English(EN) From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning

新研究表明Transformer可以实现Q学习

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-15 04:00

研究人员从理论上证明，一个双层、仅注意力机制的Transformer可以实现一种Q学习（一种强化学习算法）的形式。这种实现近似于贪婪动作选择，近似程度由温度参数控制。实验表明，在随机生成的马尔可夫决策过程任务上训练时，出现了预测的参数结构，在未见过任务上的性能与传统的Q学习相当。 AI

影响这项研究表明Transformer可以学习和执行复杂的强化学习算法，可能使AI代理更加复杂。

排序理由该集群包含一篇详细介绍AI算法理论和实验结果的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

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: 14 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍AI算法理论和实验结果的学术论文。[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, model release; 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。

arXiv cs.AI TIER_1 English(EN) · Yingjun Mo, Yajie Zhao, Fang Kong · 2026-10-15 04:00

从策略评估到离策略控制：可证明的上下文内 Q-学习的涌现

arXiv:2610.18057v1 Announce Type: cross Abstract: Transformers can implement reinforcement-learning algorithms in context, yet how off-policy control emerges within Transformers remains poorly understood. We study Q-learning, a foundational model-free control algorithm that combi…