UncoveringtheConnectionsbetweenAdversarialTransferabilityandKnowledgeTransferabilityKaizhaoLiang1JackyY.Zhang1BoxinWang1ZhuolinYang1OluwasanmiKoyejo1BoLi1Abstractmulti-lingualmachinetranslation(Don...
TheSymmetrybetweenArmsandKnapsacks:APrimal-DualApproachforBanditswithKnapsacksXiaochengLi1ChunlinSun2YinyuYe2Abstractmarkproblemfordecisionmakingunderuncertaintythathasbeenstudiedfornearlyacentury....
ExactGapbetweenGeneralizationErrorandUniformConvergenceinRandomFeatureModelsZitongYang1YuBai2SongMei3AbstractBoundsonuniformconvergenceusuallytaketheformofcomplexity/n(Vapnik,1995),wherethenumerato...
CharacterizingtheGapbetweenActor-CriticandPolicyGradientJunfengWen1SaurabhKumar2RamkiGummadi3DaleSchuurmans13Abstractonarangeofchallengingtasks.DespitethesuccessofACmethods,ACandPGhavesubtlediffere...
AccuracyontheLine:OntheStrongCorrelationbetweenOut-of-DistributionandIn-DistributionGeneralizationJohnMiller1RohanTaori2AditiRaghunathan2ShioriSagawa2PangWeiKoh2VaishaalShankar1PercyLiang2YairCarmo...
UnderstandingandMitigatingtheTradeoffbetweenRobustnessandAccuracyAditiRaghunathan1SangMichaelXie1FannyYang2JohnC.Duchi1PercyLiang1AbstractStdErr(AT)-StdErr(Std)(%)ε=1/255Figure1.Gapbetweenthestan-...
OptimalBoundsbetweenf-DivergencesandIntegralProbabilityMetricsRohitAgrawal1ThibautHorel2Abstractdivergencebetweentwodistributionsµandνasanav-eragecostofthelikelihoodratio,thatis,Iφ(µν):=Thefam...
OntheRelationbetweenQuality-DiversityEvaluationandDistribution-FittingGoalinTextGenerationJianingLi12YanyanLan12JiafengGuo12XueqiCheng12Abstractbymaximumlikelihoodestimation(MLE)(Mikolovetal.,2010)...
IsThereaTrade-OffbetweenFairnessandAccuracy?APerspectiveUsingMismatchedHypothesisTestingSanghamitraDutta12DennisWei1HazarYueksel1Pin-YuChen1SijiaLiu1KushR.Varshney1Abstract2012;Agarwaletal.,2018;Ha...
InterpolationbetweenResidualandNon-ResidualNetworksZonghanYang1YangLiu1ChenglongBao2ZuoqiangShi3Abstracteringadifferentialequationthatidentifiestheruleoftheobserveddatabasedonthestandardblockofexis...
FundamentalTradeoffsbetweenInvarianceandSensitivitytoAdversarialPerturbationsFlorianTramèr1JensBehrmann2NicholasCarlini3NicolasPapernot3Jörn-HenrikJacobsen4AbstractAdversarialexamplesarecommonlyf...
BridgingtheGapbetweenf-GANsandWassersteinGANsJiamingSong1StefanoErmon1AbstractVariousGANlearningprocedureshavebeenproposedfordifferentdiscrepancymeasures.f-GANs(NowozinGenerativeadversarialnetworks...
ASampleComplexitySeparationbetweenNon-ConvexandConvexMeta-LearningNikunjSaunshi1YiZhang1MikhailKhodak2SanjeevArora13Abstractcomplexityofanunseenbutrelatedtesttask.Althoughthereisalonghistoryofsucce...
TheoreticallyPrincipledTrade-offbetweenRobustnessandAccuracyHongyangZhang12YaodongYu3JiantaoJiao4EricP.Xing15LaurentElGhaoui4MichaelI.Jordan4Abstractblocksforarangeofsecurity-criticalsystemsandappl...
OntheConnectionbetweenAdversarialRobustnessandSaliencyMapInterpretabilityChristianEtmann12SebastianLunz3PeterMaass1Carola-BibianeScho¨nlieb3AbstractFigure1.Animageofadog(left),thesaliencymapsofahi...
LeveragingLow-RankRelationsbetweenSurrogateTasksinStructuredPredictionGiuliaLuise1DimitrisStamos1MassimilianoPontil12CarloCiliberto13AbstractAmongthemostwell-establishedstrategiesforstructuredpredi...
Optimization,FastandSlow:OptimallySwitchingbetweenLocalandBayesianOptimizationMarkMcLeod1MichaelA.Osborne12StephenJ.Roberts12AbstractposteriorsproducedbytheGParemultivariateGaussian,soprovidebothth...
OntheRelationshipbetweenDataEfficiencyandErrorforUncertaintySamplingStephenMussmann1PercyLiang1Abstractpling(Yang&Loog,2016),nogains(Schein&Ungar,2007),gainsof2x(Tong&Koller,2001),andgainsof3xWhile...