报告标题：Automata-Based Steering of Large Language Models for Diverse Structured Generation

报告时间：11月9日9:30-10:30

报告地点：理科大楼B1002

报告摘要：

      Large language models (LLMs) are increasingly tasked with generating structured outputs. While structured generation methods ensure validity, they often lack output diversity, a critical limitation that we confirm in our preliminary study. We propose a novel method to enhance diversity in automaton-based structured generation. Our approach utilizes automata traversal history to steer LLMs towards novel structural patterns. Evaluations show our method significantly improves structural and content diversity while maintaining comparable generation efficiency. Furthermore, we conduct a case study showcasing the effectiveness of our method in generating diverse test cases for testing open-source libraries.

报告人简介：

       Meng Sun received his BSdegree in Information Science and PhD degree in applied mathematics from Peking University in 1999 and 2005, respectively. He worked as a postdoc researcher at National University of Singapore from 2005 to 2006, and as a scientific staff member at CWI, the Netherlands, from 2006 to 2010. He joined Peking University in 2010 and currently is a full professor at School of Mathematical Sciences in Peking University. His research interests mainly lie in theories of programming, software formal methods, cyber-physical systems, trustworthiness guarantee of AI systems, blockchain and smart contracts. His publications appear in top-tier venues including TSE,TDSC, ICSE, FSE, FM, NDSS, NeurIPS, ICML and AAAI and got the best / distinguished paper awards for TASE 2026, FSE 2025, ICFEM 2025, SBMF 2017, TASE 2015, AMAST 2004. He has served as PC Co-chairs of ICFEM (2024 and 2018), TASE (2023), FACS (2024 and 2009), and PC member of international conferences such as FM and TACAS.