| The central idea in artificial intelligence research has been that of finding adequate representations to model human thought. To this end, research has been focused on fuzzy logic. In this thesis a new kind of fuzzy logic operators--uni-norm operators--has been introduced, which can model more faithfully such logical connectives as conjunction and disjunction. Subsequently their properties have been investigated and the general representability was established.; Another type of aggregation in fuzzy systems arise in attempts to model the way in which humans process information. Such types of aggregation--full reinforcement aggregation and self identity operators--are introduced, and their application in fuzzy systems is investigated.; Information fusion deals with the problem of combining knowledge from different information fusion deals with the problem of combining knowledge from different sources to obtain so called fused value. This work introduces the notion of a penalty function as a method for obtaining the fused value. A number of different penalty functions are investigated, and their applications are considered.; Fuzzy systems possess advantages in comparison with neural networks. In particular they consist of rules humans can understand, whereas neural networks are represented as impenetrable mass of weights. Putting neural networks in a form more amenable to human reasoning is a large step in understanding their work. The thesis shows how to represent neural networks through fuzzy systems.; Fuzzy systems are built by experts, whereas neural networks learn from examples. Enabling fuzzy systems also learn from examples greatly enhance their possibilities. In this work techniques used in building of neural networks--backpropagation method--is applied to fuzzy systems, so latter would be more consistent with both experts and examples. |
