ExaminingandCombatingSpuriousFeaturesunderDistributionShiftChuntingZhou1XuezheMa2PaulMichel1GrahamNeubig1Abstract(a)ERM(b)GroupDRO[clean](average=93.3%robust=83.4%)(average=96.5%robust=95.8%)Acentr...
RobustnesstoSpuriousCorrelationsviaHumanAnnotationsMeghaSrivastava1TatsunoriHashimoto1PercyLiang1Abstractcasestudiessuchasface-recognitionsystemsperformingpoorlyonpopulationswithdarkskincolor(Buola...
AnInvestigationofWhyOverparameterizationExacerbatesSpuriousCorrelationsShioriSagawa1AditiRaghunathan1PangWeiKoh1PercyLiang1AbstractUnderparameterizedOverparameterizedWestudywhyoverparameterization...
Passed&Spurious:DescentAlgorithmsandLocalMinimainSpikedMatrix-TensorModelsStefanoSaraoMannelli1FlorentKrzakala2PierfrancescoUrbani1LenkaZdeborova´1AbstractRecentyearsbroughtapopularlineofresearchi...
SpuriousLocalMinimaareCommoninTwo-LayerReLUNeuralNetworksItaySafran1OhadShamir1Abstractlearning,andtensordecomposition,donothaveSpuriouslocalminimaundersuitableassumptions,inwhichcaselo-Weconsidert...
NoSpuriousLocalMinimainNonconvexLowRankProblems:AUnifiedGeometricAnalysisRongGe1ChiJin2YiZheng1Abstract2016;Parketal.,2016)andmatrixcompletion(Geetal.,2016)havewell-behavedoptimizationlandscape:all...