A. Hyvarinen et Po. Hoyer, Topographic independent component analysis as a model of V1, organization and receptive fields, NEUROCOMPUT, 38, 2001, pp. 1307-1315
Independent component analysis (ICA) has been recently used as a model of n
atural image statistics and V1 simple cell receptive fields. Here we show h
ow to extend the ICA model to explain V1 topography as well, This is done b
y relaxing the independence assumption and ordering the basis vectors so th
at vectors with strong higher-order correlations are near each other. This
is a new principle of topographic organization, and may be more relevant to
natural image statistics than the more conventional topographic ordering b
ased on Euclidean distances. For example, our ordering leads to simultaneou
s emergence of complex cell properties: topographic neighborhoods act like
complex cells. (C) 2001 Elsevier Science B.V. All rights reserved.