SecAI+ Model Evasion Attack Defenses Quiz

1. What does adversarial robustness certification guarantee?

Perfect accuracy under all conditions

Provable bounds on model behavior under bounded perturbations

Immunity to gradient-based attacks

No computational overhead

2. Activation clustering detects adversarial examples by analyzing internal model representations.

True

False

3. What is a key limitation of adversarial training for defense?

It eliminates the need for other defenses

Robustness to seen attacks may not transfer to novel attacks

It requires no additional computational resources

It guarantees immunity to all perturbations

Submit

5. Which method verifies robustness by testing with random smoothing?

Randomized smoothing certification

Interval bound propagation

Linear programming verification

Symbolic execution

6. Certified defenses always maintain the original model accuracy without trade-offs.

True

False

7. What is the purpose of adversarial testing in model security?

Increase training data size

Identify model vulnerabilities before deployment

Reduce computational costs

Improve data labeling accuracy

Submit

9. Which of the following is a white-box attack?

FGSM with full gradient information

Query-based black-box attack

Random perturbation search

Confidence score estimation

10. Ensemble defenses combine multiple models to improve robustness against evasion.

True

False

11. What is an adversarial example in machine learning?

A training sample with intentional label noise

An input deliberately crafted to fool a trained model

A data point from an underrepresented class

An outlier detected during model validation

Submit

13. Which technique reduces input precision to remove adversarial perturbations?

Defensive distillation

Feature squeezing

Ensemble methods

Activation clustering

14. Gradient masking effectively prevents white-box adversarial attacks.

True

False

15. What is the primary goal of certified defenses in adversarial robustness?

Achieve 100% accuracy on clean data

Provide formal guarantees against perturbations within a radius

Eliminate all gradient information

Reduce model inference time

Submit

17. Which defense method adds random noise to model inputs before prediction?

Input sanitization

Randomized smoothing

Gradient masking

Feature squeezing

18. Adversarial training involves training a model on both clean and adversarial examples.

True

False

19. What does FGSM stand for in adversarial machine learning?

Fast Gradient Sign Method

Federated Gradient Scaling Model

Fault-Guided Security Mechanism

Fuzzy Gradient Support Module

20. Which attack type modifies input data while keeping the model fixed?

Poisoning attack

Evasion attack

Backdoor attack

Model inversion attack

SecAI+ Model Evasion Attack Defenses Quiz

1. What does adversarial robustness certification guarantee?

2.

What first name or nickname would you like us to use?

2. Activation clustering detects adversarial examples by analyzing internal model representations.

3. What is a key limitation of adversarial training for defense?

4. A ____ bound provides formal guarantees on model robustness radius.

5. Which method verifies robustness by testing with random smoothing?

6. Certified defenses always maintain the original model accuracy without trade-offs.

7. What is the purpose of adversarial testing in model security?

8. A ____ defense uses knowledge distillation to create a robust model.

9. Which of the following is a white-box attack?

10. Ensemble defenses combine multiple models to improve robustness against evasion.

11. What is an adversarial example in machine learning?

12. A ____ attack transfers adversarial examples from one model to fool another.

13. Which technique reduces input precision to remove adversarial perturbations?

14. Gradient masking effectively prevents white-box adversarial attacks.

15. What is the primary goal of certified defenses in adversarial robustness?

16. A ____ attack manipulates training data to degrade model performance.

17. Which defense method adds random noise to model inputs before prediction?

18. Adversarial training involves training a model on both clean and adversarial examples.

19. What does FGSM stand for in adversarial machine learning?

20. Which attack type modifies input data while keeping the model fixed?