Deep Learning – IIT Ropar Week 4 NPTEL Assignment Answers 2025

Need help with this week’s assignment? Get detailed and trusted solutions for Deep Learning – IIT Ropar Week 4 NPTEL Assignment Answers. Our expert-curated answers help you solve your assignments faster while deepening your conceptual clarity.

✅ Subject: Deep Learning – IIT Ropar
📅 Week: 4
🎯 Session: NPTEL 2025 July-October
🔗 Course Link: Click Here
🔍 Reliability: Verified and expert-reviewed answers
📌 Trusted By: 5000+ Students

For complete and in-depth solutions to all weekly assignments, check out 👉 NPTEL Deep Learning – IIT Ropar Week 4 NPTEL Assignment Answers

🚀 Stay ahead in your NPTEL journey with fresh, updated solutions every week!

NPTEL Deep Learning – IIT Ropar Week 4 Assignment Answers 2025

1. You are training a neural network on a dataset with 5 million samples using Mini-Batch Gradient Descent. The mini-batch size is 500, and each parameter update takes 100 milliseconds. How many seconds will it take to complete 5 epochs of training?

• 5,000
• 10,000
• 2,500
• 50,000

Answer : See Answers

2. You are comparing training times using different gradient descent algorithms on a dataset with 1,000,000 data points. Each parameter update takes 2 milliseconds. How many milliseconds longer will Stochastic Gradient Descent take compared to Vanilla (Batch) Gradient Descent to complete 2 epochs?

• 4,000,004 ms
• 4,000,000 ms
• 3,999,994 ms
• 3,999,996 ms

Answer :

3. What is the most practical benefit of using smaller batch sizes on constrained devices?

• Reduces computation time significantly
• Increases training accuracy
• Minimizes memory usage and reduces overhead
• Allows larger models to be trained

Answer :

4. You reduce the batch size from 4,000 to 1,000. What happens to the number of weight updates per epoch?

• Doubles
• Quadruples
• Remains constant
• Halves

Answer :

5. Which of the following statements are true about mini-batch gradient descent?

• It offers a compromise between computation and accuracy
• It prevents gradient vanishing completely
• It allows parallelism in training
• It can still be prone to overfitting

Answer :

6. What could be the reason for slow learning in this scenario?

• Large Learning rate
• Very small gradients
• Very high momentum
• Incorrect label noise

Answer :

7. What optimizer would help improve learning in this situation?

• Vanilla Gradient Descent
• Mini-Batch SGD
• Momentum based Gradient Descent
• Adam with bias correction and adaptive learning rate

Answer :

8. If the above model takes 300 steps per epoch, then after 5 epochs the number of weight updates is ______________
Fill in the blank: _________

Answer : See Answers

9. Which of the following helps in handling small gradients?

• Reducing learning rate
• Using Adagrad
• Using adaptive optimizers like Adam
• Using large batch sizes

Answer :

10. What are the advantages of using momentum?

• Faster convergence
• Larger steps in the wrong direction
• Helps escape shallow minima
• Avoids oscillation in steep slopes

Answer :

11. Which of the following would not help in this scenario?

• Switch to adaptive gradient descent
• Add momentum
• Reduce learning rate
• Normalize input data

Answer :

12. If the learning rate η=0.01,momentum coefficient 𝛾 = 0.9, the current gradient at step 𝑡 is ∇w_t=0.2, and the previous update was 0.1, then what is the value of the new update?

• 0.11
• 0.092
• 0.091
• 0.12

Answer :

13. A data scientist is analyzing how momentum affects learning speed over time. She initializes a weight w0=1.0, and uses the following parameters in momentum-based gradient descent:
Momentum coefficient 𝛾 = 0.8, Learning rate η=0.05 and Initial update update0=0 She receives gradients over 3 consecutive iterations: ∇w₁=−0.5, ∇w₂=−0.2, and ∇w₃=−0.3
What is the value of the update at time 𝑡 = 3?

• 0.0172
• −0.0172
• 0.0216
• −0.009

Answer :

14. What are the benefits of using mini-batch over full batch.

• Less memory usage
• More frequent weight updates
• Higher computational cost
• Better generalization

Answer :

15. What is a likely cause of oscillations?

• Too low learning rate
• Batch size too small
• Too high learning rate
• No dropout

Answer : See Answers

16. Which technique helps reduce oscillations?

• Momentum
• Adagrad
• Weight decay
• None of the above

Answer :

17. Which optimizer allows you to peek ahead before computing the gradient?

• Adam
• Vanilla SGD
• Nesterov Accelerated Gradient
• Adagrad

Answer :

18. What happens if momentum is set to 1

• Model stops updating
• Model overshoots and diverges
• Model converges quickly
• No effect

Answer :

19. Which of the following are the advantages of mini-batch gradient descent over SGD:

• Reduces variance of updates
• Requires fewer epochs
• Faster convergence
• More computation per update

Answer :

20. What does the line search algorithm aim to optimize at every step of training?

• Batch size
• The cost function value along the gradient direction
• Momentum term
• Validation accuracy

Answer :

21. What is the key computational disadvantage of applying line search in every update?

• May overfit the data
• Many more computations in each step.
• Doesn’t converge
• Reduces gradient magnitude

Answer : See Answers

22. Which of the following schedules typically require setting two hyperparameters?

• Exponential decay
• 1/t decay
• Constant learning rate
• Step decay

Answer :

23. Exponential decay adjusts learning rate using which formula?

Answer :

24. Learning rate decay is typically used to:

• Fine-tune the model toward the end of training
• Avoid oscillation near minima
• Eliminate the need for momentum
• Control the impact of noisy gradients

Answer :

25. In step decay, the learning rate changes at _______________ intervals.

• Predefined
• One
• Random
• None of the above

Answer :

26. If you have 100,000 samples and batch size is 10,000, how many parameter updates happen in one epoch?

• 10
• 100
• 1000
• 1

Answer :

27. If N = 60,000 and batch size B = 5,000, the number of weight updates per epoch = ______________.

Answer :

28. Suppose you’re using Nesterov Accelerated Gradient and are at time step 𝑡. The current gradient at the look-ahead position is ∇w_look=0.3, the previous velocity (update) is update_t−1=0.2, and the hyperparameters are: 𝛾 = 0.8, 𝜂 = 0.05 .
What is the value of the current update updatet?

• 0.175
• 0.195
• 0.18
• 0.31

Answer :

29. You’re optimizing a neural network with NAG. At iteration 𝑡, you have: Current weight: w_t=1.0, Previous update: update_t−1=0.25,γ=0.9,η=0.01, Gradient at look-ahead position: ∇w_look=−0.5.

• 0.78
• 0.775
• 0.79
• 0.77

Answer : See Answers