Duality between the rate code and the synchronous code in neural networks with chaotic inputs

摘要

It has been debated whether the brain uses the temporal spike coding or the rate coding as its dominant mode. In this report, we examine a two-layered stochastic neural network model receiving chaotic inputs generated from the Chen's attractor. With low noise, the neurons in the downstream layer fire synchronously, and intervals of synchronous firing robustly encode the signal. With moderate amount of noise, however, the neurons fire asynchronously to encode external stimuli more accurately in terms of time resolution. A possible role of noise in the brain distinct from stochastic resonance, coherence resonance, and deterministic chaos is indicated. This duality of coding also appears when other model parameters are modulated. In real situations animals may use the appropriate coding by switching from one coding scheme to another.