Noise Performance of Second-Order Bidirectional Associative Memory

Yutaka KAWABATA  Yoshimasa DAIDO  Shimmi HATTORI  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E82-D   No.5   pp.993-998
Publication Date: 1999/05/25
Online ISSN: 
DOI: 
Print ISSN: 0916-8532
Type of Manuscript: PAPER
Category: Bio-Cybernetics and Neurocomputing
Keyword: 
BAM,  noise,  capacity,  error probability,  characteristic function (CF),  Hermite Gauss,  

Full Text: PDF(408.5KB)>>
Buy this Article




Summary: 
This paper describes the error probability of the second order BAM estimated by a computer simulation and an analytical calculation method. The computer simulation suggests that the iterations to retrieve a library pattern almost converge within four times and the difference between once and twice is much larger than that between twice and four times. The error probability at the output of the second iteration is estimated by the analytical method. The effect of the noise bits is also estimated using the analytical method. The BAM with larger n is more robust for the noise. For example, the noise bits of 0.15n cause almost no degradation of the error probability when n is larger than 100. If the error probability of 10-4 is allowable, the capacity of the second order BAM can be increased by about 40% in the presence of 0.15n noise bits when n is larger than 500.