一种SVM增量学习算法

分析了SVM理论中SV（支持向量）集的特点，给出一种SVM增量学习算法，通过在增量学习中使用SV集与训练样本集的分类等价性，使得新的增量训练无需在整个训练样本空间进行，理论分析和实验结果表明，该算法能然保证分类精度的同时有效地提高训练速度。

An Approach to Incremental SVM Learning Algorithm

With the rapid growth of Internet, information updating get faster than before. The incremental learning algorithm has become one of the key techniques for handling and organizing this information. Although there are many traditional incremental learning algorithms available, it is still very difficult to find an efficient one, which can not only train online, but also has an upper bound of expected risk. Support Vector Machine (SVM) is a new emergent classification algorithm, which is based on the idea of structural risk minimization rather than empirical risk minimization. The theoretical analysis and experiment results show that SVM can find a global optimized result efficiently. Due to its perfect theoretical properties and good empirical results, SVM now attracts more attentions from researchers. It is proved that a SVM builds the final classification function on only a small part of the training samples (called SVs), and all the information about classification in the training set can be represented by SVs. However, there are still some drawbacks of SVM. For example, it is sensitive to the noisy datas and does not support incremental learning, and so on. In this paper, a new incremental SVM learning algorithm SISVM is presented. This algorithm is based on the classification equivalence between the SV set and the training set, and drastically decreases the learning time while incremental training is proceeded. The paper can be divided into 5 sections. Section 1 summarizes the properties of SVM. Section 2 introduces the theory of structural risk minimization and an upper bound of expected risk is also given. Section 3 analyzes the properties of SV set thoroughly. After that a new iterative learning method is presented to extend the SVM Classification algorithm to incremental learning area. At the end of this section, the computation complexity of the SISVM algorithm is given. Section 4 describes an experiment of incremental text classification based on the algorithm SISVM. The experiment results show that this algorithm improves the training speed efficiently and guarantees the classification precision as well. Finally, in section 5, conclusions and possible directions for future work are presented.