You are on page 1of 16

8 7

• •
lftl,utvadllble•A• _ _ _ ____.,

id'IIIZY,~RmcnemforlapltVJII It ~
4 5
1 - - - - - - Output variable •score·
---------
_ _ _ _ __,..

II •
11& e 11.26 Fuzzy Membership Functions for Outpllt V.ial •

.,. SNpmrC
NmDli
then.lttem
Hin Ip l'uncliun) II

m81ftlllld B i1 not 1t..a tl'dlr


I)/ 2 lll 0.15......1.. . . ._... . . . . . . . . . . . .
U1ll!•(lcon t, ace/Jav
MP• 0, therd'm! CiUIPdf 11tr.

1tMn (Scor~ is good)


MF= I, therefote output • (.1"le

p,,4) tlr«n (Sear~ ls-8'J(Jtl>


"ch MF= 0, therefore autput e
2 3 4 5 8 7 8

ripr e 10.28 Total Fuzzy Area for Fuzzy Output Variable Score

s Ol1& a,
for this example, crisp value corresponding of fuzzy variable Score come
cala•Jated below:
IO)x.21
(3 + 4 +5) xo.21 +(6+ 7 +s)x 0.36+ (9+--=
(o+ 1 + 2)x0. 15 + ---
CO G= --- --- --- --- --- --- --- --- 6.S
0.15+ 0.15 + 0.15 +0.21 + 0.21 +0.21 +0.36 +0.36 +0.36 +0.21 +0.ll
The Table l 0.13 consists of the overall scores (in crisp form) for different crisp
va•
It is clearly evident from the table that the score is higher for academically good
for academically poor and excellent in other activities will give overall poor
score.
ily
of the fact that the rules used in the system are framed based on higher pnor
Working for highlighted row is shown above.
Table 10.13 Summary ofResults
A(Ac....k} E (Eura Activities)
9 9
9 2
8 7
7.5 4
4 9
7 2
5 4
4 2
4
35
2
2
9
2
Fuzzy Sots and Fuzzy Logic 397

_1o.4 Neuro-Fuzzy System


10
i
.... r sysh::111 refer,
~r11ro·r•'i,__..... • tn l:omhinations
. · of ',11 ,·r·
u ... c~ u lcnrmng ~apnh11itv of ,wurnl ,
• 1 ncunil
1 1~1r1
·k
nctw 0 r~- • d r
· ' ,tn
.
uuy logic . Thi..,
,ten1 . .• 11 c.: 1wrn and rcuso111ng • p· h'I
:-) · I networks can lcnrn tnm1 du1n. hut can 1101 ·x "I·. I t.
Nrttffl l""
t. I•
' tel11' com,ist of intet prctahlc- linguislk I ules <.;an rc·1s ' h .
ca a , lly ol fuz1.y logic.
10w t icy reach to th · d · ·
. ~•r ec1.;;sons Fuvy
.,~ . . . . .h . i . . ' l , on wit ,mprcc1~c informatmn and
· , explmnrnu .... t <.'tr l cc1s1ons hut they "'' ,.. ru 1cs automatically
,.. 111 11 ,)... 1 1c,irn . Th 1 · are
, ~~e results
£l10l •
•:i,-e1e,ohed· till' co1h.:cpt
111
, Iof hyhrid
·1· intelligent systems. A neuro-t'uz.1.y.
'h Yb n.d·.tlallun amnauon!-t
ma
h hrid rcH1genl sy~tcm. t rnt ut1 tzcs both the tc<.:hniqucs by ,·omba·n 1•0 th h , i·k .
~ '' 111 . , , ·~ , ,· h ~ . . "" g c uman- 1 e reason mg
h
~t) le ()I fu11.) ~) s~c ms \\ It th c karnmg usmg neural network~. Neuro-fuzzy hybridizatmn is
"iddY renned as fuzzy N~ural Nct~ork (FNN) or Neuro-Fu.u.y System (NFS) in literature used
n1erchangeably. The learmng algonthms can learn both fuzLy sets. and fuzly rules. and can also
1
t.'S!' prior knowledge.
On1."e the fuzzy sets an~ ru!es are set up, they do not change. It does not learn from new inputs. But
since the rules ~e subjective ~nd vag~e, fuzzy systems should be designed to learn from experi-
e.nce. Fuzzy logic can be used m combmation with neural networks to produce fuzzy systems that
sre able to adapt and learn. The features distinguishing fuzzy neural networks from classical
neural network systems are that inputs, outputs, and weights are fuzzy numbers. In FNN. weighted
inputs of each neuron are not simply aggregated by summation but by aggregation operations on
fuzzy sets. Here. several fuzzy sets are combined in a desirable way to produce a single fuzzy set.
Neuro-fuzzy systems are usually represented as 3-layered feedforward neural networks (refer to
chapter on Neural Network). The first layer represents input variables, the middle (hidden) layer
represents fuzzy rules and the third layer represents output variables. Each rule of the form if x is
Ai, then y is Bi, where Ai and Bi are fuzzy sets, 1 sis n, can be interpreted as a training pattern for
amultilayer neural network, where the antecedent part of the rule is the input and the consequent
Part of the rule is the desired output of the neural net. The training set using above rules can be
Written in the form {(AI ,B 1), ... , (An,Bn)}.
Similarly, for two-input-single-output fuzzy systems of the form ifx is Ai and y is Bi. then z is Ci,
where Ai, Bi and Ci are fuzzy sets, I ~ i ~ n, the input/output training pairs for the neural net are
of the form {((A J,BI), C 1), ... , ((An,Bn), Cn)}.

