Learning algorithms for logical neural networks

Abstract

Two training methods for multilayer logical neural networks are presented and discussed. They are the probabilistic logic node (PLN) reward-penalty algorithm of I. Aleksander (1989) and the PLN backpropagation algorithm of R. Al-Alawi and T. J. Stonham (1989). They are considered within the paradigm of reward-penalty training algorithms for analog networks and are found to be capable of solving various hard learning problems in speeds which are orders of magnitude higher than error backpropagation techniques for conventional nodes