AI_ML_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaPresentation.pptx

Applications of Artificial
Intelligence in
Engineering and
Agricultural Sciences

Conten
ts
Introduction to AI & ML
01
Career in AI & ML
02
Start ups to drive growth o
ML
03
Skills for success in AI &
ML
04
Brief discussion and
Conclusions
05

What do you think Artificial
Intelligence is?

Artificial
Intelligence in
Movies
In movies, robots are able to talk,
think, have emotions, and make
decisions just like humans.

What is Artificial
Intelligence?
Artificial Intelligence is the development of
computer systems that are able to
perform tasks that would require human
intelligence.
Examples of these tasks are visual
perception, speech recognition,
decision-making, and translation between
languages.

Real Life A.I.
Examples
❏ Self Driving Cars
❏ Boston Dynamics
❏ Navigation
Systems
❏ ASIMO
❏ Chatbots
❏ Human vs
Computer Games
❏ Many More!

Weak A.I.
Machines with weak Artificial Intelligence
are made to respond to specific situations,
but can not think for themselves

Strong A.I.
A machine with strong A.I. is
able to think and act just like a
human. It is able to learn from
experiences.
Since there are no real life
examples of strong A.I. yet,
the best representation would
be how Hollywood portrays
robots.

Machine
Learning
An application of
Artificial Intelligence
that gives machines the
ability to learn and
improve without the
help of humans or new
programming.

The Turing Test
In the 1950s Alan Turing created
the Turing Test which is used to
determine the level of intelligence of
a computer.

Controversy of
Turing Test
Some people disagree with the Turing
Test. They claim it does not actually
measure a computer's intelligence.
Instead of coming up with a response
themselves, they can use the same
generic phrases to pass the test.

The Future of
A.I.
★ Military Bots
★ The perfect lawyer
★ Music
★ Business
★ Healthcare

Benefits of A.I.
The most important purpose of
A.I. is to reduce human
casualties in
➔ Wars
➔ Dangerous Workspaces
➔ Car Accidents
➔ Natural Disasters
Or to just make everyday life
easier by helping with tasks
such as:
➔ Cleaning
➔ Shopping
➔ Transportation

AI is the main tool behind new-age
innovation and discoveries like
driverless cars or disease detecting
algorithm
Generalized AI is worth thinking about
because it stretches our imaginations
and it gets us to think about our core
values and issues of choice
Artificial Intelligence will be ‘vastly
smarter’ than any human and would
overtake us by 2025.
We are now solving problems with
machine learning and AI that were…in
the realm of science fiction for the last
several decades

Artificial Intelligence and
Machine Learning
AI is trained final output machine which
mimic like human brain
Ex: Amazon Alexa
ML is a subset of AI. It is a technique to
achieve AI. Ex: Spam Detection
Artificial Intelligence Machine Learning

Normal Computer vs ML
• The difference between normal computer software
and machine learning is that a human developer
hasn’t given codes that instructs the system how to
react to situation, instead it is being trained by a
large number of data.

Artificial Intelligence and Machine Learning in Industry 4.0
Breakdowns of industrial development and the great changes in related categories
Mechanization, stream
and water power
Electronic and IT systems,
Automation
Artificial intelligence
Mass production
and Electricity
Industry
1.0
Industry
2.0
Industry
3.0
Industry
4.0
1760-1830 1870-1914 1970-2000 2015 -2050?

Applications of
AI & ML
1. Automated Customer Support
• Online shopping experience has been greatly enhanced
by chatbots because of the following reasons:
• They increase user retention by sending reminders
and notifications
• They offer instant answers compared to human
assistants, thus reducing response time
• Chatbots provide upselling opportunities through
personalized approach

2. Personalized Shopping Experience
• Implementation of artificial intelligence makes it
possible for online stores to use the smallest piece of
data about every followed link or hover to personalize
your experience on a deeper level.
• This personalization results into timely alerts,
messages, visuals that should be particularly interesting
to you, and dynamic content that modifies according to
users’ demand and supply.

3. Healthcare
• AI-enabled workflow assistants are aiding doctors free up
their schedules, reducing time and cost by streamlining
processes and opening up new avenues for the industry.
• In addition, AI-powered technology helps pathologists in
analyzing tissue samples and thus, in turn, making more
accurate diagnosis.

4. Finance
• Automated advisors powered by AI, are capable of
predicting the best portfolio or stock based on
preferences by scanning the market data.
• Actionable reports based on relevant financial data is also
being generated by scanning millions of key data points,
thus saving analysts numerous hours of work.

5. Smart Cars and Drones
• With autonomous vehicles running on the roads
and autonomous drones delivering the shipments, a
significant amount of transportation and service related issues
can be resolved faster and more effectively.

6. Travel and Navigation
• With AI-enabled mapping, it scans road information and utilizes
algorithms to identify the optimal route to take, be it in a bike, car,
bus, train, or on foot.

7. Social media
• Face book uses advanced machine learning to do
everything from serving content to you and to recognize
your face in photos to target users with advertising.
• Instagram (owned by Facebook) uses AI to identify visuals.
• LinkedIn uses AI to offer job recommendations, suggest
people you might like to connect with, and serving you
specific posts in your feed.

8. Smart Home Devices
• The connected devices of smart homes provide the data and
the AI learns from that data to perform certain tasks without
human intervention.

9. Creative Arts
• AI-powered technologies can help musicians create new
themes.

10. Security and Surveillance
• AI is making possible for humans to constantly monitor
multiple channels with feeds coming in from a huge
number of cameras at the same time.

Sophia introduced herself and spoke to the students
appearing for their exams.
India welcomes Robot Sophia for the first-ever interactive session in Kolkata
Sophia is a first AI humanoid robot developed by Hong Kong-based company Hanson Robotics

II. Career in AI
& ML
• There is a scope in developing the machines in game
playing, Speech recognition, language detection
machine, computer vision, expert systems, robotics,
and many more
• As per International Data Corporation (IDC)
Worldwide AI Guide, spending on AI systems will
accelerate over the next several years as
organizations deploy AI as part of their digital
transformation efforts & to remain competitive in
the digital economy
• Global spending on AI is forecast to double over the
next 4 years, $50.1 billion in 2020 to more than $110
billion in 2024.

