Skip to main content

Knowledge check

Quiz
Complete

Evaluate your understanding of this unit​

Questions cover lineage, metadata, bias in collection and preprocessing, and governance themes from Unit 2.

Target illustration

Action item: Knowledge check​

Use the feedback blocks to self-check your answers.

Knowledge check: 8 questions
1. What is the primary purpose of data lineage in ML operations?
  • A. To delete old data automatically
  • B. To trace origin, transformations and usage of data for debugging, compliance and trust
  • C. To replace metadata catalogues
  • D. To increase model complexity
Correct Answer: B

Feedback: Lineage connects data movement to accountability and reproducibility.

2. Data provenance most closely refers to:
  • A. Encrypting data at rest
  • B. Choosing a learning rate schedule
  • C. The origin and history of data, including changes over time
  • D. Visualising neural network layers
Correct Answer: C

Feedback: Provenance answers where data came from and how it was modified.

3. Which metadata type describes schema, keys and relationships between tables?
  • A. Structural metadata
  • B. Usage metadata only
  • C. Provenance metadata only
  • D. Network packet metadata
Correct Answer: A

Feedback: Structural metadata captures how data is organised and linked.

4. The FAIR principles emphasise datasets that are:
  • A. Fast, Accurate, Iterative, Repeatable
  • B. Fully anonymous by default
  • C. Free of charge only
  • D. Findable, Accessible, Interoperable, Reusable
Correct Answer: D

Feedback: FAIR guides metadata and publishing practices for reusable research and enterprise data.

5. Sampling bias occurs when:
  • A. The model uses too many layers
  • B. Training data does not represent the population the model will serve
  • C. Hyperparameters are tuned on the test set
  • D. GPUs are underutilised
Correct Answer: B

Feedback: Non-representative sampling skews learning toward some groups or scenarios.

6. Measurement bias often arises from:
  • A. Using too little regularisation
  • B. Choosing the wrong cloud region
  • C. Inconsistent collection or labelling that distorts recorded values
  • D. Long retention periods alone
Correct Answer: C

Feedback: Uneven measurement across groups or time warps what the model learns.

7. Impact analysis powered by lineage helps teams:
  • A. Predict how upstream data or feature changes affect model behaviour before harm spreads
  • B. Remove the need for testing
  • C. Avoid documentation
  • D. Guarantee models are unbiased
Correct Answer: A

Feedback: Lineage supports proactive checks when sources or transforms change.

8. A central metadata repository primarily helps organisations:
  • A. Store raw images only
  • B. Replace legal review
  • C. Eliminate the need for encryption
  • D. Discover, govern and audit datasets with consistent documentation
Correct Answer: D

Feedback: Catalogues and repositories operationalise metadata at enterprise scale.