ANovelMethodtoSolveNeuralKnapsackProblemsDuanshunLi1JingLiu2DongeunLee3AliSeyedmazloom4GiridharKaushik4KookjinLee5NoseongPark6Abstracttoastudy,KPsareoneofthetop-20mostpopularprob-lems(Kellereretal....
VariationalBayesianQuantizationYiboYang1RobertBamler1StephanMandt1Abstractforaparticularcompressionobjective.Here,westudyarelatedbutdifferentproblem:givenatrainedmodel,whatWeproposeanovelalgorithmf...
TransformationofReLU-basedrecurrentNeuralnetworksfromdiscrete-timetocontinuous-timeZahraMonfared1DanielDurstewitz12AbstractMorerecently,inthenaturalsciences,biologyandphysicsinparticular,RNNswereal...
TrainingNeuralNetworksforandbyInterpolationLeonardBerrada1AndrewZisserman2M.PawanKumar2Abstractdimensionalnon-convexfunctions.Inpractice,themainalgorithmsofchoiceareStochasticGradientDescent(SGD)In...
TrainingLinearNeuralNetworks:Non-LocalConvergenceandComplexityResultsArminEftekhari1AbstractbythelinearmapLinearnetworksprovidevaluableinsightsintoRdx→RdytheworkingsofNeuralnetworksingeneral.Thisp...
TrainingBinaryNeuralNetworksusingtheBayesianLearningRuleXiangmingMeng1RomanBachmann2MohammadEmtiyazKhan1Abstractbyreplacingthemultiplicationandadditionoperationswiththebit-wisexnorandbitcountoperat...
TrainingBinaryNeuralNetworksthroughLearningwithNoisySupervisionKaiHan12YunheWang2YixingXu2ChunjingXu2EnhuaWu13ChangXu4AbstractFigure1.Frameworkoflearningbinaryneuronswithnoisysuper-vision.Anetworkf...
TowardsaGeneralTheoryofInfinite-WidthLimitsofNeuralClassifiersEugeneA.Golikov1Abstractofinfinitewidth.Inparticular,(Jacotetal.,2018)showedthatifweightsareparameterizedinacertainwaythentheObtainingt...
TheNeuralTangentKernelinHighDimensions:TripleDescentandaMulti-ScaleTheoryofGeneralizationBenAdlam†1JeffreyPennington1AbstractClassicalregimeAbundantparameterizationSuperabundantparameterizationMod...
T-Basis:aCompactRepresentationforNeuralNetworksAntonObukhov1MaximRakhuba1StamatiosGeorgoulis1MenelaosKanakis1DengxinDai1LucVanGool12Abstractthatcontainmillionsoftrainableparametersandconsumelotsofm...
SuperpolynomialLowerBoundsforLearningOne-LayerNeuralNetworksusingGradientDescentSurbhiGoel1AravindGollakota1ZhihanJin2SushrutKarmalkar1AdamKlivans1AbstractOurResults.Inthispaperwegivethefirstsuperp...
SpectralClusteringwithGraphNeuralNetworksforGraphPoolingFilippoMariaBianchi1DanieleGrattarola2CesareAlippi23AbstractMessage-passingMinCutPoolMessage-passingSpectralclustering(SC)isapopularclusterin...
SDE-Net:EquippingDeepNeuralNetworkswithUncertaintyEstimatesLingkaiKong1JimengSun2ChaoZhang1Abstractbilitiesforclassification(Guoetal.,2017).Moreover,theycanmakeerroneousyetwildlyconfidentprediction...
RobustGraphRepresentationLearningviaNeuralSparsificationChengZheng1BoZong2WeiCheng2DongjinSong2JingchaoNi2WenchaoYu2HaifengChen2WeiWang1Abstracttioninsocialnetworks(Akogluetal.,2015),recommen-datio...
Retro:LearningRetrosyntheticPlanningwithNeuralGuidedASearchBinghongChen1ChengtaoLi2HanjunDai3LeSong14AbstractExistingmethodsroughlyfallintotwocategories,eithertemplate-basedortemplate-free.Eachchem...
RethinkingBias-VarianceTrade-offforGeneralizationofNeuralNetworksZitongYang1YaodongYu1ChongYou1JacobSteinhardt12YiMa1Abstractfromamismatchbetweenthemodelclassandtheunder-lyingdatadistribution,andis...
OptimizationTheoryforReLUNeuralNetworksTrainedwithNormalizationLayersYonatanDukler1QuanquanGu2GuidoMontúfar134AbstractofadditionalnormalizationmethodsfollowedBN,notablyincludingLayerNormalization(...
OptimaltransportmappingviainputconvexNeuralnetworksAshokVardhanMakkuva1AmirhosseinTaghvaei2JasonD.Lee3SewoongOh4Abstract1.IntroductionInthispaper,wepresentanovelandprincipledFindingamappingthattran...
OntheExpressivityofNeuralNetworksforDeepReinforcementLearningKefanDong1YupingLuo2TianheYu3ChelseaFinn3TengyuMa3Abstractwiththeestimateddynamics(Nagabandietal.,2018;Chuaetal.,2018;Wang&Ba,2019).Weco...
Non-AutoregressiveNeuralText-to-SpeechKainanPeng∗1WeiPing∗1ZhaoSong∗1KexinZhao∗1Abstractgram.Thispipelinerequiresmuchlessexpertknowledgeandonlyneedspairsofaudioandtranscriptastrainingdata.Inthi...