Feedback neural nets for associative memories-an overview and some new results

This paper presents an overview for the existing synthesis procedures and some new developments for associative memories using feedback neural networks. Specifically, existing results including the outer product method, the projection learning rule, the eigenstructure method, and the sparse design method are reviewed for a class of feedback neural networks. Under certain sparsity constraints on the interconnection matrix, the class of neural networks considered herein becomes neural networks (called cellular neural networks) whose dynamics can be completely specified by a cloning template which specifies the weights from every neuron to its local neighborhood. We develop a design algorithm which makes it possible to determine in a systematic manner cloning templates for neural networks with or without symmetry constraints on the interconnection weights