Graph-structured data is ubiquitous in many applications, ranging from social networks to traffic prediction. Hence, graph learning tasks such as node classification have become increasingly important. However, graph data often encode sensitive information and are often distributed across multiple machines to comply with residency and privacy laws. Motivated by these challenges, we develop privacy-preserving techniques and practical methods for fully distributed graph learning.

We introduce LPG-Net, a novel model architecture that provides differential privacy (DP) guarantees for graph edges, and provides better privacy-utility tradeoff than traditional graph neural networks (GNNs). Expanding to the federated setting, we present the first hierarchical clustering algorithm that achieves local DP guarantees. It allows for improved social recommendation systems while preserving user privacy. To support scalable deployment of GNNs, we propose RETEXO, a distributed training framework that dramatically reduces communication costs when graph data is stored across many machines, making GNN training practical across various decentralization levels.