Privacy Preserving Tree: Zero Overhead Privacy Preservation in Cross-silo Decentralized Federated Learning
Tomsy Paul, Santhosh Kumar · 2025
Federated Learning (FL) in Edge Computing generally manifests in two forms: Cross-silo FL, for networks of Edge/Cloud servers, and Cross-device FL, for IoT devices connected to a single server. Cross-silo FL, benefiting from fewer participants, robust resources, and relaxed security constraints, is particularly suited for Decentralized FL, which eliminates the need for a central server. Despite its potential, Decentralized Cross-silo FL remains a nascent field, struggling with challenges in privacy and computation & communication. While existing research addresses these issues individually, a holistic approach is lacking. To bridge this gap, we introduce Privacy Preserving Tree, a novel adaptation of the Binomial Tree employing additive Secret Sharing. This method significantly reduces both computation and communication overhead while ensuring data privacy. We also provide a container based implementation of Privacy Preserving Tree, for easy development and deployment which is quite suitable for both the academia and the industry.