Bidirectional associative memories on the matching criterion of weighted Hamming distance

A generalized bidirectional associative memory (GBAM) model is proposed. It modifies Kosko´s bidirectional associative memory (BAM) on the matching criterion of Hamming distance into the matching criterion of weighted Hamming distance, so as to retain satisfactory associative performance even though the set of the training pattern pairs is ill-structured or its continuity is not held at all. Simulation results on storing and retrieving a set of binary picture pattern pairs as an applied instance are presented. It is shown that the GBAM model has a higher level of performance than Kosko´s BAM