Fast learning algorithms for new L2 SVM based on active set iteration method

An L2 soft margin support vector machine (L2 SVM) is introduced in this paper. What is unusual for the SVM is that the dual problem for the constrained optimization of the SVM is a convex quadratic problem with simple bound constraints. The active set iteration method for this optimization problem is applied as fast learning algorithm for the SVM, and the selection of the initial active/inactive sets is discussed. For incremental learning and large-scale learning problems, a fast incremental learning algorithm for the SVM is presented. Computational experiments show the efficiency of the proposed algorithm.