VARDHAMAN COLLEGE OF ENGINEERING

(AUTONOMOUS)
Affiliated to JNTUH, Approved by AICTE, Accredited by NAAC with A++ Grade, ISO 9001:2015 Certified
Kacharam, Shamshabad, Hyderabad – 501218, Telangana, India
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI & ML)
A7709 (Deep Learning )
(CSM/CSE/IT/AIML/AIDS)
Question Bank
Unit I
Introduction

S.n Questions Course Bloom’

o. Outcom s Levels
es
1 Compare and Contrast Biological and Artificial Neurons CO1 L4
2 Discuss the architecture of an Artificial neural network (ANN)? CO1 L2
3 List and explain the various activation functions used in modeling of artificial neuron. CO1 L2
4 Describe Rectified linear units (ReLU) activation function and its use. CO1 L2
5 Describe Logistic Sigmoid and Hyperbolic Tangent activation functions and their uses. CO1 L2
6 How a deep neural network is different from ANN? CO1 L3
7 Differentiate between a regression problem and classification problem in machine learning CO1 L2
8 Differentiate between machine learning and deep learning. CO1 L4
9 List out the historical trends in Deep Learning. CO1 L1
10 Write about Mc Culloch-Pitts Neuron design for OR and AND Logic Functions CO1 L2
11 Write about Perceptron design for OR and AND Logic Functions CO1 L2
12 What do you understand by a decision surface or decision boundary in a classification CO1 L2
problem?
13 Explain linearly separable classification problem with example? CO1 L2
14 Explain linear inseparable classification problem with example? CO1 L2
15 Why the name feed forward neural networks justify with an example? CO1 L2
16 Discuss the role of hidden units in deep neural networks? CO1 L3
17 Mention the advantages of multilayer deep neural network? CO1 L1
18 What is a loss function? What is the role of loss function in improving learning of a model? CO1 L2
19 Discuss various types of loss functions in deep learning. CO1 L2
20 Discuss the Gradient Descent learning algorithm? CO1 L2
21 Explain cost function in gradient based learning CO1 L2
22 Illustrate Stochastic Gradient Descent learning algorithm. CO1 L2
23 Discuss the advantages and disadvantages of Stochastic Gradient Descent learning. CO1 L2
24 Compare Batch Gradient Descent, Mini Batch Gradient, and Stochastic Gradient Descent. CO1 L4
25 Explain how can we compute the output of ANN using Input and weight vector? CO1 L3

Abhishek Dixit
(Course Lead – A7709 - Deep learning)
Dept. of CSE (AI & ML)
 VARDHAMAN COLLEGE OF ENGINEERING
(AUTONOMOUS)
Affiliated to JNTUH, Approved by AICTE, Accredited by NAAC with A++ Grade, ISO 9001:2015 Certified
Kacharam, Shamshabad, Hyderabad – 501218, Telangana, India
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI & ML)
A7709 (Deep Learning )
(CSM/CSE/IT/AIML/AIDS)
Question Bank
Unit II
Convolution Neural Networks

S. no. Questions Course Bloom’s

Outcom Levels
es
1 Explain the architecture of a Convolution neural network (CNN)? CO2 L2
2 What are different architectures of CNN? Explain any one with example? CO2 L2
3 Examine the significance of the RELU Activation function in Convolution Neural Network CO2 L3
4 Examine the role of the Convolution Layer in CNN? Explain. CO2 L3
5 Explain the need of feature learning step in CNN? CO2 L4
6 Examine the role of kernel or filter matrix in CNN with example. CO2 L3
7 Why is striding operation used in CNN? Explain with example? CO2 L4
8 Explain how and when padding operation is implemented in CNN? CO2 L3
9 An input image has been converted into a matrix of size 12 X 12 along with a filter of size CO2 L3
3 X 3 with a Stride of 1. Determine the size of the convoluted matrix.
10 Why CNNs are better than feed-forward neural nets? Compare. CO2 L2
11 Summarize the advantages of using sparse connections in CNN. CO2 L2
12 Explain the significance of “parameter sharing” and “sparsity of connections” in CNN. CO2 L4
13 What do you understand by a tiled convolution and its implications? CO2 L2
14 Explain the significance of pooling layers (or down-sampling layer)in a CNN? CO2 L4
15 How pooling layers of CNN are used for dimensionality reduction? Explain. CO2 L4
16 List down the hyper-parameters of a Pooling Layer CO2 L1
17 Explain various types of pooling and their characteristics CO2 L2
18 Compare and contrast between weak prior and strong prior. CO2 L2
19 Discuss various types of data formats that can be used with CNN. CO2 L2
20 Analyze how the convolution operations can be made more efficient? CO2 L4
21 Construct the Architecture of Alexnet and explain in detail? CO2 L3
22 Discuss the architecture of GoogLeNet and explain. CO2 L3
21 Construct the Architecture of ResNet and explain its advantages. CO2 L3
24 With sample code explain image classification using CNN? CO2 L3
25 List out five real-life applications of CNN. CO2 L1

Abhishek Dixit
(Course Lead – A7709 - Deep learning)
Dept. of CSE (AI & ML)
 VARDHAMAN COLLEGE OF ENGINEERING
(AUTONOMOUS)
Affiliated to JNTUH, Approved by AICTE, Accredited by NAAC with A++ Grade, ISO 9001:2015 Certified
Kacharam, Shamshabad, Hyderabad – 501218, Telangana, India
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI & ML)
A7709 (Deep Learning)
(CSM/CSE/IT/AIML/AIDS)
Question Bank
Unit III
Sequence Modeling

S. Questions Course Bloom’s

no. Outcom Levels
es
1 What is sequence modeling, and why is it important in machine learning?
2 What is Recurrent neural net (RNN)? Explain the role of RNNs in sequence modeling. CO3 L2
3 Explain the architecture of a recurrent neural Network (RNN) an how does it differ from a CO3 L2
traditional feed-forward neural network? Illustrate unfolded RNN with a neat Sketch?
4 Define a loss function for a recurrent Neural Network (RNN). CO3 L2
5 How does a RNN preserve information from one RNN unit to Other unit? what are CO3 L2
advantages of such information?
6 What are the Major differences between CNN and RNN Explain CO3 L2
7 How does Recurrent Neural Network backpropagation vary from Artificial Neural Network CO3 L2
backpropagation?
8 What exactly do you mean by exploding and vanishing gradients? CO3 L2
9 How do recurrent neural networks (RNNs) process sequential data? CO3 L3
10 Illustrate the training mechanism and gradient calculation in RNNs. CO3 L2
11 How does Recurrent Neural Network backpropagation vary from Artificial Neural Network CO3 L2
backpropagation?
12 What are the issues faced while training in Recurrent Networks? CO3 L2
13 Explain how would you add an L2 regularization to a recurrent neural network to CO3 L4
overcome the vanishing gradient problem?
14 How will you implement Batch Normalization in RNN? CO3 L2
15 How would you train he RNN model and evaluate on sequence modeling tasks? CO3 L3
16 Explain the concept of gates in recurrent neural networks (RNNs) CO3 L2
17 How can RNNs be applied to predict the next word in a sentence? CO3 L3
18 Explain various RNN models used in the fields of natural language processing and speech CO3 L4
recognition.
19 How do RNNs utilize their hidden states to handle sequential data, and why is this feature CO3 L3
crucial for tasks like language modeling?

20 What is GRU (Gated Recurrent Unit) and how does it differ from traditional recurrent CO3 L2
neural networks (RNNs)?
21 Illustrate with a diagram how bidirectional RNNs architecture process input sequences? CO3 L2
22 Apply the concept of RNNs to design a simple RNN model for sentiment analysis of movie CO3 L3
reviews?
23 State and explain working of Long short term memory network (LSTM) with an example. CO3 L2
24 Define the structure of an LSTM cell, detailing the functions of the input, output, and forget CO3 L2
gates.
25 How do LSTM networks maintain long-term dependencies in sequence data? Explain. CO3 L2
26 Analyze how LSTM's gating mechanisms help in capturing temporal dependencies and CO3 L4
compare its performance with traditional RNNs?
27 Explain sequence to sequence RNN architecture with suitable diagram and mathematical CO3 L2
expressions.
28 Outline the components of an encoder-decoder sequence to sequence model and Identify the CO3 L2
major challenges associated with them.
29 Analyze the role of the encoder and decoder in handling variable-length input and output sequences. CO3 L4

30 Construct an encoder-decoder sequence to sequence model for a machine translation task. CO3 L2

Abhishek Dixit
(Course Lead – A7709 - Deep learning)
Dept. of CSE (AI & ML)
 VARDHAMAN COLLEGE OF ENGINEERING
(AUTONOMOUS)
Affiliated to JNTUH, Approved by AICTE, Accredited by NAAC with A++ Grade, ISO 9001:2015 Certified
Kacharam, Shamshabad, Hyderabad – 501218, Telangana, India
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI & ML)
A7709 (Deep Learning)
(CSM/CSE/IT/AIML/AIDS)
Question Bank
Unit IV
Auto Encoders

