GenTS: Unified Generative Modeling for Multimodal Time Series Analysis
Year: 2026
Type: conference-paper
Abstract: Modeling multimodal time series has emerged as an important research direction, aiming to incorporate auxiliary modalities such as textual descriptions to enhance temporal understanding. However, existing methods often treat external modalities as supplementary features, limiting their applicability to specific tasks and resulting in high costs in terms of design, training, and maintenance. In this paper, we propose GenTS, a unified generative framework for multimodal time series analysis that integrates multiple downstream tasks within a single model. GenTS is jointly trained for text-conditioned time series generation and multimodal forecasting, enabling effective modeling of the joint distribution between time series and external modalities. This unified formulation supports diverse applications, including conditional generation, forecasting, and time series editing. Moreover, by incorporating efficient time series captioning, GenTS alleviates the challenge of multimodal data scarcity. Extensive experiments on diverse real-world datasets demonstrate the effectiveness and generality of our approach across multiple tasks.
Authors: Shuqi Gu, Yongxiang Zhao, Baoyu Jing, Kan Ren
Institutions: ShanghaiTech University, Meta (United States)
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225328302