Daily Coding Quiz: Saturday, 7 November 2026

All 10 coding questions with answers and explanations. 0 players took part.

1. What does a convolutional layer mainly learn?

   Computer Vision · easy

   1. Sorting rules
   2. Small filters that detect local patterns like edges and textures
   3. The image size
   4. Colour histograms only

   Answer: B. Small filters that detect local patterns like edges and textures

   Explanation: Each filter slides over the image computing dot products. Early layers learn edges; deeper layers combine them into parts and objects.

2. What does max pooling with a 2×2 window and stride 2 do to a 32 × 32 feature map?

   Computer Vision · easy

   1. Keeps it 32 × 32
   2. Halves it to 16 × 16, keeping the largest value in each block
   3. Doubles it to 64 × 64
   4. Turns it into a single number

   Answer: B. Halves it to 16 × 16, keeping the largest value in each block

   Explanation: Each non-overlapping 2×2 block becomes one value, the maximum, so height and width are halved.

3. Which algorithm finds the shortest path from one source in a graph with non-negative edge weights?

   DSA · medium

   1. Kruskal’s
   2. Dijkstra’s
   3. Prim’s
   4. Topological sort

   Answer: B. Dijkstra’s

   Explanation: Dijkstra’s algorithm greedily settles the closest unvisited node using a priority queue. Kruskal’s and Prim’s build minimum spanning trees.

4. What is the logical order in which SQL evaluates these clauses?

   SQL · medium

   1. SELECT → FROM → WHERE → GROUP BY
   2. FROM → WHERE → GROUP BY → HAVING → SELECT → ORDER BY
   3. FROM → SELECT → WHERE → ORDER BY
   4. WHERE → FROM → GROUP BY → SELECT

   Answer: B. FROM → WHERE → GROUP BY → HAVING → SELECT → ORDER BY

   Explanation: That is why a column alias defined in SELECT usually can’t be used in WHERE, but can be used in ORDER BY.

5. What does Non-Maximum Suppression (NMS) do in object detection?

   Computer Vision · medium

   1. Removes low-resolution images
   2. Removes duplicate overlapping boxes for the same object, keeping the highest-scoring one
   3. Normalises pixel values
   4. Increases the number of anchors

   Answer: B. Removes duplicate overlapping boxes for the same object, keeping the highest-scoring one

   Explanation: Detectors propose many overlapping boxes. NMS keeps the top box and discards others whose IoU with it exceeds a threshold (often 0.5).

6. How many learnable parameters does a 3 × 3 convolution with 64 input channels and 128 output channels (with bias) have?

   Computer Vision · medium

   1. 73,856
   2. 73,728
   3. 1,152
   4. 8,192

   Answer: A. 73,856

   Explanation: Weights: 3 × 3 × 64 × 128 = 73,728, plus 128 biases = 73,856.

7. Depthwise separable convolution (as in MobileNet) reduces compute by:

   Computer Vision · hard

   1. Using smaller images
   2. Splitting a standard convolution into a per-channel spatial convolution followed by a 1 × 1 pointwise convolution
   3. Removing biases
   4. Quantising weights to 8 bits

   Answer: B. Splitting a standard convolution into a per-channel spatial convolution followed by a 1 × 1 pointwise convolution

   Explanation: The cost drops roughly by a factor of 1/N + 1/K² (N output channels, K kernel size): about 8–9× fewer operations for 3 × 3 kernels.

8. What is the receptive field of one output unit after two stacked 3 × 3 convolutions (stride 1)?

   Computer Vision · hard

   1. 3 × 3
   2. 5 × 5
   3. 6 × 6
   4. 9 × 9

   Answer: B. 5 × 5

   Explanation: Each extra 3 × 3 layer adds 2. Two of them see 5 × 5 with fewer parameters (2 × 9 = 18 vs 25 per channel pair) and an extra non-linearity: the VGG insight.

9. What does mAP@0.5:0.95 measure in COCO-style detection evaluation?

   Computer Vision · hard

   1. Accuracy at a single IoU of 0.5
   2. Mean average precision averaged over IoU thresholds 0.50, 0.55, …, 0.95 and over all classes
   3. Mean pixel accuracy
   4. Frames per second

   Answer: B. Mean average precision averaged over IoU thresholds 0.50, 0.55, …, 0.95 and over all classes

   Explanation: Averaging over strict IoU thresholds rewards precise localisation, not just finding the object. mAP@0.5 (the old PASCAL VOC metric) is more lenient.

10. Why do 1 × 1 convolutions appear so often in modern CNNs (Inception, ResNet bottlenecks, MobileNet)?

    Computer Vision · hard

    1. They enlarge the receptive field
    2. They mix channels cheaply, changing the channel count (e.g. to reduce it before an expensive 3 × 3)
    3. They replace pooling
    4. They act only as an activation function

    Answer: B. They mix channels cheaply, changing the channel count (e.g. to reduce it before an expensive 3 × 3)

    Explanation: A 1 × 1 conv is a per-pixel linear layer across channels. Bottlenecks shrink channels, apply the 3 × 3, then expand, cutting compute heavily.

Play the next Daily Quiz at 4 PM IST