TU 6: Machine learning

Machine learning (ML) has developed into a major new approach to scientific and technological questions. ML is located at the intersection of computer science and computational statistics and is both a technology/tool and a conceptual approach. Its success as a technology has been borne out by its now key - and often transformative - role in a rapidly broadening range of industries, sectors and research fields. Conceptually, ML places an emphasis on adapting or building models using data and on performance with respect to specific tasks. The optimality of models is defined not in the abstract, but in a goal-driven manner and with the impact of imperfect data on complex models taken into account.

To help realize the high potential impact of ML across CASCADE research units, and building on a strong base of ML research in Bonn, the Topical Unit "Machine Learning" is developing a set of activities and structures to enable CASCADE researchers to further increase their exposure to ML methods and concepts and to lower the barriers to engagement with cutting-edge ML research.

Bonn is already recognized as key location for ML in Germany, with prominent researchers at several CASCADE institutions and a number of nationally and internationally-leading initiatives. For example, Bonn is the location for the “Fraunhofer Research Center Machine Learning” (the main site for ML activity across the Fraunhofer Society) and has recently been awarded a new “Competence Center Machine Learning Rhein-Ruhr” (with Dortmund) that is one of only four such centers nationwide. ML and data science are key themes at several CASCADE institutions, including IAIS, SCAI and DZNE, and FZJ/JSC offers unique expertise and hardware for scalable ML.

The Topical Unit Machine Learning will operate via several, inter-related initiatives and structures. These include the Bonn Machine Learning Lecture Series (BMLL), which is planned as a flagship Bonn seminar series featuring world-class speakers, spanning core theory/methods, computing and applications. A CASCADE Machine Learning Meetup (CMLM), co-located with the BMLL event, will take place 3 times each year. This will provide a forum for CASCADE researchers at all levels to discuss, learn and interact in a relaxed and open setting. Junior CASCADE researchers will play a key role in organizing these events.

Specific ML activities will be offered at all career levels. To take a few examples: ML will form part of the CSCS curriculum and ML PhD projects will be integrated with other CASCADE fields via a “twinning” model that is being developed in a data science graduate school concept (HDAN; this is joint between the DZNE/UBonn, involving several CASCADE PIs). A yearly Hackathon, mainly organized by CASCADE junior scientists, will work on CASCADE data and questions, with results written up and submitted in a joint manner under the name “CASCADE Learning Collective”.