摘要
提出基于F isher距离测度的线性分类器符合统计学习理论框架的观点,结合主分量分析和遗传算法提出一种基于结构风险最小化(Structural R isk M in im ization,简称SRM)归纳原则的分类器设计方法.通过对比遗传算法和穷举法的运算量,阐明所提出的特征提取方法在采用F isher线性分类器分类时的优势.最后采用所提出的基于SRM归纳原则的方法对一组人脑慢皮层电位数据进行了分类仿真实验,并将结果与该组数据竞赛优胜者的结果进行了对比,性能得到了明显提高.
A Fisher Linear Discriminator coincident with statistical learning theory is presented. Based on principal component analysis, structural risk minimization (SRM) classifier and the genetic algorithms (GA) are proposed. It shows the advantage of the method for character extraction comparing with between GA and the general enumeration in operation. A simulation experiment is conducted for human brain slow cortical potential with the algorithm based on the SRM principle. The result shows that the new algorithm is prior to the winner of the data set in discriminator.
出处
《大连交通大学学报》
CAS
2010年第1期104-107,共4页
Journal of Dalian Jiaotong University
基金
国家自然科学基金资助项目(30570475
30170259
60172072)
中国博士后基金资助课题(20080441121)
关键词
统计学习理论
遗传算法
线性分类器
结构风险最小化
脑机接口
statistical learning principle
genetic algorithms
linear classifier
structural risk minimization
brain computer interface