1: \begin{abstract}

2:     \justifying

3:     The existing matrix completion methods focus on optimizing the relaxation of rank function 

4:     such as nuclear norm, Schatten-$p$ norm, \textit{etc}. 

5:     They usually need many iterations to converge. 

6:     Moreover, only the low-rank property of matrices is utilized in most 

7:     existing models and several methods that incorporate other knowledge 

8:     are quite time-consuming in practice. 

9:     To address these issues, we propose a novel non-convex surrogate that 

10:     can be optimized by closed-form solutions, such that it empirically 

11:     converges within dozens of iterations. 

12:     Besides, the optimization is parameter-free and the convergence is proved. 

13:     Compared with the relaxation of rank, 

14:     the surrogate is motivated by optimizing an upper-bound of rank. 

15:     We theoretically validate that it is equivalent to the existing 

16:     matrix completion models. 

17:     Besides the low-rank assumption, we intend to exploit the column-wise 

18:     correlation for matrix completion, 

19:     and thus an adaptive correlation learning, which is scaling-invariant, 

20:     is developed.

21:     More importantly, after incorporating the correlation learning, 

22:     the model can be still solved by closed-form solutions such that 

23:     it still converges fast. 

24:     Experiments show the effectiveness of the non-convex surrogate and 

25:     adaptive correlation learning. 

26: \end{abstract}

27: