Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 7

GIET UNIVERSITY, GUNUPUR – 765022

B. Tech ( 7th Semester) – Question Bank


Sub Name – Deep Learning
(2020-2024)

UNIT - I
PART – A: (Multiple Choice Questions) (1 Mark)

Q.1. Answer ALL questions [CO# [PO#


] ]
a. In machine learning, what characterizes a "learning algorithm"? 01 01
(i) A set of rules for feature engineering (ii) A mechanism for reducing bias in the
model
(iii) A program that adapts its behaviour (iv) An algorithm that only works with labeled
based on experience data
b How is the determinant of a matrix related to linear independence in the context of feature 01 01
. vectors?
(i) Determines the magnitude of the vectors (ii) Indicates whether the vectors are
orthogonal
(iii) Determines if the vectors are linearly (iv) Measures the spread of the data points
dependent
c. What is the consequence of high bias in a machine learning model? 01 01
(i) Increased sensitivity to small (ii) Overfitting to the training data
fluctuations in the training data
(iii) Underestimation of the true model (iv) Overestimation of the true model
parameters parameters
d Why is cross-validation important in model evaluation? 01 01
.
(i) It guarantees a perfect fit to the training (ii)It ensure the model has low variance.
data
(iii)It provides an unbiased estimate of model (iv)It accelerates the training process.
performance on unseen data
e. Which technique is commonly used for dimensionality reduction and feature extraction in 01 01
machine learning?
(i)Principal component Analysis (ii)K-means clustering
(iii)Decission Tree (iv)Support vector machines
f. How does regularization in the context of loss functions impact the training of a machine 01 01
learning model?
(i)Accelerates convergence (ii)Increases the risk of overfitting
(iii)Reduces the importance of outliers (iv)Introduces non linearity to the model
g What is the fundamental idea behind McCulloch's Pitt neuron in the context of artificial 01 01
. neurons?
(i)It stimulates the firing of biological (ii)It uses a linear activation function
neurons.
(iii)It incorporates multiple layers for deep (iv)It focuses on binary classification only.
learning
h What does the convergence theorem for the perceptron guarantee about the learning process? 01 01
.
(i)It ensures convergence to a global (ii)It guarantees convergence for linearly

Page 1 of 7
minimum separable data.
(iii)It prevents the model from overfitting (iv)It accelerates the convergence rate.
i. In the context of the perceptron learning algorithm, what role does the learning rate play? 01 01
(i) It determines the number of hidden (ii) It controls the step size during weight
layers in the network updates
(iii) It specifies the size of the input layer (iv) It influences the choice of activation
function
j. How does threshold logic differ from traditional binary classification in neural networks? 01 01
(i) It uses a linear activation function (ii) It allows for continuous output values
(iii) It employs multiple thresholds for (iv) It has no connection to artificial
decision-making intelligence concepts

PART – B: (Short Answer Questions) (2 Marks)

Q.2. Answer ALL questions [CO#] [PO#]

a. What is Hyperparameter Tuning? 01 01


b. What are Learning Algorithms? 01 01

c. What is Backpropagation? 01 01

d. What are the Applications of Deep Learning? 01 02


e. What is the Gradient Learning method? 01 01
f. What is Bias? 01 01
g. What is Variance? 01 02
h. What is Estimators? 01 01
i. What is Linear Perceptron? 01 02
j. What is Keras? 01 03

PART – C: (Long Answer Questions) (10 Marks)

Answer ALL questions Marks [CO#] [PO#]


1. a. What are McCulloch-Pitts units and How it Works? 05 01 01
b. What is Linear Perceptron and explain its Architecture and Working? 05 01 01
2. a. What are Perceptron Learning Algorithm and Explain its Algorithms? 05 01 01
b. What is backpropagations? Explain How it works and Mention the Benefits? 05 02 02
3. a. Difference Between Over Fitting & Under Fitting? 05 01 01
b. What is Multilayer Perceptron Algorithm and Explain its architecture and Working? 05 02 01
4. a. What is Deep feed Forward Neural Network? Explain its Key Features and Use Cases? 05 02 02
b. Explain in detail about Bias-Variance Trade-Off? 05 02 01
5.a. Explain Various Types of Activation Functions? 05 02 02
b. Explain the Gradient Learning Method? 05 02 03
6.a. Mentions the Difficulties of Training Deep Neural Networks? 10 02 01

UNIT – II
PART – A: (Multiple Choice Questions) (1 Mark)

Q.1. Answer ALL questions [CO# [PO#


] ]
a. What is a characteristic feature of a deep neural network? 02 02

Page 2 of 7
(i) Shallow architecture (ii) Limited number of layers
(iii) Single hidden unit (iv) Multiple hidden layers
b In the context of neural networks, why is the XOR problem considered challenging? 01 02
.
(i) It has a linearly separable decision (ii) It requires deep architectures to learn
boundary
(iii) It has a small input space (iv) It does not involve any non-linearity
c. Which regularization technique adds a penalty term based on the absolute values of the 02 01
weights?
(i) L1 regularization (ii) L2 regularization
(iii) Dropout regularization (iv)Elastic Net regularization
d What is a key advantage of using the Rectified Linear Unit (ReLU) activation function? 02 02
.
(i) Avoids vanishing gradient problem (ii) Smooth and differentiable
(iii) Reduces computational complexity (iv) Suitable for binary classification tasks
only
e. How does early stopping prevent overfitting in a neural network? 02 02
(i) By increasing the learning rate (ii) By reducing the number of hidden layers
(iii) By monitoring the validation loss and (iv) By applying aggressive data augmentation
stopping training when it starts to increase
f. What is the main characteristic of the Adagrad optimization algorithm? 02 02
(i) Maintains a constant learning rate (ii) Adjusts the learning rate for each
throughout training parameter based on its historical gradients
(iii) Only updates the weights after (iv) Ignores the gradient information during
evaluating the entire training dataset optimization
g What makes the Adam optimization algorithm different from traditional stochastic gradient 02 02
. descent (SGD)?
(i)It uses a fixed learning rate (ii) It adapts the learning rates for each
parameter individually
(iii) It only updates the weights every few (iv) It does not require the calculation of
epochs gradients
h What type of architecture is generally required to solve the XOR problem using a neural 02 02
. network?
(i) Shallow neural network (ii) Recurrent neural network
(iii) Deep neural network (iv) Linear regression model
i. In what scenario would dropout regularization be most beneficial? 02 02
(i) When the training dataset is very small (ii) When the model has too few parameters
(iii) When the model is prone to overfitting (iv) When the model suffers from underfitting
j. When might the Sigmoid activation function be preferable over the ReLU activation function? 02 02
(i) When dealing with image data (ii) When the output needs to be in the range
[0, 1]
(iii) When computational efficiency is a (iv) When the model requires non-linearity
priority

PART – B: (Short Answer Questions) (2 Marks)

Q.2. Answer ALL questions [CO#] [PO#]

a. What is activation function? 02 02


b. What is learning XOR? 02 02
c. What is early stopping? 02 02
Page 3 of 7
d. What is Drop Out? 02 02
e. What is Back propagation? 02 02
f. What are learning algorithms? 02 02
g. What is Hyperparameter tuning? 02 02
h. What is overfitting? 02 01
i. What is underfitting? 02 01
j. What is Backtracking? 02 01

PART – C: (Long Answer Questions) (10 Marks)

Answer ALL questions Marks [CO#] [PO#]


1. a. Write 5 different types of activation functions explain with suitable diagrams? 05 02 02
b. Explain feed forward neural network and how it works? 05 02 02
2. a. Explain different libraries of deep learning ? 05 02 02
b. Explain deep neural networks and how it works? 05 02 02
3. a. Explain the use case and key features of Deep Feed Forward neural network? 05 02 02
b. Explain in details about bias – variance trade off? 05 02 02
4. a. Explain gradient learning method? 05 02 02
b. Mention all the difficulties to train a deep neural network model? 10 02 02
5.a. Explain the optimization methods in deep learning? 05 02 02
b. Explain Greedy layer-wise training? 05 02 02

