Siraj, Fadzilah and Partridge, Derek (1999) Improving generalization of neural network using length as discriminant. Analisis, 6 (1&2). pp. 75-87. ISSN 0127-8983

Preview

PDF Download (1MB) | Preview

Abstract

This paper discusses the empirical evaluation of improving generalization performance of neural networks by systematic treatment of training and test failures. As a result of systematic treatment of failures, a discrimination technique using LENGTH was developed. The experiments presented in this paper illustrate the application of discrimination technique using LENGTH to neural networks trained to solve supervised learning tasks such as the Launch Interceptor Condition 1 problem. The discriminant LENGTH is used to discriminate between the predicted "hard-to-learn" and predicted "easy-to-learn" patterns before these patterns are fed into the networks. The experimental results reveal that the utilization of LENGTH as discriminant has improved the average generalization of the networks increased.

Item Type:	Article
Uncontrolled Keywords:	neural networks, empirical evaluations, generalization performance
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Divisions:	College of Arts and Sciences
Depositing User:	Prof Madya Fadzilah Siraj
Date Deposited:	04 Jul 2010 01:51
Last Modified:	04 Jul 2010 01:52
URI:	https://repo.uum.edu.my/id/eprint/90

Actions (login required)

View Item