Which Algorithms Can Graph Neural Networks Learn?

Name: Which Algorithms Can Graph Neural Networks Learn?
Uploaded: 2026-09-04T08:30:39.979860+00:00
Duration: 289 s
Description: The hosts discuss a paper titled "Which Algorithms Can Graph Neural Networks Learn?" which provides a theoretical framework for GNN capabilities. They explore specific algorithms that can be learned, such as shortest paths, and detail the limitations of standard MPNN architectures. The discussion also covers technical improvements like using differentiable regularization to reduce training data.

The hosts discuss a paper titled "Which Algorithms Can Graph Neural Networks Learn?" which provides a theoretical framework for GNN capabilities. They explore specific algorithms that can be learned, such as shortest paths, and detail the limitations of standard MPNN architectures. The discussion also covers technical improvements like using differentiable regularization to reduce training data.

Open the full episode