Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914550786293760 |
|---|---|
| author | Vapsi, Annita Liu, Penghang Obitayo, Saheed Aakriti Cherukumalli, Manoj Patil, Prathamesh Varshney, Amit Marchesotti, Nicolas Fons, Elizabeth Potluru, Vamsi K. Veloso, Manuela |
| author_facet | Vapsi, Annita Liu, Penghang Obitayo, Saheed Aakriti Cherukumalli, Manoj Patil, Prathamesh Varshney, Amit Marchesotti, Nicolas Fons, Elizabeth Potluru, Vamsi K. Veloso, Manuela |
| contents | Synthetic data is essential for training foundation models for time series (FMTS), but most generators assume static correlations, and are typically missing realistic inter-channel dependencies. We introduce DynLMC, a Dynamic Linear Model of Coregionalization, that incorporates time-varying, regime-switching correlations and cross-channel lag structures. Our approach produces synthetic multivariate time series with correlation dynamics that closely resemble real data. Fine-tuning three foundational models on DynLMC-generated data yields consistent zero-shot forecasting improvements across nine benchmarks. Our results demonstrate that modeling dynamic inter-channel correlations enhances FMTS transferability, highlighting the importance of data-centric pretraining. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_05064 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series Vapsi, Annita Liu, Penghang Obitayo, Saheed Aakriti Cherukumalli, Manoj Patil, Prathamesh Varshney, Amit Marchesotti, Nicolas Fons, Elizabeth Potluru, Vamsi K. Veloso, Manuela Machine Learning Artificial Intelligence Synthetic data is essential for training foundation models for time series (FMTS), but most generators assume static correlations, and are typically missing realistic inter-channel dependencies. We introduce DynLMC, a Dynamic Linear Model of Coregionalization, that incorporates time-varying, regime-switching correlations and cross-channel lag structures. Our approach produces synthetic multivariate time series with correlation dynamics that closely resemble real data. Fine-tuning three foundational models on DynLMC-generated data yields consistent zero-shot forecasting improvements across nine benchmarks. Our results demonstrate that modeling dynamic inter-channel correlations enhances FMTS transferability, highlighting the importance of data-centric pretraining. |
| title | Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2604.05064 |