Fabian Gröger

I am a final-year PhD student at the University of Basel, supervised by Alexander Navarini and Marc Pouly. I also spent time as a visiting researcher at EPFL with Maria Brbić.

Most of what we want to do with neural networks (align modalities, transfer to new domains, find what is wrong with a dataset) runs through their internal representations. That is what I am most interested in.

My recent work on the Aristotelian Representation Hypothesis (ICML ‘26) pushes back on a popular story in the field, that neural networks are converging to a shared “Platonic” representation. We show that much of the reported convergence is a measurement artifact and that what we observe agreeing across models is local neighborhood structure rather than global geometry. The claim is smaller but sharper, and it changes how we should compare and evaluate representations.

I also work on how to use representations in practice: aligning them across modalities when paired data is limited (NeurIPS ‘25), exploiting them for data quality audits in image (NeurIPS ‘24) and audio (ICASSP ‘26) collections, and applying them as a global atlas to audit the dermatology field at scale, plus other audio for hearables work (ICASSP ‘25).

Outside of research, I spend my time with my family and on the bike.