CFPGExplainer

CFPGExplainer (CounterFactual Parameterized Graph neural network Explainer) is a counterfactual explainer model for the node classification task in Graph Neural Networks (GNNs) I developed for my master thesis project together with Prof. Fabrizio Silvestri and Prof. Simone Scardapane, my advisor and co-advisor respectively.

The proposed framework expands on the idea of perturbation-based approach to achieve, at once, commonly desired properties for GNN explainers, such as model-level explanations (i.e. not tailored to a single prediction instance), a more efficient inference process and counterfactual examples for each generated explanation.