Abstract
Learning fuzzy rule-based systems can lead to very useful descriptions of several problems. Many different alternative descriptions can be generated. In many cases, a simple rule base similar to rule bases designed by humans axe preferable since it has a higher possibility of being valid in unforseen cases. Thus, the main idea of this paper is to define a minimal cost function and to generate minimal knowledge bases. Furthermore, this paper shows similarities between the generation of fuzzy systems and the generation of boolean functions on the base of minimal cost functions and it proposes criteria to learn human reasoning fuzzy rules.