Abstract

Approaches for the evaluation of Hopfield’s memory is usually based on either statistical and/ or empirical studies. We show how the internal memory representation of the information and the input representations affect the performance of the memory. A new design for the associative memory is consequently suggested which makes use of a prioriknowledge of the input vector. Some of the earlier difficulties like that of the convergence to a complementary stored vector, or the closeness of vectors that does not comply with the Hamming distance criterion, and statistical convergence to the correct vector are eliminated by our modification. The convergence of vectors is now based on much stronger ground than before, since herein the convergence mechanics is made deterministic rather than statistical.

© 1988 Optical Society of America