AI and the World's Hard Problems · Mathematics · arXiv · 2026-10-20
Convergence of Denoising Diffusion Probabilistic Models: A R\'{e}nyi Divergence Perspective
This paper analyzes convergence of denoising diffusion probabilistic models (DDPMs) using the Rényi divergence, which is more sensitive to mismatch in low-probability regions than the measures used before. The total error splits into mixing, discretization, and score-estimation parts, with an iteration complexity bound for sub-Gaussian targets. The authors also extend the guarantee to critically damped DDPMs.