跳到正文
arXiv cs.AI· You-Lin Chen, Xinyi Lisa Chen, Pedro Herrero-Vidal, Xiaojing Yu, Bingqing Song, Bin Xu·· 4 小时前AI 评分41

LeBet:学习可解释规则实现可靠的 Agent 介入

Learn then Bet: Learning Interpretable Rules for Reliable Agent Intervention

AI 导读

研究者提出 LeBet(Learn then Bet)统计学习框架,用于判断何时对长时间运行的自主 Agent 进行停止、升级或人工协助等介入。该框架学习可解释的 OR-of-AND 监控规则,通过固定投注统计量累积证据,在达到阈值时介入,并在有限交换性条件下给出假阳性控制与检测功效的有限样本保证。

正文 · 原文

View PDF HTML (experimental)

Abstract:Autonomous agents now run for hundreds of steps with little or no human supervision. In practice, practitioners use runtime monitors such as turn limits and keyword filters to decide when a run should be stopped, upgraded, or assisted. However, these heuristics may stop runs that would otherwise succeed or allow failing runs to continue. To address this limitation, we introduce LeBet (Learn then Bet), a statistical learning framework that derives its learning objective from sufficient conditions for valid and powerful intervention. In particular, LeBet learns an interpretable OR-of-AND monitor and accumulates evidence through a fixed betting statistic, intervening when the evidence reaches a prescribed threshold. Under finite within-class exchangeability and leave-one-out stability, we derive finite-sample guarantees for false-positive control (validity) and detection power conditional on the training data, using the same trajectories for learning and calibration without assuming independent and identically distributed trajectories. To address the intractable combinatorial optimization problem in our learning formulation, we use stochastic local search that leverages the monitor's Boolean structure to propose rule edits aimed at improving empirical separation. On SWE-bench Lite with Ministral 3 8B, the learned monitor saves $34.46\%$ of input-plus-output tokens and increases output precision from $18.33\%$ to $21.59\%$. The empirical false-positive rate is $13.64\%$. Our results show how to formalize the learning problem over interpretable rule sets with statistical conditions and establish training-conditional guarantees for reliable intervention based on partial histories.
Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.17662 [cs.AI]
  (or arXiv:2611.17662v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.17662

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: You-Lin Chen [view email]
[v1] Fri, 13 Nov 2026 02:14:55 UTC (357 KB)

来源:arXiv cs.AI · arxiv.org