24078922e-01 -4.06819820e-01 4.82693583e-01 6.18190646e-01 6.33654773e-01 1.00710642e+00 -1.14951789e+00 -6.86922967e-01 3.86886120e-01 1.85436442e-01 3.47542197e-01 3.21415514e-01 9.13934767e-01 -4.43092376e-01 3.81848842e-01 -5.47127783e-01 -1.07274044e+00 -1.24285138e+00 6.47644579e-01 2.71482289e-01 7.82499760e-02 -4.22221869e-01 1.09737289e+00 6.84564829e-01 -1.32252455e-01 3.41305090e-03 -4.40882325e-01 -3.67255896e-01 -1.95802122e-01 3.98067325e-01 -5.98457493e-02 -2.11965606e-01 -1.01637554e+00 -3.84849072e-01 6.81748867e-01 -1.55314773e-01 1.01759136e-01 1.20721519e+00 -3.64086062e-01 -2.83321649e-01 4.19829398e-01 9.26777482e-01 -4.15304154e-01 -1.62663615e+00 -4.95598108e-01 3.72466519e-02 -7.05582023e-01 1.44045100e-01 -8.19058359e-01 -1.17125821e+00 8.45662177e-01 6.26340508e-01 1.88549817e-01 1.24441302e+00 2.59950552e-02 8.07296336e-01 -1.30670890e-02 4.78627980e-01 -1.16078198e+00 3.36306125e-01 5.54108202e-01 9.49297011e-01 -1.63460004e+00 -1.35507241e-01 -6.52084470e-01 -8.56733501e-01 6.81668520e-01 9.44880009e-01 -3.74613941e-01 5.45813739e-01 -3.40057194e-01 -1.24424174e-01 -2.51080602e-01 -4.90219474e-01 -5.80898046e-01 5.76999068e-01 4.68660057e-01 1.69596374e-01 1.52251199e-01 -4.75985222e-02 4.51739013e-01 -1.19989216e-01 1.85204577e-02 3.64156514e-01 9.58569825e-01 -9.55282867e-01 -4.39565718e-01 -3.09532613e-01 4.59224761e-01 -3.51057023e-01 -1.49155110e-01 -2.95505792e-01 4.00203526e-01 2.89025664e-01 8.49358916e-01 1.17234528e-01 -3.65126252e-01 1.00585774e-01 -3.80697399e-01 -1.70944966e-02 -8.24945390e-01 -3.69232178e-01 1.18055701e-01 -3.76974523e-01 -6.41327322e-01 -7.24098742e-01 -6.10565960e-01 -1.39998007e+00 3.95442545e-01 -2.92131931e-01 -1.13461845e-01 4.86817598e-01 8.87097657e-01 6.50205553e-01 5.52616179e-01 7.81274259e-01 -1.22373772e+00 -2.18127578e-01 -9.18205798e-01 -5.16322494e-01 4.12074357e-01 6.11081600e-01 -8.73074889e-01 -1.56585991e-01 1.25665113e-01]
[9.794486999511719, -0.3374933898448944]
963d0afa-d26d-4645-a60b-73f73ce20b8f
rethinking-the-learning-paradigm-for-facial
2209.15402
null
https://arxiv.org/abs/2209.15402v1
https://arxiv.org/pdf/2209.15402v1.pdf
Rethinking the Learning Paradigm for Facial Expression Recognition
Due to the subjective crowdsourcing annotations and the inherent inter-class similarity of facial expressions, the real-world Facial Expression Recognition (FER) datasets usually exhibit ambiguous annotation. To simplify the learning paradigm, most previous methods convert ambiguous annotation results into precise one-hot annotations and train FER models in an end-to-end supervised manner. In this paper, we rethink the existing training paradigm and propose that it is better to use weakly supervised strategies to train FER models with original ambiguous annotation.
['Bruno Lepri', 'Nicu Sebe', 'Weijie Wang']
2022-09-30
null
null
null
null
['facial-expression-recognition']
['computer-vision']
[ 1.80397406e-01 3.08027655e-01 -2.67286181e-01 -9.21252668e-01 -7.26188958e-01 -6.79675817e-01 1.95954323e-01 -3.83585691e-01 -5.03038168e-01 1.03309631e+00 -2.98272632e-02 9.05419812e-02 3.72905344e-01 -1.40771449e-01 -3.56310427e-01 -4.68027532e-01 2.29817584e-01 4.66813862e-01 2.76645750e-01 -4.02542561e-01 -2.48407513e-01 2.88158745e-01 -1.40983891e+00 4.85839844e-01 5.51375270e-01 1.19659805e+00 -4.95341241e-01 4.87477988e-01 -4.36008453e-01 1.05932736e+00 -7.65965343e-01 -1.09694564e+00 7.36375973e-02 -4.07286435e-01 -1.15093112e+00 3.98439974e-01 1.87673345e-01 -3.21513236e-01 3.25491140e-03 1.19689310e+00 4.56685871e-01 -9.27833691e-02 2.03337878e-01 -1.56538808e+00 -5.75304270e-01 1.13153338e-01 -5.03565967e-01 -1.65999576e-01 6.43969953e-01 -1.89031869e-01 7.61827707e-01 -1.24058986e+00 7.06326187e-01 1.07493317e+00 8.65320027e-01 1.02420473e+00 -1.10031497e+00 -7.07995415e-01 3.09793502e-01 -4.08938080e-02 -1.78884709e+00 -7.18416750e-01 9.07684565e-01 -3.75389844e-01 4.53090817e-01 2.66468793e-01 4.92518991e-01 1.30382836e+00 -6.88912511e-01 9.06726003e-01 1.27257502e+00 -4.10962313e-01 2.23484308e-01 8.84809196e-02 -2.14519188e-01 6.67915404e-01 -3.76229852e-01 -2.57103682e-01 -5.80258548e-01 -4.22981650e-01 4.80600178e-01 -4.26649563e-02 -7.90475383e-02 -1.95145041e-01 -5.21491230e-01 6.88106537e-01 1.21893443e-01 3.68432611e-01 -3.22352797e-02 -7.05273077e-02 6.68948293e-01 1.65975228e-01 8.18022490e-01 1.89350501e-01 -5.35397112e-01 -2.47742042e-01 -8.10500741e-01 1.92085907e-01 7.71328628e-01 8.56249928e-01 1.06943119e+00 -1.22709675e-02 -1.61784172e-01 1.10535014e+00 2.12682679e-01 9.64120850e-02 4.44714040e-01 -9.68344867e-01 5.20138070e-02 6.30853593e-01 3.29460412e-01 -8.66307080e-01 -2.27812827e-01 1.92641377e-01 -4.69262391e-01 6.83865398e-02 4.94280636e-01 -4.50853199e-01 -8.45326126e-01 1.72712171e+00 3.34053487e-01 3.86540219e-02 -1.33586293e-02 1.17885923e+00 9.45559680e-01 1.00698352e-01 6.91616118e-01 -2.19573990e-01 1.14629054e+00 -1.04617274e+00 -1.18930876e+00 -1.84238791e-01 9.13931668e-01 -8.31919074e-01 1.09139025e+00 3.18801761e-01 -6.82102501e-01 -3.84979963e-01 -6.69530809e-01 -9.09254625e-02 -5.60710669e-01 5.71887314e-01 8.04229915e-01 5.36790133e-01 -9.25209522e-01 1.96501300e-01 -4.46179926e-01 -1.88319251e-01 7.48672664e-01 5.49537838e-01 -7.25220621e-01 1.02111481e-01 -1.23641610e+00 7.46139884e-01 4.26789165e-01 5.85022569e-01 -6.40956700e-01 -1.15483627e-01 -6.96776807e-01 -5.26551187e-01 6.23526752e-01 6.79147094e-02 1.69050634e+00 -2.17704463e+00 -1.88333416e+00 1.35820961e+00 -2.76860058e-01 5.77525422e-02 6.10633969e-01 -2.03666613e-01 -4.53577965e-01 8.17385316e-02 -1.10709824e-01 6.97682977e-01 8.44864368e-01 -1.45799959e+00 -3.41195196e-01 -3.99968177e-01 2.47790635e-01 -9.25980732e-02 -3.05617422e-01 5.31462610e-01 -4.66981411e-01 -6.60926819e-01 -1.84003666e-01 -1.01619995e+00 -1.94311142e-01 2.19720975e-01 -1.35448009e-01 -5.23070633e-01 9.17231381e-01 -3.17418486e-01 1.14236283e+00 -2.22291851e+00 -1.05962031e-01 -6.58344030e-02 4.15852040e-01 6.33392334e-01 -4.81905043e-02 -2.42056269e-02 -2.24507943e-01 3.75770469e-04 -1.76326826e-01 -7.56039202e-01 -2.16972288e-02 6.59916580e-01 -1.83630422e-01 3.07118297e-01 4.62691635e-01 8.53592753e-01 -1.15579093e+00 -8.46343398e-01 -2.54217178e-01 2.53757536e-01 -2.19498396e-01 6.12318456e-01 -2.21111909e-01 5.48216879e-01 -5.66884160e-01 1.13840294e+00 7.66025782e-01 -4.41340469e-02 2.91473955e-01 -1.92213953e-01 2.57745028e-01 -5.09781063e-01 -9.07712579e-01 1.64292943e+00 -1.73380762e-01 6.48825824e-01 3.41009259e-01 -1.03636980e+00 1.01165462e+00 5.95495164e-01 6.73720658e-01 -4.12166417e-01 3.50954771e-01 3.92572790e-01 -4.54163164e-01 -8.68921161e-01 5.24776518e-01 -5.47348976e-01 -1.09654099e-01 5.56066595e-02 4.49594676e-01 1.38233319e-01 -1.20775759e-01 -2.59968340e-01 7.40874946e-01 4.99421299e-01 2.59743273e-01 1.11963779e-01 6.71795487e-01 -1.16404571e-01 8.19453597e-01 1.73681393e-01 -7.60178208e-01 7.30811000e-01 5.91164112e-01 -8.84427786e-01 -6.49560452e-01 -6.43237233e-01 -5.98439872e-02 1.70185041e+00 -8.07965994e-02 -4.74367887e-01 -9.98928428e-01 -1.23734856e+00 -3.49494040e-01 -4.54710238e-02 -7.79175043e-01 1.83509421e-02 -2.43686229e-01 -5.36013484e-01 1.09159756e+00 6.08318329e-01 5.89126825e-01 -8.30319881e-01 -1.61157697e-01 1.17530420e-01 -3.47163916e-01 -1.49252570e+00 -3.22372735e-01 4.25042436e-02 -3.27336401e-01 -1.02413428e+00 -7.99637198e-01 -7.61324704e-01 7.86748171e-01 -1.30710721e-01 1.25292623e+00 3.43405247e-01 -1.28675729e-01 1.50797918e-01 -7.38123655e-01 -7.77019441e-01 -1.04692437e-01 -4.94397767e-02 7.38687515e-02 5.51632583e-01 9.70994532e-01 -2.28014439e-01 -2.73821801e-01 6.32915556e-01 -1.05091405e+00 -3.20724905e-01 1.61459759e-01 8.83946419e-01 6.01902008e-01 -5.59122145e-01 6.59712791e-01 -8.03079188e-01 5.05955279e-01 -2.93692887e-01 -4.52568382e-01 4.91676778e-01 -2.61535048e-01 -3.09173763e-01 4.52480316e-01 -6.84723258e-01 -1.10587835e+00 5.39683461e-01 -3.30982655e-01 -7.02924848e-01 -3.07045966e-01 1.94164693e-01 -2.47053728e-01 -4.90123928e-01 7.65806377e-01 -2.47635797e-01 -1.08595565e-01 -2.71028429e-01 7.85220563e-02 9.56657767e-01 3.88939977e-01 -7.63229847e-01 2.92818159e-01 4.08732921e-01 -3.37359190e-01 -5.47218680e-01 -1.23844290e+00 -3.23942304e-01 -8.53578448e-01 -4.90233481e-01 1.03486764e+00 -1.00206327e+00 -6.70950592e-01 4.62047666e-01 -1.33786607e+00 -3.43152583e-01 -1.16141804e-01 1.55968532e-01 -5.42674661e-01 3.36245328e-01 -3.96100581e-01 -1.03230429e+00 -1.31217986e-01 -1.08166468e+00 1.67999983e+00 2.41854653e-01 -4.21073496e-01 -7.59521306e-01 1.63867220e-01 3.53266060e-01 2.51663625e-01 2.56363153e-01 1.08372808e-01 -8.12935352e-01 1.79127529e-01 -4.48213875e-01 -3.65781397e-01 4.37217146e-01 8.21130201e-02 1.60516962e-01 -1.43881130e+00 2.34488070e-01 -3.56251389e-01 -9.77061331e-01 3.72409940e-01 -3.57259214e-01 1.52467477e+00 -2.20352441e-01 -1.37802362e-01 4.28736031e-01 1.19602334e+00 -9.16324481e-02 6.93398595e-01 6.80529401e-02 5.74600875e-01 8.31517518e-01 7.57815361e-01 4.70506966e-01 1.69227943e-01 8.99495065e-01 2.40965724e-01 -3.67604315e-01 3.70403200e-01 -3.61434102e-01 1.68448374e-01 2.26011083e-01 -3.18574041e-01 -9.59399641e-02 -8.86137784e-01 4.36469585e-01 -2.07276225e+00 -8.23599517e-01 2.10487828e-01 1.74843359e+00 1.31850588e+00 -2.63832569e-01 2.60192186e-01 -1.81090742e-01 6.31563663e-01 1.15279220e-01 -9.78256613e-02 -5.35952568e-01 -3.29843849e-01 1.05471730e-01 1.59674004e-01 4.30408776e-01 -1.20631540e+00 1.28715098e+00 7.21016216e+00 8.59571695e-01 -1.19203758e+00 4.11081582e-01 7.58968949e-01 8.39859247e-02 8.47441480e-02 -7.17397109e-02 -6.89208388e-01 4.86406296e-01 6.00792587e-01 3.39684814e-01 9.30284336e-02 1.10277891e+00 -6.12697452e-02 -1.58679470e-01 -1.03175592e+00 1.28082991e+00 2.09524900e-01 -8.08472872e-01 -1.66762784e-01 -2.93854743e-01 5.71723878e-01 -2.86246628e-01 -4.07529652e-01 4.18716908e-01 2.30337411e-01 -1.26266599e+00 6.16661608e-01 6.20268404e-01 8.50488186e-01 -3.75939757e-01 1.14782703e+00 5.99885769e-02 -9.03044760e-01 1.26266629e-01 -2.29499400e-01 -1.53159514e-01 1.19322456e-01 3.12901944e-01 -7.25048125e-01 2.69151121e-01 7.97147810e-01 3.03728431e-01 -6.28153622e-01 6.41965806e-01 -2.70473003e-01 5.58251858e-01 -2.85695195e-01 -6.93602636e-02 2.06285253e-01 -6.57870546e-02 1.68844387e-01 1.29114485e+00 -1.15531705e-01 3.78671467e-01 4.39074844e-01 5.01444519e-01 -4.28061992e-01 1.73834562e-01 -5.94029546e-01 -6.39970899e-02 1.16019152e-01 1.35268414e+00 -5.65906227e-01 -3.25535864e-01 -4.58390474e-01 1.22200477e+00 6.04540467e-01 5.38476765e-01 -9.04150844e-01 -1.36675134e-01 6.56432629e-01 -4.06930856e-02 -2.19800677e-02 -3.51439901e-02 -3.47142480e-02 -1.11789751e+00 2.99402982e-01 -8.46528828e-01 3.14939380e-01 -9.91408765e-01 -1.46239114e+00 9.35607314e-01 -1.43424854e-01 -1.16992462e+00 -2.16704503e-01 -6.43681705e-01 -3.71858180e-01 5.30507326e-01 -1.49612355e+00 -1.39364564e+00 -4.64006007e-01 6.75756991e-01 5.07964194e-01 -7.38758221e-02 1.11046600e+00 3.92450869e-01 -5.30835867e-01 9.28792477e-01 -4.17958736e-01 6.11873984e-01 9.40287888e-01 -8.92386854e-01 -5.50457835e-01 4.01129007e-01 2.57366132e-02 2.37026632e-01 7.39050984e-01 -3.19410771e-01 -8.98031950e-01 -1.00678504e+00 8.68006527e-01 -5.33055842e-01 7.25289583e-01 -4.66081798e-01 -9.85903740e-01 6.73830867e-01 1.78825811e-01 7.96087205e-01 1.06669211e+00 1.00740604e-01 -6.00078106e-01 -1.76534265e-01 -1.23052156e+00 4.25636441e-01 1.10411227e+00 -8.72378767e-01 -4.71404165e-01 4.51853842e-01 2.77786702e-01 -5.44494331e-01 -7.32359648e-01 5.57106197e-01 5.91884255e-01 -5.96409857e-01 4.19272006e-01 -1.00747490e+00 3.34133774e-01 -3.44019324e-01 -2.76399702e-01 -9.32872534e-01 2.88109750e-01 -8.50063086e-01 1.13715000e-01 1.37641919e+00 3.05114776e-01 -2.33881757e-01 8.69250178e-01 1.01680160e+00 2.24339724e-01 -7.22671866e-01 -1.01922083e+00 -5.20515621e-01 -3.26365680e-01 -1.96254089e-01 6.87627614e-01 1.24505317e+00 2.26858333e-01 2.73452193e-01 -4.78460491e-01 -8.07535127e-02 1.93742901e-01 -2.23926127e-01 7.58866549e-01 -1.26179016e+00 2.47318093e-02 -3.05834115e-01 -6.18589640e-01 -6.80853903e-01 7.99086154e-01 -3.71756911e-01 5.62673628e-01 -6.80761576e-01 -2.83407141e-02 -3.37990731e-01 -7.48968795e-02 9.66483116e-01 -3.55538696e-01 8.16370964e-01 -4.48294496e-03 1.31042793e-01 -1.32169533e+00 6.72422767e-01 7.52907634e-01 3.17675173e-02 -4.20458164e-05 -9.43762660e-02 -3.96070868e-01 1.05180585e+00 7.00459778e-01 -4.01451856e-01 -2.56724447e-01 -6.73328996e-01 3.20206434e-01 -1.13373913e-01 3.43411863e-01 -5.02612591e-01 6.48135552e-03 -2.94408917e-01 3.89667660e-01 1.92568041e-02 2.57516712e-01 -1.02879286e+00 -2.57443219e-01 -3.91007662e-01 -3.77399743e-01 -7.68861221e-03 2.03327626e-01 4.23049092e-01 -7.55940020e-01 -3.24247181e-01 6.48305416e-01 -1.81175098e-01 -7.37925053e-01 4.08308715e-01 -3.10452968e-01 8.97292495e-02 1.09331763e+00 -3.26958328e-01 6.05563149e-02 -4.91862267e-01 -1.08332014e+00 1.01475686e-01 4.81468737e-01 4.02470380e-01 3.89277816e-01 -1.50867331e+00 -4.76776749e-01 -9.76102948e-02 5.08621097e-01 1.64529812e-02 -9.24904179e-03 7.34308422e-01 -1.91399619e-01 -1.37264077e-02 -1.96073994e-01 -4.38522756e-01 -1.48384500e+00 3.14703703e-01 7.07415700e-01 3.57181802e-02 2.31310561e-01 9.76838887e-01 -9.84745026e-02 -5.34968972e-01 3.07573348e-01 2.94945300e-01 -2.04262450e-01 1.03429489e-01 6.22827053e-01 -3.48038413e-02 1.25756323e-01 -1.04451919e+00 -3.15271467e-01 5.47136366e-01 9.00399759e-02 6.66372180e-02 1.01092541e+00 -1.34752780e-01 -3.02564949e-01 2.89272875e-01 1.32294333e+00 -6.44907728e-02 -1.18586719e+00 -3.16810995e-01 2.60939062e-01 -6.25447094e-01 -2.99938917e-01 -5.97093701e-01 -1.03388810e+00 6.85381055e-01 6.75366104e-01 4.95976284e-02 1.36268663e+00 5.67989461e-02 4.85971481e-01 5.56937993e-01 5.09445131e-01 -1.52417409e+00 7.06245825e-02 4.13788825e-01 8.17680120e-01 -1.53424430e+00 -1.13447383e-01 -6.11938179e-01 -9.52161312e-01 1.10221267e+00 9.87085819e-01 2.09338978e-01 6.03153050e-01 3.64916772e-01 4.69071329e-01 -2.56609529e-01 -3.64097923e-01 -3.88598979e-01 2.46953350e-02 6.93827808e-01 6.21217787e-01 -6.29726127e-02 -5.02353489e-01 1.02039385e+00 2.02793002e-01 6.81431711e-01 1.96147770e-01 9.86776412e-01 -5.16922176e-02 -1.25966287e+00 -2.44488254e-01 -2.27866974e-02 -8.41211081e-01 3.58184934e-01 -9.98027146e-01 5.43985546e-01 4.19609666e-01 1.04558694e+00 -1.27131954e-01 -3.45828623e-01 5.57892621e-01 5.46106875e-01 3.32512289e-01 -3.31441194e-01 -5.05994737e-01 -9.69382375e-02 3.44268471e-01 -6.94769919e-01 -1.02268338e+00 -2.51571625e-01 -1.33161092e+00 8.71052817e-02 -3.64345163e-01 2.77111232e-01 2.37876013e-01 1.33067417e+00 3.75658512e-01 -8.86127129e-02 6.02779329e-01 -7.88260937e-01 -4.24826264e-01 -9.81316149e-01 -4.03870165e-01 8.54415059e-01 3.23021084e-01 -8.82989228e-01 -9.57215056e-02 3.18725973e-01]
[13.593207359313965, 1.677939534187317]
6ec2fcad-256c-4528-9672-e0adbd4b139a
how-asynchronous-events-encode-video
2206.04341
null
https://arxiv.org/abs/2206.04341v1
https://arxiv.org/pdf/2206.04341v1.pdf
How Asynchronous Events Encode Video
As event-based sensing gains in popularity, theoretical understanding is needed to harness this technology's potential. Instead of recording video by capturing frames, event-based cameras have sensors that emit events when their inputs change, thus encoding information in the timing of events. This creates new challenges in establishing reconstruction guarantees and algorithms, but also provides advantages over frame-based video. We use time encoding machines to model event-based sensors: TEMs also encode their inputs by emitting events characterized by their timing and reconstruction from time encodings is well understood. We consider the case of time encoding bandlimited video and demonstrate a dependence between spatial sensor density and overall spatial and temporal resolution. Such a dependence does not occur in frame-based video, where temporal resolution depends solely on the frame rate of the video and spatial resolution depends solely on the pixel grid. However, this dependence arises naturally in event-based video and allows oversampling in space to provide better time resolution. As such, event-based vision encourages using more sensors that emit fewer events over time.
['Martin Vetterli', 'Adam Scholefield', 'Karen Adam']
2022-06-09
null
null
null
null
['event-based-vision']
['computer-vision']
[ 8.29122186e-01 -2.74582207e-01 4.75829728e-02 -1.60277411e-01 -5.80378532e-01 -6.90828443e-01 4.96319115e-01 2.32195687e-02 -6.29124820e-01 7.02176511e-01 5.14066406e-02 1.15943896e-02 -2.61787735e-02 -8.68429720e-01 -8.31435084e-01 -6.35432899e-01 -4.02097136e-01 -1.54523656e-01 5.54419518e-01 3.94629091e-01 6.26022965e-02 5.42769194e-01 -1.75681388e+00 3.43124837e-01 1.38227552e-01 1.35197008e+00 3.49569201e-01 1.26236808e+00 6.05121143e-02 1.18089151e+00 -6.34387076e-01 2.90392011e-01 2.92562693e-01 -8.63148451e-01 -1.97827473e-01 3.91759098e-01 2.30494797e-01 -6.07493818e-01 -7.49497235e-01 8.73269916e-01 -6.70832023e-02 2.53052056e-01 3.84961873e-01 -1.10365176e+00 -3.00024301e-01 4.73017618e-02 -4.81273383e-01 4.44578588e-01 6.00450873e-01 1.14133596e-01 6.87877953e-01 -3.57150704e-01 7.64755249e-01 8.31480265e-01 6.23003364e-01 5.97549736e-01 -1.20862365e+00 -2.39222959e-01 -1.77811980e-01 2.12833256e-01 -9.56409156e-01 -5.95138967e-01 6.95531666e-01 -3.98605466e-01 9.86550033e-01 2.36409947e-01 9.29449618e-01 9.01947677e-01 3.97273123e-01 4.32816654e-01 7.95422971e-01 -4.78213370e-01 6.11187458e-01 -3.33179206e-01 -2.03708082e-01 3.75779659e-01 2.38504503e-02 4.14982527e-01 -9.52648342e-01 9.82378349e-02 1.30700159e+00 3.88209820e-01 -5.94905674e-01 -7.29665458e-02 -1.53268456e+00 4.89787161e-01 -3.65125798e-02 2.14125127e-01 -4.61441427e-01 7.55319655e-01 1.97209224e-01 3.85342956e-01 4.48379256e-02 3.11043382e-01 3.11310533e-02 -5.37975907e-01 -1.28412938e+00 7.77584836e-02 6.74724996e-01 8.79126489e-01 6.32271886e-01 1.55096740e-01 9.42922831e-02 1.69282854e-01 -7.54548013e-02 6.20389640e-01 3.46140593e-01 -1.67782009e+00 1.20240867e-01 2.41302536e-03 4.62139606e-01 -8.03620696e-01 -1.93587154e-01 2.36494303e-01 -7.05622435e-01 4.18538690e-01 6.56959116e-01 -1.69890434e-01 -6.81507587e-01 1.61778462e+00 -5.04318923e-02 6.52296662e-01 4.56524566e-02 9.41894889e-01 6.28239885e-02 1.10144281e+00 -2.81269759e-01 -9.49505925e-01 1.13179755e+00 -8.29473585e-02 -8.41113985e-01 6.17186651e-02 1.47459269e-01 -4.74881053e-01 6.05698645e-01 5.28534830e-01 -1.22447956e+00 -3.74390453e-01 -1.19944632e+00 7.95069933e-02 -3.29361223e-02 -4.78999972e-01 5.23371935e-01 4.22881097e-01 -9.85201001e-01 3.45741391e-01 -1.38269162e+00 -4.13636088e-01 1.87581882e-01 1.59016505e-01 -1.01410314e-01 3.17647308e-02 -8.51165771e-01 4.98270303e-01 1.24412283e-01 -3.72119576e-01 -6.93927646e-01 -7.65741527e-01 -7.36011565e-01 3.72372866e-02 2.16729566e-02 -4.18061137e-01 1.44343364e+00 -1.08571839e+00 -1.37922454e+00 5.07579863e-01 -4.47958350e-01 -9.40354288e-01 3.45156074e-01 1.96548477e-01 -4.80624765e-01 6.74767792e-01 -1.17674343e-01 6.25542641e-01 9.98024046e-01 -6.12326324e-01 -7.73703992e-01 -1.52792066e-01 3.42733979e-01 7.94525966e-02 -1.77458331e-01 -1.48865655e-01 -1.48934335e-01 -1.35243446e-01 2.27791965e-01 -6.71638191e-01 -8.99587870e-02 5.34347415e-01 2.46809065e-01 3.47243458e-01 1.00646579e+00 2.58842111e-02 9.55173433e-01 -2.37904453e+00 -2.88536817e-01 -1.75256059e-01 1.66498855e-01 -2.81319797e-01 1.02430962e-01 6.35459661e-01 3.04599285e-01 -3.31560373e-01 -2.33941004e-01 -1.82640299e-01 -3.89092445e-01 3.90681446e-01 -8.08177114e-01 6.76285803e-01 3.07544887e-01 4.48630393e-01 -9.79357898e-01 -5.28293312e-01 7.50971019e-01 7.44571507e-01 -5.27767241e-01 -3.05851884e-02 -4.65485126e-01 6.47334993e-01 -3.78992647e-01 4.68958765e-01 2.87594795e-01 -5.87088346e-01 1.08617954e-01 -1.85145721e-01 -6.01900101e-01 3.81056368e-02 -1.20225370e+00 1.63758779e+00 -3.33502918e-01 1.19658160e+00 2.16381058e-01 -7.98900783e-01 4.48006630e-01 6.22364581e-01 9.83922780e-01 -8.47950697e-01 4.03616903e-03 9.41086188e-03 -3.98727477e-01 -5.90119421e-01 7.26575315e-01 -3.47606838e-01 1.48287371e-01 4.32868153e-01 -2.02991515e-01 -6.42678142e-02 3.28385621e-01 1.58181846e-01 1.68943000e+00 2.76278481e-02 2.30385706e-01 1.86142400e-01 -1.42762437e-01 8.32982808e-02 4.94572729e-01 8.15777123e-01 -1.99139729e-01 7.63958454e-01 7.66209811e-02 -2.65855312e-01 -1.15741026e+00 -1.23335552e+00 -4.08114254e-01 5.87599695e-01 4.60864395e-01 -3.84027928e-01 -3.68310392e-01 2.09782660e-01 -8.15971270e-02 4.44780201e-01 -4.18722838e-01 1.91433638e-01 -4.31099832e-01 -4.92040694e-01 3.34259033e-01 6.09701097e-01 4.17974770e-01 -7.73047626e-01 -1.66242826e+00 5.77087402e-01 -7.12523833e-02 -1.38929892e+00 -3.38961422e-01 3.43852043e-01 -9.66093361e-01 -9.50434148e-01 -6.12230718e-01 -5.61917983e-02 4.35236663e-01 3.93399417e-01 9.64157999e-01 -4.57916319e-01 -3.47697675e-01 1.25277221e+00 -3.44729394e-01 -4.69866663e-01 -1.47819534e-01 -5.80257893e-01 -1.25506222e-01 1.73780784e-01 1.73049569e-01 -8.30893219e-01 -7.41567433e-01 -3.39065716e-02 -1.44083583e+00 3.90594378e-02 1.61253139e-01 4.51559216e-01 8.25067401e-01 1.35055840e-01 1.67489812e-01 -4.37806040e-01 4.37831432e-02 -5.32586694e-01 -1.01216435e+00 -4.23044488e-02 1.15276463e-02 -2.48858646e-01 5.90163410e-01 -6.31580830e-01 -8.54802907e-01 2.84225702e-01 4.13029104e-01 -6.89415753e-01 2.13380400e-02 2.65709400e-01 3.18902016e-01 5.56811020e-02 6.59628510e-01 4.02210295e-01 -1.43080205e-01 7.55035728e-02 8.24820921e-02 3.83300275e-01 8.59596312e-01 -4.01627541e-01 1.88367292e-01 1.19634831e+00 1.74782649e-01 -1.11474872e+00 -3.89089823e-01 -4.21297312e-01 -1.87735498e-01 -5.83780646e-01 9.55970883e-01 -8.89954805e-01 -8.17849636e-01 2.65483648e-01 -1.35953927e+00 -1.90381423e-01 -9.53921497e-01 8.16429436e-01 -1.00371957e+00 -5.57533577e-02 -8.16985965e-01 -1.14268863e+00 4.84883010e-01 -8.44038546e-01 1.05692947e+00 3.28236014e-01 -2.80006021e-01 -1.05703616e+00 -1.08162956e-02 -2.69890904e-01 2.71143407e-01 5.33262730e-01 2.14816108e-01 2.44607329e-01 -1.31495249e+00 -1.35191813e-01 -9.96602774e-02 7.57879065e-03 1.12617448e-01 3.47806886e-02 -9.15510297e-01 -9.52166766e-02 5.63585937e-01 4.72239070e-02 6.22382879e-01 9.27840531e-01 1.22372580e+00 -9.01363194e-02 -3.29595685e-01 6.17596567e-01 1.89392698e+00 5.93273461e-01 7.07094848e-01 3.22793610e-02 3.32131743e-01 2.27855906e-01 2.61921704e-01 7.42939711e-01 8.32619816e-02 5.50116777e-01 3.90663445e-01 2.15573698e-01 -6.34091049e-02 -1.12914458e-01 4.40221936e-01 4.74050552e-01 -1.23015605e-01 -5.05133331e-01 -5.87249994e-01 6.26767397e-01 -1.76343632e+00 -1.51694989e+00 -1.68498717e-02 2.39373875e+00 7.78227806e-01 -3.36968340e-03 1.31027415e-01 3.47333372e-01 6.96186900e-01 3.41998369e-01 -8.63509417e-01 -1.38624296e-01 -3.73408914e-01 1.95633382e-01 9.36856925e-01 4.17297006e-01 -7.06309855e-01 1.38905481e-01 7.10250282e+00 2.52580166e-01 -1.37398767e+00 1.23353682e-01 3.59739363e-01 -7.59594083e-01 -3.32760394e-01 -1.36228176e-02 -5.29171586e-01 7.74070203e-01 1.36804140e+00 -3.00817251e-01 5.52320421e-01 3.52598816e-01 4.97822583e-01 -5.17307043e-01 -1.42990053e+00 1.29104555e+00 -1.92764387e-01 -1.59472394e+00 -2.35225782e-01 2.30703652e-01 5.65732837e-01 -7.22621530e-02 -9.05929133e-02 -4.34401661e-01 9.75494757e-02 -7.50720203e-01 8.71206880e-01 7.17109442e-01 1.02855289e+00 -2.74004877e-01 -4.15086001e-02 2.73050904e-01 -1.30731940e+00 9.52148065e-02 -2.25179076e-01 -5.89142919e-01 6.55263543e-01 1.09700465e+00 -3.24987888e-01 -3.22770774e-02 5.97022772e-01 9.73786354e-01 2.63532817e-01 8.15891981e-01 3.91648948e-01 7.00973630e-01 -7.84986675e-01 2.63971627e-01 1.61014274e-02 -1.65303797e-01 6.04468882e-01 1.01210797e+00 7.12816596e-01 6.05679333e-01 5.93165457e-02 7.32827902e-01 1.32299483e-01 -8.22007418e-01 -8.48949492e-01 -1.84015125e-01 7.81302333e-01 7.18091428e-01 -8.31667244e-01 -3.58127147e-01 -9.22647119e-01 8.98900270e-01 -3.72321367e-01 6.34077191e-01 -9.60758448e-01 -2.74827719e-01 5.24714053e-01 4.81445134e-01 4.47387636e-01 -6.45283997e-01 -2.33960375e-02 -1.18790007e+00 8.73332694e-02 -3.82076144e-01 7.30330050e-02 -9.66835678e-01 -1.04549718e+00 6.04706770e-03 7.70632774e-02 -1.52633345e+00 -4.64058638e-01 -4.07954305e-01 -4.22782078e-02 4.54121232e-01 -1.16574717e+00 -4.58000690e-01 -1.82663962e-01 7.95864880e-01 5.13560236e-01 4.98828351e-01 5.73519230e-01 2.42577672e-01 1.27024099e-01 -1.27273142e-01 2.51355082e-01 -3.61480145e-03 3.74752581e-01 -9.84182894e-01 -2.45036352e-02 9.66174245e-01 4.20049578e-01 3.72122347e-01 8.16628754e-01 -3.84434789e-01 -1.99502015e+00 -9.01232541e-01 5.79150975e-01 -4.38614756e-01 7.02743769e-01 -9.09653381e-02 -7.26420522e-01 7.44213283e-01 4.48382497e-02 3.57221961e-01 5.16412914e-01 -4.81410354e-01 -2.69311756e-01 -2.69298226e-01 -1.14768302e+00 1.95217967e-01 1.08062768e+00 -1.08638597e+00 -2.87400544e-01 1.76124439e-01 4.85304326e-01 -5.41006923e-01 -6.68475866e-01 9.07581300e-02 7.32446015e-01 -1.27627242e+00 7.60109961e-01 1.81164131e-01 4.86376733e-01 -5.66064298e-01 -5.29034972e-01 -6.85527325e-01 -8.78320411e-02 -5.85920215e-01 -3.98543149e-01 8.75955462e-01 -1.23848215e-01 -7.10987210e-01 7.28268921e-01 7.33262062e-01 1.29659474e-01 -2.51536161e-01 -1.13099372e+00 -9.58201528e-01 -6.11528814e-01 -7.51002312e-01 3.02182466e-01 7.09859550e-01 1.00178279e-01 -2.00642720e-01 -1.57684550e-01 2.40864784e-01 8.29968989e-01 8.44983086e-02 6.94099218e-02 -1.00635433e+00 -4.46470678e-01 -7.61943832e-02 -7.11795986e-01 -1.44695401e+00 -3.38538587e-01 -1.62316620e-01 2.36579001e-01 -1.03131688e+00 -2.21597347e-02 -1.77143857e-01 -2.38350734e-01 -8.58235285e-02 5.10992885e-01 4.12809402e-01 1.49738923e-01 4.56001014e-01 -6.78409159e-01 2.70119160e-02 9.32570457e-01 1.35192811e-01 -7.80754462e-02 -3.80198121e-01 1.10603847e-01 6.23533130e-01 3.59526724e-01 -2.17631131e-01 -7.83875704e-01 -5.69073558e-01 4.67350632e-01 9.79648530e-01 8.16541553e-01 -1.53081918e+00 6.25761509e-01 -3.58695477e-01 5.01614034e-01 -5.07544696e-01 8.26212943e-01 -1.09180212e+00 7.80280948e-01 1.23656049e-01 -5.98276019e-01 1.65533543e-01 5.95466979e-03 9.87072051e-01 -3.45096558e-01 2.16401182e-03 8.18767011e-01 -3.46108943e-01 -7.79152453e-01 3.03621233e-01 -9.02535260e-01 -8.27149525e-02 9.79038298e-01 -8.28340530e-01 -1.87731192e-01 -5.10981679e-01 -5.67626178e-01 -4.13518280e-01 7.78501511e-01 -7.57007748e-02 6.44990087e-01 -1.23012018e+00 -1.23737417e-01 3.23485106e-01 -1.13185868e-01 -1.41268715e-01 2.01614350e-01 6.00274503e-01 -4.83478904e-01 2.76218057e-01 -2.20722198e-01 -1.06124306e+00 -9.11337078e-01 2.43331596e-01 2.53386557e-01 3.33567739e-01 -7.26941943e-01 7.38494217e-01 2.63511360e-01 7.77751386e-01 1.45455122e-01 -5.84850848e-01 4.08579081e-01 -1.61129475e-01 8.92736077e-01 3.97410721e-01 -2.43278503e-01 -1.96146458e-01 -2.15005979e-01 5.27058780e-01 3.05040628e-01 -7.92735636e-01 1.00192082e+00 -4.07629907e-01 1.19110562e-01 1.18221235e+00 1.08120334e+00 -5.46537042e-02 -1.85385394e+00 4.02773777e-03 -3.78563255e-01 -5.71178734e-01 1.70336008e-01 -2.33000770e-01 -8.11936617e-01 8.77336204e-01 5.85827589e-01 7.23608911e-01 1.48183715e+00 -5.21869212e-02 1.05139983e+00 -1.07239801e-02 8.04440022e-01 -8.89980495e-01 -7.81745017e-02 3.62038612e-01 2.86638409e-01 -9.32077467e-01 -1.63209260e-01 -2.24529535e-01 1.84402943e-01 1.16821563e+00 1.43479882e-02 -3.08611125e-01 6.05203629e-01 8.16026807e-01 -2.58957595e-01 -6.15071058e-02 -8.86435032e-01 -9.26822424e-02 -3.07358086e-01 6.59796596e-01 3.77033412e-01 1.22415863e-01 1.15367118e-03 -1.87992111e-01 -1.17015382e-02 5.81173241e-01 7.78981984e-01 1.08583176e+00 -5.63017190e-01 -8.45087826e-01 -5.14705122e-01 5.39775908e-01 -3.91873926e-01 8.85927379e-02 2.10305393e-01 5.07246196e-01 7.42476135e-02 1.03771579e+00 7.30506539e-01 -1.52124658e-01 -2.26744916e-03 -2.38543525e-01 9.75452960e-01 -4.97205168e-01 -1.37894787e-02 -4.64407615e-02 -2.47276276e-01 -8.47806931e-01 -9.99708056e-01 -8.88911486e-01 -1.55081117e+00 -5.00850439e-01 2.78676767e-02 5.31656481e-02 7.27932751e-01 8.01104724e-01 3.95174444e-01 5.88032186e-01 4.11946774e-01 -8.56144369e-01 -3.64610814e-02 -3.07170987e-01 -7.03078210e-01 2.09454253e-01 9.39258814e-01 -2.52179563e-01 -5.94684005e-01 8.70216250e-01]
[8.697234153747559, -1.3519951105117798]
182ac7d4-c1fd-4481-aee9-6f5690073d7c
concept-oriented-deep-learning-with-large
2306.17089
null
https://arxiv.org/abs/2306.17089v1
https://arxiv.org/pdf/2306.17089v1.pdf
Concept-Oriented Deep Learning with Large Language Models
Large Language Models (LLMs) have been successfully used in many natural-language tasks and applications including text generation and AI chatbots. They also are a promising new technology for concept-oriented deep learning (CODL). However, the prerequisite is that LLMs understand concepts and ensure conceptual consistency. We discuss these in this paper, as well as major uses of LLMs for CODL including concept extraction from text, concept graph extraction from text, and concept learning. Human knowledge consists of both symbolic (conceptual) knowledge and embodied (sensory) knowledge. Text-only LLMs, however, can represent only symbolic (conceptual) knowledge. Multimodal LLMs, on the other hand, are capable of representing the full range (conceptual and sensory) of human knowledge. We discuss conceptual understanding in visual-language LLMs, the most important multimodal LLMs, and major uses of them for CODL including concept extraction from image, concept graph extraction from image, and concept learning. While uses of LLMs for CODL are valuable standalone, they are particularly valuable as part of LLM applications such as AI chatbots.
['Daniel T. Chang']
2023-06-29
null
null
null
null
['text-generation']
['natural-language-processing']
[ 7.01467544e-02 5.10283351e-01 -1.45933717e-01 -1.17931046e-01 -5.08339286e-01 -7.11316168e-01 1.09441864e+00 6.61541998e-01 -4.28029805e-01 6.23883009e-01 -8.40508789e-02 -2.28717536e-01 -4.44718227e-02 -9.06619608e-01 -5.24521589e-01 -4.45745498e-01 -7.94136431e-03 6.21836901e-01 1.28726035e-01 -4.55938041e-01 3.45317245e-01 3.82581770e-01 -1.73535824e+00 7.72325456e-01 6.79224372e-01 7.88007081e-01 4.70639288e-01 6.56300068e-01 -1.16931021e+00 1.52455246e+00 -7.72382259e-01 -3.11433434e-01 -6.68003261e-01 -6.42478287e-01 -1.20837879e+00 1.68728396e-01 4.19919938e-02 3.95634249e-02 7.30804428e-02 9.20697510e-01 8.98830295e-02 2.33780235e-01 9.06597316e-01 -1.97762120e+00 -5.78415453e-01 7.01787233e-01 -1.59504399e-01 -6.64712667e-01 6.81300759e-01 -2.11120307e-01 8.11528206e-01 -9.32380795e-01 8.24988008e-01 1.77463186e+00 2.68586069e-01 8.19252431e-01 -8.02064717e-01 -3.19211990e-01 2.76744127e-01 4.99413192e-01 -1.28223562e+00 -9.87611040e-02 6.93642199e-01 -5.15302479e-01 1.01473415e+00 -8.20356533e-02 8.52164865e-01 8.63188386e-01 3.40946205e-02 1.40050089e+00 1.01893961e+00 -1.03341925e+00 4.28384960e-01 5.32147229e-01 -9.21594650e-02 9.05122697e-01 -2.24660605e-01 -2.74698257e-01 -1.10926867e+00 9.23488196e-03 1.01531339e+00 -1.39128029e-01 1.71776876e-01 -4.83625382e-01 -1.42257059e+00 1.00507331e+00 4.05044794e-01 5.37443340e-01 -1.38061360e-01 4.12384242e-01 2.83547014e-01 3.44471693e-01 -8.41407031e-02 4.77532178e-01 -1.22595087e-01 -1.85550302e-01 -7.58680463e-01 1.43487290e-01 9.43987787e-01 1.44407141e+00 9.24608707e-01 3.19118090e-02 4.48745936e-02 8.67752552e-01 3.51433992e-01 7.54363060e-01 6.63273096e-01 -1.08110833e+00 2.30388373e-01 9.49770391e-01 -6.01840466e-02 -9.32978570e-01 -4.28372502e-01 3.99724513e-01 -7.63123870e-01 1.80371433e-01 9.32237059e-02 -1.17057279e-01 -8.43723178e-01 1.51278400e+00 -5.45418821e-02 -1.86866388e-01 4.58355159e-01 6.03008091e-01 1.32769156e+00 7.87420750e-01 4.95356321e-01 -1.53029993e-01 1.33491254e+00 -6.87789381e-01 -7.94745743e-01 -2.85296321e-01 6.57320142e-01 -3.78431916e-01 1.00481081e+00 2.83659279e-01 -9.96116400e-01 -5.34569740e-01 -7.70426571e-01 -1.31845370e-01 -9.88282025e-01 -1.17193870e-02 1.01688576e+00 1.96398154e-01 -1.16489875e+00 5.83050288e-02 -5.22992194e-01 -8.67370486e-01 4.87225443e-01 2.40974486e-01 -4.65811223e-01 -1.74463570e-01 -1.30155325e+00 9.56518948e-01 5.79746842e-01 -8.93190969e-03 -8.40964496e-01 -2.12116465e-02 -1.21601796e+00 -4.37029041e-02 6.95137024e-01 -4.72150981e-01 1.33028603e+00 -1.22131503e+00 -1.25307393e+00 1.09902227e+00 -4.87107970e-02 -4.16332722e-01 2.25379437e-01 3.92989255e-03 -3.92991513e-01 6.37374341e-01 3.00169662e-02 1.39420819e+00 8.47730219e-01 -1.65452385e+00 -6.54270828e-01 7.91078731e-02 4.40692723e-01 3.64926815e-01 -2.60570049e-01 -1.28349245e-01 -4.30227816e-01 -1.90942958e-01 1.82449937e-01 -7.04401970e-01 -9.71195474e-02 2.42584661e-01 -2.65584558e-01 -3.91154081e-01 8.77639890e-01 -1.17868096e-01 7.73135424e-01 -1.77935755e+00 2.37762675e-01 1.71964511e-01 3.43293548e-01 3.06300223e-01 -5.17895877e-01 1.06010997e+00 3.99531662e-01 1.99402124e-02 -2.07891583e-01 -9.23737511e-03 1.92412421e-01 3.87839407e-01 -1.85694069e-01 -3.15096200e-01 4.34625179e-01 1.42165267e+00 -1.09576988e+00 -8.39844286e-01 6.00797534e-01 4.16070700e-01 -1.12120748e-01 1.87021345e-01 -8.96417439e-01 1.94046959e-01 -4.10976470e-01 5.07124722e-01 1.35990724e-01 -2.69495815e-01 4.62160081e-01 -5.54410256e-02 -8.36543273e-03 -2.74486899e-01 -9.61142182e-01 1.75145590e+00 -7.04838157e-01 1.11350787e+00 -1.75730199e-01 -1.05026233e+00 8.79444063e-01 5.55123866e-01 3.17462265e-01 -5.74260592e-01 -3.99091840e-02 -1.11738302e-01 -4.77992408e-02 -7.30997682e-01 4.42878783e-01 -4.88859266e-01 -1.05296284e-01 6.06967330e-01 4.91578996e-01 -6.07949734e-01 3.28327119e-01 8.28306496e-01 3.18038404e-01 2.43047878e-01 5.86452603e-01 4.54433858e-02 5.88998497e-01 3.57513636e-01 -2.40800619e-01 7.74990797e-01 1.36693045e-01 1.82518601e-01 7.00837135e-01 -1.48711160e-01 -5.58665216e-01 -9.97563720e-01 4.54004914e-01 1.14461124e+00 4.68768448e-01 -6.69025600e-01 -6.91646397e-01 -4.71264422e-01 -2.93227762e-01 7.19262898e-01 -2.24926218e-01 -7.98708797e-02 -1.65249869e-01 -1.35603279e-01 4.87698734e-01 7.01528072e-01 5.88755131e-01 -1.73166609e+00 -8.77630651e-01 2.59589493e-01 -4.95325804e-01 -1.35678124e+00 2.83219904e-01 -6.43231794e-02 -6.44046187e-01 -1.31798053e+00 -5.64268231e-01 -1.03942370e+00 7.14884520e-01 2.91935086e-01 1.16377068e+00 4.25125062e-01 -3.96713376e-01 1.30885863e+00 -6.32032692e-01 -8.50780129e-01 -6.19172454e-01 -3.15121800e-01 -1.53334603e-01 -1.35360718e-01 4.12784636e-01 -3.05979818e-01 -2.80310214e-02 -1.98743585e-02 -1.10115278e+00 4.94902998e-01 5.62001705e-01 6.22460186e-01 4.65125501e-01 3.78245525e-02 7.15362132e-01 -6.40766621e-01 9.31752324e-01 -2.37378612e-01 -2.94145584e-01 6.86008096e-01 1.81345437e-02 -2.64955442e-02 2.33751118e-01 -5.93501925e-01 -1.17923117e+00 -1.11382745e-01 -7.72777423e-02 -2.19825983e-01 -5.00879347e-01 1.01534128e+00 -7.98693225e-02 1.33532554e-01 4.99933273e-01 3.73502523e-01 -3.23426127e-02 -3.09150964e-01 8.18102777e-01 6.90867364e-01 4.01047468e-01 -8.99647355e-01 3.72950107e-01 4.97595191e-01 1.32310092e-01 -1.44467914e+00 -6.90641701e-01 -4.23587620e-01 -1.03119588e+00 -4.95314419e-01 8.44303727e-01 -7.45042324e-01 -1.10035527e+00 4.83716697e-01 -1.30037642e+00 -4.51308936e-01 -3.03491861e-01 2.64985502e-01 -8.09804559e-01 2.89958298e-01 -4.83925074e-01 -1.13033950e+00 -1.16175354e-01 -6.04632318e-01 1.00028074e+00 3.40655625e-01 -3.73362213e-01 -1.29825890e+00 -5.22437334e-01 2.35123411e-01 1.52035132e-01 3.25360149e-01 1.40644038e+00 -6.08206213e-01 -5.55408239e-01 -1.56341881e-01 -2.05469713e-01 1.02633245e-01 9.10242498e-02 -1.55165950e-02 -9.24822867e-01 -5.60803823e-02 -4.30765986e-01 -9.11230922e-01 6.55687809e-01 1.62326321e-01 7.61580944e-01 -1.88765928e-01 -4.73560691e-01 -2.48496547e-01 1.35747087e+00 3.06653351e-01 4.58764195e-01 4.13514860e-02 6.51007116e-01 1.03072834e+00 7.39170015e-01 2.03516006e-01 5.50676465e-01 3.99973392e-01 2.34275088e-01 -1.02764800e-01 -8.23307037e-02 -4.28898782e-01 4.33971614e-01 8.76640320e-01 -6.44000620e-02 -2.19016194e-01 -1.30351710e+00 5.38414001e-01 -2.21761179e+00 -1.03388691e+00 -1.27011448e-01 1.73633277e+00 8.88585627e-01 -8.64053220e-02 1.71921626e-02 -8.05604178e-03 6.30039036e-01 -2.09891781e-01 -4.84960765e-01 -4.53778774e-01 -2.59532839e-01 1.54128790e-01 -4.42293465e-01 4.70143288e-01 -8.75400007e-01 1.45934522e+00 5.85150576e+00 9.56846535e-01 -9.98010337e-01 5.84191196e-02 -3.61601636e-02 2.18027145e-01 -2.34907418e-01 5.73656559e-02 -5.04791141e-01 -1.42344996e-01 6.66443765e-01 -6.17831424e-02 1.62849978e-01 8.25275898e-01 -1.70167148e-01 -6.12459421e-01 -1.24688852e+00 1.24620569e+00 2.22444713e-01 -1.71555233e+00 5.28993845e-01 -3.12220067e-01 6.73604190e-01 -3.58943850e-01 -6.24237299e-01 5.41625917e-01 1.88148633e-01 -1.19698322e+00 8.35302472e-01 7.74231434e-01 9.48348224e-01 -7.95782447e-01 7.53923893e-01 6.72083914e-01 -1.42332745e+00 5.19036092e-02 -2.40606070e-01 -1.33289918e-01 9.34405476e-02 2.45845616e-02 -5.06705821e-01 4.38732386e-01 4.15008336e-01 6.73813343e-01 -5.59063494e-01 5.20896852e-01 -6.13825083e-01 1.26803011e-01 4.19092625e-02 -4.58364844e-01 3.13923359e-01 1.78977266e-01 6.59827441e-02 1.38880908e+00 -1.01154149e-01 1.77355856e-01 2.85644323e-01 1.03604817e+00 2.57384069e-02 7.43630975e-02 -6.23834431e-01 -7.73222268e-01 6.42478406e-01 1.10169423e+00 -8.74997735e-01 -6.54366791e-01 -5.34297943e-01 8.34887385e-01 3.26581925e-01 4.41493183e-01 -3.61281097e-01 -7.16759920e-01 3.21171463e-01 -1.31718695e-01 -1.64306968e-01 -4.86188173e-01 2.25261584e-01 -1.15242922e+00 -4.19644207e-01 -6.77097380e-01 2.33708680e-01 -1.35784674e+00 -1.32418776e+00 4.66398478e-01 4.07572955e-01 -1.17706943e+00 -6.14454150e-01 -8.73095334e-01 -4.64244753e-01 5.37042141e-01 -1.49064827e+00 -1.80426061e+00 -3.33896101e-01 9.15732920e-01 6.92741513e-01 -3.59650522e-01 1.11298084e+00 -3.89435858e-01 3.05175751e-01 -2.28694230e-02 -2.75778234e-01 3.50440949e-01 3.77500355e-01 -1.13035011e+00 8.28059688e-02 9.68481377e-02 6.42917156e-01 7.64597058e-01 1.43078431e-01 -5.70328951e-01 -1.39399672e+00 -7.61322558e-01 1.05612826e+00 -3.35146815e-01 6.13562644e-01 -5.90651274e-01 -8.47747266e-01 6.49048805e-01 2.94578910e-01 -6.46433234e-01 7.25203156e-01 1.12551982e-02 -2.18646973e-01 3.20400298e-01 -9.36998188e-01 7.74604440e-01 6.95642591e-01 -9.03146863e-01 -7.62246251e-01 5.16168892e-01 5.87479293e-01 -1.94674954e-01 -6.28461897e-01 -5.55041209e-02 6.34886742e-01 -8.69006634e-01 9.16776955e-01 -6.17943764e-01 4.40717787e-01 -2.96858460e-01 -7.49028996e-02 -1.14749062e+00 1.95565104e-01 -4.48761612e-01 -3.05216670e-01 1.31682694e+00 1.49711505e-01 -1.84483439e-01 4.45913136e-01 5.13848305e-01 2.57647038e-01 -3.48123729e-01 -5.64912379e-01 -7.35974073e-01 2.01984998e-02 -8.71046960e-01 2.69770175e-01 1.11782575e+00 7.22891331e-01 7.11080909e-01 -2.81123281e-01 -1.95444271e-01 4.37919676e-01 3.51331085e-01 6.34520829e-01 -1.41611254e+00 2.54177392e-01 -3.60152721e-01 -4.99177873e-01 -8.61666858e-01 3.17325324e-01 -7.77251244e-01 1.81617185e-01 -2.32022285e+00 2.13926956e-01 -8.52620974e-02 1.40422583e-01 8.15558434e-01 2.94576287e-01 -6.14980981e-02 7.05280721e-01 1.73462406e-01 -8.48747790e-01 3.79106492e-01 1.29244709e+00 -3.41867536e-01 -1.31960720e-01 -3.51047039e-01 -5.25094271e-01 1.05088341e+00 7.65215337e-01 -1.22735642e-01 -7.84064710e-01 -1.83520153e-01 4.77276385e-01 1.80662334e-01 8.65447879e-01 -8.86814654e-01 5.76149762e-01 -5.85421979e-01 2.84742236e-01 -5.89520276e-01 5.81776559e-01 -7.70321369e-01 -3.55390877e-01 3.20756465e-01 -5.58748484e-01 -3.27060610e-01 4.72250432e-01 4.15159613e-01 -6.75713360e-01 -4.03073370e-01 5.43921351e-01 -7.23504424e-01 -1.55378318e+00 -1.96866125e-01 -9.26294506e-01 1.24685273e-01 7.67428637e-01 -2.11522549e-01 -2.41496578e-01 -1.04791272e+00 -9.52255607e-01 4.75263596e-01 1.16510548e-01 6.93028748e-01 1.06961024e+00 -1.27664042e+00 -2.49830514e-01 5.86882755e-02 4.84770000e-01 2.72108875e-02 1.62307620e-01 4.78477359e-01 -3.88893247e-01 7.49542058e-01 -3.22277904e-01 -6.12767518e-01 -1.13731730e+00 5.17119706e-01 1.68002486e-01 3.63718927e-01 -4.17724073e-01 8.78039241e-01 3.36582273e-01 -6.29414618e-01 5.09859264e-01 -1.14583001e-01 -4.35104251e-01 2.39673659e-01 6.36386931e-01 -5.63827008e-02 -4.24905419e-01 -6.80482984e-01 -3.49059433e-01 6.97504163e-01 4.05090064e-01 -4.81922269e-01 1.11282408e+00 -2.47442678e-01 -3.59822094e-01 8.90843987e-01 8.39280725e-01 -4.51588541e-01 -8.11497867e-01 -3.24467868e-01 2.90525526e-01 9.85363349e-02 -3.54393870e-01 -9.52909529e-01 -4.95256513e-01 1.63254702e+00 1.69557676e-01 1.82289138e-01 9.78830040e-01 3.74720395e-01 4.02322561e-01 9.19982791e-01 6.50626063e-01 -1.31276238e+00 6.95063829e-01 7.18899131e-01 9.92157161e-01 -1.43797183e+00 -1.36363819e-01 -2.77845442e-01 -1.19670212e+00 1.46814907e+00 6.93837106e-01 5.12636065e-01 5.11182666e-01 1.96992040e-01 2.97558218e-01 -4.80890930e-01 -8.14815700e-01 -6.04406655e-01 2.88307518e-01 1.14586818e+00 4.74134803e-01 -2.39982139e-02 1.26133278e-01 5.02408922e-01 4.75279875e-02 7.48700127e-02 4.04414147e-01 1.27598202e+00 -6.77995086e-01 -1.06791353e+00 -2.79488117e-01 8.16692039e-02 1.94192156e-01 -2.56199658e-01 -8.20173025e-01 1.08171821e+00 2.44107708e-01 1.33116984e+00 8.32099915e-02 -1.44816294e-01 1.28214672e-01 3.36413592e-01 7.69250274e-01 -9.05169010e-01 -3.20009202e-01 -1.20685257e-01 1.45661086e-01 -2.59260505e-01 -9.77291346e-01 -2.01249748e-01 -2.00871301e+00 -1.55809805e-01 -1.99345022e-01 2.22033948e-01 7.90279984e-01 1.33769441e+00 -1.68304950e-01 1.99797809e-01 -5.72161302e-02 -8.03240716e-01 2.99339592e-01 -8.30855668e-01 -6.30117297e-01 3.39378238e-01 1.63860589e-01 -5.00414848e-01 1.10123999e-01 4.97313976e-01]
[10.59089469909668, 1.8899989128112793]
2bf207a8-1161-4b7d-a307-247e48f74723
evaluation-of-deep-segmentation-models-for
2006.02662
null
https://arxiv.org/abs/2006.02662v2
https://arxiv.org/pdf/2006.02662v2.pdf
Exploiting the Transferability of Deep Learning Systems Across Multi-modal Retinal Scans for Extracting Retinopathy Lesions
Retinal lesions play a vital role in the accurate classification of retinal abnormalities. Many researchers have proposed deep lesion-aware screening systems that analyze and grade the progression of retinopathy. However, to the best of our knowledge, no literature exploits the tendency of these systems to generalize across multiple scanner specifications and multi-modal imagery. Towards this end, this paper presents a detailed evaluation of semantic segmentation, scene parsing and hybrid deep learning systems for extracting the retinal lesions such as intra-retinal fluid, sub-retinal fluid, hard exudates, drusen, and other chorioretinal anomalies from fused fundus and optical coherence tomography (OCT) imagery. Furthermore, we present a novel strategy exploiting the transferability of these models across multiple retinal scanner specifications. A total of 363 fundus and 173,915 OCT scans from seven publicly available datasets were used in this research (from which 297 fundus and 59,593 OCT scans were used for testing purposes). Overall, a hybrid retinal analysis and grading network (RAGNet), backboned through ResNet-50, stood first for extracting the retinal lesions, achieving a mean dice coefficient score of 0.822. Moreover, the complete source code and its documentation are released at: http://biomisa.org/index.php/downloads/.
['Naoufel Werghi', 'Taimur Hassan', 'Muhammad Usman Akram']
2020-06-04
null
null
null
null
['scene-parsing']
['computer-vision']
[ 1.82818752e-02 -1.12899147e-01 1.40214413e-01 -4.81556267e-01 -5.21107674e-01 -4.95737225e-01 1.37024105e-01 2.43706815e-02 -2.07165256e-01 7.78707445e-01 1.22346118e-01 -5.62920153e-01 -4.08854276e-01 -6.67639911e-01 -1.91931486e-01 -5.02904594e-01 -1.22156203e-01 -9.22798812e-02 2.89056540e-01 3.48654896e-01 5.68002105e-01 6.93396568e-01 -1.71526396e+00 5.16603410e-01 1.47933638e+00 1.26651466e+00 1.08291097e-02 1.04057825e+00 1.41342476e-01 8.45200062e-01 -2.38318443e-01 -4.43420410e-01 5.64133942e-01 -6.04386806e-01 -7.79929936e-01 3.30849677e-01 1.21947861e+00 -7.42166042e-01 -1.70707181e-01 1.16621590e+00 9.56476986e-01 -2.70236522e-01 5.02350628e-01 -5.26429474e-01 -6.74878776e-01 -4.03541373e-03 -5.65413952e-01 7.42849946e-01 -1.32838473e-01 7.02535272e-01 5.42960465e-01 -6.13005102e-01 4.90821958e-01 7.79275537e-01 6.22379482e-01 1.67222753e-01 -6.89422548e-01 -3.52558076e-01 -2.62219161e-01 3.09117943e-01 -1.03816628e+00 -5.43063343e-01 -4.10519093e-02 -1.07172465e+00 7.47218907e-01 2.20532060e-01 1.16422880e+00 3.96895528e-01 2.51817197e-01 2.11503416e-01 1.63712192e+00 -1.86339408e-01 -1.34433284e-01 1.27829064e-03 2.72332102e-01 1.01832783e+00 7.59688437e-01 1.50111690e-01 1.11415982e-01 -6.26875961e-04 9.64269400e-01 -3.25586110e-01 -2.58950591e-01 -9.10595618e-03 -6.61565781e-01 5.68117738e-01 5.17456412e-01 -1.50092185e-01 -5.47356129e-01 -3.09587687e-01 2.89564490e-01 1.96879104e-01 5.10488510e-01 5.46546400e-01 -1.83977336e-01 1.52901441e-01 -7.95567274e-01 -4.75932546e-02 2.45567635e-01 4.13497806e-01 4.86856788e-01 -1.67744458e-01 -3.04123014e-01 1.03728676e+00 2.69323885e-01 2.96226889e-01 4.52571958e-01 -1.12483871e+00 1.71126097e-01 1.02307117e+00 2.50835363e-02 -7.54580438e-01 -7.37948954e-01 -4.62236196e-01 -7.90501893e-01 5.16989827e-01 4.85048056e-01 -6.24368429e-01 -1.38192916e+00 7.49884188e-01 1.91244081e-01 1.80042475e-01 -2.36070588e-01 1.31811726e+00 1.23364878e+00 -1.38088539e-01 8.65469724e-02 8.55120793e-02 1.55093837e+00 -8.84553730e-01 -3.46340388e-01 -1.07449576e-01 5.81178844e-01 -1.02011824e+00 6.96501493e-01 2.93923318e-01 -1.36849058e+00 -5.16480029e-01 -7.12327778e-01 -3.58060718e-01 -2.64240682e-01 7.78400838e-01 8.52320790e-01 6.05475307e-01 -1.44140351e+00 2.25017816e-01 -7.82834649e-01 -6.36947274e-01 9.57715869e-01 2.12719738e-01 -2.46715367e-01 -2.91038811e-01 -5.49195588e-01 9.44983125e-01 1.39678985e-01 2.63048142e-01 -2.92336822e-01 -7.99429595e-01 -5.15766919e-01 -4.46186453e-01 3.98586690e-02 -1.29319632e+00 9.34207141e-01 -7.94302046e-01 -1.20935595e+00 1.27411568e+00 -3.09866309e-01 -6.11119449e-01 5.58266759e-01 -3.05457592e-01 -5.85177481e-01 6.25544786e-01 5.61572553e-04 4.48416084e-01 5.95898569e-01 -6.90035224e-01 -9.88562703e-01 -5.60729921e-01 1.85508519e-01 -2.26669032e-02 2.35015020e-01 4.71260428e-01 -2.89000273e-01 -1.24201939e-01 7.17024729e-02 -7.66905487e-01 -2.21520871e-01 2.09855750e-01 -7.74781525e-01 -5.33271171e-02 1.93244871e-02 -9.53117311e-01 1.12181151e+00 -1.85684180e+00 -4.46876764e-01 1.04375653e-01 8.23701799e-01 9.69283342e-01 -1.70851246e-01 -1.99638635e-01 -2.06726715e-01 3.31542790e-01 -1.08097605e-01 4.58548777e-02 -5.47591805e-01 -3.37778926e-01 2.44154513e-01 6.39394879e-01 4.15496200e-01 9.67691302e-01 -6.93967223e-01 -4.62248802e-01 5.58872461e-01 3.30088496e-01 -4.07757968e-01 -1.37139514e-01 1.28318906e-01 4.99145150e-01 -4.03589815e-01 1.03798497e+00 8.16518247e-01 -5.59608102e-01 -1.02188297e-01 -3.24891269e-01 -5.28574347e-01 6.81199878e-02 -8.35264623e-01 1.05387366e+00 3.10049916e-04 9.57972109e-01 -1.33564323e-01 -3.72607797e-01 6.39955521e-01 -1.18261157e-02 5.89121222e-01 -6.57993019e-01 3.03811640e-01 4.05371279e-01 4.93363351e-01 -1.00384033e+00 2.65793920e-01 3.44587207e-01 9.32814538e-01 8.62832814e-02 -3.02550316e-01 2.61955172e-01 6.00578249e-01 -3.10279250e-01 1.00934231e+00 -1.21473067e-01 4.10762578e-01 2.58475900e-01 5.06344318e-01 2.34694317e-01 4.13405806e-01 5.26690245e-01 -5.94426572e-01 9.05768514e-01 7.34745562e-01 -6.47062659e-01 -1.05284965e+00 -9.92626905e-01 -5.90383410e-01 1.03904583e-01 -9.67083499e-02 -1.47377729e-01 -6.55102253e-01 -2.13900089e-01 4.88784574e-02 -5.10705598e-02 -4.37469006e-01 3.89752716e-01 -1.76664978e-01 -1.16374719e+00 3.96791667e-01 2.60647207e-01 8.77870262e-01 -6.38626635e-01 -5.17058432e-01 -5.26783206e-02 -2.16627549e-02 -9.75446224e-01 9.26588997e-02 -9.67224896e-01 -9.75464880e-01 -1.82146275e+00 -8.12698722e-01 -6.51386559e-01 7.10803151e-01 3.77181262e-01 8.87913942e-01 8.12610388e-02 -1.07219541e+00 7.40568042e-02 -8.71378481e-02 -3.81673932e-01 -2.08448395e-01 -2.51054406e-01 -2.36116603e-01 3.20248604e-01 5.25513470e-01 -2.31771767e-01 -1.21336615e+00 1.28903016e-01 -4.75337356e-01 -3.18130776e-02 1.06724942e+00 3.30808133e-01 8.03548515e-01 -1.16302967e-01 9.22441110e-02 -7.51195967e-01 5.71239948e-01 -3.93152982e-01 -8.92706156e-01 1.63335010e-01 -6.10719681e-01 -8.90408158e-01 -6.82233125e-02 1.03453182e-01 -7.81908095e-01 -2.65047222e-01 2.61071056e-01 -2.12399274e-01 -6.79439366e-01 3.65804523e-01 5.21055520e-01 -4.25155222e-01 8.61224055e-01 -1.45217463e-01 3.37496817e-01 -5.98612487e-01 2.64395803e-01 9.97188032e-01 5.11123776e-01 3.74965779e-02 1.78467229e-01 7.09578753e-01 1.71387166e-01 -9.33921933e-01 -1.11657834e+00 -6.94013119e-01 -4.98025656e-01 -1.89952448e-01 1.10951340e+00 -9.34231400e-01 -6.31088793e-01 1.08418775e+00 -8.78382742e-01 -2.33939752e-01 -4.45976406e-02 9.08301592e-01 -3.43996495e-01 3.82697701e-01 -5.05017817e-01 -5.40557802e-01 -2.54227847e-01 -1.19613838e+00 5.46801805e-01 7.27034748e-01 1.65460825e-01 -7.96231091e-01 -5.95749728e-02 7.35455155e-01 5.10516644e-01 4.19909805e-01 9.68101799e-01 -1.30148605e-01 -7.79566169e-01 -9.64576453e-02 -1.10115504e+00 7.66929090e-01 2.28695601e-01 5.45144439e-01 -8.42521489e-01 -8.21641181e-03 -4.92469847e-01 1.18492261e-01 1.07241297e+00 1.21019554e+00 1.20499671e+00 1.51497601e-02 -3.21488194e-02 1.03096616e+00 1.56711984e+00 2.47432739e-01 1.12542748e+00 4.90278661e-01 5.82440972e-01 8.04183245e-01 3.10895175e-01 4.27516848e-01 4.51146036e-01 2.94328779e-01 5.68018317e-01 -5.31659126e-01 -8.38760555e-01 3.52184296e-01 -2.04261318e-01 1.91206217e-01 -7.41255581e-01 -8.18306673e-03 -1.20329130e+00 7.22729862e-01 -1.31012928e+00 -7.44837165e-01 -7.57139266e-01 2.09378648e+00 5.42324007e-01 -3.19985121e-01 1.26345441e-01 -5.58820367e-01 9.33959723e-01 -2.96326995e-01 -8.60363007e-01 -1.99969679e-01 -4.23001438e-01 2.76810884e-01 6.96167707e-01 3.06547821e-01 -1.28306067e+00 7.98350036e-01 6.13939619e+00 1.06280208e-01 -1.25251150e+00 -8.16321000e-02 6.76155627e-01 -3.33186299e-01 3.83211821e-01 -9.96036828e-02 -6.53348446e-01 4.62250441e-01 8.28256965e-01 1.31961182e-01 2.39578068e-01 3.31213959e-02 7.31235027e-01 -4.58506405e-01 -4.55900252e-01 7.93330073e-01 -1.16731338e-01 -1.51242709e+00 9.07716379e-02 3.61162692e-01 7.82659173e-01 6.40277386e-01 2.99435198e-01 -5.40341973e-01 8.73887073e-03 -1.14477921e+00 -2.30798706e-01 1.28731704e+00 1.12945127e+00 -1.91545293e-01 9.49177384e-01 -5.48702478e-01 -5.16897261e-01 -2.23563150e-01 -3.99545014e-01 -8.61988775e-03 -5.51959500e-02 7.92304516e-01 -8.76488268e-01 4.81946737e-01 8.27443957e-01 1.23272574e+00 -1.13492644e+00 2.13501287e+00 -1.69935837e-01 6.26987755e-01 4.85136509e-02 5.93977869e-01 2.27439657e-01 -5.88130057e-01 7.13163376e-01 7.44914353e-01 5.00129938e-01 1.18171871e-01 -2.58849442e-01 9.50711668e-01 3.45258951e-01 3.24690968e-01 -3.93550038e-01 -1.57112584e-01 2.14532152e-01 1.27450335e+00 -4.05235231e-01 -1.51215017e-01 -7.46367514e-01 1.86485931e-01 -2.14531243e-01 7.11356580e-01 -5.13609588e-01 -4.52439278e-01 1.04270327e+00 4.88679975e-01 -6.82934001e-02 5.45937009e-02 -6.05589688e-01 -9.28532541e-01 1.29885331e-01 -5.93496680e-01 2.68109888e-01 -1.17694569e+00 -1.42242718e+00 5.37245572e-01 -5.01909435e-01 -1.39526796e+00 2.40778416e-01 -8.32929313e-01 -4.93030757e-01 1.43511486e+00 -2.13254690e+00 -9.83866632e-01 -7.82723546e-01 4.62923259e-01 3.33796702e-02 -6.80428445e-01 3.96693528e-01 2.58727551e-01 -9.17096853e-01 1.12599351e-01 -7.73375854e-02 3.94125015e-01 9.32747185e-01 -1.18949592e+00 1.96718052e-01 1.00423717e+00 -6.75566137e-01 8.16021323e-01 4.21664715e-02 -6.80690467e-01 -6.39601231e-01 -1.44563377e+00 7.24960744e-01 -2.68698514e-01 7.95346022e-01 8.37749541e-01 -8.11016917e-01 6.11657619e-01 1.25103101e-01 2.67398298e-01 7.77743220e-01 -1.74149305e-01 -3.09434365e-02 -6.81602955e-02 -1.10286880e+00 6.16306841e-01 8.35250616e-01 -2.71492243e-01 -2.44096622e-01 5.61854482e-01 1.88409939e-01 -7.47016013e-01 -1.19418943e+00 5.62313676e-01 4.95361686e-01 -1.56902766e+00 8.64578247e-01 -7.58167565e-01 7.24249959e-01 -2.93594182e-01 1.60772890e-01 -8.53336751e-01 -2.59165585e-01 -5.85362375e-01 8.09127912e-02 6.65411174e-01 2.28072122e-01 -1.26710057e+00 5.27310371e-01 4.59825397e-01 -6.05747938e-01 -8.26922417e-01 -5.41405916e-01 -2.87013024e-01 -3.03784423e-02 -6.65607527e-02 3.59732717e-01 7.04496026e-01 -8.91133547e-01 -3.13815594e-01 1.46396995e-01 5.19627810e-01 7.07495868e-01 1.44413754e-01 5.36507130e-01 -1.47649825e+00 2.54135817e-01 -7.73590207e-01 -9.08279538e-01 -4.19290811e-01 -2.83790350e-01 -9.57843065e-01 -7.17600286e-01 -2.08225036e+00 1.63365275e-01 -5.22241890e-01 -3.15222025e-01 4.27260935e-01 -2.11351402e-02 5.97593427e-01 -1.73723698e-01 3.56466711e-01 -8.99908766e-02 -2.18901411e-01 1.83788669e+00 1.01238407e-01 -4.82909054e-01 4.05633807e-01 -1.04825354e+00 1.01587009e+00 1.18698049e+00 1.50244549e-01 -3.63900930e-01 -5.58680177e-01 9.58293676e-02 -1.15195222e-01 1.03144622e+00 -9.92210448e-01 2.08143473e-01 -1.68966278e-02 3.11741918e-01 -4.42574203e-01 2.32719649e-02 -1.10695727e-01 -1.42862618e-01 2.12846741e-01 -1.34321466e-01 -4.44343954e-01 3.01452637e-01 2.43149325e-01 -3.41020674e-01 -2.17091432e-03 1.02511334e+00 -2.66529560e-01 -8.90273988e-01 6.45861268e-01 -3.64604235e-01 2.18249887e-01 9.82956409e-01 -6.02735460e-01 -1.10948527e+00 1.73085347e-01 -9.43142891e-01 1.93296701e-01 4.56180841e-01 1.76355690e-01 7.43047893e-01 -6.30007744e-01 -1.21096087e+00 2.83149034e-01 1.85445487e-01 -4.70672324e-02 6.60857141e-01 1.58930659e+00 -1.00034344e+00 6.41395628e-01 -6.61915481e-01 -6.62770450e-01 -1.53115284e+00 -2.69549012e-01 1.09258389e+00 3.70178401e-01 -8.19940686e-01 6.88938320e-01 -7.26285204e-02 2.30085269e-01 -1.37456447e-01 -6.25359237e-01 -5.58540344e-01 2.24775210e-01 7.22971499e-01 8.14990282e-01 2.11565554e-01 -4.10994142e-01 4.79631908e-02 1.03400135e+00 -2.71522969e-01 5.16212106e-01 1.16340935e+00 -5.16957879e-01 -8.21365714e-01 -5.98264635e-02 7.87647784e-01 -9.28100795e-02 -1.01390183e+00 -3.56140882e-01 -3.55353743e-01 -6.37910187e-01 3.91767412e-01 -1.15730369e+00 -1.21203792e+00 8.69849980e-01 1.15562153e+00 2.94423312e-01 1.33234656e+00 -2.16162398e-01 6.21268034e-01 -9.73198563e-02 -7.24395411e-03 -6.17736638e-01 -5.24048805e-01 2.55935103e-01 5.45506775e-01 -1.33861780e+00 5.82655780e-02 -7.73423493e-01 -5.72289109e-01 1.15163422e+00 5.87208569e-01 -1.54063776e-01 5.03276527e-01 -3.47924143e-01 4.55465615e-01 -3.87781084e-01 -3.95073801e-01 -7.95668781e-01 7.40425646e-01 9.34643745e-01 4.79357749e-01 1.59969836e-01 -3.26095462e-01 1.09101295e-01 -1.78563192e-01 3.68031710e-01 1.07972479e+00 2.90493339e-01 -6.25948846e-01 -6.38751864e-01 3.59256715e-02 1.18039143e+00 -7.10572481e-01 -3.29234481e-01 -4.11475092e-01 7.25978076e-01 4.88407224e-01 1.02395916e+00 2.10065708e-01 1.37571439e-01 1.72543630e-01 -2.09648430e-01 3.85163099e-01 -8.25628757e-01 -3.97956342e-01 4.09061462e-02 3.53184342e-01 -9.27231848e-01 -7.76430190e-01 -5.80134392e-01 -7.72178352e-01 -9.60867479e-02 2.44972408e-01 -5.97536564e-01 5.39470613e-01 5.49883842e-01 7.28813291e-01 6.48133218e-01 2.62856692e-01 -2.49368310e-01 -9.20338035e-02 -9.84998405e-01 -9.33976889e-01 -8.68579000e-02 8.00802588e-01 -5.81393778e-01 -2.55100608e-01 2.73935646e-01]
[15.819971084594727, -3.995516300201416]
839abc65-045e-4021-9c15-4bf679a3d224
link-prediction-without-graph-neural-networks
2305.13656
null
https://arxiv.org/abs/2305.13656v1
https://arxiv.org/pdf/2305.13656v1.pdf
Link Prediction without Graph Neural Networks
Link prediction, which consists of predicting edges based on graph features, is a fundamental task in many graph applications. As for several related problems, Graph Neural Networks (GNNs), which are based on an attribute-centric message-passing paradigm, have become the predominant framework for link prediction. GNNs have consistently outperformed traditional topology-based heuristics, but what contributes to their performance? Are there simpler approaches that achieve comparable or better results? To answer these questions, we first identify important limitations in how GNN-based link prediction methods handle the intrinsic class imbalance of the problem -- due to the graph sparsity -- in their training and evaluation. Moreover, we propose Gelato, a novel topology-centric framework that applies a topological heuristic to a graph enhanced by attribute information via graph learning. Our model is trained end-to-end with an N-pair loss on an unbiased training set to address class imbalance. Experiments show that Gelato is 145% more accurate, trains 11 times faster, infers 6,000 times faster, and has less than half of the trainable parameters compared to state-of-the-art GNNs for link prediction.
['Ambuj Singh', 'Arlei Silva', 'Mert Kosan', 'Zexi Huang']
2023-05-23
null
null
null
null
['link-prediction']
['graphs']
[ 4.26372327e-02 3.89276773e-01 -8.13996613e-01 -1.92783728e-01 -3.87893856e-01 -2.34651461e-01 2.48998225e-01 7.20210969e-01 -3.59027162e-02 9.55336750e-01 -1.95379317e-01 -7.16308832e-01 -6.29358947e-01 -1.39315248e+00 -9.05177116e-01 -2.54157394e-01 -8.20868254e-01 1.19354689e+00 4.41380918e-01 -3.10678035e-01 1.89484611e-01 3.73476297e-01 -1.17667401e+00 -1.91268753e-02 1.01915038e+00 9.21247423e-01 -3.06324154e-01 6.94883823e-01 -2.79975027e-01 7.78595209e-01 -4.45508137e-02 -8.24448884e-01 2.59040833e-01 -2.01341882e-01 -9.88163829e-01 -4.39041466e-01 8.86727095e-01 1.15978606e-01 -8.53281260e-01 1.00675118e+00 4.13037211e-01 -1.93805665e-01 4.88929868e-01 -1.68214083e+00 -4.45974767e-01 9.16491687e-01 -5.86378753e-01 2.16656938e-01 2.78487325e-01 -2.94096380e-01 1.61941874e+00 -4.16936666e-01 9.12830353e-01 1.22426391e+00 1.14514804e+00 1.61961779e-01 -1.61631751e+00 -4.80990887e-01 1.92960978e-01 5.91598272e-01 -1.28488767e+00 3.23020592e-02 9.65120256e-01 -1.56084031e-01 9.10980105e-01 2.14880809e-01 9.66413558e-01 8.01887155e-01 3.07847738e-01 7.03169286e-01 5.21215677e-01 -2.24451572e-01 -9.85523164e-02 -2.44397804e-01 3.63957733e-02 1.17718077e+00 7.28526592e-01 6.52590469e-02 -5.90881646e-01 -3.31539482e-01 6.66042686e-01 -1.87970340e-01 -3.29699069e-01 -9.54900980e-01 -1.15669358e+00 1.04647136e+00 8.59215856e-01 -1.92425177e-01 -1.03593595e-01 3.86431336e-01 6.24514163e-01 5.71284771e-01 7.39717782e-01 5.38884521e-01 -5.52223802e-01 4.84655574e-02 -6.48415089e-01 5.98811917e-02 1.40075564e+00 9.31151092e-01 8.27856421e-01 -2.06142426e-01 5.48339486e-02 7.36177921e-01 2.81670481e-01 9.48157683e-02 -2.33469531e-01 -4.30965602e-01 7.60104597e-01 9.63421524e-01 -7.46609271e-01 -1.45855868e+00 -5.96601844e-01 -9.71711099e-01 -1.12163949e+00 -1.12363659e-01 4.64178622e-01 1.41235232e-01 -8.05233061e-01 1.75754523e+00 4.32948709e-01 2.40631655e-01 -5.00825703e-01 5.26802123e-01 7.27768540e-01 3.02798271e-01 -3.43210474e-02 5.65688033e-03 7.93011785e-01 -1.01739788e+00 -3.01015735e-01 -3.43629867e-02 1.04574358e+00 -4.87286627e-01 6.73387587e-01 2.63871074e-01 -8.76644075e-01 -9.64911282e-02 -1.16853774e+00 -8.77981726e-03 -6.40195966e-01 -4.72347856e-01 1.05313528e+00 6.05412066e-01 -1.37302113e+00 1.10393465e+00 -5.35455108e-01 -7.02748418e-01 5.57440996e-01 5.57760775e-01 -4.52098399e-01 -2.45538950e-01 -1.18237078e+00 8.09171677e-01 5.80811799e-01 -1.30701303e-01 -3.86628062e-01 -8.63547683e-01 -8.61338913e-01 1.97085291e-01 5.78731537e-01 -1.05207157e+00 5.01397789e-01 -5.23937762e-01 -9.64937627e-01 6.04078710e-01 1.69045880e-01 -7.01523662e-01 4.44715530e-01 3.01219314e-01 -5.24875641e-01 2.19438002e-01 1.35650307e-01 5.18622756e-01 3.02982450e-01 -1.00805640e+00 -5.50779819e-01 -2.40722373e-01 -4.44672853e-02 -5.40526621e-02 -4.79070097e-01 -5.86591661e-01 -5.92300594e-01 -5.85985005e-01 4.33521837e-01 -1.01735508e+00 -1.99173704e-01 1.85605839e-01 -8.91143441e-01 -3.68926108e-01 7.29863286e-01 -4.55213577e-01 1.34886897e+00 -1.46239316e+00 -2.59791743e-02 9.35032010e-01 9.35613394e-01 2.69866884e-01 -4.20233130e-01 7.99669206e-01 -1.40224338e-01 2.33342424e-01 -8.47758874e-02 -4.83843014e-02 -2.70320736e-02 1.52945623e-01 4.45453748e-02 5.22799671e-01 4.14717719e-02 8.39411378e-01 -1.04435575e+00 -7.60656953e-01 3.21741216e-02 3.85742366e-01 -6.02282226e-01 -1.21817045e-01 -3.88131559e-01 4.23489101e-02 -3.13231170e-01 7.88696408e-01 6.11641526e-01 -6.74851835e-01 6.85630620e-01 -1.97974622e-01 5.21412194e-01 4.01099920e-01 -1.00734150e+00 1.41198194e+00 -2.06761524e-01 7.87759244e-01 -1.89879984e-01 -1.44093239e+00 1.00636721e+00 -5.00699766e-02 6.03758991e-01 -6.37507558e-01 -9.12569389e-02 2.66947001e-01 2.09024131e-01 -3.28462809e-01 1.29282117e-01 4.65189815e-01 3.59295428e-01 3.30000520e-01 1.29529729e-01 3.06107521e-01 6.07904792e-01 6.48717821e-01 1.86011434e+00 -5.38427159e-02 1.15309842e-01 -2.86834985e-01 2.28997499e-01 7.24281743e-02 7.20895112e-01 8.09667647e-01 -2.44066603e-02 4.24310327e-01 9.56209362e-01 -9.04117167e-01 -1.09770834e+00 -1.04490125e+00 9.73221809e-02 1.00248039e+00 2.01440737e-01 -7.67595232e-01 -4.14193898e-01 -9.78906274e-01 4.43283796e-01 2.53076196e-01 -5.56328118e-01 -2.52780020e-01 -5.54993033e-01 -8.18674266e-01 5.21346688e-01 4.89367962e-01 2.91869074e-01 -5.83712280e-01 4.68296647e-01 4.64226305e-01 -1.78497717e-01 -1.12559891e+00 -2.06789732e-01 7.52758160e-02 -1.28317738e+00 -1.53833425e+00 -8.88040885e-02 -9.84484076e-01 9.31266546e-01 1.53117791e-01 1.85781670e+00 7.43713558e-01 -3.96746129e-01 1.79312214e-01 -2.10106418e-01 -7.69334659e-02 -1.77563280e-01 6.42230868e-01 -5.35645224e-02 -2.54131019e-01 1.66411638e-01 -1.10527802e+00 -5.62838256e-01 2.72972405e-01 -3.25369954e-01 8.19583610e-02 8.90447021e-01 8.73429298e-01 4.64772582e-01 1.71292543e-01 6.06495857e-01 -1.50394213e+00 6.59024060e-01 -5.47955811e-01 -5.63196182e-01 3.53904754e-01 -1.19186270e+00 2.23673180e-01 6.64758146e-01 -8.53406414e-02 -2.15286344e-01 -1.19919404e-01 -8.18088949e-02 -9.85262394e-02 3.11932057e-01 8.42741787e-01 2.65861899e-02 -6.93099678e-01 5.23816526e-01 -1.89355224e-01 2.69796759e-01 -3.30889881e-01 2.35883653e-01 -6.35204390e-02 2.85860866e-01 -6.48945510e-01 1.16326857e+00 3.15353066e-01 7.88405240e-01 -5.49229383e-01 -6.70177162e-01 -4.35457706e-01 -5.34563065e-01 -3.07398200e-01 2.62698084e-01 -6.33202016e-01 -1.05993688e+00 2.21413791e-01 -9.98936534e-01 -3.34186137e-01 1.05778113e-01 4.03130770e-01 -3.97813678e-01 3.67096812e-01 -8.93184245e-01 -4.47712868e-01 -4.44895178e-01 -7.03045189e-01 5.38124144e-01 -3.87622900e-02 1.43828681e-02 -1.50042880e+00 4.00065817e-02 2.07745612e-01 5.37032306e-01 4.97298360e-01 1.48984754e+00 -9.78841424e-01 -9.50285971e-01 -4.54038650e-01 -6.06457114e-01 -1.60109043e-01 -1.96118340e-01 7.56264776e-02 -3.88284862e-01 -4.21298087e-01 -9.78936434e-01 -2.17958316e-01 9.87956464e-01 2.34015435e-01 1.35360813e+00 -4.21530575e-01 -9.46896017e-01 6.54681087e-01 1.68767798e+00 -5.91954887e-01 6.35026097e-01 3.21897924e-01 1.05306911e+00 5.32949448e-01 2.87622958e-01 4.97496873e-02 8.32397699e-01 4.83328015e-01 9.26603436e-01 -3.57478440e-01 -3.96666825e-01 -6.48796380e-01 -1.25439048e-01 1.00410402e+00 -1.22467853e-01 -5.56332529e-01 -9.81842279e-01 4.87702370e-01 -2.08414507e+00 -7.99566865e-01 -7.30821967e-01 2.30545020e+00 6.51296258e-01 5.58628261e-01 3.04582238e-01 1.88725948e-01 1.04196465e+00 8.85834694e-02 -4.18976337e-01 -1.36675537e-01 -4.52316962e-02 7.35065043e-02 8.09738874e-01 4.02879030e-01 -9.48666215e-01 7.71907747e-01 6.37374258e+00 9.22729135e-01 -8.87028575e-01 -3.47426176e-01 7.78248847e-01 2.18219683e-01 -4.12154108e-01 2.05806300e-01 -4.88147259e-01 3.21516693e-01 8.27959418e-01 -1.18534215e-01 5.57929397e-01 8.51858556e-01 -1.31371826e-01 2.02417746e-01 -1.41947865e+00 7.86408126e-01 3.60109508e-02 -1.78059876e+00 1.01942763e-01 3.64177436e-01 7.14049399e-01 4.10644531e-01 -1.26119047e-01 3.95579159e-01 5.67687213e-01 -1.06283712e+00 1.33847252e-01 5.16615152e-01 6.00319922e-01 -7.26353407e-01 8.51151228e-01 4.82624955e-02 -1.47431803e+00 1.17460936e-01 -5.48346341e-01 4.87321541e-02 -4.77031693e-02 1.23388672e+00 -1.14398420e+00 8.41619670e-01 5.55412352e-01 7.62957454e-01 -7.95986950e-01 1.65085626e+00 -5.60693368e-02 7.92949021e-01 -4.93313521e-01 -2.38350600e-01 -7.23361224e-02 -1.60009474e-01 6.53707981e-01 8.45423341e-01 1.15046673e-01 -4.69861954e-01 4.29406792e-01 5.61148584e-01 -6.62208796e-01 1.58813968e-01 -8.36645842e-01 5.79362847e-02 7.30338454e-01 1.37601697e+00 -7.47178674e-01 -4.28357832e-02 -3.06072474e-01 4.33009028e-01 8.96903634e-01 1.74898654e-01 -6.46093547e-01 -7.68202186e-01 3.67545336e-01 4.10332412e-01 2.36219928e-01 -6.00603521e-02 -1.73874930e-01 -6.87801957e-01 1.68127432e-01 -6.57904625e-01 7.07653165e-01 -3.55095565e-01 -1.76395226e+00 5.18621802e-01 -4.55810428e-01 -1.18446159e+00 -7.21709803e-02 -7.42509246e-01 -8.97991180e-01 4.32037085e-01 -1.74139118e+00 -1.26971972e+00 -2.67646015e-01 2.67224699e-01 -1.94502607e-01 -6.17763810e-02 7.06396461e-01 6.09982073e-01 -4.93361145e-01 7.83368349e-01 4.66149040e-02 4.38183039e-01 7.06520975e-01 -1.53112042e+00 5.70015967e-01 5.23632467e-01 2.39143163e-01 2.93676823e-01 6.15303934e-01 -8.50789845e-01 -1.57717967e+00 -1.34579897e+00 1.21345294e+00 -3.19118530e-01 8.73399734e-01 -2.70345837e-01 -8.56510699e-01 6.85445130e-01 -2.29228318e-01 6.16876841e-01 6.54875934e-01 9.28946078e-01 -5.13065338e-01 -4.75798219e-01 -1.05919611e+00 5.44543028e-01 1.67688668e+00 -3.27678800e-01 8.94800201e-02 5.88211358e-01 7.03287005e-01 -3.40535223e-01 -1.19893956e+00 6.02453411e-01 4.17885274e-01 -9.37703311e-01 9.66232836e-01 -8.06696117e-01 2.84897357e-01 -1.35953784e-01 2.40060732e-01 -1.40422499e+00 -4.91913557e-01 -6.39770210e-01 -3.33671927e-01 1.16011751e+00 8.92347336e-01 -9.08536494e-01 1.35156441e+00 -4.72403280e-02 -2.35617999e-02 -1.25481176e+00 -8.30297589e-01 -9.34240341e-01 -2.89142132e-01 -2.37699091e-01 6.34928524e-01 1.22856843e+00 1.37204692e-01 5.88069260e-01 -2.52648145e-01 1.51138246e-01 1.07243598e+00 7.93915335e-03 8.98873389e-01 -2.05327654e+00 -6.31423071e-02 -5.96373975e-01 -9.41724062e-01 -8.30773532e-01 3.46319914e-01 -1.41634285e+00 -2.76983649e-01 -1.89102399e+00 2.02802792e-01 -8.18592370e-01 -2.93371975e-01 4.09740299e-01 -1.00307092e-01 3.54601175e-01 -2.39268422e-01 -1.18716992e-02 -1.03759289e+00 4.98877704e-01 1.05350912e+00 -2.54930258e-01 1.01471685e-01 9.03183818e-02 -5.52200615e-01 5.17242730e-01 7.11850345e-01 -5.11258900e-01 -3.66356641e-01 -2.91542917e-01 8.84735107e-01 5.95582835e-02 3.57168317e-01 -1.09984577e+00 5.67563474e-01 1.34360760e-01 3.46759737e-01 -6.75303161e-01 2.07957812e-03 -7.23374665e-01 -7.97029771e-03 6.00576818e-01 -3.12681735e-01 2.83389509e-01 -1.17343418e-01 9.84711707e-01 2.50394065e-02 2.29701564e-01 2.80755401e-01 2.71942198e-01 -4.92334336e-01 9.24287021e-01 1.69978425e-01 2.47221202e-01 7.71077871e-01 -2.20608845e-01 -7.77044356e-01 -5.20532429e-01 -4.01592106e-01 5.99206328e-01 2.46687323e-01 1.54366791e-01 5.08439124e-01 -1.54600108e+00 -8.42404544e-01 -1.06439389e-01 3.12174916e-01 -1.15826733e-01 -9.99426022e-02 1.03266823e+00 -7.00711191e-01 2.89831161e-01 -2.81749200e-03 -5.92980266e-01 -1.08175325e+00 4.90245640e-01 2.79116035e-01 -6.93612278e-01 -5.89956284e-01 8.64434242e-01 -3.74512643e-01 -8.53483021e-01 3.45953822e-01 2.18450859e-01 1.53124914e-01 -2.74160147e-01 -2.07740217e-01 5.83115578e-01 3.55023354e-01 -1.70743957e-01 -2.74844438e-01 3.14347208e-01 -1.59897327e-01 5.44614136e-01 1.52842927e+00 5.06501719e-02 -5.20363390e-01 5.80103546e-02 1.20897329e+00 -9.69241858e-02 -7.47988105e-01 -3.74992430e-01 2.82880396e-01 -4.28216249e-01 -3.54399122e-02 -7.76666105e-01 -1.35496747e+00 5.57894289e-01 2.09448427e-01 7.11827159e-01 6.18164420e-01 5.47950231e-02 8.80986571e-01 5.06264389e-01 3.79249662e-01 -1.19127524e+00 3.31943342e-03 4.28198695e-01 4.79614586e-01 -1.29607093e+00 2.59870797e-01 -1.10427761e+00 1.45455837e-01 1.25410795e+00 8.49571645e-01 -5.06567582e-02 7.36744404e-01 -1.92132622e-01 -4.05462980e-01 -4.24758911e-01 -1.00638938e+00 -2.46895254e-01 3.25576663e-01 7.67445505e-01 3.80832523e-01 -1.21736284e-02 -1.77106321e-01 -6.91816583e-02 -2.08547473e-01 -3.80973965e-01 3.36886495e-01 5.03306150e-01 -3.61040384e-01 -1.36361921e+00 7.53844827e-02 1.21189344e+00 -1.24841727e-01 -2.23100349e-01 -4.43114460e-01 9.35129166e-01 -2.92568356e-01 7.64705718e-01 9.03758407e-02 -6.94707930e-01 9.92551371e-02 -3.20758671e-01 4.19154674e-01 -4.74797130e-01 -1.59884751e-01 -6.88495755e-01 7.35446692e-01 -6.97002411e-01 1.93107687e-02 -1.83938876e-01 -1.01474547e+00 -1.12790191e+00 -5.76875210e-01 3.21086019e-01 5.96787155e-01 8.21070731e-01 6.40981436e-01 5.60439408e-01 6.71375275e-01 -4.20919567e-01 -3.70472461e-01 -6.50567830e-01 -5.33228219e-01 2.97055602e-01 4.48544659e-02 -6.19591773e-01 -4.32553291e-01 -7.16785729e-01]
[7.04879093170166, 6.193018436431885]
61519088-dd04-4a14-b701-9da19a198ed3
deep-learning-on-implicit-neural-datasets
2206.01178
null
https://arxiv.org/abs/2206.01178v3
https://arxiv.org/pdf/2206.01178v3.pdf
Discretization Invariant Learning on Neural Fields
While neural fields have emerged as powerful representations of continuous data, there is a need for neural networks that can perform inference on such data without being sensitive to how the field is sampled, a property called discretization invariance. We develop DI-Net, a framework for learning discretization invariant operators on neural fields of any type. Whereas current theoretical analyses of discretization invariant networks are restricted to the limit of infinite samples, our analysis does not require infinite samples and establishes upper bounds on the variation in DI-Net outputs given different finite discretizations. Our framework leads to a family of neural networks driven by numerical integration via quasi-Monte Carlo sampling with discretizations of low discrepancy. DI-Nets manifest desirable theoretical properties such as universal approximation of a large class of maps between $L^2$ functions, and gradients that are also discretization invariant. DI-Nets can also be seen as generalizations of many existing network families as they bridge discrete and continuous network classes, such as convolutional neural networks (CNNs) and neural operators respectively. Experimentally, DI-Nets derived from CNNs can learn to classify and segment visual data represented by neural fields under various discretizations, and sometimes even generalize to new types of discretizations at test time. Code: https://github.com/clintonjwang/DI-net.
['Polina Golland', 'Clinton J. Wang']
2022-06-02
null
null
null
null
['numerical-integration']
['miscellaneous']
[ 1.78911939e-01 1.38137594e-01 -1.85523391e-01 -6.92853391e-01 -4.24998134e-01 -5.95507801e-01 3.83876741e-01 -1.62861440e-02 -3.05911869e-01 8.14296842e-01 -2.03125954e-01 -3.76660228e-01 -2.95019716e-01 -1.13579416e+00 -1.09549654e+00 -6.06176555e-01 -4.81028289e-01 3.19322884e-01 2.40620658e-01 -1.81484938e-01 -1.60143867e-01 8.05405557e-01 -1.45703065e+00 3.41211587e-01 9.09724116e-01 1.45449066e+00 -6.04480445e-01 7.48755574e-01 1.85394570e-01 6.14890218e-01 -2.23689556e-01 3.35456021e-02 6.74816728e-01 -3.58816743e-01 -8.89030218e-01 -1.10488936e-01 8.33940089e-01 -2.45006546e-01 -4.89003241e-01 1.52119029e+00 2.21290439e-01 3.62168521e-01 1.10922086e+00 -1.30353916e+00 -1.16770220e+00 6.01876974e-01 -5.10196872e-02 1.52043581e-01 -1.43495739e-01 1.06025882e-01 9.06824112e-01 -6.61491573e-01 5.45188546e-01 1.30686271e+00 1.33895135e+00 5.52393079e-01 -1.70214832e+00 -3.98399264e-01 -8.81137326e-02 -3.39221418e-01 -1.24825060e+00 -1.12740174e-01 5.46829998e-01 -7.11257041e-01 6.14755869e-01 2.70562917e-01 6.19416356e-01 7.73234904e-01 4.19811577e-01 6.22028232e-01 1.09965134e+00 -4.21162546e-01 5.76412618e-01 8.17323104e-03 3.22662234e-01 8.55430245e-01 2.41470516e-01 4.63546924e-02 -1.96353048e-02 -1.99268982e-01 1.27139544e+00 2.42956132e-01 -4.29604948e-01 -4.47521001e-01 -1.04804623e+00 1.11610103e+00 7.27440894e-01 2.37874612e-01 -2.52481252e-01 4.41123307e-01 7.10317075e-01 5.87531805e-01 5.79510450e-01 3.76109749e-01 -3.73460829e-01 2.97213346e-01 -8.07877183e-01 4.79444325e-01 7.62524962e-01 1.04462326e+00 8.41028810e-01 2.16187865e-01 -3.48389328e-01 6.11818492e-01 -1.35242820e-01 3.03166062e-01 3.47814769e-01 -1.16873562e+00 6.66615590e-02 5.34882843e-01 -1.16698764e-01 -8.43137801e-01 -3.93774122e-01 -2.35108182e-01 -1.36734223e+00 5.51431298e-01 6.19608581e-01 -2.05131680e-01 -1.08815658e+00 1.89088190e+00 -6.71114251e-02 -3.18745136e-01 -1.36733890e-01 5.68410575e-01 2.54779786e-01 5.96406519e-01 -1.89171642e-01 7.98573997e-03 8.40235353e-01 -4.21150178e-01 -3.95059615e-01 2.93752849e-01 6.95589721e-01 -1.06223553e-01 1.16192818e+00 4.01425004e-01 -1.21778965e+00 -7.52658665e-01 -1.02093291e+00 -1.89312428e-01 -7.60427535e-01 -1.32440209e-01 6.98207259e-01 4.12192523e-01 -1.33593261e+00 1.11513042e+00 -8.94635856e-01 -2.14595437e-01 9.18600619e-01 4.41352427e-01 -1.58731982e-01 1.27012268e-01 -1.30872846e+00 5.60434937e-01 6.19676352e-01 2.27123097e-01 -6.98845565e-01 -8.74671638e-01 -8.49655211e-01 4.44027893e-02 -2.51999944e-01 -3.53993714e-01 1.30055118e+00 -1.15040839e+00 -1.19445038e+00 6.78859711e-01 2.84642398e-01 -9.05026495e-01 6.58009112e-01 2.11226463e-01 -4.10911620e-01 2.75610797e-02 -1.99087784e-02 9.28495944e-01 7.76322603e-01 -7.61012375e-01 -3.99583220e-01 -9.83562917e-02 3.25541347e-01 -3.53498280e-01 -4.05014843e-01 -4.99157935e-01 3.57354403e-01 -6.33717835e-01 7.90612847e-02 -8.50915492e-01 -3.25406104e-01 7.98445642e-01 -4.31178629e-01 -3.35185289e-01 7.69620955e-01 -2.75015146e-01 8.93727243e-01 -2.09396052e+00 -1.37748167e-01 2.27815077e-01 3.02443206e-01 2.72088110e-01 7.66229630e-02 1.55182779e-01 -1.88163504e-01 2.88614601e-01 -6.77707136e-01 2.55789518e-01 3.30316812e-01 3.44349176e-01 -3.66888583e-01 7.50533283e-01 2.89979666e-01 1.01318586e+00 -7.92600989e-01 -3.40235978e-01 1.05714388e-01 4.97703522e-01 -6.90437138e-01 -1.77860111e-01 -5.35475671e-01 1.02410734e-01 -2.86864877e-01 3.83727163e-01 8.41771066e-01 -3.70809287e-01 -1.83119044e-01 -1.10702731e-01 -1.82168372e-02 -7.73694515e-02 -1.17030942e+00 1.40694618e+00 -3.28204662e-01 9.33893681e-01 4.00016941e-02 -1.67289495e+00 9.21438098e-01 3.13459843e-01 2.42969170e-01 -3.71081918e-01 2.60695100e-01 3.14880908e-01 6.55615004e-03 -2.70210415e-01 1.13159835e-01 -3.61292660e-01 -4.04274575e-02 1.14255629e-01 1.81781471e-01 -7.76528642e-02 3.80874515e-01 -1.09681241e-01 8.43645275e-01 -1.88467279e-01 1.66315347e-01 -7.53712237e-01 2.62073755e-01 -8.68424401e-02 4.69510615e-01 9.98069823e-01 -1.57412380e-01 5.20973146e-01 7.39933133e-01 -8.32536817e-01 -1.17464066e+00 -1.37574470e+00 -9.14867401e-01 8.30160677e-01 -8.87046084e-02 3.11133802e-01 -7.45232403e-01 -5.10336876e-01 2.91417897e-01 4.36556607e-01 -1.06836677e+00 -2.58563370e-01 -3.27468932e-01 -4.98251915e-01 9.33780313e-01 1.06122339e+00 8.20518017e-01 -1.02935851e+00 -6.06348991e-01 5.26160151e-02 3.96260411e-01 -8.02697837e-01 -4.35919404e-01 6.32452369e-01 -1.10393620e+00 -1.00670302e+00 -7.78506994e-01 -9.78207409e-01 7.40723968e-01 -4.60176557e-01 1.01250100e+00 -2.05609143e-01 -3.37085873e-01 2.37154543e-01 2.23123416e-01 -3.55600208e-01 -5.20505488e-01 1.12533711e-01 3.29228461e-01 5.50045930e-02 1.03118666e-01 -8.33707929e-01 -5.43546677e-01 2.23517761e-01 -1.29070771e+00 -2.92905360e-01 2.46830866e-01 1.07673609e+00 7.31797636e-01 2.17659950e-01 6.15877867e-01 -9.93877470e-01 7.01049268e-01 -3.96031231e-01 -8.15269470e-01 2.30237290e-01 -3.06028396e-01 4.96786773e-01 1.14706635e+00 -7.19556928e-01 -6.76581979e-01 6.50541931e-02 -3.08570638e-02 -5.91121137e-01 -2.77950943e-01 3.83234352e-01 1.08301878e-01 -3.36873889e-01 1.23155463e+00 -2.63338089e-02 5.44805713e-02 -1.95147529e-01 4.30477113e-01 5.71671188e-01 6.50865138e-01 -7.59168804e-01 5.42369306e-01 7.48894572e-01 3.48865569e-01 -8.59083652e-01 -8.62445712e-01 -6.43012673e-02 -8.33466053e-01 1.86215211e-02 8.61648083e-01 -4.02989447e-01 -6.91263020e-01 5.46369135e-01 -1.06136417e+00 -6.58288479e-01 -8.72247636e-01 3.44053090e-01 -7.44841039e-01 1.36854053e-02 -8.05794895e-01 -6.67991936e-01 -1.09380595e-01 -9.61321235e-01 5.84959805e-01 1.12323023e-01 -9.96424705e-02 -1.58809793e+00 -5.82003109e-02 -5.61809361e-01 5.19021630e-01 7.80744076e-01 1.04232705e+00 -6.25570714e-01 -2.83092380e-01 -4.81849134e-01 -2.62588084e-01 1.03951108e+00 3.86964940e-02 5.10590263e-02 -9.29643333e-01 -2.97819287e-01 -9.51057896e-02 -6.27240539e-01 8.85083795e-01 8.32996190e-01 1.63177025e+00 -5.26214242e-01 -1.95593014e-01 9.48906720e-01 1.53399348e+00 1.86851472e-01 4.91899252e-01 -2.26934850e-01 3.53297919e-01 1.05234183e-01 -2.03341350e-01 4.42529470e-02 -2.64659762e-01 3.17626595e-01 3.53376210e-01 -1.77114949e-01 2.20714465e-01 -1.42525733e-01 9.48816314e-02 4.92695332e-01 -7.38206059e-02 4.25418699e-03 -8.35068524e-01 5.91506064e-01 -1.70992029e+00 -8.65454733e-01 -8.66673514e-02 2.12378359e+00 9.80448425e-01 4.45175052e-01 1.66449741e-01 2.23477051e-01 8.18265557e-01 -7.17869475e-02 -9.38096285e-01 -7.70974278e-01 -8.08015913e-02 5.14497757e-01 7.73259521e-01 4.99751687e-01 -1.24326217e+00 4.72461313e-01 6.93652391e+00 8.31065238e-01 -1.24118328e+00 -1.13834003e-02 7.57311463e-01 3.81375849e-01 -1.59123018e-01 -2.70696640e-01 -5.89104712e-01 2.59367168e-01 7.79719353e-01 -6.90320656e-02 3.07640165e-01 1.07412589e+00 -2.53682464e-01 1.31837666e-01 -1.47690618e+00 6.41475201e-01 -4.22122717e-01 -1.73574710e+00 1.33592591e-01 1.45962700e-01 1.20778263e+00 5.45994472e-03 1.77710697e-01 3.42617720e-01 4.99408096e-01 -1.44940126e+00 5.56789339e-01 6.97565496e-01 1.18393016e+00 -6.42751038e-01 5.83157837e-01 3.42505872e-01 -1.05053926e+00 3.61971036e-02 -7.12681651e-01 -3.33279759e-01 -3.04275930e-01 7.36617386e-01 -3.98394257e-01 1.21419825e-01 6.47246957e-01 7.78302789e-01 -3.20116401e-01 8.70390713e-01 3.42171133e-01 6.73995078e-01 -4.66224134e-01 -1.29925922e-01 3.19801360e-01 -1.12091310e-01 2.52732068e-01 1.15007782e+00 4.99437265e-02 -1.83469281e-02 1.00712173e-01 1.35854781e+00 -1.62512973e-01 -3.54537845e-01 -9.94206429e-01 -3.53086330e-02 3.04454029e-01 9.47444081e-01 -9.00374472e-01 -2.33886689e-01 -2.67907768e-01 7.24620879e-01 4.72766370e-01 6.05398953e-01 -7.60849059e-01 -9.79444861e-01 7.20901191e-01 3.07304919e-01 4.29489762e-01 -2.30973870e-01 -4.06536639e-01 -1.03723168e+00 1.19435690e-01 -4.65691119e-01 3.02288443e-01 -4.94396776e-01 -1.46713316e+00 5.39255381e-01 2.13418871e-01 -1.18463707e+00 -1.27788484e-01 -1.26047719e+00 -6.23807132e-01 7.86752701e-01 -1.01951981e+00 -6.96355522e-01 -1.14636928e-01 6.48929358e-01 4.61291596e-02 9.49746370e-02 7.95700371e-01 8.29175115e-02 -4.40305397e-02 7.73694873e-01 4.77473557e-01 6.95970893e-01 1.10141650e-01 -1.32637072e+00 4.46650982e-01 5.82222939e-01 1.01881728e-01 6.01307750e-01 4.18411613e-01 -2.50855803e-01 -8.99734199e-01 -1.31683183e+00 4.25178438e-01 -1.80298194e-01 5.63427567e-01 -5.10607600e-01 -1.26958561e+00 9.52675164e-01 -1.62280425e-01 8.27004075e-01 5.67802526e-02 -7.89199546e-02 -5.73286176e-01 -1.51953027e-01 -1.42464554e+00 5.22099972e-01 1.18065524e+00 -6.76105559e-01 -2.72126317e-01 4.29363281e-01 7.03826189e-01 -5.37954628e-01 -1.14250374e+00 4.68777746e-01 5.35600781e-01 -1.11379945e+00 9.57854807e-01 -1.01151490e+00 4.87812340e-01 6.35246560e-02 -3.37796003e-01 -1.29871976e+00 -2.93655574e-01 -4.52589780e-01 -1.76448729e-02 7.19931841e-01 2.69565612e-01 -1.02260482e+00 5.15300930e-01 4.85337764e-01 -1.43071800e-01 -1.10893106e+00 -9.86208975e-01 -1.27226484e+00 8.69712055e-01 -5.40799379e-01 4.01393294e-01 9.24421012e-01 -1.69950098e-01 -2.88388908e-01 2.40099058e-01 -1.09451625e-03 6.02715790e-01 -8.70909020e-02 1.53208360e-01 -1.44531727e+00 -1.74211800e-01 -6.53685689e-01 -8.80784690e-01 -1.08982146e+00 2.65841812e-01 -1.34649134e+00 4.09932248e-02 -1.32579720e+00 -2.76405990e-01 -5.91797709e-01 -2.07213357e-01 5.44735610e-01 6.06725991e-01 2.88556635e-01 -3.65872681e-01 4.75480705e-02 -2.71593571e-01 4.41077590e-01 1.34619367e+00 -1.03572868e-01 -1.67247117e-01 1.57129183e-01 -3.44633013e-01 1.12773657e+00 8.52838159e-01 -2.87536263e-01 -4.14382398e-01 -1.65415049e-01 2.53396690e-01 -4.26274911e-03 8.18513215e-01 -1.42342615e+00 2.02024460e-01 -2.36248180e-01 5.46952307e-01 -2.29403153e-01 -6.75362423e-02 -5.62633514e-01 -7.19761401e-02 4.75537896e-01 -9.85594571e-01 -7.64617920e-02 2.28437141e-01 4.71147686e-01 -1.27676696e-01 -2.74270058e-01 1.26002491e+00 -1.89649001e-01 -3.87000382e-01 5.86220145e-01 -3.70159537e-01 8.19467604e-01 6.87760532e-01 -2.76888996e-01 -1.08373575e-01 -3.32844794e-01 -8.52128506e-01 -1.48572743e-01 3.00510198e-01 -1.01845063e-01 2.73620665e-01 -1.69760787e+00 -4.27624941e-01 3.00825298e-01 -1.17241614e-01 4.96265590e-01 -4.45555858e-02 5.90098917e-01 -7.10219443e-01 4.03542310e-01 -4.55436617e-01 -6.90571308e-01 -3.78997803e-01 3.64773244e-01 8.47805560e-01 -3.54765765e-02 -6.89955771e-01 8.48860621e-01 2.34354258e-01 -6.31387472e-01 4.51477528e-01 -1.20429277e+00 5.30895770e-01 -2.21022636e-01 2.04033136e-01 3.97277445e-01 1.42138740e-02 -2.00045601e-01 -1.49819225e-01 3.96108568e-01 2.00015426e-01 3.33532393e-02 1.18312263e+00 5.66475511e-01 -1.06263161e-01 7.71495283e-01 1.72686183e+00 -7.08376944e-01 -1.64198339e+00 -3.26073408e-01 -2.62402683e-01 4.17408906e-03 -1.57526657e-01 -4.02754247e-01 -1.11330950e+00 1.11384368e+00 4.81987804e-01 8.50649059e-01 1.06658721e+00 -1.08418539e-01 6.39044464e-01 6.94657207e-01 7.22611323e-02 -1.16781878e+00 -1.47272255e-02 7.38447309e-01 9.20728385e-01 -1.04558885e+00 -2.30378479e-01 -2.21872758e-02 -1.21844403e-01 1.28914344e+00 4.26928014e-01 -7.27747321e-01 1.23221183e+00 4.46757972e-01 -2.28482649e-01 -2.19049342e-02 -4.45777118e-01 -3.37874307e-03 4.09820884e-01 7.18150437e-01 4.24025536e-01 7.85013437e-02 3.13278921e-02 3.93918753e-01 -1.31596655e-01 3.44583869e-01 3.77138406e-01 8.47002506e-01 -4.12532181e-01 -4.35689598e-01 2.73037460e-02 7.55613148e-01 -1.92388624e-01 -5.47398478e-02 7.38825575e-02 1.05230367e+00 3.35127145e-01 1.50792971e-01 4.42314267e-01 -1.74485803e-01 1.64143264e-01 3.29138070e-01 5.15677392e-01 -3.85641128e-01 -1.93442866e-01 -7.09992945e-01 -3.94188583e-01 -3.00977767e-01 -4.44505632e-01 -6.33027792e-01 -1.40393722e+00 -3.54836881e-01 -1.16825514e-01 -7.17528956e-03 1.80517048e-01 8.07040513e-01 8.67050514e-02 4.67620760e-01 4.99117196e-01 -8.55764508e-01 -1.09316778e+00 -8.48594964e-01 -9.38251317e-01 4.39754874e-01 7.01443493e-01 -5.69325387e-01 -6.36667848e-01 1.86444104e-01]
[7.694057464599609, 3.5835254192352295]
9f7c205a-f4c7-49b3-9de4-4e2d4d5450fb
convert-efficient-and-accurate-conversational
1911.03688
null
https://arxiv.org/abs/1911.03688v2
https://arxiv.org/pdf/1911.03688v2.pdf
ConveRT: Efficient and Accurate Conversational Representations from Transformers
General-purpose pretrained sentence encoders such as BERT are not ideal for real-world conversational AI applications; they are computationally heavy, slow, and expensive to train. We propose ConveRT (Conversational Representations from Transformers), a pretraining framework for conversational tasks satisfying all the following requirements: it is effective, affordable, and quick to train. We pretrain using a retrieval-based response selection task, effectively leveraging quantization and subword-level parameterization in the dual encoder to build a lightweight memory- and energy-efficient model. We show that ConveRT achieves state-of-the-art performance across widely established response selection tasks. We also demonstrate that the use of extended dialog history as context yields further performance gains. Finally, we show that pretrained representations from the proposed encoder can be transferred to the intent classification task, yielding strong results across three diverse data sets. ConveRT trains substantially faster than standard sentence encoders or previous state-of-the-art dual encoders. With its reduced size and superior performance, we believe this model promises wider portability and scalability for Conversational AI applications.
['Ivan Vulić', 'Tsung-Hsien Wen', 'Pei-Hao Su', 'Nikola Mrkšić', 'Iñigo Casanueva', 'Matthew Henderson']
2019-11-09
null
https://aclanthology.org/2020.findings-emnlp.196
https://aclanthology.org/2020.findings-emnlp.196.pdf
findings-of-the-association-for-computational
['conversational-response-selection']
['natural-language-processing']
[ 3.42046529e-01 1.26579776e-01 -3.31650555e-01 -7.32770324e-01 -1.14245808e+00 -5.17303169e-01 7.55223513e-01 4.40470837e-02 -4.94582355e-01 8.44311774e-01 7.64994979e-01 -4.43650931e-01 2.02134758e-01 -7.03438163e-01 -5.37842453e-01 -3.63444477e-01 6.20458648e-02 9.51059937e-01 3.38377245e-02 -7.16398656e-01 1.31746396e-01 -1.09205812e-01 -1.32365358e+00 6.89775109e-01 6.48726642e-01 9.61043358e-01 1.61932945e-01 9.03035820e-01 7.29458258e-02 1.13621068e+00 -5.96793473e-01 -6.24674559e-01 -1.08764254e-01 -5.31332850e-01 -1.42901480e+00 -1.92959428e-01 3.37099046e-01 -8.48000646e-01 -7.49014616e-01 1.40286222e-01 6.68413103e-01 4.74202573e-01 5.20052373e-01 -9.71156657e-01 -1.16564536e+00 7.30738163e-01 -1.34036876e-02 1.72586903e-01 7.13050544e-01 2.49527723e-01 1.49725580e+00 -9.68086481e-01 4.11244839e-01 1.39917696e+00 4.87907887e-01 9.27850664e-01 -1.26919889e+00 -4.44637448e-01 1.51923105e-01 3.79358441e-01 -8.16740155e-01 -9.03580189e-01 5.23170769e-01 2.44289204e-01 1.71577346e+00 3.55987132e-01 3.12523663e-01 1.59637523e+00 1.37389734e-01 1.25413954e+00 6.84997261e-01 -3.03361297e-01 4.83857058e-02 1.01866946e-01 1.46042496e-01 8.23558271e-01 -5.04968584e-01 -3.34648371e-01 -7.79973567e-01 -2.94176221e-01 1.15643889e-01 6.95683658e-02 -4.73507136e-01 9.01011974e-02 -1.02730346e+00 1.16515934e+00 5.82672954e-01 1.43832088e-01 -2.64303595e-01 3.38984787e-01 7.49374092e-01 6.76755488e-01 7.20519125e-01 5.10339141e-01 -4.09616351e-01 -6.22883797e-01 -4.55659240e-01 2.23619640e-01 1.29858112e+00 1.03191590e+00 5.11083901e-01 -2.04214126e-01 -3.93379033e-01 1.22751379e+00 1.98344216e-02 2.36772776e-01 5.79244852e-01 -1.04879355e+00 7.04759002e-01 5.14409542e-01 -1.98648155e-01 -3.83789510e-01 -3.44282061e-01 -1.64763778e-01 -7.61582017e-01 -5.50309777e-01 2.89032096e-03 -2.24236354e-01 -5.33587337e-01 1.89778566e+00 1.02723120e-02 -3.25792171e-02 3.39019835e-01 9.16930079e-01 1.07006276e+00 8.83569002e-01 1.47837475e-01 -1.33166566e-01 1.48994434e+00 -1.42159534e+00 -6.91223800e-01 -5.21025538e-01 7.75595546e-01 -3.92471611e-01 1.38585222e+00 3.71535234e-02 -1.35334063e+00 -3.38237345e-01 -1.11322796e+00 -6.88692331e-01 -3.29313092e-02 6.75147027e-02 1.20111680e+00 3.66751879e-01 -1.20812416e+00 3.88579607e-01 -6.49298072e-01 -3.56795281e-01 7.30659142e-02 4.55200166e-01 -1.93145990e-01 -1.90630585e-01 -1.44391966e+00 1.18620849e+00 -9.61976424e-02 -3.94997783e-02 -9.25293088e-01 -5.12404203e-01 -9.76434469e-01 4.28341687e-01 2.20466748e-01 -9.84408617e-01 1.93093991e+00 -6.85818911e-01 -2.11070132e+00 8.26556265e-01 -3.02638978e-01 -6.83023334e-01 3.36629078e-02 -2.59591281e-01 -1.80497840e-01 2.76401073e-01 -9.54386368e-02 8.38988483e-01 4.66211677e-01 -6.70949996e-01 -4.32907432e-01 -6.35126308e-02 5.51912427e-01 6.93609357e-01 -7.70581484e-01 -1.80366524e-02 -4.04925108e-01 -2.13141337e-01 -4.29675728e-01 -1.11611760e+00 1.18108727e-02 -1.85789272e-01 -2.41971701e-01 -9.20181632e-01 7.22243130e-01 -5.17254591e-01 1.03190136e+00 -1.83948267e+00 4.43942934e-01 -6.16805017e-01 6.17219359e-02 1.23288780e-01 -3.27433467e-01 8.82076979e-01 4.53447908e-01 5.13828211e-02 -2.67704844e-01 -7.66746938e-01 2.12839365e-01 2.94977814e-01 -2.69199967e-01 2.58491546e-01 4.12567049e-01 1.11922896e+00 -8.41709077e-01 -2.38549560e-01 8.71229693e-02 4.48386222e-01 -8.80189478e-01 8.41289759e-01 -3.67491454e-01 1.70720771e-01 -3.82887512e-01 3.58781397e-01 1.75646767e-02 -5.15266120e-01 3.46592784e-01 -5.63097559e-02 3.77280056e-01 1.36123097e+00 -1.78804591e-01 2.12744975e+00 -1.09741187e+00 7.97150493e-01 1.12135202e-01 -9.64049876e-01 7.09288239e-01 5.85083961e-01 1.96182713e-01 -8.95446658e-01 4.45038378e-02 -5.58757829e-03 -1.12004243e-01 -3.79993916e-01 8.90589654e-01 -6.24400340e-02 -6.18832588e-01 1.01258230e+00 5.28428257e-01 -3.97079051e-01 1.03915408e-01 4.73197758e-01 1.30412853e+00 -1.20005727e-01 1.21907130e-01 -6.12550601e-02 4.04124022e-01 -2.49500006e-01 1.99786872e-01 5.54815471e-01 -4.83279042e-02 2.91645318e-01 2.45730981e-01 -4.78715718e-01 -9.04276669e-01 -8.36551130e-01 -7.55235460e-03 1.95197511e+00 -4.32665497e-02 -4.69768047e-01 -6.17563188e-01 -7.40460515e-01 -1.35521472e-01 8.13755453e-01 -4.62108076e-01 -4.31288391e-01 -8.66955280e-01 -6.38796866e-01 6.27474010e-01 6.91201568e-01 6.16487443e-01 -1.14572299e+00 -2.80200422e-01 4.10752505e-01 -6.23557746e-01 -1.29564559e+00 -8.34058285e-01 3.73802453e-01 -6.32858634e-01 -6.96661353e-01 -4.87410009e-01 -9.60520148e-01 1.77332655e-01 3.83204997e-01 1.58859158e+00 1.99622333e-01 6.93138018e-02 4.75227445e-01 -3.71516407e-01 -6.43850863e-02 -6.82487607e-01 5.95176578e-01 -5.80918901e-02 -2.69448608e-01 4.45739955e-01 -5.86496353e-01 -7.64689982e-01 2.55148709e-01 -4.82566535e-01 1.78348795e-01 5.05832374e-01 1.21018529e+00 -1.06926695e-01 -7.56259084e-01 1.06492281e+00 -9.22661245e-01 1.11215556e+00 -5.82494974e-01 7.26722032e-02 2.98828334e-01 -5.50876915e-01 3.42988312e-01 7.11388230e-01 -1.67048812e-01 -1.22177029e+00 -2.67111212e-01 -5.68347991e-01 7.64125437e-02 8.11256096e-02 4.44736451e-01 1.82196304e-01 1.98235437e-01 6.85529530e-01 3.36309135e-01 4.83401760e-04 -2.95285881e-01 5.76672137e-01 1.14924443e+00 5.19607246e-01 -6.61103725e-01 5.79834506e-02 4.01199535e-02 -5.86757898e-01 -6.24468863e-01 -8.97670567e-01 -4.05248344e-01 -2.14990780e-01 6.90168142e-02 9.03625786e-01 -9.97650206e-01 -1.16048467e+00 -2.45283619e-02 -1.26566005e+00 -5.13288915e-01 7.89871207e-04 1.95148319e-01 -7.62580156e-01 2.07860321e-01 -1.35942745e+00 -7.80960202e-01 -1.00062835e+00 -1.18677223e+00 1.28690255e+00 -1.34666665e-02 -3.22745770e-01 -1.10736847e+00 1.20709330e-01 9.20076251e-01 6.71726286e-01 -5.38056731e-01 9.01313484e-01 -8.67704809e-01 -3.76872241e-01 -2.25882530e-02 -7.77828991e-02 3.26313555e-01 -1.13702565e-03 -4.11133736e-01 -1.22705173e+00 -5.06229699e-01 2.79363617e-02 -1.25200963e+00 1.02074301e+00 -1.50798917e-01 1.15155840e+00 -5.62864602e-01 -2.46709839e-01 3.91942412e-01 8.27376127e-01 1.54806927e-01 2.74236023e-01 -3.85390967e-02 3.65815490e-01 4.10944968e-01 1.57178804e-01 4.14378256e-01 9.33611393e-01 8.93512547e-01 1.23984747e-01 1.37149364e-01 -4.23697270e-02 -2.47306824e-01 6.14147544e-01 1.44965219e+00 1.71420604e-01 -8.03360760e-01 -6.18197978e-01 6.16312742e-01 -1.95827222e+00 -9.41534519e-01 3.87879193e-01 1.69824946e+00 1.17707944e+00 4.75572385e-02 -7.03028450e-03 -2.69226253e-01 2.49950677e-01 2.99898654e-01 -5.59140563e-01 -9.67239916e-01 1.63313061e-01 3.24799597e-01 9.29633379e-02 6.26603782e-01 -1.10617542e+00 8.27838838e-01 6.97226667e+00 4.10440832e-01 -9.85611022e-01 4.34927076e-01 8.22227776e-01 -2.87227362e-01 -6.08585417e-01 -2.84568816e-01 -6.68219090e-01 2.31379434e-01 1.61915851e+00 -1.97588101e-01 6.91580176e-01 9.07923579e-01 -2.49553770e-01 2.86321610e-01 -1.61992407e+00 7.44614303e-01 3.11665267e-01 -1.54785717e+00 -1.05441786e-01 -1.04018688e-01 4.38373268e-01 2.79780507e-01 7.82748498e-03 1.03599596e+00 7.30815291e-01 -1.13455319e+00 2.30560467e-01 2.35757325e-02 7.75097132e-01 -6.84693575e-01 7.76591539e-01 3.20406705e-01 -9.97493207e-01 -2.21190825e-01 -4.16594028e-01 -2.84322232e-01 2.79294938e-01 8.87704492e-02 -1.20534372e+00 3.69700432e-01 3.63805503e-01 6.16921663e-01 -1.46700874e-01 2.28392467e-01 -1.95033759e-01 6.87986434e-01 -3.13429236e-01 -4.09934729e-01 4.65814739e-01 5.55762127e-02 1.09812260e-01 1.58398807e+00 -6.25761300e-02 2.17574254e-01 3.81891608e-01 4.98661637e-01 -7.35242963e-01 -5.02651595e-02 -6.75386310e-01 -5.19352220e-02 6.88302815e-01 1.11896086e+00 1.24894604e-01 -4.36632782e-01 -6.62899315e-01 1.30204260e+00 9.09088552e-01 8.50306377e-02 -7.39435434e-01 -2.41342038e-01 8.24710310e-01 -4.87002015e-01 1.23931289e-01 -1.50919974e-01 -4.88323085e-02 -1.27667642e+00 2.71733832e-02 -1.03956413e+00 4.75065589e-01 -4.07098651e-01 -1.57179391e+00 1.02855861e+00 -1.31762981e-01 -7.76052594e-01 -9.78550732e-01 -4.89781827e-01 -6.85222328e-01 6.93872452e-01 -1.39645505e+00 -1.27644908e+00 -1.57758817e-01 4.65785801e-01 1.13419902e+00 -2.02805564e-01 1.43550253e+00 2.95491338e-01 -5.61123550e-01 9.12468016e-01 -1.72513738e-01 2.72050854e-02 7.96397209e-01 -1.32743299e+00 7.39310324e-01 2.70730644e-01 -1.67135976e-03 5.70739746e-01 3.77672285e-01 6.42263517e-02 -1.86151123e+00 -7.77655303e-01 1.07324123e+00 -6.42287195e-01 5.14279366e-01 -8.66353035e-01 -7.32621729e-01 9.07169342e-01 9.04762447e-01 -3.17257613e-01 8.89783025e-01 6.44004881e-01 -2.78000951e-01 -4.93022837e-02 -8.44671071e-01 5.46817780e-01 1.05180013e+00 -1.16283524e+00 -7.99852490e-01 7.34227180e-01 1.15087926e+00 -4.14685935e-01 -1.00261414e+00 1.68486565e-01 5.55194438e-01 -7.68080831e-01 9.76640344e-01 -6.93526924e-01 6.40345573e-01 7.01711714e-01 -9.64535847e-02 -1.52552021e+00 -4.90856856e-01 -8.88551116e-01 -3.95600259e-01 1.09910870e+00 6.65363073e-01 -4.62171316e-01 7.60108292e-01 6.65283859e-01 -6.11971974e-01 -1.04235184e+00 -1.13061249e+00 -2.79969990e-01 2.11893171e-01 -1.91766039e-01 4.76768345e-01 7.64388263e-01 6.51569307e-01 1.29644263e+00 -5.74623227e-01 -4.36279714e-01 1.32255539e-01 3.66309226e-01 7.59341896e-01 -8.37319732e-01 -5.44213176e-01 -3.85976434e-01 8.80436003e-02 -1.78275633e+00 5.60743093e-01 -8.93455267e-01 3.51607740e-01 -1.66824269e+00 2.49083787e-01 -5.35917640e-01 -1.78680763e-01 5.74810028e-01 -3.81697595e-01 1.87526748e-01 6.19878508e-02 5.70803992e-02 -7.72857606e-01 1.12629533e+00 1.18717825e+00 -4.67536896e-01 7.04152882e-02 -4.93685566e-02 -7.50628531e-01 1.59886032e-01 5.98132193e-01 -1.67022333e-01 -7.57596970e-01 -6.92497134e-01 -7.29825795e-02 5.03502846e-01 -1.34730533e-01 -6.89020872e-01 4.28688049e-01 1.30483564e-02 -9.85348970e-02 -2.03866616e-01 1.11302924e+00 -4.10243720e-01 -5.76144278e-01 1.63994044e-01 -9.56107557e-01 1.53241813e-01 1.31556615e-01 4.92910743e-01 -2.32149854e-01 -6.28052950e-02 4.61448967e-01 -1.95061743e-01 -5.52414775e-01 2.69406259e-01 -4.91893113e-01 2.49677867e-01 6.55730069e-01 3.59967977e-01 -5.40466607e-01 -9.94608879e-01 -2.12274343e-01 5.07904589e-01 -3.13077085e-02 7.43912518e-01 6.48029506e-01 -1.13673389e+00 -9.53025997e-01 8.47473741e-03 2.51962006e-01 -2.95589924e-01 2.08595753e-01 5.17871022e-01 -2.49961197e-01 7.45307863e-01 -3.36981863e-02 -4.77205276e-01 -1.16568470e+00 4.50398415e-01 3.49716723e-01 -5.11060596e-01 -6.44437253e-01 1.13122475e+00 2.87296414e-01 -7.97764122e-01 4.16863173e-01 -2.57538795e-01 4.09928001e-02 -1.10638291e-01 5.95348179e-01 -8.25993940e-02 2.96914071e-01 -2.18847618e-01 -4.66979921e-01 -2.36748517e-01 -3.89686406e-01 -1.75048962e-01 1.47759044e+00 -1.37802839e-01 4.37539108e-02 2.92507470e-01 1.44844294e+00 -4.19572800e-01 -1.14079237e+00 -4.41169262e-01 -2.66753554e-01 -1.16706848e-01 9.97541100e-02 -7.72381306e-01 -6.86429501e-01 9.77896988e-01 1.13168970e-01 2.01940715e-01 9.32141125e-01 4.39884402e-02 1.32379043e+00 1.01576364e+00 4.42364573e-01 -1.00253260e+00 4.17308986e-01 7.50558853e-01 1.14358461e+00 -1.33037472e+00 -1.82300180e-01 -2.13727221e-01 -9.60086644e-01 1.03545702e+00 7.67008603e-01 -4.65207621e-02 2.22665131e-01 3.08253527e-01 -7.87823498e-02 -1.76244840e-01 -1.78488946e+00 3.91436443e-02 1.93206087e-01 3.22252005e-01 9.22657490e-01 5.60435280e-02 -3.54994275e-02 4.31904316e-01 -2.51036018e-01 -2.10380763e-01 3.77862185e-01 8.49166691e-01 -2.46720552e-01 -1.13072276e+00 4.68729407e-01 4.39419746e-01 -3.41892302e-01 -3.12913150e-01 -5.52539289e-01 4.04069632e-01 -6.08660579e-01 1.33321941e+00 2.41833150e-01 -6.90533340e-01 2.94526398e-01 3.54724109e-01 3.56784135e-01 -7.81238675e-01 -1.17854524e+00 -5.14312565e-01 1.04847395e+00 -5.33109486e-01 -3.09074432e-01 -2.94696838e-01 -1.00132859e+00 -7.35448480e-01 -3.80453438e-01 4.57902551e-01 6.04345322e-01 1.00764132e+00 7.99501598e-01 6.07339680e-01 8.32890809e-01 -8.75807166e-01 -1.03329563e+00 -1.36807179e+00 1.93521589e-01 3.55614930e-01 4.16217119e-01 -4.98257726e-01 -1.54256091e-01 -3.54739994e-01]
[12.304620742797852, 7.805327892303467]
09f8570c-6300-4835-9db0-4493f5724fa7
named-entity-recognition-for-social-media
2010.15458
null
https://arxiv.org/abs/2010.15458v1
https://arxiv.org/pdf/2010.15458v1.pdf
Named Entity Recognition for Social Media Texts with Semantic Augmentation
Existing approaches for named entity recognition suffer from data sparsity problems when conducted on short and informal texts, especially user-generated social media content. Semantic augmentation is a potential way to alleviate this problem. Given that rich semantic information is implicitly preserved in pre-trained word embeddings, they are potential ideal resources for semantic augmentation. In this paper, we propose a neural-based approach to NER for social media texts where both local (from running text) and augmented semantics are taken into account. In particular, we obtain the augmented semantic information from a large-scale corpus, and propose an attentive semantic augmentation module and a gate module to encode and aggregate such information, respectively. Extensive experiments are performed on three benchmark datasets collected from English and Chinese social media platforms, where the results demonstrate the superiority of our approach to previous studies across all three datasets.
['Bo Dai', 'Yan Song', 'Xiang Wan', 'Yuanhe Tian', 'Yuyang Nie']
2020-10-29
null
https://aclanthology.org/2020.emnlp-main.107
https://aclanthology.org/2020.emnlp-main.107.pdf
emnlp-2020-11
['chinese-named-entity-recognition']
['natural-language-processing']
[ 2.39992663e-01 2.58945227e-01 -3.71680379e-01 -5.58814526e-01 -4.34317440e-01 -2.79384941e-01 6.66979313e-01 5.25823116e-01 -1.12296748e+00 7.58487761e-01 9.78982270e-01 9.91526097e-02 3.21632922e-01 -9.01074350e-01 -3.49804252e-01 -1.85498714e-01 2.36190110e-01 2.26557910e-01 3.17164436e-02 -5.39433897e-01 1.25636697e-01 1.54905561e-02 -1.15555632e+00 2.70870924e-02 1.19351578e+00 7.25053668e-01 2.32534811e-01 2.00697690e-01 -9.47894573e-01 7.47421980e-01 -5.00668883e-01 -5.36499739e-01 -3.95710394e-02 -1.45804003e-01 -9.81636941e-01 4.08490486e-02 -1.64217204e-01 -2.31384374e-02 -3.56195807e-01 1.24360275e+00 6.10792041e-01 4.96584713e-01 2.11299449e-01 -1.02032125e+00 -9.73298013e-01 1.15788972e+00 -2.36966655e-01 1.00030310e-01 2.35072568e-01 -3.39199781e-01 1.25407314e+00 -1.09046972e+00 6.15037322e-01 1.07659388e+00 6.77814186e-01 7.00648546e-01 -8.04576993e-01 -5.61804891e-01 3.60405207e-01 -2.91812308e-02 -1.37178433e+00 -4.25926149e-01 7.24196136e-01 5.83542790e-03 1.07756662e+00 -4.91726510e-02 3.75862271e-01 1.15566468e+00 -5.82688272e-01 1.02116334e+00 7.90489912e-01 -4.72588122e-01 2.37404183e-01 1.51618481e-01 3.76769394e-01 4.72367525e-01 3.66244018e-01 -5.43048799e-01 -5.43653369e-01 -3.09917986e-01 4.18541968e-01 2.48560041e-01 -9.48611870e-02 7.45122954e-02 -1.12806940e+00 9.01185513e-01 4.16732788e-01 5.77411532e-01 -5.28157949e-01 -6.49171025e-02 6.39684558e-01 2.63674557e-03 1.05922377e+00 6.85163081e-01 -8.24327171e-01 -1.99525520e-01 -5.73594570e-01 -6.60996065e-02 8.30320835e-01 9.29444075e-01 8.54148209e-01 -5.80533855e-02 -1.89195558e-01 1.26899242e+00 2.11448535e-01 3.45558673e-01 1.03392768e+00 -3.00580919e-01 7.59566844e-01 1.01635122e+00 -5.82983391e-03 -9.92578804e-01 -5.34046650e-01 -3.78269553e-01 -8.58403623e-01 -7.89654851e-01 4.80725579e-02 -4.36049610e-01 -9.70180988e-01 1.86709046e+00 4.61993277e-01 6.02590561e-01 5.06978393e-01 6.59942508e-01 1.20221364e+00 6.77923679e-01 7.77275920e-01 5.53469248e-02 1.54126573e+00 -1.09066832e+00 -1.11029053e+00 -5.97350419e-01 1.12396073e+00 -4.62674916e-01 1.28629172e+00 -4.05045778e-01 -7.27677464e-01 -2.82982290e-01 -9.30595934e-01 -2.04753473e-01 -8.59323800e-01 1.51197210e-01 6.43062830e-01 5.09556115e-01 -7.07268596e-01 4.21628773e-01 -8.44288766e-01 -6.06684446e-01 6.51563704e-01 1.51816040e-01 -3.73332471e-01 -3.59412730e-02 -1.73375356e+00 6.12730324e-01 7.96753407e-01 1.26489857e-02 -1.96773082e-01 -6.65788651e-01 -1.28464115e+00 3.14971954e-01 4.30286169e-01 -4.42706436e-01 1.11636388e+00 -8.54317784e-01 -1.23905432e+00 8.44169617e-01 -2.49540105e-01 -3.52956414e-01 1.88540563e-03 -4.11426187e-01 -7.34084070e-01 -5.23422770e-02 2.64173269e-01 6.14944339e-01 3.57757509e-01 -9.87758338e-01 -4.82739985e-01 -3.87745410e-01 2.10344762e-01 4.57206815e-01 -1.41919255e+00 3.95039141e-01 -4.96484071e-01 -1.05124700e+00 4.05026600e-02 -6.12366915e-01 -7.73964286e-01 -4.96039242e-01 -6.31155193e-01 -4.56056863e-01 6.89838827e-01 -8.16054642e-01 1.45769942e+00 -2.00876856e+00 -2.60496765e-01 -4.03084941e-02 2.08629012e-01 5.48055112e-01 -4.30927426e-01 4.70503390e-01 5.26333638e-02 5.34248531e-01 -3.15852076e-01 -6.14159167e-01 7.68460035e-02 2.64237583e-01 -1.06268704e-01 -6.43303543e-02 4.37531143e-01 1.16040313e+00 -1.07797694e+00 -4.75771368e-01 -1.30492583e-01 4.90289778e-01 -4.85544473e-01 2.68342674e-01 -1.69563219e-01 3.54066432e-01 -9.74738955e-01 3.74153137e-01 6.55616820e-01 -4.10614520e-01 2.32350171e-01 -5.38233966e-02 1.48141593e-01 6.84682608e-01 -9.15078461e-01 1.96868861e+00 -5.85688293e-01 1.20215572e-01 -1.90968156e-01 -1.02341568e+00 9.10371482e-01 4.69784886e-01 4.04915899e-01 -6.95433438e-01 3.51486623e-01 1.19711123e-01 -5.27400613e-01 -4.67458367e-01 1.01247346e+00 -2.49533519e-01 -3.79939407e-01 5.65861404e-01 3.34291369e-01 2.04772398e-01 3.33189100e-01 4.43041712e-01 1.11627793e+00 -1.80752724e-01 4.87256944e-01 -2.10131288e-01 6.44678831e-01 -1.85566861e-02 8.00419390e-01 5.21093190e-01 -1.19790919e-01 4.90813851e-01 2.56984144e-01 -1.38048649e-01 -9.99199450e-01 -4.87688243e-01 1.29921347e-01 1.41308808e+00 2.24905282e-01 -6.92031324e-01 -8.37813079e-01 -1.19455516e+00 -2.20111281e-01 6.51663363e-01 -5.34974873e-01 -1.54581383e-01 -6.68589830e-01 -9.84009326e-01 6.83039546e-01 1.00011516e+00 6.12346828e-01 -1.28511465e+00 3.55493098e-01 4.18788582e-01 -4.16665226e-01 -1.68240643e+00 -5.74379444e-01 -8.49870145e-02 -9.06457901e-01 -8.22459817e-01 -4.88597244e-01 -9.66432691e-01 8.30927193e-01 3.82240564e-01 9.04568970e-01 2.41536915e-01 7.58371800e-02 1.60553962e-01 -8.09666753e-01 -5.94253302e-01 -1.68638021e-01 5.19625783e-01 -5.85280024e-02 9.47539210e-02 8.41337502e-01 -3.95501733e-01 -3.26873243e-01 1.52140036e-02 -1.21938765e+00 7.68886879e-02 3.55677456e-01 9.96890724e-01 2.73282051e-01 -1.54383823e-01 1.02861917e+00 -1.33578205e+00 7.72589207e-01 -7.89352238e-01 -1.34521171e-01 1.20846011e-01 -5.62554300e-01 -1.91074749e-03 8.03077221e-01 -2.99441695e-01 -1.46688819e+00 -1.00829989e-01 -4.31782573e-01 1.63511917e-01 -1.74307048e-01 1.00517678e+00 -4.00602192e-01 3.24379236e-01 5.31269908e-01 1.04337372e-01 -2.58917540e-01 -7.55563259e-01 5.21080434e-01 9.69572604e-01 1.83616385e-01 -5.21675646e-01 8.01027834e-01 4.93349880e-01 -6.57623172e-01 -9.37921882e-01 -1.40808690e+00 -8.39682996e-01 -5.11847019e-01 3.74475420e-01 8.80129874e-01 -1.13698745e+00 -8.88910890e-02 4.45796072e-01 -1.14990294e+00 4.31074128e-02 -5.94587207e-01 5.40785909e-01 -4.18888731e-03 4.53799099e-01 -7.79259026e-01 -4.87145573e-01 -5.82018852e-01 -6.66123867e-01 8.50261807e-01 4.42262471e-01 -1.21435225e-01 -1.52152061e+00 1.14182485e-02 5.21025181e-01 4.78173494e-01 -8.00073743e-02 6.55903757e-01 -1.55049789e+00 6.80466071e-02 -4.55962032e-01 -4.92719203e-01 6.06781662e-01 2.86815494e-01 -5.75272322e-01 -9.54180956e-01 -1.52650714e-01 -1.81436762e-01 -4.66677219e-01 8.73768926e-01 -1.81541055e-01 1.23683929e+00 -3.92372191e-01 -2.52658248e-01 3.52902740e-01 1.33182204e+00 -2.85065085e-01 4.20725882e-01 5.09999514e-01 1.04651964e+00 5.15116632e-01 6.24119163e-01 6.04067624e-01 7.30059147e-01 2.50616819e-01 1.96477011e-01 -3.38003516e-01 5.76926544e-02 -5.22678494e-01 2.03613356e-01 1.40096676e+00 1.51262134e-01 -5.15751898e-01 -8.88407946e-01 7.73199737e-01 -1.76176286e+00 -6.71052575e-01 -4.41206060e-02 1.79636240e+00 1.04327643e+00 -6.88200146e-02 -2.18182787e-01 -8.66614729e-02 1.03358114e+00 5.16698658e-01 -4.00507510e-01 2.35084351e-02 -4.36098188e-01 6.25546753e-01 5.12575626e-01 1.04907557e-01 -1.31728768e+00 1.38007593e+00 5.82079220e+00 8.19874823e-01 -8.04032207e-01 5.23049474e-01 3.96473020e-01 2.69277245e-01 -4.49850947e-01 2.48118192e-02 -9.98970926e-01 5.08844197e-01 1.04458952e+00 -4.69963700e-01 -6.68257400e-02 8.23072553e-01 1.32539019e-01 4.29217517e-01 -5.28369248e-01 5.82344890e-01 1.23433314e-01 -1.30618870e+00 1.36601090e-01 -2.35038161e-01 9.49365139e-01 2.95913607e-01 -3.13118607e-01 5.84342301e-01 5.38279653e-01 -7.95451105e-01 3.35302502e-01 1.29910588e-01 6.19461715e-01 -6.66465461e-01 1.15123785e+00 3.06782782e-01 -1.34707594e+00 3.36995833e-02 -2.88205296e-01 -8.19491148e-02 3.80271494e-01 5.76918304e-01 -7.67369568e-01 6.07085824e-01 4.22176629e-01 9.72185016e-01 -5.45991659e-01 6.55175626e-01 -6.37121677e-01 7.91368604e-01 -2.51810789e-01 -3.72311503e-01 4.07004595e-01 2.80508138e-02 3.08864802e-01 1.36995411e+00 1.13684930e-01 2.68891752e-01 2.45199606e-01 6.07149839e-01 -7.11968958e-01 7.49444366e-01 -2.89160281e-01 -7.63442516e-01 5.89045405e-01 1.45140111e+00 -6.60292327e-01 -6.63815856e-01 -9.07663047e-01 9.96030629e-01 5.90448380e-01 4.35785472e-01 -5.26005745e-01 -7.07160354e-01 7.65705645e-01 -1.52092144e-01 4.47958596e-02 -1.87889323e-01 -2.30240822e-01 -1.77775979e+00 1.01282232e-01 -5.49392521e-01 6.74520910e-01 -4.70741004e-01 -1.57702780e+00 4.95472491e-01 -5.20587921e-01 -9.14110065e-01 1.72487140e-01 -4.77666050e-01 -6.07250094e-01 6.33903623e-01 -1.77755010e+00 -1.26912689e+00 -3.01550150e-01 3.96573782e-01 6.35842562e-01 -2.17059344e-01 9.46807146e-01 6.17051899e-01 -6.82750404e-01 6.94841146e-01 8.26430172e-02 6.46183491e-01 6.27671480e-01 -1.16135824e+00 7.44206131e-01 8.39441657e-01 1.63372070e-01 7.69101977e-01 3.00613672e-01 -8.44507337e-01 -1.15435183e+00 -1.41440690e+00 1.34221780e+00 -3.84753048e-01 9.97793674e-01 -3.55465174e-01 -1.17288148e+00 7.56475925e-01 2.07779527e-01 1.24648832e-01 1.16170394e+00 3.60197753e-01 -1.80167779e-01 4.44310606e-01 -9.68236029e-01 5.86854398e-01 1.22325802e+00 -6.54011965e-01 -9.86167014e-01 3.37769777e-01 1.24408758e+00 -2.14237764e-01 -8.28358233e-01 3.75655264e-01 -2.22574174e-03 -2.10620522e-01 8.37705970e-01 -1.05362833e+00 4.57563341e-01 -5.93149439e-02 -4.24126312e-02 -1.46899986e+00 4.18546125e-02 -4.76128757e-01 -3.15094292e-02 1.65078831e+00 5.73729098e-01 -7.11783707e-01 9.80975211e-01 5.23225784e-01 -3.90827715e-01 -4.80530888e-01 -4.87729132e-01 -7.47202992e-01 5.97979128e-02 -4.96940792e-01 8.55675757e-01 1.62816966e+00 2.36878648e-01 5.57491004e-01 -2.09599316e-01 1.06800376e-02 2.96452194e-01 -4.36791569e-01 5.20823359e-01 -1.20686591e+00 2.96946436e-01 -9.93273407e-02 -4.27200586e-01 -1.08389413e+00 7.45572627e-01 -1.19142663e+00 -7.84560325e-05 -1.61497748e+00 4.73059028e-01 -6.05072498e-01 -6.36770070e-01 6.31774724e-01 -7.74327874e-01 2.07099468e-01 -9.35110003e-02 -2.84514464e-02 -8.60537410e-01 1.08442605e+00 9.84062374e-01 4.38058330e-03 -2.00690344e-01 -4.33828354e-01 -1.07703424e+00 8.97277117e-01 7.32455373e-01 -6.63128614e-01 -2.95715928e-01 -7.36315668e-01 4.12073493e-01 -4.34591562e-01 4.93788384e-02 -6.22860670e-01 2.68927127e-01 -2.35875696e-01 -1.21144712e-01 -1.69930145e-01 1.56142324e-01 -5.67600906e-01 -6.73392177e-01 7.15170056e-02 -6.38586581e-01 -1.89981863e-01 5.42123728e-02 6.49807811e-01 -5.92293620e-01 -3.25897932e-01 5.80617607e-01 -7.16366619e-02 -1.00306034e+00 6.37686133e-01 -1.21138275e-01 5.45676470e-01 5.91325700e-01 1.33762047e-01 -2.87627906e-01 -2.42682546e-01 -9.05825734e-01 4.50355977e-01 2.88858265e-01 7.78889239e-01 4.41044539e-01 -1.59426785e+00 -6.80439711e-01 6.41025156e-02 5.61942041e-01 -8.78378078e-02 4.15279925e-01 4.47312891e-01 -1.60067201e-01 2.34500527e-01 1.72793046e-01 8.21869522e-02 -8.77310634e-01 2.62695372e-01 -1.87698439e-01 -4.01330441e-01 -4.62365687e-01 9.23887968e-01 -4.39027846e-02 -9.74840045e-01 -9.75556821e-02 -1.54574290e-01 -6.55388832e-01 1.79556429e-01 6.67524159e-01 1.69479325e-01 1.29377708e-01 -8.59245658e-01 -1.76570460e-01 1.35475382e-01 -1.44055173e-01 5.89145608e-02 1.63281631e+00 -3.95810187e-01 -1.69189930e-01 1.77981913e-01 1.31633580e+00 1.45931587e-01 -6.26904309e-01 -9.86638010e-01 4.79852498e-01 -2.11298808e-01 8.20414573e-02 -3.52435052e-01 -1.06109703e+00 7.48574913e-01 6.48595393e-02 2.25053489e-01 9.31692004e-01 7.83492178e-02 1.40661645e+00 6.41218662e-01 3.05213369e-02 -1.54700434e+00 1.83166385e-01 9.64037657e-01 3.68649989e-01 -1.37730932e+00 -2.82114446e-01 -6.80823743e-01 -7.14392006e-01 8.52167964e-01 8.01303864e-01 -2.77101770e-02 7.18855679e-01 1.22423217e-01 -1.76465162e-03 -1.25552729e-01 -3.91135275e-01 -4.63545740e-01 1.40828297e-01 2.82016814e-01 6.65430546e-01 -6.33053631e-02 -5.92891932e-01 1.25714314e+00 -2.87296996e-02 -9.61559638e-02 3.68017256e-01 1.22842181e+00 -4.66200680e-01 -1.31026590e+00 1.83543697e-01 5.07395983e-01 -8.40208173e-01 -5.54951847e-01 -2.79917121e-01 4.45433766e-01 -1.02617741e-01 9.48585868e-01 -3.24460715e-02 -2.23768055e-01 3.80585283e-01 3.07752252e-01 -2.69559234e-01 -1.17367387e+00 -5.68141878e-01 -3.93695474e-01 4.39920336e-01 -4.78080481e-01 -4.99303192e-01 -4.54771131e-01 -1.74454951e+00 1.36186613e-03 -5.09840012e-01 4.80863065e-01 8.86321306e-01 1.28406489e+00 5.99968791e-01 4.20264661e-01 6.52661979e-01 -4.14811075e-01 -4.65852886e-01 -1.19226885e+00 -5.96238554e-01 9.14497018e-01 -1.66915342e-01 -4.63603884e-01 -4.72680509e-01 -2.83186659e-02]
[9.77266788482666, 9.464421272277832]
c68a62b7-2b49-4e65-9475-eef01684bf65
gshard-scaling-giant-models-with-conditional
2006.16668
null
https://arxiv.org/abs/2006.16668v1
https://arxiv.org/pdf/2006.16668v1.pdf
GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding
Neural network scaling has been critical for improving the model quality in many real-world machine learning applications with vast amounts of training data and compute. Although this trend of scaling is affirmed to be a sure-fire approach for better model quality, there are challenges on the path such as the computation cost, ease of programming, and efficient implementation on parallel devices. GShard is a module composed of a set of lightweight annotation APIs and an extension to the XLA compiler. It provides an elegant way to express a wide range of parallel computation patterns with minimal changes to the existing model code. GShard enabled us to scale up multilingual neural machine translation Transformer model with Sparsely-Gated Mixture-of-Experts beyond 600 billion parameters using automatic sharding. We demonstrate that such a giant model can efficiently be trained on 2048 TPU v3 accelerators in 4 days to achieve far superior quality for translation from 100 languages to English compared to the prior art.
