Presentation 2020-01-09
Asymptotic Behavior of Bayesian Generalization Error in Multinomial Mixtures
Takumi Watanabe, Sumio Watanabe,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Multinomial mixtures are widely used in the information engineering field. However, it is not subject to the conventional statistical theory of regular statistical models, because the models are not identifiable and their Fisher information matrices are is not positive definite. In the recent years, the generalization performance of singular learning machines was clarified based on resolution of singularities. In this report, we compute Real Log Canonical Thresholds in an algebraic geometric way and clarify asymptotic behaviors of generalization error and free energy.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) generalization error / free energy / multinomial mixtures / real log canonical thresholds
Paper # IBISML2019-18
Date of Issue 2020-01-02 (IBISML)

Conference Information
Committee IBISML
Conference Date 2020/1/9(2days)
Place (in Japanese) (See Japanese page)
Place (in English) ISM
Topics (in Japanese) (See Japanese page)
Topics (in English) Machine learning, etc.
Chair Hisashi Kashima(Kyoto Univ.)
Vice Chair Masashi Sugiyama(Univ. of Tokyo) / Koji Tsuda(Univ. of Tokyo)
Secretary Masashi Sugiyama(Nagoya Inst. of Tech.) / Koji Tsuda(AIST)
Assistant Tomoharu Iwata(NTT) / Shigeyuki Oba(Kyoto Univ.)

Paper Information
Registration To Technical Committee on Infomation-Based Induction Sciences and Machine Learning
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Asymptotic Behavior of Bayesian Generalization Error in Multinomial Mixtures
Sub Title (in English)
Keyword(1) generalization error
Keyword(2) free energy
Keyword(3) multinomial mixtures
Keyword(4) real log canonical thresholds
1st Author's Name Takumi Watanabe
1st Author's Affiliation Tokyo Institute of Technology(Tokyo Tech)
2nd Author's Name Sumio Watanabe
2nd Author's Affiliation Tokyo Institute of Technology(Tokyo Tech)
Date 2020-01-09
Paper # IBISML2019-18
Volume (vol) vol.119
Number (no) IBISML-360
Page pp.pp.1-8(IBISML),
#Pages 8
Date of Issue 2020-01-02 (IBISML)