Academic Seminar |Nonsmooth Newton methods for large scale convex and nonconvex optimization problems

Title:Nonsmooth Newton methods for large scale convex and nonconvex optimization problems

Speaker: Dr. Ying CUI, University of Southern California

Time and Date: 11:00am -12:00pm, August 22, 2018

Venue: Boardroom, Dao Yuan Building

Abstract: 

First order methods are widely used nowadays for solving large scale optimization problems. In this talk, I will discuss the possibility and necessity for incorporating second order information in solving difficult optimization problems. One example of such problems is a class of degenerate convex semidefinite programming. Guided by the error bounds theory, we propose a semismooth Newton-CG based augmented Lagrangian method to solve this class of problems with provable fast convergence rate and convincing numerical results. Another emerging area that calls for second order methods comes from modern statistical estimation problems. We will take a class of difference-convex-piecewise regression problems as an example to show how the dual semismooth Newton method can be embedded in the majorization minimization framework and its advantage over various first order subproblem solvers.

Biography:

Ying Cui is currently a postdoc research associate in Department of Industrial and Systems Engineering at University of Southern California, working with Professor Jong-Shi Pang. She received her Ph.D from Department of Mathematics, National University of Singapore in 2016, under the supervisions of Professor Defeng Sun and Professor Chenlei Leng.