COURSEPACK (Even 2023-24)

SCHEME
The scheme is an overview of work-integrated learning opportunities and gets students out
into the real world. This will give what a course entails.

Course Title Deep Learning Course Type Integrated

Course Code R1UC604C Class B. Tech.
Activity Credits Credit Hours Total Number of Assessment in
Lecture 3 3 Classes per Semester Weightage

Tutorial 0 0

Self-study
Practical
Tutorial
Theory
Instruction Practical 1 2

SEE
CIE
delivery
Self-study 1 4
Total 5 9 45 0 30 60 50% 50%
Course
Course Lead Dr. Saurabh Singh Dr. Aditya Kishor Saxena
Coordinator
Theory Practical
Names 1. Dr. Saurabh Singh 1. Dr. Saurabh Singh
Course 2. Dr. Aditya K Saxena 2. Dr. Aditya K Saxena
Instructors
3. Dr. Abdul Aleem 3. Dr. Abdul Aleem
4. Dr. Deepa Joshi 4. Dr. Deepa Joshi
5. Dr. Garima Pandey 5. Dr. Garima Pandey
6. Dr. Siksha Singh 6. Mr. Siksha Singh
7. Dr. Vimal Singh 7. Dr. Vimal Singh
8. Dr. Himanshu Sharma 8. Dr. Himanshu Sharma
9. Dr. Suman Mann 9. Dr. Suman Mann
10.Dr. Sandeep Kumar 10.Dr. Sandeep Kumar

COURSE OVERVIEW
Deep learning is a sub-field of machine learning that focuses on learning complex, hierarchical feature
representations from raw data. The dominant method for achieving this, artificialneural networks, has
revolutionized the processing of data (e.g. images, videos, text, andaudio). In this course, students will
learn the fundamental principles, underlying mathematics,and implementation details of deep learning. This
includes the concepts and methods used to optimize these highly parameterized models, the modules that
make them up (linear, convolution, and pooling layers, activation functions, etc.), and common neural
networkarchitectures (convolutional neural networks, recurrent neural networks, etc.). Through in-depth
programming assignments, students will learn how to implement these fundamental building blocks as
well as how to put them together using Python.

COURSEPACK |
PREREQUISITE COURSE
PREREQUISITE COURSE YES
REQUIRED
Prerequisite Prerequisite course name
course code
Machine Learning

COURSE OBJECTIVE
 To understand the fundamentals of neural networks.
 To apply the concept of neural network in real-time use of deep learning.
 To demonstrate real-time development of applications of deep learning.
 To apply the basic concept of deep learning in the implementation of convolutional neural
network over various datasets.
 To implement various automatic model of Deep Learning neural networks and evaluate
their performance.

COURSE OUTCOMES(COs)
After the completion of the course, the student will be able to:
CO No. Course Outcomes
R1UC604C.1 Understand the fundamentals of Neural Networks.

R1UC604C.2 Apply the concepts of Neural Networks in the development of deep learning models.

R1UC604C.3 Develop and Analyze the deep learning models like RNN, CNN and LSTM.

R1UC604C.4 To be able to evaluate and optimize deep learning models

BLOOM’S LEVEL OF THE COURSE OUTCOMES

Bloom's taxonomy is a set of hierarchical models used for the classification of educational
learning objectives into levels of complexity and specificity. The learning domains are
cognitive, affective, and psychomotor.

Remember Understand Apply Analyse Evaluate Create

CO No.
KL1 KL 2 KL 3 KL 4 KL 5 KL 6
R1UC604C.1 √ √
R1UC604C.2 √
R1UC604C.3 √ √
R1UC604C.4 √

COURSEPACK |
PROGRAM OUTCOMES (POs):
The program enhances analytical, managerial, communication and
PO1
computing Application skills besides inculcating the virtues of self-study
Analyze and synthesis computing systems through quantitative and
PO2 qualitative techniques.
Design and develop computer programs in the areas related to algorithm,
PO3 web design, networking and AI
Inculcating essential skills as demanded by Indian and Global Software
PO4 industries through latest tools of learning. This also includes team-building
skills, audio- visual presentations and personality development programs.
To develop inter-twining competence in the field of Digital Marketing and
PO5
Commerce, Computing Skill and Computational tools.
To Develop practical skills to provide solutions to industry, society and
PO6
business.
Understand environmental issues and lead a life with concerns for
PO7
environment.
Apply ethical principles and commit to professional ethics and
PO8
responsibilities and norms of the computing science practice.
To make graduates understand cross cultural, societal, professional, legal
PO9
and ethical issues prevailing in industry.
PO10 Communicate effectively in both verbal and written form
To apply standard software engineering practices and strategies in software
PO11 project development using open-source programming environment to deliver
a quality of product for business success
Recognize the need for lifelong learning for continuous enhancement and up
PO12
gradation of technological changes in society.

