arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.25858cs.AIcs.LG
收藏

自进化智能体是持续学习者吗?

Are Self-Evolving Agents Continual Learners?

Mila – Quebec AI Institute(米拉-魁北克人工智能研究所);Université de Montréal(蒙特利尔大学);Google DeepMind(谷歌DeepMind);thepa

Xiaoyin Chen, Yizheng Huang, Yinglun Xu, Ziyun Zhao, Emmett McQuinn, Julie Liu, Hope Yao

AI总结:

本文探究自进化智能体是否为有效持续学习者,评估三类方法发现其虽提升多数任务流整体性能,但存在可塑性与稳定性的权衡问题,仍是不完整的持续学习者。

摘要:

大型语言模型智能体常部署在新知识、工具和工作流不断涌现的环境中,这类智能体需持续适应,在新任务到来时学习而不遗忘旧任务,这类似持续学习问题,但速度是传统梯度更新无法维持的。非参数自进化智能体是自然替代方案:模型权重冻结,通过任务反馈重写自身的控制结构(如提示词、记忆和技能)来适应。本研究在四个主题、工作流或领域漂移的任务流(基于BIRD、BFCL和ScienceWorld构建)中,探究这类智能体是否为有效持续学习者。评估涵盖经验检索和自修改的三种代表性方法(ReasoningBank、GEPA和Meta-Harness)发现:尽管自进化提升了多数任务流的整体性能,但更高可塑性始终以稳定性为代价。具体而言,自修改方法对共享提示词或工作区进行全局编辑,会干扰未见过的任务,且随任务流变化覆盖旧知识;反之,基于相似度的记忆检索在远域定位更新,但在未见过的任务中仍会检索到无关记忆,且当相似查询需要冲突工作流时会出现检索冲突。这些发现表明,当前自进化智能体仍是不完整的持续学习者。

英文摘要:

Large language model agents are often deployed in environments where new knowledge, tools, and workflows keep arriving. Such an agent is therefore expected to adapt continually, learning new tasks as they arrive without forgetting earlier ones. This resembles the continual learning problem, but at a pace that conventional gradient updates cannot sustain. Nonparametric self-evolving agents are a natural alternative: they adapt by rewriting their own harness, such as prompts, memory, and skills, from task feedback while the model weights stay frozen. We study whether such agents are effective continual learners across four drifting task streams that shift in topic, workflow, or domain (built from BIRD, BFCL, and ScienceWorld). Evaluating three representative methods spanning experience retrieval and self-modification (ReasoningBank, GEPA, and Meta-Harness), we find that although self-evolution improves aggregate performance on most streams, higher plasticity consistently comes at the cost of stability. Specifically, self-modifying methods apply global edits to a shared prompt or workspace, interfering with unseen tasks and overwriting earlier knowledge as the stream shifts. Conversely, similarity-based memory retrieval localizes updates across distant domains, but still retrieves foreign memories on unseen tasks and suffers from retrieval collision when similar queries require conflicting workflows. These findings show that current self-evolving agents remain incomplete continual learners.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