[2604.05064] Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series

Computer Science > Machine Learning arXiv:2604.05064 (cs) [Submitted on 6 Apr 2026] Title:Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series Authors:Annita Vapsi, Penghang Liu, Saheed Obitayo, Aakriti, Manoj Cherukumalli, Prathamesh Patil, Amit Varshney, Nicolas Marchesotti, Elizabeth Fons, Vamsi K. Potluru, Manuela Veloso View a PDF of the paper titled Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series, by Annita Vapsi and 10 other authors View PDF HTML (experimental) Abstract: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. Comments: Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2604.05064 [cs.LG]   (or arXiv:2604.05064v1 [cs.LG] for this version)   https://doi.o...