AIML Rounds

QUIZ

CNNs and vision

9 questions on CNNs and vision, from 9 lessons. Pick an answer in your head, then open it.

On the roadmap for: ML Engineering.

  1. Question 1 of 9

    Why does a 3×3 convolution suit images better than a dense layer?

    1. A. It ignores neighbouring pixels to avoid picking up noise
    2. B. It's always more accurate on every kind of data, not just images
    3. C. It only works on grayscale images with one channel
    4. D. It shares local weights across positions, so far fewer parameters
    View answer

    ✓ D. It shares local weights across positions, so far fewer parameters

    Locality and weight sharing match how images work: an edge looks the same anywhere.

    Lesson: Convolution and Receptive Field

  2. Question 2 of 9

    Input 32×32, kernel 3×3, padding 1, stride 2. What is the output's height and width?

    1. A. 32×32
    2. B. 30×30
    3. C. 16×16
    4. D. 15×15
    View answer

    ✓ C. 16×16

    Output = ⌊(32 − 3 + 2·1) / 2⌋ + 1 = ⌊31/2⌋ + 1 = 16.

    Lesson: Pooling, Stride, Padding and Output Shapes

  3. Question 3 of 9

    Before ResNet, why did much deeper plain networks get worse results, even on training data?

    1. A. Too little training data for such deep models to learn anything from
    2. B. GPUs ran out of memory for deeper models
    3. C. Harder optimisation: signals faded through many layers
    4. D. Overfitting the training data with too many layers
    View answer

    ✓ C. Harder optimisation: signals faded through many layers

    Higher training error showed it was an optimisation problem; skip connections fixed it.

    Lesson: ResNets and Skip Connections

  4. Question 4 of 9

    What idea did VGG add to CNN design?

    1. A. Skip connections around each block of layers
    2. B. Attention layers between the convolution blocks
    3. C. Stacking small 3×3 convolutions for a big receptive field
    4. D. Depthwise separable convolutions to save compute
    View answer

    ✓ C. Stacking small 3×3 convolutions for a big receptive field

    Two 3×3 layers see a 5×5 area with fewer weights and an extra non-linearity.

    Lesson: CNN Architectures: AlexNet to ConvNeXt

  5. Question 5 of 9

    You have only a few hundred labelled images. What's the usual approach?

    1. A. Start pretrained, freeze most layers, fine-tune the head
    2. B. Train a CNN from scratch on the few images
    3. C. Collect no more data and rely only on data augmentation
    4. D. Use a random forest on the raw pixels
    View answer

    ✓ A. Start pretrained, freeze most layers, fine-tune the head

    Pretrained features transfer well, so small datasets can still reach good accuracy.

    Lesson: Transfer Learning and Fine-Tuning

  6. Question 6 of 9

    How does a Vision Transformer (ViT) process an image?

    1. A. It converts the image into a text caption first
    2. B. It reads the image pixel by pixel with an RNN
    3. C. It splits it into patches and runs a Transformer on them
    4. D. It uses only convolutions, with no attention layers
    View answer

    ✓ C. It splits it into patches and runs a Transformer on them

    With less built-in image bias than a CNN, ViTs usually need more data or pretraining.

    Lesson: Vision Transformers

  7. Question 7 of 9

    How is an object detector usually evaluated?

    1. A. mAP, counting boxes right above an IoU threshold
    2. B. RMSE between predicted and true box sizes
    3. C. The number of boxes it finds in each image
    4. D. Accuracy of the predicted class for each box
    View answer

    ✓ A. mAP, counting boxes right above an IoU threshold

    IoU decides whether a box matches; mAP summarises precision and recall across classes.

    Lesson: Object Detection

  8. Question 8 of 9

    Which architecture is a classic for image segmentation?

    1. A. U-Net, with skip connections for fine detail
    2. B. Word2Vec, trained on image captions
    3. C. AlexNet, the first deep network to win ImageNet
    4. D. A plain multi-layer perceptron on pixels
    View answer

    ✓ A. U-Net, with skip connections for fine detail

    The decoder upsamples to full resolution, and the skips bring back the edges lost by downsampling.

    Lesson: Image Segmentation

  9. Question 9 of 9

    How can a tiny, invisible change to an image fool a classifier?

    1. A. It removes colour information from the image
    2. B. It changes the image size that the model sees
    3. C. Tiny shifts towards higher loss add up across pixels
    4. D. It can't: classifiers simply ignore tiny changes
    View answer

    ✓ C. Tiny shifts towards higher loss add up across pixels

    Methods like FGSM use the gradient to find these adversarial examples.

    Lesson: Interpreting Networks and Adversarial Examples