{"id":"arxiv/2611.21844","url":"https://arxiv.org/abs/2611.21844","tags":["math.OC","cs.LG","stat.ML"],"score":0,"title":"Bias-Reduced and Parallel Heavy-Tailed Stochastic Convex Optimization in $\\ell_p$ Geometries","domain":"math.OC","source":"arxiv","authors":["Zhengyun Nie","Anant Shyam","Brian Bullins"],"reviews":0,"subject":"math.OC","summary":"We consider the fundamental problem of parallel stochastic convex optimization (SCO), whose aim is to approximately minimize a convex objective when given the ability to query a moment-bounded stochastic gradient oracle multiple times in parallel per round. Whereas most previous results have focused on the variance-bounded Euclidean ($\\ell_2$-norm) setting, meaning the objective is Lipschitz with respect to the $\\ell_2$-norm over an $\\ell_2$-bounded feasible set in $\\mathbb{R}^d$, in this work we study parallel SCO in $\\ell_p$ settings for $2 \\le p 0$ with at most $\\tilde{O}(d^{\\frac{p}{2p +…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-11-16","reactions":{}}