Sebastian Mair

about me

I am an assistant professor at the Division of Statistics and Machine Learning at Linköping University, Sweden. In 2026, I obtained my docent qualification (Swedish Habilitation) at Linköping University. Furthermore, I am a member of ELLIS.

Previously, I was a postdoctoral researcher at the Division of Systems and Control at Uppsala University, where I was a member of the research group of Jens Sjölund. I did my PhD at the machine learning group of Ulf Brefeld at the Leuphana University of Lüneburg. Prior to that, I received a Master of Science in computer science as well as a Bachelor of Science in mathematics from Technical University of Darmstadt and a Bachelor of Science in computer science from Hochschule Darmstadt University of Applied Sciences.

I work on data-efficient machine learning: how to train, optimise, and deploy models using less data and less compute without giving up performance or reliability. A recurrent theme in my work is representative subset selection (coresets and related data summarisation ideas), methods that identify the most informative training examples or optimisation variables so that learning and decision-making become faster, cheaper, and more sustainable. I develop algorithms with statistical and geometric grounding, and I evaluate them in methodological settings (e.g., unsupervised/representation learning and probabilistic modelling) as well as in applied pipelines where computational budgets matter. Increasingly, I am interested in how data reduction can serve as a practical lever for sustainable ML: reducing storage, training time, and energy use, while making the modelling process easier to audit and iterate on.

research interests

I am interested in representative subsets and (geometric) data summarisation, unsupervised and representation learning, generative and probabilistic modelling, responsible ML (privacy, fairness, explainability), and the sustainability of machine learning.