一种基于全局代表点的快速最小二乘支持向量机稀疏化算法

发布时间：2018-05-15 02:20

本文选题：最小二乘支持向量机 + 稀疏化　；参考：《自动化学报》2017年01期

【摘要】：非稀疏性是最小二乘支持向量机(Least squares support vector machine,LS-SVM)的主要不足,因此稀疏化是LS-SVM研究的重要内容.在目前LS-SVM稀疏化研究中,多数算法采用的是基于迭代选择的稀疏化策略,但是时间复杂度和稀疏化效果还不够理想.为了进一步改进LS-SVM稀疏化方法的性能,文中提出了一种基于全局代表点选择的快速LS-SVM稀疏化算法(Global-representation-based sparse least squares support vector machine,GRS-LSSVM).在综合考虑数据局部密度和全局离散度的基础上,给出了数据全局代表性指标来评估每个数据的全局代表性.利用该指标,在全部数据中,一次性地选择出其中最具有全局代表性的数据并构成稀疏化后的支持向量集,然后在此基础上求解决策超平面,是该算法的基本思路.该算法对LS-SVM的非迭代稀疏化研究进行了有益的探索.通过与传统的迭代稀疏化方法进行比较,实验表明GRS-LSSVM具有稀疏度高、稳定性好、计算复杂度低的优点.
[Abstract]:Non sparsity is the main deficiency of Least squares support vector machine (LS-SVM), so sparsity is an important part of LS-SVM research. In the present research of LS-SVM sparsity, most of the algorithms use the sparse strategy based on iterative selection, but the time complexity and the sparsity effect are not ideal. In order to further improve the performance of LS-SVM sparse method, a fast LS-SVM sparse algorithm based on global representation point selection (Global-representation-based sparse least squares support vector machine, GRS-LSSVM) is proposed in this paper. The global data global density and global dispersion are considered, and the global data global is given. The representative index is used to evaluate the global representativeness of each data. Using this index, the most globally representative data is selected in all data and the support vector set is made up of the sparsity. Then the decision hyperplane is solved on this basis. It is the basic idea of the algorithm. The algorithm is not iterative and sparse for LS-SVM. By comparing with the traditional iterative thinning method, the experiment shows that GRS-LSSVM has the advantages of high sparsity, good stability and low computational complexity.

【作者单位】：中国人民大学信息学院;

【相似文献】