Integrating rules and neural computation
Abstract
This paper introduces a hybrid system termed cascade ARTMAP that incorporates symbolic knowledge into neural network learning and recognition. Cascade ARTMAP, a generalization of fuzzy ARTMAP, represents rule-based knowledge explicitly and performs multistep inferencing. A rule insertion algorithm translates if-then symbolic rules into cascade ARTMAP architecture. Besides that initializing networks with prior knowledge improves learning efficiency and predictive accuracy, the inserted symbolic knowledge can be refined and enhanced by the cascade ARTMAP learning algorithm. By preserving symbolic rule form during learning, the rules extracted from cascade ARTMAP can be compared directly with the originally inserted rules. A benchmark study on a DNA promoter recognition problem shows that with the added advantages of fast and incremental learning, cascade ARTMAP produces performance superior to that of an alternative hybrid system.