A Graph Generation Network with Privacy Preserving Capabilities
Large datasets drive advancements in science, technology, and economic growth, with graph-structured data playing a pivotal role in mining and analysis. However, graph data often contains sensitive personal information, exposing it to privacy risks. To address this, the RDP-GGAN framework is proposed, integrating Rényi differential privacy (RDP) with generative adversarial networks (GANs). This framework enforces strict privacy constraints on deep graph generative models, focusing on preserving edge privacy to safeguard relational data connections. By injecting controlled noise into gradients of link-reconstruction models, RDP-GGAN enhances edge differential privacy, offering robust protection for sensitive graph-structured information.
May 21, 21210