['Maxim Krikun', 'HyoukJoong Lee', 'Noam Shazeer', 'Dehao Chen', 'Yuanzhong Xu', 'Orhan Firat', 'Zhifeng Chen', 'Yanping Huang', 'Dmitry Lepikhin']
2020-06-30
null
https://openreview.net/forum?id=qrwe7XHTmYb
https://openreview.net/pdf?id=qrwe7XHTmYb
iclr-2021-1
['2048']
['playing-games']
[ 1.69259440e-02 -1.96778178e-02 -5.28417528e-01 -6.40313327e-01 -1.09249496e+00 -4.15366411e-01 6.01287901e-01 -2.89397955e-01 -5.88363171e-01 5.68621933e-01 4.02590409e-02 -1.03410971e+00 4.94822115e-01 -5.78613579e-01 -9.79783535e-01 -5.68949044e-01 3.94889146e-01 1.18600214e+00 -3.91587242e-03 -3.76648188e-01 -3.82436514e-01 4.67863709e-01 -1.24040139e+00 4.19726163e-01 6.07487261e-01 9.43004072e-01 2.29965493e-01 4.78473932e-01 -1.70088112e-01 5.99852502e-01 -2.22836629e-01 -6.87161446e-01 2.97170132e-01 -1.35980144e-01 -7.61577606e-01 -7.78906167e-01 5.19519448e-01 -3.09454203e-01 -1.63493857e-01 9.06035423e-01 8.10680330e-01 -3.82728696e-01 3.81491095e-01 -9.79978144e-01 -3.76000762e-01 9.73236561e-01 -3.83728743e-01 2.76310205e-01 -5.45090973e-01 2.14519426e-01 7.10532010e-01 -7.76446640e-01 5.05694389e-01 1.00346398e+00 9.47069168e-01 6.53939784e-01 -1.11857045e+00 -9.66681540e-01 -4.35878247e-01 -3.54856588e-02 -9.68501270e-01 -9.16189849e-01 8.09101388e-02 -2.07386449e-01 1.83451021e+00 2.54419088e-01 6.03690565e-01 1.29799390e+00 4.69026238e-01 4.17561054e-01 9.85778868e-01 -6.50095463e-01 7.68917799e-02 3.75905484e-01 2.44561478e-01 8.84055853e-01 5.28791994e-02 8.65282491e-02 -6.49447083e-01 -2.08897591e-01 4.99240786e-01 -4.23856348e-01 1.88106194e-01 7.44018927e-02 -1.27642679e+00 6.62264049e-01 2.82848537e-01 2.50231624e-01 -2.94750128e-02 4.24652189e-01 9.03953850e-01 2.01558515e-01 5.60387909e-01 3.88327926e-01 -8.37857723e-01 -4.64723796e-01 -1.00801146e+00 5.89142442e-02 7.33479202e-01 1.02335703e+00 4.97718722e-01 4.30216551e-01 2.14672253e-01 8.37474346e-01 4.79712673e-02 7.58347094e-01 6.76442623e-01 -8.69897366e-01 5.32055318e-01 2.58409113e-01 -4.38984782e-01 -2.29643628e-01 -5.21839142e-01 -8.65110815e-01 -8.17729235e-01 1.69998482e-01 2.66752750e-01 -1.72619224e-01 -8.55037034e-01 1.75948584e+00 3.26848447e-01 -3.19324106e-01 1.00909816e-02 7.03396320e-01 5.51062644e-01 8.25733960e-01 1.41790554e-01 2.40391478e-01 1.41142082e+00 -1.32161713e+00 -2.25167572e-01 -5.28381765e-01 1.07488728e+00 -1.01557112e+00 1.32992756e+00 3.53298455e-01 -1.28085566e+00 -5.22832692e-01 -1.24790943e+00 -5.69669485e-01 -2.90947407e-01 3.62801582e-01 1.21314716e+00 9.89753962e-01 -1.33428860e+00 7.22783864e-01 -1.24759054e+00 -4.58822012e-01 3.55607599e-01 8.18831742e-01 -3.69426280e-01 -1.22047096e-01 -9.38893080e-01 1.10084856e+00 4.15095001e-01 -1.55680805e-01 -5.53426623e-01 -1.09819221e+00 -4.91991371e-01 1.35421813e-01 -2.46505484e-01 -1.01247418e+00 1.59834421e+00 -9.26490784e-01 -1.71927702e+00 9.93965864e-01 -1.09175466e-01 -6.98134124e-01 2.12884143e-01 -6.29222766e-02 -4.51468945e-01 -4.49702531e-01 -1.18283980e-01 9.07972395e-01 4.76505756e-01 -3.09777856e-01 -5.59624970e-01 -3.52109671e-01 -5.53519070e-01 1.97277352e-01 -6.78142011e-01 5.97075641e-01 -5.03624141e-01 -3.33525568e-01 -8.99683833e-02 -1.10172522e+00 -1.45431012e-01 -8.65500271e-02 1.34353682e-01 5.74611463e-02 6.70701504e-01 -8.50608766e-01 5.56045234e-01 -1.87239897e+00 -3.40770781e-02 -1.58912838e-01 -1.04709258e-02 3.25608492e-01 -1.26939630e-02 1.39269292e-01 -9.73561630e-02 -2.37643316e-01 7.36311311e-03 -4.65469837e-01 1.75965607e-01 1.33759961e-01 -5.86643159e-01 3.61603886e-01 -1.42739071e-02 1.01516676e+00 -4.32358414e-01 -2.43768498e-01 -1.29258111e-01 8.47347260e-01 -7.86183655e-01 -7.04550073e-02 -2.77802885e-01 1.87034562e-01 -1.93212572e-02 5.92297494e-01 6.07176065e-01 -4.37316746e-01 3.62735152e-01 -1.84985101e-01 -4.41057384e-02 8.41359913e-01 -5.27601480e-01 1.98954391e+00 -6.85889781e-01 5.47411799e-01 2.92407215e-01 -7.80734718e-01 5.91864884e-01 2.64436394e-01 2.57478058e-01 -9.92614925e-01 3.39024544e-01 6.83633804e-01 1.20006204e-01 5.62984543e-03 6.44936621e-01 -3.66783619e-01 -1.00185871e-01 5.65409184e-01 5.12293100e-01 -1.75976351e-01 -1.13631010e-01 -1.98414922e-02 9.66827095e-01 3.60606760e-01 -9.36046839e-02 -6.62289619e-01 -1.45090744e-01 4.49974686e-01 5.96001625e-01 4.33931112e-01 1.83734689e-02 1.45677119e-01 8.45849440e-02 -8.10766995e-01 -1.65639722e+00 -7.62476146e-01 -1.95475593e-01 1.32185161e+00 -7.17216671e-01 -5.90521276e-01 -1.14295828e+00 -2.19286799e-01 -5.54329693e-01 5.47133565e-01 3.08597386e-02 -1.03763968e-01 -8.25147450e-01 -1.47884810e+00 9.64170218e-01 6.82557106e-01 4.47256565e-01 -6.64558828e-01 -5.20320356e-01 1.65691093e-01 7.04577491e-02 -1.27288020e+00 -5.66122472e-01 7.58089006e-01 -1.01230204e+00 -3.06665331e-01 -4.25041109e-01 -9.34828818e-01 3.81708950e-01 -9.09879431e-02 1.52930605e+00 -1.72929764e-01 -2.92051941e-01 -4.06315178e-01 3.05269629e-01 -3.57749969e-01 -8.88187706e-01 5.60110867e-01 4.80079979e-01 -8.11564028e-01 3.75968546e-01 -6.15930855e-01 -3.66605967e-01 1.36325195e-01 -3.39546770e-01 6.72789574e-01 5.86470604e-01 9.35667694e-01 4.94607002e-01 -2.40888596e-01 2.50955582e-01 -9.87402618e-01 2.76698321e-01 -2.32329503e-01 -9.28676188e-01 2.72839278e-01 -1.01700246e+00 2.43728444e-01 9.94433105e-01 -3.16518515e-01 -9.15147841e-01 1.67116806e-01 -6.11791134e-01 -1.09052151e-01 2.67995685e-01 3.20529014e-01 -7.29319081e-02 -1.84225872e-01 7.75538683e-01 1.70378968e-01 9.79038179e-02 -3.96940798e-01 4.09188390e-01 6.51542187e-01 6.25526190e-01 -9.87788200e-01 1.64968491e-01 1.26255140e-01 -3.46819237e-02 -5.51286340e-01 -4.19995397e-01 2.71421280e-02 -3.47341686e-01 2.02635124e-01 7.75733292e-01 -1.35796118e+00 -5.47538340e-01 5.57240665e-01 -1.11350453e+00 -7.22756207e-01 -7.03972727e-02 5.97853899e-01 -3.61980319e-01 -9.41339508e-02 -1.14269507e+00 -1.70404866e-01 -1.08580816e+00 -1.58257496e+00 1.08245385e+00 4.01741304e-02 -1.82602629e-01 -7.65027702e-01 -1.67100448e-02 5.89560926e-01 8.84087324e-01 -4.63976234e-01 1.29146540e+00 -4.90785122e-01 -4.50513840e-01 -1.05272375e-01 -2.97337055e-01 3.55271012e-01 -4.82038110e-01 -1.21019930e-01 -1.15644753e+00 -4.40919489e-01 2.16929421e-01 -5.87917566e-01 5.97956002e-01 2.33357146e-01 8.64400446e-01 -2.12407634e-01 -4.56602901e-01 1.09376323e+00 1.22393286e+00 3.58846635e-02 3.81937116e-01 2.49311477e-02 7.35470057e-01 8.88612866e-02 1.02904672e-02 -2.42195562e-01 2.44794562e-01 9.27753210e-01 -8.29216465e-02 -3.84067863e-01 -2.32723162e-01 -1.43279865e-01 5.07315159e-01 1.75174296e+00 9.77985263e-02 1.08923376e-01 -1.30137587e+00 2.57441471e-03 -1.40769255e+00 -6.60817444e-01 -1.84726879e-01 2.13264871e+00 9.15012062e-01 2.88245797e-01 5.83791621e-02 -2.82507300e-01 3.74531031e-01 -2.83771485e-01 -4.91619557e-01 -7.18501329e-01 -1.07905962e-01 6.71893418e-01 8.19700420e-01 4.61579233e-01 -5.77591360e-01 1.22744572e+00 7.26780367e+00 1.19088280e+00 -1.58087683e+00 8.70662749e-01 1.04925656e+00 -4.28381801e-01 -1.21203892e-01 -3.62550169e-02 -1.34018934e+00 2.88671017e-01 1.79478204e+00 1.42028168e-01 6.25002623e-01 1.11312294e+00 -1.90432727e-01 3.24208260e-01 -1.07054508e+00 9.00519371e-01 1.07127219e-01 -1.77086127e+00 -1.61907062e-01 1.22484885e-01 6.06498063e-01 1.08011734e+00 1.07232705e-02 6.73208237e-01 4.95196790e-01 -1.06467307e+00 8.20963860e-01 1.20821840e-03 1.15950859e+00 -8.28681171e-01 4.23637956e-01 4.65388566e-01 -7.14597523e-01 1.77350536e-01 -6.64537013e-01 4.57764324e-03 -9.19162109e-03 5.71102381e-01 -8.60701323e-01 1.47423238e-01 9.13671196e-01 9.77714583e-02 -4.78170961e-01 3.48589033e-01 3.69022220e-01 8.08946073e-01 -6.51729524e-01 1.41251960e-03 2.25882694e-01 -2.15390250e-01 -1.56075493e-01 1.18952405e+00 5.48083425e-01 -2.80542940e-01 -2.49946669e-01 5.52895367e-01 -3.30170304e-01 9.09660235e-02 -3.06018531e-01 -1.64841097e-02 2.83792794e-01 1.41331232e+00 -8.27299297e-01 -4.46551353e-01 -4.26441282e-01 1.01868379e+00 6.60884798e-01 -2.68035680e-01 -1.15684712e+00 8.40598270e-02 6.28376007e-01 1.41216591e-01 -1.85912475e-01 -3.80936712e-01 -4.36542869e-01 -1.10176361e+00 2.96117063e-03 -1.36563265e+00 1.40744403e-01 -8.08335483e-01 -7.52831757e-01 9.99629200e-01 -4.78928447e-01 -6.81314766e-01 -1.89703882e-01 -1.04611361e+00 -3.46790314e-01 1.08298123e+00 -1.13064766e+00 -1.51950777e+00 1.98925156e-02 2.07500026e-01 3.78801435e-01 -6.29997432e-01 1.30817330e+00 8.42194855e-01 -5.78420937e-01 8.86890948e-01 3.58370304e-01 -1.67694569e-01 6.87902212e-01 -8.17346096e-01 1.13535511e+00 7.32053101e-01 3.57033849e-01 6.43166780e-01 5.64190924e-01 -4.04403925e-01 -1.80128634e+00 -9.97747421e-01 1.02140296e+00 -5.29340923e-01 6.87713325e-01 -8.19890320e-01 -5.81816316e-01 8.91019821e-01 4.63262796e-01 1.55549496e-02 7.78680325e-01 3.74083281e-01 -5.45548439e-01 -2.60210931e-01 -7.32002795e-01 6.25490069e-01 1.02727497e+00 -6.18284583e-01 2.64235842e-03 8.59308064e-01 8.90253723e-01 -7.73761451e-01 -8.86510909e-01 3.64152521e-01 6.53939962e-01 -6.68071091e-01 8.77950370e-01 -5.86659014e-01 4.40065295e-01 4.27769981e-02 -3.15046698e-01 -1.08855498e+00 -1.59157813e-01 -8.25682163e-01 5.18081486e-02 9.43345606e-01 8.80693972e-01 -6.34137630e-01 1.00193000e+00 6.72434509e-01 -4.58232164e-01 -9.04651344e-01 -1.15017855e+00 -7.17632771e-01 5.95552862e-01 -7.11736023e-01 8.01742315e-01 8.52093041e-01 5.65142743e-02 9.25005436e-01 -4.17823106e-01 -1.21116497e-01 3.81124496e-01 -9.29746851e-02 6.33845627e-01 -7.53993809e-01 -9.46727753e-01 -5.93962550e-01 -9.95601937e-02 -8.78750026e-01 1.71516716e-01 -1.58192635e+00 -2.90666193e-01 -1.16231298e+00 4.39325571e-01 -7.98085511e-01 2.48849392e-04 6.02649868e-01 2.50257522e-01 4.34357405e-01 4.77900868e-03 3.02700728e-01 -2.16291681e-01 3.63213807e-01 6.94548845e-01 -1.56043276e-01 2.95565575e-01 -2.97263354e-01 -6.79165065e-01 4.61844355e-01 7.28271067e-01 -6.82198524e-01 -2.90791661e-01 -1.17128074e+00 5.75821221e-01 -5.93944490e-02 4.02062796e-02 -1.28339815e+00 3.41520786e-01 3.85150760e-01 4.32233423e-01 -2.71148801e-01 5.20958483e-01 -6.86643600e-01 5.37990868e-01 5.43837428e-01 -1.80089340e-01 6.72983348e-01 5.58322430e-01 -2.05481246e-01 1.11583136e-01 -7.33083412e-02 9.28984463e-01 -4.27536480e-03 -2.10227847e-01 2.24344358e-01 -1.73649430e-01 -3.05056423e-02 6.03794754e-01 4.02323276e-01 -5.80204427e-01 1.09025031e-01 -2.42480680e-01 -1.44255519e-01 5.29443622e-01 3.19941103e-01 -1.40065715e-01 -1.20422518e+00 -5.63483834e-01 4.27017778e-01 -1.39225364e-01 -2.57235199e-01 8.78631175e-02 7.28729069e-01 -9.23271954e-01 7.63369560e-01 -3.21959406e-01 -6.21393681e-01 -1.16032135e+00 4.20315474e-01 5.58877885e-01 -5.60526907e-01 -8.06347370e-01 9.61515248e-01 -2.22807810e-01 -9.34264362e-01 1.58468653e-02 -3.79250497e-01 6.59578741e-01 -4.07761723e-01 5.87440550e-01 -2.69288626e-02 8.87955427e-01 -4.61182505e-01 -2.70060807e-01 2.74443865e-01 -2.57072181e-01 -3.46120715e-01 1.41623247e+00 4.53612804e-01 -2.37554237e-01 2.92769223e-01 1.20834303e+00 -1.29050046e-01 -9.23545063e-01 -4.79018725e-02 -1.71657920e-01 2.83410251e-01 3.52770567e-01 -1.02659094e+00 -1.07778597e+00 1.07235920e+00 8.42519045e-01 -7.16434181e-01 8.61866295e-01 -1.70734286e-01 1.12084854e+00 6.51939154e-01 7.51794994e-01 -1.12434375e+00 -5.23274541e-01 5.99312961e-01 3.16068947e-01 -1.18223000e+00 -3.21390145e-02 -1.04799598e-01 -4.17062402e-01 9.51957524e-01 5.12686372e-01 2.79104978e-01 5.69921255e-01 1.23468983e+00 1.86533734e-01 -3.00125033e-02 -1.02540624e+00 5.61260462e-01 -1.36895001e-01 3.08212548e-01 7.20531166e-01 2.81240255e-01 -3.83412354e-02 6.21720016e-01 -7.46811867e-01 3.27521972e-02 2.49216110e-02 5.00421703e-01 -2.23620627e-02 -1.40251601e+00 -6.37817904e-02 5.22174954e-01 -6.49457872e-01 -5.49506247e-01 2.40294933e-01 6.22691453e-01 -4.40636724e-02 4.12857324e-01 2.06906855e-01 -3.78619045e-01 1.85672063e-02 4.71288890e-01 8.14469159e-01 -4.66982454e-01 -1.16858089e+00 -3.89556997e-02 3.36534947e-01 -4.34479475e-01 3.55360389e-01 -4.89069015e-01 -1.12205160e+00 -8.26953113e-01 -2.07716763e-01 1.94249470e-02 1.39980710e+00 8.40755701e-01 6.97282493e-01 4.75716144e-01 1.90038886e-02 -8.78159702e-01 -8.24966431e-01 -7.38130629e-01 -1.00732304e-01 -3.18544060e-01 -2.49996185e-01 -9.79776382e-02 1.42428651e-01 -1.40737325e-01]
[8.664346694946289, 3.5126850605010986]
10238f4f-5cf3-4e4f-93d6-128c24a13286
uncertainty-aware-distillation-for-semi
2301.09964
null
https://arxiv.org/abs/2301.09964v1
https://arxiv.org/pdf/2301.09964v1.pdf
Uncertainty-Aware Distillation for Semi-Supervised Few-Shot Class-Incremental Learning
Given a model well-trained with a large-scale base dataset, Few-Shot Class-Incremental Learning (FSCIL) aims at incrementally learning novel classes from a few labeled samples by avoiding overfitting, without catastrophically forgetting all encountered classes previously. Currently, semi-supervised learning technique that harnesses freely-available unlabeled data to compensate for limited labeled data can boost the performance in numerous vision tasks, which heuristically can be applied to tackle issues in FSCIL, i.e., the Semi-supervised FSCIL (Semi-FSCIL). So far, very limited work focuses on the Semi-FSCIL task, leaving the adaptability issue of semi-supervised learning to the FSCIL task unresolved. In this paper, we focus on this adaptability issue and present a simple yet efficient Semi-FSCIL framework named Uncertainty-aware Distillation with Class-Equilibrium (UaD-CE), encompassing two modules UaD and CE. Specifically, when incorporating unlabeled data into each incremental session, we introduce the CE module that employs a class-balanced self-training to avoid the gradual dominance of easy-to-classified classes on pseudo-label generation. To distill reliable knowledge from the reference model, we further implement the UaD module that combines uncertainty-guided knowledge refinement with adaptive distillation. Comprehensive experiments on three benchmark datasets demonstrate that our method can boost the adaptability of unlabeled data with the semi-supervised learning technique in FSCIL tasks.
['Li Liu', 'Haoyu Chen', 'Wanxia Deng', 'Yawen Cui']
2023-01-24
null
null
null
null
['class-incremental-learning', 'few-shot-class-incremental-learning']
['computer-vision', 'methodology']
[ 4.16115493e-01 4.65440214e-01 -3.84941101e-01 -3.68218541e-01 -7.13628292e-01 -4.15375680e-01 6.96013331e-01 5.10933772e-02 -4.26136345e-01 1.04929769e+00 -2.51523107e-01 -1.35826478e-02 -7.42759481e-02 -6.66114807e-01 -6.69748545e-01 -7.74146736e-01 3.21632922e-01 6.98179245e-01 5.76569855e-01 5.95951788e-02 -1.30323190e-02 2.81387568e-01 -2.10297227e+00 1.20528117e-01 1.24767351e+00 9.26533699e-01 2.67153502e-01 2.79321611e-01 -3.14770728e-01 8.68732512e-01 -3.05912197e-01 -2.75294602e-01 2.32837796e-01 -4.73139137e-01 -7.65402794e-01 2.39115670e-01 2.91518211e-01 -1.54361516e-01 1.73036791e-02 9.73067284e-01 4.46080416e-01 3.64086181e-01 7.66255021e-01 -1.37124538e+00 -4.69554842e-01 6.49663866e-01 -4.44080323e-01 -8.12767372e-02 -6.64470121e-02 1.43837661e-01 5.24098814e-01 -1.30614173e+00 9.09907877e-01 1.09281480e+00 7.27311671e-01 9.51976895e-01 -1.21990085e+00 -6.13001287e-01 3.57156038e-01 4.55304891e-01 -1.42217255e+00 -4.06818330e-01 8.94766629e-01 -3.30404162e-01 5.83844125e-01 6.25362396e-02 5.64192057e-01 1.02461922e+00 -1.33563623e-01 1.14346564e+00 1.31669152e+00 -6.57363415e-01 8.83766830e-01 5.68864167e-01 4.34917092e-01 6.14978731e-01 3.01740378e-01 3.33546311e-01 -5.07299244e-01 -9.85705703e-02 1.62253484e-01 1.32455021e-01 -1.29884005e-01 -7.23503709e-01 -8.55658412e-01 6.21397138e-01 2.94295400e-01 1.51097119e-01 -1.98173955e-01 -3.97712916e-01 3.68995994e-01 2.60085464e-01 5.47720253e-01 3.47568691e-01 -6.95980489e-01 1.20201565e-01 -1.15710735e+00 8.47269818e-02 5.20686805e-01 1.18642855e+00 1.12952423e+00 9.13552865e-02 -4.61145997e-01 1.03473222e+00 5.27138216e-03 4.16049898e-01 8.55672181e-01 -8.65483761e-01 -1.40552938e-01 7.54421353e-01 -5.43801039e-02 -9.33888033e-02 -2.28710562e-01 -5.51989138e-01 -7.08283484e-01 4.56994236e-01 6.88001439e-02 -1.49334878e-01 -1.44989479e+00 1.69916475e+00 7.01148093e-01 4.57245171e-01 1.88233837e-01 4.25629079e-01 8.23894739e-01 4.49296951e-01 2.13726833e-01 -7.46681929e-01 7.23617017e-01 -1.25539088e+00 -4.57609981e-01 -2.42539704e-01 5.02884865e-01 -1.79599240e-01 9.92633283e-01 3.14222574e-01 -5.61052680e-01 -7.89232135e-01 -1.17799687e+00 3.36626321e-01 -6.07488453e-01 -1.94933519e-01 3.48741055e-01 5.68360806e-01 -6.91920042e-01 7.68191278e-01 -7.25602031e-01 -2.37190977e-01 7.25569189e-01 1.01067245e-01 -2.51356006e-01 -3.44009459e-01 -1.22368252e+00 7.67638981e-01 1.03349817e+00 -2.11713359e-01 -1.17900944e+00 -9.23610389e-01 -8.17573547e-01 -1.68465242e-01 9.29673791e-01 -3.66604656e-01 1.27151000e+00 -9.81370807e-01 -1.51375854e+00 6.48322761e-01 -8.96605179e-02 -5.35154343e-01 5.30159950e-01 5.24352565e-02 -3.14277977e-01 -3.52110639e-02 1.56380400e-01 1.01866543e+00 1.13066959e+00 -1.63932800e+00 -7.66640723e-01 -2.57940710e-01 -1.45935819e-01 3.05344880e-01 -2.30290934e-01 -6.87307715e-01 -3.59528154e-01 -5.56593955e-01 1.42175153e-01 -1.00447810e+00 -3.01927388e-01 1.28179099e-02 -2.21548006e-01 -3.96268964e-01 1.07315612e+00 -3.71438116e-02 1.20169365e+00 -1.88412142e+00 -3.38674411e-02 -9.60198939e-02 9.26978737e-02 8.46883416e-01 -1.78760082e-01 2.60068662e-02 -4.33858633e-02 -1.28195554e-01 -5.44741213e-01 -4.46280211e-01 -3.23240310e-01 5.52482605e-01 -2.45928451e-01 2.46844813e-02 4.43147272e-01 1.12779129e+00 -1.45738339e+00 -6.51430845e-01 4.27411079e-01 1.36144131e-01 -3.95121247e-01 1.71565771e-01 -5.85674524e-01 4.15019244e-01 -1.59888208e-01 1.01729858e+00 7.77274668e-01 -1.22826003e-01 -1.06032319e-01 -1.75497252e-02 4.97245863e-02 -6.30642951e-01 -1.18208945e+00 1.77388144e+00 -4.03993756e-01 1.02434866e-01 -5.57977736e-01 -8.96864057e-01 8.54717195e-01 8.13525468e-02 2.68520504e-01 -4.32392120e-01 4.25184034e-02 2.36533180e-01 -2.24671766e-01 -2.82770753e-01 6.60503507e-02 -3.98440331e-01 1.84110075e-01 3.42589110e-01 7.21403182e-01 -1.44632012e-01 3.61157507e-01 3.01594645e-01 8.60998392e-01 5.87950945e-01 6.24606311e-01 -6.16177283e-02 6.06138945e-01 1.23060696e-01 8.28720570e-01 1.10756481e+00 -6.36272490e-01 6.54002964e-01 -5.33967800e-02 -4.02828693e-01 -6.64704263e-01 -1.30706561e+00 -2.94151366e-01 1.11757612e+00 2.04091489e-01 -2.88167208e-01 -7.08777189e-01 -1.40788078e+00 -6.02015182e-02 1.15742707e+00 -7.56013691e-01 -6.31538570e-01 6.45889714e-02 -7.57284224e-01 8.02454650e-02 3.99363607e-01 6.33535266e-01 -1.21642399e+00 -5.97670615e-01 3.79678816e-01 1.06354967e-01 -7.54674733e-01 -1.80321604e-01 5.61975181e-01 -9.49761629e-01 -1.15714526e+00 -7.52789557e-01 -7.05119848e-01 8.06607068e-01 3.64247799e-01 9.21341598e-01 -3.73410523e-01 -4.98361111e-01 5.12148499e-01 -7.03721941e-01 -5.93074858e-01 -4.19254273e-01 1.83222108e-02 2.30657548e-01 1.44849762e-01 5.82358003e-01 -4.44908023e-01 -3.57755929e-01 3.30823988e-01 -9.45408463e-01 2.67400086e-01 5.15284359e-01 1.17252588e+00 1.02592659e+00 9.40809324e-02 1.00409520e+00 -1.35024655e+00 1.50471762e-01 -5.31922281e-01 -3.11635792e-01 6.58570826e-01 -1.20238233e+00 3.06727707e-01 7.20450580e-01 -8.29881966e-01 -1.50400400e+00 4.10708487e-01 1.66391611e-01 -8.63627911e-01 -1.94174170e-01 3.87588084e-01 -2.31958747e-01 -1.08336143e-01 9.64578688e-01 4.18877721e-01 -1.71351701e-01 -3.27552229e-01 7.92109311e-01 7.63041377e-01 6.47787273e-01 -5.16679704e-01 7.99527824e-01 4.83057499e-01 -2.05239519e-01 -6.12463236e-01 -1.40958619e+00 -5.03609478e-01 -9.02072191e-01 -5.62828422e-01 3.86139482e-01 -9.09827888e-01 6.54863343e-02 7.04786718e-01 -5.67804337e-01 -3.82570535e-01 -1.09957957e+00 2.03081235e-01 -5.93707979e-01 3.26122940e-01 -6.67138696e-02 -9.41888690e-01 -3.84035170e-01 -8.82030308e-01 8.31361413e-01 6.15788579e-01 2.32692230e-02 -8.67614329e-01 3.99761379e-01 2.46619493e-01 3.22006792e-01 1.33845344e-01 6.34042382e-01 -9.50665474e-01 -3.06670636e-01 -2.43846685e-01 -8.40421766e-02 6.56341851e-01 1.48904890e-01 -3.39004427e-01 -1.46731973e+00 -4.33658957e-01 -1.00862077e-02 -1.06610990e+00 1.19878709e+00 1.32955730e-01 1.11005139e+00 1.06197540e-02 -3.40619683e-01 4.01947826e-01 1.39848983e+00 2.89781719e-01 3.84271353e-01 2.09752545e-01 5.72807610e-01 4.64310348e-01 1.05878639e+00 5.14921784e-01 2.41809353e-01 2.38867030e-01 3.00150543e-01 2.22316146e-01 -5.73733985e-01 -4.49878544e-01 8.15647244e-02 7.59392560e-01 6.47997716e-03 3.72918658e-02 -6.85287237e-01 5.24835646e-01 -2.08827186e+00 -8.69957745e-01 4.32857513e-01 2.37357211e+00 1.06595337e+00 3.31944317e-01 -2.96551615e-01 1.39362901e-01 8.13774586e-01 -6.93066269e-02 -1.23727250e+00 8.14809799e-02 -8.56270641e-02 2.57568568e-01 2.19763875e-01 5.16660094e-01 -1.23030663e+00 1.15509391e+00 5.39143515e+00 1.31396019e+00 -8.91063333e-01 3.87438029e-01 7.18849003e-01 -1.51925802e-01 -2.62679726e-01 1.37716174e-01 -9.74996746e-01 3.31270814e-01 7.79463768e-01 -3.77757519e-01 3.54378939e-01 1.21744144e+00 -2.08576277e-01 -3.26566547e-01 -1.03908300e+00 8.99880469e-01 2.95667171e-01 -1.18569255e+00 1.71420991e-01 -4.74666268e-01 1.22606158e+00 1.16044590e-02 -7.42114633e-02 9.56972659e-01 3.58687818e-01 -3.96329582e-01 6.08077288e-01 5.67992330e-01 1.20579362e+00 -6.69523478e-01 6.11932755e-01 7.40016520e-01 -1.02102363e+00 -2.55621225e-01 -4.96794432e-01 2.20278621e-01 2.55226791e-02 7.13855803e-01 -7.94995785e-01 5.98903894e-01 6.01383090e-01 7.43636608e-01 -7.83350587e-01 1.06720173e+00 -2.85991281e-01 6.10483706e-01 -2.55280137e-01 3.23005438e-01 1.00985445e-01 5.61306207e-03 4.29574698e-01 8.70348454e-01 1.29868940e-01 1.68095961e-01 2.52514273e-01 6.84460521e-01 6.41649738e-02 -7.28943720e-02 -5.15438557e-01 1.88697249e-01 5.89925051e-01 1.16297328e+00 -8.45173359e-01 -6.68118536e-01 -2.06486806e-01 1.06681454e+00 5.03184319e-01 2.80867368e-01 -6.14258707e-01 -3.36414695e-01 -2.93992856e-03 -1.07699491e-01 2.89600044e-01 3.89311582e-01 -4.71136384e-02 -1.31459820e+00 -2.10319325e-01 -5.39985657e-01 5.63051581e-01 -8.21888208e-01 -1.45324218e+00 7.05329835e-01 2.00856194e-01 -1.27237189e+00 -3.66159081e-01 -1.68110833e-01 -5.79134643e-01 4.06736314e-01 -1.68116140e+00 -1.24231827e+00 -4.41359520e-01 6.18723333e-01 1.03035116e+00 -3.29576373e-01 7.58227229e-01 -5.97932190e-02 -5.82452834e-01 6.15358651e-01 2.88692713e-01 -5.15344381e-01 9.07364428e-01 -1.26069164e+00 4.28535230e-02 8.37734878e-01 1.19361073e-01 3.50030690e-01 5.58469474e-01 -9.54322040e-01 -7.61198163e-01 -1.69847786e+00 6.93369448e-01 -3.70998710e-01 3.81348580e-01 -2.80281186e-01 -1.12607932e+00 4.82102841e-01 -2.70653516e-01 4.91476625e-01 5.95672607e-01 -2.28438646e-01 -3.55044156e-01 -1.50394067e-01 -1.40825951e+00 5.06669939e-01 1.17338681e+00 -3.20437104e-01 -7.98330903e-01 3.19880635e-01 8.90687525e-01 -1.25231087e-01 -4.51726526e-01 7.14444578e-01 2.99186081e-01 -8.64174724e-01 8.54127586e-01 -5.05244672e-01 9.34702381e-02 -4.58490342e-01 3.49334478e-02 -1.47214949e+00 -4.05954510e-01 -3.50539714e-01 -6.53443456e-01 1.42255664e+00 2.45509431e-01 -4.18708175e-01 9.67104554e-01 5.22070229e-01 -1.05476134e-01 -8.06256115e-01 -9.77307618e-01 -1.08655965e+00 -1.50088165e-02 -3.87608945e-01 2.80815154e-01 8.95175219e-01 -1.33844569e-01 2.67088056e-01 -4.45378780e-01 -1.77891612e-01 1.04446483e+00 1.02157205e-01 5.14517009e-01 -1.46007097e+00 -2.85311610e-01 -6.07479140e-02 -2.29668349e-01 -5.40703952e-01 1.20386407e-01 -1.00246477e+00 3.69569242e-01 -1.24149895e+00 4.68786269e-01 -5.44757485e-01 -6.01216912e-01 8.12915802e-01 -5.82432985e-01 1.69289157e-01 2.74203449e-01 3.34248573e-01 -1.07811332e+00 9.38425601e-01 1.10481989e+00 -2.61148989e-01 -4.92414474e-01 8.64092261e-02 -5.65889895e-01 6.96493208e-01 6.69339418e-01 -6.63690269e-01 -9.86039698e-01 2.60301828e-01 -3.09712440e-01 -4.02509272e-01 -1.40207699e-02 -1.27185988e+00 3.36410522e-01 -2.46758029e-01 3.04863930e-01 -6.01054668e-01 4.82016206e-02 -7.83717394e-01 9.48824137e-02 4.52885985e-01 -3.62755060e-01 -9.03883338e-01 1.42655864e-01 9.66055214e-01 -6.14146627e-02 -6.01971745e-01 9.99383807e-01 -2.69371152e-01 -1.23781037e+00 4.93578613e-01 -2.61445969e-01 2.07220539e-01 1.52635336e+00 -2.18006045e-01 -1.50643200e-01 1.17461957e-01 -1.04396367e+00 3.56759131e-01 4.39462692e-01 4.54630822e-01 6.56727850e-01 -1.36883402e+00 -4.65692997e-01 3.85740221e-01 6.72894776e-01 3.51632655e-01 5.01779675e-01 3.38223308e-01 6.95197901e-04 1.24706760e-01 -1.59435987e-01 -6.10878289e-01 -9.39680219e-01 1.09439099e+00 1.99301675e-01 -3.32010865e-01 -4.47490722e-01 8.42498958e-01 8.23080391e-02 -7.46138036e-01 4.96057808e-01 4.78168689e-02 -2.55692333e-01 2.20701769e-01 7.18904376e-01 4.91015345e-01 9.65063423e-02 -1.82225943e-01 -1.58576533e-01 2.90711343e-01 -5.25480092e-01 1.67137504e-01 1.11910880e+00 -3.17534804e-01 3.10704470e-01 8.70112062e-01 7.99008369e-01 -6.13441765e-01 -1.81475413e+00 -6.82427227e-01 5.49539104e-02 -1.42955944e-01 1.15343541e-01 -1.24401557e+00 -7.58054852e-01 7.25088477e-01 8.99075627e-01 -4.01627302e-01 1.10669339e+00 -6.17785379e-02 5.27846277e-01 6.07092738e-01 7.56307781e-01 -1.37507391e+00 1.81478843e-01 4.83886003e-01 5.46484709e-01 -1.64067852e+00 -8.91238302e-02 -2.65941411e-01 -9.36040699e-01 9.30194914e-01 1.02395654e+00 2.68436849e-01 9.18297291e-01 -1.29106371e-02 -1.73822569e-03 1.08243510e-01 -8.64141703e-01 -5.22066236e-01 1.52276531e-01 8.69330168e-01 -2.58935124e-01 -6.30314946e-02 -1.13373019e-01 7.08435774e-01 4.78719085e-01 5.35816848e-01 3.16078454e-01 1.20875216e+00 -8.72081161e-01 -1.05849540e+00 -1.31294280e-01 6.12957239e-01 3.71657789e-01 -7.13912724e-03 -2.75360078e-01 5.80426097e-01 6.86108589e-01 7.59645045e-01 -2.09840804e-01 -4.80946451e-01 1.84040502e-01 5.15770972e-01 3.80145222e-01 -8.74954462e-01 -1.25359088e-01 -1.44081950e-01 -3.61606836e-01 -3.40281457e-01 -4.57219690e-01 -5.90432942e-01 -1.09958065e+00 2.82944441e-01 -5.21938205e-01 1.09195195e-01 2.99095452e-01 1.11392415e+00 3.89347434e-01 4.68788773e-01 6.95532858e-01 -9.32838559e-01 -6.59334362e-01 -8.37923586e-01 -6.19540036e-01 3.64007086e-01 1.15236789e-01 -1.00448251e+00 -4.12284434e-01 1.12739868e-01]
[9.899123191833496, 3.2209882736206055]
f0f65a90-19c1-4e3b-b41f-891cd30bb8bf
stylecarigan-caricature-generation-via
2107.04331
null
https://arxiv.org/abs/2107.04331v1
https://arxiv.org/pdf/2107.04331v1.pdf
StyleCariGAN: Caricature Generation via StyleGAN Feature Map Modulation
We present a caricature generation framework based on shape and style manipulation using StyleGAN. Our framework, dubbed StyleCariGAN, automatically creates a realistic and detailed caricature from an input photo with optional controls on shape exaggeration degree and color stylization type. The key component of our method is shape exaggeration blocks that are used for modulating coarse layer feature maps of StyleGAN to produce desirable caricature shape exaggerations. We first build a layer-mixed StyleGAN for photo-to-caricature style conversion by swapping fine layers of the StyleGAN for photos to the corresponding layers of the StyleGAN trained to generate caricatures. Given an input photo, the layer-mixed model produces detailed color stylization for a caricature but without shape exaggerations. We then append shape exaggeration blocks to the coarse layers of the layer-mixed model and train the blocks to create shape exaggerations while preserving the characteristic appearances of the input. Experimental results show that our StyleCariGAN generates realistic and detailed caricatures compared to the current state-of-the-art methods. We demonstrate StyleCariGAN also supports other StyleGAN-based image manipulations, such as facial expression control.
['Seungyong Lee', 'Xin Tong', 'Jiaolong Yang', 'Yucheol Jung', 'Gwangjin Ju', 'Wonjong Jang']
2021-07-09
null
null
null
null
['caricature']
['computer-vision']
[ 5.95193624e-01 4.67036843e-01 4.02613491e-01 -4.21456099e-01 -1.78364128e-01 -8.50684762e-01 6.13898158e-01 -8.96332622e-01 1.60179198e-01 6.22343719e-01 -6.80474862e-02 -6.20387271e-02 6.21405125e-01 -1.02524316e+00 -9.51370060e-01 -6.16660953e-01 5.57034254e-01 1.30629152e-01 -2.72367388e-01 -4.24015909e-01 3.57848480e-02 8.78244579e-01 -1.26699567e+00 5.30946672e-01 7.17954397e-01 9.29231286e-01 -1.52612329e-01 9.65838969e-01 -2.23583132e-01 4.53291923e-01 -7.35892475e-01 -6.60165489e-01 5.56026697e-01 -8.05220544e-01 -3.56330782e-01 4.33578253e-01 7.96292126e-01 -4.84372437e-01 -3.70895788e-02 1.05701780e+00 3.16850901e-01 -3.57274830e-01 7.18614340e-01 -1.34297001e+00 -1.22925401e+00 4.48324531e-01 -1.05208409e+00 -8.29638422e-01 1.26846939e-01 3.46125543e-01 5.09146631e-01 -8.02166879e-01 8.36660326e-01 1.94796586e+00 4.69628334e-01 1.21103501e+00 -1.52928638e+00 -1.09026706e+00 -4.18644175e-02 -5.24140835e-01 -1.23915386e+00 -3.20893645e-01 1.12156260e+00 -1.21380828e-01 1.16479412e-01 4.91015255e-01 8.42736304e-01 9.83324945e-01 2.87456363e-01 8.22400033e-01 1.33370137e+00 -4.07954246e-01 4.18542437e-02 7.47875944e-02 -6.92753613e-01 8.75274420e-01 9.16118249e-02 3.08193713e-01 6.89488128e-02 -3.09580099e-02 1.45238161e+00 -1.52929410e-01 -8.48158225e-02 -2.42221013e-01 -1.08075678e+00 5.86704075e-01 6.96584880e-01 -1.50245264e-01 -2.67911196e-01 6.90403044e-01 -8.58170614e-02 3.49226981e-01 3.11009020e-01 7.19467819e-01 -8.61465111e-02 2.86740363e-01 -7.67358541e-01 5.05334079e-01 4.04505402e-01 1.35365415e+00 9.91431653e-01 6.29539669e-01 -4.60544139e-01 8.34032893e-01 1.84666753e-01 9.35842693e-01 1.18763838e-02 -1.13812482e+00 1.04637153e-01 7.46569455e-01 -4.81714727e-03 -7.23928750e-01 1.38338298e-01 1.19783133e-02 -9.04156983e-01 1.11373568e+00 3.62346508e-02 -6.81580901e-01 -1.25656593e+00 1.61396611e+00 3.78170550e-01 -2.27904394e-01 -4.13471907e-02 8.02770138e-01 8.09754252e-01 7.83815861e-01 8.97554681e-03 4.53946620e-01 1.29346776e+00 -1.03604317e+00 -7.58459806e-01 -2.02317372e-01 3.08925837e-01 -1.08733666e+00 1.45065415e+00 1.89949468e-01 -1.50142348e+00 -7.13571608e-01 -1.10268486e+00 -2.88448066e-01 -2.48252705e-01 5.70351124e-01 3.49287271e-01 7.91185677e-01 -1.19038582e+00 4.95192081e-01 -2.53185302e-01 3.18114161e-02 5.79150617e-01 1.64696008e-01 -2.98496544e-01 2.19638616e-01 -8.54754746e-01 6.10187888e-01 8.78764838e-02 -7.66993687e-02 -7.65422463e-01 -9.49069798e-01 -1.12897766e+00 -1.16744183e-01 -3.20068151e-02 -8.69739294e-01 1.17099857e+00 -1.74824560e+00 -2.11456823e+00 9.95632648e-01 1.73276946e-01 2.55342685e-02 8.32518101e-01 9.63501781e-02 -3.51773471e-01 -1.01519205e-01 -3.45354736e-01 1.39442182e+00 1.44647813e+00 -1.69113171e+00 -5.38816035e-01 2.27813393e-01 1.03832342e-01 1.06775917e-01 1.10565826e-01 -7.99666047e-02 -7.47319937e-01 -1.13100970e+00 -3.29241157e-01 -1.14715946e+00 -1.01593904e-01 6.70765877e-01 -7.28393435e-01 5.52727044e-01 1.50652599e+00 -4.96590078e-01 8.93453538e-01 -2.15533423e+00 1.23163626e-01 3.00276220e-01 8.29581544e-02 3.31308424e-01 -7.46906638e-01 1.89312249e-01 -3.18371207e-01 3.71111304e-01 -3.57888520e-01 -5.77004731e-01 -6.07823431e-02 1.57850921e-01 -4.15015489e-01 1.65812746e-01 6.70874596e-01 1.29408634e+00 -6.70653045e-01 -2.85965860e-01 3.16916645e-01 8.08738947e-01 -7.77477920e-01 3.74220371e-01 -5.49916923e-01 5.65662980e-01 -2.18506202e-01 6.22302592e-01 1.08280957e+00 3.50275457e-01 -1.60713785e-03 -4.47824717e-01 -1.14896856e-02 -3.83552641e-01 -8.58806193e-01 1.37562621e+00 -6.37934268e-01 7.61931241e-01 2.10276052e-01 1.22252092e-01 1.16204000e+00 -5.03157228e-02 -1.45496666e-01 -4.23429638e-01 3.82506132e-01 -2.91356966e-02 -7.59073347e-02 -1.46603897e-01 6.21640205e-01 -3.68464828e-01 -3.43312174e-01 4.59643543e-01 -3.58892143e-01 -1.00912261e+00 -1.21526465e-01 -3.54702584e-02 3.45467389e-01 6.55296385e-01 -1.28392488e-01 -1.14084646e-01 6.67940736e-01 -3.10949057e-01 1.78005561e-01 2.47058704e-01 3.86056602e-01 1.06586814e+00 7.23270118e-01 -5.13596117e-01 -1.73180556e+00 -8.44282866e-01 4.24439371e-01 7.22628057e-01 -1.29584998e-01 -5.82486689e-02 -1.51834190e+00 -7.08164334e-01 2.48676464e-02 7.40413487e-01 -9.92276430e-01 -2.03256786e-01 -7.03173578e-01 -1.31091326e-01 8.02449822e-01 3.99567515e-01 9.10949588e-01 -1.37286901e+00 -2.93745756e-01 -1.33253977e-01 4.37232882e-01 -9.63776767e-01 -1.13394845e+00 -6.89341843e-01 -5.21011055e-01 -8.02565932e-01 -8.98968518e-01 -7.97352076e-01 1.20295131e+00 -5.38359582e-02 9.11219299e-01 2.52903372e-01 -2.72219598e-01 9.64034274e-02 -9.81795117e-02 -6.47553980e-01 -9.82820094e-01 -1.64400890e-01 -3.40613604e-01 4.38745439e-01 -3.67053479e-01 -5.08120954e-01 -7.88254857e-01 2.62080491e-01 -1.41131985e+00 7.06821322e-01 7.27959573e-01 5.78749955e-01 6.38806403e-01 -2.72751957e-01 1.11021146e-01 -1.36451888e+00 6.45927489e-01 3.27745646e-01 -8.65298271e-01 2.00664908e-01 -2.46704906e-01 3.75715971e-01 8.10727715e-01 -7.21847475e-01 -1.26278496e+00 1.21856287e-01 -1.13572530e-01 -7.39211202e-01 -6.75916374e-02 -2.53226697e-01 -5.04447460e-01 -4.01208818e-01 4.78837162e-01 6.46870509e-02 3.80107373e-01 -3.15418363e-01 9.02162611e-01 4.88769084e-01 6.74380541e-01 -5.44013143e-01 1.40473580e+00 7.08508670e-01 3.70686012e-03 -7.07918644e-01 -3.38486642e-01 6.72022462e-01 -4.64536697e-01 -1.26484722e-01 9.27933633e-01 -6.70586646e-01 -5.75794578e-01 9.10172939e-01 -1.21522069e+00 -7.08672822e-01 -5.65304637e-01 -4.03773904e-01 -6.38431549e-01 -3.20722465e-03 -3.93228084e-01 -5.50011098e-01 -6.93324625e-01 -1.06673789e+00 1.43245173e+00 4.21152681e-01 -2.43495390e-01 -7.27197587e-01 -1.06631167e-01 -6.97146505e-02 6.21243119e-01 9.19112563e-01 1.06425858e+00 5.56127548e-01 -4.93245840e-01 -1.28870517e-01 -3.65998805e-01 4.42370653e-01 4.91873592e-01 5.39520264e-01 -1.02293873e+00 -1.33958593e-01 -6.26600206e-01 -6.28141388e-02 5.30439973e-01 2.21282110e-01 1.30703795e+00 -5.87502718e-01 -5.37430756e-02 1.09653056e+00 1.36155915e+00 3.43689531e-01 1.12863100e+00 4.87751851e-04 1.09205222e+00 3.86783868e-01 2.48392627e-01 1.56000465e-01 -2.22539133e-03 6.49689913e-01 3.63060772e-01 -9.21924293e-01 -6.81580245e-01 -8.06926370e-01 4.94999528e-01 5.28726839e-02 4.02182676e-02 -8.67107287e-02 -6.85502216e-02 2.15524554e-01 -1.22538590e+00 -8.47291410e-01 8.29835907e-02 1.67130530e+00 1.10699368e+00 -1.77037448e-01 1.05891481e-01 -9.22207534e-02 8.14722896e-01 1.93921462e-01 -6.32054687e-01 -1.07043731e+00 -1.93452418e-01 3.86640817e-01 5.80174088e-01 6.61108136e-01 -7.73795247e-01 1.46408665e+00 6.02472782e+00 7.33698070e-01 -1.55181253e+00 -3.14076275e-01 8.99455309e-01 -1.05807766e-01 -8.88613224e-01 -1.44219473e-01 -5.23113072e-01 3.56053501e-01 1.17431581e-01 7.17616826e-02 6.84149563e-01 8.25516284e-01 3.98262858e-01 2.59469181e-01 -9.37965870e-01 9.88190711e-01 8.20523594e-03 -1.56065965e+00 6.67545974e-01 -1.78551767e-03 1.30230069e+00 -7.28399158e-01 5.28960824e-01 5.83600737e-02 6.55295372e-01 -1.14599633e+00 1.07973206e+00 5.43384492e-01 1.70981514e+00 -9.63076770e-01 1.73874736e-01 -3.03457856e-01 -9.21665132e-01 2.96224564e-01 -1.70748770e-01 2.17905357e-01 1.83256254e-01 2.42286563e-01 -6.86385810e-01 3.31736505e-02 3.37484062e-01 3.23834777e-01 -4.32133138e-01 5.13590455e-01 -7.18483806e-01 1.65219396e-01 9.55209881e-03 1.40707374e-01 2.85279483e-01 -4.32717264e-01 3.39863390e-01 1.18726909e+00 3.79771262e-01 -4.51708511e-02 -4.09808546e-01 1.46046162e+00 -4.18836623e-01 -2.28596792e-01 -7.59714484e-01 -1.37051612e-01 4.93099958e-01 1.46393251e+00 -1.89521760e-01 -4.85376477e-01 3.04540824e-02 1.47355318e+00 -6.14957558e-03 4.92874056e-01 -9.29310679e-01 -6.48256123e-01 9.96664703e-01 2.62124956e-01 3.41494679e-01 -7.67867416e-02 -6.03779018e-01 -8.60303879e-01 -4.75186259e-01 -1.18310177e+00 -1.74843326e-01 -1.25862014e+00 -9.68508363e-01 8.34813356e-01 -2.06136644e-01 -1.13282025e+00 -1.26916334e-01 -5.53310215e-01 -1.16525507e+00 1.22559679e+00 -1.51162004e+00 -1.86656797e+00 -5.23402691e-01 4.65008587e-01 4.52298522e-01 -1.39429763e-01 7.24305630e-01 -2.21856415e-01 -2.95966536e-01 9.42849100e-01 -3.44237477e-01 2.43417382e-01 7.58956730e-01 -1.19589806e+00 1.06574082e+00 7.34292269e-01 -3.26323092e-01 4.03086632e-01 7.52244413e-01 -7.45095909e-01 -1.47618365e+00 -1.59393859e+00 3.23287100e-01 -4.00116920e-01 8.07745978e-02 -6.19885027e-01 -5.02597749e-01 7.23634183e-01 5.18410742e-01 -8.49306583e-02 2.42838025e-01 -8.77466619e-01 -3.52226913e-01 -1.42586067e-01 -1.53087354e+00 1.13810980e+00 9.35636461e-01 -3.33611131e-01 -1.53620347e-01 -2.55751491e-01 6.27652287e-01 -6.60529435e-01 -4.66827422e-01 1.29492819e-01 9.45665419e-01 -7.15005398e-01 8.27800274e-01 -4.77790803e-01 7.47156680e-01 -6.46001458e-01 2.63585657e-01 -1.63774121e+00 -4.00692075e-01 -1.20254779e+00 3.63216639e-01 1.30903137e+00 4.87077415e-01 -2.95120925e-01 7.57062614e-01 6.14757299e-01 -7.44933682e-03 -4.38348621e-01 -1.06922321e-01 -4.78682786e-01 3.48132700e-01 8.99025239e-03 1.32370186e+00 7.55796492e-01 -5.57558656e-01 6.16787039e-02 -6.77430570e-01 -8.79209712e-02 5.90550244e-01 1.48668781e-01 1.31683362e+00 -6.15934551e-01 -1.44932643e-01 -4.77293372e-01 -2.96287928e-02 -5.85714161e-01 -5.75766973e-02 -4.93477345e-01 -2.03092426e-01 -1.19759071e+00 -1.48226649e-01 -4.29744929e-01 5.04703641e-01 6.35586262e-01 -2.25871906e-01 8.39647949e-01 6.72690928e-01 -2.28753135e-01 2.94482291e-01 5.05258322e-01 2.18468881e+00 -2.74345845e-01 -3.26074213e-01 -2.58270532e-01 -1.10536253e+00 7.84190536e-01 7.31954157e-01 -1.25304639e-01 -5.42400420e-01 -5.55358052e-01 6.64988980e-02 -2.01919556e-01 3.38754952e-01 -8.20360959e-01 -5.07672489e-01 -3.56254101e-01 8.11140418e-01 -2.79662400e-01 3.17955554e-01 -7.41395533e-01 6.53946519e-01 5.12530744e-01 -4.98387605e-01 1.36366546e-01 4.61737752e-01 5.34714721e-02 1.90574870e-01 2.42108777e-01 1.25087941e+00 -1.31457567e-01 -4.29856032e-01 4.26203251e-01 -1.05215758e-01 -2.82855719e-01 9.87061024e-01 -2.78564572e-01 -2.11920232e-01 -6.35596991e-01 -4.50118124e-01 -1.51182458e-01 1.12766409e+00 6.19115353e-01 7.95549035e-01 -1.87671936e+00 -8.24500084e-01 8.24483991e-01 5.21379523e-02 4.30578254e-02 1.10846661e-01 4.01750132e-02 -9.32929099e-01 -6.50655925e-02 -6.09826922e-01 -2.17760816e-01 -1.34214544e+00 5.44344783e-01 4.93255556e-01 1.43939435e-01 -4.77053344e-01 7.07075059e-01 9.13979292e-01 -4.34167981e-01 -2.00071871e-01 -7.02495158e-01 9.37342457e-03 -4.60505098e-01 6.15880609e-01 -6.69079646e-02 -3.30092609e-01 -5.01606286e-01 1.16297834e-01 9.60838020e-01 1.21908307e-01 -3.19317758e-01 1.11139464e+00 -5.04119247e-02 -6.17072731e-02 -2.24321887e-01 1.24530005e+00 3.51250321e-01 -1.89779603e+00 1.47198901e-01 -9.19049621e-01 -4.95663464e-01 -2.33820342e-02 -9.99534369e-01 -1.50734472e+00 8.09259176e-01 5.09442031e-01 -3.66855115e-01 1.17929399e+00 -4.35675859e-01 9.00867820e-01 -1.63093060e-01 4.50830050e-02 -9.27768826e-01 8.87074247e-02 1.85758218e-01 1.49246824e+00 -6.76672339e-01 -2.86435843e-01 -4.84941393e-01 -9.97844756e-01 1.08648300e+00 7.57840753e-01 -5.06363332e-01 2.57750094e-01 6.69921994e-01 4.61099803e-01 -1.35876149e-01 -4.54675287e-01 2.40529209e-01 5.79273939e-01 8.34078431e-01 3.22742879e-01 1.68596566e-01 -1.11749895e-01 1.95953906e-01 -5.34218311e-01 -2.86352839e-02 5.61678648e-01 6.73657000e-01 -9.48745981e-02 -1.39026451e+00 -6.99812472e-01 -6.16830252e-02 -1.91712886e-01 -1.02005370e-01 -8.69648099e-01 8.78748059e-01 3.23212177e-01 4.63629365e-01 2.48258755e-01 -5.14257431e-01 4.70399737e-01 -2.71257669e-01 6.59764111e-01 -4.29791123e-01 -7.13682294e-01 1.38272345e-01 4.79839630e-02 -5.50589263e-01 -1.06164580e-02 -1.82423994e-01 -1.12080538e+00 -6.78708613e-01 1.34617388e-01 -3.47128689e-01 6.65153265e-01 5.19155025e-01 5.20866156e-01 5.88587940e-01 8.30020308e-01 -1.25504255e+00 -7.86310211e-02 -7.85211682e-01 -5.15555561e-01 6.58403933e-01 3.48041952e-01 -2.14694813e-01 -2.59738863e-02 4.60868299e-01]
[11.976268768310547, -0.4398401975631714]
79f202da-3e4d-48a6-a0e1-2b441e5a3dd5
dash-semi-supervised-learning-with-dynamic
2109.0065
null
https://arxiv.org/abs/2109.00650v1
https://arxiv.org/pdf/2109.00650v1.pdf
Dash: Semi-Supervised Learning with Dynamic Thresholding
While semi-supervised learning (SSL) has received tremendous attentions in many machine learning tasks due to its successful use of unlabeled data, existing SSL algorithms use either all unlabeled examples or the unlabeled examples with a fixed high-confidence prediction during the training progress. However, it is possible that too many correct/wrong pseudo labeled examples are eliminated/selected. In this work we develop a simple yet powerful framework, whose key idea is to select a subset of training examples from the unlabeled data when performing existing SSL methods so that only the unlabeled examples with pseudo labels related to the labeled data will be used to train models. The selection is performed at each updating iteration by only keeping the examples whose losses are smaller than a given threshold that is dynamically adjusted through the iteration. Our proposed approach, Dash, enjoys its adaptivity in terms of unlabeled data selection and its theoretical guarantee. Specifically, we theoretically establish the convergence rate of Dash from the view of non-convex optimization. Finally, we empirically demonstrate the effectiveness of the proposed method in comparison with state-of-the-art over benchmarks.
['Rong Jin', 'Hao Li', 'Baigui Sun', 'Yu-Feng Li', 'Qi Qian', 'Jinxing Ye', 'Lei Shang', 'Yi Xu']
2021-09-01
null
null
null
null
['semi-supervised-image-classification']
['computer-vision']
[ 2.30678588e-01 9.85750556e-02 -5.30590773e-01 -6.57380939e-01 -8.05898309e-01 -3.88435453e-01 2.53843904e-01 3.69704992e-01 -6.76863194e-01 9.67401028e-01 -2.38734320e-01 -1.03371657e-01 -1.41284734e-01 -5.28544128e-01 -6.16691232e-01 -8.80773902e-01 2.49825493e-01 6.48840606e-01 2.51520157e-01 2.09547117e-01 7.81025887e-02 2.57096469e-01 -1.34623611e+00 -2.43865937e-01 1.17685306e+00 1.04483044e+00 7.25901872e-02 -2.60078143e-02 -5.95243908e-02 6.90287530e-01 -2.17540413e-01 -2.54314721e-01 5.07064342e-01 -4.43885624e-01 -4.93253410e-01 4.54138696e-01 2.56517202e-01 -7.14674518e-02 1.13824308e-01 1.19907999e+00 2.17054725e-01 3.42834264e-01 5.34604430e-01 -1.15949285e+00 -3.08551967e-01 5.25524497e-01 -7.57180870e-01 -3.22951153e-02 -1.88454613e-01 -1.25136554e-01 1.18731833e+00 -1.26669061e+00 4.15539742e-01 8.60245705e-01 4.68674511e-01 6.55905962e-01 -1.15246582e+00 -7.01085985e-01 4.67787743e-01 8.65837038e-02 -1.44523370e+00 -3.25659543e-01 6.92776084e-01 -2.21934840e-01 2.30792746e-01 2.11609334e-01 3.61414969e-01 4.48022902e-01 -3.49577516e-01 1.06400180e+00 1.00068593e+00 -6.44838572e-01 5.10546386e-01 4.78674561e-01 5.46444654e-01 6.98501170e-01 3.41628283e-01 1.14758499e-02 -4.70753998e-01 -2.58158058e-01 3.58056188e-01 1.08153246e-01 -3.06984663e-01 -6.46805823e-01 -1.00974476e+00 8.55695784e-01 2.80001223e-01 -5.75982220e-02 -4.20516551e-01 -2.36237869e-01 3.17639887e-01 2.79112756e-01 6.70113742e-01 1.94297865e-01 -5.44593751e-01 1.39954641e-01 -9.82395828e-01 -1.01461433e-01 6.03385985e-01 8.38362753e-01 9.25212860e-01 -1.55599013e-01 -2.30454773e-01 1.08631277e+00 3.52388024e-01 2.48239502e-01 4.94160593e-01 -6.59252763e-01 5.04988790e-01 8.80011737e-01 3.21039051e-01 -5.54667652e-01 -5.87859452e-02 -6.55169666e-01 -6.75763667e-01 1.67515472e-01 4.22216356e-01 -2.38546252e-01 -7.71785259e-01 1.76178944e+00 6.23297870e-01 3.55656147e-01 6.05311356e-02 1.07498682e+00 1.92109138e-01 5.73290110e-01 9.58849713e-02 -8.36826384e-01 6.80402160e-01 -1.15276408e+00 -4.37211394e-01 -2.57233560e-01 6.40957355e-01 -5.28734088e-01 1.09850562e+00 4.49096411e-01 -8.45491290e-01 -4.06806439e-01 -9.99977291e-01 3.88151139e-01 1.43903375e-01 6.41028643e-01 3.76128554e-01 7.07301676e-01 -4.59521890e-01 5.93961477e-01 -8.58047903e-01 -2.06579998e-01 4.89504039e-01 4.32287127e-01 -1.97352096e-01 -8.74047577e-02 -8.48675787e-01 5.19155860e-01 5.22632420e-01 2.73225576e-01 -8.09483588e-01 -4.53872770e-01 -7.17624426e-01 1.16336256e-01 7.78413773e-01 -1.72429651e-01 1.33473527e+00 -1.23915088e+00 -1.39656579e+00 7.98624396e-01 -2.22674355e-01 -6.40816391e-01 7.88688183e-01 -4.12231445e-01 -1.36741236e-01 6.14187855e-04 1.16811849e-01 3.34538788e-01 8.77118170e-01 -1.24583530e+00 -9.77098823e-01 -3.54207426e-01 -1.02158472e-01 4.48526412e-01 -6.69533849e-01 -2.19834894e-01 -6.58146262e-01 -5.42953551e-01 2.98494369e-01 -1.05421245e+00 -4.54986811e-01 9.30509418e-02 -4.30595160e-01 -3.65724981e-01 7.82834709e-01 -4.35225964e-02 1.26474035e+00 -2.21073294e+00 -6.80270493e-02 3.79478544e-01 1.31255582e-01 6.14658952e-01 6.90570176e-02 2.08801299e-01 9.74444896e-02 -4.05709185e-02 -4.23633844e-01 -5.93831539e-01 -1.96116343e-01 5.96698672e-02 -3.12115133e-01 6.83195829e-01 1.28422245e-01 4.55983609e-01 -1.12475514e+00 -6.23465896e-01 3.48148465e-01 -2.30723005e-02 -4.52688158e-01 2.83729613e-01 -2.70307690e-01 4.52394307e-01 -8.19754362e-01 4.16059434e-01 6.43392980e-01 -5.19889593e-01 3.57470423e-01 -2.32453719e-02 1.80948898e-02 1.65585309e-01 -1.50499964e+00 1.10859442e+00 -3.59931558e-01 1.72054753e-01 -1.15296409e-01 -1.23996127e+00 9.21635151e-01 3.25830162e-01 4.92852539e-01 -2.04105407e-01 -3.21871862e-02 3.65835756e-01 -1.82534710e-01 -2.50242531e-01 1.21476308e-01 -2.57584810e-01 2.74621516e-01 5.94502509e-01 -1.85070440e-01 3.28128397e-01 3.66047025e-01 1.42711774e-01 6.51529193e-01 7.82580376e-02 5.20780325e-01 -1.40057862e-01 7.96373606e-01 1.15253568e-01 9.44472671e-01 6.63255930e-01 -2.18094319e-01 4.18435276e-01 2.51818359e-01 -2.91249812e-01 -8.67230952e-01 -9.35381591e-01 -2.21414551e-01 9.78815556e-01 2.43445694e-01 -2.24541605e-01 -6.17197514e-01 -1.18340600e+00 -2.06223521e-02 6.30133152e-01 -5.03772378e-01 -1.57002896e-01 -4.46787983e-01 -7.43576825e-01 -9.27181318e-02 3.02191108e-01 4.60286021e-01 -9.85940695e-01 -4.67173249e-01 2.71179199e-01 1.73566580e-01 -7.50656724e-01 -5.78629792e-01 3.10450315e-01 -1.08054125e+00 -1.24437177e+00 -6.60422087e-01 -8.44355941e-01 1.23089087e+00 5.13875127e-01 7.59977996e-01 8.42807814e-02 1.13390066e-01 1.77019387e-02 -4.71395433e-01 -2.02347025e-01 -4.64999318e-01 1.37592763e-01 1.92076564e-01 4.73449469e-01 4.03611600e-01 -1.67750224e-01 -5.16068816e-01 5.46438158e-01 -7.69342363e-01 1.28539000e-02 5.86604178e-01 1.07951546e+00 1.02463114e+00 1.30358055e-01 9.15774941e-01 -1.35711610e+00 3.18615884e-01 -5.54093242e-01 -7.83442497e-01 5.19736230e-01 -9.09700930e-01 4.35435660e-02 1.05986786e+00 -4.84725326e-01 -9.67704535e-01 3.15972030e-01 1.68412805e-01 -6.25051558e-01 5.65984361e-02 5.93753040e-01 -1.20921090e-01 3.97071205e-02 5.13198853e-01 2.02952787e-01 -9.12547410e-02 -6.14188313e-01 1.42682493e-01 6.80879474e-01 1.90070763e-01 -4.74915057e-01 7.84647822e-01 4.75083977e-01 -1.95457250e-01 -5.40628135e-01 -1.43228185e+00 -6.70951366e-01 -5.36801934e-01 -1.89512819e-01 1.28512219e-01 -7.21137702e-01 -5.64621091e-01 2.32699603e-01 -4.41399544e-01 -9.98785421e-02 -4.45256829e-01 7.38536000e-01 -2.11648554e-01 5.27228951e-01 -3.11984062e-01 -1.20571697e+00 -4.05735314e-01 -9.88376200e-01 6.69839978e-01 2.70320684e-01 -5.82511202e-02 -8.51974010e-01 -4.42926064e-02 2.17594013e-01 -8.02976489e-02 2.36753956e-03 7.94462979e-01 -1.13832939e+00 -3.39405596e-01 -5.25544226e-01 -3.06135453e-02 7.56183743e-01 4.05783117e-01 -1.52405620e-01 -6.76622510e-01 -5.78441501e-01 9.70544070e-02 -7.13259161e-01 8.57610703e-01 2.93981642e-01 1.15550566e+00 -1.78758144e-01 -4.46508706e-01 2.72192091e-01 1.46594095e+00 1.91345617e-01 6.87075630e-02 1.34860843e-01 5.21272540e-01 4.90487009e-01 1.21620488e+00 5.67172945e-01 -3.37674804e-02 4.21655566e-01 4.57687020e-01 4.06062976e-02 4.53880042e-01 -2.88444668e-01 1.36332378e-01 7.36186981e-01 2.20877841e-01 -2.92059153e-01 -6.31909966e-01 5.00061333e-01 -2.11199641e+00 -6.92437232e-01 -7.39130452e-02 2.98760009e+00 9.24847364e-01 3.82686496e-01 1.37336804e-02 3.79190356e-01 9.37595129e-01 -9.55079645e-02 -8.57588530e-01 6.57995567e-02 9.26567167e-02 7.57323131e-02 5.19018710e-01 3.95287603e-01 -1.26143873e+00 7.42659926e-01 5.44590044e+00 9.53034222e-01 -1.00767112e+00 -3.63080613e-02 1.00729370e+00 -2.35079944e-01 -7.29191974e-02 -8.26524943e-02 -9.71963167e-01 4.02100861e-01 5.34001470e-01 -2.44758978e-01 2.68024057e-01 1.14733851e+00 4.77790684e-01 -2.26914257e-01 -1.13269031e+00 7.36709893e-01 -7.32647106e-02 -1.06785095e+00 -1.19822919e-01 -2.71285743e-01 1.00994015e+00 -9.11931321e-02 -2.83654071e-02 3.39262962e-01 2.23226145e-01 -6.19834602e-01 6.69523060e-01 5.32719754e-02 6.95057869e-01 -8.65525723e-01 7.03164756e-01 7.72220135e-01 -1.00745404e+00 -1.85180008e-01 -3.95910144e-01 1.12338901e-01 -1.01948455e-01 8.86841357e-01 -9.07441020e-01 4.21037227e-01 2.46937439e-01 9.07029986e-01 -3.20823848e-01 1.25453281e+00 -3.20250213e-01 1.13557363e+00 -3.89513046e-01 -1.49477854e-01 3.70374918e-01 -4.06539649e-01 3.28315914e-01 8.25680971e-01 1.91031471e-01 5.24716340e-02 6.34231091e-01 5.20704448e-01 -2.25969851e-01 5.78381002e-01 -2.08414212e-01 1.44135147e-01 6.36994243e-01 1.14403343e+00 -6.39875174e-01 -3.89089495e-01 -5.41093111e-01 6.45936191e-01 5.38521290e-01 2.46202558e-01 -6.59545779e-01 -3.94729003e-02 1.46079764e-01 -9.46194679e-03 2.40814760e-01 1.44793570e-01 -2.08958521e-01 -1.06084132e+00 3.12508225e-01 -5.33823967e-01 6.13570809e-01 -2.39041656e-01 -1.48038399e+00 4.61128742e-01 -2.57274061e-01 -1.64728045e+00 -1.13572270e-01 -2.77415544e-01 -5.43039262e-01 7.53601015e-01 -1.51402426e+00 -6.65171087e-01 -1.66978121e-01 3.45891118e-01 7.11565197e-01 -4.97693457e-02 5.98418355e-01 1.94910035e-01 -7.56596923e-01 5.86283267e-01 5.17289042e-01 5.34930602e-02 7.33350515e-01 -1.28619826e+00 -2.89878901e-02 8.52465928e-01 7.27771074e-02 5.07160544e-01 5.72881937e-01 -5.22555113e-01 -9.40414906e-01 -1.19419646e+00 8.83010685e-01 1.54279530e-01 4.38712776e-01 -1.26804397e-01 -1.02530134e+00 5.80256581e-01 -3.44485044e-01 3.43673974e-01 6.21545851e-01 -8.60496089e-02 -8.00213516e-02 -3.38010699e-01 -1.20523417e+00 5.94889939e-01 6.20224953e-01 -6.05134219e-02 -3.82645905e-01 4.92708474e-01 4.24369574e-01 -2.43421957e-01 -4.98168051e-01 6.05716527e-01 3.48552883e-01 -8.32380772e-01 5.90023637e-01 -5.44973671e-01 2.58048117e-01 -4.47960377e-01 7.69472495e-02 -1.29190779e+00 4.39454950e-02 -5.26541948e-01 -1.56403646e-01 1.11921716e+00 6.11032128e-01 -6.26514971e-01 1.17792904e+00 6.71139002e-01 1.14187233e-01 -1.09232414e+00 -8.97852302e-01 -7.84155846e-01 -2.14102075e-01 -2.53798485e-01 1.25344262e-01 8.60805273e-01 4.58005406e-02 3.42295051e-01 -5.51402032e-01 1.93945169e-01 8.11169922e-01 3.42168957e-01 5.86815596e-01 -1.32220614e+00 -4.24355775e-01 3.46901007e-02 -6.01247139e-02 -1.15096247e+00 1.47304013e-01 -7.36649036e-01 2.57400215e-01 -1.25575554e+00 3.97959739e-01 -9.91141498e-01 -7.36708581e-01 5.65689325e-01 -5.62385619e-01 1.29695684e-01 1.08700663e-01 5.01283884e-01 -8.83475482e-01 6.04121447e-01 1.23477805e+00 1.04587898e-01 -4.62976754e-01 5.69931626e-01 -4.44222182e-01 7.39163756e-01 7.44572699e-01 -6.14241540e-01 -6.79403663e-01 -1.20310426e-01 -8.23455751e-02 2.77205929e-02 -6.23004623e-02 -7.28349388e-01 7.64358835e-03 -3.07117850e-01 6.62722513e-02 -5.22727668e-01 8.87907445e-02 -9.51301336e-01 -1.77590102e-01 5.24891615e-01 -7.25351334e-01 -4.93161649e-01 -1.28981382e-01 7.96219647e-01 -2.02769354e-01 -5.47547817e-01 1.10652745e+00 1.37536988e-01 -4.87907618e-01 4.18466777e-01 -2.53742728e-02 1.59190625e-01 1.30517995e+00 -9.53575224e-02 1.20303303e-01 -2.85331726e-01 -7.68851101e-01 4.93183076e-01 5.02230704e-01 1.14975534e-01 5.14918685e-01 -1.16756225e+00 -6.66608572e-01 1.14811316e-01 2.72404224e-01 1.13150828e-01 -6.91993013e-02 8.30972254e-01 -1.48159355e-01 3.36061329e-01 3.37388307e-01 -5.25460124e-01 -1.25137758e+00 7.57979810e-01 1.68129861e-01 -4.25715238e-01 -5.44012070e-01 7.58894086e-01 1.75731882e-01 -3.38118464e-01 4.76559520e-01 -6.48819804e-02 -1.12897769e-01 -2.79827982e-01 4.37350243e-01 3.08680326e-01 5.68665117e-02 -3.53580415e-01 -2.35211372e-01 3.34338337e-01 -3.28209817e-01 8.92009214e-02 1.20389247e+00 -1.42891943e-01 1.19595356e-01 7.04693854e-01 1.17425287e+00 9.38690156e-02 -1.40504587e+00 -6.89619780e-01 1.60194680e-01 -5.59253693e-01 -1.11251501e-02 -7.04265177e-01 -1.18918085e+00 6.04944229e-01 4.64254051e-01 1.26151234e-01 1.25746632e+00 -1.30163833e-01 6.32306755e-01 5.42154253e-01 5.16914845e-01 -1.21873605e+00 -8.89752135e-02 1.15986750e-01 2.92240828e-01 -1.57219613e+00 7.47704506e-02 -6.41683698e-01 -8.18098307e-01 8.82804275e-01 6.62692666e-01 -1.72920406e-01 4.99903142e-01 -8.26062858e-02 -6.01928495e-02 2.52631456e-01 -7.86513925e-01 -9.80778970e-03 1.93499878e-01 9.62400362e-02 3.51468056e-01 -3.51364613e-02 -6.21584058e-01 4.17164922e-01 3.65811586e-01 2.14652836e-01 2.53044337e-01 9.33345735e-01 -6.17745161e-01 -1.17390776e+00 -2.60656536e-01 7.01996267e-01 -5.06697416e-01 6.32535741e-02 -1.42027035e-01 5.73166668e-01 -4.13721092e-02 9.57109869e-01 -2.13683784e-01 -5.07583693e-02 1.84644774e-01 6.32518232e-02 1.31566092e-01 -9.20042753e-01 -4.47676778e-01 2.07725421e-01 -3.03594396e-02 -1.48650184e-01 -5.27351499e-01 -6.87541068e-01 -1.39493501e+00 5.62655926e-02 -7.55278826e-01 6.12548769e-01 3.79612714e-01 1.08516848e+00 1.03723891e-01 1.13105386e-01 1.23287344e+00 -4.16267782e-01 -1.01379299e+00 -8.32651317e-01 -7.36428380e-01 4.14900690e-01 2.18180418e-01 -6.22543633e-01 -5.23552775e-01 1.04063720e-01]
[9.296937942504883, 3.9765195846557617]
5d10c330-484a-4ada-b69e-0a701ad5484f
leveraging-text-data-for-causal-inference
2307.03687
null
https://arxiv.org/abs/2307.03687v1
https://arxiv.org/pdf/2307.03687v1.pdf
Leveraging text data for causal inference using electronic health records
Text is a ubiquitous component of medical data, containing valuable information about patient characteristics and care that are often missing from structured chart data. Despite this richness, it is rarely used in clinical research, owing partly to its complexity. Using a large database of patient records and treatment histories accompanied by extensive notes by attendant physicians and nurses, we show how text data can be used to support causal inference with electronic health data in all stages, from conception and design to analysis and interpretation, with minimal additional effort. We focus on studies using matching for causal inference. We augment a classic matching analysis by incorporating text in three ways: by using text to supplement a multiple imputation procedure, we improve the fidelity of imputed values to handle missing data; by incorporating text in the matching stage, we strengthen the plausibility of the matching procedure; and by conditioning on text, we can estimate easily interpretable text-based heterogeneous treatment effects that may be stronger than those found across categories of structured covariates. Using these techniques, we hope to expand the scope of secondary analysis of clinical data to domains where quantitative data is of poor quality or nonexistent, but where text is available, such as in developing countries.
