Comparison of Inductive Versus Deductive Learning Networks

Abstr act. This paper st udies differences and similarit ies among inductive GMDH, deductive adaline, and back prop agat ion techniques. All these are considered as parallel opt imizat ion algorithms because each one minimizes the out put residual error in its own way. Selforganizing processes and criteria that help obta in the optimum output responses in the algorithms are explained through th e collect ive computational approaches of these networks. The differences in empirical analyzing capabilit ies of the pro cessing units are described. T he relevance of local minima, which depend on various act ivat ing laws and heuristics, is studied by explaining the funct iona lit ies of these algorithms. This study is helpful in und erst and ing the inductive learning mechanism in comparison with the sta ndard neural techniqu es, and in designing better and faster mechanisms for modeling and predict ions of complex systems.