Matthias Welsch

Matthias Welsch is a PhD student in the COMP3D group. He is part of the CD laboratory for Molecular Informatics in the Biosciences and holds a MSc. in Computational Science from the University of Vienna.

His PhD studies aim to better understand machine learning models in cheminformatics using representational alignment (techniques that measure how information is internally represented across models) and how this relates to predictions. These methods capture when models make the same predictions for different reasons, linking internal representations to outcomes.

His research focuses on cheminformatics-specific aspects of dataset behavior, descriptor strategies and representation learning, physics-inspired regularization, and common model types. By analyzing these factors, he seeks to reveal how models process chemical data and why they arrive at their predictions.