flwr new @charles/paris-mnist-fl

MNIST: federated learning and statistics

Flower workshop app for the Paris FLTA Summer School.

The training demonstration works as supplied. For the statistics exercises, complete the four functions in mnist_fl/analytics_student.py. Data loading, messages and validation are provided. Nodes send summaries, rather than images.

Start with the setup guide, then follow the analytics exercises or the notebooks. Those instructions supply the federation and run configurations. A statistics run raises NotImplementedError until its student functions are completed.

The app uses a fixed 6,000-image MNIST training subset and 900 separate test images, loaded through Flower Datasets. The classroom uses deployed SuperNodes; the training and analytics handlers also support Flower simulations. A local simulation needs the app's dependencies, including the Flower simulation extra. It does not connect to the classroom's deployed nodes.

Keep publish-release = false for exercise runs. Charles supplies the shared global report for the later agent exercise. Local node summaries are visible to the coordinator; this app does not add secure aggregation or differential privacy.