LATIDIA · Ciberseguridad

Desencadenantes silenciosos, huellas fuertes: un estudio de medición triespacial de la sigilosidad de la puerta trasera

arXiv: 2610.19947v1Tipo de anuncio: nuevo Resumen: los ataques de puerta trasera en redes neuronales profundas se evalúan comúnmente por la tasa de éxito del ataque y la precisión benigna, sin embargo, su sigilo sigue estando definido y medido de manera inconsistente

La noticia

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 metric while leaving conspicuous footprints elsewhere. This paper presents a tri-space measurement study of backdoor stealthiness in image classification. We distinguish latent stealthiness---the ideal but unobservable difficulty of distinguishing a backdoored system from a benign one---from measurable footprints in three spaces: input, feature, and parameter space.