OATS

Summary

OATS is an online synthetic-data augmentation method for time-series foundation-model pretraining. It uses time-series influence scores and guided diffusion to generate samples during training.

Official Artifacts

• Paper: https://arxiv.org/abs/2601.19040
• Code: https://github.com/microsoft/TimeCraft/tree/main/OATS

Role In The Wiki

OATS is a data-engine source for TSFM pretraining rather than a new forecasting backbone. Its main value is the online, training-stage-aware augmentation interface.

Related Pages

• TimeCraft
• Time-Series Generation
• Synthetic Data For Time Series
• Time-Series Foundation Models