推荐williamstein · 2023-04
“There's interesting research going into combining the LEAN theorem prover with LLMs, e.g., this recent paper: Formal Mathematics Statement Curriculum Learning”
推荐这篇把 LEAN 定理证明器与 LLM 结合的近期研究
用专家迭代让语言模型在形式数学上自学出难度课程
论文把专家迭代(证明搜索与学习交替进行)用于面向形式数学的语言建模。在相同算力预算下,专家迭代显著优于只做证明搜索;当形式命题的难度足够多样时,它还能自行发现并解决难度递增的课程,而不需要对应的标准证明。把它用于人工整理的命题集后,在 miniF2F 基准上达到当时最优,自动解出了多道高中奥赛难题。
评论者把它作为“LEAN 定理证明器加 LLM”研究的例子,说明可以不靠人写的证明,而用自生成的形式命题课程和强化学习来训练证明器。
“There's interesting research going into combining the LEAN theorem prover with LLMs, e.g., this recent paper: Formal Mathematics Statement Curriculum Learning”
“In terms of using reinforcement learning to learn beyond human written proofs, there's TacticZero ([LINK]), this paper from OpenAI”