Sistemas híbridos neurosimbólicos, estudo e implementação.

Abstract

The symbolist and connexionist approaches are utilized to guide the development of smart systems. Each approach has its advantages and its disavantages. Many a time, the disavantages of an approach are counterbalanced by the unification along with the advantages of the other approach. This is the main goal of the Neurosymbolic Hybrid Systems. This work presents SISNES as an implemented neurosymbolic hybrid system according to the subprocessing architecture. The symbolist and connexionist approaches in this system, are integrated so that an approach leads the main steps to the resolution of the problem while the other approach is subordinated to this very control. Fuzzy Logic is also used in this system, to processing data that are considered uncertain. SISNES combines great capability of explanation of the Symbolist Specialist Systems with the stoutness of the Neural Networks and with the ambiguous capability of representation of FuzzyXogic. Concepts, architectures, classification and examples of the neurosymbolic hybrid systems are presented. Experimental results, from SISNES accomplishment on building automation scope, are finally presented.