Matrix Completion With Cross-Concentrated Sampling: Bridging Uniform Sampling and CUR Sampling

被引:9
作者
Cai, HanQin [1 ,2 ]
Huang, Longxiu [3 ,4 ]
Li, Pengyu [5 ]
Needell, Deanna [6 ]
机构
[1] Univ Cent Florida, Dept Stat & Data Sci, Orlando, FL 32816 USA
[2] Univ Cent Florida, Dept Comp Sci, Orlando, FL 32816 USA
[3] Michigan State Univ, Dept Computat Math Sci & Engn, E Lansing, MI 48823 USA
[4] Michigan State Univ, Dept Math, E Lansing, MI 48823 USA
[5] Univ Michigan, Dept Elect Engn & Comp Sci, Ann Arbor, MI 48109 USA
[6] Univ Calif Los Angeles, Dept Math, Los Angeles, CA 90095 USA
关键词
Matrix decomposition; Motion pictures; Collaboration; Minimization; Data models; Complexity theory; Bridges; Cross-concentrated sampling; CUR decomposition; collaborative filtering; image recovery; low-rank matrix; link prediction; matrix completion; recommendation system; sampling strategy; LOW-RANK MATRIX; APPROXIMATION; DECOMPOSITION; ALGORITHMS;
D O I
10.1109/TPAMI.2023.3261185
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
While uniform sampling has been widely studied in the matrix completion literature, CUR sampling approximates a low-rank matrix via row and column samples. Unfortunately, both sampling models lack flexibility for various circumstances in real-world applications. In this work, we propose a novel and easy-to-implement sampling strategy, coined Cross-Concentrated Sampling (CCS). By bridging uniform sampling and CUR sampling, CCS provides extra flexibility that can potentially save sampling costs in applications. In addition, we also provide a sufficient condition for CCS-based matrix completion. Moreover, we propose a highly efficient non-convex algorithm, termed Iterative CUR Completion (ICURC), for the proposed CCS model. Numerical experiments verify the empirical advantages of CCS and ICURC against uniform sampling and its baseline algorithms, on both synthetic and real-world datasets.
引用
收藏
页码:10100 / 10113
页数:14
相关论文
共 60 条
[51]  
Tong T, 2021, J MACH LEARN RES, V22
[52]  
Tropp JA, 2009, PROCEEDINGS OF THE TWENTIETH ANNUAL ACM-SIAM SYMPOSIUM ON DISCRETE ALGORITHMS, P978
[53]   Semidefinite programming [J].
Vandenberghe, L ;
Boyd, S .
SIAM REVIEW, 1996, 38 (01) :49-95
[54]   LOW-RANK MATRIX COMPLETION BY RIEMANNIAN OPTIMIZATION [J].
Vandereycken, Bart .
SIAM JOURNAL ON OPTIMIZATION, 2013, 23 (02) :1214-1236
[55]  
Wang SS, 2013, J MACH LEARN RES, V14, P2729
[56]   GUARANTEES OF RIEMANNIAN OPTIMIZATION FOR LOW RANK MATRIX COMPLETION [J].
Wei, Ke ;
Cai, Jian-Feng ;
Chan, Tony F. ;
Leung, Shingyu .
INVERSE PROBLEMS AND IMAGING, 2020, 14 (02) :233-265
[57]  
Xu M, 2015, PR MACH LEARN RES, V37, P1412
[58]  
Yildirim H, 2008, RECSYS'08: PROCEEDINGS OF THE 2008 ACM CONFERENCE ON RECOMMENDER SYSTEMS, P131
[59]   A novel link prediction algorithm based on inductive matrix completion [J].
Zhao, Zhili ;
Gou, Zhuoyue ;
Du, Yuhong ;
Ma, Jun ;
Li, Tongfeng ;
Zhang, Ruisheng .
EXPERT SYSTEMS WITH APPLICATIONS, 2022, 188
[60]  
Zheng QQ, 2016, Arxiv, DOI arXiv:1605.07051