Abstract
We introduce a concept of a fuzzy multiplexer, discuss its use in the design of logic networks and elaborate on the role of such networks in fuzzy modeling. In essence, a fuzzy multiplexer, fMUX, acts as a fuzzy switch whose output is determined on a basis of the logic values of the information inputs being switched (selected) by the select input. Multiplexers are generic building modules in digital systems. We show that networks of fuzzy multiplexers can play a similar role in fuzzy modeling. The two general design methodologies are studied and contrasted. The first is based on gradient-based learning and helps carry out parametric optimization. The second approach exploits a global genetic optimization and supports a structural and parametric development of the network. This becomes especially attractive in case of multivariable systems where the number of system variables has to be reduced. Experimental studies are reported for synthetic Boolean data and continuous (multivalued) problems.