Fuzzy Function and the Generalized Extension Principle

Irina Perfilieva1,2 and Alexandr Šostak2

1 University of Ostrava, Institute for Research and Applications of Fuzzy Modeling,
NSC IT4Innovations, 30. dubna 22, 701 03 Ostrava 1, Czech Republic
2 Institute of Mathematics and CS, University of Latia, LV-1459, Riga, Latvia

Keywords: Fuzzy Function, Extensional Function, Fuzzy Equivalence, Extension Principle.

Abstract: The aim of this contribution is to develop a theory of such concepts as fuzzy point, fuzzy set and fuzzy function
in a similar style as is common in classical mathematical analysis. We recall some known notions and propose
new ones with the purpose to show that, similarly to the classical case, a (fuzzy) set is a collection of (fuzzy)
points or singletons. We show a relationship between a fuzzy function and its ordinary “skeleton” that can
be naturally associated with the original function. We show that any fuzzy function can be extended to the
domain of fuzzy subsets and this extension is analogous to the Extension Principle of L. A. Zadeh.

1 INTRODUCTION lection of (fuzzy) points or (fuzzy) singletons.

Last, but not least, we analyze a relationship be-
The notion of fuzzy function has at least two differ- tween a surjective fuzzy function and its ordinary
ent meanings in fuzzy literature. On the one side (see core function. The similar study has been attempted
e.g., (Hájek, 1998; Klawonn, 2000; Demirci, 1999; in (Demirci, 1999) for a perfect fuzzy function and
Demirci, 2002; Höhle et al., 2000; Šostak, 2001)), a in (Klawonn, 2000) for one particular example of a
fuzzy function is a special fuzzy relation with a gen- fuzzy function. We propose a solution in the general
eralized property of uniqueness. According to this case.
approach, each element from the ordinary domain of The present paper is organized as follows. In Sec-
thus defined fuzzy function is associated with a cer- tion 2, we give preliminary information about ex-
tain fuzzy set. Thus, a fuzzy function establishes a tension principle, residuated lattices, fuzzy sets and
“point”-to-“fuzzy set” correspondence. fuzzy spaces. Fuzzy functions and two approaches to
On the other hand (see (Novák, 1989; Perfilieva, this notion are discussed in Section 3. Section 3 con-
2004; Perfilieva, 2011; Perfilieva et al., 2012)), a tains also main results of the paper.
fuzzy function is a mapping between two universes
of fuzzy sets, i.e. establishes a “fuzzy set”-to-“fuzzy
set” correspondence. This approach is implicitly used 2 PRELIMINARIES
in many papers devoted to fuzzy IF-THEN rule mod-
els where the latter are actually partially given fuzzy
functions. 2.1 Extension Principle and Its
In this contribution, we show that both viewpoints Relational Form
can be connected by a natural generalization of the
Extension Principle of L. Zadeh (Zadeh, 1975). In An extension principle has been proposed by L.
details, a fuzzy function as a mapping is an extension Zadeh (Zadeh, 1975) in 1975 and since then it is
of a fuzzy function as a relation to the domain of fuzzy widely used in the fuzzy set theory and its applica-
sets. The similar approach has been used in (Šostak, tions. Let us recall the principle and propose its rela-
2001). tion form which will be later on used in a relationship
In order to establish the above mentioned exten- to fuzzy function.
sion, we introduce various spaces of fuzzy objects Assume that X,Y are universal sets and f : X →
with fuzzy equivalence relations on them. We show Y is a function with the domain X. Let moreover,
that similar to the classical case, a (fuzzy) set is a col- F (X), F (Y ) be respective universes of fuzzy sets on

Perfilieva I. and Šostak A..

Fuzzy Function and the Generalized Extension Principle.
169
DOI: 10.5220/0005132701690174
In Proceedings of the International Conference on Fuzzy Computation Theory and Applications (FCTA-2014), pages 169-174
ISBN: 978-989-758-053-6
Copyright c 2014 SCITEPRESS (Science and Technology Publications, Lda.)
FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

X and Y identified with their membership functions, 2.3 L-fuzzy Sets, Fuzzy Relations and
i.e. F (X) = {A : X → [0, 1]} and similarly, F (Y ). Fuzzy Spaces
By the extension principle, f induces a function f → :
F (X) → F (Y ) such that for all A ∈ F (X), Below, we recall definitions of principal notions in the
f → (A)(y) = sup A(x). (1) fuzzy set theory.
y= f (x)

Let R f be a binary relation on X × Y which corre- Fuzzy Sets with Crisp Equality. Let X be a non-
sponds to the function f , i.e. empty universal set, L a complete residuated lattice.
R f (x, y) = 1 ⇔ y = f (x). An (L-)fuzzy set A of X (fuzzy set, shortly) is a map
A : X → L that establishes a relationship between ele-
Then it is easy to see that (1) can be equivalently rep- ments of X and degrees of membership to A.
resented by Fuzzy set A is normal if there exists xA ∈ X such
_
f → (A)(y) = (A(x) · R f (x, y)). (2) that A(xA ) = 1. The (ordinary) set Core(A) = {x ∈
y∈Y
X | A(x) = 1} is a core of the normal fuzzy set A.
The (ordinary) set Supp(A) = {x ∈ X | A(x) > 0} is a
Expression (2) is the relational form of the extension support set of fuzzy set A.
principle. The meaning of expression (2) becomes A class of L-fuzzy sets of X will be denoted
more general when A is an L-fuzzy set (see Defini- LX . The couple (LX , =) is called an ordinary fuzzy
tion 3 below), binary relation R f is a fuzzy relation, space on X. The elements of (LX , =) are fuzzy sets
and multiplication · changes to a monoidal operation equipped with a crisp equality relation, i.e. for all
(see Section 2.2). In Section 3, we will discuss the A, B ∈ LX ,
proposed generalization and its relationship to fuzzy
functions. A = B if and only if (∀x ∈ X) A(x) = B(x).
In (LX , =), we strictly distinguish between fuzzy sets
2.2 Residuated Lattice even if their membership functions differ in one point.
On (LX , =), we can define the structure of resid-
Our basic algebra of operations is a residuated lattice.
uated lattice using pointwise operations over fuzzy
Definition 1. A residuated lattice is an algebra sets. Moreover, the underlying lattice hLX , ∨, ∧, 0, 1i
L = hL, ∨, ∧, ∗, →, 0, 1i. is complete, where the bottom 0 and the top 1 are con-
stant fuzzy sets, respectively.
with a support L and four binary operations and two A class of normal L-fuzzy sets of X will be de-
constants such that noted N (X). The space (N (X), =) is a subspace of
• hL, ∨, ∧, 0, 1i is a lattice where the ordering ≤ de- (LX , =).
fined using operations ∨, ∧ as usual, and 0, 1 are By identifying a point u ∈ X with a fuzzy subset
the least and the greatest elements, respectively; Iu : X → L such that Iu (u) = 1 and Iu (x) = 0 whenever
• hL, ∗, 1i is a commutative monoid, that is, ∗ is a x 6= u we may view X as a subspace of (LX , =) and as
commutative and associative operation with the a subspace of (N (X), =)
identity a ∗ 1 = a;
• the operation → is a residuation operation with Space with Fuzzy Equivalence. Fuzzy Points. Let
respect to ∗, i.e. X, Y be universal sets. Similarly to L-valued fuzzy
a ∗ b ≤ c ⇐⇒ a ≤ b → c. sets, we define (binary) (L-)fuzzy relations as fuzzy
sets of X × Y . If X = Y , then a fuzzy set of X × X is
A residuated lattice is complete if it is complete as called a (binary) (L-)fuzzy relation on X.
a lattice. A binary fuzzy relation E on X is called fuzzy
The following is a binary operation of biresidua- equivalence on X (see (Klawonn and Castro, 1995;
tion on L : Höhle, 1998; De Baets and Mesiar, 1998))1 if for all
x ↔ y = (x → y) ∧ (y → x). x, y, z ∈ X, the following holds:
The well known examples of residuated lattices 1. E(x, x) = 1, reflexivity,
are: boolean algebra, Gödel, Łukasiewicz and prod- 2. E(x, y) = E(y, x), symmetry,
uct algebras. In the particular case L = [0, 1], multi- 3. E(x, y) ∗ E(y, z) ≤ E(x, z), transitivity.
plication ∗ is a left continuous t-norm.
From now on we fix a complete residuated lattice 1 Fuzzyequivalence appears in the literature under the
L. names similarity or indistinguishability as well.

170
 FuzzyFunctionandtheGeneralizedExtensionPrinciple

If fuzzy equivalence E fulfills On the other side, if St has {t} as a one-element core
1. E(x, y) = 1 if and only if x = y, and fulfills (4), then for all x ∈ X, It (x) ≤ St (x) and
then it is called separated or a fuzzy equality on X. St (x) = St (x) ∗ St (t) ≤ E(t, x) = Et (x).
Let us remark that fuzzy equivalence E creates
fuzzy sets on X, we will call them E-fuzzy points2 of From (4) and the discussion above it follows that E-
X or simply fuzzy points if E is clear from the context. fuzzy point Et is the greatest fuzzy singleton with the
Every E-fuzzy point is a class of fuzzy equivalence E one-element core {t}. The space of all fuzzy single-
of just one point of X. In more details, if t ∈ X, then tons, considered in (X, =, E), will be denoted by SXE .
E-fuzzy point Et is a fuzzy set Et : X → L such that Obviously, SXE ⊆ LX and (SXE , =) is a subspace of
for all x ∈ X, Et (x) = E(t, x). It is easy to see that Et (LX , =).
is a normal fuzzy set and t ∈ Core(Et ). Let us discard normality in the definition of fuzzy
The set of all E-fuzzy points of X will be denoted singleton and define fuzzy sub-singleton as a fuzzy set
by U ∈ LX , such that there exists t ∈ X, so that
PXE = {Et | t ∈ X}. 0 < U(x) ≤ Et (x), x ∈ X. (5)
Obviously, PXE ⊆ LX and (PXE , =) is a subspace of
(LX , =). If E is a fuzzy equivalence on X, then it may In order to stress that a fuzzy sub-singleton is con-
happen that the same element, say Et from (PXE , =) nected with a certain fuzzy point Et , we will denote
has different representations, i.e. there exists u ∈ X it as Ut . Similarly to the above, we can prove that
such that Eu = Et . It can be shown that this holds true any fuzzy sub-singleton fulfills (4). The space of all
if and only if E(t, u) = 1, or u ∈ Core(Et ). fuzzy sub-singletons, considered in (X, =, E), will be
On the other side, if E is a fuzzy equality on X, denoted by UXE . Obviously, SXE ⊆ UXE ⊆ LX and
then the core of every E-fuzzy point consists of one (UXE , =) is a subspace of (LX , =).
element and thus, a representation of any E-fuzzy
point in the form Et is unique. Extensional Hulls. Let again our space be (X, =
, E) – a space with fuzzy equivalence and crisp equal-
Space with Fuzzy Equivalence and Crisp Equal- ity. We remind (Klawonn, 2000) that fuzzy set A is
ity. Fuzzy Singletons and Sub-singletons. Let us extensional (with respect to E) if for all x, y ∈ X,
equip the space X with both crisp = and fuzzy E
equalities and denote it by (X, =, E). In this space, we A(x) ∗ E(x, y) ≤ A(y).
are able to distinguish degrees of coincidence E(t, u) The smallest extensional fuzzy set AE containing
between any two elements t, u from X. As we dis- fuzzy set A is called the extensional hull of A. It is
cussed above, crisp and fuzzy equalities put into the not difficult to prove the following representation of
correspondence with each element t of X its charac- AE .
teristic function It and its E-fuzzy point Et . Both are
normal fuzzy sets in LX with the same one-element Lemma 1. The extensional hull AE of every fuzzy set
core. Let us consider fuzzy sets St ∈ LX , that are in A ∈ LX can be represented as follows:
between It and Et , i.e. for all x ∈ X,
AE (y) = sup A(x) ∗ E(x, y). (6)
x∈X
It (x) ≤ St (x) ≤ Et (x). (3)
We will call them fuzzy singletons. In (Klawonn, Representation (6) has been obtained in many pa-
2000), fuzzy singletons were introduced as normal pers (see e.g.,(Höhle, 1998)), therefore will not prove
fuzzy sets St ∈ LX with {t} as a one-element core, this again.
i.e. St (t) = 1, and such that for all x, y ∈ X, Lemma 1 has two important corollaries.
Corollary 1. Extensional hull of element t ∈ X iden-
St (x) ∗ St (y) ≤ E(x, y), (4)
tified with It is equal to fuzzy point Et .
where ∗ is the monoidal operation from a chosen Corollary 2. Extensional hull of fuzzy singleton St ∈
residuated lattice L. It is easy to show that this is LX , t ∈ X, is equal to the corresponding fuzzy point
equivalent to our definition. Indeed, if St fulfills (3), Et .
then it is normal, it has {t} as a one-element core, and
for all x, y ∈ X,
Decomposition of a Fuzzy Set into Fuzzy Sub-
St (x) ∗ St (y) ≤ E(t, x) ∗ E(t, y) ≤ E(x, y). singletons
2 This notion was introduced in (Klawonn, 2000) Theorem 1. Let A ∈ LX be a non-zero fuzzy set. Then

