Aggregate homotopy method for semi-supervised SVMs

Semi-supervised Support Vector Machines is an appealing method for using unlabeled data in classification. Based on a smooth approximation function named as aggregate function, a global aggregate homotopy method is presented in this paper. Compared to some existing algorithms, the new method is superior in no need of introducing extra variables or solving a sequence of subproblems. Moreover, the global convergence can make better local minima and then result in better prediction accuracy. Final numerical results reveals the efficiency of the method.