Presentation 1994/10/13
The Capabilities of A Four-layered Feedforward Neural Network
Shin'ichi Tamura, Masahiko Tateishi,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Neural network theorems state that,in the limit of infinitely many hidden units,a four-layered feedforward neural network is equivalent to a three-layered feedforward neural network.In the applications of neural networks,however,we can not use infinitely many hidden units.The capabilities of a neural network with a finite number of hidden units should be studied.In this paper,we first give another proof that a three-layered feedforward network with N-1 hidden units can realize any N input-target relations exactly.Based on the results of the proof,we then construct a four- layered feedforward network which can realize any N input-target relations with an arbitrarily small error using only N , 2+3 hidden units.This shows that a four-layered feedforward network is superior to a three-layered feedforwad network in terms of the number of parameters for a training data realization.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) feedforward neural network / mapping capabilities
Paper # NC94-32
Date of Issue

Conference Information
Committee NC
Conference Date 1994/10/13(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) The Capabilities of A Four-layered Feedforward Neural Network
Sub Title (in English)
Keyword(1) feedforward neural network
Keyword(2) mapping capabilities
1st Author's Name Shin'ichi Tamura
1st Author's Affiliation Research Laboratories,Nippondenso Co.,Ltd.()
2nd Author's Name Masahiko Tateishi
2nd Author's Affiliation Research Laboratories,Nippondenso Co.,Ltd.
Date 1994/10/13
Paper # NC94-32
Volume (vol) vol.94
Number (no) 272
Page pp.pp.-
#Pages 6
Date of Issue