Online Nonnegative Matrix Factorization With Robust Stochastic Approximation

被引:240
作者
Guan, Naiyang [1 ]
Tao, Dacheng [2 ,3 ]
Luo, Zhigang [1 ]
Yuan, Bo [4 ]
机构
[1] Natl Univ Def Technol, Sch Comp Sci, Changsha 410073, Hunan, Peoples R China
[2] Univ Technol Sydney, Ctr Quantum Computat & Intelligent Syst, Sydney, NSW 2007, Australia
[3] Univ Technol Sydney, Fac Engn & Informat Technol, Sydney, NSW 2007, Australia
[4] Shanghai Jiao Tong Univ, Dept Comp Sci & Engn, Shanghai 200240, Peoples R China
基金
澳大利亚研究理事会; 中国国家自然科学基金;
关键词
Nonnegative matrix factorization (NMF); online NMF (ONMF); robust stochastic approximation; ALGORITHMS; IMAGE;
D O I
10.1109/TNNLS.2012.2197827
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Nonnegative matrix factorization (NMF) has become a popular dimension-reduction method and has been widely applied to image processing and pattern recognition problems. However, conventional NMF learning methods require the entire dataset to reside in the memory and thus cannot be applied to large-scale or streaming datasets. In this paper, we propose an efficient online RSA-NMF algorithm (OR-NMF) that learns NMF in an incremental fashion and thus solves this problem. In particular, OR-NMF receives one sample or a chunk of samples per step and updates the bases via robust stochastic approximation. Benefitting from the smartly chosen learning rate and averaging technique, OR-NMF converges at the rate of O(1/root k) in each update of the bases. Furthermore, we prove that OR-NMF almost surely converges to a local optimal solution by using the quasi-martingale. By using a buffering strategy, we keep both the time and space complexities of one step of the OR-NMF constant and make OR-NMF suitable for large-scale or streaming datasets. Preliminary experimental results on real-world datasets show that OR-NMF outperforms the existing online NMF (ONMF) algorithms in terms of efficiency. Experimental results of face recognition and image annotation on public datasets confirm the effectiveness of OR-NMF compared with the existing ONMF algorithms.
引用
收藏
页码:1087 / 1099
页数:13
相关论文
共 47 条
[1]  
[Anonymous], 1999, Athena scientific Belmont
[2]  
[Anonymous], 2008, Advances in Neural Information Processing Systems, DOI DOI 10.7751/mitpress/8996.003.0015
[3]  
[Anonymous], 2011, P 2011 IEEE PES INN, DOI DOI 10.1109/ISGT-ASIA.2011.6167097
[4]   Matching words and pictures [J].
Barnard, K ;
Duygulu, P ;
Forsyth, D ;
de Freitas, N ;
Blei, DM ;
Jordan, MI .
JOURNAL OF MACHINE LEARNING RESEARCH, 2003, 3 (06) :1107-1135
[5]   Algorithms and applications for approximate nonnegative matrix factorization [J].
Berry, Michael W. ;
Browne, Murray ;
Langville, Amy N. ;
Pauca, V. Paul ;
Plemmons, Robert J. .
COMPUTATIONAL STATISTICS & DATA ANALYSIS, 2007, 52 (01) :155-173
[6]   Incremental subspace learning via non-negative matrix factorization [J].
Bucak, Serhat S. ;
Gunsel, Bilge .
PATTERN RECOGNITION, 2009, 42 (05) :788-797
[7]   Graph Regularized Nonnegative Matrix Factorization for Data Representation [J].
Cai, Deng ;
He, Xiaofei ;
Han, Jiawei ;
Huang, Thomas S. .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2011, 33 (08) :1548-1560
[8]  
Cao B, 2007, 20TH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, P2689
[9]  
Duchi J., 2008, P 25 INT C MACH LEAR
[10]  
Duygulu P, 2002, LECT NOTES COMPUT SC, V2353, P97