🧑🎓 关于我About Me
我是周晖林（Huilin Zhou），现为中国科学技术大学（USTC）与中国电信人工智能研究院（TeleAI）联合培养博士一年级学生，在赵健老师指导下，于李学龙教授团队开展研究。
我目前主要关注 可信与高效的智能体、多模态模型，以及文生图模型等方向，也对相关的基础问题与实际应用问题持续保持兴趣。到目前为止，我做过一些自演进越狱优化、多智能体安全评测、以及更安全的文生图生成等工作。更多信息可见 Google Scholar 与 ORCID。
I am Huilin Zhou (周晖林), a first-year Ph.D. student jointly trained by the University of Science and Technology of China (USTC) and the Institute of Artificial Intelligence, China Telecom (TeleAI). I conduct research under the supervision of Jian Zhao as a member of Prof. Xuelong Li’s team.
My current interests mainly lie in trustworthy and efficient agents, multimodal models, and text-to-image models, as well as related foundational and practical problems. So far, I have worked on self-evolving jailbreak optimization, multi-agent safety evaluation, and safer text-to-image generation. More information can be found on Google Scholar and ORCID.
🎉 动态News
- 2026 年 5 月 1 日. 🔥 Metis 与 RaGEP 被 ICML 2026（CCF-A） 接收为 regular papers。May 1, 2026. 🔥 Metis and RaGEP were accepted to ICML 2026 (CCF-A) as regular papers.
- 2025 年 12 月 5 日. 📄 TeleAI-Safety 在 arXiv 发布，聚焦 LLM 越狱攻击、防御与评测。December 5, 2025. 📄 TeleAI-Safety was released on arXiv for LLM jailbreak attacks, defenses, and evaluation.
📝 论文Selected Papers
Metis: Learning to Jailbreak LLMs via Self-Evolving Metacognitive Policy Optimization [arXiv]
Huilin Zhou, Jian Zhao, Yilu Zhong, Zhen Liang, Xiuyuan Chen, Tianle Zhang, Yuchen Yuan, Chi Zhang, Lan Zhang, Xuelong Li
International Conference on Machine Learning（ICML 2026），已接收 Regular Paper。International Conference on Machine Learning (ICML 2026), Accepted Regular Paper.
ACCORD: Training-Free Continuation Correction for Safe Text-to-Image Generation
Huilin Zhou, Jian Zhao, Wentao Hu, Yuhang Wang, Tianle Zhang, Lan Zhang, Xuelong Li
审稿中。Under review.
Decoupled Safety Control: A Safety-Control Algorithm for Training-Free Safety Guidance
Huilin Zhou, Ruoxi Cheng, Yuhang Wang, Yuming Liu, Minghao Sun, Ruolong Ma, Lan Zhang
审稿中。Under review.
RaGEP: Rank-aware Geometric Expert Pruning for Mixture-of-Experts Language Models
Wentao Hu, Zeyu Zhu, Mingkuan Zhao, Zhenhua An, Yanbo Zhai, Shanhong Yu, Huilin Zhou, Xin Lai, Xiaoyan Zhu, Jiayin Wang
International Conference on Machine Learning（ICML 2026），已接收 Regular Paper。International Conference on Machine Learning (ICML 2026), Accepted Regular Paper.
TeleAI-Safety: A comprehensive LLM jailbreaking benchmark towards attacks, defenses, and evaluations [arXiv]
Xiuyuan Chen, Jian Zhao, Yuxiang He, Yuan Xun, Xinwei Liu, Yanshu Li, Huilin Zhou, Wei Cai, Ziyan Shi, Yuchen Yuan, Tianle Zhang, Chi Zhang, Xuelong Li
arXiv 预印本 arXiv:2512.05485。arXiv preprint arXiv:2512.05485.
RADAR: A Risk-Aware Dynamic Multi-Agent Framework for LLM Safety Evaluation via Role-Specialized Collaboration [arXiv]
Xiuyuan Chen, Jian Zhao, Yuchen Yuan, Tianle Zhang, Huilin Zhou, Zheng Zhu, Linghe Kong, Chi Zhang, Weiran Huang, Xuelong Li
arXiv 预印本 arXiv:2509.25271。arXiv preprint arXiv:2509.25271.
CodeMimicry: Exploiting Safety Generalization Lag in Large Language Models via Structured Code Completion
Zhen Liang, Jian Zhao, Huilin Zhou, Tianle Zhang, Xuelong Li, Hai Huang
内部预印本。Internal preprint.
When Text Becomes Texture: Diagnosing Non-Overlay Text Recognition with FusedText
Yilu Zhong, Jian Zhao, Huilin Zhou, Tianle Zhang, Shangquan Sun, Xuelong Li
审稿中。Under review.
TrustGuard: Persistent Trust-State Governance for Personalized LLM Agent Safety
Yuming Liu, Jian Zhao, Kai Wang, Huilin Zhou, Tianle Zhang, Lan Zhang, Xuelong Li
审稿中。Under review.
Ghost in the Pixels: Unveiling Visual-Semantic Priority Inversion in Large Vision-Language Models
Yuhang Wang, Jian Zhao, Tianle Zhang, Huilin Zhou, Rui Feng
审稿中。Under review.
🎓 学术活动与服务Academic Activities & Services
- 学术服务. ICML 2026 审稿人，🏅 金牌审稿人 · Top 25%。 Service. Reviewer for ICML 2026, 🏅 Gold Reviewer · Top 25%.
- 学术活动. 担任 UbiComp 2026 TCSAUC Workshop Chair，CCF-A。 Activity. UbiComp 2026 TCSAUC Workshop Chair, CCF-A.
🏆 荣誉与奖项Honors & Awards
- 🏆2025. 第六届中国人工智能大赛获 2A1B：在“大模型幻觉挑战赛”与“智能图纸审查技术赛”获 A 级证书（最高级别），在“大模型对抗赛”获 B 级证书。 🏆2025. Received 2A1B in the 6th China Artificial Intelligence Competition: A-level certificates in the Large Model Hallucination Challenge and Intelligent Drawing Review Challenge, and a B-level certificate in the Large Model Adversarial Challenge.
- 🎓2025. 中国科学技术大学博士入学一等奖学金。 🎓2025. First-Class Ph.D. Entrance Scholarship, University of Science and Technology of China.
- 🏅2023-2024, 2024-2025. 连续两年度国家奖学金。 🏅2023-2024, 2024-2025. National Scholarship for two consecutive academic years.
- 🌟2025. 北京市优秀毕业生。 🌟2025. Beijing Outstanding Graduate.
🎓 教育经历Education
- 2025.09 – 至今，博士研究生，中国科学技术大学（USTC），合肥2025.09 – Present, Ph.D. Student, University of Science and Technology of China (USTC), Hefei, China
- 2025.09 – 至今，联合培养博士生，中国电信人工智能研究院（TeleAI），北京2025.09 – Present, Joint Ph.D. Researcher, Institute of Artificial Intelligence, China Telecom (TeleAI), Beijing, China