Matched sample selection with GANs for mitigating attribute confounding

TL;DR

This paper introduces a GAN-based matching method to create balanced datasets for bias analysis in vision systems, helping to distinguish true algorithmic bias from dataset bias.

Contribution

It proposes a novel GAN-based matching approach to mitigate attribute confounding in bias measurement, enabling more accurate bias analysis in vision datasets.

Findings

Bias persists even after matching for confounders.

The method effectively balances attributes across protected groups.

Code and tools are publicly available for replication.

Abstract

Measuring biases of vision systems with respect to protected attributes like gender and age is critical as these systems gain widespread use in society. However, significant correlations between attributes in benchmark datasets make it difficult to separate algorithmic bias from dataset bias. To mitigate such attribute confounding during bias analysis, we propose a matching approach that selects a subset of images from the full dataset with balanced attribute distributions across protected attributes. Our matching approach first projects real images onto a generative adversarial network (GAN)'s latent space in a manner that preserves semantic attributes. It then finds image matches in this latent space across a chosen protected attribute, yielding a dataset where semantic and perceptual attributes are balanced across the protected attribute. We validate projection and matching strategies with qualitative, quantitative, and human annotation experiments. We demonstrate our work in the context of gender bias in multiple open-source facial-recognition classifiers and find that bias persists after removing key confounders via matching. Code and documentation to reproduce the results here and apply the methods to new data is available at https://github.com/csinva/matching-with-gans .