<ul><li>The paper introduces SpecGap, an approach for out-of-distribution (OOD) detection on graphs by analyzing Laplacian eigenvalue gaps.</li><li>It observes that OOD graph samples often show anomalous spectral gaps, prompting the development of SpecGap that adjusts features based on the differences in eigenvalues.</li><li>SpecGap is a parameter-free post-hoc method that can be seamlessly integrated into existing graph neural network models for improved OOD detection.</li><li>The approach achieves state-of-the-art performance on various benchmark datasets, supported by ablation studies and theoretical analyses.</li></ul>

SpectralGap: Graph-Level Out-of-Distribution Detection via Laplacian Eigenvalue Gaps

Discover more