Week 4 Quiz :Convolutional Neural Networks in TensorFlow (DeepLearning.AI TensorFlow Developer Professional Certificate) Answers 2025
1. Question 1
When using image augmentation with image_dataset_from_directory, what happens to your raw image data on disk?
-
❌ A copy will be made, and the copies are augmented
-
❌ A copy will be made, and the originals will be augmented
-
✅ Nothing
-
❌ The images will be edited on disk
Explanation: Augmentation happens in-memory only. No changes are made to files on disk.
2. Question 2
What layer converts pixel values from [0,255] → [0,1]?
-
❌ Conversion
-
❌ Translation
-
❌ Resize
-
✅ Rescaling
Explanation: tf.keras.layers.Rescaling(1./255) normalizes pixel values.
3. Question 3
Traditional programming: Rules + Data → ?
-
❌ Answers
-
❌ Binary
-
❌ Machine Learning
-
❌ Bugs
Correct Answer:
✅ Answers
Explanation: Traditional programming produces answers (outputs).
In ML, data + answers → model (rules).
4. Question 4
For multiple classes with categorical_crossentropy, what parameter is required?
-
❌ label_mode=’int’
-
❌ class_mode=’int’
-
✅ label_mode=’categorical’
-
❌ class_mode=’categorical’
Explanation: For one-hot labels (needed by categorical_crossentropy), use:
label_mode='categorical'
5. Question 5
Can you use image augmentation with transfer learning?
-
❌ No
-
✅ Yes, you can augment the input images while training the new layers
Explanation: Frozen layers stay untouched, but augmentation improves learning for new classifier layers.
6. Question 6
Impact of adding convolutions on top of a DNN?
-
❌ It will be slower
-
❌ It will be faster
-
❌ No impact
-
✅ It depends (might be faster or slower)
Explanation: Convolutions can reduce computation (via downsampling) or increase it (if poorly designed).
7. Question 7
What is a convolution?
-
❌ A technique to make images smaller
-
❌ A technique to make images larger
-
✅ A technique to extract features from an image
-
❌ A technique to remove unwanted images
Explanation: Convolutions detect edges, textures, shapes, patterns.
8. Question 8
Why does Fashion-MNIST have 10 output neurons?
-
❌ To train faster
-
❌ To classify faster
-
❌ Arbitrary
-
✅ The dataset has 10 classes
Explanation: Output neuron count = number of classes.
🧾 Summary Table
| Q# | Correct Answer | Key Concept |
|---|---|---|
| 1 | Nothing | Augmentation doesn’t modify disk files |
| 2 | Rescaling | Pixel normalization |
| 3 | Answers | Traditional programming output |
| 4 | label_mode=’categorical’ | One-hot encoding for categorical loss |
| 5 | Yes | Augmentation works with transfer learning |
| 6 | It depends | Conv layers affect efficiency variably |
| 7 | Extract features | Convolution definition |
| 8 | Dataset has 10 classes | Output layer = number of classes |