COURSE C5: Soft Computing Engineering Lecturer: prof. Witold Pedrycz Syllabus The objectives of this course are to discuss the principles and applications of soft computing engineering. As the title stipulates, the focal issues emphasize a variety of synergies of the existing soft computing technologies (that are at the heart of the engineering principles of soft computing), engineering principles of design of intelligent systems in the framework of soft computing, and discuss some general application-driven environments. It is assumed that the students have been exposed to the main technologies of soft computing, not necessarily viewed as a uniform platform but more in isolation (such as fuzzy sets, neural networks, evolutionary optimization). It is also assumed that they are familiar with the basic mathematics, optimization techniques and have some generic knowledge of computer science and/or engineering. Contents A brief review of underlying technologies of fuzzy sets (and granular computing), neurocomputing and evolutionary optimization. The intent is to discuss and contrast the main technologies of soft computing engineering, namely fuzzy sets, and granular computing to be more general, evolutionary optimization and neural networks. We emphasize and elaborate in detail on the complementary nature of the individual agenda of each technology and discuss possible facets of synergy between them Engineering system design principles and soft computing: decomposition, hierarchy, synergy, semantics of information granules. The development of complex human-centric systems is a multifaceted and complex activity. We cast it in the setting of soft computing by discussing how the fundamental notions of system design being at heart of any engineering-intensive activity (such as decomposition, hierarchy, top-down, bottom-up development) are discussed in detail Overall general topologies of systems of soft computing engineering. In spite of an evident diversity of soft computing systems, we can envision a general architecture that could serve as a blueprint (skeleton) of the detailed models. We associate the main functional blocks with the underlying technologies. Heterogeneous information environments of soft computing engineering. The existence of the heterogeneous sources of information is dominant in soft computing engineering; along with numeric information we encounter granular information that is quite often relate to domain knowledge that is more qualitative. We investigate the role of granular computing in formalizing and exploiting such domain hints. Front-ends and back-ends of soft computing architectures. As the soft computing architectures are highly human-centric, it becomes inevitable that we have to develop topologies that exhibit three clearly delineated modules, that is a processing unit equipped with an input interface and output interface. Validation and verification of soft computing architectures. The objective is to discuss ways in which soft computing constructs are validated and verified. Owing to the heterogeneous character of data, the verification requirements that need to be exercised are quite distinct from those we encounter in standard (numeric) system engineering; the issue of interpretability along with the fundamental tradeoff of accuracy and interpretability requires substantial attention Metalearning as a fundamental development mechanisms. The aspect of metalearning is critical to soft computing engineering: the development environments are not straightforward to master, there exists domain knowledge that could quite easily and efficiently enhance the existing optimization mechanisms (both those occurring at the learning of neural networks and guiding genetic optimization) Logic based architectures of soft computing. We discuss an environment that seamlessly combines logic and neurocomputing by building a hybrid structure that exhibits logic transparency and adaptability coming with neurocomputing. Pattern Recognition as an example of soft computing engineering. We discuss the engineering aspects of Pattern Recognition (PR) cast in the setting of soft computing and elaborate on different ways in which it augments the underlying fundamental paradigm of PR. Bibliography