I am a Ph.D. student in the Interdisciplinary Graduate Programme at Nanyang Technological University (NTU), advised by Prof. Miao Chun Yan. Before joining NTU, I worked as a Research Engineer at Singapore Management University (SMU) with Prof. Lizi Liao. I received my master’s degree in Electronic Information (Artificial Intelligence) from Peking University (PKU), advised by Prof. Houfeng Wang, and my bachelor’s degree in Computer Science from Wuhan University (WHU).
My research focuses on building reliable and adaptive large language model systems for real-world interaction. I am particularly interested in:
- LLM evaluation and reliability, including calibration, robustness, and iterative stability;
- efficient model adaptation and post-training, including parameter-efficient fine-tuning and post-hoc model refinement; and
- reasoning in structured and interactive environments, such as spreadsheets, coding workflows, and multi-agent systems.
My work has appeared at NeurIPS, COLM, COLING, and NLPCC. Our work Sheetpedia was selected as a NeurIPS 2025 Datasets and Benchmarks Spotlight.
🔥 News
- [Aug. 2026] Our paper Overconfidence in LLM-as-a-Judge: Diagnosis and Confidence-Driven Solution was accepted to NLPCC 2026.
- [Jul. 2026] Our paper Fragility Under Pressure was accepted to COLM 2026.
- [Mar. 2026] We released Spectral Surgery, a training-free method for refining trained LoRA adapters.
- [2025] Sheetpedia was accepted to NeurIPS 2025 Datasets and Benchmarks as a Spotlight.
📝 Selected Publications
Sheetpedia: A 300K-Spreadsheet Corpus for Spreadsheet Intelligence and LLM Fine-Tuning
Zailong Tian, Zhuoheng Han, Houfeng Wang, Lizi Liao
A large-scale corpus of approximately 290K real-world spreadsheets, together with benchmarks for semantic-range understanding and formula generation.
Zhuoheng Han, Jin-Ge Yao, Zailong Tian, Xi Yang, Houfeng Wang
A study of whether LLMs remain stable and effective across repeated feedback-and-revision cycles in constrained coding environments.
Spectral Surgery: Training-Free Refinement of LoRA via Gradient-Guided Singular Value Reweighting
Zailong Tian, Yanzhe Chen, Zhuoheng Han, Lizi Liao
A training-free post-hoc method that improves trained LoRA adapters by reweighting singular components using gradients from a small calibration set.
Overconfidence in LLM-as-a-Judge: Diagnosis and Confidence-Driven Solution
Zailong Tian*, Zhuoheng Han*, Yanzhe Chen, Haozhe Xu, Xi Yang, Richeng Xuan, Houfeng Wang, Lizi Liao
A systematic diagnosis of confidence miscalibration in LLM judges and a confidence-aware fusion framework for more reliable evaluation.
* Equal contribution.
🎓 Education
-
Nanyang Technological University, Singapore
Ph.D. Student, Interdisciplinary Graduate Programme, Aug. 2026 – Present \n Advisor: Prof. Miao Chun Yan -
Peking University, Beijing, China
M.Eng. in Electronic Information (Artificial Intelligence), Sep. 2022 – Jul. 2025
Advisor: Prof. Houfeng Wang -
Wuhan University, Wuhan, China
B.Sc. in Computer Science and Technology, Sep. 2018 – Jul. 2022
💼 Research Experience
-
Research Engineer, Singapore Management University, Singapore
Aug. 2025 – Jul. 2026
Research on LLM evaluation, reliability, efficient adaptation, and multi-agent systems with Prof. Lizi Liao. -
Research Intern, Beijing Academy of Artificial Intelligence, Beijing, China
Oct. 2024 – Jul. 2025 -
Research Intern, Microsoft Research Asia, Data, Knowledge and Intelligence Group, Beijing, China
Aug. 2023 – Oct. 2024
Collaborated with the Microsoft Excel team on spreadsheet intelligence and large language model adaptation. -
Research Intern, Zhihu, Beijing, China
Apr. 2023 – Jul. 2023
Worked on evaluating the mathematical reasoning, commonsense reasoning, and general capabilities of large language models.