一种鲁棒回归支持向量机及其学习算法

为了提高支持向量机的泛化能力，给出了一个鲁棒损失函数，利用它建立了鲁棒支持向量机，并利用对偶原理推导出其对偶优化问题的形式，在此基础上设计了局部梯度算法，在这种算法中每次迭代只改变两个优化变量的值。随后分析了算法的收敛性条件，给出了学习步长的选择依据，最后用一个仿真实例来说明所提出的支持向量机的学习性能，比标准支持向量机具有更好的鲁棒性。

Robust Regression SVM and Its Learning Algorithm

In order to increase the generalization ability of SVM(support vector machine),a robust loss function is given.A robust support vector machine is put forward.The dual optimal problem formation is deduced using a dual theory.A local gradient algorithm is designed,and only two optimal variables are updated in every iteration.The convergence condition of the algorithm is analyzed,and a formula to select learning step size is given using the convergence condition.The simulation results show that the robust support vector machine performs significantly well and it possesses stronger robustness than that of the original support vector machine.