A NEW SYNTHESIS APPROACH FOR FEEDBACK NEURAL NETWORKS BASED ON THE PERCEPTRON TRAINING ALGORITHM

In this paper, a new synthesis approach is developed for associative m emories based on the perceptron training algorithm. The design (synthe sis) problem of feedback neural networks for associative memories is f ormulated as a set of linear inequalities such that the use of percept ron training is evident. The perceptron training in the synthesis algo rithms is guaranteed to converge for the design of neural networks wit hout any constraints on the connection matrix, For neural networks wit h constraints on the diagonal elements of the connection matrix, resul ts concerning the properties of such networks and concerning the exist ence of such a network design are established. For neural networks wit h sparsity and/or symmetry constraints on the connection matrix, desig n algorithms are presented. Applications of the present synthesis appr oach to the design of associative memories realized by means of other feedback neural network models are studied. To demonstrate the applica bility of the present results and to compare the present synthesis app roach with existing design methods, specific examples are considered.