Login
Sign Up

Deep Learning – IIT Ropar Week 3 NPTEL Assignment Answers 2025

By /

Need help with this week’s assignment? Get detailed and trusted solutions for Deep Learning – IIT Ropar Week 3 NPTEL Assignment Answers. Our expert-curated answers help you solve your assignments faster while deepening your conceptual clarity.

✅ Subject: Deep Learning – IIT Ropar
📅 Week: 3
🎯 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 3 NPTEL Assignment Answers

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

NPTEL Deep Learning – IIT Ropar Week 3 Assignment Answers 2025

1. What is the correct dimension of W2 in this setup?

  • 4×3
  • 3×4
  • 3×3
  • 3×1
Answer : See Answers

2. Which of the following loss–activation combinations are correctly used in AgroScan?

  • Softmax activation and squared error loss
  • Sigmoid activation and cross-entropy loss
  • Softmax activation and cross-entropy loss
  • Linear activation and cross-entropy loss
Answer :

3. Compute the pre-activation values for the hidden layer for the input x=[0.2,0.6,0.1,0.5] with

  • [0.9,0.7,0.8]
  • [0.8,0.9,0.7]
  • [0.8,0.7,0.9]
  • [0.7,0.9,0.8]
Answer :

4. In the AgroScan model, after performing the forward pass for a given input sample, the pre-activation values at the output layer (corresponding to the three classes: Healthy, Pest-infected, and Nutrient-deficient) are: 𝑧 = [0.96, −0.27, 1.19]
Use the softmax function to convert these into probabilities.
Based on the computed probabilities, what class will the network predict for this input?

  • Healthy
  • Pest-infected
  • Nutrient-deficient
  • All classes have equal probability
Answer :

5. In a classification task, the true label is “Pest-infected “, and the predicted probability for class “Pest-infected” is 0.55. Using the cross-entropy loss function, compute the loss.
Fill in the blank with the answer rounded to two decimal places:.
Fill the blank: _____________

Answer :

6. Given softmax output [0.35, 0.55, 0.10], what is the predicted class in the AgroScan model?

  • Healthy
  • Pest-infected
  • Nutrient-deficient
  • None of the above
Answer :

7. How many learnable parameters are there in the above neural network in total?

  • 12
  • 15
  • 21
  • 24
Answer : See Answers

8. If the hidden layer used 𝑡𝑎𝑛ℎ instead of 𝑠𝑖𝑔𝑚𝑜𝑖𝑑, what would change?

  • Activation outputs could be negative
  • Output layer would become invalid
  • Pre-activations would change
  • Number of parameters would increase
Answer :

9. Which of the following can lead to higher cross-entropy loss during training?

  • Predicting low probability for the true class
  • Predicting uniform probabilities for all classes
  • Predicting 1.0 for the correct class
  • Predicting a class different from the true one
Answer :

10. Which of the following are always true for softmax output?

  • Output values lie in 0, 1
  • Output values sum to 1
  • Softmax is invariant to the order of inputs
  • Softmax is sensitive to relative magnitudes of logits
Answer :

Case study for the questions from 11 to 20

A research team is developing a neural network model named LeafYield, designed to predict the crop yield (in kg) from leaf-level features extracted via sensors and imaging.

Each sample contains 5 normalized numerical features: Leaf length, Leaf width, Color intensity, Water content, Light absorption level

The neural network contains two hidden layers. First hidden layer has 4 neurons with tanh activation. Second hidden layer has 3 neurons with sigmoid activation. Output layer has one neuron with linear activation.

11. Why is the output activation in LeafYield chosen to be linear?

  • Because softmax is unsuitable for regression
  • Because we need raw values, not probabilities
  • Because sigmoid would squash the output range
  • All of the above
Answer :

12.

