摘要
当前主流分类方法在分类决策时无法同时考虑样本的全局特征和局部特征,而且大多算法仅关注各类样本的可分性,往往忽略样本之间的相对关系。为了解决上述问题,提出了基于流形判别分析的全局保序学习机。该方法引入流形判别分析来反映样本的全局特征和局部特征;通过保持各类样本中心的相对关系不变进而实现保持全体样本的先后顺序不变;借鉴核心向量机有关理论和方法,通过建立所提方法与核心向量机对偶形式的等价关系实现大规模分类。人工数据集和标准数据集上的比较实验验证了该方法的有效性。
In order to solve the problems that many traditional classification methods confronted, a global rank preservation learning machine （GRPLM） based on manifold-based diseriminant analysis is proposed in this paper. In GRPLM, the manifold-based discriminant analysis （MDA） is introduced to represent the samples＇ global and local characteristic; the relative relationship of different class centers is taken into consideration in order to preserve the samples＇ ranks; the equivalent relation between the QP form of GRPLM and core vector machine （CVM） is analyzed in order to broaden the usage of GRPLM from small- and medium-scale to large-scale. Comparative experiments on several standard datasets verify the effectiveness of the proposed methods.
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2015年第6期911-916,共6页
Journal of University of Electronic Science and Technology of China
基金
国家自然科学基金(61202311)
山西省高等学校科技创新项目(2014142)
关键词
全局保序
大规模分类
流形判别分析
支持向量机
global rank preservation
large-scale classification
manifold-based discriminant analysis （MDA）
support vector machine （SVM）
