Create a custom practice set
Pick category, difficulty, number of questions, and time limit. Start instantly with your own quiz.
Generate QuizGK Questions
Filter by category, type, and difficulty. Reading is open for everyone.
The technique that adds calibrated noise to model gradients during training to provide privacy guarantees is called ________-SGD.
Answer: Differential Privacy / DP
DP-SGD adds noise to gradients during training, providing mathematical privacy guarantees. Critical for training ML models on sensitive data while preventing membership inference.
Model inversion attacks can reconstruct sensitive training data from model predictions.
- True
- False
Answer: True
Model inversion exploits model confidence scores to infer sensitive attributes of training data. Mitigation: differential privacy, output perturbation, access controls.
Which defense technique trains ML models with adversarial examples to improve robustness?
- Adversarial Training
- Defensive Distillation
- Randomized Smoothing
- All of these
Answer: All of these
Adversarial defenses: adversarial training includes perturbed examples, defensive distillation smooths outputs, randomized smoothing adds noise. All improve robustness against adversarial attacks.
The practice of scanning container images for vulnerabilities before deployment is called ________ Scanning.
Answer: Container / Image
Container image scanning identifies known vulnerabilities, misconfigurations, and secrets in images before deployment. Critical for secure CI/CD pipelines in cloud-native environments.
Service mesh provides mutual TLS encryption for service-to-service communication in microservices.
- True
- False
Answer: True
Service mesh (Istio, Linkerd) automatically encrypts service-to-service traffic with mTLS, provides authentication, authorization, and observability without application code changes.
Which Kubernetes security practice restricts pod capabilities to minimize attack surface?
- Pod Security Policies
- Network Policies
- Resource Limits
- All of these
Answer: All of these
Kubernetes security: Pod Security Policies restrict privileges, Network Policies control traffic, Resource Limits prevent DoS. Combined for defense-in-depth in container environments.
The practice of continuously assessing device security posture before granting access is called ________ Assessment.
Answer: Device Posture
Device posture assessment evaluates endpoint security: OS patches, antivirus status, encryption, compliance. Critical for Zero Trust access decisions in hybrid environments.
Micro-segmentation in Zero Trust limits lateral movement by isolating workloads at the application level.
- True
- False
Answer: True
Micro-segmentation creates granular security policies around individual workloads/applications, containing breaches and preventing attackers from moving laterally across the network.
Which Zero Trust principle requires continuous verification of trust assumptions?
- Never Trust, Always Verify
- Assume Breach
- Least Privilege Access
- All of these
Answer: All of these
Zero Trust combines: never trust always verify (continuous authentication), assume breach (detect/contain threats), least privilege (minimum access). All principles work together for comprehensive security.
Step-by-step reasoning in LLMs is called Chain-of-________.
Answer: Thought
Chain-of-Thought prompting generates reasoning steps.
Fine-tuning adapts LLMs to specific tasks.
- True
- False
Answer: True
Fine-tuning updates weights on task-specific data.
Which architecture underpins modern LLMs?
- CNN
- RNN
- Transformer
- LSTM
Answer: Transformer
Transformer uses self-attention for parallel sequence processing.
Grouping similar data without labels is called ________.
Answer: Clustering
Clustering discovers natural groupings in unlabeled data.
Feature engineering creates new variables to improve performance.
- True
- False
Answer: True
Feature engineering: transformations, interactions, encoding.
Which technique predicts future values from historical data?
- Regression
- Classification
- Time Series Forecasting
- Clustering
Answer: Time Series Forecasting
Time series forecasting models temporal patterns.
Querying data in object storage is called ________ Query.
Answer: Serverless
Serverless query engines query object storage directly.
Data lakes store raw data; warehouses store structured data.
- True
- False
Answer: True
Data lakes store raw data; warehouses store curated data.
Which format is optimized for analytics?
- CSV
- JSON
- Parquet
- Avro
Answer: Parquet
Parquet is columnar storage optimized for analytics.
Processing data as it arrives is called ________ Processing.
Answer: Stream
Stream processing handles continuous data flows.
CDC enables real-time replication of database changes.
- True
- False
Answer: True
CDC streams database operations for real-time analytics.