FUZZY LOGIC

The term fuzzy logic refers to a logic of

approximation.
Boolean logic assumes that every fact is
either entirely true or false.
Fuzzy logic allows for varying degrees of
truth.
Computers can apply this logic to represent
vague and imprecise ideas, such as hot,
tall or balding.
5
Air conditioner (Mitsubishi)
Conventional air conditioning systems use on-off controllers.
When the temperature drops below a preset level the unit is
automatically turned off.
When the temperature rises above a preset level the unit is turned
on.
The former preset value is slightly lower than the latter preset
value, providing a dead zone, so that high-frequency on-off cycling
(chatter) is avoided.
The thermostat in the system controls the on-off action.
For example, "when the temperature rises to 25C, turn on the unit,
and when the temperature falls to 20C, turn off the unit.
" The Mitsubishi air conditioner controls by using fuzzy rules such as:
"If the ambient air is getting warmer, turn the cooling power up a
little; if the air is getting chilly, turn the power down moderately,
etc."
Air conditioner (Mitsubishi)
The machine becomes smoother as a
result. This means less wear and tear of
the air conditioner, more consistent
comfortable room temperatures, and
increased efficiency (energy savings).
Introduction
In 1965, Prof. Lofti Zadeh published the first
article Fuzzy Sets.
It becomes billions of dollars business.
America
Europe
Asia

Thousands of patents

Uncertainty: Incomplete
Ambiguity: Imprecise

Applications:

Air Conditioner Washing Machine
Subway System Camera
Aerospace Nuclear Submarine
Pattern Recognition Control
Image Processing Computer Vision
They reflect a recent trend to view fuzzy logic (FL),
neurocomputing (NC), genetic computing (GC), Rough
Sets (RS) and probabilistic computing (PC) as an
association of computing methodologies falling under
the rubric of so-called soft computing.

Among the basic concepts that underlie human
cognition, three stand out in importance: granulation,
organization, and causation.
Granulation involves a partitioning of a whole into parts;
organization involves an integration of parts into a
whole; and causation relates to an association of causes
with effects
A granule may be viewed as a clump of points (objects)
drawn together by indistinguishability, similarity, or
functionality. Modes of information granulation (IG) in
which granules are crisp play an important role in many
theories, methods and techniques, among them interval
analysis, quantization, rough set theory, qualitative
process theory, and chunking.
In fuzzy logic, fuzzy IG underlies the basic concepts of
linguistic variables, fuzzy if-then rules, and fuzzy graphs
This perception is reinforced by viewing it in the context
of generalization. More specifically, any theory, method,
technique, or problem may be fuzzified (or f-generalized)
by replacing the concept of a crisp set with that of a
fuzzy set.
Similarly, any theory, method, technique, or problem can
be granulated (g-generalized) by partitioning variables,
functions, and relations into granules.
Furthermore, we can combine fuzzification with
granulation, which gives rise to fuzzy granulation (f-
granulation). Fuzzy granulation, then, provides a basis
for what might be called f.g-generalization.
The generalization of two-valued logic leads to
multivalued logic and parts of fuzzy logic. But fuzzy logic
in its wide sensewhich is the sense in which it is used
todayresults from f.g-generalization. This crucial
difference between multivalued logic and fuzzy logic
explains why fuzzy logic has so many applications,
whereas multivalued logic does not.
Introduction
Fuzzy set theory provides a means for
representing uncertainties.
Probablity random uncertainty
But some uncertainty is non-random
In fact, a huge amount!
Natural Language is vague and imprecise.
Fuzzy set theory uses Linguistic variables, rather
than quantitative variables to represent imprecise
concepts.

Fuzzy Logic
Fuzzy Logic is suitable to
Very complex models
Judgemental
Reasoning
Perception
Decision making
Requiring precision high cost, long time
Statistics and random processes

Based on Randomness.
Fuzziness
Example.
Random Errors generally average out over time
or space
Non-random errors will not generally average out
and likely to grow with time.
Information World
Information World
Crisp set has a unique membership function

A
(x) = 1 x e A
0 x e A

A
(x) e {0, 1}

Fuzzy Set can have an infinite number of membership
functions

A
e [0,1]

Fuzziness
Examples:
A number is close to 5
Fuzziness
Examples:
He/she is tall
Fuzziness
Randomness versus Fuzziness
Drinking Water Problem
Classical Sets
Fuzzy Sets
Operations on Classical Sets
Union:
A B = {x | x e A or x e B}
Intersection:
A B = {x | x e A and x e B}
Complement:
A = {x | x e A, x e X}
X Universal Set
Set Difference:
A | B = {x | x e A and x e B}
Set difference is also denoted by A - B
Properties of Classical Sets
A B = B A
A B = B A
A (B C) = (A B) C
A (B C) = (A B) C
A (B C) = (A B) (A C)
A (B C) = (A B) (A C)
A A = A
A A = A
A X = X
A X = A
A C = A
A C = C
Properties of Classical Sets
If A _ B _ C, then A _ C
De Morgans Law:
(A B) = A B
(A B) = A B

Proof:
LHS= {x | x e (A and B)}= {x | x e A or x e B)}=

A B= RHS



Can be extended to n sets

Generalized De Morgan Law:
A A
C X
X C


Using ( ) to keep original processing order





Generalized Duality Law:
C X
X C


Using ( ) to keep original processing order
Law of the excluded middle:
A A = X
Law of the Contradiction:
A A = C
These laws are not true for Fuzzy
Sets!

Fuzzy Sets
Characteristic function X, indicating the belongingness
of x to the set A
X(x) = 1 x e A
0 x e A
or called membership
Hence,
A B X
A B
(x)
= X
A
(x) X
B
(x)
= max(X
A
(x),X
B
(x))
Note: Some books use + for , but still it is not ordinary
addition!
Some more explanations follow
Fuzzy Sets
A B X
A B
(x)
= X
A
(x) X
B
(x)
= min(X
A
(x),X
B
(x))
A X
A
(x)
= 1 X
A
(x)
A _ B X
A
(x) < X
B
(x)
A = A

Fuzzy Sets
Note (x) e [0,1]
not {0,1} like Crisp set
A = {
A
(x1) / x1 +
A
(x2) / x2 + }
= {
A
(xi) / xi}
Note: + = add
/ = divide
Only for representing element and its
membership.
Also some books use (x) for Crisp Sets too.
Fuzzy Set Operations

A B
(x) =
A
(x)
B
(x)
= max(
A
(x),
B
(x))

A B
(x) =
A
(x)
B
(x)
= min(
A
(x),
B
(x))

A
(x) = 1 -
A
(x)
De Morgans Law also holds:
(A B) = A B
(A B) = A B
But, in general
A A
A A
X =
| =
Properties of Fuzzy Sets
A B = B A
A B = B A
A (B C) = (A B) C
A (B C) = (A B) C
A (B C) = (A B) (A C)
A (B C) = (A B) (A C)
A A = A A A = A
A X = X A X = A
A C = A A C = C
If A _ B _ C, then A _ C
A = A
Sets as Points in Hypercubes
Explore to n-dimension
Classical Relations
Fuzzy relations
Logic, Approximate reasoning, Rule-based learning
systems, Nonlinear Simulation, Classification,
Pattern Recognition, etc.
Cartesian Product
A = {a,b}
B = {0,1}
A x B = { (a,0) (a,1) (b,0) (b,1) }
Ordered Pairs
Consider
A x A
or A x B x C if C is given

Based on the above, Crisp Relations are discussed
next
Crisp Relations
A subset of a Cartesian Product A1 x A2 x x Ar is
called an r-ary relation over A1,A2,,Ar
If r = 2, the relation is a subset of A1 x A2
Binary relation from A1 into A2
The strength of a relation:
Characteristic Function
X(x,y) = 1 (x,y) e X x Y
0 (x,y) e X x Y
For Classical relations, the value is 1 or 0
If the universes or sets are finite, we can use relational
matrix to represent it.
Crisp Relations
Example:
If X = {1,2,3}
Y = {a,b,c}
R = { (1 a),(1 c),(2 a),(2 b),(3 b),(3 c) }

a b c
1 1 0 1
R = 2 1 1 0
3 0 1 1
Using a diagram to represent the relation
Crisp Relations
Relations can also be defined for continuous universes
R = { (x,y) | y > 2x, x e X, y e Y}
X = 1 y > 2x
0 otherwise
Crisp Relations
Cardinality:
N: # of elements in X
M: # of elements in y
Cardinality of R
n
X x Y
= n
X
n
Y
= M N
Cardinality of the Power set of this relation
n
P(X x Y)
= 2
M-N

Operations on Crisp Relations
Complete Relation Matrix
1 1 1
1 1 1
1 1 1
Null relation Matrix
0 0 0
0 0 0
0 0 0
Operations on Crisp Relations
Union
R S X
R S
(x,y)
X
R S
(x,y) = max{ X
R
(x,y),X
S
(x,y) }
Intersection
R S X
R S
(x,y)
X
R S
(x,y) = min{ X
R
(x,y),X
S
(x,y) }
Complement
R X
R
(x,y)
X
R
(x,y) = 1 X
R
(x,y)
Containment
R _ S X
R
(x,y) < X
S
(x,y)
Identity
C 0
X E
Properties of Crisp Relations
Commutativity
Associativity
Distributivity
Idempotency All hold
De Morgan Law
Excluded middle Law
Etc.
Properties of Crisp Relations
Composition
Let R be a relation representing a mapping from X to Y
X
Y University sets
Let S be a relation, a mapping from Y to Z
Can we find T from R to S?
Properties of Crisp Relations
T: mapping from X to Z
T = R S
Two ways to compute X
T
(xz)
1. X
T
(xz) = (X
R
(xy) X
s
(yz))
= max(min{X
R
(xy),X
S
(yz)})
Max-min composition
2. X
T
(xz) = (X
R
(xy) - X
s
(yz))
Max-product composition
multiplication
y e Y
y e Y
y e Y
Properties of Crisp Relations
Using Matrix representation:
y1 y2 y3 y4
x1 1 0 1 0
R = x2 0 0 0 1
x3 0 0 0 0
z1 z2
y1 0 1 z1 z2
y2 0 0 x1 0 0
S = y3 0 1 T = x2 0 0
y4 0 0 x3 0 0

T
(x1,z1) = max[min(1,0) min(0,0) min(1,0) min(0,0)]
= max[0,0,0,0] = 0
Similar, but not the same as matrix multiplication!
Fuzzy Relations
Cardinality of Fuzzy Relations
Since the cardinality of fuzzy sets on any universe is
infinity, the cardinality of a fuzzy relation is also infinity.



Note: other books have different discussions!
Operations on Fuzzy Relations
Union:

R S
= max{
R
(x,y),
S
(x,y) }
Intersection:

R S
= min{
R
(x,y),
S
(x,y) }
Complement:

R
(x,y) = 1 -
R
(x,y)
Containment:
R _ S
R
(x,y) <
S
(x,y)
Properties of Fuzzy Relations
Commutativity
Associativity
Distributivity
Idempotency All hold
De Morgan Law
Excluded middle Law
Etc.
Note: R R = E
R R = 0
In general.
Properties of Fuzzy Relations
Fuzzy Cartesian Product and Composition

R
(x y) =
A x B
(x y) = min(
A
(x),
B
(y))
Example:
A = 0.2/x1 + 0.5/x2 + 1/x3
B = 0.3/y1 + 0.9/y2
y1 y2
0.2 x1 0.2 0.2
A x B = 0.5 0.3 0.9 = x2 0.3 0.5
1 x3 0.3 0.9
Properties of Fuzzy Relations
Vector Outer Product
If R is a fuzzy relation on the space X x Y
S is a fuzzy relation on the space Y x Z
Then, fuzzy composition is T = R S
1. Fuzzy max-min composition

T
(xz) = (
R
(xy)
s
(yz))
2. Fuzzy max-production composition

T
(xz) = (
R
(xy) -
s
(yz))
Note: R S = S R
y e Y
y e Y
Properties of Fuzzy Relations
Example:
y1 y2 z1 z2 z3
R = x1 0.7 0.5 S = y1 0.9 0.6 0.2
x2 0.8 0.4 y2 0.1 0.7 0.5
z1 z2 z3
Using max-min, T = x1 0.7 0.6 0.5
x2 0.8 0.6 0.4
z1 z2 z3
Using max-product, T = x1 0.63 0.42 0.25
x2 0.72 0.48 0.20
Note: Set, Relation, Composition
How to find new membership from the given ones!
Tolerance and Equivalence Relation
Crisp Equivalence Relation
R _ X x X
Relation has the following properties:
Reflexivity
(xi xi) e R or X
R
(xi xi) = 1
Symmetry
(xi xj) e R (xj xi) e R
or X
R
(xi xj) = X
R
(xj xi)
Transitivity
(xi xj) e R and (xj xk) e R (xi xk) e R
or X
R
(xi xj) = 1 and X
R
(xj xk) = 1 X
R
(xi xk) = 1
Tolerance and Equivalence Relation
Graph representation:
Crisp Tolerance Relation
(or proximity relation)
Only has reflexivity and symmetry
A tolerance relation, R1 can become an Equivalence
Relation by at most (n-1) compositions (< n-1), n is the
cardinal member of X.
R
1
n-1
= R1 R1 R1 = R

Crisp Tolerance Relation
(or proximity relation)
Example:
1 1 0 0 0
1 1 0 0 1
R1 = 0 0 1 0 0
0 0 0 1 0
0 1 0 0 1

1 1 0 0 1
1 1 0 0 1
Try R1
2
= 0 0 1 0 0
0 0 0 1 0
1 1 0 0 1

Note: symmetric, reflexive, but
not transitive, why?
X(x1 x2) = 1
X(x2 x5) = 1 but
X(x1 x5) = 1 (=0)
Now, it is transitive!
Fuzzy Tolerance and Equivalence Relation
A fuzzy relation R has:
1. Reflexivity
R
(xi xi) = 1
2. Symmetry
R
(xi xj) =
R
(xj xi)
3. Transitivity
R
(xi xj) = 1

