Problem
Research questions and friction points this paper is trying to address.
Fair imputation of missing protected attributes
Addressing bias propagation in GNNs
Optimizing fairness-accuracy trade-off
Better Fair than Sorry: Adversarial Missing Data Imputation for Fair GNNs
🤖 AI Summary
Existing fair graph neural network (GNN) studies typically assume either fully observable protected attributes or unbiased imputation of missing values; however, in practice, biased imputation silently amplifies discrimination, leading to severe overestimation of model fairness. This work proposes the first fairness-aware adversarial imputation framework for graphs. We formulate a three-player minimax game among an imputer, a discriminator, and a fair classifier, optimizing for *worst-case fairness*—specifically, minimizing the maximum group-wise fairness deviation (e.g., Equalized Odds difference) under adversarial imputation. Our approach jointly learns robust imputations and fair representations, ensuring fairness guarantees even under distributional shifts induced by biased missing-data mechanisms. Experiments on synthetic and real-world graph benchmarks demonstrate that our method significantly outperforms state-of-the-art fair GNNs and imputation baselines: it maintains competitive classification accuracy while substantially improving fairness—up to a 12.7% gain in Equalized Odds—effectively mitigating bias propagation.
📝 Abstract
Graph Neural Networks (GNNs) have achieved state-of-the-art results in many relevant tasks where decisions might disproportionately impact specific communities. However, existing work on fair GNNs often assumes that either protected attributes are fully observed or that the missing protected attribute imputation is fair. In practice, biases in the imputation will propagate to the model outcomes, leading them to overestimate the fairness of their predictions. We address this challenge by proposing Better Fair than Sorry (BFtS), a fair missing data imputation model for protected attributes. The key design principle behind BFtS is that imputations should approximate the worst-case scenario for fairness -- i.e. when optimizing fairness is the hardest. We implement this idea using a 3-player adversarial scheme where two adversaries collaborate against a GNN-based classifier, and the classifier minimizes the maximum bias. Experiments using synthetic and real datasets show that BFtS often achieves a better fairness x accuracy trade-off than existing alternatives.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Adversarial missing data imputation
Worst-case fairness approximation
3-player adversarial scheme
🔎 Similar Papers
No similar papers found.
