Minimizing the effects of parameter deviations on cellular neural networks
Ronald Tetzlaff, R. Kunz, D. Wolf · International Journal of Circuit Theory and Applications · 1999
The sensitivity of cellular neural networks (CNN) against random parameter deviations is discussed in detail. For different CNN with erroneous parameters the probability is estimated that all cell outputs converge to the same stable fixpoint of the corresponding error free CNN. These results are compared with approximations based on a statistical independence assumption. The influence of deviated parameters is demonstrated for different image processing templates. We propose a new parameter learning method for minimizing the effect of template and bias deviations. In all treated cases a significant improvement can be observed by using this method. Copyright © 1999 John Wiley & Sons, Ltd.