Development and applications of a Frank-Wolfe algorithms
Seminar organized by OptimiX
Presenter
Mathieu BesançonAffiliation
Zuse Institut Berlin (ZIB)
Date
June 3, 2022 at 10:00
Local
Grace Hopper Meeting Room
Abstract:
Frank-Wolfe algorithms are first-order methods that support structured convex constraints, over which linear optimization can be performed efficiently. We present FrankWolfe.jl, a library for flexible and high-performance Frank-Wolfe algorithms. In particular, we will detail design choices that allow warm starts, scaling to large problems and some use cases leveraging those features.