1: \begin{abstract}

2: We propose Shotgun, a parallel coordinate descent algorithm for minimizing $\ellone$-regularized losses.

3: Though coordinate descent seems inherently sequential, we prove convergence bounds for Shotgun which predict linear speedups,

4: up to a problem-dependent limit.

5: We present a comprehensive empirical study of Shotgun for Lasso and sparse logistic regression.

6: Our theoretical predictions on the potential for parallelism closely match behavior on real data.

7: Shotgun outperforms other published solvers on a range of large problems, proving to be one of the most scalable algorithms for $\ellone$.

8: \end{abstract}