Introduction
NVIDIA FLARE (Federated Learning Application Runtime Environment) is an open-source, domain-agnostic Python SDK designed to facilitate privacy-preserving collaborative training in production environments. Released under the Apache 2.0 license, FLARE enables researchers, data scientists, and platform developers to adapt existing machine learning and deep learning workflows to federated learning paradigms. Its architecture supports both simulation and real-world deployment, making it a versatile tool for distributed multi-party collaboration.
The framework is built on a componentized architecture that allows users to transition from research and simulation to production deployment with minimal code changes.
The project emphasizes extensibility and modularity, offering built-in support for deep learning frameworks like PyTorch and TensorFlow, as well as traditional machine learning libraries such as scikit-learn and XGBoost.
How It Works
FLARE operates on a componentized architecture that separates concerns into modular components, allowing users to customize and extend functionality as needed. The framework is structured around three primary APIs: the Client API, the ModelController API, and the FedJob API. These APIs enable users to define federated learning jobs, convert centralized training scripts into federated code, and manage model workflows without requiring separate server files.
The Client API simplifies the transition from centralized to federated training by providing tools to adapt existing machine learning scripts. The ModelController API allows for the definition of custom federated learning algorithms, such as FedAvg, FedProx, or Scaffold, while the FedJob API enables Pythonic job configuration for training and validation workflows.
The framework also includes a simulator and proof-of-concept (POC) mode for rapid prototyping and debugging, allowing users to test federated learning applications locally before deployment.
Security and privacy are central to FLARE’s design. The platform enforces federated authorization and privacy policies, integrates differential privacy and homomorphic encryption, and supports confidential computing through technologies like Intel TDX and AMD SEV-SNP.
Deployment options for FLARE include cloud and on-premise environments, with support for Kubernetes, Docker, and Slurm clusters. The framework provides Helm charts for Kubernetes deployments and scripts for cloud provisioning on AWS and Azure. A dashboard component offers project management and monitoring capabilities, while the secure provisioning tool streamlines the setup of trusted federated learning environments.
Use Cases
FLARE’s domain-agnostic design makes it applicable across industries, including healthcare, finance, autonomous systems, and drug discovery. In healthcare, the framework has been used for multi-institutional medical imaging, COVID-19 research, and federated brain tumor segmentation. For instance, projects like Federated Learning for Prostate Segmentation and Federated Learning with Differential Privacy for BraTS18 segmentation demonstrate its utility in collaborative medical AI without sharing raw patient data.
The financial sector leverages FLARE for fraud detection and credit risk modeling. Examples include Privacy-Preserving Federated Fraud Detection in Payment Transactions and End-to-End Federated XGBoost for Financial Credit Card Detection. These applications highlight FLARE’s ability to handle sensitive data while maintaining regulatory compliance and privacy standards.
Autonomous vehicles benefit from FLARE’s cross-border training capabilities, enabling collaborative model development across geographically distributed data sources. Use cases in this domain include Federated Learning in Autonomous Vehicles Using Cross-Border Training, where diverse datasets from different regions are integrated without compromising data sovereignty.
FLARE also supports large language models (LLMs) and generative AI workflows. Additionally, FLARE integrates with NVIDIA’s BioNeMo platform for federated drug discovery, enabling collaborative research in protein classification and subcellular localization.
Beyond technical applications, FLARE has been adopted in academic and industry research, as evidenced by its inclusion in publications such as Equitable Federated Learning with NCA (MICCAI 2025) and Secure Federated XGBoost with CUDA-Accelerated Homomorphic Encryption (CAI 2025).
Key Features & Capabilities
- This flexibility accommodates a wide range of use cases, from healthcare to finance.
- Built-In Algorithms and Workflows: The framework includes implementations of federated learning algorithms such as FedAvg, FedProx, FedOpt, Scaffold, and Ditto.
- These mechanisms ensure that data remains secure during collaborative training, reducing the risk of leaks or unauthorized access.
- Confidential Computing Support: Recent updates introduce support for Intel TDX and AMD SEV-SNP, enabling confidential computing environments where data is encrypted in use. This feature is critical for high-security applications, such as healthcare and financial services.
- Agent-Assisted Development: Version 2.9.0 introduces agent skills for federated development, including conversion skills for PyTorch, PyTorch Lightning, and Hugging Face Trainer projects, as well as Auto-FL optimization for tuning existing jobs.
- Scalability and Deployment: FLARE supports deployment across cloud, on-premise, and hybrid environments, with native integration for Kubernetes, Docker, and Slurm clusters.
- Simulation and Prototyping: The built-in simulator and POC mode enable users to test federated learning applications locally before deployment.
- Extensibility and Customization: FLARE’s componentized architecture allows users to extend or modify core functionalities. The framework provides low-level APIs for custom components, such as peer-to-peer communication, and supports integration with third-party libraries and tools.
- Community and Ecosystem: FLARE benefits from an active open-source community, with contributions from researchers and developers worldwide. The project maintains comprehensive documentation, tutorials, and training resources, including DLI courses and curated learning paths for beginners and advanced users.
- Integration with Other Tools: FLARE supports interoperability with frameworks like Flower, MONAI, and BioNeMo, expanding its utility in specialized domains such as medical imaging and generative AI. Recent collaborations, such as the integration with Meta’s ExecuTorch, extend its applicability to mobile and edge devices.