Huitao (David) Yang

Hello, I am Huitao.

I study foundation models, focusing on interpreting their behavior through theory and controlled experiments. I am also passionate about topics in reinforcement learning and probability.

I am currently a Master’s student in Statistics at UCLA, advised by Prof. Guang Cheng. I completed my undergraduate degree in Mathematics & Applied Mathematics at Fudan University. I am grateful to Prof. Guanting Chen, Prof. Andrea Zanette, and Prof. Zengfeng Huang for their mentorship and support.

Outside of academia, I enjoy hiking, biking, playing soccer, and diving into board games.

featured papers

Let the Target Select for Itself: Data Selection via Target-Aligned Paths

Huitao Yang, Hengzhi He, and Guang Cheng

arXiv, 2026

arXiv PDF BibTeX

Rethinks the reference path used for data influence in targeted data selection.

@misc{yang2026lettargetselectitself,
  title = {Let the Target Select for Itself: Data Selection via Target-Aligned Paths},
  author = {Yang, Huitao and He, Hengzhi and Cheng, Guang},
  year = {2026},
  eprint = {2605.09404},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2605.09404},
  tldr = {Rethinks the reference path used for data influence in targeted data selection.}
}

In-Context Curiosity: Distilling Exploration for Decision-Pretrained Transformers on Bandit Tasks

Huitao Yang and Guanting Chen

arXiv, 2025

arXiv PDF BibTeX

A study of how decision-pretrained transformers can internalize exploration behavior for bandit problems.

@misc{yang2025incontextcuriositydistillingexploration,
  title = {In-Context Curiosity: Distilling Exploration for Decision-Pretrained Transformers on Bandit Tasks},
  author = {Yang, Huitao and Chen, Guanting},
  year = {2025},
  eprint = {2510.00347},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2510.00347},
  tldr = {A study of how decision-pretrained transformers can internalize exploration behavior for bandit problems.}
}

Greedy Selection under Independent Increments: A Toy Model Analysis

Huitao Yang

arXiv, 2025

arXiv PDF BibTeX

A toy probability model showing when greedy selection succeeds under independent-increment structure.

@misc{yang2025greedyselectionindependentincrements,
  title = {Greedy Selection under Independent Increments: A Toy Model Analysis},
  author = {Yang, Huitao},
  year = {2025},
  eprint = {2506.17941},
  archiveprefix = {arXiv},
  primaryclass = {math.PR},
  url = {https://arxiv.org/abs/2506.17941},
  tldr = {A toy probability model showing when greedy selection succeeds under independent-increment structure.}
}

Fast Best-of-N Decoding via Speculative Rejection

Hanshi Sun, Momin Haider, Ruiqi Zhang, Huitao Yang, Jiahao Qiu, Ming Yin, Mengdi Wang, Peter Bartlett, and Andrea Zanette

NeurIPS, 2024

arXiv PDF BibTeX

A speculative rejection scheme that accelerates best-of-N decoding while preserving high-quality sample selection.

@misc{sun2024fastbestofndecodingspeculative,
  title = {Fast Best-of-N Decoding via Speculative Rejection},
  author = {Sun, Hanshi and Haider, Momin and Zhang, Ruiqi and Yang, Huitao and Qiu, Jiahao and Yin, Ming and Wang, Mengdi and Bartlett, Peter and Zanette, Andrea},
  year = {2024},
  eprint = {2410.20290},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL},
  venue = {NeurIPS},
  url = {https://arxiv.org/abs/2410.20290},
  tldr = {A speculative rejection scheme that accelerates best-of-N decoding while preserving high-quality sample selection.}
}