ScholarWorks@중앙대학교

초록

Graph unlearning seeks to remove the influence of specific nodes, edges, or features from trained GNNs without performing full retraining. Although recent studies propose diverse unlearning strategies, inconsistent evaluation practices limit meaningful comparison and slow progress. In this work, we introduce the GNN Unlearning Reality Checklist (GURC)—a concise framework for evaluating unlearning methods across utility, privacy, scalability, and reproducibility. We identify critical evaluation gaps in the literature and define practical reporting requirements. The checklist establishes a baseline for conducting rigorous, transparent, and comparable GNN unlearning research.

키워드