LGAIMLFeb 18, 2021

Improved Denoising Diffusion Probabilistic Models

arXiv:2102.09672v15412 citationsHas Code
AI Analysis

This work addresses practical deployment issues for generative models, offering incremental improvements in efficiency and scalability for researchers and practitioners.

The paper tackles improving denoising diffusion probabilistic models (DDPMs) by making simple modifications to achieve competitive log-likelihoods and high sample quality, while enabling sampling with an order of magnitude fewer forward passes and negligible quality loss.

Denoising diffusion probabilistic models (DDPM) are a class of generative models which have recently been shown to produce excellent samples. We show that with a few simple modifications, DDPMs can also achieve competitive log-likelihoods while maintaining high sample quality. Additionally, we find that learning variances of the reverse diffusion process allows sampling with an order of magnitude fewer forward passes with a negligible difference in sample quality, which is important for the practical deployment of these models. We additionally use precision and recall to compare how well DDPMs and GANs cover the target distribution. Finally, we show that the sample quality and likelihood of these models scale smoothly with model capacity and training compute, making them easily scalable. We release our code at https://github.com/openai/improved-diffusion

Code Implementations18 repos

Data from Papers with Code (CC-BY-SA-4.0)

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes