1: \begin{abstract}

2: \begin{quote}

3: In many learning tasks, structural models usually lead to better 

4: interpretability and higher generalization performance. In recent 

5: years, however,   the simple structural models such as lasso are frequently proved to be

6: insufficient. Accordingly, there has been a lot of work on 

7: ``superposition-structured'' models where multiple structural 

8: constraints are imposed. To efficiently solve these 

9: ``superposition-structured'' statistical models, we develop a framework 

10: based on a proximal Newton-type method.  Employing the smoothed conic dual 

11: approach with the LBFGS updating formula, we propose a scalable and extensible 

12: proximal quasi-Newton (SEP-QN) framework. Empirical analysis on various 

13: datasets shows that our framework is potentially powerful, and achieves 

14: super-linear convergence rate for optimizing some popular 

15: ``superposition-structured'' statistical models such as the fused sparse group 

16: lasso.

17: \end{quote}

18: \end{abstract}

19: