Expressive Power of Graph Neural Networks

In the last decades, database theoreticians have developed tools to analyse the expressive power of query languages. The query languages considered typically target relational, graph or other discrete and structured data. In graph machine learning, graph neural networks are currently the technique of choice to perform vertex, node or graph classification and regression tasks. In this talk we connect database theory, and query languages in particular, to graph neural networks, and hereby gain insights in the expressive power of these networks.

You can see the slides of the presentation here

Speaker

Floris Geerts is professor at the Department of Computer Science. His research concerns database theory, data quality and graph learning.

Time and Place

Friday 4/11/2022 at 11:30am in M.A.143

Registration

Participation is free, but registration is compulsory. Make sure to fill in this form.

References and Related Reading

• What are query languages and their expressive power
  • Principles of Databases (Book), Arenas, Barceló, Libkin, Martens, Pieris.
  • Foundations of Database Theory (Book), Abiteboul, Hull, Vianu.
• Graph neural networks
  • Graph Representation Learning (Book), Hamilton
  • Weisfeiler-Leman: The Story So Far (Survey), Morris, Lipman, Maron, Rieck, Kriege, Grohe, Fey, Borgwardt.
  • Expressive power of graph neural networks and the Weisfeiler-Lehman test (blog), Bronstein.