On Biased Behavior of GANs for Face Verification

TL;DR

This paper investigates biases in GAN-generated face data, revealing that such synthetic data can reinforce racial and age-related disparities in face verification systems, raising fairness concerns.

Contribution

It highlights the biased behavior of GANs trained on FFHQ, demonstrating their impact on fairness in face verification when used for data augmentation.

Findings

GANs trained on FFHQ generate biased face data towards white, young adults.

Synthetic faces cause disparate impact in race attribute for face verification.

Bias in GAN-generated data affects the fairness of verification systems.

Abstract

Deep Learning systems need large data for training. Datasets for training face verification systems are difficult to obtain and prone to privacy issues. Synthetic data generated by generative models such as GANs can be a good alternative. However, we show that data generated from GANs are prone to bias and fairness issues. Specifically, GANs trained on FFHQ dataset show biased behavior towards generating white faces in the age group of 20-29. We also demonstrate that synthetic faces cause disparate impact, specifically for race attribute, when used for fine tuning face verification systems.