Interactions in an ART neural system

Abstract

This work continues a series of papers about a neural network that incorporates adaptive resonance theory (ART) as a system of ordinary differential equations. The original, Exact ART, requires huge computational resources even for small-scale tasks. However, for technological applications it is necessary to devise more efficient computing algorithms. Here we propose simplifications of the original subsystem responsible for bottom-up and top-down data matching. In effect we formulate and solve a substitution problem that preserves only those features, which are essential for the overall system dynamics.