Presentation 1999/3/19
Statistical mechanics of factor analysis : Dynamical analysis of The Wake-Sleep algorithm
Seiichi Shimasaki, Yoshiyuki Kabashima,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The Wake-Sleep algorithm is a stochastic learning algorithm which enables Helmholtz machine, asymmetrically connected layered network to work in practice. Experimentally, several positive results have been reported so far, but theoritically, little is known about its properties even for a simple model.In this paper, we analyse the dynamics of the Wake-Sleep Algorithm applying to a single factor analys is problem in respect with statitical mechanics context. As a result, we clarify the convergence properties of WS algorithm, even for the case of using finite learning rate.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) statistical mechanics / factor analysis / unsupervised learning / helmholtz machine
Paper # NC98-154
Date of Issue

Conference Information
Committee NC
Conference Date 1999/3/19(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) Statistical mechanics of factor analysis : Dynamical analysis of The Wake-Sleep algorithm
Sub Title (in English)
Keyword(1) statistical mechanics
Keyword(2) factor analysis
Keyword(3) unsupervised learning
Keyword(4) helmholtz machine
1st Author's Name Seiichi Shimasaki
1st Author's Affiliation Dept. of Compt. Intl. and Syst. Sci. Tokyo Institute of Technology, Kabashima lab.()
2nd Author's Name Yoshiyuki Kabashima
2nd Author's Affiliation Dept. of Compt. Intl. and Syst. Sci. Tokyo Institute of Technology, Kabashima lab.
Date 1999/3/19
Paper # NC98-154
Volume (vol) vol.98
Number (no) 674
Page pp.pp.-
#Pages 8
Date of Issue