In the upcoming winter term, we offer three elective courses: Advanced Natural Language Processing (ANLP, Master), the Virtual Reality Projekt (VRP, Bachelor, new topic: health and sports) and, for the first time, KI-unterstütztes Softwaredesign und -engineering (KDE, Bachelor, 10 CP).
Come to the introductory sessions in the Schnupperwoche and see for yourself:
- Virtual Reality Projekt — Tuesday, September 22, 15:20, room A012
- Advanced Natural Language Processing — Tuesday, September 22, 17:00, room A212
- KI-unterstütztes Softwaredesign und -engineering — Friday, September 25, 15:20, room A108
Advanced Natural Language Processing
ANLP is held by Kai Eckert. When computers understand language: you work with modern AI, build applications such as chatbots or translators, and learn how machines analyze and generate text. The lecture starts from the classic methods of NLP — text representation, n-gram language models, classification and evaluation — and uses them as a point of comparison throughout. On that basis, it covers deep learning for NLP with word embeddings, sequence models and attention, the transformer architecture and the pre-training of large language models, and how these models are adapted through fine-tuning, parameter-efficient training, instruction tuning and alignment.
The second half is about building and judging real systems: inference and decoding, prompting and in-context learning, retrieval-augmented generation and tool-using agents, and how to evaluate all of that with your own test data, with metrics beyond accuracy and with a serious error analysis. Ethics, safety and regulation are part of the course, as are the current developments of a field that does not stand still. Every topic block comes with practical exercises, and a group project runs through the whole semester: the same self-chosen task is solved twice — first with classic methods as a reference, then with a language-model-based system that is measured against it. ANLP is a Master course, but it is also open to committed Bachelor students, where it replaces the Natural Language Processing course: the semester starts with a quick pass through the classic methods, and any remaining gaps in NLP are to be closed on your own. Foundations in machine learning as well as Python are expected; if single pieces are missing, that need not be a problem, but you are expected to close the gaps on your own.
Virtual Reality Projekt
The Virtual Reality Projekt is held by Kai Eckert and Marc Reisner. You develop your own concept for a VR application in a group and implement it as a working prototype on the Meta Quest 3 with the Godot Engine — headsets are provided. The course is deliberately open: each round has a new overarching topic from a different application field, and we bring in external expertise for it. What stays constant is the method, not the subject.
The topic of this round is health and sports. Both fields are particularly interesting for VR because they turn the body into the interface: movement becomes input, exertion is real, and the virtual environment decides whether an exercise motivates, bores or overwhelms. Before anything is built, interaction, space and staging are tried out without technology at all, using techniques from theatre — defining and organising a stage, using objects as placeholders, developing characters, improvising sequences. That costs hours instead of weeks and shows early whether an idea carries. Full-day workshops with guests from VR development, from theatre and improvisation, and from the subject field itself round out the term. Prior experience with Godot is recommended but not required; 3D modelling can be useful, and group work and existing assets cover the rest.
KI-unterstütztes Softwaredesign und -engineering
New this term, and taught jointly by Sebastian Steger, Adrian Lehmann, Kai Eckert and Anuja Hariharan: how far does a team get when AI is part of development from the very first idea? The course runs the entire path from concept through prototype to deployment — with locally hosted language models, your own coding harness and a direct comparison with commercial systems. Methods for conception and idea generation with AI support are developed in full-day workshops; the technical side covers the capabilities, limits and inner workings of large language models, the architecture of a coding harness (how an assistant system organises tools, context and test loops, and how to adapt it to your own requirements), and the comparison of execution environments: models running locally, scientific cloud infrastructure and commercial providers.
All of that is applied to one deliberately demanding project task, from finding a solution to cloud deployment, and it is documented, presented and critically reflected — making the team’s own use of AI in the process transparent and assessing it is part of the examination. With 10 CP this is the largest of the three courses, and it is project-based and work-intensive: it requires solid programming skills, ideally in Python, confident use of Git and the command line, and the willingness to work independently and continuously throughout the semester. Programming is not taught here, it is applied.