Unsupervised Representation Learning for Time Series: A Review

TL;DR

This paper addresses the lack of systematic analysis in unsupervised time series representation learning by providing a comprehensive taxonomy and a unified evaluation framework. It introduces ULTS, a PyTorch-based library implementing 17 state-of-the-art models across deep clustering, reconstruction-based, and self-supervised learning paradigms, enabling fair comparisons on 9 real-world datasets. The review highlights the superiority of self-supervised contrastive methods, particularly at temporal and prototype levels, while detailing practical considerations such as augmentation design, scalability, and robustness. Overall, the work offers practical guidance for researchers and practitioners and paves the way for more standardized benchmarking in time series representation learning.

Abstract

Unsupervised representation learning approaches aim to learn discriminative feature representations from unlabeled data, without the requirement of annotating every sample. Enabling unsupervised representation learning is extremely crucial for time series data, due to its unique annotation bottleneck caused by its complex characteristics and lack of visual cues compared with other data modalities. In recent years, unsupervised representation learning techniques have advanced rapidly in various domains. However, there is a lack of systematic analysis of unsupervised representation learning approaches for time series. To fill the gap, we conduct a comprehensive literature review of existing rapidly evolving unsupervised representation learning approaches for time series. Moreover, we also develop a unified and standardized library, named ULTS (i.e., Unsupervised Learning for Time Series), to facilitate fast implementations and unified evaluations on various models. With ULTS, we empirically evaluate state-of-the-art approaches, especially the rapidly evolving contrastive learning methods, on 9 diverse real-world datasets. We further discuss practical considerations as well as open research challenges on unsupervised representation learning for time series to facilitate future research in this field.

Figures (2)

• Figure 1: An up-to-date taxonomy of unsupervised representation learning methods for time series, including a) Deep Clustering Methods b) Reconstruction-based Methods and c) Self-supervised Learning Methods. The self-supervised learning methods can be further divided into adversarial methods, predictive methods and contrastive methods, depending on different types of pretext tasks employed for acquiring self-supervised signals.
• Figure 2: The organization of the research in unsupervised learning techniques for time series, based on 3 main categories outlined in our proposed taxonomy.