鲁棒性的模糊聚类神经网络

Robust Fuzzy Clustering Neural Networks

针对模糊聚类神经网络FCNN(fuzzyclusteringneuralnetwork)对例外点(outliers)敏感的缺陷,通过引入Vapnik’sε-不敏感损失函数,重新构造网络的目标函数,并根据拉格朗日优化理论推导出新的鲁棒模糊聚类神经网络及其算法(robustfuzzyclusteringneuralnetworks,简称RFCNN).RFCNN有效地克服了FCNN对例外点敏感之缺点并且能得到合理的聚类中心.仿真实验结果表明,RFCNN较之于FCNN有更好的鲁棒性.

In this paper a new robust fuzzy clustering neural networks （RFCNN） is presented to resolve the sensitivity of the fuzzy clustering neural network （FCNN） to outliers in real datasets. The new objective function of RFCNN is obtained by introducing Vapnik＇s e-insensitive loss function, and RFCNN＇s update rules are derived by using Lagrange optimization theory. Compared with the FCNN algorithm, RFCNN is much more robust to outliers in the datasets. Experimental results demonstrate the effectiveness of RFCNN.