[Paper Review] OpenFL: An open-source framework for Federated Learning

G. Anthony Reina, Alexey Gruzdev+16 more|arXiv (Cornell University)|May 13, 2021

Privacy-Preserving Technologies in Data18 references58 citations

TL;DR

OpenFL is an open-source, production-ready framework enabling data-private federated learning across collaborators and aggregators, supporting TensorFlow and PyTorch with a secure, extensible design. It demonstrates federation in healthcare and a related computational competition use case.

ABSTRACT

Federated learning (FL) is a computational paradigm that enables organizations to collaborate on machine learning (ML) projects without sharing sensitive data, such as, patient records, financial data, or classified secrets. Open Federated Learning (OpenFL https://github.com/intel/openfl) is an open-source framework for training ML algorithms using the data-private collaborative learning paradigm of FL. OpenFL works with training pipelines built with both TensorFlow and PyTorch, and can be easily extended to other ML and deep learning frameworks. Here, we summarize the motivation and development characteristics of OpenFL, with the intention of facilitating its application to existing ML model training in a production environment. Finally, we describe the first use of the OpenFL framework to train consensus ML models in a consortium of international healthcare organizations, as well as how it facilitates the first computational competition on FL.

Motivation & Objective

• Motivate the need for federated learning to enable collaboration without sharing sensitive data.
• Describe the OpenFL architecture, workflow, and security design for real-world deployments.
• Showcase healthcare-focused federated learning use cases and a computational competition.
• Explain installation, operation, and practical guidance to port existing ML workflows to FL.

Proposed method

• Describe the star-topology federation with collaborators and aggregators and the use of mutually authenticated TLS for communications.
• Introduce the concept of an FL plan (a YAML file) that defines tasks, settings, and RPCs for federation coordination.
• Explain the OpenFL components: collaborators, aggregator, backend, and how updates are aggregated into a global model.
• Detail the two primary interfaces for running federations: Python API and fx CLI, including tutorials and examples.
• Provide guidance on deployment options (bare-metal and Docker) and security enhancements (PKI, TEEs).
• Present experimental workflows including a TensorFlow/Keras MNIST tutorial and future interactive API work.

Experimental results

Research questions

• RQ1 How can an open-source framework enable secure, data-private federated learning across multiple institutions?
• RQ2 What architectural design and security mechanisms are necessary to support scalable, real-world FL deployments?
• RQ3 How can FL be used to advance healthcare collaborations, such as brain tumor segmentation, while preserving data privacy?
• RQ4 What are practical workflows for deploying and operating federated trainings using OpenFL (installation, federation setup, and execution)?

Key findings

• OpenFL supports training with TensorFlow and PyTorch across remote collaborators while keeping data local and only model updates exchanged.
• The FL plan governs federation behavior, including tasks, rounds, and aggregation, enabling repeatable, multi-institutional trainings.
• The platform emphasizes security through mutually authenticated TLS, PKI certificates, and optional trusted execution environments (TEEs).
• FeTS demonstrates OpenFL in a real-world multi-institution healthcare federation for brain tumor segmentation and collaborative model development.
• OpenFL provides tutorials and examples (including a Keras MNIST tutorial) and multiple deployment options (bare-metal and Docker) for practical adoption.
• The work outlines a framework for a first computational competition on federated learning to advance the FL research community.