Secure aggregation lets servers get finer updates without new uploads
Progressive-Resolution Secure Aggregation for Federated Learning
Cryptography and SecurityDistributed, Parallel, and Cluster ComputingInformation TheoryMachine Learning
Summary
When many devices work together on a learning task, it's important to keep their individual data private. Traditional methods require devices to send their data repeatedly at fixed precision levels. The authors present a way for devices to send their updates just once but allow the server to later unlock better, more detailed versions of the combined data without needing the devices to send anything again. They do this by cleverly encoding the data in layers that can be revealed bit by bit under the control of an independent entity.
What this means in practice
- •For federated learning engineers: Improve privacy-preserving model training by enabling servers to access progressively finer aggregated client updates without additional data uploads.
- •For data privacy officers: Enhance control over data release by authorizing improved aggregate detail only with independent approval, reducing privacy risks.
Authors
Seyed Mohammad Azimi-Abarghouyi
Abstract
Secure aggregation lets a server recover an aggregate of client updates without observing any individual update, but conventional protocols fix the aggregate precision when clients upload. We introduce and formulate a new progressive-resolution secure-aggregation functionality in which clients upload once and successively finer resolutions of the same aggregate can later be authorized without renewed client participation. To realize this functionality, we propose progressive-resolution secure aggregation (PSA): each clipped, dithered update is represented by compatible nested-lattice digits; separately releasable layers are protected by secure aggregation and an additional aggregate pad that remains unavailable to the server until a non-colluding release controller authorizes that layer.