Graph Unlearning via Embedding Reconstruction -- A Range-Null Space Decomposition Approach
This addresses the need for efficient and effective unlearning in GNNs to handle privacy or data removal requests, though it is incremental as it builds on existing unlearning concepts.
The paper tackles the problem of node unlearning in graph neural networks (GNNs), which is challenging and underexplored, by proposing a method based on embedding reconstruction and Range-Null Space Decomposition, achieving state-of-the-art performance on multiple datasets.
Graph unlearning is tailored for GNNs to handle widespread and various graph structure unlearning requests, which remain largely unexplored. The GIF (graph influence function) achieves validity under partial edge unlearning, but faces challenges in dealing with more disturbing node unlearning. To avoid the overhead of retraining and realize the model utility of unlearning, we proposed a novel node unlearning method to reverse the process of aggregation in GNN by embedding reconstruction and to adopt Range-Null Space Decomposition for the nodes' interaction learning. Experimental results on multiple representative datasets demonstrate the SOTA performance of our proposed approach.