Module 05 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. Edge-case rulings should be written…
- A. after the first labelling pass
- B. before any labelling starts
- C. only if labellers disagree
- D. by the model
2. On a dataset that is 94% one class, always predicting the majority…
- A. scores 50%
- B. scores 94% accuracy and finds none of the minority class
- C. raises an error
- D. is a reasonable baseline to deploy
3. Three photographs of the same carton should…
- A. be split randomly
- B. all be on the same side of the split
- C. be treated as three independent samples
- D. be deduplicated to one
4. A horizontal flip is unsafe for…
- A. cat vs dog
- B. reading a serial number
- C. brightness robustness
- D. aerial imagery
5. An inter-labeller agreement of 67% means…
- A. labels are good enough
- B. a hard ceiling on model performance that no architecture fixes
- C. the model will reach 67%
- D. one labeller should be replaced
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - before any labelling starts
Whatever the schema leaves undecided, labellers decide differently, and it becomes label noise.
2. B - scores 94% accuracy and finds none of the minority class
Any accuracy figure on this data has to be read against that 94%.
3. B - all be on the same side of the split
Split by the underlying object; multiple views on both sides measure memorisation.
4. B - reading a serial number
A flipped 6 is a 9 and a flipped left shoe is a right shoe - the label is now wrong and nothing notices.
5. B - a hard ceiling on model performance that no architecture fixes
It takes an afternoon to measure, and it explains a disappointing model score better than any modelling hypothesis.
Practical Check
Apply this module to your own work: complete the module activity for *Building Image Datasets and Labels*, 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.