Teaching

See TUMonline
Note: Registration is only possible via the matching platform. The preliminary meeting takes place on July 19, at 10 am.

By the end of the seminar, students have studied a specific problem in the federated machine learning domain and presented ideas on how to solve it (experimentally or with literature). In the seminar, students will learn about scientific paper writing techniques, the peer review process, and presentation of results.

Machine Learning is omnipresent and has made huge leaps forward in the last decade. More recently both scholars and practitioners have put emphasis on increasing privacy and data protection in machine learning pipelines. A technique that came up in 2016 is Federated Machine Learning (FL), which enables user groups to collaboratively train ML models without sharing their individual data with others (McMahan et al., 2016). The FLIP seminar focuses on industrial applications of FL and, more generally, critical components in FL pipelines (e.g., communication efficiency).

It is expected that students have a general understanding of ML (pipelines) and have familiarized themselves with the basic principles of distributed systems. The course topics are suitable for all experience levels.

Conference-like structure with writing your own report, reviewing one from another group, and presenting your findings with a presentation.

Students will have to present three deliverables individually or in groups of two: - Short paper - Peer review for another student group - Final presentation of their results

Introductory readings: - What is federated learning? https://ai.googleblog.com/2017/04/federated-learning-collaborative.html - The first publication on FL: https://arxiv.org/abs/1602.05629 - Advances and Open Problems in Federated Learning: https://arxiv.org/abs/1912.04977 - Comprehensive work on FL at the edge for predictive maintenance recently published: https://arxiv.org/pdf/2109.12375.pdf