Department of Electrical & Computer Engineering
University of Alberta, Edmonton Canada
Systems Research Institute, Polish Academy of Sciences
In system modeling, especially in its human-centric developments, we are faced with new challenges and opportunities that can lead to the enhancements of the ways in which the technology of fuzzy sets and Granular Computing, in general, becomes utilized. In the areas of finances, banking, business, services, and engineering we often encounter systems that are distributed and hierarchical, in which there is a significant level of knowledge generation and knowledge exchange. As a matter of fact, knowledge generation is inherently associated with the mechanisms of collaboration and knowledge sharing being realized between participating systems as well as through an interaction with the human users. In numerous ways of forming efficient conceptual and algorithmic vehicles of human-system interaction, fuzzy sets and Computational Intelligence (CI), in general, have been playing an important role.
The intent of this talk to bring into attention several ideas being of interest in the context of the modeling challenges identified above. The feature of human centricity and information granularity of CI constructs is the underlying leitmotiv of our considerations. We discuss new directions of knowledge elicitation and its quantification realized in the setting of fuzzy sets. The two main directions, that is (a) expert – based, and (b) data – based elicitation of membership functions have formed quite distinct avenues that are visible in the theory and practice of fuzzy sets. Being cognizant that fuzzy sets- information granules as being reflective of domain knowledge underpinning the essence of abstraction and dwell on numeric, data-oriented experimental evidence as well as perception of humans, we elaborate on an idea of knowledge-based clustering, which aims at the seamless realization of the data-expertise design of information granules. We emphasize the need for this unified treatment in the context of knowledge sharing where fuzzy sets are developed not only on the basis of numeric evidence available locally but in their construction we also actively engage the domain knowledge being shared by others.
To enhance the facet of human centricity of modeling with fuzzy sets, it becomes beneficial to establish a conceptual and algorithmic setup in which the predominantly numeric values of membership functions could be interpreted at the qualitative level of membership characterization. With this regard, we discuss a suite of algorithms facilitating a qualitative assessment of fuzzy sets, formulate a series of optimization tasks guided by well-formulated performance indexes and discuss the essence of the resulting solutions. It will be demonstrated that type-2 fuzzy sets emerge in this setting as a viable conceptual entity with sound algorithmic underpinnings. Proceeding with the CI models, we show how to endow the resulting modeling with an additional interpretation layer of type-2 fuzzy sets, which enhances the functionality of the existing fuzzy models and augments their human-centricity aspect.