A secure and controllable flow method of a federated learning model in a cloud-edge environment

Abstract

The application relates to a secure and controllable flow method of a federated learning model in a cloud-edge environment and belongs to the technical field of data security and artificial intelligence. The method is applied to a decentralized cloud-edge-end collaborative environment, a multi-client function encryption scheme supporting a double decryption mode is provided, a local data training task model is used by a client, a parameter gradient is encrypted by using an encryption scheme, and NIZK proof bound to encrypted data is generated, edge nodes jointly act as a distributed key management center, generate and distribute encryption keys and various decryption key shares, calculate model inner product results, calculate the similarity of each client model and the current global model in combination with NIZK proof, and filter out malicious poisoning models, and the cloud server performs model aggregation calculation in a ciphertext state, and returns to the edge node for parameter gradient aggregation and global model updating. The application improves the security, robustness and efficiency existing in the secure flow of the federated learning model in the cloud-edge environment.

Images