资源描述:
《shrink globally act locally sparse bayesian regularization and prediction bayesian models for sparse regression》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、ShrinkGlobally,ActLocally:SparseBayesianRegularizationandPrediction*UniversityPressScholarshipOnlineOxfordScholarshipOnlineBayesianStatistics9JoséM.Bernardo,M.J.Bayarri,JamesO.Berger,A.P.Dawid,DavidHeckerman,AdrianF.M.Smith,andMikeWestPrintpublicationdate:2011PrintISBN-13:9780199694587Publishedt
2、oOxfordScholarshipOnline:January2012DOI:10.1093/acprof:oso/9780199694587.001.0001ShrinkGlobally,ActLocally:SparseBayesianRegularizationandPrediction*NicholasG.PolsonJamesG.ScottDOI:10.1093/acprof:oso/9780199694587.003.0017AbstractandKeywordsWestudytheclassicproblemofchoosingapriordistributionfor
3、alocationparameterβ=(β1,…,βp)aspgrowslarge.First,westudythestandard“global‐localshrinkage”approach,basedonscalemixturesofnormals.Twotheoremsarepresentedwhichcharacterizecertaindesirablepropertiesofshrinkagepriorsforsparseproblems.Next,wereviewsomerecentresultsshowinghowLévyprocessescanbeusedtoge
4、nerateinfinite‐dimensionalversionsofstandardnormalscale‐mixturepriors,alongwithnewpriorsthathaveyettobeseriouslystudiedintheliterature.Thisapproachprovidesanintuitiveframeworkbothforgeneratingnewregularizationpenaltiesandshrinkagerules,andforperformingasymptoticanalysisonexistingmodels.Keywords:
5、LévyProcesses,Shrinkage,SparsityPage1of45ShrinkGlobally,ActLocally:SparseBayesianRegularizationandPrediction*SummaryWestudytheclassicproblemofchoosingapriordistributionforalocationparameterβ=(β1,…,βp)aspgrowslarge.First,westudythestandard“global‐localshrinkage”approach,basedonscalemixturesofnorm
6、als.Twotheoremsarepresentedwhichcharacterizecertaindesirablepropertiesofshrinkagepriorsforsparseproblems.Next,wereviewsomerecentresultsshowinghowLévyprocessescanbeusedtogenerateinfinite‐dimensionalversionsofstandardnormalscale‐mixturepriors,alongwithnewpriorsthathaveyettobeseriouslystudiedinthel
7、iterature.Thisapproachprovidesanintuitiveframeworkbothforgeneratingnewregularizationpenaltiesandshrinkagerules,andforperformingasymptoticanalysisonexistingmodels.KeywordsandPhrases:LÉVYPROCESSES;SHRINKAGE;SPARSITY1.One‐Group