Zeyi Xu

PhD candidate in Mathematics, University of California, Irvine

zeyi-picture.jpg

Department of Mathematics

University of California, Irvine

Irvine, CA 92697

zeyix1@uci.edu

I am a PhD candidate in the Department of Mathematics at the University of California, Irvine, advised by Professor Long Chen. Prior to UC Irvine, I received my B.S. in Mathematics from Fudan University in Shanghai, China. I design fast and reliable optimization algorithms for data science and machine learning. My work brings together three ideas that make first-order methods fast in practice (acceleration, adaptivity, and preconditioning) in a single framework built on continuous-time dynamics. I use it to speed up widely used algorithms, such as the Sinkhorn algorithm for optimal transport, with applications to Wasserstein gradient flows and generative modeling. You can read more on my research page. I am on the postdoc job market this year and would love to hear about opportunities. Feel free to reach out by email—I am always happy to chat about research and potential collaborations.

selected publications

  1. JOTA
    HNAG++: An Accelerated Gradient Method with a Refined Asymptotic Rate for Strongly Convex Optimization
    Long Chen and Zeyi Xu
    Journal of Optimization Theory and Applications, 2026
    DOI arXiv Bib HTML
  1. Preprint
    Acceleration of Generalized Conditional Gradient Methods
    Long Chen and Zeyi Xu
    arXiv preprint arXiv:2610.25473, 2026
    arXiv Bib
  2. Preprint
    Adaptive Accelerated Mirror Descent in Primal and Dual Spaces
    Zeyi Xu and Long Chen
    arXiv preprint arXiv:2606.02787, 2026
    arXiv Bib
  3. Preprint
    Accelerating Sinkhorn for Entropy-Regularized Optimal Transport
    Zeyi Xu and Long Chen
    arXiv preprint arXiv:2605.30267, 2026
    arXiv Bib
  1. CAMWA
    Deterministic Adam-Inspired Methods with Accelerated Convergence Rate
    Yaxin Yu, Long Chen, and Zeyi Xu
    Computers & Mathematics with Applications, 2026
    Accepted
    arXiv Bib

talks & conferences

  • SIAM MDS26 (upcoming) — SIAM Conference on Mathematics of Data Science, Salt Lake City, UT, November 16, 2026. Minisymposium talk: Accelerating Structured First-Order Methods: From Sinkhorn’s Algorithm to Conditional Gradient.
  • UC Irvine Applied & Computational Mathematics Seminar (upcoming) — University of California, Irvine, November 9, 2026. Seminar talk: Accelerating Structured First-Order Methods: From Sinkhorn’s Algorithm to Conditional Gradient.
  • SOCAMS 2026 — Southern California Applied Mathematics Symposium, Loyola Marymount University, Los Angeles, CA, April 25, 2026. Contributed talk: Accelerating Sinkhorn for Entropy-Regularized Optimal Transport.
  • SOCAMS 2025 — Southern California Applied Mathematics Symposium, University of California, Riverside, April 26, 2025. Contributed talk: Adaptive Accelerated Gradient Descent Methods for Convex Optimization.
  • NSF–CBMS Regional Research Conference — Deep Learning and Numerical PDEs, Morgan State University, Baltimore, MD, June 19–23, 2023.