Stochastic Neural Interface with Selective Synapse

This paper presents a novel implementation of artificial neural networks using stochastic logic. This model includes a new synapse model that is highly selective, thus avoiding saturation during the recognition process when using large value input signals. This selectivity allows for an unsupervised learning algorithm that is also developed. As a result, a highly integrable pure digital implementation using a FPGA is achieved. The simulations and implementation of this new ANN model validate its use for the direct processing of real neuron signals.