Tag
#evasion-attacks
3 posts tagged evasion-attacks.
- deep-dive
PLAA: What a 92.78% NIDS Evasion Rate Actually Tells You About Feature-Space Attacks
A new arXiv paper builds adversarial network traffic at the packet level instead of the flow level, hitting a 92.78% evasion rate against deep-learning NIDS. Here's why that framing matters more than the number.
- primer
Adversarial Examples vs. Data Poisoning: Timing Is Everything
Adversarial examples attack a deployed model at inference; data poisoning attacks the model before it is deployed.
- attacks
Evasion Attacks on Image Classifiers: FGSM, PGD, and C&W
The three foundational gradient-based evasion attacks, what each one actually optimizes, and what the benchmark numbers mean when you're evaluating a defense.