Data Poisoning Attack Mitigation Quiz

Submit

2. Which approach combines multiple defense mechanisms to improve robustness?

Defense stacking

Model averaging

Ensemble defenses

Gradient ascent

Submit

4. What is the advantage of using a validation set separate from training data?

To detect performance anomalies caused by poisoning

To increase training speed

To reduce memory requirements

To simplify model architecture

5. True or False: Poisoning attacks always cause obvious performance drops that are easy to detect.

True

False

6. Which defense uses statistical tests to identify poisoned samples before training?

Spectral signature analysis

Cross-entropy minimization

Batch normalization

Attention mechanisms

Submit

8. What is model validation used for in poisoning defense?

To verify model behavior on held-out clean data

To increase training speed

To reduce computational cost

To encrypt model outputs

9. True or False: Clean-label attacks modify only features while keeping original labels unchanged.

True

False

10. Which technique isolates anomalous data points using statistical distance measures?

Outlier detection

Model compression

Hyperparameter tuning

Activation maximization

11. What is data poisoning in the context of machine learning?

Intentionally corrupting training data to degrade model performance

Using toxic language in dataset labels

Removing outliers from a dataset

Encrypting sensitive training information

12. What is gradient-based poisoning designed to do?

Craft poisoned samples that maximize model loss

Prevent gradient computation

Smooth gradient updates

Encrypt model weights

13. True or False: Certified defenses can provide formal guarantees against data poisoning attacks.

True

False

14. Which metric measures how well a defense detects poisoned samples?

Precision and recall

Model accuracy only

Training time

Memory usage

Submit

16. What is the primary goal of a backdoor attack in data poisoning?

To cause random misclassification across all inputs

To trigger specific misclassifications when a trigger pattern is present

To reduce model accuracy uniformly

To expose model parameters

17. True or False: Robust training with adversarial examples can increase model resilience to data poisoning.

True

False

18. Which defense mechanism uses multiple models trained on different data subsets?

Ensemble learning

Gradient clipping

Input sanitization

Model distillation

Submit

20. Which type of data poisoning attack modifies labels in the training set?

Feature poisoning

Label flipping

Model inversion

Gradient masking

SecAI+ Data Poisoning Attack Mitigation Quiz

1. Differential privacy in training helps mitigate data poisoning by limiting ______ of any single sample.

2.

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

2. Which approach combines multiple defense mechanisms to improve robustness?

3. Certified robustness guarantees typically provide bounds on model accuracy under ______ poisoning scenarios.

4. What is the advantage of using a validation set separate from training data?

5. True or False: Poisoning attacks always cause obvious performance drops that are easy to detect.

6. Which defense uses statistical tests to identify poisoned samples before training?

7. A poisoning attack that targets a ______ class aims to misclassify only samples from that specific class.

8. What is model validation used for in poisoning defense?

9. True or False: Clean-label attacks modify only features while keeping original labels unchanged.

10. Which technique isolates anomalous data points using statistical distance measures?

11. What is data poisoning in the context of machine learning?

12. What is gradient-based poisoning designed to do?

13. True or False: Certified defenses can provide formal guarantees against data poisoning attacks.

14. Which metric measures how well a defense detects poisoned samples?

15. Data sanitization involves ______ suspicious or anomalous samples from training data.

16. What is the primary goal of a backdoor attack in data poisoning?

17. True or False: Robust training with adversarial examples can increase model resilience to data poisoning.

18. Which defense mechanism uses multiple models trained on different data subsets?

19. Data poisoning attacks are most effective when attackers have ______ access to the training process.

20. Which type of data poisoning attack modifies labels in the training set?