An Yan

I am an undergraduate researcher at the Institute for Interdisciplinary Information Sciences (IIIS, Yao Class), Tsinghua University. My research interests are LLM post-training, reasoning and agents, LLM systems.

I currently work with Prof. Hao Peng at the University of Illinois Urbana-Champaign on multi-task reinforcement learning and sequential on-policy distillation for LLM post-training. I study how task order, rollout-state distributions, and parameter-update geometry relate to cross-task transfer, interference, and long-horizon learning.

Previously, I worked with Prof. Hua Xu at Tsinghua University on LLM agents for large-scale mixed-integer linear programming and combinatorial optimization.

I am seeking Spring 2027 research opportunities in LLM post-training, reasoning, and systems.

Selected publication

• Process-Aware LNS for Large-Scale MILP via Context-Enhanced Fine-Tuned LLM-driven Selector. First author; accepted to NeurIPS 2026. The work studies process-aware, LLM-guided operator selection for large-scale mixed-integer optimization.

News

• 2026 — First-author paper accepted to NeurIPS 2026.
• 2026 — Started a research internship with Prof. Hao Peng on LLM post-training.
• 2025 — Our LLM-driven large-neighborhood-search work received an ICML 2025 Spotlight.