Decoupling Bias, Aligning Distributions: Synergistic Fairness Optimization for Deepfake Detection

ArXi:2511.10150v5 Announce Type: replace Fairness is a core element in the trustworthy deployment of deepfake detection models, especially in the field of digital identity security. Biases in detection models toward different graphic groups, such as gender and race, may lead to systemic misjudgments, exacerbating the digital divide and social inequities. However, current fairness-enhanced detectors often improve fairness at the cost of detection accuracy. To address this challenge, we propose a dual-mechanism collaborative optimization framework.