['Luke Miratrix', 'Leo A. Celi', 'Aaron R. Kaufman', 'Reagan Mozer']
2023-06-09
null
null
null
null
['imputation', 'causal-inference', 'imputation', 'causal-inference', 'imputation']
['computer-vision', 'knowledge-base', 'miscellaneous', 'miscellaneous', 'time-series']
[ 5.64653218e-01 2.12909237e-01 -8.84407938e-01 -5.46452582e-01 -7.29170620e-01 -5.09084642e-01 9.58299264e-02 1.00017035e+00 -4.26909447e-01 1.01447856e+00 1.13121200e+00 -7.79714465e-01 -5.16647518e-01 -8.26232851e-01 -6.66106522e-01 -2.24631041e-01 -4.90247719e-02 4.14610147e-01 -4.26711977e-01 2.90700078e-01 8.50716010e-02 1.25630617e-01 -8.92651141e-01 4.10671502e-01 9.92033243e-01 1.29336879e-01 -1.96265399e-01 1.88133463e-01 6.34680986e-02 9.85707343e-01 5.78242214e-03 -4.49668646e-01 3.69278304e-02 -5.08866489e-01 -5.29154181e-01 -2.37401232e-01 1.44584468e-02 -5.97210705e-01 -1.24229498e-01 4.87352520e-01 6.19331479e-01 -2.28283644e-01 6.52768075e-01 -9.40523148e-01 -4.85634089e-01 8.14827561e-01 -4.72160608e-01 -9.94399935e-02 6.74743533e-01 1.68230683e-01 7.45143473e-01 -4.14487630e-01 8.92748296e-01 1.07689333e+00 1.04449689e+00 2.59007275e-01 -1.67888069e+00 -7.57950366e-01 -1.47828877e-01 -2.59737641e-01 -9.83517051e-01 -7.24778712e-01 2.29290262e-01 -7.66434193e-01 5.55510342e-01 4.46019620e-01 5.21966577e-01 1.03008604e+00 3.59277129e-01 2.17380762e-01 1.00486624e+00 -4.26687628e-01 1.88852400e-01 -1.80948988e-01 9.20230821e-02 4.18094307e-01 5.77052653e-01 4.32289034e-01 -3.48395884e-01 -1.02117383e+00 5.98075747e-01 5.76587796e-01 -3.77077281e-01 -1.73263803e-01 -1.55893528e+00 9.08505321e-01 1.83322594e-01 -1.06402583e-01 -5.59915304e-01 -6.03008345e-02 4.53133762e-01 2.32672423e-01 3.20872664e-01 2.33107522e-01 -5.62220752e-01 -1.37772322e-01 -9.93794203e-01 2.07018122e-01 8.37002635e-01 8.35231245e-01 2.06261918e-01 -5.20910740e-01 -3.58858764e-01 6.74397349e-01 2.87510276e-01 5.41138232e-01 1.76251441e-01 -1.00062978e+00 5.90915620e-01 7.39884198e-01 3.66472453e-01 -5.82128763e-01 -6.56980097e-01 6.68204874e-02 -1.01334357e+00 -9.53551605e-02 6.21386647e-01 -6.67531252e-01 -8.83195221e-01 1.84059966e+00 4.33045268e-01 -1.89985737e-01 -6.88879415e-02 4.95378464e-01 6.40718997e-01 -1.35072276e-01 6.04085803e-01 -5.71259141e-01 1.65220380e+00 -9.87352133e-02 -9.47797179e-01 1.64445117e-02 1.04232121e+00 -7.09288776e-01 6.56629086e-01 1.05656490e-01 -1.23493218e+00 1.94496796e-01 -5.02155781e-01 -2.01620013e-01 -1.20342359e-01 -3.69032055e-01 6.12531006e-01 7.43432939e-01 -5.75544000e-01 6.72481000e-01 -9.68681276e-01 -5.47871888e-01 7.93178320e-01 2.57558078e-01 -5.72003841e-01 -3.42499346e-01 -1.05363262e+00 8.61621976e-01 6.73617376e-03 -5.51654696e-02 -1.98264107e-01 -1.32545769e+00 -1.01793325e+00 1.20275877e-01 5.38099408e-01 -1.41482496e+00 9.51234221e-01 -3.78702790e-01 -6.14282429e-01 4.59893733e-01 -5.08921921e-01 -2.41817441e-02 6.53788030e-01 2.91034043e-01 -2.05326319e-01 -3.54532823e-02 3.58706117e-01 1.89097911e-01 1.20714761e-01 -8.29003453e-01 -3.58603060e-01 -8.64532590e-01 -5.78052938e-01 1.42646626e-01 9.41296816e-02 2.82609642e-01 3.70427556e-02 -9.01111603e-01 -4.66041127e-03 -6.19194925e-01 -7.38086700e-01 1.83543533e-01 -3.36024314e-01 3.38680863e-01 1.24039985e-01 -8.31493914e-01 1.34631145e+00 -2.04876447e+00 -1.17132276e-01 1.27322480e-01 4.34356272e-01 -1.74966112e-01 1.63653359e-01 7.71240175e-01 -1.50424913e-01 4.41101074e-01 -5.11505425e-01 -1.61832690e-01 -3.22099298e-01 1.25441074e-01 -5.18576354e-02 5.45142233e-01 1.09775312e-01 1.29308057e+00 -6.84069693e-01 -6.91983521e-01 3.11159104e-01 2.73001045e-01 -9.20038164e-01 -1.41827270e-01 1.96469471e-01 5.68562567e-01 -4.53326166e-01 6.40172660e-01 4.98946577e-01 -3.53928000e-01 4.76476520e-01 2.03652695e-01 -1.91852495e-01 4.40078110e-01 -1.17718458e+00 1.48133767e+00 -1.58849373e-01 1.91737413e-01 3.83862406e-01 -7.07414329e-01 2.79389322e-01 5.97001195e-01 7.32833505e-01 -2.47438490e-01 1.43513260e-02 -1.27689108e-01 6.79454952e-02 -7.09124446e-01 6.21829778e-02 -7.51764297e-01 9.07339435e-03 6.87321007e-01 -5.26497364e-01 1.87573016e-01 -1.75513953e-01 3.33225168e-02 1.44542825e+00 -3.36790562e-01 6.10669911e-01 -1.85674191e-01 -2.54548728e-01 3.35277885e-01 8.57655108e-01 8.85408163e-01 1.08101875e-01 6.40112042e-01 6.17925823e-01 -1.42757729e-01 -1.03329825e+00 -1.11320400e+00 -9.44519103e-01 4.30282593e-01 -6.32280171e-01 -3.93224180e-01 -6.32826164e-02 -3.30225825e-01 4.17658448e-01 5.64008653e-01 -8.59733224e-01 -2.74581388e-02 -1.71390936e-01 -1.19434500e+00 4.98116642e-01 7.32728660e-01 -1.24509618e-01 -8.27284276e-01 -6.42355621e-01 5.36426723e-01 -2.18944982e-01 -6.26750588e-01 -3.74505162e-01 2.05534860e-01 -1.21666169e+00 -1.38302422e+00 -6.28959715e-01 -3.85705769e-01 7.49457657e-01 -2.21431911e-01 1.01961744e+00 1.92542806e-01 -3.68556082e-01 1.31799608e-01 -2.91011631e-01 -6.78823888e-01 -4.93682086e-01 -3.76045108e-01 -1.40732706e-01 -5.15120208e-01 4.21150714e-01 -4.83718216e-01 -7.12871194e-01 -5.96535206e-02 -1.12469268e+00 1.25811398e-01 5.99585831e-01 1.02829397e+00 1.36527658e-01 -2.37704337e-01 7.41450548e-01 -1.48583508e+00 4.38894808e-01 -8.89075756e-01 -3.20728719e-01 1.50152311e-01 -9.39052403e-01 -2.09050607e-02 2.09675714e-01 -3.77346814e-01 -9.88906145e-01 -1.77629456e-01 9.84313618e-03 1.09853879e-01 -3.56830835e-01 1.04542005e+00 3.44295017e-02 3.30132604e-01 6.43338859e-01 -4.41705048e-01 4.55483854e-01 -4.69403416e-01 8.74440074e-02 8.08955312e-01 1.56173304e-01 -4.26543206e-01 3.25052500e-01 5.19234657e-01 5.46706170e-02 -3.17748189e-01 -3.54630828e-01 -4.21918839e-01 -3.89391094e-01 4.30116981e-01 8.29013824e-01 -1.03855693e+00 -1.18558729e+00 -1.00433923e-01 -5.70076883e-01 -4.54895675e-01 -3.23942333e-01 1.07532036e+00 -3.10025841e-01 9.64098349e-02 -6.44328952e-01 -6.51591122e-01 -2.32140601e-01 -9.54956174e-01 8.34652305e-01 -2.68343270e-01 -6.81925416e-01 -1.33079875e+00 1.70109883e-01 3.71652156e-01 2.97410607e-01 6.73339963e-01 1.47468507e+00 -6.72985196e-01 -1.57538205e-01 -3.77647489e-01 -1.86663792e-01 -4.76396263e-01 6.69313848e-01 -1.26063854e-01 -5.07210255e-01 -1.54481784e-01 -1.26475450e-02 -1.27235934e-01 6.31513357e-01 1.04750526e+00 9.88233745e-01 -5.56666613e-01 -5.97764492e-01 4.28352267e-01 1.24663472e+00 1.52959347e-01 4.29041803e-01 -4.64415236e-04 5.84415555e-01 8.28214645e-01 1.60678595e-01 6.63671732e-01 7.59129524e-01 5.96596122e-01 -1.44113094e-01 -4.93016779e-01 2.26364926e-01 -3.90138894e-01 -2.39587069e-01 2.26720065e-01 1.35877430e-01 -2.74526700e-02 -9.89771187e-01 6.85307086e-01 -2.07787156e+00 -9.35932696e-01 -4.84093636e-01 2.62918997e+00 1.25839579e+00 -1.45375896e-02 1.94059163e-01 -1.55204758e-02 6.43130660e-01 -4.29689616e-01 -4.28719074e-01 -3.43339950e-01 1.39964521e-01 1.55303165e-01 7.72860229e-01 4.11648065e-01 -5.41722298e-01 8.65176320e-02 7.57220221e+00 2.73132771e-02 -5.36045849e-01 -1.25203487e-02 5.71585357e-01 -2.96048284e-01 -7.27901280e-01 3.75517964e-01 -1.37938291e-01 4.47414696e-01 9.53773439e-01 -3.02715063e-01 2.29913220e-01 1.31510973e-01 9.56326783e-01 -3.51976216e-01 -1.43470263e+00 3.93068731e-01 -4.09446865e-01 -1.41327274e+00 -3.83352578e-01 4.44935769e-01 6.43171668e-01 -1.77495778e-01 -2.51544118e-01 -1.37515560e-01 8.25181961e-01 -1.29373705e+00 3.18990380e-01 4.89078701e-01 1.00198913e+00 -3.81791830e-01 8.88756216e-01 2.55857140e-01 -5.32686710e-01 -1.53616637e-01 -7.03322813e-02 -2.94539064e-01 5.11592150e-01 9.65593874e-01 -9.88253117e-01 7.32586503e-01 4.71615434e-01 7.20081568e-01 -2.24571273e-01 1.13346839e+00 1.29341841e-01 8.20166588e-01 -1.80185273e-01 4.72062200e-01 -3.61143023e-01 -1.73176471e-02 1.49548113e-01 9.36474264e-01 2.72530526e-01 5.66790402e-01 -1.48124278e-01 9.02706742e-01 -1.19516045e-01 1.43205196e-01 -9.73175168e-01 -1.99267156e-02 5.84982157e-01 8.54270399e-01 -4.50556874e-01 -2.77817219e-01 -7.91453421e-01 3.05223018e-01 5.47328405e-02 3.47870350e-01 -2.30762601e-01 7.58543387e-02 4.63571280e-01 5.26329637e-01 -2.47077003e-01 1.96223885e-01 -7.56972492e-01 -1.13077700e+00 -1.36519358e-01 -1.05128300e+00 9.28336263e-01 -6.04655981e-01 -1.36613059e+00 -2.70692706e-01 2.09952921e-01 -7.77543068e-01 -4.67567980e-01 -5.51561117e-02 -1.94036230e-01 1.18688655e+00 -9.84363496e-01 -9.20068800e-01 1.25632927e-01 6.12810969e-01 -9.39481035e-02 3.78189802e-01 9.76445138e-01 3.30284506e-01 -6.14287794e-01 4.85084772e-01 2.67784089e-01 2.17244938e-01 1.11757874e+00 -1.02008593e+00 1.41344488e-01 3.42317462e-01 -3.59930426e-01 1.03035975e+00 4.65628058e-01 -1.36390984e+00 -1.34498739e+00 -9.00900304e-01 1.26203465e+00 -7.71012723e-01 5.49551487e-01 -1.73596710e-01 -8.97456765e-01 1.06310034e+00 -1.61463067e-01 -3.19420129e-01 1.14578199e+00 7.30756223e-01 -7.08623603e-02 2.29538292e-01 -1.49550247e+00 7.18418896e-01 8.58040333e-01 -2.58192003e-01 -7.34872937e-01 1.09878711e-01 6.76676035e-01 -3.69562089e-01 -1.45923424e+00 6.69180095e-01 7.51363695e-01 -4.71093982e-01 8.18197906e-01 -1.01623166e+00 6.54671311e-01 -1.10356295e-02 -2.66223978e-02 -9.90797997e-01 -4.66518253e-01 -4.17527020e-01 4.09933507e-01 1.16611993e+00 6.83486462e-01 -6.90939844e-01 6.75904632e-01 1.43973613e+00 2.53289063e-02 -3.58069181e-01 -8.64902854e-01 -3.60581428e-02 2.28714883e-01 -3.56112033e-01 6.33948326e-01 1.29315495e+00 4.66771513e-01 -4.81657833e-02 -2.65871257e-01 9.19060186e-02 6.67446733e-01 1.82457417e-01 4.40655410e-01 -1.31097388e+00 -2.36471921e-01 3.40975225e-02 -1.15061346e-02 -1.29543751e-01 -1.40439153e-01 -7.35300422e-01 -2.87264645e-01 -1.83523750e+00 9.27408099e-01 -7.25526214e-01 -2.42827716e-03 9.68278229e-01 -4.47150230e-01 -8.04802477e-02 -8.52267295e-02 1.78439513e-01 1.19896784e-01 1.37251452e-01 1.16255522e+00 5.32784350e-02 -5.00135243e-01 1.26484945e-01 -1.19159782e+00 5.27408719e-01 4.80276942e-01 -8.64755154e-01 -3.28303635e-01 -3.29847753e-01 2.63406932e-01 8.69661391e-01 5.56201279e-01 -1.80145711e-01 2.00545579e-01 -4.97218549e-01 6.64181054e-01 -2.31738657e-01 -2.16305062e-01 -1.07751524e+00 7.52275825e-01 6.62313700e-01 -7.35977113e-01 1.20965794e-01 4.04290825e-01 5.93712866e-01 7.48901293e-02 -1.45031393e-01 1.66844398e-01 -1.18620761e-01 2.51269877e-01 1.27597839e-01 -4.71042484e-01 2.13213056e-01 6.99296117e-01 -5.35236187e-02 -4.30433214e-01 -3.34277958e-01 -9.33412910e-01 4.74736154e-01 7.14552462e-01 1.25481814e-01 3.33242089e-01 -1.30642724e+00 -1.02980649e+00 1.77640811e-01 1.89061537e-01 -1.63461100e-02 3.75105560e-01 1.22721243e+00 -8.55539888e-02 4.16501075e-01 6.18028417e-02 -2.31186002e-01 -1.16955101e+00 8.20636451e-01 -1.90522105e-01 9.58952121e-03 -9.76798296e-01 -2.24824458e-01 1.42483935e-01 -2.55742043e-01 1.40371367e-01 -3.48070204e-01 1.48138613e-01 1.15842678e-01 5.05642295e-01 2.28026062e-01 1.09299749e-01 -2.15162128e-01 -4.66272026e-01 1.22773200e-01 5.53181581e-02 -3.00243944e-01 1.61306298e+00 -3.80117178e-01 -1.76167324e-01 4.81104970e-01 8.30278933e-01 3.46382558e-01 -9.67860043e-01 2.94027440e-02 -5.89570329e-02 -3.93919915e-01 -7.90725499e-02 -9.64496732e-01 -4.65296805e-01 3.11216950e-01 1.78814858e-01 -6.66768774e-02 1.17908740e+00 -7.15012848e-02 1.86520398e-01 -2.63992339e-01 8.73454511e-02 -4.56039131e-01 -7.32738137e-01 -2.74961680e-01 5.04718304e-01 -1.32571292e+00 2.69508243e-01 -2.48132840e-01 -4.88884419e-01 5.43168843e-01 -1.58997238e-01 5.32598913e-01 6.72198474e-01 5.73697209e-01 -6.15072362e-02 -2.76242137e-01 -1.11360347e+00 1.46428391e-01 1.22011058e-01 6.16219819e-01 7.43214428e-01 1.61536783e-01 -8.63951981e-01 8.14920187e-01 -3.58553417e-02 4.92891699e-01 8.39376211e-01 1.05745864e+00 2.16192797e-01 -1.40723968e+00 -6.10782921e-01 1.25513232e+00 -9.75820124e-01 -4.94962096e-01 -2.39487737e-01 8.28910708e-01 -7.52183124e-02 1.31174445e+00 1.27250418e-01 1.50658637e-01 4.37849164e-01 1.28687963e-01 2.06213623e-01 -6.25772119e-01 -6.13620043e-01 1.43149942e-01 4.48003590e-01 -4.02995884e-01 -5.18245280e-01 -1.14447284e+00 -1.06481838e+00 -6.89407051e-01 -2.89198071e-01 1.81081042e-01 3.80850583e-01 1.07048166e+00 4.85798180e-01 5.48849821e-01 3.31030250e-01 -1.76110744e-01 -3.78640473e-01 -6.54789507e-01 -3.78448784e-01 4.31650013e-01 6.08718753e-01 -3.89498413e-01 -1.24205768e-01 1.63991809e-01]
[7.975508689880371, 5.528327465057373]
64a76bf7-0170-493c-8391-3ada58f0c0a3
brain-structure-ages-a-new-biomarker-for
2304.06591
null
https://arxiv.org/abs/2304.06591v1
https://arxiv.org/pdf/2304.06591v1.pdf
Brain Structure Ages -- A new biomarker for multi-disease classification
Age is an important variable to describe the expected brain's anatomy status across the normal aging trajectory. The deviation from that normative aging trajectory may provide some insights into neurological diseases. In neuroimaging, predicted brain age is widely used to analyze different diseases. However, using only the brain age gap information (\ie the difference between the chronological age and the estimated age) can be not enough informative for disease classification problems. In this paper, we propose to extend the notion of global brain age by estimating brain structure ages using structural magnetic resonance imaging. To this end, an ensemble of deep learning models is first used to estimate a 3D aging map (\ie voxel-wise age estimation). Then, a 3D segmentation mask is used to obtain the final brain structure ages. This biomarker can be used in several situations. First, it enables to accurately estimate the brain age for the purpose of anomaly detection at the population level. In this situation, our approach outperforms several state-of-the-art methods. Second, brain structure ages can be used to compute the deviation from the normal aging process of each brain structure. This feature can be used in a multi-disease classification task for an accurate differential diagnosis at the subject level. Finally, the brain structure age deviations of individuals can be visualized, providing some insights about brain abnormality and helping clinicians in real medical contexts.
['Pierrick Coupé', 'Boris Mansencal', 'Michaël Clément', 'Huy-Dung Nguyen']
2023-04-13
null
null
null
null
['age-estimation', 'anatomy', 'age-estimation']
['computer-vision', 'miscellaneous', 'miscellaneous']
[-6.00848682e-02 6.82344735e-02 9.80818719e-02 -5.51208854e-01 -3.30089748e-01 -6.81700855e-02 3.55733961e-01 6.34916902e-01 -6.27447307e-01 7.34421194e-01 -6.81549534e-02 -1.91815242e-01 6.88230433e-03 -8.37106466e-01 -3.21134210e-01 -8.95038962e-01 -3.12487394e-01 6.63462758e-01 1.55610949e-01 1.23640083e-01 2.99965262e-01 5.53330243e-01 -1.50373435e+00 -2.90185779e-01 1.52920449e+00 1.11265469e+00 2.64728278e-01 2.22237378e-01 -2.55408555e-01 8.19322001e-03 -4.28113580e-01 -2.06064284e-01 -6.33622259e-02 -3.85114253e-01 -5.77960789e-01 -1.79118812e-01 4.23334122e-01 -3.19174439e-01 2.69966066e-01 1.22835743e+00 4.50561851e-01 -1.71741337e-01 9.37267840e-01 -1.02986228e+00 -9.13182572e-02 5.59765339e-01 -7.44841397e-01 4.59973484e-01 1.65571690e-01 8.02191198e-02 5.61525464e-01 -5.26630998e-01 4.11266714e-01 1.08172226e+00 5.02351642e-01 6.37002409e-01 -9.50454652e-01 -7.62485743e-01 1.07294336e-01 5.49142122e-01 -1.11087441e+00 -9.85214114e-02 6.00159883e-01 -8.99687588e-01 7.14968070e-02 4.87438552e-02 1.06967545e+00 7.69422412e-01 5.72684705e-01 4.09281313e-01 1.49867535e+00 -3.26214870e-03 2.79442847e-01 -5.92191160e-01 4.16916490e-01 8.04500818e-01 5.37849665e-01 -1.30606189e-01 -2.73870617e-01 3.72685790e-02 4.61101353e-01 -7.30215982e-02 -3.10009837e-01 -2.47406829e-02 -1.26866210e+00 3.96965206e-01 5.91900766e-01 5.27402222e-01 -5.64853251e-01 1.65262353e-02 3.65766138e-01 9.37135294e-02 9.12915230e-01 2.72553086e-01 -4.55583662e-01 6.92017004e-02 -1.13470197e+00 4.91366178e-01 1.42110407e-01 -1.99170277e-01 5.94737172e-01 9.48250964e-02 -7.51127228e-02 7.58669436e-01 3.54481488e-01 3.88349116e-01 7.59420931e-01 -8.14712226e-01 1.50913939e-01 6.24645531e-01 -2.35931948e-01 -5.08916080e-01 -9.22786117e-01 -5.72110415e-01 -1.01589978e+00 4.89210427e-01 1.09620857e+00 1.61281265e-02 -8.69249463e-01 1.77938163e+00 5.55136383e-01 6.13486767e-02 -3.74286890e-01 8.54976118e-01 4.29063469e-01 2.39659429e-01 2.28691325e-01 -2.86439717e-01 1.86518312e+00 -3.03348184e-01 -2.89261192e-01 -3.95656973e-01 5.92000782e-01 -1.15740776e-01 7.10163474e-01 4.15838003e-01 -1.00625527e+00 -4.16962117e-01 -9.57048237e-01 2.52009332e-01 -3.55431974e-01 -8.99678189e-03 3.59172016e-01 6.31036639e-01 -9.46342230e-01 8.03724945e-01 -1.28201556e+00 -4.02653694e-01 5.96782565e-01 9.96227339e-02 -3.75779986e-01 -8.12454149e-02 -1.12341976e+00 1.17673802e+00 5.51159799e-01 4.75236103e-02 -5.46957314e-01 -1.15790415e+00 -7.56919563e-01 -5.16766012e-02 -5.70080839e-02 -8.39905500e-01 6.42548025e-01 -7.16544151e-01 -8.48311186e-01 1.12310874e+00 -2.32392982e-01 -5.76089740e-01 7.20701575e-01 -2.73751885e-01 -2.84248978e-01 2.34938264e-01 2.14596123e-01 5.99987924e-01 8.53526831e-01 -8.22464287e-01 -2.93876618e-01 -1.19688034e+00 -4.20398355e-01 9.23685078e-03 -1.34540066e-01 2.16800179e-02 1.82863533e-01 -6.20279968e-01 4.62367624e-01 -6.40584469e-01 -3.22107822e-01 4.88145202e-01 -2.95658469e-01 -2.51397699e-01 5.46651721e-01 -1.47140813e+00 9.66009855e-01 -1.78804171e+00 2.26122782e-01 2.35234812e-01 8.14976692e-01 -1.52225330e-01 2.57354170e-01 -2.92392790e-01 -1.56805933e-01 3.96127440e-02 -5.72410285e-01 -2.50352174e-01 -4.45472062e-01 -4.51612651e-01 3.13173205e-01 6.68727338e-01 -4.93791653e-03 6.48908496e-01 -6.98514521e-01 -6.51309729e-01 -1.89941108e-01 3.64160210e-01 -2.37458557e-01 6.16975836e-02 6.91946000e-02 9.94774044e-01 -3.64380419e-01 5.50694764e-01 5.73664546e-01 2.96138316e-01 -1.27583623e-01 -9.91846919e-02 -1.52084365e-01 -1.98394150e-01 -4.71395701e-01 1.47237885e+00 -3.19388777e-01 4.15008992e-01 5.94859049e-02 -1.35208607e+00 1.16154718e+00 9.26818326e-02 8.12010169e-01 -6.24858260e-01 1.97349906e-01 4.93024915e-01 6.42446637e-01 -2.38376379e-01 -2.86792874e-01 -1.52905792e-01 1.42286003e-01 6.93716943e-01 -2.13865370e-01 1.59577001e-02 2.79330164e-01 -3.04007232e-01 8.40467751e-01 -1.94374830e-01 2.89502800e-01 -6.23069346e-01 8.59717667e-01 -5.48407972e-01 7.98651397e-01 1.02756456e-01 -3.98416817e-01 3.96876514e-01 7.64054418e-01 -5.43101370e-01 -1.14499509e+00 -1.30603111e+00 -3.24987531e-01 5.19432306e-01 -2.07037851e-01 7.96291381e-02 -1.15353787e+00 -5.93849003e-01 2.67231967e-02 4.79708076e-01 -5.84233403e-01 -3.92479300e-01 -6.35616124e-01 -1.04034138e+00 3.49517167e-01 6.04357660e-01 7.52766311e-01 -7.46343434e-01 -7.46023953e-01 9.06307902e-03 -2.29841977e-01 -8.70782673e-01 -2.75086731e-01 -2.41318494e-01 -1.33045745e+00 -1.15032911e+00 -1.23245513e+00 -3.33761811e-01 9.09903705e-01 -3.85615081e-01 7.61413634e-01 3.61008674e-01 -4.76330847e-01 1.33973882e-01 -1.17465489e-01 -4.46089953e-01 -5.12494922e-01 1.65468514e-01 4.34928685e-01 4.51036692e-02 3.63043509e-02 -9.02197301e-01 -1.01109219e+00 1.08694330e-01 -5.44063628e-01 -2.31526885e-02 4.98729795e-01 4.38255221e-01 4.04045254e-01 -1.52358040e-01 8.29365134e-01 -4.77927804e-01 4.53300267e-01 -4.19178337e-01 -4.58084553e-01 2.57890344e-01 -1.05057323e+00 3.08599740e-01 5.18563449e-01 -3.90656352e-01 -9.27240133e-01 -3.09649259e-01 -3.70339692e-01 5.68724275e-02 -2.06268281e-01 5.67626417e-01 -3.50090742e-01 3.47178787e-01 3.42437476e-01 2.03447357e-01 5.11673391e-01 -6.21383965e-01 6.26204088e-02 3.92362386e-01 6.17726624e-01 -6.97557151e-01 4.21555042e-01 3.56533259e-01 4.25893396e-01 -7.65713453e-01 -5.90849519e-01 -1.11315072e-01 -1.08983600e+00 -6.62922025e-01 9.55648720e-01 -5.11040628e-01 -5.60071886e-01 9.96033072e-01 -9.70780611e-01 -4.23492193e-01 8.92793015e-02 5.41140079e-01 -4.83298331e-01 5.69915414e-01 -4.23132658e-01 -6.59164131e-01 -8.00139070e-01 -1.08756912e+00 7.81510651e-01 2.73814321e-01 -2.70673275e-01 -1.16684949e+00 -1.21971481e-01 3.16253960e-01 2.95081079e-01 5.83998561e-01 1.22922897e+00 -5.95002353e-01 2.59421580e-02 -2.57801503e-01 -2.41236076e-01 2.82846242e-01 -1.02120519e-01 -9.64142978e-02 -5.30965209e-01 -1.15805075e-01 3.64033435e-03 1.91500828e-01 9.71222699e-01 8.04079711e-01 1.32088959e+00 1.82775557e-01 -3.33935291e-01 3.99542063e-01 1.06286907e+00 2.11566597e-01 5.99031210e-01 2.35866591e-01 6.15459561e-01 9.93496776e-01 5.92360795e-01 4.28464800e-01 3.98332566e-01 4.84849453e-01 3.71170372e-01 2.05176666e-01 -4.21276204e-02 2.96857476e-01 3.59478772e-01 5.25192082e-01 -4.01596278e-01 4.28392321e-01 -1.39813411e+00 4.08848703e-01 -1.40677583e+00 -7.53539681e-01 -4.42384899e-01 2.46887612e+00 6.75297379e-01 2.97829509e-01 4.01270539e-01 1.70560792e-01 1.07241964e+00 2.44902056e-02 -8.74418795e-01 -2.06694677e-01 1.24954887e-01 1.78526744e-01 2.99553514e-01 1.33164719e-01 -6.63493514e-01 3.61817181e-01 5.89376879e+00 4.29818332e-01 -1.23292375e+00 3.09719861e-01 9.38713729e-01 1.72250584e-01 -5.99753074e-02 -1.10902391e-01 -6.31874740e-01 6.55159891e-01 8.95135403e-01 -4.13390070e-01 1.60102069e-01 5.57117462e-01 3.68165821e-01 -4.95251119e-01 -9.94367898e-01 9.08914447e-01 -8.42584223e-02 -6.61191881e-01 -2.59246707e-01 2.70931125e-01 -2.00195145e-02 -2.05012575e-01 -2.33295616e-02 -1.11375097e-02 -3.27444583e-01 -7.91923225e-01 7.16917396e-01 1.12454903e+00 1.04528868e+00 -6.82603300e-01 6.24706566e-01 4.32159394e-01 -1.27339005e+00 -2.09707260e-01 1.05807371e-02 2.78544594e-02 4.60633129e-01 1.27776456e+00 -6.19617641e-01 3.91772985e-01 7.10345566e-01 5.31917214e-01 -9.21785891e-01 1.27393627e+00 -1.15962394e-01 3.14939499e-01 -2.00631380e-01 4.46337014e-01 -2.81524658e-01 -4.79060739e-01 5.76612413e-01 6.16846919e-01 8.35597157e-01 -1.73426315e-01 -1.34153053e-01 1.04973733e+00 1.74878761e-01 1.42903402e-01 -2.31140390e-01 7.18443766e-02 2.81197011e-01 1.31255519e+00 -1.15899706e+00 -2.87504226e-01 9.20149870e-03 7.56745279e-01 2.71409690e-01 1.38240047e-02 -5.48652172e-01 -1.57654598e-01 8.28869104e-01 4.73504871e-01 -3.72445226e-01 -4.49693173e-01 -4.26159382e-01 -1.06676984e+00 5.86710423e-02 -3.96873266e-01 2.79369771e-01 -6.21150494e-01 -1.05104554e+00 4.94729549e-01 1.99166864e-01 -8.41741264e-01 -2.47498587e-01 -6.33623183e-01 -1.09682095e+00 8.55880558e-01 -9.84769404e-01 -7.96479821e-01 -4.84494865e-01 7.70692825e-02 3.49493921e-01 -4.78393473e-02 4.38432842e-01 2.79138088e-01 -7.58404791e-01 3.07531983e-01 -1.67397857e-01 1.13297105e-01 6.25511289e-01 -1.50250471e+00 2.62321383e-01 1.01122522e+00 -5.15094519e-01 2.74538517e-01 7.90884793e-01 -9.55245674e-01 -4.86731470e-01 -9.08410728e-01 7.02901602e-01 -1.88278139e-01 6.49436533e-01 2.20427944e-04 -1.20800447e+00 1.93997890e-01 -3.33246201e-01 3.34648117e-02 5.68509400e-01 2.16396432e-02 -2.02679083e-01 -4.33482230e-01 -1.27386117e+00 5.26626885e-01 9.61435080e-01 -1.33711934e-01 -4.45687443e-01 -2.19264627e-01 3.37797761e-01 -5.26136234e-02 -1.27613115e+00 4.06327337e-01 1.00131965e+00 -9.66520548e-01 9.36601102e-01 -2.57669747e-01 3.47997844e-01 6.46508783e-02 4.53606218e-01 -1.44608414e+00 9.88769680e-02 2.14849025e-01 -2.73136556e-01 1.35332143e+00 6.57200739e-02 -8.46770823e-01 6.90546989e-01 7.67341077e-01 -1.18146613e-01 -1.05991399e+00 -1.04207444e+00 -7.86763191e-01 4.63272691e-01 -3.43213946e-01 7.72751153e-01 5.40197194e-01 -4.67074186e-01 -1.58283457e-01 2.81443268e-01 2.60619745e-02 9.42035794e-01 -2.71407962e-01 1.77911103e-01 -1.98308837e+00 3.30015689e-01 -8.76015604e-01 -6.07847333e-01 -3.17286611e-01 4.73670870e-01 -8.57918620e-01 -2.36016840e-01 -1.67551744e+00 1.37195438e-01 -4.46088642e-01 -3.39902133e-01 3.24807651e-02 -3.68894309e-01 -4.15899083e-02 5.75621575e-02 4.75103632e-02 1.64895073e-01 4.84327406e-01 1.16081655e+00 -1.54249460e-01 -9.38992575e-02 2.08647162e-01 -4.13442373e-01 8.85805190e-01 1.22152126e+00 -2.88850218e-01 -1.51873678e-01 1.34989381e-01 1.79178957e-02 -1.61787689e-01 4.99552459e-01 -1.35875607e+00 -1.71082214e-01 1.40027642e-01 5.63319445e-01 -6.50691807e-01 6.43905178e-02 -4.61341918e-01 -5.02215046e-03 1.03729582e+00 1.18265547e-01 1.89191505e-01 -2.53353000e-01 3.32735360e-01 5.21291532e-02 -3.92129570e-01 1.03367031e+00 -1.44088060e-01 -3.03487837e-01 6.43931210e-01 -5.30931532e-01 -1.06671616e-01 1.02575421e+00 -2.95397460e-01 -1.55093610e-01 6.71702623e-03 -8.30312252e-01 2.66307592e-01 4.89458382e-01 1.65482417e-01 5.60364187e-01 -1.16357040e+00 -7.76797593e-01 -1.45198286e-01 8.91372114e-02 -5.50072752e-02 5.00974894e-01 1.58778536e+00 -7.69831181e-01 -1.22049674e-01 -6.57761097e-01 -6.91084206e-01 -1.37193024e+00 4.58458155e-01 5.65647721e-01 -4.61385041e-01 -5.52323341e-01 3.96166563e-01 1.78172052e-01 -3.35259438e-02 -1.35144576e-01 -3.59609753e-01 -5.40239394e-01 6.43547535e-01 7.38898814e-01 5.85429192e-01 -1.16475888e-01 -7.10201085e-01 -3.15728366e-01 7.71943450e-01 -5.72608821e-02 -1.20882496e-01 1.28861213e+00 -4.05505091e-01 -5.41857600e-01 7.31636643e-01 8.85710835e-01 -3.56144190e-01 -1.12619185e+00 4.07077335e-02 3.31911951e-01 6.20881841e-03 6.39047250e-02 -4.71358359e-01 -1.37752509e+00 1.19740665e+00 1.07169437e+00 6.71711043e-02 1.22199702e+00 2.00709999e-01 9.17817235e-01 -8.97654966e-02 4.12225068e-01 -9.41463828e-01 -2.11029440e-01 -3.40485722e-02 8.22111428e-01 -1.07628918e+00 1.26008287e-01 -1.63614184e-01 -3.37133110e-01 1.34744847e+00 7.52463698e-01 1.19380131e-01 8.11418831e-01 -9.75393057e-02 -1.19375266e-01 -1.61988199e-01 -8.62543434e-02 -2.84653127e-01 4.41507727e-01 6.71308875e-01 4.77767169e-01 1.27998993e-01 -9.78406847e-01 7.59260774e-01 -3.59347790e-01 -3.75479817e-01 2.00438365e-01 2.78997570e-01 -7.16473699e-01 -1.20241475e+00 -5.81746161e-01 9.77587461e-01 -6.39559805e-01 1.67340040e-01 -1.50286332e-01 4.71426904e-01 3.58723015e-01 2.66173929e-01 2.68059939e-01 8.23540427e-03 -9.01917145e-02 4.87042248e-01 6.49617434e-01 -4.42588270e-01 -1.22097544e-01 -3.53462309e-01 -1.68741256e-01 -4.02843744e-01 -4.11647052e-01 -9.66402233e-01 -1.24298608e+00 -9.28194672e-02 2.94469367e-03 -1.30380228e-01 7.73722053e-01 1.17054403e+00 -1.58004351e-02 5.07097125e-01 4.16347891e-01 -8.17384422e-01 -5.20282052e-03 -9.17785347e-01 -8.44565630e-01 2.89831817e-01 9.54428911e-02 -1.06485713e+00 -2.60666400e-01 -7.48148814e-05]
[14.079485893249512, -1.572855830192566]
d36ca94a-eae7-4c80-af5e-5e4b554443fc
generative-prompt-tuning-for-relation-1
2210.12435
null
https://arxiv.org/abs/2210.12435v1
https://arxiv.org/pdf/2210.12435v1.pdf
Generative Prompt Tuning for Relation Classification
Using prompts to explore the knowledge contained within pre-trained language models for downstream tasks has now become an active topic. Current prompt tuning methods mostly convert the downstream tasks to masked language modeling problems by adding cloze-style phrases and mapping all labels to verbalizations with fixed length, which has proven effective for tasks with simple label spaces. However, when applied to relation classification exhibiting complex label spaces, vanilla prompt tuning methods may struggle with label verbalizations with arbitrary lengths due to rigid prompt restrictions. Inspired by the text infilling task for pre-training generative models that can flexibly predict missing spans, we propose a novel generative prompt tuning method to reformulate relation classification as an infilling problem, which frees our approach from limitations of current prompt based approaches and thus fully exploits rich semantics of entity and relation types. In addition, we design entity-guided decoding and discriminative relation scoring to generate and align relations effectively and efficiently during inference. Extensive experiments under fully supervised settings and low-resource settings demonstrate the effectiveness of our approach.
['Wei Lu', 'Shengkun Ma', 'Bo Cheng', 'Shuai Zhao', 'Jiale Han']
2022-10-22
null
null
null
null
['relation-classification', 'text-infilling']
['natural-language-processing', 'natural-language-processing']
[ 4.75784123e-01 6.19872451e-01 -6.09974384e-01 -6.24595284e-01 -9.23913121e-01 -7.71497786e-01 7.05575049e-01 2.13115692e-01 -2.56356984e-01 8.54563534e-01 5.71288407e-01 -6.95073366e-01 -5.93346842e-02 -7.84788430e-01 -4.61880535e-01 -2.45650679e-01 1.70567572e-01 8.57758284e-01 -9.47915167e-02 -1.52503937e-01 -1.15362480e-01 6.39155358e-02 -1.19458747e+00 3.80618334e-01 9.93015110e-01 5.47610641e-01 3.95541638e-02 4.77394134e-01 -4.05382335e-01 7.82013535e-01 -4.23219919e-01 -6.31925881e-01 5.52600734e-02 -3.25571179e-01 -1.20277262e+00 1.81963407e-02 2.94289440e-01 -2.73043215e-01 -2.69985646e-01 6.80352986e-01 3.23233187e-01 2.64600337e-01 8.07709038e-01 -1.13793254e+00 -8.91416490e-01 1.33630610e+00 -2.76809514e-01 4.91928570e-02 4.58078682e-01 -2.39639491e-01 1.57272673e+00 -9.94495630e-01 6.79790616e-01 1.29933834e+00 7.21492887e-01 6.80185139e-01 -1.67152238e+00 -6.53276682e-01 7.09720194e-01 1.12759948e-01 -1.46665657e+00 -5.80605149e-01 5.55824339e-01 -4.69893008e-01 1.35122228e+00 3.10842723e-01 2.51416057e-01 1.38128090e+00 -5.22916496e-01 7.64103532e-01 8.66180718e-01 -5.65576911e-01 -6.98197484e-02 -1.01756193e-02 2.56189167e-01 6.10361278e-01 1.15018547e-01 -2.03264922e-01 -5.85441887e-01 -2.14827165e-01 6.51959360e-01 -2.04842910e-01 -2.41705596e-01 -5.86518310e-02 -1.38043094e+00 8.45562041e-01 2.72745281e-01 8.89496580e-02 -1.89050868e-01 1.38853878e-01 2.75023967e-01 1.92035198e-01 7.35083044e-01 7.73579776e-01 -6.85942888e-01 -1.50182948e-01 -9.09075081e-01 3.28766793e-01 1.03350389e+00 1.48758733e+00 7.73387432e-01 -3.72034550e-01 -7.16336846e-01 1.01133871e+00 3.57224166e-01 1.76790282e-01 1.85746193e-01 -8.64164174e-01 7.94563770e-01 7.55895078e-01 -3.75785977e-02 -4.83937383e-01 -4.99277383e-01 -6.18824840e-01 -6.43192291e-01 -6.07255876e-01 4.95810091e-01 -2.07065433e-01 -8.60902667e-01 2.16797733e+00 3.88045818e-01 1.59882545e-01 1.40613183e-01 6.33391023e-01 8.44852626e-01 5.08698404e-01 4.67554778e-01 -1.97350070e-01 1.67040312e+00 -1.10751331e+00 -7.02692807e-01 -5.17331481e-01 1.22014904e+00 -6.27988935e-01 1.46014214e+00 -8.11762065e-02 -7.84005940e-01 -3.43534023e-01 -8.12552691e-01 -5.74199498e-01 -1.57315239e-01 4.32107747e-02 1.05942786e+00 4.21975404e-01 -9.22651887e-01 4.49594945e-01 -8.32205296e-01 -3.20122898e-01 3.42082292e-01 2.60131478e-01 -1.41854659e-01 -1.71453655e-01 -1.46679938e+00 9.03447151e-01 4.94270146e-01 9.33009535e-02 -5.02650917e-01 -8.76019478e-01 -1.00576878e+00 2.91518033e-01 7.34149158e-01 -1.01383841e+00 1.44318891e+00 -2.22026825e-01 -1.46811032e+00 8.64109576e-01 -4.85820711e-01 -2.61571556e-01 2.38726899e-01 -3.72084290e-01 -1.74339741e-01 -3.58456641e-01 1.76267207e-01 7.72085726e-01 3.26990098e-01 -1.07661998e+00 -3.82335156e-01 4.79766242e-02 3.07905614e-01 5.11752009e-01 -3.91230971e-01 6.31457791e-02 -3.84502858e-01 -7.37006962e-01 2.72298187e-01 -8.75648379e-01 -3.16094279e-01 -4.43242967e-01 -8.21824372e-01 -6.72017813e-01 3.47348630e-01 -4.10038829e-01 1.54812622e+00 -1.75529861e+00 2.37057716e-01 1.15461379e-01 3.64455432e-01 3.53039764e-02 -1.91911444e-01 5.57833672e-01 7.99789280e-02 3.12618613e-01 -2.66044289e-01 -6.22932017e-01 3.34473759e-01 4.18178350e-01 -5.96331537e-01 -2.30345428e-01 3.64739805e-01 1.28071606e+00 -1.07359350e+00 -6.46275938e-01 -2.94382155e-01 2.25001723e-01 -7.11779654e-01 4.10551786e-01 -5.18369019e-01 3.98924440e-01 -4.06959325e-01 6.44111216e-01 1.60787016e-01 -6.42913103e-01 4.05297071e-01 -9.09135267e-02 2.46913910e-01 1.09461665e+00 -8.45051646e-01 1.96786022e+00 -6.44451261e-01 2.52743661e-01 -1.91956162e-01 -7.93501854e-01 8.82272899e-01 4.54072297e-01 1.62542909e-02 -1.50927812e-01 -1.56161875e-01 2.25125283e-01 -2.55213697e-02 -6.02769911e-01 6.06824100e-01 -3.03422898e-01 -2.99212694e-01 6.70676291e-01 1.53405458e-01 -8.62944573e-02 2.62286901e-01 3.90812993e-01 1.31087303e+00 5.04362941e-01 3.53196293e-01 5.14401607e-02 1.11417942e-01 -3.95678170e-02 6.55291557e-01 7.16840088e-01 2.33121052e-01 4.03292745e-01 5.65855682e-01 -7.50765055e-02 -7.76229799e-01 -1.03586459e+00 7.08562974e-03 1.54170942e+00 -6.73880577e-02 -8.50830376e-01 -4.94366914e-01 -8.93796325e-01 -2.14427952e-02 8.49804103e-01 -3.55678678e-01 -1.44101262e-01 -7.51712978e-01 -7.97435820e-01 6.66822672e-01 8.09789658e-01 5.40087372e-02 -1.04540741e+00 2.83523258e-02 4.43136245e-01 -5.22811532e-01 -1.49209678e+00 -4.05058831e-01 4.91014540e-01 -7.33470559e-01 -7.20862508e-01 -3.11076313e-01 -9.36383545e-01 7.36543357e-01 -1.44362658e-01 1.43613100e+00 1.92857943e-02 3.86305079e-02 -9.52080786e-02 -3.75582069e-01 2.41137650e-02 -3.99424195e-01 7.62037814e-01 -1.52569219e-01 -3.96610498e-01 6.10262156e-01 -7.99879730e-01 -3.75403166e-01 7.78046325e-02 -5.63383520e-01 5.26914835e-01 5.94638705e-01 9.20917332e-01 3.25928241e-01 -4.89251733e-01 7.50378072e-01 -1.51641762e+00 7.89874315e-01 -5.88891149e-01 -2.28958726e-01 5.05743802e-01 -8.15935791e-01 4.87991333e-01 4.78366435e-01 -5.71936190e-01 -1.08633292e+00 -1.72462001e-01 -7.40799531e-02 -7.85994902e-02 -2.70735085e-01 6.46116078e-01 -3.36362094e-01 5.28294981e-01 6.66333020e-01 -1.91727564e-01 -3.36557329e-01 -5.80996037e-01 9.62546527e-01 6.63231015e-01 6.08261526e-01 -1.09688485e+00 8.22370827e-01 7.51357451e-02 -4.99105416e-02 -2.10030481e-01 -1.40516818e+00 -3.58426362e-01 -6.57162845e-01 3.55154872e-01 7.62242258e-01 -9.99578059e-01 -5.49068153e-01 -4.33411822e-02 -1.32116723e+00 -6.59255028e-01 -3.27055037e-01 3.58801544e-01 -4.37088311e-01 2.02901751e-01 -1.01069212e+00 -5.73645532e-01 -3.60002667e-01 -8.97130132e-01 1.21402991e+00 -9.98342559e-02 -8.71154130e-01 -1.20003629e+00 -2.58950088e-02 3.87794763e-01 3.72182131e-01 -6.27406687e-02 1.41919756e+00 -9.29887712e-01 -7.19567478e-01 1.54773951e-01 -3.59014034e-01 -1.27513126e-01 1.36001348e-01 -3.90999734e-01 -9.16974723e-01 1.24315068e-01 -5.08121431e-01 -6.35865748e-01 8.23118210e-01 -1.03672743e-01 1.01293337e+00 -4.63587582e-01 -5.99150777e-01 7.75820374e-01 8.47583890e-01 -2.26484910e-01 1.88392192e-01 1.82576135e-01 1.04889452e+00 9.28199828e-01 6.27989829e-01 1.96001142e-01 8.86763155e-01 8.22055101e-01 3.21624279e-02 -1.67569369e-01 -1.94524601e-01 -6.83633268e-01 1.19152158e-01 7.55747020e-01 1.55679345e-01 -3.66287977e-01 -8.86888146e-01 4.82409418e-01 -1.90264916e+00 -7.05295563e-01 -2.46527959e-02 1.74936926e+00 1.56256473e+00 8.57112929e-02 -1.69553503e-01 -9.39908698e-02 5.62766492e-01 9.88828838e-02 -4.25085992e-01 -2.57512808e-01 1.79862186e-01 3.75307918e-01 2.70184427e-01 6.71927929e-01 -1.05621147e+00 1.44385886e+00 5.79145575e+00 8.56871307e-01 -7.37417817e-01 2.18110040e-01 4.79528636e-01 -8.07581916e-02 -8.64765704e-01 4.35231298e-01 -1.20379448e+00 2.07918122e-01 7.69916356e-01 3.47639993e-02 4.64640230e-01 6.64003968e-01 8.44424888e-02 2.88278282e-01 -1.74712658e+00 8.42111647e-01 -2.55733520e-01 -1.18026161e+00 -3.96507867e-02 -4.24799249e-02 5.17798901e-01 -1.79964423e-01 -2.61366982e-02 8.35430205e-01 7.09590137e-01 -1.20084739e+00 6.26347482e-01 1.23596191e-01 1.20485461e+00 -2.79298574e-01 3.65411937e-01 3.65524501e-01 -1.18621624e+00 5.47555648e-02 -2.26668686e-01 -3.33639920e-01 5.21881461e-01 5.96796155e-01 -1.31588125e+00 3.83748084e-01 2.17397418e-02 5.95099270e-01 -5.19461393e-01 4.80582505e-01 -8.39270532e-01 8.89249563e-01 -3.39465588e-01 -2.55912635e-02 1.11303717e-01 -1.45561509e-02 3.34736437e-01 1.41802001e+00 2.56312996e-01 1.91579849e-01 4.28995550e-01 9.97694552e-01 -3.37969542e-01 1.32122666e-01 -4.90002245e-01 -2.43318513e-01 1.10114503e+00 1.27575302e+00 -6.57947361e-01 -4.27271754e-01 -3.58926207e-01 7.84132421e-01 8.64993155e-01 6.60604954e-01 -6.78460836e-01 -1.87277019e-01 6.45428836e-01 1.43607646e-01 1.03475913e-01 -2.65404850e-01 -5.82591236e-01 -1.33355033e+00 4.09195311e-02 -5.76464474e-01 4.58337247e-01 -5.74546874e-01 -1.26034677e+00 6.20864511e-01 2.94329017e-01 -8.38039100e-01 -6.84677184e-01 -3.17316681e-01 -4.42175448e-01 8.98721874e-01 -1.40111423e+00 -1.59559143e+00 -1.20046824e-01 2.75572211e-01 5.75324357e-01 1.05451785e-01 1.16767204e+00 2.87377328e-01 -5.39085805e-01 8.98901761e-01 -4.86198395e-01 8.24318230e-02 8.78570914e-01 -1.47153771e+00 6.42108202e-01 8.31561983e-01 4.93674099e-01 1.01613450e+00 7.14322269e-01 -7.14789033e-01 -8.80960703e-01 -1.21128857e+00 1.50969362e+00 -5.99738657e-01 7.28018939e-01 -8.53790700e-01 -1.08980381e+00 1.05768514e+00 8.22601169e-02 -8.67168605e-02 9.95393157e-01 9.88955081e-01 -6.99887693e-01 2.65019506e-01 -5.15322208e-01 7.86745667e-01 1.83056164e+00 -7.00455010e-01 -6.65275335e-01 5.56304872e-01 1.13079870e+00 -6.61469758e-01 -8.10550272e-01 5.19851446e-01 2.80820876e-01 -1.91633478e-01 8.11490178e-01 -7.86197186e-01 3.90572369e-01 -1.53928980e-01 3.91489863e-02 -1.32579124e+00 -3.86039793e-01 -8.82747352e-01 -5.23286045e-01 1.71711087e+00 9.77522135e-01 -4.77589548e-01 7.52835631e-01 7.95843422e-01 -3.27155948e-01 -9.22916412e-01 -5.00303388e-01 -5.50931752e-01 -7.12763667e-02 -4.71673280e-01 8.47435534e-01 1.01267672e+00 2.46368632e-01 8.93383503e-01 -2.31709793e-01 4.98738512e-02 3.21965456e-01 4.23498869e-01 6.15565598e-01 -1.20102656e+00 -6.97649360e-01 -3.14852774e-01 8.47951397e-02 -1.50421464e+00 6.13107264e-01 -1.46531498e+00 1.97125480e-01 -1.68581223e+00 2.44187593e-01 -1.16661024e+00 2.46587377e-02 1.06363416e+00 -6.38249218e-01 5.92524111e-02 -1.00783341e-01 2.03323200e-01 -6.42116666e-01 4.81684655e-01 9.86401618e-01 9.05655921e-02 -1.26099005e-01 -4.57070116e-03 -1.09375787e+00 5.73602259e-01 5.17858446e-01 -5.23672521e-01 -7.60194123e-01 -6.32223129e-01 6.37067616e-01 -1.05911762e-01 1.49490967e-01 -3.80953163e-01 1.96430221e-01 -2.68319339e-01 -6.74147084e-02 -3.24595690e-01 2.40253165e-01 -4.24170822e-01 1.18563389e-02 -1.75680190e-01 -9.25800979e-01 -1.17568254e-01 -2.01526552e-01 5.09761870e-01 -9.62896124e-02 -1.37974590e-01 1.70603022e-01 6.67998195e-02 -5.43689311e-01 3.05571049e-01 -3.04855518e-02 4.82499748e-01 6.93318069e-01 1.54092126e-02 -4.63400722e-01 -3.05635720e-01 -9.59000289e-01 4.45594937e-01 3.70753974e-01 4.55390424e-01 3.07449043e-01 -1.42397904e+00 -7.03448951e-01 1.57287523e-01 3.54874551e-01 5.77964008e-01 -1.43403783e-01 7.73479939e-01 -3.84423360e-02 4.29094017e-01 4.84527707e-01 -3.93744946e-01 -1.17990625e+00 4.07889992e-01 -1.27924606e-01 -6.80376410e-01 -6.16735458e-01 1.29735529e+00 3.28131109e-01 -5.85605919e-01 2.67669410e-01 -5.67613363e-01 -1.60444096e-01 6.76537456e-04 1.73108011e-01 -7.67814219e-02 1.24459900e-01 -2.58251399e-01 -2.19382465e-01 1.91773266e-01 -2.81494647e-01 -2.89911598e-01 1.19629920e+00 -2.85086900e-01 -9.80949774e-02 3.86638403e-01 9.70750690e-01 1.09600365e-01 -1.25572073e+00 -6.98960185e-01 6.23514891e-01 -1.69981625e-02 -3.47449005e-01 -8.18150759e-01 -3.37834388e-01 8.31859946e-01 -3.96123707e-01 1.93424057e-02 7.94396460e-01 4.20751095e-01 9.58399236e-01 4.56320882e-01 2.47589335e-01 -7.21413493e-01 -1.94498599e-02 7.90542245e-01 5.80288172e-01 -1.05513132e+00 -3.47484469e-01 -9.06279266e-01 -6.19653404e-01 8.42714489e-01 7.27668822e-01 3.96703899e-01 3.60999882e-01 5.70892513e-01 1.20969027e-01 -1.50681466e-01 -1.16061568e+00 -3.00639659e-01 3.43843520e-01 5.70188224e-01 9.45979953e-01 2.29501888e-01 -4.37344611e-01 7.65552998e-01 -5.21911144e-01 -2.69603908e-01 1.18366286e-01 6.99893236e-01 -2.46570900e-01 -1.58116972e+00 2.67913174e-02 5.46319783e-01 -3.33274871e-01 -5.85535407e-01 -2.21618727e-01 4.75441009e-01 9.17117745e-02 9.42082822e-01 -6.90534385e-03 -3.23573172e-01 1.33559659e-01 6.06098413e-01 5.25719225e-01 -1.40435970e+00 -4.36726421e-01 -8.65893587e-02 6.28488123e-01 -3.71775985e-01 -1.54754117e-01 -5.91419518e-01 -1.40413618e+00 1.99472252e-02 -4.38229829e-01 3.75532150e-01 2.70787239e-01 1.19951820e+00 4.91011113e-01 6.04280174e-01 1.74553946e-01 -5.55441916e-01 -8.59802186e-01 -1.35660160e+00 -1.48075476e-01 3.79948407e-01 -6.44130073e-03 -8.42754841e-01 -1.30422488e-01 1.53007761e-01]
[10.153142929077148, 8.572461128234863]
4c45723d-4418-4be6-b483-6e42a25106ce
uncertainty-aware-cascaded-dilation-filtering
2201.02366
null
https://arxiv.org/abs/2201.02366v1
https://arxiv.org/pdf/2201.02366v1.pdf
Uncertainty-Aware Cascaded Dilation Filtering for High-Efficiency Deraining
Deraining is a significant and fundamental computer vision task, aiming to remove the rain streaks and accumulations in an image or video captured under a rainy day. Existing deraining methods usually make heuristic assumptions of the rain model, which compels them to employ complex optimization or iterative refinement for high recovery quality. This, however, leads to time-consuming methods and affects the effectiveness for addressing rain patterns deviated from from the assumptions. In this paper, we propose a simple yet efficient deraining method by formulating deraining as a predictive filtering problem without complex rain model assumptions. Specifically, we identify spatially-variant predictive filtering (SPFilt) that adaptively predicts proper kernels via a deep network to filter different individual pixels. Since the filtering can be implemented via well-accelerated convolution, our method can be significantly efficient. We further propose the EfDeRain+ that contains three main contributions to address residual rain traces, multi-scale, and diverse rain patterns without harming the efficiency. First, we propose the uncertainty-aware cascaded predictive filtering (UC-PFilt) that can identify the difficulties of reconstructing clean pixels via predicted kernels and remove the residual rain traces effectively. Second, we design the weight-sharing multi-scale dilated filtering (WS-MS-DFilt) to handle multi-scale rain streaks without harming the efficiency. Third, to eliminate the gap across diverse rain patterns, we propose a novel data augmentation method (i.e., RainMix) to train our deep models. By combining all contributions with sophisticated analysis on different variants, our final method outperforms baseline methods on four single-image deraining datasets and one video deraining dataset in terms of both recovery quality and speed.
['Song Wang', 'Wei Feng', 'Di Lin', 'Lei Ma', 'Felix Juefei-Xu', 'Jingyang Sun', 'Qing Guo']
2022-01-07
null
null
null
null
['single-image-deraining']
['computer-vision']
[ 3.44395190e-02 -4.36767578e-01 4.41209465e-01 -4.39792752e-01 -6.25292838e-01 -2.62551427e-01 2.97036976e-01 -5.97099125e-01 -1.35518789e-01 7.61452198e-01 5.23075042e-03 -1.88129485e-01 -6.84713051e-02 -8.35021555e-01 -8.39741528e-01 -1.10021925e+00 1.69411704e-01 -7.12304488e-02 3.81602526e-01 -2.72867471e-01 -1.58124357e-01 5.83295226e-01 -1.67797494e+00 5.41642234e-02 1.63194072e+00 8.12867105e-01 6.25844717e-01 6.87438607e-01 4.19748016e-02 7.78364420e-01 -4.59275365e-01 -1.86047941e-01 5.01559913e-01 -5.59103966e-01 -1.57177880e-01 1.10547148e-01 9.08876956e-01 -6.93138599e-01 -4.30742085e-01 1.00286412e+00 5.14879286e-01 7.17718899e-03 2.90092498e-01 -6.57527566e-01 -4.40063298e-01 2.21499979e-01 -5.34000576e-01 2.68975556e-01 -2.71791726e-01 3.45846951e-01 5.99586725e-01 -1.18214023e+00 1.34980798e-01 1.20828295e+00 1.02258754e+00 2.75139481e-01 -1.11128294e+00 -7.37809896e-01 4.24163103e-01 2.01621726e-01 -1.18997335e+00 -5.03263175e-01 4.50022846e-01 -2.89550066e-01 4.08454776e-01 4.03203875e-01 7.35159576e-01 7.40857422e-01 -2.66467966e-02 7.56250024e-01 1.33713412e+00 -1.77628249e-01 5.22905029e-02 -3.23435515e-01 2.95878381e-01 5.16820908e-01 5.53348720e-01 2.93827355e-01 -1.64087057e-01 2.30297416e-01 6.81139827e-01 4.10722226e-01 -8.94007325e-01 1.60596907e-01 -7.60204613e-01 7.02877879e-01 5.68230629e-01 -1.47353336e-01 -4.83239949e-01 -1.39943287e-01 -2.24791333e-01 3.07095528e-01 8.40701044e-01 3.00528169e-01 -4.70625937e-01 3.82157177e-01 -1.46866739e+00 5.92337966e-01 5.89726686e-01 7.08372772e-01 1.01923478e+00 4.78656620e-01 -5.07300258e-01 7.84902930e-01 3.42211366e-01 1.24271381e+00 -2.71010458e-01 -8.79502952e-01 3.00782174e-01 9.27458182e-02 4.98157561e-01 -6.86693847e-01 -4.01631087e-01 -5.85091829e-01 -1.16838205e+00 4.17040169e-01 1.72021434e-01 -2.48772368e-01 -1.41258442e+00 1.34144676e+00 3.32292855e-01 6.22213125e-01 1.27846420e-01 1.29763269e+00 8.50586116e-01 8.96405160e-01 -1.30784974e-01 -6.29542470e-01 1.07557356e+00 -1.08171296e+00 -9.84677792e-01 -3.99540186e-01 -5.42568192e-02 -8.15404892e-01 1.03335369e+00 4.36927229e-01 -8.19612622e-01 -6.15861356e-01 -1.04375386e+00 1.08079612e-01 1.08807392e-01 1.72847480e-01 6.09818518e-01 3.49515527e-01 -8.14758360e-01 8.27331185e-01 -8.03459942e-01 -7.93720633e-02 5.02448738e-01 -8.27571154e-02 7.66737610e-02 -5.80878019e-01 -1.17977798e+00 7.96840608e-01 1.89631246e-03 7.74857819e-01 -9.52201903e-01 -9.88551438e-01 -7.43624389e-01 5.48949726e-02 2.99854040e-01 -7.97849000e-01 8.19053590e-01 -1.17411900e+00 -1.35328364e+00 2.75478482e-01 -4.38092440e-01 -4.84945208e-01 4.46585238e-01 -9.37976480e-01 -5.56841731e-01 1.12673521e-01 -1.57768488e-01 1.54584989e-01 1.56293726e+00 -1.52435923e+00 -6.03743613e-01 -1.33012414e-01 -1.06887132e-01 2.48432383e-01 1.66733295e-01 -3.10747564e-01 -6.96622670e-01 -1.06034291e+00 1.38720304e-01 -7.61271834e-01 -3.43246549e-01 -2.58559659e-02 -1.47358030e-01 4.50544626e-01 8.69929135e-01 -1.03552032e+00 1.33683205e+00 -2.16398835e+00 6.55783042e-02 -4.01269533e-02 3.64617407e-01 8.35507810e-01 -1.74138233e-01 -1.15579925e-02 1.61516294e-01 -2.13801563e-01 -7.03583837e-01 -4.74088281e-01 -3.60445172e-01 6.02414548e-01 -7.44728744e-01 5.38167596e-01 5.12676239e-01 6.23334944e-01 -6.30844057e-01 -2.09891617e-01 2.98529446e-01 6.84116960e-01 -3.67346466e-01 7.28623033e-01 -3.83513004e-01 4.98775065e-01 -2.62225211e-01 7.53321826e-01 1.53483737e+00 1.32532239e-01 -2.48988092e-01 -5.06522715e-01 -2.30616972e-01 -6.67863414e-02 -1.26815712e+00 1.09913349e+00 -5.41959941e-01 2.38423780e-01 5.00865221e-01 -5.83452582e-01 7.99704731e-01 2.42978223e-02 -3.32633480e-02 -6.05076730e-01 -3.29022586e-01 2.73712784e-01 -3.22875351e-01 -6.52754247e-01 3.96077812e-01 -3.29205334e-01 6.70420110e-01 -6.09167665e-02 -1.31517664e-01 -1.55453816e-01 -1.68713376e-01 1.58732563e-01 8.99906397e-01 4.18891609e-01 -1.99713841e-01 -1.48942575e-01 2.90102512e-01 -1.81018814e-01 1.03564882e+00 9.71100211e-01 -6.33428022e-02 1.17454517e+00 -7.59779438e-02 -6.38484240e-01 -8.37653339e-01 -1.12673378e+00 -2.90661901e-01 8.21749270e-01 4.07815158e-01 -1.02933683e-01 -4.92179841e-01 -5.45712948e-01 8.14191345e-03 5.65715432e-01 -4.91400003e-01 8.77846405e-02 -7.21312881e-01 -1.41279411e+00 2.71839559e-01 2.39266738e-01 7.58680761e-01 -9.76961017e-01 -2.22837254e-01 9.86530781e-02 -2.45719910e-01 -1.16229618e+00 -3.55475396e-01 2.23852605e-01 -9.93621647e-01 -1.00985539e+00 -5.99261165e-01 -3.67271662e-01 4.56107557e-01 7.60539055e-01 1.31773949e+00 2.70346195e-01 -2.16408879e-01 2.29556952e-02 -6.87739313e-01 -4.80962455e-01 1.33592799e-01 -4.05335605e-01 -1.45878747e-01 2.05126211e-01 -2.47421092e-03 -7.67016530e-01 -9.32402670e-01 1.40598118e-01 -1.05148304e+00 1.90754086e-01 7.75273025e-01 9.19834733e-01 7.80415952e-01 -2.35473383e-02 3.34363759e-01 -1.00274146e+00 2.48593584e-01 -5.08845866e-01 -7.56806374e-01 2.28190467e-01 -5.71922898e-01 6.98325550e-03 7.37612665e-01 -1.32472381e-01 -1.46790135e+00 7.55132139e-02 -1.68759346e-01 -6.53612435e-01 -4.73686233e-02 2.82166600e-01 -3.01313788e-01 -1.41902432e-01 5.39537072e-01 4.60557103e-01 -9.27959606e-02 -6.37685537e-01 3.86531532e-01 2.54976481e-01 6.25182450e-01 -2.23697066e-01 1.36977720e+00 7.40807056e-01 -1.21206909e-01 -8.40575457e-01 -1.50523317e+00 -2.59413332e-01 -3.43442649e-01 -1.68787494e-01 6.64652288e-01 -1.48423076e+00 -2.13808954e-01 8.59897494e-01 -9.45523143e-01 -4.76057947e-01 -8.37702528e-02 4.37289596e-01 -9.36166942e-03 4.82921451e-01 -6.49729490e-01 -1.04567719e+00 -7.92947412e-01 -8.28016341e-01 8.91463876e-01 4.39859867e-01 5.29183030e-01 -5.11696279e-01 8.23292285e-02 2.12639689e-01 8.27213466e-01 5.29670566e-02 3.71416390e-01 1.42121956e-01 -9.90580857e-01 1.74666047e-01 -4.81103837e-01 6.58529758e-01 7.24315941e-02 2.14656681e-01 -1.14016688e+00 -3.63824189e-01 1.92307889e-01 -1.11759320e-01 1.51661766e+00 6.03097081e-01 1.10385478e+00 -2.62000412e-01 -2.49087550e-02 1.34579468e+00 1.56058955e+00 -1.57744408e-01 1.04742312e+00 2.92058557e-01 8.70058596e-01 3.21198016e-01 8.39764178e-01 4.51620013e-01 2.71695524e-01 4.23938632e-01 6.86387956e-01 -4.43583488e-01 -3.42448562e-01 2.42564201e-01 4.51110095e-01 7.09713638e-01 -4.79959697e-01 -2.09272593e-01 -2.98816562e-01 5.99572837e-01 -1.94842887e+00 -1.06978965e+00 -3.76083642e-01 2.10694742e+00 8.05517673e-01 -1.87179893e-01 -4.43015575e-01 -3.48326832e-01 4.91197258e-01 4.81551677e-01 -6.21603549e-01 1.57202631e-01 -4.77491081e-01 6.54481769e-01 6.59110248e-01 6.99926734e-01 -1.35228395e+00 9.86924887e-01 5.40932703e+00 8.05263877e-01 -1.15976167e+00 1.50816903e-01 3.79633039e-01 -1.55363962e-01 -3.92812401e-01 1.09827537e-02 -1.05023146e+00 6.72081470e-01 6.08477294e-01 5.72288156e-01 6.03210449e-01 4.95010287e-01 5.90237319e-01 -2.55486071e-01 -3.90299857e-01 8.98201942e-01 -4.39179800e-02 -1.21363676e+00 -3.25888619e-02 -4.70347643e-01 7.45123029e-01 2.74486929e-01 -1.24391027e-01 3.96742046e-01 4.32172209e-01 -1.04385734e+00 5.42642057e-01 1.08874977e+00 5.10335743e-01 -5.62767088e-01 9.09735620e-01 1.79098830e-01 -9.92725790e-01 1.02345601e-01 -5.69215596e-01 4.58094031e-02 1.38911545e-01 1.43340147e+00 -8.65378454e-02 9.78277683e-01 1.21692288e+00 8.52796257e-01 -2.90510505e-01 1.28254688e+00 -5.87217450e-01 1.03728318e+00 -5.18688381e-01 7.76146054e-01 -2.10357588e-02 -5.87348998e-01 5.87042034e-01 1.31595445e+00 4.85539556e-01 4.32066381e-01 8.87616202e-02 7.49301255e-01 9.66835916e-02 -3.89181137e-01 -3.17548245e-01 4.94328320e-01 4.12644148e-01 1.32696044e+00 -2.10580423e-01 -1.93081990e-01 -4.34667468e-01 1.19868600e+00 1.17085084e-01 6.56640291e-01 -1.06492066e+00 -3.52015048e-01 1.19519842e+00 1.63525537e-01 8.31278086e-01 -1.72901034e-01 -1.91286027e-01 -1.45904887e+00 2.47278109e-01 -9.21527505e-01 3.66566912e-03 -6.95019901e-01 -1.43488526e+00 7.51912355e-01 -3.53030592e-01 -1.31745744e+00 4.96876359e-01 -2.75479347e-01 -8.92312646e-01 1.23356783e+00 -2.34913659e+00 -1.24235916e+00 -9.87769485e-01 7.09915340e-01 4.85563874e-01 2.99309075e-01 4.98310417e-01 6.03457034e-01 -7.44243085e-01 3.45889241e-01 3.08344215e-01 -2.69791812e-01 1.18037963e+00 -1.06311214e+00 2.06593886e-01 1.61549902e+00 -1.43557847e-01 3.38291049e-01 9.11499560e-01 -6.03787005e-01 -1.36099553e+00 -1.70073509e+00 5.17329633e-01 -1.45548657e-01 3.20563972e-01 -2.27556482e-01 -1.50447798e+00 4.88742203e-01 -6.42007589e-02 5.77314258e-01 7.86077380e-02 -5.02858758e-02 -4.20053303e-01 -6.48018360e-01 -9.67357099e-01 4.62116301e-01 9.78651106e-01 -1.36455879e-01 -4.03203368e-01 4.27245289e-01 9.68044102e-01 -4.67227787e-01 -5.68901658e-01 8.54986489e-01 2.43520215e-01 -1.38808692e+00 9.69863951e-01 -1.47753626e-01 4.67465967e-01 -8.38934064e-01 -2.52717167e-01 -1.38219655e+00 -4.74565983e-01 -6.35886848e-01 -6.41446531e-01 1.21756423e+00 7.83678144e-02 -5.77921152e-01 3.94883275e-01 2.61611581e-01 -4.89261329e-01 -5.74944317e-01 -5.32090664e-01 -4.78879929e-01 -2.01698065e-01 -3.47444475e-01 4.61382240e-01 8.41419578e-01 -1.06532443e+00 1.41109452e-02 -1.03279161e+00 9.64648902e-01 8.66407514e-01 6.96833968e-01 7.96444118e-01 -1.14888084e+00 -3.69540632e-01 6.45945221e-02 2.25232258e-01 -1.28557658e+00 -1.99123293e-01 -2.85417974e-01 5.85981309e-01 -1.52137744e+00 -1.32446457e-02 -5.29222488e-01 -1.92845732e-01 4.24858391e-01 -8.06284904e-01 1.65393323e-01 2.77077436e-01 6.19470477e-01 -4.55640763e-01 8.87579083e-01 1.36768544e+00 6.77283406e-02 -2.08084062e-01 1.77202523e-01 -5.16440749e-01 8.93232942e-01 6.37974620e-01 -4.37792748e-01 -1.21787451e-01 -7.71662831e-01 1.28575161e-01 -3.62987742e-02 4.63848531e-01 -1.12118161e+00 -5.63622080e-02 -2.61503369e-01 4.79644418e-01 -6.67977691e-01 3.11801344e-01 -7.41107106e-01 1.76156282e-01 1.87598988e-01 3.87274623e-01 -3.84687006e-01 1.35046124e-01 7.91298032e-01 -4.35310632e-01 9.30758789e-02 1.19516253e+00 1.63218640e-02 -8.31329703e-01 5.31845152e-01 -2.72901773e-01 -2.34028772e-01 7.01638997e-01 -8.14476907e-02 -5.13319969e-01 -2.90607929e-01 -6.92751050e-01 4.42925334e-01 4.26843911e-01 3.33097726e-02 7.21005738e-01 -8.34770799e-01 -1.02069521e+00 3.46637338e-01 -1.64868936e-01 3.67282957e-01 7.35941470e-01 1.07200086e+00 -6.97960019e-01 -1.85664043e-01 1.15250088e-01 -3.38479310e-01 -1.06501186e+00 2.68298805e-01 5.07002532e-01 -2.52417564e-01 -1.14419699e+00 9.10744190e-01 4.50112045e-01 -2.71569908e-01 7.33285174e-02 -3.65203708e-01 -1.83509588e-01 -2.74531972e-02 7.52467155e-01 9.51079205e-02 2.17855096e-01 -1.61798999e-01 -2.38657277e-03 5.07601500e-01 -9.61277932e-02 6.01439238e-01 1.70405853e+00 -3.12126517e-01 -3.06026310e-01 1.25750452e-01 5.02990007e-01 2.05475211e-01 -1.89142883e+00 -3.52121651e-01 -5.97478688e-01 -6.55720472e-01 2.92286873e-01 -8.22332859e-01 -1.64150274e+00 7.40872800e-01 8.36396337e-01 -1.88672960e-01 1.60771310e+00 -4.39348400e-01 9.51988280e-01 3.37164253e-01 6.33206666e-02 -5.99875569e-01 -3.37870806e-01 8.86255324e-01 7.54039764e-01 -1.44153082e+00 2.20939860e-01 -4.92478907e-01 -5.77828407e-01 1.07607222e+00 6.17074430e-01 -3.47297370e-01 8.46809506e-01 3.42924476e-01 2.83192575e-01 -1.17693290e-01 -3.98851871e-01 -4.78724837e-01 1.11544847e-01 4.78883833e-01 8.88152421e-02 2.03324668e-02 -1.80301204e-01 7.02060699e-01 2.87319779e-01 2.70788521e-02 4.93766904e-01 6.62428558e-01 -8.21383059e-01 -6.64777339e-01 -7.76016951e-01 6.02169216e-01 -2.79987574e-01 -7.12956250e-01 3.34323376e-01 3.88307244e-01 3.69169921e-01 8.61452639e-01 -1.56272855e-02 -1.94988400e-01 3.21784496e-01 -3.04360986e-01 2.51360536e-01 -3.08666527e-01 -3.81098896e-01 1.86137035e-01 -1.66771278e-01 -6.71637714e-01 -8.02855492e-01 -3.93477559e-01 -7.78571129e-01 -3.98314536e-01 -3.99507284e-01 2.38952022e-02 1.33755818e-01 9.97251630e-01 5.18692017e-01 5.58867991e-01 5.33291399e-01 -1.15690458e+00 -3.05049717e-01 -1.07790709e+00 -9.53823626e-01 1.92044869e-01 7.61923373e-01 -6.84623599e-01 -6.62382841e-01 3.47180367e-02]
[10.88329792022705, -3.2488625049591064]
c77da1c9-599b-4938-a4d2-786c36ec11ff
pipeline-coreference-resolution-model-for
null
null
https://aclanthology.org/2022.codi-crac.3
https://aclanthology.org/2022.codi-crac.3.pdf
Pipeline Coreference Resolution Model for Anaphoric Identity in Dialogues
CODI-CRAC 2022 Shared Task in Dialogues consists of three sub-tasks: Sub-task 1 is the resolution of anaphoric identity, sub-task 2 is the resolution of bridging references, and sub-task 3 is the resolution of discourse deixis/abstract anaphora. Anaphora resolution is the task of detecting mentions from input documents and clustering the mentions of the same entity. The end-to-end model proceeds with the pruning of the candidate mention, and the pruning has the possibility of removing the correct mention. Also, the end-to-end anaphora resolution model has high model complexity, which takes a long time to train. Therefore, we proceed with the anaphora resolution as a two-stage pipeline model. In the first mention detection step, the score of the candidate word span is calculated, and the mention is predicted without pruning. In the second anaphora resolution step, the pair of mentions of the anaphora resolution relationship is predicted using the mentions predicted in the mention detection step. We propose a two-stage anaphora resolution pipeline model that reduces model complexity and training time, and maintains similar performance to end-to-end models. As a result of the experiment, the anaphora resolution showed a performance of 68.27% in Light, 48.87% in AMI, 69.06% in Persuasion, and 60.99% on Switchboard. Our final system ranked 3rd on the leaderboard of sub-task 1.
