Knowledge refinement using fuzzy compositional neural networks

Abstract

Fuzzy relations as representational tools and fuzzy compositional operators as reasoning components, are user in this paper in order to represent knowledge expressed in semantic rules. Furthermore, neural representation and resolution of composite fuzzy relation equations provides knowledge refinement and adaptation to a specific context. A two-layer fuzzy compositional neural network is proposed in this work, with a learning algorithm changing the weights and minimize the error of the small context changes. © Springer-Verlag Berlin Heidelberg 2003.