← All researchPublication · Topological deep learning

A Remedy for Over-Squashing in Graph Learning via Forman-Ricci Curvature based Graph-to-Hypergraph Structural Lifting

Michael BanfMichael Banf et al.
August 2025 · arXiv:2508.11390, preprint

The full write-up of the lifting method, with results on four benchmark datasets and five graph and hypergraph models.

The research

Graphs squeeze information through bottlenecks. We use curvature to find a network's backbones and lift them to hyperedges, so distant parts of a graph can talk.

What it changes in practice

A company's knowledge graph has bottlenecks too. Curvature tells us where answers get squashed before they reach the person who needs them.

More in Topological deep learning

Contact

For projects, partnerships, press and open roles.

Goethestr. 68, 80336 Munich · +49 89 12137977 · contact@perelyn.com
Contact

Request a conversation. We’ll be in touch.

For projects, partnerships, press and open roles.

Topic
* Required
You are in good company:
Deutsche BahnE.ONHenkelLiebherrMarc O'PoloMunich ReREWEZEISSRocheOn