This document summarizes a research paper on generalizing deepfake detection using knowledge distillation and representation learning. The proposed FReTAL framework uses knowledge distillation to transfer knowledge from a teacher model to a student model, preventing catastrophic forgetting when adapting to new domains without source data. It also applies representation learning to leverage similar features between domains. Experiments show FReTAL outperforms baselines on deepfake benchmark datasets, achieving up to 86.97% accuracy on low-quality deepfake detection.