Citation: In H.G. Schuster & W. Singer (Eds.), Nonlinear Dynamics and Neuronal Networks, New York: Springer-Verlag, 305-334
Abstract: How the mammalian brain can rapidly but stably learn about a changing world filled with unexpected events is one of the most challenging scientific problems of our time. The brain?s ability to autonomously discover and learn appropriate representations of the world in real-time, without the intervention of an external teacher to signal that external changes have occurred or the nature of these changes, lies at the heart of this problem. Adaptive Resonance Theory, or ART, was introduced in 1976 (Grossberg, 1976a, 1976b) in order to analyze how brain networks can autonomously learn about a changing world in a rapid but stable fashion. Popular alternative models, such as back propagation, can learn only slowly, in an off-line setting, about a essentially stationary environment that includes an external teacher whose explicitly coded answers drive learning using non-local operations that seem to have non biological analog (Carpenter, 1989; Grossberg, 1988b; Park, 1982; Rumelhart, Hinton, and Williams, 1986; Werbos, 1974, 1982). The present chapter summarizes some recent results concerning how ART systems control distributed hypothesis testing and memory search in order to autonomously discover and learn predictive representations for recognition and recall.