Manifold based twin parametric-margin SVM for semi-supervised classification

作者：Chen W^*; Wang Q; Shao Y; Feng Z

来源：Journal of Computational Information Systems, 2013, 9(23): 9253-9260.

DOI：10.12733/jcis6949

• Manifold regularization
• Non-parallel hyperplanes
• Semi-supervised classification
• TPMSVM

Twin parametric-margin SVM (TPMSVM) is an excellent nonparallel-based kernel tool for supervised classification. However, TPMSVM can not be used to seimi-supervised classification (SSC) problem directly, and its generalization ability will usually deteriorate when labeled information is insufficient. In this paper, following the elegant MR framework, we propose an efficient semi-supervised classifier, termed as manifold based twin parametric-margin SVM (MTPMSVM). Our MTPMSVM is originally motivated to extend the supervised TPMSVM to deal with the SSC problem by exploiting the geometry information between labeled and unlabeled data. By adding the Manifold regularization terms, our MTPMSVM just need to solve two QPPs to obtain two parametric-margin hyperplanes. Experimental results on both artificial and real-world datasets show the benefits of the proposed approach.

• 出版日期2013
• 单位浙江大学