Report Title: Generative Capability Driven Deep Fake Defense Method

Reporter: Professor Zhang Weiming

Unit: University of Science and Technology of China

Report Time: June 23, 2023 (Friday) 15:30-17:30

Report Location: The third conference room of the Jade Science and Education Building A, Jade Lake Campus

Report Abstract: Deep fake technology represented by video face swapping may be used for economic fraud, public opinion misguidance, threatening personal information security and national political security. With the rapid development of generative artificial intelligence technology, deep fakes are evolving towards a stage with high quality and adversarial nature, dragging deep fake detection that requires a large amount of training data into an asymmetric adversarial dilemma. In this context, the ability to deeply synthesize largely determines the detection ability. Therefore, this report will introduce the defense methods of deep fake videos from the perspective of generation, and discuss the active and passive detection methods of GPT-generated text.

Reporter Introduction: Zhang Weiming is currently the deputy dean, professor, and doctoral supervisor of the School of Cyber Security at the University of Science and Technology of China. His main research interests include information hiding and artificial intelligence security. He has published more than 200 papers in internationally renowned academic journals and conferences such as IEEE TIT, TPAMI, TIFS, TIP, CVPR, S&P, ICCV, NeurIPS, AAAI. He has presided over more than 20 projects such as the National Natural Science Foundation Key, National Key R&D Topics, and National 863. He has won the first prize for military scientific and technological progress, the first prize for natural science in Anhui Province, the special prize for teaching achievements in Anhui Province, and the ACM SIGSOFT Outstanding Paper Award. He led the team to participate in the development of the most influential online face deep synthesis open source tool DeepFaceLab, with a total download of more than 120,000 times, and was selected as one of the top ten popular projects on GitHub 2020 with OpenAI's GPT-2 project; participated in the world's largest "Deep Fake Detection Challenge (DFDC)" organized by Facebook, MIT, etc., won the runner-up (a prize of 300,000 US dollars), and was rated as one of the eight Chinese innovation achievements in the field of artificial intelligence security since 2014 by the China Academy of Information and Communications Technology. He was selected as one of the top ten outstanding figures in artificial intelligence in the Yangtze River Delta in 2021.