A proposal for a Monte Carlo-based learning algorithm for multi-layer perceptron

In this paper we propose a Monte Carlo-based learning algorithm for multi-layer perceptron (MLP) which is characterized by the following features: first, the learning algorithm is able to associate to each weight of the MLP a probability distribution converging in distribution to the standardized normal one; then, the Monte Carlo-based learning algorithm performs a global search in the space of the weights.