Module-level Graded Quiz: Convolutional Neural Networks :Deep Learning with PyTorch (IBM AI Engineering Professional Certificate) Answers 2025

1. Question 1 — Purpose of convolution

❌ To increase channels
✅ To detect local patterns in the input image
❌ To apply activation
❌ To reduce image size

2. Question 2 — Zero padding effect

❌ Decreases size
✅ Increases the size of the activation map
❌ Doubles size
❌ No effect

3. Question 3 — Max pooling

❌ Adds non-linearity
❌ Enhances contrast
✅ Reduces spatial dimensions
❌ Increases channels

4. Question 4 — Activation setting negatives to 0

✅ ReLU
❌ Tanh
❌ Sigmoid
❌ Softmax

5. Question 5 — Activation with multiple channels

❌ Only last channel
❌ Only first channel
✅ Applied individually to each element in every channel
❌ Applied to sum of channels

6. Question 6 — Purpose of flattening

❌ Reduce channels
❌ Increase spatial dimensions
✅ Convert 2D feature map into 1D vector
❌ Apply pooling

7. Question 7 — Output channels

❌ Image width
✅ Number of feature maps
❌ Number of input images
❌ Image height

8. Question 8 — Benefit of pre-trained models

❌ Auto fine-tuning
❌ No dataset needed
✅ Provides a strong starting point
❌ Optimized only for speed

9. Question 9 — Why requires_grad=False

❌ Save memory
❌ Speed up forward pass
❌ Auto LR adjust
✅ To prevent modifying pre-trained weights

10. Question 10 — Why GPU

❌ Simplify code
❌ Improve visualization
❌ Reduce model size
✅ Accelerate matrix operations

🧾 Summary Table

Q#	Correct Answer	Key Concept
1	Detect local patterns	Convolutions learn features
2	Increases size	Zero padding preserves spatial dims
3	Reduce spatial dimensions	Pooling downsamples
4	ReLU	Removes negatives
5	Applied per-element per-channel	Channel-wise activation
6	Convert to 1D	Required for fully connected layers
7	Feature maps	Output channels = number of filters
8	Strong starting point	Transfer learning
9	Prevent weight updates	Freeze layers
10	Accelerate matrix ops	GPU parallel computing