FADO: A Federated Learning Attack and Defense Orchestrator
Filipe Martins Rodrigues, Rodrigo Simões, Nuno Neves · 2023
Federated Learning (FL) is a distributed machine learning approach allowing multiple parties to train a model collaboratively without sharing sensitive data. It has gained widespread popularity recently due to its ability to preserve data privacy. However, FL also poses novel security challenges since training relies on data and computations from many entities that a malicious actor might have compromised, as they are usually geographically dispersed and independently managed. Evaluations of current FL security mechanisms in the literature are often based on simplistic testing environments and demand complex programming to integrate new attacks/defenses. Therefore, this work presents an accessible platform that leverages a realistic environment to facilitate the experimentation and evaluation of new solutions in relevant FL scenarios. Comparison with already proposed approaches is also expedited since FADO provides a few out-of-the-box implementations. To demonstrate the platform’s utility, we develop a use case based on a recently published network attack.