Summary
International Symposium on Nonlinear Theory and its Applications
2008
Session Number:B2L-C
Session:
Number:B2L-C2
Weighted Blowups of Kullback Information and Application to Multinomial Distributions
Takeshi Matsuda, Sumio Watanabe,
pp.-
Publication Date:2008/9/7
Online ISSN:2188-5079
DOI:10.34385/proc.42.B2L-C2
PDF download (92.5KB)
Summary:
Singular learning machines such as mixture models, neural networks and Bayesian networks are used in many fields of information engineering. However, they are not subject to the conventional statistical theory of regular statistical models, because their Fisher information matrices are degenerate. Recently, the generalization performance of singular learning machines was clarified based on resolution of singularities. In this paper, we propose a new method to compute learning coefficients using weighted blowups and show its effectiveness by application to the mixture of multinomial distributions.