Toward Calibrated, Fair, and accurate Deepfake Detection
Two new research papers address challenges in deepfake detection, focusing on fairness and uncertainty quantification. One paper introduces Face-Fairness (FF), a framework that mitigates bias across demographic groups without requiring sensitive identity labels. The other paper proposes Correlation-Optimized Fusion (COF), an architecture-adaptive method to improve the reliability of uncertainty estimates in deepfake detection systems, particularly under distribution shifts. AI
IMPACT Advances in deepfake detection fairness and uncertainty quantification are crucial for reliable forensic applications and combating misinformation.