MEDfl, A Collaborative Framework for Federated Learning in Medicine

Train clinical AI models across hospitals without moving data. MEDfl connects sites, orchestrates real-world and simulation experiments.

Documentation Quickstart GitHub

Real-World & Simulation FL

Simplifying research collaboration

From Setup to Breakthroughs

Connect sites securely, validate datasets, design pipelines, launch federated rounds, and analyze results—end to end.

1
Connect Clients Securely
Onboard hospitals via Tailscale VPN and WebSockets. Generate auth keys and scripts, then invite collaborators.
2
Build Your Network
Discover available clients, verify socket/VPN status, and select the cohort for training.
3
Validate Compatibility
Run dataset and system checks: schema, columns, nulls, stats, OS/GPU. Be green before you train.
4
Configure Pipelines
Drag-and-drop nodes (Model, Network, Optimize, Strategy). Toggle DP/TL, set rounds and metrics.
5
Review & Launch
Inspect the final configuration, confirm client readiness, then start federated rounds when minimum criteria are met.
6
Analyze & Export
Compare runs, visualize AUC/ROC and losses, then export artifacts and persist results for reproducibility.

Onboarding clients

1 / 6

Onboarding clients

Network assembly

Data stats and checks

Pipeline builder

Review configuration

Results and analytics

Federated Training in 4 Steps.

Define the idea, run simulations, validate results, then deploy on real distributed clients.

Define the experiment idea

Set objectives, datasets, and federated assumptions.

Run simulations

Evaluate models and strategies in a simulated FL environment.

Validate results

Inspect metrics and confirm experiment stability.

Deploy federated training

Execute the experiment on real distributed clients.

Federated Training in 4 Steps.

Define the idea, run simulations, validate results, then deploy on real distributed clients.

Define the experiment idea

Set objectives, datasets, and federated assumptions.

Run simulations

Evaluate models and strategies in a simulated FL environment.

Validate results

Inspect metrics and confirm experiment stability.

Deploy federated training

Execute the experiment on real distributed clients.

Medfl

Get started

Install, run, and federate in minutes

Use MEDfl as a Python package or install the desktop application. Start a server, connect clients, and track federated rounds.

Docs Installation GitHub

Install with pip

bash

pip install MEDfl

Requires Python 3.9+. Create a virtual environment for best results.

Video tutorials

Learn MEDfl with guided videos

Short, practical walkthroughs—from client onboarding and validation to pipelines, training, and results.

Watch on YouTube Quickstart

Beginner8:12

Introduction to the Federated Learning Module

Install MEDfl, configure your environment, and run your first experiment.

Intermediate05:37

Create and run federated learning pipelines

By the end of this video, you’ll manage and run your own federated learning experiments within MEDfl.

Intermediate15:05

MEDfl | Crash tutorial

Drag-and-drop pipelines and launch federated training end-to-end.

View all tutorials

Tutorials

Learn by building, step by step

Follow focused, practical guides to get MEDfl running — from pip install to multi-site federated training.

Tutorial

Install MEDfl (Python)

Set up a virtualenv and install MEDfl via pip in minutes.

Tutorial

Run the Desktop App

Install the desktop app on Windows, macOS, or Linux and connect to a server.

Tutorial

Start a Federated Server

Spin up a FedAvg server with Strategy and track 10 rounds of training.

Tutorial

Connect a Client

Configure XGBoost params, join the federation, and report metrics.

Tutorial

Tailscale / LAN Setup

Connect clients securely over Tailscale or on a local network.

Tutorial

Tailscale / LAN Setup

Connect clients securely over Tailscale or on a local network.