Yu He 何禹

Mudd Room 3020

2233 Tech Drive

Evanston, IL

YuHe2030@u.northwestern.edu

I am a first-year Ph.D. student in the Computer Science Theory group at Northwestern University, advised by Edith Elkind. I obtained my bachelor degree in Computer Science from City University of Hong Kong.

My research interests lie in algorithmic game theory, with a particular focus on understanding strategic behavior of agents across diverse environments—ranging from social choice problem, cooperative games to learning systems, especially large language models.

I feel very fortunate to have been advised by Alexander Lam and Minming Li during my time at CityU. In the final year of my undergraduate studies, I spent a year with Haifeng Xu’s group at the University of Chicago, where I benefited greatly from the generous guidance of Haifeng Xu and Fan Yao.

news

Apr 15, 2025	Starting Fall 2025, I will join Northwestern University’s Computer Science Theory Group as a PhD student.

selected publications

AAMAS

Facility Location Games with Scaling Effects

Yu He, Alexander Lam, and Minming Li

In Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2024), May 2024

arXiv Bib

@inproceedings{Yu2024aamas,
  author = {He, Yu and Lam, Alexander and Li, Minming},
  title = {Facility Location Games with Scaling Effects},
  booktitle = {Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2024)},
  editor = {Alechina, Natasha and Dignum, Virginia and Dastani, Mehdi and Sichman, J. S.},
  address = {Auckland, New Zealand},
  month = may,
  year = {2024},
  pages = {816--824},
  publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
  howpublished = {conference}
}

Arxiv

The Complexity of Tullock Contests

Yu He, Fan Yao, Yang Yu, Xiaoyun Qiu, Minming Li, and Haifeng Xu

Apr 2025

arXiv:2412.06444v2

arXiv Bib

@misc{he2025complexity,
  title = {The Complexity of Tullock Contests},
  author = {He, Yu and Yao, Fan and Yu, Yang and Qiu, Xiaoyun and Li, Minming and Xu, Haifeng},
  year = {2025},
  month = apr,
  eprint = {2412.06444},
  archiveprefix = {arXiv},
  primaryclass = {cs.GT},
  note = {arXiv:2412.06444v2},
  howpublished = {preprint}
}

Arxiv
How Sampling Shapes LLM Alignment: From One-Shot Optima to Iterative Dynamics

(αβ) Yurong Chen, Yu He, Michael I. Jordan, and Fan Yao

Apr 2026

Abs arXiv Bib

Standard methods for aligning large language models with human preferences learn from pairwise comparisons among sampled candidate responses and regularize toward a reference policy. Despite their effectiveness, the effects of sampling and reference choices are poorly understood theoretically. We investigate these effects through Identity Preference Optimization, a widely used preference alignment framework, and show that proper instance-dependent sampling can yield stronger ranking guarantees, while skewed on-policy sampling can induce excessive concentration under structured preferences. We then analyze iterative alignment dynamics in which the learned policy feeds back into future sampling and reference policies, reflecting a common practice of model-generated preference data. We prove that these dynamics can exhibit persistent oscillations or entropy collapse for certain parameter choices, and characterize regimes that guarantee stability. Our theoretical insights extend to Direct Preference Optimization, indicating the phenomena we captured are common to a broader class of preference-alignment methods. Experiments on real-world preference data validate our findings.
@misc{chen2026sampling, title = {How Sampling Shapes LLM Alignment: From One-Shot Optima to Iterative Dynamics}, author = {Chen, Yurong and He, Yu and Jordan, Michael I. and Yao, Fan}, year = {2026}, eprint = {2602.12180}, archiveprefix = {arXiv}, primaryclass = {cs.LG}, howpublished = {preprint}, ab = {true} }