UniTS: A Universal Time Series Analysis Framework Powered by Self-Supervised Representation Learning

TL;DR

UniTS addresses practical time series analysis challenges such as partial labeling and domain shift by first learning task-agnostic representations through self-supervised pre-training, then fine-tuning lightweight task-specific heads on fused embeddings. The framework combines a versatile Pre-training Module, a flexible Feature Fusion Module, and a Task Module to support five mainstream TS tasks via sklearn-like APIs and GUI workflows, enabling efficient reuse of encoders across tasks. Empirical results show UniTS outperforms traditional end-to-end methods on multiple tasks and settings, demonstrating improved data efficiency and cross-domain generalization. The open-source design facilitates community extension and deployment in real-world scenarios.

Abstract

Machine learning has emerged as a powerful tool for time series analysis. Existing methods are usually customized for different analysis tasks and face challenges in tackling practical problems such as partial labeling and domain shift. To improve the performance and address the practical problems universally, we develop UniTS, a novel framework that incorporates self-supervised representation learning (or pre-training). The components of UniTS are designed using sklearn-like APIs to allow flexible extensions. We demonstrate how users can easily perform an analysis task using the user-friendly GUIs, and show the superior performance of UniTS over the traditional task-specific methods without self-supervised pre-training on five mainstream tasks and two practical settings.

Figures (3)

• Figure 1: Framework overview of UniTS.
• Figure 2: Demonstration overview of UniTS
• Figure 3: Overall performance of UniTS

Theorems & Definitions (1)

• Definition 2.1: Time Series Analysis