10-10.s Applications of Fuzzy Set Theory


~ ~ajor_area of application of fuay set the_ory and fuzzy logic is the fiel~ of eng'.ne.e~i_ng. All
se!;"eenng disciplines have benefited to vanous. degree~ by the met~odolog~cal p_oss1b1hues ~r~-
%' ed by fuzzy set theory. In this section, we will mention som~ of the engmeenng problems ID
ich the use of fuzzy theory has already been proven to be usetul.
thWebegm · our discussion · h are t h
with fuzzy controller systems, whtc · . ·ri
e most s1gm 1cant systems .ID
e field of electrical engineering that makes use of fuzzy sets and logic. Applications such as
1
6~ im~~e processing and electronic circuits for fuzzy logic are also related 'to this field. ~n the
of CIVIi engineering, fuzzy set theory has been found useful for the problems of evaluation or

. . ,...lntlllglnol
Ulllllftllltofulltlna conatructtons, quality of ht1hw1y pavements. phy,m.: al co
way ~ etc Fuuy systems baaed on the de1t1n experience over the cc
types of CQllltNCtions are very utteful in this era. These ~ystems arc convenaent for
lll'Yicl Ufe of a machinery or equipment for different cnn<lit10ns under whlffi daly
meclcal science, fuuy expert sy~tems can handle monitoring and diagnosi1 of atlllll
..... input parameters are of fuzzy nature. Thus. we note that fu1zy expert systelU
common use of fuzzy logic. Other relevant areas of applications of fuzzy set dleary
liaear and non-linear control. pattern recognition. financial systems, operauon
ualysis. infonnation retrieval from fuzzy databases, and so on.

10.11 Poulblllty Theory


•• •
Having reached thus far in our discussion of fuzzy set theory and fuzzy ICJ8ic. the
to discuss yet another important theory known as the possibility theory. Poui
is a mathematical theory for dealing with certain types of uncertainty and is an
probability theory. A possibility measure can be seen as plausibility measure in DelDDl111
theory of evidence. Possibility can be seen as an upper probability, i.e., any pouibilily
tion defines a unique set of admissible probability distributions by (p: VS p(S) S pol(
bility theory was first introduced by Professor L. Zadeh in the year 1978 as an extenlial1
sets and fuzzy logic. D. Dubois and H. Prade further contributed to its development.
theory uses two concepts, namely, the possibility and the necessity of the event. whel.._
ity theory uses a single number, the probability to describe how likely an event is to

10.11.1 Formallzatlon of Poaslblllty


For simplicity, assume that the universe of discourse U is a fmilo set and all
measurable. A distribution of possibility is a function pos: U ➔ [0, 1) such dial the
axioms hold true:

Axiom-3 corresponds to the additivity axiom in


practical difference. The possibility measure on finite set
individual element similar to probability:
Fuzzy Seti and Fuzzy logic 311

llOllibility theory is computationally more convenient becau!le one can


emoa fmm the posaibUity of each component II can be ~•d that
Witb. respect to union operator. However. it should be noted thar it it
llllliect to the intersection operator. Generally. the possibility of interaec.-
lO minimum of possibility of individual set

~aa.•1ty using Poulblllty


1

ore 1s defined as follows:

)] = 0 andmax[pos(S), pos(-S)] = 1. However, tbei ·


,,....

rtttpetadon. 1bese ue lisled below



Two - 1-...... ----·
lofk. n.e bave been e11di11CUM01:l lJlil
Fuzzy Seti and Fuzzy logic 401

ur-::;:
•p
lo,iic lhat is a sys1em o( formal logic dealing Willi modalilia
ability• and necessity· Logics for dealina wldi rellled tenns
, can, could, might. may. must are also called modal loaics. as t ~
~ <Benthem. Van J. et. al. 1983, 2006).

