摘要
为解决高维稀疏建模问题,本文从经验风险最小化原则出发推导出一个基于零范数约束的特征选择判据,并利用嵌入式设计模式的特点将其与支持向量机方法相结合.仿真实验和真实数据实验表明,该方法不仅具备良好的特征选择性能,而且在稀疏建模问题中表现出良好的分类准确性和泛化能力.
To deal with the high-dimensional sparse modeling problem, this paper derives a zero-norm penalized feature selection criterion based on the the empirical risk minimization principle, and combines it with support vector machines through an embedded paradigm. Numerical results on both synthetic and real data sets show that the proposed approach does not only perform well for feature selection tasks, but also shows good performance compared to the conventional sparse modeling techniques, in the sense of classification accuracy and generalization capability.
出处
《自动化学报》
EI
CSCD
北大核心
2011年第2期252-256,共5页
Acta Automatica Sinica
基金
国家高技术研究发展计划(863计划)(2006AA01Z411)资助~~
关键词
机器学习
特征选择
支持向量机
稀疏建模
正则化
Machine learning, feature selection, support vector machine （SVM）, sparse modeling, regularization
