PERCEPTRON LEARNING

Positive class = 1

Negative class = -1

Net input function = yin =w1x1+w2x2+…wmxm = w1.x1+w2.x2+…wm.xm + wb.b {b= bias , b=1}

Threshold function (Activation function) = y = f(y in) = {−11 ifotherwise

W (new) = w(old) + n (t-y)*x n= learning rate

Wb(new) = Wb(old) + n (t-y)*b

t(class_lab n=0.15
x1 x2 el) b w1 = -0.7
-1 -1 -1 1
w2 = 0.5
-1 1 -1 1
1 -1 -1 1 Wb = 0.2
1 1 1 1

yin = w1.x1+w2.x2+ wb.b

First instance (-1,-1) is producing a signal :

yin = (-0.7)(-1)+(0.5)(-1)+0.2(1) = 0.4

y = f(yin) = 1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

w1 (new) = w1(old) + n (t - y) * x1 =-0.7+ 0.15*(-1-1)*(-1) =- 0.4

w2 (new) = w2(old) + n (t - y) * x2 =0.5+0.15*(-1-1)*(-1) = 0.8

Wb(new) = Wb(old) + n (t-y)*b = 0.2+0.15*(-1-1)*1 = -0.1

Second instance is producing a signal (-1,1):

yin = (-0.4)(-1)+(0.8)(1)-0.1(1) = 1.1

y = f(yin) = 1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

w1 (new) = w1(old) + n (t - y) * x1 =-0.4+ 0.15*(-1-1)*(-1) = -0.1

w2 (new) = w2(old) + n (t - y) * x2 =0.8+0.15*(-1-1)*(1) = 0.5

Wb(new) = Wb(old) + n (t-y)*b = -0.1+0.15*(-1-1)*1 = -0.4

Third instance (1,-1) is producing a signal:

yin = (-0.1)(1)+(0.5)(-1)-0.4(1) = -1

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

The 4th instance (1,1) is producing a signal:

yin = (-0.1)(1)+(0.5)(1)-0.4(1) = 0

y = f(yin) = 1 (the prediction of the class label that the perceptron gives you)

t = 1 (actual class label (target))

target t = 1 and predicted class label y= 1. Since they are equal to each other we don’t need any
update.

WE SHOULD REPEAT THE WHOLE PROCESS

First instance (-1,-1) is producing a signal :

yin = (-0.1)(-1)+(0.5)(-1)-0.4(1) = -0.8

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

Second instance is producing a signal (-1,1):

yin = (-0.1)(-1)+(0.5)(1)-0.4(1) = 0.2

y = f(yin) = 1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

w1 (new) = w1(old) + n (t - y) * x1 =-0.1+ 0.15*(-1-1)*(-1) = 0.2

w2 (new) = w2(old) + n (t - y) * x2 =0.5+0.15*(-1-1)*(1) = 0.2

Wb(new) = Wb(old) + n (t-y)*b = -0.4+0.15*(-1-1)*1 = -0.7

w1=0.2, w2=0.2, wb= -0.7

Third instance (1,-1) is producing a signal:

yin = (0.2)(1)+(0.2)(-1)-0.7(1) = -0.7

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

w1=0.2, w2=0.2, wb= -0.7

The 4th instance (1,1) is producing a signal:

yin = (0.2)(1)+(0.2)(1)-0.7(1) = -0.3

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = 1 (actual class label (target))

target t = 1 and predicted class label y= 1. Since they are not equal to each other we need to
update the weights.

w1 (new) = w1(old) + n (t - y) * x1 =0.2+ 0.15*(1-(-1))*(1) = 0.5

w2 (new) = w2(old) + n (t - y) * x2 =0.2+0.15*(1-(-1))*(1) = 0.5

Wb(new) = Wb(old) + n (t-y)*b = -0.7+0.15*(1-(-1))*1 = -0.4

w1=0.5, w2=0.5, wb= -0.4

WE SHOULD REPEAT THE WHOLE PROCESS

First instance (-1,-1) is producing a signal :

yin = (0.5)(-1)+(0.5)(-1)-0.4(1) = -1.4

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

w1=0.5, w2=0.5, wb= -0.4

Second instance is producing a signal (-1,1):

yin = (0.5)(-1)+(0.5)(1)-0.4(1) = -0.4

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

w1=0.5, w2=0.5, wb= -0.4

Third instance (1,-1) is producing a signal:

yin = (0.5)(1)+(0.5)(-1)-0.4(1) = -0.4

y = f(yin) = -1 (the prediction of the class label that the perceptron gives you)

t = -1 (actual class label (target))

target t = -1 and predicted class label y= -1. Since they are equal to each other we don’t need
any update.

w1=0.5, w2=0.5, wb= -0.4

The 4th instance (1,1) is producing a signal:

yin = (0.5)(1)+(0.5)(1)-0.4(1) = 0.6

y = f(yin) = 1 (the prediction of the class label that the perceptron gives you)

t = 1 (actual class label (target))

target t = 1 and predicted class label y= 1. Since they are not equal to each other we need to
update the weights.

w1=0.5, w2=0.5, wb= -0.4

FINAL RESULT and predict a test instance (5,6)