中文

周期正则化Q学习

机器学习 2026-02-04 v1 人工智能

摘要

在强化学习(RL)中,Q学习是一种基本算法,其在表格设置下具有收敛保证。然而,在线性函数逼近情况下,这种收敛保证并不成立。为克服这一限制,大量研究引入了正则化技术以确保在函数逼近下的稳定收敛。本文提出了一种新算法,称为周期正则化Q学习(PRQ)。我们首先在投影算子层面引入正则化,显式构造正则化投影值迭代(RP-VI),随后将其扩展为基于样本的RL算法。通过合适地对投影算子进行正则化, resulting in the projected value iteration becomes a contraction. By extending this regularized projection into the stochastic setting, we establish the PRQ algorithm and provide a rigorous theoretical analysis that proves finite-time convergence guarantees for PRQ under linear function approximation.

关键词

引用

@article{arxiv.2602.03301,
  title  = {Periodic Regularized Q-Learning},
  author = {Hyukjun Yang and Han-Dong Lim and Donghwan Lee},
  journal= {arXiv preprint arXiv:2602.03301},
  year   = {2026}
}