Comparing Federated Learning Algorithms (FedAvg vs. FedProx) on Non-IID Data using NVIDIA FLARE

This tutorial walks through building and comparing the performance of the FedAvg and FedProx federated learning algorithms. The experiment is conducted on a non-IID CIFAR-10 setup, simulating realistic label imbalance across federated sites using a Dirichlet distribution for data splitting. The process leverages NVIDIA FLARE and the NVFlare Job API to manage the federated jobs, providing a practical guide for implementing and benchmarking these distributed training strategies.