FEDn: Federated learning from research to reality

Federated Learning: From Research to Reality

Find all the tools and resources you need in one place.

FEDn is a developer-friendly framework that supports the entire innovation process, from personal testing and proof-of-concepts to real-world, geographically distributed federated learning. Data scientists can use their ML code with FEDn to shift from centralized to federated systems without code changes and minimal DevOps effort.

Get started

Begin with the SDK using our quick start guide, or explore examples and frameworks on GitHub.

Getting Started

The best way for data scientists and ML professionals to learn FEDn is by registering for a free personal account and starting with the comprehensive Getting Started tutorial. It takes approximately 30 minutes to complete.

Get Started

FEDn GitHub

Join other developers on GitHub to explore more examples with ML frameworks like TensorFlow, PyTorch, scikit-learn, and Hugging Face, and contribute to open-source client APIs in Python, C++, and Kotlin.

Explore

A video introduction to FEDn

This video covers the following three sections related to the FEDn framework:

Where to find information about the FEDn framework.
A step-by-step guide to your first federated machine learning project using the FEDn framework.
The terminologies used in the FEDn framework.

Watch the rest of the FEDn video series here »

Learn More

FEDn is a federated learning platform that offers strong technical features and flexible deployment options. It provides scalable model coordination, improved security, and detailed monitoring. The platform supports major ML frameworks like PyTorch and TensorFlow and follows cloud-native principles. FEDn can be used as a fully managed SaaS solution for quick projects or as a self-hosted option for organizations needing more security and control. It easily deploys to Kubernetes clusters, making it suitable for public clouds and on-premise setups.

Architecture Overview

FEDn is a modular and distributed framework designed to enable federated learning across heterogeneous computing environments, from cloud infrastructure to edge devices. Its flexible architecture supports scalable model training and aggregation while maintaining control and coordination through a hierarchical system of controllers, combiners, and distributed clients.

Tier 1: Control management (cloud, near edge)The global controller serves as the central coordination point, managing service discovery across the system. Model storage is handled through S3, while a database system maintains essential operational data and state information.

‍

Tier 2: Model aggregation (cloud, near edge, far edge)Combiners play a crucial role in model aggregation, with the flexibility to deploy one or multiple instances for optimal load balancing and geographical proximity to client groups. In scenarios involving multiple Combiners, a Reducer takes on the responsibility of aggregating combiner-level models into a cohesive whole.

‍

Tier 3: Geographically distributed clients (near edge, far edge, device edge)

Core Features

Scalability and resilience

FEDn boosts federated learning by efficiently coordinating client models and aggregating them across several servers, catering to high client volumes and ensuring robust recovery from failures. It supports asynchronous training, smoothly handling client connectivity changes.

Security

FEDn enhances security in federated learning environments by eliminating the need for clients to open ingress ports and using standard encryption protocols and token authentication. This approach streamlines deployment across varied settings and ensures secure, easy integration.

Real-time monitoring and analysis

With comprehensive event logging and distributed tracing, FEDn enables real-time monitoring of events and training progress, facilitating easier troubleshooting and auditing. The API offers access to machine learning validation metrics from clients, allowing for detailed analysis of federated experiments.

Cloud native

By following cloud-native design principles, we ensure a flexible range of deployment options that cater to various needs, including private cloud environments, on-premise infrastructure, and hybrid setups. This approach ensures FEDn can be integrated across diverse deployment scenarios.

Framework Agnostic

FEDn is designed to be ML-framework agnostic, seamlessly supporting major frameworks such as Keras, PyTorch, Tensorflow, Huggingface and scikit-learn. Ready-to-use examples are provided, facilitating immediate application across different ML frameworks.

Flexible Deployment Options

FEDn Studio deploys to any standard Kubernetes cluster. This ensures flexible hosting options ranging from public and private cloud, to on-premise clusters and single workstations.

Software as a Service

We provide a fully managed cloud-hosted instance of FEDn. Requiering no server-side DevOps, this is the easiest way to get started with FL. By keeping all focus on the ML problem client-side, this is deal for early-stage projects, including pilots and proof-of-value phases.

