Time-Series Foundation Models

Adapting multivariate time-series foundation models for forecasting and anomaly detection.

Foundation models for time series promise strong zero-/few-shot performance, but real deployments face limited historical data, distribution shift, and the need for early anomaly detection. This work develops pattern- and context-aware adaptation of time-series foundation models so they refine themselves to new domains and reliably anticipate anomalies in multivariate IT operations.

Representative publication (Mishra et al., 2025).

References

2025

NeurIPS-W

Adaptive Refinement of Time Series Foundation Models via Pattern and Context-Awareness

Ashish Mishra, Tarun Kumar, Satish Kumar Mopur, and 5 more authors

In BERT2S Workshop at NeurIPS, 2025

Bib

@inproceedings{mishra2025adaptive,
  title = {Adaptive Refinement of Time Series Foundation Models via Pattern and Context-Awareness},
  author = {Mishra, Ashish and Kumar, Tarun and Mopur, Satish Kumar and Serebryakov, Sergey and Bhattacharya, Suparna and Foltin, Martin and Vogety, Ramanagopal and Koganti, Phanidhar},
  booktitle = {BERT2S Workshop at NeurIPS},
  year = {2025},
}