Study of stochastic resonance in hierarchical IF neural networks

This paper studies the stochastic resonance effect in a hierarchical integrate-and-fire (IF) neural networks characterized by the measure of average mutual information. It is shown that, as the internal noise intensity increases, the average mutual information can be optimized at an optimal non-zero noise level. It is also noted that the maximum of mutual information obtained in the resonant region can be further enhanced for a suitable signal frequency. This leads to the more efficient transduction of signals in the IF neural network. Correspondingly, the interval time of spikes, regularly established at the optimal noise level, is in accord with the input signal period. The present results are meaningful to the study of signal transduction through hierarchical IF neural networks.