Module 11 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. Discarding blurry images during collection…
- A. improves the dataset with no downside
- B. removes the rushed, poorly-lit conditions where the target may be most likely
- C. is required for training
- D. has no effect on bias
2. Per-group accuracy should be reported with…
- A. confidence scores
- B. row counts
- C. training time
- D. the model version only
3. Faces incidentally present in training images…
- A. make the model a face recogniser
- B. do not create consent for identifying people
- C. must be labelled
- D. improve generalisation
4. The first field a reviewer reads in a model card is…
- A. the architecture
- B. the baseline the model is measured against
- C. the training time
- D. the author
5. A permitted-use list should have…
- A. two categories: allowed and prohibited
- B. three, including 'not yet' with a stated condition
- C. one category
- D. no categories, just examples
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - removes the rushed, poorly-lit conditions where the target may be most likely
It sounds like quality control and silently narrows what the model generalises to.
2. B - row counts
The worst group is usually the smallest, which is exactly why the gap hides in the overall figure.
3. B - do not create consent for identifying people
The prohibition comes from the collection agreement, not from what the weights can do.
4. B - the baseline the model is measured against
91% against a 94% majority baseline is a very different claim from 91% against 50%.
5. B - three, including 'not yet' with a stated condition
Most real requests are reasonable uses that have never been evaluated, and the list needs somewhere to put them.
Practical Check
Apply this module to your own work: complete the module activity for *Bias, Privacy and Model Cards for Vision*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.