MG-TSD

Summary

MG-TSD is a multi-granularity time-series diffusion model for probabilistic forecasting. It uses coarse-grained time-series targets to guide intermediate denoising steps.

Official Artifacts

• Paper: https://arxiv.org/abs/2403.05751
• Code: https://github.com/Hundredl/MG-TSD

Role In The Wiki

MG-TSD is a diffusion-forecasting source, not a general synthetic-data generator. It is useful evidence for coarse-to-fine denoising in temporal data.

Related Pages

• Time-Series Generation
• TimeCraft
• Time-Series Foundation Models
• Time-Series Benchmark Hygiene