Experiments

1) Main experiments

We choose baseline hyperparameters and vary one at a time.

Possible baseline

• Aggregation: FedAVG
• Clients: 100
• Communication rounds: 20
• Clients sampled: 20
• Local data amount: 500
• Local epochs: 20
• alpha: 1
• Batch size: 48
• No noisy clients
• CIFAR10

Experiments

• Varying number of clients sampled
• Varying local epochs
• Varying alpha
• Varying noisy clients
• FedDF: Varying local epochs, varying alpha, varying noisy clients

2) Comparison to centralized learning

• Run baseline for "unlimited" communication rounds to show that after enough time, both FedAVG and FedDF can reach centralized

3) Different dataset

• Run baseline with a few of the variable variations