A two-layer sparse coding model learns simple and complex cell receptive fields and topography from natural images

The classical receptive fields of simple cells in the visual cortex have be en shown to emerge from the statistical properties of natural images by for cing the cell responses to be maximally sparse, i.e. significantly activate d only rarely. Here, we show that this single principle of sparseness can a lso lead to emergence of topography (columnar organization) and complex cel l properties as well. These are obtained by maximizing the sparsenesses of locally pooled energies, which correspond to complex cell outputs. Thus, we obtain a highly parsimonious model of how these properties of the visual c ortex are adapted to the characteristics of the natural input. (C) 2001 Els evier Science Ltd. All rights reserved.