Research
Vinh Tong develops efficient training and sampling methods for generative models. His work includes reducing gradient variance in equivariant diffusion models and learning time discretizations that accelerate diffusion sampling. He also contributes to symmetry-aware generation of molecular structures, connecting probabilistic methods with the geometric requirements of physical systems.
Selected publications
- SymDrift: One-Shot Generative Modeling under Symmetries 2026
- Rao-Blackwell Gradient Estimators for Equivariant Denoising Diffusion 2025
- Learning to Discretize Denoising Diffusion ODEs 2025
Information reviewed 26 September 2026 · University profile