Exemplu Perceptron NeuralNetworks1 PDF




International Baccalaureate (IB)

Artificial Neural Networks - #1
Classification using Single Layer Perceptron Model
Click here to see the video http://youtu.be/S3iQgcoQVbc
http://scholastictutors.webs.com
(ANN-001) 2
as
Classification Example
• The following table shows sample data obtained from two

different fruits.
Weight (grams) Length (cm)
Fruit 1 (Class 121 16.8
C1) 114 15.2
C2) 195 8.1
• Train a single layer perceptron model using the above parameters

to classify the two fruits.
• Using the model parameters you have obtained classify the fruit
with weight 140gm and length 17.9cm.
http://scholastictutors.webs.com 3
Classification Example
Weight (grams) Length (cm)

C1) 114 15.2
C2) 195 8.1
X2(Length in cm)
Class C1
20
(121,16.8)
16 (114,15.2)
12
(210,9.4)
8 (195,8.1)
4 Class C2
X1(Weight
http://scholastictutors.webs.com 50 100 150 200 250 in grams)
Artificial Neural Networks Theory
This space is left for

sponsors/advertisements
from
US$ 999 per advert
(Multi-book & Multi-page adverts with
web links to divert traffic to your web
sites/videos are also available)
5
Artificial Neural Networks Theory
• Artificial Neural Networks (ANN) are computers

whose architecture is modeled after the brain.
• They typically consist of many hundreds of simple
processing units which are wired together in a
complex communication network.
• Each unit or node is a simplified model of a real
neuron which fires (sends off a new signal) if it
receives a sufficiently strong input signal from the
other nodes to which it is connected.
6
Intelligent Machine
• The strength of these connections may be

varied in order for the network to perform
different tasks corresponding to different
patterns of node firing activity.
• This structure is very different from traditional
computers and hence make way to Intelligent
Machines.
7
Model of a Neuron
• The model consists of a set of synapses each of

which is characterized by a weight or strength of
its own.
• An adder, an activation function and a bias.
Bias
wk1 Activation
x1
bk Function
wk2
 (.)
Input Signals
x2 vk

Output
wk3 yk
x3
Summing
x4 wk4 Junction
Synaptic Weights 8
Some Mathematics
• In mathematical terms, a neuron k can be

described by: m
uk   wk j x j
j 1
and
yk    uk  bk 
where uk is the linear combiner output due to
input signals.
• Also .vk  uk  bk 9
Bias as an Input
• The bias is an external parameter of artificial

neuron and can be included into the equations as
follows: m
vk   wk j x j
j 0
• and
yk    vk 
• Note the change of limits of j from 1 to 0. 10
Modified Neuron Structure
Fixed Input
wk0 wk0=bk (bias)
x0 = +1
x1 wk1 Activation
Function
wk2
 (.)
x2
 vk Output
wk3 yk
x3
Summing
Junction
x4 wk4
Synaptic Weights
(including bias)
11
Types of Activation Functions
• Threshold Function or Heaviside Function:

1 if v  0
 (v )  
0 if v  0
1.0
 (v )
0.8
0.6
0.4
0.2
0
-2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 12
Types of Activation Functions (cont…)
 1
• Piecewise-Linear Function:  1, v
2

 1 1
 (v )  v,  v
 2 2
 1
 0, v
2
1.0
 (v )
0.8
0.6
0.4
0.2
0
-2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 13
Types of Activation Functions (cont…)
1
• Sigmoid Function:  (v ) 
1  exp  av 
where a is the slope parameter of the sigmoid
function.
1.0
 (v )
0.8
0.6 Increasing a
0.4
0.2
0
-8.0 -6.0 -4.0 -2.0 0 2.0 4.0 6.0 8.0 14
Single Layer Perceptron
• The neuronal model we have just discussed is

also known as a perceptron.
• The perceptron is the simplest form of a
neural network used for the classification of
patterns said to be linearly separable.
• Basically, it consists of a single neuron with
adjustable synaptic weights and bias.
• Now we will look at a method of achieving
learning in our model we have formulated.
15
Perceptron Convergence (Learning)
Algorithm
• Variables and Parameters

