{"id":"arxiv/2611.23221","url":"https://arxiv.org/abs/2611.23221","tags":["math.OC","cs.LG"],"score":0,"title":"An inexact stochastic variance-reduced gradient method for constrained convex finite-sum optimization","domain":"math.OC","source":"arxiv","authors":["Zhaosong Lu","Yifeng Xiao"],"reviews":0,"subject":"math.OC","summary":"In this paper, we study a class of constrained convex finite-sum optimization problems. In many practical applications, all or some of the constraints are hard constraints that represent imperative requirements. Consequently, any desirable approximate solution must (nearly) satisfy these constraints. Motivated by this, we focus on computing an $ε$-$surely\\ feasible\\ stochastic\\ optimal$ ($ε$-SFSO) solution, in which the constraint violation is deterministically bounded by $ε$, and the expected optimality gap is at most $ε$. While a recently proposed stochastic variance-reduced first-order…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-11-17","reactions":{}}