Diffusion models show strong potential in probabilistic time series forecasting. However, fixed noise schedules can produce intermediate states that are hard to invert and a terminal state that deviates from the near noise assumption. A new paper introduces StaTS, a diffusion model that learns the noise schedule and the denoiser through alternating updates.

StaTS Architecture

StaTS includes a Spectral Trajectory Scheduler (STS) that learns a data adaptive noise schedule with spectral regularization to improve structural preservation and stepwise invertibility. It also includes a Frequency Guided Denoiser (FGD) that estimates schedule induced spectral distortion and uses it to modulate denoising strength for heterogeneous restoration across diffusion steps and variables.

Training and Performance

A two stage training procedure stabilizes the coupling between schedule learning and denoiser optimization. Experiments on multiple real world benchmarks show consistent gains, while maintaining strong performance with fewer sampling steps. The code is available on GitHub.