Recent Advances and Challenges of Deepfake Detection

Currently, the rapid advancement of generative artificial intelligence, particularly diffusion models and large language models, has enabled the creation or manipulation of high-quality multimedia. Although beneficial in numerous fields, AI generated content presents significant risks, including potential misuse and ethical concerns. Consequently, detecting the generated multimedia has become increasingly popular in related research. This webinar presents a comprehensive overview of recent deep learning-based approaches for AI generated content detection. Our presenter also introduces their Masked Relation Learning method that formulates detection as a graph classification problem, and test-time detection method with spatial-frequency prompt learning. Furthermore, he will identify current challenges in detection and propose directions for future research.