cv
Education, research experience, selected projects, and academic service. Download the PDF using the icon above.
Research Interests
- Mechanistic interpretability of factual reasoning and knowledge editing
- Reinforcement learning for latent reasoning
- Multimodal understanding for coding tasks and learning agents
Education
-
2024 - Present M.Phil. and Ph.D. in Artificial Intelligence
HKUST (Guangzhou), Information Hub - GPA 4.1/4.3
- Full Postgraduate Student Scholarship
-
2020 - 2024 B.Sc. in Statistics, Honors Degree (Top 5%)
Chongqing University, HongShen Honor School -
2022 Summer School
University of California, Irvine
Research Experience
-
2026 - Present Ph.D. Student
HKUST (Guangzhou), LAI Lab and LARK Lab - Advised by Chengwei Qin and Zhijiang Guo
- LLM reasoning, mechanistic interpretability, and agents
-
2026 - Present Research Intern
Tencent Hunyuan - Supervised by Han Hu
- LLM reasoning and world models
-
2024 - Present M.Phil. Student
HKUST (Guangzhou), DI2 Lab and LARK Lab - Advised by Yutao Yue and Zhijiang Guo
-
2024 Research Intern
KAUST - Supervised by Di Wang
- Explainable AI and machine learning
Selected Projects
-
EMNLP 2026 SWITCH - Switchable Latent Reasoning with On-Policy Reinforcement Learning
- Learned boundary tokens make on-policy RL well-defined over hidden-state recurrence.
- Reached 79.3% on MATH-500 and 89.2% on GSM8K with Qwen3-8B.
-
ICLR 2026 ACE - Attribution-Controlled Knowledge Editing for Multi-hop Factual Recall
- Identified query-value neuron pathways in multi-hop factual recall and used them for knowledge editing.
-
TMLR CAT and ConceptGuard
- Developed concept-level backdoor attacks and the first defense tailored to them.
Honors and Awards
-
2024 - Honors Degree, Chongqing University (Top 5%)
-
2022 - Outstanding Student Scholarship, Chongqing University
- MIT Outstanding Machine Learning Project (Top 5%)
Academic Service
- Reviewer for ICLR, ICML, NeurIPS, CVPR, ICCV, ECCV, and ICRA.