<ul><li>Graph convolutional neural networks (GCNNs) are powerful for analyzing graph-structured data and have been successful in various applications.</li><li>Understanding the stability of GCNN models, i.e., how they react to small changes in graph structures, is crucial but currently limited.</li><li>This study examines how perturbations in graph topology impact GCNN outputs and presents a new method for assessing model stability.</li><li>The proposed probabilistic framework offers insights into the relationship between data properties, graph perturbations, and model stability, validated through experiments.</li></ul>

On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective

Discover more