[Ml-stat-talks] Princeton Optimization Seminar: Andrzej Ruszczynski, Thu. April 28, 4:30 PM
Amir Ali Ahmadi
a_a_a at princeton.edu
Sun Apr 24 10:46:43 EDT 2016
----- Princeton Optimization Seminar -----
DATE: Thursday, April , 2016
TIME: 4:30 PM
LOCATION: Sherrerd Hall 101
SPEAKER: Andrzej Ruszczynski, Rutgers University (on sabbatical at Princeton ORFE)
TITLE: Selective Linearization for Large-Scale Multi-Block Optimization<https://orfe.princeton.edu/abstracts/optimization-seminar/selective-linearization-large-scale-multi-block-optimization>
We consider the problem of minimizing a sum of many functions of a high-dimensional vector, under the assumption that any of these functions individually, augmented with a quadratic term, is relatively easy to optimize. The problem is well studied for the case of two functions, with several efficient methods available. After a review of existing approaches to the problem with more than two functions, we propose a new method, called selective linearization, which has theoretical convergence and rate of convergence guarantees, and exhibits superior practical performance. The main idea of the method is to solve at each iteration a regularized subproblem, in which one function is treated exactly, while all other functions are replaced by their affine minorants. The affine minorants use specially selected subgradients of the functions, which are implicitly defined at earlier steps of the method. The selection of the lead function and updates of the proximal center are governed by specific algorithmic rules. We also discuss the subproblem solution methods in typical applications. The last part of the talk will be devoted to the illustration of the performance of the method on large data analysis problems.
Andrzej Ruszczynski received his PhD and Habilitation degrees in control engineering from Warsaw University of Technology in 1976 and 1983, respectively. He has been with Warsaw University of Technology (Poland), University of Zurich (Switzerland), International Institute of Applied Systems Analysis (Laxenburg, Austria), Princeton University, University of Wisconsin-Madison, and Rutgers University, where he currently holds the rank of Distinguished Professor. Dr. Ruszczynski is one of the creators of and main contributors to the field of risk-averse optimization, author of "Nonlinear Optimization" (Princeton University Press, 2006), co-author of "Lectures on Stochastic Programming" (Society of Industrial and Applied Mathematics, 2009), "Stochastic Programming" (Elsevier, 2003), and author of more than 100 articles in the area of optimization. Dr. Ruszczynski was plenary speaker at several major international conferences and held positions in large scientific societies.
*** If you would like to subscribe to the mailing list of the optimization seminar series, please visit the link below or send an email with
SUBSCRIBE opt-seminar in the body to listserv at lists.princeton.edu<mailto:listserv at lists.princeton.edu>.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ml-stat-talks