arXiv cs.CL· Yanli Wang, Zichen Bai, Tianyue Jiang, Kunming Shao, Junhao Shi, Xueqin Liu, Jinguo Sun, Muye Li, Tingting Bi, Yanlin Wang, Shaoqiang Zheng, Xuefei Diao, Zhuwei Zeng, Nan Duan, Peng Cao·· 12 小时前AI 评分32
EvoGuard：缓解在线技能演化中的性能退化
EvoGuard: Mitigating Degradation in Online Skill Evolution
AI 导读
EvoGuard 是一种成对技能选择器，可在低延迟低成本下缓解语言智能体在线技能演化中的退化问题。它先用私有测试集为技能池打分并生成效用标签，再通过拒绝采样保留效用差距明显的技能对，训练选择器基于两个标签 token 的 logits 判断优劣，部署时单次前向即可决策。
正文 · 原文
Authors:Yanli Wang, Zichen Bai, Tianyue Jiang, Kunming Shao, Junhao Shi, Xueqin Liu, Jinguo Sun, Muye Li, Tingting Bi, Yanlin Wang, Shaoqiang Zheng, Xuefei Diao, Zhuwei Zeng, Nan Duan, Peng Cao
Abstract:Skills let language agents reuse experience across tasks as natural-language documents. Recent work evolves skills from interaction, but existing methods have limitations when skills evolve online. Prior methods update a skill offline from a batch of trajectories collected over many tasks. Some then adopt the new skill without any check, which can degrade the skill over time. Others keep it only if it improves performance on a held-out validation set, which is too slow and costly to run after every update. We introduce EvoGuard, a pairwise skill selector that mitigates skill degradation at low latency and cost. We first run the agent iteratively to build pools of skills and score each skill on a private test set to obtain utility labels. We pair the skills and use rejection sampling to keep only pairs with a clear utility gap. The selector is trained on the logits of two label tokens to pick the better skill. At deployment it decides in a single forward pass, which is significantly faster than evaluation on the private test set. Compared with adopting every new skill, EvoGuard improves success by 11.0% on ALFWorld, 5.9% on ScienceWorld, and 2.7% on LiveCodeBench.
| Comments: | 16 pages, 6 figures, 7 tables, including appendices |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.06444 [cs.CL] |
| (or arXiv:2611.06444v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.06444 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yanli Wang [view email]
[v1]
Wed, 4 Nov 2026 12:47:32 UTC (358 KB)
来源：arXiv cs.CL · arxiv.org