Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Universal Time-Series Representation Learning: A Survey

Patara Trirat, Yooju Shin, Junhyeok Kang, Youngeun Nam, Jihye Na, Minyoung Bae, Joeun Kim, Byunghyun Kim, Jae-Gil Lee

TL;DR

This survey introduces a unified view of universal time-series representation learning by organizing methods around three core design elements: training data, neural architectures, and learning objectives. It synthesizes 127 papers into a structured taxonomy that spans data-centric, architectural, and learning-focused approaches, providing guidelines for experimental design and benchmark datasets. The work highlights open challenges—from distribution shifts and irregular sampling to NAS and LLM integration—and outlines promising directions to advance robust, transferable representations across diverse time-series tasks. By bridging foundational concepts with practical evaluation practices, the paper aims to accelerate the development of generalizable time-series representations for real-world systems.

Abstract

Time-series data exists in every corner of real-world systems and services, ranging from satellites in the sky to wearable devices on human bodies. Learning representations by extracting and inferring valuable information from these time series is crucial for understanding the complex dynamics of particular phenomena and enabling informed decisions. With the learned representations, we can perform numerous downstream analyses more effectively. Among several approaches, deep learning has demonstrated remarkable performance in extracting hidden patterns and features from time-series data without manual feature engineering. This survey first presents a novel taxonomy based on three fundamental elements in designing state-of-the-art universal representation learning methods for time series. According to the proposed taxonomy, we comprehensively review existing studies and discuss their intuitions and insights into how these methods enhance the quality of learned representations. Finally, as a guideline for future studies, we summarize commonly used experimental setups and datasets and discuss several promising research directions. An up-to-date corresponding resource is available at https://github.com/itouchz/awesome-deep-time-series-representations.

Universal Time-Series Representation Learning: A Survey

TL;DR

This survey introduces a unified view of universal time-series representation learning by organizing methods around three core design elements: training data, neural architectures, and learning objectives. It synthesizes 127 papers into a structured taxonomy that spans data-centric, architectural, and learning-focused approaches, providing guidelines for experimental design and benchmark datasets. The work highlights open challenges—from distribution shifts and irregular sampling to NAS and LLM integration—and outlines promising directions to advance robust, transferable representations across diverse time-series tasks. By bridging foundational concepts with practical evaluation practices, the paper aims to accelerate the development of generalizable time-series representations for real-world systems.

Abstract

Time-series data exists in every corner of real-world systems and services, ranging from satellites in the sky to wearable devices on human bodies. Learning representations by extracting and inferring valuable information from these time series is crucial for understanding the complex dynamics of particular phenomena and enabling informed decisions. With the learned representations, we can perform numerous downstream analyses more effectively. Among several approaches, deep learning has demonstrated remarkable performance in extracting hidden patterns and features from time-series data without manual feature engineering. This survey first presents a novel taxonomy based on three fundamental elements in designing state-of-the-art universal representation learning methods for time series. According to the proposed taxonomy, we comprehensively review existing studies and discuss their intuitions and insights into how these methods enhance the quality of learned representations. Finally, as a guideline for future studies, we summarize commonly used experimental setups and datasets and discuss several promising research directions. An up-to-date corresponding resource is available at https://github.com/itouchz/awesome-deep-time-series-representations.
Paper Structure (83 sections, 7 figures, 3 tables)

This paper contains 83 sections, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Surveys
  3. Survey Scope and Literature Collection
  4. Contributions
  5. Preliminaries
  6. Definitions
  7. Unique Properties of Time Series
  8. Temporal Dependency
  9. High Noise and Dimension
  10. Inter-Relationship across Variables
  11. Variability and Nonstationarity
  12. Diverse Semantics
  13. Neural Architectures for Time Series
  14. Multi-Layer Perceptrons (MLP)
  15. Recurrent Neural Networks (RNN)
  16. ...and 68 more sections

Figures (7)

  • Figure 1: Basic concept of time-series representation methods.
  • Figure 2: Key design elements and evaluation protocols of a time-series representation learning framework.
  • Figure 3: Quantitative summary of the selected papers.
  • Figure 4: Illustrations of regularly and irregularly sampled multivariate time series ($V = 3$).
  • Figure 5: Illustrative examples of data-centric approaches.
  • ...and 2 more figures

Theorems & Definitions (11)

  • definition 1: Time Series
  • definition 2: Irregularly-Sampled Time Series
  • definition 3: Time-Series Representation Learning
  • definition 4: Forecasting
  • definition 5: Classification
  • definition 6: Extrinsic Regression
  • definition 7: Clustering
  • definition 8: Segmentation
  • definition 9: Anomaly Detection
  • definition 10: Imputation of Missing Values
  • ...and 1 more