Unbiased Least Squares Support Vector Machine with Polynomial kernel
Meng Zhang, Lihua Fu · 2006
Although least squares support vector machine (LS-SVM) has dramatically reduced the complexity of standard support vector machine (SVM), LS-SVM still costs too much time in tackling regression problems with large data sets. This paper presents an unbiased LS-SVM with inhomogeneous polynomial kernel, which shortens the training time of LS-SVM significantly without obvious loss of accuracy. This new LS-SVM is especially suitable for solving the large scale problems including relatively low dimensional input vectors. We also give an upper bound analytically. When its dimensionality is below the bound, a regression problem can be solved more efficiently by the new LS-SVM than by the standard one. The applications to a synthetic example and to an image interpolation problem show the efficiency of the new LS-SVM