A Survey of Deep Graph Learning under Distribution Shifts: from Graph Out-of-Distribution Generalization to Adaptation

TL;DR

The paper addresses the challenge of distribution shifts in graph learning by formalizing three core scenarios—graph OOD generalization, training-time OOD adaptation, and test-time OOD adaptation—and presenting a unified taxonomy that separates model-centric from data-centric approaches. It surveys a broad spectrum of methods, including invariant representation learning, causality-based learning, graph self-supervised learning, and diverse augmentation strategies, as well as both source-target alignment techniques and test-time adaptation strategies. The survey also catalogs standard benchmarks such as the Open Graph Benchmark, TU datasets, GOOD, and DrugOOD, and discusses theoretical and practical directions for improving robustness under shifts in structure, size, features, and label functions. By synthesizing these techniques across three scenarios, the work provides a practical framework and set of guidelines to advance reliable, distribution-shift-resistant graph learning with real-world impact in domains like chemistry, biology, and social networks.

Abstract

Distribution shifts on graphs -- the discrepancies in data distribution between training and employing a graph machine learning model -- are ubiquitous and often unavoidable in real-world scenarios. These shifts may severely deteriorate model performance, posing significant challenges for reliable graph machine learning. Consequently, there has been a surge in research on graph machine learning under distribution shifts, aiming to train models to achieve satisfactory performance on out-of-distribution (OOD) test data. In our survey, we provide an up-to-date and forward-looking review of deep graph learning under distribution shifts. Specifically, we cover three primary scenarios: graph OOD generalization, training-time graph OOD adaptation, and test-time graph OOD adaptation. We begin by formally formulating the problems and discussing various types of distribution shifts that can affect graph learning, such as covariate shifts and concept shifts. To provide a better understanding of the literature, we introduce a systematic taxonomy that classifies existing methods into model-centric and data-centric approaches, investigating the techniques used in each category. We also summarize commonly used datasets in this research area to facilitate further investigation. Finally, we point out promising research directions and the corresponding challenges to encourage further study in this vital domain. We also provide a continuously updated reading list at https://github.com/kaize0409/Awesome-Graph-OOD.

Figures (2)

• Figure 1: An illustration of the distinctions between graph OOD generalization, training-time graph OOD adaptation, and test-time graph OOD adaptation. The primary difference lies in the availability of target data and how source data is utilized. Graph OOD generalization assumes no target data during training, while both training-time and test-time adaptations assume access to target data. Training-time adaptation uses both source and target data during training, whereas test-time adaptation adapts a pre-trained model with only target data. The source graph can be single or multiple.
• Figure 2: Overview of our proposed taxonomy. Within each of the three scenarios, existing methods can be further categorized into model-centric and data-centric strategies.