Computational Mathematics and Scientific Computing Seminar

---

Pursuing Rapid Convergence for Gradient Sampling-like Methods for Nonsmooth Optimization

Speaker: Lucas Simões, University of Campinas, Brazil

Location: Warren Weaver Hall 1302

Date: Sept. 22, 2017, 10 a.m.

Synopsis:

Minimization problems involving nonsmooth functions appear in many real problems. For convex objective functions, Bundle Methods are important and effective tools for solving nonsmooth optimization problems. However, when the problem is not convex, the theoretical and practical aspects are not so well developed. A reliable alternative to these methods is the Gradient Sampling (GS) algorithm. Based on GS ideas, this talk presents a new method that tries to move superlinearly to the solution of the optimization problem.