Hospital and Medicine
Game Playing
Speech Recognition
Understanding
Natural Language
Computer Vision
Avenues
Cyber Security
Face Recognition
Transport
Marketing & Advertising

• AI covers many areas like medical diagnosis, stock
trading, robot control, scientific discovery.
• If you have the B.E / M. Tech degree in AI and ML, you
have the job opportunities in ISRO.
• You also have option to go in various top level microchip
manufacturer companies like Intel.
• Indian institute of Biology offers research in AI and
robotics
• Some AI job includes machine learning engineer, data
scientist, business intelligence developer, research
Scientists, and AI engineer.
• Intel offers job for AI and Robotics specialist.
• NASA is the best place to get job in AI in space
science.
• US tech companies are prepared to spend over $1
billion by 2020 in the process of poaching AI talent
from wherever they can get it.
• UK’s demand for AI skills has been growing much
faster than that in the US, Canada and Australia.
Avenues
India Abroad

World’s biggest companies heavily relying on AI & M

III. AI & ML to drive growth of
Start ups
.
• Startups ecosystem, has been nourished
with the advent of technology, and has
given rise to more evolved business
processes.
• These days Logistics, accounts,
marketing and team performance & HR
have all been supported by AI
technology.
• With the rising technologies
like AI, IoT and ML, its interesting
to watch the changing face of
Indian SMEs and startups.

Skills for success in AI & ML
• Working with AI requires an analytical thought process and the
ability to solve problems with cost effective and efficient
solutions.
• Professionals need technical skills to design, maintain and repair
technology and software programs.
• Those interested in becoming AI professionals need a education
qualification based on foundations of maths, technology, logic
and engineering prospective.
• Cognitive Science skills.

• Data dependencies
• Hardware dependencies
• Problem Solving approach
• Execution time
Overview
 Artificial Intelligence is a concept of creating
intelligent machines that stimulates human
behaviour whereas Machine learning is a
subset of Artificial intelligence that allows
machine to learn from data without being
programmed.

INTRODUCTION TO
Machine Learning

Regression, prediction, and classification are three different tasks in the field of machine
learning and statistics, each with its own objectives and methods. Here's an explanation of
the key differences between these three tasks:
1.Regression:
 Objective: The primary objective of regression is to model the relationship between a
dependent (target) variable and one or more independent (predictor) variables. It aims
to predict a continuous numeric value, making it suitable for problems where the
output is a real number.
 Output: The output of a regression model is a continuous value. For example,
predicting house prices, stock prices, or temperature would involve regression.
 Algorithm types: Linear regression, polynomial regression, support vector regression,
and neural networks (for regression) are common algorithms used for regression tasks.
 Evaluation metrics: Common evaluation metrics for regression include Mean Squared
Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-
squared (Coefficient of Determination).

2. Prediction:
 Objective: Prediction is a broad term that can encompass both regression and
classification tasks. In general, it refers to estimating future values or outcomes
based on historical or current data. It can involve predicting a wide range of values,
including numeric values (regression) or categorical outcomes (classification).
 Output: The output in prediction can be either continuous or categorical, depending
on the specific problem being addressed.
 Examples: Predicting the future stock price (regression) or classifying whether an
email is spam or not (classification) are both examples of prediction tasks.
 Algorithm types: The choice of algorithms for prediction depends on the nature of
the problem, including whether it's regression or classification.

3. Classification:
 Objective: Classification is a supervised machine learning task where the goal is to
assign input data to predefined categories or classes. It is used when the output is
categorical or discrete. The primary objective is to categorize data points into
classes based on their features.
 Output: The output of a classification model is a category or class label. For
example, classifying emails as spam or not spam, identifying handwritten digits as
numbers 0-9, or classifying images of animals into different species are
classification tasks.
 Algorithm types: Common classification algorithms include logistic regression,
decision trees, random forests, support vector machines, k-nearest neighbors, and
deep learning models like convolutional neural networks (CNNs) and recurrent
neural networks (RNNs).
 Evaluation metrics: Evaluation metrics for classification tasks include accuracy,
precision, recall, F1-score, and area under the Receiver Operating Characteristic

Linear vs Non-linear Techniques
• For some tasks, input data
can be linearly separable,
and linear classifiers can be
suitably applied
• For other tasks, linear
classifiers may have
difficulties to produce
adequate decision
boundaries

Machine Learning Basics
• Artificial Intelligence is a scientific field concerned with
the development of algorithms that allow computers to
learn without being explicitly programmed
• Machine Learning is a branch of Artificial Intelligence,
which focuses on methods that learn from data and make
predictions on unseen data
12
Labeled Data
Labeled Data
Machine Learning
algorithm
Learned model Prediction
Training
Prediction

Machine Learning Types
• Supervised: learning with labeled data
 Example: email classification, image classification
 Example: regression for predicting real-valued outputs
• Unsupervised: discover patterns in unlabeled data
 Example: cluster similar data points
• Reinforcement learning: learn to act based on feedback/reward
 Example: learn to play Go
class A
class B
Classification Regression
Clustering

Supervised Learning
• Supervised learning categories and techniques
 Numerical classifier functions
o Linear classifier, perceptron, logistic regression, support vector machines (SVM), neural
networks
 Parametric (probabilistic) functions
o Naïve Bayes, Gaussian discriminant analysis (GDA), hidden Markov models (HMM),
probabilistic graphical models
 Non-parametric (instance-based) functions
o k-nearest neighbors, kernel regression, kernel density estimation, local regression
 Symbolic functions
o Decision trees, classification and regression trees (CART)
 Aggregation (ensemble) learning
o Bagging, boosting (Adaboost), random forest

Unsupervised Learning
• Unsupervised learning categories and techniques
 Clustering
o k-means clustering
o Mean-shift clustering
o Spectral clustering
 Density estimation
o Gaussian mixture model (GMM)
o Graphical models
 Dimensionality reduction
o Principal component analysis (PCA)
o Factor analysis

45
Nearest Neighbor Classifier
• Nearest Neighbor – for each test data point, assign the class label of the nearest
training data point
 Adopt a distance function to find the nearest neighbor
o Calculate the distance to each data point in the training set, and assign the class of the nearest
data point (minimum distance)
 It does not require learning a set of weights
Test
example
Training
examples
from class 1
Training
examples
from class 2

• For image classification, the distance between all pixels is calculated (e.g., using
ℓ1 norm, or ℓ2 norm)
 Accuracy on CIFAR-10: 38.6%
• Disadvantages:
 The classifier must remember all training data and store it for future comparisons with
the test data
 Classifying a test image is expensive since it requires a comparison to all training
images
ℓ1 norm
(Manhattan distance)

Advantages of Machine Learning
• Fast, Accurate, Efficient.
• Automation of most applications.
• Wide range of real life applications.
• Enhanced cyber security and spam detection.
• No human Intervention is needed.
• Handling multi dimensional data.

