Capacity of Two-Layer Networks with Binary Weights

Abstract

The capacity C_b of two later (N - 2L - 1) feed-forward neural networks is shown to satisfy the relation O(W/(lnW)) ≤ C ≤ O(W). Here N - 2L - 1 stands for the networks with N input units, 2L hidden units and one output unit. W is the total number of weights of the networks. The weights take only binary values and the hidden units have integer thresholds.

Files

Additional details