Program Specific Outcomes (PSOs):

Have the ability to work with emerging technologies in computing requisite to

PSO1
Industry 4.0.
Demonstrate Engineering Practice learned through industry internship and research
PSO2 project to solve live problems in various domains.

COURSEPACK |
COURSE ARTICULATIONMATRIX
The Course articulation matrix indicates the correlation between Course Outcomes and Program Outcomesand
their expected strength of mapping in three levels(low,medium,and high).

PO10

PO11

PO12

PSO1

PSO2
PO1

PO2

PO3

PO4

PO5

PO6

PO7

PO8

PO9
COs#/ POs

R1UC604C.1 1

R1UC604C.2 3 1 1

R1UC604C.3 3 3 2

R1UC604C.4 3 2 3

Note: 1-Low, 2-Medium, 3-High

COURSEPACK |
COURSE ASSESSMENT
The course assessment patterns are the assessment tools used both in formative and summative
examinations.
CIE Total Marks
Type of Course Final Marks
(C) LAB @
MTE
Course-based
CIE SEE CIE*0.5+SEE*0.5
(Work+ Project^
Record)
COMPREHENSIVE 25 50 25 100 100 100

COURSE CONTENT

THEORY+ PRACTICAL

THEORY:
Machine Learning Brief: Introduction, Types of Machine Learning, Supervised Learning,
Unsupervised Learning, Reinforcement Learning. Applications of Machine Learning,
Biological Neuron, Idea of computational units, McCulloch–Pitts unit and Thresholding logic,
Linear Perceptron, Perceptron Learning Algorithm, Linear separability, Multilayer Perceptron.
ANN: Architecture of ANN, Types of Artificial Neural Network, Feed forward Neural
Network, Back propagation Algorithm, Gradient Descent, Activation Functions, Types of
Activation Functions, Deep Neural Network, Difficulty of training deep neural networks,
Greedy layer wise training, Regularization.
DL Models: Recurrent Neural Network (RNN), Long Short-Term Memory, Gated Recurrent
Units, Bidirectional LSTMs, Convolutional Neural Network, Introduction to images, Layers in
CNN, Convolution Layer, Min-Max Pooling Layer, Types of Pooling, Flattening Layer,
Padding Layer, Fully Connected Layer, Transfer Learning, Using pre-trained learning models-
YOLO.
Advance Techniques: Advanced optimization methods for neural networks (Adagrad, rmsprop,
adam), Saddle point problem in neural networks, Regularization methods (dropout, drop
connect, batch normalization). Generative Adversarial Networks, Multi-task Deep Learning,
Multi-view Deep Learning.

COURSEPACK |
LESSON PLAN
FOR THEORY 15 weeks * 3 Hours = 45 Classes) (1credit = 1Lecture Hour)
FOR PRACTICAL 15 weeks * 2Hours = 30 Hours lab sessions (1 credit = 2 lab hours)

