Cue-Invariant Geometric Structure of the Population Codes in Macaque V1 and V2

We investigated the cue-invariant representation of visual patterns associated with surface boundaries in V1 and V2. We found that individual neurons exhibited a modest degree of tuning invariance in their responses to these patterns. This tuning invariance is stronger in V1 than in V2. At a population activity level, we studied the performance of a decoder trained with one cue in decoding patterns defined in another cue. We found that cue-transfer decoding is greatly enhanced when a geometric transform is first performed to align the population activities across cues. With this geometric transform, transfer decoding can be successful even when the tuning invariance of the individual neurons are destroyed by shuffling and when the neurons are from distinct populations. These findings suggest that abstract representation of these boundary-related patterns in V1 and V2 might be primarily encoded in the geometric structures of the population codes, rather than the cue-invariant tuning properties of the individual neurons.