Beyond Soft Masks: Hard-Perturbation Mixup Explainer for Robust GNN Explainability
Researchers have developed a new framework called HPME to improve the explainability of Graph Neural Networks (GNNs). Existing methods often struggle with 'soft masks' that allow irrelevant information to persist, hindering the accuracy of explanations. HPME utilizes graph pooling to extract discrete subgraphs and a novel mixup strategy to generate explanations that are more robust and interpretable, demonstrating state-of-the-art performance on various datasets. AI
IMPACT Enhances trustworthiness of GNNs in high-stakes applications by providing more robust and interpretable explanations.