My primary research direction revolves around Quality Assurance for Trustworthy AI systems such as AI-enabled Cyber-Physical systems (AI-CPS), Large Language Models (LLMs), and Multimodal AI Agents.
My research is devoted to exploring these directions in three stages: (1) testing and evaluating the performance and reliability of generative AI in dynamic, real-world environments, (2) developing methods to enhance the safety and resilience of complex AI systems, and (3) investigating the potential of multimodal foundation models to revolutionize the capabilities of AI applications.
My long-term vision is to create a cohesive research program that pushes the boundaries of AI systems by combining foundational models and cutting-edge quality assurance techniques.
Prospective Students
I am looking for self-motivated Ph.D. and MSc. students with strong programming skills and relevant research experience. Please send me an email with your CV and transcripts if you are interested.
Please feel free to email me with any questions, discussions, or collaboration opportunities!
Our paper Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixtures is accepted at NAACL 2025 Findings. This paper introduces Multilingual Blending, a mixed-language query-response scheme designed to evaluate the safety alignment of SOTA LLMs under sophisticated, code-switching conditions.
@article{huang2023look,title={Look before you leap: An exploratory study of uncertainty measurement for large language models},author={Huang, Yuheng and Song, Jiayang and Wang, Zhijie and Zhao, Shengming and Chen, Huaming and Juefei-Xu, Felix and Ma, Lei},journal={IEEE Transactions on Software Engineering.},year={2024},organization={IEEE},}
AI-CPS
When cyber-physical systems meet AI: A benchmark, an evaluation, and a way forward
Jiayang Song, Deyun Lyu, Zhenya Zhang, and 3 more authors
In Proceedings of the 44th International Conference on Software Engineering: Software Engineering in Practice, 2022
@inproceedings{song2022cyber,title={When cyber-physical systems meet AI: A benchmark, an evaluation, and a way forward},author={Song, Jiayang and Lyu, Deyun and Zhang, Zhenya and Wang, Zhijie and Zhang, Tianyi and Ma, Lei},booktitle={Proceedings of the 44th International Conference on Software Engineering: Software Engineering in Practice},pages={343--352},year={2022},}
AI-CPS
Towards building AI-CPS with NVIDIA Isaac sim: An industrial benchmark and case study for robotics manipulation
Zhehua Zhou, Jiayang Song, Xuan Xie, and 5 more authors
In Proceedings of the 46th International Conference on Software Engineering: Software Engineering in Practice, 2024
@inproceedings{zhou2024towards,title={Towards building AI-CPS with NVIDIA Isaac sim: An industrial benchmark and case study for robotics manipulation},author={Zhou, Zhehua and Song, Jiayang and Xie, Xuan and Shu, Zhan and Ma, Lei and Liu, Dikai and Yin, Jianxiong and See, Simon},booktitle={Proceedings of the 46th International Conference on Software Engineering: Software Engineering in Practice},pages={263--274},year={2024},}
AI-CPS
SIEGE: A Semantics-Guided Safety Enhancement Framework for AI-Enabled Cyber-Physical Systems
@article{song2023mathtt,title={SIEGE: A Semantics-Guided Safety Enhancement Framework for AI-Enabled Cyber-Physical Systems},author={Song, Jiayang and Xie, Xuan and Ma, Lei},journal={IEEE Transactions on Software Engineering},volume={49},number={8},pages={4058--4080},year={2023},publisher={IEEE},}
LLM
LUNA: A Model-Based Universal Analysis Framework for Large Language Models
Da Song, Xuan Xie, Jiayang Song, and 4 more authors
@article{song2024luna,title={LUNA: A Model-Based Universal Analysis Framework for Large Language Models},author={Song, Da and Xie, Xuan and Song, Jiayang and Zhu, Derui and Huang, Yuheng and Juefei-Xu, Felix and Ma, Lei},journal={IEEE Transactions on Software Engineering},year={2024},publisher={IEEE},}
Robotics-RL
GenSafe: A Generalizable Safety Enhancer for Safe Reinforcement Learning Algorithms Based on Reduced Order Markov Decision Process Model
Zhehua Zhou, Xuan Xie, Jiayang Song, and 2 more authors
IEEE Transactions on Neural Networks and Learning Systems, 2024
@article{10766903,author={Zhou, Zhehua and Xie, Xuan and Song, Jiayang and Shu, Zhan and Ma, Lei},journal={IEEE Transactions on Neural Networks and Learning Systems},title={GenSafe: A Generalizable Safety Enhancer for Safe Reinforcement Learning Algorithms Based on Reduced Order Markov Decision Process Model},year={2024},publisher={IEEE},pages={1-15},}
