Scribd
Upload

Hebb Neural Network Design and Applications

924 views12 pages
This document discusses Hebb neural networks. It provides an overview of Hebb's rule from 1949, which is better suited for bipolar data rather than binary data. It also describes the archite…

Uploaded by

Vighnesh M
  • Module 2: Hebb Neural Network
  • Hebb Network Overview
  • Architecture
  • Hebbian Learning-Training Algorithm
  • Solved Problem 1: AND Function
  • Solved Problem 2: XOR Function
  • Exercise Problems
  • Application-Based Question

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)

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 bin
ARCHITECTURE
HEBBIAN LEARNING-TRAINING ALGORITHM Step 0: First initialize the weights. Basically in this network they may be zero, i.e.
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
Initialize : Change in weight and bias ∆wi=xiy ∆b=y w1=w2=0 and bias, b=0
SOLVED PROBLEM-1 inputs target Change in weights Change in bias New weight New weight New bias x1 x2 y ∆w1 =x1y ∆w2 =x2y
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 -
SOLVED PROBLEM-2 x1 x2 y ∆w1= x1y ∆w2 =x2y ∆b=y W1 =0 w2 =0 b =0 1 1 -1 -1 -1 -1 -1 -1 -1 1 -1 1 1 -1 1 0 -2 0 -1 1 1 -1 1 1
EXERCISE PROBLEMS 1. Design a Hebb net to implement OR function (consider bipolar inputs and targets).

You might also like