Federated learning

This course teaches theoretical foundations and efficient algorithms for federated learning (FL) applications.

FL is an umbrella term for training machine learning models in a collaborative fashion from distributed collections of data. These collections can be modelled as a graph whose nodes represent computational units (such as a smartphone). FL techniques are privacy-friendly as they do not require to share raw data between nodes but only model parameter updates. You will learn how to formulate FL applications as a regularised empirical risk minimisation and solve it using distributed implementations of gradient descent.

Course contents

• Multi-task learning
• Complex networks
• Clustering
• Privacy-preserving machine learning
• Large-scale machine learning
• Python (Flask, scikit-learn)

Learning outcomes

After successfully completing this course, the student

• can model networked data and models using concepts from graph theory.
• can formulate FL problems as optimisation problems.
• is familiar with distributed optimisation methods (gradient-descent, primal-dual).
• can implement FL methods in Python.

Course material

We provide a web server for Python programming (JupyterHub). Students only need a computer with a web browser running that is connected to the internet.

Course book: A. Jung, “Machine Learning: The Basics”, Springer, Singapore, 2022.

Completion methods

Independent study, assignments, project-work and peer-grading.

Students can collect points from different activities:

• programming assignments
• theory questions (quizzes)
• student project
• oral exam (via Zoom)