Wed short talk: Cellular Neural Networks based Systems Science
By Prof. Dr. Kyandoghere Kyamakya
Univ. Klagenfurt (Klagenfur, Austria),
Institute of Smart Systems Technologies
In various fields of science and engineering (e.g. transportation, physics, mechanics, electronics, data mining, etc.) one has to cope with systems and phenomena that are generally very complex due to a series of inherent system attributes like strong nonlinearity, time-varying system parameters, stochastic behavior, and probably chaotic behavior, just to name a few. The mastering of a robust system modeling, simulation and optimization of such complexly behaving systems requires the addressing of particularly hard system-theoretical challenges. Furthermore, the systems engineering of latest related technological developments faces a series of practical challenges of particular severity. These challenges are especially related to issues like mathematical-computational constraints or requirements where the solving of complex system models expressed generally in form of complex differential equations (e.g. ODEs and PDEs in case of continous-time problems) or graphs (in case of discrete problems) should be performed both cost- and energy-efficiently at speeds much higher (possibly up to some orders of magnitude) than the ones provided by the currently related state-of-the-art.
This talks highlights major limitations of the relevant state-of-the-art and mainly briefly reviews selected illustrative application examples of a new and efficient paradigm for systems’ modeling, simulation, control, signal & data processing, and optimization in engineering. The novel paradigm is essentially based on “cellular neural networks constructed nonlinear oscillators systems”. These nonlinear oscillators are used as very important building bricks of a new systems’ science to reliably address a series of both hard system-theoretical and practical mathematical-computational challenges faced in the frame of the comprehensive systems engineering of complexly behaving technical systems. The new paradigm does essentially integrate and upgrade the good features of three well-known old paradigms, namely: (a) nonlinear oscillatory theory, (b) analog computing, and (c) neuro-computing including cellular neural networks (CNN).
In essence, the talk also briefly reviews selected own results obtained and demonstrating that CNN can efficiently solve the following selected problem cases:
a) speeding-up computations independently of the underlying platform
b) efficient solving of differential equations
c) efficient solving of discrete problems, e.g. Shortest Path problem
d) robust classification under difficult conditions
e) robust low-level image processing
f) robust nonlinear system identification + inverse problems solving