Research Groups

NEXD investigates how the strengths of classical AI planning can be combined with the flexibility of neural reinforcement learning. The group develops methods that allow neural agents to leverage symbolic structure (e.g. domain models, formal specifications, and search) to learn faster, generalize across problem instances, and remain analyzable. Research themes include neuro-symbolic policy learning, exploration in reinforcement learing, safety, and statistical analysis of neural agents.

CaMoRe studies how causal reasoning can be brought into modern machine learning pipelines and how it can support domain experts in analyzing scientific data and designing experiments. The group works on causal discovery and its evaluation, causal abstraction, root cause analysis, and causally-informed reinforcement learning. We like to collaborate closely with domain scientists to support their research methodologically.

RAIME develops structured, philosophically grounded, application-oriented approaches to Responsible AI. The group studies value-sensitive trade-offs — fairness vs. accuracy, transparency vs. efficiency — and how to navigate them under normative uncertainty. It also works on hybrid approaches to machine ethics based on normative reasons, aiming for moral value alignment of AI agents and ensuring meaningful human oversight in ethically sensitive applications.