Saved in:
Bibliographic Details
Main Authors: Ruhe, David, Heek, Jonathan, Salimans, Tim, Hoogeboom, Emiel
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.09470
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916385339211776
author Ruhe, David
Heek, Jonathan
Salimans, Tim
Hoogeboom, Emiel
author_facet Ruhe, David
Heek, Jonathan
Salimans, Tim
Hoogeboom, Emiel
contents Diffusion models have recently been increasingly applied to temporal data such as video, fluid mechanics simulations, or climate data. These methods generally treat subsequent frames equally regarding the amount of noise in the diffusion process. This paper explores Rolling Diffusion: a new approach that uses a sliding window denoising process. It ensures that the diffusion process progressively corrupts through time by assigning more noise to frames that appear later in a sequence, reflecting greater uncertainty about the future as the generation process unfolds. Empirically, we show that when the temporal dynamics are complex, Rolling Diffusion is superior to standard diffusion. In particular, this result is demonstrated in a video prediction task using the Kinetics-600 video dataset and in a chaotic fluid dynamics forecasting experiment.
format Preprint
id arxiv_https___arxiv_org_abs_2402_09470
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Rolling Diffusion Models
Ruhe, David
Heek, Jonathan
Salimans, Tim
Hoogeboom, Emiel
Machine Learning
Diffusion models have recently been increasingly applied to temporal data such as video, fluid mechanics simulations, or climate data. These methods generally treat subsequent frames equally regarding the amount of noise in the diffusion process. This paper explores Rolling Diffusion: a new approach that uses a sliding window denoising process. It ensures that the diffusion process progressively corrupts through time by assigning more noise to frames that appear later in a sequence, reflecting greater uncertainty about the future as the generation process unfolds. Empirically, we show that when the temporal dynamics are complex, Rolling Diffusion is superior to standard diffusion. In particular, this result is demonstrated in a video prediction task using the Kinetics-600 video dataset and in a chaotic fluid dynamics forecasting experiment.
title Rolling Diffusion Models
topic Machine Learning
url https://arxiv.org/abs/2402.09470