Una Comparación entre Estrategias Evolutivas y RPROP para la Estimación de Redes Neuronales.

Rprop has been recognized as one of the most powerful algorithms for training artificial neural networks; however, the evolution strategies algorithm is a strong competitor for solving optimization problems due to its capacity for searching the global optimum without the necessity of using information about of the gradient. In this paper, we compareboth algorithms using three non-linear time series from the real world, with the aim of determinate which algorithm offers betters results in the practice. The results indicate that evolution strategies converges faster than Rprop to the local optima, but in most cases the results obtained using Rprop are better in magnitude, even in practice the values of the objective function are very near.