x( n )  ( m  1)  1 input vector
  1, x1 ( n ), x2 ( n ), , xm ( n )
T
w( n)  ( m  1)  1 weight vector
 b( n ), w1 ( n ), w2 ( n ), , wm ( n )
T
b( n )  bias
y( n )  actual response
d ( n )  desired response
  learning-rate parameter, a postive constant less than unity
16
Algorithm (cont…)
1. Initialization. Set w(0)  0. Then perform the following

computations for time step n = 1,2, …
2. Activation. At time step n, activate the perceptron by
applying input vector x( n)and desired response d ( n) .
3. Computation of Actual Response. Compute the actual
response of the perceptron:
y( n)  sgn[wT ( n)x( n)]

where sgn(.) is the signum function.
1, if x0
sgn( x )  
1, if x0 17
Algorithm (cont…)
4. Adaptation of Weight Vector. Update the
weight vector of the perceptron:
w( n  1)  w( n)    d ( n)  y( n) x( n)
where
 1 if x( n) belongs to class C1
d ( n)  
1 if x( n) belongs to class C2
5. Continuation. Increment time step n by one

and go back to step 2.
18
Decision Boundary
• The hyper-plane
x2
m
w x  b  0
i 1
i i
Class C1
or
w1 x1  w2 x2  b  0 0 x1
is the decision Class C2

boundary for a two
class classification
problem. 19
Solution to the Example Question
(with correct initial weights and bias)

from
US$ 999 per advert
20
Solution to the Example
(with correct initial weights and bias)
w1 (0)  30, w2 (0)  300, 

With correct initial
 given
weights and bias
b(0)  50,   0.01 
Fixed Input
Activation
50
Function
x0 = +1
x1 -30 Output
300
 sig (.)
y(n)
x2
1, if x0
sgn( x )  
1, if x0
21
(with correct initial weights and bias) (cont…)
w1 (0)  30, w2 (0)  300, 

 given
b(0)  50,   0.01  x2
Therefore the Initial Decision Class C1
20 (200,19.83)
Boundary for this example is: (121,16.8)
w1x1  w2 x2  b  0 16 (114,15.2)
12
30 x1  300 x2  50  0 (100,9.83) (210,9.4)
8 (195,8.1)
30  100  50 4 Class C2
x1  100, x2   9.83
300
50 100 150 200 250 x1
30  200  50
x1  200, x2   19.83
300
Initial hyper-plane does separate the two classes. 22
Classification of the Unknown Fruit
Fixed Input
Activation
50
x0 = +1 Function
x1 = 140 -30 +1220 Output
x2 = 17.9
300
 sig (.)
y(n)=+1
Class Unknown
For Class C1,
Now use the above model to classify the unknown fruit. Output = +1
x( unknown )   1, 140, 17.9
T
w(3)  50,  30, 300

T
 
y( unknown )  sgn w T (3)x( unknown )  sgn  50  1  30  140  300  17.9 
 sgn(1220)  1
 this
this unknown
unkown fruit
fruit belonge to the
belongs to the class
class C1. 23
Solution to the Example Question
(with unknown initial weights and bias)
Click here to see the video http://www.youtube.com/watch?feature=player_profilepage&v=S3iQgcoQVbc#t=1220s

from
US$ 999 per advert
24
(with unknown initial weights and bias)
w1 (0)  30, w2 (0)  300, 

With unknown initial  given
weights and bias b(0)  1230,   0.01 
Therefore the Decision Boundary for this

x2 (200,24.1)
case: 20
30 x1  300 x2  1230  0 (121,16.8)
16
(100,14.1) (114,15.2)
30  100  1230 12
x1  100, x2   14.1 (210,9.4)
300 8 (195,8.1)
30  200  1230
x1  200, x2   24.1 4
300
Initial hyperplane does not separate the two classes.

