通过迭代学习简化支持向量机决策函数

Simplify Support Vector Machines by Iterative Learning

支持向量机经过实践证明在小样本的情况下具有良好的泛化能力。但是在手写体数字识别的实验中,支持向量机被发现其在分类阶段的速度明显比神经网络要慢,因此在不影响支持向量机泛化能力的前提下简化支持向量机的决策函数,从而提高SVM的分类速度是很有意义的研究。利用迭代学习的方法来简化支持向量机的决策函数,实验证明本文的方法能够极大的简化SVM的决策函数,该方法易于实施。

Support vector machines （SVM） perform good generalization ability on various application fields in practice, especially for small sample problems. But in the SVM research on handwritten character recognition, it is substantially slower in test phase than neural networks with same generalization performance. So simplifying SVM without loss in generalization ability is important research in order to improve the classification speed of SVM. In this paper, we use iterative learning method to simplify the decision function of SVM. Computational results show that our method can simplify SVM effectively, it cam be implemented easily.