Secure Aggregation in Federated Learning Explainer
Secure aggregation is a cryptographic protocol that allows a server to compute an aggregate, such as the sum of model updates, without seeing each participant’s individual contribution. It is commonly used in federated learning.
Clients mask or secret-share their updates so individual values are unreadable to the server. When enough clients complete the protocol, the masks cancel or shares combine, revealing only the aggregate needed for a training round.
The technique reduces the exposure created by sending unprotected updates to a central coordinator. Practical protocols also handle clients that disconnect after setup, because intermittent participation is normal in federated systems.
Secure aggregation is not the same as differential privacy and does not guarantee that the final model reveals nothing about training data. It depends on threat assumptions, participation thresholds, cryptographic correctness, and defenses against malicious or poisoned updates.
Secure aggregation lets a server learn the combined value of many client updates without learning the individual update submitted by each client.
Key terms include secure aggregation, federated learning, client update, cryptographic mask, secret sharing, aggregate, dropout, threat model and differential privacy.
Clients cryptographically protect their contributions before they reach the coordinator.
The protocol reconstructs the combined aggregate only when participation requirements are met.
The final model can still leak information, and malicious clients can submit harmful updates.
Privacy mechanisms, anomaly defenses, and a clear threat model must complement the protocol.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.