Statistical analysis of information expression in V1 receipt field with sparseness constraint

Abstract

We discuss the information representation in V1 neurons including their surround modulation, with specific interests on sparseness in their coding. We compared statistical indices, kurtosis and KL divergence, between the surround-modulation model and a classical receipt field (CRF) model to evaluate the sparseness of coefficients represented in the models. The results showed a low kurtosis in the distribution of coefficient with the surround-modulation model, but a low KL distance between the models indicated their distributions fairly closed to each other. Next, we modified the nonlinear function that controls sparseness in the cost function with the aim of improving the kurtosis. The modification did not improve the distribution. These results suggest that surround modulation would decrease sparseness of the information representation in V1 cells, or that representation might not be characterized solely by sparseness.