Inexact Coordinate Descent: Complexity and Preconditioning

被引:29
作者
Tappenden, Rachael [1 ]
Richtarik, Peter [1 ]
Gondzio, Jacek [1 ]
机构
[1] Univ Edinburgh, Sch Math, Edinburgh EH9 3JZ, Midlothian, Scotland
基金
英国工程与自然科学研究理事会;
关键词
Inexact methods; Block coordinate descent; Convex optimization; Iteration complexity; Preconditioning; Conjugate gradients; GRADIENT PROJECTION; CONVERGENCE; ALGORITHM;
D O I
10.1007/s10957-016-0867-4
中图分类号
C93 [管理学]; O22 [运筹学];
学科分类号
070105 ; 12 ; 1201 ; 1202 ; 120202 ;
摘要
One of the key steps at each iteration of a randomized block coordinate descent method consists in determining the update to a block of variables. Existing algorithms assume that in order to compute the update, a particular subproblem is solved exactly. In this work, we relax this requirement and allow for the subproblem to be solved inexactly, leading to an inexact block coordinate descent method. Our approach incorporates the best known results for exact updates as a special case. Moreover, these theoretical guarantees are complemented by practical considerations: the use of iterative techniques to determine the update and the use of preconditioning for further acceleration.
引用
收藏
页码:144 / 176
页数:33
相关论文
共 43 条
[1]  
[Anonymous], 2014, Introductory Lectures on Convex Optimization: A Basic Course
[2]  
[Anonymous], 2011, Advances in neural information processing systems
[3]   The self regulation problem as an inexact steepest descent method for multicriteria optimization [J].
Bento, G. C. ;
Cruz Neto, J. X. ;
Oliveira, P. R. ;
Soubeyran, A. .
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, 2014, 235 (03) :494-502
[4]  
Benzi M, 2005, ACTA NUMER, V14, P1, DOI 10.1017/S0962492904000212
[5]   Inexact block coordinate descent methods with application to non-negative matrix factorization [J].
Bonettini, Silvia .
IMA JOURNAL OF NUMERICAL ANALYSIS, 2011, 31 (04) :1431-1452
[6]   A box constrained gradient projection algorithm for compressed sensing [J].
Broughton, R. L. ;
Coope, I. D. ;
Renaud, P. F. ;
Tappenden, R. E. H. .
SIGNAL PROCESSING, 2011, 91 (08) :1985-1992
[7]   Exact Matrix Completion via Convex Optimization [J].
Candes, Emmanuel J. ;
Recht, Benjamin .
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS, 2009, 9 (06) :717-772
[8]   On the convergence of inexact block coordinate descent methods for constrained optimization [J].
Cassioli, A. ;
Di Lorenzo, D. ;
Sciandrone, M. .
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, 2013, 231 (02) :274-281
[9]   Quadratic regularizations in an interior-point method for primal block-angular problems [J].
Castro, Jordi ;
Cuesta, Jordi .
MATHEMATICAL PROGRAMMING, 2011, 130 (02) :415-445
[10]   Algorithm 915, SuiteSparseQR: Multifrontal Multithreaded Rank-Revealing Sparse QR Factorization [J].
Davis, Timothy A. .
ACM TRANSACTIONS ON MATHEMATICAL SOFTWARE, 2011, 38 (01)