Adaptive Resonance: An Emerging Neural Theory of Cognition

Abstract

Adaptive resonance is a theory of cognitive information processing which has been realized as a family of neural network models. In recent years, these models have evolved to incorporate new capabilities in the cognitive, neural, computational, and technological domains. Minimal models provide a conceptual framework, for formulating questions about the nature of cognition; an architectural framework, for mapping cognitive functions to cortical regions; a semantic framework, for precisely defining terms; and a computational framework, for testing hypotheses. These systems are here exemplified by the distributed ART (dART) model, which generalizes localist ART systems to allow arbitrarily distributed code representations, while retaining basic capabilities such as stable fast learning and scalability. Since each component is placed in the context of a unified real-time system, analysis can move from the level of neural processes, including learning laws and rules of synaptic transmission, to cognitive processes, including attention and consciousness. Local design is driven by global functional constraints, with each network synthesizing a dynamic balance of opposing tendencies. The self-contained working ART and dART models can also be transferred to technology, in areas that include remote sensing, sensor fusion, and content-addressable information retrieval from large databases.