We develop and apply AI methods informed by domain knowledge to enable accurate simulations, parameter inference, and enhanced interpretability for applications in science and industry.
Object-centric learning and physics-informed generative models for semiconductor metrology. Collaboration with ASML and UvA Amsterdam.
Physics-informed generative AI for quantum hardware calibration, electron shuttling optimization, and digital twins of quantum devices.
AI-driven non-destructive biomass estimation using CNNs and sensor data. Multimodal models for plant phenotyping and climate adaptation research. Collaboration with University of Cologne.
Reproducible ML pipelines for scientific computing with data versioning, experiment tracking, and automated deployment. Domain knowledge embedding via LLMs.
GenAI for quantum hardware optimization. Physics-informed digital twins for electron shuttling. Partners: ML4Q (RWTH, Uni Koeln, FZ Juelich). 2027–2030. Funded by BMFTR.
Robot vision and AI cyber security for resource-efficient production systems. Deep learning camera systems for automotive manufacturing. Funded by EFRE, MWIKE and the European Commission. Details.
Orthogonal rotating features for object-centric learning in semiconductor metrology. Collaboration with UvA Amsterdam and ASML. Details.
We are looking for talented students and researchers passionate about applying AI to scientific challenges. Check out our thesis topics and open positions.
March 2026 - THK AI Journal Club established — a regular meeting for researchers and students to discuss recent papers in AI and machine learning. More information →
February 2026 - New lectures on MLOps and Distributed Systems (BVS2) now offered. More information →
January 2026 - Paper “OrthoRF: Exploring Orthogonality in Object-Centric Representations” accepted at ICLR 2026. Read more →
January 2026 - New preprint: “Data-Driven Qubit Characterization and Optimal Control using Deep Learning” on arXiv. Read paper →
Photo by Luke Jones on Unsplash