XAI for Science Lab | Ana Lucic

University of Amsterdam

AL_headshot_1MB.jpg


I am an assistant professor in machine learning at the University of Amsterdam, where I lead the Explainable AI for Science lab. I have a joint position between the Institute for Logic, Language, and Computation, and the Informatics Institute. My group is affiliated with the Amsterdam AI for Science research center, the SIAS group, and the NLP Unit. I am an ELLIS member and part of the corresponding Amsterdam ELLIS Unit.

Previously, I was a researcher at Microsoft Research AI for Science, where I worked on foundation models for Earth system modeling and at the Partnership on AI where I worked on explainable ML for healthcare. My PhD in explainable ML is from the University of Amsterdam and my MSc and BSc are both in mathematics from McMaster University in Canada.

My current research interests are primarily at the intersection of AI for science and ML interpretability. My group works methodological development in machine learning with applications in weather and climate modeling, molecular simulation, and fluid dynamics. I am also broadly interested in AI safety and am a mentor at Safe AI Netherlands. If you’re a student (e.g., UvA MSc, ELLIS PhD) looking for a supervisor related to these topics, please reach out via email explaining your research interests and include your CV and complete transcript of grades.

News

Jul 2026 I have been awarded an NWO Veni grant to work on mechanistic interpretability for graph ML models!
Jun 2026 We have one paper accepted to the Mechanistic Interpretability workshop at ICML 2026 on investigating what routers learn in Mixture-of-Depths models.
Feb 2026 We have one paper on investigating atmospheric structure in Aurora accepted to the Scientific Methods for Understanding Deep Learning workshop at ICLR 2026!
Nov 2025 We have two papers accepted to the Mechanistic Interpretability workshop at NeurIPS 2025! One is about equivariant sparse autoencoders, the other is about causal abstraction as a framework for faithfulness.
May 2025 Ege Erdogan has joined us as a new PhD student working on mechanistic interpretability. Welcome Ege!