IntroStep 1 of 3
Bonawitz et al. Secure Aggregation
Practical Secure Aggregation for Privacy-Preserving Machine Learning
arXiv:1611.04482This interactive demonstration walks you through the complete Bonawitz secure aggregation protocol, exactly as described in the original paper. You'll see how multiple clients can collaboratively compute an aggregate of their private data without revealing individual values.
Privacy Guaranteed
Server learns only the aggregate sum, never individual inputs
Dropout Resilient
Protocol succeeds even if some clients disconnect
Cryptographically Secure
Uses Diffie-Hellman key exchange and Shamir Secret Sharing