Support vectors classification and incremental learning

According to whether the slack variable of the support vector is equal to zero, the support vector is divided into two categories, one is linear separable support vector and the other is non-linear separable support vector, in this paper. Using linear separable support vector set instead of support vector set in the incremental learning, Simple ISVM 1 (Simple Incremental Support Vector Machine Algorithm) is proposed. Because the linear separable support vectors are far less than support vectors, the speed of Simple ISVM 1 is fast than SVM-Inc.[1]. But the accuracy is slightly worse than SVM — Inc. For improving the accuracy of Simple ISVM 1, generalized linear separable support vector set is used to replace linear separable support vector set in incremental learning. The Simple IS VM 2 (Simple Incremental Support Vector Machine 2) is proposed. The generalized support vector is the support vector whose slack variable is less than a positive constant. Set a proper threshold, the accuracy of Simple ISVM 2 can be no less than SVM-Inc.[1] and the speed is fast than SVM-Inc. Empirical results show that the linear separable support vector set(or generalized separable support vector set) is the minimum subset which can approximately represent the historical set in the incremental learning, which is smaller than the support vector set.