Sensitive Analysis of Radial Basis Function Networks for Fault Tolerance Purposes

This paper introduces the concept of sensitivity in radial basis function networks. By applying a fault methodology combined with the information provided by the sensitivity of the performance error to faulty elements, faulting selection method can be simplified. In addition, the relation established between the sensitivity and a measure of the system fault toleracne permit to determine the most critical neural elements in the sense of fault tolerance. The theoretical predictions are verified by simulation experiments on two groups of problems-classification and approximation problems. In summary, this paper presents the application of sensitivity analysis for determining the most critical neural elements in the sense of fault tolerance.