Department of Neurobiology, University of California at Los Angeles
Correlations: how do they affect information transmission?
Knowing how correlations among different neurons affects information transmission is critical for how we design experiments: if correlations are unimportant, single neuron recordings are sufficient to determine population codes; if correlations are important, multi-neuron recordings are essential. We examined this issue theoretically, and have identified at least one regime in which definitive statements may be made: when the signal to noise ratio is high (meaning, loosely, that the responses to each stimulus lies in a region that is small compared to the total response space), sufficiently large correlations always increase the information between stimulus and response. We comment on the significance of this for neural coding, and relate it to the well-known result that pooling correlated responses reduces information transmission.