A Clustering Approach to Learning Sparsely Used Overcomplete Dictionaries

被引：17
作者：
Agarwal, Alekh [1 ]
Anandkumar, Animashree [2 ]
Netrapalli, Praneeth [3 ]
机构：
[1] Microsoft Res, New York, NY 10011 USA
[2] Univ Calif Irvine, Ctr Pervas Commun & Comp, Elect Engn & Comp Sci Dept, Irvine, CA 92697 USA
[3] Microsoft Res, Bengaluru 10011, India
来源：
IEEE TRANSACTIONS ON INFORMATION THEORY | 2017年 / 63卷 / 01期
关键词：
Dictionary learning; sparse coding; overcomplete dictionaries; incoherence; lasso; SAMPLE COMPLEXITY;
D O I：
10.1109/TIT.2016.2614684
中图分类号：
TP [自动化技术、计算机技术];
学科分类号：
0812 ;
摘要：
We consider the problem of learning overcomplete dictionaries in the context of sparse coding, where each sample selects a sparse subset of dictionary elements. Our main result is a strategy to approximately recover the unknown dictionary using an efficient algorithm. Our algorithm is a clustering-style procedure, where each cluster is used to estimate a dictionary element. The resulting solution can often be further cleaned up to obtain a high accuracy estimate, and we provide one simple scenario where l(1)-regularized regression can be used for such a second stage.
引用
收藏
页码：575 / 592
页数：18
相关论文
共 55 条
[11]  
[Anonymous], 2006, ADV NEURAL INF PROCE
[12]  
[Anonymous], RAMSEY THEORY REVEAL
[13]  
[Anonymous], 2012, JMLR WORKSHOP C P
[14]  
[Anonymous], 2013, Proc. 30th Int. Conf. on Machine Learning
[15]  
[Anonymous], 2012, COMPRESSED SENSING T
[16]  
[Anonymous], 2010, Proceedings of the 27th International Conference on International Conference on Machine Learning
[17]  
[Anonymous], 2014, P ADV NEUR INF PROC
[18]  
[Anonymous], MORE ALGORITHMS PROV
[19]  
[Anonymous], 2015, PMLR
[20]  
Arabshahi F., 2016, LDA UNIFIED FRAMEWOR
123456