Standard machine learning methods involve storing training data on a single machine or in a data center. Federated learning is a privacy-preserving technique that is especially useful when the training data is sparse, confidential, or less diverse.
NVIDIA open-source NVIDIA FLARE, which stands for Federated Learning Application Runtime Environment. It is a software development kit that enables remote parties to collaborate for developing more generalizable AI models. NVIDIA FLARE is the underlying engine in the NVIDIA Clara Train’s federated learning software, which has been utilized for diverse 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 give consumers the distributed infrastructure needed to create a multi-party collaborative application.
Participants in federated learning collaborate to train or evaluate AI models without needing 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 Roche Digital Pathology researchers on a successful internal simulation using whole slide images for classification
- It also worked with Erasmus Medical Center in the Netherlands on an AI application identifying genetic variants associated with schizophrenia cases.
However, the server-client architecture is not appropriate for all federated learning applications. NVIDIA FLARE will make federated learning more accessible to a broader range of applications by supporting other architectures. The possible use cases include helping:
- energy corporations analyze seismic and wellbore data,
- Manufacturers optimize industrial processes
- Financial firms improve fraud detection algorithms
The ability to speed federated learning research by open-sourcing NVIDIA FLARE is especially relevant in the healthcare sector, where access to multi-institutional datasets is critical, but patient privacy concerns can hinder data sharing.
NVIDIA FLARE can work with current AI projects, such as the open-source MONAI medical imaging platform. 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 to radiology images for breast cancer and COVID-19 applications. It wants to make NVIDIA FLARE available through the ACR AI-LAB, a software platform used by society’s tens of thousands of members.
- Flywheel’s Flywheel Exchange platform allows users to access and exchange data and biomedical research techniques, manage federated projects for training and research and select their chosen federated learning solution, including NVIDIA FLARE.
- Taiwan Web Service Corporation offers a GPU-powered MLOps platform that allows users to use NVIDIA FLARE to execute federated learning.
- The NVIDIA Inception program partner, Rhino Health, has integrated NVIDIA FLARE into its federated learning solution, assisting researchers at Massachusetts General Hospital in developing an AI model that more accurately diagnoses brain aneurysms. In addition, it helps experts at the National Cancer Institute’s Early Detection Research Network in developing and validating medical imaging AI models that detect early signs of pancreatic cancer.
By making NVIDIA FLARE open source, researchers and platform developers will have additional options to personalize their federated learning solutions, enabling cutting-edge AI in practically any industry.
The post NVIDIA Open-Source ‘FLARE’ (Federated Learning Application Runtime Environment), Providing A Common Computing Foundation For Federated Learning appeared first on MarkTechPost.