A LEARNING-PROCESS OF A STOCHASTIC FEEDFORWARD NEURAL-NETWORK

A positive reinforcement type learning algorithm is formulated for a s tochastic feed-forward neural network by minimizing a relative entropi c measure, and a learning equation similar to that of the Boltzmann ma chine is obtained. The learning of the network actually shows a simila r result to that of the Boltzmann machine in the classification proble ms of AND and XOR, by numerical experiments.