EXACT ART - A COMPLETE IMPLEMENTATION OF AN ART NETWORK

In this article we introduce a continuous time implementation of adapt ive resonance theory (ART). ART designed by Grossberg concerns neural networks that self-organize stable pattern recognition categories of a rbitrary sequences of input patterns. In contrast to the current imple mentations of ART we introduce a complete implementation of an ART net work, including all regulatory and logical functions, as a system of o rdinary differential equations capable of stand-alone running in real time. This means that transient behavior is kept in tact. This impleme ntation of ART is based on ART 2 and is called Exact ART. Exact ART in cludes an implementation of a gated dipole field and an implementation of the orienting sub-system. The most important features of Exact ART , which are the design principles of ART 2, are proven mathematically. Also simulation studies show that Exact ART self-organizes stable rec ognition codes that agree with the classification behavior of ART 2. ( C) 1997 Elsevier Science Ltd.