MCQs

Disclaimer:

Educational Use Only:

The website links provided are to be used strictly for educational and knowledge-gaining purposes only.

No Misuse:

Do not misuse the links or the information contained therein for any commercial, illegal, or unethical activities.

Respect for Websites and Businesses:

Respect the intellectual property and business interests of the websites and their owners. Do not attempt to alter, copy, or redistribute content without proper authorization.

Acknowledge Value of Knowledge:

Recognize and value the knowledge and resources shared by these websites. Use the information responsibly and ethically.

No Endorsement:

The provision of these links does not imply endorsement of the linked sites, their content, or their products and services. Users are responsible for evaluating the suitability and accuracy of the information.

User Responsibility:

By accessing these links, you agree to use the information at your own risk and to comply with all applicable laws and website terms of use.

1. What type of attack occurs during the deployment phase of a machine learning model?
   a. Poisoning attack
   b. Evasion attack
   c. Model extraction attack
   d. Backdoor attack
   Answer: b. Evasion attacklumenova​
2. Which machine learning security threat involves introducing corrupted data during training?
   a. Model inversion
   b. Poisoning attack
   c. Data injection
   d. Gradient masking
   Answer: b. Poisoning attacklumenova​
3. What attack type aims to reconstruct sensitive input data through model outputs?
   a. Model inversion attack
   b. Adversarial training
   c. Property inference
   d. Prompt injection attack
   Answer: a. Model inversion attacklumenova​
4. The process of sending queries to reverse-engineer a machine learning model is called:
   a. Model extraction attack
   b. Gradient analysis
   c. Data reconstruction
   d. Robust optimization
   Answer: a. Model extraction attacklumenova​
5. Adding noise to gradient calculations to prevent adversarial perturbation is known as:
   a. Gradient masking
   b. Ensemble method
   c. Model hardening
   d. Regularization
   Answer: a. Gradient maskinglumenova​
6. Incorporating adversarial examples into training data is a defense mechanism termed:
   a. Defensive distillation
   b. Adversarial training
   c. Feature squeezing
   d. Model pruning
   Answer: b. Adversarial traininglumenova​
7. Using multiple models to enhance overall defense strength exemplifies:
   a. Ensemble methods
   b. Model inversion
   c. Gradient optimization
   d. Regularization technique
   Answer: a. Ensemble methodslumenova​
8. A proactive defense method that trains models on softened outputs for robustness is:
   a. Defensive distillation
   b. Model pruning
   c. Gradient smoothing
   d. Input normalization
   Answer: a. Defensive distillationlumenova​
9. Minimizing the precision of input data to remove adversarial noise uses:
   a. Feature squeezing
   b. Data reconstruction
   c. Model masking
   d. Robust optimization
   Answer: a. Feature squeezinglumenova​
10. What is the main benefit of real-time anomaly detection in ML systems?
    a. Early threat identification
    b. Lower computational cost
    c. Reduced model training time
    d. Increased dataset privacy
    Answer: a. Early threat identificationlumenova​
11. Which attack embeds hidden triggers in training data to alter model outputs?
    a. Backdoor attack
    b. Poisoning attack
    c. Gradient corruption
    d. Membership inference
    Answer: a. Backdoor attacksentinelone​
12. The alteration of dataset labels to mislead model learning defines:
    a. Mislabeling attack
    b. Property inference
    c. Gradient masking
    d. Model inversion
    Answer: a. Mislabeling attacksentinelone​
13. The step of removing crucial data points from training sets constitutes:
    a. Data manipulation attack
    b. Poisoning attack
    c. Adversarial evasion
    d. Prompt injection
    Answer: a. Data manipulation attacksentinelone​
14. The primary defense against adversarial perturbations is:
    a. Adversarial training
    b. Gradient amplification
    c. Dataset pruning
    d. Model inversion
    Answer: a. Adversarial traininglumenova​
15. Which attack allows reconstruction of individual training samples from model output?
    a. Model inversion attack
    b. Membership inference attack
    c. Model extraction
    d. Gradient exploitation
    Answer: a. Model inversion attacklumenova​
16. Introducing malicious queries to AI chatbots to alter output is:
    a. Prompt injection attack
    b. Data poisoning
    c. Feature squeezing
    d. Reinforcement bypassing
    Answer: a. Prompt injection attacklumenova​
17. What process monitors input anomalies and model confidence levels for security?
    a. Real-time monitoring
    b. Ensemble training
    c. Data pruning
    d. Gradient masking
    Answer: a. Real-time monitoringlumenova​
18. Model hardening primarily aims to achieve:
    a. Resistance against small perturbations
    b. Dataset expansion
    c. Unified model interpretation
    d. Reduced classifier variance
    Answer: a. Resistance against small perturbationslumenova​
19. A defense using real-time pattern analysis and automatic alert generation is:
    a. Monitoring and anomaly detection
    b. Gradient smoothing
    c. Defensive consolidation
    d. Ensemble tracking
    Answer: a. Monitoring and anomaly detectionlumenova​
20. What is the major trade-off in adversarial defense strategies?
    a. Robustness vs. accuracy
    b. Scalability vs. latency
    c. Security vs. transparency
    d. Cost vs. usability
    Answer: a. Robustness vs. accuracylumenova​
21. An attacker recovering model weights using side-channel leakage applies:
    a. Privacy side-channel attack
    b. Prompt injection
    c. Ensemble bypassing
    d. Mislabeling
    Answer: a. Privacy side-channel attackusenix​
22. Attempting to infer if an individual’s data is in a training set defines:
    a. Membership inference
    b. Model extraction
    c. Adversarial emulation
    d. Prompt optimization
    Answer: a. Membership inferencetelefonicatech​
