Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Out-of-Distribution Detection on Graphs: A Survey

Tingyi Cai, Yunliang Jiang, Yixin Liu, Ming Li, Changqin Huang, Shirui Pan

TL;DR

This survey formalizes graph out-of-distribution (GOOD) detection, defining IND vs. OOD graphs and presenting a four-way taxonomy of approaches. It comprehensively reviews enhancement-, reconstruction-, information propagation-, and classification-based methods, with representative techniques and notable examples such as VGAE-based reconstructions, energy/uncertainty propagation, and boundary-focused classifiers. The work also maps common node- and graph-level datasets, cross-domain OOD splits, and standard benchmarks like DrugOOD and UB-GOLD, while discussing practical applications, theoretical underpinnings, and future directions. By clarifying distinctions from graph anomaly/outlier detection and GOOD generalization, the paper lays a foundation for scalable, explainable, and robust graph learning under distribution shifts, and highlights promising directions such as graph foundation models with OOD awareness.

Abstract

Graph machine learning has witnessed rapid growth, driving advancements across diverse domains. However, the in-distribution assumption, where training and testing data share the same distribution, often breaks in real-world scenarios, leading to degraded model performance under distribution shifts. This challenge has catalyzed interest in graph out-of-distribution (GOOD) detection, which focuses on identifying graph data that deviates from the distribution seen during training, thereby enhancing model robustness. In this paper, we provide a rigorous definition of GOOD detection and systematically categorize existing methods into four types: enhancement-based, reconstruction-based, information propagation-based, and classification-based approaches. We analyze the principles and mechanisms of each approach and clarify the distinctions between GOOD detection and related fields, such as graph anomaly detection, outlier detection, and GOOD generalization. Beyond methodology, we discuss practical applications and theoretical foundations, highlighting the unique challenges posed by graph data. Finally, we discuss the primary challenges and propose future directions to advance this emerging field. The repository of this survey is available at https://github.com/ca1man-2022/Awesome-GOOD-Detection.

Out-of-Distribution Detection on Graphs: A Survey

TL;DR

This survey formalizes graph out-of-distribution (GOOD) detection, defining IND vs. OOD graphs and presenting a four-way taxonomy of approaches. It comprehensively reviews enhancement-, reconstruction-, information propagation-, and classification-based methods, with representative techniques and notable examples such as VGAE-based reconstructions, energy/uncertainty propagation, and boundary-focused classifiers. The work also maps common node- and graph-level datasets, cross-domain OOD splits, and standard benchmarks like DrugOOD and UB-GOLD, while discussing practical applications, theoretical underpinnings, and future directions. By clarifying distinctions from graph anomaly/outlier detection and GOOD generalization, the paper lays a foundation for scalable, explainable, and robust graph learning under distribution shifts, and highlights promising directions such as graph foundation models with OOD awareness.

Abstract

Graph machine learning has witnessed rapid growth, driving advancements across diverse domains. However, the in-distribution assumption, where training and testing data share the same distribution, often breaks in real-world scenarios, leading to degraded model performance under distribution shifts. This challenge has catalyzed interest in graph out-of-distribution (GOOD) detection, which focuses on identifying graph data that deviates from the distribution seen during training, thereby enhancing model robustness. In this paper, we provide a rigorous definition of GOOD detection and systematically categorize existing methods into four types: enhancement-based, reconstruction-based, information propagation-based, and classification-based approaches. We analyze the principles and mechanisms of each approach and clarify the distinctions between GOOD detection and related fields, such as graph anomaly detection, outlier detection, and GOOD generalization. Beyond methodology, we discuss practical applications and theoretical foundations, highlighting the unique challenges posed by graph data. Finally, we discuss the primary challenges and propose future directions to advance this emerging field. The repository of this survey is available at https://github.com/ca1man-2022/Awesome-GOOD-Detection.

Paper Structure

This paper contains 27 sections, 2 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Foundations of GOOD Detection
  3. Problem Definition
  4. Taxonomy of Approaches
  5. Enhancement-based Methods
  6. Data Enhancement
  7. Model Enhancement
  8. Reconstruction-based Methods
  9. Information Propagation-based Methods
  10. Feature Propagation
  11. Energy Propagation
  12. Uncertainty Propagation
  13. Discussion.
  14. Classification-based Methods
  15. Probabilistic Evidence Classification
  16. ...and 12 more sections

Figures (6)

  • Figure 1: Applications and rising interest in GOOD detection. This figure illustrates the diverse applications of GOOD detection, including drug discovery and robust AI systems, and highlights the increasing number of publications in this field over recent years, emphasizing its growing significance in both academia and industry.
  • Figure 2: A taxonomy of GOOD detection approaches.
  • Figure 3: General framework for the enhancement-based methods.
  • Figure 4: General framework for the reconstruction-based methods.
  • Figure 5: General framework for the information propagation-based methods.
  • ...and 1 more figures

Theorems & Definitions (2)

  • Definition 1
  • Definition 2