Generalized alternating direction method of multipliers: new theoretical insights and applications

被引：78
作者：
Fang E.X. [1 ]
He B. [2 ]
Liu H. [1 ]
Yuan X. [3 ]
机构：
[1] Department of Operations Research and Financial Engineering, Princeton University, Princeton, 08544, NJ
[2] International Centre of Management Science and Engineering, and Department of Mathematics, Nanjing University, Nanjing
[3] Department of Mathematics, Hong Kong Baptist University, Kowloon
来源：
Mathematical Programming Computation | 2015年 / 7卷 / 2期
基金：
中国国家自然科学基金;
关键词：
Alternating direction method of multipliers; Convergence rate; Convex optimization; Discriminant analysis; Statistical learning; Variable selection;
D O I：
10.1007/s12532-015-0078-2
中图分类号：
学科分类号：
摘要：
Recently, the alternating direction method of multipliers (ADMM) has received intensive attention from a broad spectrum of areas. The generalized ADMM (GADMM) proposed by Eckstein and Bertsekas is an efficient and simple acceleration scheme of ADMM. In this paper, we take a deeper look at the linearized version of GADMM where one of its subproblems is approximated by a linearization strategy. This linearized version is particularly efficient for a number of applications arising from different areas. Theoretically, we show the worst-case $${mathcal {O}}(1/k)$$O(1/k) convergence rate measured by the iteration complexity ($$k$$k represents the iteration counter) in both the ergodic and a nonergodic senses for the linearized version of GADMM. Numerically, we demonstrate the efficiency of this linearized version of GADMM by some rather new and core applications in statistical learning. Code packages in Matlab for these applications are also developed. © 2015, Springer-Verlag Berlin Heidelberg and The Mathematical Programming Society.
引用
收藏
页码：149 / 187
页数：38
相关论文
共 51 条
  • [1] Anderson T.W., An introduction to multivariate statistical analysis, 3rd edn, Wiley, (2003)
  • [2] Bertsekas D.P., Constrained optimization and Lagrange multiplier methods, (1982)
  • [3] Bickel P.J., Levina E., Some theory for Fisher’s linear discriminant function, naive Bayes’, and some alternatives when there are many more variables than observations, Bernoulli, 6, pp. 989-1010, (2004)
  • [4] Blum E., Oettli W., Mathematische Optimierung. Grundlagen und Verfahren. Ökonometrie und Unternehmensforschung, (1975)
  • [5] Boley D., Local linear convergence of ADMM on quadratic or linear programs, SIAM J. Optim., 23, 4, pp. 2183-2207, (2013)
  • [6] Boyd S., Parikh N., Chu E., Peleato B., Eckstein J., Distributed optimization and statistical learning via the alternating direction method of multipliers, Found. Trends Mach. Learn., 3, (2011)
  • [7] Cai T.T., Liu W., A Direct estimation approach to sparse linear discriminant analysis, J. Amer. Stat. Assoc., 106, pp. 1566-1577, (2011)
  • [8] Cai X., Gu G., He B., Yuan X., A proximal point algorithm revisit on alternating direction method of multipliers, Sci. China Math., 56, 10, pp. 2179-2186, (2013)
  • [9] Candes E.J., Tao T., The Dantzig selector: statistical estimation when (Formula presented.) is much larger than (Formula presented.), Ann. Stat., 35, pp. 2313-2351, (2007)
  • [10] Clemmensen L., Hastie T., Witten D., Ersboll B., Sparse discriminant analysis, Technometrics, 53, pp. 406-413, (2011)
123456