Step-by-Step Guide to Build and Compare FedAvg and FedProx Federated Learning on Non-IID CIFAR-10 with NVIDIA FLARE

A tutorial shows how to use NVIDIA FLARE to compare FedAvg and FedProx on a non-IID CIFAR-10 federated learning setup.

It partitions CIFAR-10 across simulated clients with a Dirichlet split to mimic realistic data imbalance.

The experiment uses NVIDIA FLARE’s Job and Client APIs to train, exchange, and aggregate models over multiple rounds.

It benchmarks global accuracy across communication rounds to highlight how the two algorithms behave under non-IID conditions.