USING MULTI-KOHONEN SELF-ORGANIZING MAPS FOR MODELING VISUAL PERCEPTION

Abstract

This paper presents a new organizing principle for perceptual systems based on multiple Kohonen self organizing maps. These maps are arranged in order to model the global brain activity as seen on tomography pictures. In contrast to most of neural network models, a perceived object or knowledge is not represented by the final activity of a single neuron but by a configuration of activity of the whole neurons' set.