Deep Learning – IIT Ropar Week 9 Assignment Answers

1 day ago
Sanket Kumar
3 minutes
Deep Learning - IIT Ropar

Deep Learning – IIT Ropar Week 9 Assignment Answers (Jan-Apr 2026)

1. What is the length of the one-hot vector used to represent each word in this system?

  • 3
  • 8,000
  • 10,000
  • Depends on document length
Answer : b

2. Why are one-hot encoded vectors considered highly sparse?

  • They contain negative values
  • They have repeated values
  • Only one element is non-zero
  • They are normalized vectors
Answer : c

3. How many non-zero entries are present in the one-hot representation of Document A?

  • 3
  • 1
  • 8000
  • Depends on vocabulary
Answer : a

4. Which of the following are limitations of one-hot encoding highlighted by this case study?

  • High dimensionality
  • No notion of semantic similarity
  • Large memory requirement
  • Dense representations
Answer : a, b, c

5. Why does cosine similarity work reasonably well for document comparison in this setup?

  • It is independent of document length
  • It measures overlap in word usage
  • It captures semantic meaning of words
  • Works only for dense vectors
Answer : a, b

6. Why does the team choose a context window size of 1?

  • To reduce vocabulary size
  • To capture only immediate word relationships
  • To remove stopwords automatically
  • To ensure dense representations
Answer : b

7. In the sentence
government announces new education policy,
how many context contributions does the word education make when using an a context window of 1?

  • 0
  • 1
  • 2
  • 3
Answer : c

8. Across the entire corpus, how many times does the word policy appear adjacent to the word education?

  • 1
  • 2
  • 3
  • 4
Answer : b

9. Which of the following statements about co-occurrence matrices are true in this case?

  • Matrix size depends on vocabulary size
  • Larger context windows capture broader relationships
  • Co-occurrence matrices are always dense
  • Word order within the context window is ignored
Answer : a, b, d

10. If the vocabulary size is 6, the total number of entries in the co-occurrence matrix is __________.

Answer : 36

11. Why is truncated SVD applied to the co-occurrence matrix in this system?

  • To make the matrix diagonal
  • To remove stopwords automatically
  • To reduce dimensionality while preserving structure
  • To increase sparsity
Answer : c

12. What property of truncated SVD enables semantically related words to cluster together?

  • Orthogonality of one-hot vectors
  • Preservation of exact co-occurrence counts
  • Capture of latent semantic structure
  • Removal of rare words
Answer : c

13. What is the size of the reduced word representation matrix Uk∑k ?

  • 8000 × 8000
  • 300 × 300
  • 8000 × 300
  • 300 × 8000
Answer : c

14. Why is truncated SVD suitable for the public feedback system?

  • It requires labeled training data
  • It can suppress less important components by keeping only top-k factors
  • It enables efficient similarity computation
  • It improves scalability of text processing
Answer : b, c, d

15. After truncated SVD with k = 300, each word is represented by a vector of length __________.

Answer : 300