项目名称: 具有可调节范数的支持向量机模型与算法的研究
项目编号: No.11201480
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 张春华
作者单位: 中国人民大学
项目金额: 22万元
中文摘要: 支持向量机方法是借助于最优化方法解决机器学习问题的有力工具,已经被成功地应用到数据挖掘的许多领域。在标准支持向量机的优化模型中使用的是2范数,近年来,出现了若干基于不同范数的支持向量机,由于其在特征选择等问题中的显著效果,逐渐成为支持向量机方法中新的研究热点之一。本项目拟从最优化理论与方法的角度出发,对可调节范数支持向量机模型进行研究。主要包括:(1)改进、完善现有的优化模型;(2)针对机器学习领域中常见的分类、回归、多示例、多标签等问题,构建相应的可调节范数的优化模型,并基于最优化方法研究建立简单而有效的算法(包括全局算法);(3)针对生物信息领域中的特征选择问题,研究新模型与算法的应用和改进. 本项目的实施不仅能为数据挖掘提出新理论和新方法,而且能够推动支持向量机和最优化方法的发展,具有重要的科学意义和实用价值。
中文关键词: 支持向量机;特征选择;lp-范数;最优化模型;算法
英文摘要: Support vector machines (SVMs) is a powerful tool for machine learning, and has been applied to a wide variety of fields in data mining. SVMs are based on optimization methods and in the standard SVMs, 2-norm is used. Recently, some variants of SVMs with different norms open a novel approach in the development of SVMs because they made a good performance, particularly in feature seletion. This project focuses on the SVMs with adaptive norms, via optimization theory and methods. It mainly includes the following parts: (1) improve the variants of SVMs with different norms; (2)construct new SVMs with adpative norms for classification, regression, multi-instance learning, multi-label learning problems respectively, and propose some efficient optimization methods (including global methods); (3)apply and improve our new SVMs and new optimization methods to gene selection problems in bioinformatics. We believe that this research has scientific significance and practical value since it can not only provide some new theories and methods for data mining, but also improve SVMs and numerical optimization.
英文关键词: Support Vector Machine;Feature Selection;lp-Norm;Optimization Model;Algorithm