An alternating direction method for linear‐constrained matrix nuclear norm minimization

Abstract

SUMMARY The aim of the nuclear norm minimization problem is to find a matrix that minimizes the sum of its singular values and satisfies some constraints simultaneously. Such a problem has received more attention largely because it is closely related to the affine rank minimization problem, which appears in many control applications including controller design, realization theory, and model reduction. In this paper, we first propose an exact version alternating direction method for solving the nuclear norm minimization problem with linear equality constraints. At each iteration, the method involves a singular value thresholding and linear matrix equations which are solved exactly. Convergence of the proposed algorithm is followed directly. To broaden the capacity of solving larger problems, we solve approximately the subproblem by an iterative method with the Barzilai–Borwein steplength. Some extensions to the noisy problems and nuclear norm regularized least‐square problems are also discussed. Numerical experiments and comparisons with the state‐of‐the‐art method FPCA show that the proposed method is effective and promising. Copyright © 2011 John Wiley & Sons, Ltd.

A simple and feasible method for a class of large-scalel1-problems

“…In the past decades, the ADM was developed [25][26][27][28][29] and utilized widely in signal processing and compressive sensing [30][31][32][33]. In fact, the ADM is closely related to Douglas-Rachford splitting methods [34], and split Bregman methods [35] in image processing.…”

A Proximal Fully Parallel Splitting Method for Stable Principal Component Pursuit

“…Problems (3) and (5) are both convex optimization programming, which can be equivalently reformulated as some semidefinite programming (SDP) problems and then solved by using some state-ofthe-art SDP solvers in polynomial time, such as SeDuMi [7] and SDPT3 [8]. However, these solvers are not applicable to high-dimensional (3) and (5), because a linear system needs to be solved (in)exactly at each iteration, which may be more costly and require very large amount of memory in actual implementations [9].…”

Splitting and linearizing augmented Lagrangian algorithm for subspace recovery from corrupted observations

“…Because the resulting subproblems are always sufficiently simple to have closed-form solutions, ADM recently has been exhibited as a powerful algorithmic tool to solve convex programming problems arising from various applications such as in image processing [9,12,31,35,36], compressive sensing [38], matrix completion [5,34,37], SDP [17,27,32], and multi-task feather learning [6]. In this paper, we focus on the application of ADM to solve the nuclear norm and 2,1 -mixed norm involved minimization model (1.2), and to demonstrate its remarkable effectiveness in recovering subspace structure and correcting noise as well for a given corrupted data.…”