NVIDIA FLARE for Federated Learning

NVIDIA FLARE enables federated learning to allow for multiple clients each with their own data to collaborate on training together without having to share their actual data.

Previously, a federated learning solution was built into Clara Train in versions before Clara Train 4.0, but now NVIDIA FLARE (NVIDIA Federated Learning Application Runtime Environment) is its own open-sourced and more generalized package for federated learning that can be used in contexts outside of Clara as well. See the NVIDIA FLARE documentation for details.

MMARs as Applications in NVIDIA FLARE

MMARs can be extended to work as applications in NVIDIA FLARE with the addition of the configurations for federated learning, config_fed_server.json and config_fed_client.json. Additionally, if cross site model evaluation is configured to run, config_cross_validaiton.json is needed.