What Information Matters? Graph Out-of-Distribution Detection via Tri-Component Information Decomposition
This work addresses the problem of OOD detection for graph node classification, providing a principled decomposition approach that significantly improves detection performance.
Graph neural networks are vulnerable to out-of-distribution (OOD) shifts. The proposed Tide framework decomposes node information into feature-specific, structure-specific, and joint components, filtering out spurious signals to improve OOD detection, achieving up to 34% improvement in FPR95 over baselines.
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.