Course Course Title L T P C
Code
10212EC228 DEEP LEARNING 2 0 2 3
a) Course Category
Program Elective
b) Preamble
Deep Learning is one of the most attractive and promising areas of AI and machine
learning and finds itself vastly applicable in Computer Vision tasks. This course covers
the fundamentals from Artificial Neural Network to the current trending topic of
Convolution Neural Network and adversarial deep networks. Thus this course aims to
provide basic knowledge on applying deep learning techniques to solve various real life
problems.
c) Prerequisite
-
d) Related Courses
Machine Learning
e) Course Outcomes
Upon the successful completion of the course, students will be able to:
Knowledge
CO Course Outcomes Level (Based on
Nos. Revised Bloom’s
Taxonomy)
Analyze various parameters to improve the performance of deep learning
CO1 models K4
Apply various CNN network models and transfer learning for different
CO2 applications K3
Summarize the architectural features of diverse deep CNN networks. K2
CO3
Apply various RNN network models for sequence and text processing K3
CO4
Apply various adversarial network models for image generation K3
CO5
f) Correlation of COs with POs
PO PO PO PO PO PO PO PO PO PO1 PO1 PO1 PSO PSO
1 2 3 4 5 6 7 8 9 0 1 2 1 2
CO M L L H L L L
1
CO M L L H L L L
2
CO M M M H L L M
3
CO M M M H L L M
4
CO M M M H L L M
5
g) Course Content
UNIT I OPTIMIZATION OF DEEP LEARNING MODELS 6
Review of architectural principles of deep networks, configure capacity of model with
nodes and layers, Configure gradient precision with nodes and layers, Configure what to
optimize with loss functions, Configure speed of learning with learning rate, Fix over-fitting
and penalize weights with regularization, Decouple layers with dropout, Halt training with
early stopping, K-fold cross validation.
UNIT II BASICS OF CONVOLUTIONAL NEURAL NETWORKS ` 6
CNN overview, Building blocks of Convolution Neural Networks, Convolution
operation with strides, Convolution Layers, Pooling Layers, Fully Connected Layers,
Greedy layer wise training. CNN based deep learning and Transfer learning based deep
learning.
UNIT III DEEP CONVOLUTION NEURAL NETWORKS 6
CNN architectures - LeNet, Alexnet, VGG, GoogleNet, Xception and Inception
models- Application of transfer learning with Alexnet and Googlenet models for image
classification task.
UNIT IV RECURRENT NEURAL NETWORKS 6
Support for sequences in Neural Networks, Recurrent Neural Networks, Basic
architecture, variants of RNN models - LSTM architecture, GRU architecture, Application
of RNN models for sequence and text processing.
UNIT V AUTOENCODERS AND ADVERSARIAL NETWORKS 6
Autoencoder models, Introduction to Generative Adversarial Networks- Generator,
Discriminator, Adversarial Networks, Application of autoencoder for dimensionality
reduction and adversarial model for image generation.
TOTAL: 60 periods
Text Books
1. Heaton, Jeff. "Ian goodfellow, yoshua bengio, and aaron courville: Deep learning."
(2018): 305-307.
2. Michelucci, U. Applied Deep Learning—A Case-Based Approach to Understanding
Deep Neural Networks; Apress Media, LLC: New York, NY, USA, 2018.
3. Brownlee, Jason. Better deep learning: train faster, reduce overfitting, and make
better predictions. Machine Learning Mastery, 2018.
4. Patterson, Josh, and Adam Gibson. Deep learning: A practitioner's approach. "
O'Reilly Media, Inc.", 2017.
5. Manaswi, Navin Kumar. Deep learning with applications using python: chatbots and
face, object, and speech recognition with tensorflow and keras. Apress, 2018.
Reference Books
1. Pattanayak, Santanu, Pattanayak, and Suresh John. Pro deep learning with
tensorflow. New York, NY, USA:: Apress, 2017.
2. Aggarwal CC. Neural networks and deep learning. Springer. 2018.
3. Chollet, Francois. Deep learning with Python. Simon and Schuster, 2017.
List of online resources:
1. [Link]
2. [Link]
3. [Link]
4. [Link]
List of Experiments
S.
Name of the experiment CO Skill
No Level
1 a) Understanding and working with Keras and Tensorflow CO1 S3
b) Design a dense neural network using Tensorflow
c) Example of evaluating deep neural network models with different
numbers of nodes.
d) Example of evaluating deep neural network models with different
numbers of layers
(6 Hours)
a) Design simple CNN for MNIST dataset
b) Design a larger CNN for Cifar dataset CO2
2 c) Image data generator and improve model performance with data S3
augmentation process
d) Training a convnet from scratch for small dataset
(6 Hours)
a) Use a pre-trained convent architecture for image classification
3 b) Design recent variant of CNN model for image classification CO3 S3
(6 Hours)
4 a) Design a1D CNN model for sentiment analysis and 1D signal
analysis CO4 S3
b) Design a simple LSTM model for sequence classification
c) Design a CNN model with LSTM architecture for sequence
classification
(6 Hours)
a) Design a auto encoder for image generation and dimensionality reduction
b) Design DCGAN for image generation CO5 S3
5 (6 Hours)
