

Unveiling Graph Datasets
May 8, 2025
Bastian Rieke, a tenured professor of machine learning at the University of Fribourg and leader of the Eidos Lab, dives deep into the world of graph datasets. He discusses the RINGS framework for evaluating dataset robustness and the significance of community dynamics in network analysis. Rieke highlights how topology can enhance machine learning performance by revealing data structure insights. He also addresses the ongoing challenges in graph learning and the necessity for better real-world datasets to foster innovation in research.
AI Snips
Chapters
Transcript
RINGS Framework Reveals Dataset Quality
- The RINGS framework assesses graph learning datasets by perturbing graphs to evaluate their robustness.
- If perturbations don't reduce performance, the network structure might be underutilized.
Ask the Right Questions of Networks
- The questions asked of a network dataset must align with what the network can reveal.
- Misguided homophily assumptions can distort results and mislead network analysis.
Structured Journaling on GitHub
- Bastian Rieke keeps a structured journal on GitHub in markdown files since 2018.
- He uses this to draft blog posts and papers before formalizing them.