Towards a Unified Generative Model for Scarce Time Series with Domain Experts

2026-06-13

AI Summary

Existing time series generation methods exhibit limited generalization under data-scarce conditions. To address this challenge, this work proposes TimeMoDE, a novel framework that integrates diffusion transformers with a mixture-of-experts mechanism. TimeMoDE learns universal temporal representations through multi-domain pretraining and dynamically guides expert allocation and denoising via domain prompts and diffusion-step conditioning signals. This approach effectively resolves the expert routing instability caused by noisy tokens and establishes a new paradigm for few-shot time series generation. Extensive experiments demonstrate that TimeMoDE significantly outperforms current state-of-the-art methods across various low-data regimes, confirming its superior generative performance and strong generalization capability.

Abstract

Synthesizing realistic time series with generative models has wide-ranging applications in real-world scenarios. Despite recent progress, most existing methods are trained under the assumption of abundant training data, which substantially limits their effectiveness in data-scarce settings. In this paper, we propose TimeMoDE, a novel framework that integrates Diffusion Transformers with Mixture-of-Experts to exploit both domain adaptability and diffusion-stage awareness for time series generation under data scarcity. It is pre-trained on a large-scale collection of multi-domain datasets to extract domain-agnostic temporal representations and domain-specific information benefiting generalization during fine-tuning. We propose Domain Prompts to condition expert assignment for indistinguishable noised tokens, mitigating the limitations of capturing inter-dataset relationships. Moreover, we incorporate diffusion timestep signals to equip the experts with awareness of time series degradation variations, facilitating adaptive calibrate to stage-dependent denoising requirements. Extensive experiments demonstrate that TimeMoDE outperforms existing methods under diverse low-data settings. It establishes an innovative paradigm for advanced time series few-shot generation.

Problem

• time series generation
• data scarcity
• few-shot learning
• domain adaptation
• generative modeling

Innovation

• TimeMoDE
• Diffusion Transformers
• Mixture-of-Experts
• Domain Prompts
• Few-shot Time Series Generation