MULTILAYER FEEDFORWARD NEURAL NETWORKS WITH SINGLE POWERS-OF-2 WEIGHTS

A new algorithm for designing multilayer feedforward neural networks w ith single powers-of-two weights is presented in this correspondence. By applying this algorithm, the digital hardware implementation of suc h networks becomes easier as a result of the elimination of multiplier s. This proposed algorithm consists of two stages. First, the network is trained by using the standard backpropagation algorithm. Weights ar e then quantized to single powers-of-two values, and weights and slope s of activation functions are adjusted adaptively to reduce sum of squ ared output errors to a specified level. Simulation results indicate t hat the multilayer feedforward neural networks with single powers-of-t wo weights obtained using the proposed algorithm have similar generali zation performance as the original networks with continuous weights.