Are Neural Spike Trains Deterministically Chaotic or Stochastic Processes?

Abstract

Before examining neural interspike intervals to see how they might encode information, an essential question that has first to be answered is whether, under the unstimulated condition, the apparent randomness of the neural firing paltern renects deterministic chaos or a stochastic process. Here, we use short term predictability and the structure of the prediction residual to determine the dynamic characteristics of interspike intervals. As demonstrated in given computer simulations, unlike stochastic processes, deterministic chaos is highly predictable in the short term by linear and I or nonlinear prediction techniques. interspike intervals recorded from somatosensory cortex and hippocampus were, thus, analyzed by using the same techniques. The results show that the neuml spontaneous interspike intervals are poorly predictable in the short term, and the models that best fit the interspike intenals are linear (AR or ARMA) stationary processes. Therefore, the pattern of neural spontaneous firing can be characterized as stochastic ratber tban deterministically chaotic.