Table 8 Key steps in federated learning workflow.

From: Federated learning with differential privacy for breast cancer diagnosis enabling secure data sharing and model integrity

Step

Description

Client selection

Randomly selecting a subset of clients for each training round to optimize computation efficiency

Local model training

Clients train their models independently on local datasets

Gradient clipping

Restricting gradient magnitudes to prevent excessive influence from single data points

Differential privacy (DP)

Adding Gaussian noise to gradients to ensure privacy preservation

Model update transmission

Clients send noise-protected model updates to the central server

Global model aggregation

Server aggregates all updates using the Federated Averaging (FedAvg) algorithm

Model synchronization

Updated global model is redistributed to clients for further training

Back to article page