Flamingo: Multi-Round Single-Server Secure Aggregation with Applications to Private Federated Learning

In this talk I will discuss Flamingo, a system for secure aggregation of data
across a large set of clients. In secure aggregation, a server sums up the
private inputs of clients and obtains the result without learning anything
about the individual inputs beyond what is implied by the final sum. Flamingo
works particularly well in the multi-round setting found in federated learning
in which many consecutive additions (averages) of model weights are performed
to derive a good model. Furthermore, Flamingo can tolerate the failure of
clients (e.g., clients that go offline) in the middle of the computation. Our
implementation of Flamingo shows that it can securely train a neural network on
the (Extended) MNIST and CIFAR-100 datasets significantly quicker than all
prior secure aggregation systems, and the model converges without a loss in
accuracy, compared to a non-private federated learning system.