Nice to meet you! I am Xinyue Fang (方馨悦) , currently a Master of Computer Science student in the College of Computer Science and Technology at the National University of Defense Technology (NUDT) , supervised by Prof. Zhiliang Tian (田植良) and Prof. Zhen Huang (黄震). Before joining NUDT, I received my B.E. degree in Network Engineering from the School of Computer Science and Technology at Harbin University of Science and Technology (HRBUST) in 2024 (ranked 2nd out of 115 students).
I focus on building trustworthy AI systems and enhancing model interpretability to enable the safe deployment of AI in real-world applications. My current research interests include:
- Hallucination in LLMs: detecting and mitigating hallucinations in complex settings, such as long-form generation, multiple-solution scenarios, and reasoning-chain hallucinations.
- Large Reasoning Models: understanding the internal reasoning mechanisms and behaviors of LRMs to further improve their reasoning capabilities, including mitigating overthinking and studying jailbreak attacks and defenses for reasoning processes.
- Game Theory and Collaboration among LLM-based Agents: investigating issues such as consensus hallucination, internal strategic behavior, and tool competition in multi-agent systems.
🔥 News
- 2026.04: 🎉🎉 One paper about hallucination mitigation in MoE models was accepted by ACL 2026!
- 2025.11: 🏆🏆 Honored to receive the National Scholarship (Top 1) at NUDT!
- 2025.09: 🎉🎉 One paper about sequential editing for continual knowledge updates was accepted by NeurIPS 2025!
- 2025.09: 🎉🎉 A survey paper on hallucination detection methods was accepted by Journal of Computer Research and Development!
- 2024.12: 🎉🎉 One paper about hallucination detection via contextual knowledge triples was accepted by AAAI 2025!
- 2024.06: 😎😎 I graduated from HRBUST and got the Outstanding Graduate Award of Heilongjiang Province!
📝 Publications
Knowledge Injection Exists in MoE? Exploring Expert-Aware Contrast Decoding in MoE for Mitigating LLMs’ Hallucinations
Xinyue Fang, Zhiliang Tian, Zhen Huang, Ziyi Pan, Zhihua Wen, Xi Wang, Quntian Fang, Dongsheng Li.
- Proposed an Expert-Aware Adaptive Contrastive Decoding (EAACD) method that leverages expert activation differences and reliability-aware contrastive decoding in MoE models to mitigate hallucinations without external resources or additional training.
Xinyue Fang, Zhen Huang, Zhiliang Tian, Minghui Fang, Ziyi Pan, Quntian Fang, Zhihua Wen, Hengyue Pan, Dongsheng Li.
- Proposed a graph-based context-aware hallucination detection method that leverages knowledge triple graphs and RGCN-based contextual consistency modeling to detect hallucinations in long-form text generation without external resources.
Hippocampal-like Sequential Editing for Continual Knowledge Updates in Large Language Models
Quntian Fang, Zhen Huang, Zhiliang Tian, Minghao Hu, Dongsheng Li, Yiping Yao, Xinyue Fang, Menglong Lu, Guotong Geng.
- Proposed a Hippocampal-like Sequential Editing (HSE) framework that leverages machine unlearning, Fisher Information Matrix-guided updates, and parameter replay to enable continual knowledge updates in LLMs without catastrophic forgetting or model collapse.
A Survey on Hallucination Detection Methods for Large Language Models
Zituo Li, Jianbin Sun, Guangzhou Chen, Xinyue Fang, Ruijing Cui, Zhiliang Tian, Zhen Huang, Kewei Yang.
- Provide a comprehensive survey of hallucination detection methods for large language models, systematically categorizing them into white-box and black-box approaches based on model transparency and practical application requirements.
Papers Under Review
- HUMAD: Hypergraph-Based Multi-View Fusion for Multi-Answer Hallucination Detection in LLMs. Ziyi Pan, Zhiliang Tian, Zhen Huang, Xinyue Fang, Yuquan Shu, Jingyuan Huang, Zhihua Wen, Linbo Qiao, Huaping Hu.
📖 Education
- 2024.09 - 2027.07 (Expected), Master of Computer Science. National University of Defense Technology (NUDT), Changsha, China.
- Supervisors: Prof. Zhiliang Tian & Prof. Zhen Huang.
- GPA: 3.11/4.00
- 2020.09 - 2024.07, Bachelor of Network Engineering. Harbin University of Science and Technology (HRBUST), Harbin, China.
- GPA: 4.55/5.00 (Rank 2/115)
🎖 Selected Honors and Awards
- 2025 🏆 National Scholarship (Top 1)
- 2025 🌟 Outstanding Student Award of School of Computer Science in NUDT (Top 10%)
- 2024 🎓 Outstanding Graduate of Heilongjiang Province (Top 1%)
- 2023 🥇 First Prize of the Northeast China Mathematical Modeling Competition
- 2023 💡 Utility Model Patent: “Photovoltaic Power Plant Cleaning Vehicle” - First Inventor
- 2022 🥈 Provincial Second Prize of China Undergraduate Mathematical Contest in Modeling (CUMCM)
- 2021 & 2022 🏅 National Encouragement Scholarship (Twice)
- 2021 & 2022 🎖️ HRBUST First-class scholarship (Twice)
- 2021 & 2022 🌟 HRBUST Merit Student Award (Twice)
- 2021 🥇 Provincial First Prize of China Undergraduate Mathematical Contest in Modeling (CUMCM)
🌟 Extra-Curricular Activities
- 2025.09 - 2026.02, Teaching Assistant, Computer Graphics Course of NUDT.
- Assisted instructors in mentoring students and designing course experiments.
- 2021.09 - 2023.09, Team Leader, Mathematical Modeling Competitions.
- Led a team to win two provincial first prizes and one provincial second prize.
- 2022.09 - 2023.09, Member, HRBUST Robot Club.
- Published a utility model patent as the first inventor.
🏕️ Hobbies & Contact
- Reading: I am a big fan of mystery and suspense fiction, as well as psychology-related literature, which constantly sharpens my logical reasoning and empathy.
- Outdoors: I enjoy hiking and mountaineering during my spare time to embrace nature and refresh my mind.
- Collaboration: I firmly believe that collaboration leads to win-win outcomes, and I constantly strive to be a trustworthy partner. If you are interested in my research topics or expecting any forms of collaboration, please feel free to contact me! 😊