A systematic implementation of model poisoning attacks in federated machine learning using FEDn framework.
-
Scalable. The implementation leverages the core capabilities of FEDn framework. Making it highly scalable and resilient to failures.
-
ML-framework agnostic. The attacks implemented do not rely on any particular ML framework which is another feature made possible because of FEDn framework.
-
Fast deployments. This poison attack framework provides a quick and easy way to setup new attack scenarios and to test defense mechanism with minimal changes.
-
Geo-distributed. Using FEDn architecture allows the users to deploy realistic and geo-distributed federated experiments and emulate specific attack scenarios.