Lec. Competency
No Topic Theory/Practical Skills (CO)
Machine Learning Brief: Applications,
Types of Machine Learning, Supervised Theory
1 Learning
2 Unsupervised Learning Theory
3 Reinforcement Learning Theory
Python Programming Fundamental Understan
4 Revision. Practical ding
Biological Neuron, Idea of computational Basics of
Machine CO1
units, McCulloch–Pitts unit and
5 Thresholding logic Theory Learning
Linear Perceptron, Perceptron Learning Theory
6 Algorithm, Linear separability.
7 Multilayer Perceptron Theory
8 Python Programming Libraries Revision. Practical
9 Tutorial/Revision Theory
10 Tutorial/Revision Theory
Architecture of ANN, Types of Artificial
Neural Network, Feed forward Neural Theory
11 Network
Python program to implement k-Nearest
Neighbor algorithm to classify the iris
dataset. Print both correct and wrong
12 predictions. Practical
Activation Functions, Types of Activation
13 Functions Theory
14 Back propogation Neural Network Theory
15 Gradient Descent Theory
Applying
Python program to implement the non-
Machine
parametric locally weighted regression
Learning for CO1, CO2
algorithm in order to fit the data point.
Select the appropriate data set for your
Problem
16 experiment and draw graphs. Practical Solving
Deep Neural Network, Difficulty of Theory
17 training deep neural networks
18 Greedy layer wise training Theory
Regularization methods (dropout, drop
19 connect, batch normalization) Theory
Python program to build a machine
learning model which will predict
whether or not it will rain tomorrow by
20 studying past data. Practical
21 Tutorial/Revision Theory
COURSEPACK |
 Theory
22 Tutorial/Revision
23 Tutorial/Revision Theory
Python program to build a machine
learning model which will predict
whether or not it will rain tomorrow by
24 studying past data. Practical
25 Recurrent Neural Network (RNN) Theory
26 Recurrent Neural Network (RNN) Theory
Long Short-Term Memory, Gated Theory
Basics of
27 Recurrent Units
Multi layer
Python program to implement AND, OR CO2, CO3
28 gates using perceptron. Practical
Perceptron
Learning
29 Bidirectional LSTMs Theory
Image processing basics, Types of Digital
30 Images Theory
31 Convolutional Neural Network Theory
Python program for classification of an
XOR problem with multi-layer
32 perceptron. Practical
Theory
33 Layers in CNN
34 Pooling Layer, Types of Pooling Theory
Flattening Layer, Padding Layer, Fully Solving
35 Connected Layer, Theory Real- world
Python program to implement problems
CO3
classification linearly separable data with using
36 perceptron. Practical Perceptron
37 Transfer Learning Theory Models
Theory
38 Using pre-trained learning models-YOLO
39 Tutorial/Revision Theory
Python program to recognize
40 handwritten digits using neural network. Practical
41 Tutorial/Revision Theory
42 Tutorial/Revision Theory
Advanced optimization methods for
neural networks (Adagrad, rmsprop, Theory Understandin
43 adam) g Neural
Python program to study a bank credit Networks CO3
data set and determine whether a Optimization
transaction is fraudulent or not based on Methods
44 past data. Practical
Saddle point problem in neural
45 networks, Theory
Saddle point problem in neural
46 networks, Theory
COURSEPACK |
 Python program for implementation of
47 CNN. Practical
Regularization methods (dropout, drop
48 connect, batch normalization) Theory
Regularization methods (dropout, drop Theory
49 connect, batch normalization)
50 Generative Adversarial Networks Theory
Python program for implementation of
51 RNN. Practical Understandin
52 Generative Adversarial Networks Theory g and CO3, CO4
53 Multi-task Deep Learning Theory Analyzing
Theory Variations
54 Multi-view Deep Learning. of ANNfor
Python program for implementation of different
55 LSTM. Practical tasks
56 Tutorial/Revision Theory
57 Tutorial/Revision Theory
58 Tutorial/Revision Theory
Python program for implementation of
59 Bidirectional LSTM. Practical CO4
60 Tutorial/Revision Theory

List of course based projects

AI-based Game Playing Agent: Create an AI agent capable of playing and winning games like chess, Go, or
1 Atari games using reinforcement learning.
Handwritten Digit Recognition: Implement a neural network to classify handwritten digits from the MNIST
2 dataset.
3 Sentiment Analysis: Build a model to analyze sentiment in text data, such as movie reviews or tweets.
Image Classification: Develop a convolutional neural network (CNN) to classify images from datasets like
4 CIFAR-10 or ImageNet.
Predictive Maintenance: Create a model to predict equipment failure or maintenance needs based on sensor
5 data.
Healthcare Diagnosis: Use machine learning techniques to assist in medical diagnosis based on patient data
6 or medical images.
Stock Price Prediction: Build a model to predict stock prices using historical data and technical indicators.
7
Spam Email Detection: Develop a classifier to identify spam emails from a given dataset.
8
Recommendation Systems: Create a recommendation system for movies, products, or music based on user
9 preferences.
Anomaly Detection: Build a model to detect anomalies in data, such as fraudulent transactions or network
10 intrusions.

