3 Characterization

of Fuzzy Sets

Fuzzy Systems Engineering

Toward Human-Centric Computing
Contents
3.1 Generic characterization of fuzzy sets: fundamental descriptors

3.2 Equality and inclusion relationships in fuzzy sets

3.3 Energy and entropy measures of fuzziness

3.4 Specificity of fuzzy sets

3.5 Geometric interpretation of sets and fuzzy sets

3.6 Granulation of information

3.7 Characterization of the families of fuzzy sets

3.8 Fuzzy sets, sets, and the representation theorem

Pedrycz and Gomide, FSE 2007

3.1 Generic characterization of
fuzzy sets:
Some fundamental descriptors

Pedrycz and Gomide, FSE 2007

Fuzzy sets

Fuzzy sets are membership functions A: X → [0, 1]

In principle: any function is “eligible” to describe fuzzy sets

In practice it is important to consider:

– type, shape, and properties of the function
– nature of the underlying phenomena
– semantic soundness

Pedrycz and Gomide, FSE 2007

Normality

A(x) A(x)

1.0 1.0

A
A

x x
Normal Subnormal

hgt(A) = 1 hgt(A) < 1

hgt ( A) = sup A( x)
x∈X

Pedrycz and Gomide, FSE 2007

Normalization

A(x) A(x)

1.0 1.0
Norm(A)
A

x x
Subnormal Normal

hgt(A) < 1 hgt(Norm(A)) =1

A( x)
Norm( A)( x) =
hgt ( A)
Pedrycz and Gomide, FSE 2007
Support
A(x)

1.0

A Supp( A) = {x ∈ X | A( x) > 0}

Open set

Supp(A) x

A(x)

1.0

A CSupp( A) = closure{x ∈ X | A( x) > 0}

Closed set
CSupp(A) x Pedrycz and Gomide, FSE 2007
Core

A(x)

1.0

x
Core(A)

Core( A) = {x ∈ X | A( x) = 1}

Pedrycz and Gomide, FSE 2007

α−cut

A(x) A(x)

1.0 1.0
A A
α α

x x
Aα Aα+

Aα = {x ∈ X | A( x) ≥ α} Aα = {x ∈ X | A( x) > α}

Stronger condition
Convexity A[λx1 + (1 − λ ) x2 ] ≥ min[ A( x1), A( x2 )]

A(x)
A[λx1 + (1−λ)x2)]
A(x)
1.0
1.0

x1 x2 x
x

x = λx1 + (1−λ)x2 Aα
Aα = {x ∈ X | A( x) > α}
0≤λ≤1 Convex
fuzzy set Nonconvex
Cardinality

Card( A) = ∑ A( x) X finite or countable

x∈X

Card( A) = ∫ A( x)dx
X

Card(A) = | A | sigma count (σ–count)

 3.2 Equality and inclusion
relationships for fuzzy sets

Pedrycz and Gomide, FSE 2007

Equality

A = B iff A(x) = B(x) ∀x ∈ X

Inclusion

A ⊆ B iff A(x) ≤ B(x) ∀x ∈ X

Pedrycz and Gomide, FSE 2007

Sets

A (x ) A (x )
A B A B
1.0 1.0

x x

A⊆B A⊄B

Pedrycz and Gomide, FSE 2007

Degree of inclusion

1
A( x ) ⊂ B ( x ) = ∫ ( A( x ) ⇒ B ( x )) d x
Card ( X ) X

 1 if A( x) ≤ B ( x)
A( x) ⇒ B ( x) = 
1-A( x) + B( x) otherwise
Degree of equality

1
A( x) = B( x) = ∫ [min( A( x) ⇒ B( x), B( x) ⇒ A( x))]dx
Card ( X) X
Example

Examples of fuzzy sets A and B along with their degrees of inclusion:

(a) a = 0, n = 2, b =3; m = 4 σ = 2; ||A = B|| = 0.637

(b) b = 7 ||A = B|| = 0.864

(c) a = 0, n = 2, b = 9, m = 4, σ = 0.5 || A =B || = 0.987

Pedrycz and Gomide, FSE 2007
3.3 Energy and entropy
measures of fuzziness

Pedrycz and Gomide, FSE 2007

Energy measure of fuzziness

n
E ( A) = ∑ e[ A( xi )] Card (X) = n
i =1

e : [0, 1] → [0, 1] such that

E ( A) = ∫ e[ A( x)]dx e(0) = 0
X
e(1) = 1

e: monotonically increasing

Pedrycz and Gomide, FSE 2007

Example

e(u) = u ∀u ∈ [0, 1] linear

n
E ( A) = ∑ A( xi ) = Card ( A)
i =1

n n
E ( A) = ∑ A( xi ) = ∑ | A( xi ) − φ( xi ) | = d ( A, φ)
i =1 i =1

d = Hamming distance

Pedrycz and Gomide, FSE 2007

 e(u) non-linear

e(u) e(u)
1.0 1.0

1.0 u 1.0 u

Emphasis on high Emphasis on low

membership values membership values

Pedrycz and Gomide, FSE 2007

Inclusion of probabilistic information

n
E ( A) = ∑ pi e[ A( xi )] pi: probability of xi
i =1

E ( A) = ∫ p ( x)e[ A( x)]dx p(x): probability density function

X

Pedrycz and Gomide, FSE 2007

Entropy measure of fuzziness

h : [0,1] → [0,1]
n
H ( A) = ∑ h[ A( xi )] 1-monotonically increasing [0, ½]
i =1
2-monotonically decreasing (½, 1]
H( A) = ∫ h( A( x))dx 3-boundary conditions:
X h(0) = h (1) = 0
h(½) = 1

Pedrycz and Gomide, FSE 2007

 Specificity of fuzzy sets

A (x ) A (x )
1.0
1.0

xo x x

Specific fuzzy set Lack of specificity

Pedrycz and Gomide, FSE 2007

Specificity

1-Spec(A) = 1 if and only if ∃x0 ∈ A(x0) = 1, A(x) = 0 ∀x≠ x0

2-Spec(A) = 0 if and only if A(x) = 0 ∀x ∈ X

3-Spec(A1) ≤ Spec(A2) if A1 ⊃ A2

1.0
A1

A2

x Pedrycz and Gomide, FSE 2007

Examples

α max 1
Spec( A) = ∫ dα
0 Card ( Aα )

m 1
Spec( A) = ∑ ∆αi
i =1Card ( Aα i )

Yager (1993)

Pedrycz and Gomide, FSE 2007

Geometric interpretation of sets and fuzzy sets

X = { x1, x2} P(X) = {∅, {x1}, {x2}, { x1, x2}}

x2 x2
{ x1, x2} { x1, x2}
1.0 1.0

F(X)

∅ x1 ∅ x1
1.0 1.0
Set Fuzzy set
Pedrycz and Gomide, FSE 2007
3.4 Granulation of information

Pedrycz and Gomide, FSE 2007

Motivation

Need of granulation:

– abstract information
– summarize information

Purpose:

– comprehension
– decision making
– description

Pedrycz and Gomide, FSE 2007

 Discretization, quantization, granulation

A1 A2 A3 A4 C1 C2 C3 C4 F1 F2 F3 F4
1.0 1.0 1.0

x x x

Discretatization Quantization Granulation

Pedrycz and Gomide, FSE 2007

Formal mechanisms of granulation

〈X, G, S, C 〉 S

Sm
X : universe C
S2
G: formal framework of granulation
S1
S: collection of information granules

C: transformation Fuzzy Interval Rough G

Pedrycz and Gomide, FSE 2007

3.5 Characterization of families
of fuzzy sets
Frame of cognition

Codebook of conceptual entities

– family of linguistic landmarks

Φ = {A1, A2, … ., Am}

Ai is a fuzzy set on X, i = 1, …, m

Granulation that satisfies semantic constraints

– coverage

– semantic soundness

Pedrycz and Gomide, FSE 2007

Coverage

Φ= {A1, A2, … ., Am} covers X if, for any x ∈ X

∃i ∈ I | Ai(x) > 0

∃i ∈ I | Ai(x) > δ (δ-level coverage) δ ∈ [0, 1]

Ai´s are fuzzy set on X, i ∈ I = {1, …, m}

Pedrycz and Gomide, FSE 2007

Semantic soundness

• Each Ai, i ∈ I = {1, …, m} is unimodal and normal

• Fuzzy sets Ai are disjoint enough (λ-overlapping)

• Number of elements of Φ is low

Ai
1.0

x Pedrycz and Gomide, FSE 2007

 Characteristics of frames of cognition

Specificity: Φ1 more specific than Φ2 if Spec(A1i) > Spec(A2j)

Granularity: Φ1 finer than Φ2 if |Φ1| > |Φ2|

A1 A2 A3 A1 A2 A3 A4
1.0 1.0

Φ1 Φ2

x x
Pedrycz and Gomide, FSE 2007

Focus of attention

A A
1.0

Regions of focus of attention

implied by the corresponding
fuzzy sets
Pedrycz and Gomide, FSE 2007
• Information hiding

1.0 A

a1 a2 a3 a4 x

x ∈ [a2, a3] indistinguishable for A, but not for B

3.6 Fuzzy sets, sets and
the representation
theorem

Pedrycz and Gomide, FSE 2007

Any fuzzy set can be viewed as a family of sets:

A= U αAα 1.0 A
α∈[0,1] αk
αkAαk
αj
A( x) = sup αAα ( x)
α∈[0,1] αi

x
Aαi

Pedrycz and Gomide, FSE 2007

Example
X = {1, 2, 3, 4}

A = {0/1, 0.1/2, 0.3/3, 1/4, 0.3/5} = [0, 0.1, 0.3, 1, 0.3]

A0.1 = {0/1, 1/2, 1/3, 1/4, 1/5} = [0, 1, 1, 1, 1] → 0.1A0.1 = [0, 0.1, 0.1, 0.1, 0.1]

A0.3 = {0/1, 0/2, 1/3, 1/4, 1/5} = [0, 0, 1, 1, 1] → 0.3A0.3 = [0, 0 , 0.3, 0.3, 0.3]

A1 = {0/1, 0/2, 0/3, 1/4, 0/5} = [0, 0, 0, 1, 0] → 1.0A1 = [0, 0, 0, 1, 0]

A = max (0.1A0.1 , 0.3A0.3 , 1A1 )

A = [max (0,0,0), max(0.1,0,0), max(0.1,0.3,0), max(0.1,0.3,1), max(0.1,0.3,0)]

A = [0, 0.1, 0.3, 1, 0.3]

Pedrycz and Gomide, FSE 2007