Multiplierless multilayer feedforward neural networks

Abstract

A design algorithm for multiplierless 2-layer feedforward neural networks suitable for discrete input-output mapping is proposed in this paper. By using this algorithm, the obtained network has continuous valued weights at the first layer and single-term powers-of-two valued weights at the second layer such that no multiplications are needed in the computation after training. On the other hand, step function is used at the output layer and a simplified version of sigmoid function is used at the hidden layer as activation functions which simplifies digital hardware implementation further. Simulation results showed that such networks can retain nearly identical recall performance of the corresponding networks using continuous weights, while having increased computational speed in applications and reduced cost in digital hardware implementation.< >