Transcription of Arbeitsgruppe Bülthoff - Face Rec
{{id}} {{{paragraph}}}
Max-Planck-Institutf r biologische KybernetikSpemannstra e 38 72076 T bingen GermanyArbeitsgruppe B lthoffTechnicalReportNo December NonlinearComponentAnalysisasaKernelEigen valueProblemBernhardSch olkopf AlexanderSmola andKlaus RobertM ullerAbstractWedescribeanewmethodforperf orminganonlinearformofPrincipalComponent Anal ysis Bytheuseofintegraloperatorkernelfunction s wecane cientlycomputeprincipalcomponentsinhigh dimensionalfeaturespaces relatedtoinputspacebysomenonlinearmap forinstancethespaceofallpossible pixelproductsin images Wegivethederivationofthemethod alongwithadiscussionofothertechniqueswhi chcanbemadenonlinearwiththekernelapproac h andpresent rstexperimentalresultsonnonlinearfea tureextractionforpatternrecognition ASandKRMarewithGMDF irst ForschungszentrumInformationstechnik RudowerChaussee Berlin ASandBSweresupportedbygrantsfromtheStudi enstiftungdesdeutschenVolkes BSthankstheGMDF irstforhospitalityduringtwovisits ASandBSthankV Vapnikforintroducingthemtokernelrepresen tationsofdotproductsduringjointworkonSup portVectormachines Thisworkpro tedfromdiscussionswithV Blanz L Bottou C Burges H B ultho K Gegenfurtner P Ha ner N Murata P Simard S Solla V Vapnik andT Vetter WearegratefultoV Blanz C Burges andS Sollaforreadingapreliminaryversionofthem anuscript Th
In tro duction Principal Comp onen t Analysis PCA is a p o w er ful tec hnique for extracting structure from p ossi bly highdimensional data sets It is readily p er
Domain:
Source:
Link to this page:
Please notify us if you found a problem with this document:
{{id}} {{{paragraph}}}