资源描述：
《19 Regularization and statistical learning theory for data analysis》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、ComputationalStatistics&DataAnalysis38(2002)421–432www.elsevier.com/locate/csdaRegularizationandstatisticallearningtheoryfordataanalysisTheodorosEvgenioua;∗,TomasoPoggiob,MassimilianoPontilb,AlessandroVerricaTechnologyManagement,INSEAD,Fointainebleau77305,Fran
2、cebCenterforBiologicalandComputationalLearning,MIT,45CarletonSt.,Cambridge,MA,02142,USAcINFM-DISI,UniversitadiGenova,ViaDodecaneso35,16146Genova,Italy,AbstractProblemsofdataanalysis,likeclassi/cationandregression,canbestudiedintheframeworkofRegularizationTheor
3、yasill-posedproblems,orthroughStatisticalLearningTheoryinthelearning-from-exampleparadigm.Inthispaperwehighlighttheconnectionsbetweenthesetwoapproachesanddiscusstechniques,likesupportvectormachinesandregularizationnetworks,whichcanbejusti/edinthistheoreticalfr
4、ameworkandprovedtobeusefulinanumberofimageanalysisapplications.c2002ElsevierScienceB.V.Allrightsreserved.Keywords:Statisticallearningtheory;Regularizationtheory;Supportvectormachine;Regularizationnetworks;Imageanalysisapplications1.IntroductionThegoalofthispa
5、peristoprovideabriefintroductiontothestudyofsupervisedlearningwithintheframeworkofRegularizationTheoryandStatisticalLearningTheory.ForadetailedreviewofthetheoreticalaspectsofthissubjectseeEvgeniouetal.(1999).Insupervisedlearningorlearning-from-examplesamachine
6、istrained,insteadofprogrammed,toperformagiventaskonanumberofinput–outputpairs.Accordingtothisparadigm,trainingmeanschoosingafunctionwhichbestdescribestherelationbetweentheinputsandtheoutputs.Infunctionalanalysis,thechoiceof∗Correspondingauthor.0167-9473/02/$-s
7、eefrontmatterc2002ElsevierScienceB.V.Allrightsreserved.PII:S0167-9473(01)00069-X422T.Evgeniouetal./ComputationalStatistics&DataAnalysis38(2002)421–432theoptimalfunctionisanexampleofanill-posedproblemwhichcanbeaddressedwiththemachineryofRegularizationTheory.In
8、aprobabilisticsetting,asecondfundamentalproblem,studiedbyStatisticalLearningTheory,ishowwellthechosenfunctiongeneralizes,orhowwellitestimatestheoutputfornewinputs.Thispaperisorgani
