On convergence rate of a convex relaxation of semi-supervised support vector machines

Semi-supervised support vector machine is a learning technique that comes as a compromise between supervised and unsupervised learning. Semi-super-vised support vector machines can be formulated as a mixed integer optimization problem, which is non-convex. In this paper, a cone programming relaxation for the semi-supervised support vector machine is discussed. The cone programming relaxation is used to derive results on convergence. Particularity, a discretization procedure is presented to approximate cone programming relaxation. The discretization method is shown to converge quadratically.