## Conditional gradient-based method for bilevel optimization with convex lower-level problem 02 Oct 2022, 17:24 (modified: 23 Nov 2022, 20:17)OPT 2022 PosterReaders: Everyone
Keywords: bilevel optimization, conditional gradient method
TL;DR: We propose a conditional gradient-based method for solving a class of bilevel optimization problems and provide the best-known iteration complexity in the literature.
Abstract: In this paper, we study simple bilevel optimization problems, where we minimize a smooth objective function over the optimal solution set of another convex constrained optimization problem. Several iterative methods have been developed for tackling this class of problems. Alas, their convergence guarantees are not satisfactory as they are either asymptotic for the upper-level objective, or the convergence rates are slow and sub-optimal. To address this issue, in this paper, we introduce a conditional gradient-based (CG-based) method to solve the considered problem. The main idea is to locally approximate the solution set of the lower-level problem via a cutting plane, and then run a CG-type update to decrease the upper-level objective. When the upper-level objective is convex, we show that our method requires ${\mathcal{O}}(\max\{1/\epsilon_f,1/\epsilon_g\})$ iterations to find a solution that is $\epsilon_f$-optimal for the upper-level objective and $\epsilon_g$-optimal for the lower-level objective. Moreover, when the upper-level objective is non-convex, our method requires ${\mathcal{O}}(\max\{1/\epsilon_f^2,1/(\epsilon_f\epsilon_g)\})$ iterations to find an $(\epsilon_f,\epsilon_g)$-optimal solution. To the best of our knowledge, our method achieves the best-known iteration complexity for the considered bilevel problem.
0 Replies