Adaptive Resonance Theory: Self-Organizing Networks for Stable Learning, Recognition, and Prediction

Abstract

Adaptive Resonance Theory (ART) is a neural theory of human and primate information processing and of adaptive pattern recognition and prediction for technology. Biological applications to attentive learning of visual recognition categories by inferotemporal cortex and hippocampal system, medial temporal amnesia, corticogeniculate synchronization, auditory streaming, speech recognition, and eye movement control are noted. ARTMAP systems for technology integrate neural networks, fuzzy logic, and expert production systems to carry out both unsupervised and supervised learning. Fast and slow learning are both stable response to large non stationary databases. Match tracking search conjointly maximizes learned compression while minimizing predictive error. Spatial and temporal evidence accumulation improve accuracy in 3-D object recognition. Other applications are noted.