About

Mirko received his Ph.D. in 2022 at TU Dortmund University, after finishing his M.Sc. program in Computer Science in 2018 with honors and a specialization in Data Science. His work on machine learning ranges back to 2016, when he joined our research unit as a student assistant. Before, he was working as a software developer for geo-information systems and as a student assistant at Paderborn University, where he completed his B.Sc. in 2014.

Mirko co-organized the workshops Learning to Quantify (2023), Interactive Adaptive Learning (2023), and Machine Learning for Astroparticle Physics and Astronomy (2022).

Topics

Mirko's fundamental research on machine learning is often inspired by applications in astro-particle physics. This application field is characterized by extreme class imbalances, by domain-specific down-stream tasks, and by the fact that training data must be synthesized through simulations that slightly differ from reality. Mirko's current projects concern linear inverse problems for class prevalence estimation (a.k.a. quantification learning), learning under class-conditional label noise, unsupervised domain adaptation, and a smart control of simulations through active class selection.

Social Media

• GitHub
• Google Scholar

Publications

2023

• M. Bunse and L. Pfahler, 2023: Class-conditional label noise in astroparticle physics. In: Europ. Conf. on Mach. Learn. and Knowl. Discov. in Databases.
• M. Bunse, 2023: Qunfold: composable quantification and unfolding methods in Python. In: Int. Worksh. on Learn. to Quantify.
• M. Bunse, A. Moreo, F. Sebastiani, and M. Senz, 2023: Regularization-based methods for ordinal quantification. In: CoRR.
• N. Nikolaou et al., 2023: Lessons learned from the 1st ARIEL machine learning challenge: correcting transiting exoplanet light curves for stellar spots. In: RAS Techn. and Instr.

2022

• M. Bunse, 2022: Machine learning for acquiring knowledge in astro-particle physics. TU Dortmund Univ (Ph.D.).
• M. Bunse, 2022: On multi-class extensions of adjusted classify and count. In: Int. Worksh. on Learn. to Quantify.
• M. Bunse, A. Moreo, F. Sebastiani, and M. Senz, 2022: Ordinal quantification through regularization. In: Europ. Conf. on Mach. Learn. and Knowl. Discov. in Databases.
• M. Bunse, 2022: Unification of algorithms for quantification and unfolding. In: Worksh. on Mach. Learn. for Astropart. Phys. and Astron.
• M. Senz, M. Bunse, and K. Morik, 2022: Certifiable active class selection in multi-class classification. In: Worksh. on Interact. Adapt. Learn.
• M. Senz and M. Bunse, 2022: DortmundAI at LeQua 2022: Regularized SLD. In: Conf. and Labs of the Eval. Forum.

2021

• M. Bunse and K. Morik, 2021: Active class selection with uncertain deployment class proportions. In: Worksh. on Interact. Adapt. Learn.
• M. Bunse and K. Morik, 2021: Certification of model robustness in active class selection. In: Europ. Conf. on Mach. Learn. and Knowl. Discov. in Databases.
• L. Pfahler, M. Bunse, and K. Morik, 2021: Noisy labels for weakly supervised gamma hadron classification. In: CoRR.
• M. Bunse, D. Weichert, A. Kister, and K. Morik, 2020: Optimal probabilistic classification in active class selection. In: Int. Conf. on Data Mining.

2019

• M. Bunse, A. Saadallah, and K. Morik, 2019: Towards active simulation data mining. In: Int. Tutorial and Worksh. on Interact. Adapt. Learn.
• M. Bunse and K. Morik, 2019: What can we expect from active class selection? In: Lernen, Wissen, Daten, Analysen.

2018

• M. Bunse, N. Piatkowski, K. Morik, T. Ruhe, and W. Rhode, 2018: Unification of deconvolution algorithms for Cherenkov astronomy. In: Int. Conf. on Data Sci. and Adv. Analyt.
• M. Bunse, N. Piatkowski, and K. Morik, 2018: Towards a unifying view on deconvolution in Cherenkov astronomy. In: Lernen, Wissen, Daten, Analysen.

• M. Bunse, 2018: DSEA rock-solid – regularization and comparison with other deconvolution algorithms. TU Dortmund Univ. (M.Sc.).

2017

• M. Bunse, C. Bockermann, J. Buss, K. Morik, W. Rhode, and T. Ruhe, 2017: Smart control of monte carlo simulations for astroparticle physics. In: Astron. Data Analys. Softw. and Syst.

Supervised Theses

• N. Gövert, 2023: Fisher-Konsistenz für Quantification-Algorithmen. TU Dortmund Univ. (B.Sc.).
• Z. Ye, 2023: Merkmalstransformationen in Quantification. TU Dortmund Univ. (B.Sc.).
• M. Senz, 2022: Certifiable active class selection in multi-class classification. TU Dortmund Univ. (M.Sc.).
• R.D. Drew, 2021: Deep unsupervised domain adaptation for gamma-hadron separation. TU Dortmund Univ. (B.Sc.).
• M. Schmidt, 2019: Continuous deconvolution of probability density functions in Cherenkov astronomy. TU Dortmund Univ. (M.Sc.).