Identiication of Patterns Dimensionality by Self-organization

In the human brain, beneath the symbolic-type reasoning modeled by traditional rule-based systems, there is another level of computation. It is impleented in a subsymbolic substrate, in which operations are carried out by local interaction of simple computing element, without any central guidance. This substrate is especially evident in perception although, due to the physical \implementation" of the human brain, it is underlying every aspect of reasoning. In this paper, we argue that a distributed, sub-symbolic organization can be usefully employed to overcome the weakness of traditional pattern recognition systems. We present a self-organizing model that automatically learns the topology of an input space based on samples drawn from that space. Experiments are carried out to show that this model can built a template to be used for character recognition, autonomously inferring the essential topological features from a set of character images.