Privacy-preserving Federated Learning based on Multi-key Homomorphic Encryption (2104.06824v1)

Abstract: With the advance of machine learning and the internet of things (IoT), security and privacy have become key concerns in mobile services and networks. Transferring data to a central unit violates privacy as well as protection of sensitive data while increasing bandwidth demands.Federated learning mitigates this need to transfer local data by sharing model updates only. However, data leakage still remains an issue. In this paper, we propose xMK-CKKS, a multi-key homomorphic encryption protocol to design a novel privacy-preserving federated learning scheme. In this scheme, model updates are encrypted via an aggregated public key before sharing with a server for aggregation. For decryption, collaboration between all participating devices is required. This scheme prevents privacy leakage from publicly shared information in federated learning, and is robust to collusion between $k<N-1$ participating devices and the server. Our experimental evaluation demonstrates that the scheme preserves model accuracy against traditional federated learning as well as secure federated learning with homomorphic encryption (MK-CKKS, Paillier) and reduces computational cost compared to Paillier based federated learning. The average energy consumption is 2.4 Watts, so that it is suited to IoT scenarios.

• The paper introduces xMK-CKKS, a multi-key homomorphic encryption protocol that secures federated learning model updates.
• The scheme uses distinct participant keys for encryption to prevent data leakage and resist internal collusion.
• Experimental validation in an IoT smart healthcare scenario shows comparable accuracy with reduced computational load and low energy consumption.

Privacy-preserving Federated Learning based on Multi-key Homomorphic Encryption

The paper "Privacy-preserving Federated Learning based on Multi-key Homomorphic Encryption" presents a novel approach to enhance the privacy and security of federated learning frameworks using a specifically designed encryption protocol. Federated learning (FL) primarily addresses the privacy concerns associated with transferring data by enabling local training of models on distributed devices, sharing only model updates with a central server. However, even in federated learning, there is a risk of data leakage from these updates. Traditional homomorphic encryption (HE) solutions, while encrypting these updates, often use the same keys across participants, presenting vulnerabilities to curious adversaries and collusion attacks within the training participants.

The authors propose xMK-CKKS, an enhancement to the existing MK-CKKS scheme, introducing a multi-key homomorphic encryption (MK-HE) approach suitable for the federated learning environment. This protocol ensures that different participants use distinct encryption keys, necessitating collaborative decryption involving all parties, thus mitigating the risks associated with internal attacks and collusion between devices and the server. The xMK-CKKS scheme notably refines decryption processes to avoid leaking information through shared public information.

Key Contributions

1. xMK-CKKS Scheme Design: The introduction of xMK-CKKS is a major contribution, enhancing MK-CKKS by employing an aggregated public key for encryption. This improvement provides strong privacy guarantees without necessitating interaction among participants during encryption, thus supporting distributed scenarios with disconnected topologies.
2. Implementation in Federated Learning: The xMK-CKKS protocol is integrated within a federated learning setup, enabling robust encryption of model updates. This integration ensures that the server can only decrypt aggregated information rather than individual updates, significantly boosting security.
3. Experimental Validation: The protocol was tested in a realistic IoT-based smart healthcare scenario using Jetson Nano devices for an elderly-fall detection task. The results showed comparable model accuracy to non-encrypted FL schemes while significantly reducing computational load and maintaining low energy consumption.

Theoretical and Practical Implications

The paper not only addresses direct privacy concerns within federated learning by securing model updates but also advances the applicability of homomorphic encryption techniques within resource-constrained IoT environments. The xMK-CKKS scheme provides a pathway for the integration of advanced cryptographic methods in collaborative AI environments, reinforcing data privacy while maintaining computational efficiency.

By reducing computational and communication overheads compared to prior HE-based FL systems, the xMK-CKKS protocol extends the practicality of federated learning, making it feasible for real-world IoT applications where devices often have limited processing capabilities.

Speculations on Future Work

Future research could explore improving the resiliency of the xMK-CKKS scheme against Byzantine attacks, where malicious participants attempt to bias the learning process. This would involve further designing privacy-preserving federated learning mechanisms that incorporate robust outlier detection without compromising individual data privacy.

Moreover, applying such secure federated learning frameworks in other domains, such as smart cities or autonomous vehicles, may also be explored, providing a broader evaluation of the system's effectiveness in diverse, privacy-sensitive environments. As federated learning becomes a cornerstone of distributed AI systems, ensuring its robustness against both internal and external threats will remain a high-priority research focus.