Cosyne 2005 Workshops
March 21-22, 2005
Snowbird, Utah
Workshop Title
Information coding and computation
Organizer(s)
Bruno Averbeck and Peter Latham
Abstract
The objective of this symposium is to discuss the role of correlations in neural coding and computation. Our speakers offer a broad, but well integrated selection of both empirical and theoretical approaches to this problem, which has drawn a great deal of controversy. Recently there have been multiple empirical studies which have shown that correlations between neurons, in several systems (e.g., in the retina, primary visual cortex, somatosensory barrel cortex) carry only a small amount of information about stimuli or movements. However, other studies have reported more information in correlations (e.g., in the LGN). The different results may be due to different coding strategies used in different brain areas or to different methods for measuring the importance of correlations. We aim to present the data, the metrics, and the interpretations of these studies to help clarify this critical question in systems neuroscience. The second perspective of the proposed symposium is primarily theoretical, and is represented by Bair, Pouget and Salinas. This work has shown that the correlations between neurons, within a network, can be the result of the type of processing being carried out by the network. Furthermore, the effect of the correlation on neural coding depends upon the particular task the animal is executing, and correlations can be related to features of the neuronal responses, including their integration time in the case of visual cortex. This theoretical work is beginning to broaden our understanding of the potential sources and effects of correlations, as well as integrating the study of correlations into the study of decisions, behavior, and models which relate the anatomical structure of particular cortical areas to their single cell responses. Thus, across the symposium, the speakers represent work which is measuring the size and time-scale of correlations across multiple cortical areas, and estimating the amount of information which can be conveyed by these correlations, as well as work which is modeling the potential sources of correlations at the network level.
