Reproducibility Study Of Learning Fair Graph Representations Via Automated Data Augmentations
Abstract: In this study, we undertake a reproducibility analysis of "Learning Fair Graph Representations Via Automated Data Augmentations" by Ling et al. (2022). We assess the validity of the original claims centered around node classification tasks and explore the performance of the Graphair framework in link prediction tasks. Our investigation reveals that while we can partially reproduce some of the original claims—likely impeded by unstable training and a code bug identified through collaboration with the original authors—we fully substantiate another claim. Additionally, we broaden the application of Graphair from node classification to link prediction across various datasets. This expansion demonstrates Graphair’s superior performance in fairness metrics when compared to existing models, showing only a slight reduction in accuracy. This underlines Graphair’s potential applicability in a wider array of graph-based learning contexts, showcasing its capability to maintain high fairness standards without significantly compromising accuracy. Our code base can be found on GitHub https://anonymous.4open.science/r/Reproducibility-Study-Of-Graphair-1DB6.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Guillaume_Rabusseau1
Submission Number: 2234
Loading