Quiz Space

Deep Learning for Computer Vision · Quiz 2 · 16 Aug 2026 · May 2026 term

Deep Learning for Computer Vision Quiz 2 16 Aug 2026 — Question 16

Question 16

+3 marksOne or more correct options

Which statements are true?

Select all that apply.

  1. A

    Soft attention is differentiable and can be trained with backpropagation.

  2. B

    Hard attention often needs sampling + REINFORCE (or similar) due to non- differentiability.

  3. C

    Hard attention always has higher test-time compute than soft attention.

  4. D

    Cross-attention can align decoder queries to encoder keys/values (e.g., words attending to image regions).

Show answer

Correct answers

  • A

    Soft attention is differentiable and can be trained with backpropagation.

  • B

    Hard attention often needs sampling + REINFORCE (or similar) due to non- differentiability.

  • D

    Cross-attention can align decoder queries to encoder keys/values (e.g., words attending to image regions).

See the whole paperTake it as a mock test

Question 16 of 22 in the IIT Madras BS Deep Learning for Computer Vision (Deep Learning for Computer Vision) Quiz 2 paper sat on 16 Aug 2026, in the May 2026 term (Deep Learning For Computer Vision 16 Aug 26). It carries 3 marks.

More questions from this paper

  1. Q1Why do very deep plain CNNs (without skip connections) sometimes show higher training error than shallower CNNs?
  2. Q2MobileNetV1 reduces computation primarily by using:
  3. Q3EfficientNet’s key scaling idea is:
  4. Q4Which technique most directly helps reduce catastrophic forgetting early in fine-tuning?
  5. Q5Which method produces a heatmap by weighting convolutional feature maps using gradients of a target class score?
  6. Q6Which detector is a classic two-stage detector with an RPN + ROI feature extraction?
  7. Q7RetinaNet is notable primarily because it introduced:
  8. Q8A model that predicts a class label for every pixel without separating object instances is:
  9. Q9Which statement is true?
  10. Q10In a standard ResNet-50 bottleneck block, the three convolutions are typically:
  11. Q11In Inception/GoogLeNet modules, the main purpose of 1×1 convolutions is to:
  12. Q12Which statement is most accurate?
  13. Q13Vanishing gradients in vanilla RNNs are largely caused by:
  14. Q14Which statements can be true in practice when fine-tuning with small batch sizes?
  15. Q15Which statements are true ?
  16. Q17Input feature map: 14×14×128, output: 14×14×256, kernel: 3×3, stride 1, same padding, ignore bias.\ Compute number of p…
  17. Q18Input feature map: 14×14×128, output: 14×14×256, kernel: 3×3, stride 1, same padding, ignore bias, compute total MACs (…
  18. Q19A ResNet bottleneck block takes 256 channels in and uses:\ 1×1 conv to 256 channels → 3×3 conv at 256 → 1×1 conv to 102…
  19. Q20A detector uses three pyramid levels: P3 = 80×80, P4 = 40×40, P5 = 20×20. At each location, it uses 9 anchors (3 scales…
  20. Q21Box A: top-left (2,2), bottom-right (10,12) Box B: top-left (6,5), bottom-right (14,15) Compute IoU rounded to 3 decima…
  21. Q22Let query q = [1,2]^T and keys:k1 = [1,0]^T, k2 = [0,1]^T, k3 = [1,1]^T. Scores si = q^T ki. Use softmax over scores. U…