Noise-aided signal detection in neural networks

Abstract

We demonstrate that a neural network composed of 'integrate' and 'fire' neurons can perform detection of subthreshold signal in the aid of a uniform white noise. The results are established for a neural system trained to determine the frequency components of an undetectable signal. The results are important because: 1) it permits further improvements to the already useful artificial neural networks, and 2) it accounts to the observed importance of noise in carrying out efficiently many neurophysiological processes.