23. The act of deducing sensitive attributes from overall dataset characteristics is:
    a. Property inference attack
    b. Data poisoning
    c. Prompt injection
    d. Model inversion
    Answer: a. Property inference attacktelefonicatech​
24. Reconstructing individual data points from shared model statistics is:
    a. Data reconstruction attack
    b. Model extraction
    c. Gradient tracing
    d. Ensemble attack
    Answer: a. Data reconstruction attacktelefonicatech​
25. A combined approach applying multiple models to improve robustness is:
    a. Ensemble method
    b. Gradient masking
    c. Data anonymization
    d. Label smoothing
    Answer: a. Ensemble methodlumenova​
26. Gradient masking can create what negative effect?
    a. False security perception
    b. Model completeness loss
    c. Increased prediction latency
    d. Weaker overfitting control
    Answer: a. False security perceptionlumenova​
27. Which method explicitly strengthens decision boundaries during training?
    a. Robust optimization
    b. Defensive pruning
    c. Gradient inversion
    d. Data caching
    Answer: a. Robust optimizationlumenova​
28. Excessive computational demand in ML defense arises in:
    a. Adversarial training
    b. Model extraction
    c. Gradient suppression
    d. Feature anonymization
    Answer: a. Adversarial traininglumenova​
29. Ensuring data sanitization before training helps prevent:
    a. Poisoning attacks
    b. Gradient masking
    c. Data reconstruction
    d. Model emulation
    Answer: a. Poisoning attacksarxiv​
30. Which attack manipulates model predictions using falsified training data?
    a. Data poisoning
    b. Prompt injection
    c. Model hardening
    d. Gradient tracing
    Answer: a. Data poisoningarxiv​
31. Effective defense frameworks require standardized evaluation through:
    a. Robustness metrics
    b. Sample filtering
    c. Accuracy extrapolation
    d. Confidence propagation
    Answer: a. Robustness metricslumenova​
32. Attacks that alter output confidence without visible data changes are:
    a. Evasion attacks
    b. Model inversion
    c. Gradient perturbations
    d. Data injections
    Answer: a. Evasion attackslumenova​
33. Model extraction attacks pose a risk of:
    a. Intellectual property theft
    b. False-positive anomalies
    c. Training noise generation
    d. Adversarial latency
    Answer: a. Intellectual property theftlumenova​
34. Which ML component most often exploited in black-box adversarial attacks?
    a. Output probabilities
    b. Gradient weights
    c. Internal databases
    d. Encryption layer
    Answer: a. Output probabilitieslumenova​
35. A method combining adversarial and clean inputs to generate resilient models uses:
    a. Adversarial training
    b. Label correction
    c. Feature reconstruction
    d. Ensemble balancing
    Answer: a. Adversarial traininglumenova​
36. Gradient analysis detects attacks by examining:
    a. Model sensitivity changes
    b. Access controls
    c. Dataset volume
    d. Output encryption
    Answer: a. Model sensitivity changeslumenova​
37. Model hardening promotes:
    a. Stable decision boundaries
    b. Hidden layer bypass
    c. Dataset obfuscation
    d. Output minimization
    Answer: a. Stable decision boundarieslumenova​
38. Model inversion can lead to:
    a. Privacy breaches
    b. Gradient regularization
    c. Model simplification
    d. Storage loss
    Answer: a. Privacy breacheslumenova​
39. Ensemble methods require:
    a. Synchronization of multiple models
    b. Reduction of single-model variance
    c. Elimination of multi-path inputs
    d. Propagation limiting
    Answer: a. Synchronization of multiple modelslumenova​
40. Input transformation methods clean incoming data by:
    a. Filtering adversarial perturbations
    b. Encrypting input channels
    c. Detecting overfitting
    d. Flattening layers
    Answer: a. Filtering adversarial perturbationslumenova​
41. Backdoor attacks primarily threaten:
    a. Model integrity and trust
    b. Input data precision
    c. Gradient accuracy
    d. Optimization time
    Answer: a. Model integrity and trustsentinelone​
42. Which category of attack occurs during model testing or production?
    a. Evasion attack
    b. Poisoning attack
    c. Data reconstruction
    d. Membership inference
    Answer: a. Evasion attacklumenova​
43. Adversarial machine learning in cybersecurity primarily studies:
    a. Model vulnerability and defense
    b. Dataset compression methods
    c. Supervised classification
    d. Gradient representation
    Answer: a. Model vulnerability and defenselumenova​
44. What layer of defense uses real-time deviation tracking?
    a. Monitoring and anomaly detection
    b. Model generalization
    c. Data expansion
    d. Feature fusion
    Answer: a. Monitoring and anomaly detectionlumenova​
45. Defensive distillation relies heavily on tuning of:
    a. Softmax temperature
    b. Learning rate
    c. Model bias
    d. Data diversity
    Answer: a. Softmax temperaturelumenova​
46. Combining prediction results from models reduces:
    a. Impact of single-model fraud
    b. Total training loss
    c. Gradient computation cost
    d. Memory overflow risk
    Answer: a. Impact of single-model fraudlumenova​
47. AI models susceptible to adversarial text manipulation face:
    a. Prompt injection threats
    b. Overfitting issues
    c. Transformer bleeding
    d. Batch encoding errors
    Answer: a. Prompt injection threatslumenova​
48. Poisoning models can lead to:
    a. Embedded backdoors
    b. Gradient stability
    c. Model freezing
    d. Reduced bias
    Answer: a. Embedded backdoorsarxiv​
49. Feature squeezing as a defense reduces input:
    a. Precision levels
    b. Label variance
    c. Output entropy
    d. Layer frequency
    Answer: a. Precision levelslumenova​
50. A system showing sudden prediction confidence spikes likely indicates:
    a. Adversarial attack detection signal
    b. Underfitting
    c. Data scarcity
    d. Structural redundancy
    Answer: a. Adversarial attack detection signallumenova​

