This Deep Learning presentation will help you in understanding what is Deep Learning, why do we need Deep learning, what is neural network, applications of Deep Learning, what is perceptron, implementing logic gates using perceptron, types of neural networks. At the end of the video, you will get introduced to TensorFlow along with a usecase implementation on recognizing hand-written digits. Deep Learning is inspired by the integral function of the human brain specific to artificial neural networks. These networks, which represent the decision-making process of the brain, use complex algorithms that process data in a non-linear way, learning in an unsupervised manner to make choices based on the input. Deep Learning, on the other hand, uses advanced computing power and special type of neural networks and applies them to large amounts of data to learn, understand, and identify complicated patterns. W will also understand neural networks and how they work in this Deep Learning tutorial video. This Deep Learning tutorial is ideal for professionals with beginner to intermediate level of experience. Now, let us dive deep into this topic and understand what Deep Learning actually is. Below topics are explained in this Deep Learning presentation: 1. What is Deep Learning? 2. Why do we need Deep Learning? 3. What is Neural network? 4. What is Perceptron? 5. Implementing logic gates using Perceptron 6. Types of Neural networks 7. Applications of Deep Learning 8. Working of Neural network 9. Introduction to TensorFlow 10. Use case implementation using TensorFlow Simplilearn’s Deep Learning course will transform you into an expert in deep learning techniques using TensorFlow, the open-source software library designed to conduct machine learning & deep neural network research. With our deep learning course, you’ll master deep learning and TensorFlow concepts, learn to implement algorithms, build artificial neural networks and traverse layers of data abstraction to understand the power of data and prepare you for your new role as deep learning scientist. Why Deep Learning? It is one of the most popular software platforms used for deep learning and contains powerful tools to help you build and implement artificial neural networks. Advancements in deep learning are being seen in smartphone applications, creating efficiencies in the power grid, driving advancements in healthcare, improving agricultural yields, and helping us find solutions to climate change. There is booming demand for skilled deep learning engineers across a wide range of industries, making this deep learning course with TensorFlow training well-suited for professionals at the intermediate to advanced level of experience. We recommend this deep learning online course particularly for the following professionals: 1. Software engineers 2. Data scientists 3. Data analysts 4. Statisticians with an interest in deep learning

2. Geolocation with Neural Network
How does a neural network recognize the location of an image?
Artificial Neural Network New image of a placeImages of millions of places Identifies the place

3. Geolocation with Neural Network
Pixels of millions of images Feed to Neural Network Recognizes landmarks,
landscapes, architectural
styles, road signs, etc
New image of a place Identifies the place
How does a neural network recognize the location of an image?

4. What’s in it for you?
Types of Neural Networks
Applications of Deep Learning
Working of Neural Network
Introduction to TensorFlow
Use case implementation using TensorFlow
What is Deep Learning?
What is a Neural Network?
Implementing Logic Gates using Perceptron
What is a Perceptron?
Why do we need Deep Learning?

5. What is Deep Learning?

6. What is Deep Learning?
Deep Learning is a subset of Machine Learning that has networks which are capable of learning from data
that is unstructured or unlabeled and works similar to the functioning of a human brain.
Artificial Intelligence
Machine
Learning
Deep
Learning
Ability of a machine to imitate
intelligent human behavior
Application of AI that allows a system
to automatically learn and improve
from experience
Application of Machine Learning that
uses complex algorithms and deep
neural nets to train a model

7. Why do we need Deep Learning?
Machine Learning works
only with large sets of
structured data, while
Deep Learning can work
with both structured and
unstructured data
Deep Learning
algorithms can perform
complex operations
easily while Machine
Learning Algorithms
cannot
Machine Learning
algorithms use labelled
sample data to extract
patterns, while Deep
Learning accepts large
volumes of data as
input, analyze the input
to extract features out
of an object
Performance of
Machine Learning
algorithms decreases
as the amount of data
increase, so to
maintain the
performance of the
model we need Deep
Learning
Works with unstructured data Handle complex operations Feature Extraction Achieve best performance

