Abstract of the Seminar
Graph neural networks are powerful models for machine learning on
graphs. They form a pillar of geometric deep learning and are subject of active research.
By combining elements from signal processing, partial differential
equations and differential geometry, they provide versatile tools for
data science. In this seminar, we want to give a general introduction to
graph neural networks, with an emphasis on their mathematical
underpinnings. We trace back the origins of the GCN architecture in terms of spectral filter
operations on graphs. Furthermore, we investigate modern developments
that extend and improve this foundational model in terms of ideas from
differential geometry. We discuss both theoretical and practical notions
of graph neural networks.
For more information, see also here