Muruzábal, JorgeUniversidad Carlos III de Madrid. Departamento de Estadística2009-05-132009-05-131995-04http://hdl.handle.net/10016/4201This paper is concerned with the general stimulus-response problem as addressed by a variety of simple learning c1assifier systems (CSs). We suggest a theoretical model from which the assessment of uncertainty emerges as primary concern. A number of representation schemes borrowing from fuzzy logic theory are reviewed, and sorne connections with a well-known neural architecture revisited. In pursuit of the uncertainty measuring goal, usage of explicit probability distributions in the action part of c1assifiers is advocated. Sorne ideas supporting the design of a hybrid system incorpo'rating bayesian learning on top of the CS basic algorithm are sketched.application/pdfengAtribución-NoComercial-SinDerivadas 3.0 EspañaPredictionBayesian learningFuzzy logicUncertainty measuringworking paperEstadísticaopen access