8. What is a Neural Network?

9. What is a Neural Network?
Neuron
X1
X2
Xn
Y
Output
w1
w2
w n
Input 1
Input 2
Input n
Neural Networks are modeled after biological neural networks that allow computers to learn and interact
like humans do. It has interconnected neurons with dendrites that receive inputs, and based on those
inputs, it produces an electric signal i.e. output through the axon.
Biological Neuron

10. What is a Neural Network?
Neuron
X1
X2
Xn
Y
Output
w1
w2
w n
Input 1
Input 2
Input n
Neural Network has interconnected neurons that receive some inputs, processes those inputs in layers to
produce the output.
Artificial Neuron

11. Biological Neurons Vs Artificial Neuron
Dendrites -------> Inputs
Neuron
X1
X2
Xn
Y
Output
w1
w2
w n
Dendrites Input 1
Input 2
Input n
Inputs

12. Biological Neurons Vs Artificial Neuron
Cell Nucleus
Neuron
Input 1
Input 2
Input n
Y
Output
w1
w2
w n
Nodes
X1
X2
Xn
Cell Nucleus -------> Nodes

13. X1
X2
Xn
Biological Neurons Vs Artificial Neuron
Synapse
Neuron
Input 1
Input 2
Input n
Y
Output
Weights
w1
w2
wn
Synapse -------> Weights

14. Biological Neurons Vs Artificial Neuron
X1
X2
Xn
Neuron
Input 1
Input 2
Input n
w1
w2
wn
Axon
Axon -------> Output
Y
Output

15. What is a Perceptron?

16. What is a Perceptron?
A Perceptron is the basic part of a neural network. It represents a single neuron of a human brain and is
used for binary classifiers.
X1
X2
Xn
Input 1
Input 2
Input n
w1
w2
wn
Y Output
Net Input Function Activation Function
ERROR

17. Perceptron Learning Rule
w
X1
X2
Xn
Input 1
Input 2
Input n
w2
Y Output
Activation Function
ERROR
w1
n
Net Input
Function
Inputs are multiplied with the weights and
a summation is performed, plus a bias is
added
Bias
i=1
n
w x + b
i i*
Step 1

18. Perceptron Learning Rule
w
X1
X2
Xn
Input 1
Input 2
Input n
w2
Y Output
Activation Function
ERROR
w1
n
Net Input
Function
The weighted sum of inputs is passed to
an activation function to determine if a
neuron will fire or not
Step 2

19. Perceptron Learning Rule
w
X1
X2
Xn
Input 1
Input 2
Input n
w2
Y Output
Activation Function
ERROR
w1
n
Net Input
Function
Perceptron receives input signals and if
the sum of the input signals exceeds a
certain threshold value, it either
outputs a signal or does not return an
output
Output =
0, if
i=1
n
w x + b < threshold
i i*
1, if
i=1
n
w x + b >= thresholdi i*
Step 3

20. Perceptron Learning Rule
w
X1
X2
Xn
Input 1
Input 2
Input n
w2
Y Output
Activation Function
ERROR
w1
n
Net Input
Function
The error in the output is backpropagated
and weights are updated to minimize the
error
Step 4

21. Perceptron Learning Rule
w
X1
X2
Xn
Input 1
Input 2
Input n
w2
Y Output
Activation Function
ERROR
w1
n
Net Input
Function
Repeat steps 1 to 4 in order to generate
the desired output

22. Implementing Logic Gates using Perceptron
In 1943, neuroscientist Warren McCulloch and logician Walter Pitts published a paper showing how neurons could implement
the three logical functions (AND, OR, XOR). The neurons they used were simple threshold neurons which they named it
Perceptron.
Warren McCulloch Walter Pitts

23. Implementing Logic Gates using Perceptron
A logic gate is the basic building block of a digital circuit. Most logic gates have 2 inputs and 1 output. At
any given moment, every terminal is in one of the two binary conditions low (0) or high (1), represented by
different voltage levels.
Popular Logic Gates
AND
OR
NAND
NOR
NOT
XOR

