International Journal on Artificial Intelligence Tools (IJAIT) c World Scientific Publishing Company COUPLING SUPERVISED AND UNSUPERVISED TECHNIQUES IN TRAINING FEED-FORWARD NETS * CRIS KOUTSOUGERAS and GEORGE PAPADOURAKIS† Computer Science Department, Tulane University, New Orleans, LA 70118, USA Department of Computer Science†, University of Crete, Heraklion, Crete, Greece ABSTRACT A popular approach to training feed-forward nets is to treat the problem of adapta- tion as a function approximation and to use curve fitting techniques. We discuss here the problems which the use of pure curve fitting techniques entail for the generaliza- tion capability and robustness of the net. These problems are in general inherently associated with the use of pure supervised learning techniques. We argue that a bet- ter approach to the training of feed-forward nets is to use adaptive techniques that combine properties of both supervised and unsupervised learning. A new formula- tion of the training problem is presented here. According to this formulation the net is viewed as two coupled sub-nets the first of which is trained by an unsupervised learning technique and the second by a supervised one. The same formulation gives rise to analytic expressions of the goals of the adaptation and leads to a new method for the adaptation of feed-forward nets. Keywords: neural nets, curve fitting, learning, function approximation. 1. Introduction One of the primary targets of artificial neural networks research is the develop- ment of computational models for approaching the problem of learning by examples. This problem can be stated as the automatic identification of some function where 0 *This research is supported by NSF grant MIP-8910616 and by Tulane’s COR