DEEP LEARNING
Assignment Questions — Unit 1 & Unit 2
Course: Deep Learning (23441O6) Branch: [Link] CSE (CS-Data Science)
Assignment: Unit 1 & Unit 2 Max Marks: 10 (5 each)
Unit 1: Introduction to Deep Learning
Long Answer Questions
Q1. Explain why deep learning has grown rapidly in recent years despite neural networks being a decades-old
concept. Discuss the role of Single Layer and Multiple Layer Perceptrons in solving non-linear classification
problems, using the XOR/XNOR problem as an example to show why a single-layer perceptron fails and how a
multi-layer network overcomes this limitation.
Q2. With a neat diagram, explain the architecture of a Deep Feed-Forward Network. Describe the concepts of
hidden units and architecture design, and explain the Back-Propagation algorithm used for training such networks
via stochastic gradient-based learning.
Short Answer Questions
1. Explain the historical evolution of Deep Learning.
2. Why has Deep Learning become popular in recent years?
3. Differentiate between Machine Learning and Deep Learning.
4. Define Artificial Neural Network (ANN) with a neat diagram.
5. Explain the structure of a biological neuron and compare it with an artificial neuron.
6. What is an activation function? Why is it required?
7. Explain the XOR problem. Why can't a single-layer perceptron solve it?
8. Define Feed Forward Neural Network (FFNN).
9. What are hidden layers and hidden units?
10. List the popular Deep Learning frameworks and their applications.
Unit 2: Convolutional Neural Network (CNN)
Long Answer Questions
Q1. Describe the basic architecture of a CNN, explaining the role of convolutional layers, activation functions
(Sigmoid, Tanh, ReLU, Softmax), and pooling layers (Max Pooling, Average Pooling, Global Pooling) in extracting
and reducing features from images. Illustrate with a suitable example of an image classification pipeline.
Q2. Compare the architectures of VGG, GoogLeNet, and ResNet, highlighting their key innovations and
differences. Also explain how techniques like Dropout, Batch Normalization, and Data Augmentation help improve
CNN training and prevent overfitting.
Short Answer Questions
1. Explain the architecture and working of an Artificial Neural Network with a neat diagram.
2. Explain the working of a Single Layer Perceptron with a suitable numerical example.
3. Describe the architecture of a Multi-Layer Perceptron (MLP). Explain how it overcomes the XOR problem.
4. Explain the Feed Forward Neural Network architecture with mathematical representation.
5. Derive the Backpropagation algorithm for training a neural network.
6. Explain Stochastic Gradient Descent (SGD) and compare it with Batch Gradient Descent and Mini-Batch
Gradient Descent.
7. Explain the steps involved in designing a Deep Feed Forward Network.
8. Compare TensorFlow, Keras, and PyTorch with suitable examples.
Note: Answer all questions with neat diagrams wherever applicable.