['Harksoo Kim', 'Mirae Han', 'Seongsik Park', 'Damrin Kim']
null
null
null
null
coling-codi-crac-2022-10
['coreference-resolution']
['natural-language-processing']
[-1.92046165e-02 6.22448206e-01 -2.85429507e-01 -3.12171906e-01 -1.04324210e+00 -5.34698963e-01 6.27613425e-01 2.77902126e-01 -5.58369040e-01 8.92912626e-01 6.33864641e-01 2.04862058e-02 -3.23617965e-01 -7.43582547e-01 -3.58094335e-01 -3.24272096e-01 2.02715963e-01 1.10849762e+00 5.70042372e-01 -4.81306881e-01 3.15234035e-01 2.39832044e-01 -1.09853160e+00 9.79363918e-01 6.03068888e-01 7.40331471e-01 5.48477434e-02 5.66754222e-01 -3.54664326e-01 8.81062806e-01 -9.37296271e-01 -5.17235160e-01 -2.02269837e-01 -1.91449776e-01 -1.54030716e+00 -4.23349559e-01 3.49637717e-01 -1.45419717e-01 -3.42516303e-01 1.05367172e+00 5.11675179e-01 2.05853954e-01 3.06565225e-01 -9.11641121e-01 -6.52658865e-02 1.01256418e+00 -5.53381383e-01 4.01044607e-01 6.47152185e-01 -2.62818426e-01 1.23078477e+00 -8.75530303e-01 1.01754296e+00 1.70224929e+00 5.37689209e-01 5.76709509e-01 -7.94903457e-01 -7.43589878e-01 2.69164629e-02 5.51587164e-01 -1.09816670e+00 -5.38779259e-01 5.10642707e-01 -3.90050322e-01 1.37682247e+00 3.03592533e-01 1.62491307e-01 9.90928173e-01 8.87850299e-02 5.92440426e-01 6.24648273e-01 -3.57349455e-01 8.98856949e-03 -3.19979191e-01 5.80049813e-01 4.48202908e-01 3.27542834e-02 -4.10960406e-01 -6.37427270e-01 -2.49917358e-01 2.16193125e-01 -4.18055952e-01 -2.11426333e-01 4.26010191e-01 -7.82542467e-01 9.50914502e-01 4.26246762e-01 4.38952029e-01 -5.00780642e-01 -4.58927035e-01 5.98981023e-01 1.46170750e-01 2.65964597e-01 6.24099493e-01 -3.77655387e-01 -2.38369197e-01 -7.95397878e-01 7.16367304e-01 1.10752118e+00 9.50145662e-01 2.36264542e-01 -7.37055480e-01 -4.11340237e-01 9.98295248e-01 6.35905564e-02 -3.58874425e-02 5.75059116e-01 -1.25929701e+00 1.02951014e+00 1.00652778e+00 4.29955959e-01 -8.28944147e-01 -8.24464977e-01 -1.36891767e-01 -4.31523263e-01 -2.04071224e-01 7.67232955e-01 -2.64769912e-01 -1.98229566e-01 1.83588386e+00 4.94001240e-01 4.29957844e-02 5.56974113e-01 1.33853436e+00 1.16259444e+00 6.35018170e-01 3.13753128e-01 -6.86086595e-01 2.12963080e+00 -1.10947263e+00 -1.30512166e+00 -4.71067727e-01 5.99021912e-01 -1.06761718e+00 7.93895364e-01 1.55616239e-01 -1.53566015e+00 -3.09422433e-01 -1.09115946e+00 -4.53426659e-01 -6.97035268e-02 -1.64222606e-02 4.91662502e-01 -2.78587043e-01 -1.91151232e-01 6.15475476e-01 -6.32975519e-01 -5.44587255e-01 -1.67510659e-02 2.24677995e-01 -3.33274871e-01 1.06054664e-01 -1.91023231e+00 1.32316196e+00 7.75693357e-01 -1.43720314e-01 -3.59997660e-01 -4.99906451e-01 -8.01490843e-01 4.80198592e-01 4.30710793e-01 -5.44846654e-01 1.58080363e+00 -4.04732049e-01 -1.32242393e+00 1.03793991e+00 -2.54478812e-01 -6.30800188e-01 3.60367954e-01 -6.78906143e-01 -4.82776672e-01 1.59924835e-01 5.88749170e-01 4.44344044e-01 2.50051767e-01 -6.57353640e-01 -1.19899035e+00 -4.01872486e-01 2.15365544e-01 6.46197915e-01 -2.31902319e-04 5.45959055e-01 -5.45485675e-01 -2.25667328e-01 4.02133286e-01 -7.67731905e-01 2.71948606e-01 -7.05444157e-01 -3.93799275e-01 -9.64735985e-01 8.81723583e-01 -8.16908658e-01 1.63620770e+00 -2.10524583e+00 2.85939544e-01 -4.65521276e-01 4.00636345e-02 4.20139551e-01 8.14683214e-02 3.28449279e-01 -2.87679851e-01 -1.89394832e-01 -1.31273061e-01 -2.25492910e-01 -7.56448358e-02 -1.14084065e-01 -5.58559418e-01 1.82830617e-01 3.81925255e-01 4.90615785e-01 -1.01414800e+00 -7.66197860e-01 -2.72929072e-01 -8.40528682e-02 -3.60389382e-01 4.96613860e-01 -3.06238383e-01 9.44685638e-02 -2.82834888e-01 2.34456226e-01 4.54963446e-01 -7.08720237e-02 6.73324108e-01 -4.67796415e-01 -1.47011101e-01 1.12947953e+00 -1.20690763e+00 1.68570173e+00 1.36352684e-02 4.12913412e-01 3.44422877e-01 -6.07526481e-01 9.16624904e-01 7.72055387e-01 1.42118081e-01 -6.59170568e-01 1.30499348e-01 2.87411094e-01 1.92516029e-01 -8.62267673e-01 6.37779832e-01 3.86484363e-03 -6.10574663e-01 5.15165687e-01 -3.42745930e-02 1.09706089e-01 4.61963177e-01 4.54837292e-01 1.24359632e+00 -4.16402631e-02 5.38181484e-01 -1.38498068e-01 6.75078750e-01 2.99146920e-01 9.73520577e-01 4.06603843e-01 -1.40475631e-01 3.26638907e-01 7.96184778e-01 -4.77357715e-01 -5.85544646e-01 -9.78457391e-01 1.83830392e-02 1.10416460e+00 3.24713826e-01 -7.35479534e-01 -1.11904943e+00 -8.17587554e-01 -2.21235752e-01 9.70500231e-01 -2.83449203e-01 -1.31488413e-01 -1.22886288e+00 -7.71341503e-01 8.25705469e-01 4.36788678e-01 7.10882723e-01 -1.54839563e+00 -6.53669775e-01 4.42571908e-01 -9.62080836e-01 -1.10889363e+00 -3.69821340e-01 1.22245185e-01 -5.71680486e-01 -1.54971659e+00 6.05014749e-02 -9.91233349e-01 1.25865147e-01 -1.45308048e-01 9.12719011e-01 1.35160401e-01 -2.38778759e-02 -3.93215656e-01 -1.73478015e-02 -7.52746686e-02 -2.66494066e-01 2.02021182e-01 1.18339315e-01 -4.23851788e-01 8.77720952e-01 -3.10535729e-01 -3.21346492e-01 2.44900092e-01 -2.93240726e-01 -3.51727679e-02 1.48862451e-01 8.86950076e-01 4.52451915e-01 -1.55567124e-01 7.25581348e-01 -9.83234823e-01 9.40567374e-01 -5.26128709e-01 -3.41008186e-01 1.64470509e-01 -3.67917389e-01 -2.13053644e-01 4.26990032e-01 -5.32426775e-01 -1.40913904e+00 1.19307235e-01 -2.47987118e-02 -9.72450376e-02 -4.71253805e-02 5.45622826e-01 -5.27305722e-01 9.69199359e-01 7.72102296e-01 -5.63028395e-01 -1.49143681e-01 -6.91821814e-01 2.63480932e-01 8.68911684e-01 1.19349670e+00 -6.24169230e-01 3.55844021e-01 -3.67784351e-02 -3.88933569e-01 -2.73208529e-01 -1.51377070e+00 -4.88115311e-01 -6.94438159e-01 1.77515328e-01 1.03258622e+00 -8.42425525e-01 -9.38574791e-01 1.53385729e-01 -1.90458620e+00 -6.45285994e-02 -1.82909131e-01 4.76990342e-01 -4.06895071e-01 3.10232967e-01 -1.25683069e+00 -4.62403834e-01 -8.91652286e-01 -9.95863736e-01 7.46748507e-01 5.68759501e-01 -1.04685724e+00 -5.36143541e-01 -6.93058944e-04 6.81896448e-01 -4.26251218e-02 -1.02599397e-01 1.16550815e+00 -1.33276951e+00 -5.51145449e-02 -1.29235864e-01 -3.74050200e-01 -3.69890302e-01 -1.12992302e-02 -4.45930511e-01 -8.40806186e-01 -2.36557707e-01 1.91552728e-01 -2.80607074e-01 6.06596589e-01 1.17863141e-01 3.63974571e-01 -5.16618252e-01 -6.13443375e-01 2.08048061e-01 6.47012055e-01 2.90103436e-01 6.42388284e-01 7.59918213e-01 2.35518739e-01 8.74403834e-01 1.43987787e+00 4.73122969e-02 4.05789196e-01 1.07919705e+00 3.01338047e-01 2.67211199e-01 -3.22770655e-01 -8.80262777e-02 3.21905077e-01 6.36934876e-01 2.31829658e-01 -7.45015591e-02 -9.09813762e-01 6.08866036e-01 -2.22839785e+00 -1.25723875e+00 -6.70918167e-01 1.87123764e+00 1.33664155e+00 4.11177874e-01 -1.21829227e-01 1.88378304e-01 1.15632248e+00 1.91146489e-02 -2.36675322e-01 -5.27367830e-01 -6.44085407e-02 3.18287015e-02 -1.64133310e-01 1.10726964e+00 -1.44281292e+00 1.59773254e+00 5.37086678e+00 5.94119012e-01 -5.81671298e-01 2.42124468e-01 7.16235712e-02 -1.69462621e-01 4.79017168e-01 3.94303977e-01 -1.32161534e+00 4.35196161e-01 8.83216560e-01 -1.42676651e-01 1.24802612e-01 6.58654273e-01 3.03919124e-03 -4.43994999e-02 -1.11153042e+00 5.68780124e-01 -8.12625885e-02 -1.19738734e+00 1.68024830e-03 -4.32666361e-01 1.34861827e-01 -3.47307712e-01 -6.77430511e-01 5.89776635e-01 1.95762590e-01 -8.30890000e-01 5.06546497e-01 2.95822084e-01 5.46964824e-01 -7.58322179e-01 1.12195385e+00 4.62042928e-01 -9.77524757e-01 -1.59307942e-01 -4.07627463e-01 -2.24789098e-01 5.67705274e-01 2.52007693e-01 -1.09506559e+00 2.93789119e-01 6.02082849e-01 2.70592421e-01 -1.57063887e-01 7.09441006e-01 -8.51878703e-01 4.61026460e-01 -1.45425171e-01 3.00504029e-01 4.24382165e-02 1.23134285e-01 9.53026533e-01 1.66226161e+00 -1.01619311e-01 7.21295297e-01 1.00337930e-01 7.80683577e-01 -2.69945413e-01 -5.92198223e-02 1.43571526e-01 4.81435210e-01 1.23580086e+00 1.48662651e+00 -2.64002115e-01 -5.54337978e-01 3.81282121e-02 8.69156539e-01 7.41051555e-01 -2.82176211e-03 -8.77240300e-01 -7.92368293e-01 4.52078462e-01 3.95678356e-03 -5.39834313e-02 4.47556287e-01 -1.64231524e-01 -5.08218825e-01 -1.01614676e-01 -9.72494960e-01 1.15731776e+00 -6.91338539e-01 -1.23287570e+00 5.94678044e-01 -4.03762087e-02 -6.97150767e-01 -4.72244263e-01 -2.33103320e-01 -1.06557918e+00 1.01698864e+00 -1.20464396e+00 -8.60960305e-01 -3.55045974e-01 4.19390231e-01 7.71551549e-01 -1.81147218e-01 1.17975104e+00 2.99082875e-01 -7.47511387e-01 5.51328123e-01 -6.02831125e-01 4.10963804e-01 1.26905072e+00 -1.22183216e+00 2.74635851e-01 7.06574261e-01 -5.30818522e-01 8.49601746e-01 7.78073430e-01 -9.71174896e-01 -7.09709585e-01 -1.04385328e+00 1.72504163e+00 -2.88985044e-01 7.28240490e-01 1.05086952e-01 -1.27648640e+00 7.32207358e-01 2.25502685e-01 -4.46341842e-01 3.96970898e-01 3.90802175e-01 -3.56638968e-01 2.35890776e-01 -1.17523587e+00 4.61656451e-01 8.58096480e-01 -4.38505322e-01 -1.54776895e+00 5.25738418e-01 9.75741208e-01 -9.04661298e-01 -1.09023488e+00 4.25742537e-01 1.46388590e-01 -4.21149224e-01 9.06822741e-01 -9.79398251e-01 6.82005048e-01 -1.91954032e-01 -9.85829160e-02 -7.58743048e-01 -5.75581312e-01 -7.09353924e-01 -4.99247819e-01 1.70145941e+00 5.75292170e-01 1.88884325e-02 4.58184719e-01 7.13758528e-01 -3.46778303e-01 -4.00663078e-01 -1.11498737e+00 -2.99943358e-01 -5.48355617e-02 1.55463114e-01 4.70474303e-01 1.27624929e+00 8.97370994e-01 1.25859320e+00 5.41685745e-02 3.04749936e-01 3.13426107e-01 4.83484775e-01 5.08242965e-01 -1.56310952e+00 -2.44310983e-02 -4.13977504e-01 4.02291328e-01 -8.78225267e-01 4.52998728e-01 -8.85167658e-01 9.32522789e-02 -1.61608434e+00 2.80314684e-01 -1.60767183e-01 -3.37317809e-02 6.82875276e-01 -6.06652677e-01 -3.44846994e-01 8.33884850e-02 6.34699881e-01 -8.35333347e-01 1.35882542e-01 8.33088875e-01 -7.97393918e-02 -5.21229625e-01 -1.18218385e-01 -5.57317078e-01 1.06096876e+00 8.25747609e-01 -7.01001406e-01 1.63944229e-01 -2.49388754e-01 -1.26260906e-01 4.90351737e-01 -4.93968278e-02 -5.21968186e-01 7.55179107e-01 -1.97932005e-01 4.66782786e-02 -6.84669197e-01 5.81442893e-01 -3.64366472e-01 -2.13544980e-01 5.19509733e-01 -4.85825539e-01 3.63729931e-02 1.81775615e-01 8.31955671e-02 -1.34090587e-01 -6.86662734e-01 7.80148268e-01 -7.54934698e-02 -4.59274888e-01 -3.65487814e-01 -3.96986753e-01 4.30976152e-01 7.96082973e-01 2.77807385e-01 -9.40980017e-01 1.47135044e-02 -9.12199914e-01 4.81579930e-01 -4.85527106e-02 6.83773220e-01 3.51476014e-01 -1.11850286e+00 -9.41876233e-01 -3.85615498e-01 -2.54216075e-01 3.66902530e-01 -5.58177307e-02 8.29668462e-01 -1.63346305e-01 5.04391134e-01 -1.05582640e-01 -2.57549912e-01 -1.65459025e+00 4.13903296e-01 2.73109317e-01 -7.51365006e-01 -8.00432265e-01 7.51813829e-01 -8.31781700e-02 -2.60898858e-01 4.72974330e-01 3.73120755e-01 -8.93836558e-01 3.89968634e-01 8.49534214e-01 4.75345999e-01 2.46443581e-02 -6.62228644e-01 -5.89087188e-01 1.17413729e-01 -6.03092849e-01 -1.41411707e-01 1.33151877e+00 5.43158241e-02 -5.57074964e-01 7.09282607e-02 6.15663111e-01 7.09081218e-02 -8.52543652e-01 -3.19988251e-01 6.73464775e-01 2.41504144e-02 -2.75283486e-01 -1.13321412e+00 -5.78807592e-01 6.35894299e-01 2.00379282e-01 -3.52672525e-02 7.39232540e-01 9.57761854e-02 8.57765079e-01 4.67903852e-01 -1.95894055e-02 -1.20721912e+00 -4.91793975e-02 1.15059400e+00 1.13786948e+00 -8.01396132e-01 9.19516981e-02 -7.41280079e-01 -5.74188471e-01 1.12145150e+00 1.24953008e+00 2.35326178e-02 -6.52071238e-02 5.23109436e-01 -1.05229986e-03 -5.43690085e-01 -9.48339820e-01 2.52318501e-01 5.96168637e-02 1.67366073e-01 7.83168495e-01 -1.47079714e-02 -1.05326068e+00 1.59291983e+00 -3.95980448e-01 -5.00189602e-01 2.80551314e-01 6.60332680e-01 -6.87489271e-01 -9.14799154e-01 -4.82473969e-01 -5.19342981e-02 -7.01656103e-01 -6.88621625e-02 -8.36313665e-01 7.74762690e-01 -6.43214583e-02 1.25113165e+00 3.63371253e-01 -2.41535679e-01 7.84772813e-01 4.63059634e-01 1.10933699e-01 -7.40175962e-01 -9.66633379e-01 -2.96906866e-02 8.16068828e-01 -5.84787309e-01 -1.46860018e-01 -6.37150705e-01 -2.09155655e+00 -2.61571676e-01 -6.00644469e-01 7.63541639e-01 2.05604866e-01 1.18200588e+00 3.65126163e-01 5.47063589e-01 1.60121933e-01 -5.10306954e-01 -6.83229148e-01 -1.45118213e+00 -8.04967713e-03 7.57005751e-01 -2.11879387e-01 -7.13432610e-01 -2.80614913e-01 -1.33172676e-01]
[9.341130256652832, 9.53217887878418]
df280b00-1b36-4990-b6cf-767137bd03bc
int-fp-qsim-mixed-precision-and-formats-for
2307.03712
null
https://arxiv.org/abs/2307.03712v1
https://arxiv.org/pdf/2307.03712v1.pdf
INT-FP-QSim: Mixed Precision and Formats For Large Language Models and Vision Transformers
The recent rise of large language models (LLMs) has resulted in increased efforts towards running LLMs at reduced precision. Running LLMs at lower precision supports resource constraints and furthers their democratization, enabling users to run billion-parameter LLMs on their personal devices. To supplement this ongoing effort, we propose INT-FP-QSim: an open-source simulator that enables flexible evaluation of LLMs and vision transformers at various numerical precisions and formats. INT-FP-QSim leverages existing open-source repositories such as TensorRT, QPytorch and AIMET for a combined simulator that supports various floating point and integer formats. With the help of our simulator, we survey the impact of different numerical formats on the performance of LLMs and vision transformers at 4-bit weights and 4-bit or 8-bit activations. We also compare recently proposed methods like Adaptive Block Floating Point, SmoothQuant, GPTQ and RPTQ on the model performances. We hope INT-FP-QSim will enable researchers to flexibly simulate models at various precisions to support further research in quantization of LLMs and vision transformers.
['Darius Bunandar', 'Ayon Basumallik', 'Craig Chan', 'Arulselvan Madhavan', 'Mikhail Bernadskiy', 'Lakshmi Nair']
2023-07-07
null
null
null
null
['quantization']
['methodology']
[-4.09263045e-01 -7.29153991e-01 -5.12445867e-01 -3.09561193e-01 -8.95520687e-01 -5.81106365e-01 5.11811495e-01 2.24870786e-01 -4.98408943e-01 1.65260673e-01 2.87305295e-01 -8.83591413e-01 2.33348355e-01 -8.49290073e-01 -4.82543468e-01 -2.54214872e-02 -4.15106773e-01 8.94522741e-02 3.71280164e-01 -4.93326604e-01 4.97658491e-01 4.95409250e-01 -1.50201535e+00 7.38655508e-01 5.98527789e-01 1.13393605e+00 9.51220840e-02 1.03536272e+00 -2.50403136e-01 9.73196864e-01 -5.51704645e-01 -6.32243276e-01 4.88236129e-01 3.58360201e-01 -6.06533170e-01 -6.85259283e-01 4.61963743e-01 -5.67404628e-01 -2.49596566e-01 8.65100145e-01 7.02006400e-01 -3.86186659e-01 1.82901144e-01 -1.22127616e+00 -3.49903136e-01 7.95472562e-01 -5.11321008e-01 4.81933206e-01 3.32451046e-01 8.13360453e-01 9.66723442e-01 -8.36442530e-01 1.56386673e-01 1.71896148e+00 9.27728295e-01 -1.76876962e-01 -1.20537198e+00 -8.56543779e-01 -3.84217024e-01 1.40202746e-01 -2.02100205e+00 -7.75113702e-01 -1.74908757e-01 -2.26509422e-02 1.59047127e+00 3.92530113e-01 6.45175397e-01 6.51746809e-01 7.86684334e-01 3.61305535e-01 1.15484297e+00 -3.89559537e-01 3.86295587e-01 -3.06983292e-01 1.74901083e-01 8.56335938e-01 1.33567452e-01 1.09169647e-01 -9.28802073e-01 -5.24630427e-01 8.64365339e-01 -5.97722530e-01 1.15360193e-01 2.44252130e-01 -1.35464990e+00 8.14746618e-01 1.05940267e-01 1.32865191e-01 -1.57047451e-01 9.93011594e-01 8.16276908e-01 3.90699387e-01 2.49091703e-02 2.98425257e-01 -5.31046450e-01 -7.56251752e-01 -1.32359147e+00 4.66171682e-01 8.59388590e-01 1.07849514e+00 5.53656399e-01 2.73680896e-01 -5.02280653e-01 5.35513759e-01 6.24139607e-01 7.51489282e-01 5.55226088e-01 -1.35101974e+00 6.96743369e-01 2.41777003e-01 -3.25547419e-02 -7.28399217e-01 -4.84689355e-01 -4.72252607e-01 -8.10532689e-01 1.54055744e-01 8.49574581e-02 1.49022743e-01 -7.50612259e-01 1.16750801e+00 3.37303951e-02 8.21035653e-02 -4.14704420e-02 6.54102683e-01 4.96863961e-01 9.11569178e-01 1.28498867e-01 2.06026927e-01 1.47819591e+00 -7.24372506e-01 -5.16102135e-01 6.50110319e-02 1.14807475e+00 -1.18756688e+00 1.54166162e+00 8.49752307e-01 -1.37870038e+00 -4.12032783e-01 -1.37444508e+00 -4.33066756e-01 -1.76501483e-01 1.08512744e-01 8.22116196e-01 1.04802012e+00 -1.53170943e+00 7.78880358e-01 -1.20590413e+00 3.41321737e-01 2.62186795e-01 6.73628926e-01 3.28707904e-01 1.62767321e-01 -1.27687716e+00 8.89311135e-01 9.31296423e-02 -1.13392420e-01 -6.24641478e-01 -1.01277244e+00 -5.93909979e-01 1.25235632e-01 -1.39523530e-02 -6.47825301e-01 1.56392789e+00 -4.63607579e-01 -1.66104150e+00 4.83934671e-01 -1.13793783e-01 -9.72430646e-01 1.98652178e-01 -5.71143348e-03 -7.94236064e-01 -2.15967491e-01 -1.37530982e-01 6.99970305e-01 5.53677738e-01 -2.58177370e-01 -2.48212919e-01 -1.23474121e-01 6.05065599e-02 4.03192081e-02 -3.05389732e-01 1.14182308e-01 -4.89604801e-01 -5.81675231e-01 -3.28476548e-01 -6.48042738e-01 -3.29685599e-01 1.48216337e-01 -1.15984268e-01 1.36330768e-01 4.62371588e-01 -2.67654002e-01 1.68987405e+00 -2.07064509e+00 -3.28320920e-01 3.07453513e-01 1.53911412e-01 5.51707685e-01 -1.45119935e-01 6.61181450e-01 3.82297933e-01 2.85124898e-01 2.29037732e-01 -3.11711043e-01 3.43159258e-01 2.85102159e-01 -2.82111257e-01 3.54653955e-01 -1.57796353e-01 7.92273462e-01 -6.17092788e-01 -5.57272792e-01 2.65416384e-01 5.27580678e-01 -9.58996832e-01 -3.47229213e-01 -2.39499480e-01 -2.40322575e-01 -1.94359183e-01 6.79030538e-01 8.25415432e-01 -5.14517307e-01 4.61933799e-02 -5.35290420e-01 -6.53748333e-01 8.15614104e-01 -1.29659128e+00 1.84483266e+00 -7.52733767e-01 5.86633801e-01 2.44817793e-01 -1.61172822e-01 5.81342697e-01 6.41096607e-02 -4.01578844e-03 -1.13510215e+00 -3.35559621e-02 5.16783357e-01 -1.15707619e-02 4.38340530e-02 1.27444220e+00 1.37293711e-01 -2.74302065e-01 1.90392956e-01 -1.92225307e-01 -6.39962375e-01 4.41056520e-01 4.18988109e-01 1.05858099e+00 -2.32607588e-01 8.66866484e-02 -5.82314909e-01 4.47828025e-01 -2.74486721e-01 1.14506096e-01 6.25279546e-01 -1.04401611e-01 1.56254604e-01 4.06780481e-01 -3.22551697e-01 -1.22885418e+00 -1.28198373e+00 -5.01205385e-01 1.25374162e+00 -3.12860072e-01 -1.44505966e+00 -6.41070604e-01 3.04307997e-01 1.80016197e-02 7.62963593e-01 6.94520399e-02 7.94544145e-02 -2.84562200e-01 -7.97267497e-01 1.28175223e+00 3.35673600e-01 8.22048962e-01 -3.96483898e-01 -9.83395875e-01 2.89530367e-01 3.78064871e-01 -1.11357808e+00 -6.06494308e-01 1.81807280e-01 -9.04326737e-01 -3.33444744e-01 -4.00172025e-01 -2.25486577e-01 -2.13026300e-01 -1.81049466e-01 1.40786600e+00 3.68479602e-02 -8.64530504e-02 3.82043868e-02 -8.81884694e-02 -1.77371427e-01 -6.66554630e-01 1.17951587e-01 6.87855780e-02 -5.62647283e-01 -3.61464452e-03 -5.14782012e-01 -6.86139762e-01 -8.32287245e-04 -9.88978624e-01 1.84020698e-01 2.35118836e-01 3.70847702e-01 6.16166413e-01 -2.32112944e-01 8.63073096e-02 -4.48095173e-01 9.30713117e-01 -1.90621838e-01 -9.98807132e-01 -3.92562747e-02 -6.49505079e-01 6.37114167e-01 7.10545480e-01 -1.15764268e-01 -3.57276291e-01 -3.45993042e-01 -6.88137174e-01 -3.38276654e-01 8.72878253e-01 5.45041621e-01 2.92414814e-01 -2.84921527e-01 7.70865917e-01 6.44855797e-02 -2.23185331e-01 -7.98754618e-02 6.39374197e-01 8.20418298e-01 4.28208679e-01 -6.84775352e-01 1.79219425e-01 1.95897788e-01 4.19869386e-02 -6.55691564e-01 -1.22278586e-01 -2.31905859e-02 3.78423221e-02 2.86964238e-01 3.10301244e-01 -1.24126625e+00 -8.86202812e-01 5.74080706e-01 -1.02206326e+00 -5.48125386e-01 -1.95259616e-01 1.92811683e-01 -4.34566349e-01 5.04809797e-01 -9.69038069e-01 -4.93156850e-01 -9.09597099e-01 -1.83092785e+00 1.37740445e+00 -5.46052866e-02 -4.16219205e-01 -8.77258182e-01 -2.07964271e-01 -4.25246842e-02 8.71258616e-01 -3.49890232e-01 7.08156586e-01 2.97017712e-02 -7.59823442e-01 1.87472984e-01 -2.26215079e-01 4.23744559e-01 -3.33698303e-01 3.11842442e-01 -6.77960277e-01 -3.85047972e-01 -2.51227289e-01 -4.03317600e-01 3.82652581e-01 2.76790082e-01 1.19039059e+00 -3.68250012e-01 2.00656001e-02 1.03098893e+00 1.44177115e+00 -3.36432517e-01 8.80195916e-01 1.51184648e-01 3.14486623e-01 -5.17749786e-01 4.12179530e-01 9.52489436e-01 6.36000276e-01 7.97205925e-01 2.58152097e-01 1.12167373e-01 -1.31789297e-01 5.77314710e-03 6.99735880e-01 1.33202910e+00 2.07966030e-01 -5.76218739e-02 -1.03422868e+00 4.82576750e-02 -1.08263063e+00 -5.11311352e-01 -3.49585593e-01 2.35907173e+00 1.16302037e+00 5.01131952e-01 -1.01490237e-01 3.05365562e-01 -1.35219460e-02 1.87356636e-01 -4.28755254e-01 -1.31897235e+00 4.88819852e-02 9.87847328e-01 1.34668493e+00 5.93727171e-01 -5.07478893e-01 1.09213209e+00 7.37179708e+00 1.40328276e+00 -1.47049725e+00 2.67813206e-01 7.24116385e-01 -3.21997792e-01 -3.41682732e-01 6.19736016e-02 -1.14994752e+00 6.49306297e-01 1.99510896e+00 -3.59344006e-01 6.37154996e-01 5.45925438e-01 4.33679819e-01 -2.35778898e-01 -9.58153546e-01 1.10536873e+00 -5.40292501e-01 -1.72395003e+00 2.32649207e-01 1.03605911e-01 4.22155976e-01 6.25267088e-01 4.44721431e-01 3.92295927e-01 3.87448370e-01 -1.20992362e+00 9.72955108e-01 2.79402524e-01 1.41025472e+00 -8.65487456e-01 2.30847940e-01 2.88571082e-02 -1.49259329e+00 2.62974035e-02 -3.96913081e-01 -3.87854695e-01 5.56176677e-02 5.57220638e-01 -8.89758348e-01 9.62903798e-02 5.10876894e-01 3.36419016e-01 -6.75642669e-01 7.18156815e-01 3.19718450e-01 8.30642939e-01 -7.32049167e-01 -1.21732034e-01 4.22150046e-01 2.17695817e-01 -1.61952153e-02 1.42769814e+00 5.26010513e-01 -1.34752899e-01 4.26483080e-02 7.96733499e-01 1.64233431e-01 7.85371885e-02 5.04163019e-02 -1.33988783e-01 9.37041700e-01 1.00428879e+00 -5.01405120e-01 -5.80463469e-01 -4.28145647e-01 7.69781768e-01 -9.04034916e-03 -8.76290426e-02 -1.06731105e+00 -3.13465923e-01 8.31165373e-01 3.67227465e-01 1.51433393e-01 -9.28176403e-01 -5.25876641e-01 -1.19210553e+00 -3.78376156e-01 -1.35045648e+00 1.20810792e-01 -8.94048929e-01 -5.48789024e-01 3.44631821e-01 2.33108960e-02 -8.46312106e-01 -2.77424127e-01 -6.57157838e-01 -1.36246502e-01 1.12891066e+00 -1.35454977e+00 -7.37074554e-01 1.28825277e-01 6.29522979e-01 3.48140895e-01 9.64333210e-03 9.90587294e-01 5.65066159e-01 -2.62091190e-01 1.03139234e+00 2.71020919e-01 -3.16556960e-01 4.11589414e-01 -7.99388468e-01 1.20686746e+00 4.12276328e-01 2.69718338e-02 9.05035496e-01 6.73866570e-01 -3.63059580e-01 -2.12220073e+00 -9.35124397e-01 4.91066933e-01 -1.30039647e-01 9.34782505e-01 -4.16779280e-01 -5.49084067e-01 5.12540102e-01 4.96580005e-02 2.76133984e-01 4.85411942e-01 -3.05297732e-01 -4.36342180e-01 -2.32655689e-01 -1.03446245e+00 7.26461351e-01 6.23954892e-01 -6.69363856e-01 1.60618678e-01 2.42432132e-01 8.46744359e-01 -8.45192492e-01 -1.31436753e+00 1.00199565e-01 6.10841572e-01 -1.16990280e+00 1.21703601e+00 1.49517536e-01 1.82599485e-01 -1.67524263e-01 -7.60843992e-01 -9.42058921e-01 -5.44224493e-02 -9.09814060e-01 -2.31335476e-01 7.46293485e-01 1.94669411e-01 -6.64092541e-01 4.31853145e-01 3.16693246e-01 -1.60689175e-01 -8.16016316e-01 -1.15497923e+00 -6.81843758e-01 3.80816430e-01 -1.01650405e+00 9.65730429e-01 2.53677666e-01 -8.15869793e-02 2.56810278e-01 -3.37202102e-02 -7.87856337e-03 3.07488143e-01 -3.03877562e-01 6.16124928e-01 -4.40297812e-01 -6.92698359e-01 -4.44732189e-01 -5.92610598e-01 -1.27827775e+00 -3.71211529e-01 -1.13556898e+00 -5.89140773e-01 -1.02355480e+00 -2.50550658e-01 -8.87096763e-01 6.12051114e-02 3.89341295e-01 5.45760691e-01 5.38697720e-01 3.60029012e-01 1.76095977e-01 -7.21963227e-01 4.58612412e-01 1.16952980e+00 7.73743913e-02 -4.53220978e-02 -4.03177381e-01 -5.17897427e-01 6.49638832e-01 5.33317804e-01 -8.57007354e-02 -2.57232904e-01 -5.26372433e-01 6.53220713e-01 2.35164464e-01 3.57428938e-01 -1.41429210e+00 1.85466498e-01 1.26175076e-01 2.65523940e-01 -6.33410454e-01 6.07223392e-01 -1.73019350e-01 1.43332988e-01 7.22801745e-01 -1.23524584e-01 7.05680907e-01 5.85018158e-01 -2.04236209e-01 1.71709031e-01 -1.31410643e-01 7.29555607e-01 -9.38446224e-02 -7.61346817e-01 3.70317996e-02 -5.74574232e-01 2.21780494e-01 5.02864361e-01 1.77732073e-02 -3.50500911e-01 -2.62256503e-01 -3.91915828e-01 2.60957688e-01 5.47897637e-01 1.12573251e-01 4.98068839e-01 -1.29680550e+00 -5.30128598e-01 3.94718558e-01 -2.11060688e-01 -2.85917908e-01 -3.30650471e-02 7.15119541e-01 -1.19110620e+00 8.04023623e-01 -1.42636308e-02 -7.10477471e-01 -1.14589548e+00 1.70806572e-01 3.87737602e-01 -5.88177562e-01 -3.35107476e-01 7.75644422e-01 -3.31001252e-01 -1.60166726e-01 3.77881527e-02 -1.22888374e+00 6.10085666e-01 -3.92864704e-01 7.25614488e-01 4.92892057e-01 5.18086016e-01 -5.34932196e-01 -3.65056962e-01 2.84444302e-01 1.45363837e-01 -4.17665780e-01 8.88490677e-01 2.22115591e-02 -1.61565572e-01 4.96537179e-01 1.32865655e+00 1.37196288e-01 -9.00338054e-01 9.04382914e-02 -1.05200924e-01 -1.36175856e-01 4.85614061e-01 -7.31235802e-01 -8.61087680e-01 9.11206126e-01 9.23728406e-01 -2.77741611e-01 1.10720122e+00 -5.45274556e-01 1.15954113e+00 2.81194985e-01 9.61232901e-01 -8.72220039e-01 -1.62529171e-01 8.00126791e-01 5.33516347e-01 -5.02983451e-01 4.09166873e-01 -2.07836524e-01 -2.02391848e-01 1.10628748e+00 6.98197260e-02 -1.26435295e-01 6.90189838e-01 1.07231915e+00 -1.02288604e-01 5.48785813e-02 -1.10364282e+00 2.92236894e-01 4.02540304e-02 1.43312335e-01 9.98529553e-01 2.14119256e-01 -4.60532129e-01 3.70982915e-01 -7.65996695e-01 4.74368542e-01 7.81936467e-01 9.60082889e-01 -3.54498029e-01 -1.60126436e+00 -6.97789133e-01 6.03467166e-01 -8.17275286e-01 -8.42251122e-01 3.69799465e-01 4.60171402e-01 5.56488633e-02 8.88885140e-01 2.27167159e-01 -4.25691217e-01 -3.49069908e-02 -3.03432584e-01 7.75091112e-01 -4.92677689e-01 -1.12911534e+00 -7.43330568e-02 2.15393364e-01 -9.17127490e-01 1.11612260e-01 -4.08957481e-01 -1.69791877e+00 -9.40482795e-01 -1.06181078e-01 -3.40430662e-02 1.00769925e+00 3.68546575e-01 7.97186971e-01 4.51628774e-01 2.19520181e-02 -7.78078198e-01 -1.14877212e+00 -6.82039976e-01 -4.48537916e-01 -5.19333422e-01 2.09334701e-01 -1.93425760e-01 6.36863336e-02 -2.94657797e-01]
[8.642549514770508, 3.4407434463500977]
3701846f-37d1-4960-a3a5-538f88185986
acorn-adaptive-coordinate-networks-for-neural
2105.02788
null
https://arxiv.org/abs/2105.02788v1
https://arxiv.org/pdf/2105.02788v1.pdf
ACORN: Adaptive Coordinate Networks for Neural Scene Representation
Neural representations have emerged as a new paradigm for applications in rendering, imaging, geometric modeling, and simulation. Compared to traditional representations such as meshes, point clouds, or volumes they can be flexibly incorporated into differentiable learning-based pipelines. While recent improvements to neural representations now make it possible to represent signals with fine details at moderate resolutions (e.g., for images and 3D shapes), adequately representing large-scale or complex scenes has proven a challenge. Current neural representations fail to accurately represent images at resolutions greater than a megapixel or 3D scenes with more than a few hundred thousand polygons. Here, we introduce a new hybrid implicit-explicit network architecture and training strategy that adaptively allocates resources during training and inference based on the local complexity of a signal of interest. Our approach uses a multiscale block-coordinate decomposition, similar to a quadtree or octree, that is optimized during training. The network architecture operates in two stages: using the bulk of the network parameters, a coordinate encoder generates a feature grid in a single forward pass. Then, hundreds or thousands of samples within each block can be efficiently evaluated using a lightweight feature decoder. With this hybrid implicit-explicit network architecture, we demonstrate the first experiments that fit gigapixel images to nearly 40 dB peak signal-to-noise ratio. Notably this represents an increase in scale of over 1000x compared to the resolution of previously demonstrated image-fitting experiments. Moreover, our approach is able to represent 3D shapes significantly faster and better than previous techniques; it reduces training times from days to hours or minutes and memory requirements by over an order of magnitude.
['Gordon Wetzstein', 'Marco Monteiro', 'Eric R. Chan', 'Connor Z. Lin', 'David B. Lindell', 'Julien N. P. Martel']
2021-05-06
null
null
null
null
['3d-shape-representation']
['computer-vision']
[ 3.08099002e-01 3.46580185e-02 3.55172962e-01 -2.91364014e-01 -9.34246182e-01 -1.72374517e-01 7.38177061e-01 2.52492756e-01 -2.98264176e-01 6.58676147e-01 -1.82102650e-01 -2.71654278e-01 -1.63937174e-02 -1.03064072e+00 -8.88002932e-01 -4.90655810e-01 -4.19451505e-01 6.35771573e-01 2.83726335e-01 -3.37439366e-02 6.31298125e-02 1.13912845e+00 -1.79080093e+00 3.77236396e-01 4.99270588e-01 1.39312088e+00 6.74293935e-02 8.71644080e-01 -2.69342512e-01 5.69226801e-01 -4.99721617e-01 1.61810830e-01 2.62414932e-01 1.70781553e-01 -5.35754442e-01 -3.89626510e-02 8.15550804e-01 -4.70586687e-01 -2.21360162e-01 5.85077882e-01 7.04934955e-01 4.13268507e-02 5.00960886e-01 -7.10987449e-01 -2.42778495e-01 1.16490178e-01 -5.95565259e-01 -6.18621893e-02 1.35393232e-01 2.87367821e-01 5.72108686e-01 -9.74034369e-01 4.90169346e-01 1.40082479e+00 1.08320940e+00 3.07144642e-01 -1.82503939e+00 -4.88788933e-01 8.69500414e-02 -2.76833415e-01 -1.35620964e+00 -3.96979928e-01 5.62020838e-01 -4.24447209e-01 1.24332023e+00 3.37982684e-01 8.35725188e-01 6.61269248e-01 1.27071455e-01 2.89862245e-01 1.06739235e+00 -1.77974775e-01 1.94445938e-01 -1.05780922e-02 -1.12534188e-01 7.47146428e-01 -1.19809322e-02 1.33352831e-01 -4.11362439e-01 -3.91107500e-01 1.44411266e+00 -2.21410662e-01 -1.99042901e-01 -2.43501961e-01 -1.27065170e+00 8.64536166e-01 7.40562975e-01 9.69772860e-02 -5.22552371e-01 8.63504171e-01 5.01528800e-01 2.10752860e-01 6.51379824e-01 4.47560161e-01 -4.23841804e-01 -3.38859558e-02 -1.33031237e+00 3.99373412e-01 7.08442390e-01 8.09172094e-01 9.81521308e-01 4.47968096e-01 1.33822814e-01 6.95762992e-01 5.56706861e-02 4.28066730e-01 1.29644454e-01 -1.17366266e+00 7.04183206e-02 4.08231884e-01 4.23122458e-02 -1.24263978e+00 -6.50206029e-01 -6.35247111e-01 -1.16978550e+00 7.75789261e-01 2.69205987e-01 -4.10289876e-03 -9.53688741e-01 1.33229554e+00 5.08647978e-01 3.74754131e-01 -4.88252878e-01 8.35883975e-01 7.15337038e-01 8.71729434e-01 -1.45292968e-01 4.90471311e-02 1.46651328e+00 -4.50103939e-01 -1.15334719e-01 -3.62869054e-02 3.35758477e-01 -7.17854917e-01 9.60188210e-01 4.97243732e-01 -1.40291917e+00 -6.20132744e-01 -1.13737822e+00 -2.74380893e-01 -2.94661582e-01 -1.14122912e-01 8.26206088e-01 4.43368673e-01 -1.41863775e+00 9.24302101e-01 -1.03716040e+00 1.32933753e-02 5.56651354e-01 6.04744673e-01 -1.43190399e-01 8.07532594e-02 -7.09750414e-01 5.67592263e-01 2.07943078e-02 -7.65978219e-03 -7.18324542e-01 -1.22883284e+00 -8.29496086e-01 2.85646498e-01 -1.20341606e-01 -1.05912459e+00 9.52010572e-01 -9.00387108e-01 -1.54554892e+00 7.57969439e-01 -1.30363539e-01 -7.26218641e-01 4.67556417e-01 -3.91920246e-02 -4.89138700e-02 2.11000279e-01 -2.48094067e-01 8.38267267e-01 9.18354154e-01 -1.17246974e+00 -2.72787333e-01 -2.24190533e-01 1.21976353e-01 -9.65878088e-03 7.64586627e-02 1.30906776e-02 -4.15955693e-01 -7.36432791e-01 3.31262410e-01 -5.94931304e-01 -6.30621552e-01 7.16348588e-01 -7.08271414e-02 2.43672192e-01 6.89627945e-01 -4.81709957e-01 6.51590228e-01 -2.09029961e+00 -1.47013828e-01 3.75830203e-01 3.84797454e-01 8.29230919e-02 -1.23237148e-01 2.52260536e-01 9.79283941e-04 1.46090090e-01 -3.74395281e-01 -5.31305432e-01 -6.35373965e-02 9.80326831e-02 -1.95918173e-01 4.77522582e-01 4.22483295e-01 6.95338905e-01 -5.79206228e-01 -3.34440082e-01 3.02378923e-01 1.08549905e+00 -7.45215178e-01 -9.68315750e-02 -3.03154111e-01 3.27813178e-01 -3.79345179e-01 4.65868920e-01 7.96193838e-01 -5.38361132e-01 2.44273711e-02 -3.67463380e-01 -3.55340153e-01 4.70399439e-01 -1.34930384e+00 1.74622548e+00 -8.20298374e-01 9.12329137e-01 3.92089844e-01 -1.01065397e+00 9.18691099e-01 3.38237405e-01 5.39516270e-01 -7.26085007e-01 -6.41603619e-02 2.75149792e-01 -3.53853762e-01 1.39364704e-01 4.74182278e-01 -1.10693932e-01 1.15169376e-01 2.37164751e-01 -7.34757781e-02 -5.86870134e-01 -8.94148573e-02 -1.20468870e-01 1.02796185e+00 2.04858169e-01 1.54015347e-01 -2.66382992e-01 3.24238807e-01 8.92441422e-02 2.77769446e-01 6.76533341e-01 4.45545793e-01 8.35426033e-01 4.84762281e-01 -9.97867942e-01 -1.21646214e+00 -1.00119555e+00 -4.32348341e-01 8.13803673e-01 -3.42471629e-01 -3.64926189e-01 -6.43363595e-01 2.31685817e-01 2.12394167e-02 3.54923815e-01 -5.58033526e-01 3.15042377e-01 -8.84994328e-01 -6.49290144e-01 5.45729697e-01 5.63450813e-01 3.77259970e-01 -8.65540862e-01 -9.69456911e-01 4.95201021e-01 5.83932996e-01 -1.06170213e+00 8.52015316e-02 2.65178740e-01 -1.26007462e+00 -8.20471227e-01 -6.43205106e-01 -4.90851045e-01 6.43334448e-01 -6.86133187e-03 1.50509667e+00 1.93040028e-01 -6.49359226e-01 1.70016423e-01 2.81219572e-01 -1.99494198e-01 -2.29508892e-01 -1.30657822e-01 -9.56658423e-02 -2.22939670e-01 -3.14857960e-01 -1.16217768e+00 -7.34892607e-01 -4.84344028e-02 -7.36003995e-01 3.74118388e-01 4.69790190e-01 7.74213910e-01 1.04154694e+00 -3.46361667e-01 3.28347296e-01 -8.26696098e-01 4.36746091e-01 -2.50920922e-01 -1.01721811e+00 -2.39919662e-01 -3.12254667e-01 1.51882946e-01 7.74981260e-01 -4.30228382e-01 -5.95626295e-01 2.65845001e-01 -3.60800803e-01 -5.12907803e-01 -1.54366001e-01 5.35592556e-01 3.86629373e-01 -5.06234288e-01 7.13574171e-01 1.03584729e-01 6.46878779e-02 -4.02861565e-01 3.94136667e-01 1.83079988e-01 4.92558867e-01 -7.11169720e-01 6.51729584e-01 6.04631543e-01 3.33860070e-01 -1.09253037e+00 -1.81698397e-01 3.08479997e-03 -6.63688064e-01 -1.23159550e-01 6.34194255e-01 -9.28016663e-01 -8.64375710e-01 3.21988374e-01 -1.25853038e+00 -5.45006514e-01 -5.20051539e-01 3.57541353e-01 -6.96448088e-01 1.01148272e-02 -6.38922334e-01 -5.66257179e-01 -4.04427052e-01 -1.10181165e+00 1.29091775e+00 -6.50779232e-02 -2.09628686e-01 -9.17418301e-01 -2.42928162e-01 -1.71365976e-01 8.86460483e-01 8.13219190e-01 1.10957658e+00 2.02888828e-02 -9.80795324e-01 -1.60210431e-01 -5.18163562e-01 1.01949252e-01 -1.50372222e-01 1.31012037e-01 -1.04577696e+00 -2.52790779e-01 5.84052119e-04 -4.69258666e-01 6.52033150e-01 5.10801733e-01 1.55918586e+00 -2.60475665e-01 -2.28076354e-01 1.03647876e+00 1.57272017e+00 -1.59144074e-01 6.29866242e-01 1.77775458e-01 5.01563191e-01 3.20722312e-01 2.00129356e-02 5.38744688e-01 1.17082596e-01 6.96108997e-01 5.07657230e-01 -4.11944121e-01 -2.50746518e-01 1.54839605e-01 -1.54574692e-01 7.44362593e-01 -2.42161959e-01 2.65941292e-01 -9.73171890e-01 2.64236361e-01 -1.29654396e+00 -8.52520525e-01 -1.03811577e-01 2.38162994e+00 7.21352160e-01 2.79039383e-01 -1.52901441e-01 1.64660379e-01 2.01927766e-01 2.43736267e-01 -6.00303352e-01 -6.29380465e-01 -1.06293252e-02 6.56264007e-01 5.17611861e-01 5.02660334e-01 -8.39235425e-01 5.81827819e-01 6.60161972e+00 7.03474104e-01 -1.67997336e+00 -2.45647319e-02 8.28537166e-01 -3.11420202e-01 -4.20693427e-01 -2.68529564e-01 -5.21961987e-01 1.42080784e-01 1.14955556e+00 4.78898212e-02 5.68370342e-01 8.48082960e-01 2.43132561e-01 4.84168194e-02 -1.08280206e+00 1.20302844e+00 -1.91129714e-01 -1.85588872e+00 1.50604665e-01 1.48436785e-01 4.86265242e-01 4.46506590e-01 1.15574419e-01 1.30043045e-01 8.76087695e-02 -1.57959509e+00 7.41878033e-01 5.72283566e-01 1.24304163e+00 -8.32549870e-01 1.31952688e-01 5.60247540e-01 -1.26586044e+00 2.49617875e-01 -6.19062304e-01 -9.24406052e-02 1.37539044e-01 7.58149862e-01 -7.64205813e-01 2.71283627e-01 7.18000650e-01 4.19720054e-01 -2.91573375e-01 1.01821625e+00 2.76104301e-01 5.31931400e-01 -7.91184187e-01 8.61819088e-02 3.81408781e-01 -7.86441565e-02 4.68163103e-01 1.28477824e+00 3.79857212e-01 1.75013795e-01 1.07439190e-01 1.04643393e+00 -5.41239902e-02 -1.23045504e-01 -4.93885607e-01 2.68705636e-01 3.97245169e-01 1.19604981e+00 -6.49241030e-01 -4.22233313e-01 -3.97714883e-01 7.14806676e-01 4.73396808e-01 3.19518775e-01 -7.36436248e-01 -3.97142917e-01 7.25319505e-01 5.25383115e-01 5.00717700e-01 -5.77704847e-01 -4.03734267e-01 -6.95638776e-01 -4.23955619e-02 -8.82995963e-01 -2.68191934e-01 -8.78553450e-01 -8.68571401e-01 9.08286870e-01 -9.20941979e-02 -1.07844484e+00 -4.55446839e-01 -7.59371698e-01 -4.65036988e-01 1.10709572e+00 -1.48634851e+00 -1.03298366e+00 -4.94646877e-01 3.40544105e-01 3.07313442e-01 2.22504273e-01 1.14727926e+00 3.00362080e-01 2.27815993e-02 1.93452045e-01 1.19569547e-01 4.56273817e-02 3.16837686e-03 -1.06056011e+00 8.56162488e-01 3.19573104e-01 1.04615167e-01 5.16509712e-01 4.95442748e-01 -2.01681137e-01 -1.52193904e+00 -9.57749724e-01 3.80094796e-01 5.74858710e-02 4.05109644e-01 -5.74654758e-01 -1.05760658e+00 5.10910988e-01 5.63238896e-02 5.68049908e-01 2.72674978e-01 8.23824033e-02 -3.95522326e-01 -1.93367243e-01 -1.15916681e+00 3.38281453e-01 8.43984067e-01 -4.95100081e-01 1.93624347e-02 3.15068722e-01 6.88372374e-01 -8.25385869e-01 -1.02276635e+00 4.04379785e-01 5.80805421e-01 -1.12654936e+00 1.27176940e+00 -2.54659981e-01 2.86713779e-01 -3.54440331e-01 -2.01599821e-01 -1.11333060e+00 -4.15835112e-01 -6.48075581e-01 -2.13539362e-01 7.19496965e-01 2.61841327e-01 -6.22627616e-01 7.82194078e-01 3.65766078e-01 -2.70864576e-01 -1.19914389e+00 -1.17272568e+00 -3.25534642e-01 7.46354535e-02 -5.59879541e-01 6.95135117e-01 7.29862392e-01 -6.91963077e-01 2.27841921e-02 2.95925494e-02 3.33213508e-01 7.30392754e-01 1.50495321e-01 7.17267692e-01 -1.41089523e+00 -3.38128567e-01 -5.60662687e-01 -6.71490371e-01 -1.34084296e+00 -1.43155172e-01 -7.84926891e-01 -1.79328963e-01 -1.69909894e+00 -3.81023824e-01 -9.54224586e-01 9.96296480e-02 2.44175375e-01 3.64930123e-01 4.84390765e-01 7.32230544e-02 -8.90131816e-02 4.13135476e-02 3.09113353e-01 1.10668504e+00 -1.50483996e-01 -1.48039572e-02 -1.39499918e-01 -2.92872757e-01 8.47802341e-01 6.27046287e-01 -2.55124629e-01 -3.12993586e-01 -8.92526507e-01 3.59798342e-01 2.76702613e-01 7.91892886e-01 -1.44196010e+00 2.07810313e-01 1.38990164e-01 7.97446847e-01 -7.65153348e-01 8.55863452e-01 -7.94832110e-01 5.03363907e-01 2.83920258e-01 -2.00786993e-01 5.67776971e-02 6.31054223e-01 2.82749861e-01 -1.74121827e-01 7.83897117e-02 9.27107453e-01 -2.95890808e-01 -2.46697128e-01 4.44865912e-01 -3.89539242e-01 -1.34501889e-01 6.13986254e-01 -4.83764082e-01 8.57058764e-02 -2.84898341e-01 -6.67591333e-01 -2.07412094e-02 4.78144705e-01 -9.56703871e-02 7.45913744e-01 -1.28308737e+00 -7.84871757e-01 4.24906760e-01 -4.71425861e-01 5.46839058e-01 2.78316915e-01 6.21183753e-01 -1.03129578e+00 3.64074349e-01 -1.56576753e-01 -9.31294322e-01 -1.08550632e+00 1.38488382e-01 7.26283252e-01 -2.65964687e-01 -9.42539752e-01 8.20007682e-01 2.08581045e-01 -3.76737773e-01 1.75375775e-01 -8.32748592e-01 5.96513972e-02 -2.06652939e-01 6.38407707e-01 2.99610645e-01 1.81147978e-01 -5.80335200e-01 -1.43077105e-01 7.82119095e-01 1.37323931e-01 9.59817413e-03 1.68163931e+00 3.68028551e-01 -1.57064736e-01 4.46572453e-01 1.28579092e+00 -1.29809435e-02 -1.60731006e+00 -5.13291992e-02 -4.08419818e-01 -3.60346943e-01 3.08935314e-01 -4.65376407e-01 -1.05560899e+00 1.15360332e+00 5.58897316e-01 2.54171103e-01 1.11124289e+00 -1.37973070e-01 6.54086530e-01 3.06147218e-01 5.00854790e-01 -6.04917228e-01 -1.35771230e-01 5.60417354e-01 1.11153185e+00 -7.58248985e-01 3.20403963e-01 -4.51436579e-01 9.81354937e-02 1.32078969e+00 2.07865790e-01 -5.70406616e-01 7.22328842e-01 7.17622101e-01 -1.38602003e-01 -2.96846211e-01 -8.87614071e-01 2.91534573e-01 3.89869213e-01 6.22905374e-01 5.86413383e-01 3.65233384e-02 4.02613610e-01 3.43962274e-02 -2.78090447e-01 -7.72504183e-03 2.28398025e-01 7.35365450e-01 -4.87902522e-01 -8.25290024e-01 -3.51256222e-01 6.61216438e-01 -3.04597855e-01 -1.81226239e-01 3.20031136e-01 7.97251463e-01 -1.77939180e-02 2.15106636e-01 5.91962397e-01 -3.00580841e-02 3.45149934e-01 -2.17133224e-01 5.68093300e-01 -5.22068858e-01 -7.17350364e-01 6.85715079e-02 -9.95631889e-03 -9.53363359e-01 -3.53671670e-01 -5.09275198e-01 -1.46062517e+00 -4.34834093e-01 7.15940073e-02 -1.99603125e-01 9.80687559e-01 5.09399056e-01 6.27054811e-01 7.08389640e-01 3.97877216e-01 -1.55731928e+00 -3.34339082e-01 -6.26885951e-01 -3.53933930e-01 -7.20851272e-02 5.43416381e-01 -4.60169077e-01 -1.81415111e-01 -4.75026406e-02]
[9.259081840515137, -3.157470703125]
4d957c74-ea89-4d27-9fce-923a594cd3b7
personalization-in-goal-oriented-dialog
1706.07503
null
http://arxiv.org/abs/1706.07503v3
http://arxiv.org/pdf/1706.07503v3.pdf
Personalization in Goal-Oriented Dialog
The main goal of modeling human conversation is to create agents which can interact with people in both open-ended and goal-oriented scenarios. End-to-end trained neural dialog systems are an important line of research for such generalized dialog models as they do not resort to any situation-specific handcrafting of rules. However, incorporating personalization into such systems is a largely unexplored topic as there are no existing corpora to facilitate such work. In this paper, we present a new dataset of goal-oriented dialogs which are influenced by speaker profiles attached to them. We analyze the shortcomings of an existing end-to-end dialog system based on Memory Networks and propose modifications to the architecture which enable personalization. We also investigate personalization in dialog as a multi-task learning problem, and show that a single model which shares features among various profiles outperforms separate models for each profile.
['Fei Mi', 'Boi Faltings', 'Chaitanya K. Joshi']
2017-06-22
null
null
null
null
['goal-oriented-dialog']
['natural-language-processing']
[-2.55510569e-01 6.66391611e-01 -5.29309409e-03 -9.53820407e-01 -2.08965942e-01 -6.26740575e-01 1.00336397e+00 -2.86261350e-01 -6.33162618e-01 9.74149525e-01 8.19505990e-01 -1.50086746e-01 -1.46654680e-01 -5.56459427e-01 7.15035126e-02 -3.63936871e-01 1.50153279e-01 1.40290546e+00 3.43853980e-01 -9.80547845e-01 1.39434421e-02 3.89552772e-01 -1.10423005e+00 5.35413265e-01 6.86529636e-01 3.68858755e-01 2.52419561e-01 8.86705458e-01 -2.88764745e-01 6.86225593e-01 -8.14283311e-01 -5.97448051e-01 7.76275108e-03 -3.66643935e-01 -1.24460566e+00 5.06151021e-01 2.27110788e-01 -5.06847024e-01 -2.76322186e-01 5.63793719e-01 6.82324171e-01 8.07953596e-01 6.79654598e-01 -1.04497206e+00 -7.57037878e-01 1.04915535e+00 4.20048267e-01 -3.44083868e-02 4.31774646e-01 2.44904801e-01 9.69656467e-01 -3.72049183e-01 4.28178042e-01 1.61670697e+00 4.47371453e-01 1.41697276e+00 -1.27041888e+00 3.21429595e-02 3.90566528e-01 -2.81813014e-02 -6.73498034e-01 -7.17011809e-01 8.47880840e-01 -4.34831291e-01 1.10572064e+00 2.09769309e-01 1.16810024e-01 1.56557262e+00 2.05298681e-02 6.44888401e-01 9.28734064e-01 -4.10497785e-01 2.62199670e-01 7.66812623e-01 5.91515183e-01 4.48929638e-01 -3.45269889e-01 -4.26079519e-02 -5.28847516e-01 -3.07421267e-01 4.38617051e-01 -1.48356006e-01 -1.71070844e-01 -2.50631839e-01 -1.01472032e+00 1.24672174e+00 2.41367500e-02 4.94663537e-01 -2.93361396e-01 -4.74053293e-01 3.70453537e-01 5.87768912e-01 4.20445949e-01 7.51611173e-01 -6.50085568e-01 -5.02895415e-02 -4.26790386e-01 6.27795994e-01 1.60449731e+00 9.08026218e-01 7.49260724e-01 -1.39747158e-01 -6.46204233e-01 1.30118692e+00 5.03276467e-01 -1.93299409e-02 7.95291185e-01 -1.20474052e+00 2.47496918e-01 7.72242427e-01 3.81535232e-01 -2.85160661e-01 -9.70152974e-01 6.25366271e-02 -5.70591211e-01 3.50276977e-01 6.11325741e-01 -9.58999217e-01 -4.21216547e-01 1.91381252e+00 3.11525822e-01 -4.75966603e-01 5.07293105e-01 4.76022959e-01 1.03807998e+00 5.27908802e-01 3.15097541e-01 -1.84549093e-01 1.54885161e+00 -1.43876183e+00 -8.84265840e-01 -4.14605916e-01 4.18277264e-01 -5.59175909e-01 1.19477880e+00 1.55267507e-01 -1.18789279e+00 -5.96748888e-01 -7.06649542e-01 -1.25687972e-01 -6.19183898e-01 -2.72380531e-01 7.74443388e-01 1.07323432e+00 -1.37415481e+00 5.59083760e-01 -3.23840439e-01 -8.53491783e-01 -2.92271376e-01 7.29962885e-01 -1.45587519e-01 7.69492149e-01 -1.41384137e+00 1.47367513e+00 3.83444458e-01 -3.55499536e-01 -6.32640064e-01 -3.75661790e-01 -7.35082150e-01 2.32549727e-01 3.64109010e-01 -9.87211406e-01 1.91813982e+00 -5.51921070e-01 -2.62354159e+00 8.71544063e-01 3.86291929e-02 -5.77661455e-01 3.89157027e-01 -1.80493966e-01 -3.21455508e-01 -3.49481285e-01 -3.10319871e-01 8.33778620e-01 4.81210768e-01 -1.24960530e+00 -7.41060436e-01 -2.60589093e-01 5.94953120e-01 6.02588713e-01 -6.98751450e-01 9.39081833e-02 -1.31074458e-01 -2.83668160e-01 -7.12939203e-01 -1.02855182e+00 -6.71139777e-01 -6.16721153e-01 -3.18586320e-01 -7.81138718e-01 7.22055316e-01 -3.74852300e-01 9.73372102e-01 -1.59668410e+00 3.13426167e-01 -2.74358600e-01 1.72190651e-01 4.94114906e-01 2.01091357e-02 8.30916166e-01 4.30615515e-01 -1.98161647e-01 6.97319359e-02 -9.19420660e-01 6.38412535e-01 3.97704899e-01 -1.48199379e-01 -1.39109582e-01 -3.57525229e-01 6.15134597e-01 -6.71027958e-01 -3.83693337e-01 4.96537298e-01 8.55401531e-02 -7.89457619e-01 1.02818131e+00 -7.74504483e-01 9.41930354e-01 -4.57567245e-01 -1.98177546e-01 2.20794469e-01 -1.98746510e-02 3.87390107e-01 2.32898891e-01 -1.67870730e-01 5.04617035e-01 -8.16704035e-01 1.75749683e+00 -7.03406572e-01 2.15991676e-01 3.19972277e-01 -6.52537346e-01 1.11663258e+00 6.24074697e-01 2.97759175e-01 3.81428413e-02 3.14229995e-01 -3.42828184e-01 2.30437547e-01 -6.37931585e-01 8.78888786e-01 -4.61308867e-01 -4.11251754e-01 6.97204173e-01 4.43549842e-01 -1.19025692e-01 2.18260929e-01 5.79604656e-02 7.14424253e-01 -1.70701116e-01 4.22757059e-01 -3.63958806e-01 9.35211718e-01 -3.65461558e-02 2.66880810e-01 9.30534601e-01 -4.03121442e-01 3.15089077e-01 1.72549069e-01 -4.42065209e-01 -7.34175265e-01 -8.22368264e-01 4.92380857e-02 1.91789639e+00 -1.33347869e-01 -1.56473905e-01 -8.02798867e-01 -7.87673354e-01 -2.56311953e-01 1.10489428e+00 -3.85709018e-01 3.60287316e-02 -6.10415518e-01 -7.32159019e-01 5.04353464e-01 2.65182465e-01 7.45779514e-01 -1.50997889e+00 -3.81609470e-01 4.28840905e-01 -1.00388899e-01 -1.09114766e+00 -6.74554646e-01 1.25618845e-01 -6.96751535e-01 -5.96534193e-01 -7.06991613e-01 -8.23555589e-01 2.69814953e-02 -2.08351269e-01 1.33730888e+00 -2.38718361e-01 6.27027333e-01 8.28402936e-01 -3.25560659e-01 -4.00718212e-01 -1.04226017e+00 5.42721868e-01 1.30658507e-01 -3.42049375e-02 7.27012396e-01 -7.51482785e-01 -3.61497313e-01 4.78711933e-01 -4.71032739e-01 -1.22614317e-01 3.02918285e-01 9.80295956e-01 -5.04402697e-01 -4.04414743e-01 1.00225985e+00 -1.29757881e+00 1.42826867e+00 -3.15117747e-01 -3.13723117e-01 4.04095411e-01 -3.70221078e-01 2.23265916e-01 5.96004367e-01 -4.26620185e-01 -2.08864594e+00 2.31720656e-01 -6.01113319e-01 3.03713530e-01 -7.02318013e-01 1.20905161e-01 -4.91809428e-01 1.32464319e-01 9.02148068e-01 -6.11443892e-02 2.19514012e-01 -6.45314813e-01 8.17071795e-01 1.06440651e+00 5.30638993e-01 -7.83448458e-01 3.32174242e-01 1.43851414e-01 -5.10891676e-01 -1.05262995e+00 -7.12594748e-01 -5.76557815e-01 -7.70651221e-01 -1.35366693e-01 1.09232593e+00 -5.90886116e-01 -1.19141984e+00 4.30198580e-01 -1.34043193e+00 -8.16430449e-01 -1.37364775e-01 3.49544317e-01 -8.36259305e-01 5.04224300e-01 -7.05123782e-01 -9.96567369e-01 -4.36954021e-01 -1.01880372e+00 6.72716439e-01 4.63350445e-01 -8.08281004e-01 -1.71494150e+00 5.10545194e-01 7.88032293e-01 7.13870943e-01 -5.35936832e-01 8.12818885e-01 -1.61114824e+00 -1.13871276e-01 2.38660853e-02 5.48161864e-01 3.13000441e-01 3.34795952e-01 -6.33545101e-01 -1.25148129e+00 -1.31317779e-01 4.57893968e-01 -5.88514686e-01 4.52332586e-01 1.61513492e-01 4.39080834e-01 -5.43504059e-01 -2.56486177e-01 1.38898551e-01 5.39310217e-01 3.92860681e-01 2.74731249e-01 9.53927636e-02 3.62439781e-01 1.21953118e+00 2.11101130e-01 5.62611163e-01 7.63498008e-01 1.02330709e+00 6.50732666e-02 2.29920268e-01 -1.32489800e-01 -1.76910982e-02 4.61488366e-01 5.47942281e-01 1.20688019e-04 -5.32006323e-01 -5.54635227e-01 4.51765418e-01 -2.04726434e+00 -9.15575683e-01 1.31621003e-01 1.92239690e+00 1.13953149e+00 -5.96493818e-02 5.90291798e-01 -6.53912902e-01 7.93381095e-01 3.12586755e-01 -3.14218670e-01 -6.02682769e-01 1.59771413e-01 -7.90104792e-02 -7.09136128e-02 1.10020709e+00 -1.16299963e+00 1.33223474e+00 6.48667192e+00 1.89518571e-01 -5.46016812e-01 2.82724530e-01 5.08997798e-01 3.04469347e-01 3.39850076e-02 -9.78041217e-02 -1.34043419e+00 1.53416201e-01 1.19571579e+00 -9.27434862e-02 4.32960033e-01 1.03324902e+00 1.79514959e-01 2.15332717e-01 -1.60249829e+00 4.71255779e-01 1.38618082e-01 -1.07873476e+00 1.17952619e-02 -2.56215297e-02 5.91648698e-01 -3.74574125e-01 -7.18936473e-02 7.62762547e-01 1.08981323e+00 -8.86715174e-01 2.84684100e-03 4.46292967e-01 -1.40836388e-01 -2.33962879e-01 7.06617534e-01 6.35370433e-01 -4.98709261e-01 -7.02702850e-02 -4.71110821e-01 -2.33273908e-01 6.13594413e-01 -1.35183096e-01 -1.43582439e+00 1.47790283e-01 4.03410912e-01 -2.77084708e-02 -1.95376322e-01 5.40121436e-01 -5.29634915e-02 2.45715410e-01 -9.58351642e-02 -3.49329025e-01 2.29970917e-01 -2.88252413e-01 7.29181349e-01 1.35603988e+00 -7.94102252e-02 1.08926706e-01 4.17601347e-01 7.16423929e-01 -1.90983322e-02 9.99111459e-02 -6.37964666e-01 3.02898228e-01 3.56400400e-01 1.22829032e+00 -2.08423033e-01 -4.28195000e-01 -6.19820297e-01 1.02039170e+00 5.25441945e-01 2.48771966e-01 -3.08420181e-01 4.35672067e-02 7.40613043e-01 -2.87608415e-01 1.96807683e-02 -1.66993856e-01 -2.17655510e-01 -1.19011343e+00 -6.30133271e-01 -9.85457659e-01 6.37305081e-01 -2.35908374e-01 -1.80051041e+00 9.35977876e-01 3.12702715e-01 -4.71420169e-01 -9.90951598e-01 -7.94984519e-01 -9.27080810e-01 9.23732996e-01 -9.67776954e-01 -1.23389030e+00 -1.93723574e-01 8.54932189e-01 1.04345298e+00 -7.97688365e-01 1.28057075e+00 -5.46605932e-03 -3.36222649e-01 5.61566830e-01 -1.04425520e-01 3.01401839e-02 1.10298169e+00 -1.54930675e+00 2.56462097e-01 2.57797331e-01 -1.79348022e-01 7.98588276e-01 1.15470874e+00 -6.48496568e-01 -8.12185824e-01 -8.38329554e-01 1.20718694e+00 -8.48930180e-01 5.69975913e-01 -4.94398624e-01 -8.70343685e-01 1.06634867e+00 1.12239528e+00 -9.61526334e-01 9.18718994e-01 7.40507245e-01 1.64516538e-01 1.30552828e-01 -1.12941694e+00 9.45524514e-01 9.49808300e-01 -1.72059774e-01 -1.31387067e+00 5.82300901e-01 6.89005136e-01 -3.31840038e-01 -9.01167274e-01 7.57289082e-02 1.80225715e-01 -1.28547275e+00 1.07770717e+00 -9.39198613e-01 -1.24415010e-01 3.74924898e-01 1.14911497e-01 -1.67399788e+00 -3.20401102e-01 -1.13962483e+00 -1.58783607e-02 1.58760500e+00 5.62546611e-01 -7.89364219e-01 1.04371655e+00 1.28433895e+00 -4.86929744e-01 -2.09290832e-01 -5.73570073e-01 -5.81400931e-01 5.07227957e-01 1.51593074e-01 6.61082089e-01 6.97897017e-01 6.38703108e-01 1.14277196e+00 -9.64208007e-01 -1.03407487e-01 2.27536410e-01 4.91633788e-02 1.22579265e+00 -1.42224288e+00 -6.13995552e-01 -6.21863365e-01 3.03561866e-01 -1.57965076e+00 6.13831103e-01 -5.98766625e-01 2.47538194e-01 -1.41954780e+00 -8.85176808e-02 -3.36108983e-01 2.66188174e-01 1.97050929e-01 -3.58154893e-01 -4.84005868e-01 1.90071449e-01 -2.33825184e-02 -6.14269018e-01 1.01815939e+00 9.04507935e-01 -1.37335807e-01 -8.74637902e-01 7.52438128e-01 -8.46773624e-01 9.14706469e-01 1.08559632e+00 -1.06258228e-01 -7.64991999e-01 -1.63626656e-01 -3.78421217e-01 2.33427376e-01 -6.49499223e-02 -8.04392040e-01 6.73523486e-01 -2.12668404e-01 -4.64946143e-02 -2.71486253e-01 9.56828594e-01 -6.69290781e-01 -2.33831316e-01 1.05395295e-01 -7.31017888e-01 -1.11267967e-02 -3.98540981e-02 3.79159123e-01 2.29873843e-02 -6.58279598e-01 8.60258222e-01 -6.09627783e-01 -7.73429811e-01 9.93047133e-02 -8.42197716e-01 -3.96973193e-02 8.55071664e-01 1.36669204e-01 -4.99602795e-01 -1.03285265e+00 -1.08188570e+00 5.67494333e-01 1.50387242e-01 4.96031672e-01 1.94205895e-01 -7.87832379e-01 -7.05698252e-01 -2.65245616e-01 -1.43496051e-01 -2.74771988e-01 4.52951759e-01 1.30482063e-01 2.48787291e-02 7.38005221e-01 -4.81237173e-01 -2.36218289e-01 -1.39198101e+00 6.52903795e-01 4.83043402e-01 -7.41561890e-01 -4.53748196e-01 8.35756779e-01 4.23853934e-01 -1.10415375e+00 5.88559508e-01 2.99503595e-01 -9.60962713e-01 8.69794711e-02 4.45016831e-01 2.08006680e-01 -2.05890208e-01 -5.93252778e-01 1.81681380e-01 -1.25110000e-01 -3.68790388e-01 -5.20020723e-01 1.24273467e+00 -5.04297376e-01 8.21866021e-02 5.19339561e-01 4.76017565e-01 -1.73843816e-01 -1.27710128e+00 -4.71406907e-01 2.33434275e-01 -9.69071537e-02 -6.43238783e-01 -8.82467389e-01 -5.32963037e-01 7.77078032e-01 3.76248062e-01 5.44554174e-01 5.76851130e-01 1.57471463e-01 5.79483986e-01 1.09460914e+00 1.26620844e-01 -1.19824934e+00 2.97912329e-01 1.11168098e+00 1.11119938e+00 -1.37835848e+00 -2.86377788e-01 -3.00469846e-01 -1.29382217e+00 1.00128686e+00 1.12222767e+00 1.06273927e-01 6.79243982e-01 -1.89794481e-01 4.43174720e-01 -1.16110191e-01 -9.46497023e-01 -3.70399296e-01 1.45339102e-01 1.01995885e+00 4.62187946e-01 -7.92608261e-02 -3.22448999e-01 9.20709014e-01 -5.30035973e-01 -4.46479499e-01 4.81763303e-01 6.19763672e-01 -5.38900554e-01 -1.72753561e+00 -2.89058149e-01 1.05934337e-01 -2.93083459e-01 5.14043868e-02 -7.20251203e-01 6.32568598e-01 -2.43208483e-01 1.52958167e+00 -3.14530820e-01 -4.19132024e-01 3.91058832e-01 7.15964735e-01 1.89741924e-01 -1.07027388e+00 -1.29877055e+00 -2.04656422e-01 8.30968618e-01 4.92188893e-02 -4.81240869e-01 -5.11052907e-01 -7.58614123e-01 -2.60843098e-01 -6.61143940e-03 4.86181438e-01 2.21601725e-01 1.01627731e+00 1.44290105e-01 4.03422147e-01 5.30421138e-01 -1.18166125e+00 -9.67414856e-01 -1.51702058e+00 -4.97997582e-01 6.45139277e-01 5.86976483e-02 -6.00767016e-01 -1.16487458e-01 1.05930664e-01]
[12.785675048828125, 8.020492553710938]
00c35ae9-dda7-45a9-b868-fa76a3679968
cross3dvg-baseline-and-dataset-for-cross
2305.13876
null
https://arxiv.org/abs/2305.13876v1
https://arxiv.org/pdf/2305.13876v1.pdf
Cross3DVG: Baseline and Dataset for Cross-Dataset 3D Visual Grounding on Different RGB-D Scans
We present Cross3DVG, a novel task for cross-dataset visual grounding in 3D scenes, revealing the limitations of existing 3D visual grounding models using restricted 3D resources and thus easily overfit to a specific 3D dataset. To facilitate Cross3DVG, we have created a large-scale 3D visual grounding dataset containing more than 63k diverse descriptions of 3D objects within 1,380 indoor RGB-D scans from 3RScan with human annotations, paired with the existing 52k descriptions on ScanRefer. We perform Cross3DVG by training a model on the source 3D visual grounding dataset and then evaluating it on the target dataset constructed in different ways (e.g., different sensors, 3D reconstruction methods, and language annotators) without using target labels. We conduct comprehensive experiments using established visual grounding models, as well as a CLIP-based 2D-3D integration method, designed to bridge the gaps between 3D datasets. By performing Cross3DVG tasks, we found that (i) cross-dataset 3D visual grounding has significantly lower performance than learning and evaluation with a single dataset, suggesting much room for improvement in cross-dataset generalization of 3D visual grounding, (ii) better detectors and transformer-based localization modules for 3D grounding are beneficial for enhancing 3D grounding performance and (iii) fusing 2D-3D data using CLIP demonstrates further performance improvements. Our Cross3DVG task will provide a benchmark for developing robust 3D visual grounding models capable of handling diverse 3D scenes while leveraging deep language understanding.
