Events

Jennifer Erway

Wake Forest University
Colloquia

When

Date: Wednesday, October 26, 2022
Time: 3:30 pm - 5:00 pm
Location: Petty 150
Dr. Jennifer Erway is a professor at Wake Forest University. She received her Ph.D. from University of California, San Diego in 2006. She is currently the Gale Family Faculty Fellow at Wake Forest and has won both research and teaching awards while at Wake. In addition, she has been PI on several NSF grants. Her research interests include large-scale optimization, numerical linear algebra, and applications of optimization.

Large-scale optimization refers to solving optimization problems where the number of variables is greater than 10,000.  In some applications the number of variables can be in the millions or more. In this setting, basic linear algebra tasks can either be too computationally time-consuming to perform or the storage requirements can be too demanding. The fastest optimization methods require the use of second-derivative matrices, which can be prohibitive to compute and/or store when the number of variables is large.  In this case, first-order methods may be the fastest methods available to solve general large-scale optimization problems.  The simplest first-order methods include steepest-descent and gradient-based methods, which are frequently used in engineering and machine-learning applications.  However, faster methods exist that make more extensive use of computed first derivatives.  In this talk, we review quasi-Newton methods, one of the most popular types of first-order methods. We generalize these ideas to produce multipoint symmetric secant (MSS) methods, which can be used in a trust-region framework.  The MSS method presented in this talk enjoys some auspicious linear algebra properties that make it a good candidate for solving large-scale optimization problems.