S. Questions Course Bloom’s

no. Outcomes Levels
1 Define an auto-encoder. Explain its fundamental purpose in unsupervised learning. CO4 L2
2 Construct architecture of Auto encoder and discuss various functions it perform. CO4 L2
3 Analyze the role of the encoder and decoder in handling variable-length input and CO4 L4
output sequences.
4 What do you understand by reconstruction phase in auto-encoders.? CO4 L2
5 Define what is meant by the "latent space" in the context of auto-encoders. Why is it CO4 L4
important?
6 What is feature learning, and how do auto-encoders facilitate it during the training CO4 L2
process?
7 Discuss the loss function in autoencoders. Why do we use binary cross entropy as loss CO4 L4
function in autoencoders?
8 How would you train auto-encoders? List out important hyper-parameters you need CO4 L3
to set before training an auto-encoder.
9 Explain regularization in the context of autoencoders. Why is regularization CO4 L4
important in machine learning?
10 Describe the process of training the regularized auto-encoder using suitable CO4 L2
formulation.
11 What do you understand by dimensionality reduction? CO4 L2
12 Why do we need auto encoders when there are already powerful dimensionality CO4 L4
reduction techniques like Principal Component Analysis (PCA)?
13 Explain how auto-encoders can be used for dimensionality reduction and the CO4 L3
significance of the bottleneck/hidden layer in this context?
14 Explain under-complete and over-complete autoencoder and how they achieves CO4 L2
dimensionality reduction?
15 Write short notes on variational autoencoders (VAEs). CO4 L2
16 Elaborate on the concept of contractive encoders in autoencoders. CO4 L2
17 Discuss the need for contractive auto-encoder and how would you configure its loss CO4 L2
function to incorporate the regularization term?
18 What is the Jacobian matrix? How is it used in machine learning? CO4 L2
19 Explain the role of the Jacobian matrix in contractive autoencoders. How does it CO4 L3
relate to the penalty term used for regularization?
20 Why do we need Denoising? Discuss Image Denoising using auto-encoders. CO4 L2

Abhishek Dixit
(Course Lead – A7709 - Deep learning)
Dept. of CSE (AI & ML)
 VARDHAMAN COLLEGE OF ENGINEERING
(AUTONOMOUS)
Affiliated to JNTUH, Approved by AICTE, Accredited by NAAC with A++ Grade, ISO 9001:2015 Certified
Kacharam, Shamshabad, Hyderabad – 501218, Telangana, India
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI & ML)
A7709 (Deep Learning)
(CSM/CSE/IT/AIML/AIDS)
Question Bank
Unit V
Deep Generative Models

S. Questions Course Bloom’s

no. Outcomes Levels
1
2 What are Deep Generative Models, and what is their primary objective in machine CO5 L2
learning?
3 How deep generative models differ from discriminative models (e.g. SVM) in machine CO5 L4
learning?
4 Name two common types of Deep Generative Models and briefly describe how they CO5 L3
generate new data.
5 What is a Boltzmann Machine (BM) and what is its fundamental principle? CO5 L2
6 Describe the structure and functioning of Boltzmann machines in deep learning. CO5 L2
7 What is the role of energy Function in boltzman machine? CO5 L4
Discuss the training process involved in deep boltzmann machines (DBMs). CO5 L3
8 Describe the architecture of a deep belief network (DBN). CO5 L2
9 How are DBNs constructed from Restricted Boltzmann Machines (RBMs), CO5 L3
10 what advantages do DBNs offer in unsupervised learning tasks? CO5 L2
11 Discuss the challenges in training Deep Boltzmann Machines compared to Deep Belief CO5 L4
Networks.
12 How would you apply deep belief networks to perform feature learning on a dataset CO5 L3
of handwritten digits (e.g., MNIST)?
13 What are Generative Stochastic Networks (GSNs), and how do they generate samples? CO5 L2
14 How can Generative Stochastic Networks (GSNs) be applied to generate realistic CO5 L3
images of human faces?
15 Explain the principle behind Generative Adversarial Networks (GANs). CO5 L2
16 What are the primary components of a GAN? Explain. CO5 L2
17 Give mathematical Formulation for the loss functions of generator and discriminator CO5 L2
in GANs.
18 How do the generator and discriminator in GAN work together to improve the quality CO5 L3
of generated data?
19 Discuss the challenges associated with training generative adversarial networks CO5 L2
(GANs)?
20 How would you evaluate the performance of different generative models (e.g., CO5 L4
Boltzmann Machines, GANs) on a dataset of your choice.

Abhishek Dixit
(Course Lead – A7709 - Deep learning)
Dept. of CSE (AI & ML)