['Motoki Kawanabe', 'Shuhei Kurita', 'Daichi Azuma', 'Taiki Miyanishi']
2023-05-23
null
null
null
null
['visual-grounding', '3d-reconstruction']
['computer-vision', 'computer-vision']
[-1.18143909e-01 2.25254968e-01 -9.94250104e-02 -5.72502494e-01 -1.10941303e+00 -1.12997258e+00 5.59716284e-01 3.97758543e-01 1.28966402e-02 6.02613911e-02 2.06584454e-01 -4.66512442e-01 4.75453623e-02 -8.21782410e-01 -1.13745844e+00 -2.21330151e-02 -1.89564049e-01 6.64577723e-01 4.24335390e-01 -1.39590949e-01 -1.32785827e-01 9.09572423e-01 -1.56754363e+00 3.02960128e-01 3.08756322e-01 1.15406990e+00 2.80539155e-01 4.46825117e-01 -2.99189210e-01 3.89344543e-01 -5.91737151e-01 4.03550416e-02 6.63318992e-01 -1.31819218e-01 -7.03214943e-01 2.81727552e-01 1.19612575e+00 -2.07556084e-01 -2.67129749e-01 7.02270150e-01 5.71048915e-01 2.59621143e-02 4.62009579e-01 -1.61013806e+00 -8.98627222e-01 8.99382234e-02 -4.34747338e-01 -3.34724396e-01 8.60360980e-01 2.12885052e-01 8.27967525e-01 -8.55603933e-01 8.63522232e-01 1.43301642e+00 1.11781943e+00 5.38388193e-01 -1.39957082e+00 -8.48402202e-01 2.81533360e-01 -4.83864218e-01 -1.75756431e+00 -1.27358600e-01 6.61827207e-01 -7.21096814e-01 1.31344259e+00 5.57967126e-02 8.30071747e-01 1.34495342e+00 4.93729189e-02 5.76997519e-01 1.22729695e+00 -2.17517406e-01 2.88769186e-01 -4.65793833e-02 2.03998629e-02 9.28174317e-01 2.63958395e-01 1.09313600e-01 -8.02337766e-01 -1.53779373e-01 7.34813690e-01 -3.09521649e-02 -7.86656812e-02 -9.41192329e-01 -1.37359428e+00 4.71505433e-01 9.68806505e-01 -9.68699381e-02 8.59759897e-02 3.59010786e-01 2.59324402e-01 1.32846147e-01 6.55954301e-01 4.54537988e-01 -4.69935209e-01 3.13150316e-01 -8.97781610e-01 4.89634722e-01 4.72489178e-01 1.62308669e+00 1.15199995e+00 -1.29633054e-01 5.31593263e-02 3.92955452e-01 6.00547135e-01 1.16040671e+00 -1.83253154e-01 -9.29883897e-01 8.23501050e-01 8.00986290e-01 1.47769200e-02 -9.84445989e-01 -7.39744961e-01 -1.97023585e-01 -3.92746329e-01 4.00977641e-01 2.56925076e-01 3.13206196e-01 -1.32874537e+00 1.61885405e+00 3.82666707e-01 -1.37776569e-01 4.75535206e-02 9.28352654e-01 1.37856472e+00 3.67601842e-01 5.78848980e-02 7.54223347e-01 1.08113706e+00 -5.51225126e-01 -9.41036344e-02 -4.78404343e-01 8.81421506e-01 -4.85862464e-01 1.26598692e+00 3.43164690e-02 -6.08669162e-01 -7.26165473e-01 -1.17813087e+00 -3.63484591e-01 -7.70449221e-01 -1.29590660e-01 6.43944621e-01 4.71124053e-01 -1.18773353e+00 1.40599653e-01 -8.59646499e-01 -9.73750651e-01 5.40094912e-01 1.47446156e-01 -8.83707643e-01 -3.33246112e-01 -7.18201339e-01 8.74519289e-01 5.32058954e-01 -2.23146349e-01 -1.41484833e+00 -7.02007174e-01 -1.33571100e+00 -6.92244589e-01 2.80107647e-01 -8.81458104e-01 9.02958691e-01 -1.15213357e-01 -5.77824414e-01 1.65891695e+00 1.49224982e-01 -4.71596830e-02 4.31287885e-01 -1.22120120e-01 -4.11335409e-01 -1.19278580e-01 7.22849429e-01 9.89376366e-01 3.05384547e-01 -1.71729958e+00 -2.40010217e-01 -7.16768682e-01 2.13860810e-01 2.47356817e-01 5.09195447e-01 -5.37784755e-01 -5.01348794e-01 -3.37648630e-01 8.00116420e-01 -9.51711059e-01 -1.15822917e-02 4.00072813e-01 -6.34241760e-01 1.42721385e-01 7.15599537e-01 -3.83853406e-01 3.65778148e-01 -2.23349905e+00 -8.32388401e-02 2.63968557e-01 1.74994960e-01 -1.34441301e-01 -1.48600623e-01 4.17769223e-01 9.15879905e-02 3.10539305e-01 -2.37097386e-02 -5.58844268e-01 1.00215346e-01 5.07261455e-01 -3.97681028e-01 4.72227007e-01 2.28565544e-01 9.15618420e-01 -1.02040195e+00 -5.02669394e-01 4.12170231e-01 3.40430439e-01 -5.31236291e-01 3.15165758e-01 -3.78046274e-01 3.87561172e-01 -4.15711015e-01 1.12385368e+00 7.19811380e-01 -3.98461461e-01 -1.72391504e-01 -4.34539437e-01 -5.48421964e-02 2.22883880e-01 -1.14773774e+00 2.59107065e+00 -5.00228286e-01 4.41219360e-01 -1.07793778e-01 -4.65456307e-01 1.28576636e+00 -1.14475615e-01 2.81113118e-01 -8.05284023e-01 -6.51499182e-02 1.82859525e-01 -7.99232900e-01 -4.38142478e-01 5.69527149e-01 -9.23223644e-02 -5.85071027e-01 3.66514206e-01 4.66685355e-01 -1.07444119e+00 -4.25839067e-01 2.83164114e-01 1.03513408e+00 7.18534827e-01 1.49416879e-01 -1.86329797e-01 -2.31167480e-01 6.35205328e-01 2.64544841e-02 1.05531871e+00 -1.50531709e-01 1.06064475e+00 -7.98472576e-03 -4.66349721e-01 -1.10665965e+00 -1.52121103e+00 -2.38755107e-01 6.53252900e-01 6.03945553e-01 -5.19138396e-01 -2.47026682e-01 -8.49717259e-01 5.53908050e-01 6.19291067e-01 -7.54240453e-01 4.64234781e-03 -7.79305398e-02 -1.98796064e-01 9.92691100e-01 7.34734595e-01 6.55611157e-01 -3.58605653e-01 -6.41046762e-01 -2.17599213e-01 -5.91969723e-03 -1.35135436e+00 -6.69445917e-02 7.39319146e-01 -7.65255094e-01 -1.11966777e+00 -4.83698398e-01 -6.01920605e-01 4.93945807e-01 6.86030984e-01 1.47562516e+00 -1.76168159e-01 -2.12579086e-01 9.97163713e-01 -4.74812388e-01 -5.52175105e-01 -4.01807934e-01 1.63319245e-01 1.00825526e-01 -7.64844418e-01 4.05252367e-01 -2.82102913e-01 -1.65012285e-01 4.88739192e-01 -6.14704847e-01 2.15974450e-01 1.59240171e-01 2.81449527e-01 1.03833723e+00 -4.26059127e-01 -5.81411161e-02 -4.24716741e-01 -2.57747117e-02 -5.33630967e-01 -7.49199629e-01 3.02756727e-01 -2.54929364e-01 -1.95091683e-02 -8.77982080e-02 -1.49668649e-01 -5.41509211e-01 6.93114921e-02 -7.64327645e-02 -6.57753766e-01 -4.63438570e-01 2.37069741e-01 -3.96723300e-01 -3.17480803e-01 1.06967711e+00 -2.01832175e-01 -1.20255359e-01 -6.08811557e-01 6.66471124e-01 5.48790157e-01 5.95388353e-01 -7.98656344e-01 1.03011286e+00 6.78945601e-01 1.97573245e-01 -5.25785506e-01 -1.15844429e+00 -5.59093058e-01 -8.24991107e-01 -1.71118349e-01 1.24775076e+00 -1.49124849e+00 -3.00441206e-01 2.52203226e-01 -1.13151348e+00 -6.66336238e-01 -4.02807504e-01 4.42523152e-01 -5.92252135e-01 1.09982483e-01 -9.14282650e-02 -6.10723257e-01 1.13624342e-01 -1.07152522e+00 1.92619872e+00 -2.01773658e-01 -3.54258060e-01 -7.78754950e-01 1.61765262e-01 3.40457708e-01 -8.91490579e-02 7.12240458e-01 7.49170601e-01 -5.76935470e-01 -6.57333255e-01 -1.41193748e-01 -4.32862580e-01 2.23224074e-01 2.90800482e-01 -2.87258863e-01 -1.09896886e+00 -1.01194136e-01 -4.86978024e-01 -7.15669394e-01 5.26642501e-01 1.29309222e-02 7.98512638e-01 3.76029164e-01 -5.42214453e-01 9.55025375e-01 1.57740915e+00 -7.24418536e-02 3.00406963e-01 5.50602794e-01 1.15672827e+00 4.53158110e-01 7.20101237e-01 2.51347750e-01 8.49117100e-01 7.25038528e-01 1.00511539e+00 -5.98074436e-01 -4.38088089e-01 -8.14300060e-01 4.58723791e-02 2.61064500e-01 2.16726571e-01 -4.09293920e-01 -1.40276253e+00 5.63272178e-01 -1.55631566e+00 -4.62017059e-01 -1.89903542e-01 2.15702820e+00 4.17609811e-01 8.43016356e-02 -5.06770089e-02 -1.96591839e-01 3.85610312e-01 2.68433750e-01 -6.72604382e-01 4.23959270e-02 -5.20101964e-01 1.36212379e-01 8.77643943e-01 3.48734379e-01 -1.11799812e+00 1.04382622e+00 6.63687229e+00 3.49187702e-01 -8.84425402e-01 2.17443425e-02 1.66303992e-01 -1.51360020e-01 -5.45856833e-01 1.08151644e-01 -1.01669455e+00 -3.37101892e-02 3.00155401e-01 3.92184794e-01 2.29324475e-01 1.11367643e+00 -6.81690574e-02 3.11588198e-02 -1.35723639e+00 1.24152815e+00 2.52783060e-01 -1.37582791e+00 1.15549967e-01 1.81102052e-01 7.96913981e-01 6.66027427e-01 -4.40881997e-01 1.24782920e-01 7.11239100e-01 -9.81524527e-01 1.50290108e+00 2.19761387e-01 9.02972937e-01 -2.12630898e-01 4.54642624e-01 1.21772833e-01 -1.26575458e+00 4.20033664e-01 -2.62709051e-01 1.31718621e-01 1.81856707e-01 3.02609295e-01 -8.46250534e-01 9.02184129e-01 1.15118718e+00 9.33564544e-01 -9.10375059e-01 7.86245823e-01 -2.71053165e-01 -6.66426821e-03 -5.74446917e-01 2.30674878e-01 2.79008001e-01 1.47847667e-01 4.30641472e-01 1.02051616e+00 6.25034094e-01 -3.01483393e-01 5.08847356e-01 1.12005877e+00 4.78787534e-02 -2.40176916e-01 -1.39102602e+00 1.52347848e-01 4.31589127e-01 8.89867127e-01 -7.84379780e-01 -2.42333457e-01 -4.12908554e-01 8.56838524e-01 2.19839647e-01 3.54854017e-01 -9.44958091e-01 -1.67891607e-01 7.95252144e-01 4.19232428e-01 1.42985702e-01 -9.08525705e-01 -3.31615418e-01 -1.11683774e+00 -7.84000754e-02 -4.94305253e-01 3.31004888e-01 -1.61579120e+00 -1.33385122e+00 5.42932153e-01 4.53871250e-01 -1.30941987e+00 1.14691611e-02 -7.98146844e-01 2.73485929e-01 8.47077131e-01 -1.29111719e+00 -1.81940413e+00 -9.87361372e-01 7.62889683e-01 -2.02232283e-02 1.58248186e-01 8.21992874e-01 1.89362511e-01 8.75392035e-02 2.41601944e-01 -4.23912823e-01 -2.27587321e-03 7.44086385e-01 -1.19676793e+00 7.94390261e-01 6.30368054e-01 5.32916844e-01 4.19468313e-01 3.94124568e-01 -8.09752584e-01 -1.46399605e+00 -1.42255723e+00 5.43889940e-01 -1.13421261e+00 5.34138262e-01 -9.95781779e-01 -7.06840515e-01 1.03374839e+00 -3.03615034e-01 2.99541473e-01 6.77759111e-01 1.05591953e-01 -9.20593977e-01 3.72431837e-02 -1.31463253e+00 3.23504627e-01 1.91726661e+00 -1.05778956e+00 -5.27508974e-01 3.93041134e-01 1.12149060e+00 -9.44393456e-01 -9.78863001e-01 5.32665312e-01 4.19173509e-01 -9.00892913e-01 1.23721814e+00 -4.35902238e-01 4.08536494e-02 -8.18890512e-01 -1.14609277e+00 -1.11737609e+00 -9.37490538e-02 -2.40974519e-02 3.13838452e-01 1.27786720e+00 2.18225121e-01 -3.62888932e-01 6.43344879e-01 4.33525980e-01 -6.51451051e-01 -1.20570362e-01 -9.03348982e-01 -1.15573227e+00 -8.29080194e-02 -1.18734157e+00 1.02100444e+00 9.09534633e-01 -5.70625186e-01 2.62935907e-02 2.78988685e-02 5.40436625e-01 6.06534481e-01 2.65661776e-01 1.35388041e+00 -1.11631846e+00 -3.35159078e-02 -5.01217088e-04 -8.96836042e-01 -1.21656680e+00 2.32086003e-01 -1.34145904e+00 9.98338088e-02 -1.96259785e+00 -9.38610956e-02 -8.87427747e-01 7.24110082e-02 9.86697733e-01 5.03127933e-01 6.59709036e-01 1.20725721e-01 1.29626483e-01 -7.04814970e-01 5.26010633e-01 1.06186056e+00 -4.70918775e-01 -8.62876400e-02 -5.67416310e-01 -6.32149041e-01 5.32986999e-01 3.30397546e-01 -4.91020650e-01 -4.29432720e-01 -9.09427166e-01 3.79492998e-01 -3.30987155e-01 9.98087049e-01 -1.19314659e+00 -1.35704949e-01 -6.55350089e-02 5.12636900e-01 -1.02346623e+00 4.93950188e-01 -9.89719748e-01 4.44108129e-01 -8.35595056e-02 -9.13264137e-03 6.90724328e-02 6.46416128e-01 4.42380369e-01 1.25323340e-01 2.02590153e-01 3.42008263e-01 -4.73188847e-01 -1.13920295e+00 2.72833318e-01 6.45333854e-03 2.35857874e-01 8.25881720e-01 -5.18032968e-01 -4.61746454e-01 -2.09146217e-02 -7.02356696e-01 1.48058429e-01 1.16691351e+00 7.08680034e-01 5.39749622e-01 -1.61617625e+00 -2.93486953e-01 3.44218284e-01 1.00238502e+00 5.25662959e-01 1.01015478e-01 7.47004077e-02 -6.38194501e-01 1.94339469e-01 -1.67268142e-01 -1.30789161e+00 -9.98507500e-01 4.60275471e-01 4.60668921e-01 3.20417523e-01 -6.45679474e-01 9.78650987e-01 2.63840884e-01 -1.09350216e+00 1.15849845e-01 -8.87858391e-01 5.08814216e-01 -4.56203260e-02 4.04579490e-02 1.88303590e-02 4.21911746e-01 -9.10534263e-01 -9.63161230e-01 1.03118801e+00 6.88555002e-01 -8.25888440e-02 1.16511571e+00 -1.81652382e-01 1.75176695e-01 7.23118663e-01 1.31877315e+00 -8.05007666e-02 -1.28805339e+00 -1.86092451e-01 -1.68421403e-01 -4.82083321e-01 1.24888718e-02 -8.43476713e-01 -6.70181394e-01 7.70006180e-01 8.12985957e-01 3.76621597e-02 7.75022686e-01 8.28976631e-01 2.51289219e-01 3.50903541e-01 1.20498264e+00 -5.17196774e-01 1.54612705e-01 3.93896550e-01 9.93225455e-01 -1.47521627e+00 2.73208052e-01 -2.53170013e-01 -3.91582906e-01 8.15251350e-01 5.74172974e-01 1.14571854e-01 6.01658463e-01 2.41045818e-01 3.28696907e-01 -8.53200138e-01 -8.51751044e-02 -4.50093627e-01 3.97098094e-01 1.13881350e+00 3.06877997e-02 1.03172235e-01 9.35988426e-01 2.81547397e-01 -3.24827820e-01 -1.58655152e-01 4.46053073e-02 1.00270116e+00 1.61260534e-02 -9.19182062e-01 -5.02423406e-01 -1.29264742e-01 3.35405827e-01 8.81088153e-02 -6.28563523e-01 1.34671736e+00 5.50567150e-01 8.75299931e-01 1.51471466e-01 -7.17044652e-01 7.61186779e-01 -1.80961803e-01 7.98875690e-01 -8.96713555e-01 -2.87213057e-01 -2.92846829e-01 1.72206640e-01 -9.72869992e-01 -5.68025351e-01 -4.76709604e-01 -1.20577943e+00 -6.96381107e-02 -1.65047377e-01 -3.49204093e-01 1.05099833e+00 7.23148108e-01 6.08818948e-01 2.83239633e-01 1.95767358e-01 -1.26845956e+00 7.36069912e-03 -6.03901505e-01 -7.09242523e-01 5.21810591e-01 3.25294375e-01 -1.00214970e+00 -3.02279413e-01 -9.98702496e-02]
[8.069695472717285, -3.0801408290863037]
8950000b-673d-4e9e-845a-f99fa848cb84
deep-learning-for-real-time-gravitational-1
1711.03121
null
http://arxiv.org/abs/1711.03121v1
http://arxiv.org/pdf/1711.03121v1.pdf
Deep Learning for Real-time Gravitational Wave Detection and Parameter Estimation: Results with Advanced LIGO Data
The recent Nobel-prize-winning detections of gravitational waves from merging black holes and the subsequent detection of the collision of two neutron stars in coincidence with electromagnetic observations have inaugurated a new era of multimessenger astrophysics. To enhance the scope of this emergent field of science, we pioneered the use of deep learning with convolutional neural networks, that take time-series inputs, for rapid detection and characterization of gravitational wave signals. This approach, Deep Filtering, was initially demonstrated using simulated LIGO noise. In this article, we present the extension of Deep Filtering using real data from LIGO, for both detection and parameter estimation of gravitational waves from binary black hole mergers using continuous data streams from multiple LIGO detectors. We demonstrate for the first time that machine learning can detect and estimate the true parameters of real events observed by LIGO. Our results show that Deep Filtering achieves similar sensitivities and lower errors compared to matched-filtering while being far more computationally efficient and more resilient to glitches, allowing real-time processing of weak time-series signals in non-stationary non-Gaussian noise with minimal resources, and also enables the detection of new classes of gravitational wave sources that may go unnoticed with existing detection algorithms. This unified framework for data analysis is ideally suited to enable coincident detection campaigns of gravitational waves and their multimessenger counterparts in real-time.
['E. A. Huerta', 'Daniel George']
2017-11-08
null
null
null
null
['gravitational-wave-detection']
['miscellaneous']
[-4.14244533e-01 -2.98208714e-01 5.65558672e-01 -8.26998726e-02 -5.59040189e-01 -6.28269076e-01 1.15851450e+00 -3.16671312e-01 -5.57363153e-01 3.04430038e-01 -8.72582048e-02 -6.65207744e-01 -3.09677213e-01 -1.06806946e+00 -4.32706773e-01 -8.47948492e-01 -5.69281816e-01 8.70765746e-01 3.36004823e-01 -2.39822879e-01 7.15699345e-02 8.67099524e-01 -1.30314052e+00 -1.70809373e-01 8.01567137e-02 1.08292258e+00 8.41954872e-02 8.26096773e-01 2.94724077e-01 4.37015146e-01 -6.27693772e-01 -3.05231452e-01 6.39948308e-01 -3.52713466e-01 -4.07694668e-01 -5.30079007e-01 2.73089200e-01 -1.77539423e-01 -9.85683382e-01 9.19268310e-01 8.44134808e-01 2.38268271e-01 2.52733320e-01 -7.93296158e-01 -6.30945861e-02 4.32500213e-01 -1.35837868e-01 1.03021741e+00 -3.56089324e-02 9.16606605e-01 8.55891049e-01 -6.94490314e-01 3.36964250e-01 8.74121666e-01 9.14605796e-01 4.96659130e-02 -9.93989050e-01 -7.70218372e-01 -1.00866461e+00 2.91234732e-01 -7.13858604e-01 -5.20803034e-01 5.33977270e-01 -7.53663301e-01 1.39171088e+00 1.96368605e-01 5.61503112e-01 8.87904584e-01 -1.07669914e-02 2.91753829e-01 9.02727783e-01 -5.39061010e-01 2.51589537e-01 -7.22921073e-01 -1.22045524e-01 3.82521421e-01 4.69591588e-01 1.28798258e+00 -8.00791681e-01 -3.49079162e-01 8.05022657e-01 -3.88619542e-01 -2.71620661e-01 1.80607691e-01 -1.37533581e+00 1.09444499e+00 2.04395995e-01 5.47582626e-01 -2.82880992e-01 3.82216483e-01 2.57681400e-01 4.91050303e-01 3.85018259e-01 9.08635437e-01 -6.38650596e-01 -3.35898489e-01 -1.04493618e+00 6.11750662e-01 7.01743245e-01 3.88741672e-01 5.76959848e-01 5.53498626e-01 1.84000537e-01 1.09661534e-01 2.01045528e-01 8.28938961e-01 8.19884777e-01 -5.32342196e-01 1.03247009e-01 -1.02704279e-02 1.89023972e-01 -6.49504542e-01 -1.15276170e+00 -1.00446570e+00 -6.47623837e-01 6.34580314e-01 8.07681561e-01 -4.52209622e-01 -8.46591830e-01 1.35370517e+00 2.97624767e-01 4.43296254e-01 -2.31090710e-02 1.04540217e+00 8.54752839e-01 3.26935053e-01 -2.88465053e-01 -5.72106950e-02 1.48992193e+00 -1.72834218e-01 -1.95573196e-01 -6.27831340e-01 3.49506408e-01 -9.41486478e-01 6.43566623e-02 3.46463948e-01 -7.98270643e-01 -4.60788995e-01 -1.02234066e+00 3.73719156e-01 -2.28200361e-01 -2.28107512e-01 1.32146764e+00 6.84449911e-01 -5.89725256e-01 9.13069725e-01 -1.19955850e+00 -1.72481745e-01 2.12916985e-01 2.33162835e-01 7.54904971e-02 8.14947307e-01 -1.30683827e+00 6.28853738e-01 5.19797146e-01 -6.30022883e-02 -6.88196599e-01 -9.48729575e-01 -4.19255406e-01 1.75124347e-01 -9.95404795e-02 -6.50595009e-01 1.48175406e+00 -3.29444200e-01 -1.06088805e+00 9.64971244e-01 4.71420467e-01 -1.27063620e+00 5.97447813e-01 5.83883896e-02 -8.86143982e-01 1.22902259e-01 4.12203930e-02 -2.52086878e-01 7.64905572e-01 -2.78118253e-01 -9.95881975e-01 -2.23750830e-01 -3.91632259e-01 -6.06544435e-01 4.34584469e-01 3.51568997e-01 6.70045540e-02 -6.98703527e-01 6.51563048e-01 -5.25100648e-01 -1.03141010e-01 -6.56505167e-01 1.30343199e-01 -2.75242716e-01 5.17470241e-01 -4.39391196e-01 4.49965745e-01 -2.04740691e+00 -4.03860003e-01 -4.61925603e-02 4.37980294e-01 4.04822260e-01 3.05475414e-01 5.17393231e-01 -3.68788183e-01 -2.95910329e-01 1.47607133e-01 5.73123991e-03 1.48805842e-01 -2.46038988e-01 -5.32527328e-01 7.51513720e-01 -8.36563110e-02 1.06879103e+00 -8.82694185e-01 4.59291965e-01 4.80861634e-01 6.41582236e-02 -3.18001151e-01 2.46593520e-01 -1.04964897e-01 7.58397818e-01 -8.29545632e-02 3.50854367e-01 9.25945759e-01 5.04879393e-02 -3.87264460e-01 -1.07746467e-01 -5.90607345e-01 5.64956844e-01 -1.00276244e+00 1.16400182e+00 -2.44791955e-01 1.10129142e+00 4.01197284e-01 -1.23968184e+00 1.15359950e+00 2.38853261e-01 4.20290560e-01 -1.03197551e+00 3.71629268e-01 4.92622226e-01 2.21082538e-01 -8.10292661e-01 3.24160814e-01 -8.83321881e-01 -4.05582339e-02 3.95504951e-01 5.53416491e-01 -3.84181976e-01 1.19725473e-01 -9.74217057e-02 1.41942263e+00 -2.15407535e-01 5.15634834e-04 -1.59400910e-01 -2.07074322e-02 -1.33806661e-01 5.03730536e-01 1.44775319e+00 -1.16786778e-01 7.16391206e-01 6.98119029e-02 -1.07971549e+00 -1.13301325e+00 -1.08518374e+00 -4.54632699e-01 6.85208082e-01 -2.08896711e-01 -2.18113005e-01 1.01211518e-01 -8.84173885e-02 3.72985482e-01 5.92858493e-01 -2.59741813e-01 -3.76717746e-01 -6.24612868e-01 -1.64056945e+00 6.82145536e-01 2.59050012e-01 4.03525889e-01 -1.05083501e+00 -9.02940214e-01 4.93715465e-01 1.28429249e-01 -9.59627092e-01 3.99596334e-01 6.06544971e-01 -6.29441559e-01 -1.12232411e+00 -1.07357949e-01 -3.55565399e-01 -1.77367508e-01 1.39625287e-02 1.21467185e+00 -6.48115501e-02 -6.83320701e-01 6.26966804e-02 -3.14647257e-01 -5.82314610e-01 -4.64237928e-01 -4.07178640e-01 3.78934503e-01 1.90783218e-02 7.67150521e-01 -9.92425799e-01 -4.46822196e-01 1.48969829e-01 -6.02338135e-01 -3.53697479e-01 5.01743138e-01 6.94450140e-01 -2.48769760e-01 2.65333086e-01 6.49620056e-01 -1.82912350e-01 2.43658796e-01 -6.24260843e-01 -1.44527316e+00 -3.99724662e-01 -2.30952159e-01 1.92485929e-01 5.03502488e-01 -3.12165678e-01 -7.96208024e-01 -1.56880096e-01 -4.51491177e-01 -2.09130630e-01 -1.51602447e-01 3.28960836e-01 4.58687633e-01 -4.57383126e-01 1.08807957e+00 1.35838374e-01 -2.12988243e-01 -7.91271031e-01 2.16612756e-01 4.81372058e-01 1.38150799e+00 -3.34195375e-01 1.28370810e+00 7.63719320e-01 2.63894409e-01 -7.89889514e-01 -5.23464799e-01 -8.45828474e-01 -2.63860315e-01 -1.89023703e-01 5.46405613e-01 -9.34085906e-01 -9.54523742e-01 7.19142079e-01 -1.11305404e+00 1.23545006e-01 -3.68706137e-01 1.19441319e+00 -3.83699358e-01 3.66790533e-01 -3.19122046e-01 -8.86023760e-01 -1.76681012e-01 -4.97664183e-01 6.26655340e-01 4.71529961e-01 1.63111657e-01 -7.80577242e-01 4.09709781e-01 2.05483586e-01 7.22241879e-01 3.48627567e-01 2.44525343e-01 -8.70539725e-01 -6.63121402e-01 -7.38790393e-01 -1.31605491e-01 -1.26300961e-01 -2.81712919e-01 -2.87086844e-01 -1.08825612e+00 -3.33820999e-01 7.94616997e-01 7.77776493e-03 1.31002843e+00 6.20168149e-01 3.17829758e-01 2.05109939e-01 -2.50094622e-01 1.11064851e+00 1.34617543e+00 1.98590428e-01 2.02811375e-01 6.55479193e-01 1.01442210e-01 9.17434618e-02 1.05226655e-02 7.74577916e-01 -3.48104030e-01 4.65854794e-01 5.20431459e-01 -1.13326376e-02 -2.02994570e-01 2.93287963e-01 -7.85400718e-02 4.33439136e-01 -3.09537917e-01 9.83755589e-02 -1.01666415e+00 6.92493975e-01 -1.65301347e+00 -1.40666056e+00 -7.63579428e-01 2.52159739e+00 7.49560595e-02 5.78156173e-01 1.40121132e-02 -1.53036460e-01 6.41102314e-01 4.38140780e-02 -2.10932791e-01 1.40896933e-02 -3.35410655e-01 7.50281572e-01 1.01538706e+00 4.75078881e-01 -1.24346483e+00 3.19069177e-01 6.56504297e+00 4.66181844e-01 -1.16544437e+00 4.85704124e-01 4.77812439e-02 -4.62159246e-01 -1.14741050e-01 1.50865510e-01 -5.05404353e-01 6.29959404e-01 1.06127214e+00 -2.88516760e-01 4.44389701e-01 4.68025178e-01 3.20851833e-01 -1.62758723e-01 -6.37268007e-01 1.26578009e+00 -4.75291401e-01 -1.73337960e+00 -8.74703825e-01 -1.42271236e-01 4.83410776e-01 1.05217135e+00 -4.21466231e-01 3.60732675e-01 4.22039330e-01 -5.73535919e-01 8.74717176e-01 5.61980188e-01 1.76750928e-01 -6.94246650e-01 8.71146381e-01 4.12393928e-01 -8.54602814e-01 -1.45583138e-01 -6.16236985e-01 -7.00125873e-01 5.66895962e-01 1.27556646e+00 -1.03521860e+00 7.79869795e-01 7.76270986e-01 1.85341150e-01 -4.11449552e-01 1.56142271e+00 -1.99145079e-01 1.07902813e+00 -9.51490045e-01 -4.24278751e-02 3.22969139e-01 2.30301712e-02 1.04087305e+00 1.13272870e+00 4.68387485e-01 -3.88515405e-02 -9.50246975e-02 8.50546837e-01 1.68548133e-02 -5.43012798e-01 -5.25613368e-01 6.45790622e-02 2.16256484e-01 1.31027627e+00 -7.74330676e-01 -3.15639675e-01 -6.01582527e-01 3.14445188e-03 -2.22661253e-02 1.17356144e-01 -6.76855266e-01 -4.88917589e-01 6.20583534e-01 2.05571756e-01 5.30408144e-01 -7.15402186e-01 -4.35680449e-01 -1.33128238e+00 -1.26114815e-01 -2.77755022e-01 4.78572577e-01 -4.60970163e-01 -1.23632908e+00 4.24629867e-01 -2.83290267e-01 -1.12551892e+00 -3.51486742e-01 -6.27385378e-01 -1.16981864e+00 1.01091719e+00 -1.11229587e+00 -5.66625297e-01 -4.72637340e-02 5.15941437e-03 5.83196692e-02 -6.07447565e-01 3.21687192e-01 2.99732953e-01 1.59994680e-02 -1.63629636e-01 6.15237951e-01 1.60484314e-01 5.00507712e-01 -1.18682086e+00 1.15848386e+00 1.30791366e+00 4.24679488e-01 1.16092116e-01 1.45477700e+00 -7.94279933e-01 -1.45358181e+00 -6.98768497e-01 8.22618008e-01 -3.00424814e-01 1.17125893e+00 -6.09727800e-01 -7.48741210e-01 5.77299118e-01 1.90556403e-02 3.09091449e-01 1.11212552e-01 8.88967514e-02 -1.78887486e-01 2.17231959e-02 -1.09574568e+00 -1.45601273e-01 8.13315034e-01 -5.28032482e-01 -1.03151512e+00 7.27837205e-01 3.31752211e-01 -2.44289353e-01 -1.79010704e-01 7.30557621e-01 3.84058744e-01 -1.09179652e+00 1.03142798e+00 -7.39484906e-01 -2.63746232e-01 -3.34994107e-01 2.14971378e-01 -1.14031768e+00 -6.53452277e-01 -1.35311878e+00 -7.17669167e-03 8.72297049e-01 -3.68987061e-02 -7.68729568e-01 9.06002820e-01 9.71065611e-02 -2.12948665e-01 1.92417458e-01 -1.65852308e+00 -1.14105022e+00 1.52051240e-01 -9.06062365e-01 4.26136613e-01 7.92683542e-01 -2.14972869e-01 1.59090478e-02 -3.94563705e-01 7.61430085e-01 9.00006115e-01 4.81940866e-01 5.82105935e-01 -1.45247722e+00 -8.54054272e-01 -5.48892558e-01 -9.20317829e-01 -7.76977181e-01 -4.57824200e-01 -1.06611633e+00 1.79637343e-01 -7.78002620e-01 -5.94853818e-01 -1.63030490e-01 6.32258784e-03 2.28751928e-01 1.54086679e-01 6.06719732e-01 -8.68384615e-02 2.10794106e-01 -3.83844860e-02 2.19493464e-01 3.92157048e-01 3.98163013e-02 1.52069867e-01 3.87267619e-01 1.99637166e-03 8.51647675e-01 6.66785300e-01 -6.10931754e-01 4.06894296e-01 -4.97832328e-01 6.50768518e-01 2.33044177e-01 7.04052985e-01 -1.61850429e+00 4.98557746e-01 1.97624788e-01 4.61968094e-01 -5.81355035e-01 4.37144423e-03 -1.93787187e-01 3.66002291e-01 6.53705239e-01 4.22961086e-01 -3.97802681e-01 2.28668943e-01 2.88806528e-01 -1.80035010e-01 -7.11645663e-01 1.02441871e+00 -3.69277328e-01 -6.84296370e-01 1.38489783e-01 -7.45165229e-01 8.56596902e-02 6.63938105e-01 5.33379853e-01 -3.27878386e-01 -2.79927731e-01 -9.09349918e-01 -8.13763291e-02 2.15732912e-03 3.59873444e-01 3.27503420e-02 -1.08141148e+00 -1.05634379e+00 5.13702154e-01 -2.88350284e-01 -2.97841877e-01 1.42496645e-01 5.93529224e-01 -7.62092113e-01 4.92891282e-01 -6.04812279e-02 -4.75662231e-01 -7.27510691e-01 5.52731872e-01 7.41453230e-01 -9.68491565e-03 -8.09617698e-01 1.06397915e+00 -3.55980545e-02 -4.87322867e-01 -3.01723272e-01 -1.84994996e-01 3.33508760e-01 8.37777779e-02 7.86807477e-01 2.15082213e-01 6.56404793e-01 -4.77733970e-01 -2.81578630e-01 1.25543803e-01 2.71522701e-01 2.49772822e-03 1.56201828e+00 9.74151641e-02 -2.24775430e-02 9.19758454e-02 8.22806776e-01 1.32022694e-01 -1.10144913e+00 -3.12908441e-01 1.90179467e-01 -4.80719239e-01 3.60301703e-01 -8.60979497e-01 -1.14540184e+00 7.05154538e-01 7.44596958e-01 9.54027534e-01 7.16145813e-01 6.92503393e-01 6.73907876e-01 4.01714891e-01 5.91868520e-01 -6.23629272e-01 -4.81554270e-01 6.27138317e-01 5.56403637e-01 -1.01804101e+00 -1.87143445e-01 2.57508487e-01 4.22373354e-01 1.62043619e+00 1.21526532e-01 -2.53743351e-01 6.52726293e-01 4.95917916e-01 -1.76490769e-02 -6.15702927e-01 -6.25964642e-01 -4.45024788e-01 1.58736613e-02 5.15173256e-01 -1.12203427e-01 2.24499032e-01 -4.70652431e-01 5.85468054e-01 -6.68906331e-01 -3.14790428e-01 6.81370556e-01 7.66520798e-01 -8.87089670e-01 -7.61247873e-01 -1.03514385e+00 5.02309322e-01 -6.36964679e-01 -1.24445371e-01 5.73491342e-02 4.91356045e-01 2.74521470e-01 9.89742696e-01 2.61805832e-01 -8.54174644e-02 1.40942872e-01 2.59132326e-01 3.21790993e-01 -1.97952315e-01 -6.75868154e-01 1.37984619e-01 1.64170116e-01 -2.40693182e-01 -1.81688905e-01 -8.25500250e-01 -9.09820795e-01 -2.65670568e-01 -5.16055226e-01 4.23034161e-01 8.17417502e-01 1.19553542e+00 1.34228259e-01 4.90949124e-01 7.08669186e-01 -1.25896764e+00 -6.60148978e-01 -1.13974082e+00 -6.82383716e-01 4.03074026e-01 6.11407995e-01 -6.52190506e-01 -1.03853142e+00 -2.99479038e-01]
[7.562434673309326, 3.1218490600585938]
5f59112c-8adb-407b-bd53-1f114f345fd6
model-agnostic-few-shot-open-set-recognition
2206.09236
null
https://arxiv.org/abs/2206.09236v1
https://arxiv.org/pdf/2206.09236v1.pdf
Model-Agnostic Few-Shot Open-Set Recognition
We tackle the Few-Shot Open-Set Recognition (FSOSR) problem, i.e. classifying instances among a set of classes for which we only have few labeled samples, while simultaneously detecting instances that do not belong to any known class. Departing from existing literature, we focus on developing model-agnostic inference methods that can be plugged into any existing model, regardless of its architecture or its training procedure. Through evaluating the embedding's quality of a variety of models, we quantify the intrinsic difficulty of model-agnostic FSOSR. Furthermore, a fair empirical evaluation suggests that the naive combination of a kNN detector and a prototypical classifier ranks before specialized or complex methods in the inductive setting of FSOSR. These observations motivated us to resort to transduction, as a popular and practical relaxation of standard few-shot learning problems. We introduce an Open Set Transductive Information Maximization method OSTIM, which hallucinates an outlier prototype while maximizing the mutual information between extracted features and assignments. Through extensive experiments spanning 5 datasets, we show that OSTIM surpasses both inductive and existing transductive methods in detecting open-set instances while competing with the strongest transductive methods in classifying closed-set instances. We further show that OSTIM's model agnosticity allows it to successfully leverage the strong expressive abilities of the latest architectures and training strategies without any hyperparameter modification, a promising sign that architectural advances to come will continue to positively impact OSTIM's performances.
['Ismail Ben Ayed', 'Pablo Piantanida', 'Antoine Toubhans', 'Celine Hudelot', 'Myriam Tami', 'Etienne Bennequin', 'Malik Boudiaf']
2022-06-18
null
null
null
null
['open-set-learning']
['miscellaneous']
[ 5.50733507e-01 3.11988026e-01 -4.66649055e-01 -2.07332835e-01 -9.62971091e-01 -6.59183741e-01 7.05160737e-01 1.47672549e-01 -1.59944668e-01 6.88545465e-01 -1.08522199e-01 -1.61300614e-01 -4.36703503e-01 -7.95876026e-01 -6.56917870e-01 -6.88662291e-01 -9.50945467e-02 7.75272191e-01 8.16028863e-02 -3.39623928e-01 1.12097524e-01 2.36993209e-01 -1.88653851e+00 2.40787104e-01 7.53077924e-01 1.12957239e+00 -3.24714303e-01 5.96274793e-01 -2.12339461e-02 6.59281433e-01 -3.18146288e-01 -3.90934259e-01 3.18518192e-01 -1.68569118e-01 -8.71415913e-01 3.15487683e-01 4.11871940e-01 1.10779256e-01 -1.84347689e-01 1.00033689e+00 4.19392377e-01 3.56527865e-01 9.23881352e-01 -1.36151123e+00 -8.44806433e-01 5.51818132e-01 -1.50378928e-01 3.06353688e-01 3.06512594e-01 3.79897594e-01 1.43637109e+00 -1.13506222e+00 6.91719055e-01 9.64890361e-01 8.34789932e-01 6.65888071e-01 -1.66819406e+00 -2.44905323e-01 8.37138966e-02 3.09831291e-01 -1.41513085e+00 -6.10302210e-01 5.78258753e-01 -4.90485430e-01 1.00044560e+00 5.28687418e-01 5.42536736e-01 1.29742885e+00 -2.71812648e-01 8.93264949e-01 1.14876616e+00 -6.78241909e-01 7.80223787e-01 4.55529779e-01 5.61163723e-01 6.38074756e-01 3.03721577e-01 2.43108362e-01 -4.61131185e-01 -4.09095556e-01 2.79166132e-01 1.59622893e-01 -2.26892859e-01 -5.92788756e-01 -9.34193909e-01 8.69406104e-01 3.21278423e-01 3.60393524e-01 -2.83054039e-02 5.83068840e-03 3.56501400e-01 5.54109395e-01 4.79133010e-01 8.90131533e-01 -3.98219854e-01 6.73346594e-02 -7.94882238e-01 -4.40242440e-02 9.85600412e-01 8.51851225e-01 1.07362938e+00 -1.09743424e-01 -2.29642794e-01 8.72628152e-01 1.24677494e-01 9.40855891e-02 6.22594893e-01 -8.29955935e-01 -1.48194686e-01 9.87376511e-01 -1.45919651e-01 -5.32403052e-01 -8.08589533e-02 -4.44475293e-01 -4.33769137e-01 -1.24280497e-01 2.04935268e-01 7.31295049e-02 -1.04853034e+00 1.50684321e+00 3.13640237e-01 4.89747882e-01 2.16087952e-01 6.24595761e-01 6.27734244e-01 5.17645597e-01 -1.51351452e-01 -3.30973059e-01 1.22839391e+00 -6.77641809e-01 -2.48058274e-01 -3.34570050e-01 9.79619741e-01 -2.81025231e-01 1.07372737e+00 3.92340362e-01 -5.90317547e-01 -2.13460490e-01 -1.19077277e+00 1.02986969e-01 -5.47508597e-01 -3.05377394e-01 8.35873723e-01 6.64349139e-01 -7.98137188e-01 6.95105612e-01 -5.83290279e-01 -5.70022225e-01 7.30945706e-01 2.65698910e-01 -4.76434618e-01 -3.22055191e-01 -1.16129994e+00 8.18851173e-01 4.68333542e-01 -5.90350479e-02 -1.03007627e+00 -8.88494253e-01 -6.98305786e-01 2.28352562e-01 8.14842045e-01 -7.49712348e-01 1.10717654e+00 -8.23232949e-01 -1.23702133e+00 1.10834599e+00 -1.75691992e-02 -6.43183053e-01 3.23639125e-01 1.59711558e-02 -4.59925652e-01 3.33518013e-02 1.67972986e-02 3.69287968e-01 8.81217062e-01 -1.28366613e+00 -4.32407320e-01 -4.44574624e-01 1.74474597e-01 -1.29176512e-01 -6.68836176e-01 -3.68473351e-01 -7.74280280e-02 -2.01483428e-01 1.21977001e-01 -9.14124668e-01 -4.04595941e-01 -6.39599413e-02 -7.13139653e-01 -4.58350718e-01 5.88237941e-01 2.52704710e-01 1.20533788e+00 -1.96196902e+00 1.72636896e-01 2.48450056e-01 4.16399688e-01 4.70308691e-01 -1.83884919e-01 5.86245239e-01 -2.10050717e-01 8.12373832e-02 -4.66521204e-01 -8.93777981e-02 2.94483483e-01 4.22864884e-01 -7.18401968e-01 4.84408200e-01 3.27781349e-01 1.10242355e+00 -1.01309049e+00 -4.98084098e-01 2.40027651e-01 1.09271199e-01 -5.82760751e-01 1.28504217e-01 -4.78121936e-01 -7.28311837e-02 -2.64751971e-01 8.95007133e-01 1.93091020e-01 -5.81796050e-01 5.89859076e-02 -2.98764952e-03 2.62361974e-01 -4.68666852e-02 -1.20693064e+00 1.55819809e+00 -3.40088814e-01 4.20902401e-01 -4.54188436e-01 -1.35118985e+00 9.32576120e-01 2.11459979e-01 3.66053820e-01 -2.91073054e-01 2.64758110e-01 3.26148868e-01 -1.09213106e-01 -5.58622658e-01 1.29637241e-01 -6.45355523e-01 -1.53393179e-01 4.04586673e-01 5.62397182e-01 3.15441005e-02 2.35466719e-01 3.13884109e-01 1.44829130e+00 -1.24741055e-01 5.89428186e-01 -2.55504251e-01 2.00145245e-01 8.33889395e-02 5.35089135e-01 1.16359925e+00 -3.44774783e-01 4.65204060e-01 4.66208130e-01 -4.94665325e-01 -7.72708714e-01 -1.36762249e+00 -4.71322834e-01 1.39228165e+00 -1.19926021e-01 -3.93157840e-01 -3.64114165e-01 -8.63051057e-01 9.46390033e-02 7.06548393e-01 -9.18858826e-01 -5.61445296e-01 4.53200750e-02 -9.87424493e-01 6.01313591e-01 3.18824500e-01 -1.96581930e-01 -7.45835185e-01 -3.60231817e-01 3.02442256e-02 8.26768205e-02 -7.75123596e-01 1.13421679e-03 7.66258121e-01 -7.59100139e-01 -1.11339200e+00 -3.12427878e-01 -5.41647673e-01 5.56755841e-01 1.59801319e-01 1.07846248e+00 -1.41208217e-01 -6.59702659e-01 6.10326409e-01 -4.64232832e-01 -4.89145398e-01 -3.37871283e-01 1.03013180e-01 2.82302916e-01 2.89604455e-01 8.08059812e-01 -7.53547788e-01 -1.71485692e-01 1.64616317e-01 -9.90586758e-01 -4.18313146e-01 5.15544891e-01 1.04775655e+00 4.66067910e-01 -2.27660924e-01 9.07895207e-01 -1.29092300e+00 4.44937438e-01 -7.96398818e-01 -3.44181776e-01 4.67170119e-01 -9.41006184e-01 2.63146520e-01 6.53176904e-01 -4.90921736e-01 -7.79701948e-01 -3.25433910e-02 1.93482980e-01 -7.19503164e-01 -2.34081566e-01 4.25135434e-01 -1.61877815e-02 -2.70047765e-02 1.25987506e+00 2.15496823e-01 -1.39835134e-01 -2.52646625e-01 5.61336339e-01 7.66473770e-01 4.08503354e-01 -5.74186802e-01 9.39545691e-01 7.25801706e-01 -1.54557362e-01 -1.04174840e+00 -1.45203757e+00 -8.01034570e-01 -5.73990762e-01 -5.67966066e-02 4.39089864e-01 -7.64614999e-01 -6.66492581e-01 1.58159097e-03 -6.85589671e-01 -3.04146130e-02 -8.96599948e-01 2.03392506e-01 -6.68336928e-01 2.79147178e-01 -3.92996788e-01 -9.62716281e-01 -1.36658922e-01 -7.47102141e-01 1.00107384e+00 2.12114174e-02 -3.55322063e-01 -1.14541090e+00 4.00823772e-01 4.58365798e-01 1.54229030e-01 1.20252810e-01 8.41795266e-01 -1.32634175e+00 -4.23637480e-01 -5.53763926e-01 6.01046234e-02 4.08325166e-01 -1.67114124e-01 -9.25644040e-02 -1.46857631e+00 -3.07839006e-01 -3.39191295e-02 -8.61002982e-01 1.26424718e+00 6.17153160e-02 9.00869429e-01 -2.74375588e-01 -3.82230878e-01 5.13196230e-01 1.52138793e+00 -1.48963884e-01 4.92043912e-01 1.24486722e-01 5.74330032e-01 5.27540743e-01 4.04061258e-01 5.70690632e-01 -9.05951951e-03 3.36099714e-01 2.38678336e-01 1.50456414e-01 2.39655256e-01 -3.42513472e-01 3.45474899e-01 5.60606778e-01 5.53287305e-02 -1.36169285e-01 -8.31155598e-01 5.14726639e-01 -1.86743307e+00 -1.16574609e+00 2.82684237e-01 2.28859186e+00 8.46535504e-01 3.36701274e-01 3.53759304e-02 3.40531021e-01 6.66342378e-01 8.07461515e-02 -7.57024050e-01 -3.82350981e-01 -8.49994421e-02 3.98228139e-01 2.59935975e-01 2.18285009e-01 -1.09783423e+00 8.01905394e-01 6.59891367e+00 1.05495477e+00 -8.12546611e-01 -1.14946999e-03 5.71396410e-01 -3.15797269e-01 -4.57472622e-01 3.20763230e-01 -8.12179029e-01 1.72358379e-01 1.04038930e+00 -2.12069973e-01 5.80395222e-01 1.13497233e+00 -3.69872332e-01 2.33064834e-02 -1.77577412e+00 7.99001455e-01 2.38068134e-01 -1.61592269e+00 8.44209567e-02 7.93560743e-02 8.19091022e-01 1.11225754e-01 7.16333613e-02 9.06399548e-01 4.87218946e-01 -1.14819300e+00 3.80434275e-01 5.37156165e-01 5.71197808e-01 -3.97755623e-01 4.81960654e-01 4.26027745e-01 -7.86695957e-01 -4.44051713e-01 -4.38886732e-01 -1.94312900e-01 -1.80166751e-01 6.04313135e-01 -1.00165379e+00 3.87935311e-01 3.49927127e-01 7.20221937e-01 -5.48230886e-01 9.17158723e-01 4.51764725e-02 8.53572369e-01 -4.12644088e-01 -1.25154015e-02 2.28507742e-01 -1.06948260e-02 7.59310246e-01 1.10662174e+00 -6.03597909e-02 7.72631094e-02 3.72574061e-01 1.00002444e+00 -1.88548133e-01 -2.90532559e-02 -1.05001867e+00 -2.36150354e-01 5.91213167e-01 1.27520967e+00 -5.85593998e-01 -5.51842153e-01 -2.55980283e-01 6.52132034e-01 5.78134716e-01 1.77733406e-01 -6.18891418e-01 -2.95785934e-01 4.47519124e-01 5.14944382e-02 3.73776555e-01 4.36406046e-01 -2.74194330e-01 -1.46643794e+00 4.92124679e-03 -6.81734562e-01 7.98423588e-01 -4.64399517e-01 -1.80145085e+00 2.82076389e-01 -1.35320455e-01 -1.33412254e+00 -1.24875180e-01 -6.14921689e-01 -7.07618237e-01 2.44972512e-01 -1.37468755e+00 -1.02602971e+00 1.30752623e-01 4.23719525e-01 3.71240705e-01 -5.57518378e-02 9.89329755e-01 7.46075064e-02 -7.54815578e-01 6.71653211e-01 2.12140381e-01 -6.41828356e-03 5.10222673e-01 -1.34808075e+00 -9.66843665e-02 7.71944582e-01 4.44260657e-01 7.54197299e-01 8.13473105e-01 -2.49830633e-01 -1.59485579e+00 -1.24968636e+00 6.21854484e-01 -7.72542179e-01 1.07398808e+00 -4.38343495e-01 -9.81185079e-01 9.29579496e-01 -4.13517267e-01 5.82922280e-01 1.06468785e+00 5.64072371e-01 -8.28481913e-01 -1.09440044e-01 -1.02406049e+00 4.33060914e-01 1.17913210e+00 -6.44259512e-01 -9.89109695e-01 4.06465262e-01 7.30043411e-01 2.74339288e-01 -9.38394904e-01 4.97965097e-01 2.54664481e-01 -7.93842554e-01 1.00571895e+00 -1.25195241e+00 4.22046900e-01 -6.39968291e-02 -5.99225461e-01 -1.19448054e+00 -3.29881042e-01 -6.44768536e-01 -3.63811195e-01 1.12683392e+00 4.50689703e-01 -7.17755318e-01 7.53218353e-01 6.24061465e-01 -1.48536831e-01 -9.56694961e-01 -1.06084394e+00 -1.04612076e+00 -6.90677762e-02 -6.35858357e-01 7.24705160e-02 1.12072289e+00 4.74380791e-01 7.53158391e-01 -3.01811576e-01 1.05872219e-02 7.82291234e-01 3.63988638e-01 6.94929302e-01 -1.59072649e+00 -5.75029433e-01 -2.45167598e-01 -7.28891075e-01 -4.99528795e-01 2.30720416e-01 -1.37666345e+00 4.21294272e-02 -1.13076830e+00 4.28786010e-01 -5.55576026e-01 -5.50944388e-01 6.37884021e-01 -1.21852927e-01 4.77094203e-01 -7.21050650e-02 1.59414977e-01 -1.05388904e+00 5.94937861e-01 7.42086709e-01 -2.25239560e-01 -3.35342467e-01 -1.15710385e-01 -9.55199063e-01 8.27115357e-01 3.91749531e-01 -3.95695746e-01 -3.50006968e-01 1.95571125e-01 3.26093793e-01 -8.93396363e-02 5.69528043e-01 -9.74889755e-01 1.54153287e-01 -1.42267793e-01 1.55002594e-01 7.64447227e-02 3.75286549e-01 -5.71062744e-01 -1.82746932e-01 3.78972322e-01 -5.82004070e-01 -7.86446512e-01 -1.59658760e-01 1.02971470e+00 6.44185916e-02 -4.23860162e-01 9.45178807e-01 -2.13372499e-01 -9.44595933e-01 4.31486398e-01 -3.36565137e-01 4.10361648e-01 1.32849061e+00 -3.86829317e-01 -3.39472085e-01 1.45727992e-01 -1.02759778e+00 7.26205856e-02 4.09646541e-01 3.24982911e-01 4.76002306e-01 -1.14785016e+00 -5.87767363e-01 1.90095708e-01 8.18784356e-01 -2.02763751e-01 2.74122834e-01 9.48005557e-01 1.62883028e-01 3.23004484e-01 1.52694657e-01 -7.79276729e-01 -8.42060447e-01 8.99886370e-01 1.90235987e-01 -1.63115367e-01 -5.38687766e-01 8.38486195e-01 1.40708163e-01 -6.88967288e-01 1.79318905e-01 -1.03875753e-02 6.90791011e-02 2.47008100e-01 4.58162278e-01 4.68046784e-01 2.31725007e-01 -1.97445631e-01 -3.04392934e-01 1.06860377e-01 -2.25519359e-01 2.25906476e-01 1.40558624e+00 1.17147878e-01 3.27032642e-03 1.03247213e+00 1.26343405e+00 -4.56795543e-01 -9.09690559e-01 -6.03942215e-01 1.61925793e-01 -3.05845320e-01 -5.60686290e-02 -7.41715789e-01 -3.49193990e-01 8.00161242e-01 5.08311450e-01 4.08405662e-01 7.88245261e-01 4.22569782e-01 5.19487083e-01 8.15638423e-01 3.77930999e-01 -1.15644670e+00 2.06870645e-01 4.47573900e-01 4.07495230e-01 -1.52653933e+00 -1.77594677e-01 -3.00068796e-01 -4.59009945e-01 1.01318359e+00 4.60513771e-01 -3.08316052e-01 5.46128213e-01 1.30954564e-01 -1.85846895e-01 -3.74364525e-01 -1.31689739e+00 -5.12581527e-01 2.64234155e-01 5.40621638e-01 5.22788242e-02 1.78365916e-01 1.09325409e-01 5.70283175e-01 6.56447485e-02 5.06440513e-02 4.52009439e-01 8.42907369e-01 -8.27589035e-01 -6.25083387e-01 -2.17991784e-01 9.09039736e-01 -8.17805678e-02 -1.03304327e-01 -4.98729289e-01 5.54513454e-01 1.01130195e-01 9.03495610e-01 -7.39691257e-02 -5.07120788e-01 1.24478199e-01 4.72291827e-01 4.85192567e-01 -9.60313797e-01 -3.03411871e-01 -3.47551703e-01 4.77747582e-02 -5.23308396e-01 -1.22728370e-01 -4.88672793e-01 -9.27633286e-01 3.55428569e-02 -6.61926270e-01 1.39658108e-01 8.83045495e-02 1.24389124e+00 3.38357747e-01 1.70559183e-01 7.42722869e-01 -6.51027381e-01 -1.14888060e+00 -8.27402174e-01 -5.93203306e-01 4.23808962e-01 2.62308031e-01 -7.70128667e-01 -6.74239874e-01 -4.67996784e-02]
[9.780423164367676, 2.9995365142822266]
5ffc6fc4-918e-432d-9845-973b5ae7289c
ppg-based-heart-rate-estimation-with
2303.13636
null
https://arxiv.org/abs/2303.13636v1
https://arxiv.org/pdf/2303.13636v1.pdf
PPG-based Heart Rate Estimation with Efficient Sensor Sampling and Learning Models
Recent studies showed that Photoplethysmography (PPG) sensors embedded in wearable devices can estimate heart rate (HR) with high accuracy. However, despite of prior research efforts, applying PPG sensor based HR estimation to embedded devices still faces challenges due to the energy-intensive high-frequency PPG sampling and the resource-intensive machine-learning models. In this work, we aim to explore HR estimation techniques that are more suitable for lower-power and resource-constrained embedded devices. More specifically, we seek to design techniques that could provide high-accuracy HR estimation with low-frequency PPG sampling, small model size, and fast inference time. First, we show that by combining signal processing and ML, it is possible to reduce the PPG sampling frequency from 125 Hz to only 25 Hz while providing higher HR estimation accuracy. This combination also helps to reduce the ML model feature size, leading to smaller models. Additionally, we present a comprehensive analysis on different ML models and feature sizes to compare their accuracy, model size, and inference time. The models explored include Decision Tree (DT), Random Forest (RF), K-nearest neighbor (KNN), Support vector machines (SVM), and Multi-layer perceptron (MLP). Experiments were conducted using both a widely-utilized dataset and our self-collected dataset. The experimental results show that our method by combining signal processing and ML had only 5% error for HR estimation using low-frequency PPG data. Moreover, our analysis showed that DT models with 10 to 20 input features usually have good accuracy, while are several magnitude smaller in model sizes and faster in inference time.
['Dakai Zhu', 'Jing Wang', 'Keying Ye', 'Wei Wang', 'Mimi Xie', 'Jingye Xu', 'Yuntong Zhang']
2023-03-23
null
null
null
null
['photoplethysmography-ppg', 'heart-rate-estimation']
['medical', 'medical']
[ 2.35193938e-01 -1.10733412e-01 -3.98166329e-01 -1.88801229e-01 -2.72570044e-01 -7.14599863e-02 -3.98272216e-01 1.93865895e-02 -1.88746989e-01 8.81190300e-01 -8.34919419e-03 -2.96100110e-01 -1.74256209e-02 -7.72540867e-01 -6.16401806e-02 -5.77765465e-01 -2.01856300e-01 -5.01424909e-01 -1.92141309e-01 3.87434542e-01 8.56129080e-02 4.73633260e-01 -1.56488037e+00 -1.96934879e-01 9.74354506e-01 1.35663652e+00 -1.81511864e-01 7.38580942e-01 5.64879477e-01 4.01986897e-01 -6.40016198e-01 1.09923951e-01 1.51244894e-01 -7.26450384e-01 -2.49364033e-01 -5.08812964e-01 1.91090688e-01 -3.63741189e-01 -2.40527362e-01 3.44831854e-01 1.15037870e+00 6.35199025e-02 1.81730419e-01 -1.20560324e+00 -6.74659312e-02 2.94985026e-01 -2.62153924e-01 1.51641577e-01 5.47287881e-01 1.89280555e-01 2.56434083e-01 -4.80471134e-01 -3.28116529e-02 6.77297890e-01 1.21622372e+00 5.79955041e-01 -1.24240506e+00 -8.16824853e-01 -5.80003202e-01 3.72049659e-01 -1.62956500e+00 -4.48648572e-01 1.12187314e+00 -8.96965861e-02 1.13676953e+00 7.61804640e-01 1.13841462e+00 7.74583399e-01 5.82691252e-01 9.77313742e-02 1.61716437e+00 -6.32842422e-01 3.71080905e-01 3.44405472e-01 1.53781444e-01 5.42304218e-01 5.00702560e-01 2.13087410e-01 -5.74051619e-01 -3.04886520e-01 9.33248460e-01 3.52222845e-02 -5.59301674e-01 4.08927381e-01 -8.31106365e-01 5.80546319e-01 5.03462479e-02 4.22599554e-01 -4.35781151e-01 9.51287746e-02 2.63179183e-01 -1.00592040e-02 1.55503452e-01 5.78513563e-01 -5.30439675e-01 -4.10085052e-01 -8.98274720e-01 -2.01971337e-01 1.32616711e+00 4.67237949e-01 4.97178376e-01 1.76679567e-01 -3.29483628e-01 8.07904482e-01 2.40734428e-01 7.00119913e-01 6.81118786e-01 -9.68250334e-01 1.43445969e-01 4.28508967e-01 -1.52614070e-02 -1.25633979e+00 -8.67091119e-01 -2.37341627e-01 -1.26059496e+00 -2.35802084e-01 2.50334918e-01 -3.90680522e-01 -4.65466559e-01 1.45142961e+00 2.48757526e-01 3.12475473e-01 -1.07354335e-01 9.39164639e-01 8.99499655e-01 3.95871192e-01 3.21776927e-01 -7.81560957e-01 1.57309258e+00 -5.27619958e-01 -8.90241027e-01 -1.31318077e-01 3.55493903e-01 -4.11490053e-01 1.11196637e+00 3.83113652e-01 -7.84541249e-01 -9.29927349e-01 -1.18387413e+00 4.35547233e-02 -1.83472093e-02 4.00102437e-01 7.06372797e-01 1.33026624e+00 -7.47073948e-01 1.02024758e+00 -7.40047038e-01 -4.33671594e-01 2.85062585e-02 2.73167908e-01 1.09397553e-01 3.52727890e-01 -1.44830012e+00 1.00126255e+00 5.59910536e-02 4.23451453e-01 1.06868982e-01 -6.98548436e-01 -8.65865827e-01 1.50911942e-01 -8.40260535e-02 -6.47195995e-01 6.79678261e-01 -3.69475245e-01 -2.13111401e+00 3.50425363e-01 -3.47338140e-01 -4.20881569e-01 1.69287235e-01 -2.33291924e-01 -8.46648157e-01 4.19092953e-01 -7.25775898e-01 1.73761770e-01 7.88526773e-01 -4.80350077e-01 1.14910439e-01 -4.07311618e-01 -5.25585830e-01 4.52746674e-02 -4.44887906e-01 -2.98619360e-01 1.99116573e-01 -2.09642202e-01 2.31637493e-01 -9.49011087e-01 -9.88408700e-02 -9.00351703e-02 -1.64478496e-01 -1.43983159e-02 5.93393624e-01 -1.05558670e+00 1.72055471e+00 -1.98604178e+00 -4.25138652e-01 3.47880363e-01 2.16137260e-01 3.57991040e-01 5.82896709e-01 2.72197425e-01 1.51970670e-01 1.37556404e-01 5.12218662e-02 8.48188326e-02 -2.38982484e-01 6.13844357e-02 1.52984858e-02 5.49406052e-01 -2.83420831e-01 8.72701168e-01 -3.36184144e-01 -6.47802114e-01 8.48230481e-01 6.20780587e-01 -1.31671488e-01 3.14820677e-01 5.81280053e-01 3.32242459e-01 -3.21366005e-02 8.32547188e-01 4.28361863e-01 -5.88875674e-02 3.47160876e-01 -8.60313237e-01 -1.32364267e-03 1.36376604e-01 -1.24944651e+00 1.17005825e+00 -7.29588032e-01 4.58778590e-01 -4.54724789e-01 -6.77798271e-01 1.56682396e+00 4.48289096e-01 5.82131326e-01 -7.99864054e-01 1.42457440e-01 2.39620730e-01 -1.48979966e-02 -8.87861729e-01 2.99836062e-02 -3.15692008e-01 9.55692977e-02 1.28625229e-01 -2.11457282e-01 3.61620151e-02 -3.03518593e-01 -5.78270555e-01 8.18407059e-01 2.24676244e-02 7.16947913e-01 -2.41898984e-01 4.86986041e-01 -5.01243174e-01 7.44408846e-01 6.30654156e-01 -5.61481833e-01 3.41496915e-01 8.50885138e-02 -6.20367646e-01 -4.99662250e-01 -6.85849905e-01 -3.86729091e-01 2.61833042e-01 1.86831549e-01 -3.55679452e-01 -3.86509567e-01 -2.22355896e-03 1.11306667e-01 6.86617911e-01 -2.06863269e-01 -3.93533796e-01 -5.47137320e-01 -9.74098265e-01 8.62679720e-01 6.14689171e-01 8.82134676e-01 -1.11671591e+00 -1.23987842e+00 2.55613118e-01 -2.17651665e-01 -1.00727403e+00 -1.25588655e-01 2.49985382e-02 -1.34693921e+00 -9.17423844e-01 -5.14836133e-01 -2.66525567e-01 1.94209650e-01 -3.41526717e-01 8.63870919e-01 -1.35872215e-01 -7.47155607e-01 3.25784236e-01 -9.23573822e-02 -5.41573942e-01 1.81312218e-01 -1.04728267e-01 3.40404063e-01 -2.38187447e-01 5.43714583e-01 -6.97111189e-01 -1.09849238e+00 2.70189673e-01 -4.87081185e-02 -3.68317366e-02 6.60581112e-01 5.11356473e-01 5.34318328e-01 -7.91292638e-02 8.43848705e-01 -3.98044258e-01 6.78358018e-01 -1.65176660e-01 -2.79692769e-01 1.39596492e-01 -1.26036549e+00 -3.60008448e-01 6.87949061e-01 -6.70152485e-01 -7.51778424e-01 3.51928547e-02 -3.29350382e-02 -3.81517678e-01 -1.26612946e-01 2.99699217e-01 6.36636466e-02 -2.91312605e-01 9.71556723e-01 3.40429038e-01 3.89205784e-01 -4.13229138e-01 -8.88920948e-02 8.16929162e-01 3.69407296e-01 -3.50609779e-01 3.61477643e-01 -7.20140487e-02 3.92067194e-01 -1.06809211e+00 -3.03419828e-01 -2.08210319e-01 -4.33930635e-01 -4.92670238e-01 7.23544478e-01 -9.61380303e-01 -1.23729968e+00 5.84480524e-01 -4.90988374e-01 -2.97293127e-01 -6.92818537e-02 9.87105250e-01 -4.31189239e-01 4.55447406e-01 -7.99750686e-01 -1.43832695e+00 -1.19822490e+00 -4.69550639e-01 5.50591946e-01 7.83582509e-01 -5.86500406e-01 -9.02883053e-01 -1.30181044e-01 4.40459520e-01 8.15234303e-01 6.73682332e-01 5.22504032e-01 4.21612598e-02 8.88412148e-02 -3.96086037e-01 1.40841771e-02 3.98037851e-01 2.86625534e-01 -2.55905241e-01 -1.22310388e+00 -2.88373917e-01 3.98381919e-01 -1.07087880e-01 2.91856706e-01 5.93220949e-01 1.49890471e+00 -4.77603227e-01 -3.36447448e-01 6.39481068e-01 1.64603674e+00 2.84334600e-01 1.01511657e+00 -1.02887608e-01 4.77730244e-01 3.12640250e-01 5.15021682e-01 4.48272258e-01 2.59497702e-01 6.08592629e-01 -2.29411379e-01 -5.21576226e-01 2.64837109e-02 -2.56834418e-01 1.97047874e-01 7.70407200e-01 -5.44336736e-01 3.11261654e-01 -4.52383518e-01 4.35019843e-02 -1.46848381e+00 -8.85643959e-01 -1.38839081e-01 2.61198711e+00 1.01049113e+00 -2.10488528e-01 3.91943038e-01 6.44204736e-01 5.43816388e-01 -1.49599224e-01 -6.76471353e-01 -7.95446992e-01 2.58146852e-01 5.32718956e-01 6.61559522e-01 1.10893454e-02 -8.80520999e-01 -1.74590647e-02 6.56658506e+00 3.30940872e-01 -1.37995982e+00 -1.88114285e-01 7.59935856e-01 -1.36961862e-01 3.04388046e-01 -1.64977685e-01 -8.16987276e-01 8.53489578e-01 1.47287369e+00 -4.21400130e-01 4.10718203e-01 9.17867303e-01 4.64893788e-01 -4.50698495e-01 -9.25316811e-01 1.50209355e+00 6.34149984e-02 -9.39028740e-01 -7.67848849e-01 -7.55971298e-02 5.97842410e-03 -5.89065313e-01 -6.29892707e-01 3.32721353e-01 -9.23911691e-01 -9.22051847e-01 -1.45287111e-01 7.60715783e-01 1.08410382e+00 -5.10145962e-01 8.90993059e-01 9.72653553e-02 -1.29754984e+00 -6.93059564e-02 -3.01740885e-01 -5.72435915e-01 -6.12123385e-02 1.14016771e+00 -6.55290365e-01 4.93567884e-01 8.38119626e-01 4.05717582e-01 -3.85644585e-01 1.12766004e+00 -1.13146238e-01 7.59503722e-01 -6.29552424e-01 -4.83048171e-01 -8.42000127e-01 -1.20235130e-01 5.02857417e-02 9.83711839e-01 4.46328759e-01 6.41807020e-01 6.67032301e-02 8.17095757e-01 4.44817394e-01 8.89979377e-02 -3.74935985e-01 2.75494337e-01 7.36596406e-01 1.40675294e+00 -4.24375057e-01 -1.17955640e-01 -3.87079448e-01 5.98017335e-01 -3.00341457e-01 1.76770136e-01 -1.05811167e+00 -1.01258850e+00 3.46658081e-01 3.22983474e-01 -3.77121925e-01 -2.04462141e-01 -7.88339972e-01 -9.11001205e-01 1.51747972e-01 -4.60189223e-01 4.08633947e-01 -5.28068900e-01 -1.04211509e+00 4.02379781e-01 -4.52535367e-03 -1.29306924e+00 -2.16609851e-01 -1.49656072e-01 -5.94473779e-01 1.22443843e+00 -1.40778089e+00 -4.34319198e-01 -7.80152261e-01 5.12903333e-01 7.99362361e-02 4.63841677e-01 1.11750329e+00 4.38601196e-01 -8.61073613e-01 7.72289574e-01 -5.17660081e-01 -1.83645517e-01 5.56516826e-01 -8.89964044e-01 -2.79479742e-01 5.30148387e-01 -5.49780190e-01 6.68110728e-01 4.69866872e-01 -4.45137382e-01 -1.80052030e+00 -7.69398391e-01 1.05573010e+00 -3.03816870e-02 -4.28465828e-02 -1.80035964e-01 -8.73759329e-01 -2.02227291e-02 -3.24761510e-01 2.86090404e-01 9.02481675e-01 1.14540190e-01 1.68836892e-01 -7.14346528e-01 -1.66365361e+00 2.88906008e-01 6.25112832e-01 -4.69215780e-01 -4.25288796e-01 -8.61893222e-02 2.36272410e-01 -5.85044444e-01 -1.69720995e+00 7.05636680e-01 1.18533659e+00 -6.73825741e-01 8.47080708e-01 2.67094433e-01 -3.94015387e-02 -2.76889473e-01 2.30823830e-01 -8.61452937e-01 -3.39982182e-01 -6.50380611e-01 -7.08448946e-01 1.13565397e+00 1.57739058e-01 -1.06234658e+00 5.69059730e-01 1.44356048e+00 2.67611653e-01 -9.62760150e-01 -8.54005754e-01 -8.51575315e-01 -6.63170636e-01 -8.32553059e-02 1.82023287e-01 7.91691601e-01 6.65488183e-01 2.51493722e-01 -7.84037113e-01 -4.42142710e-02 7.24332809e-01 3.38174105e-01 5.02080500e-01 -1.22018456e+00 -3.51644307e-01 9.37889293e-02 -3.84504884e-01 -4.04736966e-01 -4.10677105e-01 -2.34683976e-01 -7.01551698e-03 -1.34834075e+00 2.07419991e-02 -5.37529469e-01 -5.37677824e-01 8.15532267e-01 -3.87813807e-01 5.12849271e-01 1.69736221e-02 5.62054403e-02 1.60641894e-01 2.61397719e-01 8.39426398e-01 3.45263988e-01 -9.70177114e-01 2.38270581e-01 -4.49563831e-01 2.98799932e-01 1.09207404e+00 -3.18761975e-01 -3.76104236e-01 3.77911091e-01 -2.20271438e-01 4.19690102e-01 3.45472097e-01 -1.38944769e+00 1.92475140e-01 -6.91546500e-02 1.02054644e+00 -2.09994569e-01 3.58023554e-01 -6.64739192e-01 6.64438426e-01 9.68104005e-01 9.49949920e-02 -3.49680305e-01 4.19494182e-01 1.16568021e-01 1.12758130e-01 1.14236481e-01 7.80343294e-01 2.94366535e-02 -4.07486111e-01 -2.98122540e-02 -1.91331625e-01 -4.66238290e-01 9.72999930e-01 -5.63098311e-01 -3.73007476e-01 -3.27865064e-01 -6.13476157e-01 -1.25586897e-01 6.38638437e-02 2.41287112e-01 6.67181969e-01 -1.17155933e+00 -1.67223766e-01 3.67708415e-01 -1.08253203e-01 -5.96142352e-01 5.65360427e-01 1.19702518e+00 -3.61281335e-01 4.97694790e-01 -4.71691787e-01 -5.66584289e-01 -1.39323485e+00 3.39802772e-01 5.31256735e-01 3.13146822e-02 -7.70571411e-01 1.18079357e-01 -8.19724321e-01 6.51605874e-02 1.52880475e-01 -4.84127194e-01 -2.80455530e-01 -1.90319628e-01 4.82081681e-01 9.49708521e-01 3.83301228e-02 1.36216015e-01 -5.53831816e-01 9.15862918e-01 8.09875488e-01 3.76813412e-01 8.12104344e-01 -2.38058284e-01 9.96172130e-02 5.42236209e-01 1.00613225e+00 -1.46030545e-01 -7.61776030e-01 1.81367159e-01 -2.68822283e-01 -4.65704381e-01 2.58344144e-01 -9.46189880e-01 -9.86464500e-01 5.71967900e-01 1.16266966e+00 1.86914191e-01 1.77469957e+00 -5.50163865e-01 9.51147318e-01 1.62888989e-01 6.42190695e-01 -1.19458997e+00 -3.95842969e-01 -4.19567496e-01 4.54221696e-01 -8.73494804e-01 4.82888162e-01 -5.94585359e-01 -4.97770101e-01 1.31846416e+00 8.43424141e-01 9.18983594e-02 8.36842775e-01 7.24855512e-02 -5.28969392e-02 2.17601746e-01 -4.94744658e-01 1.24501206e-01 3.19183320e-01 5.44268966e-01 5.34065485e-01 3.38858873e-01 -9.67993677e-01 8.31848979e-01 -1.58042580e-01 8.07561815e-01 4.55020338e-01 7.07952321e-01 -3.92825872e-01 -7.50419080e-01 -2.14961186e-01 9.06953454e-01 -5.06299973e-01 3.40937898e-02 1.99512467e-01 5.88905692e-01 -3.18495668e-02 1.30310249e+00 -1.12405837e-01 -6.43071651e-01 5.17821908e-01 3.14019471e-01 5.95931828e-01 -1.35216191e-01 -6.39237344e-01 -1.22401277e-02 1.62846804e-01 -8.71830046e-01 -3.97197872e-01 -3.39608818e-01 -9.68501389e-01 -2.88720787e-01 -3.64368439e-01 8.98553990e-03 9.43976343e-01 6.61276400e-01 7.33947277e-01 3.74952286e-01 7.85372257e-01 -5.87616861e-01 -3.15010309e-01 -1.16140604e+00 -8.55619967e-01 -4.18643877e-02 5.65439798e-02 -3.88996661e-01 -6.86616123e-01 -1.14841379e-01]
[13.924928665161133, 3.050413131713867]
7d4b9959-1036-4e84-a39a-1dec10ad95a1
fusing-multimodal-signals-on-hyper-complex
2306.13968
null
https://arxiv.org/abs/2306.13968v1
https://arxiv.org/pdf/2306.13968v1.pdf
Fusing Multimodal Signals on Hyper-complex Space for Extreme Abstractive Text Summarization (TL;DR) of Scientific Contents
The realm of scientific text summarization has experienced remarkable progress due to the availability of annotated brief summaries and ample data. However, the utilization of multiple input modalities, such as videos and audio, has yet to be thoroughly explored. At present, scientific multimodal-input-based text summarization systems tend to employ longer target summaries like abstracts, leading to an underwhelming performance in the task of text summarization. In this paper, we deal with a novel task of extreme abstractive text summarization (aka TL;DR generation) by leveraging multiple input modalities. To this end, we introduce mTLDR, a first-of-its-kind dataset for the aforementioned task, comprising videos, audio, and text, along with both author-composed summaries and expert-annotated summaries. The mTLDR dataset accompanies a total of 4,182 instances collected from various academic conference proceedings, such as ICLR, ACL, and CVPR. Subsequently, we present mTLDRgen, an encoder-decoder-based model that employs a novel dual-fused hyper-complex Transformer combined with a Wasserstein Riemannian Encoder Transformer, to dexterously capture the intricacies between different modalities in a hyper-complex latent geometric space. The hyper-complex Transformer captures the intrinsic properties between the modalities, while the Wasserstein Riemannian Encoder Transformer captures the latent structure of the modalities in the latent space geometry, thereby enabling the model to produce diverse sentences. mTLDRgen outperforms 20 baselines on mTLDR as well as another non-scientific dataset (How2) across three Rouge-based evaluation measures. Furthermore, based on the qualitative metrics, BERTScore and FEQA, and human evaluations, we demonstrate that the summaries generated by mTLDRgen are fluent and congruent to the original source material.
