Comparison Study on Using of BP and Genetic NN for Digital Logic Circuit Application

Abstract

Neural networks are facing many problems when they employ a backpropagation algorithm. These problems are characterized by long training time and trapping the network into local minima. For these reasons the trend, in recent years, started toward the application of the genetic algorithm because of its ability to discover wide and complex search spaces. In the present work, a number of comparisons between BP and GA have been carried out. The results regarding training speed and performance, show that GA is more suitable than BP for training neural networks (ANN). with respect to the results obtained, a novel approach for designing a multiplayer artificial neural network system has been introduced and implemented. The new system uses GA for updating and modification of the architecture and weight coefficients of the neural network.