Reproducibility study of FairAC

TL;DR

This work conducts a thorough reproducibility study of FairAC, a fair attribute completion framework for graphs, and extends evaluation to assess generalizability and individual fairness. It confirms that FairAC can achieve strong group-fairness improvements with competitive accuracy and demonstrates that the method generalizes across several datasets and sensitive attributes, albeit with attribute-dependent trade-offs. The authors refactor the FairAC codebase into an ergonomic library, integrate additional baselines, and introduce an individual-fairness metric (Consistency) to broaden the fairness assessment. Overall, the study supports the original claims while highlighting practical considerations in reproducibility, generalizability, and the nuanced balance between group and individual fairness.

Abstract

This work aims to reproduce the findings of the paper "Fair Attribute Completion on Graph with Missing Attributes" written by Guo, Chu, and Li arXiv:2302.12977 by investigating the claims made in the paper. This paper suggests that the results of the original paper are reproducible and thus, the claims hold. However, the claim that FairAC is a generic framework for many downstream tasks is very broad and could therefore only be partially tested. Moreover, we show that FairAC is generalizable to various datasets and sensitive attributes and show evidence that the improvement in group fairness of the FairAC framework does not come at the expense of individual fairness. Lastly, the codebase of FairAC has been refactored and is now easily applicable for various datasets and models.

Figures (2)

• Figure 1: The FairAC framework consists of mainly three components: An auto-encoder to generate embeddings, an attention-based mechanism for attribute completion and a sensitive classifier to apply adversarial learning fairac.
• Figure 2: Hyperparameter value study for $\beta$, which influences the trade-off between accuracy and fairness.