[#25] Machine learning for graphs: Hot trends and emerging frontiers

She delved into the history of GNNs, explaining their functioning based on the message-passing algorithm and their expressive power. She also discussed the current limitations of GNNs (Oversmoothing and Oversquashing), their applications and potential future improvements. You can find her slides here.

Veronica made it clear that Graph Neural Networks (GNNs) are powerful tools for extracting information from complex structured data, but are they the holy grail, or is a more complex system approach still needed?

Machine learning for graphs still faces the explainability problem, but complex systems analysis hardly depends on interpretation. Both disciplines are necessary, and their applicability can be contingent on the specific task.