FasBack: matching-error based learning for automatic generation of fuzzy logic systems

Many research works have been reported with respect to the relation between neural and fuzzy systems. Looking for a synergistic relation of these technologies, an important property of neural network-based systems is their learning capacity, that permits to embed self-organization in fuzzy logic systems. In this paper, a new neuro-fuzzy system, called FasBack, is proposed, that combines learning based on prediction error minimization and pattern matching. FasBack adds error-based learning to a previously proposed model, called FasArt, which extended and formalized neural networks models of the ART family, as fuzzy logic systems. Experimental results are presented in nonlinear systems identification problems, typically used in the literature.