Introduction

Bidirectional Associative Memory (BAM) is an extension of the Hopfield network introduced by Kosko [7]. The principal difference between BAM and the Hopfield network is that the units are divided into two sets and there are no connections between the units in each set. The BAM is described as being capable of learning pattern pairs, but this is a misleading description of its operation; it simply recalls stored patterns, which have been previously been divided into two parts by the network designer.