About me
Hi, I am Yu-Han (吳雨翰). I have been a PhD student in LPSM at Sorbonne University since September 2024, supervised by Gérard Biau, Claire Boyer and Pierre Marion. I am also part of the co-advised PhD student program at Google DeepMind, co-advised by Quentin Berthet and Romuald Elie.
News
- New preprint: Kastor, an efficient fine-tuning strategy for generative emulation of PDE simulations, with colleagues at Google DeepMind.
- The DiffusionGemma technical report is out: an open-weight discrete-diffusion language model obtained by fine-tuning Gemma 4.
- Talk at MathSTIC (USPN) on Understanding diffusion models requires rethinking (again) generalization.
- Two new preprints: Understanding diffusion models requires rethinking (again) generalization (with Pierre Marion) and MIND, a sample-efficient alternative to FID.
- Optimal Stopping in Latent Diffusion Models accepted at ICML 2026.
Research focus
- Memorization and generalization in diffusion models Taking a Big Step · Rethinking generalization
- Sampling in latent diffusion models Optimal Stopping
- Evaluating generative models MIND
- Generative models for science and language Kastor · DiffusionGemma
- Implicit regularization in deep learning ResNets and neural ODEs
Selected publications
arXiv preprint · 2026
Kastor: An efficient fine-tuning strategy for generative emulation of PDE simulation
Turns a deterministic physics foundation model into a fast and accurate generative emulator, with a two-stage inference scheme and mean-prediction regularization, improving forecasts across The Well benchmark.
arXiv preprint · 2026
Understanding diffusion models requires rethinking (again) generalization
A position paper arguing that generalization in diffusion models needs new theory: memorization and generalization are incompatible, so the question is what a model learns before it memorizes.
COLT 2025
Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization
Shows that the large learning rates used in practice implicitly regularize denoising score matching and keep training away from the memorizing empirical optimal score.
Brief curriculum vitae
- 2024 – Sorbonne University and Google DeepMind, co-advised PhD student
- 2020 – 2024 Ecole Normale Supérieure
- 2022 – 2023 University Paris-Saclay, Master degree (Mathematics of Randomness)
- 2018 – 2020 Lycée Louis-le-Grand, preparatory classes
Internships
- 2024 Sorbonne University, Research Internship
- 2023 Owkin, Research Internship
- 2023 Sorbonne University, Research Internship
- 2022 Caltech, Research Internship