171
FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

• A can be represented as a supremum of fuzzy sub- 3 FUZZY FUNCTIONS

singletons UtA , t ∈ Supp(A), such that
The notion of fuzzy function has many definitions in
UtA (x) = A(x) ∧ Et (x), x ∈ X, (7)
the literature, see e.g. (Hájek, 1998; Klawonn, 2000;
• A can be represented as a supremum of fuzzy sub- Demirci, 2002; Perfilieva, 2004). In (Hájek, 1998;
singletons WtA , t ∈ Supp(A), such that Klawonn, 2000; Demirci, 2002), a fuzzy function is
considered as a special fuzzy relation. Below, we re-
WtA (x) = A(x) ∗ Et (x), x ∈ X, (8) mind the notion of fuzzy function as it appeared (in-
In both cases, for all x ∈ X, dependently) in (Klawonn, 2000), (Höhle et al., 2000)
and (Demirci, 2002):
A(x) = sup UtA (x) = sup (A(x) ∧ Et (x)), (9)
t∈Supp(A) t∈Supp(A) Definition 2. Let E, F be fuzzy equivalences on X
and Y , respectively. A fuzzy function is a binary fuzzy
and
relation ρ on X ×Y such that for all x, x′ ∈ X, y, y′ ∈ Y
A(x) = sup WtA (x) = sup (A(x) ∗ Et (x)). the following axioms hold true:
t∈Supp(A) t∈Supp(A)
1. ρ(x, y) ∗ E(x, x′ ) ≤ ρ(x′ , y),
(10)
2. ρ(x, y) ∗ F(y, y′ ) ≤ ρ(x, y′ ),
Proof. At first, we will prove that for all t ∈ Supp(A), 3. ρ(x, y) ∗ ρ(x, y′ ) ≤ F(y, y′ ),
UtA and WtA are fuzzy sub-singletons, i.e. UtA and WtA
are non-zero and less than Et . The first assertion fol- A fuzzy function is called perfect (Demirci, 1999), (cf
lows from the assumption t ∈ Supp(A), so that also (Höhle et al., 2000, Section 3.2)) if it additionally
fulfills
UtA (t) =A(t) ∧ Et (t) = A(t) > 0, 1. for all x ∈ X, there exists y ∈ Y , such that ρ(x, y) =
WtA (t) =A(t) ∗ Et (t) = A(t) > 0. 1.
The second assertion easily follows from (7) and (8). A fuzzy function is called (strong) surjective
To prove (9) and (10), we first notice that both of (Demirci, 1999), cf also (Höhle et al., 2000, Section
them are trivially valid for x 6∈ Supp(A). Therefore, 4.2) if
we assume that x ∈ Supp(A). Then (9) follows from 1. for all y ∈ Y , there exists x ∈ X, such that ρ(x, y) =
the two inequalities below: 1.
A(x) = sup UtA (x) = Actually, a fuzzy function ρ establishes a double ex-
t∈Supp(A) tensional correspondence between the space (X, =, E)
= sup (A(x) ∧ Et (x)) ≥ A(x) ∧ Ex (x) = A(x), and the space of (Y, =, F) (axioms FF.1, FF.2) which
t∈Supp(A) is weakly functional (axioms FF.3). Moreover, it is a
point-to-(fuzzy set) mapping between X and LY such
and that for all x ∈ X, ρ(x, ·) is a fuzzy set on Y . If for
A(x) = sup UtA (x) = sup (A(x) ∧ Et (x)) ≤ all x ∈ X, ρ(x, ·) is a normal fuzzy set then ρ is per-
t∈Supp(A) t∈Supp(A) fect, and there is an ordinary function g : X → Y such
that for all y ∈ Y , ρ(x, y) = F(g(x), y) (see (Demirci,
≤ sup A(x) = A(x).
t∈Supp(A) 2002)). This means that every F-fuzzy point Fg(x) of
Y determined by g(x) is a fuzzy value of ρ at x ∈ X.
To prove (10), we recall that in every complete In our study, we will consider the case where ρ is
residuated lattice the following holds true: surjective and defined everywhere on X, i.e.
sup (A(x) ∗ Et (x)) = A(x) ∗ sup Et (x). (∀x ∈ X)(∃y ∈ Y ) ρ(x, y) > 0. (11)
t∈Supp(A) t∈Supp(A)

