Bridging Rate Coding and Temporal Spike Coding by Effect of Noise

Abstract

It is controversial whether temporal spike coding or rate coding is dominant in the information processing of the brain. We show by a two-layered neural network model with noise that, when noise is small, cortical neurons fire synchronously and intervals of synchronous firing robustly encode the signal information, but that the neurons desynchronize with moderately strong noise to encode waveforms of the signal more accurately. Further increase of noise just deteriorates the encoding. A positive role of noise in the brain is suggested in a meaning different from stochastic resonance, coherence resonance, and deterministic chaos.