A superlinearly convergent first-order method for nonsmooth optimization
Abstract
Nonsmooth optimization problems appear throughout machine learning and signal processing. However, standard first-order methods for nonsmooth optimization can be slow for "poorly conditioned" problems. In this talk, I will present a locally accelerated first-order method that is less sensitive to conditioning and achieves superlinear (i.e., double-exponential) convergence near solutions for a broad family of problems. The algorithm is inspired by Newton's method for solving nonlinear equations. Bio
Vasilis is an AI & Science postdoctoral scholar at the University of Chicago Data Science Institute. He is broadly interested in developing numerical optimization methods for machine learning, signal processing and scientific computing. He holds a PhD in Operations Research & Information Engineering from Cornell University, where he was advised by Damek Davis. Vasilis was recognized as a Rising Star in Computational and Data Sciences by the UT Austin Oden Institute in 2023 and received the Cornelia Ye Outstanding Teaching Assistant Award at Cornell University in 2021. |