Research profile dr. K. (Kerstin) Bunte

Description of research:

In July 2016 I started my tenure track Rosalind Franklin Fellowship in Computer Science at the University of Groningen. My research focuses on interdisciplinary projects and development of interpretable machine learning techniques for integrative data analysis.

To date most successful machine learning techniques for the analysis of complex interdisciplinary data predominantly use significant amounts of vectorial measurements as input to a statistical system. The domain expert knowledge is often only used in data preprocessing and the subsequently trained technique appears as a black-box, which is difficult to interpret or judge and rarely allows insight into the underlying natural process. However, in many applications, like for example in the bio-medical domain, the underlying biological process is complex and the amount of measurements is limited due to the costs and inconvenience for the patient.

In my individual Marie Curie Fellowship under the title ‘Learning in the Space of Dynamic Models for Adrenal Steroidogenesis (LeSoDyMAS)’ we started to tackle the above mentioned problems by combining the forces of Computer Science, Medicine, Engineering, Applied Mathematics and pharmaceutical industry to work on high impact bio-medical problems. We are aiming at the design of machine learning methods which allow the principled integration of expert knowledge for example in form of dynamic models.

Furthermore, I am part of the Innovative Training Network SUNDIAL, which combines training in computer science and astrophysics. The young scientists in computer science will study topics such as detecting ultrafaint galaxy signals, developing automated models for galaxy recognition and classification, and developing new methods to compare observations and galaxy simulations as well as visualization. My project will focus on galaxy simulations and visualization. With galaxy simulations continuously becoming larger and more detailed, comparison with observation data is far from trivial. Failing to reproduce one ore more observable properties may be indicative that an important astrophysical process is not taken into account in the models. The ability to compare simulations with observations in a well-defined and well-understood quantitative way is a crucial step in calibrating simulations with observed data, thus allowing for a much deeper understanding of galaxy formation.

Keywords: machine learning, interpretable models, interdisciplinary data analysis, dimensionality reduction, classification, clustering, visualization

Websites:
Employee (corporate) webpage: http://www.rug.nl/staff/k.bunte/
Personal webpage: http://www.cs.rug.nl/~kbunte/
Rosaling Franklin profile webpage: http://www.rug.nl/about-us/work-with-us/rff/rosalind-franklin/rff-portraits/rff-2015-2016/kerstin-bunte