Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Guo, Xingzhuo, Zhang, Yu, Chen, Baixu, Xu, Haoran, Wang, Jianmin, Long, Mingsheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916639187927040
author Guo, Xingzhuo
Zhang, Yu
Chen, Baixu
Xu, Haoran
Wang, Jianmin
Long, Mingsheng
author_facet Guo, Xingzhuo
Zhang, Yu
Chen, Baixu
Xu, Haoran
Wang, Jianmin
Long, Mingsheng
contents Diffusion models have emerged as powerful generative frameworks by progressively adding noise to data through a forward process and then reversing this process to generate realistic samples. While these models have achieved strong performance across various tasks and modalities, their application to temporal predictive learning remains underexplored. Existing approaches treat predictive learning as a conditional generation problem, but often fail to fully exploit the temporal dynamics inherent in the data, leading to challenges in generating temporally coherent sequences. To address this, we introduce Dynamical Diffusion (DyDiff), a theoretically sound framework that incorporates temporally aware forward and reverse processes. Dynamical Diffusion explicitly models temporal transitions at each diffusion step, establishing dependencies on preceding states to better capture temporal dynamics. Through the reparameterization trick, Dynamical Diffusion achieves efficient training and inference similar to any standard diffusion model. Extensive experiments across scientific spatiotemporal forecasting, video prediction, and time series forecasting demonstrate that Dynamical Diffusion consistently improves performance in temporal predictive tasks, filling a crucial gap in existing methodologies. Code is available at this repository: https://github.com/thuml/dynamical-diffusion.
format Preprint
id arxiv_https___arxiv_org_abs_2503_00951
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models
Guo, Xingzhuo
Zhang, Yu
Chen, Baixu
Xu, Haoran
Wang, Jianmin
Long, Mingsheng
Machine Learning
Computer Vision and Pattern Recognition
Diffusion models have emerged as powerful generative frameworks by progressively adding noise to data through a forward process and then reversing this process to generate realistic samples. While these models have achieved strong performance across various tasks and modalities, their application to temporal predictive learning remains underexplored. Existing approaches treat predictive learning as a conditional generation problem, but often fail to fully exploit the temporal dynamics inherent in the data, leading to challenges in generating temporally coherent sequences. To address this, we introduce Dynamical Diffusion (DyDiff), a theoretically sound framework that incorporates temporally aware forward and reverse processes. Dynamical Diffusion explicitly models temporal transitions at each diffusion step, establishing dependencies on preceding states to better capture temporal dynamics. Through the reparameterization trick, Dynamical Diffusion achieves efficient training and inference similar to any standard diffusion model. Extensive experiments across scientific spatiotemporal forecasting, video prediction, and time series forecasting demonstrate that Dynamical Diffusion consistently improves performance in temporal predictive tasks, filling a crucial gap in existing methodologies. Code is available at this repository: https://github.com/thuml/dynamical-diffusion.
title Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.00951