IE-OPIM Joint Seminar 12 November 2014 at 13.40 at FENS G035
On Block Active Set and Stochastic Gradient Algorithms
Abstract: In this talk, we focus on two mathematical programming problems encountered in certain machine learning applications. The first problem is the minimization of a piecewise-quadratic function. We propose a block active set method to solve this problem, explain the derivation of the method from different perspectives, and discuss convergence issues. Then we show how the proposed approach can be applied to non-quadratic convex problems of the same type. Second, we consider unconstrained minimization of the expected value of a convex function. An old but still popular approach for solving this problem is stochastic approximation. We give a very brief summary of known properties and suggested variants of the stochastic gradient algorithm, and complete the talk by a discussion on the two extensions that are subject to our ongoing research.
Bio: Figen Öztoprak is an assistant professor at Istanbul Bilgi University Industrial Engineering Department. Her research interests are mainly in nonlinear programming and its applications; more specifically, in the development of sequential and parallel algorithms for large-scale continuous optimization problems. She received her Ph.D. from Sabanci University in 2011, was a research fellow at Northwestern University between 2010-2012, and at Istanbul Technical University Informatics Institute between 2012-2014.