Reproduce “Scaling Riemannian Diffusion 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

Aaron Lou, Minkai Xu, Stefano Ermon

Improved heat kernel estimation and sampling techniques for Riemannian symmetric spaces enable Riemannian Diffusion Models to scale to high-dimensional tasks. The approach successfully models SU(3) lattice configurations and enhances out-of-distribution detection for hyperspherical embeddings, achieving competitive performance compared to state-of-the-art methods.

Key insights

  • Restricting the analysis to Riemannian symmetric spaces allows for more tractable and accurate heat kernel computations, as these spaces possess simplifying geometric properties like maximal tori.
  • A hybrid approach for heat kernel estimation, combining various approximations and expansions based on the diffusion time 't', ensures numerical stability and accuracy across the entire diffusion process.
  • Exact heat kernel sampling, particularly using probability flow ODEs, provides a robust method for high-dimensional manifolds by leveraging the simplified vector field dynamics on the maximal torus.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.