Improved Sample Complexity Bounds for Diffusion Model Training

Neural Information Processing Systems 

Diffusion models have become the most popular approach to deep generative modeling of images, largely due to their empirical performance and reliability.