The role of rhythm in nervous processing

Summary form only given. Throughout the history of neuroscience, interneuron communication has usually been considered under the simplifying assumption of frequency-modulated (FM) encoding. In newer artificial neural networks, this same view has justified the abstraction of real neural pulse trains into hypothesized neuron activation levels: continuously varying scalars representing the short-term average firing rate, or firing probability, of the original pulse train. After reviewing conventional FM and digital paradigms of neural computation, the author presents new physiological evidence in support of a different view: temporally fine-grained computation based on the coincident interaction and history-dependent processing of individual impulses. A biologically inspired artificial network that can learn to categorize temporally fine-grained impulse patterns is described, demonstrating the viability of this nontraditional approach