A Randomized Algorithm for the Exact Solution of Transductive Support Vector Machines

Figures & data

FIGURE 1 Smallest enclosing ball: extremes (red circled in red) are the points essential for the solution, violators (in blue) are the points lying outside the ball, a basis is a minimal set of points having the same ball.

TABLE 1 Error rate on datasets for supervised SVM, state-of-the-art transduction algorithms (as reported in Chapelle 2008), and STSVM

datasets	SVM		cS^3VM	CCCP	S^3VM^light		Newton	STSVM
TwoMoons	50.2	61	37.7	63.1	68.8	22.5	11	0
Coil3	66.6	61.6	61	56.6	56.7	61.6	61.5	0
Coil20	24.1	25.6	30.7	26.6	25.3	12.3	24.1	1.4

FIGURE 2 (a) Distribution and exact solution for the two moons dataset (4.000 unlabeled, 2 labeled as triangle and cross). (b) Weights distribution for the whole set of points at the final round. Support vectors of the optimal solution are encircled.

Chapelle, O. 2008. Optimization techniques for semi supervised support vector machines. Journal of Machine Learning Research 9:203–233.

 Web of Science ®Google Scholar