Modeling distributed concept representation in Hopfield neural networks

Learning in neural networks is usually identified with alterations in the networks connections. Learning rules perform these alterations gradually and are based on the deviation between the desired and effective responses to a stimulus. Learning can also be accomplished by synthesis methods, which determine the connections directly, without incurring the cost of gradual training. We introduce a synthesis method for binary Hopfield neural networks according to which learning is viewed as an optimization problem. A theory for concept representation is developed and synthesis criteria, used to define the optimization problems objective function and constraints, are presented. Experimental results are provided based on the use of simulated annealing to solve the optimisation problem. (C) 1999 Elsevier Science Ltd. All rights reserved.