Spatiotemporal encoding of a bar's direction of motion by neural ensembles in cat primary visual cortex.

Directionally selective neurons strongly fire when presented with a preferred direction of motion of a bar and only weakly respond otherwise. Intuition suggests these "specialist" neurons would be better suited to report this stimulus feature to higher visual centers than "generalist" neurons, neurons that broadly modulate their activity to the feature. However, as stimuli are encoded not by one cell but by large neural ensembles, we have studied the role of single-cell receptive field properties in stimulus representation. Using regression error statistics, we compared the performance of direction-of-motion estimators, using ensembles of neurons selectively responding to direction of motion and estimators using ensembles not specializing to direction. We found that direction-selective ensembles were no better at representing a bar's direction of motion than nonselective ensembles. Quite the opposite, the nonselective unit ensembles provided a better estimate of the direction (standard deviation of the error of 33 degrees) than the direction-selective ensembles (standard deviation of the error of 42 degrees). The nonselective neurons provided information through a latency code that is apparent only when a neuron's activity is considered in the context of the responses of neighboring neurons. These results suggest that models utilizing both generalist and specialist neurons may better reflect the encoding mechanisms that take place in sensory pathways.