Introduction to Deep Learning and
Its Applications

Where is deep learning applied?
 Computer Vision: for applications like vehicle number plate identification and facial recognition.
 Information Retrieval: for applications like search engines, both text search, and image search.
 Marketing: for applications like automated email marketing, target identification
 Medical Diagnosis: for applications like cancer identification, anomaly detection
 Natural Language Processing: for applications like sentiment analysis, photo tagging

How deep learning works?
• Most deep learning methods use neural network architectures, which is why deep
learning models are often referred to as deep neural networks.
• The term “deep” usually refers to the number of hidden layers in the neural
network. Traditional neural networks only contain 2-3 hidden layers, while deep
networks can have as many as 150.
• Deep learning models are trained using large sets of labelled data and neural
network architectures that learn features directly from the data without requiring
manual extraction. feature extraction.

What is Classification ?
Classification is the task of
● Identifying to which of a set of categories
(sub-populations) a new observation
belongs.
● It is decided based on a training set of data
containing observations (or instances) whose
category membership is known.

Examples of Classification
● Classifying emails as spam or not spam
● Classifying flowers of a particular species like the Iris Dataset
● Classifying a credit card transaction as fraudulent or not
● Face recognition

Binary vs Multi-class Classification
• A classification problem with only 2 classes is referred to as binary classification
 The output labels are 0 or 1
 E.g., benign or malignant tumor, spam or no-spam email
• A problem with 3 or more classes is referred to as multi-class classification

Binary and Multiclass
Classification
5
Not 5
Binary
Classification
Multiclass
Classification
Classification is
done
between 2 classes
Classification is done
between multiple
classes

source:developer.nvidia.com/deep-‐learning-‐courses

Learning
Input: X Output: Y
Label”motorcycle”

Why is it
hard?
But the camera sees this:
You see this

Raw Image Representation
pixel 2
Cars
“Non”-‐Cars
Learning
Algorithm
pixel 2
pixel 1
pixel 1

Feature
Representations
Expert Knowledge!
Source: feature representations incomputer vision(Honglak lee)

Deep Learning: learn
representations!
Source:Lee et.al. ICML2009

So, 1. what exactly is deep learning ?
And, 2. why is it generally better than other methods on
image, speech and certain other types of data?
The short answers
1. ‘Deep Learning’ means using a neural network
with several layers of nodes between input and output
2.the series of layersbetween input & output do feature
identification and processingin a series of stages, just as our
brains seem to.

hmmm… OK, but:
3. multilayer neural networks have been around for
25 years. What’sactually new?
we have always had good algorithms for learning the
weights in networks with 1 hidden layer
but these algorithms are not good at learning the weights for
networks with more hidden layers
what’s new is: algorithms for training many-‐laternetworks

Input x1
Input x2
Bias x0
w0
w1
w2
f(x) output
Single Unit, Input, weights, activation function, output
Activation functions:
1. linear
2. Sigmoid
3. Tanh
4. Relu
5.Softmax
etc.
f(x) = g(w0 x0+ w1 x1+ w2 x2 )

A dataset
Fields class
1.4 2.7 1.9 0
3.8 3.4 3.2 0
6.4 2.8 1.7 1
4.1 0.1 0.2 0
etc …
Train the deep neural network

A dataset
Fields class
1.4 2.7 1.9 0
3.8 3.4 3.2 0
6.4 2.8 1.7 1
4.1 0.1 0.2 0
etc …
Initialize with random weights
1.4
2.7
1.9
0.7 (0)
Error=0.7
Comparewith the
target output

Adjust weights based on error
1.4
2.7
1.9
0.7 (0)
Error=0.7
Repeat this thousands, maybe millionsof times – each time
taking a random training instance, and making slight
weight adjustments
Algorithms for weight adjustment are designed to make
changes that will reduce the error

Learning Rule :
•
•
w1
w2
Neuron
(Perceptron)
a(k)
d
-
Weight
adjustment
•
hardlimiter
x1
x2
xn
wn
b
y
a

x1, x2, ……, xn are inputs

In a vector form X=[x1, x2, ……, xn]
Output from the perceptron
n
n
2
2
1
1 x
w
....
x
w
x
w
y 



 
 )
b
wx
( T
]
w
....
w
w
[
w n
2
1

 
 

 b
T
wx
lim
hard
)
y
lim(
hard
a
If y > 0 , a = 1
If y < 0 , a = 0

d= desired output
a=actual output from the perceptron
e= error = d – a
w1(new) = w1(old) + e.x1
wn(new) = wn(old) + e. xn
Learning rule :

In a vector form
w(new)=w(old)+ex
w=[w1 w2 …..wn]
One can also introduce a learning rate
w(new)=w(old)+ e X
where = learning rate
0.2    2
b(new) = b(old) + e

Simple network
The learning rule is similar to the multiple perceptron network.
0 = hardlim (y)
b = hardlim (wx+b)
•
x1
a1
•
x2
xn
y1

a2
y2

•

• Perceptrons are used for simple
pattern classification problems
Consider again the four class decision problem as :








































0
0
t
,
2
1
p
,
0
0
t
,
1
1
p 2
2
1
1








































1
0
t
,
0
2
p
,
1
0
t
,
1
2
p 4
4
3
3








































0
1
t
,
1
2
p
,
0
1
t
,
2
1
p 6
6
5
5












































1
1
t
,
2
2
p
,
1
1
t
,
1
1
p 8
8
7
7

• Perceptron
d
Learning
Rule
•
p1, …. , p8
e = d-a
y a

y>0, a=1
y<0, a=0

Initial weights and biases














1
1
)
0
(
b
,
1
0
0
1
)
0
(
w
1st iteration : 













0
0
t
,
1
1
p 1
1









1
1
))
0
(
b
p
)
0
(
w
(
lim
hard
a
1











1
1
a
t
e 1
  ,
0
1
1
0
1
1
1
1
1
0
0
1
ep
)
0
(
w
)
1
(
w T
1 



































0
0
e
)
0
(
b
)
1
(
b

2nd iteration : 













0
0
t
,
2
1
p 2
2
,
0
0
))
1
(
b
p
)
1
(
w
(
lim
hard
a
2 

















0
0
a
t
e 2











0
1
1
0
ep
)
1
(
w
)
2
(
w T
2









0
0
e
)
1
(
b
)
2
(
b

3rd iteration: : 














1
0
t
,
1
2
p 3
3









0
1
))
2
(
b
p
)
(
w
(
lim
hard
a
3
2









1
1
a
t
e 3











1
1
0
2
ep
)
2
(
w
)
3
(
w T
3









1
1
e
)
2
(
b
)
3
(
b

Iteration 4 through 8 produces no changes in weights
w(8)=w(7)=w(6)=w(5)=w(4)=w(3)
b(8)=b(7)=b(6)=b(5)=b(4)=b(3)
,
1
0
))
8
(
b
p
)
8
(
w
(
lim
hard
a
1 







 









1
0
a
t
e 1











2
0
0
2
ep
)
8
(
w
)
9
(
w T
1









0
1
e
)
8
(
b
)
9
(
b
The 9th iteration produces

PRESENTED BY: DR. SAMIT ARI
NIT, ROURKELA,
ODISHA, 769008
Evolution of Different
Deep Learning
Architectures

A Typical Convolution Neural Network (CNN)

A convolutional layer
A filter
A CNN is a neural network with some convolutional layers
(and some other layers). A convolutional layer has a number
of filters that does convolutional operation.
Beak detector

Convolution
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
-1 1 -1
-1 1 -1
-1 1 -1
Filter 2
…
…
These are the network
parameters to be learned.
Each filter detects a
small pattern (3 x 3).

