Academic Activities

General Information

Full Name Chieh-Hsin (Jesse) LAI
Tags AI research scientist; Applied mathematician
Languages Mandarin, English, Japanese

Experience

  • 2024 - Present
    Visiting Assistant Professor
    Department of Applied Mathematics, National Yang Ming Chiao Tung University, Taiwan
    • Deep learning approaches for numerical methods and partial differential equations.
  • 2022 - Present
    Research Scientist
    Sony AI, Tokyo
    • Deep generative models (especially diffusion models), mathematically grounded deep learning, robustness.
  • 2021 - 2022
    Senior Research Engineer
    Sony USA
    • Robustness of deep learning.
  • 2015 - 2016
    Research Assistant
    Institute of Mathematics, Academia Sinica, Taiwan
    • Harmonic analysis and microlocal analysis of PDEs.

Education

  • 2021
    PhD in Mathematics
    University of Minnesota, Twin Cities
    • Theory and applications of deep learning, anomaly detection, robustness, generative models
      • Advisor. Gilad Lerman
      • Period. August 2018 – May 2021
    • Reaction-diffusion partial differential equations (PDEs), harmonic analysis
      • Advisor. Wei-Ming Ni
      • Period. June 2017 – June 2018
  • 2015
    Bachelor in Mathematics
    National Tsing Hua University, Taiwan

Events and Invited Talks

  • August 9, 2024
    Invited talks at Cognitive Developmental Robotics Lab at University of Tokyo
    • Introduction to Diffusion Models and Its Applications in Other Disciplines
  • April 26, 2024
    Invited talks at Crunch Seminar of Division of Applied Mathematics at Brown University
    • Exploring the Intersection of Diffusion Models and (Partial) Differential Equation Solving
    • YouTube Link
  • April 19, 2024
    Invited talks on Scientific Machine Learning at National Yang Ming Chiao Tung University (Part-II)
    • Diffusion Models, Insights into Their Connection with PDE solving
    • Webinar Page
  • March 29, 2024
    Invited talks on Scientific Machine Learning at National Yang Ming Chiao Tung University (Part-I)
  • Feb. 29, 2024
    Guest lecture at Duke Kushan University
    • Evolution of Diffusion Models, From Birth to Enhanced Efficiency and Controllability
  • Feb. 27, 2024
    Invited talk at Department of Mathematics, National Tsing Hua University, Taiwan
    • Theory of Diffusion Models
    • YouTube Link (in Mandarin). You can find a list of credits for the materials used in the video here (under construction...).
  • Feb. 23, 2024
    Invited talk at Appier, Taiwan
    • Evolution of Diffusion Models, Its Birth and Applications
  • Feb. 22, 2024
    Invited talk at Department of Mathematics, National Central University, Taiwan
    • Theory of Diffusion Models
  • Feb. 22, 2024
    Invited talk at Robotic Search Lab, National Central University, Taiwan
    • Evolution of Diffusion Models, Its Birth and Applications
  • Feb. 21, 2024
    Invited talk at Department of EE, National Taiwan University, Taiwan
    • Evolution of Diffusion Models, From Birth to Enhanced Efficiency and Controllability
  • Feb. 21, 2024
    Invited talk at NVIDIA, Taiwan
    • Evolution of Diffusion Models, From Birth to Enhanced Efficiency and Controllability
  • Feb. 20, 2024
    Presented at Learning on Graphs & Geometry (LoGG) reading group
    • Consistency Trajectory Models, Learning Probability Flow ODE Trajectory of Diffusion
    • YouTube Link. You can find a list of credits for the materials used in the video here (under construction...).
  • Oct. 21, 2019
    Presented at NSF ATD and AMPS workshop 2019, Washington D.C.
    • Robust Subspace Recovery Layer for Unsupervised Anomaly Detection

Academic Services

  • TBD, 2024
    Tutorial at ISMIR 2024
  • May 7, 2024
    Organized social event at ICLR 2024, Vienna
    • Recent advances on diffusion and GAN
    • ICLR Page
  • Dec. 10, 2023
    Organized Expo workshop at NeurIPS 2023, New Orleans
    • Media Content Restoration and Editing with Deep Generative Models and Beyond
    • NeurIPS Page
  • 2024
    IEEE TPAMI, ACM TKDD Reviewer
  • 2021-
    ICLR, ICML, NeurIPS Reviewer

Open Source Projects

  • 2024-
    Consistency Trajectory Model (CTM)
    • For single-step diffusion model sampling, CTM achieves SOTA on CIFAR-10 and ImageNet 64x64. CTM offers diverse sampling options and balances computational budget with sample fidelity effectively.
  • 2024-
    Slicing Adversarial Network (SAN)
    • A theoretically grounded and simple modification scheme for discriminators of almost any GANs to enhance GAN performance. Applying SAN to StyleGAN-XL results in SOTA performance on ImageNet 256x256.
  • 2023-
    FP-Diffusion
    • Improving density estimation of diffusion models by regularizing with the underlying equation describing the temporal evolution of scores, theoretically supported.
  • 2023-
    GibbsDDRM
    • Solving blind linear inverse problems by utilizing the pre-trained diffusion models in a Gibbs sampling manner.