1: \begin{abstract}

2: Accurate signal recovery or image reconstruction from indirect and

3: possibly undersampled data is a topic of considerable interest; 

4: for example, the literature in the recent field of compressed

5: sensing is already quite immense.

6: Inspired by recent breakthroughs in the development of novel

7: first-order methods in convex optimization, most notably Nesterov's

8: smoothing technique, this paper introduces a fast and accurate

9: algorithm for solving common recovery problems in signal

10: processing. In the spirit of Nesterov's work,  one of the key

11: ideas of this algorithm is

12: a subtle averaging of sequences of

13: iterates, which has been shown to improve the convergence properties

14: of standard gradient-descent algorithms. This paper demonstrates

15: that this approach is ideally suited for solving large-scale

16: compressed sensing reconstruction problems as 1) it is

17: computationally efficient, 2) it is accurate and returns solutions

18: with several correct digits, 3) it is flexible and amenable to many

19: kinds of reconstruction problems, and 4) it is robust in the sense

20: that its excellent performance across a wide range of problems does

21: not depend on the fine tuning of several parameters.  Comprehensive

22: numerical experiments on realistic signals exhibiting a large

23: dynamic range show that this algorithm compares favorably with

24: recently proposed state-of-the-art methods. We also apply the

25: algorithm to solve other problems for which there are fewer

26: alternatives, such as total-variation minimization, and convex

27: programs seeking to minimize the $\ell_1$ norm of $Wx$ under

28: constraints, in which $W$ is not diagonal.

29: \end{abstract}

30: