SRI KRISHNA INSTITUTE OF TECHNOLOGY

(Accredited by NAAC, Approved by A.I.C.T.E. New Delhi, Recognised by Govt. of Karnataka & Affiliated to V.T U., Belagavi)
#29, Chimney Hills, Hesaraghatta Main Road, Chikkabanavara Post, Bengaluru- 560090

Dept. of Information Science & Engineering

Deep Learning (21CS743)

Question Bank

Module 2 - Regularization

QUESTIONS COS BT
1. What is regularization? List and explain the different strategies to
CO2 L2
create regularized model.
2. What is L2 Normalization? Explain its weight decay strategy and CO2 L2
also show its effects on the optimal value of w.
3. What is L1 regularization? Explain the difference between L1 and CO2 L2
L2 normal forms.
4. What is Hessian Matrix? Explain. CO2 L2

5. Explain parameter norm penalties. CO2 L2

6. Explain Dataset augmentation. CO2 L2
7. Explain the following: CO2 L2
a. Multi Task Learning
b. Semi-Supervised Learning
c. Noise Robustness
d. Early Stopping
e. Parameter Typing and Parameter Sharing
f. Sparse Representation
g. Bagging
h. Adversarial Training

8. What is dropout? What is its significance in regularization? Explain CO2 L2

with an example.

Ragini Krishna, Dept. ISE, SKIT. Page 1

 SRI KRISHNA INSTITUTE OF TECHNOLOGY
(Accredited by NAAC, Approved by A.I.C.T.E. New Delhi, Recognised by Govt. of Karnataka & Affiliated to V.T U., Belagavi)
#29, Chimney Hills, Hesaraghatta Main Road, Chikkabanavara Post, Bengaluru- 560090

Dept. of Information Science & Engineering

Deep Learning (21CS743)

Question Bank

Module 3 - Optimization

QUESTIONS COS BT
1. How learning differs from pure Optimization. CO3 L2

2. What is Surrogate Loss Functions and Early Stopping? CO3 L2

3. Explain Batch and Mini Batch Algorithms. What are the factors on CO3 L2
which Mini Batch size is decided?
4. List and Explain the Challenges involved in Neural Network Optim- CO3 L2
ization? Explain with examples.

5. Write the algorithm for Stochastic Gradient Descent. CO3 L2

6. What is Momentum? Write and Explain SGD with Momentum Al- CO3 L2
gorithm and why is it needed.
7. Write and explain Nesterov Momentum (NAG) algorithm. CO3 L2

8. Explain the following adaptive optimization learning Algorithms. CO3 L2

a. AdaGrad
b. RMSProp
c. Adam
9. Explain the following Approximate Second Order Methods: CO3 L2
a. Newton’s Method
b. Conjugate Gradients
c. Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm
10. Explain the following Optimization Strategies and Meta-Algorithms: CO3 L2
a. Batch Normalization
b. Coordinate Descent
c. Polyak Averaging
d. Supervised Pretraining

Ragini Krishna, Dept. ISE, SKIT. Page 2

 SRI KRISHNA INSTITUTE OF TECHNOLOGY
(Accredited by NAAC, Approved by A.I.C.T.E. New Delhi, Recognised by Govt. of Karnataka & Affiliated to V.T U., Belagavi)
#29, Chimney Hills, Hesaraghatta Main Road, Chikkabanavara Post, Bengaluru- 560090

Dept. of Information Science & Engineering

Deep Learning (21CS743)

Question Bank

Module 4 - CNN

QUESTIONS COS BT
1. What is Convolution Neural Network? Explain with a diagram. CO2 L2

2. Explain Convolution Operation. With a neat diagram explain the 2- CO2 L2

D convolution without kernel-flipping
3. With a neat diagram explain how sparse connectivity of data is CO2 L2
handled in CNN
4. With a neat diagram explain the components of a typical convolu- CO2 L2
tional neural network layer.

5. Explain the significance of pooling in CNN. Explain the working of CO2 L2

pooling with a neat diagram.
6. CO2 L2
rd CO2 L2
3 Internals
7. What are variants of Convolution Function? Explain Strides with ex- CO2 L2
amples. Also Show the effect of zero padding on network size.
8. What is Unshared Convolution? Explain the same mathematically CO2 L2
and with a neat diagram.
9. What is tiled Convolution? Give a Comparison of locally connected CO2 L2
layers, titled convolution and standard convolution.
10. What is transpose Convolution? CO2 L2
11. What are the different data formats supported by Convolution neural CO2 L2
network? Give examples.
12. What is Gabor function? Explain it with a variety of parameter with CO2 L2
a diagram.

Module 5 – RNN and other Applications

QUESTIONS COS BT
1. Define RNN. Explain unfolding of computational graphs. CO2 L2

2. What is RNN? Explain the important design patterns of Recurrent CO2 L2

Neural Networks.

Ragini Krishna, Dept. ISE, SKIT. Page 3

 SRI KRISHNA INSTITUTE OF TECHNOLOGY
(Accredited by NAAC, Approved by A.I.C.T.E. New Delhi, Recognised by Govt. of Karnataka & Affiliated to V.T U., Belagavi)
#29, Chimney Hills, Hesaraghatta Main Road, Chikkabanavara Post, Bengaluru- 560090

Dept. of Information Science & Engineering

3. Explain the concept of teacher forcing in Recurrent Neural Network CO2 L2

with a neat diagram.
4. What is Back Propagation Through Time (BPTT) algorithm? Ex- CO2 L2
plain how BPTT behaves.

5. State the various ways of representing RNN as various directed CO2 L2

graphical model.
6. Explain Modeling sequences conditioned in context with RNN CO2 L2
7. Explain the computation of bi directional Recurrent Neural Network. CO2 L2
8. Explain how an RNN can map an input sequence to an output se- CO2 L2
quence of the same length.
9. Explain how an RNN can map a fixed-size vector to a sequence. CO2 L2
10. Explain encoder-decoder sequence-sequence RNN architecture. CO2 L2
11. What are the various ways in which a recurrent neural network can CO2 L2
be made deep in many ways? Explain each with a neat diagram.
12. What are the various parameters that needs to be considered for CO2 L2
Large Scale Deep Learning? Explain them.
13. What are steps involved in Computer Vision? Explain them. CO2 L2
14. What is Natural Language Processing? List and explain the steps in- CO2 L2
volved in NLP.

Ragini Krishna, Dept. ISE, SKIT. Page 4