Deepfake detection improves fairness across gender and skin tones
Intersectional Invariant Learning for Fair Deepfake Detection
Computer Vision and Pattern Recognition
Summary
Deepfake detectors can mistakenly accuse certain groups of people more often due to biases in their features, like skin texture. The authors propose a new way to train these detectors so their decision-making ignores demographic differences and focuses on truly detecting fakes. Their method looks for the hardest group splits and learns features that stay consistent across those groups. Tested on over a million images, their approach reduces unfair mistakes while keeping high accuracy.
What this means in practice
- •For security teams: Enhance deepfake detection systems used in content verification to reduce demographic bias and false alarms across gender and skin tone groups.
- •For social media companies: Implement fairer deepfake detection algorithms in automated moderation pipelines to ensure more consistent treatment of diverse users’ content.
Authors
Uzoamaka Ezeakunne, Taiwo Olabamipe, Deevesh Jain, Xiuwen Liu
Abstract
Deepfake detection models achieve high overall accuracy but exhibit large performance disparities across demographic groups, causing some groups to experience substantially higher false-positive rates than others. Existing fairness-aware methods address this by constraining model outputs to equalize group predictions, but do not tackle the underlying feature-level bias. As a result, models may still rely on demographic characteristics, such as skin texture, as indirect cues for detecting manipulated content. To address this problem, we propose Intersectional Invariant Learning (IIL), a framework that improves feature learning rather than only adjusting model outputs. IIL alternates between identifying the most challenging demographic partitions and learning features that remain consistent across those partitions. The framework also includes a false-positive-rate fairness objective and group-balanced sampling across six gender x skin-tone intersectional groups. We evaluate IIL on a large-scale fairness benchmark containing over one million images, 10 baseline detectors, and 21 fairness and utility metrics. Results show that IIL achieves Pareto-optimal performance, simultaneously attaining an AUC above 99%, a false-positive rate under 2.5%, and the lowest intersectional fairness disparity (FMEO) among all compared methods. IIL outperforms prior fairness-aware methods on the large majority of the 21 evaluation metrics, while reducing both false-positive rates and intersectional fairness disparities by several-fold compared with the strongest competing method.