摘要
结构化矩阵重构是机器学习中的重要问题之一,矩阵补全是目前研究的热点,笔者重点研究了矩阵补全的推广形式,即矩阵是由低秩矩阵、稀疏矩阵以及噪声叠加而成。有学者已经从数值实验证明:基于交替方向加权主成分追踪算法重构效果优于基于交替方向主成分追踪算法,同时加权算法对白噪声更加稳健。笔者将从理论上推出此结论。
This paper addresses the general case of matrix completion in which the matrix can be modeled as a low-rank plus a sparse component and a noisy component.Numerical results showed that both the recovery accuracy and the rank of recovery matrix of the Weighted Algorithm for Principal Component Pursuit of Alternating Direction Algorithm WPCP-ADM algorithm are competitive with PCP-ADM.The WPCP-ADM algorithm is also more stable for while noise than PCP-ADM.In this paper,We provide a performance analysis of WPCP-ADM.
作者
游庆山
You Qingshan(College of Computer Science and Technology,Civil Aviation Flight University of China,Guanghan Sichuan 618307,China)
出处
《信息与电脑》
2020年第11期59-62,共4页
Information & Computer
基金
中央高校教育教学改革专项资金(项目编号:E2020043,E2019051)。
关键词
矩阵补全
匹配追踪
低秩矩阵
加权算法
稀疏
交替投影法
matrix completion
principal component pursuit
low-rank matrices
weighted algorithm
sparsity
alternating direction method
