Abstract: It is a crucial issue to analyze the generalization performance of support vector machines in the field of machine learning. In this paper, following the idea that support vector machine can be constructed by minimizing the number of the essential support vectors, from the view of greedy method, a new kind of support vector machine named greedy support vector machine is presented combined with the problem of sparse learning. Applying the breast cancer data in UCI database, the performance of this kind of support vector machine on making a balance between estimation accuracy and sparsity of the solution is tested. Moreover for the greedy support vector machine designed, the generalization performance of this kind of support vector machine is derived in terms of the technique in empirical process.

Key words: greedy support vector machine, generalization performance, sparse learning

摘要： 支持向量机推广性能的分析是机器学习中的一项重要内容。依据可通过最小化本性支持向量个数来构造支持向量机的思路，结合稀疏学习，从贪婪方法的角度出发，提出了一种新的支持向量机，称之为贪婪支持向量机。利用UCI数据库中的乳腺癌数据集来测试贪婪支持向量机算法在平衡估计精确性和解的稀疏性方面的性能。针对设计的贪婪支持向量机，利用经验过程中的方法，得到这一类型支持向量机的推广性能。

关键词: 贪婪支持向量机, 推广性能, 稀疏学习