L T P C

Course Course DEEP LEARNING FOR DATA Course

20ITE612J E Professional Elective
Code Name ANALYTICS Category 3 0 2 4

Pre-requisite Co-requisite Progressive

20ITC504J Nil
Courses Courses Courses
Course Offering Department Information Technology Data Book / Codes/Standards Nil

Course Learning Rationale

The purpose of learning this course is to: Learning Program Learning Outcomes (PLO)
(CLR):
CLR-1 : Understand a Mathematical models to build a Deep learning Models. 1 2 3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
CLR-2 : Gain knowledge on basic key concepts of Deep learning

Level of Thinking (Bloom)

Multicultural Competence
Expected Proficiency (%)
CLR-3 : Understand the linear models and stochastic optimization methods for training deep neural networks

Expected Attainment (%)

Community Engagement
Disciplinary Knowledge

Self-Directed Learning
CLR-4 : Understand the intuition behind Convolutional Neural Networks and Recurrent Neural Networks

Analytical Reasoning

Scientific Reasoning
Reflective Thinking

Life Long Learning

Ethical Reasoning
CLR-5 : Introduce the basic principles of auto encoders

Leadership Skills
Problem Solving
Critical Thinking

Research Skills
CLR-6 : Understand the various deep architectures and applications of computer vision

Team Work

ICT Skills
Course Learning Outcomes
At the end of this course, learners will be able to:
(CLO):
CLO-1 : Understand the basic concept of linear algebra and able to develop basic Deep neural network model 1 80 70 M M H H M - - - L L - H - - H
CLO-2 : Apply the knowledge of gradient descent and regularization techniques in feed forward networks 1 85 75 M M - M M - - - M L - H - - H
CLO-3 : Create a Convolutional neural network for processing high dimensional data like images and Text 2 75 70 M H M H H - - - M L - H - - H
CLO-4 : Understand the sequence modeling and able to develop Recurrent models 2 85 80 M H M H H - - - M L - H - - H
CLO-5 : Acquire knowledge of auto encoders 3 85 75 H H - H M - - - M L - H - - H
Implement deep neural network for various tasks of computer vision like object detection, image
CLO-6 : 3 80 70 H H H H M - - - L L - H - - H
captioning.

Duration (hour) 15 15 15 15 15
Convolutional
Introduction--Machine Learning Basics: Deep feed forward networks- Auto encoders -
SLO-1 Networks - Deep Architectures in Vision
Learning Algorithms Introduction- Introduction
S-1 Introduction
Shallow Neural Network, Deep Introduction to imagenet and imagenet Large Scale
SLO-2 Supervised and Unsupervised Training Motivation Auto encoder Architecture
Neural Network Visual Recognition Challenge (ILSVRC)
Hyperparameter Tuning, Batch Convolutional under complete,
SLO-1 Linear Algebra for machine Learning Alexnet
Normalization operation regularized auto encoder
S-2
Dimensionality reduction-Principal stochastic, denoising,
SLO-2 Learning XOR Pooling VGGNet
Component Analysis-PCA contractive auto encoder

Professional Elective Course 957

Duration (hour) 15 15 15 15 15
Introduction to
SLO-1 Over/Under-fitting, Gradient-Based Learning Variational Autoencoders Resnet
Normalization
S-3
Various types of Gradient Various types of Applications of Auto
SLO-2 Cross-Validation - types Googlenet
Descent functions Normalization in CNN encoders
SLO-1 Lab 4: To build a simple feed Lab 10: To Perfom
Lab 7 : To Build a
Lab 1: To implement data classification forward neural network to compression on mnist Lab 13: To implement a Pre-trained CNN model as a
S4-5 CNN model to classify
SLO-2 using simple ML algorithm (Revisit) recognize handwritten dataset using auto Feature Extractor using Transfer Learning
Cat and dog image
character.(MNIST Dataset) encoder
Back propagation in Deep Optimization for Deep
SLO-1 Hyper parameters Vs Parameters Transfer learning,
neural network Applications in Learning
S-6
validation sets - Training set, Test set Mathematical derivation to Computer Vision
SLO-2 Optimization functions Siamese Networks
and validation set calculate weight updation
SLO-1 Estimators, Bias, Variance, Activation Functions Time Series Data RMS Prop for RNNs, CNN-RNN
S-7
SLO-2 Regularization Linear and non linear Activation Sequence Modeling: SGD for CNNs Applications of CNN-RNN
Recurrent Neural Image Compression -
SLO-1 Challenges Motivating Deep Learning Error Functions 3D CNN
Networks introduction
S-8 Difference between Auto
Maximum likelihood and cross Difficulty in Training Encoders and PCA-its
SLO-2 The Curse of Dimensionality Applications in video Tasks
entropy RNN Applications in Image
Compression
SLO-1 Lab 8: TO Build a Lab 11: To perform
Lab 5: To implement implement Recurrent Neural compression on Images
Lab 2: To implement Planer data different variants of gradient Network - Step by (collection of face images) Lab 14: To implement 3D CNN to classification of Data
S9-10
SLO-2 classification with a hidden layer descent algorithm and generate Step. (Spam using auto encoders and ( Apply Medical Image Processing )-
3D and 2D animation plots. classification, PCA and compare the
sentiment analysis) results.
Adversarial Generative
S-11 SLO-1 Historical Trends in Deep Learning Architecture Design LSTM Architecture Metric Learning
Networks - Introduction
Universal Approximation LSTM Applications,
SLO-2 Introduction to a simple DNN Motivation - why GAN
Properties and Depth RNN VS LSTM
GRU Architecture -
Generator and
S-12 SLO-1 Platform for deep learning differentiation algorithms Steps - update gate, Object Detection and Object Localization
Discriminator Functions
Reset gate
First order and second order Steps: Current GAN and Convolution RCNN, Fast RCNN, Faster RCNN- Working and
SLO-2 Introduction, requirements to build DNN
optimization algorithm memory content, Final networks comparison

