Deep Learning

Unit 4
1. What is the primary function of a Hopfield Network?
A) Pattern recognition
B) Associative memory retrieval
C) Image segmentation
D) Time-series forecasting
2. In a Hopfield Network, the weights between neurons are:
A) Asymmetric and binary
B) Symmetric and real-valued
C) Symmetric and binary
D) Asymmetric and real-valued
3. Which property is essential for a Hopfield Network to converge to a stable state?
A) Symmetric weights and no self-loops
B) Asymmetric weights and activation functions
C) Backpropagation of errors
D) Random weight initialization
4. A Boltzmann Machine is primarily used for:
A) Supervised learning tasks
B) Energy-based modeling and unsupervised learning
C) Image classification tasks
D) Sequence generation in RNNs
5. How does a Restricted Boltzmann Machine (RBM) differ from a standard Boltzmann Machine?
A) RBM removes connections between visible nodes and hidden nodes.
B) RBM restricts connections between hidden nodes and visible nodes only.
C) RBM allows visible and hidden nodes to interact freely.
D) RBM restricts connections between hidden nodes and between visible nodes.
6. The sigmoid activation function maps input values to:
A) A range between -1 and 1
B) A range between 0 and 1
C) A range between -∞ and +∞
D) A range between 0 and ∞
7. What is a common drawback of the sigmoid activation function?
A) It is computationally expensive for large datasets.
B) It cannot handle non-linear relationships.
C) It suffers from the vanishing gradient problem.
D) It cannot map negative values.
8. Autoencoders are commonly used for:
A) Predictive modeling
B) Dimensionality reduction and feature extraction
C) Reinforcement learning tasks
D) Time-series forecasting
9. The bottleneck layer in an autoencoder:
A) Performs classification tasks
B) Compresses input data to its latent representation
C) Adds noise to the input for training
D) Expands input features for reconstruction
10. Which of the following networks is most suitable for initializing weights in deep learning
models?
A) Hopfield Network
B) Restricted Boltzmann Machine (RBM)
C) Sigmoid Net
D) Autoencoder
Fill-in-the-Blanks Questions
1. A Hopfield Network is primarily used for __________ memory retrieval.
2. A Hopfield Network is designed for __________ tasks rather than supervised learning.
3. A Boltzmann Machine is a __________ learning algorithm.
4. The learning process in Boltzmann Machines involves minimizing the __________
function.
5. An RBM restricts connections between __________ and __________ nodes.

6. The __________ function in an RBM determines the probability of a particular state.

 7. The sigmoid activation function maps inputs to a range of __________.
8. Sigmoid functions are prone to the __________ gradient problem.

9. Autoencoders are primarily used for __________ extraction.

10. Autoencoders reconstruct their input by learning a __________ representation.
UNIT IV 5 marks

1. Hopfield Networks are a type of Probabilistic Neural Network that can be used for pattern
recognition and associative memory tasks. Explain the fundamental concepts of Hopfield
Networks and how they operate.Discuss the key components of Hopfield Networks, such as
the energy function and the update rule. [CO4,L2]

2. Boltzmann Machines are powerful generative models that can learn complex distributions of
data. Explain the fundamental concepts of Boltzmann Machines and how they differ from
other neural network architectures? Discuss the key components of Boltzmann Machines,
such as visible and hidden units, and the energy-based framework. How do these components
enable Boltzmann Machines to model the joint probability distribution of the data? [CO4,L2]

3. Boltzmann Machines have been widely used in various applications, such as image
recognition, recommendation systems, and generative modeling. Discuss the applications and
advantages of Boltzmann Machines in these domains. [CO4,L3]

4. Explain the different types of RBMs, and discuss their advantages and disadvantages.
[CO4,L2]
5. Explain how RBMs can learn meaningful representations and patterns from unlabeled data
and how these representations can be used for tasks. [CO4,L2]

6. Sigmoid nets, also known as sigmoidal neural networks, are a type of artificial neural
network that use sigmoid activation functions. Explain the concept of sigmoid nets and how
they differ from other neural network architectures? [CO4,L2]

7. Explain how sigmoid nets can be used to solve problems such as classification. [CO4,L3]
8. Sigmoid nets have been widely used for classification tasks. Discuss the suitability of
sigmoid nets for binary classification problems. Explain how sigmoid activation functions
map inputs to probabilities and how they can be interpreted as representing class
probabilities. [CO4,L2]

9. Explain the different types of autoencoders, and discuss their advantages and disadvantages.
 [CO4,L2]

10. Autoencoders are unsupervised learning models that are widely used for feature extraction,
dimensionality reduction, and generative modeling. Explain the concept of autoencoders and
how they differ from other neural network architectures. Discuss the components of
autoencoders, such as the encoder and decoder, and how they work together to reconstruct
the input data. [CO4,L2]

