Abstract
We are concerned with a fundamental idea of fuzzy neurocomputing manifesting as a realization of the synergy of fuzzy sets and neural networks. These two frameworks are complementary to a significant extent. The architectures of neurofuzzy systems benefit from mechanisms of explicit knowledge representation supported by fuzzy sets and a spectrum of learning methods being a genuine forte of neurocomputing. Several categories of fuzzy (logic) neurons are introduced, the topologies of the networks built with the aid of these neurons are discussed and learning schemes are presented. The linkages with Boolean networks are highlighted and mechanisms enhancing the interpretability of the network. The gradient-based learning is discussed as a generic mechanism of supervised learning. Selected architectures of neurofuzzy systems involving autoencoders and relational factorization are put forward.