) is a technique used to deceive artificial intelligence models by manipulating their internal data representations. ScienceDirect.com +1 Rather than changing the final output directly, these attacks target the "features"—the specific characteristics or patterns a model uses to make decisions. arXiv +1 Key Types of Feature Attacks Feature Disruption Attack (FDA): This method optimizes adversarial examples by disrupting the intermediate features extracted by deep neural networks (DNNs), causing the model to lose its ability to recognize objects correctly. Feature Importance-Aware Attack (FIA): This technique identifies which features are most "important" to a model’s decision and specifically targets those to maximize the chances of a successful attack. Feature Permutation Attack (FPA): This rearranges pixel values in selected feature maps to improve the attack's "transferability," meaning the attack works even on different types of AI architectures. Decoupled Feature Attack (DEFEAT): This attack manipulates internal representations to ensure that an adversarial image created for one model can successfully "fool" a different, even more robust model. ScienceDirect.com +5 Common Use Cases Phishing Detection Evasion: Researchers use Feature Importance Guided Attacks (FIGA) to bypass security filters by modifying the most critical data points used to identify malicious emails. Object Detection Sabotage: Attacks like GHFA target object detectors (like those in self-driving cars) by weighting specific feature maps to hide objects from the AI. Authentication Bypassing: Some attacks target features related to "styles" (like colors or outlines) to create images that look natural to humans but are seen as completely different objects by a computer. ScienceDirect.com +2 Related Security Features If you are looking for "Attack" features in specific software: 12 sites Gradient-guided hierarchical feature attack for object detector Jan 15, 2024 —
Every attack follows a specific lifecycle, often referred to as the . This process involves identifying a vulnerability and using an attack vector (the path or means) to deliver a payload. ) is a technique used to deceive artificial
If you enjoy the "shopping experience"—wide aisles, in-store bakeries, and aesthetic lighting—ATAC will disappoint you. ScienceDirect
: The method used to exploit that weakness, such as an email attachment, a malicious website, or a compromised USB drive. such as an email attachment