24. Implementing AND Gate
Using the Logic Gates, Neural Networks can learn on their own without having to manually code the logic
Input (A) Input (B) Output
0 0 0
0 1 0
1 0 0
1 1 1
AND
The output of Perceptron is true or positive if both
the inputs are true, else false

25. Implementing AND Gate
A and B are input neurons. The red neuron is our output neuron. The threshold value is 1. If the sum of the
weighted input neurons is greater than the threshold, the output is 1 else 0.
Input (A) Input (B) Output
0.7*0 0.7*0 0
0.7*0 0.7*1 0.7
0.7*1 0.7*0 0.7
0.7*1 0.7*1 1.4
A
B
0.7
0.7
= 1
This exceeds
the threshold, so
output is 1
We have designed a neuron which implements a logical
AND gate
w x1 1* w x
2 2* w x +1 1* w x
2 2*

26. Implementing OR Gate
Using the Logic Gates, Neural Networks can learn on their own without having to manually code the logic
Input (A) Input (B) Output
0 0 0
0 1 1
1 0 1
1 1 1
OR
The output of Perceptron is true or positive if
either of the inputs are true, else false

27. Implementing OR Gate
A and B are input neurons. The red neuron is our output neuron. The threshold value is 1. If the sum of the
weighted input neurons is greater than the threshold, the output is 1 else 0.
Input (A) Input (B) Output
1.2*0 1.2*0 0
1.2*0 1.2*1 1.2
1.2*1 1.2*0 1.2
1.2*1 1.2*1 2.4
A
B
1.2
1.2
= 1
These exceed
the threshold, so
output is 1
w x1 1* w x
2 2* w x +1 1* w x
2 2*
We have designed a neuron which implements a logical
OR gate

28. Implementing XOR Gate
Using the Logic Gates, Neural Networks can learn on their own without having to manually code the logic
Input (A) Input (B) Output
0 0 0
0 1 1
1 0 1
1 1 0
XOR
The output of Perceptron is true or positive if one
of the inputs is true, else false

29. Implementing XOR Gate
XOR gate requires an intermediate hidden layer to achieve the logic of a XOR gate
XOR
Implementation
5
X
X
h
h
O
1
2
3
4
w13
w14
w23
w24
w45
w35
5
• X1 and X2 are inputs and h3 and h4 are the
hidden layers. B3, B4 and B5 are the biases
for h3, h4 and O5.
• h3 = sigmoid (X1*w13 + X2*w23 – B3)
• h4 = sigmoid (X2*w14 + X2*w24 – B4)
• O5 = sigmoid (h3*w35 + h4*w45 – B5)

30. Implementing XOR Gate
XOR gate requires an intermediate hidden layer to achieve the logic of a XOR gate
5
X
X
h
h
O
1
2
3
4
20
-20
20
-20
20
20
5
Sigmoid(20*0+20*0-10) 0
Sigmoid(20*1+20*1-10) 1
Sigmoid(20*0+20*1-10) 1
Sigmoid(20*1+20*0-10) 1
~~
~~
~~
~~
Sigmoid(-20*0-20*0+30) 1
Sigmoid(-20*1-20*1+30) 0
Sigmoid(-20*0-20*1+30) 1
Sigmoid(-20*1-20*0+30) 1
~~
~~
~~
~~
b= -10
Sigmoid(20*0+20*1-30) 0
Sigmoid(20*1+20*0-30) 0
Sigmoid(20*1+20*1-30) 1
Sigmoid(20*1+20*1-30) 1
~~
~~
~~
~~
h3 h4
Output
These exceed the
threshold, so
output is 1
b= 30
b= -30
= 1
= 1
= 1

31. Types of Neural Network

32. Types of Neural Networks
Pattern
Recognition
ANN CNN RNN DNN DBN
Image Processing Acoustic
Modeling
Recursive
Neural Network
Cancer Detection
Neural Networks are mainly classified into 5 types
Artificial Neural
Network
Convolution
Neural Network
Speech Recognition
Deep Neural
Network
Deep Belief
Network

