CGRL: Causal-Guided Representation Learning for Graph Out-of-Distribution Generalization

Bowen Lu, Lianqiang Yang, Teng Li · 2026

Read CGRL: Causal-Guided Representation Learning for Graph Out-of-Distribution Generalization by Bowen Lu, Lianqiang Yang, Teng Li (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-2603-24304.txt.

Description

Lu and co-authors present CGRL, a causal method that helps graph networks generalize to new distributions. Graph networks often learn noise and spurious links instead of stable causal patterns. The authors also name a training instability, Info-Jitter, where mutual information with the labels swings during training. Their method builds a causal graph for node classification and blocks non-causal paths with do-calculus. A re-weighted module boosts causal signals during message passing. Tests across distribution shifts beat strong baselines and calm the Info-Jitter effect.

Catalog record

CREATORBowen Lu, Lianqiang Yang, Teng Li
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75460
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.24304

Browse the full library · About the Tower of Babel