Federated Learning in Healthcare Using AWS FedML, Amazon EKS, and Amazon SageMaker
Healthcare organizations needed to collaboratively train machine learning models on distributed heart disease patient data across multiple organizations without compromising data privacy and security. FedML implemented federated learning using the FedML framework on AWS, employing Amazon EKS for infrastructure and Amazon SageMaker for experiment tracking and model monitoring. This approach enabled training models locally on sensitive data, sharing only model parameters, thereby enhancing data privacy while improving patient outcomes by accurately predicting heart disease without exposing sensitive data.
- Organization
- FedML
- Industry
- Healthcare
- Location
- United States
- Published
- March 2024
Reported outcomes
Strategic outcomes
Primary read
Use case focus
Showing 3 of 3
- 1Federated Learning
- 2Machine Learning Model Training
- 3Healthcare AI
- Healthcare organizations face challenges in collaboratively training machine learning models on distributed sensitive patient data while maintaining strict data privacy and security compliance, especially under regulations like HIPAA.
- Sharing raw patient data across organizations risks privacy leaks, potentially limiting the accuracy of predictive models due to restricted data availability.
- There is a need for scalable, secure, and compliant infrastructure to support federated learning that collaborates without exposing raw data.
- FedML leveraged its open-source federated learning framework deployed on AWS using multiple Amazon EKS clusters to securely train ML models across organizations without sharing raw data.
- Amazon SageMaker was integrated for experiment tracking and monitoring performance of client and centralized aggregator models.
- The solution used FedML Octopus and MLOps platforms to enable scalable, secure distributed training and collaborative ML development between multiple healthcare organizations.
- Infrastructure deployment was automated with EKS Blueprints for Terraform, enabling consistent environments across AWS accounts and regions.
- Improved patient outcomes by enabling accurate heart disease prediction models trained on distributed data sources without compromising patient privacy.
- Enhanced data privacy and security in healthcare ML workflows through decentralized federated learning.
- Fostered collaboration among healthcare organizations overcoming data silos while complying with industry regulations.
Architecture
The architecture involves deploying FedML federated learning framework across multiple Amazon EKS clusters, integrating with Amazon SageMaker for model experiment tracking and monitoring. FedML Octopus provides secure distributed training. Infrastructure is provisioned with EKS Blueprints for Terraform in multiple AWS accounts and regions.
Sources & evidence1
AI-generated summary. Verify important details with the linked sources before relying on this case.