Secure Aggregation
A cryptographic protocol that allows a server to compute aggregate values from individual contributions without seeing the individual values.
Secure Aggregation enables computing sums over individual contributions without seeing individual values – key building block for privacy-preserving Federated Learning.
Explanation
In Federated Learning, each client masks their model updates with random masks. The masks cancel out during aggregation. The server sees only the sum, never individual updates.
Marketing Relevance
Essential for privacy-compliant Federated Learning. Prevents the aggregating server from inferring individual gradients (and thus training data).
Example
Apple uses Secure Aggregation for iCloud ML: Emoji usage statistics are aggregated without Apple ever seeing which emojis an individual user uses.
Common Pitfalls
High communication overhead. Dropouts (clients going offline) must be handled securely. Scaling with thousands of clients is complex.
Origin & History
Bonawitz et al. (Google, 2017) introduced Practical Secure Aggregation for Federated Learning. Since then a standard component in Google's and Apple's on-device ML systems.
Comparisons & Differences
Secure Aggregation vs. Differential Privacy
Differential Privacy adds noise to results; Secure Aggregation cryptographically hides individual contributions from the server.
Secure Aggregation vs. Federated Learning
Federated Learning is the training approach; Secure Aggregation is a protection layer within it.