Yixiao HUANG

prof_pic.jpg

Berkeley, CA, United States

yixiaoh@berkeley.edu
How to pronounce?
My first name can be pronounced as “e-/ɕjɑʊ/”:

I’m Yixiao HUANG (Chinese name: 黄一笑) and I’m a first-year PhD student at UC Berkeley EECS supervised by Prof. Somayeh Sojoudi. I’m interested in establishing the theoretical foundations for generative models to enhance their efficiency and reliability.

I received my Bachelor’s degree in Informaton Engineering in June, 2023 from Department of Electrical Engineering, City University of Hong Kong. I was fortunate to work with Prof. Rosa CHAN who patiently guided me to start my research adventure. I also had a great time at University of Michigan, working with Prof. Samet Oymak on the theoretical foundations of self-attention and hybrid models.

news

Aug 1, 2025 Two new preprints are out!
Jun 13, 2025 Excited to share two new preprints:
Dec 9, 2024 Our paper on the power of hybrid architecture is accepted to AAAI 2025 as an oral presentation! You can find it at arXiv.

selected publications

* indicates equal contribution
  1. Preprint
    Reinforcement Learning for Flow-Matching Policies
    Samuel Pfrommer, Yixiao Huang, and Somayeh Sojoudi
    arXiv preprint arXiv:2507.15073, 2025
    arXiv
  2. Preprint
    Spooky Action at a Distance: Normalization Layers Enable Side-Channel Spatial Communication
    Samuel Pfrommer, George Ma, Yixiao Huang, and Somayeh Sojoudi
    arXiv preprint arXiv:2507.04709, 2025
  3. Preprint
    Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers
    Yixiao Huang*, Hanlin Zhu*, Tianyu Guo*, Jiantao Jiao, Somayeh Sojoudi, and 3 more authors
    2025
    arXiv
  4. Preprint
    OVERT: A Benchmark for Over-Refusal Evaluation on Text-to-Image Models
    Ziheng Cheng*, Yixiao Huang*, Hui Xu, Somayeh Sojoudi, Xuandong Zhao, and 2 more authors
    2025
    arXiv
  5. AAAI 2025
    On the Power of Convolution Augmented Transformer
    Mingchen Li, Xuechen Zhang, Yixiao Huang, and Samet Oymak
    AAAI Conference on Artificial Intelligence (AAAI), 2025
    Oral Presentation
    arXiv
  6. ICML 2024
    From Self-Attention to Markov Models: Unveiling the Dynamics of Generative Transformers
    M. Emrullah Ildiz, Yixiao Huang, Yingcong Li, Ankit Singh Rawat, and Samet Oymak
    International Conference on Machine Learning (ICML), 2024
    arXiv
  7. AISTATS 2024
    Mechanics of Next Token Prediction with Self-Attention
    Yingcong Li*, Yixiao Huang*, M. Emrullah Ildiz, Ankit Singh Rawat, and Samet Oymak
    International Conference on Artificial Intelligence and Statistics (AISTATS), 2024
    arXiv PDF
  8. ICML 2024
    Towards Efficient Training and Evaluation of Robust Models against ł_0 Bounded Adversarial Perturbations
    Xuyang Zhong, Yixiao Huang, and Chen Liu
    International Conference on Machine Learning (ICML), 2024
    arXiv