e~prasive po~er of traditional logic by adding modal operaton and


meanmgs of relations for concepts such as consistency, possibility, neces-
wn truths. and temporal situations, such as past, present, and future. In
I/Jfllsettt possible worlds in addition to the actual world in which we live.
where interpretation of a well-formed formula results in an assignment of
-ct.enly. an interpretation in modal logic would be given for each pouible
no well-formed formula may be true in one world and may be fae ia
with reasoning that involves the use of the expressions MCUNrily aod
.......Jqgic is used more broadly to cover a family of logics with similar sules
symbols. The most familiar logics in the modal family are consrrueted
tf~tem proposed by Saul Kripke. Under this family~ dlOdat IDlic ia
and possibility as listed in Table 10.14.
Modal Logic Operators
Expressions Symbolized
It is necessary that ...
It is possible that.•.

may be developed for such logics U~llf K-s:


-sygtem include - for not and ➔ for IF-
med using - and ➔ as dofi ·
.. modal operators.
with an accilknt..
ibiJity,. la modal
f . . .
nt If and only if it is no
aa follows:

IJllllator l• defined in term• of nece11ity


•According to this, some statement ot
t, n,cessarily true.

po,sibly Q is equivalent to not necessarily not 1'~

u,,,..,,
-:allifill••• fru pouib~ that Delhi is older than Chandi1arll If...4'11'1
r·•-
.•u•· II_, oltMr
lu
than Chandigarh. We can repreaent It

PNcilcly, the axioms and rules which must be added to the tnllOldm1181 all...
_.lie lyam of modal logic is a matter of philosophical opinioJL Tky as
ea••t GIii- wishes to prove. Many modal logics, known coltecdV~ ••••11111•
-=---•followhig rule and moms:

Adding axioms
K-system that if'

Other well-

In K-system, the operators □ and ◊ beha1e.;


example, the definition of ◊ from □ mlrran Iii
lo&ic. Purtbmnore, following results alio llelf
. . Ir& 711111.i PIE '
..... ff•,...._ ... ,,,.••• r ,,,,_fl•PA-•~,,,,.
.....mQ

.....
. . 11 Mat ........MW. . . . . . . , , , , . . , . . ad otlMn ra1c1 are «r·

CODCladeOR.
CONfudo OR
OQtbaloom b,de
➔ OR.ad OQ then nwMtndf! OR

· ~ ➔OR.and DQdmCXIDt:1trdD

IJ-.ple 10.4 Consider the follol\ing ~ons:


i Ram is a man.
i <> (Ram is a child) [i.e., Ram is pombly a child]
ii o (Ram is a child) ➔ (Ramis a child)
iv. □ [(Ram is a man) ➔-, (Ram is a child)]
Then. ?- -, (Ram is a child)
.w ,,,;,,,, Using Q ➔ o Q, and (iv). we get
,·. [(Ram is a man) ➔-, (Ram is a child)]
Now using Modus ponen rule again on (v) and (i), we get
-, (Ram is a child)

10.12.2 Temporal Logic


The term temporal log,c is used to describe any systan of
and reasoning about, propositions qualified in terms of tilllL
logic. More precisely, we can say that temporal logic . .u
tors in rdation to concepts of time, such as
particular modal logic-based system oftemporal
The INIIIIVVnl logic contains, in addition to the
i'"'"'84r,d weninp as given in Table 10.15. Out
tbtuR. wllilc dw: operaton H and P arc usa1 fer mtiaa
temo lc)lie (llc:w:lw- and Urquhart. 1971).
1' LaaloOl,11 ,.
,_, .._
llprc 11 • a, tat I
It will alwaya be dlia - 111M
It Will at...._. bl._ . . . . .
ltba1at....,,.mwlhe011e..._

II the weak tame operators, while die opw PafJ\H •


•n. the two pairs are generally regardccl II ialer
Cf•
M.,.., lij11fi/
Q is some closed formula in CODVadioaal lope. n. Iola 1 SMP
tietween the past and future operators:

~--!d . ._.. .
ngsAlil, Arthur used the operators to bai1cl
yprdillag · , which have been takm as axiclM
-111-•
•'4••••1••·

....,.._ta.._. atetJl:IQ, HQ, and FQ are theorems.


