The role of correlations in information coding in the primary visual cortex

F. MONTANI; A. KOHN; S. R. SCHULTZ

Washington DC (USA)

Congreso; Neurons and Sensory Systems 2005; 2005

Neuron

Spikes fired by nearby neurons are often correlated, but the significance of this correlation for perception is unclear.  To evaluate the effect of correlation on the population coding of sensory information, we recorded responses of pairs of single neurons in primary visual cortex (V1) of the anaesthetised macaque monkey to stimuli of varying orientation and contrast.  Both trial-to-trial variability and synchrony were found to depend on both stimulus orientation and contrast (A. Kohn et al, J Neurosci 25:3661-73, 2005).  We used information theory to examine whether this stimulus-dependent correlation could contribute to the neural coding of orientation by pairs of V1 cells. We computed mutual information using the Nemenman-Shafee-Bialek estimator, and compared the ensemble information available from the pair of cells with a spike pooling code, and with the sum of the single cell information values. This allowed us to assess the degree of synergy (or conversely redundancy) in the coding. We paid particular attention to stimulus-dependent correlation because in its absence a pair of correlated cells with overlapping tuning would tend to interact redundantly and, consequently, the number of neurons that could be usefully pooled would be limited. The extent to which neurons interacted synergistically was found to be determined by this “stimulus-dependent correlation” component of the information. We found that this synergy compensated for redundancy effects due to overlapping tuning (signal similarity) and the stimulus-independent correlation term, leading on average to an independent coding of orientation in V1, despite the presence of substantial correlations. Furthermore, the coding for orientation is more effective if spikes are not pooled across cells, but instead the identity of the individual cells is preserved (in agreement with Reich et al., Science 295: 2566-68, 2001).  Our finding indicates that the size of the relevant neuronal pool is unlikely to be limited by trial to trial correlations in the response variability of V1 neurons.