1: \begin{abstract} 

2: This paper addresses the problem of scalable optimization for $l_1$-regularized 

3: conditional Gaussian graphical models.  Conditional Gaussian graphical models 

4: generalize the well-known Gaussian graphical models to conditional distributions 

5: to model the output network influenced by conditioning input variables. 

6: While highly scalable optimization methods exist for sparse 

7: Gaussian graphical model estimation, state-of-the-art methods for 

8: conditional Gaussian graphical models are not efficient enough 

9: and more importantly, fail due to memory constraints for very large problems. 

10: In this paper, we propose a new optimization procedure based on a Newton method 

11: that efficiently iterates over two sub-problems, leading to drastic improvement 

12: in computation time compared to the previous methods. 

13: We then extend our method to scale to large problems under memory constraints, 

14: using block coordinate descent to limit memory usage while achieving fast convergence. 

15: Using synthetic and genomic data, we show that our methods can solve problems with millions of variables and tens of billions of parameters to high accuracy on a single machine.

16: \end{abstract}