Analysis of the noise performance of associative memories

The performances of two commonly used associative memories, generalized inverse (GI) and correlation memory matrix (CMM), are studied analytically and compared. The effects of crosstalk and a novel type of noise (memory noise appearing due to noisy training patterns) are also considered. Theoretical expressions are determined for the SNR (signal to noise ratio) gain of the GI and CMM memories in the auto-associative and heteroassociative modes of operation. It is found that the GI method performance degrades significantly in the presence of memory noise, while the CMM method is relatively unaffected. In overall performance, the GI method is superior to that of the CMM method, except at high values of input stimulus noise, or when the number of associations stored is close to the dimension of the key vectors. Theoretical expressions are plotted and compared with the results obtained from Monte Carlo simulations, and the two are found to be in excellent agreement