A Fast Robust Learning Algorithm for RBF Network Against Outliers
Training data set often contains outliers, which can cause substantial deterioration of the approximation realized by a neural network. In this paper, a fast robust learning algorithm against outliers for RBF network is presented. The algorithm uses the subtractive clustering(SC) method to select hidden node centers of RBF network, and the gradient descent method with the scaled robust loss function(SRLF) as the objective function to adjust hidden node widths and the connection weights of the network. Therefore, the learning of RBF network has robustness on dealing with outliers and fast rate of convergence. The experimental results show the advantages of the learning algorithm over traditional learning algorithms for RBF network.
Unable to display preview. Download preview PDF.
- 1.Haykin, S.: Neural Networks, A Comprehensive Foundation, 2nd edn., pp. 298–305. Tsinghua University Press & Prentice Hall (2001)Google Scholar
- 6.Chien-Cheng, L., Pau-Choo, C., Jea-Rong, T., Chein-I, C.: Robust Radial Basis Function Neural Networks. IEEE Transactions on Systems, Man, and Cybernetics 29(6), 674–685 (1999)Google Scholar
- 7.Meiqin, L., Xiaoxin, L.: A Robust Learning Algorithm for RBF Neural Networks. Journal of Huazhong University of Science and Technology (Nature Science) 28(2), 8–10 (2000) (In Chinese)Google Scholar
- 12.Chiu, S.L.: Fuzzy Model Identification Based on Cluster Estimation. J. Intell. Fuzzy Systems 2(3), 267–278 (1994)Google Scholar
- 13.Weixiang, Z., Dezhao, C., Shangxu, H.: Detection of Outlier and A Robust BP Algorithm Against Outlier. Computers and Chemical Engineering 28, 1403–1408 (2004)Google Scholar