arXiv：cs.LG· Yaniv Oren, Viliam Vadocz, Joery A. de Vries, Wendelin B\"ohmer, Matthijs T. J. Spaan, Hendrik Baier·· 4 小时前AI 评分42
Particle MCTS：面向 GPU 批量并行的并行蒙特卡洛树搜索算法
Particle Monte Carlo Tree Search
AI 导读
研究者提出 Particle MCTS（PMCTS），一种面向神经网络评估、支持 GPU 批量并行的并行 MCTS 算法，并保留 MCTS 的近似策略改进解释。PMCTS 随并行算力扩展良好，在 Chess、19x19 Go、9x9 Go、Gardner Chess、Snake、Brax 经典控制环境及 Sokoban 中的 LLM 推理等离散与连续动作基准上，持续优于或比肩主流启发式基线。
正文
Abstract:Monte Carlo Tree Search (MCTS) is a widely used approach for policy improvement and action selection in Reinforcement Learning. Due to its sequential and deterministic nature, principled runtime-scaling of MCTS with parallel compute remains a major challenge. We introduce Particle MCTS (PMCTS), a parallel MCTS algorithm which is suited for neural network evaluations, designed for GPU-acceleration with batch-parallelization and retains MCTS's principled approximate policy improvement interpretation. Empirically, PMCTS scales well with parallel compute and consistently outperforms or compares well to the popular heuristic-based baselines across a range of popular discrete- and continuous-action benchmark domains, including Chess, 19x19 Go, 9x9 Go, Gardner Chess, Snake, classical control environments from Brax and LLM reasoning in Sokoban.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.08982 [cs.LG] |
| (or arXiv:2605.08982v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.08982 arXiv-issued DOI via DataCite |
Submission history
From: Yaniv Oren [view email]
[v1]
Sat, 9 May 2026 14:54:07 UTC (156 KB)
[v2]
Thu, 21 May 2026 09:52:14 UTC (152 KB)
[v3]
Mon, 7 Sep 2026 08:47:23 UTC (209 KB)
[v4]
Tue, 6 Oct 2026 13:59:56 UTC (205 KB)
来源：arXiv：cs.LG · arxiv.org