In one-dimensional density estimation on i.i.d. observations we suggest an adaptive cross-validation technique for the selection of a kernel estimator. This estimator is both asymptotic MISE-efficient with respect to the monotone oracle, and sharp minimax-
EXACTORACLEINEQUALITYFORASHARPADAPTIVEKERNELDENSITYESTIMATOR
ClementineDalelane
LaboratoiredeProbabilit´esetMod`elesAl´eatoires
Universit´ePierreetMarieCurieParisVI
dalelane@ccr.jussieu.fr
arXiv:math/0504382v1 [math.ST] 19 Apr 2005
February1,2008
Abstract
Inone-dimensionaldensityestimationoni.i.d.observationswesuggestanadaptivecross-validationtechniquefortheselectionofakernelestimator.Thisestimatorisbothasymp-toticMISE-e cientwithrespecttothemonotoneoracle,andsharpminimax-adaptiveoverthewholescaleofSobolevspaceswithsmoothnessindexgreaterthan1/2.Theproofofthecentralconcentrationinequalityavoids“chaining”andreliesonanadditivedecompositionoftheempiricalprocessesinvolved.
Keywords:kerneldensityestimator,MISE-optimalkernel,monotoneoracle,minimax-adaptivityMathematicalSubjectClassi cation:62G07,62G20
1Introduction
Formanyyears,adaptiveestimationprocedureshavestimulatedthestatisticalinterest.Suchestimatesachieveminimaxconvergenceratesrelyingonverylittlepriorknowledgeaboutthepropertiesofthecurvestobeestimated.Oracleinequalities tintothisframework,butgivemuchmorepreciseinformationabouttheperformanceofanestimate.Theycomparetheriskofanadaptivecandidatenottotheminimax,buttothebestpossiblerisk.Oracleinequalitieshavebeenproposedforavarietyofproblemsandestimator,Kneip(1994)andDonoho,Johnstone(1994)presumablybeingthe rstpaperstostatesome.MorerecentexamplesareHall,Kerkyacharian,Picard(1999),Cavalier,Tsybakov(2001),Cay(2003),Efromovich(2004).
Inthefollowingwewillconsideroracleinequalitiesindensityestimation,anditwaswaveletestimatorsthathavereceivedthemainattentioninthiscontext.Duringthe90’s,authorslikeDonoho,Johnstone,Hall,KerkyacharianandPicarddevelopedvariousestimationtechniquesthatsatis edmoreandmorere nedoracleinequalities.EfromovichalsoexaminedFourierseriesestimates.Ofcourse,eventhecaseofdatacontrolledbandwidthselectioninvestigatedinthe80’s,canberegardedaskindofanoracleproblem.ButtheonlysourceforamoregeneraloracleinequalityforakerneldensityestimatorisRigollet(2004).Remarkably,unliketheotheroracleinequalitiesondensityestimators,Rigollet’sisanexactone.Ourcontributiongivesanotherexactoracleinequalityforkerneldensityestimation,butthetwodonotcoveroneanotherinneitherdirection.
Rigollet’sapplicationofStein’sblockwiseestimatortonon-parametricdensityestimationisasharpminimax-adaptivekernelselectionrule.Theprocedureapproximatestheso-called
1
Exact oracle inequality ... 暂无评价 20页 免费 Adaptive, multi-paramete.... A Variant of EFICA Algorithm with Adaptive Parametric Density Estimator ...
Exact oracle inequality ... 暂无评价 20页 免费...WAVELET DENSITY ESTIMATOR FOR STOCHASTIC PROCESSES ?...We provide an adaptive version of the algorithm,...
高线 absolute inequality 绝不等式 ambiguous case ...取值至 consistent estimator 相容估计量 correlation ...exact 真确 energy 能; 能量 exact differential ...
Exact oracle inequality ... 暂无评价 20页 免费...WAVELET DENSITY ESTIMATOR FOR STOCHASTIC PROCESSES ?...We provide an adaptive version of the algorithm,...
热门文档