Training multilayer perceptron using differential evolution algorithm for signature recognition application

In this work, multilayer perceptron (MLP) has been trained by differential evolution algorithm (DEA) and the performance of the neural network has been analyzed by using high-dimensional and non-linear signature recognition data base. DEA, which doesn´t depend on the initial weight values and doesn´t stick in local minimums, carries out the global optimization. The performance of the DGA which is the heuristic algorithm to training of the network has been compared to the performance of the error back-propagation algorithm (EBPA) based on gradient. Simulation results show that the performance of the training MLP using DEA is outperforms the training MLP using EBPA.