论文信息 - Constructing Support Vector Classifiers with Unlabeled Data

Constructing Support Vector Classifiers with Unlabeled Data

In this paper, a new method is presented to improve the speed and accuracy of SVMs with unlabeled data respectively: one method is to build SVMs with grid points which can be expected to speed SVMs in test phase; another method is to build SVMs with unlabeled data and it was shown that it can improve the accuracy of SVMs when there have a very few labeled data. These two methods are in the frame of quadric programming and no need to increase the computation cost of SVMs greatly, so it is expected to play an important role in some fields for the future.

Tao Wu | Hanqing Zhao

[1] Christopher J. C. Burges,et al. Simplified Support Vector Decision Rules , 1996, ICML.

[2] O. Mangasarian,et al. Semi-Supervised Support Vector Machines for Unlabeled Data Classiﬁcation , 2001 .

[3] Bernhard Schölkopf,et al. Improving the Accuracy and Speed of Support Vector Machines , 1996, NIPS.

[4] Ayhan Demiriz,et al. Semi-Supervised Support Vector Machines , 1998, NIPS.

[5] Sebastian Thrun,et al. Text Classification from Labeled and Unlabeled Documents using EM , 2000, Machine Learning.