arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2604.05064cs.LGcs.AI
收藏

动态线性共区域化用于真实合成多变量时间序列

Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series

JPMorganChase(摩根大通)

Annita Vapsi, Penghang Liu, Saheed Obitayo, Aakriti, Manoj Cherukumalli, Prathamesh Patil, Amit Varshney, Nicolas Marchesotti, Elizabeth Fons, Vamsi K. Potluru, Manuela Veloso

AI总结:

本文提出DynLMC模型,通过引入时变和制度切换相关性及跨通道滞后结构,生成与真实数据相似的多变量时间序列,提升FMTS的迁移能力。

摘要:

合成数据对于训练时间序列基础模型(FMTS)至关重要,但大多数生成器假设静态相关性,通常缺乏真实跨通道依赖关系。我们引入了动态线性共区域化模型(DynLMC),该模型结合了时间变化、制度切换相关性和跨通道滞后结构。我们的方法生成具有与真实数据相似相关动态的合成多变量时间序列。在三个基础模型上微调DynLMC生成的数据,在九个基准测试中实现了一致的零样本预测改进。我们的结果表明,建模动态跨通道相关性增强了FMTS的迁移性,突显了数据驱动预训练的重要性。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