What’sintheBox?ExploringtheInnerLifeofNeuralNetworkswithRobustRulesJonasFischer1AnnaOla´h1JillesVreeken2AbstractYaretypicallyactivewhenneuronsXare.Forrobustnessweexplicitlyallowfornoise,andtoensu...
Variational(Gradient)EstimateoftheScoreFunctioninEnergy-basedLatentVariableModelsFanBao12KunXu1ChongxuanLi1LanqingHong2JunZhu1BoZhang1Abstractsivepower(Salakhutdinov&Hinton,2009;Baoetal.,2020)andto...
UnderstandingtheDynamicsofGradientFlowinOverparameterizedLinearModelsSalmaTarmoun12GuilhermeFranc¸a13BenjaminHaeffele14Rene´Vidal14Abstractathoroughanalysisofthedynamicsofoptimizationmeth-ods,suc...
UncoveringtheConnectionsBetweenAdversarialTransferabilityandKnowledgeTransferabilityKaizhaoLiang1JackyY.Zhang1BoxinWang1ZhuolinYang1OluwasanmiKoyejo1BoLi1Abstractmulti-lingualmachinetranslation(Don...
UCBMomentumQ-learning:CorrectingthebiaswithoutforgettingPierreMénard1OmarDarwicheDomingues2XuedongShang23MichalValko234Abstractbalancetheexplorationoftheenvironmentandexploitationofthecurrentknowl...
TowardsTightBoundsontheSampleComplexityofAverage-rewardMDPsYujiaJin1AaronSidford1Abstractmakingunderuncertaintyandreinforcementlearning(Puter-man,2014;Sutton&Barto,2018).Itisaprominenttheoret-Wepro...
TowardstheUnificationandRobustnessofPerturbationandGradientBasedExplanationsSushantAgarwal1ShahinJabbari2ChiragAgarwal2SohiniUpadhyay2ZhiweiStevenWu3HimabinduLakkaraju2Abstractcanunderstandandconse...
TowardUnderstandingtheFeatureLearningProcessofSelf-supervisedContrastiveLearningZixinWen1YuanzhiLi2Abstractcanevenoutperformthoselearnedbysupervisedlearninginseveraldownstreamtasks.theremakablepote...
Tiltingtheplayingfield:DynamicallossfunctionsformachinelearningMiguelRuiz-Garcia12GeZhang1SamuelS.Schoenholz3AndreaJ.Liu1Abstracttionscale(Glorot&Bengio,2010;Xiaoetal.,2018)orlearningrateschedule(H...
TighterBoundsontheLogMarginalLikelihoodofGaussianProcessRegressionusingConjugateGradientsArtemArtemev12DavidR.Burt3MarkvanderWilk1Abstractdientbasedmethodsinordertoautomaticallyselectmodelhyperpara...
TighteningtheDependenceonHorizonintheSampleComplexityofQ-LearningGenLi1ChangxiaoCai2YuxinChen2YuantaoGu1YutingWei3YuejieChi4AbstractQ-learning(Borkar&Meyn,2000;Jaakkolaetal.,1994;Szepesva´ri,1998;...
TightBoundsontheSmallestEigenvalueoftheNeuralTangentKernelforDeepReLUNetworksQuynhNguyen1MarcoMondelli2GuidoMontufar13AbstractWeassumethatthenetworkhasasingleoutput,namelynL=1andWL∈RnL−1.Forconsi...
theSymmetrybetweenArmsandKnapsacks:APrimal-DualApproachforBanditswithKnapsacksXiaochengLi1ChunlinSun2YinyuYe2Abstractmarkproblemfordecisionmakingunderuncertaintythathasbeenstudiedfornearlyacentury....
thePowerofAdaptivityforStochasticSubmodularCoverRohanGhuge1AnupamGupta2ViswanathNagarajan1Abstractsoleimanetal.,2015;Batenietal.,2018):herearetwoexamplesfromsensordeploymentandmedicaldiagnosis.Inth...
thePowerofLog-Sum-Exp:SequentialDensityRatioMatrixEstimationforSpeed-AccuracyOptimizationTaikiMiyagawa1AkinoriF.Ebihara1AbstractMorietal.,2018).Earlyclassificationoftimeseriesisamulti-objectiveopti...
theLipschitzConstantofSelf-AttentionHyunjikKim1GeorgePapamakarios1AndriyMnih1Abstractconstraintforneuralnetworks,tocontrolhowmuchanet-work’soutputcanchangerelativetoitsinput.SuchLips-Lipschitzcons...
theLimitsofMin-MaxOptimizationAlgorithms:ConvergencetoSpuriousNon-CriticalSetsYa-PingHsieh1PanayotisMertikopoulos23VolkanCevher4AbstractGivenanalgorithmforsolving(SP),itisthennaturaltoComparedtoord...
theImplicitRegularizationforAdaptiveOptimizationAlgorithmsonHomogeneousNeuralNetworksBohanWang1QiMeng1WeiChen1Tie-YanLiu1Abstractprocessing(Youngetal.,2018).Inpractice,deepneuralnetworks(DNN)learne...
theHeavy-TailPhenomenoninSGDMertGürbüzbalaban1UmutS¸ims¸ekli2LingjiongZhu3Abstract1.IntroductionInrecentyears,variousnotionsofcapacityandthelearningprobleminneuralnetworkscanbeexpressedascomple...