數據科學系列演講Data Science Seminar

講題:Confronting the Evolution of DeepFake: From Generative Models to Detection Challenges

日期:2025.09.10 (Wed.) 15:30-17:20

地點:線上演講(會議連結：https://meet.google.com/xjn-bjro-znh)

主持人：曾新穆 教授

Abstract

This talk addresses the rapid evolution of DeepFake technology, from the progress of generative models such as GANs and diffusion methods to the increasing challenges in reliable detection. We first introduce the development of DeepFake generation, highlighting its accessibility and the threats it poses to multimedia authenticity, security, and trust. We then review recent advances in detection techniques, including classification-based, multi-modal, and robust approaches against compression and adversarial attacks. The presentation also discusses our contributions in semi-supervised and self-supervised learning frameworks, graph-regularized methods, and adversarial defense strategies, which aim to enhance generalization and real-world applicability. Finally, we explore open problems in dataset diversity, explainability, and cross-domain generalization, and point to future directions for building practical, lightweight, and trustworthy detection systems.

Speaker Bio

Prof. Chih-Chung Hsu received his Ph.D. degree in Electrical Engineering from National Tsing Hua University in 2014. He co-founded AI.SKOPY in 2017, serving as Chief Technology Officer, and previously held the role of R&D Director at Vanda Technology. He was an Assistant Professor at National Pingtung University of Science and Technology (2018–2021) and later joined the Institute of Data Science at National Cheng Kung University as an Associate Professor (2021–2025). Since 2025, he has been an Associate Professor with the College of Artificial Intelligence at National Yang Ming Chiao Tung University. Dr. Hsu’s research lies in computer vision and robust deep learning, with applications spanning DeepFake detection, super-resolution, hyperspectral imaging, medical image analysis, and autonomous driving vision. His work has been published in prestigious venues, including IEEE TPAMI, IEEE TIP, IEEE TMM, IEEE TGRS, ACM Multimedia, IEEE ICIP, IEEE ICCV, and IEEE CVPR. He has received multiple honors, including the Future Tech Award, the Y.Z. Hsu Scientific Paper Award, the IEEE Tainan Section Best Professional Member Award, and the Best Student Paper Award at IEEE ICIP. Together with his Advanced Computer Vision Lab (ACVLab), he has led students to achieve over 40 international competition awards, including championships and top-three finishes at major venues such as CVPR, ICCV, ICPR, WACV, and IGARSS. Beyond academic achievements, Dr. Hsu actively bridges research and industry through national and corporate collaborations, particularly in the fields of multimedia forensics, satellite imaging, and AI security. He continues to drive impactful projects integrating trustworthy AI, cross-modal learning, and sustainable computing.