StudentSpecializationinDeepRectifiedNetworksWithFiniteWidthandInputDimensionYuandongTian1Abstract1.IntroductionWeconsideradeepReLU/LeakyReLUstu-WhileDeepLearninghasachievedgreatsuccessindiffer-dent...
SpectralClusteringwithGraphNeuralNetworksforGraphPoolingFilippoMariaBianchi1DanieleGrattarola2CesareAlippi23AbstractMessage-passingMinCutPoolMessage-passingSpectralclustering(SC)isapopularclusterin...
SimpleandDeepGraphConvolutionalNetworksMingChen1ZheweiWei234ZengfengHuang5BolinDing6YaliangLi6AbstractDespitetheirenormoussuccess,mostofthecurrentGCNmodelsareshallow.Mostoftherecentmodels,suchasGra...
SimGANs:Simulator-BasedGenerativeAdversarialNetworksforECGSynthesistoImproveDeepECGClassificationTomerGolany1DanielFreedman2KiraRadinsky1Abstractonvarioustasks(Gabrilovich&Markovitch,2009;Zhangetal...
SGDLearnsOne-LayerNetworksinWGANsQiLei1JasonD.Lee2AlexandrosG.Dimakis1ConstantinosDaskalakis3Abstractbutionwithlowsamplecomplexity.However,theseresultscannotbealgorithmicallyattainedviapracticalGAN...
SDE-Net:EquippingDeepNeuralNetworkswithUncertaintyEstimatesLingkaiKong1JimengSun2ChaoZhang1Abstractbilitiesforclassification(Guoetal.,2017).Moreover,theycanmakeerroneousyetwildlyconfidentprediction...
RobustOne-BitRecoveryviaReLUGenerativeNetworks:Near-OptimalStatisticalRateandGlobalLandscapeAnalysisShuangQiu1XiaohanWei2ZhuoranYang3Abstract1.IntroductionWestudytherobustone-bitcompressedsens-Quan...
Reverse-EngineeringDeepReLUNetworksDavidRolnick1KonradP.Ko¨rding1Abstractconstructadversarialattacks.TheoutputofaneuralnetworkdependsonitsarTherelatedquestionofreverse-engineeringbiologicalneu...
OptimizationTheoryforReLUNeuralNetworksTrainedwithNormalizationLayersYonatanDukler1QuanquanGu2GuidoMontúfar134AbstractofadditionalnormalizationmethodsfollowedBN,notablyincludingLayerNormalization(...
OntheExpressivityofNeuralNetworksforDeepReinforcementLearningKefanDong1YupingLuo2TianheYu3ChelseaFinn3TengyuMa3Abstractwiththeestimateddynamics(Nagabandietal.,2018;Chuaetal.,2018;Wang&Ba,2019).Weco...
NeuralNetworksareConvexRegularizers:ExactPolynomial-timeConvexOptimizationFormulationsforTwo-layerNetworksMertPilanci1TolgaErgen1Abstractconnectionstokernelmethods,andshowedthatrandomlyinitializedg...
Maximum-and-ConcatenationNetworksXingyuXie1HaoKong1JianlongWu2WayneZhang3GuangcanLiu4ZhouchenLin1Abstractnumberoftrainingsamples;namely,Whilesuccessfulinmanyfields,deepneuralnet-minL(θ):=1n(fθ(xi...
Let’sAgreetoAgree:NeuralNetworksShareClassificationOrderonRealDatasetsGuyHacohen12LeshemChoshen1DaphnaWeinshall1AbstractFigure1.Imageslearnedatthebeginning(toprow)andtheendofthetraining(bottomrow)...
LearningQuadraticGamesonNetworksYanLeng1XiaowenDong2JunfengWu3AlexPantland4AbstractFacebook,onlytofindoutthatanypairofFacebookuserscanactuallybeconnectedviaapproximatelythreeandahalfIndividuals,oro...
LearningOpinionsinSocialNetworksVincentConitzer1DebmalyaPanigrahi1HanruiZhang1Abstractstronglyaffectsuserv,andweknowthatuseruisawareofourproduct,thenwearequitesurethatuservisalsoWestudytheproblemof...
LearningAlgebraicMultigridUsingGraphNeuralNetworksIlayLuz1MeiravGalun1HaggaiMaron2RonenBasri1IradYavneh3Abstract(2013);Napov&Notay(2016);Fox&Manteuffel(2018)).WhileAMGismathematicallywellgrounded,i...
InfoGAN-CRandModelCentrality:Self-supervisedModelTrainingandSelectionforDisentanglingGANsZinanLin1KiranK.Thekumparampil2GiuliaFanti1SewoongOh3Abstract1.IntroductionDisentangledgenerativemodelsmapal...
IncidenceNetworksforGeometricDeepLearningMarjanAlbooyeh1DanieleBertoliniSiamakRavanbakhsh23Abstracttwo(edges).Moreovereachedge(column)isincidenttoexactlytwonodes(rows).Thesparsitypatternoftheinci-S...
HowrecurrentNetworksimplementcontextualprocessinginsentimentanalysisNiruMaheswaranathan1DavidSussillo1AbstractRigorouslyunderstandinghowNetworkssolveimportanttasksisacentralchallengeindeeplearning....
HarmonicDecompositionsofConvolutionalNetworksMeyerScetbon12ZaidHarchaoui2Abstractlyzeamulti-layerperceptron(MLP)insteadactingonvec-torizedimages,whichhasthedownsidethesetasidetheWepresentadescripti...