securefederatedai/openfederatedlearning

What it solves

Open Federated Learning provides a way for multiple organizations to collaboratively train and validate machine learning models without sharing their sensitive raw data. This eliminates the need to move private information—such as medical records or financial data—to a central server, reducing privacy risks and data movement.

How it works

Instead of centralized learning, the framework uses a distributed approach where only model parameters and updates are exchanged between a central aggregator and local collaborators. It is backend-agnostic, supporting frameworks like PyTorch, TensorFlow, and Jax. Users can set up experiments via a TaskRunner API for secure communication (including mTLS and TEE support) or a Workflow API for simulating and scaling experiments.

Who it’s for

It is designed for researchers and organizations handling sensitive data across different locations, particularly in fields like medical imaging (e.g., the FeTS platform), though it is intended to be agnostic to the specific industry or use case.

Highlights

• Backend Agnostic: Works with PyTorch, TensorFlow, and Jax.
• Privacy-Preserving: Ensures raw data never leaves the local environment.
• Multiple Aggregation Algorithms: Built-in support for FedAvg, FedOpt, FedProx, and FedCurv.
• Secure Communication: Supports mTLS-based channels and confidential computing environments.