DiffStyleTS: Diffusion Model for Style Transfer in Time Series

2025-10-13

AI Summary

To address the challenge of entangled content and style in time series—hindering data augmentation and scenario simulation—this paper proposes the first self-supervised attention-based diffusion framework for time-series style transfer. The method employs a dual-branch convolutional encoder to disentangle content and style representations, and introduces a conditional attention diffusion model to achieve interpretable, controllable style decoupling and recomposition. Its key innovation lies in enabling the first explainable and separable time-series style transfer in non-image domains without requiring paired annotations. Experiments demonstrate that the generated samples achieve significantly higher visual fidelity and lower Fréchet Inception Distance (FID) than baselines. When applied to anomaly detection via data augmentation, it improves F1-score by up to 18.7% in few-shot settings (<100 samples). This work establishes a novel paradigm for low-data time-series modeling.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Data transparency and provenance

Abstract

Style transfer combines the content of one signal with the style of another. It supports applications such as data augmentation and scenario simulation, helping machine learning models generalize in data-scarce domains. While well developed in vision and language, style transfer methods for time series data remain limited. We introduce DiffTSST, a diffusion-based framework that disentangles a time series into content and style representations via convolutional encoders and recombines them through a self-supervised attention-based diffusion process. At inference, encoders extract content and style from two distinct series, enabling conditional generation of novel samples to achieve style transfer. We demonstrate both qualitatively and quantitatively that DiffTSST achieves effective style transfer. We further validate its real-world utility by showing that data augmentation with DiffTSST improves anomaly detection in data-scarce regimes.

Problem

• Achieving style transfer in time series data
• Disentangling time series into content and style representations
• Improving anomaly detection through data augmentation

Innovation

• Diffusion model for time series style transfer
• Convolutional encoders disentangle content and style
• Self-supervised attention-based diffusion recombines representations