Home  /  Courses  /  IOE899 Seminar Schedule

IOE 899: Seminar in Industrial and Operations Engineering
Wed Oct 21, 2009, 4:00-5:00pm, 1680 IOE

Katya Scheinberg, Columbia University
"Recent advances in model based derivative free optimization"

Abstract
Derivative free optimization addresses general nonlinear optimization problems in the cases when obtaining derivative information for the objective and/or the constraint functions is impractical due to computational cost or numerical inaccuracies. Applications of derivative free optimization arise often in engineering design such as circuit tuning, aircraft configuration, water pipe calibration, oil reservoir modeling, etc. Traditional approaches to derivative free optimization until late 1990's have been based on sampling of the objective function, without any attempt to build models of the function or its derivatives. While the most traditional of these approaches - the Nelder-Mead algorithm - is known to be relatively successful in practice, it is also know to fail to converge even on very simple problems. A class of sampling methods known as pattern search methods were developed in the 90's by Dennis and Torczon with the appropriate convergence theory. Unfortunately their practical performance is far inferior to that of the Nelder-Mead algorithm. In the late 90's model based trust region derivative free methods started to gain popularity, pioneered by Powell and further advanced by Conn, Scheinberg and Toint. These methods build linear or quadratic interpolation model of the objective function and hence can exploit some first and second oder information. The model based methods have been consistently gaining a wider audience, due to their superior performance. However convergence theory of these method was far from perfect - the assumptions imposed on the methods were not practical. In the last several years the general convergence theory for these methods, under reasonable assumptions, was developed by Conn, Scheinberg and Vicente. Moreover, recently Scheinberg and Toint have discovered the "self-correcting" property which helps explain the good performance observed in these methods and have shown convergence under very mild requirements. Based on the paper: http://www.optimization-online.org/DB_HTML/2009/02/2216.html and some parts of the book: http://www.ec-securehost.com/SIAM/MP08.html
 
Bio
Katya Scheinberg received her PhD from the department of Industrial Engineering and Operations Research at Columbia University in 1997. Since then she has worked as a Research Staff Member in the Mathematical Sciences Department of IBM Research, Yorktown Height, New York. She is currently visiting Columbia as a research scientist. Katya's research interest lie in the field of continuous optimization and cover various areas from derivative free optimization to convex optimization, interior point methods, first order methods for convex optimization and applications in machines learning and statistics. She is the author of three open source optimization packages as well as numerous papers covering these subjects. In January 2009 she has published a book "Introduction to Derivative Free Optimization" which she co-authored with A. R. Conn and L. N. Vicente and which is published in the SIAM/MPS Optimization series. Recently Katya's research has focused on first order methods for large scale semidefinite programming problems.

back to top

Topic-Based Site-Wide Navigation

back to top

Search

Audience-Based Site-Wide Navigation:

back to top