Investigating GNN Convergence on Large Randomly Generated Graphs with Realistic Node Feature Correlations

Mohammed Zain Ali Ahmed · 2026

Read Investigating GNN Convergence on Large Randomly Generated Graphs with Realistic Node Feature Correlations by Mohammed Zain Ali Ahmed (2026) 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-2602-16145.txt.

Description

Mohammed Zain Ali Ahmed studies how graph neural networks behave on large random graphs with realistic node features. Prior studies of convergence on large random graphs mostly ignore correlations between node features, yet such correlations arise naturally in real networks. That gap matters: limits derived without them may not reflect how expressive these networks truly are on realistic graphs. The paper introduces a new method for generating random graphs whose node features correlate between neighbors.

The sampling scheme draws motivation from traits of real graphs, notably traits captured by the Barabasi-Albert model. Theoretical analysis strongly indicates that convergence can be avoided in some cases, and experiments on large random graphs built with the new method validate that claim. The observed divergent behavior suggests graph neural networks may prove more expressive than early studies implied, especially where node features correlate as they do in the wild.

Catalog record

CREATORMohammed Zain Ali Ahmed
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-16145

Browse the full library · About the Tower of Babel