Soft Computing

A definition by Lotfi A. Zadeh reads: "Soft computing differs from convetional (hard) computing in that, unlike hard computing, it is tolerant of imprecision, uncertainty and partial truth. In effect, the role model for soft computing is the human mind. "The guiding principle of soft computing is: exploit the tolerance for imprecision, uncertainty and partiality to achieve tractability, robustness and low solution cost." A slightly shorter description (from the Wikipedia article on Lotfi A. Zadeh): "Soft computing dispenses with an exact analysis of a system in favor of a qualitative and, above all, interpretable description." Essentially (but not exclusively), soft computing deals with problem-solving paradigms that are modeled on biological fundamentals. This includes, for example, the human brain, which very quickly can find very good - not always optimal, but problem-solving - solutions (eg also on rules that are verbally formulated). Methods of soft computing include neural networks (eg perceptres or self-organizing maps of Kohonen maps), fuzzy systems and evolutionary algorithms. The application of these techniques allows problem solving as described above: for insecure output data (eg inaccurate or "blurred"), efficient and inexpensive. The "machine IQ" of technical systems can be significantly increased.

The lecture has above all the o. A. Paradigms of content, ie neural networks, fuzzy logic and evolutionary algorithms. This area is commonly referred to as "soft computing". The following topics will be discussed: biological basics, supervising learning neural networks (eg perceptres, multilayer percepts, radial basis funtion networks), unsupervised learning neural networks (eg competitive learning, self-organizing maps), first-order learning, Second Order Learning, Fuzzy Logic and Fuzzy Systems, Genetic Algorithms and Evolutionary Techniques. In each case, application examples are discussed and combinations of different methods are presented.

the exercises consist of both paper and computer exercises. The computer exercises work with MATLAB software and suitable toolboxes (for example on evolutionary algorithms or fuzzy logic).

• Contact persons for the lecture are Prof. Dr. med. Bernhard Sick and M.Sc. Benjamin Herwig.
• Formal, general information about the lecture (eg assignment to areas of application, credits, examination type) can be found in the module handbook. Degree Program -> Examination Regulations -> Module
• The lecture is also credited in electrical engineering!
• Up-to-date information about the current lecture (eg slides, bibliographical references, examination dates) will be provided in the Moodle of the University of Kassel.
• Link to the course catalog.