UNIT – III
PART – A: (Multiple Choice Questions) (1 Mark)

Q.1. Answer ALL questions [CO# [PO#


] ]
a. Why are RNNs suitable for sequence data? 03 01
(i) They only process fixed-length (ii) They have a fixed number of layers
sequences
(iii) They can maintain a hidden state that (iv) They are resistant to vanishing gradients
captures temporal information
b What is the key improvement of LSTM networks over traditional RNNs? 03 02
.
(i) Higher computational efficiency (ii) Ability to handle only short sequences
(iii) Avoidance of vanishing gradient (iv) Exponential increase in the number of
problem parameters
c. What is the purpose of the activation function in a convolutional layer? 02 01
(i) Introduce non-linearity to the model (ii) Regularize the weights during training
(iii) Increase the interpretability of the model (iv) Enhance the model's ability to handle
spatial features
d Which normalization technique is commonly used in CNNs to improve convergence and 03 01
. generalization?
(i) Batch Normalization (ii) L1 Normalization
(iii)Group Normalization (iv) Local Response Normalization
e. What is the primary advantage of using convolutional layers in a Convolutional Neural 03 02
Network (CNN)?
(i) Increased interpretability (ii) Parameter sharing and translation
invariance
(iii) Reduction of overfitting (iv) Faster convergence
Page 4 of 7
f. In medical imaging, what is a common application of computer vision? 03 02
(i) Sentiment analysis of patient reviews (ii) Object detection in outdoor scenes
(iii) Tumor detection in X-ray images (iv) Handwriting recognition in medical
records
g How does attention mechanism improve sequence modeling in neural networks? 03 02
.
(i) By reducing the model's capacity (ii) By allowing the model to focus on
specific parts of the input sequence
(iii) By increasing the number of parameters (iv) By removing the need for recurrent
in the model connections
h What is the primary challenge in sequence modeling tasks? 03 02
.
(i) Handling spatial relationships (ii) Capturing temporal dependencies
(iii) Balancing class distribution (iv) Feature extraction
i. Which of the following is a common application of computer vision? 03 02
(i) Sentiment analysis (ii) Speech recognition
(iii) Object detection (iv) Text summarization
j. In a CNN, what is the purpose of pooling layers? 03 02
(i) Increase the spatial dimensions of the (ii) Reduce the spatial dimensions of the input
input
(iii) Introduce non-linearity to the model (iv) Regularize the weights during training

PART – B: (Short Answer Questions) (2 Marks)

Q.2. Answer ALL questions [CO#] [PO#]

a. Give a brief description of CNN? 03 02


b. What are the Applications of Computer Vision? 03 02
c. What is Pooling? 03 01
d. What is the Purpose of Normalisation in CNN? 03 02
e. What is LSTM? 03 02
f. What is RNN? 03 02
g. What are the Challenges in Sequence Modeling? 03 02
h. What is the Architecture of VGGNet? 03 01
i. What is Bidirectional in RNN? 03 02
j. What is Normalization? 03 01

PART – C: (Long Answer Questions) (10 Marks)

Answer ALL questions Marks [CO#] [PO#]


1. a. What is CNN? Draw and Explain the Architecture of CNN? 05 03 02
b. Difficulties of training a recurrent neural network? 10 03 03
2. a. What is Pooling and Purpose of Pooling? 05 03 02
b. What is normalization and Explain the purpose and types of Normalization in CNN? 05 03 02
3. a. What is RNN, Explain in detail about RNN? 05 03 01
b. What are the Applications of Computer Vision in CNN? 05 03 02
4. a. Difference between LSTM and RNN? 05 03 02
b. Give a detailed overview of ImageNet? 05 03 02
5.a. What is polling and the purpose of pooling? 05 03 02
b. Give a detailed overview of VGGNet and LeNet? 05 03 02
Page 5 of 7
6.a. Explain the architecture of Long short-term memory? 05 03 02
b. What is Bidirectional RNN? Explain in detail about its working and its need with an 05 03 02
Example.

