1: \begin{abstract}

2: This paper deals with the trace regression model where $n$ entries

3: or linear combinations of entries of an unknown $m_1\times m_2$

4: matrix $A_0$ corrupted by noise are observed. We propose a new

5: nuclear norm penalized estimator of $A_0$ and establish a

6: general sharp oracle inequality for this estimator

7: for arbitrary values of $n,m_1,m_2$

8: under the condition of isometry in expectation. Then this

9: method is applied to the matrix completion problem. In this case, the estimator

10: admits a simple explicit form and we prove that it satisfies oracle

11: inequalities with faster rates of convergence than in the previous

12: works. They are valid, in particular, in the

13: high-dimensional setting $m_1m_2\gg n$. We show that the obtained

14: rates are optimal up to logarithmic factors in a minimax sense and

15: also derive, for any fixed matrix $A_0$, a non-minimax lower bound

16: on the rate of convergence  of our estimator, which coincides with

17: the upper bound up to a constant factor. Finally, we show that our

18: procedure provides an exact recovery of the rank of $A_0$ with

19: probability close to 1. We also discuss the statistical learning

20: setting where there is no underlying model determined by $A_0$ and

21: the aim is to find the best trace regression model approximating the

22: data.

23: \end{abstract}

24: