Position-based competition learning of neural-networks array

In this paper, we propose a model of neural-network array composed of a number of multilayer perceptrons (MLP) each of which can be automatically trained to recognize the different dynamics of time series data. The proposed array adopts a position-based competitive learning methods that puts members with similar dynamics close to each other. The proposed array model intends to deal effectively with switching dynamics problems and produce a map of the dynamics.