Assesing the Noise Immunity of Radial Basis Function Neural Networks
Previous works have demonstrated that Mean Squared Sensitivity (MSS) constitutes a good aproximation to the performance degradation of a MLP affected by perturbations. In the present paper, the expression of MSS for Radial Basis Function Neural Networks affected by additive noise in their inputs is obtained. This expression is experimentally validated, allowing us to propose MSS as an effective measurement of noise immunity of RBFNs.
Unable to display preview. Download preview PDF.