AbstractMany applications require recovering a ground truth low-rank matrix from noisy observations of the entries. In practice, this is typically formulated as
weighted low-rank approximation problem and solved using non-convex optimization heuristics such as
→