Presentation 2002/7/19
On the Effects of applying Tsallis mutual entropy to Independent Component Analysis
Hiroki SUYARI, Masumi DAKEMOTO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) When we apply independent component analysis (ICA for short) to extraction of some information sources from given observations, the results often depend on initial settings of the algorithm. In somes cases, we cannot find the required resuluts by any setting in ICA. In this paper, we apply Tsallis mutual entropy, one-parameter generalization of Shannon mutual entropy, to ICA and reports the effects of the new parameter (q ∈ [0,1]) in iCA. In particular, we take the examples which the usual ICA using Shannon mutual entropy cannot succeed to extract. It is shown that the application of Tsallis mutual entropy to ICA for such examples enables us to succeed in the extraction. The effect of the new parameter q in the learning is also discussed.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) independent component analysis / Tsallis entropy / Tsallis mutual entropy
Paper # NC2002-34
Date of Issue

Conference Information
Committee NC
Conference Date 2002/7/19(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) On the Effects of applying Tsallis mutual entropy to Independent Component Analysis
Sub Title (in English)
Keyword(1) independent component analysis
Keyword(2) Tsallis entropy
Keyword(3) Tsallis mutual entropy
1st Author's Name Hiroki SUYARI
1st Author's Affiliation Faculty of Engineering, Chiba University()
2nd Author's Name Masumi DAKEMOTO
2nd Author's Affiliation Graduate School of Science and Technology, Chiba University
Date 2002/7/19
Paper # NC2002-34
Volume (vol) vol.102
Number (no) 253
Page pp.pp.-
#Pages 5
Date of Issue