ID-Based Multireceiver Homomorphic Proxy Re-Encryption in Federated Learning
Chun‐I Fan, Ya-Wen Hsu, Cheng-Han Shie, Yi‐Fan Tseng
Abstract
Data privacy has become a growing concern with advances in machine learning. Federated learning (FL) is a type of machine learning invented by Google in 2016. In FL, the main aim is to train a high-accuracy global model by aggregating the local models uploaded by participants, and all data in the process are kept locally. However, compromises to security in the cloud server or among participants render this process insufficiently secure. To solve the problem, this article presents an identity-based multireceiver homomorphic proxy re-encryption (IMHPRE) scheme that utilizes homomorphism operations and re-encryption to provide improved encrypted-data processing and access control. When this scheme is employed, participants can directly use public identities for encryption. The IMHPRE scheme is also secure against the chosen-plaintext attacks. Comparison results indicated that the IMHPRE outperforms its counterparts because it allows a cloud server to perform model aggregation on re-encrypted models for multiple receivers.