A delta-rule encoding for bidirectional associative memory

Summary form only given. A novel encoding scheme, called bidirectional learning (BL), was proposed to enhance the performance of the discrete bidirectional associative memory (BAM). The proposed encoding scheme is based on the perceptron learning algorithm in forward and backward directions. It is shown that BL yields a solution weight matrix within a finite number of iterations if there exists one with which each training pair can be recalled. The BL encoding scheme ensures a retrieval of a correct pattern pair for all training pairs when a perfect cue is given.<
>