Through a fair looking-glass: mitigating bias in image datasets

TL;DR

This paper introduces a fast, effective method for reducing bias in image datasets by reconstructing images and minimizing dependence between target and protected attributes, improving fairness without high computational costs.

Contribution

The study presents a novel de-biasing model combining a U-net for image reconstruction with a pre-trained classifier to minimize statistical dependence, offering a computationally efficient alternative.

Findings

Achieves a good fairness-accuracy trade-off on CelebA dataset.

Outperforms existing de-biasing methods in efficiency and effectiveness.

Reduces bias reflected in trained models without significant loss of accuracy.

Abstract

With the recent growth in computer vision applications, the question of how fair and unbiased they are has yet to be explored. There is abundant evidence that the bias present in training data is reflected in the models, or even amplified. Many previous methods for image dataset de-biasing, including models based on augmenting datasets, are computationally expensive to implement. In this study, we present a fast and effective model to de-bias an image dataset through reconstruction and minimizing the statistical dependence between intended variables. Our architecture includes a U-net to reconstruct images, combined with a pre-trained classifier which penalizes the statistical dependence between target attribute and the protected attribute. We evaluate our proposed model on CelebA dataset, compare the results with a state-of-the-art de-biasing method, and show that the model achieves a promising fairness-accuracy combination.