Federated Learning: Train Without Sharing Data (2D)

Federation setup

Client devices 10

Round speed 1.0×

Dropout rate 18%

Local epochs / round 2

Live stats

Round

Accuracy

—

Active clients

0 / 0

Gain / round

—

How a round works

Broadcast — the server sends the current global model to every client.
Local train — each client trains on its own private data for the chosen number of epochs; some drop out (stragglers / offline devices).
Upload — surviving clients send back only their model updates, never raw data.
Aggregate — the server averages the updates (FedAvg) into an improved global model, and accuracy climbs toward a noisy ceiling.

Broadcast

Server (global model) Active client Dropped-out client