1. https://www.lumenova.ai/blog/adversarial-attacks-ml-detection-defense-strategies/
2. https://www.sentinelone.com/cybersecurity-101/cybersecurity/data-poisoning/
3. https://www.usenix.org/system/files/usenixsecurity24-debenedetti.pdf
4. https://telefonicatech.com/en/blog/attacks-ai-privacy-attacks
5. https://arxiv.org/html/2204.05986v3
6. https://www.sanfoundry.com/1000-cyber-security-questions-answers/
7. https://examradar.com/artificial-intelligence-ai-and-machine-learning-ml-in-cyber-security-mcqs/
8. https://www.igmguru.com/blog/cyber-security-mcqs
9. https://interviewkickstart.com/blogs/articles/machine-learning-mcqs-answers
10. https://www.interviewbit.com/cyber-security-mcq/
11. https://www.tripwire.com/state-of-security/understanding-machine-learning-attacks-techniques-and-defenses
12. https://testbook.com/objective-questions/mcq-on-cyber-security-and-emerging-technologies–5eea6a0d39140f30f369e309
13. https://www.aionlinecourse.com/ai-quiz-questions/machine-learning/adversarial-learning
14. https://www.sciencedirect.com/science/article/abs/pii/S0140366422002146
15. https://interviewkickstart.com/blogs/articles/cyber-security-mcqs-answers
16. https://hirevire.com/pre-screening-interview-questions/adversarial-machine-learning-specialist
17. https://www.darkreading.com/application-security/only-250-documents-poison-any-ai-model
18. https://dl.acm.org/doi/10.1145/3624010
19. https://gca.isa.org/blog/how-to-secure-machine-learning-data
20. https://www.theknowledgeacademy.com/blog/cyber-security-mcq-and-answers/