Presentation 2008-03-13
A Statistical Analysis of Support Vector Machines of Forgetting Factor
Yoshihiko NOMURA, Hiroyuki FUNAYA, Kazushi IKEDA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Support Vector Machines (SVMs) are, in general, trained in batch but any trick is necessary when the target to be trained is time-varying. In this study, we introduced the idea of forgetting factor (FF) of the RLS algorithm for adaptive filters and analyzed its effect on the SVM performance. More concretely, we derived the average generalization error of the algorithm in a simple case where input space is one-dimensional. The average generalization error of the SVMs with FF does not converge to zero, differently from the SVM in batch or online. We confirmed our results by computer simulations.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) support vector machine / forgetting factor / average generalization error
Paper # NC2007-168
Date of Issue

Conference Information
Committee NC
Conference Date 2008/3/5(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Statistical Analysis of Support Vector Machines of Forgetting Factor
Sub Title (in English)
Keyword(1) support vector machine
Keyword(2) forgetting factor
Keyword(3) average generalization error
1st Author's Name Yoshihiko NOMURA
1st Author's Affiliation Department of Systems Science, Graduate School of Informatics, Kyoto-University()
2nd Author's Name Hiroyuki FUNAYA
2nd Author's Affiliation Department of Systems Science, Graduate School of Informatics, Kyoto-University
3rd Author's Name Kazushi IKEDA
3rd Author's Affiliation Department of Systems Science, Graduate School of Informatics, Kyoto-University
Date 2008-03-13
Paper # NC2007-168
Volume (vol) vol.107
Number (no) 542
Page pp.pp.-
#Pages 6
Date of Issue