1: \begin{abstract}

2: The randomized Kaczmarz ($\RK$) algorithm is a simple but powerful

3: approach for solving consistent linear systems $Ax=b$. This paper

4: proposes an accelerated randomized Kaczmarz ($\ARK$) algorithm with

5: better convergence than the standard $\RK$ algorithm on ill

6: conditioned problems.  The per-iteration cost of $\RK$ and $\ARK$

7: are similar if $A$ is dense, but $\RK$ is much more able to exploit

8: sparsity in $A$ than is $\ARK$. To deal with the sparse case, an

9: efficient implementation for $\ARK$, called $\SARK$, is proposed. A

10: comparison of convergence rates and average per-iteration complexities

11: among $\RK$, $\ARK$, and $\SARK$ is given, taking into account

12: different levels of sparseness and conditioning. Comparisons with the

13: leading deterministic algorithm --- conjugate gradient applied to the

14: normal equations --- are also given. Finally, the analysis is

15: validated via computational testing.

16: \end{abstract}

17: