A Verifiable Federated Learning Scheme Based on Secure Multi-party Computation

Federated learning ensures that the quality of the model is uncompromised while the resulting global model is consistent with the model trained by directly collecting user data. However, the risk of inferring data considered in federated learning. Furthermore, the inference to the learning outcome considered in a federated learning environment must satisfy that data cannot be inferred from any outcome except the owner of the data. In this paper, we propose a new federated learning scheme based on secure multi-party computation (SMC) and differential privacy. The scheme prevents inference during the learning process as well as inference of the output. Meanwhile, the scheme protects the user's local data during the learning process to ensure the correctness of the results after users' midway exits through the process.