Authors: Yash Sinha, Murari Mandal, Mohan Kankanhalli

D2DGN is a model-agnostic framework for graph unlearning using knowledge distillation. It removes the influence of deleted nodes, edges, or features from a trained GNN without retraining from scratch.

Key features:

• Response-based and embedding-based distillation
• No retraining required
• Better efficiency and unlearning performance

✅ Improves over GNNDelete by +2.4% AUC, uses 10.2×10⁶ fewer FLOPs, and is up to 3.2× faster.

This repository is adapted from GNNDelete (ICLR 2023).