正在加载图片...
第3期 花小朋,等:一种改进的投影孪生支持向量机 .389. 表4非线性模式下3种算法的实验比较 Table 4 Experimental comparision of three algorithms on UCI datasets with nonlinear kernel TWSVM PTSVM WPTSVM Dataset 正确率/% 训练时间/s 正确率/% 训练时间/s 正确率/% 训练时间/s Spectf(267×46) 83.43±7.15 14.63 82.49±8.36 14.65 84.03±5.17 1.73 Cleve (296x13) 85.12±5.77 7.22 85.93±4.90 7.82 85.24±4.56 3.40 Wpbc(198×33) 79.36±5.52 4.12 78.03±6.59 1.72 79.98±6.62 0.59 P_gene(106×57) 65.50±18.23 2.25 64.13±21.17 2.22 68.13±16.71 1.19 Sonar(208×60) 63.29±18.65 7.55 64.50±13.50 8.63 68.43±13.26 2.78 Spect (267x22) 84.42±9.04 11.42 84.42±7.43 4.01 84.80±8.27 5.54 Monks2(432×6) 68.77±15.24 5.02 68.71±3.43 6.43 68.99±11.07 4.57 Vertebral (310x6) 85.81±6.32 5.63 84.52±6.42 6.02 86.13±6.46 2.87 Monks3(432×6) 98.70±3.89 11.09 98.52±4.44 9.97 99.26±2.22 8.03 Breast gy(277×9) 75.86±5.89 5.24 73.57±4.10 4.04 75.94±6.69 0.87 结束语 [5]MANGASARIAN O L,WILD E W.MultisurFace proximal support vector machine classification via generalized eigen- 本文基于投影孪生支持向量机(PTSVM)提出一 values [J].IEEE transactions on pattern analysis and ma- chine intelligence,2006,28 (1):69-74. 种新的的非平行超平面分类器方法:加权投影孪生支 [6]PENG Xinjun,XU Dong.Bi-density twin support vector ma- 持向量机(WPTSVM)。WPTSVM不仅继承了PTS chines for pattern recognition[J].Neurocomputing,2013, VM方法的优点,而且在一定程度上提高了算法局部 99:134-143. 学习能力。除此之外,WPTSVM通过类间近邻图选 [7]CHEN Xiaobo,YANG Jian,YE Qiaolin,et al.Recursive 择少量边界样本构造优化问题约束条件,相当程度上 projection twin support vector machine via within-class vari- ance minimization[J].Pattern recognition,2011,44 (10/ 降低了二次规划求解时间复杂度。理论分析及实验 11):2643-2655. 结果均验证了本文所提算法的有效性。诚然,WPTS- [8 ]SHAO Yuanhai,WANG Zhen,CHEN Weijie,et al.A reg- VM在构造类内及类间近邻图时,需要花费额外的计 ularization for the projection twin support vector machine 算开销,特别是在学习样本数目较大时,算法计算复 [J].Knowledge-based systems,2013,37 (1):203-210. 杂度会偏高,这也是今后进一步研究的目标。 [9]YANG Xubing,CHEN Songcan,CHEN Bin,et al.Proxi- mal support vector machine using local information [J]. 参考文献: Neurocomputing,2009,73(1):357-365. [10]COVER T M,HART P E.Nearest neighbor pattern classi- [1]皋军,王士同,邓赵红.基于全局和局部保持的半监督 fication [J ]IEEE transactions on information theory, 支持向量机[J].电子学报,2010,38(7):1626-1633. 1967,13(1):21-27. GAO Jun,WANG Shitong,DENG Zhaohong.Global and local [11]WANG Xiaoming,CHUNG Fulai,WANG Shitong.On preserving based semi-supervised support vector machine minimum class locality preserving variance support vector [J].Acta electronica sinica,2010,38(7):1626-1633. machine[J].Patter recognition,2010,43(8):2753- [2]花小朋,丁世飞.局部保持对支持向量机[J].计算机研 2762. 究与发展,2014,51(3):590-597. [12]YE Qiaolin,ZhAO Chunxia,YE Ning,et al.Localized HUAxiaopeng,DING Shifei.Locality preserving twin sup- twin svm via convex minimization[J].Neurocomputing, port vector machines [J].Journal of computer research and 2011.74(4):580-587. development,2014,51(3):590-597. [13]皋军,黄丽莉,王士同.基于局部子域的最大间距判别 [3]DING Shifei,HUA Xiaopeng,YU Junzhao.An overview on 分析[J].控制与决策,2014,29(5):827-832. nonparallel hyperplane support vector machines[J].Neural GAO Jun,HUANG Lili,WANG Shitong.Local sub-do- computing and applications,2014,25(5):975-982. mains based maximum margin criterion [J].Control and [4]JAYADEVA,KHEMCHAND R,CHANDRA S.Twin sup- decision,2014,29(5):827-832. port vector machines for pattern classification [J].IEEE [14]邓乃杨,田英杰.支持向量机一理论、算法与拓展[M] transaction on pattern analysis and machine intelligence, 北京:科学出版社,2009:164-223. 