Classifying Artificial Neural Network Architecture

Abstract

Most Artificial Neural Network architecture appears to be heuristic in conception; devised by the skill and experience of the creator rather than derived from first principles. In contrast, however, we feel that it would be preferable to develop a formal taxonomy for Neural Networks in order that network size and designs may be matched more closely to the underlying problem. To this end, a theoretic and algebraic classification of Neural Network architecture has been developed.