-
DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting
We introduce a novel diffusion model-based framework, DYffusion, for large-scale probabilistic forecasting. We propose to couple the diffusion steps with the physical timesteps of the data, leading to temporal forward and reverse processes that we represent through an interpolator and a forecaster network, respectively. DYffusion is faster than standard diffusion models during sampling, has low memory needs, and effectively addresses the challenges of generating stable, accurate and probabilistic rollout forecasts.