RepresentationMatters:OfflinePretrainingforSequentialDecisionMakingMengjiaoYang1OfirNachum1AbstractFigure1.Asummaryoftheadvantagesofrepresentationlearningviacontrastiveself-prediction,acrossavariet...
RepresentationMatters:AssessingtheImportanceofSubgroupAllocationsinTrainingDataEstherRolf1TheodoraWorledge1BenjaminRecht1MichaelI.Jordan12AbstractOurworkaimstodevelopageneralunifyingperspectiveonth...
OrderMatters:ProbabilisticModelingofNodeSequenceforGraphGenerationXiaohuiChen1XuHan1JiajingHu1FranciscoJ.R.Ruiz2LipingLiu1Abstract2018;Liaoetal.,2019;Daietal.,2020;Goyaletal.,2020;Yuanetal.,2020;Sh...
RandomizationMattersHowtodefendagainststrongadversarialattacksRafaelPinot12RaphaelEttedgui1GeovaniRizk1YannChevaleyre1JamalAtif1Abstractand(Szegedyetal.,2014)itbecameincreasinglyimportanttoundersta...
Goal-AwarePrediction:LearningtoModelWhatMattersSurajNair1SilvioSavarese1ChelseaFinn1Abstractheavysupervisionistoleveragedatathattheagentcancollectautonomouslywithoutrewardsorlabels,termedLearneddyn...
BiasAlsoMatters:BiasAttributionforDeepNeuralNetworkExplanationShengjieWang1TianyiZhou1JefferyA.Bilmes2Abstractfunctions,utilizingmulti-layerarchitecturesandarichsetofpossiblehiddennon-linearities,m...
WidthProvablyMattersinOptimizationforDeepLinearNeuralNetworksSimonS.Du1WeiHu2Abstractconvergestoglobalminimumunderfurtherassumptionsonbothdataandglobalminimum.TheseresultsrequireWeprovethatforanL-l...