Preserving Fairness Generalization in Deepfake Detection | Read Paper on Bytez