StochasticOptimizationforDCFunctionsandNon-smoothNon-convexRegularizerswithNon-asymptoticConvergenceYiXu1QiQi1QihangLin2RongJin3TianbaoYang1Abstractwhereg(·)andh(·)arereal-valuedlower-semicontinu...
ShapeConstraintsforSetFunctionsAndrewCotter1MayaR.Gupta1HeinrichJiang1ErezLouidor1JamesMuller1TamanNarayan1SerenaWang1TaoZhu1AbstractForexample,supposeonewantstopredicthowgoodarestaurantisbasedonit...
SeparatingvalueFunctionsacrosstime-scalesJoshuaRomoff12PeterHenderson3AhmedTouati42EmmaBrunskill3JoellePineau12YannOllivier2Abstractvergenceproperties,makinglearningmoreefficientandstable(Bertsekas...
PACLearnabilityofNodeFunctionsinNetworkedDynamicalSystemsAbhijinAdiga1ChrisJ.Kuhlman1MadhavV.Marathe12S.S.Ravi13AnilK.Vullikanti12Abstract(Beietal.,2016;Kleinbergetal.,2017;Adigaetal.,2018).Weconsi...
OntheLimitationsofRepresentingFunctionsonSetsEdwardWagstaff1FabianB.Fuchs1MartinEngelcke1IngmarPosner1MichaelOsborne1AbstractInputOutputRecentworkontherepresentationofFunctionsϕ+ρf(x1,…,xM)onset...
EfficientlearningofsmoothprobabilityFunctionsfromBernoullitestswithguaranteesPaulRolland1AliKavis1AlexImmer1AdishSingla2VolkanCevher1Abstractgrowscubiclywiththenumberoftests,andcanthusbeinapplicabl...
ComposingValueFunctionsinReinforcementLearningBenjaminvanNiekerk1StevenJames1AdamEarle1BenjaminRosman12Abstractpreviousabilities.Animportantpropertyforlifelong-learningInreinforcementlearning(RL),o...
CharacterizationofConvexObjectiveFunctionsandOptimalExpectedConvergenceRatesforSGDMartenvanDijk1LamM.Nguyen2PhuongHaNguyen1DzungT.Phan2AbstractAlgorithm1StochasticGradientDescent(SGD)MethodWestudyS...
BayesianOptimizationofCompositeFunctionsRaulAstudillo1PeterI.Frazier12AbstractSokolov,2018);inmaterialsanddrugdesign(Kapetanovic,2008;Frazier&Wang,2016)whenseekingtodesignaWeconsideroptimizationofc...
AdaptiveRegretofConvexandSmoothFunctionsLijunZhang1Tie-YanLiu2Zhi-HuaZhou1Abstractreal-worldapplication,wearealsofacinganotherdynamicchallenge—theoptimalsolutionmaychangecontinuously.Weinvestigate...
SmoothedActionValueFunctionsforLearningGaussianPoliciesOfirNachum1MohammadNorouzi1GeorgeTucker1DaleSchuurmans12Abstracthard-maxnotionofQ-value,definedastheexpectedreturnoffollowinganoptimalpolicy.S...
LyapunovFunctionsforFirst-OrderMethods:TightAutomatedConvergenceGuaranteesAdrienTaylor1BryanVanScoy2LaurentLessard23Abstractmethod,withanenergyofzerocorrespondingtoreachingtheoptimalsolutionof(P).T...
LearningtoOptimizeCombinatorialFunctionsNirRosenfeld1EricBalkanski1AmirGloberson2YaronSinger1Abstractrithmsforlearninginpractice.EncouragingresultsshowthatinmanycasessubmodularFunctionscanbeefficie...
FastMaximizationofNon-Submodular,MonotonicFunctionsontheIntegerLatticeAlanKuhnle1J.DavidSmith1VictoriaG.Crawford1MyT.Thai1AbstractKempe,2011;Horel&Singer,2016).Inthiswork,weTheoptimizationofsubmodu...
UnderstandingBlack-boxPredictionsviaInfluenceFunctionsPangWeiKoh1PercyLiang1Abstractpoint(Ribeiroetal.,2016)orbyperturbingthetestpointtoseehowthepredictionchanges(Simonyanetal.,2013;LiHowcanweexpla...
RobustBudgetAllocationviaContinuousSubmodularFunctionsMatthewStaib1StefanieJegelka1Abstractpersontisweightedwithaprobabilitypstthat,e.g.,anadvertisementonradiostationswillinfluencepersonttoTheoptim...
NeuralNetworksandRationalFunctionsMatusTelgarsky1Abstract4spikeNeuralnetworksandrationalFunctionseffi-ratcientlyapproximateeachother.Inmorede-tail,itisshownherethatforanyReLUnet-3polywork,thereexis...
GuaranteesforGreedyMaximizationofNon-submodularFunctionswithApplicationsAndrewAnBian1JoachimM.Buhmann1AndreasKrause1SebastianTschiatschek1AbstractwhereV={v1,...,vn}isthegroundset.Specifically,inexp...
GlobaloptimizationofLipschitzFunctionsCe´dricMalherbe1NicolasVayatis1Abstractglobaloptimization(Pinte´r,1991),black-boxoptimization(Jonesetal.,1998)orderivative-freeoptimization(Rios&Thegoalofthe...
Depth-WidthTradeoffsinApproximatingNaturalFunctionswithNeuralNetworksItaySafran1OhadShamir1Abstractcentempiricalevidencesuggeststhatstandardfeedforwarddeepnetworksarehardertooptimizethanshallowerne...