[Home]   [  News]   [  Events]   [  People]   [  Research]   [  Education]   [Visitor Info]   [UCSD Only]   [Admin]
Home > Events > CCoM > Abstract
Search this site:

A Shifted Primal-Dual Trust-Region Interior-Point Algorithm

Alex Guldemond
UCSD

Abstract:

Interior-point methods are some of the most effective and widely used methods to finding local minimizers of large-scale non-convex optimization problems. In this talk, we introduce three different mechanisms for ensuring global convergence to second-order local minimizers from arbitrary feasible starting points by solving a sequence of trust-region subproblems defined by quadratic models of a shifted primal-dual penalty-barrier merit function. Each of these methods begins by solving the trust-region subproblem to form a new trial point, and proceeds to refine the trial iterate until a sufficient-decrease condition is met. We suggest two different definitions of the trust region, and provide numerical results comparing each of the different approaches.

Tuesday, March 8, 2022
11:00AM Zoom ID 922 9012 0877