WhenAllWeNeedisaPieceofthePie:AGenericFrameworkforOptimizingTwo-wayPartialAUCZhiyongYang12QianqianXu3ShilongBao12YuanHe4XiaochunCao12QingmingHuang3567Abstractframework.TheAreaUndertheROCCurve(AUC)i...
OneforOne,orAllforAll:EquilibriaandOptimalityofCollaborationinFederatedLearningAvrimBlum1NikaHaghtalab2RichardLanasPhillips3HanShao1Abstractpitals(Wenetal.,2019;Powell,2019)anddevices(McMa-han&Rama...
NotAllMemoriesareCreatedEqual:LearningtoForgetbyExpiringSainbayarSukhbaatar1DaJu1SpencerPoff1StephenRoller1ArthurSzlam1JasonWeston1AngelaFan12AbstractSukhbaataretal.,2019a).However,acriticalcompone...
IsSpace-TimeAttentionAllYouNeedforVideoUnderstanding?GedasBertasius1HengWang1LorenzoTorresani12AbstractVideounderstandingsharesseveralhigh-levelsimilaritieswithNLP.FirstofAll,videosandsentencesareb...
AttentionisnotAllyouneed:pureattentionlosesrankdoublyexponentiAllywithdepthYiheDong1Jean-BaptisteCordonnier2AndreasLoukas3Abstractattentionlayers.Surprisingly,wefindthatpureself-attentionnetworks(S...
RiggingtheLottery:MakingAllTicketsWinnersUtkuEvci1TrevorGale1JacobMenick2PabloSamelCastro1ErichElsen2AbstractFigure1.RigLimprovestheoptimizationofsparseneuralnet-worksbyleveragingweightmagnitudeand...
RandomizedSmoothingofAllShapesandSizesGregYang1TonyDuan12J.EdwardHu12HadiSalman1IlyaRazenshteyn1JerryLi1Abstractheuristicdefensesthatarerobusttospecificclassesofper-turbations,butmanywouldlaterbebr...
OneSizeFitsAll:CanWeTrainOneDenoiserforAllNoiseLevels?AbhiramGnansambandam1StanleyH.Chan12AbstractarguablyuniversalforAlllearning-basedestimators.Whensuchaproblemarises,themoststraight-forwardsolut...
AllintheExponentialFamily:BregmanDualityinThermodynamicVariationalInferenceRobBrekelmans1VadenMasrani2FrankWood2GregVerSteeg1AramGalstyan1AbstractFigure1.TheoriginalTVOpaperrecommendedusingtwoparti...
SubmodularStreaminginAllItsGlory:TightApproximation,MinimumMemoryandLowAdaptiveComplexityEhsanKazemi1MarkoMitrovic1MortezaZadimoghaddam2SilvioLattanzi2AminKarbasi1Abstractnon-negativesetfunctionf:2...
NotAllSamplesAreCreatedEqual:DeepLearningwithImportanceSamplingAngelosKatharopoulos12Franc¸oisFleuret12Abstractmodel.Tothisend,weproposeanovelimportancesamplingschemethatacceleratesthetrainingofan...
DeepLinearNetworkswithArbitraryLoss:AllLocalMinimaAreGlobalThomasLaurent1JamesH.vonBrecht2Abstractdeepnonlinearnetworksduringtraining(Saxeetal.,2014).ResultsofthissortprovideasmAllpieceofevidenceth...