Quiet Triggers, Loud Footprints: A Tri-Space Measurement Study of Backdoor Stealthiness

본문 미리보기

arXiv:2610.19947v1 Announce Type: new Abstract: Backdoor attacks on deep neural networks are commonly evaluated by attack success rate and benign accuracy, yet their stealthiness remains inconsistently defined and measured. Existing work often studies only one observable aspect of stealthiness, such as trigger imperceptibility, feature-space separability, parameter-space concentration, or evasion of particular defenses. This creates a fragmented picture: an attack may appear stealthy under one