Penalty Decomposition Methods for Rank Minimization
In this paper we consider general rank minimization problems with rank appearing in either objective function or constraint. We first establish that a class of special rank minimization problems has closed-form solutions. Using this result, we then propose penalty decomposition methods for general rank minimization problems in which each subproblem is solved by a block coordinate descend method. Under some suitable assumptions, we show that any accumulation point of the sequence generated by the penalty decomposition methods satisfies the first-order optimality conditions of a nonlinear reformulation of the problems. Finally, we test the performance of our methods by applying them to the matrix completion and nearest low-rank correlation matrix problems. The computational results demonstrate that our methods are generally comparable or superior to the existing methods in terms of solution quality.
Year of publication: |
2010-08
|
---|---|
Authors: | Lu, Zhaosong ; Zhang, Yong |
Institutions: | arXiv.org |
Saved in:
freely available
Saved in favorites
Similar items by person
-
Description of dynamics of stock prices by a Langevin approach
Huang, Zi-Gang, (2005)
-
Dimension reduction and coefficient estimation in multivariate linear regression
Yuan, Ming, (2007)
-
Optimal Solutions for the Closest-String Problem via Integer Programming
Meneses, Cláudio N., (2004)
- More ...