Unifying perspectives on neuronal codes and processing

Abstract

This talk will review a rapidly growing body of work on modeling neuronal ensembles that provides an understanding of a diverse body of experimental neurobiologiesl observations and has generated new insights into how neurons can implement a rich repertoire of powerful information-processing functions. The starting point is the Georgopoulos ‘Population Vector’ for modeling motor cortex experimental results, which has been expanded upon by Abbott, Sanger, Salinas, and others. Miller’s studies of the cricket cereal system provides the most detailed experimental example of a population code. This simple system exhibits many of the characteristics seen in primate retinal ganglion cells; visual, parietal and motor cortex pyramidal cells; as well as cells in the subcortical systems that generate eye movements. One form of population codes, which is closely related to Specht’s probability neural networks, is based on the assumption that neuronal firing rates encode the amplitudes of representations of probability density functions (PDFs). The PDF framework can be utilized to generate models of very large neuronal cortical circuits that incorporate the statistical inference ‘Pattern Theory’ of Grenander as well as generalizations of the Anderson, Van Essen and Olshausen routing circuits.