PhD Opportunity in Robust & Certifiable AI
I am looking for PhD candidates interested in the foundations of machine learning evaluation under distribution shift, with direct links to validation and certification procedures in safety-critical applications (e.g., in collaboration with TÜV Austria). The focus is on mathematically grounded methods (learning theory, estimation, guarantees, uncertainty quantification), rather than purely empirical modeling.
Research directions include:
quantification of distribution shift (e.g., density ratios, classifier-based approaches)
target risk estimation under shift (e.g., importance weighting, aggregation)
validation protocols and decision criteria for reliable deployment
You are a good fit if you are interested in developing work at the level of the following papers, which define the expected level and style of a PhD in this direction:
M.-C. Dinu et al., “Addressing parameter choice issues in unsupervised domain adaptation by aggregation,” ICLR (oral), 2023
P. Setinek et al., “SIMSHIFT: A benchmark for adapting neural surrogates to distribution shifts,” preprint, 2025
W. Zellinger, “Binary losses for density ratio estimation,” ICLR, 2025
W. Zellinger et al., "Adaptive learning of density ratios in RKHS," JMLR, 2023.
K. Schweighofer et al., “Safe and certifiable AI systems: Concepts, challenges, and lessons learned,” TÜV AUSTRIA Report, 2025
Positions are typically embedded in larger institute activities and third-party projects. If this aligns with your interests, please reach out with a short note explaining how your background connects to these topics.