OptimizationofGraphNeuralNetworks:ImplicitAccelerationbySkipConnectionsandMoreDepthKeyuluXu1MozhiZhang2StefanieJegelka1KenjiKawaguchi3AbstracttheoreticalaspectsofGNNstounderstandtheirsuccessandlimi...
FastMarginMaximizationviaDualAccelerationZiweiJi1NathanSrebro2MatusTelgarsky1Abstractmargin0.050.00Wepresentandanalyzeamomentum-basedgra-−0.05Alg1/eq(1.1)dientmethodfortraininglinearclassifierswit...
AccelerationviaFractalLearningRateSchedulesNamanAgarwal1SurbhiGoel2CyrilZhang2AbstractChebyshevnodesγtstepsizesγt−1fractalscheduleηtInpracticalapplicationsofiterativefirst-orderFigure1:Visualiz...
AModularAnalysisofProvableAccelerationviaPolyak’sMomentum:TrainingaWideReLUNetworkandaDeepLinearNetworkJun-KunWang1Chi-HengLin2JacobAbernethy1Abstract1.IntroductionIncorporatingaso-called“momentu...
VarianceReducedCoordinateDescentwithAcceleration:NewMethodWithaSurprisingApplicationtoFinite-SumProblemsFilipHanzely1DmitryKovalev1PeterRichta´rik1Abstractcontrast,ifψisnotseparable,thecorrespond...
AndersonAccelerationofProximalGradientMethodsVienV.Mai1MikaelJohansson1Abstractrameters;slightlyover-orunder-estimatingthestrongcon-vexityconstantcanhaveasevereeffectontheoverallper-Andersonacceler...
AccelerationforCompressedGradientDescentinDistributedandFederatedOptimizationZhizeLi1DmitryKovalev1XunQian1PeterRichta´rik1Abstract1.IntroductionDuetothehighcommunicationcostindistributedWiththepr...
Average-CaseAccelerationThroughSpectralDensityEstimationFabianPedregosa1DamienScieur2Abstractworst-caseaverage-caseWedevelopaframeworkfortheaverage-caseSuboptimalityanalysisofrandomquadraticproblem...
Curvature-ExploitingAccelerationofElasticNetComputationsVienV.Mai1MikaelJohansson1Abstractimprovetheperformancewhenfeaturesarehighlycorre-lated(Tibshiranietal.,2015;Zou&Hastie,2005).Thispaperintrod...
AnytimeOnline-to-Batch,OptimismandAccelerationAshokCutkosky1Abstractoptimalornear-optimalguarantees.ThishashelpedfuelthewidespreadadoptionofonlinelearningalgorithmsasAstandardwaytoobtainconvergence...
AccelerationofSVRGandKatyushaXbyInexactPreconditioningYanliLiu1FeiFeng1WotaoYin1Abstractregularizerψ(x)isproper,closed,andconvex,butmaybenonsmooth.Anonzeroψ(x)isdesirableinmanyapplica-Empiricalri...
ADynamicalSystemsPerspectiveonNesterovAccelerationMichaelMuehlebach1MichaelI.Jordan1Abstractbeenmanyattemptstounderstandandcharacterizethephe-nomenon.Bubecketal.(2015)suggestamodificationofWepresen...
OnAccelerationwithNoise-CorruptedGradientsMichaelB.Cohen1JelenaDiakonikolas2LorenzoOrecchia2AbstractAccelerationisinterestingbecauseityieldsfasteralgorithmsthanclassicalsteepest-descentalgorithms,o...
“ConvexUntilProvenGuilty”:Dimension-FreeAccelerationofGradientDescentonNon-ConvexFunctionsYairCarmonJohnC.DuchiOliverHinderAaronSidford1AbstractOptimizationbecomesmoredifficultwithoutconvexity,as...