COURSEPACK |
 Natural Language Processing (NLP): Implement NLP techniques for tasks like text summarization, named
11 entity recognition, or language translation.

Facial Recognition: Develop a system to recognize faces in images or videos using deep learning techniques.
12
Autonomous Vehicles Simulation: Create a simulated environment where autonomous vehicles navigate
13 using reinforcement learning algorithms.

Gesture Recognition: Build a model to recognize hand gestures from images or videos, which can be used
14 for human-computer interaction.

Customer Churn Prediction: Develop a model to predict customer churn for businesses based on historical
15 customer data.
Fake News Detection: Implement a model to classify news articles as real or fake based on their content.
16
Predictive Analytics for Energy Consumption: Use machine learning to predict energy consumption patterns
17 and optimize energy usage.

Credit Risk Assessment: Build a model to assess credit risk for loan applicants using historical financial
18 data.
Object Detection: Develop a model to detect and localize objects within images or videos.
19
Human Activity Recognition: Create a system to recognize different human activities, such as walking,
20 running, or sitting, from sensor data.

Brain-Computer Interface: Explore EEG data and build a model to interpret brain signals for controlling
21 devices or applications.
Customer Segmentation: Use clustering algorithms to segment customers based on their behavior or
22 demographics.
Language Translation: Implement a neural machine translation system to translate text between different
23 languages.
Document Classification: Build a model to classify documents into predefined categories based on their
24 content.
Image Style Transfer: Develop a model to transfer the style of one image onto another, creating artistic
25 effects.
Time Series Forecasting: Use recurrent neural networks (RNNs) or long short-term memory networks
26 (LSTMs) to forecast time series data, such as stock prices or weather patterns.

Emotion Recognition: Build a model to recognize emotions from facial expressions in images or videos.
27
Chatbot Development: Create a conversational AI chatbot that can answer user queries or assist in tasks.
28
Semantic Segmentation: Develop a model to segment images at the pixel level, identifying different objects
29 or regions within the image.

Speech Recognition: Build a system to transcribe spoken language into text using deep learning techniques.
30
Customer Lifetime Value Prediction: Develop a model to predict the lifetime value of customers for a
31 business based on their past behavior.

COURSEPACK |
 Video Activity Recognition: Extend human activity recognition to videos, identifying activities performed
32 over a sequence of frames.

Automated Essay Scoring: Build a model to automatically score essays based on their content, coherence,
33 and grammar.
Fraud Detection: Develop a model to detect fraudulent transactions in banking or e-commerce systems.
34
Music Genre Classification: Create a model to classify music into different genres based on audio features.
35
Topic Modeling: Use techniques like Latent Dirichlet Allocation (LDA) to identify topics within a collection
36 of documents.
Object Tracking: Implement algorithms to track objects in videos across frames, useful for surveillance or
37 self-driving cars.
Satellite Image Analysis: Analyze satellite imagery for tasks like land cover classification, urban
38 development monitoring, or disaster response.

Hand Gesture Recognition: Extend gesture recognition to recognize hand gestures in real-time for
39 controlling devices or applications.

E-commerce Recommendation: Build a personalized recommendation system for an e-commerce platform

40 based on user browsing and purchase history.

Drug Discovery: Use machine learning to predict the biological activity of molecules and assist in drug
41 discovery processes.
Smart Home Automation: Develop a system to automate tasks in a smart home environment based on sensor
42 data and user preferences.

Facial Attribute Detection: Build a model to detect facial attributes such as age, gender, or facial hair from
43 images.
Speech Emotion Recognition: Develop a model to recognize emotions from speech signals, useful for
44 applications like customer service or mental health monitoring.

Urban Mobility Prediction: Predict traffic flow or transportation demand in urban areas using machine
45 learning techniques.