50 100 150 200 250 x1
Therefore we need to Train the Neural Network
25
Training with known Fruit
(121,16.8)
Fixed Input
Activation
-1230
x0 = +1 Function
x1 = 121 -30 Output


+180
sig (.) y(n)=+1
x2 = 16.8
300
Class C1
x( n )  x(0)   1, 121, 16.8
T
and d (0)  1 For Class C1,
w( n)  w( 0)   1230,  30, 300 Output = +1
T

y( n )  y(0)  sgn w T ( 0)x( 0) 
 sgn  1230  1  30  121  300  16.8 
 sgn (180)  1  d (0)
Hence no need to recalculate the weights.
 w( n  1)  w(1)   1230,  30, 300
T
26
(114,15.2)
Fixed Input
Activation
-1230
x0 = +1 Function
x1 = 114 -30 Output

-90 y(n)=-1
sig (.)
x2 = 15.2
300
Class C1
For Class C1, the
x(1)   1, 114, 15.2

T output should be +1,
and d (1)  1 but what we get is -1.
Hence we have to
w(1)   1230,  30, 300
T
recalculate the weights.
 
y(1)  sgn w T (1)x(1)  sgn  1230  1  30  114  300  15.2 
 sgn( 90)  1  d (1)
Hence we have to recalculate the weights.
Adaptation of Weight Vector
Here we use Adaptation of Weight Vector (Step 4) to update the weight

vector of the perceptron.
w( n  1)  w( n)    d ( n)  y( n) x( n)
w(1)   1230,  30, 300
T
x(1)   1, 114, 15.2

T
d (1)  1, y(1)  1,  0.01

w(1  1)  w( 2)   1230,  30, 300  0.01  1   1   1, 114, 15.2
T T
  1230,  30, 300   0.02, 2.28, 0.304

T T
 w( 2)   1229.08,  27.72, 300.304

T
28
(210,9.4)
Fixed Input
-1229.08 Activation
x0 = +1 Function
x1 = 210 -27.72
Output
x2 = 9.4  -4227.4224
sig (.)
y(n)=-1
300.304
Class C2
w( 2)   1229.08,  27.72, 300.304

T For Class C2,
Output = -1
x( 2)   1, 210, 9.4
T
and d ( 2)  1
 
y( 2)  sgn w T ( 2)x( 2)  sgn  1229.08  1  27.72  210  300.304  9.4 
 sgn ( 4227.4224)  1  d ( 2)
 w( n  1)  w(3)   1229.08,  27.72, 300.304
T
29
(195,8.1)
Fixed Input
-1229.08 Activation
x0 = +1 Function
x1 = 195 -27.72
Output
x2 = 8.1  -4202.0176
sig (.)
y(n)=-1
300.304
Class C2
w(3)   1229.08,  27.72, 300.304

T For Class C2,
Output = -1
x(3)   1, 195, 8.1
T
and d (3)  1
 
y(3)  sgn w T (3)x(3)  sgn  1229.08  1  27.72  195  300.304  8.1
 sgn ( 4202.0176)  1  d (3)
 w( n  1)  w( 4)   1229.08,  27.72, 300.304
T
30
(121,16.8)
Fixed Input
-1229.08 Activation
x0 = +1 Function
x1 = 121 -27.72
Output
x2 = 16.8  461.91
sig (.)
y(n)=+1
300.304
Class C1
w( 4)   1229.08,  27.72, 300.304

T For Class C1,
Output = +1
x( 4)   1, 121, 16.8
T
and d ( 4)  1
 
y( 4)  sgn w T ( 4)x( 4)  sgn  1229.08  1  27.72  121  300.304  16.8 
 sgn ( 461.91)  1  d ( 4)
 w( n  1)  w(5)   1229.08,  27.72, 300.304
T
31
(114,15.2)
Fixed Input
-1229.08 Activation
x0 = +1 Function
x1 = 114 -27.72
Output
x2 = 15.2  175.46
sig (.)
y(n)=+1
300.304
Class C1
w(5)   1229.08,  27.72, 300.304

