FedE: Protecting Training Data of Federated Learning Based on Multi-Precision Functional Encryption
TL;DR
FedE, a multi-precision, multi-source, heterogeneous privacy-preserving federated learning training method based on functional encryption that enhances numerical adaptation during ciphertext computation and prevents model parameter updates from easily compromising privacy in cross-institutional federated learning.
Abstract
: With the rapid development of artificial intelligence technologies in machine learning-as-a-service (MLaaS), deep learning-based intelligent models have demonstrated high value in applications such as trend prediction and risk assessment. However, MLaaS data are typically highly sensitive and contain private information. In cross-institutional collaborative modeling scenarios, different departments and local centers often hold partial, heterogeneous data resources on MLaaS platforms. Given data security, privacy, and compliance requirements, raw data are difficult to share directly, which makes it challenging to apply centralized model training methods. This paper proposes FedE, a multi-precision, multi-source, heterogeneous privacy-preserving federated learning training method based on functional encryption. By introducing a multi-precision joint-computation mechanism, this method enhances numerical adaptation during ciphertext computation. Besides, combined with differential privacy techniques, it prevents model parameter updates from easily compromising privacy in cross-institutional federated learning. Based on the federated learning training process, a secure training framework for sensitive MLaaS data is constructed. Finally, experiments demonstrate the security and efficiency of the proposed approach.