Professional Elective Course 958

Duration (hour) 15 15 15 15 15
memory at current
time step
Various GAN - DCGANs,
Regularization for Deep Encoder Decoder - conditional GANs, Pix2Pix,
S-13 SLO-1 Deep learning software libraries Object Detection in Video
learning introduction CycleGANs, MalGAN,
TextGAN
Introduction to Tensorflow, Keras, Architecture and
SLO-2 Early Stopping, Drop out. GAN Applications YOLO Architecture
PyTorch applications
SLO-1 Lab 9: TO Build a
Lab 12: To Implement a
Lab 6: To implement gradient Recurrent Neural
Lab 3: To Build a deep neural network - Deep Convolutional GAN
S14-15 descent and backpropagation in Network - to generate Lab 15: To implement a YOLO model to detect object.
SLO-2 step by step to generate complex, color
deep neural network caption from the
images
Scene Images

1. Ian Goodfellow, Yoshua Bengio, Aaron Courville, “Deep Learning”, MIT Press, 5. Christopher and M. Bishop, “Pattern Recognition and Machine Learning”, Springer Science
2016 (available at http://www.deeplearningbook.org) Business Media, 2006.
2. Kevin P. Murphy, “Machine Learning: A Probabilistic Perspective”, MIT Press, 6. Jason Brownlee , “Deep Learning with Python” , ebook, 2016
Learning 2012 7. N. D. Lewis , “Deep Learning Step by Step with Python: A Very Gentle Introduction to Deep
Resources 3. Michael Nielsen, “Neural Networks and Deep Learning”, Online book, 2016 Neural Networks for Practical Data Science
(http://neuralnetworksanddeeplearning.com/)
4. Li Deng, Dong Yu, “Deep Learning: Methods and Applications”, Foundations and
Trends in Signal Processing.

Continuous Learning Assessment (CLA) (60% weightage)

Final Examination
Bloom’s CLA-1 CLA-2
(40% weightage)
Level of Thinking (20%) (25%) #CLA-3 (15%)
Theory Practice Theory Practice Theory Practice
Remember
Level 1 20% 20% 15% 15% 20% 15% 10%
Understand
Apply
Level 2 20% 20% 15% 15% 40% 20% 20%
Analyze
Evaluate
Level 3 10% 10% 20% 20% 40% 15% 20%
Create
Total 100 % 100 % 100 % 100 %
#CLA-3 will be a Self-Learning Component and is generally a combination from among one or more of these options:

Professional Elective Course 959

Assignments Surprise Tests Seminars Multiple Choice Quizzes
Tech. Talks Field Visits Self-Study NPTEL/MOOC/Swayam
Mini-Projects Case-Study Group Activities Online Certifications
Presentations Debates Conference Papers Group Discussions

Course Designers
Experts from Industry Experts from Higher Technical Institutions Internal Experts
1. Mr. Gautham Suriya Mathialagan, Head of Engineering, AI ,ML VR, Anavi 1. Dr. J. Umarani , IIITDM Kancheepuram 1. Ms. Meenakshi K, Asst. Prof. Department of Information
smart VR solutions Technology

Professional Elective Course 960