Presentation 1993/11/24
Improving Learning Performance of Layered Neural Networks with Multi Resolutional Structure
Yoichi Motomura,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) On a standard Back Propagation learning method,if the number of hidden units and initialparameters are not suitable for the target function which is approximated by the neural network,we see its learning performance can be bad.This is because the system constructed by the basis functions is uncomplete,when we see the layered neural network as a non-linear functional expansion. In this paper,it is shown that a multi-resolutional structure comes over the uncompleteness of the system through discrete wavelets decompositions,which are constructed by some sigmoidal functions in the layered neural network and also given some coments.We will see that learning performance,in the sence of learning speed and accuracy,of the networks with that structure is better than that of the networks with usually used initial structure of ordinary BP.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Multi Layer Perceptron / Multi Resolutional Structure / Wavelet / Layered Neural network
Paper # NC93-50
Date of Issue

Conference Information
Committee NC
Conference Date 1993/11/24(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) Improving Learning Performance of Layered Neural Networks with Multi Resolutional Structure
Sub Title (in English)
Keyword(1) Multi Layer Perceptron
Keyword(2) Multi Resolutional Structure
Keyword(3) Wavelet
Keyword(4) Layered Neural network
1st Author's Name Yoichi Motomura
1st Author's Affiliation Electro technical Laboratory()
Date 1993/11/24
Paper # NC93-50
Volume (vol) vol.93
Number (no) 341
Page pp.pp.-
#Pages 7
Date of Issue