Each node generates model updates (gradients or weights) after completing an epoch of training on local data.
Encryption of Model Updates:
The model updates are encrypted using the symmetric encryption key derived from the ECDH shared secret. This encryption step ensures that updates remain secure in transit.
Metadata for Secure Aggregation:
Along with the encrypted updates, metadata (e.g., node ID, epoch number) is transmitted to ensure the server can properly track and aggregate updates.
Transmission of Encrypted Updates:
The encrypted model updates, along with metadata and initialization vector (IV), are securely transmitted to Unity Network’s central server.