Presentation | 2009-01-19 Structure estimation using time-dependent data in hidden Markov models Masashi MATSUMOTO, Sumio WATANABE, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | A lot of learning machines used in information science, for example, mixture models, artificial neural networks, Bayesian networks, hidden Markov models and Boltzmann machines are non-identifiable and their Fisher information matrices are not positive definite. Therefore, they are not regular but singular. Recently, the learning theory for singular models was constructed under the condition that data are generated from the true distribution identically independent. However, if the training data are time-dependent, learning theory is not yet established. In this paper, we define an Ergodic generalization error for time-dependent learning and study its behavior by numerical experiments in hidden Markov models. As the result, It is clarified that the generalization error is in inverse proportion to the number of training data and the learning coefficient was strongly depends on time-dependency of the true distribution. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | hidden Markov models / Bayes learning / learning curve |
Paper # | NC2008-86 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2009/1/12(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) | Structure estimation using time-dependent data in hidden Markov models |
Sub Title (in English) | |
Keyword(1) | hidden Markov models |
Keyword(2) | Bayes learning |
Keyword(3) | learning curve |
1st Author's Name | Masashi MATSUMOTO |
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 | 2009-01-19 |
Paper # | NC2008-86 |
Volume (vol) | vol.108 |
Number (no) | 383 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |