You are on page 1of 17

# Artificial Neural Networks - 6

Today

## Pattern Association. continued

Fundamentals revisited
Autoassociative Nets
Hopfield Nets
BAMs
MATLAB coding

Pattern Association
Learning process of forming associations
between related patterns
Heteroassociative NNs
Autoassociative NNs
Hopfield Net

## Linear Vector Spaces

A linear vector space X set of vectors
defined over scalar field F,
1.Vector addition x X , y X , x y X
2. x y y x
3. ( x y) z x ( y z )
4.There is a unique vector called zero
vector, x x, x X

## 5. For each vector, x ( x) 0, x X

6. a F , x X , ax X
7. a F , b F , x X , a(bx) (ab) x
8. (a b) x ax bx
9. a( x y ) ax ay

## Linear Vector Spaces

Linear Independence
Spanning a space
Inner Product
Norm
Orthogonality

## Linear Vector Spaces - Ex

LI: x1=[1 1]T, x2=[1 -1]T
LI: [1 1 1], [1 0 1], [1 2 1] Ts
LI: [1 1 1 1], [1 0 1 1], [1 2 1 1] Ts

## Linear Vector Spaces

Eigenvalues
Eigenvectors
Significance:

A: X X

Az z
Ex: A cos sin
sin cos

1 1
A

Autoassociative Net

Concept
Ex: [1 1 1 1]
Test with one mistakes / missing
Test with two mistakes / missing
Store [1 1 -1 -1] and [-1 1 1 -1]
Limit!!!

Autoassociative Net

## Store [1 1 -1 -1] and [-1 1 1 -1]

Limit!!!
Store [1 -1 -1 1] and [1 1 1 1]
With no self-connection Ex: [1 1 1 -1]
Ex: [1 1 -1 -1], [-1 1 1 -1], [-1 1 -1 1], [1 1 1
1]
Storage capacity theorem

Autoassociative Net

Autoassociative Net

Recurrent Nets
[1 1 1 -1]
Three missing components!!
2 mistakes: [-1 -1 1 -1]
Fixed point cycle!!

Discrete Hopfield

Discrete Hopfield

Discrete Hopfield

Ex: [1 1 1 0]
Let W be bipolar
Let update bin order be y1,y4,y3,y2
Test: [0 0 1 0]

BAM

BAM