Whether FedML support homomorphic encryption and multi-party secure computing functions

[Doodlera]

I wonder whether FedML support homomorphic encryption and multi-party secure computing functions. I found "norm_diff_clipping" and "weak_dp" part in RobustAggregator while it is not enough for the applications. Whether more security modules already exist while I miss them, or I need to finish them myself. Thank you.

[JLU-Neal]

Got the same question...Seems nothing changed after a year

[chaoyanghe]

@Doodlera @JLU-Neal we do have MPC-based algorithm, check this example: https://github.com/FedML-AI/FedML/tree/master/python/examples/cross_silo/light_sec_agg_example. It's our new paper published at MLSys 2022: https://proceedings.mlsys.org/paper/2022/hash/d2ddea18f00665ce8623e36bd4e3c7c5-Abstract.html

[mstc-xqp]

@Doodlera @JLU-Neal we do have MPC-based algorithm, check this example: https://github.com/FedML-AI/FedML/tree/master/python/examples/cross_silo/light_sec_agg_example. It's our new paper published at MLSys 2022: https://proceedings.mlsys.org/paper/2022/hash/d2ddea18f00665ce8623e36bd4e3c7c5-Abstract.html

FedML-HE: An Efficient Homomorphic-Encryption-Based Privacy-Preserving Federated Learning System what about fhe

[fedml-dimitris]

As @chaoyanghe already mentioned, FedML supports both MPC and HE: MPC

• https://github.com/FedML-AI/FedML/tree/master/python/examples/cross_silo/light_sec_agg_example
• https://github.com/FedML-AI/FedML/tree/a9b04b50a34bba60e55d351526702317586d416f/python/fedml/core/mpc

HE

• https://github.com/wzjin2017/FedML/tree/he-tenseal/python/examples/cross_silo/mqtt_s3_fedavg_fhe_mnist_lr_example (the above branch will soon be merged in the latest release)

[kamelferrahi]

Hi can I have the link of the paper since the one published here does not work