Simulate Federated Learning with compressed communication on a large number of Clients.

Recreate experiments described in Sattler, F., Wiedemann, S., Müller, K. R., & Samek, W. (2019). Robust and Communication-Efficient Federated Learning from Non-IID Data. arXiv preprint arXiv:1903.02891.

First, set environment variable 'TRAINING_DATA' to point to the directory where you want your training data to be stored. MNIST, FASHION-MNIST and CIFAR10 will download automatically.

python federated_learning.py

will run the Federated Learning experiment specified in

federated_learning.json.

You can specify:

• "dataset" : Choose from ["mnist", "cifar10", "kws", "fashionmnist"]
• "net" : Choose from ["logistic", "lstm", "cnn", "vgg11", "vgg11s"]

• "n_clients" : Number of Clients
• "classes_per_client" : Number of different Classes every Client holds in it's local data
• "participation_rate" : Fraction of Clients which participate in every Communication Round
• "batch_size" : Batch-size used by the Clients
• "balancedness" : Default 1.0, if <1.0 data will be more concentrated on some clients
• "iterations" : Total number of training iterations
• "momentum" : Momentum used during training on the clients

• "compression" : Choose from [["none", {}], ["fedavg", {"n" : ?}], ["signsgd", {"lr" : ?}], ["stc_updown", [{"p_up" : ?, "p_down" : ?}]], ["stc_up", {"p_up" : ?}], ["dgc_updown", [{"p_up" : ?, "p_down" : ?}]], ["dgc_up", {"p_up" : ?}] ]

• "log_frequency" : Number of communication rounds after which results are logged and saved to disk
• "log_path" : e.g. "results/experiment1/"

Run multiple experiments by listing different configurations.

• --schedule : specify which batch of experiments to run, defaults to "main"

Paper

Sattler, F., Wiedemann, S., Müller, K. R., & Samek, W. (2019). Robust and Communication-Efficient Federated Learning from Non-IID Data. arXiv preprint arXiv:1903.02891.