392041 Learning from Distributed Data (S) (SoSe 2026)

This seminar focuses on learning from distributed data, an increasingly important setting in which data is stored across mulitple locations or devices and cannot be centrally aggregated due to privacy, ownership, or communication constraints.

The course introduces fundamental and advanced methods for learning in such distributed settings, including distributed learning, federated learning, and federated analytics. Central issues include the impact of statistical heterogeneity, privacy-preserving techniques, as well as trade-offs between accuracy, efficiency, and robustness. The seminar buils on a brief recap of machine learning fundamentals and emphasizes the connection of current research and real-world deployments.

Each session consists of an instructor-led introduction to a specific paradigm or method, followed by different interactive formats such as guided hands-on experiments, discussion of selected research papers, or group work. In addition, students work independently on a project in which they address a self-selected real-world or research problem related to learning from distributed data. The project includes a short mid-semester pitch, a final presentation, and a written report.

After completing the seminar, students will be able to understand and compare different paradigms for learning from distributed data, select appropriate methods for concrete problems, and critically assess recent research in this field.

Requirements for participation, required level

Prerequisite:
- Familiarity with Python and ideally with PyTorch
- Fundamentals of Machine Learning and Deep Learning