L1-范数子空间技术的鲁棒建模综述(英)  

L1-norm based subspace techniques for robust modelling:a brief review

作　　者：胡姿岚[1] 王海贤[2] 

机构地区：[1]安徽工业大学数理科学与工程学院,安徽马鞍山243032 [2]东南大学学习科学研究中心,江苏南京210096

出　　处：《安徽大学学报（自然科学版）》2017年第5期9-16,共8页Journal of Anhui University(Natural Science Edition)

基　　金：Foundation item:Supported by the National Natural Science Foundation of China(61375118);the Research Foundation for Young Teachers in Anhui University of Technology(QZ201516);the Key Natural Science Foundation of Anhui Province(KJ2015ZD44)

摘　　要：子空间学习可以通过多种技术来开展,对一些流行且被广泛使用的子空间学习方法,简要回顾其基于L1-范数的鲁棒建模.从主成分分析开始介绍子空间学习技术、线性判别分析以及更一般的图嵌入框架.作为L1-范数的综合利用,进一步讨论具有稀疏性的鲁棒建模.此外,还论述一些应用在神经科学中的相关子空间学习技术.最后,针对基于L1-范数的子空间学习的求解问题,介绍一个有力工具,即边界优化技术.Subspace learning can be performed with plenty of techniques. The paper was to briefly review Ll-norm based robust modelling for the most popular and widely used subspace learning methods. We commenced the subspace learning technique from principal component analysis, and proceeded to linear discriminant analysis and then the more general framework called graph embedding. As a comprehensive utilization of the Ll-norm, robust modelling plus sparsity was further discussed. Besides, some relative subspace learning techniques applied in neuroscience were gone through. Finally, the bound optimization technique, a useful tool for tackling the Ll-norm based subspace learning, was described.

关 键 词：子空间学习 L1-范数 鲁棒建模 稀疏建模 边界优化 脑机接口 

分 类 号：O151[理学—数学]