Liu, P., Tao, Q., & Zhou, J. (2024, October). Robust deepfake detection by addressing generalization and trustworthiness challenges: A short survey. In Proceedings of the 32nd ACM International Conference on Multimedia (MM '24) (pp. 3–11). ACM. https://doi.org/10.1145/3689090.3689386
Abstract:
The rapid advancement of deep learning technologies has led to a proliferation of deepfake content, posing serious threats to security, privacy, and trust in digital information. This paper explores the evolution of robust deepfake detection methods over recent years from two perspectives, specifically, generalized and trustworthy detection. As numerous generative techniques continuously emerge, generalized detection emphasizes the robustness against data distribution shift represented by unseen manipulation at testing time. By systematically reviewing generalized detection methods, we categorize these approaches into input-level, model-level, and learning-level. Trustworthy detection aims to enhance robustness against attacks that maliciously fail the detection system, including adversarial and backdoor attacks. To address these threats, researchers have developed robust defense strategies, including adversarial feature similarity learning and ensemble methods. By providing an overview of robust detection methods, attack techniques, and defense strategies, this paper highlights the challenges and advancements in creating reliable and generalizable deepfake detection systems.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the National Research Foundation Singapore and the Ministry of Digital Development and Information - Online Trust and Safety (OTS) Research Programme
Grant Reference no. : MCI-OTS-001
This research / project is supported by the A*STAR Science and Engineering Research Council - Central Research Fund (Use-Inspired Basic Research)
Grant Reference no. :
Description:
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org