Self-hosted

For organizations with stringent cybersecurity needs, FEDn can be deployed on private clouds / VPCs or on-premise, allowing full control over the deployment and enhanced security and privacy.

Tutorials

Highlighted examples and guides.

Custom project

A guide on how a FEDn project is structured, and details how to develop your own project. Basic understanding of the FEDn needed.

My Project

Flower ClientApps in FEDn

This guide will show you how to run a Flower ‘ClientApp’ on FEDn and integrate it with FEDn’s federated learning framework.

View Guide

Spam Detection

This example project demonstrates how one can make use of the popular Hugging Face ‘Transformers’ library in FEDn.

Spam Detection

Backed by Industry Leaders

Accelerating federated learning innovation through strategic partnerships with technology leaders.

NVIDIA® Inception Member

As a member of NVIDIA® Inception, we leverage cutting-edge technology and resources to accelerate our innovation in federated learning and AI. This program connects us with technical expertise and a global network of AI pioneers.

Intel® Partner Alliance Member

Our Intel® Partner Alliance membership provides access to advanced technical resources and optimization tools, enabling us to deliver high-performance federated learning solutions across diverse computing environments.

Questions & Answers

FAQ in the documentation »

Why should we choose your FL framework over other options?
Our framework offers an easy-to-use interface, visual aids, and collaboration tools for ML/FL projects, with features like distributed tracing and event logging for debugging and performance analysis. It ensures security through client identity management and authentication, and has scalable architecture with multiple servers and load-balancers. FEDn also allows flexible experimentation, session management, and deployment on any cloud or on-premises infrastructure.
Is this yet another ML platform we have to install?
FEDn is a versatile framework that can be extended, configured, and integrated into existing systems to tailored to your environment. For effective Federated Learning (FL) management, deployment of server-side components and charts is necessary. It enhances rather than replaces your current setup.
What deployment options does FEDn offer?
FEDn offers two main deployment options to cater to different organizational needs and project stages. The fully-managed SaaS (Software as a Service) model simplifies access to federated learning technology, making it ideal for early-stage projects, pilots, and proof-of-value phases. For organizations with strict cybersecurity requirements, FEDn can be deployed on private clouds or on-premise, providing full control over deployment, security, and privacy. This self-managed option is particularly suitable for advanced security needs, such as protecting IP-sensitive ML models.
What is federated learning?
Federated learning (FL) overcomes the limitations of centralized machine learning by training models on data spread across different locations, preserving privacy and complying with regulations. This decentralized approach enables secure, efficient, and scalable machine learning without moving the data, useful for managing the growing complexity and volume of data in a connected world. Learn more »

Can we build our own IP using your framework?
Absolutely. You can develop your own IP without any conflict. Utilize our framework and Scaleout’s expertise to accelerate your project. There's no risk of lock-in, as our Software Development Kit (SDK) for integration is licensed under Apache2. We're confident you'll find value in our support services, warranty, indemnification, and comprehensive toolkit.
How can I explore FL without deep technical expertise?
We offer a cloud-hosted FL platform for easy FL exploration, optimized for cost and ideal for R&D. Scaleout enables data scientists to investigate FL without initial IT/DevOps resources. We provide a smooth transition to self-hosted production with enterprise integrations, ensuring your PoC is scalable, secure, and representative of real-world scenarios.
What capabilities does FEDn offer to support my organization's federated learning needs?
FEDn supports a range of capabilities to meet diverse organizational demands. It offers both multi-tenant and single-tenant options, allowing organizations to choose the configuration that best suits their needs. Additionally, FEDn takes care of complex operations management, including server aggregation, data storage, user authentication, network configuration, and system monitoring. This reduces the operational burden on users, enabling them to focus on developing and refining their federated learning.
What is edge AI?
Edge AI is an emerging field that combines artificial intelligence with edge computing, enabling AI processing directly on local edge devices. It enables real-time data processing and analysis without constant reliance on cloud infrastructure. This approach has gained momentum as the tech landscape undergoes a major transformation, driven by the exponential increase in data generated at the edge, thanks to IoT devices, sensors, and other connected technologies. Learn more »