Mathieu Besançon, Inria Grenoble

Abstract: In this talk, I will introduce Frank-Wolfe, a family of algorithms for nonlinear optimization over convex sets. A noteworthy property of these algorithms is their reliance on an oracle on the feasible region, instead of algebraic expressions of the constraints. After highlighting interesting properties of the algorithms, we will showcase several applications in network design, optimal design of experiments, and objective-robust optimization, where the constraint structure makes Frank-Wolfe a well-suited approach.