New learning method using prior information of neural networks

Baiquan Lu, Kotaro Hirasawa, Junichi Murata, Takayuki Furuzuki

Research output: Contribution to journalArticle

Abstract

In this paper, we present a new learning method using prior information for three-layer neural networks. Usually when neural networks are used for identification of systems, all of their weights are trained independently, without considering their inter-relation of weights values. Thus the training results are not usually good. The reason for this is that each parameter has its influence on others during the learning. To overcome this problem, first, we give exact mathematical equation that describes the relation between weight values given a set of data conveying prior information. Then we present a new learning method that trains the part of the weights and calculates the others by using these exact mathematical equations. This method often keeps a priori given mathematical structure exactly during the learning, in other words, training is done so that the network follows predetermined trajectory. Numerical computer simulation results are provided to support the present approaches.

Original languageEnglish
Pages (from-to)29-35
Number of pages7
JournalResearch Reports on Information Science and Electrical Engineering of Kyushu University
Volume4
Issue number1
Publication statusPublished - 1999 Mar
Externally publishedYes

Fingerprint

Neural networks
Conveying
Trajectories
Computer simulation

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Hardware and Architecture
  • Engineering (miscellaneous)

Cite this

New learning method using prior information of neural networks. / Lu, Baiquan; Hirasawa, Kotaro; Murata, Junichi; Furuzuki, Takayuki.

In: Research Reports on Information Science and Electrical Engineering of Kyushu University, Vol. 4, No. 1, 03.1999, p. 29-35.

Research output: Contribution to journalArticle

@article{eaf9844679854cd4bc51bfa91347e132,
title = "New learning method using prior information of neural networks",
abstract = "In this paper, we present a new learning method using prior information for three-layer neural networks. Usually when neural networks are used for identification of systems, all of their weights are trained independently, without considering their inter-relation of weights values. Thus the training results are not usually good. The reason for this is that each parameter has its influence on others during the learning. To overcome this problem, first, we give exact mathematical equation that describes the relation between weight values given a set of data conveying prior information. Then we present a new learning method that trains the part of the weights and calculates the others by using these exact mathematical equations. This method often keeps a priori given mathematical structure exactly during the learning, in other words, training is done so that the network follows predetermined trajectory. Numerical computer simulation results are provided to support the present approaches.",
author = "Baiquan Lu and Kotaro Hirasawa and Junichi Murata and Takayuki Furuzuki",
year = "1999",
month = "3",
language = "English",
volume = "4",
pages = "29--35",
journal = "Research Reports on Information Science and Electrical Engineering of Kyushu University",
issn = "1342-3819",
publisher = "Kyushu University, Faculty of Science",
number = "1",

}

TY - JOUR

T1 - New learning method using prior information of neural networks

AU - Lu, Baiquan

AU - Hirasawa, Kotaro

AU - Murata, Junichi

AU - Furuzuki, Takayuki

PY - 1999/3

Y1 - 1999/3

N2 - In this paper, we present a new learning method using prior information for three-layer neural networks. Usually when neural networks are used for identification of systems, all of their weights are trained independently, without considering their inter-relation of weights values. Thus the training results are not usually good. The reason for this is that each parameter has its influence on others during the learning. To overcome this problem, first, we give exact mathematical equation that describes the relation between weight values given a set of data conveying prior information. Then we present a new learning method that trains the part of the weights and calculates the others by using these exact mathematical equations. This method often keeps a priori given mathematical structure exactly during the learning, in other words, training is done so that the network follows predetermined trajectory. Numerical computer simulation results are provided to support the present approaches.

AB - In this paper, we present a new learning method using prior information for three-layer neural networks. Usually when neural networks are used for identification of systems, all of their weights are trained independently, without considering their inter-relation of weights values. Thus the training results are not usually good. The reason for this is that each parameter has its influence on others during the learning. To overcome this problem, first, we give exact mathematical equation that describes the relation between weight values given a set of data conveying prior information. Then we present a new learning method that trains the part of the weights and calculates the others by using these exact mathematical equations. This method often keeps a priori given mathematical structure exactly during the learning, in other words, training is done so that the network follows predetermined trajectory. Numerical computer simulation results are provided to support the present approaches.

UR - http://www.scopus.com/inward/record.url?scp=0032653149&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0032653149&partnerID=8YFLogxK

M3 - Article

VL - 4

SP - 29

EP - 35

JO - Research Reports on Information Science and Electrical Engineering of Kyushu University

JF - Research Reports on Information Science and Electrical Engineering of Kyushu University

SN - 1342-3819

IS - 1

ER -