adap / flower

Flower: A Friendly Federated AI Framework
https://flower.ai
Apache License 2.0
5.13k stars 881 forks source link

Require a secure aggregation baseline based on homomorphic encryption #4179

Open LeoLee1116 opened 2 months ago

LeoLee1116 commented 2 months ago

Describe the type of feature and its functionality.

Hello, the efficiency of the Flower framework is amazing and it greatly meets my research needs. Thank you for your contributions! I noticed that secret sharing and symmetric encryption techniques have been considered in terms of secure aggregation. However, as mentioned in #907 , I still haven't found a secure aggregation baseline that explicitly uses homomorphic encryption.

Describe step by step what files and adjustments are you planning to include.

In my opinion, in the simplest case, a secure aggregation strategy based on homomorphic encryption allows all clients to share a pair of keys and encrypt the transmission parameters or gradients. This scheme supports basic operations such as encrypted addition and multiplication during the aggregation process, ensuring the accuracy of server decryption results while providing strong privacy protection. On this basis, the baseline can also be extended to scenarios where each client holds different key pairs, which is of great significance for the development of the Flower framework.

Is there something else you want to add?

No response

petrosDemetrakopoulos commented 2 months ago

I am also extremely interested in such a feature and I am also willing to help towards developing it.