{"id":"arxiv/2607.25785","url":"https://arxiv.org/abs/2607.25785","tags":["math.OC","cs.LG"],"score":0,"title":"Variance-Reduced Conditional Gradient Methods under Markovian Sampling for Nonconvex Composite Optimization","domain":"math.OC","source":"arxiv","authors":["Zhaojun Peng"],"reviews":0,"subject":"math.OC","summary":"We study stochastic composite nonconvex optimization over a compact convex set when gradient samples arrive along a single trajectory of a fixed ergodic Markov chain. Existing single-trajectory variance-reduction theory covers smooth unconstrained objectives; we address the projection-free composite setting using the generalized Frank-Wolfe gap. We propose MC-ALFCG, which combines a momentum conditional-gradient method with coupled capped multilevel Monte Carlo estimation and per-iteration clipping. The deepest nested average uses consecutive states from the same trajectory, yielding…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-28","reactions":{}}