Convolution
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
3 -1
stride=1
Dot
product

1
Convolutional Neural Networks (CNNs)
• When the convolutional filters are scanned over the image, they capture useful
features
 E.g., edge detection by convolutions
Filter
1 1 1 1 1 1 0.015686 0.015686 0.011765 0.015686 0.015686 0.015686 0.015686 0.964706 0.988235 0.964706 0.866667 0.031373 0.023529 0.007843
0.007843 0.741176 1 1 0.984314 0.023529 0.019608 0.015686 0.015686 0.015686 0.011765 0.101961 0.972549 1 1 0.996078 0.996078 0.996078 0.058824 0.015686
0.019608 0.513726 1 1 1 0.019608 0.015686 0.015686 0.015686 0.007843 0.011765 1 1 1 0.996078 0.031373 0.015686 0.019608 1 0.011765
0.015686 0.733333 1 1 0.996078 0.019608 0.019608 0.015686 0.015686 0.011765 0.984314 1 1 0.988235 0.027451 0.015686 0.007843 0.007843 1 0.352941
0.015686 0.823529 1 1 0.988235 0.019608 0.019608 0.015686 0.015686 0.019608 1 1 0.980392 0.015686 0.015686 0.015686 0.015686 0.996078 1 0.996078
0.015686 0.913726 1 1 0.996078 0.019608 0.019608 0.019608 0.019608 1 1 0.984314 0.015686 0.015686 0.015686 0.015686 0.952941 1 1 0.992157
0.019608 0.913726 1 1 0.988235 0.019608 0.019608 0.019608 0.039216 0.996078 1 0.015686 0.015686 0.015686 0.015686 0.996078 1 1 1 0.007843
0.019608 0.898039 1 1 0.988235 0.019608 0.015686 0.019608 0.968628 0.996078 0.980392 0.027451 0.015686 0.019608 0.980392 0.972549 1 1 1 0.019608
0.043137 0.905882 1 1 1 0.015686 0.035294 0.968628 1 1 0.023529 1 0.792157 0.996078 1 1 0.980392 0.992157 0.039216 0.023529
1 1 1 1 1 0.992157 0.992157 1 1 0.984314 0.015686 0.015686 0.858824 0.996078 1 0.992157 0.501961 0.019608 0.019608 0.023529
0.996078 0.992157 1 1 1 0.933333 0.003922 0.996078 1 0.988235 1 0.992157 1 1 1 0.988235 1 1 1 1
0.015686 0.74902 1 1 0.984314 0.019608 0.019608 0.031373 0.984314 0.023529 0.015686 0.015686 1 1 1 0 0.003922 0.027451 0.980392 1
0.019608 0.023529 1 1 1 0.019608 0.019608 0.564706 0.894118 0.019608 0.015686 0.015686 1 1 1 0.015686 0.015686 0.015686 0.05098 1
0.015686 0.015686 1 1 1 0.047059 0.019608 0.992157 0.007843 0.011765 0.011765 0.015686 1 1 1 0.015686 0.019608 0.996078 0.023529 0.996078
0.019608 0.015686 0.243137 1 1 0.976471 0.035294 1 0.003922 0.011765 0.011765 0.015686 1 1 1 0.988235 0.988235 1 0.003922 0.015686
0.019608 0.019608 0.027451 1 1 0.992157 0.223529 0.662745 0.011765 0.011765 0.011765 0.015686 1 1 1 0.015686 0.023529 0.996078 0.011765 0.011765
0.015686 0.015686 0.011765 1 1 1 1 0.035294 0.011765 0.011765 0.011765 0.015686 1 1 1 0.015686 0.015686 0.964706 0.003922 0.996078
0.007843 0.019608 0.011765 0.054902 1 1 0.988235 0.007843 0.011765 0.011765 0.015686 0.011765 1 1 1 0.015686 0.015686 0.015686 0.023529 1
0.007843 0.007843 0.015686 0.015686 0.960784 1 0.490196 0.015686 0.015686 0.015686 0.007843 0.027451 1 1 1 0.011765 0.011765 0.043137 1 1
0.023529 0.003922 0.007843 0.023529 0.980392 0.976471 0.039216 0.019608 0.007843 0.019608 0.015686 1 1 1 1 1 1 1 1 1
0 1 0
1 -4 1
0 1 0
Input Image Convoluted Image

Convolution
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
3 -3
If stride=2

Convolution
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
3 -1 -3 -1
-3 1 0 -3
-3 -3 0 1
3 -2 -2 -1
stride=1

Convolution
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
3 -1 -3 -1
-3 1 0 -3
-3 -3 0 1
3 -2 -2 -1
-1 1 -1
-1 1 -1
-1 1 -1
Filter 2
-1 -1 -1 -1
-1 -1 -2 1
-1 -1 -2 1
-1 0 -4 3
Repeat this for each filter
stride=1
Feature
Map

Color image: RGB 3 channels
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
-1 1 -1
-1 1 -1
-1 1 -1
Filter 2
1 -1 -1
-1 1 -1
-1 -1 1
1 -1 -1
-1 1 -1
-1 -1 1
-1 1 -1
-1 1 -1
-1 1 -1
-1 1 -1
-1 1 -1
-1 1 -1
Color image

1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
1
2
3
…
8
9
…
1
3
14
15
…
Only connect to 9
inputs, not fully
connected
4:
10:
16
1
0
0
0
0
1
0
0
0
0
1
1
3
fewer parameters!

