UniTS

Summary

UniTS is a unified multi-task time-series model that uses task tokenization and one shared backbone for forecasting, classification, imputation, and anomaly detection.

Role In The Wiki

UniTS is the main requested example of a pretrained time-series model that tries to unify task interfaces, not just forecasting horizons. It is especially useful when comparing separate-head TSFMs with task-token and prompt-token designs.

Official Artifacts

• Project page: https://zitniklab.hms.harvard.edu/projects/UniTS/
• Official code: https://github.com/mims-harvard/UniTS
• Pretrained weights: https://github.com/mims-harvard/UniTS/releases/tag/ckpt

Evidence

• UniTS: A Unified Multi-Task Time Series Model

Relation To Foundation TSFM Agenda

Use the source-level agenda mapping in units-2024 rather than duplicating verdict rows here.

At the entity level, UniTS is the main requested example of a pretrained time-series model that tries to unify task interfaces, not just forecasting horizons. It is especially useful when comparing separate-head TSFMs with task-token and prompt-token designs. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.

Related Pages

• Foundation Time-Series Model Research Agenda
• Time-Series Foundation Models
• Time-Series Classification Foundation Models
• Time-Series Benchmark Hygiene