Skip to content

HicResearch/MELODY

Repository files navigation

MELODY : federated Machine Learning fOr dermatologY

MELODy Logo MELODY is a project funded under the DARE UK Real World Exemplar Programme. The aim of the project is to develop and test federated machine learning approaches for dermatology research using clinical images from NHS Tayside and Oxford University Hospitals. By training AI models across multiple TREs without centralising data, the project aims to support the development of more inclusive and representative dermatology AI systems.

Preamble & Terminology

MELODY uses the Flower Framework to enable federated machine learning (FML). Details on how to set up a Flower network are detailed in the Flower.AI Setup document.

MELODY uses a custom wrapper around the existing flwr tooling to provide

  • Enhanced auditing
  • Information to improve TRE/SDE Egress

Benefits

Enhanced Auditing

TRE/SDE Egress Information

Installation & Use

The MELODY cli tool is available from the Release section of this Github repo. It is built as a linux standalone application, and can be used simply by running

./cli

Configuration

To maximise the benefit of the MELODY cli, there are several configuration options to set up:

RO-Crates

RO-Crates are a method to package up your ML jobs with their metadata. To set up RO-Crates to work with MELODY:

  1. Create a .toml config file with your RO-Crate details See the sample config example for more details.
  2. When you next run a MELODY job, an RO-Crate will be generated detailing everything about the run.

Logging

MELODY provides advanced logging for FML jobs. These logs are automatically generated and stored in ~/.mldy/melody.log This can be overwiitten via the config file.

CLI Commands

Run

List

Log

Pull

Stop

Contributing

We welcome contributions. Please see our contributing guide to get started.

Acknowledgements

About

MELODY: federated MachiEne Learning fOr dermatologY // TREvolution 2026 Real World Exemplar

Resources

License

Contributing

Stars

0 stars

Watchers

1 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages