AsynchronousByzantineMachineLearning(thecaseofSGD)GeorgiosDamaskinos1ElMahdiElMhamdi1RachidGuerraoui1RhicheekPatra1MahsaTaziki1Abstract1.IntroductionAsynchronousdistributedmachinelearningso-Tokeepu...
AnalysisofMinimaxErrorRateforCrowdsourcingandItsApplicationtoWorkerClusteringModelHideakiImamura12IsseiSato12MasashiSugiyama21Abstractsolvethisproblembyprovidingredundancyforlabeling,i.e.,bycollect...
AnalyzingtheRobustnessofNearestNeighborstoAdversarialExamplesYizhenWang1SomeshJha2KamalikaChaudhuri1Abstractattacksviaadversarialexamples.Here,anadversaryhastheabilitytoprovidemodifiedtestinputstoa...
AnAlgorithmicFrameworkofVariableMetricOver-RelaxedHybridProximalExtra-GradientMethodLiShen1PengSun1YitongWang1WeiLiu1TongZhang1Abstracttimizationandconvex-concavesaddle-pointoptimization,encompasse...
AdversarialRiskandtheDangersofEvaluatingAgainstWeakAttacksJonathanUesato1BrendanO’Donoghue1AaronvandenOord1PushmeetKohli1Abstractadversarialexamplescanbecomputedrelativelyeasilybyusingoptimization...
AdversarialDistillationofBayesianNeuralNetworkPosteriorsKuan-ChiehWang12PaulVicol12JamesLucas12LiGu1RogerGrosse12RichardZemel12AbstractUncertaintyisimportantinmanyscenarios.Forexam-ple,designersofa...
ActiveLearningforAccurateEstimationofLinearModelsCarlosRiquelme1MohammadGhavamzadeh2AlessandroLazaric3Abstractassumethattheobservationsarecorruptedbynoiselevelsthatareproblem-dependentandmustbelear...
AUnifiedViewofMulti-LabelPerformanceMeasuresXi-ZhuWu1Zhi-HuaZhou1Abstracteachinstancecanbeassociatedwithmultiplelabelssimul-taneously.Forexample,itisdifficulttotellwhichmistakeMulti-labelclassifica...
ARicherTheoryofConvexConstrainedOptimizationwithReducedProjectionsandImprovedRatesTianbaoYang1QihangLin1LijunZhang2Abstract1.IntroductionThispaperfocusesonconvexconstrainedopti-Inthispaper,weaimats...
ADivergenceBoundforHybridsofMCMCandVariationalInferenceandanApplicationtoLangevinDynamicsandSGVIJustinDomke1Abstractthem.Computingp(z)thusrequiresafullpassoverthedataset.TheideaofStochasticGradient...
“ConvexUntilProvenGuilty”:Dimension-FreeAccelerationofGradientDescentonNon-ConvexFunctionsYairCarmonJohnC.DuchiOliverHinderAaronSidford1AbstractOptimizationbecomesmoredifficultwithoutconvexity,as...
WorldofBits:AnOpen-DomainPlatformforWeb-BasedAgentsTianlin(Tim)Shi12AndrejKarpathy2Linxi(Jim)Fan1JonathanHernandez2PercyLiang1AbstractFigure1.AgentsintheWorldofBitsperceivethescreenpixels,theDOM(wi...
VariantsofRMSPropandAdagradwithLogarithmicRegretBoundsMaheshChandraMukkamala12MatthiasHein1AbstractThegoalofthispaperistwofold.First,weproposeSC-AdagradwhichisavariantofAdagradadaptedtotheAdaptiveg...
UnderstandingtheRepresentationandComputationofMultilayerPerceptrons:ACaseStudyinSpeechRecognitionTashaNagamine1NimaMesgarani1Abstracthiddenlayerareuniversalapproximators(Cybenko,1989;K.Hornik&White...
TowardControlledGenerationofTextZhitingHu12ZichaoYang1XiaodanLiang12RuslanSalakhutdinov1EricP.Xing12Abstractrequiredtocapturecomplexsemanticstructuresunderly-ingsentences.Previousworkhavebeenmostly...
TheSampleComplexityofOnlineOne-ClassCollaborativeFilteringReinhardHeckel1KannanRamchandran1Abstractuserlikes,basedonratingsthatthisuserandalargenum-berofotherusershaveprovidedinthepast.Tothisend,We...
ThePriceofDifferentialPrivacyforOnlineLearningNamanAgarwal1KaranSingh1Abstractthefullinformationandpartialinformation(bandit)set-tings.ThisresultimprovestheknownbestregretboundsWedesigndifferential...
TheLossSurfaceofDeepandWideNeuralNetworksQuynhNguyen1MatthiasHein1Abstractdoesnotencounterproblemswithsuboptimallocalmin-ima.However,astheauthorsadmitthemselvesin(Good-Whiletheoptimizationproblembe...
StochasticDCAfortheLarge-sumofNon-convexFunctionsProblemanditsApplicationtoGroupVariableSelectioninClassificationHoaiAnLeThi1HoaiMinhLe1DuyNhatPhan1BachTran1AbstractNowadays,thegrowthoftechnologies...
SPLICE:FullyTractableHierarchicalExtensionofICAwithPoolingJun-ichiroHirayama12AapoHyva¨rinen34MotoakiKawanabe21Abstractaselectro-ormagneto-encephalography(EEG/MEG)sig-nalanalysisandnaturalimagesta...