Presentation 2004-06-21
Accelerating Boltzmann machine
Daisuke Itoh, Keita Torii, Tomo Munehisa,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Learning in a Boltzmann machine requires a very long calculation time because of statistical averaging. In order to reduce the time, the following improvements are performed on two steps where statistical averaging is taken in previous methods. At the first step, values of input units, output units and added hidden units are fixed so that they are separated linearly. By this separation, the necessity of taking statistics in the fixed state of input units and output units is lost. At the next step, statistical averaging in the state of fixed input units, freed output units and hidden units is approximated by the mean field theory. By this analytical technique, the calculation time can be shortened sharply. We show these two techniques in the n-bit-parity problem.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Boltzmann machine / linear separation / hidden unit / mean field theory
Paper # AI2004-11
Date of Issue

Conference Information
Committee AI
Conference Date 2004/6/14(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 Artificial Intelligence and Knowledge-Based Processing (AI)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Accelerating Boltzmann machine
Sub Title (in English)
Keyword(1) Boltzmann machine
Keyword(2) linear separation
Keyword(3) hidden unit
Keyword(4) mean field theory
1st Author's Name Daisuke Itoh
1st Author's Affiliation Faculty of Engineering, University of Yamanashi()
2nd Author's Name Keita Torii
2nd Author's Affiliation Faculty of Engineering, University of Yamanashi
3rd Author's Name Tomo Munehisa
3rd Author's Affiliation Faculty of Engineering, University of Yamanashi
Date 2004-06-21
Paper # AI2004-11
Volume (vol) vol.104
Number (no) 133
Page pp.pp.-
#Pages 6
Date of Issue