Recurrent correlation associative memories with multiple-value

We propose a modified multiple-valued recurrent correlation associative memory with exponential weighting function (MEMRCAM) which is a network model with high storage capacity and strong error correction capability. The operations of this kind of networks depend on the similarity-measure computation and the input-output relation of each neuron being step function. This new model adopts the same principle as that proposed in the previous work (Chiueh and Goodman, 1991). We show the convergent property of the MEMRCAM and simulation result verifies the strong error correction capability with large storage capacity of the MEMRCAM.<>