OpeningtheBlackbox:AcceleratingNeuralDifferentialEquationsbyRegularizingInternalSolverHeuristicsAvikPal12YingboMa2ViralShah2ChristopherRackauckas2345AbstractFigure1.TrainingandPredictionPerformance...
On-PolicyDeepReinforcementLearningfortheAverage-RewardCriterionYimingZhang1KeithW.Ross21AbstractHaarnojaetal.,2018)orinaqueuingscenario(Tadepalli&Ok,1994;Sutton&Barto,2018),thereisnonaturalsep-Wede...
RecomposingtheReinforcementLearningBuildingBlockswithHypernetworksEladSarafian1ShaiKeynan1SaritKraus1AbstractResBlockmetavariablePrimarynetLinearBlock256ResBlocktheReinforcementLearning(RL)building...
OntheOptimalityofBatchPolicyOptimizationAlgorithmsChenjunXiao12YifanWu3TorLittlemore4BoDai2JinchengMei12LihongLi†5CsabaSzepesvari14DaleSchuurmans12Abstractafixeddatasetofpreviouslycollectedexperie...
OnthePowerofLocalizedPerceptronforLabel-OptimalLearningofHalfspaceswithAdversarialNoiseJieShen1Abstractthelearnermustpayforeachlabelitwishestoberevealed.thegoalistodesignqueryingstrategiestoavoidle...
OntheInherentRegularizationEffectsofNoiseInjectionDuringTrainingOussamaDhifallah1YueM.Lu1Abstract2008)underGaussianinputandperturbationvectors.OuranalysisparticularlyshowsthatGaussiannoiseinjection...
OntheImplicitBiasofInitializationShape:BeyondInfinitesimalMirrorDescentShaharAzulay1EdwardMoroshko2MorShpigelNacson2BlakeWoodworth3NathanSrebro3AmirGloberson1DanielSoudry2Abstractparameterizedmodel...
OntheGeneralizationPowerofOverfittedTwo-LayerNeuralTangentKernelModelsPeizhongJu1XiaojunLin1NessB.Shroff2AbstractMuthukumaretal.,2019;Juetal.,2020),aninteresting“double-descent”phenomenonhasbeeno...
OntheDifficultyofUnbiasedAlphaDivergenceMinimizationTomasGeffner1JustinDomke1AbstractExistingalpha-divergenceminimizationalgorithmscanbeclassifiedintotwobroadgroups:biasedmethods(Li&Severalapproxim...
OntheExplicitRoleofInitializationontheConvergenceandImplicitBiasofOverparametrizedLinearNetworksHanchengMin12SalmaTarmoun13Rene´Vidal14EnriqueMallada12Abstractwithoutexplicitregularization,enjoysg...
OntheConvergenceofHamiltonianMonteCarlowithStochasticGradientsDifanZou1QuanquanGu1AbstracttionssuchasBayesianinference,reinforcementlearning,andcomputervision.Inthepastdecades,manyMCMCHamiltonianMo...
OnPerceptualLossyCompression:theCostofPerceptualReconstructionandAnOptimalTrainingFrameworkZeyuYan1FeiWen1RendongYing1ChaoMa1PeilinLiu1Abstract2017;Santurkaretal.,2018;Shaham&Michaeli,2018).Forloss...
OntheRandomConjugateKernelandNeuralTangentKernelZhengmianHu1HengHuang1Abstractduetothedifficultiesraisedbythenon-convexityofthelossfunctionandthecomplicationofoptimizationmethods.Weinvestigatethedi...
OntheProofofGlobalConvergenceofGradientDescentforDeepReLUNetworkswithLinearWidthsQuynhNguyen1Abstracttrainingdata,thentheoutputatlayerlisgivenbyWegiveasimpleprooffortheglobalconver-genceofgradien...
OntheProblemofUnderrankinginGroup-FairRankingSruthiGorantla1AmitDeshpande2AnandLouis1Abstractethicalconcernsandcanpotentiallycauselong-termeco-nomicandsocietalharmtodemographicsandbusinessesBiasinr...
OnthepriceofexplainabilityforsomeclusteringproblemsEduardoLaber1LucasMurtinho1Abstractdecisiontreewith3leaves.Asanexample,theblueclustercanbeexplainedasthesetofpointsthatsatisfyFeaturethepriceofexp...
OnthePredictabilityofPruningAcrossScalesJonathanRosenfeld1JonathanFrankle1MichaelCarbin1NirShavit1AbstractAsafirsttry,wecouldattempttoanswerthisquestionusingbruteforce:wecouldpruneeverymemberofanet...
LowerBoundsonCross-EntropyLossinthePresenceofTest-timeAdversariesArjunNitinBhagoji1DanielCullina2VikashSehwag3PrateekMittal3Abstractonestablishingfundamentalboundsonlearninginthepres-enceoftest-tim...
MindtheBox:l1-APGDforSparseAdversarialAttacksonImageClassifiersFrancescoCroce1MatthiasHein1Abstractexistasetofl1-basedattacks(Chenetal.,2018;Modasetal.,2019;Brendeletal.,2019;Croce&Hein,2020a;Wesho...
MixedNashEquilibriaintheAdversarialExamplesGameLaurentMeunier12MeyerScetbon3RafaelPinot4JamalAtif1YannChevaleyre1AbstractAlongthisline,(Pinotetal.,2020)demonstrated,usinggametheory,thatrandomizedcl...