摘要
线性判别回归分类算法没有考虑到类内距离和类间距离,为此提出一种基于线性判别回归与局部判别分析的维数约简算法(LDRFDR),同时利用类内和类间重构误差、以及类内和局部类间距离获得投影矩阵。其物理含义是为各个类尽量寻找相互之间离得最远的线性子空间,其中类内距离与类间距离还考虑数据的局部性,避免最大化相离太远的类间样本对优化目标造成影响。实验结果表明,LDRFDR算法的维数约简性能优于其它半监督维数约减算法。
To overcome the problem that linear discriminant regression classification （LDRC） does not consider the within class distances and across classes distances, a dimensionality reduction method named linear discriminant regression and local discrimi-nant analysis based dimensionality reduction （LDRFDR） was proposed. The within-class reconstruction error, across classes re-construction error, within class distances, and across classes distances were utilized at the same time. The physical meaning of LDRFDR was to find the farthest linear space among all classes. In LDRFDR, the locality was also considered by within class distances and across classes distances, which avoided the impact of the distance between two far-away-samples. Experimental re-sults demonstrate that LDRFDR is better than other semi-supervised dimensionality reduction algorithms.
作者
林平荣
孙亚新
文贵华
LIN Ping-rong SUN Ya-xin WEN Gui-hua(Software Engineering Department, South China Institute of Software Engineering, Guangzhou University, Guangzhou 510990, China School of Computer Science and Engineering,South China University of Technology, Guangzhou 510006, China)
出处
《计算机工程与设计》
北大核心
2017年第5期1371-1376,共6页
Computer Engineering and Design
基金
2015年度广东高校省级重点平台和重大科研基金项目(2015KQNCX201)
关键词
线性判别回归
局部判别分析
局部重构
维数约减
线性子空间
linear discriminant regression
local discriminant analysis
local reconstruction
dimensionality reduction
linear sub-space
