QueryComplexityofAdversarialAttacksGrzegorzGłuch1Ru¨digerUrbanke1Abstract2018;Hayes&Danezis,2017)canonlyobservetheinput-outputbehaviorofthemodel.ManydefenseshavebeenTherearetwomainattackmodelscon...
QuantitativeUnderstandingofVAEasaNon-linearlyScaledIsometricEmbeddingAkiraNakagawa1KeizoKato1TaijiSuzuki23Abstractconvergestoanappropriatevalue,andsoon.Alemietal.(2018)introducedtheRDtrade-offbased...
QuantifyingtheBenefitofUsingDifferentiableLearningoverTangentKernelsEranMalach1PritishKamath2EmmanuelAbbe3NathanSrebro2AbstractCanallthesuccessofdeeplearningbeexplainedusingtheNTK?Thiswouldimplytha...
ProvablyEfficientLearningofTransferableRewardsAlbertoMariaMetelli1GiorgiaRamponi1AlessandroConcetti1MarcelloRestelli1Abstracttheoretically,underthestrongassumptionofrewardunique-ness(Abbeel&Ng,2004...
ProvableMeta-LearningofLinearRepresentationsNileshTripuraneni1ChiJin2MichaelI.Jordan1Abstractrangingfromdeepreinforcementlearning(Baevskietal.,2019)tonaturallanguageprocessing(Ando&Zhang,2005;Meta-...
ProvableRobustnessofAdversarialTrainingforLearningHalfspaceswithNoiseDifanZou1SpencerFrei2QuanquanGu1AbstractToformalizetheabovecomment,letusdefinethero-Weanalyzethepropertiesofadversarialtrain-bus...
ProvableGeneralizationofSGD-trainedNeuralNetworksofAnyWidthinthePresenceofAdversarialLabelNoiseSpencerFrei1YuanCao2QuanquanGu2AbstractdefineWeconsideraone-hidden-layerleakyReLUnet-OPT:=minP(x,y)∼D...
Prediction-CentricLearningofIndependentCascadeDynamicsfromPartialObservationsMateuszWilinski1AndreyY.Lokhov1AbstractThismotivatesthedevelopmentofefficientalgorithmsthatcaninfertheeffectivespreading...
ParetoGAN:ExtendingtheRepresentationalPowerofGANstoHeavy-TailedDistributionsToddHuster1JeremyE.J.Cohen1ZinanLin2KevinChan3CharlesKamhoua3NandiLeslie4Cho-YuJasonChiang1VyasSekar2AbstractEvans&Erland...
OrderMatters:ProbabilisticModelingofNodeSequenceforGraphGenerationXiaohuiChen1XuHan1JiajingHu1FranciscoJ.R.Ruiz2LipingLiu1Abstract2018;Liaoetal.,2019;Daietal.,2020;Goyaletal.,2020;Yuanetal.,2020;Sh...
OptimizationofGraphNeuralNetworks:ImplicitAccelerationbySkipConnectionsandMoreDepthKeyuluXu1MozhiZhang2StefanieJegelka1KenjiKawaguchi3AbstracttheoreticalaspectsofGNNstounderstandtheirsuccessandlimi...
OptimalEstimationofHighDimensionalSmoothAdditiveFunctionBasedonNoisyObservationsFanZhou,PingLiCognitiveComputingLabBaiduResearch10900NE8thStBellevueWA98004USA{fanzhou,liping11}@baidu.comAbstractana...
OperationalizingComplexCauses:APragmaticViewofMediationLimorGultchin12DavidS.Watson3MattJ.Kusner4RicardoSilva32AbstractHistoricalDataLimitedNewDataWeexaminetheproblemofcausalresponsees-Z<latexitsha...
OnlineLearningforLoadBalancingofUnknownMonotoneResourceAllocationGamesAbstractIlaiBistritz1NicholasBambos12018;Boursier&Perchet,2019;Mehrabianetal.,2020;ConsiderNplayersthateachusesamixtureofNayyar...
Rate-DistortionAnalysisofMinimumExcessRiskinBayesianLearningHassanHafez-Kolahi1BehradMoniri2ShohrehKasaei1MahdiehSoleymaniBaghshah1Abstractinthisregardistostudylowerboundsontheexcessriskwhichcouldb...
OntheOptimalityofBatchPolicyOptimizationAlgorithmsChenjunXiao12YifanWu3TorLittlemore4BoDai2JinchengMei12LihongLi†5CsabaSzepesvari14DaleSchuurmans12Abstractafixeddatasetofpreviouslycollectedexperie...
OnthePowerofLocalizedPerceptronforLabel-OptimalLearningofHalfspaceswithAdversarialNoiseJieShen1Abstractthelearnermustpayforeachlabelitwishestoberevealed.Thegoalistodesignqueryingstrategiestoavoidle...
OntheInherentRegularizationEffectsofNoiseInjectionDuringTrainingOussamaDhifallah1YueM.Lu1Abstract2008)underGaussianinputandperturbationvectors.OuranalysisparticularlyshowsthatGaussiannoiseinjection...
OntheImplicitBiasofInitializationShape:BeyondInfinitesimalMirrorDescentShaharAzulay1EdwardMoroshko2MorShpigelNacson2BlakeWoodworth3NathanSrebro3AmirGloberson1DanielSoudry2Abstractparameterizedmodel...
OntheGeneralizationPowerofOverfittedTwo-LayerNeuralTangentKernelModelsPeizhongJu1XiaojunLin1NessB.Shroff2AbstractMuthukumaretal.,2019;Juetal.,2020),aninteresting“double-descent”phenomenonhasbeeno...