Because for x ∈ Supp(A), sup Et (x) = 1, we eas- In this case, we will propose an analytic representa-
t∈Supp(A) tion of ρ and use ρ in the generalized extension prin-
ily get ciple. Moreover, we will discover a relationship be-
tween a fuzzy function, its ordinary core function and
A(x) = sup WtA (x) = sup (A(x) ∗ Et (x)) = its extension to a mapping over the domain of fuzzy
t∈Supp(A) t∈Supp(A)
sets.
= A(x) ∗ sup Et (x) = A(x).
t∈Supp(A)

172
 FuzzyFunctionandtheGeneralizedExtensionPrinciple

3.1 Fuzzy Function and Its Core 1. for all x ∈ X, y ∈ Y ,

_
In this Section, we will show that each surjective ρ(x, y) = (Ex′ ′ (x) ∗ Fg(x′ ) (y)), (16)
x′ ∈X ′
fuzzy function ρ on X × Y determines a correspond-
ing ordinary core function g : X ′ → Y , where X ′ ⊆ X, 2. for all t ∈ X ′ , y ∈ Y ,
such that at any x′ ∈ X ′ , the value ρ(x′ , ·) is equal to
the F-fuzzy point Fg(x′ )(·) . The proofs of the below ρ(t, y) = Fg(t) (y). (17)
given Theorems 2 and 3 are in (Perfilieva, 2011).
Theorem 2. Let fuzzy relations E on X and F on
3.2 Generalized Extension Principle
Y be fuzzy equivalences and moreover, F be a fuzzy
In this Section, we will show that every fuzzy func-
equality. Let fuzzy relation ρ on X ×Y be a surjective
tion ρ that establishes a point-to-(fuzzy set) mapping
fuzzy function. For every y ∈ Y , let us choose and fix
between X and LY can be extended (via the Gener-
xy ∈ Core(ρ(x, y)). Denote X ′ = {xy | xy ∈ X, y ∈ Y }.
alized extension principle) to a (fuzzy set)-to-(fuzzy
Then the following fuzzy relation on X
set) mapping between LX and LY . We will use ex-
^
E ′ (x, x′ ) = (ρ(x, y) ↔ ρ(x′ , y)), (12) pression (2), where we replace · by ∗ and use fuzzy
y∈Y relation ρ instead of ordinary R f . Moreover, we will
use our representation (10) of a non-zero fuzzy set
is a fuzzy equivalence E ′ such that and show that the extended mapping between LX and
′

