Hi, I’m Min Cai (蔡旻). I’m a PhD student at the University of Alberta, supervised by Dr. Xi Ye. Previously, I was an M.S. student graduated from Shenzhen University, where I was supervised by Prof. Haodi Zhang. Before that, I obtained my B.A. in Translation from Beijing Language and Culture University. I also work closely with Dr. Dan Zhang, Dr. Ziniu Hu, Dr. Shichang Zhang, and Dr. Difan Zou.

I have broad interests in ML and NLP, particularly in understanding the mechanisms behind neural language models, developing LLM agents capable of solving complex problems, and enhancing LLM reasoning. My current focus is on teaching LLMs to continually learn (e.g., MGS for resolving gradient conflicts and mitigating interference when learning across domains) — stay tuned for more work coming soon!

News

Jun 2025 Starting my PhD at the University of Alberta, advised by Dr. Xi Ye.
Apr 2025 How Post-Training Reshapes LLMs accepted to COLM 2025; also received Outstanding Paper at the New England NLP Workshop.
Jan 2025 Strategist accepted to ICLR 2025. Featured in the State of AI Report 2024.
Jun 2024 SelfControl accepted to the Mechanistic Interpretability Workshop & Foundation Models in the Wild Workshop at ICML 2024.
Oct 2023 AvalonBench accepted to the Foundation Models for Decision Making Workshop at NeurIPS 2023.

Experience

Oct 2025 – Mar 2026 Research Intern, Baidu.
Aug 2024 – Jul 2025 Research Intern, Z.ai.

Research

Agent
Evaluation
NeurIPS 2023 Workshop AvalonBench: Evaluating LLMs Playing the Game of Avalon
ACL 2026 Findings DataSciBench: An LLM Agent Benchmark for Data Science
Strategic Reasoning
ICLR 2025 Strategist: Learning Strategic Skills by LLMs via Bi-Level Tree Search
Mechanistic Interpretability
Post-Training Analysis
COLM 2025 How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence ★ Outstanding Paper — New England NLP Workshop
Behavior Control
ICML 2024 Workshop Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller
Training & Alignment
Reward Modeling
arXiv 2025 TDRM: Smooth Reward Models with Temporal Difference for LLM RL and Inference
Reasoning Models
arXiv 2026 Advancing General-Purpose Reasoning Models with Modular Gradient Surgery
EMNLP 2026 Tyler: Typed Latent Reasoning for Language Models — When to Think, What to Compute, and How Much to Allocate

Beyond Academics

In my spare time, I enjoy playing and listening to music — jazz, classical, R&B, and more. I also play the Pokémon Trading Card Game (PTCG) and love exploring new foods.