B TECH ARTIFICIAL INTELLIGENCE AND DATA SCIENCE

AIL DEEP LEARNING LAB CATEGORY L T P CREDIT

411 LAB 0 0 3 2

Preamble: This course aims to offer students hands-on experience on deep learning
algorithms. Students will be able to familiarize basic python packages for deep learning,
computer vision concepts for deep learning, sequence modelling, recurrent neural network also.
This course helps the learners to enhance the capability to design and implement a deep
learning architecture for a real time application.

Prerequisite: A sound knowledge in python programming, machine learning concepts, deep

learning algorithms.

Course Outcomes: After the completion of the course the student will be able to

CO 1 Implement advanced machine learning concepts using python.(Cognitive Knowledge

Level: Apply)

CO 2 Applybasic data pre-processing and tuning techniques. (Cognitive Knowledge Level:

Apply)

CO 3 Implement basic neural network and CNN on standard datasets. (Cognitive

Knowledge Level: Apply)

CO 4 Design and Implement sequence modelling schemes.(Cognitive Knowledge Level:

Apply)

CO 5 Implement auto encoders on standard datasets and analyse the performance.

(Cognitive Knowledge Level: Apply)

Mapping of course outcomes with program outcomes

PO 1 PO 2 PO 3 PO 4 PO 5 PO 6 PO 7 PO 8 PO 9 PO 10 PO 11 PO 12
CO 1

CO 2

CO 3

CO 4

CO 5
Assessment Pattern

Continuous Assessment
Bloom’s Category End Semester Examination %
Test %

Remember 20 20

Understand 20 20

Apply 60 60

Analyze
Evaluate
Create

Mark distribution

Total CIE ESE ESE

Marks Duration

150 75 75 3 hours

Continuous Internal Evaluation Pattern:

Attendance : 15 marks
Continuous Evaluation in Lab : 30 marks
Continuous Assessment Test : 15 marks
Viva-voce : 15 marks

Internal Examination Pattern: The marks will be distributed as Algorithm 30 marks,

Program 20 marks, Output 20 marks and Viva 30 marks. Total 100 marks which will be
converted out of 15 while calculating Internal Evaluation marks.

End Semester Examination Pattern: The marks will be distributed as Algorithm 30 marks,
Program 20 marks, Output 20 marks and Viva 30 marks. Total 100 marks will be converted
out of 75 for End Semester Examination.

Operating System to Use in Lab :Linux/Windows

Programming Language/Software to Use in Lab :matlab or python

Fair Lab Record:

All Students attending the Deep Learning Lab should have a Fair Record. The fair record
should be produced in the University Lab Examination. Every experiment conducted in the lab
should be noted in the fair record. For every experiment in the fair record the right hand page
should contain Experiment Heading, Experiment Number, Date of Experiment, Aim of
Experiment, Details of Experiment including algorithm and Result of Experiment. The left
hand page should contain a print out of the code used for the experiment and sample output
obtained for a set of input.

SYLLABUS

Familiarize python frameworks for deep learning, Data Preprocessing, Supervised

Unsupervised Learning, Design and Implementation of Simple Neural Networks, Back
Propagation, Regularization, Dropout, Build and analyze deep learning architectures like
CNN, RNN, LSTM, GRU, Autoencoders.

.
Reference Books
1. Deep Learning with Python, by François Chollet, Manning, 2021
2. Goodfellow, I., Bengio,Y., and Courville, A., Deep Learning, MIT Press, 2016.
3. Neural Networks and Deep Learning, Aggarwal, Charu C., c Springer International Pub-
lishing AG, part of Springer Nature 2018

LIST OF PRACTICE QUESTIONS

1. Basic python programs in machine learning.

2. Implement Simple Linear Regression with Synthetic Data.**
3. Implement basic image enhancement operations such as histogram equalization,
morphological operations. **
4. Data pre-processing operations such as duplicate or missing value management.
5. Implement Feed forward neural network with three hidden layers for classification on
CIFAR-10 dataset. Design and train a neural network that achieves high accuracy in clas-
sifying the images into their respective classes. Test different hyper-parameters
6. Implement a feed forward neural network with three hidden layers for the CIFAR-10
dataset.Train the network using a baseline optimization algorithm, such as stochastic
gradient descent (SGD) or Adam, without any specific weight initialization technique
or regularization technique. Record the accuracy and loss during training. (a) Repeat
 the training process with Xavier initialization for weight initialization. Compare the
convergence speed and accuracy of the network with the baseline results. Analyze the
impact of Xavier initialization on the network's performance. (b) Repeat the training
process with Kaiming initialization for weight initialization. Compare the convergence
speed and accuracy of the network with the baseline results. Analyze the impact of
Kaiming initialization on the network's performance. (c) Implement dropout regulariza-
tion by applying dropout to the hidden layers of the network. Train the network with
dropout regularization and compare its performance with the baseline results. Analyze
the impact of dropout on the network's performance in terms of accuracy and overfit-
ting (d) Implement L1 or L2 regularization techniques by adding a regularization term
to the loss function during training. Train the network with regularization and compare
its performance with the baseline results. Analyze the impact of regularization on the
network's performance in terms of accuracy and prevention of overfitting.**
7. Implement a Convolutional Neural Network (CNN) architecture for digit classification
on the MNIST dataset. Design and train a CNN model that achieves high accuracy in
recognizing handwritten digits.**
8. Digit classification using pre-trained networks like VGGnet-19 for MNIST dataset and
analyse and visualize performance improvement. Explore transfer learning using Con-
volutional Neural Networks (ConvNets) as fixed feature extractors and fine-tuning for
image classification. Analyze their performance on a new image classification task
while comparing the fixed feature extractor approach with fine-tuning.**
9. Implement a Recurrent Neural Network (RNN) for review classification on the IMDB
dataset. Design and train an RNN model to classify movie reviews as positive or nega-
tive based on their sentiment.**
10. Analyze and visualize the performance change while using LSTM (Long Short-Term
Memory) and GRU (Gated Recurrent Unit) instead of a standard RNN (Recurrent Neu-
ral Network) for sentiment analysis on the IMDB dataset. Compare the performance of
different RNN architectures and understand their impact on sentiment classification.**
11. Implement time series forecasting for the NIFTY-50 dataset. Design and train a model
to predict future values of the NIFTY-50 stock market index based on historical data.**
12. Implement a shallow autoencoder and decoder network for machine translation using
the Kaggle English to Hindi Neural Translation Dataset. Design and train a model to
translate English sentences to Hindi by leveraging the power of autoencoders and de-
coders.**