BIBLIOGRAPHY

Text Book:
1. Deep learning (Hard Cover)- Ian Goodfellow, Yoshua Bengio, Aaron
ReferenceBooks:
1. Dive into Deep Learning - Aston Zhang, Zachary C. Lipton, Mu Li, and Alexander J.
Smola, Open community
2. Pattern Classification- Richard O. Duda, Peter E. Hart, David G. Stork, John Wiley &
Sons Inc.
3. Deep Learning: A practical Approach, R. Chopra.
COURSEPACK |
Webliography:
1. https://course.fast.ai/
2. https://www.youtube.com/watch?v=7PiK4wtfvbA&list=PLBAGcD3siRDguyYYzhVw
Z3tLvOyyG5k6K
SWAYAM/NPTEL/MOOCs Certification:
1. Course Name: Deep Learning, By Prof. Prabir Kumar Biswas, IIT Kharagpur.
https://nptel.ac.in/courses/

COURSEPACK |
PROBLEM-BASED LEARNING

Exercises in Problem-based Learning (Assignments) (Min 45 Problems*)

SNo Problem KL
1 Download the fashion-MNIST dataset and plot 1 sample image for each class. Use "from K3
keras.datasets import fashion_mnist" for getting the fashion mnist dataset.
2 Implement a feedforward neural network which takes images from the fashion-mnist data K6
as input and outputs a probability distribution over the 10 classes.
3 Implement the backpropagation algorithm with support for the following optimisation K6
functions:
 sgd
 momentum based gradient descent
 nesterov accelerated gradient descent
 rmsprop
 adam
4 Use the train_test_split to divide the dataset and perform the classification. K3
5 Evaluate the performance of the model using Precision, Recall and F1-score measures. K5
6 Draw the confusion matrix. K4
7 In the model above, we would have used cross-entropy loss. Now compare the cross- K4
entropy loss with the squared error loss.
8 Display the comparison Charts for MSE and Cross-Entropy loss. K3

9 What is Deep Learning? How is it different than Machine Learning? K2

10 What is Neural Network? Describe the Multi-Layer Perceptron. K3

.11 What Is Data Normalization, and Why Do We Need It? K2

12 What is the Boltzmann Machine? K3

13 What Is the Role of Activation Functions in a Neural Network? K2

14 What Is the Cost Function? K2

15 Discuss the Gradient Descent. K2

16 Discuss and demonstrate the process of Backpropagation? K2

17 What Is the Difference Between a Feedforward Neural Network and Recurrent Neural K3
Network?
18 What Are the Applications of a Recurrent Neural Network (RNN)? K3

19 Describe and compare the Softmax and ReLU Functions? K3

20 What Are Hyperparameters tuned in the development of a deep neural network? K3

21 What Will Happen If the Learning Rate Is Set Too Low or Too High? K3

22 Explain the Dropout and Batch Normalization with example? K3

23 What Is the Difference Between Batch Gradient Descent and Stochastic Gradient K3
Descent?

COURSEPACK |
24 Discuss Overfitting and Underfitting, and How to Combat Them? K3

25 Demonstrate how Are Weights Initialized in a Network? K4

26 Explain the Different Layers on CNN? K3

27 Describe is Pooling on CNN, and How Does It Work? K4

28 How Does an LSTM Network Work? K4

29 What are Vanishing and Exploding Gradients? Explain with example. K3

30 What is the Difference Between Epoch, Batch, and Iteration in Deep Learning? K2

31 What do you mean by a tensor? What Are the Programming Elements in Tensorflow? K3

32 What Is an Auto-encoder? K3

33 Compare and contrast Bagging and Boosting with example? K4

34 What do you understand by transfer learning? Name a few commonly used transfer K4
learning models.
35 Discuss the difference between SAME and VALID padding in Tensorflow? K3

36 What are some of the uses of Autoencoders in Deep Learning? K3

37 What are the reasons for mini-batch gradient being so useful? K3

38 What do you understand by Leaky ReLU activation function? K3

39 What is Data Augmentation in Deep Learning? K3

40 Explain the Adam optimization algorithm. K3

41 Why is a convolutional neural network preferred over a dense neural network for an K4
image classification task?
42 Which of the following strategy does not prevent a model from over-fitting to the training K5
data?
1. Dropout
2. Pooling
3. Data augmentation
4. Early stopping
43 Explain two ways to deal with the vanishing gradient problem in a deep neural network. K4

44 Why is a deep neural network better than a shallow neural network? K5

45 What is the need to add randomness in the weight initialization process? K4

46 How can you train hyperparameters in a neural network? K4

COURSEPACK |