Yuekun Yao

I am Yuekun Yao, a postdoctoral researcher at The Ohio State University, where I work with Prof. Huan Sun. I completed my Ph.D. in the Department of Language Science and Technology at Saarland University, advised by Prof. Alexander Koller.

My research asks how neural networks learn and generalize tasks. Lately, I study the limitations of transformers, and how data, training recipes, and model architectures shape these behaviors. Areas I am interested in include interpretability, LLM knowledge, and reasoning. Before that, I worked on semantic parsing, with a focus on out-of-distribution and compositional generalization.

I am happy to collaborate on research projects. Please don’t hesitate to reach out if you are interested!

selected publications

Preprint

Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers

Harsh Kohli, Srinivasan Parthasarathy, Huan Sun, and 1 more author

2026

arXiv preprint

arXiv Bib PDF

@misc{kohli2026loopthinkgeneralize,
  title = {Loop, Think, \& Generalize: Implicit Reasoning in Recurrent-Depth Transformers},
  author = {Kohli, Harsh and Parthasarathy, Srinivasan and Sun, Huan and Yao, Yuekun},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL},
  note = {arXiv preprint},
  url = {https://arxiv.org/abs/2604.07822},
}

EMNLP

Language models can learn implicit multi-hop reasoning, but only if they have lots of training data

Yuekun Yao, Yupei Du, Dawei Zhu, and 2 more authors

In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025

arXiv Bib PDF

@inproceedings{yao2025implicit,
  title = {Language models can learn implicit multi-hop reasoning, but only if they have lots of training data},
  author = {Yao, Yuekun and Du, Yupei and Zhu, Dawei and Hahn, Michael and Koller, Alexander},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
}

EMNLP

Structural generalization is hard for sequence-to-sequence models

Yuekun Yao and Alexander Koller

In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, 2022

Bib PDF Code Website

@inproceedings{yao2022structural,
  title = {Structural generalization is hard for sequence-to-sequence models},
  author = {Yao, Yuekun and Koller, Alexander},
  booktitle = {Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing},
  year = {2022},
}