摘要
为了得到核Bayes函数作为分类器的核主元分析(KPCA)与核Fisher判据分析(KFDA)的算法模式稳定性,利用Rademacher复杂度的概念及相关定理,推导出了核Bayes函数Rademacher复杂度的界以及其作为分类器的算法发生错误分类的概率的界,说明了模式稳定性与样本长度、降维矩阵的维数等关系,提出了两种衡量模式稳定性的直观指标,误分差和百分比和误分均值偏离度.仿真结果不仅验证了几个定理,也表明所提出的衡量指标是有效的、可行的.
It is obtained pattern stability of KPCA and KFDA algorithm which use kernel Bayes function as classifier.Firstly,the concept of Rademacher complexity and several theorems are introduced.Then,the upper bounds of Rademacher complexity of kernel Bayes function is deduced,and the upper bound of mis-classification probability which uses kernel Bayes function as classifier is obtained.Also,two indexes which measure pattern stability are proposed.One is ratio of mis-classification differences summation,the other is deviation ratio of mis-classification mean.Simulation results not only verify the deduced theorem,but also indicate that the proposed measure indexes are effective and practicable.
出处
《湖南师范大学自然科学学报》
CAS
北大核心
2010年第3期16-21,共6页
Journal of Natural Science of Hunan Normal University
基金
国家自然科学基金资助项目(60634030
60702066)
高校博士点专项基金资助项目(20060699032)