IAM-PIMS-MITACS Distinguished Colloquium Series: Stephen Wright (University of Wisconsin)

In many applications of optimization, an exact solution is less useful than a simple, well structured approximate solution. An example is found in compressed sensing, where we prefer a sparse signal (e.g. containing few frequencies) that matches the observations well to a more complex signal that matches the observations even more closely. The need for simple, approximate solutions has a profound effect on the way that optimization problems are formulated and solved. Regularization terms can be introduced into the formulation to induce the desired structure, but such terms are often nonsmooth and thus may complicate the algorithms. On the other hand, an algorithm that is too slow for finding exact solutions may become competitive and even superior when we need only an approximate solution. In this talk we outline the range of applications of sparse optimization, then sketch some techniques for formulating and solving such problems, with a particular focus on applications such as compressed sensing and data analysis.

-----

Steve Wright is a Professor in the Computer Sciences Department at University of Wisconsin in Madison, with an additional appointment in the Department of Industrial and Systems Engineering. He is a member of the Optimization Group, the Decision Science/Operations Research Group, and the Committee on Optimization and its Applications at UW-Madison. He also serves as Chair of the Mathematical Optimization Society (formerly Mathematical Programming Society) and is a member of the Board of Trustees of the Society for Industrial and Applied Mathematics. Steve's research interests include algorithms for nonlinear optimization, as well as applications of optimization to signal and image processing, process control, computational statistics, computational biology, cancer radiotherapy, weather forecasting, and other areas. He is also involved in the development of optimization and compressed sensing software.