['Tanmoy Chakraborty', 'Vikram Goyal', 'Yash Kumar Atri']
2023-06-24
null
null
null
null
['abstractive-text-summarization', 'text-summarization']
['natural-language-processing', 'natural-language-processing']
[ 2.23623663e-01 6.71175122e-02 7.29071274e-02 -1.30896941e-01 -1.41755593e+00 -6.41889334e-01 8.15450490e-01 1.33068994e-01 -1.98195204e-02 6.92460954e-01 1.02875757e+00 1.76993594e-01 -3.94529104e-02 -2.07228824e-01 -6.15056932e-01 -5.74512601e-01 8.37254003e-02 2.60794401e-01 -3.60871494e-01 -1.51543155e-01 3.80105078e-01 2.97884922e-02 -1.36298549e+00 4.90208596e-01 1.25402761e+00 6.56984985e-01 1.40555486e-01 1.01832879e+00 -1.36378212e-02 5.54987788e-01 -9.16053534e-01 -7.83373535e-01 -2.49945477e-01 -7.07908869e-01 -6.30379915e-01 8.91302228e-02 7.67751396e-01 -2.41125330e-01 -6.83583498e-01 7.72530377e-01 8.27700913e-01 2.76219249e-01 1.04559839e+00 -9.90272105e-01 -7.91789651e-01 8.49579275e-01 -5.26141942e-01 -6.23082668e-02 9.79499161e-01 1.31100073e-01 1.37168419e+00 -1.03355455e+00 7.77994335e-01 1.39959967e+00 3.81430149e-01 4.81000811e-01 -1.01776850e+00 -2.70410240e-01 2.47417670e-02 6.82128817e-02 -1.10574055e+00 -5.19415259e-01 7.79133320e-01 -2.95558184e-01 7.17709899e-01 5.55788636e-01 5.66081345e-01 1.59071159e+00 1.82008952e-01 1.28784692e+00 5.73823094e-01 -5.67099117e-02 4.94517498e-02 -1.21995352e-01 3.21269548e-03 4.28391308e-01 1.78780332e-01 -4.84730691e-01 -1.00282729e+00 -1.18660651e-01 2.64528245e-01 -1.22111641e-01 -7.33205020e-01 4.46650833e-02 -1.72789907e+00 5.48984885e-01 1.03456624e-01 1.89643666e-01 -3.06012511e-01 -1.45464376e-01 7.27974474e-01 2.07437024e-01 6.05565310e-01 6.63853467e-01 6.20267959e-03 -5.58884978e-01 -1.21758080e+00 4.85113829e-01 1.04632795e+00 1.22363520e+00 2.25381926e-01 9.22212079e-02 -7.80841231e-01 8.22094738e-01 1.72291979e-01 7.42932081e-01 4.38744664e-01 -9.89356041e-01 9.66727912e-01 6.23638093e-01 -1.58317268e-01 -1.13468552e+00 -2.09172338e-01 -3.13525915e-01 -1.22672784e+00 -6.64212286e-01 -7.40863988e-03 -1.52560607e-01 -4.57061172e-01 1.61332285e+00 -5.39318770e-02 -3.29217352e-02 5.42678952e-01 6.91341698e-01 1.57435274e+00 1.10602736e+00 -3.45964700e-01 -2.62631625e-01 1.20039594e+00 -9.37041759e-01 -1.03467691e+00 1.57756791e-01 5.70355296e-01 -7.50489593e-01 1.10658097e+00 3.54792386e-01 -1.37312555e+00 -3.95156235e-01 -1.17582989e+00 -3.90204996e-01 -1.92541346e-01 4.89105046e-01 7.66469315e-02 1.09682508e-01 -1.06130695e+00 5.78597844e-01 -5.48028946e-01 -3.80488843e-01 2.19066858e-01 -1.57105476e-01 -6.00917637e-01 -2.95089334e-01 -1.06413674e+00 6.06455684e-01 3.82281810e-01 1.16384014e-01 -6.52329803e-01 -7.57150829e-01 -1.01792991e+00 1.24426357e-01 2.59813607e-01 -9.73528981e-01 1.04704762e+00 -2.77781397e-01 -1.65767789e+00 5.49188495e-01 -1.87955394e-01 -1.68760821e-01 8.05392921e-01 -4.35852468e-01 -3.88370872e-01 4.92891490e-01 1.25447407e-01 6.06991410e-01 7.38417149e-01 -1.14034581e+00 -2.82686085e-01 -2.92171985e-01 -2.35941455e-01 5.71294963e-01 -4.64100391e-01 -3.16969335e-01 -4.93142605e-01 -9.28571880e-01 -3.82421851e-01 -7.21691489e-01 1.70824394e-01 -5.93805790e-01 -9.88515496e-01 -3.68798226e-01 9.35936689e-01 -8.80462408e-01 1.37365115e+00 -2.30951977e+00 8.40619326e-01 -3.22201461e-01 4.41943765e-01 7.64578059e-02 -4.01654989e-01 8.03689122e-01 2.07600772e-01 1.84389785e-01 -3.45079213e-01 -8.52442384e-01 2.27926582e-01 -1.74303815e-01 -5.68286717e-01 3.42278391e-01 8.87156203e-02 1.02158010e+00 -1.06295931e+00 -7.21131682e-01 2.21186131e-02 4.12298769e-01 -3.76314342e-01 3.67157906e-01 -1.29433915e-01 4.06712174e-01 -5.52335262e-01 6.03512704e-01 2.39891753e-01 -1.25265941e-01 -2.71211773e-01 -4.91778433e-01 -1.41404569e-01 1.81032181e-01 -7.22325325e-01 2.16293502e+00 -2.57264286e-01 9.56869960e-01 -1.55140460e-01 -7.82542288e-01 8.97763550e-01 5.39915264e-01 4.16984528e-01 -2.57458299e-01 3.34671438e-02 3.08778226e-01 -5.47241569e-01 -5.49607933e-01 1.18123388e+00 3.13328385e-01 -5.06153643e-01 4.31389511e-01 2.72195250e-01 -5.92833042e-01 5.86618841e-01 8.36021423e-01 1.25529134e+00 9.88178104e-02 1.09632172e-01 2.45765820e-02 4.61095840e-01 -9.25320014e-02 1.75701886e-01 8.06250870e-01 1.38214424e-01 1.04071605e+00 5.69155812e-01 2.77080894e-01 -8.99039209e-01 -1.08396709e+00 -5.90461195e-02 7.74148107e-01 1.01953730e-01 -9.38059270e-01 -7.18956649e-01 -5.83142757e-01 -1.34598643e-01 8.81608605e-01 -4.03909951e-01 -4.14172262e-01 -3.40103239e-01 -3.93935919e-01 8.61392379e-01 2.68919826e-01 5.34293413e-01 -9.43022192e-01 -3.05728763e-01 -4.50681802e-03 -7.53490865e-01 -1.27139354e+00 -9.67883646e-01 -3.82409215e-01 -8.02242994e-01 -8.32538307e-01 -1.06498575e+00 -4.12405998e-01 2.74014711e-01 2.50992864e-01 1.10678637e+00 -4.55836207e-01 -2.15797380e-01 8.19872797e-01 -5.31050205e-01 -1.99772701e-01 -4.92812485e-01 2.62245983e-01 -3.12273353e-02 3.48386206e-02 -1.14819117e-01 -4.60602880e-01 -4.31419820e-01 -3.91800217e-02 -1.09216070e+00 3.20338190e-01 7.93473601e-01 8.06520224e-01 2.83344328e-01 -5.07730603e-01 8.26372802e-01 -6.07129216e-01 1.09882474e+00 -5.08565187e-01 1.28015071e-01 4.05060619e-01 4.79633622e-02 1.07647806e-01 6.62637889e-01 -5.07597983e-01 -1.10969520e+00 -4.58893150e-01 9.17946398e-02 -4.31566447e-01 9.57863405e-02 8.96642625e-01 -4.18374538e-01 6.94453418e-01 4.82485563e-01 4.74099427e-01 -1.35315046e-01 -4.53435242e-01 6.73245966e-01 9.86018777e-01 9.54868555e-01 -5.80257058e-01 5.50860822e-01 2.00352460e-01 -1.06686935e-01 -1.24030840e+00 -9.42560256e-01 -4.35368508e-01 -3.86830002e-01 -2.57061452e-01 8.81185353e-01 -1.04050255e+00 -5.54988265e-01 4.00416315e-01 -1.47801518e+00 1.02546193e-01 -5.21321595e-01 5.92451811e-01 -7.52657592e-01 7.20549226e-01 -5.77207744e-01 -5.19069314e-01 -6.22963905e-01 -1.04966986e+00 1.55563974e+00 1.48347825e-01 -4.60243374e-01 -9.50504303e-01 2.02425346e-01 5.16663790e-01 2.01636389e-01 4.93587971e-01 7.76504874e-01 -8.75374198e-01 -3.01851600e-01 -4.05321658e-01 -1.36174574e-01 3.50267291e-01 4.07572426e-02 2.24167928e-01 -7.57846951e-01 -2.65347004e-01 -1.86597526e-01 -3.94862622e-01 1.14762032e+00 3.28158438e-01 1.02772045e+00 -5.74622154e-01 -8.84429924e-03 4.76115137e-01 6.91482484e-01 -2.90052623e-01 6.28819525e-01 -1.82651818e-01 9.64587569e-01 6.33901179e-01 4.67034191e-01 5.68486750e-01 6.89950287e-01 5.16945124e-01 5.68187349e-02 1.31923378e-01 -1.82957709e-01 -4.76213038e-01 8.49754333e-01 1.58414972e+00 -1.13510840e-01 -6.84387982e-01 -4.81542617e-01 4.32041109e-01 -2.02290320e+00 -1.18059790e+00 -5.60907796e-02 1.96207309e+00 1.00806928e+00 -1.84847757e-01 -5.88441491e-02 -1.52853936e-01 6.23330593e-01 5.40138185e-01 -5.19160450e-01 -3.38015586e-01 -4.85895008e-01 -4.33476418e-01 -4.91424389e-02 2.09071070e-01 -9.01729763e-01 5.76420903e-01 5.58679295e+00 1.08492565e+00 -8.18570733e-01 -2.23345682e-01 3.31702054e-01 -2.06072852e-01 -5.38318276e-01 -2.49950051e-01 -5.99180579e-01 4.68000352e-01 1.00311244e+00 -6.43114686e-01 1.60487682e-01 4.86969501e-01 2.91107982e-01 -2.52216379e-03 -1.42714667e+00 1.16122258e+00 5.35893738e-01 -1.36752439e+00 4.64635193e-01 -3.74542773e-02 8.49134207e-01 -2.09784314e-01 1.10407643e-01 4.73467469e-01 8.02545249e-02 -1.11325824e+00 6.59610033e-01 8.02075028e-01 1.06773293e+00 -5.09990335e-01 6.83520079e-01 3.18295836e-01 -1.02759326e+00 2.79633403e-01 -1.63636446e-01 4.41687971e-01 3.07068467e-01 6.91059351e-01 -5.08033991e-01 1.16115379e+00 5.08654594e-01 1.45545661e+00 -5.42252958e-01 7.89797902e-01 -9.21059623e-02 5.33839822e-01 -3.76956873e-02 -1.10894531e-01 6.74906299e-02 -3.55547696e-01 1.25269091e+00 1.68822050e+00 5.66752076e-01 5.11185341e-02 9.54624638e-02 8.71697128e-01 -5.06151140e-01 1.41225889e-01 -8.09425652e-01 -5.20161748e-01 3.65898371e-01 1.27025032e+00 -1.10563137e-01 -5.21153390e-01 -2.01477930e-01 1.04242897e+00 4.77262735e-02 5.66882372e-01 -5.91204643e-01 -5.99824309e-01 2.37250179e-01 -2.87014693e-01 -6.15863763e-02 -2.02277869e-01 -1.84061602e-01 -1.70092559e+00 1.77070245e-01 -1.01314783e+00 2.54984349e-01 -8.38145435e-01 -1.31014788e+00 6.25997543e-01 2.77665734e-01 -1.32472301e+00 -3.61914486e-01 -7.96689987e-02 -7.97320366e-01 7.20877588e-01 -1.10390317e+00 -1.11229408e+00 -4.13557917e-01 3.78598452e-01 9.07539964e-01 -3.35512042e-01 6.72241569e-01 1.01892434e-01 -8.01270962e-01 6.04631126e-01 3.86095911e-01 -8.67005438e-02 9.92815316e-01 -1.31652629e+00 1.71593487e-01 7.09637225e-01 -1.70216709e-02 4.30167049e-01 9.12257731e-01 -6.24309182e-01 -1.73461843e+00 -1.10580885e+00 7.83516347e-01 -5.10215998e-01 5.33472955e-01 -2.37647951e-01 -9.27280545e-01 5.03394544e-01 5.23703694e-01 -4.87573147e-01 5.65672576e-01 -1.58676937e-01 -2.89815038e-01 1.04194492e-01 -7.25109041e-01 9.03886259e-01 9.60698187e-01 -4.95083481e-01 -7.86346197e-01 3.09799910e-01 9.06989813e-01 -5.58051109e-01 -1.06303883e+00 4.41479832e-01 4.94033575e-01 -6.79000914e-01 8.92752349e-01 -3.78407001e-01 1.08573091e+00 -1.45786162e-02 -2.32448012e-01 -1.70120013e+00 2.00515866e-01 -9.58522022e-01 -4.36538607e-01 1.66018903e+00 2.65869379e-01 -1.53946698e-01 1.93947896e-01 2.78498322e-01 -5.73143244e-01 -7.15043962e-01 -8.60413253e-01 -4.69930530e-01 -8.68079662e-02 -1.18602045e-01 4.24960971e-01 5.95766485e-01 3.02753776e-01 8.39884579e-01 -4.64186430e-01 -3.15835387e-01 5.43745339e-01 1.44323155e-01 9.15715098e-01 -1.03354371e+00 -4.40415181e-02 -6.74394429e-01 -2.46162191e-01 -1.24890280e+00 3.21199268e-01 -9.56499040e-01 7.74309933e-02 -1.86068702e+00 4.96447980e-01 2.20159009e-01 2.59858996e-01 8.85304362e-02 -3.17243248e-01 -9.14019346e-02 3.39058608e-01 3.63633752e-01 -1.07875943e+00 1.24657404e+00 1.38320100e+00 -3.95412713e-01 -2.32377246e-01 -9.72696021e-02 -7.91639686e-01 5.37297904e-01 3.77406389e-01 -6.37502968e-02 -2.78150856e-01 -2.99544007e-01 2.18175441e-01 4.95837629e-01 2.65058488e-01 -8.14434767e-01 2.91862994e-01 1.37033239e-01 1.76641524e-01 -9.58353579e-01 4.62820947e-01 -2.14555532e-01 2.21098028e-03 -1.44737497e-01 -7.40088701e-01 1.55670807e-01 2.31430642e-02 8.07020545e-01 -5.18072486e-01 6.26336643e-03 2.37480685e-01 1.71793252e-01 -7.84684196e-02 1.26681313e-01 -2.92070687e-01 5.50468862e-01 6.06945336e-01 -6.35830387e-02 -6.22419357e-01 -7.67006934e-01 -3.90392244e-01 4.78665501e-01 3.24590594e-01 4.89252985e-01 8.43836129e-01 -1.24418116e+00 -1.33916593e+00 -2.57808417e-01 1.76197812e-01 2.29935259e-01 5.17618001e-01 1.05883551e+00 -1.73683047e-01 4.09771323e-01 1.27601624e-01 -7.26987600e-01 -1.12022567e+00 1.40139610e-01 -1.29130960e-01 -4.78363544e-01 -9.38081443e-01 4.00340229e-01 3.29042673e-01 -3.92251074e-01 4.10790801e-01 -2.64876872e-01 -3.58693689e-01 4.01022375e-01 5.87379217e-01 7.30459630e-01 -2.83623729e-02 -6.87718630e-01 -7.87442401e-02 4.16017532e-01 -6.48451969e-02 -2.46054128e-01 1.26560152e+00 -2.82146692e-01 -2.35211793e-02 7.42093682e-01 1.37097716e+00 2.20330015e-01 -1.08345246e+00 -2.10436001e-01 -9.90449786e-02 -1.02518417e-01 -2.15050966e-01 -6.41219556e-01 -6.69781983e-01 9.19132054e-01 -2.21389413e-01 2.21363723e-01 9.78700221e-01 1.05400138e-01 9.07175839e-01 4.38857913e-01 -2.61323601e-01 -9.85302985e-01 5.11245847e-01 6.46123111e-01 1.41373599e+00 -1.06620705e+00 2.45703638e-01 -1.27626389e-01 -1.08219361e+00 1.11105788e+00 2.00809866e-01 1.42088562e-01 4.86890636e-02 -1.68761745e-01 -3.65292996e-01 -1.57883048e-01 -9.22221780e-01 2.11884007e-01 7.71873355e-01 3.23365659e-01 5.08313656e-01 1.02339417e-01 -1.47057250e-01 7.92835057e-01 -4.63547558e-01 -2.43886501e-01 8.24530482e-01 6.34054661e-01 -1.21279381e-01 -6.17471635e-01 -3.09219033e-01 5.70205212e-01 -4.45663840e-01 5.11911605e-03 -6.86011672e-01 5.84799647e-01 -6.78538084e-01 9.90714669e-01 -1.23054400e-01 -3.48661155e-01 5.31573236e-01 -2.14354932e-01 3.13933432e-01 -4.97663528e-01 -4.13044930e-01 6.07289281e-03 1.62652627e-01 -2.72450089e-01 -4.35248762e-01 -8.12271476e-01 -9.70118523e-01 -4.37891752e-01 -1.10622503e-01 3.87804985e-01 6.60688698e-01 8.63731623e-01 6.00805998e-01 7.64221013e-01 6.19819522e-01 -1.29370010e+00 -7.32150316e-01 -1.23999560e+00 -5.46341836e-01 4.05666083e-01 5.02740026e-01 -3.37311238e-01 -5.34916043e-01 5.84801920e-02]
[10.656126976013184, 0.6710078716278076]
e3d13657-7f93-4dbe-9dd4-861b9a87a323
an-interpretable-machine-vision-approach-to
1812.00668
null
http://arxiv.org/abs/1812.00668v1
http://arxiv.org/pdf/1812.00668v1.pdf
An Interpretable Machine Vision Approach to Human Activity Recognition using Photoplethysmograph Sensor Data
The current gold standard for human activity recognition (HAR) is based on the use of cameras. However, the poor scalability of camera systems renders them impractical in pursuit of the goal of wider adoption of HAR in mobile computing contexts. Consequently, researchers instead rely on wearable sensors and in particular inertial sensors. A particularly prevalent wearable is the smart watch which due to its integrated inertial and optical sensing capabilities holds great potential for realising better HAR in a non-obtrusive way. This paper seeks to simplify the wearable approach to HAR through determining if the wrist-mounted optical sensor alone typically found in a smartwatch or similar device can be used as a useful source of data for activity recognition. The approach has the potential to eliminate the need for the inertial sensing element which would in turn reduce the cost of and complexity of smartwatches and fitness trackers. This could potentially commoditise the hardware requirements for HAR while retaining the functionality of both heart rate monitoring and activity capture all from a single optical sensor. Our approach relies on the adoption of machine vision for activity recognition based on suitably scaled plots of the optical signals. We take this approach so as to produce classifications that are easily explainable and interpretable by non-technical users. More specifically, images of photoplethysmography signal time series are used to retrain the penultimate layer of a convolutional neural network which has initially been trained on the ImageNet database. We then use the 2048 dimensional features from the penultimate layer as input to a support vector machine. Results from the experiment yielded an average classification accuracy of 92.3%. This result outperforms that of an optical and inertial sensor combined (78%) and illustrates the capability of HAR systems using...
['José Juan Dominguez Veiga', 'Eoin Brophy', 'Zhengwei Wang', 'Tomas E. Ward', 'Alan F. Smeaton']
2018-12-03
null
null
null
null
['2048']
['playing-games']
[ 4.88211691e-01 6.02531573e-03 -5.09311929e-02 -7.16938302e-02 -3.17024350e-01 -4.81970727e-01 1.75674468e-01 1.88371558e-02 -7.42270470e-01 6.69219911e-01 1.02395721e-01 -3.89156550e-01 4.30665351e-02 -5.31888366e-01 -4.21446413e-01 -6.60952330e-01 1.65536076e-01 -3.44528645e-01 -2.40183473e-01 1.79427460e-01 -4.16033762e-03 4.36337322e-01 -1.59486413e+00 -1.58943966e-01 4.80799347e-01 1.19186962e+00 -1.09486558e-01 9.09070849e-01 4.15135235e-01 3.45057964e-01 -6.89109445e-01 -5.51985428e-02 2.78011411e-01 -4.66035426e-01 -6.79715797e-02 2.12511912e-01 5.77648222e-01 -5.73338330e-01 -6.01248331e-02 4.59063530e-01 7.11988032e-01 8.66047442e-02 1.80812567e-01 -8.98578703e-01 -1.73091754e-01 -9.11343098e-02 -2.55780071e-01 4.04592782e-01 8.44677329e-01 3.50385696e-01 4.07657534e-01 -5.09563088e-01 2.10018665e-01 4.61927623e-01 1.00094497e+00 2.41184682e-01 -1.18510318e+00 -3.43521863e-01 -5.09147704e-01 -8.26336369e-02 -1.41295707e+00 -6.47034287e-01 8.14615726e-01 -2.99206734e-01 1.24050558e+00 5.83524048e-01 1.17692494e+00 1.14154136e+00 3.79191786e-01 1.88451126e-01 1.15666676e+00 -4.52454180e-01 2.88646638e-01 3.50897640e-01 -1.08157195e-01 7.48116851e-01 7.87315965e-01 -2.08637819e-01 -4.46488768e-01 -4.19941312e-03 9.14353848e-01 2.54709452e-01 -2.33936965e-01 -2.25785643e-01 -1.13834500e+00 3.45760047e-01 2.10832298e-01 4.94848877e-01 -6.15505099e-01 1.92329139e-01 2.54447699e-01 1.52272671e-01 1.12567760e-01 5.57721674e-01 -1.85530812e-01 -5.80438316e-01 -9.96199667e-01 -2.59394646e-01 8.37039292e-01 2.35970527e-01 2.08657026e-01 2.10439175e-01 3.70406359e-01 4.94245231e-01 4.38002020e-01 3.83535832e-01 7.27739453e-01 -8.85474026e-01 1.54762805e-01 9.07283485e-01 2.29963794e-01 -1.01796710e+00 -7.51490414e-01 -3.21032137e-01 -5.07022858e-01 2.27538124e-01 6.48117602e-01 -4.47778851e-01 -6.08945429e-01 1.18700039e+00 3.98384809e-01 2.69134879e-01 -1.52347609e-01 9.38904166e-01 4.19730067e-01 2.77997315e-01 1.78545237e-01 -2.51141608e-01 1.50228155e+00 -3.40127856e-01 -6.37562156e-01 -2.50134557e-01 5.27267933e-01 -5.21177173e-01 1.02074862e+00 4.62506056e-01 -8.79027367e-01 -5.92004597e-01 -1.65835965e+00 -5.14135100e-02 -4.66028035e-01 2.98645139e-01 5.96073925e-01 1.28993285e+00 -8.94959629e-01 5.12207627e-01 -1.17444491e+00 -7.47042835e-01 2.46421769e-01 6.98848605e-01 -4.58598167e-01 2.81816632e-01 -7.47097433e-01 1.05790699e+00 1.60620779e-01 3.59257072e-01 -7.70658329e-02 -3.56431663e-01 -8.79752040e-01 1.13477949e-02 7.31048509e-02 -6.39365017e-01 8.50412369e-01 -1.05048752e+00 -1.70948732e+00 6.09011769e-01 9.33825970e-02 -4.02232587e-01 5.09479821e-01 -2.89176881e-01 -5.46494186e-01 4.19628471e-01 -2.35104337e-01 1.90735772e-01 8.35045815e-01 -4.22922432e-01 -2.79889524e-01 -3.78572553e-01 -3.69090103e-02 3.28102887e-01 -5.56895554e-01 -2.64611870e-01 -6.42283633e-02 -2.20660001e-01 7.15834349e-02 -1.10790539e+00 1.76242933e-01 9.19905026e-03 -5.53270318e-02 1.82685658e-01 7.35335052e-01 -6.36147916e-01 1.20607090e+00 -2.15922356e+00 -3.90655130e-01 2.02337712e-01 2.11707279e-01 5.50064683e-01 4.45939332e-01 1.70464560e-01 -1.37390226e-01 -2.02015370e-01 1.95123613e-01 -2.61129122e-02 -3.66116047e-01 1.02637693e-01 3.02970827e-01 8.87453079e-01 -1.28855174e-02 7.75307834e-01 -6.40292346e-01 -2.85271108e-01 7.22592533e-01 8.08760047e-01 -2.31372416e-01 3.45909260e-02 5.39449513e-01 3.61968935e-01 -1.48268893e-01 7.16825843e-01 4.72378545e-02 -9.14508998e-02 3.51057857e-01 -3.00870121e-01 -1.56287149e-01 2.93203592e-01 -1.27759075e+00 1.53639746e+00 -4.77606356e-01 6.50232673e-01 -1.92122623e-01 -8.93150568e-01 9.59432840e-01 5.50782621e-01 7.63870001e-01 -6.78495765e-01 2.76770651e-01 1.92835137e-01 1.45964861e-01 -7.53480196e-01 3.19911063e-01 -1.46968916e-01 -1.67574305e-02 3.41879338e-01 -1.26357600e-01 5.39667189e-01 -7.02860504e-02 -2.91938215e-01 1.16931009e+00 5.03305614e-01 5.89473069e-01 -1.59565032e-01 4.69049513e-01 -2.47789100e-01 2.14628384e-01 5.53784490e-01 -4.18709487e-01 5.27451634e-01 -7.56277591e-02 -6.58398628e-01 -8.56573045e-01 -9.66708899e-01 -9.55965444e-02 7.48765826e-01 -1.60301402e-01 -2.62372941e-01 -6.21489108e-01 -2.71516293e-01 8.88419989e-03 2.98710167e-01 -4.55799431e-01 -1.97437689e-01 -4.45613056e-01 -6.90151870e-01 6.80574000e-01 6.67548478e-01 5.30578554e-01 -7.30773747e-01 -1.57481611e+00 4.27497178e-01 1.52391598e-01 -9.20665324e-01 -3.22036371e-02 2.56668925e-01 -9.77626204e-01 -9.82870519e-01 -5.99509537e-01 -1.05177075e-01 5.18396735e-01 3.00164670e-01 4.69127268e-01 6.44792244e-03 -4.40000445e-01 8.62870514e-01 -7.98977688e-02 -6.19277418e-01 1.49032041e-01 1.26940101e-01 4.15399969e-01 1.76016822e-01 7.42980599e-01 -7.91860282e-01 -9.37576413e-01 8.12303200e-02 -5.51549077e-01 -1.48089767e-01 4.99813944e-01 3.95629615e-01 2.27634296e-01 -4.02499586e-01 5.56532741e-01 -3.68894488e-01 6.97413325e-01 -3.02716762e-01 -4.14477587e-01 -4.34167832e-02 -8.52318406e-01 -2.74270773e-01 5.94786227e-01 -5.35974383e-01 -6.79578304e-01 4.70770776e-01 1.41831860e-01 -1.72641650e-01 -2.46024981e-01 3.83783877e-01 1.17053173e-01 -1.52222008e-01 8.11755061e-01 -6.83806511e-03 2.85264045e-01 -3.34122866e-01 -1.57463774e-01 9.65532184e-01 6.62837207e-01 -3.53918448e-02 3.90009791e-01 3.62109989e-01 1.74842089e-01 -1.26391304e+00 -2.84309834e-01 -5.67327499e-01 -5.59440136e-01 -5.83772421e-01 8.97718668e-01 -9.82483149e-01 -1.01115227e+00 2.88104683e-01 -5.79956234e-01 2.21631303e-01 -1.43751577e-01 7.79674053e-01 -3.79717708e-01 2.73606271e-01 -1.87508970e-01 -1.14411950e+00 -4.15975749e-01 -7.01843798e-01 7.84970522e-01 6.68257833e-01 -7.40271389e-01 -9.36982512e-01 -2.91535873e-02 7.22871721e-01 5.33505678e-01 7.08237112e-01 1.89116001e-01 -3.39929044e-01 -2.01635361e-01 -9.12340999e-01 1.94487199e-01 3.58171523e-01 3.22324067e-01 -2.84938544e-01 -1.36352456e+00 -2.46748075e-01 1.94388777e-01 -1.40127257e-01 2.06901222e-01 3.84570032e-01 5.38685262e-01 -2.02902481e-01 -8.58977139e-02 4.91373509e-01 1.52684152e+00 4.24148589e-01 8.04789245e-01 4.84433591e-01 6.62485242e-01 1.51860371e-01 3.00315637e-02 3.15336019e-01 3.09435070e-01 6.46564484e-01 1.00063853e-01 -2.88434505e-01 8.34641755e-02 -1.17445171e-01 5.53241253e-01 5.56698978e-01 -6.08622313e-01 1.51489571e-01 -7.50926495e-01 2.91094929e-01 -1.57301021e+00 -9.39813793e-01 -1.60033908e-02 2.56956005e+00 4.18190807e-01 5.85916415e-02 4.07136917e-01 4.98658091e-01 3.03709686e-01 -9.89347249e-02 -4.26665395e-01 -9.03736174e-01 4.29742783e-01 2.99968243e-01 7.66675055e-01 3.30112614e-02 -1.14479017e+00 -3.00083458e-02 5.97484732e+00 -2.10628092e-01 -1.24057138e+00 -1.27777115e-01 3.56498480e-01 -4.82573897e-01 5.03511190e-01 -3.22010368e-01 -4.82320458e-01 6.80159569e-01 1.54817080e+00 -3.10402513e-02 4.66717482e-01 7.83663571e-01 5.27867377e-01 -7.70602584e-01 -1.19459379e+00 1.24229670e+00 1.85874715e-01 -9.91416276e-01 -6.96362197e-01 3.13546687e-01 2.17300698e-01 -1.20445102e-01 -1.52211547e-01 -2.13119145e-02 -6.18161440e-01 -9.42003012e-01 1.99560180e-01 8.10443521e-01 7.91878283e-01 -4.23641711e-01 7.34890342e-01 1.91534519e-01 -1.01614654e+00 -1.63943142e-01 4.73250402e-03 -6.77709162e-01 -8.12357962e-02 2.23832920e-01 -9.96952772e-01 2.61059999e-01 5.46346307e-01 3.17420244e-01 -5.01320660e-01 1.10619354e+00 2.01570258e-01 5.35585225e-01 -7.72419095e-01 -1.74190193e-01 -4.64062802e-02 -7.90264830e-02 2.44187549e-01 1.11926830e+00 4.52296495e-01 -3.24329510e-02 5.48090274e-03 2.83102572e-01 1.56049207e-01 -1.49704646e-02 -8.86985064e-01 -2.30024487e-01 2.46180415e-01 1.34663713e+00 -8.02857518e-01 -1.60435557e-01 -7.87256420e-01 7.97480404e-01 -2.32010707e-01 4.20089141e-02 -6.37097418e-01 -5.27810574e-01 3.97443503e-01 4.62169468e-01 1.11438580e-01 -3.42430472e-01 -4.75920588e-01 -1.11449611e+00 2.82529861e-01 -7.74216712e-01 2.81553835e-01 -6.96798980e-01 -6.61487818e-01 2.17218190e-01 -2.17945397e-01 -1.30679560e+00 -5.13691962e-01 -6.12957895e-01 -5.24396122e-01 1.03830683e+00 -8.86234701e-01 -8.64697397e-01 -6.06676042e-01 5.15634894e-01 1.97297141e-01 2.29459211e-01 1.06925607e+00 3.53902787e-01 -7.06300557e-01 4.62735415e-01 -1.62356108e-01 9.48897675e-02 4.89566416e-01 -1.13057065e+00 -3.27853821e-02 1.02428901e+00 6.25038072e-02 8.15677762e-01 5.25587499e-01 -4.90349650e-01 -1.81942463e+00 -5.93872488e-01 8.96522284e-01 -7.05056667e-01 4.43642467e-01 -1.58066511e-01 -4.97281253e-01 5.02343774e-01 -3.70843452e-04 3.28364340e-03 9.92605507e-01 -5.73748127e-02 1.73613369e-01 -4.38763082e-01 -1.15833879e+00 4.98563200e-01 5.18315196e-01 -6.89535260e-01 -6.62606716e-01 -7.91028291e-02 -2.49819025e-01 -3.41414481e-01 -1.03413653e+00 -1.61634404e-02 1.34375882e+00 -8.44902694e-01 8.09185684e-01 -2.58795440e-01 3.67871374e-02 -3.53683531e-01 1.04945041e-01 -8.77765596e-01 -7.64243081e-02 -6.57658100e-01 -1.52826726e-01 9.21915352e-01 1.94014117e-01 -7.89966702e-01 8.41095030e-01 1.17073071e+00 1.60732105e-01 -6.08864307e-01 -1.10202074e+00 -6.31286740e-01 -8.32665861e-01 -3.94204170e-01 9.21018329e-03 7.26139724e-01 4.71597344e-01 3.60253125e-01 -4.26495761e-01 -9.16323289e-02 3.99270117e-01 -2.87488788e-01 7.48151481e-01 -1.27011716e+00 -3.48639935e-01 1.03608556e-01 -9.12083626e-01 -4.44660336e-01 -7.16359735e-01 -5.20136297e-01 -2.76920080e-01 -1.31526434e+00 -2.05809638e-01 -8.37944634e-03 -4.75988418e-01 5.81946015e-01 5.12290299e-02 8.34422767e-01 2.16565862e-01 6.23455308e-02 -2.63047665e-01 -2.36037821e-01 8.29950809e-01 2.29008988e-01 -6.64539218e-01 1.52303446e-02 -7.74589241e-01 7.11078107e-01 8.80776107e-01 -2.17428207e-01 -5.89272439e-01 -5.21194302e-02 2.65681326e-01 4.52381670e-02 5.17472088e-01 -1.52959108e+00 1.67375460e-01 3.00723314e-01 1.02584016e+00 1.69288576e-01 4.91345167e-01 -1.11640322e+00 5.40808856e-01 5.66474736e-01 5.46990298e-02 1.89741701e-01 1.95016906e-01 3.51057619e-01 2.61962682e-01 -7.64462538e-03 4.81901169e-01 -1.63786203e-01 -3.94939631e-01 -4.17544037e-01 -5.87152898e-01 -6.07557297e-01 9.47452009e-01 -1.09989715e+00 -1.82165071e-01 -2.90280730e-01 -7.66588271e-01 -2.86754757e-01 3.75370085e-01 3.29727918e-01 3.14271629e-01 -1.01236820e+00 -2.39053853e-02 4.73353893e-01 -9.97921303e-02 -5.92447579e-01 6.38073459e-02 1.27458513e+00 -5.80164194e-01 5.52027941e-01 -6.29961252e-01 -5.37180066e-01 -1.32678854e+00 2.68743962e-01 4.89938885e-01 2.41461471e-01 -8.11870754e-01 1.96049795e-01 -7.04725981e-01 4.58151877e-01 1.84615150e-01 -3.49908501e-01 -2.21578091e-01 1.26732424e-01 6.37966394e-01 5.45854747e-01 2.91077524e-01 -3.83163989e-01 -5.85270345e-01 2.78929502e-01 1.66349605e-01 -3.46590988e-02 1.14713347e+00 -2.40435451e-01 3.38482708e-01 6.38309419e-01 9.86592114e-01 -4.70998287e-02 -1.24878120e+00 3.46411258e-01 -3.44080180e-02 -3.75448585e-01 2.70524800e-01 -9.65641558e-01 -6.76472127e-01 5.92748284e-01 1.24591053e+00 3.37353796e-01 1.26415741e+00 -5.42312324e-01 5.79332352e-01 4.71618652e-01 2.48966798e-01 -1.30661976e+00 -2.81288654e-01 -3.10163975e-01 3.10179144e-01 -9.12350714e-01 2.95362562e-01 1.81284230e-02 -3.93131346e-01 1.23934305e+00 3.08487743e-01 -1.70940727e-01 3.12999338e-01 2.18945459e-01 1.56897143e-01 -9.66991708e-02 -2.69371629e-01 -7.41878375e-02 1.59220830e-01 6.50575697e-01 7.47379303e-01 1.54178113e-01 -7.15199232e-01 3.92678142e-01 -3.35906222e-02 6.47365928e-01 6.79597437e-01 1.26653624e+00 -3.14884663e-01 -7.70165801e-01 -4.28270161e-01 6.79819643e-01 -6.82328463e-01 2.85040975e-01 -1.96439788e-01 6.75749183e-01 3.50847512e-01 1.05240583e+00 -6.76490664e-02 -4.07818794e-01 4.18496251e-01 5.51436961e-01 4.85132366e-01 -2.90406346e-01 -7.62402833e-01 1.86394677e-01 3.12835127e-01 -6.93577528e-01 -6.67811334e-01 -7.80637920e-01 -7.48921037e-01 -1.34668231e-01 4.51930463e-02 -2.58542269e-01 1.07821929e+00 1.03938091e+00 4.67421651e-01 3.92676204e-01 3.77357930e-01 -9.10093665e-01 -3.58979762e-01 -8.73171031e-01 -5.34280002e-01 1.57908246e-01 3.76048565e-01 -5.71122110e-01 -2.07230598e-02 3.23289126e-01]
[13.67502212524414, 2.9760146141052246]
1a92db05-b0b1-4f0d-9edb-285839d0659a
super-resolution-of-bvoc-emission-maps-via
2306.12796
null
https://arxiv.org/abs/2306.12796v1
https://arxiv.org/pdf/2306.12796v1.pdf
Super-Resolution of BVOC Emission Maps Via Domain Adaptation
Enhancing the resolution of Biogenic Volatile Organic Compound (BVOC) emission maps is a critical task in remote sensing. Recently, some Super-Resolution (SR) methods based on Deep Learning (DL) have been proposed, leveraging data from numerical simulations for their training process. However, when dealing with data derived from satellite observations, the reconstruction is particularly challenging due to the scarcity of measurements to train SR algorithms with. In our work, we aim at super-resolving low resolution emission maps derived from satellite observations by leveraging the information of emission maps obtained through numerical simulations. To do this, we combine a SR method based on DL with Domain Adaptation (DA) techniques, harmonizing the different aggregation strategies and spatial information used in simulated and observed domains to ensure compatibility. We investigate the effectiveness of DA strategies at different stages by systematically varying the number of simulated and observed emissions used, exploring the implications of data scarcity on the adaptation strategies. To the best of our knowledge, there are no prior investigations of DA in satellite-derived BVOC maps enhancement. Our work represents a first step toward the development of robust strategies for the reconstruction of observed BVOC emissions.
['Stefano Tubaro', 'Marco Marcon', 'Paolo Bestagini', 'Sara Mandelli', 'Antonio Giganti']
2023-06-22
null
null
null
null
['super-resolution']
['computer-vision']
[ 3.77031505e-01 -4.48238820e-01 1.94402367e-01 -1.75910696e-01 -7.86698818e-01 -5.19064784e-01 8.56680095e-01 5.56149334e-02 -4.82202619e-01 1.29556429e+00 1.58246100e-01 -3.57759267e-01 -4.76603240e-01 -1.14469743e+00 -4.87092793e-01 -1.03932202e+00 -1.86490715e-01 3.65298897e-01 3.25058430e-01 -4.80047673e-01 -6.88036308e-02 9.49591815e-01 -1.63889921e+00 1.16605155e-01 1.30646694e+00 7.64458179e-01 5.92355728e-01 4.20426607e-01 -2.06732690e-01 2.32804805e-01 -5.17364860e-01 4.32157636e-01 5.64494312e-01 -5.57274222e-01 -3.94167721e-01 -3.72721739e-02 3.26920331e-01 -1.02417715e-01 1.81202844e-01 9.51683462e-01 5.52842975e-01 4.47699815e-01 6.58973038e-01 -7.04512537e-01 -1.47832245e-01 3.85886103e-01 -2.92588741e-01 4.38838512e-01 -3.51503432e-01 1.74407102e-02 5.64921737e-01 -7.16785967e-01 4.08335358e-01 1.00023806e+00 5.82992673e-01 1.71242937e-01 -1.48629892e+00 -7.73779511e-01 1.96441874e-01 -2.89569758e-02 -1.47038937e+00 -4.20276016e-01 4.65846926e-01 -6.41311884e-01 8.30491662e-01 2.41196930e-01 5.79431474e-01 9.55547512e-01 -7.54154474e-02 -7.47730955e-02 1.61299419e+00 -2.95436651e-01 3.53159726e-01 4.49500740e-01 -3.33940715e-01 -1.49806082e-01 3.55482310e-01 6.71305299e-01 -2.55147517e-01 -5.05186766e-02 6.83370471e-01 -2.62535304e-01 -1.98333293e-01 2.57897656e-02 -6.85623765e-01 1.02372694e+00 6.29692316e-01 4.08191413e-01 -6.15387678e-01 -3.06467384e-01 1.39570134e-02 3.27842802e-01 1.04914784e+00 6.43086612e-01 -4.05656368e-01 3.50569963e-01 -1.25963378e+00 4.79465693e-01 3.82018745e-01 4.85559344e-01 8.09117675e-01 4.70362306e-01 1.54307619e-01 9.47676241e-01 2.19994843e-01 7.60164797e-01 1.54527172e-01 -8.84330630e-01 2.46956870e-01 3.71714711e-01 4.80067968e-01 -7.78735697e-01 -3.57135326e-01 -6.61307096e-01 -9.30992126e-01 5.00301123e-01 5.59015684e-02 -4.51527297e-01 -8.47304225e-01 1.60040510e+00 4.05783892e-01 1.08719729e-02 3.34769309e-01 1.02919674e+00 4.87213254e-01 1.04196966e+00 3.89243633e-01 -3.15446556e-01 8.65534484e-01 -4.48951811e-01 -6.97808564e-01 -8.54832232e-02 3.44861776e-01 -3.69688928e-01 5.92597246e-01 1.64974406e-01 -4.57362473e-01 -7.64664769e-01 -9.91996109e-01 4.00946677e-01 -1.07639253e+00 -9.65542626e-03 3.78695220e-01 4.63297009e-01 -8.65614295e-01 8.28500926e-01 -9.12618279e-01 -2.96748847e-01 1.27411142e-01 1.81464568e-01 -9.33427289e-02 3.57774675e-01 -1.65929830e+00 1.10277104e+00 6.41581297e-01 4.02530611e-01 -9.77772951e-01 -7.72869468e-01 -6.19200349e-01 -2.31033806e-02 2.45522380e-01 -3.09945822e-01 6.45115137e-01 -1.08656025e+00 -1.22522652e+00 3.75137538e-01 -6.47708029e-02 -5.62674105e-01 5.91221273e-01 -1.24701522e-01 -8.14867496e-01 -1.49320334e-01 1.81308627e-01 5.55575132e-01 7.46474683e-01 -1.57211721e+00 -7.71444738e-01 -4.07592237e-01 -1.82567071e-02 2.21967399e-01 -1.06354289e-01 1.18622616e-01 4.68734145e-01 -4.94551927e-01 -1.44635201e-01 -9.45386708e-01 -4.43114579e-01 -2.31578082e-01 3.60861748e-01 1.53121784e-01 7.89797604e-01 -7.93489099e-01 1.00004268e+00 -2.16119313e+00 2.38019377e-01 2.53481656e-01 -1.35875970e-01 6.38252795e-01 -1.96991354e-01 4.95495975e-01 1.10686973e-01 2.57642657e-01 -8.58111143e-01 -8.77633914e-02 -1.50363907e-01 4.53653574e-01 -5.21591544e-01 3.60446543e-01 5.47404051e-01 3.00798327e-01 -7.12414026e-01 -2.89824069e-01 4.14444566e-01 6.78058267e-01 -3.42661105e-02 3.95137519e-01 -4.54010546e-01 1.01480174e+00 -3.30727011e-01 1.60516113e-01 1.22632825e+00 1.99930057e-01 1.29846931e-01 1.61975533e-01 -8.18129659e-01 3.74548793e-01 -1.41682744e+00 1.01402116e+00 -6.05549932e-01 5.02125800e-01 4.18522090e-01 -8.75391066e-01 1.15450299e+00 2.32808247e-01 3.09833527e-01 -9.06799972e-01 -3.83479148e-01 4.05532777e-01 1.58055082e-01 -4.06805694e-01 5.82282364e-01 -6.88787103e-01 3.96252006e-01 1.99664980e-01 -2.65757561e-01 -4.15511966e-01 4.66094762e-02 -2.44994521e-01 2.07495376e-01 3.91105711e-01 3.08377501e-02 -4.43209469e-01 7.90518284e-01 2.85752207e-01 5.40291667e-01 7.42961228e-01 1.04985662e-01 6.10137105e-01 1.38727292e-01 -3.99619818e-01 -1.37344444e+00 -9.82419610e-01 -6.76186740e-01 1.09617543e+00 -2.40503460e-01 1.70601755e-01 -3.86849046e-01 -9.30599198e-02 8.40397850e-02 7.43584335e-01 -4.65462059e-01 2.33805403e-01 -4.98618811e-01 -1.55157566e+00 6.21349812e-01 4.86404300e-01 7.18787730e-01 -1.08295858e+00 -5.68802416e-01 5.46976507e-01 3.81555781e-02 -1.21972144e+00 6.51494801e-01 3.81067008e-01 -1.08017325e+00 -6.11360073e-01 -5.62613547e-01 -3.86065878e-02 2.35241354e-01 9.32010859e-02 7.48837054e-01 -3.56665522e-01 6.93900809e-02 -2.83664137e-01 -4.39283013e-01 -5.72752357e-01 -6.79669917e-01 4.48256582e-01 5.93721047e-02 -5.12835421e-02 5.87098151e-02 -7.77320325e-01 -3.01420808e-01 2.73790747e-01 -1.36208689e+00 -4.11605723e-02 5.99522829e-01 3.59474063e-01 4.91989911e-01 3.12485993e-01 7.36348093e-01 -8.48414838e-01 5.96515775e-01 -7.55162358e-01 -1.10723293e+00 -6.10451698e-02 -6.98689163e-01 6.98012337e-02 6.46386862e-01 -3.34122144e-02 -1.57996058e+00 -1.01459868e-01 -2.46412933e-01 -7.22307339e-02 -5.23084879e-01 7.78284848e-01 1.29593208e-01 -1.84776053e-01 8.31343889e-01 4.43531722e-01 -9.23273712e-02 -6.84368491e-01 1.73848063e-01 9.61332262e-01 1.87399432e-01 -5.17418444e-01 8.57898176e-01 6.40071630e-01 1.77596584e-01 -1.16295576e+00 -9.03883159e-01 -4.18026745e-01 -7.10376501e-01 -4.07109372e-02 7.50232100e-01 -1.12829423e+00 2.64011323e-01 4.01488394e-01 -6.91773236e-01 -5.35640121e-01 -6.70441017e-02 6.29384518e-01 -1.61170736e-01 2.16355294e-01 -6.49501337e-03 -1.03660882e+00 -3.79577309e-01 -9.84237254e-01 8.86661291e-01 8.36099610e-02 2.73928702e-01 -9.62890208e-01 5.15517473e-01 3.42995077e-02 8.85718703e-01 6.62039280e-01 6.56310618e-01 -1.51013181e-01 -5.39323926e-01 3.21648031e-01 -4.54663277e-01 6.69267595e-01 4.46297377e-01 8.20094123e-02 -1.18064785e+00 -3.51537794e-01 9.80711952e-02 -3.50763947e-02 1.13721180e+00 2.65942752e-01 9.80453372e-01 -1.27503932e-01 2.19454225e-02 7.43170619e-01 1.74647355e+00 6.30545840e-02 5.08268952e-01 7.07094669e-01 4.23258096e-01 8.77881289e-01 7.22542346e-01 4.93312925e-01 -7.66505897e-02 5.22036850e-01 6.40381634e-01 -2.79395938e-01 1.89052336e-02 2.57600993e-01 1.58833191e-01 1.65740341e-01 -5.93065083e-01 3.19484547e-02 -9.63488042e-01 7.12843478e-01 -1.72108459e+00 -9.22950864e-01 -2.26204574e-01 2.17039466e+00 6.42075241e-01 -7.40640983e-02 4.94264532e-03 -1.30457565e-01 5.79777241e-01 6.64286673e-01 -4.27385539e-01 -3.05870086e-01 -3.85063767e-01 3.98068339e-01 8.89431119e-01 6.02402806e-01 -1.20615125e+00 8.77563596e-01 5.78307772e+00 4.54973727e-01 -1.39566350e+00 2.66834080e-01 3.72453481e-02 -3.19151469e-02 -3.32701415e-01 -8.21456835e-02 -1.02344859e+00 4.63748068e-01 1.33832502e+00 1.42537981e-01 6.14847541e-01 4.60097820e-01 8.10320914e-01 -1.38328061e-01 -3.91272455e-01 2.52569437e-01 -4.03033078e-01 -1.30753410e+00 6.43752664e-02 1.21345259e-01 1.00300813e+00 5.98084807e-01 5.52389771e-02 2.15497747e-01 2.01800123e-01 -9.49891150e-01 3.91401172e-01 6.91033721e-01 6.69587433e-01 -7.89526641e-01 8.13020051e-01 4.24147636e-01 -1.14991486e+00 -1.90154254e-01 -6.42854810e-01 -1.59949780e-01 -4.47320119e-02 6.74844563e-01 -6.55057847e-01 9.51641619e-01 8.41157675e-01 6.22094691e-01 -4.60932821e-01 7.84860373e-01 -2.01320261e-01 6.36711895e-01 -5.76906919e-01 4.33295995e-01 6.39367223e-01 -5.21672368e-01 6.16705775e-01 1.22374296e+00 4.10999209e-01 2.45188504e-01 -1.70091823e-01 1.25577831e+00 3.95529628e-01 8.06986541e-02 -8.16739976e-01 -2.19874252e-02 5.36397099e-01 1.03995848e+00 -2.03478307e-01 -2.04636678e-01 -3.33399832e-01 2.89394408e-01 8.33475292e-02 4.63927180e-01 -7.20493197e-01 -3.72147001e-02 9.24720526e-01 3.14882666e-01 3.75534475e-01 -5.69190860e-01 -9.15350094e-02 -8.74220908e-01 -1.52779132e-01 -9.57094014e-01 3.71077329e-01 -7.28180349e-01 -1.10340953e+00 6.40215576e-01 5.85724533e-01 -1.31117284e+00 -1.72735542e-01 -4.54516709e-01 -5.22335470e-01 1.35266352e+00 -2.17459583e+00 -8.11782181e-01 -5.51399410e-01 2.66455054e-01 1.95931107e-01 3.46864536e-02 6.80557787e-01 7.94905648e-02 -3.65561724e-01 -3.84554088e-01 9.11401391e-01 -3.63294601e-01 6.03549957e-01 -1.12147737e+00 3.52563471e-01 8.77277613e-01 -3.06104004e-01 2.79359877e-01 7.27814794e-01 -6.10996008e-01 -6.92794800e-01 -1.52012324e+00 5.93814373e-01 -3.32514569e-02 5.72043300e-01 -1.50558308e-01 -1.39496720e+00 4.91022557e-01 -9.16797668e-02 3.49706784e-02 4.93575782e-01 -1.32170439e-01 1.29796490e-01 -5.75877309e-01 -1.23259461e+00 2.50434816e-01 6.51507854e-01 -5.59726775e-01 -5.82596481e-01 2.16873854e-01 5.34175038e-01 -2.40169853e-01 -1.13478553e+00 9.05255377e-01 4.41220313e-01 -8.60214114e-01 9.03409541e-01 -5.35016119e-01 3.05281609e-01 -6.54428363e-01 -3.68144572e-01 -1.55670249e+00 -4.68025863e-01 2.29747087e-01 3.30628306e-01 1.11524868e+00 4.02245820e-01 -8.09622645e-01 3.21496844e-01 3.14991653e-01 -6.08846359e-03 4.89861295e-02 -9.29921448e-01 -9.69544351e-01 4.19468313e-01 -1.61303982e-01 8.56808901e-01 8.51446986e-01 -8.46111953e-01 -1.49207905e-01 -3.52202594e-01 8.61350238e-01 5.22168636e-01 2.29807109e-01 6.33288026e-01 -1.48323691e+00 -2.50929985e-02 -1.93436161e-01 3.24968308e-01 -1.88797206e-01 9.61659476e-02 -5.52293599e-01 1.01809934e-01 -1.38567507e+00 -2.33068839e-01 -5.47815144e-01 -4.60233361e-01 2.49233276e-01 -6.01142943e-02 9.05813724e-02 2.19982505e-01 2.72258222e-01 1.45032153e-01 7.87185311e-01 1.00793695e+00 3.47227119e-02 -4.03133720e-01 -2.46907532e-01 -2.22835243e-01 4.39937651e-01 1.08413887e+00 -7.24543989e-01 -1.73561320e-01 -5.93344092e-01 1.85802147e-01 -5.94848953e-02 3.39379042e-01 -1.10635662e+00 -2.77773947e-01 -5.21710217e-01 4.28612202e-01 -7.12621033e-01 2.18960613e-01 -8.72806013e-01 6.74763083e-01 2.79974759e-01 -2.42043450e-01 -2.28335842e-01 6.08341455e-01 2.69544512e-01 -3.32523763e-01 -2.12869376e-01 1.12412500e+00 -4.62222964e-01 -8.69110703e-01 1.38229832e-01 -6.14196658e-01 -3.02111119e-01 5.77656806e-01 5.94003201e-02 -2.10303053e-01 -9.91225243e-05 -7.13932037e-01 2.34197393e-01 3.77047837e-01 2.64045089e-01 3.00823301e-01 -9.25842166e-01 -1.12420630e+00 3.27596396e-01 8.74398276e-02 3.77943397e-01 1.72567263e-01 6.38927519e-01 -5.40724397e-01 6.65057242e-01 -4.58698481e-01 -4.83606607e-01 -8.81823838e-01 4.00777161e-01 7.93490708e-01 -3.36709201e-01 -3.39219689e-01 3.33907872e-01 4.56417501e-02 -8.44470859e-01 -3.73829812e-01 -5.96476421e-02 -4.14011270e-01 4.46418554e-01 3.16060275e-01 5.95125377e-01 2.42837623e-01 -6.46577835e-01 -2.07491860e-01 4.56916988e-01 3.37118983e-01 -3.02297562e-01 1.62942266e+00 -2.60538518e-01 -2.08559811e-01 4.40596491e-01 8.08982730e-01 -8.07712004e-02 -1.37690473e+00 -2.61706263e-01 -4.27305847e-02 -5.78180254e-01 4.56230700e-01 -7.29612589e-01 -1.03697777e+00 8.40025783e-01 7.81525910e-01 1.41063750e-01 1.17099810e+00 -3.88156444e-01 2.69750327e-01 4.56406385e-01 9.92385074e-02 -1.13793290e+00 -5.24151742e-01 4.16917354e-01 1.02328289e+00 -1.54151893e+00 8.28048959e-02 -2.20799651e-02 -3.97155017e-01 9.41121578e-01 4.29472864e-01 -8.30947012e-02 5.82095981e-01 1.00171678e-02 1.27471805e-01 -3.12329661e-02 -4.35513139e-01 -6.31291628e-01 3.02197188e-02 5.71229339e-01 1.76412970e-01 2.49533981e-01 -3.78798842e-01 -3.40028815e-02 -5.83317271e-03 1.40527040e-01 4.20315653e-01 6.53532624e-01 -5.68592906e-01 -1.02427197e+00 -7.87615001e-01 2.10466295e-01 -3.44035268e-01 -2.83851087e-01 -2.54795253e-01 8.23672056e-01 3.13472390e-01 8.30128968e-01 5.93641885e-02 1.03192858e-01 1.30878299e-01 -2.63203727e-03 -1.58723742e-02 -5.98490655e-01 -3.14435899e-01 5.03401831e-02 2.32630610e-01 -1.24388754e-01 -9.06753957e-01 -8.11602950e-01 -7.93304682e-01 -2.32948333e-01 -2.37409562e-01 3.97875816e-01 8.52183938e-01 8.70260775e-01 1.72979802e-01 4.82535183e-01 7.56980240e-01 -1.04057717e+00 -5.43201029e-01 -1.23021567e+00 -8.96542490e-01 9.23299640e-02 7.23145604e-01 -8.28732431e-01 -6.47598803e-01 -4.77472216e-01]
[9.77794361114502, -1.6997723579406738]
3b228ba3-faf6-477c-9e2e-7d11e39b68bf
a-simple-and-robust-convolutional-attention
1904.01375
null
https://arxiv.org/abs/1904.01375v5
https://arxiv.org/pdf/1904.01375v5.pdf
A Holistic Representation Guided Attention Network for Scene Text Recognition
Reading irregular scene text of arbitrary shape in natural images is still a challenging problem, despite the progress made recently. Many existing approaches incorporate sophisticated network structures to handle various shapes, use extra annotations for stronger supervision, or employ hard-to-train recurrent neural networks for sequence modeling. In this work, we propose a simple yet strong approach for scene text recognition. With no need to convert input images to sequence representations, we directly connect two-dimensional CNN features to an attention-based sequence decoder which guided by holistic representation. The holistic representation can guide the attention-based decoder focus on more accurate area. As no recurrent module is adopted, our model can be trained in parallel. It achieves 1.5x to 9.4x acceleration to backward pass and 1.3x to 7.9x acceleration to forward pass, compared with the RNN counterparts. The proposed model is trained with only word-level annotations. With this simple design, our method achieves state-of-the-art or competitive recognition performance on the evaluated regular and irregular scene text benchmark datasets.
['Yanning Zhang', 'Zhen Li', 'Hui Li', 'Peng Wang', 'Fan Dang', 'Lu Yang']
2019-04-02
null
null
null
null
['irregular-text-recognition']
['computer-vision']
[ 8.45542669e-01 -1.88126788e-01 -2.70281043e-02 -4.34864312e-01 -7.05677986e-01 -3.12818140e-01 8.33806396e-01 4.87948023e-03 -6.26013041e-01 2.39224598e-01 2.96949565e-01 -4.99594778e-01 5.81373513e-01 -6.22553766e-01 -8.59207749e-01 -6.24803424e-01 7.04925656e-01 1.71528414e-01 2.91985512e-01 -2.24767998e-01 6.00428939e-01 2.84488261e-01 -1.28928602e+00 4.74003702e-01 6.47904873e-01 9.20952082e-01 6.53438032e-01 1.04534137e+00 -4.15435821e-01 1.25292683e+00 -4.00051028e-01 -2.83049643e-01 -1.59514081e-02 -5.91876447e-01 -7.83638775e-01 3.07133883e-01 4.01285142e-01 -4.56760287e-01 -6.43675864e-01 8.47959995e-01 5.42664945e-01 1.64896138e-02 6.72394216e-01 -3.09283942e-01 -8.94369781e-01 5.56386232e-01 -8.03914905e-01 2.34383807e-01 4.23027843e-01 2.59095222e-01 9.04360592e-01 -1.07174230e+00 2.53408045e-01 1.04923773e+00 3.79653305e-01 4.94876444e-01 -9.62481916e-01 -2.23004565e-01 4.15632993e-01 3.67330253e-01 -1.35588408e+00 -5.98298609e-01 5.73584497e-01 -3.05478483e-01 1.45820904e+00 1.56785697e-01 3.08777809e-01 1.22070980e+00 1.83102749e-02 9.99908745e-01 7.48203695e-01 -5.72931170e-01 -1.06846727e-01 -2.77673095e-01 3.08542252e-01 8.15017581e-01 5.80102876e-02 -4.94427204e-01 -3.37955713e-01 3.87144476e-01 6.09911561e-01 2.56768793e-01 -3.85963798e-01 1.10443428e-01 -1.09899449e+00 6.69973850e-01 5.08451581e-01 2.68237978e-01 -2.82258272e-01 3.21011752e-01 7.11598694e-01 1.30979940e-01 2.95176268e-01 9.32289138e-02 -1.96440130e-01 -4.36166376e-01 -9.89260018e-01 -3.04526687e-01 4.97921318e-01 9.53447223e-01 4.43884194e-01 3.63361895e-01 -2.63710052e-01 9.92983341e-01 2.52765208e-01 4.87001866e-01 8.32412302e-01 -9.16964486e-02 9.38653350e-01 6.89765811e-01 -2.84775525e-01 -9.20564294e-01 -3.33612829e-01 -4.23315048e-01 -1.12438273e+00 -3.03000510e-01 2.26130754e-01 1.57214731e-01 -1.27565038e+00 1.17585516e+00 -1.05982713e-01 -1.17961019e-02 1.07261509e-01 9.76905584e-01 6.79771125e-01 1.07953537e+00 -2.39305139e-01 1.47033438e-01 1.36602724e+00 -1.47560549e+00 -5.30768931e-01 -3.58425349e-01 8.52510273e-01 -9.64219570e-01 1.37971652e+00 3.67238641e-01 -1.16775334e+00 -5.57409465e-01 -1.24623346e+00 -5.16146123e-01 -1.46098286e-01 6.07968628e-01 2.53500957e-02 4.88856584e-01 -9.26923573e-01 2.89942861e-01 -8.70348155e-01 -5.03887415e-01 4.63969976e-01 2.69155502e-01 -7.20174164e-02 -3.22888553e-01 -6.42296493e-01 7.95268774e-01 2.35477105e-01 3.10467035e-01 -7.79177129e-01 -2.57903367e-01 -9.90435123e-01 2.92643517e-01 3.87930125e-01 -5.60463428e-01 1.31941831e+00 -1.05159676e+00 -1.89839828e+00 7.06525207e-01 -3.97407979e-01 -6.05604887e-01 4.27481145e-01 -3.63803387e-01 -1.75675705e-01 1.56646237e-01 -2.63123035e-01 4.29327041e-01 9.25207734e-01 -8.24636757e-01 -3.39198291e-01 -1.83077097e-01 -1.35699138e-01 3.42434734e-01 -4.31330204e-01 2.08396375e-01 -7.23669231e-01 -9.33689475e-01 1.55166045e-01 -8.22734177e-01 -2.56473124e-01 -2.77148187e-02 -4.85400766e-01 -1.67258829e-01 8.85773718e-01 -7.62841463e-01 1.04297352e+00 -2.07101679e+00 1.97045043e-01 -1.22004218e-01 3.21370028e-02 5.02303183e-01 -2.75965929e-01 4.48396146e-01 2.00783312e-01 6.61032423e-02 -3.87106746e-01 -4.69070017e-01 -5.23925163e-02 1.51205817e-02 -4.50230509e-01 5.43268263e-01 4.14719403e-01 1.02653766e+00 -4.62143898e-01 -2.40670726e-01 2.67023295e-01 4.81862336e-01 -4.72006083e-01 3.48941386e-01 -4.07282412e-01 2.63855428e-01 -3.14447373e-01 3.68320584e-01 6.19366944e-01 -6.73751414e-01 2.18942761e-01 -1.04912862e-01 -1.00381665e-01 6.76154554e-01 -6.58455729e-01 1.93765724e+00 -6.68100595e-01 9.45927560e-01 -3.22677016e-01 -1.28827775e+00 1.09291124e+00 8.53950605e-02 -1.34781480e-01 -9.67790306e-01 3.57760131e-01 8.81178007e-02 -3.77800083e-03 -5.19684970e-01 6.59078062e-01 2.81045824e-01 -1.11861397e-02 5.42393506e-01 -1.13952108e-01 2.84087628e-01 -5.06796651e-02 6.92578331e-02 1.15375149e+00 1.43317431e-01 2.65711904e-01 -1.82282422e-02 9.17373717e-01 -2.82387227e-01 2.20209017e-01 6.38322830e-01 3.29117686e-01 7.77211726e-01 3.78461659e-01 -4.32989001e-01 -1.33009148e+00 -5.28856218e-01 1.40959218e-01 1.35482013e+00 3.06703206e-02 -4.36362535e-01 -5.93836069e-01 -4.67893064e-01 -5.25943577e-01 5.23411036e-01 -4.63575363e-01 2.15618704e-02 -9.35095668e-01 -6.09387517e-01 8.28000367e-01 7.76331723e-01 7.93559015e-01 -9.18341577e-01 -5.33419847e-01 2.22396359e-01 -9.87386480e-02 -1.30804873e+00 -7.54103899e-01 1.28983349e-01 -7.94954479e-01 -5.91939569e-01 -1.08354568e+00 -1.02484345e+00 8.52147698e-01 3.97571504e-01 8.33592772e-01 1.87944397e-01 -2.81308949e-01 1.35542313e-02 -6.39747500e-01 -4.17099446e-02 -2.78207868e-01 4.14296567e-01 -5.16428471e-01 2.42957637e-01 1.89132690e-01 -4.26878273e-01 -5.69318593e-01 2.19833836e-01 -9.14104164e-01 4.53027666e-01 8.02315652e-01 1.02003312e+00 3.22429985e-01 -6.51742041e-01 3.69924814e-01 -7.69574702e-01 4.22536224e-01 -1.25130177e-01 -6.95149541e-01 3.66245210e-01 -3.02270681e-01 3.87772471e-01 1.07587719e+00 -3.56969088e-01 -1.12378013e+00 3.57169986e-01 -3.92543197e-01 -3.00478995e-01 -1.99588865e-01 5.08523822e-01 -1.62607152e-02 1.15574896e-02 5.77014565e-01 8.91065240e-01 -2.18321905e-01 -6.28873348e-01 4.27304745e-01 9.40939963e-01 3.93514097e-01 -3.17909122e-01 5.08397400e-01 3.67040306e-01 -3.23310554e-01 -9.70006347e-01 -9.02627468e-01 -4.30843115e-01 -7.74331450e-01 1.82439223e-01 9.16907310e-01 -1.07500994e+00 -6.46217883e-01 6.07452750e-01 -1.33286071e+00 -4.77992207e-01 1.74867600e-01 2.78854162e-01 -3.79637599e-01 7.37177730e-01 -7.84313977e-01 -6.38085365e-01 -7.93529451e-01 -1.12364471e+00 1.17985630e+00 3.75308134e-02 7.77150691e-02 -6.74755573e-01 -2.09018603e-01 4.00036842e-01 5.56370020e-01 -2.50445247e-01 7.97408581e-01 -5.32880366e-01 -7.99451232e-01 -1.14257172e-01 -6.88072443e-01 2.61879355e-01 4.70313467e-02 -9.78443623e-02 -9.21807289e-01 -3.01691502e-01 -1.77984834e-02 -5.74844778e-01 1.18721986e+00 1.51825594e-02 1.54183567e+00 -4.44753110e-01 -1.40310535e-02 7.44660795e-01 1.38800490e+00 1.71391398e-01 9.01412845e-01 2.10817695e-01 1.03223574e+00 2.07714140e-01 2.20375910e-01 6.49633646e-01 5.38814783e-01 6.31985903e-01 3.74662280e-01 -1.70724049e-01 -2.12914631e-01 -1.96317062e-01 5.93218565e-01 1.25715518e+00 5.49133085e-02 -4.86684322e-01 -1.06274700e+00 4.57219303e-01 -1.80978000e+00 -8.50474954e-01 -1.94093898e-01 1.94829082e+00 7.38102734e-01 2.22470447e-01 -5.27648665e-02 1.10489093e-01 6.94175661e-01 2.93585956e-01 -5.58786452e-01 -6.94827497e-01 -1.21083513e-01 1.30928949e-01 6.26103044e-01 4.14866686e-01 -8.01257908e-01 1.18833721e+00 5.38124466e+00 9.10263717e-01 -1.54266524e+00 -5.04143573e-02 7.34275281e-01 -9.98559967e-02 -8.41676071e-02 -1.37540102e-01 -8.07739973e-01 3.67031723e-01 9.86807048e-01 1.40718017e-02 4.19986457e-01 6.46033347e-01 1.56946048e-01 1.06259726e-01 -1.07795310e+00 1.25972986e+00 5.30331254e-01 -1.46984267e+00 3.45633209e-01 -5.05355038e-02 6.43153429e-01 3.18979561e-01 -1.77969951e-02 2.63492823e-01 -5.60333245e-02 -1.33143866e+00 6.70582354e-01 4.49711561e-01 1.03955901e+00 -6.23150349e-01 7.40614951e-01 5.13359725e-01 -1.29078567e+00 7.85658658e-02 -5.93468070e-01 -1.74022734e-01 1.34135187e-01 3.63274068e-01 -9.38640654e-01 5.19929707e-01 3.59639436e-01 9.84506428e-01 -6.99201167e-01 8.54461253e-01 -9.68299210e-02 8.40994239e-01 -2.24384651e-01 -4.56432104e-01 5.73888540e-01 -1.11684566e-02 6.54733405e-02 1.70204103e+00 3.01688969e-01 1.20760715e-02 1.76753595e-01 4.70111430e-01 -4.31000739e-01 3.25643629e-01 -5.64909458e-01 -1.90098956e-01 -5.06894253e-02 1.07382345e+00 -6.51002884e-01 -6.01148546e-01 -6.60433888e-01 1.58211446e+00 5.03292441e-01 2.72275418e-01 -9.96290863e-01 -6.31948054e-01 1.66525096e-01 -1.27073273e-01 8.86745870e-01 -3.58216196e-01 -4.36362535e-01 -1.41789758e+00 2.48705447e-01 -1.05049145e+00 1.73107162e-01 -8.23282063e-01 -1.02439284e+00 8.63747478e-01 -7.32983053e-01 -1.06859469e+00 -1.28917873e-01 -7.64806628e-01 -6.34769857e-01 8.51108432e-01 -1.58056927e+00 -1.03424466e+00 -3.70734781e-01 6.78925574e-01 1.07120633e+00 -1.76893219e-01 6.54325306e-01 3.21174204e-01 -8.44131351e-01 7.56571412e-01 2.86016613e-01 3.78338516e-01 4.19867724e-01 -8.70869756e-01 7.86487937e-01 9.59775567e-01 2.04287186e-01 3.04528177e-01 2.74973899e-01 -3.60813230e-01 -1.68556106e+00 -1.30210698e+00 9.09205377e-01 -2.18220502e-01 6.17863238e-01 -6.89652622e-01 -1.09853303e+00 6.24224663e-01 5.57808936e-01 -4.81367372e-02 4.77321416e-01 -1.00575790e-01 -6.44546449e-01 1.01073489e-01 -6.36188745e-01 7.86163449e-01 1.06776679e+00 -7.01972306e-01 -4.08336759e-01 2.86978006e-01 8.02158773e-01 -4.50096518e-01 -5.22385120e-01 1.32968530e-01 3.33355635e-01 -5.87922812e-01 5.98593593e-01 -4.34743762e-01 7.99239755e-01 -2.73240566e-01 -2.56172329e-01 -7.18729556e-01 -3.58968407e-01 -6.26653492e-01 -4.31367289e-03 9.42570746e-01 6.05308235e-01 -4.68467176e-01 7.54296362e-01 3.44017744e-02 -2.56886393e-01 -8.88043046e-01 -6.76636875e-01 -5.73262215e-01 2.10907832e-02 -3.78116369e-01 2.67760456e-01 7.28145003e-01 -1.18364841e-01 7.89626300e-01 -6.63660824e-01 3.68074887e-02 2.41213843e-01 2.98455596e-01 7.49406695e-01 -6.88752592e-01 -3.51346016e-01 -6.08757973e-01 -4.24295753e-01 -2.00846100e+00 -5.61467893e-02 -1.01765692e+00 1.57916933e-01 -1.65532005e+00 3.23789090e-01 8.55625421e-03 -1.05321757e-01 5.16073227e-01 -6.97023347e-02 2.32017234e-01 3.78911644e-01 1.65347785e-01 -8.26617122e-01 1.00198841e+00 1.10215151e+00 -3.77333552e-01 1.06780529e-01 -3.26910257e-01 -5.73754609e-01 5.91930270e-01 1.13443375e+00 -2.40952194e-01 -3.62654001e-01 -1.03266203e+00 1.84490412e-01 2.97472794e-02 1.87595814e-01 -8.66386712e-01 4.48112518e-01 1.52437806e-01 3.81424904e-01 -8.80009711e-01 4.32653815e-01 -4.92940366e-01 -4.18748647e-01 3.68208975e-01 -5.83036780e-01 1.87907696e-01 3.13882768e-01 5.43567121e-01 -2.22081736e-01 -4.71312284e-01 7.51947045e-01 3.56190428e-02 -4.92679268e-01 1.25458211e-01 -5.87793589e-01 -1.09592438e-01 6.79194629e-01 -2.72318810e-01 -3.00849795e-01 -4.95258212e-01 -4.23420608e-01 2.97290571e-02 3.80529583e-01 3.57800514e-01 8.78943861e-01 -9.61472213e-01 -8.77375424e-01 1.00657381e-01 7.78299198e-02 1.83644108e-02 2.16646269e-01 7.31169105e-01 -6.96858048e-01 8.01750839e-01 -1.95094179e-02 -7.96567976e-01 -1.30381739e+00 5.13929069e-01 2.07900524e-01 -3.94340068e-01 -8.23814332e-01 7.58155882e-01 3.68264765e-01 -2.28564531e-01 2.69694030e-01 -4.51093853e-01 -1.30730331e-01 -3.30124497e-01 5.99152267e-01 5.51314764e-02 1.01856597e-01 -6.07621372e-01 -2.23189250e-01 7.65533447e-01 -5.23261607e-01 1.00319244e-01 1.20958757e+00 -1.36441782e-01 9.33173019e-03 4.03876960e-01 1.43432391e+00 -2.29075193e-01 -1.25189424e+00 -4.57830638e-01 3.59036028e-02 -2.66367733e-01 -3.37101780e-02 -6.88221157e-01 -9.27415013e-01 1.35590625e+00 3.55655342e-01 -3.72139141e-02 1.18109965e+00 -2.00591490e-01 8.08227181e-01 7.91073203e-01 -7.55407661e-02 -8.81804526e-01 3.26916784e-01 9.88839209e-01 1.03299248e+00 -1.25058818e+00 -8.78140107e-02 -2.94374913e-01 -7.49116242e-01 1.32572699e+00 6.71568096e-01 -2.12100163e-01 1.81491971e-01 2.24854693e-01 -1.13656215e-01 1.99064519e-02 -1.07548344e+00 -2.21023709e-01 2.42985860e-01 7.26046786e-02 7.69209385e-01 -3.46865356e-01 -1.38140485e-01 2.44730428e-01 1.62578281e-02 -2.41156533e-01 7.00083733e-01 8.30762982e-01 -4.85945731e-01 -9.88675058e-01 -9.25887153e-02 5.40159166e-01 -4.94877130e-01 -5.40732503e-01 -2.42797315e-01 3.22362512e-01 -6.10811055e-01 7.22460508e-01 2.21910536e-01 -2.19237357e-01 2.82324314e-01 2.77004596e-02 4.06678587e-01 -5.64698160e-01 -6.02128983e-01 2.19790488e-01 3.16754133e-02 -3.78264397e-01 -2.57975996e-01 -5.01900494e-01 -1.22907400e+00 -2.52245575e-01 -4.63772237e-01 -3.07410091e-01 6.99588954e-01 9.34706748e-01 5.56052566e-01 6.04431570e-01 7.10159540e-01 -7.88360655e-01 -5.19218683e-01 -1.12961996e+00 1.07738793e-01 7.18488097e-02 3.94215345e-01 1.09551456e-02 3.16919386e-02 3.29991192e-01]
[11.900084495544434, 2.2219715118408203]
e79b9710-3f34-448c-aa6c-11ab4aa643a3
a-new-android-malware-detection-approach
1608.00848
null
http://arxiv.org/abs/1608.00848v1
http://arxiv.org/pdf/1608.00848v1.pdf
A New Android Malware Detection Approach Using Bayesian Classification
Mobile malware has been growing in scale and complexity as smartphone usage continues to rise. Android has surpassed other mobile platforms as the most popular whilst also witnessing a dramatic increase in malware targeting the platform. A worrying trend that is emerging is the increasing sophistication of Android malware to evade detection by traditional signature-based scanners. As such, Android app marketplaces remain at risk of hosting malicious apps that could evade detection before being downloaded by unsuspecting users. Hence, in this paper we present an effective approach to alleviate this problem based on Bayesian classification models obtained from static code analysis. The models are built from a collection of code and app characteristics that provide indicators of potential malicious activities. The models are evaluated with real malware samples in the wild and results of experiments are presented to demonstrate the effectiveness of the proposed approach.
