Complex systems are ubiquitous, whether in environmental processes, human behavior or biomedical applications. Understanding the underlying processes of such systems better and being able to make adequate predictions is a major challenge. Machine learning methods and in particular deep learning models deal with this mainly by relying on huge amounts of data. This often leads to models that are difficult to understand and from which it is not easily possible to gain deeper insights into the corresponding domain. At the same time, background knowledge and a deeper understanding of complex systems are essential for practical decision-making and planning. This highlights the central role of background knowledge, whose integration with machine learning has the potential to enable explanatory as well as predictive models with significantly increased predictive power, robustness, generalizability and fairness.
Goals and approach
Therefore, the AI junior research group Themis deals with knowledge extraction and knowledge integration for machine learning. The project aims to develop knowledge extraction methods to explain and understand complex systems along three axes: horizontal (interactions between components of the system), vertical (differences in the collected data, such as for different groups of people), and temporal (over time). Furthermore, the limits and the potential of knowledge integration in practical applications are examined and optimized. The methods developed are to be tested in case studies and projects with users.
Innovation und perspectives
The methods developed will thus form the basis for gaining new insights in a wide variety of application scenarios and optimally using existing knowledge for comprehensible machine learning. In the medium term, this should enable the practical application of knowledge-based machine learning in a variety of industries such as environmental modelling, behavioral analysis or biomedical development.