Artificial Neural Network

1
 Artificial neural network
◼ Artificial neural networks are
computing systems inspired
by the biological neural
networks that is inspired by
the way biological nervous
systems, such as brain,
process information

2
 Artificial neural network
◼ Biological neurons have dendrites to receive signals, a cell
body to process them, and an axon to send signals out to other
neurons.
◼ Artificial neurons have a number of input channels, a
processing stage, and one output that can fan out to multiple
other artificial neurons.

3
Artificial neural network

4
 Artificial neural network
◼ Rosenblatt proposed the earliest
algorithm of artificial neural
networks called perceptron

5
 Artificial neural network
◼ The linear activation function determines the output of the unit

6
 Perceptron Algorithm
◼ Components and Notation

7
 Perceptron Algorithm
◼ Steps of the Perceptron Learning Algorithm

8
 Perceptron Algorithm
◼ Steps of the Perceptron Learning Algorithm

9
 Perceptron Algorithm
◼ Steps of the Perceptron Learning Algorithm

10
 Perceptron Learning Example
◼ Perceptron Learning Example

11
 Perceptron Learning Example
◼ Perceptron Learning Example

12
 Perceptron Learning Example
◼ Perceptron Learning Example

13
 Perceptron Learning Example
◼ Perceptron Learning Example

14
 Perceptron Learning Example
◼ Perceptron Learning Example

15
 Perceptron Learning Example
◼ Perceptron Learning Example

16
 Perceptron Learning Example
◼ Perceptron Learning Example

17
 Perceptron Learning Example
◼ Perceptron Learning Example

18
 Perceptron Learning Example
◼ Perceptron Learning Example

19
 Perceptron Learning Example
◼ Perceptron Learning Example

20
 Perceptron Learning Example
◼ Perceptron Learning Example

21
 Perceptron Learning Example
◼ Researchers discovered that
Perceptron cannot approximate
many nonlinear decision functions,
for example, the XOR problem

22
 Artificial neural network
◼ Researchers found a solution to that
problem by stacking multiple layers
of linear classifiers called
multilayer perceptron to
approximate nonlinear decision
functions.

23
 Artificial neural network
◼ Werbos effectively solved the
exclusive-or problem and more
generally accelerated the training of
multi-layer networks using
Backpropagation algorithm.

24
 Artificial neural network
◼ Forward propagation: This is the
process of feeding data through a
neural network to get an output. It
involves multiplying the input with
weights, adding bias, and applying
an activation function at each node.
◼ Backpropagation: is a method
used in artificial neural networks to
calculate the error contribution of
each neuron after a batch of data.

25
 Artificial neural network
◼ Three concepts behind Backpropagation (From Calculus)
1. Derivative
2. Partial Derivative
3. Chain Rule

26
Artificial neural network

27
Artificial neural network

28
Artificial neural network

29
 Artificial neural network
◼ The linear and
nonlinear
activation
function
determines the
output of the unit

30
Artificial neural network training process

31
Forward propagation

32
Forward propagation

33
Forward propagation

34
Forward propagation

35
Backpropagation

36
Backpropagation

37
Forward propagation

38
Forward propagation

39
Forward propagation

40
Forward propagation

41
Forward propagation

42
Forward propagation

43
Forward propagation

44
Forward propagation

45
 Backpropagation
◼ Our goal with back-propagation is to update each of the weights
in the network so that they cause the actual output to be closer
to the target output by minimizing the error for each output
neuron and the network as a whole.

46
Backpropagation

47
Backpropagation

48
Backpropagation

49
Backpropagation

50
Backpropagation

51
Backpropagation

52
Backpropagation

53
Backpropagation

54
Backpropagation

55
Backpropagation

56
Backpropagation

57
Backpropagation

58
Backpropagation

59
Classification of Handwritten Digits

60
Classification of Handwritten Digits

61
Classification of Handwritten Digits

62
Classification of Handwritten Digits

63
Classification of Handwritten Digits

64
Classification of Handwritten Digits

65
Classification of Handwritten Digits

66
Neural Network Zero

67
Neural Network Zero

68
Neural Network Zero

69
Neural Network Zero

70
Neural Network Zero

71
Neural Network Zero

72
 Application of ANN
◼ Image recognition: ANN can classify objects in images, like
faces, cars, etc.
◼ Speech recognition: ANN can convert speech into text and
vice versa.
◼ Natural language processing: ANN can understand and
generate natural language, like translation, text summarization,
etc.
◼ Weather prediction: ANN can predict future weather
conditions based on historical data.
◼ Medical analytics: ANN can assist in disease diagnosis,
disease risk prediction, etc.
◼ Finance: ANN can predict stock prices, analyze risks, and
more.
◼ Manufacturing: ANN can control the manufacturing process, 73