Attention Mask Guided PGD Adversarial Attacks Improve Stealth, Efficiency, and Explainability
This research paper introduces a novel attention mask-guided PGD adversarial attack method that outperforms existing methods in achieving a balance between stealth, efficiency, and explainability, effectively fooling XAI-based safety monitors for image classification.
The Vulnerability of Image and Text Classification Algorithms to Adversarial Attacks Using GANs, SMOTE, FGSM, and GradCAM
Machine learning models, particularly those used for text and image classification, are highly susceptible to adversarial attacks, which can significantly reduce their accuracy and reliability.
Edge-Attack: A Novel Generative Adversarial Patch Attack on Cross-Modal Pedestrian Re-Identification Models
This research paper introduces Edge-Attack, a novel method for generating physically realizable adversarial patches that exploit the vulnerability of cross-modal pedestrian re-identification (VI-ReID) models by targeting their reliance on shallow edge features.
Hard-label based Small Query Black-box Adversarial Attack: Improving Query Efficiency
Proposing a new practical setting of hard-label based attack with an optimization process guided by a pre-trained surrogate model significantly improves query efficiency in black-box attacks.
Auswirkung der Umgebungs-Intrinsischen Dimensionenlücke auf die anfälligkeit für Angriffe
Die Existenz von Angriffen auf maschinelles Lernen, die für Menschen unmerklich sind, ist eine natürliche Konsequenz der Dimensionenlücke zwischen den intrinsischen und Umgebungsdimensionen der Daten.
Verbesserung von Adversarialen Angriffen auf Latent Diffusion Model
Verbesserung von Adversarialen Angriffen auf das Latent Diffusion Model durch die Einführung von konsistenten Fehlermustern.
© 2024 by Linnk AI