Experimenting with federated learning: implementing state-of-the-art methods with fluke

Federated Learning (FL) is a transformative paradigm that enables collaborative model training across decentralized data sources while preserving data privacy. By reducing the need for raw data exchange, FL offers a scalable and efficient approach to machine learning in distributed environments. 

However, beginners usually face a huge problem: How to approach FL? Nowadays, there are plenty of theoretical results that continuously push the frontier of FL systems understanding. At the same time, beginners and experienced practitioners are bothered by which tool should be used to approach this topic or to test a novel algorithm easily. In this sense, also system design research goes on, with the ongoing introduction of new software tools. In this scenario, many people can be confused about how to approach FL. We would like to address these issues with this tutorial. 

In first place, we will introduce the topic of FL for beginners, to create a common knowledge base. Thanks to this common knowledge base, the attendees will have clear in mind which are all the components of a FL system that can be tackled to design new methodologies. In particular, we will present also some state-of-the-art algorithms, that will be used in the second part of the tutorial. 

In the second part of the tutorial, we will introduce fluke (Federated Learning Utility frameworK for Experimentation and research), a simulation framework made by researchers for researchers. This second part represents an interesting opportunity for both beginners and experienced practitioners to gain experience with a tool that makes the deployment of an FL algorithm very simple. In fact, fluke is designed to be flexible, easy to use, and easy to extend, and can be used to benchmark a wide variety of federated learning algorithms. Attendees will gain a comprehensive understanding of Fluke's architecture and functionality, followed by a hands-on session where we implement a state-of-the-art FL method live. This tutorial is designed for researchers, practitioners, and developers interested in leveraging FL for privacy-preserving machine learning applications. 

Materials available at this GitHub repository!

schedule

Introduction to Federated Learning (45 minutes)

Overview of federated learning and its significance

 Key challenges and approaches

Applications in various domains

State-of-the-art methods and novel algorithms       

Introduction to Fluke (30 minutes)

Motivation and goals of the Fluke framework

    Core architecture and design principles

Key functionalities and extensibility

Hands-On Session: Implementing FL Methods (90 minutes)

 Setting up Fluke: Installation and basic configuration

Implementing a modern federated learning technique

Running and evaluating the implementation

Q&A and Discussion (15 minutes)

Target Audience

This tutorial is ideal for both beginners and experienced AI scientists/researchers interested in federated learning and its applications. Familiarity with Python and machine learning concepts is recommended but not required.

EXPECTED OUTCOMES

• Understanding of Federated Learning’s fundamental concepts.

• Implement and evaluate a state-of-the-art federated learning method using Fluke.

• Be equipped with the knowledge to extend and apply federated learning to their own projects.

TUTORIAL'S ORGANIZERS