arXiv：cs.LG· Apurva Gandhi, Satyaki Chakraborty, Xiangjun Wang, Aviral Kumar, Graham Neubig·· 5 小时前AI 评分40
递归智能体优化（RAO）：用强化学习训练可递归生成子任务的智能体
Recursive Agent Optimization
AI 导读
研究者提出 Recursive Agent Optimization（RAO），一种用强化学习训练递归智能体的方法，这类智能体可递归生成自身新实例并委派子任务。RAO 教会智能体何时以及如何委派与通信，训练效率更高，能处理超出模型上下文窗口的任务，并泛化到比训练任务更难的问题，相比单智能体系统还可减少实际运行时间。
正文
Abstract:We introduce Recursive Agent Optimization (RAO), a reinforcement learning approach for training recursive agents: agents that can spawn and delegate sub-tasks to new instantiations of themselves recursively. Recursive agents implement an inference-time scaling algorithm that naturally allows agents to scale to longer contexts and generalize to more difficult problems via divide-and-conquer. RAO provides a method to train models to best take advantage of such recursive inference, teaching agents when and how to delegate and communicate. We find that recursive agents trained in this way enjoy better training efficiency, can scale to tasks that go beyond the model's context window, generalize to tasks much harder than the ones the agent was trained on, and can enjoy reduced wall-clock time compared to single-agent systems.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2605.06639 [cs.LG] |
| (or arXiv:2605.06639v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.06639 arXiv-issued DOI via DataCite |
Submission history
From: Apurva Gandhi [view email]
[v1]
Thu, 7 May 2026 17:49:09 UTC (388 KB)
[v2]
Fri, 2 Oct 2026 17:54:50 UTC (439 KB)
来源：arXiv：cs.LG · arxiv.org