Pattern recognition using finite-iteration cellular systems

Cellular systems are defined by cells that have an internal state and local interactions between cells that govern the dynamics of the system. We propose to use a special kind of cellular neural networks (CNNs) which operates in finite iteration discrete-time mode and mimics the processing of visual perception in biological systems for digit recognition. We propose also a solution to another type of pattern recognition problem using a non-standard cellular neural networks called molecular graph networks (MGNs) which offer direct mapping from compound to property of interest such as physico-chemical, toxicity, logP, inhibitory activity MGNs translate molecular topology to network topology. We show how to design/train by backpropagation CNNs and MGNs in their discrete-time and finite-iteration versions to perform classification on real-world data sets.