1: \begin{abstract}

2: Alternating minimization is a widely used and empirically successful heuristic

3: for matrix completion and related low-rank optimization problems. Theoretical

4: guarantees for alternating minimization have been hard to come by and are

5: still poorly understood. This is in part because the heuristic is iterative

6: and non-convex in nature. We give a new algorithm based on alternating

7: minimization that provably recovers an unknown low-rank matrix from a random

8: subsample of its entries under a standard incoherence assumption.  Our results

9: reduce the sample size requirements of the alternating minimization approach

10: by at least a quartic factor in the rank and the condition number of the

11: unknown matrix. These improvements apply even if the matrix is only close to

12: low-rank in the Frobenius norm. Our algorithm runs in nearly linear time in

13: the dimension of the matrix and, in a broad range of parameters, gives the

14: strongest sample bounds among all subquadratic time algorithms that we are aware

15: of. 

16: 

17: Underlying our work is a new robust convergence analysis of the well-known

18: Power Method for computing the dominant singular vectors of a matrix.  This

19: viewpoint leads to a conceptually simple understanding of alternating

20: minimization. In addition, we contribute a new technique for controlling the

21: coherence of intermediate solutions arising in iterative algorithms based on a

22: smoothed analysis of the QR factorization. These techniques may be of interest

23: beyond their application here.

24: \end{abstract}

25: