ISE seminar- Tuesday 3-1

INDUSTRIAL AND SYSTEMS ENGINEERING DEPARTMENT
SPRING 2011 SEMINAR SERIES

TITLE:                 Using Active-Set Phases to Accelerate Optimization Algorithms

SPEAKER:         Dr. Daniel Robinson
                              Northwestern University, Illinois

DATE / TIME:    Tuesday, March 1, 2011
                             1:30 – 2:30 p.m.

LOCATION:      Room 375 Mohler Lab, 200 W. Packer Avenue

ABSTRACT: In the first part of the talk I will clarify the meaning of “active-set phases” by giving an overview of three optimization algorithms that are currently being developed. The first algorithm is designed for solving linear complementarity problems and is motivated by a financial optimization problem. The second algorithm is used for solving (convex) stochastic optimization problems and will be described in the context of a speech recognition application. The third algorithm is a general-purpose second- derivative sequential quadratic programming method that is designed for solving sparse large-scale nonlinear and nonconvex problems. Since this method benefits from so-called “warm-starting”, I will motivate its development by considering a simple trajectory optimization problem.
I will dedicate the second part of the talk to a more detailed description of the sequential quadratic programming algorithm. In particular, I will (i) describe various difficulties associated with developing second-derivative sequential quadratic programming methods; (ii) explain the strategies that I used to avoid these hurdles; (iii) provide global and local convergence results; and (iv) furnish numerical tests that elucidate the benefits of an active-set phase.

BIOGRAPHY: Daniel Robinson received his Ph.D. from the Mathematics and Statistics Department at the University of California, San Diego in 2007. In the summer of 2006 he worked for Northrop Grumman as a consultant with his advisor Philip Gill and developed algorithms for trajectory optimization problems. From 2007-2010, he worked as a research assistant to Nick Gould at the University of Oxford where he developed and implemented optimization algorithms for large-scale nonlinear and convex optimization. During this time, he also maintained a visiting position at the Rutherford Appleton Laboratory, formally known as Harwell. Daniel is currently a post-doc for Jorge Nocedal in the Industrial Engineering and Management Sciences Department at Northwestern University, and his current research interests include algorithms for machine learning, linear complementarity problems, and large-scale nonlinear optimization.


ALL FULL-TIME ISE PHD STUDENTS ARE REQUIRED TO ATTEND AND

ALL ISE GRADUATE STUDENTS ARE ENCOURAGED TO ATTEND
REFRESHMENTS WILL BE SERVED FOLLOWING THE SEMINAR – IN THE 4TH FLOOR GOTT LOUNGE AREA