1: \begin{abstract}

2: %Matrix completion aims to recover an unknown low-rank matrix from a small subset of its entries. 

3: %In many applications, the rank of the unknown target matrix is known in advance. 

4: %In this paper, we propose a two-phase algorithm that leverages the rank information to compute both a suitable value for the regularization parameter and a warm-start for an accelerated Soft-Impute algorithm. 

5: %Properties inherited from proximal gradient algorithms are exploited to propose a parameter tuning to accelerate the method and also to establish a convergence analysis. Numerical experiments with both synthetic and real data show that the proposed algorithm can recover low-rank matrices, with high precision, faster than other well-established matrix completion algorithms.

6: 

7: Matrix completion aims to recover an unknown low-rank matrix from a small subset of its entries. 

8: In many applications, the rank of the unknown target matrix is known in advance. 

9: In this paper, first we revisit a recently proposed rank-based heuristic for ``known-rank'' matrix completion %which keeps the iterates with the correct target rank.  

10: and establish a condition under which the generated sequence is quasi-Fejér convergent to the solution set. 

11: Then, by including an acceleration mechanism similar to Nesterov's acceleration, we obtain a new heuristic. 

12: Even though the convergence of such heuristic cannot be granted in general, it turns out that it can be very useful as a warm-start phase,  

13: providing a suitable estimate for the regularization parameter and a good starting-point, to an accelerated Soft-Impute algorithm. 

14: Numerical experiments with both synthetic and real data show that the resulting two-phase rank-based algorithm can recover low-rank matrices, 

15: with relatively high precision, faster than other well-established matrix completion algorithms.

16: 

17: \end{abstract}

18: