Sparsely symmetric interconnected associative memories

This paper presents and theoretically justifies a design procedure for an associative memory, with a cellular neural network structure, using singular values decomposition. By assuring the condition of symmetry of the interconnections, the suggested method generates a completely stable architecture. This guarantees that the obtained neural network never oscillates or becomes chaotic. The performance of the obtained associative memory are presented. The simulation results are compared with other different associative memory designs.