Presentation 1999/7/19
Mapping Ability of Multilayered Neural Network Introduced Linear Dependent Constraints among Weights
Masaki ISHII, Itsuo KUMAZAWA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In order to improve generalization ability, it is effective to restrict degrees of freedom of representation model based on a priori knowledge about target. A method we proposed here is to introduce linear dependent constraints among weights, as an approach of restricting degrees of freedom. The linear dependency is determined based on some invarience structure of input patterns. Before discussing its generalization ability, it is necessary to clarify how its mapping ability decreases by introducing such constraints. Therefore for multilayerd neural network with threshold units, we clarify the bounds on the VC dimension as its mapping ability.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) neural network / a priori knowledge / dependent constraints / VC dimension
Paper # NC99-36
Date of Issue

Conference Information
Committee NC
Conference Date 1999/7/19(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) Mapping Ability of Multilayered Neural Network Introduced Linear Dependent Constraints among Weights
Sub Title (in English)
Keyword(1) neural network
Keyword(2) a priori knowledge
Keyword(3) dependent constraints
Keyword(4) VC dimension
1st Author's Name Masaki ISHII
1st Author's Affiliation Department of Computer Science Graduate School of Information Science and Engineering Tokyo Institute of Technology()
2nd Author's Name Itsuo KUMAZAWA
2nd Author's Affiliation Department of Computer Science Graduate School of Information Science and Engineering Tokyo Institute of Technology
Date 1999/7/19
Paper # NC99-36
Volume (vol) vol.99
Number (no) 193
Page pp.pp.-
#Pages 6
Date of Issue