1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1
1:
2:
3:
…
7:
8:
9:
…
1
3:
14:
15:
…
4:
10:
16:
1
0
0
0
0
1
0
0
0
0
1
1
3
-1
Shared weights
6 x 6 image
Fewer parameters
Even fewer parameters

The whole CNN
Fully Connected
Feedforward network
cat dog ……
Convolution
Max Pooling
Convolution
Max Pooling
Flattened
Can
repeat
many
times

Max Pooling
3 -1 -3 -1
-3 1 0 -3
-3 -3 0 1
3 -2 -2 -1
-1 1 -1
-1 1 -1
-1 1 -1
Filter 2
-1 -1 -1 -1
-1 -1 -2 1
-1 -1 -2 1
-1 0 -4 3
1 -1 -1
-1 1 -1
-1 -1 1
Filter 1

Why Pooling
• Subsampling pixels will not change the object
Subsampling
bird
bird
We can subsample the pixels to make image
smaller
fewer parameters to characterize the image

Max
Pooling
1 0 0 0 0 1
0 1 0 0 1 0
0 0 1 1 0 0
1 0 0 0 1 0
0 1 0 0 1 0
0 0 1 0 1 0
6 x 6 image
3 0
1
3
-1 1
3
0
2 x 2 image
Each filter
is a channel
New image
but smaller
Conv
Max
Pooling

The whole CNN
Convolution
Max Pooling
Convolution
Max Pooling
Can
repeat
many
times
A new image
The number of channels
is the number of filters
Smaller than the original
image
3 0
1
3
-1 1
3
0

The whole CNN
Fully Connected
Feedforward network
cat dog ……
Convolution
Max Pooling
Convolution
Max Pooling
Flattened
A new image
A new image

Flattening
3 0
1
3
-1 1
3
0 Flattened
3
0
1
3
-1
1
0
3
Fully Connected
Feedforward network

Only modified the network structure and input
format (vector -> 3-D tensor)
CNN in Keras
Convolution
Max Pooling
Convolution
Max Pooling
input
1 -1 -1
-1 1 -1
-1 -1 1
-1 1 -1
-1 1 -1
-1 1 -1
There are
25 3x3
filters.
…
…
Input_shape = ( 28 , 28 , 1)
1: black/white, 3: RGB
28 x 28 pixels
3 -1
-3 1
3

format (vector -> 3-D array)
CNN in Keras
Convolution
Max Pooling
Convolution
Max Pooling
Input
1 x 28 x 28
25 x 26 x 26
25 x 13 x 13
50 x 11 x 11
50 x 5 x 5

format (vector -> 3-D array)
CNN in Keras
Convolution
Max Pooling
Convolution
Max Pooling
Input
1 x 28 x 28
25 x 26 x 26
25 x 13 x 13
50 x 11 x 11
50 x 5 x 5
Flattened
1250
Fully connected feed-forward
network
Output

Possibility of implementation without
CoDiNg

Possibility of implementation with CoDiNg

Important steps in implementation of any classification
problems
Dr. Samit Ari, NIT Rourkela, Progress Report on Research Proposal titled "Development of advanced Holter monitor with extended recording and episode detection
U
n
d
e
r
s
ta
n
d
in
ga
b
o
u
t
th
es
tr
u
c
tu
r
eo
f th
ein
p
u
td
a
ta
P
r
e
-
p
r
o
c
e
s
s
in
gte
c
h
n
iq
u
e
s
if a
n
yr
e
q
u
ir
e
d
n
u
m
b
e
ro
f c
la
s
s
e
s
Id
e
n
tifyth
e
r
e
q
u
ir
e
dp
a
c
k
a
g
e
s
fo
rth
eim
p
le
m
e
n
ta
tio
n
E
x
p
e
r
im
e
n
tw
ith
d
iffe
r
e
n
tA
r
c
h
ite
c
tu
r
e
s
fo
rb
e
tte
rc
la
s
s
ific
a
tio
n
p
e
r
fo
r
m
a
n
c
e
E
ith
e
rtu
n
in
go
f
h
y
p
e
r
-
p
a
r
a
m
e
te
r
s(
o
r
)
C
h
a
n
g
in
gth
es
tr
u
c
tu
r
eo
f
th
en
e
tw
o
r
ka
r
c
h
ite
c
tu
r
e

Important steps in implementation of any classification
problems
Training and Test dataset
● We split the data into
○ Training set
○ Test set
We train the model on training set
● And evaluate the performance of the model on test set

Required packages for implementation
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, confusion_matrix
from keras import models
from keras.layers import Dense, Conv1D, Flatten,Dropout, MaxPool1D, BatchNormalization, LSTM,
Reshape,GlobalAveragePooling1D, MaxPooling1D, Input, concatenate
from keras.models import Sequential, Model
from sklearn.preprocessing import LabelEncoder, OneHotEncoder
from keras import utils
import tensorflow as tf
import pydot

Different Architectures
# Multilayer Perceptron
from keras.utils import plot_model
from keras.models import Model
from keras.layers import Input
from keras.layers import Dense
visible = Input(shape=(10,))
hidden1 = Dense(10, activation='relu')(visible)
hidden2 = Dense(20, activation='relu')(hidden1)
output = Dense(1, activation='sigmoid')(hidden3)
model = Model(inputs=visible, outputs=output)
# summarize layers
print(model.summary())
# plot graph
plot_model(model, to_file='multilayer_perceptron_graph.png')

# Convolutional Neural Network
from keras.layers import Flatten
from keras.layers.convolutional import Conv2D
from keras.layers.pooling import MaxPooling2D
visible = Input(shape=(64,64,1))
conv1 = Conv2D(32, kernel_size=4, activation='relu')(visible)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(16, kernel_size=4, activation='relu')(pool1)
flat = Flatten()(pool2)
hidden1 = Dense(10, activation='relu')(flat)
# summarize layers
# plot graph
plot_model(model, to_file='convolutional_neural_network.png')

# Recurrent Neural Network
from keras.layers.recurrent import LSTM
visible = Input(shape=(100,1))
hidden1 = LSTM(10)(visible)
# summarize layers
# plot graph
plot_model(model, to_file='recurrent_neural_network.png')

# Shared Input Layer
from keras.layers.merge import concatenate
# input layer
visible = Input(shape=(64,64,1))
# first feature extractor
flat1 = Flatten()(pool1)
# second feature extractor
# merge feature extractors
merge = concatenate([flat1, flat2])
# interpretation layer
hidden1 = Dense(10, activation='relu')(merge)
# prediction output
# summarize layers
# plot graph
plot_model(model, to_file='shared_input_layer.png')

# Multiple Inputs
from keras.layers.merge import concatenate
# first input model
visible1 = Input(shape=(64,64,1))
conv11 = Conv2D(32, kernel_size=4, activation='relu')(visible1)
# second input model
visible2 = Input(shape=(32,32,3))
conv21 = Conv2D(32, kernel_size=4, activation='relu')(visible2)
# merge input models
merge = concatenate([flat1, flat2])
# interpretation model
hidden1 = Dense(10, activation='relu')(merge)
model = Model(inputs=[visible1, visible2], outputs=output)
# summarize layers
# plot graph
plot_model(model, to_file='multiple_inputs.png')