R
(xj xk) = 2
R
(xi xk) =
where > min{1, 2}
Fuzzy tolerance relation R1 has reflexivity, symmetry. It
can be transformed into a fuzzy equivalence relation by
at most (n-1) (< n-1) compositions.
R
1
n-1
= R1 R1 R1 = R
Fuzzy Tolerance and Equivalence Relation
Example:
1 0.8 0 0.1 0.2
0.8 1 0.4 0 0.9
R1 = 0 0.4 1 0 0
0.1 0 0 1 0.5
0.2 0.9 0 0.5 1

R1
(x1 x2) = 0.8

R1
(x2 x5) = 0.9
But
R1
(x1 x5) = 0.2 < min(0.8,0.9) not transitive
Fuzzy Tolerance and Equivalence Relation
Value Assignment
How to find the membership values for the relation?
1. Cartesian Production
Note: you have to know the membership value for the
sets! Will discuss in chapter 4.
2. Y = f(x) X input vector
Y output vector

3. Look up table y1 y2 y3
x1
x2
x3
Fuzzy Tolerance and Equivalence Relation
Value Assignment
4. Linguistic rule of knowledge chapters 7 9
5. Classification chapter 11
6. Similarity methods in data manipulation

The more robust a data set, the more accurate the
relation entries!
Cosine Amplitude
X = {x
1
,x
2
,,x
n
} each element is also a vector
X
i
= {x
i1
,x
i2
,,x
im
}

ij
=
R
(x
i
,x
j
)
It will be n x n symmetric,reflexive i.e. a tolerance
relation!
Note: this relates to the vector dot product for cosine
function
Cosine Amplitude
Example:
x1 x2 x3 x4 x5
xi1 0.3 0.2 0.1 0.7 0.4
xi2 0.6 0.4 0.6 0.2 0.6
xi3 0.1 0.4 0.3 0.1 0
Using the above formula:
1 symm
0.836 1
R1 = 0.914 0.934 1
0.682 0.6 0.441 1
0.982 0.74 0.818 0.7741 1
Cosine Amplitude
1 symm
0.538 1
R1 = 0.667 0.667 1
0.429 0.333 0.25 1
0.818 0.429 0.538 0.429 1
Computationally simple!
Max-min Method:
Other Similarity Methods
Absolute Exponential:
Exponential Similarity Coefficient:
Where, S
k
= any general measure for all the data i.e.
(S
k
)
2
0
Other Similarity Methods
Other methods produce scalar quantities which are similar
to the cosine amplitude, such as the following:
Geometric average minimum:
Scalar Product:

Where:
Other Similarity Methods
Some methods are analogous to popular statistical
quantities, such as:
Correlation Coefficient:
Where: and
Arithmetic Average Minimum:
Other Similarity Methods
Some methods are based on the inverse relationships,
for example:
Absolute Reciprocal:
Where M is selected to make 0 r
ij
1
Absolute subtrahend:
Where c is selected to make 0 r
ij
1
Other Similarity Methods
Other methods are nonparametric, such as:
Nonparametric:
where x
ik
= x
ik
x
i
and x
jk
x
j

n
+
= number of elements > 0 in
{x
i1
x
j1
,x
i2
x
j2
,,x
im
x
jm
}
n
-
= number of elements < 0 in
{x
i1
x
j1
,x
i2
,x
j2
,,x
im
,x
jm
}

In the above equations, terms such as x
i1
x
j1
are
products of data elements.

Membership Function
Membership Functions characterize the fuzziness of
fuzzy sets. There are an infinite # of ways to
characterize fuzzy infinite ways to define fuzzy
membership functions.

Membership function essentially embodies all fuzziness
for a particular fuzzy set, its description is essential to
fuzzy property or operation.
Features of Membership Function
Core: comprises of elements x of the universe, such that

A
(x) = 1
Support: comprises of elements x of universe, such that

A
(x) > 0
Boundaries: comprise the elements x of the universe
0 <
A
(x) < 1
A normal fuzzy set has at least one element with
membership 1
For fuzzy set, if one and only one element has a
membership = 1, this element is called as the prototype
of set.
A subnormal fuzzy set has no element with
membership=1.
Features of Membership Function
Graphically,
Features of Membership Function
A convex fuzzy set has a membership whose value is:
1. strictly monotonically increasing, or
2. strictly monotonically decreasing, or
3. strictly monotonically increasing, then strictly
monotonically decreasing
Or another way to describe:
(y) min[(x), (z)], if x < y < z
If A and B are convex sets, then A B is also a convex set
Crossover points have membership 0.5
Height of a Fuzzy set is the maximum value of the
membership: max{
A
(x)}
Features of Membership Function
If height < 1, the fuzzy set is subnormal.
Fuzzy number: like a number is close to 5. It has to have
the properties:
1. A must be a normal fuzzy set.
2. oA must be closed for all o(0,1].
3. The support, 0A must be bounded.