文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Exact oracle inequality for a sharp adaptive kernel density estimator
侵权投诉

Exact oracle inequality for a sharp adaptive kernel density estimator

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-

monotoneoraclebytheuseofkernelfunctionswithpiecewiseconstantFouriertransform.Themonotoneoracleisapseudo-estimator,whichminimizesthequadraticrisk(MISE)overtheclassofallkernelfunctions,whoseFouriertransformisreal,symmetricanddecreasesmonotonouslyonR+.

Whenconsideringtheconceptofcurvesmoothingfromtheviewpointofsignalrecogni-tion,amonotoneFouriertransformappearstobeanaturalassumptiontoakernel.Giventhattheunknowndensityissquare-integrable,itisequivalentwithrespecttoMISEeithertoestimatethedensityitselfortoreconstructitfromanestimateofitsFouriertransform.Ontheotherhanditisknownthatwithincreasingfrequency,randomin uencesoverbal-ancethetruevalueintheempiricalFouriertransform.Forthisreason,theFourierseriesprojectionestimatoromitsempiricalFouriercoe cientsbeyondacriticalfrequency.InthefamousPinsker- lter,therigidcut-o isweakenedtoamonotoneshrinkageoftheunreliablecoe cientsbythePinsker-wheights.Thefocusonkernelswithmonotonouslydecreasing,butotherwisearbitraryFouriertransformisjustafurthergeneralizationofthethisnotion.

Theobjectiveofthepresentworkistoproposeapurelydatadependentestimatorthatapproximatesthemonotoneoracleinanexactoracleinequality.IncomparisontoRigollet(2004),weabandontheassumptionofthekernel’spiecewiseconstantFouriertransform,ourkernelsonlybeingband-limitedto[ n,n]andhavingamonotoneFouriertransform.AsymptoticexactMISE-e ciencyisshowntoholdoverthesetofallbounded,L2-integrabledensities,whicharenotin nitelydi erenciable.Sharpasymptoticminimax-adaptivityonthewholescaleofSobolevspaceswithsmoothnessindexgreaterthan1/2followsautomatically.

Thereareessentiallytwoquantitiestodeterminethestatementofanoracleinequality:thesetofestimatorsdisposabletotheminimizationofrisk;andthesetoftrueparameters,overwhichtheoracleinequalityissupposedtohold.Evidently,thelargerthesesets,thestrongertheoracleinequality.Inanon-parametricsetting,regularityconditionsarethenaturalwaytospecifythespaceofparameters.Classesofestimatorsconsideredhavebeenquitediverseandcovermanyfamiliarnon-parametricestimationmethods.However,alltheclasses,forwhichoracleinequalitieswereprovensofar,shareanimportantproperty–whether xedorgrowingwiththenumberofobservations,theirdimensionis nite.

Thisisnatural,whendealingwithwaveletorFouriercoe cients.InorderedlinearsmoothersandblockwiseStein’smethod,theassumptions“ordered”and“blockwise”,respec-tively,assurethe nitedimensionality.Forpenalizedleastsquarsestimators,thedimensionisalwaysexplicitlydetermined.

Notethatoracleinequalitiesrelyonspecialconcentrationinequalities,becauseitisnec-essarytoapproximatethemaximumofanempiricalprocessindexedbyaclassoffunctions;functionallimittheoremsareimperative.With nitedimensionwehaveaccesstotheuni-formentropyoftheestimatorclassandchainingargumentsprovideuswithasuitableboundfortheprocess.Yettheseapproximationsunavoidablycontainthefactordimensioninonewayoranother.

EstimatorsindexedbykernelswithmonotoneFouriertransformisaclassthathasobvi-ouslynot nitedimension.Butitisknownthatthesetofmonotonefunctionsalsoallowsforanapproximationofitsuniformcoveringnumber.Unfortunately,theapproximationsdonotcarryoverfromtheFouriertothespacedomain.Andsothechainingapproachisobstructedtous.

Instead,wepursueanalternativewaytoapproximateourempiricalprocess,namelyanadditivedecomposition.Theprocess,indexedbytheclassofkernels,isdecomposedintoalinearcombinationofcountablymanybasisprocesses.Separateargumentsasregardsthebasisprocesses(exponentialinequalities)andthesizeofthenon-randomcoe cientsare

2

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email