On the Prediction Instability of Graph Neural Networks

Max Klabunde, Florian Lemmerich

Instability of trained models, i.e., the dependence of individual node predictions on random factors, can affect reproducibility, reliability, and trust in machine learning systems…