In federated learning, what security concern is specifically addressed by secure aggregation?

• A. Ensuring device updates are always positive.
• B. To prevent leakage of individual updates while still learning an aggregate model.
• C. Reducing communication costs by compressing updates.
• D. Increasing the training data on central server.

Answer: (B)

In federated learning, secure aggregation targets privacy of each device’s updates. Devices generate updates on their local data and send them to a central server, but if the server sees individual updates, sensitive information about local datasets could be inferred. Secure aggregation uses cryptographic techniques so the server can compute the overall sum (or average) of all updates without learning any single participant’s update. This lets the global model improve based on all clients while keeping each device’s data private. The other options don’t address this privacy protection: positivity of updates, compressing updates for fewer messages, and increasing data on the central server relate to correctness or efficiency, not safeguarding individual contributions.