Self-Organizing Neural Networks for Supervised and Unsupervised Learning and Prediction

Abstract

A neural network architecture for fast, stable, incremental learning of recognition categories and multidimensional maps is described. The architecture, called fuzzy ARTMAP, achieves a synthesis of fuzzy logic and Adaptive Resonance Theory (ART) neural networks. Fuzzy ARTMAP realizes a Minimax Learning Rule that conjointly minimizes predictive error and maximizes code compression. The system automatically learns a minimal number of recognition categories, or “hidden units”, to meet accuracy criteria, and the final code can include both fine and coarse categories. At each learning stage, system weights may be translated into a set of if-then rules that characterize the decision making process. Prediction is improved by training the system several times using different orderings of the input set, then voting on the outcomes. ART and ARTMAP networks are being applied to problems such as medical prediction, airplane design, electrocardiogram analysis, seismic recognition, adaptive software, and radar scene analysis.