Network analysis of multilayer perceptrons using stochastic computing techniques

A new approach to modeling the architecture of a pulse-mode digital multilayer neural network (DMNN) is presented. A stochastic model in which synaptic weights and neuron states are represented as probabilities and estimated as average rates of pulse occurrence in corresponding pseudo-random pulse sequences is developed. A random error (or noise) model is developed to estimate inaccuracy associated with stochastic computing in terms of mean and variance. DMNN feedforward architectures are modeled in VHDL using character recognition problems as testbeds. Computational accuracy is analyzed and the results of the statistical model are compared with the actual simulation results. Experiments show that the statistical model successfully predicts the accuracy of the operations performed in the DMNN