Figure 3:

Evaluation of single-instance explanations. A-B. Shown are exemplar explanation subgraphs for node classification task on four synthetic datasets. Each method provides explanation for the red node’s prediction.