Presentation 2012-11-07
A Calculating Method for Estimation Accuracy of Latent Variables Based on the Free Energy Functions
Keisuke YAMAZAKI, Kazuho WATANABE, Daisuke KAJI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Hierarchical models with latent variables such as a Gaussian mixture and a hidden Markov model are widely used in machine learning and data mining. Estimation accuracy of the variables has not been studied well from the theoretical point of view while the generalization ability has been analyzed for long. Recently, the asymptotic form of an error function on the latent variable density has been derived in the Bayes estimation. However, the value cannot be calculated from the given observable data because the variables are unobservable and the definition of the error consists of the true density assumed to generate the data. In the present paper, we propose a computational method for the error value based on the asymptotic form, which has a connection to the free energy functions of the Bayes and the variational Bayes methods. Numerical experiments show the validity in Bernoulli mixture models.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Latent Variable Estimation / Bayes Method / Variational Bayes Method / Free Energy
Paper # IBISML2012-44
Date of Issue

Conference Information
Committee IBISML
Conference Date 2012/10/31(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 Information-Based Induction Sciences and Machine Learning (IBISML)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Calculating Method for Estimation Accuracy of Latent Variables Based on the Free Energy Functions
Sub Title (in English)
Keyword(1) Latent Variable Estimation
Keyword(2) Bayes Method
Keyword(3) Variational Bayes Method
Keyword(4) Free Energy
1st Author's Name Keisuke YAMAZAKI
1st Author's Affiliation Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology()
2nd Author's Name Kazuho WATANABE
2nd Author's Affiliation Graduate School of Information Science, Nara Institute of Science and Technology
3rd Author's Name Daisuke KAJI
3rd Author's Affiliation IT R & D Department, Konicaminolta Medical & Graphic, INC.
Date 2012-11-07
Paper # IBISML2012-44
Volume (vol) vol.112
Number (no) 279
Page pp.pp.-
#Pages 7
Date of Issue