FedNC: A Secure and Efficient Federated Learning Method with Network Coding

Federated Learning (FL) is a promising distributed learning mechanism whichstill faces two major challenges, namely privacy breaches and systemefficiency. In this work, we reconceptualize the FL system from the perspectiveof network information theory, and formulate an original FL communicationframework, FedNC, which is inspired by Network Coding (NC). The main idea ofFedNC is mixing the information of the local models by making random linearcombinations of the original parameters, before uploading for furtheraggregation. Due to the benefits of the coding scheme, both theoretical andexperimental analysis indicate that FedNC improves the performance oftraditional FL in several important ways, including security, efficiency, androbustness. To the best of our knowledge, this is the first framework where NCis introduced in FL. As FL continues to evolve within practical networkframeworks, more variants can be further designed based on FedNC.