A pivotal difference between Artificial Neural Networks and Fuzzy Cognitive Maps (FCMs) is that the latter allow modeling a physical system in terms of concepts and causal relations, thus equipping the network with interpretability features. However, such components are normally described by quantitative terms, which may be difficult to handle by domain experts. In this paper, we explore a reasoning mechanism for FCMs based on the Computing with Words paradigm where numerical concepts and relations are replaced with linguistic terms. More explicitly, we include triangular fuzzy numbers into the qualitative reasoning process attached to our model, thus proving further interpretability and transparency. The simulations show the potential behind the symbolic reasoning mechanism proposed in this study.

No abstract

In this paper, a review about the quality of the similarity measure and its applications in machine learning is presented. This measure is analyzed from the perspective of the granular computing. The granular computing allows analyzing the information at different levels of abstraction and from different approaches. The analysis shows that this measure is based on two basic aspects on the universe of objects: the granularity of the information and the principle that, similar problems have similar solutions. Using the measure, a method was formulated to build relations of similarity; these relations and other results have been used in improving machine learning techniques.

In this paper, a new method for solving classification problems based on prototypes is proposed. When using similarity relations for the granulation of the universe, similarity classes are generated and a prototypes is selected for each similarity class. Experimental results showing the performance of our method and comparing accuracy against other prototype selection methods reported.

No abstract