{"id":"arxiv/2604.06525","url":"https://arxiv.org/abs/2604.06525","tags":["math.OC","cs.LG"],"score":0,"title":"Stochastic Auto-conditioned Fast Gradient Methods with Optimal Rates","domain":"math.OC","source":"arxiv","authors":["Yao Ji","Guanghui Lan"],"reviews":0,"subject":"math.OC","summary":"Achieving optimal rates for stochastic composite convex optimization without prior knowledge of problem parameters remains a central challenge. In the deterministic setting, the auto-conditioned fast gradient method has recently been proposed to attain optimal accelerated rates without line-search procedures or prior knowledge of the Lipschitz smoothness constant, providing a natural prototype for parameter-free acceleration. However, extending this approach to the stochastic setting has proven technically challenging and remains open. Existing parameter-free stochastic methods either fail…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-04-07","reactions":{}}