To solve the problems of SVM in dealing with large sample size and asymmetric distributed samples, a support vector classification algorithm based on variable parameter linear programming is proposed. In the proposed ...To solve the problems of SVM in dealing with large sample size and asymmetric distributed samples, a support vector classification algorithm based on variable parameter linear programming is proposed. In the proposed algorithm, linear programming is employed to solve the optimization problem of classification to decrease the computation time and to reduce its complexity when compared with the original model. The adjusted punishment parameter greatly reduced the classification error resulting from asymmetric distributed samples and the detailed procedure of the proposed algorithm is given. An experiment is conducted to verify whether the proposed algorithm is suitable for asymmetric distributed samples.展开更多
研究路径跟踪线性规划支持向量机(path following linear programming support vector machine,PF-LPSVM)分类算法,利用路径跟踪法求解线性规划的高效性,提高线性规划支持向量机在大规模数据集上的学习效率。给出线性规划支持向量机的...研究路径跟踪线性规划支持向量机(path following linear programming support vector machine,PF-LPSVM)分类算法,利用路径跟踪法求解线性规划的高效性,提高线性规划支持向量机在大规模数据集上的学习效率。给出线性规划支持向量机的模型并将其标准化,导出用路径跟踪法求解线性规划向量机的关键公式,给出完整的算法流程。在随机数据集及UCI数据集上,将所提算法与LibSVM和牛顿法线性规划向量机(Newton-LPSVM,N-LPSVM)做比较,实验结果表明,所提算法用路径跟踪法提高LPSVM的学习效率是可行的,其适用于大规模数据集的学习。展开更多
以L1范数为例,设计了一个L1范数的大间隔分类器L1MMC(L1-norm Maximum Margin Classifier),主要特点如下:(1)间隔由L1范数的点到平面距离解析表示;(2)该分类器与SVM一样,通过最大化L1间隔,达到同时最小化经验风险和结构风险的目的;(3)...以L1范数为例,设计了一个L1范数的大间隔分类器L1MMC(L1-norm Maximum Margin Classifier),主要特点如下:(1)间隔由L1范数的点到平面距离解析表示;(2)该分类器与SVM一样,通过最大化L1间隔,达到同时最小化经验风险和结构风险的目的;(3)只需要通过线性规划进行求解,避免了SVM的二次规划问题;(4)分类精度达到甚至超过SVM.最后,在人工数据和国际标准UCI数据集上,验证了该方法的有效性.展开更多
基金the National Natural Science Foundation of China (70471074)China Postdoctoral Science Foundation(2005038042)Department of Science and Technology of Guangdong Province(2004B36001051).
文摘To solve the problems of SVM in dealing with large sample size and asymmetric distributed samples, a support vector classification algorithm based on variable parameter linear programming is proposed. In the proposed algorithm, linear programming is employed to solve the optimization problem of classification to decrease the computation time and to reduce its complexity when compared with the original model. The adjusted punishment parameter greatly reduced the classification error resulting from asymmetric distributed samples and the detailed procedure of the proposed algorithm is given. An experiment is conducted to verify whether the proposed algorithm is suitable for asymmetric distributed samples.
文摘研究路径跟踪线性规划支持向量机(path following linear programming support vector machine,PF-LPSVM)分类算法,利用路径跟踪法求解线性规划的高效性,提高线性规划支持向量机在大规模数据集上的学习效率。给出线性规划支持向量机的模型并将其标准化,导出用路径跟踪法求解线性规划向量机的关键公式,给出完整的算法流程。在随机数据集及UCI数据集上,将所提算法与LibSVM和牛顿法线性规划向量机(Newton-LPSVM,N-LPSVM)做比较,实验结果表明,所提算法用路径跟踪法提高LPSVM的学习效率是可行的,其适用于大规模数据集的学习。
文摘以L1范数为例,设计了一个L1范数的大间隔分类器L1MMC(L1-norm Maximum Margin Classifier),主要特点如下:(1)间隔由L1范数的点到平面距离解析表示;(2)该分类器与SVM一样,通过最大化L1间隔,达到同时最小化经验风险和结构风险的目的;(3)只需要通过线性规划进行求解,避免了SVM的二次规划问题;(4)分类精度达到甚至超过SVM.最后,在人工数据和国际标准UCI数据集上,验证了该方法的有效性.
