I am a Ph.D. student in Computer Science at the University of Mannheim. I am excited about understanding the emergent internal mechanisms of neural networks on complex tasks, such as multi-step reasoning, and using these insights to develop better methods. I completed a M.Sc. in Data Science (with distinction) at the University of Mannheim, and B.Sc. in Computer Science.
Here are some representative projects which I have worked on. The complete list of publications can be found on Google Scholar.
A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task
A Multidimensional Analysis of Social Biases in Vision Transformers
A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task [Paper]
Jannik Brinkmann*, Abhay Sheshadri*, Victor Levoso*, Paul Swoboda, Christian Bartelt. arXiv preprint.
Backward Chaining Circuits in a Transformer Trained on a Symbolic Reasoning Task
Jannik Brinkmann*, Abhay Sheshadri*, Victor Levoso*, Paul Swoboda, Christian Bartelt. Workshop on Mathematical and Empirical Understanding of Foundation Models (ME-FoMo), at International Conference on Learning Representations (ICLR).
Unsupervised Extraction of Test Scenarios from Time-Series Sensor Data using Trace Graphs [Paper]
Jannik Brinkmann, Noah Metzger, Christian Bartelt. Hawaii International Conference on Systems Science (HICSS).
Jannik Brinkmann, Paul Swoboda, Christian Bartelt. International Conference on Computer Vision (ICCV).