4 Learning Rules

• Supervised Learning
Network is provided with a set of examples
of proper network behavior (inputs/targets)
{p1, t 1} , {p2, t 2} , … , {pQ, tQ}

• Reinforcement Learning
Network is only provided with a grade, or score,
which indicates network performance

• Unsupervised Learning
Only network inputs are available to the learning
algorithm. Network learns to categorize (cluster)
the inputs.
2
 4 Perceptron Architecture
w 1, 1 w 1, 2 … w 1, R
Input Hard Limit Layer w 2, 1 w 2, 2 … w 2, R
W =

p
Rx1 A
A
W
n AA
AA
a
Sx1
w S, 1 w S, 2 … w S, R

A AA
SxR
T
1w
Sx1 wi, 1
1 b wi, 2 T
Sx1 iw = W = 2w
R S
wi, R T
a = hardlim (Wp + b) Sw

T
a i = hardlim ( n i ) = hardlim ( iw p + b i )

3
4 Single-Neuron Perceptron
w 1, 1 = 1 w 1, 2 = 1 b = –1
Inputs Two-Input Neuron
p2

p1

p2
AA AA
AAAA
w1,1

w1,2
Σ
b
n a 1w p + b = 0
T 1 1 w
a=1

p1
1
a=0 1
a = hardlim (Wp + b)

T
a = hardlim ( 1w p + b ) = hardlim ( w 1, 1 p 1 + w1, 2 p 2 + b )

4
4 Decision Boundary
T T
1w p + b = 0 1w p = – b

• All points on the decision boundary have the same inner

product with the weight vector.
• Therefore they have the same projection onto the weight
vector, and they must lie on a line orthogonal to the
weight vector
1w Tp + b = 0

1 w

1 w 1 w

5
 4 Example - OR

 0 , t = 0  0 , t = 1  1 , t = 1  1 , t = 1
 1
p = 1   2
p = 2   3
p = 3   4
p = 4 
 0   1   0   1 

6
4 OR Solution

OR
1w

Weight vector should be orthogonal to the decision boundary.

w = 0.5
1
0.5

Pick a point on the decision boundary to find the bias.

T 0 + b = 0.25 + b = 0 ⇒
1w p + b = 0.5 0.5 b = – 0.25
0.5

7
4 Multiple-Neuron Perceptron

Each neuron will have its own decision boundary.

T
iw p + b i = 0

A single neuron can classify input vectors

into two categories.

A multi-neuron perceptron can classify

input vectors into 2S categories.

8
4 Learning Rule Test Problem
{p 1, t 1} , { p 2, t 2} , …, {pQ, tQ}

 1 , t = 1  –1 , t = 0   0 , t = 0
 1
p = 1   2
p = 2   3
p = 3 
 2   2   –1 

Inputs No-Bias Neuron

p1

p2
AAAA
AAAA
w1,1

w1,2
Σ n a

a = hardlim(Wp)
9
4 Starting Point
2 1

Random initial weight:

1.0
1w =
– 0.8

1w
3

Present p1 to the network:

T  1 
a = hardlim ( 1w p 1 ) = hardlim  1.0 – 0.8 
 2

a = hardlim ( – 0.6 ) = 0

Incorrect Classification.
10
 4 Tentative Learning Rule
• Set 1w to p1
– Not stable

• Add p1 to 1w

new old
Tentative Rule: If t = 1 and a = 0, then 1w = 1w +p

2 1

w
1

new old 1.0 + 1 = 2.0

1w = 1w + p1 =
– 0.8 2 1.2

3
11
 4 Second Input Vector
 
a = hardlim ( 1w p 2 ) = hardlim  2.0 1.2 – 1 
T
 2 

a = hardlim ( 0.4 ) = 1 (Incorrect Classification)

new old
Modification to Rule: If t = 0 and a = 1, then 1w = 1w –p

2 1

new old 2.0 –1 3.0

1w = 1w – p2 = – =
1.2 2 – 0.8

1w
3
12
 4 Third Input Vector
T  0 
a = hardlim ( 1w p 3 ) = hardlim  3.0 – 0.8 
 –1 
a = hardlim ( 0.8 ) = 1 (Incorrect Classification)
2 1

new old 3.0 – 0 = 3.0

1w = 1w – p3 =
– 0.8 –1 0.2 1w

Patterns are now correctly classified.

new old
If t = a, then 1w = 1w .
13
4 Unified Learning Rule
new old
If t = 1 and a = 0, then 1w = 1w +p
new old
If t = 0 and a = 1, then 1w = 1w –p
new old
If t = a, then 1w = 1w

e = t–a

new old
If e = 1, then 1w = 1w +p
new old
If e = – 1, then 1w = 1w –p
new old
If e = 0, then 1w = 1w

1w
new
= 1w
old
+ e p = 1w
old
+ (t – a)p A bias is a
weight with
new old
b = b +e an input of 1.
14
4 Multiple-Neuron Perceptrons
To update the ith row of the weight matrix:
new old
iw = iw + ei p

new old
bi = bi + ei

Matrix form:

Wnew = W old + ep T

new old
b = b +e

15
4 Apple/Banana Example
Training Set
 – 1   1 
   
 1
p = ,
1 1 t = 1   2
p = ,
1 2 t = 0 
   
 –1   –1 

Initial Weights
W = 0.5 – 1 – 0.5 b = 0.5

First Iteration
 –1 
 
a = hardlim ( Wp 1 + b ) = hardlim  0.5 – 1 – 0.5 1 + 0.5
 
 – 1 
a = hardlim ( – 0.5 ) = 0 e = t1 – a = 1 – 0 = 1
new old T
W = W + ep = 0.5 – 1 – 0.5 + ( 1 ) – 1 1 – 1 = – 0.5 0 – 1.5
new old
b = b + e = 0.5 + ( 1 ) = 1.5
16
4 Second Iteration

1
a = hardlim (Wp 2 + b) = hardlim ( – 0.5 0 – 1.5 1 + ( 1.5 ))
–1

a = hardlim (2.5) = 1

e = t2 – a = 0 – 1 = –1

new old T
W = W + ep = – 0.5 0 – 1.5 + ( – 1 ) 1 1 – 1 = – 1.5 – 1 – 0.5

new old
b = b + e = 1.5 + ( – 1 ) = 0.5

17
4 Check

–1
a = hardlim (Wp 1 + b) = hardlim ( – 1.5 – 1 – 0.5 1 + 0.5)
–1

a = hardlim (1.5) = 1 = t 1

1
a = hardlim (Wp 2 + b) = hardlim ( – 1.5 – 1 – 0.5 1 + 0.5)
–1

a = hardlim (– 1.5) = 0 = t 2

18
4 Perceptron Rule Capability

The perceptron rule will always

converge to weights which accomplish
the desired classification, assuming that
such weights exist.

19