Quiz Space

Deep Learning for Computer Vision · End Term · 28 Apr 2024 · January 2024 term

Question 33: Given word embeddings for two words and weight matrix to…

Question 33

+0.7 marksNumerical answer

Given word embeddings for two words and weight matrix to generate queries, keys and values as below (in practice, these would be more complex or learned; for simplicity, we provide them here):

Calculate the query, key, and value to calculate z1. Consider only the first word for calculating the scores; remember that attention output is calculated by a weighted sum of all value vectors. Also, the key vector dimension dk = 3
Based on the above data, answer the given subquestions.

Given word embeddings for two words and weight matrix to generate queries, keys and values as below (in practice, these would be more complex or learned; for simplicity, we provide them here):

Second element of attention vector is _____________

Show answer

Correct answer: 1 (accepted within ±0.02)

See the whole paperTake it as a mock test

Question 33 of 44 in the IIT Madras BS Deep Learning for Computer Vision (Deep Learning for Computer Vision) End Term paper sat on 28 Apr 2024, in the January 2024 term (IIT M DEGREE AN EXAM QDB3 28 Apr 2024). It carries 0.7 marks.

More questions from this paper

  1. Q1Given word embeddings for two words and weight matrix to generate queries, keys and values as below (in practice, these…
  2. Q2What is the role of the Gaussian blur in the Canny edge detection algorithm, and how does the choice of the standard de…
  3. Q3Which property of the SIFT descriptor is primarily responsible for its resilience to changes in scale, viewpoint, and i…
  4. Q4In the Maximally Stable External Regions (MSER) algorithm, how does the tree structure contribute to the efficiency of …
  5. Q5In which scenarios is the Random Sample Consensus (RANSAC) algorithm more suitable compared to least squares?
  6. Q6In what way does VLAD (Vector of Locally Aggregated Descriptors) differ from Bag of Words (BoW) in representing visual …
  7. Q7How does pyramid matching differ from Speeded-Up Robust Features (SURF) in image matching?
  8. Q8In the context of Convolutional Neural Networks (CNNs), what is the primary challenge associated with backpropagation c…
  9. Q9What is the primary advantage of Depthwise Separable Convolutions compared to standard Convolutional Neural Network (CN…
  10. Q10Which of the following strategies can help mitigate the risk of overfitting when fine-tuning pretrained models on a sma…
  11. Q11What is the primary advantage of visualizing patches of maximally activating neurons in neural network interpretation?
  12. Q12Which layer of the neural network does Grad-CAM typically operate on?
  13. Q13Which component of Grad-CAM is responsible for weighting the importance of each feature map?
  14. Q14What is the primary role of the Gram matrix in neural style transfer?
  15. Q15What is a potential limitation of non-maximum suppression (NMS) algorithms in object detection?
  16. Q16What is a key feature of focal loss that distinguishes it from traditional loss functions in object detection?
  17. Q17What is a unique challenge faced by Siamese networks in comparison to traditional neural networks?
  18. Q18Consider the following assertion and reason, and state which of them is true.\ Assertion: ProgressiveGAN uses Minibatch…
  19. Q19Figure question
  20. Q20Which of the following statements is false?
  21. Q21Match the following: | 1)VAE | i) z = \mu + \sigma \odot \epsilon;\ \epsilon \sim \mathcal{N}(0, 1) | |---|---| | 2)GAN…
  22. Q22Figure question
  23. Q23Figure question
  24. Q24What is the advantage of the progressive training methodology proposed in the Progressive GAN paper?
  25. Q25Contrastive Language Image Pre-training (CLIP) Can be used as a zero-shot classifier.
  26. Q26Figure question
  27. Q27The cornerness score of the center pixel for the image represented by the matrix below is . | 1 | 0 | 0 | 0 | 1 | |---|…
  28. Q28Consider a 3 \times 3 grayscale image in terms of pixel intensities as shown in matrix form below. | 9 | 1 | 6 | |---|-…
  29. Q29Figure question
  30. Q30Consider a Faster R-CNN object detection framework, which is being trained with a mini-batch size of 2. The RPN produce…
  31. Q31Transformer encoder requires a sequence of as input.
  32. Q32Given word embeddings for two words and weight matrix to generate queries, keys and values as below (in practice, these…
  33. Q34Given word embeddings for two words and weight matrix to generate queries, keys and values as below (in practice, these…
  34. Q35NOTE: Enter your answer in two decimal places.
  35. Q36NOTE: Enter your answer in two decimal places.
  36. Q37NOTE: Enter your answer in two decimal places.
  37. Q38Based on the above data, answer the given subquestions.
  38. Q39Based on the above data, answer the given subquestions.
  39. Q40Based on the above data, answer the given subquestions.
  40. Q41Enter the correct answer for blank(a):
  41. Q42Enter the correct answer for blank(b):
  42. Q43Enter the correct answer for blank(a):
  43. Q44Enter the correct answer for blank(b):