Finite iteration DT-CNN - new design and operating principles

Abstract

In this paper we propose to use the discrete-time cellular neural network (DT-CNN) in a finite iterate mode. In such a mode of operation no special requirements on template stability properties are needed. We propose a constructive back propagation based algorithm for template design. For a given number of iterations we can find optimal sequence of templates for a given problem to be solved. Our novel approach is demonstrated by a design of a digit recognition DT-CNN.