ABSTRACT
Neural networks are distributed structures designed to perform specific tasks. Thus, their design should be task or objective oriented, where then concrete analytic relationships between various network weights and parameters may be developed, with the subsequent dramatic reduction in the free parameters to be “learned” during the network training process (see Pados et al. [27-31]). When training is attained via interaction with the environment it corresponds to supervised learning, where in objectiveoriented network designs supervised learning algorithms should be dictated by the performance criterion pertinent to the objective (see Pados et al. [27-31]).