1. E ≤ E ′ and ρ is a fuzzy function with respect to LY is fully determined by its reduction to a certain set
fuzzy equivalences E ′ and F, of fuzzy sub-singletons, and in particular, to sets of
2. for all x ∈ X, y ∈ Y , E ′ -fuzzy points of X.
Definition 3 (Generalized extension principle). Let
ρ(x, y) = E ′ (x, xy ), (13) L be a complete residuated lattice and (LX , =),
3. for all y, y′ ∈ Y , (LY , =) fuzzy spaces. Let E, F be fuzzy equivalences
on X and Y , respectively, and fuzzy relation ρ on
E ′ (xy , xy′ ) = F(y, y′ ), (14) X × Y be a fuzzy function. Then ρ induces the map
fρ→ : LX → LY such that for every A ∈ LX ,
4. the mapping g : X ′ → Y such that g(xy ) = y is sur-
_
jective and extensional with respect to E ′ and F, fρ→ (A)(y) = (A(x) ∗ ρ(x, y)). (18)
i.e. for all x,t ∈ X ′ , x∈X

E ′ (x,t) ≤ F(g(x), g(t)). (15) Theorem 4. Let fuzzy relations E on X and F on Y be

fuzzy equivalences and moreover, F be a fuzzy equal-
Corollary 3. Fuzzy equivalence E ′,
given by (12), is
ity. Let fuzzy relation ρ on X ×Y be a surjective fuzzy
the greatest one (in the sense of ≤) that fulfils Theo-
function and E ′ be fuzzy equivalence given by (12).
rem 2.
Then for any A ∈ LX ,
Corollary 4. Fuzzy equivalence E ′ , given by (12), _
covers X, i.e. for all x ∈ X there exists xy ∈ X ′ such fρ→ (A) = fρ→ (WtA ), (19)
that E ′ (x, xy ) > 0. t∈Supp(A)

Proof. By (11), for arbitrary x ∈ X there exists y ∈ Y , where WtA is a fuzzy sub-singleton (8) in the space
such that ρ(x, y) > 0. By (13), ρ(x, y) = E ′ (x, xy ), and (X, =, E ′ ).
therefore, E ′ (x, xy ) > 0. In particular, if A is represented
W as a supremum of
fuzzy points Et′ , i.e. A = t∈Supp(A) Et′ , then
The meaning of the assertions below is that a sur- _ _
jective fuzzy function ρ is indeed a fuzzified version fρ→ (A)(y) = fρ→ (Et′ )(y) = ρ(t, y).
of its core function g : X ′ → Y , where X ′ ⊆ X. If x ∈ X, t∈Supp(A) t∈Supp(A)
then the fuzzy value of ρ(x, ·) is a “linear”-like com- (20)
bination of F-fuzzy points Fg(x′ ) (·). In particular, if
Proof. By Theorem 1, A can be represented as a
x′ ∈ X ′ - domain of g, then the fuzzy value of ρ(x′ , ·)
supremum of fuzzy sub-singletons WtA , t ∈ Supp(A),
is equal to the corresponding F-fuzzy point Fg(x′ ) (·).
where WtA (x) = A(x) ∗ Et (x), x ∈ X. Thence,
Theorem 3. Let fuzzy relations E, E ′ , F, ρ and func-
tion g : X ′ → Y where X ′ = {xy | y ∈ Y } fulfil assump-
tions and conclusions of Theorem 2. Then

