UnderstandingNoiseInjectioninGANsRuiliFeng1DeliZhao2Zheng-JunZha†1Abstractquantizingneuralnetworks(Baskinetal.,2018).Inrecentyears,Noiseinjection1hasattractedmoreandmoreatten-Noiseinjectionisaneff...
UnderstandingInstance-LevelLabelNoise:DisparateImpactsandTreatmentsYangLiu1Abstractframework(Feldman,2020),wedemonstratethevaryingeffectsofmemorizingnoisylabelsassociatedwithinstancesThispaperaimst...
SolvingInverseProblemswithaFlow-basedNoiseModelJayWhang1QiLei2AlexandrosG.Dimakis3Abstractplausiblesolutionamongthem.SparsityhasclassicallybeenaveryinfluentialstructuralpriorforvariousinverseWestud...
Positive-NegativeMomentum:ManipulatingStochasticGradientNoisetoImproveGeneralizationZekeXie1LiYuan2ZhanxingZhu3MasashiSugiyama41AbstractItiswell-knownthatstochasticgradientNoise(SGN)instochasticopt...
OntheInherentRegularizationEffectsofNoiseInjectionDuringTrainingOussamaDhifallah1YueM.Lu1Abstract2008)underGaussianinputandperturbationvectors.OuranalysisparticularlyshowsthatGaussianNoiseinjection...
NoiseandFluctuationofFiniteLearningRateStochasticGradientDescentKangqiaoLiu1LiuZiyin1MasahitoUeda123AbstractandTeh,2011).WhentheNoiseisduetominibatchsam-pling,theNoiseiscalledtheSGDNoiseorminibatch...
MultiplicativeNoiseandHeavyTailsinStochasticOptimizationLiamHodgkinson1MichaelW.Mahoney1Abstract(orquenching)perspective,examinesconvergenceeitherinexpectation(Chenetal.,2019;Zhouetal.,2018;GowerAl...
LearningtoGenerateNoiseforMulti-AttackRobustnessDivyamMadaan1JinwooShin23SungJuHwang134Abstractasaccurate,sinceincorrectpredictionsmayleadtosevereconsequences.Notably,itiswell-knownthattheexistingA...
LearningNoiseTransitionMatrixfromOnlyNoisyLabelsviaTotalVariationRegularizationYivanZhang12GangNiu2MasashiSugiyama21AbstractBackground.Earlystudiesonlearningfromnoisylabelscanbetracedbacktotherando...
HomomorphicSensing:SparsityandNoiseLiangzuPeng1BoshiWang1ManolisC.Tsakiris1AbstractAssumingtheexistenceofak-sparsesolutionx∗to(1),thefirstquestioniswhetherx∗isunique.TheanswerisUnlabeledsensingis...
Class2Simi:ANoiseReductionPerspectiveonLearningwithNoisyLabelsSonghuaWu1XiaoboXia1TongliangLiu1BoHan2MingmingGong3NannanWang4HaifengLiu5GangNiu6AbstractdealwiththelabelNoiseprobleminpointwisemanner...
Streamingk-SubmodularMaximizationunderNoisesubjecttoSizeConstraintLanN.Nguyen1MyT.Thai1AbstractAnotherreadilyapplicationisSensorPlacementwithktypesofmeasures(Ohsaka&Yoshida,2015;Qianetal.,Maximizin...
OntheGeneralizationBenefitofNoiseinStochasticGradientDescentSamuelL.Smith1ErichElsen1SohamDe1Abstractbatches,orbecauseauthorsoftencomparedifferentbatchsizesunderaconstantepochbudget(suchthatsmallba...
FeatureNoiseInducesLossDiscrepancyAcrossGroupsFereshteKhani1PercyLiang1Abstractdiscrepancy(Liptonetal.,2018)andadverselong-termimpacts(Liuetal.,2018).Therefore,itisimportanttoTheperformanceofstanda...
EfficientlyLearningAdversariallyRobustHalfspaceswithNoiseOmarMontasser1SurbhiGoel2IliasDiakonikolas3NathanSrebro1AbstractTheinformation-theoreticaspectsofadversariallyrobustlearninghavebeenstudiedi...
DoesLabelSmoothingMitigateLabelNoise?MichalLukasik1SrinadhBhojanapalli1AdityaKrishnaMenon1SanjivKumar1AbstractInterestingly,therearetwocompetingintuitions.Ontheonehand,smoothingmightmitigatetheprob...
BeyondSyntheticNoise:DeepLearningonControlledNoisyLabelsLuJiang1DiHuang2MasonLiu3WeilongYang1AbstractHowever,duetothelackofsuitabledatasets,previousworkhasonlyexaminedDNNsoncontrolledsyntheticlabel...
UnsupervisedLabelNoiseModelingandLossCorrectionEricArazo1DiegoOrtego1PaulAlbert1NoelE.O’Connor1KevinMcGuinness1Abstract3Clean4DespitebeingrobusttosmallamountsoflabelLoss2Noisy3Noise,convolutionaln...
TowardsUnderstandingtheImportanceofNoiseinTrainingNeuralNetworksMoZhou1TianyiLiu2YanLi2DachaoLin1EnluZhou2TuoZhao2AbstractSimplefirstorderalgorithmssuchasStochasticGradientDescent(SGD)anditsvariant...
TheAnisotropicNoiseinStochasticGradientDescent:ItsBehaviorofEscapingfromSharpMinimaandRegularizationEffectsZhanxingZhu123JingfengWu1BingYu1LeiWu1JinwenMa1Abstract90Understandingthebehaviorofstochas...