International Association
for Cryptologic Research

This paper introduces Flamingo, a system for secure aggregation of data across a large set of clients that fits the stringent needs of federated learning settings. 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 focuses on the multi-round setting found in federated learning in which many consecutive summations (averages) of model weights are performed to derive a good model. Prior works, designed for a single round, when adapted to the multi-round setting, require all clients to establish pairwise secrets per round, which is onerous when the number of clients is large and clients have varying network conditions. Flamingo introduces a novel protocol for reusing pairwise secrets to reduce the overall communication-round complexity and a new lightweight dropout resilience protocol to ensure that if clients leave in the middle of a sum the server can still obtain a meaningful result. These techniques help Flamingo reduce the end-to-end runtime and the number of interactions between clients and the server for a full training session over prior work. We implement and evaluate Flamingo and show that it can effectively be used to securely train a neural network on the (Extended) MNIST and CIFAR-100 datasets, and the model converges without a loss in accuracy, compared to a non-private federated learning system.