33. Applications of Deep Learning

34. Applications of Deep Learning
Deep Learning allows us to build machines that can play games
Playing Games

35. Applications of Deep Learning
Composing Music
Deep Neural Nets can be used to produce music by making computers
learn the patterns in a composition

36. Applications of Deep Learning
Autonomous Driving Cars
Distinguishes different types of objects, people, road signs and drives
without human intervention

37. Applications of Deep Learning
Building Robots
Deep Learning is used to train robots to perform human tasks

38. Applications of Deep Learning
Medical Diagnosis
Deep Neural Nets are used to identify suspicious lesions and nodules in
lung cancer patients

39. Working of Neural Network

40. Working of Neural Network
Let’s find out how an Artificial Neural Network can be used to identify different shapes like Squares, Circles and Triangles
How to identify
various shapes
using a neural
network?
Feed the shapes to a neural network as
input
Artificial Neural Network

41. Working of Neural Network
Artificial Neural Network
28
28
Strengths
28 28 = 784*
• 28*28 pixels of the input image is
taken as input i.e. 784 neurons
Lets consider the shape of a square
Shape of a Square

42. Working of Neural Network
28
28
Strengths
• 28*28 pixels of the input image is
taken as input i.e. 784 neurons
28 28 = 784*
0.78 Activation
• Each neuron holds a number called
Activation that represents grayscale
value of the corresponding pixel ranging
from 0 to 1. 1 for white pixel and 0 for
black pixel
Lets consider the shape of a square
Artificial Neural Network

43. Working of Neural Network
Lets consider the shape of a square
28
28
Strengths
• 28*28 pixels of the input image is
taken as input i.e. 784 neurons
• Each neuron is lit up when its
activation is close to 1
28 28 = 784*
• Each neuron holds a number called
Activation that represents grayscale
value of the corresponding pixel ranging
from 0 to 1. 1 for white pixel and 0 for
black pixel
Artificial Neural Network

44. Working of Neural Network

45. Working of Neural Network

46. Working of Neural Network

47. Working of Neural Network
X1
X2
Xn
w1
w2
wn
Step 1
i=1
n
w x + b
i i*
Activation Function
Step 2
i=1
n
w x + b
ii*( )
Lets find out how an Artificial Neural Network can be used to identify different shapes
Identifies different shapes
Squares Circles
Triangles
Pixels of images
fed as input

48. Working of Neural Network
Lets find out how an Artificial Neural Network can be used to identify different shapes
X1
X2
Xn
Y
W1
W2
W3
.
.
.
.
.
.
.
.
.
Wn
W22
W23
W33
.
.
.
.
.
.
.
.
.
Wn2
W11
W21
W31
.
.
.
.
.
.
.
.
.
Wn1
Apply the weights
Identifies different shapes
Squares Circles
Triangles^
Pixels of images
fed as input

49. Working of Neural Network
Lets find out how an Artificial Neural Network can be used to identify different shapes
X1
X2
Xn
Y^
Apply the activation functions
Identifies different shapes
Squares Circles
Triangles

50. Working of Neural Network
Lets find out how an Artificial Neural Network can be used to identify different shapes
X1
X2
Xn
Y^
Compare it with
actual output
Predicted Output
Actual
Output
Y

51. Working of Neural Network
Applying the cost function to minimize the difference between predicted and actual output using gradient descent algorithm
Predicted Output
X1
X2
Xn
Y^
Actual
Output
Compare it with
actual output
Y
Cost Function: C = ½( Y – Y )
2

52. Working of Neural Network
Neural Networks use Backpropagation method along with improve the performance of the Neural Net. A cost function is used
to reduce the error rate between predicted and actual output.
X
X
X
Y^
Cost Function: C = ½( Y – Y )
2
1
2
n
Y

53. Cost Function
The Cost value is the difference between the neural nets predicted output and the actual output from a set of
labelled training data. The least cost value is obtained by making adjustments to the weights and biases
iteratively throughout the training process.
X
X
Xn
Y Y^
W1
W2
W3
.
.
.
.
.
.
.
.
.
Wn
W22
W23
W33
.
.
.
.
.
.
.
.
.
Wn2
W11
W21
W31
.
.
.
.
.
.
.
.
.
Wn1
1
2
Cost Function: C = ½( Y – Y )
2

54. Gradient Descent
Slower Faster
The person will take more time to reach the base of the mountain if the
slope is gentle and will come down faster if the slope is steep.
Likewise, a Neural Net will train slowly if the Gradient is small and it will
train quickly if the Gradient is large.
Gradient Descent is an optimization algorithm for finding the minimum of a function
A person trying to reach the base of a
mountain

55. Gradient Descent
Gradient Descent is an optimization algorithm for finding the minimum of a functionGradient Descent is an optimization algorithm for finding the minimum of a function
C = ½( Y – Y )^ 2
C
W
C
W
Best
Gradient
Global Minimum
cost
Initial Weight

56. Stochastic Gradient Descent
C
W
Best
Local Minimum Global Minimum
It does not require the
cost function to be
convex
Takes the rows one by
one, runs the neural
network and then adjusts
the weights
Helps you avoid the
local minimum as it
performs 1 iteration at a
time

57. Deep Learning Platforms
Torch
Keras
4 main platforms
TensorFlow
DeepLearning4J
(java)

58. Introduction to TensorFlow

59. Introduction to TensorFlow
• TensorFlow is a Deep Learning tool to define and run computations involving tensors.
• A tensor is a generalization of vectors and matrices to potentially higher dimensions.
• The array of data passed in each layer of nodes is known as Tensor.
a
m
k
q
d
2
4
8
1
1
9
3
2
5
4
4
6
6
3
3
7
8
2
9
5
Tensor of Dimensions[5] Tensor of Dimensions[5,4]Tensor of Dimension[3,3,3]

60. Introduction to TensorFlow
• TensorFlow is a Deep Learning tool to define and run computations involving tensors.
• A tensor is a generalization of vectors and matrices to potentially higher dimensions.
• The array of data passed in each layer of nodes is known as Tensor.
1.2 3.5 2.2…
4.7 10 4.5…
2.2 7.8 8.1…
3.6 2.5 4.5…
Arrays of data with different
dimensions and ranks go as
input to the network

61. Tensor Ranks
m=V=[1,2,3],[4,5,6]
v=[10,20,30]
t=[[[1],[2],[3]],[[4],[5],[6]],[[7],[8],[9]]]
s= [107]
Tensor of Rank 0 Tensor of Rank 1
Tensor of Rank 2Tensor of Rank 3

62. Introduction to TensorFlow
Open source software
library developed by
Google
Most popular library in
Deep Learning
Can run on either CPU
or GPU Can create data flow
graphs that have nodes
and edges
Used for Machine
Learning applications
such as Neural
Networks

63. Use case Implementation using TensorFlow
Lets build a neural network to identify hand written digits using MNIST Database.
Hand written digits
from MNIST Database
Modified National Institute of Standards and Technology
Database
Has a collection of 70,000 handwritten digits
Digit labels identify each of the digits from 0 to 9

64. Use case Implementation using TensorFlow
Lets build a neural network to identify hand written digits using MNIST Database.
Hand written digits
from MNIST Database
Pixels of digits fed as
input to the network
New image of a digit Identifies the digit
That’s
digit 3

65. Softmax Function
…………….
0 1 2 9
28*28=784 pixels
…………….
Softmax (Ln) =
Ln
e
e L

66. Use case Implementation using TensorFlow
1. Import MNIST Data using TensorFlow
2. Check the type of Dataset

67. Use case Implementation using TensorFlow
3. Array of Training images
4. Number of images for Training, Testing and Validation

68. Use case Implementation using TensorFlow
5. Visualizing the Data

69. Use case Implementation using TensorFlow
6. Maximum and minimum value of the pixels in the image

70. Use case Implementation using TensorFlow
7. Create the Model

71. Use case Implementation using TensorFlow
8. Create the Session
9. Evaluate the Trained model on Test data