Trainsimultaneously,generalizebetter:Stabilityofgradient-basedminimaxlearnersFarzanFarnia1AsumanOzdaglar1Abstract2014)andadversarialtraining(Madryetal.,2017)haveachievedgreatsuccessoverawidearrayof...
StabilityandConvergenceofStochasticGradientClipping:BeyondLipschitzContinuityandSmoothnessVienV.Mai1MikaelJohansson1Abstractproblemsareatthecoreofmanymachine-learningappli-cations,andareoftensolved...
StabilityandGeneralizationofStochasticGradientMethodsforMinimaxProblemsYunwenLei1ZhenhuanYang2TianbaoYang3YimingYing2Abstracting(Goodfellowetal.,2014),robustoptimization(Chenetal.,2017;Namkoong&Duc...
InterpretableStabilityBoundsforSpectralGraphFiltersHenryKenlay1DorinaThanou2XiaowenDong1Abstractaubiquitoustooldesignedtoprocessgraph-structureddata.Inadditiontovarioussignalprocessingtasks(Shumane...
LandscapeConnectivityandDropoutStabilityofSGDSolutionsforOver-parameterizedNeuralNetworksAlexanderShevchenko1MarcoMondelli1Abstractanover-parameterizednetworkviaSGDtypicallyleadstoasolutionthathass...
Fine-GrainedAnalysisofStabilityandGeneralizationforStochasticGradientDescentYunwenLei12YimingYing3Abstractaniterativealgorithm,SGDupdatesthemodelsequentiallyuponreceivinganewdatumwithacheapper-iter...
StabilityandGeneralizationofLearningAlgorithmsthatConvergetoGlobalOptimaZacharyCharles1DimitrisPapailiopoulos1Abstractrecentworkinanalyzingthegeneralizationcapacityofvar-iouslearningalgorithms,ourt...
Data-DependentStabilityofStochasticGradientDescentIljaKuzborskij1ChristophH.Lampert2Abstractticeonemightnotevenreachaminimum,yetneverthelessobservesexcellentperformance.Weestablishadata-dependentno...
AlgorithmicStabilityandHypothesisComplexityTongliangLiu1Ga´borLugosi234GergelyNeu5DachengTao1AbstractHardtetal.(2015)showedthatparametricmodelstrainedbystochasticgradientdescentalgorithmsareunifor...