Fed-DNN-Debugger: Automatically Debugging Deep Neural Network Models in Federated Learning

Figure 2

Architecture of Fed-DNN-Debugger. Debugging the client model to repair its weights. The gradients after local debugging are encrypted by homomorphic encryption and sent to the server. The server receives the encrypted gradients for secure aggregation and then returns the aggregated gradients to each client. The clients receive the returned gradients, update their local models, and perform tests.