Module 06 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. What does flattening an image before a dense layer discard?
- A. Colour information
- B. Which pixels are adjacent to which
- C. The batch dimension
- D. Pixel intensity
2. A 32x32 input through Conv2d(kernel_size=3, padding=1, stride=1) produces what size?
- A. 30x30
- B. 32x32
- C. 16x16
- D. 34x34
3. How does a convolution's parameter count change with image size?
- A. It grows linearly
- B. It grows quadratically
- C. It does not change
- D. It shrinks
4. Which split should augmentation be applied to?
- A. All splits
- B. Training only
- C. Validation only
- D. Test only
5. Accuracy is 0.85 on a three-class problem where one class has recall 0.00. What went wrong with the reporting?
- A. Nothing — 0.85 is good
- B. Accuracy is dominated by the majority class and hid a class the model never predicts
- C. The confusion matrix was computed wrongly
- D. The model needs more epochs
---
Answer Key and Explanations
Check these only after attempting every question.
1. B — Which pixels are adjacent to which
A dense layer treats every position as unrelated, so a pattern learned in one corner tells it nothing about the same pattern elsewhere.
2. B — 32x32
out = floor((32 + 2*1 - 3)/1) + 1 = 32. Kernel 3 with padding 1 is the standard size-preserving choice.
3. C — It does not change
Parameters depend on channels and kernel size only. The same filter slides over an image of any size.
4. B — Training only
Augmenting validation makes the score depend on which random transformations were drawn, so two evaluations of the same model disagree.
5. B — Accuracy is dominated by the majority class and hid a class the model never predicts
On imbalanced data, per-class recall is the minimum reporting standard. Accuracy summarises the majority class.
Practical Check
Apply this module to your own work: complete the module activity for *Convolutional Neural Networks for Images*, 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.