CIFAR 10 and Convolutional Neural Network
CIFAR 10 dataset:
50,000 training images
10,000 testing images
10 categories (classes)
Accuraciesfrom different methods:
Human: ~94%
Whitening K-‐mean: 80%
……
Deep CNN: 95.5%

Deep Convolutional Neural Networks on CIFAR10
convolution2D MaxPooling2D convolution2D
MaxPooling2D
Fully-‐connected
output
Convolutional Layer: filters work on every
part of the image, therefore, they are
searching for the same feature everywhere
in the image.
Input image Convolutional output
Dropout

MaxPooling2D
Fully-‐connected
output
Convolutional output
MaxPooling MaxPooling: usually present after
the convolutional layer. It provides a
down-‐sampling of the convolutional
output
(2,2)
Dropout

MaxPooling2D
Fully-‐connected
output
Dropout
Dropout: randomly drop units along with their
connections during training. It helps to learn more
robust features by reducing complex co-‐adaptations of
units and alleviate overfitting issue as well.
Srivastava et al. Dropout: A Simple wayto Prevent Neural Networks from Overfitting. Journal of Machine Learning Research15(2014):1929-‐1958

MaxPooling2D
Fully-‐connected
output
Dropout
input
output
hidden
Fully-‐connected layer (dense): each node is fully
connected to all input nodes, each node computes
weighted sum of all input nodes. It has one-‐
dimensional structure. It helps to classify input pattern
with high-‐level features extractedby previous layers.

Why GPU Matters in Deep Learning?
vs
Running time without GPU Running time with GPU
With GPU, the running time is 733/27=27.1 times faster then the running time without GPU!!!
Again, WHY GPUs?
1. Every set of weights canbe stored as a matrix (m,n)
2.GPUs are made to do common parallel problems fast. All similar calculations are done at the
same time. This extremely boosts the performance in parallel computations.

Summary: Deep Learning
• Make it deep (many layers)
• Way more labeled data (1 million)
• A lot better computing power (GPU clusters)

Deep Learning For
Recommender
Systems

Recommender System
Image courtesy of Netflix

Recommender System
Image courtesy of Amazon

Recommender Systems: Software
tools and techniques providing
suggestions for items to be of use to
a user.
Input Data:
1. A set of users U={u1,u2,…, um}
2. A set of items V={v1,v2,…,vn}
3. The history preference ratings Rij
Output Data:
Given user u and item v
Predict the rating or preference ruv
Ricci et al. Introduction to RecommenderSystems Handbook. 2011

Various Real-Time AI
Applications

ECG Basics
 Electrocardiogram (ECG) shown in Fig.1 is an advanced
technique that is used as a diagnostic tool for finding
abnormalities in the heart.
 ECG signal is widely used as a basic tool for the
detection and diagnosis of heart disorders, ECG is the
record of alteration of bioelectric potential concerning
time as the human heartbeats.
 Early detection of heart diseases can extend life and
improves the quality of living through proper treatment.
It is very difficult for doctors to analyze long ECG
records in short time duration.
 Therefore, a strong and robust computer-aided diagnosis
(CAD) system is required for early detection of cardiac
abnormalities .
ECG Signal

Details about the data
 Each row in ECG signal contains 128 samples (one full beat)
 Last row represents label of that particular beat type
 Total Number of classes in ECG are 4 as shown in Figure. 2
Different types of ECG beats

ECG Beat Classification using Deep Learning
07-July-22
of ECE, NIT Rourkela
1

ECG and Disorders
07-July-22
3
Some of the diseases diagnosed by ECG are:
 Myocardial Ischemia/Infarction.
 Arrhythmias.
 Hypertrophy and enlargement of heart.
 Conduction Blocks.
 Pre-excitation Syndromes.
 Other cardiac disorders.

2D CONVOLUTIONAL NEURAL NETWORK (CNN)
07-July-22
One of the most popular deep neural networks types is convolutional neural networks
(CNN or ConvNet). A CNN convolves learned features with input data and uses 2D
convolutional layers, making this architecture well-suited to processing 2D data, such
as images.
CNNs learn to detect different features of an image using tens or hundreds of hidden
layers. Every hidden layer increases the complexity of the learned image features.
For example, the first hidden layer could learn how to detect edges, and the last
learns how to detect more complex shapes specifically scatered to the shape of the
object we are trying to recognize.
07-July-22 Dr. Samit Ari, dept. of ECE, NIT Rourkela 24

1D CONVOLUTIONAL NEURAL NETWORK(CNN)
07-July-22
5
• The conventional deep CNNs presented in the previous slide
are designed to operate exclusively on 2D data such as images
and videos. This is why they are often referred to as, ‘‘2D
CNNs”. As an alternative, a modified version of 2D CNN scaled
1D CNNs have recently been developed
• There is a significant difference in terms of computational
complexities of 1D and 2D convolutions, i.e., an image with
NxN dimensions convolve with KxK kernel will have a
computational complexity ~O(N2K2) while in the
corresponding 1D convolution (with the same dimensions, N
and K) this is ~O(NK). This means that under equivalent
conditions (same configuration, network and hyper
parameters) the computational complexity of a 1D CNN is
significantly lower than the 2DCNN.

1D Convolution Operation
07-July-22
3

Database
 The MIT-BIH arrhythmia, and real-time ECG database are
used to estimate the performance of the proposed technique.
MITBIH arrhythmia database consists of normal and
abnormal heart beats.
 This database is observed as the benchmark for cardiac beat
detection and classification.
 An ECG database was made in real-time with the
collaboration of Ispat General Hospital (IGH), Rourkela using
the EDAN SE-1010 PC ECG acquisition system operating at
1000 samples per second with a frequency response of 0.05–
150 Hz.
 It consists of 34 ECG recordings from a group of 17
individuals which comprises of 13 males and four females
within the age limit of 25–50.
07-July-22
6
Data Collection

Database
07-July-22
7
AAMI Standard label Label
Non-ectopic beat (N)
Normal beat (n)
Left bundle branch block beat (L)
Right bundle branch block beat (R)
Atrial escape beat (e)
Nodal (junctional) escape beat (j)
Supraventricular ectopic beat (S)
Atrial premature beat (A)
Aberrated atrial premature beat (a)
Nodal (junctional) premature beat (J)
Supraventricular premature or ectopic beat (S)
Ventricular ectopic beat (V) Premature ventricular contraction (PVC)
Ventricular escape beat (E)
Fusion beat (F) Fusion of ventricular and normal beat (F)
Unknown beat (Q) Unknown beat (Q)

Database
07-July-22
8
 The proposed method utilizes a limited amount of
training data for ECG beat classification. In this
work, 8672 segmented ECG beats are considered
a limited training dataset.
 The total dataset is divided into training and
testing, which helps in modelling and evaluating
the model. Training and testing data (DS1 and
DS2) consists of 8672, 49,564 ECG beat images.
 Reported deep learning beat classification
techniques in literature utilizes a large dataset for
training compared to the proposed system

Methodology I
07-July-22
9
Block Diagram of proposed technique for ECG beat
classification.
The following steps are followed to
detect ECG beat type.
 Pre-processing & beat segmentation
 Time-frequency representation using
Stockwell transform (ST)
 ECG beat classification
ECG Signal
S-Transform Images
(Spectrograms)
ECG beat
Pre-processing Stage
(Beat segmentation & Transformation)
Input Data
Deep
Resnet
Train & Validation
data set
Detected Arrhythmias
Test data set
Classification Stage
*Allam Jaya Prakash, Samantray Saunak, and Samit Ari., “SpEC: A system for patient-specific ECG beat classification using the deep
residual network.,” Bio-cybernetics and Biomedical Engineering, 40(4), pp.1446-1457 (2020).

