A Practical Implementation of Medical Privacy-Preserving Federated Learning Using Multi-Key Homomorphic Encryption and Flower Framework

Round 1

Reviewer 1 Report

This work present a privacy-preserving FL scheme based on multi-key fully homomorphic  encryption based on RLWE and Flower.  The motivation is interesting and the experiments are sufficient.  However, the security analysis is to be improved. In particular, the authors introduce improvements towards the referred FHE scheme in ref.[30], thus the security proof or analysis on new FHE scheme is expected. 

Author Response

We appreciate the reviewer's feedback and their recognition of the motivation and experiments in our work. We tried to address this comment in Section 2.4, Homomorphic Encryption. Regarding the security analysis, we understand the importance of thoroughly assessing the new FHE scheme introduced in our study. However, it's important to clarify that our primary focus was on the practical implementation and evaluation of this privacy-preserving FL scheme using the xMK-CKKS encryption system.

The xMK-CKKS encryption system represents a significant advancement in preserving the privacy of federated learning processes. It enhances efficiency through collaborative public key generation, reducing the computational burden on individual clients. This system encrypts model updates using a combined public key derived from individual client public keys, thereby strengthening privacy protection. Collaborative encryption ensures that model updates remain concealed from the server and other clients, safeguarding data privacy. Additionally, the partial decryption process, involving multiple clients, adds an extra layer of security.

While we acknowledge the importance of providing security proof or an in-depth analysis of the new FHE scheme, a comprehensive examination of this scheme goes beyond the scope of our current research. Our contribution primarily lies in implementing and evaluating this enhanced privacy-preserving FL framework. This work sets the stage for future in-depth security analyses of the xMK-CKKS encryption system, which could be a valuable direction for further research.

Once again, we appreciate the reviewer's feedback and will consider their suggestions for future investigations into the security aspects of the xMK-CKKS encryption system.

Reviewer 2 Report

Comments for author File: Comments.pdf

Author Response

We appreciate the reviewer's valuable input regarding including authentication protocols in Chapter 2 of our manuscript. In response to this comment, we have carefully revised and expanded our related work section (Section 2) to incorporate references to the Privacy Preserving Mutual Authentication Protocol, as suggested by the reviewer.

1. We have now included references to the relevant literature that primarily presents this protocol, ensuring that our readers have access to a comprehensive overview of authentication protocols in the context of privacy-preserving federated learning. These additions enhance the depth and relevance of our related work section, providing a more extensive background for our research.

2. In response to this valuable input, we have included a dedicated section in Chapter 2 that elaborates on using PUFs for secure key generation. This section provides an in-depth explanation of how PUFs can enhance key generation security, drawing from relevant literature, including the works mentioned by the reviewer. This addition enriches our manuscript by addressing the importance of PUFs in healthcare data privacy within the context of federated learning.

3. We have revised the contribution section in the introduction (Section 1) to provide a more precise and more explicit comparison of our approach with previous works in the literature. We have highlighted our method's main strengths and unique contributions, facilitating a better understanding of its novelty within the context of privacy-preserving federated learning.