A NSGA-II with alternating direction method of multipliers mutation for solving multiobjective robust principal component analysis problem
文献类型：期刊论文
| 作者 | Yuan, Weitao; Liang XD(梁晓丹); Chen HN(陈瀚宁); Lin N(蔺娜); Zou T(邹涛)
|
| 刊名 | Journal of Computational and Theoretical Nanoscience
 |
| 出版日期 | 2016 |
| 卷号 | 13期号:6页码:3722-3733 |
| 关键词 | EVOLUTIONARY ALGORITHM MULTIOBJECTIVE OPTIMIZATION MUTATION ROBUST PRINCIPAL COMPONENT ANALYSIS |
| ISSN号 | 1546-1955 |
| 产权排序 | 3 |
| 通讯作者 | 陈瀚宁 |
| 中文摘要 | Robust Principal Component Analysis (RPCA), which is a popular parsimony model, is becoming increasingly important for researchers to do data analysis and prediction. The RPCA formulation is made of two components: sparse penalty and low rank penalty. These two competing terms are balanced with one parameter, which is essential for the effectiveness of RPCA. However, in real-world applications, the lack of data adaptive methods for choosing the right parameter hinders the popularization of RPCA. In this work, RPCA is generalized to a multiobjective optimization problem without any balancing parameter. The new model is named as Multiobjective Robust Principal Component Analysis (MRPCA). We aim to solve MRPCA via Evolutionary Algorithm. To the best knowledge of authors, this is the first attempt to use evolutionary algorithm to solve RPCA problem, which is a high dimensional convex optimization problem. Specifically, one of the popular evolutionary algorithm, NSGA-II, is tested on MRPCA problem. The curse of dimensionality is observed when the dimension of MRPCA problem increases. To handle this dimensionality problem, we introduce a novel mutation, termed as Alternating Direction Method of Multipliers mutation (ADMM mutation), that works well in high dimensional decision space. Numerical experiments show that this modified NSGA-II, which converges much faster than the standard one, can deal with the curse of dimensionality well. Furthermore, numerical image reconstruction test confirms that the reconstruction performance of our modified NSGA-II is better than the traditional proximal algorithm, which is usually used to solve RPCA problem. |
| 收录类别 | EI |
| 语种 | 英语 |
| 源URL | [http://ir.sia.cn/handle/173321/19927]  |
| 专题 | 沈阳自动化研究所_信息服务与智能控制技术研究室 |
| 推荐引用方式 GB/T 7714 | Yuan, Weitao,Liang XD,Chen HN,et al. A NSGA-II with alternating direction method of multipliers mutation for solving multiobjective robust principal component analysis problem[J]. Journal of Computational and Theoretical Nanoscience,2016,13(6):3722-3733. |
| APA | Yuan, Weitao,Liang XD,Chen HN,Lin N,&Zou T.(2016).A NSGA-II with alternating direction method of multipliers mutation for solving multiobjective robust principal component analysis problem.Journal of Computational and Theoretical Nanoscience,13(6),3722-3733. |
| MLA | Yuan, Weitao,et al."A NSGA-II with alternating direction method of multipliers mutation for solving multiobjective robust principal component analysis problem".Journal of Computational and Theoretical Nanoscience 13.6(2016):3722-3733. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。