Secondary descent active set algorithm based on SVM

To solve the slow convergence rate of the existing active set methods applied into optimization formulation of support vector machine, an active set algorithm based on the secondary descent method and the speculative assignment method is proposed. At each iteration of the algorithm, a projection operator is used to restrict the iterative vector onto the inequality constraints of optimization formulation, and then an adjustable step size is used to ensure the functional value of optimization formulation make further descent compared to the traditional active set method. As functional value ensure rapid and strictly descent at the end of each iteration, the global optimum solution can be obtained with rapid convergence rate. Experimental results show that iterations time and training time of the proposed method have been decreased obviously.