2007,29(5):905-910. [l5]丁立中,廖士中.KMA-a:一个支持向量机核矩阵的近表 4 非线性模式下 3 种算法的实验比较 Table 4 Experimental comparision of three algorithms on UCI datasets with nonlinear kernel Dataset TWSVM PTSVM WPTSVM 正确率/ % 训练时间/ s 正确率/ % 训练时间/ s 正确率/ % 训练时间/ s Spectf (267×46) 83.43±7.15 14.63 82.49±8.36 14.65 84.03±5.17 1.73 Cleve (296×13) 85.12±5.77 7.22 85.93±4.90 7.82 85.24±4.56 3.40 Wpbc (198×33) 79.36±5.52 4.12 78.03±6.59 1.72 79.98±6.62 0.59 P_gene (106×57) 65.50±18.23 2.25 64.13±21.17 2.22 68.13±16.71 1.19 Sonar (208×60) 63.29±18.65 7.55 64.50±13.50 8.63 68.43±13.26 2.78 Spect (267×22) 84.42±9.04 11.42 84.42±7.43 4.01 84.80±8.27 5.54 Monks2 (432×6) 68.77±15.24 5.02 68.71±3.43 6.43 68.99±11.07 4.57 Vertebral (310×6) 85.81±6.32 5.63 84.52±6.42 6.02 86.13±6.46 2.87 Monks3 (432×6) 98.70±3.89 11.09 98.52±4.44 9.97 99.26±2.22 8.03 Breast_gy(277×9) 75.86±5.89 5.24 73.57±4.10 4.04 75.94±6.69 0.87 4 结束语 本文基于投影孪生支持向量机(PTSVM )提出一 种新的的非平行超平面分类器方法:加权投影孪生支 持向量机(WPTSVM )。 WPTSVM 不仅继承了 PTS⁃ VM 方法的优点,而且在一定程度上提高了算法局部 学习能力。 除此之外,WPTSVM 通过类间近邻图选 择少量边界样本构造优化问题约束条件,相当程度上 降低了二次规划求解时间复杂度。 理论分析及实验 结果均验证了本文所提算法的有效性。 诚然,WPTS⁃ VM 在构造类内及类间近邻图时,需要花费额外的计 算开销,特别是在学习样本数目较大时,算法计算复 杂度会偏高,这也是今后进一步研究的目标。 参考文献: [1]皋军, 王士同, 邓赵红. 基于全局和局部保持的半监督 支持向量机[J]. 电子学报, 2010, 38(7): 1626⁃1633. GAO Jun, WANG Shitong, DENG Zhaohong. Global and local preserving based semi⁃supervised support vector machine [J]. Acta electronica sinica, 2010, 38(7): 1626⁃1633. [2]花小朋, 丁世飞. 局部保持对支持向量机[ J]. 计算机研 究与发展, 2014, 51(3): 590⁃597. HUAxiaopeng, DING Shifei. Locality preserving twin sup⁃ port vector machines [J]. Journal of computer research and development, 2014, 51(3): 590⁃597. [3]DING Shifei, HUA Xiaopeng, YU Junzhao. An overview on nonparallel hyperplane support vector machines[ J]. Neural computing and applications, 2014, 25(5): 975⁃982. [4]JAYADEVA, KHEMCHAND R, CHANDRA S. Twin sup⁃ port vector machines for pattern classification [ J]. IEEE transaction on pattern analysis and machine intelligence, 2007, 29 (5): 905⁃910. [5] MANGASARIAN O L, WILD E W. MultisurFace proximal support vector machine classification via generalized eigen⁃ values [ J]. IEEE transactions on pattern analysis and ma⁃ chine intelligence, 2006, 28 (1): 69⁃74. [6]PENG Xinjun,XU Dong. Bi-density twin support vector ma⁃ chines for pattern recognition[ J]. Neurocomputing, 2013, 99: 134⁃143. [7] CHEN Xiaobo, YANG Jian, YE Qiaolin, et al. Recursive projection twin support vector machine via within⁃class vari⁃ ance minimization[J]. Pattern recognition, 2011, 44 (10 / 11): 2643⁃2655. [8]SHAO Yuanhai, WANG Zhen, CHEN Weijie, et al. A reg⁃ ularization for the projection twin support vector machine [J]. Knowledge⁃based systems, 2013, 37 (1): 203⁃210. [9]YANG Xubing, CHEN Songcan, CHEN Bin, et al. Proxi⁃ mal support vector machine using local information [ J]. Neurocomputing, 2009, 73(1): 357⁃365. [10]COVER T M, HART P E. Nearest neighbor pattern classi⁃ fication [ J ]. IEEE transactions on information theory, 1967, 13 (1): 21⁃27. [11] WANG Xiaoming, CHUNG Fulai, WANG Shitong. On minimum class locality preserving variance support vector machine[ J]. Patter recognition, 2010, 43 ( 8): 2753⁃ 2762. [12] YE Qiaolin, ZhAO Chunxia, YE Ning, et al. Localized twin svm via convex minimization [ J]. Neurocomputing, 2011, 74(4): 580⁃587. [13]皋军, 黄丽莉, 王士同. 基于局部子域的最大间距判别 分析 [J ]. 控制与决策, 2014, 29 (5): 827⁃832. GAO Jun, HUANG Lili, WANG Shitong. Local sub -do⁃ mains based maximum margin criterion [ J]. Control and decision, 2014, 29 (5): 827⁃832. [14]邓乃杨, 田英杰. 支持向量机—理论、算法与拓展[M]. 北京: 科学出版社, 2009: 164⁃223. [15]丁立中, 廖士中. KMA-a: 一个支持向量机核矩阵的近 第 3 期 花小朋,等:一种改进的投影孪生支持向量机 ·389·
<<向上翻页向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有