Enhancing Distribution and Label Consistency for Graph Out-of-Distribution Generalization

Song Wang 2 , Xiaodong Yang 3 , Rashidul Islam 3 , Huiyuan Chen 3 , Minghua Xu 3 , Jundong Li 2 , Yiwei Cai 3 · 2025

Read Enhancing Distribution and Label Consistency for Graph Out-of-Distribution Generalization by Song Wang 2 , Xiaodong Yang 3 , Rashidul Islam 3 , Huiyuan Chen 3 , Minghua Xu 3 , Jundong Li 2 , Yiwei Cai 3 (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-2501-04102.txt.

Description

Song Wang and six co-authors tackled graph learning under distribution shift in twenty twenty-five. Graph neural networks falter when test graphs differ from training ones. Popular defenses augment the training graphs to create new environments, then extract the subgraph that stays predictive across them. The authors argue both steps can break consistency: augmented graphs may be unrealistic, and hacked-up subgraphs may lose their honest link to the label.

Their framework uses a single modifier to produce both augmented and invariant graphs in a unified way. Augmentation enriches training while preserving label-graph relationships, and a label-consistency mechanism protects supervision inside the invariant graph. Extensive experiments on real-world data sets showed the approach beating state-of-the-art baselines for out-of-distribution graph generalization.

Catalog record

CREATORSong Wang 2 , Xiaodong Yang 3 , Rashidul Islam 3 , Huiyuan Chen 3 , Minghua Xu 3 , Jundong Li 2 , Yiwei Cai 3
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-04102

Browse the full library · About the Tower of Babel