A self-organizing network for mixed category perception

Abstract

A neural network model capable of self-organizing in presence of multiple or mixed categories is presented. A certainty factor is derived about the decision on how well the features (due to single or mixed categories) have been interpreted by the network. One part of the model, the, monitor, controls the performance of the other part, the, categorizer in the self-organization process. The network automatically adjusts the number of nodes in the hidden and output layers, depending on the nature of overlap between the patterns from different categories. Mathematical derivations of the bounds on the number of nodes have been presented. The capability of the model is demonstrated experimentally both on one-dimensional binary strings and visual patterns.