1: \begin{abstract}

2: Minimizing the rank of a matrix subject to affine constraints is a

3: fundamental problem with many important applications in machine

4: learning and statistics. In this paper we propose a simple and fast

5: algorithm $\mathsf{SVP}$ (Singular Value Projection) for rank

6: minimization with affine constraints ($\mathsf{ARMP}$) and show that SVP recovers the minimum rank solution for affine constraints that satisfy the {\sl restricted isometry property}. We show robustness of our method to noise with a strong geometric convergence rate even for noisy measurements. Our results improve upon a recent breakthrough by Recht, Fazel and Parillo \cite{RechtFP2007} and Lee and Bresler \cite{LeeB2009} in three significant ways: 1) our method ($\mathsf{SVP}$) is significantly simpler to analyze and easier to implement, 2) we give recovery guarantees under strictly weaker isometry assumptions 3) we give geometric convergence guarantees for  $\mathsf{SVP}$ and, as demonstrated empirically, $\mathsf{SVP}$ is significantly faster on real-world and synthetic problems. In addition, we address the practically important problem of low-rank matrix completion, which can be seen as a special case of $\mathsf{ARMP}$. However, the affine constraints defining the matrix-completion problem do not obey the {\sl restricted isometry property} in general. We empirically demonstrate that our algorithm recovers low-rank {\sl incoherent} matrices from an almost optimal number of uniformly sampled entries. We make partial progress towards proving exact recovery and provide some intuition for the performance of $\mathsf{SVP}$ applied to matrix completion by showing a more restricted isometry property. Our algorithm outperforms existing methods, such as those of \cite{RechtFP2007,CandesR2008,CandesT2009,CaiCS2008,KeshavanOM2009}, for $\mathsf{ARMP}$ and the matrix-completion problem by an order of magnitude and is also significantly more robust to noise.

7: \end{abstract}

8: