AutoICE: 通过LLM驱动的进化自动合成可验证的C代码
AutoICE: Automatically Synthesizing Verifiable C Code via LLM-driven Evolution
AI总结:
AutoICE通过LLM驱动的进化搜索方法,自动合成可验证的C代码,验证成功率高达90.36%,优于现有方法。
摘要:
自动从自然语言需求中合成可验证的代码可以确保软件的正确性和可靠性,同时显著降低采用形式化方法技术的门槛。随着大语言模型(LLMs)的兴起,长期存在的自动形式化努力获得了新的活力。然而,现有方法由于领域特定预训练语料匮乏,导致严重的语法和语义错误,且往往无法有效形式化隐含知识。在本文中,我们提出了AutoICE,一种基于LLM的进化搜索方法,用于合成可验证的C代码。它引入了多样个体初始化和协作交叉,以实现多样化的迭代更新,从而减轻单体迭代中固有的错误传播。此外,它采用自我反思突变,以促进隐含知识的发现。评估结果证明了AutoICE的有效性:它成功验证了90.36%的代码,优于最先进的(SOTA)方法。此外,在一个开发者友好的数据集变体上,AutoICE实现了88.33%的验证成功率,显著超过了SOTA方法的65%成功率。
英文摘要:
Automatically synthesizing verifiable code from natural language requirements ensures software correctness and reliability while significantly lowering the barrier to adopting the techniques of formal methods. With the rise of large language models (LLMs), long-standing efforts at autoformalization have gained new momentum. However, existing approaches suffer from severe syntactic and semantic errors due to the scarcity of domain-specific pre-training corpora and often fail to formalize implicit knowledge effectively. In this paper, we propose AutoICE, an LLM-driven evolutionary search for synthesizing verifiable C code. It introduces the diverse individual initialization and the collaborative crossover to enable diverse iterative updates, thereby mitigating error propagation inherent in single-agent iterations. Besides, it employs the self-reflective mutation to facilitate the discovery of implicit knowledge. Evaluation results demonstrate the effectiveness of AutoICE: it successfully verifies $90.36$\% of code, outperforming the state-of-the-art (SOTA) approach. Besides, on a developer-friendly dataset variant, AutoICE achieves a $88.33$\% verification success rate, significantly surpassing the $65$\% success rate of the SOTA approach.
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。