Retinotopy and spatial phase in topographic maps

Self-organizing maps have been successfully applied as models of map formation in the mammalian primary visual cortex. We study a map of high-dimensional stimuli such as patches of filtered natural images onto a cortical layer of neurons. Since most map algorithms reproduce essentially distance relations the representation of stimuli of opposite spatial phase causes a distortion of the retinotopy of the map. We suggest that a related algorithm, the so-called adaptive subspace map, can produce topographic maps by converting the spatial phase to an internal property of local networks, such that phase is not topographically mapped at the experimental level. Finally, we compare other properties of this model with its biological counterpart. (C) 2000 Published by Elsevier Science B.V. All rights reserved.