  • 0.70
  • 0.71
  • 0.72
  • 0.73
Answer :

13. Which loss function is most appropriate for LeafYield?

  • Cross-entropy
  • Binary cross-entropy
  • Mean squared error
  • Kullback–Leibler divergence
Answer :

14. Given the number of neurons in the layers and that bias is used in all layers, compute the total number of learnable parameters.
Fill in the blank : ____________

Answer :

15. Given output layer activation is linear and final pre-activation value is 3.75, what will be the model’s prediction?
Fill in the blank : ______________

Answer : See Answers

16. Which of the following statements are true about the activation functions used in LeafYield?

  • tanh outputs values between -1 and 1
  • sigmoid outputs values between 0 and 1
  • linear activation has no bounds
  • sigmoid is more centered around zero than tanh
Answer :

17. Given the input vector x=[0.1,0.5,0.3,0.7,0.2]T And

Which is the pre-activation vector a[1] for hidden layer 1.

  • [0.6,1.2,0.5,1.5]
  • [0.6,1.2,0.3,1.5]
  • [0.6,1.3,0.5,1.5]
  • [0.6,1.3,0.3,1.5]
Answer :

18. Apply tanh activation to the above vector obtained in Question: 17 and round to two decimals.

  • [0.54,0.83,0.46,0.91]
  • [0.54,0.83,0.29,0.91]
  • [0.54,0.86,0.46,0.91]
  • [0.54,0.86,0.29,0.91]
Answer :

19. The LeafYield model was tested on 5 plant samples. The true and predicted crop yields (in kg) are:

What is the Mean Squared Error?

  • 0.12
  • 0.15
  • 0.25
  • 0.30
Answer :

20. Which layer’s weights will receive gradients first during backpropagation?

  • Output layer
  • Hidden layer 2
  • Hidden layer 1
  • Input layer
Answer :

Case study for Questions 21 to 29

A sports analytics company has developed PlayPredict, a neural network model that classifies football players into one of four roles based on physical and tactical attributes, such as Defender, Midfielder, Forward, Goalkeeper.

From video and tracking data, 5 numerical features are extracted: Speed, Pass accuracy, Defensive actions, Dribble attempts, Position heatmap score.

The model has one hidden layer with 4 neurons using the tanh activation function, followed by an output layer.

21. Which activation is best suited for the above scenario?

  • Linear activation
  • Sigmoid activation
  • Tanh activation
  • Softmax activation
Answer :

22. What is the predicted class if the output vector contains [0.2,0.1,0.6,0.1]?

  • Defender
  • Midfielder
  • Forward
  • Goalkeeper
Answer : See Answers

23. Given logits: [1.2,0.8,2.0,1.0], compute softmax output (rounded to 2 decimals)

  • [0.22,0.14,0.48,0.17]
  • [0.21,0.14,0.47,0.17]
  • [0.28,0.16,0.58,0.27]
  • [0.51,0.54,0.47,0.17]
Answer :

24. Why use tanh in the hidden layer?

  • Keeps outputs in [0,1]
  • Introduces non-linearity
  • Allows negative outputs
  • Helps output softmax
Answer :

25.

Compute the pre-activation vector a[1].

  • [0.9,0.9,0.8,0.7]
  • [0.9,0.8,1.1,0.6]
  • [0.9,0.9,1.1,0.7]
  • [1.0,0.9,1.1,0.8]
Answer :

26. Compute the hidden vector h[1] from the above computed a[1].

  • [0.72,0.72,0.66,0.60]
  • [0.72,0.66,0.80,0.54]
  • [0.72,0.72,0.80,0.60]
  • [0.76,0.72,0.80,0.66]
Answer :

27. The PlayPredict model produced the following softmax output for a football player sample:
y^=[0.1,0.7,0.1,0.1]
What is the categorical cross-entropy loss for this prediction (rounded to 3 decimals)?

  • 0.105
  • 0.357
  • 0.500
  • 0.845
Answer :

28. When bias is used in both the layers, what is the total number of learnable parameters in the playpreidct network?
Fill in the blank _________________

Answer :

29. Which of the following can be tanh activation output?

  • [–0.9, 0.2, 0.8, –0.4]
  • [1.5, –1.2, 2.3, 0.5]
  • [0.0, 1.0, 2.0, –1.0]
  • [–2, –1, 0, 1]
Answer : See Answers

Related Posts