Presentation 1996/2/2
A Study of Learning Paths and Topological properties of Information Measures
Hiroyuki SHIOYA, Takanori KONDO, Tetsuya KOJIMA, Tsutomu DA-TE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) From the view of information measures,the Boltzman machine learning makes the distribution close to the target distribution in the sense of kullback divergence. Hypothesises from the Boltzman machine learning with two units using the several divergences are always same as the target distribution. α-divergence is one of distance-like criteria between two probability distributions and equal to kullback divergence in the limited behavior. But both divergences have the different characteristics on toplogical properties in a space of all probability ditributions. In this study, we show the relation between information measures used in the Boltzman machine learning and learning paths in the parameter space.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Boltzman machine / learning path / information Measure
Paper # NC95-93
Date of Issue

Conference Information
Committee NC
Conference Date 1996/2/2(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) A Study of Learning Paths and Topological properties of Information Measures
Sub Title (in English)
Keyword(1) Boltzman machine
Keyword(2) learning path
Keyword(3) information Measure
1st Author's Name Hiroyuki SHIOYA
1st Author's Affiliation Faculity of Engineering, Hokkaidou University()
2nd Author's Name Takanori KONDO
2nd Author's Affiliation Faculity of Engineering, Hokkaidou University
3rd Author's Name Tetsuya KOJIMA
3rd Author's Affiliation Faculity of Engineering, Hokkaidou University
4th Author's Name Tsutomu DA-TE
4th Author's Affiliation Faculity of Engineering, Hokkaidou University
Date 1996/2/2
Paper # NC95-93
Volume (vol) vol.95
Number (no) 505
Page pp.pp.-
#Pages 5
Date of Issue