NVIDIA Open Source “FLARE” (Federated Learning Application Runtime Environment), providing a common computing foundation for federated learning
Standard machine learning methods involve storing training data on a single machine or in a data center. Federated learning is a confidentiality preservation technique that is particularly useful when training data is scarce, confidential, or less diverse.
NVIDIA open source NVIDIA FLARE, which stands for Federated Learning Application Runtime Environment. It is a software development kit that allows remote parties to collaborate to develop more generalizable AI models. NVIDIA FLARE is the underlying engine for NVIDIA Clara Train’s federated learning software, which has been used for various AI applications such as medical imaging, genetic analysis, cancer, and COVID-19 research.
Researchers can use the SDK to customize their method for domain-specific applications by choosing from a variety of federated learning architectures. NVIDIA FLARE can also be used by platform developers to provide consumers with the distributed infrastructure needed to build a multi-party collaborative application.
Federated learning participants collaborate to train or evaluate AI models without the need to share or pool their private datasets. NVIDIA FLARE supports a variety of distributed architectures, including peer-to-peer, cyclic, and server-client techniques, among others.
NVIDIA FLARE has been used in two federated learning collaborations:
- NVIDIA collaborated with researchers at Roche Digital Pathology on a successful internal simulation using whole slide images for classification
- He also worked with Erasmus Medical Center in the Netherlands on an AI application identifying genetic variants associated with cases of schizophrenia.
However, the server-client architecture is not appropriate for all federated learning applications. NVIDIA FLARE will make federated learning more accessible to a wider range of applications by supporting other architectures. Possible use cases include help:
- energy companies analyze seismic and wellbore data,
- Manufacturers optimize industrial processes
- Financial firms improve fraud detection algorithms
The ability to accelerate federated learning research through open source NVIDIA FLARE is particularly relevant in the healthcare industry, where access to multi-institutional datasets is critical, but privacy concerns of patients can interfere with data sharing.
NVIDIA FLARE can work with current AI projects, such as the open source medical imaging platform MONAI. It will also be deployed in the following areas to power federated learning solutions:
- The American College of Radiography (ACR) has collaborated with NVIDIA on federated learning research that uses artificial intelligence for radiological images for breast cancer and COVID-19 applications. He wants to make NVIDIA FLARE available through ACR AI-LAB, a software platform used by tens of thousands of members of the company.
- Flywheel’s Flywheel Exchange platform enables users to access and exchange biomedical research data and techniques, manage federated training and research projects and select the federated learning solution of their choice, including NVIDIA FLARE.
- Taiwan Web Service Corporation offers a GPU-powered MLOps platform that allows users to use NVIDIA FLARE to perform federated learning.
- NVIDIA Inception program partner Rhino Health has integrated NVIDIA FLARE into its federated learning solution, helping researchers at Massachusetts General Hospital develop an AI model that more accurately diagnoses brain aneurysms. Additionally, he helps experts at the National Cancer Institute’s Early Detection Research Network develop and validate AI medical imaging models that detect early signs of pancreatic cancer.
By making NVIDIA FLARE open source, researchers and platform developers will have additional options to customize their federated learning solutions, enabling cutting-edge AI in virtually any industry.