Distributed Polytope ARTMAP: A Vigilance-Free ART Network for Distributed Supervised Learning

The polytope ARTMAP (PTAM) suggests that irregular polytopes are more flexible than the predefined category geometries to approximate the borders among the desired output predictions. However, the categories cannot cover input space efficiently for the limited category expansion. This paper proposes distributed polytope ARTMAP (DPTAM), which seeks to combine the advantages of distributed coding and PTAM. DPTAM not only allows different polytopes expanding towards the input pattern simultaneously, but also permits of simplex overlap which is from the same desired prediction. Simulations show that DPTAM retains PTAM accuracy while ameliorating memory compression and region cover efficiency with less sensitivity to the variation of minimum simplex angle.