DiffSTG: Probabilistic Spatio-Temporal Graph Forecasting with Denoising Diffusion Models

TL;DR

This study presents the first attempt to generalize the popular de-noising diffusion probabilistic models to STGs, leading to a novel non-autoregressive framework called DiffSTG, along with the first denoising network UGnet for STG in the framework.

Abstract

Spatio-temporal graph neural networks (STGNN) have emerged as the dominant model for spatio-temporal graph (STG) forecasting. Despite their success, they fail to model intrinsic uncertainties within STG data, which cripples their practicality in downstream tasks for decision-making. To this end, this paper focuses on probabilistic STG forecasting, which is challenging due to the difficulty in modeling uncertainties and complex ST dependencies. In this study, we present the first attempt to generalize the popular denoising diffusion probabilistic models to STGs, leading to a novel non-autoregressive framework called DiffSTG, along with the first denoising network UGnet for STG in the framework. Our approach combines the spatio-temporal learning capabilities of STGNNs with the uncertainty measurements of diffusion models. Extensive experiments validate that DiffSTG reduces the Continuous Ranked Probability Score (CRPS) by 4%-14%, and Root Mean Squared Error (RMSE) by 2%-7% over existing methods on three real-world datasets.

Figures (10)

• Figure 1: Illustration of probabilistic STG forecasting. (a): Spatio-temporal graph forecasting; (b): Motivation of probabilistic prediction.
• Figure 2: Illustration of proposed DiffSTG and denoising network UGnet. The inference process of DiffSTG utilizes the trained denoising function ${\bm{\epsilon}}_{\theta}$ (i.e., UGnet) to sample ${\bm x}^{\rm all}_{n-1}$ step by step, under the guidance of ${\bm x}^{\rm all}_{\rm msk}$ and ${\mathcal{G}}$. UGnet leverages an Unet-based architecture to capture multi-scale temporal dependencies and the Graph Neural Network (GNN) to model spatial correlations.
• Figure 3: Overview of different models.
• Figure 4: Ablation study.
• Figure 5: Example of probabilistic spatio-temporal graph forecasting for air quality and traffic dataset.