1: \begin{abstract}

2:   We consider the problem of optimizing a high-dimensional convex function

3:   using stochastic zeroth-order queries.

4:   %Such problems arise naturally in a variety of practical applications, including optimizing experimental or simulation parameters

5:   %with many variables.

6:   Under sparsity assumptions on the gradients or function values, 

7:   we present 

8:   two algorithms: a successive component/feature selection algorithm and a noisy mirror descent algorithm

9:   using Lasso gradient estimates, 

10:   and show that both algorithms have convergence rates that depend only logarithmically

11:   on the ambient  dimension of the problem.

12:   Empirical results confirm our theoretical findings and show that the algorithms we design outperform classical zeroth-order optimization

13:   methods in the high-dimensional setting.

14: \end{abstract}

15: