Understanding Generalization in Node and Link Prediction

Antonis Vasileiou 1 Timo Stoll 1 Christopher Morris 1 1 Department of Computer Science, RWTH Aachen University, Germany [email protected] · 2025

Read Understanding Generalization in Node and Link Prediction by Antonis Vasileiou 1 Timo Stoll 1 Christopher Morris 1 1 Department of Computer Science, RWTH Aachen University, Germany [email protected] (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-2507-00927.txt.

Description

Graph neural networks pass messages between linked points to predict labels and missing links, but how well they generalize is poorly mapped. Antonis Vasileiou, Timo Stoll, and Christopher Morris build one framework to answer that. Their analysis covers learning from labeled graphs and filling in partly known ones. It folds in design choices, loss types, and the graph's own shape. Older studies assumed neat, unrelated data, which real networks rarely give.

Experiments back the theory's main claims. The framework also applies to plain classification tasks beyond graphs. Knowing when message passing will carry to new nodes helps builders trust it in fraud, biology, and social data. The work replaces rule-of-thumb confidence with measured bounds.

Catalog record

CREATORAntonis Vasileiou 1 Timo Stoll 1 Christopher Morris 1 1 Department of Computer Science, RWTH Aachen University, Germany [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE148 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-00927

Browse the full library · About the Tower of Babel