Storing pattern pairs in a bicameral neural network

A model of a bicameral neural network proposed by S.C. Kak and M.C. Stinson (Electronics Letters, vol.25, p.203-5, 1989) is examined for its performance as a heteroassociative memory and for its ability to store pairs of patterns. The model is bidirectional, as any pattern of each pair can be used as a probe to retrieve the other pattern associated with it. Issues relating to the error-correcting capability of this model have been investigated experimentally. A method of detecting the errors using a technique called indexing is discussed. A method of upgrading the model to store patterns sets of more than two patterns each, such that any one of the patterns in the set could be used as a probe to retrieve all other patterns associated with it, is discussed. The results of comparing the error-correcting ability of the bicameral model to that of bidirectional associative memory show that the bicameral model compares favorably