Presentation 2006-07-14
Asymptotic Behavior of Free Energy of General Boltzmann Machines in Mean Field Approximation
Yu NISHIYAMA, Sumio WATANABE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In the Bayesian learning, which generally requires huge computational costs, the algorithms based on the mean field approximation have shown us the effectiveness in the practical information systems. Recently, the generalization error or free energy in the mean field approximation has been theoretically studied. The theoretical results enable us to know the accuracy of the approximation and contribute to the foundation of a model selection in statistical singular machines. In this paper, we show that the upper bounds of the asymptotic free energies are theoretically obtained by counting the number of non-0 eigenvalues of Fisher information matrices and derive the upper bound in the learning model of general Boltzmann machines.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Singular Learning Machines / Free Energy / Mean Field Approximation / Boltzmann Machines
Paper # NC2006-38
Date of Issue

Conference Information
Committee NC
Conference Date 2006/7/7(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) Asymptotic Behavior of Free Energy of General Boltzmann Machines in Mean Field Approximation
Sub Title (in English)
Keyword(1) Singular Learning Machines
Keyword(2) Free Energy
Keyword(3) Mean Field Approximation
Keyword(4) Boltzmann Machines
1st Author's Name Yu NISHIYAMA
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 2006-07-14
Paper # NC2006-38
Volume (vol) vol.106
Number (no) 163
Page pp.pp.-
#Pages 6
Date of Issue