Presentation 2011-07-25
An Incremental Learning Algorithm of Kernel Principal Component Analysis for Chunk Data
Takaomi TOKUMOTO, Seiichi OZAWA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, a new algorithm for Kernel Principal Component Analysis (KPCA) is proposed. We extended Takeuchi et al's Incremental KPCA to learn multiple data given at the same time by solving an eigenvalue problem at once. In our method, one or more linear independent data are selected from data in chunk based on the accumulation ratio. After the selection, the eigenspace is rotated by solving an eigenvalue problem at last. This rotation is done only once. So proposed IKPCA can learn faster than Takeuchi et al's IKPCA in which an eigenvalue problem should be solved for individual data. The experimental results shows that the proposed IKPCA can learn faster than Takeuchi et al's IKPCA without losing classification accuracy seriously.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) incremental learning / kernel method / principal component analysis / feature extraction
Paper # NC2011-29
Date of Issue

Conference Information
Committee NC
Conference Date 2011/7/18(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 JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) An Incremental Learning Algorithm of Kernel Principal Component Analysis for Chunk Data
Sub Title (in English)
Keyword(1) incremental learning
Keyword(2) kernel method
Keyword(3) principal component analysis
Keyword(4) feature extraction
1st Author's Name Takaomi TOKUMOTO
1st Author's Affiliation Faculty of engineering, Kobe University()
2nd Author's Name Seiichi OZAWA
2nd Author's Affiliation Faculty of engineering, Kobe University
Date 2011-07-25
Paper # NC2011-29
Volume (vol) vol.111
Number (no) 157
Page pp.pp.-
#Pages 6
Date of Issue