Facing combinatory explosion in NAC networks

Abstract

A previous paper introduced a model of architecture called Neuromorphic Autonomous Controller (NAC) conformed by interconnected modules, each one in charge of growing its own network of cells. The integration of all internal structures results from automatic mechanisms which, triggering elementary actions, leads to correlations learning between local states. Though concluding in the viability of this approach, aimed at performing autonomous training, adaptation and control, combinatory explosion was mentioned as the main challenge to face. Focused on reducing this side effect, this paper acquaints with some enhancements to the NAC model, including: improving internal module structure, generating an alternative representation and defining a more accurate cooperation between modules. © Springer-Verlag 2004.