Learning the Appropriate Representation Paradigm
by Circular Processing Units

Sandro Ridella, Stefano Rovetta, and Rodolfo Zunino

DIBE - Department of Biophysical and Electronic Engineering
University of Genoa - Italy


Proceedings of the 1995 European Symposium on Artificial Neural Networks
Brussels, Belgium, April 1995, pp. 297-302.
(c) 1995 De Facto



Abstract

This paper presents a neural model that exhibits, in a single
architecture, the properties of other different, widely used schemes.
Such a model can be regarded as the result of generalization of
different feedforward architectures. The way its properties are
achieved is very inexpensive in terms of computational load. Moreover,
although the behavior of the single unit is conceptually different
from that of the well-known “multi-layer perceptron”, the network
layout is almost identical, thus allowing one to port all the
knowledge available about the classical model onto the new scheme
at very low cost.