Summary

International Symposium on Nonlinear Theory and its Applications

2005

Session Number:3-1-5

Session:

Number:3-1-5-2

A New Method of Model Selection Based on Learning Coefficient

Keisuke Yamazaki,  Kenji Nagata,  Sumio Watanabe,  

pp.389-392

Publication Date:2005/10/18

Online ISSN:2188-5079

DOI:10.34385/proc.40.3-1-5-2

PDF download (108.8KB)

Summary:
In the information engineering field, many practical learning machines, e.g. neural networks, mixture models and hidden Markov models, have been developed. In spite of their wide-range applications, there is no theoretical method to select the optimal sized model in such models i.e. singular models. Recent years, an approach to analyze the singular models was established based on algebraic geometry. In this paper, we propose a new model selection criterion, Singular Information Criterion (SingIC), based on the algebraic geometrical method.