ABSTRACT
Introduction
The primary mode of neural signaling involves the generation, transmission and transformation of spike trains by neurons in the central nervous system. The debate concerning spike timing versus rate coding has received increasing attention among neuroscientists, particularly within computational neuroscience (Shadlen & Newsome, 1994; Softky, 1995; The-unissen & Miller, 1995). It is widely believed that neurons in the nervous system are noisy, and that therefore only the rate or frequency of firing could reliably serve as a neural code. Recently this long held view has been directly challenged by a number of studies. Under tightly controlled input conditions, cortical neurons in the mammalian brain are capable of generating rather precisely timed spike trains, in both rat brain slices and behaving monkeys (Mainen & Sejnowski, 1994; Gallant, 1996). Information theoretic analysis has been successfully applied to this coding problem, and provides answers to the question of how much information is contained in a spike train (Bialek, Rieke, de Ruyter van Stevenick, & Warland, 1991). Although powerful, information theory has not yet answered the crucial question of whether the information measured is relevant to the neural system in the generation of behavioral output.