UNIT – IV
PART – A: (Multiple Choice Questions) (1 Mark)

Q.1. Answer ALL questions [CO# [PO#


] ]
a. What is the primary objective of an autoencoder? 04 01
(i) Classification (ii) Regression
(iii) Dimensionality reduction (iv) Clustering
b Which of the following is a common application of autoencoders? 04 01
.
(i) Image classification (ii) Speech recognition
(iii) Dimensionality reduction (iv) All of the above
c. What is the purpose of dimensionality reduction in the context of autoencoders? 04 01
(i) Increasing computational complexity (ii) Reducing noise in data
(iii) Improving model interpretability (iv) None of the above
d Which type of autoencoder is well-suited for handling sequential data, such as time series? 04 01
.
(i) Denoising autoencoder (ii) Variational autoencoder
(iii) Recurrent autoencoder (iv Sparse autoencoder
e. In dimensionality reduction, what does the term "encoder" refer to in an autoencoder 04 01
architecture?
(i)The part of the network responsible for (ii) The activation function
reducing dimensionality
(iii) The loss function (iv)The output layer
f. What is a denoising autoencoder designed to do? 04 01
(i)Add noise to input data (ii) Remove noise from input data
(iii) Increase the complexity of the model (iv) None of the above
g Which type of autoencoder is known for its ability to generate new data samples similar to the 04 01
. training data?
(i) Contractive autoencoder (ii) Variational autoencoder
(iii) Sparse autoencoder (iv) Stacked autoencoder
h How does a variational autoencoder differ from a traditional autoencoder? 04 01
.
(i) It introduces randomness in the encoding (ii) It has more layers in the neural network
process
(iii) It uses a different activation function (iv) It only works with image data
i. What is a common challenge associated with autoencoders in practice? 04 01
(i)Difficulty in training deep architectures (ii) Inability to handle categorical data
(iii Limited applications in computer vision (iv) None of the above
j. Which dimensionality reduction technique is often used as a preprocessing step before 04 01
applying autoencoders?
(i) Principal Component Analysis (PCA) (ii) Support Vector Machines (SVM)
(iii) K-Means clustering (iv) Decision Trees

PART – B: (Short Answer Questions) (2 Marks)


Page 6 of 7
Q.2. Answer ALL questions [CO#] [PO#]

a. What do you mean by Encoder? 04 01


b. What is the need for an encoder? 04 02
c. What are the types of autoencoders? 04 01
d. What is the fundamental purpose of an autoencoder in neural networks? 04 02
e. What is Stochastic Autoencoders? 04 03
f. Why do we need Autoencoder? 04 02
g. Name one type of autoencoder that is specifically designed to introduce noise to the input data 04 01
during training?
h. Define LSTM (Long Short-Term Memory) networks in the context of recurrent neural 04 03
networks, and mention a specific advantage they have over traditional RNNs.
i. Briefly define what an autoencoder is and explain its main objective in the field of neural 04 02
networks.
j. What is Convergence Theorem for Perceptron? 04 02

PART – C: (Long Answer Questions) (10 Marks)

Answer ALL questions Marks [CO#] [PO#]


1. a. What is Autoencoder and explain its Architecture. 05 04 03
b. Why optimization is needed for Deep Learning and Explain the optimizer SGD for 10 04 02
CNN?
2. a. What is Dimensionality Reduction and How it can be done with Autoencoder? 05 04 03
b. What is under complete Autoencoder and Denoising Autoencoder? 05 04 04
3. a. What are the Types of Autoencoders and explain each of them? 05 04 02
b. How does an autoencoder differ from traditional feedforward neural networks in terms 05 04 03
of architecture and functionality?
4. a. Justify the advantage of Autoencoder over principal component analysis for 05 04 02
Dimensionality Reduction?
b. Discuss the role of the encoder and decoder in an autoencoder, and explain how they 10 04 03
contribute to the overall learning process.
5.a. What is Image Segmentation and How it can be done with Autoencoder? 05 04 03
b. What are the Applications of Autoencoder and Applications of LSTM ? 05 04 03

--- End of Paper ---

Page 7 of 7

You might also like