A fast constrained learning algorithm based on the construction of suitable internal representations

A novel approach to the task of training feedforward networks is presented based on the concept of constrained learning and on principles of optimal control theory. Minimization of the usual mean square error cost function is performed under a condition whose purpose is to facilitate the formation of suitable internal representations and thus accelerate learning. The algorithm is applied to binary benchmarks. Its performance, in terms of learning speed, is evaluated and found superior to the performance of the backpropagation algorithm and variants thereof.