Learning with kernel.pdf

《Learning with kernel.pdf》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、LearningwithKernelsSch¨olkopfandSmola:LearningwithKernels—Confidentialdraft,pleasedonotcirculate—2001/03/0220:32LearningwithKernelsbyBernhardSch¨olkopfAlexanderJ.SmolaTheMITPressCambridge,MassachusettsLondon,EnglandSch¨olkopfandSmola:LearningwithKernels—Confidentialdraft,pleased

2、onotcirculate—2001/03/0220:32c2000MassachusettsInstituteofTechnologyAllrightsreserved.Nopartofthisbookmaybereproducedinanyformbyanyelectronicormechanicalmeans(includingphotocopying,recording,orinformationstorageandretrieval)withoutpermissioninwritingfromthepublisher.Printedand

3、boundintheUnitedStatesofAmericaLibraryofCongressCataloging-in-PublicationDataLearningwithKernels/byBernhardSch¨olkopf,AlexanderJ.Smola.p.cm.Includesbibliographicalreferencesandindex.ISBN0-xxx-xxxxx-x(alk.paper)1.Machinelearning.2.Algorithms.3.KernelfunctionsI.Sch¨olkopf,Bernha

4、rd.II.Smola,AlexanderJ.xxxx.x.xxx2000xxx.x’x–xxxx00.xxxxxCIPContents1ATutorialIntroduction11.1DataRepresentationandSimilarity...................11.2ASimplePatternRecognitionAlgorithm...............31.3SomeInsightsFromStatisticalLearningTheory............61.4HyperplaneClassifier

5、s..........................101.5SupportVectorClassification......................131.6SupportVectorRegression........................161.7KernelPrincipalComponentAnalysis.................181.8EmpiricalResultsandImplementations................19References22Index26Sch¨olkopfandSm

6、ola:LearningwithKernels—Confidentialdraft,pleasedonotcirculate—2001/03/0220:321ATutorialIntroductionThischapterdescribesthecentralideasofsupportvector(SV)learninginanutshell.Itsgoalistoprovideanoverviewofthebasicconcepts.Oneoftheseconceptsisthatofakernel.Ratherthanimmediatelygo

7、ingintoOverviewmathematicaldetail,weintroducekernelsinformallyassimilaritymeasuresthatarisefromaparticularrepresentationofpatterns(Section1.1),anddescribeasimplekernelalgorithmforpatternrecognition(Section1.2).Followingthat,wereportsomebasicinsightsfromstatisticallearningtheor

8、y,themathematicaltheorythatunderliesthebasicideaofSVlearning(