173
FCTA2014-InternationalConferenceonFuzzyComputationTheoryandApplications

_ REFERENCES
fρ→ (A)(y) = (A(x) ∗ ρ(x, y)) =
x∈X
  De Baets, B. and Mesiar, R. (1998). T-partitions. Fuzzy Sets
_ _ Systems, 97:211–223.
=  WtA (x) ∗ ρ(x, y) = Demirci, M. (1999). Fuzzy functions and their fundamental
x∈X t∈Supp(A) properties. Fuzzy Sets and Systems, 106:239 – 246.
_ _ _ Demirci, M. (2002). Fundamentals of m-vague algebra
WtA (x)∗ρ(x, y) = fρ→ (WtA )(y).
and m-vague arithmetic operations. Int. Journ. Un-
t∈Supp(A) x∈X t∈Supp(A) certainty, Fuzziness and Knowledge-Based Systems,
10:25–75.
To prove (20), we first decompose
_ Hájek, P. (1998). Metamathematics of Fuzzy Logic. Kluwer,
fρ→ (A)(y) = fρ→ (Et′ )(y), Dordrecht.
t∈Supp(A) Höhle, U. (1998). Many-valued equalities, singletons and
fuzzy partitions. Soft Computing, 2:134–140.
and then continue with the following chain of equali- Höhle, U., Porst, H., and Šostak, A. (2000). Fuzzy func-
ties tions: a fuzzy extension of the category set and some
_ related categories. Applied General Topology, 1:115–
fρ→ (Et′ )(y) = (Et′ (x) ∗ ρ(x, y)) = 127.
x∈X Klawonn, F. (2000). Fuzzy points, fuzzy relations and fuzzy
_ _
= ′
Et (x) ∗ (Ex′ ′ (x) ∗ Fg(x′) (y)) = functions. In Novák, V. and Perfilieva, I., editors, Dis-
covering the World with Fuzzy Logic, pages 431–453.
x∈X x′ ∈X ′
! Springer, Berlin.
_ _ Klawonn, F. and Castro, J. L. (1995). Similarity in fuzzy
Ex′ ′ (x) ∗ Et′ (x) ∗ Fg(x′) (y) = reasoning. Mathware Soft Comput., pages 197–228.
x′ ∈X ′ x∈X
Novák, V. (1989). Fuzzy Sets and Their Applications. Adam
_
= Ex′ ′ (x) ∗ Fg(x′ ) (y) = ρ(t, y), Hilger, Bristol.
x′ ∈X ′ Perfilieva, I. (2004). Fuzzy function as an approximate so-
lution to a system of fuzzy relation equations. Fuzzy
where we made use of representation ρ by (16). Sets and Systems, 147:363–383.
Perfilieva, I. (2011). Fuzzy function: Theoretical and practi-
cal point of view. In Proc. 7th conf. European Society
for Fuzzy Logic and Technology, EUSFLAT 2011, Aix-
4 CONCLUSION Les-Bains, France, July 18-22, 2011, pages 480–486.
Atlantis Press.
In this contribution we started a mathematical analy- Perfilieva, I., Dubois, D., Prade, H., Godo, L., Esteva, F.,
sis of basic concepts in the fuzzy set theory such as and Hoďáková, P. (2012). Interpolation of fuzzy data:
fuzzy point, fuzzy set and fuzzy function. We intro- analytical approach and overview. Fuzzy Sets and Sys-
duced various spaces of elements equipped with crisp tems, 192:134158.
and fuzzy equivalences with the purpose to show that Šostak, A. (2001). Fuzzy functions and an extension of the
similar to the classical case, a (fuzzy) set is a collec- category l-top of chang-goguen l-topological spaces.
tion of (fuzzy) points or singletons. We recalled the In Proc. 9th Prague Topological Symp., pages 271–
294, Prague, Czech Republlic.
notion of a fuzzy function as a special fuzzy relation
and showed that similarly to the classical case, any Zadeh, L. A. (1975). The concept of a linguistic variable
and its application to approximate reasoning I, II, III.
fuzzy function can be extended to the domain of fuzzy Information Sciences, 8-9:199–257, 301–357, 43–80.
subsets and this extension is similar to the Extension
Principle of L. Zadeh. We clarified a relationship be-
tween a fuzzy function and its ordinary core function.

ACKNOWLEDGEMENTS
The support of the ESF project 2013/0024/1DP/
1.1.1.2.0/13/APIA/VIAA/045 is kindly announced.
Further support comes from the European Regional
Development Fund in the IT4Innovations Centre of
Excellence project (CZ.1.05/1.1.00/02.0070).

174