406 Artificial Intelligence

• Q -+ HFQ
H('lwc, er. Q -+ HFQ 11my appear to IHI\ c unacceptably determ111ist1c. for it claims, Uiat what is true now
has nh,:t)S been such thut it \\ill occur in the future (IIFQ). rts interpretation is What ,s. has alwlI\lr
gomg to be

Ho\\e, er, possibk world semantics for temporal logic reveals that the two intcrattion axi01111
equally acceptable. It should be noted that the characteristic axiom of moda] logic. □Q ➔ Q, •
acceptable for either H or G, since Q docs not follow from it always was the case that Q.
from 1t always w1/l be the case that Q. However, it is acceptable in a closely related
logic where G is read as 11 1s and always will be, and H is read as 11 1s and always was

Depending on \vhich assumptions we make about the structure of time, further axioms may
added to temporal logics . A set of axioms commonly adopted in temporal logics are as fol

• GQ ➔ GGQ and HQ ➔ HHQ


• GGQ ➔ GQ and ffilQ ➔ HQ
• GQ ➔ FQ and HQ ➔ PQ

It is interesting to note that certain combinations of past and future tense operators may be u
express complex tenses in English. For example, FPQ, corresponds to sentence Q in the
perfect tense. Similarly, PPQ expresses the past perfect tense . To be more precise. some
axtoms and inference rules in temporal logic are stated below.

Axioms in Temporal Logic

Herc we are listing six useful axioms in Temporal Logic:

L It has always been the case that Q is true is equivalent to It is not in the past that Q was
HQ = ~P--Q
ii // will always be the case that Q will be true is equivalent to It is not ,n future that Q will
GQ = ~F-Q
iii. It will be lhe case in future that Q will be true is equivalent to It will not be always that Q
false
FQ : -G-Q
1,•. It was the case that Q was true is equivalent to It has not always be the case m past that Q
PO = ~H-Q
v. It has not always heen the case in past that Q was true is equivalent to It was the case m
Q wasfalse
~HQ = P-Q
vi. It will not always be the.,case that Q will be true is equivalent to It will be the cas~ in the
Q will be.false.
-GQ : F-Q
emporal Logic can be u8eCf for iftlltalCIJIS -.,,,,,,.• • • •

......... I.Ogle
.,_
area of natural language processing. Reiche
iag to which the function of each tense is
three times, namely the speech time (S),
utterance. In this way Reichenbach co
ecL The statement / .faw a snaa, for
event of my seeing snake. while the •-•N
&Is p,cseeit time, relative to w.hi~h my ~ --• _ , •
·vea by Reichenbach l6 not s
• •11119111• Sut,soquently, a lot of
~ioJI& in lan-illalO
•ng. until l. u.1ing llie

· • - · " ' witb Ii .


4111 L1 ■ Pi•J21 II

TIie cul. CII II 5-al 1111 e ha ill Al._ been of importence from the point of view
far lllf - - - - - I I Ir, bow. Gr be able to deduce, thole properties of tbe
-t r • • w at fl • 11& n.1 •Pt of any event or action. In everyday life, we 1181111tl
__.. feels c~ 11 1 ?!ly wiwlwNI comciously being aware of them. For example, the COllDe
wllllllcl: I aa arfi-•iolloflntes.
lb I I 11,te CJI • SjNaal logic bu found extensive application in the area of CoqMla
• J C 1lle sp:cilv•N wt verification of programs concurrently w-o.ting in parallel
• Clllllllle Wiled bellniour of such a program it is necessary to specify interrelatioa
-1 a lk N.iom processors and the relative timing of the actions.
Te ife. llpl.ticl4QII of tcmpotal logic include its use as a formalism for c · ·
•Yf' · al · • .._,. lime, as a framework within which to define the semaoiics of
eap=a · r• ia r +al language, as a language for encoding temporal knowledge in am·na
fig a e, all• a tool for haoclling the temporal aspects of the execution of computer

Exercises
•••• •

- MU LdPI _. P2 be two fuzzy sets whose membership functions are defined by die followint

l
(x+4)/4, for-4<xSOI l(x-4)/4, for4<x:S41
Pl(a)= (4-x)/4, for0<x<4 F2(x)= (6-x)/2, fm0<x$6
O. otbeawise 0, odlerwilc
C• ◄ a die fuzzy sets Fl + F2, Fl - F2, Fl . Fl, Min(Fl, Pl). and Max(Pl~ Fl).
lt..2 Aza,me Dnffene of ducoune for set of aalary structures (in terma of. K) per IIJOillh is
U = 12. S, 10, 20, JI), 40, 50, 60, 70, SOJ.
The fuzzy ldsdefiaedoa U bated on aalariel are
Poor• ((2, I), (5, I), (10, 0.8). (~ 0.5), (30. 0.2})
Avera,e • ( (10, 0.2), (20, 0.5), (30, 0.6), («), QA <IA ct6). • • ea
Rlclt • 1(40. 0.2). (50. 0.5), (60, 0.8), (10. t)f
Draw die graphs of all three fuzzy ICU oa lbe ume coo,f!c
yourage,
11.J CoDlida lhe depec of adon a& e,_,.,,_,,
1elloo'l,
tuapittic variable ICYel_of_edacaaion• • {not_lai.P.IJ.J.l ~I~

You might also like