A shunting multilayer perceptron network for confusing/composite pattern recognition

Chung-Hsien Wu, Jhing Fa Wang, Wen Horng Wu

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

In this paper, a shunting multilayer perceptron (SMLP) network trained with a modified backpropagation algorithm is proposed. In the SMLP, an MLP network shunts another MLP network in order to learn the relationship between any two training patterns. The SMLP network is endowed with two abilities in relation to patterns in complicated circumstances. Firstly, the SMLP network can perform confusing pattern recognition by weighting the node responses which are important for distinguishing two confusing patterns. Secondly, the SMLP network can perform composite pattern recognition by paying selective attention to the node responses which are distinctive for each individual pattern. The determination of the hidden node number for constructing a suitably sized SMLP network is also discussed here. For the experimental evaluations, ten spoken and handprinted digits and nine spoken and handprinted English alphabets were used. Comparisons to MLP, dynamic time warping algorithm, and nearest neighbor classifier showed satisfactory improvements in recognition accuracy for confusing/composite patterns.

Original languageEnglish
Pages (from-to)1093-1103
Number of pages11
JournalPattern Recognition
Volume24
Issue number11
DOIs
Publication statusPublished - 1991 Jan 1

Fingerprint

Multilayer neural networks
Pattern recognition
Composite materials
Backpropagation algorithms
Classifiers

All Science Journal Classification (ASJC) codes

  • Software
  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Cite this

Wu, Chung-Hsien ; Wang, Jhing Fa ; Wu, Wen Horng. / A shunting multilayer perceptron network for confusing/composite pattern recognition. In: Pattern Recognition. 1991 ; Vol. 24, No. 11. pp. 1093-1103.
@article{af7c240b631c4d6fa8b0b4e5484e64a1,
title = "A shunting multilayer perceptron network for confusing/composite pattern recognition",
abstract = "In this paper, a shunting multilayer perceptron (SMLP) network trained with a modified backpropagation algorithm is proposed. In the SMLP, an MLP network shunts another MLP network in order to learn the relationship between any two training patterns. The SMLP network is endowed with two abilities in relation to patterns in complicated circumstances. Firstly, the SMLP network can perform confusing pattern recognition by weighting the node responses which are important for distinguishing two confusing patterns. Secondly, the SMLP network can perform composite pattern recognition by paying selective attention to the node responses which are distinctive for each individual pattern. The determination of the hidden node number for constructing a suitably sized SMLP network is also discussed here. For the experimental evaluations, ten spoken and handprinted digits and nine spoken and handprinted English alphabets were used. Comparisons to MLP, dynamic time warping algorithm, and nearest neighbor classifier showed satisfactory improvements in recognition accuracy for confusing/composite patterns.",
author = "Chung-Hsien Wu and Wang, {Jhing Fa} and Wu, {Wen Horng}",
year = "1991",
month = "1",
day = "1",
doi = "10.1016/0031-3203(91)90124-N",
language = "English",
volume = "24",
pages = "1093--1103",
journal = "Pattern Recognition",
issn = "0031-3203",
publisher = "Elsevier Limited",
number = "11",

}

A shunting multilayer perceptron network for confusing/composite pattern recognition. / Wu, Chung-Hsien; Wang, Jhing Fa; Wu, Wen Horng.

In: Pattern Recognition, Vol. 24, No. 11, 01.01.1991, p. 1093-1103.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A shunting multilayer perceptron network for confusing/composite pattern recognition

AU - Wu, Chung-Hsien

AU - Wang, Jhing Fa

AU - Wu, Wen Horng

PY - 1991/1/1

Y1 - 1991/1/1

N2 - In this paper, a shunting multilayer perceptron (SMLP) network trained with a modified backpropagation algorithm is proposed. In the SMLP, an MLP network shunts another MLP network in order to learn the relationship between any two training patterns. The SMLP network is endowed with two abilities in relation to patterns in complicated circumstances. Firstly, the SMLP network can perform confusing pattern recognition by weighting the node responses which are important for distinguishing two confusing patterns. Secondly, the SMLP network can perform composite pattern recognition by paying selective attention to the node responses which are distinctive for each individual pattern. The determination of the hidden node number for constructing a suitably sized SMLP network is also discussed here. For the experimental evaluations, ten spoken and handprinted digits and nine spoken and handprinted English alphabets were used. Comparisons to MLP, dynamic time warping algorithm, and nearest neighbor classifier showed satisfactory improvements in recognition accuracy for confusing/composite patterns.

AB - In this paper, a shunting multilayer perceptron (SMLP) network trained with a modified backpropagation algorithm is proposed. In the SMLP, an MLP network shunts another MLP network in order to learn the relationship between any two training patterns. The SMLP network is endowed with two abilities in relation to patterns in complicated circumstances. Firstly, the SMLP network can perform confusing pattern recognition by weighting the node responses which are important for distinguishing two confusing patterns. Secondly, the SMLP network can perform composite pattern recognition by paying selective attention to the node responses which are distinctive for each individual pattern. The determination of the hidden node number for constructing a suitably sized SMLP network is also discussed here. For the experimental evaluations, ten spoken and handprinted digits and nine spoken and handprinted English alphabets were used. Comparisons to MLP, dynamic time warping algorithm, and nearest neighbor classifier showed satisfactory improvements in recognition accuracy for confusing/composite patterns.

UR - http://www.scopus.com/inward/record.url?scp=0026371512&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0026371512&partnerID=8YFLogxK

U2 - 10.1016/0031-3203(91)90124-N

DO - 10.1016/0031-3203(91)90124-N

M3 - Article

VL - 24

SP - 1093

EP - 1103

JO - Pattern Recognition

JF - Pattern Recognition

SN - 0031-3203

IS - 11

ER -