Abstract: A general model for multi-valued bi-directional associative memory (BAM) is presented. It is based on the concept of similarity measure. From the general model, we can derive Wang's multivalued exponential BAM (MV-eBAM), the polynomial bi-directional hetero-corrector (PBHC) and several new multi-valued BAM models, such as the hyperbolic tangent BAM (HTBAM) and Caushy BAM. Among these models, we place special emphasis on the comparisons between PBHC and HTBAM. Simulation results show that the proposed HTBAM model has a competitive storage capacity and much greater error-correcting capability than the PBHC model. Finally, we apply the HTBAM to IP address recognition and obtain a novel algorithm for associative IP routing lookups.

Key words: multi-value, bi-directional associative memory, similarity measure, IP address recognition