WhatAreBayesianNeuralNetworkPosteriorsReallyLike?PavelIzmailov1SharadVikram2MatthewD.Hoffman2AndrewGordonWilson1AbstractforNeuralnetworkspromisesimprovedpredictions,reli-ableuncertaintyestimates,an...
Weight-CovarianceAlignmentforAdversariallyRobustNeuralNetworksPanagiotisEustratiadis1HenryGouk1DaLi12TimothyHospedales12Abstract2013).Inordertotacklethisproblem,manyadversarialdefensealgorithmshave...
WatermarkingDeepNeuralNetworkswithGreedyResidualsHanwenLiu1ZhenyuWeng1YueshengZhu1Abstractandprivacyissues(Shokrietal.,2017;Songetal.,2017;Salemetal.,2019),andthusthemodelsareconsideredtoDeepNeural...
UnsupervisedRepresentationLearningviaNeuralActivationCodingYookoonPark1SanghoLee2GunheeKim2DavidM.Blei1Abstract(a)Atinitialization(b)AfterNACtrainingWepresentNeuralactivationcoding(NAC)asaFigure1.D...
TrainingRecurrentNeuralNetworksviaForwardPropagationThroughTimeAnilKag1VenkateshSaligrama1Abstractempiricalriskfunction:Back-propagationthroughtime(BPTT)hasbeen[W∗,v∗]=argminL(W,v)=1NT(yi,yˆti)w...
TrainingQuantizedNeuralNetworkstoGlobalOptimalityviaSemidefiniteProgrammingBurakBartan1MertPilanci1Abstractdimensionsusingsemidefiniteprogramming(Bartan&Pi-lanci,2021).Inthiswork,wetakeasimilarconv...
TrainingGraphNeuralNetworkswith1000LayersGuohaoLi12MatthiasMu¨ller1BernardGhanem2VladlenKoltun1AbstractRevGNN-WideResGNN-224DeepgraphNeuralnetworks(GNNs)have88achievedexcellentresultsonvarioustask...
TowardsUnderstandingLearninginNeuralNetworkswithLinearTeachersRoeiSarussi1AlonBrutzkus1AmirGloberson1Abstractwillseparatethedata.Whichofthesewillbefoundbygradientdescent?CanaNeuralnetworkminimizing...
TowardsCertifying∞RobustnessusingNeuralNetworkswith∞-distNeuronsBohangZhang1TianleCai23ZhouLu4DiHe5LiweiWang16Abstract1.IntroductionItiswell-knownthatstandardNeuralnetworks,ModernNeuralnetworksar...
TheHintonsinyourNeuralNetwork:aQuantumFieldTheoryViewofDeepLearningRobertoBondesan1MaxWelling1AbstractInthispaperweproposeadirectmappingofadeepneu-ralnetworkontoaopticalquantumcomputerthroughtheInt...
TensorProgramsIIb:ArchitecturalUniversalityofNeuralTangentKernelTrainingDynamicsGregYang1EtaiLittwin2AbstractIn(Yang,2020a),theNTKINITpropertywasproventoholdforstandardarchitectures,meaninganycompo...
RevealingtheStructureofDeepNeuralNetworksviaConvexDualityTolgaErgen1MertPilanci1Abstractwheretwo-layerReLUnetworkswiththeminimumEu-clideannormsolutionandzerotrainingerrorareprovenWestudyregularized...
RethinkingNeuralvs.Matrix-FactorizationCollaborativeFiltering:theTheoreticalPerspectivesDaXu1ChuanweiRuan2EvrenKorpeoglu1SushantKumar1KannanAchan1Abstractexploredthankstotheirinterpretabilityandcom...
ProvableGeneralizationofSGD-trainedNeuralNetworksofAnyWidthinthePresenceofAdversarialLabelNoiseSpencerFrei1YuanCao2QuanquanGu2AbstractdefineWeconsideraone-hidden-layerleakyReLUnet-OPT:=minP(x,y)∼D...
Privacy-PreservingVideoClassificationwithConvolutionalNeuralNetworksSikhaPentyala1RafaelDowsley2MartineDeCock13AbstractMeinich-Bacheetal.,2020;Pengetal.,2020;Shahetal.,2019;Wuetal.,2019).Manyexisti...
PrincipledSimplicialNeuralNetworksforTrajectoryPredictionT.MitchellRoddenberry1NicholasGlaze1SantiagoSegarra1Abstractintheirabilitytoincorporatearbitrarypairwiserelationalstructuresintheircomputati...
PhaseTransitions,DistanceFunctions,andImplicitNeuralRepresentationsYaronLipman12AbstractFigure1.ImplicitNeuralrepresentationslearnedwiththenovelPHASEloss.Thisnewlossunifiesoccupancyandsigneddistanc...
OptimizationofGraphNeuralNetworks:ImplicitAccelerationbySkipConnectionsandMoreDepthKeyuluXu1MozhiZhang2StefanieJegelka1KenjiKawaguchi3AbstracttheoreticalaspectsofGNNstounderstandtheirsuccessandlimi...
OpeningtheBlackbox:AcceleratingNeuralDifferentialEquationsbyRegularizingInternalSolverHeuristicsAvikPal12YingboMa2ViralShah2ChristopherRackauckas2345AbstractFigure1.TrainingandPredictionPerformance...