Fig. 9.

The model architecture of federated learning. a Global model (in central server) implemented with average CNN model. b 50 local models made as a copy of the global model (same weight). c 50 clients built with local data. d Training & testing with clients’ data, scaling the weights, and storing these into a variable. e Aggregating all different weights and making average. Last, sending the updated weight to the central server. The iteration went through 5 communication rounds. f For each communication round, the test accuracy of the global model is checked