FLorist logo FLorist

FLorist is a platform to launch and monitor Federated Learning (FL) training jobs. Its goal is to bridge the gap between state-of-the-art FL algorithm implementations and their applications by providing a system to easily kick off, orchestrate, manage, collect, and summarize the results of FL4Health training jobs.

As Federated Learning has a client and a server side, FLorist also has client and server-side “modes” to orchestrate the training process on both sides. When FLorist’s client long-running process is started, they will be waiting for instructions from FLorist’s server to start FL clients for training. Once FLorist’s server starts the FL server, it sends instructions to FLorist’s clients to start their own FL clients. Then, FLorist’s server monitors the FL server and clients processes while collecting their progress to be displayed in the web UI.

At the end of training, it saves the results in a database and also provide access to the training artifacts (e.g. model files). For a visual representation of the system, please check the diagram below.

system_diagram.png

Use Cases

1. Facilitate the orchestration of FL4Health training process

Scenario: The process of training an FL4Health model is manual and cumbersome, requiring a lot of technical and programming knowledge.

Steps:

  • Provide an easy-to-use UI to set up the parameters of a training job

  • Display the progress and preliminary metrics while training is happening

  • Display the results if successful, or centralized error messages and logs if not

  • Make the training artifacts (e.g. models) easily accessible in a centralized place

2. Facilitate the use of state-of-the-art FL implementations

Scenario: State-of-the-art algorithms implemented in FL4Health are not easy to be adopted because of the relatively high learning curve.

Steps:

  • By providing an easy-to-use UI, the learning curve to adopt FL4Health is reduced

  • Providing a centralized place to see the progress, access the logs and the training artifacts also lowers the barrier of adoption

  • Having a system where training jobs can be easily restarted with different parameters facilitates experimentation

3. Provide a useful tool to the open source community

Scenario: There is a scarcity of tools for orchestrating and managing FL training jobs in the OSS community.

Steps:

  • Develop a high-quality, easy-to-use and extensible tool for managing FL training jobs in FL4Health and later Flower in general

  • Use the latest and greatest front-end and back-end technologies and code practices

  • Provide a robust and visually enticing user interface