TheoryofSpectralMethodforUnionofSubspaces-BasedRandomGeometryGraphGenLi1YuantaoGu1AbstracttaskinprocessingdatapointsinUoSistoclusterthesedatapoints,whichisknownasSubspaceClustering(SC).Appli-Spectr...
OnlineOptimizationinGamesviaControlTheory:ConnectingRegret,PassivityandPoincare´RecurrenceYunKuenCheung1GeorgiosPiliouras2AbstractTheoryandonlineoptimization.Wepresentanovelcontrol-theoreticunders...
MessagePassingAdaptiveResonanceTheoryforOnlineActiveSemi-supervisedLearningTaehyeongKim12InjuneHwang1HyundoLee2HyunseoKim2Won-SeokChoi2JosephJ.Lim13Byoung-TakZhang2Abstractuseofdeeplearninginreal-w...
Few-shotLanguageCoordinationbyModelingTheoryofMindHaoZhu1GrahamNeubig1YonatanBisk1Abstractprocess,withawidevarietyofworksexaminingcommuni-cationbetweenagentsviaeithercompletelyartificialemer-Nomani...
f-Domain-AdversarialLearning:TheoryandAlgorithmsDavidAcuna123GuojunZhang43MarcT.Law1SanjaFidler123Abstractlabeledandunlabeleddatasets,anditisexpectedtogen-eralizewelltothetargetdatasetifthegapbetwe...
ApproximationTheoryofConvolutionalArchitecturesforTimeSeriesModellingHaotianJiang1ZhongLi2QianxiaoLi13AbstractTherecentworkofLietal.(2021)formulatedthetempo-ralmodellingtaskintoafunctionalanalysisp...
ApproximationTheoryBasedMethodsforRKHSBanditsShoTakemori1MasahiroSato1AbstracttheadversarialRKHSbanditproblem,wherealearnerinter-actswithasequenceofanyfunctionsfromtheRKHSwithTheRKHSbanditproblem(a...
ATheoryofHighDimensionalRegressionwithArbitraryCorrelationsbetweenInputFeaturesandTargetFunctions:SampleComplexity,MultipleDescentCurvesandaHierarchyofPhaseTransitionsGabrielC.Mel1SuryaGanguli2Abst...
ATheoryofLabelPropagationforSubpopulationShiftTianleCai12RuiqiGao12JasonD.Lee1QiLei1Abstracttees.Manyoftheworksarebasedonthecovariateshiftas-sumption(Heckman,1979;Shimodaira,2000),whichstatesOneoft...
TowardsaGeneralTheoryofInfinite-WidthLimitsofNeuralClassifiersEugeneA.Golikov1Abstractofinfinitewidth.Inparticular,(Jacotetal.,2018)showedthatifweightsareparameterizedinacertainwaythentheObtainingt...
RandomMatrixTheoryProvesthatDeepLearningRepresentationsofGAN-dataBehaveasGaussianMixturesMohamedElAmineSeddik12CosmeLouart13MohamedTamaazousti1RomainCouillet23Abstractontopofitthatareaccurateforthe...
OptimizationTheoryforReLUNeuralNetworksTrainedwithNormalizationLayersYonatanDukler1QuanquanGu2GuidoMontúfar134AbstractofadditionalnormalizationmethodsfollowedBN,notablyincludingLayerNormalization(...
AUnifiedTheoryofDecentralizedSGDwithChangingTopologyandLocalUpdatesAnastasiaKoloskova1NicolasLoizou2SadraBoreiri1MartinJaggi1SebastianU.Stich1Abstractetal.,2016;2017;Kairouzetal.,2019)hasemerged,bu...
RateDistortionForModelCompression:FromTheoryToPracticeWeihaoGao1Yu-HanLiu2ChongWang3SewoongOh4Abstracttationalpower,thenumberofweightsinpracticalneuralnetworkmodelalsogrowsrapidly.Forexample,inthea...
ObtainingFairnessusingOptimalTransportTheoryEustasiodelBarrio1FabriceGamboa2PaulaGordaliza21Jean-MichelLoubes2Abstractwereferto(Zafaretal.,2017)or(Friedleretal.,2018).Inthefairclassificationsetup,w...
BridgingTheoryandAlgorithmforDomainAdaptationYuchenZhang12TianleLiu13MingshengLong12MichaelI.Jordan4AbstractRemarkabletheoreticaladvanceshavebeenachievedindo-mainadaptation.Mansouretal.(2009c);Ben-...
ATheoryofRegularizedMarkovDecisionProcessesMatthieuGeist1BrunoScherrer2OlivierPietquin1AbstractTsallisentropy(Leeetal.,2018),withthemotivationofhavingasparseregularizedgreedypolicy.OtherapproachesM...
AKernelTheoryofModernDataAugmentationTriDao1AlbertGu1AlexanderJ.Ratner1VirginiaSmith2ChristopherDeSa3ChristopherRe´1Abstractasregularizertomaketheresultingmodelmorerobust,andprovideresourcestodata...
AConvergenceTheoryforDeepLearningviaOver-ParameterizationZeyuanAllen-Zhu1YuanzhiLi23ZhaoSong456Abstract1IntroductionDeepneuralnetworks(DNNs)havedemon-Neuralnetworkshavedemonstratedagreatsuccessinst...
OntheTheoryofVarianceReductionforStochasticGradientMonteCarloNiladriS.Chatterji1NicolasFlammarion2Yi-AnMa2PeterL.Bartlett23MichaelI.Jordan23AbstractsothatefficientapproachestoMonteCarlosamplingarer...