Graph Reconstruction Attack

Graph reconstruction attacks (GRAs) aim to recover the underlying structure of a graph from information leaked by graph neural networks (GNNs) or their outputs, such as node embeddings. Current research focuses on developing more effective GRA methods, leveraging techniques like Markov chain approximations and analyzing the impact of different GNN architectures (e.g., GCNs, GATs, SNNs) and explanation methods on attack success. These attacks highlight significant privacy vulnerabilities in graph-based machine learning, impacting applications where sensitive relationships are encoded in graph data, and prompting the development of defense mechanisms to mitigate information leakage.