Optimally Tuned Iterative Reconstruction Algorithms for Compressed Sensing

被引:184
作者
Maleki, Arian [1 ]
Donoho, David L. [2 ]
机构
[1] Stanford Univ, Dept Elect Engn, Stanford, CA 94305 USA
[2] Stanford Univ, Dept Stat, Stanford, CA 94305 USA
关键词
Compressed sensing; iterative algorithms; phase transition; random matrices; thresholding; MORPHOLOGICAL COMPONENT ANALYSIS; SIGNAL RECOVERY; SPARSE; REPRESENTATIONS; NEIGHBORLINESS; DECOMPOSITION; PROJECTION; POLYTOPES; PURSUIT;
D O I
10.1109/JSTSP.2009.2039176
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
We conducted an extensive computational experiment, lasting multiple CPU-years, to optimally select parameters for two important classes of algorithms for finding sparse solutions of underdetermined systems of linear equations. We make the optimally tuned implementations available at sparselab.stanford.edu; they run "out of the box" with no user tuning: it is not necessary to select thresholds or know the likely degree of sparsity. Our class of algorithms includes iterative hard and soft thresholding with or without relaxation, as well as CoSaMP, subspace pursuit and some natural extensions. As a result, our optimally tuned algorithms dominate such proposals. Our notion of optimality is defined in terms of phase transitions, i.e., we maximize the number of nonzeros at which the algorithm can successfully operate. We show that the phase transition is a well-defined quantity with our suite of random underdetermined linear systems. Our tuning gives the highest transition possible within each class of algorithms. We verify by extensive computation the robustness of our recommendations to the amplitude distribution of the nonzero coefficients as well as the matrix ensemble defining the underdetermined system. Our findings include the following. 1) For all algorithms, the worst amplitude distribution for nonzeros is generally the constant-amplitude random-sign distribution, where all nonzeros are the same amplitude. 2) Various random matrix ensembles give the same phase transitions; random partial isometries may give different transitions and require different tuning. 3) Optimally tuned subspace pursuit dominates optimally tuned CoSaMP, particularly so when the system is almost square.
引用
收藏
页码:330 / 341
页数:12
相关论文
共 50 条
[11]   Atomic decomposition by basis pursuit [J].
Chen, SSB ;
Donoho, DL ;
Saunders, MA .
SIAM JOURNAL ON SCIENTIFIC COMPUTING, 1998, 20 (01) :33-61
[12]   Signal recovery by proximal forward-backward splitting [J].
Combettes, PL ;
Wajs, VR .
MULTISCALE MODELING & SIMULATION, 2005, 4 (04) :1168-1200
[13]   Subspace Pursuit for Compressive Sensing Signal Reconstruction [J].
Dai, Wei ;
Milenkovic, Olgica .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2009, 55 (05) :2230-2249
[14]   An iterative thresholding algorithm for linear inverse problems with a sparsity constraint [J].
Daubechies, I ;
Defrise, M ;
De Mol, C .
COMMUNICATIONS ON PURE AND APPLIED MATHEMATICS, 2004, 57 (11) :1413-1457
[15]  
Donoho D., 2006, SPARSE SOLUTION UNDE
[16]  
Donoho D. L., IEEE T INF THE UNPUB
[17]   Observed universality of phase transitions in high-dimensional geometry, with implications for modern data analysis and signal processing [J].
Donoho, David ;
Tanner, Jared .
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES, 2009, 367 (1906) :4273-4293
[18]   Fast Solution of l1-Norm Minimization Problems When the Solution May Be Sparse [J].
Donoho, David L. ;
Tsaig, Yaakov .
IEEE TRANSACTIONS ON INFORMATION THEORY, 2008, 54 (11) :4789-4812
[19]  
Donoho DL, 2009, J AM MATH SOC, V22, P1
[20]   Reproducible Research in Computational Harmonic Analysis [J].
Donoho, David L. ;
Maleki, Arian ;
Shahram, Morteza ;
Rahman, Inam Ur ;
Stodden, Victoria .
COMPUTING IN SCIENCE & ENGINEERING, 2009, 11 (01) :8-18