ECG beat classification using 2D ResNet model
07-July-22
0
* Allam Jaya Prakash, Samantray Saunak, and Samit Ari., “SpEC: A system for patient-specific ECG beat classification using a deep residual network.,” Bio-cybernetics and
Biomedical Engineering, 40(4), pp.1446-1457 (2020).

Parameter tuning of the proposed deep learning
model
07-July-22
1
 The number of residual paths, learning rate, and batch size is the essential parameters. The optimization of
these three parameters plays a vital role in the system classification performance.
 A series of experiments are conducted to optimize these three parameters. Initially, the network is tested with
a different number of residual paths while keeping the learning rate, and batch size constant.
 Finally, residual paths, learning rate, and batch size are 12, 0.001, and 32 sequentially observed as optimized
parameters of the ResNet, where the model reached the highest overall accuracy.
Overall accuracy of the ResNet with the learning rate of 0.3, and batch size of 256.

Parameter tuning of the proposed deep learning model
07-July-22
2
The overall accuracy of the ResNet with
a batch size of 256.
The overall accuracy of the ResNet with
the learning rate of 0.001.

Performance Analysis Metrics
07-July-22
3
 The following parameters were used to evaluate the performance of the proposed method
 Where, TP: True Positive, TN: True Negative
FN
TP
TP
Sen
y
Sensitivit


)
(
FN
TN
FP
TP
TN
TP
Acc
Accuracy





)
(
FP
TN
TN
Spe
y
Specificit


)
(
FP
TP
TP
Ppr
edictivity
Positive


)
(
Pr

Experimental Results
07-July-22
4
 The proposed system is evaluated on a standard MIT-BIH database

Experimental Results(contd.)
07-July-22
5
 The proposed system is evaluated on acquired real-time ECG database

Performance comparison of proposed method and earlier reported
methods
07-July-22
6
Comparative Performances of the proposed method with earlier reported literature methods for ECG beat
classification

Performance comparison of the proposed method and earlier
reported methods
07-July-22
7
Performance comparison of discrete ROC for the proposed
method and earlier reported literature methods

Conclusion I
07-July-22
8
 This work introduces a novel system termed SpEC, based on ST and 2D-ResNet model for patient-specific ECG beat
classification with a limited amount of training dataset.
 In the present work, the ST-based 2D ResNet model is proposed to make use of frequency invariant amplitude
response and progressive resolution of the ECG beats.
 To utilize gradient information efficiently, and fine-tune the weights with a limited training dataset 2D ResNet with
12 residual paths is proposed in this work.
 The proposed SpEC system which does not utilize handcrafted features enjoys the benefits of the ST and 2D-ResNet
model to detect the ECG beats automatically.
 Experiments are conducted on the MIT-BIH arrhythmia database and real-time acquired ECG dataset where the
proposed SpEC system achieves better performance compared to the state-of-art techniques for the detection of all
five different beats including important beats like S and V beats as these beats are clinically crucial for early
detection of cardiac abnormalities.

Technique based on Deep learning: Methodology II
07-July-22
9
Block Diagram of proposed technique for ECG beat
classification using EMD and Deep Learning.
 In this system, pre-processing and
classification are the two crucial stages for
de-noising and classification of the ECG beat.
 R-peak detection is helpful in finding the
specific QRS complex to segment the
whole ECG signal into individual ECG
beats.
 The EMD technique deconstructs the raw ECG
beats into intrinsic mode functions (IMFs)
components, and significant IMF components
are added to reconstruct the noise-free ECG
beat.
 These resultant beats are utilized for training
and testing the deep learning model.

Methodology II
07-July-22
0
 EMD is a well-known method to analyze non-stationary data like ECG, EEG, etc. Any non-stationary data can be
decomposed into finite IMF components.
 EMD is a powerful technique that decomposes the signal without distorting the time domain features.
 The EMD is able to produce a different number of IMFs for applied signal. These IMFs consist of individual parts,
which when added up reproduce the original signal.
 The original ECG signal ECG(t) represented with IMF components 𝑆𝑛(𝑡), and residue 𝑟𝑒𝑠(𝑡)are as follows:
𝐸𝐶𝐺 𝑡 =
𝑛
𝑆𝑛 𝑡 + 𝑟𝑒𝑠(𝑡)
 The following conditions are used to identify significant IMF components: (i) At any point, the mean value of the
envelope formed by the local maxima and the envelope defined by the local minima is zero, and (ii) the number
of extrema and zero crossings must either equal or differ by one.

Methodology II (contd.)
07-July-22
1
N, S, V, F, and Q beats and IMF components of #100, #102, #104, #105 and #208
ECG records (Number of samples is represented along the X-axis whereas
amplitude is shown along the Y-axis).

07-July-22
2
Detailed dataset description used in this work
 The database is prepared with 58,236 noise-
free beats.
 A typical common training dataset is prepared
randomly with 245 ECG beats, including 75 N,
75 S, 75V beats, 13 types of F beats, and 7 type
Q beats from the 100 series ECG records.
 In addition to this 245 beats data, 5min
patient-specific data from 200 series ECG
records are also added to the training data.
 The remaining 25 min data in 200 series
records are used to test the network, which is
entirely new to the network.
 It is concluded that the training set (TR1), and
testing set (TS2) contain 8654, and 49371 beats
respectively.

07-July-22
3
Deep learning model for ECG beat detection
 Datasets are prepared individually to train the
three CNN blocks of the proposed network
from the segmented ECG beat data.
 The whole segmented individual ECG beat data
of size (1x256) is used to train the first CNN
block, i.e., -128 to +128.
 The second CNN block is trained with the size
of (1x128) by considering -128 (left side) to the
R-peak location, i.e., -128 to 0.
 The third block of the CNN is trained with the
R-peak location to +128 (Right side), i.e., 0 to
+128.
 All three datasets are applied simultaneously to
the three parallel blocks of the CNN to extract
the in-detail features of the ECG beat.

