When Denoising Hurts: Rethinking the Terminal Step of Diffusion Time Series Forecasters -- Extended Version

A new arXiv study (2608.14067v1) finds that in diffusion-based time series forecasting, continued low-noise refinement can introduce statistical drift and degrade forecasts, whereas general temporal structure is recovered at higher noise levels. The authors propose a label-free global stopping criterion and a Bernoulli timestep sampler to terminate inference earlier, improving accuracy and speeding up inference across eight real-world datasets.

arXiv:2608.14067v1 Announce Type: new Abstract: Diffusion models offer a natural way to model uncertainty in time series forecasting, yet their iterative sampling process is often treated as a uniformly beneficial refinement procedure. Our study challenges this view by examining how forecast quality evolves throughout reverse diffusion. We find that general temporal structure is often recovered at relatively high noise levels, whereas continued low-noise refinement can introduce statistical drift and degrade the final forecast. Our analysis further suggests that this behavior explains why prior methods often favor relatively narrow diffusion architecture and schedule design. Building on this observation, we propose a label-free global stopping criterion that detects the optimal termination point, eventually speeding up inference and improving predictive accuracy. Additionally, since early stopping terminates inference in high-noise regions, we propose a Bernoulli timestep sampler that concentrates training on this region while preserving coverage of the full diffusion process. Extensive experiments conducted across eight real-world datasets demonstrate the superior performance of our method compared to existing approaches.