Fabian Jogl

Fabian Jogl is a PreDoc Researcher at the Vienna University of Technology (TU Wien) within the Faculty of Informatics, holding dual affiliations in the Department of Databases and Artificial Intelligence (Institute E192) and the Department of Machine Learning (Institute E194). His work centers on theoretical and applied aspects of graph neural networks under the StruDL (2023–2027) and VHH (2019–2023) research projects.

His research focuses on the expressivity limits of graph neural networks, particularly investigating whether enhanced expressivity translates to better predictive performance. Key areas include outerplanar graph analysis, path-based GNN architectures, global feature integration, and connections to the Weisfeiler-Lehman hierarchy. He examines structural properties of GNNs through homomorphisms and graph transformations to advance graph representation learning.

Recent publications reveal trends toward unifying theoretical expressivity frameworks with empirical validation across diverse graph datasets. His work bridges geometric deep learning for cell complexes and practical applications like historical film analysis via the Historian dataset. Jogl actively contributes to major conferences including NeurIPS, ICML, and LoG while co-supervising student projects in machine learning algorithms.

As a core member of the StruDL project, he investigates deep learning model structures for graph data, extending prior work on historical film annotation under the VHH project. His technical contributions include novel GNN simulation techniques and methods for enhancing message-passing architectures through graph transformations.