APrecisePerformanceAnalysisofSupportVectorRegressionHoussemSifaou1AblaKammoun1Mohamed-SlimAlouini1AbstractofNakkiranetal.(Nakkiranetal.,2020a;b)indicatesthatmoredatamayharmtheperformance.Onepotenti...
ANewRepresentationofSuccessorFeaturesforTransferacrossDissimilarEnvironmentsMajidAbdolshah1HungLe1ThommenKarimpanalGeorge1SunilGupta1SantuRana1SvethaVenkatesh1Abstractintoindependentsub-domains.How...
VisualGroundingofLearnedPhysicalModelsYunzhuLi1ToruLin1KexinYi2DanielM.Bear3DanielL.K.Yamins3JiajunWu4JoshuaB.Tenenbaum5AntonioTorralba5Abstractdynamicsofphysicalsystemsappliestonotonlyrigidbod-ies...
VoiceSeparationwithanUnknownNumberofMultipleSpeakersEliyaNachmani12YossiAdi1LiorWolf12Abstractthispartialrepresentations.Sincetheorderofthespeakersisconsideredarbitrary(itishardtosortvoices),oneuse...
UnsupervisedDiscoveryofInterpretableDirectionsintheGANLatentSpaceAndreyVoynov1ArtemBabenko12Abstract-Thickness+ThelatentspacesofGANmodelsoftenhave-Hair+semanticallymeaningfuldirections.Mov-inginthe...
UniversalAverage-CaseOptimalityofPolyakMomentumDamienScieur1FabianPedregosa2Abstractmizationmethods.Thesimplexmethod,forexample,hasaworst-caseexponentialcomplexity,thatbecomespolyno-Polyakmomentum(...
UniquePropertiesofFlatMinimainDeepNetworksRotemMulayoff1TomerMichaeli1Abstractmodels(Gunasekaretal.,2018b)anddeepnonlinearnet-workswithhomogeneousactivationfunctions(Lyu&Li,Itiswellknownthat(stocha...
UniformConvergenceofRank-weightedLearningJustinKhim1LiuLeqi1AdarshPrasad1PradeepRavikumar1Abstractlearning,wherepercentilebasedriskmeasureshavebeenusedtoquantifythetail-riskofmodels.ArecentlineofTh...
UnderstandingtheImpactofModelIncoherenceonConvergenceofIncrementalSGDwithRandomReshuffeShaocongMa1YiZhou1Abstractthesamplelossofthei-thdatasample.Suchaproblemformulationcoversavarietyofmachinelearn...
UnderstandingtheCurseofHorizoninoff-PolicyEvaluationviaConditionalImportanceSamplingYaoLiu1Pierre-LucBacon2EmmaBrunskill1Abstractincreasinginterestindevelopingaccurateandefficientalgo-rithmsforoff-...
EstimatingGeneralizationunderDistributionShiftsviaDomain-InvariantRepresentationsChing-YaoChuang1AntonioTorralba1StefanieJegelka1AbstractAISafety(Amodeietal.,2016),motivatingrecentworkonestimatingt...
TransformationofReLU-basedrecurrentneuralnetworksfromdiscrete-timetocontinuous-timeZahraMonfared1DanielDurstewitz12AbstractMorerecently,inthenaturalsciences,biologyandphysicsinparticular,RNNswereal...
TowardsUnderstandingtheRegularizationofAdversarialRobustnessonNeuralNetworksYuxinWen12ShuaiLi1KuiJia12Abstract(Szegedyetal.,2014).Moreintriguingly,theseadversarialexamplesaregeneratedbyaddingimperc...
TowardsUnderstandingtheDynamicsoftheFirst-OrderAdversariesZhunDeng1HangfengHe2JiaoyangHuang3WeijieJ.Su2Abstractattackssuchasinputde-noising(Guoetal.,2017),random-izedsmoothing(Ilyasetal.,2019),grad...
TowardsaGeneralTheoryofInfinite-WidthLimitsofNeuralClassifiersEugeneA.Golikov1Abstractofinfinitewidth.Inparticular,(Jacotetal.,2018)showedthatifweightsareparameterizedinacertainwaythentheObtainingt...
TheRoleofRegularizationinClassificationofHigh-dimensionalNoisyGaussianMixtureFrancescaMignacco1FlorentKrzakala2YueM.Lu3PierfrancescoUrbani1LenkaZdeborova´1Abstractgeneratedasv∗√Weconsiderahigh-d...
TheSampleComplexityofBest-kItemsSelectionfromPairwiseComparisonsWenboRen1JiaLiu2NessB.Shroff13Abstract1.IntroductionThispaperstudiesthesamplecomplexity(aka1.1.BackgroundandMotivationnumberofcompari...
TheNon-IIDDataQuagmireofDecentralizedMachineLearningKevinHsieh12AmarPhanishayee1OnurMutlu32PhillipB.Gibbons2AbstractTheseconstraintsmotivatetheneedforMLtrainingoverwidelydistributeddata(decentraliz...
Thek-tiedNormalDistribution:ACompactParameterizationofGaussianMeanFieldPosteriorsinBayesianNeuralNetworksJakubSwiatkowski1+KevinRoth2+BastiaanS.Veeling34+LinhTran5+JoshuaV.Dillon4JasperSnoek4Stepha...
TheIntrinsicRobustnessofStochasticBanditstoStrategicManipulationZheFeng1DavidC.Parkes1HaifengXu2Abstractabletomodulateitsownrewardfeedbackinordertofurtheritsownobjective,e.g.,increasingthenumberoft...