Title：On the worst-case complexity of nonlinear stepsize control algorithms for unconstrained optimization

Speaker： Academician Yaxiang Yuan （Chinese Academy of Sciences）

Time：5:00-6:00P.M. Sep 18th, 2015

Location：Room 4131, Building No.4, Wushan Campus

Abstract：A nonlinear stepsize control framework for unconstrained optimization was pro-posed by Toint (Optim Methods Softw 28:8295, 2013), generalizing many trust-region and regularization algorithms. More recently, Grapiglia, Yuan and Yuan (Math. Program. DOI 10.1007/s10107-014-0794-9) proved worst-case complexity bounds for this family of algorithms in the context of nonconvex problems. In this paper, certain classes of nonlinear stepsize control algorithms are shown to have improved worst-case complexity when applied to more speci_c types of problems, including convex and strongly convex ones.