{"id":"arxiv/2609.03870","url":"https://arxiv.org/abs/2609.03870","tags":["math.OC"],"score":0,"title":"Data-Driven Brownian Reflection Control","domain":"math.OC","source":"arxiv","authors":["Guodong Pang","Dacheng Yao","Hao Yin"],"reviews":0,"subject":"math.OC","summary":"We study a data-driven reflection control problem for a Brownian model with unknown drift and volatility. We first propose a learn-then-optimize (LTO) algorithm: it estimates the policy-relevant parameter during exploration, plugs the estimate into the optimality equation, and exploits the resulting policy---achieving an $O(\\sqrt{T})$ finite-time expected regret bound. We further propose two algorithms, adaptive-updating (AU) and full-history adaptive-updating (AU-FH), which continuously update the estimator and reflecting level, attaining an improved $O(\\log T)$ regret bound. Notably, AU-FH…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-03","reactions":{}}