GOODAT: Towards Test-time Graph Out-of-Distribution Detection

GOODAT: Towards Test-time Graph Out-of-Distribution Detection

10 Jan 2024 | Luzhi Wang, Dongxiao He, He Zhang, Yixin Liu, Wenjie Wang, Shirui Pan, Di Jin, Tat-Seng Chua
The paper introduces GOODAT, a novel method for detecting Graph Out-of-Distribution (OOD) samples at test-time. GOODAT is a data-centric, unsupervised, and plug-and-play solution that operates independently of training data and does not require modifications to the GNN architecture. The method uses a lightweight graph masker to learn informative subgraphs from test samples, enabling the capture of distinct graph patterns between OOD and in-distribution (ID) samples. Three unsupervised objective functions based on the graph information bottleneck (GIB) principle are designed to optimize the graph masker, guiding it to capture compact yet informative subgraphs for OOD detection. Comprehensive evaluations on various real-world datasets demonstrate that GOODAT outperforms state-of-the-art benchmarks in graph OOD detection tasks. The main contributions of the paper include a new paradigm for test-time graph OOD detection, a novel method (GOODAT), and extensive experiments validating its effectiveness.The paper introduces GOODAT, a novel method for detecting Graph Out-of-Distribution (OOD) samples at test-time. GOODAT is a data-centric, unsupervised, and plug-and-play solution that operates independently of training data and does not require modifications to the GNN architecture. The method uses a lightweight graph masker to learn informative subgraphs from test samples, enabling the capture of distinct graph patterns between OOD and in-distribution (ID) samples. Three unsupervised objective functions based on the graph information bottleneck (GIB) principle are designed to optimize the graph masker, guiding it to capture compact yet informative subgraphs for OOD detection. Comprehensive evaluations on various real-world datasets demonstrate that GOODAT outperforms state-of-the-art benchmarks in graph OOD detection tasks. The main contributions of the paper include a new paradigm for test-time graph OOD detection, a novel method (GOODAT), and extensive experiments validating its effectiveness.
Reach us at info@study.space