T
For Class C1,
x(5)   1, 114, 15.2
T Output = +1
and d (5)  1
 
y(5)  sgn w T (5)x(5)  sgn  1229.08  1  27.72  114  300.304  15.2 
 sgn (175.46)  1  d (5)
 w( n  1)  w(6)   1229.08,  27.72, 300.304
T
32
Decision Boundary After Training
Therefore the decision boundary obtained

after Neural Network training is:
27.72 x1  300.304 x2  1229.08  0

x2
27.72  100  1229.08
Class C1 (200,22.55)
x1  100, x2   13.32 20
300.30
(121,16.8)
27.72  200  1229.08 16 (114,15.2)
x1  200, x2   22.55
300.30 (100,13.32)
12
(210,9.4)
8 (195,8.1)
The new decision boundary
can now be used to classify 4 Class C2
any unknown item.
50 100 150 200 250 x1
33
Classification of the Unknown Fruit
using the New Decision Boundary
Fixed Input
-1229.08 Activation
x0 = +1 Function
x1 = 140 -27.72
Output
x2 = 17.9  175.46
sig (.)
y(n)=+1
300.304
Class Unknown
For Class C1,

Now use the above model to classify the unknown fruit. Output = +1
𝐱 unknown = +1,140,17.9 𝑇
𝐰 5 = −1229.08, −27.72,300.304 𝑇
y unkown = sgn 𝐰 𝑇 (5)𝐱(unknown)
= sgn −1229.08 × 1 − 27.72 × 140 + 300.304 × 17.9
= sgn 265.56 = +1
∴ this unknown fruit belongs to the class 𝐶1 35

Exemplu Perceptron NeuralNetworks1 PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Exemplu Perceptron NeuralNetworks1 PDF

Hochgeladen von

Copyright:

Verfügbare Formate



International Baccalaureate (IB)

Click here to see the video http://youtu.be/S3iQgcoQVbc

• The following table shows sample data obtained from two

• Train a single layer perceptron model using the above parameters

Weight (grams) Length (cm)

Click here to see the video http://youtu.be/S3iQgcoQVbc

This space is left for

• Artificial Neural Networks (ANN) are computers

• The strength of these connections may be

• The model consists of a set of synapses each of

• In mathematical terms, a neuron k can be

• The bias is an external parameter of artificial

• Threshold Function or Heaviside Function:

• The neuronal model we have just discussed is

• Variables and Parameters

1. Initialization. Set w(0)  0. Then perform the following

y( n)  sgn[wT ( n)x( n)]

5. Continuation. Increment time step n by one

is the decision Class C2

Click here to see the video http://youtu.be/S3iQgcoQVbc

This space is left for

w1 (0)  30, w2 (0)  300, 

w1 (0)  30, w2 (0)  300, 

x1 = 140 -30 +1220 Output

w(3)  50,  30, 300

Click here to see the video http://www.youtube.com/watch?feature=player_profilepage&v=S3iQgcoQVbc#t=1220s

This space is left for

w1 (0)  30, w2 (0)  300, 

Therefore the Decision Boundary for this

Initial hyperplane does not separate the two classes.

x1 = 121 -30 Output

x1 = 114 -30 Output

x(1)   1, 114, 15.2

Here we use Adaptation of Weight Vector (Step 4) to update the weight

x(1)   1, 114, 15.2

d (1)  1, y(1)  1,  0.01

  1230,  30, 300   0.02, 2.28, 0.304

 w( 2)   1229.08,  27.72, 300.304

w( 2)   1229.08,  27.72, 300.304

w(3)   1229.08,  27.72, 300.304

w( 4)   1229.08,  27.72, 300.304

w(5)   1229.08,  27.72, 300.304

Therefore the decision boundary obtained

27.72 x1  300.304 x2  1229.08  0

For Class C1,

∴ this unknown fruit belongs to the class 𝐶1 35

Das könnte Ihnen auch gefallen