无偏置v- 支持向量回归优化问题研究.pdfVIP

第27 卷 第6 期 控 制 与 决 策 2012 年 6 月 Vol. 27 No. 6 Control and Decision Jun. 2012 文章编号: 1001-0920 (2012) 06-0866-05 无偏置- 支持向量回归优化问题研究 丁晓剑, 赵银亮 (西安交通大学电子与信息工程学院，西安710049) 摘 要: 在高维特征空间中, 具有支持向量机形式的学习机的决策超平面倾向于通过原点, 并不需要偏置. 但在- 支 持向量回归机(-SVR) 中存在偏置, 为了研究偏置在-SVR 中的作用, 提出了无偏置的-SVR 优化问题并给出其 求解方法. 在标准数据集上的实验表明, 无偏置-SVR 的泛化性能好于-SVR. 根据对偶优化问题的解空间分析, 偏 置 不应包含在-SVR 优化问题中, -SVR 的决策超平面在高维特征空间中应通过原点. 关键词: 偏置；- 支持向量回归机；优化问题 中图分类号: TP273 文献标识码: A Study on - support vector regression optimization problem without bias DING Xiao-jian, ZHAO Yin-liang (School of Electronics and Information Engineering，Xi’an Jiaotong University ，Xi’an 710049 ，China. Correspondent: ZHAO Yin-liang ，E-mail ：zhaoy@) Abstract: In the high-dimensional feature space, the decision hyperplane of learning machine with support vector machine style tends to pass through the origin and the bias is not need. However, bias exists in -support vector regression (-SVR). To study the role of bias in -SVR, optimization formulation of -SVR without bias is proposed and the corresponding method of solving the dual optimization formulation is presented. The experimental results on benchmark data sets show that the generalization ability of -SVR without bias is better than -SVR. Based on the analysis of solution space on dual optimization formulation, the bias should not be contained in the optimization formulation of -SVR, and the hyperplane of -SVR should pass through the origin. Key words: bias ；-support vector regression ；optimization formulation 1 引引引 言言言