Design and development of dynamic collaborative frameworks using concepts of knowledge-based networks

We have developed a suite of dynamic frameworks for clustering, classification, and regression problems of knowledge-based networks using collaborative approaches. For solving clustering problems, we provide a formulation of the model-integration problem using the principles of sharing prototypes and membership-functions and describe iterative algorithms that converge to an optimal solution. We show that the measure of proximity-distance is a suitable vehicle for quantifying the consensus of collaborative data sites.;For classification and regression problems, we present a new experience-consistent framework. By extending the performance index, we show that the domain knowledge captured by regression and classification models plays a regularization role in system identification problems. We demonstrate that the achieved consistency between collaborative sites can be quantified through fuzzy sets related to the parameters of the model.;In the development of an approach to fuzzy rule-based model identification realized in a collaborative framework of experiential evidence (data) and knowledge evidence (past experience), we demonstrate how to reconcile these two essential sources of guidance in the form of local regression models.;Using a radial-basis function neural networks approach, consistency is achieved using a connection value framework to reconcile data with past experience by considering gradient-based neural networks method.;The study provides architectural considerations, elaborates on essential communication mechanisms, and covers underlying algorithmic aspects of knowledge-based networks. We explain how the collaboration mechanism gives rise to higher order granular constructs such as type-2 fuzzy sets that emerge in a highly legitimate manner in distributed fuzzy modeling. We evaluate our methods with type-2 fuzzy sets.;The theoretical and algorithmic approaches to collaborative frameworks investigated in this study can be used as a foundation for further research in the area of distributed fuzzy modeling.