Mitigate Adversarial Risks CY0-001 Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   Which practice is a key component of a secure AI development lifecycle?

   • AConducting regular security reviews and threat modeling at each phase
   • BKeeping the model architecture secret to prevent attacks
   • COnly testing the model on clean, unperturbed data
   • DDeploying the model immediately after initial training
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   Which of the following is a common framework used for threat modeling that involves drawing a data flow diagram and identifying threats using a structured approach?

   • ACVSS
   • BOWASP Top 10
   • CSTRIDE
   • DISO 27001
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   Which defense mechanism uses multiple models and aggregates their predictions to reduce the impact of an adversarial example that might fool a single model?

   • AModel ensembling
   • BModel distillation
   • CModel quantization
   • DModel pruning
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A pharmaceutical company uses an AI model to predict drug interactions. The regulatory body requires that the model's decisions be explainable, but the data science team wants to use a deep learning model that is highly accurate but not inherently interpretable. Which approach would best satisfy both requirements?

   • AMonitor the model's performance on a validation set to ensure accuracy.
   • BRetrain the model with more data to improve its accuracy.
   • CUse a simpler model like logistic regression that is inherently interpretable.
   • DUse a post-hoc explainability technique like LIME to generate explanations for each prediction.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A telecom company uses an AI model to predict customer churn. The security team has set up monitoring that tracks the model's performance metrics. They notice that the model's accuracy has dropped, but they are unsure if it is due to an adversarial attack or a natural shift in customer behavior. Which monitoring technique would best help them differentiate between the two?

   • ADeploy the model to a test environment to see if it performs better.
   • BRetrain the model with the latest data to see if accuracy improves.
   • CIncrease the model's complexity to better capture customer behavior.
   • DAnalyze the model's prediction confidence and feature attributions over time to detect anomalies.
   Reveal Solution Discussion Select an answer first