Talk by Max Müller-Eberstein

16 May 2025, by DS Group

We are more than happy to have Max Müller-Eberstein as a guest in our lab. He will be presenting several exciting projects.

Title: How Language Model Learning Dynamics Shape Social Inclusion

Abstract: It's an exciting time for studying the Machine Learning theory underlying Language Model training, as well as the social implications of these models' real-world use. The connections between these areas, however, often remain unexplored. In this talk, we'll journey through the full LM training pipeline to show how fundamental learning dynamics influence downstream capabilities related to social inclusion, such as gender bias mitigation and cultural adaptation. By examining consistent learning patterns across different training runs, we'll show how these insights can help address scenarios where collecting adequate training data is physically impossible—enabling new ML applications for underserved communities.

Who: Max is a postdoc at the IT University of Copenhagen's NLPnorth Lab and the Pioneer Centre for AI, studying consistencies across the learning dynamics of Machine Learning models. Towards the goal of enabling AI training in extremely low-resource scenarios, he has applied these findings to Language Model adaptation (e.g., to Danish culture), and to improved accessibility (e.g., atypical speech recognition, mapping visual art to music). Previously, he completed a PhD on "Quantifying Linguistic Variation", within the ELLIS network, conducting joint research at ITU Copenhagen, LMU Munich, and the University of Edinburgh.