Guaranteed recall of all training pairs for exponential bidirectional associative memory

Abstract

Several methods of guaranteeing the recall of all training pairs for exponential bidirectional associative memory are presented. If the minimum Hamming distance between patterns is known, the radix can be determined to provide a certain basin of attraction. Otherwise the statistical minimum distance can be applied. A weight learning algorithm based on a pseudo-relaxation algorithm is also developed to make all training pairs recallable with a small radix.