Explainability

We extend and test KEdge, an interpretable-by-design approach for graph neural networks, and compare it to gradient-based attribution techniques.