Deep Learning – IIT Ropar Week 7 Assignment Answers

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

Deep Learning – IIT Ropar Week 7 Assignment Answers (Jan-Apr 2026)

1. What is the primary advantage of using an ensemble model for AQI prediction in this case?

  • It guarantees zero prediction error
  • It reduces model bias
  • It reduces variance by combining diverse models
  • It eliminates the need for training data
Answer : c

2. Despite Model B having the lowest individual MSE, why does the ensemble outperform it?

  • Model B is underfitting
  • Ensemble combines uncorrelated errors
  • Ensemble increases model complexity
  • Model B has the highest bias
Answer : b

3. What is the average prediction of the ensemble for the severe pollution day, using the mean prediction of each model?

  • 300
  • 305
  • 309
  • 320
Answer : c

4. Which of the following are valid reasons for deploying the ensemble AQI model in a Smart City context?

  • Improved robustness to unusual pollution patterns
  • Reduced sensitivity to noisy sensor readings
  • Guaranteed perfect AQI forecasts
  • Better generalization across seasons
Answer : a, b, d

5. Compared to individual models, the ensemble model is most likely to have:

  • Lower variance
  • Higher variance
  • Similar or slightly higher bias
  • Improved generalization
Answer : a, c, d

6. The large gap between training accuracy (98%) and validation accuracy (72%) before augmentation indicates:

  • High bias
  • High variance
  • Data leakage
  • Underfitting
Answer : b

7. Out of the original 1,200 images, how many images are used for training before augmentation?

  • 720
  • 840
  • 900
  • 1,200
Answer : b

8. If each training image generates three additional augmented versions, what is the effective size of the training dataset after augmentation?

  • 1,680
  • 2,520
  • 3,360
  • 4,200
Answer : c

9. Why does training accuracy slightly decrease after applying dataset augmentation?

  • The model capacity is reduced
  • Augmentation introduces label noise
  • The learning rate is too small
  • The task becomes harder due to increased variability
Answer : d

10. Which augmentation strategies used in this case are appropriate for crop leaf images?

  • Horizontal flipping
  • Rotation within ±15°
  • Vertical flipping
  • Random zoom
Answer : a, b, d

11. The initial model (without regularization) shows very high training accuracy but lower test accuracy. This behavior most strongly indicates:

  • Underfitting
  • High bias
  • Overfitting
  • Data leakage
Answer : c

12. What is the primary role of L2 regularization in this fraud detection model?

  • It removes irrelevant features
  • It shrinks model weights toward zero
  • It increases the learning rate
  • It increases model variance
Answer : b

13. Given
∇J=4,α=2,w=5
What is the new gradient after adding L2 regularization?

  • 6
  • 9
  • 14
  • 15
Answer : c

14. Using the updated gradient from Q13 and learning rate η=0.1, what is the updated weight value after one step?

  • 4.6
  • 4.1
  • 3.6
  • 3.0
Answer : c

15. Which of the following are true effects of applying L2 regularization in this case?

  • Reduction in overfitting
  • Slight increase in training error
  • Increase in model variance
  • Improvement in test performance
Answer : a, b, d