Presentation 2012-11-07
Regularization of Restricted Boltzmann Machine Learning through Entropy Minimization
Taichi KIWAKI, Takaki MAKINO, Kazuyuki AIHARA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose a learning scheme for Restricted Boltzmann Machines (RBMs) that suppresses over-fitting, where the entropy of the hidden variables is used as a regularization term. Since conventional RBMs lack mechanisms to reduce their redundant complexity, they tend to suffer from over-fitting, that is, learning excessively complex expressions regardless of the complexity of the data. In our approach, the complexity of RBMs is dynamically adjusted to be right for expressing the given data. This can be regarded as an implementation of so-called model selection on RBMs. In this paper, we analytically derived a learning algorithm based on the regularization term using techniques of statistical mechanics.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Unsupervised learning / Statistical-mechanical informatics / Restricted Boltzmann Machines / Entropy minimization
Paper # IBISML2012-48
Date of Issue

Conference Information
Committee IBISML
Conference Date 2012/10/31(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 Information-Based Induction Sciences and Machine Learning (IBISML)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Regularization of Restricted Boltzmann Machine Learning through Entropy Minimization
Sub Title (in English)
Keyword(1) Unsupervised learning
Keyword(2) Statistical-mechanical informatics
Keyword(3) Restricted Boltzmann Machines
Keyword(4) Entropy minimization
1st Author's Name Taichi KIWAKI
1st Author's Affiliation Graduated School of Engineering, The University of Tokyo()
2nd Author's Name Takaki MAKINO
2nd Author's Affiliation Institute of Industrial Science, The University of Tokyo
3rd Author's Name Kazuyuki AIHARA
3rd Author's Affiliation Institute of Industrial Science, The University of Tokyo
Date 2012-11-07
Paper # IBISML2012-48
Volume (vol) vol.112
Number (no) 279
Page pp.pp.-
#Pages 4
Date of Issue