A root location training method for polynomial cellular neural networks that implements totalistic cellular automata

The Polynomial Cellular Neural Network (PCNN) is a powerful non-linear processor that is capable of classifying non-linearly separable data points with a single neuron. Despite the capabilities of this model, the determination of the synaptic weights is not a trivial task. In this paper we present the root location training method as an effective, straightforward and high-speed procedure. Such method obtains the synaptic weights of a PCNN that implements any totalistic cellular automata behavior, dispensing the usage of heuristic methods such as genetic algorithms or numerical approaches such as quadratic programming procedures.