Improvement of bidirectional associative memories by using correlation significance

A correlation significance technique is introduced, which uses the significances between the training associations and between the neurons of the different fields, such that the storage capacity of Kosko bidirectional associative memories (BAMs) is improved. Based on the gradient descent method, an error function is constructed such that the maximal number of stored associations which are successfully recalled can be obtained. Finally computer simulations demonstrate that the proposed method has much better storage capacity than the existing methods.