Reproduce “Poisson Flow Generative Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yilun Xu, Ziming Liu, Max Tegmark, Tommi Jaakkola
Researchers at MIT introduce Poisson Flow Generative Models (PFGM), a physics-inspired framework that interprets data as electrical charges in an augmented space, leveraging the Poisson equation for generative modeling. This approach achieves state-of-the-art sample quality among normalizing flows while delivering 10x to 20x faster and more robust sampling compared to existing diffusion models.
Key insights
- Interpreting data points as electrical charges, whose flow along electric field lines transforms them into a uniform distribution on a hemisphere, provides a powerful and invertible generative mechanism.
- Normalizing the Poisson field makes its dynamics stable and learnable by a neural network, overcoming stiffness challenges.
- An anchored backward ODE formulation with a logarithmic time variable enables highly efficient and robust sampling by directly integrating from a known prior to the data manifold.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.