Critical Windows: Non-Asymptotic Theory for Feature Emergence in Diffusion Models

Published in International Conference on Machine Learning, 2024

We provide non-asymptotic analyses that identify precise timesteps where semantic features emerge during diffusion sampling.

Recommended citation: Marvin Li and Sitan Chen. (2024). "Critical Windows: Non-Asymptotic Theory for Feature Emergence in Diffusion Models." Proceedings of the 41st International Conference on Machine Learning (ICML).
Download Paper