👋 About Me
Hi! I am Jiayue Pu (浦嘉越), currently a PhD student at the Gaoling School of Artificial Intelligence, Renmin University of China, co-advised by Prof. Jun Xu and Prof. Xiao Zhang. Prior to this, I received my bachelor’s degree in Computer Science from the University of Chinese Academy of Sciences (UCAS), where I was fortunate to be supervised by Prof. Xueqi Cheng and Prof. Fei Sun at the Institute of Computing Technology (ICT, CAS). I also spent time as a visiting student at UC Berkeley, working on LLM agent hallucinations under the guidance of Dr. Yiyou Sun in Prof. Dawn Song’s group.
My primary research interests lie in Trustworthy AI Agents and AI for Science (AI4Science).
📖 Educations
- 2026.09 - Present, PhD in Artificial Intelligence, Gaoling School of Artificial Intelligence, Renmin University of China.
- 2022.09 - 2026.08, B.S. in Computer Science, School of Computer Science and Technology, University of Chinese Academy of Sciences.
- 2025.01 - 2025.05, Visiting Student in EECS, University of California, Berkeley.
📝 Publications
HomeSafe-Bench: Evaluating Vision-Language Models on Unsafe Action Detection for Embodied Agents in Household Scenarios
Jiayue Pu*, Zhongxiang Sun*, Zilu Zhang*, Xiao Zhang, Jun Xu
Gaoling School of Artificial Intelligence, Renmin University of China
(* Equal contribution)
| [Project Page] | [Paper] | [Code] | [Dataset] | [Leaderboard] |
We introduce HomeSafe-Bench, a challenging benchmark designed to evaluate Vision-Language Models (VLMs) on unsafe action detection in household scenarios, featuring 438 diverse cases across six functional areas with fine-grained multidimensional annotations. We also propose HD-Guard, a hierarchical streaming architecture for real-time safety monitoring that coordinates a lightweight FastBrain for continuous high-frequency screening with an asynchronous SlowBrain for deep multimodal reasoning.
MIRAGE-Bench: LLM Agent is Hallucinating and Where to Find Them
Weichen Zhang*, Yiyou Sun*, Pohao Huang†, Jiayue Pu†, Heyue Lin†, Dawn Song
University of California, Berkeley
(* Equal contribution, † Equal second contribution)
| [Paper] | [Code & Data] |
We present MIRAGE-Bench, the first unified benchmark for eliciting and evaluating hallucinations in interactive LLM-agent scenarios. It introduces a three-part taxonomy and adopts a fine-grained-level LLM-as-a-Judge paradigm with tailored risk-aware prompts to systematically assess agent reliability in dynamic environments.
A Survey on Unlearning in Large Language Models
Ruichen Qiu, Jiajun Tan, Jiayue Pu, Honglin Wang, Xiao-Shan Gao, Fei Sun
Institute of Computing Technology, Chinese Academy of Sciences
This comprehensive survey provides a systematic review of over 180 papers on LLM unlearning published since 2021. We introduce a novel taxonomy categorizing unlearning methods and offer a multidimensional analysis of evaluation paradigms, datasets, and metrics, providing crucial insights for developing more effective unlearning techniques in large language models.
💻 Internships
- 2023.09 - Present, Institute of Computing Technology, Chinese Academy of Sciences,Research Intern, advised by Prof. Fei Sun.
- 2024.09 - 2025.05, Berkeley AI Safety Initiative, group members.
- 2024.09 - 2025.05, University of California, Berkeley, Research Collaborator, advised by Postdoc. Yiyou Sun.
- 2025.08 - Present, Gaoling School of Artificial Intelligence, Renmin University of China, co-advised by Prof. XiaoZhang and Prof. Jun Xu.
🥇 Honors and Awards
- National Scholarship, Ministry of Education of P.R.China
- Awarded in November 2024
- Outstanding Triple-A Student, University of Chinese Academy of Sciences
- Awarded in June 2024
- Triple-A Student, University of Chinese Academy of Sciences
- Awarded in June 2023 and June 2024
- Outstanding Communist Youth League Member, University of Chinese Academy of Sciences
- Awarded in June 2023 and June 2024
- First Class Academic Scholarship, University of Chinese Academy of Sciences
- Awarded in October 2023
🤝 Meetings
- ACL(The Association for Computational Linguistics) 2024, Bangkok, Thailand
- EMNLP (Empirical Methods in Natural Language Processing) 2025, Su Zhou, China
👩 Public Affairs
- Peer Mentor of Class 2412
- Principal Cellist of the UCAS Philharmonic Orchestra
- President of the Undergraduate Choir, University of Chinese Academy of Sciences
- Officer of the Student Societies Center, Undergraduate Student Union, University of Chinese Academy of Sciences
- Vice Class Representative, Class of 2206, Computer Science and Technology, University of Chinese Academy of Sciences