发表机构
AAAI Press（AAAI出版社）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究发现基础模型的特征幅度可区分真实与伪造媒体，将深度伪造检测建模为异常检测，简单统计指标即可达到竞争性能，且模型规模越大判别信号越强。
AI 中文摘要
预训练的自监督表示已成为当前深度伪造检测方法的核心组成部分，但目前尚不清楚其哪些特性能区分真实媒体与伪造媒体。在本研究中，我们发现了一个惊人一致的现象：在多个预训练模型、数据集以及图像和视频领域中，伪造样本的表示幅度系统性低于真实样本。基于这一发现，我们将深度伪造检测建模为异常检测问题，并证明特征幅度的简单统计指标能达到与复杂得多的深度伪造检测方法相当的性能。我们进一步探究了该效应的起源，证明特征幅度降低主要与伪造内容引入的语义偏移相关，而低级生成指纹的作用相对较小。最后，我们表明该判别信号随基础基础模型的规模增大而增强，这表明表示学习的进展可自然转化为更强的零样本深度伪造检测器。
英文摘要
Pretrained self-supervised representations have emerged as a core component of current deepfake detection methods, yet it remains unclear which of their properties make real and fake media distinguishable. In this work, we uncover a surprisingly consistent phenomenon: across multiple pretrained models, datasets, and both image and video domains, fake samples systematically produce lower-magnitude representations than their real counterparts. Motivated by this finding, we formulate deepfake detection as an anomaly detection problem and show that simple statistics of feature magnitude achieve competitive performance with far more sophisticated deepfake detection methods. We further investigate the origin of this effect and demonstrate that reduced feature magnitude is primarily associated with semantic shifts introduced by fake content, while low-level generative fingerprints play a comparatively smaller role. Finally, we show that this discriminative signal strengthens as the size of the underlying foundation model grows, suggesting that advances in representation learning naturally translate into stronger zero-shot deepfake detectors.
