Presentation 2010-09-06
A Study of Variances of Cross Validation and Generalization Error in Variational Bayes Method
Shinji OYAMA, Sumio WATANABE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Variational Bayes method provides high generalization performance as the Bayes method using a small computational cost as the EM algorithm, therefore, it is widely being used in machine learning. In general, the average of the cross validation is asymptotically equal to that of the generalization error. However, in variational Bayes learning, the relationship between them has not yet been clarified. In this paper, we study variational Bayes method in a Gaussian mixture and experimentally show that the variances of the cross validation and the generalization error depend on the condition of regularity, and that the variance of the cross validation is greater than that of the generalization error, especially when the hyperparameter is smaller than the critical point of the phase transition.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) variational Bayes method / cross validation / Gaussian mixture / hyperparameter
Paper # PRMU2010-74,IBISML2010-46
Date of Issue

Conference Information
Committee PRMU
Conference Date 2010/8/29(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 Pattern Recognition and Media Understanding (PRMU)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Study of Variances of Cross Validation and Generalization Error in Variational Bayes Method
Sub Title (in English)
Keyword(1) variational Bayes method
Keyword(2) cross validation
Keyword(3) Gaussian mixture
Keyword(4) hyperparameter
1st Author's Name Shinji OYAMA
1st Author's Affiliation Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology()
2nd Author's Name Sumio WATANABE
2nd Author's Affiliation Precision and Intelligence Laboratory, Tokyo Institute of Technology
Date 2010-09-06
Paper # PRMU2010-74,IBISML2010-46
Volume (vol) vol.110
Number (no) 187
Page pp.pp.-
#Pages 8
Date of Issue