NewtonMethodoverNetworksisFastuptotheStatisticalPrecisionAmirDaneshmand1GesualdoScutari1PavelDvurechensky23AlexanderGasnikov43Abstractwhere:Rd×Z→Risthelossfunction,assumedtobe(strongly)convexinx,...
AdaptiveNewtonSketch:Linear-timeOptimizationwithQuadraticConvergenceandEffectiveHessianDimensionalityJonathanLacotte1YifeiWang1MertPilanci1Abstract1.IntroductionWeproposearandomizedalgorithmwithWec...
StochasticSubspaceCubicNewtonMethodFilipHanzely1NikitaDoikov2PeterRichta´rik1YuriiNesterov2Abstractareinterestedintheregimewherethedimensiondisverylarge,whicharisesinmanycontexts,suchasthetraining...
SemismoothNewtonAlgorithmforEfficientProjectionsonto1,∞-normBallDejunChu1ChangshuiZhang1ShiliangSun2QingTao3AbstracttionproblemThestructuredsparsity-inducing1,∞-norm,asminL(W)+λR(W)(1)ageneraliz...
MulticlassNeuralNetworkMinimizationviaTropicalNewtonPolytopeApproximationGeorgiosSmyrnis1PetrosMaragos1Abstracttherecentadventofneuralnetworkshasprovideditwithan-otherapplicationofgreatinterest,whi...
EstimatingtheErrorofRandomizedNewtonMethods:ABootstrapApproachJessieX.T.Chen1MilesE.Lopes2AbstractresearchhasbeendrivenbythefactthatcomputinganinverseHessiantomachineprecisioncanbeverycostlyRandomi...
StochasticVariance-ReducedCubicRegularizedNewtonMethodsDongruoZhou1PanXu1QuanquanGu1Abstracttheirextensions(Ghadimi&Lan,2013;Reddietal.,2016a;Allen-Zhu&Hazan,2016;Ghadimi&Lan,2016).Never-Weproposea...
RandomizedBlockCubicNewtonMethodNikitaDoikov1PeterRichta´rik234AbstractwhereQ⊆RNisaclosedconvexset,andg,φandψareconvexfunctionswithdifferentsmoothnessandstructuralWestudytheproblemofminimizingt...
AnEfficientSemismoothNewtonBasedAlgorithmforConvexClusteringYanchengYuan1DefengSun2Kim-ChuanToh3Abstractwhichisnotavailableinmanyrealapplications.There-fore,inrealapplications,k-meansistypicallytri...
ASemismoothNewtonMethodforFast,GenericConvexProgrammingAlnurAli1EricWong1J.ZicoKolter2Abstracti.e.,thespaceofm×mpositivesemidefinitematricesSm+,yieldsasemidefiniteprogram;andtakingKasthesecond-Wei...
ApproximateNewtonMethodsandTheirLocalConvergenceHaishanYe1LuoLuo1ZhihuaZhang2Abstractwhereηt>0isthesteplength.IfQtistheidentitymatrixandg(x(t))=∇F(x(t)),theresultingprocedureiscalledManymachinele...