Abstract

   We present a C application through which we train a MLP (multilayer perceptron ) neural network, and, using the backpropagation algorithm, we obtain more valid sets of weights. Although each from these sets of weights assures the convergence of the network, we choose only one set of weights, the one for which the neural network has the best fault tolerance. We analyse all the p - fault hidden neurons combinations , where p is a natural number less than the number of neurons from the hidden layer.