Adversarial Machine Learning Techniques COASP Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A security analyst is evaluating the robustness of a malware detection model. They want to generate adversarial malware samples that evade detection. Which technique should they use?

   • AFeature selection
   • BModel pruning
   • CDifferential privacy
   • DGradient-based perturbation
   Reveal Solution Discussion Select an answer first
2. 17expert · hard

   A security team is defending a production image classifier. They have a limited budget and need to choose a defense that reduces the success rate of both white-box and black-box evasion attacks while minimizing the impact on model accuracy. Which approach should they choose?

   • AAdversarial training with a diverse set of attack methods
   • BModel inversion defense
   • CInput sanitization using a single filter
   • DGradient masking
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is the main idea behind adversarial training as a defensive technique?

   • AIncluding adversarial examples in the training set so the model learns to handle them.
   • BUsing a more complex model architecture to confuse attackers.
   • CRemoving all input validation to increase processing speed.
   • DTraining the model exclusively on clean data to avoid noise.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A healthcare organization uses an AI model to triage X-ray images. They are concerned about an attacker subtly altering images to change the diagnosis. What is the most important first step in defending against this threat?

   • AConduct a threat model to identify attack vectors and potential impact.
   • BUse a more complex model architecture.
   • CRestrict access to the model's API.
   • DImmediately implement adversarial training on all images.
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A company's AI model for detecting network intrusions is being attacked. The attackers have access to the model's confidence scores (not just the final label). The company wants to deploy a defense that does not require retraining the model. Which defensive technique is most suitable?

   • AAdding a detection layer that flags inputs with high confidence but unusual patterns.
   • BAdversarial training with new examples.
   • CInput sanitization to remove suspicious patterns before feeding the model.
   • DModel distillation to simplify the model.
   Reveal Solution Discussion Select an answer first