07-July-22
3
Model Kernel Size Stride Number of Filters
Model-1 36 with a length of 7
sampling points
8 256
sampling points
4 128
sampling points
4 128
sampling points
2 64
sampling points
4 128
sampling points
2 64
 The 1D convolutional layer creates a convolution kernel that is convolved with the input layer over a single dimension
to produce a tensor of output. The kernel size was set to 36 in the first model and decreased to 12 in the subsequent
model, in order to reduce computational costs

07-July-22
4
Different layers in the individual model
(Model 1-6 in the above architecture)
 Various features from ECG beat are extracted using
different convolutional layers in the architecture.
 The first branch is mostly concentrated on the
morphological nature of the QRS complex of the ECG
beat.
 The second branch has extracted the features of the P-
wave.
 The third branch helps in finding the nature of the T-
wave.
 Specific feature extraction from the P-wave, and T-
wave is helpful in improving the detection of S, and V
beats, which are clinically significant.
 The three branches are individually extracted features
from the input data.

07-July-22
5
 The performance of any deep learning model mainly
depends upon the selection of optimized parameters.
 In this model, the learning rate (𝜂) and batch size are
the crucial parameters that affect the system’s
performance. A progression of trials is conducted to
streamline these two parameters.
 At first, the network performance is tested with various
learning rates while keeping the batch size constant.
 The network performance is tested with the different
batch sizes by maintaining the learning rate constant,
and 32 batch size is the minimum requirement for the
best performance.
Tuning of learning rate for the proposed network
Tuning of batch size rate for the proposed network

Performance of the Methodology II
07-July-22
6
Confusion Matrix of the Methodology 2
Performance of the Methodology 2
* Allam Jaya Prakash, Samantray Saunak, and Samit Ari., “Patient-Specific ECG Beat Classification using
EMD and Deep Learning-based Technique.,” Advanced Methods in Biomedical Signal Processing and
Analysis, Elsevier (2022).

Performance comparison of the methodology II
07-July-22
7
Performance comparison of the proposed method with literature
Comparison of the proposed method ROC with earlier reported techniques.

Conclusion of the Methodology II
 The proposed method is used to classify five different types of ECG beats N, S, V, F, and Q, which followed
the AAMI standard.
 In this work, the suggested approach for beat detection consisted of two steps: pre-processing and
classification. In pre-processing, applying EMD on ECG signal to extract significant IMF components are
crucial to extracting the relevant beat information from the signal.
 These significant IMF components are beneficial in removing high-frequency noise components. To extract
the morphological information about the QRS complex, P, and T-wave, three different datasets are
prepared in this work separately.
 These three datasets are processed individually through three parallel CNN architectures. The experimental
results show that the proposed EMD-based deep learning successfully identified ECG beats by extracting
exact morphology information from the beat segments.
 The performance parameters of the proposed method show it provides better performance than the
earlier reported techniques.

Technique based on Deep learning: Methodology III
07-July-22
 In this system, pre-processing and classification
are the two crucial stages for de-noising and
classification of the ECG beat.
 The EWT technique decompose the raw ECG
beats into modes, and significant modes are
added to reconstruct the noise-free ECG beat.
 EWT is an advanced technique to decompose
the signal better than empirical mode
decomposition (EMD).
 After successful decomposition of ECG signal,
the specific low-frequency modes are added
and form a noise-free signal. The resultant
noise-free.
 ECG beats are used as input to the customized
deep learning network for further
classification.

Methodology III
07-July-22
of ECE, NIT Rourkela
0
N, S beat and its corresponding EWT modes for ECG record number has #124 is represented as an example respectively.
Empirical Wavelet Transform (EWT)
ECG beat and Corresponding EWT modes Spectrum Partitioning of ECG beat Comparison of original beat with denoised beat
ECG beat and Corresponding EWT modes Spectrum Partitioning of ECG beat Comparison of original beat with denoised beat

Methodology III
Deep learning model for ECG beat detection
 The deep learning architecture shown in Figure consists of three serial
CNN blocks. Each CNN block consists of one convolutional layer, batch
normalization, and activation.
 In the suggested approach, three deep convolutional blocks are
employed since they exhibit a proper balance between computational
efficiency and the validity of the findings.
 The kernel in the 1D convolutional layer is convolved with the single
dimension input vector to produce the output tensor.
 The kernel size of 40 is used in the initial layer, but gradually it is
decreased to 4, which reduces the computational cost of the network.
 The input was managed through batch standardization. It was used to
boost performance and stabilize the learning process of the deep
neural network after each convolutional layer and before pooling.

Parameters of the Network
A detailed description of the proposed network

Performance of the Methodology III
Confusion Matrix
* Allam Jaya Prakash, Samantray Saunak, and Samit Ari., “Empirical Wavelet Transform and Deep Learning-
based Technique for ECG Beat Detection.,” Advanced Methods in Biomedical Signal Processing and
Analysis, Elsevier (2022).

Performance of the Methodology III
Performance comparison of the proposed method with existing techniques
 The proposed automatic ECG beat classification system performance is compared with the existing techniques
in the literature. The proposed method effectively identifies applied ECG beats in a patient-specific way with a
performance accuracy of 99.75% which is better than the earlier techniques.
 The major advantages of the proposed deep learning-based classification system compared to the state-of-
the-art techniques are as follows:
 the EWT-based pre-processing technique is very helpful in removing low-frequency and
high-frequency noise components from the ECG signal.
 automatic feature extraction
 less computation time for the prediction
 model complexity is very less
 high accuracy in detection in S, and V beats.

[1] Develop a cutting-edge smart farming system that integrates drone technology, IoT, and AI-enabled spectral
imaging to enhance crop health monitoring and management.
[2] Create a real-time analysis platform that utilizes advanced algorithms to provide accurate and reliable data on
crop health, insect detection, and spraying of pesticide, enabling farmers to make data-driven decisions.
[3] Enable precision agriculture techniques by providing farmers with valuable insights on crop health, soil quality,
and other relevant factors, leading to improved crop yields and reduced costs.
[4] Facilitate sustainable agriculture practices by reducing the use of pesticides and fertilizers, and minimizing
waste through targeted application of resources.
[5] Establish a scalable and adaptable system that can be customized to fit the specific needs of individual farmers
and agricultural operations.
Objectives:

AI_ML_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaPresentation.pptx

Recommended

Recommended

More Related Content

Similar to AI_ML_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaPresentation.pptx

Similar to AI_ML_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaPresentation.pptx (20)

Recently uploaded

Recently uploaded (20)

AI_ML_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaPresentation.pptx