In many real-world applications, unlabeled examples are inexpensive and easy to obtain. Semi-supervised approaches try to utilize such examples to boost the predictive performance. But previous research mainly focuses on classification problem, and semi-supervised regression remains largely under-studied. In this work, a novel semi-supervised regression method, semi-supervised LS-SVR (S²LS-SVR), is proposed on the basis of LS-SVR. Similar to the LS-SVR, one only solves a convex linear system in the training phrase too, thus largely speeding up training. Experimental results on corn data set indicate that our approach is feasible and efficient.

全文见：2011_8_6_885_892.pdf