Comparison of Inductive Versus Deductive Learning Networks
H. Madala
Department of Mathematics and Computer Science, Clarkson University,
Potsdam, NY 13699, USA
Abstract
This paper studies differences and similarities among inductive GMDH, deductive adaline, and back propagation techniques. All these are considered as parallel optimization algorithms because each one minimizes the output residual error in its own way. Self-organizing processes and criteria that help obtain the optimum output responses in the algorithms are explained through the collective computational approaches of these networks. The differences in empirical analyzing capabilities of the processing units are described. The relevance of local minima, which depend on various activating laws and heuristics, is studied by explaining the functionalities of these algorithms. This study is helpful in understanding the inductive learning mechanism in comparison with the standard neural techniques, and in designing better and faster mechanisms for modeling and predictions of complex systems.
