What Information Matters? Graph Out-of-Distribution Detection via Tri-Component Information Decomposition

2026-05-13

AI Summary

Graph neural networks for node classification are highly susceptible to out-of-distribution (OOD) shifts in both node features and graph structure, and standard supervised learning often captures spurious correlations that degrade generalization. To address this, this work proposes Tide, a novel framework that, for the first time, incorporates trisecting information decomposition into graph-based OOD detection. Tide explicitly disentangles feature-specific, structure-specific, and joint information, retaining only the label-relevant joint component to filter out spurious signals. Grounded in the information bottleneck principle, the method is theoretically superior to conventional supervised learning, yielding higher in-distribution (ID) confidence and a wider prediction entropy gap between ID and OOD samples. Evaluated on seven benchmark datasets, Tide substantially improves OOD detection performance—reducing FPR95 by up to 34%—while maintaining strong ID classification accuracy.

Abstract

Graph neural networks are widely used for node classification, but they remain vulnerable to out-of-distribution (OOD) shifts in node features and graph structure. Prior work established that methods trained with standard supervised learning (SL) objectives tend to capture spurious signals from either features and/or structure, leaving the model fragile under distributional changes. To address this, we propose textscTide, a textbfnovel and effective underlineTri-Component underlineInformation underlineDecomposition framework that textbfexplicitly decomposes information into textitfeature-specific, structure-specific and joint components. textscTide aims to textbfpreserve only the label-relevant part of the joint information while textbffiltering out spurious feature- and structure-specific information, thereby enhancing the separation between in-distribution (ID) and OOD nodes. Beyond the framework, we provide theoretical and empirical analyses showing that an information bottleneck objective is preferable to standard SL for graph OOD detection, with higher ID confidence and a greater entropy gap between ID and OOD data. Extensive experiments across seven datasets confirm the efficacy of textscTide, achieving up to a 34% improvement in FPR95 over strong baselines while maintaining competitive ID accuracy.

Problem

• out-of-distribution detection
• graph neural networks
• distributional shift
• spurious correlation
• node classification

Innovation

• Graph Out-of-Distribution Detection
• Information Decomposition
• Graph Neural Networks
• Information Bottleneck
• Spurious Correlation