The capacity of a two layer network with binary weights

A statistical approach is used to determine the information capacity of a two-layer feedforward neural network with binary interconnections and integer thresholds. It is shown that the probability distribution of correct classifications of the two-layer network with binary weights is approximately Poisson, which is monotonic and exhibits a sharp transition at the capacity point. This behavior is similar to that observed in the Hopfield associative memory as well as some other linear classifiers. The capacity of the network was determined explicitly