A biologically plausible maturation of an ART network

Abstract

We present a numerical bifurcation analysis of a shunting neural network (SNN). Fold bifurcations appear to occur in the plane of parameters that are subject to post-natal maturation: the range and strength of lateral connections. The SNN is implemented as a content addressable memory (CAM) in Exact ART, a complete implementation of Adaptive Resonance Theory. The stability and functionality of Exact ART with the CAM in different dynamic regimes is maintained. Moreover, through a bifurcation, the learning behavior of Exact ART changes from forming local representations to forming distributed representations. Presently, we extend Exact ART by adding biologically plausible evolution equations for activation dependent maturation of lateral connections in a SNN (Van Ooyen et al., 1995). The resulting model is an epigenetic model of both morphological development and cognitive behavioral development.