1: \begin{abstract}

2:   Sparse high dimensional graphical model selection is a popular topic

3:   in contemporary machine learning. To this end, various useful

4:   approaches have been proposed in the context of $\ell_1$-penalized

5:   estimation in the Gaussian framework. Though many of these inverse

6:   covariance estimation approaches are demonstrably scalable and have

7:   leveraged recent advances in convex optimization, they still depend

8:   on the Gaussian functional form. To address this gap, a convex

9:   pseudo-likelihood based partial correlation graph estimation method

10:   (CONCORD) has been recently proposed. This method uses 

11:   coordinate-wise minimization of a regression based

12:   pseudo-likelihood, and has been shown to have robust model selection

13:   properties in comparison with the Gaussian approach. In direct

14:   contrast to the parallel work in the Gaussian setting however, this

15:   new convex pseudo-likelihood framework has not leveraged the

16:   extensive array of methods that have been proposed in the machine

17:   learning literature for convex optimization. In this paper, we

18:   address this crucial gap by proposing two proximal gradient methods

19:   (CONCORD-ISTA and CONCORD-FISTA) for performing $\ell_1$-regularized

20:   inverse covariance matrix estimation in the pseudo-likelihood

21:   framework. We present timing comparisons with coordinate-wise

22:   minimization and demonstrate that our approach yields tremendous

23:   payoffs for $\ell_1$-penalized partial correlation graph estimation

24:   outside the Gaussian setting, thus yielding the fastest and most

25:   scalable approach for such problems. We undertake a theoretical

26:   analysis of our approach and rigorously demonstrate convergence, and

27:   also derive rates thereof.

28: \end{abstract}

29: