Category ART: A variation on adaptive resonance theory neural networks

In this paper we describe Category ART, a variation on the adaptive resonance theory (ART) neural network models. Category ART is a predictive ART architecture because it incorporates an ART module to be able to learn to predict a prescribed category given a prescribed n-dimensional input vector a. In contrast to ARTMAP, Category ART contains only one ART module and the map field algorithm has been simplified. The remaining ART module in a Categoty ART can be either Fuzzy ART or ART2-A. Its performance is demonstrated on a benchmark neural network test, the two spiraals problem.

• Abstract
• 1. Introduction
• 2. Basic features of ART systems
• 3. Fuzzy ART algorithm
  • 3.1. Preprocessing
  • 3.2. Category choice
  • 3.3 Learning
• 4. Category ART algorithm
• 5. Simulation: Learning to tell two spirals apart
• 5. Conclusions
• 6. Acknowledgement
• 7. References