Marius Winter

• Semantic segmentation of human muscles in MR images
• Extraction of relevant muscle features (biomarkers)
• Investigation of the association of biomarkers with muscular changes and diseases

If you are interested in working on a thesis (BA/FA/MA) focused on topics such as feature selection/extraction, uncertainty estimation and explainability in machine learning, or related areas, we can arrange a meeting to discuss possibilities.

Update July, 4th 2025:

We are offering a thesis on the following topic:

In cooperative game theory, one task is to fairly distribute the credit (or blame) for a coaltion's performance among its players.
Among game theorists, it is widely accepted that a fair distribution has to fulfill four mathematical axioms: 
(1) efficiency, (2) symmetry, (3) additivity, (4) null player property

It has been mathematically proven that the "Shapley values" are the only solution that satisfies all four axioms.

For machine learning, a Shapley value of an input feature can be interpreted as its feature importance, i.e., we fairly distribute the credit (or blame) of a model's prediction across all input features.

The ISS (Liao et al.) developed a deep learning-based feature selection approach that effectively captures the input features' importance scores in the form of a feature mask m.

In this thesis, we aim to investigate to what extent this feature mask method does or does not satisfy the mathematical axioms proposed by game theorists.