A Modified Neural Network Learning Approach and Its Application to Bengali Character Recognition

Main Article Content

Md. Enamul Karim
Mahmood Hossain
Md. Abdul Mottalib

Abstract

Presents a new learning model for neural networks. It combines two previous modifications to the back propagation algorithm with a new scheme for pruning less contributory training cycles to achieve faster training. The total training time is reduced by predicting future weights at regular intervals from the nature of the previous weight changes. The oscillations among different patterns are reduced by updating the weights as a function of sum of errors of all input patterns. The predefined error level is adjusted by aborting the training session at an early stage when the weight changes become insignificant with respect to changes in iterations. This model was used to recognize Bengali alphabets and a significant reduction in training time was observed.

Downloads

Download data is not yet available.

Article Details

How to Cite
Karim, M. E., Hossain, M., & Mottalib, M. A. (1998). A Modified Neural Network Learning Approach and Its Application to Bengali Character Recognition. Malaysian Journal of Computer Science, 11(2), 68–73. Retrieved from http://jice.um.edu.my/index.php/MJCS/article/view/5737
Section
Articles