EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Latent Spectral Space Diffusion Models for Spatio-Temporal Predictive Learning

Ming Zhang, Jie Wang, Qiao Liang

IEEE Access · 2026 · 14 · 48114-48126

DOI: 10.1109/access.2026.3675874 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Ming Zhang und 2 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Spatio-temporal forecasting is a critical challenge in various domains such as weather prediction, traffic flow analysis, and environmental monitoring, where understanding both spatial and temporal dependencies is key to accurate predictions. In this paper, we introduce a novel framework, Latent Spectral Diffusion (LSD), which combines spectral representations with diffusion models to improve prediction accuracy and robustness. Specifically, LSD utilizes Discrete Fourier Transform (DFT) to map the input data into a latent spectral space, where both local and global dependencies can be more effectively captured. A U-Net-based diffusion model is then employed to simulate the dynamic evolution of these latent features over time. This two-step framework facilitates both short-term and long-term forecasting by capturing intricate spatio-temporal patterns while ensuring computational stability and scalability. Extensive experiments on both synthetic and real-world datasets show that LSD outperforms existing spatio-temporal forecasting methods across several performance metrics. The results demonstrate the potential of latent spectral spaces coupled with diffusion dynamics in addressing the challenges posed by spatio-temporal prediction tasks, particularly when dealing with long-term trends and high variability.

Zitieren

APA

Zhang, M., Wang, J., & Liang, Q. (2026). Latent Spectral Space Diffusion Models for Spatio-Temporal Predictive Learning. IEEE Access, 14, 48114-48126. https://doi.org/10.1109/access.2026.3675874

Vancouver

Zhang M, Wang J, Liang Q. Latent Spectral Space Diffusion Models for Spatio-Temporal Predictive Learning. IEEE Access. 2026;14:48114-48126. doi:10.1109/access.2026.3675874

BibTeX
@article{Zhang2026,
  author = {Zhang, Ming and Wang, Jie and Liang, Qiao},
  title = {Latent Spectral Space Diffusion Models for Spatio-Temporal Predictive Learning},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {48114--48126},
  doi = {10.1109/access.2026.3675874}
}
Kostenlos starten →