Kernel Space Conditional Distribution Alignment for Improving Group Fairness in Deepfake Detection

TMLR Paper5170 Authors

Abstract: We introduce FairAlign, a new method to reduce bias and improve group fairness in deepfake detection by aligning conditional distributions of embeddings in a high-dimensional kernel space. Our approach reduces information related to sensitive attributes in the embedding space that could potentially bias the detection process, thus promoting fairness. FairAlign is a versatile plug-and-play loss term compatible with various deepfake detection networks and is capable of enhancing group fairness without compromising detection performance. In addition to applying FairAlign for reducing gender bias, we implement a systematic pipeline for the annotation of skin tones and promotion of fairness in deepfake detection related to this sensitive attribute. Finally, we perform the first comprehensive study toward quantifying and understanding the trade-off between fairness and accuracy in the context of deepfake detection. We use three public deepfake datasets FaceForensics++, CelebDF, and WildDeepfake to evaluate our method. Through various experiments, we observe that FairAlign outperforms other bias-mitigating methods across various deepfake detection backbones for both gender and skin tone, setting a new state-of-the-art. Moreover, our fairness-accuracy trade-off analysis demonstrates that our approach demonstrates the best overall performance when considering effectiveness in both deepfake detection and reducing bias. We release the code at: https://anonymous.4open.science/r/FairAlign-170F.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Samira_Ebrahimi_Kahou1

Submission Number: 5170