Modeling the parallel development of multiple featuremaps and topography in visual cortex

Abstract

We study the development of a high dimensional feature map in a retinotopical organised visual cortex. The presented model is an extension of the elastic-net approach employing two basic terms: a stimulus dependent term and a ”regularising” constraint term. Interpreting each cortical element as a feature coding vector with selectivity across multiple dimensions, the parallel development of an optimised map with multiple features can be studied. The feature distributions of two feature maps show a structure similar to spatial frequency columns in cat visual cortex supporting the concept of a columnar architecture.