A novel federated learning scheme based on cloud-edge collaborative computing

In the process of federated learning, how to efficiently protect the privacy and the integrity of local and global training models is an urgent problem to be solved. Traditional secure federated learning methods based on differential privacy have shortcomings such as high computational overhead, high communication energy consumption and long execution time. Therefore, a novel secure and efficient federated learning scheme (SEFL) based on “cloud edge” collaborative computing was proposed. SEFL ensured the security of model aggregation by configuring the Intel SGX-based TEE (Trusted Execution Environment) on the Cloud Server (CS). It combined the symmetric and the asymmetric encryption technologies to protect the communication security between CS and Edge Servers (ESs), and improved the security of model storage by constructing a chained storage structure on ESs. Theoretical analyses and experimental results showed that SEFL could secure FL and effectively improve the FL training efficiency.