Author:

Keywords:

nonlinear, non-linear, system theory, neural networks, CNN, cellular neural networks, circuit theory, complex systems, SISTA

Abstract:

Cellular neural networks (CNNs) is a powerful class of chip- implementable neural networks with many applications such as e.g. in image processing, associative memories and cryptography. Many existing implementations are with analog design, which requires more robust designs for template learning and optimization of CNN for the future. We propose to study these issues for single layer CNNs as well as further explore the potential of novel multilayer (two-layer) CNN chips. The extension to multilayered CNNs also poses new challenges towards stability analysis of such systems which will be deeper investigated. Optimization issues in CNNs CNNs require the study of global optimization techniques where the use of coupled minimizers will be further explored, such as for the design of coupled simulated annealing processes.