Splitting and linearizing augmented Lagrangian algorithm for subspace recovery from corrupted observations (Q1955536)

For a given set of corrupted data samples approximately drawn from a union of multiple independent subspaces, the task of recovering low-rank components and to correct possible noise simultaneously can be accomplished by solving a convex minimization problem. The authors propose a primal splitting and linearizing augmented Lagrangian alternative direction algorithm to solve the minimization problem. The task of minimizing the corresponding augmented Lagrangian function is split into two subproblems. At each iteration, the resulting subproblems with respect to each variable is determined explicitly by using shrinkage operators. Two matrix multiplications are involved at each step. Global convergence of the algorithm is established under certain assumptions. Some numerical results on some simulated data and a real data set are provided to illustrate the effectively of the algorithm.