Newest papers

MIG-COW: Transferable Adversarial Attacks on Deepfake Detectors via Gradient Decomposition

Wonjune Seo, Joonhyuk Baek, Y. Jung, Saerom Park · 2025 · ACM International Conference on Multimedia (ACM MM)

DOI: 10.1145/3746027.3761986 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Apesar dos avanços recentes, os detectores de falsificação profunda permanecem vulneráveis a exemplos adversários, particularmente em ambientes reais diversos. Proponemos o MIG-COW, um novo framework de ataque adversário que gera exemplos adversários altamente generalizáveis e visualmente imperceptíveis. Ao combinar gradientes integrados de impulso com uma decomposição consensual-ortogonal, o MIG-COW captura vulnerabilidades compartilhadas e específicas de modelos em detectores heterogêneos CNN e ViT. No AADD-2025 Challenge benchmarks, o MIG-COW atinge uma taxa de sucesso de ataque de caixa branca (ASR) de 99,96% com alta semelhança perceptual.

Machine translation (Meta M2M100); the English original follows.

Abstract

Despite recent advances, deepfake detectors remain vulnerable to adversarial examples, particularly in diverse, real-world settings. We propose MIG-COW, a novel adversarial attack framework that generates highly generalizable and visually imperceptible adversarial examples. By combining momentum-integrated gradients with a consensus-orthogonal decomposition, MIG-COW captures both shared and model-specific vulnerabilities across heterogeneous CNN and ViT detectors. On the AADD-2025 Challenge benchmarks, MIG-COW achieves a 99.96% white-box attack success rate (ASR) with high perceptual similarit

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Adversarial system, Computer science, Algorithm, Detector, Key (lock), Decomposition

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange