On the condition for fast neural computation

A fundamental question in theoretical neuro-science is to answer why neural systems can process information extremely fast. Here we investigate the effect of noise and neuronal collaborative activity on speeding up population decoding. We consider a one-dimensional stimulus encoded by a number of integrate-and-fire neurons. We find that 1) when noise is Poissionian, i.e., its variance is proportional to the mean, and 2) when a neural ensemble is at its stochastic equilibrium state, noise has the ‘best’ effect of accelerating computation, in the sense that the strength of external inputs is linearly encoded by the number of neurons firing in a short-time window, and that the neural system can use a simple strategy to decode the input rapidly and accurately. Interestingly, we also observe that under this noisy environment, the accuracy of neural decoding in short-time window is insensitive to the noise strength.