Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reproducibility Study Of Learning Fair Graph Representations Via Automated Data Augmentations

Thijmen Nijdam, Juell Sprott, Taiki Papandreou-Lazos, Jurgen de Heus

TL;DR

It is indicated that, while Graphair demonstrates a comparable fairness-accuracy trade-off to baseline models for mixed dyadic-level fairness, it has a superior trade-off for subgroup dyadic-level fairness.

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 focused on node classification tasks and explore the performance of the Graphair framework in link prediction tasks. Our investigation reveals that we can partially reproduce one of the original three claims and fully substantiate the other two. Additionally, we broaden the application of Graphair from node classification to link prediction across various datasets. Our findings indicate that, while Graphair demonstrates a comparable fairness-accuracy trade-off to baseline models for mixed dyadic-level fairness, it has a superior trade-off for subgroup dyadic-level fairness. These findings underscore Graphair's potential for wider adoption in graph-based learning. Our code base can be found on GitHub at https://github.com/juellsprott/graphair-reproducibility.

Reproducibility Study Of Learning Fair Graph Representations Via Automated Data Augmentations

TL;DR

It is indicated that, while Graphair demonstrates a comparable fairness-accuracy trade-off to baseline models for mixed dyadic-level fairness, it has a superior trade-off for subgroup dyadic-level fairness.

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 focused on node classification tasks and explore the performance of the Graphair framework in link prediction tasks. Our investigation reveals that we can partially reproduce one of the original three claims and fully substantiate the other two. Additionally, we broaden the application of Graphair from node classification to link prediction across various datasets. Our findings indicate that, while Graphair demonstrates a comparable fairness-accuracy trade-off to baseline models for mixed dyadic-level fairness, it has a superior trade-off for subgroup dyadic-level fairness. These findings underscore Graphair's potential for wider adoption in graph-based learning. Our code base can be found on GitHub at https://github.com/juellsprott/graphair-reproducibility.
Paper Structure (27 sections, 9 equations, 6 figures, 9 tables)

This paper contains 27 sections, 9 equations, 6 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Scope of reproducibility
  3. Methodology
  4. Graphair
  5. Adversarial Training for Fairness
  6. Contrastive Learning for Informativeness
  7. Reconstruction Based Regularization to Ensure Graph Consistency
  8. Integrated Training Objective
  9. Datasets
  10. Experimental Setup
  11. Link Prediction
  12. Baselines
  13. Hyperparameters
  14. Node Classification
  15. Link Prediction
  16. ...and 12 more sections

Figures (6)

  • Figure 1: ACC and DP trade-off for baselines, Graphair and our results for Graphair. Upper-left corner (high accuracy, low demographic parity) is preferable.
  • Figure 2: Node sensitive homophily distributions in the original and the fair graph data.
  • Figure 3: Spearman correlation between node features and the sensitive attribute in the original and the fair graph data.
  • Figure 4: ACC and DP trade-off for the baselines and our Graphair for link prediction. The top row shows the $\Delta \text{DP}_m$ metric, and the bottom row shows the $\Delta \text{DP}_s$ metric. Points in the upper-left corner are desired.
  • Figure 5: Overview of the Graphair framework ling2022learning
  • ...and 1 more figures