['Gavin McWilliams', 'Suleiman Y. Yerima', 'Sakir Sezer', 'Igor Muttik']
2016-08-02
null
null
null
null
['android-malware-detection']
['miscellaneous']
[ 4.06961173e-01 -2.22606152e-01 -4.17803258e-01 5.70161715e-02 -5.19099832e-01 -9.45794702e-01 7.27688849e-01 2.95724291e-02 -2.07070664e-01 6.71887577e-01 -2.66447932e-01 -8.26972306e-01 1.85937330e-01 -6.02904499e-01 -5.18150985e-01 -4.08986986e-01 -2.46389642e-01 1.23249702e-01 6.17487788e-01 1.11629479e-01 5.61649919e-01 2.63864577e-01 -1.50709450e+00 1.06737316e-01 8.19490314e-01 8.90799999e-01 2.15921283e-01 8.55005205e-01 1.37061223e-01 4.89394486e-01 -8.35091770e-01 -4.20740664e-01 2.56425291e-01 -1.82289124e-01 -3.32528174e-01 -1.55928239e-01 -1.37518212e-01 -4.77850080e-01 -6.17000833e-02 1.41631174e+00 -2.42053922e-02 -4.47418302e-01 3.76636714e-01 -1.21240377e+00 -7.94275999e-02 1.94940150e-01 -8.28713596e-01 5.20854652e-01 6.47838771e-01 1.11402921e-01 4.87588078e-01 -3.39093983e-01 2.89025098e-01 8.35431218e-01 6.89536273e-01 2.96824127e-01 -9.24772501e-01 -6.94700658e-01 -4.08889204e-01 1.62509501e-01 -1.26698601e+00 -3.31615835e-01 7.58421600e-01 -6.18497074e-01 8.89684319e-01 3.80161047e-01 6.47844732e-01 1.31675804e+00 6.93706572e-01 3.22126627e-01 1.34426332e+00 -1.40219554e-01 3.20741236e-01 4.99575615e-01 2.44716629e-01 6.73832059e-01 9.89217579e-01 -1.30274128e-02 -6.59501925e-02 -9.24371004e-01 -1.52736716e-02 2.78179854e-01 -3.01473830e-02 -2.32057273e-01 -4.82375801e-01 8.37677240e-01 -2.16030806e-01 3.85680974e-01 -3.02058071e-01 -5.57328425e-02 5.95668435e-01 -2.95922786e-01 2.47997463e-01 1.38051122e-01 -4.86795783e-01 -9.22472954e-01 -9.70937252e-01 1.03120714e-01 7.57174134e-01 4.41680610e-01 4.62621063e-01 1.45117752e-02 7.74599493e-01 5.51343858e-01 7.49292374e-01 7.21368372e-01 1.04526365e+00 -4.14766103e-01 1.42508686e-01 9.92511213e-01 -2.60015042e-03 -1.37862146e+00 3.10926527e-01 -2.71805644e-01 -2.26030722e-01 1.12645537e-01 2.26059422e-01 1.77554823e-02 -8.14407349e-01 1.19084489e+00 1.37890100e-01 4.51055259e-01 -2.88489372e-01 -1.79386333e-01 -9.86143202e-02 6.59782529e-01 3.16249877e-02 -4.87119228e-01 1.46830511e+00 -2.64373958e-01 -4.91091460e-01 -1.17613085e-01 4.81532931e-01 -6.05149686e-01 9.69695866e-01 6.03586495e-01 -4.39542621e-01 -1.53439045e-01 -1.28994536e+00 6.68905139e-01 -6.05404794e-01 -1.33736774e-01 3.49977851e-01 1.59390140e+00 -6.26442373e-01 1.88323662e-01 -1.17081428e+00 -4.33438301e-01 7.02261567e-01 6.17519200e-01 -4.56638299e-02 2.05472037e-01 -6.61892295e-01 4.07187313e-01 2.50346899e-01 -3.26611787e-01 -9.63890910e-01 -2.03356385e-01 -4.44148630e-01 -1.83641240e-01 4.68643188e-01 3.29068661e-01 1.34914410e+00 -9.83747542e-01 -1.29429173e+00 5.86197615e-01 -1.67250782e-01 -5.62002420e-01 2.66469955e-01 -1.39139920e-01 -7.04299331e-01 -2.74913432e-03 7.39813671e-02 -2.50270814e-01 1.24500060e+00 -1.12748456e+00 -6.76220834e-01 -6.01940930e-01 -7.55131319e-02 -3.89854401e-01 -6.59493089e-01 2.09671974e-01 -1.31012827e-01 -2.82455772e-01 -4.15938795e-01 -1.32950759e+00 1.73353449e-01 -8.84719253e-01 -6.31589234e-01 2.44353935e-01 1.89542389e+00 -8.41734707e-01 1.50153232e+00 -2.03177691e+00 -4.37401712e-01 3.94036055e-01 1.78835779e-01 9.28935349e-01 5.70524275e-01 3.51118535e-01 2.44775027e-01 6.55751884e-01 -4.21038002e-01 6.53152168e-02 -5.34946620e-01 3.11892271e-01 -4.50811654e-01 3.76137078e-01 -2.60852247e-01 4.89514321e-01 -8.33631277e-01 -3.17756444e-01 -9.92523879e-03 5.82444429e-01 -3.54939699e-01 5.60226552e-02 -1.32713303e-01 3.25432479e-01 -6.08074069e-01 9.35692668e-01 6.20686471e-01 -1.99714214e-01 4.60175008e-01 4.34216738e-01 1.20614134e-01 4.26983297e-01 -7.25487530e-01 4.77432638e-01 -3.17719042e-01 5.93683183e-01 4.76375259e-02 -7.27487206e-01 4.98076737e-01 1.06298558e-01 2.58137792e-01 -7.38026798e-02 4.60685909e-01 3.34655017e-01 3.80595863e-01 -5.69557369e-01 4.44244862e-01 3.85346115e-01 -1.34384772e-02 4.63743418e-01 -4.79195386e-01 3.62940580e-01 -6.65230826e-02 1.89602762e-01 1.51115406e+00 -7.88073316e-02 7.58589745e-01 -1.79004267e-01 1.08832932e+00 5.07904813e-02 3.63988638e-01 4.65618074e-01 -5.64574778e-01 -2.32984081e-01 6.97467506e-01 2.79719476e-02 -5.52780747e-01 -1.07128787e+00 1.72922481e-02 7.08450317e-01 8.60510208e-03 -6.20119274e-01 -1.18734407e+00 -1.26560962e+00 -1.64812833e-01 3.99151713e-01 -4.08566743e-01 -8.47569853e-03 -5.04152060e-01 -8.74747396e-01 7.56267726e-01 5.02900127e-03 8.10763180e-01 -5.95362782e-01 -9.91595864e-01 -2.16117245e-03 4.07756492e-02 -1.09778941e+00 -8.56660008e-02 -2.47152269e-01 -1.10094059e+00 -1.44510794e+00 -2.87571013e-01 -3.52144897e-01 6.73857808e-01 5.09970963e-01 4.88305390e-01 3.99319112e-01 -4.72280502e-01 5.12528062e-01 -3.54564190e-01 -6.30505145e-01 -9.90758717e-01 2.29947358e-01 1.79353341e-01 1.84820980e-01 4.72881705e-01 -4.78285551e-01 -3.28487962e-01 4.50490415e-01 -8.78027916e-01 -7.26258278e-01 3.99373025e-01 2.94907540e-01 2.39882797e-01 6.80871367e-01 5.19835651e-01 -1.24300373e+00 7.86827326e-01 -1.00122750e+00 -7.07616329e-01 -2.35781059e-01 -8.01761568e-01 -5.46985805e-01 7.26741672e-01 -7.24966645e-01 -1.04662144e+00 2.11779192e-01 -1.23928607e-01 2.96157002e-01 -3.84823352e-01 3.04905832e-01 -3.68261874e-01 -3.17606241e-01 6.40829444e-01 3.08567643e-01 1.59338042e-01 -3.06397319e-01 -1.37272596e-01 1.06753051e+00 1.70246184e-01 -3.35666090e-02 9.46895957e-01 5.68484128e-01 -2.08138362e-01 -1.44726610e+00 -9.96521190e-02 -3.93539608e-01 -9.07080770e-02 -3.46070617e-01 7.05368817e-01 -4.20251220e-01 -6.26731932e-01 7.77150810e-01 -8.14516485e-01 2.36020625e-01 5.55682242e-01 7.54696354e-02 2.04856936e-02 8.35158348e-01 -4.18089747e-01 -1.10553706e+00 1.07550144e-01 -1.42122376e+00 7.41289794e-01 2.22088546e-01 -3.22957993e-01 -9.83147144e-01 2.78551191e-01 5.56293607e-01 6.35526478e-01 1.55830517e-01 8.57464612e-01 -9.62932229e-01 -5.36344409e-01 -8.51767004e-01 -1.20241102e-02 4.41258311e-01 6.15908206e-01 3.78044665e-01 -9.11322713e-01 -1.79414287e-01 6.97071850e-01 3.25389981e-01 1.53400525e-01 1.11188747e-01 1.05993688e+00 -6.01888478e-01 -8.40153754e-01 1.22958243e-01 1.53812826e+00 8.10589910e-01 6.86437845e-01 2.60541201e-01 6.67559266e-01 3.05345446e-01 4.73099411e-01 2.13728741e-01 -7.22274259e-02 6.12586856e-01 7.96861172e-01 6.93412483e-01 3.56550306e-01 -3.89768004e-01 6.83482647e-01 6.77862704e-01 -9.43207890e-02 -6.03949390e-02 -1.08669519e+00 1.59892797e-01 -1.38280642e+00 -8.73922944e-01 -4.16313738e-01 2.48632717e+00 2.87541628e-01 4.47963744e-01 5.57748556e-01 4.14686054e-01 7.88759649e-01 8.26597735e-02 -2.85658568e-01 -4.97331142e-01 4.95036930e-01 1.66995764e-01 7.08233297e-01 5.76682389e-01 -9.08979833e-01 6.21475935e-01 6.29683065e+00 8.11923206e-01 -1.21085429e+00 4.81679857e-01 7.73316860e-01 4.79025811e-01 1.13108113e-01 1.29874825e-01 -9.74021435e-01 1.12783146e+00 1.35278559e+00 1.34520698e-02 3.24288875e-01 1.25232434e+00 2.57706165e-01 -6.88336313e-01 -3.69888067e-01 9.18196857e-01 2.00537249e-01 -9.67277110e-01 -2.75442988e-01 8.49231362e-01 5.37955761e-01 -1.31183369e-02 2.56368607e-01 1.39069185e-01 -9.53587070e-02 -7.01999366e-01 1.04216658e-01 -1.10718198e-01 3.08134913e-01 -6.53925836e-01 7.73152292e-01 5.24373174e-01 -9.93768334e-01 -5.52157760e-01 2.09172741e-01 -2.92343229e-01 1.75726667e-01 3.94580513e-01 -1.19789279e+00 -7.46298358e-02 5.02265811e-01 4.31980014e-01 -1.03483534e+00 8.25701058e-01 -1.86766731e-03 1.21330810e+00 -2.77348489e-01 -3.54989737e-01 -4.42961939e-02 -2.03337774e-01 6.55048847e-01 9.30639327e-01 5.30932188e-01 -5.57848811e-01 -3.50342333e-01 4.35897529e-01 1.39339343e-01 -3.32752541e-02 -1.34647048e+00 -6.02954686e-01 3.89006376e-01 1.39475310e+00 -1.15316951e+00 -2.35816717e-01 -5.00781655e-01 7.84057081e-01 -3.25934738e-01 7.35906735e-02 -9.88841474e-01 -1.90754935e-01 6.23948276e-01 7.18084395e-01 1.24254704e-01 -6.28906846e-01 -4.90089431e-02 -8.09305131e-01 -1.12766691e-01 -1.23461390e+00 -3.70035209e-02 -1.12865113e-01 -8.92961204e-01 5.72155237e-01 1.37914732e-01 -1.06162047e+00 -4.79834139e-01 -8.45424533e-01 -6.27399087e-01 3.30223024e-01 -8.25311303e-01 -7.61654079e-01 -1.00164384e-01 3.69239897e-01 4.96466845e-01 -3.53780955e-01 5.09767830e-01 3.06547076e-01 -4.36965138e-01 4.09176916e-01 7.45339096e-02 -1.05331533e-01 1.93985533e-02 -8.17027152e-01 2.41119161e-01 1.10516477e+00 2.45644301e-02 1.01204073e+00 6.66399837e-01 -1.36364484e+00 -1.45974886e+00 -7.31010199e-01 4.87125963e-01 -8.39105248e-01 9.26754415e-01 -6.00054443e-01 -7.49957800e-01 6.82633519e-01 4.83629443e-02 -4.52995092e-01 7.74143755e-01 -4.66144621e-01 -3.64486694e-01 1.09425541e-02 -1.38987136e+00 4.96733814e-01 4.85080034e-01 -6.51718080e-01 -1.90597445e-01 6.79665729e-02 1.13023125e-01 2.05302477e-01 -2.40200728e-01 2.67443776e-01 6.31970406e-01 -1.32929778e+00 5.80589533e-01 -3.80221717e-02 -4.68615703e-02 -2.45789558e-01 -2.41115972e-01 -5.46798289e-01 5.37646174e-01 -9.29086804e-01 -6.01866186e-01 1.15429342e+00 3.03332984e-01 -9.14376795e-01 1.07878232e+00 2.51774400e-01 6.36150777e-01 -7.97173202e-01 -1.00608885e+00 -8.95931304e-01 -4.16893721e-01 -4.64097232e-01 2.50590324e-01 7.16823578e-01 -2.36706808e-02 1.10103585e-01 -3.31009507e-01 4.70694713e-02 7.60209739e-01 -8.41678023e-01 9.47677076e-01 -1.20549452e+00 -7.18789518e-01 -2.48073533e-01 -6.94120526e-01 -7.24084675e-01 -1.71918035e-01 -4.04304206e-01 -1.58264264e-01 -6.11194432e-01 4.69397247e-01 -2.61060763e-02 2.47419700e-01 2.50526220e-02 -4.56752116e-03 4.98424023e-01 -1.49195731e-01 3.74534816e-01 -2.16761142e-01 -2.45725617e-01 2.87977427e-01 1.56901985e-01 -3.59363794e-01 5.12519002e-01 -5.49243033e-01 1.04768002e+00 1.09629250e+00 -5.74836195e-01 -7.58568466e-01 5.45326769e-01 5.21750987e-01 -3.48519713e-01 1.23309776e-01 -1.08771622e+00 -3.57941449e-01 1.77714333e-01 -9.86180678e-02 -3.19807410e-01 2.80029327e-01 -1.03726196e+00 2.04342797e-01 1.00351787e+00 3.00321460e-01 4.26472276e-01 1.31928131e-01 1.01781273e+00 1.39936134e-01 -3.91931266e-01 7.93065310e-01 4.11423296e-02 -3.18258047e-01 -7.83495679e-02 -1.10901999e+00 -2.21264243e-01 1.45063472e+00 -6.71816409e-01 -3.42481941e-01 -4.32227880e-01 -1.70802385e-01 -6.28506601e-01 8.32229018e-01 5.09895802e-01 4.84591931e-01 -6.81519806e-01 1.59855455e-01 2.97507495e-01 -1.14878222e-01 -1.01214683e+00 -1.20467782e-01 7.26840973e-01 -6.53326631e-01 3.77848625e-01 -1.91229522e-01 -6.08285904e-01 -1.93617547e+00 6.97411180e-01 -7.36007392e-02 -3.48144323e-01 -1.55502319e-01 2.29663089e-01 -4.13470805e-01 -8.07093573e-04 -2.33059563e-02 -1.12749554e-01 -1.58842161e-01 -4.13571179e-01 7.03974903e-01 8.78123701e-01 -1.04972981e-01 -1.06047785e+00 -5.55100679e-01 2.82095939e-01 -3.01288366e-01 -6.32776842e-02 6.76105440e-01 -1.53183313e-02 -3.89690101e-01 3.73966485e-01 1.10819590e+00 7.76136458e-01 -8.19556773e-01 5.89378774e-01 4.43142086e-01 -8.05214345e-01 -3.33892673e-01 -6.74984157e-01 -6.00954354e-01 6.66993618e-01 9.46238637e-01 9.68991637e-01 7.78537691e-01 -6.46355227e-02 9.66569662e-01 2.11358160e-01 6.85186446e-01 -7.02404797e-01 1.27940476e-01 1.92139894e-01 8.86342302e-03 -1.24366963e+00 -3.58884037e-02 -6.54917717e-01 -2.30065107e-01 7.17771292e-01 3.62281412e-01 -1.95924118e-01 1.05254757e+00 3.32682639e-01 -2.08234414e-01 -1.49490267e-01 -5.64600565e-02 4.62399751e-01 -9.88406241e-02 9.21003819e-01 3.17130476e-01 2.16256663e-01 -5.84996283e-01 3.52356344e-01 1.67948321e-01 -2.09295750e-01 9.27820921e-01 1.22408366e+00 -5.58881402e-01 -1.60781634e+00 -4.49791163e-01 7.98278630e-01 -1.27342260e+00 1.62241012e-01 -8.69326830e-01 8.79714429e-01 1.87573969e-01 1.23942304e+00 -4.75044072e-01 -7.29900599e-01 -3.39298785e-01 1.02025129e-01 1.11086950e-01 -6.77307069e-01 -3.97677422e-01 -1.83021963e-01 5.23041748e-02 -3.87534261e-01 -8.27794522e-02 -6.60280168e-01 -9.61383998e-01 -1.53605700e-01 -4.10596669e-01 8.23653638e-02 1.12725532e+00 7.59122610e-01 5.13071120e-01 7.72409588e-02 5.80047846e-01 -6.66901946e-01 -4.96860355e-01 -7.91258633e-01 -3.52073908e-01 4.82960865e-02 1.57958522e-01 -7.77698815e-01 -5.35926580e-01 -1.20173758e-02]
[14.422342300415039, 9.679590225219727]
cf181fae-34f3-4b8d-ad2b-f6527f97c0c3
wasserstein-gaussianization-and-efficient
2305.14746
null
https://arxiv.org/abs/2305.14746v1
https://arxiv.org/pdf/2305.14746v1.pdf
Wasserstein Gaussianization and Efficient Variational Bayes for Robust Bayesian Synthetic Likelihood
The Bayesian Synthetic Likelihood (BSL) method is a widely-used tool for likelihood-free Bayesian inference. This method assumes that some summary statistics are normally distributed, which can be incorrect in many applications. We propose a transformation, called the Wasserstein Gaussianization transformation, that uses a Wasserstein gradient flow to approximately transform the distribution of the summary statistics into a Gaussian distribution. BSL also implicitly requires compatibility between simulated summary statistics under the working model and the observed summary statistics. A robust BSL variant which achieves this has been developed in the recent literature. We combine the Wasserstein Gaussianization transformation with robust BSL, and an efficient Variational Bayes procedure for posterior approximation, to develop a highly efficient and reliable approximate Bayesian inference method for likelihood-free problems.
['David Nott', 'Christopher Drovandi', 'Minh-Ngoc Tran', 'Nhat-Minh Nguyen']
2023-05-24
null
null
null
null
['bayesian-inference']
['methodology']
[-6.30290955e-02 -1.49750367e-01 3.80322337e-02 -6.88088059e-01 -1.22358000e+00 -2.90859550e-01 7.15882361e-01 -9.56215113e-02 -2.62943834e-01 1.18402958e+00 9.04242173e-02 -2.15215564e-01 -1.73232064e-01 -7.29437411e-01 -5.41082978e-01 -8.06437671e-01 2.59317696e-01 4.98408794e-01 2.80373603e-01 3.93194169e-01 3.74401003e-01 5.35604417e-01 -1.18427193e+00 -5.73818624e-01 1.23275137e+00 4.73162025e-01 2.09243551e-01 8.50693047e-01 -8.91019851e-02 3.38420749e-01 -5.88090181e-01 -4.61632013e-01 -8.45969990e-02 -6.30118787e-01 -2.42750362e-01 -7.86070600e-02 2.66815603e-01 -5.06767452e-01 -1.88991129e-01 1.45559156e+00 3.94505411e-01 4.23723608e-01 1.35336030e+00 -1.37004054e+00 -3.72324884e-01 4.84821409e-01 -7.38813937e-01 -1.52387559e-01 2.75445908e-01 -2.38649353e-01 6.62933648e-01 -1.04010534e+00 4.34701711e-01 1.76942778e+00 6.65147126e-01 1.24883614e-01 -1.30939102e+00 -5.65493047e-01 -2.78345406e-01 -3.30261774e-02 -1.68701291e+00 -3.11434120e-01 5.20125866e-01 -4.41600889e-01 3.30125093e-01 2.77344495e-01 6.22692168e-01 1.05656922e+00 4.66327637e-01 9.12493348e-01 1.26494205e+00 -3.98415804e-01 6.17564678e-01 1.43176764e-01 4.26054120e-01 6.23516023e-01 8.56420398e-01 5.41402772e-02 -4.47742194e-01 -6.88717484e-01 8.83949459e-01 -1.30333692e-01 -1.99601144e-01 -3.55599135e-01 -1.03144932e+00 9.93520916e-01 -3.83328944e-01 -1.69456452e-01 -3.40984255e-01 5.08402050e-01 8.22072029e-02 -2.39831537e-01 7.65813112e-01 -1.74341187e-01 -5.02588637e-02 -1.15998760e-01 -1.21093953e+00 1.00350070e+00 9.55371141e-01 1.09337342e+00 8.02516758e-01 2.23470181e-01 -7.03115225e-01 6.42609835e-01 1.12136447e+00 1.24789810e+00 -1.44844115e-01 -1.26509225e+00 5.58640659e-02 -1.63761482e-01 8.16905797e-01 -7.86956608e-01 1.28487432e-02 -2.86667079e-01 -8.12104344e-01 1.01740383e-01 4.62402254e-01 -2.65823156e-01 -7.55547285e-01 1.72017634e+00 6.86558187e-01 4.93958831e-01 -4.54453472e-03 6.42911315e-01 4.40102816e-01 8.88719022e-01 -1.94563419e-01 -5.84935486e-01 9.11378324e-01 -3.23439986e-01 -1.22789967e+00 8.44489112e-02 -3.21392007e-02 -7.09742308e-01 9.79253650e-01 4.09771085e-01 -1.03730083e+00 -3.12485278e-01 -1.12966418e+00 2.20340252e-01 1.14095554e-01 -8.74107182e-02 4.51910466e-01 1.01700759e+00 -8.93282294e-01 6.48532987e-01 -1.09004021e+00 -1.58317119e-01 1.44619197e-01 -2.36174867e-01 -2.84326896e-02 -5.53099737e-02 -1.07440770e+00 8.94340277e-01 4.60411429e-01 1.61506087e-01 -1.13601708e+00 -6.45394266e-01 -9.12332058e-01 2.14005813e-01 4.23929542e-01 -5.85964203e-01 1.52062201e+00 1.16654374e-02 -1.72447097e+00 1.57263026e-01 -6.04512274e-01 -5.66642247e-02 7.37541914e-01 -3.59297872e-01 -1.00776397e-01 -1.23348907e-01 1.96294516e-01 2.98966542e-02 1.14945352e+00 -1.12944937e+00 -3.45558107e-01 -3.41555119e-01 -7.40003228e-01 -1.15558259e-01 2.51774669e-01 2.56643537e-02 -3.82783681e-01 -7.55695701e-01 4.34430778e-01 -5.53953767e-01 -2.31183350e-01 -1.61883682e-01 -5.48130035e-01 -4.43719715e-01 5.70937991e-01 -7.67831504e-01 1.16855991e+00 -1.95779157e+00 -6.86301813e-02 5.12064934e-01 -1.49386868e-01 -1.00373179e-01 1.00329556e-01 4.02862906e-01 4.14276272e-01 4.26389538e-02 -7.55837977e-01 -6.90499485e-01 3.21647108e-01 1.90477043e-01 -4.33770061e-01 7.94639587e-01 7.29225725e-02 6.54683053e-01 -1.22223926e+00 -6.98715627e-01 2.35468745e-01 3.89254808e-01 -4.45648283e-01 5.97936690e-01 -2.43463501e-01 4.23100680e-01 -5.75696528e-01 3.90066266e-01 1.24667859e+00 7.90083110e-02 4.00496572e-02 -8.86248276e-02 -2.45164305e-01 -1.48041576e-01 -1.58697593e+00 1.31133926e+00 5.16075548e-03 3.75166506e-01 -2.30896361e-02 -9.21685100e-01 1.17555237e+00 3.04577857e-01 9.33509246e-02 2.99839407e-01 1.33199885e-01 3.28476757e-01 -6.44729733e-01 -4.47450429e-01 4.66231048e-01 -5.12258708e-01 -6.66222870e-02 3.72247547e-01 1.46978572e-01 -7.94149578e-01 2.06489399e-01 2.87126034e-01 6.18820906e-01 5.64034581e-01 5.38281202e-01 -4.98591214e-01 2.63454109e-01 -2.75463194e-01 8.64189386e-01 1.42286146e+00 -1.12764768e-01 8.03739488e-01 5.52288175e-01 8.39590132e-02 -1.04093385e+00 -1.83466077e+00 -4.51431960e-01 4.48610187e-01 1.91253498e-02 -3.83054554e-01 -8.59843552e-01 -4.86215532e-01 9.98502970e-03 1.22137761e+00 -2.28338674e-01 -1.64932504e-01 -3.91963683e-02 -1.00984895e+00 4.67923224e-01 2.73411274e-01 6.09152675e-01 -5.12212396e-01 -5.26431203e-02 4.76887137e-01 -2.48939440e-01 -8.68727326e-01 -3.71972889e-01 -1.45657882e-01 -7.92358875e-01 -8.15088391e-01 -1.03777993e+00 1.26551166e-01 3.87165278e-01 -2.81478819e-02 5.91911614e-01 -6.26887679e-01 2.11064607e-01 3.52814257e-01 -2.24088758e-01 -5.19355416e-01 -6.78263426e-01 -5.74230015e-01 1.94071457e-01 1.29537791e-01 3.82618397e-01 -2.59131998e-01 -9.72820744e-02 1.27413914e-01 -9.23048317e-01 -1.20003842e-01 2.51546890e-01 8.15655351e-01 6.39412582e-01 -5.10410219e-02 6.06042385e-01 -8.37051988e-01 6.54528022e-01 -4.54886883e-01 -1.00876927e+00 3.74902517e-01 -4.39447552e-01 4.27784503e-01 2.23000064e-01 -2.22825736e-01 -1.62376785e+00 -2.74530649e-01 -1.78781018e-01 -3.56079340e-01 -2.94562541e-02 7.49582231e-01 -2.85637289e-01 2.00376779e-01 4.79668498e-01 2.12986112e-01 -7.32194353e-03 -6.83624566e-01 2.78515100e-01 8.94964576e-01 6.99220896e-01 -8.05065513e-01 6.31639302e-01 6.16820514e-01 2.64635831e-01 -1.01238894e+00 -1.23204815e+00 -3.30819428e-01 -3.96381259e-01 -1.51774287e-01 8.38703454e-01 -5.78057408e-01 -5.68712473e-01 6.26478791e-01 -1.41379893e+00 -2.50609845e-01 -1.88780993e-01 1.11971748e+00 -8.59790266e-01 7.49591529e-01 -9.06984806e-02 -1.49218249e+00 4.79657464e-02 -1.08973908e+00 9.79149342e-01 1.75662264e-01 -2.36054316e-01 -1.01899755e+00 5.14637291e-01 -1.53542370e-01 2.01827407e-01 2.15835720e-01 6.41503751e-01 -3.66447538e-01 -3.88727844e-01 -2.35069573e-01 -4.28626925e-01 5.23513496e-01 6.34098575e-02 5.50337255e-01 -8.69743168e-01 -3.23174655e-01 3.34778011e-01 1.38295189e-01 8.55196178e-01 9.44967210e-01 9.44562018e-01 -1.46508679e-01 -2.80825526e-01 6.59040987e-01 1.28749585e+00 4.05741681e-04 5.08722723e-01 -2.75561631e-01 5.12014747e-01 3.79914224e-01 8.88037980e-01 8.46671283e-01 1.65201828e-01 4.17641491e-01 -1.49767175e-01 5.27985692e-01 2.36716330e-01 -4.07946676e-01 4.34199810e-01 7.66781271e-01 1.72113508e-01 -3.23602676e-01 -6.84736729e-01 3.71771902e-01 -2.14062119e+00 -9.92900968e-01 -4.47773457e-01 2.45068264e+00 8.14178765e-01 -9.12360549e-02 -6.54192194e-02 -9.98631865e-02 1.11421335e+00 7.75785893e-02 -5.66780865e-01 -3.54806989e-01 4.11998965e-02 9.75245684e-02 5.01023829e-01 9.25671339e-01 -7.93807685e-01 7.46103287e-01 8.09985924e+00 1.27713513e+00 -4.16197777e-01 3.65795702e-01 1.65973932e-01 5.30809700e-01 -6.80741727e-01 2.80813515e-01 -1.00671756e+00 5.48581302e-01 9.94275272e-01 -4.35353249e-01 -1.50843011e-02 7.25738227e-01 6.54901743e-01 -5.54714799e-01 -8.59161794e-01 9.11139786e-01 -1.77368462e-01 -9.98522997e-01 8.22991282e-02 -6.93252981e-02 8.48947942e-01 -3.24376911e-01 -1.53889433e-01 2.36492410e-01 7.30315089e-01 -8.07599247e-01 7.88208306e-01 1.09604776e+00 7.10477650e-01 -7.81907022e-01 8.20842028e-01 4.78600591e-01 -9.53665853e-01 4.98341590e-01 -7.35414684e-01 1.20577671e-01 6.60048366e-01 1.17406344e+00 -7.18864858e-01 4.53200191e-01 5.58984220e-01 6.01358175e-01 8.20073858e-03 1.25295341e+00 -5.55188894e-01 9.26700175e-01 -5.88643014e-01 -3.07291299e-01 9.86735299e-02 -7.90852666e-01 1.10186327e+00 1.16549039e+00 8.26740742e-01 -1.34770498e-01 -5.21771982e-02 1.43259478e+00 4.13616091e-01 -1.19605422e-01 -4.88034517e-01 -4.63698767e-02 7.89719403e-01 8.06433082e-01 -6.05469644e-01 -6.67730272e-01 -2.24411964e-01 5.79649687e-01 4.85766958e-03 8.27386260e-01 -7.85238922e-01 -4.46665138e-01 4.14386570e-01 -2.58930683e-01 2.13474169e-01 -4.54285026e-01 -3.12547714e-01 -1.36106122e+00 -1.91220820e-01 -3.24330598e-01 2.37937272e-01 -1.01847613e+00 -1.28754985e+00 -1.40876815e-01 8.29987884e-01 -8.72450769e-01 -4.61444527e-01 -4.18929160e-01 -7.71241844e-01 1.18535638e+00 -1.19111288e+00 -8.52401257e-01 -2.03493431e-01 2.23737299e-01 2.59027511e-01 -1.73225075e-01 5.54281712e-01 -2.82526970e-01 -5.41000783e-01 3.26280177e-01 7.49016345e-01 -4.62318510e-01 7.44938374e-01 -1.49443281e+00 4.53429341e-01 9.55187440e-01 -3.58162969e-01 7.37097442e-01 1.35331333e+00 -1.13770509e+00 -1.22197592e+00 -9.44302797e-01 7.74101138e-01 -1.69364527e-01 6.21023417e-01 -2.24249348e-01 -9.81598914e-01 7.41701007e-01 -1.31989077e-01 -3.26692075e-01 3.88313413e-01 -3.51902023e-02 -2.32234687e-01 6.72613550e-03 -1.38154674e+00 8.12094569e-01 5.55804253e-01 -2.77560622e-01 -8.20421934e-01 3.42391074e-01 5.11943638e-01 4.03526500e-02 -8.44579756e-01 4.95829076e-01 6.85141623e-01 -9.10069287e-01 7.75495827e-01 -3.67636293e-01 -1.15800694e-01 -5.49777746e-01 -4.20085043e-01 -1.42127776e+00 -2.02846497e-01 -9.84067619e-01 -2.30175436e-01 1.28987324e+00 7.76615292e-02 -8.36718380e-01 5.24004161e-01 5.62931061e-01 -1.56000173e-02 -2.51088858e-01 -1.04950607e+00 -9.94929075e-01 7.74643198e-02 -8.45422804e-01 5.92408955e-01 3.85044008e-01 -2.58580863e-01 -4.11442854e-02 -5.40978611e-01 2.16619849e-01 1.46336806e+00 -2.66422272e-01 8.75989914e-01 -1.27538359e+00 -2.86252916e-01 -1.42098188e-01 -2.02589095e-01 -1.19025588e+00 3.44402790e-01 -5.11360109e-01 4.90357041e-01 -1.57492757e+00 5.84969938e-01 -8.02464336e-02 1.52823910e-01 -1.47054464e-01 -4.80160594e-01 -3.95896770e-02 -3.68217438e-01 -4.75773215e-02 -2.54076630e-01 9.44907188e-01 1.27579057e+00 1.62347496e-01 -1.02769304e-02 2.83578187e-01 -3.46784413e-01 9.07805026e-01 5.52739382e-01 -8.05753708e-01 -2.88025737e-01 1.50084034e-01 3.77904385e-01 2.26154551e-01 3.79700691e-01 -6.37155771e-01 9.00331885e-02 -4.25121188e-01 1.26349121e-01 -1.15927792e+00 3.13778311e-01 -4.73497272e-01 3.26323271e-01 2.42867440e-01 -1.54683337e-01 -4.15421665e-01 -2.16281638e-01 8.21380734e-01 -1.96637899e-01 -8.86574626e-01 9.77259099e-01 8.35111141e-02 -1.61637038e-01 2.50685096e-01 -7.07102954e-01 3.69961075e-02 5.90102255e-01 4.63463739e-03 4.17642519e-02 -7.27910876e-01 -7.17196286e-01 1.29724145e-01 3.36542159e-01 -4.15215671e-01 8.07469964e-01 -1.46429014e+00 -1.13699734e+00 6.42690212e-02 -1.75875098e-01 7.35429395e-03 2.47113124e-01 8.76021028e-01 -5.94674706e-01 1.08223289e-01 3.08981329e-01 -7.81326473e-01 -6.58363044e-01 1.83894575e-01 1.55242041e-01 8.32763612e-02 -4.08991933e-01 5.09107888e-01 6.94579184e-02 -5.79091072e-01 -1.31312162e-02 -3.25976342e-01 1.80052042e-01 -2.46218264e-01 7.05668807e-01 1.11543739e+00 -3.65924060e-01 -3.02269936e-01 -1.07684270e-01 4.42680717e-01 1.66173846e-01 -9.49651599e-01 1.03584456e+00 -3.81572992e-01 -2.40565851e-01 1.01948977e+00 1.18199241e+00 4.10889909e-02 -1.37517536e+00 -1.98213711e-01 7.53642395e-02 -6.64445341e-01 1.60158306e-01 -2.20380068e-01 -2.93491960e-01 7.89016604e-01 1.81662634e-01 2.74321884e-01 4.29446101e-01 -1.41395420e-01 3.49369824e-01 3.43469977e-01 2.37875789e-01 -9.86508012e-01 -2.49689639e-01 5.29288411e-01 9.31704164e-01 -1.25760806e+00 2.15431660e-01 -4.67804492e-01 -2.70165801e-01 9.70820904e-01 5.58806770e-02 -2.98957288e-01 9.59243178e-01 2.41267934e-01 -1.80489212e-01 1.89899746e-02 -3.29217613e-01 6.52731955e-02 3.56794357e-01 7.98868597e-01 2.61489749e-01 9.26564783e-02 -6.91519320e-01 7.37295389e-01 -6.24369346e-02 -5.89444712e-02 5.96526325e-01 6.20442390e-01 -6.70460343e-01 -7.39564955e-01 -7.30418563e-01 4.99415874e-01 -2.79976517e-01 2.08418611e-02 4.63523790e-02 5.90222597e-01 -4.33766872e-01 1.07472932e+00 1.51251376e-01 2.12083027e-01 -4.62910235e-02 1.11714318e-01 5.57207823e-01 -5.49051583e-01 3.95954609e-01 6.24139845e-01 -1.43920392e-01 -3.51724416e-01 -4.51825708e-01 -1.02321720e+00 -8.92126858e-01 -4.34453517e-01 -5.43100893e-01 4.94773120e-01 7.45413482e-01 1.16164732e+00 -2.28950843e-01 2.68820465e-01 5.08628070e-01 -8.72346282e-01 -1.18524873e+00 -1.28294086e+00 -1.22784102e+00 6.69600889e-02 2.30340898e-01 -8.86299968e-01 -7.55053580e-01 -2.48859301e-01]
[7.0150933265686035, 3.9851067066192627]
bb677445-8c6f-4f7e-875d-8c19068e560c
a-robust-predictive-model-for-stock-price
1912.077
null
https://arxiv.org/abs/1912.07700v1
https://arxiv.org/pdf/1912.07700v1.pdf
A Robust Predictive Model for Stock Price Prediction Using Deep Learning and Natural Language Processing
Prediction of future movement of stock prices has been a subject matter of many research work. There is a gamut of literature of technical analysis of stock prices where the objective is to identify patterns in stock price movements and derive profit from it. Improving the prediction accuracy remains the single most challenge in this area of research. We propose a hybrid approach for stock price movement prediction using machine learning, deep learning, and natural language processing. We select the NIFTY 50 index values of the National Stock Exchange of India, and collect its daily price movement over a period of three years (2015 to 2017). Based on the data of 2015 to 2017, we build various predictive models using machine learning, and then use those models to predict the closing value of NIFTY 50 for the period January 2018 till June 2019 with a prediction horizon of one week. For predicting the price movement patterns, we use a number of classification techniques, while for predicting the actual closing price of the stock, various regression models have been used. We also build a Long and Short-Term Memory - based deep learning network for predicting the closing price of the stocks and compare the prediction accuracies of the machine learning models with the LSTM model. We further augment the predictive model by integrating a sentiment analysis module on twitter data to correlate the public sentiment of stock prices with the market sentiment. This has been done using twitter sentiment and previous week closing values to predict stock price movement for the next week. We tested our proposed scheme using a cross validation method based on Self Organizing Fuzzy Neural Networks and found extremely interesting results.
['Sidra Mehtab', 'Jaydip Sen']
2019-12-09
null
null
null
null
['stock-price-prediction']
['time-series']
[-7.47520924e-01 -5.11615515e-01 -3.17637205e-01 -3.12229156e-01 -2.17029184e-01 -6.40425801e-01 5.88811696e-01 8.83250684e-02 -5.94534934e-01 9.28519547e-01 3.14036340e-01 -5.90870976e-01 -1.10389784e-01 -1.57814205e+00 -5.59016466e-01 -4.97572303e-01 -2.82727808e-01 2.81727970e-01 7.59491250e-02 -8.25577676e-01 8.52201641e-01 4.33192939e-01 -1.48330450e+00 3.18600178e-01 3.33463788e-01 1.67182875e+00 -5.91500811e-02 2.28506848e-01 -5.23109376e-01 1.16820514e+00 -5.41961193e-01 -2.37778589e-01 9.07221913e-01 1.24587603e-01 -4.88956243e-01 -3.95247638e-01 -3.24665666e-01 -4.49336052e-01 7.89813846e-02 9.90539134e-01 8.10186937e-02 -2.91434936e-02 2.52781153e-01 -1.11682987e+00 -7.61715710e-01 1.04433978e+00 -4.76532370e-01 8.17637265e-01 -1.93793595e-01 -2.83388525e-01 1.14111304e+00 -8.08335900e-01 3.62161368e-01 6.90362990e-01 8.13231587e-01 -1.67269319e-01 -6.66832149e-01 -1.07448816e+00 -6.59064651e-02 1.04795590e-01 -9.87541258e-01 9.16749015e-02 7.36976683e-01 -7.22077608e-01 1.13570249e+00 3.49411070e-02 1.11439645e+00 3.21549326e-01 7.22956717e-01 4.29818749e-01 1.40807974e+00 -2.65711486e-01 2.82284051e-01 4.87081289e-01 2.91583061e-01 8.90010074e-02 2.24839419e-01 2.29791895e-01 -1.78345650e-01 -1.66003704e-01 3.33303273e-01 5.33209622e-01 2.21235380e-01 7.87739456e-01 -9.40895915e-01 1.14113235e+00 5.65126359e-01 7.75172651e-01 -8.28003824e-01 -2.11469010e-01 3.59427243e-01 8.73049021e-01 9.77285206e-01 5.64400733e-01 -1.05670190e+00 -1.59734502e-01 -1.46811748e+00 6.69879079e-01 9.95611370e-01 1.78647593e-01 7.28629172e-01 4.76652801e-01 3.86328548e-01 3.70518416e-01 2.97443002e-01 5.14676809e-01 1.20892668e+00 -1.78884640e-01 2.84500837e-01 7.59337902e-01 3.46031785e-01 -1.25169659e+00 -4.80964035e-01 -4.25348461e-01 -4.76936936e-01 5.26018858e-01 1.63239226e-01 -6.80967689e-01 -6.24992907e-01 9.99221623e-01 -3.25356036e-01 3.96838099e-01 4.61836219e-01 4.43299502e-01 4.16401088e-01 1.24200380e+00 -3.23133320e-02 -3.29419196e-01 1.00068378e+00 -5.48961818e-01 -7.42074966e-01 1.48709565e-01 5.31689227e-01 -6.13306344e-01 2.98141718e-01 2.49008030e-01 -9.86105561e-01 -4.70492810e-01 -1.07674861e+00 5.12868285e-01 -1.29196048e+00 -1.33816585e-01 3.91155362e-01 2.86151648e-01 -1.06625652e+00 1.14344227e+00 -6.60286903e-01 1.79454699e-01 6.13965914e-02 6.51430488e-01 2.87024051e-01 1.06216621e+00 -1.75347042e+00 1.18491626e+00 7.61944890e-01 4.34192605e-02 3.05862248e-01 -7.34747052e-01 -4.84830767e-01 9.86800995e-03 -2.78837174e-01 -4.52789431e-03 1.02211130e+00 -1.29750884e+00 -1.39853919e+00 5.38362622e-01 3.00070256e-01 -1.37178123e+00 3.86294067e-01 6.48313612e-02 -9.32100058e-01 -4.37589228e-01 8.55139866e-02 2.71948010e-01 4.85381842e-01 -5.06638944e-01 -1.21167767e+00 -3.08941007e-01 -2.39273254e-02 -1.39117941e-01 -5.46542525e-01 7.38070011e-02 5.34345567e-01 -1.02805758e+00 -2.81426888e-02 -8.60636353e-01 -5.93957119e-02 -9.11361992e-01 7.96478465e-02 -3.92788500e-01 9.36834812e-01 -8.45290661e-01 1.39322138e+00 -1.96138108e+00 -7.24995017e-01 6.06343865e-01 -3.27559561e-01 1.96371481e-01 4.39732969e-01 4.28733885e-01 -3.90179038e-01 2.94753432e-01 2.65911706e-02 2.02845961e-01 2.28843298e-02 -2.85207778e-02 -1.17771566e+00 1.66684881e-01 1.75629660e-01 1.09607863e+00 -2.15025052e-01 1.65563330e-01 1.82578877e-01 1.09202623e-01 -7.13560805e-02 -9.39421058e-02 -3.85628074e-01 -2.55346715e-01 -2.76238203e-01 5.31212151e-01 7.85052061e-01 -2.74374962e-01 -3.17144156e-01 -6.42847046e-02 -8.59346747e-01 4.89198923e-01 -1.01187539e+00 6.28073454e-01 -2.57016718e-01 5.70886135e-01 -6.23800278e-01 -8.97876740e-01 1.32372344e+00 3.23973447e-01 6.28912687e-01 -7.75653958e-01 4.03049231e-01 6.47974014e-01 1.67218626e-01 -1.88388526e-01 6.85955942e-01 -7.64296293e-01 -7.01065511e-02 7.93277144e-01 -3.69100451e-01 2.58705556e-01 2.22486466e-01 -5.10098517e-01 4.58555818e-01 -6.54273927e-02 2.12069035e-01 -5.19115269e-01 5.30726492e-01 2.70349860e-01 3.31132144e-01 1.69925645e-01 -1.19957469e-01 3.17770503e-02 2.47952580e-01 -1.21908629e+00 -9.38206196e-01 -5.10246396e-01 -3.25839967e-01 9.23429668e-01 -3.15317541e-01 3.47888887e-01 -2.71455884e-01 -5.13685122e-02 4.28620845e-01 9.40833807e-01 -8.47320557e-01 3.34825784e-01 -3.54116857e-01 -1.29124594e+00 2.76081026e-01 3.49742860e-01 7.31724381e-01 -1.51836872e+00 -6.97135627e-01 4.63268787e-01 4.49422359e-01 -7.57563829e-01 2.24041790e-01 4.26515698e-01 -8.87280405e-01 -6.90928280e-01 -6.50920093e-01 -8.15129399e-01 3.88808139e-02 -3.15196723e-01 9.46227074e-01 -2.19374020e-02 5.68125129e-01 -2.93719113e-01 -1.51301175e-01 -1.23525941e+00 -1.13408133e-01 1.70661300e-01 1.64385334e-01 1.28989726e-01 7.75342882e-01 -4.55054611e-01 -4.04063761e-01 -9.67310891e-02 -8.92509401e-01 -2.57910937e-01 4.29935902e-01 3.05945635e-01 3.65005344e-01 5.86741447e-01 1.12233961e+00 -7.23286748e-01 8.60074401e-01 -1.01660550e+00 -1.13560200e+00 -1.67662874e-01 -1.06414282e+00 7.03756977e-03 5.91568053e-01 -2.71334928e-02 -8.03503513e-01 -2.94251740e-01 -2.47302786e-01 -2.29233563e-01 3.46304476e-01 1.15530276e+00 7.43226945e-01 1.69222161e-01 3.99255529e-02 5.15784562e-01 1.66622862e-01 -3.50914210e-01 -2.66717553e-01 6.82460845e-01 2.13035166e-01 1.58942610e-01 8.99612546e-01 4.64799911e-01 -3.41048837e-01 -4.24484640e-01 -8.86553347e-01 -2.11435542e-01 -6.39398694e-01 -1.74480513e-01 6.37978137e-01 -1.12828481e+00 -6.92242861e-01 8.25970590e-01 -7.18364537e-01 6.82437345e-02 -2.94538081e-01 4.67620850e-01 -3.61679308e-02 -4.65502888e-01 -8.13227773e-01 -9.91530120e-01 -8.83200407e-01 -8.56921911e-01 2.99321562e-01 2.63637424e-01 -2.88765997e-01 -1.47985601e+00 4.09037948e-01 -7.24716187e-02 8.32163632e-01 5.06195903e-01 7.49879122e-01 -1.29933834e+00 -2.72367001e-01 -6.28316998e-01 -3.03993858e-02 5.18879831e-01 3.03288281e-01 7.57115260e-02 -7.20404148e-01 2.24940218e-02 4.65459555e-01 -7.82070756e-02 9.21225667e-01 6.30368948e-01 6.27104938e-01 -5.52317262e-01 2.29228705e-01 3.92310828e-01 1.79812169e+00 8.30151379e-01 5.38799107e-01 1.36404085e+00 2.21812502e-01 4.57627118e-01 5.25240541e-01 6.05743527e-01 5.35957217e-01 -9.49635208e-02 1.51600182e-01 2.09374741e-01 9.17135656e-01 3.26411612e-02 6.53980792e-01 1.05436659e+00 -2.56202310e-01 2.65497088e-01 -9.69520032e-01 4.36743051e-01 -1.53178203e+00 -1.33352554e+00 1.28550977e-01 1.85507417e+00 5.70403874e-01 7.10160494e-01 3.76194268e-01 5.38480163e-01 5.46516538e-01 4.56999868e-01 -4.88347352e-01 -5.80861926e-01 -2.47796461e-01 3.76046240e-01 1.08392358e+00 2.22610280e-01 -1.30963373e+00 7.84834623e-01 6.03360128e+00 4.38057035e-01 -1.89917612e+00 -3.88888985e-01 1.01626825e+00 -7.85624236e-02 -2.94080883e-01 -3.25999647e-01 -9.88530755e-01 9.31812465e-01 1.52814150e+00 -6.33863568e-01 1.76656753e-01 7.97879338e-01 5.87002754e-01 1.22239068e-02 -3.52771699e-01 5.74732184e-01 -1.94469601e-01 -2.00821996e+00 1.04964055e-01 1.92849725e-01 8.97250414e-01 3.95987988e-01 3.36049646e-01 3.85546118e-01 8.24084580e-02 -7.99373925e-01 8.97059143e-01 9.31883991e-01 -3.48616764e-02 -1.03168797e+00 1.24561536e+00 4.53916013e-01 -1.13847685e+00 -4.14758086e-01 -2.25293681e-01 -8.54101121e-01 -9.14130658e-02 5.30147552e-01 -6.73270464e-01 2.69045502e-01 9.90353942e-01 1.00004923e+00 -2.12381005e-01 5.48167944e-01 4.25903112e-01 5.52565098e-01 -3.90826464e-01 -4.52298790e-01 7.31498182e-01 -5.28057396e-01 8.75274539e-02 7.14373529e-01 6.76757932e-01 1.09168828e-01 -8.97323415e-02 8.00529838e-01 -1.13180570e-01 3.23494941e-01 -6.96250975e-01 -1.47196487e-01 2.53576398e-01 8.53537798e-01 -7.28716135e-01 -6.03708684e-01 -7.37718940e-01 1.57128096e-01 -2.13576779e-01 -6.98678643e-02 -4.88956869e-01 -4.21288908e-01 5.61769009e-01 3.63148421e-01 3.89857113e-01 5.38441911e-02 -8.02642703e-01 -1.06325376e+00 -1.12265401e-01 -3.58268201e-01 3.11334491e-01 -5.35783470e-01 -1.44088781e+00 6.69137239e-01 -1.88537896e-01 -1.63673663e+00 -4.83650446e-01 -6.39877796e-01 -1.12560821e+00 1.12787557e+00 -1.97527897e+00 -4.58032876e-01 4.87776250e-01 4.77000833e-01 4.30907279e-01 -9.14674997e-01 6.27619147e-01 4.52695563e-02 -3.76862347e-01 -1.40977174e-01 5.07043242e-01 6.62656784e-01 6.86861053e-02 -1.26677680e+00 7.23012328e-01 5.41096270e-01 -1.11902028e-01 4.60130543e-01 6.09179974e-01 -9.64601338e-01 -8.81399751e-01 -1.24626851e+00 1.15084219e+00 2.53467709e-02 1.54990602e+00 2.30689123e-01 -9.08690453e-01 8.88092220e-01 5.05833209e-01 -1.54342458e-01 8.06168973e-01 -5.10632098e-01 1.37982041e-01 -2.93606520e-01 -1.24497974e+00 1.18718699e-01 -3.13696086e-01 -2.18388453e-01 -1.09723175e+00 3.81836891e-02 4.43078637e-01 -4.81190383e-02 -1.13367093e+00 2.49735042e-01 7.80925095e-01 -1.10771406e+00 6.96126580e-01 -3.05237979e-01 3.42304736e-01 -1.77662715e-01 -2.92880647e-02 -1.41441047e+00 -2.46355489e-01 -1.78238809e-01 2.68191785e-01 9.96565044e-01 8.87554407e-01 -1.19568300e+00 9.30240333e-01 9.34493721e-01 3.27657461e-01 -9.17940080e-01 -6.45595193e-01 -4.65596974e-01 5.35947204e-01 -3.93408149e-01 1.12333536e+00 1.10088825e+00 -9.77333542e-03 -1.29974812e-01 -2.42583722e-01 4.27694954e-02 2.13213071e-01 7.25903153e-01 2.63691008e-01 -1.45665860e+00 1.51615351e-01 -6.04820490e-01 -3.25336099e-01 -2.47978792e-01 2.78013200e-01 -8.00583661e-01 -6.54828131e-01 -1.22388935e+00 -4.85822052e-01 -2.20831946e-01 -8.50360155e-01 4.38727885e-01 4.21493590e-01 1.41206414e-01 2.93720961e-01 6.70270026e-01 2.65499562e-01 2.72377729e-01 8.39906156e-01 -1.02628097e-01 -4.76719856e-01 3.41891527e-01 -5.66099882e-01 9.93697226e-01 1.20820653e+00 -1.75425708e-01 5.07160537e-02 -4.89499932e-03 6.62340522e-01 3.72038223e-03 -9.66988280e-02 -1.03596270e+00 9.73171219e-02 -2.34288454e-01 8.02372456e-01 -1.15359437e+00 -3.43760885e-02 -9.24164414e-01 2.10546464e-01 7.51309335e-01 -3.09633493e-01 7.53929496e-01 3.28574061e-01 6.77729473e-02 -8.76245558e-01 -1.39107749e-01 5.30603170e-01 -4.34072703e-01 -9.07310903e-01 3.27191770e-01 -5.83190501e-01 -3.25988501e-01 1.30617869e+00 -3.59288573e-01 -2.97235996e-02 -2.54797131e-01 -7.29277790e-01 2.74091244e-01 -8.47951416e-03 5.74316323e-01 4.38267618e-01 -1.38190842e+00 -6.92613006e-01 3.44361722e-01 -4.43003744e-01 -5.76880693e-01 -1.97983727e-01 4.51340199e-01 -9.53194022e-01 6.37353480e-01 -5.68779409e-01 1.09175012e-01 -5.19703448e-01 3.77680093e-01 4.90643829e-01 -3.99765044e-01 -3.47430736e-01 4.98849660e-01 -5.63674390e-01 -4.14842330e-02 -1.64854035e-01 -7.89194465e-01 -1.04119074e+00 1.01288331e+00 7.57875025e-01 3.29431385e-01 2.39756003e-01 -8.67725134e-01 -1.33788899e-01 6.86222553e-01 -6.04601689e-02 -8.87938067e-02 2.13381934e+00 5.68140969e-02 -4.09507513e-01 1.08619320e+00 1.32516456e+00 -2.30840057e-01 -7.50148118e-01 1.82687882e-02 7.62513101e-01 -2.40616258e-02 2.97598660e-01 -6.63494706e-01 -1.55010378e+00 3.41827303e-01 7.14114070e-01 8.89995039e-01 1.05818701e+00 -6.39166415e-01 1.36685956e+00 3.54691476e-01 2.44388387e-01 -1.30948234e+00 -7.43406832e-01 7.96332896e-01 7.25013077e-01 -1.39258242e+00 -7.79964551e-02 3.92921716e-01 -5.49015105e-01 1.41359973e+00 -4.37056646e-02 -9.10802007e-01 1.63923669e+00 5.68242192e-01 4.75529820e-01 -1.10826820e-01 -8.67077112e-01 4.74020429e-02 1.23550437e-01 -2.20565572e-01 4.22230273e-01 1.72654748e-01 -2.65568644e-01 9.21845257e-01 -9.32914674e-01 4.38799560e-01 7.00594306e-01 9.68056619e-01 -6.20920777e-01 -8.73197198e-01 -5.63411951e-01 9.74600434e-01 -1.23234177e+00 -3.62857789e-01 -1.18271679e-01 7.64239132e-01 -3.69897159e-03 6.67698741e-01 7.36561596e-01 -5.98171294e-01 1.27237797e-01 3.76848549e-01 -6.03880644e-01 -2.75566041e-01 -1.13931918e+00 -4.99082077e-03 -2.62508482e-01 7.89728835e-02 -6.86026931e-01 -8.67761910e-01 -1.46638966e+00 -7.38298059e-01 1.74695533e-02 2.70469993e-01 6.80368245e-01 1.33649874e+00 3.33461985e-02 1.22544602e-01 1.07336974e+00 -9.40238416e-01 -6.32953048e-01 -9.61320400e-01 -1.10110176e+00 2.50842124e-01 6.10912561e-01 -3.83547008e-01 -5.71950316e-01 4.35813628e-02]
[4.449123382568359, 4.246147632598877]
9fca886b-a416-4718-abf0-1af63ae2903c
quantifying-the-intrinsic-usefulness-of
2305.15961
null
https://arxiv.org/abs/2305.15961v1
https://arxiv.org/pdf/2305.15961v1.pdf
Quantifying the Intrinsic Usefulness of Attributional Explanations for Graph Neural Networks with Artificial Simulatability Studies
Despite the increasing relevance of explainable AI, assessing the quality of explanations remains a challenging issue. Due to the high costs associated with human-subject experiments, various proxy metrics are often used to approximately quantify explanation quality. Generally, one possible interpretation of the quality of an explanation is its inherent value for teaching a related concept to a student. In this work, we extend artificial simulatability studies to the domain of graph neural networks. Instead of costly human trials, we use explanation-supervisable graph neural networks to perform simulatability studies to quantify the inherent usefulness of attributional graph explanations. We perform an extensive ablation study to investigate the conditions under which the proposed analyses are most meaningful. We additionally validate our methods applicability on real-world graph classification and regression datasets. We find that relevant explanations can significantly boost the sample efficiency of graph neural networks and analyze the robustness towards noise and bias in the explanations. We believe that the notion of usefulness obtained from our proposed simulatability analysis provides a dimension of explanation quality that is largely orthogonal to the common practice of faithfulness and has great potential to expand the toolbox of explanation quality assessments, specifically for graph explanations.
['Pascal Friederich', 'Luca Torresi', 'Jonas Teufel']
2023-05-25
null
null
null
null
['graph-classification']
['graphs']
[ 3.81714165e-01 8.70394349e-01 -3.69264692e-01 -4.20788884e-01 4.42615300e-02 -5.26804745e-01 8.02109480e-01 6.38631344e-01 -1.52618244e-01 5.67451358e-01 7.41030350e-02 -8.11092913e-01 -7.22542942e-01 -8.71349573e-01 -7.34611213e-01 -3.02872568e-01 -5.57828322e-02 2.39672884e-01 -1.54273286e-01 -1.61288515e-01 4.55700666e-01 5.67503452e-01 -1.57034433e+00 -2.21863285e-01 1.35021544e+00 7.70583510e-01 -3.38936806e-01 4.89964426e-01 -2.48255823e-02 8.53601933e-01 -7.30547607e-01 -7.25193322e-01 -6.54468089e-02 -7.16562748e-01 -9.25505161e-01 6.93965778e-02 6.06039166e-01 -7.08120316e-02 -1.74132697e-02 1.18770576e+00 5.02558500e-02 2.59986758e-01 7.54984736e-01 -1.74371552e+00 -1.10738385e+00 1.16103733e+00 1.39329672e-01 3.19965094e-01 4.76050705e-01 1.56160966e-01 1.28714216e+00 -4.51195776e-01 3.48124206e-01 1.19160903e+00 6.61457896e-01 3.59297186e-01 -1.26479959e+00 -5.67212522e-01 2.61940241e-01 3.12539369e-01 -7.78910160e-01 -1.51679173e-01 9.55706537e-01 -3.94273549e-01 4.98391986e-01 3.73735279e-01 8.58613372e-01 1.16201234e+00 1.40783340e-01 3.55765849e-01 1.31753683e+00 -6.65357351e-01 5.04134655e-01 4.25669461e-01 8.08149099e-01 8.32030058e-01 8.94308507e-01 2.99742281e-01 -5.61778903e-01 -1.96911488e-02 7.90647805e-01 -4.09807898e-02 -4.24880236e-01 -5.21904707e-01 -9.97679591e-01 1.14156842e+00 7.50906467e-01 2.05998480e-01 -2.80860007e-01 3.71386796e-01 8.45810920e-02 4.36591566e-01 3.31780195e-01 1.05071187e+00 -1.33484244e-01 -5.89526184e-02 -5.61965168e-01 1.10771947e-01 8.35667551e-01 6.25564039e-01 6.99609578e-01 4.55497116e-01 -2.89452933e-02 4.47580040e-01 3.68023604e-01 4.30215672e-02 4.23253030e-01 -1.02919459e+00 5.43567091e-02 9.69177842e-01 -1.91198960e-01 -1.31069815e+00 -3.91432017e-01 -8.34303677e-01 -6.29291654e-01 3.15275252e-01 6.71414554e-01 7.86529034e-02 -5.27963638e-01 1.83263218e+00 1.22312613e-01 1.25389956e-02 7.73294047e-02 9.59638119e-01 9.58235264e-01 2.38620453e-02 2.28090182e-01 5.39185889e-02 1.20641792e+00 -8.98271382e-01 -7.01213062e-01 -4.18613374e-01 5.76310039e-01 -4.02389139e-01 1.58692431e+00 2.25797430e-01 -9.11061227e-01 -4.63550061e-01 -1.15859032e+00 8.28502998e-02 -4.49078023e-01 -2.18057662e-01 1.23665404e+00 9.66893792e-01 -9.03265715e-01 9.04164791e-01 -4.55298632e-01 -4.19771880e-01 2.56021738e-01 4.47901905e-01 -3.44003439e-01 5.50065227e-02 -1.25144732e+00 1.10421932e+00 2.33956024e-01 -5.04198112e-03 -5.65544307e-01 -5.82266808e-01 -9.32380497e-01 5.89639962e-01 3.78020853e-01 -7.66518593e-01 9.03086901e-01 -1.29477024e+00 -1.28850091e+00 4.19120878e-01 1.49367571e-01 -5.43419063e-01 2.43122995e-01 7.00320080e-02 -3.41336429e-02 3.88267376e-02 -7.86382481e-02 4.07731205e-01 5.77718556e-01 -1.24396694e+00 1.52275681e-01 -2.18136787e-01 5.22484124e-01 -7.34575242e-02 -5.85087180e-01 -5.49082696e-01 1.82022408e-01 -6.85017228e-01 2.01906398e-01 -1.09393811e+00 -1.70910031e-01 -8.66646841e-02 -3.88445675e-01 -2.34073982e-01 1.88726231e-01 -8.25610533e-02 1.04237115e+00 -1.83856332e+00 1.30453959e-01 3.32842052e-01 6.24688208e-01 -1.16361424e-01 -1.28010303e-01 2.99742311e-01 -4.15881038e-01 3.74909461e-01 -9.11270902e-02 -4.22586352e-02 3.24047804e-01 1.11935467e-01 -2.83089340e-01 3.80062521e-01 1.16477631e-01 1.09863269e+00 -8.35541487e-01 -2.22717285e-01 2.84896553e-01 5.02328694e-01 -6.06736422e-01 1.73803240e-01 2.18591671e-02 1.41606005e-02 -4.89703685e-01 3.20816427e-01 1.70845643e-01 -5.32646477e-01 9.95838046e-02 -4.73808907e-02 3.41044664e-01 4.92838204e-01 -9.99394476e-01 9.80815828e-01 -3.49505484e-01 9.61064696e-01 -4.79001850e-01 -1.01730335e+00 9.55073297e-01 1.32761136e-01 -2.22878337e-01 -5.69950640e-01 3.54370058e-01 2.59511340e-02 5.64876020e-01 -1.97998807e-01 6.52513146e-01 -3.49358976e-01 1.72161117e-01 7.47132659e-01 1.06392682e-01 -3.39239895e-01 -9.66313947e-03 2.95177996e-01 9.83280897e-01 -8.39228630e-02 7.88765848e-01 -5.59662819e-01 3.46831322e-01 -3.78633663e-03 8.96804035e-02 8.33256125e-01 -3.12380373e-01 2.55159229e-01 8.81790221e-01 -4.94538665e-01 -6.91715598e-01 -9.14597690e-01 5.27760871e-02 7.99314380e-01 1.87247675e-02 -2.94426829e-01 -8.37154150e-01 -8.41348112e-01 4.14142683e-02 1.21491587e+00 -1.08641613e+00 -5.68991482e-01 -1.43111032e-02 -4.06432778e-01 5.03494561e-01 4.98105675e-01 1.74788907e-01 -1.15518963e+00 -6.94773734e-01 -1.30682439e-01 5.23201451e-02 -9.07597125e-01 -2.45169494e-02 3.54359388e-01 -1.14872110e+00 -1.23146117e+00 -2.91074008e-01 -2.28298247e-01 9.03643548e-01 5.38740754e-01 1.32563365e+00 8.89547884e-01 2.52286971e-01 6.56708002e-01 -3.63437504e-01 -4.09656614e-01 -6.18208110e-01 -3.89959551e-02 2.07046438e-02 -5.31226993e-01 3.44225645e-01 -7.65367806e-01 -3.57042819e-01 1.93601698e-01 -8.76053274e-01 -9.21252370e-02 6.32216632e-01 8.32037747e-01 1.22550584e-01 9.71013457e-02 6.72552645e-01 -1.16257238e+00 1.26588368e+00 -4.53817159e-01 -5.33001304e-01 1.82752803e-01 -1.48050952e+00 4.08053100e-01 7.42889881e-01 -6.10583603e-01 -6.88033342e-01 -4.52825665e-01 3.06432128e-01 -1.70040131e-01 -2.60982484e-01 7.86163032e-01 1.46943673e-01 -4.49723989e-01 1.11220336e+00 -1.55317321e-01 1.27977028e-01 2.01113015e-01 6.07104003e-01 1.59596521e-02 1.59832403e-01 -6.65695488e-01 1.08113933e+00 -1.00756846e-01 3.07916939e-01 -6.18558586e-01 -8.37308824e-01 2.42027953e-01 -2.92197108e-01 -5.42353272e-01 6.35655999e-01 -3.73291820e-01 -1.06884205e+00 -2.80283511e-01 -9.86909688e-01 -3.74280274e-01 -3.81080985e-01 5.01470804e-01 -4.07102913e-01 5.12432158e-01 -4.03520912e-01 -7.84003079e-01 -5.94892688e-02 -1.41812241e+00 4.39112782e-01 3.46791476e-01 -7.50341594e-01 -1.48496354e+00 -1.59142569e-01 4.69617516e-01 4.84969884e-01 2.96041369e-01 1.25661922e+00 -9.45486188e-01 -7.34032571e-01 -2.85974383e-01 -2.13218138e-01 6.45083338e-02 6.54548481e-02 -5.43965921e-02 -9.84094739e-01 1.03256881e-01 -1.49557889e-01 -4.15016025e-01 7.05649197e-01 3.70648146e-01 9.75910962e-01 -5.27177989e-01 8.52892026e-02 9.60465670e-02 1.25064087e+00 -1.80380210e-01 4.13334936e-01 5.59199750e-01 6.75045252e-01 1.02613473e+00 4.35135603e-01 7.48907924e-02 3.64242613e-01 5.28429627e-01 7.50721872e-01 -2.76730582e-02 -1.59878805e-01 -4.09962147e-01 2.54211783e-01 5.64060450e-01 -1.54572785e-01 -5.22132814e-01 -9.11678195e-01 3.50465804e-01 -1.58944273e+00 -8.38763177e-01 -3.98181677e-01 2.19288683e+00 3.21229666e-01 4.10519719e-01 1.52038904e-02 4.10671443e-01 4.48434204e-01 1.61159895e-02 -2.43835330e-01 -5.93685687e-01 2.10168049e-01 -7.13069439e-02 2.80343890e-01 7.80796587e-01 -3.76380682e-01 7.07354486e-01 6.91597795e+00 3.78339678e-01 -8.23831201e-01 -3.30527306e-01 5.82438350e-01 4.35612619e-01 -8.48838568e-01 2.15552166e-01 -1.19200416e-01 -1.30994925e-02 1.09051692e+00 -3.56436729e-01 5.33965111e-01 1.07075620e+00 1.34703279e-01 6.27193004e-02 -1.42184830e+00 3.68029356e-01 -2.46716440e-02 -1.26579261e+00 2.15297207e-01 2.66409189e-01 4.91416931e-01 -5.93352079e-01 2.72223711e-01 2.26569206e-01 3.27007234e-01 -1.32676136e+00 7.31353164e-01 2.55721658e-01 2.09865719e-01 -5.65813661e-01 8.48865390e-01 2.20574826e-01 -6.65883839e-01 5.65871634e-02 -3.16414893e-01 -6.78397238e-01 -4.66056287e-01 5.22506058e-01 -1.17004049e+00 3.30260277e-01 1.81078643e-01 5.09694219e-01 -1.09693813e+00 6.44470632e-01 -7.61031151e-01 9.30968523e-01 3.75386775e-02 -4.30362463e-01 5.28113842e-02 -2.30995789e-01 3.76269609e-01 7.92659879e-01 2.48500273e-01 1.16253927e-01 -3.54783863e-01 1.33439410e+00 4.10030829e-04 1.12020299e-01 -9.48051691e-01 -4.10620362e-01 5.54879189e-01 1.25528133e+00 -1.14237905e+00 -2.34466851e-01 -1.79677591e-01 6.49710655e-01 5.98504066e-01 3.66027743e-01 -8.28390241e-01 -8.98323804e-02 3.48966479e-01 -7.33925328e-02 -1.66288912e-01 -8.10585991e-02 -7.43350267e-01 -9.24750090e-01 -2.28336543e-01 -1.13772428e+00 3.09324533e-01 -8.97462487e-01 -1.14567184e+00 7.02029824e-01 6.11382090e-02 -1.06017160e+00 -3.39904130e-01 -6.17596328e-01 -9.03920650e-01 7.05407381e-01 -1.30453014e+00 -8.75146985e-01 -7.42487550e-01 1.53680176e-01 3.16419333e-01 -4.70462181e-02 8.57037902e-01 -3.04125994e-01 -6.03579283e-01 6.61995292e-01 -6.93338037e-01 -2.43972957e-01 2.79225171e-01 -1.58334315e+00 3.68397593e-01 9.26944971e-01 3.80204231e-01 1.17631745e+00 1.32698643e+00 -5.09300828e-01 -1.11803019e+00 -6.02117717e-01 7.58897126e-01 -5.18909216e-01 7.16558993e-01 -8.11884478e-02 -1.15427566e+00 8.17409754e-01 2.83594161e-01 -2.66681582e-01 7.46772885e-01 5.97702503e-01 -3.53386611e-01 3.35298806e-01 -9.22696233e-01 8.35621774e-01 9.66543019e-01 -4.01992172e-01 -8.48757744e-01 1.42784998e-01 9.23139572e-01 -3.45496647e-02 -9.81380045e-01 8.48011076e-02 4.61280674e-01 -1.32130611e+00 8.84642899e-01 -6.53924406e-01 8.12116265e-01 2.79857777e-02 7.22500607e-02 -1.59856021e+00 -4.21193331e-01 -3.00297499e-01 1.56841606e-01 1.24012828e+00 6.57615840e-01 -8.36081684e-01 1.00061703e+00 1.10184884e+00 -5.76581135e-02 -4.88702089e-01 -5.89038789e-01 -6.84930265e-01 1.08157776e-01 -5.46892405e-01 7.38327205e-01 1.30925286e+00 2.35994682e-01 5.15562415e-01 -9.83346924e-02 2.51111329e-01 8.01761329e-01 9.80266184e-02 8.60302687e-01 -1.64937925e+00 -3.88969064e-01 -7.32968330e-01 -4.22212660e-01 -6.22878432e-01 6.32635176e-01 -1.06576490e+00 -2.31244683e-01 -1.45777690e+00 1.10632218e-01 -2.54689902e-01 -1.43072739e-01 4.61077124e-01 -4.53882366e-01 -1.56340224e-03 2.32234761e-01 -1.90323312e-02 -2.83564717e-01 6.68619990e-01 1.15847862e+00 5.86201362e-02 3.64937298e-02 -1.08735271e-01 -1.06034541e+00 7.78982520e-01 7.99580693e-01 -6.57680988e-01 -8.83543193e-01 -1.40580386e-01 4.98478442e-01 4.52079736e-02 7.38388538e-01 -7.73448229e-01 4.72556651e-02 -3.68445009e-01 1.72662005e-01 2.27448002e-01 4.22289446e-02 -9.77166414e-01 1.47775918e-01 6.07597172e-01 -8.06583524e-01 2.57508993e-01 1.43547088e-01 6.01105690e-01 -4.68986481e-02 -4.47458029e-01 4.38019574e-01 1.02098905e-01 -4.09644485e-01 -1.02777317e-01 -3.05066109e-01 -1.29079044e-01 7.01469362e-01 -4.01652336e-01 -6.15483701e-01 -8.35416079e-01 -4.09850866e-01 -1.68620214e-01 5.95994830e-01 2.94580936e-01 6.81354165e-01 -1.24282169e+00 -2.37423182e-01 -3.53797600e-02 2.29562581e-01 -6.57060087e-01 -8.75128880e-02 6.64723277e-01 -2.88998842e-01 4.24423337e-01 -3.90064150e-01 -3.40109646e-01 -1.09155393e+00 6.61728740e-01 4.09887463e-01 -5.18885590e-02 -4.72422004e-01 4.89967495e-01 2.08817512e-01 -4.78714526e-01 2.21578717e-01 -5.56415975e-01 -4.41570699e-01 -1.48025289e-01 1.17118508e-01 5.32483935e-01 -7.89094046e-02 -2.23617986e-01 -2.52040118e-01 1.91792294e-01 3.29863876e-01 -4.15501148e-02 1.19148934e+00 7.81618878e-02 6.42330870e-02 5.12051940e-01 5.49333751e-01 7.48362839e-02 -9.07244086e-01 3.51036608e-01 1.15595147e-01 -3.81423414e-01 1.50035083e-01 -8.41763198e-01 -1.00368631e+00 9.10887361e-01 2.32090935e-01 8.03900003e-01 8.38850439e-01 -2.43932694e-01 -6.94190040e-02 5.55599511e-01 8.96246433e-02 -4.70957190e-01 2.97301710e-01 5.93586527e-02 9.24306214e-01 -1.33486140e+00 3.34627271e-01 -6.18553460e-01 -5.58693171e-01 1.33405590e+00 6.02072239e-01 -1.14348739e-01 2.29737282e-01 -2.73080766e-01 5.70127070e-02 -5.90523005e-01 -6.97504938e-01 4.56549693e-03 6.45184338e-01 6.21885478e-01 9.08441126e-01 1.30167469e-01 -4.50366557e-01 7.19964564e-01 -6.77775979e-01 -1.82494745e-01 1.13669038e+00 2.21587151e-01 -3.12602729e-01 -8.89252365e-01 -3.54870856e-01 3.45850050e-01 -1.21073708e-01 -1.55311033e-01 -9.04322624e-01 1.25359273e+00 -4.11372721e-01 1.24937844e+00 -3.75487059e-01 -4.40655828e-01 3.49600047e-01 1.28337562e-01 3.13522369e-01 -4.60473806e-01 -7.41267145e-01 -6.60426915e-01 2.50045359e-01 -5.82149327e-01 -4.95170444e-01 -3.53798062e-01 -1.25324404e+00 -5.79168439e-01 -6.40984714e-01 4.03237283e-01 5.47409713e-01 1.12428045e+00 1.80676311e-01 7.70848155e-01 1.56654462e-01 -3.71530414e-01 -5.45179009e-01 -9.53937292e-01 -3.18722248e-01 4.69737232e-01 2.00732186e-01 -1.05818820e+00 -9.00173426e-01 -4.22230512e-01]
[8.611167907714844, 5.94423246383667]