arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.04883cs.CLcs.LO
收藏

优化 Lean 中智能定理证明器的成本-质量权衡

Optimizing the Cost-Quality Tradeoff of Agentic Theorem Provers in Lean

University of Washington(华盛顿大学);University of California, Berkeley(加州大学伯克利分校)

Kári Rögnvaldsson, Chenhao Sun, Jasper Dekoninck, Martin Vechev

AI总结:

提出一种包含数据平面和控制平面的动作路由智能体,通过观察失败轨迹并估计成功概率与成本来动态决定继续证明或重新分解,在 PutnamBench 子集上平均降低 25.8% 成本且保持性能。

摘要:

大型语言模型(LLMs)越来越多地用于在 Lean 中生成形式化证明的工作流程。这些工作流程通常将问题分解为更小的引理,采样许多证明尝试,并使用编译器反馈来指导搜索。然而,它们可能成本高昂,往往在最终失败的尝试上花费大量计算。在这项工作中,我们通过一个包含数据平面和控制平面的动作路由智能体来解决这个问题。数据平面生成自然语言的引理分解,在 Lean 中形式化它们,并为由此产生的定理和引理目标采样证明尝试。控制平面观察之前失败的 Lean 尝试,估计成功可能性和另一次尝试的成本,并决定是继续证明当前目标还是从新的分解重新开始。在 PutnamBench 的一个子集上,我们的智能体平均比固定步长基线降低 25.8% 的成本,在显著减少计算量的同时保持性能。这些结果表明,失败的 Lean 轨迹为智能定理证明中的成本感知资源分配提供了可操作的信号。

英文摘要:

Large language models (LLMs) are increasingly used in workflows for generating formal proofs in Lean. These workflows often decompose problems into smaller lemmas, sample many proof attempts, and use compiler feedback to guide search. However, they can be prohibitively expensive, often spending substantial compute on attempts that ultimately fail. In this work, we address this problem with an action routing agent that consists of a data plane and a control plane. The data plane generates natural-language lemma decompositions, formalizes them in Lean, and samples proof attempts for the resulting theorem and lemma targets. The control plane observes previous failed Lean attempts, estimates both the likelihood of success and the cost of another attempt, and decides whether to continue proving the current target or restart from a new breakdown. On a subset of PutnamBench, our agent decreases the cost by 28.9% over a fixed-step baseline on average, preserving performance while using substantially less compute. These results suggest that failed Lean trajectories provide actionable signals for cost-aware resource allocation in agentic theorem proving.

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。

↑