{"id":"arxiv/2605.07104","url":"https://arxiv.org/abs/2605.07104","tags":["cs.LG","math.OC","stat.ML"],"score":0,"title":"Almost Sure Convergence Rates of Stochastic Approximation and Reinforcement Learning via a Poisson-Moreau Drift","domain":"cs.LG","source":"arxiv","authors":["Xinyu Liu","Zixuan Xie","Shangtong Zhang"],"reviews":0,"subject":"cs.LG","summary":"Establishing almost sure convergence rates for stochastic approximation and reinforcement learning under Markovian noise is a fundamental theoretical challenge. We make progress towards this challenge for a class of stochastic approximation algorithms whose expected updates are contractive, a setting that arises in many reinforcement learning algorithms such as $Q$-learning and linear temporal difference learning. Specifically, for a power-law learning rate $O(n^{-η})$ with $η\\in (1/2, 1)$, we obtain an almost sure convergence rate arbitrarily close to $o(n^{1 - 2η})$. For a harmonic…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-05-08","reactions":{}}