LLMs and post-training
RLHF, PPO variants, alignment, test-time reasoning, adaptive inference, long-context modeling, and domain-specific fine-tuning.
McGill University | Mila - Quebec AI Institute
Dun Yuan
I work on large language models for technical domains, with a focus on post-training, knowledge-grounded generation, telecommunications, networked systems, and distributed AI.
- Affiliation
- Mila - Quebec AI Institute
- Current focus
- Stable LLM fine-tuning and knowledge-grounded telecom systems
- Also
- Student Researcher, Samsung Research America
About
Language models, networks, and trustworthy systems.
I am a Ph.D. candidate in Computer Science at McGill University, supervised by Prof. Xue Liu, and affiliated with Mila - Quebec AI Institute. I study how language models can be trained, grounded, and deployed in technical environments where accuracy and reliability matter.
Recent projects include contraction-aware reinforcement learning for language model fine-tuning, knowledge-graph-enhanced retrieval for telecommunications, semantic alignment in agent communication protocols, and optimization for wireless and data-center systems.
Research Areas
Research themes
Knowledge-centric AI
Retrieval-augmented generation, knowledge graphs, provenance-aware question answering, information extraction, and grounded generation.
AI for telecom and systems
LLMs for wireless networks, telecom question answering, digital twins, multi-agent reinforcement learning, and edge intelligence.
Trustworthy distributed AI
Web3, decentralized systems, cryptographic protocols, secure data infrastructures, and privacy-aware distributed computing.
Featured Project
Beyond Message Passing
Project page for my survey on agent communication protocols, organized around communication, syntactic, and semantic layers.
Recent Publications
Recent work
2026
Escaping Policy Contraction: Contraction-Aware PPO (CaPPO) for Stable Language Model Fine-Tuning
Dun Yuan, Di Wu, Xue Liu. ICLR 2026 Poster.
OpenReview2026
Beyond Message Passing: A Semantic View of Agent Communication Protocols
Dun Yuan, Fuyuan Lyu, Ye Yuan, Weixu Zhang, et al. arXiv:2604.02369.
Project page2025
Enhancing Large Language Models for Telecommunications using Knowledge Graphs and Retrieval-Augmented Generation
Dun Yuan, Hao Zhou, Di Wu, Xue Liu, Hao Chen, Yan Xin, Jianzhong Charlie Zhang. ICC Workshops 2025.
DBLPFor the complete publication list, see Google Scholar.