Reproduce “Sensitivity Analysis for 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
Christopher Scarvelis, Justin Solomon
The authors estimate how changing a diffusion model’s training distribution shifts samples without retraining. They differentiate the score function with respect to an infinitesimal distribution change and propagate that sensitivity through sampling. In image experiments, predictions correlate with sample changes after retraining and fine-tuning, outperforming entropic optimal transport; this local estimate does not guarantee predictions for large changes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.