1: \begin{abstract}

2: We consider the problem of sparse coding, where each sample consists

3: of a sparse linear combination of a set of dictionary atoms, and the task is

4: to learn both the dictionary elements and the mixing

5: coefficients. Alternating minimization is a popular heuristic for

6: sparse coding, where the dictionary and the coefficients are estimated

7: in alternate steps, keeping the other fixed. Typically, the

8: coefficients are estimated via $\ell_1$ minimization, keeping the

9: dictionary fixed, and the dictionary is estimated through least

10: squares, keeping the coefficients fixed.  In this paper, we establish

11: local linear convergence for this variant of alternating minimization

12: and establish that the basin of attraction for the global optimum

13: (corresponding to the true dictionary and the coefficients) is

14: $\order{1/s^2}$, where $s$ is the sparsity level in each sample and

15: the dictionary satisfies RIP. Combined with the recent

16: results of approximate dictionary estimation, this yields provable

17: guarantees for exact recovery of both the dictionary elements and the

18: coefficients, when the dictionary elements are incoherent.

19: \end{abstract}

20: