Professional Documents
Culture Documents
Email: Zadeh@cs.berkeley.edu
October 2, 2002
LOTFI A. ZADEH
COMPUTER SCIENCE DIVISION, DEPARTMENT
OF EECS
UNIVERSITY OF CALIFORNIA
BERKELEY, CA 94720-1776
TEL: (510) 642-4959
FAX: (510) 642-1712
SECRETARY: (510) 642-8271
HOME FAX: (510) 526-2433
E-MAIL: zadeh@cs.berkeley.edu
LAZ 9/23/2002
WHAT IS FUZZY LOGIC AND WHAT ARE ITS
APPLICATIONS
L.A. Zadeh
URL:
http://www.cs.berkeley.edu/People/Faculty/Homepages/zadeh
LAZ 9/23/2002
LAZ 9/23/2002
WHAT IS FUZZY LOGIC?
in essence, fuzzy logic (FL) is
focused on modes of reasoning
which are approximate rather than
exact.
fuzzy logic is aimed at precisiation of
approximate reasoning
in fuzzy logic, everything, including
truth, is or is allowed to be a matter
of degree
in bivalent logic, everything is
either true or false
there is a fundamental conflict
LAZ 9/23/2002
WHAT IS FUZZY LOGIC?
LAZ 9/23/2002
FUZZY LOGIC
The real world is pervaded with
imprecision, uncertainty and partiality –
especially partiality of truth, certainty
and possibility
In the real world, almost everything is a
matter of degree. Absolutes are few and
far between
It is this reality that is the point of
departure in fuzzy logic
The cornerstones of fuzzy logic are
verity, possibility and probability
The role model for fuzzy logic is the
human mind and its remarkable
capability to operate on perception-
based information without any
LAZ 9/23/2002
REASONING WITH WORDS
Business Week 9-18-95
a lower deficit leads to a lower dollar and a higher
deficit pushes the dollar higher. Here is the logic behind
it:
LAZ 9/23/2002
WHAT IS FUZZY LOGIC?
LAZ 9/23/2002
Professor William Kahan (1975)
“Fuzzy theory is wrong, wrong, and pernicious.” says
William Kahan, a professor of computer sciences and
mathematics at Cal whose Evans Hall office is a few
doors from Zadeh’s. “I can not think of any problem that
could not be solved better by ordinary logic.”
“What Zadeh is saying is the same sort of things
‘Technology got us into this mess and now it can’t get us
out.’” Kahan says. “Well, technology did not get us into
this mess. Greed and weakness and ambivalence got us
into this mess. What we need is more logical thinking, not
less. The danger of fuzzy theory is that it will encourage
the sort of imprecise thinking that has brought us so
much trouble.”
LAZ 9/23/2002
STATISTICS
Count of papers containing the word
“fuzzy” in the title, as cited in INSPEC and
MATH.SCI.NET databases. (data for 2001 are
not complete)
LAZ 9/23/2002
“In physical science the first essential step in
the direction of learning any subject is to find
principles of numerical reckoning and
practicable methods for measuring some
quality connected with it. I often say that
when you can measure what you are
speaking about and express it in numbers,
you know something about it; but when you
cannot measure it, when you cannot express
it in numbers, your knowledge is of a meager
and unsatisfactory kind: it may be the
beginning of knowledge but you have
scarcely, in your thoughts, advanced to the
state of science, whatever the matter may
be.”
LAZ 9/23/2002
IN QUEST OF PRECISION
The risk of a 6.0 quake—which could be more
damaging, with one-tenth the destructive power of
the October 17 quake—is 11 percent during the next
two months, the survey’s scientists say.
LAZ 9/23/2002
THE QUEST FOR PRECISION
LAZ 9/23/2002
THE GAP BETWEEN THEORY AND REALITY
LAZ 9/23/2002
IN QUEST OF PRECISION
LAZ 9/23/2002
WHAT IS FUZZY LOGIC? WHY IS IT NEEDED?
(a) B
A D
(b) C
if I choose (a), I will arrive in D at time t1
if I choose (b), I will arrive in D at time t2
t1 is earlier than t2
therefore, I should choose (a) ?
LAZ 9/23/2002
CONTINUED
now, let us take a closer look at the problem
the connection time, cB , in B is short
should I miss the connecting flight from B to D, the
next flight will bring me to D at t3
t3 is later than t2
what should I do?
Dana is young
Tandy is a few years older than Dana
Tandy is ?A
Y is several times larger than X
Y is large
X is ?A
simple examples
Dana is young
Tandy is a few years older than Dana
Tandy is (young + few)
FL/R relational
F: fuzziness/ fuzzification
G: granularity/ granulation
F.G: F and G
LAZ 9/23/2002
The logical facet, FL/L, is focused on
logical systems in which truth is a
matter of degree – a degree which is
allowed to be a fuzzy set
The set-theoretic facet, FL/S, is
concerned, in the main, with the
theory of fuzzy sets. Most of the
mathematical literature on fuzzy
logic relates to FL/S
The relational facet, FL/R, is focused
on fuzzy dependencies, granulation,
linguistic variables and fuzzy rule
sets. Most practical applications of
fuzzy logic relate to FL/R
LAZ 9/23/2002
The epistemic facet, FL/E, is
concerned, in the main, with
knowledge representation, natural
languages, semantics and expert
systems. Probabilistic and
possibilistic modes of reasoning are
a part of this facet as well as FL/L
and FL/R
LAZ 9/23/2002
FROM NUMBERS TO WORDS
LAZ 9/23/2002
VARIABLES AND LINGUISTIC VARIABLES
very old
0
Age
LAZ 9/23/2002
EXAMPLES OF F-GRANULATION (LINGUISTIC
VARIABLES)
color: red, blue, green, yellow, …
age: young, middle-aged, old, very old
size: small, big, very big, …
distance: near, far, very, not very far, …
µ
young middle-aged
old
1
0 age
100
Age
µ µ
1 1
…
0 0
1 2 130 years young middl old
e-
µ refinement aged
attribute value modifiers:
1 very, not very, quite
1 2 12
…
0
1 2 12 months
LAZ 9/23/2002
F-GRANULARITY AND F-GRANULATION
perceptions are f-granular (fuzzy and granular)
fuzzy: unsharp class boundaries
gradual transition from membership to non-
membership
granular: class elements are grouped into granules, with
a granule being a clump of elements drawn together
by indistinguishability, similarity, proximity or
functionality
f-granular is a manifestation of a fundamental limitation
on the cognitive ability of humans to resolve detail and
store information
f-granulation serves two major purposes:
(a) Data compression
(a') Suppression of decision-irrelevant detail
(b) Divide and conquer
LAZ 9/23/2002
A NEGATIVE VIEW
R.E. Kalman (1972)
I would like to comment briefly on Professor Zadeh’s
presentation. His proposals could be severely,
ferociously, even brutally criticized from a technical point
view. This would be out of place here. But a blunt
question remains: Is Professor Zadeh presenting
important ideas or is he indulging in wishful thinking?
The most serious objection of “fuzzification” of system
analysis is that lack of methods of systems analysis is
not the principal scientific problem in the “systems”
field. That problem is one of developing basic concepts
and deep insight into the nature of “systems,” perhaps
trying to find something akin to the “laws” of Newton. In
my opinion, Professor Zadeh’s suggestions have no
chance to contribute to the solution of this basic
problem.
LAZ 9/23/2002
PRINCIPAL APPLICATIONS OF FUZZY
LOGIC
FL
control
consumer
products
industrial
systems
automotive
CFR
decision
analysis
medicine CFR: calculus of fuzzy rules
geology
pattern
recognition
LAZ 9/23/2002
EMERGING APPLICATIONS OF FUZZY LOGIC
computational theory of perceptions
financial engineering
biomedicine
legal reasoning
forecasting
LAZ 9/23/2002
CALCULUS OF FUZZY RULES (CFR)
syntax:legal forms of rules
if X is A then Y is B
if X is A then Y is B unless Z is C
taxonomy: classification of rules
categorical
if X is then Y is B
qualified
if X is A then usually (Y is B)
semantics: meaning of rules
single rule
collection of rules
LAZ 9/23/2002
FUZZY IF-THEN RULES
LAZ 9/23/2002
DEPENDENCY AND COMMAND
Dependency
Y is large if X is small
Y is medium if X is medium
Y is small if X is large
Command
LAZ 9/23/2002
TAXONOMY OF RULES IN FDCL
categorical (examples)
X is A (fact)
if X is A then Y is B or equivalently Y is B if X is A
if X is A and Y is B then U is C and W is D
if X is A then Y is f(A)
if X is A then Action is B (command)
if X is A and Context is B then replace X is A with X is
C (replacement)
if X is A then delete (if X is B then Y is C) (metarule)
if X is A then add (if X is B then Y is C) (metarule)
the more X is A the more Y is B (gradual)
…
LAZ 9/23/2002
TAXONOMY OF RULES IN FDCL
qualified (examples)
if X is A then Y is B unless Z is E (exception)
if X is A then usually (Y is B) (usuality qualified)
usually (if X is A then Y is B)
if X is A and Prob {Y is B|X is A} is C then Action is D
if X is A then possibly (Y is B) (possibility qualified)
(if X is A then Y is B) is possible α (possibilistic)
(if X is A then Y is B) is true α (truth qualified)
…
hybrid (examples)
categorical
If X1 is A1 and … Xn is An then Y is B1 and Yn
is Bn
(sugeno) If X1 is A1 and … Xn is An then Y is (b0 + Σi bi
Xi)
qualified
exception if X is A then Y is B unless Z is E
truth qualified if X is A then Y is B is very true
probability-qualified if X is A then Y is B is likely
possibility-qualified if X is A then Y is B is quite
possible
LAZ 9/23/2002
FUZZY IF-THEN RULES
LAZ 9/23/2002
HONDA FUZZY LOGIC TRANSMISSION
Fuzzy Set Not Very Low
High
1 1 Close 1
High Low High
Grade
Grade
Grade
Low
Not Low
0 30 130 180 0 54 0 5
Speed Throttle Shift
Control Rules:
2. If (speed is low) and (shift is high) then (-3)
3. If (speed is high) and (shift is low) then (+3)
4. If (throt is low) and (speed is high) then (+3)
5. If (throt is low) and (speed is low) then (+1)
6. If (throt is high) and (speed is high) then (-1)
7. If (throt is high) and (speed is low) then (-3)
LAZ 9/23/2002
INTERPOLATION
Y is B1 if X is A1
Y is B2 if X is A2
………..
Y is Bn if X is An
Y is ?B if X is A A≠ A1, …, An
LAZ 9/23/2002
THE “IT IS POSSIBLE BUT NOT PROBABLE”
DILEMMA—THE ROCK ON WHICH MANY CRISP
THEORIES FOUNDER
decision is based on information
in most real-world settings, decision-relevant
information is incomplete, uncertain and
imprecise
to assess the consequences of a decision
when decision-relevant information is not
complete, requires consideration of all
possible scenarios
among such scenarios, a scenario that plays
a pivotal role is the worst-case scenario
LAZ 9/23/2002
THE DILEMMA
LAZ 9/23/2002
NEW TOOLS
computing computing
with numbers with words
CN + + + CW
IA GrC PNL
computing computing precisiated
with intervals with granules natural
language
• a granule is defined
CTP: computational CTP PTp THD
theory of perceptions
by a generalized PTp: perception-based
constraint probability theory
THD: theory of hierarchical
definability
LAZ 9/23/2002
COMPUTATIONAL THEORY OF PERCEPTIONS
imprecision of perceptions is a
manifestation of the bounded ability of
sensory organs and, ultimately, the
brain, to resolve detail and store
information
perceptions are f-granular in the sense
that (a) the boundaries of perceived
classes are fuzzy; and (b) the values of
perceived attributes are granular, with a
granule being a clump of values drawn
together by indistinguishability,
similarity, proximity or functionality
it is not possible to construct a
computational theory of perceptions
within the conceptual structure of
LAZ 9/23/2002
THE ROBERT EXAMPLE
Version 1.
My perception is that
Robert usually
returns from work at
about 6:00pm
LAZ 9/23/2002
DEFINITION OF p: ABOUT 20-25
µ MINUTES
1
c-definition:
0 20 25 time
µ
1
f-definition:
0 20 25 time
µ
1
f.g-definition:
0 20 25 time
P
PNL-definition: B Prob (Time is A) is B
6
LAZ 9/23/2002 A time
WHAT IS A RANDOM SAMPLE?
LAZ 9/23/2002
WHAT IS PRECISIATED NATURAL LANGUAGE (PNL)?
PRELIMINARIES
GCL PFL
abstraction
GC(p) PF(p)
precisiation
of perception
translation
p X isr R
precisiation
explicitation
generalized
constraint form of
type r (GC(p))
copula
type identifier
constraining relation
constrained variable
= (X1 , …, Xn )
may have a structure: X=Location (Residence(Carol))
may be a function of another variable: X=f(Y)
may be conditioned: (X/Y)
r := / ≤ / ... / ⊂/ ⊃/ blank / v / p / u / rs / fg / ps / ...
LAZ 9/23/2002 LAZ 8-14-02
GC-FORM (GENERALIZED CONSTRAINT
FORM OF TYPE r)
X isr R
2: protoform
precisiation
p* (GC-form) PF(p*)
Perception-description language
Definition language
LAZ 9/23/2002
BRITTLENESS OF DEFINITIONS
(THE SORITES PARADOX)
statistical independence
A and B are independent PA(B) = P(B)
suppose that (a) PA(B) and P(B) differ by an
epsilon; (b) epsilon increases
at which point will A and B cease to be
independent?
statistical independence is a matter of
degree
degree of independence is context-
dependent
brittleness is a consequence of bivalence
LAZ 9/23/2002
STABILITY IS A FUZZY CONCEPT
length
volume
edge
cluster
summary
relevance
density
LAZ 9/23/2002
Y
MAXIMUM ?
maximum Y maximum (possibilistic)
interval-valued
0 X 0 X
Y Pareto maximum Y
interval-valued fuzzy-interval-valued
0 X 0 X
Y
fuzzy graph
Bi
Y isfg (∑ iAi×Bi)
0 9/23/2002
LAZ X
WHY IS EXPRESSIVE POWER AN
IMPORTANT FACTOR?
Definition of a concept, construct or metric
may be viewed as a precisiation of perception
of the definiendum
The language in which a definition is
expressed is a definition language
B language bivalent-logic-based
NL
…
0 20 25 time
P
PNL definition: f-granular probability distributio
0
20 25 time
LAZ 9/23/2002 LAZ 7-22-02
DEFINITION OF OPTIMALITY
OPTIMIZATION=MAXIMIZATION?
gain g
yes unsure
ain
0 0
a X a b X
gain no gain
hard to tell
0 a b
X 0 a b c X
definition of optimal X requires use of PNL
LAZ 9/23/2002 LAZ 7-15-02
EXAMPLE
granular P
probability
distribution
likely
unlikely
Time
LAZ 9/23/2002 20 25 LAZ 7-22-02
BRITTLENESS OF DEFINITIONS
statisticalindependence
P (A, B) = P(A) P(B)
stationarity
Kolmogorov, Chaitin, …
in PTp, statistical independence, stationarity,
etc are a matter of degree
P2
P1
X
A1 A2 An
∑ Count ( Bi / Ai ) = ( Σ k1 µ Ai ( ak ) ∧ µ Bi ( b )) / ∑ k µ Ai ( a k )
∑ Count ( Bi / Ai ) ≅ ( ΣCount ( Bi )
PNL-defined
LAZ 9/23/2002 LAZ 8-30-02
PNL-BASED DEFINITION OF STATISTICAL
Y INDEPENDENCE
contingency table
L Σ C(M/L)
3 L/S L/M L/L
M
2 M/S M/M M/L
S Σ C(S/S)
LAZ 9/23/2002
THE ROBERT EXAMPLE
0 • •
6:00 t time
µ ≤ t *
1
t*
time
0 6 pm t
PF(q): Prob(C) is ? D
LAZ 9/23/2002 LAZ 7-22-02
PROTOFORMAL DEDUCTION
THE ROBERT EXAMPLE
IDS p: usually Robert returns from work at about 6 pm.
TDS q: what is the probability that Robert is home at
about 6:15 pm?
5. precisiation:
p Prob {Time (Robert returns from work is
about 6 pm} is usually
q Prob {Time (Robert is home) is about 6:15 pm}
is ?D
• calibration: µusually , µt* , t* = about t
• abstraction:
p* Prob {X is A} is B
q* Prob {Y is C} is ?D
LAZ 9/23/2002 LAZ 9-21-02
CONTINUED
Prob {X is A} is B
not found
Prob {Y is C} is D
t* (about t pm)
0 • •
T t time
time of return
Prob {X is A} is B
Prob {X is C} is D
µ D ( v ) = sup g ( µ B ( ∫ µ A ( u ) g ( u )du ))
U
v = ∫ µ c ( u ) g ( u )du
U