Cellular neural networks based on terminal dynamics

Cellular neural networks (CNNs) based on the paradigm of the terminal dynamics are considered in the paper. The motivation for this has been the attempt to remove one of the most fundamental limitations of artificial neural networks (NN); their prescribed behavior compared with biological systems. The so called terminal dynamics approach is used, based on the concept of terminal attractors and repellers. Because of the violations of the Lipschitz condition at equilibrium points, terminal dynamics attains two fundamental properties: it is irreversible and non-deterministic. Based on this, a substantially new class of systems-unpredictable systems-have been introduced in the literature. In this contribution, CNNs are presented as unpredictable systems, and it is shown that sign strings at the critical points are able to control them. A cellular neural network has been able to reproduce a prescribed behavior with a prescribed accuracy by changing a combination of signs, due to interconnections between neighbouring neurones. That is why the presented CNNs are extremely flexible and could be used for different applications. The design of CNN template coefficients is considered as well