Title: Let Them Drop in Federated Learning (making it not a big deal!)
Abstract: Secure model aggregation is nowadays recognized as the key component for Federated Learning (FL). It enables the collaborative training of a global machine learning model without leaking any information about FL clients’ local models. It is shown that clients who fail to complete the protocol, referred to as dropped clients, can seriously affect the correct computation of the global machine learning model. While the literature counts multiple fault-tolerant secure aggregation protocols, they rely on secret sharing techniques to reconstruct the inputs of dropped clients. As a consequence, the performance of these solutions decreases with increasing dropout rates. In this talk, we present recent fault-tolerant secure aggregation contributions, based on the combination of homomorphic encryption and secret sharing mechanisms, which only rely on the inputs of online clients. These contributions are agnostic to client failures and, therefore, outperform existing solutions.
Français 
English (UK)