- Research into fundamental methods and techniques with broad application in data science and machine learning. In the open-source project DuckDB we investigate embeddable analytics: database systems that can be embedded in data science processing pipelines. In the safe statistics project, we are completely re-designing common methods in statistics. Current statistical methods are inflexible and this often leads to methodological errors in scientific studies. As a result, incorrect conclusions are drawn and results of scientific research are often not reproducible. Finally, we are creating new machine learning techniques, inspired by the human brain (neuromorphic systems), that consume significantly less energy than present-day AI systems.
- Strategic agents in cyber-systems and cyber-physical systems. In a cyber system (e.g. an online market or a smart grid), the individual participants have preferences and goals that are not necessarily aligned with those of the system as a whole. We model and analyze systems in which strategic agents digitally represent the participants. We want to design integrated methods and solutions that combine AI, data, algorithms and game theory. To include the ethical, legal and societal aspects of this research, we collaborate with other groups and institutes.