MODULE:2

HEBB NEURAL NETWORK

HEBB NETWORK
 Donald Hebb in 1949
 Hebb rule is more suited for bipolar data

 limitations : Hebb rule not applicable for binary data

 weight update in Hebb rule:

wi(new) = wi(old) + ∆w
∆w weight change
ARCHITECTURE
HEBBIAN LEARNING-TRAINING
ALGORITHM
 Step 0: First initialize the weights. Basically in this
network they may be zero, i.e., w= 0 for i= 1to n
where "n" may be the total number of input neurons.
 Step 1: Steps 2-4 have to be performed for each input
training vector and target output pair, s: t.
 Step 2: Input units activations are set.

xi =si for i= 1to n

 Step 3: Output units activations are set: y= t
 Step 4: Weight & bias calculations:

wi(new) = wi(old)+ xiy

b(new) = b(old) + y
SOLVED PROBLEM-1
 Realize a Hebb net for the AND function with bipolar
inputs and targets.

x1 x2 y(target)
1 1 1

1 -1 -1

-1 1 -1

-1 -1 -1
 Initialize :
w1=w2=0 and bias, b=0

 Change in weight and bias

∆wi=xiy

∆b=y
SOLVED PROBLEM-1
inputs target Change in Change in bias New New New
weights weight weight bias

x1 x2 y ∆w1 ∆w2 ∆b=y w1 w2 b

=x1y =x2y

1 1 1 1 1 1 1 1 1
1 -1 -1 -1 1 -1 0 2 0
-1 1 -1 1 -1 -1 1 1 -1
-1 -1 -1 1 1 -1 2 2 -2

testing W1=2, w2=2, b=-2 (1, 1)

X1w1+x2w2+b=2  y=f(yin) = 1-- y=t

W1=2, w2=2, b=-2 (1, -1)

X1w1+x2w2+b= -2  y=f(yin) = -1-- y=t
SOLVED PROBLEM-2
 Realize a Hebb net for the XOR function with bipolar
inputs and targets.

x1 x2 y (target)
1 1 -1

1 -1 1

-1 1 1

-1 -1 -1
SOLVED PROBLEM-2

x1 x2 y ∆w1= ∆w2 ∆b=y W1 =0 w2 =0 b =0

x1 y =x2y

1 1 -1 -1 -1 -1 -1 -1 -1
1 -1 1 1 -1 1 0 -2 0
-1 1 1 -1 1 1 -1 -1 1
-1 -1 -1 1 1 -1 0 0 0
EXERCISE PROBLEMS
1. Design a Hebb net to implement OR function (consider
bipolar inputs and targets).
 EXERCISE PROBLEMS
1. Using the hebb rule, find the weights required to
perform the following classifications:
vectors (1 1 1 1) and (-1 1 -1 -1) are the members of class
(with target value 1) ;
vectors (1 1 1 -1) and (1 -1 -1 1) are the members of class
(with target value -1)
x1 x2 x3 x4 t
1 1 1 1 1
-1 1 -1 -1 1
1 1 1 -1 -1
1 -1 -1 1 -1
APPLICATION BASED QUESTION
 Classify the two-dimensional input pattern using Hebb
neural network (letters I-J pattern)