Graph Privacy

Graph privacy research focuses on protecting sensitive information embedded within graph-structured data while preserving the utility of the data for machine learning tasks, particularly those using graph neural networks (GNNs). Current efforts concentrate on developing differentially private algorithms that add noise to graph data or model parameters to limit privacy leakage, often employing techniques like singular value decomposition or adaptive noise injection tailored to node importance. These methods aim to balance the trade-off between strong privacy guarantees (e.g., differential privacy) and maintaining the accuracy of GNN models for various applications, such as node classification and graph generation. The field's impact lies in enabling the responsible use of sensitive graph data in diverse domains while mitigating privacy risks.