给定经验风险水平的支持向量回归机  

Support vector regression machines with given empirical risk

作　　者：罗林开[1] 叶凌君[1] 彭洪[1] 杨帆[1] 

机构地区：[1]厦门大学自动化系,福建厦门361005

出　　处：《华中科技大学学报（自然科学版）》2010年第10期47-51,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60704042);福建省自然科学基金资助项目(2008J0033);厦门大学211三期信息创新平台资助项目

摘　　要：针对传统支持向量回归机(SVR)中经验风险和置信风险的折中系数C难以选择的问题,提出了2种给定经验风险水平的支持向量回归新模型.相比于传统的SVR模型,新模型给出了结构风险化原则的另一种实现方式,为经验风险和置信风险的控制提供了更容易处理的方案.此方案一方面可满足对经验风险的具体要求;另一方面又避免了折中系数C的选取,减少了模型参数的选择时间.此外,新模型还可通过设置各个样本点上经验风险的大小,自然地处理样本点重要性不一样的问题.标准数据集上的实验验证了新模型的有效性.The trade-off coefficient C between the empirical risk and the confidence risk in traditional support vector regression machines SVR （support vector regression） is hard to choose. Thus, two new support vector regression models with given empirical risk are proposed. Compared with tradi- tional SVR, another type of realizing the structural risk minimization principle are given in the new SVR models, providing a more manageable solution for the control of the empirical risk and the confidence risk. These new SVR models not only satisfy the requirements of empirical risk, but also save the time of selecting model＇s parameters by avoiding the selection of the trade-off coefficient C. In ad- dition, the new models can easily solve the problem with different importance of each sample by setting the different empirical risk for each sample. The experiment results of some benchmark datasets illustrate the effectiveness of the new SVR models.

关 键 词：支持向量回归机 回归模型 经验风险 置信风险 结构风险最小化原则 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]