Genetic algorithms like learning rule for neural networks

The paper explores the use of the genetic algorithm (GA) as a learning rule for feedforward neural networks (FNN). An extended comparison between GA and backpropagation error (BPE) are presented and the potential of information induction and reduction from the training set is revised for both methods. The GA was based only in the crossover and mutation genetic operators. Three case studies were used to evaluate the performance of both learning rules. F From the point of view of convergence time and the reduction of the cost function, BPE showed a better performance, but the GA extracted relevant information for the training set, finding an optimal solution. Given the adaptations shown by the GA approach, an online application was developed and evaluated. The results show the potential of this learning rule to learn and improve the response of the controller