ARE YOU CERTAIN THAT IT IS A DEEPFAKE? DETECTION, GENERATION, AND SOURCE DETECTION FROM AN UNCERTAINTY PERSPECTIVE
Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: deepfake detection, deepfake source detection, uncertainty quantification
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: As generative models are advancing in quality and quantity for creating synthetic content, deepfakes begin to cause online mistrust. Deepfake detectors are proposed to counter this effect, however, misuse of detectors claiming fake content as real or vice versa further fuels this misinformation problem. In this paper, we evaluate, compare, and analyze the uncertainty of these deepfake detectors. As reflected in detectors' responses, deepfake generators also contribute to this uncertainty as their generative residues vary, so we cross the uncertainty analysis of deepfake detectors and generators. Based on our observations, the uncertainty manifold holds enough consistent information to leverage uncertainty for deepfake source detection. We evaluate uncertainty on two datasets with nine generators, with four blind and two biological detectors, compare different uncertainty methods, explore region- and pixel-based uncertainty, and conduct ablation studies. We conduct and analyze binary real/fake, multi-class real/fake, source detection, and leave-one-out experiments between the generator/detector combinations to share their generalization capability, model calibration, uncertainty, and robustness against adversarial attacks. This comprehensive, uncertainty-forward analysis addresses a critical gap in current deepfake detection understanding that we believe can help drive future improvements to detectors and thus restore trust in media in the age of generative AI.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6193
Loading