The relative effectiveness of using the unipolar and bipolar binaries is compared and the concept of hamming distance used for information storage and retrieval in neural network models is described. It is shown that the use of the bipolar retrieval key has the effect of doubling the length of the probe and stored vectors. This suggests useful modifications for improved performance and shows necessary conditions for a proper convergence when an unknown and erroneous unipolar or bipolar probe vector is used for retrieval. A new measure of distance defined over a field of real numbers is described.
Abdul Aziz Bhatti,
"Analysis of effectiveness of using unipolar and bipolar binaries and hamming distance in neural network computing," Optical Engineering 31(9), (1 September 1992). https://doi.org/10.1117/12.58856