Michael Kirchhof
University of Tübingen
Dpt. of Computer Science
Human-Computer Interaction
Sand 14
72076 Tübingen
Germany
- Telefax
- +49 - (0) 70 71 - 29 - 50 62
- michael.kirchhof@uni-tuebingen.de
- Office
- Maria von Linden Strasse 6, 1st floor open space
Research Interests
- Uncertainty
- Representation Learning
I’m a PhD student in the International Max-Planck Research School for Intelligent Systems (IMPRS-IS), co-supervised by Enkelejda Kasneci and Seong Joon Oh at the University of Tübingen.
My goal is making machine learning more trustworthy by delivering pretrained uncertainty estimates along with each prediction. To this end, I’m developing probabilistic embeddings that represent a model’s uncertainty directly in its embedding space. I love to understand and prove things first from a theoretical perspective first (like MCInfoNCE) and then scale them to large datasets as in the new URL benchmark.
I received my BSc (2018) and MSc (2021) in Statistics with distinction at TU Dortmund University where I focussed on probabilistic modeling and machine learning. In 2019, I was a research intern at BMW Group, Munich.
Education
- Ph.D. Candidate in Computer Science, Eberhard Karls University of Tübingen, Germany, 2021 - Present
- M.Sc. in Statistics (with distinction, GPA 4.00/4.00), TU Dortmund University, Germany, 2021
- B.Sc. in Statistics (with distinction, GPA 3.91/4.00), TU Dortmund University, Germany, 2018
Publications
2023
Probabilistic Contrastive Learning Recovers the Correct Aleatoric Uncertainty of Ambiguous Inputs
Michael Kirchhof, Enkelejda Kasneci, and Seong Joon Oh. International Conference on Machine Learning (ICML), 2023.
When are post-hoc conceptual explanations identifiable?
Tobias Leemann, Michael Kirchhof, Yao Rong, Enkelejda Kasneci, and Gjergji Kasneci. Uncertainty in Artificial Intelligence (UAI), 2023.
URL: A Representation Learning Benchmark for Transferable Uncertainty Estimates
Michael Kirchhof, Bálint Mucsányi, Seong Joon Oh, and Enkelejda Kasneci. NeurIPS Datasets and Benchmarks, 2023.
2022
A Non-isotropic Probabilistic Take on Proxy-based Deep Metric Learning
Michael Kirchhof, Karsten Roth, Zeynep Akata, and Enkelejda Kasneci. European Conference on Computer Vision (ECCV), 2022.