English(EN) Policy Iteration Is Not Strongly Polynomial for Deterministic Markov Decision Processes: The Price of Algorithmic Anarchy

新研究表明Howard的策略迭代对于MDP并非强多项式

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-02 04:00

一篇新发表在arXiv上的论文表明，即使在动作数量有限的情况下，Howard的策略迭代算法对于确定性马尔可夫决策过程也可能表现出指数级的迭代下界。这一发现确立了当折扣因子是输入的一部分时，该算法并非强多项式。该研究突显了去中心化的、自私的改进与协调的动作选择之间存在的显著差距，阐释了算法无政府状态的“代价”。 AI

影响强调了决策算法的理论局限性，可能影响强化学习的未来研究。

排序理由发表在arXiv上的学术论文，详细介绍了算法的理论局限性。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 13 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 发表在arXiv上的学术论文，详细介绍了算法的理论局限性。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, other; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Han Zhong, Yinyu Ye · 2026-10-02 04:00

策略迭代并非确定性马尔可夫决策过程的强多项式：算法无政府状态的代价

arXiv:2609.40147v1 Announce Type: new Abstract: We establish an exponential iteration lower bound in the number of states for Howard's policy iteration on deterministic discounted Markov decision processes, with at most two actions per state. This rules out strong polynomiality o…

本文如何被排名

报道来源 [1]

策略迭代并非确定性马尔可夫决策过程的强多项式：算法无政府状态的代价

相关实体

相关话题