My research asks how learning systems keep performing when the world drifts away from their training data. With my group at Aarhus University, we develop test-time adaptation methods that update models on unlabelled streams, self-supervised representations that transfer across domains and tasks, and uncertainty estimates that tell a model when it should not be trusted. Much of this work is tested in medical imaging, where silent errors are costly.
A growing part of my research takes these ideas into the physical world. In the STORM project, funded by the Independent Research Fund Denmark and starting in 2027, we will study how robots can use touch to notice unexpected physical changes, remember contact, and correct or stop when uncertainty is too high. With vision–language models that reason over images and text, this points towards agents that observe, adapt and act safely in changing real-world conditions.
Before joining Aarhus University in 2026, I was Head of the Computer Vision Team and Lecturer at École Polytechnique Fédérale de Lausanne (EPFL) from 2017 to 2025, with a joint position as Senior Researcher in the Department of Radiology at Lausanne University Hospital (CHUV) from 2019 to 2025, and earlier a Postdoctoral Research Fellow at IBM Research Australia. I received my PhD from the University of Canberra in 2016. I am also a Guest Professor at EPFL.
I am an elected Member of the European Laboratory for Learning and Intelligent Systems (ELLIS) since 2021 and, since 2026, an ELLIS Scholar and Advisor in the ELLIS PhD & Postdoc Programme, affiliated with the ELLIS Unit Lausanne at EPFL, where I support joint and co-supervised PhD positions across the ELLIS network.
