Comparación del Desempe o de Funciones de Activación en Redes Feedforward para aproximar Funciones de Datos con y sin Ruido.

The neural networks in many cases have given good results in the approach of functions in many applications, even so exist many problems that have not been able to solve. The selection of activation functions is made in agreement with the problem and to criterion of the investigator, sometimes by test and error. Commonly, the function of logistic activation has been more frequently used bringing good results. In Literature a standard criterion for the selection of these functions of activation in the neuronal networks does not exist, nor exists either an exhaustive investigation in this subject. It is therefore that the primary target of this I articulate is to obtain acriterion of selection for three functions of activation in a neuralnetwork feedforward with an hidden layer, comparing his performance with multiple neurons, to approximate the propose functions objective in [4] which were designed to evaluate the capacity of regression of models of neuronal networks.