Hongru Yang 杨鸿儒

headshot.jpg

I am a final year PhD candidate in Computer Science at The University of Texas at Austin, where I am extremely fortunate to be advised by Prof. Atlas Zhangyang Wang in the VITA research group. I am a visiting student at Princeton University from September 2023 to May 2025, hosted by Prof. Jason D. Lee. During my PhD study, I have worked closely with Prof. Yingbin Liang from The Ohio State University. Prior to my PhD study, I obtained my Bachelor degree in Statistics and Computer Science and Mathematics from University of Illinois Urbana-Champaign in 2019. My research interest mainly lies in deep learning theory and optimization.

I am looking for full-time opportunities.

selected publications

  1. Transformers Provably Learn Two-Mixture of Linear Classification via Gradient Flow
    Hongru Yang, Zhangyang Wang, Jason D. Lee, and Yingbin Liang
    The Thirteenth International Conference on Learning Representations, 2025
  2. Training Dynamics of Transformers to Recognize Word Co-occurrence via Gradient Flow Analysis
    Hongru Yang, Bhavya Kailkhura, Zhangyang Wang, and Yingbin Liang
    Advances in Neural Information Processing Systems, 2024
  3. Neural Networks with Sparse Activation Induced by Large Bias: Tighter Analysis with Bias-Generalized NTK
    Hongru Yang, Ziyu Jiang, Ruizhe Zhang, Yingbin Liang, and Zhangyang Wang
    Journal of Machine Learning Research, 2024
  4. Pruning before training may improve generalization
    Hongru Yang, Yingbin Liang, Xiaojie Guo, Lingfei Wu, and Zhangyang Wang
    Journal of Machine Learning Research, 2024
  5. On the neural tangent kernel analysis of randomly pruned neural networks
    Hongru Yang, and Zhangyang Wang
    In International Conference on Artificial Intelligence and Statistics, 2023