Abstract
In previous work we have derived a magnitude termed the 'Mean Squared Sensitivity' (MSS) to predict the performance degradation of a MLP affected by perturbations in different parameters. The present Letter continues the same line of researching, applying a similar methodology to deal with RBF networks and to study the implications when they are affected by input noise. We obtain the corresponding analytical expression for MSS in RBF networks and validate it experimentally, using two different models for perturbations: an additive and a multiplicative model. We discuss the relationship between MSS and the generalization ability. MSS is proposed as a quantitative measurement to evaluate the noise immunity and generalization ability of a RBFN configuration, giving even more generalization to our approach.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.References
Broomhead, D. S. and Lowe, D.: Multivariate functional interpolation and adaptive networks, Complex Systems, 2 321–355.
Haykin, S.: Neural Networks. A Comprehensive Foundation, (second edition, Prentice Hall, 1999).
Hochreiter, H. and Schmidhuber, J.: Flat Minima, Neural Computation, 9 (1997), 1–42.
Minai, A. A. and Williams, R. D.: Perturbation response in feedforward networks, Neural Networks, 7(5) (1994), 783–796.
Edwards, P. J. and Murray, A. F.: Towards optimally distributed computation, Neural Computation, 10 (1998), 997–1015.
Bernier, J. L., Ortega, J., Rodriguez, M. M., Rojas, I. and Prieto, A.: An accurate measure for multilayer perceptron tolerance to weight deviations, Neural Processing Letters, 10(2) (1999), 121–130.
Bernier, J. L., Ortega, J., Ros, E., Rojas, I. and Prieto, A.: A quantitative study of fault tolerance,noise immunity and generalization ability of MLPs, Neural Computation, 12 (2000), 2941–2964.
Segee, B. E. and Carter, M. J.: Comparative fault tolerance of generalized radial basis function and multilayer perceptron networks, IEEE Int. Conference on Neural Networks, 3 (1993), 1847–1852.
Townsend, N. W. and Tarassenko, L.: Estimations of error bounds for neural-network function approximators, IEEE Trans. on Neural Networks, 10(2) (1999), 217–230.
Parra, X. and Catala, A.: Fault tolerance in the learning algorithm of radial basis function networks, Proc. IEEE-INNS-ENNS Int. Joint Conference on Neural Networks, 3 (2000), 527–532.
Hagiwara, K., Hayasaka, T., Toda, N., Usui, S. and Kuno, K.: Upper bound of the expected trainig error of neural regression for a gaussian noise sequence, Neural Networks, 14 (2001), 1419–1429.
Catala, A. and Parra, X.: Fault tolerance parameter model of radial basis function networks, IEEE Int. Conference on Neural Networks, 2 (1996), 1384–1389.
Hegde, M. V., Naraghi-Pour, M. and Bapat, P.: Learning algorithms for fault tolerance in radial basis function networks, Proc 37th Midwest Symposium on Circuits and Systems, 1 (1995), 535–538.
Wang, Z. and Zhu, T.: An efficient learning algorithm for improving generalization performance of radial basis function neural networks, Neural Networks, 13 (2000), 545–553.
Bernier, J. L., Ortega, J., Rojas, I., Ros, E. and Prieto, A.: Obtaining fault tolerant multilayer perceptrons using an explicit regularization. Neural Processing Letters. 12(2) (2000), 107–113.
Choi, J. Y. and Choi, C.: Sensitivity analysis of multilayer perceptron with differentiable activation functions, IEEE Trans. on Neural Networks, 3(1) (1992), 101–107.
Bishop, C.: Training with noise is equivalent to Tikhonov regularization, Neural Computation, 7(1) (1995), 108–116.
Mackey, M. C. and Glass, L.: Oscillation and chaos in physiological control systems, Science, 197 (1977), 287–289.
Cherkassky, V., Gehring, D. and Mulier, F.: Comparison of adaptive methods for function estimation from samples, IEEE Trans. on Neural Networks, 7(4) (1996), 969–984.
Box, G. E. P. and Jenkins, G. M.: Time Series Analysis, Forecasting and Control. San Francisco, Holden-Day, CA, 2nd ed., 1976.
Gonzalez, J., Rojas, I., Pomares, H., Fernandez, F. J. and Diaz, A. F.: Multiobjective evolutionary optimization of the size, shape and position parameters of radial basis function networks for function approximation, IEEE Trans. on Neural Networks (in press).
Author information
Authors and Affiliations
Rights and permissions
About this article
Cite this article
Bernier, J.L., Díaz, A.F., Fernández, F.J. et al. Assessing the Noise Immunity and Generalization of Radial Basis Function Networks. Neural Processing Letters 18, 35–48 (2003). https://doi.org/10.1023/A:1026275522974
Issue Date:
DOI: https://doi.org/10.1023/A:1026275522974