The Two-level Quantization Strategy of Quadratic Hebbian-type Associative Memories

The strategy of two-level quantization for quadratic Hebbian-type associative memories are proposed and their performances are analyzed. The strategy reduces the interconnection values and makes the hardware implementation of Hebbian-type associative memory more feasible. The probabilities of direct convergence of the quantized networks are explored and simulations are also used to verify the proposed strategies. The results show that two-level quantization of quadratic Hebbian-type associative memories have approximately the same convergent capability as their original networks if the original recall capacity is higher. The performance of the first order and second order networks are also analyzed in this paper. The results demonstrate that the second order networks have better performance than the first order networks after either two-level or three-level quantization.