Presentation | 2012-01-27 On the Relationship between Learning Coefficient for Bayesian Estimation and Average of Acceptance Rate for Metropolis Algorithm Kenji NAGATA, Masato OKADA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Singular learning machines such as neural networks, mixture of gaussians and hidden Markov models are difficult to clarify the generalization performance because the asymptotic normality does not hold. In recent studies, the algebraic geometrical method has been established, and clarify the learning coefficient, which characterize the generalization performance, for Bayesian estimation for some singular learning machines. However, applying the algebraic geometrical method is not generally easy, and there are many singular learning machines not to clarify the learning coefficient. On the other hand, in the previous study, we analytically calculate the average of acceptance rate for the Metropolis algorithm, one of the Markov chain Monte Carlo method. In this study, we show that the learning coefficient for Bayesian estimation has a great deal to do with the average of acceptance rate for the Metropolis algorithm. Moreover, we propose a new method to numerically calculate the learning coefficient for the singular learning machine which is difficult to apply the algebraic geometrical method. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Bayesian estimation / Learning coefficient / Metropolis algorithm / Average of acceptance rate |
Paper # | NC2011-111 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2012/1/19(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 Relationship between Learning Coefficient for Bayesian Estimation and Average of Acceptance Rate for Metropolis Algorithm |
Sub Title (in English) | |
Keyword(1) | Bayesian estimation |
Keyword(2) | Learning coefficient |
Keyword(3) | Metropolis algorithm |
Keyword(4) | Average of acceptance rate |
1st Author's Name | Kenji NAGATA |
1st Author's Affiliation | Graduate School of Frontier Sciences, The University of Tokyo() |
2nd Author's Name | Masato OKADA |
2nd Author's Affiliation | Graduate School of Frontier Sciences, The University of Tokyo:RIKEN Brain Science Institute |
Date | 2012-01-27 |
Paper # | NC2011-111 |
Volume (vol) | vol.111 |
Number (no) | 419 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |