Gaussian correlation associative memory

Abstract

This paper presents a high-capacity correlation-type associative memory neural network called the Gaussian correlation associative memory (GCAM). The Gaussian function is used as a weighting function. Using the Gaussian function has the same effectivity in discriminating correlations as the exponential function in the ECAM (Exponential Correlation Associative Memory), but has no limitation on the dynamic range in the real circuit implementation from which the ECAM suffers. The GCAM not only has high storage capacity and powerful error-correcting ability but also controllability of the basins of attractions of fundamental memories through adjusting the parameters of the Gaussian function.