GARDiff: Graph-Aligned Residual Diffusion for Probabilistic Multivariate Time-Series Forecasting

2026-09-29

AI Summary

This study addresses the structural misalignment between deterministic graphs and residual dependencies in decoupled diffusion models by proposing the GARDiff framework. This method introduces a dynamic evolution mechanism to replace fixed graph conditioning, achieving progressive alignment from graph conditions to residual generation through uncertainty-aware refinement and timestep-wise edge sparsification techniques. Experiments conducted on six benchmark datasets demonstrate that GARDiff significantly improves probabilistic forecasting performance and uncertainty calibration accuracy. By effectively bridging the gap between static graph structures and dynamic generative processes, this work establishes a reliable structural alignment paradigm for graph-conditioned diffusion generation.

Abstract

Diffusion models have recently shown strong potential for probabilistic multivariate time-series forecasting by modeling complex conditional distributions. Recent decoupled diffusion frameworks further separate forecasting into deterministic prediction and stochastic residual generation, making it natural to derive dependency graphs from deterministic representations and use them to guide residual diffusion. However, we show that this direct structural transfer is unreliable. Although deterministic-derived graphs encode useful global dependency priors, they exhibit substantial edge-level misalignment with residual dependency structures, introducing inaccurate or redundant conditions during residual generation. This reveals a previously overlooked deterministic-to-residual structural alignment problem in decoupled diffusion forecasting. To address this problem, we propose GARDiff, a Graph-Aligned Residual Diffusion framework for probabilistic multivariate time-series forecasting. Instead of treating deterministic-derived graphs as fixed diffusion conditions, GARDiff progressively adapts them to residual generation. Specifically, GARDiff estimates residual uncertainty to distinguish high- and low-uncertainty regions, enabling uncertainty-aware structural refinement, and further performs timestep-aware edge sparsification during reverse diffusion to evolve graph conditions from broad dependency aggregation to localized residual refinement. Extensive experiments on six real-world benchmarks demonstrate that GARDiff consistently improves probabilistic forecasting performance and uncertainty calibration over strong baselines.

Problem

• probabilistic multivariate time-series forecasting
• decoupled diffusion
• structural alignment
• residual generation
• dependency graph

Innovation

• Graph-Aligned Residual Diffusion
• Probabilistic Multivariate Time-Series Forecasting
• Decoupled Diffusion Frameworks
• Uncertainty-Aware Structural Refinement
• Timestep-Aware Edge Sparsification