Limits of message passing for node classification: How class-bottlenecks restrict signal-to-noise ratio
Jonathan Rubin, Sahil Loomba, Nick S. Jones · 2025
Read Limits of message passing for node classification: How class-bottlenecks restrict signal-to-noise ratio by Jonathan Rubin, Sahil Loomba, Nick S. Jones (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-17822.txt.
Description
Jonathan Rubin, Sahil Loomba, and Nick Jones set out limits of message passing for node classification on graphs. Their 2025 paper deals with two known weaknesses: heterophily, where neighbors often differ in class, and structural bottlenecks.
Their statistical framework ties the two through the signal-to-noise ratio of the learned representations. That ratio splits performance into feature-dependent parts and feature-independent sensitivities. They prove the sensitivity to class signals is bounded by a multi-hop generalization of homophily. Where that quantity is low, bottlenecks and labels interact locally to choke the signal. Analysis of graph ensembles quantifies the effect further.
Catalog record
| CREATOR | Jonathan Rubin, Sahil Loomba, Nick S. Jones |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 207 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17822