Artificial neural network weights optimization design based on MEC algorithm

Author

He, Xiao-Juan ; Zeng, Jian-chao ; Jie, Jing

Author_Institution

Dept. of Math., Taiyuan Heavy Machinery Inst., China

Volume

6

fYear

2004

fDate

26-29 Aug. 2004

Firstpage

3361

Abstract

Mind evolutionary computation (MEC) is a new approach of evolutionary computation. In this paper, it is adopted to train the weights of artificial neural network (ANN) to solve premature convergence problem of BP algorithm and genetic algorithm. The coding method of taking individual weights as the center of normal distribution is proposed, and information of network weights is used. Dynamic searching method is used, and weights are trained successfully. The simulation result shows that the new method is better than the common BP algorithm and genetic algorithm.

Keywords

convergence; evolutionary computation; learning (artificial intelligence); neural nets; BP algorithm; MEC algorithm; artificial neural network weights optimization design; dynamic searching method; genetic algorithm; mind evolutionary computation; premature convergence problem; Algorithm design and analysis; Artificial neural networks; Computational modeling; Computer applications; Computer simulation; Convergence; Design optimization; Genetic algorithms; Machinery; Neural networks;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning and Cybernetics, 2004. Proceedings of 2004 International Conference on

Print_ISBN

0-7803-8403-2

Type

conf

DOI

10.1109/ICMLC.2004.1380361

Filename

1380361