On this page, you can find information related to our regularly taught classes, as well as student projects and theses.


The below table shows an overview of classes we generally teach each year.





Introduction to Deep Learning LSF

! From 2025 during summer terms

Learning Generative Models LSF

! From 2025 during winter terms


Music Information Retrieval LSF

Neuronale Netze LSF

! Wird ab 2025 nicht mehr angeboten


 Deep Learning für Ingenieure

ab WS2024/2025

KI Prototyping LSF


Erklärbare und Sichere KI

ab SS2025

For reference, we also have a list of all classes we have taught in the past.  

Work with Us

Process for Projects and BA/MA Theses

  1. Familiarize yourself with the topic you like to work on
  2. Apply via our Application Form
  3. Upon acceptance, start working on the topic
  4. Give a kickoff presentation
  5. Write a project report/thesis
  6. Give a final presentation/defense

More details: Theses and Projects

Research Topics

You can find a list of possible topics in our OVGU cloud.

Some topics with one-sentence summaries can also be found below (to be updated):

  • XAI-Based Neural Network Pruning  [PDF]
    We will investigate, how explainable AI can guide in making AI models smaller.
  • Explainable AI for Transformers  [PDF]
    We will review techniques for explaining Transformer models and potentially develop novel explainable AI methods.
  • Spatial Canonicalisation  [PDF]
    Even marginally transformed input signals can break downstream performance of many deep neural nets. You can test this yourself by training a classifier on Fashion-MNIST and test it on rotated images. This spatial brittleness can have severe impacts in the real-world. Think about the imperfect landscape of street signs and a vision model to recognise them. Canonicalisation aims to rectify the inputs of the downstream model by removing spatial distortions during test time. 
  • Traffic Light Control  [PDF]
    Optimising traffic lights in cities using deep learning is a new but highly promising avenue. We use traffic simulation to train Deep Reinforcement Learning agents and Graph Neural Networks on traffic densities to predict the next traffic light states (green or red) to optimise traffic flow and minimise traffic jams.

Last Modification: 20.03.2024 - Contact Person: Webmaster