UNIT – IV 10 marks
1. Explain the different types of Hopfield nets, and discuss their advantages and disadvantages.
[CO4,L2]
2. Explain how Boltzmann machines can be used to model probability distributions. [CO4,L3]
3. Explain how RBMs can be used to learn deep representations of data. [CO4,L3]
4. Explain how sigmoid nets can be used to model linear regression and logistic regression
problems. [CO4,L3]
5. Explain how autoencoders can be used for tasks such as dimensionality reduction,
unsupervised learning, and anomaly detection. [CO4,L3]
 Unit 5
1. What is the main purpose of object recognition in computer vision?
A) Generating synthetic images
B) Identifying and classifying objects within images or videos
C) Enhancing image resolution
D) Storing large datasets
2. Which technique is commonly used in sparse coding?
A) Backpropagation
B) L1 regularization
C) Cross-entropy loss
D) Fourier transforms
3. What is a primary challenge in computer vision tasks?
A) Training time is minimal
B) Lack of large datasets
C) Handling variations in lighting, scale, and orientation
D) Inability to use neural networks
4. Natural Language Processing (NLP) is primarily focused on:
A) Processing numerical datasets
B) Interpreting, analyzing, and generating human language
C) Compressing large datasets into smaller formats
D) Optimizing network bandwidth
5. Which deep learning tool was developed by Facebook and is widely used for research in
deep learning?
A) TensorFlow
B) Torch
C) Theano
D) Caffe
6. What is the primary goal of sparse coding in machine learning?
A) To improve GPU performance
B) To represent data with as few active components as possible
C) To increase dataset size
D) To minimize gradient computation
7. Which of the following is NOT a tool used in deep learning?
A) Theano
B) PyTorch
C) Scikit-learn
D) Hadoop
8. Which deep learning tool is known for its focus on efficiency and use in embedded
systems?
A) Caffe
B) TensorFlow
C) Torch
D) Keras
9. What is a primary application of computer vision?
A) Analyzing genomic sequences
B) Real-time object detection and facial recognition
C) Synthesizing human voices
D) Training reinforcement learning agents
10. Theano is a deep learning framework known for:
A) High-level neural network APIs
B) Efficient numerical computation on GPUs
C) Real-time language translation
D) Pre-trained natural language models

Fill in the blanks:

1. Object recognition involves identifying and classifying objects within __________.
2. A commonly used algorithm for object recognition is the __________ convolutional
neural network (CNN).
3. Sparse coding aims to represent data with a __________ number of active
components.
4. Sparse coding is widely used in __________ signal processing.
5. Computer vision focuses on enabling machines to interpret and understand
__________.
6. Tasks like object detection, segmentation, and recognition are part of __________
vision.
 7. NLP involves processing and understanding human __________ by machines.
8. Tasks like sentiment analysis and translation are part of __________.
9. Tokenization splits text into __________ or sentences for analysis.
10. Caffe is widely used for __________ vision tasks like image classification.
UNIT – V
5marks
1. Explain the different types of object recognition tasks, such as image classification,
object detection, and localization.
[CO5,L2]
2. Give an example of a challenging object recognition task and explain how you would
solve it using deep learning. [CO5,L3]

3. Explain what sparse coding is and why it is useful for computer vision and natural
language processing tasks. [CO5,L2]

4. Give an example of how sparse coding can be used to solve a computer vision or
natural language processing task. [CO5,L3]

5. Sparse coding is a technique used in signal processing and machine learning for finding
sparse representations of data. Explain the concept of sparse coding and how it differs from
traditional dense representations? [CO5,L2]

6. Explain the different types of computer vision tasks, such as image classification,
object detection, semantic segmentation, and image generation.
[CO5,L3]
7. Discuss the different approaches to computer vision, such as traditional methods
based on hand-crafted features and deep learning-based methods. [CO5,L3]

8. Explain the different types of natural language processing tasks, such as machine
translation, and text summarization. [CO5,L3]

9. Compare and contrast the different deep learning tools available, such as Caffe,
Theano, and Torch.
[CO5,L2]
10. Discuss the advantages and disadvantages of Caffe, Theano, and Torch.
[CO5,L2]

UNIT – V 10 marks
1. Explain the different types of object recognition tasks, such as image classification.
[CO5,L3]
2. Explain what sparse coding is and why it is useful for computer vision and natural
language processing tasks. [CO5,L3]
3. Explain the different types of computer vision tasks.
[CO5,L2]
4. Explain the different types of natural language processing tasks, such as machine
translation, text summarization, and question answering. [CO5,L3]

5. Compare and contrast the different deep learning tools such as Caffe, Theano, and
Torch.[CO5,L2]