Release_Phase2.0

Sequence Overview

Client:

Train the local model. Split the updated weights into shares using Shamir’s secret sharing mechanism. Apply Differential Privacy (add noise) to each share. Send noisy shares to different servers.

Server:

Receive noisy shares from clients. Perform aggregation using SMPC technique. Reconstruct the global model from the aggregated shares. Update the global model with new weights.