OptimizationofGraphNeuralNetworks:ImplicitAccelerationbySkipConnectionsandMoreDepthKeyuluXu1MozhiZhang2StefanieJegelka1KenjiKawaguchi3AbstracttheoreticalaspectsofGNNstounderstandtheirsuccessandlimi...
OnExplainabilityofGraphNeuralNetworksviaSubgraphExplorationsHaoYuan1HaiyangYu1JieWang2KangLi3ShuiwangJi1Abstract2018;Wangetal.,2019),andgraphpooling(Yuan&Ji,2020;Gao&Ji,2019;Zhangetal.,2018).Howeve...
NewtonMethodoverNetworksisFastuptotheStatisticalPrecisionAmirDaneshmand1GesualdoScutari1PavelDvurechensky23AlexanderGasnikov43Abstractwhere:Rd×Z→Risthelossfunction,assumedtobe(strongly)convexinx,...
MultiscaleInvertibleGenerativeNetworksforHigh-DimensionalBayesianInferenceShumaoZhang1PengchuanZhang2ThomasY.Hou1AbstractwhereListhelikelihoodgivenasWeproposeaMultiscaleInvertibleGenerativeL(yx)=N(...
LearningDiverse-StructuredNetworksforAdversarialRobustnessXuefengDu12JingfengZhang3BoHan1TongliangLiu4YuRong5GangNiu3JunzhouHuang5MasashiSugiyama36AbstractTable1.Performancemisalignmentfordifferent...
LearningCurvesforAnalysisofDeepNetworksDerekHoiem1TanmayGupta2ZhizhongLi1MichalM.Shlapentokh-Rothman1Abstracttrapolateratherthanevaluateandhavetypicallyrequiredlarge-scaleexperimentsthatareoutsidet...
LearningDeepNeuralNetworksunderAgnosticCorruptedSupervisionBoyangLiu1MengyingSun1DingWang1Pang-NingTan1JiayuZhou1Abstractduringtraining.Bydetectingandfiltering(orre-weighting)thepossiblecorruptedsa...
InformationObfuscationofGraphNeuralNetworksPeiyuanLiao1HanZhao2KeyuluXu3TommiJaakkola3GeoffreyGordon1StefanieJegelka3RuslanSalakhutdinov1Abstractthesocialnetwork.Bytakingthestructuredinformationoff...
ImprovingGradientRegularizationusingComplex-ValuedNeuralNetworksEricYeats1YiranChen1HaiLi1Abstractandthattheseinputsaredifficulttodistinguishfromnor-malinputs.SubsequentinvestigationbyGoodfelloweta...
ImprovingBreadth-WiseBackpropagationinGraphNeuralNetworksHelpsLearningLong-RangeDependenciesDenisLukovnikov1AsjaFischer1Abstract2020),(2)over-smoothing(Lietal.,2018;Chenetal.,2019;Zhao&Akoglu,2020;...
HowFrameletsEnhanceGraphNeuralNetworksXuebinZheng1BingxinZhou1JunbinGao1YuGuangWang234PietroLiò5MingLi6GuidoMontúfar72Abstract1.IntroductionThispaperpresentsanewapproachforassem-Graphneuralnetwor...
HowcouldNeuralNetworksunderstandPrograms?DinglanPeng1ShuxinZheng2YataoLi2GuolinKe2DiHe2Tie-YanLiu2Abstract1.IntroductionSemanticunderstandingofprogramsisafunda-Modernsoftwaretypicallycontainstonsof...
Grid-FunctionedNeuralNetworksJavierDehesa12AndrewVidler2JulianPadget1ChristofLutteroth1Abstractpointsscatteredthroughoutthewholeproblemspace.Forexample,predictingageographicalfeaturegivenalatitudeW...
GraphNeuralNetworksInspiredbyClassicalIterativeAlgorithmsYongyiYang1†TangLiu1†YangkunWang2†JinjingZhou3QuanGan3ZheweiWei4ZhengZhang3ZengfengHuang1DavidWipf3Abstract1.IntroductionDespitetherecent...
GNNAutoScale:ScalableandExpressiveGraphNeuralNetworksviaHistoricalEmbeddingsMatthiasFey1JanEricLenssen1FrankWeichert1JureLeskovec2AbstractWhilethefull-gradientinaGNNisstraightforwardtocom-pute,assu...
Globally-RobustNeuralNetworksKlasLeino1ZifanWang1MattFredrikson1AbstractFigure1:Illustrationofglobalrobustness.ThemodelabstainsfrompredictingonthemarginbetweentheclassesThethreatofadversarialexampl...
GenerativeCausalExplanationsforGraphNeuralNetworksWanyuLin12HaoLan2BaochunLi2AbstractDespitetheirpracticalsuccess,mostGNNsaredeployedasblackboxes,lackingexplicitdeclarativeknowledgerepre-Thispaperp...
GenerativeAdversarialNetworksforMarkovianTemporalDynamics:StochasticContinuousDataGenerationSungWooPark1DongWookShu1JunseokKwon1AbstractprobabilitydistributionXt∼pt,whichhasthefollowingformulation...