NVIDIA FLARE

NVIDIA FLARE is a Python SDK for researchers and data scientists who want to train models across participating sites while keeping training data local. Platform developers can also use it to build systems for collaboration between organizations. It supports on-premise and cloud deployments and is open source under the Apache 2.0 license.

Teams can adapt existing machine learning workflows through its Client API rather than rewrite their training code around a new framework. FLARE works with PyTorch, PyTorch Lightning, TensorFlow, scikit-learn and XGBoost, covering deep learning as well as traditional machine learning. Clients train and validate locally, then send model updates back for shared training.

Its scope extends beyond training. FLARE supports federated statistics, global model evaluation and cross-site validation. It handles horizontal and vertical federated learning, with built-in algorithms including FedAvg, FedProx, Scaffold and Ditto. Teams can choose server-controlled or client-controlled workflows to suit how participants collaborate.

A simulator and proof-of-concept mode let teams test workloads before deploying them across sites. Its modular architecture supports custom components, while a dashboard handles project management and deployment. Privacy capabilities include differential privacy, homomorphic encryption and private set intersection. Federated authorization and privacy policies govern participation, and built-in fault tolerance helps workloads cope with failures.