项目名称: 提高支持向量机处理复杂数据效能的方法研究
项目编号: No.61273291
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 王文剑
作者单位: 山西大学
项目金额: 79万元
中文摘要: 支持向量机(SVM)作为一种通用有效的机器学习方法,在许多领域得到成功的应用,但随着所处理问题的数据规模日趋增大、数据表示和结构的日趋复杂,SVM的学习效率和泛化能力受到极大制约,进而限制了它的进一步应用。本项目将系统研究提高SVM处理复杂数据效能的方法,主要内容包括:(1)基于半参数的SVM核函数构造方法。(2)SVM与粒度计算理论的深度融合,包括学习问题的重构和训练后超平面的几何分析与调整。(3)SVM框架下的高效降维算法,建立基于SVM的线性PCA和非线性PCA的线性规划求解机制和算法。(4)非结构化数据中语义特征函数的构造方法及结构化SVM的学习机理。(5)模型及环境参数相关的学习算法评价方法。(6)研制一个基于SVM的复杂数据处理平台。本项目研究成果将丰富和完善SVM的理论和算法研究、拓展其应用领域,对大规模复杂数据处理的研究有重要的理论意义和应用价值。
中文关键词: 支持向量机;复杂数据;效率;性能;
英文摘要: Support vector machine(SVM), as a general and effective machine learning approach, has been applied successfully in many fields. With the fleetly growing data size and more complex data expression and structure in actual application problems, the learning efficiency and generalization performance of SVM are restricted greatly, and then the application areas are also limited. In this project, the theory and approaches on improving the learning efficiency and generalization performance of SVM for complex data will be studied deeply. The main research contents include: (1) Kernel function selection approach based on semi-parameter. (2) Depth of fusion on SVM and granular computing, which includes reconstructing the optimal quadratic programming problem, analyzing the decision hyperplane in geometry after SVM training and correcting the hyperplane. (3) Effective dimension reduction approach under SVM-like frame. Linear PCA will be learned with linear programming, and the method will be extended to nonlinear/kernel PCA. Also, the algorithms for learning linear and nonlinear PCA will be provided. (4) Researches on constructing semantic feature function for non-structure data and exploring the learning mechanism of structural SVM. (5) Establishment of SVM evaluation system based on parameters of SVM model and learning
英文关键词: Support vector machine;Complex data;Efficiency;Performance;