Reproduce “Probability flow ODEs in score-based and reflected 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

Rama CONT

The authors analyze probability-flow ODEs (PF-ODEs) as deterministic diffusion samplers and distinguish density evolution from well-behaved trajectories. They link stability to unweighted velocity error and flow regularity, showing score loss alone can miss tail errors. At cutoff δ=10⁻², an exact-score flat-manifold example has normal RMS 0.1, inverse sensitivity 10, and codimension-8 volume factor 10⁻⁸; this is a toy endpoint tradeoff.

Summary by alphaXiv.

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