Jul 2026
Variance-Reduced Conditional Gradient Methods under Markovian Sampling for Nonconvex Composite Optimization
This work proposes MC-ALFCG, which combines a momentum conditional-gradient method with coupled capped multilevel Monte Carlo estimation and per-iteration clipping, and addresses the projection-free composite setting using the generalized Frank-Wolfe gap.
Zhaojun Peng
· arXiv.org · 0 citations