Training back propagation neural networks using asexual reproduction optimization

Abstract

Training a back propagation neural network is an optimization problem to find optimal weight set in training process. The back propagation neural network can fall into a local minimum point during learning of training patterns. Therefore, evolutionary algorithms can be used to train this neural network to obtain suitable initial weight set. In this paper, a novel approach is proposed to train the back propagation neural network based on asexual reproduction optimization (ARO) algorithm. The basic idea of the proposed method is to apply ARO algorithm at the first step to search for the global initial connection weights. Then, the back propagation algorithm is used to thoroughly search for the optimal weight set. The performance of the proposed method is evaluated using a number of problems. Experimental results show that the proposed method is better than the genetic and back propagation algorithms in convergent speed and convergent accuracy.