BRIDGE: Bootstrapping Text to Control Time-Series Generation via Multi-Agent Iterative Optimization and Diffusion Modeling

Hao Li, Yu-Hao Huang, Chang Xu, Viktor Schlegel, Renhe Jiang, Riza Batista-Navarro, Goran Nenadic, Jiang Bian

TL;DR: A text-controlled time series generation model and a LLM-based multi-agents framework for text instructions building.

Abstract:

Time-series Generation (TSG) is a prominent research area with broad applications in simulations, data augmentation, and counterfactual analysis. While existing methods have shown promise in unconditional single-domain TSG, real-world applications demand for cross-domain approaches capable of controlled generation tailored to domain-specific constraints and instance-level requirements.
In this paper, we argue that text can provide semantic insights, domain information and instance-specific temporal patterns, to guide and improve TSG. We introduce ``Text-Controlled TSG'', a task focused on generating realistic time series by incorporating textual descriptions.
To address data scarcity in this setting, we propose a novel LLM-based Multi-Agent framework that synthesizes diverse, realistic text-to-TS datasets. Furthermore, we introduce Bridge, a hybrid text-controlled TSG framework that integrates semantic prototypes with text description for supporting domain-level guidance. This approach achieves state-of-the-art generation fidelity on 11 of 12 datasets, and improves controllability by up to 12% on MSE and 6% MAE compared to no text input generation, highlighting its potential for generating tailored time-series data.

Lay Summary: Time series data—such as electricity usage, stock prices, or patient heart rates—are essential in fields like finance, energy, and healthcare. However, creating realistic synthetic time series that meet specific needs is a major challenge. Our research introduces a new method, called BRIDGE, that allows users to control time series generation using plain-language descriptions. Think of it as describing a pattern in words and receiving a matching time series in return. To do this, we designed a system where multiple AI agents work together to generate, evaluate, and refine these descriptions. We also developed a powerful model that learns both from text and from common time series patterns to generate high-quality, tailored outputs. BRIDGE not only outperforms existing methods on 11 of 12 benchmark datasets but also works well in new, unseen domains.

Primary Area: Applications->Time Series

Keywords: Multi-agents Systems, Time Series Generation, Large Language Models

Submission Number: 3188