Large Language Models as Falsifiers for Cyber-Physical Systems

AI 导读

文章提出LLM-Falsifier方法，将大语言模型用于网络物理系统验证，通过优化STL鲁棒性度提升反驳效率，在基准测试中超越传统工具。

This paper introduces LLM-Falsifier, a method using large language models to enhance cyber-physical system verification by optimizing STL robustness, outperforming existing tools in benchmark tests.

重点速览

• LLM-Falsifier结合LLM与反驳方法，通过最小化STL鲁棒性度寻找规范反例 LLM-Falsifier integrates large language models with falsification to minimize STL robustness degree
• 引入自然语言输入输出及时间见证信息，提升优化效率 Incorporates natural language inputs/outputs and time witnesses for smarter optimization
• 在ARCH-COMP基准测试中，14/21规范表现优于传统优化范式 Outperforms traditional methods in 14/21 ARCH-COMP benchmarks
• 突破传统数值优化器限制，利用语言模型语义优势 Breaks numerical optimizer limitations via language model semantics
• 为CPS形式验证提供新范式，降低反例搜索样本量 Provides new paradigm for CPS formal verification

一句话：LLM-Falsifier通过自然语言信息优化，显著提升CPS验证效率。 / LLM-Falsifier enhances CPS verification efficiency through semantic-aware optimization.

推荐理由 展示LLM在系统验证中的潜力，提供新方法论。