Adaptive Pattern Recognition with a Self-Organizing Learning Architecture

Abstract

Biological systems have enormous adaptability. We have developed a biologically motivated computer model, called the artificial neuromolecular (ANM) system, that is capable of differentiating patterns and tolerating a certain degree of noise in a self-organizing manner. Two biological features, biological-like structure-function relationship and evolution-friendliness, that facilitate self-organizing learning have been built into the system. With these two important features, the system can be molded to perform coherent functions in specific task. Three pattern sets were used to test the system, ranging from comparatively dissimilar (randomly generated patterns) to comparatively similar (printed Chinese characters). Each consists of one thousand patterns. Experimental results show that the system is able to achieve a high degree of pattern differentiation and degrade gracefully in the face of increasing noise.