Training of artificial neural networks using differential evolution algorithm

Abstract

In the paper an application of differential evolution algorithm to training of artificial neural networks is presented. The adaptive selection of control parameters has been introduced in the algorithm; due to this property only one parameter is set at the start of proposed algorithm. The artificial neural networks to classification of parity-p problem have been trained using proposed algorithm. Results obtained using proposed algorithm have been compared to the results obtained using other evolutionary method, and gradient training methods such as: error back-propagation, and Levenberg-Marquardt method. It has been shown in this paper that application of differential evolution algorithm to artificial neural networks training can be an alternative to other training methods.